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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2013, 2017 by Delphix. All rights reserved. 24 * Copyright 2014 HybridCluster. All rights reserved. 25 */ 26 27 #include <sys/dmu.h> 28 #include <sys/dmu_objset.h> 29 #include <sys/dmu_tx.h> 30 #include <sys/dnode.h> 31 #include <sys/zap.h> 32 #include <sys/zfeature.h> 33 34 uint64_t 35 dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize, 36 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 37 { 38 uint64_t object; 39 uint64_t L1_dnode_count = DNODES_PER_BLOCK << 40 (DMU_META_DNODE(os)->dn_indblkshift - SPA_BLKPTRSHIFT); 41 dnode_t *dn = NULL; 42 43 mutex_enter(&os->os_obj_lock); 44 for (;;) { 45 object = os->os_obj_next; 46 /* 47 * Each time we polish off a L1 bp worth of dnodes (2^12 48 * objects), move to another L1 bp that's still reasonably 49 * sparse (at most 1/4 full). Look from the beginning at most 50 * once per txg, but after that keep looking from here. 51 * os_scan_dnodes is set during txg sync if enough objects 52 * have been freed since the previous rescan to justify 53 * backfilling again. If we can't find a suitable block, just 54 * keep going from here. 55 * 56 * Note that dmu_traverse depends on the behavior that we use 57 * multiple blocks of the dnode object before going back to 58 * reuse objects. Any change to this algorithm should preserve 59 * that property or find another solution to the issues 60 * described in traverse_visitbp. 61 */ 62 63 if (P2PHASE(object, L1_dnode_count) == 0) { 64 uint64_t offset; 65 int error; 66 if (os->os_rescan_dnodes) { 67 offset = 0; 68 os->os_rescan_dnodes = B_FALSE; 69 } else { 70 offset = object << DNODE_SHIFT; 71 } 72 error = dnode_next_offset(DMU_META_DNODE(os), 73 DNODE_FIND_HOLE, 74 &offset, 2, DNODES_PER_BLOCK >> 2, 0); 75 if (error == 0) 76 object = offset >> DNODE_SHIFT; 77 } 78 os->os_obj_next = ++object; 79 80 /* 81 * XXX We should check for an i/o error here and return 82 * up to our caller. Actually we should pre-read it in 83 * dmu_tx_assign(), but there is currently no mechanism 84 * to do so. 85 */ 86 (void) dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, 87 FTAG, &dn); 88 if (dn) 89 break; 90 91 if (dmu_object_next(os, &object, B_TRUE, 0) == 0) 92 os->os_obj_next = object - 1; 93 } 94 95 dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx); 96 mutex_exit(&os->os_obj_lock); 97 98 dmu_tx_add_new_object(tx, dn); 99 dnode_rele(dn, FTAG); 100 101 return (object); 102 } 103 104 int 105 dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot, 106 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 107 { 108 dnode_t *dn; 109 int err; 110 111 if (object == DMU_META_DNODE_OBJECT && !dmu_tx_private_ok(tx)) 112 return (SET_ERROR(EBADF)); 113 114 err = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, FTAG, &dn); 115 if (err) 116 return (err); 117 dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx); 118 dmu_tx_add_new_object(tx, dn); 119 120 dnode_rele(dn, FTAG); 121 122 return (0); 123 } 124 125 int 126 dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot, 127 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) 128 { 129 dnode_t *dn; 130 int err; 131 132 if (object == DMU_META_DNODE_OBJECT) 133 return (SET_ERROR(EBADF)); 134 135 err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 136 FTAG, &dn); 137 if (err) 138 return (err); 139 140 dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, tx); 141 142 dnode_rele(dn, FTAG); 143 return (err); 144 } 145 146 int 147 dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx) 148 { 149 dnode_t *dn; 150 int err; 151 152 ASSERT(object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); 153 154 err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 155 FTAG, &dn); 156 if (err) 157 return (err); 158 159 ASSERT(dn->dn_type != DMU_OT_NONE); 160 dnode_free_range(dn, 0, DMU_OBJECT_END, tx); 161 dnode_free(dn, tx); 162 dnode_rele(dn, FTAG); 163 164 return (0); 165 } 166 167 /* 168 * Return (in *objectp) the next object which is allocated (or a hole) 169 * after *object, taking into account only objects that may have been modified 170 * after the specified txg. 171 */ 172 int 173 dmu_object_next(objset_t *os, uint64_t *objectp, boolean_t hole, uint64_t txg) 174 { 175 uint64_t offset = (*objectp + 1) << DNODE_SHIFT; 176 int error; 177 178 error = dnode_next_offset(DMU_META_DNODE(os), 179 (hole ? DNODE_FIND_HOLE : 0), &offset, 0, DNODES_PER_BLOCK, txg); 180 181 *objectp = offset >> DNODE_SHIFT; 182 183 return (error); 184 } 185 186 /* 187 * Turn this object from old_type into DMU_OTN_ZAP_METADATA, and bump the 188 * refcount on SPA_FEATURE_EXTENSIBLE_DATASET. 189 * 190 * Only for use from syncing context, on MOS objects. 191 */ 192 void 193 dmu_object_zapify(objset_t *mos, uint64_t object, dmu_object_type_t old_type, 194 dmu_tx_t *tx) 195 { 196 dnode_t *dn; 197 198 ASSERT(dmu_tx_is_syncing(tx)); 199 200 VERIFY0(dnode_hold(mos, object, FTAG, &dn)); 201 if (dn->dn_type == DMU_OTN_ZAP_METADATA) { 202 dnode_rele(dn, FTAG); 203 return; 204 } 205 ASSERT3U(dn->dn_type, ==, old_type); 206 ASSERT0(dn->dn_maxblkid); 207 208 /* 209 * We must initialize the ZAP data before changing the type, 210 * so that concurrent calls to *_is_zapified() can determine if 211 * the object has been completely zapified by checking the type. 212 */ 213 mzap_create_impl(mos, object, 0, 0, tx); 214 215 dn->dn_next_type[tx->tx_txg & TXG_MASK] = dn->dn_type = 216 DMU_OTN_ZAP_METADATA; 217 dnode_setdirty(dn, tx); 218 dnode_rele(dn, FTAG); 219 220 spa_feature_incr(dmu_objset_spa(mos), 221 SPA_FEATURE_EXTENSIBLE_DATASET, tx); 222 } 223 224 void 225 dmu_object_free_zapified(objset_t *mos, uint64_t object, dmu_tx_t *tx) 226 { 227 dnode_t *dn; 228 dmu_object_type_t t; 229 230 ASSERT(dmu_tx_is_syncing(tx)); 231 232 VERIFY0(dnode_hold(mos, object, FTAG, &dn)); 233 t = dn->dn_type; 234 dnode_rele(dn, FTAG); 235 236 if (t == DMU_OTN_ZAP_METADATA) { 237 spa_feature_decr(dmu_objset_spa(mos), 238 SPA_FEATURE_EXTENSIBLE_DATASET, tx); 239 } 240 VERIFY0(dmu_object_free(mos, object, tx)); 241 } 242