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/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2017 Nexenta Systems, Inc.
28 */
29
30#include <sys/zio.h>
31#include <sys/spa.h>
32#include <sys/dmu.h>
33#include <sys/zfs_context.h>
34#include <sys/zap.h>
35#include <sys/refcount.h>
36#include <sys/zap_impl.h>
37#include <sys/zap_leaf.h>
38#include <sys/avl.h>
39#include <sys/arc.h>
40#include <sys/dmu_objset.h>
41
42#ifdef _KERNEL
43#include <sys/sunddi.h>
44#endif
45
46extern inline mzap_phys_t *zap_m_phys(zap_t *zap);
47
48static int mzap_upgrade(zap_t **zapp,
49    void *tag, dmu_tx_t *tx, zap_flags_t flags);
50
51uint64_t
52zap_getflags(zap_t *zap)
53{
54	if (zap->zap_ismicro)
55		return (0);
56	return (zap_f_phys(zap)->zap_flags);
57}
58
59int
60zap_hashbits(zap_t *zap)
61{
62	if (zap_getflags(zap) & ZAP_FLAG_HASH64)
63		return (48);
64	else
65		return (28);
66}
67
68uint32_t
69zap_maxcd(zap_t *zap)
70{
71	if (zap_getflags(zap) & ZAP_FLAG_HASH64)
72		return ((1<<16)-1);
73	else
74		return (-1U);
75}
76
77static uint64_t
78zap_hash(zap_name_t *zn)
79{
80	zap_t *zap = zn->zn_zap;
81	uint64_t h = 0;
82
83	if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
84		ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
85		h = *(uint64_t *)zn->zn_key_orig;
86	} else {
87		h = zap->zap_salt;
88		ASSERT(h != 0);
89		ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
90
91		if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
92			const uint64_t *wp = zn->zn_key_norm;
93
94			ASSERT(zn->zn_key_intlen == 8);
95			for (int i = 0; i < zn->zn_key_norm_numints;
96			    wp++, i++) {
97				uint64_t word = *wp;
98
99				for (int j = 0; j < zn->zn_key_intlen; j++) {
100					h = (h >> 8) ^
101					    zfs_crc64_table[(h ^ word) & 0xFF];
102					word >>= NBBY;
103				}
104			}
105		} else {
106			const uint8_t *cp = zn->zn_key_norm;
107
108			/*
109			 * We previously stored the terminating null on
110			 * disk, but didn't hash it, so we need to
111			 * continue to not hash it.  (The
112			 * zn_key_*_numints includes the terminating
113			 * null for non-binary keys.)
114			 */
115			int len = zn->zn_key_norm_numints - 1;
116
117			ASSERT(zn->zn_key_intlen == 1);
118			for (int i = 0; i < len; cp++, i++) {
119				h = (h >> 8) ^
120				    zfs_crc64_table[(h ^ *cp) & 0xFF];
121			}
122		}
123	}
124	/*
125	 * Don't use all 64 bits, since we need some in the cookie for
126	 * the collision differentiator.  We MUST use the high bits,
127	 * since those are the ones that we first pay attention to when
128	 * chosing the bucket.
129	 */
130	h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
131
132	return (h);
133}
134
135static int
136zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags)
137{
138	ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
139
140	size_t inlen = strlen(name) + 1;
141	size_t outlen = ZAP_MAXNAMELEN;
142
143	int err = 0;
144	(void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
145	    normflags | U8_TEXTPREP_IGNORE_NULL | U8_TEXTPREP_IGNORE_INVALID,
146	    U8_UNICODE_LATEST, &err);
147
148	return (err);
149}
150
151boolean_t
152zap_match(zap_name_t *zn, const char *matchname)
153{
154	ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
155
156	if (zn->zn_matchtype & MT_NORMALIZE) {
157		char norm[ZAP_MAXNAMELEN];
158
159		if (zap_normalize(zn->zn_zap, matchname, norm,
160		    zn->zn_normflags) != 0)
161			return (B_FALSE);
162
163		return (strcmp(zn->zn_key_norm, norm) == 0);
164	} else {
165		return (strcmp(zn->zn_key_orig, matchname) == 0);
166	}
167}
168
169void
170zap_name_free(zap_name_t *zn)
171{
172	kmem_free(zn, sizeof (zap_name_t));
173}
174
175zap_name_t *
176zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
177{
178	zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
179
180	zn->zn_zap = zap;
181	zn->zn_key_intlen = sizeof (*key);
182	zn->zn_key_orig = key;
183	zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
184	zn->zn_matchtype = mt;
185	zn->zn_normflags = zap->zap_normflags;
186
187	/*
188	 * If we're dealing with a case sensitive lookup on a mixed or
189	 * insensitive fs, remove U8_TEXTPREP_TOUPPER or the lookup
190	 * will fold case to all caps overriding the lookup request.
191	 */
192	if (mt & MT_MATCH_CASE)
193		zn->zn_normflags &= ~U8_TEXTPREP_TOUPPER;
194
195	if (zap->zap_normflags) {
196		/*
197		 * We *must* use zap_normflags because this normalization is
198		 * what the hash is computed from.
199		 */
200		if (zap_normalize(zap, key, zn->zn_normbuf,
201		    zap->zap_normflags) != 0) {
202			zap_name_free(zn);
203			return (NULL);
204		}
205		zn->zn_key_norm = zn->zn_normbuf;
206		zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
207	} else {
208		if (mt != 0) {
209			zap_name_free(zn);
210			return (NULL);
211		}
212		zn->zn_key_norm = zn->zn_key_orig;
213		zn->zn_key_norm_numints = zn->zn_key_orig_numints;
214	}
215
216	zn->zn_hash = zap_hash(zn);
217
218	if (zap->zap_normflags != zn->zn_normflags) {
219		/*
220		 * We *must* use zn_normflags because this normalization is
221		 * what the matching is based on.  (Not the hash!)
222		 */
223		if (zap_normalize(zap, key, zn->zn_normbuf,
224		    zn->zn_normflags) != 0) {
225			zap_name_free(zn);
226			return (NULL);
227		}
228		zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
229	}
230
231	return (zn);
232}
233
234zap_name_t *
235zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
236{
237	zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
238
239	ASSERT(zap->zap_normflags == 0);
240	zn->zn_zap = zap;
241	zn->zn_key_intlen = sizeof (*key);
242	zn->zn_key_orig = zn->zn_key_norm = key;
243	zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
244	zn->zn_matchtype = 0;
245
246	zn->zn_hash = zap_hash(zn);
247	return (zn);
248}
249
250static void
251mzap_byteswap(mzap_phys_t *buf, size_t size)
252{
253	buf->mz_block_type = BSWAP_64(buf->mz_block_type);
254	buf->mz_salt = BSWAP_64(buf->mz_salt);
255	buf->mz_normflags = BSWAP_64(buf->mz_normflags);
256	int max = (size / MZAP_ENT_LEN) - 1;
257	for (int i = 0; i < max; i++) {
258		buf->mz_chunk[i].mze_value =
259		    BSWAP_64(buf->mz_chunk[i].mze_value);
260		buf->mz_chunk[i].mze_cd =
261		    BSWAP_32(buf->mz_chunk[i].mze_cd);
262	}
263}
264
265void
266zap_byteswap(void *buf, size_t size)
267{
268	uint64_t block_type = *(uint64_t *)buf;
269
270	if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
271		/* ASSERT(magic == ZAP_LEAF_MAGIC); */
272		mzap_byteswap(buf, size);
273	} else {
274		fzap_byteswap(buf, size);
275	}
276}
277
278static int
279mze_compare(const void *arg1, const void *arg2)
280{
281	const mzap_ent_t *mze1 = arg1;
282	const mzap_ent_t *mze2 = arg2;
283
284	int cmp = AVL_CMP(mze1->mze_hash, mze2->mze_hash);
285	if (likely(cmp))
286		return (cmp);
287
288	return (AVL_CMP(mze1->mze_cd, mze2->mze_cd));
289}
290
291static int
292mze_insert(zap_t *zap, int chunkid, uint64_t hash)
293{
294	avl_index_t idx;
295
296	ASSERT(zap->zap_ismicro);
297	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
298
299	mzap_ent_t *mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
300	mze->mze_chunkid = chunkid;
301	mze->mze_hash = hash;
302	mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
303	ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
304	if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
305		kmem_free(mze, sizeof (mzap_ent_t));
306		return (EEXIST);
307	}
308	avl_insert(&zap->zap_m.zap_avl, mze, idx);
309	return (0);
310}
311
312static mzap_ent_t *
313mze_find(zap_name_t *zn)
314{
315	mzap_ent_t mze_tofind;
316	mzap_ent_t *mze;
317	avl_index_t idx;
318	avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
319
320	ASSERT(zn->zn_zap->zap_ismicro);
321	ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
322
323	mze_tofind.mze_hash = zn->zn_hash;
324	mze_tofind.mze_cd = 0;
325
326	mze = avl_find(avl, &mze_tofind, &idx);
327	if (mze == NULL)
328		mze = avl_nearest(avl, idx, AVL_AFTER);
329	for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
330		ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
331		if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
332			return (mze);
333	}
334
335	return (NULL);
336}
337
338static uint32_t
339mze_find_unused_cd(zap_t *zap, uint64_t hash)
340{
341	mzap_ent_t mze_tofind;
342	avl_index_t idx;
343	avl_tree_t *avl = &zap->zap_m.zap_avl;
344
345	ASSERT(zap->zap_ismicro);
346	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
347
348	mze_tofind.mze_hash = hash;
349	mze_tofind.mze_cd = 0;
350
351	uint32_t cd = 0;
352	for (mzap_ent_t *mze = avl_find(avl, &mze_tofind, &idx);
353	    mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
354		if (mze->mze_cd != cd)
355			break;
356		cd++;
357	}
358
359	return (cd);
360}
361
362static void
363mze_remove(zap_t *zap, mzap_ent_t *mze)
364{
365	ASSERT(zap->zap_ismicro);
366	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
367
368	avl_remove(&zap->zap_m.zap_avl, mze);
369	kmem_free(mze, sizeof (mzap_ent_t));
370}
371
372static void
373mze_destroy(zap_t *zap)
374{
375	mzap_ent_t *mze;
376	void *avlcookie = NULL;
377
378	while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
379		kmem_free(mze, sizeof (mzap_ent_t));
380	avl_destroy(&zap->zap_m.zap_avl);
381}
382
383static zap_t *
384mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
385{
386	zap_t *winner;
387	uint64_t *zap_hdr = (uint64_t *)db->db_data;
388	uint64_t zap_block_type = zap_hdr[0];
389	uint64_t zap_magic = zap_hdr[1];
390
391	ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
392
393	zap_t *zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
394	rw_init(&zap->zap_rwlock, 0, 0, 0);
395	rw_enter(&zap->zap_rwlock, RW_WRITER);
396	zap->zap_objset = os;
397	zap->zap_object = obj;
398	zap->zap_dbuf = db;
399
400	if (zap_block_type != ZBT_MICRO) {
401		mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
402		zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
403		if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
404			winner = NULL;	/* No actual winner here... */
405			goto handle_winner;
406		}
407	} else {
408		zap->zap_ismicro = TRUE;
409	}
410
411	/*
412	 * Make sure that zap_ismicro is set before we let others see
413	 * it, because zap_lockdir() checks zap_ismicro without the lock
414	 * held.
415	 */
416	dmu_buf_init_user(&zap->zap_dbu, zap_evict_sync, NULL, &zap->zap_dbuf);
417	winner = dmu_buf_set_user(db, &zap->zap_dbu);
418
419	if (winner != NULL)
420		goto handle_winner;
421
422	if (zap->zap_ismicro) {
423		zap->zap_salt = zap_m_phys(zap)->mz_salt;
424		zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
425		zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
426		avl_create(&zap->zap_m.zap_avl, mze_compare,
427		    sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
428
429		for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
430			mzap_ent_phys_t *mze =
431			    &zap_m_phys(zap)->mz_chunk[i];
432			if (mze->mze_name[0]) {
433				zap_name_t *zn;
434
435				zn = zap_name_alloc(zap, mze->mze_name, 0);
436				if (mze_insert(zap, i, zn->zn_hash) == 0)
437					zap->zap_m.zap_num_entries++;
438				else {
439					printf("ZFS WARNING: Duplicated ZAP "
440					    "entry detected (%s).\n",
441					    mze->mze_name);
442				}
443				zap_name_free(zn);
444			}
445		}
446	} else {
447		zap->zap_salt = zap_f_phys(zap)->zap_salt;
448		zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
449
450		ASSERT3U(sizeof (struct zap_leaf_header), ==,
451		    2*ZAP_LEAF_CHUNKSIZE);
452
453		/*
454		 * The embedded pointer table should not overlap the
455		 * other members.
456		 */
457		ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
458		    &zap_f_phys(zap)->zap_salt);
459
460		/*
461		 * The embedded pointer table should end at the end of
462		 * the block
463		 */
464		ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
465		    1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
466		    (uintptr_t)zap_f_phys(zap), ==,
467		    zap->zap_dbuf->db_size);
468	}
469	rw_exit(&zap->zap_rwlock);
470	return (zap);
471
472handle_winner:
473	rw_exit(&zap->zap_rwlock);
474	rw_destroy(&zap->zap_rwlock);
475	if (!zap->zap_ismicro)
476		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
477	kmem_free(zap, sizeof (zap_t));
478	return (winner);
479}
480
481/*
482 * This routine "consumes" the caller's hold on the dbuf, which must
483 * have the specified tag.
484 */
485static int
486zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx,
487    krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
488{
489	ASSERT0(db->db_offset);
490	objset_t *os = dmu_buf_get_objset(db);
491	uint64_t obj = db->db_object;
492
493	*zapp = NULL;
494
495	zap_t *zap = dmu_buf_get_user(db);
496	if (zap == NULL) {
497		zap = mzap_open(os, obj, db);
498		if (zap == NULL) {
499			/*
500			 * mzap_open() didn't like what it saw on-disk.
501			 * Check for corruption!
502			 */
503			return (SET_ERROR(EIO));
504		}
505	}
506
507	/*
508	 * We're checking zap_ismicro without the lock held, in order to
509	 * tell what type of lock we want.  Once we have some sort of
510	 * lock, see if it really is the right type.  In practice this
511	 * can only be different if it was upgraded from micro to fat,
512	 * and micro wanted WRITER but fat only needs READER.
513	 */
514	krw_t lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
515	rw_enter(&zap->zap_rwlock, lt);
516	if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
517		/* it was upgraded, now we only need reader */
518		ASSERT(lt == RW_WRITER);
519		ASSERT(RW_READER ==
520		    (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
521		rw_downgrade(&zap->zap_rwlock);
522		lt = RW_READER;
523	}
524
525	zap->zap_objset = os;
526
527	if (lt == RW_WRITER)
528		dmu_buf_will_dirty(db, tx);
529
530	ASSERT3P(zap->zap_dbuf, ==, db);
531
532	ASSERT(!zap->zap_ismicro ||
533	    zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
534	if (zap->zap_ismicro && tx && adding &&
535	    zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
536		uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
537		if (newsz > MZAP_MAX_BLKSZ) {
538			dprintf("upgrading obj %llu: num_entries=%u\n",
539			    obj, zap->zap_m.zap_num_entries);
540			*zapp = zap;
541			int err = mzap_upgrade(zapp, tag, tx, 0);
542			if (err != 0)
543				rw_exit(&zap->zap_rwlock);
544			return (err);
545		}
546		VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx));
547		zap->zap_m.zap_num_chunks =
548		    db->db_size / MZAP_ENT_LEN - 1;
549	}
550
551	*zapp = zap;
552	return (0);
553}
554
555static int
556zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
557    krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
558{
559	dmu_buf_t *db;
560
561	int err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
562	if (err != 0) {
563		return (err);
564	}
565#ifdef ZFS_DEBUG
566	{
567		dmu_object_info_t doi;
568		dmu_object_info_from_db(db, &doi);
569		ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
570	}
571#endif
572
573	err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
574	if (err != 0) {
575		dmu_buf_rele(db, tag);
576	}
577	return (err);
578}
579
580int
581zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
582    krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
583{
584	dmu_buf_t *db;
585
586	int err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH);
587	if (err != 0)
588		return (err);
589#ifdef ZFS_DEBUG
590	{
591		dmu_object_info_t doi;
592		dmu_object_info_from_db(db, &doi);
593		ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
594	}
595#endif
596	err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
597	if (err != 0)
598		dmu_buf_rele(db, tag);
599	return (err);
600}
601
602void
603zap_unlockdir(zap_t *zap, void *tag)
604{
605	rw_exit(&zap->zap_rwlock);
606	dmu_buf_rele(zap->zap_dbuf, tag);
607}
608
609static int
610mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags)
611{
612	int err = 0;
613	zap_t *zap = *zapp;
614
615	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
616
617	int sz = zap->zap_dbuf->db_size;
618	mzap_phys_t *mzp = zio_buf_alloc(sz);
619	bcopy(zap->zap_dbuf->db_data, mzp, sz);
620	int nchunks = zap->zap_m.zap_num_chunks;
621
622	if (!flags) {
623		err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
624		    1ULL << fzap_default_block_shift, 0, tx);
625		if (err != 0) {
626			zio_buf_free(mzp, sz);
627			return (err);
628		}
629	}
630
631	dprintf("upgrading obj=%llu with %u chunks\n",
632	    zap->zap_object, nchunks);
633	/* XXX destroy the avl later, so we can use the stored hash value */
634	mze_destroy(zap);
635
636	fzap_upgrade(zap, tx, flags);
637
638	for (int i = 0; i < nchunks; i++) {
639		mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
640		if (mze->mze_name[0] == 0)
641			continue;
642		dprintf("adding %s=%llu\n",
643		    mze->mze_name, mze->mze_value);
644		zap_name_t *zn = zap_name_alloc(zap, mze->mze_name, 0);
645		err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
646		    tag, tx);
647		zap = zn->zn_zap;	/* fzap_add_cd() may change zap */
648		zap_name_free(zn);
649		if (err != 0)
650			break;
651	}
652	zio_buf_free(mzp, sz);
653	*zapp = zap;
654	return (err);
655}
656
657/*
658 * The "normflags" determine the behavior of the matchtype_t which is
659 * passed to zap_lookup_norm().  Names which have the same normalized
660 * version will be stored with the same hash value, and therefore we can
661 * perform normalization-insensitive lookups.  We can be Unicode form-
662 * insensitive and/or case-insensitive.  The following flags are valid for
663 * "normflags":
664 *
665 * U8_TEXTPREP_NFC
666 * U8_TEXTPREP_NFD
667 * U8_TEXTPREP_NFKC
668 * U8_TEXTPREP_NFKD
669 * U8_TEXTPREP_TOUPPER
670 *
671 * The *_NF* (Normalization Form) flags are mutually exclusive; at most one
672 * of them may be supplied.
673 */
674void
675mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
676    dmu_tx_t *tx)
677{
678	dmu_buf_t *db;
679
680	VERIFY0(dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
681
682	dmu_buf_will_dirty(db, tx);
683	mzap_phys_t *zp = db->db_data;
684	zp->mz_block_type = ZBT_MICRO;
685	zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
686	zp->mz_normflags = normflags;
687
688	if (flags != 0) {
689		zap_t *zap;
690		/* Only fat zap supports flags; upgrade immediately. */
691		VERIFY0(zap_lockdir_impl(db, FTAG, tx, RW_WRITER,
692		    B_FALSE, B_FALSE, &zap));
693		VERIFY0(mzap_upgrade(&zap, FTAG, tx, flags));
694		zap_unlockdir(zap, FTAG);
695	} else {
696		dmu_buf_rele(db, FTAG);
697	}
698}
699
700int
701zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
702    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
703{
704	return (zap_create_claim_norm(os, obj,
705	    0, ot, bonustype, bonuslen, tx));
706}
707
708int
709zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
710    dmu_object_type_t ot,
711    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
712{
713	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
714	int err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
715	if (err != 0)
716		return (err);
717	mzap_create_impl(os, obj, normflags, 0, tx);
718	return (0);
719}
720
721uint64_t
722zap_create(objset_t *os, dmu_object_type_t ot,
723    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
724{
725	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
726}
727
728uint64_t
729zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
730    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
731{
732	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
733	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
734
735	mzap_create_impl(os, obj, normflags, 0, tx);
736	return (obj);
737}
738
739uint64_t
740zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
741    dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
742    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
743{
744	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
745	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
746
747	ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
748	    leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
749	    indirect_blockshift >= SPA_MINBLOCKSHIFT &&
750	    indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
751
752	VERIFY(dmu_object_set_blocksize(os, obj,
753	    1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
754
755	mzap_create_impl(os, obj, normflags, flags, tx);
756	return (obj);
757}
758
759int
760zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
761{
762	/*
763	 * dmu_object_free will free the object number and free the
764	 * data.  Freeing the data will cause our pageout function to be
765	 * called, which will destroy our data (zap_leaf_t's and zap_t).
766	 */
767
768	return (dmu_object_free(os, zapobj, tx));
769}
770
771void
772zap_evict_sync(void *dbu)
773{
774	zap_t *zap = dbu;
775
776	rw_destroy(&zap->zap_rwlock);
777
778	if (zap->zap_ismicro)
779		mze_destroy(zap);
780	else
781		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
782
783	kmem_free(zap, sizeof (zap_t));
784}
785
786int
787zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
788{
789	zap_t *zap;
790
791	int err =
792	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
793	if (err != 0)
794		return (err);
795	if (!zap->zap_ismicro) {
796		err = fzap_count(zap, count);
797	} else {
798		*count = zap->zap_m.zap_num_entries;
799	}
800	zap_unlockdir(zap, FTAG);
801	return (err);
802}
803
804/*
805 * zn may be NULL; if not specified, it will be computed if needed.
806 * See also the comment above zap_entry_normalization_conflict().
807 */
808static boolean_t
809mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
810{
811	int direction = AVL_BEFORE;
812	boolean_t allocdzn = B_FALSE;
813
814	if (zap->zap_normflags == 0)
815		return (B_FALSE);
816
817again:
818	for (mzap_ent_t *other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
819	    other && other->mze_hash == mze->mze_hash;
820	    other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
821
822		if (zn == NULL) {
823			zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
824			    MT_NORMALIZE);
825			allocdzn = B_TRUE;
826		}
827		if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
828			if (allocdzn)
829				zap_name_free(zn);
830			return (B_TRUE);
831		}
832	}
833
834	if (direction == AVL_BEFORE) {
835		direction = AVL_AFTER;
836		goto again;
837	}
838
839	if (allocdzn)
840		zap_name_free(zn);
841	return (B_FALSE);
842}
843
844/*
845 * Routines for manipulating attributes.
846 */
847
848int
849zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
850    uint64_t integer_size, uint64_t num_integers, void *buf)
851{
852	return (zap_lookup_norm(os, zapobj, name, integer_size,
853	    num_integers, buf, 0, NULL, 0, NULL));
854}
855
856static int
857zap_lookup_impl(zap_t *zap, const char *name,
858    uint64_t integer_size, uint64_t num_integers, void *buf,
859    matchtype_t mt, char *realname, int rn_len,
860    boolean_t *ncp)
861{
862	int err = 0;
863
864	zap_name_t *zn = zap_name_alloc(zap, name, mt);
865	if (zn == NULL)
866		return (SET_ERROR(ENOTSUP));
867
868	if (!zap->zap_ismicro) {
869		err = fzap_lookup(zn, integer_size, num_integers, buf,
870		    realname, rn_len, ncp);
871	} else {
872		mzap_ent_t *mze = mze_find(zn);
873		if (mze == NULL) {
874			err = SET_ERROR(ENOENT);
875		} else {
876			if (num_integers < 1) {
877				err = SET_ERROR(EOVERFLOW);
878			} else if (integer_size != 8) {
879				err = SET_ERROR(EINVAL);
880			} else {
881				*(uint64_t *)buf =
882				    MZE_PHYS(zap, mze)->mze_value;
883				(void) strlcpy(realname,
884				    MZE_PHYS(zap, mze)->mze_name, rn_len);
885				if (ncp) {
886					*ncp = mzap_normalization_conflict(zap,
887					    zn, mze);
888				}
889			}
890		}
891	}
892	zap_name_free(zn);
893	return (err);
894}
895
896int
897zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
898    uint64_t integer_size, uint64_t num_integers, void *buf,
899    matchtype_t mt, char *realname, int rn_len,
900    boolean_t *ncp)
901{
902	zap_t *zap;
903
904	int err =
905	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
906	if (err != 0)
907		return (err);
908	err = zap_lookup_impl(zap, name, integer_size,
909	    num_integers, buf, mt, realname, rn_len, ncp);
910	zap_unlockdir(zap, FTAG);
911	return (err);
912}
913
914int
915zap_lookup_by_dnode(dnode_t *dn, const char *name,
916    uint64_t integer_size, uint64_t num_integers, void *buf)
917{
918	return (zap_lookup_norm_by_dnode(dn, name, integer_size,
919	    num_integers, buf, 0, NULL, 0, NULL));
920}
921
922int
923zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
924    uint64_t integer_size, uint64_t num_integers, void *buf,
925    matchtype_t mt, char *realname, int rn_len,
926    boolean_t *ncp)
927{
928	zap_t *zap;
929
930	int err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
931	    FTAG, &zap);
932	if (err != 0)
933		return (err);
934	err = zap_lookup_impl(zap, name, integer_size,
935	    num_integers, buf, mt, realname, rn_len, ncp);
936	zap_unlockdir(zap, FTAG);
937	return (err);
938}
939
940int
941zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
942    int key_numints)
943{
944	zap_t *zap;
945
946	int err =
947	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
948	if (err != 0)
949		return (err);
950	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
951	if (zn == NULL) {
952		zap_unlockdir(zap, FTAG);
953		return (SET_ERROR(ENOTSUP));
954	}
955
956	fzap_prefetch(zn);
957	zap_name_free(zn);
958	zap_unlockdir(zap, FTAG);
959	return (err);
960}
961
962int
963zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
964    int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
965{
966	zap_t *zap;
967
968	int err =
969	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
970	if (err != 0)
971		return (err);
972	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
973	if (zn == NULL) {
974		zap_unlockdir(zap, FTAG);
975		return (SET_ERROR(ENOTSUP));
976	}
977
978	err = fzap_lookup(zn, integer_size, num_integers, buf,
979	    NULL, 0, NULL);
980	zap_name_free(zn);
981	zap_unlockdir(zap, FTAG);
982	return (err);
983}
984
985int
986zap_contains(objset_t *os, uint64_t zapobj, const char *name)
987{
988	int err = zap_lookup_norm(os, zapobj, name, 0,
989	    0, NULL, 0, NULL, 0, NULL);
990	if (err == EOVERFLOW || err == EINVAL)
991		err = 0; /* found, but skipped reading the value */
992	return (err);
993}
994
995int
996zap_length(objset_t *os, uint64_t zapobj, const char *name,
997    uint64_t *integer_size, uint64_t *num_integers)
998{
999	zap_t *zap;
1000
1001	int err =
1002	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1003	if (err != 0)
1004		return (err);
1005	zap_name_t *zn = zap_name_alloc(zap, name, 0);
1006	if (zn == NULL) {
1007		zap_unlockdir(zap, FTAG);
1008		return (SET_ERROR(ENOTSUP));
1009	}
1010	if (!zap->zap_ismicro) {
1011		err = fzap_length(zn, integer_size, num_integers);
1012	} else {
1013		mzap_ent_t *mze = mze_find(zn);
1014		if (mze == NULL) {
1015			err = SET_ERROR(ENOENT);
1016		} else {
1017			if (integer_size)
1018				*integer_size = 8;
1019			if (num_integers)
1020				*num_integers = 1;
1021		}
1022	}
1023	zap_name_free(zn);
1024	zap_unlockdir(zap, FTAG);
1025	return (err);
1026}
1027
1028int
1029zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1030    int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1031{
1032	zap_t *zap;
1033
1034	int err =
1035	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1036	if (err != 0)
1037		return (err);
1038	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1039	if (zn == NULL) {
1040		zap_unlockdir(zap, FTAG);
1041		return (SET_ERROR(ENOTSUP));
1042	}
1043	err = fzap_length(zn, integer_size, num_integers);
1044	zap_name_free(zn);
1045	zap_unlockdir(zap, FTAG);
1046	return (err);
1047}
1048
1049static void
1050mzap_addent(zap_name_t *zn, uint64_t value)
1051{
1052	zap_t *zap = zn->zn_zap;
1053	int start = zap->zap_m.zap_alloc_next;
1054
1055	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1056
1057#ifdef ZFS_DEBUG
1058	for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1059		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1060		ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1061	}
1062#endif
1063
1064	uint32_t cd = mze_find_unused_cd(zap, zn->zn_hash);
1065	/* given the limited size of the microzap, this can't happen */
1066	ASSERT(cd < zap_maxcd(zap));
1067
1068again:
1069	for (int i = start; i < zap->zap_m.zap_num_chunks; i++) {
1070		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1071		if (mze->mze_name[0] == 0) {
1072			mze->mze_value = value;
1073			mze->mze_cd = cd;
1074			(void) strcpy(mze->mze_name, zn->zn_key_orig);
1075			zap->zap_m.zap_num_entries++;
1076			zap->zap_m.zap_alloc_next = i+1;
1077			if (zap->zap_m.zap_alloc_next ==
1078			    zap->zap_m.zap_num_chunks)
1079				zap->zap_m.zap_alloc_next = 0;
1080			VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
1081			return;
1082		}
1083	}
1084	if (start != 0) {
1085		start = 0;
1086		goto again;
1087	}
1088	ASSERT(!"out of entries!");
1089}
1090
1091static int
1092zap_add_impl(zap_t *zap, const char *key,
1093    int integer_size, uint64_t num_integers,
1094    const void *val, dmu_tx_t *tx, void *tag)
1095{
1096	const uint64_t *intval = val;
1097	int err = 0;
1098
1099	zap_name_t *zn = zap_name_alloc(zap, key, 0);
1100	if (zn == NULL) {
1101		zap_unlockdir(zap, tag);
1102		return (SET_ERROR(ENOTSUP));
1103	}
1104	if (!zap->zap_ismicro) {
1105		err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1106		zap = zn->zn_zap;	/* fzap_add() may change zap */
1107	} else if (integer_size != 8 || num_integers != 1 ||
1108	    strlen(key) >= MZAP_NAME_LEN) {
1109		err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
1110		if (err == 0) {
1111			err = fzap_add(zn, integer_size, num_integers, val,
1112			    tag, tx);
1113		}
1114		zap = zn->zn_zap;	/* fzap_add() may change zap */
1115	} else {
1116		if (mze_find(zn) != NULL) {
1117			err = SET_ERROR(EEXIST);
1118		} else {
1119			mzap_addent(zn, *intval);
1120		}
1121	}
1122	ASSERT(zap == zn->zn_zap);
1123	zap_name_free(zn);
1124	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1125		zap_unlockdir(zap, tag);
1126	return (err);
1127}
1128
1129int
1130zap_add(objset_t *os, uint64_t zapobj, const char *key,
1131    int integer_size, uint64_t num_integers,
1132    const void *val, dmu_tx_t *tx)
1133{
1134	zap_t *zap;
1135	int err;
1136
1137	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1138	if (err != 0)
1139		return (err);
1140	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1141	/* zap_add_impl() calls zap_unlockdir() */
1142	return (err);
1143}
1144
1145int
1146zap_add_by_dnode(dnode_t *dn, const char *key,
1147    int integer_size, uint64_t num_integers,
1148    const void *val, dmu_tx_t *tx)
1149{
1150	zap_t *zap;
1151	int err;
1152
1153	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1154	if (err != 0)
1155		return (err);
1156	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1157	/* zap_add_impl() calls zap_unlockdir() */
1158	return (err);
1159}
1160
1161int
1162zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1163    int key_numints, int integer_size, uint64_t num_integers,
1164    const void *val, dmu_tx_t *tx)
1165{
1166	zap_t *zap;
1167
1168	int err =
1169	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1170	if (err != 0)
1171		return (err);
1172	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1173	if (zn == NULL) {
1174		zap_unlockdir(zap, FTAG);
1175		return (SET_ERROR(ENOTSUP));
1176	}
1177	err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1178	zap = zn->zn_zap;	/* fzap_add() may change zap */
1179	zap_name_free(zn);
1180	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1181		zap_unlockdir(zap, FTAG);
1182	return (err);
1183}
1184
1185int
1186zap_update(objset_t *os, uint64_t zapobj, const char *name,
1187    int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1188{
1189	zap_t *zap;
1190	uint64_t oldval;
1191	const uint64_t *intval = val;
1192
1193#ifdef ZFS_DEBUG
1194	/*
1195	 * If there is an old value, it shouldn't change across the
1196	 * lockdir (eg, due to bprewrite's xlation).
1197	 */
1198	if (integer_size == 8 && num_integers == 1)
1199		(void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1200#endif
1201
1202	int err =
1203	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1204	if (err != 0)
1205		return (err);
1206	zap_name_t *zn = zap_name_alloc(zap, name, 0);
1207	if (zn == NULL) {
1208		zap_unlockdir(zap, FTAG);
1209		return (SET_ERROR(ENOTSUP));
1210	}
1211	if (!zap->zap_ismicro) {
1212		err = fzap_update(zn, integer_size, num_integers, val,
1213		    FTAG, tx);
1214		zap = zn->zn_zap;	/* fzap_update() may change zap */
1215	} else if (integer_size != 8 || num_integers != 1 ||
1216	    strlen(name) >= MZAP_NAME_LEN) {
1217		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1218		    zapobj, integer_size, num_integers, name);
1219		err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1220		if (err == 0) {
1221			err = fzap_update(zn, integer_size, num_integers,
1222			    val, FTAG, tx);
1223		}
1224		zap = zn->zn_zap;	/* fzap_update() may change zap */
1225	} else {
1226		mzap_ent_t *mze = mze_find(zn);
1227		if (mze != NULL) {
1228			ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1229			MZE_PHYS(zap, mze)->mze_value = *intval;
1230		} else {
1231			mzap_addent(zn, *intval);
1232		}
1233	}
1234	ASSERT(zap == zn->zn_zap);
1235	zap_name_free(zn);
1236	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1237		zap_unlockdir(zap, FTAG);
1238	return (err);
1239}
1240
1241int
1242zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1243    int key_numints,
1244    int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1245{
1246	zap_t *zap;
1247
1248	int err =
1249	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1250	if (err != 0)
1251		return (err);
1252	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1253	if (zn == NULL) {
1254		zap_unlockdir(zap, FTAG);
1255		return (SET_ERROR(ENOTSUP));
1256	}
1257	err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx);
1258	zap = zn->zn_zap;	/* fzap_update() may change zap */
1259	zap_name_free(zn);
1260	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1261		zap_unlockdir(zap, FTAG);
1262	return (err);
1263}
1264
1265int
1266zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1267{
1268	return (zap_remove_norm(os, zapobj, name, 0, tx));
1269}
1270
1271static int
1272zap_remove_impl(zap_t *zap, const char *name,
1273    matchtype_t mt, dmu_tx_t *tx)
1274{
1275	int err = 0;
1276
1277	zap_name_t *zn = zap_name_alloc(zap, name, mt);
1278	if (zn == NULL)
1279		return (SET_ERROR(ENOTSUP));
1280	if (!zap->zap_ismicro) {
1281		err = fzap_remove(zn, tx);
1282	} else {
1283		mzap_ent_t *mze = mze_find(zn);
1284		if (mze == NULL) {
1285			err = SET_ERROR(ENOENT);
1286		} else {
1287			zap->zap_m.zap_num_entries--;
1288			bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1289			    sizeof (mzap_ent_phys_t));
1290			mze_remove(zap, mze);
1291		}
1292	}
1293	zap_name_free(zn);
1294	return (err);
1295}
1296
1297int
1298zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1299    matchtype_t mt, dmu_tx_t *tx)
1300{
1301	zap_t *zap;
1302	int err;
1303
1304	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1305	if (err)
1306		return (err);
1307	err = zap_remove_impl(zap, name, mt, tx);
1308	zap_unlockdir(zap, FTAG);
1309	return (err);
1310}
1311
1312int
1313zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
1314{
1315	zap_t *zap;
1316	int err;
1317
1318	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1319	if (err)
1320		return (err);
1321	err = zap_remove_impl(zap, name, 0, tx);
1322	zap_unlockdir(zap, FTAG);
1323	return (err);
1324}
1325
1326int
1327zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1328    int key_numints, dmu_tx_t *tx)
1329{
1330	zap_t *zap;
1331
1332	int err =
1333	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1334	if (err != 0)
1335		return (err);
1336	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1337	if (zn == NULL) {
1338		zap_unlockdir(zap, FTAG);
1339		return (SET_ERROR(ENOTSUP));
1340	}
1341	err = fzap_remove(zn, tx);
1342	zap_name_free(zn);
1343	zap_unlockdir(zap, FTAG);
1344	return (err);
1345}
1346
1347/*
1348 * Routines for iterating over the attributes.
1349 */
1350
1351void
1352zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1353    uint64_t serialized)
1354{
1355	zc->zc_objset = os;
1356	zc->zc_zap = NULL;
1357	zc->zc_leaf = NULL;
1358	zc->zc_zapobj = zapobj;
1359	zc->zc_serialized = serialized;
1360	zc->zc_hash = 0;
1361	zc->zc_cd = 0;
1362}
1363
1364void
1365zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1366{
1367	zap_cursor_init_serialized(zc, os, zapobj, 0);
1368}
1369
1370void
1371zap_cursor_fini(zap_cursor_t *zc)
1372{
1373	if (zc->zc_zap) {
1374		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1375		zap_unlockdir(zc->zc_zap, NULL);
1376		zc->zc_zap = NULL;
1377	}
1378	if (zc->zc_leaf) {
1379		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1380		zap_put_leaf(zc->zc_leaf);
1381		zc->zc_leaf = NULL;
1382	}
1383	zc->zc_objset = NULL;
1384}
1385
1386uint64_t
1387zap_cursor_serialize(zap_cursor_t *zc)
1388{
1389	if (zc->zc_hash == -1ULL)
1390		return (-1ULL);
1391	if (zc->zc_zap == NULL)
1392		return (zc->zc_serialized);
1393	ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1394	ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1395
1396	/*
1397	 * We want to keep the high 32 bits of the cursor zero if we can, so
1398	 * that 32-bit programs can access this.  So usually use a small
1399	 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1400	 * of the cursor.
1401	 *
1402	 * [ collision differentiator | zap_hashbits()-bit hash value ]
1403	 */
1404	return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1405	    ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1406}
1407
1408int
1409zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1410{
1411	int err;
1412
1413	if (zc->zc_hash == -1ULL)
1414		return (SET_ERROR(ENOENT));
1415
1416	if (zc->zc_zap == NULL) {
1417		int hb;
1418		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1419		    RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1420		if (err != 0)
1421			return (err);
1422
1423		/*
1424		 * To support zap_cursor_init_serialized, advance, retrieve,
1425		 * we must add to the existing zc_cd, which may already
1426		 * be 1 due to the zap_cursor_advance.
1427		 */
1428		ASSERT(zc->zc_hash == 0);
1429		hb = zap_hashbits(zc->zc_zap);
1430		zc->zc_hash = zc->zc_serialized << (64 - hb);
1431		zc->zc_cd += zc->zc_serialized >> hb;
1432		if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1433			zc->zc_cd = 0;
1434	} else {
1435		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1436	}
1437	if (!zc->zc_zap->zap_ismicro) {
1438		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1439	} else {
1440		avl_index_t idx;
1441		mzap_ent_t mze_tofind;
1442
1443		mze_tofind.mze_hash = zc->zc_hash;
1444		mze_tofind.mze_cd = zc->zc_cd;
1445
1446		mzap_ent_t *mze =
1447		    avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1448		if (mze == NULL) {
1449			mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1450			    idx, AVL_AFTER);
1451		}
1452		if (mze) {
1453			mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1454			ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1455			za->za_normalization_conflict =
1456			    mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1457			za->za_integer_length = 8;
1458			za->za_num_integers = 1;
1459			za->za_first_integer = mzep->mze_value;
1460			(void) strcpy(za->za_name, mzep->mze_name);
1461			zc->zc_hash = mze->mze_hash;
1462			zc->zc_cd = mze->mze_cd;
1463			err = 0;
1464		} else {
1465			zc->zc_hash = -1ULL;
1466			err = SET_ERROR(ENOENT);
1467		}
1468	}
1469	rw_exit(&zc->zc_zap->zap_rwlock);
1470	return (err);
1471}
1472
1473void
1474zap_cursor_advance(zap_cursor_t *zc)
1475{
1476	if (zc->zc_hash == -1ULL)
1477		return;
1478	zc->zc_cd++;
1479}
1480
1481int
1482zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1483{
1484	int err = 0;
1485	mzap_ent_t *mze;
1486	zap_name_t *zn;
1487
1488	if (zc->zc_zap == NULL) {
1489		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1490		    RW_READER, TRUE, FALSE, FTAG, &zc->zc_zap);
1491		if (err)
1492			return (err);
1493	} else {
1494		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1495	}
1496
1497	zn = zap_name_alloc(zc->zc_zap, name, mt);
1498	if (zn == NULL) {
1499		rw_exit(&zc->zc_zap->zap_rwlock);
1500		return (SET_ERROR(ENOTSUP));
1501	}
1502
1503	if (!zc->zc_zap->zap_ismicro) {
1504		err = fzap_cursor_move_to_key(zc, zn);
1505	} else {
1506		mze = mze_find(zn);
1507		if (mze == NULL) {
1508			err = SET_ERROR(ENOENT);
1509			goto out;
1510		}
1511		zc->zc_hash = mze->mze_hash;
1512		zc->zc_cd = mze->mze_cd;
1513	}
1514
1515out:
1516	zap_name_free(zn);
1517	rw_exit(&zc->zc_zap->zap_rwlock);
1518	return (err);
1519}
1520
1521int
1522zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1523{
1524	zap_t *zap;
1525
1526	int err =
1527	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1528	if (err != 0)
1529		return (err);
1530
1531	bzero(zs, sizeof (zap_stats_t));
1532
1533	if (zap->zap_ismicro) {
1534		zs->zs_blocksize = zap->zap_dbuf->db_size;
1535		zs->zs_num_entries = zap->zap_m.zap_num_entries;
1536		zs->zs_num_blocks = 1;
1537	} else {
1538		fzap_get_stats(zap, zs);
1539	}
1540	zap_unlockdir(zap, FTAG);
1541	return (0);
1542}
1543