dnode_sync.c revision 288538
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 */ 26 27#include <sys/zfs_context.h> 28#include <sys/dbuf.h> 29#include <sys/dnode.h> 30#include <sys/dmu.h> 31#include <sys/dmu_tx.h> 32#include <sys/dmu_objset.h> 33#include <sys/dsl_dataset.h> 34#include <sys/spa.h> 35#include <sys/range_tree.h> 36#include <sys/zfeature.h> 37 38static void 39dnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) 40{ 41 dmu_buf_impl_t *db; 42 int txgoff = tx->tx_txg & TXG_MASK; 43 int nblkptr = dn->dn_phys->dn_nblkptr; 44 int old_toplvl = dn->dn_phys->dn_nlevels - 1; 45 int new_level = dn->dn_next_nlevels[txgoff]; 46 int i; 47 48 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 49 50 /* this dnode can't be paged out because it's dirty */ 51 ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 52 ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); 53 ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); 54 55 db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); 56 ASSERT(db != NULL); 57 58 dn->dn_phys->dn_nlevels = new_level; 59 dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, 60 dn->dn_object, dn->dn_phys->dn_nlevels); 61 62 /* check for existing blkptrs in the dnode */ 63 for (i = 0; i < nblkptr; i++) 64 if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[i])) 65 break; 66 if (i != nblkptr) { 67 /* transfer dnode's block pointers to new indirect block */ 68 (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); 69 ASSERT(db->db.db_data); 70 ASSERT(arc_released(db->db_buf)); 71 ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); 72 bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, 73 sizeof (blkptr_t) * nblkptr); 74 arc_buf_freeze(db->db_buf); 75 } 76 77 /* set dbuf's parent pointers to new indirect buf */ 78 for (i = 0; i < nblkptr; i++) { 79 dmu_buf_impl_t *child = 80 dbuf_find(dn->dn_objset, dn->dn_object, old_toplvl, i); 81 82 if (child == NULL) 83 continue; 84#ifdef DEBUG 85 DB_DNODE_ENTER(child); 86 ASSERT3P(DB_DNODE(child), ==, dn); 87 DB_DNODE_EXIT(child); 88#endif /* DEBUG */ 89 if (child->db_parent && child->db_parent != dn->dn_dbuf) { 90 ASSERT(child->db_parent->db_level == db->db_level); 91 ASSERT(child->db_blkptr != 92 &dn->dn_phys->dn_blkptr[child->db_blkid]); 93 mutex_exit(&child->db_mtx); 94 continue; 95 } 96 ASSERT(child->db_parent == NULL || 97 child->db_parent == dn->dn_dbuf); 98 99 child->db_parent = db; 100 dbuf_add_ref(db, child); 101 if (db->db.db_data) 102 child->db_blkptr = (blkptr_t *)db->db.db_data + i; 103 else 104 child->db_blkptr = NULL; 105 dprintf_dbuf_bp(child, child->db_blkptr, 106 "changed db_blkptr to new indirect %s", ""); 107 108 mutex_exit(&child->db_mtx); 109 } 110 111 bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); 112 113 dbuf_rele(db, FTAG); 114 115 rw_exit(&dn->dn_struct_rwlock); 116} 117 118static void 119free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) 120{ 121 dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; 122 uint64_t bytesfreed = 0; 123 124 dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); 125 126 for (int i = 0; i < num; i++, bp++) { 127 if (BP_IS_HOLE(bp)) 128 continue; 129 130 bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE); 131 ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); 132 133 /* 134 * Save some useful information on the holes being 135 * punched, including logical size, type, and indirection 136 * level. Retaining birth time enables detection of when 137 * holes are punched for reducing the number of free 138 * records transmitted during a zfs send. 139 */ 140 141 uint64_t lsize = BP_GET_LSIZE(bp); 142 dmu_object_type_t type = BP_GET_TYPE(bp); 143 uint64_t lvl = BP_GET_LEVEL(bp); 144 145 bzero(bp, sizeof (blkptr_t)); 146 147 if (spa_feature_is_active(dn->dn_objset->os_spa, 148 SPA_FEATURE_HOLE_BIRTH)) { 149 BP_SET_LSIZE(bp, lsize); 150 BP_SET_TYPE(bp, type); 151 BP_SET_LEVEL(bp, lvl); 152 BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0); 153 } 154 } 155 dnode_diduse_space(dn, -bytesfreed); 156} 157 158#ifdef ZFS_DEBUG 159static void 160free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) 161{ 162 int off, num; 163 int i, err, epbs; 164 uint64_t txg = tx->tx_txg; 165 dnode_t *dn; 166 167 DB_DNODE_ENTER(db); 168 dn = DB_DNODE(db); 169 epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 170 off = start - (db->db_blkid * 1<<epbs); 171 num = end - start + 1; 172 173 ASSERT3U(off, >=, 0); 174 ASSERT3U(num, >=, 0); 175 ASSERT3U(db->db_level, >, 0); 176 ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); 177 ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); 178 ASSERT(db->db_blkptr != NULL); 179 180 for (i = off; i < off+num; i++) { 181 uint64_t *buf; 182 dmu_buf_impl_t *child; 183 dbuf_dirty_record_t *dr; 184 int j; 185 186 ASSERT(db->db_level == 1); 187 188 rw_enter(&dn->dn_struct_rwlock, RW_READER); 189 err = dbuf_hold_impl(dn, db->db_level-1, 190 (db->db_blkid << epbs) + i, TRUE, FTAG, &child); 191 rw_exit(&dn->dn_struct_rwlock); 192 if (err == ENOENT) 193 continue; 194 ASSERT(err == 0); 195 ASSERT(child->db_level == 0); 196 dr = child->db_last_dirty; 197 while (dr && dr->dr_txg > txg) 198 dr = dr->dr_next; 199 ASSERT(dr == NULL || dr->dr_txg == txg); 200 201 /* data_old better be zeroed */ 202 if (dr) { 203 buf = dr->dt.dl.dr_data->b_data; 204 for (j = 0; j < child->db.db_size >> 3; j++) { 205 if (buf[j] != 0) { 206 panic("freed data not zero: " 207 "child=%p i=%d off=%d num=%d\n", 208 (void *)child, i, off, num); 209 } 210 } 211 } 212 213 /* 214 * db_data better be zeroed unless it's dirty in a 215 * future txg. 216 */ 217 mutex_enter(&child->db_mtx); 218 buf = child->db.db_data; 219 if (buf != NULL && child->db_state != DB_FILL && 220 child->db_last_dirty == NULL) { 221 for (j = 0; j < child->db.db_size >> 3; j++) { 222 if (buf[j] != 0) { 223 panic("freed data not zero: " 224 "child=%p i=%d off=%d num=%d\n", 225 (void *)child, i, off, num); 226 } 227 } 228 } 229 mutex_exit(&child->db_mtx); 230 231 dbuf_rele(child, FTAG); 232 } 233 DB_DNODE_EXIT(db); 234} 235#endif 236 237static void 238free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, 239 dmu_tx_t *tx) 240{ 241 dnode_t *dn; 242 blkptr_t *bp; 243 dmu_buf_impl_t *subdb; 244 uint64_t start, end, dbstart, dbend, i; 245 int epbs, shift; 246 247 /* 248 * There is a small possibility that this block will not be cached: 249 * 1 - if level > 1 and there are no children with level <= 1 250 * 2 - if this block was evicted since we read it from 251 * dmu_tx_hold_free(). 252 */ 253 if (db->db_state != DB_CACHED) 254 (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); 255 256 dbuf_release_bp(db); 257 bp = db->db.db_data; 258 259 DB_DNODE_ENTER(db); 260 dn = DB_DNODE(db); 261 epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 262 shift = (db->db_level - 1) * epbs; 263 dbstart = db->db_blkid << epbs; 264 start = blkid >> shift; 265 if (dbstart < start) { 266 bp += start - dbstart; 267 } else { 268 start = dbstart; 269 } 270 dbend = ((db->db_blkid + 1) << epbs) - 1; 271 end = (blkid + nblks - 1) >> shift; 272 if (dbend <= end) 273 end = dbend; 274 275 ASSERT3U(start, <=, end); 276 277 if (db->db_level == 1) { 278 FREE_VERIFY(db, start, end, tx); 279 free_blocks(dn, bp, end-start+1, tx); 280 } else { 281 for (i = start; i <= end; i++, bp++) { 282 if (BP_IS_HOLE(bp)) 283 continue; 284 rw_enter(&dn->dn_struct_rwlock, RW_READER); 285 VERIFY0(dbuf_hold_impl(dn, db->db_level - 1, 286 i, B_TRUE, FTAG, &subdb)); 287 rw_exit(&dn->dn_struct_rwlock); 288 ASSERT3P(bp, ==, subdb->db_blkptr); 289 290 free_children(subdb, blkid, nblks, tx); 291 dbuf_rele(subdb, FTAG); 292 } 293 } 294 295 /* If this whole block is free, free ourself too. */ 296 for (i = 0, bp = db->db.db_data; i < 1 << epbs; i++, bp++) { 297 if (!BP_IS_HOLE(bp)) 298 break; 299 } 300 if (i == 1 << epbs) { 301 /* didn't find any non-holes */ 302 bzero(db->db.db_data, db->db.db_size); 303 free_blocks(dn, db->db_blkptr, 1, tx); 304 } else { 305 /* 306 * Partial block free; must be marked dirty so that it 307 * will be written out. 308 */ 309 ASSERT(db->db_dirtycnt > 0); 310 } 311 312 DB_DNODE_EXIT(db); 313 arc_buf_freeze(db->db_buf); 314} 315 316/* 317 * Traverse the indicated range of the provided file 318 * and "free" all the blocks contained there. 319 */ 320static void 321dnode_sync_free_range_impl(dnode_t *dn, uint64_t blkid, uint64_t nblks, 322 dmu_tx_t *tx) 323{ 324 blkptr_t *bp = dn->dn_phys->dn_blkptr; 325 int dnlevel = dn->dn_phys->dn_nlevels; 326 boolean_t trunc = B_FALSE; 327 328 if (blkid > dn->dn_phys->dn_maxblkid) 329 return; 330 331 ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); 332 if (blkid + nblks > dn->dn_phys->dn_maxblkid) { 333 nblks = dn->dn_phys->dn_maxblkid - blkid + 1; 334 trunc = B_TRUE; 335 } 336 337 /* There are no indirect blocks in the object */ 338 if (dnlevel == 1) { 339 if (blkid >= dn->dn_phys->dn_nblkptr) { 340 /* this range was never made persistent */ 341 return; 342 } 343 ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); 344 free_blocks(dn, bp + blkid, nblks, tx); 345 } else { 346 int shift = (dnlevel - 1) * 347 (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); 348 int start = blkid >> shift; 349 int end = (blkid + nblks - 1) >> shift; 350 dmu_buf_impl_t *db; 351 352 ASSERT(start < dn->dn_phys->dn_nblkptr); 353 bp += start; 354 for (int i = start; i <= end; i++, bp++) { 355 if (BP_IS_HOLE(bp)) 356 continue; 357 rw_enter(&dn->dn_struct_rwlock, RW_READER); 358 VERIFY0(dbuf_hold_impl(dn, dnlevel - 1, i, 359 TRUE, FTAG, &db)); 360 rw_exit(&dn->dn_struct_rwlock); 361 362 free_children(db, blkid, nblks, tx); 363 dbuf_rele(db, FTAG); 364 } 365 } 366 367 if (trunc) { 368 dn->dn_phys->dn_maxblkid = blkid == 0 ? 0 : blkid - 1; 369 370 uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * 371 (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); 372 ASSERT(off < dn->dn_phys->dn_maxblkid || 373 dn->dn_phys->dn_maxblkid == 0 || 374 dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); 375 } 376} 377 378typedef struct dnode_sync_free_range_arg { 379 dnode_t *dsfra_dnode; 380 dmu_tx_t *dsfra_tx; 381} dnode_sync_free_range_arg_t; 382 383static void 384dnode_sync_free_range(void *arg, uint64_t blkid, uint64_t nblks) 385{ 386 dnode_sync_free_range_arg_t *dsfra = arg; 387 dnode_t *dn = dsfra->dsfra_dnode; 388 389 mutex_exit(&dn->dn_mtx); 390 dnode_sync_free_range_impl(dn, blkid, nblks, dsfra->dsfra_tx); 391 mutex_enter(&dn->dn_mtx); 392} 393 394/* 395 * Try to kick all the dnode's dbufs out of the cache... 396 */ 397void 398dnode_evict_dbufs(dnode_t *dn) 399{ 400 int progress; 401 int pass = 0; 402 403 do { 404 dmu_buf_impl_t *db, *db_next; 405 int evicting = FALSE; 406 407 progress = FALSE; 408 mutex_enter(&dn->dn_dbufs_mtx); 409 for (db = avl_first(&dn->dn_dbufs); db != NULL; db = db_next) { 410 db_next = AVL_NEXT(&dn->dn_dbufs, db); 411#ifdef DEBUG 412 DB_DNODE_ENTER(db); 413 ASSERT3P(DB_DNODE(db), ==, dn); 414 DB_DNODE_EXIT(db); 415#endif /* DEBUG */ 416 417 mutex_enter(&db->db_mtx); 418 if (db->db_state == DB_EVICTING) { 419 progress = TRUE; 420 evicting = TRUE; 421 mutex_exit(&db->db_mtx); 422 } else if (refcount_is_zero(&db->db_holds)) { 423 progress = TRUE; 424 dbuf_clear(db); /* exits db_mtx for us */ 425 } else { 426 mutex_exit(&db->db_mtx); 427 } 428 429 } 430 /* 431 * NB: we need to drop dn_dbufs_mtx between passes so 432 * that any DB_EVICTING dbufs can make progress. 433 * Ideally, we would have some cv we could wait on, but 434 * since we don't, just wait a bit to give the other 435 * thread a chance to run. 436 */ 437 mutex_exit(&dn->dn_dbufs_mtx); 438 if (evicting) 439 delay(1); 440 pass++; 441 ASSERT(pass < 100); /* sanity check */ 442 } while (progress); 443 444 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 445 if (dn->dn_bonus && refcount_is_zero(&dn->dn_bonus->db_holds)) { 446 mutex_enter(&dn->dn_bonus->db_mtx); 447 dbuf_evict(dn->dn_bonus); 448 dn->dn_bonus = NULL; 449 } 450 rw_exit(&dn->dn_struct_rwlock); 451} 452 453static void 454dnode_undirty_dbufs(list_t *list) 455{ 456 dbuf_dirty_record_t *dr; 457 458 while (dr = list_head(list)) { 459 dmu_buf_impl_t *db = dr->dr_dbuf; 460 uint64_t txg = dr->dr_txg; 461 462 if (db->db_level != 0) 463 dnode_undirty_dbufs(&dr->dt.di.dr_children); 464 465 mutex_enter(&db->db_mtx); 466 /* XXX - use dbuf_undirty()? */ 467 list_remove(list, dr); 468 ASSERT(db->db_last_dirty == dr); 469 db->db_last_dirty = NULL; 470 db->db_dirtycnt -= 1; 471 if (db->db_level == 0) { 472 ASSERT(db->db_blkid == DMU_BONUS_BLKID || 473 dr->dt.dl.dr_data == db->db_buf); 474 dbuf_unoverride(dr); 475 } else { 476 mutex_destroy(&dr->dt.di.dr_mtx); 477 list_destroy(&dr->dt.di.dr_children); 478 } 479 kmem_free(dr, sizeof (dbuf_dirty_record_t)); 480 dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); 481 } 482} 483 484static void 485dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) 486{ 487 int txgoff = tx->tx_txg & TXG_MASK; 488 489 ASSERT(dmu_tx_is_syncing(tx)); 490 491 /* 492 * Our contents should have been freed in dnode_sync() by the 493 * free range record inserted by the caller of dnode_free(). 494 */ 495 ASSERT0(DN_USED_BYTES(dn->dn_phys)); 496 ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); 497 498 dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); 499 dnode_evict_dbufs(dn); 500 ASSERT(avl_is_empty(&dn->dn_dbufs)); 501 ASSERT3P(dn->dn_bonus, ==, NULL); 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 */ 547void 548dnode_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_space(dn->dn_free_ranges[txgoff]) != 0); 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 mutex_enter(&dn->dn_mtx); 675 range_tree_vacate(dn->dn_free_ranges[txgoff], 676 dnode_sync_free_range, &dsfra); 677 range_tree_destroy(dn->dn_free_ranges[txgoff]); 678 dn->dn_free_ranges[txgoff] = NULL; 679 mutex_exit(&dn->dn_mtx); 680 } 681 682 if (freeing_dnode) { 683 dnode_sync_free(dn, tx); 684 return; 685 } 686 687 if (dn->dn_next_nlevels[txgoff]) { 688 dnode_increase_indirection(dn, tx); 689 dn->dn_next_nlevels[txgoff] = 0; 690 } 691 692 if (dn->dn_next_nblkptr[txgoff]) { 693 /* this should only happen on a realloc */ 694 ASSERT(dn->dn_allocated_txg == tx->tx_txg); 695 if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { 696 /* zero the new blkptrs we are gaining */ 697 bzero(dnp->dn_blkptr + dnp->dn_nblkptr, 698 sizeof (blkptr_t) * 699 (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); 700#ifdef ZFS_DEBUG 701 } else { 702 int i; 703 ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); 704 /* the blkptrs we are losing better be unallocated */ 705 for (i = dn->dn_next_nblkptr[txgoff]; 706 i < dnp->dn_nblkptr; i++) 707 ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); 708#endif 709 } 710 mutex_enter(&dn->dn_mtx); 711 dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; 712 dn->dn_next_nblkptr[txgoff] = 0; 713 mutex_exit(&dn->dn_mtx); 714 } 715 716 dbuf_sync_list(list, dn->dn_phys->dn_nlevels - 1, tx); 717 718 if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 719 ASSERT3P(list_head(list), ==, NULL); 720 dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 721 } 722 723 /* 724 * Although we have dropped our reference to the dnode, it 725 * can't be evicted until its written, and we haven't yet 726 * initiated the IO for the dnode's dbuf. 727 */ 728} 729