zap_micro.c revision 185029
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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26#pragma ident "%Z%%M% %I% %E% SMI" 27 28#include <sys/spa.h> 29#include <sys/dmu.h> 30#include <sys/zfs_context.h> 31#include <sys/zap.h> 32#include <sys/refcount.h> 33#include <sys/zap_impl.h> 34#include <sys/zap_leaf.h> 35#include <sys/avl.h> 36 37#ifdef _KERNEL 38#include <sys/sunddi.h> 39#endif 40 41static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx); 42 43 44static uint64_t 45zap_hash(zap_t *zap, const char *normname) 46{ 47 const uint8_t *cp; 48 uint8_t c; 49 uint64_t crc = zap->zap_salt; 50 51 /* NB: name must already be normalized, if necessary */ 52 53 ASSERT(crc != 0); 54 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 55 for (cp = (const uint8_t *)normname; (c = *cp) != '\0'; cp++) { 56 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ c) & 0xFF]; 57 } 58 59 /* 60 * Only use 28 bits, since we need 4 bits in the cookie for the 61 * collision differentiator. We MUST use the high bits, since 62 * those are the ones that we first pay attention to when 63 * chosing the bucket. 64 */ 65 crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1); 66 67 return (crc); 68} 69 70static int 71zap_normalize(zap_t *zap, const char *name, char *namenorm) 72{ 73 size_t inlen, outlen; 74 int err; 75 76 inlen = strlen(name) + 1; 77 outlen = ZAP_MAXNAMELEN; 78 79 err = 0; 80 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen, 81 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL, U8_UNICODE_LATEST, 82 &err); 83 84 return (err); 85} 86 87boolean_t 88zap_match(zap_name_t *zn, const char *matchname) 89{ 90 if (zn->zn_matchtype == MT_FIRST) { 91 char norm[ZAP_MAXNAMELEN]; 92 93 if (zap_normalize(zn->zn_zap, matchname, norm) != 0) 94 return (B_FALSE); 95 96 return (strcmp(zn->zn_name_norm, norm) == 0); 97 } else { 98 /* MT_BEST or MT_EXACT */ 99 return (strcmp(zn->zn_name_orij, matchname) == 0); 100 } 101} 102 103void 104zap_name_free(zap_name_t *zn) 105{ 106 kmem_free(zn, sizeof (zap_name_t)); 107} 108 109/* XXX combine this with zap_lockdir()? */ 110zap_name_t * 111zap_name_alloc(zap_t *zap, const char *name, matchtype_t mt) 112{ 113 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP); 114 115 zn->zn_zap = zap; 116 zn->zn_name_orij = name; 117 zn->zn_matchtype = mt; 118 if (zap->zap_normflags) { 119 if (zap_normalize(zap, name, zn->zn_normbuf) != 0) { 120 zap_name_free(zn); 121 return (NULL); 122 } 123 zn->zn_name_norm = zn->zn_normbuf; 124 } else { 125 if (mt != MT_EXACT) { 126 zap_name_free(zn); 127 return (NULL); 128 } 129 zn->zn_name_norm = zn->zn_name_orij; 130 } 131 132 zn->zn_hash = zap_hash(zap, zn->zn_name_norm); 133 return (zn); 134} 135 136static void 137mzap_byteswap(mzap_phys_t *buf, size_t size) 138{ 139 int i, max; 140 buf->mz_block_type = BSWAP_64(buf->mz_block_type); 141 buf->mz_salt = BSWAP_64(buf->mz_salt); 142 buf->mz_normflags = BSWAP_64(buf->mz_normflags); 143 max = (size / MZAP_ENT_LEN) - 1; 144 for (i = 0; i < max; i++) { 145 buf->mz_chunk[i].mze_value = 146 BSWAP_64(buf->mz_chunk[i].mze_value); 147 buf->mz_chunk[i].mze_cd = 148 BSWAP_32(buf->mz_chunk[i].mze_cd); 149 } 150} 151 152void 153zap_byteswap(void *buf, size_t size) 154{ 155 uint64_t block_type; 156 157 block_type = *(uint64_t *)buf; 158 159 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) { 160 /* ASSERT(magic == ZAP_LEAF_MAGIC); */ 161 mzap_byteswap(buf, size); 162 } else { 163 fzap_byteswap(buf, size); 164 } 165} 166 167static int 168mze_compare(const void *arg1, const void *arg2) 169{ 170 const mzap_ent_t *mze1 = arg1; 171 const mzap_ent_t *mze2 = arg2; 172 173 if (mze1->mze_hash > mze2->mze_hash) 174 return (+1); 175 if (mze1->mze_hash < mze2->mze_hash) 176 return (-1); 177 if (mze1->mze_phys.mze_cd > mze2->mze_phys.mze_cd) 178 return (+1); 179 if (mze1->mze_phys.mze_cd < mze2->mze_phys.mze_cd) 180 return (-1); 181 return (0); 182} 183 184static void 185mze_insert(zap_t *zap, int chunkid, uint64_t hash, mzap_ent_phys_t *mzep) 186{ 187 mzap_ent_t *mze; 188 189 ASSERT(zap->zap_ismicro); 190 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 191 ASSERT(mzep->mze_cd < ZAP_MAXCD); 192 193 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP); 194 mze->mze_chunkid = chunkid; 195 mze->mze_hash = hash; 196 mze->mze_phys = *mzep; 197 avl_add(&zap->zap_m.zap_avl, mze); 198} 199 200static mzap_ent_t * 201mze_find(zap_name_t *zn) 202{ 203 mzap_ent_t mze_tofind; 204 mzap_ent_t *mze; 205 avl_index_t idx; 206 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl; 207 208 ASSERT(zn->zn_zap->zap_ismicro); 209 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock)); 210 211 if (strlen(zn->zn_name_norm) >= sizeof (mze_tofind.mze_phys.mze_name)) 212 return (NULL); 213 214 mze_tofind.mze_hash = zn->zn_hash; 215 mze_tofind.mze_phys.mze_cd = 0; 216 217again: 218 mze = avl_find(avl, &mze_tofind, &idx); 219 if (mze == NULL) 220 mze = avl_nearest(avl, idx, AVL_AFTER); 221 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) { 222 if (zap_match(zn, mze->mze_phys.mze_name)) 223 return (mze); 224 } 225 if (zn->zn_matchtype == MT_BEST) { 226 zn->zn_matchtype = MT_FIRST; 227 goto again; 228 } 229 return (NULL); 230} 231 232static uint32_t 233mze_find_unused_cd(zap_t *zap, uint64_t hash) 234{ 235 mzap_ent_t mze_tofind; 236 mzap_ent_t *mze; 237 avl_index_t idx; 238 avl_tree_t *avl = &zap->zap_m.zap_avl; 239 uint32_t cd; 240 241 ASSERT(zap->zap_ismicro); 242 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 243 244 mze_tofind.mze_hash = hash; 245 mze_tofind.mze_phys.mze_cd = 0; 246 247 cd = 0; 248 for (mze = avl_find(avl, &mze_tofind, &idx); 249 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) { 250 if (mze->mze_phys.mze_cd != cd) 251 break; 252 cd++; 253 } 254 255 return (cd); 256} 257 258static void 259mze_remove(zap_t *zap, mzap_ent_t *mze) 260{ 261 ASSERT(zap->zap_ismicro); 262 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 263 264 avl_remove(&zap->zap_m.zap_avl, mze); 265 kmem_free(mze, sizeof (mzap_ent_t)); 266} 267 268static void 269mze_destroy(zap_t *zap) 270{ 271 mzap_ent_t *mze; 272 void *avlcookie = NULL; 273 274 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie)) 275 kmem_free(mze, sizeof (mzap_ent_t)); 276 avl_destroy(&zap->zap_m.zap_avl); 277} 278 279static zap_t * 280mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db) 281{ 282 zap_t *winner; 283 zap_t *zap; 284 int i; 285 286 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t)); 287 288 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP); 289 rw_init(&zap->zap_rwlock, NULL, RW_DEFAULT, 0); 290 rw_enter(&zap->zap_rwlock, RW_WRITER); 291 zap->zap_objset = os; 292 zap->zap_object = obj; 293 zap->zap_dbuf = db; 294 295 if (*(uint64_t *)db->db_data != ZBT_MICRO) { 296 mutex_init(&zap->zap_f.zap_num_entries_mtx, NULL, 297 MUTEX_DEFAULT, 0); 298 zap->zap_f.zap_block_shift = highbit(db->db_size) - 1; 299 } else { 300 zap->zap_ismicro = TRUE; 301 } 302 303 /* 304 * Make sure that zap_ismicro is set before we let others see 305 * it, because zap_lockdir() checks zap_ismicro without the lock 306 * held. 307 */ 308 winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict); 309 310 if (winner != NULL) { 311 rw_exit(&zap->zap_rwlock); 312 rw_destroy(&zap->zap_rwlock); 313 if (!zap->zap_ismicro) 314 mutex_destroy(&zap->zap_f.zap_num_entries_mtx); 315 kmem_free(zap, sizeof (zap_t)); 316 return (winner); 317 } 318 319 if (zap->zap_ismicro) { 320 zap->zap_salt = zap->zap_m.zap_phys->mz_salt; 321 zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags; 322 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1; 323 avl_create(&zap->zap_m.zap_avl, mze_compare, 324 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node)); 325 326 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) { 327 mzap_ent_phys_t *mze = 328 &zap->zap_m.zap_phys->mz_chunk[i]; 329 if (mze->mze_name[0]) { 330 zap_name_t *zn; 331 332 zap->zap_m.zap_num_entries++; 333 zn = zap_name_alloc(zap, mze->mze_name, 334 MT_EXACT); 335 mze_insert(zap, i, zn->zn_hash, mze); 336 zap_name_free(zn); 337 } 338 } 339 } else { 340 zap->zap_salt = zap->zap_f.zap_phys->zap_salt; 341 zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags; 342 343 ASSERT3U(sizeof (struct zap_leaf_header), ==, 344 2*ZAP_LEAF_CHUNKSIZE); 345 346 /* 347 * The embedded pointer table should not overlap the 348 * other members. 349 */ 350 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >, 351 &zap->zap_f.zap_phys->zap_salt); 352 353 /* 354 * The embedded pointer table should end at the end of 355 * the block 356 */ 357 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap, 358 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) - 359 (uintptr_t)zap->zap_f.zap_phys, ==, 360 zap->zap_dbuf->db_size); 361 } 362 rw_exit(&zap->zap_rwlock); 363 return (zap); 364} 365 366int 367zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx, 368 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp) 369{ 370 zap_t *zap; 371 dmu_buf_t *db; 372 krw_t lt; 373 int err; 374 375 *zapp = NULL; 376 377 err = dmu_buf_hold(os, obj, 0, NULL, &db); 378 if (err) 379 return (err); 380 381#ifdef ZFS_DEBUG 382 { 383 dmu_object_info_t doi; 384 dmu_object_info_from_db(db, &doi); 385 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap); 386 } 387#endif 388 389 zap = dmu_buf_get_user(db); 390 if (zap == NULL) 391 zap = mzap_open(os, obj, db); 392 393 /* 394 * We're checking zap_ismicro without the lock held, in order to 395 * tell what type of lock we want. Once we have some sort of 396 * lock, see if it really is the right type. In practice this 397 * can only be different if it was upgraded from micro to fat, 398 * and micro wanted WRITER but fat only needs READER. 399 */ 400 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti; 401 rw_enter(&zap->zap_rwlock, lt); 402 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) { 403 /* it was upgraded, now we only need reader */ 404 ASSERT(lt == RW_WRITER); 405 ASSERT(RW_READER == 406 (!zap->zap_ismicro && fatreader) ? RW_READER : lti); 407 rw_downgrade(&zap->zap_rwlock); 408 lt = RW_READER; 409 } 410 411 zap->zap_objset = os; 412 413 if (lt == RW_WRITER) 414 dmu_buf_will_dirty(db, tx); 415 416 ASSERT3P(zap->zap_dbuf, ==, db); 417 418 ASSERT(!zap->zap_ismicro || 419 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks); 420 if (zap->zap_ismicro && tx && adding && 421 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) { 422 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE; 423 if (newsz > MZAP_MAX_BLKSZ) { 424 dprintf("upgrading obj %llu: num_entries=%u\n", 425 obj, zap->zap_m.zap_num_entries); 426 *zapp = zap; 427 return (mzap_upgrade(zapp, tx)); 428 } 429 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx); 430 ASSERT3U(err, ==, 0); 431 zap->zap_m.zap_num_chunks = 432 db->db_size / MZAP_ENT_LEN - 1; 433 } 434 435 *zapp = zap; 436 return (0); 437} 438 439void 440zap_unlockdir(zap_t *zap) 441{ 442 rw_exit(&zap->zap_rwlock); 443 dmu_buf_rele(zap->zap_dbuf, NULL); 444} 445 446static int 447mzap_upgrade(zap_t **zapp, dmu_tx_t *tx) 448{ 449 mzap_phys_t *mzp; 450 int i, sz, nchunks, err; 451 zap_t *zap = *zapp; 452 453 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 454 455 sz = zap->zap_dbuf->db_size; 456 mzp = kmem_alloc(sz, KM_SLEEP); 457 bcopy(zap->zap_dbuf->db_data, mzp, sz); 458 nchunks = zap->zap_m.zap_num_chunks; 459 460 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object, 461 1ULL << fzap_default_block_shift, 0, tx); 462 if (err) { 463 kmem_free(mzp, sz); 464 return (err); 465 } 466 467 dprintf("upgrading obj=%llu with %u chunks\n", 468 zap->zap_object, nchunks); 469 /* XXX destroy the avl later, so we can use the stored hash value */ 470 mze_destroy(zap); 471 472 fzap_upgrade(zap, tx); 473 474 for (i = 0; i < nchunks; i++) { 475 int err; 476 mzap_ent_phys_t *mze = &mzp->mz_chunk[i]; 477 zap_name_t *zn; 478 if (mze->mze_name[0] == 0) 479 continue; 480 dprintf("adding %s=%llu\n", 481 mze->mze_name, mze->mze_value); 482 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT); 483 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx); 484 zap = zn->zn_zap; /* fzap_add_cd() may change zap */ 485 zap_name_free(zn); 486 if (err) 487 break; 488 } 489 kmem_free(mzp, sz); 490 *zapp = zap; 491 return (err); 492} 493 494static void 495mzap_create_impl(objset_t *os, uint64_t obj, int normflags, dmu_tx_t *tx) 496{ 497 dmu_buf_t *db; 498 mzap_phys_t *zp; 499 500 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db)); 501 502#ifdef ZFS_DEBUG 503 { 504 dmu_object_info_t doi; 505 dmu_object_info_from_db(db, &doi); 506 ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap); 507 } 508#endif 509 510 dmu_buf_will_dirty(db, tx); 511 zp = db->db_data; 512 zp->mz_block_type = ZBT_MICRO; 513 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL; 514 zp->mz_normflags = normflags; 515 dmu_buf_rele(db, FTAG); 516} 517 518int 519zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot, 520 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 521{ 522 return (zap_create_claim_norm(os, obj, 523 0, ot, bonustype, bonuslen, tx)); 524} 525 526int 527zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags, 528 dmu_object_type_t ot, 529 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 530{ 531 int err; 532 533 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx); 534 if (err != 0) 535 return (err); 536 mzap_create_impl(os, obj, normflags, tx); 537 return (0); 538} 539 540uint64_t 541zap_create(objset_t *os, dmu_object_type_t ot, 542 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 543{ 544 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx)); 545} 546 547uint64_t 548zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot, 549 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 550{ 551 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx); 552 553 mzap_create_impl(os, obj, normflags, tx); 554 return (obj); 555} 556 557int 558zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx) 559{ 560 /* 561 * dmu_object_free will free the object number and free the 562 * data. Freeing the data will cause our pageout function to be 563 * called, which will destroy our data (zap_leaf_t's and zap_t). 564 */ 565 566 return (dmu_object_free(os, zapobj, tx)); 567} 568 569_NOTE(ARGSUSED(0)) 570void 571zap_evict(dmu_buf_t *db, void *vzap) 572{ 573 zap_t *zap = vzap; 574 575 rw_destroy(&zap->zap_rwlock); 576 577 if (zap->zap_ismicro) 578 mze_destroy(zap); 579 else 580 mutex_destroy(&zap->zap_f.zap_num_entries_mtx); 581 582 kmem_free(zap, sizeof (zap_t)); 583} 584 585int 586zap_count(objset_t *os, uint64_t zapobj, uint64_t *count) 587{ 588 zap_t *zap; 589 int err; 590 591 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 592 if (err) 593 return (err); 594 if (!zap->zap_ismicro) { 595 err = fzap_count(zap, count); 596 } else { 597 *count = zap->zap_m.zap_num_entries; 598 } 599 zap_unlockdir(zap); 600 return (err); 601} 602 603/* 604 * zn may be NULL; if not specified, it will be computed if needed. 605 * See also the comment above zap_entry_normalization_conflict(). 606 */ 607static boolean_t 608mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze) 609{ 610 mzap_ent_t *other; 611 int direction = AVL_BEFORE; 612 boolean_t allocdzn = B_FALSE; 613 614 if (zap->zap_normflags == 0) 615 return (B_FALSE); 616 617again: 618 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction); 619 other && other->mze_hash == mze->mze_hash; 620 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) { 621 622 if (zn == NULL) { 623 zn = zap_name_alloc(zap, mze->mze_phys.mze_name, 624 MT_FIRST); 625 allocdzn = B_TRUE; 626 } 627 if (zap_match(zn, other->mze_phys.mze_name)) { 628 if (allocdzn) 629 zap_name_free(zn); 630 return (B_TRUE); 631 } 632 } 633 634 if (direction == AVL_BEFORE) { 635 direction = AVL_AFTER; 636 goto again; 637 } 638 639 if (allocdzn) 640 zap_name_free(zn); 641 return (B_FALSE); 642} 643 644/* 645 * Routines for manipulating attributes. 646 */ 647 648int 649zap_lookup(objset_t *os, uint64_t zapobj, const char *name, 650 uint64_t integer_size, uint64_t num_integers, void *buf) 651{ 652 return (zap_lookup_norm(os, zapobj, name, integer_size, 653 num_integers, buf, MT_EXACT, NULL, 0, NULL)); 654} 655 656int 657zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name, 658 uint64_t integer_size, uint64_t num_integers, void *buf, 659 matchtype_t mt, char *realname, int rn_len, 660 boolean_t *ncp) 661{ 662 zap_t *zap; 663 int err; 664 mzap_ent_t *mze; 665 zap_name_t *zn; 666 667 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 668 if (err) 669 return (err); 670 zn = zap_name_alloc(zap, name, mt); 671 if (zn == NULL) { 672 zap_unlockdir(zap); 673 return (ENOTSUP); 674 } 675 676 if (!zap->zap_ismicro) { 677 err = fzap_lookup(zn, integer_size, num_integers, buf, 678 realname, rn_len, ncp); 679 } else { 680 mze = mze_find(zn); 681 if (mze == NULL) { 682 err = ENOENT; 683 } else { 684 if (num_integers < 1) { 685 err = EOVERFLOW; 686 } else if (integer_size != 8) { 687 err = EINVAL; 688 } else { 689 *(uint64_t *)buf = mze->mze_phys.mze_value; 690 (void) strlcpy(realname, 691 mze->mze_phys.mze_name, rn_len); 692 if (ncp) { 693 *ncp = mzap_normalization_conflict(zap, 694 zn, mze); 695 } 696 } 697 } 698 } 699 zap_name_free(zn); 700 zap_unlockdir(zap); 701 return (err); 702} 703 704int 705zap_length(objset_t *os, uint64_t zapobj, const char *name, 706 uint64_t *integer_size, uint64_t *num_integers) 707{ 708 zap_t *zap; 709 int err; 710 mzap_ent_t *mze; 711 zap_name_t *zn; 712 713 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 714 if (err) 715 return (err); 716 zn = zap_name_alloc(zap, name, MT_EXACT); 717 if (zn == NULL) { 718 zap_unlockdir(zap); 719 return (ENOTSUP); 720 } 721 if (!zap->zap_ismicro) { 722 err = fzap_length(zn, integer_size, num_integers); 723 } else { 724 mze = mze_find(zn); 725 if (mze == NULL) { 726 err = ENOENT; 727 } else { 728 if (integer_size) 729 *integer_size = 8; 730 if (num_integers) 731 *num_integers = 1; 732 } 733 } 734 zap_name_free(zn); 735 zap_unlockdir(zap); 736 return (err); 737} 738 739static void 740mzap_addent(zap_name_t *zn, uint64_t value) 741{ 742 int i; 743 zap_t *zap = zn->zn_zap; 744 int start = zap->zap_m.zap_alloc_next; 745 uint32_t cd; 746 747 dprintf("obj=%llu %s=%llu\n", zap->zap_object, 748 zn->zn_name_orij, value); 749 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 750 751#ifdef ZFS_DEBUG 752 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) { 753 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i]; 754 ASSERT(strcmp(zn->zn_name_orij, mze->mze_name) != 0); 755 } 756#endif 757 758 cd = mze_find_unused_cd(zap, zn->zn_hash); 759 /* given the limited size of the microzap, this can't happen */ 760 ASSERT(cd != ZAP_MAXCD); 761 762again: 763 for (i = start; i < zap->zap_m.zap_num_chunks; i++) { 764 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i]; 765 if (mze->mze_name[0] == 0) { 766 mze->mze_value = value; 767 mze->mze_cd = cd; 768 (void) strcpy(mze->mze_name, zn->zn_name_orij); 769 zap->zap_m.zap_num_entries++; 770 zap->zap_m.zap_alloc_next = i+1; 771 if (zap->zap_m.zap_alloc_next == 772 zap->zap_m.zap_num_chunks) 773 zap->zap_m.zap_alloc_next = 0; 774 mze_insert(zap, i, zn->zn_hash, mze); 775 return; 776 } 777 } 778 if (start != 0) { 779 start = 0; 780 goto again; 781 } 782 ASSERT(!"out of entries!"); 783} 784 785int 786zap_add(objset_t *os, uint64_t zapobj, const char *name, 787 int integer_size, uint64_t num_integers, 788 const void *val, dmu_tx_t *tx) 789{ 790 zap_t *zap; 791 int err; 792 mzap_ent_t *mze; 793 const uint64_t *intval = val; 794 zap_name_t *zn; 795 796 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap); 797 if (err) 798 return (err); 799 zn = zap_name_alloc(zap, name, MT_EXACT); 800 if (zn == NULL) { 801 zap_unlockdir(zap); 802 return (ENOTSUP); 803 } 804 if (!zap->zap_ismicro) { 805 err = fzap_add(zn, integer_size, num_integers, val, tx); 806 zap = zn->zn_zap; /* fzap_add() may change zap */ 807 } else if (integer_size != 8 || num_integers != 1 || 808 strlen(name) >= MZAP_NAME_LEN) { 809 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n", 810 zapobj, integer_size, num_integers, name); 811 err = mzap_upgrade(&zn->zn_zap, tx); 812 if (err == 0) 813 err = fzap_add(zn, integer_size, num_integers, val, tx); 814 zap = zn->zn_zap; /* fzap_add() may change zap */ 815 } else { 816 mze = mze_find(zn); 817 if (mze != NULL) { 818 err = EEXIST; 819 } else { 820 mzap_addent(zn, *intval); 821 } 822 } 823 ASSERT(zap == zn->zn_zap); 824 zap_name_free(zn); 825 if (zap != NULL) /* may be NULL if fzap_add() failed */ 826 zap_unlockdir(zap); 827 return (err); 828} 829 830int 831zap_update(objset_t *os, uint64_t zapobj, const char *name, 832 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx) 833{ 834 zap_t *zap; 835 mzap_ent_t *mze; 836 const uint64_t *intval = val; 837 zap_name_t *zn; 838 int err; 839 840 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap); 841 if (err) 842 return (err); 843 zn = zap_name_alloc(zap, name, MT_EXACT); 844 if (zn == NULL) { 845 zap_unlockdir(zap); 846 return (ENOTSUP); 847 } 848 if (!zap->zap_ismicro) { 849 err = fzap_update(zn, integer_size, num_integers, val, tx); 850 zap = zn->zn_zap; /* fzap_update() may change zap */ 851 } else if (integer_size != 8 || num_integers != 1 || 852 strlen(name) >= MZAP_NAME_LEN) { 853 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n", 854 zapobj, integer_size, num_integers, name); 855 err = mzap_upgrade(&zn->zn_zap, tx); 856 if (err == 0) 857 err = fzap_update(zn, integer_size, num_integers, 858 val, tx); 859 zap = zn->zn_zap; /* fzap_update() may change zap */ 860 } else { 861 mze = mze_find(zn); 862 if (mze != NULL) { 863 mze->mze_phys.mze_value = *intval; 864 zap->zap_m.zap_phys->mz_chunk 865 [mze->mze_chunkid].mze_value = *intval; 866 } else { 867 mzap_addent(zn, *intval); 868 } 869 } 870 ASSERT(zap == zn->zn_zap); 871 zap_name_free(zn); 872 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */ 873 zap_unlockdir(zap); 874 return (err); 875} 876 877int 878zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx) 879{ 880 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx)); 881} 882 883int 884zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name, 885 matchtype_t mt, dmu_tx_t *tx) 886{ 887 zap_t *zap; 888 int err; 889 mzap_ent_t *mze; 890 zap_name_t *zn; 891 892 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap); 893 if (err) 894 return (err); 895 zn = zap_name_alloc(zap, name, mt); 896 if (zn == NULL) { 897 zap_unlockdir(zap); 898 return (ENOTSUP); 899 } 900 if (!zap->zap_ismicro) { 901 err = fzap_remove(zn, tx); 902 } else { 903 mze = mze_find(zn); 904 if (mze == NULL) { 905 err = ENOENT; 906 } else { 907 zap->zap_m.zap_num_entries--; 908 bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid], 909 sizeof (mzap_ent_phys_t)); 910 mze_remove(zap, mze); 911 } 912 } 913 zap_name_free(zn); 914 zap_unlockdir(zap); 915 return (err); 916} 917 918/* 919 * Routines for iterating over the attributes. 920 */ 921 922/* 923 * We want to keep the high 32 bits of the cursor zero if we can, so 924 * that 32-bit programs can access this. So use a small hash value so 925 * we can fit 4 bits of cd into the 32-bit cursor. 926 * 927 * [ 4 zero bits | 32-bit collision differentiator | 28-bit hash value ] 928 */ 929void 930zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj, 931 uint64_t serialized) 932{ 933 zc->zc_objset = os; 934 zc->zc_zap = NULL; 935 zc->zc_leaf = NULL; 936 zc->zc_zapobj = zapobj; 937 if (serialized == -1ULL) { 938 zc->zc_hash = -1ULL; 939 zc->zc_cd = 0; 940 } else { 941 zc->zc_hash = serialized << (64-ZAP_HASHBITS); 942 zc->zc_cd = serialized >> ZAP_HASHBITS; 943 if (zc->zc_cd >= ZAP_MAXCD) /* corrupt serialized */ 944 zc->zc_cd = 0; 945 } 946} 947 948void 949zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj) 950{ 951 zap_cursor_init_serialized(zc, os, zapobj, 0); 952} 953 954void 955zap_cursor_fini(zap_cursor_t *zc) 956{ 957 if (zc->zc_zap) { 958 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 959 zap_unlockdir(zc->zc_zap); 960 zc->zc_zap = NULL; 961 } 962 if (zc->zc_leaf) { 963 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 964 zap_put_leaf(zc->zc_leaf); 965 zc->zc_leaf = NULL; 966 } 967 zc->zc_objset = NULL; 968} 969 970uint64_t 971zap_cursor_serialize(zap_cursor_t *zc) 972{ 973 if (zc->zc_hash == -1ULL) 974 return (-1ULL); 975 ASSERT((zc->zc_hash & (ZAP_MAXCD-1)) == 0); 976 ASSERT(zc->zc_cd < ZAP_MAXCD); 977 return ((zc->zc_hash >> (64-ZAP_HASHBITS)) | 978 ((uint64_t)zc->zc_cd << ZAP_HASHBITS)); 979} 980 981int 982zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za) 983{ 984 int err; 985 avl_index_t idx; 986 mzap_ent_t mze_tofind; 987 mzap_ent_t *mze; 988 989 if (zc->zc_hash == -1ULL) 990 return (ENOENT); 991 992 if (zc->zc_zap == NULL) { 993 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL, 994 RW_READER, TRUE, FALSE, &zc->zc_zap); 995 if (err) 996 return (err); 997 } else { 998 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER); 999 } 1000 if (!zc->zc_zap->zap_ismicro) { 1001 err = fzap_cursor_retrieve(zc->zc_zap, zc, za); 1002 } else { 1003 err = ENOENT; 1004 1005 mze_tofind.mze_hash = zc->zc_hash; 1006 mze_tofind.mze_phys.mze_cd = zc->zc_cd; 1007 1008 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx); 1009 if (mze == NULL) { 1010 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl, 1011 idx, AVL_AFTER); 1012 } 1013 if (mze) { 1014 ASSERT(0 == bcmp(&mze->mze_phys, 1015 &zc->zc_zap->zap_m.zap_phys->mz_chunk 1016 [mze->mze_chunkid], sizeof (mze->mze_phys))); 1017 1018 za->za_normalization_conflict = 1019 mzap_normalization_conflict(zc->zc_zap, NULL, mze); 1020 za->za_integer_length = 8; 1021 za->za_num_integers = 1; 1022 za->za_first_integer = mze->mze_phys.mze_value; 1023 (void) strcpy(za->za_name, mze->mze_phys.mze_name); 1024 zc->zc_hash = mze->mze_hash; 1025 zc->zc_cd = mze->mze_phys.mze_cd; 1026 err = 0; 1027 } else { 1028 zc->zc_hash = -1ULL; 1029 } 1030 } 1031 rw_exit(&zc->zc_zap->zap_rwlock); 1032 return (err); 1033} 1034 1035void 1036zap_cursor_advance(zap_cursor_t *zc) 1037{ 1038 if (zc->zc_hash == -1ULL) 1039 return; 1040 zc->zc_cd++; 1041 if (zc->zc_cd >= ZAP_MAXCD) { 1042 zc->zc_cd = 0; 1043 zc->zc_hash += 1ULL<<(64-ZAP_HASHBITS); 1044 if (zc->zc_hash == 0) /* EOF */ 1045 zc->zc_hash = -1ULL; 1046 } 1047} 1048 1049int 1050zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs) 1051{ 1052 int err; 1053 zap_t *zap; 1054 1055 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap); 1056 if (err) 1057 return (err); 1058 1059 bzero(zs, sizeof (zap_stats_t)); 1060 1061 if (zap->zap_ismicro) { 1062 zs->zs_blocksize = zap->zap_dbuf->db_size; 1063 zs->zs_num_entries = zap->zap_m.zap_num_entries; 1064 zs->zs_num_blocks = 1; 1065 } else { 1066 fzap_get_stats(zap, zs); 1067 } 1068 zap_unlockdir(zap); 1069 return (0); 1070} 1071