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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
28 #include <sys/zfs_context.h>
30 #include <sys/refcount.h>
31 #include <sys/zap_impl.h>
32 #include <sys/zap_leaf.h>
36 #include <sys/sunddi.h>
39 static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx);
43 zap_hash(zap_t *zap, const char *normname)
47 uint64_t crc = zap->zap_salt;
49 /* NB: name must already be normalized, if necessary */
52 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
53 for (cp = (const uint8_t *)normname; (c = *cp) != '\0'; cp++) {
54 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ c) & 0xFF];
58 * Only use 28 bits, since we need 4 bits in the cookie for the
59 * collision differentiator. We MUST use the high bits, since
60 * those are the ones that we first pay attention to when
63 crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1);
69 zap_normalize(zap_t *zap, const char *name, char *namenorm)
74 inlen = strlen(name) + 1;
75 outlen = ZAP_MAXNAMELEN;
78 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
79 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
80 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
86 zap_match(zap_name_t *zn, const char *matchname)
88 if (zn->zn_matchtype == MT_FIRST) {
89 char norm[ZAP_MAXNAMELEN];
91 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
94 return (strcmp(zn->zn_name_norm, norm) == 0);
96 /* MT_BEST or MT_EXACT */
97 return (strcmp(zn->zn_name_orij, matchname) == 0);
102 zap_name_free(zap_name_t *zn)
104 kmem_free(zn, sizeof (zap_name_t));
107 /* XXX combine this with zap_lockdir()? */
109 zap_name_alloc(zap_t *zap, const char *name, matchtype_t mt)
111 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
114 zn->zn_name_orij = name;
115 zn->zn_matchtype = mt;
116 if (zap->zap_normflags) {
117 if (zap_normalize(zap, name, zn->zn_normbuf) != 0) {
121 zn->zn_name_norm = zn->zn_normbuf;
123 if (mt != MT_EXACT) {
127 zn->zn_name_norm = zn->zn_name_orij;
130 zn->zn_hash = zap_hash(zap, zn->zn_name_norm);
135 mzap_byteswap(mzap_phys_t *buf, size_t size)
138 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
139 buf->mz_salt = BSWAP_64(buf->mz_salt);
140 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
141 max = (size / MZAP_ENT_LEN) - 1;
142 for (i = 0; i < max; i++) {
143 buf->mz_chunk[i].mze_value =
144 BSWAP_64(buf->mz_chunk[i].mze_value);
145 buf->mz_chunk[i].mze_cd =
146 BSWAP_32(buf->mz_chunk[i].mze_cd);
151 zap_byteswap(void *buf, size_t size)
155 block_type = *(uint64_t *)buf;
157 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
158 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
159 mzap_byteswap(buf, size);
161 fzap_byteswap(buf, size);
166 mze_compare(const void *arg1, const void *arg2)
168 const mzap_ent_t *mze1 = arg1;
169 const mzap_ent_t *mze2 = arg2;
171 if (mze1->mze_hash > mze2->mze_hash)
173 if (mze1->mze_hash < mze2->mze_hash)
175 if (mze1->mze_phys.mze_cd > mze2->mze_phys.mze_cd)
177 if (mze1->mze_phys.mze_cd < mze2->mze_phys.mze_cd)
183 mze_insert(zap_t *zap, int chunkid, uint64_t hash, mzap_ent_phys_t *mzep)
187 ASSERT(zap->zap_ismicro);
188 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
189 ASSERT(mzep->mze_cd < ZAP_MAXCD);
191 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
192 mze->mze_chunkid = chunkid;
193 mze->mze_hash = hash;
194 mze->mze_phys = *mzep;
195 avl_add(&zap->zap_m.zap_avl, mze);
199 mze_find(zap_name_t *zn)
201 mzap_ent_t mze_tofind;
204 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
206 ASSERT(zn->zn_zap->zap_ismicro);
207 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
209 if (strlen(zn->zn_name_norm) >= sizeof (mze_tofind.mze_phys.mze_name))
212 mze_tofind.mze_hash = zn->zn_hash;
213 mze_tofind.mze_phys.mze_cd = 0;
216 mze = avl_find(avl, &mze_tofind, &idx);
218 mze = avl_nearest(avl, idx, AVL_AFTER);
219 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
220 if (zap_match(zn, mze->mze_phys.mze_name))
223 if (zn->zn_matchtype == MT_BEST) {
224 zn->zn_matchtype = MT_FIRST;
231 mze_find_unused_cd(zap_t *zap, uint64_t hash)
233 mzap_ent_t mze_tofind;
236 avl_tree_t *avl = &zap->zap_m.zap_avl;
239 ASSERT(zap->zap_ismicro);
240 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
242 mze_tofind.mze_hash = hash;
243 mze_tofind.mze_phys.mze_cd = 0;
246 for (mze = avl_find(avl, &mze_tofind, &idx);
247 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
248 if (mze->mze_phys.mze_cd != cd)
257 mze_remove(zap_t *zap, mzap_ent_t *mze)
259 ASSERT(zap->zap_ismicro);
260 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
262 avl_remove(&zap->zap_m.zap_avl, mze);
263 kmem_free(mze, sizeof (mzap_ent_t));
267 mze_destroy(zap_t *zap)
270 void *avlcookie = NULL;
272 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
273 kmem_free(mze, sizeof (mzap_ent_t));
274 avl_destroy(&zap->zap_m.zap_avl);
278 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
284 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
286 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
287 rw_init(&zap->zap_rwlock, 0, 0, 0);
288 rw_enter(&zap->zap_rwlock, RW_WRITER);
289 zap->zap_objset = os;
290 zap->zap_object = obj;
293 if (*(uint64_t *)db->db_data != ZBT_MICRO) {
294 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
295 zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
297 zap->zap_ismicro = TRUE;
301 * Make sure that zap_ismicro is set before we let others see
302 * it, because zap_lockdir() checks zap_ismicro without the lock
305 winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict);
307 if (winner != NULL) {
308 rw_exit(&zap->zap_rwlock);
309 rw_destroy(&zap->zap_rwlock);
310 if (!zap->zap_ismicro)
311 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
312 kmem_free(zap, sizeof (zap_t));
316 if (zap->zap_ismicro) {
317 zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
318 zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags;
319 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
320 avl_create(&zap->zap_m.zap_avl, mze_compare,
321 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
323 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
324 mzap_ent_phys_t *mze =
325 &zap->zap_m.zap_phys->mz_chunk[i];
326 if (mze->mze_name[0]) {
329 zap->zap_m.zap_num_entries++;
330 zn = zap_name_alloc(zap, mze->mze_name,
332 mze_insert(zap, i, zn->zn_hash, mze);
337 zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
338 zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags;
340 ASSERT3U(sizeof (struct zap_leaf_header), ==,
341 2*ZAP_LEAF_CHUNKSIZE);
344 * The embedded pointer table should not overlap the
347 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
348 &zap->zap_f.zap_phys->zap_salt);
351 * The embedded pointer table should end at the end of
354 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
355 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
356 (uintptr_t)zap->zap_f.zap_phys, ==,
357 zap->zap_dbuf->db_size);
359 rw_exit(&zap->zap_rwlock);
364 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
365 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
374 err = dmu_buf_hold(os, obj, 0, NULL, &db);
380 dmu_object_info_t doi;
381 dmu_object_info_from_db(db, &doi);
382 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
386 zap = dmu_buf_get_user(db);
388 zap = mzap_open(os, obj, db);
391 * We're checking zap_ismicro without the lock held, in order to
392 * tell what type of lock we want. Once we have some sort of
393 * lock, see if it really is the right type. In practice this
394 * can only be different if it was upgraded from micro to fat,
395 * and micro wanted WRITER but fat only needs READER.
397 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
398 rw_enter(&zap->zap_rwlock, lt);
399 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
400 /* it was upgraded, now we only need reader */
401 ASSERT(lt == RW_WRITER);
403 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
404 rw_downgrade(&zap->zap_rwlock);
408 zap->zap_objset = os;
411 dmu_buf_will_dirty(db, tx);
413 ASSERT3P(zap->zap_dbuf, ==, db);
415 ASSERT(!zap->zap_ismicro ||
416 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
417 if (zap->zap_ismicro && tx && adding &&
418 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
419 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
420 if (newsz > MZAP_MAX_BLKSZ) {
421 dprintf("upgrading obj %llu: num_entries=%u\n",
422 obj, zap->zap_m.zap_num_entries);
424 return (mzap_upgrade(zapp, tx));
426 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
427 ASSERT3U(err, ==, 0);
428 zap->zap_m.zap_num_chunks =
429 db->db_size / MZAP_ENT_LEN - 1;
437 zap_unlockdir(zap_t *zap)
439 rw_exit(&zap->zap_rwlock);
440 dmu_buf_rele(zap->zap_dbuf, NULL);
444 mzap_upgrade(zap_t **zapp, dmu_tx_t *tx)
447 int i, sz, nchunks, err;
450 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
452 sz = zap->zap_dbuf->db_size;
453 mzp = kmem_alloc(sz, KM_SLEEP);
454 bcopy(zap->zap_dbuf->db_data, mzp, sz);
455 nchunks = zap->zap_m.zap_num_chunks;
457 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
458 1ULL << fzap_default_block_shift, 0, tx);
464 dprintf("upgrading obj=%llu with %u chunks\n",
465 zap->zap_object, nchunks);
466 /* XXX destroy the avl later, so we can use the stored hash value */
469 fzap_upgrade(zap, tx);
471 for (i = 0; i < nchunks; i++) {
473 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
475 if (mze->mze_name[0] == 0)
477 dprintf("adding %s=%llu\n",
478 mze->mze_name, mze->mze_value);
479 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
480 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
481 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
492 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, dmu_tx_t *tx)
497 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db));
501 dmu_object_info_t doi;
502 dmu_object_info_from_db(db, &doi);
503 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
507 dmu_buf_will_dirty(db, tx);
509 zp->mz_block_type = ZBT_MICRO;
510 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
511 zp->mz_normflags = normflags;
512 dmu_buf_rele(db, FTAG);
516 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
517 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
519 return (zap_create_claim_norm(os, obj,
520 0, ot, bonustype, bonuslen, tx));
524 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
525 dmu_object_type_t ot,
526 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
530 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
533 mzap_create_impl(os, obj, normflags, tx);
538 zap_create(objset_t *os, dmu_object_type_t ot,
539 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
541 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
545 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
546 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
548 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
550 mzap_create_impl(os, obj, normflags, tx);
555 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
558 * dmu_object_free will free the object number and free the
559 * data. Freeing the data will cause our pageout function to be
560 * called, which will destroy our data (zap_leaf_t's and zap_t).
563 return (dmu_object_free(os, zapobj, tx));
568 zap_evict(dmu_buf_t *db, void *vzap)
572 rw_destroy(&zap->zap_rwlock);
574 if (zap->zap_ismicro)
577 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
579 kmem_free(zap, sizeof (zap_t));
583 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
588 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
591 if (!zap->zap_ismicro) {
592 err = fzap_count(zap, count);
594 *count = zap->zap_m.zap_num_entries;
601 * zn may be NULL; if not specified, it will be computed if needed.
602 * See also the comment above zap_entry_normalization_conflict().
605 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
608 int direction = AVL_BEFORE;
609 boolean_t allocdzn = B_FALSE;
611 if (zap->zap_normflags == 0)
615 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
616 other && other->mze_hash == mze->mze_hash;
617 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
620 zn = zap_name_alloc(zap, mze->mze_phys.mze_name,
624 if (zap_match(zn, other->mze_phys.mze_name)) {
631 if (direction == AVL_BEFORE) {
632 direction = AVL_AFTER;
642 * Routines for manipulating attributes.
646 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
647 uint64_t integer_size, uint64_t num_integers, void *buf)
649 return (zap_lookup_norm(os, zapobj, name, integer_size,
650 num_integers, buf, MT_EXACT, NULL, 0, NULL));
654 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
655 uint64_t integer_size, uint64_t num_integers, void *buf,
656 matchtype_t mt, char *realname, int rn_len,
664 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
667 zn = zap_name_alloc(zap, name, mt);
673 if (!zap->zap_ismicro) {
674 err = fzap_lookup(zn, integer_size, num_integers, buf,
675 realname, rn_len, ncp);
681 if (num_integers < 1) {
683 } else if (integer_size != 8) {
686 *(uint64_t *)buf = mze->mze_phys.mze_value;
687 (void) strlcpy(realname,
688 mze->mze_phys.mze_name, rn_len);
690 *ncp = mzap_normalization_conflict(zap,
702 zap_length(objset_t *os, uint64_t zapobj, const char *name,
703 uint64_t *integer_size, uint64_t *num_integers)
710 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
713 zn = zap_name_alloc(zap, name, MT_EXACT);
718 if (!zap->zap_ismicro) {
719 err = fzap_length(zn, integer_size, num_integers);
737 mzap_addent(zap_name_t *zn, uint64_t value)
740 zap_t *zap = zn->zn_zap;
741 int start = zap->zap_m.zap_alloc_next;
744 dprintf("obj=%llu %s=%llu\n", zap->zap_object,
745 zn->zn_name_orij, value);
746 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
749 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
750 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
751 ASSERT(strcmp(zn->zn_name_orij, mze->mze_name) != 0);
755 cd = mze_find_unused_cd(zap, zn->zn_hash);
756 /* given the limited size of the microzap, this can't happen */
757 ASSERT(cd != ZAP_MAXCD);
760 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
761 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
762 if (mze->mze_name[0] == 0) {
763 mze->mze_value = value;
765 (void) strcpy(mze->mze_name, zn->zn_name_orij);
766 zap->zap_m.zap_num_entries++;
767 zap->zap_m.zap_alloc_next = i+1;
768 if (zap->zap_m.zap_alloc_next ==
769 zap->zap_m.zap_num_chunks)
770 zap->zap_m.zap_alloc_next = 0;
771 mze_insert(zap, i, zn->zn_hash, mze);
779 ASSERT(!"out of entries!");
783 zap_add(objset_t *os, uint64_t zapobj, const char *name,
784 int integer_size, uint64_t num_integers,
785 const void *val, dmu_tx_t *tx)
790 const uint64_t *intval = val;
793 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
796 zn = zap_name_alloc(zap, name, MT_EXACT);
801 if (!zap->zap_ismicro) {
802 err = fzap_add(zn, integer_size, num_integers, val, tx);
803 zap = zn->zn_zap; /* fzap_add() may change zap */
804 } else if (integer_size != 8 || num_integers != 1 ||
805 strlen(name) >= MZAP_NAME_LEN) {
806 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
807 zapobj, integer_size, num_integers, name);
808 err = mzap_upgrade(&zn->zn_zap, tx);
810 err = fzap_add(zn, integer_size, num_integers, val, tx);
811 zap = zn->zn_zap; /* fzap_add() may change zap */
817 mzap_addent(zn, *intval);
820 ASSERT(zap == zn->zn_zap);
822 if (zap != NULL) /* may be NULL if fzap_add() failed */
828 zap_update(objset_t *os, uint64_t zapobj, const char *name,
829 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
833 const uint64_t *intval = val;
837 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
840 zn = zap_name_alloc(zap, name, MT_EXACT);
845 if (!zap->zap_ismicro) {
846 err = fzap_update(zn, integer_size, num_integers, val, tx);
847 zap = zn->zn_zap; /* fzap_update() may change zap */
848 } else if (integer_size != 8 || num_integers != 1 ||
849 strlen(name) >= MZAP_NAME_LEN) {
850 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
851 zapobj, integer_size, num_integers, name);
852 err = mzap_upgrade(&zn->zn_zap, tx);
854 err = fzap_update(zn, integer_size, num_integers,
856 zap = zn->zn_zap; /* fzap_update() may change zap */
860 mze->mze_phys.mze_value = *intval;
861 zap->zap_m.zap_phys->mz_chunk
862 [mze->mze_chunkid].mze_value = *intval;
864 mzap_addent(zn, *intval);
867 ASSERT(zap == zn->zn_zap);
869 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
875 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
877 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
881 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
882 matchtype_t mt, dmu_tx_t *tx)
889 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
892 zn = zap_name_alloc(zap, name, mt);
897 if (!zap->zap_ismicro) {
898 err = fzap_remove(zn, tx);
904 zap->zap_m.zap_num_entries--;
905 bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
906 sizeof (mzap_ent_phys_t));
907 mze_remove(zap, mze);
916 * Routines for iterating over the attributes.
920 * We want to keep the high 32 bits of the cursor zero if we can, so
921 * that 32-bit programs can access this. So use a small hash value so
922 * we can fit 4 bits of cd into the 32-bit cursor.
924 * [ 4 zero bits | 32-bit collision differentiator | 28-bit hash value ]
927 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
933 zc->zc_zapobj = zapobj;
934 if (serialized == -1ULL) {
938 zc->zc_hash = serialized << (64-ZAP_HASHBITS);
939 zc->zc_cd = serialized >> ZAP_HASHBITS;
940 if (zc->zc_cd >= ZAP_MAXCD) /* corrupt serialized */
946 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
948 zap_cursor_init_serialized(zc, os, zapobj, 0);
952 zap_cursor_fini(zap_cursor_t *zc)
955 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
956 zap_unlockdir(zc->zc_zap);
960 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
961 zap_put_leaf(zc->zc_leaf);
964 zc->zc_objset = NULL;
968 zap_cursor_serialize(zap_cursor_t *zc)
970 if (zc->zc_hash == -1ULL)
972 ASSERT((zc->zc_hash & (ZAP_MAXCD-1)) == 0);
973 ASSERT(zc->zc_cd < ZAP_MAXCD);
974 return ((zc->zc_hash >> (64-ZAP_HASHBITS)) |
975 ((uint64_t)zc->zc_cd << ZAP_HASHBITS));
979 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
983 mzap_ent_t mze_tofind;
986 if (zc->zc_hash == -1ULL)
989 if (zc->zc_zap == NULL) {
990 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
991 RW_READER, TRUE, FALSE, &zc->zc_zap);
995 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
997 if (!zc->zc_zap->zap_ismicro) {
998 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1002 mze_tofind.mze_hash = zc->zc_hash;
1003 mze_tofind.mze_phys.mze_cd = zc->zc_cd;
1005 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1007 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1011 ASSERT(0 == bcmp(&mze->mze_phys,
1012 &zc->zc_zap->zap_m.zap_phys->mz_chunk
1013 [mze->mze_chunkid], sizeof (mze->mze_phys)));
1015 za->za_normalization_conflict =
1016 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1017 za->za_integer_length = 8;
1018 za->za_num_integers = 1;
1019 za->za_first_integer = mze->mze_phys.mze_value;
1020 (void) strcpy(za->za_name, mze->mze_phys.mze_name);
1021 zc->zc_hash = mze->mze_hash;
1022 zc->zc_cd = mze->mze_phys.mze_cd;
1025 zc->zc_hash = -1ULL;
1028 rw_exit(&zc->zc_zap->zap_rwlock);
1033 zap_cursor_advance(zap_cursor_t *zc)
1035 if (zc->zc_hash == -1ULL)
1038 if (zc->zc_cd >= ZAP_MAXCD) {
1040 zc->zc_hash += 1ULL<<(64-ZAP_HASHBITS);
1041 if (zc->zc_hash == 0) /* EOF */
1042 zc->zc_hash = -1ULL;
1047 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1052 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1056 bzero(zs, sizeof (zap_stats_t));
1058 if (zap->zap_ismicro) {
1059 zs->zs_blocksize = zap->zap_dbuf->db_size;
1060 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1061 zs->zs_num_blocks = 1;
1063 fzap_get_stats(zap, zs);
1070 zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
1071 uint64_t *towrite, uint64_t *tooverwrite, uint64_t dn_datablkshift)
1078 * Since, we don't have a name, we cannot figure out which blocks will
1079 * be affected in this operation. So, account for the worst case :
1080 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1081 * - 4 new blocks written if adding:
1082 * - 2 blocks for possibly split leaves,
1083 * - 2 grown ptrtbl blocks
1085 * This also accomodates the case where an add operation to a fairly
1086 * large microzap results in a promotion to fatzap.
1089 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1094 * We lock the zap with adding == FALSE. Because, if we pass
1095 * the actual value of add, it could trigger a mzap_upgrade().
1096 * At present we are just evaluating the possibility of this operation
1097 * and hence we donot want to trigger an upgrade.
1099 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1103 if (!zap->zap_ismicro) {
1104 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1106 err = fzap_count_write(zn, add, towrite,
1111 * We treat this case as similar to (name == NULL)
1113 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1117 if (dmu_buf_freeable(zap->zap_dbuf))
1118 *tooverwrite += SPA_MAXBLOCKSIZE;
1120 *towrite += SPA_MAXBLOCKSIZE;
1123 * We are here if we are adding and (name != NULL).
1124 * It is hard to find out if this add will promote this
1125 * microzap to fatzap. Hence, we assume the worst case
1126 * and account for the blocks assuming this microzap
1127 * would be promoted to a fatzap.
1129 * 1 block overwritten : header block
1130 * 4 new blocks written : 2 new split leaf, 2 grown
1133 if (dmu_buf_freeable(zap->zap_dbuf))
1134 *tooverwrite += 1 << dn_datablkshift;
1136 *towrite += 1 << dn_datablkshift;
1137 *towrite += 4 << dn_datablkshift;