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.
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.
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]
22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #pragma ident "%Z%%M% %I% %E% SMI"
30 #include <sys/zfs_context.h>
32 #include <sys/refcount.h>
33 #include <sys/zap_impl.h>
34 #include <sys/zap_leaf.h>
38 #include <sys/sunddi.h>
41 static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx);
45 zap_hash(zap_t *zap, const char *normname)
49 uint64_t crc = zap->zap_salt;
51 /* NB: name must already be normalized, if necessary */
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];
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
65 crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1);
71 zap_normalize(zap_t *zap, const char *name, char *namenorm)
76 inlen = strlen(name) + 1;
77 outlen = ZAP_MAXNAMELEN;
80 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
81 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL, U8_UNICODE_LATEST,
88 zap_match(zap_name_t *zn, const char *matchname)
90 if (zn->zn_matchtype == MT_FIRST) {
91 char norm[ZAP_MAXNAMELEN];
93 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
96 return (strcmp(zn->zn_name_norm, norm) == 0);
98 /* MT_BEST or MT_EXACT */
99 return (strcmp(zn->zn_name_orij, matchname) == 0);
104 zap_name_free(zap_name_t *zn)
106 kmem_free(zn, sizeof (zap_name_t));
109 /* XXX combine this with zap_lockdir()? */
111 zap_name_alloc(zap_t *zap, const char *name, matchtype_t mt)
113 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
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) {
123 zn->zn_name_norm = zn->zn_normbuf;
125 if (mt != MT_EXACT) {
129 zn->zn_name_norm = zn->zn_name_orij;
132 zn->zn_hash = zap_hash(zap, zn->zn_name_norm);
137 mzap_byteswap(mzap_phys_t *buf, size_t size)
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);
153 zap_byteswap(void *buf, size_t size)
157 block_type = *(uint64_t *)buf;
159 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
160 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
161 mzap_byteswap(buf, size);
163 fzap_byteswap(buf, size);
168 mze_compare(const void *arg1, const void *arg2)
170 const mzap_ent_t *mze1 = arg1;
171 const mzap_ent_t *mze2 = arg2;
173 if (mze1->mze_hash > mze2->mze_hash)
175 if (mze1->mze_hash < mze2->mze_hash)
177 if (mze1->mze_phys.mze_cd > mze2->mze_phys.mze_cd)
179 if (mze1->mze_phys.mze_cd < mze2->mze_phys.mze_cd)
185 mze_insert(zap_t *zap, int chunkid, uint64_t hash, mzap_ent_phys_t *mzep)
189 ASSERT(zap->zap_ismicro);
190 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
191 ASSERT(mzep->mze_cd < ZAP_MAXCD);
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);
201 mze_find(zap_name_t *zn)
203 mzap_ent_t mze_tofind;
206 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
208 ASSERT(zn->zn_zap->zap_ismicro);
209 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
211 if (strlen(zn->zn_name_norm) >= sizeof (mze_tofind.mze_phys.mze_name))
214 mze_tofind.mze_hash = zn->zn_hash;
215 mze_tofind.mze_phys.mze_cd = 0;
218 mze = avl_find(avl, &mze_tofind, &idx);
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))
225 if (zn->zn_matchtype == MT_BEST) {
226 zn->zn_matchtype = MT_FIRST;
233 mze_find_unused_cd(zap_t *zap, uint64_t hash)
235 mzap_ent_t mze_tofind;
238 avl_tree_t *avl = &zap->zap_m.zap_avl;
241 ASSERT(zap->zap_ismicro);
242 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
244 mze_tofind.mze_hash = hash;
245 mze_tofind.mze_phys.mze_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)
259 mze_remove(zap_t *zap, mzap_ent_t *mze)
261 ASSERT(zap->zap_ismicro);
262 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
264 avl_remove(&zap->zap_m.zap_avl, mze);
265 kmem_free(mze, sizeof (mzap_ent_t));
269 mze_destroy(zap_t *zap)
272 void *avlcookie = NULL;
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);
280 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
286 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
288 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
289 rw_init(&zap->zap_rwlock, NULL, RW_DEFAULT, 0);
290 rw_enter(&zap->zap_rwlock, RW_WRITER);
291 zap->zap_objset = os;
292 zap->zap_object = obj;
295 if (*(uint64_t *)db->db_data != ZBT_MICRO) {
296 mutex_init(&zap->zap_f.zap_num_entries_mtx, NULL,
298 zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
300 zap->zap_ismicro = TRUE;
304 * Make sure that zap_ismicro is set before we let others see
305 * it, because zap_lockdir() checks zap_ismicro without the lock
308 winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict);
310 if (winner != NULL) {
311 rw_exit(&zap->zap_rwlock);
312 rw_destroy(&zap->zap_rwlock);
313 if (!zap->zap_ismicro)
314 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
315 kmem_free(zap, sizeof (zap_t));
319 if (zap->zap_ismicro) {
320 zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
321 zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags;
322 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
323 avl_create(&zap->zap_m.zap_avl, mze_compare,
324 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
326 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
327 mzap_ent_phys_t *mze =
328 &zap->zap_m.zap_phys->mz_chunk[i];
329 if (mze->mze_name[0]) {
332 zap->zap_m.zap_num_entries++;
333 zn = zap_name_alloc(zap, mze->mze_name,
335 mze_insert(zap, i, zn->zn_hash, mze);
340 zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
341 zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags;
343 ASSERT3U(sizeof (struct zap_leaf_header), ==,
344 2*ZAP_LEAF_CHUNKSIZE);
347 * The embedded pointer table should not overlap the
350 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
351 &zap->zap_f.zap_phys->zap_salt);
354 * The embedded pointer table should end at the end of
357 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
358 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
359 (uintptr_t)zap->zap_f.zap_phys, ==,
360 zap->zap_dbuf->db_size);
362 rw_exit(&zap->zap_rwlock);
367 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
368 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
377 err = dmu_buf_hold(os, obj, 0, NULL, &db);
383 dmu_object_info_t doi;
384 dmu_object_info_from_db(db, &doi);
385 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
389 zap = dmu_buf_get_user(db);
391 zap = mzap_open(os, obj, db);
394 * We're checking zap_ismicro without the lock held, in order to
395 * tell what type of lock we want. Once we have some sort of
396 * lock, see if it really is the right type. In practice this
397 * can only be different if it was upgraded from micro to fat,
398 * and micro wanted WRITER but fat only needs READER.
400 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
401 rw_enter(&zap->zap_rwlock, lt);
402 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
403 /* it was upgraded, now we only need reader */
404 ASSERT(lt == RW_WRITER);
406 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
407 rw_downgrade(&zap->zap_rwlock);
411 zap->zap_objset = os;
414 dmu_buf_will_dirty(db, tx);
416 ASSERT3P(zap->zap_dbuf, ==, db);
418 ASSERT(!zap->zap_ismicro ||
419 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
420 if (zap->zap_ismicro && tx && adding &&
421 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
422 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
423 if (newsz > MZAP_MAX_BLKSZ) {
424 dprintf("upgrading obj %llu: num_entries=%u\n",
425 obj, zap->zap_m.zap_num_entries);
427 return (mzap_upgrade(zapp, tx));
429 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
430 ASSERT3U(err, ==, 0);
431 zap->zap_m.zap_num_chunks =
432 db->db_size / MZAP_ENT_LEN - 1;
440 zap_unlockdir(zap_t *zap)
442 rw_exit(&zap->zap_rwlock);
443 dmu_buf_rele(zap->zap_dbuf, NULL);
447 mzap_upgrade(zap_t **zapp, dmu_tx_t *tx)
450 int i, sz, nchunks, err;
453 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
455 sz = zap->zap_dbuf->db_size;
456 mzp = kmem_alloc(sz, KM_SLEEP);
457 bcopy(zap->zap_dbuf->db_data, mzp, sz);
458 nchunks = zap->zap_m.zap_num_chunks;
460 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
461 1ULL << fzap_default_block_shift, 0, tx);
467 dprintf("upgrading obj=%llu with %u chunks\n",
468 zap->zap_object, nchunks);
469 /* XXX destroy the avl later, so we can use the stored hash value */
472 fzap_upgrade(zap, tx);
474 for (i = 0; i < nchunks; i++) {
476 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
478 if (mze->mze_name[0] == 0)
480 dprintf("adding %s=%llu\n",
481 mze->mze_name, mze->mze_value);
482 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
483 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
484 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
495 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, dmu_tx_t *tx)
500 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db));
504 dmu_object_info_t doi;
505 dmu_object_info_from_db(db, &doi);
506 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
510 dmu_buf_will_dirty(db, tx);
512 zp->mz_block_type = ZBT_MICRO;
513 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
514 zp->mz_normflags = normflags;
515 dmu_buf_rele(db, FTAG);
519 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
520 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
522 return (zap_create_claim_norm(os, obj,
523 0, ot, bonustype, bonuslen, tx));
527 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
528 dmu_object_type_t ot,
529 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
533 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
536 mzap_create_impl(os, obj, normflags, tx);
541 zap_create(objset_t *os, dmu_object_type_t ot,
542 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
544 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
548 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
549 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
551 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
553 mzap_create_impl(os, obj, normflags, tx);
558 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
561 * dmu_object_free will free the object number and free the
562 * data. Freeing the data will cause our pageout function to be
563 * called, which will destroy our data (zap_leaf_t's and zap_t).
566 return (dmu_object_free(os, zapobj, tx));
571 zap_evict(dmu_buf_t *db, void *vzap)
575 rw_destroy(&zap->zap_rwlock);
577 if (zap->zap_ismicro)
580 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
582 kmem_free(zap, sizeof (zap_t));
586 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
591 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
594 if (!zap->zap_ismicro) {
595 err = fzap_count(zap, count);
597 *count = zap->zap_m.zap_num_entries;
604 * zn may be NULL; if not specified, it will be computed if needed.
605 * See also the comment above zap_entry_normalization_conflict().
608 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
611 int direction = AVL_BEFORE;
612 boolean_t allocdzn = B_FALSE;
614 if (zap->zap_normflags == 0)
618 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
619 other && other->mze_hash == mze->mze_hash;
620 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
623 zn = zap_name_alloc(zap, mze->mze_phys.mze_name,
627 if (zap_match(zn, other->mze_phys.mze_name)) {
634 if (direction == AVL_BEFORE) {
635 direction = AVL_AFTER;
645 * Routines for manipulating attributes.
649 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
650 uint64_t integer_size, uint64_t num_integers, void *buf)
652 return (zap_lookup_norm(os, zapobj, name, integer_size,
653 num_integers, buf, MT_EXACT, NULL, 0, NULL));
657 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
658 uint64_t integer_size, uint64_t num_integers, void *buf,
659 matchtype_t mt, char *realname, int rn_len,
667 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
670 zn = zap_name_alloc(zap, name, mt);
676 if (!zap->zap_ismicro) {
677 err = fzap_lookup(zn, integer_size, num_integers, buf,
678 realname, rn_len, ncp);
684 if (num_integers < 1) {
686 } else if (integer_size != 8) {
689 *(uint64_t *)buf = mze->mze_phys.mze_value;
690 (void) strlcpy(realname,
691 mze->mze_phys.mze_name, rn_len);
693 *ncp = mzap_normalization_conflict(zap,
705 zap_length(objset_t *os, uint64_t zapobj, const char *name,
706 uint64_t *integer_size, uint64_t *num_integers)
713 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
716 zn = zap_name_alloc(zap, name, MT_EXACT);
721 if (!zap->zap_ismicro) {
722 err = fzap_length(zn, integer_size, num_integers);
740 mzap_addent(zap_name_t *zn, uint64_t value)
743 zap_t *zap = zn->zn_zap;
744 int start = zap->zap_m.zap_alloc_next;
747 dprintf("obj=%llu %s=%llu\n", zap->zap_object,
748 zn->zn_name_orij, value);
749 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
752 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
753 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
754 ASSERT(strcmp(zn->zn_name_orij, mze->mze_name) != 0);
758 cd = mze_find_unused_cd(zap, zn->zn_hash);
759 /* given the limited size of the microzap, this can't happen */
760 ASSERT(cd != ZAP_MAXCD);
763 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
764 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
765 if (mze->mze_name[0] == 0) {
766 mze->mze_value = value;
768 (void) strcpy(mze->mze_name, zn->zn_name_orij);
769 zap->zap_m.zap_num_entries++;
770 zap->zap_m.zap_alloc_next = i+1;
771 if (zap->zap_m.zap_alloc_next ==
772 zap->zap_m.zap_num_chunks)
773 zap->zap_m.zap_alloc_next = 0;
774 mze_insert(zap, i, zn->zn_hash, mze);
782 ASSERT(!"out of entries!");
786 zap_add(objset_t *os, uint64_t zapobj, const char *name,
787 int integer_size, uint64_t num_integers,
788 const void *val, dmu_tx_t *tx)
793 const uint64_t *intval = val;
796 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
799 zn = zap_name_alloc(zap, name, MT_EXACT);
804 if (!zap->zap_ismicro) {
805 err = fzap_add(zn, integer_size, num_integers, val, tx);
806 zap = zn->zn_zap; /* fzap_add() may change zap */
807 } else if (integer_size != 8 || num_integers != 1 ||
808 strlen(name) >= MZAP_NAME_LEN) {
809 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
810 zapobj, integer_size, num_integers, name);
811 err = mzap_upgrade(&zn->zn_zap, tx);
813 err = fzap_add(zn, integer_size, num_integers, val, tx);
814 zap = zn->zn_zap; /* fzap_add() may change zap */
820 mzap_addent(zn, *intval);
823 ASSERT(zap == zn->zn_zap);
825 if (zap != NULL) /* may be NULL if fzap_add() failed */
831 zap_update(objset_t *os, uint64_t zapobj, const char *name,
832 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
836 const uint64_t *intval = val;
840 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
843 zn = zap_name_alloc(zap, name, MT_EXACT);
848 if (!zap->zap_ismicro) {
849 err = fzap_update(zn, integer_size, num_integers, val, tx);
850 zap = zn->zn_zap; /* fzap_update() may change zap */
851 } else if (integer_size != 8 || num_integers != 1 ||
852 strlen(name) >= MZAP_NAME_LEN) {
853 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
854 zapobj, integer_size, num_integers, name);
855 err = mzap_upgrade(&zn->zn_zap, tx);
857 err = fzap_update(zn, integer_size, num_integers,
859 zap = zn->zn_zap; /* fzap_update() may change zap */
863 mze->mze_phys.mze_value = *intval;
864 zap->zap_m.zap_phys->mz_chunk
865 [mze->mze_chunkid].mze_value = *intval;
867 mzap_addent(zn, *intval);
870 ASSERT(zap == zn->zn_zap);
872 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
878 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
880 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
884 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
885 matchtype_t mt, dmu_tx_t *tx)
892 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
895 zn = zap_name_alloc(zap, name, mt);
900 if (!zap->zap_ismicro) {
901 err = fzap_remove(zn, tx);
907 zap->zap_m.zap_num_entries--;
908 bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
909 sizeof (mzap_ent_phys_t));
910 mze_remove(zap, mze);
919 * Routines for iterating over the attributes.
923 * We want to keep the high 32 bits of the cursor zero if we can, so
924 * that 32-bit programs can access this. So use a small hash value so
925 * we can fit 4 bits of cd into the 32-bit cursor.
927 * [ 4 zero bits | 32-bit collision differentiator | 28-bit hash value ]
930 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
936 zc->zc_zapobj = zapobj;
937 if (serialized == -1ULL) {
941 zc->zc_hash = serialized << (64-ZAP_HASHBITS);
942 zc->zc_cd = serialized >> ZAP_HASHBITS;
943 if (zc->zc_cd >= ZAP_MAXCD) /* corrupt serialized */
949 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
951 zap_cursor_init_serialized(zc, os, zapobj, 0);
955 zap_cursor_fini(zap_cursor_t *zc)
958 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
959 zap_unlockdir(zc->zc_zap);
963 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
964 zap_put_leaf(zc->zc_leaf);
967 zc->zc_objset = NULL;
971 zap_cursor_serialize(zap_cursor_t *zc)
973 if (zc->zc_hash == -1ULL)
975 ASSERT((zc->zc_hash & (ZAP_MAXCD-1)) == 0);
976 ASSERT(zc->zc_cd < ZAP_MAXCD);
977 return ((zc->zc_hash >> (64-ZAP_HASHBITS)) |
978 ((uint64_t)zc->zc_cd << ZAP_HASHBITS));
982 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
986 mzap_ent_t mze_tofind;
989 if (zc->zc_hash == -1ULL)
992 if (zc->zc_zap == NULL) {
993 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
994 RW_READER, TRUE, FALSE, &zc->zc_zap);
998 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1000 if (!zc->zc_zap->zap_ismicro) {
1001 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1005 mze_tofind.mze_hash = zc->zc_hash;
1006 mze_tofind.mze_phys.mze_cd = zc->zc_cd;
1008 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1010 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1014 ASSERT(0 == bcmp(&mze->mze_phys,
1015 &zc->zc_zap->zap_m.zap_phys->mz_chunk
1016 [mze->mze_chunkid], sizeof (mze->mze_phys)));
1018 za->za_normalization_conflict =
1019 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1020 za->za_integer_length = 8;
1021 za->za_num_integers = 1;
1022 za->za_first_integer = mze->mze_phys.mze_value;
1023 (void) strcpy(za->za_name, mze->mze_phys.mze_name);
1024 zc->zc_hash = mze->mze_hash;
1025 zc->zc_cd = mze->mze_phys.mze_cd;
1028 zc->zc_hash = -1ULL;
1031 rw_exit(&zc->zc_zap->zap_rwlock);
1036 zap_cursor_advance(zap_cursor_t *zc)
1038 if (zc->zc_hash == -1ULL)
1041 if (zc->zc_cd >= ZAP_MAXCD) {
1043 zc->zc_hash += 1ULL<<(64-ZAP_HASHBITS);
1044 if (zc->zc_hash == 0) /* EOF */
1045 zc->zc_hash = -1ULL;
1050 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1055 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1059 bzero(zs, sizeof (zap_stats_t));
1061 if (zap->zap_ismicro) {
1062 zs->zs_blocksize = zap->zap_dbuf->db_size;
1063 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1064 zs->zs_num_blocks = 1;
1066 fzap_get_stats(zap, zs);