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, 0, 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 (%lluk 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, 0, 0, 0);
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, 0, 0, 0);
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), ==,
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));
496 int bs = FZAP_BLOCK_SHIFT(zap);
497 dnode_t *dn = dmu_buf_dnode_enter(zap->zap_dbuf);
498 int err = dmu_buf_hold_by_dnode(dn,
499 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
500 dmu_buf_dnode_exit(zap->zap_dbuf);
504 ASSERT3U(db->db_object, ==, zap->zap_object);
505 ASSERT3U(db->db_offset, ==, blkid << bs);
506 ASSERT3U(db->db_size, ==, 1 << bs);
509 zap_leaf_t *l = dmu_buf_get_user(db);
512 l = zap_open_leaf(blkid, db);
514 rw_enter(&l->l_rwlock, lt);
516 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
517 * causing ASSERT below to fail.
520 dmu_buf_will_dirty(db, tx);
521 ASSERT3U(l->l_blkid, ==, blkid);
522 ASSERT3P(l->l_dbuf, ==, db);
523 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF);
524 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
531 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
533 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
535 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
537 (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift));
538 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
541 return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl,
547 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
550 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
552 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
553 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
556 return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl,
562 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
566 ASSERT(zap->zap_dbuf == NULL ||
567 zap_f_phys(zap) == zap->zap_dbuf->db_data);
569 /* Reality check for corrupt zap objects (leaf or header). */
570 if ((zap_f_phys(zap)->zap_block_type != ZBT_LEAF &&
571 zap_f_phys(zap)->zap_block_type != ZBT_HEADER) ||
572 zap_f_phys(zap)->zap_magic != ZAP_MAGIC) {
573 return (SET_ERROR(EIO));
576 uint64_t idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
577 int err = zap_idx_to_blk(zap, idx, &blk);
580 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
583 ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) ==
584 zap_leaf_phys(*lp)->l_hdr.lh_prefix);
589 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l,
590 void *tag, dmu_tx_t *tx, zap_leaf_t **lp)
592 zap_t *zap = zn->zn_zap;
593 uint64_t hash = zn->zn_hash;
595 int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len;
597 ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
598 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
600 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
601 zap_leaf_phys(l)->l_hdr.lh_prefix);
603 if (zap_tryupgradedir(zap, tx) == 0 ||
604 old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
605 /* We failed to upgrade, or need to grow the pointer table */
606 objset_t *os = zap->zap_objset;
607 uint64_t object = zap->zap_object;
610 zap_unlockdir(zap, tag);
611 err = zap_lockdir(os, object, tx, RW_WRITER,
612 FALSE, FALSE, tag, &zn->zn_zap);
616 ASSERT(!zap->zap_ismicro);
618 while (old_prefix_len ==
619 zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
620 err = zap_grow_ptrtbl(zap, tx);
625 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
629 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) {
630 /* it split while our locks were down */
635 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
636 ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
637 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
638 zap_leaf_phys(l)->l_hdr.lh_prefix);
640 int prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
641 (old_prefix_len + 1);
643 (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
645 /* check for i/o errors before doing zap_leaf_split */
646 for (int i = 0; i < (1ULL << prefix_diff); i++) {
648 err = zap_idx_to_blk(zap, sibling + i, &blk);
651 ASSERT3U(blk, ==, l->l_blkid);
654 zap_leaf_t *nl = zap_create_leaf(zap, tx);
655 zap_leaf_split(l, nl, zap->zap_normflags != 0);
657 /* set sibling pointers */
658 for (int i = 0; i < (1ULL << prefix_diff); i++) {
659 err = zap_set_idx_to_blk(zap, sibling + i, nl->l_blkid, tx);
660 ASSERT0(err); /* we checked for i/o errors above */
663 if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) {
664 /* we want the sibling */
676 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l,
677 void *tag, dmu_tx_t *tx)
679 zap_t *zap = zn->zn_zap;
680 int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
681 int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift &&
682 zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
686 if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) {
688 * We are in the middle of growing the pointer table, or
689 * this leaf will soon make us grow it.
691 if (zap_tryupgradedir(zap, tx) == 0) {
692 objset_t *os = zap->zap_objset;
693 uint64_t zapobj = zap->zap_object;
695 zap_unlockdir(zap, tag);
696 int err = zap_lockdir(os, zapobj, tx,
697 RW_WRITER, FALSE, FALSE, tag, &zn->zn_zap);
703 /* could have finished growing while our locks were down */
704 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift)
705 (void) zap_grow_ptrtbl(zap, tx);
710 fzap_checkname(zap_name_t *zn)
712 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
713 return (SET_ERROR(ENAMETOOLONG));
718 fzap_checksize(uint64_t integer_size, uint64_t num_integers)
720 /* Only integer sizes supported by C */
721 switch (integer_size) {
728 return (SET_ERROR(EINVAL));
731 if (integer_size * num_integers > ZAP_MAXVALUELEN)
738 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
740 int err = fzap_checkname(zn);
743 return (fzap_checksize(integer_size, num_integers));
747 * Routines for manipulating attributes.
750 fzap_lookup(zap_name_t *zn,
751 uint64_t integer_size, uint64_t num_integers, void *buf,
752 char *realname, int rn_len, boolean_t *ncp)
755 zap_entry_handle_t zeh;
757 int err = fzap_checkname(zn);
761 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
764 err = zap_leaf_lookup(l, zn, &zeh);
766 if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
771 err = zap_entry_read(&zeh, integer_size, num_integers, buf);
772 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
774 *ncp = zap_entry_normalization_conflict(&zeh,
775 zn, NULL, zn->zn_zap);
784 fzap_add_cd(zap_name_t *zn,
785 uint64_t integer_size, uint64_t num_integers,
786 const void *val, uint32_t cd, void *tag, dmu_tx_t *tx)
790 zap_entry_handle_t zeh;
791 zap_t *zap = zn->zn_zap;
793 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
794 ASSERT(!zap->zap_ismicro);
795 ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
797 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
801 err = zap_leaf_lookup(l, zn, &zeh);
803 err = SET_ERROR(EEXIST);
809 err = zap_entry_create(l, zn, cd,
810 integer_size, num_integers, val, &zeh);
813 zap_increment_num_entries(zap, 1, tx);
814 } else if (err == EAGAIN) {
815 err = zap_expand_leaf(zn, l, tag, tx, &l);
816 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
823 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
828 fzap_add(zap_name_t *zn,
829 uint64_t integer_size, uint64_t num_integers,
830 const void *val, void *tag, dmu_tx_t *tx)
832 int err = fzap_check(zn, integer_size, num_integers);
836 return (fzap_add_cd(zn, integer_size, num_integers,
837 val, ZAP_NEED_CD, tag, tx));
841 fzap_update(zap_name_t *zn,
842 int integer_size, uint64_t num_integers, const void *val,
843 void *tag, dmu_tx_t *tx)
848 zap_entry_handle_t zeh;
849 zap_t *zap = zn->zn_zap;
851 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
852 err = fzap_check(zn, integer_size, num_integers);
856 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
860 err = zap_leaf_lookup(l, zn, &zeh);
861 create = (err == ENOENT);
862 ASSERT(err == 0 || err == ENOENT);
865 err = zap_entry_create(l, zn, ZAP_NEED_CD,
866 integer_size, num_integers, val, &zeh);
868 zap_increment_num_entries(zap, 1, tx);
870 err = zap_entry_update(&zeh, integer_size, num_integers, val);
874 err = zap_expand_leaf(zn, l, tag, tx, &l);
875 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
881 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
886 fzap_length(zap_name_t *zn,
887 uint64_t *integer_size, uint64_t *num_integers)
891 zap_entry_handle_t zeh;
893 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
896 err = zap_leaf_lookup(l, zn, &zeh);
900 if (integer_size != 0)
901 *integer_size = zeh.zeh_integer_size;
902 if (num_integers != 0)
903 *num_integers = zeh.zeh_num_integers;
910 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
914 zap_entry_handle_t zeh;
916 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
919 err = zap_leaf_lookup(l, zn, &zeh);
921 zap_entry_remove(&zeh);
922 zap_increment_num_entries(zn->zn_zap, -1, tx);
929 fzap_prefetch(zap_name_t *zn)
932 zap_t *zap = zn->zn_zap;
934 uint64_t idx = ZAP_HASH_IDX(zn->zn_hash,
935 zap_f_phys(zap)->zap_ptrtbl.zt_shift);
936 if (zap_idx_to_blk(zap, idx, &blk) != 0)
938 int bs = FZAP_BLOCK_SHIFT(zap);
939 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs,
940 ZIO_PRIORITY_SYNC_READ);
944 * Helper functions for consumers.
948 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
949 const char *name, dmu_tx_t *tx)
951 return (zap_create_link_dnsize(os, ot, parent_obj, name, 0, tx));
955 zap_create_link_dnsize(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
956 const char *name, int dnodesize, dmu_tx_t *tx)
960 VERIFY((new_obj = zap_create_dnsize(os, ot, DMU_OT_NONE, 0,
961 dnodesize, tx)) > 0);
962 VERIFY0(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
969 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
978 zap_attribute_t *za = kmem_alloc(sizeof (*za), 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 (*za));
993 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
998 zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP);
999 for (zap_cursor_init(&zc, os, fromobj);
1000 zap_cursor_retrieve(&zc, za) == 0;
1001 (void) zap_cursor_advance(&zc)) {
1002 if (za->za_integer_length != 8 || za->za_num_integers != 1) {
1003 err = SET_ERROR(EINVAL);
1006 err = zap_add(os, intoobj, za->za_name,
1007 8, 1, &za->za_first_integer, tx);
1011 zap_cursor_fini(&zc);
1012 kmem_free(za, sizeof (*za));
1017 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1018 uint64_t value, dmu_tx_t *tx)
1023 zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP);
1024 for (zap_cursor_init(&zc, os, fromobj);
1025 zap_cursor_retrieve(&zc, za) == 0;
1026 (void) zap_cursor_advance(&zc)) {
1027 if (za->za_integer_length != 8 || za->za_num_integers != 1) {
1028 err = SET_ERROR(EINVAL);
1031 err = zap_add(os, intoobj, za->za_name,
1036 zap_cursor_fini(&zc);
1037 kmem_free(za, sizeof (*za));
1042 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1048 zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP);
1049 for (zap_cursor_init(&zc, os, fromobj);
1050 zap_cursor_retrieve(&zc, za) == 0;
1051 (void) zap_cursor_advance(&zc)) {
1054 if (za->za_integer_length != 8 || za->za_num_integers != 1) {
1055 err = SET_ERROR(EINVAL);
1059 err = zap_lookup(os, intoobj, za->za_name, 8, 1, &delta);
1060 if (err != 0 && err != ENOENT)
1062 delta += za->za_first_integer;
1063 err = zap_update(os, intoobj, za->za_name, 8, 1, &delta, tx);
1067 zap_cursor_fini(&zc);
1068 kmem_free(za, sizeof (*za));
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,
1137 int err = zap_lookup(os, obj, name, 8, 1, &value);
1138 if (err != 0 && err != ENOENT)
1142 err = zap_remove(os, obj, name, tx);
1144 err = zap_update(os, obj, name, 8, 1, &value, tx);
1149 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1154 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1155 return (zap_increment(os, obj, name, delta, tx));
1159 * Routines for iterating over the attributes.
1163 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1166 zap_entry_handle_t zeh;
1169 /* retrieve the next entry at or after zc_hash/zc_cd */
1170 /* if no entry, return ENOENT */
1173 (ZAP_HASH_IDX(zc->zc_hash,
1174 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) !=
1175 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) {
1176 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1177 zap_put_leaf(zc->zc_leaf);
1182 if (zc->zc_leaf == NULL) {
1183 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1188 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1192 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1194 if (err == ENOENT) {
1196 (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1;
1197 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1199 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 ||
1201 zc->zc_hash = -1ULL;
1203 zap_put_leaf(zc->zc_leaf);
1210 zc->zc_hash = zeh.zeh_hash;
1211 zc->zc_cd = zeh.zeh_cd;
1212 za->za_integer_length = zeh.zeh_integer_size;
1213 za->za_num_integers = zeh.zeh_num_integers;
1214 if (zeh.zeh_num_integers == 0) {
1215 za->za_first_integer = 0;
1217 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1218 ASSERT(err == 0 || err == EOVERFLOW);
1220 err = zap_entry_read_name(zap, &zeh,
1221 sizeof (za->za_name), za->za_name);
1224 za->za_normalization_conflict =
1225 zap_entry_normalization_conflict(&zeh,
1226 NULL, za->za_name, zap);
1228 rw_exit(&zc->zc_leaf->l_rwlock);
1233 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1235 uint64_t lastblk = 0;
1238 * NB: if a leaf has more pointers than an entire ptrtbl block
1239 * can hold, then it'll be accounted for more than once, since
1240 * we won't have lastblk.
1242 for (int i = 0; i < len; i++) {
1245 if (tbl[i] == lastblk)
1249 int err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1251 zap_leaf_stats(zap, l, zs);
1258 fzap_cursor_move_to_key(zap_cursor_t *zc, zap_name_t *zn)
1262 zap_entry_handle_t zeh;
1264 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
1265 return (SET_ERROR(ENAMETOOLONG));
1267 err = zap_deref_leaf(zc->zc_zap, zn->zn_hash, NULL, RW_READER, &l);
1271 err = zap_leaf_lookup(l, zn, &zeh);
1276 zc->zc_hash = zeh.zeh_hash;
1277 zc->zc_cd = zeh.zeh_cd;
1283 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1285 int bs = FZAP_BLOCK_SHIFT(zap);
1286 zs->zs_blocksize = 1ULL << bs;
1289 * Set zap_phys_t fields
1291 zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs;
1292 zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries;
1293 zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk;
1294 zs->zs_block_type = zap_f_phys(zap)->zap_block_type;
1295 zs->zs_magic = zap_f_phys(zap)->zap_magic;
1296 zs->zs_salt = zap_f_phys(zap)->zap_salt;
1299 * Set zap_ptrtbl fields
1301 zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1302 zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk;
1303 zs->zs_ptrtbl_blks_copied =
1304 zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied;
1305 zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk;
1306 zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1307 zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1309 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
1310 /* the ptrtbl is entirely in the header block. */
1311 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1312 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1314 dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
1315 zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs,
1316 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs,
1317 ZIO_PRIORITY_SYNC_READ);
1319 for (int b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1324 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1325 (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs,
1326 FTAG, &db, DMU_READ_NO_PREFETCH);
1328 zap_stats_ptrtbl(zap, db->db_data,
1330 dmu_buf_rele(db, FTAG);