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 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved. 25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 26 */ 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_tx.h> 33 #include <sys/dmu_objset.h> 34 #include <sys/dsl_dataset.h> 35 #include <sys/spa.h> 36 #include <sys/range_tree.h> 37 #include <sys/zfeature.h> 38 39 static void 40 dnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) 41 { 42 dmu_buf_impl_t *db; 43 int txgoff = tx->tx_txg & TXG_MASK; 44 int nblkptr = dn->dn_phys->dn_nblkptr; 45 int old_toplvl = dn->dn_phys->dn_nlevels - 1; 46 int new_level = dn->dn_next_nlevels[txgoff]; 47 int i; 48 49 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 50 51 /* this dnode can't be paged out because it's dirty */ 52 ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 53 ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); 54 ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); 55 56 db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); 57 ASSERT(db != NULL); 58 59 dn->dn_phys->dn_nlevels = new_level; 60 dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, 61 dn->dn_object, dn->dn_phys->dn_nlevels); 62 63 /* check for existing blkptrs in the dnode */ 64 for (i = 0; i < nblkptr; i++) 65 if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[i])) 66 break; 67 if (i != nblkptr) { 68 /* transfer dnode's block pointers to new indirect block */ 69 (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); 70 ASSERT(db->db.db_data); 71 ASSERT(arc_released(db->db_buf)); 72 ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); 73 bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, 74 sizeof (blkptr_t) * nblkptr); 75 arc_buf_freeze(db->db_buf); 76 } 77 78 /* set dbuf's parent pointers to new indirect buf */ 79 for (i = 0; i < nblkptr; i++) { 80 dmu_buf_impl_t *child = 81 dbuf_find(dn->dn_objset, dn->dn_object, old_toplvl, i); 82 83 if (child == NULL) 84 continue; 85 #ifdef DEBUG 86 DB_DNODE_ENTER(child); 87 ASSERT3P(DB_DNODE(child), ==, dn); 88 DB_DNODE_EXIT(child); 89 #endif /* DEBUG */ 90 if (child->db_parent && child->db_parent != dn->dn_dbuf) { 91 ASSERT(child->db_parent->db_level == db->db_level); 92 ASSERT(child->db_blkptr != 93 &dn->dn_phys->dn_blkptr[child->db_blkid]); 94 mutex_exit(&child->db_mtx); 95 continue; 96 } 97 ASSERT(child->db_parent == NULL || 98 child->db_parent == dn->dn_dbuf); 99 100 child->db_parent = db; 101 dbuf_add_ref(db, child); 102 if (db->db.db_data) 103 child->db_blkptr = (blkptr_t *)db->db.db_data + i; 104 else 105 child->db_blkptr = NULL; 106 dprintf_dbuf_bp(child, child->db_blkptr, 107 "changed db_blkptr to new indirect %s", ""); 108 109 mutex_exit(&child->db_mtx); 110 } 111 112 bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); 113 114 dbuf_rele(db, FTAG); 115 116 rw_exit(&dn->dn_struct_rwlock); 117 } 118 119 static void 120 free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) 121 { 122 dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; 123 uint64_t bytesfreed = 0; 124 125 dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); 126 127 for (int i = 0; i < num; i++, bp++) { 128 if (BP_IS_HOLE(bp)) 129 continue; 130 131 bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE); 132 ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); 133 134 /* 135 * Save some useful information on the holes being 136 * punched, including logical size, type, and indirection 137 * level. Retaining birth time enables detection of when 138 * holes are punched for reducing the number of free 139 * records transmitted during a zfs send. 140 */ 141 142 uint64_t lsize = BP_GET_LSIZE(bp); 143 dmu_object_type_t type = BP_GET_TYPE(bp); 144 uint64_t lvl = BP_GET_LEVEL(bp); 145 146 bzero(bp, sizeof (blkptr_t)); 147 148 if (spa_feature_is_active(dn->dn_objset->os_spa, 149 SPA_FEATURE_HOLE_BIRTH)) { 150 BP_SET_LSIZE(bp, lsize); 151 BP_SET_TYPE(bp, type); 152 BP_SET_LEVEL(bp, lvl); 153 BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0); 154 } 155 } 156 dnode_diduse_space(dn, -bytesfreed); 157 } 158 159 #ifdef ZFS_DEBUG 160 static void 161 free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) 162 { 163 int off, num; 164 int i, err, epbs; 165 uint64_t txg = tx->tx_txg; 166 dnode_t *dn; 167 168 DB_DNODE_ENTER(db); 169 dn = DB_DNODE(db); 170 epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 171 off = start - (db->db_blkid * 1<<epbs); 172 num = end - start + 1; 173 174 ASSERT3U(off, >=, 0); 175 ASSERT3U(num, >=, 0); 176 ASSERT3U(db->db_level, >, 0); 177 ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); 178 ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); 179 ASSERT(db->db_blkptr != NULL); 180 181 for (i = off; i < off+num; i++) { 182 uint64_t *buf; 183 dmu_buf_impl_t *child; 184 dbuf_dirty_record_t *dr; 185 int j; 186 187 ASSERT(db->db_level == 1); 188 189 rw_enter(&dn->dn_struct_rwlock, RW_READER); 190 err = dbuf_hold_impl(dn, db->db_level-1, 191 (db->db_blkid << epbs) + i, TRUE, FTAG, &child); 192 rw_exit(&dn->dn_struct_rwlock); 193 if (err == ENOENT) 194 continue; 195 ASSERT(err == 0); 196 ASSERT(child->db_level == 0); 197 dr = child->db_last_dirty; 198 while (dr && dr->dr_txg > txg) 199 dr = dr->dr_next; 200 ASSERT(dr == NULL || dr->dr_txg == txg); 201 202 /* data_old better be zeroed */ 203 if (dr) { 204 buf = dr->dt.dl.dr_data->b_data; 205 for (j = 0; j < child->db.db_size >> 3; j++) { 206 if (buf[j] != 0) { 207 panic("freed data not zero: " 208 "child=%p i=%d off=%d num=%d\n", 209 (void *)child, i, off, num); 210 } 211 } 212 } 213 214 /* 215 * db_data better be zeroed unless it's dirty in a 216 * future txg. 217 */ 218 mutex_enter(&child->db_mtx); 219 buf = child->db.db_data; 220 if (buf != NULL && child->db_state != DB_FILL && 221 child->db_last_dirty == NULL) { 222 for (j = 0; j < child->db.db_size >> 3; j++) { 223 if (buf[j] != 0) { 224 panic("freed data not zero: " 225 "child=%p i=%d off=%d num=%d\n", 226 (void *)child, i, off, num); 227 } 228 } 229 } 230 mutex_exit(&child->db_mtx); 231 232 dbuf_rele(child, FTAG); 233 } 234 DB_DNODE_EXIT(db); 235 } 236 #endif 237 238 static void 239 free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, 240 dmu_tx_t *tx) 241 { 242 dnode_t *dn; 243 blkptr_t *bp; 244 dmu_buf_impl_t *subdb; 245 uint64_t start, end, dbstart, dbend, i; 246 int epbs, shift; 247 248 /* 249 * There is a small possibility that this block will not be cached: 250 * 1 - if level > 1 and there are no children with level <= 1 251 * 2 - if this block was evicted since we read it from 252 * dmu_tx_hold_free(). 253 */ 254 if (db->db_state != DB_CACHED) 255 (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); 256 257 dbuf_release_bp(db); 258 bp = db->db.db_data; 259 260 DB_DNODE_ENTER(db); 261 dn = DB_DNODE(db); 262 epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 263 shift = (db->db_level - 1) * epbs; 264 dbstart = db->db_blkid << epbs; 265 start = blkid >> shift; 266 if (dbstart < start) { 267 bp += start - dbstart; 268 } else { 269 start = dbstart; 270 } 271 dbend = ((db->db_blkid + 1) << epbs) - 1; 272 end = (blkid + nblks - 1) >> shift; 273 if (dbend <= end) 274 end = dbend; 275 276 ASSERT3U(start, <=, end); 277 278 if (db->db_level == 1) { 279 FREE_VERIFY(db, start, end, tx); 280 free_blocks(dn, bp, end-start+1, tx); 281 } else { 282 for (i = start; i <= end; i++, bp++) { 283 if (BP_IS_HOLE(bp)) 284 continue; 285 rw_enter(&dn->dn_struct_rwlock, RW_READER); 286 VERIFY0(dbuf_hold_impl(dn, db->db_level - 1, 287 i, B_TRUE, FTAG, &subdb)); 288 rw_exit(&dn->dn_struct_rwlock); 289 ASSERT3P(bp, ==, subdb->db_blkptr); 290 291 free_children(subdb, blkid, nblks, tx); 292 dbuf_rele(subdb, FTAG); 293 } 294 } 295 296 /* If this whole block is free, free ourself too. */ 297 for (i = 0, bp = db->db.db_data; i < 1 << epbs; i++, bp++) { 298 if (!BP_IS_HOLE(bp)) 299 break; 300 } 301 if (i == 1 << epbs) { 302 /* didn't find any non-holes */ 303 bzero(db->db.db_data, db->db.db_size); 304 free_blocks(dn, db->db_blkptr, 1, tx); 305 } else { 306 /* 307 * Partial block free; must be marked dirty so that it 308 * will be written out. 309 */ 310 ASSERT(db->db_dirtycnt > 0); 311 } 312 313 DB_DNODE_EXIT(db); 314 arc_buf_freeze(db->db_buf); 315 } 316 317 /* 318 * Traverse the indicated range of the provided file 319 * and "free" all the blocks contained there. 320 */ 321 static void 322 dnode_sync_free_range_impl(dnode_t *dn, uint64_t blkid, uint64_t nblks, 323 dmu_tx_t *tx) 324 { 325 blkptr_t *bp = dn->dn_phys->dn_blkptr; 326 int dnlevel = dn->dn_phys->dn_nlevels; 327 boolean_t trunc = B_FALSE; 328 329 if (blkid > dn->dn_phys->dn_maxblkid) 330 return; 331 332 ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); 333 if (blkid + nblks > dn->dn_phys->dn_maxblkid) { 334 nblks = dn->dn_phys->dn_maxblkid - blkid + 1; 335 trunc = B_TRUE; 336 } 337 338 /* There are no indirect blocks in the object */ 339 if (dnlevel == 1) { 340 if (blkid >= dn->dn_phys->dn_nblkptr) { 341 /* this range was never made persistent */ 342 return; 343 } 344 ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); 345 free_blocks(dn, bp + blkid, nblks, tx); 346 } else { 347 int shift = (dnlevel - 1) * 348 (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); 349 int start = blkid >> shift; 350 int end = (blkid + nblks - 1) >> shift; 351 dmu_buf_impl_t *db; 352 353 ASSERT(start < dn->dn_phys->dn_nblkptr); 354 bp += start; 355 for (int i = start; i <= end; i++, bp++) { 356 if (BP_IS_HOLE(bp)) 357 continue; 358 rw_enter(&dn->dn_struct_rwlock, RW_READER); 359 VERIFY0(dbuf_hold_impl(dn, dnlevel - 1, i, 360 TRUE, FTAG, &db)); 361 rw_exit(&dn->dn_struct_rwlock); 362 363 free_children(db, blkid, nblks, tx); 364 dbuf_rele(db, FTAG); 365 } 366 } 367 368 if (trunc) { 369 dn->dn_phys->dn_maxblkid = blkid == 0 ? 0 : blkid - 1; 370 371 uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * 372 (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); 373 ASSERT(off < dn->dn_phys->dn_maxblkid || 374 dn->dn_phys->dn_maxblkid == 0 || 375 dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); 376 } 377 } 378 379 typedef struct dnode_sync_free_range_arg { 380 dnode_t *dsfra_dnode; 381 dmu_tx_t *dsfra_tx; 382 } dnode_sync_free_range_arg_t; 383 384 static void 385 dnode_sync_free_range(void *arg, uint64_t blkid, uint64_t nblks) 386 { 387 dnode_sync_free_range_arg_t *dsfra = arg; 388 dnode_t *dn = dsfra->dsfra_dnode; 389 390 mutex_exit(&dn->dn_mtx); 391 dnode_sync_free_range_impl(dn, blkid, nblks, dsfra->dsfra_tx); 392 mutex_enter(&dn->dn_mtx); 393 } 394 395 /* 396 * Try to kick all the dnode's dbufs out of the cache... 397 */ 398 void 399 dnode_evict_dbufs(dnode_t *dn) 400 { 401 dmu_buf_impl_t db_marker; 402 dmu_buf_impl_t *db, *db_next; 403 404 mutex_enter(&dn->dn_dbufs_mtx); 405 for (db = avl_first(&dn->dn_dbufs); db != NULL; db = db_next) { 406 407 #ifdef DEBUG 408 DB_DNODE_ENTER(db); 409 ASSERT3P(DB_DNODE(db), ==, dn); 410 DB_DNODE_EXIT(db); 411 #endif /* DEBUG */ 412 413 mutex_enter(&db->db_mtx); 414 if (db->db_state != DB_EVICTING && 415 refcount_is_zero(&db->db_holds)) { 416 db_marker.db_level = db->db_level; 417 db_marker.db_blkid = db->db_blkid; 418 db_marker.db_state = DB_SEARCH; 419 avl_insert_here(&dn->dn_dbufs, &db_marker, db, 420 AVL_BEFORE); 421 422 dbuf_clear(db); 423 424 db_next = AVL_NEXT(&dn->dn_dbufs, &db_marker); 425 avl_remove(&dn->dn_dbufs, &db_marker); 426 } else { 427 mutex_exit(&db->db_mtx); 428 db_next = AVL_NEXT(&dn->dn_dbufs, db); 429 } 430 } 431 mutex_exit(&dn->dn_dbufs_mtx); 432 433 dnode_evict_bonus(dn); 434 } 435 436 void 437 dnode_evict_bonus(dnode_t *dn) 438 { 439 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 440 if (dn->dn_bonus && refcount_is_zero(&dn->dn_bonus->db_holds)) { 441 mutex_enter(&dn->dn_bonus->db_mtx); 442 dbuf_evict(dn->dn_bonus); 443 dn->dn_bonus = NULL; 444 } 445 rw_exit(&dn->dn_struct_rwlock); 446 } 447 448 static void 449 dnode_undirty_dbufs(list_t *list) 450 { 451 dbuf_dirty_record_t *dr; 452 453 while (dr = list_head(list)) { 454 dmu_buf_impl_t *db = dr->dr_dbuf; 455 uint64_t txg = dr->dr_txg; 456 457 if (db->db_level != 0) 458 dnode_undirty_dbufs(&dr->dt.di.dr_children); 459 460 mutex_enter(&db->db_mtx); 461 /* XXX - use dbuf_undirty()? */ 462 list_remove(list, dr); 463 ASSERT(db->db_last_dirty == dr); 464 db->db_last_dirty = NULL; 465 db->db_dirtycnt -= 1; 466 if (db->db_level == 0) { 467 ASSERT(db->db_blkid == DMU_BONUS_BLKID || 468 dr->dt.dl.dr_data == db->db_buf); 469 dbuf_unoverride(dr); 470 } else { 471 mutex_destroy(&dr->dt.di.dr_mtx); 472 list_destroy(&dr->dt.di.dr_children); 473 } 474 kmem_free(dr, sizeof (dbuf_dirty_record_t)); 475 dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); 476 } 477 } 478 479 static void 480 dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) 481 { 482 int txgoff = tx->tx_txg & TXG_MASK; 483 484 ASSERT(dmu_tx_is_syncing(tx)); 485 486 /* 487 * Our contents should have been freed in dnode_sync() by the 488 * free range record inserted by the caller of dnode_free(). 489 */ 490 ASSERT0(DN_USED_BYTES(dn->dn_phys)); 491 ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); 492 493 dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); 494 dnode_evict_dbufs(dn); 495 ASSERT(avl_is_empty(&dn->dn_dbufs)); 496 497 /* 498 * XXX - It would be nice to assert this, but we may still 499 * have residual holds from async evictions from the arc... 500 * 501 * zfs_obj_to_path() also depends on this being 502 * commented out. 503 * 504 * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); 505 */ 506 507 /* Undirty next bits */ 508 dn->dn_next_nlevels[txgoff] = 0; 509 dn->dn_next_indblkshift[txgoff] = 0; 510 dn->dn_next_blksz[txgoff] = 0; 511 512 /* ASSERT(blkptrs are zero); */ 513 ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 514 ASSERT(dn->dn_type != DMU_OT_NONE); 515 516 ASSERT(dn->dn_free_txg > 0); 517 if (dn->dn_allocated_txg != dn->dn_free_txg) 518 dmu_buf_will_dirty(&dn->dn_dbuf->db, tx); 519 bzero(dn->dn_phys, sizeof (dnode_phys_t)); 520 521 mutex_enter(&dn->dn_mtx); 522 dn->dn_type = DMU_OT_NONE; 523 dn->dn_maxblkid = 0; 524 dn->dn_allocated_txg = 0; 525 dn->dn_free_txg = 0; 526 dn->dn_have_spill = B_FALSE; 527 mutex_exit(&dn->dn_mtx); 528 529 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 530 531 dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 532 /* 533 * Now that we've released our hold, the dnode may 534 * be evicted, so we musn't access it. 535 */ 536 } 537 538 /* 539 * Write out the dnode's dirty buffers. 540 */ 541 void 542 dnode_sync(dnode_t *dn, dmu_tx_t *tx) 543 { 544 dnode_phys_t *dnp = dn->dn_phys; 545 int txgoff = tx->tx_txg & TXG_MASK; 546 list_t *list = &dn->dn_dirty_records[txgoff]; 547 static const dnode_phys_t zerodn = { 0 }; 548 boolean_t kill_spill = B_FALSE; 549 550 ASSERT(dmu_tx_is_syncing(tx)); 551 ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); 552 ASSERT(dnp->dn_type != DMU_OT_NONE || 553 bcmp(dnp, &zerodn, DNODE_SIZE) == 0); 554 DNODE_VERIFY(dn); 555 556 ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); 557 558 if (dmu_objset_userused_enabled(dn->dn_objset) && 559 !DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 560 mutex_enter(&dn->dn_mtx); 561 dn->dn_oldused = DN_USED_BYTES(dn->dn_phys); 562 dn->dn_oldflags = dn->dn_phys->dn_flags; 563 dn->dn_phys->dn_flags |= DNODE_FLAG_USERUSED_ACCOUNTED; 564 mutex_exit(&dn->dn_mtx); 565 dmu_objset_userquota_get_ids(dn, B_FALSE, tx); 566 } else { 567 /* Once we account for it, we should always account for it. */ 568 ASSERT(!(dn->dn_phys->dn_flags & 569 DNODE_FLAG_USERUSED_ACCOUNTED)); 570 } 571 572 mutex_enter(&dn->dn_mtx); 573 if (dn->dn_allocated_txg == tx->tx_txg) { 574 /* The dnode is newly allocated or reallocated */ 575 if (dnp->dn_type == DMU_OT_NONE) { 576 /* this is a first alloc, not a realloc */ 577 dnp->dn_nlevels = 1; 578 dnp->dn_nblkptr = dn->dn_nblkptr; 579 } 580 581 dnp->dn_type = dn->dn_type; 582 dnp->dn_bonustype = dn->dn_bonustype; 583 dnp->dn_bonuslen = dn->dn_bonuslen; 584 } 585 ASSERT(dnp->dn_nlevels > 1 || 586 BP_IS_HOLE(&dnp->dn_blkptr[0]) || 587 BP_IS_EMBEDDED(&dnp->dn_blkptr[0]) || 588 BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 589 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 590 ASSERT(dnp->dn_nlevels < 2 || 591 BP_IS_HOLE(&dnp->dn_blkptr[0]) || 592 BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 1 << dnp->dn_indblkshift); 593 594 if (dn->dn_next_type[txgoff] != 0) { 595 dnp->dn_type = dn->dn_type; 596 dn->dn_next_type[txgoff] = 0; 597 } 598 599 if (dn->dn_next_blksz[txgoff] != 0) { 600 ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], 601 SPA_MINBLOCKSIZE) == 0); 602 ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || 603 dn->dn_maxblkid == 0 || list_head(list) != NULL || 604 dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == 605 dnp->dn_datablkszsec || 606 range_tree_space(dn->dn_free_ranges[txgoff]) != 0); 607 dnp->dn_datablkszsec = 608 dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; 609 dn->dn_next_blksz[txgoff] = 0; 610 } 611 612 if (dn->dn_next_bonuslen[txgoff] != 0) { 613 if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) 614 dnp->dn_bonuslen = 0; 615 else 616 dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; 617 ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN); 618 dn->dn_next_bonuslen[txgoff] = 0; 619 } 620 621 if (dn->dn_next_bonustype[txgoff] != 0) { 622 ASSERT(DMU_OT_IS_VALID(dn->dn_next_bonustype[txgoff])); 623 dnp->dn_bonustype = dn->dn_next_bonustype[txgoff]; 624 dn->dn_next_bonustype[txgoff] = 0; 625 } 626 627 boolean_t freeing_dnode = dn->dn_free_txg > 0 && 628 dn->dn_free_txg <= tx->tx_txg; 629 630 /* 631 * Remove the spill block if we have been explicitly asked to 632 * remove it, or if the object is being removed. 633 */ 634 if (dn->dn_rm_spillblk[txgoff] || freeing_dnode) { 635 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) 636 kill_spill = B_TRUE; 637 dn->dn_rm_spillblk[txgoff] = 0; 638 } 639 640 if (dn->dn_next_indblkshift[txgoff] != 0) { 641 ASSERT(dnp->dn_nlevels == 1); 642 dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; 643 dn->dn_next_indblkshift[txgoff] = 0; 644 } 645 646 /* 647 * Just take the live (open-context) values for checksum and compress. 648 * Strictly speaking it's a future leak, but nothing bad happens if we 649 * start using the new checksum or compress algorithm a little early. 650 */ 651 dnp->dn_checksum = dn->dn_checksum; 652 dnp->dn_compress = dn->dn_compress; 653 654 mutex_exit(&dn->dn_mtx); 655 656 if (kill_spill) { 657 free_blocks(dn, &dn->dn_phys->dn_spill, 1, tx); 658 mutex_enter(&dn->dn_mtx); 659 dnp->dn_flags &= ~DNODE_FLAG_SPILL_BLKPTR; 660 mutex_exit(&dn->dn_mtx); 661 } 662 663 /* process all the "freed" ranges in the file */ 664 if (dn->dn_free_ranges[txgoff] != NULL) { 665 dnode_sync_free_range_arg_t dsfra; 666 dsfra.dsfra_dnode = dn; 667 dsfra.dsfra_tx = tx; 668 mutex_enter(&dn->dn_mtx); 669 range_tree_vacate(dn->dn_free_ranges[txgoff], 670 dnode_sync_free_range, &dsfra); 671 range_tree_destroy(dn->dn_free_ranges[txgoff]); 672 dn->dn_free_ranges[txgoff] = NULL; 673 mutex_exit(&dn->dn_mtx); 674 } 675 676 if (freeing_dnode) { 677 dnode_sync_free(dn, tx); 678 return; 679 } 680 681 if (dn->dn_next_nlevels[txgoff]) { 682 dnode_increase_indirection(dn, tx); 683 dn->dn_next_nlevels[txgoff] = 0; 684 } 685 686 if (dn->dn_next_nblkptr[txgoff]) { 687 /* this should only happen on a realloc */ 688 ASSERT(dn->dn_allocated_txg == tx->tx_txg); 689 if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { 690 /* zero the new blkptrs we are gaining */ 691 bzero(dnp->dn_blkptr + dnp->dn_nblkptr, 692 sizeof (blkptr_t) * 693 (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); 694 #ifdef ZFS_DEBUG 695 } else { 696 int i; 697 ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); 698 /* the blkptrs we are losing better be unallocated */ 699 for (i = dn->dn_next_nblkptr[txgoff]; 700 i < dnp->dn_nblkptr; i++) 701 ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); 702 #endif 703 } 704 mutex_enter(&dn->dn_mtx); 705 dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; 706 dn->dn_next_nblkptr[txgoff] = 0; 707 mutex_exit(&dn->dn_mtx); 708 } 709 710 dbuf_sync_list(list, dn->dn_phys->dn_nlevels - 1, tx); 711 712 if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 713 ASSERT3P(list_head(list), ==, NULL); 714 dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 715 } 716 717 /* 718 * Although we have dropped our reference to the dnode, it 719 * can't be evicted until its written, and we haven't yet 720 * initiated the IO for the dnode's dbuf. 721 */ 722 } 723