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 2010 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #include <sys/zfs_context.h> 27 #include <sys/dbuf.h> 28 #include <sys/dnode.h> 29 #include <sys/dmu.h> 30 #include <sys/dmu_tx.h> 31 #include <sys/dmu_objset.h> 32 #include <sys/dsl_dataset.h> 33 #include <sys/spa.h> 34 35 static void 36 dnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) 37 { 38 dmu_buf_impl_t *db; 39 int txgoff = tx->tx_txg & TXG_MASK; 40 int nblkptr = dn->dn_phys->dn_nblkptr; 41 int old_toplvl = dn->dn_phys->dn_nlevels - 1; 42 int new_level = dn->dn_next_nlevels[txgoff]; 43 int i; 44 45 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 46 47 /* this dnode can't be paged out because it's dirty */ 48 ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 49 ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); 50 ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); 51 52 db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); 53 ASSERT(db != NULL); 54 55 dn->dn_phys->dn_nlevels = new_level; 56 dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, 57 dn->dn_object, dn->dn_phys->dn_nlevels); 58 59 /* check for existing blkptrs in the dnode */ 60 for (i = 0; i < nblkptr; i++) 61 if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[i])) 62 break; 63 if (i != nblkptr) { 64 /* transfer dnode's block pointers to new indirect block */ 65 (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); 66 ASSERT(db->db.db_data); 67 ASSERT(arc_released(db->db_buf)); 68 ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); 69 bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, 70 sizeof (blkptr_t) * nblkptr); 71 arc_buf_freeze(db->db_buf); 72 } 73 74 /* set dbuf's parent pointers to new indirect buf */ 75 for (i = 0; i < nblkptr; i++) { 76 dmu_buf_impl_t *child = dbuf_find(dn, old_toplvl, i); 77 78 if (child == NULL) 79 continue; 80 ASSERT3P(child->db_dnode, ==, dn); 81 if (child->db_parent && child->db_parent != dn->dn_dbuf) { 82 ASSERT(child->db_parent->db_level == db->db_level); 83 ASSERT(child->db_blkptr != 84 &dn->dn_phys->dn_blkptr[child->db_blkid]); 85 mutex_exit(&child->db_mtx); 86 continue; 87 } 88 ASSERT(child->db_parent == NULL || 89 child->db_parent == dn->dn_dbuf); 90 91 child->db_parent = db; 92 dbuf_add_ref(db, child); 93 if (db->db.db_data) 94 child->db_blkptr = (blkptr_t *)db->db.db_data + i; 95 else 96 child->db_blkptr = NULL; 97 dprintf_dbuf_bp(child, child->db_blkptr, 98 "changed db_blkptr to new indirect %s", ""); 99 100 mutex_exit(&child->db_mtx); 101 } 102 103 bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); 104 105 dbuf_rele(db, FTAG); 106 107 rw_exit(&dn->dn_struct_rwlock); 108 } 109 110 static int 111 free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) 112 { 113 dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; 114 uint64_t bytesfreed = 0; 115 int i, blocks_freed = 0; 116 117 dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); 118 119 for (i = 0; i < num; i++, bp++) { 120 if (BP_IS_HOLE(bp)) 121 continue; 122 123 bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE); 124 ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); 125 bzero(bp, sizeof (blkptr_t)); 126 blocks_freed += 1; 127 } 128 dnode_diduse_space(dn, -bytesfreed); 129 return (blocks_freed); 130 } 131 132 #ifdef ZFS_DEBUG 133 static void 134 free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) 135 { 136 int off, num; 137 int i, err, epbs; 138 uint64_t txg = tx->tx_txg; 139 140 epbs = db->db_dnode->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 141 off = start - (db->db_blkid * 1<<epbs); 142 num = end - start + 1; 143 144 ASSERT3U(off, >=, 0); 145 ASSERT3U(num, >=, 0); 146 ASSERT3U(db->db_level, >, 0); 147 ASSERT3U(db->db.db_size, ==, 1<<db->db_dnode->dn_phys->dn_indblkshift); 148 ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); 149 ASSERT(db->db_blkptr != NULL); 150 151 for (i = off; i < off+num; i++) { 152 uint64_t *buf; 153 dmu_buf_impl_t *child; 154 dbuf_dirty_record_t *dr; 155 int j; 156 157 ASSERT(db->db_level == 1); 158 159 rw_enter(&db->db_dnode->dn_struct_rwlock, RW_READER); 160 err = dbuf_hold_impl(db->db_dnode, db->db_level-1, 161 (db->db_blkid << epbs) + i, TRUE, FTAG, &child); 162 rw_exit(&db->db_dnode->dn_struct_rwlock); 163 if (err == ENOENT) 164 continue; 165 ASSERT(err == 0); 166 ASSERT(child->db_level == 0); 167 dr = child->db_last_dirty; 168 while (dr && dr->dr_txg > txg) 169 dr = dr->dr_next; 170 ASSERT(dr == NULL || dr->dr_txg == txg); 171 172 /* data_old better be zeroed */ 173 if (dr) { 174 buf = dr->dt.dl.dr_data->b_data; 175 for (j = 0; j < child->db.db_size >> 3; j++) { 176 if (buf[j] != 0) { 177 panic("freed data not zero: " 178 "child=%p i=%d off=%d num=%d\n", 179 (void *)child, i, off, num); 180 } 181 } 182 } 183 184 /* 185 * db_data better be zeroed unless it's dirty in a 186 * future txg. 187 */ 188 mutex_enter(&child->db_mtx); 189 buf = child->db.db_data; 190 if (buf != NULL && child->db_state != DB_FILL && 191 child->db_last_dirty == NULL) { 192 for (j = 0; j < child->db.db_size >> 3; j++) { 193 if (buf[j] != 0) { 194 panic("freed data not zero: " 195 "child=%p i=%d off=%d num=%d\n", 196 (void *)child, i, off, num); 197 } 198 } 199 } 200 mutex_exit(&child->db_mtx); 201 202 dbuf_rele(child, FTAG); 203 } 204 } 205 #endif 206 207 #define ALL -1 208 209 static int 210 free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, int trunc, 211 dmu_tx_t *tx) 212 { 213 dnode_t *dn = db->db_dnode; 214 blkptr_t *bp; 215 dmu_buf_impl_t *subdb; 216 uint64_t start, end, dbstart, dbend, i; 217 int epbs, shift, err; 218 int all = TRUE; 219 int blocks_freed = 0; 220 221 /* 222 * There is a small possibility that this block will not be cached: 223 * 1 - if level > 1 and there are no children with level <= 1 224 * 2 - if we didn't get a dirty hold (because this block had just 225 * finished being written -- and so had no holds), and then this 226 * block got evicted before we got here. 227 */ 228 if (db->db_state != DB_CACHED) 229 (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); 230 231 arc_release(db->db_buf, db); 232 bp = (blkptr_t *)db->db.db_data; 233 234 epbs = db->db_dnode->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 235 shift = (db->db_level - 1) * epbs; 236 dbstart = db->db_blkid << epbs; 237 start = blkid >> shift; 238 if (dbstart < start) { 239 bp += start - dbstart; 240 all = FALSE; 241 } else { 242 start = dbstart; 243 } 244 dbend = ((db->db_blkid + 1) << epbs) - 1; 245 end = (blkid + nblks - 1) >> shift; 246 if (dbend <= end) 247 end = dbend; 248 else if (all) 249 all = trunc; 250 ASSERT3U(start, <=, end); 251 252 if (db->db_level == 1) { 253 FREE_VERIFY(db, start, end, tx); 254 blocks_freed = free_blocks(dn, bp, end-start+1, tx); 255 arc_buf_freeze(db->db_buf); 256 ASSERT(all || blocks_freed == 0 || db->db_last_dirty); 257 return (all ? ALL : blocks_freed); 258 } 259 260 for (i = start; i <= end; i++, bp++) { 261 if (BP_IS_HOLE(bp)) 262 continue; 263 rw_enter(&dn->dn_struct_rwlock, RW_READER); 264 err = dbuf_hold_impl(dn, db->db_level-1, i, TRUE, FTAG, &subdb); 265 ASSERT3U(err, ==, 0); 266 rw_exit(&dn->dn_struct_rwlock); 267 268 if (free_children(subdb, blkid, nblks, trunc, tx) == ALL) { 269 ASSERT3P(subdb->db_blkptr, ==, bp); 270 blocks_freed += free_blocks(dn, bp, 1, tx); 271 } else { 272 all = FALSE; 273 } 274 dbuf_rele(subdb, FTAG); 275 } 276 arc_buf_freeze(db->db_buf); 277 #ifdef ZFS_DEBUG 278 bp -= (end-start)+1; 279 for (i = start; i <= end; i++, bp++) { 280 if (i == start && blkid != 0) 281 continue; 282 else if (i == end && !trunc) 283 continue; 284 ASSERT3U(bp->blk_birth, ==, 0); 285 } 286 #endif 287 ASSERT(all || blocks_freed == 0 || db->db_last_dirty); 288 return (all ? ALL : blocks_freed); 289 } 290 291 /* 292 * free_range: Traverse the indicated range of the provided file 293 * and "free" all the blocks contained there. 294 */ 295 static void 296 dnode_sync_free_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) 297 { 298 blkptr_t *bp = dn->dn_phys->dn_blkptr; 299 dmu_buf_impl_t *db; 300 int trunc, start, end, shift, i, err; 301 int dnlevel = dn->dn_phys->dn_nlevels; 302 303 if (blkid > dn->dn_phys->dn_maxblkid) 304 return; 305 306 ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); 307 trunc = blkid + nblks > dn->dn_phys->dn_maxblkid; 308 if (trunc) 309 nblks = dn->dn_phys->dn_maxblkid - blkid + 1; 310 311 /* There are no indirect blocks in the object */ 312 if (dnlevel == 1) { 313 if (blkid >= dn->dn_phys->dn_nblkptr) { 314 /* this range was never made persistent */ 315 return; 316 } 317 ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); 318 (void) free_blocks(dn, bp + blkid, nblks, tx); 319 if (trunc) { 320 uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * 321 (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); 322 dn->dn_phys->dn_maxblkid = (blkid ? blkid - 1 : 0); 323 ASSERT(off < dn->dn_phys->dn_maxblkid || 324 dn->dn_phys->dn_maxblkid == 0 || 325 dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); 326 } 327 return; 328 } 329 330 shift = (dnlevel - 1) * (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); 331 start = blkid >> shift; 332 ASSERT(start < dn->dn_phys->dn_nblkptr); 333 end = (blkid + nblks - 1) >> shift; 334 bp += start; 335 for (i = start; i <= end; i++, bp++) { 336 if (BP_IS_HOLE(bp)) 337 continue; 338 rw_enter(&dn->dn_struct_rwlock, RW_READER); 339 err = dbuf_hold_impl(dn, dnlevel-1, i, TRUE, FTAG, &db); 340 ASSERT3U(err, ==, 0); 341 rw_exit(&dn->dn_struct_rwlock); 342 343 if (free_children(db, blkid, nblks, trunc, tx) == ALL) { 344 ASSERT3P(db->db_blkptr, ==, bp); 345 (void) free_blocks(dn, bp, 1, tx); 346 } 347 dbuf_rele(db, FTAG); 348 } 349 if (trunc) { 350 uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * 351 (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); 352 dn->dn_phys->dn_maxblkid = (blkid ? blkid - 1 : 0); 353 ASSERT(off < dn->dn_phys->dn_maxblkid || 354 dn->dn_phys->dn_maxblkid == 0 || 355 dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); 356 } 357 } 358 359 /* 360 * Try to kick all the dnodes dbufs out of the cache... 361 */ 362 void 363 dnode_evict_dbufs(dnode_t *dn) 364 { 365 int progress; 366 int pass = 0; 367 368 do { 369 dmu_buf_impl_t *db, marker; 370 int evicting = FALSE; 371 372 progress = FALSE; 373 mutex_enter(&dn->dn_dbufs_mtx); 374 list_insert_tail(&dn->dn_dbufs, &marker); 375 db = list_head(&dn->dn_dbufs); 376 for (; db != ▮ db = list_head(&dn->dn_dbufs)) { 377 list_remove(&dn->dn_dbufs, db); 378 list_insert_tail(&dn->dn_dbufs, db); 379 ASSERT3P(db->db_dnode, ==, dn); 380 381 mutex_enter(&db->db_mtx); 382 if (db->db_state == DB_EVICTING) { 383 progress = TRUE; 384 evicting = TRUE; 385 mutex_exit(&db->db_mtx); 386 } else if (refcount_is_zero(&db->db_holds)) { 387 progress = TRUE; 388 dbuf_clear(db); /* exits db_mtx for us */ 389 } else { 390 mutex_exit(&db->db_mtx); 391 } 392 393 } 394 list_remove(&dn->dn_dbufs, &marker); 395 /* 396 * NB: we need to drop dn_dbufs_mtx between passes so 397 * that any DB_EVICTING dbufs can make progress. 398 * Ideally, we would have some cv we could wait on, but 399 * since we don't, just wait a bit to give the other 400 * thread a chance to run. 401 */ 402 mutex_exit(&dn->dn_dbufs_mtx); 403 if (evicting) 404 delay(1); 405 pass++; 406 ASSERT(pass < 100); /* sanity check */ 407 } while (progress); 408 409 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 410 if (dn->dn_bonus && refcount_is_zero(&dn->dn_bonus->db_holds)) { 411 mutex_enter(&dn->dn_bonus->db_mtx); 412 dbuf_evict(dn->dn_bonus); 413 dn->dn_bonus = NULL; 414 } 415 rw_exit(&dn->dn_struct_rwlock); 416 } 417 418 static void 419 dnode_undirty_dbufs(list_t *list) 420 { 421 dbuf_dirty_record_t *dr; 422 423 while (dr = list_head(list)) { 424 dmu_buf_impl_t *db = dr->dr_dbuf; 425 uint64_t txg = dr->dr_txg; 426 427 if (db->db_level != 0) 428 dnode_undirty_dbufs(&dr->dt.di.dr_children); 429 430 mutex_enter(&db->db_mtx); 431 /* XXX - use dbuf_undirty()? */ 432 list_remove(list, dr); 433 ASSERT(db->db_last_dirty == dr); 434 db->db_last_dirty = NULL; 435 db->db_dirtycnt -= 1; 436 if (db->db_level == 0) { 437 ASSERT(db->db_blkid == DMU_BONUS_BLKID || 438 dr->dt.dl.dr_data == db->db_buf); 439 dbuf_unoverride(dr); 440 } 441 kmem_free(dr, sizeof (dbuf_dirty_record_t)); 442 dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); 443 } 444 } 445 446 static void 447 dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) 448 { 449 int txgoff = tx->tx_txg & TXG_MASK; 450 451 ASSERT(dmu_tx_is_syncing(tx)); 452 453 /* 454 * Our contents should have been freed in dnode_sync() by the 455 * free range record inserted by the caller of dnode_free(). 456 */ 457 ASSERT3U(DN_USED_BYTES(dn->dn_phys), ==, 0); 458 ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); 459 460 dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); 461 dnode_evict_dbufs(dn); 462 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); 463 464 /* 465 * XXX - It would be nice to assert this, but we may still 466 * have residual holds from async evictions from the arc... 467 * 468 * zfs_obj_to_path() also depends on this being 469 * commented out. 470 * 471 * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); 472 */ 473 474 /* Undirty next bits */ 475 dn->dn_next_nlevels[txgoff] = 0; 476 dn->dn_next_indblkshift[txgoff] = 0; 477 dn->dn_next_blksz[txgoff] = 0; 478 479 /* ASSERT(blkptrs are zero); */ 480 ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 481 ASSERT(dn->dn_type != DMU_OT_NONE); 482 483 ASSERT(dn->dn_free_txg > 0); 484 if (dn->dn_allocated_txg != dn->dn_free_txg) 485 dbuf_will_dirty(dn->dn_dbuf, tx); 486 bzero(dn->dn_phys, sizeof (dnode_phys_t)); 487 488 mutex_enter(&dn->dn_mtx); 489 dn->dn_type = DMU_OT_NONE; 490 dn->dn_maxblkid = 0; 491 dn->dn_allocated_txg = 0; 492 dn->dn_free_txg = 0; 493 dn->dn_have_spill = B_FALSE; 494 mutex_exit(&dn->dn_mtx); 495 496 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 497 498 dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 499 /* 500 * Now that we've released our hold, the dnode may 501 * be evicted, so we musn't access it. 502 */ 503 } 504 505 /* 506 * Write out the dnode's dirty buffers. 507 */ 508 void 509 dnode_sync(dnode_t *dn, dmu_tx_t *tx) 510 { 511 free_range_t *rp; 512 dnode_phys_t *dnp = dn->dn_phys; 513 int txgoff = tx->tx_txg & TXG_MASK; 514 list_t *list = &dn->dn_dirty_records[txgoff]; 515 static const dnode_phys_t zerodn = { 0 }; 516 boolean_t kill_spill = B_FALSE; 517 518 ASSERT(dmu_tx_is_syncing(tx)); 519 ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); 520 ASSERT(dnp->dn_type != DMU_OT_NONE || 521 bcmp(dnp, &zerodn, DNODE_SIZE) == 0); 522 DNODE_VERIFY(dn); 523 524 ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); 525 526 if (dmu_objset_userused_enabled(dn->dn_objset) && 527 !DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 528 mutex_enter(&dn->dn_mtx); 529 dn->dn_oldused = DN_USED_BYTES(dn->dn_phys); 530 dn->dn_oldflags = dn->dn_phys->dn_flags; 531 dn->dn_id_flags |= DN_ID_SYNC; 532 dn->dn_phys->dn_flags |= DNODE_FLAG_USERUSED_ACCOUNTED; 533 mutex_exit(&dn->dn_mtx); 534 dmu_objset_userquota_get_ids(dn, B_FALSE); 535 } else { 536 /* Once we account for it, we should always account for it. */ 537 ASSERT(!(dn->dn_phys->dn_flags & 538 DNODE_FLAG_USERUSED_ACCOUNTED)); 539 } 540 541 mutex_enter(&dn->dn_mtx); 542 if (dn->dn_allocated_txg == tx->tx_txg) { 543 /* The dnode is newly allocated or reallocated */ 544 if (dnp->dn_type == DMU_OT_NONE) { 545 /* this is a first alloc, not a realloc */ 546 dnp->dn_nlevels = 1; 547 dnp->dn_nblkptr = dn->dn_nblkptr; 548 } 549 550 dnp->dn_type = dn->dn_type; 551 dnp->dn_bonustype = dn->dn_bonustype; 552 dnp->dn_bonuslen = dn->dn_bonuslen; 553 } 554 555 ASSERT(dnp->dn_nlevels > 1 || 556 BP_IS_HOLE(&dnp->dn_blkptr[0]) || 557 BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 558 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 559 560 if (dn->dn_next_blksz[txgoff]) { 561 ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], 562 SPA_MINBLOCKSIZE) == 0); 563 ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || 564 dn->dn_maxblkid == 0 || list_head(list) != NULL || 565 dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == 566 dnp->dn_datablkszsec); 567 dnp->dn_datablkszsec = 568 dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; 569 dn->dn_next_blksz[txgoff] = 0; 570 } 571 572 if (dn->dn_next_bonuslen[txgoff]) { 573 if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) 574 dnp->dn_bonuslen = 0; 575 else 576 dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; 577 ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN); 578 dn->dn_next_bonuslen[txgoff] = 0; 579 } 580 581 if (dn->dn_next_bonustype[txgoff]) { 582 ASSERT(dn->dn_next_bonustype[txgoff] < DMU_OT_NUMTYPES); 583 dnp->dn_bonustype = dn->dn_next_bonustype[txgoff]; 584 dn->dn_next_bonustype[txgoff] = 0; 585 } 586 587 /* 588 * We will either remove a spill block when a file is being removed 589 * or we have been asked to remove it. 590 */ 591 if (dn->dn_rm_spillblk[txgoff] || 592 ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) && 593 dn->dn_free_txg > 0 && dn->dn_free_txg <= tx->tx_txg)) { 594 if ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) 595 kill_spill = B_TRUE; 596 dn->dn_rm_spillblk[txgoff] = 0; 597 } 598 599 if (dn->dn_next_indblkshift[txgoff]) { 600 ASSERT(dnp->dn_nlevels == 1); 601 dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; 602 dn->dn_next_indblkshift[txgoff] = 0; 603 } 604 605 /* 606 * Just take the live (open-context) values for checksum and compress. 607 * Strictly speaking it's a future leak, but nothing bad happens if we 608 * start using the new checksum or compress algorithm a little early. 609 */ 610 dnp->dn_checksum = dn->dn_checksum; 611 dnp->dn_compress = dn->dn_compress; 612 613 mutex_exit(&dn->dn_mtx); 614 615 if (kill_spill) { 616 dmu_buf_impl_t *spilldb; 617 (void) free_blocks(dn, &dn->dn_phys->dn_spill, 1, tx); 618 mutex_enter(&dn->dn_mtx); 619 dnp->dn_flags &= ~DNODE_FLAG_SPILL_BLKPTR; 620 mutex_exit(&dn->dn_mtx); 621 rw_enter(&dn->dn_struct_rwlock, RW_READER); 622 spilldb = dbuf_find(dn, 0, DMU_SPILL_BLKID); 623 if (spilldb) { 624 spilldb->db_blkptr = NULL; 625 mutex_exit(&spilldb->db_mtx); 626 } 627 rw_exit(&dn->dn_struct_rwlock); 628 } 629 630 /* process all the "freed" ranges in the file */ 631 while (rp = avl_last(&dn->dn_ranges[txgoff])) { 632 dnode_sync_free_range(dn, rp->fr_blkid, rp->fr_nblks, tx); 633 /* grab the mutex so we don't race with dnode_block_freed() */ 634 mutex_enter(&dn->dn_mtx); 635 avl_remove(&dn->dn_ranges[txgoff], rp); 636 mutex_exit(&dn->dn_mtx); 637 kmem_free(rp, sizeof (free_range_t)); 638 } 639 640 if (dn->dn_free_txg > 0 && dn->dn_free_txg <= tx->tx_txg) { 641 dnode_sync_free(dn, tx); 642 return; 643 } 644 645 if (dn->dn_next_nblkptr[txgoff]) { 646 /* this should only happen on a realloc */ 647 ASSERT(dn->dn_allocated_txg == tx->tx_txg); 648 if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { 649 /* zero the new blkptrs we are gaining */ 650 bzero(dnp->dn_blkptr + dnp->dn_nblkptr, 651 sizeof (blkptr_t) * 652 (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); 653 #ifdef ZFS_DEBUG 654 } else { 655 int i; 656 ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); 657 /* the blkptrs we are losing better be unallocated */ 658 for (i = dn->dn_next_nblkptr[txgoff]; 659 i < dnp->dn_nblkptr; i++) 660 ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); 661 #endif 662 } 663 mutex_enter(&dn->dn_mtx); 664 dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; 665 dn->dn_next_nblkptr[txgoff] = 0; 666 mutex_exit(&dn->dn_mtx); 667 } 668 669 if (dn->dn_next_nlevels[txgoff]) { 670 dnode_increase_indirection(dn, tx); 671 dn->dn_next_nlevels[txgoff] = 0; 672 } 673 674 dbuf_sync_list(list, tx); 675 676 if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 677 ASSERT3P(list_head(list), ==, NULL); 678 dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 679 } 680 681 /* 682 * Although we have dropped our reference to the dnode, it 683 * can't be evicted until its written, and we haven't yet 684 * initiated the IO for the dnode's dbuf. 685 */ 686 } 687