1168404Spjd/* 2168404Spjd * CDDL HEADER START 3168404Spjd * 4168404Spjd * The contents of this file are subject to the terms of the 5168404Spjd * Common Development and Distribution License (the "License"). 6168404Spjd * You may not use this file except in compliance with the License. 7168404Spjd * 8168404Spjd * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9168404Spjd * or http://www.opensolaris.org/os/licensing. 10168404Spjd * See the License for the specific language governing permissions 11168404Spjd * and limitations under the License. 12168404Spjd * 13168404Spjd * When distributing Covered Code, include this CDDL HEADER in each 14168404Spjd * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15168404Spjd * If applicable, add the following below this CDDL HEADER, with the 16168404Spjd * fields enclosed by brackets "[]" replaced with your own identifying 17168404Spjd * information: Portions Copyright [yyyy] [name of copyright owner] 18168404Spjd * 19168404Spjd * CDDL HEADER END 20168404Spjd */ 21236884Smm 22168404Spjd/* 23219089Spjd * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24265740Sdelphij * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 25168404Spjd */ 26168404Spjd 27168404Spjd#include <sys/zfs_context.h> 28168404Spjd#include <sys/dbuf.h> 29168404Spjd#include <sys/dnode.h> 30168404Spjd#include <sys/dmu.h> 31168404Spjd#include <sys/dmu_tx.h> 32168404Spjd#include <sys/dmu_objset.h> 33168404Spjd#include <sys/dsl_dataset.h> 34168404Spjd#include <sys/spa.h> 35265740Sdelphij#include <sys/range_tree.h> 36263397Sdelphij#include <sys/zfeature.h> 37168404Spjd 38168404Spjdstatic void 39168404Spjddnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) 40168404Spjd{ 41168404Spjd dmu_buf_impl_t *db; 42168404Spjd int txgoff = tx->tx_txg & TXG_MASK; 43168404Spjd int nblkptr = dn->dn_phys->dn_nblkptr; 44168404Spjd int old_toplvl = dn->dn_phys->dn_nlevels - 1; 45168404Spjd int new_level = dn->dn_next_nlevels[txgoff]; 46168404Spjd int i; 47168404Spjd 48168404Spjd rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 49168404Spjd 50168404Spjd /* this dnode can't be paged out because it's dirty */ 51168404Spjd ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 52168404Spjd ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); 53168404Spjd ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); 54168404Spjd 55168404Spjd db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); 56168404Spjd ASSERT(db != NULL); 57168404Spjd 58168404Spjd dn->dn_phys->dn_nlevels = new_level; 59185029Spjd dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, 60185029Spjd dn->dn_object, dn->dn_phys->dn_nlevels); 61168404Spjd 62168404Spjd /* check for existing blkptrs in the dnode */ 63168404Spjd for (i = 0; i < nblkptr; i++) 64168404Spjd if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[i])) 65168404Spjd break; 66168404Spjd if (i != nblkptr) { 67168404Spjd /* transfer dnode's block pointers to new indirect block */ 68168404Spjd (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); 69168404Spjd ASSERT(db->db.db_data); 70168404Spjd ASSERT(arc_released(db->db_buf)); 71168404Spjd ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); 72168404Spjd bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, 73168404Spjd sizeof (blkptr_t) * nblkptr); 74168404Spjd arc_buf_freeze(db->db_buf); 75168404Spjd } 76168404Spjd 77168404Spjd /* set dbuf's parent pointers to new indirect buf */ 78168404Spjd for (i = 0; i < nblkptr; i++) { 79168404Spjd dmu_buf_impl_t *child = dbuf_find(dn, old_toplvl, i); 80168404Spjd 81168404Spjd if (child == NULL) 82168404Spjd continue; 83219089Spjd#ifdef DEBUG 84219089Spjd DB_DNODE_ENTER(child); 85219089Spjd ASSERT3P(DB_DNODE(child), ==, dn); 86219089Spjd DB_DNODE_EXIT(child); 87219089Spjd#endif /* DEBUG */ 88168404Spjd if (child->db_parent && child->db_parent != dn->dn_dbuf) { 89168404Spjd ASSERT(child->db_parent->db_level == db->db_level); 90168404Spjd ASSERT(child->db_blkptr != 91168404Spjd &dn->dn_phys->dn_blkptr[child->db_blkid]); 92168404Spjd mutex_exit(&child->db_mtx); 93168404Spjd continue; 94168404Spjd } 95168404Spjd ASSERT(child->db_parent == NULL || 96168404Spjd child->db_parent == dn->dn_dbuf); 97168404Spjd 98168404Spjd child->db_parent = db; 99168404Spjd dbuf_add_ref(db, child); 100168404Spjd if (db->db.db_data) 101168404Spjd child->db_blkptr = (blkptr_t *)db->db.db_data + i; 102168404Spjd else 103168404Spjd child->db_blkptr = NULL; 104168404Spjd dprintf_dbuf_bp(child, child->db_blkptr, 105168404Spjd "changed db_blkptr to new indirect %s", ""); 106168404Spjd 107168404Spjd mutex_exit(&child->db_mtx); 108168404Spjd } 109168404Spjd 110168404Spjd bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); 111168404Spjd 112168404Spjd dbuf_rele(db, FTAG); 113168404Spjd 114168404Spjd rw_exit(&dn->dn_struct_rwlock); 115168404Spjd} 116168404Spjd 117263397Sdelphijstatic void 118168404Spjdfree_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) 119168404Spjd{ 120185029Spjd dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; 121168404Spjd uint64_t bytesfreed = 0; 122168404Spjd 123185029Spjd dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); 124168404Spjd 125263397Sdelphij for (int i = 0; i < num; i++, bp++) { 126168404Spjd if (BP_IS_HOLE(bp)) 127168404Spjd continue; 128168404Spjd 129219089Spjd bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE); 130168404Spjd ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); 131263397Sdelphij 132263397Sdelphij /* 133263397Sdelphij * Save some useful information on the holes being 134263397Sdelphij * punched, including logical size, type, and indirection 135263397Sdelphij * level. Retaining birth time enables detection of when 136263397Sdelphij * holes are punched for reducing the number of free 137263397Sdelphij * records transmitted during a zfs send. 138263397Sdelphij */ 139263397Sdelphij 140263397Sdelphij uint64_t lsize = BP_GET_LSIZE(bp); 141263397Sdelphij dmu_object_type_t type = BP_GET_TYPE(bp); 142263397Sdelphij uint64_t lvl = BP_GET_LEVEL(bp); 143263397Sdelphij 144168404Spjd bzero(bp, sizeof (blkptr_t)); 145263397Sdelphij 146263397Sdelphij if (spa_feature_is_active(dn->dn_objset->os_spa, 147263397Sdelphij SPA_FEATURE_HOLE_BIRTH)) { 148263397Sdelphij BP_SET_LSIZE(bp, lsize); 149263397Sdelphij BP_SET_TYPE(bp, type); 150263397Sdelphij BP_SET_LEVEL(bp, lvl); 151263397Sdelphij BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0); 152263397Sdelphij } 153168404Spjd } 154168404Spjd dnode_diduse_space(dn, -bytesfreed); 155168404Spjd} 156168404Spjd 157168404Spjd#ifdef ZFS_DEBUG 158168404Spjdstatic void 159168404Spjdfree_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) 160168404Spjd{ 161168404Spjd int off, num; 162168404Spjd int i, err, epbs; 163168404Spjd uint64_t txg = tx->tx_txg; 164219089Spjd dnode_t *dn; 165168404Spjd 166219089Spjd DB_DNODE_ENTER(db); 167219089Spjd dn = DB_DNODE(db); 168219089Spjd epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 169168404Spjd off = start - (db->db_blkid * 1<<epbs); 170168404Spjd num = end - start + 1; 171168404Spjd 172168404Spjd ASSERT3U(off, >=, 0); 173168404Spjd ASSERT3U(num, >=, 0); 174168404Spjd ASSERT3U(db->db_level, >, 0); 175219089Spjd ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); 176168404Spjd ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); 177168404Spjd ASSERT(db->db_blkptr != NULL); 178168404Spjd 179168404Spjd for (i = off; i < off+num; i++) { 180168404Spjd uint64_t *buf; 181168404Spjd dmu_buf_impl_t *child; 182168404Spjd dbuf_dirty_record_t *dr; 183168404Spjd int j; 184168404Spjd 185168404Spjd ASSERT(db->db_level == 1); 186168404Spjd 187219089Spjd rw_enter(&dn->dn_struct_rwlock, RW_READER); 188219089Spjd err = dbuf_hold_impl(dn, db->db_level-1, 189185029Spjd (db->db_blkid << epbs) + i, TRUE, FTAG, &child); 190219089Spjd rw_exit(&dn->dn_struct_rwlock); 191168404Spjd if (err == ENOENT) 192168404Spjd continue; 193168404Spjd ASSERT(err == 0); 194168404Spjd ASSERT(child->db_level == 0); 195168404Spjd dr = child->db_last_dirty; 196168404Spjd while (dr && dr->dr_txg > txg) 197168404Spjd dr = dr->dr_next; 198168404Spjd ASSERT(dr == NULL || dr->dr_txg == txg); 199168404Spjd 200168404Spjd /* data_old better be zeroed */ 201168404Spjd if (dr) { 202168404Spjd buf = dr->dt.dl.dr_data->b_data; 203168404Spjd for (j = 0; j < child->db.db_size >> 3; j++) { 204168404Spjd if (buf[j] != 0) { 205168404Spjd panic("freed data not zero: " 206168404Spjd "child=%p i=%d off=%d num=%d\n", 207185029Spjd (void *)child, i, off, num); 208168404Spjd } 209168404Spjd } 210168404Spjd } 211168404Spjd 212168404Spjd /* 213168404Spjd * db_data better be zeroed unless it's dirty in a 214168404Spjd * future txg. 215168404Spjd */ 216168404Spjd mutex_enter(&child->db_mtx); 217168404Spjd buf = child->db.db_data; 218168404Spjd if (buf != NULL && child->db_state != DB_FILL && 219168404Spjd child->db_last_dirty == NULL) { 220168404Spjd for (j = 0; j < child->db.db_size >> 3; j++) { 221168404Spjd if (buf[j] != 0) { 222168404Spjd panic("freed data not zero: " 223168404Spjd "child=%p i=%d off=%d num=%d\n", 224185029Spjd (void *)child, i, off, num); 225168404Spjd } 226168404Spjd } 227168404Spjd } 228168404Spjd mutex_exit(&child->db_mtx); 229168404Spjd 230168404Spjd dbuf_rele(child, FTAG); 231168404Spjd } 232219089Spjd DB_DNODE_EXIT(db); 233168404Spjd} 234168404Spjd#endif 235168404Spjd 236263397Sdelphijstatic void 237263397Sdelphijfree_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, 238168404Spjd dmu_tx_t *tx) 239168404Spjd{ 240219089Spjd dnode_t *dn; 241168404Spjd blkptr_t *bp; 242168404Spjd dmu_buf_impl_t *subdb; 243168404Spjd uint64_t start, end, dbstart, dbend, i; 244263397Sdelphij int epbs, shift; 245168404Spjd 246185029Spjd /* 247185029Spjd * There is a small possibility that this block will not be cached: 248185029Spjd * 1 - if level > 1 and there are no children with level <= 1 249263397Sdelphij * 2 - if this block was evicted since we read it from 250263397Sdelphij * dmu_tx_hold_free(). 251185029Spjd */ 252185029Spjd if (db->db_state != DB_CACHED) 253185029Spjd (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); 254185029Spjd 255219089Spjd dbuf_release_bp(db); 256263397Sdelphij bp = db->db.db_data; 257168404Spjd 258219089Spjd DB_DNODE_ENTER(db); 259219089Spjd dn = DB_DNODE(db); 260219089Spjd epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 261168404Spjd shift = (db->db_level - 1) * epbs; 262168404Spjd dbstart = db->db_blkid << epbs; 263168404Spjd start = blkid >> shift; 264168404Spjd if (dbstart < start) { 265168404Spjd bp += start - dbstart; 266168404Spjd } else { 267168404Spjd start = dbstart; 268168404Spjd } 269168404Spjd dbend = ((db->db_blkid + 1) << epbs) - 1; 270168404Spjd end = (blkid + nblks - 1) >> shift; 271168404Spjd if (dbend <= end) 272168404Spjd end = dbend; 273263397Sdelphij 274168404Spjd ASSERT3U(start, <=, end); 275168404Spjd 276168404Spjd if (db->db_level == 1) { 277168404Spjd FREE_VERIFY(db, start, end, tx); 278263397Sdelphij free_blocks(dn, bp, end-start+1, tx); 279263397Sdelphij } else { 280263397Sdelphij for (i = start; i <= end; i++, bp++) { 281263397Sdelphij if (BP_IS_HOLE(bp)) 282263397Sdelphij continue; 283263397Sdelphij rw_enter(&dn->dn_struct_rwlock, RW_READER); 284263397Sdelphij VERIFY0(dbuf_hold_impl(dn, db->db_level - 1, 285263397Sdelphij i, B_TRUE, FTAG, &subdb)); 286263397Sdelphij rw_exit(&dn->dn_struct_rwlock); 287263397Sdelphij ASSERT3P(bp, ==, subdb->db_blkptr); 288263397Sdelphij 289263397Sdelphij free_children(subdb, blkid, nblks, tx); 290263397Sdelphij dbuf_rele(subdb, FTAG); 291263397Sdelphij } 292168404Spjd } 293168404Spjd 294263397Sdelphij /* If this whole block is free, free ourself too. */ 295263397Sdelphij for (i = 0, bp = db->db.db_data; i < 1 << epbs; i++, bp++) { 296263397Sdelphij if (!BP_IS_HOLE(bp)) 297263397Sdelphij break; 298263397Sdelphij } 299263397Sdelphij if (i == 1 << epbs) { 300263397Sdelphij /* didn't find any non-holes */ 301263397Sdelphij bzero(db->db.db_data, db->db.db_size); 302263397Sdelphij free_blocks(dn, db->db_blkptr, 1, tx); 303263397Sdelphij } else { 304263397Sdelphij /* 305263397Sdelphij * Partial block free; must be marked dirty so that it 306263397Sdelphij * will be written out. 307263397Sdelphij */ 308263397Sdelphij ASSERT(db->db_dirtycnt > 0); 309263397Sdelphij } 310168404Spjd 311219089Spjd DB_DNODE_EXIT(db); 312168404Spjd arc_buf_freeze(db->db_buf); 313168404Spjd} 314168404Spjd 315168404Spjd/* 316251631Sdelphij * Traverse the indicated range of the provided file 317168404Spjd * and "free" all the blocks contained there. 318168404Spjd */ 319168404Spjdstatic void 320265740Sdelphijdnode_sync_free_range_impl(dnode_t *dn, uint64_t blkid, uint64_t nblks, 321263397Sdelphij dmu_tx_t *tx) 322168404Spjd{ 323168404Spjd blkptr_t *bp = dn->dn_phys->dn_blkptr; 324168404Spjd int dnlevel = dn->dn_phys->dn_nlevels; 325263397Sdelphij boolean_t trunc = B_FALSE; 326168404Spjd 327168404Spjd if (blkid > dn->dn_phys->dn_maxblkid) 328168404Spjd return; 329168404Spjd 330168404Spjd ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); 331263397Sdelphij if (blkid + nblks > dn->dn_phys->dn_maxblkid) { 332168404Spjd nblks = dn->dn_phys->dn_maxblkid - blkid + 1; 333263397Sdelphij trunc = B_TRUE; 334263397Sdelphij } 335168404Spjd 336168404Spjd /* There are no indirect blocks in the object */ 337168404Spjd if (dnlevel == 1) { 338168404Spjd if (blkid >= dn->dn_phys->dn_nblkptr) { 339168404Spjd /* this range was never made persistent */ 340168404Spjd return; 341168404Spjd } 342168404Spjd ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); 343263397Sdelphij free_blocks(dn, bp + blkid, nblks, tx); 344263397Sdelphij } else { 345263397Sdelphij int shift = (dnlevel - 1) * 346263397Sdelphij (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); 347263397Sdelphij int start = blkid >> shift; 348263397Sdelphij int end = (blkid + nblks - 1) >> shift; 349263397Sdelphij dmu_buf_impl_t *db; 350168404Spjd 351263397Sdelphij ASSERT(start < dn->dn_phys->dn_nblkptr); 352263397Sdelphij bp += start; 353263397Sdelphij for (int i = start; i <= end; i++, bp++) { 354263397Sdelphij if (BP_IS_HOLE(bp)) 355263397Sdelphij continue; 356263397Sdelphij rw_enter(&dn->dn_struct_rwlock, RW_READER); 357263397Sdelphij VERIFY0(dbuf_hold_impl(dn, dnlevel - 1, i, 358263397Sdelphij TRUE, FTAG, &db)); 359263397Sdelphij rw_exit(&dn->dn_struct_rwlock); 360168404Spjd 361263397Sdelphij free_children(db, blkid, nblks, tx); 362263397Sdelphij dbuf_rele(db, FTAG); 363168404Spjd } 364168404Spjd } 365263397Sdelphij 366168404Spjd if (trunc) { 367263397Sdelphij dn->dn_phys->dn_maxblkid = blkid == 0 ? 0 : blkid - 1; 368263397Sdelphij 369168404Spjd uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * 370168404Spjd (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); 371168404Spjd ASSERT(off < dn->dn_phys->dn_maxblkid || 372168404Spjd dn->dn_phys->dn_maxblkid == 0 || 373185029Spjd dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); 374168404Spjd } 375168404Spjd} 376168404Spjd 377265740Sdelphijtypedef struct dnode_sync_free_range_arg { 378265740Sdelphij dnode_t *dsfra_dnode; 379265740Sdelphij dmu_tx_t *dsfra_tx; 380265740Sdelphij} dnode_sync_free_range_arg_t; 381265740Sdelphij 382265740Sdelphijstatic void 383265740Sdelphijdnode_sync_free_range(void *arg, uint64_t blkid, uint64_t nblks) 384265740Sdelphij{ 385265740Sdelphij dnode_sync_free_range_arg_t *dsfra = arg; 386265740Sdelphij dnode_t *dn = dsfra->dsfra_dnode; 387265740Sdelphij 388265740Sdelphij mutex_exit(&dn->dn_mtx); 389265740Sdelphij dnode_sync_free_range_impl(dn, blkid, nblks, dsfra->dsfra_tx); 390265740Sdelphij mutex_enter(&dn->dn_mtx); 391265740Sdelphij} 392265740Sdelphij 393168404Spjd/* 394251631Sdelphij * Try to kick all the dnode's dbufs out of the cache... 395168404Spjd */ 396185029Spjdvoid 397185029Spjddnode_evict_dbufs(dnode_t *dn) 398168404Spjd{ 399168404Spjd int progress; 400168404Spjd int pass = 0; 401168404Spjd 402168404Spjd do { 403269845Sdelphij dmu_buf_impl_t *db, *db_next; 404168404Spjd int evicting = FALSE; 405168404Spjd 406168404Spjd progress = FALSE; 407168404Spjd mutex_enter(&dn->dn_dbufs_mtx); 408269845Sdelphij for (db = avl_first(&dn->dn_dbufs); db != NULL; db = db_next) { 409269845Sdelphij db_next = AVL_NEXT(&dn->dn_dbufs, db); 410219089Spjd#ifdef DEBUG 411219089Spjd DB_DNODE_ENTER(db); 412219089Spjd ASSERT3P(DB_DNODE(db), ==, dn); 413219089Spjd DB_DNODE_EXIT(db); 414219089Spjd#endif /* DEBUG */ 415168404Spjd 416168404Spjd mutex_enter(&db->db_mtx); 417168404Spjd if (db->db_state == DB_EVICTING) { 418168404Spjd progress = TRUE; 419168404Spjd evicting = TRUE; 420168404Spjd mutex_exit(&db->db_mtx); 421168404Spjd } else if (refcount_is_zero(&db->db_holds)) { 422168404Spjd progress = TRUE; 423168404Spjd dbuf_clear(db); /* exits db_mtx for us */ 424168404Spjd } else { 425168404Spjd mutex_exit(&db->db_mtx); 426168404Spjd } 427168404Spjd 428168404Spjd } 429168404Spjd /* 430168404Spjd * NB: we need to drop dn_dbufs_mtx between passes so 431168404Spjd * that any DB_EVICTING dbufs can make progress. 432168404Spjd * Ideally, we would have some cv we could wait on, but 433168404Spjd * since we don't, just wait a bit to give the other 434168404Spjd * thread a chance to run. 435168404Spjd */ 436168404Spjd mutex_exit(&dn->dn_dbufs_mtx); 437168404Spjd if (evicting) 438168404Spjd delay(1); 439168404Spjd pass++; 440168404Spjd ASSERT(pass < 100); /* sanity check */ 441168404Spjd } while (progress); 442168404Spjd 443168404Spjd rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 444168404Spjd if (dn->dn_bonus && refcount_is_zero(&dn->dn_bonus->db_holds)) { 445168404Spjd mutex_enter(&dn->dn_bonus->db_mtx); 446168404Spjd dbuf_evict(dn->dn_bonus); 447168404Spjd dn->dn_bonus = NULL; 448168404Spjd } 449168404Spjd rw_exit(&dn->dn_struct_rwlock); 450168404Spjd} 451168404Spjd 452168404Spjdstatic void 453168404Spjddnode_undirty_dbufs(list_t *list) 454168404Spjd{ 455168404Spjd dbuf_dirty_record_t *dr; 456168404Spjd 457168404Spjd while (dr = list_head(list)) { 458168404Spjd dmu_buf_impl_t *db = dr->dr_dbuf; 459168404Spjd uint64_t txg = dr->dr_txg; 460168404Spjd 461219089Spjd if (db->db_level != 0) 462219089Spjd dnode_undirty_dbufs(&dr->dt.di.dr_children); 463219089Spjd 464168404Spjd mutex_enter(&db->db_mtx); 465168404Spjd /* XXX - use dbuf_undirty()? */ 466168404Spjd list_remove(list, dr); 467168404Spjd ASSERT(db->db_last_dirty == dr); 468168404Spjd db->db_last_dirty = NULL; 469168404Spjd db->db_dirtycnt -= 1; 470168404Spjd if (db->db_level == 0) { 471219089Spjd ASSERT(db->db_blkid == DMU_BONUS_BLKID || 472168404Spjd dr->dt.dl.dr_data == db->db_buf); 473168404Spjd dbuf_unoverride(dr); 474168404Spjd } else { 475269218Sdelphij mutex_destroy(&dr->dt.di.dr_mtx); 476169325Spjd list_destroy(&dr->dt.di.dr_children); 477168404Spjd } 478168404Spjd kmem_free(dr, sizeof (dbuf_dirty_record_t)); 479219089Spjd dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); 480168404Spjd } 481168404Spjd} 482168404Spjd 483168404Spjdstatic void 484168404Spjddnode_sync_free(dnode_t *dn, dmu_tx_t *tx) 485168404Spjd{ 486168404Spjd int txgoff = tx->tx_txg & TXG_MASK; 487168404Spjd 488168404Spjd ASSERT(dmu_tx_is_syncing(tx)); 489168404Spjd 490185029Spjd /* 491185029Spjd * Our contents should have been freed in dnode_sync() by the 492185029Spjd * free range record inserted by the caller of dnode_free(). 493185029Spjd */ 494240415Smm ASSERT0(DN_USED_BYTES(dn->dn_phys)); 495185029Spjd ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); 496185029Spjd 497168404Spjd dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); 498185029Spjd dnode_evict_dbufs(dn); 499269845Sdelphij ASSERT(avl_is_empty(&dn->dn_dbufs)); 500248571Smm ASSERT3P(dn->dn_bonus, ==, NULL); 501168404Spjd 502168404Spjd /* 503168404Spjd * XXX - It would be nice to assert this, but we may still 504168404Spjd * have residual holds from async evictions from the arc... 505168404Spjd * 506168404Spjd * zfs_obj_to_path() also depends on this being 507168404Spjd * commented out. 508168404Spjd * 509168404Spjd * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); 510168404Spjd */ 511168404Spjd 512168404Spjd /* Undirty next bits */ 513168404Spjd dn->dn_next_nlevels[txgoff] = 0; 514168404Spjd dn->dn_next_indblkshift[txgoff] = 0; 515168404Spjd dn->dn_next_blksz[txgoff] = 0; 516168404Spjd 517168404Spjd /* ASSERT(blkptrs are zero); */ 518168404Spjd ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 519168404Spjd ASSERT(dn->dn_type != DMU_OT_NONE); 520168404Spjd 521168404Spjd ASSERT(dn->dn_free_txg > 0); 522168404Spjd if (dn->dn_allocated_txg != dn->dn_free_txg) 523263397Sdelphij dmu_buf_will_dirty(&dn->dn_dbuf->db, tx); 524168404Spjd bzero(dn->dn_phys, sizeof (dnode_phys_t)); 525168404Spjd 526168404Spjd mutex_enter(&dn->dn_mtx); 527168404Spjd dn->dn_type = DMU_OT_NONE; 528168404Spjd dn->dn_maxblkid = 0; 529168404Spjd dn->dn_allocated_txg = 0; 530185029Spjd dn->dn_free_txg = 0; 531219089Spjd dn->dn_have_spill = B_FALSE; 532168404Spjd mutex_exit(&dn->dn_mtx); 533168404Spjd 534168404Spjd ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 535168404Spjd 536168404Spjd dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 537168404Spjd /* 538168404Spjd * Now that we've released our hold, the dnode may 539168404Spjd * be evicted, so we musn't access it. 540168404Spjd */ 541168404Spjd} 542168404Spjd 543168404Spjd/* 544168404Spjd * Write out the dnode's dirty buffers. 545168404Spjd */ 546168404Spjdvoid 547168404Spjddnode_sync(dnode_t *dn, dmu_tx_t *tx) 548168404Spjd{ 549168404Spjd dnode_phys_t *dnp = dn->dn_phys; 550168404Spjd int txgoff = tx->tx_txg & TXG_MASK; 551168404Spjd list_t *list = &dn->dn_dirty_records[txgoff]; 552209962Smm static const dnode_phys_t zerodn = { 0 }; 553219089Spjd boolean_t kill_spill = B_FALSE; 554168404Spjd 555168404Spjd ASSERT(dmu_tx_is_syncing(tx)); 556168404Spjd ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); 557209962Smm ASSERT(dnp->dn_type != DMU_OT_NONE || 558209962Smm bcmp(dnp, &zerodn, DNODE_SIZE) == 0); 559168404Spjd DNODE_VERIFY(dn); 560168404Spjd 561168404Spjd ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); 562168404Spjd 563209962Smm if (dmu_objset_userused_enabled(dn->dn_objset) && 564209962Smm !DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 565219089Spjd mutex_enter(&dn->dn_mtx); 566219089Spjd dn->dn_oldused = DN_USED_BYTES(dn->dn_phys); 567219089Spjd dn->dn_oldflags = dn->dn_phys->dn_flags; 568209962Smm dn->dn_phys->dn_flags |= DNODE_FLAG_USERUSED_ACCOUNTED; 569219089Spjd mutex_exit(&dn->dn_mtx); 570219089Spjd dmu_objset_userquota_get_ids(dn, B_FALSE, tx); 571209962Smm } else { 572209962Smm /* Once we account for it, we should always account for it. */ 573209962Smm ASSERT(!(dn->dn_phys->dn_flags & 574209962Smm DNODE_FLAG_USERUSED_ACCOUNTED)); 575209962Smm } 576209962Smm 577168404Spjd mutex_enter(&dn->dn_mtx); 578168404Spjd if (dn->dn_allocated_txg == tx->tx_txg) { 579168404Spjd /* The dnode is newly allocated or reallocated */ 580168404Spjd if (dnp->dn_type == DMU_OT_NONE) { 581168404Spjd /* this is a first alloc, not a realloc */ 582168404Spjd dnp->dn_nlevels = 1; 583196703Spjd dnp->dn_nblkptr = dn->dn_nblkptr; 584168404Spjd } 585168404Spjd 586168404Spjd dnp->dn_type = dn->dn_type; 587168404Spjd dnp->dn_bonustype = dn->dn_bonustype; 588168404Spjd dnp->dn_bonuslen = dn->dn_bonuslen; 589168404Spjd } 590168404Spjd ASSERT(dnp->dn_nlevels > 1 || 591168404Spjd BP_IS_HOLE(&dnp->dn_blkptr[0]) || 592268649Sdelphij BP_IS_EMBEDDED(&dnp->dn_blkptr[0]) || 593168404Spjd BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 594168404Spjd dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 595268649Sdelphij ASSERT(dnp->dn_nlevels < 2 || 596268649Sdelphij BP_IS_HOLE(&dnp->dn_blkptr[0]) || 597268649Sdelphij BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 1 << dnp->dn_indblkshift); 598168404Spjd 599263390Sdelphij if (dn->dn_next_type[txgoff] != 0) { 600263390Sdelphij dnp->dn_type = dn->dn_type; 601263390Sdelphij dn->dn_next_type[txgoff] = 0; 602263390Sdelphij } 603263390Sdelphij 604263390Sdelphij if (dn->dn_next_blksz[txgoff] != 0) { 605168404Spjd ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], 606168404Spjd SPA_MINBLOCKSIZE) == 0); 607168404Spjd ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || 608185029Spjd dn->dn_maxblkid == 0 || list_head(list) != NULL || 609168404Spjd dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == 610265740Sdelphij dnp->dn_datablkszsec || 611265740Sdelphij range_tree_space(dn->dn_free_ranges[txgoff]) != 0); 612168404Spjd dnp->dn_datablkszsec = 613168404Spjd dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; 614168404Spjd dn->dn_next_blksz[txgoff] = 0; 615168404Spjd } 616168404Spjd 617263390Sdelphij if (dn->dn_next_bonuslen[txgoff] != 0) { 618185029Spjd if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) 619185029Spjd dnp->dn_bonuslen = 0; 620185029Spjd else 621185029Spjd dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; 622185029Spjd ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN); 623185029Spjd dn->dn_next_bonuslen[txgoff] = 0; 624185029Spjd } 625185029Spjd 626263390Sdelphij if (dn->dn_next_bonustype[txgoff] != 0) { 627236884Smm ASSERT(DMU_OT_IS_VALID(dn->dn_next_bonustype[txgoff])); 628219089Spjd dnp->dn_bonustype = dn->dn_next_bonustype[txgoff]; 629219089Spjd dn->dn_next_bonustype[txgoff] = 0; 630219089Spjd } 631219089Spjd 632263397Sdelphij boolean_t freeing_dnode = dn->dn_free_txg > 0 && 633263397Sdelphij dn->dn_free_txg <= tx->tx_txg; 634263397Sdelphij 635219089Spjd /* 636219089Spjd * We will either remove a spill block when a file is being removed 637219089Spjd * or we have been asked to remove it. 638219089Spjd */ 639219089Spjd if (dn->dn_rm_spillblk[txgoff] || 640263397Sdelphij ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) && freeing_dnode)) { 641219089Spjd if ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) 642219089Spjd kill_spill = B_TRUE; 643219089Spjd dn->dn_rm_spillblk[txgoff] = 0; 644219089Spjd } 645219089Spjd 646263390Sdelphij if (dn->dn_next_indblkshift[txgoff] != 0) { 647168404Spjd ASSERT(dnp->dn_nlevels == 1); 648168404Spjd dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; 649168404Spjd dn->dn_next_indblkshift[txgoff] = 0; 650168404Spjd } 651168404Spjd 652168404Spjd /* 653168404Spjd * Just take the live (open-context) values for checksum and compress. 654168404Spjd * Strictly speaking it's a future leak, but nothing bad happens if we 655168404Spjd * start using the new checksum or compress algorithm a little early. 656168404Spjd */ 657168404Spjd dnp->dn_checksum = dn->dn_checksum; 658168404Spjd dnp->dn_compress = dn->dn_compress; 659168404Spjd 660168404Spjd mutex_exit(&dn->dn_mtx); 661168404Spjd 662219089Spjd if (kill_spill) { 663263397Sdelphij free_blocks(dn, &dn->dn_phys->dn_spill, 1, tx); 664219089Spjd mutex_enter(&dn->dn_mtx); 665219089Spjd dnp->dn_flags &= ~DNODE_FLAG_SPILL_BLKPTR; 666219089Spjd mutex_exit(&dn->dn_mtx); 667219089Spjd } 668219089Spjd 669168404Spjd /* process all the "freed" ranges in the file */ 670265740Sdelphij if (dn->dn_free_ranges[txgoff] != NULL) { 671265740Sdelphij dnode_sync_free_range_arg_t dsfra; 672265740Sdelphij dsfra.dsfra_dnode = dn; 673265740Sdelphij dsfra.dsfra_tx = tx; 674185029Spjd mutex_enter(&dn->dn_mtx); 675265740Sdelphij range_tree_vacate(dn->dn_free_ranges[txgoff], 676265740Sdelphij dnode_sync_free_range, &dsfra); 677265740Sdelphij range_tree_destroy(dn->dn_free_ranges[txgoff]); 678265740Sdelphij dn->dn_free_ranges[txgoff] = NULL; 679185029Spjd mutex_exit(&dn->dn_mtx); 680168404Spjd } 681168404Spjd 682263397Sdelphij if (freeing_dnode) { 683168404Spjd dnode_sync_free(dn, tx); 684168404Spjd return; 685168404Spjd } 686168404Spjd 687271392Sdelphij if (dn->dn_next_nlevels[txgoff]) { 688271392Sdelphij dnode_increase_indirection(dn, tx); 689271392Sdelphij dn->dn_next_nlevels[txgoff] = 0; 690271392Sdelphij } 691271392Sdelphij 692196703Spjd if (dn->dn_next_nblkptr[txgoff]) { 693196703Spjd /* this should only happen on a realloc */ 694196703Spjd ASSERT(dn->dn_allocated_txg == tx->tx_txg); 695196703Spjd if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { 696196703Spjd /* zero the new blkptrs we are gaining */ 697196703Spjd bzero(dnp->dn_blkptr + dnp->dn_nblkptr, 698196703Spjd sizeof (blkptr_t) * 699196703Spjd (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); 700196703Spjd#ifdef ZFS_DEBUG 701196703Spjd } else { 702196703Spjd int i; 703196703Spjd ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); 704196703Spjd /* the blkptrs we are losing better be unallocated */ 705196703Spjd for (i = dn->dn_next_nblkptr[txgoff]; 706196703Spjd i < dnp->dn_nblkptr; i++) 707196703Spjd ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); 708196703Spjd#endif 709196703Spjd } 710196703Spjd mutex_enter(&dn->dn_mtx); 711196703Spjd dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; 712196703Spjd dn->dn_next_nblkptr[txgoff] = 0; 713196703Spjd mutex_exit(&dn->dn_mtx); 714196703Spjd } 715196703Spjd 716168404Spjd dbuf_sync_list(list, tx); 717168404Spjd 718209962Smm if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 719168404Spjd ASSERT3P(list_head(list), ==, NULL); 720168404Spjd dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 721168404Spjd } 722168404Spjd 723168404Spjd /* 724168404Spjd * Although we have dropped our reference to the dnode, it 725168404Spjd * can't be evicted until its written, and we haven't yet 726168404Spjd * initiated the IO for the dnode's dbuf. 727168404Spjd */ 728168404Spjd} 729