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