1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2019 by Delphix. All rights reserved. 24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 25 * Copyright (c) 2014 Integros [integros.com] 26 * Copyright 2017 RackTop Systems. 27 */ 28 29 #include <sys/zfs_context.h> 30 #include <sys/dbuf.h> 31 #include <sys/dnode.h> 32 #include <sys/dmu.h> 33 #include <sys/dmu_impl.h> 34 #include <sys/dmu_tx.h> 35 #include <sys/dmu_objset.h> 36 #include <sys/dsl_dir.h> 37 #include <sys/dsl_dataset.h> 38 #include <sys/spa.h> 39 #include <sys/zio.h> 40 #include <sys/dmu_zfetch.h> 41 #include <sys/range_tree.h> 42 #include <sys/zfs_project.h> 43 44 dnode_stats_t dnode_stats = { 45 { "dnode_hold_dbuf_hold", KSTAT_DATA_UINT64 }, 46 { "dnode_hold_dbuf_read", KSTAT_DATA_UINT64 }, 47 { "dnode_hold_alloc_hits", KSTAT_DATA_UINT64 }, 48 { "dnode_hold_alloc_misses", KSTAT_DATA_UINT64 }, 49 { "dnode_hold_alloc_interior", KSTAT_DATA_UINT64 }, 50 { "dnode_hold_alloc_lock_retry", KSTAT_DATA_UINT64 }, 51 { "dnode_hold_alloc_lock_misses", KSTAT_DATA_UINT64 }, 52 { "dnode_hold_alloc_type_none", KSTAT_DATA_UINT64 }, 53 { "dnode_hold_free_hits", KSTAT_DATA_UINT64 }, 54 { "dnode_hold_free_misses", KSTAT_DATA_UINT64 }, 55 { "dnode_hold_free_lock_misses", KSTAT_DATA_UINT64 }, 56 { "dnode_hold_free_lock_retry", KSTAT_DATA_UINT64 }, 57 { "dnode_hold_free_overflow", KSTAT_DATA_UINT64 }, 58 { "dnode_hold_free_refcount", KSTAT_DATA_UINT64 }, 59 { "dnode_free_interior_lock_retry", KSTAT_DATA_UINT64 }, 60 { "dnode_allocate", KSTAT_DATA_UINT64 }, 61 { "dnode_reallocate", KSTAT_DATA_UINT64 }, 62 { "dnode_buf_evict", KSTAT_DATA_UINT64 }, 63 { "dnode_alloc_next_chunk", KSTAT_DATA_UINT64 }, 64 { "dnode_alloc_race", KSTAT_DATA_UINT64 }, 65 { "dnode_alloc_next_block", KSTAT_DATA_UINT64 }, 66 { "dnode_move_invalid", KSTAT_DATA_UINT64 }, 67 { "dnode_move_recheck1", KSTAT_DATA_UINT64 }, 68 { "dnode_move_recheck2", KSTAT_DATA_UINT64 }, 69 { "dnode_move_special", KSTAT_DATA_UINT64 }, 70 { "dnode_move_handle", KSTAT_DATA_UINT64 }, 71 { "dnode_move_rwlock", KSTAT_DATA_UINT64 }, 72 { "dnode_move_active", KSTAT_DATA_UINT64 }, 73 }; 74 75 static kstat_t *dnode_ksp; 76 static kmem_cache_t *dnode_cache; 77 78 static dnode_phys_t dnode_phys_zero; 79 80 int zfs_default_bs = SPA_MINBLOCKSHIFT; 81 int zfs_default_ibs = DN_MAX_INDBLKSHIFT; 82 83 #ifdef _KERNEL 84 static kmem_cbrc_t dnode_move(void *, void *, size_t, void *); 85 #endif /* _KERNEL */ 86 87 static int 88 dbuf_compare(const void *x1, const void *x2) 89 { 90 const dmu_buf_impl_t *d1 = x1; 91 const dmu_buf_impl_t *d2 = x2; 92 93 int cmp = TREE_CMP(d1->db_level, d2->db_level); 94 if (likely(cmp)) 95 return (cmp); 96 97 cmp = TREE_CMP(d1->db_blkid, d2->db_blkid); 98 if (likely(cmp)) 99 return (cmp); 100 101 if (d1->db_state == DB_SEARCH) { 102 ASSERT3S(d2->db_state, !=, DB_SEARCH); 103 return (-1); 104 } else if (d2->db_state == DB_SEARCH) { 105 ASSERT3S(d1->db_state, !=, DB_SEARCH); 106 return (1); 107 } 108 109 return (TREE_PCMP(d1, d2)); 110 } 111 112 /* ARGSUSED */ 113 static int 114 dnode_cons(void *arg, void *unused, int kmflag) 115 { 116 dnode_t *dn = arg; 117 int i; 118 119 rw_init(&dn->dn_struct_rwlock, NULL, RW_DEFAULT, NULL); 120 mutex_init(&dn->dn_mtx, NULL, MUTEX_DEFAULT, NULL); 121 mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL); 122 cv_init(&dn->dn_notxholds, NULL, CV_DEFAULT, NULL); 123 cv_init(&dn->dn_nodnholds, NULL, CV_DEFAULT, NULL); 124 125 /* 126 * Every dbuf has a reference, and dropping a tracked reference is 127 * O(number of references), so don't track dn_holds. 128 */ 129 zfs_refcount_create_untracked(&dn->dn_holds); 130 zfs_refcount_create(&dn->dn_tx_holds); 131 list_link_init(&dn->dn_link); 132 133 bzero(&dn->dn_next_nblkptr[0], sizeof (dn->dn_next_nblkptr)); 134 bzero(&dn->dn_next_nlevels[0], sizeof (dn->dn_next_nlevels)); 135 bzero(&dn->dn_next_indblkshift[0], sizeof (dn->dn_next_indblkshift)); 136 bzero(&dn->dn_next_bonustype[0], sizeof (dn->dn_next_bonustype)); 137 bzero(&dn->dn_rm_spillblk[0], sizeof (dn->dn_rm_spillblk)); 138 bzero(&dn->dn_next_bonuslen[0], sizeof (dn->dn_next_bonuslen)); 139 bzero(&dn->dn_next_blksz[0], sizeof (dn->dn_next_blksz)); 140 bzero(&dn->dn_next_maxblkid[0], sizeof (dn->dn_next_maxblkid)); 141 142 for (i = 0; i < TXG_SIZE; i++) { 143 multilist_link_init(&dn->dn_dirty_link[i]); 144 dn->dn_free_ranges[i] = NULL; 145 list_create(&dn->dn_dirty_records[i], 146 sizeof (dbuf_dirty_record_t), 147 offsetof(dbuf_dirty_record_t, dr_dirty_node)); 148 } 149 150 dn->dn_allocated_txg = 0; 151 dn->dn_free_txg = 0; 152 dn->dn_assigned_txg = 0; 153 dn->dn_dirty_txg = 0; 154 dn->dn_dirtyctx = 0; 155 dn->dn_dirtyctx_firstset = NULL; 156 dn->dn_bonus = NULL; 157 dn->dn_have_spill = B_FALSE; 158 dn->dn_zio = NULL; 159 dn->dn_oldused = 0; 160 dn->dn_oldflags = 0; 161 dn->dn_olduid = 0; 162 dn->dn_oldgid = 0; 163 dn->dn_oldprojid = ZFS_DEFAULT_PROJID; 164 dn->dn_newuid = 0; 165 dn->dn_newgid = 0; 166 dn->dn_newprojid = ZFS_DEFAULT_PROJID; 167 dn->dn_id_flags = 0; 168 169 dn->dn_dbufs_count = 0; 170 avl_create(&dn->dn_dbufs, dbuf_compare, sizeof (dmu_buf_impl_t), 171 offsetof(dmu_buf_impl_t, db_link)); 172 173 dn->dn_moved = 0; 174 return (0); 175 } 176 177 /* ARGSUSED */ 178 static void 179 dnode_dest(void *arg, void *unused) 180 { 181 int i; 182 dnode_t *dn = arg; 183 184 rw_destroy(&dn->dn_struct_rwlock); 185 mutex_destroy(&dn->dn_mtx); 186 mutex_destroy(&dn->dn_dbufs_mtx); 187 cv_destroy(&dn->dn_notxholds); 188 cv_destroy(&dn->dn_nodnholds); 189 zfs_refcount_destroy(&dn->dn_holds); 190 zfs_refcount_destroy(&dn->dn_tx_holds); 191 ASSERT(!list_link_active(&dn->dn_link)); 192 193 for (i = 0; i < TXG_SIZE; i++) { 194 ASSERT(!multilist_link_active(&dn->dn_dirty_link[i])); 195 ASSERT3P(dn->dn_free_ranges[i], ==, NULL); 196 list_destroy(&dn->dn_dirty_records[i]); 197 ASSERT0(dn->dn_next_nblkptr[i]); 198 ASSERT0(dn->dn_next_nlevels[i]); 199 ASSERT0(dn->dn_next_indblkshift[i]); 200 ASSERT0(dn->dn_next_bonustype[i]); 201 ASSERT0(dn->dn_rm_spillblk[i]); 202 ASSERT0(dn->dn_next_bonuslen[i]); 203 ASSERT0(dn->dn_next_blksz[i]); 204 ASSERT0(dn->dn_next_maxblkid[i]); 205 } 206 207 ASSERT0(dn->dn_allocated_txg); 208 ASSERT0(dn->dn_free_txg); 209 ASSERT0(dn->dn_assigned_txg); 210 ASSERT0(dn->dn_dirty_txg); 211 ASSERT0(dn->dn_dirtyctx); 212 ASSERT3P(dn->dn_dirtyctx_firstset, ==, NULL); 213 ASSERT3P(dn->dn_bonus, ==, NULL); 214 ASSERT(!dn->dn_have_spill); 215 ASSERT3P(dn->dn_zio, ==, NULL); 216 ASSERT0(dn->dn_oldused); 217 ASSERT0(dn->dn_oldflags); 218 ASSERT0(dn->dn_olduid); 219 ASSERT0(dn->dn_oldgid); 220 ASSERT0(dn->dn_oldprojid); 221 ASSERT0(dn->dn_newuid); 222 ASSERT0(dn->dn_newgid); 223 ASSERT0(dn->dn_newprojid); 224 ASSERT0(dn->dn_id_flags); 225 226 ASSERT0(dn->dn_dbufs_count); 227 avl_destroy(&dn->dn_dbufs); 228 } 229 230 void 231 dnode_init(void) 232 { 233 ASSERT(dnode_cache == NULL); 234 dnode_cache = kmem_cache_create("dnode_t", 235 sizeof (dnode_t), 236 0, dnode_cons, dnode_dest, NULL, NULL, NULL, 0); 237 #ifdef _KERNEL 238 kmem_cache_set_move(dnode_cache, dnode_move); 239 240 dnode_ksp = kstat_create("zfs", 0, "dnodestats", "misc", 241 KSTAT_TYPE_NAMED, sizeof (dnode_stats) / sizeof (kstat_named_t), 242 KSTAT_FLAG_VIRTUAL); 243 if (dnode_ksp != NULL) { 244 dnode_ksp->ks_data = &dnode_stats; 245 kstat_install(dnode_ksp); 246 } 247 #endif /* _KERNEL */ 248 } 249 250 void 251 dnode_fini(void) 252 { 253 if (dnode_ksp != NULL) { 254 kstat_delete(dnode_ksp); 255 dnode_ksp = NULL; 256 } 257 258 kmem_cache_destroy(dnode_cache); 259 dnode_cache = NULL; 260 } 261 262 263 #ifdef ZFS_DEBUG 264 void 265 dnode_verify(dnode_t *dn) 266 { 267 int drop_struct_lock = FALSE; 268 269 ASSERT(dn->dn_phys); 270 ASSERT(dn->dn_objset); 271 ASSERT(dn->dn_handle->dnh_dnode == dn); 272 273 ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); 274 275 if (!(zfs_flags & ZFS_DEBUG_DNODE_VERIFY)) 276 return; 277 278 if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { 279 rw_enter(&dn->dn_struct_rwlock, RW_READER); 280 drop_struct_lock = TRUE; 281 } 282 if (dn->dn_phys->dn_type != DMU_OT_NONE || dn->dn_allocated_txg != 0) { 283 int i; 284 int max_bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots); 285 ASSERT3U(dn->dn_indblkshift, >=, 0); 286 ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT); 287 if (dn->dn_datablkshift) { 288 ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT); 289 ASSERT3U(dn->dn_datablkshift, <=, SPA_MAXBLOCKSHIFT); 290 ASSERT3U(1<<dn->dn_datablkshift, ==, dn->dn_datablksz); 291 } 292 ASSERT3U(dn->dn_nlevels, <=, 30); 293 ASSERT(DMU_OT_IS_VALID(dn->dn_type)); 294 ASSERT3U(dn->dn_nblkptr, >=, 1); 295 ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 296 ASSERT3U(dn->dn_bonuslen, <=, max_bonuslen); 297 ASSERT3U(dn->dn_datablksz, ==, 298 dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); 299 ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0); 300 ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) + 301 dn->dn_bonuslen, <=, max_bonuslen); 302 for (i = 0; i < TXG_SIZE; i++) { 303 ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels); 304 } 305 } 306 if (dn->dn_phys->dn_type != DMU_OT_NONE) 307 ASSERT3U(dn->dn_phys->dn_nlevels, <=, dn->dn_nlevels); 308 ASSERT(DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_dbuf != NULL); 309 if (dn->dn_dbuf != NULL) { 310 ASSERT3P(dn->dn_phys, ==, 311 (dnode_phys_t *)dn->dn_dbuf->db.db_data + 312 (dn->dn_object % (dn->dn_dbuf->db.db_size >> DNODE_SHIFT))); 313 } 314 if (drop_struct_lock) 315 rw_exit(&dn->dn_struct_rwlock); 316 } 317 #endif 318 319 void 320 dnode_byteswap(dnode_phys_t *dnp) 321 { 322 uint64_t *buf64 = (void*)&dnp->dn_blkptr; 323 int i; 324 325 if (dnp->dn_type == DMU_OT_NONE) { 326 bzero(dnp, sizeof (dnode_phys_t)); 327 return; 328 } 329 330 dnp->dn_datablkszsec = BSWAP_16(dnp->dn_datablkszsec); 331 dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen); 332 dnp->dn_extra_slots = BSWAP_8(dnp->dn_extra_slots); 333 dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid); 334 dnp->dn_used = BSWAP_64(dnp->dn_used); 335 336 /* 337 * dn_nblkptr is only one byte, so it's OK to read it in either 338 * byte order. We can't read dn_bouslen. 339 */ 340 ASSERT(dnp->dn_indblkshift <= SPA_MAXBLOCKSHIFT); 341 ASSERT(dnp->dn_nblkptr <= DN_MAX_NBLKPTR); 342 for (i = 0; i < dnp->dn_nblkptr * sizeof (blkptr_t)/8; i++) 343 buf64[i] = BSWAP_64(buf64[i]); 344 345 /* 346 * OK to check dn_bonuslen for zero, because it won't matter if 347 * we have the wrong byte order. This is necessary because the 348 * dnode dnode is smaller than a regular dnode. 349 */ 350 if (dnp->dn_bonuslen != 0) { 351 dmu_object_byteswap_t byteswap; 352 ASSERT(DMU_OT_IS_VALID(dnp->dn_bonustype)); 353 byteswap = DMU_OT_BYTESWAP(dnp->dn_bonustype); 354 dmu_ot_byteswap[byteswap].ob_func(DN_BONUS(dnp), 355 DN_MAX_BONUS_LEN(dnp)); 356 } 357 358 /* Swap SPILL block if we have one */ 359 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) 360 byteswap_uint64_array(DN_SPILL_BLKPTR(dnp), sizeof (blkptr_t)); 361 362 } 363 364 void 365 dnode_buf_byteswap(void *vbuf, size_t size) 366 { 367 int i = 0; 368 369 ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT)); 370 ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0); 371 372 while (i < size) { 373 dnode_phys_t *dnp = (void *)(((char *)vbuf) + i); 374 dnode_byteswap(dnp); 375 376 i += DNODE_MIN_SIZE; 377 if (dnp->dn_type != DMU_OT_NONE) 378 i += dnp->dn_extra_slots * DNODE_MIN_SIZE; 379 } 380 } 381 382 void 383 dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx) 384 { 385 ASSERT3U(zfs_refcount_count(&dn->dn_holds), >=, 1); 386 387 dnode_setdirty(dn, tx); 388 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 389 ASSERT3U(newsize, <=, DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots) - 390 (dn->dn_nblkptr-1) * sizeof (blkptr_t)); 391 dn->dn_bonuslen = newsize; 392 if (newsize == 0) 393 dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = DN_ZERO_BONUSLEN; 394 else 395 dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; 396 rw_exit(&dn->dn_struct_rwlock); 397 } 398 399 void 400 dnode_setbonus_type(dnode_t *dn, dmu_object_type_t newtype, dmu_tx_t *tx) 401 { 402 ASSERT3U(zfs_refcount_count(&dn->dn_holds), >=, 1); 403 dnode_setdirty(dn, tx); 404 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 405 dn->dn_bonustype = newtype; 406 dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; 407 rw_exit(&dn->dn_struct_rwlock); 408 } 409 410 void 411 dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx) 412 { 413 ASSERT3U(zfs_refcount_count(&dn->dn_holds), >=, 1); 414 ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); 415 dnode_setdirty(dn, tx); 416 dn->dn_rm_spillblk[tx->tx_txg&TXG_MASK] = DN_KILL_SPILLBLK; 417 dn->dn_have_spill = B_FALSE; 418 } 419 420 static void 421 dnode_setdblksz(dnode_t *dn, int size) 422 { 423 ASSERT0(P2PHASE(size, SPA_MINBLOCKSIZE)); 424 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 425 ASSERT3U(size, >=, SPA_MINBLOCKSIZE); 426 ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, 427 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); 428 dn->dn_datablksz = size; 429 dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; 430 dn->dn_datablkshift = ISP2(size) ? highbit64(size - 1) : 0; 431 } 432 433 static dnode_t * 434 dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, 435 uint64_t object, dnode_handle_t *dnh) 436 { 437 dnode_t *dn; 438 439 dn = kmem_cache_alloc(dnode_cache, KM_SLEEP); 440 #ifdef _KERNEL 441 ASSERT(!POINTER_IS_VALID(dn->dn_objset)); 442 #endif /* _KERNEL */ 443 dn->dn_moved = 0; 444 445 /* 446 * Defer setting dn_objset until the dnode is ready to be a candidate 447 * for the dnode_move() callback. 448 */ 449 dn->dn_object = object; 450 dn->dn_dbuf = db; 451 dn->dn_handle = dnh; 452 dn->dn_phys = dnp; 453 454 if (dnp->dn_datablkszsec) { 455 dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 456 } else { 457 dn->dn_datablksz = 0; 458 dn->dn_datablkszsec = 0; 459 dn->dn_datablkshift = 0; 460 } 461 dn->dn_indblkshift = dnp->dn_indblkshift; 462 dn->dn_nlevels = dnp->dn_nlevels; 463 dn->dn_type = dnp->dn_type; 464 dn->dn_nblkptr = dnp->dn_nblkptr; 465 dn->dn_checksum = dnp->dn_checksum; 466 dn->dn_compress = dnp->dn_compress; 467 dn->dn_bonustype = dnp->dn_bonustype; 468 dn->dn_bonuslen = dnp->dn_bonuslen; 469 dn->dn_num_slots = dnp->dn_extra_slots + 1; 470 dn->dn_maxblkid = dnp->dn_maxblkid; 471 dn->dn_have_spill = ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0); 472 dn->dn_id_flags = 0; 473 474 dmu_zfetch_init(&dn->dn_zfetch, dn); 475 476 ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); 477 ASSERT(zrl_is_locked(&dnh->dnh_zrlock)); 478 ASSERT(!DN_SLOT_IS_PTR(dnh->dnh_dnode)); 479 480 mutex_enter(&os->os_lock); 481 482 /* 483 * Exclude special dnodes from os_dnodes so an empty os_dnodes 484 * signifies that the special dnodes have no references from 485 * their children (the entries in os_dnodes). This allows 486 * dnode_destroy() to easily determine if the last child has 487 * been removed and then complete eviction of the objset. 488 */ 489 if (!DMU_OBJECT_IS_SPECIAL(object)) 490 list_insert_head(&os->os_dnodes, dn); 491 membar_producer(); 492 493 /* 494 * Everything else must be valid before assigning dn_objset 495 * makes the dnode eligible for dnode_move(). 496 */ 497 dn->dn_objset = os; 498 499 dnh->dnh_dnode = dn; 500 mutex_exit(&os->os_lock); 501 502 arc_space_consume(sizeof (dnode_t), ARC_SPACE_OTHER); 503 504 return (dn); 505 } 506 507 /* 508 * Caller must be holding the dnode handle, which is released upon return. 509 */ 510 static void 511 dnode_destroy(dnode_t *dn) 512 { 513 objset_t *os = dn->dn_objset; 514 boolean_t complete_os_eviction = B_FALSE; 515 516 ASSERT((dn->dn_id_flags & DN_ID_NEW_EXIST) == 0); 517 518 mutex_enter(&os->os_lock); 519 POINTER_INVALIDATE(&dn->dn_objset); 520 if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 521 list_remove(&os->os_dnodes, dn); 522 complete_os_eviction = 523 list_is_empty(&os->os_dnodes) && 524 list_link_active(&os->os_evicting_node); 525 } 526 mutex_exit(&os->os_lock); 527 528 /* the dnode can no longer move, so we can release the handle */ 529 if (!zrl_is_locked(&dn->dn_handle->dnh_zrlock)) 530 zrl_remove(&dn->dn_handle->dnh_zrlock); 531 532 dn->dn_allocated_txg = 0; 533 dn->dn_free_txg = 0; 534 dn->dn_assigned_txg = 0; 535 dn->dn_dirty_txg = 0; 536 537 dn->dn_dirtyctx = 0; 538 if (dn->dn_dirtyctx_firstset != NULL) { 539 kmem_free(dn->dn_dirtyctx_firstset, 1); 540 dn->dn_dirtyctx_firstset = NULL; 541 } 542 if (dn->dn_bonus != NULL) { 543 mutex_enter(&dn->dn_bonus->db_mtx); 544 dbuf_destroy(dn->dn_bonus); 545 dn->dn_bonus = NULL; 546 } 547 dn->dn_zio = NULL; 548 549 dn->dn_have_spill = B_FALSE; 550 dn->dn_oldused = 0; 551 dn->dn_oldflags = 0; 552 dn->dn_olduid = 0; 553 dn->dn_oldgid = 0; 554 dn->dn_oldprojid = ZFS_DEFAULT_PROJID; 555 dn->dn_newuid = 0; 556 dn->dn_newgid = 0; 557 dn->dn_newprojid = ZFS_DEFAULT_PROJID; 558 dn->dn_id_flags = 0; 559 560 dmu_zfetch_fini(&dn->dn_zfetch); 561 kmem_cache_free(dnode_cache, dn); 562 arc_space_return(sizeof (dnode_t), ARC_SPACE_OTHER); 563 564 if (complete_os_eviction) 565 dmu_objset_evict_done(os); 566 } 567 568 void 569 dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 570 dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx) 571 { 572 int i; 573 574 ASSERT3U(dn_slots, >, 0); 575 ASSERT3U(dn_slots << DNODE_SHIFT, <=, 576 spa_maxdnodesize(dmu_objset_spa(dn->dn_objset))); 577 ASSERT3U(blocksize, <=, 578 spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); 579 if (blocksize == 0) 580 blocksize = 1 << zfs_default_bs; 581 else 582 blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); 583 584 if (ibs == 0) 585 ibs = zfs_default_ibs; 586 587 ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); 588 589 dprintf("os=%p obj=%" PRIu64 " txg=%" PRIu64 590 " blocksize=%d ibs=%d dn_slots=%d\n", 591 dn->dn_objset, dn->dn_object, tx->tx_txg, blocksize, ibs, dn_slots); 592 DNODE_STAT_BUMP(dnode_allocate); 593 594 ASSERT(dn->dn_type == DMU_OT_NONE); 595 ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); 596 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); 597 ASSERT(ot != DMU_OT_NONE); 598 ASSERT(DMU_OT_IS_VALID(ot)); 599 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 600 (bonustype == DMU_OT_SA && bonuslen == 0) || 601 (bonustype != DMU_OT_NONE && bonuslen != 0)); 602 ASSERT(DMU_OT_IS_VALID(bonustype)); 603 ASSERT3U(bonuslen, <=, DN_SLOTS_TO_BONUSLEN(dn_slots)); 604 ASSERT(dn->dn_type == DMU_OT_NONE); 605 ASSERT0(dn->dn_maxblkid); 606 ASSERT0(dn->dn_allocated_txg); 607 ASSERT0(dn->dn_dirty_txg); 608 ASSERT0(dn->dn_assigned_txg); 609 ASSERT(zfs_refcount_is_zero(&dn->dn_tx_holds)); 610 ASSERT3U(zfs_refcount_count(&dn->dn_holds), <=, 1); 611 ASSERT(avl_is_empty(&dn->dn_dbufs)); 612 613 for (i = 0; i < TXG_SIZE; i++) { 614 ASSERT0(dn->dn_next_nblkptr[i]); 615 ASSERT0(dn->dn_next_nlevels[i]); 616 ASSERT0(dn->dn_next_indblkshift[i]); 617 ASSERT0(dn->dn_next_bonuslen[i]); 618 ASSERT0(dn->dn_next_bonustype[i]); 619 ASSERT0(dn->dn_rm_spillblk[i]); 620 ASSERT0(dn->dn_next_blksz[i]); 621 ASSERT0(dn->dn_next_maxblkid[i]); 622 ASSERT(!multilist_link_active(&dn->dn_dirty_link[i])); 623 ASSERT3P(list_head(&dn->dn_dirty_records[i]), ==, NULL); 624 ASSERT3P(dn->dn_free_ranges[i], ==, NULL); 625 } 626 627 dn->dn_type = ot; 628 dnode_setdblksz(dn, blocksize); 629 dn->dn_indblkshift = ibs; 630 dn->dn_nlevels = 1; 631 dn->dn_num_slots = dn_slots; 632 if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ 633 dn->dn_nblkptr = 1; 634 else { 635 dn->dn_nblkptr = MIN(DN_MAX_NBLKPTR, 636 1 + ((DN_SLOTS_TO_BONUSLEN(dn_slots) - bonuslen) >> 637 SPA_BLKPTRSHIFT)); 638 } 639 640 dn->dn_bonustype = bonustype; 641 dn->dn_bonuslen = bonuslen; 642 dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 643 dn->dn_compress = ZIO_COMPRESS_INHERIT; 644 dn->dn_dirtyctx = 0; 645 646 dn->dn_free_txg = 0; 647 if (dn->dn_dirtyctx_firstset) { 648 kmem_free(dn->dn_dirtyctx_firstset, 1); 649 dn->dn_dirtyctx_firstset = NULL; 650 } 651 652 dn->dn_allocated_txg = tx->tx_txg; 653 dn->dn_id_flags = 0; 654 655 dnode_setdirty(dn, tx); 656 dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs; 657 dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; 658 dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; 659 dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz; 660 } 661 662 void 663 dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 664 dmu_object_type_t bonustype, int bonuslen, int dn_slots, 665 boolean_t keep_spill, dmu_tx_t *tx) 666 { 667 int nblkptr; 668 669 ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); 670 ASSERT3U(blocksize, <=, 671 spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); 672 ASSERT0(blocksize % SPA_MINBLOCKSIZE); 673 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); 674 ASSERT(tx->tx_txg != 0); 675 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || 676 (bonustype != DMU_OT_NONE && bonuslen != 0) || 677 (bonustype == DMU_OT_SA && bonuslen == 0)); 678 ASSERT(DMU_OT_IS_VALID(bonustype)); 679 ASSERT3U(bonuslen, <=, 680 DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(dn->dn_objset)))); 681 ASSERT3U(bonuslen, <=, DN_BONUS_SIZE(dn_slots << DNODE_SHIFT)); 682 683 dnode_free_interior_slots(dn); 684 DNODE_STAT_BUMP(dnode_reallocate); 685 686 /* clean up any unreferenced dbufs */ 687 dnode_evict_dbufs(dn); 688 689 dn->dn_id_flags = 0; 690 691 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 692 dnode_setdirty(dn, tx); 693 if (dn->dn_datablksz != blocksize) { 694 /* change blocksize */ 695 ASSERT(dn->dn_maxblkid == 0 && 696 (BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || 697 dnode_block_freed(dn, 0))); 698 dnode_setdblksz(dn, blocksize); 699 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize; 700 } 701 if (dn->dn_bonuslen != bonuslen) 702 dn->dn_next_bonuslen[tx->tx_txg&TXG_MASK] = bonuslen; 703 704 if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ 705 nblkptr = 1; 706 else 707 nblkptr = MIN(DN_MAX_NBLKPTR, 708 1 + ((DN_SLOTS_TO_BONUSLEN(dn_slots) - bonuslen) >> 709 SPA_BLKPTRSHIFT)); 710 if (dn->dn_bonustype != bonustype) 711 dn->dn_next_bonustype[tx->tx_txg&TXG_MASK] = bonustype; 712 if (dn->dn_nblkptr != nblkptr) 713 dn->dn_next_nblkptr[tx->tx_txg&TXG_MASK] = nblkptr; 714 if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR && !keep_spill) { 715 dbuf_rm_spill(dn, tx); 716 dnode_rm_spill(dn, tx); 717 } 718 rw_exit(&dn->dn_struct_rwlock); 719 720 /* change type */ 721 dn->dn_type = ot; 722 723 /* change bonus size and type */ 724 mutex_enter(&dn->dn_mtx); 725 dn->dn_bonustype = bonustype; 726 dn->dn_bonuslen = bonuslen; 727 dn->dn_num_slots = dn_slots; 728 dn->dn_nblkptr = nblkptr; 729 dn->dn_checksum = ZIO_CHECKSUM_INHERIT; 730 dn->dn_compress = ZIO_COMPRESS_INHERIT; 731 ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); 732 733 /* fix up the bonus db_size */ 734 if (dn->dn_bonus) { 735 dn->dn_bonus->db.db_size = 736 DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots) - 737 (dn->dn_nblkptr - 1) * sizeof (blkptr_t); 738 ASSERT(dn->dn_bonuslen <= dn->dn_bonus->db.db_size); 739 } 740 741 dn->dn_allocated_txg = tx->tx_txg; 742 mutex_exit(&dn->dn_mtx); 743 } 744 745 #ifdef _KERNEL 746 static void 747 dnode_move_impl(dnode_t *odn, dnode_t *ndn) 748 { 749 int i; 750 751 ASSERT(!RW_LOCK_HELD(&odn->dn_struct_rwlock)); 752 ASSERT(MUTEX_NOT_HELD(&odn->dn_mtx)); 753 ASSERT(MUTEX_NOT_HELD(&odn->dn_dbufs_mtx)); 754 ASSERT(!RW_LOCK_HELD(&odn->dn_zfetch.zf_rwlock)); 755 756 /* Copy fields. */ 757 ndn->dn_objset = odn->dn_objset; 758 ndn->dn_object = odn->dn_object; 759 ndn->dn_dbuf = odn->dn_dbuf; 760 ndn->dn_handle = odn->dn_handle; 761 ndn->dn_phys = odn->dn_phys; 762 ndn->dn_type = odn->dn_type; 763 ndn->dn_bonuslen = odn->dn_bonuslen; 764 ndn->dn_bonustype = odn->dn_bonustype; 765 ndn->dn_nblkptr = odn->dn_nblkptr; 766 ndn->dn_checksum = odn->dn_checksum; 767 ndn->dn_compress = odn->dn_compress; 768 ndn->dn_nlevels = odn->dn_nlevels; 769 ndn->dn_indblkshift = odn->dn_indblkshift; 770 ndn->dn_datablkshift = odn->dn_datablkshift; 771 ndn->dn_datablkszsec = odn->dn_datablkszsec; 772 ndn->dn_datablksz = odn->dn_datablksz; 773 ndn->dn_maxblkid = odn->dn_maxblkid; 774 ndn->dn_num_slots = odn->dn_num_slots; 775 bcopy(&odn->dn_next_type[0], &ndn->dn_next_type[0], 776 sizeof (odn->dn_next_type)); 777 bcopy(&odn->dn_next_nblkptr[0], &ndn->dn_next_nblkptr[0], 778 sizeof (odn->dn_next_nblkptr)); 779 bcopy(&odn->dn_next_nlevels[0], &ndn->dn_next_nlevels[0], 780 sizeof (odn->dn_next_nlevels)); 781 bcopy(&odn->dn_next_indblkshift[0], &ndn->dn_next_indblkshift[0], 782 sizeof (odn->dn_next_indblkshift)); 783 bcopy(&odn->dn_next_bonustype[0], &ndn->dn_next_bonustype[0], 784 sizeof (odn->dn_next_bonustype)); 785 bcopy(&odn->dn_rm_spillblk[0], &ndn->dn_rm_spillblk[0], 786 sizeof (odn->dn_rm_spillblk)); 787 bcopy(&odn->dn_next_bonuslen[0], &ndn->dn_next_bonuslen[0], 788 sizeof (odn->dn_next_bonuslen)); 789 bcopy(&odn->dn_next_blksz[0], &ndn->dn_next_blksz[0], 790 sizeof (odn->dn_next_blksz)); 791 bcopy(&odn->dn_next_maxblkid[0], &ndn->dn_next_maxblkid[0], 792 sizeof (odn->dn_next_maxblkid)); 793 for (i = 0; i < TXG_SIZE; i++) { 794 list_move_tail(&ndn->dn_dirty_records[i], 795 &odn->dn_dirty_records[i]); 796 } 797 bcopy(&odn->dn_free_ranges[0], &ndn->dn_free_ranges[0], 798 sizeof (odn->dn_free_ranges)); 799 ndn->dn_allocated_txg = odn->dn_allocated_txg; 800 ndn->dn_free_txg = odn->dn_free_txg; 801 ndn->dn_assigned_txg = odn->dn_assigned_txg; 802 ndn->dn_dirty_txg = odn->dn_dirty_txg; 803 ndn->dn_dirtyctx = odn->dn_dirtyctx; 804 ndn->dn_dirtyctx_firstset = odn->dn_dirtyctx_firstset; 805 ASSERT(zfs_refcount_count(&odn->dn_tx_holds) == 0); 806 zfs_refcount_transfer(&ndn->dn_holds, &odn->dn_holds); 807 ASSERT(avl_is_empty(&ndn->dn_dbufs)); 808 avl_swap(&ndn->dn_dbufs, &odn->dn_dbufs); 809 ndn->dn_dbufs_count = odn->dn_dbufs_count; 810 ndn->dn_bonus = odn->dn_bonus; 811 ndn->dn_have_spill = odn->dn_have_spill; 812 ndn->dn_zio = odn->dn_zio; 813 ndn->dn_oldused = odn->dn_oldused; 814 ndn->dn_oldflags = odn->dn_oldflags; 815 ndn->dn_olduid = odn->dn_olduid; 816 ndn->dn_oldgid = odn->dn_oldgid; 817 ndn->dn_oldprojid = odn->dn_oldprojid; 818 ndn->dn_newuid = odn->dn_newuid; 819 ndn->dn_newgid = odn->dn_newgid; 820 ndn->dn_newprojid = odn->dn_newprojid; 821 ndn->dn_id_flags = odn->dn_id_flags; 822 dmu_zfetch_init(&ndn->dn_zfetch, NULL); 823 list_move_tail(&ndn->dn_zfetch.zf_stream, &odn->dn_zfetch.zf_stream); 824 ndn->dn_zfetch.zf_dnode = odn->dn_zfetch.zf_dnode; 825 826 /* 827 * Update back pointers. Updating the handle fixes the back pointer of 828 * every descendant dbuf as well as the bonus dbuf. 829 */ 830 ASSERT(ndn->dn_handle->dnh_dnode == odn); 831 ndn->dn_handle->dnh_dnode = ndn; 832 if (ndn->dn_zfetch.zf_dnode == odn) { 833 ndn->dn_zfetch.zf_dnode = ndn; 834 } 835 836 /* 837 * Invalidate the original dnode by clearing all of its back pointers. 838 */ 839 odn->dn_dbuf = NULL; 840 odn->dn_handle = NULL; 841 avl_create(&odn->dn_dbufs, dbuf_compare, sizeof (dmu_buf_impl_t), 842 offsetof(dmu_buf_impl_t, db_link)); 843 odn->dn_dbufs_count = 0; 844 odn->dn_bonus = NULL; 845 odn->dn_zfetch.zf_dnode = NULL; 846 847 /* 848 * Set the low bit of the objset pointer to ensure that dnode_move() 849 * recognizes the dnode as invalid in any subsequent callback. 850 */ 851 POINTER_INVALIDATE(&odn->dn_objset); 852 853 /* 854 * Satisfy the destructor. 855 */ 856 for (i = 0; i < TXG_SIZE; i++) { 857 list_create(&odn->dn_dirty_records[i], 858 sizeof (dbuf_dirty_record_t), 859 offsetof(dbuf_dirty_record_t, dr_dirty_node)); 860 odn->dn_free_ranges[i] = NULL; 861 odn->dn_next_nlevels[i] = 0; 862 odn->dn_next_indblkshift[i] = 0; 863 odn->dn_next_bonustype[i] = 0; 864 odn->dn_rm_spillblk[i] = 0; 865 odn->dn_next_bonuslen[i] = 0; 866 odn->dn_next_blksz[i] = 0; 867 } 868 odn->dn_allocated_txg = 0; 869 odn->dn_free_txg = 0; 870 odn->dn_assigned_txg = 0; 871 odn->dn_dirty_txg = 0; 872 odn->dn_dirtyctx = 0; 873 odn->dn_dirtyctx_firstset = NULL; 874 odn->dn_have_spill = B_FALSE; 875 odn->dn_zio = NULL; 876 odn->dn_oldused = 0; 877 odn->dn_oldflags = 0; 878 odn->dn_olduid = 0; 879 odn->dn_oldgid = 0; 880 odn->dn_oldprojid = ZFS_DEFAULT_PROJID; 881 odn->dn_newuid = 0; 882 odn->dn_newgid = 0; 883 odn->dn_newprojid = ZFS_DEFAULT_PROJID; 884 odn->dn_id_flags = 0; 885 886 /* 887 * Mark the dnode. 888 */ 889 ndn->dn_moved = 1; 890 odn->dn_moved = (uint8_t)-1; 891 } 892 893 /*ARGSUSED*/ 894 static kmem_cbrc_t 895 dnode_move(void *buf, void *newbuf, size_t size, void *arg) 896 { 897 dnode_t *odn = buf, *ndn = newbuf; 898 objset_t *os; 899 int64_t refcount; 900 uint32_t dbufs; 901 902 /* 903 * The dnode is on the objset's list of known dnodes if the objset 904 * pointer is valid. We set the low bit of the objset pointer when 905 * freeing the dnode to invalidate it, and the memory patterns written 906 * by kmem (baddcafe and deadbeef) set at least one of the two low bits. 907 * A newly created dnode sets the objset pointer last of all to indicate 908 * that the dnode is known and in a valid state to be moved by this 909 * function. 910 */ 911 os = odn->dn_objset; 912 if (!POINTER_IS_VALID(os)) { 913 DNODE_STAT_BUMP(dnode_move_invalid); 914 return (KMEM_CBRC_DONT_KNOW); 915 } 916 917 /* 918 * Ensure that the objset does not go away during the move. 919 */ 920 rw_enter(&os_lock, RW_WRITER); 921 if (os != odn->dn_objset) { 922 rw_exit(&os_lock); 923 DNODE_STAT_BUMP(dnode_move_recheck1); 924 return (KMEM_CBRC_DONT_KNOW); 925 } 926 927 /* 928 * If the dnode is still valid, then so is the objset. We know that no 929 * valid objset can be freed while we hold os_lock, so we can safely 930 * ensure that the objset remains in use. 931 */ 932 mutex_enter(&os->os_lock); 933 934 /* 935 * Recheck the objset pointer in case the dnode was removed just before 936 * acquiring the lock. 937 */ 938 if (os != odn->dn_objset) { 939 mutex_exit(&os->os_lock); 940 rw_exit(&os_lock); 941 DNODE_STAT_BUMP(dnode_move_recheck2); 942 return (KMEM_CBRC_DONT_KNOW); 943 } 944 945 /* 946 * At this point we know that as long as we hold os->os_lock, the dnode 947 * cannot be freed and fields within the dnode can be safely accessed. 948 * The objset listing this dnode cannot go away as long as this dnode is 949 * on its list. 950 */ 951 rw_exit(&os_lock); 952 if (DMU_OBJECT_IS_SPECIAL(odn->dn_object)) { 953 mutex_exit(&os->os_lock); 954 DNODE_STAT_BUMP(dnode_move_special); 955 return (KMEM_CBRC_NO); 956 } 957 ASSERT(odn->dn_dbuf != NULL); /* only "special" dnodes have no parent */ 958 959 /* 960 * Lock the dnode handle to prevent the dnode from obtaining any new 961 * holds. This also prevents the descendant dbufs and the bonus dbuf 962 * from accessing the dnode, so that we can discount their holds. The 963 * handle is safe to access because we know that while the dnode cannot 964 * go away, neither can its handle. Once we hold dnh_zrlock, we can 965 * safely move any dnode referenced only by dbufs. 966 */ 967 if (!zrl_tryenter(&odn->dn_handle->dnh_zrlock)) { 968 mutex_exit(&os->os_lock); 969 DNODE_STAT_BUMP(dnode_move_handle); 970 return (KMEM_CBRC_LATER); 971 } 972 973 /* 974 * Ensure a consistent view of the dnode's holds and the dnode's dbufs. 975 * We need to guarantee that there is a hold for every dbuf in order to 976 * determine whether the dnode is actively referenced. Falsely matching 977 * a dbuf to an active hold would lead to an unsafe move. It's possible 978 * that a thread already having an active dnode hold is about to add a 979 * dbuf, and we can't compare hold and dbuf counts while the add is in 980 * progress. 981 */ 982 if (!rw_tryenter(&odn->dn_struct_rwlock, RW_WRITER)) { 983 zrl_exit(&odn->dn_handle->dnh_zrlock); 984 mutex_exit(&os->os_lock); 985 DNODE_STAT_BUMP(dnode_move_rwlock); 986 return (KMEM_CBRC_LATER); 987 } 988 989 /* 990 * A dbuf may be removed (evicted) without an active dnode hold. In that 991 * case, the dbuf count is decremented under the handle lock before the 992 * dbuf's hold is released. This order ensures that if we count the hold 993 * after the dbuf is removed but before its hold is released, we will 994 * treat the unmatched hold as active and exit safely. If we count the 995 * hold before the dbuf is removed, the hold is discounted, and the 996 * removal is blocked until the move completes. 997 */ 998 refcount = zfs_refcount_count(&odn->dn_holds); 999 ASSERT(refcount >= 0); 1000 dbufs = odn->dn_dbufs_count; 1001 1002 /* We can't have more dbufs than dnode holds. */ 1003 ASSERT3U(dbufs, <=, refcount); 1004 DTRACE_PROBE3(dnode__move, dnode_t *, odn, int64_t, refcount, 1005 uint32_t, dbufs); 1006 1007 if (refcount > dbufs) { 1008 rw_exit(&odn->dn_struct_rwlock); 1009 zrl_exit(&odn->dn_handle->dnh_zrlock); 1010 mutex_exit(&os->os_lock); 1011 DNODE_STAT_BUMP(dnode_move_active); 1012 return (KMEM_CBRC_LATER); 1013 } 1014 1015 rw_exit(&odn->dn_struct_rwlock); 1016 1017 /* 1018 * At this point we know that anyone with a hold on the dnode is not 1019 * actively referencing it. The dnode is known and in a valid state to 1020 * move. We're holding the locks needed to execute the critical section. 1021 */ 1022 dnode_move_impl(odn, ndn); 1023 1024 list_link_replace(&odn->dn_link, &ndn->dn_link); 1025 /* If the dnode was safe to move, the refcount cannot have changed. */ 1026 ASSERT(refcount == zfs_refcount_count(&ndn->dn_holds)); 1027 ASSERT(dbufs == ndn->dn_dbufs_count); 1028 zrl_exit(&ndn->dn_handle->dnh_zrlock); /* handle has moved */ 1029 mutex_exit(&os->os_lock); 1030 1031 return (KMEM_CBRC_YES); 1032 } 1033 #endif /* _KERNEL */ 1034 1035 static void 1036 dnode_slots_hold(dnode_children_t *children, int idx, int slots) 1037 { 1038 ASSERT3S(idx + slots, <=, DNODES_PER_BLOCK); 1039 1040 for (int i = idx; i < idx + slots; i++) { 1041 dnode_handle_t *dnh = &children->dnc_children[i]; 1042 zrl_add(&dnh->dnh_zrlock); 1043 } 1044 } 1045 1046 static void 1047 dnode_slots_rele(dnode_children_t *children, int idx, int slots) 1048 { 1049 ASSERT3S(idx + slots, <=, DNODES_PER_BLOCK); 1050 1051 for (int i = idx; i < idx + slots; i++) { 1052 dnode_handle_t *dnh = &children->dnc_children[i]; 1053 1054 if (zrl_is_locked(&dnh->dnh_zrlock)) 1055 zrl_exit(&dnh->dnh_zrlock); 1056 else 1057 zrl_remove(&dnh->dnh_zrlock); 1058 } 1059 } 1060 1061 static int 1062 dnode_slots_tryenter(dnode_children_t *children, int idx, int slots) 1063 { 1064 ASSERT3S(idx + slots, <=, DNODES_PER_BLOCK); 1065 1066 for (int i = idx; i < idx + slots; i++) { 1067 dnode_handle_t *dnh = &children->dnc_children[i]; 1068 1069 if (!zrl_tryenter(&dnh->dnh_zrlock)) { 1070 for (int j = idx; j < i; j++) { 1071 dnh = &children->dnc_children[j]; 1072 zrl_exit(&dnh->dnh_zrlock); 1073 } 1074 1075 return (0); 1076 } 1077 } 1078 1079 return (1); 1080 } 1081 1082 static void 1083 dnode_set_slots(dnode_children_t *children, int idx, int slots, void *ptr) 1084 { 1085 ASSERT3S(idx + slots, <=, DNODES_PER_BLOCK); 1086 1087 for (int i = idx; i < idx + slots; i++) { 1088 dnode_handle_t *dnh = &children->dnc_children[i]; 1089 dnh->dnh_dnode = ptr; 1090 } 1091 } 1092 1093 static boolean_t 1094 dnode_check_slots_free(dnode_children_t *children, int idx, int slots) 1095 { 1096 ASSERT3S(idx + slots, <=, DNODES_PER_BLOCK); 1097 1098 /* 1099 * If all dnode slots are either already free or 1100 * evictable return B_TRUE. 1101 */ 1102 for (int i = idx; i < idx + slots; i++) { 1103 dnode_handle_t *dnh = &children->dnc_children[i]; 1104 dnode_t *dn = dnh->dnh_dnode; 1105 1106 if (dn == DN_SLOT_FREE) { 1107 continue; 1108 } else if (DN_SLOT_IS_PTR(dn)) { 1109 mutex_enter(&dn->dn_mtx); 1110 boolean_t can_free = (dn->dn_type == DMU_OT_NONE && 1111 zfs_refcount_is_zero(&dn->dn_holds) && 1112 !DNODE_IS_DIRTY(dn)); 1113 mutex_exit(&dn->dn_mtx); 1114 1115 if (!can_free) 1116 return (B_FALSE); 1117 else 1118 continue; 1119 } else { 1120 return (B_FALSE); 1121 } 1122 } 1123 1124 return (B_TRUE); 1125 } 1126 1127 static void 1128 dnode_reclaim_slots(dnode_children_t *children, int idx, int slots) 1129 { 1130 ASSERT3S(idx + slots, <=, DNODES_PER_BLOCK); 1131 1132 for (int i = idx; i < idx + slots; i++) { 1133 dnode_handle_t *dnh = &children->dnc_children[i]; 1134 1135 ASSERT(zrl_is_locked(&dnh->dnh_zrlock)); 1136 1137 if (DN_SLOT_IS_PTR(dnh->dnh_dnode)) { 1138 ASSERT3S(dnh->dnh_dnode->dn_type, ==, DMU_OT_NONE); 1139 dnode_destroy(dnh->dnh_dnode); 1140 dnh->dnh_dnode = DN_SLOT_FREE; 1141 } 1142 } 1143 } 1144 1145 void 1146 dnode_free_interior_slots(dnode_t *dn) 1147 { 1148 dnode_children_t *children = dmu_buf_get_user(&dn->dn_dbuf->db); 1149 int epb = dn->dn_dbuf->db.db_size >> DNODE_SHIFT; 1150 int idx = (dn->dn_object & (epb - 1)) + 1; 1151 int slots = dn->dn_num_slots - 1; 1152 1153 if (slots == 0) 1154 return; 1155 1156 ASSERT3S(idx + slots, <=, DNODES_PER_BLOCK); 1157 1158 while (!dnode_slots_tryenter(children, idx, slots)) 1159 DNODE_STAT_BUMP(dnode_free_interior_lock_retry); 1160 1161 dnode_set_slots(children, idx, slots, DN_SLOT_FREE); 1162 dnode_slots_rele(children, idx, slots); 1163 } 1164 1165 void 1166 dnode_special_close(dnode_handle_t *dnh) 1167 { 1168 dnode_t *dn = dnh->dnh_dnode; 1169 1170 /* 1171 * Ensure dnode_rele_and_unlock() has released dn_mtx, after final 1172 * zfs_refcount_remove() 1173 */ 1174 mutex_enter(&dn->dn_mtx); 1175 if (zfs_refcount_count(&dn->dn_holds) > 0) 1176 cv_wait(&dn->dn_nodnholds, &dn->dn_mtx); 1177 mutex_exit(&dn->dn_mtx); 1178 ASSERT3U(zfs_refcount_count(&dn->dn_holds), ==, 0); 1179 1180 ASSERT(dn->dn_dbuf == NULL || 1181 dmu_buf_get_user(&dn->dn_dbuf->db) == NULL); 1182 zrl_add(&dnh->dnh_zrlock); 1183 dnode_destroy(dn); /* implicit zrl_remove() */ 1184 zrl_destroy(&dnh->dnh_zrlock); 1185 dnh->dnh_dnode = NULL; 1186 } 1187 1188 void 1189 dnode_special_open(objset_t *os, dnode_phys_t *dnp, uint64_t object, 1190 dnode_handle_t *dnh) 1191 { 1192 dnode_t *dn; 1193 1194 zrl_init(&dnh->dnh_zrlock); 1195 VERIFY3U(1, ==, zrl_tryenter(&dnh->dnh_zrlock)); 1196 1197 dn = dnode_create(os, dnp, NULL, object, dnh); 1198 DNODE_VERIFY(dn); 1199 1200 zrl_exit(&dnh->dnh_zrlock); 1201 } 1202 1203 static void 1204 dnode_buf_evict_async(void *dbu) 1205 { 1206 dnode_children_t *dnc = dbu; 1207 1208 DNODE_STAT_BUMP(dnode_buf_evict); 1209 1210 for (int i = 0; i < dnc->dnc_count; i++) { 1211 dnode_handle_t *dnh = &dnc->dnc_children[i]; 1212 dnode_t *dn; 1213 1214 /* 1215 * The dnode handle lock guards against the dnode moving to 1216 * another valid address, so there is no need here to guard 1217 * against changes to or from NULL. 1218 */ 1219 if (!DN_SLOT_IS_PTR(dnh->dnh_dnode)) { 1220 zrl_destroy(&dnh->dnh_zrlock); 1221 dnh->dnh_dnode = DN_SLOT_UNINIT; 1222 continue; 1223 } 1224 1225 zrl_add(&dnh->dnh_zrlock); 1226 dn = dnh->dnh_dnode; 1227 /* 1228 * If there are holds on this dnode, then there should 1229 * be holds on the dnode's containing dbuf as well; thus 1230 * it wouldn't be eligible for eviction and this function 1231 * would not have been called. 1232 */ 1233 ASSERT(zfs_refcount_is_zero(&dn->dn_holds)); 1234 ASSERT(zfs_refcount_is_zero(&dn->dn_tx_holds)); 1235 1236 dnode_destroy(dn); /* implicit zrl_remove() for first slot */ 1237 zrl_destroy(&dnh->dnh_zrlock); 1238 dnh->dnh_dnode = DN_SLOT_UNINIT; 1239 } 1240 kmem_free(dnc, sizeof (dnode_children_t) + 1241 dnc->dnc_count * sizeof (dnode_handle_t)); 1242 } 1243 1244 /* 1245 * When the DNODE_MUST_BE_FREE flag is set, the "slots" parameter is used 1246 * to ensure the hole at the specified object offset is large enough to 1247 * hold the dnode being created. The slots parameter is also used to ensure 1248 * a dnode does not span multiple dnode blocks. In both of these cases, if 1249 * a failure occurs, ENOSPC is returned. Keep in mind, these failure cases 1250 * are only possible when using DNODE_MUST_BE_FREE. 1251 * 1252 * If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. 1253 * dnode_hold_impl() will check if the requested dnode is already consumed 1254 * as an extra dnode slot by an large dnode, in which case it returns 1255 * ENOENT. 1256 * 1257 * If the DNODE_DRY_RUN flag is set, we don't actually hold the dnode, just 1258 * return whether the hold would succeed or not. tag and dnp should set to 1259 * NULL in this case. 1260 * 1261 * errors: 1262 * EINVAL - invalid object number or flags. 1263 * ENOSPC - hole too small to fulfill "slots" request (DNODE_MUST_BE_FREE) 1264 * EEXIST - Refers to an allocated dnode (DNODE_MUST_BE_FREE) 1265 * - Refers to a freeing dnode (DNODE_MUST_BE_FREE) 1266 * - Refers to an interior dnode slot (DNODE_MUST_BE_ALLOCATED) 1267 * ENOENT - The requested dnode is not allocated (DNODE_MUST_BE_ALLOCATED) 1268 * - The requested dnode is being freed (DNODE_MUST_BE_ALLOCATED) 1269 * EIO - i/o error error when reading the meta dnode dbuf. 1270 * succeeds even for free dnodes. 1271 */ 1272 int 1273 dnode_hold_impl(objset_t *os, uint64_t object, int flag, int slots, 1274 void *tag, dnode_t **dnp) 1275 { 1276 int epb, idx, err; 1277 int drop_struct_lock = FALSE; 1278 int type; 1279 uint64_t blk; 1280 dnode_t *mdn, *dn; 1281 dmu_buf_impl_t *db; 1282 dnode_children_t *dnc; 1283 dnode_phys_t *dn_block; 1284 dnode_handle_t *dnh; 1285 1286 ASSERT(!(flag & DNODE_MUST_BE_ALLOCATED) || (slots == 0)); 1287 ASSERT(!(flag & DNODE_MUST_BE_FREE) || (slots > 0)); 1288 IMPLY(flag & DNODE_DRY_RUN, (tag == NULL) && (dnp == NULL)); 1289 1290 /* 1291 * If you are holding the spa config lock as writer, you shouldn't 1292 * be asking the DMU to do *anything* unless it's the root pool 1293 * which may require us to read from the root filesystem while 1294 * holding some (not all) of the locks as writer. 1295 */ 1296 ASSERT(spa_config_held(os->os_spa, SCL_ALL, RW_WRITER) == 0 || 1297 (spa_is_root(os->os_spa) && 1298 spa_config_held(os->os_spa, SCL_STATE, RW_WRITER))); 1299 1300 ASSERT((flag & DNODE_MUST_BE_ALLOCATED) || (flag & DNODE_MUST_BE_FREE)); 1301 1302 if (object == DMU_USERUSED_OBJECT || object == DMU_GROUPUSED_OBJECT || 1303 object == DMU_PROJECTUSED_OBJECT) { 1304 if (object == DMU_USERUSED_OBJECT) 1305 dn = DMU_USERUSED_DNODE(os); 1306 else if (object == DMU_GROUPUSED_OBJECT) 1307 dn = DMU_GROUPUSED_DNODE(os); 1308 else 1309 dn = DMU_PROJECTUSED_DNODE(os); 1310 if (dn == NULL) 1311 return (SET_ERROR(ENOENT)); 1312 type = dn->dn_type; 1313 if ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) 1314 return (SET_ERROR(ENOENT)); 1315 if ((flag & DNODE_MUST_BE_FREE) && type != DMU_OT_NONE) 1316 return (SET_ERROR(EEXIST)); 1317 DNODE_VERIFY(dn); 1318 /* Don't actually hold if dry run, just return 0 */ 1319 if (!(flag & DNODE_DRY_RUN)) { 1320 (void) zfs_refcount_add(&dn->dn_holds, tag); 1321 *dnp = dn; 1322 } 1323 return (0); 1324 } 1325 1326 if (object == 0 || object >= DN_MAX_OBJECT) 1327 return (SET_ERROR(EINVAL)); 1328 1329 mdn = DMU_META_DNODE(os); 1330 ASSERT(mdn->dn_object == DMU_META_DNODE_OBJECT); 1331 1332 DNODE_VERIFY(mdn); 1333 1334 if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { 1335 rw_enter(&mdn->dn_struct_rwlock, RW_READER); 1336 drop_struct_lock = TRUE; 1337 } 1338 1339 blk = dbuf_whichblock(mdn, 0, object * sizeof (dnode_phys_t)); 1340 db = dbuf_hold(mdn, blk, FTAG); 1341 if (drop_struct_lock) 1342 rw_exit(&mdn->dn_struct_rwlock); 1343 if (db == NULL) { 1344 DNODE_STAT_BUMP(dnode_hold_dbuf_hold); 1345 return (SET_ERROR(EIO)); 1346 } 1347 /* 1348 * We do not need to decrypt to read the dnode so it doesn't matter 1349 * if we get the encrypted or decrypted version. 1350 */ 1351 err = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_NO_DECRYPT); 1352 if (err) { 1353 DNODE_STAT_BUMP(dnode_hold_dbuf_read); 1354 dbuf_rele(db, FTAG); 1355 return (err); 1356 } 1357 1358 ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT); 1359 epb = db->db.db_size >> DNODE_SHIFT; 1360 1361 idx = object & (epb - 1); 1362 dn_block = (dnode_phys_t *)db->db.db_data; 1363 1364 ASSERT(DB_DNODE(db)->dn_type == DMU_OT_DNODE); 1365 dnc = dmu_buf_get_user(&db->db); 1366 dnh = NULL; 1367 if (dnc == NULL) { 1368 dnode_children_t *winner; 1369 int skip = 0; 1370 1371 dnc = kmem_zalloc(sizeof (dnode_children_t) + 1372 epb * sizeof (dnode_handle_t), KM_SLEEP); 1373 dnc->dnc_count = epb; 1374 dnh = &dnc->dnc_children[0]; 1375 1376 /* Initialize dnode slot status from dnode_phys_t */ 1377 for (int i = 0; i < epb; i++) { 1378 zrl_init(&dnh[i].dnh_zrlock); 1379 1380 if (skip) { 1381 skip--; 1382 continue; 1383 } 1384 1385 if (dn_block[i].dn_type != DMU_OT_NONE) { 1386 int interior = dn_block[i].dn_extra_slots; 1387 1388 dnode_set_slots(dnc, i, 1, DN_SLOT_ALLOCATED); 1389 dnode_set_slots(dnc, i + 1, interior, 1390 DN_SLOT_INTERIOR); 1391 skip = interior; 1392 } else { 1393 dnh[i].dnh_dnode = DN_SLOT_FREE; 1394 skip = 0; 1395 } 1396 } 1397 1398 dmu_buf_init_user(&dnc->dnc_dbu, NULL, 1399 dnode_buf_evict_async, NULL); 1400 winner = dmu_buf_set_user(&db->db, &dnc->dnc_dbu); 1401 if (winner != NULL) { 1402 1403 for (int i = 0; i < epb; i++) 1404 zrl_destroy(&dnh[i].dnh_zrlock); 1405 1406 kmem_free(dnc, sizeof (dnode_children_t) + 1407 epb * sizeof (dnode_handle_t)); 1408 dnc = winner; 1409 } 1410 } 1411 1412 ASSERT(dnc->dnc_count == epb); 1413 dn = DN_SLOT_UNINIT; 1414 1415 if (flag & DNODE_MUST_BE_ALLOCATED) { 1416 slots = 1; 1417 1418 while (dn == DN_SLOT_UNINIT) { 1419 dnode_slots_hold(dnc, idx, slots); 1420 dnh = &dnc->dnc_children[idx]; 1421 1422 if (DN_SLOT_IS_PTR(dnh->dnh_dnode)) { 1423 dn = dnh->dnh_dnode; 1424 break; 1425 } else if (dnh->dnh_dnode == DN_SLOT_INTERIOR) { 1426 DNODE_STAT_BUMP(dnode_hold_alloc_interior); 1427 dnode_slots_rele(dnc, idx, slots); 1428 dbuf_rele(db, FTAG); 1429 return (SET_ERROR(EEXIST)); 1430 } else if (dnh->dnh_dnode != DN_SLOT_ALLOCATED) { 1431 DNODE_STAT_BUMP(dnode_hold_alloc_misses); 1432 dnode_slots_rele(dnc, idx, slots); 1433 dbuf_rele(db, FTAG); 1434 return (SET_ERROR(ENOENT)); 1435 } 1436 1437 dnode_slots_rele(dnc, idx, slots); 1438 if (!dnode_slots_tryenter(dnc, idx, slots)) { 1439 DNODE_STAT_BUMP(dnode_hold_alloc_lock_retry); 1440 continue; 1441 } 1442 1443 /* 1444 * Someone else won the race and called dnode_create() 1445 * after we checked DN_SLOT_IS_PTR() above but before 1446 * we acquired the lock. 1447 */ 1448 if (DN_SLOT_IS_PTR(dnh->dnh_dnode)) { 1449 DNODE_STAT_BUMP(dnode_hold_alloc_lock_misses); 1450 dn = dnh->dnh_dnode; 1451 } else { 1452 dn = dnode_create(os, dn_block + idx, db, 1453 object, dnh); 1454 } 1455 } 1456 1457 mutex_enter(&dn->dn_mtx); 1458 if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg != 0) { 1459 DNODE_STAT_BUMP(dnode_hold_alloc_type_none); 1460 mutex_exit(&dn->dn_mtx); 1461 dnode_slots_rele(dnc, idx, slots); 1462 dbuf_rele(db, FTAG); 1463 return (SET_ERROR(ENOENT)); 1464 } 1465 1466 /* Don't actually hold if dry run, just return 0 */ 1467 if (flag & DNODE_DRY_RUN) { 1468 mutex_exit(&dn->dn_mtx); 1469 dnode_slots_rele(dnc, idx, slots); 1470 dbuf_rele(db, FTAG); 1471 return (0); 1472 } 1473 1474 DNODE_STAT_BUMP(dnode_hold_alloc_hits); 1475 } else if (flag & DNODE_MUST_BE_FREE) { 1476 1477 if (idx + slots - 1 >= DNODES_PER_BLOCK) { 1478 DNODE_STAT_BUMP(dnode_hold_free_overflow); 1479 dbuf_rele(db, FTAG); 1480 return (SET_ERROR(ENOSPC)); 1481 } 1482 1483 while (dn == DN_SLOT_UNINIT) { 1484 dnode_slots_hold(dnc, idx, slots); 1485 1486 if (!dnode_check_slots_free(dnc, idx, slots)) { 1487 DNODE_STAT_BUMP(dnode_hold_free_misses); 1488 dnode_slots_rele(dnc, idx, slots); 1489 dbuf_rele(db, FTAG); 1490 return (SET_ERROR(ENOSPC)); 1491 } 1492 1493 dnode_slots_rele(dnc, idx, slots); 1494 if (!dnode_slots_tryenter(dnc, idx, slots)) { 1495 DNODE_STAT_BUMP(dnode_hold_free_lock_retry); 1496 continue; 1497 } 1498 1499 if (!dnode_check_slots_free(dnc, idx, slots)) { 1500 DNODE_STAT_BUMP(dnode_hold_free_lock_misses); 1501 dnode_slots_rele(dnc, idx, slots); 1502 dbuf_rele(db, FTAG); 1503 return (SET_ERROR(ENOSPC)); 1504 } 1505 1506 /* 1507 * Allocated but otherwise free dnodes which would 1508 * be in the interior of a multi-slot dnodes need 1509 * to be freed. Single slot dnodes can be safely 1510 * re-purposed as a performance optimization. 1511 */ 1512 if (slots > 1) 1513 dnode_reclaim_slots(dnc, idx + 1, slots - 1); 1514 1515 dnh = &dnc->dnc_children[idx]; 1516 if (DN_SLOT_IS_PTR(dnh->dnh_dnode)) { 1517 dn = dnh->dnh_dnode; 1518 } else { 1519 dn = dnode_create(os, dn_block + idx, db, 1520 object, dnh); 1521 } 1522 } 1523 1524 mutex_enter(&dn->dn_mtx); 1525 if (!zfs_refcount_is_zero(&dn->dn_holds) || dn->dn_free_txg) { 1526 DNODE_STAT_BUMP(dnode_hold_free_refcount); 1527 mutex_exit(&dn->dn_mtx); 1528 dnode_slots_rele(dnc, idx, slots); 1529 dbuf_rele(db, FTAG); 1530 return (SET_ERROR(EEXIST)); 1531 } 1532 1533 /* Don't actually hold if dry run, just return 0 */ 1534 if (flag & DNODE_DRY_RUN) { 1535 mutex_exit(&dn->dn_mtx); 1536 dnode_slots_rele(dnc, idx, slots); 1537 dbuf_rele(db, FTAG); 1538 return (0); 1539 } 1540 1541 dnode_set_slots(dnc, idx + 1, slots - 1, DN_SLOT_INTERIOR); 1542 DNODE_STAT_BUMP(dnode_hold_free_hits); 1543 } else { 1544 dbuf_rele(db, FTAG); 1545 return (SET_ERROR(EINVAL)); 1546 } 1547 1548 ASSERT0(dn->dn_free_txg); 1549 1550 if (zfs_refcount_add(&dn->dn_holds, tag) == 1) 1551 dbuf_add_ref(db, dnh); 1552 1553 mutex_exit(&dn->dn_mtx); 1554 1555 /* Now we can rely on the hold to prevent the dnode from moving. */ 1556 dnode_slots_rele(dnc, idx, slots); 1557 1558 DNODE_VERIFY(dn); 1559 ASSERT3P(dn->dn_dbuf, ==, db); 1560 ASSERT3U(dn->dn_object, ==, object); 1561 dbuf_rele(db, FTAG); 1562 1563 *dnp = dn; 1564 return (0); 1565 } 1566 1567 /* 1568 * Return held dnode if the object is allocated, NULL if not. 1569 */ 1570 int 1571 dnode_hold(objset_t *os, uint64_t object, void *tag, dnode_t **dnp) 1572 { 1573 return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0, tag, 1574 dnp)); 1575 } 1576 1577 /* 1578 * Can only add a reference if there is already at least one 1579 * reference on the dnode. Returns FALSE if unable to add a 1580 * new reference. 1581 */ 1582 boolean_t 1583 dnode_add_ref(dnode_t *dn, void *tag) 1584 { 1585 mutex_enter(&dn->dn_mtx); 1586 if (zfs_refcount_is_zero(&dn->dn_holds)) { 1587 mutex_exit(&dn->dn_mtx); 1588 return (FALSE); 1589 } 1590 VERIFY(1 < zfs_refcount_add(&dn->dn_holds, tag)); 1591 mutex_exit(&dn->dn_mtx); 1592 return (TRUE); 1593 } 1594 1595 void 1596 dnode_rele(dnode_t *dn, void *tag) 1597 { 1598 mutex_enter(&dn->dn_mtx); 1599 dnode_rele_and_unlock(dn, tag, B_FALSE); 1600 } 1601 1602 void 1603 dnode_rele_and_unlock(dnode_t *dn, void *tag, boolean_t evicting) 1604 { 1605 uint64_t refs; 1606 /* Get while the hold prevents the dnode from moving. */ 1607 dmu_buf_impl_t *db = dn->dn_dbuf; 1608 dnode_handle_t *dnh = dn->dn_handle; 1609 1610 refs = zfs_refcount_remove(&dn->dn_holds, tag); 1611 if (refs == 0) 1612 cv_broadcast(&dn->dn_nodnholds); 1613 mutex_exit(&dn->dn_mtx); 1614 /* dnode could get destroyed at this point, so don't use it anymore */ 1615 1616 /* 1617 * It's unsafe to release the last hold on a dnode by dnode_rele() or 1618 * indirectly by dbuf_rele() while relying on the dnode handle to 1619 * prevent the dnode from moving, since releasing the last hold could 1620 * result in the dnode's parent dbuf evicting its dnode handles. For 1621 * that reason anyone calling dnode_rele() or dbuf_rele() without some 1622 * other direct or indirect hold on the dnode must first drop the dnode 1623 * handle. 1624 */ 1625 ASSERT(refs > 0 || dnh->dnh_zrlock.zr_owner != curthread); 1626 1627 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ 1628 if (refs == 0 && db != NULL) { 1629 /* 1630 * Another thread could add a hold to the dnode handle in 1631 * dnode_hold_impl() while holding the parent dbuf. Since the 1632 * hold on the parent dbuf prevents the handle from being 1633 * destroyed, the hold on the handle is OK. We can't yet assert 1634 * that the handle has zero references, but that will be 1635 * asserted anyway when the handle gets destroyed. 1636 */ 1637 mutex_enter(&db->db_mtx); 1638 dbuf_rele_and_unlock(db, dnh, evicting); 1639 } 1640 } 1641 1642 /* 1643 * Test whether we can create a dnode at the specified location. 1644 */ 1645 int 1646 dnode_try_claim(objset_t *os, uint64_t object, int slots) 1647 { 1648 return (dnode_hold_impl(os, object, DNODE_MUST_BE_FREE | DNODE_DRY_RUN, 1649 slots, NULL, NULL)); 1650 } 1651 1652 void 1653 dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) 1654 { 1655 objset_t *os = dn->dn_objset; 1656 uint64_t txg = tx->tx_txg; 1657 1658 if (DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 1659 dsl_dataset_dirty(os->os_dsl_dataset, tx); 1660 return; 1661 } 1662 1663 DNODE_VERIFY(dn); 1664 1665 #ifdef ZFS_DEBUG 1666 mutex_enter(&dn->dn_mtx); 1667 ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); 1668 ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); 1669 mutex_exit(&dn->dn_mtx); 1670 #endif 1671 1672 /* 1673 * Determine old uid/gid when necessary 1674 */ 1675 dmu_objset_userquota_get_ids(dn, B_TRUE, tx); 1676 1677 multilist_t *dirtylist = os->os_dirty_dnodes[txg & TXG_MASK]; 1678 multilist_sublist_t *mls = multilist_sublist_lock_obj(dirtylist, dn); 1679 1680 /* 1681 * If we are already marked dirty, we're done. 1682 */ 1683 if (multilist_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) { 1684 multilist_sublist_unlock(mls); 1685 return; 1686 } 1687 1688 ASSERT(!zfs_refcount_is_zero(&dn->dn_holds) || 1689 !avl_is_empty(&dn->dn_dbufs)); 1690 ASSERT(dn->dn_datablksz != 0); 1691 ASSERT0(dn->dn_next_bonuslen[txg&TXG_MASK]); 1692 ASSERT0(dn->dn_next_blksz[txg&TXG_MASK]); 1693 ASSERT0(dn->dn_next_bonustype[txg&TXG_MASK]); 1694 1695 dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", 1696 dn->dn_object, txg); 1697 1698 multilist_sublist_insert_head(mls, dn); 1699 1700 multilist_sublist_unlock(mls); 1701 1702 /* 1703 * The dnode maintains a hold on its containing dbuf as 1704 * long as there are holds on it. Each instantiated child 1705 * dbuf maintains a hold on the dnode. When the last child 1706 * drops its hold, the dnode will drop its hold on the 1707 * containing dbuf. We add a "dirty hold" here so that the 1708 * dnode will hang around after we finish processing its 1709 * children. 1710 */ 1711 VERIFY(dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg)); 1712 1713 (void) dbuf_dirty(dn->dn_dbuf, tx); 1714 1715 dsl_dataset_dirty(os->os_dsl_dataset, tx); 1716 } 1717 1718 void 1719 dnode_free(dnode_t *dn, dmu_tx_t *tx) 1720 { 1721 mutex_enter(&dn->dn_mtx); 1722 if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { 1723 mutex_exit(&dn->dn_mtx); 1724 return; 1725 } 1726 dn->dn_free_txg = tx->tx_txg; 1727 mutex_exit(&dn->dn_mtx); 1728 1729 dnode_setdirty(dn, tx); 1730 } 1731 1732 /* 1733 * Try to change the block size for the indicated dnode. This can only 1734 * succeed if there are no blocks allocated or dirty beyond first block 1735 */ 1736 int 1737 dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) 1738 { 1739 dmu_buf_impl_t *db; 1740 int err; 1741 1742 ASSERT3U(size, <=, spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); 1743 if (size == 0) 1744 size = SPA_MINBLOCKSIZE; 1745 else 1746 size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); 1747 1748 if (ibs == dn->dn_indblkshift) 1749 ibs = 0; 1750 1751 if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && ibs == 0) 1752 return (0); 1753 1754 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 1755 1756 /* Check for any allocated blocks beyond the first */ 1757 if (dn->dn_maxblkid != 0) 1758 goto fail; 1759 1760 mutex_enter(&dn->dn_dbufs_mtx); 1761 for (db = avl_first(&dn->dn_dbufs); db != NULL; 1762 db = AVL_NEXT(&dn->dn_dbufs, db)) { 1763 if (db->db_blkid != 0 && db->db_blkid != DMU_BONUS_BLKID && 1764 db->db_blkid != DMU_SPILL_BLKID) { 1765 mutex_exit(&dn->dn_dbufs_mtx); 1766 goto fail; 1767 } 1768 } 1769 mutex_exit(&dn->dn_dbufs_mtx); 1770 1771 if (ibs && dn->dn_nlevels != 1) 1772 goto fail; 1773 1774 /* resize the old block */ 1775 err = dbuf_hold_impl(dn, 0, 0, TRUE, FALSE, FTAG, &db); 1776 if (err == 0) { 1777 dbuf_new_size(db, size, tx); 1778 } else if (err != ENOENT) { 1779 goto fail; 1780 } 1781 1782 dnode_setdblksz(dn, size); 1783 dnode_setdirty(dn, tx); 1784 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size; 1785 if (ibs) { 1786 dn->dn_indblkshift = ibs; 1787 dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; 1788 } 1789 /* rele after we have fixed the blocksize in the dnode */ 1790 if (db) 1791 dbuf_rele(db, FTAG); 1792 1793 rw_exit(&dn->dn_struct_rwlock); 1794 return (0); 1795 1796 fail: 1797 rw_exit(&dn->dn_struct_rwlock); 1798 return (SET_ERROR(ENOTSUP)); 1799 } 1800 1801 static void 1802 dnode_set_nlevels_impl(dnode_t *dn, int new_nlevels, dmu_tx_t *tx) 1803 { 1804 uint64_t txgoff = tx->tx_txg & TXG_MASK; 1805 int old_nlevels = dn->dn_nlevels; 1806 dmu_buf_impl_t *db; 1807 list_t *list; 1808 dbuf_dirty_record_t *new, *dr, *dr_next; 1809 1810 ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); 1811 1812 dn->dn_nlevels = new_nlevels; 1813 1814 ASSERT3U(new_nlevels, >, dn->dn_next_nlevels[txgoff]); 1815 dn->dn_next_nlevels[txgoff] = new_nlevels; 1816 1817 /* dirty the left indirects */ 1818 db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); 1819 ASSERT(db != NULL); 1820 new = dbuf_dirty(db, tx); 1821 dbuf_rele(db, FTAG); 1822 1823 /* transfer the dirty records to the new indirect */ 1824 mutex_enter(&dn->dn_mtx); 1825 mutex_enter(&new->dt.di.dr_mtx); 1826 list = &dn->dn_dirty_records[txgoff]; 1827 for (dr = list_head(list); dr; dr = dr_next) { 1828 dr_next = list_next(&dn->dn_dirty_records[txgoff], dr); 1829 if (dr->dr_dbuf->db_level != new_nlevels-1 && 1830 dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID && 1831 dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) { 1832 ASSERT(dr->dr_dbuf->db_level == old_nlevels-1); 1833 list_remove(&dn->dn_dirty_records[txgoff], dr); 1834 list_insert_tail(&new->dt.di.dr_children, dr); 1835 dr->dr_parent = new; 1836 } 1837 } 1838 mutex_exit(&new->dt.di.dr_mtx); 1839 mutex_exit(&dn->dn_mtx); 1840 } 1841 1842 int 1843 dnode_set_nlevels(dnode_t *dn, int nlevels, dmu_tx_t *tx) 1844 { 1845 int ret = 0; 1846 1847 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 1848 1849 if (dn->dn_nlevels == nlevels) { 1850 ret = 0; 1851 goto out; 1852 } else if (nlevels < dn->dn_nlevels) { 1853 ret = SET_ERROR(EINVAL); 1854 goto out; 1855 } 1856 1857 dnode_set_nlevels_impl(dn, nlevels, tx); 1858 1859 out: 1860 rw_exit(&dn->dn_struct_rwlock); 1861 return (ret); 1862 } 1863 1864 /* read-holding callers must not rely on the lock being continuously held */ 1865 void 1866 dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t have_read, 1867 boolean_t force) 1868 { 1869 int epbs, new_nlevels; 1870 uint64_t sz; 1871 1872 ASSERT(blkid != DMU_BONUS_BLKID); 1873 1874 ASSERT(have_read ? 1875 RW_READ_HELD(&dn->dn_struct_rwlock) : 1876 RW_WRITE_HELD(&dn->dn_struct_rwlock)); 1877 1878 /* 1879 * if we have a read-lock, check to see if we need to do any work 1880 * before upgrading to a write-lock. 1881 */ 1882 if (have_read) { 1883 if (blkid <= dn->dn_maxblkid) 1884 return; 1885 1886 if (!rw_tryupgrade(&dn->dn_struct_rwlock)) { 1887 rw_exit(&dn->dn_struct_rwlock); 1888 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 1889 } 1890 } 1891 1892 /* 1893 * Raw sends (indicated by the force flag) require that we take the 1894 * given blkid even if the value is lower than the current value. 1895 */ 1896 if (!force && blkid <= dn->dn_maxblkid) 1897 goto out; 1898 1899 /* 1900 * We use the (otherwise unused) top bit of dn_next_maxblkid[txgoff] 1901 * to indicate that this field is set. This allows us to set the 1902 * maxblkid to 0 on an existing object in dnode_sync(). 1903 */ 1904 dn->dn_maxblkid = blkid; 1905 dn->dn_next_maxblkid[tx->tx_txg & TXG_MASK] = 1906 blkid | DMU_NEXT_MAXBLKID_SET; 1907 1908 /* 1909 * Compute the number of levels necessary to support the new maxblkid. 1910 * Raw sends will ensure nlevels is set correctly for us. 1911 */ 1912 new_nlevels = 1; 1913 epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 1914 for (sz = dn->dn_nblkptr; 1915 sz <= blkid && sz >= dn->dn_nblkptr; sz <<= epbs) 1916 new_nlevels++; 1917 1918 if (!force) { 1919 if (new_nlevels > dn->dn_nlevels) 1920 dnode_set_nlevels_impl(dn, new_nlevels, tx); 1921 } else { 1922 ASSERT3U(dn->dn_nlevels, >=, new_nlevels); 1923 } 1924 1925 out: 1926 if (have_read) 1927 rw_downgrade(&dn->dn_struct_rwlock); 1928 } 1929 1930 static void 1931 dnode_dirty_l1(dnode_t *dn, uint64_t l1blkid, dmu_tx_t *tx) 1932 { 1933 dmu_buf_impl_t *db = dbuf_hold_level(dn, 1, l1blkid, FTAG); 1934 if (db != NULL) { 1935 dmu_buf_will_dirty(&db->db, tx); 1936 dbuf_rele(db, FTAG); 1937 } 1938 } 1939 1940 /* 1941 * Dirty all the in-core level-1 dbufs in the range specified by start_blkid 1942 * and end_blkid. 1943 */ 1944 static void 1945 dnode_dirty_l1range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid, 1946 dmu_tx_t *tx) 1947 { 1948 dmu_buf_impl_t db_search; 1949 dmu_buf_impl_t *db; 1950 avl_index_t where; 1951 1952 mutex_enter(&dn->dn_dbufs_mtx); 1953 1954 db_search.db_level = 1; 1955 db_search.db_blkid = start_blkid + 1; 1956 db_search.db_state = DB_SEARCH; 1957 for (;;) { 1958 1959 db = avl_find(&dn->dn_dbufs, &db_search, &where); 1960 if (db == NULL) 1961 db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER); 1962 1963 if (db == NULL || db->db_level != 1 || 1964 db->db_blkid >= end_blkid) { 1965 break; 1966 } 1967 1968 /* 1969 * Setup the next blkid we want to search for. 1970 */ 1971 db_search.db_blkid = db->db_blkid + 1; 1972 ASSERT3U(db->db_blkid, >=, start_blkid); 1973 1974 /* 1975 * If the dbuf transitions to DB_EVICTING while we're trying 1976 * to dirty it, then we will be unable to discover it in 1977 * the dbuf hash table. This will result in a call to 1978 * dbuf_create() which needs to acquire the dn_dbufs_mtx 1979 * lock. To avoid a deadlock, we drop the lock before 1980 * dirtying the level-1 dbuf. 1981 */ 1982 mutex_exit(&dn->dn_dbufs_mtx); 1983 dnode_dirty_l1(dn, db->db_blkid, tx); 1984 mutex_enter(&dn->dn_dbufs_mtx); 1985 } 1986 1987 #ifdef ZFS_DEBUG 1988 /* 1989 * Walk all the in-core level-1 dbufs and verify they have been dirtied. 1990 */ 1991 db_search.db_level = 1; 1992 db_search.db_blkid = start_blkid + 1; 1993 db_search.db_state = DB_SEARCH; 1994 db = avl_find(&dn->dn_dbufs, &db_search, &where); 1995 if (db == NULL) 1996 db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER); 1997 for (; db != NULL; db = AVL_NEXT(&dn->dn_dbufs, db)) { 1998 if (db->db_level != 1 || db->db_blkid >= end_blkid) 1999 break; 2000 ASSERT(db->db_dirtycnt > 0); 2001 } 2002 #endif 2003 mutex_exit(&dn->dn_dbufs_mtx); 2004 } 2005 2006 void 2007 dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) 2008 { 2009 dmu_buf_impl_t *db; 2010 uint64_t blkoff, blkid, nblks; 2011 int blksz, blkshift, head, tail; 2012 int trunc = FALSE; 2013 int epbs; 2014 2015 blksz = dn->dn_datablksz; 2016 blkshift = dn->dn_datablkshift; 2017 epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 2018 2019 if (len == DMU_OBJECT_END) { 2020 len = UINT64_MAX - off; 2021 trunc = TRUE; 2022 } 2023 2024 /* 2025 * First, block align the region to free: 2026 */ 2027 if (ISP2(blksz)) { 2028 head = P2NPHASE(off, blksz); 2029 blkoff = P2PHASE(off, blksz); 2030 if ((off >> blkshift) > dn->dn_maxblkid) 2031 return; 2032 } else { 2033 ASSERT(dn->dn_maxblkid == 0); 2034 if (off == 0 && len >= blksz) { 2035 /* 2036 * Freeing the whole block; fast-track this request. 2037 */ 2038 blkid = 0; 2039 nblks = 1; 2040 if (dn->dn_nlevels > 1) { 2041 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 2042 dnode_dirty_l1(dn, 0, tx); 2043 rw_exit(&dn->dn_struct_rwlock); 2044 } 2045 goto done; 2046 } else if (off >= blksz) { 2047 /* Freeing past end-of-data */ 2048 return; 2049 } else { 2050 /* Freeing part of the block. */ 2051 head = blksz - off; 2052 ASSERT3U(head, >, 0); 2053 } 2054 blkoff = off; 2055 } 2056 /* zero out any partial block data at the start of the range */ 2057 if (head) { 2058 int res; 2059 ASSERT3U(blkoff + head, ==, blksz); 2060 if (len < head) 2061 head = len; 2062 rw_enter(&dn->dn_struct_rwlock, RW_READER); 2063 res = dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, 0, off), 2064 TRUE, FALSE, FTAG, &db); 2065 rw_exit(&dn->dn_struct_rwlock); 2066 if (res == 0) { 2067 caddr_t data; 2068 boolean_t dirty; 2069 2070 db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_READER, 2071 FTAG); 2072 /* don't dirty if it isn't on disk and isn't dirty */ 2073 dirty = db->db_last_dirty || 2074 (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr)); 2075 dmu_buf_unlock_parent(db, dblt, FTAG); 2076 if (dirty) { 2077 dmu_buf_will_dirty(&db->db, tx); 2078 data = db->db.db_data; 2079 bzero(data + blkoff, head); 2080 } 2081 dbuf_rele(db, FTAG); 2082 } 2083 off += head; 2084 len -= head; 2085 } 2086 2087 /* If the range was less than one block, we're done */ 2088 if (len == 0) 2089 return; 2090 2091 /* If the remaining range is past end of file, we're done */ 2092 if ((off >> blkshift) > dn->dn_maxblkid) 2093 return; 2094 2095 ASSERT(ISP2(blksz)); 2096 if (trunc) 2097 tail = 0; 2098 else 2099 tail = P2PHASE(len, blksz); 2100 2101 ASSERT0(P2PHASE(off, blksz)); 2102 /* zero out any partial block data at the end of the range */ 2103 if (tail) { 2104 int res; 2105 if (len < tail) 2106 tail = len; 2107 rw_enter(&dn->dn_struct_rwlock, RW_READER); 2108 res = dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, 0, off+len), 2109 TRUE, FALSE, FTAG, &db); 2110 rw_exit(&dn->dn_struct_rwlock); 2111 if (res == 0) { 2112 boolean_t dirty; 2113 /* don't dirty if not on disk and not dirty */ 2114 db_lock_type_t type = dmu_buf_lock_parent(db, RW_READER, 2115 FTAG); 2116 dirty = db->db_last_dirty || 2117 (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr)); 2118 dmu_buf_unlock_parent(db, type, FTAG); 2119 if (dirty) { 2120 dmu_buf_will_dirty(&db->db, tx); 2121 bzero(db->db.db_data, tail); 2122 } 2123 dbuf_rele(db, FTAG); 2124 } 2125 len -= tail; 2126 } 2127 2128 /* If the range did not include a full block, we are done */ 2129 if (len == 0) 2130 return; 2131 2132 ASSERT(IS_P2ALIGNED(off, blksz)); 2133 ASSERT(trunc || IS_P2ALIGNED(len, blksz)); 2134 blkid = off >> blkshift; 2135 nblks = len >> blkshift; 2136 if (trunc) 2137 nblks += 1; 2138 2139 /* 2140 * Dirty all the indirect blocks in this range. Note that only 2141 * the first and last indirect blocks can actually be written 2142 * (if they were partially freed) -- they must be dirtied, even if 2143 * they do not exist on disk yet. The interior blocks will 2144 * be freed by free_children(), so they will not actually be written. 2145 * Even though these interior blocks will not be written, we 2146 * dirty them for two reasons: 2147 * 2148 * - It ensures that the indirect blocks remain in memory until 2149 * syncing context. (They have already been prefetched by 2150 * dmu_tx_hold_free(), so we don't have to worry about reading 2151 * them serially here.) 2152 * 2153 * - The dirty space accounting will put pressure on the txg sync 2154 * mechanism to begin syncing, and to delay transactions if there 2155 * is a large amount of freeing. Even though these indirect 2156 * blocks will not be written, we could need to write the same 2157 * amount of space if we copy the freed BPs into deadlists. 2158 */ 2159 if (dn->dn_nlevels > 1) { 2160 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 2161 uint64_t first, last; 2162 2163 first = blkid >> epbs; 2164 dnode_dirty_l1(dn, first, tx); 2165 if (trunc) 2166 last = dn->dn_maxblkid >> epbs; 2167 else 2168 last = (blkid + nblks - 1) >> epbs; 2169 if (last != first) 2170 dnode_dirty_l1(dn, last, tx); 2171 2172 dnode_dirty_l1range(dn, first, last, tx); 2173 2174 int shift = dn->dn_datablkshift + dn->dn_indblkshift - 2175 SPA_BLKPTRSHIFT; 2176 for (uint64_t i = first + 1; i < last; i++) { 2177 /* 2178 * Set i to the blockid of the next non-hole 2179 * level-1 indirect block at or after i. Note 2180 * that dnode_next_offset() operates in terms of 2181 * level-0-equivalent bytes. 2182 */ 2183 uint64_t ibyte = i << shift; 2184 int err = dnode_next_offset(dn, DNODE_FIND_HAVELOCK, 2185 &ibyte, 2, 1, 0); 2186 i = ibyte >> shift; 2187 if (i >= last) 2188 break; 2189 2190 /* 2191 * Normally we should not see an error, either 2192 * from dnode_next_offset() or dbuf_hold_level() 2193 * (except for ESRCH from dnode_next_offset). 2194 * If there is an i/o error, then when we read 2195 * this block in syncing context, it will use 2196 * ZIO_FLAG_MUSTSUCCEED, and thus hang/panic according 2197 * to the "failmode" property. dnode_next_offset() 2198 * doesn't have a flag to indicate MUSTSUCCEED. 2199 */ 2200 if (err != 0) 2201 break; 2202 2203 dnode_dirty_l1(dn, i, tx); 2204 } 2205 rw_exit(&dn->dn_struct_rwlock); 2206 } 2207 2208 done: 2209 /* 2210 * Add this range to the dnode range list. 2211 * We will finish up this free operation in the syncing phase. 2212 */ 2213 mutex_enter(&dn->dn_mtx); 2214 int txgoff = tx->tx_txg & TXG_MASK; 2215 if (dn->dn_free_ranges[txgoff] == NULL) { 2216 dn->dn_free_ranges[txgoff] = range_tree_create(NULL, 2217 RANGE_SEG64, NULL, 0, 0); 2218 } 2219 range_tree_clear(dn->dn_free_ranges[txgoff], blkid, nblks); 2220 range_tree_add(dn->dn_free_ranges[txgoff], blkid, nblks); 2221 dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", 2222 blkid, nblks, tx->tx_txg); 2223 mutex_exit(&dn->dn_mtx); 2224 2225 dbuf_free_range(dn, blkid, blkid + nblks - 1, tx); 2226 dnode_setdirty(dn, tx); 2227 } 2228 2229 static boolean_t 2230 dnode_spill_freed(dnode_t *dn) 2231 { 2232 int i; 2233 2234 mutex_enter(&dn->dn_mtx); 2235 for (i = 0; i < TXG_SIZE; i++) { 2236 if (dn->dn_rm_spillblk[i] == DN_KILL_SPILLBLK) 2237 break; 2238 } 2239 mutex_exit(&dn->dn_mtx); 2240 return (i < TXG_SIZE); 2241 } 2242 2243 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */ 2244 uint64_t 2245 dnode_block_freed(dnode_t *dn, uint64_t blkid) 2246 { 2247 void *dp = spa_get_dsl(dn->dn_objset->os_spa); 2248 int i; 2249 2250 if (blkid == DMU_BONUS_BLKID) 2251 return (FALSE); 2252 2253 /* 2254 * If we're in the process of opening the pool, dp will not be 2255 * set yet, but there shouldn't be anything dirty. 2256 */ 2257 if (dp == NULL) 2258 return (FALSE); 2259 2260 if (dn->dn_free_txg) 2261 return (TRUE); 2262 2263 if (blkid == DMU_SPILL_BLKID) 2264 return (dnode_spill_freed(dn)); 2265 2266 mutex_enter(&dn->dn_mtx); 2267 for (i = 0; i < TXG_SIZE; i++) { 2268 if (dn->dn_free_ranges[i] != NULL && 2269 range_tree_contains(dn->dn_free_ranges[i], blkid, 1)) 2270 break; 2271 } 2272 mutex_exit(&dn->dn_mtx); 2273 return (i < TXG_SIZE); 2274 } 2275 2276 /* call from syncing context when we actually write/free space for this dnode */ 2277 void 2278 dnode_diduse_space(dnode_t *dn, int64_t delta) 2279 { 2280 uint64_t space; 2281 dprintf_dnode(dn, "dn=%p dnp=%p used=%llu delta=%lld\n", 2282 dn, dn->dn_phys, 2283 (u_longlong_t)dn->dn_phys->dn_used, 2284 (longlong_t)delta); 2285 2286 mutex_enter(&dn->dn_mtx); 2287 space = DN_USED_BYTES(dn->dn_phys); 2288 if (delta > 0) { 2289 ASSERT3U(space + delta, >=, space); /* no overflow */ 2290 } else { 2291 ASSERT3U(space, >=, -delta); /* no underflow */ 2292 } 2293 space += delta; 2294 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_DNODE_BYTES) { 2295 ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0); 2296 ASSERT0(P2PHASE(space, 1<<DEV_BSHIFT)); 2297 dn->dn_phys->dn_used = space >> DEV_BSHIFT; 2298 } else { 2299 dn->dn_phys->dn_used = space; 2300 dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES; 2301 } 2302 mutex_exit(&dn->dn_mtx); 2303 } 2304 2305 /* 2306 * Scans a block at the indicated "level" looking for a hole or data, 2307 * depending on 'flags'. 2308 * 2309 * If level > 0, then we are scanning an indirect block looking at its 2310 * pointers. If level == 0, then we are looking at a block of dnodes. 2311 * 2312 * If we don't find what we are looking for in the block, we return ESRCH. 2313 * Otherwise, return with *offset pointing to the beginning (if searching 2314 * forwards) or end (if searching backwards) of the range covered by the 2315 * block pointer we matched on (or dnode). 2316 * 2317 * The basic search algorithm used below by dnode_next_offset() is to 2318 * use this function to search up the block tree (widen the search) until 2319 * we find something (i.e., we don't return ESRCH) and then search back 2320 * down the tree (narrow the search) until we reach our original search 2321 * level. 2322 */ 2323 static int 2324 dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset, 2325 int lvl, uint64_t blkfill, uint64_t txg) 2326 { 2327 dmu_buf_impl_t *db = NULL; 2328 void *data = NULL; 2329 uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 2330 uint64_t epb = 1ULL << epbs; 2331 uint64_t minfill, maxfill; 2332 boolean_t hole; 2333 int i, inc, error, span; 2334 2335 dprintf("probing object %llu offset %llx level %d of %u\n", 2336 dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); 2337 2338 ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); 2339 2340 hole = ((flags & DNODE_FIND_HOLE) != 0); 2341 inc = (flags & DNODE_FIND_BACKWARDS) ? -1 : 1; 2342 ASSERT(txg == 0 || !hole); 2343 2344 if (lvl == dn->dn_phys->dn_nlevels) { 2345 error = 0; 2346 epb = dn->dn_phys->dn_nblkptr; 2347 data = dn->dn_phys->dn_blkptr; 2348 } else { 2349 uint64_t blkid = dbuf_whichblock(dn, lvl, *offset); 2350 error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FALSE, FTAG, &db); 2351 if (error) { 2352 if (error != ENOENT) 2353 return (error); 2354 if (hole) 2355 return (0); 2356 /* 2357 * This can only happen when we are searching up 2358 * the block tree for data. We don't really need to 2359 * adjust the offset, as we will just end up looking 2360 * at the pointer to this block in its parent, and its 2361 * going to be unallocated, so we will skip over it. 2362 */ 2363 return (SET_ERROR(ESRCH)); 2364 } 2365 error = dbuf_read(db, NULL, 2366 DB_RF_CANFAIL | DB_RF_HAVESTRUCT | DB_RF_NO_DECRYPT); 2367 if (error) { 2368 dbuf_rele(db, FTAG); 2369 return (error); 2370 } 2371 data = db->db.db_data; 2372 rw_enter(&db->db_rwlock, RW_READER); 2373 } 2374 2375 if (db != NULL && txg != 0 && (db->db_blkptr == NULL || 2376 db->db_blkptr->blk_birth <= txg || 2377 BP_IS_HOLE(db->db_blkptr))) { 2378 /* 2379 * This can only happen when we are searching up the tree 2380 * and these conditions mean that we need to keep climbing. 2381 */ 2382 error = SET_ERROR(ESRCH); 2383 } else if (lvl == 0) { 2384 dnode_phys_t *dnp = data; 2385 2386 ASSERT(dn->dn_type == DMU_OT_DNODE); 2387 ASSERT(!(flags & DNODE_FIND_BACKWARDS)); 2388 2389 for (i = (*offset >> DNODE_SHIFT) & (blkfill - 1); 2390 i < blkfill; i += dnp[i].dn_extra_slots + 1) { 2391 if ((dnp[i].dn_type == DMU_OT_NONE) == hole) 2392 break; 2393 } 2394 2395 if (i == blkfill) 2396 error = SET_ERROR(ESRCH); 2397 2398 *offset = (*offset & ~(DNODE_BLOCK_SIZE - 1)) + 2399 (i << DNODE_SHIFT); 2400 } else { 2401 blkptr_t *bp = data; 2402 uint64_t start = *offset; 2403 span = (lvl - 1) * epbs + dn->dn_datablkshift; 2404 minfill = 0; 2405 maxfill = blkfill << ((lvl - 1) * epbs); 2406 2407 if (hole) 2408 maxfill--; 2409 else 2410 minfill++; 2411 2412 *offset = *offset >> span; 2413 for (i = BF64_GET(*offset, 0, epbs); 2414 i >= 0 && i < epb; i += inc) { 2415 if (BP_GET_FILL(&bp[i]) >= minfill && 2416 BP_GET_FILL(&bp[i]) <= maxfill && 2417 (hole || bp[i].blk_birth > txg)) 2418 break; 2419 if (inc > 0 || *offset > 0) 2420 *offset += inc; 2421 } 2422 *offset = *offset << span; 2423 if (inc < 0) { 2424 /* traversing backwards; position offset at the end */ 2425 ASSERT3U(*offset, <=, start); 2426 *offset = MIN(*offset + (1ULL << span) - 1, start); 2427 } else if (*offset < start) { 2428 *offset = start; 2429 } 2430 if (i < 0 || i >= epb) 2431 error = SET_ERROR(ESRCH); 2432 } 2433 2434 if (db != NULL) { 2435 rw_exit(&db->db_rwlock); 2436 dbuf_rele(db, FTAG); 2437 } 2438 2439 return (error); 2440 } 2441 2442 /* 2443 * Find the next hole, data, or sparse region at or after *offset. 2444 * The value 'blkfill' tells us how many items we expect to find 2445 * in an L0 data block; this value is 1 for normal objects, 2446 * DNODES_PER_BLOCK for the meta dnode, and some fraction of 2447 * DNODES_PER_BLOCK when searching for sparse regions thereof. 2448 * 2449 * Examples: 2450 * 2451 * dnode_next_offset(dn, flags, offset, 1, 1, 0); 2452 * Finds the next/previous hole/data in a file. 2453 * Used in dmu_offset_next(). 2454 * 2455 * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg); 2456 * Finds the next free/allocated dnode an objset's meta-dnode. 2457 * Only finds objects that have new contents since txg (ie. 2458 * bonus buffer changes and content removal are ignored). 2459 * Used in dmu_object_next(). 2460 * 2461 * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0); 2462 * Finds the next L2 meta-dnode bp that's at most 1/4 full. 2463 * Used in dmu_object_alloc(). 2464 */ 2465 int 2466 dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset, 2467 int minlvl, uint64_t blkfill, uint64_t txg) 2468 { 2469 uint64_t initial_offset = *offset; 2470 int lvl, maxlvl; 2471 int error = 0; 2472 2473 if (!(flags & DNODE_FIND_HAVELOCK)) 2474 rw_enter(&dn->dn_struct_rwlock, RW_READER); 2475 2476 if (dn->dn_phys->dn_nlevels == 0) { 2477 error = SET_ERROR(ESRCH); 2478 goto out; 2479 } 2480 2481 if (dn->dn_datablkshift == 0) { 2482 if (*offset < dn->dn_datablksz) { 2483 if (flags & DNODE_FIND_HOLE) 2484 *offset = dn->dn_datablksz; 2485 } else { 2486 error = SET_ERROR(ESRCH); 2487 } 2488 goto out; 2489 } 2490 2491 maxlvl = dn->dn_phys->dn_nlevels; 2492 2493 for (lvl = minlvl; lvl <= maxlvl; lvl++) { 2494 error = dnode_next_offset_level(dn, 2495 flags, offset, lvl, blkfill, txg); 2496 if (error != ESRCH) 2497 break; 2498 } 2499 2500 while (error == 0 && --lvl >= minlvl) { 2501 error = dnode_next_offset_level(dn, 2502 flags, offset, lvl, blkfill, txg); 2503 } 2504 2505 /* 2506 * There's always a "virtual hole" at the end of the object, even 2507 * if all BP's which physically exist are non-holes. 2508 */ 2509 if ((flags & DNODE_FIND_HOLE) && error == ESRCH && txg == 0 && 2510 minlvl == 1 && blkfill == 1 && !(flags & DNODE_FIND_BACKWARDS)) { 2511 error = 0; 2512 } 2513 2514 if (error == 0 && (flags & DNODE_FIND_BACKWARDS ? 2515 initial_offset < *offset : initial_offset > *offset)) 2516 error = SET_ERROR(ESRCH); 2517 out: 2518 if (!(flags & DNODE_FIND_HAVELOCK)) 2519 rw_exit(&dn->dn_struct_rwlock); 2520 2521 return (error); 2522 } 2523