xref: /illumos-gate/usr/src/uts/common/fs/zfs/zap_micro.c (revision 52abb70e)
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 
46 extern inline mzap_phys_t *zap_m_phys(zap_t *zap);
47 
48 static int mzap_upgrade(zap_t **zapp,
49     void *tag, dmu_tx_t *tx, zap_flags_t flags);
50 
51 uint64_t
52 zap_getflags(zap_t *zap)
53 {
54 	if (zap->zap_ismicro)
55 		return (0);
56 	return (zap_f_phys(zap)->zap_flags);
57 }
58 
59 int
60 zap_hashbits(zap_t *zap)
61 {
62 	if (zap_getflags(zap) & ZAP_FLAG_HASH64)
63 		return (48);
64 	else
65 		return (28);
66 }
67 
68 uint32_t
69 zap_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 
77 static uint64_t
78 zap_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 
135 static int
136 zap_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 
151 boolean_t
152 zap_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 
169 void
170 zap_name_free(zap_name_t *zn)
171 {
172 	kmem_free(zn, sizeof (zap_name_t));
173 }
174 
175 zap_name_t *
176 zap_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 
234 zap_name_t *
235 zap_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 
250 static void
251 mzap_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 
265 void
266 zap_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 
278 static int
279 mze_compare(const void *arg1, const void *arg2)
280 {
281 	const mzap_ent_t *mze1 = arg1;
282 	const mzap_ent_t *mze2 = arg2;
283 
284 	if (mze1->mze_hash > mze2->mze_hash)
285 		return (+1);
286 	if (mze1->mze_hash < mze2->mze_hash)
287 		return (-1);
288 	if (mze1->mze_cd > mze2->mze_cd)
289 		return (+1);
290 	if (mze1->mze_cd < mze2->mze_cd)
291 		return (-1);
292 	return (0);
293 }
294 
295 static void
296 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
297 {
298 	ASSERT(zap->zap_ismicro);
299 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
300 
301 	mzap_ent_t *mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
302 	mze->mze_chunkid = chunkid;
303 	mze->mze_hash = hash;
304 	mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
305 	ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
306 	avl_add(&zap->zap_m.zap_avl, mze);
307 }
308 
309 static mzap_ent_t *
310 mze_find(zap_name_t *zn)
311 {
312 	mzap_ent_t mze_tofind;
313 	mzap_ent_t *mze;
314 	avl_index_t idx;
315 	avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
316 
317 	ASSERT(zn->zn_zap->zap_ismicro);
318 	ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
319 
320 	mze_tofind.mze_hash = zn->zn_hash;
321 	mze_tofind.mze_cd = 0;
322 
323 	mze = avl_find(avl, &mze_tofind, &idx);
324 	if (mze == NULL)
325 		mze = avl_nearest(avl, idx, AVL_AFTER);
326 	for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
327 		ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
328 		if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
329 			return (mze);
330 	}
331 
332 	return (NULL);
333 }
334 
335 static uint32_t
336 mze_find_unused_cd(zap_t *zap, uint64_t hash)
337 {
338 	mzap_ent_t mze_tofind;
339 	avl_index_t idx;
340 	avl_tree_t *avl = &zap->zap_m.zap_avl;
341 
342 	ASSERT(zap->zap_ismicro);
343 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
344 
345 	mze_tofind.mze_hash = hash;
346 	mze_tofind.mze_cd = 0;
347 
348 	uint32_t cd = 0;
349 	for (mzap_ent_t *mze = avl_find(avl, &mze_tofind, &idx);
350 	    mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
351 		if (mze->mze_cd != cd)
352 			break;
353 		cd++;
354 	}
355 
356 	return (cd);
357 }
358 
359 static void
360 mze_remove(zap_t *zap, mzap_ent_t *mze)
361 {
362 	ASSERT(zap->zap_ismicro);
363 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
364 
365 	avl_remove(&zap->zap_m.zap_avl, mze);
366 	kmem_free(mze, sizeof (mzap_ent_t));
367 }
368 
369 static void
370 mze_destroy(zap_t *zap)
371 {
372 	mzap_ent_t *mze;
373 	void *avlcookie = NULL;
374 
375 	while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
376 		kmem_free(mze, sizeof (mzap_ent_t));
377 	avl_destroy(&zap->zap_m.zap_avl);
378 }
379 
380 static zap_t *
381 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
382 {
383 	zap_t *winner;
384 	uint64_t *zap_hdr = (uint64_t *)db->db_data;
385 	uint64_t zap_block_type = zap_hdr[0];
386 	uint64_t zap_magic = zap_hdr[1];
387 
388 	ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
389 
390 	zap_t *zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
391 	rw_init(&zap->zap_rwlock, 0, 0, 0);
392 	rw_enter(&zap->zap_rwlock, RW_WRITER);
393 	zap->zap_objset = os;
394 	zap->zap_object = obj;
395 	zap->zap_dbuf = db;
396 
397 	if (zap_block_type != ZBT_MICRO) {
398 		mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
399 		zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
400 		if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
401 			winner = NULL;	/* No actual winner here... */
402 			goto handle_winner;
403 		}
404 	} else {
405 		zap->zap_ismicro = TRUE;
406 	}
407 
408 	/*
409 	 * Make sure that zap_ismicro is set before we let others see
410 	 * it, because zap_lockdir() checks zap_ismicro without the lock
411 	 * held.
412 	 */
413 	dmu_buf_init_user(&zap->zap_dbu, zap_evict_sync, NULL, &zap->zap_dbuf);
414 	winner = dmu_buf_set_user(db, &zap->zap_dbu);
415 
416 	if (winner != NULL)
417 		goto handle_winner;
418 
419 	if (zap->zap_ismicro) {
420 		zap->zap_salt = zap_m_phys(zap)->mz_salt;
421 		zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
422 		zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
423 		avl_create(&zap->zap_m.zap_avl, mze_compare,
424 		    sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
425 
426 		for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
427 			mzap_ent_phys_t *mze =
428 			    &zap_m_phys(zap)->mz_chunk[i];
429 			if (mze->mze_name[0]) {
430 				zap_name_t *zn;
431 
432 				zap->zap_m.zap_num_entries++;
433 				zn = zap_name_alloc(zap, mze->mze_name, 0);
434 				mze_insert(zap, i, zn->zn_hash);
435 				zap_name_free(zn);
436 			}
437 		}
438 	} else {
439 		zap->zap_salt = zap_f_phys(zap)->zap_salt;
440 		zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
441 
442 		ASSERT3U(sizeof (struct zap_leaf_header), ==,
443 		    2*ZAP_LEAF_CHUNKSIZE);
444 
445 		/*
446 		 * The embedded pointer table should not overlap the
447 		 * other members.
448 		 */
449 		ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
450 		    &zap_f_phys(zap)->zap_salt);
451 
452 		/*
453 		 * The embedded pointer table should end at the end of
454 		 * the block
455 		 */
456 		ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
457 		    1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
458 		    (uintptr_t)zap_f_phys(zap), ==,
459 		    zap->zap_dbuf->db_size);
460 	}
461 	rw_exit(&zap->zap_rwlock);
462 	return (zap);
463 
464 handle_winner:
465 	rw_exit(&zap->zap_rwlock);
466 	rw_destroy(&zap->zap_rwlock);
467 	if (!zap->zap_ismicro)
468 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
469 	kmem_free(zap, sizeof (zap_t));
470 	return (winner);
471 }
472 
473 /*
474  * This routine "consumes" the caller's hold on the dbuf, which must
475  * have the specified tag.
476  */
477 static int
478 zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx,
479     krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
480 {
481 	ASSERT0(db->db_offset);
482 	objset_t *os = dmu_buf_get_objset(db);
483 	uint64_t obj = db->db_object;
484 
485 	*zapp = NULL;
486 
487 	zap_t *zap = dmu_buf_get_user(db);
488 	if (zap == NULL) {
489 		zap = mzap_open(os, obj, db);
490 		if (zap == NULL) {
491 			/*
492 			 * mzap_open() didn't like what it saw on-disk.
493 			 * Check for corruption!
494 			 */
495 			return (SET_ERROR(EIO));
496 		}
497 	}
498 
499 	/*
500 	 * We're checking zap_ismicro without the lock held, in order to
501 	 * tell what type of lock we want.  Once we have some sort of
502 	 * lock, see if it really is the right type.  In practice this
503 	 * can only be different if it was upgraded from micro to fat,
504 	 * and micro wanted WRITER but fat only needs READER.
505 	 */
506 	krw_t lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
507 	rw_enter(&zap->zap_rwlock, lt);
508 	if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
509 		/* it was upgraded, now we only need reader */
510 		ASSERT(lt == RW_WRITER);
511 		ASSERT(RW_READER ==
512 		    (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
513 		rw_downgrade(&zap->zap_rwlock);
514 		lt = RW_READER;
515 	}
516 
517 	zap->zap_objset = os;
518 
519 	if (lt == RW_WRITER)
520 		dmu_buf_will_dirty(db, tx);
521 
522 	ASSERT3P(zap->zap_dbuf, ==, db);
523 
524 	ASSERT(!zap->zap_ismicro ||
525 	    zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
526 	if (zap->zap_ismicro && tx && adding &&
527 	    zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
528 		uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
529 		if (newsz > MZAP_MAX_BLKSZ) {
530 			dprintf("upgrading obj %llu: num_entries=%u\n",
531 			    obj, zap->zap_m.zap_num_entries);
532 			*zapp = zap;
533 			int err = mzap_upgrade(zapp, tag, tx, 0);
534 			if (err != 0)
535 				rw_exit(&zap->zap_rwlock);
536 			return (err);
537 		}
538 		VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx));
539 		zap->zap_m.zap_num_chunks =
540 		    db->db_size / MZAP_ENT_LEN - 1;
541 	}
542 
543 	*zapp = zap;
544 	return (0);
545 }
546 
547 static int
548 zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
549     krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
550 {
551 	dmu_buf_t *db;
552 
553 	int err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
554 	if (err != 0) {
555 		return (err);
556 	}
557 #ifdef ZFS_DEBUG
558 	{
559 		dmu_object_info_t doi;
560 		dmu_object_info_from_db(db, &doi);
561 		ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
562 	}
563 #endif
564 
565 	err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
566 	if (err != 0) {
567 		dmu_buf_rele(db, tag);
568 	}
569 	return (err);
570 }
571 
572 int
573 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
574     krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
575 {
576 	dmu_buf_t *db;
577 
578 	int err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH);
579 	if (err != 0)
580 		return (err);
581 #ifdef ZFS_DEBUG
582 	{
583 		dmu_object_info_t doi;
584 		dmu_object_info_from_db(db, &doi);
585 		ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
586 	}
587 #endif
588 	err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
589 	if (err != 0)
590 		dmu_buf_rele(db, tag);
591 	return (err);
592 }
593 
594 void
595 zap_unlockdir(zap_t *zap, void *tag)
596 {
597 	rw_exit(&zap->zap_rwlock);
598 	dmu_buf_rele(zap->zap_dbuf, tag);
599 }
600 
601 static int
602 mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags)
603 {
604 	int err = 0;
605 	zap_t *zap = *zapp;
606 
607 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
608 
609 	int sz = zap->zap_dbuf->db_size;
610 	mzap_phys_t *mzp = zio_buf_alloc(sz);
611 	bcopy(zap->zap_dbuf->db_data, mzp, sz);
612 	int nchunks = zap->zap_m.zap_num_chunks;
613 
614 	if (!flags) {
615 		err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
616 		    1ULL << fzap_default_block_shift, 0, tx);
617 		if (err != 0) {
618 			zio_buf_free(mzp, sz);
619 			return (err);
620 		}
621 	}
622 
623 	dprintf("upgrading obj=%llu with %u chunks\n",
624 	    zap->zap_object, nchunks);
625 	/* XXX destroy the avl later, so we can use the stored hash value */
626 	mze_destroy(zap);
627 
628 	fzap_upgrade(zap, tx, flags);
629 
630 	for (int i = 0; i < nchunks; i++) {
631 		mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
632 		if (mze->mze_name[0] == 0)
633 			continue;
634 		dprintf("adding %s=%llu\n",
635 		    mze->mze_name, mze->mze_value);
636 		zap_name_t *zn = zap_name_alloc(zap, mze->mze_name, 0);
637 		err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
638 		    tag, tx);
639 		zap = zn->zn_zap;	/* fzap_add_cd() may change zap */
640 		zap_name_free(zn);
641 		if (err != 0)
642 			break;
643 	}
644 	zio_buf_free(mzp, sz);
645 	*zapp = zap;
646 	return (err);
647 }
648 
649 /*
650  * The "normflags" determine the behavior of the matchtype_t which is
651  * passed to zap_lookup_norm().  Names which have the same normalized
652  * version will be stored with the same hash value, and therefore we can
653  * perform normalization-insensitive lookups.  We can be Unicode form-
654  * insensitive and/or case-insensitive.  The following flags are valid for
655  * "normflags":
656  *
657  * U8_TEXTPREP_NFC
658  * U8_TEXTPREP_NFD
659  * U8_TEXTPREP_NFKC
660  * U8_TEXTPREP_NFKD
661  * U8_TEXTPREP_TOUPPER
662  *
663  * The *_NF* (Normalization Form) flags are mutually exclusive; at most one
664  * of them may be supplied.
665  */
666 void
667 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
668     dmu_tx_t *tx)
669 {
670 	dmu_buf_t *db;
671 
672 	VERIFY0(dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
673 
674 	dmu_buf_will_dirty(db, tx);
675 	mzap_phys_t *zp = db->db_data;
676 	zp->mz_block_type = ZBT_MICRO;
677 	zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
678 	zp->mz_normflags = normflags;
679 
680 	if (flags != 0) {
681 		zap_t *zap;
682 		/* Only fat zap supports flags; upgrade immediately. */
683 		VERIFY0(zap_lockdir_impl(db, FTAG, tx, RW_WRITER,
684 		    B_FALSE, B_FALSE, &zap));
685 		VERIFY0(mzap_upgrade(&zap, FTAG, tx, flags));
686 		zap_unlockdir(zap, FTAG);
687 	} else {
688 		dmu_buf_rele(db, FTAG);
689 	}
690 }
691 
692 int
693 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
694     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
695 {
696 	return (zap_create_claim_dnsize(os, obj, ot, bonustype, bonuslen,
697 	    0, tx));
698 }
699 
700 int
701 zap_create_claim_dnsize(objset_t *os, uint64_t obj, dmu_object_type_t ot,
702     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
703 {
704 	return (zap_create_claim_norm_dnsize(os, obj,
705 	    0, ot, bonustype, bonuslen, dnodesize, tx));
706 }
707 
708 int
709 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
710     dmu_object_type_t ot,
711     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
712 {
713 	return (zap_create_claim_norm_dnsize(os, obj, normflags, ot, bonustype,
714 	    bonuslen, 0, tx));
715 }
716 
717 int
718 zap_create_claim_norm_dnsize(objset_t *os, uint64_t obj, int normflags,
719     dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
720     int dnodesize, dmu_tx_t *tx)
721 {
722 	int err;
723 
724 	err = dmu_object_claim_dnsize(os, obj, ot, 0, bonustype, bonuslen,
725 	    dnodesize, tx);
726 	if (err != 0)
727 		return (err);
728 	mzap_create_impl(os, obj, normflags, 0, tx);
729 	return (0);
730 }
731 
732 uint64_t
733 zap_create(objset_t *os, dmu_object_type_t ot,
734     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
735 {
736 	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
737 }
738 
739 uint64_t
740 zap_create_dnsize(objset_t *os, dmu_object_type_t ot,
741     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
742 {
743 	return (zap_create_norm_dnsize(os, 0, ot, bonustype, bonuslen,
744 	    dnodesize, tx));
745 }
746 
747 uint64_t
748 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
749     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
750 {
751 	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
752 	return (zap_create_norm_dnsize(os, normflags, ot, bonustype, bonuslen,
753 	    0, tx));
754 }
755 
756 uint64_t
757 zap_create_norm_dnsize(objset_t *os, int normflags, dmu_object_type_t ot,
758     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
759 {
760 	uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
761 	    dnodesize, tx);
762 
763 	mzap_create_impl(os, obj, normflags, 0, tx);
764 	return (obj);
765 }
766 
767 uint64_t
768 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
769     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
770     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
771 {
772 	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
773 	return (zap_create_flags_dnsize(os, normflags, flags, ot,
774 	    leaf_blockshift, indirect_blockshift, bonustype, bonuslen, 0, tx));
775 }
776 
777 uint64_t
778 zap_create_flags_dnsize(objset_t *os, int normflags, zap_flags_t flags,
779     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
780     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
781 {
782 	uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
783 	    dnodesize, tx);
784 
785 	ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
786 	    leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
787 	    indirect_blockshift >= SPA_MINBLOCKSHIFT &&
788 	    indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
789 
790 	VERIFY(dmu_object_set_blocksize(os, obj,
791 	    1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
792 
793 	mzap_create_impl(os, obj, normflags, flags, tx);
794 	return (obj);
795 }
796 
797 int
798 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
799 {
800 	/*
801 	 * dmu_object_free will free the object number and free the
802 	 * data.  Freeing the data will cause our pageout function to be
803 	 * called, which will destroy our data (zap_leaf_t's and zap_t).
804 	 */
805 
806 	return (dmu_object_free(os, zapobj, tx));
807 }
808 
809 void
810 zap_evict_sync(void *dbu)
811 {
812 	zap_t *zap = dbu;
813 
814 	rw_destroy(&zap->zap_rwlock);
815 
816 	if (zap->zap_ismicro)
817 		mze_destroy(zap);
818 	else
819 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
820 
821 	kmem_free(zap, sizeof (zap_t));
822 }
823 
824 int
825 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
826 {
827 	zap_t *zap;
828 
829 	int err =
830 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
831 	if (err != 0)
832 		return (err);
833 	if (!zap->zap_ismicro) {
834 		err = fzap_count(zap, count);
835 	} else {
836 		*count = zap->zap_m.zap_num_entries;
837 	}
838 	zap_unlockdir(zap, FTAG);
839 	return (err);
840 }
841 
842 /*
843  * zn may be NULL; if not specified, it will be computed if needed.
844  * See also the comment above zap_entry_normalization_conflict().
845  */
846 static boolean_t
847 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
848 {
849 	int direction = AVL_BEFORE;
850 	boolean_t allocdzn = B_FALSE;
851 
852 	if (zap->zap_normflags == 0)
853 		return (B_FALSE);
854 
855 again:
856 	for (mzap_ent_t *other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
857 	    other && other->mze_hash == mze->mze_hash;
858 	    other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
859 
860 		if (zn == NULL) {
861 			zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
862 			    MT_NORMALIZE);
863 			allocdzn = B_TRUE;
864 		}
865 		if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
866 			if (allocdzn)
867 				zap_name_free(zn);
868 			return (B_TRUE);
869 		}
870 	}
871 
872 	if (direction == AVL_BEFORE) {
873 		direction = AVL_AFTER;
874 		goto again;
875 	}
876 
877 	if (allocdzn)
878 		zap_name_free(zn);
879 	return (B_FALSE);
880 }
881 
882 /*
883  * Routines for manipulating attributes.
884  */
885 
886 int
887 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
888     uint64_t integer_size, uint64_t num_integers, void *buf)
889 {
890 	return (zap_lookup_norm(os, zapobj, name, integer_size,
891 	    num_integers, buf, 0, NULL, 0, NULL));
892 }
893 
894 static int
895 zap_lookup_impl(zap_t *zap, const char *name,
896     uint64_t integer_size, uint64_t num_integers, void *buf,
897     matchtype_t mt, char *realname, int rn_len,
898     boolean_t *ncp)
899 {
900 	int err = 0;
901 
902 	zap_name_t *zn = zap_name_alloc(zap, name, mt);
903 	if (zn == NULL)
904 		return (SET_ERROR(ENOTSUP));
905 
906 	if (!zap->zap_ismicro) {
907 		err = fzap_lookup(zn, integer_size, num_integers, buf,
908 		    realname, rn_len, ncp);
909 	} else {
910 		mzap_ent_t *mze = mze_find(zn);
911 		if (mze == NULL) {
912 			err = SET_ERROR(ENOENT);
913 		} else {
914 			if (num_integers < 1) {
915 				err = SET_ERROR(EOVERFLOW);
916 			} else if (integer_size != 8) {
917 				err = SET_ERROR(EINVAL);
918 			} else {
919 				*(uint64_t *)buf =
920 				    MZE_PHYS(zap, mze)->mze_value;
921 				(void) strlcpy(realname,
922 				    MZE_PHYS(zap, mze)->mze_name, rn_len);
923 				if (ncp) {
924 					*ncp = mzap_normalization_conflict(zap,
925 					    zn, mze);
926 				}
927 			}
928 		}
929 	}
930 	zap_name_free(zn);
931 	return (err);
932 }
933 
934 int
935 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
936     uint64_t integer_size, uint64_t num_integers, void *buf,
937     matchtype_t mt, char *realname, int rn_len,
938     boolean_t *ncp)
939 {
940 	zap_t *zap;
941 
942 	int err =
943 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
944 	if (err != 0)
945 		return (err);
946 	err = zap_lookup_impl(zap, name, integer_size,
947 	    num_integers, buf, mt, realname, rn_len, ncp);
948 	zap_unlockdir(zap, FTAG);
949 	return (err);
950 }
951 
952 int
953 zap_lookup_by_dnode(dnode_t *dn, const char *name,
954     uint64_t integer_size, uint64_t num_integers, void *buf)
955 {
956 	return (zap_lookup_norm_by_dnode(dn, name, integer_size,
957 	    num_integers, buf, 0, NULL, 0, NULL));
958 }
959 
960 int
961 zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
962     uint64_t integer_size, uint64_t num_integers, void *buf,
963     matchtype_t mt, char *realname, int rn_len,
964     boolean_t *ncp)
965 {
966 	zap_t *zap;
967 
968 	int err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
969 	    FTAG, &zap);
970 	if (err != 0)
971 		return (err);
972 	err = zap_lookup_impl(zap, name, integer_size,
973 	    num_integers, buf, mt, realname, rn_len, ncp);
974 	zap_unlockdir(zap, FTAG);
975 	return (err);
976 }
977 
978 int
979 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
980     int key_numints)
981 {
982 	zap_t *zap;
983 
984 	int err =
985 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
986 	if (err != 0)
987 		return (err);
988 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
989 	if (zn == NULL) {
990 		zap_unlockdir(zap, FTAG);
991 		return (SET_ERROR(ENOTSUP));
992 	}
993 
994 	fzap_prefetch(zn);
995 	zap_name_free(zn);
996 	zap_unlockdir(zap, FTAG);
997 	return (err);
998 }
999 
1000 int
1001 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1002     int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
1003 {
1004 	zap_t *zap;
1005 
1006 	int err =
1007 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1008 	if (err != 0)
1009 		return (err);
1010 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1011 	if (zn == NULL) {
1012 		zap_unlockdir(zap, FTAG);
1013 		return (SET_ERROR(ENOTSUP));
1014 	}
1015 
1016 	err = fzap_lookup(zn, integer_size, num_integers, buf,
1017 	    NULL, 0, NULL);
1018 	zap_name_free(zn);
1019 	zap_unlockdir(zap, FTAG);
1020 	return (err);
1021 }
1022 
1023 int
1024 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
1025 {
1026 	int err = zap_lookup_norm(os, zapobj, name, 0,
1027 	    0, NULL, 0, NULL, 0, NULL);
1028 	if (err == EOVERFLOW || err == EINVAL)
1029 		err = 0; /* found, but skipped reading the value */
1030 	return (err);
1031 }
1032 
1033 int
1034 zap_length(objset_t *os, uint64_t zapobj, const char *name,
1035     uint64_t *integer_size, uint64_t *num_integers)
1036 {
1037 	zap_t *zap;
1038 
1039 	int err =
1040 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1041 	if (err != 0)
1042 		return (err);
1043 	zap_name_t *zn = zap_name_alloc(zap, name, 0);
1044 	if (zn == NULL) {
1045 		zap_unlockdir(zap, FTAG);
1046 		return (SET_ERROR(ENOTSUP));
1047 	}
1048 	if (!zap->zap_ismicro) {
1049 		err = fzap_length(zn, integer_size, num_integers);
1050 	} else {
1051 		mzap_ent_t *mze = mze_find(zn);
1052 		if (mze == NULL) {
1053 			err = SET_ERROR(ENOENT);
1054 		} else {
1055 			if (integer_size)
1056 				*integer_size = 8;
1057 			if (num_integers)
1058 				*num_integers = 1;
1059 		}
1060 	}
1061 	zap_name_free(zn);
1062 	zap_unlockdir(zap, FTAG);
1063 	return (err);
1064 }
1065 
1066 int
1067 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1068     int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1069 {
1070 	zap_t *zap;
1071 
1072 	int err =
1073 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1074 	if (err != 0)
1075 		return (err);
1076 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1077 	if (zn == NULL) {
1078 		zap_unlockdir(zap, FTAG);
1079 		return (SET_ERROR(ENOTSUP));
1080 	}
1081 	err = fzap_length(zn, integer_size, num_integers);
1082 	zap_name_free(zn);
1083 	zap_unlockdir(zap, FTAG);
1084 	return (err);
1085 }
1086 
1087 static void
1088 mzap_addent(zap_name_t *zn, uint64_t value)
1089 {
1090 	zap_t *zap = zn->zn_zap;
1091 	int start = zap->zap_m.zap_alloc_next;
1092 
1093 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1094 
1095 #ifdef ZFS_DEBUG
1096 	for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1097 		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1098 		ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1099 	}
1100 #endif
1101 
1102 	uint32_t cd = mze_find_unused_cd(zap, zn->zn_hash);
1103 	/* given the limited size of the microzap, this can't happen */
1104 	ASSERT(cd < zap_maxcd(zap));
1105 
1106 again:
1107 	for (int i = start; i < zap->zap_m.zap_num_chunks; i++) {
1108 		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1109 		if (mze->mze_name[0] == 0) {
1110 			mze->mze_value = value;
1111 			mze->mze_cd = cd;
1112 			(void) strcpy(mze->mze_name, zn->zn_key_orig);
1113 			zap->zap_m.zap_num_entries++;
1114 			zap->zap_m.zap_alloc_next = i+1;
1115 			if (zap->zap_m.zap_alloc_next ==
1116 			    zap->zap_m.zap_num_chunks)
1117 				zap->zap_m.zap_alloc_next = 0;
1118 			mze_insert(zap, i, zn->zn_hash);
1119 			return;
1120 		}
1121 	}
1122 	if (start != 0) {
1123 		start = 0;
1124 		goto again;
1125 	}
1126 	ASSERT(!"out of entries!");
1127 }
1128 
1129 static int
1130 zap_add_impl(zap_t *zap, const char *key,
1131     int integer_size, uint64_t num_integers,
1132     const void *val, dmu_tx_t *tx, void *tag)
1133 {
1134 	const uint64_t *intval = val;
1135 	int err = 0;
1136 
1137 	zap_name_t *zn = zap_name_alloc(zap, key, 0);
1138 	if (zn == NULL) {
1139 		zap_unlockdir(zap, tag);
1140 		return (SET_ERROR(ENOTSUP));
1141 	}
1142 	if (!zap->zap_ismicro) {
1143 		err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1144 		zap = zn->zn_zap;	/* fzap_add() may change zap */
1145 	} else if (integer_size != 8 || num_integers != 1 ||
1146 	    strlen(key) >= MZAP_NAME_LEN) {
1147 		err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
1148 		if (err == 0) {
1149 			err = fzap_add(zn, integer_size, num_integers, val,
1150 			    tag, tx);
1151 		}
1152 		zap = zn->zn_zap;	/* fzap_add() may change zap */
1153 	} else {
1154 		if (mze_find(zn) != NULL) {
1155 			err = SET_ERROR(EEXIST);
1156 		} else {
1157 			mzap_addent(zn, *intval);
1158 		}
1159 	}
1160 	ASSERT(zap == zn->zn_zap);
1161 	zap_name_free(zn);
1162 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1163 		zap_unlockdir(zap, tag);
1164 	return (err);
1165 }
1166 
1167 int
1168 zap_add(objset_t *os, uint64_t zapobj, const char *key,
1169     int integer_size, uint64_t num_integers,
1170     const void *val, dmu_tx_t *tx)
1171 {
1172 	zap_t *zap;
1173 	int err;
1174 
1175 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1176 	if (err != 0)
1177 		return (err);
1178 	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1179 	/* zap_add_impl() calls zap_unlockdir() */
1180 	return (err);
1181 }
1182 
1183 int
1184 zap_add_by_dnode(dnode_t *dn, const char *key,
1185     int integer_size, uint64_t num_integers,
1186     const void *val, dmu_tx_t *tx)
1187 {
1188 	zap_t *zap;
1189 	int err;
1190 
1191 	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1192 	if (err != 0)
1193 		return (err);
1194 	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1195 	/* zap_add_impl() calls zap_unlockdir() */
1196 	return (err);
1197 }
1198 
1199 int
1200 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1201     int key_numints, int integer_size, uint64_t num_integers,
1202     const void *val, dmu_tx_t *tx)
1203 {
1204 	zap_t *zap;
1205 
1206 	int err =
1207 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1208 	if (err != 0)
1209 		return (err);
1210 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1211 	if (zn == NULL) {
1212 		zap_unlockdir(zap, FTAG);
1213 		return (SET_ERROR(ENOTSUP));
1214 	}
1215 	err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1216 	zap = zn->zn_zap;	/* fzap_add() may change zap */
1217 	zap_name_free(zn);
1218 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1219 		zap_unlockdir(zap, FTAG);
1220 	return (err);
1221 }
1222 
1223 int
1224 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1225     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1226 {
1227 	zap_t *zap;
1228 	uint64_t oldval;
1229 	const uint64_t *intval = val;
1230 
1231 #ifdef ZFS_DEBUG
1232 	/*
1233 	 * If there is an old value, it shouldn't change across the
1234 	 * lockdir (eg, due to bprewrite's xlation).
1235 	 */
1236 	if (integer_size == 8 && num_integers == 1)
1237 		(void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1238 #endif
1239 
1240 	int err =
1241 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1242 	if (err != 0)
1243 		return (err);
1244 	zap_name_t *zn = zap_name_alloc(zap, name, 0);
1245 	if (zn == NULL) {
1246 		zap_unlockdir(zap, FTAG);
1247 		return (SET_ERROR(ENOTSUP));
1248 	}
1249 	if (!zap->zap_ismicro) {
1250 		err = fzap_update(zn, integer_size, num_integers, val,
1251 		    FTAG, tx);
1252 		zap = zn->zn_zap;	/* fzap_update() may change zap */
1253 	} else if (integer_size != 8 || num_integers != 1 ||
1254 	    strlen(name) >= MZAP_NAME_LEN) {
1255 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1256 		    zapobj, integer_size, num_integers, name);
1257 		err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1258 		if (err == 0) {
1259 			err = fzap_update(zn, integer_size, num_integers,
1260 			    val, FTAG, tx);
1261 		}
1262 		zap = zn->zn_zap;	/* fzap_update() may change zap */
1263 	} else {
1264 		mzap_ent_t *mze = mze_find(zn);
1265 		if (mze != NULL) {
1266 			ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1267 			MZE_PHYS(zap, mze)->mze_value = *intval;
1268 		} else {
1269 			mzap_addent(zn, *intval);
1270 		}
1271 	}
1272 	ASSERT(zap == zn->zn_zap);
1273 	zap_name_free(zn);
1274 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1275 		zap_unlockdir(zap, FTAG);
1276 	return (err);
1277 }
1278 
1279 int
1280 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1281     int key_numints,
1282     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1283 {
1284 	zap_t *zap;
1285 
1286 	int err =
1287 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1288 	if (err != 0)
1289 		return (err);
1290 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1291 	if (zn == NULL) {
1292 		zap_unlockdir(zap, FTAG);
1293 		return (SET_ERROR(ENOTSUP));
1294 	}
1295 	err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx);
1296 	zap = zn->zn_zap;	/* fzap_update() may change zap */
1297 	zap_name_free(zn);
1298 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1299 		zap_unlockdir(zap, FTAG);
1300 	return (err);
1301 }
1302 
1303 int
1304 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1305 {
1306 	return (zap_remove_norm(os, zapobj, name, 0, tx));
1307 }
1308 
1309 static int
1310 zap_remove_impl(zap_t *zap, const char *name,
1311     matchtype_t mt, dmu_tx_t *tx)
1312 {
1313 	int err = 0;
1314 
1315 	zap_name_t *zn = zap_name_alloc(zap, name, mt);
1316 	if (zn == NULL)
1317 		return (SET_ERROR(ENOTSUP));
1318 	if (!zap->zap_ismicro) {
1319 		err = fzap_remove(zn, tx);
1320 	} else {
1321 		mzap_ent_t *mze = mze_find(zn);
1322 		if (mze == NULL) {
1323 			err = SET_ERROR(ENOENT);
1324 		} else {
1325 			zap->zap_m.zap_num_entries--;
1326 			bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1327 			    sizeof (mzap_ent_phys_t));
1328 			mze_remove(zap, mze);
1329 		}
1330 	}
1331 	zap_name_free(zn);
1332 	return (err);
1333 }
1334 
1335 int
1336 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1337     matchtype_t mt, dmu_tx_t *tx)
1338 {
1339 	zap_t *zap;
1340 	int err;
1341 
1342 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1343 	if (err)
1344 		return (err);
1345 	err = zap_remove_impl(zap, name, mt, tx);
1346 	zap_unlockdir(zap, FTAG);
1347 	return (err);
1348 }
1349 
1350 int
1351 zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
1352 {
1353 	zap_t *zap;
1354 	int err;
1355 
1356 	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1357 	if (err)
1358 		return (err);
1359 	err = zap_remove_impl(zap, name, 0, tx);
1360 	zap_unlockdir(zap, FTAG);
1361 	return (err);
1362 }
1363 
1364 int
1365 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1366     int key_numints, dmu_tx_t *tx)
1367 {
1368 	zap_t *zap;
1369 
1370 	int err =
1371 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1372 	if (err != 0)
1373 		return (err);
1374 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1375 	if (zn == NULL) {
1376 		zap_unlockdir(zap, FTAG);
1377 		return (SET_ERROR(ENOTSUP));
1378 	}
1379 	err = fzap_remove(zn, tx);
1380 	zap_name_free(zn);
1381 	zap_unlockdir(zap, FTAG);
1382 	return (err);
1383 }
1384 
1385 /*
1386  * Routines for iterating over the attributes.
1387  */
1388 
1389 static void
1390 zap_cursor_init_impl(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1391     uint64_t serialized, boolean_t prefetch)
1392 {
1393 	zc->zc_objset = os;
1394 	zc->zc_zap = NULL;
1395 	zc->zc_leaf = NULL;
1396 	zc->zc_zapobj = zapobj;
1397 	zc->zc_serialized = serialized;
1398 	zc->zc_hash = 0;
1399 	zc->zc_cd = 0;
1400 	zc->zc_prefetch = prefetch;
1401 }
1402 void
1403 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1404     uint64_t serialized)
1405 {
1406 	zap_cursor_init_impl(zc, os, zapobj, serialized, B_TRUE);
1407 }
1408 
1409 /*
1410  * Initialize a cursor at the beginning of the ZAP object.  The entire
1411  * ZAP object will be prefetched.
1412  */
1413 void
1414 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1415 {
1416 	zap_cursor_init_impl(zc, os, zapobj, 0, B_TRUE);
1417 }
1418 
1419 /*
1420  * Initialize a cursor at the beginning, but request that we not prefetch
1421  * the entire ZAP object.
1422  */
1423 void
1424 zap_cursor_init_noprefetch(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1425 {
1426 	zap_cursor_init_impl(zc, os, zapobj, 0, B_FALSE);
1427 }
1428 
1429 void
1430 zap_cursor_fini(zap_cursor_t *zc)
1431 {
1432 	if (zc->zc_zap) {
1433 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1434 		zap_unlockdir(zc->zc_zap, NULL);
1435 		zc->zc_zap = NULL;
1436 	}
1437 	if (zc->zc_leaf) {
1438 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1439 		zap_put_leaf(zc->zc_leaf);
1440 		zc->zc_leaf = NULL;
1441 	}
1442 	zc->zc_objset = NULL;
1443 }
1444 
1445 uint64_t
1446 zap_cursor_serialize(zap_cursor_t *zc)
1447 {
1448 	if (zc->zc_hash == -1ULL)
1449 		return (-1ULL);
1450 	if (zc->zc_zap == NULL)
1451 		return (zc->zc_serialized);
1452 	ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1453 	ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1454 
1455 	/*
1456 	 * We want to keep the high 32 bits of the cursor zero if we can, so
1457 	 * that 32-bit programs can access this.  So usually use a small
1458 	 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1459 	 * of the cursor.
1460 	 *
1461 	 * [ collision differentiator | zap_hashbits()-bit hash value ]
1462 	 */
1463 	return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1464 	    ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1465 }
1466 
1467 int
1468 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1469 {
1470 	int err;
1471 
1472 	if (zc->zc_hash == -1ULL)
1473 		return (SET_ERROR(ENOENT));
1474 
1475 	if (zc->zc_zap == NULL) {
1476 		int hb;
1477 		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1478 		    RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1479 		if (err != 0)
1480 			return (err);
1481 
1482 		/*
1483 		 * To support zap_cursor_init_serialized, advance, retrieve,
1484 		 * we must add to the existing zc_cd, which may already
1485 		 * be 1 due to the zap_cursor_advance.
1486 		 */
1487 		ASSERT(zc->zc_hash == 0);
1488 		hb = zap_hashbits(zc->zc_zap);
1489 		zc->zc_hash = zc->zc_serialized << (64 - hb);
1490 		zc->zc_cd += zc->zc_serialized >> hb;
1491 		if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1492 			zc->zc_cd = 0;
1493 	} else {
1494 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1495 	}
1496 	if (!zc->zc_zap->zap_ismicro) {
1497 		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1498 	} else {
1499 		avl_index_t idx;
1500 		mzap_ent_t mze_tofind;
1501 
1502 		mze_tofind.mze_hash = zc->zc_hash;
1503 		mze_tofind.mze_cd = zc->zc_cd;
1504 
1505 		mzap_ent_t *mze =
1506 		    avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1507 		if (mze == NULL) {
1508 			mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1509 			    idx, AVL_AFTER);
1510 		}
1511 		if (mze) {
1512 			mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1513 			ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1514 			za->za_normalization_conflict =
1515 			    mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1516 			za->za_integer_length = 8;
1517 			za->za_num_integers = 1;
1518 			za->za_first_integer = mzep->mze_value;
1519 			(void) strcpy(za->za_name, mzep->mze_name);
1520 			zc->zc_hash = mze->mze_hash;
1521 			zc->zc_cd = mze->mze_cd;
1522 			err = 0;
1523 		} else {
1524 			zc->zc_hash = -1ULL;
1525 			err = SET_ERROR(ENOENT);
1526 		}
1527 	}
1528 	rw_exit(&zc->zc_zap->zap_rwlock);
1529 	return (err);
1530 }
1531 
1532 void
1533 zap_cursor_advance(zap_cursor_t *zc)
1534 {
1535 	if (zc->zc_hash == -1ULL)
1536 		return;
1537 	zc->zc_cd++;
1538 }
1539 
1540 int
1541 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1542 {
1543 	zap_t *zap;
1544 
1545 	int err =
1546 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1547 	if (err != 0)
1548 		return (err);
1549 
1550 	bzero(zs, sizeof (zap_stats_t));
1551 
1552 	if (zap->zap_ismicro) {
1553 		zs->zs_blocksize = zap->zap_dbuf->db_size;
1554 		zs->zs_num_entries = zap->zap_m.zap_num_entries;
1555 		zs->zs_num_blocks = 1;
1556 	} else {
1557 		fzap_get_stats(zap, zs);
1558 	}
1559 	zap_unlockdir(zap, FTAG);
1560 	return (0);
1561 }
1562