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) 2012 by Delphix. All rights reserved.
  24  */
  25 
  26 /*
  27  * This file contains the top half of the zfs directory structure
  28  * implementation. The bottom half is in zap_leaf.c.
  29  *
  30  * The zdir is an extendable hash data structure. There is a table of
  31  * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are
  32  * each a constant size and hold a variable number of directory entries.
  33  * The buckets (aka "leaf nodes") are implemented in zap_leaf.c.
  34  *
  35  * The pointer table holds a power of 2 number of pointers.
  36  * (1<<zap_t->zd_data->zd_phys->zd_prefix_len).  The bucket pointed to
  37  * by the pointer at index i in the table holds entries whose hash value
  38  * has a zd_prefix_len - bit prefix
  39  */
  40 
  41 #include <sys/spa.h>
  42 #include <sys/dmu.h>
  43 #include <sys/zfs_context.h>
  44 #include <sys/zfs_znode.h>
  45 #include <sys/fs/zfs.h>
  46 #include <sys/zap.h>
  47 #include <sys/refcount.h>
  48 #include <sys/zap_impl.h>
  49 #include <sys/zap_leaf.h>
  50 
  51 int fzap_default_block_shift = 14; /* 16k blocksize */
  52 
  53 static void zap_leaf_pageout(dmu_buf_t *db, void *vl);
  54 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
  55 
  56 
  57 void
  58 fzap_byteswap(void *vbuf, size_t size)
  59 {
  60         uint64_t block_type;
  61 
  62         block_type = *(uint64_t *)vbuf;
  63 
  64         if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF))
  65                 zap_leaf_byteswap(vbuf, size);
  66         else {
  67                 /* it's a ptrtbl block */
  68                 byteswap_uint64_array(vbuf, size);
  69         }
  70 }
  71 
  72 void
  73 fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags)
  74 {
  75         dmu_buf_t *db;
  76         zap_leaf_t *l;
  77         int i;
  78         zap_phys_t *zp;
  79 
  80         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
  81         zap->zap_ismicro = FALSE;
  82 
  83         (void) dmu_buf_update_user(zap->zap_dbuf, zap, zap,
  84             &zap->zap_f.zap_phys, zap_evict);
  85 
  86         mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
  87         zap->zap_f.zap_block_shift = highbit(zap->zap_dbuf->db_size) - 1;
  88 
  89         zp = zap->zap_f.zap_phys;
  90         /*
  91          * explicitly zero it since it might be coming from an
  92          * initialized microzap
  93          */
  94         bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
  95         zp->zap_block_type = ZBT_HEADER;
  96         zp->zap_magic = ZAP_MAGIC;
  97 
  98         zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);
  99 
 100         zp->zap_freeblk = 2;         /* block 1 will be the first leaf */
 101         zp->zap_num_leafs = 1;
 102         zp->zap_num_entries = 0;
 103         zp->zap_salt = zap->zap_salt;
 104         zp->zap_normflags = zap->zap_normflags;
 105         zp->zap_flags = flags;
 106 
 107         /* block 1 will be the first leaf */
 108         for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
 109                 ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;
 110 
 111         /*
 112          * set up block 1 - the first leaf
 113          */
 114         VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
 115             1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
 116         dmu_buf_will_dirty(db, tx);
 117 
 118         l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
 119         l->l_dbuf = db;
 120         l->l_phys = db->db_data;
 121 
 122         zap_leaf_init(l, zp->zap_normflags != 0);
 123 
 124         kmem_free(l, sizeof (zap_leaf_t));
 125         dmu_buf_rele(db, FTAG);
 126 }
 127 
 128 static int
 129 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
 130 {
 131         if (RW_WRITE_HELD(&zap->zap_rwlock))
 132                 return (1);
 133         if (rw_tryupgrade(&zap->zap_rwlock)) {
 134                 dmu_buf_will_dirty(zap->zap_dbuf, tx);
 135                 return (1);
 136         }
 137         return (0);
 138 }
 139 
 140 /*
 141  * Generic routines for dealing with the pointer & cookie tables.
 142  */
 143 
 144 static int
 145 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl,
 146     void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n),
 147     dmu_tx_t *tx)
 148 {
 149         uint64_t b, newblk;
 150         dmu_buf_t *db_old, *db_new;
 151         int err;
 152         int bs = FZAP_BLOCK_SHIFT(zap);
 153         int hepb = 1<<(bs-4);
 154         /* hepb = half the number of entries in a block */
 155 
 156         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 157         ASSERT(tbl->zt_blk != 0);
 158         ASSERT(tbl->zt_numblks > 0);
 159 
 160         if (tbl->zt_nextblk != 0) {
 161                 newblk = tbl->zt_nextblk;
 162         } else {
 163                 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2);
 164                 tbl->zt_nextblk = newblk;
 165                 ASSERT3U(tbl->zt_blks_copied, ==, 0);
 166                 dmu_prefetch(zap->zap_objset, zap->zap_object,
 167                     tbl->zt_blk << bs, tbl->zt_numblks << bs);
 168         }
 169 
 170         /*
 171          * Copy the ptrtbl from the old to new location.
 172          */
 173 
 174         b = tbl->zt_blks_copied;
 175         err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
 176             (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH);
 177         if (err)
 178                 return (err);
 179 
 180         /* first half of entries in old[b] go to new[2*b+0] */
 181         VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
 182             (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
 183         dmu_buf_will_dirty(db_new, tx);
 184         transfer_func(db_old->db_data, db_new->db_data, hepb);
 185         dmu_buf_rele(db_new, FTAG);
 186 
 187         /* second half of entries in old[b] go to new[2*b+1] */
 188         VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
 189             (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
 190         dmu_buf_will_dirty(db_new, tx);
 191         transfer_func((uint64_t *)db_old->db_data + hepb,
 192             db_new->db_data, hepb);
 193         dmu_buf_rele(db_new, FTAG);
 194 
 195         dmu_buf_rele(db_old, FTAG);
 196 
 197         tbl->zt_blks_copied++;
 198 
 199         dprintf("copied block %llu of %llu\n",
 200             tbl->zt_blks_copied, tbl->zt_numblks);
 201 
 202         if (tbl->zt_blks_copied == tbl->zt_numblks) {
 203                 (void) dmu_free_range(zap->zap_objset, zap->zap_object,
 204                     tbl->zt_blk << bs, tbl->zt_numblks << bs, tx);
 205 
 206                 tbl->zt_blk = newblk;
 207                 tbl->zt_numblks *= 2;
 208                 tbl->zt_shift++;
 209                 tbl->zt_nextblk = 0;
 210                 tbl->zt_blks_copied = 0;
 211 
 212                 dprintf("finished; numblocks now %llu (%lluk entries)\n",
 213                     tbl->zt_numblks, 1<<(tbl->zt_shift-10));
 214         }
 215 
 216         return (0);
 217 }
 218 
 219 static int
 220 zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val,
 221     dmu_tx_t *tx)
 222 {
 223         int err;
 224         uint64_t blk, off;
 225         int bs = FZAP_BLOCK_SHIFT(zap);
 226         dmu_buf_t *db;
 227 
 228         ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 229         ASSERT(tbl->zt_blk != 0);
 230 
 231         dprintf("storing %llx at index %llx\n", val, idx);
 232 
 233         blk = idx >> (bs-3);
 234         off = idx & ((1<<(bs-3))-1);
 235 
 236         err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
 237             (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
 238         if (err)
 239                 return (err);
 240         dmu_buf_will_dirty(db, tx);
 241 
 242         if (tbl->zt_nextblk != 0) {
 243                 uint64_t idx2 = idx * 2;
 244                 uint64_t blk2 = idx2 >> (bs-3);
 245                 uint64_t off2 = idx2 & ((1<<(bs-3))-1);
 246                 dmu_buf_t *db2;
 247 
 248                 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
 249                     (tbl->zt_nextblk + blk2) << bs, FTAG, &db2,
 250                     DMU_READ_NO_PREFETCH);
 251                 if (err) {
 252                         dmu_buf_rele(db, FTAG);
 253                         return (err);
 254                 }
 255                 dmu_buf_will_dirty(db2, tx);
 256                 ((uint64_t *)db2->db_data)[off2] = val;
 257                 ((uint64_t *)db2->db_data)[off2+1] = val;
 258                 dmu_buf_rele(db2, FTAG);
 259         }
 260 
 261         ((uint64_t *)db->db_data)[off] = val;
 262         dmu_buf_rele(db, FTAG);
 263 
 264         return (0);
 265 }
 266 
 267 static int
 268 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
 269 {
 270         uint64_t blk, off;
 271         int err;
 272         dmu_buf_t *db;
 273         int bs = FZAP_BLOCK_SHIFT(zap);
 274 
 275         ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 276 
 277         blk = idx >> (bs-3);
 278         off = idx & ((1<<(bs-3))-1);
 279 
 280         err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
 281             (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
 282         if (err)
 283                 return (err);
 284         *valp = ((uint64_t *)db->db_data)[off];
 285         dmu_buf_rele(db, FTAG);
 286 
 287         if (tbl->zt_nextblk != 0) {
 288                 /*
 289                  * read the nextblk for the sake of i/o error checking,
 290                  * so that zap_table_load() will catch errors for
 291                  * zap_table_store.
 292                  */
 293                 blk = (idx*2) >> (bs-3);
 294 
 295                 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
 296                     (tbl->zt_nextblk + blk) << bs, FTAG, &db,
 297                     DMU_READ_NO_PREFETCH);
 298                 dmu_buf_rele(db, FTAG);
 299         }
 300         return (err);
 301 }
 302 
 303 /*
 304  * Routines for growing the ptrtbl.
 305  */
 306 
 307 static void
 308 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n)
 309 {
 310         int i;
 311         for (i = 0; i < n; i++) {
 312                 uint64_t lb = src[i];
 313                 dst[2*i+0] = lb;
 314                 dst[2*i+1] = lb;
 315         }
 316 }
 317 
 318 static int
 319 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
 320 {
 321         /*
 322          * The pointer table should never use more hash bits than we
 323          * have (otherwise we'd be using useless zero bits to index it).
 324          * If we are within 2 bits of running out, stop growing, since
 325          * this is already an aberrant condition.
 326          */
 327         if (zap->zap_f.zap_phys->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2)
 328                 return (ENOSPC);
 329 
 330         if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) {
 331                 /*
 332                  * We are outgrowing the "embedded" ptrtbl (the one
 333                  * stored in the header block).  Give it its own entire
 334                  * block, which will double the size of the ptrtbl.
 335                  */
 336                 uint64_t newblk;
 337                 dmu_buf_t *db_new;
 338                 int err;
 339 
 340                 ASSERT3U(zap->zap_f.zap_phys->zap_ptrtbl.zt_shift, ==,
 341                     ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
 342                 ASSERT3U(zap->zap_f.zap_phys->zap_ptrtbl.zt_blk, ==, 0);
 343 
 344                 newblk = zap_allocate_blocks(zap, 1);
 345                 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
 346                     newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new,
 347                     DMU_READ_NO_PREFETCH);
 348                 if (err)
 349                         return (err);
 350                 dmu_buf_will_dirty(db_new, tx);
 351                 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
 352                     db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
 353                 dmu_buf_rele(db_new, FTAG);
 354 
 355                 zap->zap_f.zap_phys->zap_ptrtbl.zt_blk = newblk;
 356                 zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks = 1;
 357                 zap->zap_f.zap_phys->zap_ptrtbl.zt_shift++;
 358 
 359                 ASSERT3U(1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift, ==,
 360                     zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks <<
 361                     (FZAP_BLOCK_SHIFT(zap)-3));
 362 
 363                 return (0);
 364         } else {
 365                 return (zap_table_grow(zap, &zap->zap_f.zap_phys->zap_ptrtbl,
 366                     zap_ptrtbl_transfer, tx));
 367         }
 368 }
 369 
 370 static void
 371 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
 372 {
 373         dmu_buf_will_dirty(zap->zap_dbuf, tx);
 374         mutex_enter(&zap->zap_f.zap_num_entries_mtx);
 375         ASSERT(delta > 0 || zap->zap_f.zap_phys->zap_num_entries >= -delta);
 376         zap->zap_f.zap_phys->zap_num_entries += delta;
 377         mutex_exit(&zap->zap_f.zap_num_entries_mtx);
 378 }
 379 
 380 static uint64_t
 381 zap_allocate_blocks(zap_t *zap, int nblocks)
 382 {
 383         uint64_t newblk;
 384         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 385         newblk = zap->zap_f.zap_phys->zap_freeblk;
 386         zap->zap_f.zap_phys->zap_freeblk += nblocks;
 387         return (newblk);
 388 }
 389 
 390 static zap_leaf_t *
 391 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
 392 {
 393         void *winner;
 394         zap_leaf_t *l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
 395 
 396         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 397 
 398         rw_init(&l->l_rwlock, 0, 0, 0);
 399         rw_enter(&l->l_rwlock, RW_WRITER);
 400         l->l_blkid = zap_allocate_blocks(zap, 1);
 401         l->l_dbuf = NULL;
 402         l->l_phys = NULL;
 403 
 404         VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
 405             l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
 406             DMU_READ_NO_PREFETCH));
 407         winner = dmu_buf_set_user(l->l_dbuf, l, &l->l_phys, zap_leaf_pageout);
 408         ASSERT(winner == NULL);
 409         dmu_buf_will_dirty(l->l_dbuf, tx);
 410 
 411         zap_leaf_init(l, zap->zap_normflags != 0);
 412 
 413         zap->zap_f.zap_phys->zap_num_leafs++;
 414 
 415         return (l);
 416 }
 417 
 418 int
 419 fzap_count(zap_t *zap, uint64_t *count)
 420 {
 421         ASSERT(!zap->zap_ismicro);
 422         mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
 423         *count = zap->zap_f.zap_phys->zap_num_entries;
 424         mutex_exit(&zap->zap_f.zap_num_entries_mtx);
 425         return (0);
 426 }
 427 
 428 /*
 429  * Routines for obtaining zap_leaf_t's
 430  */
 431 
 432 void
 433 zap_put_leaf(zap_leaf_t *l)
 434 {
 435         rw_exit(&l->l_rwlock);
 436         dmu_buf_rele(l->l_dbuf, NULL);
 437 }
 438 
 439 _NOTE(ARGSUSED(0))
 440 static void
 441 zap_leaf_pageout(dmu_buf_t *db, void *vl)
 442 {
 443         zap_leaf_t *l = vl;
 444 
 445         rw_destroy(&l->l_rwlock);
 446         kmem_free(l, sizeof (zap_leaf_t));
 447 }
 448 
 449 static zap_leaf_t *
 450 zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
 451 {
 452         zap_leaf_t *l, *winner;
 453 
 454         ASSERT(blkid != 0);
 455 
 456         l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
 457         rw_init(&l->l_rwlock, 0, 0, 0);
 458         rw_enter(&l->l_rwlock, RW_WRITER);
 459         l->l_blkid = blkid;
 460         l->l_bs = highbit(db->db_size)-1;
 461         l->l_dbuf = db;
 462         l->l_phys = NULL;
 463 
 464         winner = dmu_buf_set_user(db, l, &l->l_phys, zap_leaf_pageout);
 465 
 466         rw_exit(&l->l_rwlock);
 467         if (winner != NULL) {
 468                 /* someone else set it first */
 469                 zap_leaf_pageout(NULL, l);
 470                 l = winner;
 471         }
 472 
 473         /*
 474          * lhr_pad was previously used for the next leaf in the leaf
 475          * chain.  There should be no chained leafs (as we have removed
 476          * support for them).
 477          */
 478         ASSERT3U(l->l_phys->l_hdr.lh_pad1, ==, 0);
 479 
 480         /*
 481          * There should be more hash entries than there can be
 482          * chunks to put in the hash table
 483          */
 484         ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
 485 
 486         /* The chunks should begin at the end of the hash table */
 487         ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
 488             &l->l_phys->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
 489 
 490         /* The chunks should end at the end of the block */
 491         ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
 492             (uintptr_t)l->l_phys, ==, l->l_dbuf->db_size);
 493 
 494         return (l);
 495 }
 496 
 497 static int
 498 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
 499     zap_leaf_t **lp)
 500 {
 501         dmu_buf_t *db;
 502         zap_leaf_t *l;
 503         int bs = FZAP_BLOCK_SHIFT(zap);
 504         int err;
 505 
 506         ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 507 
 508         err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
 509             blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
 510         if (err)
 511                 return (err);
 512 
 513         ASSERT3U(db->db_object, ==, zap->zap_object);
 514         ASSERT3U(db->db_offset, ==, blkid << bs);
 515         ASSERT3U(db->db_size, ==, 1 << bs);
 516         ASSERT(blkid != 0);
 517 
 518         l = dmu_buf_get_user(db);
 519 
 520         if (l == NULL)
 521                 l = zap_open_leaf(blkid, db);
 522 
 523         rw_enter(&l->l_rwlock, lt);
 524         /*
 525          * Must lock before dirtying, otherwise l->l_phys could change,
 526          * causing ASSERT below to fail.
 527          */
 528         if (lt == RW_WRITER)
 529                 dmu_buf_will_dirty(db, tx);
 530         ASSERT3U(l->l_blkid, ==, blkid);
 531         ASSERT3P(l->l_dbuf, ==, db);
 532         ASSERT3P(l->l_phys, ==, l->l_dbuf->db_data);
 533         ASSERT3U(l->l_phys->l_hdr.lh_block_type, ==, ZBT_LEAF);
 534         ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
 535 
 536         *lp = l;
 537         return (0);
 538 }
 539 
 540 static int
 541 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
 542 {
 543         ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 544 
 545         if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) {
 546                 ASSERT3U(idx, <,
 547                     (1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift));
 548                 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
 549                 return (0);
 550         } else {
 551                 return (zap_table_load(zap, &zap->zap_f.zap_phys->zap_ptrtbl,
 552                     idx, valp));
 553         }
 554 }
 555 
 556 static int
 557 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
 558 {
 559         ASSERT(tx != NULL);
 560         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 561 
 562         if (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk == 0) {
 563                 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
 564                 return (0);
 565         } else {
 566                 return (zap_table_store(zap, &zap->zap_f.zap_phys->zap_ptrtbl,
 567                     idx, blk, tx));
 568         }
 569 }
 570 
 571 static int
 572 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
 573 {
 574         uint64_t idx, blk;
 575         int err;
 576 
 577         ASSERT(zap->zap_dbuf == NULL ||
 578             zap->zap_f.zap_phys == zap->zap_dbuf->db_data);
 579         ASSERT3U(zap->zap_f.zap_phys->zap_magic, ==, ZAP_MAGIC);
 580         idx = ZAP_HASH_IDX(h, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift);
 581         err = zap_idx_to_blk(zap, idx, &blk);
 582         if (err != 0)
 583                 return (err);
 584         err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
 585 
 586         ASSERT(err || ZAP_HASH_IDX(h, (*lp)->l_phys->l_hdr.lh_prefix_len) ==
 587             (*lp)->l_phys->l_hdr.lh_prefix);
 588         return (err);
 589 }
 590 
 591 static int
 592 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx, zap_leaf_t **lp)
 593 {
 594         zap_t *zap = zn->zn_zap;
 595         uint64_t hash = zn->zn_hash;
 596         zap_leaf_t *nl;
 597         int prefix_diff, i, err;
 598         uint64_t sibling;
 599         int old_prefix_len = l->l_phys->l_hdr.lh_prefix_len;
 600 
 601         ASSERT3U(old_prefix_len, <=, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift);
 602         ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 603 
 604         ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
 605             l->l_phys->l_hdr.lh_prefix);
 606 
 607         if (zap_tryupgradedir(zap, tx) == 0 ||
 608             old_prefix_len == zap->zap_f.zap_phys->zap_ptrtbl.zt_shift) {
 609                 /* We failed to upgrade, or need to grow the pointer table */
 610                 objset_t *os = zap->zap_objset;
 611                 uint64_t object = zap->zap_object;
 612 
 613                 zap_put_leaf(l);
 614                 zap_unlockdir(zap);
 615                 err = zap_lockdir(os, object, tx, RW_WRITER,
 616                     FALSE, FALSE, &zn->zn_zap);
 617                 zap = zn->zn_zap;
 618                 if (err)
 619                         return (err);
 620                 ASSERT(!zap->zap_ismicro);
 621 
 622                 while (old_prefix_len ==
 623                     zap->zap_f.zap_phys->zap_ptrtbl.zt_shift) {
 624                         err = zap_grow_ptrtbl(zap, tx);
 625                         if (err)
 626                                 return (err);
 627                 }
 628 
 629                 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
 630                 if (err)
 631                         return (err);
 632 
 633                 if (l->l_phys->l_hdr.lh_prefix_len != old_prefix_len) {
 634                         /* it split while our locks were down */
 635                         *lp = l;
 636                         return (0);
 637                 }
 638         }
 639         ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 640         ASSERT3U(old_prefix_len, <, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift);
 641         ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
 642             l->l_phys->l_hdr.lh_prefix);
 643 
 644         prefix_diff = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift -
 645             (old_prefix_len + 1);
 646         sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
 647 
 648         /* check for i/o errors before doing zap_leaf_split */
 649         for (i = 0; i < (1ULL<<prefix_diff); i++) {
 650                 uint64_t blk;
 651                 err = zap_idx_to_blk(zap, sibling+i, &blk);
 652                 if (err)
 653                         return (err);
 654                 ASSERT3U(blk, ==, l->l_blkid);
 655         }
 656 
 657         nl = zap_create_leaf(zap, tx);
 658         zap_leaf_split(l, nl, zap->zap_normflags != 0);
 659 
 660         /* set sibling pointers */
 661         for (i = 0; i < (1ULL<<prefix_diff); i++) {
 662                 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx);
 663                 ASSERT3U(err, ==, 0); /* we checked for i/o errors above */
 664         }
 665 
 666         if (hash & (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len))) {
 667                 /* we want the sibling */
 668                 zap_put_leaf(l);
 669                 *lp = nl;
 670         } else {
 671                 zap_put_leaf(nl);
 672                 *lp = l;
 673         }
 674 
 675         return (0);
 676 }
 677 
 678 static void
 679 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx)
 680 {
 681         zap_t *zap = zn->zn_zap;
 682         int shift = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift;
 683         int leaffull = (l->l_phys->l_hdr.lh_prefix_len == shift &&
 684             l->l_phys->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
 685 
 686         zap_put_leaf(l);
 687 
 688         if (leaffull || zap->zap_f.zap_phys->zap_ptrtbl.zt_nextblk) {
 689                 int err;
 690 
 691                 /*
 692                  * We are in the middle of growing the pointer table, or
 693                  * this leaf will soon make us grow it.
 694                  */
 695                 if (zap_tryupgradedir(zap, tx) == 0) {
 696                         objset_t *os = zap->zap_objset;
 697                         uint64_t zapobj = zap->zap_object;
 698 
 699                         zap_unlockdir(zap);
 700                         err = zap_lockdir(os, zapobj, tx,
 701                             RW_WRITER, FALSE, FALSE, &zn->zn_zap);
 702                         zap = zn->zn_zap;
 703                         if (err)
 704                                 return;
 705                 }
 706 
 707                 /* could have finished growing while our locks were down */
 708                 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_shift == shift)
 709                         (void) zap_grow_ptrtbl(zap, tx);
 710         }
 711 }
 712 
 713 static int
 714 fzap_checkname(zap_name_t *zn)
 715 {
 716         if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
 717                 return (ENAMETOOLONG);
 718         return (0);
 719 }
 720 
 721 static int
 722 fzap_checksize(uint64_t integer_size, uint64_t num_integers)
 723 {
 724         /* Only integer sizes supported by C */
 725         switch (integer_size) {
 726         case 1:
 727         case 2:
 728         case 4:
 729         case 8:
 730                 break;
 731         default:
 732                 return (EINVAL);
 733         }
 734 
 735         if (integer_size * num_integers > ZAP_MAXVALUELEN)
 736                 return (E2BIG);
 737 
 738         return (0);
 739 }
 740 
 741 static int
 742 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
 743 {
 744         int err;
 745 
 746         if ((err = fzap_checkname(zn)) != 0)
 747                 return (err);
 748         return (fzap_checksize(integer_size, num_integers));
 749 }
 750 
 751 /*
 752  * Routines for manipulating attributes.
 753  */
 754 int
 755 fzap_lookup(zap_name_t *zn,
 756     uint64_t integer_size, uint64_t num_integers, void *buf,
 757     char *realname, int rn_len, boolean_t *ncp)
 758 {
 759         zap_leaf_t *l;
 760         int err;
 761         zap_entry_handle_t zeh;
 762 
 763         if ((err = fzap_checkname(zn)) != 0)
 764                 return (err);
 765 
 766         err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
 767         if (err != 0)
 768                 return (err);
 769         err = zap_leaf_lookup(l, zn, &zeh);
 770         if (err == 0) {
 771                 if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
 772                         zap_put_leaf(l);
 773                         return (err);
 774                 }
 775 
 776                 err = zap_entry_read(&zeh, integer_size, num_integers, buf);
 777                 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
 778                 if (ncp) {
 779                         *ncp = zap_entry_normalization_conflict(&zeh,
 780                             zn, NULL, zn->zn_zap);
 781                 }
 782         }
 783 
 784         zap_put_leaf(l);
 785         return (err);
 786 }
 787 
 788 int
 789 fzap_add_cd(zap_name_t *zn,
 790     uint64_t integer_size, uint64_t num_integers,
 791     const void *val, uint32_t cd, dmu_tx_t *tx)
 792 {
 793         zap_leaf_t *l;
 794         int err;
 795         zap_entry_handle_t zeh;
 796         zap_t *zap = zn->zn_zap;
 797 
 798         ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 799         ASSERT(!zap->zap_ismicro);
 800         ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
 801 
 802         err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
 803         if (err != 0)
 804                 return (err);
 805 retry:
 806         err = zap_leaf_lookup(l, zn, &zeh);
 807         if (err == 0) {
 808                 err = EEXIST;
 809                 goto out;
 810         }
 811         if (err != ENOENT)
 812                 goto out;
 813 
 814         err = zap_entry_create(l, zn, cd,
 815             integer_size, num_integers, val, &zeh);
 816 
 817         if (err == 0) {
 818                 zap_increment_num_entries(zap, 1, tx);
 819         } else if (err == EAGAIN) {
 820                 err = zap_expand_leaf(zn, l, tx, &l);
 821                 zap = zn->zn_zap;    /* zap_expand_leaf() may change zap */
 822                 if (err == 0)
 823                         goto retry;
 824         }
 825 
 826 out:
 827         if (zap != NULL)
 828                 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
 829         return (err);
 830 }
 831 
 832 int
 833 fzap_add(zap_name_t *zn,
 834     uint64_t integer_size, uint64_t num_integers,
 835     const void *val, dmu_tx_t *tx)
 836 {
 837         int err = fzap_check(zn, integer_size, num_integers);
 838         if (err != 0)
 839                 return (err);
 840 
 841         return (fzap_add_cd(zn, integer_size, num_integers,
 842             val, ZAP_NEED_CD, tx));
 843 }
 844 
 845 int
 846 fzap_update(zap_name_t *zn,
 847     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
 848 {
 849         zap_leaf_t *l;
 850         int err, create;
 851         zap_entry_handle_t zeh;
 852         zap_t *zap = zn->zn_zap;
 853 
 854         ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 855         err = fzap_check(zn, integer_size, num_integers);
 856         if (err != 0)
 857                 return (err);
 858 
 859         err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
 860         if (err != 0)
 861                 return (err);
 862 retry:
 863         err = zap_leaf_lookup(l, zn, &zeh);
 864         create = (err == ENOENT);
 865         ASSERT(err == 0 || err == ENOENT);
 866 
 867         if (create) {
 868                 err = zap_entry_create(l, zn, ZAP_NEED_CD,
 869                     integer_size, num_integers, val, &zeh);
 870                 if (err == 0)
 871                         zap_increment_num_entries(zap, 1, tx);
 872         } else {
 873                 err = zap_entry_update(&zeh, integer_size, num_integers, val);
 874         }
 875 
 876         if (err == EAGAIN) {
 877                 err = zap_expand_leaf(zn, l, tx, &l);
 878                 zap = zn->zn_zap;    /* zap_expand_leaf() may change zap */
 879                 if (err == 0)
 880                         goto retry;
 881         }
 882 
 883         if (zap != NULL)
 884                 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
 885         return (err);
 886 }
 887 
 888 int
 889 fzap_length(zap_name_t *zn,
 890     uint64_t *integer_size, uint64_t *num_integers)
 891 {
 892         zap_leaf_t *l;
 893         int err;
 894         zap_entry_handle_t zeh;
 895 
 896         err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
 897         if (err != 0)
 898                 return (err);
 899         err = zap_leaf_lookup(l, zn, &zeh);
 900         if (err != 0)
 901                 goto out;
 902 
 903         if (integer_size)
 904                 *integer_size = zeh.zeh_integer_size;
 905         if (num_integers)
 906                 *num_integers = zeh.zeh_num_integers;
 907 out:
 908         zap_put_leaf(l);
 909         return (err);
 910 }
 911 
 912 int
 913 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
 914 {
 915         zap_leaf_t *l;
 916         int err;
 917         zap_entry_handle_t zeh;
 918 
 919         err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
 920         if (err != 0)
 921                 return (err);
 922         err = zap_leaf_lookup(l, zn, &zeh);
 923         if (err == 0) {
 924                 zap_entry_remove(&zeh);
 925                 zap_increment_num_entries(zn->zn_zap, -1, tx);
 926         }
 927         zap_put_leaf(l);
 928         return (err);
 929 }
 930 
 931 void
 932 fzap_prefetch(zap_name_t *zn)
 933 {
 934         uint64_t idx, blk;
 935         zap_t *zap = zn->zn_zap;
 936         int bs;
 937 
 938         idx = ZAP_HASH_IDX(zn->zn_hash,
 939             zap->zap_f.zap_phys->zap_ptrtbl.zt_shift);
 940         if (zap_idx_to_blk(zap, idx, &blk) != 0)
 941                 return;
 942         bs = FZAP_BLOCK_SHIFT(zap);
 943         dmu_prefetch(zap->zap_objset, zap->zap_object, blk << bs, 1 << bs);
 944 }
 945 
 946 /*
 947  * Helper functions for consumers.
 948  */
 949 
 950 uint64_t
 951 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
 952     const char *name, dmu_tx_t *tx)
 953 {
 954         uint64_t new_obj;
 955 
 956         VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
 957         VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
 958             tx) == 0);
 959 
 960         return (new_obj);
 961 }
 962 
 963 int
 964 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
 965     char *name)
 966 {
 967         zap_cursor_t zc;
 968         zap_attribute_t *za;
 969         int err;
 970 
 971         if (mask == 0)
 972                 mask = -1ULL;
 973 
 974         za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
 975         for (zap_cursor_init(&zc, os, zapobj);
 976             (err = zap_cursor_retrieve(&zc, za)) == 0;
 977             zap_cursor_advance(&zc)) {
 978                 if ((za->za_first_integer & mask) == (value & mask)) {
 979                         (void) strcpy(name, za->za_name);
 980                         break;
 981                 }
 982         }
 983         zap_cursor_fini(&zc);
 984         kmem_free(za, sizeof (zap_attribute_t));
 985         return (err);
 986 }
 987 
 988 int
 989 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
 990 {
 991         zap_cursor_t zc;
 992         zap_attribute_t za;
 993         int err;
 994 
 995         for (zap_cursor_init(&zc, os, fromobj);
 996             zap_cursor_retrieve(&zc, &za) == 0;
 997             (void) zap_cursor_advance(&zc)) {
 998                 if (za.za_integer_length != 8 || za.za_num_integers != 1)
 999                         return (EINVAL);
1000                 err = zap_add(os, intoobj, za.za_name,
1001                     8, 1, &za.za_first_integer, tx);
1002                 if (err)
1003                         return (err);
1004         }
1005         zap_cursor_fini(&zc);
1006         return (0);
1007 }
1008 
1009 int
1010 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1011     uint64_t value, dmu_tx_t *tx)
1012 {
1013         zap_cursor_t zc;
1014         zap_attribute_t za;
1015         int err;
1016 
1017         for (zap_cursor_init(&zc, os, fromobj);
1018             zap_cursor_retrieve(&zc, &za) == 0;
1019             (void) zap_cursor_advance(&zc)) {
1020                 if (za.za_integer_length != 8 || za.za_num_integers != 1)
1021                         return (EINVAL);
1022                 err = zap_add(os, intoobj, za.za_name,
1023                     8, 1, &value, tx);
1024                 if (err)
1025                         return (err);
1026         }
1027         zap_cursor_fini(&zc);
1028         return (0);
1029 }
1030 
1031 int
1032 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1033     dmu_tx_t *tx)
1034 {
1035         zap_cursor_t zc;
1036         zap_attribute_t za;
1037         int err;
1038 
1039         for (zap_cursor_init(&zc, os, fromobj);
1040             zap_cursor_retrieve(&zc, &za) == 0;
1041             (void) zap_cursor_advance(&zc)) {
1042                 uint64_t delta = 0;
1043 
1044                 if (za.za_integer_length != 8 || za.za_num_integers != 1)
1045                         return (EINVAL);
1046 
1047                 err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta);
1048                 if (err != 0 && err != ENOENT)
1049                         return (err);
1050                 delta += za.za_first_integer;
1051                 err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx);
1052                 if (err)
1053                         return (err);
1054         }
1055         zap_cursor_fini(&zc);
1056         return (0);
1057 }
1058 
1059 int
1060 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1061 {
1062         char name[20];
1063 
1064         (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1065         return (zap_add(os, obj, name, 8, 1, &value, tx));
1066 }
1067 
1068 int
1069 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1070 {
1071         char name[20];
1072 
1073         (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1074         return (zap_remove(os, obj, name, tx));
1075 }
1076 
1077 int
1078 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1079 {
1080         char name[20];
1081 
1082         (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1083         return (zap_lookup(os, obj, name, 8, 1, &value));
1084 }
1085 
1086 int
1087 zap_add_int_key(objset_t *os, uint64_t obj,
1088     uint64_t key, uint64_t value, dmu_tx_t *tx)
1089 {
1090         char name[20];
1091 
1092         (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1093         return (zap_add(os, obj, name, 8, 1, &value, tx));
1094 }
1095 
1096 int
1097 zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1098 {
1099         char name[20];
1100 
1101         (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1102         return (zap_lookup(os, obj, name, 8, 1, valuep));
1103 }
1104 
1105 int
1106 zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1107     dmu_tx_t *tx)
1108 {
1109         uint64_t value = 0;
1110         int err;
1111 
1112         if (delta == 0)
1113                 return (0);
1114 
1115         err = zap_lookup(os, obj, name, 8, 1, &value);
1116         if (err != 0 && err != ENOENT)
1117                 return (err);
1118         value += delta;
1119         if (value == 0)
1120                 err = zap_remove(os, obj, name, tx);
1121         else
1122                 err = zap_update(os, obj, name, 8, 1, &value, tx);
1123         return (err);
1124 }
1125 
1126 int
1127 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1128     dmu_tx_t *tx)
1129 {
1130         char name[20];
1131 
1132         (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1133         return (zap_increment(os, obj, name, delta, tx));
1134 }
1135 
1136 /*
1137  * Routines for iterating over the attributes.
1138  */
1139 
1140 int
1141 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1142 {
1143         int err = ENOENT;
1144         zap_entry_handle_t zeh;
1145         zap_leaf_t *l;
1146 
1147         /* retrieve the next entry at or after zc_hash/zc_cd */
1148         /* if no entry, return ENOENT */
1149 
1150         if (zc->zc_leaf &&
1151             (ZAP_HASH_IDX(zc->zc_hash,
1152             zc->zc_leaf->l_phys->l_hdr.lh_prefix_len) !=
1153             zc->zc_leaf->l_phys->l_hdr.lh_prefix)) {
1154                 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1155                 zap_put_leaf(zc->zc_leaf);
1156                 zc->zc_leaf = NULL;
1157         }
1158 
1159 again:
1160         if (zc->zc_leaf == NULL) {
1161                 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1162                     &zc->zc_leaf);
1163                 if (err != 0)
1164                         return (err);
1165         } else {
1166                 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1167         }
1168         l = zc->zc_leaf;
1169 
1170         err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1171 
1172         if (err == ENOENT) {
1173                 uint64_t nocare =
1174                     (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len)) - 1;
1175                 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1176                 zc->zc_cd = 0;
1177                 if (l->l_phys->l_hdr.lh_prefix_len == 0 || zc->zc_hash == 0) {
1178                         zc->zc_hash = -1ULL;
1179                 } else {
1180                         zap_put_leaf(zc->zc_leaf);
1181                         zc->zc_leaf = NULL;
1182                         goto again;
1183                 }
1184         }
1185 
1186         if (err == 0) {
1187                 zc->zc_hash = zeh.zeh_hash;
1188                 zc->zc_cd = zeh.zeh_cd;
1189                 za->za_integer_length = zeh.zeh_integer_size;
1190                 za->za_num_integers = zeh.zeh_num_integers;
1191                 if (zeh.zeh_num_integers == 0) {
1192                         za->za_first_integer = 0;
1193                 } else {
1194                         err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1195                         ASSERT(err == 0 || err == EOVERFLOW);
1196                 }
1197                 err = zap_entry_read_name(zap, &zeh,
1198                     sizeof (za->za_name), za->za_name);
1199                 ASSERT(err == 0);
1200 
1201                 za->za_normalization_conflict =
1202                     zap_entry_normalization_conflict(&zeh,
1203                     NULL, za->za_name, zap);
1204         }
1205         rw_exit(&zc->zc_leaf->l_rwlock);
1206         return (err);
1207 }
1208 
1209 static void
1210 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1211 {
1212         int i, err;
1213         uint64_t lastblk = 0;
1214 
1215         /*
1216          * NB: if a leaf has more pointers than an entire ptrtbl block
1217          * can hold, then it'll be accounted for more than once, since
1218          * we won't have lastblk.
1219          */
1220         for (i = 0; i < len; i++) {
1221                 zap_leaf_t *l;
1222 
1223                 if (tbl[i] == lastblk)
1224                         continue;
1225                 lastblk = tbl[i];
1226 
1227                 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1228                 if (err == 0) {
1229                         zap_leaf_stats(zap, l, zs);
1230                         zap_put_leaf(l);
1231                 }
1232         }
1233 }
1234 
1235 int
1236 fzap_cursor_move_to_key(zap_cursor_t *zc, zap_name_t *zn)
1237 {
1238         int err;
1239         zap_leaf_t *l;
1240         zap_entry_handle_t zeh;
1241 
1242         if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
1243                 return (ENAMETOOLONG);
1244 
1245         err = zap_deref_leaf(zc->zc_zap, zn->zn_hash, NULL, RW_READER, &l);
1246         if (err != 0)
1247                 return (err);
1248 
1249         err = zap_leaf_lookup(l, zn, &zeh);
1250         if (err != 0)
1251                 return (err);
1252 
1253         zc->zc_leaf = l;
1254         zc->zc_hash = zeh.zeh_hash;
1255         zc->zc_cd = zeh.zeh_cd;
1256 
1257         return (err);
1258 }
1259 
1260 void
1261 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1262 {
1263         int bs = FZAP_BLOCK_SHIFT(zap);
1264         zs->zs_blocksize = 1ULL << bs;
1265 
1266         /*
1267          * Set zap_phys_t fields
1268          */
1269         zs->zs_num_leafs = zap->zap_f.zap_phys->zap_num_leafs;
1270         zs->zs_num_entries = zap->zap_f.zap_phys->zap_num_entries;
1271         zs->zs_num_blocks = zap->zap_f.zap_phys->zap_freeblk;
1272         zs->zs_block_type = zap->zap_f.zap_phys->zap_block_type;
1273         zs->zs_magic = zap->zap_f.zap_phys->zap_magic;
1274         zs->zs_salt = zap->zap_f.zap_phys->zap_salt;
1275 
1276         /*
1277          * Set zap_ptrtbl fields
1278          */
1279         zs->zs_ptrtbl_len = 1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift;
1280         zs->zs_ptrtbl_nextblk = zap->zap_f.zap_phys->zap_ptrtbl.zt_nextblk;
1281         zs->zs_ptrtbl_blks_copied =
1282             zap->zap_f.zap_phys->zap_ptrtbl.zt_blks_copied;
1283         zs->zs_ptrtbl_zt_blk = zap->zap_f.zap_phys->zap_ptrtbl.zt_blk;
1284         zs->zs_ptrtbl_zt_numblks = zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks;
1285         zs->zs_ptrtbl_zt_shift = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift;
1286 
1287         if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) {
1288                 /* the ptrtbl is entirely in the header block. */
1289                 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1290                     1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1291         } else {
1292                 int b;
1293 
1294                 dmu_prefetch(zap->zap_objset, zap->zap_object,
1295                     zap->zap_f.zap_phys->zap_ptrtbl.zt_blk << bs,
1296                     zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks << bs);
1297 
1298                 for (b = 0; b < zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks;
1299                     b++) {
1300                         dmu_buf_t *db;
1301                         int err;
1302 
1303                         err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1304                             (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk + b) << bs,
1305                             FTAG, &db, DMU_READ_NO_PREFETCH);
1306                         if (err == 0) {
1307                                 zap_stats_ptrtbl(zap, db->db_data,
1308                                     1<<(bs-3), zs);
1309                                 dmu_buf_rele(db, FTAG);
1310                         }
1311                 }
1312         }
1313 }
1314 
1315 int
1316 fzap_count_write(zap_name_t *zn, int add, uint64_t *towrite,
1317     uint64_t *tooverwrite)
1318 {
1319         zap_t *zap = zn->zn_zap;
1320         zap_leaf_t *l;
1321         int err;
1322 
1323         /*
1324          * Account for the header block of the fatzap.
1325          */
1326         if (!add && dmu_buf_freeable(zap->zap_dbuf)) {
1327                 *tooverwrite += zap->zap_dbuf->db_size;
1328         } else {
1329                 *towrite += zap->zap_dbuf->db_size;
1330         }
1331 
1332         /*
1333          * Account for the pointer table blocks.
1334          * If we are adding we need to account for the following cases :
1335          * - If the pointer table is embedded, this operation could force an
1336          *   external pointer table.
1337          * - If this already has an external pointer table this operation
1338          *   could extend the table.
1339          */
1340         if (add) {
1341                 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk == 0)
1342                         *towrite += zap->zap_dbuf->db_size;
1343                 else
1344                         *towrite += (zap->zap_dbuf->db_size * 3);
1345         }
1346 
1347         /*
1348          * Now, check if the block containing leaf is freeable
1349          * and account accordingly.
1350          */
1351         err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l);
1352         if (err != 0) {
1353                 return (err);
1354         }
1355 
1356         if (!add && dmu_buf_freeable(l->l_dbuf)) {
1357                 *tooverwrite += l->l_dbuf->db_size;
1358         } else {
1359                 /*
1360                  * If this an add operation, the leaf block could split.
1361                  * Hence, we need to account for an additional leaf block.
1362                  */
1363                 *towrite += (add ? 2 : 1) * l->l_dbuf->db_size;
1364         }
1365 
1366         zap_put_leaf(l);
1367         return (0);
1368 }