1 /*
   2  * CDDL HEADER START
   3  *
   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.
   7  *
   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.
  12  *
  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]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2013 by Delphix. All rights reserved.
  24  */
  25 
  26 #include <sys/zio.h>
  27 #include <sys/spa.h>
  28 #include <sys/dmu.h>
  29 #include <sys/zfs_context.h>
  30 #include <sys/zap.h>
  31 #include <sys/refcount.h>
  32 #include <sys/zap_impl.h>
  33 #include <sys/zap_leaf.h>
  34 #include <sys/avl.h>
  35 #include <sys/arc.h>
  36 
  37 #ifdef _KERNEL
  38 #include <sys/sunddi.h>
  39 #endif
  40 
  41 static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags);
  42 
  43 uint64_t
  44 zap_getflags(zap_t *zap)
  45 {
  46         if (zap->zap_ismicro)
  47                 return (0);
  48         return (zap->zap_f_phys->zap_flags);
  49 }
  50 
  51 int
  52 zap_hashbits(zap_t *zap)
  53 {
  54         if (zap_getflags(zap) & ZAP_FLAG_HASH64)
  55                 return (48);
  56         else
  57                 return (28);
  58 }
  59 
  60 uint32_t
  61 zap_maxcd(zap_t *zap)
  62 {
  63         if (zap_getflags(zap) & ZAP_FLAG_HASH64)
  64                 return ((1<<16)-1);
  65         else
  66                 return (-1U);
  67 }
  68 
  69 static uint64_t
  70 zap_hash(zap_name_t *zn)
  71 {
  72         zap_t *zap = zn->zn_zap;
  73         uint64_t h = 0;
  74 
  75         if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
  76                 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
  77                 h = *(uint64_t *)zn->zn_key_orig;
  78         } else {
  79                 h = zap->zap_salt;
  80                 ASSERT(h != 0);
  81                 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
  82 
  83                 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
  84                         int i;
  85                         const uint64_t *wp = zn->zn_key_norm;
  86 
  87                         ASSERT(zn->zn_key_intlen == 8);
  88                         for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
  89                                 int j;
  90                                 uint64_t word = *wp;
  91 
  92                                 for (j = 0; j < zn->zn_key_intlen; j++) {
  93                                         h = (h >> 8) ^
  94                                             zfs_crc64_table[(h ^ word) & 0xFF];
  95                                         word >>= NBBY;
  96                                 }
  97                         }
  98                 } else {
  99                         int i, len;
 100                         const uint8_t *cp = zn->zn_key_norm;
 101 
 102                         /*
 103                          * We previously stored the terminating null on
 104                          * disk, but didn't hash it, so we need to
 105                          * continue to not hash it.  (The
 106                          * zn_key_*_numints includes the terminating
 107                          * null for non-binary keys.)
 108                          */
 109                         len = zn->zn_key_norm_numints - 1;
 110 
 111                         ASSERT(zn->zn_key_intlen == 1);
 112                         for (i = 0; i < len; cp++, i++) {
 113                                 h = (h >> 8) ^
 114                                     zfs_crc64_table[(h ^ *cp) & 0xFF];
 115                         }
 116                 }
 117         }
 118         /*
 119          * Don't use all 64 bits, since we need some in the cookie for
 120          * the collision differentiator.  We MUST use the high bits,
 121          * since those are the ones that we first pay attention to when
 122          * chosing the bucket.
 123          */
 124         h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
 125 
 126         return (h);
 127 }
 128 
 129 static int
 130 zap_normalize(zap_t *zap, const char *name, char *namenorm)
 131 {
 132         size_t inlen, outlen;
 133         int err;
 134 
 135         ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
 136 
 137         inlen = strlen(name) + 1;
 138         outlen = ZAP_MAXNAMELEN;
 139 
 140         err = 0;
 141         (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
 142             zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
 143             U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
 144 
 145         return (err);
 146 }
 147 
 148 boolean_t
 149 zap_match(zap_name_t *zn, const char *matchname)
 150 {
 151         ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
 152 
 153         if (zn->zn_matchtype == MT_FIRST) {
 154                 char norm[ZAP_MAXNAMELEN];
 155 
 156                 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
 157                         return (B_FALSE);
 158 
 159                 return (strcmp(zn->zn_key_norm, norm) == 0);
 160         } else {
 161                 /* MT_BEST or MT_EXACT */
 162                 return (strcmp(zn->zn_key_orig, matchname) == 0);
 163         }
 164 }
 165 
 166 void
 167 zap_name_free(zap_name_t *zn)
 168 {
 169         kmem_free(zn, sizeof (zap_name_t));
 170 }
 171 
 172 zap_name_t *
 173 zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
 174 {
 175         zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
 176 
 177         zn->zn_zap = zap;
 178         zn->zn_key_intlen = sizeof (*key);
 179         zn->zn_key_orig = key;
 180         zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
 181         zn->zn_matchtype = mt;
 182         if (zap->zap_normflags) {
 183                 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
 184                         zap_name_free(zn);
 185                         return (NULL);
 186                 }
 187                 zn->zn_key_norm = zn->zn_normbuf;
 188                 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
 189         } else {
 190                 if (mt != MT_EXACT) {
 191                         zap_name_free(zn);
 192                         return (NULL);
 193                 }
 194                 zn->zn_key_norm = zn->zn_key_orig;
 195                 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
 196         }
 197 
 198         zn->zn_hash = zap_hash(zn);
 199         return (zn);
 200 }
 201 
 202 zap_name_t *
 203 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
 204 {
 205         zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
 206 
 207         ASSERT(zap->zap_normflags == 0);
 208         zn->zn_zap = zap;
 209         zn->zn_key_intlen = sizeof (*key);
 210         zn->zn_key_orig = zn->zn_key_norm = key;
 211         zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
 212         zn->zn_matchtype = MT_EXACT;
 213 
 214         zn->zn_hash = zap_hash(zn);
 215         return (zn);
 216 }
 217 
 218 static void
 219 mzap_byteswap(mzap_phys_t *buf, size_t size)
 220 {
 221         int i, max;
 222         buf->mz_block_type = BSWAP_64(buf->mz_block_type);
 223         buf->mz_salt = BSWAP_64(buf->mz_salt);
 224         buf->mz_normflags = BSWAP_64(buf->mz_normflags);
 225         max = (size / MZAP_ENT_LEN) - 1;
 226         for (i = 0; i < max; i++) {
 227                 buf->mz_chunk[i].mze_value =
 228                     BSWAP_64(buf->mz_chunk[i].mze_value);
 229                 buf->mz_chunk[i].mze_cd =
 230                     BSWAP_32(buf->mz_chunk[i].mze_cd);
 231         }
 232 }
 233 
 234 void
 235 zap_byteswap(void *buf, size_t size)
 236 {
 237         uint64_t block_type;
 238 
 239         block_type = *(uint64_t *)buf;
 240 
 241         if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
 242                 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
 243                 mzap_byteswap(buf, size);
 244         } else {
 245                 fzap_byteswap(buf, size);
 246         }
 247 }
 248 
 249 static int
 250 mze_compare(const void *arg1, const void *arg2)
 251 {
 252         const mzap_ent_t *mze1 = arg1;
 253         const mzap_ent_t *mze2 = arg2;
 254 
 255         if (mze1->mze_hash > mze2->mze_hash)
 256                 return (+1);
 257         if (mze1->mze_hash < mze2->mze_hash)
 258                 return (-1);
 259         if (mze1->mze_cd > mze2->mze_cd)
 260                 return (+1);
 261         if (mze1->mze_cd < mze2->mze_cd)
 262                 return (-1);
 263         return (0);
 264 }
 265 
 266 static void
 267 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
 268 {
 269         mzap_ent_t *mze;
 270 
 271         ASSERT(zap->zap_ismicro);
 272         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 273 
 274         mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
 275         mze->mze_chunkid = chunkid;
 276         mze->mze_hash = hash;
 277         mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
 278         ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
 279         avl_add(&zap->zap_m.zap_avl, mze);
 280 }
 281 
 282 static mzap_ent_t *
 283 mze_find(zap_name_t *zn)
 284 {
 285         mzap_ent_t mze_tofind;
 286         mzap_ent_t *mze;
 287         avl_index_t idx;
 288         avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
 289 
 290         ASSERT(zn->zn_zap->zap_ismicro);
 291         ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
 292 
 293         mze_tofind.mze_hash = zn->zn_hash;
 294         mze_tofind.mze_cd = 0;
 295 
 296 again:
 297         mze = avl_find(avl, &mze_tofind, &idx);
 298         if (mze == NULL)
 299                 mze = avl_nearest(avl, idx, AVL_AFTER);
 300         for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
 301                 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
 302                 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
 303                         return (mze);
 304         }
 305         if (zn->zn_matchtype == MT_BEST) {
 306                 zn->zn_matchtype = MT_FIRST;
 307                 goto again;
 308         }
 309         return (NULL);
 310 }
 311 
 312 static uint32_t
 313 mze_find_unused_cd(zap_t *zap, uint64_t hash)
 314 {
 315         mzap_ent_t mze_tofind;
 316         mzap_ent_t *mze;
 317         avl_index_t idx;
 318         avl_tree_t *avl = &zap->zap_m.zap_avl;
 319         uint32_t cd;
 320 
 321         ASSERT(zap->zap_ismicro);
 322         ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 323 
 324         mze_tofind.mze_hash = hash;
 325         mze_tofind.mze_cd = 0;
 326 
 327         cd = 0;
 328         for (mze = avl_find(avl, &mze_tofind, &idx);
 329             mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
 330                 if (mze->mze_cd != cd)
 331                         break;
 332                 cd++;
 333         }
 334 
 335         return (cd);
 336 }
 337 
 338 static void
 339 mze_remove(zap_t *zap, mzap_ent_t *mze)
 340 {
 341         ASSERT(zap->zap_ismicro);
 342         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 343 
 344         avl_remove(&zap->zap_m.zap_avl, mze);
 345         kmem_free(mze, sizeof (mzap_ent_t));
 346 }
 347 
 348 static void
 349 mze_destroy(zap_t *zap)
 350 {
 351         mzap_ent_t *mze;
 352         void *avlcookie = NULL;
 353 
 354         while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
 355                 kmem_free(mze, sizeof (mzap_ent_t));
 356         avl_destroy(&zap->zap_m.zap_avl);
 357 }
 358 
 359 static zap_t *
 360 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
 361 {
 362         zap_t *winner;
 363         zap_t *zap;
 364         int i;
 365 
 366         ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
 367 
 368         zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
 369         rw_init(&zap->zap_rwlock, 0, 0, 0);
 370         rw_enter(&zap->zap_rwlock, RW_WRITER);
 371         zap->zap_objset = os;
 372         zap->zap_object = obj;
 373         zap->zap_dbuf = db;
 374 
 375         if (*(uint64_t *)db->db_data != ZBT_MICRO) {
 376                 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
 377                 zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
 378         } else {
 379                 zap->zap_ismicro = TRUE;
 380         }
 381 
 382         /*
 383          * Make sure that zap_ismicro is set before we let others see
 384          * it, because zap_lockdir() checks zap_ismicro without the lock
 385          * held.
 386          */
 387         dmu_buf_init_user(&zap->db_evict, zap_evict);
 388         winner = (zap_t *)dmu_buf_set_user(db, &zap->db_evict);
 389 
 390         if (winner != NULL) {
 391                 rw_exit(&zap->zap_rwlock);
 392                 rw_destroy(&zap->zap_rwlock);
 393                 if (!zap->zap_ismicro)
 394                         mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
 395                 kmem_free(zap, sizeof (zap_t));
 396                 return (winner);
 397         }
 398 
 399         if (zap->zap_ismicro) {
 400                 zap->zap_salt = zap->zap_m_phys->mz_salt;
 401                 zap->zap_normflags = zap->zap_m_phys->mz_normflags;
 402                 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
 403                 avl_create(&zap->zap_m.zap_avl, mze_compare,
 404                     sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
 405 
 406                 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
 407                         mzap_ent_phys_t *mze =
 408                             &zap->zap_m_phys->mz_chunk[i];
 409                         if (mze->mze_name[0]) {
 410                                 zap_name_t *zn;
 411 
 412                                 zap->zap_m.zap_num_entries++;
 413                                 zn = zap_name_alloc(zap, mze->mze_name,
 414                                     MT_EXACT);
 415                                 mze_insert(zap, i, zn->zn_hash);
 416                                 zap_name_free(zn);
 417                         }
 418                 }
 419         } else {
 420                 zap->zap_salt = zap->zap_f_phys->zap_salt;
 421                 zap->zap_normflags = zap->zap_f_phys->zap_normflags;
 422 
 423                 ASSERT3U(sizeof (struct zap_leaf_header), ==,
 424                     2*ZAP_LEAF_CHUNKSIZE);
 425 
 426                 /*
 427                  * The embedded pointer table should not overlap the
 428                  * other members.
 429                  */
 430                 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
 431                     &zap->zap_f_phys->zap_salt);
 432 
 433                 /*
 434                  * The embedded pointer table should end at the end of
 435                  * the block
 436                  */
 437                 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
 438                     1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
 439                     (uintptr_t)zap->zap_f_phys, ==,
 440                     zap->zap_dbuf->db_size);
 441         }
 442         rw_exit(&zap->zap_rwlock);
 443         return (zap);
 444 }
 445 
 446 int
 447 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
 448     krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
 449 {
 450         zap_t *zap;
 451         dmu_buf_t *db;
 452         krw_t lt;
 453         int err;
 454 
 455         *zapp = NULL;
 456 
 457         err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH);
 458         if (err)
 459                 return (err);
 460 
 461 #ifdef ZFS_DEBUG
 462         {
 463                 dmu_object_info_t doi;
 464                 dmu_object_info_from_db(db, &doi);
 465                 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
 466         }
 467 #endif
 468 
 469         zap = (zap_t *)dmu_buf_get_user(db);
 470         if (zap == NULL)
 471                 zap = mzap_open(os, obj, db);
 472 
 473         /*
 474          * We're checking zap_ismicro without the lock held, in order to
 475          * tell what type of lock we want.  Once we have some sort of
 476          * lock, see if it really is the right type.  In practice this
 477          * can only be different if it was upgraded from micro to fat,
 478          * and micro wanted WRITER but fat only needs READER.
 479          */
 480         lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
 481         rw_enter(&zap->zap_rwlock, lt);
 482         if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
 483                 /* it was upgraded, now we only need reader */
 484                 ASSERT(lt == RW_WRITER);
 485                 ASSERT(RW_READER ==
 486                     (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
 487                 rw_downgrade(&zap->zap_rwlock);
 488                 lt = RW_READER;
 489         }
 490 
 491         zap->zap_objset = os;
 492 
 493         if (lt == RW_WRITER)
 494                 dmu_buf_will_dirty(db, tx);
 495 
 496         ASSERT3P(zap->zap_dbuf, ==, db);
 497 
 498         ASSERT(!zap->zap_ismicro ||
 499             zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
 500         if (zap->zap_ismicro && tx && adding &&
 501             zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
 502                 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
 503                 if (newsz > MZAP_MAX_BLKSZ) {
 504                         dprintf("upgrading obj %llu: num_entries=%u\n",
 505                             obj, zap->zap_m.zap_num_entries);
 506                         *zapp = zap;
 507                         return (mzap_upgrade(zapp, tx, 0));
 508                 }
 509                 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
 510                 ASSERT0(err);
 511                 zap->zap_m.zap_num_chunks =
 512                     db->db_size / MZAP_ENT_LEN - 1;
 513         }
 514 
 515         *zapp = zap;
 516         return (0);
 517 }
 518 
 519 void
 520 zap_unlockdir(zap_t *zap)
 521 {
 522         rw_exit(&zap->zap_rwlock);
 523         dmu_buf_rele(zap->zap_dbuf, NULL);
 524 }
 525 
 526 static int
 527 mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
 528 {
 529         mzap_phys_t *mzp;
 530         int i, sz, nchunks;
 531         int err = 0;
 532         zap_t *zap = *zapp;
 533 
 534         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 535 
 536         sz = zap->zap_dbuf->db_size;
 537         mzp = kmem_alloc(sz, KM_SLEEP);
 538         bcopy(zap->zap_dbuf->db_data, mzp, sz);
 539         nchunks = zap->zap_m.zap_num_chunks;
 540 
 541         if (!flags) {
 542                 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
 543                     1ULL << fzap_default_block_shift, 0, tx);
 544                 if (err) {
 545                         kmem_free(mzp, sz);
 546                         return (err);
 547                 }
 548         }
 549 
 550         dprintf("upgrading obj=%llu with %u chunks\n",
 551             zap->zap_object, nchunks);
 552         /* XXX destroy the avl later, so we can use the stored hash value */
 553         mze_destroy(zap);
 554 
 555         fzap_upgrade(zap, tx, flags);
 556 
 557         for (i = 0; i < nchunks; i++) {
 558                 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
 559                 zap_name_t *zn;
 560                 if (mze->mze_name[0] == 0)
 561                         continue;
 562                 dprintf("adding %s=%llu\n",
 563                     mze->mze_name, mze->mze_value);
 564                 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
 565                 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
 566                 zap = zn->zn_zap;    /* fzap_add_cd() may change zap */
 567                 zap_name_free(zn);
 568                 if (err)
 569                         break;
 570         }
 571         kmem_free(mzp, sz);
 572         *zapp = zap;
 573         return (err);
 574 }
 575 
 576 static void
 577 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
 578     dmu_tx_t *tx)
 579 {
 580         dmu_buf_t *db;
 581         mzap_phys_t *zp;
 582 
 583         VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
 584 
 585 #ifdef ZFS_DEBUG
 586         {
 587                 dmu_object_info_t doi;
 588                 dmu_object_info_from_db(db, &doi);
 589                 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
 590         }
 591 #endif
 592 
 593         dmu_buf_will_dirty(db, tx);
 594         zp = db->db_data;
 595         zp->mz_block_type = ZBT_MICRO;
 596         zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
 597         zp->mz_normflags = normflags;
 598         dmu_buf_rele(db, FTAG);
 599 
 600         if (flags != 0) {
 601                 zap_t *zap;
 602                 /* Only fat zap supports flags; upgrade immediately. */
 603                 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
 604                     B_FALSE, B_FALSE, &zap));
 605                 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags));
 606                 zap_unlockdir(zap);
 607         }
 608 }
 609 
 610 int
 611 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
 612     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 613 {
 614         return (zap_create_claim_norm(os, obj,
 615             0, ot, bonustype, bonuslen, tx));
 616 }
 617 
 618 int
 619 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
 620     dmu_object_type_t ot,
 621     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 622 {
 623         int err;
 624 
 625         err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
 626         if (err != 0)
 627                 return (err);
 628         mzap_create_impl(os, obj, normflags, 0, tx);
 629         return (0);
 630 }
 631 
 632 uint64_t
 633 zap_create(objset_t *os, dmu_object_type_t ot,
 634     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 635 {
 636         return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
 637 }
 638 
 639 uint64_t
 640 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
 641     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 642 {
 643         uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
 644 
 645         mzap_create_impl(os, obj, normflags, 0, tx);
 646         return (obj);
 647 }
 648 
 649 uint64_t
 650 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
 651     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
 652     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 653 {
 654         uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
 655 
 656         ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
 657             leaf_blockshift <= SPA_MAXBLOCKSHIFT &&
 658             indirect_blockshift >= SPA_MINBLOCKSHIFT &&
 659             indirect_blockshift <= SPA_MAXBLOCKSHIFT);
 660 
 661         VERIFY(dmu_object_set_blocksize(os, obj,
 662             1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
 663 
 664         mzap_create_impl(os, obj, normflags, flags, tx);
 665         return (obj);
 666 }
 667 
 668 int
 669 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
 670 {
 671         /*
 672          * dmu_object_free will free the object number and free the
 673          * data.  Freeing the data will cause our pageout function to be
 674          * called, which will destroy our data (zap_leaf_t's and zap_t).
 675          */
 676 
 677         return (dmu_object_free(os, zapobj, tx));
 678 }
 679 
 680 void
 681 zap_evict(dmu_buf_user_t *dbu)
 682 {
 683         zap_t *zap = (zap_t *)dbu;
 684 
 685         rw_destroy(&zap->zap_rwlock);
 686 
 687         if (zap->zap_ismicro)
 688                 mze_destroy(zap);
 689         else
 690                 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
 691 
 692         kmem_free(zap, sizeof (zap_t));
 693 }
 694 
 695 int
 696 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
 697 {
 698         zap_t *zap;
 699         int err;
 700 
 701         err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
 702         if (err)
 703                 return (err);
 704         if (!zap->zap_ismicro) {
 705                 err = fzap_count(zap, count);
 706         } else {
 707                 *count = zap->zap_m.zap_num_entries;
 708         }
 709         zap_unlockdir(zap);
 710         return (err);
 711 }
 712 
 713 /*
 714  * zn may be NULL; if not specified, it will be computed if needed.
 715  * See also the comment above zap_entry_normalization_conflict().
 716  */
 717 static boolean_t
 718 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
 719 {
 720         mzap_ent_t *other;
 721         int direction = AVL_BEFORE;
 722         boolean_t allocdzn = B_FALSE;
 723 
 724         if (zap->zap_normflags == 0)
 725                 return (B_FALSE);
 726 
 727 again:
 728         for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
 729             other && other->mze_hash == mze->mze_hash;
 730             other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
 731 
 732                 if (zn == NULL) {
 733                         zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
 734                             MT_FIRST);
 735                         allocdzn = B_TRUE;
 736                 }
 737                 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
 738                         if (allocdzn)
 739                                 zap_name_free(zn);
 740                         return (B_TRUE);
 741                 }
 742         }
 743 
 744         if (direction == AVL_BEFORE) {
 745                 direction = AVL_AFTER;
 746                 goto again;
 747         }
 748 
 749         if (allocdzn)
 750                 zap_name_free(zn);
 751         return (B_FALSE);
 752 }
 753 
 754 /*
 755  * Routines for manipulating attributes.
 756  */
 757 
 758 int
 759 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
 760     uint64_t integer_size, uint64_t num_integers, void *buf)
 761 {
 762         return (zap_lookup_norm(os, zapobj, name, integer_size,
 763             num_integers, buf, MT_EXACT, NULL, 0, NULL));
 764 }
 765 
 766 int
 767 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
 768     uint64_t integer_size, uint64_t num_integers, void *buf,
 769     matchtype_t mt, char *realname, int rn_len,
 770     boolean_t *ncp)
 771 {
 772         zap_t *zap;
 773         int err;
 774         mzap_ent_t *mze;
 775         zap_name_t *zn;
 776 
 777         err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
 778         if (err)
 779                 return (err);
 780         zn = zap_name_alloc(zap, name, mt);
 781         if (zn == NULL) {
 782                 zap_unlockdir(zap);
 783                 return (SET_ERROR(ENOTSUP));
 784         }
 785 
 786         if (!zap->zap_ismicro) {
 787                 err = fzap_lookup(zn, integer_size, num_integers, buf,
 788                     realname, rn_len, ncp);
 789         } else {
 790                 mze = mze_find(zn);
 791                 if (mze == NULL) {
 792                         err = SET_ERROR(ENOENT);
 793                 } else {
 794                         if (num_integers < 1) {
 795                                 err = SET_ERROR(EOVERFLOW);
 796                         } else if (integer_size != 8) {
 797                                 err = SET_ERROR(EINVAL);
 798                         } else {
 799                                 *(uint64_t *)buf =
 800                                     MZE_PHYS(zap, mze)->mze_value;
 801                                 (void) strlcpy(realname,
 802                                     MZE_PHYS(zap, mze)->mze_name, rn_len);
 803                                 if (ncp) {
 804                                         *ncp = mzap_normalization_conflict(zap,
 805                                             zn, mze);
 806                                 }
 807                         }
 808                 }
 809         }
 810         zap_name_free(zn);
 811         zap_unlockdir(zap);
 812         return (err);
 813 }
 814 
 815 int
 816 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
 817     int key_numints)
 818 {
 819         zap_t *zap;
 820         int err;
 821         zap_name_t *zn;
 822 
 823         err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
 824         if (err)
 825                 return (err);
 826         zn = zap_name_alloc_uint64(zap, key, key_numints);
 827         if (zn == NULL) {
 828                 zap_unlockdir(zap);
 829                 return (SET_ERROR(ENOTSUP));
 830         }
 831 
 832         fzap_prefetch(zn);
 833         zap_name_free(zn);
 834         zap_unlockdir(zap);
 835         return (err);
 836 }
 837 
 838 int
 839 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
 840     int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
 841 {
 842         zap_t *zap;
 843         int err;
 844         zap_name_t *zn;
 845 
 846         err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
 847         if (err)
 848                 return (err);
 849         zn = zap_name_alloc_uint64(zap, key, key_numints);
 850         if (zn == NULL) {
 851                 zap_unlockdir(zap);
 852                 return (SET_ERROR(ENOTSUP));
 853         }
 854 
 855         err = fzap_lookup(zn, integer_size, num_integers, buf,
 856             NULL, 0, NULL);
 857         zap_name_free(zn);
 858         zap_unlockdir(zap);
 859         return (err);
 860 }
 861 
 862 int
 863 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
 864 {
 865         int err = (zap_lookup_norm(os, zapobj, name, 0,
 866             0, NULL, MT_EXACT, NULL, 0, NULL));
 867         if (err == EOVERFLOW || err == EINVAL)
 868                 err = 0; /* found, but skipped reading the value */
 869         return (err);
 870 }
 871 
 872 int
 873 zap_length(objset_t *os, uint64_t zapobj, const char *name,
 874     uint64_t *integer_size, uint64_t *num_integers)
 875 {
 876         zap_t *zap;
 877         int err;
 878         mzap_ent_t *mze;
 879         zap_name_t *zn;
 880 
 881         err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
 882         if (err)
 883                 return (err);
 884         zn = zap_name_alloc(zap, name, MT_EXACT);
 885         if (zn == NULL) {
 886                 zap_unlockdir(zap);
 887                 return (SET_ERROR(ENOTSUP));
 888         }
 889         if (!zap->zap_ismicro) {
 890                 err = fzap_length(zn, integer_size, num_integers);
 891         } else {
 892                 mze = mze_find(zn);
 893                 if (mze == NULL) {
 894                         err = SET_ERROR(ENOENT);
 895                 } else {
 896                         if (integer_size)
 897                                 *integer_size = 8;
 898                         if (num_integers)
 899                                 *num_integers = 1;
 900                 }
 901         }
 902         zap_name_free(zn);
 903         zap_unlockdir(zap);
 904         return (err);
 905 }
 906 
 907 int
 908 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
 909     int key_numints, uint64_t *integer_size, uint64_t *num_integers)
 910 {
 911         zap_t *zap;
 912         int err;
 913         zap_name_t *zn;
 914 
 915         err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
 916         if (err)
 917                 return (err);
 918         zn = zap_name_alloc_uint64(zap, key, key_numints);
 919         if (zn == NULL) {
 920                 zap_unlockdir(zap);
 921                 return (SET_ERROR(ENOTSUP));
 922         }
 923         err = fzap_length(zn, integer_size, num_integers);
 924         zap_name_free(zn);
 925         zap_unlockdir(zap);
 926         return (err);
 927 }
 928 
 929 static void
 930 mzap_addent(zap_name_t *zn, uint64_t value)
 931 {
 932         int i;
 933         zap_t *zap = zn->zn_zap;
 934         int start = zap->zap_m.zap_alloc_next;
 935         uint32_t cd;
 936 
 937         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 938 
 939 #ifdef ZFS_DEBUG
 940         for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
 941                 mzap_ent_phys_t *mze = &zap->zap_m_phys->mz_chunk[i];
 942                 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
 943         }
 944 #endif
 945 
 946         cd = mze_find_unused_cd(zap, zn->zn_hash);
 947         /* given the limited size of the microzap, this can't happen */
 948         ASSERT(cd < zap_maxcd(zap));
 949 
 950 again:
 951         for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
 952                 mzap_ent_phys_t *mze = &zap->zap_m_phys->mz_chunk[i];
 953                 if (mze->mze_name[0] == 0) {
 954                         mze->mze_value = value;
 955                         mze->mze_cd = cd;
 956                         (void) strcpy(mze->mze_name, zn->zn_key_orig);
 957                         zap->zap_m.zap_num_entries++;
 958                         zap->zap_m.zap_alloc_next = i+1;
 959                         if (zap->zap_m.zap_alloc_next ==
 960                             zap->zap_m.zap_num_chunks)
 961                                 zap->zap_m.zap_alloc_next = 0;
 962                         mze_insert(zap, i, zn->zn_hash);
 963                         return;
 964                 }
 965         }
 966         if (start != 0) {
 967                 start = 0;
 968                 goto again;
 969         }
 970         ASSERT(!"out of entries!");
 971 }
 972 
 973 int
 974 zap_add(objset_t *os, uint64_t zapobj, const char *key,
 975     int integer_size, uint64_t num_integers,
 976     const void *val, dmu_tx_t *tx)
 977 {
 978         zap_t *zap;
 979         int err;
 980         mzap_ent_t *mze;
 981         const uint64_t *intval = val;
 982         zap_name_t *zn;
 983 
 984         err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
 985         if (err)
 986                 return (err);
 987         zn = zap_name_alloc(zap, key, MT_EXACT);
 988         if (zn == NULL) {
 989                 zap_unlockdir(zap);
 990                 return (SET_ERROR(ENOTSUP));
 991         }
 992         if (!zap->zap_ismicro) {
 993                 err = fzap_add(zn, integer_size, num_integers, val, tx);
 994                 zap = zn->zn_zap;    /* fzap_add() may change zap */
 995         } else if (integer_size != 8 || num_integers != 1 ||
 996             strlen(key) >= MZAP_NAME_LEN) {
 997                 err = mzap_upgrade(&zn->zn_zap, tx, 0);
 998                 if (err == 0)
 999                         err = fzap_add(zn, integer_size, num_integers, val, tx);
1000                 zap = zn->zn_zap;    /* fzap_add() may change zap */
1001         } else {
1002                 mze = mze_find(zn);
1003                 if (mze != NULL) {
1004                         err = SET_ERROR(EEXIST);
1005                 } else {
1006                         mzap_addent(zn, *intval);
1007                 }
1008         }
1009         ASSERT(zap == zn->zn_zap);
1010         zap_name_free(zn);
1011         if (zap != NULL)        /* may be NULL if fzap_add() failed */
1012                 zap_unlockdir(zap);
1013         return (err);
1014 }
1015 
1016 int
1017 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1018     int key_numints, int integer_size, uint64_t num_integers,
1019     const void *val, dmu_tx_t *tx)
1020 {
1021         zap_t *zap;
1022         int err;
1023         zap_name_t *zn;
1024 
1025         err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1026         if (err)
1027                 return (err);
1028         zn = zap_name_alloc_uint64(zap, key, key_numints);
1029         if (zn == NULL) {
1030                 zap_unlockdir(zap);
1031                 return (SET_ERROR(ENOTSUP));
1032         }
1033         err = fzap_add(zn, integer_size, num_integers, val, tx);
1034         zap = zn->zn_zap;    /* fzap_add() may change zap */
1035         zap_name_free(zn);
1036         if (zap != NULL)        /* may be NULL if fzap_add() failed */
1037                 zap_unlockdir(zap);
1038         return (err);
1039 }
1040 
1041 int
1042 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1043     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1044 {
1045         zap_t *zap;
1046         mzap_ent_t *mze;
1047         uint64_t oldval;
1048         const uint64_t *intval = val;
1049         zap_name_t *zn;
1050         int err;
1051 
1052 #ifdef ZFS_DEBUG
1053         /*
1054          * If there is an old value, it shouldn't change across the
1055          * lockdir (eg, due to bprewrite's xlation).
1056          */
1057         if (integer_size == 8 && num_integers == 1)
1058                 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1059 #endif
1060 
1061         err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1062         if (err)
1063                 return (err);
1064         zn = zap_name_alloc(zap, name, MT_EXACT);
1065         if (zn == NULL) {
1066                 zap_unlockdir(zap);
1067                 return (SET_ERROR(ENOTSUP));
1068         }
1069         if (!zap->zap_ismicro) {
1070                 err = fzap_update(zn, integer_size, num_integers, val, tx);
1071                 zap = zn->zn_zap;    /* fzap_update() may change zap */
1072         } else if (integer_size != 8 || num_integers != 1 ||
1073             strlen(name) >= MZAP_NAME_LEN) {
1074                 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1075                     zapobj, integer_size, num_integers, name);
1076                 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1077                 if (err == 0)
1078                         err = fzap_update(zn, integer_size, num_integers,
1079                             val, tx);
1080                 zap = zn->zn_zap;    /* fzap_update() may change zap */
1081         } else {
1082                 mze = mze_find(zn);
1083                 if (mze != NULL) {
1084                         ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1085                         MZE_PHYS(zap, mze)->mze_value = *intval;
1086                 } else {
1087                         mzap_addent(zn, *intval);
1088                 }
1089         }
1090         ASSERT(zap == zn->zn_zap);
1091         zap_name_free(zn);
1092         if (zap != NULL)        /* may be NULL if fzap_upgrade() failed */
1093                 zap_unlockdir(zap);
1094         return (err);
1095 }
1096 
1097 int
1098 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1099     int key_numints,
1100     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1101 {
1102         zap_t *zap;
1103         zap_name_t *zn;
1104         int err;
1105 
1106         err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1107         if (err)
1108                 return (err);
1109         zn = zap_name_alloc_uint64(zap, key, key_numints);
1110         if (zn == NULL) {
1111                 zap_unlockdir(zap);
1112                 return (SET_ERROR(ENOTSUP));
1113         }
1114         err = fzap_update(zn, integer_size, num_integers, val, tx);
1115         zap = zn->zn_zap;    /* fzap_update() may change zap */
1116         zap_name_free(zn);
1117         if (zap != NULL)        /* may be NULL if fzap_upgrade() failed */
1118                 zap_unlockdir(zap);
1119         return (err);
1120 }
1121 
1122 int
1123 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1124 {
1125         return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1126 }
1127 
1128 int
1129 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1130     matchtype_t mt, dmu_tx_t *tx)
1131 {
1132         zap_t *zap;
1133         int err;
1134         mzap_ent_t *mze;
1135         zap_name_t *zn;
1136 
1137         err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1138         if (err)
1139                 return (err);
1140         zn = zap_name_alloc(zap, name, mt);
1141         if (zn == NULL) {
1142                 zap_unlockdir(zap);
1143                 return (SET_ERROR(ENOTSUP));
1144         }
1145         if (!zap->zap_ismicro) {
1146                 err = fzap_remove(zn, tx);
1147         } else {
1148                 mze = mze_find(zn);
1149                 if (mze == NULL) {
1150                         err = SET_ERROR(ENOENT);
1151                 } else {
1152                         zap->zap_m.zap_num_entries--;
1153                         bzero(&zap->zap_m_phys->mz_chunk[mze->mze_chunkid],
1154                             sizeof (mzap_ent_phys_t));
1155                         mze_remove(zap, mze);
1156                 }
1157         }
1158         zap_name_free(zn);
1159         zap_unlockdir(zap);
1160         return (err);
1161 }
1162 
1163 int
1164 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1165     int key_numints, dmu_tx_t *tx)
1166 {
1167         zap_t *zap;
1168         int err;
1169         zap_name_t *zn;
1170 
1171         err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1172         if (err)
1173                 return (err);
1174         zn = zap_name_alloc_uint64(zap, key, key_numints);
1175         if (zn == NULL) {
1176                 zap_unlockdir(zap);
1177                 return (SET_ERROR(ENOTSUP));
1178         }
1179         err = fzap_remove(zn, tx);
1180         zap_name_free(zn);
1181         zap_unlockdir(zap);
1182         return (err);
1183 }
1184 
1185 /*
1186  * Routines for iterating over the attributes.
1187  */
1188 
1189 void
1190 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1191     uint64_t serialized)
1192 {
1193         zc->zc_objset = os;
1194         zc->zc_zap = NULL;
1195         zc->zc_leaf = NULL;
1196         zc->zc_zapobj = zapobj;
1197         zc->zc_serialized = serialized;
1198         zc->zc_hash = 0;
1199         zc->zc_cd = 0;
1200 }
1201 
1202 void
1203 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1204 {
1205         zap_cursor_init_serialized(zc, os, zapobj, 0);
1206 }
1207 
1208 void
1209 zap_cursor_fini(zap_cursor_t *zc)
1210 {
1211         if (zc->zc_zap) {
1212                 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1213                 zap_unlockdir(zc->zc_zap);
1214                 zc->zc_zap = NULL;
1215         }
1216         if (zc->zc_leaf) {
1217                 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1218                 zap_put_leaf(zc->zc_leaf);
1219                 zc->zc_leaf = NULL;
1220         }
1221         zc->zc_objset = NULL;
1222 }
1223 
1224 uint64_t
1225 zap_cursor_serialize(zap_cursor_t *zc)
1226 {
1227         if (zc->zc_hash == -1ULL)
1228                 return (-1ULL);
1229         if (zc->zc_zap == NULL)
1230                 return (zc->zc_serialized);
1231         ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1232         ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1233 
1234         /*
1235          * We want to keep the high 32 bits of the cursor zero if we can, so
1236          * that 32-bit programs can access this.  So usually use a small
1237          * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1238          * of the cursor.
1239          *
1240          * [ collision differentiator | zap_hashbits()-bit hash value ]
1241          */
1242         return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1243             ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1244 }
1245 
1246 int
1247 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1248 {
1249         int err;
1250         avl_index_t idx;
1251         mzap_ent_t mze_tofind;
1252         mzap_ent_t *mze;
1253 
1254         if (zc->zc_hash == -1ULL)
1255                 return (SET_ERROR(ENOENT));
1256 
1257         if (zc->zc_zap == NULL) {
1258                 int hb;
1259                 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1260                     RW_READER, TRUE, FALSE, &zc->zc_zap);
1261                 if (err)
1262                         return (err);
1263 
1264                 /*
1265                  * To support zap_cursor_init_serialized, advance, retrieve,
1266                  * we must add to the existing zc_cd, which may already
1267                  * be 1 due to the zap_cursor_advance.
1268                  */
1269                 ASSERT(zc->zc_hash == 0);
1270                 hb = zap_hashbits(zc->zc_zap);
1271                 zc->zc_hash = zc->zc_serialized << (64 - hb);
1272                 zc->zc_cd += zc->zc_serialized >> hb;
1273                 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1274                         zc->zc_cd = 0;
1275         } else {
1276                 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1277         }
1278         if (!zc->zc_zap->zap_ismicro) {
1279                 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1280         } else {
1281                 mze_tofind.mze_hash = zc->zc_hash;
1282                 mze_tofind.mze_cd = zc->zc_cd;
1283 
1284                 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1285                 if (mze == NULL) {
1286                         mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1287                             idx, AVL_AFTER);
1288                 }
1289                 if (mze) {
1290                         mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1291                         ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1292                         za->za_normalization_conflict =
1293                             mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1294                         za->za_integer_length = 8;
1295                         za->za_num_integers = 1;
1296                         za->za_first_integer = mzep->mze_value;
1297                         (void) strcpy(za->za_name, mzep->mze_name);
1298                         zc->zc_hash = mze->mze_hash;
1299                         zc->zc_cd = mze->mze_cd;
1300                         err = 0;
1301                 } else {
1302                         zc->zc_hash = -1ULL;
1303                         err = SET_ERROR(ENOENT);
1304                 }
1305         }
1306         rw_exit(&zc->zc_zap->zap_rwlock);
1307         return (err);
1308 }
1309 
1310 void
1311 zap_cursor_advance(zap_cursor_t *zc)
1312 {
1313         if (zc->zc_hash == -1ULL)
1314                 return;
1315         zc->zc_cd++;
1316 }
1317 
1318 int
1319 zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1320 {
1321         int err = 0;
1322         mzap_ent_t *mze;
1323         zap_name_t *zn;
1324 
1325         if (zc->zc_zap == NULL) {
1326                 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1327                     RW_READER, TRUE, FALSE, &zc->zc_zap);
1328                 if (err)
1329                         return (err);
1330         } else {
1331                 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1332         }
1333 
1334         zn = zap_name_alloc(zc->zc_zap, name, mt);
1335         if (zn == NULL) {
1336                 rw_exit(&zc->zc_zap->zap_rwlock);
1337                 return (SET_ERROR(ENOTSUP));
1338         }
1339 
1340         if (!zc->zc_zap->zap_ismicro) {
1341                 err = fzap_cursor_move_to_key(zc, zn);
1342         } else {
1343                 mze = mze_find(zn);
1344                 if (mze == NULL) {
1345                         err = SET_ERROR(ENOENT);
1346                         goto out;
1347                 }
1348                 zc->zc_hash = mze->mze_hash;
1349                 zc->zc_cd = mze->mze_cd;
1350         }
1351 
1352 out:
1353         zap_name_free(zn);
1354         rw_exit(&zc->zc_zap->zap_rwlock);
1355         return (err);
1356 }
1357 
1358 int
1359 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1360 {
1361         int err;
1362         zap_t *zap;
1363 
1364         err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1365         if (err)
1366                 return (err);
1367 
1368         bzero(zs, sizeof (zap_stats_t));
1369 
1370         if (zap->zap_ismicro) {
1371                 zs->zs_blocksize = zap->zap_dbuf->db_size;
1372                 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1373                 zs->zs_num_blocks = 1;
1374         } else {
1375                 fzap_get_stats(zap, zs);
1376         }
1377         zap_unlockdir(zap);
1378         return (0);
1379 }
1380 
1381 int
1382 zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
1383     uint64_t *towrite, uint64_t *tooverwrite)
1384 {
1385         zap_t *zap;
1386         int err = 0;
1387 
1388 
1389         /*
1390          * Since, we don't have a name, we cannot figure out which blocks will
1391          * be affected in this operation. So, account for the worst case :
1392          * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1393          * - 4 new blocks written if adding:
1394          *      - 2 blocks for possibly split leaves,
1395          *      - 2 grown ptrtbl blocks
1396          *
1397          * This also accomodates the case where an add operation to a fairly
1398          * large microzap results in a promotion to fatzap.
1399          */
1400         if (name == NULL) {
1401                 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1402                 return (err);
1403         }
1404 
1405         /*
1406          * We lock the zap with adding == FALSE. Because, if we pass
1407          * the actual value of add, it could trigger a mzap_upgrade().
1408          * At present we are just evaluating the possibility of this operation
1409          * and hence we donot want to trigger an upgrade.
1410          */
1411         err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1412         if (err)
1413                 return (err);
1414 
1415         if (!zap->zap_ismicro) {
1416                 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1417                 if (zn) {
1418                         err = fzap_count_write(zn, add, towrite,
1419                             tooverwrite);
1420                         zap_name_free(zn);
1421                 } else {
1422                         /*
1423                          * We treat this case as similar to (name == NULL)
1424                          */
1425                         *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1426                 }
1427         } else {
1428                 /*
1429                  * We are here if (name != NULL) and this is a micro-zap.
1430                  * We account for the header block depending on whether it
1431                  * is freeable.
1432                  *
1433                  * Incase of an add-operation it is hard to find out
1434                  * if this add will promote this microzap to fatzap.
1435                  * Hence, we consider the worst case and account for the
1436                  * blocks assuming this microzap would be promoted to a
1437                  * fatzap.
1438                  *
1439                  * 1 block overwritten  : header block
1440                  * 4 new blocks written : 2 new split leaf, 2 grown
1441                  *                      ptrtbl blocks
1442                  */
1443                 if (dmu_buf_freeable(zap->zap_dbuf))
1444                         *tooverwrite += SPA_MAXBLOCKSIZE;
1445                 else
1446                         *towrite += SPA_MAXBLOCKSIZE;
1447 
1448                 if (add) {
1449                         *towrite += 4 * SPA_MAXBLOCKSIZE;
1450                 }
1451         }
1452 
1453         zap_unlockdir(zap);
1454         return (err);
1455 }