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, 2017 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 /* transfer dnode's block pointers to new indirect block */ 64 (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); 65 ASSERT(db->db.db_data); 66 ASSERT(arc_released(db->db_buf)); 67 ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); 68 bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, 69 sizeof (blkptr_t) * nblkptr); 70 arc_buf_freeze(db->db_buf); 71 72 /* set dbuf's parent pointers to new indirect buf */ 73 for (i = 0; i < nblkptr; i++) { 74 dmu_buf_impl_t *child = 75 dbuf_find(dn->dn_objset, dn->dn_object, old_toplvl, i); 76 77 if (child == NULL) 78 continue; 79 #ifdef DEBUG 80 DB_DNODE_ENTER(child); 81 ASSERT3P(DB_DNODE(child), ==, dn); 82 DB_DNODE_EXIT(child); 83 #endif /* DEBUG */ 84 if (child->db_parent && child->db_parent != dn->dn_dbuf) { 85 ASSERT(child->db_parent->db_level == db->db_level); 86 ASSERT(child->db_blkptr != 87 &dn->dn_phys->dn_blkptr[child->db_blkid]); 88 mutex_exit(&child->db_mtx); 89 continue; 90 } 91 ASSERT(child->db_parent == NULL || 92 child->db_parent == dn->dn_dbuf); 93 94 child->db_parent = db; 95 dbuf_add_ref(db, child); 96 if (db->db.db_data) 97 child->db_blkptr = (blkptr_t *)db->db.db_data + i; 98 else 99 child->db_blkptr = NULL; 100 dprintf_dbuf_bp(child, child->db_blkptr, 101 "changed db_blkptr to new indirect %s", ""); 102 103 mutex_exit(&child->db_mtx); 104 } 105 106 bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); 107 108 dbuf_rele(db, FTAG); 109 110 rw_exit(&dn->dn_struct_rwlock); 111 } 112 113 static void 114 free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) 115 { 116 dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; 117 uint64_t bytesfreed = 0; 118 119 dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); 120 121 for (int i = 0; i < num; i++, bp++) { 122 if (BP_IS_HOLE(bp)) 123 continue; 124 125 bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE); 126 ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); 127 128 /* 129 * Save some useful information on the holes being 130 * punched, including logical size, type, and indirection 131 * level. Retaining birth time enables detection of when 132 * holes are punched for reducing the number of free 133 * records transmitted during a zfs send. 134 */ 135 136 uint64_t lsize = BP_GET_LSIZE(bp); 137 dmu_object_type_t type = BP_GET_TYPE(bp); 138 uint64_t lvl = BP_GET_LEVEL(bp); 139 140 bzero(bp, sizeof (blkptr_t)); 141 142 if (spa_feature_is_active(dn->dn_objset->os_spa, 143 SPA_FEATURE_HOLE_BIRTH)) { 144 BP_SET_LSIZE(bp, lsize); 145 BP_SET_TYPE(bp, type); 146 BP_SET_LEVEL(bp, lvl); 147 BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0); 148 } 149 } 150 dnode_diduse_space(dn, -bytesfreed); 151 } 152 153 #ifdef ZFS_DEBUG 154 static void 155 free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) 156 { 157 int off, num; 158 int i, err, epbs; 159 uint64_t txg = tx->tx_txg; 160 dnode_t *dn; 161 162 DB_DNODE_ENTER(db); 163 dn = DB_DNODE(db); 164 epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 165 off = start - (db->db_blkid * 1<<epbs); 166 num = end - start + 1; 167 168 ASSERT3U(off, >=, 0); 169 ASSERT3U(num, >=, 0); 170 ASSERT3U(db->db_level, >, 0); 171 ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); 172 ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); 173 ASSERT(db->db_blkptr != NULL); 174 175 for (i = off; i < off+num; i++) { 176 uint64_t *buf; 177 dmu_buf_impl_t *child; 178 dbuf_dirty_record_t *dr; 179 int j; 180 181 ASSERT(db->db_level == 1); 182 183 rw_enter(&dn->dn_struct_rwlock, RW_READER); 184 err = dbuf_hold_impl(dn, db->db_level-1, 185 (db->db_blkid << epbs) + i, TRUE, FALSE, FTAG, &child); 186 rw_exit(&dn->dn_struct_rwlock); 187 if (err == ENOENT) 188 continue; 189 ASSERT(err == 0); 190 ASSERT(child->db_level == 0); 191 dr = child->db_last_dirty; 192 while (dr && dr->dr_txg > txg) 193 dr = dr->dr_next; 194 ASSERT(dr == NULL || dr->dr_txg == txg); 195 196 /* data_old better be zeroed */ 197 if (dr) { 198 buf = dr->dt.dl.dr_data->b_data; 199 for (j = 0; j < child->db.db_size >> 3; j++) { 200 if (buf[j] != 0) { 201 panic("freed data not zero: " 202 "child=%p i=%d off=%d num=%d\n", 203 (void *)child, i, off, num); 204 } 205 } 206 } 207 208 /* 209 * db_data better be zeroed unless it's dirty in a 210 * future txg. 211 */ 212 mutex_enter(&child->db_mtx); 213 buf = child->db.db_data; 214 if (buf != NULL && child->db_state != DB_FILL && 215 child->db_last_dirty == NULL) { 216 for (j = 0; j < child->db.db_size >> 3; j++) { 217 if (buf[j] != 0) { 218 panic("freed data not zero: " 219 "child=%p i=%d off=%d num=%d\n", 220 (void *)child, i, off, num); 221 } 222 } 223 } 224 mutex_exit(&child->db_mtx); 225 226 dbuf_rele(child, FTAG); 227 } 228 DB_DNODE_EXIT(db); 229 } 230 #endif 231 232 /* 233 * We don't usually free the indirect blocks here. If in one txg we have a 234 * free_range and a write to the same indirect block, it's important that we 235 * preserve the hole's birth times. Therefore, we don't free any any indirect 236 * blocks in free_children(). If an indirect block happens to turn into all 237 * holes, it will be freed by dbuf_write_children_ready, which happens at a 238 * point in the syncing process where we know for certain the contents of the 239 * indirect block. 240 * 241 * However, if we're freeing a dnode, its space accounting must go to zero 242 * before we actually try to free the dnode, or we will trip an assertion. In 243 * addition, we know the case described above cannot occur, because the dnode is 244 * being freed. Therefore, we free the indirect blocks immediately in that 245 * case. 246 */ 247 static void 248 free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, 249 boolean_t free_indirects, dmu_tx_t *tx) 250 { 251 dnode_t *dn; 252 blkptr_t *bp; 253 dmu_buf_impl_t *subdb; 254 uint64_t start, end, dbstart, dbend; 255 unsigned int epbs, shift, i; 256 257 /* 258 * There is a small possibility that this block will not be cached: 259 * 1 - if level > 1 and there are no children with level <= 1 260 * 2 - if this block was evicted since we read it from 261 * dmu_tx_hold_free(). 262 */ 263 if (db->db_state != DB_CACHED) 264 (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); 265 266 dbuf_release_bp(db); 267 bp = db->db.db_data; 268 269 DB_DNODE_ENTER(db); 270 dn = DB_DNODE(db); 271 epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 272 ASSERT3U(epbs, <, 31); 273 shift = (db->db_level - 1) * epbs; 274 dbstart = db->db_blkid << epbs; 275 start = blkid >> shift; 276 if (dbstart < start) { 277 bp += start - dbstart; 278 } else { 279 start = dbstart; 280 } 281 dbend = ((db->db_blkid + 1) << epbs) - 1; 282 end = (blkid + nblks - 1) >> shift; 283 if (dbend <= end) 284 end = dbend; 285 286 ASSERT3U(start, <=, end); 287 288 if (db->db_level == 1) { 289 FREE_VERIFY(db, start, end, tx); 290 free_blocks(dn, bp, end-start+1, tx); 291 } else { 292 for (uint64_t id = start; id <= end; id++, bp++) { 293 if (BP_IS_HOLE(bp)) 294 continue; 295 rw_enter(&dn->dn_struct_rwlock, RW_READER); 296 VERIFY0(dbuf_hold_impl(dn, db->db_level - 1, 297 id, TRUE, FALSE, FTAG, &subdb)); 298 rw_exit(&dn->dn_struct_rwlock); 299 ASSERT3P(bp, ==, subdb->db_blkptr); 300 301 free_children(subdb, blkid, nblks, free_indirects, tx); 302 dbuf_rele(subdb, FTAG); 303 } 304 } 305 306 if (free_indirects) { 307 for (i = 0, bp = db->db.db_data; i < 1 << epbs; i++, bp++) 308 ASSERT(BP_IS_HOLE(bp)); 309 bzero(db->db.db_data, db->db.db_size); 310 free_blocks(dn, db->db_blkptr, 1, tx); 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 boolean_t free_indirects, 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, FALSE, FTAG, &db)); 361 rw_exit(&dn->dn_struct_rwlock); 362 363 free_children(db, blkid, nblks, free_indirects, 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 boolean_t dsfra_free_indirects; 383 } dnode_sync_free_range_arg_t; 384 385 static void 386 dnode_sync_free_range(void *arg, uint64_t blkid, uint64_t nblks) 387 { 388 dnode_sync_free_range_arg_t *dsfra = arg; 389 dnode_t *dn = dsfra->dsfra_dnode; 390 391 mutex_exit(&dn->dn_mtx); 392 dnode_sync_free_range_impl(dn, blkid, nblks, 393 dsfra->dsfra_free_indirects, dsfra->dsfra_tx); 394 mutex_enter(&dn->dn_mtx); 395 } 396 397 /* 398 * Try to kick all the dnode's dbufs out of the cache... 399 */ 400 void 401 dnode_evict_dbufs(dnode_t *dn) 402 { 403 dmu_buf_impl_t db_marker; 404 dmu_buf_impl_t *db, *db_next; 405 406 mutex_enter(&dn->dn_dbufs_mtx); 407 for (db = avl_first(&dn->dn_dbufs); db != NULL; db = db_next) { 408 409 #ifdef DEBUG 410 DB_DNODE_ENTER(db); 411 ASSERT3P(DB_DNODE(db), ==, dn); 412 DB_DNODE_EXIT(db); 413 #endif /* DEBUG */ 414 415 mutex_enter(&db->db_mtx); 416 if (db->db_state != DB_EVICTING && 417 refcount_is_zero(&db->db_holds)) { 418 db_marker.db_level = db->db_level; 419 db_marker.db_blkid = db->db_blkid; 420 db_marker.db_state = DB_SEARCH; 421 avl_insert_here(&dn->dn_dbufs, &db_marker, db, 422 AVL_BEFORE); 423 424 dbuf_destroy(db); 425 426 db_next = AVL_NEXT(&dn->dn_dbufs, &db_marker); 427 avl_remove(&dn->dn_dbufs, &db_marker); 428 } else { 429 db->db_pending_evict = TRUE; 430 mutex_exit(&db->db_mtx); 431 db_next = AVL_NEXT(&dn->dn_dbufs, db); 432 } 433 } 434 mutex_exit(&dn->dn_dbufs_mtx); 435 436 dnode_evict_bonus(dn); 437 } 438 439 void 440 dnode_evict_bonus(dnode_t *dn) 441 { 442 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 443 if (dn->dn_bonus != NULL) { 444 if (refcount_is_zero(&dn->dn_bonus->db_holds)) { 445 mutex_enter(&dn->dn_bonus->db_mtx); 446 dbuf_destroy(dn->dn_bonus); 447 dn->dn_bonus = NULL; 448 } else { 449 dn->dn_bonus->db_pending_evict = TRUE; 450 } 451 } 452 rw_exit(&dn->dn_struct_rwlock); 453 } 454 455 static void 456 dnode_undirty_dbufs(list_t *list) 457 { 458 dbuf_dirty_record_t *dr; 459 460 while (dr = list_head(list)) { 461 dmu_buf_impl_t *db = dr->dr_dbuf; 462 uint64_t txg = dr->dr_txg; 463 464 if (db->db_level != 0) 465 dnode_undirty_dbufs(&dr->dt.di.dr_children); 466 467 mutex_enter(&db->db_mtx); 468 /* XXX - use dbuf_undirty()? */ 469 list_remove(list, dr); 470 ASSERT(db->db_last_dirty == dr); 471 db->db_last_dirty = NULL; 472 db->db_dirtycnt -= 1; 473 if (db->db_level == 0) { 474 ASSERT(db->db_blkid == DMU_BONUS_BLKID || 475 dr->dt.dl.dr_data == db->db_buf); 476 dbuf_unoverride(dr); 477 } else { 478 mutex_destroy(&dr->dt.di.dr_mtx); 479 list_destroy(&dr->dt.di.dr_children); 480 } 481 kmem_free(dr, sizeof (dbuf_dirty_record_t)); 482 dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); 483 } 484 } 485 486 static void 487 dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) 488 { 489 int txgoff = tx->tx_txg & TXG_MASK; 490 491 ASSERT(dmu_tx_is_syncing(tx)); 492 493 /* 494 * Our contents should have been freed in dnode_sync() by the 495 * free range record inserted by the caller of dnode_free(). 496 */ 497 ASSERT0(DN_USED_BYTES(dn->dn_phys)); 498 ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); 499 500 dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); 501 dnode_evict_dbufs(dn); 502 503 /* 504 * XXX - It would be nice to assert this, but we may still 505 * have residual holds from async evictions from the arc... 506 * 507 * zfs_obj_to_path() also depends on this being 508 * commented out. 509 * 510 * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); 511 */ 512 513 /* Undirty next bits */ 514 dn->dn_next_nlevels[txgoff] = 0; 515 dn->dn_next_indblkshift[txgoff] = 0; 516 dn->dn_next_blksz[txgoff] = 0; 517 518 /* ASSERT(blkptrs are zero); */ 519 ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 520 ASSERT(dn->dn_type != DMU_OT_NONE); 521 522 ASSERT(dn->dn_free_txg > 0); 523 if (dn->dn_allocated_txg != dn->dn_free_txg) 524 dmu_buf_will_dirty(&dn->dn_dbuf->db, tx); 525 bzero(dn->dn_phys, sizeof (dnode_phys_t)); 526 527 mutex_enter(&dn->dn_mtx); 528 dn->dn_type = DMU_OT_NONE; 529 dn->dn_maxblkid = 0; 530 dn->dn_allocated_txg = 0; 531 dn->dn_free_txg = 0; 532 dn->dn_have_spill = B_FALSE; 533 mutex_exit(&dn->dn_mtx); 534 535 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 536 537 dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 538 /* 539 * Now that we've released our hold, the dnode may 540 * be evicted, so we musn't access it. 541 */ 542 } 543 544 /* 545 * Write out the dnode's dirty buffers. 546 */ 547 void 548 dnode_sync(dnode_t *dn, dmu_tx_t *tx) 549 { 550 dnode_phys_t *dnp = dn->dn_phys; 551 int txgoff = tx->tx_txg & TXG_MASK; 552 list_t *list = &dn->dn_dirty_records[txgoff]; 553 static const dnode_phys_t zerodn = { 0 }; 554 boolean_t kill_spill = B_FALSE; 555 556 ASSERT(dmu_tx_is_syncing(tx)); 557 ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); 558 ASSERT(dnp->dn_type != DMU_OT_NONE || 559 bcmp(dnp, &zerodn, DNODE_SIZE) == 0); 560 DNODE_VERIFY(dn); 561 562 ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); 563 564 if (dmu_objset_userused_enabled(dn->dn_objset) && 565 !DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 566 mutex_enter(&dn->dn_mtx); 567 dn->dn_oldused = DN_USED_BYTES(dn->dn_phys); 568 dn->dn_oldflags = dn->dn_phys->dn_flags; 569 dn->dn_phys->dn_flags |= DNODE_FLAG_USERUSED_ACCOUNTED; 570 mutex_exit(&dn->dn_mtx); 571 dmu_objset_userquota_get_ids(dn, B_FALSE, tx); 572 } else { 573 /* Once we account for it, we should always account for it. */ 574 ASSERT(!(dn->dn_phys->dn_flags & 575 DNODE_FLAG_USERUSED_ACCOUNTED)); 576 } 577 578 mutex_enter(&dn->dn_mtx); 579 if (dn->dn_allocated_txg == tx->tx_txg) { 580 /* The dnode is newly allocated or reallocated */ 581 if (dnp->dn_type == DMU_OT_NONE) { 582 /* this is a first alloc, not a realloc */ 583 dnp->dn_nlevels = 1; 584 dnp->dn_nblkptr = dn->dn_nblkptr; 585 } 586 587 dnp->dn_type = dn->dn_type; 588 dnp->dn_bonustype = dn->dn_bonustype; 589 dnp->dn_bonuslen = dn->dn_bonuslen; 590 } 591 ASSERT(dnp->dn_nlevels > 1 || 592 BP_IS_HOLE(&dnp->dn_blkptr[0]) || 593 BP_IS_EMBEDDED(&dnp->dn_blkptr[0]) || 594 BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 595 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 596 ASSERT(dnp->dn_nlevels < 2 || 597 BP_IS_HOLE(&dnp->dn_blkptr[0]) || 598 BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 1 << dnp->dn_indblkshift); 599 600 if (dn->dn_next_type[txgoff] != 0) { 601 dnp->dn_type = dn->dn_type; 602 dn->dn_next_type[txgoff] = 0; 603 } 604 605 if (dn->dn_next_blksz[txgoff] != 0) { 606 ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], 607 SPA_MINBLOCKSIZE) == 0); 608 ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || 609 dn->dn_maxblkid == 0 || list_head(list) != NULL || 610 dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == 611 dnp->dn_datablkszsec || 612 !range_tree_is_empty(dn->dn_free_ranges[txgoff])); 613 dnp->dn_datablkszsec = 614 dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; 615 dn->dn_next_blksz[txgoff] = 0; 616 } 617 618 if (dn->dn_next_bonuslen[txgoff] != 0) { 619 if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) 620 dnp->dn_bonuslen = 0; 621 else 622 dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; 623 ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN); 624 dn->dn_next_bonuslen[txgoff] = 0; 625 } 626 627 if (dn->dn_next_bonustype[txgoff] != 0) { 628 ASSERT(DMU_OT_IS_VALID(dn->dn_next_bonustype[txgoff])); 629 dnp->dn_bonustype = dn->dn_next_bonustype[txgoff]; 630 dn->dn_next_bonustype[txgoff] = 0; 631 } 632 633 boolean_t freeing_dnode = dn->dn_free_txg > 0 && 634 dn->dn_free_txg <= tx->tx_txg; 635 636 /* 637 * Remove the spill block if we have been explicitly asked to 638 * remove it, or if the object is being removed. 639 */ 640 if (dn->dn_rm_spillblk[txgoff] || freeing_dnode) { 641 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) 642 kill_spill = B_TRUE; 643 dn->dn_rm_spillblk[txgoff] = 0; 644 } 645 646 if (dn->dn_next_indblkshift[txgoff] != 0) { 647 ASSERT(dnp->dn_nlevels == 1); 648 dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; 649 dn->dn_next_indblkshift[txgoff] = 0; 650 } 651 652 /* 653 * Just take the live (open-context) values for checksum and compress. 654 * Strictly speaking it's a future leak, but nothing bad happens if we 655 * start using the new checksum or compress algorithm a little early. 656 */ 657 dnp->dn_checksum = dn->dn_checksum; 658 dnp->dn_compress = dn->dn_compress; 659 660 mutex_exit(&dn->dn_mtx); 661 662 if (kill_spill) { 663 free_blocks(dn, &dn->dn_phys->dn_spill, 1, tx); 664 mutex_enter(&dn->dn_mtx); 665 dnp->dn_flags &= ~DNODE_FLAG_SPILL_BLKPTR; 666 mutex_exit(&dn->dn_mtx); 667 } 668 669 /* process all the "freed" ranges in the file */ 670 if (dn->dn_free_ranges[txgoff] != NULL) { 671 dnode_sync_free_range_arg_t dsfra; 672 dsfra.dsfra_dnode = dn; 673 dsfra.dsfra_tx = tx; 674 dsfra.dsfra_free_indirects = freeing_dnode; 675 if (freeing_dnode) { 676 ASSERT(range_tree_contains(dn->dn_free_ranges[txgoff], 677 0, dn->dn_maxblkid + 1)); 678 } 679 mutex_enter(&dn->dn_mtx); 680 range_tree_vacate(dn->dn_free_ranges[txgoff], 681 dnode_sync_free_range, &dsfra); 682 range_tree_destroy(dn->dn_free_ranges[txgoff]); 683 dn->dn_free_ranges[txgoff] = NULL; 684 mutex_exit(&dn->dn_mtx); 685 } 686 687 if (freeing_dnode) { 688 dn->dn_objset->os_freed_dnodes++; 689 dnode_sync_free(dn, tx); 690 return; 691 } 692 693 if (dn->dn_next_nlevels[txgoff]) { 694 dnode_increase_indirection(dn, tx); 695 dn->dn_next_nlevels[txgoff] = 0; 696 } 697 698 if (dn->dn_next_nblkptr[txgoff]) { 699 /* this should only happen on a realloc */ 700 ASSERT(dn->dn_allocated_txg == tx->tx_txg); 701 if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { 702 /* zero the new blkptrs we are gaining */ 703 bzero(dnp->dn_blkptr + dnp->dn_nblkptr, 704 sizeof (blkptr_t) * 705 (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); 706 #ifdef ZFS_DEBUG 707 } else { 708 int i; 709 ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); 710 /* the blkptrs we are losing better be unallocated */ 711 for (i = dn->dn_next_nblkptr[txgoff]; 712 i < dnp->dn_nblkptr; i++) 713 ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); 714 #endif 715 } 716 mutex_enter(&dn->dn_mtx); 717 dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; 718 dn->dn_next_nblkptr[txgoff] = 0; 719 mutex_exit(&dn->dn_mtx); 720 } 721 722 dbuf_sync_list(list, dn->dn_phys->dn_nlevels - 1, tx); 723 724 if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 725 ASSERT3P(list_head(list), ==, NULL); 726 dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 727 } 728 729 /* 730 * Although we have dropped our reference to the dnode, it 731 * can't be evicted until its written, and we haven't yet 732 * initiated the IO for the dnode's dbuf. 733 */ 734 } 735