xref: /illumos-gate/usr/src/uts/common/fs/zfs/zap_micro.c (revision 4c28a617e3922d92a58e813a5b955eb526b9c386)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2017 Nexenta Systems, Inc.
28  */
29 
30 #include <sys/zio.h>
31 #include <sys/spa.h>
32 #include <sys/dmu.h>
33 #include <sys/zfs_context.h>
34 #include <sys/zap.h>
35 #include <sys/refcount.h>
36 #include <sys/zap_impl.h>
37 #include <sys/zap_leaf.h>
38 #include <sys/avl.h>
39 #include <sys/arc.h>
40 #include <sys/dmu_objset.h>
41 
42 #ifdef _KERNEL
43 #include <sys/sunddi.h>
44 #endif
45 
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_norm(os, obj,
697 	    0, ot, bonustype, bonuslen, tx));
698 }
699 
700 int
701 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
702     dmu_object_type_t ot,
703     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
704 {
705 	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
706 	int err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
707 	if (err != 0)
708 		return (err);
709 	mzap_create_impl(os, obj, normflags, 0, tx);
710 	return (0);
711 }
712 
713 uint64_t
714 zap_create(objset_t *os, dmu_object_type_t ot,
715     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
716 {
717 	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
718 }
719 
720 uint64_t
721 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
722     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
723 {
724 	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
725 	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
726 
727 	mzap_create_impl(os, obj, normflags, 0, tx);
728 	return (obj);
729 }
730 
731 uint64_t
732 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
733     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
734     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
735 {
736 	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
737 	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
738 
739 	ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
740 	    leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
741 	    indirect_blockshift >= SPA_MINBLOCKSHIFT &&
742 	    indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
743 
744 	VERIFY(dmu_object_set_blocksize(os, obj,
745 	    1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
746 
747 	mzap_create_impl(os, obj, normflags, flags, tx);
748 	return (obj);
749 }
750 
751 int
752 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
753 {
754 	/*
755 	 * dmu_object_free will free the object number and free the
756 	 * data.  Freeing the data will cause our pageout function to be
757 	 * called, which will destroy our data (zap_leaf_t's and zap_t).
758 	 */
759 
760 	return (dmu_object_free(os, zapobj, tx));
761 }
762 
763 void
764 zap_evict_sync(void *dbu)
765 {
766 	zap_t *zap = dbu;
767 
768 	rw_destroy(&zap->zap_rwlock);
769 
770 	if (zap->zap_ismicro)
771 		mze_destroy(zap);
772 	else
773 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
774 
775 	kmem_free(zap, sizeof (zap_t));
776 }
777 
778 int
779 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
780 {
781 	zap_t *zap;
782 
783 	int err =
784 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
785 	if (err != 0)
786 		return (err);
787 	if (!zap->zap_ismicro) {
788 		err = fzap_count(zap, count);
789 	} else {
790 		*count = zap->zap_m.zap_num_entries;
791 	}
792 	zap_unlockdir(zap, FTAG);
793 	return (err);
794 }
795 
796 /*
797  * zn may be NULL; if not specified, it will be computed if needed.
798  * See also the comment above zap_entry_normalization_conflict().
799  */
800 static boolean_t
801 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
802 {
803 	int direction = AVL_BEFORE;
804 	boolean_t allocdzn = B_FALSE;
805 
806 	if (zap->zap_normflags == 0)
807 		return (B_FALSE);
808 
809 again:
810 	for (mzap_ent_t *other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
811 	    other && other->mze_hash == mze->mze_hash;
812 	    other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
813 
814 		if (zn == NULL) {
815 			zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
816 			    MT_NORMALIZE);
817 			allocdzn = B_TRUE;
818 		}
819 		if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
820 			if (allocdzn)
821 				zap_name_free(zn);
822 			return (B_TRUE);
823 		}
824 	}
825 
826 	if (direction == AVL_BEFORE) {
827 		direction = AVL_AFTER;
828 		goto again;
829 	}
830 
831 	if (allocdzn)
832 		zap_name_free(zn);
833 	return (B_FALSE);
834 }
835 
836 /*
837  * Routines for manipulating attributes.
838  */
839 
840 int
841 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
842     uint64_t integer_size, uint64_t num_integers, void *buf)
843 {
844 	return (zap_lookup_norm(os, zapobj, name, integer_size,
845 	    num_integers, buf, 0, NULL, 0, NULL));
846 }
847 
848 static int
849 zap_lookup_impl(zap_t *zap, const char *name,
850     uint64_t integer_size, uint64_t num_integers, void *buf,
851     matchtype_t mt, char *realname, int rn_len,
852     boolean_t *ncp)
853 {
854 	int err = 0;
855 
856 	zap_name_t *zn = zap_name_alloc(zap, name, mt);
857 	if (zn == NULL)
858 		return (SET_ERROR(ENOTSUP));
859 
860 	if (!zap->zap_ismicro) {
861 		err = fzap_lookup(zn, integer_size, num_integers, buf,
862 		    realname, rn_len, ncp);
863 	} else {
864 		mzap_ent_t *mze = mze_find(zn);
865 		if (mze == NULL) {
866 			err = SET_ERROR(ENOENT);
867 		} else {
868 			if (num_integers < 1) {
869 				err = SET_ERROR(EOVERFLOW);
870 			} else if (integer_size != 8) {
871 				err = SET_ERROR(EINVAL);
872 			} else {
873 				*(uint64_t *)buf =
874 				    MZE_PHYS(zap, mze)->mze_value;
875 				(void) strlcpy(realname,
876 				    MZE_PHYS(zap, mze)->mze_name, rn_len);
877 				if (ncp) {
878 					*ncp = mzap_normalization_conflict(zap,
879 					    zn, mze);
880 				}
881 			}
882 		}
883 	}
884 	zap_name_free(zn);
885 	return (err);
886 }
887 
888 int
889 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
890     uint64_t integer_size, uint64_t num_integers, void *buf,
891     matchtype_t mt, char *realname, int rn_len,
892     boolean_t *ncp)
893 {
894 	zap_t *zap;
895 
896 	int err =
897 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
898 	if (err != 0)
899 		return (err);
900 	err = zap_lookup_impl(zap, name, integer_size,
901 	    num_integers, buf, mt, realname, rn_len, ncp);
902 	zap_unlockdir(zap, FTAG);
903 	return (err);
904 }
905 
906 int
907 zap_lookup_by_dnode(dnode_t *dn, const char *name,
908     uint64_t integer_size, uint64_t num_integers, void *buf)
909 {
910 	return (zap_lookup_norm_by_dnode(dn, name, integer_size,
911 	    num_integers, buf, 0, NULL, 0, NULL));
912 }
913 
914 int
915 zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
916     uint64_t integer_size, uint64_t num_integers, void *buf,
917     matchtype_t mt, char *realname, int rn_len,
918     boolean_t *ncp)
919 {
920 	zap_t *zap;
921 
922 	int err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
923 	    FTAG, &zap);
924 	if (err != 0)
925 		return (err);
926 	err = zap_lookup_impl(zap, name, integer_size,
927 	    num_integers, buf, mt, realname, rn_len, ncp);
928 	zap_unlockdir(zap, FTAG);
929 	return (err);
930 }
931 
932 int
933 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
934     int key_numints)
935 {
936 	zap_t *zap;
937 
938 	int err =
939 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
940 	if (err != 0)
941 		return (err);
942 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
943 	if (zn == NULL) {
944 		zap_unlockdir(zap, FTAG);
945 		return (SET_ERROR(ENOTSUP));
946 	}
947 
948 	fzap_prefetch(zn);
949 	zap_name_free(zn);
950 	zap_unlockdir(zap, FTAG);
951 	return (err);
952 }
953 
954 int
955 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
956     int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
957 {
958 	zap_t *zap;
959 
960 	int err =
961 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
962 	if (err != 0)
963 		return (err);
964 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
965 	if (zn == NULL) {
966 		zap_unlockdir(zap, FTAG);
967 		return (SET_ERROR(ENOTSUP));
968 	}
969 
970 	err = fzap_lookup(zn, integer_size, num_integers, buf,
971 	    NULL, 0, NULL);
972 	zap_name_free(zn);
973 	zap_unlockdir(zap, FTAG);
974 	return (err);
975 }
976 
977 int
978 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
979 {
980 	int err = zap_lookup_norm(os, zapobj, name, 0,
981 	    0, NULL, 0, NULL, 0, NULL);
982 	if (err == EOVERFLOW || err == EINVAL)
983 		err = 0; /* found, but skipped reading the value */
984 	return (err);
985 }
986 
987 int
988 zap_length(objset_t *os, uint64_t zapobj, const char *name,
989     uint64_t *integer_size, uint64_t *num_integers)
990 {
991 	zap_t *zap;
992 
993 	int err =
994 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
995 	if (err != 0)
996 		return (err);
997 	zap_name_t *zn = zap_name_alloc(zap, name, 0);
998 	if (zn == NULL) {
999 		zap_unlockdir(zap, FTAG);
1000 		return (SET_ERROR(ENOTSUP));
1001 	}
1002 	if (!zap->zap_ismicro) {
1003 		err = fzap_length(zn, integer_size, num_integers);
1004 	} else {
1005 		mzap_ent_t *mze = mze_find(zn);
1006 		if (mze == NULL) {
1007 			err = SET_ERROR(ENOENT);
1008 		} else {
1009 			if (integer_size)
1010 				*integer_size = 8;
1011 			if (num_integers)
1012 				*num_integers = 1;
1013 		}
1014 	}
1015 	zap_name_free(zn);
1016 	zap_unlockdir(zap, FTAG);
1017 	return (err);
1018 }
1019 
1020 int
1021 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1022     int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1023 {
1024 	zap_t *zap;
1025 
1026 	int err =
1027 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1028 	if (err != 0)
1029 		return (err);
1030 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1031 	if (zn == NULL) {
1032 		zap_unlockdir(zap, FTAG);
1033 		return (SET_ERROR(ENOTSUP));
1034 	}
1035 	err = fzap_length(zn, integer_size, num_integers);
1036 	zap_name_free(zn);
1037 	zap_unlockdir(zap, FTAG);
1038 	return (err);
1039 }
1040 
1041 static void
1042 mzap_addent(zap_name_t *zn, uint64_t value)
1043 {
1044 	zap_t *zap = zn->zn_zap;
1045 	int start = zap->zap_m.zap_alloc_next;
1046 
1047 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1048 
1049 #ifdef ZFS_DEBUG
1050 	for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1051 		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1052 		ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1053 	}
1054 #endif
1055 
1056 	uint32_t cd = mze_find_unused_cd(zap, zn->zn_hash);
1057 	/* given the limited size of the microzap, this can't happen */
1058 	ASSERT(cd < zap_maxcd(zap));
1059 
1060 again:
1061 	for (int i = start; i < zap->zap_m.zap_num_chunks; i++) {
1062 		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1063 		if (mze->mze_name[0] == 0) {
1064 			mze->mze_value = value;
1065 			mze->mze_cd = cd;
1066 			(void) strcpy(mze->mze_name, zn->zn_key_orig);
1067 			zap->zap_m.zap_num_entries++;
1068 			zap->zap_m.zap_alloc_next = i+1;
1069 			if (zap->zap_m.zap_alloc_next ==
1070 			    zap->zap_m.zap_num_chunks)
1071 				zap->zap_m.zap_alloc_next = 0;
1072 			mze_insert(zap, i, zn->zn_hash);
1073 			return;
1074 		}
1075 	}
1076 	if (start != 0) {
1077 		start = 0;
1078 		goto again;
1079 	}
1080 	ASSERT(!"out of entries!");
1081 }
1082 
1083 static int
1084 zap_add_impl(zap_t *zap, const char *key,
1085     int integer_size, uint64_t num_integers,
1086     const void *val, dmu_tx_t *tx, void *tag)
1087 {
1088 	const uint64_t *intval = val;
1089 	int err = 0;
1090 
1091 	zap_name_t *zn = zap_name_alloc(zap, key, 0);
1092 	if (zn == NULL) {
1093 		zap_unlockdir(zap, tag);
1094 		return (SET_ERROR(ENOTSUP));
1095 	}
1096 	if (!zap->zap_ismicro) {
1097 		err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1098 		zap = zn->zn_zap;	/* fzap_add() may change zap */
1099 	} else if (integer_size != 8 || num_integers != 1 ||
1100 	    strlen(key) >= MZAP_NAME_LEN) {
1101 		err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
1102 		if (err == 0) {
1103 			err = fzap_add(zn, integer_size, num_integers, val,
1104 			    tag, tx);
1105 		}
1106 		zap = zn->zn_zap;	/* fzap_add() may change zap */
1107 	} else {
1108 		if (mze_find(zn) != NULL) {
1109 			err = SET_ERROR(EEXIST);
1110 		} else {
1111 			mzap_addent(zn, *intval);
1112 		}
1113 	}
1114 	ASSERT(zap == zn->zn_zap);
1115 	zap_name_free(zn);
1116 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1117 		zap_unlockdir(zap, tag);
1118 	return (err);
1119 }
1120 
1121 int
1122 zap_add(objset_t *os, uint64_t zapobj, const char *key,
1123     int integer_size, uint64_t num_integers,
1124     const void *val, dmu_tx_t *tx)
1125 {
1126 	zap_t *zap;
1127 	int err;
1128 
1129 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1130 	if (err != 0)
1131 		return (err);
1132 	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1133 	/* zap_add_impl() calls zap_unlockdir() */
1134 	return (err);
1135 }
1136 
1137 int
1138 zap_add_by_dnode(dnode_t *dn, const char *key,
1139     int integer_size, uint64_t num_integers,
1140     const void *val, dmu_tx_t *tx)
1141 {
1142 	zap_t *zap;
1143 	int err;
1144 
1145 	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1146 	if (err != 0)
1147 		return (err);
1148 	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1149 	/* zap_add_impl() calls zap_unlockdir() */
1150 	return (err);
1151 }
1152 
1153 int
1154 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1155     int key_numints, int integer_size, uint64_t num_integers,
1156     const void *val, dmu_tx_t *tx)
1157 {
1158 	zap_t *zap;
1159 
1160 	int err =
1161 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1162 	if (err != 0)
1163 		return (err);
1164 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1165 	if (zn == NULL) {
1166 		zap_unlockdir(zap, FTAG);
1167 		return (SET_ERROR(ENOTSUP));
1168 	}
1169 	err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1170 	zap = zn->zn_zap;	/* fzap_add() may change zap */
1171 	zap_name_free(zn);
1172 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1173 		zap_unlockdir(zap, FTAG);
1174 	return (err);
1175 }
1176 
1177 int
1178 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1179     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1180 {
1181 	zap_t *zap;
1182 	uint64_t oldval;
1183 	const uint64_t *intval = val;
1184 
1185 #ifdef ZFS_DEBUG
1186 	/*
1187 	 * If there is an old value, it shouldn't change across the
1188 	 * lockdir (eg, due to bprewrite's xlation).
1189 	 */
1190 	if (integer_size == 8 && num_integers == 1)
1191 		(void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1192 #endif
1193 
1194 	int err =
1195 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1196 	if (err != 0)
1197 		return (err);
1198 	zap_name_t *zn = zap_name_alloc(zap, name, 0);
1199 	if (zn == NULL) {
1200 		zap_unlockdir(zap, FTAG);
1201 		return (SET_ERROR(ENOTSUP));
1202 	}
1203 	if (!zap->zap_ismicro) {
1204 		err = fzap_update(zn, integer_size, num_integers, val,
1205 		    FTAG, tx);
1206 		zap = zn->zn_zap;	/* fzap_update() may change zap */
1207 	} else if (integer_size != 8 || num_integers != 1 ||
1208 	    strlen(name) >= MZAP_NAME_LEN) {
1209 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1210 		    zapobj, integer_size, num_integers, name);
1211 		err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1212 		if (err == 0) {
1213 			err = fzap_update(zn, integer_size, num_integers,
1214 			    val, FTAG, tx);
1215 		}
1216 		zap = zn->zn_zap;	/* fzap_update() may change zap */
1217 	} else {
1218 		mzap_ent_t *mze = mze_find(zn);
1219 		if (mze != NULL) {
1220 			ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1221 			MZE_PHYS(zap, mze)->mze_value = *intval;
1222 		} else {
1223 			mzap_addent(zn, *intval);
1224 		}
1225 	}
1226 	ASSERT(zap == zn->zn_zap);
1227 	zap_name_free(zn);
1228 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1229 		zap_unlockdir(zap, FTAG);
1230 	return (err);
1231 }
1232 
1233 int
1234 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1235     int key_numints,
1236     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1237 {
1238 	zap_t *zap;
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_uint64(zap, key, key_numints);
1245 	if (zn == NULL) {
1246 		zap_unlockdir(zap, FTAG);
1247 		return (SET_ERROR(ENOTSUP));
1248 	}
1249 	err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx);
1250 	zap = zn->zn_zap;	/* fzap_update() may change zap */
1251 	zap_name_free(zn);
1252 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1253 		zap_unlockdir(zap, FTAG);
1254 	return (err);
1255 }
1256 
1257 int
1258 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1259 {
1260 	return (zap_remove_norm(os, zapobj, name, 0, tx));
1261 }
1262 
1263 static int
1264 zap_remove_impl(zap_t *zap, const char *name,
1265     matchtype_t mt, dmu_tx_t *tx)
1266 {
1267 	int err = 0;
1268 
1269 	zap_name_t *zn = zap_name_alloc(zap, name, mt);
1270 	if (zn == NULL)
1271 		return (SET_ERROR(ENOTSUP));
1272 	if (!zap->zap_ismicro) {
1273 		err = fzap_remove(zn, tx);
1274 	} else {
1275 		mzap_ent_t *mze = mze_find(zn);
1276 		if (mze == NULL) {
1277 			err = SET_ERROR(ENOENT);
1278 		} else {
1279 			zap->zap_m.zap_num_entries--;
1280 			bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1281 			    sizeof (mzap_ent_phys_t));
1282 			mze_remove(zap, mze);
1283 		}
1284 	}
1285 	zap_name_free(zn);
1286 	return (err);
1287 }
1288 
1289 int
1290 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1291     matchtype_t mt, dmu_tx_t *tx)
1292 {
1293 	zap_t *zap;
1294 	int err;
1295 
1296 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1297 	if (err)
1298 		return (err);
1299 	err = zap_remove_impl(zap, name, mt, tx);
1300 	zap_unlockdir(zap, FTAG);
1301 	return (err);
1302 }
1303 
1304 int
1305 zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
1306 {
1307 	zap_t *zap;
1308 	int err;
1309 
1310 	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1311 	if (err)
1312 		return (err);
1313 	err = zap_remove_impl(zap, name, 0, tx);
1314 	zap_unlockdir(zap, FTAG);
1315 	return (err);
1316 }
1317 
1318 int
1319 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1320     int key_numints, dmu_tx_t *tx)
1321 {
1322 	zap_t *zap;
1323 
1324 	int err =
1325 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1326 	if (err != 0)
1327 		return (err);
1328 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1329 	if (zn == NULL) {
1330 		zap_unlockdir(zap, FTAG);
1331 		return (SET_ERROR(ENOTSUP));
1332 	}
1333 	err = fzap_remove(zn, tx);
1334 	zap_name_free(zn);
1335 	zap_unlockdir(zap, FTAG);
1336 	return (err);
1337 }
1338 
1339 /*
1340  * Routines for iterating over the attributes.
1341  */
1342 
1343 void
1344 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1345     uint64_t serialized)
1346 {
1347 	zc->zc_objset = os;
1348 	zc->zc_zap = NULL;
1349 	zc->zc_leaf = NULL;
1350 	zc->zc_zapobj = zapobj;
1351 	zc->zc_serialized = serialized;
1352 	zc->zc_hash = 0;
1353 	zc->zc_cd = 0;
1354 }
1355 
1356 void
1357 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1358 {
1359 	zap_cursor_init_serialized(zc, os, zapobj, 0);
1360 }
1361 
1362 void
1363 zap_cursor_fini(zap_cursor_t *zc)
1364 {
1365 	if (zc->zc_zap) {
1366 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1367 		zap_unlockdir(zc->zc_zap, NULL);
1368 		zc->zc_zap = NULL;
1369 	}
1370 	if (zc->zc_leaf) {
1371 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1372 		zap_put_leaf(zc->zc_leaf);
1373 		zc->zc_leaf = NULL;
1374 	}
1375 	zc->zc_objset = NULL;
1376 }
1377 
1378 uint64_t
1379 zap_cursor_serialize(zap_cursor_t *zc)
1380 {
1381 	if (zc->zc_hash == -1ULL)
1382 		return (-1ULL);
1383 	if (zc->zc_zap == NULL)
1384 		return (zc->zc_serialized);
1385 	ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1386 	ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1387 
1388 	/*
1389 	 * We want to keep the high 32 bits of the cursor zero if we can, so
1390 	 * that 32-bit programs can access this.  So usually use a small
1391 	 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1392 	 * of the cursor.
1393 	 *
1394 	 * [ collision differentiator | zap_hashbits()-bit hash value ]
1395 	 */
1396 	return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1397 	    ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1398 }
1399 
1400 int
1401 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1402 {
1403 	int err;
1404 
1405 	if (zc->zc_hash == -1ULL)
1406 		return (SET_ERROR(ENOENT));
1407 
1408 	if (zc->zc_zap == NULL) {
1409 		int hb;
1410 		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1411 		    RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1412 		if (err != 0)
1413 			return (err);
1414 
1415 		/*
1416 		 * To support zap_cursor_init_serialized, advance, retrieve,
1417 		 * we must add to the existing zc_cd, which may already
1418 		 * be 1 due to the zap_cursor_advance.
1419 		 */
1420 		ASSERT(zc->zc_hash == 0);
1421 		hb = zap_hashbits(zc->zc_zap);
1422 		zc->zc_hash = zc->zc_serialized << (64 - hb);
1423 		zc->zc_cd += zc->zc_serialized >> hb;
1424 		if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1425 			zc->zc_cd = 0;
1426 	} else {
1427 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1428 	}
1429 	if (!zc->zc_zap->zap_ismicro) {
1430 		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1431 	} else {
1432 		avl_index_t idx;
1433 		mzap_ent_t mze_tofind;
1434 
1435 		mze_tofind.mze_hash = zc->zc_hash;
1436 		mze_tofind.mze_cd = zc->zc_cd;
1437 
1438 		mzap_ent_t *mze =
1439 		    avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1440 		if (mze == NULL) {
1441 			mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1442 			    idx, AVL_AFTER);
1443 		}
1444 		if (mze) {
1445 			mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1446 			ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1447 			za->za_normalization_conflict =
1448 			    mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1449 			za->za_integer_length = 8;
1450 			za->za_num_integers = 1;
1451 			za->za_first_integer = mzep->mze_value;
1452 			(void) strcpy(za->za_name, mzep->mze_name);
1453 			zc->zc_hash = mze->mze_hash;
1454 			zc->zc_cd = mze->mze_cd;
1455 			err = 0;
1456 		} else {
1457 			zc->zc_hash = -1ULL;
1458 			err = SET_ERROR(ENOENT);
1459 		}
1460 	}
1461 	rw_exit(&zc->zc_zap->zap_rwlock);
1462 	return (err);
1463 }
1464 
1465 void
1466 zap_cursor_advance(zap_cursor_t *zc)
1467 {
1468 	if (zc->zc_hash == -1ULL)
1469 		return;
1470 	zc->zc_cd++;
1471 }
1472 
1473 int
1474 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1475 {
1476 	zap_t *zap;
1477 
1478 	int err =
1479 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1480 	if (err != 0)
1481 		return (err);
1482 
1483 	bzero(zs, sizeof (zap_stats_t));
1484 
1485 	if (zap->zap_ismicro) {
1486 		zs->zs_blocksize = zap->zap_dbuf->db_size;
1487 		zs->zs_num_entries = zap->zap_m.zap_num_entries;
1488 		zs->zs_num_blocks = 1;
1489 	} else {
1490 		fzap_get_stats(zap, zs);
1491 	}
1492 	zap_unlockdir(zap, FTAG);
1493 	return (0);
1494 }
1495