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.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
29 #include <sys/zfs_context.h>
31 #include <sys/refcount.h>
32 #include <sys/zap_impl.h>
33 #include <sys/zap_leaf.h>
38 #include <sys/sunddi.h>
41 static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags);
44 zap_getflags(zap_t *zap)
48 return (zap->zap_u.zap_fat.zap_phys->zap_flags);
52 zap_hashbits(zap_t *zap)
54 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
63 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
70 zap_hash(zap_name_t *zn)
72 zap_t *zap = zn->zn_zap;
75 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
76 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
77 h = *(uint64_t *)zn->zn_key_orig;
81 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
83 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
85 const uint64_t *wp = zn->zn_key_norm;
87 ASSERT(zn->zn_key_intlen == 8);
88 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
92 for (j = 0; j < zn->zn_key_intlen; j++) {
94 zfs_crc64_table[(h ^ word) & 0xFF];
100 const uint8_t *cp = zn->zn_key_norm;
103 * We previously stored the terminating null on
104 * disk, but didn't hash it, so we need to
105 * continue to not hash it. (The
106 * zn_key_*_numints includes the terminating
107 * null for non-binary keys.)
109 len = zn->zn_key_norm_numints - 1;
111 ASSERT(zn->zn_key_intlen == 1);
112 for (i = 0; i < len; cp++, i++) {
114 zfs_crc64_table[(h ^ *cp) & 0xFF];
119 * Don't use all 64 bits, since we need some in the cookie for
120 * the collision differentiator. We MUST use the high bits,
121 * since those are the ones that we first pay attention to when
122 * chosing the bucket.
124 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
130 zap_normalize(zap_t *zap, const char *name, char *namenorm)
132 size_t inlen, outlen;
135 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
137 inlen = strlen(name) + 1;
138 outlen = ZAP_MAXNAMELEN;
141 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
142 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
143 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
149 zap_match(zap_name_t *zn, const char *matchname)
151 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
153 if (zn->zn_matchtype == MT_FIRST) {
154 char norm[ZAP_MAXNAMELEN];
156 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
159 return (strcmp(zn->zn_key_norm, norm) == 0);
161 /* MT_BEST or MT_EXACT */
162 return (strcmp(zn->zn_key_orig, matchname) == 0);
167 zap_name_free(zap_name_t *zn)
169 kmem_free(zn, sizeof (zap_name_t));
173 zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
175 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
178 zn->zn_key_intlen = sizeof (*key);
179 zn->zn_key_orig = key;
180 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
181 zn->zn_matchtype = mt;
182 if (zap->zap_normflags) {
183 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
187 zn->zn_key_norm = zn->zn_normbuf;
188 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
190 if (mt != MT_EXACT) {
194 zn->zn_key_norm = zn->zn_key_orig;
195 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
198 zn->zn_hash = zap_hash(zn);
203 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
205 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
207 ASSERT(zap->zap_normflags == 0);
209 zn->zn_key_intlen = sizeof (*key);
210 zn->zn_key_orig = zn->zn_key_norm = key;
211 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
212 zn->zn_matchtype = MT_EXACT;
214 zn->zn_hash = zap_hash(zn);
219 mzap_byteswap(mzap_phys_t *buf, size_t size)
222 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
223 buf->mz_salt = BSWAP_64(buf->mz_salt);
224 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
225 max = (size / MZAP_ENT_LEN) - 1;
226 for (i = 0; i < max; i++) {
227 buf->mz_chunk[i].mze_value =
228 BSWAP_64(buf->mz_chunk[i].mze_value);
229 buf->mz_chunk[i].mze_cd =
230 BSWAP_32(buf->mz_chunk[i].mze_cd);
235 zap_byteswap(void *buf, size_t size)
239 block_type = *(uint64_t *)buf;
241 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
242 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
243 mzap_byteswap(buf, size);
245 fzap_byteswap(buf, size);
250 mze_compare(const void *arg1, const void *arg2)
252 const mzap_ent_t *mze1 = arg1;
253 const mzap_ent_t *mze2 = arg2;
255 if (mze1->mze_hash > mze2->mze_hash)
257 if (mze1->mze_hash < mze2->mze_hash)
259 if (mze1->mze_cd > mze2->mze_cd)
261 if (mze1->mze_cd < mze2->mze_cd)
267 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
272 ASSERT(zap->zap_ismicro);
273 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
275 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
276 mze->mze_chunkid = chunkid;
277 mze->mze_hash = hash;
278 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
279 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
280 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
281 kmem_free(mze, sizeof (mzap_ent_t));
284 avl_insert(&zap->zap_m.zap_avl, mze, idx);
289 mze_find(zap_name_t *zn)
291 mzap_ent_t mze_tofind;
294 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
296 ASSERT(zn->zn_zap->zap_ismicro);
297 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
299 mze_tofind.mze_hash = zn->zn_hash;
300 mze_tofind.mze_cd = 0;
303 mze = avl_find(avl, &mze_tofind, &idx);
305 mze = avl_nearest(avl, idx, AVL_AFTER);
306 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
307 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
308 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
311 if (zn->zn_matchtype == MT_BEST) {
312 zn->zn_matchtype = MT_FIRST;
319 mze_find_unused_cd(zap_t *zap, uint64_t hash)
321 mzap_ent_t mze_tofind;
324 avl_tree_t *avl = &zap->zap_m.zap_avl;
327 ASSERT(zap->zap_ismicro);
328 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
330 mze_tofind.mze_hash = hash;
331 mze_tofind.mze_cd = 0;
334 for (mze = avl_find(avl, &mze_tofind, &idx);
335 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
336 if (mze->mze_cd != cd)
345 mze_remove(zap_t *zap, mzap_ent_t *mze)
347 ASSERT(zap->zap_ismicro);
348 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
350 avl_remove(&zap->zap_m.zap_avl, mze);
351 kmem_free(mze, sizeof (mzap_ent_t));
355 mze_destroy(zap_t *zap)
358 void *avlcookie = NULL;
360 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
361 kmem_free(mze, sizeof (mzap_ent_t));
362 avl_destroy(&zap->zap_m.zap_avl);
366 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
372 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
374 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
375 rw_init(&zap->zap_rwlock, 0, 0, 0);
376 rw_enter(&zap->zap_rwlock, RW_WRITER);
377 zap->zap_objset = os;
378 zap->zap_object = obj;
381 if (*(uint64_t *)db->db_data != ZBT_MICRO) {
382 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
383 zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
385 zap->zap_ismicro = TRUE;
389 * Make sure that zap_ismicro is set before we let others see
390 * it, because zap_lockdir() checks zap_ismicro without the lock
393 winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict);
395 if (winner != NULL) {
396 rw_exit(&zap->zap_rwlock);
397 rw_destroy(&zap->zap_rwlock);
398 if (!zap->zap_ismicro)
399 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
400 kmem_free(zap, sizeof (zap_t));
404 if (zap->zap_ismicro) {
405 zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
406 zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags;
407 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
408 avl_create(&zap->zap_m.zap_avl, mze_compare,
409 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
411 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
412 mzap_ent_phys_t *mze =
413 &zap->zap_m.zap_phys->mz_chunk[i];
414 if (mze->mze_name[0]) {
417 zn = zap_name_alloc(zap, mze->mze_name,
419 if (mze_insert(zap, i, zn->zn_hash) == 0)
420 zap->zap_m.zap_num_entries++;
422 printf("ZFS WARNING: Duplicated ZAP "
423 "entry detected (%s).\n",
430 zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
431 zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags;
433 ASSERT3U(sizeof (struct zap_leaf_header), ==,
434 2*ZAP_LEAF_CHUNKSIZE);
437 * The embedded pointer table should not overlap the
440 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
441 &zap->zap_f.zap_phys->zap_salt);
444 * The embedded pointer table should end at the end of
447 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
448 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
449 (uintptr_t)zap->zap_f.zap_phys, ==,
450 zap->zap_dbuf->db_size);
452 rw_exit(&zap->zap_rwlock);
457 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
458 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
467 err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH);
473 dmu_object_info_t doi;
474 dmu_object_info_from_db(db, &doi);
475 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
479 zap = dmu_buf_get_user(db);
481 zap = mzap_open(os, obj, db);
484 * We're checking zap_ismicro without the lock held, in order to
485 * tell what type of lock we want. Once we have some sort of
486 * lock, see if it really is the right type. In practice this
487 * can only be different if it was upgraded from micro to fat,
488 * and micro wanted WRITER but fat only needs READER.
490 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
491 rw_enter(&zap->zap_rwlock, lt);
492 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
493 /* it was upgraded, now we only need reader */
494 ASSERT(lt == RW_WRITER);
496 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
497 rw_downgrade(&zap->zap_rwlock);
501 zap->zap_objset = os;
504 dmu_buf_will_dirty(db, tx);
506 ASSERT3P(zap->zap_dbuf, ==, db);
508 ASSERT(!zap->zap_ismicro ||
509 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
510 if (zap->zap_ismicro && tx && adding &&
511 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
512 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
513 if (newsz > MZAP_MAX_BLKSZ) {
514 dprintf("upgrading obj %llu: num_entries=%u\n",
515 obj, zap->zap_m.zap_num_entries);
517 return (mzap_upgrade(zapp, tx, 0));
519 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
521 zap->zap_m.zap_num_chunks =
522 db->db_size / MZAP_ENT_LEN - 1;
530 zap_unlockdir(zap_t *zap)
532 rw_exit(&zap->zap_rwlock);
533 dmu_buf_rele(zap->zap_dbuf, NULL);
537 mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
544 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
546 sz = zap->zap_dbuf->db_size;
547 mzp = kmem_alloc(sz, KM_SLEEP);
548 bcopy(zap->zap_dbuf->db_data, mzp, sz);
549 nchunks = zap->zap_m.zap_num_chunks;
552 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
553 1ULL << fzap_default_block_shift, 0, tx);
560 dprintf("upgrading obj=%llu with %u chunks\n",
561 zap->zap_object, nchunks);
562 /* XXX destroy the avl later, so we can use the stored hash value */
565 fzap_upgrade(zap, tx, flags);
567 for (i = 0; i < nchunks; i++) {
568 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
570 if (mze->mze_name[0] == 0)
572 dprintf("adding %s=%llu\n",
573 mze->mze_name, mze->mze_value);
574 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
575 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
576 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
587 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
593 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
597 dmu_object_info_t doi;
598 dmu_object_info_from_db(db, &doi);
599 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
603 dmu_buf_will_dirty(db, tx);
605 zp->mz_block_type = ZBT_MICRO;
606 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
607 zp->mz_normflags = normflags;
608 dmu_buf_rele(db, FTAG);
612 /* Only fat zap supports flags; upgrade immediately. */
613 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
614 B_FALSE, B_FALSE, &zap));
615 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags));
621 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
622 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
624 return (zap_create_claim_norm(os, obj,
625 0, ot, bonustype, bonuslen, tx));
629 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
630 dmu_object_type_t ot,
631 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
635 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
638 mzap_create_impl(os, obj, normflags, 0, tx);
643 zap_create(objset_t *os, dmu_object_type_t ot,
644 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
646 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
650 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
651 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
653 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
655 mzap_create_impl(os, obj, normflags, 0, tx);
660 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
661 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
662 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
664 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
666 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
667 leaf_blockshift <= SPA_MAXBLOCKSHIFT &&
668 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
669 indirect_blockshift <= SPA_MAXBLOCKSHIFT);
671 VERIFY(dmu_object_set_blocksize(os, obj,
672 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
674 mzap_create_impl(os, obj, normflags, flags, tx);
679 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
682 * dmu_object_free will free the object number and free the
683 * data. Freeing the data will cause our pageout function to be
684 * called, which will destroy our data (zap_leaf_t's and zap_t).
687 return (dmu_object_free(os, zapobj, tx));
692 zap_evict(dmu_buf_t *db, void *vzap)
696 rw_destroy(&zap->zap_rwlock);
698 if (zap->zap_ismicro)
701 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
703 kmem_free(zap, sizeof (zap_t));
707 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
712 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
715 if (!zap->zap_ismicro) {
716 err = fzap_count(zap, count);
718 *count = zap->zap_m.zap_num_entries;
725 * zn may be NULL; if not specified, it will be computed if needed.
726 * See also the comment above zap_entry_normalization_conflict().
729 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
732 int direction = AVL_BEFORE;
733 boolean_t allocdzn = B_FALSE;
735 if (zap->zap_normflags == 0)
739 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
740 other && other->mze_hash == mze->mze_hash;
741 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
744 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
748 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
755 if (direction == AVL_BEFORE) {
756 direction = AVL_AFTER;
766 * Routines for manipulating attributes.
770 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
771 uint64_t integer_size, uint64_t num_integers, void *buf)
773 return (zap_lookup_norm(os, zapobj, name, integer_size,
774 num_integers, buf, MT_EXACT, NULL, 0, NULL));
778 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
779 uint64_t integer_size, uint64_t num_integers, void *buf,
780 matchtype_t mt, char *realname, int rn_len,
788 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
791 zn = zap_name_alloc(zap, name, mt);
794 return (SET_ERROR(ENOTSUP));
797 if (!zap->zap_ismicro) {
798 err = fzap_lookup(zn, integer_size, num_integers, buf,
799 realname, rn_len, ncp);
803 err = SET_ERROR(ENOENT);
805 if (num_integers < 1) {
806 err = SET_ERROR(EOVERFLOW);
807 } else if (integer_size != 8) {
808 err = SET_ERROR(EINVAL);
811 MZE_PHYS(zap, mze)->mze_value;
812 (void) strlcpy(realname,
813 MZE_PHYS(zap, mze)->mze_name, rn_len);
815 *ncp = mzap_normalization_conflict(zap,
827 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
834 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
837 zn = zap_name_alloc_uint64(zap, key, key_numints);
840 return (SET_ERROR(ENOTSUP));
850 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
851 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
857 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
860 zn = zap_name_alloc_uint64(zap, key, key_numints);
863 return (SET_ERROR(ENOTSUP));
866 err = fzap_lookup(zn, integer_size, num_integers, buf,
874 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
876 int err = (zap_lookup_norm(os, zapobj, name, 0,
877 0, NULL, MT_EXACT, NULL, 0, NULL));
878 if (err == EOVERFLOW || err == EINVAL)
879 err = 0; /* found, but skipped reading the value */
884 zap_length(objset_t *os, uint64_t zapobj, const char *name,
885 uint64_t *integer_size, uint64_t *num_integers)
892 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
895 zn = zap_name_alloc(zap, name, MT_EXACT);
898 return (SET_ERROR(ENOTSUP));
900 if (!zap->zap_ismicro) {
901 err = fzap_length(zn, integer_size, num_integers);
905 err = SET_ERROR(ENOENT);
919 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
920 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
926 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
929 zn = zap_name_alloc_uint64(zap, key, key_numints);
932 return (SET_ERROR(ENOTSUP));
934 err = fzap_length(zn, integer_size, num_integers);
941 mzap_addent(zap_name_t *zn, uint64_t value)
944 zap_t *zap = zn->zn_zap;
945 int start = zap->zap_m.zap_alloc_next;
948 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
951 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
952 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
953 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
957 cd = mze_find_unused_cd(zap, zn->zn_hash);
958 /* given the limited size of the microzap, this can't happen */
959 ASSERT(cd < zap_maxcd(zap));
962 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
963 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
964 if (mze->mze_name[0] == 0) {
965 mze->mze_value = value;
967 (void) strcpy(mze->mze_name, zn->zn_key_orig);
968 zap->zap_m.zap_num_entries++;
969 zap->zap_m.zap_alloc_next = i+1;
970 if (zap->zap_m.zap_alloc_next ==
971 zap->zap_m.zap_num_chunks)
972 zap->zap_m.zap_alloc_next = 0;
973 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
981 ASSERT(!"out of entries!");
985 zap_add(objset_t *os, uint64_t zapobj, const char *key,
986 int integer_size, uint64_t num_integers,
987 const void *val, dmu_tx_t *tx)
992 const uint64_t *intval = val;
995 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
998 zn = zap_name_alloc(zap, key, MT_EXACT);
1001 return (SET_ERROR(ENOTSUP));
1003 if (!zap->zap_ismicro) {
1004 err = fzap_add(zn, integer_size, num_integers, val, tx);
1005 zap = zn->zn_zap; /* fzap_add() may change zap */
1006 } else if (integer_size != 8 || num_integers != 1 ||
1007 strlen(key) >= MZAP_NAME_LEN) {
1008 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1010 err = fzap_add(zn, integer_size, num_integers, val, tx);
1011 zap = zn->zn_zap; /* fzap_add() may change zap */
1015 err = SET_ERROR(EEXIST);
1017 mzap_addent(zn, *intval);
1020 ASSERT(zap == zn->zn_zap);
1022 if (zap != NULL) /* may be NULL if fzap_add() failed */
1028 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1029 int key_numints, int integer_size, uint64_t num_integers,
1030 const void *val, dmu_tx_t *tx)
1036 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1039 zn = zap_name_alloc_uint64(zap, key, key_numints);
1042 return (SET_ERROR(ENOTSUP));
1044 err = fzap_add(zn, integer_size, num_integers, val, tx);
1045 zap = zn->zn_zap; /* fzap_add() may change zap */
1047 if (zap != NULL) /* may be NULL if fzap_add() failed */
1053 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1054 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1059 const uint64_t *intval = val;
1065 * If there is an old value, it shouldn't change across the
1066 * lockdir (eg, due to bprewrite's xlation).
1068 if (integer_size == 8 && num_integers == 1)
1069 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1072 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1075 zn = zap_name_alloc(zap, name, MT_EXACT);
1078 return (SET_ERROR(ENOTSUP));
1080 if (!zap->zap_ismicro) {
1081 err = fzap_update(zn, integer_size, num_integers, val, tx);
1082 zap = zn->zn_zap; /* fzap_update() may change zap */
1083 } else if (integer_size != 8 || num_integers != 1 ||
1084 strlen(name) >= MZAP_NAME_LEN) {
1085 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1086 zapobj, integer_size, num_integers, name);
1087 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1089 err = fzap_update(zn, integer_size, num_integers,
1091 zap = zn->zn_zap; /* fzap_update() may change zap */
1095 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1096 MZE_PHYS(zap, mze)->mze_value = *intval;
1098 mzap_addent(zn, *intval);
1101 ASSERT(zap == zn->zn_zap);
1103 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1109 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1111 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1117 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1120 zn = zap_name_alloc_uint64(zap, key, key_numints);
1123 return (SET_ERROR(ENOTSUP));
1125 err = fzap_update(zn, integer_size, num_integers, val, tx);
1126 zap = zn->zn_zap; /* fzap_update() may change zap */
1128 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1134 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1136 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1140 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1141 matchtype_t mt, dmu_tx_t *tx)
1148 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1151 zn = zap_name_alloc(zap, name, mt);
1154 return (SET_ERROR(ENOTSUP));
1156 if (!zap->zap_ismicro) {
1157 err = fzap_remove(zn, tx);
1161 err = SET_ERROR(ENOENT);
1163 zap->zap_m.zap_num_entries--;
1164 bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
1165 sizeof (mzap_ent_phys_t));
1166 mze_remove(zap, mze);
1175 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1176 int key_numints, dmu_tx_t *tx)
1182 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1185 zn = zap_name_alloc_uint64(zap, key, key_numints);
1188 return (SET_ERROR(ENOTSUP));
1190 err = fzap_remove(zn, tx);
1197 * Routines for iterating over the attributes.
1201 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1202 uint64_t serialized)
1207 zc->zc_zapobj = zapobj;
1208 zc->zc_serialized = serialized;
1214 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1216 zap_cursor_init_serialized(zc, os, zapobj, 0);
1220 zap_cursor_fini(zap_cursor_t *zc)
1223 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1224 zap_unlockdir(zc->zc_zap);
1228 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1229 zap_put_leaf(zc->zc_leaf);
1232 zc->zc_objset = NULL;
1236 zap_cursor_serialize(zap_cursor_t *zc)
1238 if (zc->zc_hash == -1ULL)
1240 if (zc->zc_zap == NULL)
1241 return (zc->zc_serialized);
1242 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1243 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1246 * We want to keep the high 32 bits of the cursor zero if we can, so
1247 * that 32-bit programs can access this. So usually use a small
1248 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1251 * [ collision differentiator | zap_hashbits()-bit hash value ]
1253 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1254 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1258 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1262 mzap_ent_t mze_tofind;
1265 if (zc->zc_hash == -1ULL)
1266 return (SET_ERROR(ENOENT));
1268 if (zc->zc_zap == NULL) {
1270 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1271 RW_READER, TRUE, FALSE, &zc->zc_zap);
1276 * To support zap_cursor_init_serialized, advance, retrieve,
1277 * we must add to the existing zc_cd, which may already
1278 * be 1 due to the zap_cursor_advance.
1280 ASSERT(zc->zc_hash == 0);
1281 hb = zap_hashbits(zc->zc_zap);
1282 zc->zc_hash = zc->zc_serialized << (64 - hb);
1283 zc->zc_cd += zc->zc_serialized >> hb;
1284 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1287 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1289 if (!zc->zc_zap->zap_ismicro) {
1290 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1292 mze_tofind.mze_hash = zc->zc_hash;
1293 mze_tofind.mze_cd = zc->zc_cd;
1295 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1297 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1301 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1302 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1303 za->za_normalization_conflict =
1304 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1305 za->za_integer_length = 8;
1306 za->za_num_integers = 1;
1307 za->za_first_integer = mzep->mze_value;
1308 (void) strcpy(za->za_name, mzep->mze_name);
1309 zc->zc_hash = mze->mze_hash;
1310 zc->zc_cd = mze->mze_cd;
1313 zc->zc_hash = -1ULL;
1314 err = SET_ERROR(ENOENT);
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);
1348 return (SET_ERROR(ENOTSUP));
1351 if (!zc->zc_zap->zap_ismicro) {
1352 err = fzap_cursor_move_to_key(zc, zn);
1356 err = SET_ERROR(ENOENT);
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;