module/zfs/dmu_object.c

   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, 2015 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 #include <sys/dsl_dataset.h>
  34
  35 uint64_t
  36 dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
  37     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
  38 {
  39         return dmu_object_alloc_dnsize(os, ot, blocksize, bonustype, bonuslen,
  40             0, tx);
  41 }
  42
  43 uint64_t
  44 dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot, int blocksize,
  45     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
  46 {
  47         uint64_t object;
  48         uint64_t L1_dnode_count = DNODES_PER_BLOCK <<
  49             (DMU_META_DNODE(os)->dn_indblkshift - SPA_BLKPTRSHIFT);
  50         dnode_t *dn = NULL;
  51         int dn_slots = dnodesize >> DNODE_SHIFT;
  52         boolean_t restarted = B_FALSE;
  53
  54         if (dn_slots == 0) {
  55                 dn_slots = DNODE_MIN_SLOTS;
  56         } else {
  57                 ASSERT3S(dn_slots, >=, DNODE_MIN_SLOTS);
  58                 ASSERT3S(dn_slots, <=, DNODE_MAX_SLOTS);
  59         }
  60
  61         mutex_enter(&os->os_obj_lock);
  62         for (;;) {
  63                 object = os->os_obj_next;
  64                 /*
  65                  * Each time we polish off a L1 bp worth of dnodes (2^12
  66                  * objects), move to another L1 bp that's still
  67                  * reasonably sparse (at most 1/4 full). Look from the
  68                  * beginning at most once per txg. If we still can't
  69                  * allocate from that L1 block, search for an empty L0
  70                  * block, which will quickly skip to the end of the
  71                  * metadnode if the no nearby L0 blocks are empty. This
  72                  * fallback avoids a pathology where full dnode blocks
  73                  * containing large dnodes appear sparse because they
  74                  * have a low blk_fill, leading to many failed
  75                  * allocation attempts. In the long term a better
  76                  * mechanism to search for sparse metadnode regions,
  77                  * such as spacemaps, could be implemented.
  78                  *
  79                  * os_scan_dnodes is set during txg sync if enough objects
  80                  * have been freed since the previous rescan to justify
  81                  * backfilling again.
  82                  *
  83                  * Note that dmu_traverse depends on the behavior that we use
  84                  * multiple blocks of the dnode object before going back to
  85                  * reuse objects.  Any change to this algorithm should preserve
  86                  * that property or find another solution to the issues
  87                  * described in traverse_visitbp.
  88                  */
  89                 if (P2PHASE(object, L1_dnode_count) == 0) {
  90                         uint64_t offset;
  91                         uint64_t blkfill;
  92                         int minlvl;
  93                         int error;
  94                         if (os->os_rescan_dnodes) {
  95                                 offset = 0;
  96                                 os->os_rescan_dnodes = B_FALSE;
  97                         } else {
  98                                 offset = object << DNODE_SHIFT;
  99                         }
 100                         blkfill = restarted ? 1 : DNODES_PER_BLOCK >> 2;
 101                         minlvl = restarted ? 1 : 2;
 102                         restarted = B_TRUE;
 103                         error = dnode_next_offset(DMU_META_DNODE(os),
 104                             DNODE_FIND_HOLE, &offset, minlvl, blkfill, 0);
 105                         if (error == 0)
 106                                 object = offset >> DNODE_SHIFT;
 107                 }
 108                 os->os_obj_next = object + dn_slots;
 109
 110                 /*
 111                  * XXX We should check for an i/o error here and return
 112                  * up to our caller.  Actually we should pre-read it in
 113                  * dmu_tx_assign(), but there is currently no mechanism
 114                  * to do so.
 115                  */
 116                 (void) dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, dn_slots,
 117                     FTAG, &dn);
 118                 if (dn)
 119                         break;
 120
 121                 if (dmu_object_next(os, &object, B_TRUE, 0) == 0)
 122                         os->os_obj_next = object;
 123                 else
 124                         /*
 125                          * Skip to next known valid starting point for a dnode.
 126                          */
 127                         os->os_obj_next = P2ROUNDUP(object + 1,
 128                             DNODES_PER_BLOCK);
 129         }
 130
 131         dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, dn_slots, tx);
 132         dnode_rele(dn, FTAG);
 133
 134         mutex_exit(&os->os_obj_lock);
 135
 136         dmu_tx_add_new_object(tx, os, object);
 137         return (object);
 138 }
 139
 140 int
 141 dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
 142     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 143 {
 144         return (dmu_object_claim_dnsize(os, object, ot, blocksize, bonustype,
 145             bonuslen, 0, tx));
 146 }
 147
 148 int
 149 dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
 150     int blocksize, dmu_object_type_t bonustype, int bonuslen,
 151     int dnodesize, dmu_tx_t *tx)
 152 {
 153         dnode_t *dn;
 154         int dn_slots = dnodesize >> DNODE_SHIFT;
 155         int err;
 156
 157         if (dn_slots == 0)
 158                 dn_slots = DNODE_MIN_SLOTS;
 159         ASSERT3S(dn_slots, >=, DNODE_MIN_SLOTS);
 160         ASSERT3S(dn_slots, <=, DNODE_MAX_SLOTS);
 161
 162         if (object == DMU_META_DNODE_OBJECT && !dmu_tx_private_ok(tx))
 163                 return (SET_ERROR(EBADF));
 164
 165         err = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, dn_slots,
 166             FTAG, &dn);
 167         if (err)
 168                 return (err);
 169
 170         dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, dn_slots, tx);
 171         dnode_rele(dn, FTAG);
 172
 173         dmu_tx_add_new_object(tx, os, object);
 174         return (0);
 175 }
 176
 177 int
 178 dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
 179     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 180 {
 181         return (dmu_object_reclaim_dnsize(os, object, ot, blocksize, bonustype,
 182             bonuslen, 0, tx));
 183 }
 184
 185 int
 186 dmu_object_reclaim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
 187     int blocksize, dmu_object_type_t bonustype, int bonuslen, int dnodesize,
 188     dmu_tx_t *tx)
 189 {
 190         dnode_t *dn;
 191         int dn_slots = dnodesize >> DNODE_SHIFT;
 192         int err;
 193
 194         if (object == DMU_META_DNODE_OBJECT)
 195                 return (SET_ERROR(EBADF));
 196
 197         err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0,
 198             FTAG, &dn);
 199         if (err)
 200                 return (err);
 201
 202         dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, dn_slots, tx);
 203
 204         dnode_rele(dn, FTAG);
 205         return (err);
 206 }
 207
 208
 209 int
 210 dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
 211 {
 212         dnode_t *dn;
 213         int err;
 214
 215         ASSERT(object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx));
 216
 217         err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0,
 218             FTAG, &dn);
 219         if (err)
 220                 return (err);
 221
 222         ASSERT(dn->dn_type != DMU_OT_NONE);
 223         dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
 224         dnode_free(dn, tx);
 225         dnode_rele(dn, FTAG);
 226
 227         return (0);
 228 }
 229
 230 /*
 231  * Return (in *objectp) the next object which is allocated (or a hole)
 232  * after *object, taking into account only objects that may have been modified
 233  * after the specified txg.
 234  */
 235 int
 236 dmu_object_next(objset_t *os, uint64_t *objectp, boolean_t hole, uint64_t txg)
 237 {
 238         uint64_t offset;
 239         dmu_object_info_t doi;
 240         struct dsl_dataset *ds = os->os_dsl_dataset;
 241         int dnodesize;
 242         int error;
 243
 244         /*
 245          * Avoid expensive dnode hold if this dataset doesn't use large dnodes.
 246          */
 247         if (ds && ds->ds_feature_inuse[SPA_FEATURE_LARGE_DNODE]) {
 248                 error = dmu_object_info(os, *objectp, &doi);
 249                 if (error && !(error == EINVAL && *objectp == 0))
 250                         return (SET_ERROR(error));
 251                 else
 252                         dnodesize = doi.doi_dnodesize;
 253         } else {
 254                 dnodesize = DNODE_MIN_SIZE;
 255         }
 256
 257         if (*objectp == 0)
 258                 offset = 1 << DNODE_SHIFT;
 259         else
 260                 offset = (*objectp << DNODE_SHIFT) + dnodesize;
 261
 262         error = dnode_next_offset(DMU_META_DNODE(os),
 263             (hole ? DNODE_FIND_HOLE : 0), &offset, 0, DNODES_PER_BLOCK, txg);
 264
 265         *objectp = offset >> DNODE_SHIFT;
 266
 267         return (error);
 268 }
 269
 270 /*
 271  * Turn this object from old_type into DMU_OTN_ZAP_METADATA, and bump the
 272  * refcount on SPA_FEATURE_EXTENSIBLE_DATASET.
 273  *
 274  * Only for use from syncing context, on MOS objects.
 275  */
 276 void
 277 dmu_object_zapify(objset_t *mos, uint64_t object, dmu_object_type_t old_type,
 278     dmu_tx_t *tx)
 279 {
 280         dnode_t *dn;
 281
 282         ASSERT(dmu_tx_is_syncing(tx));
 283
 284         VERIFY0(dnode_hold(mos, object, FTAG, &dn));
 285         if (dn->dn_type == DMU_OTN_ZAP_METADATA) {
 286                 dnode_rele(dn, FTAG);
 287                 return;
 288         }
 289         ASSERT3U(dn->dn_type, ==, old_type);
 290         ASSERT0(dn->dn_maxblkid);
 291         dn->dn_next_type[tx->tx_txg & TXG_MASK] = dn->dn_type =
 292             DMU_OTN_ZAP_METADATA;
 293         dnode_setdirty(dn, tx);
 294         dnode_rele(dn, FTAG);
 295
 296         mzap_create_impl(mos, object, 0, 0, tx);
 297
 298         spa_feature_incr(dmu_objset_spa(mos),
 299             SPA_FEATURE_EXTENSIBLE_DATASET, tx);
 300 }
 301
 302 void
 303 dmu_object_free_zapified(objset_t *mos, uint64_t object, dmu_tx_t *tx)
 304 {
 305         dnode_t *dn;
 306         dmu_object_type_t t;
 307
 308         ASSERT(dmu_tx_is_syncing(tx));
 309
 310         VERIFY0(dnode_hold(mos, object, FTAG, &dn));
 311         t = dn->dn_type;
 312         dnode_rele(dn, FTAG);
 313
 314         if (t == DMU_OTN_ZAP_METADATA) {
 315                 spa_feature_decr(dmu_objset_spa(mos),
 316                     SPA_FEATURE_EXTENSIBLE_DATASET, tx);
 317         }
 318         VERIFY0(dmu_object_free(mos, object, tx));
 319 }
 320
 321 #if defined(_KERNEL) && defined(HAVE_SPL)
 322 EXPORT_SYMBOL(dmu_object_alloc);
 323 EXPORT_SYMBOL(dmu_object_alloc_dnsize);
 324 EXPORT_SYMBOL(dmu_object_claim);
 325 EXPORT_SYMBOL(dmu_object_claim_dnsize);
 326 EXPORT_SYMBOL(dmu_object_reclaim);
 327 EXPORT_SYMBOL(dmu_object_reclaim_dnsize);
 328 EXPORT_SYMBOL(dmu_object_free);
 329 EXPORT_SYMBOL(dmu_object_next);
 330 EXPORT_SYMBOL(dmu_object_zapify);
 331 EXPORT_SYMBOL(dmu_object_free_zapified);
 332 #endif