xref: /titanic_44/usr/src/uts/common/fs/zfs/zap_micro.c (revision 587032cf0967234b39ccb50adca936a367841063)
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 2006 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 
38 static uint64_t mzap_write_cookie(zap_t *zap, uint64_t cookie,
39     uint64_t entptr);
40 static void mzap_upgrade(zap_t *zap, dmu_tx_t *tx);
41 
42 
43 static void
44 mzap_byteswap(mzap_phys_t *buf, size_t size)
45 {
46 	int i, max;
47 	buf->mz_block_type = BSWAP_64(buf->mz_block_type);
48 	buf->mz_salt = BSWAP_64(buf->mz_salt);
49 	max = (size / MZAP_ENT_LEN) - 1;
50 	for (i = 0; i < max; i++) {
51 		buf->mz_chunk[i].mze_value =
52 		    BSWAP_64(buf->mz_chunk[i].mze_value);
53 		buf->mz_chunk[i].mze_cd =
54 		    BSWAP_32(buf->mz_chunk[i].mze_cd);
55 	}
56 }
57 
58 void
59 zap_byteswap(void *buf, size_t size)
60 {
61 	uint64_t block_type;
62 
63 	block_type = *(uint64_t *)buf;
64 
65 	switch (block_type) {
66 	case ZBT_MICRO:
67 	case BSWAP_64(ZBT_MICRO):
68 		/* ASSERT(magic == ZAP_LEAF_MAGIC); */
69 		mzap_byteswap(buf, size);
70 		return;
71 	default:
72 		fzap_byteswap(buf, size);
73 		return;
74 	}
75 }
76 
77 static int
78 mze_compare(const void *arg1, const void *arg2)
79 {
80 	const mzap_ent_t *mze1 = arg1;
81 	const mzap_ent_t *mze2 = arg2;
82 
83 	if (mze1->mze_hash > mze2->mze_hash)
84 		return (+1);
85 	if (mze1->mze_hash < mze2->mze_hash)
86 		return (-1);
87 	if (mze1->mze_phys.mze_cd > mze2->mze_phys.mze_cd)
88 		return (+1);
89 	if (mze1->mze_phys.mze_cd < mze2->mze_phys.mze_cd)
90 		return (-1);
91 	return (0);
92 }
93 
94 static void
95 mze_insert(zap_t *zap, int chunkid, uint64_t hash, mzap_ent_phys_t *mzep)
96 {
97 	mzap_ent_t *mze;
98 
99 	ASSERT(zap->zap_ismicro);
100 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
101 	ASSERT(mzep->mze_cd < ZAP_MAXCD);
102 	ASSERT3U(zap_hash(zap, mzep->mze_name), ==, hash);
103 
104 	mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
105 	mze->mze_chunkid = chunkid;
106 	mze->mze_hash = hash;
107 	mze->mze_phys = *mzep;
108 	avl_add(&zap->zap_m.zap_avl, mze);
109 }
110 
111 static mzap_ent_t *
112 mze_find(zap_t *zap, const char *name, uint64_t hash)
113 {
114 	mzap_ent_t mze_tofind;
115 	mzap_ent_t *mze;
116 	avl_index_t idx;
117 	avl_tree_t *avl = &zap->zap_m.zap_avl;
118 
119 	ASSERT(zap->zap_ismicro);
120 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
121 	ASSERT3U(zap_hash(zap, name), ==, hash);
122 
123 	if (strlen(name) >= sizeof (mze_tofind.mze_phys.mze_name))
124 		return (NULL);
125 
126 	mze_tofind.mze_hash = hash;
127 	mze_tofind.mze_phys.mze_cd = 0;
128 
129 	mze = avl_find(avl, &mze_tofind, &idx);
130 	if (mze == NULL)
131 		mze = avl_nearest(avl, idx, AVL_AFTER);
132 	for (; mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
133 		if (strcmp(name, mze->mze_phys.mze_name) == 0)
134 			return (mze);
135 	}
136 	return (NULL);
137 }
138 
139 static uint32_t
140 mze_find_unused_cd(zap_t *zap, uint64_t hash)
141 {
142 	mzap_ent_t mze_tofind;
143 	mzap_ent_t *mze;
144 	avl_index_t idx;
145 	avl_tree_t *avl = &zap->zap_m.zap_avl;
146 	uint32_t cd;
147 
148 	ASSERT(zap->zap_ismicro);
149 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
150 
151 	mze_tofind.mze_hash = hash;
152 	mze_tofind.mze_phys.mze_cd = 0;
153 
154 	cd = 0;
155 	for (mze = avl_find(avl, &mze_tofind, &idx);
156 	    mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
157 		if (mze->mze_phys.mze_cd != cd)
158 			break;
159 		cd++;
160 	}
161 
162 	return (cd);
163 }
164 
165 static void
166 mze_remove(zap_t *zap, mzap_ent_t *mze)
167 {
168 	ASSERT(zap->zap_ismicro);
169 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
170 
171 	avl_remove(&zap->zap_m.zap_avl, mze);
172 	kmem_free(mze, sizeof (mzap_ent_t));
173 }
174 
175 static void
176 mze_destroy(zap_t *zap)
177 {
178 	mzap_ent_t *mze;
179 	void *avlcookie = NULL;
180 
181 	while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
182 		kmem_free(mze, sizeof (mzap_ent_t));
183 	avl_destroy(&zap->zap_m.zap_avl);
184 }
185 
186 static zap_t *
187 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
188 {
189 	zap_t *winner;
190 	zap_t *zap;
191 	int i;
192 
193 	ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
194 
195 	zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
196 	rw_init(&zap->zap_rwlock, 0, 0, 0);
197 	rw_enter(&zap->zap_rwlock, RW_WRITER);
198 	zap->zap_objset = os;
199 	zap->zap_object = obj;
200 	zap->zap_dbuf = db;
201 
202 	if (((uint64_t *)db->db_data)[0] != ZBT_MICRO) {
203 		mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
204 		zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
205 	} else {
206 		zap->zap_ismicro = TRUE;
207 	}
208 
209 	/*
210 	 * Make sure that zap_ismicro is set before we let others see
211 	 * it, because zap_lockdir() checks zap_ismicro without the lock
212 	 * held.
213 	 */
214 	winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_pageout);
215 
216 	if (winner != NULL) {
217 		kmem_free(zap, sizeof (zap_t));
218 		return (winner);
219 	}
220 
221 	if (zap->zap_ismicro) {
222 		zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
223 		zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
224 		avl_create(&zap->zap_m.zap_avl, mze_compare,
225 		    sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
226 
227 		for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
228 			mzap_ent_phys_t *mze =
229 			    &zap->zap_m.zap_phys->mz_chunk[i];
230 			if (mze->mze_name[0]) {
231 				zap->zap_m.zap_num_entries++;
232 				mze_insert(zap, i,
233 				    zap_hash(zap, mze->mze_name), mze);
234 			}
235 		}
236 	} else {
237 		zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
238 
239 		ASSERT3U(sizeof (struct zap_leaf_header), ==,
240 		    2*ZAP_LEAF_CHUNKSIZE);
241 
242 		/*
243 		 * The embedded pointer table should not overlap the
244 		 * other members.
245 		 */
246 		ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
247 		    &zap->zap_f.zap_phys->zap_salt);
248 
249 		/*
250 		 * The embedded pointer table should end at the end of
251 		 * the block
252 		 */
253 		ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
254 		    1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
255 		    (uintptr_t)zap->zap_f.zap_phys, ==,
256 		    zap->zap_dbuf->db_size);
257 	}
258 	rw_exit(&zap->zap_rwlock);
259 	return (zap);
260 }
261 
262 int
263 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
264     krw_t lti, int fatreader, zap_t **zapp)
265 {
266 	zap_t *zap;
267 	dmu_buf_t *db;
268 	krw_t lt;
269 	int err;
270 
271 	*zapp = NULL;
272 
273 	err = dmu_buf_hold(os, obj, 0, NULL, &db);
274 	if (err)
275 		return (err);
276 
277 #ifdef ZFS_DEBUG
278 	{
279 		dmu_object_info_t doi;
280 		dmu_object_info_from_db(db, &doi);
281 		ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
282 	}
283 #endif
284 
285 	zap = dmu_buf_get_user(db);
286 	if (zap == NULL)
287 		zap = mzap_open(os, obj, db);
288 
289 	/*
290 	 * We're checking zap_ismicro without the lock held, in order to
291 	 * tell what type of lock we want.  Once we have some sort of
292 	 * lock, see if it really is the right type.  In practice this
293 	 * can only be different if it was upgraded from micro to fat,
294 	 * and micro wanted WRITER but fat only needs READER.
295 	 */
296 	lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
297 	rw_enter(&zap->zap_rwlock, lt);
298 	if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
299 		/* it was upgraded, now we only need reader */
300 		ASSERT(lt == RW_WRITER);
301 		ASSERT(RW_READER ==
302 		    (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
303 		rw_downgrade(&zap->zap_rwlock);
304 		lt = RW_READER;
305 	}
306 
307 	zap->zap_objset = os;
308 
309 	if (lt == RW_WRITER)
310 		dmu_buf_will_dirty(db, tx);
311 
312 	ASSERT3P(zap->zap_dbuf, ==, db);
313 
314 	ASSERT(!zap->zap_ismicro ||
315 	    zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
316 	if (zap->zap_ismicro && tx &&
317 	    zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
318 		uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
319 		if (newsz > MZAP_MAX_BLKSZ) {
320 			dprintf("upgrading obj %llu: num_entries=%u\n",
321 			    obj, zap->zap_m.zap_num_entries);
322 			mzap_upgrade(zap, tx);
323 			*zapp = zap;
324 			return (0);
325 		}
326 		err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
327 		ASSERT3U(err, ==, 0);
328 		zap->zap_m.zap_num_chunks =
329 		    db->db_size / MZAP_ENT_LEN - 1;
330 	}
331 
332 	*zapp = zap;
333 	return (0);
334 }
335 
336 void
337 zap_unlockdir(zap_t *zap)
338 {
339 	rw_exit(&zap->zap_rwlock);
340 	dmu_buf_rele(zap->zap_dbuf, NULL);
341 }
342 
343 static void
344 mzap_upgrade(zap_t *zap, dmu_tx_t *tx)
345 {
346 	mzap_phys_t *mzp;
347 	int i, sz, nchunks, err;
348 
349 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
350 
351 	sz = zap->zap_dbuf->db_size;
352 	mzp = kmem_alloc(sz, KM_SLEEP);
353 	bcopy(zap->zap_dbuf->db_data, mzp, sz);
354 	nchunks = zap->zap_m.zap_num_chunks;
355 
356 	err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
357 	    1ULL << fzap_default_block_shift, 0, tx);
358 	ASSERT(err == 0);
359 
360 	dprintf("upgrading obj=%llu with %u chunks\n",
361 	    zap->zap_object, nchunks);
362 	mze_destroy(zap);
363 
364 	fzap_upgrade(zap, tx);
365 
366 	for (i = 0; i < nchunks; i++) {
367 		int err;
368 		mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
369 		if (mze->mze_name[0] == 0)
370 			continue;
371 		dprintf("adding %s=%llu\n",
372 		    mze->mze_name, mze->mze_value);
373 		err = fzap_add_cd(zap,
374 		    mze->mze_name, 8, 1, &mze->mze_value,
375 		    mze->mze_cd, tx);
376 		ASSERT3U(err, ==, 0);
377 	}
378 	kmem_free(mzp, sz);
379 }
380 
381 uint64_t
382 zap_hash(zap_t *zap, const char *name)
383 {
384 	const uint8_t *cp;
385 	uint8_t c;
386 	uint64_t crc = zap->zap_salt;
387 
388 	ASSERT(crc != 0);
389 	ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
390 	for (cp = (const uint8_t *)name; (c = *cp) != '\0'; cp++)
391 		crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ c) & 0xFF];
392 
393 	/*
394 	 * Only use 28 bits, since we need 4 bits in the cookie for the
395 	 * collision differentiator.  We MUST use the high bits, since
396 	 * those are the onces that we first pay attention to when
397 	 * chosing the bucket.
398 	 */
399 	crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1);
400 
401 	return (crc);
402 }
403 
404 
405 static void
406 mzap_create_impl(objset_t *os, uint64_t obj, dmu_tx_t *tx)
407 {
408 	dmu_buf_t *db;
409 	mzap_phys_t *zp;
410 
411 	VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db));
412 
413 #ifdef ZFS_DEBUG
414 	{
415 		dmu_object_info_t doi;
416 		dmu_object_info_from_db(db, &doi);
417 		ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
418 	}
419 #endif
420 
421 	dmu_buf_will_dirty(db, tx);
422 	zp = db->db_data;
423 	zp->mz_block_type = ZBT_MICRO;
424 	zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
425 	ASSERT(zp->mz_salt != 0);
426 	dmu_buf_rele(db, FTAG);
427 }
428 
429 int
430 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
431     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
432 {
433 	int err;
434 
435 	err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
436 	if (err != 0)
437 		return (err);
438 	mzap_create_impl(os, obj, tx);
439 	return (0);
440 }
441 
442 uint64_t
443 zap_create(objset_t *os, dmu_object_type_t ot,
444     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
445 {
446 	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
447 
448 	mzap_create_impl(os, obj, tx);
449 	return (obj);
450 }
451 
452 int
453 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
454 {
455 	/*
456 	 * dmu_object_free will free the object number and free the
457 	 * data.  Freeing the data will cause our pageout function to be
458 	 * called, which will destroy our data (zap_leaf_t's and zap_t).
459 	 */
460 
461 	return (dmu_object_free(os, zapobj, tx));
462 }
463 
464 _NOTE(ARGSUSED(0))
465 void
466 zap_pageout(dmu_buf_t *db, void *vmzap)
467 {
468 	zap_t *zap = vmzap;
469 
470 	rw_destroy(&zap->zap_rwlock);
471 
472 	if (zap->zap_ismicro)
473 		mze_destroy(zap);
474 
475 	kmem_free(zap, sizeof (zap_t));
476 }
477 
478 
479 int
480 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
481 {
482 	zap_t *zap;
483 	int err;
484 
485 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, &zap);
486 	if (err)
487 		return (err);
488 	if (!zap->zap_ismicro) {
489 		err = fzap_count(zap, count);
490 	} else {
491 		*count = zap->zap_m.zap_num_entries;
492 	}
493 	zap_unlockdir(zap);
494 	return (err);
495 }
496 
497 /*
498  * Routines for maniplulating attributes.
499  */
500 
501 int
502 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
503     uint64_t integer_size, uint64_t num_integers, void *buf)
504 {
505 	zap_t *zap;
506 	int err;
507 	mzap_ent_t *mze;
508 
509 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, &zap);
510 	if (err)
511 		return (err);
512 	if (!zap->zap_ismicro) {
513 		err = fzap_lookup(zap, name,
514 		    integer_size, num_integers, buf);
515 	} else {
516 		mze = mze_find(zap, name, zap_hash(zap, name));
517 		if (mze == NULL) {
518 			err = ENOENT;
519 		} else {
520 			if (num_integers < 1)
521 				err = EOVERFLOW;
522 			else if (integer_size != 8)
523 				err = EINVAL;
524 			else
525 				*(uint64_t *)buf = mze->mze_phys.mze_value;
526 		}
527 	}
528 	zap_unlockdir(zap);
529 	return (err);
530 }
531 
532 int
533 zap_length(objset_t *os, uint64_t zapobj, const char *name,
534     uint64_t *integer_size, uint64_t *num_integers)
535 {
536 	zap_t *zap;
537 	int err;
538 	mzap_ent_t *mze;
539 
540 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, &zap);
541 	if (err)
542 		return (err);
543 	if (!zap->zap_ismicro) {
544 		err = fzap_length(zap, name, integer_size, num_integers);
545 	} else {
546 		mze = mze_find(zap, name, zap_hash(zap, name));
547 		if (mze == NULL) {
548 			err = ENOENT;
549 		} else {
550 			if (integer_size)
551 				*integer_size = 8;
552 			if (num_integers)
553 				*num_integers = 1;
554 		}
555 	}
556 	zap_unlockdir(zap);
557 	return (err);
558 }
559 
560 static void
561 mzap_addent(zap_t *zap, const char *name, uint64_t hash, uint64_t value)
562 {
563 	int i;
564 	int start = zap->zap_m.zap_alloc_next;
565 	uint32_t cd;
566 
567 	dprintf("obj=%llu %s=%llu\n", zap->zap_object, name, value);
568 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
569 
570 #ifdef ZFS_DEBUG
571 	for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
572 		mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
573 		ASSERT(strcmp(name, mze->mze_name) != 0);
574 	}
575 #endif
576 
577 	cd = mze_find_unused_cd(zap, hash);
578 	/* given the limited size of the microzap, this can't happen */
579 	ASSERT(cd != ZAP_MAXCD);
580 
581 again:
582 	for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
583 		mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
584 		if (mze->mze_name[0] == 0) {
585 			mze->mze_value = value;
586 			mze->mze_cd = cd;
587 			(void) strcpy(mze->mze_name, name);
588 			zap->zap_m.zap_num_entries++;
589 			zap->zap_m.zap_alloc_next = i+1;
590 			if (zap->zap_m.zap_alloc_next ==
591 			    zap->zap_m.zap_num_chunks)
592 				zap->zap_m.zap_alloc_next = 0;
593 			mze_insert(zap, i, hash, mze);
594 			return;
595 		}
596 	}
597 	if (start != 0) {
598 		start = 0;
599 		goto again;
600 	}
601 	ASSERT(!"out of entries!");
602 }
603 
604 int
605 zap_add(objset_t *os, uint64_t zapobj, const char *name,
606     int integer_size, uint64_t num_integers,
607     const void *val, dmu_tx_t *tx)
608 {
609 	zap_t *zap;
610 	int err;
611 	mzap_ent_t *mze;
612 	const uint64_t *intval = val;
613 	uint64_t hash;
614 
615 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, &zap);
616 	if (err)
617 		return (err);
618 	if (!zap->zap_ismicro) {
619 		err = fzap_add(zap, name, integer_size, num_integers, val, tx);
620 	} else if (integer_size != 8 || num_integers != 1 ||
621 	    strlen(name) >= MZAP_NAME_LEN) {
622 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
623 		    zapobj, integer_size, num_integers, name);
624 		mzap_upgrade(zap, tx);
625 		err = fzap_add(zap, name, integer_size, num_integers, val, tx);
626 	} else {
627 		hash = zap_hash(zap, name);
628 		mze = mze_find(zap, name, hash);
629 		if (mze != NULL) {
630 			err = EEXIST;
631 		} else {
632 			mzap_addent(zap, name, hash, *intval);
633 		}
634 	}
635 	zap_unlockdir(zap);
636 	return (err);
637 }
638 
639 int
640 zap_update(objset_t *os, uint64_t zapobj, const char *name,
641     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
642 {
643 	zap_t *zap;
644 	mzap_ent_t *mze;
645 	const uint64_t *intval = val;
646 	uint64_t hash;
647 	int err;
648 
649 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, &zap);
650 	if (err)
651 		return (err);
652 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
653 	if (!zap->zap_ismicro) {
654 		err = fzap_update(zap, name,
655 		    integer_size, num_integers, val, tx);
656 	} else if (integer_size != 8 || num_integers != 1 ||
657 	    strlen(name) >= MZAP_NAME_LEN) {
658 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
659 		    zapobj, integer_size, num_integers, name);
660 		mzap_upgrade(zap, tx);
661 		err = fzap_update(zap, name,
662 		    integer_size, num_integers, val, tx);
663 	} else {
664 		hash = zap_hash(zap, name);
665 		mze = mze_find(zap, name, hash);
666 		if (mze != NULL) {
667 			mze->mze_phys.mze_value = *intval;
668 			zap->zap_m.zap_phys->mz_chunk
669 			    [mze->mze_chunkid].mze_value = *intval;
670 		} else {
671 			mzap_addent(zap, name, hash, *intval);
672 		}
673 	}
674 	zap_unlockdir(zap);
675 	return (0);
676 }
677 
678 int
679 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
680 {
681 	zap_t *zap;
682 	int err;
683 	mzap_ent_t *mze;
684 
685 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, &zap);
686 	if (err)
687 		return (err);
688 	if (!zap->zap_ismicro) {
689 		err = fzap_remove(zap, name, tx);
690 	} else {
691 		mze = mze_find(zap, name, zap_hash(zap, name));
692 		if (mze == NULL) {
693 			dprintf("fail: %s\n", name);
694 			err = ENOENT;
695 		} else {
696 			dprintf("success: %s\n", name);
697 			zap->zap_m.zap_num_entries--;
698 			bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
699 			    sizeof (mzap_ent_phys_t));
700 			mze_remove(zap, mze);
701 		}
702 	}
703 	zap_unlockdir(zap);
704 	return (err);
705 }
706 
707 
708 /*
709  * Routines for iterating over the attributes.
710  */
711 
712 /*
713  * We want to keep the high 32 bits of the cursor zero if we can, so
714  * that 32-bit programs can access this.  So use a small hash value so
715  * we can fit 4 bits of cd into the 32-bit cursor.
716  *
717  * [ 4 zero bits | 32-bit collision differentiator | 28-bit hash value ]
718  */
719 void
720 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
721     uint64_t serialized)
722 {
723 	zc->zc_objset = os;
724 	zc->zc_zap = NULL;
725 	zc->zc_leaf = NULL;
726 	zc->zc_zapobj = zapobj;
727 	if (serialized == -1ULL) {
728 		zc->zc_hash = -1ULL;
729 		zc->zc_cd = 0;
730 	} else {
731 		zc->zc_hash = serialized << (64-ZAP_HASHBITS);
732 		zc->zc_cd = serialized >> ZAP_HASHBITS;
733 		if (zc->zc_cd >= ZAP_MAXCD) /* corrupt serialized */
734 			zc->zc_cd = 0;
735 	}
736 }
737 
738 void
739 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
740 {
741 	zap_cursor_init_serialized(zc, os, zapobj, 0);
742 }
743 
744 void
745 zap_cursor_fini(zap_cursor_t *zc)
746 {
747 	if (zc->zc_zap) {
748 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
749 		zap_unlockdir(zc->zc_zap);
750 		zc->zc_zap = NULL;
751 	}
752 	if (zc->zc_leaf) {
753 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
754 		zap_put_leaf(zc->zc_leaf);
755 		zc->zc_leaf = NULL;
756 	}
757 	zc->zc_objset = NULL;
758 }
759 
760 uint64_t
761 zap_cursor_serialize(zap_cursor_t *zc)
762 {
763 	if (zc->zc_hash == -1ULL)
764 		return (-1ULL);
765 	ASSERT((zc->zc_hash & (ZAP_MAXCD-1)) == 0);
766 	ASSERT(zc->zc_cd < ZAP_MAXCD);
767 	return ((zc->zc_hash >> (64-ZAP_HASHBITS)) |
768 	    ((uint64_t)zc->zc_cd << ZAP_HASHBITS));
769 }
770 
771 int
772 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
773 {
774 	int err;
775 	avl_index_t idx;
776 	mzap_ent_t mze_tofind;
777 	mzap_ent_t *mze;
778 
779 	if (zc->zc_hash == -1ULL)
780 		return (ENOENT);
781 
782 	if (zc->zc_zap == NULL) {
783 		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
784 		    RW_READER, TRUE, &zc->zc_zap);
785 		if (err)
786 			return (err);
787 	} else {
788 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
789 	}
790 	if (!zc->zc_zap->zap_ismicro) {
791 		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
792 	} else {
793 		err = ENOENT;
794 
795 		mze_tofind.mze_hash = zc->zc_hash;
796 		mze_tofind.mze_phys.mze_cd = zc->zc_cd;
797 
798 		mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
799 		ASSERT(mze == NULL || 0 == bcmp(&mze->mze_phys,
800 		    &zc->zc_zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
801 		    sizeof (mze->mze_phys)));
802 		if (mze == NULL) {
803 			mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
804 			    idx, AVL_AFTER);
805 		}
806 		if (mze) {
807 			za->za_integer_length = 8;
808 			za->za_num_integers = 1;
809 			za->za_first_integer = mze->mze_phys.mze_value;
810 			(void) strcpy(za->za_name, mze->mze_phys.mze_name);
811 			zc->zc_hash = mze->mze_hash;
812 			zc->zc_cd = mze->mze_phys.mze_cd;
813 			err = 0;
814 		} else {
815 			zc->zc_hash = -1ULL;
816 		}
817 	}
818 	rw_exit(&zc->zc_zap->zap_rwlock);
819 	return (err);
820 }
821 
822 void
823 zap_cursor_advance(zap_cursor_t *zc)
824 {
825 	if (zc->zc_hash == -1ULL)
826 		return;
827 	zc->zc_cd++;
828 	if (zc->zc_cd >= ZAP_MAXCD) {
829 		zc->zc_cd = 0;
830 		zc->zc_hash += 1ULL<<(64-ZAP_HASHBITS);
831 		if (zc->zc_hash == 0) /* EOF */
832 			zc->zc_hash = -1ULL;
833 	}
834 }
835 
836 int
837 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
838 {
839 	int err;
840 	zap_t *zap;
841 
842 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, &zap);
843 	if (err)
844 		return (err);
845 
846 	bzero(zs, sizeof (zap_stats_t));
847 
848 	if (zap->zap_ismicro) {
849 		zs->zs_blocksize = zap->zap_dbuf->db_size;
850 		zs->zs_num_entries = zap->zap_m.zap_num_entries;
851 		zs->zs_num_blocks = 1;
852 	} else {
853 		fzap_get_stats(zap, zs);
854 	}
855 	zap_unlockdir(zap);
856 	return (0);
857 }
858