sys/cddl/contrib/opensolaris/uts/common/fs/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 2011 Nexenta Systems, Inc. All rights reserved.
  24  * Copyright (c) 2013 by Delphix. All rights reserved.
  25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2012, Martin Matuska <mm@FreeBSD.org>. 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/zfs_ioctl.h>
  43 #include <sys/zap.h>
  44 #include <sys/zio_checksum.h>
  45 #include <sys/zfs_znode.h>
  46 #include <zfs_fletcher.h>
  47 #include <sys/avl.h>
  48 #include <sys/ddt.h>
  49 #include <sys/zfs_onexit.h>
  50 #include <sys/dmu_send.h>
  51 #include <sys/dsl_destroy.h>
  52
  53 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
  54 int zfs_send_corrupt_data = B_FALSE;
  55
  56 static char *dmu_recv_tag = "dmu_recv_tag";
  57 static const char *recv_clone_name = "%recv";
  58
  59 static int
  60 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
  61 {
  62         dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
  63         struct uio auio;
  64         struct iovec aiov;
  65         ASSERT0(len % 8);
  66
  67         fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
  68         aiov.iov_base = buf;
  69         aiov.iov_len = len;
  70         auio.uio_iov = &aiov;
  71         auio.uio_iovcnt = 1;
  72         auio.uio_resid = len;
  73         auio.uio_segflg = UIO_SYSSPACE;
  74         auio.uio_rw = UIO_WRITE;
  75         auio.uio_offset = (off_t)-1;
  76         auio.uio_td = dsp->dsa_td;
  77 #ifdef _KERNEL
  78         if (dsp->dsa_fp->f_type == DTYPE_VNODE)
  79                 bwillwrite();
  80         dsp->dsa_err = fo_write(dsp->dsa_fp, &auio, dsp->dsa_td->td_ucred, 0,
  81             dsp->dsa_td);
  82 #else
  83         fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
  84         dsp->dsa_err = EOPNOTSUPP;
  85 #endif
  86         mutex_enter(&ds->ds_sendstream_lock);
  87         *dsp->dsa_off += len;
  88         mutex_exit(&ds->ds_sendstream_lock);
  89
  90         return (dsp->dsa_err);
  91 }
  92
  93 static int
  94 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
  95     uint64_t length)
  96 {
  97         struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
  98
  99         /*
 100          * When we receive a free record, dbuf_free_range() assumes
 101          * that the receiving system doesn't have any dbufs in the range
 102          * being freed.  This is always true because there is a one-record
 103          * constraint: we only send one WRITE record for any given
 104          * object+offset.  We know that the one-record constraint is
 105          * true because we always send data in increasing order by
 106          * object,offset.
 107          *
 108          * If the increasing-order constraint ever changes, we should find
 109          * another way to assert that the one-record constraint is still
 110          * satisfied.
 111          */
 112         ASSERT(object > dsp->dsa_last_data_object ||
 113             (object == dsp->dsa_last_data_object &&
 114             offset > dsp->dsa_last_data_offset));
 115
 116         /*
 117          * If we are doing a non-incremental send, then there can't
 118          * be any data in the dataset we're receiving into.  Therefore
 119          * a free record would simply be a no-op.  Save space by not
 120          * sending it to begin with.
 121          */
 122         if (!dsp->dsa_incremental)
 123                 return (0);
 124
 125         if (length != -1ULL && offset + length < offset)
 126                 length = -1ULL;
 127
 128         /*
 129          * If there is a pending op, but it's not PENDING_FREE, push it out,
 130          * since free block aggregation can only be done for blocks of the
 131          * same type (i.e., DRR_FREE records can only be aggregated with
 132          * other DRR_FREE records.  DRR_FREEOBJECTS records can only be
 133          * aggregated with other DRR_FREEOBJECTS records.
 134          */
 135         if (dsp->dsa_pending_op != PENDING_NONE &&
 136             dsp->dsa_pending_op != PENDING_FREE) {
 137                 if (dump_bytes(dsp, dsp->dsa_drr,
 138                     sizeof (dmu_replay_record_t)) != 0)
 139                         return (SET_ERROR(EINTR));
 140                 dsp->dsa_pending_op = PENDING_NONE;
 141         }
 142
 143         if (dsp->dsa_pending_op == PENDING_FREE) {
 144                 /*
 145                  * There should never be a PENDING_FREE if length is -1
 146                  * (because dump_dnode is the only place where this
 147                  * function is called with a -1, and only after flushing
 148                  * any pending record).
 149                  */
 150                 ASSERT(length != -1ULL);
 151                 /*
 152                  * Check to see whether this free block can be aggregated
 153                  * with pending one.
 154                  */
 155                 if (drrf->drr_object == object && drrf->drr_offset +
 156                     drrf->drr_length == offset) {
 157                         drrf->drr_length += length;
 158                         return (0);
 159                 } else {
 160                         /* not a continuation.  Push out pending record */
 161                         if (dump_bytes(dsp, dsp->dsa_drr,
 162                             sizeof (dmu_replay_record_t)) != 0)
 163                                 return (SET_ERROR(EINTR));
 164                         dsp->dsa_pending_op = PENDING_NONE;
 165                 }
 166         }
 167         /* create a FREE record and make it pending */
 168         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 169         dsp->dsa_drr->drr_type = DRR_FREE;
 170         drrf->drr_object = object;
 171         drrf->drr_offset = offset;
 172         drrf->drr_length = length;
 173         drrf->drr_toguid = dsp->dsa_toguid;
 174         if (length == -1ULL) {
 175                 if (dump_bytes(dsp, dsp->dsa_drr,
 176                     sizeof (dmu_replay_record_t)) != 0)
 177                         return (SET_ERROR(EINTR));
 178         } else {
 179                 dsp->dsa_pending_op = PENDING_FREE;
 180         }
 181
 182         return (0);
 183 }
 184
 185 static int
 186 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
 187     uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
 188 {
 189         struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
 190
 191         /*
 192          * We send data in increasing object, offset order.
 193          * See comment in dump_free() for details.
 194          */
 195         ASSERT(object > dsp->dsa_last_data_object ||
 196             (object == dsp->dsa_last_data_object &&
 197             offset > dsp->dsa_last_data_offset));
 198         dsp->dsa_last_data_object = object;
 199         dsp->dsa_last_data_offset = offset + blksz - 1;
 200
 201         /*
 202          * If there is any kind of pending aggregation (currently either
 203          * a grouping of free objects or free blocks), push it out to
 204          * the stream, since aggregation can't be done across operations
 205          * of different types.
 206          */
 207         if (dsp->dsa_pending_op != PENDING_NONE) {
 208                 if (dump_bytes(dsp, dsp->dsa_drr,
 209                     sizeof (dmu_replay_record_t)) != 0)
 210                         return (SET_ERROR(EINTR));
 211                 dsp->dsa_pending_op = PENDING_NONE;
 212         }
 213         /* write a DATA record */
 214         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 215         dsp->dsa_drr->drr_type = DRR_WRITE;
 216         drrw->drr_object = object;
 217         drrw->drr_type = type;
 218         drrw->drr_offset = offset;
 219         drrw->drr_length = blksz;
 220         drrw->drr_toguid = dsp->dsa_toguid;
 221         drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
 222         if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
 223                 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
 224         DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
 225         DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
 226         DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
 227         drrw->drr_key.ddk_cksum = bp->blk_cksum;
 228
 229         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 230                 return (SET_ERROR(EINTR));
 231         if (dump_bytes(dsp, data, blksz) != 0)
 232                 return (SET_ERROR(EINTR));
 233         return (0);
 234 }
 235
 236 static int
 237 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
 238 {
 239         struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
 240
 241         if (dsp->dsa_pending_op != PENDING_NONE) {
 242                 if (dump_bytes(dsp, dsp->dsa_drr,
 243                     sizeof (dmu_replay_record_t)) != 0)
 244                         return (SET_ERROR(EINTR));
 245                 dsp->dsa_pending_op = PENDING_NONE;
 246         }
 247
 248         /* write a SPILL record */
 249         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 250         dsp->dsa_drr->drr_type = DRR_SPILL;
 251         drrs->drr_object = object;
 252         drrs->drr_length = blksz;
 253         drrs->drr_toguid = dsp->dsa_toguid;
 254
 255         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
 256                 return (SET_ERROR(EINTR));
 257         if (dump_bytes(dsp, data, blksz))
 258                 return (SET_ERROR(EINTR));
 259         return (0);
 260 }
 261
 262 static int
 263 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
 264 {
 265         struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
 266
 267         /* See comment in dump_free(). */
 268         if (!dsp->dsa_incremental)
 269                 return (0);
 270
 271         /*
 272          * If there is a pending op, but it's not PENDING_FREEOBJECTS,
 273          * push it out, since free block aggregation can only be done for
 274          * blocks of the same type (i.e., DRR_FREE records can only be
 275          * aggregated with other DRR_FREE records.  DRR_FREEOBJECTS records
 276          * can only be aggregated with other DRR_FREEOBJECTS records.
 277          */
 278         if (dsp->dsa_pending_op != PENDING_NONE &&
 279             dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
 280                 if (dump_bytes(dsp, dsp->dsa_drr,
 281                     sizeof (dmu_replay_record_t)) != 0)
 282                         return (SET_ERROR(EINTR));
 283                 dsp->dsa_pending_op = PENDING_NONE;
 284         }
 285         if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
 286                 /*
 287                  * See whether this free object array can be aggregated
 288                  * with pending one
 289                  */
 290                 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
 291                         drrfo->drr_numobjs += numobjs;
 292                         return (0);
 293                 } else {
 294                         /* can't be aggregated.  Push out pending record */
 295                         if (dump_bytes(dsp, dsp->dsa_drr,
 296                             sizeof (dmu_replay_record_t)) != 0)
 297                                 return (SET_ERROR(EINTR));
 298                         dsp->dsa_pending_op = PENDING_NONE;
 299                 }
 300         }
 301
 302         /* write a FREEOBJECTS record */
 303         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 304         dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
 305         drrfo->drr_firstobj = firstobj;
 306         drrfo->drr_numobjs = numobjs;
 307         drrfo->drr_toguid = dsp->dsa_toguid;
 308
 309         dsp->dsa_pending_op = PENDING_FREEOBJECTS;
 310
 311         return (0);
 312 }
 313
 314 static int
 315 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
 316 {
 317         struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
 318
 319         if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
 320                 return (dump_freeobjects(dsp, object, 1));
 321
 322         if (dsp->dsa_pending_op != PENDING_NONE) {
 323                 if (dump_bytes(dsp, dsp->dsa_drr,
 324                     sizeof (dmu_replay_record_t)) != 0)
 325                         return (SET_ERROR(EINTR));
 326                 dsp->dsa_pending_op = PENDING_NONE;
 327         }
 328
 329         /* write an OBJECT record */
 330         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 331         dsp->dsa_drr->drr_type = DRR_OBJECT;
 332         drro->drr_object = object;
 333         drro->drr_type = dnp->dn_type;
 334         drro->drr_bonustype = dnp->dn_bonustype;
 335         drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
 336         drro->drr_bonuslen = dnp->dn_bonuslen;
 337         drro->drr_checksumtype = dnp->dn_checksum;
 338         drro->drr_compress = dnp->dn_compress;
 339         drro->drr_toguid = dsp->dsa_toguid;
 340
 341         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 342                 return (SET_ERROR(EINTR));
 343
 344         if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
 345                 return (SET_ERROR(EINTR));
 346
 347         /* Free anything past the end of the file. */
 348         if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
 349             (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
 350                 return (SET_ERROR(EINTR));
 351         if (dsp->dsa_err != 0)
 352                 return (SET_ERROR(EINTR));
 353         return (0);
 354 }
 355
 356 #define BP_SPAN(dnp, level) \
 357         (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
 358         (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
 359
 360 /* ARGSUSED */
 361 static int
 362 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 363     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
 364 {
 365         dmu_sendarg_t *dsp = arg;
 366         dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
 367         int err = 0;
 368
 369         if (issig(JUSTLOOKING) && issig(FORREAL))
 370                 return (SET_ERROR(EINTR));
 371
 372         if (zb->zb_object != DMU_META_DNODE_OBJECT &&
 373             DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
 374                 return (0);
 375         } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
 376                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 377                 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
 378                 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
 379         } else if (bp == NULL) {
 380                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 381                 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
 382         } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
 383                 return (0);
 384         } else if (type == DMU_OT_DNODE) {
 385                 dnode_phys_t *blk;
 386                 int i;
 387                 int blksz = BP_GET_LSIZE(bp);
 388                 uint32_t aflags = ARC_WAIT;
 389                 arc_buf_t *abuf;
 390
 391                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 392                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 393                     &aflags, zb) != 0)
 394                         return (SET_ERROR(EIO));
 395
 396                 blk = abuf->b_data;
 397                 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
 398                         uint64_t dnobj = (zb->zb_blkid <<
 399                             (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
 400                         err = dump_dnode(dsp, dnobj, blk+i);
 401                         if (err != 0)
 402                                 break;
 403                 }
 404                 (void) arc_buf_remove_ref(abuf, &abuf);
 405         } else if (type == DMU_OT_SA) {
 406                 uint32_t aflags = ARC_WAIT;
 407                 arc_buf_t *abuf;
 408                 int blksz = BP_GET_LSIZE(bp);
 409
 410                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 411                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 412                     &aflags, zb) != 0)
 413                         return (SET_ERROR(EIO));
 414
 415                 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
 416                 (void) arc_buf_remove_ref(abuf, &abuf);
 417         } else { /* it's a level-0 block of a regular object */
 418                 uint32_t aflags = ARC_WAIT;
 419                 arc_buf_t *abuf;
 420                 int blksz = BP_GET_LSIZE(bp);
 421
 422                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 423                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 424                     &aflags, zb) != 0) {
 425                         if (zfs_send_corrupt_data) {
 426                                 /* Send a block filled with 0x"zfs badd bloc" */
 427                                 abuf = arc_buf_alloc(spa, blksz, &abuf,
 428                                     ARC_BUFC_DATA);
 429                                 uint64_t *ptr;
 430                                 for (ptr = abuf->b_data;
 431                                     (char *)ptr < (char *)abuf->b_data + blksz;
 432                                     ptr++)
 433                                         *ptr = 0x2f5baddb10c;
 434                         } else {
 435                                 return (SET_ERROR(EIO));
 436                         }
 437                 }
 438
 439                 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
 440                     blksz, bp, abuf->b_data);
 441                 (void) arc_buf_remove_ref(abuf, &abuf);
 442         }
 443
 444         ASSERT(err == 0 || err == EINTR);
 445         return (err);
 446 }
 447
 448 /*
 449  * Releases dp, ds, and fromds, using the specified tag.
 450  */
 451 static int
 452 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 453 #ifdef illumos
 454     dsl_dataset_t *fromds, int outfd, vnode_t *vp, offset_t *off)
 455 #else
 456     dsl_dataset_t *fromds, int outfd, struct file *fp, offset_t *off)
 457 #endif
 458 {
 459         objset_t *os;
 460         dmu_replay_record_t *drr;
 461         dmu_sendarg_t *dsp;
 462         int err;
 463         uint64_t fromtxg = 0;
 464
 465         if (fromds != NULL && !dsl_dataset_is_before(ds, fromds)) {
 466                 dsl_dataset_rele(fromds, tag);
 467                 dsl_dataset_rele(ds, tag);
 468                 dsl_pool_rele(dp, tag);
 469                 return (SET_ERROR(EXDEV));
 470         }
 471
 472         err = dmu_objset_from_ds(ds, &os);
 473         if (err != 0) {
 474                 if (fromds != NULL)
 475                         dsl_dataset_rele(fromds, tag);
 476                 dsl_dataset_rele(ds, tag);
 477                 dsl_pool_rele(dp, tag);
 478                 return (err);
 479         }
 480
 481         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
 482         drr->drr_type = DRR_BEGIN;
 483         drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
 484         DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
 485             DMU_SUBSTREAM);
 486
 487 #ifdef _KERNEL
 488         if (dmu_objset_type(os) == DMU_OST_ZFS) {
 489                 uint64_t version;
 490                 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
 491                         kmem_free(drr, sizeof (dmu_replay_record_t));
 492                         if (fromds != NULL)
 493                                 dsl_dataset_rele(fromds, tag);
 494                         dsl_dataset_rele(ds, tag);
 495                         dsl_pool_rele(dp, tag);
 496                         return (SET_ERROR(EINVAL));
 497                 }
 498                 if (version >= ZPL_VERSION_SA) {
 499                         DMU_SET_FEATUREFLAGS(
 500                             drr->drr_u.drr_begin.drr_versioninfo,
 501                             DMU_BACKUP_FEATURE_SA_SPILL);
 502                 }
 503         }
 504 #endif
 505
 506         drr->drr_u.drr_begin.drr_creation_time =
 507             ds->ds_phys->ds_creation_time;
 508         drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
 509         if (fromds != NULL && ds->ds_dir != fromds->ds_dir)
 510                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
 511         drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
 512         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 513                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
 514
 515         if (fromds != NULL)
 516                 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
 517         dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
 518
 519         if (fromds != NULL) {
 520                 fromtxg = fromds->ds_phys->ds_creation_txg;
 521                 dsl_dataset_rele(fromds, tag);
 522                 fromds = NULL;
 523         }
 524
 525         dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
 526
 527         dsp->dsa_drr = drr;
 528         dsp->dsa_outfd = outfd;
 529         dsp->dsa_proc = curproc;
 530         dsp->dsa_td = curthread;
 531         dsp->dsa_fp = fp;
 532         dsp->dsa_os = os;
 533         dsp->dsa_off = off;
 534         dsp->dsa_toguid = ds->ds_phys->ds_guid;
 535         ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
 536         dsp->dsa_pending_op = PENDING_NONE;
 537         dsp->dsa_incremental = (fromtxg != 0);
 538
 539         mutex_enter(&ds->ds_sendstream_lock);
 540         list_insert_head(&ds->ds_sendstreams, dsp);
 541         mutex_exit(&ds->ds_sendstream_lock);
 542
 543         dsl_dataset_long_hold(ds, FTAG);
 544         dsl_pool_rele(dp, tag);
 545
 546         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 547                 err = dsp->dsa_err;
 548                 goto out;
 549         }
 550
 551         err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
 552             backup_cb, dsp);
 553
 554         if (dsp->dsa_pending_op != PENDING_NONE)
 555                 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
 556                         err = SET_ERROR(EINTR);
 557
 558         if (err != 0) {
 559                 if (err == EINTR && dsp->dsa_err != 0)
 560                         err = dsp->dsa_err;
 561                 goto out;
 562         }
 563
 564         bzero(drr, sizeof (dmu_replay_record_t));
 565         drr->drr_type = DRR_END;
 566         drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
 567         drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
 568
 569         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 570                 err = dsp->dsa_err;
 571                 goto out;
 572         }
 573
 574 out:
 575         mutex_enter(&ds->ds_sendstream_lock);
 576         list_remove(&ds->ds_sendstreams, dsp);
 577         mutex_exit(&ds->ds_sendstream_lock);
 578
 579         kmem_free(drr, sizeof (dmu_replay_record_t));
 580         kmem_free(dsp, sizeof (dmu_sendarg_t));
 581
 582         dsl_dataset_long_rele(ds, FTAG);
 583         dsl_dataset_rele(ds, tag);
 584
 585         return (err);
 586 }
 587
 588 int
 589 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
 590 #ifdef illumos
 591     int outfd, vnode_t *vp, offset_t *off)
 592 #else
 593     int outfd, struct file *fp, offset_t *off)
 594 #endif
 595 {
 596         dsl_pool_t *dp;
 597         dsl_dataset_t *ds;
 598         dsl_dataset_t *fromds = NULL;
 599         int err;
 600
 601         err = dsl_pool_hold(pool, FTAG, &dp);
 602         if (err != 0)
 603                 return (err);
 604
 605         err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
 606         if (err != 0) {
 607                 dsl_pool_rele(dp, FTAG);
 608                 return (err);
 609         }
 610
 611         if (fromsnap != 0) {
 612                 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
 613                 if (err != 0) {
 614                         dsl_dataset_rele(ds, FTAG);
 615                         dsl_pool_rele(dp, FTAG);
 616                         return (err);
 617                 }
 618         }
 619
 620         return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, fp, off));
 621 }
 622
 623 int
 624 dmu_send(const char *tosnap, const char *fromsnap,
 625 #ifdef illumos
 626     int outfd, vnode_t *vp, offset_t *off)
 627 #else
 628     int outfd, struct file *fp, offset_t *off)
 629 #endif
 630 {
 631         dsl_pool_t *dp;
 632         dsl_dataset_t *ds;
 633         dsl_dataset_t *fromds = NULL;
 634         int err;
 635
 636         if (strchr(tosnap, '@') == NULL)
 637                 return (SET_ERROR(EINVAL));
 638         if (fromsnap != NULL && strchr(fromsnap, '@') == NULL)
 639                 return (SET_ERROR(EINVAL));
 640
 641         err = dsl_pool_hold(tosnap, FTAG, &dp);
 642         if (err != 0)
 643                 return (err);
 644
 645         err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
 646         if (err != 0) {
 647                 dsl_pool_rele(dp, FTAG);
 648                 return (err);
 649         }
 650
 651         if (fromsnap != NULL) {
 652                 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
 653                 if (err != 0) {
 654                         dsl_dataset_rele(ds, FTAG);
 655                         dsl_pool_rele(dp, FTAG);
 656                         return (err);
 657                 }
 658         }
 659         return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, fp, off));
 660 }
 661
 662 int
 663 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
 664 {
 665         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 666         int err;
 667         uint64_t size;
 668
 669         ASSERT(dsl_pool_config_held(dp));
 670
 671         /* tosnap must be a snapshot */
 672         if (!dsl_dataset_is_snapshot(ds))
 673                 return (SET_ERROR(EINVAL));
 674
 675         /*
 676          * fromsnap must be an earlier snapshot from the same fs as tosnap,
 677          * or the origin's fs.
 678          */
 679         if (fromds != NULL && !dsl_dataset_is_before(ds, fromds))
 680                 return (SET_ERROR(EXDEV));
 681
 682         /* Get uncompressed size estimate of changed data. */
 683         if (fromds == NULL) {
 684                 size = ds->ds_phys->ds_uncompressed_bytes;
 685         } else {
 686                 uint64_t used, comp;
 687                 err = dsl_dataset_space_written(fromds, ds,
 688                     &used, &comp, &size);
 689                 if (err != 0)
 690                         return (err);
 691         }
 692
 693         /*
 694          * Assume that space (both on-disk and in-stream) is dominated by
 695          * data.  We will adjust for indirect blocks and the copies property,
 696          * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
 697          */
 698
 699         /*
 700          * Subtract out approximate space used by indirect blocks.
 701          * Assume most space is used by data blocks (non-indirect, non-dnode).
 702          * Assume all blocks are recordsize.  Assume ditto blocks and
 703          * internal fragmentation counter out compression.
 704          *
 705          * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
 706          * block, which we observe in practice.
 707          */
 708         uint64_t recordsize;
 709         err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
 710         if (err != 0)
 711                 return (err);
 712         size -= size / recordsize * sizeof (blkptr_t);
 713
 714         /* Add in the space for the record associated with each block. */
 715         size += size / recordsize * sizeof (dmu_replay_record_t);
 716
 717         *sizep = size;
 718
 719         return (0);
 720 }
 721
 722 typedef struct dmu_recv_begin_arg {
 723         const char *drba_origin;
 724         dmu_recv_cookie_t *drba_cookie;
 725         cred_t *drba_cred;
 726         uint64_t drba_snapobj;
 727 } dmu_recv_begin_arg_t;
 728
 729 static int
 730 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
 731     uint64_t fromguid)
 732 {
 733         uint64_t val;
 734         int error;
 735         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 736
 737         /* temporary clone name must not exist */
 738         error = zap_lookup(dp->dp_meta_objset,
 739             ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
 740             8, 1, &val);
 741         if (error != ENOENT)
 742                 return (error == 0 ? EBUSY : error);
 743
 744         /* new snapshot name must not exist */
 745         error = zap_lookup(dp->dp_meta_objset,
 746             ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
 747             8, 1, &val);
 748         if (error != ENOENT)
 749                 return (error == 0 ? EEXIST : error);
 750
 751         if (fromguid != 0) {
 752                 dsl_dataset_t *snap;
 753                 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
 754
 755                 /* Find snapshot in this dir that matches fromguid. */
 756                 while (obj != 0) {
 757                         error = dsl_dataset_hold_obj(dp, obj, FTAG,
 758                             &snap);
 759                         if (error != 0)
 760                                 return (SET_ERROR(ENODEV));
 761                         if (snap->ds_dir != ds->ds_dir) {
 762                                 dsl_dataset_rele(snap, FTAG);
 763                                 return (SET_ERROR(ENODEV));
 764                         }
 765                         if (snap->ds_phys->ds_guid == fromguid)
 766                                 break;
 767                         obj = snap->ds_phys->ds_prev_snap_obj;
 768                         dsl_dataset_rele(snap, FTAG);
 769                 }
 770                 if (obj == 0)
 771                         return (SET_ERROR(ENODEV));
 772
 773                 if (drba->drba_cookie->drc_force) {
 774                         drba->drba_snapobj = obj;
 775                 } else {
 776                         /*
 777                          * If we are not forcing, there must be no
 778                          * changes since fromsnap.
 779                          */
 780                         if (dsl_dataset_modified_since_snap(ds, snap)) {
 781                                 dsl_dataset_rele(snap, FTAG);
 782                                 return (SET_ERROR(ETXTBSY));
 783                         }
 784                         drba->drba_snapobj = ds->ds_prev->ds_object;
 785                 }
 786
 787                 dsl_dataset_rele(snap, FTAG);
 788         } else {
 789                 /* if full, most recent snapshot must be $ORIGIN */
 790                 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
 791                         return (SET_ERROR(ENODEV));
 792                 drba->drba_snapobj = ds->ds_phys->ds_prev_snap_obj;
 793         }
 794
 795         return (0);
 796
 797 }
 798
 799 static int
 800 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 801 {
 802         dmu_recv_begin_arg_t *drba = arg;
 803         dsl_pool_t *dp = dmu_tx_pool(tx);
 804         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 805         uint64_t fromguid = drrb->drr_fromguid;
 806         int flags = drrb->drr_flags;
 807         int error;
 808         dsl_dataset_t *ds;
 809         const char *tofs = drba->drba_cookie->drc_tofs;
 810
 811         /* already checked */
 812         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 813
 814         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 815             DMU_COMPOUNDSTREAM ||
 816             drrb->drr_type >= DMU_OST_NUMTYPES ||
 817             ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
 818                 return (SET_ERROR(EINVAL));
 819
 820         /* Verify pool version supports SA if SA_SPILL feature set */
 821         if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
 822             DMU_BACKUP_FEATURE_SA_SPILL) &&
 823             spa_version(dp->dp_spa) < SPA_VERSION_SA) {
 824                 return (SET_ERROR(ENOTSUP));
 825         }
 826
 827         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 828         if (error == 0) {
 829                 /* target fs already exists; recv into temp clone */
 830
 831                 /* Can't recv a clone into an existing fs */
 832                 if (flags & DRR_FLAG_CLONE) {
 833                         dsl_dataset_rele(ds, FTAG);
 834                         return (SET_ERROR(EINVAL));
 835                 }
 836
 837                 error = recv_begin_check_existing_impl(drba, ds, fromguid);
 838                 dsl_dataset_rele(ds, FTAG);
 839         } else if (error == ENOENT) {
 840                 /* target fs does not exist; must be a full backup or clone */
 841                 char buf[MAXNAMELEN];
 842
 843                 /*
 844                  * If it's a non-clone incremental, we are missing the
 845                  * target fs, so fail the recv.
 846                  */
 847                 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
 848                         return (SET_ERROR(ENOENT));
 849
 850                 /* Open the parent of tofs */
 851                 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
 852                 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
 853                 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
 854                 if (error != 0)
 855                         return (error);
 856
 857                 if (drba->drba_origin != NULL) {
 858                         dsl_dataset_t *origin;
 859                         error = dsl_dataset_hold(dp, drba->drba_origin,
 860                             FTAG, &origin);
 861                         if (error != 0) {
 862                                 dsl_dataset_rele(ds, FTAG);
 863                                 return (error);
 864                         }
 865                         if (!dsl_dataset_is_snapshot(origin)) {
 866                                 dsl_dataset_rele(origin, FTAG);
 867                                 dsl_dataset_rele(ds, FTAG);
 868                                 return (SET_ERROR(EINVAL));
 869                         }
 870                         if (origin->ds_phys->ds_guid != fromguid) {
 871                                 dsl_dataset_rele(origin, FTAG);
 872                                 dsl_dataset_rele(ds, FTAG);
 873                                 return (SET_ERROR(ENODEV));
 874                         }
 875                         dsl_dataset_rele(origin, FTAG);
 876                 }
 877                 dsl_dataset_rele(ds, FTAG);
 878                 error = 0;
 879         }
 880         return (error);
 881 }
 882
 883 static void
 884 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
 885 {
 886         dmu_recv_begin_arg_t *drba = arg;
 887         dsl_pool_t *dp = dmu_tx_pool(tx);
 888         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 889         const char *tofs = drba->drba_cookie->drc_tofs;
 890         dsl_dataset_t *ds, *newds;
 891         uint64_t dsobj;
 892         int error;
 893         uint64_t crflags;
 894
 895         crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
 896             DS_FLAG_CI_DATASET : 0;
 897
 898         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 899         if (error == 0) {
 900                 /* create temporary clone */
 901                 dsl_dataset_t *snap = NULL;
 902                 if (drba->drba_snapobj != 0) {
 903                         VERIFY0(dsl_dataset_hold_obj(dp,
 904                             drba->drba_snapobj, FTAG, &snap));
 905                 }
 906                 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
 907                     snap, crflags, drba->drba_cred, tx);
 908                 dsl_dataset_rele(snap, FTAG);
 909                 dsl_dataset_rele(ds, FTAG);
 910         } else {
 911                 dsl_dir_t *dd;
 912                 const char *tail;
 913                 dsl_dataset_t *origin = NULL;
 914
 915                 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
 916
 917                 if (drba->drba_origin != NULL) {
 918                         VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
 919                             FTAG, &origin));
 920                 }
 921
 922                 /* Create new dataset. */
 923                 dsobj = dsl_dataset_create_sync(dd,
 924                     strrchr(tofs, '/') + 1,
 925                     origin, crflags, drba->drba_cred, tx);
 926                 if (origin != NULL)
 927                         dsl_dataset_rele(origin, FTAG);
 928                 dsl_dir_rele(dd, FTAG);
 929                 drba->drba_cookie->drc_newfs = B_TRUE;
 930         }
 931         VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
 932
 933         dmu_buf_will_dirty(newds->ds_dbuf, tx);
 934         newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
 935
 936         /*
 937          * If we actually created a non-clone, we need to create the
 938          * objset in our new dataset.
 939          */
 940         if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
 941                 (void) dmu_objset_create_impl(dp->dp_spa,
 942                     newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
 943         }
 944
 945         drba->drba_cookie->drc_ds = newds;
 946
 947         spa_history_log_internal_ds(newds, "receive", tx, "");
 948 }
 949
 950 /*
 951  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
 952  * succeeds; otherwise we will leak the holds on the datasets.
 953  */
 954 int
 955 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
 956     boolean_t force, char *origin, dmu_recv_cookie_t *drc)
 957 {
 958         dmu_recv_begin_arg_t drba = { 0 };
 959         dmu_replay_record_t *drr;
 960
 961         bzero(drc, sizeof (dmu_recv_cookie_t));
 962         drc->drc_drrb = drrb;
 963         drc->drc_tosnap = tosnap;
 964         drc->drc_tofs = tofs;
 965         drc->drc_force = force;
 966
 967         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
 968                 drc->drc_byteswap = B_TRUE;
 969         else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
 970                 return (SET_ERROR(EINVAL));
 971
 972         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
 973         drr->drr_type = DRR_BEGIN;
 974         drr->drr_u.drr_begin = *drc->drc_drrb;
 975         if (drc->drc_byteswap) {
 976                 fletcher_4_incremental_byteswap(drr,
 977                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
 978         } else {
 979                 fletcher_4_incremental_native(drr,
 980                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
 981         }
 982         kmem_free(drr, sizeof (dmu_replay_record_t));
 983
 984         if (drc->drc_byteswap) {
 985                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
 986                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
 987                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
 988                 drrb->drr_type = BSWAP_32(drrb->drr_type);
 989                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
 990                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
 991         }
 992
 993         drba.drba_origin = origin;
 994         drba.drba_cookie = drc;
 995         drba.drba_cred = CRED();
 996
 997         return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
 998             &drba, 5));
 999 }
1000
1001 struct restorearg {
1002         int err;
1003         boolean_t byteswap;
1004         kthread_t *td;
1005         struct file *fp;
1006         char *buf;
1007         uint64_t voff;
1008         int bufsize; /* amount of memory allocated for buf */
1009         zio_cksum_t cksum;
1010         avl_tree_t *guid_to_ds_map;
1011 };
1012
1013 typedef struct guid_map_entry {
1014         uint64_t        guid;
1015         dsl_dataset_t   *gme_ds;
1016         avl_node_t      avlnode;
1017 } guid_map_entry_t;
1018
1019 static int
1020 guid_compare(const void *arg1, const void *arg2)
1021 {
1022         const guid_map_entry_t *gmep1 = arg1;
1023         const guid_map_entry_t *gmep2 = arg2;
1024
1025         if (gmep1->guid < gmep2->guid)
1026                 return (-1);
1027         else if (gmep1->guid > gmep2->guid)
1028                 return (1);
1029         return (0);
1030 }
1031
1032 static void
1033 free_guid_map_onexit(void *arg)
1034 {
1035         avl_tree_t *ca = arg;
1036         void *cookie = NULL;
1037         guid_map_entry_t *gmep;
1038
1039         while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1040                 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1041                 dsl_dataset_rele(gmep->gme_ds, gmep);
1042                 kmem_free(gmep, sizeof (guid_map_entry_t));
1043         }
1044         avl_destroy(ca);
1045         kmem_free(ca, sizeof (avl_tree_t));
1046 }
1047
1048 static int
1049 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
1050 {
1051         struct uio auio;
1052         struct iovec aiov;
1053         int error;
1054
1055         aiov.iov_base = buf;
1056         aiov.iov_len = len;
1057         auio.uio_iov = &aiov;
1058         auio.uio_iovcnt = 1;
1059         auio.uio_resid = len;
1060         auio.uio_segflg = UIO_SYSSPACE;
1061         auio.uio_rw = UIO_READ;
1062         auio.uio_offset = off;
1063         auio.uio_td = ra->td;
1064 #ifdef _KERNEL
1065         error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
1066 #else
1067         fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
1068         error = EOPNOTSUPP;
1069 #endif
1070         *resid = auio.uio_resid;
1071         return (error);
1072 }
1073
1074 static void *
1075 restore_read(struct restorearg *ra, int len)
1076 {
1077         void *rv;
1078         int done = 0;
1079
1080         /* some things will require 8-byte alignment, so everything must */
1081         ASSERT0(len % 8);
1082
1083         while (done < len) {
1084                 ssize_t resid;
1085
1086                 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
1087                     len - done, ra->voff, &resid);
1088
1089                 if (resid == len - done)
1090                         ra->err = SET_ERROR(EINVAL);
1091                 ra->voff += len - done - resid;
1092                 done = len - resid;
1093                 if (ra->err != 0)
1094                         return (NULL);
1095         }
1096
1097         ASSERT3U(done, ==, len);
1098         rv = ra->buf;
1099         if (ra->byteswap)
1100                 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1101         else
1102                 fletcher_4_incremental_native(rv, len, &ra->cksum);
1103         return (rv);
1104 }
1105
1106 static void
1107 backup_byteswap(dmu_replay_record_t *drr)
1108 {
1109 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1110 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1111         drr->drr_type = BSWAP_32(drr->drr_type);
1112         drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1113         switch (drr->drr_type) {
1114         case DRR_BEGIN:
1115                 DO64(drr_begin.drr_magic);
1116                 DO64(drr_begin.drr_versioninfo);
1117                 DO64(drr_begin.drr_creation_time);
1118                 DO32(drr_begin.drr_type);
1119                 DO32(drr_begin.drr_flags);
1120                 DO64(drr_begin.drr_toguid);
1121                 DO64(drr_begin.drr_fromguid);
1122                 break;
1123         case DRR_OBJECT:
1124                 DO64(drr_object.drr_object);
1125                 /* DO64(drr_object.drr_allocation_txg); */
1126                 DO32(drr_object.drr_type);
1127                 DO32(drr_object.drr_bonustype);
1128                 DO32(drr_object.drr_blksz);
1129                 DO32(drr_object.drr_bonuslen);
1130                 DO64(drr_object.drr_toguid);
1131                 break;
1132         case DRR_FREEOBJECTS:
1133                 DO64(drr_freeobjects.drr_firstobj);
1134                 DO64(drr_freeobjects.drr_numobjs);
1135                 DO64(drr_freeobjects.drr_toguid);
1136                 break;
1137         case DRR_WRITE:
1138                 DO64(drr_write.drr_object);
1139                 DO32(drr_write.drr_type);
1140                 DO64(drr_write.drr_offset);
1141                 DO64(drr_write.drr_length);
1142                 DO64(drr_write.drr_toguid);
1143                 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1144                 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1145                 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1146                 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1147                 DO64(drr_write.drr_key.ddk_prop);
1148                 break;
1149         case DRR_WRITE_BYREF:
1150                 DO64(drr_write_byref.drr_object);
1151                 DO64(drr_write_byref.drr_offset);
1152                 DO64(drr_write_byref.drr_length);
1153                 DO64(drr_write_byref.drr_toguid);
1154                 DO64(drr_write_byref.drr_refguid);
1155                 DO64(drr_write_byref.drr_refobject);
1156                 DO64(drr_write_byref.drr_refoffset);
1157                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1158                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1159                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1160                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1161                 DO64(drr_write_byref.drr_key.ddk_prop);
1162                 break;
1163         case DRR_FREE:
1164                 DO64(drr_free.drr_object);
1165                 DO64(drr_free.drr_offset);
1166                 DO64(drr_free.drr_length);
1167                 DO64(drr_free.drr_toguid);
1168                 break;
1169         case DRR_SPILL:
1170                 DO64(drr_spill.drr_object);
1171                 DO64(drr_spill.drr_length);
1172                 DO64(drr_spill.drr_toguid);
1173                 break;
1174         case DRR_END:
1175                 DO64(drr_end.drr_checksum.zc_word[0]);
1176                 DO64(drr_end.drr_checksum.zc_word[1]);
1177                 DO64(drr_end.drr_checksum.zc_word[2]);
1178                 DO64(drr_end.drr_checksum.zc_word[3]);
1179                 DO64(drr_end.drr_toguid);
1180                 break;
1181         }
1182 #undef DO64
1183 #undef DO32
1184 }
1185
1186 static int
1187 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1188 {
1189         int err;
1190         dmu_tx_t *tx;
1191         void *data = NULL;
1192
1193         if (drro->drr_type == DMU_OT_NONE ||
1194             !DMU_OT_IS_VALID(drro->drr_type) ||
1195             !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1196             drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1197             drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1198             P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1199             drro->drr_blksz < SPA_MINBLOCKSIZE ||
1200             drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1201             drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1202                 return (SET_ERROR(EINVAL));
1203         }
1204
1205         err = dmu_object_info(os, drro->drr_object, NULL);
1206
1207         if (err != 0 && err != ENOENT)
1208                 return (SET_ERROR(EINVAL));
1209
1210         if (drro->drr_bonuslen) {
1211                 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1212                 if (ra->err != 0)
1213                         return (ra->err);
1214         }
1215
1216         if (err == ENOENT) {
1217                 /* currently free, want to be allocated */
1218                 tx = dmu_tx_create(os);
1219                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1220                 err = dmu_tx_assign(tx, TXG_WAIT);
1221                 if (err != 0) {
1222                         dmu_tx_abort(tx);
1223                         return (err);
1224                 }
1225                 err = dmu_object_claim(os, drro->drr_object,
1226                     drro->drr_type, drro->drr_blksz,
1227                     drro->drr_bonustype, drro->drr_bonuslen, tx);
1228                 dmu_tx_commit(tx);
1229         } else {
1230                 /* currently allocated, want to be allocated */
1231                 err = dmu_object_reclaim(os, drro->drr_object,
1232                     drro->drr_type, drro->drr_blksz,
1233                     drro->drr_bonustype, drro->drr_bonuslen);
1234         }
1235         if (err != 0) {
1236                 return (SET_ERROR(EINVAL));
1237         }
1238
1239         tx = dmu_tx_create(os);
1240         dmu_tx_hold_bonus(tx, drro->drr_object);
1241         err = dmu_tx_assign(tx, TXG_WAIT);
1242         if (err != 0) {
1243                 dmu_tx_abort(tx);
1244                 return (err);
1245         }
1246
1247         dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1248             tx);
1249         dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1250
1251         if (data != NULL) {
1252                 dmu_buf_t *db;
1253
1254                 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1255                 dmu_buf_will_dirty(db, tx);
1256
1257                 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1258                 bcopy(data, db->db_data, drro->drr_bonuslen);
1259                 if (ra->byteswap) {
1260                         dmu_object_byteswap_t byteswap =
1261                             DMU_OT_BYTESWAP(drro->drr_bonustype);
1262                         dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1263                             drro->drr_bonuslen);
1264                 }
1265                 dmu_buf_rele(db, FTAG);
1266         }
1267         dmu_tx_commit(tx);
1268         return (0);
1269 }
1270
1271 /* ARGSUSED */
1272 static int
1273 restore_freeobjects(struct restorearg *ra, objset_t *os,
1274     struct drr_freeobjects *drrfo)
1275 {
1276         uint64_t obj;
1277
1278         if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1279                 return (SET_ERROR(EINVAL));
1280
1281         for (obj = drrfo->drr_firstobj;
1282             obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1283             (void) dmu_object_next(os, &obj, FALSE, 0)) {
1284                 int err;
1285
1286                 if (dmu_object_info(os, obj, NULL) != 0)
1287                         continue;
1288
1289                 err = dmu_free_long_object(os, obj);
1290                 if (err != 0)
1291                         return (err);
1292         }
1293         return (0);
1294 }
1295
1296 static int
1297 restore_write(struct restorearg *ra, objset_t *os,
1298     struct drr_write *drrw)
1299 {
1300         dmu_tx_t *tx;
1301         void *data;
1302         int err;
1303
1304         if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1305             !DMU_OT_IS_VALID(drrw->drr_type))
1306                 return (SET_ERROR(EINVAL));
1307
1308         data = restore_read(ra, drrw->drr_length);
1309         if (data == NULL)
1310                 return (ra->err);
1311
1312         if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1313                 return (SET_ERROR(EINVAL));
1314
1315         tx = dmu_tx_create(os);
1316
1317         dmu_tx_hold_write(tx, drrw->drr_object,
1318             drrw->drr_offset, drrw->drr_length);
1319         err = dmu_tx_assign(tx, TXG_WAIT);
1320         if (err != 0) {
1321                 dmu_tx_abort(tx);
1322                 return (err);
1323         }
1324         if (ra->byteswap) {
1325                 dmu_object_byteswap_t byteswap =
1326                     DMU_OT_BYTESWAP(drrw->drr_type);
1327                 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1328         }
1329         dmu_write(os, drrw->drr_object,
1330             drrw->drr_offset, drrw->drr_length, data, tx);
1331         dmu_tx_commit(tx);
1332         return (0);
1333 }
1334
1335 /*
1336  * Handle a DRR_WRITE_BYREF record.  This record is used in dedup'ed
1337  * streams to refer to a copy of the data that is already on the
1338  * system because it came in earlier in the stream.  This function
1339  * finds the earlier copy of the data, and uses that copy instead of
1340  * data from the stream to fulfill this write.
1341  */
1342 static int
1343 restore_write_byref(struct restorearg *ra, objset_t *os,
1344     struct drr_write_byref *drrwbr)
1345 {
1346         dmu_tx_t *tx;
1347         int err;
1348         guid_map_entry_t gmesrch;
1349         guid_map_entry_t *gmep;
1350         avl_index_t     where;
1351         objset_t *ref_os = NULL;
1352         dmu_buf_t *dbp;
1353
1354         if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1355                 return (SET_ERROR(EINVAL));
1356
1357         /*
1358          * If the GUID of the referenced dataset is different from the
1359          * GUID of the target dataset, find the referenced dataset.
1360          */
1361         if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1362                 gmesrch.guid = drrwbr->drr_refguid;
1363                 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1364                     &where)) == NULL) {
1365                         return (SET_ERROR(EINVAL));
1366                 }
1367                 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1368                         return (SET_ERROR(EINVAL));
1369         } else {
1370                 ref_os = os;
1371         }
1372
1373         if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1374             drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1375                 return (err);
1376
1377         tx = dmu_tx_create(os);
1378
1379         dmu_tx_hold_write(tx, drrwbr->drr_object,
1380             drrwbr->drr_offset, drrwbr->drr_length);
1381         err = dmu_tx_assign(tx, TXG_WAIT);
1382         if (err != 0) {
1383                 dmu_tx_abort(tx);
1384                 return (err);
1385         }
1386         dmu_write(os, drrwbr->drr_object,
1387             drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1388         dmu_buf_rele(dbp, FTAG);
1389         dmu_tx_commit(tx);
1390         return (0);
1391 }
1392
1393 static int
1394 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1395 {
1396         dmu_tx_t *tx;
1397         void *data;
1398         dmu_buf_t *db, *db_spill;
1399         int err;
1400
1401         if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1402             drrs->drr_length > SPA_MAXBLOCKSIZE)
1403                 return (SET_ERROR(EINVAL));
1404
1405         data = restore_read(ra, drrs->drr_length);
1406         if (data == NULL)
1407                 return (ra->err);
1408
1409         if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1410                 return (SET_ERROR(EINVAL));
1411
1412         VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1413         if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1414                 dmu_buf_rele(db, FTAG);
1415                 return (err);
1416         }
1417
1418         tx = dmu_tx_create(os);
1419
1420         dmu_tx_hold_spill(tx, db->db_object);
1421
1422         err = dmu_tx_assign(tx, TXG_WAIT);
1423         if (err != 0) {
1424                 dmu_buf_rele(db, FTAG);
1425                 dmu_buf_rele(db_spill, FTAG);
1426                 dmu_tx_abort(tx);
1427                 return (err);
1428         }
1429         dmu_buf_will_dirty(db_spill, tx);
1430
1431         if (db_spill->db_size < drrs->drr_length)
1432                 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1433                     drrs->drr_length, tx));
1434         bcopy(data, db_spill->db_data, drrs->drr_length);
1435
1436         dmu_buf_rele(db, FTAG);
1437         dmu_buf_rele(db_spill, FTAG);
1438
1439         dmu_tx_commit(tx);
1440         return (0);
1441 }
1442
1443 /* ARGSUSED */
1444 static int
1445 restore_free(struct restorearg *ra, objset_t *os,
1446     struct drr_free *drrf)
1447 {
1448         int err;
1449
1450         if (drrf->drr_length != -1ULL &&
1451             drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1452                 return (SET_ERROR(EINVAL));
1453
1454         if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1455                 return (SET_ERROR(EINVAL));
1456
1457         err = dmu_free_long_range(os, drrf->drr_object,
1458             drrf->drr_offset, drrf->drr_length);
1459         return (err);
1460 }
1461
1462 /* used to destroy the drc_ds on error */
1463 static void
1464 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1465 {
1466         char name[MAXNAMELEN];
1467         dsl_dataset_name(drc->drc_ds, name);
1468         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1469         (void) dsl_destroy_head(name);
1470 }
1471
1472 /*
1473  * NB: callers *must* call dmu_recv_end() if this succeeds.
1474  */
1475 int
1476 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1477     int cleanup_fd, uint64_t *action_handlep)
1478 {
1479         struct restorearg ra = { 0 };
1480         dmu_replay_record_t *drr;
1481         objset_t *os;
1482         zio_cksum_t pcksum;
1483         int featureflags;
1484
1485         ra.byteswap = drc->drc_byteswap;
1486         ra.cksum = drc->drc_cksum;
1487         ra.td = curthread;
1488         ra.fp = fp;
1489         ra.voff = *voffp;
1490         ra.bufsize = 1<<20;
1491         ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1492
1493         /* these were verified in dmu_recv_begin */
1494         ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1495             DMU_SUBSTREAM);
1496         ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1497
1498         /*
1499          * Open the objset we are modifying.
1500          */
1501         VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1502
1503         ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1504
1505         featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1506
1507         /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1508         if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1509                 minor_t minor;
1510
1511                 if (cleanup_fd == -1) {
1512                         ra.err = SET_ERROR(EBADF);
1513                         goto out;
1514                 }
1515                 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1516                 if (ra.err != 0) {
1517                         cleanup_fd = -1;
1518                         goto out;
1519                 }
1520
1521                 if (*action_handlep == 0) {
1522                         ra.guid_to_ds_map =
1523                             kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1524                         avl_create(ra.guid_to_ds_map, guid_compare,
1525                             sizeof (guid_map_entry_t),
1526                             offsetof(guid_map_entry_t, avlnode));
1527                         ra.err = zfs_onexit_add_cb(minor,
1528                             free_guid_map_onexit, ra.guid_to_ds_map,
1529                             action_handlep);
1530                         if (ra.err != 0)
1531                                 goto out;
1532                 } else {
1533                         ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1534                             (void **)&ra.guid_to_ds_map);
1535                         if (ra.err != 0)
1536                                 goto out;
1537                 }
1538
1539                 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1540         }
1541
1542         /*
1543          * Read records and process them.
1544          */
1545         pcksum = ra.cksum;
1546         while (ra.err == 0 &&
1547             NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1548                 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1549                         ra.err = SET_ERROR(EINTR);
1550                         goto out;
1551                 }
1552
1553                 if (ra.byteswap)
1554                         backup_byteswap(drr);
1555
1556                 switch (drr->drr_type) {
1557                 case DRR_OBJECT:
1558                 {
1559                         /*
1560                          * We need to make a copy of the record header,
1561                          * because restore_{object,write} may need to
1562                          * restore_read(), which will invalidate drr.
1563                          */
1564                         struct drr_object drro = drr->drr_u.drr_object;
1565                         ra.err = restore_object(&ra, os, &drro);
1566                         break;
1567                 }
1568                 case DRR_FREEOBJECTS:
1569                 {
1570                         struct drr_freeobjects drrfo =
1571                             drr->drr_u.drr_freeobjects;
1572                         ra.err = restore_freeobjects(&ra, os, &drrfo);
1573                         break;
1574                 }
1575                 case DRR_WRITE:
1576                 {
1577                         struct drr_write drrw = drr->drr_u.drr_write;
1578                         ra.err = restore_write(&ra, os, &drrw);
1579                         break;
1580                 }
1581                 case DRR_WRITE_BYREF:
1582                 {
1583                         struct drr_write_byref drrwbr =
1584                             drr->drr_u.drr_write_byref;
1585                         ra.err = restore_write_byref(&ra, os, &drrwbr);
1586                         break;
1587                 }
1588                 case DRR_FREE:
1589                 {
1590                         struct drr_free drrf = drr->drr_u.drr_free;
1591                         ra.err = restore_free(&ra, os, &drrf);
1592                         break;
1593                 }
1594                 case DRR_END:
1595                 {
1596                         struct drr_end drre = drr->drr_u.drr_end;
1597                         /*
1598                          * We compare against the *previous* checksum
1599                          * value, because the stored checksum is of
1600                          * everything before the DRR_END record.
1601                          */
1602                         if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1603                                 ra.err = SET_ERROR(ECKSUM);
1604                         goto out;
1605                 }
1606                 case DRR_SPILL:
1607                 {
1608                         struct drr_spill drrs = drr->drr_u.drr_spill;
1609                         ra.err = restore_spill(&ra, os, &drrs);
1610                         break;
1611                 }
1612                 default:
1613                         ra.err = SET_ERROR(EINVAL);
1614                         goto out;
1615                 }
1616                 pcksum = ra.cksum;
1617         }
1618         ASSERT(ra.err != 0);
1619
1620 out:
1621         if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1622                 zfs_onexit_fd_rele(cleanup_fd);
1623
1624         if (ra.err != 0) {
1625                 /*
1626                  * destroy what we created, so we don't leave it in the
1627                  * inconsistent restoring state.
1628                  */
1629                 dmu_recv_cleanup_ds(drc);
1630         }
1631
1632         kmem_free(ra.buf, ra.bufsize);
1633         *voffp = ra.voff;
1634         return (ra.err);
1635 }
1636
1637 static int
1638 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1639 {
1640         dmu_recv_cookie_t *drc = arg;
1641         dsl_pool_t *dp = dmu_tx_pool(tx);
1642         int error;
1643
1644         ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1645
1646         if (!drc->drc_newfs) {
1647                 dsl_dataset_t *origin_head;
1648
1649                 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1650                 if (error != 0)
1651                         return (error);
1652                 if (drc->drc_force) {
1653                         /*
1654                          * We will destroy any snapshots in tofs (i.e. before
1655                          * origin_head) that are after the origin (which is
1656                          * the snap before drc_ds, because drc_ds can not
1657                          * have any snaps of its own).
1658                          */
1659                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1660                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1661                                 dsl_dataset_t *snap;
1662                                 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1663                                     &snap);
1664                                 if (error != 0)
1665                                         return (error);
1666                                 if (snap->ds_dir != origin_head->ds_dir)
1667                                         error = SET_ERROR(EINVAL);
1668                                 if (error == 0)  {
1669                                         error = dsl_destroy_snapshot_check_impl(
1670                                             snap, B_FALSE);
1671                                 }
1672                                 obj = snap->ds_phys->ds_prev_snap_obj;
1673                                 dsl_dataset_rele(snap, FTAG);
1674                                 if (error != 0)
1675                                         return (error);
1676                         }
1677                 }
1678                 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1679                     origin_head, drc->drc_force, drc->drc_owner, tx);
1680                 if (error != 0) {
1681                         dsl_dataset_rele(origin_head, FTAG);
1682                         return (error);
1683                 }
1684                 error = dsl_dataset_snapshot_check_impl(origin_head,
1685                     drc->drc_tosnap, tx, B_TRUE);
1686                 dsl_dataset_rele(origin_head, FTAG);
1687                 if (error != 0)
1688                         return (error);
1689
1690                 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1691         } else {
1692                 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1693                     drc->drc_tosnap, tx, B_TRUE);
1694         }
1695         return (error);
1696 }
1697
1698 static void
1699 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1700 {
1701         dmu_recv_cookie_t *drc = arg;
1702         dsl_pool_t *dp = dmu_tx_pool(tx);
1703
1704         spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1705             tx, "snap=%s", drc->drc_tosnap);
1706
1707         if (!drc->drc_newfs) {
1708                 dsl_dataset_t *origin_head;
1709
1710                 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1711                     &origin_head));
1712
1713                 if (drc->drc_force) {
1714                         /*
1715                          * Destroy any snapshots of drc_tofs (origin_head)
1716                          * after the origin (the snap before drc_ds).
1717                          */
1718                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1719                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1720                                 dsl_dataset_t *snap;
1721                                 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1722                                     &snap));
1723                                 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1724                                 obj = snap->ds_phys->ds_prev_snap_obj;
1725                                 dsl_destroy_snapshot_sync_impl(snap,
1726                                     B_FALSE, tx);
1727                                 dsl_dataset_rele(snap, FTAG);
1728                         }
1729                 }
1730                 VERIFY3P(drc->drc_ds->ds_prev, ==,
1731                     origin_head->ds_prev);
1732
1733                 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1734                     origin_head, tx);
1735                 dsl_dataset_snapshot_sync_impl(origin_head,
1736                     drc->drc_tosnap, tx);
1737
1738                 /* set snapshot's creation time and guid */
1739                 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1740                 origin_head->ds_prev->ds_phys->ds_creation_time =
1741                     drc->drc_drrb->drr_creation_time;
1742                 origin_head->ds_prev->ds_phys->ds_guid =
1743                     drc->drc_drrb->drr_toguid;
1744                 origin_head->ds_prev->ds_phys->ds_flags &=
1745                     ~DS_FLAG_INCONSISTENT;
1746
1747                 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1748                 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1749
1750                 dsl_dataset_rele(origin_head, FTAG);
1751                 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1752
1753                 if (drc->drc_owner != NULL)
1754                         VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
1755         } else {
1756                 dsl_dataset_t *ds = drc->drc_ds;
1757
1758                 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1759
1760                 /* set snapshot's creation time and guid */
1761                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1762                 ds->ds_prev->ds_phys->ds_creation_time =
1763                     drc->drc_drrb->drr_creation_time;
1764                 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1765                 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1766
1767                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1768                 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1769         }
1770         drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1771         /*
1772          * Release the hold from dmu_recv_begin.  This must be done before
1773          * we return to open context, so that when we free the dataset's dnode,
1774          * we can evict its bonus buffer.
1775          */
1776         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1777         drc->drc_ds = NULL;
1778 }
1779
1780 static int
1781 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1782 {
1783         dsl_pool_t *dp;
1784         dsl_dataset_t *snapds;
1785         guid_map_entry_t *gmep;
1786         int err;
1787
1788         ASSERT(guid_map != NULL);
1789
1790         err = dsl_pool_hold(name, FTAG, &dp);
1791         if (err != 0)
1792                 return (err);
1793         gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
1794         err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
1795         if (err == 0) {
1796                 gmep->guid = snapds->ds_phys->ds_guid;
1797                 gmep->gme_ds = snapds;
1798                 avl_add(guid_map, gmep);
1799                 dsl_dataset_long_hold(snapds, gmep);
1800         } else
1801                 kmem_free(gmep, sizeof (*gmep));
1802
1803         dsl_pool_rele(dp, FTAG);
1804         return (err);
1805 }
1806
1807 static int dmu_recv_end_modified_blocks = 3;
1808
1809 static int
1810 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1811 {
1812         int error;
1813         char name[MAXNAMELEN];
1814
1815 #ifdef _KERNEL
1816         /*
1817          * We will be destroying the ds; make sure its origin is unmounted if
1818          * necessary.
1819          */
1820         dsl_dataset_name(drc->drc_ds, name);
1821         zfs_destroy_unmount_origin(name);
1822 #endif
1823
1824         error = dsl_sync_task(drc->drc_tofs,
1825             dmu_recv_end_check, dmu_recv_end_sync, drc,
1826             dmu_recv_end_modified_blocks);
1827
1828         if (error != 0)
1829                 dmu_recv_cleanup_ds(drc);
1830         return (error);
1831 }
1832
1833 static int
1834 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1835 {
1836         int error;
1837
1838         error = dsl_sync_task(drc->drc_tofs,
1839             dmu_recv_end_check, dmu_recv_end_sync, drc,
1840             dmu_recv_end_modified_blocks);
1841
1842         if (error != 0) {
1843                 dmu_recv_cleanup_ds(drc);
1844         } else if (drc->drc_guid_to_ds_map != NULL) {
1845                 (void) add_ds_to_guidmap(drc->drc_tofs,
1846                     drc->drc_guid_to_ds_map,
1847                     drc->drc_newsnapobj);
1848         }
1849         return (error);
1850 }
1851
1852 int
1853 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
1854 {
1855         drc->drc_owner = owner;
1856
1857         if (drc->drc_newfs)
1858                 return (dmu_recv_new_end(drc));
1859         else
1860                 return (dmu_recv_existing_end(drc));
1861 }
1862
1863 /*
1864  * Return TRUE if this objset is currently being received into.
1865  */
1866 boolean_t
1867 dmu_objset_is_receiving(objset_t *os)
1868 {
1869         return (os->os_dsl_dataset != NULL &&
1870             os->os_dsl_dataset->ds_owner == dmu_recv_tag);
1871 }