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) 2011, 2016 by Delphix. All rights reserved. 24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 25 * Copyright (c) 2014 Integros [integros.com] 26 */ 27 28#include <sys/zio.h> 29#include <sys/spa.h> 30#include <sys/dmu.h> 31#include <sys/zfs_context.h> 32#include <sys/zap.h> 33#include <sys/refcount.h> 34#include <sys/zap_impl.h> 35#include <sys/zap_leaf.h> 36#include <sys/avl.h> 37#include <sys/arc.h> 38#include <sys/dmu_objset.h> 39 40#ifdef _KERNEL 41#include <sys/sunddi.h> 42#endif 43 44extern inline mzap_phys_t *zap_m_phys(zap_t *zap); 45 46static int mzap_upgrade(zap_t **zapp, 47 void *tag, dmu_tx_t *tx, zap_flags_t flags); 48 49uint64_t 50zap_getflags(zap_t *zap) 51{ 52 if (zap->zap_ismicro) 53 return (0); 54 return (zap_f_phys(zap)->zap_flags); 55} 56 57int 58zap_hashbits(zap_t *zap) 59{ 60 if (zap_getflags(zap) & ZAP_FLAG_HASH64) 61 return (48); 62 else 63 return (28); 64} 65 66uint32_t 67zap_maxcd(zap_t *zap) 68{ 69 if (zap_getflags(zap) & ZAP_FLAG_HASH64) 70 return ((1<<16)-1); 71 else 72 return (-1U); 73} 74 75static uint64_t 76zap_hash(zap_name_t *zn) 77{ 78 zap_t *zap = zn->zn_zap; 79 uint64_t h = 0; 80 81 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) { 82 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY); 83 h = *(uint64_t *)zn->zn_key_orig; 84 } else { 85 h = zap->zap_salt; 86 ASSERT(h != 0); 87 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 88 89 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) { 90 int i; 91 const uint64_t *wp = zn->zn_key_norm; 92 93 ASSERT(zn->zn_key_intlen == 8); 94 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) { 95 int j; 96 uint64_t word = *wp; 97 98 for (j = 0; j < zn->zn_key_intlen; j++) { 99 h = (h >> 8) ^ 100 zfs_crc64_table[(h ^ word) & 0xFF]; 101 word >>= NBBY; 102 } 103 } 104 } else { 105 int i, len; 106 const uint8_t *cp = zn->zn_key_norm; 107 108 /* 109 * We previously stored the terminating null on 110 * disk, but didn't hash it, so we need to 111 * continue to not hash it. (The 112 * zn_key_*_numints includes the terminating 113 * null for non-binary keys.) 114 */ 115 len = zn->zn_key_norm_numints - 1; 116 117 ASSERT(zn->zn_key_intlen == 1); 118 for (i = 0; i < len; cp++, i++) { 119 h = (h >> 8) ^ 120 zfs_crc64_table[(h ^ *cp) & 0xFF]; 121 } 122 } 123 } 124 /* 125 * Don't use all 64 bits, since we need some in the cookie for 126 * the collision differentiator. We MUST use the high bits, 127 * since those are the ones that we first pay attention to when 128 * chosing the bucket. 129 */ 130 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1); 131 132 return (h); 133} 134 135static int 136zap_normalize(zap_t *zap, const char *name, char *namenorm) 137{ 138 size_t inlen, outlen; 139 int err; 140 141 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY)); 142 143 inlen = strlen(name) + 1; 144 outlen = ZAP_MAXNAMELEN; 145 146 err = 0; 147 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen, 148 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL | 149 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err); 150 151 return (err); 152} 153 154boolean_t 155zap_match(zap_name_t *zn, const char *matchname) 156{ 157 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY)); 158 159 if (zn->zn_matchtype == MT_FIRST) { 160 char norm[ZAP_MAXNAMELEN]; 161 162 if (zap_normalize(zn->zn_zap, matchname, norm) != 0) 163 return (B_FALSE); 164 165 return (strcmp(zn->zn_key_norm, norm) == 0); 166 } else { 167 /* MT_BEST or MT_EXACT */ 168 return (strcmp(zn->zn_key_orig, matchname) == 0); 169 } 170} 171 172void 173zap_name_free(zap_name_t *zn) 174{ 175 kmem_free(zn, sizeof (zap_name_t)); 176} 177 178zap_name_t * 179zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt) 180{ 181 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP); 182 183 zn->zn_zap = zap; 184 zn->zn_key_intlen = sizeof (*key); 185 zn->zn_key_orig = key; 186 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1; 187 zn->zn_matchtype = mt; 188 if (zap->zap_normflags) { 189 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) { 190 zap_name_free(zn); 191 return (NULL); 192 } 193 zn->zn_key_norm = zn->zn_normbuf; 194 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1; 195 } else { 196 if (mt != MT_EXACT) { 197 zap_name_free(zn); 198 return (NULL); 199 } 200 zn->zn_key_norm = zn->zn_key_orig; 201 zn->zn_key_norm_numints = zn->zn_key_orig_numints; 202 } 203 204 zn->zn_hash = zap_hash(zn); 205 return (zn); 206} 207 208zap_name_t * 209zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints) 210{ 211 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP); 212 213 ASSERT(zap->zap_normflags == 0); 214 zn->zn_zap = zap; 215 zn->zn_key_intlen = sizeof (*key); 216 zn->zn_key_orig = zn->zn_key_norm = key; 217 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints; 218 zn->zn_matchtype = MT_EXACT; 219 220 zn->zn_hash = zap_hash(zn); 221 return (zn); 222} 223 224static void 225mzap_byteswap(mzap_phys_t *buf, size_t size) 226{ 227 int i, max; 228 buf->mz_block_type = BSWAP_64(buf->mz_block_type); 229 buf->mz_salt = BSWAP_64(buf->mz_salt); 230 buf->mz_normflags = BSWAP_64(buf->mz_normflags); 231 max = (size / MZAP_ENT_LEN) - 1; 232 for (i = 0; i < max; i++) { 233 buf->mz_chunk[i].mze_value = 234 BSWAP_64(buf->mz_chunk[i].mze_value); 235 buf->mz_chunk[i].mze_cd = 236 BSWAP_32(buf->mz_chunk[i].mze_cd); 237 } 238} 239 240void 241zap_byteswap(void *buf, size_t size) 242{ 243 uint64_t block_type; 244 245 block_type = *(uint64_t *)buf; 246 247 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) { 248 /* ASSERT(magic == ZAP_LEAF_MAGIC); */ 249 mzap_byteswap(buf, size); 250 } else { 251 fzap_byteswap(buf, size); 252 } 253} 254 255static int 256mze_compare(const void *arg1, const void *arg2) 257{ 258 const mzap_ent_t *mze1 = arg1; 259 const mzap_ent_t *mze2 = arg2; 260 261 if (mze1->mze_hash > mze2->mze_hash) 262 return (+1); 263 if (mze1->mze_hash < mze2->mze_hash) 264 return (-1); 265 if (mze1->mze_cd > mze2->mze_cd) 266 return (+1); 267 if (mze1->mze_cd < mze2->mze_cd) 268 return (-1); 269 return (0); 270} 271 272static int 273mze_insert(zap_t *zap, int chunkid, uint64_t hash) 274{ 275 mzap_ent_t *mze; 276 avl_index_t idx; 277 278 ASSERT(zap->zap_ismicro); 279 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 280 281 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP); 282 mze->mze_chunkid = chunkid; 283 mze->mze_hash = hash; 284 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd; 285 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0); 286 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) { 287 kmem_free(mze, sizeof (mzap_ent_t)); 288 return (EEXIST); 289 } 290 avl_insert(&zap->zap_m.zap_avl, mze, idx); 291 return (0); 292} 293 294static mzap_ent_t * 295mze_find(zap_name_t *zn) 296{ 297 mzap_ent_t mze_tofind; 298 mzap_ent_t *mze; 299 avl_index_t idx; 300 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl; 301 302 ASSERT(zn->zn_zap->zap_ismicro); 303 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock)); 304 305 mze_tofind.mze_hash = zn->zn_hash; 306 mze_tofind.mze_cd = 0; 307 308again: 309 mze = avl_find(avl, &mze_tofind, &idx); 310 if (mze == NULL) 311 mze = avl_nearest(avl, idx, AVL_AFTER); 312 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) { 313 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd); 314 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name)) 315 return (mze); 316 } 317 if (zn->zn_matchtype == MT_BEST) { 318 zn->zn_matchtype = MT_FIRST; 319 goto again; 320 } 321 return (NULL); 322} 323 324static uint32_t 325mze_find_unused_cd(zap_t *zap, uint64_t hash) 326{ 327 mzap_ent_t mze_tofind; 328 mzap_ent_t *mze; 329 avl_index_t idx; 330 avl_tree_t *avl = &zap->zap_m.zap_avl; 331 uint32_t cd; 332 333 ASSERT(zap->zap_ismicro); 334 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 335 336 mze_tofind.mze_hash = hash; 337 mze_tofind.mze_cd = 0; 338 339 cd = 0; 340 for (mze = avl_find(avl, &mze_tofind, &idx); 341 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) { 342 if (mze->mze_cd != cd) 343 break; 344 cd++; 345 } 346 347 return (cd); 348} 349 350static void 351mze_remove(zap_t *zap, mzap_ent_t *mze) 352{ 353 ASSERT(zap->zap_ismicro); 354 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 355 356 avl_remove(&zap->zap_m.zap_avl, mze); 357 kmem_free(mze, sizeof (mzap_ent_t)); 358} 359 360static void 361mze_destroy(zap_t *zap) 362{ 363 mzap_ent_t *mze; 364 void *avlcookie = NULL; 365 366 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie)) 367 kmem_free(mze, sizeof (mzap_ent_t)); 368 avl_destroy(&zap->zap_m.zap_avl); 369} 370 371static zap_t * 372mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db) 373{ 374 zap_t *winner; 375 zap_t *zap; 376 int i; 377 378 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t)); 379 380 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP); 381 rw_init(&zap->zap_rwlock, 0, 0, 0); 382 rw_enter(&zap->zap_rwlock, RW_WRITER); 383 zap->zap_objset = os; 384 zap->zap_object = obj; 385 zap->zap_dbuf = db; 386 387 if (*(uint64_t *)db->db_data != ZBT_MICRO) { 388 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0); 389 zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1; 390 } else { 391 zap->zap_ismicro = TRUE; 392 } 393 394 /* 395 * Make sure that zap_ismicro is set before we let others see 396 * it, because zap_lockdir() checks zap_ismicro without the lock 397 * held. 398 */ 399 dmu_buf_init_user(&zap->zap_dbu, zap_evict, &zap->zap_dbuf); 400 winner = dmu_buf_set_user(db, &zap->zap_dbu); 401 402 if (winner != NULL) { 403 rw_exit(&zap->zap_rwlock); 404 rw_destroy(&zap->zap_rwlock); 405 if (!zap->zap_ismicro) 406 mutex_destroy(&zap->zap_f.zap_num_entries_mtx); 407 kmem_free(zap, sizeof (zap_t)); 408 return (winner); 409 } 410 411 if (zap->zap_ismicro) { 412 zap->zap_salt = zap_m_phys(zap)->mz_salt; 413 zap->zap_normflags = zap_m_phys(zap)->mz_normflags; 414 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1; 415 avl_create(&zap->zap_m.zap_avl, mze_compare, 416 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node)); 417 418 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) { 419 mzap_ent_phys_t *mze = 420 &zap_m_phys(zap)->mz_chunk[i]; 421 if (mze->mze_name[0]) { 422 zap_name_t *zn; 423 424 zn = zap_name_alloc(zap, mze->mze_name, 425 MT_EXACT); 426 if (mze_insert(zap, i, zn->zn_hash) == 0) 427 zap->zap_m.zap_num_entries++; 428 else { 429 printf("ZFS WARNING: Duplicated ZAP " 430 "entry detected (%s).\n", 431 mze->mze_name); 432 } 433 zap_name_free(zn); 434 } 435 } 436 } else { 437 zap->zap_salt = zap_f_phys(zap)->zap_salt; 438 zap->zap_normflags = zap_f_phys(zap)->zap_normflags; 439 440 ASSERT3U(sizeof (struct zap_leaf_header), ==, 441 2*ZAP_LEAF_CHUNKSIZE); 442 443 /* 444 * The embedded pointer table should not overlap the 445 * other members. 446 */ 447 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >, 448 &zap_f_phys(zap)->zap_salt); 449 450 /* 451 * The embedded pointer table should end at the end of 452 * the block 453 */ 454 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap, 455 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) - 456 (uintptr_t)zap_f_phys(zap), ==, 457 zap->zap_dbuf->db_size); 458 } 459 rw_exit(&zap->zap_rwlock); 460 return (zap); 461} 462 463static int 464zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx, 465 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp) 466{ 467 zap_t *zap; 468 krw_t lt; 469 470 ASSERT0(db->db_offset); 471 objset_t *os = dmu_buf_get_objset(db); 472 uint64_t obj = db->db_object; 473 474 *zapp = NULL; 475 476#ifdef ZFS_DEBUG 477 { 478 dmu_object_info_t doi; 479 dmu_object_info_from_db(db, &doi); 480 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP); 481 } 482#endif 483 484 zap = dmu_buf_get_user(db); 485 if (zap == NULL) 486 zap = mzap_open(os, obj, db); 487 488 /* 489 * We're checking zap_ismicro without the lock held, in order to 490 * tell what type of lock we want. Once we have some sort of 491 * lock, see if it really is the right type. In practice this 492 * can only be different if it was upgraded from micro to fat, 493 * and micro wanted WRITER but fat only needs READER. 494 */ 495 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti; 496 rw_enter(&zap->zap_rwlock, lt); 497 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) { 498 /* it was upgraded, now we only need reader */ 499 ASSERT(lt == RW_WRITER); 500 ASSERT(RW_READER == 501 (!zap->zap_ismicro && fatreader) ? RW_READER : lti); 502 rw_downgrade(&zap->zap_rwlock); 503 lt = RW_READER; 504 } 505 506 zap->zap_objset = os; 507 508 if (lt == RW_WRITER) 509 dmu_buf_will_dirty(db, tx); 510 511 ASSERT3P(zap->zap_dbuf, ==, db); 512 513 ASSERT(!zap->zap_ismicro || 514 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks); 515 if (zap->zap_ismicro && tx && adding && 516 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) { 517 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE; 518 if (newsz > MZAP_MAX_BLKSZ) { 519 dprintf("upgrading obj %llu: num_entries=%u\n", 520 obj, zap->zap_m.zap_num_entries); 521 *zapp = zap; 522 int err = mzap_upgrade(zapp, tag, tx, 0); 523 if (err != 0) 524 rw_exit(&zap->zap_rwlock); 525 return (err); 526 } 527 VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx)); 528 zap->zap_m.zap_num_chunks = 529 db->db_size / MZAP_ENT_LEN - 1; 530 } 531 532 *zapp = zap; 533 return (0); 534} 535 536static int 537zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx, 538 krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp) 539{ 540 dmu_buf_t *db; 541 int err; 542 543 err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH); 544 if (err != 0) { 545 return (err); 546 } 547 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp); 548 if (err != 0) { 549 dmu_buf_rele(db, tag); 550 } 551 return (err); 552} 553 554int 555zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx, 556 krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp) 557{ 558 dmu_buf_t *db; 559 int err; 560 561 err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH); 562 if (err != 0) 563 return (err); 564 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp); 565 if (err != 0) 566 dmu_buf_rele(db, tag); 567 return (err); 568} 569 570void 571zap_unlockdir(zap_t *zap, void *tag) 572{ 573 rw_exit(&zap->zap_rwlock); 574 dmu_buf_rele(zap->zap_dbuf, tag); 575} 576 577static int 578mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags) 579{ 580 mzap_phys_t *mzp; 581 int i, sz, nchunks; 582 int err = 0; 583 zap_t *zap = *zapp; 584 585 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 586 587 sz = zap->zap_dbuf->db_size; 588 mzp = zio_buf_alloc(sz); 589 bcopy(zap->zap_dbuf->db_data, mzp, sz); 590 nchunks = zap->zap_m.zap_num_chunks; 591 592 if (!flags) { 593 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object, 594 1ULL << fzap_default_block_shift, 0, tx); 595 if (err) { 596 zio_buf_free(mzp, sz); 597 return (err); 598 } 599 } 600 601 dprintf("upgrading obj=%llu with %u chunks\n", 602 zap->zap_object, nchunks); 603 /* XXX destroy the avl later, so we can use the stored hash value */ 604 mze_destroy(zap); 605 606 fzap_upgrade(zap, tx, flags); 607 608 for (i = 0; i < nchunks; i++) { 609 mzap_ent_phys_t *mze = &mzp->mz_chunk[i]; 610 zap_name_t *zn; 611 if (mze->mze_name[0] == 0) 612 continue; 613 dprintf("adding %s=%llu\n", 614 mze->mze_name, mze->mze_value); 615 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT); 616 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, 617 tag, tx); 618 zap = zn->zn_zap; /* fzap_add_cd() may change zap */ 619 zap_name_free(zn); 620 if (err) 621 break; 622 } 623 zio_buf_free(mzp, sz); 624 *zapp = zap; 625 return (err); 626} 627 628void 629mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags, 630 dmu_tx_t *tx) 631{ 632 dmu_buf_t *db; 633 mzap_phys_t *zp; 634 635 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH)); 636 637#ifdef ZFS_DEBUG 638 { 639 dmu_object_info_t doi; 640 dmu_object_info_from_db(db, &doi); 641 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP); 642 } 643#endif 644 645 dmu_buf_will_dirty(db, tx); 646 zp = db->db_data; 647 zp->mz_block_type = ZBT_MICRO; 648 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL; 649 zp->mz_normflags = normflags; 650 dmu_buf_rele(db, FTAG); 651 652 if (flags != 0) { 653 zap_t *zap; 654 /* Only fat zap supports flags; upgrade immediately. */ 655 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER, 656 B_FALSE, B_FALSE, FTAG, &zap)); 657 VERIFY3U(0, ==, mzap_upgrade(&zap, FTAG, tx, flags)); 658 zap_unlockdir(zap, FTAG); 659 } 660} 661 662int 663zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot, 664 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 665{ 666 return (zap_create_claim_norm(os, obj, 667 0, ot, bonustype, bonuslen, tx)); 668} 669 670int 671zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags, 672 dmu_object_type_t ot, 673 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 674{ 675 int err; 676 677 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx); 678 if (err != 0) 679 return (err); 680 mzap_create_impl(os, obj, normflags, 0, tx); 681 return (0); 682} 683 684uint64_t 685zap_create(objset_t *os, dmu_object_type_t ot, 686 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 687{ 688 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx)); 689} 690 691uint64_t 692zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot, 693 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 694{ 695 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx); 696 697 mzap_create_impl(os, obj, normflags, 0, tx); 698 return (obj); 699} 700 701uint64_t 702zap_create_flags(objset_t *os, int normflags, zap_flags_t flags, 703 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift, 704 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 705{ 706 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx); 707 708 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT && 709 leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT && 710 indirect_blockshift >= SPA_MINBLOCKSHIFT && 711 indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT); 712 713 VERIFY(dmu_object_set_blocksize(os, obj, 714 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0); 715 716 mzap_create_impl(os, obj, normflags, flags, tx); 717 return (obj); 718} 719 720int 721zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx) 722{ 723 /* 724 * dmu_object_free will free the object number and free the 725 * data. Freeing the data will cause our pageout function to be 726 * called, which will destroy our data (zap_leaf_t's and zap_t). 727 */ 728 729 return (dmu_object_free(os, zapobj, tx)); 730} 731 732void 733zap_evict(void *dbu) 734{ 735 zap_t *zap = dbu; 736 737 rw_destroy(&zap->zap_rwlock); 738 739 if (zap->zap_ismicro) 740 mze_destroy(zap); 741 else 742 mutex_destroy(&zap->zap_f.zap_num_entries_mtx); 743 744 kmem_free(zap, sizeof (zap_t)); 745} 746 747int 748zap_count(objset_t *os, uint64_t zapobj, uint64_t *count) 749{ 750 zap_t *zap; 751 int err; 752 753 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 754 if (err) 755 return (err); 756 if (!zap->zap_ismicro) { 757 err = fzap_count(zap, count); 758 } else { 759 *count = zap->zap_m.zap_num_entries; 760 } 761 zap_unlockdir(zap, FTAG); 762 return (err); 763} 764 765/* 766 * zn may be NULL; if not specified, it will be computed if needed. 767 * See also the comment above zap_entry_normalization_conflict(). 768 */ 769static boolean_t 770mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze) 771{ 772 mzap_ent_t *other; 773 int direction = AVL_BEFORE; 774 boolean_t allocdzn = B_FALSE; 775 776 if (zap->zap_normflags == 0) 777 return (B_FALSE); 778 779again: 780 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction); 781 other && other->mze_hash == mze->mze_hash; 782 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) { 783 784 if (zn == NULL) { 785 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name, 786 MT_FIRST); 787 allocdzn = B_TRUE; 788 } 789 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) { 790 if (allocdzn) 791 zap_name_free(zn); 792 return (B_TRUE); 793 } 794 } 795 796 if (direction == AVL_BEFORE) { 797 direction = AVL_AFTER; 798 goto again; 799 } 800 801 if (allocdzn) 802 zap_name_free(zn); 803 return (B_FALSE); 804} 805 806/* 807 * Routines for manipulating attributes. 808 */ 809 810int 811zap_lookup(objset_t *os, uint64_t zapobj, const char *name, 812 uint64_t integer_size, uint64_t num_integers, void *buf) 813{ 814 return (zap_lookup_norm(os, zapobj, name, integer_size, 815 num_integers, buf, MT_EXACT, NULL, 0, NULL)); 816} 817 818static int 819zap_lookup_impl(zap_t *zap, const char *name, 820 uint64_t integer_size, uint64_t num_integers, void *buf, 821 matchtype_t mt, char *realname, int rn_len, 822 boolean_t *ncp) 823{ 824 int err = 0; 825 mzap_ent_t *mze; 826 zap_name_t *zn; 827 828 zn = zap_name_alloc(zap, name, mt); 829 if (zn == NULL) 830 return (SET_ERROR(ENOTSUP)); 831 832 if (!zap->zap_ismicro) { 833 err = fzap_lookup(zn, integer_size, num_integers, buf, 834 realname, rn_len, ncp); 835 } else { 836 mze = mze_find(zn); 837 if (mze == NULL) { 838 err = SET_ERROR(ENOENT); 839 } else { 840 if (num_integers < 1) { 841 err = SET_ERROR(EOVERFLOW); 842 } else if (integer_size != 8) { 843 err = SET_ERROR(EINVAL); 844 } else { 845 *(uint64_t *)buf = 846 MZE_PHYS(zap, mze)->mze_value; 847 (void) strlcpy(realname, 848 MZE_PHYS(zap, mze)->mze_name, rn_len); 849 if (ncp) { 850 *ncp = mzap_normalization_conflict(zap, 851 zn, mze); 852 } 853 } 854 } 855 } 856 zap_name_free(zn); 857 return (err); 858} 859 860int 861zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name, 862 uint64_t integer_size, uint64_t num_integers, void *buf, 863 matchtype_t mt, char *realname, int rn_len, 864 boolean_t *ncp) 865{ 866 zap_t *zap; 867 int err; 868 869 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 870 if (err != 0) 871 return (err); 872 err = zap_lookup_impl(zap, name, integer_size, 873 num_integers, buf, mt, realname, rn_len, ncp); 874 zap_unlockdir(zap, FTAG); 875 return (err); 876} 877 878int 879zap_lookup_by_dnode(dnode_t *dn, const char *name, 880 uint64_t integer_size, uint64_t num_integers, void *buf) 881{ 882 return (zap_lookup_norm_by_dnode(dn, name, integer_size, 883 num_integers, buf, MT_EXACT, NULL, 0, NULL)); 884} 885 886int 887zap_lookup_norm_by_dnode(dnode_t *dn, const char *name, 888 uint64_t integer_size, uint64_t num_integers, void *buf, 889 matchtype_t mt, char *realname, int rn_len, 890 boolean_t *ncp) 891{ 892 zap_t *zap; 893 int err; 894 895 err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE, 896 FTAG, &zap); 897 if (err != 0) 898 return (err); 899 err = zap_lookup_impl(zap, name, integer_size, 900 num_integers, buf, mt, realname, rn_len, ncp); 901 zap_unlockdir(zap, FTAG); 902 return (err); 903} 904 905int 906zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 907 int key_numints) 908{ 909 zap_t *zap; 910 int err; 911 zap_name_t *zn; 912 913 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 914 if (err) 915 return (err); 916 zn = zap_name_alloc_uint64(zap, key, key_numints); 917 if (zn == NULL) { 918 zap_unlockdir(zap, FTAG); 919 return (SET_ERROR(ENOTSUP)); 920 } 921 922 fzap_prefetch(zn); 923 zap_name_free(zn); 924 zap_unlockdir(zap, FTAG); 925 return (err); 926} 927 928int 929zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 930 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf) 931{ 932 zap_t *zap; 933 int err; 934 zap_name_t *zn; 935 936 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 937 if (err) 938 return (err); 939 zn = zap_name_alloc_uint64(zap, key, key_numints); 940 if (zn == NULL) { 941 zap_unlockdir(zap, FTAG); 942 return (SET_ERROR(ENOTSUP)); 943 } 944 945 err = fzap_lookup(zn, integer_size, num_integers, buf, 946 NULL, 0, NULL); 947 zap_name_free(zn); 948 zap_unlockdir(zap, FTAG); 949 return (err); 950} 951 952int 953zap_contains(objset_t *os, uint64_t zapobj, const char *name) 954{ 955 int err = zap_lookup_norm(os, zapobj, name, 0, 956 0, NULL, MT_EXACT, NULL, 0, NULL); 957 if (err == EOVERFLOW || err == EINVAL) 958 err = 0; /* found, but skipped reading the value */ 959 return (err); 960} 961 962int 963zap_length(objset_t *os, uint64_t zapobj, const char *name, 964 uint64_t *integer_size, uint64_t *num_integers) 965{ 966 zap_t *zap; 967 int err; 968 mzap_ent_t *mze; 969 zap_name_t *zn; 970 971 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 972 if (err) 973 return (err); 974 zn = zap_name_alloc(zap, name, MT_EXACT); 975 if (zn == NULL) { 976 zap_unlockdir(zap, FTAG); 977 return (SET_ERROR(ENOTSUP)); 978 } 979 if (!zap->zap_ismicro) { 980 err = fzap_length(zn, integer_size, num_integers); 981 } else { 982 mze = mze_find(zn); 983 if (mze == NULL) { 984 err = SET_ERROR(ENOENT); 985 } else { 986 if (integer_size) 987 *integer_size = 8; 988 if (num_integers) 989 *num_integers = 1; 990 } 991 } 992 zap_name_free(zn); 993 zap_unlockdir(zap, FTAG); 994 return (err); 995} 996 997int 998zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 999 int key_numints, uint64_t *integer_size, uint64_t *num_integers) 1000{ 1001 zap_t *zap; 1002 int err; 1003 zap_name_t *zn; 1004 1005 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 1006 if (err) 1007 return (err); 1008 zn = zap_name_alloc_uint64(zap, key, key_numints); 1009 if (zn == NULL) { 1010 zap_unlockdir(zap, FTAG); 1011 return (SET_ERROR(ENOTSUP)); 1012 } 1013 err = fzap_length(zn, integer_size, num_integers); 1014 zap_name_free(zn); 1015 zap_unlockdir(zap, FTAG); 1016 return (err); 1017} 1018 1019static void 1020mzap_addent(zap_name_t *zn, uint64_t value) 1021{ 1022 int i; 1023 zap_t *zap = zn->zn_zap; 1024 int start = zap->zap_m.zap_alloc_next; 1025 uint32_t cd; 1026 1027 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 1028 1029#ifdef ZFS_DEBUG 1030 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) { 1031 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i]; 1032 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0); 1033 } 1034#endif 1035 1036 cd = mze_find_unused_cd(zap, zn->zn_hash); 1037 /* given the limited size of the microzap, this can't happen */ 1038 ASSERT(cd < zap_maxcd(zap)); 1039 1040again: 1041 for (i = start; i < zap->zap_m.zap_num_chunks; i++) { 1042 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i]; 1043 if (mze->mze_name[0] == 0) { 1044 mze->mze_value = value; 1045 mze->mze_cd = cd; 1046 (void) strcpy(mze->mze_name, zn->zn_key_orig); 1047 zap->zap_m.zap_num_entries++; 1048 zap->zap_m.zap_alloc_next = i+1; 1049 if (zap->zap_m.zap_alloc_next == 1050 zap->zap_m.zap_num_chunks) 1051 zap->zap_m.zap_alloc_next = 0; 1052 VERIFY(0 == mze_insert(zap, i, zn->zn_hash)); 1053 return; 1054 } 1055 } 1056 if (start != 0) { 1057 start = 0; 1058 goto again; 1059 } 1060 ASSERT(!"out of entries!"); 1061} 1062 1063int 1064zap_add(objset_t *os, uint64_t zapobj, const char *key, 1065 int integer_size, uint64_t num_integers, 1066 const void *val, dmu_tx_t *tx) 1067{ 1068 zap_t *zap; 1069 int err; 1070 mzap_ent_t *mze; 1071 const uint64_t *intval = val; 1072 zap_name_t *zn; 1073 1074 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap); 1075 if (err) 1076 return (err); 1077 zn = zap_name_alloc(zap, key, MT_EXACT); 1078 if (zn == NULL) { 1079 zap_unlockdir(zap, FTAG); 1080 return (SET_ERROR(ENOTSUP)); 1081 } 1082 if (!zap->zap_ismicro) { 1083 err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx); 1084 zap = zn->zn_zap; /* fzap_add() may change zap */ 1085 } else if (integer_size != 8 || num_integers != 1 || 1086 strlen(key) >= MZAP_NAME_LEN) { 1087 err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0); 1088 if (err == 0) { 1089 err = fzap_add(zn, integer_size, num_integers, val, 1090 FTAG, tx); 1091 } 1092 zap = zn->zn_zap; /* fzap_add() may change zap */ 1093 } else { 1094 mze = mze_find(zn); 1095 if (mze != NULL) { 1096 err = SET_ERROR(EEXIST); 1097 } else { 1098 mzap_addent(zn, *intval); 1099 } 1100 } 1101 ASSERT(zap == zn->zn_zap); 1102 zap_name_free(zn); 1103 if (zap != NULL) /* may be NULL if fzap_add() failed */ 1104 zap_unlockdir(zap, FTAG); 1105 return (err); 1106} 1107 1108int 1109zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1110 int key_numints, int integer_size, uint64_t num_integers, 1111 const void *val, dmu_tx_t *tx) 1112{ 1113 zap_t *zap; 1114 int err; 1115 zap_name_t *zn; 1116 1117 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap); 1118 if (err) 1119 return (err); 1120 zn = zap_name_alloc_uint64(zap, key, key_numints); 1121 if (zn == NULL) { 1122 zap_unlockdir(zap, FTAG); 1123 return (SET_ERROR(ENOTSUP)); 1124 } 1125 err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx); 1126 zap = zn->zn_zap; /* fzap_add() may change zap */ 1127 zap_name_free(zn); 1128 if (zap != NULL) /* may be NULL if fzap_add() failed */ 1129 zap_unlockdir(zap, FTAG); 1130 return (err); 1131} 1132 1133int 1134zap_update(objset_t *os, uint64_t zapobj, const char *name, 1135 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx) 1136{ 1137 zap_t *zap; 1138 mzap_ent_t *mze; 1139 uint64_t oldval; 1140 const uint64_t *intval = val; 1141 zap_name_t *zn; 1142 int err; 1143 1144#ifdef ZFS_DEBUG 1145 /* 1146 * If there is an old value, it shouldn't change across the 1147 * lockdir (eg, due to bprewrite's xlation). 1148 */ 1149 if (integer_size == 8 && num_integers == 1) 1150 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval); 1151#endif 1152 1153 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap); 1154 if (err) 1155 return (err); 1156 zn = zap_name_alloc(zap, name, MT_EXACT); 1157 if (zn == NULL) { 1158 zap_unlockdir(zap, FTAG); 1159 return (SET_ERROR(ENOTSUP)); 1160 } 1161 if (!zap->zap_ismicro) { 1162 err = fzap_update(zn, integer_size, num_integers, val, 1163 FTAG, tx); 1164 zap = zn->zn_zap; /* fzap_update() may change zap */ 1165 } else if (integer_size != 8 || num_integers != 1 || 1166 strlen(name) >= MZAP_NAME_LEN) { 1167 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n", 1168 zapobj, integer_size, num_integers, name); 1169 err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0); 1170 if (err == 0) { 1171 err = fzap_update(zn, integer_size, num_integers, 1172 val, FTAG, tx); 1173 } 1174 zap = zn->zn_zap; /* fzap_update() may change zap */ 1175 } else { 1176 mze = mze_find(zn); 1177 if (mze != NULL) { 1178 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval); 1179 MZE_PHYS(zap, mze)->mze_value = *intval; 1180 } else { 1181 mzap_addent(zn, *intval); 1182 } 1183 } 1184 ASSERT(zap == zn->zn_zap); 1185 zap_name_free(zn); 1186 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */ 1187 zap_unlockdir(zap, FTAG); 1188 return (err); 1189} 1190 1191int 1192zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1193 int key_numints, 1194 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx) 1195{ 1196 zap_t *zap; 1197 zap_name_t *zn; 1198 int err; 1199 1200 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap); 1201 if (err) 1202 return (err); 1203 zn = zap_name_alloc_uint64(zap, key, key_numints); 1204 if (zn == NULL) { 1205 zap_unlockdir(zap, FTAG); 1206 return (SET_ERROR(ENOTSUP)); 1207 } 1208 err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx); 1209 zap = zn->zn_zap; /* fzap_update() may change zap */ 1210 zap_name_free(zn); 1211 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */ 1212 zap_unlockdir(zap, FTAG); 1213 return (err); 1214} 1215 1216int 1217zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx) 1218{ 1219 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx)); 1220} 1221 1222int 1223zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name, 1224 matchtype_t mt, dmu_tx_t *tx) 1225{ 1226 zap_t *zap; 1227 int err; 1228 mzap_ent_t *mze; 1229 zap_name_t *zn; 1230 1231 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap); 1232 if (err) 1233 return (err); 1234 zn = zap_name_alloc(zap, name, mt); 1235 if (zn == NULL) { 1236 zap_unlockdir(zap, FTAG); 1237 return (SET_ERROR(ENOTSUP)); 1238 } 1239 if (!zap->zap_ismicro) { 1240 err = fzap_remove(zn, tx); 1241 } else { 1242 mze = mze_find(zn); 1243 if (mze == NULL) { 1244 err = SET_ERROR(ENOENT); 1245 } else { 1246 zap->zap_m.zap_num_entries--; 1247 bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid], 1248 sizeof (mzap_ent_phys_t)); 1249 mze_remove(zap, mze); 1250 } 1251 } 1252 zap_name_free(zn); 1253 zap_unlockdir(zap, FTAG); 1254 return (err); 1255} 1256 1257int 1258zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1259 int key_numints, dmu_tx_t *tx) 1260{ 1261 zap_t *zap; 1262 int err; 1263 zap_name_t *zn; 1264 1265 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap); 1266 if (err) 1267 return (err); 1268 zn = zap_name_alloc_uint64(zap, key, key_numints); 1269 if (zn == NULL) { 1270 zap_unlockdir(zap, FTAG); 1271 return (SET_ERROR(ENOTSUP)); 1272 } 1273 err = fzap_remove(zn, tx); 1274 zap_name_free(zn); 1275 zap_unlockdir(zap, FTAG); 1276 return (err); 1277} 1278 1279/* 1280 * Routines for iterating over the attributes. 1281 */ 1282 1283void 1284zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj, 1285 uint64_t serialized) 1286{ 1287 zc->zc_objset = os; 1288 zc->zc_zap = NULL; 1289 zc->zc_leaf = NULL; 1290 zc->zc_zapobj = zapobj; 1291 zc->zc_serialized = serialized; 1292 zc->zc_hash = 0; 1293 zc->zc_cd = 0; 1294} 1295 1296void 1297zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj) 1298{ 1299 zap_cursor_init_serialized(zc, os, zapobj, 0); 1300} 1301 1302void 1303zap_cursor_fini(zap_cursor_t *zc) 1304{ 1305 if (zc->zc_zap) { 1306 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 1307 zap_unlockdir(zc->zc_zap, NULL); 1308 zc->zc_zap = NULL; 1309 } 1310 if (zc->zc_leaf) { 1311 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1312 zap_put_leaf(zc->zc_leaf); 1313 zc->zc_leaf = NULL; 1314 } 1315 zc->zc_objset = NULL; 1316} 1317 1318uint64_t 1319zap_cursor_serialize(zap_cursor_t *zc) 1320{ 1321 if (zc->zc_hash == -1ULL) 1322 return (-1ULL); 1323 if (zc->zc_zap == NULL) 1324 return (zc->zc_serialized); 1325 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0); 1326 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap)); 1327 1328 /* 1329 * We want to keep the high 32 bits of the cursor zero if we can, so 1330 * that 32-bit programs can access this. So usually use a small 1331 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits 1332 * of the cursor. 1333 * 1334 * [ collision differentiator | zap_hashbits()-bit hash value ] 1335 */ 1336 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) | 1337 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap))); 1338} 1339 1340int 1341zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za) 1342{ 1343 int err; 1344 avl_index_t idx; 1345 mzap_ent_t mze_tofind; 1346 mzap_ent_t *mze; 1347 1348 if (zc->zc_hash == -1ULL) 1349 return (SET_ERROR(ENOENT)); 1350 1351 if (zc->zc_zap == NULL) { 1352 int hb; 1353 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL, 1354 RW_READER, TRUE, FALSE, NULL, &zc->zc_zap); 1355 if (err) 1356 return (err); 1357 1358 /* 1359 * To support zap_cursor_init_serialized, advance, retrieve, 1360 * we must add to the existing zc_cd, which may already 1361 * be 1 due to the zap_cursor_advance. 1362 */ 1363 ASSERT(zc->zc_hash == 0); 1364 hb = zap_hashbits(zc->zc_zap); 1365 zc->zc_hash = zc->zc_serialized << (64 - hb); 1366 zc->zc_cd += zc->zc_serialized >> hb; 1367 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */ 1368 zc->zc_cd = 0; 1369 } else { 1370 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 1371 } 1372 if (!zc->zc_zap->zap_ismicro) { 1373 err = fzap_cursor_retrieve(zc->zc_zap, zc, za); 1374 } else { 1375 mze_tofind.mze_hash = zc->zc_hash; 1376 mze_tofind.mze_cd = zc->zc_cd; 1377 1378 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx); 1379 if (mze == NULL) { 1380 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl, 1381 idx, AVL_AFTER); 1382 } 1383 if (mze) { 1384 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze); 1385 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd); 1386 za->za_normalization_conflict = 1387 mzap_normalization_conflict(zc->zc_zap, NULL, mze); 1388 za->za_integer_length = 8; 1389 za->za_num_integers = 1; 1390 za->za_first_integer = mzep->mze_value; 1391 (void) strcpy(za->za_name, mzep->mze_name); 1392 zc->zc_hash = mze->mze_hash; 1393 zc->zc_cd = mze->mze_cd; 1394 err = 0; 1395 } else { 1396 zc->zc_hash = -1ULL; 1397 err = SET_ERROR(ENOENT); 1398 } 1399 } 1400 rw_exit(&zc->zc_zap->zap_rwlock); 1401 return (err); 1402} 1403 1404void 1405zap_cursor_advance(zap_cursor_t *zc) 1406{ 1407 if (zc->zc_hash == -1ULL) 1408 return; 1409 zc->zc_cd++; 1410} 1411 1412int 1413zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt) 1414{ 1415 int err = 0; 1416 mzap_ent_t *mze; 1417 zap_name_t *zn; 1418 1419 if (zc->zc_zap == NULL) { 1420 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL, 1421 RW_READER, TRUE, FALSE, FTAG, &zc->zc_zap); 1422 if (err) 1423 return (err); 1424 } else { 1425 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 1426 } 1427 1428 zn = zap_name_alloc(zc->zc_zap, name, mt); 1429 if (zn == NULL) { 1430 rw_exit(&zc->zc_zap->zap_rwlock); 1431 return (SET_ERROR(ENOTSUP)); 1432 } 1433 1434 if (!zc->zc_zap->zap_ismicro) { 1435 err = fzap_cursor_move_to_key(zc, zn); 1436 } else { 1437 mze = mze_find(zn); 1438 if (mze == NULL) { 1439 err = SET_ERROR(ENOENT); 1440 goto out; 1441 } 1442 zc->zc_hash = mze->mze_hash; 1443 zc->zc_cd = mze->mze_cd; 1444 } 1445 1446out: 1447 zap_name_free(zn); 1448 rw_exit(&zc->zc_zap->zap_rwlock); 1449 return (err); 1450} 1451 1452int 1453zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs) 1454{ 1455 int err; 1456 zap_t *zap; 1457 1458 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 1459 if (err) 1460 return (err); 1461 1462 bzero(zs, sizeof (zap_stats_t)); 1463 1464 if (zap->zap_ismicro) { 1465 zs->zs_blocksize = zap->zap_dbuf->db_size; 1466 zs->zs_num_entries = zap->zap_m.zap_num_entries; 1467 zs->zs_num_blocks = 1; 1468 } else { 1469 fzap_get_stats(zap, zs); 1470 } 1471 zap_unlockdir(zap, FTAG); 1472 return (0); 1473} 1474 1475int 1476zap_count_write_by_dnode(dnode_t *dn, const char *name, int add, 1477 refcount_t *towrite, refcount_t *tooverwrite) 1478{ 1479 zap_t *zap; 1480 int err = 0; 1481 1482 /* 1483 * Since, we don't have a name, we cannot figure out which blocks will 1484 * be affected in this operation. So, account for the worst case : 1485 * - 3 blocks overwritten: target leaf, ptrtbl block, header block 1486 * - 4 new blocks written if adding: 1487 * - 2 blocks for possibly split leaves, 1488 * - 2 grown ptrtbl blocks 1489 * 1490 * This also accommodates the case where an add operation to a fairly 1491 * large microzap results in a promotion to fatzap. 1492 */ 1493 if (name == NULL) { 1494 (void) refcount_add_many(towrite, 1495 (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE, FTAG); 1496 return (err); 1497 } 1498 1499 /* 1500 * We lock the zap with adding == FALSE. Because, if we pass 1501 * the actual value of add, it could trigger a mzap_upgrade(). 1502 * At present we are just evaluating the possibility of this operation 1503 * and hence we do not want to trigger an upgrade. 1504 */ 1505 err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE, 1506 FTAG, &zap); 1507 if (err != 0) 1508 return (err); 1509 1510 if (!zap->zap_ismicro) { 1511 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT); 1512 if (zn) { 1513 err = fzap_count_write(zn, add, towrite, 1514 tooverwrite); 1515 zap_name_free(zn); 1516 } else { 1517 /* 1518 * We treat this case as similar to (name == NULL) 1519 */ 1520 (void) refcount_add_many(towrite, 1521 (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE, FTAG); 1522 } 1523 } else { 1524 /* 1525 * We are here if (name != NULL) and this is a micro-zap. 1526 * We account for the header block depending on whether it 1527 * is freeable. 1528 * 1529 * Incase of an add-operation it is hard to find out 1530 * if this add will promote this microzap to fatzap. 1531 * Hence, we consider the worst case and account for the 1532 * blocks assuming this microzap would be promoted to a 1533 * fatzap. 1534 * 1535 * 1 block overwritten : header block 1536 * 4 new blocks written : 2 new split leaf, 2 grown 1537 * ptrtbl blocks 1538 */ 1539 if (dmu_buf_freeable(zap->zap_dbuf)) { 1540 (void) refcount_add_many(tooverwrite, 1541 MZAP_MAX_BLKSZ, FTAG); 1542 } else { 1543 (void) refcount_add_many(towrite, 1544 MZAP_MAX_BLKSZ, FTAG); 1545 } 1546 1547 if (add) { 1548 (void) refcount_add_many(towrite, 1549 4 * MZAP_MAX_BLKSZ, FTAG); 1550 } 1551 } 1552 1553 zap_unlockdir(zap, FTAG); 1554 return (err); 1555} 1556