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 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* 27 * This file contains the top half of the zfs directory structure 28 * implementation. The bottom half is in zap_leaf.c. 29 * 30 * The zdir is an extendable hash data structure. There is a table of 31 * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are 32 * each a constant size and hold a variable number of directory entries. 33 * The buckets (aka "leaf nodes") are implemented in zap_leaf.c. 34 * 35 * The pointer table holds a power of 2 number of pointers. 36 * (1<<zap_t->zd_data->zd_phys->zd_prefix_len). The bucket pointed to 37 * by the pointer at index i in the table holds entries whose hash value 38 * has a zd_prefix_len - bit prefix 39 */ 40 41 #include <sys/spa.h> 42 #include <sys/dmu.h> 43 #include <sys/zfs_context.h> 44 #include <sys/zfs_znode.h> 45 #include <sys/fs/zfs.h> 46 #include <sys/zap.h> 47 #include <sys/refcount.h> 48 #include <sys/zap_impl.h> 49 #include <sys/zap_leaf.h> 50 51 int fzap_default_block_shift = 14; /* 16k blocksize */ 52 53 static void zap_leaf_pageout(dmu_buf_t *db, void *vl); 54 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks); 55 56 57 void 58 fzap_byteswap(void *vbuf, size_t size) 59 { 60 uint64_t block_type; 61 62 block_type = *(uint64_t *)vbuf; 63 64 if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF)) 65 zap_leaf_byteswap(vbuf, size); 66 else { 67 /* it's a ptrtbl block */ 68 byteswap_uint64_array(vbuf, size); 69 } 70 } 71 72 void 73 fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags) 74 { 75 dmu_buf_t *db; 76 zap_leaf_t *l; 77 int i; 78 zap_phys_t *zp; 79 80 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 81 zap->zap_ismicro = FALSE; 82 83 (void) dmu_buf_update_user(zap->zap_dbuf, zap, zap, 84 &zap->zap_f.zap_phys, zap_evict); 85 86 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0); 87 zap->zap_f.zap_block_shift = highbit(zap->zap_dbuf->db_size) - 1; 88 89 zp = zap->zap_f.zap_phys; 90 /* 91 * explicitly zero it since it might be coming from an 92 * initialized microzap 93 */ 94 bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size); 95 zp->zap_block_type = ZBT_HEADER; 96 zp->zap_magic = ZAP_MAGIC; 97 98 zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap); 99 100 zp->zap_freeblk = 2; /* block 1 will be the first leaf */ 101 zp->zap_num_leafs = 1; 102 zp->zap_num_entries = 0; 103 zp->zap_salt = zap->zap_salt; 104 zp->zap_normflags = zap->zap_normflags; 105 zp->zap_flags = flags; 106 107 /* block 1 will be the first leaf */ 108 for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++) 109 ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1; 110 111 /* 112 * set up block 1 - the first leaf 113 */ 114 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 115 1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db)); 116 dmu_buf_will_dirty(db, tx); 117 118 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); 119 l->l_dbuf = db; 120 l->l_phys = db->db_data; 121 122 zap_leaf_init(l, zp->zap_normflags != 0); 123 124 kmem_free(l, sizeof (zap_leaf_t)); 125 dmu_buf_rele(db, FTAG); 126 } 127 128 static int 129 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx) 130 { 131 if (RW_WRITE_HELD(&zap->zap_rwlock)) 132 return (1); 133 if (rw_tryupgrade(&zap->zap_rwlock)) { 134 dmu_buf_will_dirty(zap->zap_dbuf, tx); 135 return (1); 136 } 137 return (0); 138 } 139 140 /* 141 * Generic routines for dealing with the pointer & cookie tables. 142 */ 143 144 static int 145 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl, 146 void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n), 147 dmu_tx_t *tx) 148 { 149 uint64_t b, newblk; 150 dmu_buf_t *db_old, *db_new; 151 int err; 152 int bs = FZAP_BLOCK_SHIFT(zap); 153 int hepb = 1<<(bs-4); 154 /* hepb = half the number of entries in a block */ 155 156 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 157 ASSERT(tbl->zt_blk != 0); 158 ASSERT(tbl->zt_numblks > 0); 159 160 if (tbl->zt_nextblk != 0) { 161 newblk = tbl->zt_nextblk; 162 } else { 163 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2); 164 tbl->zt_nextblk = newblk; 165 ASSERT3U(tbl->zt_blks_copied, ==, 0); 166 dmu_prefetch(zap->zap_objset, zap->zap_object, 167 tbl->zt_blk << bs, tbl->zt_numblks << bs); 168 } 169 170 /* 171 * Copy the ptrtbl from the old to new location. 172 */ 173 174 b = tbl->zt_blks_copied; 175 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 176 (tbl->zt_blk + b) << bs, FTAG, &db_old); 177 if (err) 178 return (err); 179 180 /* first half of entries in old[b] go to new[2*b+0] */ 181 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 182 (newblk + 2*b+0) << bs, FTAG, &db_new)); 183 dmu_buf_will_dirty(db_new, tx); 184 transfer_func(db_old->db_data, db_new->db_data, hepb); 185 dmu_buf_rele(db_new, FTAG); 186 187 /* second half of entries in old[b] go to new[2*b+1] */ 188 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 189 (newblk + 2*b+1) << bs, FTAG, &db_new)); 190 dmu_buf_will_dirty(db_new, tx); 191 transfer_func((uint64_t *)db_old->db_data + hepb, 192 db_new->db_data, hepb); 193 dmu_buf_rele(db_new, FTAG); 194 195 dmu_buf_rele(db_old, FTAG); 196 197 tbl->zt_blks_copied++; 198 199 dprintf("copied block %llu of %llu\n", 200 tbl->zt_blks_copied, tbl->zt_numblks); 201 202 if (tbl->zt_blks_copied == tbl->zt_numblks) { 203 (void) dmu_free_range(zap->zap_objset, zap->zap_object, 204 tbl->zt_blk << bs, tbl->zt_numblks << bs, tx); 205 206 tbl->zt_blk = newblk; 207 tbl->zt_numblks *= 2; 208 tbl->zt_shift++; 209 tbl->zt_nextblk = 0; 210 tbl->zt_blks_copied = 0; 211 212 dprintf("finished; numblocks now %llu (%lluk entries)\n", 213 tbl->zt_numblks, 1<<(tbl->zt_shift-10)); 214 } 215 216 return (0); 217 } 218 219 static int 220 zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val, 221 dmu_tx_t *tx) 222 { 223 int err; 224 uint64_t blk, off; 225 int bs = FZAP_BLOCK_SHIFT(zap); 226 dmu_buf_t *db; 227 228 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 229 ASSERT(tbl->zt_blk != 0); 230 231 dprintf("storing %llx at index %llx\n", val, idx); 232 233 blk = idx >> (bs-3); 234 off = idx & ((1<<(bs-3))-1); 235 236 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 237 (tbl->zt_blk + blk) << bs, FTAG, &db); 238 if (err) 239 return (err); 240 dmu_buf_will_dirty(db, tx); 241 242 if (tbl->zt_nextblk != 0) { 243 uint64_t idx2 = idx * 2; 244 uint64_t blk2 = idx2 >> (bs-3); 245 uint64_t off2 = idx2 & ((1<<(bs-3))-1); 246 dmu_buf_t *db2; 247 248 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 249 (tbl->zt_nextblk + blk2) << bs, FTAG, &db2); 250 if (err) { 251 dmu_buf_rele(db, FTAG); 252 return (err); 253 } 254 dmu_buf_will_dirty(db2, tx); 255 ((uint64_t *)db2->db_data)[off2] = val; 256 ((uint64_t *)db2->db_data)[off2+1] = val; 257 dmu_buf_rele(db2, FTAG); 258 } 259 260 ((uint64_t *)db->db_data)[off] = val; 261 dmu_buf_rele(db, FTAG); 262 263 return (0); 264 } 265 266 static int 267 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp) 268 { 269 uint64_t blk, off; 270 int err; 271 dmu_buf_t *db; 272 int bs = FZAP_BLOCK_SHIFT(zap); 273 274 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 275 276 blk = idx >> (bs-3); 277 off = idx & ((1<<(bs-3))-1); 278 279 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 280 (tbl->zt_blk + blk) << bs, FTAG, &db); 281 if (err) 282 return (err); 283 *valp = ((uint64_t *)db->db_data)[off]; 284 dmu_buf_rele(db, FTAG); 285 286 if (tbl->zt_nextblk != 0) { 287 /* 288 * read the nextblk for the sake of i/o error checking, 289 * so that zap_table_load() will catch errors for 290 * zap_table_store. 291 */ 292 blk = (idx*2) >> (bs-3); 293 294 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 295 (tbl->zt_nextblk + blk) << bs, FTAG, &db); 296 dmu_buf_rele(db, FTAG); 297 } 298 return (err); 299 } 300 301 /* 302 * Routines for growing the ptrtbl. 303 */ 304 305 static void 306 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n) 307 { 308 int i; 309 for (i = 0; i < n; i++) { 310 uint64_t lb = src[i]; 311 dst[2*i+0] = lb; 312 dst[2*i+1] = lb; 313 } 314 } 315 316 static int 317 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx) 318 { 319 /* 320 * The pointer table should never use more hash bits than we 321 * have (otherwise we'd be using useless zero bits to index it). 322 * If we are within 2 bits of running out, stop growing, since 323 * this is already an aberrant condition. 324 */ 325 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2) 326 return (ENOSPC); 327 328 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) { 329 /* 330 * We are outgrowing the "embedded" ptrtbl (the one 331 * stored in the header block). Give it its own entire 332 * block, which will double the size of the ptrtbl. 333 */ 334 uint64_t newblk; 335 dmu_buf_t *db_new; 336 int err; 337 338 ASSERT3U(zap->zap_f.zap_phys->zap_ptrtbl.zt_shift, ==, 339 ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); 340 ASSERT3U(zap->zap_f.zap_phys->zap_ptrtbl.zt_blk, ==, 0); 341 342 newblk = zap_allocate_blocks(zap, 1); 343 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 344 newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new); 345 if (err) 346 return (err); 347 dmu_buf_will_dirty(db_new, tx); 348 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), 349 db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); 350 dmu_buf_rele(db_new, FTAG); 351 352 zap->zap_f.zap_phys->zap_ptrtbl.zt_blk = newblk; 353 zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks = 1; 354 zap->zap_f.zap_phys->zap_ptrtbl.zt_shift++; 355 356 ASSERT3U(1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift, ==, 357 zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks << 358 (FZAP_BLOCK_SHIFT(zap)-3)); 359 360 return (0); 361 } else { 362 return (zap_table_grow(zap, &zap->zap_f.zap_phys->zap_ptrtbl, 363 zap_ptrtbl_transfer, tx)); 364 } 365 } 366 367 static void 368 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx) 369 { 370 dmu_buf_will_dirty(zap->zap_dbuf, tx); 371 mutex_enter(&zap->zap_f.zap_num_entries_mtx); 372 ASSERT(delta > 0 || zap->zap_f.zap_phys->zap_num_entries >= -delta); 373 zap->zap_f.zap_phys->zap_num_entries += delta; 374 mutex_exit(&zap->zap_f.zap_num_entries_mtx); 375 } 376 377 static uint64_t 378 zap_allocate_blocks(zap_t *zap, int nblocks) 379 { 380 uint64_t newblk; 381 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 382 newblk = zap->zap_f.zap_phys->zap_freeblk; 383 zap->zap_f.zap_phys->zap_freeblk += nblocks; 384 return (newblk); 385 } 386 387 static zap_leaf_t * 388 zap_create_leaf(zap_t *zap, dmu_tx_t *tx) 389 { 390 void *winner; 391 zap_leaf_t *l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP); 392 393 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 394 395 rw_init(&l->l_rwlock, 0, 0, 0); 396 rw_enter(&l->l_rwlock, RW_WRITER); 397 l->l_blkid = zap_allocate_blocks(zap, 1); 398 l->l_dbuf = NULL; 399 l->l_phys = NULL; 400 401 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 402 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf)); 403 winner = dmu_buf_set_user(l->l_dbuf, l, &l->l_phys, zap_leaf_pageout); 404 ASSERT(winner == NULL); 405 dmu_buf_will_dirty(l->l_dbuf, tx); 406 407 zap_leaf_init(l, zap->zap_normflags != 0); 408 409 zap->zap_f.zap_phys->zap_num_leafs++; 410 411 return (l); 412 } 413 414 int 415 fzap_count(zap_t *zap, uint64_t *count) 416 { 417 ASSERT(!zap->zap_ismicro); 418 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */ 419 *count = zap->zap_f.zap_phys->zap_num_entries; 420 mutex_exit(&zap->zap_f.zap_num_entries_mtx); 421 return (0); 422 } 423 424 /* 425 * Routines for obtaining zap_leaf_t's 426 */ 427 428 void 429 zap_put_leaf(zap_leaf_t *l) 430 { 431 rw_exit(&l->l_rwlock); 432 dmu_buf_rele(l->l_dbuf, NULL); 433 } 434 435 _NOTE(ARGSUSED(0)) 436 static void 437 zap_leaf_pageout(dmu_buf_t *db, void *vl) 438 { 439 zap_leaf_t *l = vl; 440 441 rw_destroy(&l->l_rwlock); 442 kmem_free(l, sizeof (zap_leaf_t)); 443 } 444 445 static zap_leaf_t * 446 zap_open_leaf(uint64_t blkid, dmu_buf_t *db) 447 { 448 zap_leaf_t *l, *winner; 449 450 ASSERT(blkid != 0); 451 452 l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP); 453 rw_init(&l->l_rwlock, 0, 0, 0); 454 rw_enter(&l->l_rwlock, RW_WRITER); 455 l->l_blkid = blkid; 456 l->l_bs = highbit(db->db_size)-1; 457 l->l_dbuf = db; 458 l->l_phys = NULL; 459 460 winner = dmu_buf_set_user(db, l, &l->l_phys, zap_leaf_pageout); 461 462 rw_exit(&l->l_rwlock); 463 if (winner != NULL) { 464 /* someone else set it first */ 465 zap_leaf_pageout(NULL, l); 466 l = winner; 467 } 468 469 /* 470 * lhr_pad was previously used for the next leaf in the leaf 471 * chain. There should be no chained leafs (as we have removed 472 * support for them). 473 */ 474 ASSERT3U(l->l_phys->l_hdr.lh_pad1, ==, 0); 475 476 /* 477 * There should be more hash entries than there can be 478 * chunks to put in the hash table 479 */ 480 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3); 481 482 /* The chunks should begin at the end of the hash table */ 483 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==, 484 &l->l_phys->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]); 485 486 /* The chunks should end at the end of the block */ 487 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) - 488 (uintptr_t)l->l_phys, ==, l->l_dbuf->db_size); 489 490 return (l); 491 } 492 493 static int 494 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt, 495 zap_leaf_t **lp) 496 { 497 dmu_buf_t *db; 498 zap_leaf_t *l; 499 int bs = FZAP_BLOCK_SHIFT(zap); 500 int err; 501 502 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 503 504 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 505 blkid << bs, NULL, &db); 506 if (err) 507 return (err); 508 509 ASSERT3U(db->db_object, ==, zap->zap_object); 510 ASSERT3U(db->db_offset, ==, blkid << bs); 511 ASSERT3U(db->db_size, ==, 1 << bs); 512 ASSERT(blkid != 0); 513 514 l = dmu_buf_get_user(db); 515 516 if (l == NULL) 517 l = zap_open_leaf(blkid, db); 518 519 rw_enter(&l->l_rwlock, lt); 520 /* 521 * Must lock before dirtying, otherwise l->l_phys could change, 522 * causing ASSERT below to fail. 523 */ 524 if (lt == RW_WRITER) 525 dmu_buf_will_dirty(db, tx); 526 ASSERT3U(l->l_blkid, ==, blkid); 527 ASSERT3P(l->l_dbuf, ==, db); 528 ASSERT3P(l->l_phys, ==, l->l_dbuf->db_data); 529 ASSERT3U(l->l_phys->l_hdr.lh_block_type, ==, ZBT_LEAF); 530 ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC); 531 532 *lp = l; 533 return (0); 534 } 535 536 static int 537 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp) 538 { 539 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 540 541 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) { 542 ASSERT3U(idx, <, 543 (1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift)); 544 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx); 545 return (0); 546 } else { 547 return (zap_table_load(zap, &zap->zap_f.zap_phys->zap_ptrtbl, 548 idx, valp)); 549 } 550 } 551 552 static int 553 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx) 554 { 555 ASSERT(tx != NULL); 556 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 557 558 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk == 0) { 559 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk; 560 return (0); 561 } else { 562 return (zap_table_store(zap, &zap->zap_f.zap_phys->zap_ptrtbl, 563 idx, blk, tx)); 564 } 565 } 566 567 static int 568 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp) 569 { 570 uint64_t idx, blk; 571 int err; 572 573 ASSERT(zap->zap_dbuf == NULL || 574 zap->zap_f.zap_phys == zap->zap_dbuf->db_data); 575 ASSERT3U(zap->zap_f.zap_phys->zap_magic, ==, ZAP_MAGIC); 576 idx = ZAP_HASH_IDX(h, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift); 577 err = zap_idx_to_blk(zap, idx, &blk); 578 if (err != 0) 579 return (err); 580 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp); 581 582 ASSERT(err || ZAP_HASH_IDX(h, (*lp)->l_phys->l_hdr.lh_prefix_len) == 583 (*lp)->l_phys->l_hdr.lh_prefix); 584 return (err); 585 } 586 587 static int 588 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx, zap_leaf_t **lp) 589 { 590 zap_t *zap = zn->zn_zap; 591 uint64_t hash = zn->zn_hash; 592 zap_leaf_t *nl; 593 int prefix_diff, i, err; 594 uint64_t sibling; 595 int old_prefix_len = l->l_phys->l_hdr.lh_prefix_len; 596 597 ASSERT3U(old_prefix_len, <=, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift); 598 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 599 600 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, 601 l->l_phys->l_hdr.lh_prefix); 602 603 if (zap_tryupgradedir(zap, tx) == 0 || 604 old_prefix_len == zap->zap_f.zap_phys->zap_ptrtbl.zt_shift) { 605 /* We failed to upgrade, or need to grow the pointer table */ 606 objset_t *os = zap->zap_objset; 607 uint64_t object = zap->zap_object; 608 609 zap_put_leaf(l); 610 zap_unlockdir(zap); 611 err = zap_lockdir(os, object, tx, RW_WRITER, 612 FALSE, FALSE, &zn->zn_zap); 613 zap = zn->zn_zap; 614 if (err) 615 return (err); 616 ASSERT(!zap->zap_ismicro); 617 618 while (old_prefix_len == 619 zap->zap_f.zap_phys->zap_ptrtbl.zt_shift) { 620 err = zap_grow_ptrtbl(zap, tx); 621 if (err) 622 return (err); 623 } 624 625 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l); 626 if (err) 627 return (err); 628 629 if (l->l_phys->l_hdr.lh_prefix_len != old_prefix_len) { 630 /* it split while our locks were down */ 631 *lp = l; 632 return (0); 633 } 634 } 635 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 636 ASSERT3U(old_prefix_len, <, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift); 637 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, 638 l->l_phys->l_hdr.lh_prefix); 639 640 prefix_diff = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift - 641 (old_prefix_len + 1); 642 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff; 643 644 /* check for i/o errors before doing zap_leaf_split */ 645 for (i = 0; i < (1ULL<<prefix_diff); i++) { 646 uint64_t blk; 647 err = zap_idx_to_blk(zap, sibling+i, &blk); 648 if (err) 649 return (err); 650 ASSERT3U(blk, ==, l->l_blkid); 651 } 652 653 nl = zap_create_leaf(zap, tx); 654 zap_leaf_split(l, nl, zap->zap_normflags != 0); 655 656 /* set sibling pointers */ 657 for (i = 0; i < (1ULL<<prefix_diff); i++) { 658 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx); 659 ASSERT3U(err, ==, 0); /* we checked for i/o errors above */ 660 } 661 662 if (hash & (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len))) { 663 /* we want the sibling */ 664 zap_put_leaf(l); 665 *lp = nl; 666 } else { 667 zap_put_leaf(nl); 668 *lp = l; 669 } 670 671 return (0); 672 } 673 674 static void 675 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx) 676 { 677 zap_t *zap = zn->zn_zap; 678 int shift = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift; 679 int leaffull = (l->l_phys->l_hdr.lh_prefix_len == shift && 680 l->l_phys->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER); 681 682 zap_put_leaf(l); 683 684 if (leaffull || zap->zap_f.zap_phys->zap_ptrtbl.zt_nextblk) { 685 int err; 686 687 /* 688 * We are in the middle of growing the pointer table, or 689 * this leaf will soon make us grow it. 690 */ 691 if (zap_tryupgradedir(zap, tx) == 0) { 692 objset_t *os = zap->zap_objset; 693 uint64_t zapobj = zap->zap_object; 694 695 zap_unlockdir(zap); 696 err = zap_lockdir(os, zapobj, tx, 697 RW_WRITER, FALSE, FALSE, &zn->zn_zap); 698 zap = zn->zn_zap; 699 if (err) 700 return; 701 } 702 703 /* could have finished growing while our locks were down */ 704 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_shift == shift) 705 (void) zap_grow_ptrtbl(zap, tx); 706 } 707 } 708 709 710 static int 711 fzap_checksize(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers) 712 { 713 if (zn->zn_key_orig_len > ZAP_MAXNAMELEN) 714 return (E2BIG); 715 716 /* Only integer sizes supported by C */ 717 switch (integer_size) { 718 case 1: 719 case 2: 720 case 4: 721 case 8: 722 break; 723 default: 724 return (EINVAL); 725 } 726 727 if (integer_size * num_integers > ZAP_MAXVALUELEN) 728 return (E2BIG); 729 730 return (0); 731 } 732 733 /* 734 * Routines for manipulating attributes. 735 */ 736 int 737 fzap_lookup(zap_name_t *zn, 738 uint64_t integer_size, uint64_t num_integers, void *buf, 739 char *realname, int rn_len, boolean_t *ncp) 740 { 741 zap_leaf_t *l; 742 int err; 743 zap_entry_handle_t zeh; 744 745 err = fzap_checksize(zn, integer_size, num_integers); 746 if (err != 0) 747 return (err); 748 749 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); 750 if (err != 0) 751 return (err); 752 err = zap_leaf_lookup(l, zn, &zeh); 753 if (err == 0) { 754 err = zap_entry_read(&zeh, integer_size, num_integers, buf); 755 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname); 756 if (ncp) { 757 *ncp = zap_entry_normalization_conflict(&zeh, 758 zn, NULL, zn->zn_zap); 759 } 760 } 761 762 zap_put_leaf(l); 763 return (err); 764 } 765 766 int 767 fzap_add_cd(zap_name_t *zn, 768 uint64_t integer_size, uint64_t num_integers, 769 const void *val, uint32_t cd, dmu_tx_t *tx) 770 { 771 zap_leaf_t *l; 772 int err; 773 zap_entry_handle_t zeh; 774 zap_t *zap = zn->zn_zap; 775 776 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 777 ASSERT(!zap->zap_ismicro); 778 ASSERT(fzap_checksize(zn, integer_size, num_integers) == 0); 779 780 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); 781 if (err != 0) 782 return (err); 783 retry: 784 err = zap_leaf_lookup(l, zn, &zeh); 785 if (err == 0) { 786 err = EEXIST; 787 goto out; 788 } 789 if (err != ENOENT) 790 goto out; 791 792 err = zap_entry_create(l, zn, cd, 793 integer_size, num_integers, val, &zeh); 794 795 if (err == 0) { 796 zap_increment_num_entries(zap, 1, tx); 797 } else if (err == EAGAIN) { 798 err = zap_expand_leaf(zn, l, tx, &l); 799 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ 800 if (err == 0) 801 goto retry; 802 } 803 804 out: 805 if (zap != NULL) 806 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx); 807 return (err); 808 } 809 810 int 811 fzap_add(zap_name_t *zn, 812 uint64_t integer_size, uint64_t num_integers, 813 const void *val, dmu_tx_t *tx) 814 { 815 int err = fzap_checksize(zn, integer_size, num_integers); 816 if (err != 0) 817 return (err); 818 819 return (fzap_add_cd(zn, integer_size, num_integers, 820 val, ZAP_NEED_CD, tx)); 821 } 822 823 int 824 fzap_update(zap_name_t *zn, 825 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx) 826 { 827 zap_leaf_t *l; 828 int err, create; 829 zap_entry_handle_t zeh; 830 zap_t *zap = zn->zn_zap; 831 832 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 833 err = fzap_checksize(zn, integer_size, num_integers); 834 if (err != 0) 835 return (err); 836 837 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); 838 if (err != 0) 839 return (err); 840 retry: 841 err = zap_leaf_lookup(l, zn, &zeh); 842 create = (err == ENOENT); 843 ASSERT(err == 0 || err == ENOENT); 844 845 if (create) { 846 err = zap_entry_create(l, zn, ZAP_NEED_CD, 847 integer_size, num_integers, val, &zeh); 848 if (err == 0) 849 zap_increment_num_entries(zap, 1, tx); 850 } else { 851 err = zap_entry_update(&zeh, integer_size, num_integers, val); 852 } 853 854 if (err == EAGAIN) { 855 err = zap_expand_leaf(zn, l, tx, &l); 856 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ 857 if (err == 0) 858 goto retry; 859 } 860 861 if (zap != NULL) 862 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx); 863 return (err); 864 } 865 866 int 867 fzap_length(zap_name_t *zn, 868 uint64_t *integer_size, uint64_t *num_integers) 869 { 870 zap_leaf_t *l; 871 int err; 872 zap_entry_handle_t zeh; 873 874 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); 875 if (err != 0) 876 return (err); 877 err = zap_leaf_lookup(l, zn, &zeh); 878 if (err != 0) 879 goto out; 880 881 if (integer_size) 882 *integer_size = zeh.zeh_integer_size; 883 if (num_integers) 884 *num_integers = zeh.zeh_num_integers; 885 out: 886 zap_put_leaf(l); 887 return (err); 888 } 889 890 int 891 fzap_remove(zap_name_t *zn, dmu_tx_t *tx) 892 { 893 zap_leaf_t *l; 894 int err; 895 zap_entry_handle_t zeh; 896 897 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l); 898 if (err != 0) 899 return (err); 900 err = zap_leaf_lookup(l, zn, &zeh); 901 if (err == 0) { 902 zap_entry_remove(&zeh); 903 zap_increment_num_entries(zn->zn_zap, -1, tx); 904 } 905 zap_put_leaf(l); 906 return (err); 907 } 908 909 /* 910 * Helper functions for consumers. 911 */ 912 913 int 914 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask, 915 char *name) 916 { 917 zap_cursor_t zc; 918 zap_attribute_t *za; 919 int err; 920 921 if (mask == 0) 922 mask = -1ULL; 923 924 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 925 for (zap_cursor_init(&zc, os, zapobj); 926 (err = zap_cursor_retrieve(&zc, za)) == 0; 927 zap_cursor_advance(&zc)) { 928 if ((za->za_first_integer & mask) == (value & mask)) { 929 (void) strcpy(name, za->za_name); 930 break; 931 } 932 } 933 zap_cursor_fini(&zc); 934 kmem_free(za, sizeof (zap_attribute_t)); 935 return (err); 936 } 937 938 int 939 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx) 940 { 941 zap_cursor_t zc; 942 zap_attribute_t za; 943 int err; 944 945 for (zap_cursor_init(&zc, os, fromobj); 946 zap_cursor_retrieve(&zc, &za) == 0; 947 (void) zap_cursor_advance(&zc)) { 948 if (za.za_integer_length != 8 || za.za_num_integers != 1) 949 return (EINVAL); 950 err = zap_add(os, intoobj, za.za_name, 951 8, 1, &za.za_first_integer, tx); 952 if (err) 953 return (err); 954 } 955 zap_cursor_fini(&zc); 956 return (0); 957 } 958 959 int 960 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) 961 { 962 char name[20]; 963 964 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 965 return (zap_add(os, obj, name, 8, 1, &value, tx)); 966 } 967 968 int 969 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) 970 { 971 char name[20]; 972 973 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 974 return (zap_remove(os, obj, name, tx)); 975 } 976 977 int 978 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value) 979 { 980 char name[20]; 981 982 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 983 return (zap_lookup(os, obj, name, 8, 1, &value)); 984 } 985 986 int 987 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta, 988 dmu_tx_t *tx) 989 { 990 char name[20]; 991 uint64_t value = 0; 992 int err; 993 994 if (delta == 0) 995 return (0); 996 997 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 998 err = zap_lookup(os, obj, name, 8, 1, &value); 999 if (err != 0 && err != ENOENT) 1000 return (err); 1001 value += delta; 1002 if (value == 0) 1003 err = zap_remove(os, obj, name, tx); 1004 else 1005 err = zap_update(os, obj, name, 8, 1, &value, tx); 1006 return (err); 1007 } 1008 1009 1010 /* 1011 * Routines for iterating over the attributes. 1012 */ 1013 1014 int 1015 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za) 1016 { 1017 int err = ENOENT; 1018 zap_entry_handle_t zeh; 1019 zap_leaf_t *l; 1020 1021 /* memset(za, 0xba, sizeof (zap_attribute_t)); */ 1022 1023 /* retrieve the next entry at or after zc_hash/zc_cd */ 1024 /* if no entry, return ENOENT */ 1025 1026 if (zc->zc_leaf && 1027 (ZAP_HASH_IDX(zc->zc_hash, 1028 zc->zc_leaf->l_phys->l_hdr.lh_prefix_len) != 1029 zc->zc_leaf->l_phys->l_hdr.lh_prefix)) { 1030 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1031 zap_put_leaf(zc->zc_leaf); 1032 zc->zc_leaf = NULL; 1033 } 1034 1035 again: 1036 if (zc->zc_leaf == NULL) { 1037 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER, 1038 &zc->zc_leaf); 1039 if (err != 0) 1040 return (err); 1041 } else { 1042 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1043 } 1044 l = zc->zc_leaf; 1045 1046 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh); 1047 1048 if (err == ENOENT) { 1049 uint64_t nocare = 1050 (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len)) - 1; 1051 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1; 1052 zc->zc_cd = 0; 1053 if (l->l_phys->l_hdr.lh_prefix_len == 0 || zc->zc_hash == 0) { 1054 zc->zc_hash = -1ULL; 1055 } else { 1056 zap_put_leaf(zc->zc_leaf); 1057 zc->zc_leaf = NULL; 1058 goto again; 1059 } 1060 } 1061 1062 if (err == 0) { 1063 zc->zc_hash = zeh.zeh_hash; 1064 zc->zc_cd = zeh.zeh_cd; 1065 za->za_integer_length = zeh.zeh_integer_size; 1066 za->za_num_integers = zeh.zeh_num_integers; 1067 if (zeh.zeh_num_integers == 0) { 1068 za->za_first_integer = 0; 1069 } else { 1070 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer); 1071 ASSERT(err == 0 || err == EOVERFLOW); 1072 } 1073 err = zap_entry_read_name(zap, &zeh, 1074 sizeof (za->za_name), za->za_name); 1075 ASSERT(err == 0); 1076 1077 za->za_normalization_conflict = 1078 zap_entry_normalization_conflict(&zeh, 1079 NULL, za->za_name, zap); 1080 } 1081 rw_exit(&zc->zc_leaf->l_rwlock); 1082 return (err); 1083 } 1084 1085 1086 static void 1087 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs) 1088 { 1089 int i, err; 1090 uint64_t lastblk = 0; 1091 1092 /* 1093 * NB: if a leaf has more pointers than an entire ptrtbl block 1094 * can hold, then it'll be accounted for more than once, since 1095 * we won't have lastblk. 1096 */ 1097 for (i = 0; i < len; i++) { 1098 zap_leaf_t *l; 1099 1100 if (tbl[i] == lastblk) 1101 continue; 1102 lastblk = tbl[i]; 1103 1104 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l); 1105 if (err == 0) { 1106 zap_leaf_stats(zap, l, zs); 1107 zap_put_leaf(l); 1108 } 1109 } 1110 } 1111 1112 int 1113 fzap_cursor_move_to_key(zap_cursor_t *zc, zap_name_t *zn) 1114 { 1115 int err; 1116 zap_leaf_t *l; 1117 zap_entry_handle_t zeh; 1118 1119 if (zn->zn_key_orig_len > ZAP_MAXNAMELEN) 1120 return (E2BIG); 1121 1122 err = zap_deref_leaf(zc->zc_zap, zn->zn_hash, NULL, RW_READER, &l); 1123 if (err != 0) 1124 return (err); 1125 1126 err = zap_leaf_lookup(l, zn, &zeh); 1127 if (err != 0) 1128 return (err); 1129 1130 zc->zc_leaf = l; 1131 zc->zc_hash = zeh.zeh_hash; 1132 zc->zc_cd = zeh.zeh_cd; 1133 1134 return (err); 1135 } 1136 1137 void 1138 fzap_get_stats(zap_t *zap, zap_stats_t *zs) 1139 { 1140 int bs = FZAP_BLOCK_SHIFT(zap); 1141 zs->zs_blocksize = 1ULL << bs; 1142 1143 /* 1144 * Set zap_phys_t fields 1145 */ 1146 zs->zs_num_leafs = zap->zap_f.zap_phys->zap_num_leafs; 1147 zs->zs_num_entries = zap->zap_f.zap_phys->zap_num_entries; 1148 zs->zs_num_blocks = zap->zap_f.zap_phys->zap_freeblk; 1149 zs->zs_block_type = zap->zap_f.zap_phys->zap_block_type; 1150 zs->zs_magic = zap->zap_f.zap_phys->zap_magic; 1151 zs->zs_salt = zap->zap_f.zap_phys->zap_salt; 1152 1153 /* 1154 * Set zap_ptrtbl fields 1155 */ 1156 zs->zs_ptrtbl_len = 1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift; 1157 zs->zs_ptrtbl_nextblk = zap->zap_f.zap_phys->zap_ptrtbl.zt_nextblk; 1158 zs->zs_ptrtbl_blks_copied = 1159 zap->zap_f.zap_phys->zap_ptrtbl.zt_blks_copied; 1160 zs->zs_ptrtbl_zt_blk = zap->zap_f.zap_phys->zap_ptrtbl.zt_blk; 1161 zs->zs_ptrtbl_zt_numblks = zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks; 1162 zs->zs_ptrtbl_zt_shift = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift; 1163 1164 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) { 1165 /* the ptrtbl is entirely in the header block. */ 1166 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), 1167 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs); 1168 } else { 1169 int b; 1170 1171 dmu_prefetch(zap->zap_objset, zap->zap_object, 1172 zap->zap_f.zap_phys->zap_ptrtbl.zt_blk << bs, 1173 zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks << bs); 1174 1175 for (b = 0; b < zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks; 1176 b++) { 1177 dmu_buf_t *db; 1178 int err; 1179 1180 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 1181 (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk + b) << bs, 1182 FTAG, &db); 1183 if (err == 0) { 1184 zap_stats_ptrtbl(zap, db->db_data, 1185 1<<(bs-3), zs); 1186 dmu_buf_rele(db, FTAG); 1187 } 1188 } 1189 } 1190 } 1191 1192 int 1193 fzap_count_write(zap_name_t *zn, int add, uint64_t *towrite, 1194 uint64_t *tooverwrite) 1195 { 1196 zap_t *zap = zn->zn_zap; 1197 zap_leaf_t *l; 1198 int err; 1199 1200 /* 1201 * Account for the header block of the fatzap. 1202 */ 1203 if (!add && dmu_buf_freeable(zap->zap_dbuf)) { 1204 *tooverwrite += zap->zap_dbuf->db_size; 1205 } else { 1206 *towrite += zap->zap_dbuf->db_size; 1207 } 1208 1209 /* 1210 * Account for the pointer table blocks. 1211 * If we are adding we need to account for the following cases : 1212 * - If the pointer table is embedded, this operation could force an 1213 * external pointer table. 1214 * - If this already has an external pointer table this operation 1215 * could extend the table. 1216 */ 1217 if (add) { 1218 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk == 0) 1219 *towrite += zap->zap_dbuf->db_size; 1220 else 1221 *towrite += (zap->zap_dbuf->db_size * 3); 1222 } 1223 1224 /* 1225 * Now, check if the block containing leaf is freeable 1226 * and account accordingly. 1227 */ 1228 err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l); 1229 if (err != 0) { 1230 return (err); 1231 } 1232 1233 if (!add && dmu_buf_freeable(l->l_dbuf)) { 1234 *tooverwrite += l->l_dbuf->db_size; 1235 } else { 1236 /* 1237 * If this an add operation, the leaf block could split. 1238 * Hence, we need to account for an additional leaf block. 1239 */ 1240 *towrite += (add ? 2 : 1) * l->l_dbuf->db_size; 1241 } 1242 1243 zap_put_leaf(l); 1244 return (0); 1245 } 1246