1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21ad135b5dSChristopher Siden 22fa9e4066Sahrens /* 2306e0070dSMark Shellenbaum * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24bf16b11eSMatthew Ahrens * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 25bc9014e6SJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 26fa9e4066Sahrens */ 27fa9e4066Sahrens 28fa9e4066Sahrens #include <sys/zfs_context.h> 29fa9e4066Sahrens #include <sys/dbuf.h> 30fa9e4066Sahrens #include <sys/dnode.h> 31fa9e4066Sahrens #include <sys/dmu.h> 32fa9e4066Sahrens #include <sys/dmu_tx.h> 33fa9e4066Sahrens #include <sys/dmu_objset.h> 34fa9e4066Sahrens #include <sys/dsl_dataset.h> 35fa9e4066Sahrens #include <sys/spa.h> 36bf16b11eSMatthew Ahrens #include <sys/range_tree.h> 3743466aaeSMax Grossman #include <sys/zfeature.h> 38fa9e4066Sahrens 39fa9e4066Sahrens static void 40fa9e4066Sahrens dnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) 41fa9e4066Sahrens { 42fa9e4066Sahrens dmu_buf_impl_t *db; 43c717a561Smaybee int txgoff = tx->tx_txg & TXG_MASK; 44c717a561Smaybee int nblkptr = dn->dn_phys->dn_nblkptr; 45c717a561Smaybee int old_toplvl = dn->dn_phys->dn_nlevels - 1; 46c717a561Smaybee int new_level = dn->dn_next_nlevels[txgoff]; 47fa9e4066Sahrens int i; 48fa9e4066Sahrens 49c717a561Smaybee rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 50c717a561Smaybee 51c717a561Smaybee /* this dnode can't be paged out because it's dirty */ 52fa9e4066Sahrens ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 53fa9e4066Sahrens ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); 54c717a561Smaybee ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); 55fa9e4066Sahrens 56fa9e4066Sahrens db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); 57ea8dc4b6Seschrock ASSERT(db != NULL); 58c717a561Smaybee 59c717a561Smaybee dn->dn_phys->dn_nlevels = new_level; 60f82bfe17Sgw dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, 61f82bfe17Sgw dn->dn_object, dn->dn_phys->dn_nlevels); 62c717a561Smaybee 63c717a561Smaybee /* check for existing blkptrs in the dnode */ 64c717a561Smaybee for (i = 0; i < nblkptr; i++) 65fa9e4066Sahrens if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[i])) 66fa9e4066Sahrens break; 67c717a561Smaybee if (i != nblkptr) { 68c717a561Smaybee /* transfer dnode's block pointers to new indirect block */ 69c717a561Smaybee (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); 70c717a561Smaybee ASSERT(db->db.db_data); 71c717a561Smaybee ASSERT(arc_released(db->db_buf)); 72c717a561Smaybee ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); 73fa9e4066Sahrens bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, 74c717a561Smaybee sizeof (blkptr_t) * nblkptr); 756b4acc8bSahrens arc_buf_freeze(db->db_buf); 76fa9e4066Sahrens } 77fa9e4066Sahrens 78fa9e4066Sahrens /* set dbuf's parent pointers to new indirect buf */ 79c717a561Smaybee for (i = 0; i < nblkptr; i++) { 80e57a022bSJustin T. Gibbs dmu_buf_impl_t *child = 81e57a022bSJustin T. Gibbs dbuf_find(dn->dn_objset, dn->dn_object, old_toplvl, i); 82c717a561Smaybee 83fa9e4066Sahrens if (child == NULL) 84fa9e4066Sahrens continue; 85744947dcSTom Erickson #ifdef DEBUG 86744947dcSTom Erickson DB_DNODE_ENTER(child); 87744947dcSTom Erickson ASSERT3P(DB_DNODE(child), ==, dn); 88744947dcSTom Erickson DB_DNODE_EXIT(child); 89744947dcSTom Erickson #endif /* DEBUG */ 90c717a561Smaybee if (child->db_parent && child->db_parent != dn->dn_dbuf) { 91c717a561Smaybee ASSERT(child->db_parent->db_level == db->db_level); 92c717a561Smaybee ASSERT(child->db_blkptr != 93c717a561Smaybee &dn->dn_phys->dn_blkptr[child->db_blkid]); 94fa9e4066Sahrens mutex_exit(&child->db_mtx); 95fa9e4066Sahrens continue; 96fa9e4066Sahrens } 97c717a561Smaybee ASSERT(child->db_parent == NULL || 98c717a561Smaybee child->db_parent == dn->dn_dbuf); 99c717a561Smaybee 100c717a561Smaybee child->db_parent = db; 101c717a561Smaybee dbuf_add_ref(db, child); 102c717a561Smaybee if (db->db.db_data) 103c717a561Smaybee child->db_blkptr = (blkptr_t *)db->db.db_data + i; 104c717a561Smaybee else 105c717a561Smaybee child->db_blkptr = NULL; 106c717a561Smaybee dprintf_dbuf_bp(child, child->db_blkptr, 107c717a561Smaybee "changed db_blkptr to new indirect %s", ""); 108fa9e4066Sahrens 109fa9e4066Sahrens mutex_exit(&child->db_mtx); 110fa9e4066Sahrens } 111fa9e4066Sahrens 112c717a561Smaybee bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); 113fa9e4066Sahrens 114ea8dc4b6Seschrock dbuf_rele(db, FTAG); 115c717a561Smaybee 116c717a561Smaybee rw_exit(&dn->dn_struct_rwlock); 117fa9e4066Sahrens } 118fa9e4066Sahrens 11943466aaeSMax Grossman static void 120fa9e4066Sahrens free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) 121fa9e4066Sahrens { 122cdb0ab79Smaybee dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; 123fa9e4066Sahrens uint64_t bytesfreed = 0; 124fa9e4066Sahrens 125cdb0ab79Smaybee dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); 126fa9e4066Sahrens 12743466aaeSMax Grossman for (int i = 0; i < num; i++, bp++) { 128fa9e4066Sahrens if (BP_IS_HOLE(bp)) 129fa9e4066Sahrens continue; 130fa9e4066Sahrens 131b24ab676SJeff Bonwick bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE); 13299653d4eSeschrock ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); 13343466aaeSMax Grossman 13443466aaeSMax Grossman /* 13543466aaeSMax Grossman * Save some useful information on the holes being 13643466aaeSMax Grossman * punched, including logical size, type, and indirection 13743466aaeSMax Grossman * level. Retaining birth time enables detection of when 13843466aaeSMax Grossman * holes are punched for reducing the number of free 13943466aaeSMax Grossman * records transmitted during a zfs send. 14043466aaeSMax Grossman */ 14143466aaeSMax Grossman 14243466aaeSMax Grossman uint64_t lsize = BP_GET_LSIZE(bp); 14343466aaeSMax Grossman dmu_object_type_t type = BP_GET_TYPE(bp); 14443466aaeSMax Grossman uint64_t lvl = BP_GET_LEVEL(bp); 14543466aaeSMax Grossman 146c717a561Smaybee bzero(bp, sizeof (blkptr_t)); 14743466aaeSMax Grossman 14843466aaeSMax Grossman if (spa_feature_is_active(dn->dn_objset->os_spa, 14943466aaeSMax Grossman SPA_FEATURE_HOLE_BIRTH)) { 15043466aaeSMax Grossman BP_SET_LSIZE(bp, lsize); 15143466aaeSMax Grossman BP_SET_TYPE(bp, type); 15243466aaeSMax Grossman BP_SET_LEVEL(bp, lvl); 15343466aaeSMax Grossman BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0); 15443466aaeSMax Grossman } 155fa9e4066Sahrens } 156fa9e4066Sahrens dnode_diduse_space(dn, -bytesfreed); 157fa9e4066Sahrens } 158fa9e4066Sahrens 1599c9dc39aSek #ifdef ZFS_DEBUG 160fa9e4066Sahrens static void 161fa9e4066Sahrens free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) 162fa9e4066Sahrens { 163fa9e4066Sahrens int off, num; 164fa9e4066Sahrens int i, err, epbs; 165fa9e4066Sahrens uint64_t txg = tx->tx_txg; 166744947dcSTom Erickson dnode_t *dn; 167fa9e4066Sahrens 168744947dcSTom Erickson DB_DNODE_ENTER(db); 169744947dcSTom Erickson dn = DB_DNODE(db); 170744947dcSTom Erickson epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 171fa9e4066Sahrens off = start - (db->db_blkid * 1<<epbs); 172fa9e4066Sahrens num = end - start + 1; 173fa9e4066Sahrens 174fa9e4066Sahrens ASSERT3U(off, >=, 0); 175fa9e4066Sahrens ASSERT3U(num, >=, 0); 176fa9e4066Sahrens ASSERT3U(db->db_level, >, 0); 177744947dcSTom Erickson ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); 178fa9e4066Sahrens ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); 179fa9e4066Sahrens ASSERT(db->db_blkptr != NULL); 180fa9e4066Sahrens 181fa9e4066Sahrens for (i = off; i < off+num; i++) { 182fa9e4066Sahrens uint64_t *buf; 183fa9e4066Sahrens dmu_buf_impl_t *child; 184c717a561Smaybee dbuf_dirty_record_t *dr; 185c717a561Smaybee int j; 186fa9e4066Sahrens 187fa9e4066Sahrens ASSERT(db->db_level == 1); 188fa9e4066Sahrens 189744947dcSTom Erickson rw_enter(&dn->dn_struct_rwlock, RW_READER); 190744947dcSTom Erickson err = dbuf_hold_impl(dn, db->db_level-1, 191f82bfe17Sgw (db->db_blkid << epbs) + i, TRUE, FTAG, &child); 192744947dcSTom Erickson rw_exit(&dn->dn_struct_rwlock); 193fa9e4066Sahrens if (err == ENOENT) 194fa9e4066Sahrens continue; 195fa9e4066Sahrens ASSERT(err == 0); 196fa9e4066Sahrens ASSERT(child->db_level == 0); 197c717a561Smaybee dr = child->db_last_dirty; 198c717a561Smaybee while (dr && dr->dr_txg > txg) 199c717a561Smaybee dr = dr->dr_next; 200c717a561Smaybee ASSERT(dr == NULL || dr->dr_txg == txg); 201c717a561Smaybee 202c717a561Smaybee /* data_old better be zeroed */ 203c717a561Smaybee if (dr) { 204c717a561Smaybee buf = dr->dt.dl.dr_data->b_data; 205fa9e4066Sahrens for (j = 0; j < child->db.db_size >> 3; j++) { 206fa9e4066Sahrens if (buf[j] != 0) { 207fa9e4066Sahrens panic("freed data not zero: " 208fa9e4066Sahrens "child=%p i=%d off=%d num=%d\n", 209903a11ebSrh (void *)child, i, off, num); 210fa9e4066Sahrens } 211fa9e4066Sahrens } 212fa9e4066Sahrens } 213fa9e4066Sahrens 214fa9e4066Sahrens /* 215fa9e4066Sahrens * db_data better be zeroed unless it's dirty in a 216fa9e4066Sahrens * future txg. 217fa9e4066Sahrens */ 218fa9e4066Sahrens mutex_enter(&child->db_mtx); 219fa9e4066Sahrens buf = child->db.db_data; 220fa9e4066Sahrens if (buf != NULL && child->db_state != DB_FILL && 221c717a561Smaybee child->db_last_dirty == NULL) { 222fa9e4066Sahrens for (j = 0; j < child->db.db_size >> 3; j++) { 223fa9e4066Sahrens if (buf[j] != 0) { 224fa9e4066Sahrens panic("freed data not zero: " 225fa9e4066Sahrens "child=%p i=%d off=%d num=%d\n", 226903a11ebSrh (void *)child, i, off, num); 227fa9e4066Sahrens } 228fa9e4066Sahrens } 229fa9e4066Sahrens } 230fa9e4066Sahrens mutex_exit(&child->db_mtx); 231fa9e4066Sahrens 232ea8dc4b6Seschrock dbuf_rele(child, FTAG); 233fa9e4066Sahrens } 234744947dcSTom Erickson DB_DNODE_EXIT(db); 235fa9e4066Sahrens } 2369c9dc39aSek #endif 237fa9e4066Sahrens 23843466aaeSMax Grossman static void 23943466aaeSMax Grossman free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, 240fa9e4066Sahrens dmu_tx_t *tx) 241fa9e4066Sahrens { 242744947dcSTom Erickson dnode_t *dn; 243fa9e4066Sahrens blkptr_t *bp; 244fa9e4066Sahrens dmu_buf_impl_t *subdb; 245fa9e4066Sahrens uint64_t start, end, dbstart, dbend, i; 24643466aaeSMax Grossman int epbs, shift; 247cdb0ab79Smaybee 248cdb0ab79Smaybee /* 249cdb0ab79Smaybee * There is a small possibility that this block will not be cached: 250cdb0ab79Smaybee * 1 - if level > 1 and there are no children with level <= 1 25143466aaeSMax Grossman * 2 - if this block was evicted since we read it from 25243466aaeSMax Grossman * dmu_tx_hold_free(). 253cdb0ab79Smaybee */ 254cdb0ab79Smaybee if (db->db_state != DB_CACHED) 255cdb0ab79Smaybee (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); 256fa9e4066Sahrens 2573f9d6ad7SLin Ling dbuf_release_bp(db); 25843466aaeSMax Grossman bp = db->db.db_data; 259fa9e4066Sahrens 260744947dcSTom Erickson DB_DNODE_ENTER(db); 261744947dcSTom Erickson dn = DB_DNODE(db); 262744947dcSTom Erickson epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; 263fa9e4066Sahrens shift = (db->db_level - 1) * epbs; 264fa9e4066Sahrens dbstart = db->db_blkid << epbs; 265fa9e4066Sahrens start = blkid >> shift; 266fa9e4066Sahrens if (dbstart < start) { 267fa9e4066Sahrens bp += start - dbstart; 268fa9e4066Sahrens } else { 269fa9e4066Sahrens start = dbstart; 270fa9e4066Sahrens } 271fa9e4066Sahrens dbend = ((db->db_blkid + 1) << epbs) - 1; 272fa9e4066Sahrens end = (blkid + nblks - 1) >> shift; 273fa9e4066Sahrens if (dbend <= end) 274fa9e4066Sahrens end = dbend; 27543466aaeSMax Grossman 276fa9e4066Sahrens ASSERT3U(start, <=, end); 277fa9e4066Sahrens 278fa9e4066Sahrens if (db->db_level == 1) { 2799c9dc39aSek FREE_VERIFY(db, start, end, tx); 28043466aaeSMax Grossman free_blocks(dn, bp, end-start+1, tx); 28143466aaeSMax Grossman } else { 28243466aaeSMax Grossman for (i = start; i <= end; i++, bp++) { 28343466aaeSMax Grossman if (BP_IS_HOLE(bp)) 28443466aaeSMax Grossman continue; 28543466aaeSMax Grossman rw_enter(&dn->dn_struct_rwlock, RW_READER); 28643466aaeSMax Grossman VERIFY0(dbuf_hold_impl(dn, db->db_level - 1, 28743466aaeSMax Grossman i, B_TRUE, FTAG, &subdb)); 28843466aaeSMax Grossman rw_exit(&dn->dn_struct_rwlock); 28943466aaeSMax Grossman ASSERT3P(bp, ==, subdb->db_blkptr); 29043466aaeSMax Grossman 29143466aaeSMax Grossman free_children(subdb, blkid, nblks, tx); 29243466aaeSMax Grossman dbuf_rele(subdb, FTAG); 29343466aaeSMax Grossman } 294fa9e4066Sahrens } 295fa9e4066Sahrens 29643466aaeSMax Grossman /* If this whole block is free, free ourself too. */ 29743466aaeSMax Grossman for (i = 0, bp = db->db.db_data; i < 1 << epbs; i++, bp++) { 29843466aaeSMax Grossman if (!BP_IS_HOLE(bp)) 29943466aaeSMax Grossman break; 30043466aaeSMax Grossman } 30143466aaeSMax Grossman if (i == 1 << epbs) { 30243466aaeSMax Grossman /* didn't find any non-holes */ 30343466aaeSMax Grossman bzero(db->db.db_data, db->db.db_size); 30443466aaeSMax Grossman free_blocks(dn, db->db_blkptr, 1, tx); 30543466aaeSMax Grossman } else { 30643466aaeSMax Grossman /* 30743466aaeSMax Grossman * Partial block free; must be marked dirty so that it 30843466aaeSMax Grossman * will be written out. 30943466aaeSMax Grossman */ 31043466aaeSMax Grossman ASSERT(db->db_dirtycnt > 0); 311fa9e4066Sahrens } 31243466aaeSMax Grossman 313744947dcSTom Erickson DB_DNODE_EXIT(db); 3146b4acc8bSahrens arc_buf_freeze(db->db_buf); 315fa9e4066Sahrens } 316fa9e4066Sahrens 317fa9e4066Sahrens /* 318f7170741SWill Andrews * Traverse the indicated range of the provided file 319fa9e4066Sahrens * and "free" all the blocks contained there. 320fa9e4066Sahrens */ 321fa9e4066Sahrens static void 322bf16b11eSMatthew Ahrens dnode_sync_free_range_impl(dnode_t *dn, uint64_t blkid, uint64_t nblks, 32343466aaeSMax Grossman dmu_tx_t *tx) 324fa9e4066Sahrens { 325fa9e4066Sahrens blkptr_t *bp = dn->dn_phys->dn_blkptr; 326fa9e4066Sahrens int dnlevel = dn->dn_phys->dn_nlevels; 32743466aaeSMax Grossman boolean_t trunc = B_FALSE; 328fa9e4066Sahrens 329fa9e4066Sahrens if (blkid > dn->dn_phys->dn_maxblkid) 330fa9e4066Sahrens return; 331fa9e4066Sahrens 332fa9e4066Sahrens ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); 33343466aaeSMax Grossman if (blkid + nblks > dn->dn_phys->dn_maxblkid) { 334fa9e4066Sahrens nblks = dn->dn_phys->dn_maxblkid - blkid + 1; 33543466aaeSMax Grossman trunc = B_TRUE; 33643466aaeSMax Grossman } 337fa9e4066Sahrens 338fa9e4066Sahrens /* There are no indirect blocks in the object */ 339fa9e4066Sahrens if (dnlevel == 1) { 340fa9e4066Sahrens if (blkid >= dn->dn_phys->dn_nblkptr) { 341fa9e4066Sahrens /* this range was never made persistent */ 342fa9e4066Sahrens return; 343fa9e4066Sahrens } 344fa9e4066Sahrens ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); 34543466aaeSMax Grossman free_blocks(dn, bp + blkid, nblks, tx); 34643466aaeSMax Grossman } else { 34743466aaeSMax Grossman int shift = (dnlevel - 1) * 34843466aaeSMax Grossman (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); 34943466aaeSMax Grossman int start = blkid >> shift; 35043466aaeSMax Grossman int end = (blkid + nblks - 1) >> shift; 35143466aaeSMax Grossman dmu_buf_impl_t *db; 35243466aaeSMax Grossman 35343466aaeSMax Grossman ASSERT(start < dn->dn_phys->dn_nblkptr); 35443466aaeSMax Grossman bp += start; 35543466aaeSMax Grossman for (int i = start; i <= end; i++, bp++) { 35643466aaeSMax Grossman if (BP_IS_HOLE(bp)) 35743466aaeSMax Grossman continue; 35843466aaeSMax Grossman rw_enter(&dn->dn_struct_rwlock, RW_READER); 35943466aaeSMax Grossman VERIFY0(dbuf_hold_impl(dn, dnlevel - 1, i, 36043466aaeSMax Grossman TRUE, FTAG, &db)); 36143466aaeSMax Grossman rw_exit(&dn->dn_struct_rwlock); 36243466aaeSMax Grossman 36343466aaeSMax Grossman free_children(db, blkid, nblks, tx); 36443466aaeSMax Grossman dbuf_rele(db, FTAG); 365fa9e4066Sahrens } 366fa9e4066Sahrens } 36743466aaeSMax Grossman 368fa9e4066Sahrens if (trunc) { 36943466aaeSMax Grossman dn->dn_phys->dn_maxblkid = blkid == 0 ? 0 : blkid - 1; 37043466aaeSMax Grossman 371fa9e4066Sahrens uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * 372fa9e4066Sahrens (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); 373fa9e4066Sahrens ASSERT(off < dn->dn_phys->dn_maxblkid || 374fa9e4066Sahrens dn->dn_phys->dn_maxblkid == 0 || 375cdb0ab79Smaybee dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); 376fa9e4066Sahrens } 377fa9e4066Sahrens } 378fa9e4066Sahrens 379bf16b11eSMatthew Ahrens typedef struct dnode_sync_free_range_arg { 380bf16b11eSMatthew Ahrens dnode_t *dsfra_dnode; 381bf16b11eSMatthew Ahrens dmu_tx_t *dsfra_tx; 382bf16b11eSMatthew Ahrens } dnode_sync_free_range_arg_t; 383bf16b11eSMatthew Ahrens 384bf16b11eSMatthew Ahrens static void 385bf16b11eSMatthew Ahrens dnode_sync_free_range(void *arg, uint64_t blkid, uint64_t nblks) 386bf16b11eSMatthew Ahrens { 387bf16b11eSMatthew Ahrens dnode_sync_free_range_arg_t *dsfra = arg; 388bf16b11eSMatthew Ahrens dnode_t *dn = dsfra->dsfra_dnode; 389bf16b11eSMatthew Ahrens 390bf16b11eSMatthew Ahrens mutex_exit(&dn->dn_mtx); 391bf16b11eSMatthew Ahrens dnode_sync_free_range_impl(dn, blkid, nblks, dsfra->dsfra_tx); 392bf16b11eSMatthew Ahrens mutex_enter(&dn->dn_mtx); 393bf16b11eSMatthew Ahrens } 394bf16b11eSMatthew Ahrens 395ea8dc4b6Seschrock /* 396f7170741SWill Andrews * Try to kick all the dnode's dbufs out of the cache... 397ea8dc4b6Seschrock */ 3981934e92fSmaybee void 3991934e92fSmaybee dnode_evict_dbufs(dnode_t *dn) 400ea8dc4b6Seschrock { 401bc9014e6SJustin Gibbs dmu_buf_impl_t db_marker; 402bc9014e6SJustin Gibbs dmu_buf_impl_t *db, *db_next; 403c543ec06Sahrens 404bc9014e6SJustin Gibbs mutex_enter(&dn->dn_dbufs_mtx); 405bc9014e6SJustin Gibbs for (db = avl_first(&dn->dn_dbufs); db != NULL; db = db_next) { 406c543ec06Sahrens 407744947dcSTom Erickson #ifdef DEBUG 408bc9014e6SJustin Gibbs DB_DNODE_ENTER(db); 409bc9014e6SJustin Gibbs ASSERT3P(DB_DNODE(db), ==, dn); 410bc9014e6SJustin Gibbs DB_DNODE_EXIT(db); 411744947dcSTom Erickson #endif /* DEBUG */ 412ea8dc4b6Seschrock 413bc9014e6SJustin Gibbs mutex_enter(&db->db_mtx); 414bc9014e6SJustin Gibbs if (db->db_state != DB_EVICTING && 415bc9014e6SJustin Gibbs refcount_is_zero(&db->db_holds)) { 416bc9014e6SJustin Gibbs db_marker.db_level = db->db_level; 417bc9014e6SJustin Gibbs db_marker.db_blkid = db->db_blkid; 418bc9014e6SJustin Gibbs db_marker.db_state = DB_SEARCH; 419bc9014e6SJustin Gibbs avl_insert_here(&dn->dn_dbufs, &db_marker, db, 420bc9014e6SJustin Gibbs AVL_BEFORE); 421bc9014e6SJustin Gibbs 422bc9014e6SJustin Gibbs dbuf_clear(db); 423bc9014e6SJustin Gibbs 424bc9014e6SJustin Gibbs db_next = AVL_NEXT(&dn->dn_dbufs, &db_marker); 425bc9014e6SJustin Gibbs avl_remove(&dn->dn_dbufs, &db_marker); 426bc9014e6SJustin Gibbs } else { 427bc9014e6SJustin Gibbs mutex_exit(&db->db_mtx); 428bc9014e6SJustin Gibbs db_next = AVL_NEXT(&dn->dn_dbufs, db); 429ea8dc4b6Seschrock } 430bc9014e6SJustin Gibbs } 431bc9014e6SJustin Gibbs mutex_exit(&dn->dn_dbufs_mtx); 432c543ec06Sahrens 433*cd485b49SJustin T. Gibbs dnode_evict_bonus(dn); 434*cd485b49SJustin T. Gibbs } 435*cd485b49SJustin T. Gibbs 436*cd485b49SJustin T. Gibbs void 437*cd485b49SJustin T. Gibbs dnode_evict_bonus(dnode_t *dn) 438*cd485b49SJustin T. Gibbs { 439ea8dc4b6Seschrock rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 440ea8dc4b6Seschrock if (dn->dn_bonus && refcount_is_zero(&dn->dn_bonus->db_holds)) { 441ea8dc4b6Seschrock mutex_enter(&dn->dn_bonus->db_mtx); 442ea8dc4b6Seschrock dbuf_evict(dn->dn_bonus); 443ea8dc4b6Seschrock dn->dn_bonus = NULL; 444ea8dc4b6Seschrock } 445ea8dc4b6Seschrock rw_exit(&dn->dn_struct_rwlock); 446ea8dc4b6Seschrock } 447ea8dc4b6Seschrock 448c717a561Smaybee static void 449c717a561Smaybee dnode_undirty_dbufs(list_t *list) 450fa9e4066Sahrens { 451c717a561Smaybee dbuf_dirty_record_t *dr; 452fa9e4066Sahrens 453c717a561Smaybee while (dr = list_head(list)) { 454c717a561Smaybee dmu_buf_impl_t *db = dr->dr_dbuf; 455c717a561Smaybee uint64_t txg = dr->dr_txg; 456fa9e4066Sahrens 457b24ab676SJeff Bonwick if (db->db_level != 0) 458b24ab676SJeff Bonwick dnode_undirty_dbufs(&dr->dt.di.dr_children); 459b24ab676SJeff Bonwick 460fa9e4066Sahrens mutex_enter(&db->db_mtx); 461fa9e4066Sahrens /* XXX - use dbuf_undirty()? */ 462c717a561Smaybee list_remove(list, dr); 463c717a561Smaybee ASSERT(db->db_last_dirty == dr); 464c717a561Smaybee db->db_last_dirty = NULL; 465c717a561Smaybee db->db_dirtycnt -= 1; 466fa9e4066Sahrens if (db->db_level == 0) { 4670a586ceaSMark Shellenbaum ASSERT(db->db_blkid == DMU_BONUS_BLKID || 468c717a561Smaybee dr->dt.dl.dr_data == db->db_buf); 469c717a561Smaybee dbuf_unoverride(dr); 470d2b3cbbdSJorgen Lundman } else { 471d2b3cbbdSJorgen Lundman mutex_destroy(&dr->dt.di.dr_mtx); 472d2b3cbbdSJorgen Lundman list_destroy(&dr->dt.di.dr_children); 473fa9e4066Sahrens } 474c717a561Smaybee kmem_free(dr, sizeof (dbuf_dirty_record_t)); 475b24ab676SJeff Bonwick dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); 476fa9e4066Sahrens } 477c717a561Smaybee } 478fa9e4066Sahrens 479c717a561Smaybee static void 480c717a561Smaybee dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) 481c717a561Smaybee { 482c717a561Smaybee int txgoff = tx->tx_txg & TXG_MASK; 483c717a561Smaybee 484c717a561Smaybee ASSERT(dmu_tx_is_syncing(tx)); 485c717a561Smaybee 486cdb0ab79Smaybee /* 487cdb0ab79Smaybee * Our contents should have been freed in dnode_sync() by the 488cdb0ab79Smaybee * free range record inserted by the caller of dnode_free(). 489cdb0ab79Smaybee */ 490fb09f5aaSMadhav Suresh ASSERT0(DN_USED_BYTES(dn->dn_phys)); 491cdb0ab79Smaybee ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); 492cdb0ab79Smaybee 493c717a561Smaybee dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); 4941934e92fSmaybee dnode_evict_dbufs(dn); 4950f6d88adSAlex Reece ASSERT(avl_is_empty(&dn->dn_dbufs)); 496ea8dc4b6Seschrock 497ea8dc4b6Seschrock /* 498ea8dc4b6Seschrock * XXX - It would be nice to assert this, but we may still 499ea8dc4b6Seschrock * have residual holds from async evictions from the arc... 500ea8dc4b6Seschrock * 50155434c77Sek * zfs_obj_to_path() also depends on this being 50255434c77Sek * commented out. 50355434c77Sek * 504ea8dc4b6Seschrock * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); 505ea8dc4b6Seschrock */ 506fa9e4066Sahrens 507fa9e4066Sahrens /* Undirty next bits */ 508fa9e4066Sahrens dn->dn_next_nlevels[txgoff] = 0; 509fa9e4066Sahrens dn->dn_next_indblkshift[txgoff] = 0; 510c543ec06Sahrens dn->dn_next_blksz[txgoff] = 0; 511fa9e4066Sahrens 512fa9e4066Sahrens /* ASSERT(blkptrs are zero); */ 513fa9e4066Sahrens ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); 514fa9e4066Sahrens ASSERT(dn->dn_type != DMU_OT_NONE); 515fa9e4066Sahrens 516fa9e4066Sahrens ASSERT(dn->dn_free_txg > 0); 517fa9e4066Sahrens if (dn->dn_allocated_txg != dn->dn_free_txg) 51843466aaeSMax Grossman dmu_buf_will_dirty(&dn->dn_dbuf->db, tx); 519fa9e4066Sahrens bzero(dn->dn_phys, sizeof (dnode_phys_t)); 520fa9e4066Sahrens 521fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 522fa9e4066Sahrens dn->dn_type = DMU_OT_NONE; 523fa9e4066Sahrens dn->dn_maxblkid = 0; 524fa9e4066Sahrens dn->dn_allocated_txg = 0; 525758f6e0bSgw dn->dn_free_txg = 0; 5260a586ceaSMark Shellenbaum dn->dn_have_spill = B_FALSE; 527fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 528fa9e4066Sahrens 529ea8dc4b6Seschrock ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 530fa9e4066Sahrens 531fa9e4066Sahrens dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 532fa9e4066Sahrens /* 533fa9e4066Sahrens * Now that we've released our hold, the dnode may 534fa9e4066Sahrens * be evicted, so we musn't access it. 535fa9e4066Sahrens */ 536fa9e4066Sahrens } 537fa9e4066Sahrens 538fa9e4066Sahrens /* 539c717a561Smaybee * Write out the dnode's dirty buffers. 540fa9e4066Sahrens */ 541c717a561Smaybee void 542c717a561Smaybee dnode_sync(dnode_t *dn, dmu_tx_t *tx) 543fa9e4066Sahrens { 544fa9e4066Sahrens dnode_phys_t *dnp = dn->dn_phys; 545c717a561Smaybee int txgoff = tx->tx_txg & TXG_MASK; 546c717a561Smaybee list_t *list = &dn->dn_dirty_records[txgoff]; 54714843421SMatthew Ahrens static const dnode_phys_t zerodn = { 0 }; 5480a586ceaSMark Shellenbaum boolean_t kill_spill = B_FALSE; 549fa9e4066Sahrens 550fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 551fa9e4066Sahrens ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); 55214843421SMatthew Ahrens ASSERT(dnp->dn_type != DMU_OT_NONE || 55314843421SMatthew Ahrens bcmp(dnp, &zerodn, DNODE_SIZE) == 0); 5549c9dc39aSek DNODE_VERIFY(dn); 555c543ec06Sahrens 556c717a561Smaybee ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); 557fa9e4066Sahrens 55814843421SMatthew Ahrens if (dmu_objset_userused_enabled(dn->dn_objset) && 55914843421SMatthew Ahrens !DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 5600a586ceaSMark Shellenbaum mutex_enter(&dn->dn_mtx); 5610a586ceaSMark Shellenbaum dn->dn_oldused = DN_USED_BYTES(dn->dn_phys); 5620a586ceaSMark Shellenbaum dn->dn_oldflags = dn->dn_phys->dn_flags; 56314843421SMatthew Ahrens dn->dn_phys->dn_flags |= DNODE_FLAG_USERUSED_ACCOUNTED; 5640a586ceaSMark Shellenbaum mutex_exit(&dn->dn_mtx); 56506e0070dSMark Shellenbaum dmu_objset_userquota_get_ids(dn, B_FALSE, tx); 56614843421SMatthew Ahrens } else { 56714843421SMatthew Ahrens /* Once we account for it, we should always account for it. */ 56814843421SMatthew Ahrens ASSERT(!(dn->dn_phys->dn_flags & 56914843421SMatthew Ahrens DNODE_FLAG_USERUSED_ACCOUNTED)); 57014843421SMatthew Ahrens } 57114843421SMatthew Ahrens 572fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 573fa9e4066Sahrens if (dn->dn_allocated_txg == tx->tx_txg) { 574fa9e4066Sahrens /* The dnode is newly allocated or reallocated */ 575fa9e4066Sahrens if (dnp->dn_type == DMU_OT_NONE) { 576fa9e4066Sahrens /* this is a first alloc, not a realloc */ 577fa9e4066Sahrens dnp->dn_nlevels = 1; 578da03de99SMark Maybee dnp->dn_nblkptr = dn->dn_nblkptr; 579fa9e4066Sahrens } 580fa9e4066Sahrens 581fa9e4066Sahrens dnp->dn_type = dn->dn_type; 582fa9e4066Sahrens dnp->dn_bonustype = dn->dn_bonustype; 583fa9e4066Sahrens dnp->dn_bonuslen = dn->dn_bonuslen; 584fa9e4066Sahrens } 585c717a561Smaybee ASSERT(dnp->dn_nlevels > 1 || 586f676ed34Sahrens BP_IS_HOLE(&dnp->dn_blkptr[0]) || 5875d7b4d43SMatthew Ahrens BP_IS_EMBEDDED(&dnp->dn_blkptr[0]) || 588f676ed34Sahrens BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 589f676ed34Sahrens dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 5905d7b4d43SMatthew Ahrens ASSERT(dnp->dn_nlevels < 2 || 5915d7b4d43SMatthew Ahrens BP_IS_HOLE(&dnp->dn_blkptr[0]) || 5925d7b4d43SMatthew Ahrens BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 1 << dnp->dn_indblkshift); 593f676ed34Sahrens 5942acef22dSMatthew Ahrens if (dn->dn_next_type[txgoff] != 0) { 5952acef22dSMatthew Ahrens dnp->dn_type = dn->dn_type; 5962acef22dSMatthew Ahrens dn->dn_next_type[txgoff] = 0; 5972acef22dSMatthew Ahrens } 5982acef22dSMatthew Ahrens 5992acef22dSMatthew Ahrens if (dn->dn_next_blksz[txgoff] != 0) { 600c543ec06Sahrens ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], 601fa9e4066Sahrens SPA_MINBLOCKSIZE) == 0); 602f676ed34Sahrens ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || 603cdb0ab79Smaybee dn->dn_maxblkid == 0 || list_head(list) != NULL || 604347a31bcSahrens dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == 605bf16b11eSMatthew Ahrens dnp->dn_datablkszsec || 606bf16b11eSMatthew Ahrens range_tree_space(dn->dn_free_ranges[txgoff]) != 0); 607fa9e4066Sahrens dnp->dn_datablkszsec = 608c543ec06Sahrens dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; 609c543ec06Sahrens dn->dn_next_blksz[txgoff] = 0; 610fa9e4066Sahrens } 611fa9e4066Sahrens 6122acef22dSMatthew Ahrens if (dn->dn_next_bonuslen[txgoff] != 0) { 6131934e92fSmaybee if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) 6141934e92fSmaybee dnp->dn_bonuslen = 0; 6151934e92fSmaybee else 6161934e92fSmaybee dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; 6171934e92fSmaybee ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN); 6181934e92fSmaybee dn->dn_next_bonuslen[txgoff] = 0; 6191934e92fSmaybee } 6201934e92fSmaybee 6212acef22dSMatthew Ahrens if (dn->dn_next_bonustype[txgoff] != 0) { 622ad135b5dSChristopher Siden ASSERT(DMU_OT_IS_VALID(dn->dn_next_bonustype[txgoff])); 6230a586ceaSMark Shellenbaum dnp->dn_bonustype = dn->dn_next_bonustype[txgoff]; 6240a586ceaSMark Shellenbaum dn->dn_next_bonustype[txgoff] = 0; 6250a586ceaSMark Shellenbaum } 6260a586ceaSMark Shellenbaum 62743466aaeSMax Grossman boolean_t freeing_dnode = dn->dn_free_txg > 0 && 62843466aaeSMax Grossman dn->dn_free_txg <= tx->tx_txg; 62943466aaeSMax Grossman 6300a586ceaSMark Shellenbaum /* 631e6518318SMatthew Ahrens * Remove the spill block if we have been explicitly asked to 632e6518318SMatthew Ahrens * remove it, or if the object is being removed. 6330a586ceaSMark Shellenbaum */ 634e6518318SMatthew Ahrens if (dn->dn_rm_spillblk[txgoff] || freeing_dnode) { 635e6518318SMatthew Ahrens if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) 6360a586ceaSMark Shellenbaum kill_spill = B_TRUE; 6370a586ceaSMark Shellenbaum dn->dn_rm_spillblk[txgoff] = 0; 6380a586ceaSMark Shellenbaum } 6390a586ceaSMark Shellenbaum 6402acef22dSMatthew Ahrens if (dn->dn_next_indblkshift[txgoff] != 0) { 641fa9e4066Sahrens ASSERT(dnp->dn_nlevels == 1); 642fa9e4066Sahrens dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; 643fa9e4066Sahrens dn->dn_next_indblkshift[txgoff] = 0; 644fa9e4066Sahrens } 645fa9e4066Sahrens 646fa9e4066Sahrens /* 647fa9e4066Sahrens * Just take the live (open-context) values for checksum and compress. 648fa9e4066Sahrens * Strictly speaking it's a future leak, but nothing bad happens if we 649fa9e4066Sahrens * start using the new checksum or compress algorithm a little early. 650fa9e4066Sahrens */ 651fa9e4066Sahrens dnp->dn_checksum = dn->dn_checksum; 652fa9e4066Sahrens dnp->dn_compress = dn->dn_compress; 653fa9e4066Sahrens 654fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 655fa9e4066Sahrens 6560a586ceaSMark Shellenbaum if (kill_spill) { 65743466aaeSMax Grossman free_blocks(dn, &dn->dn_phys->dn_spill, 1, tx); 6580a586ceaSMark Shellenbaum mutex_enter(&dn->dn_mtx); 6590a586ceaSMark Shellenbaum dnp->dn_flags &= ~DNODE_FLAG_SPILL_BLKPTR; 6600a586ceaSMark Shellenbaum mutex_exit(&dn->dn_mtx); 6610a586ceaSMark Shellenbaum } 6620a586ceaSMark Shellenbaum 663fa9e4066Sahrens /* process all the "freed" ranges in the file */ 664bf16b11eSMatthew Ahrens if (dn->dn_free_ranges[txgoff] != NULL) { 665bf16b11eSMatthew Ahrens dnode_sync_free_range_arg_t dsfra; 666bf16b11eSMatthew Ahrens dsfra.dsfra_dnode = dn; 667bf16b11eSMatthew Ahrens dsfra.dsfra_tx = tx; 668cdb0ab79Smaybee mutex_enter(&dn->dn_mtx); 669bf16b11eSMatthew Ahrens range_tree_vacate(dn->dn_free_ranges[txgoff], 670bf16b11eSMatthew Ahrens dnode_sync_free_range, &dsfra); 671bf16b11eSMatthew Ahrens range_tree_destroy(dn->dn_free_ranges[txgoff]); 672bf16b11eSMatthew Ahrens dn->dn_free_ranges[txgoff] = NULL; 673cdb0ab79Smaybee mutex_exit(&dn->dn_mtx); 674fa9e4066Sahrens } 6751934e92fSmaybee 67643466aaeSMax Grossman if (freeing_dnode) { 677c717a561Smaybee dnode_sync_free(dn, tx); 678c717a561Smaybee return; 679fa9e4066Sahrens } 680fa9e4066Sahrens 681e503a685SGeorge Wilson if (dn->dn_next_nlevels[txgoff]) { 682e503a685SGeorge Wilson dnode_increase_indirection(dn, tx); 683e503a685SGeorge Wilson dn->dn_next_nlevels[txgoff] = 0; 684e503a685SGeorge Wilson } 685e503a685SGeorge Wilson 686da03de99SMark Maybee if (dn->dn_next_nblkptr[txgoff]) { 687da03de99SMark Maybee /* this should only happen on a realloc */ 688da03de99SMark Maybee ASSERT(dn->dn_allocated_txg == tx->tx_txg); 689da03de99SMark Maybee if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { 690da03de99SMark Maybee /* zero the new blkptrs we are gaining */ 691da03de99SMark Maybee bzero(dnp->dn_blkptr + dnp->dn_nblkptr, 692da03de99SMark Maybee sizeof (blkptr_t) * 693da03de99SMark Maybee (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); 694da03de99SMark Maybee #ifdef ZFS_DEBUG 695da03de99SMark Maybee } else { 696da03de99SMark Maybee int i; 697da03de99SMark Maybee ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); 698da03de99SMark Maybee /* the blkptrs we are losing better be unallocated */ 699da03de99SMark Maybee for (i = dn->dn_next_nblkptr[txgoff]; 700da03de99SMark Maybee i < dnp->dn_nblkptr; i++) 701da03de99SMark Maybee ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); 702da03de99SMark Maybee #endif 703da03de99SMark Maybee } 704da03de99SMark Maybee mutex_enter(&dn->dn_mtx); 705da03de99SMark Maybee dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; 706da03de99SMark Maybee dn->dn_next_nblkptr[txgoff] = 0; 707da03de99SMark Maybee mutex_exit(&dn->dn_mtx); 708da03de99SMark Maybee } 709da03de99SMark Maybee 710c717a561Smaybee dbuf_sync_list(list, tx); 711fa9e4066Sahrens 71214843421SMatthew Ahrens if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { 713c717a561Smaybee ASSERT3P(list_head(list), ==, NULL); 714c717a561Smaybee dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); 715fa9e4066Sahrens } 716c717a561Smaybee 717c717a561Smaybee /* 718c717a561Smaybee * Although we have dropped our reference to the dnode, it 719c717a561Smaybee * can't be evicted until its written, and we haven't yet 720c717a561Smaybee * initiated the IO for the dnode's dbuf. 721c717a561Smaybee */ 722fa9e4066Sahrens } 723