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 */ 21168404Spjd/* 22219089Spjd * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23249195Smm * Copyright (c) 2013 by Delphix. All rights reserved. 24168404Spjd */ 25168404Spjd 26168404Spjd#include <sys/dmu.h> 27168404Spjd#include <sys/dmu_objset.h> 28168404Spjd#include <sys/dmu_tx.h> 29168404Spjd#include <sys/dnode.h> 30263390Sdelphij#include <sys/zap.h> 31263390Sdelphij#include <sys/zfeature.h> 32168404Spjd 33168404Spjduint64_t 34168404Spjddmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize, 35168404Spjd dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 36168404Spjd{ 37168404Spjd uint64_t object; 38168404Spjd uint64_t L2_dnode_count = DNODES_PER_BLOCK << 39219089Spjd (DMU_META_DNODE(os)->dn_indblkshift - SPA_BLKPTRSHIFT); 40168404Spjd dnode_t *dn = NULL; 41168404Spjd int restarted = B_FALSE; 42168404Spjd 43219089Spjd mutex_enter(&os->os_obj_lock); 44168404Spjd for (;;) { 45219089Spjd object = os->os_obj_next; 46168404Spjd /* 47168404Spjd * Each time we polish off an L2 bp worth of dnodes 48168404Spjd * (2^13 objects), move to another L2 bp that's still 49168404Spjd * reasonably sparse (at most 1/4 full). Look from the 50168404Spjd * beginning once, but after that keep looking from here. 51168404Spjd * If we can't find one, just keep going from here. 52168404Spjd */ 53168404Spjd if (P2PHASE(object, L2_dnode_count) == 0) { 54168404Spjd uint64_t offset = restarted ? object << DNODE_SHIFT : 0; 55219089Spjd int error = dnode_next_offset(DMU_META_DNODE(os), 56185029Spjd DNODE_FIND_HOLE, 57185029Spjd &offset, 2, DNODES_PER_BLOCK >> 2, 0); 58168404Spjd restarted = B_TRUE; 59168404Spjd if (error == 0) 60168404Spjd object = offset >> DNODE_SHIFT; 61168404Spjd } 62219089Spjd os->os_obj_next = ++object; 63168404Spjd 64168404Spjd /* 65168404Spjd * XXX We should check for an i/o error here and return 66168404Spjd * up to our caller. Actually we should pre-read it in 67168404Spjd * dmu_tx_assign(), but there is currently no mechanism 68168404Spjd * to do so. 69168404Spjd */ 70219089Spjd (void) dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, 71168404Spjd FTAG, &dn); 72168404Spjd if (dn) 73168404Spjd break; 74168404Spjd 75168404Spjd if (dmu_object_next(os, &object, B_TRUE, 0) == 0) 76219089Spjd os->os_obj_next = object - 1; 77168404Spjd } 78168404Spjd 79168404Spjd dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx); 80168404Spjd dnode_rele(dn, FTAG); 81168404Spjd 82219089Spjd mutex_exit(&os->os_obj_lock); 83168404Spjd 84168404Spjd dmu_tx_add_new_object(tx, os, object); 85168404Spjd return (object); 86168404Spjd} 87168404Spjd 88168404Spjdint 89168404Spjddmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot, 90168404Spjd int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 91168404Spjd{ 92168404Spjd dnode_t *dn; 93168404Spjd int err; 94168404Spjd 95168404Spjd if (object == DMU_META_DNODE_OBJECT && !dmu_tx_private_ok(tx)) 96249195Smm return (SET_ERROR(EBADF)); 97168404Spjd 98219089Spjd err = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, FTAG, &dn); 99168404Spjd if (err) 100168404Spjd return (err); 101168404Spjd dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx); 102168404Spjd dnode_rele(dn, FTAG); 103168404Spjd 104168404Spjd dmu_tx_add_new_object(tx, os, object); 105168404Spjd return (0); 106168404Spjd} 107168404Spjd 108168404Spjdint 109168404Spjddmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot, 110200726Sdelphij int blocksize, dmu_object_type_t bonustype, int bonuslen) 111168404Spjd{ 112168404Spjd dnode_t *dn; 113200726Sdelphij dmu_tx_t *tx; 114200726Sdelphij int nblkptr; 115168404Spjd int err; 116168404Spjd 117200726Sdelphij if (object == DMU_META_DNODE_OBJECT) 118249195Smm return (SET_ERROR(EBADF)); 119168404Spjd 120219089Spjd err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 121168404Spjd FTAG, &dn); 122168404Spjd if (err) 123168404Spjd return (err); 124200726Sdelphij 125200726Sdelphij if (dn->dn_type == ot && dn->dn_datablksz == blocksize && 126200726Sdelphij dn->dn_bonustype == bonustype && dn->dn_bonuslen == bonuslen) { 127200726Sdelphij /* nothing is changing, this is a noop */ 128200726Sdelphij dnode_rele(dn, FTAG); 129200726Sdelphij return (0); 130200726Sdelphij } 131200726Sdelphij 132219089Spjd if (bonustype == DMU_OT_SA) { 133219089Spjd nblkptr = 1; 134219089Spjd } else { 135219089Spjd nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); 136219089Spjd } 137200726Sdelphij 138200726Sdelphij /* 139200726Sdelphij * If we are losing blkptrs or changing the block size this must 140200726Sdelphij * be a new file instance. We must clear out the previous file 141200726Sdelphij * contents before we can change this type of metadata in the dnode. 142200726Sdelphij */ 143207624Smm if (dn->dn_nblkptr > nblkptr || dn->dn_datablksz != blocksize) { 144207624Smm err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); 145207624Smm if (err) 146207624Smm goto out; 147207624Smm } 148200726Sdelphij 149207624Smm tx = dmu_tx_create(os); 150207624Smm dmu_tx_hold_bonus(tx, object); 151207624Smm err = dmu_tx_assign(tx, TXG_WAIT); 152207624Smm if (err) { 153207624Smm dmu_tx_abort(tx); 154207624Smm goto out; 155207624Smm } 156207624Smm 157168404Spjd dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, tx); 158200726Sdelphij 159200726Sdelphij dmu_tx_commit(tx); 160207624Smmout: 161168404Spjd dnode_rele(dn, FTAG); 162168404Spjd 163207624Smm return (err); 164168404Spjd} 165168404Spjd 166168404Spjdint 167168404Spjddmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx) 168168404Spjd{ 169168404Spjd dnode_t *dn; 170168404Spjd int err; 171168404Spjd 172168404Spjd ASSERT(object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); 173168404Spjd 174219089Spjd err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 175168404Spjd FTAG, &dn); 176168404Spjd if (err) 177168404Spjd return (err); 178168404Spjd 179168404Spjd ASSERT(dn->dn_type != DMU_OT_NONE); 180185029Spjd dnode_free_range(dn, 0, DMU_OBJECT_END, tx); 181168404Spjd dnode_free(dn, tx); 182168404Spjd dnode_rele(dn, FTAG); 183168404Spjd 184168404Spjd return (0); 185168404Spjd} 186168404Spjd 187168404Spjdint 188168404Spjddmu_object_next(objset_t *os, uint64_t *objectp, boolean_t hole, uint64_t txg) 189168404Spjd{ 190168404Spjd uint64_t offset = (*objectp + 1) << DNODE_SHIFT; 191168404Spjd int error; 192168404Spjd 193219089Spjd error = dnode_next_offset(DMU_META_DNODE(os), 194185029Spjd (hole ? DNODE_FIND_HOLE : 0), &offset, 0, DNODES_PER_BLOCK, txg); 195168404Spjd 196168404Spjd *objectp = offset >> DNODE_SHIFT; 197168404Spjd 198168404Spjd return (error); 199168404Spjd} 200263390Sdelphij 201263390Sdelphij/* 202263390Sdelphij * Turn this object from old_type into DMU_OTN_ZAP_METADATA, and bump the 203263390Sdelphij * refcount on SPA_FEATURE_EXTENSIBLE_DATASET. 204263390Sdelphij * 205263390Sdelphij * Only for use from syncing context, on MOS objects. 206263390Sdelphij */ 207263390Sdelphijvoid 208263390Sdelphijdmu_object_zapify(objset_t *mos, uint64_t object, dmu_object_type_t old_type, 209263390Sdelphij dmu_tx_t *tx) 210263390Sdelphij{ 211263390Sdelphij dnode_t *dn; 212263390Sdelphij 213263390Sdelphij ASSERT(dmu_tx_is_syncing(tx)); 214263390Sdelphij 215263390Sdelphij VERIFY0(dnode_hold(mos, object, FTAG, &dn)); 216263390Sdelphij if (dn->dn_type == DMU_OTN_ZAP_METADATA) { 217263390Sdelphij dnode_rele(dn, FTAG); 218263390Sdelphij return; 219263390Sdelphij } 220263390Sdelphij ASSERT3U(dn->dn_type, ==, old_type); 221263390Sdelphij ASSERT0(dn->dn_maxblkid); 222263390Sdelphij dn->dn_next_type[tx->tx_txg & TXG_MASK] = dn->dn_type = 223263390Sdelphij DMU_OTN_ZAP_METADATA; 224263390Sdelphij dnode_setdirty(dn, tx); 225263390Sdelphij dnode_rele(dn, FTAG); 226263390Sdelphij 227263390Sdelphij mzap_create_impl(mos, object, 0, 0, tx); 228263390Sdelphij 229263390Sdelphij spa_feature_incr(dmu_objset_spa(mos), 230263390Sdelphij SPA_FEATURE_EXTENSIBLE_DATASET, tx); 231263390Sdelphij} 232263390Sdelphij 233263390Sdelphijvoid 234263390Sdelphijdmu_object_free_zapified(objset_t *mos, uint64_t object, dmu_tx_t *tx) 235263390Sdelphij{ 236263390Sdelphij dnode_t *dn; 237263390Sdelphij dmu_object_type_t t; 238263390Sdelphij 239263390Sdelphij ASSERT(dmu_tx_is_syncing(tx)); 240263390Sdelphij 241263390Sdelphij VERIFY0(dnode_hold(mos, object, FTAG, &dn)); 242263390Sdelphij t = dn->dn_type; 243263390Sdelphij dnode_rele(dn, FTAG); 244263390Sdelphij 245263390Sdelphij if (t == DMU_OTN_ZAP_METADATA) { 246263390Sdelphij spa_feature_decr(dmu_objset_spa(mos), 247263390Sdelphij SPA_FEATURE_EXTENSIBLE_DATASET, tx); 248263390Sdelphij } 249263390Sdelphij VERIFY0(dmu_object_free(mos, object, tx)); 250263390Sdelphij} 251