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_datablkszsec = BSWAP_16(dnp->dn_datablkszsec); 181 dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen); 182 dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid); 183 dnp->dn_used = BSWAP_64(dnp->dn_used); 184 185 /* 186 * dn_nblkptr is only one byte, so it's OK to read it in either 187 * byte order. We can't read dn_bouslen. 188 */ 189 ASSERT(dnp->dn_indblkshift <= SPA_MAXBLOCKSHIFT); 190 ASSERT(dnp->dn_nblkptr <= DN_MAX_NBLKPTR); 191 for (i = 0; i < dnp->dn_nblkptr * sizeof (blkptr_t)/8; i++) 192 buf64[i] = BSWAP_64(buf64[i]); 193 194 /* 195 * OK to check dn_bonuslen for zero, because it won't matter if 196 * we have the wrong byte order. This is necessary because the 197 * dnode dnode is smaller than a regular dnode. 198 */ 199 if (dnp->dn_bonuslen != 0) { 200 /* 201 * Note that the bonus length calculated here may be 202 * longer than the actual bonus buffer. This is because 203 * we always put the bonus buffer after the last block 204 * pointer (instead of packing it against the end of the 205 * dnode buffer). 206 */ 207 int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t); 208 size_t len = DN_MAX_BONUSLEN - off; 209 dmu_ot[dnp->dn_bonustype].ot_byteswap(dnp->dn_bonus + off, len); 210 } 211 } 212 213 void 214 dnode_buf_byteswap(void *vbuf, size_t size) 215 { 216 dnode_phys_t *buf = vbuf; 217 int i; 218 219 ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT)); 220 ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0); 221 222 size >>= DNODE_SHIFT; 223 for (i = 0; i < size; i++) { 224 dnode_byteswap(buf); 225 buf++; 226 } 227 } 228 229 static int 230 free_range_compar(const void *node1, const void *node2) 231 { 232 const free_range_t *rp1 = node1; 233 const free_range_t *rp2 = node2; 234 235 if (rp1->fr_blkid < rp2->fr_blkid) 236 return (-1); 237 else if (rp1->fr_blkid > rp2->fr_blkid) 238 return (1); 239 else return (0); 240 } 241 242 static void 243 dnode_setdblksz(dnode_t *dn, int size) 244 { 245 ASSERT3U(P2PHASE(size, SPA_MINBLOCKSIZE), ==, 0); 246 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 247 ASSERT3U(size, >=, SPA_MINBLOCKSIZE); 248 ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, 249 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); 250 dn->dn_datablksz = size; 251 dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; 252 dn->dn_datablkshift = ISP2(size) ? highbit(size - 1) : 0; 253 } 254 255 static dnode_t * 256 dnode_create(objset_impl_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, 257 uint64_t object) 258 { 259 dnode_t *dn = kmem_cache_alloc(dnode_cache, KM_SLEEP); 260 (void) dnode_cons(dn, NULL, 0); /* XXX */ 261 262 dn->dn_objset = os; 263 dn->dn_object = object; 264 dn->dn_dbuf = db; 265 dn->dn_phys = dnp; 266 267 if (dnp->dn_datablkszsec) 268 dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 269 dn->dn_indblkshift = dnp->dn_indblkshift; 270 dn->dn_nlevels = dnp->dn_nlevels; 271 dn->dn_type = dnp->dn_type; 272 dn->dn_nblkptr = dnp->dn_nblkptr; 273 dn->dn_checksum = dnp->dn_checksum; 274 dn->dn_compress = dnp->dn_compress; 275 dn->dn_bonustype = dnp->dn_bonustype; 276 dn->dn_bonuslen = dnp->dn_bonuslen; 277 dn->dn_maxblkid = dnp->dn_maxblkid; 278 279 dmu_zfetch_init(&dn->dn_zfetch, dn); 280 281 ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); 282 mutex_enter(&os->os_lock); 283 list_insert_head(&os->os_dnodes, dn); 284 mutex_exit(&os->os_lock); 285 286 return (dn); 287 } 288 289 static void 290 dnode_destroy(dnode_t *dn) 291 { 292 objset_impl_t *os = dn->dn_objset; 293 294 #ifdef ZFS_DEBUG 295 int i; 296 297 for (i = 0; i < TXG_SIZE; i++) { 298 ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 299 ASSERT(NULL == list_head(&dn->dn_dirty_dbufs[i])); 300 ASSERT(0 == avl_numnodes(&dn->dn_ranges[i])); 301 } 302 ASSERT(NULL == list_head(&dn->dn_dbufs)); 303 #endif 304 305 mutex_enter(&os->os_lock); 306 list_remove(&os->os_dnodes, dn); 307 mutex_exit(&os->os_lock); 308 309 if (dn->dn_dirtyctx_firstset) { 310 kmem_free(dn->dn_dirtyctx_firstset, 1); 311 dn->dn_dirtyctx_firstset = NULL; 312 } 313 dmu_zfetch_rele(&dn->dn_zfetch); 314 if (dn->dn_bonus) { 315 mutex_enter(&dn->dn_bonus->db_mtx); 316 dbuf_evict(dn->dn_bonus); 317 dn->dn_bonus = NULL; 318 } 319 kmem_cache_free(dnode_cache, dn); 320 } 321 322 void 323 dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 324 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 325 { 326 int i; 327 328 if (blocksize == 0) 329 blocksize = 1 << zfs_default_bs; 330 else if (blocksize > SPA_MAXBLOCKSIZE) 331 blocksize = SPA_MAXBLOCKSIZE; 332 else 333 blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); 334 335 if (ibs == 0) 336 ibs = zfs_default_ibs; 337 338 ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); 339 340 dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset, 341 dn->dn_object, tx->tx_txg, blocksize, ibs); 342 343 ASSERT(dn->dn_type == DMU_OT_NONE); 344 ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); 345 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); 346 ASSERT(ot != DMU_OT_NONE); 347 ASSERT3U(ot, <, DMU_OT_NUMTYPES); 348 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 349 (bonustype != DMU_OT_NONE && bonuslen != 0)); 350 ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 351 ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 352 ASSERT(dn->dn_type == DMU_OT_NONE); 353 ASSERT3U(dn->dn_maxblkid, ==, 0); 354 ASSERT3U(dn->dn_allocated_txg, ==, 0); 355 ASSERT3U(dn->dn_assigned_txg, ==, 0); 356 ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 357 ASSERT3U(refcount_count(&dn->dn_holds), <=, 1); 358 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 359 360 for (i = 0; i < TXG_SIZE; i++) { 361 ASSERT3U(dn->dn_next_nlevels[i], ==, 0); 362 ASSERT3U(dn->dn_next_indblkshift[i], ==, 0); 363 ASSERT3U(dn->dn_next_blksz[i], ==, 0); 364 ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 365 ASSERT3P(list_head(&dn->dn_dirty_dbufs[i]), ==, NULL); 366 ASSERT3U(avl_numnodes(&dn->dn_ranges[i]), ==, 0); 367 } 368 369 dn->dn_type = ot; 370 dnode_setdblksz(dn, blocksize); 371 dn->dn_indblkshift = ibs; 372 dn->dn_nlevels = 1; 373 dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 374 dn->dn_bonustype = bonustype; 375 dn->dn_bonuslen = bonuslen; 376 dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 377 dn->dn_compress = ZIO_COMPRESS_INHERIT; 378 dn->dn_dirtyctx = 0; 379 380 dn->dn_free_txg = 0; 381 if (dn->dn_dirtyctx_firstset) { 382 kmem_free(dn->dn_dirtyctx_firstset, 1); 383 dn->dn_dirtyctx_firstset = NULL; 384 } 385 386 dn->dn_allocated_txg = tx->tx_txg; 387 388 dnode_setdirty(dn, tx); 389 dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs; 390 dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz; 391 } 392 393 void 394 dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 395 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 396 { 397 int i; 398 dmu_buf_impl_t *db = NULL; 399 400 ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); 401 ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE); 402 ASSERT3U(blocksize % SPA_MINBLOCKSIZE, ==, 0); 403 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); 404 ASSERT(tx->tx_txg != 0); 405 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 406 (bonustype != DMU_OT_NONE && bonuslen != 0)); 407 ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 408 ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 409 410 for (i = 0; i < TXG_SIZE; i++) 411 ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 412 413 /* clean up any unreferenced dbufs */ 414 (void) dnode_evict_dbufs(dn, 0); 415 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 416 417 /* 418 * XXX I should really have a generation number to tell if we 419 * need to do this... 420 */ 421 if (blocksize != dn->dn_datablksz || 422 dn->dn_bonustype != bonustype || dn->dn_bonuslen != bonuslen) { 423 /* free all old data */ 424 dnode_free_range(dn, 0, -1ULL, tx); 425 } 426 427 /* change blocksize */ 428 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 429 if (blocksize != dn->dn_datablksz && 430 (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || 431 list_head(&dn->dn_dbufs) != NULL)) { 432 db = dbuf_hold(dn, 0, FTAG); 433 dbuf_new_size(db, blocksize, tx); 434 } 435 dnode_setdblksz(dn, blocksize); 436 dnode_setdirty(dn, tx); 437 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize; 438 rw_exit(&dn->dn_struct_rwlock); 439 if (db) { 440 dbuf_rele(db, FTAG); 441 db = NULL; 442 } 443 444 /* change type */ 445 dn->dn_type = ot; 446 447 if (dn->dn_bonuslen != bonuslen) { 448 /* change bonus size */ 449 if (bonuslen == 0) 450 bonuslen = 1; /* XXX */ 451 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 452 if (dn->dn_bonus == NULL) 453 dn->dn_bonus = dbuf_create_bonus(dn); 454 db = dn->dn_bonus; 455 rw_exit(&dn->dn_struct_rwlock); 456 if (refcount_add(&db->db_holds, FTAG) == 1) 457 dnode_add_ref(dn, db); 458 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED)); 459 mutex_enter(&db->db_mtx); 460 ASSERT3U(db->db.db_size, ==, dn->dn_bonuslen); 461 ASSERT(db->db.db_data != NULL); 462 db->db.db_size = bonuslen; 463 mutex_exit(&db->db_mtx); 464 dbuf_dirty(db, tx); 465 } 466 467 /* change bonus size and type */ 468 mutex_enter(&dn->dn_mtx); 469 dn->dn_bonustype = bonustype; 470 dn->dn_bonuslen = bonuslen; 471 dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 472 dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 473 dn->dn_compress = ZIO_COMPRESS_INHERIT; 474 ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 475 476 /* 477 * NB: we have to do the dbuf_rele after we've changed the 478 * dn_bonuslen, for the sake of dbuf_verify(). 479 */ 480 if (db) 481 dbuf_rele(db, FTAG); 482 483 dn->dn_allocated_txg = tx->tx_txg; 484 mutex_exit(&dn->dn_mtx); 485 } 486 487 void 488 dnode_special_close(dnode_t *dn) 489 { 490 /* 491 * Wait for final references to the dnode to clear. This can 492 * only happen if the arc is asyncronously evicting state that 493 * has a hold on this dnode while we are trying to evict this 494 * dnode. 495 */ 496 while (refcount_count(&dn->dn_holds) > 0) 497 delay(1); 498 dnode_destroy(dn); 499 } 500 501 dnode_t * 502 dnode_special_open(objset_impl_t *os, dnode_phys_t *dnp, uint64_t object) 503 { 504 dnode_t *dn = dnode_create(os, dnp, NULL, object); 505 DNODE_VERIFY(dn); 506 return (dn); 507 } 508 509 static void 510 dnode_buf_pageout(dmu_buf_t *db, void *arg) 511 { 512 dnode_t **children_dnodes = arg; 513 int i; 514 int epb = db->db_size >> DNODE_SHIFT; 515 516 for (i = 0; i < epb; i++) { 517 dnode_t *dn = children_dnodes[i]; 518 int n; 519 520 if (dn == NULL) 521 continue; 522 #ifdef ZFS_DEBUG 523 /* 524 * If there are holds on this dnode, then there should 525 * be holds on the dnode's containing dbuf as well; thus 526 * it wouldn't be eligable for eviction and this function 527 * would not have been called. 528 */ 529 ASSERT(refcount_is_zero(&dn->dn_holds)); 530 ASSERT(list_head(&dn->dn_dbufs) == NULL); 531 ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 532 533 for (n = 0; n < TXG_SIZE; n++) 534 ASSERT(!list_link_active(&dn->dn_dirty_link[n])); 535 #endif 536 children_dnodes[i] = NULL; 537 dnode_destroy(dn); 538 } 539 kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 540 } 541 542 /* 543 * errors: 544 * EINVAL - invalid object number. 545 * EIO - i/o error. 546 * succeeds even for free dnodes. 547 */ 548 int 549 dnode_hold_impl(objset_impl_t *os, uint64_t object, int flag, 550 void *tag, dnode_t **dnp) 551 { 552 int epb, idx, err; 553 int drop_struct_lock = FALSE; 554 int type; 555 uint64_t blk; 556 dnode_t *mdn, *dn; 557 dmu_buf_impl_t *db; 558 dnode_t **children_dnodes; 559 560 if (object == 0 || object >= DN_MAX_OBJECT) 561 return (EINVAL); 562 563 mdn = os->os_meta_dnode; 564 565 DNODE_VERIFY(mdn); 566 567 if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { 568 rw_enter(&mdn->dn_struct_rwlock, RW_READER); 569 drop_struct_lock = TRUE; 570 } 571 572 blk = dbuf_whichblock(mdn, object * sizeof (dnode_phys_t)); 573 574 db = dbuf_hold(mdn, blk, FTAG); 575 if (drop_struct_lock) 576 rw_exit(&mdn->dn_struct_rwlock); 577 if (db == NULL) 578 return (EIO); 579 err = dbuf_read(db, NULL, DB_RF_CANFAIL); 580 if (err) { 581 dbuf_rele(db, FTAG); 582 return (err); 583 } 584 585 ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT); 586 epb = db->db.db_size >> DNODE_SHIFT; 587 588 idx = object & (epb-1); 589 590 children_dnodes = dmu_buf_get_user(&db->db); 591 if (children_dnodes == NULL) { 592 dnode_t **winner; 593 children_dnodes = kmem_zalloc(epb * sizeof (dnode_t *), 594 KM_SLEEP); 595 if (winner = dmu_buf_set_user(&db->db, children_dnodes, NULL, 596 dnode_buf_pageout)) { 597 kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 598 children_dnodes = winner; 599 } 600 } 601 602 if ((dn = children_dnodes[idx]) == NULL) { 603 dnode_t *winner; 604 dn = dnode_create(os, (dnode_phys_t *)db->db.db_data+idx, 605 db, object); 606 winner = atomic_cas_ptr(&children_dnodes[idx], NULL, dn); 607 if (winner != NULL) { 608 dnode_destroy(dn); 609 dn = winner; 610 } 611 } 612 613 mutex_enter(&dn->dn_mtx); 614 type = dn->dn_type; 615 if (dn->dn_free_txg || 616 ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) || 617 ((flag & DNODE_MUST_BE_FREE) && type != DMU_OT_NONE)) { 618 mutex_exit(&dn->dn_mtx); 619 dbuf_rele(db, FTAG); 620 return (type == DMU_OT_NONE ? ENOENT : EEXIST); 621 } 622 mutex_exit(&dn->dn_mtx); 623 624 if (refcount_add(&dn->dn_holds, tag) == 1) 625 dbuf_add_ref(db, dn); 626 627 DNODE_VERIFY(dn); 628 ASSERT3P(dn->dn_dbuf, ==, db); 629 ASSERT3U(dn->dn_object, ==, object); 630 dbuf_rele(db, FTAG); 631 632 *dnp = dn; 633 return (0); 634 } 635 636 /* 637 * Return held dnode if the object is allocated, NULL if not. 638 */ 639 int 640 dnode_hold(objset_impl_t *os, uint64_t object, void *tag, dnode_t **dnp) 641 { 642 return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp)); 643 } 644 645 void 646 dnode_add_ref(dnode_t *dn, void *tag) 647 { 648 ASSERT(refcount_count(&dn->dn_holds) > 0); 649 (void) refcount_add(&dn->dn_holds, tag); 650 } 651 652 void 653 dnode_rele(dnode_t *dn, void *tag) 654 { 655 uint64_t refs; 656 657 refs = refcount_remove(&dn->dn_holds, tag); 658 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ 659 if (refs == 0 && dn->dn_dbuf) 660 dbuf_rele(dn->dn_dbuf, dn); 661 } 662 663 void 664 dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) 665 { 666 objset_impl_t *os = dn->dn_objset; 667 uint64_t txg = tx->tx_txg; 668 669 if (dn->dn_object == DMU_META_DNODE_OBJECT) 670 return; 671 672 DNODE_VERIFY(dn); 673 674 #ifdef ZFS_DEBUG 675 mutex_enter(&dn->dn_mtx); 676 ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); 677 /* ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); */ 678 mutex_exit(&dn->dn_mtx); 679 #endif 680 681 mutex_enter(&os->os_lock); 682 683 /* 684 * If we are already marked dirty, we're done. 685 */ 686 if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) { 687 mutex_exit(&os->os_lock); 688 return; 689 } 690 691 ASSERT(!refcount_is_zero(&dn->dn_holds) || list_head(&dn->dn_dbufs)); 692 ASSERT(dn->dn_datablksz != 0); 693 ASSERT3U(dn->dn_next_blksz[txg&TXG_MASK], ==, 0); 694 695 dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", 696 dn->dn_object, txg); 697 698 if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) { 699 list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn); 700 } else { 701 list_insert_tail(&os->os_dirty_dnodes[txg&TXG_MASK], dn); 702 } 703 704 mutex_exit(&os->os_lock); 705 706 /* 707 * The dnode maintains a hold on its containing dbuf as 708 * long as there are holds on it. Each instantiated child 709 * dbuf maintaines a hold on the dnode. When the last child 710 * drops its hold, the dnode will drop its hold on the 711 * containing dbuf. We add a "dirty hold" here so that the 712 * dnode will hang around after we finish processing its 713 * children. 714 */ 715 dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg); 716 717 dbuf_dirty(dn->dn_dbuf, tx); 718 719 dsl_dataset_dirty(os->os_dsl_dataset, tx); 720 } 721 722 void 723 dnode_free(dnode_t *dn, dmu_tx_t *tx) 724 { 725 int txgoff = tx->tx_txg & TXG_MASK; 726 727 dprintf("dn=%p txg=%llu\n", dn, tx->tx_txg); 728 729 /* we should be the only holder... hopefully */ 730 /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ 731 732 mutex_enter(&dn->dn_mtx); 733 if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { 734 mutex_exit(&dn->dn_mtx); 735 return; 736 } 737 dn->dn_free_txg = tx->tx_txg; 738 mutex_exit(&dn->dn_mtx); 739 740 /* 741 * If the dnode is already dirty, it needs to be moved from 742 * the dirty list to the free list. 743 */ 744 mutex_enter(&dn->dn_objset->os_lock); 745 if (list_link_active(&dn->dn_dirty_link[txgoff])) { 746 list_remove(&dn->dn_objset->os_dirty_dnodes[txgoff], dn); 747 list_insert_tail(&dn->dn_objset->os_free_dnodes[txgoff], dn); 748 mutex_exit(&dn->dn_objset->os_lock); 749 } else { 750 mutex_exit(&dn->dn_objset->os_lock); 751 dnode_setdirty(dn, tx); 752 } 753 } 754 755 /* 756 * Try to change the block size for the indicated dnode. This can only 757 * succeed if there are no blocks allocated or dirty beyond first block 758 */ 759 int 760 dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) 761 { 762 dmu_buf_impl_t *db, *db_next; 763 int have_db0 = FALSE; 764 765 if (size == 0) 766 size = SPA_MINBLOCKSIZE; 767 if (size > SPA_MAXBLOCKSIZE) 768 size = SPA_MAXBLOCKSIZE; 769 else 770 size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); 771 772 if (ibs == dn->dn_indblkshift) 773 ibs = 0; 774 775 if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && ibs == 0) 776 return (0); 777 778 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 779 780 /* Check for any allocated blocks beyond the first */ 781 if (dn->dn_phys->dn_maxblkid != 0) 782 goto fail; 783 784 mutex_enter(&dn->dn_dbufs_mtx); 785 for (db = list_head(&dn->dn_dbufs); db; db = db_next) { 786 db_next = list_next(&dn->dn_dbufs, db); 787 788 if (db->db_blkid == 0) { 789 have_db0 = TRUE; 790 } else if (db->db_blkid != DB_BONUS_BLKID) { 791 mutex_exit(&dn->dn_dbufs_mtx); 792 goto fail; 793 } 794 } 795 mutex_exit(&dn->dn_dbufs_mtx); 796 797 if (ibs && dn->dn_nlevels != 1) 798 goto fail; 799 800 db = NULL; 801 if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || have_db0) { 802 /* obtain the old block */ 803 db = dbuf_hold(dn, 0, FTAG); 804 dbuf_new_size(db, size, tx); 805 } 806 807 dnode_setdblksz(dn, size); 808 dnode_setdirty(dn, tx); 809 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size; 810 if (ibs) { 811 dn->dn_indblkshift = ibs; 812 dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; 813 } 814 815 if (db) 816 dbuf_rele(db, FTAG); 817 818 rw_exit(&dn->dn_struct_rwlock); 819 return (0); 820 821 fail: 822 rw_exit(&dn->dn_struct_rwlock); 823 return (ENOTSUP); 824 } 825 826 void 827 dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx) 828 { 829 uint64_t txgoff = tx->tx_txg & TXG_MASK; 830 int drop_struct_lock = FALSE; 831 int epbs, new_nlevels; 832 uint64_t sz; 833 834 ASSERT(blkid != DB_BONUS_BLKID); 835 836 if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { 837 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 838 drop_struct_lock = TRUE; 839 } 840 841 if (blkid <= dn->dn_maxblkid) 842 goto out; 843 844 dn->dn_maxblkid = blkid; 845 846 /* 847 * Compute the number of levels necessary to support the new maxblkid. 848 */ 849 new_nlevels = 1; 850 epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 851 for (sz = dn->dn_nblkptr; 852 sz <= blkid && sz >= dn->dn_nblkptr; sz <<= epbs) 853 new_nlevels++; 854 855 if (new_nlevels > dn->dn_nlevels) { 856 int old_nlevels = dn->dn_nlevels; 857 dmu_buf_impl_t *db; 858 859 dn->dn_nlevels = new_nlevels; 860 861 ASSERT3U(new_nlevels, >, dn->dn_next_nlevels[txgoff]); 862 dn->dn_next_nlevels[txgoff] = new_nlevels; 863 864 /* Dirty the left indirects. */ 865 db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); 866 dbuf_dirty(db, tx); 867 dbuf_rele(db, FTAG); 868 869 } 870 871 out: 872 if (drop_struct_lock) 873 rw_exit(&dn->dn_struct_rwlock); 874 } 875 876 void 877 dnode_clear_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) 878 { 879 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 880 avl_index_t where; 881 free_range_t *rp; 882 free_range_t rp_tofind; 883 uint64_t endblk = blkid + nblks; 884 885 ASSERT(MUTEX_HELD(&dn->dn_mtx)); 886 ASSERT(nblks <= UINT64_MAX - blkid); /* no overflow */ 887 888 dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 889 blkid, nblks, tx->tx_txg); 890 rp_tofind.fr_blkid = blkid; 891 rp = avl_find(tree, &rp_tofind, &where); 892 if (rp == NULL) 893 rp = avl_nearest(tree, where, AVL_BEFORE); 894 if (rp == NULL) 895 rp = avl_nearest(tree, where, AVL_AFTER); 896 897 while (rp && (rp->fr_blkid <= blkid + nblks)) { 898 uint64_t fr_endblk = rp->fr_blkid + rp->fr_nblks; 899 free_range_t *nrp = AVL_NEXT(tree, rp); 900 901 if (blkid <= rp->fr_blkid && endblk >= fr_endblk) { 902 /* clear this entire range */ 903 avl_remove(tree, rp); 904 kmem_free(rp, sizeof (free_range_t)); 905 } else if (blkid <= rp->fr_blkid && 906 endblk > rp->fr_blkid && endblk < fr_endblk) { 907 /* clear the beginning of this range */ 908 rp->fr_blkid = endblk; 909 rp->fr_nblks = fr_endblk - endblk; 910 } else if (blkid > rp->fr_blkid && blkid < fr_endblk && 911 endblk >= fr_endblk) { 912 /* clear the end of this range */ 913 rp->fr_nblks = blkid - rp->fr_blkid; 914 } else if (blkid > rp->fr_blkid && endblk < fr_endblk) { 915 /* clear a chunk out of this range */ 916 free_range_t *new_rp = 917 kmem_alloc(sizeof (free_range_t), KM_SLEEP); 918 919 new_rp->fr_blkid = endblk; 920 new_rp->fr_nblks = fr_endblk - endblk; 921 avl_insert_here(tree, new_rp, rp, AVL_AFTER); 922 rp->fr_nblks = blkid - rp->fr_blkid; 923 } 924 /* there may be no overlap */ 925 rp = nrp; 926 } 927 } 928 929 void 930 dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) 931 { 932 dmu_buf_impl_t *db; 933 uint64_t blkoff, blkid, nblks; 934 int blksz, head; 935 int trunc = FALSE; 936 937 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 938 blksz = dn->dn_datablksz; 939 940 /* If the range is past the end of the file, this is a no-op */ 941 if (off >= blksz * (dn->dn_maxblkid+1)) 942 goto out; 943 if (len == -1ULL) { 944 len = UINT64_MAX - off; 945 trunc = TRUE; 946 } 947 948 /* 949 * First, block align the region to free: 950 */ 951 if (ISP2(blksz)) { 952 head = P2NPHASE(off, blksz); 953 blkoff = P2PHASE(off, blksz); 954 } else { 955 ASSERT(dn->dn_maxblkid == 0); 956 if (off == 0 && len >= blksz) { 957 /* Freeing the whole block; don't do any head. */ 958 head = 0; 959 } else { 960 /* Freeing part of the block. */ 961 head = blksz - off; 962 ASSERT3U(head, >, 0); 963 } 964 blkoff = off; 965 } 966 /* zero out any partial block data at the start of the range */ 967 if (head) { 968 ASSERT3U(blkoff + head, ==, blksz); 969 if (len < head) 970 head = len; 971 if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off), TRUE, 972 FTAG, &db) == 0) { 973 caddr_t data; 974 975 /* don't dirty if it isn't on disk and isn't dirty */ 976 if (db->db_dirtied || 977 (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { 978 rw_exit(&dn->dn_struct_rwlock); 979 dbuf_will_dirty(db, tx); 980 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 981 data = db->db.db_data; 982 bzero(data + blkoff, head); 983 } 984 dbuf_rele(db, FTAG); 985 } 986 off += head; 987 len -= head; 988 } 989 990 /* If the range was less than one block, we're done */ 991 if (len == 0 || off >= blksz * (dn->dn_maxblkid+1)) 992 goto out; 993 994 if (!ISP2(blksz)) { 995 /* 996 * They are freeing the whole block of a 997 * non-power-of-two blocksize file. Skip all the messy 998 * math. 999 */ 1000 ASSERT3U(off, ==, 0); 1001 ASSERT3U(len, >=, blksz); 1002 blkid = 0; 1003 nblks = 1; 1004 } else { 1005 int tail; 1006 int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 1007 int blkshift = dn->dn_datablkshift; 1008 1009 /* If the remaining range is past end of file, we're done */ 1010 if (off > dn->dn_maxblkid << blkshift) 1011 goto out; 1012 1013 if (off + len == UINT64_MAX) 1014 tail = 0; 1015 else 1016 tail = P2PHASE(len, blksz); 1017 1018 ASSERT3U(P2PHASE(off, blksz), ==, 0); 1019 /* zero out any partial block data at the end of the range */ 1020 if (tail) { 1021 if (len < tail) 1022 tail = len; 1023 if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len), 1024 TRUE, FTAG, &db) == 0) { 1025 /* don't dirty if not on disk and not dirty */ 1026 if (db->db_dirtied || 1027 (db->db_blkptr && 1028 !BP_IS_HOLE(db->db_blkptr))) { 1029 rw_exit(&dn->dn_struct_rwlock); 1030 dbuf_will_dirty(db, tx); 1031 rw_enter(&dn->dn_struct_rwlock, 1032 RW_WRITER); 1033 bzero(db->db.db_data, tail); 1034 } 1035 dbuf_rele(db, FTAG); 1036 } 1037 len -= tail; 1038 } 1039 /* If the range did not include a full block, we are done */ 1040 if (len == 0) 1041 goto out; 1042 1043 /* dirty the left indirects */ 1044 if (dn->dn_nlevels > 1 && off != 0) { 1045 db = dbuf_hold_level(dn, 1, 1046 (off - head) >> (blkshift + epbs), FTAG); 1047 dbuf_will_dirty(db, tx); 1048 dbuf_rele(db, FTAG); 1049 } 1050 1051 /* dirty the right indirects */ 1052 if (dn->dn_nlevels > 1 && !trunc) { 1053 db = dbuf_hold_level(dn, 1, 1054 (off + len + tail - 1) >> (blkshift + epbs), FTAG); 1055 dbuf_will_dirty(db, tx); 1056 dbuf_rele(db, FTAG); 1057 } 1058 1059 /* 1060 * Finally, add this range to the dnode range list, we 1061 * will finish up this free operation in the syncing phase. 1062 */ 1063 ASSERT(IS_P2ALIGNED(off, 1<<blkshift)); 1064 ASSERT(off + len == UINT64_MAX || 1065 IS_P2ALIGNED(len, 1<<blkshift)); 1066 blkid = off >> blkshift; 1067 nblks = len >> blkshift; 1068 1069 if (trunc) 1070 dn->dn_maxblkid = (blkid ? blkid - 1 : 0); 1071 } 1072 1073 mutex_enter(&dn->dn_mtx); 1074 dnode_clear_range(dn, blkid, nblks, tx); 1075 { 1076 free_range_t *rp, *found; 1077 avl_index_t where; 1078 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 1079 1080 /* Add new range to dn_ranges */ 1081 rp = kmem_alloc(sizeof (free_range_t), KM_SLEEP); 1082 rp->fr_blkid = blkid; 1083 rp->fr_nblks = nblks; 1084 found = avl_find(tree, rp, &where); 1085 ASSERT(found == NULL); 1086 avl_insert(tree, rp, where); 1087 dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 1088 blkid, nblks, tx->tx_txg); 1089 } 1090 mutex_exit(&dn->dn_mtx); 1091 1092 dbuf_free_range(dn, blkid, nblks, tx); 1093 dnode_setdirty(dn, tx); 1094 out: 1095 rw_exit(&dn->dn_struct_rwlock); 1096 } 1097 1098 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */ 1099 uint64_t 1100 dnode_block_freed(dnode_t *dn, uint64_t blkid) 1101 { 1102 free_range_t range_tofind; 1103 void *dp = spa_get_dsl(dn->dn_objset->os_spa); 1104 int i; 1105 1106 if (blkid == DB_BONUS_BLKID) 1107 return (FALSE); 1108 1109 /* 1110 * If we're in the process of opening the pool, dp will not be 1111 * set yet, but there shouldn't be anything dirty. 1112 */ 1113 if (dp == NULL) 1114 return (FALSE); 1115 1116 if (dn->dn_free_txg) 1117 return (TRUE); 1118 1119 /* 1120 * If dn_datablkshift is not set, then there's only a single 1121 * block, in which case there will never be a free range so it 1122 * won't matter. 1123 */ 1124 range_tofind.fr_blkid = blkid; 1125 mutex_enter(&dn->dn_mtx); 1126 for (i = 0; i < TXG_SIZE; i++) { 1127 free_range_t *range_found; 1128 avl_index_t idx; 1129 1130 range_found = avl_find(&dn->dn_ranges[i], &range_tofind, &idx); 1131 if (range_found) { 1132 ASSERT(range_found->fr_nblks > 0); 1133 break; 1134 } 1135 range_found = avl_nearest(&dn->dn_ranges[i], idx, AVL_BEFORE); 1136 if (range_found && 1137 range_found->fr_blkid + range_found->fr_nblks > blkid) 1138 break; 1139 } 1140 mutex_exit(&dn->dn_mtx); 1141 return (i < TXG_SIZE); 1142 } 1143 1144 /* call from syncing context when we actually write/free space for this dnode */ 1145 void 1146 dnode_diduse_space(dnode_t *dn, int64_t delta) 1147 { 1148 uint64_t space; 1149 dprintf_dnode(dn, "dn=%p dnp=%p used=%llu delta=%lld\n", 1150 dn, dn->dn_phys, 1151 (u_longlong_t)dn->dn_phys->dn_used, 1152 (longlong_t)delta); 1153 1154 mutex_enter(&dn->dn_mtx); 1155 space = DN_USED_BYTES(dn->dn_phys); 1156 if (delta > 0) { 1157 ASSERT3U(space + delta, >=, space); /* no overflow */ 1158 } else { 1159 ASSERT3U(space, >=, -delta); /* no underflow */ 1160 } 1161 space += delta; 1162 if (spa_version(dn->dn_objset->os_spa) < ZFS_VERSION_DNODE_BYTES) { 1163 ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0); 1164 ASSERT3U(P2PHASE(space, 1<<DEV_BSHIFT), ==, 0); 1165 dn->dn_phys->dn_used = space >> DEV_BSHIFT; 1166 } else { 1167 dn->dn_phys->dn_used = space; 1168 dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES; 1169 } 1170 mutex_exit(&dn->dn_mtx); 1171 } 1172 1173 /* 1174 * Call when we think we're going to write/free space in open context. 1175 * Be conservative (ie. OK to write less than this or free more than 1176 * this, but don't write more or free less). 1177 */ 1178 void 1179 dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx) 1180 { 1181 objset_impl_t *os = dn->dn_objset; 1182 dsl_dataset_t *ds = os->os_dsl_dataset; 1183 1184 if (space > 0) 1185 space = spa_get_asize(os->os_spa, space); 1186 1187 if (ds) 1188 dsl_dir_willuse_space(ds->ds_dir, space, tx); 1189 1190 dmu_tx_willuse_space(tx, space); 1191 } 1192 1193 static int 1194 dnode_next_offset_level(dnode_t *dn, boolean_t hole, uint64_t *offset, 1195 int lvl, uint64_t blkfill, uint64_t txg) 1196 { 1197 dmu_buf_impl_t *db = NULL; 1198 void *data = NULL; 1199 uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 1200 uint64_t epb = 1ULL << epbs; 1201 uint64_t minfill, maxfill; 1202 int i, error, span; 1203 1204 dprintf("probing object %llu offset %llx level %d of %u\n", 1205 dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); 1206 1207 if (lvl == dn->dn_phys->dn_nlevels) { 1208 error = 0; 1209 epb = dn->dn_phys->dn_nblkptr; 1210 data = dn->dn_phys->dn_blkptr; 1211 } else { 1212 uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl); 1213 error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db); 1214 if (error) { 1215 if (error == ENOENT) 1216 return (hole ? 0 : ESRCH); 1217 return (error); 1218 } 1219 error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT); 1220 if (error) { 1221 dbuf_rele(db, FTAG); 1222 return (error); 1223 } 1224 data = db->db.db_data; 1225 } 1226 1227 if (db && txg && 1228 (db->db_blkptr == NULL || db->db_blkptr->blk_birth <= txg)) { 1229 error = ESRCH; 1230 } else if (lvl == 0) { 1231 dnode_phys_t *dnp = data; 1232 span = DNODE_SHIFT; 1233 ASSERT(dn->dn_type == DMU_OT_DNODE); 1234 1235 for (i = (*offset >> span) & (blkfill - 1); i < blkfill; i++) { 1236 boolean_t newcontents = B_TRUE; 1237 if (txg) { 1238 int j; 1239 newcontents = B_FALSE; 1240 for (j = 0; j < dnp[i].dn_nblkptr; j++) { 1241 if (dnp[i].dn_blkptr[j].blk_birth > txg) 1242 newcontents = B_TRUE; 1243 } 1244 } 1245 if (!dnp[i].dn_type == hole && newcontents) 1246 break; 1247 *offset += 1ULL << span; 1248 } 1249 if (i == blkfill) 1250 error = ESRCH; 1251 } else { 1252 blkptr_t *bp = data; 1253 span = (lvl - 1) * epbs + dn->dn_datablkshift; 1254 minfill = 0; 1255 maxfill = blkfill << ((lvl - 1) * epbs); 1256 1257 if (hole) 1258 maxfill--; 1259 else 1260 minfill++; 1261 1262 for (i = (*offset >> span) & ((1ULL << epbs) - 1); 1263 i < epb; i++) { 1264 if (bp[i].blk_fill >= minfill && 1265 bp[i].blk_fill <= maxfill && 1266 bp[i].blk_birth > txg) 1267 break; 1268 *offset += 1ULL << span; 1269 } 1270 if (i >= epb) 1271 error = ESRCH; 1272 } 1273 1274 if (db) 1275 dbuf_rele(db, FTAG); 1276 1277 return (error); 1278 } 1279 1280 /* 1281 * Find the next hole, data, or sparse region at or after *offset. 1282 * The value 'blkfill' tells us how many items we expect to find 1283 * in an L0 data block; this value is 1 for normal objects, 1284 * DNODES_PER_BLOCK for the meta dnode, and some fraction of 1285 * DNODES_PER_BLOCK when searching for sparse regions thereof. 1286 * 1287 * Examples: 1288 * 1289 * dnode_next_offset(dn, hole, offset, 1, 1, 0); 1290 * Finds the next hole/data in a file. 1291 * Used in dmu_offset_next(). 1292 * 1293 * dnode_next_offset(mdn, hole, offset, 0, DNODES_PER_BLOCK, txg); 1294 * Finds the next free/allocated dnode an objset's meta-dnode. 1295 * Only finds objects that have new contents since txg (ie. 1296 * bonus buffer changes and content removal are ignored). 1297 * Used in dmu_object_next(). 1298 * 1299 * dnode_next_offset(mdn, TRUE, offset, 2, DNODES_PER_BLOCK >> 2, 0); 1300 * Finds the next L2 meta-dnode bp that's at most 1/4 full. 1301 * Used in dmu_object_alloc(). 1302 */ 1303 int 1304 dnode_next_offset(dnode_t *dn, boolean_t hole, uint64_t *offset, 1305 int minlvl, uint64_t blkfill, uint64_t txg) 1306 { 1307 int lvl, maxlvl; 1308 int error = 0; 1309 uint64_t initial_offset = *offset; 1310 1311 rw_enter(&dn->dn_struct_rwlock, RW_READER); 1312 1313 if (dn->dn_phys->dn_nlevels == 0) { 1314 rw_exit(&dn->dn_struct_rwlock); 1315 return (ESRCH); 1316 } 1317 1318 if (dn->dn_datablkshift == 0) { 1319 if (*offset < dn->dn_datablksz) { 1320 if (hole) 1321 *offset = dn->dn_datablksz; 1322 } else { 1323 error = ESRCH; 1324 } 1325 rw_exit(&dn->dn_struct_rwlock); 1326 return (error); 1327 } 1328 1329 maxlvl = dn->dn_phys->dn_nlevels; 1330 1331 for (lvl = minlvl; lvl <= maxlvl; lvl++) { 1332 error = dnode_next_offset_level(dn, 1333 hole, offset, lvl, blkfill, txg); 1334 if (error != ESRCH) 1335 break; 1336 } 1337 1338 while (--lvl >= minlvl && error == 0) { 1339 error = dnode_next_offset_level(dn, 1340 hole, offset, lvl, blkfill, txg); 1341 } 1342 1343 rw_exit(&dn->dn_struct_rwlock); 1344 1345 if (error == 0 && initial_offset > *offset) 1346 error = ESRCH; 1347 1348 return (error); 1349 } 1350