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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
28 #include <sys/zfs_context.h>
30 #include <sys/refcount.h>
31 #include <sys/zap_impl.h>
32 #include <sys/zap_leaf.h>
37 #include <sys/sunddi.h>
40 static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags);
43 zap_getflags(zap_t *zap)
47 return (zap->zap_u.zap_fat.zap_phys->zap_flags);
51 zap_hashbits(zap_t *zap)
53 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
62 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
69 zap_hash(zap_name_t *zn)
71 zap_t *zap = zn->zn_zap;
74 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
75 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
76 h = *(uint64_t *)zn->zn_key_orig;
80 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
82 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
84 const uint64_t *wp = zn->zn_key_norm;
86 ASSERT(zn->zn_key_intlen == 8);
87 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
91 for (j = 0; j < zn->zn_key_intlen; j++) {
93 zfs_crc64_table[(h ^ word) & 0xFF];
99 const uint8_t *cp = zn->zn_key_norm;
102 * We previously stored the terminating null on
103 * disk, but didn't hash it, so we need to
104 * continue to not hash it. (The
105 * zn_key_*_numints includes the terminating
106 * null for non-binary keys.)
108 len = zn->zn_key_norm_numints - 1;
110 ASSERT(zn->zn_key_intlen == 1);
111 for (i = 0; i < len; cp++, i++) {
113 zfs_crc64_table[(h ^ *cp) & 0xFF];
118 * Don't use all 64 bits, since we need some in the cookie for
119 * the collision differentiator. We MUST use the high bits,
120 * since those are the ones that we first pay attention to when
121 * chosing the bucket.
123 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
129 zap_normalize(zap_t *zap, const char *name, char *namenorm)
131 size_t inlen, outlen;
134 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
136 inlen = strlen(name) + 1;
137 outlen = ZAP_MAXNAMELEN;
140 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
141 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
142 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
148 zap_match(zap_name_t *zn, const char *matchname)
150 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
152 if (zn->zn_matchtype == MT_FIRST) {
153 char norm[ZAP_MAXNAMELEN];
155 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
158 return (strcmp(zn->zn_key_norm, norm) == 0);
160 /* MT_BEST or MT_EXACT */
161 return (strcmp(zn->zn_key_orig, matchname) == 0);
166 zap_name_free(zap_name_t *zn)
168 kmem_free(zn, sizeof (zap_name_t));
172 zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
174 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
177 zn->zn_key_intlen = sizeof (*key);
178 zn->zn_key_orig = key;
179 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
180 zn->zn_matchtype = mt;
181 if (zap->zap_normflags) {
182 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
186 zn->zn_key_norm = zn->zn_normbuf;
187 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
189 if (mt != MT_EXACT) {
193 zn->zn_key_norm = zn->zn_key_orig;
194 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
197 zn->zn_hash = zap_hash(zn);
202 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
204 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
206 ASSERT(zap->zap_normflags == 0);
208 zn->zn_key_intlen = sizeof (*key);
209 zn->zn_key_orig = zn->zn_key_norm = key;
210 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
211 zn->zn_matchtype = MT_EXACT;
213 zn->zn_hash = zap_hash(zn);
218 mzap_byteswap(mzap_phys_t *buf, size_t size)
221 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
222 buf->mz_salt = BSWAP_64(buf->mz_salt);
223 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
224 max = (size / MZAP_ENT_LEN) - 1;
225 for (i = 0; i < max; i++) {
226 buf->mz_chunk[i].mze_value =
227 BSWAP_64(buf->mz_chunk[i].mze_value);
228 buf->mz_chunk[i].mze_cd =
229 BSWAP_32(buf->mz_chunk[i].mze_cd);
234 zap_byteswap(void *buf, size_t size)
238 block_type = *(uint64_t *)buf;
240 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
241 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
242 mzap_byteswap(buf, size);
244 fzap_byteswap(buf, size);
249 mze_compare(const void *arg1, const void *arg2)
251 const mzap_ent_t *mze1 = arg1;
252 const mzap_ent_t *mze2 = arg2;
254 if (mze1->mze_hash > mze2->mze_hash)
256 if (mze1->mze_hash < mze2->mze_hash)
258 if (mze1->mze_cd > mze2->mze_cd)
260 if (mze1->mze_cd < mze2->mze_cd)
266 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
271 ASSERT(zap->zap_ismicro);
272 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
274 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
275 mze->mze_chunkid = chunkid;
276 mze->mze_hash = hash;
277 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
278 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
279 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
280 kmem_free(mze, sizeof (mzap_ent_t));
283 avl_insert(&zap->zap_m.zap_avl, mze, idx);
288 mze_find(zap_name_t *zn)
290 mzap_ent_t mze_tofind;
293 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
295 ASSERT(zn->zn_zap->zap_ismicro);
296 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
298 mze_tofind.mze_hash = zn->zn_hash;
299 mze_tofind.mze_cd = 0;
302 mze = avl_find(avl, &mze_tofind, &idx);
304 mze = avl_nearest(avl, idx, AVL_AFTER);
305 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
306 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
307 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
310 if (zn->zn_matchtype == MT_BEST) {
311 zn->zn_matchtype = MT_FIRST;
318 mze_find_unused_cd(zap_t *zap, uint64_t hash)
320 mzap_ent_t mze_tofind;
323 avl_tree_t *avl = &zap->zap_m.zap_avl;
326 ASSERT(zap->zap_ismicro);
327 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
329 mze_tofind.mze_hash = hash;
330 mze_tofind.mze_cd = 0;
333 for (mze = avl_find(avl, &mze_tofind, &idx);
334 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
335 if (mze->mze_cd != cd)
344 mze_remove(zap_t *zap, mzap_ent_t *mze)
346 ASSERT(zap->zap_ismicro);
347 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
349 avl_remove(&zap->zap_m.zap_avl, mze);
350 kmem_free(mze, sizeof (mzap_ent_t));
354 mze_destroy(zap_t *zap)
357 void *avlcookie = NULL;
359 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
360 kmem_free(mze, sizeof (mzap_ent_t));
361 avl_destroy(&zap->zap_m.zap_avl);
365 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
371 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
373 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
374 rw_init(&zap->zap_rwlock, 0, 0, 0);
375 rw_enter(&zap->zap_rwlock, RW_WRITER);
376 zap->zap_objset = os;
377 zap->zap_object = obj;
380 if (*(uint64_t *)db->db_data != ZBT_MICRO) {
381 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
382 zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
384 zap->zap_ismicro = TRUE;
388 * Make sure that zap_ismicro is set before we let others see
389 * it, because zap_lockdir() checks zap_ismicro without the lock
392 winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict);
394 if (winner != NULL) {
395 rw_exit(&zap->zap_rwlock);
396 rw_destroy(&zap->zap_rwlock);
397 if (!zap->zap_ismicro)
398 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
399 kmem_free(zap, sizeof (zap_t));
403 if (zap->zap_ismicro) {
404 zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
405 zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags;
406 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
407 avl_create(&zap->zap_m.zap_avl, mze_compare,
408 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
410 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
411 mzap_ent_phys_t *mze =
412 &zap->zap_m.zap_phys->mz_chunk[i];
413 if (mze->mze_name[0]) {
416 zn = zap_name_alloc(zap, mze->mze_name,
418 if (mze_insert(zap, i, zn->zn_hash) == 0)
419 zap->zap_m.zap_num_entries++;
421 printf("ZFS WARNING: Duplicated ZAP "
422 "entry detected (%s).\n",
429 zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
430 zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags;
432 ASSERT3U(sizeof (struct zap_leaf_header), ==,
433 2*ZAP_LEAF_CHUNKSIZE);
436 * The embedded pointer table should not overlap the
439 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
440 &zap->zap_f.zap_phys->zap_salt);
443 * The embedded pointer table should end at the end of
446 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
447 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
448 (uintptr_t)zap->zap_f.zap_phys, ==,
449 zap->zap_dbuf->db_size);
451 rw_exit(&zap->zap_rwlock);
456 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
457 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
466 err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH);
472 dmu_object_info_t doi;
473 dmu_object_info_from_db(db, &doi);
474 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
478 zap = dmu_buf_get_user(db);
480 zap = mzap_open(os, obj, db);
483 * We're checking zap_ismicro without the lock held, in order to
484 * tell what type of lock we want. Once we have some sort of
485 * lock, see if it really is the right type. In practice this
486 * can only be different if it was upgraded from micro to fat,
487 * and micro wanted WRITER but fat only needs READER.
489 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
490 rw_enter(&zap->zap_rwlock, lt);
491 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
492 /* it was upgraded, now we only need reader */
493 ASSERT(lt == RW_WRITER);
495 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
496 rw_downgrade(&zap->zap_rwlock);
500 zap->zap_objset = os;
503 dmu_buf_will_dirty(db, tx);
505 ASSERT3P(zap->zap_dbuf, ==, db);
507 ASSERT(!zap->zap_ismicro ||
508 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
509 if (zap->zap_ismicro && tx && adding &&
510 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
511 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
512 if (newsz > MZAP_MAX_BLKSZ) {
513 dprintf("upgrading obj %llu: num_entries=%u\n",
514 obj, zap->zap_m.zap_num_entries);
516 return (mzap_upgrade(zapp, tx, 0));
518 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
519 ASSERT3U(err, ==, 0);
520 zap->zap_m.zap_num_chunks =
521 db->db_size / MZAP_ENT_LEN - 1;
529 zap_unlockdir(zap_t *zap)
531 rw_exit(&zap->zap_rwlock);
532 dmu_buf_rele(zap->zap_dbuf, NULL);
536 mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
543 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
545 sz = zap->zap_dbuf->db_size;
546 mzp = kmem_alloc(sz, KM_SLEEP);
547 bcopy(zap->zap_dbuf->db_data, mzp, sz);
548 nchunks = zap->zap_m.zap_num_chunks;
551 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
552 1ULL << fzap_default_block_shift, 0, tx);
559 dprintf("upgrading obj=%llu with %u chunks\n",
560 zap->zap_object, nchunks);
561 /* XXX destroy the avl later, so we can use the stored hash value */
564 fzap_upgrade(zap, tx, flags);
566 for (i = 0; i < nchunks; i++) {
567 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
569 if (mze->mze_name[0] == 0)
571 dprintf("adding %s=%llu\n",
572 mze->mze_name, mze->mze_value);
573 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
574 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
575 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
586 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
592 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
596 dmu_object_info_t doi;
597 dmu_object_info_from_db(db, &doi);
598 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
602 dmu_buf_will_dirty(db, tx);
604 zp->mz_block_type = ZBT_MICRO;
605 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
606 zp->mz_normflags = normflags;
607 dmu_buf_rele(db, FTAG);
611 /* Only fat zap supports flags; upgrade immediately. */
612 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
613 B_FALSE, B_FALSE, &zap));
614 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags));
620 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
621 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
623 return (zap_create_claim_norm(os, obj,
624 0, ot, bonustype, bonuslen, tx));
628 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
629 dmu_object_type_t ot,
630 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
634 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
637 mzap_create_impl(os, obj, normflags, 0, tx);
642 zap_create(objset_t *os, dmu_object_type_t ot,
643 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
645 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
649 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
650 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
652 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
654 mzap_create_impl(os, obj, normflags, 0, tx);
659 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
660 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
661 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
663 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
665 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
666 leaf_blockshift <= SPA_MAXBLOCKSHIFT &&
667 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
668 indirect_blockshift <= SPA_MAXBLOCKSHIFT);
670 VERIFY(dmu_object_set_blocksize(os, obj,
671 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
673 mzap_create_impl(os, obj, normflags, flags, tx);
678 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
681 * dmu_object_free will free the object number and free the
682 * data. Freeing the data will cause our pageout function to be
683 * called, which will destroy our data (zap_leaf_t's and zap_t).
686 return (dmu_object_free(os, zapobj, tx));
691 zap_evict(dmu_buf_t *db, void *vzap)
695 rw_destroy(&zap->zap_rwlock);
697 if (zap->zap_ismicro)
700 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
702 kmem_free(zap, sizeof (zap_t));
706 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
711 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
714 if (!zap->zap_ismicro) {
715 err = fzap_count(zap, count);
717 *count = zap->zap_m.zap_num_entries;
724 * zn may be NULL; if not specified, it will be computed if needed.
725 * See also the comment above zap_entry_normalization_conflict().
728 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
731 int direction = AVL_BEFORE;
732 boolean_t allocdzn = B_FALSE;
734 if (zap->zap_normflags == 0)
738 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
739 other && other->mze_hash == mze->mze_hash;
740 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
743 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
747 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
754 if (direction == AVL_BEFORE) {
755 direction = AVL_AFTER;
765 * Routines for manipulating attributes.
769 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
770 uint64_t integer_size, uint64_t num_integers, void *buf)
772 return (zap_lookup_norm(os, zapobj, name, integer_size,
773 num_integers, buf, MT_EXACT, NULL, 0, NULL));
777 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
778 uint64_t integer_size, uint64_t num_integers, void *buf,
779 matchtype_t mt, char *realname, int rn_len,
787 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
790 zn = zap_name_alloc(zap, name, mt);
796 if (!zap->zap_ismicro) {
797 err = fzap_lookup(zn, integer_size, num_integers, buf,
798 realname, rn_len, ncp);
804 if (num_integers < 1) {
806 } else if (integer_size != 8) {
810 MZE_PHYS(zap, mze)->mze_value;
811 (void) strlcpy(realname,
812 MZE_PHYS(zap, mze)->mze_name, rn_len);
814 *ncp = mzap_normalization_conflict(zap,
826 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
833 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
836 zn = zap_name_alloc_uint64(zap, key, key_numints);
849 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
850 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
856 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
859 zn = zap_name_alloc_uint64(zap, key, key_numints);
865 err = fzap_lookup(zn, integer_size, num_integers, buf,
873 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
875 int err = (zap_lookup_norm(os, zapobj, name, 0,
876 0, NULL, MT_EXACT, NULL, 0, NULL));
877 if (err == EOVERFLOW || err == EINVAL)
878 err = 0; /* found, but skipped reading the value */
883 zap_length(objset_t *os, uint64_t zapobj, const char *name,
884 uint64_t *integer_size, uint64_t *num_integers)
891 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
894 zn = zap_name_alloc(zap, name, MT_EXACT);
899 if (!zap->zap_ismicro) {
900 err = fzap_length(zn, integer_size, num_integers);
918 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
919 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
925 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
928 zn = zap_name_alloc_uint64(zap, key, key_numints);
933 err = fzap_length(zn, integer_size, num_integers);
940 mzap_addent(zap_name_t *zn, uint64_t value)
943 zap_t *zap = zn->zn_zap;
944 int start = zap->zap_m.zap_alloc_next;
947 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
950 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
951 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
952 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
956 cd = mze_find_unused_cd(zap, zn->zn_hash);
957 /* given the limited size of the microzap, this can't happen */
958 ASSERT(cd < zap_maxcd(zap));
961 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
962 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
963 if (mze->mze_name[0] == 0) {
964 mze->mze_value = value;
966 (void) strcpy(mze->mze_name, zn->zn_key_orig);
967 zap->zap_m.zap_num_entries++;
968 zap->zap_m.zap_alloc_next = i+1;
969 if (zap->zap_m.zap_alloc_next ==
970 zap->zap_m.zap_num_chunks)
971 zap->zap_m.zap_alloc_next = 0;
972 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
980 ASSERT(!"out of entries!");
984 zap_add(objset_t *os, uint64_t zapobj, const char *key,
985 int integer_size, uint64_t num_integers,
986 const void *val, dmu_tx_t *tx)
991 const uint64_t *intval = val;
994 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
997 zn = zap_name_alloc(zap, key, MT_EXACT);
1002 if (!zap->zap_ismicro) {
1003 err = fzap_add(zn, integer_size, num_integers, val, tx);
1004 zap = zn->zn_zap; /* fzap_add() may change zap */
1005 } else if (integer_size != 8 || num_integers != 1 ||
1006 strlen(key) >= MZAP_NAME_LEN) {
1007 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1009 err = fzap_add(zn, integer_size, num_integers, val, tx);
1010 zap = zn->zn_zap; /* fzap_add() may change zap */
1016 mzap_addent(zn, *intval);
1019 ASSERT(zap == zn->zn_zap);
1021 if (zap != NULL) /* may be NULL if fzap_add() failed */
1027 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1028 int key_numints, int integer_size, uint64_t num_integers,
1029 const void *val, dmu_tx_t *tx)
1035 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1038 zn = zap_name_alloc_uint64(zap, key, key_numints);
1043 err = fzap_add(zn, integer_size, num_integers, val, tx);
1044 zap = zn->zn_zap; /* fzap_add() may change zap */
1046 if (zap != NULL) /* may be NULL if fzap_add() failed */
1052 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1053 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1058 const uint64_t *intval = val;
1064 * If there is an old value, it shouldn't change across the
1065 * lockdir (eg, due to bprewrite's xlation).
1067 if (integer_size == 8 && num_integers == 1)
1068 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1071 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1074 zn = zap_name_alloc(zap, name, MT_EXACT);
1079 if (!zap->zap_ismicro) {
1080 err = fzap_update(zn, integer_size, num_integers, val, tx);
1081 zap = zn->zn_zap; /* fzap_update() may change zap */
1082 } else if (integer_size != 8 || num_integers != 1 ||
1083 strlen(name) >= MZAP_NAME_LEN) {
1084 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1085 zapobj, integer_size, num_integers, name);
1086 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1088 err = fzap_update(zn, integer_size, num_integers,
1090 zap = zn->zn_zap; /* fzap_update() may change zap */
1094 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1095 MZE_PHYS(zap, mze)->mze_value = *intval;
1097 mzap_addent(zn, *intval);
1100 ASSERT(zap == zn->zn_zap);
1102 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1108 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1110 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1116 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1119 zn = zap_name_alloc_uint64(zap, key, key_numints);
1124 err = fzap_update(zn, integer_size, num_integers, val, tx);
1125 zap = zn->zn_zap; /* fzap_update() may change zap */
1127 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1133 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1135 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1139 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1140 matchtype_t mt, dmu_tx_t *tx)
1147 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1150 zn = zap_name_alloc(zap, name, mt);
1155 if (!zap->zap_ismicro) {
1156 err = fzap_remove(zn, tx);
1162 zap->zap_m.zap_num_entries--;
1163 bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
1164 sizeof (mzap_ent_phys_t));
1165 mze_remove(zap, mze);
1174 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1175 int key_numints, dmu_tx_t *tx)
1181 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1184 zn = zap_name_alloc_uint64(zap, key, key_numints);
1189 err = fzap_remove(zn, tx);
1196 * Routines for iterating over the attributes.
1200 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1201 uint64_t serialized)
1206 zc->zc_zapobj = zapobj;
1207 zc->zc_serialized = serialized;
1213 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1215 zap_cursor_init_serialized(zc, os, zapobj, 0);
1219 zap_cursor_fini(zap_cursor_t *zc)
1222 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1223 zap_unlockdir(zc->zc_zap);
1227 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1228 zap_put_leaf(zc->zc_leaf);
1231 zc->zc_objset = NULL;
1235 zap_cursor_serialize(zap_cursor_t *zc)
1237 if (zc->zc_hash == -1ULL)
1239 if (zc->zc_zap == NULL)
1240 return (zc->zc_serialized);
1241 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1242 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1245 * We want to keep the high 32 bits of the cursor zero if we can, so
1246 * that 32-bit programs can access this. So usually use a small
1247 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1250 * [ collision differentiator | zap_hashbits()-bit hash value ]
1252 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1253 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1257 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1261 mzap_ent_t mze_tofind;
1264 if (zc->zc_hash == -1ULL)
1267 if (zc->zc_zap == NULL) {
1269 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1270 RW_READER, TRUE, FALSE, &zc->zc_zap);
1275 * To support zap_cursor_init_serialized, advance, retrieve,
1276 * we must add to the existing zc_cd, which may already
1277 * be 1 due to the zap_cursor_advance.
1279 ASSERT(zc->zc_hash == 0);
1280 hb = zap_hashbits(zc->zc_zap);
1281 zc->zc_hash = zc->zc_serialized << (64 - hb);
1282 zc->zc_cd += zc->zc_serialized >> hb;
1283 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1286 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1288 if (!zc->zc_zap->zap_ismicro) {
1289 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1293 mze_tofind.mze_hash = zc->zc_hash;
1294 mze_tofind.mze_cd = zc->zc_cd;
1296 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1298 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1302 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1303 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1304 za->za_normalization_conflict =
1305 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1306 za->za_integer_length = 8;
1307 za->za_num_integers = 1;
1308 za->za_first_integer = mzep->mze_value;
1309 (void) strcpy(za->za_name, mzep->mze_name);
1310 zc->zc_hash = mze->mze_hash;
1311 zc->zc_cd = mze->mze_cd;
1314 zc->zc_hash = -1ULL;
1317 rw_exit(&zc->zc_zap->zap_rwlock);
1322 zap_cursor_advance(zap_cursor_t *zc)
1324 if (zc->zc_hash == -1ULL)
1330 zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1336 if (zc->zc_zap == NULL) {
1337 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1338 RW_READER, TRUE, FALSE, &zc->zc_zap);
1342 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1345 zn = zap_name_alloc(zc->zc_zap, name, mt);
1347 rw_exit(&zc->zc_zap->zap_rwlock);
1351 if (!zc->zc_zap->zap_ismicro) {
1352 err = fzap_cursor_move_to_key(zc, zn);
1359 zc->zc_hash = mze->mze_hash;
1360 zc->zc_cd = mze->mze_cd;
1365 rw_exit(&zc->zc_zap->zap_rwlock);
1370 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1375 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1379 bzero(zs, sizeof (zap_stats_t));
1381 if (zap->zap_ismicro) {
1382 zs->zs_blocksize = zap->zap_dbuf->db_size;
1383 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1384 zs->zs_num_blocks = 1;
1386 fzap_get_stats(zap, zs);
1393 zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
1394 uint64_t *towrite, uint64_t *tooverwrite)
1401 * Since, we don't have a name, we cannot figure out which blocks will
1402 * be affected in this operation. So, account for the worst case :
1403 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1404 * - 4 new blocks written if adding:
1405 * - 2 blocks for possibly split leaves,
1406 * - 2 grown ptrtbl blocks
1408 * This also accomodates the case where an add operation to a fairly
1409 * large microzap results in a promotion to fatzap.
1412 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1417 * We lock the zap with adding == FALSE. Because, if we pass
1418 * the actual value of add, it could trigger a mzap_upgrade().
1419 * At present we are just evaluating the possibility of this operation
1420 * and hence we donot want to trigger an upgrade.
1422 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1426 if (!zap->zap_ismicro) {
1427 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1429 err = fzap_count_write(zn, add, towrite,
1434 * We treat this case as similar to (name == NULL)
1436 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1440 * We are here if (name != NULL) and this is a micro-zap.
1441 * We account for the header block depending on whether it
1444 * Incase of an add-operation it is hard to find out
1445 * if this add will promote this microzap to fatzap.
1446 * Hence, we consider the worst case and account for the
1447 * blocks assuming this microzap would be promoted to a
1450 * 1 block overwritten : header block
1451 * 4 new blocks written : 2 new split leaf, 2 grown
1454 if (dmu_buf_freeable(zap->zap_dbuf))
1455 *tooverwrite += SPA_MAXBLOCKSIZE;
1457 *towrite += SPA_MAXBLOCKSIZE;
1460 *towrite += 4 * SPA_MAXBLOCKSIZE;