dmu_object.c revision 249195
1219820Sjeff/* 2219820Sjeff * CDDL HEADER START 3219820Sjeff * 4219820Sjeff * The contents of this file are subject to the terms of the 5219820Sjeff * Common Development and Distribution License (the "License"). 6219820Sjeff * You may not use this file except in compliance with the License. 7219820Sjeff * 8219820Sjeff * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9219820Sjeff * or http://www.opensolaris.org/os/licensing. 10219820Sjeff * See the License for the specific language governing permissions 11219820Sjeff * and limitations under the License. 12219820Sjeff * 13219820Sjeff * When distributing Covered Code, include this CDDL HEADER in each 14219820Sjeff * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15219820Sjeff * If applicable, add the following below this CDDL HEADER, with the 16219820Sjeff * fields enclosed by brackets "[]" replaced with your own identifying 17219820Sjeff * information: Portions Copyright [yyyy] [name of copyright owner] 18219820Sjeff * 19219820Sjeff * CDDL HEADER END 20219820Sjeff */ 21219820Sjeff/* 22219820Sjeff * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23219820Sjeff * Copyright (c) 2013 by Delphix. All rights reserved. 24219820Sjeff */ 25219820Sjeff 26219820Sjeff#include <sys/dmu.h> 27219820Sjeff#include <sys/dmu_objset.h> 28219820Sjeff#include <sys/dmu_tx.h> 29219820Sjeff#include <sys/dnode.h> 30219820Sjeff 31219820Sjeffuint64_t 32219820Sjeffdmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize, 33219820Sjeff dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 34219820Sjeff{ 35219820Sjeff uint64_t object; 36219820Sjeff uint64_t L2_dnode_count = DNODES_PER_BLOCK << 37219820Sjeff (DMU_META_DNODE(os)->dn_indblkshift - SPA_BLKPTRSHIFT); 38219820Sjeff dnode_t *dn = NULL; 39219820Sjeff int restarted = B_FALSE; 40219820Sjeff 41219820Sjeff mutex_enter(&os->os_obj_lock); 42219820Sjeff for (;;) { 43219820Sjeff object = os->os_obj_next; 44219820Sjeff /* 45219820Sjeff * Each time we polish off an L2 bp worth of dnodes 46219820Sjeff * (2^13 objects), move to another L2 bp that's still 47219820Sjeff * reasonably sparse (at most 1/4 full). Look from the 48219820Sjeff * beginning once, but after that keep looking from here. 49219820Sjeff * If we can't find one, just keep going from here. 50219820Sjeff */ 51219820Sjeff if (P2PHASE(object, L2_dnode_count) == 0) { 52219820Sjeff uint64_t offset = restarted ? object << DNODE_SHIFT : 0; 53219820Sjeff int error = dnode_next_offset(DMU_META_DNODE(os), 54219820Sjeff DNODE_FIND_HOLE, 55219820Sjeff &offset, 2, DNODES_PER_BLOCK >> 2, 0); 56219820Sjeff restarted = B_TRUE; 57219820Sjeff if (error == 0) 58219820Sjeff object = offset >> DNODE_SHIFT; 59219820Sjeff } 60219820Sjeff os->os_obj_next = ++object; 61219820Sjeff 62219820Sjeff /* 63219820Sjeff * XXX We should check for an i/o error here and return 64219820Sjeff * up to our caller. Actually we should pre-read it in 65219820Sjeff * dmu_tx_assign(), but there is currently no mechanism 66219820Sjeff * to do so. 67219820Sjeff */ 68219820Sjeff (void) dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, 69219820Sjeff FTAG, &dn); 70219820Sjeff if (dn) 71219820Sjeff break; 72219820Sjeff 73219820Sjeff if (dmu_object_next(os, &object, B_TRUE, 0) == 0) 74219820Sjeff os->os_obj_next = object - 1; 75219820Sjeff } 76219820Sjeff 77219820Sjeff dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx); 78219820Sjeff dnode_rele(dn, FTAG); 79219820Sjeff 80219820Sjeff mutex_exit(&os->os_obj_lock); 81219820Sjeff 82219820Sjeff dmu_tx_add_new_object(tx, os, object); 83219820Sjeff return (object); 84219820Sjeff} 85219820Sjeff 86219820Sjeffint 87219820Sjeffdmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot, 88219820Sjeff int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 89219820Sjeff{ 90219820Sjeff dnode_t *dn; 91219820Sjeff int err; 92219820Sjeff 93219820Sjeff if (object == DMU_META_DNODE_OBJECT && !dmu_tx_private_ok(tx)) 94219820Sjeff return (SET_ERROR(EBADF)); 95219820Sjeff 96219820Sjeff err = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, FTAG, &dn); 97219820Sjeff if (err) 98219820Sjeff return (err); 99219820Sjeff dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx); 100219820Sjeff dnode_rele(dn, FTAG); 101219820Sjeff 102219820Sjeff dmu_tx_add_new_object(tx, os, object); 103219820Sjeff return (0); 104219820Sjeff} 105219820Sjeff 106219820Sjeffint 107219820Sjeffdmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot, 108219820Sjeff int blocksize, dmu_object_type_t bonustype, int bonuslen) 109219820Sjeff{ 110219820Sjeff dnode_t *dn; 111219820Sjeff dmu_tx_t *tx; 112219820Sjeff int nblkptr; 113219820Sjeff int err; 114219820Sjeff 115219820Sjeff if (object == DMU_META_DNODE_OBJECT) 116219820Sjeff return (SET_ERROR(EBADF)); 117219820Sjeff 118219820Sjeff err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 119219820Sjeff FTAG, &dn); 120219820Sjeff if (err) 121219820Sjeff return (err); 122219820Sjeff 123219820Sjeff if (dn->dn_type == ot && dn->dn_datablksz == blocksize && 124219820Sjeff dn->dn_bonustype == bonustype && dn->dn_bonuslen == bonuslen) { 125219820Sjeff /* nothing is changing, this is a noop */ 126219820Sjeff dnode_rele(dn, FTAG); 127219820Sjeff return (0); 128219820Sjeff } 129219820Sjeff 130219820Sjeff if (bonustype == DMU_OT_SA) { 131219820Sjeff nblkptr = 1; 132219820Sjeff } else { 133219820Sjeff nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 134219820Sjeff } 135219820Sjeff 136219820Sjeff /* 137219820Sjeff * If we are losing blkptrs or changing the block size this must 138219820Sjeff * be a new file instance. We must clear out the previous file 139219820Sjeff * contents before we can change this type of metadata in the dnode. 140219820Sjeff */ 141219820Sjeff if (dn->dn_nblkptr > nblkptr || dn->dn_datablksz != blocksize) { 142219820Sjeff err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); 143219820Sjeff if (err) 144219820Sjeff goto out; 145219820Sjeff } 146219820Sjeff 147219820Sjeff tx = dmu_tx_create(os); 148219820Sjeff dmu_tx_hold_bonus(tx, object); 149219820Sjeff err = dmu_tx_assign(tx, TXG_WAIT); 150219820Sjeff if (err) { 151219820Sjeff dmu_tx_abort(tx); 152219820Sjeff goto out; 153219820Sjeff } 154219820Sjeff 155219820Sjeff dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, tx); 156219820Sjeff 157219820Sjeff dmu_tx_commit(tx); 158219820Sjeffout: 159219820Sjeff dnode_rele(dn, FTAG); 160219820Sjeff 161219820Sjeff return (err); 162219820Sjeff} 163219820Sjeff 164219820Sjeffint 165219820Sjeffdmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx) 166219820Sjeff{ 167219820Sjeff dnode_t *dn; 168219820Sjeff int err; 169219820Sjeff 170219820Sjeff ASSERT(object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); 171219820Sjeff 172219820Sjeff err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 173219820Sjeff FTAG, &dn); 174219820Sjeff if (err) 175219820Sjeff return (err); 176219820Sjeff 177219820Sjeff ASSERT(dn->dn_type != DMU_OT_NONE); 178219820Sjeff dnode_free_range(dn, 0, DMU_OBJECT_END, tx); 179219820Sjeff dnode_free(dn, tx); 180219820Sjeff dnode_rele(dn, FTAG); 181219820Sjeff 182219820Sjeff return (0); 183219820Sjeff} 184219820Sjeff 185219820Sjeffint 186219820Sjeffdmu_object_next(objset_t *os, uint64_t *objectp, boolean_t hole, uint64_t txg) 187219820Sjeff{ 188219820Sjeff uint64_t offset = (*objectp + 1) << DNODE_SHIFT; 189219820Sjeff int error; 190219820Sjeff 191219820Sjeff error = dnode_next_offset(DMU_META_DNODE(os), 192219820Sjeff (hole ? DNODE_FIND_HOLE : 0), &offset, 0, DNODES_PER_BLOCK, txg); 193219820Sjeff 194219820Sjeff *objectp = offset >> DNODE_SHIFT; 195219820Sjeff 196219820Sjeff return (error); 197219820Sjeff} 198219820Sjeff