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 2006 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/zfs_context.h> 29 #include <sys/dbuf.h> 30 #include <sys/dnode.h> 31 #include <sys/dmu.h> 32 #include <sys/dmu_impl.h> 33 #include <sys/dmu_tx.h> 34 #include <sys/dmu_objset.h> 35 #include <sys/dsl_dir.h> 36 #include <sys/dsl_dataset.h> 37 #include <sys/spa.h> 38 #include <sys/zio.h> 39 #include <sys/dmu_zfetch.h> 40 41 static int free_range_compar(const void *node1, const void *node2); 42 43 static kmem_cache_t *dnode_cache; 44 45 static dnode_phys_t dnode_phys_zero; 46 47 int zfs_default_bs = SPA_MINBLOCKSHIFT; 48 int zfs_default_ibs = DN_MAX_INDBLKSHIFT; 49 50 /* ARGSUSED */ 51 static int 52 dnode_cons(void *arg, void *unused, int kmflag) 53 { 54 int i; 55 dnode_t *dn = arg; 56 bzero(dn, sizeof (dnode_t)); 57 58 rw_init(&dn->dn_struct_rwlock, NULL, RW_DEFAULT, NULL); 59 mutex_init(&dn->dn_mtx, NULL, MUTEX_DEFAULT, NULL); 60 mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL); 61 refcount_create(&dn->dn_holds); 62 refcount_create(&dn->dn_tx_holds); 63 64 for (i = 0; i < TXG_SIZE; i++) { 65 avl_create(&dn->dn_ranges[i], free_range_compar, 66 sizeof (free_range_t), 67 offsetof(struct free_range, fr_node)); 68 list_create(&dn->dn_dirty_dbufs[i], 69 sizeof (dmu_buf_impl_t), 70 offsetof(dmu_buf_impl_t, db_dirty_node[i])); 71 } 72 73 list_create(&dn->dn_dbufs, sizeof (dmu_buf_impl_t), 74 offsetof(dmu_buf_impl_t, db_link)); 75 76 return (0); 77 } 78 79 /* ARGSUSED */ 80 static void 81 dnode_dest(void *arg, void *unused) 82 { 83 int i; 84 dnode_t *dn = arg; 85 86 rw_destroy(&dn->dn_struct_rwlock); 87 mutex_destroy(&dn->dn_mtx); 88 mutex_destroy(&dn->dn_dbufs_mtx); 89 refcount_destroy(&dn->dn_holds); 90 refcount_destroy(&dn->dn_tx_holds); 91 92 for (i = 0; i < TXG_SIZE; i++) { 93 avl_destroy(&dn->dn_ranges[i]); 94 list_destroy(&dn->dn_dirty_dbufs[i]); 95 } 96 97 list_destroy(&dn->dn_dbufs); 98 } 99 100 void 101 dnode_init(void) 102 { 103 dnode_cache = kmem_cache_create("dnode_t", 104 sizeof (dnode_t), 105 0, dnode_cons, dnode_dest, NULL, NULL, NULL, 0); 106 } 107 108 void 109 dnode_fini(void) 110 { 111 kmem_cache_destroy(dnode_cache); 112 } 113 114 115 #ifdef ZFS_DEBUG 116 void 117 dnode_verify(dnode_t *dn) 118 { 119 int drop_struct_lock = FALSE; 120 121 ASSERT(dn->dn_phys); 122 ASSERT(dn->dn_objset); 123 124 ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); 125 126 if (!(zfs_flags & ZFS_DEBUG_DNODE_VERIFY)) 127 return; 128 129 if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { 130 rw_enter(&dn->dn_struct_rwlock, RW_READER); 131 drop_struct_lock = TRUE; 132 } 133 if (dn->dn_phys->dn_type != DMU_OT_NONE || dn->dn_allocated_txg != 0) { 134 int i; 135 ASSERT3U(dn->dn_indblkshift, >=, 0); 136 ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT); 137 if (dn->dn_datablkshift) { 138 ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT); 139 ASSERT3U(dn->dn_datablkshift, <=, SPA_MAXBLOCKSHIFT); 140 ASSERT3U(1<<dn->dn_datablkshift, ==, dn->dn_datablksz); 141 } 142 ASSERT3U(dn->dn_nlevels, <=, 30); 143 ASSERT3U(dn->dn_type, <=, DMU_OT_NUMTYPES); 144 ASSERT3U(dn->dn_nblkptr, >=, 1); 145 ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 146 ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); 147 ASSERT3U(dn->dn_datablksz, ==, 148 dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); 149 ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0); 150 ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) + 151 dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); 152 for (i = 0; i < TXG_SIZE; i++) { 153 ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels); 154 } 155 } 156 if (dn->dn_phys->dn_type != DMU_OT_NONE) 157 ASSERT3U(dn->dn_phys->dn_nlevels, <=, dn->dn_nlevels); 158 ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || dn->dn_dbuf != NULL); 159 if (dn->dn_dbuf != NULL) { 160 ASSERT3P(dn->dn_phys, ==, 161 (dnode_phys_t *)dn->dn_dbuf->db.db_data + 162 (dn->dn_object % (dn->dn_dbuf->db.db_size >> DNODE_SHIFT))); 163 } 164 if (drop_struct_lock) 165 rw_exit(&dn->dn_struct_rwlock); 166 } 167 #endif 168 169 void 170 dnode_byteswap(dnode_phys_t *dnp) 171 { 172 uint64_t *buf64 = (void*)&dnp->dn_blkptr; 173 int i; 174 175 if (dnp->dn_type == DMU_OT_NONE) { 176 bzero(dnp, sizeof (dnode_phys_t)); 177 return; 178 } 179 180 dnp->dn_type = BSWAP_8(dnp->dn_type); 181 dnp->dn_indblkshift = BSWAP_8(dnp->dn_indblkshift); 182 dnp->dn_nlevels = BSWAP_8(dnp->dn_nlevels); 183 dnp->dn_nblkptr = BSWAP_8(dnp->dn_nblkptr); 184 dnp->dn_bonustype = BSWAP_8(dnp->dn_bonustype); 185 dnp->dn_checksum = BSWAP_8(dnp->dn_checksum); 186 dnp->dn_compress = BSWAP_8(dnp->dn_compress); 187 dnp->dn_datablkszsec = BSWAP_16(dnp->dn_datablkszsec); 188 dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen); 189 dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid); 190 dnp->dn_secphys = BSWAP_64(dnp->dn_secphys); 191 192 /* 193 * dn_nblkptr is only one byte, so it's OK to read it in either 194 * byte order. We can't read dn_bouslen. 195 */ 196 ASSERT(dnp->dn_indblkshift <= SPA_MAXBLOCKSHIFT); 197 ASSERT(dnp->dn_nblkptr <= DN_MAX_NBLKPTR); 198 for (i = 0; i < dnp->dn_nblkptr * sizeof (blkptr_t)/8; i++) 199 buf64[i] = BSWAP_64(buf64[i]); 200 201 /* 202 * OK to check dn_bonuslen for zero, because it won't matter if 203 * we have the wrong byte order. This is necessary because the 204 * dnode dnode is smaller than a regular dnode. 205 */ 206 if (dnp->dn_bonuslen != 0) { 207 /* 208 * Note that the bonus length calculated here may be 209 * longer than the actual bonus buffer. This is because 210 * we always put the bonus buffer after the last block 211 * pointer (instead of packing it against the end of the 212 * dnode buffer). 213 */ 214 int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t); 215 size_t len = DN_MAX_BONUSLEN - off; 216 dmu_ot[dnp->dn_bonustype].ot_byteswap(dnp->dn_bonus + off, len); 217 } 218 } 219 220 void 221 dnode_buf_byteswap(void *vbuf, size_t size) 222 { 223 dnode_phys_t *buf = vbuf; 224 int i; 225 226 ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT)); 227 ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0); 228 229 size >>= DNODE_SHIFT; 230 for (i = 0; i < size; i++) { 231 dnode_byteswap(buf); 232 buf++; 233 } 234 } 235 236 static int 237 free_range_compar(const void *node1, const void *node2) 238 { 239 const free_range_t *rp1 = node1; 240 const free_range_t *rp2 = node2; 241 242 if (rp1->fr_blkid < rp2->fr_blkid) 243 return (-1); 244 else if (rp1->fr_blkid > rp2->fr_blkid) 245 return (1); 246 else return (0); 247 } 248 249 static void 250 dnode_setdblksz(dnode_t *dn, int size) 251 { 252 ASSERT3U(P2PHASE(size, SPA_MINBLOCKSIZE), ==, 0); 253 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 254 ASSERT3U(size, >=, SPA_MINBLOCKSIZE); 255 ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, 256 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); 257 dn->dn_datablksz = size; 258 dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; 259 dn->dn_datablkshift = ISP2(size) ? highbit(size - 1) : 0; 260 } 261 262 static dnode_t * 263 dnode_create(objset_impl_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, 264 uint64_t object) 265 { 266 dnode_t *dn = kmem_cache_alloc(dnode_cache, KM_SLEEP); 267 (void) dnode_cons(dn, NULL, 0); /* XXX */ 268 269 dn->dn_objset = os; 270 dn->dn_object = object; 271 dn->dn_dbuf = db; 272 dn->dn_phys = dnp; 273 274 if (dnp->dn_datablkszsec) 275 dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 276 dn->dn_indblkshift = dnp->dn_indblkshift; 277 dn->dn_nlevels = dnp->dn_nlevels; 278 dn->dn_type = dnp->dn_type; 279 dn->dn_nblkptr = dnp->dn_nblkptr; 280 dn->dn_checksum = dnp->dn_checksum; 281 dn->dn_compress = dnp->dn_compress; 282 dn->dn_bonustype = dnp->dn_bonustype; 283 dn->dn_bonuslen = dnp->dn_bonuslen; 284 dn->dn_maxblkid = dnp->dn_maxblkid; 285 286 dmu_zfetch_init(&dn->dn_zfetch, dn); 287 288 ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); 289 mutex_enter(&os->os_lock); 290 list_insert_head(&os->os_dnodes, dn); 291 mutex_exit(&os->os_lock); 292 293 return (dn); 294 } 295 296 static void 297 dnode_destroy(dnode_t *dn) 298 { 299 objset_impl_t *os = dn->dn_objset; 300 301 mutex_enter(&os->os_lock); 302 list_remove(&os->os_dnodes, dn); 303 mutex_exit(&os->os_lock); 304 305 if (dn->dn_dirtyctx_firstset) { 306 kmem_free(dn->dn_dirtyctx_firstset, 1); 307 dn->dn_dirtyctx_firstset = NULL; 308 } 309 dmu_zfetch_rele(&dn->dn_zfetch); 310 if (dn->dn_bonus) { 311 mutex_enter(&dn->dn_bonus->db_mtx); 312 dbuf_evict(dn->dn_bonus); 313 dn->dn_bonus = NULL; 314 } 315 kmem_cache_free(dnode_cache, dn); 316 } 317 318 void 319 dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 320 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 321 { 322 int i; 323 324 if (blocksize == 0) 325 blocksize = 1 << zfs_default_bs; 326 else if (blocksize > SPA_MAXBLOCKSIZE) 327 blocksize = SPA_MAXBLOCKSIZE; 328 else 329 blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); 330 331 if (ibs == 0) 332 ibs = zfs_default_ibs; 333 334 ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); 335 336 dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset, 337 dn->dn_object, tx->tx_txg, blocksize, ibs); 338 339 ASSERT(dn->dn_type == DMU_OT_NONE); 340 ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); 341 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); 342 ASSERT(ot != DMU_OT_NONE); 343 ASSERT3U(ot, <, DMU_OT_NUMTYPES); 344 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 345 (bonustype != DMU_OT_NONE && bonuslen != 0)); 346 ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 347 ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 348 ASSERT(dn->dn_type == DMU_OT_NONE); 349 ASSERT3U(dn->dn_maxblkid, ==, 0); 350 ASSERT3U(dn->dn_allocated_txg, ==, 0); 351 ASSERT3U(dn->dn_assigned_txg, ==, 0); 352 ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 353 ASSERT3U(refcount_count(&dn->dn_holds), <=, 1); 354 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 355 356 for (i = 0; i < TXG_SIZE; i++) { 357 ASSERT3U(dn->dn_next_nlevels[i], ==, 0); 358 ASSERT3U(dn->dn_next_indblkshift[i], ==, 0); 359 ASSERT3U(dn->dn_next_blksz[i], ==, 0); 360 ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 361 ASSERT3P(list_head(&dn->dn_dirty_dbufs[i]), ==, NULL); 362 ASSERT3U(avl_numnodes(&dn->dn_ranges[i]), ==, 0); 363 } 364 365 dn->dn_type = ot; 366 dnode_setdblksz(dn, blocksize); 367 dn->dn_indblkshift = ibs; 368 dn->dn_nlevels = 1; 369 dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 370 dn->dn_bonustype = bonustype; 371 dn->dn_bonuslen = bonuslen; 372 dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 373 dn->dn_compress = ZIO_COMPRESS_INHERIT; 374 dn->dn_dirtyctx = 0; 375 376 dn->dn_free_txg = 0; 377 if (dn->dn_dirtyctx_firstset) { 378 kmem_free(dn->dn_dirtyctx_firstset, 1); 379 dn->dn_dirtyctx_firstset = NULL; 380 } 381 382 dn->dn_allocated_txg = tx->tx_txg; 383 384 dnode_setdirty(dn, tx); 385 dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs; 386 dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz; 387 } 388 389 void 390 dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 391 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 392 { 393 int i; 394 395 ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); 396 ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE); 397 ASSERT3U(blocksize % SPA_MINBLOCKSIZE, ==, 0); 398 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); 399 ASSERT(tx->tx_txg != 0); 400 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 401 (bonustype != DMU_OT_NONE && bonuslen != 0)); 402 ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 403 ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 404 405 for (i = 0; i < TXG_SIZE; i++) 406 ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 407 408 /* clean up any unreferenced dbufs */ 409 (void) dnode_evict_dbufs(dn, 0); 410 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 411 412 /* 413 * XXX I should really have a generation number to tell if we 414 * need to do this... 415 */ 416 if (blocksize != dn->dn_datablksz || 417 dn->dn_bonustype != bonustype || dn->dn_bonuslen != bonuslen) { 418 /* free all old data */ 419 dnode_free_range(dn, 0, -1ULL, tx); 420 } 421 422 /* change blocksize */ 423 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 424 dnode_setdblksz(dn, blocksize); 425 dnode_setdirty(dn, tx); 426 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize; 427 rw_exit(&dn->dn_struct_rwlock); 428 429 /* change type */ 430 dn->dn_type = ot; 431 432 if (dn->dn_bonuslen != bonuslen) { 433 dmu_buf_impl_t *db = NULL; 434 435 /* change bonus size */ 436 if (bonuslen == 0) 437 bonuslen = 1; /* XXX */ 438 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 439 if (dn->dn_bonus == NULL) 440 dn->dn_bonus = dbuf_create_bonus(dn); 441 db = dn->dn_bonus; 442 rw_exit(&dn->dn_struct_rwlock); 443 if (refcount_add(&db->db_holds, FTAG) == 1) 444 dnode_add_ref(dn, db); 445 mutex_enter(&db->db_mtx); 446 ASSERT3U(db->db.db_size, ==, dn->dn_bonuslen); 447 ASSERT(db->db.db_data != NULL); 448 db->db.db_size = bonuslen; 449 mutex_exit(&db->db_mtx); 450 dbuf_dirty(db, tx); 451 dbuf_rele(db, FTAG); 452 } 453 454 /* change bonus size and type */ 455 mutex_enter(&dn->dn_mtx); 456 dn->dn_bonustype = bonustype; 457 dn->dn_bonuslen = bonuslen; 458 dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 459 dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 460 dn->dn_compress = ZIO_COMPRESS_INHERIT; 461 ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 462 463 dn->dn_allocated_txg = tx->tx_txg; 464 mutex_exit(&dn->dn_mtx); 465 } 466 467 void 468 dnode_special_close(dnode_t *dn) 469 { 470 /* 471 * Wait for final references to the dnode to clear. This can 472 * only happen if the arc is asyncronously evicting state that 473 * has a hold on this dnode while we are trying to evict this 474 * dnode. 475 */ 476 while (refcount_count(&dn->dn_holds) > 0) 477 delay(1); 478 dnode_destroy(dn); 479 } 480 481 dnode_t * 482 dnode_special_open(objset_impl_t *os, dnode_phys_t *dnp, uint64_t object) 483 { 484 dnode_t *dn = dnode_create(os, dnp, NULL, object); 485 DNODE_VERIFY(dn); 486 return (dn); 487 } 488 489 static void 490 dnode_buf_pageout(dmu_buf_t *db, void *arg) 491 { 492 dnode_t **children_dnodes = arg; 493 int i; 494 int epb = db->db_size >> DNODE_SHIFT; 495 496 for (i = 0; i < epb; i++) { 497 dnode_t *dn = children_dnodes[i]; 498 int n; 499 500 if (dn == NULL) 501 continue; 502 #ifdef ZFS_DEBUG 503 /* 504 * If there are holds on this dnode, then there should 505 * be holds on the dnode's containing dbuf as well; thus 506 * it wouldn't be eligable for eviction and this function 507 * would not have been called. 508 */ 509 ASSERT(refcount_is_zero(&dn->dn_holds)); 510 ASSERT(list_head(&dn->dn_dbufs) == NULL); 511 ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 512 513 for (n = 0; n < TXG_SIZE; n++) 514 ASSERT(!list_link_active(&dn->dn_dirty_link[n])); 515 #endif 516 children_dnodes[i] = NULL; 517 dnode_destroy(dn); 518 } 519 kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 520 } 521 522 /* 523 * errors: 524 * EINVAL - invalid object number. 525 * EIO - i/o error. 526 * succeeds even for free dnodes. 527 */ 528 int 529 dnode_hold_impl(objset_impl_t *os, uint64_t object, int flag, 530 void *tag, dnode_t **dnp) 531 { 532 int epb, idx, err; 533 int drop_struct_lock = FALSE; 534 int type; 535 uint64_t blk; 536 dnode_t *mdn, *dn; 537 dmu_buf_impl_t *db; 538 dnode_t **children_dnodes; 539 540 if (object == 0 || object >= DN_MAX_OBJECT) 541 return (EINVAL); 542 543 mdn = os->os_meta_dnode; 544 545 DNODE_VERIFY(mdn); 546 547 if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { 548 rw_enter(&mdn->dn_struct_rwlock, RW_READER); 549 drop_struct_lock = TRUE; 550 } 551 552 blk = dbuf_whichblock(mdn, object * sizeof (dnode_phys_t)); 553 554 db = dbuf_hold(mdn, blk, FTAG); 555 if (drop_struct_lock) 556 rw_exit(&mdn->dn_struct_rwlock); 557 if (db == NULL) 558 return (EIO); 559 err = dbuf_read(db, NULL, DB_RF_CANFAIL); 560 if (err) { 561 dbuf_rele(db, FTAG); 562 return (err); 563 } 564 565 ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT); 566 epb = db->db.db_size >> DNODE_SHIFT; 567 568 idx = object & (epb-1); 569 570 children_dnodes = dmu_buf_get_user(&db->db); 571 if (children_dnodes == NULL) { 572 dnode_t **winner; 573 children_dnodes = kmem_zalloc(epb * sizeof (dnode_t *), 574 KM_SLEEP); 575 if (winner = dmu_buf_set_user(&db->db, children_dnodes, NULL, 576 dnode_buf_pageout)) { 577 kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 578 children_dnodes = winner; 579 } 580 } 581 582 if ((dn = children_dnodes[idx]) == NULL) { 583 dnode_t *winner; 584 dn = dnode_create(os, (dnode_phys_t *)db->db.db_data+idx, 585 db, object); 586 winner = atomic_cas_ptr(&children_dnodes[idx], NULL, dn); 587 if (winner != NULL) { 588 dnode_destroy(dn); 589 dn = winner; 590 } 591 } 592 593 mutex_enter(&dn->dn_mtx); 594 type = dn->dn_type; 595 if (dn->dn_free_txg || 596 ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) || 597 ((flag & DNODE_MUST_BE_FREE) && type != DMU_OT_NONE)) { 598 mutex_exit(&dn->dn_mtx); 599 dbuf_rele(db, FTAG); 600 return (type == DMU_OT_NONE ? ENOENT : EEXIST); 601 } 602 mutex_exit(&dn->dn_mtx); 603 604 if (refcount_add(&dn->dn_holds, tag) == 1) 605 dbuf_add_ref(db, dn); 606 607 DNODE_VERIFY(dn); 608 ASSERT3P(dn->dn_dbuf, ==, db); 609 ASSERT3U(dn->dn_object, ==, object); 610 dbuf_rele(db, FTAG); 611 612 *dnp = dn; 613 return (0); 614 } 615 616 /* 617 * Return held dnode if the object is allocated, NULL if not. 618 */ 619 int 620 dnode_hold(objset_impl_t *os, uint64_t object, void *tag, dnode_t **dnp) 621 { 622 return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp)); 623 } 624 625 void 626 dnode_add_ref(dnode_t *dn, void *tag) 627 { 628 ASSERT(refcount_count(&dn->dn_holds) > 0); 629 (void) refcount_add(&dn->dn_holds, tag); 630 } 631 632 void 633 dnode_rele(dnode_t *dn, void *tag) 634 { 635 uint64_t refs; 636 637 refs = refcount_remove(&dn->dn_holds, tag); 638 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ 639 if (refs == 0 && dn->dn_dbuf) 640 dbuf_rele(dn->dn_dbuf, dn); 641 } 642 643 void 644 dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) 645 { 646 objset_impl_t *os = dn->dn_objset; 647 uint64_t txg = tx->tx_txg; 648 649 if (dn->dn_object == DMU_META_DNODE_OBJECT) 650 return; 651 652 DNODE_VERIFY(dn); 653 654 #ifdef ZFS_DEBUG 655 mutex_enter(&dn->dn_mtx); 656 ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); 657 /* ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); */ 658 mutex_exit(&dn->dn_mtx); 659 #endif 660 661 mutex_enter(&os->os_lock); 662 663 /* 664 * If we are already marked dirty, we're done. 665 */ 666 if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) { 667 mutex_exit(&os->os_lock); 668 return; 669 } 670 671 ASSERT(!refcount_is_zero(&dn->dn_holds) || list_head(&dn->dn_dbufs)); 672 ASSERT(dn->dn_datablksz != 0); 673 ASSERT3U(dn->dn_next_blksz[txg&TXG_MASK], ==, 0); 674 675 dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", 676 dn->dn_object, txg); 677 678 if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) { 679 list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn); 680 } else { 681 list_insert_tail(&os->os_dirty_dnodes[txg&TXG_MASK], dn); 682 } 683 684 mutex_exit(&os->os_lock); 685 686 /* 687 * The dnode maintains a hold on its containing dbuf as 688 * long as there are holds on it. Each instantiated child 689 * dbuf maintaines a hold on the dnode. When the last child 690 * drops its hold, the dnode will drop its hold on the 691 * containing dbuf. We add a "dirty hold" here so that the 692 * dnode will hang around after we finish processing its 693 * children. 694 */ 695 dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg); 696 697 dbuf_dirty(dn->dn_dbuf, tx); 698 699 dsl_dataset_dirty(os->os_dsl_dataset, tx); 700 } 701 702 void 703 dnode_free(dnode_t *dn, dmu_tx_t *tx) 704 { 705 int txgoff = tx->tx_txg & TXG_MASK; 706 707 dprintf("dn=%p txg=%llu\n", dn, tx->tx_txg); 708 709 /* we should be the only holder... hopefully */ 710 /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ 711 712 mutex_enter(&dn->dn_mtx); 713 if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { 714 mutex_exit(&dn->dn_mtx); 715 return; 716 } 717 dn->dn_free_txg = tx->tx_txg; 718 mutex_exit(&dn->dn_mtx); 719 720 /* 721 * If the dnode is already dirty, it needs to be moved from 722 * the dirty list to the free list. 723 */ 724 mutex_enter(&dn->dn_objset->os_lock); 725 if (list_link_active(&dn->dn_dirty_link[txgoff])) { 726 list_remove(&dn->dn_objset->os_dirty_dnodes[txgoff], dn); 727 list_insert_tail(&dn->dn_objset->os_free_dnodes[txgoff], dn); 728 mutex_exit(&dn->dn_objset->os_lock); 729 } else { 730 mutex_exit(&dn->dn_objset->os_lock); 731 dnode_setdirty(dn, tx); 732 } 733 } 734 735 /* 736 * Try to change the block size for the indicated dnode. This can only 737 * succeed if there are no blocks allocated or dirty beyond first block 738 */ 739 int 740 dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) 741 { 742 dmu_buf_impl_t *db, *db_next; 743 int have_db0 = FALSE; 744 int err = ENOTSUP; 745 746 if (size == 0) 747 size = SPA_MINBLOCKSIZE; 748 if (size > SPA_MAXBLOCKSIZE) 749 size = SPA_MAXBLOCKSIZE; 750 else 751 size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); 752 753 if (ibs == 0) 754 ibs = dn->dn_indblkshift; 755 756 if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && 757 ibs == dn->dn_indblkshift) 758 return (0); 759 760 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 761 762 /* Check for any allocated blocks beyond the first */ 763 if (dn->dn_phys->dn_maxblkid != 0) 764 goto end; 765 766 mutex_enter(&dn->dn_dbufs_mtx); 767 for (db = list_head(&dn->dn_dbufs); db; db = db_next) { 768 db_next = list_next(&dn->dn_dbufs, db); 769 770 if (db->db_blkid == 0) { 771 have_db0 = TRUE; 772 } else if (db->db_blkid != DB_BONUS_BLKID) { 773 mutex_exit(&dn->dn_dbufs_mtx); 774 goto end; 775 } 776 } 777 mutex_exit(&dn->dn_dbufs_mtx); 778 779 db = NULL; 780 if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || have_db0) { 781 /* obtain the old block */ 782 db = dbuf_hold(dn, 0, FTAG); 783 dbuf_new_size(db, size, tx); 784 } 785 786 dnode_setdblksz(dn, size); 787 dn->dn_indblkshift = ibs; 788 dnode_setdirty(dn, tx); 789 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size; 790 dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; 791 792 if (db) 793 dbuf_rele(db, FTAG); 794 795 err = 0; 796 end: 797 rw_exit(&dn->dn_struct_rwlock); 798 return (err); 799 } 800 801 uint64_t 802 dnode_max_nonzero_offset(dnode_t *dn) 803 { 804 if (dn->dn_phys->dn_maxblkid == 0 && 805 BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0])) 806 return (0); 807 else 808 return ((dn->dn_phys->dn_maxblkid+1) * dn->dn_datablksz); 809 } 810 811 void 812 dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx) 813 { 814 uint64_t txgoff = tx->tx_txg & TXG_MASK; 815 int drop_struct_lock = FALSE; 816 int epbs, new_nlevels; 817 uint64_t sz; 818 819 ASSERT(blkid != DB_BONUS_BLKID); 820 821 if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { 822 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 823 drop_struct_lock = TRUE; 824 } 825 826 if (blkid <= dn->dn_maxblkid) 827 goto out; 828 829 dn->dn_maxblkid = blkid; 830 831 /* 832 * Compute the number of levels necessary to support the new maxblkid. 833 */ 834 new_nlevels = 1; 835 epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 836 for (sz = dn->dn_nblkptr; 837 sz <= blkid && sz >= dn->dn_nblkptr; sz <<= epbs) 838 new_nlevels++; 839 840 if (new_nlevels > dn->dn_nlevels) { 841 int old_nlevels = dn->dn_nlevels; 842 dmu_buf_impl_t *db; 843 844 dn->dn_nlevels = new_nlevels; 845 846 ASSERT3U(new_nlevels, >, dn->dn_next_nlevels[txgoff]); 847 dn->dn_next_nlevels[txgoff] = new_nlevels; 848 849 /* Dirty the left indirects. */ 850 db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); 851 dbuf_dirty(db, tx); 852 dbuf_rele(db, FTAG); 853 854 } 855 856 out: 857 if (drop_struct_lock) 858 rw_exit(&dn->dn_struct_rwlock); 859 } 860 861 void 862 dnode_clear_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) 863 { 864 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 865 avl_index_t where; 866 free_range_t *rp; 867 free_range_t rp_tofind; 868 uint64_t endblk = blkid + nblks; 869 870 ASSERT(MUTEX_HELD(&dn->dn_mtx)); 871 ASSERT(nblks <= UINT64_MAX - blkid); /* no overflow */ 872 873 dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 874 blkid, nblks, tx->tx_txg); 875 rp_tofind.fr_blkid = blkid; 876 rp = avl_find(tree, &rp_tofind, &where); 877 if (rp == NULL) 878 rp = avl_nearest(tree, where, AVL_BEFORE); 879 if (rp == NULL) 880 rp = avl_nearest(tree, where, AVL_AFTER); 881 882 while (rp && (rp->fr_blkid <= blkid + nblks)) { 883 uint64_t fr_endblk = rp->fr_blkid + rp->fr_nblks; 884 free_range_t *nrp = AVL_NEXT(tree, rp); 885 886 if (blkid <= rp->fr_blkid && endblk >= fr_endblk) { 887 /* clear this entire range */ 888 avl_remove(tree, rp); 889 kmem_free(rp, sizeof (free_range_t)); 890 } else if (blkid <= rp->fr_blkid && 891 endblk > rp->fr_blkid && endblk < fr_endblk) { 892 /* clear the beginning of this range */ 893 rp->fr_blkid = endblk; 894 rp->fr_nblks = fr_endblk - endblk; 895 } else if (blkid > rp->fr_blkid && blkid < fr_endblk && 896 endblk >= fr_endblk) { 897 /* clear the end of this range */ 898 rp->fr_nblks = blkid - rp->fr_blkid; 899 } else if (blkid > rp->fr_blkid && endblk < fr_endblk) { 900 /* clear a chunk out of this range */ 901 free_range_t *new_rp = 902 kmem_alloc(sizeof (free_range_t), KM_SLEEP); 903 904 new_rp->fr_blkid = endblk; 905 new_rp->fr_nblks = fr_endblk - endblk; 906 avl_insert_here(tree, new_rp, rp, AVL_AFTER); 907 rp->fr_nblks = blkid - rp->fr_blkid; 908 } 909 /* there may be no overlap */ 910 rp = nrp; 911 } 912 } 913 914 void 915 dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) 916 { 917 dmu_buf_impl_t *db; 918 uint64_t start, objsize, blkid, nblks; 919 int blkshift, blksz, tail, head, epbs; 920 int trunc = FALSE; 921 922 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 923 blksz = dn->dn_datablksz; 924 blkshift = dn->dn_datablkshift; 925 epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 926 927 /* If the range is past the end of the file, this is a no-op */ 928 objsize = blksz * (dn->dn_maxblkid+1); 929 if (off >= objsize) 930 goto out; 931 if (len == -1ULL) { 932 len = UINT64_MAX - off; 933 trunc = TRUE; 934 } 935 936 /* 937 * First, block align the region to free: 938 */ 939 if (dn->dn_maxblkid == 0) { 940 if (off == 0) { 941 head = 0; 942 } else { 943 head = blksz - off; 944 ASSERT3U(head, >, 0); 945 } 946 start = off; 947 } else { 948 ASSERT(ISP2(blksz)); 949 head = P2NPHASE(off, blksz); 950 start = P2PHASE(off, blksz); 951 } 952 /* zero out any partial block data at the start of the range */ 953 if (head) { 954 ASSERT3U(start + head, ==, blksz); 955 if (len < head) 956 head = len; 957 if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off), TRUE, 958 FTAG, &db) == 0) { 959 caddr_t data; 960 961 /* don't dirty if it isn't on disk and isn't dirty */ 962 if (db->db_dirtied || 963 (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { 964 rw_exit(&dn->dn_struct_rwlock); 965 dbuf_will_dirty(db, tx); 966 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 967 data = db->db.db_data; 968 bzero(data + start, head); 969 } 970 dbuf_rele(db, FTAG); 971 } 972 off += head; 973 len -= head; 974 } 975 /* If the range was less than one block, we are done */ 976 if (len == 0) 977 goto out; 978 979 /* If the remaining range is past the end of the file, we are done */ 980 if (off > dn->dn_maxblkid << blkshift) 981 goto out; 982 983 if (off + len == UINT64_MAX) 984 tail = 0; 985 else 986 tail = P2PHASE(len, blksz); 987 988 ASSERT3U(P2PHASE(off, blksz), ==, 0); 989 /* zero out any partial block data at the end of the range */ 990 if (tail) { 991 if (len < tail) 992 tail = len; 993 if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len), 994 TRUE, FTAG, &db) == 0) { 995 /* don't dirty if it isn't on disk and isn't dirty */ 996 if (db->db_dirtied || 997 (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { 998 rw_exit(&dn->dn_struct_rwlock); 999 dbuf_will_dirty(db, tx); 1000 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 1001 bzero(db->db.db_data, tail); 1002 } 1003 dbuf_rele(db, FTAG); 1004 } 1005 len -= tail; 1006 } 1007 /* If the range did not include a full block, we are done */ 1008 if (len == 0) 1009 goto out; 1010 1011 /* dirty the left indirects */ 1012 if (dn->dn_nlevels > 1 && off != 0) { 1013 db = dbuf_hold_level(dn, 1, 1014 (off - head) >> (blkshift + epbs), FTAG); 1015 dbuf_will_dirty(db, tx); 1016 dbuf_rele(db, FTAG); 1017 } 1018 1019 /* dirty the right indirects */ 1020 if (dn->dn_nlevels > 1 && !trunc) { 1021 db = dbuf_hold_level(dn, 1, 1022 (off + len + tail - 1) >> (blkshift + epbs), FTAG); 1023 dbuf_will_dirty(db, tx); 1024 dbuf_rele(db, FTAG); 1025 } 1026 1027 /* 1028 * Finally, add this range to the dnode range list, we 1029 * will finish up this free operation in the syncing phase. 1030 */ 1031 ASSERT(IS_P2ALIGNED(off, 1<<blkshift)); 1032 ASSERT(off + len == UINT64_MAX || IS_P2ALIGNED(len, 1<<blkshift)); 1033 blkid = off >> blkshift; 1034 nblks = len >> blkshift; 1035 1036 if (trunc) 1037 dn->dn_maxblkid = (blkid ? blkid - 1 : 0); 1038 1039 mutex_enter(&dn->dn_mtx); 1040 dnode_clear_range(dn, blkid, nblks, tx); 1041 { 1042 free_range_t *rp, *found; 1043 avl_index_t where; 1044 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 1045 1046 /* Add new range to dn_ranges */ 1047 rp = kmem_alloc(sizeof (free_range_t), KM_SLEEP); 1048 rp->fr_blkid = blkid; 1049 rp->fr_nblks = nblks; 1050 found = avl_find(tree, rp, &where); 1051 ASSERT(found == NULL); 1052 avl_insert(tree, rp, where); 1053 dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 1054 blkid, nblks, tx->tx_txg); 1055 } 1056 mutex_exit(&dn->dn_mtx); 1057 1058 dbuf_free_range(dn, blkid, nblks, tx); 1059 dnode_setdirty(dn, tx); 1060 out: 1061 rw_exit(&dn->dn_struct_rwlock); 1062 } 1063 1064 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */ 1065 uint64_t 1066 dnode_block_freed(dnode_t *dn, uint64_t blkid) 1067 { 1068 free_range_t range_tofind; 1069 void *dp = spa_get_dsl(dn->dn_objset->os_spa); 1070 int i; 1071 1072 if (blkid == DB_BONUS_BLKID) 1073 return (FALSE); 1074 1075 /* 1076 * If we're in the process of opening the pool, dp will not be 1077 * set yet, but there shouldn't be anything dirty. 1078 */ 1079 if (dp == NULL) 1080 return (FALSE); 1081 1082 if (dn->dn_free_txg) 1083 return (TRUE); 1084 1085 /* 1086 * If dn_datablkshift is not set, then there's only a single 1087 * block, in which case there will never be a free range so it 1088 * won't matter. 1089 */ 1090 range_tofind.fr_blkid = blkid; 1091 mutex_enter(&dn->dn_mtx); 1092 for (i = 0; i < TXG_SIZE; i++) { 1093 free_range_t *range_found; 1094 avl_index_t idx; 1095 1096 range_found = avl_find(&dn->dn_ranges[i], &range_tofind, &idx); 1097 if (range_found) { 1098 ASSERT(range_found->fr_nblks > 0); 1099 break; 1100 } 1101 range_found = avl_nearest(&dn->dn_ranges[i], idx, AVL_BEFORE); 1102 if (range_found && 1103 range_found->fr_blkid + range_found->fr_nblks > blkid) 1104 break; 1105 } 1106 mutex_exit(&dn->dn_mtx); 1107 return (i < TXG_SIZE); 1108 } 1109 1110 /* call from syncing context when we actually write/free space for this dnode */ 1111 void 1112 dnode_diduse_space(dnode_t *dn, int64_t space) 1113 { 1114 uint64_t sectors; 1115 1116 dprintf_dnode(dn, "dn=%p dnp=%p secphys=%llu space=%lld\n", 1117 dn, dn->dn_phys, 1118 (u_longlong_t)dn->dn_phys->dn_secphys, 1119 (longlong_t)space); 1120 1121 ASSERT(P2PHASE(space, 1<<DEV_BSHIFT) == 0); 1122 1123 mutex_enter(&dn->dn_mtx); 1124 if (space > 0) { 1125 sectors = space >> DEV_BSHIFT; 1126 ASSERT3U(dn->dn_phys->dn_secphys + sectors, >=, 1127 dn->dn_phys->dn_secphys); 1128 dn->dn_phys->dn_secphys += sectors; 1129 } else { 1130 sectors = -space >> DEV_BSHIFT; 1131 ASSERT3U(dn->dn_phys->dn_secphys, >=, sectors); 1132 dn->dn_phys->dn_secphys -= sectors; 1133 } 1134 mutex_exit(&dn->dn_mtx); 1135 } 1136 1137 /* 1138 * Call when we think we're going to write/free space in open context. 1139 * Be conservative (ie. OK to write less than this or free more than 1140 * this, but don't write more or free less). 1141 */ 1142 void 1143 dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx) 1144 { 1145 objset_impl_t *os = dn->dn_objset; 1146 dsl_dataset_t *ds = os->os_dsl_dataset; 1147 1148 if (space > 0) 1149 space = spa_get_asize(os->os_spa, space); 1150 1151 if (ds) 1152 dsl_dir_willuse_space(ds->ds_dir, space, tx); 1153 1154 dmu_tx_willuse_space(tx, space); 1155 } 1156 1157 static int 1158 dnode_next_offset_level(dnode_t *dn, boolean_t hole, uint64_t *offset, 1159 int lvl, uint64_t blkfill) 1160 { 1161 dmu_buf_impl_t *db = NULL; 1162 void *data = NULL; 1163 uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 1164 uint64_t epb = 1ULL << epbs; 1165 uint64_t minfill, maxfill; 1166 int i, error, span; 1167 1168 dprintf("probing object %llu offset %llx level %d of %u\n", 1169 dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); 1170 1171 if (lvl == dn->dn_phys->dn_nlevels) { 1172 error = 0; 1173 epb = dn->dn_phys->dn_nblkptr; 1174 data = dn->dn_phys->dn_blkptr; 1175 } else { 1176 uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl); 1177 error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db); 1178 if (error) { 1179 if (error == ENOENT) 1180 return (hole ? 0 : ESRCH); 1181 return (error); 1182 } 1183 error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT); 1184 if (error) { 1185 dbuf_rele(db, FTAG); 1186 return (error); 1187 } 1188 data = db->db.db_data; 1189 } 1190 1191 if (lvl == 0) { 1192 dnode_phys_t *dnp = data; 1193 span = DNODE_SHIFT; 1194 ASSERT(dn->dn_type == DMU_OT_DNODE); 1195 1196 for (i = (*offset >> span) & (blkfill - 1); i < blkfill; i++) { 1197 if (!dnp[i].dn_type == hole) 1198 break; 1199 *offset += 1ULL << span; 1200 } 1201 if (i == blkfill) 1202 error = ESRCH; 1203 } else { 1204 blkptr_t *bp = data; 1205 span = (lvl - 1) * epbs + dn->dn_datablkshift; 1206 minfill = 0; 1207 maxfill = blkfill << ((lvl - 1) * epbs); 1208 1209 if (hole) 1210 maxfill--; 1211 else 1212 minfill++; 1213 1214 for (i = (*offset >> span) & ((1ULL << epbs) - 1); 1215 i < epb; i++) { 1216 if (bp[i].blk_fill >= minfill && 1217 bp[i].blk_fill <= maxfill) 1218 break; 1219 *offset += 1ULL << span; 1220 } 1221 if (i >= epb) 1222 error = ESRCH; 1223 } 1224 1225 if (db) 1226 dbuf_rele(db, FTAG); 1227 1228 return (error); 1229 } 1230 1231 /* 1232 * Find the next hole, data, or sparse region at or after *offset. 1233 * The value 'blkfill' tells us how many items we expect to find 1234 * in an L0 data block; this value is 1 for normal objects, 1235 * DNODES_PER_BLOCK for the meta dnode, and some fraction of 1236 * DNODES_PER_BLOCK when searching for sparse regions thereof. 1237 * Examples: 1238 * 1239 * dnode_next_offset(dn, hole, offset, 1, 1); 1240 * Finds the next hole/data in a file. 1241 * Used in dmu_offset_next(). 1242 * 1243 * dnode_next_offset(mdn, hole, offset, 0, DNODES_PER_BLOCK); 1244 * Finds the next free/allocated dnode an objset's meta-dnode. 1245 * Used in dmu_object_next(). 1246 * 1247 * dnode_next_offset(mdn, TRUE, offset, 2, DNODES_PER_BLOCK >> 2); 1248 * Finds the next L2 meta-dnode bp that's at most 1/4 full. 1249 * Used in dmu_object_alloc(). 1250 */ 1251 int 1252 dnode_next_offset(dnode_t *dn, boolean_t hole, uint64_t *offset, 1253 int minlvl, uint64_t blkfill) 1254 { 1255 int lvl, maxlvl; 1256 int error = 0; 1257 uint64_t initial_offset = *offset; 1258 1259 rw_enter(&dn->dn_struct_rwlock, RW_READER); 1260 1261 if (dn->dn_phys->dn_nlevels == 0) { 1262 rw_exit(&dn->dn_struct_rwlock); 1263 return (ESRCH); 1264 } 1265 1266 if (dn->dn_datablkshift == 0) { 1267 if (*offset < dn->dn_datablksz) { 1268 if (hole) 1269 *offset = dn->dn_datablksz; 1270 } else { 1271 error = ESRCH; 1272 } 1273 rw_exit(&dn->dn_struct_rwlock); 1274 return (error); 1275 } 1276 1277 maxlvl = dn->dn_phys->dn_nlevels; 1278 1279 for (lvl = minlvl; lvl <= maxlvl; lvl++) { 1280 error = dnode_next_offset_level(dn, hole, offset, lvl, blkfill); 1281 if (error != ESRCH) 1282 break; 1283 } 1284 1285 while (--lvl >= minlvl && error == 0) 1286 error = dnode_next_offset_level(dn, hole, offset, lvl, blkfill); 1287 1288 rw_exit(&dn->dn_struct_rwlock); 1289 1290 if (error == 0 && initial_offset > *offset) 1291 error = ESRCH; 1292 1293 return (error); 1294 } 1295