xref: /freebsd/sys/contrib/openzfs/module/zfs/zap_micro.c (revision 5b59b0c61e29f684a019afdd2848ffe2d5604e0c)
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 https://opensource.org/licenses/CDDL-1.0.
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 2017 Nexenta Systems, Inc.
27  * Copyright (c) 2024, Klara, 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/zap_impl.h>
36 #include <sys/zap_leaf.h>
37 #include <sys/btree.h>
38 #include <sys/arc.h>
39 #include <sys/dmu_objset.h>
40 #include <sys/spa_impl.h>
41 
42 #ifdef _KERNEL
43 #include <sys/sunddi.h>
44 #endif
45 
46 /*
47  * The maximum size (in bytes) of a microzap before it is converted to a
48  * fatzap. It will be rounded up to next multiple of 512 (SPA_MINBLOCKSIZE).
49  *
50  * By definition, a microzap must fit into a single block, so this has
51  * traditionally been SPA_OLD_MAXBLOCKSIZE, and is set to that by default.
52  * Setting this higher requires both the large_blocks feature (to even create
53  * blocks that large) and the large_microzap feature (to enable the stream
54  * machinery to understand not to try to split a microzap block).
55  *
56  * If large_microzap is enabled, this value will be clamped to
57  * spa_maxblocksize(). If not, it will be clamped to SPA_OLD_MAXBLOCKSIZE.
58  */
59 static int zap_micro_max_size = SPA_OLD_MAXBLOCKSIZE;
60 
61 uint64_t
62 zap_get_micro_max_size(spa_t *spa)
63 {
64 	uint64_t maxsz = P2ROUNDUP(zap_micro_max_size, SPA_MINBLOCKSIZE);
65 	if (maxsz <= SPA_OLD_MAXBLOCKSIZE)
66 		return (maxsz);
67 	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_MICROZAP))
68 		return (MIN(maxsz, spa_maxblocksize(spa)));
69 	return (SPA_OLD_MAXBLOCKSIZE);
70 }
71 
72 static int mzap_upgrade(zap_t **zapp,
73     const void *tag, dmu_tx_t *tx, zap_flags_t flags);
74 
75 uint64_t
76 zap_getflags(zap_t *zap)
77 {
78 	if (zap->zap_ismicro)
79 		return (0);
80 	return (zap_f_phys(zap)->zap_flags);
81 }
82 
83 int
84 zap_hashbits(zap_t *zap)
85 {
86 	if (zap_getflags(zap) & ZAP_FLAG_HASH64)
87 		return (48);
88 	else
89 		return (28);
90 }
91 
92 uint32_t
93 zap_maxcd(zap_t *zap)
94 {
95 	if (zap_getflags(zap) & ZAP_FLAG_HASH64)
96 		return ((1<<16)-1);
97 	else
98 		return (-1U);
99 }
100 
101 static uint64_t
102 zap_hash(zap_name_t *zn)
103 {
104 	zap_t *zap = zn->zn_zap;
105 	uint64_t h = 0;
106 
107 	if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
108 		ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
109 		h = *(uint64_t *)zn->zn_key_orig;
110 	} else {
111 		h = zap->zap_salt;
112 		ASSERT(h != 0);
113 		ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
114 
115 		if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
116 			const uint64_t *wp = zn->zn_key_norm;
117 
118 			ASSERT(zn->zn_key_intlen == 8);
119 			for (int i = 0; i < zn->zn_key_norm_numints;
120 			    wp++, i++) {
121 				uint64_t word = *wp;
122 
123 				for (int j = 0; j < 8; j++) {
124 					h = (h >> 8) ^
125 					    zfs_crc64_table[(h ^ word) & 0xFF];
126 					word >>= NBBY;
127 				}
128 			}
129 		} else {
130 			const uint8_t *cp = zn->zn_key_norm;
131 
132 			/*
133 			 * We previously stored the terminating null on
134 			 * disk, but didn't hash it, so we need to
135 			 * continue to not hash it.  (The
136 			 * zn_key_*_numints includes the terminating
137 			 * null for non-binary keys.)
138 			 */
139 			int len = zn->zn_key_norm_numints - 1;
140 
141 			ASSERT(zn->zn_key_intlen == 1);
142 			for (int i = 0; i < len; cp++, i++) {
143 				h = (h >> 8) ^
144 				    zfs_crc64_table[(h ^ *cp) & 0xFF];
145 			}
146 		}
147 	}
148 	/*
149 	 * Don't use all 64 bits, since we need some in the cookie for
150 	 * the collision differentiator.  We MUST use the high bits,
151 	 * since those are the ones that we first pay attention to when
152 	 * choosing the bucket.
153 	 */
154 	h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
155 
156 	return (h);
157 }
158 
159 static int
160 zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags,
161     size_t outlen)
162 {
163 	ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
164 
165 	size_t inlen = strlen(name) + 1;
166 
167 	int err = 0;
168 	(void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
169 	    normflags | U8_TEXTPREP_IGNORE_NULL | U8_TEXTPREP_IGNORE_INVALID,
170 	    U8_UNICODE_LATEST, &err);
171 
172 	return (err);
173 }
174 
175 boolean_t
176 zap_match(zap_name_t *zn, const char *matchname)
177 {
178 	boolean_t res = B_FALSE;
179 	ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
180 
181 	if (zn->zn_matchtype & MT_NORMALIZE) {
182 		size_t namelen = zn->zn_normbuf_len;
183 		char normbuf[ZAP_MAXNAMELEN];
184 		char *norm = normbuf;
185 
186 		/*
187 		 * Cannot allocate this on-stack as it exceed the stack-limit of
188 		 * 1024.
189 		 */
190 		if (namelen > ZAP_MAXNAMELEN)
191 			norm = kmem_alloc(namelen, KM_SLEEP);
192 
193 		if (zap_normalize(zn->zn_zap, matchname, norm,
194 		    zn->zn_normflags, namelen) != 0) {
195 			res = B_FALSE;
196 		} else {
197 			res = (strcmp(zn->zn_key_norm, norm) == 0);
198 		}
199 		if (norm != normbuf)
200 			kmem_free(norm, namelen);
201 	} else {
202 		res = (strcmp(zn->zn_key_orig, matchname) == 0);
203 	}
204 	return (res);
205 }
206 
207 static kmem_cache_t *zap_name_cache;
208 static kmem_cache_t *zap_attr_cache;
209 static kmem_cache_t *zap_name_long_cache;
210 static kmem_cache_t *zap_attr_long_cache;
211 
212 void
213 zap_init(void)
214 {
215 	zap_name_cache = kmem_cache_create("zap_name",
216 	    sizeof (zap_name_t) + ZAP_MAXNAMELEN, 0, NULL, NULL,
217 	    NULL, NULL, NULL, 0);
218 
219 	zap_attr_cache = kmem_cache_create("zap_attr_cache",
220 	    sizeof (zap_attribute_t) + ZAP_MAXNAMELEN,  0, NULL,
221 	    NULL, NULL, NULL, NULL, 0);
222 
223 	zap_name_long_cache = kmem_cache_create("zap_name_long",
224 	    sizeof (zap_name_t) + ZAP_MAXNAMELEN_NEW, 0, NULL, NULL,
225 	    NULL, NULL, NULL, 0);
226 
227 	zap_attr_long_cache = kmem_cache_create("zap_attr_long_cache",
228 	    sizeof (zap_attribute_t) + ZAP_MAXNAMELEN_NEW,  0, NULL,
229 	    NULL, NULL, NULL, NULL, 0);
230 }
231 
232 void
233 zap_fini(void)
234 {
235 	kmem_cache_destroy(zap_name_cache);
236 	kmem_cache_destroy(zap_attr_cache);
237 	kmem_cache_destroy(zap_name_long_cache);
238 	kmem_cache_destroy(zap_attr_long_cache);
239 }
240 
241 static zap_name_t *
242 zap_name_alloc(zap_t *zap, boolean_t longname)
243 {
244 	kmem_cache_t *cache = longname ? zap_name_long_cache : zap_name_cache;
245 	zap_name_t *zn = kmem_cache_alloc(cache, KM_SLEEP);
246 
247 	zn->zn_zap = zap;
248 	zn->zn_normbuf_len = longname ? ZAP_MAXNAMELEN_NEW : ZAP_MAXNAMELEN;
249 	return (zn);
250 }
251 
252 void
253 zap_name_free(zap_name_t *zn)
254 {
255 	if (zn->zn_normbuf_len == ZAP_MAXNAMELEN) {
256 		kmem_cache_free(zap_name_cache, zn);
257 	} else {
258 		ASSERT3U(zn->zn_normbuf_len, ==, ZAP_MAXNAMELEN_NEW);
259 		kmem_cache_free(zap_name_long_cache, zn);
260 	}
261 }
262 
263 static int
264 zap_name_init_str(zap_name_t *zn, const char *key, matchtype_t mt)
265 {
266 	zap_t *zap = zn->zn_zap;
267 	size_t key_len = strlen(key) + 1;
268 
269 	/* Make sure zn is allocated for longname if key is long */
270 	IMPLY(key_len > ZAP_MAXNAMELEN,
271 	    zn->zn_normbuf_len == ZAP_MAXNAMELEN_NEW);
272 
273 	zn->zn_key_intlen = sizeof (*key);
274 	zn->zn_key_orig = key;
275 	zn->zn_key_orig_numints = key_len;
276 	zn->zn_matchtype = mt;
277 	zn->zn_normflags = zap->zap_normflags;
278 
279 	/*
280 	 * If we're dealing with a case sensitive lookup on a mixed or
281 	 * insensitive fs, remove U8_TEXTPREP_TOUPPER or the lookup
282 	 * will fold case to all caps overriding the lookup request.
283 	 */
284 	if (mt & MT_MATCH_CASE)
285 		zn->zn_normflags &= ~U8_TEXTPREP_TOUPPER;
286 
287 	if (zap->zap_normflags) {
288 		/*
289 		 * We *must* use zap_normflags because this normalization is
290 		 * what the hash is computed from.
291 		 */
292 		if (zap_normalize(zap, key, zn->zn_normbuf,
293 		    zap->zap_normflags, zn->zn_normbuf_len) != 0)
294 			return (SET_ERROR(ENOTSUP));
295 		zn->zn_key_norm = zn->zn_normbuf;
296 		zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
297 	} else {
298 		if (mt != 0)
299 			return (SET_ERROR(ENOTSUP));
300 		zn->zn_key_norm = zn->zn_key_orig;
301 		zn->zn_key_norm_numints = zn->zn_key_orig_numints;
302 	}
303 
304 	zn->zn_hash = zap_hash(zn);
305 
306 	if (zap->zap_normflags != zn->zn_normflags) {
307 		/*
308 		 * We *must* use zn_normflags because this normalization is
309 		 * what the matching is based on.  (Not the hash!)
310 		 */
311 		if (zap_normalize(zap, key, zn->zn_normbuf,
312 		    zn->zn_normflags, zn->zn_normbuf_len) != 0)
313 			return (SET_ERROR(ENOTSUP));
314 		zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
315 	}
316 
317 	return (0);
318 }
319 
320 zap_name_t *
321 zap_name_alloc_str(zap_t *zap, const char *key, matchtype_t mt)
322 {
323 	size_t key_len = strlen(key) + 1;
324 	zap_name_t *zn = zap_name_alloc(zap, (key_len > ZAP_MAXNAMELEN));
325 	if (zap_name_init_str(zn, key, mt) != 0) {
326 		zap_name_free(zn);
327 		return (NULL);
328 	}
329 	return (zn);
330 }
331 
332 static zap_name_t *
333 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
334 {
335 	zap_name_t *zn = kmem_cache_alloc(zap_name_cache, KM_SLEEP);
336 
337 	ASSERT(zap->zap_normflags == 0);
338 	zn->zn_zap = zap;
339 	zn->zn_key_intlen = sizeof (*key);
340 	zn->zn_key_orig = zn->zn_key_norm = key;
341 	zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
342 	zn->zn_matchtype = 0;
343 	zn->zn_normbuf_len = ZAP_MAXNAMELEN;
344 
345 	zn->zn_hash = zap_hash(zn);
346 	return (zn);
347 }
348 
349 static void
350 mzap_byteswap(mzap_phys_t *buf, size_t size)
351 {
352 	buf->mz_block_type = BSWAP_64(buf->mz_block_type);
353 	buf->mz_salt = BSWAP_64(buf->mz_salt);
354 	buf->mz_normflags = BSWAP_64(buf->mz_normflags);
355 	int max = (size / MZAP_ENT_LEN) - 1;
356 	for (int i = 0; i < max; i++) {
357 		buf->mz_chunk[i].mze_value =
358 		    BSWAP_64(buf->mz_chunk[i].mze_value);
359 		buf->mz_chunk[i].mze_cd =
360 		    BSWAP_32(buf->mz_chunk[i].mze_cd);
361 	}
362 }
363 
364 void
365 zap_byteswap(void *buf, size_t size)
366 {
367 	uint64_t block_type = *(uint64_t *)buf;
368 
369 	if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
370 		/* ASSERT(magic == ZAP_LEAF_MAGIC); */
371 		mzap_byteswap(buf, size);
372 	} else {
373 		fzap_byteswap(buf, size);
374 	}
375 }
376 
377 __attribute__((always_inline)) inline
378 static int
379 mze_compare(const void *arg1, const void *arg2)
380 {
381 	const mzap_ent_t *mze1 = arg1;
382 	const mzap_ent_t *mze2 = arg2;
383 
384 	return (TREE_CMP((uint64_t)(mze1->mze_hash) << 32 | mze1->mze_cd,
385 	    (uint64_t)(mze2->mze_hash) << 32 | mze2->mze_cd));
386 }
387 
388 ZFS_BTREE_FIND_IN_BUF_FUNC(mze_find_in_buf, mzap_ent_t,
389     mze_compare)
390 
391 static void
392 mze_insert(zap_t *zap, uint16_t chunkid, uint64_t hash)
393 {
394 	mzap_ent_t mze;
395 
396 	ASSERT(zap->zap_ismicro);
397 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
398 
399 	mze.mze_chunkid = chunkid;
400 	ASSERT0(hash & 0xffffffff);
401 	mze.mze_hash = hash >> 32;
402 	ASSERT3U(MZE_PHYS(zap, &mze)->mze_cd, <=, 0xffff);
403 	mze.mze_cd = (uint16_t)MZE_PHYS(zap, &mze)->mze_cd;
404 	ASSERT(MZE_PHYS(zap, &mze)->mze_name[0] != 0);
405 	zfs_btree_add(&zap->zap_m.zap_tree, &mze);
406 }
407 
408 static mzap_ent_t *
409 mze_find(zap_name_t *zn, zfs_btree_index_t *idx)
410 {
411 	mzap_ent_t mze_tofind;
412 	mzap_ent_t *mze;
413 	zfs_btree_t *tree = &zn->zn_zap->zap_m.zap_tree;
414 
415 	ASSERT(zn->zn_zap->zap_ismicro);
416 	ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
417 
418 	ASSERT0(zn->zn_hash & 0xffffffff);
419 	mze_tofind.mze_hash = zn->zn_hash >> 32;
420 	mze_tofind.mze_cd = 0;
421 
422 	mze = zfs_btree_find(tree, &mze_tofind, idx);
423 	if (mze == NULL)
424 		mze = zfs_btree_next(tree, idx, idx);
425 	for (; mze && mze->mze_hash == mze_tofind.mze_hash;
426 	    mze = zfs_btree_next(tree, idx, idx)) {
427 		ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
428 		if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
429 			return (mze);
430 	}
431 
432 	return (NULL);
433 }
434 
435 static uint32_t
436 mze_find_unused_cd(zap_t *zap, uint64_t hash)
437 {
438 	mzap_ent_t mze_tofind;
439 	zfs_btree_index_t idx;
440 	zfs_btree_t *tree = &zap->zap_m.zap_tree;
441 
442 	ASSERT(zap->zap_ismicro);
443 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
444 
445 	ASSERT0(hash & 0xffffffff);
446 	hash >>= 32;
447 	mze_tofind.mze_hash = hash;
448 	mze_tofind.mze_cd = 0;
449 
450 	uint32_t cd = 0;
451 	for (mzap_ent_t *mze = zfs_btree_find(tree, &mze_tofind, &idx);
452 	    mze && mze->mze_hash == hash;
453 	    mze = zfs_btree_next(tree, &idx, &idx)) {
454 		if (mze->mze_cd != cd)
455 			break;
456 		cd++;
457 	}
458 
459 	return (cd);
460 }
461 
462 /*
463  * Each mzap entry requires at max : 4 chunks
464  * 3 chunks for names + 1 chunk for value.
465  */
466 #define	MZAP_ENT_CHUNKS	(1 + ZAP_LEAF_ARRAY_NCHUNKS(MZAP_NAME_LEN) + \
467 	ZAP_LEAF_ARRAY_NCHUNKS(sizeof (uint64_t)))
468 
469 /*
470  * Check if the current entry keeps the colliding entries under the fatzap leaf
471  * size.
472  */
473 static boolean_t
474 mze_canfit_fzap_leaf(zap_name_t *zn, uint64_t hash)
475 {
476 	zap_t *zap = zn->zn_zap;
477 	mzap_ent_t mze_tofind;
478 	zfs_btree_index_t idx;
479 	zfs_btree_t *tree = &zap->zap_m.zap_tree;
480 	uint32_t mzap_ents = 0;
481 
482 	ASSERT0(hash & 0xffffffff);
483 	hash >>= 32;
484 	mze_tofind.mze_hash = hash;
485 	mze_tofind.mze_cd = 0;
486 
487 	for (mzap_ent_t *mze = zfs_btree_find(tree, &mze_tofind, &idx);
488 	    mze && mze->mze_hash == hash;
489 	    mze = zfs_btree_next(tree, &idx, &idx)) {
490 		mzap_ents++;
491 	}
492 
493 	/* Include the new entry being added */
494 	mzap_ents++;
495 
496 	return (ZAP_LEAF_NUMCHUNKS_DEF > (mzap_ents * MZAP_ENT_CHUNKS));
497 }
498 
499 static void
500 mze_destroy(zap_t *zap)
501 {
502 	zfs_btree_clear(&zap->zap_m.zap_tree);
503 	zfs_btree_destroy(&zap->zap_m.zap_tree);
504 }
505 
506 static zap_t *
507 mzap_open(dmu_buf_t *db)
508 {
509 	zap_t *winner;
510 	uint64_t *zap_hdr = (uint64_t *)db->db_data;
511 	uint64_t zap_block_type = zap_hdr[0];
512 	uint64_t zap_magic = zap_hdr[1];
513 
514 	ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
515 
516 	zap_t *zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
517 	rw_init(&zap->zap_rwlock, NULL, RW_DEFAULT, NULL);
518 	rw_enter(&zap->zap_rwlock, RW_WRITER);
519 	zap->zap_objset = dmu_buf_get_objset(db);
520 	zap->zap_object = db->db_object;
521 	zap->zap_dbuf = db;
522 
523 	if (zap_block_type != ZBT_MICRO) {
524 		mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, MUTEX_DEFAULT,
525 		    0);
526 		zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
527 		if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
528 			winner = NULL;	/* No actual winner here... */
529 			goto handle_winner;
530 		}
531 	} else {
532 		zap->zap_ismicro = TRUE;
533 	}
534 
535 	/*
536 	 * Make sure that zap_ismicro is set before we let others see
537 	 * it, because zap_lockdir() checks zap_ismicro without the lock
538 	 * held.
539 	 */
540 	dmu_buf_init_user(&zap->zap_dbu, zap_evict_sync, NULL, &zap->zap_dbuf);
541 	winner = dmu_buf_set_user(db, &zap->zap_dbu);
542 
543 	if (winner != NULL)
544 		goto handle_winner;
545 
546 	if (zap->zap_ismicro) {
547 		zap->zap_salt = zap_m_phys(zap)->mz_salt;
548 		zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
549 		zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
550 
551 		/*
552 		 * Reduce B-tree leaf from 4KB to 512 bytes to reduce memmove()
553 		 * overhead on massive inserts below.  It still allows to store
554 		 * 62 entries before we have to add 2KB B-tree core node.
555 		 */
556 		zfs_btree_create_custom(&zap->zap_m.zap_tree, mze_compare,
557 		    mze_find_in_buf, sizeof (mzap_ent_t), 512);
558 
559 		zap_name_t *zn = zap_name_alloc(zap, B_FALSE);
560 		for (uint16_t i = 0; i < zap->zap_m.zap_num_chunks; i++) {
561 			mzap_ent_phys_t *mze =
562 			    &zap_m_phys(zap)->mz_chunk[i];
563 			if (mze->mze_name[0]) {
564 				zap->zap_m.zap_num_entries++;
565 				zap_name_init_str(zn, mze->mze_name, 0);
566 				mze_insert(zap, i, zn->zn_hash);
567 			}
568 		}
569 		zap_name_free(zn);
570 	} else {
571 		zap->zap_salt = zap_f_phys(zap)->zap_salt;
572 		zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
573 
574 		ASSERT3U(sizeof (struct zap_leaf_header), ==,
575 		    2*ZAP_LEAF_CHUNKSIZE);
576 
577 		/*
578 		 * The embedded pointer table should not overlap the
579 		 * other members.
580 		 */
581 		ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
582 		    &zap_f_phys(zap)->zap_salt);
583 
584 		/*
585 		 * The embedded pointer table should end at the end of
586 		 * the block
587 		 */
588 		ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
589 		    1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
590 		    (uintptr_t)zap_f_phys(zap), ==,
591 		    zap->zap_dbuf->db_size);
592 	}
593 	rw_exit(&zap->zap_rwlock);
594 	return (zap);
595 
596 handle_winner:
597 	rw_exit(&zap->zap_rwlock);
598 	rw_destroy(&zap->zap_rwlock);
599 	if (!zap->zap_ismicro)
600 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
601 	kmem_free(zap, sizeof (zap_t));
602 	return (winner);
603 }
604 
605 /*
606  * This routine "consumes" the caller's hold on the dbuf, which must
607  * have the specified tag.
608  */
609 static int
610 zap_lockdir_impl(dnode_t *dn, dmu_buf_t *db, const void *tag, dmu_tx_t *tx,
611     krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
612 {
613 	ASSERT0(db->db_offset);
614 	objset_t *os = dmu_buf_get_objset(db);
615 	uint64_t obj = db->db_object;
616 	dmu_object_info_t doi;
617 
618 	*zapp = NULL;
619 
620 	dmu_object_info_from_dnode(dn, &doi);
621 	if (DMU_OT_BYTESWAP(doi.doi_type) != DMU_BSWAP_ZAP)
622 		return (SET_ERROR(EINVAL));
623 
624 	zap_t *zap = dmu_buf_get_user(db);
625 	if (zap == NULL) {
626 		zap = mzap_open(db);
627 		if (zap == NULL) {
628 			/*
629 			 * mzap_open() didn't like what it saw on-disk.
630 			 * Check for corruption!
631 			 */
632 			return (SET_ERROR(EIO));
633 		}
634 	}
635 
636 	/*
637 	 * We're checking zap_ismicro without the lock held, in order to
638 	 * tell what type of lock we want.  Once we have some sort of
639 	 * lock, see if it really is the right type.  In practice this
640 	 * can only be different if it was upgraded from micro to fat,
641 	 * and micro wanted WRITER but fat only needs READER.
642 	 */
643 	krw_t lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
644 	rw_enter(&zap->zap_rwlock, lt);
645 	if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
646 		/* it was upgraded, now we only need reader */
647 		ASSERT(lt == RW_WRITER);
648 		ASSERT(RW_READER ==
649 		    ((!zap->zap_ismicro && fatreader) ? RW_READER : lti));
650 		rw_downgrade(&zap->zap_rwlock);
651 		lt = RW_READER;
652 	}
653 
654 	zap->zap_objset = os;
655 	zap->zap_dnode = dn;
656 
657 	if (lt == RW_WRITER)
658 		dmu_buf_will_dirty(db, tx);
659 
660 	ASSERT3P(zap->zap_dbuf, ==, db);
661 
662 	ASSERT(!zap->zap_ismicro ||
663 	    zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
664 	if (zap->zap_ismicro && tx && adding &&
665 	    zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
666 		uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
667 		if (newsz > zap_get_micro_max_size(dmu_objset_spa(os))) {
668 			dprintf("upgrading obj %llu: num_entries=%u\n",
669 			    (u_longlong_t)obj, zap->zap_m.zap_num_entries);
670 			*zapp = zap;
671 			int err = mzap_upgrade(zapp, tag, tx, 0);
672 			if (err != 0)
673 				rw_exit(&zap->zap_rwlock);
674 			return (err);
675 		}
676 		VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx));
677 		zap->zap_m.zap_num_chunks =
678 		    db->db_size / MZAP_ENT_LEN - 1;
679 
680 		if (newsz > SPA_OLD_MAXBLOCKSIZE) {
681 			dsl_dataset_t *ds = dmu_objset_ds(os);
682 			if (!dsl_dataset_feature_is_active(ds,
683 			    SPA_FEATURE_LARGE_MICROZAP)) {
684 				/*
685 				 * A microzap just grew beyond the old limit
686 				 * for the first time, so we have to ensure the
687 				 * feature flag is activated.
688 				 * zap_get_micro_max_size() won't let us get
689 				 * here if the feature is not enabled, so we
690 				 * don't need any other checks beforehand.
691 				 *
692 				 * Since we're in open context, we can't
693 				 * activate the feature directly, so we instead
694 				 * flag it on the dataset for next sync.
695 				 */
696 				dsl_dataset_dirty(ds, tx);
697 				mutex_enter(&ds->ds_lock);
698 				ds->ds_feature_activation
699 				    [SPA_FEATURE_LARGE_MICROZAP] =
700 				    (void *)B_TRUE;
701 				mutex_exit(&ds->ds_lock);
702 			}
703 		}
704 	}
705 
706 	*zapp = zap;
707 	return (0);
708 }
709 
710 static int
711 zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
712     krw_t lti, boolean_t fatreader, boolean_t adding, const void *tag,
713     zap_t **zapp)
714 {
715 	dmu_buf_t *db;
716 	int err;
717 
718 	err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
719 	if (err != 0)
720 		return (err);
721 	err = zap_lockdir_impl(dn, db, tag, tx, lti, fatreader, adding, zapp);
722 	if (err != 0)
723 		dmu_buf_rele(db, tag);
724 	else
725 		VERIFY(dnode_add_ref(dn, tag));
726 	return (err);
727 }
728 
729 int
730 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
731     krw_t lti, boolean_t fatreader, boolean_t adding, const void *tag,
732     zap_t **zapp)
733 {
734 	dnode_t *dn;
735 	dmu_buf_t *db;
736 	int err;
737 
738 	err = dnode_hold(os, obj, tag, &dn);
739 	if (err != 0)
740 		return (err);
741 	err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
742 	if (err != 0) {
743 		dnode_rele(dn, tag);
744 		return (err);
745 	}
746 	err = zap_lockdir_impl(dn, db, tag, tx, lti, fatreader, adding, zapp);
747 	if (err != 0) {
748 		dmu_buf_rele(db, tag);
749 		dnode_rele(dn, tag);
750 	}
751 	return (err);
752 }
753 
754 void
755 zap_unlockdir(zap_t *zap, const void *tag)
756 {
757 	rw_exit(&zap->zap_rwlock);
758 	dnode_rele(zap->zap_dnode, tag);
759 	dmu_buf_rele(zap->zap_dbuf, tag);
760 }
761 
762 static int
763 mzap_upgrade(zap_t **zapp, const void *tag, dmu_tx_t *tx, zap_flags_t flags)
764 {
765 	int err = 0;
766 	zap_t *zap = *zapp;
767 
768 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
769 
770 	int sz = zap->zap_dbuf->db_size;
771 	mzap_phys_t *mzp = vmem_alloc(sz, KM_SLEEP);
772 	memcpy(mzp, zap->zap_dbuf->db_data, sz);
773 	int nchunks = zap->zap_m.zap_num_chunks;
774 
775 	if (!flags) {
776 		err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
777 		    1ULL << fzap_default_block_shift, 0, tx);
778 		if (err != 0) {
779 			vmem_free(mzp, sz);
780 			return (err);
781 		}
782 	}
783 
784 	dprintf("upgrading obj=%llu with %u chunks\n",
785 	    (u_longlong_t)zap->zap_object, nchunks);
786 	/* XXX destroy the tree later, so we can use the stored hash value */
787 	mze_destroy(zap);
788 
789 	fzap_upgrade(zap, tx, flags);
790 
791 	zap_name_t *zn = zap_name_alloc(zap, B_FALSE);
792 	for (int i = 0; i < nchunks; i++) {
793 		mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
794 		if (mze->mze_name[0] == 0)
795 			continue;
796 		dprintf("adding %s=%llu\n",
797 		    mze->mze_name, (u_longlong_t)mze->mze_value);
798 		zap_name_init_str(zn, mze->mze_name, 0);
799 		/* If we fail here, we would end up losing entries */
800 		VERIFY0(fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
801 		    tag, tx));
802 		zap = zn->zn_zap;	/* fzap_add_cd() may change zap */
803 	}
804 	zap_name_free(zn);
805 	vmem_free(mzp, sz);
806 	*zapp = zap;
807 	return (0);
808 }
809 
810 /*
811  * The "normflags" determine the behavior of the matchtype_t which is
812  * passed to zap_lookup_norm().  Names which have the same normalized
813  * version will be stored with the same hash value, and therefore we can
814  * perform normalization-insensitive lookups.  We can be Unicode form-
815  * insensitive and/or case-insensitive.  The following flags are valid for
816  * "normflags":
817  *
818  * U8_TEXTPREP_NFC
819  * U8_TEXTPREP_NFD
820  * U8_TEXTPREP_NFKC
821  * U8_TEXTPREP_NFKD
822  * U8_TEXTPREP_TOUPPER
823  *
824  * The *_NF* (Normalization Form) flags are mutually exclusive; at most one
825  * of them may be supplied.
826  */
827 void
828 mzap_create_impl(dnode_t *dn, int normflags, zap_flags_t flags, dmu_tx_t *tx)
829 {
830 	dmu_buf_t *db;
831 
832 	VERIFY0(dmu_buf_hold_by_dnode(dn, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
833 
834 	dmu_buf_will_dirty(db, tx);
835 	mzap_phys_t *zp = db->db_data;
836 	zp->mz_block_type = ZBT_MICRO;
837 	zp->mz_salt =
838 	    ((uintptr_t)db ^ (uintptr_t)tx ^ (dn->dn_object << 1)) | 1ULL;
839 	zp->mz_normflags = normflags;
840 
841 	if (flags != 0) {
842 		zap_t *zap;
843 		/* Only fat zap supports flags; upgrade immediately. */
844 		VERIFY(dnode_add_ref(dn, FTAG));
845 		VERIFY0(zap_lockdir_impl(dn, db, FTAG, tx, RW_WRITER,
846 		    B_FALSE, B_FALSE, &zap));
847 		VERIFY0(mzap_upgrade(&zap, FTAG, tx, flags));
848 		zap_unlockdir(zap, FTAG);
849 	} else {
850 		dmu_buf_rele(db, FTAG);
851 	}
852 }
853 
854 static uint64_t
855 zap_create_impl(objset_t *os, int normflags, zap_flags_t flags,
856     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
857     dmu_object_type_t bonustype, int bonuslen, int dnodesize,
858     dnode_t **allocated_dnode, const void *tag, dmu_tx_t *tx)
859 {
860 	uint64_t obj;
861 
862 	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
863 
864 	if (allocated_dnode == NULL) {
865 		dnode_t *dn;
866 		obj = dmu_object_alloc_hold(os, ot, 1ULL << leaf_blockshift,
867 		    indirect_blockshift, bonustype, bonuslen, dnodesize,
868 		    &dn, FTAG, tx);
869 		mzap_create_impl(dn, normflags, flags, tx);
870 		dnode_rele(dn, FTAG);
871 	} else {
872 		obj = dmu_object_alloc_hold(os, ot, 1ULL << leaf_blockshift,
873 		    indirect_blockshift, bonustype, bonuslen, dnodesize,
874 		    allocated_dnode, tag, tx);
875 		mzap_create_impl(*allocated_dnode, normflags, flags, tx);
876 	}
877 
878 	return (obj);
879 }
880 
881 int
882 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
883     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
884 {
885 	return (zap_create_claim_dnsize(os, obj, ot, bonustype, bonuslen,
886 	    0, tx));
887 }
888 
889 int
890 zap_create_claim_dnsize(objset_t *os, uint64_t obj, dmu_object_type_t ot,
891     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
892 {
893 	return (zap_create_claim_norm_dnsize(os, obj,
894 	    0, ot, bonustype, bonuslen, dnodesize, tx));
895 }
896 
897 int
898 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
899     dmu_object_type_t ot,
900     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
901 {
902 	return (zap_create_claim_norm_dnsize(os, obj, normflags, ot, bonustype,
903 	    bonuslen, 0, tx));
904 }
905 
906 int
907 zap_create_claim_norm_dnsize(objset_t *os, uint64_t obj, int normflags,
908     dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
909     int dnodesize, dmu_tx_t *tx)
910 {
911 	dnode_t *dn;
912 	int error;
913 
914 	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
915 	error = dmu_object_claim_dnsize(os, obj, ot, 0, bonustype, bonuslen,
916 	    dnodesize, tx);
917 	if (error != 0)
918 		return (error);
919 
920 	error = dnode_hold(os, obj, FTAG, &dn);
921 	if (error != 0)
922 		return (error);
923 
924 	mzap_create_impl(dn, normflags, 0, tx);
925 
926 	dnode_rele(dn, FTAG);
927 
928 	return (0);
929 }
930 
931 uint64_t
932 zap_create(objset_t *os, dmu_object_type_t ot,
933     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
934 {
935 	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
936 }
937 
938 uint64_t
939 zap_create_dnsize(objset_t *os, dmu_object_type_t ot,
940     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
941 {
942 	return (zap_create_norm_dnsize(os, 0, ot, bonustype, bonuslen,
943 	    dnodesize, tx));
944 }
945 
946 uint64_t
947 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
948     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
949 {
950 	return (zap_create_norm_dnsize(os, normflags, ot, bonustype, bonuslen,
951 	    0, tx));
952 }
953 
954 uint64_t
955 zap_create_norm_dnsize(objset_t *os, int normflags, dmu_object_type_t ot,
956     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
957 {
958 	return (zap_create_impl(os, normflags, 0, ot, 0, 0,
959 	    bonustype, bonuslen, dnodesize, NULL, NULL, tx));
960 }
961 
962 uint64_t
963 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
964     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
965     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
966 {
967 	return (zap_create_flags_dnsize(os, normflags, flags, ot,
968 	    leaf_blockshift, indirect_blockshift, bonustype, bonuslen, 0, tx));
969 }
970 
971 uint64_t
972 zap_create_flags_dnsize(objset_t *os, int normflags, zap_flags_t flags,
973     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
974     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
975 {
976 	return (zap_create_impl(os, normflags, flags, ot, leaf_blockshift,
977 	    indirect_blockshift, bonustype, bonuslen, dnodesize, NULL, NULL,
978 	    tx));
979 }
980 
981 /*
982  * Create a zap object and return a pointer to the newly allocated dnode via
983  * the allocated_dnode argument.  The returned dnode will be held and the
984  * caller is responsible for releasing the hold by calling dnode_rele().
985  */
986 uint64_t
987 zap_create_hold(objset_t *os, int normflags, zap_flags_t flags,
988     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
989     dmu_object_type_t bonustype, int bonuslen, int dnodesize,
990     dnode_t **allocated_dnode, const void *tag, dmu_tx_t *tx)
991 {
992 	return (zap_create_impl(os, normflags, flags, ot, leaf_blockshift,
993 	    indirect_blockshift, bonustype, bonuslen, dnodesize,
994 	    allocated_dnode, tag, tx));
995 }
996 
997 int
998 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
999 {
1000 	/*
1001 	 * dmu_object_free will free the object number and free the
1002 	 * data.  Freeing the data will cause our pageout function to be
1003 	 * called, which will destroy our data (zap_leaf_t's and zap_t).
1004 	 */
1005 
1006 	return (dmu_object_free(os, zapobj, tx));
1007 }
1008 
1009 void
1010 zap_evict_sync(void *dbu)
1011 {
1012 	zap_t *zap = dbu;
1013 
1014 	rw_destroy(&zap->zap_rwlock);
1015 
1016 	if (zap->zap_ismicro)
1017 		mze_destroy(zap);
1018 	else
1019 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
1020 
1021 	kmem_free(zap, sizeof (zap_t));
1022 }
1023 
1024 int
1025 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
1026 {
1027 	zap_t *zap;
1028 
1029 	int err =
1030 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1031 	if (err != 0)
1032 		return (err);
1033 	if (!zap->zap_ismicro) {
1034 		err = fzap_count(zap, count);
1035 	} else {
1036 		*count = zap->zap_m.zap_num_entries;
1037 	}
1038 	zap_unlockdir(zap, FTAG);
1039 	return (err);
1040 }
1041 
1042 /*
1043  * zn may be NULL; if not specified, it will be computed if needed.
1044  * See also the comment above zap_entry_normalization_conflict().
1045  */
1046 static boolean_t
1047 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze,
1048     zfs_btree_index_t *idx)
1049 {
1050 	boolean_t allocdzn = B_FALSE;
1051 	mzap_ent_t *other;
1052 	zfs_btree_index_t oidx;
1053 
1054 	if (zap->zap_normflags == 0)
1055 		return (B_FALSE);
1056 
1057 	for (other = zfs_btree_prev(&zap->zap_m.zap_tree, idx, &oidx);
1058 	    other && other->mze_hash == mze->mze_hash;
1059 	    other = zfs_btree_prev(&zap->zap_m.zap_tree, &oidx, &oidx)) {
1060 
1061 		if (zn == NULL) {
1062 			zn = zap_name_alloc_str(zap,
1063 			    MZE_PHYS(zap, mze)->mze_name, MT_NORMALIZE);
1064 			allocdzn = B_TRUE;
1065 		}
1066 		if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
1067 			if (allocdzn)
1068 				zap_name_free(zn);
1069 			return (B_TRUE);
1070 		}
1071 	}
1072 
1073 	for (other = zfs_btree_next(&zap->zap_m.zap_tree, idx, &oidx);
1074 	    other && other->mze_hash == mze->mze_hash;
1075 	    other = zfs_btree_next(&zap->zap_m.zap_tree, &oidx, &oidx)) {
1076 
1077 		if (zn == NULL) {
1078 			zn = zap_name_alloc_str(zap,
1079 			    MZE_PHYS(zap, mze)->mze_name, MT_NORMALIZE);
1080 			allocdzn = B_TRUE;
1081 		}
1082 		if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
1083 			if (allocdzn)
1084 				zap_name_free(zn);
1085 			return (B_TRUE);
1086 		}
1087 	}
1088 
1089 	if (allocdzn)
1090 		zap_name_free(zn);
1091 	return (B_FALSE);
1092 }
1093 
1094 /*
1095  * Routines for manipulating attributes.
1096  */
1097 
1098 int
1099 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
1100     uint64_t integer_size, uint64_t num_integers, void *buf)
1101 {
1102 	return (zap_lookup_norm(os, zapobj, name, integer_size,
1103 	    num_integers, buf, 0, NULL, 0, NULL));
1104 }
1105 
1106 static int
1107 zap_lookup_impl(zap_t *zap, const char *name,
1108     uint64_t integer_size, uint64_t num_integers, void *buf,
1109     matchtype_t mt, char *realname, int rn_len,
1110     boolean_t *ncp)
1111 {
1112 	int err = 0;
1113 
1114 	zap_name_t *zn = zap_name_alloc_str(zap, name, mt);
1115 	if (zn == NULL)
1116 		return (SET_ERROR(ENOTSUP));
1117 
1118 	if (!zap->zap_ismicro) {
1119 		err = fzap_lookup(zn, integer_size, num_integers, buf,
1120 		    realname, rn_len, ncp);
1121 	} else {
1122 		zfs_btree_index_t idx;
1123 		mzap_ent_t *mze = mze_find(zn, &idx);
1124 		if (mze == NULL) {
1125 			err = SET_ERROR(ENOENT);
1126 		} else {
1127 			if (num_integers < 1) {
1128 				err = SET_ERROR(EOVERFLOW);
1129 			} else if (integer_size != 8) {
1130 				err = SET_ERROR(EINVAL);
1131 			} else {
1132 				*(uint64_t *)buf =
1133 				    MZE_PHYS(zap, mze)->mze_value;
1134 				if (realname != NULL)
1135 					(void) strlcpy(realname,
1136 					    MZE_PHYS(zap, mze)->mze_name,
1137 					    rn_len);
1138 				if (ncp) {
1139 					*ncp = mzap_normalization_conflict(zap,
1140 					    zn, mze, &idx);
1141 				}
1142 			}
1143 		}
1144 	}
1145 	zap_name_free(zn);
1146 	return (err);
1147 }
1148 
1149 int
1150 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
1151     uint64_t integer_size, uint64_t num_integers, void *buf,
1152     matchtype_t mt, char *realname, int rn_len,
1153     boolean_t *ncp)
1154 {
1155 	zap_t *zap;
1156 
1157 	int err =
1158 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1159 	if (err != 0)
1160 		return (err);
1161 	err = zap_lookup_impl(zap, name, integer_size,
1162 	    num_integers, buf, mt, realname, rn_len, ncp);
1163 	zap_unlockdir(zap, FTAG);
1164 	return (err);
1165 }
1166 
1167 int
1168 zap_prefetch(objset_t *os, uint64_t zapobj, const char *name)
1169 {
1170 	zap_t *zap;
1171 	int err;
1172 	zap_name_t *zn;
1173 
1174 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1175 	if (err)
1176 		return (err);
1177 	zn = zap_name_alloc_str(zap, name, 0);
1178 	if (zn == NULL) {
1179 		zap_unlockdir(zap, FTAG);
1180 		return (SET_ERROR(ENOTSUP));
1181 	}
1182 
1183 	fzap_prefetch(zn);
1184 	zap_name_free(zn);
1185 	zap_unlockdir(zap, FTAG);
1186 	return (err);
1187 }
1188 
1189 int
1190 zap_prefetch_object(objset_t *os, uint64_t zapobj)
1191 {
1192 	int error;
1193 	dmu_object_info_t doi;
1194 
1195 	error = dmu_object_info(os, zapobj, &doi);
1196 	if (error == 0 && DMU_OT_BYTESWAP(doi.doi_type) != DMU_BSWAP_ZAP)
1197 		error = SET_ERROR(EINVAL);
1198 	if (error == 0)
1199 		dmu_prefetch_wait(os, zapobj, 0, doi.doi_max_offset);
1200 
1201 	return (error);
1202 }
1203 
1204 int
1205 zap_lookup_by_dnode(dnode_t *dn, const char *name,
1206     uint64_t integer_size, uint64_t num_integers, void *buf)
1207 {
1208 	return (zap_lookup_norm_by_dnode(dn, name, integer_size,
1209 	    num_integers, buf, 0, NULL, 0, NULL));
1210 }
1211 
1212 int
1213 zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
1214     uint64_t integer_size, uint64_t num_integers, void *buf,
1215     matchtype_t mt, char *realname, int rn_len,
1216     boolean_t *ncp)
1217 {
1218 	zap_t *zap;
1219 
1220 	int err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
1221 	    FTAG, &zap);
1222 	if (err != 0)
1223 		return (err);
1224 	err = zap_lookup_impl(zap, name, integer_size,
1225 	    num_integers, buf, mt, realname, rn_len, ncp);
1226 	zap_unlockdir(zap, FTAG);
1227 	return (err);
1228 }
1229 
1230 static int
1231 zap_prefetch_uint64_impl(zap_t *zap, const uint64_t *key, int key_numints)
1232 {
1233 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1234 	if (zn == NULL) {
1235 		zap_unlockdir(zap, FTAG);
1236 		return (SET_ERROR(ENOTSUP));
1237 	}
1238 
1239 	fzap_prefetch(zn);
1240 	zap_name_free(zn);
1241 	zap_unlockdir(zap, FTAG);
1242 	return (0);
1243 }
1244 
1245 int
1246 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1247     int key_numints)
1248 {
1249 	zap_t *zap;
1250 
1251 	int err =
1252 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1253 	if (err != 0)
1254 		return (err);
1255 	err = zap_prefetch_uint64_impl(zap, key, key_numints);
1256 	/* zap_prefetch_uint64_impl() calls zap_unlockdir() */
1257 	return (err);
1258 }
1259 
1260 int
1261 zap_prefetch_uint64_by_dnode(dnode_t *dn, const uint64_t *key, int key_numints)
1262 {
1263 	zap_t *zap;
1264 
1265 	int err =
1266 	    zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1267 	if (err != 0)
1268 		return (err);
1269 	err = zap_prefetch_uint64_impl(zap, key, key_numints);
1270 	/* zap_prefetch_uint64_impl() calls zap_unlockdir() */
1271 	return (err);
1272 }
1273 
1274 static int
1275 zap_lookup_uint64_impl(zap_t *zap, const uint64_t *key,
1276     int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
1277 {
1278 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1279 	if (zn == NULL) {
1280 		zap_unlockdir(zap, FTAG);
1281 		return (SET_ERROR(ENOTSUP));
1282 	}
1283 
1284 	int err = fzap_lookup(zn, integer_size, num_integers, buf,
1285 	    NULL, 0, NULL);
1286 	zap_name_free(zn);
1287 	zap_unlockdir(zap, FTAG);
1288 	return (err);
1289 }
1290 
1291 int
1292 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1293     int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
1294 {
1295 	zap_t *zap;
1296 
1297 	int err =
1298 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1299 	if (err != 0)
1300 		return (err);
1301 	err = zap_lookup_uint64_impl(zap, key, key_numints, integer_size,
1302 	    num_integers, buf);
1303 	/* zap_lookup_uint64_impl() calls zap_unlockdir() */
1304 	return (err);
1305 }
1306 
1307 int
1308 zap_lookup_uint64_by_dnode(dnode_t *dn, const uint64_t *key,
1309     int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
1310 {
1311 	zap_t *zap;
1312 
1313 	int err =
1314 	    zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1315 	if (err != 0)
1316 		return (err);
1317 	err = zap_lookup_uint64_impl(zap, key, key_numints, integer_size,
1318 	    num_integers, buf);
1319 	/* zap_lookup_uint64_impl() calls zap_unlockdir() */
1320 	return (err);
1321 }
1322 
1323 int
1324 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
1325 {
1326 	int err = zap_lookup_norm(os, zapobj, name, 0,
1327 	    0, NULL, 0, NULL, 0, NULL);
1328 	if (err == EOVERFLOW || err == EINVAL)
1329 		err = 0; /* found, but skipped reading the value */
1330 	return (err);
1331 }
1332 
1333 int
1334 zap_length(objset_t *os, uint64_t zapobj, const char *name,
1335     uint64_t *integer_size, uint64_t *num_integers)
1336 {
1337 	zap_t *zap;
1338 
1339 	int err =
1340 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1341 	if (err != 0)
1342 		return (err);
1343 	zap_name_t *zn = zap_name_alloc_str(zap, name, 0);
1344 	if (zn == NULL) {
1345 		zap_unlockdir(zap, FTAG);
1346 		return (SET_ERROR(ENOTSUP));
1347 	}
1348 	if (!zap->zap_ismicro) {
1349 		err = fzap_length(zn, integer_size, num_integers);
1350 	} else {
1351 		zfs_btree_index_t idx;
1352 		mzap_ent_t *mze = mze_find(zn, &idx);
1353 		if (mze == NULL) {
1354 			err = SET_ERROR(ENOENT);
1355 		} else {
1356 			if (integer_size)
1357 				*integer_size = 8;
1358 			if (num_integers)
1359 				*num_integers = 1;
1360 		}
1361 	}
1362 	zap_name_free(zn);
1363 	zap_unlockdir(zap, FTAG);
1364 	return (err);
1365 }
1366 
1367 int
1368 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1369     int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1370 {
1371 	zap_t *zap;
1372 
1373 	int err =
1374 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1375 	if (err != 0)
1376 		return (err);
1377 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1378 	if (zn == NULL) {
1379 		zap_unlockdir(zap, FTAG);
1380 		return (SET_ERROR(ENOTSUP));
1381 	}
1382 	err = fzap_length(zn, integer_size, num_integers);
1383 	zap_name_free(zn);
1384 	zap_unlockdir(zap, FTAG);
1385 	return (err);
1386 }
1387 
1388 static void
1389 mzap_addent(zap_name_t *zn, uint64_t value)
1390 {
1391 	zap_t *zap = zn->zn_zap;
1392 	uint16_t start = zap->zap_m.zap_alloc_next;
1393 
1394 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1395 
1396 #ifdef ZFS_DEBUG
1397 	for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1398 		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1399 		ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1400 	}
1401 #endif
1402 
1403 	uint32_t cd = mze_find_unused_cd(zap, zn->zn_hash);
1404 	/* given the limited size of the microzap, this can't happen */
1405 	ASSERT(cd < zap_maxcd(zap));
1406 
1407 again:
1408 	for (uint16_t i = start; i < zap->zap_m.zap_num_chunks; i++) {
1409 		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1410 		if (mze->mze_name[0] == 0) {
1411 			mze->mze_value = value;
1412 			mze->mze_cd = cd;
1413 			(void) strlcpy(mze->mze_name, zn->zn_key_orig,
1414 			    sizeof (mze->mze_name));
1415 			zap->zap_m.zap_num_entries++;
1416 			zap->zap_m.zap_alloc_next = i+1;
1417 			if (zap->zap_m.zap_alloc_next ==
1418 			    zap->zap_m.zap_num_chunks)
1419 				zap->zap_m.zap_alloc_next = 0;
1420 			mze_insert(zap, i, zn->zn_hash);
1421 			return;
1422 		}
1423 	}
1424 	if (start != 0) {
1425 		start = 0;
1426 		goto again;
1427 	}
1428 	cmn_err(CE_PANIC, "out of entries!");
1429 }
1430 
1431 static int
1432 zap_add_impl(zap_t *zap, const char *key,
1433     int integer_size, uint64_t num_integers,
1434     const void *val, dmu_tx_t *tx, const void *tag)
1435 {
1436 	const uint64_t *intval = val;
1437 	int err = 0;
1438 
1439 	zap_name_t *zn = zap_name_alloc_str(zap, key, 0);
1440 	if (zn == NULL) {
1441 		zap_unlockdir(zap, tag);
1442 		return (SET_ERROR(ENOTSUP));
1443 	}
1444 	if (!zap->zap_ismicro) {
1445 		err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1446 		zap = zn->zn_zap;	/* fzap_add() may change zap */
1447 	} else if (integer_size != 8 || num_integers != 1 ||
1448 	    strlen(key) >= MZAP_NAME_LEN ||
1449 	    !mze_canfit_fzap_leaf(zn, zn->zn_hash)) {
1450 		err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
1451 		if (err == 0) {
1452 			err = fzap_add(zn, integer_size, num_integers, val,
1453 			    tag, tx);
1454 		}
1455 		zap = zn->zn_zap;	/* fzap_add() may change zap */
1456 	} else {
1457 		zfs_btree_index_t idx;
1458 		if (mze_find(zn, &idx) != NULL) {
1459 			err = SET_ERROR(EEXIST);
1460 		} else {
1461 			mzap_addent(zn, *intval);
1462 		}
1463 	}
1464 	ASSERT(zap == zn->zn_zap);
1465 	zap_name_free(zn);
1466 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1467 		zap_unlockdir(zap, tag);
1468 	return (err);
1469 }
1470 
1471 int
1472 zap_add(objset_t *os, uint64_t zapobj, const char *key,
1473     int integer_size, uint64_t num_integers,
1474     const void *val, dmu_tx_t *tx)
1475 {
1476 	zap_t *zap;
1477 	int err;
1478 
1479 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1480 	if (err != 0)
1481 		return (err);
1482 	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1483 	/* zap_add_impl() calls zap_unlockdir() */
1484 	return (err);
1485 }
1486 
1487 int
1488 zap_add_by_dnode(dnode_t *dn, const char *key,
1489     int integer_size, uint64_t num_integers,
1490     const void *val, dmu_tx_t *tx)
1491 {
1492 	zap_t *zap;
1493 	int err;
1494 
1495 	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1496 	if (err != 0)
1497 		return (err);
1498 	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1499 	/* zap_add_impl() calls zap_unlockdir() */
1500 	return (err);
1501 }
1502 
1503 static int
1504 zap_add_uint64_impl(zap_t *zap, const uint64_t *key,
1505     int key_numints, int integer_size, uint64_t num_integers,
1506     const void *val, dmu_tx_t *tx, const void *tag)
1507 {
1508 	int err;
1509 
1510 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1511 	if (zn == NULL) {
1512 		zap_unlockdir(zap, tag);
1513 		return (SET_ERROR(ENOTSUP));
1514 	}
1515 	err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1516 	zap = zn->zn_zap;	/* fzap_add() may change zap */
1517 	zap_name_free(zn);
1518 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1519 		zap_unlockdir(zap, tag);
1520 	return (err);
1521 }
1522 
1523 int
1524 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1525     int key_numints, int integer_size, uint64_t num_integers,
1526     const void *val, dmu_tx_t *tx)
1527 {
1528 	zap_t *zap;
1529 
1530 	int err =
1531 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1532 	if (err != 0)
1533 		return (err);
1534 	err = zap_add_uint64_impl(zap, key, key_numints,
1535 	    integer_size, num_integers, val, tx, FTAG);
1536 	/* zap_add_uint64_impl() calls zap_unlockdir() */
1537 	return (err);
1538 }
1539 
1540 int
1541 zap_add_uint64_by_dnode(dnode_t *dn, const uint64_t *key,
1542     int key_numints, int integer_size, uint64_t num_integers,
1543     const void *val, dmu_tx_t *tx)
1544 {
1545 	zap_t *zap;
1546 
1547 	int err =
1548 	    zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1549 	if (err != 0)
1550 		return (err);
1551 	err = zap_add_uint64_impl(zap, key, key_numints,
1552 	    integer_size, num_integers, val, tx, FTAG);
1553 	/* zap_add_uint64_impl() calls zap_unlockdir() */
1554 	return (err);
1555 }
1556 
1557 int
1558 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1559     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1560 {
1561 	zap_t *zap;
1562 	const uint64_t *intval = val;
1563 
1564 	int err =
1565 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1566 	if (err != 0)
1567 		return (err);
1568 	zap_name_t *zn = zap_name_alloc_str(zap, name, 0);
1569 	if (zn == NULL) {
1570 		zap_unlockdir(zap, FTAG);
1571 		return (SET_ERROR(ENOTSUP));
1572 	}
1573 	if (!zap->zap_ismicro) {
1574 		err = fzap_update(zn, integer_size, num_integers, val,
1575 		    FTAG, tx);
1576 		zap = zn->zn_zap;	/* fzap_update() may change zap */
1577 	} else if (integer_size != 8 || num_integers != 1 ||
1578 	    strlen(name) >= MZAP_NAME_LEN) {
1579 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1580 		    (u_longlong_t)zapobj, integer_size,
1581 		    (u_longlong_t)num_integers, name);
1582 		err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1583 		if (err == 0) {
1584 			err = fzap_update(zn, integer_size, num_integers,
1585 			    val, FTAG, tx);
1586 		}
1587 		zap = zn->zn_zap;	/* fzap_update() may change zap */
1588 	} else {
1589 		zfs_btree_index_t idx;
1590 		mzap_ent_t *mze = mze_find(zn, &idx);
1591 		if (mze != NULL) {
1592 			MZE_PHYS(zap, mze)->mze_value = *intval;
1593 		} else {
1594 			mzap_addent(zn, *intval);
1595 		}
1596 	}
1597 	ASSERT(zap == zn->zn_zap);
1598 	zap_name_free(zn);
1599 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1600 		zap_unlockdir(zap, FTAG);
1601 	return (err);
1602 }
1603 
1604 static int
1605 zap_update_uint64_impl(zap_t *zap, const uint64_t *key, int key_numints,
1606     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx,
1607     const void *tag)
1608 {
1609 	int err;
1610 
1611 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1612 	if (zn == NULL) {
1613 		zap_unlockdir(zap, tag);
1614 		return (SET_ERROR(ENOTSUP));
1615 	}
1616 	err = fzap_update(zn, integer_size, num_integers, val, tag, tx);
1617 	zap = zn->zn_zap;	/* fzap_update() may change zap */
1618 	zap_name_free(zn);
1619 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1620 		zap_unlockdir(zap, tag);
1621 	return (err);
1622 }
1623 
1624 int
1625 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1626     int key_numints, int integer_size, uint64_t num_integers, const void *val,
1627     dmu_tx_t *tx)
1628 {
1629 	zap_t *zap;
1630 
1631 	int err =
1632 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1633 	if (err != 0)
1634 		return (err);
1635 	err = zap_update_uint64_impl(zap, key, key_numints,
1636 	    integer_size, num_integers, val, tx, FTAG);
1637 	/* zap_update_uint64_impl() calls zap_unlockdir() */
1638 	return (err);
1639 }
1640 
1641 int
1642 zap_update_uint64_by_dnode(dnode_t *dn, const uint64_t *key, int key_numints,
1643     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1644 {
1645 	zap_t *zap;
1646 
1647 	int err =
1648 	    zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1649 	if (err != 0)
1650 		return (err);
1651 	err = zap_update_uint64_impl(zap, key, key_numints,
1652 	    integer_size, num_integers, val, tx, FTAG);
1653 	/* zap_update_uint64_impl() calls zap_unlockdir() */
1654 	return (err);
1655 }
1656 
1657 int
1658 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1659 {
1660 	return (zap_remove_norm(os, zapobj, name, 0, tx));
1661 }
1662 
1663 static int
1664 zap_remove_impl(zap_t *zap, const char *name,
1665     matchtype_t mt, dmu_tx_t *tx)
1666 {
1667 	int err = 0;
1668 
1669 	zap_name_t *zn = zap_name_alloc_str(zap, name, mt);
1670 	if (zn == NULL)
1671 		return (SET_ERROR(ENOTSUP));
1672 	if (!zap->zap_ismicro) {
1673 		err = fzap_remove(zn, tx);
1674 	} else {
1675 		zfs_btree_index_t idx;
1676 		mzap_ent_t *mze = mze_find(zn, &idx);
1677 		if (mze == NULL) {
1678 			err = SET_ERROR(ENOENT);
1679 		} else {
1680 			zap->zap_m.zap_num_entries--;
1681 			memset(MZE_PHYS(zap, mze), 0, sizeof (mzap_ent_phys_t));
1682 			zfs_btree_remove_idx(&zap->zap_m.zap_tree, &idx);
1683 		}
1684 	}
1685 	zap_name_free(zn);
1686 	return (err);
1687 }
1688 
1689 int
1690 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1691     matchtype_t mt, dmu_tx_t *tx)
1692 {
1693 	zap_t *zap;
1694 	int err;
1695 
1696 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1697 	if (err)
1698 		return (err);
1699 	err = zap_remove_impl(zap, name, mt, tx);
1700 	zap_unlockdir(zap, FTAG);
1701 	return (err);
1702 }
1703 
1704 int
1705 zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
1706 {
1707 	zap_t *zap;
1708 	int err;
1709 
1710 	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1711 	if (err)
1712 		return (err);
1713 	err = zap_remove_impl(zap, name, 0, tx);
1714 	zap_unlockdir(zap, FTAG);
1715 	return (err);
1716 }
1717 
1718 static int
1719 zap_remove_uint64_impl(zap_t *zap, const uint64_t *key, int key_numints,
1720     dmu_tx_t *tx, const void *tag)
1721 {
1722 	int err;
1723 
1724 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1725 	if (zn == NULL) {
1726 		zap_unlockdir(zap, tag);
1727 		return (SET_ERROR(ENOTSUP));
1728 	}
1729 	err = fzap_remove(zn, tx);
1730 	zap_name_free(zn);
1731 	zap_unlockdir(zap, tag);
1732 	return (err);
1733 }
1734 
1735 int
1736 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1737     int key_numints, dmu_tx_t *tx)
1738 {
1739 	zap_t *zap;
1740 
1741 	int err =
1742 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1743 	if (err != 0)
1744 		return (err);
1745 	err = zap_remove_uint64_impl(zap, key, key_numints, tx, FTAG);
1746 	/* zap_remove_uint64_impl() calls zap_unlockdir() */
1747 	return (err);
1748 }
1749 
1750 int
1751 zap_remove_uint64_by_dnode(dnode_t *dn, const uint64_t *key, int key_numints,
1752     dmu_tx_t *tx)
1753 {
1754 	zap_t *zap;
1755 
1756 	int err =
1757 	    zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1758 	if (err != 0)
1759 		return (err);
1760 	err = zap_remove_uint64_impl(zap, key, key_numints, tx, FTAG);
1761 	/* zap_remove_uint64_impl() calls zap_unlockdir() */
1762 	return (err);
1763 }
1764 
1765 
1766 static zap_attribute_t *
1767 zap_attribute_alloc_impl(boolean_t longname)
1768 {
1769 	zap_attribute_t *za;
1770 
1771 	za = kmem_cache_alloc((longname)? zap_attr_long_cache : zap_attr_cache,
1772 	    KM_SLEEP);
1773 	za->za_name_len = (longname)? ZAP_MAXNAMELEN_NEW : ZAP_MAXNAMELEN;
1774 	return (za);
1775 }
1776 
1777 zap_attribute_t *
1778 zap_attribute_alloc(void)
1779 {
1780 	return (zap_attribute_alloc_impl(B_FALSE));
1781 }
1782 
1783 zap_attribute_t *
1784 zap_attribute_long_alloc(void)
1785 {
1786 	return (zap_attribute_alloc_impl(B_TRUE));
1787 }
1788 
1789 void
1790 zap_attribute_free(zap_attribute_t *za)
1791 {
1792 	if (za->za_name_len == ZAP_MAXNAMELEN) {
1793 		kmem_cache_free(zap_attr_cache, za);
1794 	} else {
1795 		ASSERT3U(za->za_name_len, ==, ZAP_MAXNAMELEN_NEW);
1796 		kmem_cache_free(zap_attr_long_cache, za);
1797 	}
1798 }
1799 
1800 /*
1801  * Routines for iterating over the attributes.
1802  */
1803 
1804 static void
1805 zap_cursor_init_impl(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1806     uint64_t serialized, boolean_t prefetch)
1807 {
1808 	zc->zc_objset = os;
1809 	zc->zc_zap = NULL;
1810 	zc->zc_leaf = NULL;
1811 	zc->zc_zapobj = zapobj;
1812 	zc->zc_serialized = serialized;
1813 	zc->zc_hash = 0;
1814 	zc->zc_cd = 0;
1815 	zc->zc_prefetch = prefetch;
1816 }
1817 void
1818 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1819     uint64_t serialized)
1820 {
1821 	zap_cursor_init_impl(zc, os, zapobj, serialized, B_TRUE);
1822 }
1823 
1824 /*
1825  * Initialize a cursor at the beginning of the ZAP object.  The entire
1826  * ZAP object will be prefetched.
1827  */
1828 void
1829 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1830 {
1831 	zap_cursor_init_impl(zc, os, zapobj, 0, B_TRUE);
1832 }
1833 
1834 /*
1835  * Initialize a cursor at the beginning, but request that we not prefetch
1836  * the entire ZAP object.
1837  */
1838 void
1839 zap_cursor_init_noprefetch(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1840 {
1841 	zap_cursor_init_impl(zc, os, zapobj, 0, B_FALSE);
1842 }
1843 
1844 void
1845 zap_cursor_fini(zap_cursor_t *zc)
1846 {
1847 	if (zc->zc_zap) {
1848 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1849 		zap_unlockdir(zc->zc_zap, NULL);
1850 		zc->zc_zap = NULL;
1851 	}
1852 	if (zc->zc_leaf) {
1853 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1854 		zap_put_leaf(zc->zc_leaf);
1855 		zc->zc_leaf = NULL;
1856 	}
1857 	zc->zc_objset = NULL;
1858 }
1859 
1860 uint64_t
1861 zap_cursor_serialize(zap_cursor_t *zc)
1862 {
1863 	if (zc->zc_hash == -1ULL)
1864 		return (-1ULL);
1865 	if (zc->zc_zap == NULL)
1866 		return (zc->zc_serialized);
1867 	ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1868 	ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1869 
1870 	/*
1871 	 * We want to keep the high 32 bits of the cursor zero if we can, so
1872 	 * that 32-bit programs can access this.  So usually use a small
1873 	 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1874 	 * of the cursor.
1875 	 *
1876 	 * [ collision differentiator | zap_hashbits()-bit hash value ]
1877 	 */
1878 	return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1879 	    ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1880 }
1881 
1882 int
1883 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1884 {
1885 	int err;
1886 
1887 	if (zc->zc_hash == -1ULL)
1888 		return (SET_ERROR(ENOENT));
1889 
1890 	if (zc->zc_zap == NULL) {
1891 		int hb;
1892 		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1893 		    RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1894 		if (err != 0)
1895 			return (err);
1896 
1897 		/*
1898 		 * To support zap_cursor_init_serialized, advance, retrieve,
1899 		 * we must add to the existing zc_cd, which may already
1900 		 * be 1 due to the zap_cursor_advance.
1901 		 */
1902 		ASSERT(zc->zc_hash == 0);
1903 		hb = zap_hashbits(zc->zc_zap);
1904 		zc->zc_hash = zc->zc_serialized << (64 - hb);
1905 		zc->zc_cd += zc->zc_serialized >> hb;
1906 		if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1907 			zc->zc_cd = 0;
1908 	} else {
1909 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1910 	}
1911 	if (!zc->zc_zap->zap_ismicro) {
1912 		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1913 	} else {
1914 		zfs_btree_index_t idx;
1915 		mzap_ent_t mze_tofind;
1916 
1917 		mze_tofind.mze_hash = zc->zc_hash >> 32;
1918 		mze_tofind.mze_cd = zc->zc_cd;
1919 
1920 		mzap_ent_t *mze = zfs_btree_find(&zc->zc_zap->zap_m.zap_tree,
1921 		    &mze_tofind, &idx);
1922 		if (mze == NULL) {
1923 			mze = zfs_btree_next(&zc->zc_zap->zap_m.zap_tree,
1924 			    &idx, &idx);
1925 		}
1926 		if (mze) {
1927 			mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1928 			ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1929 			za->za_normalization_conflict =
1930 			    mzap_normalization_conflict(zc->zc_zap, NULL,
1931 			    mze, &idx);
1932 			za->za_integer_length = 8;
1933 			za->za_num_integers = 1;
1934 			za->za_first_integer = mzep->mze_value;
1935 			(void) strlcpy(za->za_name, mzep->mze_name,
1936 			    za->za_name_len);
1937 			zc->zc_hash = (uint64_t)mze->mze_hash << 32;
1938 			zc->zc_cd = mze->mze_cd;
1939 			err = 0;
1940 		} else {
1941 			zc->zc_hash = -1ULL;
1942 			err = SET_ERROR(ENOENT);
1943 		}
1944 	}
1945 	rw_exit(&zc->zc_zap->zap_rwlock);
1946 	return (err);
1947 }
1948 
1949 void
1950 zap_cursor_advance(zap_cursor_t *zc)
1951 {
1952 	if (zc->zc_hash == -1ULL)
1953 		return;
1954 	zc->zc_cd++;
1955 }
1956 
1957 int
1958 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1959 {
1960 	zap_t *zap;
1961 
1962 	int err =
1963 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1964 	if (err != 0)
1965 		return (err);
1966 
1967 	memset(zs, 0, sizeof (zap_stats_t));
1968 
1969 	if (zap->zap_ismicro) {
1970 		zs->zs_blocksize = zap->zap_dbuf->db_size;
1971 		zs->zs_num_entries = zap->zap_m.zap_num_entries;
1972 		zs->zs_num_blocks = 1;
1973 	} else {
1974 		fzap_get_stats(zap, zs);
1975 	}
1976 	zap_unlockdir(zap, FTAG);
1977 	return (0);
1978 }
1979 
1980 #if defined(_KERNEL)
1981 EXPORT_SYMBOL(zap_create);
1982 EXPORT_SYMBOL(zap_create_dnsize);
1983 EXPORT_SYMBOL(zap_create_norm);
1984 EXPORT_SYMBOL(zap_create_norm_dnsize);
1985 EXPORT_SYMBOL(zap_create_flags);
1986 EXPORT_SYMBOL(zap_create_flags_dnsize);
1987 EXPORT_SYMBOL(zap_create_claim);
1988 EXPORT_SYMBOL(zap_create_claim_norm);
1989 EXPORT_SYMBOL(zap_create_claim_norm_dnsize);
1990 EXPORT_SYMBOL(zap_create_hold);
1991 EXPORT_SYMBOL(zap_destroy);
1992 EXPORT_SYMBOL(zap_lookup);
1993 EXPORT_SYMBOL(zap_lookup_by_dnode);
1994 EXPORT_SYMBOL(zap_lookup_norm);
1995 EXPORT_SYMBOL(zap_lookup_uint64);
1996 EXPORT_SYMBOL(zap_contains);
1997 EXPORT_SYMBOL(zap_prefetch);
1998 EXPORT_SYMBOL(zap_prefetch_uint64);
1999 EXPORT_SYMBOL(zap_prefetch_object);
2000 EXPORT_SYMBOL(zap_add);
2001 EXPORT_SYMBOL(zap_add_by_dnode);
2002 EXPORT_SYMBOL(zap_add_uint64);
2003 EXPORT_SYMBOL(zap_add_uint64_by_dnode);
2004 EXPORT_SYMBOL(zap_update);
2005 EXPORT_SYMBOL(zap_update_uint64);
2006 EXPORT_SYMBOL(zap_update_uint64_by_dnode);
2007 EXPORT_SYMBOL(zap_length);
2008 EXPORT_SYMBOL(zap_length_uint64);
2009 EXPORT_SYMBOL(zap_remove);
2010 EXPORT_SYMBOL(zap_remove_by_dnode);
2011 EXPORT_SYMBOL(zap_remove_norm);
2012 EXPORT_SYMBOL(zap_remove_uint64);
2013 EXPORT_SYMBOL(zap_remove_uint64_by_dnode);
2014 EXPORT_SYMBOL(zap_count);
2015 EXPORT_SYMBOL(zap_value_search);
2016 EXPORT_SYMBOL(zap_join);
2017 EXPORT_SYMBOL(zap_join_increment);
2018 EXPORT_SYMBOL(zap_add_int);
2019 EXPORT_SYMBOL(zap_remove_int);
2020 EXPORT_SYMBOL(zap_lookup_int);
2021 EXPORT_SYMBOL(zap_increment_int);
2022 EXPORT_SYMBOL(zap_add_int_key);
2023 EXPORT_SYMBOL(zap_lookup_int_key);
2024 EXPORT_SYMBOL(zap_increment);
2025 EXPORT_SYMBOL(zap_cursor_init);
2026 EXPORT_SYMBOL(zap_cursor_fini);
2027 EXPORT_SYMBOL(zap_cursor_retrieve);
2028 EXPORT_SYMBOL(zap_cursor_advance);
2029 EXPORT_SYMBOL(zap_cursor_serialize);
2030 EXPORT_SYMBOL(zap_cursor_init_serialized);
2031 EXPORT_SYMBOL(zap_get_stats);
2032 
2033 ZFS_MODULE_PARAM(zfs, , zap_micro_max_size, INT, ZMOD_RW,
2034 	"Maximum micro ZAP size, before converting to a fat ZAP, in bytes");
2035 #endif
2036