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         struct dsl_ds_destroyarg dsda = {0};
 397
 398         /* must be a head ds */
 399         if (ds->ds_phys->ds_next_snap_obj != 0)
 400                 return (EINVAL);
 401
 402         /* must not be a clone ds */
 403         if (dsl_dir_is_clone(ds->ds_dir))
 404                 return (EINVAL);
 405
 406         dsda.ds = ds;
 407         err = dsl_dataset_destroy_check(&dsda, rbsa->tag, tx);
 408         if (err)
 409                 return (err);
 410
 411         if (rbsa->origin) {
 412                 /* make sure it's a snap in the same pool */
 413                 if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool)
 414                         return (EXDEV);
 415                 if (rbsa->origin->ds_phys->ds_num_children == 0)
 416                         return (EINVAL);
 417                 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
 418                         return (ENODEV);
 419         }
 420
 421         return (0);
 422 }
 423
 424 static void
 425 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 426 {
 427         dsl_dataset_t *ds = arg1;
 428         struct recvbeginsyncarg *rbsa = arg2;
 429         dsl_dir_t *dd = ds->ds_dir;
 430         uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
 431         uint64_t dsobj;
 432         struct dsl_ds_destroyarg dsda = {0};
 433
 434         /*
 435          * NB: caller must provide an extra hold on the dsl_dir_t, so it
 436          * won't go away when dsl_dataset_destroy_sync() closes the
 437          * dataset.
 438          */
 439         dsda.ds = ds;
 440         dsl_dataset_destroy_sync(&dsda, rbsa->tag, cr, tx);
 441         ASSERT3P(dsda.rm_origin, ==, NULL);
 442
 443         dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
 444
 445         rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
 446             rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
 447 }
 448
 449 /* ARGSUSED */
 450 static int
 451 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
 452 {
 453         dsl_dataset_t *ds = arg1;
 454         struct recvbeginsyncarg *rbsa = arg2;
 455         int err;
 456         uint64_t val;
 457
 458         /* must not have any changes since most recent snapshot */
 459         if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
 460                 return (ETXTBSY);
 461
 462         /* must already be a snapshot of this fs */
 463         if (ds->ds_phys->ds_prev_snap_obj == 0)
 464                 return (ENODEV);
 465
 466         /* most recent snapshot must match fromguid */
 467         if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
 468                 return (ENODEV);
 469
 470         /* temporary clone name must not exist */
 471         err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
 472             ds->ds_dir->dd_phys->dd_child_dir_zapobj,
 473             rbsa->clonelastname, 8, 1, &val);
 474         if (err == 0)
 475                 return (EEXIST);
 476         if (err != ENOENT)
 477                 return (err);
 478
 479         /* new snapshot name must not exist */
 480         err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
 481             ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
 482         if (err == 0)
 483                 return (EEXIST);
 484         if (err != ENOENT)
 485                 return (err);
 486         return (0);
 487 }
 488
 489 /* ARGSUSED */
 490 static void
 491 recv_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 492 {
 493         dsl_dataset_t *ohds = arg1;
 494         struct recvbeginsyncarg *rbsa = arg2;
 495         dsl_pool_t *dp = ohds->ds_dir->dd_pool;
 496         dsl_dataset_t *ods, *cds;
 497         uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
 498         uint64_t dsobj;
 499
 500         /* create the temporary clone */
 501         VERIFY(0 == dsl_dataset_hold_obj(dp, ohds->ds_phys->ds_prev_snap_obj,
 502             FTAG, &ods));
 503         dsobj = dsl_dataset_create_sync(ohds->ds_dir,
 504             rbsa->clonelastname, ods, flags, cr, tx);
 505         dsl_dataset_rele(ods, FTAG);
 506
 507         /* open the temporary clone */
 508         VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
 509             DS_MODE_INCONSISTENT, dmu_recv_tag, &cds));
 510
 511         /* copy the refquota from the target fs to the clone */
 512         if (ohds->ds_quota > 0)
 513                 dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
 514
 515         rbsa->ds = cds;
 516
 517         spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
 518             dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
 519 }
 520
 521 /*
 522  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
 523  * succeeds; otherwise we will leak the holds on the datasets.
 524  */
 525 int
 526 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
 527     boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
 528 {
 529         int err = 0;
 530         boolean_t byteswap;
 531         struct recvbeginsyncarg rbsa;
 532         uint64_t version;
 533         int flags;
 534         dsl_dataset_t *ds;
 535
 536         if (drrb->drr_magic == DMU_BACKUP_MAGIC)
 537                 byteswap = FALSE;
 538         else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
 539                 byteswap = TRUE;
 540         else
 541                 return (EINVAL);
 542
 543         rbsa.tofs = tofs;
 544         rbsa.tosnap = tosnap;
 545         rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL;
 546         rbsa.fromguid = drrb->drr_fromguid;
 547         rbsa.type = drrb->drr_type;
 548         rbsa.tag = FTAG;
 549         rbsa.dsflags = 0;
 550         version = drrb->drr_version;
 551         flags = drrb->drr_flags;
 552
 553         if (byteswap) {
 554                 rbsa.type = BSWAP_32(rbsa.type);
 555                 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
 556                 version = BSWAP_64(version);
 557                 flags = BSWAP_32(flags);
 558         }
 559
 560         if (version != DMU_BACKUP_STREAM_VERSION ||
 561             rbsa.type >= DMU_OST_NUMTYPES ||
 562             ((flags & DRR_FLAG_CLONE) && origin == NULL))
 563                 return (EINVAL);
 564
 565         if (flags & DRR_FLAG_CI_DATA)
 566                 rbsa.dsflags = DS_FLAG_CI_DATASET;
 567
 568         bzero(drc, sizeof (dmu_recv_cookie_t));
 569         drc->drc_drrb = drrb;
 570         drc->drc_tosnap = tosnap;
 571         drc->drc_force = force;
 572
 573         /*
 574          * Process the begin in syncing context.
 575          */
 576         if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
 577                 /* incremental receive */
 578
 579                 /* tmp clone name is: tofs/%tosnap" */
 580                 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
 581                     "%%%s", tosnap);
 582
 583                 /* open the dataset we are logically receiving into */
 584                 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
 585                 if (err)
 586                         return (err);
 587
 588                 /* must not have an incremental recv already in progress */
 589                 if (!mutex_tryenter(&ds->ds_recvlock)) {
 590                         dsl_dataset_rele(ds, dmu_recv_tag);
 591                         return (EBUSY);
 592                 }
 593
 594                 rbsa.force = force;
 595                 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
 596                     recv_incremental_check,
 597                     recv_incremental_sync, ds, &rbsa, 5);
 598                 if (err) {
 599                         mutex_exit(&ds->ds_recvlock);
 600                         dsl_dataset_rele(ds, dmu_recv_tag);
 601                         return (err);
 602                 }
 603                 drc->drc_logical_ds = ds;
 604                 drc->drc_real_ds = rbsa.ds;
 605         } else {
 606                 /* create new fs -- full backup or clone */
 607                 dsl_dir_t *dd = NULL;
 608                 const char *tail;
 609
 610                 err = dsl_dir_open(tofs, FTAG, &dd, &tail);
 611                 if (err)
 612                         return (err);
 613                 if (tail == NULL) {
 614                         if (!force) {
 615                                 dsl_dir_close(dd, FTAG);
 616                                 return (EEXIST);
 617                         }
 618
 619                         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
 620                         err = dsl_dataset_own_obj(dd->dd_pool,
 621                             dd->dd_phys->dd_head_dataset_obj,
 622                             DS_MODE_INCONSISTENT, FTAG, &ds);
 623                         rw_exit(&dd->dd_pool->dp_config_rwlock);
 624                         if (err) {
 625                                 dsl_dir_close(dd, FTAG);
 626                                 return (err);
 627                         }
 628
 629                         dsl_dataset_make_exclusive(ds, FTAG);
 630                         err = dsl_sync_task_do(dd->dd_pool,
 631                             recv_full_existing_check,
 632                             recv_full_existing_sync, ds, &rbsa, 5);
 633                         dsl_dataset_disown(ds, FTAG);
 634                 } else {
 635                         err = dsl_sync_task_do(dd->dd_pool, recv_full_check,
 636                             recv_full_sync, dd, &rbsa, 5);
 637                 }
 638                 dsl_dir_close(dd, FTAG);
 639                 if (err)
 640                         return (err);
 641                 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
 642                 drc->drc_newfs = B_TRUE;
 643         }
 644
 645         return (0);
 646 }
 647
 648 struct restorearg {
 649         int err;
 650         int byteswap;
 651         vnode_t *vp;
 652         char *buf;
 653         uint64_t voff;
 654         int bufsize; /* amount of memory allocated for buf */
 655         zio_cksum_t cksum;
 656 };
 657
 658 static void *
 659 restore_read(struct restorearg *ra, int len)
 660 {
 661         void *rv;
 662         int done = 0;
 663
 664         /* some things will require 8-byte alignment, so everything must */
 665         ASSERT3U(len % 8, ==, 0);
 666
 667         while (done < len) {
 668                 ssize_t resid;
 669
 670                 ra->err = vn_rdwr(UIO_READ, ra->vp,
 671                     (caddr_t)ra->buf + done, len - done,
 672                     ra->voff, UIO_SYSSPACE, FAPPEND,
 673                     RLIM64_INFINITY, CRED(), &resid);
 674
 675                 if (resid == len - done)
 676                         ra->err = EINVAL;
 677                 ra->voff += len - done - resid;
 678                 done = len - resid;
 679                 if (ra->err)
 680                         return (NULL);
 681         }
 682
 683         ASSERT3U(done, ==, len);
 684         rv = ra->buf;
 685         if (ra->byteswap)
 686                 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
 687         else
 688                 fletcher_4_incremental_native(rv, len, &ra->cksum);
 689         return (rv);
 690 }
 691
 692 static void
 693 backup_byteswap(dmu_replay_record_t *drr)
 694 {
 695 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
 696 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
 697         drr->drr_type = BSWAP_32(drr->drr_type);
 698         drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
 699         switch (drr->drr_type) {
 700         case DRR_BEGIN:
 701                 DO64(drr_begin.drr_magic);
 702                 DO64(drr_begin.drr_version);
 703                 DO64(drr_begin.drr_creation_time);
 704                 DO32(drr_begin.drr_type);
 705                 DO32(drr_begin.drr_flags);
 706                 DO64(drr_begin.drr_toguid);
 707                 DO64(drr_begin.drr_fromguid);
 708                 break;
 709         case DRR_OBJECT:
 710                 DO64(drr_object.drr_object);
 711                 /* DO64(drr_object.drr_allocation_txg); */
 712                 DO32(drr_object.drr_type);
 713                 DO32(drr_object.drr_bonustype);
 714                 DO32(drr_object.drr_blksz);
 715                 DO32(drr_object.drr_bonuslen);
 716                 break;
 717         case DRR_FREEOBJECTS:
 718                 DO64(drr_freeobjects.drr_firstobj);
 719                 DO64(drr_freeobjects.drr_numobjs);
 720                 break;
 721         case DRR_WRITE:
 722                 DO64(drr_write.drr_object);
 723                 DO32(drr_write.drr_type);
 724                 DO64(drr_write.drr_offset);
 725                 DO64(drr_write.drr_length);
 726                 break;
 727         case DRR_FREE:
 728                 DO64(drr_free.drr_object);
 729                 DO64(drr_free.drr_offset);
 730                 DO64(drr_free.drr_length);
 731                 break;
 732         case DRR_END:
 733                 DO64(drr_end.drr_checksum.zc_word[0]);
 734                 DO64(drr_end.drr_checksum.zc_word[1]);
 735                 DO64(drr_end.drr_checksum.zc_word[2]);
 736                 DO64(drr_end.drr_checksum.zc_word[3]);
 737                 break;
 738         }
 739 #undef DO64
 740 #undef DO32
 741 }
 742
 743 static int
 744 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
 745 {
 746         int err;
 747         dmu_tx_t *tx;
 748         void *data = NULL;
 749
 750         if (drro->drr_type == DMU_OT_NONE ||
 751             drro->drr_type >= DMU_OT_NUMTYPES ||
 752             drro->drr_bonustype >= DMU_OT_NUMTYPES ||
 753             drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
 754             drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
 755             P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
 756             drro->drr_blksz < SPA_MINBLOCKSIZE ||
 757             drro->drr_blksz > SPA_MAXBLOCKSIZE ||
 758             drro->drr_bonuslen > DN_MAX_BONUSLEN) {
 759                 return (EINVAL);
 760         }
 761
 762         err = dmu_object_info(os, drro->drr_object, NULL);
 763
 764         if (err != 0 && err != ENOENT)
 765                 return (EINVAL);
 766
 767         if (drro->drr_bonuslen) {
 768                 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
 769                 if (ra->err)
 770                         return (ra->err);
 771         }
 772
 773         if (err == ENOENT) {
 774                 /* currently free, want to be allocated */
 775                 tx = dmu_tx_create(os);
 776                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
 777                 err = dmu_tx_assign(tx, TXG_WAIT);
 778                 if (err) {
 779                         dmu_tx_abort(tx);
 780                         return (err);
 781                 }
 782                 err = dmu_object_claim(os, drro->drr_object,
 783                     drro->drr_type, drro->drr_blksz,
 784                     drro->drr_bonustype, drro->drr_bonuslen, tx);
 785                 dmu_tx_commit(tx);
 786         } else {
 787                 /* currently allocated, want to be allocated */
 788                 err = dmu_object_reclaim(os, drro->drr_object,
 789                     drro->drr_type, drro->drr_blksz,
 790                     drro->drr_bonustype, drro->drr_bonuslen);
 791         }
 792         if (err)
 793                 return (EINVAL);
 794
 795         tx = dmu_tx_create(os);
 796         dmu_tx_hold_bonus(tx, drro->drr_object);
 797         err = dmu_tx_assign(tx, TXG_WAIT);
 798         if (err) {
 799                 dmu_tx_abort(tx);
 800                 return (err);
 801         }
 802
 803         dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
 804         dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
 805
 806         if (data != NULL) {
 807                 dmu_buf_t *db;
 808
 809                 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
 810                 dmu_buf_will_dirty(db, tx);
 811
 812                 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
 813                 bcopy(data, db->db_data, drro->drr_bonuslen);
 814                 if (ra->byteswap) {
 815                         dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
 816                             drro->drr_bonuslen);
 817                 }
 818                 dmu_buf_rele(db, FTAG);
 819         }
 820         dmu_tx_commit(tx);
 821         return (0);
 822 }
 823
 824 /* ARGSUSED */
 825 static int
 826 restore_freeobjects(struct restorearg *ra, objset_t *os,
 827     struct drr_freeobjects *drrfo)
 828 {
 829         uint64_t obj;
 830
 831         if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
 832                 return (EINVAL);
 833
 834         for (obj = drrfo->drr_firstobj;
 835             obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
 836             (void) dmu_object_next(os, &obj, FALSE, 0)) {
 837                 int err;
 838
 839                 if (dmu_object_info(os, obj, NULL) != 0)
 840                         continue;
 841
 842                 err = dmu_free_object(os, obj);
 843                 if (err)
 844                         return (err);
 845         }
 846         return (0);
 847 }
 848
 849 static int
 850 restore_write(struct restorearg *ra, objset_t *os,
 851     struct drr_write *drrw)
 852 {
 853         dmu_tx_t *tx;
 854         void *data;
 855         int err;
 856
 857         if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
 858             drrw->drr_type >= DMU_OT_NUMTYPES)
 859                 return (EINVAL);
 860
 861         data = restore_read(ra, drrw->drr_length);
 862         if (data == NULL)
 863                 return (ra->err);
 864
 865         if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
 866                 return (EINVAL);
 867
 868         tx = dmu_tx_create(os);
 869
 870         dmu_tx_hold_write(tx, drrw->drr_object,
 871             drrw->drr_offset, drrw->drr_length);
 872         err = dmu_tx_assign(tx, TXG_WAIT);
 873         if (err) {
 874                 dmu_tx_abort(tx);
 875                 return (err);
 876         }
 877         if (ra->byteswap)
 878                 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
 879         dmu_write(os, drrw->drr_object,
 880             drrw->drr_offset, drrw->drr_length, data, tx);
 881         dmu_tx_commit(tx);
 882         return (0);
 883 }
 884
 885 /* ARGSUSED */
 886 static int
 887 restore_free(struct restorearg *ra, objset_t *os,
 888     struct drr_free *drrf)
 889 {
 890         int err;
 891
 892         if (drrf->drr_length != -1ULL &&
 893             drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
 894                 return (EINVAL);
 895
 896         if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
 897                 return (EINVAL);
 898
 899         err = dmu_free_long_range(os, drrf->drr_object,
 900             drrf->drr_offset, drrf->drr_length);
 901         return (err);
 902 }
 903
 904 /*
 905  * NB: callers *must* call dmu_recv_end() if this succeeds.
 906  */
 907 int
 908 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp)
 909 {
 910         struct restorearg ra = { 0 };
 911         dmu_replay_record_t *drr;
 912         objset_t *os;
 913         zio_cksum_t pcksum;
 914
 915         if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
 916                 ra.byteswap = TRUE;
 917
 918         {
 919                 /* compute checksum of drr_begin record */
 920                 dmu_replay_record_t *drr;
 921                 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
 922
 923                 drr->drr_type = DRR_BEGIN;
 924                 drr->drr_u.drr_begin = *drc->drc_drrb;
 925                 if (ra.byteswap) {
 926                         fletcher_4_incremental_byteswap(drr,
 927                             sizeof (dmu_replay_record_t), &ra.cksum);
 928                 } else {
 929                         fletcher_4_incremental_native(drr,
 930                             sizeof (dmu_replay_record_t), &ra.cksum);
 931                 }
 932                 kmem_free(drr, sizeof (dmu_replay_record_t));
 933         }
 934
 935         if (ra.byteswap) {
 936                 struct drr_begin *drrb = drc->drc_drrb;
 937                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
 938                 drrb->drr_version = BSWAP_64(drrb->drr_version);
 939                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
 940                 drrb->drr_type = BSWAP_32(drrb->drr_type);
 941                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
 942                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
 943         }
 944
 945         ra.vp = vp;
 946         ra.voff = *voffp;
 947         ra.bufsize = 1<<20;
 948         ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
 949
 950         /* these were verified in dmu_recv_begin */
 951         ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
 952         ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
 953
 954         /*
 955          * Open the objset we are modifying.
 956          */
 957         VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0);
 958
 959         ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
 960
 961         /*
 962          * Read records and process them.
 963          */
 964         pcksum = ra.cksum;
 965         while (ra.err == 0 &&
 966             NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
 967                 if (issig(JUSTLOOKING) && issig(FORREAL)) {
 968                         ra.err = EINTR;
 969                         goto out;
 970                 }
 971
 972                 if (ra.byteswap)
 973                         backup_byteswap(drr);
 974
 975                 switch (drr->drr_type) {
 976                 case DRR_OBJECT:
 977                 {
 978                         /*
 979                          * We need to make a copy of the record header,
 980                          * because restore_{object,write} may need to
 981                          * restore_read(), which will invalidate drr.
 982                          */
 983                         struct drr_object drro = drr->drr_u.drr_object;
 984                         ra.err = restore_object(&ra, os, &drro);
 985                         break;
 986                 }
 987                 case DRR_FREEOBJECTS:
 988                 {
 989                         struct drr_freeobjects drrfo =
 990                             drr->drr_u.drr_freeobjects;
 991                         ra.err = restore_freeobjects(&ra, os, &drrfo);
 992                         break;
 993                 }
 994                 case DRR_WRITE:
 995                 {
 996                         struct drr_write drrw = drr->drr_u.drr_write;
 997                         ra.err = restore_write(&ra, os, &drrw);
 998                         break;
 999                 }
1000                 case DRR_FREE:
1001                 {
1002                         struct drr_free drrf = drr->drr_u.drr_free;
1003                         ra.err = restore_free(&ra, os, &drrf);
1004                         break;
1005                 }
1006                 case DRR_END:
1007                 {
1008                         struct drr_end drre = drr->drr_u.drr_end;
1009                         /*
1010                          * We compare against the *previous* checksum
1011                          * value, because the stored checksum is of
1012                          * everything before the DRR_END record.
1013                          */
1014                         if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1015                                 ra.err = ECKSUM;
1016                         goto out;
1017                 }
1018                 default:
1019                         ra.err = EINVAL;
1020                         goto out;
1021                 }
1022                 pcksum = ra.cksum;
1023         }
1024         ASSERT(ra.err != 0);
1025
1026 out:
1027         dmu_objset_close(os);
1028
1029         if (ra.err != 0) {
1030                 /*
1031                  * destroy what we created, so we don't leave it in the
1032                  * inconsistent restoring state.
1033                  */
1034                 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1035
1036                 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1037                     B_FALSE);
1038                 if (drc->drc_real_ds != drc->drc_logical_ds) {
1039                         mutex_exit(&drc->drc_logical_ds->ds_recvlock);
1040                         dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
1041                 }
1042         }
1043
1044         kmem_free(ra.buf, ra.bufsize);
1045         *voffp = ra.voff;
1046         return (ra.err);
1047 }
1048
1049 struct recvendsyncarg {
1050         char *tosnap;
1051         uint64_t creation_time;
1052         uint64_t toguid;
1053 };
1054
1055 static int
1056 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1057 {
1058         dsl_dataset_t *ds = arg1;
1059         struct recvendsyncarg *resa = arg2;
1060
1061         return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1062 }
1063
1064 static void
1065 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1066 {
1067         dsl_dataset_t *ds = arg1;
1068         struct recvendsyncarg *resa = arg2;
1069
1070         dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
1071
1072         /* set snapshot's creation time and guid */
1073         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1074         ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1075         ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1076         ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1077
1078         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1079         ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1080 }
1081
1082 int
1083 dmu_recv_end(dmu_recv_cookie_t *drc)
1084 {
1085         struct recvendsyncarg resa;
1086         dsl_dataset_t *ds = drc->drc_logical_ds;
1087         int err;
1088
1089         /*
1090          * XXX hack; seems the ds is still dirty and
1091          * dsl_pool_zil_clean() expects it to have a ds_user_ptr
1092          * (and zil), but clone_swap() can close it.
1093          */
1094         txg_wait_synced(ds->ds_dir->dd_pool, 0);
1095
1096         if (ds != drc->drc_real_ds) {
1097                 /* we are doing an online recv */
1098                 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1099                         err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1100                             drc->drc_force);
1101                         if (err)
1102                                 dsl_dataset_disown(ds, dmu_recv_tag);
1103                 } else {
1104                         err = EBUSY;
1105                         dsl_dataset_rele(ds, dmu_recv_tag);
1106                 }
1107                 /* dsl_dataset_destroy() will disown the ds */
1108                 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1109                     B_FALSE);
1110                 mutex_exit(&drc->drc_logical_ds->ds_recvlock);
1111                 if (err)
1112                         return (err);
1113         }
1114
1115         resa.creation_time = drc->drc_drrb->drr_creation_time;
1116         resa.toguid = drc->drc_drrb->drr_toguid;
1117         resa.tosnap = drc->drc_tosnap;
1118
1119         err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1120             recv_end_check, recv_end_sync, ds, &resa, 3);
1121         if (err) {
1122                 if (drc->drc_newfs) {
1123                         ASSERT(ds == drc->drc_real_ds);
1124                         (void) dsl_dataset_destroy(ds, dmu_recv_tag,
1125                             B_FALSE);
1126                         return (err);
1127                 } else {
1128                         (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
1129                 }
1130         }
1131
1132         /* release the hold from dmu_recv_begin */
1133         dsl_dataset_disown(ds, dmu_recv_tag);
1134         return (err);
1135 }