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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2017 Nexenta Systems, Inc.
33 #include <sys/zfs_context.h>
35 #include <sys/refcount.h>
36 #include <sys/zap_impl.h>
37 #include <sys/zap_leaf.h>
40 #include <sys/dmu_objset.h>
43 #include <sys/sunddi.h>
46 extern inline mzap_phys_t *zap_m_phys(zap_t *zap);
48 static int mzap_upgrade(zap_t **zapp,
49 void *tag, dmu_tx_t *tx, zap_flags_t flags);
52 zap_getflags(zap_t *zap)
56 return (zap_f_phys(zap)->zap_flags);
60 zap_hashbits(zap_t *zap)
62 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
71 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
78 zap_hash(zap_name_t *zn)
80 zap_t *zap = zn->zn_zap;
83 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
84 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
85 h = *(uint64_t *)zn->zn_key_orig;
89 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
91 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
92 const uint64_t *wp = zn->zn_key_norm;
94 ASSERT(zn->zn_key_intlen == 8);
95 for (int i = 0; i < zn->zn_key_norm_numints;
99 for (int j = 0; j < zn->zn_key_intlen; j++) {
101 zfs_crc64_table[(h ^ word) & 0xFF];
106 const uint8_t *cp = zn->zn_key_norm;
109 * We previously stored the terminating null on
110 * disk, but didn't hash it, so we need to
111 * continue to not hash it. (The
112 * zn_key_*_numints includes the terminating
113 * null for non-binary keys.)
115 int len = zn->zn_key_norm_numints - 1;
117 ASSERT(zn->zn_key_intlen == 1);
118 for (int i = 0; i < len; cp++, i++) {
120 zfs_crc64_table[(h ^ *cp) & 0xFF];
125 * Don't use all 64 bits, since we need some in the cookie for
126 * the collision differentiator. We MUST use the high bits,
127 * since those are the ones that we first pay attention to when
128 * chosing the bucket.
130 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
136 zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags)
138 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
140 size_t inlen = strlen(name) + 1;
141 size_t outlen = ZAP_MAXNAMELEN;
144 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
145 normflags | U8_TEXTPREP_IGNORE_NULL | U8_TEXTPREP_IGNORE_INVALID,
146 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_NORMALIZE) {
157 char norm[ZAP_MAXNAMELEN];
159 if (zap_normalize(zn->zn_zap, matchname, norm,
160 zn->zn_normflags) != 0)
163 return (strcmp(zn->zn_key_norm, norm) == 0);
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 zn->zn_normflags = zap->zap_normflags;
188 * If we're dealing with a case sensitive lookup on a mixed or
189 * insensitive fs, remove U8_TEXTPREP_TOUPPER or the lookup
190 * will fold case to all caps overriding the lookup request.
192 if (mt & MT_MATCH_CASE)
193 zn->zn_normflags &= ~U8_TEXTPREP_TOUPPER;
195 if (zap->zap_normflags) {
197 * We *must* use zap_normflags because this normalization is
198 * what the hash is computed from.
200 if (zap_normalize(zap, key, zn->zn_normbuf,
201 zap->zap_normflags) != 0) {
205 zn->zn_key_norm = zn->zn_normbuf;
206 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
212 zn->zn_key_norm = zn->zn_key_orig;
213 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
216 zn->zn_hash = zap_hash(zn);
218 if (zap->zap_normflags != zn->zn_normflags) {
220 * We *must* use zn_normflags because this normalization is
221 * what the matching is based on. (Not the hash!)
223 if (zap_normalize(zap, key, zn->zn_normbuf,
224 zn->zn_normflags) != 0) {
228 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
235 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
237 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
239 ASSERT(zap->zap_normflags == 0);
241 zn->zn_key_intlen = sizeof (*key);
242 zn->zn_key_orig = zn->zn_key_norm = key;
243 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
244 zn->zn_matchtype = 0;
246 zn->zn_hash = zap_hash(zn);
251 mzap_byteswap(mzap_phys_t *buf, size_t size)
253 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
254 buf->mz_salt = BSWAP_64(buf->mz_salt);
255 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
256 int max = (size / MZAP_ENT_LEN) - 1;
257 for (int i = 0; i < max; i++) {
258 buf->mz_chunk[i].mze_value =
259 BSWAP_64(buf->mz_chunk[i].mze_value);
260 buf->mz_chunk[i].mze_cd =
261 BSWAP_32(buf->mz_chunk[i].mze_cd);
266 zap_byteswap(void *buf, size_t size)
268 uint64_t block_type = *(uint64_t *)buf;
270 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
271 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
272 mzap_byteswap(buf, size);
274 fzap_byteswap(buf, size);
279 mze_compare(const void *arg1, const void *arg2)
281 const mzap_ent_t *mze1 = arg1;
282 const mzap_ent_t *mze2 = arg2;
284 int cmp = AVL_CMP(mze1->mze_hash, mze2->mze_hash);
288 return (AVL_CMP(mze1->mze_cd, mze2->mze_cd));
292 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
296 ASSERT(zap->zap_ismicro);
297 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
299 mzap_ent_t *mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
300 mze->mze_chunkid = chunkid;
301 mze->mze_hash = hash;
302 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
303 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
304 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
305 kmem_free(mze, sizeof (mzap_ent_t));
308 avl_insert(&zap->zap_m.zap_avl, mze, idx);
313 mze_find(zap_name_t *zn)
315 mzap_ent_t mze_tofind;
318 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
320 ASSERT(zn->zn_zap->zap_ismicro);
321 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
323 mze_tofind.mze_hash = zn->zn_hash;
324 mze_tofind.mze_cd = 0;
326 mze = avl_find(avl, &mze_tofind, &idx);
328 mze = avl_nearest(avl, idx, AVL_AFTER);
329 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
330 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
331 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
339 mze_find_unused_cd(zap_t *zap, uint64_t hash)
341 mzap_ent_t mze_tofind;
343 avl_tree_t *avl = &zap->zap_m.zap_avl;
345 ASSERT(zap->zap_ismicro);
346 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
348 mze_tofind.mze_hash = hash;
349 mze_tofind.mze_cd = 0;
352 for (mzap_ent_t *mze = avl_find(avl, &mze_tofind, &idx);
353 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
354 if (mze->mze_cd != cd)
363 mze_remove(zap_t *zap, mzap_ent_t *mze)
365 ASSERT(zap->zap_ismicro);
366 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
368 avl_remove(&zap->zap_m.zap_avl, mze);
369 kmem_free(mze, sizeof (mzap_ent_t));
373 mze_destroy(zap_t *zap)
376 void *avlcookie = NULL;
378 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
379 kmem_free(mze, sizeof (mzap_ent_t));
380 avl_destroy(&zap->zap_m.zap_avl);
384 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
387 uint64_t *zap_hdr = (uint64_t *)db->db_data;
388 uint64_t zap_block_type = zap_hdr[0];
389 uint64_t zap_magic = zap_hdr[1];
391 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
393 zap_t *zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
394 rw_init(&zap->zap_rwlock, 0, 0, 0);
395 rw_enter(&zap->zap_rwlock, RW_WRITER);
396 zap->zap_objset = os;
397 zap->zap_object = obj;
400 if (zap_block_type != ZBT_MICRO) {
401 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
402 zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
403 if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
404 winner = NULL; /* No actual winner here... */
408 zap->zap_ismicro = TRUE;
412 * Make sure that zap_ismicro is set before we let others see
413 * it, because zap_lockdir() checks zap_ismicro without the lock
416 dmu_buf_init_user(&zap->zap_dbu, zap_evict_sync, NULL, &zap->zap_dbuf);
417 winner = dmu_buf_set_user(db, &zap->zap_dbu);
422 if (zap->zap_ismicro) {
423 zap->zap_salt = zap_m_phys(zap)->mz_salt;
424 zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
425 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
426 avl_create(&zap->zap_m.zap_avl, mze_compare,
427 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
429 for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
430 mzap_ent_phys_t *mze =
431 &zap_m_phys(zap)->mz_chunk[i];
432 if (mze->mze_name[0]) {
435 zn = zap_name_alloc(zap, mze->mze_name, 0);
436 if (mze_insert(zap, i, zn->zn_hash) == 0)
437 zap->zap_m.zap_num_entries++;
439 printf("ZFS WARNING: Duplicated ZAP "
440 "entry detected (%s).\n",
447 zap->zap_salt = zap_f_phys(zap)->zap_salt;
448 zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
450 ASSERT3U(sizeof (struct zap_leaf_header), ==,
451 2*ZAP_LEAF_CHUNKSIZE);
454 * The embedded pointer table should not overlap the
457 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
458 &zap_f_phys(zap)->zap_salt);
461 * The embedded pointer table should end at the end of
464 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
465 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
466 (uintptr_t)zap_f_phys(zap), ==,
467 zap->zap_dbuf->db_size);
469 rw_exit(&zap->zap_rwlock);
473 rw_exit(&zap->zap_rwlock);
474 rw_destroy(&zap->zap_rwlock);
475 if (!zap->zap_ismicro)
476 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
477 kmem_free(zap, sizeof (zap_t));
482 * This routine "consumes" the caller's hold on the dbuf, which must
483 * have the specified tag.
486 zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx,
487 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
489 ASSERT0(db->db_offset);
490 objset_t *os = dmu_buf_get_objset(db);
491 uint64_t obj = db->db_object;
495 zap_t *zap = dmu_buf_get_user(db);
497 zap = mzap_open(os, obj, db);
500 * mzap_open() didn't like what it saw on-disk.
501 * Check for corruption!
503 return (SET_ERROR(EIO));
508 * We're checking zap_ismicro without the lock held, in order to
509 * tell what type of lock we want. Once we have some sort of
510 * lock, see if it really is the right type. In practice this
511 * can only be different if it was upgraded from micro to fat,
512 * and micro wanted WRITER but fat only needs READER.
514 krw_t lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
515 rw_enter(&zap->zap_rwlock, lt);
516 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
517 /* it was upgraded, now we only need reader */
518 ASSERT(lt == RW_WRITER);
520 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
521 rw_downgrade(&zap->zap_rwlock);
525 zap->zap_objset = os;
528 dmu_buf_will_dirty(db, tx);
530 ASSERT3P(zap->zap_dbuf, ==, db);
532 ASSERT(!zap->zap_ismicro ||
533 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
534 if (zap->zap_ismicro && tx && adding &&
535 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
536 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
537 if (newsz > MZAP_MAX_BLKSZ) {
538 dprintf("upgrading obj %llu: num_entries=%u\n",
539 obj, zap->zap_m.zap_num_entries);
541 int err = mzap_upgrade(zapp, tag, tx, 0);
543 rw_exit(&zap->zap_rwlock);
546 VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx));
547 zap->zap_m.zap_num_chunks =
548 db->db_size / MZAP_ENT_LEN - 1;
556 zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
557 krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
561 int err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
567 dmu_object_info_t doi;
568 dmu_object_info_from_db(db, &doi);
569 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
573 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
575 dmu_buf_rele(db, tag);
581 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
582 krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
586 int err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH);
591 dmu_object_info_t doi;
592 dmu_object_info_from_db(db, &doi);
593 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
596 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
598 dmu_buf_rele(db, tag);
603 zap_unlockdir(zap_t *zap, void *tag)
605 rw_exit(&zap->zap_rwlock);
606 dmu_buf_rele(zap->zap_dbuf, tag);
610 mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags)
615 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
617 int sz = zap->zap_dbuf->db_size;
618 mzap_phys_t *mzp = zio_buf_alloc(sz);
619 bcopy(zap->zap_dbuf->db_data, mzp, sz);
620 int nchunks = zap->zap_m.zap_num_chunks;
623 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
624 1ULL << fzap_default_block_shift, 0, tx);
626 zio_buf_free(mzp, sz);
631 dprintf("upgrading obj=%llu with %u chunks\n",
632 zap->zap_object, nchunks);
633 /* XXX destroy the avl later, so we can use the stored hash value */
636 fzap_upgrade(zap, tx, flags);
638 for (int i = 0; i < nchunks; i++) {
639 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
640 if (mze->mze_name[0] == 0)
642 dprintf("adding %s=%llu\n",
643 mze->mze_name, mze->mze_value);
644 zap_name_t *zn = zap_name_alloc(zap, mze->mze_name, 0);
645 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
647 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
652 zio_buf_free(mzp, sz);
658 * The "normflags" determine the behavior of the matchtype_t which is
659 * passed to zap_lookup_norm(). Names which have the same normalized
660 * version will be stored with the same hash value, and therefore we can
661 * perform normalization-insensitive lookups. We can be Unicode form-
662 * insensitive and/or case-insensitive. The following flags are valid for
669 * U8_TEXTPREP_TOUPPER
671 * The *_NF* (Normalization Form) flags are mutually exclusive; at most one
672 * of them may be supplied.
675 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
680 VERIFY0(dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
682 dmu_buf_will_dirty(db, tx);
683 mzap_phys_t *zp = db->db_data;
684 zp->mz_block_type = ZBT_MICRO;
685 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
686 zp->mz_normflags = normflags;
690 /* Only fat zap supports flags; upgrade immediately. */
691 VERIFY0(zap_lockdir_impl(db, FTAG, tx, RW_WRITER,
692 B_FALSE, B_FALSE, &zap));
693 VERIFY0(mzap_upgrade(&zap, FTAG, tx, flags));
694 zap_unlockdir(zap, FTAG);
696 dmu_buf_rele(db, FTAG);
701 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
702 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
704 return (zap_create_claim_dnsize(os, obj, ot, bonustype, bonuslen,
709 zap_create_claim_dnsize(objset_t *os, uint64_t obj, dmu_object_type_t ot,
710 dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
712 return (zap_create_claim_norm_dnsize(os, obj,
713 0, ot, bonustype, bonuslen, dnodesize, tx));
717 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
718 dmu_object_type_t ot,
719 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
721 return (zap_create_claim_norm_dnsize(os, obj, normflags, ot, bonustype,
726 zap_create_claim_norm_dnsize(objset_t *os, uint64_t obj, int normflags,
727 dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
728 int dnodesize, dmu_tx_t *tx)
732 err = dmu_object_claim_dnsize(os, obj, ot, 0, bonustype, bonuslen,
736 mzap_create_impl(os, obj, normflags, 0, tx);
741 zap_create(objset_t *os, dmu_object_type_t ot,
742 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
744 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
748 zap_create_dnsize(objset_t *os, dmu_object_type_t ot,
749 dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
751 return (zap_create_norm_dnsize(os, 0, ot, bonustype, bonuslen,
756 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
757 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
759 ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
760 return (zap_create_norm_dnsize(os, normflags, ot, bonustype, bonuslen,
765 zap_create_norm_dnsize(objset_t *os, int normflags, dmu_object_type_t ot,
766 dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
768 uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
771 mzap_create_impl(os, obj, normflags, 0, tx);
776 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
777 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
778 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
780 ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
781 return (zap_create_flags_dnsize(os, normflags, flags, ot,
782 leaf_blockshift, indirect_blockshift, bonustype, bonuslen, 0, tx));
786 zap_create_flags_dnsize(objset_t *os, int normflags, zap_flags_t flags,
787 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
788 dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
790 uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
793 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
794 leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
795 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
796 indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
798 VERIFY(dmu_object_set_blocksize(os, obj,
799 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
801 mzap_create_impl(os, obj, normflags, flags, tx);
806 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
809 * dmu_object_free will free the object number and free the
810 * data. Freeing the data will cause our pageout function to be
811 * called, which will destroy our data (zap_leaf_t's and zap_t).
814 return (dmu_object_free(os, zapobj, tx));
818 zap_evict_sync(void *dbu)
822 rw_destroy(&zap->zap_rwlock);
824 if (zap->zap_ismicro)
827 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
829 kmem_free(zap, sizeof (zap_t));
833 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
838 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
841 if (!zap->zap_ismicro) {
842 err = fzap_count(zap, count);
844 *count = zap->zap_m.zap_num_entries;
846 zap_unlockdir(zap, FTAG);
851 * zn may be NULL; if not specified, it will be computed if needed.
852 * See also the comment above zap_entry_normalization_conflict().
855 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
857 int direction = AVL_BEFORE;
858 boolean_t allocdzn = B_FALSE;
860 if (zap->zap_normflags == 0)
864 for (mzap_ent_t *other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
865 other && other->mze_hash == mze->mze_hash;
866 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
869 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
873 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
880 if (direction == AVL_BEFORE) {
881 direction = AVL_AFTER;
891 * Routines for manipulating attributes.
895 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
896 uint64_t integer_size, uint64_t num_integers, void *buf)
898 return (zap_lookup_norm(os, zapobj, name, integer_size,
899 num_integers, buf, 0, NULL, 0, NULL));
903 zap_lookup_impl(zap_t *zap, const char *name,
904 uint64_t integer_size, uint64_t num_integers, void *buf,
905 matchtype_t mt, char *realname, int rn_len,
910 zap_name_t *zn = zap_name_alloc(zap, name, mt);
912 return (SET_ERROR(ENOTSUP));
914 if (!zap->zap_ismicro) {
915 err = fzap_lookup(zn, integer_size, num_integers, buf,
916 realname, rn_len, ncp);
918 mzap_ent_t *mze = mze_find(zn);
920 err = SET_ERROR(ENOENT);
922 if (num_integers < 1) {
923 err = SET_ERROR(EOVERFLOW);
924 } else if (integer_size != 8) {
925 err = SET_ERROR(EINVAL);
928 MZE_PHYS(zap, mze)->mze_value;
929 (void) strlcpy(realname,
930 MZE_PHYS(zap, mze)->mze_name, rn_len);
932 *ncp = mzap_normalization_conflict(zap,
943 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
944 uint64_t integer_size, uint64_t num_integers, void *buf,
945 matchtype_t mt, char *realname, int rn_len,
951 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
954 err = zap_lookup_impl(zap, name, integer_size,
955 num_integers, buf, mt, realname, rn_len, ncp);
956 zap_unlockdir(zap, FTAG);
961 zap_lookup_by_dnode(dnode_t *dn, const char *name,
962 uint64_t integer_size, uint64_t num_integers, void *buf)
964 return (zap_lookup_norm_by_dnode(dn, name, integer_size,
965 num_integers, buf, 0, NULL, 0, NULL));
969 zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
970 uint64_t integer_size, uint64_t num_integers, void *buf,
971 matchtype_t mt, char *realname, int rn_len,
976 int err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
980 err = zap_lookup_impl(zap, name, integer_size,
981 num_integers, buf, mt, realname, rn_len, ncp);
982 zap_unlockdir(zap, FTAG);
987 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
993 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
996 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
998 zap_unlockdir(zap, FTAG);
999 return (SET_ERROR(ENOTSUP));
1004 zap_unlockdir(zap, FTAG);
1009 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1010 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
1015 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1018 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1020 zap_unlockdir(zap, FTAG);
1021 return (SET_ERROR(ENOTSUP));
1024 err = fzap_lookup(zn, integer_size, num_integers, buf,
1027 zap_unlockdir(zap, FTAG);
1032 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
1034 int err = zap_lookup_norm(os, zapobj, name, 0,
1035 0, NULL, 0, NULL, 0, NULL);
1036 if (err == EOVERFLOW || err == EINVAL)
1037 err = 0; /* found, but skipped reading the value */
1042 zap_length(objset_t *os, uint64_t zapobj, const char *name,
1043 uint64_t *integer_size, uint64_t *num_integers)
1048 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1051 zap_name_t *zn = zap_name_alloc(zap, name, 0);
1053 zap_unlockdir(zap, FTAG);
1054 return (SET_ERROR(ENOTSUP));
1056 if (!zap->zap_ismicro) {
1057 err = fzap_length(zn, integer_size, num_integers);
1059 mzap_ent_t *mze = mze_find(zn);
1061 err = SET_ERROR(ENOENT);
1070 zap_unlockdir(zap, FTAG);
1075 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1076 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1081 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1084 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1086 zap_unlockdir(zap, FTAG);
1087 return (SET_ERROR(ENOTSUP));
1089 err = fzap_length(zn, integer_size, num_integers);
1091 zap_unlockdir(zap, FTAG);
1096 mzap_addent(zap_name_t *zn, uint64_t value)
1098 zap_t *zap = zn->zn_zap;
1099 int start = zap->zap_m.zap_alloc_next;
1101 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1104 for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1105 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1106 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1110 uint32_t cd = mze_find_unused_cd(zap, zn->zn_hash);
1111 /* given the limited size of the microzap, this can't happen */
1112 ASSERT(cd < zap_maxcd(zap));
1115 for (int i = start; i < zap->zap_m.zap_num_chunks; i++) {
1116 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1117 if (mze->mze_name[0] == 0) {
1118 mze->mze_value = value;
1120 (void) strcpy(mze->mze_name, zn->zn_key_orig);
1121 zap->zap_m.zap_num_entries++;
1122 zap->zap_m.zap_alloc_next = i+1;
1123 if (zap->zap_m.zap_alloc_next ==
1124 zap->zap_m.zap_num_chunks)
1125 zap->zap_m.zap_alloc_next = 0;
1126 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
1134 ASSERT(!"out of entries!");
1138 zap_add_impl(zap_t *zap, const char *key,
1139 int integer_size, uint64_t num_integers,
1140 const void *val, dmu_tx_t *tx, void *tag)
1142 const uint64_t *intval = val;
1145 zap_name_t *zn = zap_name_alloc(zap, key, 0);
1147 zap_unlockdir(zap, tag);
1148 return (SET_ERROR(ENOTSUP));
1150 if (!zap->zap_ismicro) {
1151 err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1152 zap = zn->zn_zap; /* fzap_add() may change zap */
1153 } else if (integer_size != 8 || num_integers != 1 ||
1154 strlen(key) >= MZAP_NAME_LEN) {
1155 err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
1157 err = fzap_add(zn, integer_size, num_integers, val,
1160 zap = zn->zn_zap; /* fzap_add() may change zap */
1162 if (mze_find(zn) != NULL) {
1163 err = SET_ERROR(EEXIST);
1165 mzap_addent(zn, *intval);
1168 ASSERT(zap == zn->zn_zap);
1170 if (zap != NULL) /* may be NULL if fzap_add() failed */
1171 zap_unlockdir(zap, tag);
1176 zap_add(objset_t *os, uint64_t zapobj, const char *key,
1177 int integer_size, uint64_t num_integers,
1178 const void *val, dmu_tx_t *tx)
1183 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1186 err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1187 /* zap_add_impl() calls zap_unlockdir() */
1192 zap_add_by_dnode(dnode_t *dn, const char *key,
1193 int integer_size, uint64_t num_integers,
1194 const void *val, dmu_tx_t *tx)
1199 err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1202 err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1203 /* zap_add_impl() calls zap_unlockdir() */
1208 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1209 int key_numints, int integer_size, uint64_t num_integers,
1210 const void *val, dmu_tx_t *tx)
1215 zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1218 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1220 zap_unlockdir(zap, FTAG);
1221 return (SET_ERROR(ENOTSUP));
1223 err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1224 zap = zn->zn_zap; /* fzap_add() may change zap */
1226 if (zap != NULL) /* may be NULL if fzap_add() failed */
1227 zap_unlockdir(zap, FTAG);
1232 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1233 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1237 const uint64_t *intval = val;
1241 * If there is an old value, it shouldn't change across the
1242 * lockdir (eg, due to bprewrite's xlation).
1244 if (integer_size == 8 && num_integers == 1)
1245 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1249 zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1252 zap_name_t *zn = zap_name_alloc(zap, name, 0);
1254 zap_unlockdir(zap, FTAG);
1255 return (SET_ERROR(ENOTSUP));
1257 if (!zap->zap_ismicro) {
1258 err = fzap_update(zn, integer_size, num_integers, val,
1260 zap = zn->zn_zap; /* fzap_update() may change zap */
1261 } else if (integer_size != 8 || num_integers != 1 ||
1262 strlen(name) >= MZAP_NAME_LEN) {
1263 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1264 zapobj, integer_size, num_integers, name);
1265 err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1267 err = fzap_update(zn, integer_size, num_integers,
1270 zap = zn->zn_zap; /* fzap_update() may change zap */
1272 mzap_ent_t *mze = mze_find(zn);
1274 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1275 MZE_PHYS(zap, mze)->mze_value = *intval;
1277 mzap_addent(zn, *intval);
1280 ASSERT(zap == zn->zn_zap);
1282 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1283 zap_unlockdir(zap, FTAG);
1288 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1290 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1295 zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1298 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1300 zap_unlockdir(zap, FTAG);
1301 return (SET_ERROR(ENOTSUP));
1303 err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx);
1304 zap = zn->zn_zap; /* fzap_update() may change zap */
1306 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1307 zap_unlockdir(zap, FTAG);
1312 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1314 return (zap_remove_norm(os, zapobj, name, 0, tx));
1318 zap_remove_impl(zap_t *zap, const char *name,
1319 matchtype_t mt, dmu_tx_t *tx)
1323 zap_name_t *zn = zap_name_alloc(zap, name, mt);
1325 return (SET_ERROR(ENOTSUP));
1326 if (!zap->zap_ismicro) {
1327 err = fzap_remove(zn, tx);
1329 mzap_ent_t *mze = mze_find(zn);
1331 err = SET_ERROR(ENOENT);
1333 zap->zap_m.zap_num_entries--;
1334 bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1335 sizeof (mzap_ent_phys_t));
1336 mze_remove(zap, mze);
1344 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1345 matchtype_t mt, dmu_tx_t *tx)
1350 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1353 err = zap_remove_impl(zap, name, mt, tx);
1354 zap_unlockdir(zap, FTAG);
1359 zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
1364 err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1367 err = zap_remove_impl(zap, name, 0, tx);
1368 zap_unlockdir(zap, FTAG);
1373 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1374 int key_numints, dmu_tx_t *tx)
1379 zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1382 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1384 zap_unlockdir(zap, FTAG);
1385 return (SET_ERROR(ENOTSUP));
1387 err = fzap_remove(zn, tx);
1389 zap_unlockdir(zap, FTAG);
1394 * Routines for iterating over the attributes.
1398 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1399 uint64_t serialized)
1404 zc->zc_zapobj = zapobj;
1405 zc->zc_serialized = serialized;
1411 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1413 zap_cursor_init_serialized(zc, os, zapobj, 0);
1417 zap_cursor_fini(zap_cursor_t *zc)
1420 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1421 zap_unlockdir(zc->zc_zap, NULL);
1425 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1426 zap_put_leaf(zc->zc_leaf);
1429 zc->zc_objset = NULL;
1433 zap_cursor_serialize(zap_cursor_t *zc)
1435 if (zc->zc_hash == -1ULL)
1437 if (zc->zc_zap == NULL)
1438 return (zc->zc_serialized);
1439 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1440 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1443 * We want to keep the high 32 bits of the cursor zero if we can, so
1444 * that 32-bit programs can access this. So usually use a small
1445 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1448 * [ collision differentiator | zap_hashbits()-bit hash value ]
1450 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1451 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1455 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1459 if (zc->zc_hash == -1ULL)
1460 return (SET_ERROR(ENOENT));
1462 if (zc->zc_zap == NULL) {
1464 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1465 RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1470 * To support zap_cursor_init_serialized, advance, retrieve,
1471 * we must add to the existing zc_cd, which may already
1472 * be 1 due to the zap_cursor_advance.
1474 ASSERT(zc->zc_hash == 0);
1475 hb = zap_hashbits(zc->zc_zap);
1476 zc->zc_hash = zc->zc_serialized << (64 - hb);
1477 zc->zc_cd += zc->zc_serialized >> hb;
1478 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1481 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1483 if (!zc->zc_zap->zap_ismicro) {
1484 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1487 mzap_ent_t mze_tofind;
1489 mze_tofind.mze_hash = zc->zc_hash;
1490 mze_tofind.mze_cd = zc->zc_cd;
1493 avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1495 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1499 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1500 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1501 za->za_normalization_conflict =
1502 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1503 za->za_integer_length = 8;
1504 za->za_num_integers = 1;
1505 za->za_first_integer = mzep->mze_value;
1506 (void) strcpy(za->za_name, mzep->mze_name);
1507 zc->zc_hash = mze->mze_hash;
1508 zc->zc_cd = mze->mze_cd;
1511 zc->zc_hash = -1ULL;
1512 err = SET_ERROR(ENOENT);
1515 rw_exit(&zc->zc_zap->zap_rwlock);
1520 zap_cursor_advance(zap_cursor_t *zc)
1522 if (zc->zc_hash == -1ULL)
1528 zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1534 if (zc->zc_zap == NULL) {
1535 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1536 RW_READER, TRUE, FALSE, FTAG, &zc->zc_zap);
1540 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1543 zn = zap_name_alloc(zc->zc_zap, name, mt);
1545 rw_exit(&zc->zc_zap->zap_rwlock);
1546 return (SET_ERROR(ENOTSUP));
1549 if (!zc->zc_zap->zap_ismicro) {
1550 err = fzap_cursor_move_to_key(zc, zn);
1554 err = SET_ERROR(ENOENT);
1557 zc->zc_hash = mze->mze_hash;
1558 zc->zc_cd = mze->mze_cd;
1563 rw_exit(&zc->zc_zap->zap_rwlock);
1568 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1573 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1577 bzero(zs, sizeof (zap_stats_t));
1579 if (zap->zap_ismicro) {
1580 zs->zs_blocksize = zap->zap_dbuf->db_size;
1581 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1582 zs->zs_num_blocks = 1;
1584 fzap_get_stats(zap, zs);
1586 zap_unlockdir(zap, FTAG);