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 = zap_evict;
86 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
87 zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1;
91 * explicitly zero it since it might be coming from an
92 * initialized microzap
94 bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
95 zp->zap_block_type = ZBT_HEADER;
96 zp->zap_magic = ZAP_MAGIC;
98 zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);
100 zp->zap_freeblk = 2; /* block 1 will be the first leaf */
101 zp->zap_num_leafs = 1;
102 zp->zap_num_entries = 0;
103 zp->zap_salt = zap->zap_salt;
104 zp->zap_normflags = zap->zap_normflags;
105 zp->zap_flags = flags;
107 /* block 1 will be the first leaf */
108 for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
109 ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;
112 * set up block 1 - the first leaf
114 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
115 1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
116 dmu_buf_will_dirty(db, tx);
118 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
121 zap_leaf_init(l, zp->zap_normflags != 0);
123 kmem_free(l, sizeof (zap_leaf_t));
124 dmu_buf_rele(db, FTAG);
128 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
130 if (RW_WRITE_HELD(&zap->zap_rwlock))
132 if (rw_tryupgrade(&zap->zap_rwlock)) {
133 dmu_buf_will_dirty(zap->zap_dbuf, tx);
140 * Generic routines for dealing with the pointer & cookie tables.
144 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl,
145 void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n),
149 dmu_buf_t *db_old, *db_new;
151 int bs = FZAP_BLOCK_SHIFT(zap);
152 int hepb = 1<<(bs-4);
153 /* hepb = half the number of entries in a block */
155 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
156 ASSERT(tbl->zt_blk != 0);
157 ASSERT(tbl->zt_numblks > 0);
159 if (tbl->zt_nextblk != 0) {
160 newblk = tbl->zt_nextblk;
162 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2);
163 tbl->zt_nextblk = newblk;
164 ASSERT0(tbl->zt_blks_copied);
165 dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
166 tbl->zt_blk << bs, tbl->zt_numblks << bs,
167 ZIO_PRIORITY_SYNC_READ);
171 * Copy the ptrtbl from the old to new location.
174 b = tbl->zt_blks_copied;
175 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
176 (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH);
180 /* first half of entries in old[b] go to new[2*b+0] */
181 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
182 (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
183 dmu_buf_will_dirty(db_new, tx);
184 transfer_func(db_old->db_data, db_new->db_data, hepb);
185 dmu_buf_rele(db_new, FTAG);
187 /* second half of entries in old[b] go to new[2*b+1] */
188 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
189 (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
190 dmu_buf_will_dirty(db_new, tx);
191 transfer_func((uint64_t *)db_old->db_data + hepb,
192 db_new->db_data, hepb);
193 dmu_buf_rele(db_new, FTAG);
195 dmu_buf_rele(db_old, FTAG);
197 tbl->zt_blks_copied++;
199 dprintf("copied block %llu of %llu\n",
200 tbl->zt_blks_copied, tbl->zt_numblks);
202 if (tbl->zt_blks_copied == tbl->zt_numblks) {
203 (void) dmu_free_range(zap->zap_objset, zap->zap_object,
204 tbl->zt_blk << bs, tbl->zt_numblks << bs, tx);
206 tbl->zt_blk = newblk;
207 tbl->zt_numblks *= 2;
210 tbl->zt_blks_copied = 0;
212 dprintf("finished; numblocks now %llu (%lluk entries)\n",
213 tbl->zt_numblks, 1<<(tbl->zt_shift-10));
220 zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val,
225 int bs = FZAP_BLOCK_SHIFT(zap);
228 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
229 ASSERT(tbl->zt_blk != 0);
231 dprintf("storing %llx at index %llx\n", val, idx);
234 off = idx & ((1<<(bs-3))-1);
236 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
237 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
240 dmu_buf_will_dirty(db, tx);
242 if (tbl->zt_nextblk != 0) {
243 uint64_t idx2 = idx * 2;
244 uint64_t blk2 = idx2 >> (bs-3);
245 uint64_t off2 = idx2 & ((1<<(bs-3))-1);
248 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
249 (tbl->zt_nextblk + blk2) << bs, FTAG, &db2,
250 DMU_READ_NO_PREFETCH);
252 dmu_buf_rele(db, FTAG);
255 dmu_buf_will_dirty(db2, tx);
256 ((uint64_t *)db2->db_data)[off2] = val;
257 ((uint64_t *)db2->db_data)[off2+1] = val;
258 dmu_buf_rele(db2, FTAG);
261 ((uint64_t *)db->db_data)[off] = val;
262 dmu_buf_rele(db, FTAG);
268 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
273 int bs = FZAP_BLOCK_SHIFT(zap);
275 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
278 off = idx & ((1<<(bs-3))-1);
280 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
281 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
284 *valp = ((uint64_t *)db->db_data)[off];
285 dmu_buf_rele(db, FTAG);
287 if (tbl->zt_nextblk != 0) {
289 * read the nextblk for the sake of i/o error checking,
290 * so that zap_table_load() will catch errors for
293 blk = (idx*2) >> (bs-3);
295 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
296 (tbl->zt_nextblk + blk) << bs, FTAG, &db,
297 DMU_READ_NO_PREFETCH);
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)
312 for (i = 0; i < n; i++) {
313 uint64_t lb = src[i];
320 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
323 * The pointer table should never use more hash bits than we
324 * have (otherwise we'd be using useless zero bits to index it).
325 * If we are within 2 bits of running out, stop growing, since
326 * this is already an aberrant condition.
328 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2)
329 return (SET_ERROR(ENOSPC));
331 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
333 * We are outgrowing the "embedded" ptrtbl (the one
334 * stored in the header block). Give it its own entire
335 * block, which will double the size of the ptrtbl.
341 ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
342 ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
343 ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk);
345 newblk = zap_allocate_blocks(zap, 1);
346 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
347 newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new,
348 DMU_READ_NO_PREFETCH);
351 dmu_buf_will_dirty(db_new, tx);
352 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
353 db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
354 dmu_buf_rele(db_new, FTAG);
356 zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk;
357 zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1;
358 zap_f_phys(zap)->zap_ptrtbl.zt_shift++;
360 ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
361 zap_f_phys(zap)->zap_ptrtbl.zt_numblks <<
362 (FZAP_BLOCK_SHIFT(zap)-3));
366 return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl,
367 zap_ptrtbl_transfer, tx));
372 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
374 dmu_buf_will_dirty(zap->zap_dbuf, tx);
375 mutex_enter(&zap->zap_f.zap_num_entries_mtx);
376 ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta);
377 zap_f_phys(zap)->zap_num_entries += delta;
378 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
382 zap_allocate_blocks(zap_t *zap, int nblocks)
385 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
386 newblk = zap_f_phys(zap)->zap_freeblk;
387 zap_f_phys(zap)->zap_freeblk += nblocks;
392 zap_leaf_pageout(void *dbu)
396 rw_destroy(&l->l_rwlock);
397 kmem_free(l, sizeof (zap_leaf_t));
401 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
404 zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
406 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
408 rw_init(&l->l_rwlock, 0, 0, 0);
409 rw_enter(&l->l_rwlock, RW_WRITER);
410 l->l_blkid = zap_allocate_blocks(zap, 1);
413 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
414 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
415 DMU_READ_NO_PREFETCH));
416 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
417 winner = dmu_buf_set_user(l->l_dbuf, &l->l_dbu);
418 ASSERT(winner == NULL);
419 dmu_buf_will_dirty(l->l_dbuf, tx);
421 zap_leaf_init(l, zap->zap_normflags != 0);
423 zap_f_phys(zap)->zap_num_leafs++;
429 fzap_count(zap_t *zap, uint64_t *count)
431 ASSERT(!zap->zap_ismicro);
432 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
433 *count = zap_f_phys(zap)->zap_num_entries;
434 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
439 * Routines for obtaining zap_leaf_t's
443 zap_put_leaf(zap_leaf_t *l)
445 rw_exit(&l->l_rwlock);
446 dmu_buf_rele(l->l_dbuf, NULL);
450 zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
452 zap_leaf_t *l, *winner;
456 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
457 rw_init(&l->l_rwlock, 0, 0, 0);
458 rw_enter(&l->l_rwlock, RW_WRITER);
460 l->l_bs = highbit64(db->db_size) - 1;
463 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
464 winner = dmu_buf_set_user(db, &l->l_dbu);
466 rw_exit(&l->l_rwlock);
467 if (winner != NULL) {
468 /* someone else set it first */
469 zap_leaf_pageout(&l->l_dbu);
474 * lhr_pad was previously used for the next leaf in the leaf
475 * chain. There should be no chained leafs (as we have removed
478 ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1);
481 * There should be more hash entries than there can be
482 * chunks to put in the hash table
484 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
486 /* The chunks should begin at the end of the hash table */
487 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
488 &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
490 /* The chunks should end at the end of the block */
491 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
492 (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size);
498 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
503 int bs = FZAP_BLOCK_SHIFT(zap);
506 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
508 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
509 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
513 ASSERT3U(db->db_object, ==, zap->zap_object);
514 ASSERT3U(db->db_offset, ==, blkid << bs);
515 ASSERT3U(db->db_size, ==, 1 << bs);
518 l = dmu_buf_get_user(db);
521 l = zap_open_leaf(blkid, db);
523 rw_enter(&l->l_rwlock, lt);
525 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
526 * causing ASSERT below to fail.
529 dmu_buf_will_dirty(db, tx);
530 ASSERT3U(l->l_blkid, ==, blkid);
531 ASSERT3P(l->l_dbuf, ==, db);
532 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF);
533 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
540 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
542 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
544 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
546 (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift));
547 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
550 return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl,
556 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
559 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
561 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
562 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
565 return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl,
571 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);
578 ASSERT3U(zap_f_phys(zap)->zap_magic, ==, ZAP_MAGIC);
579 idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
580 err = zap_idx_to_blk(zap, idx, &blk);
583 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
586 ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) ==
587 zap_leaf_phys(*lp)->l_hdr.lh_prefix);
592 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l,
593 void *tag, dmu_tx_t *tx, zap_leaf_t **lp)
595 zap_t *zap = zn->zn_zap;
596 uint64_t hash = zn->zn_hash;
598 int prefix_diff, i, err;
600 int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len;
602 ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
603 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
605 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
606 zap_leaf_phys(l)->l_hdr.lh_prefix);
608 if (zap_tryupgradedir(zap, tx) == 0 ||
609 old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
610 /* We failed to upgrade, or need to grow the pointer table */
611 objset_t *os = zap->zap_objset;
612 uint64_t object = zap->zap_object;
615 zap_unlockdir(zap, tag);
616 err = zap_lockdir(os, object, tx, RW_WRITER,
617 FALSE, FALSE, tag, &zn->zn_zap);
621 ASSERT(!zap->zap_ismicro);
623 while (old_prefix_len ==
624 zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
625 err = zap_grow_ptrtbl(zap, tx);
630 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
634 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) {
635 /* it split while our locks were down */
640 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
641 ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
642 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
643 zap_leaf_phys(l)->l_hdr.lh_prefix);
645 prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
646 (old_prefix_len + 1);
647 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
649 /* check for i/o errors before doing zap_leaf_split */
650 for (i = 0; i < (1ULL<<prefix_diff); i++) {
652 err = zap_idx_to_blk(zap, sibling+i, &blk);
655 ASSERT3U(blk, ==, l->l_blkid);
658 nl = zap_create_leaf(zap, tx);
659 zap_leaf_split(l, nl, zap->zap_normflags != 0);
661 /* set sibling pointers */
662 for (i = 0; i < (1ULL << prefix_diff); i++) {
663 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx);
664 ASSERT0(err); /* we checked for i/o errors above */
667 if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) {
668 /* we want the sibling */
680 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l,
681 void *tag, dmu_tx_t *tx)
683 zap_t *zap = zn->zn_zap;
684 int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
685 int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift &&
686 zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
690 if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) {
694 * We are in the middle of growing the pointer table, or
695 * this leaf will soon make us grow it.
697 if (zap_tryupgradedir(zap, tx) == 0) {
698 objset_t *os = zap->zap_objset;
699 uint64_t zapobj = zap->zap_object;
701 zap_unlockdir(zap, tag);
702 err = zap_lockdir(os, zapobj, tx,
703 RW_WRITER, FALSE, FALSE, tag, &zn->zn_zap);
709 /* could have finished growing while our locks were down */
710 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift)
711 (void) zap_grow_ptrtbl(zap, tx);
716 fzap_checkname(zap_name_t *zn)
718 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
719 return (SET_ERROR(ENAMETOOLONG));
724 fzap_checksize(uint64_t integer_size, uint64_t num_integers)
726 /* Only integer sizes supported by C */
727 switch (integer_size) {
734 return (SET_ERROR(EINVAL));
737 if (integer_size * num_integers > ZAP_MAXVALUELEN)
744 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
748 if ((err = fzap_checkname(zn)) != 0)
750 return (fzap_checksize(integer_size, num_integers));
754 * Routines for manipulating attributes.
757 fzap_lookup(zap_name_t *zn,
758 uint64_t integer_size, uint64_t num_integers, void *buf,
759 char *realname, int rn_len, boolean_t *ncp)
763 zap_entry_handle_t zeh;
765 if ((err = fzap_checkname(zn)) != 0)
768 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
771 err = zap_leaf_lookup(l, zn, &zeh);
773 if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
778 err = zap_entry_read(&zeh, integer_size, num_integers, buf);
779 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
781 *ncp = zap_entry_normalization_conflict(&zeh,
782 zn, NULL, zn->zn_zap);
791 fzap_add_cd(zap_name_t *zn,
792 uint64_t integer_size, uint64_t num_integers,
793 const void *val, uint32_t cd, void *tag, dmu_tx_t *tx)
797 zap_entry_handle_t zeh;
798 zap_t *zap = zn->zn_zap;
800 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
801 ASSERT(!zap->zap_ismicro);
802 ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
804 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
808 err = zap_leaf_lookup(l, zn, &zeh);
810 err = SET_ERROR(EEXIST);
816 err = zap_entry_create(l, zn, cd,
817 integer_size, num_integers, val, &zeh);
820 zap_increment_num_entries(zap, 1, tx);
821 } else if (err == EAGAIN) {
822 err = zap_expand_leaf(zn, l, tag, tx, &l);
823 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
830 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
835 fzap_add(zap_name_t *zn,
836 uint64_t integer_size, uint64_t num_integers,
837 const void *val, void *tag, dmu_tx_t *tx)
839 int err = fzap_check(zn, integer_size, num_integers);
843 return (fzap_add_cd(zn, integer_size, num_integers,
844 val, ZAP_NEED_CD, tag, tx));
848 fzap_update(zap_name_t *zn,
849 int integer_size, uint64_t num_integers, const void *val,
850 void *tag, dmu_tx_t *tx)
854 zap_entry_handle_t zeh;
855 zap_t *zap = zn->zn_zap;
857 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
858 err = fzap_check(zn, integer_size, num_integers);
862 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
866 err = zap_leaf_lookup(l, zn, &zeh);
867 create = (err == ENOENT);
868 ASSERT(err == 0 || err == ENOENT);
871 err = zap_entry_create(l, zn, ZAP_NEED_CD,
872 integer_size, num_integers, val, &zeh);
874 zap_increment_num_entries(zap, 1, tx);
876 err = zap_entry_update(&zeh, integer_size, num_integers, val);
880 err = zap_expand_leaf(zn, l, tag, tx, &l);
881 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
887 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
892 fzap_length(zap_name_t *zn,
893 uint64_t *integer_size, uint64_t *num_integers)
897 zap_entry_handle_t zeh;
899 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
902 err = zap_leaf_lookup(l, zn, &zeh);
907 *integer_size = zeh.zeh_integer_size;
909 *num_integers = zeh.zeh_num_integers;
916 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
920 zap_entry_handle_t zeh;
922 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
925 err = zap_leaf_lookup(l, zn, &zeh);
927 zap_entry_remove(&zeh);
928 zap_increment_num_entries(zn->zn_zap, -1, tx);
935 fzap_prefetch(zap_name_t *zn)
938 zap_t *zap = zn->zn_zap;
941 idx = ZAP_HASH_IDX(zn->zn_hash,
942 zap_f_phys(zap)->zap_ptrtbl.zt_shift);
943 if (zap_idx_to_blk(zap, idx, &blk) != 0)
945 bs = FZAP_BLOCK_SHIFT(zap);
946 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs,
947 ZIO_PRIORITY_SYNC_READ);
951 * Helper functions for consumers.
955 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
956 const char *name, dmu_tx_t *tx)
960 VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
961 VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
968 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
978 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
979 for (zap_cursor_init(&zc, os, zapobj);
980 (err = zap_cursor_retrieve(&zc, za)) == 0;
981 zap_cursor_advance(&zc)) {
982 if ((za->za_first_integer & mask) == (value & mask)) {
983 (void) strcpy(name, za->za_name);
987 zap_cursor_fini(&zc);
988 kmem_free(za, sizeof (zap_attribute_t));
993 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
1000 for (zap_cursor_init(&zc, os, fromobj);
1001 zap_cursor_retrieve(&zc, &za) == 0;
1002 (void) zap_cursor_advance(&zc)) {
1003 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1004 err = SET_ERROR(EINVAL);
1007 err = zap_add(os, intoobj, za.za_name,
1008 8, 1, &za.za_first_integer, tx);
1012 zap_cursor_fini(&zc);
1017 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1018 uint64_t value, dmu_tx_t *tx)
1025 for (zap_cursor_init(&zc, os, fromobj);
1026 zap_cursor_retrieve(&zc, &za) == 0;
1027 (void) zap_cursor_advance(&zc)) {
1028 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1029 err = SET_ERROR(EINVAL);
1032 err = zap_add(os, intoobj, za.za_name,
1037 zap_cursor_fini(&zc);
1042 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1050 for (zap_cursor_init(&zc, os, fromobj);
1051 zap_cursor_retrieve(&zc, &za) == 0;
1052 (void) zap_cursor_advance(&zc)) {
1055 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1056 err = SET_ERROR(EINVAL);
1060 err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta);
1061 if (err != 0 && err != ENOENT)
1063 delta += za.za_first_integer;
1064 err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx);
1068 zap_cursor_fini(&zc);
1073 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1077 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1078 return (zap_add(os, obj, name, 8, 1, &value, tx));
1082 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1086 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1087 return (zap_remove(os, obj, name, tx));
1091 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1095 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1096 return (zap_lookup(os, obj, name, 8, 1, &value));
1100 zap_add_int_key(objset_t *os, uint64_t obj,
1101 uint64_t key, uint64_t value, dmu_tx_t *tx)
1105 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1106 return (zap_add(os, obj, name, 8, 1, &value, tx));
1110 zap_update_int_key(objset_t *os, uint64_t obj,
1111 uint64_t key, uint64_t value, dmu_tx_t *tx)
1115 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1116 return (zap_update(os, obj, name, 8, 1, &value, tx));
1120 zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1124 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1125 return (zap_lookup(os, obj, name, 8, 1, valuep));
1129 zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1138 err = zap_lookup(os, obj, name, 8, 1, &value);
1139 if (err != 0 && err != ENOENT)
1143 err = zap_remove(os, obj, name, tx);
1145 err = zap_update(os, obj, name, 8, 1, &value, tx);
1150 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1155 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1156 return (zap_increment(os, obj, name, delta, tx));
1160 * Routines for iterating over the attributes.
1164 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1167 zap_entry_handle_t zeh;
1170 /* retrieve the next entry at or after zc_hash/zc_cd */
1171 /* if no entry, return ENOENT */
1174 (ZAP_HASH_IDX(zc->zc_hash,
1175 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) !=
1176 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) {
1177 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1178 zap_put_leaf(zc->zc_leaf);
1183 if (zc->zc_leaf == NULL) {
1184 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1189 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1193 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1195 if (err == ENOENT) {
1197 (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1;
1198 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1200 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 ||
1202 zc->zc_hash = -1ULL;
1204 zap_put_leaf(zc->zc_leaf);
1211 zc->zc_hash = zeh.zeh_hash;
1212 zc->zc_cd = zeh.zeh_cd;
1213 za->za_integer_length = zeh.zeh_integer_size;
1214 za->za_num_integers = zeh.zeh_num_integers;
1215 if (zeh.zeh_num_integers == 0) {
1216 za->za_first_integer = 0;
1218 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1219 ASSERT(err == 0 || err == EOVERFLOW);
1221 err = zap_entry_read_name(zap, &zeh,
1222 sizeof (za->za_name), za->za_name);
1225 za->za_normalization_conflict =
1226 zap_entry_normalization_conflict(&zeh,
1227 NULL, za->za_name, zap);
1229 rw_exit(&zc->zc_leaf->l_rwlock);
1234 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1237 uint64_t lastblk = 0;
1240 * NB: if a leaf has more pointers than an entire ptrtbl block
1241 * can hold, then it'll be accounted for more than once, since
1242 * we won't have lastblk.
1244 for (i = 0; i < len; i++) {
1247 if (tbl[i] == lastblk)
1251 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1253 zap_leaf_stats(zap, l, zs);
1260 fzap_cursor_move_to_key(zap_cursor_t *zc, zap_name_t *zn)
1264 zap_entry_handle_t zeh;
1266 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
1267 return (SET_ERROR(ENAMETOOLONG));
1269 err = zap_deref_leaf(zc->zc_zap, zn->zn_hash, NULL, RW_READER, &l);
1273 err = zap_leaf_lookup(l, zn, &zeh);
1278 zc->zc_hash = zeh.zeh_hash;
1279 zc->zc_cd = zeh.zeh_cd;
1285 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1287 int bs = FZAP_BLOCK_SHIFT(zap);
1288 zs->zs_blocksize = 1ULL << bs;
1291 * Set zap_phys_t fields
1293 zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs;
1294 zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries;
1295 zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk;
1296 zs->zs_block_type = zap_f_phys(zap)->zap_block_type;
1297 zs->zs_magic = zap_f_phys(zap)->zap_magic;
1298 zs->zs_salt = zap_f_phys(zap)->zap_salt;
1301 * Set zap_ptrtbl fields
1303 zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1304 zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk;
1305 zs->zs_ptrtbl_blks_copied =
1306 zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied;
1307 zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk;
1308 zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1309 zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1311 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
1312 /* the ptrtbl is entirely in the header block. */
1313 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1314 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1318 dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
1319 zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs,
1320 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs,
1321 ZIO_PRIORITY_SYNC_READ);
1323 for (b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1328 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1329 (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs,
1330 FTAG, &db, DMU_READ_NO_PREFETCH);
1332 zap_stats_ptrtbl(zap, db->db_data,
1334 dmu_buf_rele(db, FTAG);
1341 fzap_count_write(zap_name_t *zn, int add, refcount_t *towrite,
1342 refcount_t *tooverwrite)
1344 zap_t *zap = zn->zn_zap;
1349 * Account for the header block of the fatzap.
1351 if (!add && dmu_buf_freeable(zap->zap_dbuf)) {
1352 (void) refcount_add_many(tooverwrite,
1353 zap->zap_dbuf->db_size, FTAG);
1355 (void) refcount_add_many(towrite,
1356 zap->zap_dbuf->db_size, FTAG);
1360 * Account for the pointer table blocks.
1361 * If we are adding we need to account for the following cases :
1362 * - If the pointer table is embedded, this operation could force an
1363 * external pointer table.
1364 * - If this already has an external pointer table this operation
1365 * could extend the table.
1368 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
1369 (void) refcount_add_many(towrite,
1370 zap->zap_dbuf->db_size, FTAG);
1372 (void) refcount_add_many(towrite,
1373 zap->zap_dbuf->db_size * 3, FTAG);
1378 * Now, check if the block containing leaf is freeable
1379 * and account accordingly.
1381 err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l);
1386 if (!add && dmu_buf_freeable(l->l_dbuf)) {
1387 (void) refcount_add_many(tooverwrite, l->l_dbuf->db_size, FTAG);
1390 * If this an add operation, the leaf block could split.
1391 * Hence, we need to account for an additional leaf block.
1393 (void) refcount_add_many(towrite,
1394 (add ? 2 : 1) * l->l_dbuf->db_size, FTAG);