1/*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 */
26
27/*
28 * This file contains the top half of the zfs directory structure
29 * implementation. The bottom half is in zap_leaf.c.
30 *
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.
35 *
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
40 */
41
42#include <sys/spa.h>
43#include <sys/dmu.h>
44#include <sys/zfs_context.h>
45#include <sys/zfs_znode.h>
46#include <sys/fs/zfs.h>
47#include <sys/zap.h>
48#include <sys/refcount.h>
49#include <sys/zap_impl.h>
50#include <sys/zap_leaf.h>
51
52/*
53 * If zap_iterate_prefetch is set, we will prefetch the entire ZAP object
54 * (all leaf blocks) when we start iterating over it.
55 *
56 * For zap_cursor_init(), the callers all intend to iterate through all the
57 * entries.  There are a few cases where an error (typically i/o error) could
58 * cause it to bail out early.
59 *
60 * For zap_cursor_init_serialized(), there are callers that do the iteration
61 * outside of ZFS.  Typically they would iterate over everything, but we
62 * don't have control of that.  E.g. zfs_ioc_snapshot_list_next(),
63 * zcp_snapshots_iter(), and other iterators over things in the MOS - these
64 * are called by /sbin/zfs and channel programs.  The other example is
65 * zfs_readdir() which iterates over directory entries for the getdents()
66 * syscall.  /sbin/ls iterates to the end (unless it receives a signal), but
67 * userland doesn't have to.
68 *
69 * Given that the ZAP entries aren't returned in a specific order, the only
70 * legitimate use cases for partial iteration would be:
71 *
72 * 1. Pagination: e.g. you only want to display 100 entries at a time, so you
73 *    get the first 100 and then wait for the user to hit "next page", which
74 *    they may never do).
75 *
76 * 2. You want to know if there are more than X entries, without relying on
77 *    the zfs-specific implementation of the directory's st_size (which is
78 *    the number of entries).
79 */
80boolean_t zap_iterate_prefetch = B_TRUE;
81
82int fzap_default_block_shift = 14; /* 16k blocksize */
83
84extern inline zap_phys_t *zap_f_phys(zap_t *zap);
85
86static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
87
88void
89fzap_byteswap(void *vbuf, size_t size)
90{
91	uint64_t block_type = *(uint64_t *)vbuf;
92
93	if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF))
94		zap_leaf_byteswap(vbuf, size);
95	else {
96		/* it's a ptrtbl block */
97		byteswap_uint64_array(vbuf, size);
98	}
99}
100
101void
102fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags)
103{
104	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
105	zap->zap_ismicro = FALSE;
106
107	zap->zap_dbu.dbu_evict_func_sync = zap_evict_sync;
108	zap->zap_dbu.dbu_evict_func_async = NULL;
109
110	mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
111	zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1;
112
113	zap_phys_t *zp = zap_f_phys(zap);
114	/*
115	 * explicitly zero it since it might be coming from an
116	 * initialized microzap
117	 */
118	bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
119	zp->zap_block_type = ZBT_HEADER;
120	zp->zap_magic = ZAP_MAGIC;
121
122	zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);
123
124	zp->zap_freeblk = 2;		/* block 1 will be the first leaf */
125	zp->zap_num_leafs = 1;
126	zp->zap_num_entries = 0;
127	zp->zap_salt = zap->zap_salt;
128	zp->zap_normflags = zap->zap_normflags;
129	zp->zap_flags = flags;
130
131	/* block 1 will be the first leaf */
132	for (int i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
133		ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;
134
135	/*
136	 * set up block 1 - the first leaf
137	 */
138	dmu_buf_t *db;
139	VERIFY0(dmu_buf_hold(zap->zap_objset, zap->zap_object,
140	    1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
141	dmu_buf_will_dirty(db, tx);
142
143	zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
144	l->l_dbuf = db;
145
146	zap_leaf_init(l, zp->zap_normflags != 0);
147
148	kmem_free(l, sizeof (zap_leaf_t));
149	dmu_buf_rele(db, FTAG);
150}
151
152static int
153zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
154{
155	if (RW_WRITE_HELD(&zap->zap_rwlock))
156		return (1);
157	if (rw_tryupgrade(&zap->zap_rwlock)) {
158		dmu_buf_will_dirty(zap->zap_dbuf, tx);
159		return (1);
160	}
161	return (0);
162}
163
164/*
165 * Generic routines for dealing with the pointer & cookie tables.
166 */
167
168static int
169zap_table_grow(zap_t *zap, zap_table_phys_t *tbl,
170    void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n),
171    dmu_tx_t *tx)
172{
173	uint64_t newblk;
174	int bs = FZAP_BLOCK_SHIFT(zap);
175	int hepb = 1<<(bs-4);
176	/* hepb = half the number of entries in a block */
177
178	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
179	ASSERT(tbl->zt_blk != 0);
180	ASSERT(tbl->zt_numblks > 0);
181
182	if (tbl->zt_nextblk != 0) {
183		newblk = tbl->zt_nextblk;
184	} else {
185		newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2);
186		tbl->zt_nextblk = newblk;
187		ASSERT0(tbl->zt_blks_copied);
188		dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
189		    tbl->zt_blk << bs, tbl->zt_numblks << bs,
190		    ZIO_PRIORITY_SYNC_READ);
191	}
192
193	/*
194	 * Copy the ptrtbl from the old to new location.
195	 */
196
197	uint64_t b = tbl->zt_blks_copied;
198	dmu_buf_t *db_old;
199	int err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
200	    (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH);
201	if (err != 0)
202		return (err);
203
204	/* first half of entries in old[b] go to new[2*b+0] */
205	dmu_buf_t *db_new;
206	VERIFY0(dmu_buf_hold(zap->zap_objset, zap->zap_object,
207	    (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
208	dmu_buf_will_dirty(db_new, tx);
209	transfer_func(db_old->db_data, db_new->db_data, hepb);
210	dmu_buf_rele(db_new, FTAG);
211
212	/* second half of entries in old[b] go to new[2*b+1] */
213	VERIFY0(dmu_buf_hold(zap->zap_objset, zap->zap_object,
214	    (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
215	dmu_buf_will_dirty(db_new, tx);
216	transfer_func((uint64_t *)db_old->db_data + hepb,
217	    db_new->db_data, hepb);
218	dmu_buf_rele(db_new, FTAG);
219
220	dmu_buf_rele(db_old, FTAG);
221
222	tbl->zt_blks_copied++;
223
224	dprintf("copied block %llu of %llu\n",
225	    tbl->zt_blks_copied, tbl->zt_numblks);
226
227	if (tbl->zt_blks_copied == tbl->zt_numblks) {
228		(void) dmu_free_range(zap->zap_objset, zap->zap_object,
229		    tbl->zt_blk << bs, tbl->zt_numblks << bs, tx);
230
231		tbl->zt_blk = newblk;
232		tbl->zt_numblks *= 2;
233		tbl->zt_shift++;
234		tbl->zt_nextblk = 0;
235		tbl->zt_blks_copied = 0;
236
237		dprintf("finished; numblocks now %llu (%lluk entries)\n",
238		    tbl->zt_numblks, 1<<(tbl->zt_shift-10));
239	}
240
241	return (0);
242}
243
244static int
245zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val,
246    dmu_tx_t *tx)
247{
248	int bs = FZAP_BLOCK_SHIFT(zap);
249
250	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
251	ASSERT(tbl->zt_blk != 0);
252
253	dprintf("storing %llx at index %llx\n", val, idx);
254
255	uint64_t blk = idx >> (bs-3);
256	uint64_t off = idx & ((1<<(bs-3))-1);
257
258	dmu_buf_t *db;
259	int err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
260	    (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
261	if (err != 0)
262		return (err);
263	dmu_buf_will_dirty(db, tx);
264
265	if (tbl->zt_nextblk != 0) {
266		uint64_t idx2 = idx * 2;
267		uint64_t blk2 = idx2 >> (bs-3);
268		uint64_t off2 = idx2 & ((1<<(bs-3))-1);
269		dmu_buf_t *db2;
270
271		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
272		    (tbl->zt_nextblk + blk2) << bs, FTAG, &db2,
273		    DMU_READ_NO_PREFETCH);
274		if (err != 0) {
275			dmu_buf_rele(db, FTAG);
276			return (err);
277		}
278		dmu_buf_will_dirty(db2, tx);
279		((uint64_t *)db2->db_data)[off2] = val;
280		((uint64_t *)db2->db_data)[off2+1] = val;
281		dmu_buf_rele(db2, FTAG);
282	}
283
284	((uint64_t *)db->db_data)[off] = val;
285	dmu_buf_rele(db, FTAG);
286
287	return (0);
288}
289
290static int
291zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
292{
293	int bs = FZAP_BLOCK_SHIFT(zap);
294
295	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
296
297	uint64_t blk = idx >> (bs-3);
298	uint64_t off = idx & ((1<<(bs-3))-1);
299
300	/*
301	 * Note: this is equivalent to dmu_buf_hold(), but we use
302	 * _dnode_enter / _by_dnode because it's faster because we don't
303	 * have to hold the dnode.
304	 */
305	dnode_t *dn = dmu_buf_dnode_enter(zap->zap_dbuf);
306	dmu_buf_t *db;
307	int err = dmu_buf_hold_by_dnode(dn,
308	    (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
309	dmu_buf_dnode_exit(zap->zap_dbuf);
310	if (err != 0)
311		return (err);
312	*valp = ((uint64_t *)db->db_data)[off];
313	dmu_buf_rele(db, FTAG);
314
315	if (tbl->zt_nextblk != 0) {
316		/*
317		 * read the nextblk for the sake of i/o error checking,
318		 * so that zap_table_load() will catch errors for
319		 * zap_table_store.
320		 */
321		blk = (idx*2) >> (bs-3);
322
323		dn = dmu_buf_dnode_enter(zap->zap_dbuf);
324		err = dmu_buf_hold_by_dnode(dn,
325		    (tbl->zt_nextblk + blk) << bs, FTAG, &db,
326		    DMU_READ_NO_PREFETCH);
327		dmu_buf_dnode_exit(zap->zap_dbuf);
328		if (err == 0)
329			dmu_buf_rele(db, FTAG);
330	}
331	return (err);
332}
333
334/*
335 * Routines for growing the ptrtbl.
336 */
337
338static void
339zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n)
340{
341	for (int i = 0; i < n; i++) {
342		uint64_t lb = src[i];
343		dst[2 * i + 0] = lb;
344		dst[2 * i + 1] = lb;
345	}
346}
347
348static int
349zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
350{
351	/*
352	 * The pointer table should never use more hash bits than we
353	 * have (otherwise we'd be using useless zero bits to index it).
354	 * If we are within 2 bits of running out, stop growing, since
355	 * this is already an aberrant condition.
356	 */
357	if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2)
358		return (SET_ERROR(ENOSPC));
359
360	if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
361		/*
362		 * We are outgrowing the "embedded" ptrtbl (the one
363		 * stored in the header block).  Give it its own entire
364		 * block, which will double the size of the ptrtbl.
365		 */
366		ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
367		    ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
368		ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk);
369
370		uint64_t newblk = zap_allocate_blocks(zap, 1);
371		dmu_buf_t *db_new;
372		int err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
373		    newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new,
374		    DMU_READ_NO_PREFETCH);
375		if (err != 0)
376			return (err);
377		dmu_buf_will_dirty(db_new, tx);
378		zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
379		    db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
380		dmu_buf_rele(db_new, FTAG);
381
382		zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk;
383		zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1;
384		zap_f_phys(zap)->zap_ptrtbl.zt_shift++;
385
386		ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
387		    zap_f_phys(zap)->zap_ptrtbl.zt_numblks <<
388		    (FZAP_BLOCK_SHIFT(zap)-3));
389
390		return (0);
391	} else {
392		return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl,
393		    zap_ptrtbl_transfer, tx));
394	}
395}
396
397static void
398zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
399{
400	dmu_buf_will_dirty(zap->zap_dbuf, tx);
401	mutex_enter(&zap->zap_f.zap_num_entries_mtx);
402	ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta);
403	zap_f_phys(zap)->zap_num_entries += delta;
404	mutex_exit(&zap->zap_f.zap_num_entries_mtx);
405}
406
407static uint64_t
408zap_allocate_blocks(zap_t *zap, int nblocks)
409{
410	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
411	uint64_t newblk = zap_f_phys(zap)->zap_freeblk;
412	zap_f_phys(zap)->zap_freeblk += nblocks;
413	return (newblk);
414}
415
416static void
417zap_leaf_evict_sync(void *dbu)
418{
419	zap_leaf_t *l = dbu;
420
421	rw_destroy(&l->l_rwlock);
422	kmem_free(l, sizeof (zap_leaf_t));
423}
424
425static zap_leaf_t *
426zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
427{
428	zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
429
430	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
431
432	rw_init(&l->l_rwlock, 0, 0, 0);
433	rw_enter(&l->l_rwlock, RW_WRITER);
434	l->l_blkid = zap_allocate_blocks(zap, 1);
435	l->l_dbuf = NULL;
436
437	VERIFY0(dmu_buf_hold(zap->zap_objset, zap->zap_object,
438	    l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
439	    DMU_READ_NO_PREFETCH));
440	dmu_buf_init_user(&l->l_dbu, zap_leaf_evict_sync, NULL, &l->l_dbuf);
441	VERIFY3P(NULL, ==, dmu_buf_set_user(l->l_dbuf, &l->l_dbu));
442	dmu_buf_will_dirty(l->l_dbuf, tx);
443
444	zap_leaf_init(l, zap->zap_normflags != 0);
445
446	zap_f_phys(zap)->zap_num_leafs++;
447
448	return (l);
449}
450
451int
452fzap_count(zap_t *zap, uint64_t *count)
453{
454	ASSERT(!zap->zap_ismicro);
455	mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
456	*count = zap_f_phys(zap)->zap_num_entries;
457	mutex_exit(&zap->zap_f.zap_num_entries_mtx);
458	return (0);
459}
460
461/*
462 * Routines for obtaining zap_leaf_t's
463 */
464
465void
466zap_put_leaf(zap_leaf_t *l)
467{
468	rw_exit(&l->l_rwlock);
469	dmu_buf_rele(l->l_dbuf, NULL);
470}
471
472static zap_leaf_t *
473zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
474{
475	ASSERT(blkid != 0);
476
477	zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
478	rw_init(&l->l_rwlock, 0, 0, 0);
479	rw_enter(&l->l_rwlock, RW_WRITER);
480	l->l_blkid = blkid;
481	l->l_bs = highbit64(db->db_size) - 1;
482	l->l_dbuf = db;
483
484	dmu_buf_init_user(&l->l_dbu, zap_leaf_evict_sync, NULL, &l->l_dbuf);
485	zap_leaf_t *winner = dmu_buf_set_user(db, &l->l_dbu);
486
487	rw_exit(&l->l_rwlock);
488	if (winner != NULL) {
489		/* someone else set it first */
490		zap_leaf_evict_sync(&l->l_dbu);
491		l = winner;
492	}
493
494	/*
495	 * lhr_pad was previously used for the next leaf in the leaf
496	 * chain.  There should be no chained leafs (as we have removed
497	 * support for them).
498	 */
499	ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1);
500
501	/*
502	 * There should be more hash entries than there can be
503	 * chunks to put in the hash table
504	 */
505	ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
506
507	/* The chunks should begin at the end of the hash table */
508	ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
509	    &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
510
511	/* The chunks should end at the end of the block */
512	ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
513	    (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size);
514
515	return (l);
516}
517
518static int
519zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
520    zap_leaf_t **lp)
521{
522	dmu_buf_t *db;
523
524	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
525
526	int bs = FZAP_BLOCK_SHIFT(zap);
527	dnode_t *dn = dmu_buf_dnode_enter(zap->zap_dbuf);
528	int err = dmu_buf_hold_by_dnode(dn,
529	    blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
530	dmu_buf_dnode_exit(zap->zap_dbuf);
531	if (err != 0)
532		return (err);
533
534	ASSERT3U(db->db_object, ==, zap->zap_object);
535	ASSERT3U(db->db_offset, ==, blkid << bs);
536	ASSERT3U(db->db_size, ==, 1 << bs);
537	ASSERT(blkid != 0);
538
539	zap_leaf_t *l = dmu_buf_get_user(db);
540
541	if (l == NULL)
542		l = zap_open_leaf(blkid, db);
543
544	rw_enter(&l->l_rwlock, lt);
545	/*
546	 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
547	 * causing ASSERT below to fail.
548	 */
549	if (lt == RW_WRITER)
550		dmu_buf_will_dirty(db, tx);
551	ASSERT3U(l->l_blkid, ==, blkid);
552	ASSERT3P(l->l_dbuf, ==, db);
553	ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF);
554	ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
555
556	*lp = l;
557	return (0);
558}
559
560static int
561zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
562{
563	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
564
565	if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
566		ASSERT3U(idx, <,
567		    (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift));
568		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
569		return (0);
570	} else {
571		return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl,
572		    idx, valp));
573	}
574}
575
576static int
577zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
578{
579	ASSERT(tx != NULL);
580	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
581
582	if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
583		ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
584		return (0);
585	} else {
586		return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl,
587		    idx, blk, tx));
588	}
589}
590
591static int
592zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
593{
594	uint64_t blk;
595
596	ASSERT(zap->zap_dbuf == NULL ||
597	    zap_f_phys(zap) == zap->zap_dbuf->db_data);
598
599	/* Reality check for corrupt zap objects (leaf or header). */
600	if ((zap_f_phys(zap)->zap_block_type != ZBT_LEAF &&
601	    zap_f_phys(zap)->zap_block_type != ZBT_HEADER) ||
602	    zap_f_phys(zap)->zap_magic != ZAP_MAGIC) {
603		return (SET_ERROR(EIO));
604	}
605
606	uint64_t idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
607	int err = zap_idx_to_blk(zap, idx, &blk);
608	if (err != 0)
609		return (err);
610	err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
611
612	ASSERT(err ||
613	    ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) ==
614	    zap_leaf_phys(*lp)->l_hdr.lh_prefix);
615	return (err);
616}
617
618static int
619zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l,
620    void *tag, dmu_tx_t *tx, zap_leaf_t **lp)
621{
622	zap_t *zap = zn->zn_zap;
623	uint64_t hash = zn->zn_hash;
624	int err;
625	int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len;
626
627	ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
628	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
629
630	ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
631	    zap_leaf_phys(l)->l_hdr.lh_prefix);
632
633	if (zap_tryupgradedir(zap, tx) == 0 ||
634	    old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
635		/* We failed to upgrade, or need to grow the pointer table */
636		objset_t *os = zap->zap_objset;
637		uint64_t object = zap->zap_object;
638
639		zap_put_leaf(l);
640		zap_unlockdir(zap, tag);
641		err = zap_lockdir(os, object, tx, RW_WRITER,
642		    FALSE, FALSE, tag, &zn->zn_zap);
643		zap = zn->zn_zap;
644		if (err != 0)
645			return (err);
646		ASSERT(!zap->zap_ismicro);
647
648		while (old_prefix_len ==
649		    zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
650			err = zap_grow_ptrtbl(zap, tx);
651			if (err != 0)
652				return (err);
653		}
654
655		err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
656		if (err != 0)
657			return (err);
658
659		if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) {
660			/* it split while our locks were down */
661			*lp = l;
662			return (0);
663		}
664	}
665	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
666	ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
667	ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
668	    zap_leaf_phys(l)->l_hdr.lh_prefix);
669
670	int prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
671	    (old_prefix_len + 1);
672	uint64_t sibling =
673	    (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
674
675	/* check for i/o errors before doing zap_leaf_split */
676	for (int i = 0; i < (1ULL << prefix_diff); i++) {
677		uint64_t blk;
678		err = zap_idx_to_blk(zap, sibling + i, &blk);
679		if (err != 0)
680			return (err);
681		ASSERT3U(blk, ==, l->l_blkid);
682	}
683
684	zap_leaf_t *nl = zap_create_leaf(zap, tx);
685	zap_leaf_split(l, nl, zap->zap_normflags != 0);
686
687	/* set sibling pointers */
688	for (int i = 0; i < (1ULL << prefix_diff); i++) {
689		err = zap_set_idx_to_blk(zap, sibling + i, nl->l_blkid, tx);
690		ASSERT0(err); /* we checked for i/o errors above */
691	}
692
693	if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) {
694		/* we want the sibling */
695		zap_put_leaf(l);
696		*lp = nl;
697	} else {
698		zap_put_leaf(nl);
699		*lp = l;
700	}
701
702	return (0);
703}
704
705static void
706zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l,
707    void *tag, dmu_tx_t *tx)
708{
709	zap_t *zap = zn->zn_zap;
710	int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
711	int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift &&
712	    zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
713
714	zap_put_leaf(l);
715
716	if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) {
717		/*
718		 * We are in the middle of growing the pointer table, or
719		 * this leaf will soon make us grow it.
720		 */
721		if (zap_tryupgradedir(zap, tx) == 0) {
722			objset_t *os = zap->zap_objset;
723			uint64_t zapobj = zap->zap_object;
724
725			zap_unlockdir(zap, tag);
726			int err = zap_lockdir(os, zapobj, tx,
727			    RW_WRITER, FALSE, FALSE, tag, &zn->zn_zap);
728			zap = zn->zn_zap;
729			if (err != 0)
730				return;
731		}
732
733		/* could have finished growing while our locks were down */
734		if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift)
735			(void) zap_grow_ptrtbl(zap, tx);
736	}
737}
738
739static int
740fzap_checkname(zap_name_t *zn)
741{
742	if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
743		return (SET_ERROR(ENAMETOOLONG));
744	return (0);
745}
746
747static int
748fzap_checksize(uint64_t integer_size, uint64_t num_integers)
749{
750	/* Only integer sizes supported by C */
751	switch (integer_size) {
752	case 1:
753	case 2:
754	case 4:
755	case 8:
756		break;
757	default:
758		return (SET_ERROR(EINVAL));
759	}
760
761	if (integer_size * num_integers > ZAP_MAXVALUELEN)
762		return (E2BIG);
763
764	return (0);
765}
766
767static int
768fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
769{
770	int err = fzap_checkname(zn);
771	if (err != 0)
772		return (err);
773	return (fzap_checksize(integer_size, num_integers));
774}
775
776/*
777 * Routines for manipulating attributes.
778 */
779int
780fzap_lookup(zap_name_t *zn,
781    uint64_t integer_size, uint64_t num_integers, void *buf,
782    char *realname, int rn_len, boolean_t *ncp)
783{
784	zap_leaf_t *l;
785	zap_entry_handle_t zeh;
786
787	int err = fzap_checkname(zn);
788	if (err != 0)
789		return (err);
790
791	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
792	if (err != 0)
793		return (err);
794	err = zap_leaf_lookup(l, zn, &zeh);
795	if (err == 0) {
796		if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
797			zap_put_leaf(l);
798			return (err);
799		}
800
801		err = zap_entry_read(&zeh, integer_size, num_integers, buf);
802		(void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
803		if (ncp) {
804			*ncp = zap_entry_normalization_conflict(&zeh,
805			    zn, NULL, zn->zn_zap);
806		}
807	}
808
809	zap_put_leaf(l);
810	return (err);
811}
812
813int
814fzap_add_cd(zap_name_t *zn,
815    uint64_t integer_size, uint64_t num_integers,
816    const void *val, uint32_t cd, void *tag, dmu_tx_t *tx)
817{
818	zap_leaf_t *l;
819	int err;
820	zap_entry_handle_t zeh;
821	zap_t *zap = zn->zn_zap;
822
823	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
824	ASSERT(!zap->zap_ismicro);
825	ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
826
827	err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
828	if (err != 0)
829		return (err);
830retry:
831	err = zap_leaf_lookup(l, zn, &zeh);
832	if (err == 0) {
833		err = SET_ERROR(EEXIST);
834		goto out;
835	}
836	if (err != ENOENT)
837		goto out;
838
839	err = zap_entry_create(l, zn, cd,
840	    integer_size, num_integers, val, &zeh);
841
842	if (err == 0) {
843		zap_increment_num_entries(zap, 1, tx);
844	} else if (err == EAGAIN) {
845		err = zap_expand_leaf(zn, l, tag, tx, &l);
846		zap = zn->zn_zap;	/* zap_expand_leaf() may change zap */
847		if (err == 0)
848			goto retry;
849	}
850
851out:
852	if (zap != NULL)
853		zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
854	return (err);
855}
856
857int
858fzap_add(zap_name_t *zn,
859    uint64_t integer_size, uint64_t num_integers,
860    const void *val, void *tag, dmu_tx_t *tx)
861{
862	int err = fzap_check(zn, integer_size, num_integers);
863	if (err != 0)
864		return (err);
865
866	return (fzap_add_cd(zn, integer_size, num_integers,
867	    val, ZAP_NEED_CD, tag, tx));
868}
869
870int
871fzap_update(zap_name_t *zn,
872    int integer_size, uint64_t num_integers, const void *val,
873    void *tag, dmu_tx_t *tx)
874{
875	zap_leaf_t *l;
876	int err;
877	boolean_t create;
878	zap_entry_handle_t zeh;
879	zap_t *zap = zn->zn_zap;
880
881	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
882	err = fzap_check(zn, integer_size, num_integers);
883	if (err != 0)
884		return (err);
885
886	err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
887	if (err != 0)
888		return (err);
889retry:
890	err = zap_leaf_lookup(l, zn, &zeh);
891	create = (err == ENOENT);
892	ASSERT(err == 0 || err == ENOENT);
893
894	if (create) {
895		err = zap_entry_create(l, zn, ZAP_NEED_CD,
896		    integer_size, num_integers, val, &zeh);
897		if (err == 0)
898			zap_increment_num_entries(zap, 1, tx);
899	} else {
900		err = zap_entry_update(&zeh, integer_size, num_integers, val);
901	}
902
903	if (err == EAGAIN) {
904		err = zap_expand_leaf(zn, l, tag, tx, &l);
905		zap = zn->zn_zap;	/* zap_expand_leaf() may change zap */
906		if (err == 0)
907			goto retry;
908	}
909
910	if (zap != NULL)
911		zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
912	return (err);
913}
914
915int
916fzap_length(zap_name_t *zn,
917    uint64_t *integer_size, uint64_t *num_integers)
918{
919	zap_leaf_t *l;
920	int err;
921	zap_entry_handle_t zeh;
922
923	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
924	if (err != 0)
925		return (err);
926	err = zap_leaf_lookup(l, zn, &zeh);
927	if (err != 0)
928		goto out;
929
930	if (integer_size != 0)
931		*integer_size = zeh.zeh_integer_size;
932	if (num_integers != 0)
933		*num_integers = zeh.zeh_num_integers;
934out:
935	zap_put_leaf(l);
936	return (err);
937}
938
939int
940fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
941{
942	zap_leaf_t *l;
943	int err;
944	zap_entry_handle_t zeh;
945
946	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
947	if (err != 0)
948		return (err);
949	err = zap_leaf_lookup(l, zn, &zeh);
950	if (err == 0) {
951		zap_entry_remove(&zeh);
952		zap_increment_num_entries(zn->zn_zap, -1, tx);
953	}
954	zap_put_leaf(l);
955	return (err);
956}
957
958void
959fzap_prefetch(zap_name_t *zn)
960{
961	uint64_t blk;
962	zap_t *zap = zn->zn_zap;
963
964	uint64_t idx = ZAP_HASH_IDX(zn->zn_hash,
965	    zap_f_phys(zap)->zap_ptrtbl.zt_shift);
966	if (zap_idx_to_blk(zap, idx, &blk) != 0)
967		return;
968	int bs = FZAP_BLOCK_SHIFT(zap);
969	dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs,
970	    ZIO_PRIORITY_SYNC_READ);
971}
972
973/*
974 * Helper functions for consumers.
975 */
976
977uint64_t
978zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
979    const char *name, dmu_tx_t *tx)
980{
981	return (zap_create_link_dnsize(os, ot, parent_obj, name, 0, tx));
982}
983
984uint64_t
985zap_create_link_dnsize(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
986    const char *name, int dnodesize, dmu_tx_t *tx)
987{
988	uint64_t new_obj;
989
990	VERIFY((new_obj = zap_create_dnsize(os, ot, DMU_OT_NONE, 0,
991	    dnodesize, tx)) > 0);
992	VERIFY0(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
993	    tx));
994
995	return (new_obj);
996}
997
998int
999zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
1000    char *name)
1001{
1002	zap_cursor_t zc;
1003	int err;
1004
1005	if (mask == 0)
1006		mask = -1ULL;
1007
1008	zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP);
1009	for (zap_cursor_init(&zc, os, zapobj);
1010	    (err = zap_cursor_retrieve(&zc, za)) == 0;
1011	    zap_cursor_advance(&zc)) {
1012		if ((za->za_first_integer & mask) == (value & mask)) {
1013			(void) strcpy(name, za->za_name);
1014			break;
1015		}
1016	}
1017	zap_cursor_fini(&zc);
1018	kmem_free(za, sizeof (*za));
1019	return (err);
1020}
1021
1022int
1023zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
1024{
1025	zap_cursor_t zc;
1026	int err = 0;
1027
1028	zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP);
1029	for (zap_cursor_init(&zc, os, fromobj);
1030	    zap_cursor_retrieve(&zc, za) == 0;
1031	    (void) zap_cursor_advance(&zc)) {
1032		if (za->za_integer_length != 8 || za->za_num_integers != 1) {
1033			err = SET_ERROR(EINVAL);
1034			break;
1035		}
1036		err = zap_add(os, intoobj, za->za_name,
1037		    8, 1, &za->za_first_integer, tx);
1038		if (err != 0)
1039			break;
1040	}
1041	zap_cursor_fini(&zc);
1042	kmem_free(za, sizeof (*za));
1043	return (err);
1044}
1045
1046int
1047zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1048    uint64_t value, dmu_tx_t *tx)
1049{
1050	zap_cursor_t zc;
1051	int err = 0;
1052
1053	zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP);
1054	for (zap_cursor_init(&zc, os, fromobj);
1055	    zap_cursor_retrieve(&zc, za) == 0;
1056	    (void) zap_cursor_advance(&zc)) {
1057		if (za->za_integer_length != 8 || za->za_num_integers != 1) {
1058			err = SET_ERROR(EINVAL);
1059			break;
1060		}
1061		err = zap_add(os, intoobj, za->za_name,
1062		    8, 1, &value, tx);
1063		if (err != 0)
1064			break;
1065	}
1066	zap_cursor_fini(&zc);
1067	kmem_free(za, sizeof (*za));
1068	return (err);
1069}
1070
1071int
1072zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1073    dmu_tx_t *tx)
1074{
1075	zap_cursor_t zc;
1076	int err = 0;
1077
1078	zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP);
1079	for (zap_cursor_init(&zc, os, fromobj);
1080	    zap_cursor_retrieve(&zc, za) == 0;
1081	    (void) zap_cursor_advance(&zc)) {
1082		uint64_t delta = 0;
1083
1084		if (za->za_integer_length != 8 || za->za_num_integers != 1) {
1085			err = SET_ERROR(EINVAL);
1086			break;
1087		}
1088
1089		err = zap_lookup(os, intoobj, za->za_name, 8, 1, &delta);
1090		if (err != 0 && err != ENOENT)
1091			break;
1092		delta += za->za_first_integer;
1093		err = zap_update(os, intoobj, za->za_name, 8, 1, &delta, tx);
1094		if (err != 0)
1095			break;
1096	}
1097	zap_cursor_fini(&zc);
1098	kmem_free(za, sizeof (*za));
1099	return (err);
1100}
1101
1102int
1103zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1104{
1105	char name[20];
1106
1107	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1108	return (zap_add(os, obj, name, 8, 1, &value, tx));
1109}
1110
1111int
1112zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1113{
1114	char name[20];
1115
1116	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1117	return (zap_remove(os, obj, name, tx));
1118}
1119
1120int
1121zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1122{
1123	char name[20];
1124
1125	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1126	return (zap_lookup(os, obj, name, 8, 1, &value));
1127}
1128
1129int
1130zap_add_int_key(objset_t *os, uint64_t obj,
1131    uint64_t key, uint64_t value, dmu_tx_t *tx)
1132{
1133	char name[20];
1134
1135	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1136	return (zap_add(os, obj, name, 8, 1, &value, tx));
1137}
1138
1139int
1140zap_update_int_key(objset_t *os, uint64_t obj,
1141    uint64_t key, uint64_t value, dmu_tx_t *tx)
1142{
1143	char name[20];
1144
1145	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1146	return (zap_update(os, obj, name, 8, 1, &value, tx));
1147}
1148
1149int
1150zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1151{
1152	char name[20];
1153
1154	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1155	return (zap_lookup(os, obj, name, 8, 1, valuep));
1156}
1157
1158int
1159zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1160    dmu_tx_t *tx)
1161{
1162	uint64_t value = 0;
1163
1164	if (delta == 0)
1165		return (0);
1166
1167	int err = zap_lookup(os, obj, name, 8, 1, &value);
1168	if (err != 0 && err != ENOENT)
1169		return (err);
1170	value += delta;
1171	if (value == 0)
1172		err = zap_remove(os, obj, name, tx);
1173	else
1174		err = zap_update(os, obj, name, 8, 1, &value, tx);
1175	return (err);
1176}
1177
1178int
1179zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1180    dmu_tx_t *tx)
1181{
1182	char name[20];
1183
1184	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1185	return (zap_increment(os, obj, name, delta, tx));
1186}
1187
1188/*
1189 * Routines for iterating over the attributes.
1190 */
1191
1192int
1193fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1194{
1195	int err = ENOENT;
1196	zap_entry_handle_t zeh;
1197	zap_leaf_t *l;
1198
1199	/* retrieve the next entry at or after zc_hash/zc_cd */
1200	/* if no entry, return ENOENT */
1201
1202	/*
1203	 * If we are reading from the beginning, we're almost
1204	 * certain to iterate over the entire ZAP object.  If there are
1205	 * multiple leaf blocks (freeblk > 2), prefetch the whole
1206	 * object, so that we read the leaf blocks concurrently.
1207	 * (Unless noprefetch was requested via zap_cursor_init_noprefetch()).
1208	 */
1209	if (zc->zc_hash == 0 && zap_iterate_prefetch &&
1210	    zc->zc_prefetch && zap_f_phys(zap)->zap_freeblk > 2) {
1211		dmu_prefetch(zc->zc_objset, zc->zc_zapobj, 0, 0,
1212		    zap_f_phys(zap)->zap_freeblk << FZAP_BLOCK_SHIFT(zap),
1213		    ZIO_PRIORITY_ASYNC_READ);
1214	}
1215
1216	if (zc->zc_leaf &&
1217	    (ZAP_HASH_IDX(zc->zc_hash,
1218	    zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) !=
1219	    zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) {
1220		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1221		zap_put_leaf(zc->zc_leaf);
1222		zc->zc_leaf = NULL;
1223	}
1224
1225again:
1226	if (zc->zc_leaf == NULL) {
1227		err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1228		    &zc->zc_leaf);
1229		if (err != 0)
1230			return (err);
1231	} else {
1232		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1233	}
1234	l = zc->zc_leaf;
1235
1236	err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1237
1238	if (err == ENOENT) {
1239		uint64_t nocare =
1240		    (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1;
1241		zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1242		zc->zc_cd = 0;
1243		if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 ||
1244		    zc->zc_hash == 0) {
1245			zc->zc_hash = -1ULL;
1246		} else {
1247			zap_put_leaf(zc->zc_leaf);
1248			zc->zc_leaf = NULL;
1249			goto again;
1250		}
1251	}
1252
1253	if (err == 0) {
1254		zc->zc_hash = zeh.zeh_hash;
1255		zc->zc_cd = zeh.zeh_cd;
1256		za->za_integer_length = zeh.zeh_integer_size;
1257		za->za_num_integers = zeh.zeh_num_integers;
1258		if (zeh.zeh_num_integers == 0) {
1259			za->za_first_integer = 0;
1260		} else {
1261			err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1262			ASSERT(err == 0 || err == EOVERFLOW);
1263		}
1264		err = zap_entry_read_name(zap, &zeh,
1265		    sizeof (za->za_name), za->za_name);
1266		ASSERT(err == 0);
1267
1268		za->za_normalization_conflict =
1269		    zap_entry_normalization_conflict(&zeh,
1270		    NULL, za->za_name, zap);
1271	}
1272	rw_exit(&zc->zc_leaf->l_rwlock);
1273	return (err);
1274}
1275
1276static void
1277zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1278{
1279	uint64_t lastblk = 0;
1280
1281	/*
1282	 * NB: if a leaf has more pointers than an entire ptrtbl block
1283	 * can hold, then it'll be accounted for more than once, since
1284	 * we won't have lastblk.
1285	 */
1286	for (int i = 0; i < len; i++) {
1287		zap_leaf_t *l;
1288
1289		if (tbl[i] == lastblk)
1290			continue;
1291		lastblk = tbl[i];
1292
1293		int err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1294		if (err == 0) {
1295			zap_leaf_stats(zap, l, zs);
1296			zap_put_leaf(l);
1297		}
1298	}
1299}
1300
1301void
1302fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1303{
1304	int bs = FZAP_BLOCK_SHIFT(zap);
1305	zs->zs_blocksize = 1ULL << bs;
1306
1307	/*
1308	 * Set zap_phys_t fields
1309	 */
1310	zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs;
1311	zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries;
1312	zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk;
1313	zs->zs_block_type = zap_f_phys(zap)->zap_block_type;
1314	zs->zs_magic = zap_f_phys(zap)->zap_magic;
1315	zs->zs_salt = zap_f_phys(zap)->zap_salt;
1316
1317	/*
1318	 * Set zap_ptrtbl fields
1319	 */
1320	zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1321	zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk;
1322	zs->zs_ptrtbl_blks_copied =
1323	    zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied;
1324	zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk;
1325	zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1326	zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1327
1328	if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
1329		/* the ptrtbl is entirely in the header block. */
1330		zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1331		    1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1332	} else {
1333		dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
1334		    zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs,
1335		    zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs,
1336		    ZIO_PRIORITY_SYNC_READ);
1337
1338		for (int b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1339		    b++) {
1340			dmu_buf_t *db;
1341			int err;
1342
1343			err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1344			    (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs,
1345			    FTAG, &db, DMU_READ_NO_PREFETCH);
1346			if (err == 0) {
1347				zap_stats_ptrtbl(zap, db->db_data,
1348				    1<<(bs-3), zs);
1349				dmu_buf_rele(db, FTAG);
1350			}
1351		}
1352	}
1353}
1354