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 static int 543 zap_lockdir_by_dnode(dnode_t *dn, 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_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH); 550 if (err != 0) { 551 return (err); 552 } 553 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp); 554 if (err != 0) { 555 dmu_buf_rele(db, tag); 556 } 557 return (err); 558 } 559 560 int 561 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx, 562 krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp) 563 { 564 dmu_buf_t *db; 565 int err; 566 567 err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH); 568 if (err != 0) 569 return (err); 570 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp); 571 if (err != 0) 572 dmu_buf_rele(db, tag); 573 return (err); 574 } 575 576 void 577 zap_unlockdir(zap_t *zap, void *tag) 578 { 579 rw_exit(&zap->zap_rwlock); 580 dmu_buf_rele(zap->zap_dbuf, tag); 581 } 582 583 static int 584 mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags) 585 { 586 mzap_phys_t *mzp; 587 int i, sz, nchunks; 588 int err = 0; 589 zap_t *zap = *zapp; 590 591 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 592 593 sz = zap->zap_dbuf->db_size; 594 mzp = zio_buf_alloc(sz); 595 bcopy(zap->zap_dbuf->db_data, mzp, sz); 596 nchunks = zap->zap_m.zap_num_chunks; 597 598 if (!flags) { 599 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object, 600 1ULL << fzap_default_block_shift, 0, tx); 601 if (err) { 602 zio_buf_free(mzp, sz); 603 return (err); 604 } 605 } 606 607 dprintf("upgrading obj=%llu with %u chunks\n", 608 zap->zap_object, nchunks); 609 /* XXX destroy the avl later, so we can use the stored hash value */ 610 mze_destroy(zap); 611 612 fzap_upgrade(zap, tx, flags); 613 614 for (i = 0; i < nchunks; i++) { 615 mzap_ent_phys_t *mze = &mzp->mz_chunk[i]; 616 zap_name_t *zn; 617 if (mze->mze_name[0] == 0) 618 continue; 619 dprintf("adding %s=%llu\n", 620 mze->mze_name, mze->mze_value); 621 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT); 622 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, 623 tag, tx); 624 zap = zn->zn_zap; /* fzap_add_cd() may change zap */ 625 zap_name_free(zn); 626 if (err) 627 break; 628 } 629 zio_buf_free(mzp, sz); 630 *zapp = zap; 631 return (err); 632 } 633 634 void 635 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags, 636 dmu_tx_t *tx) 637 { 638 dmu_buf_t *db; 639 mzap_phys_t *zp; 640 641 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH)); 642 643 #ifdef ZFS_DEBUG 644 { 645 dmu_object_info_t doi; 646 dmu_object_info_from_db(db, &doi); 647 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP); 648 } 649 #endif 650 651 dmu_buf_will_dirty(db, tx); 652 zp = db->db_data; 653 zp->mz_block_type = ZBT_MICRO; 654 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL; 655 zp->mz_normflags = normflags; 656 dmu_buf_rele(db, FTAG); 657 658 if (flags != 0) { 659 zap_t *zap; 660 /* Only fat zap supports flags; upgrade immediately. */ 661 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER, 662 B_FALSE, B_FALSE, FTAG, &zap)); 663 VERIFY3U(0, ==, mzap_upgrade(&zap, FTAG, tx, flags)); 664 zap_unlockdir(zap, FTAG); 665 } 666 } 667 668 int 669 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot, 670 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 671 { 672 return (zap_create_claim_norm(os, obj, 673 0, ot, bonustype, bonuslen, tx)); 674 } 675 676 int 677 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags, 678 dmu_object_type_t ot, 679 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 680 { 681 int err; 682 683 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx); 684 if (err != 0) 685 return (err); 686 mzap_create_impl(os, obj, normflags, 0, tx); 687 return (0); 688 } 689 690 uint64_t 691 zap_create(objset_t *os, dmu_object_type_t ot, 692 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 693 { 694 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx)); 695 } 696 697 uint64_t 698 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot, 699 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 700 { 701 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx); 702 703 mzap_create_impl(os, obj, normflags, 0, tx); 704 return (obj); 705 } 706 707 uint64_t 708 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags, 709 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift, 710 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 711 { 712 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx); 713 714 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT && 715 leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT && 716 indirect_blockshift >= SPA_MINBLOCKSHIFT && 717 indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT); 718 719 VERIFY(dmu_object_set_blocksize(os, obj, 720 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0); 721 722 mzap_create_impl(os, obj, normflags, flags, tx); 723 return (obj); 724 } 725 726 int 727 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx) 728 { 729 /* 730 * dmu_object_free will free the object number and free the 731 * data. Freeing the data will cause our pageout function to be 732 * called, which will destroy our data (zap_leaf_t's and zap_t). 733 */ 734 735 return (dmu_object_free(os, zapobj, tx)); 736 } 737 738 void 739 zap_evict(void *dbu) 740 { 741 zap_t *zap = dbu; 742 743 rw_destroy(&zap->zap_rwlock); 744 745 if (zap->zap_ismicro) 746 mze_destroy(zap); 747 else 748 mutex_destroy(&zap->zap_f.zap_num_entries_mtx); 749 750 kmem_free(zap, sizeof (zap_t)); 751 } 752 753 int 754 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count) 755 { 756 zap_t *zap; 757 int err; 758 759 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 760 if (err) 761 return (err); 762 if (!zap->zap_ismicro) { 763 err = fzap_count(zap, count); 764 } else { 765 *count = zap->zap_m.zap_num_entries; 766 } 767 zap_unlockdir(zap, FTAG); 768 return (err); 769 } 770 771 /* 772 * zn may be NULL; if not specified, it will be computed if needed. 773 * See also the comment above zap_entry_normalization_conflict(). 774 */ 775 static boolean_t 776 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze) 777 { 778 mzap_ent_t *other; 779 int direction = AVL_BEFORE; 780 boolean_t allocdzn = B_FALSE; 781 782 if (zap->zap_normflags == 0) 783 return (B_FALSE); 784 785 again: 786 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction); 787 other && other->mze_hash == mze->mze_hash; 788 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) { 789 790 if (zn == NULL) { 791 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name, 792 MT_FIRST); 793 allocdzn = B_TRUE; 794 } 795 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) { 796 if (allocdzn) 797 zap_name_free(zn); 798 return (B_TRUE); 799 } 800 } 801 802 if (direction == AVL_BEFORE) { 803 direction = AVL_AFTER; 804 goto again; 805 } 806 807 if (allocdzn) 808 zap_name_free(zn); 809 return (B_FALSE); 810 } 811 812 /* 813 * Routines for manipulating attributes. 814 */ 815 816 int 817 zap_lookup(objset_t *os, uint64_t zapobj, const char *name, 818 uint64_t integer_size, uint64_t num_integers, void *buf) 819 { 820 return (zap_lookup_norm(os, zapobj, name, integer_size, 821 num_integers, buf, MT_EXACT, NULL, 0, NULL)); 822 } 823 824 static int 825 zap_lookup_impl(zap_t *zap, const char *name, 826 uint64_t integer_size, uint64_t num_integers, void *buf, 827 matchtype_t mt, char *realname, int rn_len, 828 boolean_t *ncp) 829 { 830 int err = 0; 831 mzap_ent_t *mze; 832 zap_name_t *zn; 833 834 zn = zap_name_alloc(zap, name, mt); 835 if (zn == NULL) 836 return (SET_ERROR(ENOTSUP)); 837 838 if (!zap->zap_ismicro) { 839 err = fzap_lookup(zn, integer_size, num_integers, buf, 840 realname, rn_len, ncp); 841 } else { 842 mze = mze_find(zn); 843 if (mze == NULL) { 844 err = SET_ERROR(ENOENT); 845 } else { 846 if (num_integers < 1) { 847 err = SET_ERROR(EOVERFLOW); 848 } else if (integer_size != 8) { 849 err = SET_ERROR(EINVAL); 850 } else { 851 *(uint64_t *)buf = 852 MZE_PHYS(zap, mze)->mze_value; 853 (void) strlcpy(realname, 854 MZE_PHYS(zap, mze)->mze_name, rn_len); 855 if (ncp) { 856 *ncp = mzap_normalization_conflict(zap, 857 zn, mze); 858 } 859 } 860 } 861 } 862 zap_name_free(zn); 863 return (err); 864 } 865 866 int 867 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name, 868 uint64_t integer_size, uint64_t num_integers, void *buf, 869 matchtype_t mt, char *realname, int rn_len, 870 boolean_t *ncp) 871 { 872 zap_t *zap; 873 int err; 874 875 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 876 if (err != 0) 877 return (err); 878 err = zap_lookup_impl(zap, name, integer_size, 879 num_integers, buf, mt, realname, rn_len, ncp); 880 zap_unlockdir(zap, FTAG); 881 return (err); 882 } 883 884 int 885 zap_lookup_by_dnode(dnode_t *dn, const char *name, 886 uint64_t integer_size, uint64_t num_integers, void *buf) 887 { 888 return (zap_lookup_norm_by_dnode(dn, name, integer_size, 889 num_integers, buf, MT_EXACT, NULL, 0, NULL)); 890 } 891 892 int 893 zap_lookup_norm_by_dnode(dnode_t *dn, const char *name, 894 uint64_t integer_size, uint64_t num_integers, void *buf, 895 matchtype_t mt, char *realname, int rn_len, 896 boolean_t *ncp) 897 { 898 zap_t *zap; 899 int err; 900 901 err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE, 902 FTAG, &zap); 903 if (err != 0) 904 return (err); 905 err = zap_lookup_impl(zap, name, integer_size, 906 num_integers, buf, mt, realname, rn_len, ncp); 907 zap_unlockdir(zap, FTAG); 908 return (err); 909 } 910 911 int 912 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 913 int key_numints) 914 { 915 zap_t *zap; 916 int err; 917 zap_name_t *zn; 918 919 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 920 if (err) 921 return (err); 922 zn = zap_name_alloc_uint64(zap, key, key_numints); 923 if (zn == NULL) { 924 zap_unlockdir(zap, FTAG); 925 return (SET_ERROR(ENOTSUP)); 926 } 927 928 fzap_prefetch(zn); 929 zap_name_free(zn); 930 zap_unlockdir(zap, FTAG); 931 return (err); 932 } 933 934 int 935 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 936 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf) 937 { 938 zap_t *zap; 939 int err; 940 zap_name_t *zn; 941 942 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 943 if (err) 944 return (err); 945 zn = zap_name_alloc_uint64(zap, key, key_numints); 946 if (zn == NULL) { 947 zap_unlockdir(zap, FTAG); 948 return (SET_ERROR(ENOTSUP)); 949 } 950 951 err = fzap_lookup(zn, integer_size, num_integers, buf, 952 NULL, 0, NULL); 953 zap_name_free(zn); 954 zap_unlockdir(zap, FTAG); 955 return (err); 956 } 957 958 int 959 zap_contains(objset_t *os, uint64_t zapobj, const char *name) 960 { 961 int err = zap_lookup_norm(os, zapobj, name, 0, 962 0, NULL, MT_EXACT, NULL, 0, NULL); 963 if (err == EOVERFLOW || err == EINVAL) 964 err = 0; /* found, but skipped reading the value */ 965 return (err); 966 } 967 968 int 969 zap_length(objset_t *os, uint64_t zapobj, const char *name, 970 uint64_t *integer_size, uint64_t *num_integers) 971 { 972 zap_t *zap; 973 int err; 974 mzap_ent_t *mze; 975 zap_name_t *zn; 976 977 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 978 if (err) 979 return (err); 980 zn = zap_name_alloc(zap, name, MT_EXACT); 981 if (zn == NULL) { 982 zap_unlockdir(zap, FTAG); 983 return (SET_ERROR(ENOTSUP)); 984 } 985 if (!zap->zap_ismicro) { 986 err = fzap_length(zn, integer_size, num_integers); 987 } else { 988 mze = mze_find(zn); 989 if (mze == NULL) { 990 err = SET_ERROR(ENOENT); 991 } else { 992 if (integer_size) 993 *integer_size = 8; 994 if (num_integers) 995 *num_integers = 1; 996 } 997 } 998 zap_name_free(zn); 999 zap_unlockdir(zap, FTAG); 1000 return (err); 1001 } 1002 1003 int 1004 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1005 int key_numints, uint64_t *integer_size, uint64_t *num_integers) 1006 { 1007 zap_t *zap; 1008 int err; 1009 zap_name_t *zn; 1010 1011 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 1012 if (err) 1013 return (err); 1014 zn = zap_name_alloc_uint64(zap, key, key_numints); 1015 if (zn == NULL) { 1016 zap_unlockdir(zap, FTAG); 1017 return (SET_ERROR(ENOTSUP)); 1018 } 1019 err = fzap_length(zn, integer_size, num_integers); 1020 zap_name_free(zn); 1021 zap_unlockdir(zap, FTAG); 1022 return (err); 1023 } 1024 1025 static void 1026 mzap_addent(zap_name_t *zn, uint64_t value) 1027 { 1028 int i; 1029 zap_t *zap = zn->zn_zap; 1030 int start = zap->zap_m.zap_alloc_next; 1031 uint32_t cd; 1032 1033 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 1034 1035 #ifdef ZFS_DEBUG 1036 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) { 1037 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i]; 1038 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0); 1039 } 1040 #endif 1041 1042 cd = mze_find_unused_cd(zap, zn->zn_hash); 1043 /* given the limited size of the microzap, this can't happen */ 1044 ASSERT(cd < zap_maxcd(zap)); 1045 1046 again: 1047 for (i = start; i < zap->zap_m.zap_num_chunks; i++) { 1048 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i]; 1049 if (mze->mze_name[0] == 0) { 1050 mze->mze_value = value; 1051 mze->mze_cd = cd; 1052 (void) strcpy(mze->mze_name, zn->zn_key_orig); 1053 zap->zap_m.zap_num_entries++; 1054 zap->zap_m.zap_alloc_next = i+1; 1055 if (zap->zap_m.zap_alloc_next == 1056 zap->zap_m.zap_num_chunks) 1057 zap->zap_m.zap_alloc_next = 0; 1058 mze_insert(zap, i, zn->zn_hash); 1059 return; 1060 } 1061 } 1062 if (start != 0) { 1063 start = 0; 1064 goto again; 1065 } 1066 ASSERT(!"out of entries!"); 1067 } 1068 1069 int 1070 zap_add(objset_t *os, uint64_t zapobj, const char *key, 1071 int integer_size, uint64_t num_integers, 1072 const void *val, dmu_tx_t *tx) 1073 { 1074 zap_t *zap; 1075 int err; 1076 mzap_ent_t *mze; 1077 const uint64_t *intval = val; 1078 zap_name_t *zn; 1079 1080 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap); 1081 if (err) 1082 return (err); 1083 zn = zap_name_alloc(zap, key, MT_EXACT); 1084 if (zn == NULL) { 1085 zap_unlockdir(zap, FTAG); 1086 return (SET_ERROR(ENOTSUP)); 1087 } 1088 if (!zap->zap_ismicro) { 1089 err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx); 1090 zap = zn->zn_zap; /* fzap_add() may change zap */ 1091 } else if (integer_size != 8 || num_integers != 1 || 1092 strlen(key) >= MZAP_NAME_LEN) { 1093 err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0); 1094 if (err == 0) { 1095 err = fzap_add(zn, integer_size, num_integers, val, 1096 FTAG, tx); 1097 } 1098 zap = zn->zn_zap; /* fzap_add() may change zap */ 1099 } else { 1100 mze = mze_find(zn); 1101 if (mze != NULL) { 1102 err = SET_ERROR(EEXIST); 1103 } else { 1104 mzap_addent(zn, *intval); 1105 } 1106 } 1107 ASSERT(zap == zn->zn_zap); 1108 zap_name_free(zn); 1109 if (zap != NULL) /* may be NULL if fzap_add() failed */ 1110 zap_unlockdir(zap, FTAG); 1111 return (err); 1112 } 1113 1114 int 1115 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1116 int key_numints, int integer_size, uint64_t num_integers, 1117 const void *val, dmu_tx_t *tx) 1118 { 1119 zap_t *zap; 1120 int err; 1121 zap_name_t *zn; 1122 1123 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap); 1124 if (err) 1125 return (err); 1126 zn = zap_name_alloc_uint64(zap, key, key_numints); 1127 if (zn == NULL) { 1128 zap_unlockdir(zap, FTAG); 1129 return (SET_ERROR(ENOTSUP)); 1130 } 1131 err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx); 1132 zap = zn->zn_zap; /* fzap_add() may change zap */ 1133 zap_name_free(zn); 1134 if (zap != NULL) /* may be NULL if fzap_add() failed */ 1135 zap_unlockdir(zap, FTAG); 1136 return (err); 1137 } 1138 1139 int 1140 zap_update(objset_t *os, uint64_t zapobj, const char *name, 1141 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx) 1142 { 1143 zap_t *zap; 1144 mzap_ent_t *mze; 1145 uint64_t oldval; 1146 const uint64_t *intval = val; 1147 zap_name_t *zn; 1148 int err; 1149 1150 #ifdef ZFS_DEBUG 1151 /* 1152 * If there is an old value, it shouldn't change across the 1153 * lockdir (eg, due to bprewrite's xlation). 1154 */ 1155 if (integer_size == 8 && num_integers == 1) 1156 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval); 1157 #endif 1158 1159 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap); 1160 if (err) 1161 return (err); 1162 zn = zap_name_alloc(zap, name, MT_EXACT); 1163 if (zn == NULL) { 1164 zap_unlockdir(zap, FTAG); 1165 return (SET_ERROR(ENOTSUP)); 1166 } 1167 if (!zap->zap_ismicro) { 1168 err = fzap_update(zn, integer_size, num_integers, val, 1169 FTAG, tx); 1170 zap = zn->zn_zap; /* fzap_update() may change zap */ 1171 } else if (integer_size != 8 || num_integers != 1 || 1172 strlen(name) >= MZAP_NAME_LEN) { 1173 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n", 1174 zapobj, integer_size, num_integers, name); 1175 err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0); 1176 if (err == 0) { 1177 err = fzap_update(zn, integer_size, num_integers, 1178 val, FTAG, tx); 1179 } 1180 zap = zn->zn_zap; /* fzap_update() may change zap */ 1181 } else { 1182 mze = mze_find(zn); 1183 if (mze != NULL) { 1184 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval); 1185 MZE_PHYS(zap, mze)->mze_value = *intval; 1186 } else { 1187 mzap_addent(zn, *intval); 1188 } 1189 } 1190 ASSERT(zap == zn->zn_zap); 1191 zap_name_free(zn); 1192 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */ 1193 zap_unlockdir(zap, FTAG); 1194 return (err); 1195 } 1196 1197 int 1198 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1199 int key_numints, 1200 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx) 1201 { 1202 zap_t *zap; 1203 zap_name_t *zn; 1204 int err; 1205 1206 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap); 1207 if (err) 1208 return (err); 1209 zn = zap_name_alloc_uint64(zap, key, key_numints); 1210 if (zn == NULL) { 1211 zap_unlockdir(zap, FTAG); 1212 return (SET_ERROR(ENOTSUP)); 1213 } 1214 err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx); 1215 zap = zn->zn_zap; /* fzap_update() may change zap */ 1216 zap_name_free(zn); 1217 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */ 1218 zap_unlockdir(zap, FTAG); 1219 return (err); 1220 } 1221 1222 int 1223 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx) 1224 { 1225 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx)); 1226 } 1227 1228 int 1229 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name, 1230 matchtype_t mt, dmu_tx_t *tx) 1231 { 1232 zap_t *zap; 1233 int err; 1234 mzap_ent_t *mze; 1235 zap_name_t *zn; 1236 1237 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap); 1238 if (err) 1239 return (err); 1240 zn = zap_name_alloc(zap, name, mt); 1241 if (zn == NULL) { 1242 zap_unlockdir(zap, FTAG); 1243 return (SET_ERROR(ENOTSUP)); 1244 } 1245 if (!zap->zap_ismicro) { 1246 err = fzap_remove(zn, tx); 1247 } else { 1248 mze = mze_find(zn); 1249 if (mze == NULL) { 1250 err = SET_ERROR(ENOENT); 1251 } else { 1252 zap->zap_m.zap_num_entries--; 1253 bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid], 1254 sizeof (mzap_ent_phys_t)); 1255 mze_remove(zap, mze); 1256 } 1257 } 1258 zap_name_free(zn); 1259 zap_unlockdir(zap, FTAG); 1260 return (err); 1261 } 1262 1263 int 1264 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key, 1265 int key_numints, dmu_tx_t *tx) 1266 { 1267 zap_t *zap; 1268 int err; 1269 zap_name_t *zn; 1270 1271 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap); 1272 if (err) 1273 return (err); 1274 zn = zap_name_alloc_uint64(zap, key, key_numints); 1275 if (zn == NULL) { 1276 zap_unlockdir(zap, FTAG); 1277 return (SET_ERROR(ENOTSUP)); 1278 } 1279 err = fzap_remove(zn, tx); 1280 zap_name_free(zn); 1281 zap_unlockdir(zap, FTAG); 1282 return (err); 1283 } 1284 1285 /* 1286 * Routines for iterating over the attributes. 1287 */ 1288 1289 void 1290 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj, 1291 uint64_t serialized) 1292 { 1293 zc->zc_objset = os; 1294 zc->zc_zap = NULL; 1295 zc->zc_leaf = NULL; 1296 zc->zc_zapobj = zapobj; 1297 zc->zc_serialized = serialized; 1298 zc->zc_hash = 0; 1299 zc->zc_cd = 0; 1300 } 1301 1302 void 1303 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj) 1304 { 1305 zap_cursor_init_serialized(zc, os, zapobj, 0); 1306 } 1307 1308 void 1309 zap_cursor_fini(zap_cursor_t *zc) 1310 { 1311 if (zc->zc_zap) { 1312 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 1313 zap_unlockdir(zc->zc_zap, NULL); 1314 zc->zc_zap = NULL; 1315 } 1316 if (zc->zc_leaf) { 1317 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1318 zap_put_leaf(zc->zc_leaf); 1319 zc->zc_leaf = NULL; 1320 } 1321 zc->zc_objset = NULL; 1322 } 1323 1324 uint64_t 1325 zap_cursor_serialize(zap_cursor_t *zc) 1326 { 1327 if (zc->zc_hash == -1ULL) 1328 return (-1ULL); 1329 if (zc->zc_zap == NULL) 1330 return (zc->zc_serialized); 1331 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0); 1332 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap)); 1333 1334 /* 1335 * We want to keep the high 32 bits of the cursor zero if we can, so 1336 * that 32-bit programs can access this. So usually use a small 1337 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits 1338 * of the cursor. 1339 * 1340 * [ collision differentiator | zap_hashbits()-bit hash value ] 1341 */ 1342 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) | 1343 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap))); 1344 } 1345 1346 int 1347 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za) 1348 { 1349 int err; 1350 avl_index_t idx; 1351 mzap_ent_t mze_tofind; 1352 mzap_ent_t *mze; 1353 1354 if (zc->zc_hash == -1ULL) 1355 return (SET_ERROR(ENOENT)); 1356 1357 if (zc->zc_zap == NULL) { 1358 int hb; 1359 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL, 1360 RW_READER, TRUE, FALSE, NULL, &zc->zc_zap); 1361 if (err) 1362 return (err); 1363 1364 /* 1365 * To support zap_cursor_init_serialized, advance, retrieve, 1366 * we must add to the existing zc_cd, which may already 1367 * be 1 due to the zap_cursor_advance. 1368 */ 1369 ASSERT(zc->zc_hash == 0); 1370 hb = zap_hashbits(zc->zc_zap); 1371 zc->zc_hash = zc->zc_serialized << (64 - hb); 1372 zc->zc_cd += zc->zc_serialized >> hb; 1373 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */ 1374 zc->zc_cd = 0; 1375 } else { 1376 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 1377 } 1378 if (!zc->zc_zap->zap_ismicro) { 1379 err = fzap_cursor_retrieve(zc->zc_zap, zc, za); 1380 } else { 1381 mze_tofind.mze_hash = zc->zc_hash; 1382 mze_tofind.mze_cd = zc->zc_cd; 1383 1384 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx); 1385 if (mze == NULL) { 1386 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl, 1387 idx, AVL_AFTER); 1388 } 1389 if (mze) { 1390 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze); 1391 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd); 1392 za->za_normalization_conflict = 1393 mzap_normalization_conflict(zc->zc_zap, NULL, mze); 1394 za->za_integer_length = 8; 1395 za->za_num_integers = 1; 1396 za->za_first_integer = mzep->mze_value; 1397 (void) strcpy(za->za_name, mzep->mze_name); 1398 zc->zc_hash = mze->mze_hash; 1399 zc->zc_cd = mze->mze_cd; 1400 err = 0; 1401 } else { 1402 zc->zc_hash = -1ULL; 1403 err = SET_ERROR(ENOENT); 1404 } 1405 } 1406 rw_exit(&zc->zc_zap->zap_rwlock); 1407 return (err); 1408 } 1409 1410 void 1411 zap_cursor_advance(zap_cursor_t *zc) 1412 { 1413 if (zc->zc_hash == -1ULL) 1414 return; 1415 zc->zc_cd++; 1416 } 1417 1418 int 1419 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs) 1420 { 1421 int err; 1422 zap_t *zap; 1423 1424 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap); 1425 if (err) 1426 return (err); 1427 1428 bzero(zs, sizeof (zap_stats_t)); 1429 1430 if (zap->zap_ismicro) { 1431 zs->zs_blocksize = zap->zap_dbuf->db_size; 1432 zs->zs_num_entries = zap->zap_m.zap_num_entries; 1433 zs->zs_num_blocks = 1; 1434 } else { 1435 fzap_get_stats(zap, zs); 1436 } 1437 zap_unlockdir(zap, FTAG); 1438 return (0); 1439 } 1440 1441 int 1442 zap_count_write_by_dnode(dnode_t *dn, const char *name, int add, 1443 refcount_t *towrite, refcount_t *tooverwrite) 1444 { 1445 zap_t *zap; 1446 int err = 0; 1447 1448 /* 1449 * Since, we don't have a name, we cannot figure out which blocks will 1450 * be affected in this operation. So, account for the worst case : 1451 * - 3 blocks overwritten: target leaf, ptrtbl block, header block 1452 * - 4 new blocks written if adding: 1453 * - 2 blocks for possibly split leaves, 1454 * - 2 grown ptrtbl blocks 1455 * 1456 * This also accommodates the case where an add operation to a fairly 1457 * large microzap results in a promotion to fatzap. 1458 */ 1459 if (name == NULL) { 1460 (void) refcount_add_many(towrite, 1461 (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE, FTAG); 1462 return (err); 1463 } 1464 1465 /* 1466 * We lock the zap with adding == FALSE. Because, if we pass 1467 * the actual value of add, it could trigger a mzap_upgrade(). 1468 * At present we are just evaluating the possibility of this operation 1469 * and hence we do not want to trigger an upgrade. 1470 */ 1471 err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE, 1472 FTAG, &zap); 1473 if (err != 0) 1474 return (err); 1475 1476 if (!zap->zap_ismicro) { 1477 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT); 1478 if (zn) { 1479 err = fzap_count_write(zn, add, towrite, 1480 tooverwrite); 1481 zap_name_free(zn); 1482 } else { 1483 /* 1484 * We treat this case as similar to (name == NULL) 1485 */ 1486 (void) refcount_add_many(towrite, 1487 (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE, FTAG); 1488 } 1489 } else { 1490 /* 1491 * We are here if (name != NULL) and this is a micro-zap. 1492 * We account for the header block depending on whether it 1493 * is freeable. 1494 * 1495 * Incase of an add-operation it is hard to find out 1496 * if this add will promote this microzap to fatzap. 1497 * Hence, we consider the worst case and account for the 1498 * blocks assuming this microzap would be promoted to a 1499 * fatzap. 1500 * 1501 * 1 block overwritten : header block 1502 * 4 new blocks written : 2 new split leaf, 2 grown 1503 * ptrtbl blocks 1504 */ 1505 if (dmu_buf_freeable(zap->zap_dbuf)) { 1506 (void) refcount_add_many(tooverwrite, 1507 MZAP_MAX_BLKSZ, FTAG); 1508 } else { 1509 (void) refcount_add_many(towrite, 1510 MZAP_MAX_BLKSZ, FTAG); 1511 } 1512 1513 if (add) { 1514 (void) refcount_add_many(towrite, 1515 4 * MZAP_MAX_BLKSZ, FTAG); 1516 } 1517 } 1518 1519 zap_unlockdir(zap, FTAG); 1520 return (err); 1521 } 1522