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) 2011, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30 #include <sys/zfs_context.h>
32 #include <sys/refcount.h>
33 #include <sys/zap_impl.h>
34 #include <sys/zap_leaf.h>
37 #include <sys/dmu_objset.h>
40 #include <sys/sunddi.h>
43 extern inline mzap_phys_t *zap_m_phys(zap_t *zap);
45 static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags);
48 zap_getflags(zap_t *zap)
52 return (zap_f_phys(zap)->zap_flags);
56 zap_hashbits(zap_t *zap)
58 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
67 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
74 zap_hash(zap_name_t *zn)
76 zap_t *zap = zn->zn_zap;
79 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
80 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
81 h = *(uint64_t *)zn->zn_key_orig;
85 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
87 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
89 const uint64_t *wp = zn->zn_key_norm;
91 ASSERT(zn->zn_key_intlen == 8);
92 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
96 for (j = 0; j < zn->zn_key_intlen; j++) {
98 zfs_crc64_table[(h ^ word) & 0xFF];
104 const uint8_t *cp = zn->zn_key_norm;
107 * We previously stored the terminating null on
108 * disk, but didn't hash it, so we need to
109 * continue to not hash it. (The
110 * zn_key_*_numints includes the terminating
111 * null for non-binary keys.)
113 len = zn->zn_key_norm_numints - 1;
115 ASSERT(zn->zn_key_intlen == 1);
116 for (i = 0; i < len; cp++, i++) {
118 zfs_crc64_table[(h ^ *cp) & 0xFF];
123 * Don't use all 64 bits, since we need some in the cookie for
124 * the collision differentiator. We MUST use the high bits,
125 * since those are the ones that we first pay attention to when
126 * chosing the bucket.
128 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
134 zap_normalize(zap_t *zap, const char *name, char *namenorm)
136 size_t inlen, outlen;
139 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
141 inlen = strlen(name) + 1;
142 outlen = ZAP_MAXNAMELEN;
145 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
146 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
147 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
153 zap_match(zap_name_t *zn, const char *matchname)
155 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
157 if (zn->zn_matchtype == MT_FIRST) {
158 char norm[ZAP_MAXNAMELEN];
160 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
163 return (strcmp(zn->zn_key_norm, norm) == 0);
165 /* MT_BEST or MT_EXACT */
166 return (strcmp(zn->zn_key_orig, matchname) == 0);
171 zap_name_free(zap_name_t *zn)
173 kmem_free(zn, sizeof (zap_name_t));
177 zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
179 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
182 zn->zn_key_intlen = sizeof (*key);
183 zn->zn_key_orig = key;
184 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
185 zn->zn_matchtype = mt;
186 if (zap->zap_normflags) {
187 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
191 zn->zn_key_norm = zn->zn_normbuf;
192 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
194 if (mt != MT_EXACT) {
198 zn->zn_key_norm = zn->zn_key_orig;
199 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
202 zn->zn_hash = zap_hash(zn);
207 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
209 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
211 ASSERT(zap->zap_normflags == 0);
213 zn->zn_key_intlen = sizeof (*key);
214 zn->zn_key_orig = zn->zn_key_norm = key;
215 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
216 zn->zn_matchtype = MT_EXACT;
218 zn->zn_hash = zap_hash(zn);
223 mzap_byteswap(mzap_phys_t *buf, size_t size)
226 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
227 buf->mz_salt = BSWAP_64(buf->mz_salt);
228 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
229 max = (size / MZAP_ENT_LEN) - 1;
230 for (i = 0; i < max; i++) {
231 buf->mz_chunk[i].mze_value =
232 BSWAP_64(buf->mz_chunk[i].mze_value);
233 buf->mz_chunk[i].mze_cd =
234 BSWAP_32(buf->mz_chunk[i].mze_cd);
239 zap_byteswap(void *buf, size_t size)
243 block_type = *(uint64_t *)buf;
245 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
246 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
247 mzap_byteswap(buf, size);
249 fzap_byteswap(buf, size);
254 mze_compare(const void *arg1, const void *arg2)
256 const mzap_ent_t *mze1 = arg1;
257 const mzap_ent_t *mze2 = arg2;
259 if (mze1->mze_hash > mze2->mze_hash)
261 if (mze1->mze_hash < mze2->mze_hash)
263 if (mze1->mze_cd > mze2->mze_cd)
265 if (mze1->mze_cd < mze2->mze_cd)
271 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
276 ASSERT(zap->zap_ismicro);
277 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
279 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
280 mze->mze_chunkid = chunkid;
281 mze->mze_hash = hash;
282 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
283 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
284 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
285 kmem_free(mze, sizeof (mzap_ent_t));
288 avl_insert(&zap->zap_m.zap_avl, mze, idx);
293 mze_find(zap_name_t *zn)
295 mzap_ent_t mze_tofind;
298 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
300 ASSERT(zn->zn_zap->zap_ismicro);
301 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
303 mze_tofind.mze_hash = zn->zn_hash;
304 mze_tofind.mze_cd = 0;
307 mze = avl_find(avl, &mze_tofind, &idx);
309 mze = avl_nearest(avl, idx, AVL_AFTER);
310 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
311 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
312 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
315 if (zn->zn_matchtype == MT_BEST) {
316 zn->zn_matchtype = MT_FIRST;
323 mze_find_unused_cd(zap_t *zap, uint64_t hash)
325 mzap_ent_t mze_tofind;
328 avl_tree_t *avl = &zap->zap_m.zap_avl;
331 ASSERT(zap->zap_ismicro);
332 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
334 mze_tofind.mze_hash = hash;
335 mze_tofind.mze_cd = 0;
338 for (mze = avl_find(avl, &mze_tofind, &idx);
339 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
340 if (mze->mze_cd != cd)
349 mze_remove(zap_t *zap, mzap_ent_t *mze)
351 ASSERT(zap->zap_ismicro);
352 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
354 avl_remove(&zap->zap_m.zap_avl, mze);
355 kmem_free(mze, sizeof (mzap_ent_t));
359 mze_destroy(zap_t *zap)
362 void *avlcookie = NULL;
364 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
365 kmem_free(mze, sizeof (mzap_ent_t));
366 avl_destroy(&zap->zap_m.zap_avl);
370 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
376 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
378 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
379 rw_init(&zap->zap_rwlock, 0, 0, 0);
380 rw_enter(&zap->zap_rwlock, RW_WRITER);
381 zap->zap_objset = os;
382 zap->zap_object = obj;
385 if (*(uint64_t *)db->db_data != ZBT_MICRO) {
386 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
387 zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
389 zap->zap_ismicro = TRUE;
393 * Make sure that zap_ismicro is set before we let others see
394 * it, because zap_lockdir() checks zap_ismicro without the lock
397 dmu_buf_init_user(&zap->zap_dbu, zap_evict, &zap->zap_dbuf);
398 winner = dmu_buf_set_user(db, &zap->zap_dbu);
400 if (winner != NULL) {
401 rw_exit(&zap->zap_rwlock);
402 rw_destroy(&zap->zap_rwlock);
403 if (!zap->zap_ismicro)
404 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
405 kmem_free(zap, sizeof (zap_t));
409 if (zap->zap_ismicro) {
410 zap->zap_salt = zap_m_phys(zap)->mz_salt;
411 zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
412 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
413 avl_create(&zap->zap_m.zap_avl, mze_compare,
414 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
416 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
417 mzap_ent_phys_t *mze =
418 &zap_m_phys(zap)->mz_chunk[i];
419 if (mze->mze_name[0]) {
422 zn = zap_name_alloc(zap, mze->mze_name,
424 if (mze_insert(zap, i, zn->zn_hash) == 0)
425 zap->zap_m.zap_num_entries++;
427 printf("ZFS WARNING: Duplicated ZAP "
428 "entry detected (%s).\n",
435 zap->zap_salt = zap_f_phys(zap)->zap_salt;
436 zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
438 ASSERT3U(sizeof (struct zap_leaf_header), ==,
439 2*ZAP_LEAF_CHUNKSIZE);
442 * The embedded pointer table should not overlap the
445 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
446 &zap_f_phys(zap)->zap_salt);
449 * The embedded pointer table should end at the end of
452 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
453 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
454 (uintptr_t)zap_f_phys(zap), ==,
455 zap->zap_dbuf->db_size);
457 rw_exit(&zap->zap_rwlock);
462 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
463 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
472 err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH);
478 dmu_object_info_t doi;
479 dmu_object_info_from_db(db, &doi);
480 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
484 zap = dmu_buf_get_user(db);
486 zap = mzap_open(os, obj, db);
489 * We're checking zap_ismicro without the lock held, in order to
490 * tell what type of lock we want. Once we have some sort of
491 * lock, see if it really is the right type. In practice this
492 * can only be different if it was upgraded from micro to fat,
493 * and micro wanted WRITER but fat only needs READER.
495 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
496 rw_enter(&zap->zap_rwlock, lt);
497 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
498 /* it was upgraded, now we only need reader */
499 ASSERT(lt == RW_WRITER);
501 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
502 rw_downgrade(&zap->zap_rwlock);
506 zap->zap_objset = os;
509 dmu_buf_will_dirty(db, tx);
511 ASSERT3P(zap->zap_dbuf, ==, db);
513 ASSERT(!zap->zap_ismicro ||
514 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
515 if (zap->zap_ismicro && tx && adding &&
516 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
517 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
518 if (newsz > MZAP_MAX_BLKSZ) {
519 dprintf("upgrading obj %llu: num_entries=%u\n",
520 obj, zap->zap_m.zap_num_entries);
522 return (mzap_upgrade(zapp, tx, 0));
524 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
526 zap->zap_m.zap_num_chunks =
527 db->db_size / MZAP_ENT_LEN - 1;
535 zap_unlockdir(zap_t *zap)
537 rw_exit(&zap->zap_rwlock);
538 dmu_buf_rele(zap->zap_dbuf, NULL);
542 mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
549 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
551 sz = zap->zap_dbuf->db_size;
552 mzp = zio_buf_alloc(sz);
553 bcopy(zap->zap_dbuf->db_data, mzp, sz);
554 nchunks = zap->zap_m.zap_num_chunks;
557 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
558 1ULL << fzap_default_block_shift, 0, tx);
560 zio_buf_free(mzp, sz);
565 dprintf("upgrading obj=%llu with %u chunks\n",
566 zap->zap_object, nchunks);
567 /* XXX destroy the avl later, so we can use the stored hash value */
570 fzap_upgrade(zap, tx, flags);
572 for (i = 0; i < nchunks; i++) {
573 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
575 if (mze->mze_name[0] == 0)
577 dprintf("adding %s=%llu\n",
578 mze->mze_name, mze->mze_value);
579 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
580 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
581 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
586 zio_buf_free(mzp, sz);
592 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
598 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
602 dmu_object_info_t doi;
603 dmu_object_info_from_db(db, &doi);
604 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
608 dmu_buf_will_dirty(db, tx);
610 zp->mz_block_type = ZBT_MICRO;
611 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
612 zp->mz_normflags = normflags;
613 dmu_buf_rele(db, FTAG);
617 /* Only fat zap supports flags; upgrade immediately. */
618 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
619 B_FALSE, B_FALSE, &zap));
620 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags));
626 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
627 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
629 return (zap_create_claim_norm(os, obj,
630 0, ot, bonustype, bonuslen, tx));
634 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
635 dmu_object_type_t ot,
636 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
640 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
643 mzap_create_impl(os, obj, normflags, 0, tx);
648 zap_create(objset_t *os, dmu_object_type_t ot,
649 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
651 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
655 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
656 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
658 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
660 mzap_create_impl(os, obj, normflags, 0, tx);
665 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
666 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
667 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
669 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
671 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
672 leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
673 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
674 indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
676 VERIFY(dmu_object_set_blocksize(os, obj,
677 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
679 mzap_create_impl(os, obj, normflags, flags, tx);
684 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
687 * dmu_object_free will free the object number and free the
688 * data. Freeing the data will cause our pageout function to be
689 * called, which will destroy our data (zap_leaf_t's and zap_t).
692 return (dmu_object_free(os, zapobj, tx));
700 rw_destroy(&zap->zap_rwlock);
702 if (zap->zap_ismicro)
705 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
707 kmem_free(zap, sizeof (zap_t));
711 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
716 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
719 if (!zap->zap_ismicro) {
720 err = fzap_count(zap, count);
722 *count = zap->zap_m.zap_num_entries;
729 * zn may be NULL; if not specified, it will be computed if needed.
730 * See also the comment above zap_entry_normalization_conflict().
733 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
736 int direction = AVL_BEFORE;
737 boolean_t allocdzn = B_FALSE;
739 if (zap->zap_normflags == 0)
743 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
744 other && other->mze_hash == mze->mze_hash;
745 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
748 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
752 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
759 if (direction == AVL_BEFORE) {
760 direction = AVL_AFTER;
770 * Routines for manipulating attributes.
774 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
775 uint64_t integer_size, uint64_t num_integers, void *buf)
777 return (zap_lookup_norm(os, zapobj, name, integer_size,
778 num_integers, buf, MT_EXACT, NULL, 0, NULL));
782 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
783 uint64_t integer_size, uint64_t num_integers, void *buf,
784 matchtype_t mt, char *realname, int rn_len,
792 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
795 zn = zap_name_alloc(zap, name, mt);
798 return (SET_ERROR(ENOTSUP));
801 if (!zap->zap_ismicro) {
802 err = fzap_lookup(zn, integer_size, num_integers, buf,
803 realname, rn_len, ncp);
807 err = SET_ERROR(ENOENT);
809 if (num_integers < 1) {
810 err = SET_ERROR(EOVERFLOW);
811 } else if (integer_size != 8) {
812 err = SET_ERROR(EINVAL);
815 MZE_PHYS(zap, mze)->mze_value;
816 (void) strlcpy(realname,
817 MZE_PHYS(zap, mze)->mze_name, rn_len);
819 *ncp = mzap_normalization_conflict(zap,
831 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
838 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
841 zn = zap_name_alloc_uint64(zap, key, key_numints);
844 return (SET_ERROR(ENOTSUP));
854 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
855 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
861 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
864 zn = zap_name_alloc_uint64(zap, key, key_numints);
867 return (SET_ERROR(ENOTSUP));
870 err = fzap_lookup(zn, integer_size, num_integers, buf,
878 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
880 int err = zap_lookup_norm(os, zapobj, name, 0,
881 0, NULL, MT_EXACT, NULL, 0, NULL);
882 if (err == EOVERFLOW || err == EINVAL)
883 err = 0; /* found, but skipped reading the value */
888 zap_length(objset_t *os, uint64_t zapobj, const char *name,
889 uint64_t *integer_size, uint64_t *num_integers)
896 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
899 zn = zap_name_alloc(zap, name, MT_EXACT);
902 return (SET_ERROR(ENOTSUP));
904 if (!zap->zap_ismicro) {
905 err = fzap_length(zn, integer_size, num_integers);
909 err = SET_ERROR(ENOENT);
923 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
924 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
930 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
933 zn = zap_name_alloc_uint64(zap, key, key_numints);
936 return (SET_ERROR(ENOTSUP));
938 err = fzap_length(zn, integer_size, num_integers);
945 mzap_addent(zap_name_t *zn, uint64_t value)
948 zap_t *zap = zn->zn_zap;
949 int start = zap->zap_m.zap_alloc_next;
952 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
955 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
956 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
957 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
961 cd = mze_find_unused_cd(zap, zn->zn_hash);
962 /* given the limited size of the microzap, this can't happen */
963 ASSERT(cd < zap_maxcd(zap));
966 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
967 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
968 if (mze->mze_name[0] == 0) {
969 mze->mze_value = value;
971 (void) strcpy(mze->mze_name, zn->zn_key_orig);
972 zap->zap_m.zap_num_entries++;
973 zap->zap_m.zap_alloc_next = i+1;
974 if (zap->zap_m.zap_alloc_next ==
975 zap->zap_m.zap_num_chunks)
976 zap->zap_m.zap_alloc_next = 0;
977 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
985 ASSERT(!"out of entries!");
989 zap_add(objset_t *os, uint64_t zapobj, const char *key,
990 int integer_size, uint64_t num_integers,
991 const void *val, dmu_tx_t *tx)
996 const uint64_t *intval = val;
999 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1002 zn = zap_name_alloc(zap, key, MT_EXACT);
1005 return (SET_ERROR(ENOTSUP));
1007 if (!zap->zap_ismicro) {
1008 err = fzap_add(zn, integer_size, num_integers, val, tx);
1009 zap = zn->zn_zap; /* fzap_add() may change zap */
1010 } else if (integer_size != 8 || num_integers != 1 ||
1011 strlen(key) >= MZAP_NAME_LEN) {
1012 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1014 err = fzap_add(zn, integer_size, num_integers, val, tx);
1015 zap = zn->zn_zap; /* fzap_add() may change zap */
1019 err = SET_ERROR(EEXIST);
1021 mzap_addent(zn, *intval);
1024 ASSERT(zap == zn->zn_zap);
1026 if (zap != NULL) /* may be NULL if fzap_add() failed */
1032 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1033 int key_numints, int integer_size, uint64_t num_integers,
1034 const void *val, dmu_tx_t *tx)
1040 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1043 zn = zap_name_alloc_uint64(zap, key, key_numints);
1046 return (SET_ERROR(ENOTSUP));
1048 err = fzap_add(zn, integer_size, num_integers, val, tx);
1049 zap = zn->zn_zap; /* fzap_add() may change zap */
1051 if (zap != NULL) /* may be NULL if fzap_add() failed */
1057 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1058 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1063 const uint64_t *intval = val;
1069 * If there is an old value, it shouldn't change across the
1070 * lockdir (eg, due to bprewrite's xlation).
1072 if (integer_size == 8 && num_integers == 1)
1073 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1076 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1079 zn = zap_name_alloc(zap, name, MT_EXACT);
1082 return (SET_ERROR(ENOTSUP));
1084 if (!zap->zap_ismicro) {
1085 err = fzap_update(zn, integer_size, num_integers, val, tx);
1086 zap = zn->zn_zap; /* fzap_update() may change zap */
1087 } else if (integer_size != 8 || num_integers != 1 ||
1088 strlen(name) >= MZAP_NAME_LEN) {
1089 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1090 zapobj, integer_size, num_integers, name);
1091 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1093 err = fzap_update(zn, integer_size, num_integers,
1095 zap = zn->zn_zap; /* fzap_update() may change zap */
1099 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1100 MZE_PHYS(zap, mze)->mze_value = *intval;
1102 mzap_addent(zn, *intval);
1105 ASSERT(zap == zn->zn_zap);
1107 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1113 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1115 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1121 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1124 zn = zap_name_alloc_uint64(zap, key, key_numints);
1127 return (SET_ERROR(ENOTSUP));
1129 err = fzap_update(zn, integer_size, num_integers, val, tx);
1130 zap = zn->zn_zap; /* fzap_update() may change zap */
1132 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1138 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1140 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1144 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1145 matchtype_t mt, dmu_tx_t *tx)
1152 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1155 zn = zap_name_alloc(zap, name, mt);
1158 return (SET_ERROR(ENOTSUP));
1160 if (!zap->zap_ismicro) {
1161 err = fzap_remove(zn, tx);
1165 err = SET_ERROR(ENOENT);
1167 zap->zap_m.zap_num_entries--;
1168 bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1169 sizeof (mzap_ent_phys_t));
1170 mze_remove(zap, mze);
1179 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1180 int key_numints, dmu_tx_t *tx)
1186 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1189 zn = zap_name_alloc_uint64(zap, key, key_numints);
1192 return (SET_ERROR(ENOTSUP));
1194 err = fzap_remove(zn, tx);
1201 * Routines for iterating over the attributes.
1205 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1206 uint64_t serialized)
1211 zc->zc_zapobj = zapobj;
1212 zc->zc_serialized = serialized;
1218 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1220 zap_cursor_init_serialized(zc, os, zapobj, 0);
1224 zap_cursor_fini(zap_cursor_t *zc)
1227 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1228 zap_unlockdir(zc->zc_zap);
1232 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1233 zap_put_leaf(zc->zc_leaf);
1236 zc->zc_objset = NULL;
1240 zap_cursor_serialize(zap_cursor_t *zc)
1242 if (zc->zc_hash == -1ULL)
1244 if (zc->zc_zap == NULL)
1245 return (zc->zc_serialized);
1246 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1247 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1250 * We want to keep the high 32 bits of the cursor zero if we can, so
1251 * that 32-bit programs can access this. So usually use a small
1252 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1255 * [ collision differentiator | zap_hashbits()-bit hash value ]
1257 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1258 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1262 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1266 mzap_ent_t mze_tofind;
1269 if (zc->zc_hash == -1ULL)
1270 return (SET_ERROR(ENOENT));
1272 if (zc->zc_zap == NULL) {
1274 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1275 RW_READER, TRUE, FALSE, &zc->zc_zap);
1280 * To support zap_cursor_init_serialized, advance, retrieve,
1281 * we must add to the existing zc_cd, which may already
1282 * be 1 due to the zap_cursor_advance.
1284 ASSERT(zc->zc_hash == 0);
1285 hb = zap_hashbits(zc->zc_zap);
1286 zc->zc_hash = zc->zc_serialized << (64 - hb);
1287 zc->zc_cd += zc->zc_serialized >> hb;
1288 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1291 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1293 if (!zc->zc_zap->zap_ismicro) {
1294 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1296 mze_tofind.mze_hash = zc->zc_hash;
1297 mze_tofind.mze_cd = zc->zc_cd;
1299 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1301 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1305 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1306 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1307 za->za_normalization_conflict =
1308 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1309 za->za_integer_length = 8;
1310 za->za_num_integers = 1;
1311 za->za_first_integer = mzep->mze_value;
1312 (void) strcpy(za->za_name, mzep->mze_name);
1313 zc->zc_hash = mze->mze_hash;
1314 zc->zc_cd = mze->mze_cd;
1317 zc->zc_hash = -1ULL;
1318 err = SET_ERROR(ENOENT);
1321 rw_exit(&zc->zc_zap->zap_rwlock);
1326 zap_cursor_advance(zap_cursor_t *zc)
1328 if (zc->zc_hash == -1ULL)
1334 zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1340 if (zc->zc_zap == NULL) {
1341 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1342 RW_READER, TRUE, FALSE, &zc->zc_zap);
1346 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1349 zn = zap_name_alloc(zc->zc_zap, name, mt);
1351 rw_exit(&zc->zc_zap->zap_rwlock);
1352 return (SET_ERROR(ENOTSUP));
1355 if (!zc->zc_zap->zap_ismicro) {
1356 err = fzap_cursor_move_to_key(zc, zn);
1360 err = SET_ERROR(ENOENT);
1363 zc->zc_hash = mze->mze_hash;
1364 zc->zc_cd = mze->mze_cd;
1369 rw_exit(&zc->zc_zap->zap_rwlock);
1374 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1379 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1383 bzero(zs, sizeof (zap_stats_t));
1385 if (zap->zap_ismicro) {
1386 zs->zs_blocksize = zap->zap_dbuf->db_size;
1387 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1388 zs->zs_num_blocks = 1;
1390 fzap_get_stats(zap, zs);
1397 zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
1398 uint64_t *towrite, uint64_t *tooverwrite)
1404 * Since, we don't have a name, we cannot figure out which blocks will
1405 * be affected in this operation. So, account for the worst case :
1406 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1407 * - 4 new blocks written if adding:
1408 * - 2 blocks for possibly split leaves,
1409 * - 2 grown ptrtbl blocks
1411 * This also accomodates the case where an add operation to a fairly
1412 * large microzap results in a promotion to fatzap.
1415 *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE;
1420 * We lock the zap with adding == FALSE. Because, if we pass
1421 * the actual value of add, it could trigger a mzap_upgrade().
1422 * At present we are just evaluating the possibility of this operation
1423 * and hence we donot want to trigger an upgrade.
1425 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1429 if (!zap->zap_ismicro) {
1430 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1432 err = fzap_count_write(zn, add, towrite,
1437 * We treat this case as similar to (name == NULL)
1439 *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE;
1443 * We are here if (name != NULL) and this is a micro-zap.
1444 * We account for the header block depending on whether it
1447 * Incase of an add-operation it is hard to find out
1448 * if this add will promote this microzap to fatzap.
1449 * Hence, we consider the worst case and account for the
1450 * blocks assuming this microzap would be promoted to a
1453 * 1 block overwritten : header block
1454 * 4 new blocks written : 2 new split leaf, 2 grown
1457 if (dmu_buf_freeable(zap->zap_dbuf))
1458 *tooverwrite += MZAP_MAX_BLKSZ;
1460 *towrite += MZAP_MAX_BLKSZ;
1463 *towrite += 4 * MZAP_MAX_BLKSZ;