module/zfs/dmu_send.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 2009 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25
  26 #include <sys/dmu.h>
  27 #include <sys/dmu_impl.h>
  28 #include <sys/dmu_tx.h>
  29 #include <sys/dbuf.h>
  30 #include <sys/dnode.h>
  31 #include <sys/zfs_context.h>
  32 #include <sys/dmu_objset.h>
  33 #include <sys/dmu_traverse.h>
  34 #include <sys/dsl_dataset.h>
  35 #include <sys/dsl_dir.h>
  36 #include <sys/dsl_pool.h>
  37 #include <sys/dsl_synctask.h>
  38 #include <sys/zfs_ioctl.h>
  39 #include <sys/zap.h>
  40 #include <sys/zio_checksum.h>
  41
  42 static char *dmu_recv_tag = "dmu_recv_tag";
  43
  44 struct backuparg {
  45         dmu_replay_record_t *drr;
  46         vnode_t *vp;
  47         offset_t *off;
  48         objset_t *os;
  49         zio_cksum_t zc;
  50         int err;
  51 };
  52
  53 static int
  54 dump_bytes(struct backuparg *ba, void *buf, int len)
  55 {
  56         ssize_t resid; /* have to get resid to get detailed errno */
  57         ASSERT3U(len % 8, ==, 0);
  58
  59         fletcher_4_incremental_native(buf, len, &ba->zc);
  60         ba->err = vn_rdwr(UIO_WRITE, ba->vp,
  61             (caddr_t)buf, len,
  62             0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
  63         *ba->off += len;
  64         return (ba->err);
  65 }
  66
  67 static int
  68 dump_free(struct backuparg *ba, uint64_t object, uint64_t offset,
  69     uint64_t length)
  70 {
  71         /* write a FREE record */
  72         bzero(ba->drr, sizeof (dmu_replay_record_t));
  73         ba->drr->drr_type = DRR_FREE;
  74         ba->drr->drr_u.drr_free.drr_object = object;
  75         ba->drr->drr_u.drr_free.drr_offset = offset;
  76         ba->drr->drr_u.drr_free.drr_length = length;
  77
  78         if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
  79                 return (EINTR);
  80         return (0);
  81 }
  82
  83 static int
  84 dump_data(struct backuparg *ba, dmu_object_type_t type,
  85     uint64_t object, uint64_t offset, int blksz, void *data)
  86 {
  87         /* write a DATA record */
  88         bzero(ba->drr, sizeof (dmu_replay_record_t));
  89         ba->drr->drr_type = DRR_WRITE;
  90         ba->drr->drr_u.drr_write.drr_object = object;
  91         ba->drr->drr_u.drr_write.drr_type = type;
  92         ba->drr->drr_u.drr_write.drr_offset = offset;
  93         ba->drr->drr_u.drr_write.drr_length = blksz;
  94
  95         if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
  96                 return (EINTR);
  97         if (dump_bytes(ba, data, blksz))
  98                 return (EINTR);
  99         return (0);
 100 }
 101
 102 static int
 103 dump_freeobjects(struct backuparg *ba, uint64_t firstobj, uint64_t numobjs)
 104 {
 105         /* write a FREEOBJECTS record */
 106         bzero(ba->drr, sizeof (dmu_replay_record_t));
 107         ba->drr->drr_type = DRR_FREEOBJECTS;
 108         ba->drr->drr_u.drr_freeobjects.drr_firstobj = firstobj;
 109         ba->drr->drr_u.drr_freeobjects.drr_numobjs = numobjs;
 110
 111         if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
 112                 return (EINTR);
 113         return (0);
 114 }
 115
 116 static int
 117 dump_dnode(struct backuparg *ba, uint64_t object, dnode_phys_t *dnp)
 118 {
 119         if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
 120                 return (dump_freeobjects(ba, object, 1));
 121
 122         /* write an OBJECT record */
 123         bzero(ba->drr, sizeof (dmu_replay_record_t));
 124         ba->drr->drr_type = DRR_OBJECT;
 125         ba->drr->drr_u.drr_object.drr_object = object;
 126         ba->drr->drr_u.drr_object.drr_type = dnp->dn_type;
 127         ba->drr->drr_u.drr_object.drr_bonustype = dnp->dn_bonustype;
 128         ba->drr->drr_u.drr_object.drr_blksz =
 129             dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
 130         ba->drr->drr_u.drr_object.drr_bonuslen = dnp->dn_bonuslen;
 131         ba->drr->drr_u.drr_object.drr_checksum = dnp->dn_checksum;
 132         ba->drr->drr_u.drr_object.drr_compress = dnp->dn_compress;
 133
 134         if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
 135                 return (EINTR);
 136
 137         if (dump_bytes(ba, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)))
 138                 return (EINTR);
 139
 140         /* free anything past the end of the file */
 141         if (dump_free(ba, object, (dnp->dn_maxblkid + 1) *
 142             (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
 143                 return (EINTR);
 144         if (ba->err)
 145                 return (EINTR);
 146         return (0);
 147 }
 148
 149 #define BP_SPAN(dnp, level) \
 150         (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
 151         (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
 152
 153 static int
 154 backup_cb(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
 155     const dnode_phys_t *dnp, void *arg)
 156 {
 157         struct backuparg *ba = arg;
 158         dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
 159         int err = 0;
 160
 161         if (issig(JUSTLOOKING) && issig(FORREAL))
 162                 return (EINTR);
 163
 164         if (zb->zb_object != 0 && DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
 165                 return (0);
 166         } else if (bp == NULL && zb->zb_object == 0) {
 167                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 168                 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
 169                 err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
 170         } else if (bp == NULL) {
 171                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 172                 err = dump_free(ba, zb->zb_object, zb->zb_blkid * span, span);
 173         } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
 174                 return (0);
 175         } else if (type == DMU_OT_DNODE) {
 176                 dnode_phys_t *blk;
 177                 int i;
 178                 int blksz = BP_GET_LSIZE(bp);
 179                 uint32_t aflags = ARC_WAIT;
 180                 arc_buf_t *abuf;
 181
 182                 if (arc_read_nolock(NULL, spa, bp,
 183                     arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
 184                     ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
 185                         return (EIO);
 186
 187                 blk = abuf->b_data;
 188                 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
 189                         uint64_t dnobj = (zb->zb_blkid <<
 190                             (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
 191                         err = dump_dnode(ba, dnobj, blk+i);
 192                         if (err)
 193                                 break;
 194                 }
 195                 (void) arc_buf_remove_ref(abuf, &abuf);
 196         } else { /* it's a level-0 block of a regular object */
 197                 uint32_t aflags = ARC_WAIT;
 198                 arc_buf_t *abuf;
 199                 int blksz = BP_GET_LSIZE(bp);
 200
 201                 if (arc_read_nolock(NULL, spa, bp,
 202                     arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
 203                     ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
 204                         return (EIO);
 205
 206                 err = dump_data(ba, type, zb->zb_object, zb->zb_blkid * blksz,
 207                     blksz, abuf->b_data);
 208                 (void) arc_buf_remove_ref(abuf, &abuf);
 209         }
 210
 211         ASSERT(err == 0 || err == EINTR);
 212         return (err);
 213 }
 214
 215 int
 216 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
 217     vnode_t *vp, offset_t *off)
 218 {
 219         dsl_dataset_t *ds = tosnap->os->os_dsl_dataset;
 220         dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL;
 221         dmu_replay_record_t *drr;
 222         struct backuparg ba;
 223         int err;
 224         uint64_t fromtxg = 0;
 225
 226         /* tosnap must be a snapshot */
 227         if (ds->ds_phys->ds_next_snap_obj == 0)
 228                 return (EINVAL);
 229
 230         /* fromsnap must be an earlier snapshot from the same fs as tosnap */
 231         if (fromds && (ds->ds_dir != fromds->ds_dir ||
 232             fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
 233                 return (EXDEV);
 234
 235         if (fromorigin) {
 236                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
 237
 238                 if (fromsnap)
 239                         return (EINVAL);
 240
 241                 if (dsl_dir_is_clone(ds->ds_dir)) {
 242                         rw_enter(&dp->dp_config_rwlock, RW_READER);
 243                         err = dsl_dataset_hold_obj(dp,
 244                             ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
 245                         rw_exit(&dp->dp_config_rwlock);
 246                         if (err)
 247                                 return (err);
 248                 } else {
 249                         fromorigin = B_FALSE;
 250                 }
 251         }
 252
 253
 254         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
 255         drr->drr_type = DRR_BEGIN;
 256         drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
 257         drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION;
 258         drr->drr_u.drr_begin.drr_creation_time =
 259             ds->ds_phys->ds_creation_time;
 260         drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type;
 261         if (fromorigin)
 262                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
 263         drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
 264         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 265                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
 266
 267         if (fromds)
 268                 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
 269         dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
 270
 271         if (fromds)
 272                 fromtxg = fromds->ds_phys->ds_creation_txg;
 273         if (fromorigin)
 274                 dsl_dataset_rele(fromds, FTAG);
 275
 276         ba.drr = drr;
 277         ba.vp = vp;
 278         ba.os = tosnap;
 279         ba.off = off;
 280         ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
 281
 282         if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
 283                 kmem_free(drr, sizeof (dmu_replay_record_t));
 284                 return (ba.err);
 285         }
 286
 287         err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
 288             backup_cb, &ba);
 289
 290         if (err) {
 291                 if (err == EINTR && ba.err)
 292                         err = ba.err;
 293                 kmem_free(drr, sizeof (dmu_replay_record_t));
 294                 return (err);
 295         }
 296
 297         bzero(drr, sizeof (dmu_replay_record_t));
 298         drr->drr_type = DRR_END;
 299         drr->drr_u.drr_end.drr_checksum = ba.zc;
 300
 301         if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
 302                 kmem_free(drr, sizeof (dmu_replay_record_t));
 303                 return (ba.err);
 304         }
 305
 306         kmem_free(drr, sizeof (dmu_replay_record_t));
 307
 308         return (0);
 309 }
 310
 311 struct recvbeginsyncarg {
 312         const char *tofs;
 313         const char *tosnap;
 314         dsl_dataset_t *origin;
 315         uint64_t fromguid;
 316         dmu_objset_type_t type;
 317         void *tag;
 318         boolean_t force;
 319         uint64_t dsflags;
 320         char clonelastname[MAXNAMELEN];
 321         dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
 322 };
 323
 324 static dsl_dataset_t *
 325 recv_full_sync_impl(dsl_pool_t *dp, uint64_t dsobj, dmu_objset_type_t type,
 326     cred_t *cr, dmu_tx_t *tx)
 327 {
 328         dsl_dataset_t *ds;
 329
 330         /* This should always work, since we just created it */
 331         /* XXX - create should return an owned ds */
 332         VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
 333             DS_MODE_INCONSISTENT, dmu_recv_tag, &ds));
 334
 335         if (type != DMU_OST_NONE) {
 336                 (void) dmu_objset_create_impl(dp->dp_spa,
 337                     ds, &ds->ds_phys->ds_bp, type, tx);
 338         }
 339
 340         spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC,
 341             dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
 342
 343         return (ds);
 344 }
 345
 346 /* ARGSUSED */
 347 static int
 348 recv_full_check(void *arg1, void *arg2, dmu_tx_t *tx)
 349 {
 350         dsl_dir_t *dd = arg1;
 351         struct recvbeginsyncarg *rbsa = arg2;
 352         objset_t *mos = dd->dd_pool->dp_meta_objset;
 353         uint64_t val;
 354         int err;
 355
 356         err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
 357             strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
 358
 359         if (err != ENOENT)
 360                 return (err ? err : EEXIST);
 361
 362         if (rbsa->origin) {
 363                 /* make sure it's a snap in the same pool */
 364                 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
 365                         return (EXDEV);
 366                 if (rbsa->origin->ds_phys->ds_num_children == 0)
 367                         return (EINVAL);
 368                 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
 369                         return (ENODEV);
 370         }
 371
 372         return (0);
 373 }
 374
 375 static void
 376 recv_full_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 377 {
 378         dsl_dir_t *dd = arg1;
 379         struct recvbeginsyncarg *rbsa = arg2;
 380         uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
 381         uint64_t dsobj;
 382
 383         dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
 384             rbsa->origin, flags, cr, tx);
 385
 386         rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
 387             rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
 388 }
 389
 390 static int
 391 recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
 392 {
 393         dsl_dataset_t *ds = arg1;
 394         struct recvbeginsyncarg *rbsa = arg2;
 395         int err;
 396
 397         /* must be a head ds */
 398         if (ds->ds_phys->ds_next_snap_obj != 0)
 399                 return (EINVAL);
 400
 401         /* must not be a clone ds */
 402         if (dsl_dir_is_clone(ds->ds_dir))
 403                 return (EINVAL);
 404
 405         err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
 406         if (err)
 407                 return (err);
 408
 409         if (rbsa->origin) {
 410                 /* make sure it's a snap in the same pool */
 411                 if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool)
 412                         return (EXDEV);
 413                 if (rbsa->origin->ds_phys->ds_num_children == 0)
 414                         return (EINVAL);
 415                 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
 416                         return (ENODEV);
 417         }
 418
 419         return (0);
 420 }
 421
 422 static void
 423 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 424 {
 425         dsl_dataset_t *ds = arg1;
 426         struct recvbeginsyncarg *rbsa = arg2;
 427         dsl_dir_t *dd = ds->ds_dir;
 428         uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
 429         uint64_t dsobj;
 430
 431         /*
 432          * NB: caller must provide an extra hold on the dsl_dir_t, so it
 433          * won't go away when dsl_dataset_destroy_sync() closes the
 434          * dataset.
 435          */
 436         dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
 437
 438         dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
 439
 440         rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
 441             rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
 442 }
 443
 444 /* ARGSUSED */
 445 static int
 446 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
 447 {
 448         dsl_dataset_t *ds = arg1;
 449         struct recvbeginsyncarg *rbsa = arg2;
 450         int err;
 451         uint64_t val;
 452
 453         /* must not have any changes since most recent snapshot */
 454         if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
 455                 return (ETXTBSY);
 456
 457         /* must already be a snapshot of this fs */
 458         if (ds->ds_phys->ds_prev_snap_obj == 0)
 459                 return (ENODEV);
 460
 461         /* most recent snapshot must match fromguid */
 462         if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
 463                 return (ENODEV);
 464
 465         /* temporary clone name must not exist */
 466         err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
 467             ds->ds_dir->dd_phys->dd_child_dir_zapobj,
 468             rbsa->clonelastname, 8, 1, &val);
 469         if (err == 0)
 470                 return (EEXIST);
 471         if (err != ENOENT)
 472                 return (err);
 473
 474         /* new snapshot name must not exist */
 475         err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
 476             ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
 477         if (err == 0)
 478                 return (EEXIST);
 479         if (err != ENOENT)
 480                 return (err);
 481         return (0);
 482 }
 483
 484 /* ARGSUSED */
 485 static void
 486 recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 487 {
 488         dsl_dataset_t *ohds = arg1;
 489         struct recvbeginsyncarg *rbsa = arg2;
 490         dsl_pool_t *dp = ohds->ds_dir->dd_pool;
 491         dsl_dataset_t *ods, *cds;
 492         uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
 493         uint64_t dsobj;
 494
 495         /* create the temporary clone */
 496         VERIFY(0 == dsl_dataset_hold_obj(dp, ohds->ds_phys->ds_prev_snap_obj,
 497             FTAG, &ods));
 498         dsobj = dsl_dataset_create_sync(ohds->ds_dir,
 499             rbsa->clonelastname, ods, flags, cr, tx);
 500         dsl_dataset_rele(ods, FTAG);
 501
 502         /* open the temporary clone */
 503         VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
 504             DS_MODE_INCONSISTENT, dmu_recv_tag, &cds));
 505
 506         /* copy the refquota from the target fs to the clone */
 507         if (ohds->ds_quota > 0)
 508                 dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
 509
 510         rbsa->ds = cds;
 511
 512         spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
 513             dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
 514 }
 515
 516 /* ARGSUSED */
 517 static void
 518 recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 519 {
 520         dsl_dataset_t *ds = arg1;
 521
 522         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 523         ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
 524
 525         spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
 526             ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
 527             ds->ds_object);
 528 }
 529
 530 /*
 531  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
 532  * succeeds; otherwise we will leak the holds on the datasets.
 533  */
 534 int
 535 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
 536     boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
 537 {
 538         int err = 0;
 539         boolean_t byteswap;
 540         struct recvbeginsyncarg rbsa;
 541         uint64_t version;
 542         int flags;
 543         dsl_dataset_t *ds;
 544
 545         if (drrb->drr_magic == DMU_BACKUP_MAGIC)
 546                 byteswap = FALSE;
 547         else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
 548                 byteswap = TRUE;
 549         else
 550                 return (EINVAL);
 551
 552         rbsa.tofs = tofs;
 553         rbsa.tosnap = tosnap;
 554         rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL;
 555         rbsa.fromguid = drrb->drr_fromguid;
 556         rbsa.type = drrb->drr_type;
 557         rbsa.tag = FTAG;
 558         rbsa.dsflags = 0;
 559         version = drrb->drr_version;
 560         flags = drrb->drr_flags;
 561
 562         if (byteswap) {
 563                 rbsa.type = BSWAP_32(rbsa.type);
 564                 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
 565                 version = BSWAP_64(version);
 566                 flags = BSWAP_32(flags);
 567         }
 568
 569         if (version != DMU_BACKUP_STREAM_VERSION ||
 570             rbsa.type >= DMU_OST_NUMTYPES ||
 571             ((flags & DRR_FLAG_CLONE) && origin == NULL))
 572                 return (EINVAL);
 573
 574         if (flags & DRR_FLAG_CI_DATA)
 575                 rbsa.dsflags = DS_FLAG_CI_DATASET;
 576
 577         bzero(drc, sizeof (dmu_recv_cookie_t));
 578         drc->drc_drrb = drrb;
 579         drc->drc_tosnap = tosnap;
 580         drc->drc_force = force;
 581
 582         /*
 583          * Process the begin in syncing context.
 584          */
 585         if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
 586                 /* offline incremental receive */
 587                 err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds);
 588                 if (err)
 589                         return (err);
 590
 591                 /*
 592                  * Only do the rollback if the most recent snapshot
 593                  * matches the incremental source
 594                  */
 595                 if (force) {
 596                         if (ds->ds_prev == NULL ||
 597                             ds->ds_prev->ds_phys->ds_guid !=
 598                             rbsa.fromguid) {
 599                                 dsl_dataset_disown(ds, dmu_recv_tag);
 600                                 return (ENODEV);
 601                         }
 602                         (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
 603                 }
 604                 rbsa.force = B_FALSE;
 605                 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
 606                     recv_incremental_check,
 607                     recv_offline_incremental_sync, ds, &rbsa, 1);
 608                 if (err) {
 609                         dsl_dataset_disown(ds, dmu_recv_tag);
 610                         return (err);
 611                 }
 612                 drc->drc_logical_ds = drc->drc_real_ds = ds;
 613         } else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
 614                 /* online incremental receive */
 615
 616                 /* tmp clone name is: tofs/%tosnap" */
 617                 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
 618                     "%%%s", tosnap);
 619
 620                 /* open the dataset we are logically receiving into */
 621                 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
 622                 if (err)
 623                         return (err);
 624
 625                 rbsa.force = force;
 626                 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
 627                     recv_incremental_check,
 628                     recv_online_incremental_sync, ds, &rbsa, 5);
 629                 if (err) {
 630                         dsl_dataset_rele(ds, dmu_recv_tag);
 631                         return (err);
 632                 }
 633                 drc->drc_logical_ds = ds;
 634                 drc->drc_real_ds = rbsa.ds;
 635         } else {
 636                 /* create new fs -- full backup or clone */
 637                 dsl_dir_t *dd = NULL;
 638                 const char *tail;
 639
 640                 err = dsl_dir_open(tofs, FTAG, &dd, &tail);
 641                 if (err)
 642                         return (err);
 643                 if (tail == NULL) {
 644                         if (!force) {
 645                                 dsl_dir_close(dd, FTAG);
 646                                 return (EEXIST);
 647                         }
 648
 649                         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
 650                         err = dsl_dataset_own_obj(dd->dd_pool,
 651                             dd->dd_phys->dd_head_dataset_obj,
 652                             DS_MODE_INCONSISTENT, FTAG, &ds);
 653                         rw_exit(&dd->dd_pool->dp_config_rwlock);
 654                         if (err) {
 655                                 dsl_dir_close(dd, FTAG);
 656                                 return (err);
 657                         }
 658
 659                         dsl_dataset_make_exclusive(ds, FTAG);
 660                         err = dsl_sync_task_do(dd->dd_pool,
 661                             recv_full_existing_check,
 662                             recv_full_existing_sync, ds, &rbsa, 5);
 663                         dsl_dataset_disown(ds, FTAG);
 664                 } else {
 665                         err = dsl_sync_task_do(dd->dd_pool, recv_full_check,
 666                             recv_full_sync, dd, &rbsa, 5);
 667                 }
 668                 dsl_dir_close(dd, FTAG);
 669                 if (err)
 670                         return (err);
 671                 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
 672                 drc->drc_newfs = B_TRUE;
 673         }
 674
 675         return (0);
 676 }
 677
 678 struct restorearg {
 679         int err;
 680         int byteswap;
 681         vnode_t *vp;
 682         char *buf;
 683         uint64_t voff;
 684         int bufsize; /* amount of memory allocated for buf */
 685         zio_cksum_t cksum;
 686 };
 687
 688 static void *
 689 restore_read(struct restorearg *ra, int len)
 690 {
 691         void *rv;
 692         int done = 0;
 693
 694         /* some things will require 8-byte alignment, so everything must */
 695         ASSERT3U(len % 8, ==, 0);
 696
 697         while (done < len) {
 698                 ssize_t resid;
 699
 700                 ra->err = vn_rdwr(UIO_READ, ra->vp,
 701                     (caddr_t)ra->buf + done, len - done,
 702                     ra->voff, UIO_SYSSPACE, FAPPEND,
 703                     RLIM64_INFINITY, CRED(), &resid);
 704
 705                 if (resid == len - done)
 706                         ra->err = EINVAL;
 707                 ra->voff += len - done - resid;
 708                 done = len - resid;
 709                 if (ra->err)
 710                         return (NULL);
 711         }
 712
 713         ASSERT3U(done, ==, len);
 714         rv = ra->buf;
 715         if (ra->byteswap)
 716                 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
 717         else
 718                 fletcher_4_incremental_native(rv, len, &ra->cksum);
 719         return (rv);
 720 }
 721
 722 static void
 723 backup_byteswap(dmu_replay_record_t *drr)
 724 {
 725 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
 726 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
 727         drr->drr_type = BSWAP_32(drr->drr_type);
 728         drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
 729         switch (drr->drr_type) {
 730         case DRR_BEGIN:
 731                 DO64(drr_begin.drr_magic);
 732                 DO64(drr_begin.drr_version);
 733                 DO64(drr_begin.drr_creation_time);
 734                 DO32(drr_begin.drr_type);
 735                 DO32(drr_begin.drr_flags);
 736                 DO64(drr_begin.drr_toguid);
 737                 DO64(drr_begin.drr_fromguid);
 738                 break;
 739         case DRR_OBJECT:
 740                 DO64(drr_object.drr_object);
 741                 /* DO64(drr_object.drr_allocation_txg); */
 742                 DO32(drr_object.drr_type);
 743                 DO32(drr_object.drr_bonustype);
 744                 DO32(drr_object.drr_blksz);
 745                 DO32(drr_object.drr_bonuslen);
 746                 break;
 747         case DRR_FREEOBJECTS:
 748                 DO64(drr_freeobjects.drr_firstobj);
 749                 DO64(drr_freeobjects.drr_numobjs);
 750                 break;
 751         case DRR_WRITE:
 752                 DO64(drr_write.drr_object);
 753                 DO32(drr_write.drr_type);
 754                 DO64(drr_write.drr_offset);
 755                 DO64(drr_write.drr_length);
 756                 break;
 757         case DRR_FREE:
 758                 DO64(drr_free.drr_object);
 759                 DO64(drr_free.drr_offset);
 760                 DO64(drr_free.drr_length);
 761                 break;
 762         case DRR_END:
 763                 DO64(drr_end.drr_checksum.zc_word[0]);
 764                 DO64(drr_end.drr_checksum.zc_word[1]);
 765                 DO64(drr_end.drr_checksum.zc_word[2]);
 766                 DO64(drr_end.drr_checksum.zc_word[3]);
 767                 break;
 768         }
 769 #undef DO64
 770 #undef DO32
 771 }
 772
 773 static int
 774 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
 775 {
 776         int err;
 777         dmu_tx_t *tx;
 778         void *data = NULL;
 779
 780         if (drro->drr_type == DMU_OT_NONE ||
 781             drro->drr_type >= DMU_OT_NUMTYPES ||
 782             drro->drr_bonustype >= DMU_OT_NUMTYPES ||
 783             drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
 784             drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
 785             P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
 786             drro->drr_blksz < SPA_MINBLOCKSIZE ||
 787             drro->drr_blksz > SPA_MAXBLOCKSIZE ||
 788             drro->drr_bonuslen > DN_MAX_BONUSLEN) {
 789                 return (EINVAL);
 790         }
 791
 792         err = dmu_object_info(os, drro->drr_object, NULL);
 793
 794         if (err != 0 && err != ENOENT)
 795                 return (EINVAL);
 796
 797         if (drro->drr_bonuslen) {
 798                 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
 799                 if (ra->err)
 800                         return (ra->err);
 801         }
 802
 803         if (err == ENOENT) {
 804                 /* currently free, want to be allocated */
 805                 tx = dmu_tx_create(os);
 806                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
 807                 err = dmu_tx_assign(tx, TXG_WAIT);
 808                 if (err) {
 809                         dmu_tx_abort(tx);
 810                         return (err);
 811                 }
 812                 err = dmu_object_claim(os, drro->drr_object,
 813                     drro->drr_type, drro->drr_blksz,
 814                     drro->drr_bonustype, drro->drr_bonuslen, tx);
 815                 dmu_tx_commit(tx);
 816         } else {
 817                 /* currently allocated, want to be allocated */
 818                 err = dmu_object_reclaim(os, drro->drr_object,
 819                     drro->drr_type, drro->drr_blksz,
 820                     drro->drr_bonustype, drro->drr_bonuslen);
 821         }
 822         if (err)
 823                 return (EINVAL);
 824
 825         tx = dmu_tx_create(os);
 826         dmu_tx_hold_bonus(tx, drro->drr_object);
 827         err = dmu_tx_assign(tx, TXG_WAIT);
 828         if (err) {
 829                 dmu_tx_abort(tx);
 830                 return (err);
 831         }
 832
 833         dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
 834         dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
 835
 836         if (data != NULL) {
 837                 dmu_buf_t *db;
 838
 839                 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
 840                 dmu_buf_will_dirty(db, tx);
 841
 842                 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
 843                 bcopy(data, db->db_data, drro->drr_bonuslen);
 844                 if (ra->byteswap) {
 845                         dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
 846                             drro->drr_bonuslen);
 847                 }
 848                 dmu_buf_rele(db, FTAG);
 849         }
 850         dmu_tx_commit(tx);
 851         return (0);
 852 }
 853
 854 /* ARGSUSED */
 855 static int
 856 restore_freeobjects(struct restorearg *ra, objset_t *os,
 857     struct drr_freeobjects *drrfo)
 858 {
 859         uint64_t obj;
 860
 861         if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
 862                 return (EINVAL);
 863
 864         for (obj = drrfo->drr_firstobj;
 865             obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
 866             (void) dmu_object_next(os, &obj, FALSE, 0)) {
 867                 int err;
 868
 869                 if (dmu_object_info(os, obj, NULL) != 0)
 870                         continue;
 871
 872                 err = dmu_free_object(os, obj);
 873                 if (err)
 874                         return (err);
 875         }
 876         return (0);
 877 }
 878
 879 static int
 880 restore_write(struct restorearg *ra, objset_t *os,
 881     struct drr_write *drrw)
 882 {
 883         dmu_tx_t *tx;
 884         void *data;
 885         int err;
 886
 887         if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
 888             drrw->drr_type >= DMU_OT_NUMTYPES)
 889                 return (EINVAL);
 890
 891         data = restore_read(ra, drrw->drr_length);
 892         if (data == NULL)
 893                 return (ra->err);
 894
 895         if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
 896                 return (EINVAL);
 897
 898         tx = dmu_tx_create(os);
 899
 900         dmu_tx_hold_write(tx, drrw->drr_object,
 901             drrw->drr_offset, drrw->drr_length);
 902         err = dmu_tx_assign(tx, TXG_WAIT);
 903         if (err) {
 904                 dmu_tx_abort(tx);
 905                 return (err);
 906         }
 907         if (ra->byteswap)
 908                 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
 909         dmu_write(os, drrw->drr_object,
 910             drrw->drr_offset, drrw->drr_length, data, tx);
 911         dmu_tx_commit(tx);
 912         return (0);
 913 }
 914
 915 /* ARGSUSED */
 916 static int
 917 restore_free(struct restorearg *ra, objset_t *os,
 918     struct drr_free *drrf)
 919 {
 920         int err;
 921
 922         if (drrf->drr_length != -1ULL &&
 923             drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
 924                 return (EINVAL);
 925
 926         if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
 927                 return (EINVAL);
 928
 929         err = dmu_free_long_range(os, drrf->drr_object,
 930             drrf->drr_offset, drrf->drr_length);
 931         return (err);
 932 }
 933
 934 void
 935 dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
 936 {
 937         if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
 938                 /*
 939                  * online incremental or new fs: destroy the fs (which
 940                  * may be a clone) that we created
 941                  */
 942                 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
 943                 if (drc->drc_real_ds != drc->drc_logical_ds)
 944                         dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
 945         } else {
 946                 /*
 947                  * offline incremental: rollback to most recent snapshot.
 948                  */
 949                 (void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE);
 950                 dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag);
 951         }
 952 }
 953
 954 /*
 955  * NB: callers *must* call dmu_recv_end() if this succeeds.
 956  */
 957 int
 958 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp)
 959 {
 960         struct restorearg ra = { 0 };
 961         dmu_replay_record_t *drr;
 962         objset_t *os;
 963         zio_cksum_t pcksum;
 964
 965         if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
 966                 ra.byteswap = TRUE;
 967
 968         {
 969                 /* compute checksum of drr_begin record */
 970                 dmu_replay_record_t *drr;
 971                 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
 972
 973                 drr->drr_type = DRR_BEGIN;
 974                 drr->drr_u.drr_begin = *drc->drc_drrb;
 975                 if (ra.byteswap) {
 976                         fletcher_4_incremental_byteswap(drr,
 977                             sizeof (dmu_replay_record_t), &ra.cksum);
 978                 } else {
 979                         fletcher_4_incremental_native(drr,
 980                             sizeof (dmu_replay_record_t), &ra.cksum);
 981                 }
 982                 kmem_free(drr, sizeof (dmu_replay_record_t));
 983         }
 984
 985         if (ra.byteswap) {
 986                 struct drr_begin *drrb = drc->drc_drrb;
 987                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
 988                 drrb->drr_version = BSWAP_64(drrb->drr_version);
 989                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
 990                 drrb->drr_type = BSWAP_32(drrb->drr_type);
 991                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
 992                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
 993         }
 994
 995         ra.vp = vp;
 996         ra.voff = *voffp;
 997         ra.bufsize = 1<<20;
 998         ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
 999
1000         /* these were verified in dmu_recv_begin */
1001         ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
1002         ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1003
1004         /*
1005          * Open the objset we are modifying.
1006          */
1007         VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0);
1008
1009         ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1010
1011         /*
1012          * Read records and process them.
1013          */
1014         pcksum = ra.cksum;
1015         while (ra.err == 0 &&
1016             NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1017                 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1018                         ra.err = EINTR;
1019                         goto out;
1020                 }
1021
1022                 if (ra.byteswap)
1023                         backup_byteswap(drr);
1024
1025                 switch (drr->drr_type) {
1026                 case DRR_OBJECT:
1027                 {
1028                         /*
1029                          * We need to make a copy of the record header,
1030                          * because restore_{object,write} may need to
1031                          * restore_read(), which will invalidate drr.
1032                          */
1033                         struct drr_object drro = drr->drr_u.drr_object;
1034                         ra.err = restore_object(&ra, os, &drro);
1035                         break;
1036                 }
1037                 case DRR_FREEOBJECTS:
1038                 {
1039                         struct drr_freeobjects drrfo =
1040                             drr->drr_u.drr_freeobjects;
1041                         ra.err = restore_freeobjects(&ra, os, &drrfo);
1042                         break;
1043                 }
1044                 case DRR_WRITE:
1045                 {
1046                         struct drr_write drrw = drr->drr_u.drr_write;
1047                         ra.err = restore_write(&ra, os, &drrw);
1048                         break;
1049                 }
1050                 case DRR_FREE:
1051                 {
1052                         struct drr_free drrf = drr->drr_u.drr_free;
1053                         ra.err = restore_free(&ra, os, &drrf);
1054                         break;
1055                 }
1056                 case DRR_END:
1057                 {
1058                         struct drr_end drre = drr->drr_u.drr_end;
1059                         /*
1060                          * We compare against the *previous* checksum
1061                          * value, because the stored checksum is of
1062                          * everything before the DRR_END record.
1063                          */
1064                         if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1065                                 ra.err = ECKSUM;
1066                         goto out;
1067                 }
1068                 default:
1069                         ra.err = EINVAL;
1070                         goto out;
1071                 }
1072                 pcksum = ra.cksum;
1073         }
1074         ASSERT(ra.err != 0);
1075
1076 out:
1077         dmu_objset_close(os);
1078
1079         if (ra.err != 0) {
1080                 /*
1081                  * rollback or destroy what we created, so we don't
1082                  * leave it in the restoring state.
1083                  */
1084                 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1085                 dmu_recv_abort_cleanup(drc);
1086         }
1087
1088         kmem_free(ra.buf, ra.bufsize);
1089         *voffp = ra.voff;
1090         return (ra.err);
1091 }
1092
1093 struct recvendsyncarg {
1094         char *tosnap;
1095         uint64_t creation_time;
1096         uint64_t toguid;
1097 };
1098
1099 static int
1100 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1101 {
1102         dsl_dataset_t *ds = arg1;
1103         struct recvendsyncarg *resa = arg2;
1104
1105         return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1106 }
1107
1108 static void
1109 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1110 {
1111         dsl_dataset_t *ds = arg1;
1112         struct recvendsyncarg *resa = arg2;
1113
1114         dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
1115
1116         /* set snapshot's creation time and guid */
1117         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1118         ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1119         ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1120         ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1121
1122         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1123         ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1124 }
1125
1126 int
1127 dmu_recv_end(dmu_recv_cookie_t *drc)
1128 {
1129         struct recvendsyncarg resa;
1130         dsl_dataset_t *ds = drc->drc_logical_ds;
1131         int err;
1132
1133         /*
1134          * XXX hack; seems the ds is still dirty and
1135          * dsl_pool_zil_clean() expects it to have a ds_user_ptr
1136          * (and zil), but clone_swap() can close it.
1137          */
1138         txg_wait_synced(ds->ds_dir->dd_pool, 0);
1139
1140         if (ds != drc->drc_real_ds) {
1141                 /* we are doing an online recv */
1142                 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1143                         err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1144                             drc->drc_force);
1145                         if (err)
1146                                 dsl_dataset_disown(ds, dmu_recv_tag);
1147                 } else {
1148                         err = EBUSY;
1149                         dsl_dataset_rele(ds, dmu_recv_tag);
1150                 }
1151                 /* dsl_dataset_destroy() will disown the ds */
1152                 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
1153                 if (err)
1154                         return (err);
1155         }
1156
1157         resa.creation_time = drc->drc_drrb->drr_creation_time;
1158         resa.toguid = drc->drc_drrb->drr_toguid;
1159         resa.tosnap = drc->drc_tosnap;
1160
1161         err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1162             recv_end_check, recv_end_sync, ds, &resa, 3);
1163         if (err) {
1164                 if (drc->drc_newfs) {
1165                         ASSERT(ds == drc->drc_real_ds);
1166                         (void) dsl_dataset_destroy(ds, dmu_recv_tag);
1167                         return (err);
1168                 } else {
1169                         (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
1170                 }
1171         }
1172
1173         /* release the hold from dmu_recv_begin */
1174         dsl_dataset_disown(ds, dmu_recv_tag);
1175         return (err);
1176 }