zap.c revision 307292
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, 2016 by Delphix. All rights reserved. 24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 25 */ 26 27/* 28 * This file contains the top half of the zfs directory structure 29 * implementation. The bottom half is in zap_leaf.c. 30 * 31 * The zdir is an extendable hash data structure. There is a table of 32 * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are 33 * each a constant size and hold a variable number of directory entries. 34 * The buckets (aka "leaf nodes") are implemented in zap_leaf.c. 35 * 36 * The pointer table holds a power of 2 number of pointers. 37 * (1<<zap_t->zd_data->zd_phys->zd_prefix_len). The bucket pointed to 38 * by the pointer at index i in the table holds entries whose hash value 39 * has a zd_prefix_len - bit prefix 40 */ 41 42#include <sys/spa.h> 43#include <sys/dmu.h> 44#include <sys/zfs_context.h> 45#include <sys/zfs_znode.h> 46#include <sys/fs/zfs.h> 47#include <sys/zap.h> 48#include <sys/refcount.h> 49#include <sys/zap_impl.h> 50#include <sys/zap_leaf.h> 51 52int fzap_default_block_shift = 14; /* 16k blocksize */ 53 54extern inline zap_phys_t *zap_f_phys(zap_t *zap); 55 56static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks); 57 58void 59fzap_byteswap(void *vbuf, size_t size) 60{ 61 uint64_t block_type; 62 63 block_type = *(uint64_t *)vbuf; 64 65 if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF)) 66 zap_leaf_byteswap(vbuf, size); 67 else { 68 /* it's a ptrtbl block */ 69 byteswap_uint64_array(vbuf, size); 70 } 71} 72 73void 74fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags) 75{ 76 dmu_buf_t *db; 77 zap_leaf_t *l; 78 int i; 79 zap_phys_t *zp; 80 81 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 82 zap->zap_ismicro = FALSE; 83 84 zap->zap_dbu.dbu_evict_func = zap_evict; 85 86 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0); 87 zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1; 88 89 zp = zap_f_phys(zap); 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; 120 121 zap_leaf_init(l, zp->zap_normflags != 0); 122 123 kmem_free(l, sizeof (zap_leaf_t)); 124 dmu_buf_rele(db, FTAG); 125} 126 127static int 128zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx) 129{ 130 if (RW_WRITE_HELD(&zap->zap_rwlock)) 131 return (1); 132 if (rw_tryupgrade(&zap->zap_rwlock)) { 133 dmu_buf_will_dirty(zap->zap_dbuf, tx); 134 return (1); 135 } 136 return (0); 137} 138 139/* 140 * Generic routines for dealing with the pointer & cookie tables. 141 */ 142 143static int 144zap_table_grow(zap_t *zap, zap_table_phys_t *tbl, 145 void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n), 146 dmu_tx_t *tx) 147{ 148 uint64_t b, newblk; 149 dmu_buf_t *db_old, *db_new; 150 int err; 151 int bs = FZAP_BLOCK_SHIFT(zap); 152 int hepb = 1<<(bs-4); 153 /* hepb = half the number of entries in a block */ 154 155 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 156 ASSERT(tbl->zt_blk != 0); 157 ASSERT(tbl->zt_numblks > 0); 158 159 if (tbl->zt_nextblk != 0) { 160 newblk = tbl->zt_nextblk; 161 } else { 162 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2); 163 tbl->zt_nextblk = newblk; 164 ASSERT0(tbl->zt_blks_copied); 165 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, 166 tbl->zt_blk << bs, tbl->zt_numblks << bs, 167 ZIO_PRIORITY_SYNC_READ); 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 219static int 220zap_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 267static int 268zap_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 dnode_t *dn; 274 int bs = FZAP_BLOCK_SHIFT(zap); 275 276 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 277 278 blk = idx >> (bs-3); 279 off = idx & ((1<<(bs-3))-1); 280 281 /* 282 * Note: this is equivalent to dmu_buf_hold(), but we use 283 * _dnode_enter / _by_dnode because it's faster because we don't 284 * have to hold the dnode. 285 */ 286 dn = dmu_buf_dnode_enter(zap->zap_dbuf); 287 err = dmu_buf_hold_by_dnode(dn, 288 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH); 289 dmu_buf_dnode_exit(zap->zap_dbuf); 290 if (err) 291 return (err); 292 *valp = ((uint64_t *)db->db_data)[off]; 293 dmu_buf_rele(db, FTAG); 294 295 if (tbl->zt_nextblk != 0) { 296 /* 297 * read the nextblk for the sake of i/o error checking, 298 * so that zap_table_load() will catch errors for 299 * zap_table_store. 300 */ 301 blk = (idx*2) >> (bs-3); 302 303 dn = dmu_buf_dnode_enter(zap->zap_dbuf); 304 err = dmu_buf_hold_by_dnode(dn, 305 (tbl->zt_nextblk + blk) << bs, FTAG, &db, 306 DMU_READ_NO_PREFETCH); 307 dmu_buf_dnode_exit(zap->zap_dbuf); 308 if (err == 0) 309 dmu_buf_rele(db, FTAG); 310 } 311 return (err); 312} 313 314/* 315 * Routines for growing the ptrtbl. 316 */ 317 318static void 319zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n) 320{ 321 int i; 322 for (i = 0; i < n; i++) { 323 uint64_t lb = src[i]; 324 dst[2*i+0] = lb; 325 dst[2*i+1] = lb; 326 } 327} 328 329static int 330zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx) 331{ 332 /* 333 * The pointer table should never use more hash bits than we 334 * have (otherwise we'd be using useless zero bits to index it). 335 * If we are within 2 bits of running out, stop growing, since 336 * this is already an aberrant condition. 337 */ 338 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2) 339 return (SET_ERROR(ENOSPC)); 340 341 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) { 342 /* 343 * We are outgrowing the "embedded" ptrtbl (the one 344 * stored in the header block). Give it its own entire 345 * block, which will double the size of the ptrtbl. 346 */ 347 uint64_t newblk; 348 dmu_buf_t *db_new; 349 int err; 350 351 ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==, 352 ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); 353 ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk); 354 355 newblk = zap_allocate_blocks(zap, 1); 356 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 357 newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new, 358 DMU_READ_NO_PREFETCH); 359 if (err) 360 return (err); 361 dmu_buf_will_dirty(db_new, tx); 362 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), 363 db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); 364 dmu_buf_rele(db_new, FTAG); 365 366 zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk; 367 zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1; 368 zap_f_phys(zap)->zap_ptrtbl.zt_shift++; 369 370 ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==, 371 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << 372 (FZAP_BLOCK_SHIFT(zap)-3)); 373 374 return (0); 375 } else { 376 return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl, 377 zap_ptrtbl_transfer, tx)); 378 } 379} 380 381static void 382zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx) 383{ 384 dmu_buf_will_dirty(zap->zap_dbuf, tx); 385 mutex_enter(&zap->zap_f.zap_num_entries_mtx); 386 ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta); 387 zap_f_phys(zap)->zap_num_entries += delta; 388 mutex_exit(&zap->zap_f.zap_num_entries_mtx); 389} 390 391static uint64_t 392zap_allocate_blocks(zap_t *zap, int nblocks) 393{ 394 uint64_t newblk; 395 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 396 newblk = zap_f_phys(zap)->zap_freeblk; 397 zap_f_phys(zap)->zap_freeblk += nblocks; 398 return (newblk); 399} 400 401static void 402zap_leaf_pageout(void *dbu) 403{ 404 zap_leaf_t *l = dbu; 405 406 rw_destroy(&l->l_rwlock); 407 kmem_free(l, sizeof (zap_leaf_t)); 408} 409 410static zap_leaf_t * 411zap_create_leaf(zap_t *zap, dmu_tx_t *tx) 412{ 413 void *winner; 414 zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); 415 416 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 417 418 rw_init(&l->l_rwlock, 0, 0, 0); 419 rw_enter(&l->l_rwlock, RW_WRITER); 420 l->l_blkid = zap_allocate_blocks(zap, 1); 421 l->l_dbuf = NULL; 422 423 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 424 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf, 425 DMU_READ_NO_PREFETCH)); 426 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf); 427 winner = dmu_buf_set_user(l->l_dbuf, &l->l_dbu); 428 ASSERT(winner == NULL); 429 dmu_buf_will_dirty(l->l_dbuf, tx); 430 431 zap_leaf_init(l, zap->zap_normflags != 0); 432 433 zap_f_phys(zap)->zap_num_leafs++; 434 435 return (l); 436} 437 438int 439fzap_count(zap_t *zap, uint64_t *count) 440{ 441 ASSERT(!zap->zap_ismicro); 442 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */ 443 *count = zap_f_phys(zap)->zap_num_entries; 444 mutex_exit(&zap->zap_f.zap_num_entries_mtx); 445 return (0); 446} 447 448/* 449 * Routines for obtaining zap_leaf_t's 450 */ 451 452void 453zap_put_leaf(zap_leaf_t *l) 454{ 455 rw_exit(&l->l_rwlock); 456 dmu_buf_rele(l->l_dbuf, NULL); 457} 458 459static zap_leaf_t * 460zap_open_leaf(uint64_t blkid, dmu_buf_t *db) 461{ 462 zap_leaf_t *l, *winner; 463 464 ASSERT(blkid != 0); 465 466 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); 467 rw_init(&l->l_rwlock, 0, 0, 0); 468 rw_enter(&l->l_rwlock, RW_WRITER); 469 l->l_blkid = blkid; 470 l->l_bs = highbit64(db->db_size) - 1; 471 l->l_dbuf = db; 472 473 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf); 474 winner = dmu_buf_set_user(db, &l->l_dbu); 475 476 rw_exit(&l->l_rwlock); 477 if (winner != NULL) { 478 /* someone else set it first */ 479 zap_leaf_pageout(&l->l_dbu); 480 l = winner; 481 } 482 483 /* 484 * lhr_pad was previously used for the next leaf in the leaf 485 * chain. There should be no chained leafs (as we have removed 486 * support for them). 487 */ 488 ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1); 489 490 /* 491 * There should be more hash entries than there can be 492 * chunks to put in the hash table 493 */ 494 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3); 495 496 /* The chunks should begin at the end of the hash table */ 497 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==, 498 &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]); 499 500 /* The chunks should end at the end of the block */ 501 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) - 502 (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size); 503 504 return (l); 505} 506 507static int 508zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt, 509 zap_leaf_t **lp) 510{ 511 dmu_buf_t *db; 512 zap_leaf_t *l; 513 int bs = FZAP_BLOCK_SHIFT(zap); 514 int err; 515 516 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 517 518 dnode_t *dn = dmu_buf_dnode_enter(zap->zap_dbuf); 519 err = dmu_buf_hold_by_dnode(dn, 520 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH); 521 dmu_buf_dnode_exit(zap->zap_dbuf); 522 if (err) 523 return (err); 524 525 ASSERT3U(db->db_object, ==, zap->zap_object); 526 ASSERT3U(db->db_offset, ==, blkid << bs); 527 ASSERT3U(db->db_size, ==, 1 << bs); 528 ASSERT(blkid != 0); 529 530 l = dmu_buf_get_user(db); 531 532 if (l == NULL) 533 l = zap_open_leaf(blkid, db); 534 535 rw_enter(&l->l_rwlock, lt); 536 /* 537 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change, 538 * causing ASSERT below to fail. 539 */ 540 if (lt == RW_WRITER) 541 dmu_buf_will_dirty(db, tx); 542 ASSERT3U(l->l_blkid, ==, blkid); 543 ASSERT3P(l->l_dbuf, ==, db); 544 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF); 545 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC); 546 547 *lp = l; 548 return (0); 549} 550 551static int 552zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp) 553{ 554 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 555 556 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) { 557 ASSERT3U(idx, <, 558 (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift)); 559 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx); 560 return (0); 561 } else { 562 return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl, 563 idx, valp)); 564 } 565} 566 567static int 568zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx) 569{ 570 ASSERT(tx != NULL); 571 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 572 573 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) { 574 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk; 575 return (0); 576 } else { 577 return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl, 578 idx, blk, tx)); 579 } 580} 581 582static int 583zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp) 584{ 585 uint64_t idx, blk; 586 int err; 587 588 ASSERT(zap->zap_dbuf == NULL || 589 zap_f_phys(zap) == zap->zap_dbuf->db_data); 590 ASSERT3U(zap_f_phys(zap)->zap_magic, ==, ZAP_MAGIC); 591 idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift); 592 err = zap_idx_to_blk(zap, idx, &blk); 593 if (err != 0) 594 return (err); 595 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp); 596 597 ASSERT(err || 598 ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) == 599 zap_leaf_phys(*lp)->l_hdr.lh_prefix); 600 return (err); 601} 602 603static int 604zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, 605 void *tag, dmu_tx_t *tx, zap_leaf_t **lp) 606{ 607 zap_t *zap = zn->zn_zap; 608 uint64_t hash = zn->zn_hash; 609 zap_leaf_t *nl; 610 int prefix_diff, i, err; 611 uint64_t sibling; 612 int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len; 613 614 ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift); 615 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 616 617 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, 618 zap_leaf_phys(l)->l_hdr.lh_prefix); 619 620 if (zap_tryupgradedir(zap, tx) == 0 || 621 old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) { 622 /* We failed to upgrade, or need to grow the pointer table */ 623 objset_t *os = zap->zap_objset; 624 uint64_t object = zap->zap_object; 625 626 zap_put_leaf(l); 627 zap_unlockdir(zap, tag); 628 err = zap_lockdir(os, object, tx, RW_WRITER, 629 FALSE, FALSE, tag, &zn->zn_zap); 630 zap = zn->zn_zap; 631 if (err) 632 return (err); 633 ASSERT(!zap->zap_ismicro); 634 635 while (old_prefix_len == 636 zap_f_phys(zap)->zap_ptrtbl.zt_shift) { 637 err = zap_grow_ptrtbl(zap, tx); 638 if (err) 639 return (err); 640 } 641 642 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l); 643 if (err) 644 return (err); 645 646 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) { 647 /* it split while our locks were down */ 648 *lp = l; 649 return (0); 650 } 651 } 652 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 653 ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift); 654 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, 655 zap_leaf_phys(l)->l_hdr.lh_prefix); 656 657 prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift - 658 (old_prefix_len + 1); 659 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff; 660 661 /* check for i/o errors before doing zap_leaf_split */ 662 for (i = 0; i < (1ULL<<prefix_diff); i++) { 663 uint64_t blk; 664 err = zap_idx_to_blk(zap, sibling+i, &blk); 665 if (err) 666 return (err); 667 ASSERT3U(blk, ==, l->l_blkid); 668 } 669 670 nl = zap_create_leaf(zap, tx); 671 zap_leaf_split(l, nl, zap->zap_normflags != 0); 672 673 /* set sibling pointers */ 674 for (i = 0; i < (1ULL << prefix_diff); i++) { 675 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx); 676 ASSERT0(err); /* we checked for i/o errors above */ 677 } 678 679 if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) { 680 /* we want the sibling */ 681 zap_put_leaf(l); 682 *lp = nl; 683 } else { 684 zap_put_leaf(nl); 685 *lp = l; 686 } 687 688 return (0); 689} 690 691static void 692zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, 693 void *tag, dmu_tx_t *tx) 694{ 695 zap_t *zap = zn->zn_zap; 696 int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift; 697 int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift && 698 zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER); 699 700 zap_put_leaf(l); 701 702 if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) { 703 int err; 704 705 /* 706 * We are in the middle of growing the pointer table, or 707 * this leaf will soon make us grow it. 708 */ 709 if (zap_tryupgradedir(zap, tx) == 0) { 710 objset_t *os = zap->zap_objset; 711 uint64_t zapobj = zap->zap_object; 712 713 zap_unlockdir(zap, tag); 714 err = zap_lockdir(os, zapobj, tx, 715 RW_WRITER, FALSE, FALSE, tag, &zn->zn_zap); 716 zap = zn->zn_zap; 717 if (err) 718 return; 719 } 720 721 /* could have finished growing while our locks were down */ 722 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift) 723 (void) zap_grow_ptrtbl(zap, tx); 724 } 725} 726 727static int 728fzap_checkname(zap_name_t *zn) 729{ 730 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN) 731 return (SET_ERROR(ENAMETOOLONG)); 732 return (0); 733} 734 735static int 736fzap_checksize(uint64_t integer_size, uint64_t num_integers) 737{ 738 /* Only integer sizes supported by C */ 739 switch (integer_size) { 740 case 1: 741 case 2: 742 case 4: 743 case 8: 744 break; 745 default: 746 return (SET_ERROR(EINVAL)); 747 } 748 749 if (integer_size * num_integers > ZAP_MAXVALUELEN) 750 return (E2BIG); 751 752 return (0); 753} 754 755static int 756fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers) 757{ 758 int err; 759 760 if ((err = fzap_checkname(zn)) != 0) 761 return (err); 762 return (fzap_checksize(integer_size, num_integers)); 763} 764 765/* 766 * Routines for manipulating attributes. 767 */ 768int 769fzap_lookup(zap_name_t *zn, 770 uint64_t integer_size, uint64_t num_integers, void *buf, 771 char *realname, int rn_len, boolean_t *ncp) 772{ 773 zap_leaf_t *l; 774 int err; 775 zap_entry_handle_t zeh; 776 777 if ((err = fzap_checkname(zn)) != 0) 778 return (err); 779 780 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); 781 if (err != 0) 782 return (err); 783 err = zap_leaf_lookup(l, zn, &zeh); 784 if (err == 0) { 785 if ((err = fzap_checksize(integer_size, num_integers)) != 0) { 786 zap_put_leaf(l); 787 return (err); 788 } 789 790 err = zap_entry_read(&zeh, integer_size, num_integers, buf); 791 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname); 792 if (ncp) { 793 *ncp = zap_entry_normalization_conflict(&zeh, 794 zn, NULL, zn->zn_zap); 795 } 796 } 797 798 zap_put_leaf(l); 799 return (err); 800} 801 802int 803fzap_add_cd(zap_name_t *zn, 804 uint64_t integer_size, uint64_t num_integers, 805 const void *val, uint32_t cd, void *tag, dmu_tx_t *tx) 806{ 807 zap_leaf_t *l; 808 int err; 809 zap_entry_handle_t zeh; 810 zap_t *zap = zn->zn_zap; 811 812 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 813 ASSERT(!zap->zap_ismicro); 814 ASSERT(fzap_check(zn, integer_size, num_integers) == 0); 815 816 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); 817 if (err != 0) 818 return (err); 819retry: 820 err = zap_leaf_lookup(l, zn, &zeh); 821 if (err == 0) { 822 err = SET_ERROR(EEXIST); 823 goto out; 824 } 825 if (err != ENOENT) 826 goto out; 827 828 err = zap_entry_create(l, zn, cd, 829 integer_size, num_integers, val, &zeh); 830 831 if (err == 0) { 832 zap_increment_num_entries(zap, 1, tx); 833 } else if (err == EAGAIN) { 834 err = zap_expand_leaf(zn, l, tag, tx, &l); 835 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ 836 if (err == 0) 837 goto retry; 838 } 839 840out: 841 if (zap != NULL) 842 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx); 843 return (err); 844} 845 846int 847fzap_add(zap_name_t *zn, 848 uint64_t integer_size, uint64_t num_integers, 849 const void *val, void *tag, dmu_tx_t *tx) 850{ 851 int err = fzap_check(zn, integer_size, num_integers); 852 if (err != 0) 853 return (err); 854 855 return (fzap_add_cd(zn, integer_size, num_integers, 856 val, ZAP_NEED_CD, tag, tx)); 857} 858 859int 860fzap_update(zap_name_t *zn, 861 int integer_size, uint64_t num_integers, const void *val, 862 void *tag, dmu_tx_t *tx) 863{ 864 zap_leaf_t *l; 865 int err, create; 866 zap_entry_handle_t zeh; 867 zap_t *zap = zn->zn_zap; 868 869 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 870 err = fzap_check(zn, integer_size, num_integers); 871 if (err != 0) 872 return (err); 873 874 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); 875 if (err != 0) 876 return (err); 877retry: 878 err = zap_leaf_lookup(l, zn, &zeh); 879 create = (err == ENOENT); 880 ASSERT(err == 0 || err == ENOENT); 881 882 if (create) { 883 err = zap_entry_create(l, zn, ZAP_NEED_CD, 884 integer_size, num_integers, val, &zeh); 885 if (err == 0) 886 zap_increment_num_entries(zap, 1, tx); 887 } else { 888 err = zap_entry_update(&zeh, integer_size, num_integers, val); 889 } 890 891 if (err == EAGAIN) { 892 err = zap_expand_leaf(zn, l, tag, tx, &l); 893 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ 894 if (err == 0) 895 goto retry; 896 } 897 898 if (zap != NULL) 899 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx); 900 return (err); 901} 902 903int 904fzap_length(zap_name_t *zn, 905 uint64_t *integer_size, uint64_t *num_integers) 906{ 907 zap_leaf_t *l; 908 int err; 909 zap_entry_handle_t zeh; 910 911 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); 912 if (err != 0) 913 return (err); 914 err = zap_leaf_lookup(l, zn, &zeh); 915 if (err != 0) 916 goto out; 917 918 if (integer_size) 919 *integer_size = zeh.zeh_integer_size; 920 if (num_integers) 921 *num_integers = zeh.zeh_num_integers; 922out: 923 zap_put_leaf(l); 924 return (err); 925} 926 927int 928fzap_remove(zap_name_t *zn, dmu_tx_t *tx) 929{ 930 zap_leaf_t *l; 931 int err; 932 zap_entry_handle_t zeh; 933 934 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l); 935 if (err != 0) 936 return (err); 937 err = zap_leaf_lookup(l, zn, &zeh); 938 if (err == 0) { 939 zap_entry_remove(&zeh); 940 zap_increment_num_entries(zn->zn_zap, -1, tx); 941 } 942 zap_put_leaf(l); 943 return (err); 944} 945 946void 947fzap_prefetch(zap_name_t *zn) 948{ 949 uint64_t idx, blk; 950 zap_t *zap = zn->zn_zap; 951 int bs; 952 953 idx = ZAP_HASH_IDX(zn->zn_hash, 954 zap_f_phys(zap)->zap_ptrtbl.zt_shift); 955 if (zap_idx_to_blk(zap, idx, &blk) != 0) 956 return; 957 bs = FZAP_BLOCK_SHIFT(zap); 958 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs, 959 ZIO_PRIORITY_SYNC_READ); 960} 961 962/* 963 * Helper functions for consumers. 964 */ 965 966uint64_t 967zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj, 968 const char *name, dmu_tx_t *tx) 969{ 970 uint64_t new_obj; 971 972 VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0); 973 VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj, 974 tx) == 0); 975 976 return (new_obj); 977} 978 979int 980zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask, 981 char *name) 982{ 983 zap_cursor_t zc; 984 zap_attribute_t *za; 985 int err; 986 987 if (mask == 0) 988 mask = -1ULL; 989 990 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 991 for (zap_cursor_init(&zc, os, zapobj); 992 (err = zap_cursor_retrieve(&zc, za)) == 0; 993 zap_cursor_advance(&zc)) { 994 if ((za->za_first_integer & mask) == (value & mask)) { 995 (void) strcpy(name, za->za_name); 996 break; 997 } 998 } 999 zap_cursor_fini(&zc); 1000 kmem_free(za, sizeof (zap_attribute_t)); 1001 return (err); 1002} 1003 1004int 1005zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx) 1006{ 1007 zap_cursor_t zc; 1008 zap_attribute_t za; 1009 int err; 1010 1011 err = 0; 1012 for (zap_cursor_init(&zc, os, fromobj); 1013 zap_cursor_retrieve(&zc, &za) == 0; 1014 (void) zap_cursor_advance(&zc)) { 1015 if (za.za_integer_length != 8 || za.za_num_integers != 1) { 1016 err = SET_ERROR(EINVAL); 1017 break; 1018 } 1019 err = zap_add(os, intoobj, za.za_name, 1020 8, 1, &za.za_first_integer, tx); 1021 if (err) 1022 break; 1023 } 1024 zap_cursor_fini(&zc); 1025 return (err); 1026} 1027 1028int 1029zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj, 1030 uint64_t value, dmu_tx_t *tx) 1031{ 1032 zap_cursor_t zc; 1033 zap_attribute_t za; 1034 int err; 1035 1036 err = 0; 1037 for (zap_cursor_init(&zc, os, fromobj); 1038 zap_cursor_retrieve(&zc, &za) == 0; 1039 (void) zap_cursor_advance(&zc)) { 1040 if (za.za_integer_length != 8 || za.za_num_integers != 1) { 1041 err = SET_ERROR(EINVAL); 1042 break; 1043 } 1044 err = zap_add(os, intoobj, za.za_name, 1045 8, 1, &value, tx); 1046 if (err) 1047 break; 1048 } 1049 zap_cursor_fini(&zc); 1050 return (err); 1051} 1052 1053int 1054zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj, 1055 dmu_tx_t *tx) 1056{ 1057 zap_cursor_t zc; 1058 zap_attribute_t za; 1059 int err; 1060 1061 err = 0; 1062 for (zap_cursor_init(&zc, os, fromobj); 1063 zap_cursor_retrieve(&zc, &za) == 0; 1064 (void) zap_cursor_advance(&zc)) { 1065 uint64_t delta = 0; 1066 1067 if (za.za_integer_length != 8 || za.za_num_integers != 1) { 1068 err = SET_ERROR(EINVAL); 1069 break; 1070 } 1071 1072 err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta); 1073 if (err != 0 && err != ENOENT) 1074 break; 1075 delta += za.za_first_integer; 1076 err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx); 1077 if (err) 1078 break; 1079 } 1080 zap_cursor_fini(&zc); 1081 return (err); 1082} 1083 1084int 1085zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) 1086{ 1087 char name[20]; 1088 1089 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 1090 return (zap_add(os, obj, name, 8, 1, &value, tx)); 1091} 1092 1093int 1094zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) 1095{ 1096 char name[20]; 1097 1098 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 1099 return (zap_remove(os, obj, name, tx)); 1100} 1101 1102int 1103zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value) 1104{ 1105 char name[20]; 1106 1107 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 1108 return (zap_lookup(os, obj, name, 8, 1, &value)); 1109} 1110 1111int 1112zap_add_int_key(objset_t *os, uint64_t obj, 1113 uint64_t key, uint64_t value, dmu_tx_t *tx) 1114{ 1115 char name[20]; 1116 1117 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 1118 return (zap_add(os, obj, name, 8, 1, &value, tx)); 1119} 1120 1121int 1122zap_update_int_key(objset_t *os, uint64_t obj, 1123 uint64_t key, uint64_t value, dmu_tx_t *tx) 1124{ 1125 char name[20]; 1126 1127 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 1128 return (zap_update(os, obj, name, 8, 1, &value, tx)); 1129} 1130 1131int 1132zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep) 1133{ 1134 char name[20]; 1135 1136 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 1137 return (zap_lookup(os, obj, name, 8, 1, valuep)); 1138} 1139 1140int 1141zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta, 1142 dmu_tx_t *tx) 1143{ 1144 uint64_t value = 0; 1145 int err; 1146 1147 if (delta == 0) 1148 return (0); 1149 1150 err = zap_lookup(os, obj, name, 8, 1, &value); 1151 if (err != 0 && err != ENOENT) 1152 return (err); 1153 value += delta; 1154 if (value == 0) 1155 err = zap_remove(os, obj, name, tx); 1156 else 1157 err = zap_update(os, obj, name, 8, 1, &value, tx); 1158 return (err); 1159} 1160 1161int 1162zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta, 1163 dmu_tx_t *tx) 1164{ 1165 char name[20]; 1166 1167 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 1168 return (zap_increment(os, obj, name, delta, tx)); 1169} 1170 1171/* 1172 * Routines for iterating over the attributes. 1173 */ 1174 1175int 1176fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za) 1177{ 1178 int err = ENOENT; 1179 zap_entry_handle_t zeh; 1180 zap_leaf_t *l; 1181 1182 /* retrieve the next entry at or after zc_hash/zc_cd */ 1183 /* if no entry, return ENOENT */ 1184 1185 if (zc->zc_leaf && 1186 (ZAP_HASH_IDX(zc->zc_hash, 1187 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) != 1188 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) { 1189 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1190 zap_put_leaf(zc->zc_leaf); 1191 zc->zc_leaf = NULL; 1192 } 1193 1194again: 1195 if (zc->zc_leaf == NULL) { 1196 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER, 1197 &zc->zc_leaf); 1198 if (err != 0) 1199 return (err); 1200 } else { 1201 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1202 } 1203 l = zc->zc_leaf; 1204 1205 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh); 1206 1207 if (err == ENOENT) { 1208 uint64_t nocare = 1209 (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1; 1210 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1; 1211 zc->zc_cd = 0; 1212 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 || 1213 zc->zc_hash == 0) { 1214 zc->zc_hash = -1ULL; 1215 } else { 1216 zap_put_leaf(zc->zc_leaf); 1217 zc->zc_leaf = NULL; 1218 goto again; 1219 } 1220 } 1221 1222 if (err == 0) { 1223 zc->zc_hash = zeh.zeh_hash; 1224 zc->zc_cd = zeh.zeh_cd; 1225 za->za_integer_length = zeh.zeh_integer_size; 1226 za->za_num_integers = zeh.zeh_num_integers; 1227 if (zeh.zeh_num_integers == 0) { 1228 za->za_first_integer = 0; 1229 } else { 1230 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer); 1231 ASSERT(err == 0 || err == EOVERFLOW); 1232 } 1233 err = zap_entry_read_name(zap, &zeh, 1234 sizeof (za->za_name), za->za_name); 1235 ASSERT(err == 0); 1236 1237 za->za_normalization_conflict = 1238 zap_entry_normalization_conflict(&zeh, 1239 NULL, za->za_name, zap); 1240 } 1241 rw_exit(&zc->zc_leaf->l_rwlock); 1242 return (err); 1243} 1244 1245static void 1246zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs) 1247{ 1248 int i, err; 1249 uint64_t lastblk = 0; 1250 1251 /* 1252 * NB: if a leaf has more pointers than an entire ptrtbl block 1253 * can hold, then it'll be accounted for more than once, since 1254 * we won't have lastblk. 1255 */ 1256 for (i = 0; i < len; i++) { 1257 zap_leaf_t *l; 1258 1259 if (tbl[i] == lastblk) 1260 continue; 1261 lastblk = tbl[i]; 1262 1263 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l); 1264 if (err == 0) { 1265 zap_leaf_stats(zap, l, zs); 1266 zap_put_leaf(l); 1267 } 1268 } 1269} 1270 1271int 1272fzap_cursor_move_to_key(zap_cursor_t *zc, zap_name_t *zn) 1273{ 1274 int err; 1275 zap_leaf_t *l; 1276 zap_entry_handle_t zeh; 1277 1278 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN) 1279 return (SET_ERROR(ENAMETOOLONG)); 1280 1281 err = zap_deref_leaf(zc->zc_zap, zn->zn_hash, NULL, RW_READER, &l); 1282 if (err != 0) 1283 return (err); 1284 1285 err = zap_leaf_lookup(l, zn, &zeh); 1286 if (err != 0) 1287 return (err); 1288 1289 zc->zc_leaf = l; 1290 zc->zc_hash = zeh.zeh_hash; 1291 zc->zc_cd = zeh.zeh_cd; 1292 1293 return (err); 1294} 1295 1296void 1297fzap_get_stats(zap_t *zap, zap_stats_t *zs) 1298{ 1299 int bs = FZAP_BLOCK_SHIFT(zap); 1300 zs->zs_blocksize = 1ULL << bs; 1301 1302 /* 1303 * Set zap_phys_t fields 1304 */ 1305 zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs; 1306 zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries; 1307 zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk; 1308 zs->zs_block_type = zap_f_phys(zap)->zap_block_type; 1309 zs->zs_magic = zap_f_phys(zap)->zap_magic; 1310 zs->zs_salt = zap_f_phys(zap)->zap_salt; 1311 1312 /* 1313 * Set zap_ptrtbl fields 1314 */ 1315 zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift; 1316 zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk; 1317 zs->zs_ptrtbl_blks_copied = 1318 zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied; 1319 zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk; 1320 zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks; 1321 zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift; 1322 1323 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) { 1324 /* the ptrtbl is entirely in the header block. */ 1325 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), 1326 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs); 1327 } else { 1328 int b; 1329 1330 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, 1331 zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs, 1332 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs, 1333 ZIO_PRIORITY_SYNC_READ); 1334 1335 for (b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks; 1336 b++) { 1337 dmu_buf_t *db; 1338 int err; 1339 1340 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 1341 (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs, 1342 FTAG, &db, DMU_READ_NO_PREFETCH); 1343 if (err == 0) { 1344 zap_stats_ptrtbl(zap, db->db_data, 1345 1<<(bs-3), zs); 1346 dmu_buf_rele(db, FTAG); 1347 } 1348 } 1349 } 1350} 1351 1352int 1353fzap_count_write(zap_name_t *zn, int add, refcount_t *towrite, 1354 refcount_t *tooverwrite) 1355{ 1356 zap_t *zap = zn->zn_zap; 1357 zap_leaf_t *l; 1358 int err; 1359 1360 /* 1361 * Account for the header block of the fatzap. 1362 */ 1363 if (!add && dmu_buf_freeable(zap->zap_dbuf)) { 1364 (void) refcount_add_many(tooverwrite, 1365 zap->zap_dbuf->db_size, FTAG); 1366 } else { 1367 (void) refcount_add_many(towrite, 1368 zap->zap_dbuf->db_size, FTAG); 1369 } 1370 1371 /* 1372 * Account for the pointer table blocks. 1373 * If we are adding we need to account for the following cases : 1374 * - If the pointer table is embedded, this operation could force an 1375 * external pointer table. 1376 * - If this already has an external pointer table this operation 1377 * could extend the table. 1378 */ 1379 if (add) { 1380 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) { 1381 (void) refcount_add_many(towrite, 1382 zap->zap_dbuf->db_size, FTAG); 1383 } else { 1384 (void) refcount_add_many(towrite, 1385 zap->zap_dbuf->db_size * 3, FTAG); 1386 } 1387 } 1388 1389 /* 1390 * Now, check if the block containing leaf is freeable 1391 * and account accordingly. 1392 */ 1393 err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l); 1394 if (err != 0) { 1395 return (err); 1396 } 1397 1398 if (!add && dmu_buf_freeable(l->l_dbuf)) { 1399 (void) refcount_add_many(tooverwrite, l->l_dbuf->db_size, FTAG); 1400 } else { 1401 /* 1402 * If this an add operation, the leaf block could split. 1403 * Hence, we need to account for an additional leaf block. 1404 */ 1405 (void) refcount_add_many(towrite, 1406 (add ? 2 : 1) * l->l_dbuf->db_size, FTAG); 1407 } 1408 1409 zap_put_leaf(l); 1410 return (0); 1411} 1412