xref: /titanic_52/usr/src/uts/common/fs/zfs/zap_micro.c (revision 71269a2275bf5a143dad6461eee2710a344e7261)
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 2008 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 
37 #ifdef _KERNEL
38 #include <sys/sunddi.h>
39 #endif
40 
41 static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx);
42 
43 
44 static uint64_t
45 zap_hash(zap_t *zap, const char *normname)
46 {
47 	const uint8_t *cp;
48 	uint8_t c;
49 	uint64_t crc = zap->zap_salt;
50 
51 	/* NB: name must already be normalized, if necessary */
52 
53 	ASSERT(crc != 0);
54 	ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
55 	for (cp = (const uint8_t *)normname; (c = *cp) != '\0'; cp++) {
56 		crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ c) & 0xFF];
57 	}
58 
59 	/*
60 	 * Only use 28 bits, since we need 4 bits in the cookie for the
61 	 * collision differentiator.  We MUST use the high bits, since
62 	 * those are the ones that we first pay attention to when
63 	 * chosing the bucket.
64 	 */
65 	crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1);
66 
67 	return (crc);
68 }
69 
70 static int
71 zap_normalize(zap_t *zap, const char *name, char *namenorm)
72 {
73 	size_t inlen, outlen;
74 	int err;
75 
76 	inlen = strlen(name) + 1;
77 	outlen = ZAP_MAXNAMELEN;
78 
79 	err = 0;
80 	(void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
81 	    zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL, U8_UNICODE_LATEST,
82 	    &err);
83 
84 	return (err);
85 }
86 
87 boolean_t
88 zap_match(zap_name_t *zn, const char *matchname)
89 {
90 	if (zn->zn_matchtype == MT_FIRST) {
91 		char norm[ZAP_MAXNAMELEN];
92 
93 		if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
94 			return (B_FALSE);
95 
96 		return (strcmp(zn->zn_name_norm, norm) == 0);
97 	} else {
98 		/* MT_BEST or MT_EXACT */
99 		return (strcmp(zn->zn_name_orij, matchname) == 0);
100 	}
101 }
102 
103 void
104 zap_name_free(zap_name_t *zn)
105 {
106 	kmem_free(zn, sizeof (zap_name_t));
107 }
108 
109 /* XXX combine this with zap_lockdir()? */
110 zap_name_t *
111 zap_name_alloc(zap_t *zap, const char *name, matchtype_t mt)
112 {
113 	zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
114 
115 	zn->zn_zap = zap;
116 	zn->zn_name_orij = name;
117 	zn->zn_matchtype = mt;
118 	if (zap->zap_normflags) {
119 		if (zap_normalize(zap, name, zn->zn_normbuf) != 0) {
120 			zap_name_free(zn);
121 			return (NULL);
122 		}
123 		zn->zn_name_norm = zn->zn_normbuf;
124 	} else {
125 		if (mt != MT_EXACT) {
126 			zap_name_free(zn);
127 			return (NULL);
128 		}
129 		zn->zn_name_norm = zn->zn_name_orij;
130 	}
131 
132 	zn->zn_hash = zap_hash(zap, zn->zn_name_norm);
133 	return (zn);
134 }
135 
136 static void
137 mzap_byteswap(mzap_phys_t *buf, size_t size)
138 {
139 	int i, max;
140 	buf->mz_block_type = BSWAP_64(buf->mz_block_type);
141 	buf->mz_salt = BSWAP_64(buf->mz_salt);
142 	buf->mz_normflags = BSWAP_64(buf->mz_normflags);
143 	max = (size / MZAP_ENT_LEN) - 1;
144 	for (i = 0; i < max; i++) {
145 		buf->mz_chunk[i].mze_value =
146 		    BSWAP_64(buf->mz_chunk[i].mze_value);
147 		buf->mz_chunk[i].mze_cd =
148 		    BSWAP_32(buf->mz_chunk[i].mze_cd);
149 	}
150 }
151 
152 void
153 zap_byteswap(void *buf, size_t size)
154 {
155 	uint64_t block_type;
156 
157 	block_type = *(uint64_t *)buf;
158 
159 	if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
160 		/* ASSERT(magic == ZAP_LEAF_MAGIC); */
161 		mzap_byteswap(buf, size);
162 	} else {
163 		fzap_byteswap(buf, size);
164 	}
165 }
166 
167 static int
168 mze_compare(const void *arg1, const void *arg2)
169 {
170 	const mzap_ent_t *mze1 = arg1;
171 	const mzap_ent_t *mze2 = arg2;
172 
173 	if (mze1->mze_hash > mze2->mze_hash)
174 		return (+1);
175 	if (mze1->mze_hash < mze2->mze_hash)
176 		return (-1);
177 	if (mze1->mze_phys.mze_cd > mze2->mze_phys.mze_cd)
178 		return (+1);
179 	if (mze1->mze_phys.mze_cd < mze2->mze_phys.mze_cd)
180 		return (-1);
181 	return (0);
182 }
183 
184 static void
185 mze_insert(zap_t *zap, int chunkid, uint64_t hash, mzap_ent_phys_t *mzep)
186 {
187 	mzap_ent_t *mze;
188 
189 	ASSERT(zap->zap_ismicro);
190 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
191 	ASSERT(mzep->mze_cd < ZAP_MAXCD);
192 
193 	mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
194 	mze->mze_chunkid = chunkid;
195 	mze->mze_hash = hash;
196 	mze->mze_phys = *mzep;
197 	avl_add(&zap->zap_m.zap_avl, mze);
198 }
199 
200 static mzap_ent_t *
201 mze_find(zap_name_t *zn)
202 {
203 	mzap_ent_t mze_tofind;
204 	mzap_ent_t *mze;
205 	avl_index_t idx;
206 	avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
207 
208 	ASSERT(zn->zn_zap->zap_ismicro);
209 	ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
210 
211 	if (strlen(zn->zn_name_norm) >= sizeof (mze_tofind.mze_phys.mze_name))
212 		return (NULL);
213 
214 	mze_tofind.mze_hash = zn->zn_hash;
215 	mze_tofind.mze_phys.mze_cd = 0;
216 
217 again:
218 	mze = avl_find(avl, &mze_tofind, &idx);
219 	if (mze == NULL)
220 		mze = avl_nearest(avl, idx, AVL_AFTER);
221 	for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
222 		if (zap_match(zn, mze->mze_phys.mze_name))
223 			return (mze);
224 	}
225 	if (zn->zn_matchtype == MT_BEST) {
226 		zn->zn_matchtype = MT_FIRST;
227 		goto again;
228 	}
229 	return (NULL);
230 }
231 
232 static uint32_t
233 mze_find_unused_cd(zap_t *zap, uint64_t hash)
234 {
235 	mzap_ent_t mze_tofind;
236 	mzap_ent_t *mze;
237 	avl_index_t idx;
238 	avl_tree_t *avl = &zap->zap_m.zap_avl;
239 	uint32_t cd;
240 
241 	ASSERT(zap->zap_ismicro);
242 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
243 
244 	mze_tofind.mze_hash = hash;
245 	mze_tofind.mze_phys.mze_cd = 0;
246 
247 	cd = 0;
248 	for (mze = avl_find(avl, &mze_tofind, &idx);
249 	    mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
250 		if (mze->mze_phys.mze_cd != cd)
251 			break;
252 		cd++;
253 	}
254 
255 	return (cd);
256 }
257 
258 static void
259 mze_remove(zap_t *zap, mzap_ent_t *mze)
260 {
261 	ASSERT(zap->zap_ismicro);
262 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
263 
264 	avl_remove(&zap->zap_m.zap_avl, mze);
265 	kmem_free(mze, sizeof (mzap_ent_t));
266 }
267 
268 static void
269 mze_destroy(zap_t *zap)
270 {
271 	mzap_ent_t *mze;
272 	void *avlcookie = NULL;
273 
274 	while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
275 		kmem_free(mze, sizeof (mzap_ent_t));
276 	avl_destroy(&zap->zap_m.zap_avl);
277 }
278 
279 static zap_t *
280 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
281 {
282 	zap_t *winner;
283 	zap_t *zap;
284 	int i;
285 
286 	ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
287 
288 	zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
289 	rw_init(&zap->zap_rwlock, 0, 0, 0);
290 	rw_enter(&zap->zap_rwlock, RW_WRITER);
291 	zap->zap_objset = os;
292 	zap->zap_object = obj;
293 	zap->zap_dbuf = db;
294 
295 	if (*(uint64_t *)db->db_data != ZBT_MICRO) {
296 		mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
297 		zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
298 	} else {
299 		zap->zap_ismicro = TRUE;
300 	}
301 
302 	/*
303 	 * Make sure that zap_ismicro is set before we let others see
304 	 * it, because zap_lockdir() checks zap_ismicro without the lock
305 	 * held.
306 	 */
307 	winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict);
308 
309 	if (winner != NULL) {
310 		rw_exit(&zap->zap_rwlock);
311 		rw_destroy(&zap->zap_rwlock);
312 		if (!zap->zap_ismicro)
313 			mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
314 		kmem_free(zap, sizeof (zap_t));
315 		return (winner);
316 	}
317 
318 	if (zap->zap_ismicro) {
319 		zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
320 		zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags;
321 		zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
322 		avl_create(&zap->zap_m.zap_avl, mze_compare,
323 		    sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
324 
325 		for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
326 			mzap_ent_phys_t *mze =
327 			    &zap->zap_m.zap_phys->mz_chunk[i];
328 			if (mze->mze_name[0]) {
329 				zap_name_t *zn;
330 
331 				zap->zap_m.zap_num_entries++;
332 				zn = zap_name_alloc(zap, mze->mze_name,
333 				    MT_EXACT);
334 				mze_insert(zap, i, zn->zn_hash, mze);
335 				zap_name_free(zn);
336 			}
337 		}
338 	} else {
339 		zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
340 		zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags;
341 
342 		ASSERT3U(sizeof (struct zap_leaf_header), ==,
343 		    2*ZAP_LEAF_CHUNKSIZE);
344 
345 		/*
346 		 * The embedded pointer table should not overlap the
347 		 * other members.
348 		 */
349 		ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
350 		    &zap->zap_f.zap_phys->zap_salt);
351 
352 		/*
353 		 * The embedded pointer table should end at the end of
354 		 * the block
355 		 */
356 		ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
357 		    1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
358 		    (uintptr_t)zap->zap_f.zap_phys, ==,
359 		    zap->zap_dbuf->db_size);
360 	}
361 	rw_exit(&zap->zap_rwlock);
362 	return (zap);
363 }
364 
365 int
366 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
367     krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
368 {
369 	zap_t *zap;
370 	dmu_buf_t *db;
371 	krw_t lt;
372 	int err;
373 
374 	*zapp = NULL;
375 
376 	err = dmu_buf_hold(os, obj, 0, NULL, &db);
377 	if (err)
378 		return (err);
379 
380 #ifdef ZFS_DEBUG
381 	{
382 		dmu_object_info_t doi;
383 		dmu_object_info_from_db(db, &doi);
384 		ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
385 	}
386 #endif
387 
388 	zap = dmu_buf_get_user(db);
389 	if (zap == NULL)
390 		zap = mzap_open(os, obj, db);
391 
392 	/*
393 	 * We're checking zap_ismicro without the lock held, in order to
394 	 * tell what type of lock we want.  Once we have some sort of
395 	 * lock, see if it really is the right type.  In practice this
396 	 * can only be different if it was upgraded from micro to fat,
397 	 * and micro wanted WRITER but fat only needs READER.
398 	 */
399 	lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
400 	rw_enter(&zap->zap_rwlock, lt);
401 	if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
402 		/* it was upgraded, now we only need reader */
403 		ASSERT(lt == RW_WRITER);
404 		ASSERT(RW_READER ==
405 		    (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
406 		rw_downgrade(&zap->zap_rwlock);
407 		lt = RW_READER;
408 	}
409 
410 	zap->zap_objset = os;
411 
412 	if (lt == RW_WRITER)
413 		dmu_buf_will_dirty(db, tx);
414 
415 	ASSERT3P(zap->zap_dbuf, ==, db);
416 
417 	ASSERT(!zap->zap_ismicro ||
418 	    zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
419 	if (zap->zap_ismicro && tx && adding &&
420 	    zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
421 		uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
422 		if (newsz > MZAP_MAX_BLKSZ) {
423 			dprintf("upgrading obj %llu: num_entries=%u\n",
424 			    obj, zap->zap_m.zap_num_entries);
425 			*zapp = zap;
426 			return (mzap_upgrade(zapp, tx));
427 		}
428 		err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
429 		ASSERT3U(err, ==, 0);
430 		zap->zap_m.zap_num_chunks =
431 		    db->db_size / MZAP_ENT_LEN - 1;
432 	}
433 
434 	*zapp = zap;
435 	return (0);
436 }
437 
438 void
439 zap_unlockdir(zap_t *zap)
440 {
441 	rw_exit(&zap->zap_rwlock);
442 	dmu_buf_rele(zap->zap_dbuf, NULL);
443 }
444 
445 static int
446 mzap_upgrade(zap_t **zapp, dmu_tx_t *tx)
447 {
448 	mzap_phys_t *mzp;
449 	int i, sz, nchunks, err;
450 	zap_t *zap = *zapp;
451 
452 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
453 
454 	sz = zap->zap_dbuf->db_size;
455 	mzp = kmem_alloc(sz, KM_SLEEP);
456 	bcopy(zap->zap_dbuf->db_data, mzp, sz);
457 	nchunks = zap->zap_m.zap_num_chunks;
458 
459 	err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
460 	    1ULL << fzap_default_block_shift, 0, tx);
461 	if (err) {
462 		kmem_free(mzp, sz);
463 		return (err);
464 	}
465 
466 	dprintf("upgrading obj=%llu with %u chunks\n",
467 	    zap->zap_object, nchunks);
468 	/* XXX destroy the avl later, so we can use the stored hash value */
469 	mze_destroy(zap);
470 
471 	fzap_upgrade(zap, tx);
472 
473 	for (i = 0; i < nchunks; i++) {
474 		int err;
475 		mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
476 		zap_name_t *zn;
477 		if (mze->mze_name[0] == 0)
478 			continue;
479 		dprintf("adding %s=%llu\n",
480 		    mze->mze_name, mze->mze_value);
481 		zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
482 		err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
483 		zap = zn->zn_zap;	/* fzap_add_cd() may change zap */
484 		zap_name_free(zn);
485 		if (err)
486 			break;
487 	}
488 	kmem_free(mzp, sz);
489 	*zapp = zap;
490 	return (err);
491 }
492 
493 static void
494 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, dmu_tx_t *tx)
495 {
496 	dmu_buf_t *db;
497 	mzap_phys_t *zp;
498 
499 	VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db));
500 
501 #ifdef ZFS_DEBUG
502 	{
503 		dmu_object_info_t doi;
504 		dmu_object_info_from_db(db, &doi);
505 		ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
506 	}
507 #endif
508 
509 	dmu_buf_will_dirty(db, tx);
510 	zp = db->db_data;
511 	zp->mz_block_type = ZBT_MICRO;
512 	zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
513 	zp->mz_normflags = normflags;
514 	dmu_buf_rele(db, FTAG);
515 }
516 
517 int
518 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
519     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
520 {
521 	return (zap_create_claim_norm(os, obj,
522 	    0, ot, bonustype, bonuslen, tx));
523 }
524 
525 int
526 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
527     dmu_object_type_t ot,
528     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
529 {
530 	int err;
531 
532 	err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
533 	if (err != 0)
534 		return (err);
535 	mzap_create_impl(os, obj, normflags, tx);
536 	return (0);
537 }
538 
539 uint64_t
540 zap_create(objset_t *os, dmu_object_type_t ot,
541     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
542 {
543 	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
544 }
545 
546 uint64_t
547 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
548     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
549 {
550 	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
551 
552 	mzap_create_impl(os, obj, normflags, tx);
553 	return (obj);
554 }
555 
556 int
557 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
558 {
559 	/*
560 	 * dmu_object_free will free the object number and free the
561 	 * data.  Freeing the data will cause our pageout function to be
562 	 * called, which will destroy our data (zap_leaf_t's and zap_t).
563 	 */
564 
565 	return (dmu_object_free(os, zapobj, tx));
566 }
567 
568 _NOTE(ARGSUSED(0))
569 void
570 zap_evict(dmu_buf_t *db, void *vzap)
571 {
572 	zap_t *zap = vzap;
573 
574 	rw_destroy(&zap->zap_rwlock);
575 
576 	if (zap->zap_ismicro)
577 		mze_destroy(zap);
578 	else
579 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
580 
581 	kmem_free(zap, sizeof (zap_t));
582 }
583 
584 int
585 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
586 {
587 	zap_t *zap;
588 	int err;
589 
590 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
591 	if (err)
592 		return (err);
593 	if (!zap->zap_ismicro) {
594 		err = fzap_count(zap, count);
595 	} else {
596 		*count = zap->zap_m.zap_num_entries;
597 	}
598 	zap_unlockdir(zap);
599 	return (err);
600 }
601 
602 /*
603  * zn may be NULL; if not specified, it will be computed if needed.
604  * See also the comment above zap_entry_normalization_conflict().
605  */
606 static boolean_t
607 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
608 {
609 	mzap_ent_t *other;
610 	int direction = AVL_BEFORE;
611 	boolean_t allocdzn = B_FALSE;
612 
613 	if (zap->zap_normflags == 0)
614 		return (B_FALSE);
615 
616 again:
617 	for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
618 	    other && other->mze_hash == mze->mze_hash;
619 	    other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
620 
621 		if (zn == NULL) {
622 			zn = zap_name_alloc(zap, mze->mze_phys.mze_name,
623 			    MT_FIRST);
624 			allocdzn = B_TRUE;
625 		}
626 		if (zap_match(zn, other->mze_phys.mze_name)) {
627 			if (allocdzn)
628 				zap_name_free(zn);
629 			return (B_TRUE);
630 		}
631 	}
632 
633 	if (direction == AVL_BEFORE) {
634 		direction = AVL_AFTER;
635 		goto again;
636 	}
637 
638 	if (allocdzn)
639 		zap_name_free(zn);
640 	return (B_FALSE);
641 }
642 
643 /*
644  * Routines for manipulating attributes.
645  */
646 
647 int
648 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
649     uint64_t integer_size, uint64_t num_integers, void *buf)
650 {
651 	return (zap_lookup_norm(os, zapobj, name, integer_size,
652 	    num_integers, buf, MT_EXACT, NULL, 0, NULL));
653 }
654 
655 int
656 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
657     uint64_t integer_size, uint64_t num_integers, void *buf,
658     matchtype_t mt, char *realname, int rn_len,
659     boolean_t *ncp)
660 {
661 	zap_t *zap;
662 	int err;
663 	mzap_ent_t *mze;
664 	zap_name_t *zn;
665 
666 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
667 	if (err)
668 		return (err);
669 	zn = zap_name_alloc(zap, name, mt);
670 	if (zn == NULL) {
671 		zap_unlockdir(zap);
672 		return (ENOTSUP);
673 	}
674 
675 	if (!zap->zap_ismicro) {
676 		err = fzap_lookup(zn, integer_size, num_integers, buf,
677 		    realname, rn_len, ncp);
678 	} else {
679 		mze = mze_find(zn);
680 		if (mze == NULL) {
681 			err = ENOENT;
682 		} else {
683 			if (num_integers < 1) {
684 				err = EOVERFLOW;
685 			} else if (integer_size != 8) {
686 				err = EINVAL;
687 			} else {
688 				*(uint64_t *)buf = mze->mze_phys.mze_value;
689 				(void) strlcpy(realname,
690 				    mze->mze_phys.mze_name, rn_len);
691 				if (ncp) {
692 					*ncp = mzap_normalization_conflict(zap,
693 					    zn, mze);
694 				}
695 			}
696 		}
697 	}
698 	zap_name_free(zn);
699 	zap_unlockdir(zap);
700 	return (err);
701 }
702 
703 int
704 zap_length(objset_t *os, uint64_t zapobj, const char *name,
705     uint64_t *integer_size, uint64_t *num_integers)
706 {
707 	zap_t *zap;
708 	int err;
709 	mzap_ent_t *mze;
710 	zap_name_t *zn;
711 
712 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
713 	if (err)
714 		return (err);
715 	zn = zap_name_alloc(zap, name, MT_EXACT);
716 	if (zn == NULL) {
717 		zap_unlockdir(zap);
718 		return (ENOTSUP);
719 	}
720 	if (!zap->zap_ismicro) {
721 		err = fzap_length(zn, integer_size, num_integers);
722 	} else {
723 		mze = mze_find(zn);
724 		if (mze == NULL) {
725 			err = ENOENT;
726 		} else {
727 			if (integer_size)
728 				*integer_size = 8;
729 			if (num_integers)
730 				*num_integers = 1;
731 		}
732 	}
733 	zap_name_free(zn);
734 	zap_unlockdir(zap);
735 	return (err);
736 }
737 
738 static void
739 mzap_addent(zap_name_t *zn, uint64_t value)
740 {
741 	int i;
742 	zap_t *zap = zn->zn_zap;
743 	int start = zap->zap_m.zap_alloc_next;
744 	uint32_t cd;
745 
746 	dprintf("obj=%llu %s=%llu\n", zap->zap_object,
747 	    zn->zn_name_orij, value);
748 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
749 
750 #ifdef ZFS_DEBUG
751 	for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
752 		mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
753 		ASSERT(strcmp(zn->zn_name_orij, mze->mze_name) != 0);
754 	}
755 #endif
756 
757 	cd = mze_find_unused_cd(zap, zn->zn_hash);
758 	/* given the limited size of the microzap, this can't happen */
759 	ASSERT(cd != ZAP_MAXCD);
760 
761 again:
762 	for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
763 		mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
764 		if (mze->mze_name[0] == 0) {
765 			mze->mze_value = value;
766 			mze->mze_cd = cd;
767 			(void) strcpy(mze->mze_name, zn->zn_name_orij);
768 			zap->zap_m.zap_num_entries++;
769 			zap->zap_m.zap_alloc_next = i+1;
770 			if (zap->zap_m.zap_alloc_next ==
771 			    zap->zap_m.zap_num_chunks)
772 				zap->zap_m.zap_alloc_next = 0;
773 			mze_insert(zap, i, zn->zn_hash, mze);
774 			return;
775 		}
776 	}
777 	if (start != 0) {
778 		start = 0;
779 		goto again;
780 	}
781 	ASSERT(!"out of entries!");
782 }
783 
784 int
785 zap_add(objset_t *os, uint64_t zapobj, const char *name,
786     int integer_size, uint64_t num_integers,
787     const void *val, dmu_tx_t *tx)
788 {
789 	zap_t *zap;
790 	int err;
791 	mzap_ent_t *mze;
792 	const uint64_t *intval = val;
793 	zap_name_t *zn;
794 
795 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
796 	if (err)
797 		return (err);
798 	zn = zap_name_alloc(zap, name, MT_EXACT);
799 	if (zn == NULL) {
800 		zap_unlockdir(zap);
801 		return (ENOTSUP);
802 	}
803 	if (!zap->zap_ismicro) {
804 		err = fzap_add(zn, integer_size, num_integers, val, tx);
805 		zap = zn->zn_zap;	/* fzap_add() may change zap */
806 	} else if (integer_size != 8 || num_integers != 1 ||
807 	    strlen(name) >= MZAP_NAME_LEN) {
808 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
809 		    zapobj, integer_size, num_integers, name);
810 		err = mzap_upgrade(&zn->zn_zap, tx);
811 		if (err == 0)
812 			err = fzap_add(zn, integer_size, num_integers, val, tx);
813 		zap = zn->zn_zap;	/* fzap_add() may change zap */
814 	} else {
815 		mze = mze_find(zn);
816 		if (mze != NULL) {
817 			err = EEXIST;
818 		} else {
819 			mzap_addent(zn, *intval);
820 		}
821 	}
822 	ASSERT(zap == zn->zn_zap);
823 	zap_name_free(zn);
824 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
825 		zap_unlockdir(zap);
826 	return (err);
827 }
828 
829 int
830 zap_update(objset_t *os, uint64_t zapobj, const char *name,
831     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
832 {
833 	zap_t *zap;
834 	mzap_ent_t *mze;
835 	const uint64_t *intval = val;
836 	zap_name_t *zn;
837 	int err;
838 
839 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
840 	if (err)
841 		return (err);
842 	zn = zap_name_alloc(zap, name, MT_EXACT);
843 	if (zn == NULL) {
844 		zap_unlockdir(zap);
845 		return (ENOTSUP);
846 	}
847 	if (!zap->zap_ismicro) {
848 		err = fzap_update(zn, integer_size, num_integers, val, tx);
849 		zap = zn->zn_zap;	/* fzap_update() may change zap */
850 	} else if (integer_size != 8 || num_integers != 1 ||
851 	    strlen(name) >= MZAP_NAME_LEN) {
852 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
853 		    zapobj, integer_size, num_integers, name);
854 		err = mzap_upgrade(&zn->zn_zap, tx);
855 		if (err == 0)
856 			err = fzap_update(zn, integer_size, num_integers,
857 			    val, tx);
858 		zap = zn->zn_zap;	/* fzap_update() may change zap */
859 	} else {
860 		mze = mze_find(zn);
861 		if (mze != NULL) {
862 			mze->mze_phys.mze_value = *intval;
863 			zap->zap_m.zap_phys->mz_chunk
864 			    [mze->mze_chunkid].mze_value = *intval;
865 		} else {
866 			mzap_addent(zn, *intval);
867 		}
868 	}
869 	ASSERT(zap == zn->zn_zap);
870 	zap_name_free(zn);
871 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
872 		zap_unlockdir(zap);
873 	return (err);
874 }
875 
876 int
877 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
878 {
879 	return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
880 }
881 
882 int
883 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
884     matchtype_t mt, dmu_tx_t *tx)
885 {
886 	zap_t *zap;
887 	int err;
888 	mzap_ent_t *mze;
889 	zap_name_t *zn;
890 
891 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
892 	if (err)
893 		return (err);
894 	zn = zap_name_alloc(zap, name, mt);
895 	if (zn == NULL) {
896 		zap_unlockdir(zap);
897 		return (ENOTSUP);
898 	}
899 	if (!zap->zap_ismicro) {
900 		err = fzap_remove(zn, tx);
901 	} else {
902 		mze = mze_find(zn);
903 		if (mze == NULL) {
904 			err = ENOENT;
905 		} else {
906 			zap->zap_m.zap_num_entries--;
907 			bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
908 			    sizeof (mzap_ent_phys_t));
909 			mze_remove(zap, mze);
910 		}
911 	}
912 	zap_name_free(zn);
913 	zap_unlockdir(zap);
914 	return (err);
915 }
916 
917 /*
918  * Routines for iterating over the attributes.
919  */
920 
921 /*
922  * We want to keep the high 32 bits of the cursor zero if we can, so
923  * that 32-bit programs can access this.  So use a small hash value so
924  * we can fit 4 bits of cd into the 32-bit cursor.
925  *
926  * [ 4 zero bits | 32-bit collision differentiator | 28-bit hash value ]
927  */
928 void
929 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
930     uint64_t serialized)
931 {
932 	zc->zc_objset = os;
933 	zc->zc_zap = NULL;
934 	zc->zc_leaf = NULL;
935 	zc->zc_zapobj = zapobj;
936 	if (serialized == -1ULL) {
937 		zc->zc_hash = -1ULL;
938 		zc->zc_cd = 0;
939 	} else {
940 		zc->zc_hash = serialized << (64-ZAP_HASHBITS);
941 		zc->zc_cd = serialized >> ZAP_HASHBITS;
942 		if (zc->zc_cd >= ZAP_MAXCD) /* corrupt serialized */
943 			zc->zc_cd = 0;
944 	}
945 }
946 
947 void
948 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
949 {
950 	zap_cursor_init_serialized(zc, os, zapobj, 0);
951 }
952 
953 void
954 zap_cursor_fini(zap_cursor_t *zc)
955 {
956 	if (zc->zc_zap) {
957 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
958 		zap_unlockdir(zc->zc_zap);
959 		zc->zc_zap = NULL;
960 	}
961 	if (zc->zc_leaf) {
962 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
963 		zap_put_leaf(zc->zc_leaf);
964 		zc->zc_leaf = NULL;
965 	}
966 	zc->zc_objset = NULL;
967 }
968 
969 uint64_t
970 zap_cursor_serialize(zap_cursor_t *zc)
971 {
972 	if (zc->zc_hash == -1ULL)
973 		return (-1ULL);
974 	ASSERT((zc->zc_hash & (ZAP_MAXCD-1)) == 0);
975 	ASSERT(zc->zc_cd < ZAP_MAXCD);
976 	return ((zc->zc_hash >> (64-ZAP_HASHBITS)) |
977 	    ((uint64_t)zc->zc_cd << ZAP_HASHBITS));
978 }
979 
980 int
981 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
982 {
983 	int err;
984 	avl_index_t idx;
985 	mzap_ent_t mze_tofind;
986 	mzap_ent_t *mze;
987 
988 	if (zc->zc_hash == -1ULL)
989 		return (ENOENT);
990 
991 	if (zc->zc_zap == NULL) {
992 		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
993 		    RW_READER, TRUE, FALSE, &zc->zc_zap);
994 		if (err)
995 			return (err);
996 	} else {
997 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
998 	}
999 	if (!zc->zc_zap->zap_ismicro) {
1000 		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1001 	} else {
1002 		err = ENOENT;
1003 
1004 		mze_tofind.mze_hash = zc->zc_hash;
1005 		mze_tofind.mze_phys.mze_cd = zc->zc_cd;
1006 
1007 		mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1008 		if (mze == NULL) {
1009 			mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1010 			    idx, AVL_AFTER);
1011 		}
1012 		if (mze) {
1013 			ASSERT(0 == bcmp(&mze->mze_phys,
1014 			    &zc->zc_zap->zap_m.zap_phys->mz_chunk
1015 			    [mze->mze_chunkid], sizeof (mze->mze_phys)));
1016 
1017 			za->za_normalization_conflict =
1018 			    mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1019 			za->za_integer_length = 8;
1020 			za->za_num_integers = 1;
1021 			za->za_first_integer = mze->mze_phys.mze_value;
1022 			(void) strcpy(za->za_name, mze->mze_phys.mze_name);
1023 			zc->zc_hash = mze->mze_hash;
1024 			zc->zc_cd = mze->mze_phys.mze_cd;
1025 			err = 0;
1026 		} else {
1027 			zc->zc_hash = -1ULL;
1028 		}
1029 	}
1030 	rw_exit(&zc->zc_zap->zap_rwlock);
1031 	return (err);
1032 }
1033 
1034 void
1035 zap_cursor_advance(zap_cursor_t *zc)
1036 {
1037 	if (zc->zc_hash == -1ULL)
1038 		return;
1039 	zc->zc_cd++;
1040 	if (zc->zc_cd >= ZAP_MAXCD) {
1041 		zc->zc_cd = 0;
1042 		zc->zc_hash += 1ULL<<(64-ZAP_HASHBITS);
1043 		if (zc->zc_hash == 0) /* EOF */
1044 			zc->zc_hash = -1ULL;
1045 	}
1046 }
1047 
1048 int
1049 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1050 {
1051 	int err;
1052 	zap_t *zap;
1053 
1054 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1055 	if (err)
1056 		return (err);
1057 
1058 	bzero(zs, sizeof (zap_stats_t));
1059 
1060 	if (zap->zap_ismicro) {
1061 		zs->zs_blocksize = zap->zap_dbuf->db_size;
1062 		zs->zs_num_entries = zap->zap_m.zap_num_entries;
1063 		zs->zs_num_blocks = 1;
1064 	} else {
1065 		fzap_get_stats(zap, zs);
1066 	}
1067 	zap_unlockdir(zap);
1068 	return (0);
1069 }
1070