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, 2018 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_dnsize(os, obj, ot, bonustype, bonuslen,
705	    0, tx));
706}
707
708int
709zap_create_claim_dnsize(objset_t *os, uint64_t obj, dmu_object_type_t ot,
710    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
711{
712	return (zap_create_claim_norm_dnsize(os, obj,
713	    0, ot, bonustype, bonuslen, dnodesize, tx));
714}
715
716int
717zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
718    dmu_object_type_t ot,
719    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
720{
721	return (zap_create_claim_norm_dnsize(os, obj, normflags, ot, bonustype,
722	    bonuslen, 0, tx));
723}
724
725int
726zap_create_claim_norm_dnsize(objset_t *os, uint64_t obj, int normflags,
727    dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
728    int dnodesize, dmu_tx_t *tx)
729{
730	int err;
731
732	err = dmu_object_claim_dnsize(os, obj, ot, 0, bonustype, bonuslen,
733	    dnodesize, tx);
734	if (err != 0)
735		return (err);
736	mzap_create_impl(os, obj, normflags, 0, tx);
737	return (0);
738}
739
740uint64_t
741zap_create(objset_t *os, dmu_object_type_t ot,
742    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
743{
744	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
745}
746
747uint64_t
748zap_create_dnsize(objset_t *os, dmu_object_type_t ot,
749    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
750{
751	return (zap_create_norm_dnsize(os, 0, ot, bonustype, bonuslen,
752	    dnodesize, tx));
753}
754
755uint64_t
756zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
757    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
758{
759	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
760	return (zap_create_norm_dnsize(os, normflags, ot, bonustype, bonuslen,
761	    0, tx));
762}
763
764uint64_t
765zap_create_norm_dnsize(objset_t *os, int normflags, dmu_object_type_t ot,
766    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
767{
768	uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
769	    dnodesize, tx);
770
771	mzap_create_impl(os, obj, normflags, 0, tx);
772	return (obj);
773}
774
775uint64_t
776zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
777    dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
778    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
779{
780	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
781	return (zap_create_flags_dnsize(os, normflags, flags, ot,
782	    leaf_blockshift, indirect_blockshift, bonustype, bonuslen, 0, tx));
783}
784
785uint64_t
786zap_create_flags_dnsize(objset_t *os, int normflags, zap_flags_t flags,
787    dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
788    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
789{
790	uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
791	    dnodesize, tx);
792
793	ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
794	    leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
795	    indirect_blockshift >= SPA_MINBLOCKSHIFT &&
796	    indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
797
798	VERIFY(dmu_object_set_blocksize(os, obj,
799	    1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
800
801	mzap_create_impl(os, obj, normflags, flags, tx);
802	return (obj);
803}
804
805int
806zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
807{
808	/*
809	 * dmu_object_free will free the object number and free the
810	 * data.  Freeing the data will cause our pageout function to be
811	 * called, which will destroy our data (zap_leaf_t's and zap_t).
812	 */
813
814	return (dmu_object_free(os, zapobj, tx));
815}
816
817void
818zap_evict_sync(void *dbu)
819{
820	zap_t *zap = dbu;
821
822	rw_destroy(&zap->zap_rwlock);
823
824	if (zap->zap_ismicro)
825		mze_destroy(zap);
826	else
827		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
828
829	kmem_free(zap, sizeof (zap_t));
830}
831
832int
833zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
834{
835	zap_t *zap;
836
837	int err =
838	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
839	if (err != 0)
840		return (err);
841	if (!zap->zap_ismicro) {
842		err = fzap_count(zap, count);
843	} else {
844		*count = zap->zap_m.zap_num_entries;
845	}
846	zap_unlockdir(zap, FTAG);
847	return (err);
848}
849
850/*
851 * zn may be NULL; if not specified, it will be computed if needed.
852 * See also the comment above zap_entry_normalization_conflict().
853 */
854static boolean_t
855mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
856{
857	int direction = AVL_BEFORE;
858	boolean_t allocdzn = B_FALSE;
859
860	if (zap->zap_normflags == 0)
861		return (B_FALSE);
862
863again:
864	for (mzap_ent_t *other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
865	    other && other->mze_hash == mze->mze_hash;
866	    other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
867
868		if (zn == NULL) {
869			zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
870			    MT_NORMALIZE);
871			allocdzn = B_TRUE;
872		}
873		if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
874			if (allocdzn)
875				zap_name_free(zn);
876			return (B_TRUE);
877		}
878	}
879
880	if (direction == AVL_BEFORE) {
881		direction = AVL_AFTER;
882		goto again;
883	}
884
885	if (allocdzn)
886		zap_name_free(zn);
887	return (B_FALSE);
888}
889
890/*
891 * Routines for manipulating attributes.
892 */
893
894int
895zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
896    uint64_t integer_size, uint64_t num_integers, void *buf)
897{
898	return (zap_lookup_norm(os, zapobj, name, integer_size,
899	    num_integers, buf, 0, NULL, 0, NULL));
900}
901
902static int
903zap_lookup_impl(zap_t *zap, const char *name,
904    uint64_t integer_size, uint64_t num_integers, void *buf,
905    matchtype_t mt, char *realname, int rn_len,
906    boolean_t *ncp)
907{
908	int err = 0;
909
910	zap_name_t *zn = zap_name_alloc(zap, name, mt);
911	if (zn == NULL)
912		return (SET_ERROR(ENOTSUP));
913
914	if (!zap->zap_ismicro) {
915		err = fzap_lookup(zn, integer_size, num_integers, buf,
916		    realname, rn_len, ncp);
917	} else {
918		mzap_ent_t *mze = mze_find(zn);
919		if (mze == NULL) {
920			err = SET_ERROR(ENOENT);
921		} else {
922			if (num_integers < 1) {
923				err = SET_ERROR(EOVERFLOW);
924			} else if (integer_size != 8) {
925				err = SET_ERROR(EINVAL);
926			} else {
927				*(uint64_t *)buf =
928				    MZE_PHYS(zap, mze)->mze_value;
929				(void) strlcpy(realname,
930				    MZE_PHYS(zap, mze)->mze_name, rn_len);
931				if (ncp) {
932					*ncp = mzap_normalization_conflict(zap,
933					    zn, mze);
934				}
935			}
936		}
937	}
938	zap_name_free(zn);
939	return (err);
940}
941
942int
943zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
944    uint64_t integer_size, uint64_t num_integers, void *buf,
945    matchtype_t mt, char *realname, int rn_len,
946    boolean_t *ncp)
947{
948	zap_t *zap;
949
950	int err =
951	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
952	if (err != 0)
953		return (err);
954	err = zap_lookup_impl(zap, name, integer_size,
955	    num_integers, buf, mt, realname, rn_len, ncp);
956	zap_unlockdir(zap, FTAG);
957	return (err);
958}
959
960int
961zap_lookup_by_dnode(dnode_t *dn, const char *name,
962    uint64_t integer_size, uint64_t num_integers, void *buf)
963{
964	return (zap_lookup_norm_by_dnode(dn, name, integer_size,
965	    num_integers, buf, 0, NULL, 0, NULL));
966}
967
968int
969zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
970    uint64_t integer_size, uint64_t num_integers, void *buf,
971    matchtype_t mt, char *realname, int rn_len,
972    boolean_t *ncp)
973{
974	zap_t *zap;
975
976	int err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
977	    FTAG, &zap);
978	if (err != 0)
979		return (err);
980	err = zap_lookup_impl(zap, name, integer_size,
981	    num_integers, buf, mt, realname, rn_len, ncp);
982	zap_unlockdir(zap, FTAG);
983	return (err);
984}
985
986int
987zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
988    int key_numints)
989{
990	zap_t *zap;
991
992	int err =
993	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
994	if (err != 0)
995		return (err);
996	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
997	if (zn == NULL) {
998		zap_unlockdir(zap, FTAG);
999		return (SET_ERROR(ENOTSUP));
1000	}
1001
1002	fzap_prefetch(zn);
1003	zap_name_free(zn);
1004	zap_unlockdir(zap, FTAG);
1005	return (err);
1006}
1007
1008int
1009zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1010    int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
1011{
1012	zap_t *zap;
1013
1014	int err =
1015	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1016	if (err != 0)
1017		return (err);
1018	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1019	if (zn == NULL) {
1020		zap_unlockdir(zap, FTAG);
1021		return (SET_ERROR(ENOTSUP));
1022	}
1023
1024	err = fzap_lookup(zn, integer_size, num_integers, buf,
1025	    NULL, 0, NULL);
1026	zap_name_free(zn);
1027	zap_unlockdir(zap, FTAG);
1028	return (err);
1029}
1030
1031int
1032zap_contains(objset_t *os, uint64_t zapobj, const char *name)
1033{
1034	int err = zap_lookup_norm(os, zapobj, name, 0,
1035	    0, NULL, 0, NULL, 0, NULL);
1036	if (err == EOVERFLOW || err == EINVAL)
1037		err = 0; /* found, but skipped reading the value */
1038	return (err);
1039}
1040
1041int
1042zap_length(objset_t *os, uint64_t zapobj, const char *name,
1043    uint64_t *integer_size, uint64_t *num_integers)
1044{
1045	zap_t *zap;
1046
1047	int err =
1048	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1049	if (err != 0)
1050		return (err);
1051	zap_name_t *zn = zap_name_alloc(zap, name, 0);
1052	if (zn == NULL) {
1053		zap_unlockdir(zap, FTAG);
1054		return (SET_ERROR(ENOTSUP));
1055	}
1056	if (!zap->zap_ismicro) {
1057		err = fzap_length(zn, integer_size, num_integers);
1058	} else {
1059		mzap_ent_t *mze = mze_find(zn);
1060		if (mze == NULL) {
1061			err = SET_ERROR(ENOENT);
1062		} else {
1063			if (integer_size)
1064				*integer_size = 8;
1065			if (num_integers)
1066				*num_integers = 1;
1067		}
1068	}
1069	zap_name_free(zn);
1070	zap_unlockdir(zap, FTAG);
1071	return (err);
1072}
1073
1074int
1075zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1076    int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1077{
1078	zap_t *zap;
1079
1080	int err =
1081	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1082	if (err != 0)
1083		return (err);
1084	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1085	if (zn == NULL) {
1086		zap_unlockdir(zap, FTAG);
1087		return (SET_ERROR(ENOTSUP));
1088	}
1089	err = fzap_length(zn, integer_size, num_integers);
1090	zap_name_free(zn);
1091	zap_unlockdir(zap, FTAG);
1092	return (err);
1093}
1094
1095static void
1096mzap_addent(zap_name_t *zn, uint64_t value)
1097{
1098	zap_t *zap = zn->zn_zap;
1099	int start = zap->zap_m.zap_alloc_next;
1100
1101	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1102
1103#ifdef ZFS_DEBUG
1104	for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1105		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1106		ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1107	}
1108#endif
1109
1110	uint32_t cd = mze_find_unused_cd(zap, zn->zn_hash);
1111	/* given the limited size of the microzap, this can't happen */
1112	ASSERT(cd < zap_maxcd(zap));
1113
1114again:
1115	for (int i = start; i < zap->zap_m.zap_num_chunks; i++) {
1116		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1117		if (mze->mze_name[0] == 0) {
1118			mze->mze_value = value;
1119			mze->mze_cd = cd;
1120			(void) strcpy(mze->mze_name, zn->zn_key_orig);
1121			zap->zap_m.zap_num_entries++;
1122			zap->zap_m.zap_alloc_next = i+1;
1123			if (zap->zap_m.zap_alloc_next ==
1124			    zap->zap_m.zap_num_chunks)
1125				zap->zap_m.zap_alloc_next = 0;
1126			VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
1127			return;
1128		}
1129	}
1130	if (start != 0) {
1131		start = 0;
1132		goto again;
1133	}
1134	ASSERT(!"out of entries!");
1135}
1136
1137static int
1138zap_add_impl(zap_t *zap, const char *key,
1139    int integer_size, uint64_t num_integers,
1140    const void *val, dmu_tx_t *tx, void *tag)
1141{
1142	const uint64_t *intval = val;
1143	int err = 0;
1144
1145	zap_name_t *zn = zap_name_alloc(zap, key, 0);
1146	if (zn == NULL) {
1147		zap_unlockdir(zap, tag);
1148		return (SET_ERROR(ENOTSUP));
1149	}
1150	if (!zap->zap_ismicro) {
1151		err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1152		zap = zn->zn_zap;	/* fzap_add() may change zap */
1153	} else if (integer_size != 8 || num_integers != 1 ||
1154	    strlen(key) >= MZAP_NAME_LEN) {
1155		err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
1156		if (err == 0) {
1157			err = fzap_add(zn, integer_size, num_integers, val,
1158			    tag, tx);
1159		}
1160		zap = zn->zn_zap;	/* fzap_add() may change zap */
1161	} else {
1162		if (mze_find(zn) != NULL) {
1163			err = SET_ERROR(EEXIST);
1164		} else {
1165			mzap_addent(zn, *intval);
1166		}
1167	}
1168	ASSERT(zap == zn->zn_zap);
1169	zap_name_free(zn);
1170	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1171		zap_unlockdir(zap, tag);
1172	return (err);
1173}
1174
1175int
1176zap_add(objset_t *os, uint64_t zapobj, const char *key,
1177    int integer_size, uint64_t num_integers,
1178    const void *val, dmu_tx_t *tx)
1179{
1180	zap_t *zap;
1181	int err;
1182
1183	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1184	if (err != 0)
1185		return (err);
1186	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1187	/* zap_add_impl() calls zap_unlockdir() */
1188	return (err);
1189}
1190
1191int
1192zap_add_by_dnode(dnode_t *dn, const char *key,
1193    int integer_size, uint64_t num_integers,
1194    const void *val, dmu_tx_t *tx)
1195{
1196	zap_t *zap;
1197	int err;
1198
1199	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1200	if (err != 0)
1201		return (err);
1202	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1203	/* zap_add_impl() calls zap_unlockdir() */
1204	return (err);
1205}
1206
1207int
1208zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1209    int key_numints, int integer_size, uint64_t num_integers,
1210    const void *val, dmu_tx_t *tx)
1211{
1212	zap_t *zap;
1213
1214	int err =
1215	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1216	if (err != 0)
1217		return (err);
1218	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1219	if (zn == NULL) {
1220		zap_unlockdir(zap, FTAG);
1221		return (SET_ERROR(ENOTSUP));
1222	}
1223	err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1224	zap = zn->zn_zap;	/* fzap_add() may change zap */
1225	zap_name_free(zn);
1226	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1227		zap_unlockdir(zap, FTAG);
1228	return (err);
1229}
1230
1231int
1232zap_update(objset_t *os, uint64_t zapobj, const char *name,
1233    int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1234{
1235	zap_t *zap;
1236	uint64_t oldval;
1237	const uint64_t *intval = val;
1238
1239#ifdef ZFS_DEBUG
1240	/*
1241	 * If there is an old value, it shouldn't change across the
1242	 * lockdir (eg, due to bprewrite's xlation).
1243	 */
1244	if (integer_size == 8 && num_integers == 1)
1245		(void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1246#endif
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(zap, name, 0);
1253	if (zn == NULL) {
1254		zap_unlockdir(zap, FTAG);
1255		return (SET_ERROR(ENOTSUP));
1256	}
1257	if (!zap->zap_ismicro) {
1258		err = fzap_update(zn, integer_size, num_integers, val,
1259		    FTAG, tx);
1260		zap = zn->zn_zap;	/* fzap_update() may change zap */
1261	} else if (integer_size != 8 || num_integers != 1 ||
1262	    strlen(name) >= MZAP_NAME_LEN) {
1263		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1264		    zapobj, integer_size, num_integers, name);
1265		err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1266		if (err == 0) {
1267			err = fzap_update(zn, integer_size, num_integers,
1268			    val, FTAG, tx);
1269		}
1270		zap = zn->zn_zap;	/* fzap_update() may change zap */
1271	} else {
1272		mzap_ent_t *mze = mze_find(zn);
1273		if (mze != NULL) {
1274			ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1275			MZE_PHYS(zap, mze)->mze_value = *intval;
1276		} else {
1277			mzap_addent(zn, *intval);
1278		}
1279	}
1280	ASSERT(zap == zn->zn_zap);
1281	zap_name_free(zn);
1282	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1283		zap_unlockdir(zap, FTAG);
1284	return (err);
1285}
1286
1287int
1288zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1289    int key_numints,
1290    int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1291{
1292	zap_t *zap;
1293
1294	int err =
1295	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1296	if (err != 0)
1297		return (err);
1298	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1299	if (zn == NULL) {
1300		zap_unlockdir(zap, FTAG);
1301		return (SET_ERROR(ENOTSUP));
1302	}
1303	err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx);
1304	zap = zn->zn_zap;	/* fzap_update() may change zap */
1305	zap_name_free(zn);
1306	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1307		zap_unlockdir(zap, FTAG);
1308	return (err);
1309}
1310
1311int
1312zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1313{
1314	return (zap_remove_norm(os, zapobj, name, 0, tx));
1315}
1316
1317static int
1318zap_remove_impl(zap_t *zap, const char *name,
1319    matchtype_t mt, dmu_tx_t *tx)
1320{
1321	int err = 0;
1322
1323	zap_name_t *zn = zap_name_alloc(zap, name, mt);
1324	if (zn == NULL)
1325		return (SET_ERROR(ENOTSUP));
1326	if (!zap->zap_ismicro) {
1327		err = fzap_remove(zn, tx);
1328	} else {
1329		mzap_ent_t *mze = mze_find(zn);
1330		if (mze == NULL) {
1331			err = SET_ERROR(ENOENT);
1332		} else {
1333			zap->zap_m.zap_num_entries--;
1334			bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1335			    sizeof (mzap_ent_phys_t));
1336			mze_remove(zap, mze);
1337		}
1338	}
1339	zap_name_free(zn);
1340	return (err);
1341}
1342
1343int
1344zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1345    matchtype_t mt, dmu_tx_t *tx)
1346{
1347	zap_t *zap;
1348	int err;
1349
1350	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1351	if (err)
1352		return (err);
1353	err = zap_remove_impl(zap, name, mt, tx);
1354	zap_unlockdir(zap, FTAG);
1355	return (err);
1356}
1357
1358int
1359zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
1360{
1361	zap_t *zap;
1362	int err;
1363
1364	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1365	if (err)
1366		return (err);
1367	err = zap_remove_impl(zap, name, 0, tx);
1368	zap_unlockdir(zap, FTAG);
1369	return (err);
1370}
1371
1372int
1373zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1374    int key_numints, dmu_tx_t *tx)
1375{
1376	zap_t *zap;
1377
1378	int err =
1379	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1380	if (err != 0)
1381		return (err);
1382	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1383	if (zn == NULL) {
1384		zap_unlockdir(zap, FTAG);
1385		return (SET_ERROR(ENOTSUP));
1386	}
1387	err = fzap_remove(zn, tx);
1388	zap_name_free(zn);
1389	zap_unlockdir(zap, FTAG);
1390	return (err);
1391}
1392
1393/*
1394 * Routines for iterating over the attributes.
1395 */
1396
1397static void
1398zap_cursor_init_impl(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1399    uint64_t serialized, boolean_t prefetch)
1400{
1401	zc->zc_objset = os;
1402	zc->zc_zap = NULL;
1403	zc->zc_leaf = NULL;
1404	zc->zc_zapobj = zapobj;
1405	zc->zc_serialized = serialized;
1406	zc->zc_hash = 0;
1407	zc->zc_cd = 0;
1408	zc->zc_prefetch = prefetch;
1409}
1410void
1411zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1412    uint64_t serialized)
1413{
1414	zap_cursor_init_impl(zc, os, zapobj, serialized, B_TRUE);
1415}
1416
1417/*
1418 * Initialize a cursor at the beginning of the ZAP object.  The entire
1419 * ZAP object will be prefetched.
1420 */
1421void
1422zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1423{
1424	zap_cursor_init_impl(zc, os, zapobj, 0, B_TRUE);
1425}
1426
1427/*
1428 * Initialize a cursor at the beginning, but request that we not prefetch
1429 * the entire ZAP object.
1430 */
1431void
1432zap_cursor_init_noprefetch(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1433{
1434	zap_cursor_init_impl(zc, os, zapobj, 0, B_FALSE);
1435}
1436
1437void
1438zap_cursor_fini(zap_cursor_t *zc)
1439{
1440	if (zc->zc_zap) {
1441		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1442		zap_unlockdir(zc->zc_zap, NULL);
1443		zc->zc_zap = NULL;
1444	}
1445	if (zc->zc_leaf) {
1446		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1447		zap_put_leaf(zc->zc_leaf);
1448		zc->zc_leaf = NULL;
1449	}
1450	zc->zc_objset = NULL;
1451}
1452
1453uint64_t
1454zap_cursor_serialize(zap_cursor_t *zc)
1455{
1456	if (zc->zc_hash == -1ULL)
1457		return (-1ULL);
1458	if (zc->zc_zap == NULL)
1459		return (zc->zc_serialized);
1460	ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1461	ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1462
1463	/*
1464	 * We want to keep the high 32 bits of the cursor zero if we can, so
1465	 * that 32-bit programs can access this.  So usually use a small
1466	 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1467	 * of the cursor.
1468	 *
1469	 * [ collision differentiator | zap_hashbits()-bit hash value ]
1470	 */
1471	return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1472	    ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1473}
1474
1475int
1476zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1477{
1478	int err;
1479
1480	if (zc->zc_hash == -1ULL)
1481		return (SET_ERROR(ENOENT));
1482
1483	if (zc->zc_zap == NULL) {
1484		int hb;
1485		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1486		    RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1487		if (err != 0)
1488			return (err);
1489
1490		/*
1491		 * To support zap_cursor_init_serialized, advance, retrieve,
1492		 * we must add to the existing zc_cd, which may already
1493		 * be 1 due to the zap_cursor_advance.
1494		 */
1495		ASSERT(zc->zc_hash == 0);
1496		hb = zap_hashbits(zc->zc_zap);
1497		zc->zc_hash = zc->zc_serialized << (64 - hb);
1498		zc->zc_cd += zc->zc_serialized >> hb;
1499		if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1500			zc->zc_cd = 0;
1501	} else {
1502		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1503	}
1504	if (!zc->zc_zap->zap_ismicro) {
1505		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1506	} else {
1507		avl_index_t idx;
1508		mzap_ent_t mze_tofind;
1509
1510		mze_tofind.mze_hash = zc->zc_hash;
1511		mze_tofind.mze_cd = zc->zc_cd;
1512
1513		mzap_ent_t *mze =
1514		    avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1515		if (mze == NULL) {
1516			mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1517			    idx, AVL_AFTER);
1518		}
1519		if (mze) {
1520			mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1521			ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1522			za->za_normalization_conflict =
1523			    mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1524			za->za_integer_length = 8;
1525			za->za_num_integers = 1;
1526			za->za_first_integer = mzep->mze_value;
1527			(void) strcpy(za->za_name, mzep->mze_name);
1528			zc->zc_hash = mze->mze_hash;
1529			zc->zc_cd = mze->mze_cd;
1530			err = 0;
1531		} else {
1532			zc->zc_hash = -1ULL;
1533			err = SET_ERROR(ENOENT);
1534		}
1535	}
1536	rw_exit(&zc->zc_zap->zap_rwlock);
1537	return (err);
1538}
1539
1540void
1541zap_cursor_advance(zap_cursor_t *zc)
1542{
1543	if (zc->zc_hash == -1ULL)
1544		return;
1545	zc->zc_cd++;
1546}
1547
1548int
1549zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1550{
1551	int err = 0;
1552	mzap_ent_t *mze;
1553	zap_name_t *zn;
1554
1555	if (zc->zc_zap == NULL) {
1556		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1557		    RW_READER, TRUE, FALSE, FTAG, &zc->zc_zap);
1558		if (err)
1559			return (err);
1560	} else {
1561		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1562	}
1563
1564	zn = zap_name_alloc(zc->zc_zap, name, mt);
1565	if (zn == NULL) {
1566		rw_exit(&zc->zc_zap->zap_rwlock);
1567		return (SET_ERROR(ENOTSUP));
1568	}
1569
1570	if (!zc->zc_zap->zap_ismicro) {
1571		err = fzap_cursor_move_to_key(zc, zn);
1572	} else {
1573		mze = mze_find(zn);
1574		if (mze == NULL) {
1575			err = SET_ERROR(ENOENT);
1576			goto out;
1577		}
1578		zc->zc_hash = mze->mze_hash;
1579		zc->zc_cd = mze->mze_cd;
1580	}
1581
1582out:
1583	zap_name_free(zn);
1584	rw_exit(&zc->zc_zap->zap_rwlock);
1585	return (err);
1586}
1587
1588int
1589zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1590{
1591	zap_t *zap;
1592
1593	int err =
1594	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1595	if (err != 0)
1596		return (err);
1597
1598	bzero(zs, sizeof (zap_stats_t));
1599
1600	if (zap->zap_ismicro) {
1601		zs->zs_blocksize = zap->zap_dbuf->db_size;
1602		zs->zs_num_entries = zap->zap_m.zap_num_entries;
1603		zs->zs_num_blocks = 1;
1604	} else {
1605		fzap_get_stats(zap, zs);
1606	}
1607	zap_unlockdir(zap, FTAG);
1608	return (0);
1609}
1610