1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/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_records[i], 69 sizeof (dbuf_dirty_record_t), 70 offsetof(dbuf_dirty_record_t, dr_dirty_node)); 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_records[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 ASSERT3U(dnp->dn_bonustype, <, DMU_OT_NUMTYPES); 210 dmu_ot[dnp->dn_bonustype].ot_byteswap(dnp->dn_bonus + off, len); 211 } 212 } 213 214 void 215 dnode_buf_byteswap(void *vbuf, size_t size) 216 { 217 dnode_phys_t *buf = vbuf; 218 int i; 219 220 ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT)); 221 ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0); 222 223 size >>= DNODE_SHIFT; 224 for (i = 0; i < size; i++) { 225 dnode_byteswap(buf); 226 buf++; 227 } 228 } 229 230 static int 231 free_range_compar(const void *node1, const void *node2) 232 { 233 const free_range_t *rp1 = node1; 234 const free_range_t *rp2 = node2; 235 236 if (rp1->fr_blkid < rp2->fr_blkid) 237 return (-1); 238 else if (rp1->fr_blkid > rp2->fr_blkid) 239 return (1); 240 else return (0); 241 } 242 243 void 244 dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx) 245 { 246 ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); 247 248 dnode_setdirty(dn, tx); 249 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 250 ASSERT3U(newsize, <=, DN_MAX_BONUSLEN - 251 (dn->dn_nblkptr-1) * sizeof (blkptr_t)); 252 dn->dn_bonuslen = newsize; 253 if (newsize == 0) 254 dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = DN_ZERO_BONUSLEN; 255 else 256 dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; 257 rw_exit(&dn->dn_struct_rwlock); 258 } 259 260 static void 261 dnode_setdblksz(dnode_t *dn, int size) 262 { 263 ASSERT3U(P2PHASE(size, SPA_MINBLOCKSIZE), ==, 0); 264 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 265 ASSERT3U(size, >=, SPA_MINBLOCKSIZE); 266 ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, 267 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); 268 dn->dn_datablksz = size; 269 dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; 270 dn->dn_datablkshift = ISP2(size) ? highbit(size - 1) : 0; 271 } 272 273 static dnode_t * 274 dnode_create(objset_impl_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, 275 uint64_t object) 276 { 277 dnode_t *dn = kmem_cache_alloc(dnode_cache, KM_SLEEP); 278 (void) dnode_cons(dn, NULL, 0); /* XXX */ 279 280 dn->dn_objset = os; 281 dn->dn_object = object; 282 dn->dn_dbuf = db; 283 dn->dn_phys = dnp; 284 285 if (dnp->dn_datablkszsec) 286 dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 287 dn->dn_indblkshift = dnp->dn_indblkshift; 288 dn->dn_nlevels = dnp->dn_nlevels; 289 dn->dn_type = dnp->dn_type; 290 dn->dn_nblkptr = dnp->dn_nblkptr; 291 dn->dn_checksum = dnp->dn_checksum; 292 dn->dn_compress = dnp->dn_compress; 293 dn->dn_bonustype = dnp->dn_bonustype; 294 dn->dn_bonuslen = dnp->dn_bonuslen; 295 dn->dn_maxblkid = dnp->dn_maxblkid; 296 297 dmu_zfetch_init(&dn->dn_zfetch, dn); 298 299 ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); 300 mutex_enter(&os->os_lock); 301 list_insert_head(&os->os_dnodes, dn); 302 mutex_exit(&os->os_lock); 303 304 arc_space_consume(sizeof (dnode_t)); 305 return (dn); 306 } 307 308 static void 309 dnode_destroy(dnode_t *dn) 310 { 311 objset_impl_t *os = dn->dn_objset; 312 313 #ifdef ZFS_DEBUG 314 int i; 315 316 for (i = 0; i < TXG_SIZE; i++) { 317 ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 318 ASSERT(NULL == list_head(&dn->dn_dirty_records[i])); 319 ASSERT(0 == avl_numnodes(&dn->dn_ranges[i])); 320 } 321 ASSERT(NULL == list_head(&dn->dn_dbufs)); 322 #endif 323 324 mutex_enter(&os->os_lock); 325 list_remove(&os->os_dnodes, dn); 326 mutex_exit(&os->os_lock); 327 328 if (dn->dn_dirtyctx_firstset) { 329 kmem_free(dn->dn_dirtyctx_firstset, 1); 330 dn->dn_dirtyctx_firstset = NULL; 331 } 332 dmu_zfetch_rele(&dn->dn_zfetch); 333 if (dn->dn_bonus) { 334 mutex_enter(&dn->dn_bonus->db_mtx); 335 dbuf_evict(dn->dn_bonus); 336 dn->dn_bonus = NULL; 337 } 338 kmem_cache_free(dnode_cache, dn); 339 arc_space_return(sizeof (dnode_t)); 340 } 341 342 void 343 dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 344 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 345 { 346 int i; 347 348 if (blocksize == 0) 349 blocksize = 1 << zfs_default_bs; 350 else if (blocksize > SPA_MAXBLOCKSIZE) 351 blocksize = SPA_MAXBLOCKSIZE; 352 else 353 blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); 354 355 if (ibs == 0) 356 ibs = zfs_default_ibs; 357 358 ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); 359 360 dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset, 361 dn->dn_object, tx->tx_txg, blocksize, ibs); 362 363 ASSERT(dn->dn_type == DMU_OT_NONE); 364 ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); 365 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); 366 ASSERT(ot != DMU_OT_NONE); 367 ASSERT3U(ot, <, DMU_OT_NUMTYPES); 368 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 369 (bonustype != DMU_OT_NONE && bonuslen != 0)); 370 ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 371 ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 372 ASSERT(dn->dn_type == DMU_OT_NONE); 373 ASSERT3U(dn->dn_maxblkid, ==, 0); 374 ASSERT3U(dn->dn_allocated_txg, ==, 0); 375 ASSERT3U(dn->dn_assigned_txg, ==, 0); 376 ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 377 ASSERT3U(refcount_count(&dn->dn_holds), <=, 1); 378 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 379 380 for (i = 0; i < TXG_SIZE; i++) { 381 ASSERT3U(dn->dn_next_nlevels[i], ==, 0); 382 ASSERT3U(dn->dn_next_indblkshift[i], ==, 0); 383 ASSERT3U(dn->dn_next_bonuslen[i], ==, 0); 384 ASSERT3U(dn->dn_next_blksz[i], ==, 0); 385 ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 386 ASSERT3P(list_head(&dn->dn_dirty_records[i]), ==, NULL); 387 ASSERT3U(avl_numnodes(&dn->dn_ranges[i]), ==, 0); 388 } 389 390 dn->dn_type = ot; 391 dnode_setdblksz(dn, blocksize); 392 dn->dn_indblkshift = ibs; 393 dn->dn_nlevels = 1; 394 dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 395 dn->dn_bonustype = bonustype; 396 dn->dn_bonuslen = bonuslen; 397 dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 398 dn->dn_compress = ZIO_COMPRESS_INHERIT; 399 dn->dn_dirtyctx = 0; 400 401 dn->dn_free_txg = 0; 402 if (dn->dn_dirtyctx_firstset) { 403 kmem_free(dn->dn_dirtyctx_firstset, 1); 404 dn->dn_dirtyctx_firstset = NULL; 405 } 406 407 dn->dn_allocated_txg = tx->tx_txg; 408 409 dnode_setdirty(dn, tx); 410 dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs; 411 dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; 412 dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz; 413 } 414 415 void 416 dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 417 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 418 { 419 int i, old_nblkptr; 420 dmu_buf_impl_t *db = NULL; 421 422 ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); 423 ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE); 424 ASSERT3U(blocksize % SPA_MINBLOCKSIZE, ==, 0); 425 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); 426 ASSERT(tx->tx_txg != 0); 427 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 428 (bonustype != DMU_OT_NONE && bonuslen != 0)); 429 ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); 430 ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); 431 432 for (i = 0; i < TXG_SIZE; i++) 433 ASSERT(!list_link_active(&dn->dn_dirty_link[i])); 434 435 /* clean up any unreferenced dbufs */ 436 dnode_evict_dbufs(dn); 437 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 438 439 /* 440 * XXX I should really have a generation number to tell if we 441 * need to do this... 442 */ 443 if (blocksize != dn->dn_datablksz || 444 dn->dn_bonustype != bonustype || dn->dn_bonuslen != bonuslen) { 445 /* free all old data */ 446 dnode_free_range(dn, 0, -1ULL, tx); 447 } 448 449 /* change blocksize */ 450 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 451 if (blocksize != dn->dn_datablksz && 452 (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || 453 list_head(&dn->dn_dbufs) != NULL)) { 454 db = dbuf_hold(dn, 0, FTAG); 455 dbuf_new_size(db, blocksize, tx); 456 } 457 dnode_setdblksz(dn, blocksize); 458 dnode_setdirty(dn, tx); 459 dn->dn_next_bonuslen[tx->tx_txg&TXG_MASK] = bonuslen; 460 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize; 461 rw_exit(&dn->dn_struct_rwlock); 462 if (db) 463 dbuf_rele(db, FTAG); 464 465 /* change type */ 466 dn->dn_type = ot; 467 468 /* change bonus size and type */ 469 mutex_enter(&dn->dn_mtx); 470 old_nblkptr = dn->dn_nblkptr; 471 dn->dn_bonustype = bonustype; 472 dn->dn_bonuslen = bonuslen; 473 dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 474 dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 475 dn->dn_compress = ZIO_COMPRESS_INHERIT; 476 ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 477 478 /* XXX - for now, we can't make nblkptr smaller */ 479 ASSERT3U(dn->dn_nblkptr, >=, old_nblkptr); 480 481 /* fix up the bonus db_size if dn_nblkptr has changed */ 482 if (dn->dn_bonus && dn->dn_bonuslen != old_nblkptr) { 483 dn->dn_bonus->db.db_size = 484 DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t); 485 ASSERT(dn->dn_bonuslen <= dn->dn_bonus->db.db_size); 486 } 487 488 dn->dn_allocated_txg = tx->tx_txg; 489 mutex_exit(&dn->dn_mtx); 490 } 491 492 void 493 dnode_special_close(dnode_t *dn) 494 { 495 /* 496 * Wait for final references to the dnode to clear. This can 497 * only happen if the arc is asyncronously evicting state that 498 * has a hold on this dnode while we are trying to evict this 499 * dnode. 500 */ 501 while (refcount_count(&dn->dn_holds) > 0) 502 delay(1); 503 dnode_destroy(dn); 504 } 505 506 dnode_t * 507 dnode_special_open(objset_impl_t *os, dnode_phys_t *dnp, uint64_t object) 508 { 509 dnode_t *dn = dnode_create(os, dnp, NULL, object); 510 DNODE_VERIFY(dn); 511 return (dn); 512 } 513 514 static void 515 dnode_buf_pageout(dmu_buf_t *db, void *arg) 516 { 517 dnode_t **children_dnodes = arg; 518 int i; 519 int epb = db->db_size >> DNODE_SHIFT; 520 521 for (i = 0; i < epb; i++) { 522 dnode_t *dn = children_dnodes[i]; 523 int n; 524 525 if (dn == NULL) 526 continue; 527 #ifdef ZFS_DEBUG 528 /* 529 * If there are holds on this dnode, then there should 530 * be holds on the dnode's containing dbuf as well; thus 531 * it wouldn't be eligable for eviction and this function 532 * would not have been called. 533 */ 534 ASSERT(refcount_is_zero(&dn->dn_holds)); 535 ASSERT(list_head(&dn->dn_dbufs) == NULL); 536 ASSERT(refcount_is_zero(&dn->dn_tx_holds)); 537 538 for (n = 0; n < TXG_SIZE; n++) 539 ASSERT(!list_link_active(&dn->dn_dirty_link[n])); 540 #endif 541 children_dnodes[i] = NULL; 542 dnode_destroy(dn); 543 } 544 kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 545 } 546 547 /* 548 * errors: 549 * EINVAL - invalid object number. 550 * EIO - i/o error. 551 * succeeds even for free dnodes. 552 */ 553 int 554 dnode_hold_impl(objset_impl_t *os, uint64_t object, int flag, 555 void *tag, dnode_t **dnp) 556 { 557 int epb, idx, err; 558 int drop_struct_lock = FALSE; 559 int type; 560 uint64_t blk; 561 dnode_t *mdn, *dn; 562 dmu_buf_impl_t *db; 563 dnode_t **children_dnodes; 564 565 if (object == 0 || object >= DN_MAX_OBJECT) 566 return (EINVAL); 567 568 mdn = os->os_meta_dnode; 569 570 DNODE_VERIFY(mdn); 571 572 if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { 573 rw_enter(&mdn->dn_struct_rwlock, RW_READER); 574 drop_struct_lock = TRUE; 575 } 576 577 blk = dbuf_whichblock(mdn, object * sizeof (dnode_phys_t)); 578 579 db = dbuf_hold(mdn, blk, FTAG); 580 if (drop_struct_lock) 581 rw_exit(&mdn->dn_struct_rwlock); 582 if (db == NULL) 583 return (EIO); 584 err = dbuf_read(db, NULL, DB_RF_CANFAIL); 585 if (err) { 586 dbuf_rele(db, FTAG); 587 return (err); 588 } 589 590 ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT); 591 epb = db->db.db_size >> DNODE_SHIFT; 592 593 idx = object & (epb-1); 594 595 children_dnodes = dmu_buf_get_user(&db->db); 596 if (children_dnodes == NULL) { 597 dnode_t **winner; 598 children_dnodes = kmem_zalloc(epb * sizeof (dnode_t *), 599 KM_SLEEP); 600 if (winner = dmu_buf_set_user(&db->db, children_dnodes, NULL, 601 dnode_buf_pageout)) { 602 kmem_free(children_dnodes, epb * sizeof (dnode_t *)); 603 children_dnodes = winner; 604 } 605 } 606 607 if ((dn = children_dnodes[idx]) == NULL) { 608 dnode_phys_t *dnp = (dnode_phys_t *)db->db.db_data+idx; 609 dnode_t *winner; 610 611 dn = dnode_create(os, dnp, db, object); 612 winner = atomic_cas_ptr(&children_dnodes[idx], NULL, dn); 613 if (winner != NULL) { 614 dnode_destroy(dn); 615 dn = winner; 616 } 617 } 618 619 mutex_enter(&dn->dn_mtx); 620 type = dn->dn_type; 621 if (dn->dn_free_txg || 622 ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) || 623 ((flag & DNODE_MUST_BE_FREE) && type != DMU_OT_NONE)) { 624 mutex_exit(&dn->dn_mtx); 625 dbuf_rele(db, FTAG); 626 return (type == DMU_OT_NONE ? ENOENT : EEXIST); 627 } 628 mutex_exit(&dn->dn_mtx); 629 630 if (refcount_add(&dn->dn_holds, tag) == 1) 631 dbuf_add_ref(db, dn); 632 633 DNODE_VERIFY(dn); 634 ASSERT3P(dn->dn_dbuf, ==, db); 635 ASSERT3U(dn->dn_object, ==, object); 636 dbuf_rele(db, FTAG); 637 638 *dnp = dn; 639 return (0); 640 } 641 642 /* 643 * Return held dnode if the object is allocated, NULL if not. 644 */ 645 int 646 dnode_hold(objset_impl_t *os, uint64_t object, void *tag, dnode_t **dnp) 647 { 648 return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp)); 649 } 650 651 /* 652 * Can only add a reference if there is already at least one 653 * reference on the dnode. Returns FALSE if unable to add a 654 * new reference. 655 */ 656 boolean_t 657 dnode_add_ref(dnode_t *dn, void *tag) 658 { 659 mutex_enter(&dn->dn_mtx); 660 if (refcount_is_zero(&dn->dn_holds)) { 661 mutex_exit(&dn->dn_mtx); 662 return (FALSE); 663 } 664 VERIFY(1 < refcount_add(&dn->dn_holds, tag)); 665 mutex_exit(&dn->dn_mtx); 666 return (TRUE); 667 } 668 669 void 670 dnode_rele(dnode_t *dn, void *tag) 671 { 672 uint64_t refs; 673 674 mutex_enter(&dn->dn_mtx); 675 refs = refcount_remove(&dn->dn_holds, tag); 676 mutex_exit(&dn->dn_mtx); 677 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ 678 if (refs == 0 && dn->dn_dbuf) 679 dbuf_rele(dn->dn_dbuf, dn); 680 } 681 682 void 683 dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) 684 { 685 objset_impl_t *os = dn->dn_objset; 686 uint64_t txg = tx->tx_txg; 687 688 if (dn->dn_object == DMU_META_DNODE_OBJECT) 689 return; 690 691 DNODE_VERIFY(dn); 692 693 #ifdef ZFS_DEBUG 694 mutex_enter(&dn->dn_mtx); 695 ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); 696 /* ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); */ 697 mutex_exit(&dn->dn_mtx); 698 #endif 699 700 mutex_enter(&os->os_lock); 701 702 /* 703 * If we are already marked dirty, we're done. 704 */ 705 if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) { 706 mutex_exit(&os->os_lock); 707 return; 708 } 709 710 ASSERT(!refcount_is_zero(&dn->dn_holds) || list_head(&dn->dn_dbufs)); 711 ASSERT(dn->dn_datablksz != 0); 712 ASSERT3U(dn->dn_next_bonuslen[txg&TXG_MASK], ==, 0); 713 ASSERT3U(dn->dn_next_blksz[txg&TXG_MASK], ==, 0); 714 715 dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", 716 dn->dn_object, txg); 717 718 if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) { 719 list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn); 720 } else { 721 list_insert_tail(&os->os_dirty_dnodes[txg&TXG_MASK], dn); 722 } 723 724 mutex_exit(&os->os_lock); 725 726 /* 727 * The dnode maintains a hold on its containing dbuf as 728 * long as there are holds on it. Each instantiated child 729 * dbuf maintaines a hold on the dnode. When the last child 730 * drops its hold, the dnode will drop its hold on the 731 * containing dbuf. We add a "dirty hold" here so that the 732 * dnode will hang around after we finish processing its 733 * children. 734 */ 735 VERIFY(dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg)); 736 737 (void) dbuf_dirty(dn->dn_dbuf, tx); 738 739 dsl_dataset_dirty(os->os_dsl_dataset, tx); 740 } 741 742 void 743 dnode_free(dnode_t *dn, dmu_tx_t *tx) 744 { 745 int txgoff = tx->tx_txg & TXG_MASK; 746 747 dprintf("dn=%p txg=%llu\n", dn, tx->tx_txg); 748 749 /* we should be the only holder... hopefully */ 750 /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ 751 752 mutex_enter(&dn->dn_mtx); 753 if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { 754 mutex_exit(&dn->dn_mtx); 755 return; 756 } 757 dn->dn_free_txg = tx->tx_txg; 758 mutex_exit(&dn->dn_mtx); 759 760 /* 761 * If the dnode is already dirty, it needs to be moved from 762 * the dirty list to the free list. 763 */ 764 mutex_enter(&dn->dn_objset->os_lock); 765 if (list_link_active(&dn->dn_dirty_link[txgoff])) { 766 list_remove(&dn->dn_objset->os_dirty_dnodes[txgoff], dn); 767 list_insert_tail(&dn->dn_objset->os_free_dnodes[txgoff], dn); 768 mutex_exit(&dn->dn_objset->os_lock); 769 } else { 770 mutex_exit(&dn->dn_objset->os_lock); 771 dnode_setdirty(dn, tx); 772 } 773 } 774 775 /* 776 * Try to change the block size for the indicated dnode. This can only 777 * succeed if there are no blocks allocated or dirty beyond first block 778 */ 779 int 780 dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) 781 { 782 dmu_buf_impl_t *db, *db_next; 783 int err; 784 785 if (size == 0) 786 size = SPA_MINBLOCKSIZE; 787 if (size > SPA_MAXBLOCKSIZE) 788 size = SPA_MAXBLOCKSIZE; 789 else 790 size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); 791 792 if (ibs == dn->dn_indblkshift) 793 ibs = 0; 794 795 if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && ibs == 0) 796 return (0); 797 798 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 799 800 /* Check for any allocated blocks beyond the first */ 801 if (dn->dn_phys->dn_maxblkid != 0) 802 goto fail; 803 804 mutex_enter(&dn->dn_dbufs_mtx); 805 for (db = list_head(&dn->dn_dbufs); db; db = db_next) { 806 db_next = list_next(&dn->dn_dbufs, db); 807 808 if (db->db_blkid != 0 && db->db_blkid != DB_BONUS_BLKID) { 809 mutex_exit(&dn->dn_dbufs_mtx); 810 goto fail; 811 } 812 } 813 mutex_exit(&dn->dn_dbufs_mtx); 814 815 if (ibs && dn->dn_nlevels != 1) 816 goto fail; 817 818 /* resize the old block */ 819 err = dbuf_hold_impl(dn, 0, 0, TRUE, FTAG, &db); 820 if (err == 0) 821 dbuf_new_size(db, size, tx); 822 else if (err != ENOENT) 823 goto fail; 824 825 dnode_setdblksz(dn, size); 826 dnode_setdirty(dn, tx); 827 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size; 828 if (ibs) { 829 dn->dn_indblkshift = ibs; 830 dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; 831 } 832 /* rele after we have fixed the blocksize in the dnode */ 833 if (db) 834 dbuf_rele(db, FTAG); 835 836 rw_exit(&dn->dn_struct_rwlock); 837 return (0); 838 839 fail: 840 rw_exit(&dn->dn_struct_rwlock); 841 return (ENOTSUP); 842 } 843 844 void 845 dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx) 846 { 847 uint64_t txgoff = tx->tx_txg & TXG_MASK; 848 int drop_struct_lock = FALSE; 849 int epbs, new_nlevels; 850 uint64_t sz; 851 852 ASSERT(blkid != DB_BONUS_BLKID); 853 854 if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { 855 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 856 drop_struct_lock = TRUE; 857 } 858 859 if (blkid <= dn->dn_maxblkid) 860 goto out; 861 862 dn->dn_maxblkid = blkid; 863 864 /* 865 * Compute the number of levels necessary to support the new maxblkid. 866 */ 867 new_nlevels = 1; 868 epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 869 for (sz = dn->dn_nblkptr; 870 sz <= blkid && sz >= dn->dn_nblkptr; sz <<= epbs) 871 new_nlevels++; 872 873 if (new_nlevels > dn->dn_nlevels) { 874 int old_nlevels = dn->dn_nlevels; 875 dmu_buf_impl_t *db; 876 list_t *list; 877 dbuf_dirty_record_t *new, *dr, *dr_next; 878 879 dn->dn_nlevels = new_nlevels; 880 881 ASSERT3U(new_nlevels, >, dn->dn_next_nlevels[txgoff]); 882 dn->dn_next_nlevels[txgoff] = new_nlevels; 883 884 /* dirty the left indirects */ 885 db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); 886 new = dbuf_dirty(db, tx); 887 dbuf_rele(db, FTAG); 888 889 /* transfer the dirty records to the new indirect */ 890 mutex_enter(&dn->dn_mtx); 891 mutex_enter(&new->dt.di.dr_mtx); 892 list = &dn->dn_dirty_records[txgoff]; 893 for (dr = list_head(list); dr; dr = dr_next) { 894 dr_next = list_next(&dn->dn_dirty_records[txgoff], dr); 895 if (dr->dr_dbuf->db_level != new_nlevels-1 && 896 dr->dr_dbuf->db_blkid != DB_BONUS_BLKID) { 897 ASSERT(dr->dr_dbuf->db_level == old_nlevels-1); 898 list_remove(&dn->dn_dirty_records[txgoff], dr); 899 list_insert_tail(&new->dt.di.dr_children, dr); 900 dr->dr_parent = new; 901 } 902 } 903 mutex_exit(&new->dt.di.dr_mtx); 904 mutex_exit(&dn->dn_mtx); 905 } 906 907 out: 908 if (drop_struct_lock) 909 rw_exit(&dn->dn_struct_rwlock); 910 } 911 912 void 913 dnode_clear_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) 914 { 915 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 916 avl_index_t where; 917 free_range_t *rp; 918 free_range_t rp_tofind; 919 uint64_t endblk = blkid + nblks; 920 921 ASSERT(MUTEX_HELD(&dn->dn_mtx)); 922 ASSERT(nblks <= UINT64_MAX - blkid); /* no overflow */ 923 924 dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 925 blkid, nblks, tx->tx_txg); 926 rp_tofind.fr_blkid = blkid; 927 rp = avl_find(tree, &rp_tofind, &where); 928 if (rp == NULL) 929 rp = avl_nearest(tree, where, AVL_BEFORE); 930 if (rp == NULL) 931 rp = avl_nearest(tree, where, AVL_AFTER); 932 933 while (rp && (rp->fr_blkid <= blkid + nblks)) { 934 uint64_t fr_endblk = rp->fr_blkid + rp->fr_nblks; 935 free_range_t *nrp = AVL_NEXT(tree, rp); 936 937 if (blkid <= rp->fr_blkid && endblk >= fr_endblk) { 938 /* clear this entire range */ 939 avl_remove(tree, rp); 940 kmem_free(rp, sizeof (free_range_t)); 941 } else if (blkid <= rp->fr_blkid && 942 endblk > rp->fr_blkid && endblk < fr_endblk) { 943 /* clear the beginning of this range */ 944 rp->fr_blkid = endblk; 945 rp->fr_nblks = fr_endblk - endblk; 946 } else if (blkid > rp->fr_blkid && blkid < fr_endblk && 947 endblk >= fr_endblk) { 948 /* clear the end of this range */ 949 rp->fr_nblks = blkid - rp->fr_blkid; 950 } else if (blkid > rp->fr_blkid && endblk < fr_endblk) { 951 /* clear a chunk out of this range */ 952 free_range_t *new_rp = 953 kmem_alloc(sizeof (free_range_t), KM_SLEEP); 954 955 new_rp->fr_blkid = endblk; 956 new_rp->fr_nblks = fr_endblk - endblk; 957 avl_insert_here(tree, new_rp, rp, AVL_AFTER); 958 rp->fr_nblks = blkid - rp->fr_blkid; 959 } 960 /* there may be no overlap */ 961 rp = nrp; 962 } 963 } 964 965 void 966 dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) 967 { 968 dmu_buf_impl_t *db; 969 uint64_t blkoff, blkid, nblks; 970 int blksz, blkshift, head, tail; 971 int trunc = FALSE; 972 int epbs; 973 974 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 975 blksz = dn->dn_datablksz; 976 blkshift = dn->dn_datablkshift; 977 epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 978 979 if (len == -1ULL) { 980 len = UINT64_MAX - off; 981 trunc = TRUE; 982 } 983 984 /* 985 * First, block align the region to free: 986 */ 987 if (ISP2(blksz)) { 988 head = P2NPHASE(off, blksz); 989 blkoff = P2PHASE(off, blksz); 990 if ((off >> blkshift) > dn->dn_maxblkid) 991 goto out; 992 } else { 993 ASSERT(dn->dn_maxblkid == 0); 994 if (off == 0 && len >= blksz) { 995 /* Freeing the whole block; fast-track this request */ 996 blkid = 0; 997 nblks = 1; 998 goto done; 999 } else if (off > blkid) { 1000 /* Freeing past end-of-data */ 1001 goto out; 1002 } else { 1003 /* Freeing part of the block. */ 1004 head = blksz - off; 1005 ASSERT3U(head, >, 0); 1006 } 1007 blkoff = off; 1008 } 1009 /* zero out any partial block data at the start of the range */ 1010 if (head) { 1011 ASSERT3U(blkoff + head, ==, blksz); 1012 if (len < head) 1013 head = len; 1014 if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off), TRUE, 1015 FTAG, &db) == 0) { 1016 caddr_t data; 1017 1018 /* don't dirty if it isn't on disk and isn't dirty */ 1019 if (db->db_last_dirty || 1020 (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { 1021 rw_exit(&dn->dn_struct_rwlock); 1022 dbuf_will_dirty(db, tx); 1023 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 1024 data = db->db.db_data; 1025 bzero(data + blkoff, head); 1026 } 1027 dbuf_rele(db, FTAG); 1028 } 1029 off += head; 1030 len -= head; 1031 } 1032 1033 /* If the range was less than one block, we're done */ 1034 if (len == 0) 1035 goto out; 1036 1037 ASSERT(ISP2(blksz)); 1038 /* If the remaining range is past end of file, we're done */ 1039 if ((off >> blkshift) > dn->dn_maxblkid) 1040 goto out; 1041 1042 if (trunc) 1043 tail = 0; 1044 else 1045 tail = P2PHASE(len, blksz); 1046 1047 ASSERT3U(P2PHASE(off, blksz), ==, 0); 1048 /* zero out any partial block data at the end of the range */ 1049 if (tail) { 1050 if (len < tail) 1051 tail = len; 1052 if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len), 1053 TRUE, FTAG, &db) == 0) { 1054 /* don't dirty if not on disk and not dirty */ 1055 if (db->db_last_dirty || 1056 (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { 1057 rw_exit(&dn->dn_struct_rwlock); 1058 dbuf_will_dirty(db, tx); 1059 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 1060 bzero(db->db.db_data, tail); 1061 } 1062 dbuf_rele(db, FTAG); 1063 } 1064 len -= tail; 1065 } 1066 1067 /* If the range did not include a full block, we are done */ 1068 if (len == 0) 1069 goto out; 1070 1071 ASSERT(IS_P2ALIGNED(off, blksz)); 1072 ASSERT(trunc || IS_P2ALIGNED(len, blksz)); 1073 blkid = off >> blkshift; 1074 nblks = len >> blkshift; 1075 if (trunc) 1076 nblks += 1; 1077 1078 /* 1079 * Read in and mark all the level-1 indirects dirty, 1080 * so that they will stay in memory until syncing phase. 1081 */ 1082 if (dn->dn_nlevels > 1) { 1083 uint64_t i, first, last; 1084 int shift = epbs + dn->dn_datablkshift; 1085 1086 first = blkid >> epbs; 1087 if (trunc) 1088 last = dn->dn_maxblkid >> epbs; 1089 else 1090 last = (blkid + nblks - 1) >> epbs; 1091 for (i = first; i <= last; i++) { 1092 uint64_t ibyte = i << shift; 1093 int err; 1094 1095 err = dnode_next_offset(dn, 1096 DNODE_FIND_HAVELOCK, &ibyte, 1, 1, 0); 1097 i = ibyte >> shift; 1098 if (err == ESRCH || i > last) 1099 break; 1100 ASSERT(err == 0); 1101 db = dbuf_hold_level(dn, 1, i, FTAG); 1102 if (db) { 1103 dbuf_will_dirty(db, tx); 1104 dbuf_rele(db, FTAG); 1105 } 1106 } 1107 } 1108 done: 1109 /* 1110 * Add this range to the dnode range list. 1111 * We will finish up this free operation in the syncing phase. 1112 */ 1113 mutex_enter(&dn->dn_mtx); 1114 dnode_clear_range(dn, blkid, nblks, tx); 1115 { 1116 free_range_t *rp, *found; 1117 avl_index_t where; 1118 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; 1119 1120 /* Add new range to dn_ranges */ 1121 rp = kmem_alloc(sizeof (free_range_t), KM_SLEEP); 1122 rp->fr_blkid = blkid; 1123 rp->fr_nblks = nblks; 1124 found = avl_find(tree, rp, &where); 1125 ASSERT(found == NULL); 1126 avl_insert(tree, rp, where); 1127 dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 1128 blkid, nblks, tx->tx_txg); 1129 } 1130 mutex_exit(&dn->dn_mtx); 1131 1132 dbuf_free_range(dn, blkid, blkid + nblks - 1, tx); 1133 dnode_setdirty(dn, tx); 1134 out: 1135 if (trunc && dn->dn_maxblkid >= (off >> blkshift)) 1136 dn->dn_maxblkid = (off >> blkshift ? (off >> blkshift) - 1 : 0); 1137 1138 rw_exit(&dn->dn_struct_rwlock); 1139 } 1140 1141 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */ 1142 uint64_t 1143 dnode_block_freed(dnode_t *dn, uint64_t blkid) 1144 { 1145 free_range_t range_tofind; 1146 void *dp = spa_get_dsl(dn->dn_objset->os_spa); 1147 int i; 1148 1149 if (blkid == DB_BONUS_BLKID) 1150 return (FALSE); 1151 1152 /* 1153 * If we're in the process of opening the pool, dp will not be 1154 * set yet, but there shouldn't be anything dirty. 1155 */ 1156 if (dp == NULL) 1157 return (FALSE); 1158 1159 if (dn->dn_free_txg) 1160 return (TRUE); 1161 1162 /* 1163 * If dn_datablkshift is not set, then there's only a single 1164 * block, in which case there will never be a free range so it 1165 * won't matter. 1166 */ 1167 range_tofind.fr_blkid = blkid; 1168 mutex_enter(&dn->dn_mtx); 1169 for (i = 0; i < TXG_SIZE; i++) { 1170 free_range_t *range_found; 1171 avl_index_t idx; 1172 1173 range_found = avl_find(&dn->dn_ranges[i], &range_tofind, &idx); 1174 if (range_found) { 1175 ASSERT(range_found->fr_nblks > 0); 1176 break; 1177 } 1178 range_found = avl_nearest(&dn->dn_ranges[i], idx, AVL_BEFORE); 1179 if (range_found && 1180 range_found->fr_blkid + range_found->fr_nblks > blkid) 1181 break; 1182 } 1183 mutex_exit(&dn->dn_mtx); 1184 return (i < TXG_SIZE); 1185 } 1186 1187 /* call from syncing context when we actually write/free space for this dnode */ 1188 void 1189 dnode_diduse_space(dnode_t *dn, int64_t delta) 1190 { 1191 uint64_t space; 1192 dprintf_dnode(dn, "dn=%p dnp=%p used=%llu delta=%lld\n", 1193 dn, dn->dn_phys, 1194 (u_longlong_t)dn->dn_phys->dn_used, 1195 (longlong_t)delta); 1196 1197 mutex_enter(&dn->dn_mtx); 1198 space = DN_USED_BYTES(dn->dn_phys); 1199 if (delta > 0) { 1200 ASSERT3U(space + delta, >=, space); /* no overflow */ 1201 } else { 1202 ASSERT3U(space, >=, -delta); /* no underflow */ 1203 } 1204 space += delta; 1205 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_DNODE_BYTES) { 1206 ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0); 1207 ASSERT3U(P2PHASE(space, 1<<DEV_BSHIFT), ==, 0); 1208 dn->dn_phys->dn_used = space >> DEV_BSHIFT; 1209 } else { 1210 dn->dn_phys->dn_used = space; 1211 dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES; 1212 } 1213 mutex_exit(&dn->dn_mtx); 1214 } 1215 1216 /* 1217 * Call when we think we're going to write/free space in open context. 1218 * Be conservative (ie. OK to write less than this or free more than 1219 * this, but don't write more or free less). 1220 */ 1221 void 1222 dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx) 1223 { 1224 objset_impl_t *os = dn->dn_objset; 1225 dsl_dataset_t *ds = os->os_dsl_dataset; 1226 1227 if (space > 0) 1228 space = spa_get_asize(os->os_spa, space); 1229 1230 if (ds) 1231 dsl_dir_willuse_space(ds->ds_dir, space, tx); 1232 1233 dmu_tx_willuse_space(tx, space); 1234 } 1235 1236 static int 1237 dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset, 1238 int lvl, uint64_t blkfill, uint64_t txg) 1239 { 1240 dmu_buf_impl_t *db = NULL; 1241 void *data = NULL; 1242 uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 1243 uint64_t epb = 1ULL << epbs; 1244 uint64_t minfill, maxfill; 1245 boolean_t hole; 1246 int i, inc, error, span; 1247 1248 dprintf("probing object %llu offset %llx level %d of %u\n", 1249 dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); 1250 1251 hole = flags & DNODE_FIND_HOLE; 1252 inc = (flags & DNODE_FIND_BACKWARDS) ? -1 : 1; 1253 1254 if (lvl == dn->dn_phys->dn_nlevels) { 1255 error = 0; 1256 epb = dn->dn_phys->dn_nblkptr; 1257 data = dn->dn_phys->dn_blkptr; 1258 } else { 1259 uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl); 1260 error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db); 1261 if (error) { 1262 if (error == ENOENT) 1263 return (hole ? 0 : ESRCH); 1264 return (error); 1265 } 1266 error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT); 1267 if (error) { 1268 dbuf_rele(db, FTAG); 1269 return (error); 1270 } 1271 data = db->db.db_data; 1272 } 1273 1274 if (db && txg && 1275 (db->db_blkptr == NULL || db->db_blkptr->blk_birth <= txg)) { 1276 error = ESRCH; 1277 } else if (lvl == 0) { 1278 dnode_phys_t *dnp = data; 1279 span = DNODE_SHIFT; 1280 ASSERT(dn->dn_type == DMU_OT_DNODE); 1281 1282 for (i = (*offset >> span) & (blkfill - 1); 1283 i >= 0 && i < blkfill; i += inc) { 1284 boolean_t newcontents = B_TRUE; 1285 if (txg) { 1286 int j; 1287 newcontents = B_FALSE; 1288 for (j = 0; j < dnp[i].dn_nblkptr; j++) { 1289 if (dnp[i].dn_blkptr[j].blk_birth > txg) 1290 newcontents = B_TRUE; 1291 } 1292 } 1293 if (!dnp[i].dn_type == hole && newcontents) 1294 break; 1295 *offset += (1ULL << span) * inc; 1296 } 1297 if (i < 0 || i == blkfill) 1298 error = ESRCH; 1299 } else { 1300 blkptr_t *bp = data; 1301 span = (lvl - 1) * epbs + dn->dn_datablkshift; 1302 minfill = 0; 1303 maxfill = blkfill << ((lvl - 1) * epbs); 1304 1305 if (hole) 1306 maxfill--; 1307 else 1308 minfill++; 1309 1310 for (i = (*offset >> span) & ((1ULL << epbs) - 1); 1311 i >= 0 && i < epb; i += inc) { 1312 if (bp[i].blk_fill >= minfill && 1313 bp[i].blk_fill <= maxfill && 1314 bp[i].blk_birth > txg) 1315 break; 1316 *offset += (1ULL << span) * inc; 1317 } 1318 if (i < 0 || i == epb) 1319 error = ESRCH; 1320 } 1321 1322 if (db) 1323 dbuf_rele(db, FTAG); 1324 1325 return (error); 1326 } 1327 1328 /* 1329 * Find the next hole, data, or sparse region at or after *offset. 1330 * The value 'blkfill' tells us how many items we expect to find 1331 * in an L0 data block; this value is 1 for normal objects, 1332 * DNODES_PER_BLOCK for the meta dnode, and some fraction of 1333 * DNODES_PER_BLOCK when searching for sparse regions thereof. 1334 * 1335 * Examples: 1336 * 1337 * dnode_next_offset(dn, flags, offset, 1, 1, 0); 1338 * Finds the next/previous hole/data in a file. 1339 * Used in dmu_offset_next(). 1340 * 1341 * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg); 1342 * Finds the next free/allocated dnode an objset's meta-dnode. 1343 * Only finds objects that have new contents since txg (ie. 1344 * bonus buffer changes and content removal are ignored). 1345 * Used in dmu_object_next(). 1346 * 1347 * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0); 1348 * Finds the next L2 meta-dnode bp that's at most 1/4 full. 1349 * Used in dmu_object_alloc(). 1350 */ 1351 int 1352 dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset, 1353 int minlvl, uint64_t blkfill, uint64_t txg) 1354 { 1355 uint64_t initial_offset = *offset; 1356 int lvl, maxlvl; 1357 int error = 0; 1358 1359 if (!(flags & DNODE_FIND_HAVELOCK)) 1360 rw_enter(&dn->dn_struct_rwlock, RW_READER); 1361 1362 if (dn->dn_phys->dn_nlevels == 0) { 1363 error = ESRCH; 1364 goto out; 1365 } 1366 1367 if (dn->dn_datablkshift == 0) { 1368 if (*offset < dn->dn_datablksz) { 1369 if (flags & DNODE_FIND_HOLE) 1370 *offset = dn->dn_datablksz; 1371 } else { 1372 error = ESRCH; 1373 } 1374 goto out; 1375 } 1376 1377 maxlvl = dn->dn_phys->dn_nlevels; 1378 1379 for (lvl = minlvl; lvl <= maxlvl; lvl++) { 1380 error = dnode_next_offset_level(dn, 1381 flags, offset, lvl, blkfill, txg); 1382 if (error != ESRCH) 1383 break; 1384 } 1385 1386 while (error == 0 && --lvl >= minlvl) { 1387 error = dnode_next_offset_level(dn, 1388 flags, offset, lvl, blkfill, txg); 1389 } 1390 1391 if (error == 0 && (flags & DNODE_FIND_BACKWARDS ? 1392 initial_offset < *offset : initial_offset > *offset)) 1393 error = ESRCH; 1394 out: 1395 if (!(flags & DNODE_FIND_HAVELOCK)) 1396 rw_exit(&dn->dn_struct_rwlock); 1397 1398 return (error); 1399 } 1400