xref: /freebsd/sys/contrib/openzfs/module/zfs/zap_micro.c (revision 1f1e2261e341e6ca6862f82261066ef1705f0a7a)
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 2017 Nexenta Systems, Inc.
27  */
28 
29 #include <sys/zio.h>
30 #include <sys/spa.h>
31 #include <sys/dmu.h>
32 #include <sys/zfs_context.h>
33 #include <sys/zap.h>
34 #include <sys/zap_impl.h>
35 #include <sys/zap_leaf.h>
36 #include <sys/avl.h>
37 #include <sys/arc.h>
38 #include <sys/dmu_objset.h>
39 
40 #ifdef _KERNEL
41 #include <sys/sunddi.h>
42 #endif
43 
44 static int mzap_upgrade(zap_t **zapp,
45     void *tag, dmu_tx_t *tx, zap_flags_t flags);
46 
47 uint64_t
48 zap_getflags(zap_t *zap)
49 {
50 	if (zap->zap_ismicro)
51 		return (0);
52 	return (zap_f_phys(zap)->zap_flags);
53 }
54 
55 int
56 zap_hashbits(zap_t *zap)
57 {
58 	if (zap_getflags(zap) & ZAP_FLAG_HASH64)
59 		return (48);
60 	else
61 		return (28);
62 }
63 
64 uint32_t
65 zap_maxcd(zap_t *zap)
66 {
67 	if (zap_getflags(zap) & ZAP_FLAG_HASH64)
68 		return ((1<<16)-1);
69 	else
70 		return (-1U);
71 }
72 
73 static uint64_t
74 zap_hash(zap_name_t *zn)
75 {
76 	zap_t *zap = zn->zn_zap;
77 	uint64_t h = 0;
78 
79 	if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
80 		ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
81 		h = *(uint64_t *)zn->zn_key_orig;
82 	} else {
83 		h = zap->zap_salt;
84 		ASSERT(h != 0);
85 		ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
86 
87 		if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
88 			const uint64_t *wp = zn->zn_key_norm;
89 
90 			ASSERT(zn->zn_key_intlen == 8);
91 			for (int i = 0; i < zn->zn_key_norm_numints;
92 			    wp++, i++) {
93 				uint64_t word = *wp;
94 
95 				for (int j = 0; j < zn->zn_key_intlen; j++) {
96 					h = (h >> 8) ^
97 					    zfs_crc64_table[(h ^ word) & 0xFF];
98 					word >>= NBBY;
99 				}
100 			}
101 		} else {
102 			const uint8_t *cp = zn->zn_key_norm;
103 
104 			/*
105 			 * We previously stored the terminating null on
106 			 * disk, but didn't hash it, so we need to
107 			 * continue to not hash it.  (The
108 			 * zn_key_*_numints includes the terminating
109 			 * null for non-binary keys.)
110 			 */
111 			int len = zn->zn_key_norm_numints - 1;
112 
113 			ASSERT(zn->zn_key_intlen == 1);
114 			for (int i = 0; i < len; cp++, i++) {
115 				h = (h >> 8) ^
116 				    zfs_crc64_table[(h ^ *cp) & 0xFF];
117 			}
118 		}
119 	}
120 	/*
121 	 * Don't use all 64 bits, since we need some in the cookie for
122 	 * the collision differentiator.  We MUST use the high bits,
123 	 * since those are the ones that we first pay attention to when
124 	 * choosing the bucket.
125 	 */
126 	h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
127 
128 	return (h);
129 }
130 
131 static int
132 zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags)
133 {
134 	ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
135 
136 	size_t inlen = strlen(name) + 1;
137 	size_t outlen = ZAP_MAXNAMELEN;
138 
139 	int err = 0;
140 	(void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
141 	    normflags | U8_TEXTPREP_IGNORE_NULL | U8_TEXTPREP_IGNORE_INVALID,
142 	    U8_UNICODE_LATEST, &err);
143 
144 	return (err);
145 }
146 
147 boolean_t
148 zap_match(zap_name_t *zn, const char *matchname)
149 {
150 	ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
151 
152 	if (zn->zn_matchtype & MT_NORMALIZE) {
153 		char norm[ZAP_MAXNAMELEN];
154 
155 		if (zap_normalize(zn->zn_zap, matchname, norm,
156 		    zn->zn_normflags) != 0)
157 			return (B_FALSE);
158 
159 		return (strcmp(zn->zn_key_norm, norm) == 0);
160 	} else {
161 		return (strcmp(zn->zn_key_orig, matchname) == 0);
162 	}
163 }
164 
165 void
166 zap_name_free(zap_name_t *zn)
167 {
168 	kmem_free(zn, sizeof (zap_name_t));
169 }
170 
171 zap_name_t *
172 zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
173 {
174 	zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
175 
176 	zn->zn_zap = zap;
177 	zn->zn_key_intlen = sizeof (*key);
178 	zn->zn_key_orig = key;
179 	zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
180 	zn->zn_matchtype = mt;
181 	zn->zn_normflags = zap->zap_normflags;
182 
183 	/*
184 	 * If we're dealing with a case sensitive lookup on a mixed or
185 	 * insensitive fs, remove U8_TEXTPREP_TOUPPER or the lookup
186 	 * will fold case to all caps overriding the lookup request.
187 	 */
188 	if (mt & MT_MATCH_CASE)
189 		zn->zn_normflags &= ~U8_TEXTPREP_TOUPPER;
190 
191 	if (zap->zap_normflags) {
192 		/*
193 		 * We *must* use zap_normflags because this normalization is
194 		 * what the hash is computed from.
195 		 */
196 		if (zap_normalize(zap, key, zn->zn_normbuf,
197 		    zap->zap_normflags) != 0) {
198 			zap_name_free(zn);
199 			return (NULL);
200 		}
201 		zn->zn_key_norm = zn->zn_normbuf;
202 		zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
203 	} else {
204 		if (mt != 0) {
205 			zap_name_free(zn);
206 			return (NULL);
207 		}
208 		zn->zn_key_norm = zn->zn_key_orig;
209 		zn->zn_key_norm_numints = zn->zn_key_orig_numints;
210 	}
211 
212 	zn->zn_hash = zap_hash(zn);
213 
214 	if (zap->zap_normflags != zn->zn_normflags) {
215 		/*
216 		 * We *must* use zn_normflags because this normalization is
217 		 * what the matching is based on.  (Not the hash!)
218 		 */
219 		if (zap_normalize(zap, key, zn->zn_normbuf,
220 		    zn->zn_normflags) != 0) {
221 			zap_name_free(zn);
222 			return (NULL);
223 		}
224 		zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
225 	}
226 
227 	return (zn);
228 }
229 
230 static zap_name_t *
231 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
232 {
233 	zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
234 
235 	ASSERT(zap->zap_normflags == 0);
236 	zn->zn_zap = zap;
237 	zn->zn_key_intlen = sizeof (*key);
238 	zn->zn_key_orig = zn->zn_key_norm = key;
239 	zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
240 	zn->zn_matchtype = 0;
241 
242 	zn->zn_hash = zap_hash(zn);
243 	return (zn);
244 }
245 
246 static void
247 mzap_byteswap(mzap_phys_t *buf, size_t size)
248 {
249 	buf->mz_block_type = BSWAP_64(buf->mz_block_type);
250 	buf->mz_salt = BSWAP_64(buf->mz_salt);
251 	buf->mz_normflags = BSWAP_64(buf->mz_normflags);
252 	int max = (size / MZAP_ENT_LEN) - 1;
253 	for (int i = 0; i < max; i++) {
254 		buf->mz_chunk[i].mze_value =
255 		    BSWAP_64(buf->mz_chunk[i].mze_value);
256 		buf->mz_chunk[i].mze_cd =
257 		    BSWAP_32(buf->mz_chunk[i].mze_cd);
258 	}
259 }
260 
261 void
262 zap_byteswap(void *buf, size_t size)
263 {
264 	uint64_t block_type = *(uint64_t *)buf;
265 
266 	if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
267 		/* ASSERT(magic == ZAP_LEAF_MAGIC); */
268 		mzap_byteswap(buf, size);
269 	} else {
270 		fzap_byteswap(buf, size);
271 	}
272 }
273 
274 static int
275 mze_compare(const void *arg1, const void *arg2)
276 {
277 	const mzap_ent_t *mze1 = arg1;
278 	const mzap_ent_t *mze2 = arg2;
279 
280 	int cmp = TREE_CMP(mze1->mze_hash, mze2->mze_hash);
281 	if (likely(cmp))
282 		return (cmp);
283 
284 	return (TREE_CMP(mze1->mze_cd, mze2->mze_cd));
285 }
286 
287 static void
288 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
289 {
290 	ASSERT(zap->zap_ismicro);
291 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
292 
293 	mzap_ent_t *mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
294 	mze->mze_chunkid = chunkid;
295 	mze->mze_hash = hash;
296 	mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
297 	ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
298 	avl_add(&zap->zap_m.zap_avl, mze);
299 }
300 
301 static mzap_ent_t *
302 mze_find(zap_name_t *zn)
303 {
304 	mzap_ent_t mze_tofind;
305 	mzap_ent_t *mze;
306 	avl_index_t idx;
307 	avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
308 
309 	ASSERT(zn->zn_zap->zap_ismicro);
310 	ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
311 
312 	mze_tofind.mze_hash = zn->zn_hash;
313 	mze_tofind.mze_cd = 0;
314 
315 	mze = avl_find(avl, &mze_tofind, &idx);
316 	if (mze == NULL)
317 		mze = avl_nearest(avl, idx, AVL_AFTER);
318 	for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
319 		ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
320 		if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
321 			return (mze);
322 	}
323 
324 	return (NULL);
325 }
326 
327 static uint32_t
328 mze_find_unused_cd(zap_t *zap, uint64_t hash)
329 {
330 	mzap_ent_t mze_tofind;
331 	avl_index_t idx;
332 	avl_tree_t *avl = &zap->zap_m.zap_avl;
333 
334 	ASSERT(zap->zap_ismicro);
335 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
336 
337 	mze_tofind.mze_hash = hash;
338 	mze_tofind.mze_cd = 0;
339 
340 	uint32_t cd = 0;
341 	for (mzap_ent_t *mze = avl_find(avl, &mze_tofind, &idx);
342 	    mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
343 		if (mze->mze_cd != cd)
344 			break;
345 		cd++;
346 	}
347 
348 	return (cd);
349 }
350 
351 /*
352  * Each mzap entry requires at max : 4 chunks
353  * 3 chunks for names + 1 chunk for value.
354  */
355 #define	MZAP_ENT_CHUNKS	(1 + ZAP_LEAF_ARRAY_NCHUNKS(MZAP_NAME_LEN) + \
356 	ZAP_LEAF_ARRAY_NCHUNKS(sizeof (uint64_t)))
357 
358 /*
359  * Check if the current entry keeps the colliding entries under the fatzap leaf
360  * size.
361  */
362 static boolean_t
363 mze_canfit_fzap_leaf(zap_name_t *zn, uint64_t hash)
364 {
365 	zap_t *zap = zn->zn_zap;
366 	mzap_ent_t mze_tofind;
367 	mzap_ent_t *mze;
368 	avl_index_t idx;
369 	avl_tree_t *avl = &zap->zap_m.zap_avl;
370 	uint32_t mzap_ents = 0;
371 
372 	mze_tofind.mze_hash = hash;
373 	mze_tofind.mze_cd = 0;
374 
375 	for (mze = avl_find(avl, &mze_tofind, &idx);
376 	    mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
377 		mzap_ents++;
378 	}
379 
380 	/* Include the new entry being added */
381 	mzap_ents++;
382 
383 	return (ZAP_LEAF_NUMCHUNKS_DEF > (mzap_ents * MZAP_ENT_CHUNKS));
384 }
385 
386 static void
387 mze_remove(zap_t *zap, mzap_ent_t *mze)
388 {
389 	ASSERT(zap->zap_ismicro);
390 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
391 
392 	avl_remove(&zap->zap_m.zap_avl, mze);
393 	kmem_free(mze, sizeof (mzap_ent_t));
394 }
395 
396 static void
397 mze_destroy(zap_t *zap)
398 {
399 	mzap_ent_t *mze;
400 	void *avlcookie = NULL;
401 
402 	while ((mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie)))
403 		kmem_free(mze, sizeof (mzap_ent_t));
404 	avl_destroy(&zap->zap_m.zap_avl);
405 }
406 
407 static zap_t *
408 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
409 {
410 	zap_t *winner;
411 	uint64_t *zap_hdr = (uint64_t *)db->db_data;
412 	uint64_t zap_block_type = zap_hdr[0];
413 	uint64_t zap_magic = zap_hdr[1];
414 
415 	ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
416 
417 	zap_t *zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
418 	rw_init(&zap->zap_rwlock, NULL, RW_DEFAULT, NULL);
419 	rw_enter(&zap->zap_rwlock, RW_WRITER);
420 	zap->zap_objset = os;
421 	zap->zap_object = obj;
422 	zap->zap_dbuf = db;
423 
424 	if (zap_block_type != ZBT_MICRO) {
425 		mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, MUTEX_DEFAULT,
426 		    0);
427 		zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
428 		if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
429 			winner = NULL;	/* No actual winner here... */
430 			goto handle_winner;
431 		}
432 	} else {
433 		zap->zap_ismicro = TRUE;
434 	}
435 
436 	/*
437 	 * Make sure that zap_ismicro is set before we let others see
438 	 * it, because zap_lockdir() checks zap_ismicro without the lock
439 	 * held.
440 	 */
441 	dmu_buf_init_user(&zap->zap_dbu, zap_evict_sync, NULL, &zap->zap_dbuf);
442 	winner = dmu_buf_set_user(db, &zap->zap_dbu);
443 
444 	if (winner != NULL)
445 		goto handle_winner;
446 
447 	if (zap->zap_ismicro) {
448 		zap->zap_salt = zap_m_phys(zap)->mz_salt;
449 		zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
450 		zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
451 		avl_create(&zap->zap_m.zap_avl, mze_compare,
452 		    sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
453 
454 		for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
455 			mzap_ent_phys_t *mze =
456 			    &zap_m_phys(zap)->mz_chunk[i];
457 			if (mze->mze_name[0]) {
458 				zap_name_t *zn;
459 
460 				zap->zap_m.zap_num_entries++;
461 				zn = zap_name_alloc(zap, mze->mze_name, 0);
462 				mze_insert(zap, i, zn->zn_hash);
463 				zap_name_free(zn);
464 			}
465 		}
466 	} else {
467 		zap->zap_salt = zap_f_phys(zap)->zap_salt;
468 		zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
469 
470 		ASSERT3U(sizeof (struct zap_leaf_header), ==,
471 		    2*ZAP_LEAF_CHUNKSIZE);
472 
473 		/*
474 		 * The embedded pointer table should not overlap the
475 		 * other members.
476 		 */
477 		ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
478 		    &zap_f_phys(zap)->zap_salt);
479 
480 		/*
481 		 * The embedded pointer table should end at the end of
482 		 * the block
483 		 */
484 		ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
485 		    1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
486 		    (uintptr_t)zap_f_phys(zap), ==,
487 		    zap->zap_dbuf->db_size);
488 	}
489 	rw_exit(&zap->zap_rwlock);
490 	return (zap);
491 
492 handle_winner:
493 	rw_exit(&zap->zap_rwlock);
494 	rw_destroy(&zap->zap_rwlock);
495 	if (!zap->zap_ismicro)
496 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
497 	kmem_free(zap, sizeof (zap_t));
498 	return (winner);
499 }
500 
501 /*
502  * This routine "consumes" the caller's hold on the dbuf, which must
503  * have the specified tag.
504  */
505 static int
506 zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx,
507     krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
508 {
509 	ASSERT0(db->db_offset);
510 	objset_t *os = dmu_buf_get_objset(db);
511 	uint64_t obj = db->db_object;
512 	dmu_object_info_t doi;
513 
514 	*zapp = NULL;
515 
516 	dmu_object_info_from_db(db, &doi);
517 	if (DMU_OT_BYTESWAP(doi.doi_type) != DMU_BSWAP_ZAP)
518 		return (SET_ERROR(EINVAL));
519 
520 	zap_t *zap = dmu_buf_get_user(db);
521 	if (zap == NULL) {
522 		zap = mzap_open(os, obj, db);
523 		if (zap == NULL) {
524 			/*
525 			 * mzap_open() didn't like what it saw on-disk.
526 			 * Check for corruption!
527 			 */
528 			return (SET_ERROR(EIO));
529 		}
530 	}
531 
532 	/*
533 	 * We're checking zap_ismicro without the lock held, in order to
534 	 * tell what type of lock we want.  Once we have some sort of
535 	 * lock, see if it really is the right type.  In practice this
536 	 * can only be different if it was upgraded from micro to fat,
537 	 * and micro wanted WRITER but fat only needs READER.
538 	 */
539 	krw_t lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
540 	rw_enter(&zap->zap_rwlock, lt);
541 	if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
542 		/* it was upgraded, now we only need reader */
543 		ASSERT(lt == RW_WRITER);
544 		ASSERT(RW_READER ==
545 		    ((!zap->zap_ismicro && fatreader) ? RW_READER : lti));
546 		rw_downgrade(&zap->zap_rwlock);
547 		lt = RW_READER;
548 	}
549 
550 	zap->zap_objset = os;
551 
552 	if (lt == RW_WRITER)
553 		dmu_buf_will_dirty(db, tx);
554 
555 	ASSERT3P(zap->zap_dbuf, ==, db);
556 
557 	ASSERT(!zap->zap_ismicro ||
558 	    zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
559 	if (zap->zap_ismicro && tx && adding &&
560 	    zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
561 		uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
562 		if (newsz > MZAP_MAX_BLKSZ) {
563 			dprintf("upgrading obj %llu: num_entries=%u\n",
564 			    (u_longlong_t)obj, zap->zap_m.zap_num_entries);
565 			*zapp = zap;
566 			int err = mzap_upgrade(zapp, tag, tx, 0);
567 			if (err != 0)
568 				rw_exit(&zap->zap_rwlock);
569 			return (err);
570 		}
571 		VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx));
572 		zap->zap_m.zap_num_chunks =
573 		    db->db_size / MZAP_ENT_LEN - 1;
574 	}
575 
576 	*zapp = zap;
577 	return (0);
578 }
579 
580 static int
581 zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
582     krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
583 {
584 	dmu_buf_t *db;
585 
586 	int err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
587 	if (err != 0) {
588 		return (err);
589 	}
590 #ifdef ZFS_DEBUG
591 	{
592 		dmu_object_info_t doi;
593 		dmu_object_info_from_db(db, &doi);
594 		ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
595 	}
596 #endif
597 
598 	err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
599 	if (err != 0) {
600 		dmu_buf_rele(db, tag);
601 	}
602 	return (err);
603 }
604 
605 int
606 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
607     krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
608 {
609 	dmu_buf_t *db;
610 
611 	int err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH);
612 	if (err != 0)
613 		return (err);
614 #ifdef ZFS_DEBUG
615 	{
616 		dmu_object_info_t doi;
617 		dmu_object_info_from_db(db, &doi);
618 		ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
619 	}
620 #endif
621 	err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
622 	if (err != 0)
623 		dmu_buf_rele(db, tag);
624 	return (err);
625 }
626 
627 void
628 zap_unlockdir(zap_t *zap, void *tag)
629 {
630 	rw_exit(&zap->zap_rwlock);
631 	dmu_buf_rele(zap->zap_dbuf, tag);
632 }
633 
634 static int
635 mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags)
636 {
637 	int err = 0;
638 	zap_t *zap = *zapp;
639 
640 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
641 
642 	int sz = zap->zap_dbuf->db_size;
643 	mzap_phys_t *mzp = vmem_alloc(sz, KM_SLEEP);
644 	memcpy(mzp, zap->zap_dbuf->db_data, sz);
645 	int nchunks = zap->zap_m.zap_num_chunks;
646 
647 	if (!flags) {
648 		err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
649 		    1ULL << fzap_default_block_shift, 0, tx);
650 		if (err != 0) {
651 			vmem_free(mzp, sz);
652 			return (err);
653 		}
654 	}
655 
656 	dprintf("upgrading obj=%llu with %u chunks\n",
657 	    (u_longlong_t)zap->zap_object, nchunks);
658 	/* XXX destroy the avl later, so we can use the stored hash value */
659 	mze_destroy(zap);
660 
661 	fzap_upgrade(zap, tx, flags);
662 
663 	for (int i = 0; i < nchunks; i++) {
664 		mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
665 		if (mze->mze_name[0] == 0)
666 			continue;
667 		dprintf("adding %s=%llu\n",
668 		    mze->mze_name, (u_longlong_t)mze->mze_value);
669 		zap_name_t *zn = zap_name_alloc(zap, mze->mze_name, 0);
670 		/* If we fail here, we would end up losing entries */
671 		VERIFY0(fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
672 		    tag, tx));
673 		zap = zn->zn_zap;	/* fzap_add_cd() may change zap */
674 		zap_name_free(zn);
675 	}
676 	vmem_free(mzp, sz);
677 	*zapp = zap;
678 	return (0);
679 }
680 
681 /*
682  * The "normflags" determine the behavior of the matchtype_t which is
683  * passed to zap_lookup_norm().  Names which have the same normalized
684  * version will be stored with the same hash value, and therefore we can
685  * perform normalization-insensitive lookups.  We can be Unicode form-
686  * insensitive and/or case-insensitive.  The following flags are valid for
687  * "normflags":
688  *
689  * U8_TEXTPREP_NFC
690  * U8_TEXTPREP_NFD
691  * U8_TEXTPREP_NFKC
692  * U8_TEXTPREP_NFKD
693  * U8_TEXTPREP_TOUPPER
694  *
695  * The *_NF* (Normalization Form) flags are mutually exclusive; at most one
696  * of them may be supplied.
697  */
698 void
699 mzap_create_impl(dnode_t *dn, int normflags, zap_flags_t flags, dmu_tx_t *tx)
700 {
701 	dmu_buf_t *db;
702 
703 	VERIFY0(dmu_buf_hold_by_dnode(dn, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
704 
705 	dmu_buf_will_dirty(db, tx);
706 	mzap_phys_t *zp = db->db_data;
707 	zp->mz_block_type = ZBT_MICRO;
708 	zp->mz_salt =
709 	    ((uintptr_t)db ^ (uintptr_t)tx ^ (dn->dn_object << 1)) | 1ULL;
710 	zp->mz_normflags = normflags;
711 
712 	if (flags != 0) {
713 		zap_t *zap;
714 		/* Only fat zap supports flags; upgrade immediately. */
715 		VERIFY0(zap_lockdir_impl(db, FTAG, tx, RW_WRITER,
716 		    B_FALSE, B_FALSE, &zap));
717 		VERIFY0(mzap_upgrade(&zap, FTAG, tx, flags));
718 		zap_unlockdir(zap, FTAG);
719 	} else {
720 		dmu_buf_rele(db, FTAG);
721 	}
722 }
723 
724 static uint64_t
725 zap_create_impl(objset_t *os, int normflags, zap_flags_t flags,
726     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
727     dmu_object_type_t bonustype, int bonuslen, int dnodesize,
728     dnode_t **allocated_dnode, void *tag, dmu_tx_t *tx)
729 {
730 	uint64_t obj;
731 
732 	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
733 
734 	if (allocated_dnode == NULL) {
735 		dnode_t *dn;
736 		obj = dmu_object_alloc_hold(os, ot, 1ULL << leaf_blockshift,
737 		    indirect_blockshift, bonustype, bonuslen, dnodesize,
738 		    &dn, FTAG, tx);
739 		mzap_create_impl(dn, normflags, flags, tx);
740 		dnode_rele(dn, FTAG);
741 	} else {
742 		obj = dmu_object_alloc_hold(os, ot, 1ULL << leaf_blockshift,
743 		    indirect_blockshift, bonustype, bonuslen, dnodesize,
744 		    allocated_dnode, tag, tx);
745 		mzap_create_impl(*allocated_dnode, normflags, flags, tx);
746 	}
747 
748 	return (obj);
749 }
750 
751 int
752 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
753     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
754 {
755 	return (zap_create_claim_dnsize(os, obj, ot, bonustype, bonuslen,
756 	    0, tx));
757 }
758 
759 int
760 zap_create_claim_dnsize(objset_t *os, uint64_t obj, dmu_object_type_t ot,
761     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
762 {
763 	return (zap_create_claim_norm_dnsize(os, obj,
764 	    0, ot, bonustype, bonuslen, dnodesize, tx));
765 }
766 
767 int
768 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
769     dmu_object_type_t ot,
770     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
771 {
772 	return (zap_create_claim_norm_dnsize(os, obj, normflags, ot, bonustype,
773 	    bonuslen, 0, tx));
774 }
775 
776 int
777 zap_create_claim_norm_dnsize(objset_t *os, uint64_t obj, int normflags,
778     dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
779     int dnodesize, dmu_tx_t *tx)
780 {
781 	dnode_t *dn;
782 	int error;
783 
784 	ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
785 	error = dmu_object_claim_dnsize(os, obj, ot, 0, bonustype, bonuslen,
786 	    dnodesize, tx);
787 	if (error != 0)
788 		return (error);
789 
790 	error = dnode_hold(os, obj, FTAG, &dn);
791 	if (error != 0)
792 		return (error);
793 
794 	mzap_create_impl(dn, normflags, 0, tx);
795 
796 	dnode_rele(dn, FTAG);
797 
798 	return (0);
799 }
800 
801 uint64_t
802 zap_create(objset_t *os, dmu_object_type_t ot,
803     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
804 {
805 	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
806 }
807 
808 uint64_t
809 zap_create_dnsize(objset_t *os, dmu_object_type_t ot,
810     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
811 {
812 	return (zap_create_norm_dnsize(os, 0, ot, bonustype, bonuslen,
813 	    dnodesize, tx));
814 }
815 
816 uint64_t
817 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
818     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
819 {
820 	return (zap_create_norm_dnsize(os, normflags, ot, bonustype, bonuslen,
821 	    0, tx));
822 }
823 
824 uint64_t
825 zap_create_norm_dnsize(objset_t *os, int normflags, dmu_object_type_t ot,
826     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
827 {
828 	return (zap_create_impl(os, normflags, 0, ot, 0, 0,
829 	    bonustype, bonuslen, dnodesize, NULL, NULL, tx));
830 }
831 
832 uint64_t
833 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
834     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
835     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
836 {
837 	return (zap_create_flags_dnsize(os, normflags, flags, ot,
838 	    leaf_blockshift, indirect_blockshift, bonustype, bonuslen, 0, tx));
839 }
840 
841 uint64_t
842 zap_create_flags_dnsize(objset_t *os, int normflags, zap_flags_t flags,
843     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
844     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
845 {
846 	return (zap_create_impl(os, normflags, flags, ot, leaf_blockshift,
847 	    indirect_blockshift, bonustype, bonuslen, dnodesize, NULL, NULL,
848 	    tx));
849 }
850 
851 /*
852  * Create a zap object and return a pointer to the newly allocated dnode via
853  * the allocated_dnode argument.  The returned dnode will be held and the
854  * caller is responsible for releasing the hold by calling dnode_rele().
855  */
856 uint64_t
857 zap_create_hold(objset_t *os, int normflags, zap_flags_t flags,
858     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
859     dmu_object_type_t bonustype, int bonuslen, int dnodesize,
860     dnode_t **allocated_dnode, void *tag, dmu_tx_t *tx)
861 {
862 	return (zap_create_impl(os, normflags, flags, ot, leaf_blockshift,
863 	    indirect_blockshift, bonustype, bonuslen, dnodesize,
864 	    allocated_dnode, tag, tx));
865 }
866 
867 int
868 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
869 {
870 	/*
871 	 * dmu_object_free will free the object number and free the
872 	 * data.  Freeing the data will cause our pageout function to be
873 	 * called, which will destroy our data (zap_leaf_t's and zap_t).
874 	 */
875 
876 	return (dmu_object_free(os, zapobj, tx));
877 }
878 
879 void
880 zap_evict_sync(void *dbu)
881 {
882 	zap_t *zap = dbu;
883 
884 	rw_destroy(&zap->zap_rwlock);
885 
886 	if (zap->zap_ismicro)
887 		mze_destroy(zap);
888 	else
889 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
890 
891 	kmem_free(zap, sizeof (zap_t));
892 }
893 
894 int
895 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
896 {
897 	zap_t *zap;
898 
899 	int err =
900 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
901 	if (err != 0)
902 		return (err);
903 	if (!zap->zap_ismicro) {
904 		err = fzap_count(zap, count);
905 	} else {
906 		*count = zap->zap_m.zap_num_entries;
907 	}
908 	zap_unlockdir(zap, FTAG);
909 	return (err);
910 }
911 
912 /*
913  * zn may be NULL; if not specified, it will be computed if needed.
914  * See also the comment above zap_entry_normalization_conflict().
915  */
916 static boolean_t
917 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
918 {
919 	int direction = AVL_BEFORE;
920 	boolean_t allocdzn = B_FALSE;
921 
922 	if (zap->zap_normflags == 0)
923 		return (B_FALSE);
924 
925 again:
926 	for (mzap_ent_t *other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
927 	    other && other->mze_hash == mze->mze_hash;
928 	    other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
929 
930 		if (zn == NULL) {
931 			zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
932 			    MT_NORMALIZE);
933 			allocdzn = B_TRUE;
934 		}
935 		if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
936 			if (allocdzn)
937 				zap_name_free(zn);
938 			return (B_TRUE);
939 		}
940 	}
941 
942 	if (direction == AVL_BEFORE) {
943 		direction = AVL_AFTER;
944 		goto again;
945 	}
946 
947 	if (allocdzn)
948 		zap_name_free(zn);
949 	return (B_FALSE);
950 }
951 
952 /*
953  * Routines for manipulating attributes.
954  */
955 
956 int
957 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
958     uint64_t integer_size, uint64_t num_integers, void *buf)
959 {
960 	return (zap_lookup_norm(os, zapobj, name, integer_size,
961 	    num_integers, buf, 0, NULL, 0, NULL));
962 }
963 
964 static int
965 zap_lookup_impl(zap_t *zap, const char *name,
966     uint64_t integer_size, uint64_t num_integers, void *buf,
967     matchtype_t mt, char *realname, int rn_len,
968     boolean_t *ncp)
969 {
970 	int err = 0;
971 
972 	zap_name_t *zn = zap_name_alloc(zap, name, mt);
973 	if (zn == NULL)
974 		return (SET_ERROR(ENOTSUP));
975 
976 	if (!zap->zap_ismicro) {
977 		err = fzap_lookup(zn, integer_size, num_integers, buf,
978 		    realname, rn_len, ncp);
979 	} else {
980 		mzap_ent_t *mze = mze_find(zn);
981 		if (mze == NULL) {
982 			err = SET_ERROR(ENOENT);
983 		} else {
984 			if (num_integers < 1) {
985 				err = SET_ERROR(EOVERFLOW);
986 			} else if (integer_size != 8) {
987 				err = SET_ERROR(EINVAL);
988 			} else {
989 				*(uint64_t *)buf =
990 				    MZE_PHYS(zap, mze)->mze_value;
991 				(void) strlcpy(realname,
992 				    MZE_PHYS(zap, mze)->mze_name, rn_len);
993 				if (ncp) {
994 					*ncp = mzap_normalization_conflict(zap,
995 					    zn, mze);
996 				}
997 			}
998 		}
999 	}
1000 	zap_name_free(zn);
1001 	return (err);
1002 }
1003 
1004 int
1005 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
1006     uint64_t integer_size, uint64_t num_integers, void *buf,
1007     matchtype_t mt, char *realname, int rn_len,
1008     boolean_t *ncp)
1009 {
1010 	zap_t *zap;
1011 
1012 	int err =
1013 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1014 	if (err != 0)
1015 		return (err);
1016 	err = zap_lookup_impl(zap, name, integer_size,
1017 	    num_integers, buf, mt, realname, rn_len, ncp);
1018 	zap_unlockdir(zap, FTAG);
1019 	return (err);
1020 }
1021 
1022 int
1023 zap_prefetch(objset_t *os, uint64_t zapobj, const char *name)
1024 {
1025 	zap_t *zap;
1026 	int err;
1027 	zap_name_t *zn;
1028 
1029 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1030 	if (err)
1031 		return (err);
1032 	zn = zap_name_alloc(zap, name, 0);
1033 	if (zn == NULL) {
1034 		zap_unlockdir(zap, FTAG);
1035 		return (SET_ERROR(ENOTSUP));
1036 	}
1037 
1038 	fzap_prefetch(zn);
1039 	zap_name_free(zn);
1040 	zap_unlockdir(zap, FTAG);
1041 	return (err);
1042 }
1043 
1044 int
1045 zap_lookup_by_dnode(dnode_t *dn, const char *name,
1046     uint64_t integer_size, uint64_t num_integers, void *buf)
1047 {
1048 	return (zap_lookup_norm_by_dnode(dn, name, integer_size,
1049 	    num_integers, buf, 0, NULL, 0, NULL));
1050 }
1051 
1052 int
1053 zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
1054     uint64_t integer_size, uint64_t num_integers, void *buf,
1055     matchtype_t mt, char *realname, int rn_len,
1056     boolean_t *ncp)
1057 {
1058 	zap_t *zap;
1059 
1060 	int err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
1061 	    FTAG, &zap);
1062 	if (err != 0)
1063 		return (err);
1064 	err = zap_lookup_impl(zap, name, integer_size,
1065 	    num_integers, buf, mt, realname, rn_len, ncp);
1066 	zap_unlockdir(zap, FTAG);
1067 	return (err);
1068 }
1069 
1070 int
1071 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1072     int key_numints)
1073 {
1074 	zap_t *zap;
1075 
1076 	int err =
1077 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1078 	if (err != 0)
1079 		return (err);
1080 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1081 	if (zn == NULL) {
1082 		zap_unlockdir(zap, FTAG);
1083 		return (SET_ERROR(ENOTSUP));
1084 	}
1085 
1086 	fzap_prefetch(zn);
1087 	zap_name_free(zn);
1088 	zap_unlockdir(zap, FTAG);
1089 	return (err);
1090 }
1091 
1092 int
1093 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1094     int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
1095 {
1096 	zap_t *zap;
1097 
1098 	int err =
1099 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1100 	if (err != 0)
1101 		return (err);
1102 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1103 	if (zn == NULL) {
1104 		zap_unlockdir(zap, FTAG);
1105 		return (SET_ERROR(ENOTSUP));
1106 	}
1107 
1108 	err = fzap_lookup(zn, integer_size, num_integers, buf,
1109 	    NULL, 0, NULL);
1110 	zap_name_free(zn);
1111 	zap_unlockdir(zap, FTAG);
1112 	return (err);
1113 }
1114 
1115 int
1116 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
1117 {
1118 	int err = zap_lookup_norm(os, zapobj, name, 0,
1119 	    0, NULL, 0, NULL, 0, NULL);
1120 	if (err == EOVERFLOW || err == EINVAL)
1121 		err = 0; /* found, but skipped reading the value */
1122 	return (err);
1123 }
1124 
1125 int
1126 zap_length(objset_t *os, uint64_t zapobj, const char *name,
1127     uint64_t *integer_size, uint64_t *num_integers)
1128 {
1129 	zap_t *zap;
1130 
1131 	int err =
1132 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1133 	if (err != 0)
1134 		return (err);
1135 	zap_name_t *zn = zap_name_alloc(zap, name, 0);
1136 	if (zn == NULL) {
1137 		zap_unlockdir(zap, FTAG);
1138 		return (SET_ERROR(ENOTSUP));
1139 	}
1140 	if (!zap->zap_ismicro) {
1141 		err = fzap_length(zn, integer_size, num_integers);
1142 	} else {
1143 		mzap_ent_t *mze = mze_find(zn);
1144 		if (mze == NULL) {
1145 			err = SET_ERROR(ENOENT);
1146 		} else {
1147 			if (integer_size)
1148 				*integer_size = 8;
1149 			if (num_integers)
1150 				*num_integers = 1;
1151 		}
1152 	}
1153 	zap_name_free(zn);
1154 	zap_unlockdir(zap, FTAG);
1155 	return (err);
1156 }
1157 
1158 int
1159 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1160     int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1161 {
1162 	zap_t *zap;
1163 
1164 	int err =
1165 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1166 	if (err != 0)
1167 		return (err);
1168 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1169 	if (zn == NULL) {
1170 		zap_unlockdir(zap, FTAG);
1171 		return (SET_ERROR(ENOTSUP));
1172 	}
1173 	err = fzap_length(zn, integer_size, num_integers);
1174 	zap_name_free(zn);
1175 	zap_unlockdir(zap, FTAG);
1176 	return (err);
1177 }
1178 
1179 static void
1180 mzap_addent(zap_name_t *zn, uint64_t value)
1181 {
1182 	zap_t *zap = zn->zn_zap;
1183 	int start = zap->zap_m.zap_alloc_next;
1184 
1185 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1186 
1187 #ifdef ZFS_DEBUG
1188 	for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1189 		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1190 		ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1191 	}
1192 #endif
1193 
1194 	uint32_t cd = mze_find_unused_cd(zap, zn->zn_hash);
1195 	/* given the limited size of the microzap, this can't happen */
1196 	ASSERT(cd < zap_maxcd(zap));
1197 
1198 again:
1199 	for (int i = start; i < zap->zap_m.zap_num_chunks; i++) {
1200 		mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1201 		if (mze->mze_name[0] == 0) {
1202 			mze->mze_value = value;
1203 			mze->mze_cd = cd;
1204 			(void) strlcpy(mze->mze_name, zn->zn_key_orig,
1205 			    sizeof (mze->mze_name));
1206 			zap->zap_m.zap_num_entries++;
1207 			zap->zap_m.zap_alloc_next = i+1;
1208 			if (zap->zap_m.zap_alloc_next ==
1209 			    zap->zap_m.zap_num_chunks)
1210 				zap->zap_m.zap_alloc_next = 0;
1211 			mze_insert(zap, i, zn->zn_hash);
1212 			return;
1213 		}
1214 	}
1215 	if (start != 0) {
1216 		start = 0;
1217 		goto again;
1218 	}
1219 	cmn_err(CE_PANIC, "out of entries!");
1220 }
1221 
1222 static int
1223 zap_add_impl(zap_t *zap, const char *key,
1224     int integer_size, uint64_t num_integers,
1225     const void *val, dmu_tx_t *tx, void *tag)
1226 {
1227 	const uint64_t *intval = val;
1228 	int err = 0;
1229 
1230 	zap_name_t *zn = zap_name_alloc(zap, key, 0);
1231 	if (zn == NULL) {
1232 		zap_unlockdir(zap, tag);
1233 		return (SET_ERROR(ENOTSUP));
1234 	}
1235 	if (!zap->zap_ismicro) {
1236 		err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1237 		zap = zn->zn_zap;	/* fzap_add() may change zap */
1238 	} else if (integer_size != 8 || num_integers != 1 ||
1239 	    strlen(key) >= MZAP_NAME_LEN ||
1240 	    !mze_canfit_fzap_leaf(zn, zn->zn_hash)) {
1241 		err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
1242 		if (err == 0) {
1243 			err = fzap_add(zn, integer_size, num_integers, val,
1244 			    tag, tx);
1245 		}
1246 		zap = zn->zn_zap;	/* fzap_add() may change zap */
1247 	} else {
1248 		if (mze_find(zn) != NULL) {
1249 			err = SET_ERROR(EEXIST);
1250 		} else {
1251 			mzap_addent(zn, *intval);
1252 		}
1253 	}
1254 	ASSERT(zap == zn->zn_zap);
1255 	zap_name_free(zn);
1256 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1257 		zap_unlockdir(zap, tag);
1258 	return (err);
1259 }
1260 
1261 int
1262 zap_add(objset_t *os, uint64_t zapobj, const char *key,
1263     int integer_size, uint64_t num_integers,
1264     const void *val, dmu_tx_t *tx)
1265 {
1266 	zap_t *zap;
1267 	int err;
1268 
1269 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1270 	if (err != 0)
1271 		return (err);
1272 	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1273 	/* zap_add_impl() calls zap_unlockdir() */
1274 	return (err);
1275 }
1276 
1277 int
1278 zap_add_by_dnode(dnode_t *dn, const char *key,
1279     int integer_size, uint64_t num_integers,
1280     const void *val, dmu_tx_t *tx)
1281 {
1282 	zap_t *zap;
1283 	int err;
1284 
1285 	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1286 	if (err != 0)
1287 		return (err);
1288 	err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1289 	/* zap_add_impl() calls zap_unlockdir() */
1290 	return (err);
1291 }
1292 
1293 int
1294 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1295     int key_numints, int integer_size, uint64_t num_integers,
1296     const void *val, dmu_tx_t *tx)
1297 {
1298 	zap_t *zap;
1299 
1300 	int err =
1301 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1302 	if (err != 0)
1303 		return (err);
1304 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1305 	if (zn == NULL) {
1306 		zap_unlockdir(zap, FTAG);
1307 		return (SET_ERROR(ENOTSUP));
1308 	}
1309 	err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1310 	zap = zn->zn_zap;	/* fzap_add() may change zap */
1311 	zap_name_free(zn);
1312 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
1313 		zap_unlockdir(zap, FTAG);
1314 	return (err);
1315 }
1316 
1317 int
1318 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1319     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1320 {
1321 	zap_t *zap;
1322 	const uint64_t *intval = val;
1323 
1324 	int err =
1325 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1326 	if (err != 0)
1327 		return (err);
1328 	zap_name_t *zn = zap_name_alloc(zap, name, 0);
1329 	if (zn == NULL) {
1330 		zap_unlockdir(zap, FTAG);
1331 		return (SET_ERROR(ENOTSUP));
1332 	}
1333 	if (!zap->zap_ismicro) {
1334 		err = fzap_update(zn, integer_size, num_integers, val,
1335 		    FTAG, tx);
1336 		zap = zn->zn_zap;	/* fzap_update() may change zap */
1337 	} else if (integer_size != 8 || num_integers != 1 ||
1338 	    strlen(name) >= MZAP_NAME_LEN) {
1339 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1340 		    (u_longlong_t)zapobj, integer_size,
1341 		    (u_longlong_t)num_integers, name);
1342 		err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1343 		if (err == 0) {
1344 			err = fzap_update(zn, integer_size, num_integers,
1345 			    val, FTAG, tx);
1346 		}
1347 		zap = zn->zn_zap;	/* fzap_update() may change zap */
1348 	} else {
1349 		mzap_ent_t *mze = mze_find(zn);
1350 		if (mze != NULL) {
1351 			MZE_PHYS(zap, mze)->mze_value = *intval;
1352 		} else {
1353 			mzap_addent(zn, *intval);
1354 		}
1355 	}
1356 	ASSERT(zap == zn->zn_zap);
1357 	zap_name_free(zn);
1358 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1359 		zap_unlockdir(zap, FTAG);
1360 	return (err);
1361 }
1362 
1363 int
1364 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1365     int key_numints,
1366     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1367 {
1368 	zap_t *zap;
1369 
1370 	int err =
1371 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, 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_update(zn, integer_size, num_integers, val, FTAG, tx);
1380 	zap = zn->zn_zap;	/* fzap_update() may change zap */
1381 	zap_name_free(zn);
1382 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
1383 		zap_unlockdir(zap, FTAG);
1384 	return (err);
1385 }
1386 
1387 int
1388 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1389 {
1390 	return (zap_remove_norm(os, zapobj, name, 0, tx));
1391 }
1392 
1393 static int
1394 zap_remove_impl(zap_t *zap, const char *name,
1395     matchtype_t mt, dmu_tx_t *tx)
1396 {
1397 	int err = 0;
1398 
1399 	zap_name_t *zn = zap_name_alloc(zap, name, mt);
1400 	if (zn == NULL)
1401 		return (SET_ERROR(ENOTSUP));
1402 	if (!zap->zap_ismicro) {
1403 		err = fzap_remove(zn, tx);
1404 	} else {
1405 		mzap_ent_t *mze = mze_find(zn);
1406 		if (mze == NULL) {
1407 			err = SET_ERROR(ENOENT);
1408 		} else {
1409 			zap->zap_m.zap_num_entries--;
1410 			memset(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid], 0,
1411 			    sizeof (mzap_ent_phys_t));
1412 			mze_remove(zap, mze);
1413 		}
1414 	}
1415 	zap_name_free(zn);
1416 	return (err);
1417 }
1418 
1419 int
1420 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1421     matchtype_t mt, dmu_tx_t *tx)
1422 {
1423 	zap_t *zap;
1424 	int err;
1425 
1426 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1427 	if (err)
1428 		return (err);
1429 	err = zap_remove_impl(zap, name, mt, tx);
1430 	zap_unlockdir(zap, FTAG);
1431 	return (err);
1432 }
1433 
1434 int
1435 zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
1436 {
1437 	zap_t *zap;
1438 	int err;
1439 
1440 	err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1441 	if (err)
1442 		return (err);
1443 	err = zap_remove_impl(zap, name, 0, tx);
1444 	zap_unlockdir(zap, FTAG);
1445 	return (err);
1446 }
1447 
1448 int
1449 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1450     int key_numints, dmu_tx_t *tx)
1451 {
1452 	zap_t *zap;
1453 
1454 	int err =
1455 	    zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1456 	if (err != 0)
1457 		return (err);
1458 	zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1459 	if (zn == NULL) {
1460 		zap_unlockdir(zap, FTAG);
1461 		return (SET_ERROR(ENOTSUP));
1462 	}
1463 	err = fzap_remove(zn, tx);
1464 	zap_name_free(zn);
1465 	zap_unlockdir(zap, FTAG);
1466 	return (err);
1467 }
1468 
1469 /*
1470  * Routines for iterating over the attributes.
1471  */
1472 
1473 static void
1474 zap_cursor_init_impl(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1475     uint64_t serialized, boolean_t prefetch)
1476 {
1477 	zc->zc_objset = os;
1478 	zc->zc_zap = NULL;
1479 	zc->zc_leaf = NULL;
1480 	zc->zc_zapobj = zapobj;
1481 	zc->zc_serialized = serialized;
1482 	zc->zc_hash = 0;
1483 	zc->zc_cd = 0;
1484 	zc->zc_prefetch = prefetch;
1485 }
1486 void
1487 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1488     uint64_t serialized)
1489 {
1490 	zap_cursor_init_impl(zc, os, zapobj, serialized, B_TRUE);
1491 }
1492 
1493 /*
1494  * Initialize a cursor at the beginning of the ZAP object.  The entire
1495  * ZAP object will be prefetched.
1496  */
1497 void
1498 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1499 {
1500 	zap_cursor_init_impl(zc, os, zapobj, 0, B_TRUE);
1501 }
1502 
1503 /*
1504  * Initialize a cursor at the beginning, but request that we not prefetch
1505  * the entire ZAP object.
1506  */
1507 void
1508 zap_cursor_init_noprefetch(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1509 {
1510 	zap_cursor_init_impl(zc, os, zapobj, 0, B_FALSE);
1511 }
1512 
1513 void
1514 zap_cursor_fini(zap_cursor_t *zc)
1515 {
1516 	if (zc->zc_zap) {
1517 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1518 		zap_unlockdir(zc->zc_zap, NULL);
1519 		zc->zc_zap = NULL;
1520 	}
1521 	if (zc->zc_leaf) {
1522 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1523 		zap_put_leaf(zc->zc_leaf);
1524 		zc->zc_leaf = NULL;
1525 	}
1526 	zc->zc_objset = NULL;
1527 }
1528 
1529 uint64_t
1530 zap_cursor_serialize(zap_cursor_t *zc)
1531 {
1532 	if (zc->zc_hash == -1ULL)
1533 		return (-1ULL);
1534 	if (zc->zc_zap == NULL)
1535 		return (zc->zc_serialized);
1536 	ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1537 	ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1538 
1539 	/*
1540 	 * We want to keep the high 32 bits of the cursor zero if we can, so
1541 	 * that 32-bit programs can access this.  So usually use a small
1542 	 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1543 	 * of the cursor.
1544 	 *
1545 	 * [ collision differentiator | zap_hashbits()-bit hash value ]
1546 	 */
1547 	return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1548 	    ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1549 }
1550 
1551 int
1552 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1553 {
1554 	int err;
1555 
1556 	if (zc->zc_hash == -1ULL)
1557 		return (SET_ERROR(ENOENT));
1558 
1559 	if (zc->zc_zap == NULL) {
1560 		int hb;
1561 		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1562 		    RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1563 		if (err != 0)
1564 			return (err);
1565 
1566 		/*
1567 		 * To support zap_cursor_init_serialized, advance, retrieve,
1568 		 * we must add to the existing zc_cd, which may already
1569 		 * be 1 due to the zap_cursor_advance.
1570 		 */
1571 		ASSERT(zc->zc_hash == 0);
1572 		hb = zap_hashbits(zc->zc_zap);
1573 		zc->zc_hash = zc->zc_serialized << (64 - hb);
1574 		zc->zc_cd += zc->zc_serialized >> hb;
1575 		if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1576 			zc->zc_cd = 0;
1577 	} else {
1578 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1579 	}
1580 	if (!zc->zc_zap->zap_ismicro) {
1581 		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1582 	} else {
1583 		avl_index_t idx;
1584 		mzap_ent_t mze_tofind;
1585 
1586 		mze_tofind.mze_hash = zc->zc_hash;
1587 		mze_tofind.mze_cd = zc->zc_cd;
1588 
1589 		mzap_ent_t *mze =
1590 		    avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1591 		if (mze == NULL) {
1592 			mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1593 			    idx, AVL_AFTER);
1594 		}
1595 		if (mze) {
1596 			mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1597 			ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1598 			za->za_normalization_conflict =
1599 			    mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1600 			za->za_integer_length = 8;
1601 			za->za_num_integers = 1;
1602 			za->za_first_integer = mzep->mze_value;
1603 			(void) strlcpy(za->za_name, mzep->mze_name,
1604 			    sizeof (za->za_name));
1605 			zc->zc_hash = mze->mze_hash;
1606 			zc->zc_cd = mze->mze_cd;
1607 			err = 0;
1608 		} else {
1609 			zc->zc_hash = -1ULL;
1610 			err = SET_ERROR(ENOENT);
1611 		}
1612 	}
1613 	rw_exit(&zc->zc_zap->zap_rwlock);
1614 	return (err);
1615 }
1616 
1617 void
1618 zap_cursor_advance(zap_cursor_t *zc)
1619 {
1620 	if (zc->zc_hash == -1ULL)
1621 		return;
1622 	zc->zc_cd++;
1623 }
1624 
1625 int
1626 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1627 {
1628 	zap_t *zap;
1629 
1630 	int err =
1631 	    zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1632 	if (err != 0)
1633 		return (err);
1634 
1635 	memset(zs, 0, sizeof (zap_stats_t));
1636 
1637 	if (zap->zap_ismicro) {
1638 		zs->zs_blocksize = zap->zap_dbuf->db_size;
1639 		zs->zs_num_entries = zap->zap_m.zap_num_entries;
1640 		zs->zs_num_blocks = 1;
1641 	} else {
1642 		fzap_get_stats(zap, zs);
1643 	}
1644 	zap_unlockdir(zap, FTAG);
1645 	return (0);
1646 }
1647 
1648 #if defined(_KERNEL)
1649 EXPORT_SYMBOL(zap_create);
1650 EXPORT_SYMBOL(zap_create_dnsize);
1651 EXPORT_SYMBOL(zap_create_norm);
1652 EXPORT_SYMBOL(zap_create_norm_dnsize);
1653 EXPORT_SYMBOL(zap_create_flags);
1654 EXPORT_SYMBOL(zap_create_flags_dnsize);
1655 EXPORT_SYMBOL(zap_create_claim);
1656 EXPORT_SYMBOL(zap_create_claim_norm);
1657 EXPORT_SYMBOL(zap_create_claim_norm_dnsize);
1658 EXPORT_SYMBOL(zap_create_hold);
1659 EXPORT_SYMBOL(zap_destroy);
1660 EXPORT_SYMBOL(zap_lookup);
1661 EXPORT_SYMBOL(zap_lookup_by_dnode);
1662 EXPORT_SYMBOL(zap_lookup_norm);
1663 EXPORT_SYMBOL(zap_lookup_uint64);
1664 EXPORT_SYMBOL(zap_contains);
1665 EXPORT_SYMBOL(zap_prefetch);
1666 EXPORT_SYMBOL(zap_prefetch_uint64);
1667 EXPORT_SYMBOL(zap_add);
1668 EXPORT_SYMBOL(zap_add_by_dnode);
1669 EXPORT_SYMBOL(zap_add_uint64);
1670 EXPORT_SYMBOL(zap_update);
1671 EXPORT_SYMBOL(zap_update_uint64);
1672 EXPORT_SYMBOL(zap_length);
1673 EXPORT_SYMBOL(zap_length_uint64);
1674 EXPORT_SYMBOL(zap_remove);
1675 EXPORT_SYMBOL(zap_remove_by_dnode);
1676 EXPORT_SYMBOL(zap_remove_norm);
1677 EXPORT_SYMBOL(zap_remove_uint64);
1678 EXPORT_SYMBOL(zap_count);
1679 EXPORT_SYMBOL(zap_value_search);
1680 EXPORT_SYMBOL(zap_join);
1681 EXPORT_SYMBOL(zap_join_increment);
1682 EXPORT_SYMBOL(zap_add_int);
1683 EXPORT_SYMBOL(zap_remove_int);
1684 EXPORT_SYMBOL(zap_lookup_int);
1685 EXPORT_SYMBOL(zap_increment_int);
1686 EXPORT_SYMBOL(zap_add_int_key);
1687 EXPORT_SYMBOL(zap_lookup_int_key);
1688 EXPORT_SYMBOL(zap_increment);
1689 EXPORT_SYMBOL(zap_cursor_init);
1690 EXPORT_SYMBOL(zap_cursor_fini);
1691 EXPORT_SYMBOL(zap_cursor_retrieve);
1692 EXPORT_SYMBOL(zap_cursor_advance);
1693 EXPORT_SYMBOL(zap_cursor_serialize);
1694 EXPORT_SYMBOL(zap_cursor_init_serialized);
1695 EXPORT_SYMBOL(zap_get_stats);
1696 #endif
1697