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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  27  */
  28
  29 #include <sys/dmu.h>
  30 #include <sys/dmu_impl.h>
  31 #include <sys/dmu_tx.h>
  32 #include <sys/dbuf.h>
  33 #include <sys/dnode.h>
  34 #include <sys/zfs_context.h>
  35 #include <sys/dmu_objset.h>
  36 #include <sys/dmu_traverse.h>
  37 #include <sys/dsl_dataset.h>
  38 #include <sys/dsl_dir.h>
  39 #include <sys/dsl_prop.h>
  40 #include <sys/dsl_pool.h>
  41 #include <sys/dsl_synctask.h>
  42 #include <sys/spa_impl.h>
  43 #include <sys/zfs_ioctl.h>
  44 #include <sys/zap.h>
  45 #include <sys/zio_checksum.h>
  46 #include <sys/zfs_znode.h>
  47 #include <zfs_fletcher.h>
  48 #include <sys/avl.h>
  49 #include <sys/ddt.h>
  50 #include <sys/zfs_onexit.h>
  51 #include <sys/dmu_send.h>
  52 #include <sys/dsl_destroy.h>
  53 #include <sys/blkptr.h>
  54 #include <sys/dsl_bookmark.h>
  55 #include <sys/zfeature.h>
  56
  57 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
  58 int zfs_send_corrupt_data = B_FALSE;
  59
  60 static char *dmu_recv_tag = "dmu_recv_tag";
  61 static const char *recv_clone_name = "%recv";
  62
  63 typedef struct dump_bytes_io {
  64         dmu_sendarg_t   *dbi_dsp;
  65         void            *dbi_buf;
  66         int             dbi_len;
  67 } dump_bytes_io_t;
  68
  69 static void
  70 dump_bytes_cb(void *arg)
  71 {
  72         dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg;
  73         dmu_sendarg_t *dsp = dbi->dbi_dsp;
  74         dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
  75         ssize_t resid; /* have to get resid to get detailed errno */
  76         ASSERT0(dbi->dbi_len % 8);
  77
  78         fletcher_4_incremental_native(dbi->dbi_buf, dbi->dbi_len, &dsp->dsa_zc);
  79         dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
  80             (caddr_t)dbi->dbi_buf, dbi->dbi_len,
  81             0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
  82
  83         mutex_enter(&ds->ds_sendstream_lock);
  84         *dsp->dsa_off += dbi->dbi_len;
  85         mutex_exit(&ds->ds_sendstream_lock);
  86 }
  87
  88 static int
  89 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
  90 {
  91         dump_bytes_io_t dbi;
  92
  93         dbi.dbi_dsp = dsp;
  94         dbi.dbi_buf = buf;
  95         dbi.dbi_len = len;
  96
  97 #if defined(HAVE_LARGE_STACKS)
  98         dump_bytes_cb(&dbi);
  99 #else
 100         /*
 101          * The vn_rdwr() call is performed in a taskq to ensure that there is
 102          * always enough stack space to write safely to the target filesystem.
 103          * The ZIO_TYPE_FREE threads are used because there can be a lot of
 104          * them and they are used in vdev_file.c for a similar purpose.
 105          */
 106         spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE,
 107             ZIO_TASKQ_ISSUE, dump_bytes_cb, &dbi, TQ_SLEEP);
 108 #endif /* HAVE_LARGE_STACKS */
 109
 110         return (dsp->dsa_err);
 111 }
 112
 113 static int
 114 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
 115     uint64_t length)
 116 {
 117         struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
 118
 119         /*
 120          * When we receive a free record, dbuf_free_range() assumes
 121          * that the receiving system doesn't have any dbufs in the range
 122          * being freed.  This is always true because there is a one-record
 123          * constraint: we only send one WRITE record for any given
 124          * object+offset.  We know that the one-record constraint is
 125          * true because we always send data in increasing order by
 126          * object,offset.
 127          *
 128          * If the increasing-order constraint ever changes, we should find
 129          * another way to assert that the one-record constraint is still
 130          * satisfied.
 131          */
 132         ASSERT(object > dsp->dsa_last_data_object ||
 133             (object == dsp->dsa_last_data_object &&
 134             offset > dsp->dsa_last_data_offset));
 135
 136         /*
 137          * If we are doing a non-incremental send, then there can't
 138          * be any data in the dataset we're receiving into.  Therefore
 139          * a free record would simply be a no-op.  Save space by not
 140          * sending it to begin with.
 141          */
 142         if (!dsp->dsa_incremental)
 143                 return (0);
 144
 145         if (length != -1ULL && offset + length < offset)
 146                 length = -1ULL;
 147
 148         /*
 149          * If there is a pending op, but it's not PENDING_FREE, push it out,
 150          * since free block aggregation can only be done for blocks of the
 151          * same type (i.e., DRR_FREE records can only be aggregated with
 152          * other DRR_FREE records.  DRR_FREEOBJECTS records can only be
 153          * aggregated with other DRR_FREEOBJECTS records.
 154          */
 155         if (dsp->dsa_pending_op != PENDING_NONE &&
 156             dsp->dsa_pending_op != PENDING_FREE) {
 157                 if (dump_bytes(dsp, dsp->dsa_drr,
 158                     sizeof (dmu_replay_record_t)) != 0)
 159                         return (SET_ERROR(EINTR));
 160                 dsp->dsa_pending_op = PENDING_NONE;
 161         }
 162
 163         if (dsp->dsa_pending_op == PENDING_FREE) {
 164                 /*
 165                  * There should never be a PENDING_FREE if length is -1
 166                  * (because dump_dnode is the only place where this
 167                  * function is called with a -1, and only after flushing
 168                  * any pending record).
 169                  */
 170                 ASSERT(length != -1ULL);
 171                 /*
 172                  * Check to see whether this free block can be aggregated
 173                  * with pending one.
 174                  */
 175                 if (drrf->drr_object == object && drrf->drr_offset +
 176                     drrf->drr_length == offset) {
 177                         drrf->drr_length += length;
 178                         return (0);
 179                 } else {
 180                         /* not a continuation.  Push out pending record */
 181                         if (dump_bytes(dsp, dsp->dsa_drr,
 182                             sizeof (dmu_replay_record_t)) != 0)
 183                                 return (SET_ERROR(EINTR));
 184                         dsp->dsa_pending_op = PENDING_NONE;
 185                 }
 186         }
 187         /* create a FREE record and make it pending */
 188         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 189         dsp->dsa_drr->drr_type = DRR_FREE;
 190         drrf->drr_object = object;
 191         drrf->drr_offset = offset;
 192         drrf->drr_length = length;
 193         drrf->drr_toguid = dsp->dsa_toguid;
 194         if (length == -1ULL) {
 195                 if (dump_bytes(dsp, dsp->dsa_drr,
 196                     sizeof (dmu_replay_record_t)) != 0)
 197                         return (SET_ERROR(EINTR));
 198         } else {
 199                 dsp->dsa_pending_op = PENDING_FREE;
 200         }
 201
 202         return (0);
 203 }
 204
 205 static int
 206 dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type,
 207     uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
 208 {
 209         struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
 210
 211         /*
 212          * We send data in increasing object, offset order.
 213          * See comment in dump_free() for details.
 214          */
 215         ASSERT(object > dsp->dsa_last_data_object ||
 216             (object == dsp->dsa_last_data_object &&
 217             offset > dsp->dsa_last_data_offset));
 218         dsp->dsa_last_data_object = object;
 219         dsp->dsa_last_data_offset = offset + blksz - 1;
 220
 221         /*
 222          * If there is any kind of pending aggregation (currently either
 223          * a grouping of free objects or free blocks), push it out to
 224          * the stream, since aggregation can't be done across operations
 225          * of different types.
 226          */
 227         if (dsp->dsa_pending_op != PENDING_NONE) {
 228                 if (dump_bytes(dsp, dsp->dsa_drr,
 229                     sizeof (dmu_replay_record_t)) != 0)
 230                         return (SET_ERROR(EINTR));
 231                 dsp->dsa_pending_op = PENDING_NONE;
 232         }
 233         /* write a DATA record */
 234         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 235         dsp->dsa_drr->drr_type = DRR_WRITE;
 236         drrw->drr_object = object;
 237         drrw->drr_type = type;
 238         drrw->drr_offset = offset;
 239         drrw->drr_length = blksz;
 240         drrw->drr_toguid = dsp->dsa_toguid;
 241         if (bp == NULL || BP_IS_EMBEDDED(bp)) {
 242                 /*
 243                  * There's no pre-computed checksum for partial-block
 244                  * writes or embedded BP's, so (like
 245                  * fletcher4-checkummed blocks) userland will have to
 246                  * compute a dedup-capable checksum itself.
 247                  */
 248                 drrw->drr_checksumtype = ZIO_CHECKSUM_OFF;
 249         } else {
 250                 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
 251                 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
 252                         drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
 253                 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
 254                 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
 255                 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
 256                 drrw->drr_key.ddk_cksum = bp->blk_cksum;
 257         }
 258
 259         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 260                 return (SET_ERROR(EINTR));
 261         if (dump_bytes(dsp, data, blksz) != 0)
 262                 return (SET_ERROR(EINTR));
 263         return (0);
 264 }
 265
 266 static int
 267 dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
 268     int blksz, const blkptr_t *bp)
 269 {
 270         char buf[BPE_PAYLOAD_SIZE];
 271         struct drr_write_embedded *drrw =
 272             &(dsp->dsa_drr->drr_u.drr_write_embedded);
 273
 274         if (dsp->dsa_pending_op != PENDING_NONE) {
 275                 if (dump_bytes(dsp, dsp->dsa_drr,
 276                     sizeof (dmu_replay_record_t)) != 0)
 277                         return (EINTR);
 278                 dsp->dsa_pending_op = PENDING_NONE;
 279         }
 280
 281         ASSERT(BP_IS_EMBEDDED(bp));
 282
 283         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 284         dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED;
 285         drrw->drr_object = object;
 286         drrw->drr_offset = offset;
 287         drrw->drr_length = blksz;
 288         drrw->drr_toguid = dsp->dsa_toguid;
 289         drrw->drr_compression = BP_GET_COMPRESS(bp);
 290         drrw->drr_etype = BPE_GET_ETYPE(bp);
 291         drrw->drr_lsize = BPE_GET_LSIZE(bp);
 292         drrw->drr_psize = BPE_GET_PSIZE(bp);
 293
 294         decode_embedded_bp_compressed(bp, buf);
 295
 296         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 297                 return (EINTR);
 298         if (dump_bytes(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0)
 299                 return (EINTR);
 300         return (0);
 301 }
 302
 303 static int
 304 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
 305 {
 306         struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
 307
 308         if (dsp->dsa_pending_op != PENDING_NONE) {
 309                 if (dump_bytes(dsp, dsp->dsa_drr,
 310                     sizeof (dmu_replay_record_t)) != 0)
 311                         return (SET_ERROR(EINTR));
 312                 dsp->dsa_pending_op = PENDING_NONE;
 313         }
 314
 315         /* write a SPILL record */
 316         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 317         dsp->dsa_drr->drr_type = DRR_SPILL;
 318         drrs->drr_object = object;
 319         drrs->drr_length = blksz;
 320         drrs->drr_toguid = dsp->dsa_toguid;
 321
 322         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
 323                 return (SET_ERROR(EINTR));
 324         if (dump_bytes(dsp, data, blksz))
 325                 return (SET_ERROR(EINTR));
 326         return (0);
 327 }
 328
 329 static int
 330 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
 331 {
 332         struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
 333
 334         /* See comment in dump_free(). */
 335         if (!dsp->dsa_incremental)
 336                 return (0);
 337
 338         /*
 339          * If there is a pending op, but it's not PENDING_FREEOBJECTS,
 340          * push it out, since free block aggregation can only be done for
 341          * blocks of the same type (i.e., DRR_FREE records can only be
 342          * aggregated with other DRR_FREE records.  DRR_FREEOBJECTS records
 343          * can only be aggregated with other DRR_FREEOBJECTS records.
 344          */
 345         if (dsp->dsa_pending_op != PENDING_NONE &&
 346             dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
 347                 if (dump_bytes(dsp, dsp->dsa_drr,
 348                     sizeof (dmu_replay_record_t)) != 0)
 349                         return (SET_ERROR(EINTR));
 350                 dsp->dsa_pending_op = PENDING_NONE;
 351         }
 352         if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
 353                 /*
 354                  * See whether this free object array can be aggregated
 355                  * with pending one
 356                  */
 357                 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
 358                         drrfo->drr_numobjs += numobjs;
 359                         return (0);
 360                 } else {
 361                         /* can't be aggregated.  Push out pending record */
 362                         if (dump_bytes(dsp, dsp->dsa_drr,
 363                             sizeof (dmu_replay_record_t)) != 0)
 364                                 return (SET_ERROR(EINTR));
 365                         dsp->dsa_pending_op = PENDING_NONE;
 366                 }
 367         }
 368
 369         /* write a FREEOBJECTS record */
 370         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 371         dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
 372         drrfo->drr_firstobj = firstobj;
 373         drrfo->drr_numobjs = numobjs;
 374         drrfo->drr_toguid = dsp->dsa_toguid;
 375
 376         dsp->dsa_pending_op = PENDING_FREEOBJECTS;
 377
 378         return (0);
 379 }
 380
 381 static int
 382 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
 383 {
 384         struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
 385
 386         if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
 387                 return (dump_freeobjects(dsp, object, 1));
 388
 389         if (dsp->dsa_pending_op != PENDING_NONE) {
 390                 if (dump_bytes(dsp, dsp->dsa_drr,
 391                     sizeof (dmu_replay_record_t)) != 0)
 392                         return (SET_ERROR(EINTR));
 393                 dsp->dsa_pending_op = PENDING_NONE;
 394         }
 395
 396         /* write an OBJECT record */
 397         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 398         dsp->dsa_drr->drr_type = DRR_OBJECT;
 399         drro->drr_object = object;
 400         drro->drr_type = dnp->dn_type;
 401         drro->drr_bonustype = dnp->dn_bonustype;
 402         drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
 403         drro->drr_bonuslen = dnp->dn_bonuslen;
 404         drro->drr_checksumtype = dnp->dn_checksum;
 405         drro->drr_compress = dnp->dn_compress;
 406         drro->drr_toguid = dsp->dsa_toguid;
 407
 408         if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
 409             drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE)
 410                 drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE;
 411
 412         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 413                 return (SET_ERROR(EINTR));
 414
 415         if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
 416                 return (SET_ERROR(EINTR));
 417
 418         /* Free anything past the end of the file. */
 419         if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
 420             (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
 421                 return (SET_ERROR(EINTR));
 422         if (dsp->dsa_err != 0)
 423                 return (SET_ERROR(EINTR));
 424         return (0);
 425 }
 426
 427 static boolean_t
 428 backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp)
 429 {
 430         if (!BP_IS_EMBEDDED(bp))
 431                 return (B_FALSE);
 432
 433         /*
 434          * Compression function must be legacy, or explicitly enabled.
 435          */
 436         if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS &&
 437             !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4)))
 438                 return (B_FALSE);
 439
 440         /*
 441          * Embed type must be explicitly enabled.
 442          */
 443         switch (BPE_GET_ETYPE(bp)) {
 444         case BP_EMBEDDED_TYPE_DATA:
 445                 if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)
 446                         return (B_TRUE);
 447                 break;
 448         default:
 449                 return (B_FALSE);
 450         }
 451         return (B_FALSE);
 452 }
 453
 454 #define BP_SPAN(dnp, level) \
 455         (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
 456         (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
 457
 458 /* ARGSUSED */
 459 static int
 460 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 461     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
 462 {
 463         dmu_sendarg_t *dsp = arg;
 464         dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
 465         int err = 0;
 466
 467         if (issig(JUSTLOOKING) && issig(FORREAL))
 468                 return (SET_ERROR(EINTR));
 469
 470         if (zb->zb_object != DMU_META_DNODE_OBJECT &&
 471             DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
 472                 return (0);
 473         } else if (zb->zb_level == ZB_ZIL_LEVEL) {
 474                 /*
 475                  * If we are sending a non-snapshot (which is allowed on
 476                  * read-only pools), it may have a ZIL, which must be ignored.
 477                  */
 478                 return (0);
 479         } else if (BP_IS_HOLE(bp) &&
 480             zb->zb_object == DMU_META_DNODE_OBJECT) {
 481                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 482                 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
 483                 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
 484         } else if (BP_IS_HOLE(bp)) {
 485                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 486                 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
 487         } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
 488                 return (0);
 489         } else if (type == DMU_OT_DNODE) {
 490                 dnode_phys_t *blk;
 491                 int i;
 492                 int blksz = BP_GET_LSIZE(bp);
 493                 arc_flags_t aflags = ARC_FLAG_WAIT;
 494                 arc_buf_t *abuf;
 495
 496                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 497                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 498                     &aflags, zb) != 0)
 499                         return (SET_ERROR(EIO));
 500
 501                 blk = abuf->b_data;
 502                 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
 503                         uint64_t dnobj = (zb->zb_blkid <<
 504                             (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
 505                         err = dump_dnode(dsp, dnobj, blk+i);
 506                         if (err != 0)
 507                                 break;
 508                 }
 509                 (void) arc_buf_remove_ref(abuf, &abuf);
 510         } else if (type == DMU_OT_SA) {
 511                 arc_flags_t aflags = ARC_FLAG_WAIT;
 512                 arc_buf_t *abuf;
 513                 int blksz = BP_GET_LSIZE(bp);
 514
 515                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 516                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 517                     &aflags, zb) != 0)
 518                         return (SET_ERROR(EIO));
 519
 520                 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
 521                 (void) arc_buf_remove_ref(abuf, &abuf);
 522         } else if (backup_do_embed(dsp, bp)) {
 523                 /* it's an embedded level-0 block of a regular object */
 524                 int blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
 525                 err = dump_write_embedded(dsp, zb->zb_object,
 526                     zb->zb_blkid * blksz, blksz, bp);
 527         } else { /* it's a level-0 block of a regular object */
 528                 uint64_t offset;
 529                 arc_flags_t aflags = ARC_FLAG_WAIT;
 530                 arc_buf_t *abuf;
 531                 int blksz = BP_GET_LSIZE(bp);
 532
 533                 ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
 534                 ASSERT0(zb->zb_level);
 535                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 536                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 537                     &aflags, zb) != 0) {
 538                         if (zfs_send_corrupt_data) {
 539                                 uint64_t *ptr;
 540                                 /* Send a block filled with 0x"zfs badd bloc" */
 541                                 abuf = arc_buf_alloc(spa, blksz, &abuf,
 542                                     ARC_BUFC_DATA);
 543                                 for (ptr = abuf->b_data;
 544                                     (char *)ptr < (char *)abuf->b_data + blksz;
 545                                     ptr++)
 546                                         *ptr = 0x2f5baddb10cULL;
 547                         } else {
 548                                 return (SET_ERROR(EIO));
 549                         }
 550                 }
 551
 552                 offset = zb->zb_blkid * blksz;
 553
 554                 if (!(dsp->dsa_featureflags &
 555                     DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
 556                     blksz > SPA_OLD_MAXBLOCKSIZE) {
 557                         char *buf = abuf->b_data;
 558                         while (blksz > 0 && err == 0) {
 559                                 int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE);
 560                                 err = dump_write(dsp, type, zb->zb_object,
 561                                     offset, n, NULL, buf);
 562                                 offset += n;
 563                                 buf += n;
 564                                 blksz -= n;
 565                         }
 566                 } else {
 567                         err = dump_write(dsp, type, zb->zb_object,
 568                             offset, blksz, bp, abuf->b_data);
 569                 }
 570                 (void) arc_buf_remove_ref(abuf, &abuf);
 571         }
 572
 573         ASSERT(err == 0 || err == EINTR);
 574         return (err);
 575 }
 576
 577 /*
 578  * Releases dp using the specified tag.
 579  */
 580 static int
 581 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 582     zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok,
 583     boolean_t large_block_ok, int outfd, vnode_t *vp, offset_t *off)
 584 {
 585         objset_t *os;
 586         dmu_replay_record_t *drr;
 587         dmu_sendarg_t *dsp;
 588         int err;
 589         uint64_t fromtxg = 0;
 590         uint64_t featureflags = 0;
 591
 592         err = dmu_objset_from_ds(ds, &os);
 593         if (err != 0) {
 594                 dsl_pool_rele(dp, tag);
 595                 return (err);
 596         }
 597
 598         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
 599         drr->drr_type = DRR_BEGIN;
 600         drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
 601         DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
 602             DMU_SUBSTREAM);
 603
 604 #ifdef _KERNEL
 605         if (dmu_objset_type(os) == DMU_OST_ZFS) {
 606                 uint64_t version;
 607                 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
 608                         kmem_free(drr, sizeof (dmu_replay_record_t));
 609                         dsl_pool_rele(dp, tag);
 610                         return (SET_ERROR(EINVAL));
 611                 }
 612                 if (version >= ZPL_VERSION_SA) {
 613                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
 614                 }
 615         }
 616 #endif
 617
 618         if (large_block_ok && ds->ds_feature_inuse[SPA_FEATURE_LARGE_BLOCKS])
 619                 featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS;
 620         if (embedok &&
 621             spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) {
 622                 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA;
 623                 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
 624                         featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4;
 625         } else {
 626                 embedok = B_FALSE;
 627         }
 628
 629         DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo,
 630             featureflags);
 631
 632         drr->drr_u.drr_begin.drr_creation_time =
 633             dsl_dataset_phys(ds)->ds_creation_time;
 634         drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
 635         if (is_clone)
 636                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
 637         drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(ds)->ds_guid;
 638         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 639                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
 640
 641         if (fromzb != NULL) {
 642                 drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid;
 643                 fromtxg = fromzb->zbm_creation_txg;
 644         }
 645         dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
 646         if (!ds->ds_is_snapshot) {
 647                 (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
 648                     sizeof (drr->drr_u.drr_begin.drr_toname));
 649         }
 650
 651         dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
 652
 653         dsp->dsa_drr = drr;
 654         dsp->dsa_vp = vp;
 655         dsp->dsa_outfd = outfd;
 656         dsp->dsa_proc = curproc;
 657         dsp->dsa_os = os;
 658         dsp->dsa_off = off;
 659         dsp->dsa_toguid = dsl_dataset_phys(ds)->ds_guid;
 660         ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
 661         dsp->dsa_pending_op = PENDING_NONE;
 662         dsp->dsa_incremental = (fromzb != NULL);
 663         dsp->dsa_featureflags = featureflags;
 664
 665         mutex_enter(&ds->ds_sendstream_lock);
 666         list_insert_head(&ds->ds_sendstreams, dsp);
 667         mutex_exit(&ds->ds_sendstream_lock);
 668
 669         dsl_dataset_long_hold(ds, FTAG);
 670         dsl_pool_rele(dp, tag);
 671
 672         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 673                 err = dsp->dsa_err;
 674                 goto out;
 675         }
 676
 677         err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
 678             backup_cb, dsp);
 679
 680         if (dsp->dsa_pending_op != PENDING_NONE)
 681                 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
 682                         err = SET_ERROR(EINTR);
 683
 684         if (err != 0) {
 685                 if (err == EINTR && dsp->dsa_err != 0)
 686                         err = dsp->dsa_err;
 687                 goto out;
 688         }
 689
 690         bzero(drr, sizeof (dmu_replay_record_t));
 691         drr->drr_type = DRR_END;
 692         drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
 693         drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
 694
 695         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 696                 err = dsp->dsa_err;
 697                 goto out;
 698         }
 699
 700 out:
 701         mutex_enter(&ds->ds_sendstream_lock);
 702         list_remove(&ds->ds_sendstreams, dsp);
 703         mutex_exit(&ds->ds_sendstream_lock);
 704
 705         kmem_free(drr, sizeof (dmu_replay_record_t));
 706         kmem_free(dsp, sizeof (dmu_sendarg_t));
 707
 708         dsl_dataset_long_rele(ds, FTAG);
 709
 710         return (err);
 711 }
 712
 713 int
 714 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
 715     boolean_t embedok, boolean_t large_block_ok,
 716     int outfd, vnode_t *vp, offset_t *off)
 717 {
 718         dsl_pool_t *dp;
 719         dsl_dataset_t *ds;
 720         dsl_dataset_t *fromds = NULL;
 721         int err;
 722
 723         err = dsl_pool_hold(pool, FTAG, &dp);
 724         if (err != 0)
 725                 return (err);
 726
 727         err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
 728         if (err != 0) {
 729                 dsl_pool_rele(dp, FTAG);
 730                 return (err);
 731         }
 732
 733         if (fromsnap != 0) {
 734                 zfs_bookmark_phys_t zb;
 735                 boolean_t is_clone;
 736
 737                 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
 738                 if (err != 0) {
 739                         dsl_dataset_rele(ds, FTAG);
 740                         dsl_pool_rele(dp, FTAG);
 741                         return (err);
 742                 }
 743                 if (!dsl_dataset_is_before(ds, fromds, 0))
 744                         err = SET_ERROR(EXDEV);
 745                 zb.zbm_creation_time =
 746                     dsl_dataset_phys(fromds)->ds_creation_time;
 747                 zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg;
 748                 zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid;
 749                 is_clone = (fromds->ds_dir != ds->ds_dir);
 750                 dsl_dataset_rele(fromds, FTAG);
 751                 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
 752                     embedok, large_block_ok, outfd, vp, off);
 753         } else {
 754                 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
 755                     embedok, large_block_ok, outfd, vp, off);
 756         }
 757         dsl_dataset_rele(ds, FTAG);
 758         return (err);
 759 }
 760
 761 int
 762 dmu_send(const char *tosnap, const char *fromsnap,
 763     boolean_t embedok, boolean_t large_block_ok,
 764     int outfd, vnode_t *vp, offset_t *off)
 765 {
 766         dsl_pool_t *dp;
 767         dsl_dataset_t *ds;
 768         int err;
 769         boolean_t owned = B_FALSE;
 770
 771         if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
 772                 return (SET_ERROR(EINVAL));
 773
 774         err = dsl_pool_hold(tosnap, FTAG, &dp);
 775         if (err != 0)
 776                 return (err);
 777
 778         if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) {
 779                 /*
 780                  * We are sending a filesystem or volume.  Ensure
 781                  * that it doesn't change by owning the dataset.
 782                  */
 783                 err = dsl_dataset_own(dp, tosnap, FTAG, &ds);
 784                 owned = B_TRUE;
 785         } else {
 786                 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
 787         }
 788         if (err != 0) {
 789                 dsl_pool_rele(dp, FTAG);
 790                 return (err);
 791         }
 792
 793         if (fromsnap != NULL) {
 794                 zfs_bookmark_phys_t zb;
 795                 boolean_t is_clone = B_FALSE;
 796                 int fsnamelen = strchr(tosnap, '@') - tosnap;
 797
 798                 /*
 799                  * If the fromsnap is in a different filesystem, then
 800                  * mark the send stream as a clone.
 801                  */
 802                 if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
 803                     (fromsnap[fsnamelen] != '@' &&
 804                     fromsnap[fsnamelen] != '#')) {
 805                         is_clone = B_TRUE;
 806                 }
 807
 808                 if (strchr(fromsnap, '@')) {
 809                         dsl_dataset_t *fromds;
 810                         err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
 811                         if (err == 0) {
 812                                 if (!dsl_dataset_is_before(ds, fromds, 0))
 813                                         err = SET_ERROR(EXDEV);
 814                                 zb.zbm_creation_time =
 815                                     dsl_dataset_phys(fromds)->ds_creation_time;
 816                                 zb.zbm_creation_txg =
 817                                     dsl_dataset_phys(fromds)->ds_creation_txg;
 818                                 zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid;
 819                                 is_clone = (ds->ds_dir != fromds->ds_dir);
 820                                 dsl_dataset_rele(fromds, FTAG);
 821                         }
 822                 } else {
 823                         err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb);
 824                 }
 825                 if (err != 0) {
 826                         dsl_dataset_rele(ds, FTAG);
 827                         dsl_pool_rele(dp, FTAG);
 828                         return (err);
 829                 }
 830                 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
 831                     embedok, large_block_ok, outfd, vp, off);
 832         } else {
 833                 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
 834                     embedok, large_block_ok, outfd, vp, off);
 835         }
 836         if (owned)
 837                 dsl_dataset_disown(ds, FTAG);
 838         else
 839                 dsl_dataset_rele(ds, FTAG);
 840         return (err);
 841 }
 842
 843 static int
 844 dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t size,
 845     uint64_t *sizep)
 846 {
 847         int err;
 848         /*
 849          * Assume that space (both on-disk and in-stream) is dominated by
 850          * data.  We will adjust for indirect blocks and the copies property,
 851          * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
 852          */
 853
 854         /*
 855          * Subtract out approximate space used by indirect blocks.
 856          * Assume most space is used by data blocks (non-indirect, non-dnode).
 857          * Assume all blocks are recordsize.  Assume ditto blocks and
 858          * internal fragmentation counter out compression.
 859          *
 860          * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
 861          * block, which we observe in practice.
 862          */
 863         uint64_t recordsize;
 864         err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
 865         if (err != 0)
 866                 return (err);
 867         size -= size / recordsize * sizeof (blkptr_t);
 868
 869         /* Add in the space for the record associated with each block. */
 870         size += size / recordsize * sizeof (dmu_replay_record_t);
 871
 872         *sizep = size;
 873
 874         return (0);
 875 }
 876
 877 int
 878 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
 879 {
 880         int err;
 881         uint64_t size;
 882
 883         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 884
 885         /* tosnap must be a snapshot */
 886         if (!ds->ds_is_snapshot)
 887                 return (SET_ERROR(EINVAL));
 888
 889         /* fromsnap, if provided, must be a snapshot */
 890         if (fromds != NULL && !fromds->ds_is_snapshot)
 891                 return (SET_ERROR(EINVAL));
 892
 893         /*
 894          * fromsnap must be an earlier snapshot from the same fs as tosnap,
 895          * or the origin's fs.
 896          */
 897         if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
 898                 return (SET_ERROR(EXDEV));
 899
 900         /* Get uncompressed size estimate of changed data. */
 901         if (fromds == NULL) {
 902                 size = dsl_dataset_phys(ds)->ds_uncompressed_bytes;
 903         } else {
 904                 uint64_t used, comp;
 905                 err = dsl_dataset_space_written(fromds, ds,
 906                     &used, &comp, &size);
 907                 if (err != 0)
 908                         return (err);
 909         }
 910
 911         err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep);
 912         return (err);
 913 }
 914
 915 /*
 916  * Simple callback used to traverse the blocks of a snapshot and sum their
 917  * uncompressed size
 918  */
 919 /* ARGSUSED */
 920 static int
 921 dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 922     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
 923 {
 924         uint64_t *spaceptr = arg;
 925         if (bp != NULL && !BP_IS_HOLE(bp)) {
 926                 *spaceptr += BP_GET_UCSIZE(bp);
 927         }
 928         return (0);
 929 }
 930
 931 /*
 932  * Given a desination snapshot and a TXG, calculate the approximate size of a
 933  * send stream sent from that TXG. from_txg may be zero, indicating that the
 934  * whole snapshot will be sent.
 935  */
 936 int
 937 dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg,
 938     uint64_t *sizep)
 939 {
 940         int err;
 941         uint64_t size = 0;
 942
 943         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 944
 945         /* tosnap must be a snapshot */
 946         if (!dsl_dataset_is_snapshot(ds))
 947                 return (SET_ERROR(EINVAL));
 948
 949         /* verify that from_txg is before the provided snapshot was taken */
 950         if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) {
 951                 return (SET_ERROR(EXDEV));
 952         }
 953         /*
 954          * traverse the blocks of the snapshot with birth times after
 955          * from_txg, summing their uncompressed size
 956          */
 957         err = traverse_dataset(ds, from_txg, TRAVERSE_POST,
 958             dmu_calculate_send_traversal, &size);
 959         if (err)
 960                 return (err);
 961
 962         err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep);
 963         return (err);
 964 }
 965
 966 typedef struct dmu_recv_begin_arg {
 967         const char *drba_origin;
 968         dmu_recv_cookie_t *drba_cookie;
 969         cred_t *drba_cred;
 970         uint64_t drba_snapobj;
 971 } dmu_recv_begin_arg_t;
 972
 973 static int
 974 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
 975     uint64_t fromguid)
 976 {
 977         uint64_t val;
 978         int error;
 979         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 980
 981         /* temporary clone name must not exist */
 982         error = zap_lookup(dp->dp_meta_objset,
 983             dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name,
 984             8, 1, &val);
 985         if (error != ENOENT)
 986                 return (error == 0 ? EBUSY : error);
 987
 988         /* new snapshot name must not exist */
 989         error = zap_lookup(dp->dp_meta_objset,
 990             dsl_dataset_phys(ds)->ds_snapnames_zapobj,
 991             drba->drba_cookie->drc_tosnap, 8, 1, &val);
 992         if (error != ENOENT)
 993                 return (error == 0 ? EEXIST : error);
 994
 995         /*
 996          * Check snapshot limit before receiving. We'll recheck again at the
 997          * end, but might as well abort before receiving if we're already over
 998          * the limit.
 999          *
1000          * Note that we do not check the file system limit with
1001          * dsl_dir_fscount_check because the temporary %clones don't count
1002          * against that limit.
1003          */
1004         error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT,
1005             NULL, drba->drba_cred);
1006         if (error != 0)
1007                 return (error);
1008
1009         if (fromguid != 0) {
1010                 dsl_dataset_t *snap;
1011                 uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1012
1013                 /* Find snapshot in this dir that matches fromguid. */
1014                 while (obj != 0) {
1015                         error = dsl_dataset_hold_obj(dp, obj, FTAG,
1016                             &snap);
1017                         if (error != 0)
1018                                 return (SET_ERROR(ENODEV));
1019                         if (snap->ds_dir != ds->ds_dir) {
1020                                 dsl_dataset_rele(snap, FTAG);
1021                                 return (SET_ERROR(ENODEV));
1022                         }
1023                         if (dsl_dataset_phys(snap)->ds_guid == fromguid)
1024                                 break;
1025                         obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
1026                         dsl_dataset_rele(snap, FTAG);
1027                 }
1028                 if (obj == 0)
1029                         return (SET_ERROR(ENODEV));
1030
1031                 if (drba->drba_cookie->drc_force) {
1032                         drba->drba_snapobj = obj;
1033                 } else {
1034                         /*
1035                          * If we are not forcing, there must be no
1036                          * changes since fromsnap.
1037                          */
1038                         if (dsl_dataset_modified_since_snap(ds, snap)) {
1039                                 dsl_dataset_rele(snap, FTAG);
1040                                 return (SET_ERROR(ETXTBSY));
1041                         }
1042                         drba->drba_snapobj = ds->ds_prev->ds_object;
1043                 }
1044
1045                 dsl_dataset_rele(snap, FTAG);
1046         } else {
1047                 /* if full, then must be forced */
1048                 if (!drba->drba_cookie->drc_force)
1049                         return (SET_ERROR(EEXIST));
1050                 /* start from $ORIGIN@$ORIGIN, if supported */
1051                 drba->drba_snapobj = dp->dp_origin_snap != NULL ?
1052                     dp->dp_origin_snap->ds_object : 0;
1053         }
1054
1055         return (0);
1056
1057 }
1058
1059 static int
1060 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
1061 {
1062         dmu_recv_begin_arg_t *drba = arg;
1063         dsl_pool_t *dp = dmu_tx_pool(tx);
1064         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1065         uint64_t fromguid = drrb->drr_fromguid;
1066         int flags = drrb->drr_flags;
1067         int error;
1068         uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
1069         dsl_dataset_t *ds;
1070         const char *tofs = drba->drba_cookie->drc_tofs;
1071
1072         /* already checked */
1073         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
1074
1075         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
1076             DMU_COMPOUNDSTREAM ||
1077             drrb->drr_type >= DMU_OST_NUMTYPES ||
1078             ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
1079                 return (SET_ERROR(EINVAL));
1080
1081         /* Verify pool version supports SA if SA_SPILL feature set */
1082         if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
1083             spa_version(dp->dp_spa) < SPA_VERSION_SA)
1084                 return (SET_ERROR(ENOTSUP));
1085
1086         /*
1087          * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1088          * record to a plan WRITE record, so the pool must have the
1089          * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1090          * records.  Same with WRITE_EMBEDDED records that use LZ4 compression.
1091          */
1092         if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) &&
1093             !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA))
1094                 return (SET_ERROR(ENOTSUP));
1095         if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) &&
1096             !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
1097                 return (SET_ERROR(ENOTSUP));
1098
1099         /*
1100          * The receiving code doesn't know how to translate large blocks
1101          * to smaller ones, so the pool must have the LARGE_BLOCKS
1102          * feature enabled if the stream has LARGE_BLOCKS.
1103          */
1104         if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
1105             !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
1106                 return (SET_ERROR(ENOTSUP));
1107
1108         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1109         if (error == 0) {
1110                 /* target fs already exists; recv into temp clone */
1111
1112                 /* Can't recv a clone into an existing fs */
1113                 if (flags & DRR_FLAG_CLONE) {
1114                         dsl_dataset_rele(ds, FTAG);
1115                         return (SET_ERROR(EINVAL));
1116                 }
1117
1118                 error = recv_begin_check_existing_impl(drba, ds, fromguid);
1119                 dsl_dataset_rele(ds, FTAG);
1120         } else if (error == ENOENT) {
1121                 /* target fs does not exist; must be a full backup or clone */
1122                 char buf[MAXNAMELEN];
1123
1124                 /*
1125                  * If it's a non-clone incremental, we are missing the
1126                  * target fs, so fail the recv.
1127                  */
1128                 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
1129                         return (SET_ERROR(ENOENT));
1130
1131                 /* Open the parent of tofs */
1132                 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
1133                 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
1134                 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
1135                 if (error != 0)
1136                         return (error);
1137
1138                 /*
1139                  * Check filesystem and snapshot limits before receiving. We'll
1140                  * recheck snapshot limits again at the end (we create the
1141                  * filesystems and increment those counts during begin_sync).
1142                  */
1143                 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1144                     ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred);
1145                 if (error != 0) {
1146                         dsl_dataset_rele(ds, FTAG);
1147                         return (error);
1148                 }
1149
1150                 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1151                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred);
1152                 if (error != 0) {
1153                         dsl_dataset_rele(ds, FTAG);
1154                         return (error);
1155                 }
1156
1157                 if (drba->drba_origin != NULL) {
1158                         dsl_dataset_t *origin;
1159                         error = dsl_dataset_hold(dp, drba->drba_origin,
1160                             FTAG, &origin);
1161                         if (error != 0) {
1162                                 dsl_dataset_rele(ds, FTAG);
1163                                 return (error);
1164                         }
1165                         if (!origin->ds_is_snapshot) {
1166                                 dsl_dataset_rele(origin, FTAG);
1167                                 dsl_dataset_rele(ds, FTAG);
1168                                 return (SET_ERROR(EINVAL));
1169                         }
1170                         if (dsl_dataset_phys(origin)->ds_guid != fromguid) {
1171                                 dsl_dataset_rele(origin, FTAG);
1172                                 dsl_dataset_rele(ds, FTAG);
1173                                 return (SET_ERROR(ENODEV));
1174                         }
1175                         dsl_dataset_rele(origin, FTAG);
1176                 }
1177                 dsl_dataset_rele(ds, FTAG);
1178                 error = 0;
1179         }
1180         return (error);
1181 }
1182
1183 static void
1184 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
1185 {
1186         dmu_recv_begin_arg_t *drba = arg;
1187         dsl_pool_t *dp = dmu_tx_pool(tx);
1188         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1189         const char *tofs = drba->drba_cookie->drc_tofs;
1190         dsl_dataset_t *ds, *newds;
1191         uint64_t dsobj;
1192         int error;
1193         uint64_t crflags;
1194
1195         crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
1196             DS_FLAG_CI_DATASET : 0;
1197
1198         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1199         if (error == 0) {
1200                 /* create temporary clone */
1201                 dsl_dataset_t *snap = NULL;
1202                 if (drba->drba_snapobj != 0) {
1203                         VERIFY0(dsl_dataset_hold_obj(dp,
1204                             drba->drba_snapobj, FTAG, &snap));
1205                 }
1206                 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
1207                     snap, crflags, drba->drba_cred, tx);
1208                 dsl_dataset_rele(snap, FTAG);
1209                 dsl_dataset_rele(ds, FTAG);
1210         } else {
1211                 dsl_dir_t *dd;
1212                 const char *tail;
1213                 dsl_dataset_t *origin = NULL;
1214
1215                 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1216
1217                 if (drba->drba_origin != NULL) {
1218                         VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1219                             FTAG, &origin));
1220                 }
1221
1222                 /* Create new dataset. */
1223                 dsobj = dsl_dataset_create_sync(dd,
1224                     strrchr(tofs, '/') + 1,
1225                     origin, crflags, drba->drba_cred, tx);
1226                 if (origin != NULL)
1227                         dsl_dataset_rele(origin, FTAG);
1228                 dsl_dir_rele(dd, FTAG);
1229                 drba->drba_cookie->drc_newfs = B_TRUE;
1230         }
1231         VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1232
1233         dmu_buf_will_dirty(newds->ds_dbuf, tx);
1234         dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT;
1235
1236         /*
1237          * If we actually created a non-clone, we need to create the
1238          * objset in our new dataset.
1239          */
1240         if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1241                 (void) dmu_objset_create_impl(dp->dp_spa,
1242                     newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1243         }
1244
1245         drba->drba_cookie->drc_ds = newds;
1246
1247         spa_history_log_internal_ds(newds, "receive", tx, "");
1248 }
1249
1250 /*
1251  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1252  * succeeds; otherwise we will leak the holds on the datasets.
1253  */
1254 int
1255 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1256     boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1257 {
1258         dmu_recv_begin_arg_t drba = { 0 };
1259         dmu_replay_record_t *drr;
1260
1261         bzero(drc, sizeof (dmu_recv_cookie_t));
1262         drc->drc_drrb = drrb;
1263         drc->drc_tosnap = tosnap;
1264         drc->drc_tofs = tofs;
1265         drc->drc_force = force;
1266         drc->drc_cred = CRED();
1267
1268         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1269                 drc->drc_byteswap = B_TRUE;
1270         else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1271                 return (SET_ERROR(EINVAL));
1272
1273         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1274         drr->drr_type = DRR_BEGIN;
1275         drr->drr_u.drr_begin = *drc->drc_drrb;
1276         if (drc->drc_byteswap) {
1277                 fletcher_4_incremental_byteswap(drr,
1278                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1279         } else {
1280                 fletcher_4_incremental_native(drr,
1281                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1282         }
1283         kmem_free(drr, sizeof (dmu_replay_record_t));
1284
1285         if (drc->drc_byteswap) {
1286                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1287                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1288                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1289                 drrb->drr_type = BSWAP_32(drrb->drr_type);
1290                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1291                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1292         }
1293
1294         drba.drba_origin = origin;
1295         drba.drba_cookie = drc;
1296         drba.drba_cred = CRED();
1297
1298         return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1299             &drba, 5, ZFS_SPACE_CHECK_NORMAL));
1300 }
1301
1302 struct restorearg {
1303         int err;
1304         boolean_t byteswap;
1305         vnode_t *vp;
1306         char *buf;
1307         uint64_t voff;
1308         int bufsize; /* amount of memory allocated for buf */
1309         zio_cksum_t cksum;
1310         avl_tree_t *guid_to_ds_map;
1311 };
1312
1313 typedef struct guid_map_entry {
1314         uint64_t        guid;
1315         dsl_dataset_t   *gme_ds;
1316         avl_node_t      avlnode;
1317 } guid_map_entry_t;
1318
1319 static int
1320 guid_compare(const void *arg1, const void *arg2)
1321 {
1322         const guid_map_entry_t *gmep1 = arg1;
1323         const guid_map_entry_t *gmep2 = arg2;
1324
1325         if (gmep1->guid < gmep2->guid)
1326                 return (-1);
1327         else if (gmep1->guid > gmep2->guid)
1328                 return (1);
1329         return (0);
1330 }
1331
1332 static void
1333 free_guid_map_onexit(void *arg)
1334 {
1335         avl_tree_t *ca = arg;
1336         void *cookie = NULL;
1337         guid_map_entry_t *gmep;
1338
1339         while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1340                 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1341                 dsl_dataset_rele(gmep->gme_ds, gmep);
1342                 kmem_free(gmep, sizeof (guid_map_entry_t));
1343         }
1344         avl_destroy(ca);
1345         kmem_free(ca, sizeof (avl_tree_t));
1346 }
1347
1348 static void *
1349 restore_read(struct restorearg *ra, int len, char *buf)
1350 {
1351         int done = 0;
1352
1353         if (buf == NULL)
1354                 buf = ra->buf;
1355
1356         /* some things will require 8-byte alignment, so everything must */
1357         ASSERT0(len % 8);
1358         ASSERT3U(len, <=, ra->bufsize);
1359
1360         while (done < len) {
1361                 ssize_t resid;
1362
1363                 ra->err = vn_rdwr(UIO_READ, ra->vp,
1364                     buf + done, len - done,
1365                     ra->voff, UIO_SYSSPACE, FAPPEND,
1366                     RLIM64_INFINITY, CRED(), &resid);
1367
1368                 if (resid == len - done)
1369                         ra->err = SET_ERROR(EINVAL);
1370                 ra->voff += len - done - resid;
1371                 done = len - resid;
1372                 if (ra->err != 0)
1373                         return (NULL);
1374         }
1375
1376         ASSERT3U(done, ==, len);
1377         if (ra->byteswap)
1378                 fletcher_4_incremental_byteswap(buf, len, &ra->cksum);
1379         else
1380                 fletcher_4_incremental_native(buf, len, &ra->cksum);
1381         return (buf);
1382 }
1383
1384 noinline static void
1385 backup_byteswap(dmu_replay_record_t *drr)
1386 {
1387 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1388 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1389         drr->drr_type = BSWAP_32(drr->drr_type);
1390         drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1391         switch (drr->drr_type) {
1392         case DRR_BEGIN:
1393                 DO64(drr_begin.drr_magic);
1394                 DO64(drr_begin.drr_versioninfo);
1395                 DO64(drr_begin.drr_creation_time);
1396                 DO32(drr_begin.drr_type);
1397                 DO32(drr_begin.drr_flags);
1398                 DO64(drr_begin.drr_toguid);
1399                 DO64(drr_begin.drr_fromguid);
1400                 break;
1401         case DRR_OBJECT:
1402                 DO64(drr_object.drr_object);
1403                 DO32(drr_object.drr_type);
1404                 DO32(drr_object.drr_bonustype);
1405                 DO32(drr_object.drr_blksz);
1406                 DO32(drr_object.drr_bonuslen);
1407                 DO64(drr_object.drr_toguid);
1408                 break;
1409         case DRR_FREEOBJECTS:
1410                 DO64(drr_freeobjects.drr_firstobj);
1411                 DO64(drr_freeobjects.drr_numobjs);
1412                 DO64(drr_freeobjects.drr_toguid);
1413                 break;
1414         case DRR_WRITE:
1415                 DO64(drr_write.drr_object);
1416                 DO32(drr_write.drr_type);
1417                 DO64(drr_write.drr_offset);
1418                 DO64(drr_write.drr_length);
1419                 DO64(drr_write.drr_toguid);
1420                 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1421                 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1422                 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1423                 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1424                 DO64(drr_write.drr_key.ddk_prop);
1425                 break;
1426         case DRR_WRITE_BYREF:
1427                 DO64(drr_write_byref.drr_object);
1428                 DO64(drr_write_byref.drr_offset);
1429                 DO64(drr_write_byref.drr_length);
1430                 DO64(drr_write_byref.drr_toguid);
1431                 DO64(drr_write_byref.drr_refguid);
1432                 DO64(drr_write_byref.drr_refobject);
1433                 DO64(drr_write_byref.drr_refoffset);
1434                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1435                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1436                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1437                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1438                 DO64(drr_write_byref.drr_key.ddk_prop);
1439                 break;
1440         case DRR_WRITE_EMBEDDED:
1441                 DO64(drr_write_embedded.drr_object);
1442                 DO64(drr_write_embedded.drr_offset);
1443                 DO64(drr_write_embedded.drr_length);
1444                 DO64(drr_write_embedded.drr_toguid);
1445                 DO32(drr_write_embedded.drr_lsize);
1446                 DO32(drr_write_embedded.drr_psize);
1447                 break;
1448         case DRR_FREE:
1449                 DO64(drr_free.drr_object);
1450                 DO64(drr_free.drr_offset);
1451                 DO64(drr_free.drr_length);
1452                 DO64(drr_free.drr_toguid);
1453                 break;
1454         case DRR_SPILL:
1455                 DO64(drr_spill.drr_object);
1456                 DO64(drr_spill.drr_length);
1457                 DO64(drr_spill.drr_toguid);
1458                 break;
1459         case DRR_END:
1460                 DO64(drr_end.drr_checksum.zc_word[0]);
1461                 DO64(drr_end.drr_checksum.zc_word[1]);
1462                 DO64(drr_end.drr_checksum.zc_word[2]);
1463                 DO64(drr_end.drr_checksum.zc_word[3]);
1464                 DO64(drr_end.drr_toguid);
1465                 break;
1466         default:
1467                 break;
1468         }
1469 #undef DO64
1470 #undef DO32
1471 }
1472
1473 static inline uint8_t
1474 deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size)
1475 {
1476         if (bonus_type == DMU_OT_SA) {
1477                 return (1);
1478         } else {
1479                 return (1 +
1480                     ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT));
1481         }
1482 }
1483
1484 noinline static int
1485 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1486 {
1487         dmu_object_info_t doi;
1488         dmu_tx_t *tx;
1489         void *data = NULL;
1490         uint64_t object;
1491         int err;
1492
1493         if (drro->drr_type == DMU_OT_NONE ||
1494             !DMU_OT_IS_VALID(drro->drr_type) ||
1495             !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1496             drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1497             drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1498             P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1499             drro->drr_blksz < SPA_MINBLOCKSIZE ||
1500             drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(os)) ||
1501             drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1502                 return (SET_ERROR(EINVAL));
1503         }
1504
1505         err = dmu_object_info(os, drro->drr_object, &doi);
1506
1507         if (err != 0 && err != ENOENT)
1508                 return (SET_ERROR(EINVAL));
1509         object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT;
1510
1511         if (drro->drr_bonuslen) {
1512                 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8), NULL);
1513                 if (ra->err != 0)
1514                         return (ra->err);
1515         }
1516
1517         /*
1518          * If we are losing blkptrs or changing the block size this must
1519          * be a new file instance.  We must clear out the previous file
1520          * contents before we can change this type of metadata in the dnode.
1521          */
1522         if (err == 0) {
1523                 int nblkptr;
1524
1525                 nblkptr = deduce_nblkptr(drro->drr_bonustype,
1526                     drro->drr_bonuslen);
1527
1528                 if (drro->drr_blksz != doi.doi_data_block_size ||
1529                     nblkptr < doi.doi_nblkptr) {
1530                         err = dmu_free_long_range(os, drro->drr_object,
1531                             0, DMU_OBJECT_END);
1532                         if (err != 0)
1533                                 return (SET_ERROR(EINVAL));
1534                 }
1535         }
1536
1537         tx = dmu_tx_create(os);
1538         dmu_tx_hold_bonus(tx, object);
1539         err = dmu_tx_assign(tx, TXG_WAIT);
1540         if (err != 0) {
1541                 dmu_tx_abort(tx);
1542                 return (err);
1543         }
1544
1545         if (object == DMU_NEW_OBJECT) {
1546                 /* currently free, want to be allocated */
1547                 err = dmu_object_claim(os, drro->drr_object,
1548                     drro->drr_type, drro->drr_blksz,
1549                     drro->drr_bonustype, drro->drr_bonuslen, tx);
1550         } else if (drro->drr_type != doi.doi_type ||
1551             drro->drr_blksz != doi.doi_data_block_size ||
1552             drro->drr_bonustype != doi.doi_bonus_type ||
1553             drro->drr_bonuslen != doi.doi_bonus_size) {
1554                 /* currently allocated, but with different properties */
1555                 err = dmu_object_reclaim(os, drro->drr_object,
1556                     drro->drr_type, drro->drr_blksz,
1557                     drro->drr_bonustype, drro->drr_bonuslen, tx);
1558         }
1559         if (err != 0) {
1560                 dmu_tx_commit(tx);
1561                 return (SET_ERROR(EINVAL));
1562         }
1563
1564         dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1565             tx);
1566         dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1567
1568         if (data != NULL) {
1569                 dmu_buf_t *db;
1570
1571                 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1572                 dmu_buf_will_dirty(db, tx);
1573
1574                 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1575                 bcopy(data, db->db_data, drro->drr_bonuslen);
1576                 if (ra->byteswap) {
1577                         dmu_object_byteswap_t byteswap =
1578                             DMU_OT_BYTESWAP(drro->drr_bonustype);
1579                         dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1580                             drro->drr_bonuslen);
1581                 }
1582                 dmu_buf_rele(db, FTAG);
1583         }
1584         dmu_tx_commit(tx);
1585         return (0);
1586 }
1587
1588 /* ARGSUSED */
1589 noinline static int
1590 restore_freeobjects(struct restorearg *ra, objset_t *os,
1591     struct drr_freeobjects *drrfo)
1592 {
1593         uint64_t obj;
1594
1595         if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1596                 return (SET_ERROR(EINVAL));
1597
1598         for (obj = drrfo->drr_firstobj;
1599             obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1600             (void) dmu_object_next(os, &obj, FALSE, 0)) {
1601                 int err;
1602
1603                 if (dmu_object_info(os, obj, NULL) != 0)
1604                         continue;
1605
1606                 err = dmu_free_long_object(os, obj);
1607                 if (err != 0)
1608                         return (err);
1609         }
1610         return (0);
1611 }
1612
1613 noinline static int
1614 restore_write(struct restorearg *ra, objset_t *os,
1615     struct drr_write *drrw)
1616 {
1617         dmu_tx_t *tx;
1618         dmu_buf_t *bonus;
1619         arc_buf_t *abuf;
1620         void *data;
1621         int err;
1622
1623         if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1624             !DMU_OT_IS_VALID(drrw->drr_type))
1625                 return (SET_ERROR(EINVAL));
1626
1627         if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1628                 return (SET_ERROR(EINVAL));
1629
1630         if (dmu_bonus_hold(os, drrw->drr_object, FTAG, &bonus) != 0)
1631                 return (SET_ERROR(EINVAL));
1632
1633         abuf = dmu_request_arcbuf(bonus, drrw->drr_length);
1634
1635         data = restore_read(ra, drrw->drr_length, abuf->b_data);
1636         if (data == NULL) {
1637                 dmu_return_arcbuf(abuf);
1638                 dmu_buf_rele(bonus, FTAG);
1639                 return (ra->err);
1640         }
1641
1642         tx = dmu_tx_create(os);
1643
1644         dmu_tx_hold_write(tx, drrw->drr_object,
1645             drrw->drr_offset, drrw->drr_length);
1646         err = dmu_tx_assign(tx, TXG_WAIT);
1647         if (err != 0) {
1648                 dmu_return_arcbuf(abuf);
1649                 dmu_buf_rele(bonus, FTAG);
1650                 dmu_tx_abort(tx);
1651                 return (err);
1652         }
1653         if (ra->byteswap) {
1654                 dmu_object_byteswap_t byteswap =
1655                     DMU_OT_BYTESWAP(drrw->drr_type);
1656                 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1657         }
1658         dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx);
1659         dmu_tx_commit(tx);
1660         dmu_buf_rele(bonus, FTAG);
1661         return (0);
1662 }
1663
1664 /*
1665  * Handle a DRR_WRITE_BYREF record.  This record is used in dedup'ed
1666  * streams to refer to a copy of the data that is already on the
1667  * system because it came in earlier in the stream.  This function
1668  * finds the earlier copy of the data, and uses that copy instead of
1669  * data from the stream to fulfill this write.
1670  */
1671 static int
1672 restore_write_byref(struct restorearg *ra, objset_t *os,
1673     struct drr_write_byref *drrwbr)
1674 {
1675         dmu_tx_t *tx;
1676         int err;
1677         guid_map_entry_t gmesrch;
1678         guid_map_entry_t *gmep;
1679         avl_index_t where;
1680         objset_t *ref_os = NULL;
1681         dmu_buf_t *dbp;
1682
1683         if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1684                 return (SET_ERROR(EINVAL));
1685
1686         /*
1687          * If the GUID of the referenced dataset is different from the
1688          * GUID of the target dataset, find the referenced dataset.
1689          */
1690         if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1691                 gmesrch.guid = drrwbr->drr_refguid;
1692                 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1693                     &where)) == NULL) {
1694                         return (SET_ERROR(EINVAL));
1695                 }
1696                 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1697                         return (SET_ERROR(EINVAL));
1698         } else {
1699                 ref_os = os;
1700         }
1701
1702         err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1703             drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1704         if (err != 0)
1705                 return (err);
1706
1707         tx = dmu_tx_create(os);
1708
1709         dmu_tx_hold_write(tx, drrwbr->drr_object,
1710             drrwbr->drr_offset, drrwbr->drr_length);
1711         err = dmu_tx_assign(tx, TXG_WAIT);
1712         if (err != 0) {
1713                 dmu_tx_abort(tx);
1714                 return (err);
1715         }
1716         dmu_write(os, drrwbr->drr_object,
1717             drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1718         dmu_buf_rele(dbp, FTAG);
1719         dmu_tx_commit(tx);
1720         return (0);
1721 }
1722
1723 static int
1724 restore_write_embedded(struct restorearg *ra, objset_t *os,
1725     struct drr_write_embedded *drrwnp)
1726 {
1727         dmu_tx_t *tx;
1728         int err;
1729         void *data;
1730
1731         if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset)
1732                 return (EINVAL);
1733
1734         if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE)
1735                 return (EINVAL);
1736
1737         if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES)
1738                 return (EINVAL);
1739         if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS)
1740                 return (EINVAL);
1741
1742         data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8), NULL);
1743         if (data == NULL)
1744                 return (ra->err);
1745
1746         tx = dmu_tx_create(os);
1747
1748         dmu_tx_hold_write(tx, drrwnp->drr_object,
1749             drrwnp->drr_offset, drrwnp->drr_length);
1750         err = dmu_tx_assign(tx, TXG_WAIT);
1751         if (err != 0) {
1752                 dmu_tx_abort(tx);
1753                 return (err);
1754         }
1755
1756         dmu_write_embedded(os, drrwnp->drr_object,
1757             drrwnp->drr_offset, data, drrwnp->drr_etype,
1758             drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize,
1759             ra->byteswap ^ ZFS_HOST_BYTEORDER, tx);
1760
1761         dmu_tx_commit(tx);
1762         return (0);
1763 }
1764
1765 static int
1766 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1767 {
1768         dmu_tx_t *tx;
1769         void *data;
1770         dmu_buf_t *db, *db_spill;
1771         int err;
1772
1773         if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1774             drrs->drr_length > spa_maxblocksize(dmu_objset_spa(os)))
1775                 return (SET_ERROR(EINVAL));
1776
1777         data = restore_read(ra, drrs->drr_length, NULL);
1778         if (data == NULL)
1779                 return (ra->err);
1780
1781         if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1782                 return (SET_ERROR(EINVAL));
1783
1784         VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1785         if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1786                 dmu_buf_rele(db, FTAG);
1787                 return (err);
1788         }
1789
1790         tx = dmu_tx_create(os);
1791
1792         dmu_tx_hold_spill(tx, db->db_object);
1793
1794         err = dmu_tx_assign(tx, TXG_WAIT);
1795         if (err != 0) {
1796                 dmu_buf_rele(db, FTAG);
1797                 dmu_buf_rele(db_spill, FTAG);
1798                 dmu_tx_abort(tx);
1799                 return (err);
1800         }
1801         dmu_buf_will_dirty(db_spill, tx);
1802
1803         if (db_spill->db_size < drrs->drr_length)
1804                 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1805                     drrs->drr_length, tx));
1806         bcopy(data, db_spill->db_data, drrs->drr_length);
1807
1808         dmu_buf_rele(db, FTAG);
1809         dmu_buf_rele(db_spill, FTAG);
1810
1811         dmu_tx_commit(tx);
1812         return (0);
1813 }
1814
1815 /* ARGSUSED */
1816 noinline static int
1817 restore_free(struct restorearg *ra, objset_t *os,
1818     struct drr_free *drrf)
1819 {
1820         int err;
1821
1822         if (drrf->drr_length != -1ULL &&
1823             drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1824                 return (SET_ERROR(EINVAL));
1825
1826         if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1827                 return (SET_ERROR(EINVAL));
1828
1829         err = dmu_free_long_range(os, drrf->drr_object,
1830             drrf->drr_offset, drrf->drr_length);
1831         return (err);
1832 }
1833
1834 /* used to destroy the drc_ds on error */
1835 static void
1836 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1837 {
1838         char name[MAXNAMELEN];
1839         dsl_dataset_name(drc->drc_ds, name);
1840         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1841         (void) dsl_destroy_head(name);
1842 }
1843
1844 /*
1845  * NB: callers *must* call dmu_recv_end() if this succeeds.
1846  */
1847 int
1848 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1849     int cleanup_fd, uint64_t *action_handlep)
1850 {
1851         struct restorearg ra = { 0 };
1852         dmu_replay_record_t *drr;
1853         objset_t *os;
1854         zio_cksum_t pcksum;
1855         int featureflags;
1856
1857         ra.byteswap = drc->drc_byteswap;
1858         ra.cksum = drc->drc_cksum;
1859         ra.vp = vp;
1860         ra.voff = *voffp;
1861         ra.bufsize = SPA_MAXBLOCKSIZE;
1862         ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1863
1864         /* these were verified in dmu_recv_begin */
1865         ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1866             DMU_SUBSTREAM);
1867         ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1868
1869         /*
1870          * Open the objset we are modifying.
1871          */
1872         VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1873
1874         ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT);
1875
1876         featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1877
1878         /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1879         if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1880                 minor_t minor;
1881
1882                 if (cleanup_fd == -1) {
1883                         ra.err = SET_ERROR(EBADF);
1884                         goto out;
1885                 }
1886                 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1887                 if (ra.err != 0) {
1888                         cleanup_fd = -1;
1889                         goto out;
1890                 }
1891
1892                 if (*action_handlep == 0) {
1893                         ra.guid_to_ds_map =
1894                             kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1895                         avl_create(ra.guid_to_ds_map, guid_compare,
1896                             sizeof (guid_map_entry_t),
1897                             offsetof(guid_map_entry_t, avlnode));
1898                         ra.err = zfs_onexit_add_cb(minor,
1899                             free_guid_map_onexit, ra.guid_to_ds_map,
1900                             action_handlep);
1901                         if (ra.err != 0)
1902                                 goto out;
1903                 } else {
1904                         ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1905                             (void **)&ra.guid_to_ds_map);
1906                         if (ra.err != 0)
1907                                 goto out;
1908                 }
1909
1910                 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1911         }
1912
1913         /*
1914          * Read records and process them.
1915          */
1916         pcksum = ra.cksum;
1917         while (ra.err == 0 &&
1918             NULL != (drr = restore_read(&ra, sizeof (*drr), NULL))) {
1919                 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1920                         ra.err = SET_ERROR(EINTR);
1921                         goto out;
1922                 }
1923
1924                 if (ra.byteswap)
1925                         backup_byteswap(drr);
1926
1927                 switch (drr->drr_type) {
1928                 case DRR_OBJECT:
1929                 {
1930                         /*
1931                          * We need to make a copy of the record header,
1932                          * because restore_{object,write} may need to
1933                          * restore_read(), which will invalidate drr.
1934                          */
1935                         struct drr_object drro = drr->drr_u.drr_object;
1936                         ra.err = restore_object(&ra, os, &drro);
1937                         break;
1938                 }
1939                 case DRR_FREEOBJECTS:
1940                 {
1941                         struct drr_freeobjects drrfo =
1942                             drr->drr_u.drr_freeobjects;
1943                         ra.err = restore_freeobjects(&ra, os, &drrfo);
1944                         break;
1945                 }
1946                 case DRR_WRITE:
1947                 {
1948                         struct drr_write drrw = drr->drr_u.drr_write;
1949                         ra.err = restore_write(&ra, os, &drrw);
1950                         break;
1951                 }
1952                 case DRR_WRITE_BYREF:
1953                 {
1954                         struct drr_write_byref drrwbr =
1955                             drr->drr_u.drr_write_byref;
1956                         ra.err = restore_write_byref(&ra, os, &drrwbr);
1957                         break;
1958                 }
1959                 case DRR_WRITE_EMBEDDED:
1960                 {
1961                         struct drr_write_embedded drrwe =
1962                             drr->drr_u.drr_write_embedded;
1963                         ra.err = restore_write_embedded(&ra, os, &drrwe);
1964                         break;
1965                 }
1966                 case DRR_FREE:
1967                 {
1968                         struct drr_free drrf = drr->drr_u.drr_free;
1969                         ra.err = restore_free(&ra, os, &drrf);
1970                         break;
1971                 }
1972                 case DRR_END:
1973                 {
1974                         struct drr_end drre = drr->drr_u.drr_end;
1975                         /*
1976                          * We compare against the *previous* checksum
1977                          * value, because the stored checksum is of
1978                          * everything before the DRR_END record.
1979                          */
1980                         if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1981                                 ra.err = SET_ERROR(ECKSUM);
1982                         goto out;
1983                 }
1984                 case DRR_SPILL:
1985                 {
1986                         struct drr_spill drrs = drr->drr_u.drr_spill;
1987                         ra.err = restore_spill(&ra, os, &drrs);
1988                         break;
1989                 }
1990                 default:
1991                         ra.err = SET_ERROR(EINVAL);
1992                         goto out;
1993                 }
1994                 pcksum = ra.cksum;
1995         }
1996         ASSERT(ra.err != 0);
1997
1998 out:
1999         if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
2000                 zfs_onexit_fd_rele(cleanup_fd);
2001
2002         if (ra.err != 0) {
2003                 /*
2004                  * destroy what we created, so we don't leave it in the
2005                  * inconsistent restoring state.
2006                  */
2007                 dmu_recv_cleanup_ds(drc);
2008         }
2009
2010         vmem_free(ra.buf, ra.bufsize);
2011         *voffp = ra.voff;
2012         return (ra.err);
2013 }
2014
2015 static int
2016 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
2017 {
2018         dmu_recv_cookie_t *drc = arg;
2019         dsl_pool_t *dp = dmu_tx_pool(tx);
2020         int error;
2021
2022         ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
2023
2024         if (!drc->drc_newfs) {
2025                 dsl_dataset_t *origin_head;
2026
2027                 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
2028                 if (error != 0)
2029                         return (error);
2030                 if (drc->drc_force) {
2031                         /*
2032                          * We will destroy any snapshots in tofs (i.e. before
2033                          * origin_head) that are after the origin (which is
2034                          * the snap before drc_ds, because drc_ds can not
2035                          * have any snaps of its own).
2036                          */
2037                         uint64_t obj;
2038
2039                         obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
2040                         while (obj !=
2041                             dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
2042                                 dsl_dataset_t *snap;
2043                                 error = dsl_dataset_hold_obj(dp, obj, FTAG,
2044                                     &snap);
2045                                 if (error != 0)
2046                                         break;
2047                                 if (snap->ds_dir != origin_head->ds_dir)
2048                                         error = SET_ERROR(EINVAL);
2049                                 if (error == 0)  {
2050                                         error = dsl_destroy_snapshot_check_impl(
2051                                             snap, B_FALSE);
2052                                 }
2053                                 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
2054                                 dsl_dataset_rele(snap, FTAG);
2055                                 if (error != 0)
2056                                         break;
2057                         }
2058                         if (error != 0) {
2059                                 dsl_dataset_rele(origin_head, FTAG);
2060                                 return (error);
2061                         }
2062                 }
2063                 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
2064                     origin_head, drc->drc_force, drc->drc_owner, tx);
2065                 if (error != 0) {
2066                         dsl_dataset_rele(origin_head, FTAG);
2067                         return (error);
2068                 }
2069                 error = dsl_dataset_snapshot_check_impl(origin_head,
2070                     drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2071                 dsl_dataset_rele(origin_head, FTAG);
2072                 if (error != 0)
2073                         return (error);
2074
2075                 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
2076         } else {
2077                 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
2078                     drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2079         }
2080         return (error);
2081 }
2082
2083 static void
2084 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
2085 {
2086         dmu_recv_cookie_t *drc = arg;
2087         dsl_pool_t *dp = dmu_tx_pool(tx);
2088
2089         spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
2090             tx, "snap=%s", drc->drc_tosnap);
2091
2092         if (!drc->drc_newfs) {
2093                 dsl_dataset_t *origin_head;
2094
2095                 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
2096                     &origin_head));
2097
2098                 if (drc->drc_force) {
2099                         /*
2100                          * Destroy any snapshots of drc_tofs (origin_head)
2101                          * after the origin (the snap before drc_ds).
2102                          */
2103                         uint64_t obj;
2104
2105                         obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
2106                         while (obj !=
2107                             dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
2108                                 dsl_dataset_t *snap;
2109                                 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
2110                                     &snap));
2111                                 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
2112                                 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
2113                                 dsl_destroy_snapshot_sync_impl(snap,
2114                                     B_FALSE, tx);
2115                                 dsl_dataset_rele(snap, FTAG);
2116                         }
2117                 }
2118                 VERIFY3P(drc->drc_ds->ds_prev, ==,
2119                     origin_head->ds_prev);
2120
2121                 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
2122                     origin_head, tx);
2123                 dsl_dataset_snapshot_sync_impl(origin_head,
2124                     drc->drc_tosnap, tx);
2125
2126                 /* set snapshot's creation time and guid */
2127                 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
2128                 dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time =
2129                     drc->drc_drrb->drr_creation_time;
2130                 dsl_dataset_phys(origin_head->ds_prev)->ds_guid =
2131                     drc->drc_drrb->drr_toguid;
2132                 dsl_dataset_phys(origin_head->ds_prev)->ds_flags &=
2133                     ~DS_FLAG_INCONSISTENT;
2134
2135                 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2136                 dsl_dataset_phys(origin_head)->ds_flags &=
2137                     ~DS_FLAG_INCONSISTENT;
2138
2139                 dsl_dataset_rele(origin_head, FTAG);
2140                 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
2141
2142                 if (drc->drc_owner != NULL)
2143                         VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
2144         } else {
2145                 dsl_dataset_t *ds = drc->drc_ds;
2146
2147                 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
2148
2149                 /* set snapshot's creation time and guid */
2150                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2151                 dsl_dataset_phys(ds->ds_prev)->ds_creation_time =
2152                     drc->drc_drrb->drr_creation_time;
2153                 dsl_dataset_phys(ds->ds_prev)->ds_guid =
2154                     drc->drc_drrb->drr_toguid;
2155                 dsl_dataset_phys(ds->ds_prev)->ds_flags &=
2156                     ~DS_FLAG_INCONSISTENT;
2157
2158                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2159                 dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT;
2160         }
2161         drc->drc_newsnapobj = dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj;
2162         /*
2163          * Release the hold from dmu_recv_begin.  This must be done before
2164          * we return to open context, so that when we free the dataset's dnode,
2165          * we can evict its bonus buffer.
2166          */
2167         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
2168         drc->drc_ds = NULL;
2169 }
2170
2171 static int
2172 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
2173 {
2174         dsl_pool_t *dp;
2175         dsl_dataset_t *snapds;
2176         guid_map_entry_t *gmep;
2177         int err;
2178
2179         ASSERT(guid_map != NULL);
2180
2181         err = dsl_pool_hold(name, FTAG, &dp);
2182         if (err != 0)
2183                 return (err);
2184         gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
2185         err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
2186         if (err == 0) {
2187                 gmep->guid = dsl_dataset_phys(snapds)->ds_guid;
2188                 gmep->gme_ds = snapds;
2189                 avl_add(guid_map, gmep);
2190                 dsl_dataset_long_hold(snapds, gmep);
2191         } else {
2192                 kmem_free(gmep, sizeof (*gmep));
2193         }
2194
2195         dsl_pool_rele(dp, FTAG);
2196         return (err);
2197 }
2198
2199 static int dmu_recv_end_modified_blocks = 3;
2200
2201 static int
2202 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
2203 {
2204         int error;
2205
2206 #ifdef _KERNEL
2207         char *name;
2208
2209         /*
2210          * We will be destroying the ds; make sure its origin is unmounted if
2211          * necessary.
2212          */
2213         name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2214         dsl_dataset_name(drc->drc_ds, name);
2215         zfs_destroy_unmount_origin(name);
2216         kmem_free(name, MAXNAMELEN);
2217 #endif
2218
2219         error = dsl_sync_task(drc->drc_tofs,
2220             dmu_recv_end_check, dmu_recv_end_sync, drc,
2221             dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2222
2223         if (error != 0)
2224                 dmu_recv_cleanup_ds(drc);
2225         return (error);
2226 }
2227
2228 static int
2229 dmu_recv_new_end(dmu_recv_cookie_t *drc)
2230 {
2231         int error;
2232
2233         error = dsl_sync_task(drc->drc_tofs,
2234             dmu_recv_end_check, dmu_recv_end_sync, drc,
2235             dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2236
2237         if (error != 0) {
2238                 dmu_recv_cleanup_ds(drc);
2239         } else if (drc->drc_guid_to_ds_map != NULL) {
2240                 (void) add_ds_to_guidmap(drc->drc_tofs,
2241                     drc->drc_guid_to_ds_map,
2242                     drc->drc_newsnapobj);
2243         }
2244         return (error);
2245 }
2246
2247 int
2248 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
2249 {
2250         drc->drc_owner = owner;
2251
2252         if (drc->drc_newfs)
2253                 return (dmu_recv_new_end(drc));
2254         else
2255                 return (dmu_recv_existing_end(drc));
2256 }
2257
2258 /*
2259  * Return TRUE if this objset is currently being received into.
2260  */
2261 boolean_t
2262 dmu_objset_is_receiving(objset_t *os)
2263 {
2264         return (os->os_dsl_dataset != NULL &&
2265             os->os_dsl_dataset->ds_owner == dmu_recv_tag);
2266 }
2267
2268 #if defined(_KERNEL)
2269 module_param(zfs_send_corrupt_data, int, 0644);
2270 MODULE_PARM_DESC(zfs_send_corrupt_data, "Allow sending corrupt data");
2271 #endif