xref: /titanic_52/usr/src/uts/common/fs/zfs/zap_micro.c (revision 5763ba1e357fad1d57b5875c499307b7ea6e2cd4)
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  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/spa.h>
29 #include <sys/dmu.h>
30 #include <sys/zfs_context.h>
31 #include <sys/zap.h>
32 #include <sys/refcount.h>
33 #include <sys/zap_impl.h>
34 #include <sys/zap_leaf.h>
35 #include <sys/avl.h>
36 #include <sys/zfs_i18n.h>
37 
38 static void mzap_upgrade(zap_t *zap, dmu_tx_t *tx);
39 
40 
41 static uint64_t
42 zap_hash(zap_t *zap, const char *normname)
43 {
44 	const uint8_t *cp;
45 	uint8_t c;
46 	uint64_t crc = zap->zap_salt;
47 
48 	/* NB: name must already be normalized, if necessary */
49 
50 	ASSERT(crc != 0);
51 	ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
52 	for (cp = (const uint8_t *)normname; (c = *cp) != '\0'; cp++) {
53 		crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ c) & 0xFF];
54 	}
55 
56 	/*
57 	 * Only use 28 bits, since we need 4 bits in the cookie for the
58 	 * collision differentiator.  We MUST use the high bits, since
59 	 * those are the ones that we first pay attention to when
60 	 * chosing the bucket.
61 	 */
62 	crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1);
63 
64 	return (crc);
65 }
66 
67 static int
68 zap_normalize(zap_t *zap, const char *name, char *namenorm)
69 {
70 	size_t inlen, outlen;
71 	int err;
72 
73 	inlen = strlen(name) + 1;
74 	outlen = ZAP_MAXNAMELEN;
75 
76 	err = 0;
77 	(void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
78 	    zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL, U8_UNICODE_LATEST,
79 	    &err);
80 
81 	return (err);
82 }
83 
84 boolean_t
85 zap_match(zap_name_t *zn, const char *matchname)
86 {
87 	if (zn->zn_matchtype == MT_FIRST) {
88 		char norm[ZAP_MAXNAMELEN];
89 
90 		if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
91 			return (B_FALSE);
92 
93 		return (strcmp(zn->zn_name_norm, norm) == 0);
94 	} else {
95 		/* MT_BEST or MT_EXACT */
96 		return (strcmp(zn->zn_name_orij, matchname) == 0);
97 	}
98 }
99 
100 void
101 zap_name_free(zap_name_t *zn)
102 {
103 	kmem_free(zn, sizeof (zap_name_t));
104 }
105 
106 /* XXX combine this with zap_lockdir()? */
107 zap_name_t *
108 zap_name_alloc(zap_t *zap, const char *name, matchtype_t mt)
109 {
110 	zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
111 
112 	zn->zn_zap = zap;
113 	zn->zn_name_orij = name;
114 	zn->zn_matchtype = mt;
115 	if (zap->zap_normflags) {
116 		if (zap_normalize(zap, name, zn->zn_normbuf) != 0) {
117 			zap_name_free(zn);
118 			return (NULL);
119 		}
120 		zn->zn_name_norm = zn->zn_normbuf;
121 	} else {
122 		if (mt != MT_EXACT) {
123 			zap_name_free(zn);
124 			return (NULL);
125 		}
126 		zn->zn_name_norm = zn->zn_name_orij;
127 	}
128 
129 	zn->zn_hash = zap_hash(zap, zn->zn_name_norm);
130 	return (zn);
131 }
132 
133 static void
134 mzap_byteswap(mzap_phys_t *buf, size_t size)
135 {
136 	int i, max;
137 	buf->mz_block_type = BSWAP_64(buf->mz_block_type);
138 	buf->mz_salt = BSWAP_64(buf->mz_salt);
139 	max = (size / MZAP_ENT_LEN) - 1;
140 	for (i = 0; i < max; i++) {
141 		buf->mz_chunk[i].mze_value =
142 		    BSWAP_64(buf->mz_chunk[i].mze_value);
143 		buf->mz_chunk[i].mze_cd =
144 		    BSWAP_32(buf->mz_chunk[i].mze_cd);
145 	}
146 }
147 
148 void
149 zap_byteswap(void *buf, size_t size)
150 {
151 	uint64_t block_type;
152 
153 	block_type = *(uint64_t *)buf;
154 
155 	if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
156 		/* ASSERT(magic == ZAP_LEAF_MAGIC); */
157 		mzap_byteswap(buf, size);
158 	} else {
159 		fzap_byteswap(buf, size);
160 	}
161 }
162 
163 static int
164 mze_compare(const void *arg1, const void *arg2)
165 {
166 	const mzap_ent_t *mze1 = arg1;
167 	const mzap_ent_t *mze2 = arg2;
168 
169 	if (mze1->mze_hash > mze2->mze_hash)
170 		return (+1);
171 	if (mze1->mze_hash < mze2->mze_hash)
172 		return (-1);
173 	if (mze1->mze_phys.mze_cd > mze2->mze_phys.mze_cd)
174 		return (+1);
175 	if (mze1->mze_phys.mze_cd < mze2->mze_phys.mze_cd)
176 		return (-1);
177 	return (0);
178 }
179 
180 static void
181 mze_insert(zap_t *zap, int chunkid, uint64_t hash, mzap_ent_phys_t *mzep)
182 {
183 	mzap_ent_t *mze;
184 
185 	ASSERT(zap->zap_ismicro);
186 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
187 	ASSERT(mzep->mze_cd < ZAP_MAXCD);
188 
189 	mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
190 	mze->mze_chunkid = chunkid;
191 	mze->mze_hash = hash;
192 	mze->mze_phys = *mzep;
193 	avl_add(&zap->zap_m.zap_avl, mze);
194 }
195 
196 static mzap_ent_t *
197 mze_find(zap_name_t *zn)
198 {
199 	mzap_ent_t mze_tofind;
200 	mzap_ent_t *mze;
201 	avl_index_t idx;
202 	avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
203 
204 	ASSERT(zn->zn_zap->zap_ismicro);
205 	ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
206 
207 	if (strlen(zn->zn_name_norm) >= sizeof (mze_tofind.mze_phys.mze_name))
208 		return (NULL);
209 
210 	mze_tofind.mze_hash = zn->zn_hash;
211 	mze_tofind.mze_phys.mze_cd = 0;
212 
213 again:
214 	mze = avl_find(avl, &mze_tofind, &idx);
215 	if (mze == NULL)
216 		mze = avl_nearest(avl, idx, AVL_AFTER);
217 	for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
218 		if (zap_match(zn, mze->mze_phys.mze_name))
219 			return (mze);
220 	}
221 	if (zn->zn_matchtype == MT_BEST) {
222 		zn->zn_matchtype = MT_FIRST;
223 		goto again;
224 	}
225 	return (NULL);
226 }
227 
228 static uint32_t
229 mze_find_unused_cd(zap_t *zap, uint64_t hash)
230 {
231 	mzap_ent_t mze_tofind;
232 	mzap_ent_t *mze;
233 	avl_index_t idx;
234 	avl_tree_t *avl = &zap->zap_m.zap_avl;
235 	uint32_t cd;
236 
237 	ASSERT(zap->zap_ismicro);
238 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
239 
240 	mze_tofind.mze_hash = hash;
241 	mze_tofind.mze_phys.mze_cd = 0;
242 
243 	cd = 0;
244 	for (mze = avl_find(avl, &mze_tofind, &idx);
245 	    mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
246 		if (mze->mze_phys.mze_cd != cd)
247 			break;
248 		cd++;
249 	}
250 
251 	return (cd);
252 }
253 
254 static void
255 mze_remove(zap_t *zap, mzap_ent_t *mze)
256 {
257 	ASSERT(zap->zap_ismicro);
258 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
259 
260 	avl_remove(&zap->zap_m.zap_avl, mze);
261 	kmem_free(mze, sizeof (mzap_ent_t));
262 }
263 
264 static void
265 mze_destroy(zap_t *zap)
266 {
267 	mzap_ent_t *mze;
268 	void *avlcookie = NULL;
269 
270 	while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
271 		kmem_free(mze, sizeof (mzap_ent_t));
272 	avl_destroy(&zap->zap_m.zap_avl);
273 }
274 
275 static zap_t *
276 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
277 {
278 	zap_t *winner;
279 	zap_t *zap;
280 	int i;
281 
282 	ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
283 
284 	zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
285 	rw_init(&zap->zap_rwlock, 0, 0, 0);
286 	rw_enter(&zap->zap_rwlock, RW_WRITER);
287 	zap->zap_objset = os;
288 	zap->zap_object = obj;
289 	zap->zap_dbuf = db;
290 
291 	if (*(uint64_t *)db->db_data != ZBT_MICRO) {
292 		mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
293 		zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
294 	} else {
295 		zap->zap_ismicro = TRUE;
296 	}
297 
298 	/*
299 	 * Make sure that zap_ismicro is set before we let others see
300 	 * it, because zap_lockdir() checks zap_ismicro without the lock
301 	 * held.
302 	 */
303 	winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict);
304 
305 	if (winner != NULL) {
306 		rw_exit(&zap->zap_rwlock);
307 		rw_destroy(&zap->zap_rwlock);
308 		if (!zap->zap_ismicro)
309 			mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
310 		kmem_free(zap, sizeof (zap_t));
311 		return (winner);
312 	}
313 
314 	if (zap->zap_ismicro) {
315 		zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
316 		zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags;
317 		zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
318 		avl_create(&zap->zap_m.zap_avl, mze_compare,
319 		    sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
320 
321 		for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
322 			mzap_ent_phys_t *mze =
323 			    &zap->zap_m.zap_phys->mz_chunk[i];
324 			if (mze->mze_name[0]) {
325 				zap_name_t *zn;
326 
327 				zap->zap_m.zap_num_entries++;
328 				zn = zap_name_alloc(zap, mze->mze_name,
329 				    MT_EXACT);
330 				mze_insert(zap, i, zn->zn_hash, mze);
331 				zap_name_free(zn);
332 			}
333 		}
334 	} else {
335 		zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
336 		zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags;
337 
338 		ASSERT3U(sizeof (struct zap_leaf_header), ==,
339 		    2*ZAP_LEAF_CHUNKSIZE);
340 
341 		/*
342 		 * The embedded pointer table should not overlap the
343 		 * other members.
344 		 */
345 		ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
346 		    &zap->zap_f.zap_phys->zap_salt);
347 
348 		/*
349 		 * The embedded pointer table should end at the end of
350 		 * the block
351 		 */
352 		ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
353 		    1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
354 		    (uintptr_t)zap->zap_f.zap_phys, ==,
355 		    zap->zap_dbuf->db_size);
356 	}
357 	rw_exit(&zap->zap_rwlock);
358 	return (zap);
359 }
360 
361 int
362 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
363     krw_t lti, int fatreader, zap_t **zapp)
364 {
365 	zap_t *zap;
366 	dmu_buf_t *db;
367 	krw_t lt;
368 	int err;
369 
370 	*zapp = NULL;
371 
372 	err = dmu_buf_hold(os, obj, 0, NULL, &db);
373 	if (err)
374 		return (err);
375 
376 #ifdef ZFS_DEBUG
377 	{
378 		dmu_object_info_t doi;
379 		dmu_object_info_from_db(db, &doi);
380 		ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
381 	}
382 #endif
383 
384 	zap = dmu_buf_get_user(db);
385 	if (zap == NULL)
386 		zap = mzap_open(os, obj, db);
387 
388 	/*
389 	 * We're checking zap_ismicro without the lock held, in order to
390 	 * tell what type of lock we want.  Once we have some sort of
391 	 * lock, see if it really is the right type.  In practice this
392 	 * can only be different if it was upgraded from micro to fat,
393 	 * and micro wanted WRITER but fat only needs READER.
394 	 */
395 	lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
396 	rw_enter(&zap->zap_rwlock, lt);
397 	if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
398 		/* it was upgraded, now we only need reader */
399 		ASSERT(lt == RW_WRITER);
400 		ASSERT(RW_READER ==
401 		    (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
402 		rw_downgrade(&zap->zap_rwlock);
403 		lt = RW_READER;
404 	}
405 
406 	zap->zap_objset = os;
407 
408 	if (lt == RW_WRITER)
409 		dmu_buf_will_dirty(db, tx);
410 
411 	ASSERT3P(zap->zap_dbuf, ==, db);
412 
413 	ASSERT(!zap->zap_ismicro ||
414 	    zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
415 	if (zap->zap_ismicro && tx &&
416 	    zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
417 		uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
418 		if (newsz > MZAP_MAX_BLKSZ) {
419 			dprintf("upgrading obj %llu: num_entries=%u\n",
420 			    obj, zap->zap_m.zap_num_entries);
421 			mzap_upgrade(zap, tx);
422 			*zapp = zap;
423 			return (0);
424 		}
425 		err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
426 		ASSERT3U(err, ==, 0);
427 		zap->zap_m.zap_num_chunks =
428 		    db->db_size / MZAP_ENT_LEN - 1;
429 	}
430 
431 	*zapp = zap;
432 	return (0);
433 }
434 
435 void
436 zap_unlockdir(zap_t *zap)
437 {
438 	rw_exit(&zap->zap_rwlock);
439 	dmu_buf_rele(zap->zap_dbuf, NULL);
440 }
441 
442 static void
443 mzap_upgrade(zap_t *zap, dmu_tx_t *tx)
444 {
445 	mzap_phys_t *mzp;
446 	int i, sz, nchunks, err;
447 
448 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
449 
450 	sz = zap->zap_dbuf->db_size;
451 	mzp = kmem_alloc(sz, KM_SLEEP);
452 	bcopy(zap->zap_dbuf->db_data, mzp, sz);
453 	nchunks = zap->zap_m.zap_num_chunks;
454 
455 	err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
456 	    1ULL << fzap_default_block_shift, 0, tx);
457 	ASSERT(err == 0);
458 
459 	dprintf("upgrading obj=%llu with %u chunks\n",
460 	    zap->zap_object, nchunks);
461 	/* XXX destroy the avl later, so we can use the stored hash value */
462 	mze_destroy(zap);
463 
464 	fzap_upgrade(zap, tx);
465 
466 	for (i = 0; i < nchunks; i++) {
467 		int err;
468 		mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
469 		zap_name_t *zn;
470 		if (mze->mze_name[0] == 0)
471 			continue;
472 		dprintf("adding %s=%llu\n",
473 		    mze->mze_name, mze->mze_value);
474 		zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
475 		err = fzap_add_cd(zn, 8, 1, &mze->mze_value,
476 		    mze->mze_cd, tx);
477 		zap_name_free(zn);
478 		ASSERT3U(err, ==, 0);
479 	}
480 	kmem_free(mzp, sz);
481 }
482 
483 static void
484 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, dmu_tx_t *tx)
485 {
486 	dmu_buf_t *db;
487 	mzap_phys_t *zp;
488 
489 	ASSERT(normflags == 0 ||
490 	    spa_version(dmu_objset_spa(os)) >= SPA_VERSION_NORMALIZATION);
491 	VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db));
492 
493 #ifdef ZFS_DEBUG
494 	{
495 		dmu_object_info_t doi;
496 		dmu_object_info_from_db(db, &doi);
497 		ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
498 	}
499 #endif
500 
501 	dmu_buf_will_dirty(db, tx);
502 	zp = db->db_data;
503 	zp->mz_block_type = ZBT_MICRO;
504 	zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
505 	zp->mz_normflags = normflags;
506 	dmu_buf_rele(db, FTAG);
507 }
508 
509 int
510 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
511     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
512 {
513 	return (zap_create_claim_norm(os, obj,
514 	    0, ot, bonustype, bonuslen, tx));
515 }
516 
517 int
518 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
519     dmu_object_type_t ot,
520     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
521 {
522 	int err;
523 
524 	err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
525 	if (err != 0)
526 		return (err);
527 	mzap_create_impl(os, obj, normflags, tx);
528 	return (0);
529 }
530 
531 uint64_t
532 zap_create(objset_t *os, dmu_object_type_t ot,
533     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
534 {
535 	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
536 }
537 
538 uint64_t
539 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
540     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
541 {
542 	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
543 
544 	mzap_create_impl(os, obj, normflags, tx);
545 	return (obj);
546 }
547 
548 int
549 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
550 {
551 	/*
552 	 * dmu_object_free will free the object number and free the
553 	 * data.  Freeing the data will cause our pageout function to be
554 	 * called, which will destroy our data (zap_leaf_t's and zap_t).
555 	 */
556 
557 	return (dmu_object_free(os, zapobj, tx));
558 }
559 
560 _NOTE(ARGSUSED(0))
561 void
562 zap_evict(dmu_buf_t *db, void *vzap)
563 {
564 	zap_t *zap = vzap;
565 
566 	rw_destroy(&zap->zap_rwlock);
567 
568 	if (zap->zap_ismicro)
569 		mze_destroy(zap);
570 	else
571 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
572 
573 	kmem_free(zap, sizeof (zap_t));
574 }
575 
576 int
577 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
578 {
579 	zap_t *zap;
580 	int err;
581 
582 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, &zap);
583 	if (err)
584 		return (err);
585 	if (!zap->zap_ismicro) {
586 		err = fzap_count(zap, count);
587 	} else {
588 		*count = zap->zap_m.zap_num_entries;
589 	}
590 	zap_unlockdir(zap);
591 	return (err);
592 }
593 
594 /*
595  * zn may be NULL; if not specified, it will be computed if needed.
596  * See also the comment above zap_entry_normalization_conflict().
597  */
598 static boolean_t
599 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
600 {
601 	mzap_ent_t *other;
602 	int direction = AVL_BEFORE;
603 	boolean_t allocdzn = B_FALSE;
604 
605 	if (zap->zap_normflags == 0)
606 		return (B_FALSE);
607 
608 again:
609 	for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
610 	    other && other->mze_hash == mze->mze_hash;
611 	    other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
612 
613 		if (zn == NULL) {
614 			zn = zap_name_alloc(zap, mze->mze_phys.mze_name,
615 			    MT_FIRST);
616 			allocdzn = B_TRUE;
617 		}
618 		if (zap_match(zn, other->mze_phys.mze_name)) {
619 			if (allocdzn)
620 				zap_name_free(zn);
621 			return (B_TRUE);
622 		}
623 	}
624 
625 	if (direction == AVL_BEFORE) {
626 		direction = AVL_AFTER;
627 		goto again;
628 	}
629 
630 	if (allocdzn)
631 		zap_name_free(zn);
632 	return (B_FALSE);
633 }
634 
635 /*
636  * Routines for manipulating attributes.
637  */
638 
639 int
640 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
641     uint64_t integer_size, uint64_t num_integers, void *buf)
642 {
643 	return (zap_lookup_norm(os, zapobj, name, integer_size,
644 	    num_integers, buf, MT_EXACT, NULL, 0, NULL));
645 }
646 
647 int
648 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
649     uint64_t integer_size, uint64_t num_integers, void *buf,
650     matchtype_t mt, char *realname, int rn_len,
651     boolean_t *ncp)
652 {
653 	zap_t *zap;
654 	int err;
655 	mzap_ent_t *mze;
656 	zap_name_t *zn;
657 
658 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, &zap);
659 	if (err)
660 		return (err);
661 	zn = zap_name_alloc(zap, name, mt);
662 	if (zn == NULL) {
663 		zap_unlockdir(zap);
664 		return (ENOTSUP);
665 	}
666 
667 	if (!zap->zap_ismicro) {
668 		err = fzap_lookup(zn, integer_size, num_integers, buf,
669 		    realname, rn_len, ncp);
670 	} else {
671 		mze = mze_find(zn);
672 		if (mze == NULL) {
673 			err = ENOENT;
674 		} else {
675 			if (num_integers < 1) {
676 				err = EOVERFLOW;
677 			} else if (integer_size != 8) {
678 				err = EINVAL;
679 			} else {
680 				*(uint64_t *)buf = mze->mze_phys.mze_value;
681 				(void) strlcpy(realname,
682 				    mze->mze_phys.mze_name, rn_len);
683 				if (ncp) {
684 					*ncp = mzap_normalization_conflict(zap,
685 					    zn, mze);
686 				}
687 			}
688 		}
689 	}
690 	zap_name_free(zn);
691 	zap_unlockdir(zap);
692 	return (err);
693 }
694 
695 int
696 zap_length(objset_t *os, uint64_t zapobj, const char *name,
697     uint64_t *integer_size, uint64_t *num_integers)
698 {
699 	zap_t *zap;
700 	int err;
701 	mzap_ent_t *mze;
702 	zap_name_t *zn;
703 
704 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, &zap);
705 	if (err)
706 		return (err);
707 	zn = zap_name_alloc(zap, name, MT_EXACT);
708 	if (zn == NULL) {
709 		zap_unlockdir(zap);
710 		return (ENOTSUP);
711 	}
712 	if (!zap->zap_ismicro) {
713 		err = fzap_length(zn, integer_size, num_integers);
714 	} else {
715 		mze = mze_find(zn);
716 		if (mze == NULL) {
717 			err = ENOENT;
718 		} else {
719 			if (integer_size)
720 				*integer_size = 8;
721 			if (num_integers)
722 				*num_integers = 1;
723 		}
724 	}
725 	zap_name_free(zn);
726 	zap_unlockdir(zap);
727 	return (err);
728 }
729 
730 static void
731 mzap_addent(zap_name_t *zn, uint64_t value)
732 {
733 	int i;
734 	zap_t *zap = zn->zn_zap;
735 	int start = zap->zap_m.zap_alloc_next;
736 	uint32_t cd;
737 
738 	dprintf("obj=%llu %s=%llu\n", zap->zap_object,
739 	    zn->zn_name_orij, value);
740 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
741 
742 #ifdef ZFS_DEBUG
743 	for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
744 		mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
745 		ASSERT(strcmp(zn->zn_name_orij, mze->mze_name) != 0);
746 	}
747 #endif
748 
749 	cd = mze_find_unused_cd(zap, zn->zn_hash);
750 	/* given the limited size of the microzap, this can't happen */
751 	ASSERT(cd != ZAP_MAXCD);
752 
753 again:
754 	for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
755 		mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
756 		if (mze->mze_name[0] == 0) {
757 			mze->mze_value = value;
758 			mze->mze_cd = cd;
759 			(void) strcpy(mze->mze_name, zn->zn_name_orij);
760 			zap->zap_m.zap_num_entries++;
761 			zap->zap_m.zap_alloc_next = i+1;
762 			if (zap->zap_m.zap_alloc_next ==
763 			    zap->zap_m.zap_num_chunks)
764 				zap->zap_m.zap_alloc_next = 0;
765 			mze_insert(zap, i, zn->zn_hash, mze);
766 			return;
767 		}
768 	}
769 	if (start != 0) {
770 		start = 0;
771 		goto again;
772 	}
773 	ASSERT(!"out of entries!");
774 }
775 
776 int
777 zap_add(objset_t *os, uint64_t zapobj, const char *name,
778     int integer_size, uint64_t num_integers,
779     const void *val, dmu_tx_t *tx)
780 {
781 	zap_t *zap;
782 	int err;
783 	mzap_ent_t *mze;
784 	const uint64_t *intval = val;
785 	zap_name_t *zn;
786 
787 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, &zap);
788 	if (err)
789 		return (err);
790 	zn = zap_name_alloc(zap, name, MT_EXACT);
791 	if (zn == NULL) {
792 		zap_unlockdir(zap);
793 		return (ENOTSUP);
794 	}
795 	if (!zap->zap_ismicro) {
796 		err = fzap_add(zn, integer_size, num_integers, val, tx);
797 	} else if (integer_size != 8 || num_integers != 1 ||
798 	    strlen(name) >= MZAP_NAME_LEN) {
799 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
800 		    zapobj, integer_size, num_integers, name);
801 		mzap_upgrade(zap, tx);
802 		err = fzap_add(zn, integer_size, num_integers, val, tx);
803 	} else {
804 		mze = mze_find(zn);
805 		if (mze != NULL) {
806 			err = EEXIST;
807 		} else {
808 			mzap_addent(zn, *intval);
809 		}
810 	}
811 	zap_name_free(zn);
812 	zap_unlockdir(zap);
813 	return (err);
814 }
815 
816 int
817 zap_update(objset_t *os, uint64_t zapobj, const char *name,
818     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
819 {
820 	zap_t *zap;
821 	mzap_ent_t *mze;
822 	const uint64_t *intval = val;
823 	zap_name_t *zn;
824 	int err;
825 
826 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, &zap);
827 	if (err)
828 		return (err);
829 	zn = zap_name_alloc(zap, name, MT_EXACT);
830 	if (zn == NULL) {
831 		zap_unlockdir(zap);
832 		return (ENOTSUP);
833 	}
834 	if (!zap->zap_ismicro) {
835 		err = fzap_update(zn, integer_size, num_integers, val, tx);
836 	} else if (integer_size != 8 || num_integers != 1 ||
837 	    strlen(name) >= MZAP_NAME_LEN) {
838 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
839 		    zapobj, integer_size, num_integers, name);
840 		mzap_upgrade(zap, tx);
841 		err = fzap_update(zn, integer_size, num_integers, val, tx);
842 	} else {
843 		mze = mze_find(zn);
844 		if (mze != NULL) {
845 			mze->mze_phys.mze_value = *intval;
846 			zap->zap_m.zap_phys->mz_chunk
847 			    [mze->mze_chunkid].mze_value = *intval;
848 		} else {
849 			mzap_addent(zn, *intval);
850 		}
851 	}
852 	zap_name_free(zn);
853 	zap_unlockdir(zap);
854 	return (err);
855 }
856 
857 int
858 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
859 {
860 	return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
861 }
862 
863 int
864 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
865     matchtype_t mt, dmu_tx_t *tx)
866 {
867 	zap_t *zap;
868 	int err;
869 	mzap_ent_t *mze;
870 	zap_name_t *zn;
871 
872 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, &zap);
873 	if (err)
874 		return (err);
875 	zn = zap_name_alloc(zap, name, mt);
876 	if (zn == NULL) {
877 		zap_unlockdir(zap);
878 		return (ENOTSUP);
879 	}
880 	if (!zap->zap_ismicro) {
881 		err = fzap_remove(zn, tx);
882 	} else {
883 		mze = mze_find(zn);
884 		if (mze == NULL) {
885 			err = ENOENT;
886 		} else {
887 			zap->zap_m.zap_num_entries--;
888 			bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
889 			    sizeof (mzap_ent_phys_t));
890 			mze_remove(zap, mze);
891 		}
892 	}
893 	zap_name_free(zn);
894 	zap_unlockdir(zap);
895 	return (err);
896 }
897 
898 
899 /*
900  * Routines for iterating over the attributes.
901  */
902 
903 /*
904  * We want to keep the high 32 bits of the cursor zero if we can, so
905  * that 32-bit programs can access this.  So use a small hash value so
906  * we can fit 4 bits of cd into the 32-bit cursor.
907  *
908  * [ 4 zero bits | 32-bit collision differentiator | 28-bit hash value ]
909  */
910 void
911 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
912     uint64_t serialized)
913 {
914 	zc->zc_objset = os;
915 	zc->zc_zap = NULL;
916 	zc->zc_leaf = NULL;
917 	zc->zc_zapobj = zapobj;
918 	if (serialized == -1ULL) {
919 		zc->zc_hash = -1ULL;
920 		zc->zc_cd = 0;
921 	} else {
922 		zc->zc_hash = serialized << (64-ZAP_HASHBITS);
923 		zc->zc_cd = serialized >> ZAP_HASHBITS;
924 		if (zc->zc_cd >= ZAP_MAXCD) /* corrupt serialized */
925 			zc->zc_cd = 0;
926 	}
927 }
928 
929 void
930 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
931 {
932 	zap_cursor_init_serialized(zc, os, zapobj, 0);
933 }
934 
935 void
936 zap_cursor_fini(zap_cursor_t *zc)
937 {
938 	if (zc->zc_zap) {
939 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
940 		zap_unlockdir(zc->zc_zap);
941 		zc->zc_zap = NULL;
942 	}
943 	if (zc->zc_leaf) {
944 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
945 		zap_put_leaf(zc->zc_leaf);
946 		zc->zc_leaf = NULL;
947 	}
948 	zc->zc_objset = NULL;
949 }
950 
951 uint64_t
952 zap_cursor_serialize(zap_cursor_t *zc)
953 {
954 	if (zc->zc_hash == -1ULL)
955 		return (-1ULL);
956 	ASSERT((zc->zc_hash & (ZAP_MAXCD-1)) == 0);
957 	ASSERT(zc->zc_cd < ZAP_MAXCD);
958 	return ((zc->zc_hash >> (64-ZAP_HASHBITS)) |
959 	    ((uint64_t)zc->zc_cd << ZAP_HASHBITS));
960 }
961 
962 int
963 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
964 {
965 	int err;
966 	avl_index_t idx;
967 	mzap_ent_t mze_tofind;
968 	mzap_ent_t *mze;
969 
970 	if (zc->zc_hash == -1ULL)
971 		return (ENOENT);
972 
973 	if (zc->zc_zap == NULL) {
974 		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
975 		    RW_READER, TRUE, &zc->zc_zap);
976 		if (err)
977 			return (err);
978 	} else {
979 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
980 	}
981 	if (!zc->zc_zap->zap_ismicro) {
982 		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
983 	} else {
984 		err = ENOENT;
985 
986 		mze_tofind.mze_hash = zc->zc_hash;
987 		mze_tofind.mze_phys.mze_cd = zc->zc_cd;
988 
989 		mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
990 		if (mze == NULL) {
991 			mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
992 			    idx, AVL_AFTER);
993 		}
994 		if (mze) {
995 			ASSERT(0 == bcmp(&mze->mze_phys,
996 			    &zc->zc_zap->zap_m.zap_phys->mz_chunk
997 			    [mze->mze_chunkid], sizeof (mze->mze_phys)));
998 
999 			za->za_normalization_conflict =
1000 			    mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1001 			za->za_integer_length = 8;
1002 			za->za_num_integers = 1;
1003 			za->za_first_integer = mze->mze_phys.mze_value;
1004 			(void) strcpy(za->za_name, mze->mze_phys.mze_name);
1005 			zc->zc_hash = mze->mze_hash;
1006 			zc->zc_cd = mze->mze_phys.mze_cd;
1007 			err = 0;
1008 		} else {
1009 			zc->zc_hash = -1ULL;
1010 		}
1011 	}
1012 	rw_exit(&zc->zc_zap->zap_rwlock);
1013 	return (err);
1014 }
1015 
1016 void
1017 zap_cursor_advance(zap_cursor_t *zc)
1018 {
1019 	if (zc->zc_hash == -1ULL)
1020 		return;
1021 	zc->zc_cd++;
1022 	if (zc->zc_cd >= ZAP_MAXCD) {
1023 		zc->zc_cd = 0;
1024 		zc->zc_hash += 1ULL<<(64-ZAP_HASHBITS);
1025 		if (zc->zc_hash == 0) /* EOF */
1026 			zc->zc_hash = -1ULL;
1027 	}
1028 }
1029 
1030 int
1031 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1032 {
1033 	int err;
1034 	zap_t *zap;
1035 
1036 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, &zap);
1037 	if (err)
1038 		return (err);
1039 
1040 	bzero(zs, sizeof (zap_stats_t));
1041 
1042 	if (zap->zap_ismicro) {
1043 		zs->zs_blocksize = zap->zap_dbuf->db_size;
1044 		zs->zs_num_entries = zap->zap_m.zap_num_entries;
1045 		zs->zs_num_blocks = 1;
1046 	} else {
1047 		fzap_get_stats(zap, zs);
1048 	}
1049 	zap_unlockdir(zap);
1050 	return (0);
1051 }
1052