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, 2016 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) {
93 const uint64_t *wp = zn->zn_key_norm;
95 ASSERT(zn->zn_key_intlen == 8);
96 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
100 for (j = 0; j < zn->zn_key_intlen; j++) {
102 zfs_crc64_table[(h ^ word) & 0xFF];
108 const uint8_t *cp = zn->zn_key_norm;
111 * We previously stored the terminating null on
112 * disk, but didn't hash it, so we need to
113 * continue to not hash it. (The
114 * zn_key_*_numints includes the terminating
115 * null for non-binary keys.)
117 len = zn->zn_key_norm_numints - 1;
119 ASSERT(zn->zn_key_intlen == 1);
120 for (i = 0; i < len; cp++, i++) {
122 zfs_crc64_table[(h ^ *cp) & 0xFF];
127 * Don't use all 64 bits, since we need some in the cookie for
128 * the collision differentiator. We MUST use the high bits,
129 * since those are the ones that we first pay attention to when
130 * chosing the bucket.
132 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
138 zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags)
140 size_t inlen, outlen;
143 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
145 inlen = strlen(name) + 1;
146 outlen = ZAP_MAXNAMELEN;
149 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
150 normflags | U8_TEXTPREP_IGNORE_NULL | U8_TEXTPREP_IGNORE_INVALID,
151 U8_UNICODE_LATEST, &err);
157 zap_match(zap_name_t *zn, const char *matchname)
159 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
161 if (zn->zn_matchtype & MT_NORMALIZE) {
162 char norm[ZAP_MAXNAMELEN];
164 if (zap_normalize(zn->zn_zap, matchname, norm,
165 zn->zn_normflags) != 0)
168 return (strcmp(zn->zn_key_norm, norm) == 0);
170 return (strcmp(zn->zn_key_orig, matchname) == 0);
175 zap_name_free(zap_name_t *zn)
177 kmem_free(zn, sizeof (zap_name_t));
181 zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
183 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
186 zn->zn_key_intlen = sizeof (*key);
187 zn->zn_key_orig = key;
188 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
189 zn->zn_matchtype = mt;
190 zn->zn_normflags = zap->zap_normflags;
193 * If we're dealing with a case sensitive lookup on a mixed or
194 * insensitive fs, remove U8_TEXTPREP_TOUPPER or the lookup
195 * will fold case to all caps overriding the lookup request.
197 if (mt & MT_MATCH_CASE)
198 zn->zn_normflags &= ~U8_TEXTPREP_TOUPPER;
200 if (zap->zap_normflags) {
202 * We *must* use zap_normflags because this normalization is
203 * what the hash is computed from.
205 if (zap_normalize(zap, key, zn->zn_normbuf,
206 zap->zap_normflags) != 0) {
210 zn->zn_key_norm = zn->zn_normbuf;
211 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
217 zn->zn_key_norm = zn->zn_key_orig;
218 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
221 zn->zn_hash = zap_hash(zn);
223 if (zap->zap_normflags != zn->zn_normflags) {
225 * We *must* use zn_normflags because this normalization is
226 * what the matching is based on. (Not the hash!)
228 if (zap_normalize(zap, key, zn->zn_normbuf,
229 zn->zn_normflags) != 0) {
233 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
240 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
242 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
244 ASSERT(zap->zap_normflags == 0);
246 zn->zn_key_intlen = sizeof (*key);
247 zn->zn_key_orig = zn->zn_key_norm = key;
248 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
249 zn->zn_matchtype = 0;
251 zn->zn_hash = zap_hash(zn);
256 mzap_byteswap(mzap_phys_t *buf, size_t size)
259 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
260 buf->mz_salt = BSWAP_64(buf->mz_salt);
261 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
262 max = (size / MZAP_ENT_LEN) - 1;
263 for (i = 0; i < max; i++) {
264 buf->mz_chunk[i].mze_value =
265 BSWAP_64(buf->mz_chunk[i].mze_value);
266 buf->mz_chunk[i].mze_cd =
267 BSWAP_32(buf->mz_chunk[i].mze_cd);
272 zap_byteswap(void *buf, size_t size)
276 block_type = *(uint64_t *)buf;
278 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
279 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
280 mzap_byteswap(buf, size);
282 fzap_byteswap(buf, size);
287 mze_compare(const void *arg1, const void *arg2)
289 const mzap_ent_t *mze1 = arg1;
290 const mzap_ent_t *mze2 = arg2;
292 if (mze1->mze_hash > mze2->mze_hash)
294 if (mze1->mze_hash < mze2->mze_hash)
296 if (mze1->mze_cd > mze2->mze_cd)
298 if (mze1->mze_cd < mze2->mze_cd)
304 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
309 ASSERT(zap->zap_ismicro);
310 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
312 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
313 mze->mze_chunkid = chunkid;
314 mze->mze_hash = hash;
315 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
316 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
317 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
318 kmem_free(mze, sizeof (mzap_ent_t));
321 avl_insert(&zap->zap_m.zap_avl, mze, idx);
326 mze_find(zap_name_t *zn)
328 mzap_ent_t mze_tofind;
331 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
333 ASSERT(zn->zn_zap->zap_ismicro);
334 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
336 mze_tofind.mze_hash = zn->zn_hash;
337 mze_tofind.mze_cd = 0;
339 mze = avl_find(avl, &mze_tofind, &idx);
341 mze = avl_nearest(avl, idx, AVL_AFTER);
342 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
343 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
344 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
352 mze_find_unused_cd(zap_t *zap, uint64_t hash)
354 mzap_ent_t mze_tofind;
357 avl_tree_t *avl = &zap->zap_m.zap_avl;
360 ASSERT(zap->zap_ismicro);
361 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
363 mze_tofind.mze_hash = hash;
364 mze_tofind.mze_cd = 0;
367 for (mze = avl_find(avl, &mze_tofind, &idx);
368 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
369 if (mze->mze_cd != cd)
378 mze_remove(zap_t *zap, mzap_ent_t *mze)
380 ASSERT(zap->zap_ismicro);
381 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
383 avl_remove(&zap->zap_m.zap_avl, mze);
384 kmem_free(mze, sizeof (mzap_ent_t));
388 mze_destroy(zap_t *zap)
391 void *avlcookie = NULL;
393 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
394 kmem_free(mze, sizeof (mzap_ent_t));
395 avl_destroy(&zap->zap_m.zap_avl);
399 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
404 uint64_t *zap_hdr = (uint64_t *)db->db_data;
405 uint64_t zap_block_type = zap_hdr[0];
406 uint64_t zap_magic = zap_hdr[1];
408 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
410 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
411 rw_init(&zap->zap_rwlock, 0, 0, 0);
412 rw_enter(&zap->zap_rwlock, RW_WRITER);
413 zap->zap_objset = os;
414 zap->zap_object = obj;
417 if (zap_block_type != ZBT_MICRO) {
418 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
419 zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
420 if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
421 winner = NULL; /* No actual winner here... */
425 zap->zap_ismicro = TRUE;
429 * Make sure that zap_ismicro is set before we let others see
430 * it, because zap_lockdir() checks zap_ismicro without the lock
433 dmu_buf_init_user(&zap->zap_dbu, zap_evict_sync, NULL, &zap->zap_dbuf);
434 winner = dmu_buf_set_user(db, &zap->zap_dbu);
439 if (zap->zap_ismicro) {
440 zap->zap_salt = zap_m_phys(zap)->mz_salt;
441 zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
442 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
443 avl_create(&zap->zap_m.zap_avl, mze_compare,
444 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
446 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
447 mzap_ent_phys_t *mze =
448 &zap_m_phys(zap)->mz_chunk[i];
449 if (mze->mze_name[0]) {
452 zn = zap_name_alloc(zap, mze->mze_name, 0);
453 if (mze_insert(zap, i, zn->zn_hash) == 0)
454 zap->zap_m.zap_num_entries++;
456 printf("ZFS WARNING: Duplicated ZAP "
457 "entry detected (%s).\n",
464 zap->zap_salt = zap_f_phys(zap)->zap_salt;
465 zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
467 ASSERT3U(sizeof (struct zap_leaf_header), ==,
468 2*ZAP_LEAF_CHUNKSIZE);
471 * The embedded pointer table should not overlap the
474 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
475 &zap_f_phys(zap)->zap_salt);
478 * The embedded pointer table should end at the end of
481 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
482 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
483 (uintptr_t)zap_f_phys(zap), ==,
484 zap->zap_dbuf->db_size);
486 rw_exit(&zap->zap_rwlock);
490 rw_exit(&zap->zap_rwlock);
491 rw_destroy(&zap->zap_rwlock);
492 if (!zap->zap_ismicro)
493 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
494 kmem_free(zap, sizeof (zap_t));
499 zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx,
500 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
505 ASSERT0(db->db_offset);
506 objset_t *os = dmu_buf_get_objset(db);
507 uint64_t obj = db->db_object;
513 dmu_object_info_t doi;
514 dmu_object_info_from_db(db, &doi);
515 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
519 zap = dmu_buf_get_user(db);
521 zap = mzap_open(os, obj, db);
524 * mzap_open() didn't like what it saw on-disk.
525 * Check for corruption!
527 return (SET_ERROR(EIO));
532 * We're checking zap_ismicro without the lock held, in order to
533 * tell what type of lock we want. Once we have some sort of
534 * lock, see if it really is the right type. In practice this
535 * can only be different if it was upgraded from micro to fat,
536 * and micro wanted WRITER but fat only needs READER.
538 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
539 rw_enter(&zap->zap_rwlock, lt);
540 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
541 /* it was upgraded, now we only need reader */
542 ASSERT(lt == RW_WRITER);
544 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
545 rw_downgrade(&zap->zap_rwlock);
549 zap->zap_objset = os;
552 dmu_buf_will_dirty(db, tx);
554 ASSERT3P(zap->zap_dbuf, ==, db);
556 ASSERT(!zap->zap_ismicro ||
557 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
558 if (zap->zap_ismicro && tx && adding &&
559 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
560 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
561 if (newsz > MZAP_MAX_BLKSZ) {
562 dprintf("upgrading obj %llu: num_entries=%u\n",
563 obj, zap->zap_m.zap_num_entries);
565 int err = mzap_upgrade(zapp, tag, tx, 0);
567 rw_exit(&zap->zap_rwlock);
570 VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx));
571 zap->zap_m.zap_num_chunks =
572 db->db_size / MZAP_ENT_LEN - 1;
580 zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
581 krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
586 err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
590 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
592 dmu_buf_rele(db, tag);
598 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
599 krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
604 err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH);
607 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
609 dmu_buf_rele(db, tag);
614 zap_unlockdir(zap_t *zap, void *tag)
616 rw_exit(&zap->zap_rwlock);
617 dmu_buf_rele(zap->zap_dbuf, tag);
621 mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags)
628 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
630 sz = zap->zap_dbuf->db_size;
631 mzp = zio_buf_alloc(sz);
632 bcopy(zap->zap_dbuf->db_data, mzp, sz);
633 nchunks = zap->zap_m.zap_num_chunks;
636 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
637 1ULL << fzap_default_block_shift, 0, tx);
639 zio_buf_free(mzp, sz);
644 dprintf("upgrading obj=%llu with %u chunks\n",
645 zap->zap_object, nchunks);
646 /* XXX destroy the avl later, so we can use the stored hash value */
649 fzap_upgrade(zap, tx, flags);
651 for (i = 0; i < nchunks; i++) {
652 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
654 if (mze->mze_name[0] == 0)
656 dprintf("adding %s=%llu\n",
657 mze->mze_name, mze->mze_value);
658 zn = zap_name_alloc(zap, mze->mze_name, 0);
659 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
661 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
666 zio_buf_free(mzp, sz);
672 * The "normflags" determine the behavior of the matchtype_t which is
673 * passed to zap_lookup_norm(). Names which have the same normalized
674 * version will be stored with the same hash value, and therefore we can
675 * perform normalization-insensitive lookups. We can be Unicode form-
676 * insensitive and/or case-insensitive. The following flags are valid for
683 * U8_TEXTPREP_TOUPPER
685 * The *_NF* (Normalization Form) flags are mutually exclusive; at most one
686 * of them may be supplied.
689 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
695 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
699 dmu_object_info_t doi;
700 dmu_object_info_from_db(db, &doi);
701 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
705 dmu_buf_will_dirty(db, tx);
707 zp->mz_block_type = ZBT_MICRO;
708 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
709 zp->mz_normflags = normflags;
710 dmu_buf_rele(db, FTAG);
714 /* Only fat zap supports flags; upgrade immediately. */
715 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
716 B_FALSE, B_FALSE, FTAG, &zap));
717 VERIFY3U(0, ==, mzap_upgrade(&zap, FTAG, tx, flags));
718 zap_unlockdir(zap, FTAG);
723 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
724 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
726 return (zap_create_claim_norm(os, obj,
727 0, ot, bonustype, bonuslen, tx));
731 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
732 dmu_object_type_t ot,
733 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
737 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
740 mzap_create_impl(os, obj, normflags, 0, tx);
745 zap_create(objset_t *os, dmu_object_type_t ot,
746 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
748 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
752 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
753 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
755 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
757 mzap_create_impl(os, obj, normflags, 0, tx);
762 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
763 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
764 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
766 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
768 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
769 leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
770 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
771 indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
773 VERIFY(dmu_object_set_blocksize(os, obj,
774 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
776 mzap_create_impl(os, obj, normflags, flags, tx);
781 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
784 * dmu_object_free will free the object number and free the
785 * data. Freeing the data will cause our pageout function to be
786 * called, which will destroy our data (zap_leaf_t's and zap_t).
789 return (dmu_object_free(os, zapobj, tx));
793 zap_evict_sync(void *dbu)
797 rw_destroy(&zap->zap_rwlock);
799 if (zap->zap_ismicro)
802 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
804 kmem_free(zap, sizeof (zap_t));
808 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
813 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
816 if (!zap->zap_ismicro) {
817 err = fzap_count(zap, count);
819 *count = zap->zap_m.zap_num_entries;
821 zap_unlockdir(zap, FTAG);
826 * zn may be NULL; if not specified, it will be computed if needed.
827 * See also the comment above zap_entry_normalization_conflict().
830 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
833 int direction = AVL_BEFORE;
834 boolean_t allocdzn = B_FALSE;
836 if (zap->zap_normflags == 0)
840 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
841 other && other->mze_hash == mze->mze_hash;
842 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
845 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
849 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
856 if (direction == AVL_BEFORE) {
857 direction = AVL_AFTER;
867 * Routines for manipulating attributes.
871 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
872 uint64_t integer_size, uint64_t num_integers, void *buf)
874 return (zap_lookup_norm(os, zapobj, name, integer_size,
875 num_integers, buf, 0, NULL, 0, NULL));
879 zap_lookup_impl(zap_t *zap, const char *name,
880 uint64_t integer_size, uint64_t num_integers, void *buf,
881 matchtype_t mt, char *realname, int rn_len,
888 zn = zap_name_alloc(zap, name, mt);
890 return (SET_ERROR(ENOTSUP));
892 if (!zap->zap_ismicro) {
893 err = fzap_lookup(zn, integer_size, num_integers, buf,
894 realname, rn_len, ncp);
898 err = SET_ERROR(ENOENT);
900 if (num_integers < 1) {
901 err = SET_ERROR(EOVERFLOW);
902 } else if (integer_size != 8) {
903 err = SET_ERROR(EINVAL);
906 MZE_PHYS(zap, mze)->mze_value;
907 (void) strlcpy(realname,
908 MZE_PHYS(zap, mze)->mze_name, rn_len);
910 *ncp = mzap_normalization_conflict(zap,
921 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
922 uint64_t integer_size, uint64_t num_integers, void *buf,
923 matchtype_t mt, char *realname, int rn_len,
929 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
932 err = zap_lookup_impl(zap, name, integer_size,
933 num_integers, buf, mt, realname, rn_len, ncp);
934 zap_unlockdir(zap, FTAG);
939 zap_lookup_by_dnode(dnode_t *dn, const char *name,
940 uint64_t integer_size, uint64_t num_integers, void *buf)
942 return (zap_lookup_norm_by_dnode(dn, name, integer_size,
943 num_integers, buf, 0, NULL, 0, NULL));
947 zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
948 uint64_t integer_size, uint64_t num_integers, void *buf,
949 matchtype_t mt, char *realname, int rn_len,
955 err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
959 err = zap_lookup_impl(zap, name, integer_size,
960 num_integers, buf, mt, realname, rn_len, ncp);
961 zap_unlockdir(zap, FTAG);
966 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
973 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
976 zn = zap_name_alloc_uint64(zap, key, key_numints);
978 zap_unlockdir(zap, FTAG);
979 return (SET_ERROR(ENOTSUP));
984 zap_unlockdir(zap, FTAG);
989 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
990 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
996 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
999 zn = zap_name_alloc_uint64(zap, key, key_numints);
1001 zap_unlockdir(zap, FTAG);
1002 return (SET_ERROR(ENOTSUP));
1005 err = fzap_lookup(zn, integer_size, num_integers, buf,
1008 zap_unlockdir(zap, FTAG);
1013 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
1015 int err = zap_lookup_norm(os, zapobj, name, 0,
1016 0, NULL, 0, NULL, 0, NULL);
1017 if (err == EOVERFLOW || err == EINVAL)
1018 err = 0; /* found, but skipped reading the value */
1023 zap_length(objset_t *os, uint64_t zapobj, const char *name,
1024 uint64_t *integer_size, uint64_t *num_integers)
1031 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1034 zn = zap_name_alloc(zap, name, 0);
1036 zap_unlockdir(zap, FTAG);
1037 return (SET_ERROR(ENOTSUP));
1039 if (!zap->zap_ismicro) {
1040 err = fzap_length(zn, integer_size, num_integers);
1044 err = SET_ERROR(ENOENT);
1053 zap_unlockdir(zap, FTAG);
1058 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1059 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1065 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1068 zn = zap_name_alloc_uint64(zap, key, key_numints);
1070 zap_unlockdir(zap, FTAG);
1071 return (SET_ERROR(ENOTSUP));
1073 err = fzap_length(zn, integer_size, num_integers);
1075 zap_unlockdir(zap, FTAG);
1080 mzap_addent(zap_name_t *zn, uint64_t value)
1083 zap_t *zap = zn->zn_zap;
1084 int start = zap->zap_m.zap_alloc_next;
1087 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1090 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1091 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1092 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1096 cd = mze_find_unused_cd(zap, zn->zn_hash);
1097 /* given the limited size of the microzap, this can't happen */
1098 ASSERT(cd < zap_maxcd(zap));
1101 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
1102 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1103 if (mze->mze_name[0] == 0) {
1104 mze->mze_value = value;
1106 (void) strcpy(mze->mze_name, zn->zn_key_orig);
1107 zap->zap_m.zap_num_entries++;
1108 zap->zap_m.zap_alloc_next = i+1;
1109 if (zap->zap_m.zap_alloc_next ==
1110 zap->zap_m.zap_num_chunks)
1111 zap->zap_m.zap_alloc_next = 0;
1112 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
1120 ASSERT(!"out of entries!");
1124 zap_add_impl(zap_t *zap, const char *key,
1125 int integer_size, uint64_t num_integers,
1126 const void *val, dmu_tx_t *tx, void *tag)
1130 const uint64_t *intval = val;
1133 zn = zap_name_alloc(zap, key, 0);
1135 zap_unlockdir(zap, tag);
1136 return (SET_ERROR(ENOTSUP));
1138 if (!zap->zap_ismicro) {
1139 err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1140 zap = zn->zn_zap; /* fzap_add() may change zap */
1141 } else if (integer_size != 8 || num_integers != 1 ||
1142 strlen(key) >= MZAP_NAME_LEN) {
1143 err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
1145 err = fzap_add(zn, integer_size, num_integers, val,
1148 zap = zn->zn_zap; /* fzap_add() may change zap */
1152 err = SET_ERROR(EEXIST);
1154 mzap_addent(zn, *intval);
1157 ASSERT(zap == zn->zn_zap);
1159 if (zap != NULL) /* may be NULL if fzap_add() failed */
1160 zap_unlockdir(zap, tag);
1165 zap_add(objset_t *os, uint64_t zapobj, const char *key,
1166 int integer_size, uint64_t num_integers,
1167 const void *val, dmu_tx_t *tx)
1172 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1175 err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1176 /* zap_add_impl() calls zap_unlockdir() */
1181 zap_add_by_dnode(dnode_t *dn, const char *key,
1182 int integer_size, uint64_t num_integers,
1183 const void *val, dmu_tx_t *tx)
1188 err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1191 err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1192 /* zap_add_impl() calls zap_unlockdir() */
1197 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1198 int key_numints, int integer_size, uint64_t num_integers,
1199 const void *val, dmu_tx_t *tx)
1205 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1208 zn = zap_name_alloc_uint64(zap, key, key_numints);
1210 zap_unlockdir(zap, FTAG);
1211 return (SET_ERROR(ENOTSUP));
1213 err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1214 zap = zn->zn_zap; /* fzap_add() may change zap */
1216 if (zap != NULL) /* may be NULL if fzap_add() failed */
1217 zap_unlockdir(zap, FTAG);
1222 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1223 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1228 const uint64_t *intval = val;
1234 * If there is an old value, it shouldn't change across the
1235 * lockdir (eg, due to bprewrite's xlation).
1237 if (integer_size == 8 && num_integers == 1)
1238 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1241 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1244 zn = zap_name_alloc(zap, name, 0);
1246 zap_unlockdir(zap, FTAG);
1247 return (SET_ERROR(ENOTSUP));
1249 if (!zap->zap_ismicro) {
1250 err = fzap_update(zn, integer_size, num_integers, val,
1252 zap = zn->zn_zap; /* fzap_update() may change zap */
1253 } else if (integer_size != 8 || num_integers != 1 ||
1254 strlen(name) >= MZAP_NAME_LEN) {
1255 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1256 zapobj, integer_size, num_integers, name);
1257 err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1259 err = fzap_update(zn, integer_size, num_integers,
1262 zap = zn->zn_zap; /* fzap_update() may change zap */
1266 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1267 MZE_PHYS(zap, mze)->mze_value = *intval;
1269 mzap_addent(zn, *intval);
1272 ASSERT(zap == zn->zn_zap);
1274 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1275 zap_unlockdir(zap, FTAG);
1280 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1282 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1288 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1291 zn = zap_name_alloc_uint64(zap, key, key_numints);
1293 zap_unlockdir(zap, FTAG);
1294 return (SET_ERROR(ENOTSUP));
1296 err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx);
1297 zap = zn->zn_zap; /* fzap_update() may change zap */
1299 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1300 zap_unlockdir(zap, FTAG);
1305 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1307 return (zap_remove_norm(os, zapobj, name, 0, tx));
1311 zap_remove_impl(zap_t *zap, const char *name,
1312 matchtype_t mt, dmu_tx_t *tx)
1318 zn = zap_name_alloc(zap, name, mt);
1320 return (SET_ERROR(ENOTSUP));
1321 if (!zap->zap_ismicro) {
1322 err = fzap_remove(zn, tx);
1326 err = SET_ERROR(ENOENT);
1328 zap->zap_m.zap_num_entries--;
1329 bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1330 sizeof (mzap_ent_phys_t));
1331 mze_remove(zap, mze);
1339 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1340 matchtype_t mt, dmu_tx_t *tx)
1345 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1348 err = zap_remove_impl(zap, name, mt, tx);
1349 zap_unlockdir(zap, FTAG);
1354 zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
1359 err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1362 err = zap_remove_impl(zap, name, 0, tx);
1363 zap_unlockdir(zap, FTAG);
1368 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1369 int key_numints, dmu_tx_t *tx)
1375 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1378 zn = zap_name_alloc_uint64(zap, key, key_numints);
1380 zap_unlockdir(zap, FTAG);
1381 return (SET_ERROR(ENOTSUP));
1383 err = fzap_remove(zn, tx);
1385 zap_unlockdir(zap, FTAG);
1390 * Routines for iterating over the attributes.
1394 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1395 uint64_t serialized)
1400 zc->zc_zapobj = zapobj;
1401 zc->zc_serialized = serialized;
1407 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1409 zap_cursor_init_serialized(zc, os, zapobj, 0);
1413 zap_cursor_fini(zap_cursor_t *zc)
1416 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1417 zap_unlockdir(zc->zc_zap, NULL);
1421 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1422 zap_put_leaf(zc->zc_leaf);
1425 zc->zc_objset = NULL;
1429 zap_cursor_serialize(zap_cursor_t *zc)
1431 if (zc->zc_hash == -1ULL)
1433 if (zc->zc_zap == NULL)
1434 return (zc->zc_serialized);
1435 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1436 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1439 * We want to keep the high 32 bits of the cursor zero if we can, so
1440 * that 32-bit programs can access this. So usually use a small
1441 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1444 * [ collision differentiator | zap_hashbits()-bit hash value ]
1446 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1447 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1451 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1455 mzap_ent_t mze_tofind;
1458 if (zc->zc_hash == -1ULL)
1459 return (SET_ERROR(ENOENT));
1461 if (zc->zc_zap == NULL) {
1463 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1464 RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1469 * To support zap_cursor_init_serialized, advance, retrieve,
1470 * we must add to the existing zc_cd, which may already
1471 * be 1 due to the zap_cursor_advance.
1473 ASSERT(zc->zc_hash == 0);
1474 hb = zap_hashbits(zc->zc_zap);
1475 zc->zc_hash = zc->zc_serialized << (64 - hb);
1476 zc->zc_cd += zc->zc_serialized >> hb;
1477 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1480 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1482 if (!zc->zc_zap->zap_ismicro) {
1483 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1485 mze_tofind.mze_hash = zc->zc_hash;
1486 mze_tofind.mze_cd = zc->zc_cd;
1488 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1490 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1494 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1495 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1496 za->za_normalization_conflict =
1497 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1498 za->za_integer_length = 8;
1499 za->za_num_integers = 1;
1500 za->za_first_integer = mzep->mze_value;
1501 (void) strcpy(za->za_name, mzep->mze_name);
1502 zc->zc_hash = mze->mze_hash;
1503 zc->zc_cd = mze->mze_cd;
1506 zc->zc_hash = -1ULL;
1507 err = SET_ERROR(ENOENT);
1510 rw_exit(&zc->zc_zap->zap_rwlock);
1515 zap_cursor_advance(zap_cursor_t *zc)
1517 if (zc->zc_hash == -1ULL)
1523 zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1529 if (zc->zc_zap == NULL) {
1530 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1531 RW_READER, TRUE, FALSE, FTAG, &zc->zc_zap);
1535 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1538 zn = zap_name_alloc(zc->zc_zap, name, mt);
1540 rw_exit(&zc->zc_zap->zap_rwlock);
1541 return (SET_ERROR(ENOTSUP));
1544 if (!zc->zc_zap->zap_ismicro) {
1545 err = fzap_cursor_move_to_key(zc, zn);
1549 err = SET_ERROR(ENOENT);
1552 zc->zc_hash = mze->mze_hash;
1553 zc->zc_cd = mze->mze_cd;
1558 rw_exit(&zc->zc_zap->zap_rwlock);
1563 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1568 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1572 bzero(zs, sizeof (zap_stats_t));
1574 if (zap->zap_ismicro) {
1575 zs->zs_blocksize = zap->zap_dbuf->db_size;
1576 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1577 zs->zs_num_blocks = 1;
1579 fzap_get_stats(zap, zs);
1581 zap_unlockdir(zap, FTAG);