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.
29 #include <sys/zfs_context.h>
31 #include <sys/refcount.h>
32 #include <sys/zap_impl.h>
33 #include <sys/zap_leaf.h>
36 #include <sys/dmu_objset.h>
39 #include <sys/sunddi.h>
42 extern inline mzap_phys_t *zap_m_phys(zap_t *zap);
44 static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags);
47 zap_getflags(zap_t *zap)
51 return (zap_f_phys(zap)->zap_flags);
55 zap_hashbits(zap_t *zap)
57 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
66 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
73 zap_hash(zap_name_t *zn)
75 zap_t *zap = zn->zn_zap;
78 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
79 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
80 h = *(uint64_t *)zn->zn_key_orig;
84 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
86 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
88 const uint64_t *wp = zn->zn_key_norm;
90 ASSERT(zn->zn_key_intlen == 8);
91 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
95 for (j = 0; j < zn->zn_key_intlen; j++) {
97 zfs_crc64_table[(h ^ word) & 0xFF];
103 const uint8_t *cp = zn->zn_key_norm;
106 * We previously stored the terminating null on
107 * disk, but didn't hash it, so we need to
108 * continue to not hash it. (The
109 * zn_key_*_numints includes the terminating
110 * null for non-binary keys.)
112 len = zn->zn_key_norm_numints - 1;
114 ASSERT(zn->zn_key_intlen == 1);
115 for (i = 0; i < len; cp++, i++) {
117 zfs_crc64_table[(h ^ *cp) & 0xFF];
122 * Don't use all 64 bits, since we need some in the cookie for
123 * the collision differentiator. We MUST use the high bits,
124 * since those are the ones that we first pay attention to when
125 * chosing the bucket.
127 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
133 zap_normalize(zap_t *zap, const char *name, char *namenorm)
135 size_t inlen, outlen;
138 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
140 inlen = strlen(name) + 1;
141 outlen = ZAP_MAXNAMELEN;
144 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
145 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
146 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
152 zap_match(zap_name_t *zn, const char *matchname)
154 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
156 if (zn->zn_matchtype == MT_FIRST) {
157 char norm[ZAP_MAXNAMELEN];
159 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
162 return (strcmp(zn->zn_key_norm, norm) == 0);
164 /* MT_BEST or MT_EXACT */
165 return (strcmp(zn->zn_key_orig, matchname) == 0);
170 zap_name_free(zap_name_t *zn)
172 kmem_free(zn, sizeof (zap_name_t));
176 zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
178 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
181 zn->zn_key_intlen = sizeof (*key);
182 zn->zn_key_orig = key;
183 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
184 zn->zn_matchtype = mt;
185 if (zap->zap_normflags) {
186 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
190 zn->zn_key_norm = zn->zn_normbuf;
191 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
193 if (mt != MT_EXACT) {
197 zn->zn_key_norm = zn->zn_key_orig;
198 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
201 zn->zn_hash = zap_hash(zn);
206 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
208 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
210 ASSERT(zap->zap_normflags == 0);
212 zn->zn_key_intlen = sizeof (*key);
213 zn->zn_key_orig = zn->zn_key_norm = key;
214 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
215 zn->zn_matchtype = MT_EXACT;
217 zn->zn_hash = zap_hash(zn);
222 mzap_byteswap(mzap_phys_t *buf, size_t size)
225 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
226 buf->mz_salt = BSWAP_64(buf->mz_salt);
227 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
228 max = (size / MZAP_ENT_LEN) - 1;
229 for (i = 0; i < max; i++) {
230 buf->mz_chunk[i].mze_value =
231 BSWAP_64(buf->mz_chunk[i].mze_value);
232 buf->mz_chunk[i].mze_cd =
233 BSWAP_32(buf->mz_chunk[i].mze_cd);
238 zap_byteswap(void *buf, size_t size)
242 block_type = *(uint64_t *)buf;
244 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
245 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
246 mzap_byteswap(buf, size);
248 fzap_byteswap(buf, size);
253 mze_compare(const void *arg1, const void *arg2)
255 const mzap_ent_t *mze1 = arg1;
256 const mzap_ent_t *mze2 = arg2;
258 if (mze1->mze_hash > mze2->mze_hash)
260 if (mze1->mze_hash < mze2->mze_hash)
262 if (mze1->mze_cd > mze2->mze_cd)
264 if (mze1->mze_cd < mze2->mze_cd)
270 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
275 ASSERT(zap->zap_ismicro);
276 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
278 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
279 mze->mze_chunkid = chunkid;
280 mze->mze_hash = hash;
281 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
282 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
283 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
284 kmem_free(mze, sizeof (mzap_ent_t));
287 avl_insert(&zap->zap_m.zap_avl, mze, idx);
292 mze_find(zap_name_t *zn)
294 mzap_ent_t mze_tofind;
297 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
299 ASSERT(zn->zn_zap->zap_ismicro);
300 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
302 mze_tofind.mze_hash = zn->zn_hash;
303 mze_tofind.mze_cd = 0;
306 mze = avl_find(avl, &mze_tofind, &idx);
308 mze = avl_nearest(avl, idx, AVL_AFTER);
309 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
310 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
311 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
314 if (zn->zn_matchtype == MT_BEST) {
315 zn->zn_matchtype = MT_FIRST;
322 mze_find_unused_cd(zap_t *zap, uint64_t hash)
324 mzap_ent_t mze_tofind;
327 avl_tree_t *avl = &zap->zap_m.zap_avl;
330 ASSERT(zap->zap_ismicro);
331 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
333 mze_tofind.mze_hash = hash;
334 mze_tofind.mze_cd = 0;
337 for (mze = avl_find(avl, &mze_tofind, &idx);
338 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
339 if (mze->mze_cd != cd)
348 mze_remove(zap_t *zap, mzap_ent_t *mze)
350 ASSERT(zap->zap_ismicro);
351 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
353 avl_remove(&zap->zap_m.zap_avl, mze);
354 kmem_free(mze, sizeof (mzap_ent_t));
358 mze_destroy(zap_t *zap)
361 void *avlcookie = NULL;
363 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
364 kmem_free(mze, sizeof (mzap_ent_t));
365 avl_destroy(&zap->zap_m.zap_avl);
369 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
375 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
377 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
378 rw_init(&zap->zap_rwlock, 0, 0, 0);
379 rw_enter(&zap->zap_rwlock, RW_WRITER);
380 zap->zap_objset = os;
381 zap->zap_object = obj;
384 if (*(uint64_t *)db->db_data != ZBT_MICRO) {
385 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
386 zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
388 zap->zap_ismicro = TRUE;
392 * Make sure that zap_ismicro is set before we let others see
393 * it, because zap_lockdir() checks zap_ismicro without the lock
396 winner = dmu_buf_set_user(db, zap, zap_evict);
398 if (winner != NULL) {
399 rw_exit(&zap->zap_rwlock);
400 rw_destroy(&zap->zap_rwlock);
401 if (!zap->zap_ismicro)
402 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
403 kmem_free(zap, sizeof (zap_t));
407 if (zap->zap_ismicro) {
408 zap->zap_salt = zap_m_phys(zap)->mz_salt;
409 zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
410 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
411 avl_create(&zap->zap_m.zap_avl, mze_compare,
412 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
414 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
415 mzap_ent_phys_t *mze =
416 &zap_m_phys(zap)->mz_chunk[i];
417 if (mze->mze_name[0]) {
420 zn = zap_name_alloc(zap, mze->mze_name,
422 if (mze_insert(zap, i, zn->zn_hash) == 0)
423 zap->zap_m.zap_num_entries++;
425 printf("ZFS WARNING: Duplicated ZAP "
426 "entry detected (%s).\n",
433 zap->zap_salt = zap_f_phys(zap)->zap_salt;
434 zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
436 ASSERT3U(sizeof (struct zap_leaf_header), ==,
437 2*ZAP_LEAF_CHUNKSIZE);
440 * The embedded pointer table should not overlap the
443 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
444 &zap_f_phys(zap)->zap_salt);
447 * The embedded pointer table should end at the end of
450 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
451 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
452 (uintptr_t)zap_f_phys(zap), ==,
453 zap->zap_dbuf->db_size);
455 rw_exit(&zap->zap_rwlock);
460 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
461 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
470 err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH);
476 dmu_object_info_t doi;
477 dmu_object_info_from_db(db, &doi);
478 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
482 zap = dmu_buf_get_user(db);
484 zap = mzap_open(os, obj, db);
487 * We're checking zap_ismicro without the lock held, in order to
488 * tell what type of lock we want. Once we have some sort of
489 * lock, see if it really is the right type. In practice this
490 * can only be different if it was upgraded from micro to fat,
491 * and micro wanted WRITER but fat only needs READER.
493 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
494 rw_enter(&zap->zap_rwlock, lt);
495 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
496 /* it was upgraded, now we only need reader */
497 ASSERT(lt == RW_WRITER);
499 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
500 rw_downgrade(&zap->zap_rwlock);
504 zap->zap_objset = os;
507 dmu_buf_will_dirty(db, tx);
509 ASSERT3P(zap->zap_dbuf, ==, db);
511 ASSERT(!zap->zap_ismicro ||
512 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
513 if (zap->zap_ismicro && tx && adding &&
514 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
515 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
516 if (newsz > MZAP_MAX_BLKSZ) {
517 dprintf("upgrading obj %llu: num_entries=%u\n",
518 obj, zap->zap_m.zap_num_entries);
520 return (mzap_upgrade(zapp, tx, 0));
522 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
524 zap->zap_m.zap_num_chunks =
525 db->db_size / MZAP_ENT_LEN - 1;
533 zap_unlockdir(zap_t *zap)
535 rw_exit(&zap->zap_rwlock);
536 dmu_buf_rele(zap->zap_dbuf, NULL);
540 mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
547 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
549 sz = zap->zap_dbuf->db_size;
550 mzp = zio_buf_alloc(sz);
551 bcopy(zap->zap_dbuf->db_data, mzp, sz);
552 nchunks = zap->zap_m.zap_num_chunks;
555 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
556 1ULL << fzap_default_block_shift, 0, tx);
558 zio_buf_free(mzp, sz);
563 dprintf("upgrading obj=%llu with %u chunks\n",
564 zap->zap_object, nchunks);
565 /* XXX destroy the avl later, so we can use the stored hash value */
568 fzap_upgrade(zap, tx, flags);
570 for (i = 0; i < nchunks; i++) {
571 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
573 if (mze->mze_name[0] == 0)
575 dprintf("adding %s=%llu\n",
576 mze->mze_name, mze->mze_value);
577 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
578 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
579 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
584 zio_buf_free(mzp, sz);
590 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
596 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
600 dmu_object_info_t doi;
601 dmu_object_info_from_db(db, &doi);
602 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
606 dmu_buf_will_dirty(db, tx);
608 zp->mz_block_type = ZBT_MICRO;
609 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
610 zp->mz_normflags = normflags;
611 dmu_buf_rele(db, FTAG);
615 /* Only fat zap supports flags; upgrade immediately. */
616 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
617 B_FALSE, B_FALSE, &zap));
618 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags));
624 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
625 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
627 return (zap_create_claim_norm(os, obj,
628 0, ot, bonustype, bonuslen, tx));
632 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
633 dmu_object_type_t ot,
634 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
638 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
641 mzap_create_impl(os, obj, normflags, 0, tx);
646 zap_create(objset_t *os, dmu_object_type_t ot,
647 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
649 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
653 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
654 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
656 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
658 mzap_create_impl(os, obj, normflags, 0, tx);
663 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
664 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
665 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
667 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
669 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
670 leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
671 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
672 indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
674 VERIFY(dmu_object_set_blocksize(os, obj,
675 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
677 mzap_create_impl(os, obj, normflags, flags, tx);
682 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
685 * dmu_object_free will free the object number and free the
686 * data. Freeing the data will cause our pageout function to be
687 * called, which will destroy our data (zap_leaf_t's and zap_t).
690 return (dmu_object_free(os, zapobj, tx));
695 zap_evict(dmu_buf_t *db, void *vzap)
699 rw_destroy(&zap->zap_rwlock);
701 if (zap->zap_ismicro)
704 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
706 kmem_free(zap, sizeof (zap_t));
710 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
715 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
718 if (!zap->zap_ismicro) {
719 err = fzap_count(zap, count);
721 *count = zap->zap_m.zap_num_entries;
728 * zn may be NULL; if not specified, it will be computed if needed.
729 * See also the comment above zap_entry_normalization_conflict().
732 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
735 int direction = AVL_BEFORE;
736 boolean_t allocdzn = B_FALSE;
738 if (zap->zap_normflags == 0)
742 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
743 other && other->mze_hash == mze->mze_hash;
744 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
747 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
751 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
758 if (direction == AVL_BEFORE) {
759 direction = AVL_AFTER;
769 * Routines for manipulating attributes.
773 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
774 uint64_t integer_size, uint64_t num_integers, void *buf)
776 return (zap_lookup_norm(os, zapobj, name, integer_size,
777 num_integers, buf, MT_EXACT, NULL, 0, NULL));
781 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
782 uint64_t integer_size, uint64_t num_integers, void *buf,
783 matchtype_t mt, char *realname, int rn_len,
791 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
794 zn = zap_name_alloc(zap, name, mt);
797 return (SET_ERROR(ENOTSUP));
800 if (!zap->zap_ismicro) {
801 err = fzap_lookup(zn, integer_size, num_integers, buf,
802 realname, rn_len, ncp);
806 err = SET_ERROR(ENOENT);
808 if (num_integers < 1) {
809 err = SET_ERROR(EOVERFLOW);
810 } else if (integer_size != 8) {
811 err = SET_ERROR(EINVAL);
814 MZE_PHYS(zap, mze)->mze_value;
815 (void) strlcpy(realname,
816 MZE_PHYS(zap, mze)->mze_name, rn_len);
818 *ncp = mzap_normalization_conflict(zap,
830 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
837 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
840 zn = zap_name_alloc_uint64(zap, key, key_numints);
843 return (SET_ERROR(ENOTSUP));
853 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
854 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
860 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
863 zn = zap_name_alloc_uint64(zap, key, key_numints);
866 return (SET_ERROR(ENOTSUP));
869 err = fzap_lookup(zn, integer_size, num_integers, buf,
877 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
879 int err = zap_lookup_norm(os, zapobj, name, 0,
880 0, NULL, MT_EXACT, NULL, 0, NULL);
881 if (err == EOVERFLOW || err == EINVAL)
882 err = 0; /* found, but skipped reading the value */
887 zap_length(objset_t *os, uint64_t zapobj, const char *name,
888 uint64_t *integer_size, uint64_t *num_integers)
895 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
898 zn = zap_name_alloc(zap, name, MT_EXACT);
901 return (SET_ERROR(ENOTSUP));
903 if (!zap->zap_ismicro) {
904 err = fzap_length(zn, integer_size, num_integers);
908 err = SET_ERROR(ENOENT);
922 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
923 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
929 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
932 zn = zap_name_alloc_uint64(zap, key, key_numints);
935 return (SET_ERROR(ENOTSUP));
937 err = fzap_length(zn, integer_size, num_integers);
944 mzap_addent(zap_name_t *zn, uint64_t value)
947 zap_t *zap = zn->zn_zap;
948 int start = zap->zap_m.zap_alloc_next;
951 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
954 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
955 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
956 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
960 cd = mze_find_unused_cd(zap, zn->zn_hash);
961 /* given the limited size of the microzap, this can't happen */
962 ASSERT(cd < zap_maxcd(zap));
965 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
966 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
967 if (mze->mze_name[0] == 0) {
968 mze->mze_value = value;
970 (void) strcpy(mze->mze_name, zn->zn_key_orig);
971 zap->zap_m.zap_num_entries++;
972 zap->zap_m.zap_alloc_next = i+1;
973 if (zap->zap_m.zap_alloc_next ==
974 zap->zap_m.zap_num_chunks)
975 zap->zap_m.zap_alloc_next = 0;
976 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
984 ASSERT(!"out of entries!");
988 zap_add(objset_t *os, uint64_t zapobj, const char *key,
989 int integer_size, uint64_t num_integers,
990 const void *val, dmu_tx_t *tx)
995 const uint64_t *intval = val;
998 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1001 zn = zap_name_alloc(zap, key, MT_EXACT);
1004 return (SET_ERROR(ENOTSUP));
1006 if (!zap->zap_ismicro) {
1007 err = fzap_add(zn, integer_size, num_integers, val, tx);
1008 zap = zn->zn_zap; /* fzap_add() may change zap */
1009 } else if (integer_size != 8 || num_integers != 1 ||
1010 strlen(key) >= MZAP_NAME_LEN) {
1011 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1013 err = fzap_add(zn, integer_size, num_integers, val, tx);
1014 zap = zn->zn_zap; /* fzap_add() may change zap */
1018 err = SET_ERROR(EEXIST);
1020 mzap_addent(zn, *intval);
1023 ASSERT(zap == zn->zn_zap);
1025 if (zap != NULL) /* may be NULL if fzap_add() failed */
1031 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1032 int key_numints, int integer_size, uint64_t num_integers,
1033 const void *val, dmu_tx_t *tx)
1039 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1042 zn = zap_name_alloc_uint64(zap, key, key_numints);
1045 return (SET_ERROR(ENOTSUP));
1047 err = fzap_add(zn, integer_size, num_integers, val, tx);
1048 zap = zn->zn_zap; /* fzap_add() may change zap */
1050 if (zap != NULL) /* may be NULL if fzap_add() failed */
1056 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1057 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1062 const uint64_t *intval = val;
1068 * If there is an old value, it shouldn't change across the
1069 * lockdir (eg, due to bprewrite's xlation).
1071 if (integer_size == 8 && num_integers == 1)
1072 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1075 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1078 zn = zap_name_alloc(zap, name, MT_EXACT);
1081 return (SET_ERROR(ENOTSUP));
1083 if (!zap->zap_ismicro) {
1084 err = fzap_update(zn, integer_size, num_integers, val, tx);
1085 zap = zn->zn_zap; /* fzap_update() may change zap */
1086 } else if (integer_size != 8 || num_integers != 1 ||
1087 strlen(name) >= MZAP_NAME_LEN) {
1088 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1089 zapobj, integer_size, num_integers, name);
1090 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1092 err = fzap_update(zn, integer_size, num_integers,
1094 zap = zn->zn_zap; /* fzap_update() may change zap */
1098 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1099 MZE_PHYS(zap, mze)->mze_value = *intval;
1101 mzap_addent(zn, *intval);
1104 ASSERT(zap == zn->zn_zap);
1106 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1112 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1114 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1120 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1123 zn = zap_name_alloc_uint64(zap, key, key_numints);
1126 return (SET_ERROR(ENOTSUP));
1128 err = fzap_update(zn, integer_size, num_integers, val, tx);
1129 zap = zn->zn_zap; /* fzap_update() may change zap */
1131 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1137 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1139 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1143 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1144 matchtype_t mt, dmu_tx_t *tx)
1151 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1154 zn = zap_name_alloc(zap, name, mt);
1157 return (SET_ERROR(ENOTSUP));
1159 if (!zap->zap_ismicro) {
1160 err = fzap_remove(zn, tx);
1164 err = SET_ERROR(ENOENT);
1166 zap->zap_m.zap_num_entries--;
1167 bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1168 sizeof (mzap_ent_phys_t));
1169 mze_remove(zap, mze);
1178 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1179 int key_numints, dmu_tx_t *tx)
1185 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1188 zn = zap_name_alloc_uint64(zap, key, key_numints);
1191 return (SET_ERROR(ENOTSUP));
1193 err = fzap_remove(zn, tx);
1200 * Routines for iterating over the attributes.
1204 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1205 uint64_t serialized)
1210 zc->zc_zapobj = zapobj;
1211 zc->zc_serialized = serialized;
1217 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1219 zap_cursor_init_serialized(zc, os, zapobj, 0);
1223 zap_cursor_fini(zap_cursor_t *zc)
1226 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1227 zap_unlockdir(zc->zc_zap);
1231 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1232 zap_put_leaf(zc->zc_leaf);
1235 zc->zc_objset = NULL;
1239 zap_cursor_serialize(zap_cursor_t *zc)
1241 if (zc->zc_hash == -1ULL)
1243 if (zc->zc_zap == NULL)
1244 return (zc->zc_serialized);
1245 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1246 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1249 * We want to keep the high 32 bits of the cursor zero if we can, so
1250 * that 32-bit programs can access this. So usually use a small
1251 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1254 * [ collision differentiator | zap_hashbits()-bit hash value ]
1256 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1257 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1261 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1265 mzap_ent_t mze_tofind;
1268 if (zc->zc_hash == -1ULL)
1269 return (SET_ERROR(ENOENT));
1271 if (zc->zc_zap == NULL) {
1273 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1274 RW_READER, TRUE, FALSE, &zc->zc_zap);
1279 * To support zap_cursor_init_serialized, advance, retrieve,
1280 * we must add to the existing zc_cd, which may already
1281 * be 1 due to the zap_cursor_advance.
1283 ASSERT(zc->zc_hash == 0);
1284 hb = zap_hashbits(zc->zc_zap);
1285 zc->zc_hash = zc->zc_serialized << (64 - hb);
1286 zc->zc_cd += zc->zc_serialized >> hb;
1287 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1290 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1292 if (!zc->zc_zap->zap_ismicro) {
1293 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1295 mze_tofind.mze_hash = zc->zc_hash;
1296 mze_tofind.mze_cd = zc->zc_cd;
1298 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1300 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1304 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1305 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1306 za->za_normalization_conflict =
1307 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1308 za->za_integer_length = 8;
1309 za->za_num_integers = 1;
1310 za->za_first_integer = mzep->mze_value;
1311 (void) strcpy(za->za_name, mzep->mze_name);
1312 zc->zc_hash = mze->mze_hash;
1313 zc->zc_cd = mze->mze_cd;
1316 zc->zc_hash = -1ULL;
1317 err = SET_ERROR(ENOENT);
1320 rw_exit(&zc->zc_zap->zap_rwlock);
1325 zap_cursor_advance(zap_cursor_t *zc)
1327 if (zc->zc_hash == -1ULL)
1333 zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1339 if (zc->zc_zap == NULL) {
1340 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1341 RW_READER, TRUE, FALSE, &zc->zc_zap);
1345 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1348 zn = zap_name_alloc(zc->zc_zap, name, mt);
1350 rw_exit(&zc->zc_zap->zap_rwlock);
1351 return (SET_ERROR(ENOTSUP));
1354 if (!zc->zc_zap->zap_ismicro) {
1355 err = fzap_cursor_move_to_key(zc, zn);
1359 err = SET_ERROR(ENOENT);
1362 zc->zc_hash = mze->mze_hash;
1363 zc->zc_cd = mze->mze_cd;
1368 rw_exit(&zc->zc_zap->zap_rwlock);
1373 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1378 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1382 bzero(zs, sizeof (zap_stats_t));
1384 if (zap->zap_ismicro) {
1385 zs->zs_blocksize = zap->zap_dbuf->db_size;
1386 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1387 zs->zs_num_blocks = 1;
1389 fzap_get_stats(zap, zs);
1396 zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
1397 uint64_t *towrite, uint64_t *tooverwrite)
1403 * Since, we don't have a name, we cannot figure out which blocks will
1404 * be affected in this operation. So, account for the worst case :
1405 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1406 * - 4 new blocks written if adding:
1407 * - 2 blocks for possibly split leaves,
1408 * - 2 grown ptrtbl blocks
1410 * This also accomodates the case where an add operation to a fairly
1411 * large microzap results in a promotion to fatzap.
1414 *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE;
1419 * We lock the zap with adding == FALSE. Because, if we pass
1420 * the actual value of add, it could trigger a mzap_upgrade().
1421 * At present we are just evaluating the possibility of this operation
1422 * and hence we donot want to trigger an upgrade.
1424 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1428 if (!zap->zap_ismicro) {
1429 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1431 err = fzap_count_write(zn, add, towrite,
1436 * We treat this case as similar to (name == NULL)
1438 *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE;
1442 * We are here if (name != NULL) and this is a micro-zap.
1443 * We account for the header block depending on whether it
1446 * Incase of an add-operation it is hard to find out
1447 * if this add will promote this microzap to fatzap.
1448 * Hence, we consider the worst case and account for the
1449 * blocks assuming this microzap would be promoted to a
1452 * 1 block overwritten : header block
1453 * 4 new blocks written : 2 new split leaf, 2 grown
1456 if (dmu_buf_freeable(zap->zap_dbuf))
1457 *tooverwrite += MZAP_MAX_BLKSZ;
1459 *towrite += MZAP_MAX_BLKSZ;
1462 *towrite += 4 * MZAP_MAX_BLKSZ;