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) 2012, 2016 by Delphix. All rights reserved. 24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 25 */ 26 27 /* 28 * This file contains the top half of the zfs directory structure 29 * implementation. The bottom half is in zap_leaf.c. 30 * 31 * The zdir is an extendable hash data structure. There is a table of 32 * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are 33 * each a constant size and hold a variable number of directory entries. 34 * The buckets (aka "leaf nodes") are implemented in zap_leaf.c. 35 * 36 * The pointer table holds a power of 2 number of pointers. 37 * (1<<zap_t->zd_data->zd_phys->zd_prefix_len). The bucket pointed to 38 * by the pointer at index i in the table holds entries whose hash value 39 * has a zd_prefix_len - bit prefix 40 */ 41 42 #include <sys/spa.h> 43 #include <sys/dmu.h> 44 #include <sys/zfs_context.h> 45 #include <sys/zfs_znode.h> 46 #include <sys/fs/zfs.h> 47 #include <sys/zap.h> 48 #include <sys/refcount.h> 49 #include <sys/zap_impl.h> 50 #include <sys/zap_leaf.h> 51 52 int fzap_default_block_shift = 14; /* 16k blocksize */ 53 54 extern inline zap_phys_t *zap_f_phys(zap_t *zap); 55 56 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks); 57 58 void 59 fzap_byteswap(void *vbuf, size_t size) 60 { 61 uint64_t block_type; 62 63 block_type = *(uint64_t *)vbuf; 64 65 if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF)) 66 zap_leaf_byteswap(vbuf, size); 67 else { 68 /* it's a ptrtbl block */ 69 byteswap_uint64_array(vbuf, size); 70 } 71 } 72 73 void 74 fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags) 75 { 76 dmu_buf_t *db; 77 zap_leaf_t *l; 78 int i; 79 zap_phys_t *zp; 80 81 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 82 zap->zap_ismicro = FALSE; 83 84 zap->zap_dbu.dbu_evict_func = zap_evict; 85 86 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0); 87 zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1; 88 89 zp = zap_f_phys(zap); 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, DMU_READ_NO_PREFETCH)); 116 dmu_buf_will_dirty(db, tx); 117 118 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); 119 l->l_dbuf = db; 120 121 zap_leaf_init(l, zp->zap_normflags != 0); 122 123 kmem_free(l, sizeof (zap_leaf_t)); 124 dmu_buf_rele(db, FTAG); 125 } 126 127 static int 128 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx) 129 { 130 if (RW_WRITE_HELD(&zap->zap_rwlock)) 131 return (1); 132 if (rw_tryupgrade(&zap->zap_rwlock)) { 133 dmu_buf_will_dirty(zap->zap_dbuf, tx); 134 return (1); 135 } 136 return (0); 137 } 138 139 /* 140 * Generic routines for dealing with the pointer & cookie tables. 141 */ 142 143 static int 144 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl, 145 void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n), 146 dmu_tx_t *tx) 147 { 148 uint64_t b, newblk; 149 dmu_buf_t *db_old, *db_new; 150 int err; 151 int bs = FZAP_BLOCK_SHIFT(zap); 152 int hepb = 1<<(bs-4); 153 /* hepb = half the number of entries in a block */ 154 155 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 156 ASSERT(tbl->zt_blk != 0); 157 ASSERT(tbl->zt_numblks > 0); 158 159 if (tbl->zt_nextblk != 0) { 160 newblk = tbl->zt_nextblk; 161 } else { 162 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2); 163 tbl->zt_nextblk = newblk; 164 ASSERT0(tbl->zt_blks_copied); 165 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, 166 tbl->zt_blk << bs, tbl->zt_numblks << bs, 167 ZIO_PRIORITY_SYNC_READ); 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, DMU_READ_NO_PREFETCH); 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, DMU_READ_NO_PREFETCH)); 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, DMU_READ_NO_PREFETCH)); 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, DMU_READ_NO_PREFETCH); 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 DMU_READ_NO_PREFETCH); 251 if (err) { 252 dmu_buf_rele(db, FTAG); 253 return (err); 254 } 255 dmu_buf_will_dirty(db2, tx); 256 ((uint64_t *)db2->db_data)[off2] = val; 257 ((uint64_t *)db2->db_data)[off2+1] = val; 258 dmu_buf_rele(db2, FTAG); 259 } 260 261 ((uint64_t *)db->db_data)[off] = val; 262 dmu_buf_rele(db, FTAG); 263 264 return (0); 265 } 266 267 static int 268 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp) 269 { 270 uint64_t blk, off; 271 int err; 272 dmu_buf_t *db; 273 int bs = FZAP_BLOCK_SHIFT(zap); 274 275 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 276 277 blk = idx >> (bs-3); 278 off = idx & ((1<<(bs-3))-1); 279 280 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 281 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH); 282 if (err) 283 return (err); 284 *valp = ((uint64_t *)db->db_data)[off]; 285 dmu_buf_rele(db, FTAG); 286 287 if (tbl->zt_nextblk != 0) { 288 /* 289 * read the nextblk for the sake of i/o error checking, 290 * so that zap_table_load() will catch errors for 291 * zap_table_store. 292 */ 293 blk = (idx*2) >> (bs-3); 294 295 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 296 (tbl->zt_nextblk + blk) << bs, FTAG, &db, 297 DMU_READ_NO_PREFETCH); 298 if (err == 0) 299 dmu_buf_rele(db, FTAG); 300 } 301 return (err); 302 } 303 304 /* 305 * Routines for growing the ptrtbl. 306 */ 307 308 static void 309 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n) 310 { 311 int i; 312 for (i = 0; i < n; i++) { 313 uint64_t lb = src[i]; 314 dst[2*i+0] = lb; 315 dst[2*i+1] = lb; 316 } 317 } 318 319 static int 320 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx) 321 { 322 /* 323 * The pointer table should never use more hash bits than we 324 * have (otherwise we'd be using useless zero bits to index it). 325 * If we are within 2 bits of running out, stop growing, since 326 * this is already an aberrant condition. 327 */ 328 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2) 329 return (SET_ERROR(ENOSPC)); 330 331 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) { 332 /* 333 * We are outgrowing the "embedded" ptrtbl (the one 334 * stored in the header block). Give it its own entire 335 * block, which will double the size of the ptrtbl. 336 */ 337 uint64_t newblk; 338 dmu_buf_t *db_new; 339 int err; 340 341 ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==, 342 ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); 343 ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk); 344 345 newblk = zap_allocate_blocks(zap, 1); 346 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 347 newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new, 348 DMU_READ_NO_PREFETCH); 349 if (err) 350 return (err); 351 dmu_buf_will_dirty(db_new, tx); 352 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), 353 db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); 354 dmu_buf_rele(db_new, FTAG); 355 356 zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk; 357 zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1; 358 zap_f_phys(zap)->zap_ptrtbl.zt_shift++; 359 360 ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==, 361 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << 362 (FZAP_BLOCK_SHIFT(zap)-3)); 363 364 return (0); 365 } else { 366 return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl, 367 zap_ptrtbl_transfer, tx)); 368 } 369 } 370 371 static void 372 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx) 373 { 374 dmu_buf_will_dirty(zap->zap_dbuf, tx); 375 mutex_enter(&zap->zap_f.zap_num_entries_mtx); 376 ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta); 377 zap_f_phys(zap)->zap_num_entries += delta; 378 mutex_exit(&zap->zap_f.zap_num_entries_mtx); 379 } 380 381 static uint64_t 382 zap_allocate_blocks(zap_t *zap, int nblocks) 383 { 384 uint64_t newblk; 385 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 386 newblk = zap_f_phys(zap)->zap_freeblk; 387 zap_f_phys(zap)->zap_freeblk += nblocks; 388 return (newblk); 389 } 390 391 static void 392 zap_leaf_pageout(void *dbu) 393 { 394 zap_leaf_t *l = dbu; 395 396 rw_destroy(&l->l_rwlock); 397 kmem_free(l, sizeof (zap_leaf_t)); 398 } 399 400 static zap_leaf_t * 401 zap_create_leaf(zap_t *zap, dmu_tx_t *tx) 402 { 403 void *winner; 404 zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); 405 406 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 407 408 rw_init(&l->l_rwlock, 0, 0, 0); 409 rw_enter(&l->l_rwlock, RW_WRITER); 410 l->l_blkid = zap_allocate_blocks(zap, 1); 411 l->l_dbuf = NULL; 412 413 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 414 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf, 415 DMU_READ_NO_PREFETCH)); 416 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf); 417 winner = dmu_buf_set_user(l->l_dbuf, &l->l_dbu); 418 ASSERT(winner == NULL); 419 dmu_buf_will_dirty(l->l_dbuf, tx); 420 421 zap_leaf_init(l, zap->zap_normflags != 0); 422 423 zap_f_phys(zap)->zap_num_leafs++; 424 425 return (l); 426 } 427 428 int 429 fzap_count(zap_t *zap, uint64_t *count) 430 { 431 ASSERT(!zap->zap_ismicro); 432 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */ 433 *count = zap_f_phys(zap)->zap_num_entries; 434 mutex_exit(&zap->zap_f.zap_num_entries_mtx); 435 return (0); 436 } 437 438 /* 439 * Routines for obtaining zap_leaf_t's 440 */ 441 442 void 443 zap_put_leaf(zap_leaf_t *l) 444 { 445 rw_exit(&l->l_rwlock); 446 dmu_buf_rele(l->l_dbuf, NULL); 447 } 448 449 static zap_leaf_t * 450 zap_open_leaf(uint64_t blkid, dmu_buf_t *db) 451 { 452 zap_leaf_t *l, *winner; 453 454 ASSERT(blkid != 0); 455 456 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); 457 rw_init(&l->l_rwlock, 0, 0, 0); 458 rw_enter(&l->l_rwlock, RW_WRITER); 459 l->l_blkid = blkid; 460 l->l_bs = highbit64(db->db_size) - 1; 461 l->l_dbuf = db; 462 463 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf); 464 winner = dmu_buf_set_user(db, &l->l_dbu); 465 466 rw_exit(&l->l_rwlock); 467 if (winner != NULL) { 468 /* someone else set it first */ 469 zap_leaf_pageout(&l->l_dbu); 470 l = winner; 471 } 472 473 /* 474 * lhr_pad was previously used for the next leaf in the leaf 475 * chain. There should be no chained leafs (as we have removed 476 * support for them). 477 */ 478 ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1); 479 480 /* 481 * There should be more hash entries than there can be 482 * chunks to put in the hash table 483 */ 484 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3); 485 486 /* The chunks should begin at the end of the hash table */ 487 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==, 488 &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]); 489 490 /* The chunks should end at the end of the block */ 491 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) - 492 (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size); 493 494 return (l); 495 } 496 497 static int 498 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt, 499 zap_leaf_t **lp) 500 { 501 dmu_buf_t *db; 502 zap_leaf_t *l; 503 int bs = FZAP_BLOCK_SHIFT(zap); 504 int err; 505 506 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 507 508 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 509 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH); 510 if (err) 511 return (err); 512 513 ASSERT3U(db->db_object, ==, zap->zap_object); 514 ASSERT3U(db->db_offset, ==, blkid << bs); 515 ASSERT3U(db->db_size, ==, 1 << bs); 516 ASSERT(blkid != 0); 517 518 l = dmu_buf_get_user(db); 519 520 if (l == NULL) 521 l = zap_open_leaf(blkid, db); 522 523 rw_enter(&l->l_rwlock, lt); 524 /* 525 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change, 526 * causing ASSERT below to fail. 527 */ 528 if (lt == RW_WRITER) 529 dmu_buf_will_dirty(db, tx); 530 ASSERT3U(l->l_blkid, ==, blkid); 531 ASSERT3P(l->l_dbuf, ==, db); 532 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF); 533 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC); 534 535 *lp = l; 536 return (0); 537 } 538 539 static int 540 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp) 541 { 542 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 543 544 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) { 545 ASSERT3U(idx, <, 546 (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift)); 547 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx); 548 return (0); 549 } else { 550 return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl, 551 idx, valp)); 552 } 553 } 554 555 static int 556 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx) 557 { 558 ASSERT(tx != NULL); 559 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 560 561 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) { 562 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk; 563 return (0); 564 } else { 565 return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl, 566 idx, blk, tx)); 567 } 568 } 569 570 static int 571 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp) 572 { 573 uint64_t idx, blk; 574 int err; 575 576 ASSERT(zap->zap_dbuf == NULL || 577 zap_f_phys(zap) == zap->zap_dbuf->db_data); 578 579 /* Reality check for corrupt zap objects (leaf or header). */ 580 if ((zap_f_phys(zap)->zap_block_type != ZBT_LEAF && 581 zap_f_phys(zap)->zap_block_type != ZBT_HEADER) || 582 zap_f_phys(zap)->zap_magic != ZAP_MAGIC) { 583 return (SET_ERROR(EIO)); 584 } 585 586 idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift); 587 err = zap_idx_to_blk(zap, idx, &blk); 588 if (err != 0) 589 return (err); 590 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp); 591 592 ASSERT(err || 593 ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) == 594 zap_leaf_phys(*lp)->l_hdr.lh_prefix); 595 return (err); 596 } 597 598 static int 599 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, 600 void *tag, dmu_tx_t *tx, zap_leaf_t **lp) 601 { 602 zap_t *zap = zn->zn_zap; 603 uint64_t hash = zn->zn_hash; 604 zap_leaf_t *nl; 605 int prefix_diff, i, err; 606 uint64_t sibling; 607 int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len; 608 609 ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift); 610 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 611 612 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, 613 zap_leaf_phys(l)->l_hdr.lh_prefix); 614 615 if (zap_tryupgradedir(zap, tx) == 0 || 616 old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) { 617 /* We failed to upgrade, or need to grow the pointer table */ 618 objset_t *os = zap->zap_objset; 619 uint64_t object = zap->zap_object; 620 621 zap_put_leaf(l); 622 zap_unlockdir(zap, tag); 623 err = zap_lockdir(os, object, tx, RW_WRITER, 624 FALSE, FALSE, tag, &zn->zn_zap); 625 zap = zn->zn_zap; 626 if (err) 627 return (err); 628 ASSERT(!zap->zap_ismicro); 629 630 while (old_prefix_len == 631 zap_f_phys(zap)->zap_ptrtbl.zt_shift) { 632 err = zap_grow_ptrtbl(zap, tx); 633 if (err) 634 return (err); 635 } 636 637 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l); 638 if (err) 639 return (err); 640 641 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) { 642 /* it split while our locks were down */ 643 *lp = l; 644 return (0); 645 } 646 } 647 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 648 ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift); 649 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, 650 zap_leaf_phys(l)->l_hdr.lh_prefix); 651 652 prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift - 653 (old_prefix_len + 1); 654 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff; 655 656 /* check for i/o errors before doing zap_leaf_split */ 657 for (i = 0; i < (1ULL<<prefix_diff); i++) { 658 uint64_t blk; 659 err = zap_idx_to_blk(zap, sibling+i, &blk); 660 if (err) 661 return (err); 662 ASSERT3U(blk, ==, l->l_blkid); 663 } 664 665 nl = zap_create_leaf(zap, tx); 666 zap_leaf_split(l, nl, zap->zap_normflags != 0); 667 668 /* set sibling pointers */ 669 for (i = 0; i < (1ULL << prefix_diff); i++) { 670 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx); 671 ASSERT0(err); /* we checked for i/o errors above */ 672 } 673 674 if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) { 675 /* we want the sibling */ 676 zap_put_leaf(l); 677 *lp = nl; 678 } else { 679 zap_put_leaf(nl); 680 *lp = l; 681 } 682 683 return (0); 684 } 685 686 static void 687 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, 688 void *tag, dmu_tx_t *tx) 689 { 690 zap_t *zap = zn->zn_zap; 691 int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift; 692 int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift && 693 zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER); 694 695 zap_put_leaf(l); 696 697 if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) { 698 int err; 699 700 /* 701 * We are in the middle of growing the pointer table, or 702 * this leaf will soon make us grow it. 703 */ 704 if (zap_tryupgradedir(zap, tx) == 0) { 705 objset_t *os = zap->zap_objset; 706 uint64_t zapobj = zap->zap_object; 707 708 zap_unlockdir(zap, tag); 709 err = zap_lockdir(os, zapobj, tx, 710 RW_WRITER, FALSE, FALSE, tag, &zn->zn_zap); 711 zap = zn->zn_zap; 712 if (err) 713 return; 714 } 715 716 /* could have finished growing while our locks were down */ 717 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift) 718 (void) zap_grow_ptrtbl(zap, tx); 719 } 720 } 721 722 static int 723 fzap_checkname(zap_name_t *zn) 724 { 725 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN) 726 return (SET_ERROR(ENAMETOOLONG)); 727 return (0); 728 } 729 730 static int 731 fzap_checksize(uint64_t integer_size, uint64_t num_integers) 732 { 733 /* Only integer sizes supported by C */ 734 switch (integer_size) { 735 case 1: 736 case 2: 737 case 4: 738 case 8: 739 break; 740 default: 741 return (SET_ERROR(EINVAL)); 742 } 743 744 if (integer_size * num_integers > ZAP_MAXVALUELEN) 745 return (E2BIG); 746 747 return (0); 748 } 749 750 static int 751 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers) 752 { 753 int err; 754 755 if ((err = fzap_checkname(zn)) != 0) 756 return (err); 757 return (fzap_checksize(integer_size, num_integers)); 758 } 759 760 /* 761 * Routines for manipulating attributes. 762 */ 763 int 764 fzap_lookup(zap_name_t *zn, 765 uint64_t integer_size, uint64_t num_integers, void *buf, 766 char *realname, int rn_len, boolean_t *ncp) 767 { 768 zap_leaf_t *l; 769 int err; 770 zap_entry_handle_t zeh; 771 772 if ((err = fzap_checkname(zn)) != 0) 773 return (err); 774 775 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); 776 if (err != 0) 777 return (err); 778 err = zap_leaf_lookup(l, zn, &zeh); 779 if (err == 0) { 780 if ((err = fzap_checksize(integer_size, num_integers)) != 0) { 781 zap_put_leaf(l); 782 return (err); 783 } 784 785 err = zap_entry_read(&zeh, integer_size, num_integers, buf); 786 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname); 787 if (ncp) { 788 *ncp = zap_entry_normalization_conflict(&zeh, 789 zn, NULL, zn->zn_zap); 790 } 791 } 792 793 zap_put_leaf(l); 794 return (err); 795 } 796 797 int 798 fzap_add_cd(zap_name_t *zn, 799 uint64_t integer_size, uint64_t num_integers, 800 const void *val, uint32_t cd, void *tag, dmu_tx_t *tx) 801 { 802 zap_leaf_t *l; 803 int err; 804 zap_entry_handle_t zeh; 805 zap_t *zap = zn->zn_zap; 806 807 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 808 ASSERT(!zap->zap_ismicro); 809 ASSERT(fzap_check(zn, integer_size, num_integers) == 0); 810 811 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); 812 if (err != 0) 813 return (err); 814 retry: 815 err = zap_leaf_lookup(l, zn, &zeh); 816 if (err == 0) { 817 err = SET_ERROR(EEXIST); 818 goto out; 819 } 820 if (err != ENOENT) 821 goto out; 822 823 err = zap_entry_create(l, zn, cd, 824 integer_size, num_integers, val, &zeh); 825 826 if (err == 0) { 827 zap_increment_num_entries(zap, 1, tx); 828 } else if (err == EAGAIN) { 829 err = zap_expand_leaf(zn, l, tag, tx, &l); 830 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ 831 if (err == 0) 832 goto retry; 833 } 834 835 out: 836 if (zap != NULL) 837 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx); 838 return (err); 839 } 840 841 int 842 fzap_add(zap_name_t *zn, 843 uint64_t integer_size, uint64_t num_integers, 844 const void *val, void *tag, dmu_tx_t *tx) 845 { 846 int err = fzap_check(zn, integer_size, num_integers); 847 if (err != 0) 848 return (err); 849 850 return (fzap_add_cd(zn, integer_size, num_integers, 851 val, ZAP_NEED_CD, tag, tx)); 852 } 853 854 int 855 fzap_update(zap_name_t *zn, 856 int integer_size, uint64_t num_integers, const void *val, 857 void *tag, dmu_tx_t *tx) 858 { 859 zap_leaf_t *l; 860 int err, create; 861 zap_entry_handle_t zeh; 862 zap_t *zap = zn->zn_zap; 863 864 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 865 err = fzap_check(zn, integer_size, num_integers); 866 if (err != 0) 867 return (err); 868 869 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); 870 if (err != 0) 871 return (err); 872 retry: 873 err = zap_leaf_lookup(l, zn, &zeh); 874 create = (err == ENOENT); 875 ASSERT(err == 0 || err == ENOENT); 876 877 if (create) { 878 err = zap_entry_create(l, zn, ZAP_NEED_CD, 879 integer_size, num_integers, val, &zeh); 880 if (err == 0) 881 zap_increment_num_entries(zap, 1, tx); 882 } else { 883 err = zap_entry_update(&zeh, integer_size, num_integers, val); 884 } 885 886 if (err == EAGAIN) { 887 err = zap_expand_leaf(zn, l, tag, tx, &l); 888 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ 889 if (err == 0) 890 goto retry; 891 } 892 893 if (zap != NULL) 894 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx); 895 return (err); 896 } 897 898 int 899 fzap_length(zap_name_t *zn, 900 uint64_t *integer_size, uint64_t *num_integers) 901 { 902 zap_leaf_t *l; 903 int err; 904 zap_entry_handle_t zeh; 905 906 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); 907 if (err != 0) 908 return (err); 909 err = zap_leaf_lookup(l, zn, &zeh); 910 if (err != 0) 911 goto out; 912 913 if (integer_size) 914 *integer_size = zeh.zeh_integer_size; 915 if (num_integers) 916 *num_integers = zeh.zeh_num_integers; 917 out: 918 zap_put_leaf(l); 919 return (err); 920 } 921 922 int 923 fzap_remove(zap_name_t *zn, dmu_tx_t *tx) 924 { 925 zap_leaf_t *l; 926 int err; 927 zap_entry_handle_t zeh; 928 929 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l); 930 if (err != 0) 931 return (err); 932 err = zap_leaf_lookup(l, zn, &zeh); 933 if (err == 0) { 934 zap_entry_remove(&zeh); 935 zap_increment_num_entries(zn->zn_zap, -1, tx); 936 } 937 zap_put_leaf(l); 938 return (err); 939 } 940 941 void 942 fzap_prefetch(zap_name_t *zn) 943 { 944 uint64_t idx, blk; 945 zap_t *zap = zn->zn_zap; 946 int bs; 947 948 idx = ZAP_HASH_IDX(zn->zn_hash, 949 zap_f_phys(zap)->zap_ptrtbl.zt_shift); 950 if (zap_idx_to_blk(zap, idx, &blk) != 0) 951 return; 952 bs = FZAP_BLOCK_SHIFT(zap); 953 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs, 954 ZIO_PRIORITY_SYNC_READ); 955 } 956 957 /* 958 * Helper functions for consumers. 959 */ 960 961 uint64_t 962 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj, 963 const char *name, dmu_tx_t *tx) 964 { 965 uint64_t new_obj; 966 967 VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0); 968 VERIFY0(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj, 969 tx)); 970 971 return (new_obj); 972 } 973 974 int 975 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask, 976 char *name) 977 { 978 zap_cursor_t zc; 979 zap_attribute_t *za; 980 int err; 981 982 if (mask == 0) 983 mask = -1ULL; 984 985 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 986 for (zap_cursor_init(&zc, os, zapobj); 987 (err = zap_cursor_retrieve(&zc, za)) == 0; 988 zap_cursor_advance(&zc)) { 989 if ((za->za_first_integer & mask) == (value & mask)) { 990 (void) strcpy(name, za->za_name); 991 break; 992 } 993 } 994 zap_cursor_fini(&zc); 995 kmem_free(za, sizeof (zap_attribute_t)); 996 return (err); 997 } 998 999 int 1000 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx) 1001 { 1002 zap_cursor_t zc; 1003 zap_attribute_t za; 1004 int err; 1005 1006 err = 0; 1007 for (zap_cursor_init(&zc, os, fromobj); 1008 zap_cursor_retrieve(&zc, &za) == 0; 1009 (void) zap_cursor_advance(&zc)) { 1010 if (za.za_integer_length != 8 || za.za_num_integers != 1) { 1011 err = SET_ERROR(EINVAL); 1012 break; 1013 } 1014 err = zap_add(os, intoobj, za.za_name, 1015 8, 1, &za.za_first_integer, tx); 1016 if (err) 1017 break; 1018 } 1019 zap_cursor_fini(&zc); 1020 return (err); 1021 } 1022 1023 int 1024 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj, 1025 uint64_t value, dmu_tx_t *tx) 1026 { 1027 zap_cursor_t zc; 1028 zap_attribute_t za; 1029 int err; 1030 1031 err = 0; 1032 for (zap_cursor_init(&zc, os, fromobj); 1033 zap_cursor_retrieve(&zc, &za) == 0; 1034 (void) zap_cursor_advance(&zc)) { 1035 if (za.za_integer_length != 8 || za.za_num_integers != 1) { 1036 err = SET_ERROR(EINVAL); 1037 break; 1038 } 1039 err = zap_add(os, intoobj, za.za_name, 1040 8, 1, &value, tx); 1041 if (err) 1042 break; 1043 } 1044 zap_cursor_fini(&zc); 1045 return (err); 1046 } 1047 1048 int 1049 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj, 1050 dmu_tx_t *tx) 1051 { 1052 zap_cursor_t zc; 1053 zap_attribute_t za; 1054 int err; 1055 1056 err = 0; 1057 for (zap_cursor_init(&zc, os, fromobj); 1058 zap_cursor_retrieve(&zc, &za) == 0; 1059 (void) zap_cursor_advance(&zc)) { 1060 uint64_t delta = 0; 1061 1062 if (za.za_integer_length != 8 || za.za_num_integers != 1) { 1063 err = SET_ERROR(EINVAL); 1064 break; 1065 } 1066 1067 err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta); 1068 if (err != 0 && err != ENOENT) 1069 break; 1070 delta += za.za_first_integer; 1071 err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx); 1072 if (err) 1073 break; 1074 } 1075 zap_cursor_fini(&zc); 1076 return (err); 1077 } 1078 1079 int 1080 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) 1081 { 1082 char name[20]; 1083 1084 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 1085 return (zap_add(os, obj, name, 8, 1, &value, tx)); 1086 } 1087 1088 int 1089 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) 1090 { 1091 char name[20]; 1092 1093 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 1094 return (zap_remove(os, obj, name, tx)); 1095 } 1096 1097 int 1098 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value) 1099 { 1100 char name[20]; 1101 1102 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 1103 return (zap_lookup(os, obj, name, 8, 1, &value)); 1104 } 1105 1106 int 1107 zap_add_int_key(objset_t *os, uint64_t obj, 1108 uint64_t key, uint64_t value, dmu_tx_t *tx) 1109 { 1110 char name[20]; 1111 1112 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 1113 return (zap_add(os, obj, name, 8, 1, &value, tx)); 1114 } 1115 1116 int 1117 zap_update_int_key(objset_t *os, uint64_t obj, 1118 uint64_t key, uint64_t value, dmu_tx_t *tx) 1119 { 1120 char name[20]; 1121 1122 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 1123 return (zap_update(os, obj, name, 8, 1, &value, tx)); 1124 } 1125 1126 int 1127 zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep) 1128 { 1129 char name[20]; 1130 1131 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 1132 return (zap_lookup(os, obj, name, 8, 1, valuep)); 1133 } 1134 1135 int 1136 zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta, 1137 dmu_tx_t *tx) 1138 { 1139 uint64_t value = 0; 1140 int err; 1141 1142 if (delta == 0) 1143 return (0); 1144 1145 err = zap_lookup(os, obj, name, 8, 1, &value); 1146 if (err != 0 && err != ENOENT) 1147 return (err); 1148 value += delta; 1149 if (value == 0) 1150 err = zap_remove(os, obj, name, tx); 1151 else 1152 err = zap_update(os, obj, name, 8, 1, &value, tx); 1153 return (err); 1154 } 1155 1156 int 1157 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta, 1158 dmu_tx_t *tx) 1159 { 1160 char name[20]; 1161 1162 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 1163 return (zap_increment(os, obj, name, delta, tx)); 1164 } 1165 1166 /* 1167 * Routines for iterating over the attributes. 1168 */ 1169 1170 int 1171 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za) 1172 { 1173 int err = ENOENT; 1174 zap_entry_handle_t zeh; 1175 zap_leaf_t *l; 1176 1177 /* retrieve the next entry at or after zc_hash/zc_cd */ 1178 /* if no entry, return ENOENT */ 1179 1180 if (zc->zc_leaf && 1181 (ZAP_HASH_IDX(zc->zc_hash, 1182 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) != 1183 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) { 1184 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1185 zap_put_leaf(zc->zc_leaf); 1186 zc->zc_leaf = NULL; 1187 } 1188 1189 again: 1190 if (zc->zc_leaf == NULL) { 1191 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER, 1192 &zc->zc_leaf); 1193 if (err != 0) 1194 return (err); 1195 } else { 1196 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1197 } 1198 l = zc->zc_leaf; 1199 1200 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh); 1201 1202 if (err == ENOENT) { 1203 uint64_t nocare = 1204 (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1; 1205 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1; 1206 zc->zc_cd = 0; 1207 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 || 1208 zc->zc_hash == 0) { 1209 zc->zc_hash = -1ULL; 1210 } else { 1211 zap_put_leaf(zc->zc_leaf); 1212 zc->zc_leaf = NULL; 1213 goto again; 1214 } 1215 } 1216 1217 if (err == 0) { 1218 zc->zc_hash = zeh.zeh_hash; 1219 zc->zc_cd = zeh.zeh_cd; 1220 za->za_integer_length = zeh.zeh_integer_size; 1221 za->za_num_integers = zeh.zeh_num_integers; 1222 if (zeh.zeh_num_integers == 0) { 1223 za->za_first_integer = 0; 1224 } else { 1225 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer); 1226 ASSERT(err == 0 || err == EOVERFLOW); 1227 } 1228 err = zap_entry_read_name(zap, &zeh, 1229 sizeof (za->za_name), za->za_name); 1230 ASSERT(err == 0); 1231 1232 za->za_normalization_conflict = 1233 zap_entry_normalization_conflict(&zeh, 1234 NULL, za->za_name, zap); 1235 } 1236 rw_exit(&zc->zc_leaf->l_rwlock); 1237 return (err); 1238 } 1239 1240 static void 1241 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs) 1242 { 1243 int i, err; 1244 uint64_t lastblk = 0; 1245 1246 /* 1247 * NB: if a leaf has more pointers than an entire ptrtbl block 1248 * can hold, then it'll be accounted for more than once, since 1249 * we won't have lastblk. 1250 */ 1251 for (i = 0; i < len; i++) { 1252 zap_leaf_t *l; 1253 1254 if (tbl[i] == lastblk) 1255 continue; 1256 lastblk = tbl[i]; 1257 1258 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l); 1259 if (err == 0) { 1260 zap_leaf_stats(zap, l, zs); 1261 zap_put_leaf(l); 1262 } 1263 } 1264 } 1265 1266 void 1267 fzap_get_stats(zap_t *zap, zap_stats_t *zs) 1268 { 1269 int bs = FZAP_BLOCK_SHIFT(zap); 1270 zs->zs_blocksize = 1ULL << bs; 1271 1272 /* 1273 * Set zap_phys_t fields 1274 */ 1275 zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs; 1276 zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries; 1277 zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk; 1278 zs->zs_block_type = zap_f_phys(zap)->zap_block_type; 1279 zs->zs_magic = zap_f_phys(zap)->zap_magic; 1280 zs->zs_salt = zap_f_phys(zap)->zap_salt; 1281 1282 /* 1283 * Set zap_ptrtbl fields 1284 */ 1285 zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift; 1286 zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk; 1287 zs->zs_ptrtbl_blks_copied = 1288 zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied; 1289 zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk; 1290 zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks; 1291 zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift; 1292 1293 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) { 1294 /* the ptrtbl is entirely in the header block. */ 1295 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), 1296 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs); 1297 } else { 1298 int b; 1299 1300 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, 1301 zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs, 1302 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs, 1303 ZIO_PRIORITY_SYNC_READ); 1304 1305 for (b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks; 1306 b++) { 1307 dmu_buf_t *db; 1308 int err; 1309 1310 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 1311 (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs, 1312 FTAG, &db, DMU_READ_NO_PREFETCH); 1313 if (err == 0) { 1314 zap_stats_ptrtbl(zap, db->db_data, 1315 1<<(bs-3), zs); 1316 dmu_buf_rele(db, FTAG); 1317 } 1318 } 1319 } 1320 } 1321 1322 int 1323 fzap_count_write(zap_name_t *zn, int add, refcount_t *towrite, 1324 refcount_t *tooverwrite) 1325 { 1326 zap_t *zap = zn->zn_zap; 1327 zap_leaf_t *l; 1328 int err; 1329 1330 /* 1331 * Account for the header block of the fatzap. 1332 */ 1333 if (!add && dmu_buf_freeable(zap->zap_dbuf)) { 1334 (void) refcount_add_many(tooverwrite, 1335 zap->zap_dbuf->db_size, FTAG); 1336 } else { 1337 (void) refcount_add_many(towrite, 1338 zap->zap_dbuf->db_size, FTAG); 1339 } 1340 1341 /* 1342 * Account for the pointer table blocks. 1343 * If we are adding we need to account for the following cases : 1344 * - If the pointer table is embedded, this operation could force an 1345 * external pointer table. 1346 * - If this already has an external pointer table this operation 1347 * could extend the table. 1348 */ 1349 if (add) { 1350 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) { 1351 (void) refcount_add_many(towrite, 1352 zap->zap_dbuf->db_size, FTAG); 1353 } else { 1354 (void) refcount_add_many(towrite, 1355 zap->zap_dbuf->db_size * 3, FTAG); 1356 } 1357 } 1358 1359 /* 1360 * Now, check if the block containing leaf is freeable 1361 * and account accordingly. 1362 */ 1363 err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l); 1364 if (err != 0) { 1365 return (err); 1366 } 1367 1368 if (!add && dmu_buf_freeable(l->l_dbuf)) { 1369 (void) refcount_add_many(tooverwrite, l->l_dbuf->db_size, FTAG); 1370 } else { 1371 /* 1372 * If this an add operation, the leaf block could split. 1373 * Hence, we need to account for an additional leaf block. 1374 */ 1375 (void) refcount_add_many(towrite, 1376 (add ? 2 : 1) * l->l_dbuf->db_size, FTAG); 1377 } 1378 1379 zap_put_leaf(l); 1380 return (0); 1381 } 1382