zap_micro.c revision 219089
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 */ 24 25#include <sys/zio.h> 26#include <sys/spa.h> 27#include <sys/dmu.h> 28#include <sys/zfs_context.h> 29#include <sys/zap.h> 30#include <sys/refcount.h> 31#include <sys/zap_impl.h> 32#include <sys/zap_leaf.h> 33#include <sys/avl.h> 34#include <sys/arc.h> 35 36#ifdef _KERNEL 37#include <sys/sunddi.h> 38#endif 39 40static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags); 41 42uint64_t 43zap_getflags(zap_t *zap) 44{ 45 if (zap->zap_ismicro) 46 return (0); 47 return (zap->zap_u.zap_fat.zap_phys->zap_flags); 48} 49 50int 51zap_hashbits(zap_t *zap) 52{ 53 if (zap_getflags(zap) & ZAP_FLAG_HASH64) 54 return (48); 55 else 56 return (28); 57} 58 59uint32_t 60zap_maxcd(zap_t *zap) 61{ 62 if (zap_getflags(zap) & ZAP_FLAG_HASH64) 63 return ((1<<16)-1); 64 else 65 return (-1U); 66} 67 68static uint64_t 69zap_hash(zap_name_t *zn) 70{ 71 zap_t *zap = zn->zn_zap; 72 uint64_t h = 0; 73 74 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) { 75 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY); 76 h = *(uint64_t *)zn->zn_key_orig; 77 } else { 78 h = zap->zap_salt; 79 ASSERT(h != 0); 80 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 81 82 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) { 83 int i; 84 const uint64_t *wp = zn->zn_key_norm; 85 86 ASSERT(zn->zn_key_intlen == 8); 87 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) { 88 int j; 89 uint64_t word = *wp; 90 91 for (j = 0; j < zn->zn_key_intlen; j++) { 92 h = (h >> 8) ^ 93 zfs_crc64_table[(h ^ word) & 0xFF]; 94 word >>= NBBY; 95 } 96 } 97 } else { 98 int i, len; 99 const uint8_t *cp = zn->zn_key_norm; 100 101 /* 102 * We previously stored the terminating null on 103 * disk, but didn't hash it, so we need to 104 * continue to not hash it. (The 105 * zn_key_*_numints includes the terminating 106 * null for non-binary keys.) 107 */ 108 len = zn->zn_key_norm_numints - 1; 109 110 ASSERT(zn->zn_key_intlen == 1); 111 for (i = 0; i < len; cp++, i++) { 112 h = (h >> 8) ^ 113 zfs_crc64_table[(h ^ *cp) & 0xFF]; 114 } 115 } 116 } 117 /* 118 * Don't use all 64 bits, since we need some in the cookie for 119 * the collision differentiator. We MUST use the high bits, 120 * since those are the ones that we first pay attention to when 121 * chosing the bucket. 122 */ 123 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1); 124 125 return (h); 126} 127 128static int 129zap_normalize(zap_t *zap, const char *name, char *namenorm) 130{ 131 size_t inlen, outlen; 132 int err; 133 134 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY)); 135 136 inlen = strlen(name) + 1; 137 outlen = ZAP_MAXNAMELEN; 138 139 err = 0; 140 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen, 141 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL | 142 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err); 143 144 return (err); 145} 146 147boolean_t 148zap_match(zap_name_t *zn, const char *matchname) 149{ 150 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY)); 151 152 if (zn->zn_matchtype == MT_FIRST) { 153 char norm[ZAP_MAXNAMELEN]; 154 155 if (zap_normalize(zn->zn_zap, matchname, norm) != 0) 156 return (B_FALSE); 157 158 return (strcmp(zn->zn_key_norm, norm) == 0); 159 } else { 160 /* MT_BEST or MT_EXACT */ 161 return (strcmp(zn->zn_key_orig, matchname) == 0); 162 } 163} 164 165void 166zap_name_free(zap_name_t *zn) 167{ 168 kmem_free(zn, sizeof (zap_name_t)); 169} 170 171zap_name_t * 172zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt) 173{ 174 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP); 175 176 zn->zn_zap = zap; 177 zn->zn_key_intlen = sizeof (*key); 178 zn->zn_key_orig = key; 179 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1; 180 zn->zn_matchtype = mt; 181 if (zap->zap_normflags) { 182 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) { 183 zap_name_free(zn); 184 return (NULL); 185 } 186 zn->zn_key_norm = zn->zn_normbuf; 187 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1; 188 } else { 189 if (mt != MT_EXACT) { 190 zap_name_free(zn); 191 return (NULL); 192 } 193 zn->zn_key_norm = zn->zn_key_orig; 194 zn->zn_key_norm_numints = zn->zn_key_orig_numints; 195 } 196 197 zn->zn_hash = zap_hash(zn); 198 return (zn); 199} 200 201zap_name_t * 202zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints) 203{ 204 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP); 205 206 ASSERT(zap->zap_normflags == 0); 207 zn->zn_zap = zap; 208 zn->zn_key_intlen = sizeof (*key); 209 zn->zn_key_orig = zn->zn_key_norm = key; 210 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints; 211 zn->zn_matchtype = MT_EXACT; 212 213 zn->zn_hash = zap_hash(zn); 214 return (zn); 215} 216 217static void 218mzap_byteswap(mzap_phys_t *buf, size_t size) 219{ 220 int i, max; 221 buf->mz_block_type = BSWAP_64(buf->mz_block_type); 222 buf->mz_salt = BSWAP_64(buf->mz_salt); 223 buf->mz_normflags = BSWAP_64(buf->mz_normflags); 224 max = (size / MZAP_ENT_LEN) - 1; 225 for (i = 0; i < max; i++) { 226 buf->mz_chunk[i].mze_value = 227 BSWAP_64(buf->mz_chunk[i].mze_value); 228 buf->mz_chunk[i].mze_cd = 229 BSWAP_32(buf->mz_chunk[i].mze_cd); 230 } 231} 232 233void 234zap_byteswap(void *buf, size_t size) 235{ 236 uint64_t block_type; 237 238 block_type = *(uint64_t *)buf; 239 240 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) { 241 /* ASSERT(magic == ZAP_LEAF_MAGIC); */ 242 mzap_byteswap(buf, size); 243 } else { 244 fzap_byteswap(buf, size); 245 } 246} 247 248static int 249mze_compare(const void *arg1, const void *arg2) 250{ 251 const mzap_ent_t *mze1 = arg1; 252 const mzap_ent_t *mze2 = arg2; 253 254 if (mze1->mze_hash > mze2->mze_hash) 255 return (+1); 256 if (mze1->mze_hash < mze2->mze_hash) 257 return (-1); 258 if (mze1->mze_cd > mze2->mze_cd) 259 return (+1); 260 if (mze1->mze_cd < mze2->mze_cd) 261 return (-1); 262 return (0); 263} 264 265static int 266mze_insert(zap_t *zap, int chunkid, uint64_t hash) 267{ 268 mzap_ent_t *mze; 269 avl_index_t idx; 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 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) { 280 kmem_free(mze, sizeof (mzap_ent_t)); 281 return (EEXIST); 282 } 283 avl_insert(&zap->zap_m.zap_avl, mze, idx); 284 return (0); 285} 286 287static mzap_ent_t * 288mze_find(zap_name_t *zn) 289{ 290 mzap_ent_t mze_tofind; 291 mzap_ent_t *mze; 292 avl_index_t idx; 293 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl; 294 295 ASSERT(zn->zn_zap->zap_ismicro); 296 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock)); 297 298 mze_tofind.mze_hash = zn->zn_hash; 299 mze_tofind.mze_cd = 0; 300 301again: 302 mze = avl_find(avl, &mze_tofind, &idx); 303 if (mze == NULL) 304 mze = avl_nearest(avl, idx, AVL_AFTER); 305 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) { 306 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd); 307 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name)) 308 return (mze); 309 } 310 if (zn->zn_matchtype == MT_BEST) { 311 zn->zn_matchtype = MT_FIRST; 312 goto again; 313 } 314 return (NULL); 315} 316 317static uint32_t 318mze_find_unused_cd(zap_t *zap, uint64_t hash) 319{ 320 mzap_ent_t mze_tofind; 321 mzap_ent_t *mze; 322 avl_index_t idx; 323 avl_tree_t *avl = &zap->zap_m.zap_avl; 324 uint32_t cd; 325 326 ASSERT(zap->zap_ismicro); 327 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 328 329 mze_tofind.mze_hash = hash; 330 mze_tofind.mze_cd = 0; 331 332 cd = 0; 333 for (mze = avl_find(avl, &mze_tofind, &idx); 334 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) { 335 if (mze->mze_cd != cd) 336 break; 337 cd++; 338 } 339 340 return (cd); 341} 342 343static void 344mze_remove(zap_t *zap, mzap_ent_t *mze) 345{ 346 ASSERT(zap->zap_ismicro); 347 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 348 349 avl_remove(&zap->zap_m.zap_avl, mze); 350 kmem_free(mze, sizeof (mzap_ent_t)); 351} 352 353static void 354mze_destroy(zap_t *zap) 355{ 356 mzap_ent_t *mze; 357 void *avlcookie = NULL; 358 359 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie)) 360 kmem_free(mze, sizeof (mzap_ent_t)); 361 avl_destroy(&zap->zap_m.zap_avl); 362} 363 364static zap_t * 365mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db) 366{ 367 zap_t *winner; 368 zap_t *zap; 369 int i; 370 371 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t)); 372 373 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP); 374 rw_init(&zap->zap_rwlock, 0, 0, 0); 375 rw_enter(&zap->zap_rwlock, RW_WRITER); 376 zap->zap_objset = os; 377 zap->zap_object = obj; 378 zap->zap_dbuf = db; 379 380 if (*(uint64_t *)db->db_data != ZBT_MICRO) { 381 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0); 382 zap->zap_f.zap_block_shift = highbit(db->db_size) - 1; 383 } else { 384 zap->zap_ismicro = TRUE; 385 } 386 387 /* 388 * Make sure that zap_ismicro is set before we let others see 389 * it, because zap_lockdir() checks zap_ismicro without the lock 390 * held. 391 */ 392 winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict); 393 394 if (winner != NULL) { 395 rw_exit(&zap->zap_rwlock); 396 rw_destroy(&zap->zap_rwlock); 397 if (!zap->zap_ismicro) 398 mutex_destroy(&zap->zap_f.zap_num_entries_mtx); 399 kmem_free(zap, sizeof (zap_t)); 400 return (winner); 401 } 402 403 if (zap->zap_ismicro) { 404 zap->zap_salt = zap->zap_m.zap_phys->mz_salt; 405 zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags; 406 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1; 407 avl_create(&zap->zap_m.zap_avl, mze_compare, 408 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node)); 409 410 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) { 411 mzap_ent_phys_t *mze = 412 &zap->zap_m.zap_phys->mz_chunk[i]; 413 if (mze->mze_name[0]) { 414 zap_name_t *zn; 415 416 zn = zap_name_alloc(zap, mze->mze_name, 417 MT_EXACT); 418 if (mze_insert(zap, i, zn->zn_hash) == 0) 419 zap->zap_m.zap_num_entries++; 420 else { 421 printf("ZFS WARNING: Duplicated ZAP " 422 "entry detected (%s).\n", 423 mze->mze_name); 424 } 425 zap_name_free(zn); 426 } 427 } 428 } else { 429 zap->zap_salt = zap->zap_f.zap_phys->zap_salt; 430 zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags; 431 432 ASSERT3U(sizeof (struct zap_leaf_header), ==, 433 2*ZAP_LEAF_CHUNKSIZE); 434 435 /* 436 * The embedded pointer table should not overlap the 437 * other members. 438 */ 439 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >, 440 &zap->zap_f.zap_phys->zap_salt); 441 442 /* 443 * The embedded pointer table should end at the end of 444 * the block 445 */ 446 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap, 447 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) - 448 (uintptr_t)zap->zap_f.zap_phys, ==, 449 zap->zap_dbuf->db_size); 450 } 451 rw_exit(&zap->zap_rwlock); 452 return (zap); 453} 454 455int 456zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx, 457 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp) 458{ 459 zap_t *zap; 460 dmu_buf_t *db; 461 krw_t lt; 462 int err; 463 464 *zapp = NULL; 465 466 err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH); 467 if (err) 468 return (err); 469 470#ifdef ZFS_DEBUG 471 { 472 dmu_object_info_t doi; 473 dmu_object_info_from_db(db, &doi); 474 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap); 475 } 476#endif 477 478 zap = dmu_buf_get_user(db); 479 if (zap == NULL) 480 zap = mzap_open(os, obj, db); 481 482 /* 483 * We're checking zap_ismicro without the lock held, in order to 484 * tell what type of lock we want. Once we have some sort of 485 * lock, see if it really is the right type. In practice this 486 * can only be different if it was upgraded from micro to fat, 487 * and micro wanted WRITER but fat only needs READER. 488 */ 489 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti; 490 rw_enter(&zap->zap_rwlock, lt); 491 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) { 492 /* it was upgraded, now we only need reader */ 493 ASSERT(lt == RW_WRITER); 494 ASSERT(RW_READER == 495 (!zap->zap_ismicro && fatreader) ? RW_READER : lti); 496 rw_downgrade(&zap->zap_rwlock); 497 lt = RW_READER; 498 } 499 500 zap->zap_objset = os; 501 502 if (lt == RW_WRITER) 503 dmu_buf_will_dirty(db, tx); 504 505 ASSERT3P(zap->zap_dbuf, ==, db); 506 507 ASSERT(!zap->zap_ismicro || 508 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks); 509 if (zap->zap_ismicro && tx && adding && 510 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) { 511 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE; 512 if (newsz > MZAP_MAX_BLKSZ) { 513 dprintf("upgrading obj %llu: num_entries=%u\n", 514 obj, zap->zap_m.zap_num_entries); 515 *zapp = zap; 516 return (mzap_upgrade(zapp, tx, 0)); 517 } 518 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx); 519 ASSERT3U(err, ==, 0); 520 zap->zap_m.zap_num_chunks = 521 db->db_size / MZAP_ENT_LEN - 1; 522 } 523 524 *zapp = zap; 525 return (0); 526} 527 528void 529zap_unlockdir(zap_t *zap) 530{ 531 rw_exit(&zap->zap_rwlock); 532 dmu_buf_rele(zap->zap_dbuf, NULL); 533} 534 535static int 536mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags) 537{ 538 mzap_phys_t *mzp; 539 int i, sz, nchunks; 540 int err = 0; 541 zap_t *zap = *zapp; 542 543 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 544 545 sz = zap->zap_dbuf->db_size; 546 mzp = kmem_alloc(sz, KM_SLEEP); 547 bcopy(zap->zap_dbuf->db_data, mzp, sz); 548 nchunks = zap->zap_m.zap_num_chunks; 549 550 if (!flags) { 551 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object, 552 1ULL << fzap_default_block_shift, 0, tx); 553 if (err) { 554 kmem_free(mzp, sz); 555 return (err); 556 } 557 } 558 559 dprintf("upgrading obj=%llu with %u chunks\n", 560 zap->zap_object, nchunks); 561 /* XXX destroy the avl later, so we can use the stored hash value */ 562 mze_destroy(zap); 563 564 fzap_upgrade(zap, tx, flags); 565 566 for (i = 0; i < nchunks; i++) { 567 mzap_ent_phys_t *mze = &mzp->mz_chunk[i]; 568 zap_name_t *zn; 569 if (mze->mze_name[0] == 0) 570 continue; 571 dprintf("adding %s=%llu\n", 572 mze->mze_name, mze->mze_value); 573 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT); 574 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx); 575 zap = zn->zn_zap; /* fzap_add_cd() may change zap */ 576 zap_name_free(zn); 577 if (err) 578 break; 579 } 580 kmem_free(mzp, sz); 581 *zapp = zap; 582 return (err); 583} 584 585static void 586mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags, 587 dmu_tx_t *tx) 588{ 589 dmu_buf_t *db; 590 mzap_phys_t *zp; 591 592 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH)); 593 594#ifdef ZFS_DEBUG 595 { 596 dmu_object_info_t doi; 597 dmu_object_info_from_db(db, &doi); 598 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap); 599 } 600#endif 601 602 dmu_buf_will_dirty(db, tx); 603 zp = db->db_data; 604 zp->mz_block_type = ZBT_MICRO; 605 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL; 606 zp->mz_normflags = normflags; 607 dmu_buf_rele(db, FTAG); 608 609 if (flags != 0) { 610 zap_t *zap; 611 /* Only fat zap supports flags; upgrade immediately. */ 612 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER, 613 B_FALSE, B_FALSE, &zap)); 614 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags)); 615 zap_unlockdir(zap); 616 } 617} 618 619int 620zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot, 621 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 622{ 623 return (zap_create_claim_norm(os, obj, 624 0, ot, bonustype, bonuslen, tx)); 625} 626 627int 628zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags, 629 dmu_object_type_t ot, 630 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 631{ 632 int err; 633 634 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx); 635 if (err != 0) 636 return (err); 637 mzap_create_impl(os, obj, normflags, 0, tx); 638 return (0); 639} 640 641uint64_t 642zap_create(objset_t *os, dmu_object_type_t ot, 643 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 644{ 645 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx)); 646} 647 648uint64_t 649zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot, 650 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 651{ 652 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx); 653 654 mzap_create_impl(os, obj, normflags, 0, tx); 655 return (obj); 656} 657 658uint64_t 659zap_create_flags(objset_t *os, int normflags, zap_flags_t flags, 660 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift, 661 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 662{ 663 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx); 664 665 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT && 666 leaf_blockshift <= SPA_MAXBLOCKSHIFT && 667 indirect_blockshift >= SPA_MINBLOCKSHIFT && 668 indirect_blockshift <= SPA_MAXBLOCKSHIFT); 669 670 VERIFY(dmu_object_set_blocksize(os, obj, 671 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0); 672 673 mzap_create_impl(os, obj, normflags, flags, tx); 674 return (obj); 675} 676 677int 678zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx) 679{ 680 /* 681 * dmu_object_free will free the object number and free the 682 * data. Freeing the data will cause our pageout function to be 683 * called, which will destroy our data (zap_leaf_t's and zap_t). 684 */ 685 686 return (dmu_object_free(os, zapobj, tx)); 687} 688 689_NOTE(ARGSUSED(0)) 690void 691zap_evict(dmu_buf_t *db, void *vzap) 692{ 693 zap_t *zap = vzap; 694 695 rw_destroy(&zap->zap_rwlock); 696 697 if (zap->zap_ismicro) 698 mze_destroy(zap); 699 else 700 mutex_destroy(&zap->zap_f.zap_num_entries_mtx); 701 702 kmem_free(zap, sizeof (zap_t)); 703} 704 705int 706zap_count(objset_t *os, uint64_t zapobj, uint64_t *count) 707{ 708 zap_t *zap; 709 int err; 710 711 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 712 if (err) 713 return (err); 714 if (!zap->zap_ismicro) { 715 err = fzap_count(zap, count); 716 } else { 717 *count = zap->zap_m.zap_num_entries; 718 } 719 zap_unlockdir(zap); 720 return (err); 721} 722 723/* 724 * zn may be NULL; if not specified, it will be computed if needed. 725 * See also the comment above zap_entry_normalization_conflict(). 726 */ 727static boolean_t 728mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze) 729{ 730 mzap_ent_t *other; 731 int direction = AVL_BEFORE; 732 boolean_t allocdzn = B_FALSE; 733 734 if (zap->zap_normflags == 0) 735 return (B_FALSE); 736 737again: 738 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction); 739 other && other->mze_hash == mze->mze_hash; 740 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) { 741 742 if (zn == NULL) { 743 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name, 744 MT_FIRST); 745 allocdzn = B_TRUE; 746 } 747 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) { 748 if (allocdzn) 749 zap_name_free(zn); 750 return (B_TRUE); 751 } 752 } 753 754 if (direction == AVL_BEFORE) { 755 direction = AVL_AFTER; 756 goto again; 757 } 758 759 if (allocdzn) 760 zap_name_free(zn); 761 return (B_FALSE); 762} 763 764/* 765 * Routines for manipulating attributes. 766 */ 767 768int 769zap_lookup(objset_t *os, uint64_t zapobj, const char *name, 770 uint64_t integer_size, uint64_t num_integers, void *buf) 771{ 772 return (zap_lookup_norm(os, zapobj, name, integer_size, 773 num_integers, buf, MT_EXACT, NULL, 0, NULL)); 774} 775 776int 777zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name, 778 uint64_t integer_size, uint64_t num_integers, void *buf, 779 matchtype_t mt, char *realname, int rn_len, 780 boolean_t *ncp) 781{ 782 zap_t *zap; 783 int err; 784 mzap_ent_t *mze; 785 zap_name_t *zn; 786 787 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 788 if (err) 789 return (err); 790 zn = zap_name_alloc(zap, name, mt); 791 if (zn == NULL) { 792 zap_unlockdir(zap); 793 return (ENOTSUP); 794 } 795 796 if (!zap->zap_ismicro) { 797 err = fzap_lookup(zn, integer_size, num_integers, buf, 798 realname, rn_len, ncp); 799 } else { 800 mze = mze_find(zn); 801 if (mze == NULL) { 802 err = ENOENT; 803 } else { 804 if (num_integers < 1) { 805 err = EOVERFLOW; 806 } else if (integer_size != 8) { 807 err = EINVAL; 808 } else { 809 *(uint64_t *)buf = 810 MZE_PHYS(zap, mze)->mze_value; 811 (void) strlcpy(realname, 812 MZE_PHYS(zap, mze)->mze_name, rn_len); 813 if (ncp) { 814 *ncp = mzap_normalization_conflict(zap, 815 zn, mze); 816 } 817 } 818 } 819 } 820 zap_name_free(zn); 821 zap_unlockdir(zap); 822 return (err); 823} 824 825int 826zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 827 int key_numints) 828{ 829 zap_t *zap; 830 int err; 831 zap_name_t *zn; 832 833 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 834 if (err) 835 return (err); 836 zn = zap_name_alloc_uint64(zap, key, key_numints); 837 if (zn == NULL) { 838 zap_unlockdir(zap); 839 return (ENOTSUP); 840 } 841 842 fzap_prefetch(zn); 843 zap_name_free(zn); 844 zap_unlockdir(zap); 845 return (err); 846} 847 848int 849zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 850 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf) 851{ 852 zap_t *zap; 853 int err; 854 zap_name_t *zn; 855 856 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 857 if (err) 858 return (err); 859 zn = zap_name_alloc_uint64(zap, key, key_numints); 860 if (zn == NULL) { 861 zap_unlockdir(zap); 862 return (ENOTSUP); 863 } 864 865 err = fzap_lookup(zn, integer_size, num_integers, buf, 866 NULL, 0, NULL); 867 zap_name_free(zn); 868 zap_unlockdir(zap); 869 return (err); 870} 871 872int 873zap_contains(objset_t *os, uint64_t zapobj, const char *name) 874{ 875 int err = (zap_lookup_norm(os, zapobj, name, 0, 876 0, NULL, MT_EXACT, NULL, 0, NULL)); 877 if (err == EOVERFLOW || err == EINVAL) 878 err = 0; /* found, but skipped reading the value */ 879 return (err); 880} 881 882int 883zap_length(objset_t *os, uint64_t zapobj, const char *name, 884 uint64_t *integer_size, uint64_t *num_integers) 885{ 886 zap_t *zap; 887 int err; 888 mzap_ent_t *mze; 889 zap_name_t *zn; 890 891 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 892 if (err) 893 return (err); 894 zn = zap_name_alloc(zap, name, MT_EXACT); 895 if (zn == NULL) { 896 zap_unlockdir(zap); 897 return (ENOTSUP); 898 } 899 if (!zap->zap_ismicro) { 900 err = fzap_length(zn, integer_size, num_integers); 901 } else { 902 mze = mze_find(zn); 903 if (mze == NULL) { 904 err = ENOENT; 905 } else { 906 if (integer_size) 907 *integer_size = 8; 908 if (num_integers) 909 *num_integers = 1; 910 } 911 } 912 zap_name_free(zn); 913 zap_unlockdir(zap); 914 return (err); 915} 916 917int 918zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 919 int key_numints, uint64_t *integer_size, uint64_t *num_integers) 920{ 921 zap_t *zap; 922 int err; 923 zap_name_t *zn; 924 925 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 926 if (err) 927 return (err); 928 zn = zap_name_alloc_uint64(zap, key, key_numints); 929 if (zn == NULL) { 930 zap_unlockdir(zap); 931 return (ENOTSUP); 932 } 933 err = fzap_length(zn, integer_size, num_integers); 934 zap_name_free(zn); 935 zap_unlockdir(zap); 936 return (err); 937} 938 939static void 940mzap_addent(zap_name_t *zn, uint64_t value) 941{ 942 int i; 943 zap_t *zap = zn->zn_zap; 944 int start = zap->zap_m.zap_alloc_next; 945 uint32_t cd; 946 947 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 948 949#ifdef ZFS_DEBUG 950 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) { 951 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i]; 952 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0); 953 } 954#endif 955 956 cd = mze_find_unused_cd(zap, zn->zn_hash); 957 /* given the limited size of the microzap, this can't happen */ 958 ASSERT(cd < zap_maxcd(zap)); 959 960again: 961 for (i = start; i < zap->zap_m.zap_num_chunks; i++) { 962 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i]; 963 if (mze->mze_name[0] == 0) { 964 mze->mze_value = value; 965 mze->mze_cd = cd; 966 (void) strcpy(mze->mze_name, zn->zn_key_orig); 967 zap->zap_m.zap_num_entries++; 968 zap->zap_m.zap_alloc_next = i+1; 969 if (zap->zap_m.zap_alloc_next == 970 zap->zap_m.zap_num_chunks) 971 zap->zap_m.zap_alloc_next = 0; 972 VERIFY(0 == mze_insert(zap, i, zn->zn_hash)); 973 return; 974 } 975 } 976 if (start != 0) { 977 start = 0; 978 goto again; 979 } 980 ASSERT(!"out of entries!"); 981} 982 983int 984zap_add(objset_t *os, uint64_t zapobj, const char *key, 985 int integer_size, uint64_t num_integers, 986 const void *val, dmu_tx_t *tx) 987{ 988 zap_t *zap; 989 int err; 990 mzap_ent_t *mze; 991 const uint64_t *intval = val; 992 zap_name_t *zn; 993 994 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap); 995 if (err) 996 return (err); 997 zn = zap_name_alloc(zap, key, MT_EXACT); 998 if (zn == NULL) { 999 zap_unlockdir(zap); 1000 return (ENOTSUP); 1001 } 1002 if (!zap->zap_ismicro) { 1003 err = fzap_add(zn, integer_size, num_integers, val, tx); 1004 zap = zn->zn_zap; /* fzap_add() may change zap */ 1005 } else if (integer_size != 8 || num_integers != 1 || 1006 strlen(key) >= MZAP_NAME_LEN) { 1007 err = mzap_upgrade(&zn->zn_zap, tx, 0); 1008 if (err == 0) 1009 err = fzap_add(zn, integer_size, num_integers, val, tx); 1010 zap = zn->zn_zap; /* fzap_add() may change zap */ 1011 } else { 1012 mze = mze_find(zn); 1013 if (mze != NULL) { 1014 err = EEXIST; 1015 } else { 1016 mzap_addent(zn, *intval); 1017 } 1018 } 1019 ASSERT(zap == zn->zn_zap); 1020 zap_name_free(zn); 1021 if (zap != NULL) /* may be NULL if fzap_add() failed */ 1022 zap_unlockdir(zap); 1023 return (err); 1024} 1025 1026int 1027zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1028 int key_numints, int integer_size, uint64_t num_integers, 1029 const void *val, dmu_tx_t *tx) 1030{ 1031 zap_t *zap; 1032 int err; 1033 zap_name_t *zn; 1034 1035 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap); 1036 if (err) 1037 return (err); 1038 zn = zap_name_alloc_uint64(zap, key, key_numints); 1039 if (zn == NULL) { 1040 zap_unlockdir(zap); 1041 return (ENOTSUP); 1042 } 1043 err = fzap_add(zn, integer_size, num_integers, val, tx); 1044 zap = zn->zn_zap; /* fzap_add() may change zap */ 1045 zap_name_free(zn); 1046 if (zap != NULL) /* may be NULL if fzap_add() failed */ 1047 zap_unlockdir(zap); 1048 return (err); 1049} 1050 1051int 1052zap_update(objset_t *os, uint64_t zapobj, const char *name, 1053 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx) 1054{ 1055 zap_t *zap; 1056 mzap_ent_t *mze; 1057 uint64_t oldval; 1058 const uint64_t *intval = val; 1059 zap_name_t *zn; 1060 int err; 1061 1062#ifdef ZFS_DEBUG 1063 /* 1064 * If there is an old value, it shouldn't change across the 1065 * lockdir (eg, due to bprewrite's xlation). 1066 */ 1067 if (integer_size == 8 && num_integers == 1) 1068 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval); 1069#endif 1070 1071 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap); 1072 if (err) 1073 return (err); 1074 zn = zap_name_alloc(zap, name, MT_EXACT); 1075 if (zn == NULL) { 1076 zap_unlockdir(zap); 1077 return (ENOTSUP); 1078 } 1079 if (!zap->zap_ismicro) { 1080 err = fzap_update(zn, integer_size, num_integers, val, tx); 1081 zap = zn->zn_zap; /* fzap_update() may change zap */ 1082 } else if (integer_size != 8 || num_integers != 1 || 1083 strlen(name) >= MZAP_NAME_LEN) { 1084 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n", 1085 zapobj, integer_size, num_integers, name); 1086 err = mzap_upgrade(&zn->zn_zap, tx, 0); 1087 if (err == 0) 1088 err = fzap_update(zn, integer_size, num_integers, 1089 val, tx); 1090 zap = zn->zn_zap; /* fzap_update() may change zap */ 1091 } else { 1092 mze = mze_find(zn); 1093 if (mze != NULL) { 1094 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval); 1095 MZE_PHYS(zap, mze)->mze_value = *intval; 1096 } else { 1097 mzap_addent(zn, *intval); 1098 } 1099 } 1100 ASSERT(zap == zn->zn_zap); 1101 zap_name_free(zn); 1102 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */ 1103 zap_unlockdir(zap); 1104 return (err); 1105} 1106 1107int 1108zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1109 int key_numints, 1110 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx) 1111{ 1112 zap_t *zap; 1113 zap_name_t *zn; 1114 int err; 1115 1116 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap); 1117 if (err) 1118 return (err); 1119 zn = zap_name_alloc_uint64(zap, key, key_numints); 1120 if (zn == NULL) { 1121 zap_unlockdir(zap); 1122 return (ENOTSUP); 1123 } 1124 err = fzap_update(zn, integer_size, num_integers, val, tx); 1125 zap = zn->zn_zap; /* fzap_update() may change zap */ 1126 zap_name_free(zn); 1127 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */ 1128 zap_unlockdir(zap); 1129 return (err); 1130} 1131 1132int 1133zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx) 1134{ 1135 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx)); 1136} 1137 1138int 1139zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name, 1140 matchtype_t mt, dmu_tx_t *tx) 1141{ 1142 zap_t *zap; 1143 int err; 1144 mzap_ent_t *mze; 1145 zap_name_t *zn; 1146 1147 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap); 1148 if (err) 1149 return (err); 1150 zn = zap_name_alloc(zap, name, mt); 1151 if (zn == NULL) { 1152 zap_unlockdir(zap); 1153 return (ENOTSUP); 1154 } 1155 if (!zap->zap_ismicro) { 1156 err = fzap_remove(zn, tx); 1157 } else { 1158 mze = mze_find(zn); 1159 if (mze == NULL) { 1160 err = ENOENT; 1161 } else { 1162 zap->zap_m.zap_num_entries--; 1163 bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid], 1164 sizeof (mzap_ent_phys_t)); 1165 mze_remove(zap, mze); 1166 } 1167 } 1168 zap_name_free(zn); 1169 zap_unlockdir(zap); 1170 return (err); 1171} 1172 1173int 1174zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1175 int key_numints, dmu_tx_t *tx) 1176{ 1177 zap_t *zap; 1178 int err; 1179 zap_name_t *zn; 1180 1181 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap); 1182 if (err) 1183 return (err); 1184 zn = zap_name_alloc_uint64(zap, key, key_numints); 1185 if (zn == NULL) { 1186 zap_unlockdir(zap); 1187 return (ENOTSUP); 1188 } 1189 err = fzap_remove(zn, tx); 1190 zap_name_free(zn); 1191 zap_unlockdir(zap); 1192 return (err); 1193} 1194 1195/* 1196 * Routines for iterating over the attributes. 1197 */ 1198 1199void 1200zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj, 1201 uint64_t serialized) 1202{ 1203 zc->zc_objset = os; 1204 zc->zc_zap = NULL; 1205 zc->zc_leaf = NULL; 1206 zc->zc_zapobj = zapobj; 1207 zc->zc_serialized = serialized; 1208 zc->zc_hash = 0; 1209 zc->zc_cd = 0; 1210} 1211 1212void 1213zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj) 1214{ 1215 zap_cursor_init_serialized(zc, os, zapobj, 0); 1216} 1217 1218void 1219zap_cursor_fini(zap_cursor_t *zc) 1220{ 1221 if (zc->zc_zap) { 1222 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 1223 zap_unlockdir(zc->zc_zap); 1224 zc->zc_zap = NULL; 1225 } 1226 if (zc->zc_leaf) { 1227 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1228 zap_put_leaf(zc->zc_leaf); 1229 zc->zc_leaf = NULL; 1230 } 1231 zc->zc_objset = NULL; 1232} 1233 1234uint64_t 1235zap_cursor_serialize(zap_cursor_t *zc) 1236{ 1237 if (zc->zc_hash == -1ULL) 1238 return (-1ULL); 1239 if (zc->zc_zap == NULL) 1240 return (zc->zc_serialized); 1241 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0); 1242 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap)); 1243 1244 /* 1245 * We want to keep the high 32 bits of the cursor zero if we can, so 1246 * that 32-bit programs can access this. So usually use a small 1247 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits 1248 * of the cursor. 1249 * 1250 * [ collision differentiator | zap_hashbits()-bit hash value ] 1251 */ 1252 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) | 1253 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap))); 1254} 1255 1256int 1257zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za) 1258{ 1259 int err; 1260 avl_index_t idx; 1261 mzap_ent_t mze_tofind; 1262 mzap_ent_t *mze; 1263 1264 if (zc->zc_hash == -1ULL) 1265 return (ENOENT); 1266 1267 if (zc->zc_zap == NULL) { 1268 int hb; 1269 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL, 1270 RW_READER, TRUE, FALSE, &zc->zc_zap); 1271 if (err) 1272 return (err); 1273 1274 /* 1275 * To support zap_cursor_init_serialized, advance, retrieve, 1276 * we must add to the existing zc_cd, which may already 1277 * be 1 due to the zap_cursor_advance. 1278 */ 1279 ASSERT(zc->zc_hash == 0); 1280 hb = zap_hashbits(zc->zc_zap); 1281 zc->zc_hash = zc->zc_serialized << (64 - hb); 1282 zc->zc_cd += zc->zc_serialized >> hb; 1283 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */ 1284 zc->zc_cd = 0; 1285 } else { 1286 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 1287 } 1288 if (!zc->zc_zap->zap_ismicro) { 1289 err = fzap_cursor_retrieve(zc->zc_zap, zc, za); 1290 } else { 1291 err = ENOENT; 1292 1293 mze_tofind.mze_hash = zc->zc_hash; 1294 mze_tofind.mze_cd = zc->zc_cd; 1295 1296 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx); 1297 if (mze == NULL) { 1298 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl, 1299 idx, AVL_AFTER); 1300 } 1301 if (mze) { 1302 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze); 1303 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd); 1304 za->za_normalization_conflict = 1305 mzap_normalization_conflict(zc->zc_zap, NULL, mze); 1306 za->za_integer_length = 8; 1307 za->za_num_integers = 1; 1308 za->za_first_integer = mzep->mze_value; 1309 (void) strcpy(za->za_name, mzep->mze_name); 1310 zc->zc_hash = mze->mze_hash; 1311 zc->zc_cd = mze->mze_cd; 1312 err = 0; 1313 } else { 1314 zc->zc_hash = -1ULL; 1315 } 1316 } 1317 rw_exit(&zc->zc_zap->zap_rwlock); 1318 return (err); 1319} 1320 1321void 1322zap_cursor_advance(zap_cursor_t *zc) 1323{ 1324 if (zc->zc_hash == -1ULL) 1325 return; 1326 zc->zc_cd++; 1327} 1328 1329int 1330zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt) 1331{ 1332 int err = 0; 1333 mzap_ent_t *mze; 1334 zap_name_t *zn; 1335 1336 if (zc->zc_zap == NULL) { 1337 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL, 1338 RW_READER, TRUE, FALSE, &zc->zc_zap); 1339 if (err) 1340 return (err); 1341 } else { 1342 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 1343 } 1344 1345 zn = zap_name_alloc(zc->zc_zap, name, mt); 1346 if (zn == NULL) { 1347 rw_exit(&zc->zc_zap->zap_rwlock); 1348 return (ENOTSUP); 1349 } 1350 1351 if (!zc->zc_zap->zap_ismicro) { 1352 err = fzap_cursor_move_to_key(zc, zn); 1353 } else { 1354 mze = mze_find(zn); 1355 if (mze == NULL) { 1356 err = ENOENT; 1357 goto out; 1358 } 1359 zc->zc_hash = mze->mze_hash; 1360 zc->zc_cd = mze->mze_cd; 1361 } 1362 1363out: 1364 zap_name_free(zn); 1365 rw_exit(&zc->zc_zap->zap_rwlock); 1366 return (err); 1367} 1368 1369int 1370zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs) 1371{ 1372 int err; 1373 zap_t *zap; 1374 1375 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 1376 if (err) 1377 return (err); 1378 1379 bzero(zs, sizeof (zap_stats_t)); 1380 1381 if (zap->zap_ismicro) { 1382 zs->zs_blocksize = zap->zap_dbuf->db_size; 1383 zs->zs_num_entries = zap->zap_m.zap_num_entries; 1384 zs->zs_num_blocks = 1; 1385 } else { 1386 fzap_get_stats(zap, zs); 1387 } 1388 zap_unlockdir(zap); 1389 return (0); 1390} 1391 1392int 1393zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add, 1394 uint64_t *towrite, uint64_t *tooverwrite) 1395{ 1396 zap_t *zap; 1397 int err = 0; 1398 1399 1400 /* 1401 * Since, we don't have a name, we cannot figure out which blocks will 1402 * be affected in this operation. So, account for the worst case : 1403 * - 3 blocks overwritten: target leaf, ptrtbl block, header block 1404 * - 4 new blocks written if adding: 1405 * - 2 blocks for possibly split leaves, 1406 * - 2 grown ptrtbl blocks 1407 * 1408 * This also accomodates the case where an add operation to a fairly 1409 * large microzap results in a promotion to fatzap. 1410 */ 1411 if (name == NULL) { 1412 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE; 1413 return (err); 1414 } 1415 1416 /* 1417 * We lock the zap with adding == FALSE. Because, if we pass 1418 * the actual value of add, it could trigger a mzap_upgrade(). 1419 * At present we are just evaluating the possibility of this operation 1420 * and hence we donot want to trigger an upgrade. 1421 */ 1422 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 1423 if (err) 1424 return (err); 1425 1426 if (!zap->zap_ismicro) { 1427 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT); 1428 if (zn) { 1429 err = fzap_count_write(zn, add, towrite, 1430 tooverwrite); 1431 zap_name_free(zn); 1432 } else { 1433 /* 1434 * We treat this case as similar to (name == NULL) 1435 */ 1436 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE; 1437 } 1438 } else { 1439 /* 1440 * We are here if (name != NULL) and this is a micro-zap. 1441 * We account for the header block depending on whether it 1442 * is freeable. 1443 * 1444 * Incase of an add-operation it is hard to find out 1445 * if this add will promote this microzap to fatzap. 1446 * Hence, we consider the worst case and account for the 1447 * blocks assuming this microzap would be promoted to a 1448 * fatzap. 1449 * 1450 * 1 block overwritten : header block 1451 * 4 new blocks written : 2 new split leaf, 2 grown 1452 * ptrtbl blocks 1453 */ 1454 if (dmu_buf_freeable(zap->zap_dbuf)) 1455 *tooverwrite += SPA_MAXBLOCKSIZE; 1456 else 1457 *towrite += SPA_MAXBLOCKSIZE; 1458 1459 if (add) { 1460 *towrite += 4 * SPA_MAXBLOCKSIZE; 1461 } 1462 } 1463 1464 zap_unlockdir(zap); 1465 return (err); 1466} 1467