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_IS_HOLE(bp) &&
 376             zb->zb_object == DMU_META_DNODE_OBJECT) {
 377                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 378                 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
 379                 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
 380         } else if (BP_IS_HOLE(bp)) {
 381                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 382                 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
 383         } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
 384                 return (0);
 385         } else if (type == DMU_OT_DNODE) {
 386                 dnode_phys_t *blk;
 387                 int i;
 388                 int blksz = BP_GET_LSIZE(bp);
 389                 uint32_t aflags = ARC_WAIT;
 390                 arc_buf_t *abuf;
 391
 392                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 393                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 394                     &aflags, zb) != 0)
 395                         return (SET_ERROR(EIO));
 396
 397                 blk = abuf->b_data;
 398                 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
 399                         uint64_t dnobj = (zb->zb_blkid <<
 400                             (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
 401                         err = dump_dnode(dsp, dnobj, blk+i);
 402                         if (err != 0)
 403                                 break;
 404                 }
 405                 (void) arc_buf_remove_ref(abuf, &abuf);
 406         } else if (type == DMU_OT_SA) {
 407                 uint32_t aflags = ARC_WAIT;
 408                 arc_buf_t *abuf;
 409                 int blksz = BP_GET_LSIZE(bp);
 410
 411                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 412                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 413                     &aflags, zb) != 0)
 414                         return (SET_ERROR(EIO));
 415
 416                 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
 417                 (void) arc_buf_remove_ref(abuf, &abuf);
 418         } else { /* it's a level-0 block of a regular object */
 419                 uint32_t aflags = ARC_WAIT;
 420                 arc_buf_t *abuf;
 421                 int blksz = BP_GET_LSIZE(bp);
 422
 423                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 424                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 425                     &aflags, zb) != 0) {
 426                         if (zfs_send_corrupt_data) {
 427                                 /* Send a block filled with 0x"zfs badd bloc" */
 428                                 abuf = arc_buf_alloc(spa, blksz, &abuf,
 429                                     ARC_BUFC_DATA);
 430                                 uint64_t *ptr;
 431                                 for (ptr = abuf->b_data;
 432                                     (char *)ptr < (char *)abuf->b_data + blksz;
 433                                     ptr++)
 434                                         *ptr = 0x2f5baddb10c;
 435                         } else {
 436                                 return (SET_ERROR(EIO));
 437                         }
 438                 }
 439
 440                 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
 441                     blksz, bp, abuf->b_data);
 442                 (void) arc_buf_remove_ref(abuf, &abuf);
 443         }
 444
 445         ASSERT(err == 0 || err == EINTR);
 446         return (err);
 447 }
 448
 449 /*
 450  * Releases dp, ds, and fromds, using the specified tag.
 451  */
 452 static int
 453 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 454 #ifdef illumos
 455     dsl_dataset_t *fromds, int outfd, vnode_t *vp, offset_t *off)
 456 #else
 457     dsl_dataset_t *fromds, int outfd, struct file *fp, offset_t *off)
 458 #endif
 459 {
 460         objset_t *os;
 461         dmu_replay_record_t *drr;
 462         dmu_sendarg_t *dsp;
 463         int err;
 464         uint64_t fromtxg = 0;
 465
 466         if (fromds != NULL && !dsl_dataset_is_before(ds, fromds)) {
 467                 dsl_dataset_rele(fromds, tag);
 468                 dsl_dataset_rele(ds, tag);
 469                 dsl_pool_rele(dp, tag);
 470                 return (SET_ERROR(EXDEV));
 471         }
 472
 473         err = dmu_objset_from_ds(ds, &os);
 474         if (err != 0) {
 475                 if (fromds != NULL)
 476                         dsl_dataset_rele(fromds, tag);
 477                 dsl_dataset_rele(ds, tag);
 478                 dsl_pool_rele(dp, tag);
 479                 return (err);
 480         }
 481
 482         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
 483         drr->drr_type = DRR_BEGIN;
 484         drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
 485         DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
 486             DMU_SUBSTREAM);
 487
 488 #ifdef _KERNEL
 489         if (dmu_objset_type(os) == DMU_OST_ZFS) {
 490                 uint64_t version;
 491                 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
 492                         kmem_free(drr, sizeof (dmu_replay_record_t));
 493                         if (fromds != NULL)
 494                                 dsl_dataset_rele(fromds, tag);
 495                         dsl_dataset_rele(ds, tag);
 496                         dsl_pool_rele(dp, tag);
 497                         return (SET_ERROR(EINVAL));
 498                 }
 499                 if (version >= ZPL_VERSION_SA) {
 500                         DMU_SET_FEATUREFLAGS(
 501                             drr->drr_u.drr_begin.drr_versioninfo,
 502                             DMU_BACKUP_FEATURE_SA_SPILL);
 503                 }
 504         }
 505 #endif
 506
 507         drr->drr_u.drr_begin.drr_creation_time =
 508             ds->ds_phys->ds_creation_time;
 509         drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
 510         if (fromds != NULL && ds->ds_dir != fromds->ds_dir)
 511                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
 512         drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
 513         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 514                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
 515
 516         if (fromds != NULL)
 517                 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
 518         dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
 519
 520         if (fromds != NULL) {
 521                 fromtxg = fromds->ds_phys->ds_creation_txg;
 522                 dsl_dataset_rele(fromds, tag);
 523                 fromds = NULL;
 524         }
 525
 526         dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
 527
 528         dsp->dsa_drr = drr;
 529         dsp->dsa_outfd = outfd;
 530         dsp->dsa_proc = curproc;
 531         dsp->dsa_td = curthread;
 532         dsp->dsa_fp = fp;
 533         dsp->dsa_os = os;
 534         dsp->dsa_off = off;
 535         dsp->dsa_toguid = ds->ds_phys->ds_guid;
 536         ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
 537         dsp->dsa_pending_op = PENDING_NONE;
 538         dsp->dsa_incremental = (fromtxg != 0);
 539
 540         mutex_enter(&ds->ds_sendstream_lock);
 541         list_insert_head(&ds->ds_sendstreams, dsp);
 542         mutex_exit(&ds->ds_sendstream_lock);
 543
 544         dsl_dataset_long_hold(ds, FTAG);
 545         dsl_pool_rele(dp, tag);
 546
 547         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 548                 err = dsp->dsa_err;
 549                 goto out;
 550         }
 551
 552         err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
 553             backup_cb, dsp);
 554
 555         if (dsp->dsa_pending_op != PENDING_NONE)
 556                 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
 557                         err = SET_ERROR(EINTR);
 558
 559         if (err != 0) {
 560                 if (err == EINTR && dsp->dsa_err != 0)
 561                         err = dsp->dsa_err;
 562                 goto out;
 563         }
 564
 565         bzero(drr, sizeof (dmu_replay_record_t));
 566         drr->drr_type = DRR_END;
 567         drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
 568         drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
 569
 570         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 571                 err = dsp->dsa_err;
 572                 goto out;
 573         }
 574
 575 out:
 576         mutex_enter(&ds->ds_sendstream_lock);
 577         list_remove(&ds->ds_sendstreams, dsp);
 578         mutex_exit(&ds->ds_sendstream_lock);
 579
 580         kmem_free(drr, sizeof (dmu_replay_record_t));
 581         kmem_free(dsp, sizeof (dmu_sendarg_t));
 582
 583         dsl_dataset_long_rele(ds, FTAG);
 584         dsl_dataset_rele(ds, tag);
 585
 586         return (err);
 587 }
 588
 589 int
 590 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
 591 #ifdef illumos
 592     int outfd, vnode_t *vp, offset_t *off)
 593 #else
 594     int outfd, struct file *fp, offset_t *off)
 595 #endif
 596 {
 597         dsl_pool_t *dp;
 598         dsl_dataset_t *ds;
 599         dsl_dataset_t *fromds = NULL;
 600         int err;
 601
 602         err = dsl_pool_hold(pool, FTAG, &dp);
 603         if (err != 0)
 604                 return (err);
 605
 606         err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
 607         if (err != 0) {
 608                 dsl_pool_rele(dp, FTAG);
 609                 return (err);
 610         }
 611
 612         if (fromsnap != 0) {
 613                 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
 614                 if (err != 0) {
 615                         dsl_dataset_rele(ds, FTAG);
 616                         dsl_pool_rele(dp, FTAG);
 617                         return (err);
 618                 }
 619         }
 620
 621         return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, fp, off));
 622 }
 623
 624 int
 625 dmu_send(const char *tosnap, const char *fromsnap,
 626 #ifdef illumos
 627     int outfd, vnode_t *vp, offset_t *off)
 628 #else
 629     int outfd, struct file *fp, offset_t *off)
 630 #endif
 631 {
 632         dsl_pool_t *dp;
 633         dsl_dataset_t *ds;
 634         dsl_dataset_t *fromds = NULL;
 635         int err;
 636
 637         if (strchr(tosnap, '@') == NULL)
 638                 return (SET_ERROR(EINVAL));
 639         if (fromsnap != NULL && strchr(fromsnap, '@') == NULL)
 640                 return (SET_ERROR(EINVAL));
 641
 642         err = dsl_pool_hold(tosnap, FTAG, &dp);
 643         if (err != 0)
 644                 return (err);
 645
 646         err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
 647         if (err != 0) {
 648                 dsl_pool_rele(dp, FTAG);
 649                 return (err);
 650         }
 651
 652         if (fromsnap != NULL) {
 653                 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
 654                 if (err != 0) {
 655                         dsl_dataset_rele(ds, FTAG);
 656                         dsl_pool_rele(dp, FTAG);
 657                         return (err);
 658                 }
 659         }
 660         return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, fp, off));
 661 }
 662
 663 int
 664 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
 665 {
 666         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 667         int err;
 668         uint64_t size;
 669
 670         ASSERT(dsl_pool_config_held(dp));
 671
 672         /* tosnap must be a snapshot */
 673         if (!dsl_dataset_is_snapshot(ds))
 674                 return (SET_ERROR(EINVAL));
 675
 676         /*
 677          * fromsnap must be an earlier snapshot from the same fs as tosnap,
 678          * or the origin's fs.
 679          */
 680         if (fromds != NULL && !dsl_dataset_is_before(ds, fromds))
 681                 return (SET_ERROR(EXDEV));
 682
 683         /* Get uncompressed size estimate of changed data. */
 684         if (fromds == NULL) {
 685                 size = ds->ds_phys->ds_uncompressed_bytes;
 686         } else {
 687                 uint64_t used, comp;
 688                 err = dsl_dataset_space_written(fromds, ds,
 689                     &used, &comp, &size);
 690                 if (err != 0)
 691                         return (err);
 692         }
 693
 694         /*
 695          * Assume that space (both on-disk and in-stream) is dominated by
 696          * data.  We will adjust for indirect blocks and the copies property,
 697          * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
 698          */
 699
 700         /*
 701          * Subtract out approximate space used by indirect blocks.
 702          * Assume most space is used by data blocks (non-indirect, non-dnode).
 703          * Assume all blocks are recordsize.  Assume ditto blocks and
 704          * internal fragmentation counter out compression.
 705          *
 706          * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
 707          * block, which we observe in practice.
 708          */
 709         uint64_t recordsize;
 710         err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
 711         if (err != 0)
 712                 return (err);
 713         size -= size / recordsize * sizeof (blkptr_t);
 714
 715         /* Add in the space for the record associated with each block. */
 716         size += size / recordsize * sizeof (dmu_replay_record_t);
 717
 718         *sizep = size;
 719
 720         return (0);
 721 }
 722
 723 typedef struct dmu_recv_begin_arg {
 724         const char *drba_origin;
 725         dmu_recv_cookie_t *drba_cookie;
 726         cred_t *drba_cred;
 727         uint64_t drba_snapobj;
 728 } dmu_recv_begin_arg_t;
 729
 730 static int
 731 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
 732     uint64_t fromguid)
 733 {
 734         uint64_t val;
 735         int error;
 736         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 737
 738         /* temporary clone name must not exist */
 739         error = zap_lookup(dp->dp_meta_objset,
 740             ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
 741             8, 1, &val);
 742         if (error != ENOENT)
 743                 return (error == 0 ? EBUSY : error);
 744
 745         /* new snapshot name must not exist */
 746         error = zap_lookup(dp->dp_meta_objset,
 747             ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
 748             8, 1, &val);
 749         if (error != ENOENT)
 750                 return (error == 0 ? EEXIST : error);
 751
 752         if (fromguid != 0) {
 753                 dsl_dataset_t *snap;
 754                 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
 755
 756                 /* Find snapshot in this dir that matches fromguid. */
 757                 while (obj != 0) {
 758                         error = dsl_dataset_hold_obj(dp, obj, FTAG,
 759                             &snap);
 760                         if (error != 0)
 761                                 return (SET_ERROR(ENODEV));
 762                         if (snap->ds_dir != ds->ds_dir) {
 763                                 dsl_dataset_rele(snap, FTAG);
 764                                 return (SET_ERROR(ENODEV));
 765                         }
 766                         if (snap->ds_phys->ds_guid == fromguid)
 767                                 break;
 768                         obj = snap->ds_phys->ds_prev_snap_obj;
 769                         dsl_dataset_rele(snap, FTAG);
 770                 }
 771                 if (obj == 0)
 772                         return (SET_ERROR(ENODEV));
 773
 774                 if (drba->drba_cookie->drc_force) {
 775                         drba->drba_snapobj = obj;
 776                 } else {
 777                         /*
 778                          * If we are not forcing, there must be no
 779                          * changes since fromsnap.
 780                          */
 781                         if (dsl_dataset_modified_since_snap(ds, snap)) {
 782                                 dsl_dataset_rele(snap, FTAG);
 783                                 return (SET_ERROR(ETXTBSY));
 784                         }
 785                         drba->drba_snapobj = ds->ds_prev->ds_object;
 786                 }
 787
 788                 dsl_dataset_rele(snap, FTAG);
 789         } else {
 790                 /* if full, most recent snapshot must be $ORIGIN */
 791                 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
 792                         return (SET_ERROR(ENODEV));
 793                 drba->drba_snapobj = ds->ds_phys->ds_prev_snap_obj;
 794         }
 795
 796         return (0);
 797
 798 }
 799
 800 static int
 801 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 802 {
 803         dmu_recv_begin_arg_t *drba = arg;
 804         dsl_pool_t *dp = dmu_tx_pool(tx);
 805         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 806         uint64_t fromguid = drrb->drr_fromguid;
 807         int flags = drrb->drr_flags;
 808         int error;
 809         dsl_dataset_t *ds;
 810         const char *tofs = drba->drba_cookie->drc_tofs;
 811
 812         /* already checked */
 813         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 814
 815         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 816             DMU_COMPOUNDSTREAM ||
 817             drrb->drr_type >= DMU_OST_NUMTYPES ||
 818             ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
 819                 return (SET_ERROR(EINVAL));
 820
 821         /* Verify pool version supports SA if SA_SPILL feature set */
 822         if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
 823             DMU_BACKUP_FEATURE_SA_SPILL) &&
 824             spa_version(dp->dp_spa) < SPA_VERSION_SA) {
 825                 return (SET_ERROR(ENOTSUP));
 826         }
 827
 828         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 829         if (error == 0) {
 830                 /* target fs already exists; recv into temp clone */
 831
 832                 /* Can't recv a clone into an existing fs */
 833                 if (flags & DRR_FLAG_CLONE) {
 834                         dsl_dataset_rele(ds, FTAG);
 835                         return (SET_ERROR(EINVAL));
 836                 }
 837
 838                 error = recv_begin_check_existing_impl(drba, ds, fromguid);
 839                 dsl_dataset_rele(ds, FTAG);
 840         } else if (error == ENOENT) {
 841                 /* target fs does not exist; must be a full backup or clone */
 842                 char buf[MAXNAMELEN];
 843
 844                 /*
 845                  * If it's a non-clone incremental, we are missing the
 846                  * target fs, so fail the recv.
 847                  */
 848                 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
 849                         return (SET_ERROR(ENOENT));
 850
 851                 /* Open the parent of tofs */
 852                 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
 853                 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
 854                 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
 855                 if (error != 0)
 856                         return (error);
 857
 858                 if (drba->drba_origin != NULL) {
 859                         dsl_dataset_t *origin;
 860                         error = dsl_dataset_hold(dp, drba->drba_origin,
 861                             FTAG, &origin);
 862                         if (error != 0) {
 863                                 dsl_dataset_rele(ds, FTAG);
 864                                 return (error);
 865                         }
 866                         if (!dsl_dataset_is_snapshot(origin)) {
 867                                 dsl_dataset_rele(origin, FTAG);
 868                                 dsl_dataset_rele(ds, FTAG);
 869                                 return (SET_ERROR(EINVAL));
 870                         }
 871                         if (origin->ds_phys->ds_guid != fromguid) {
 872                                 dsl_dataset_rele(origin, FTAG);
 873                                 dsl_dataset_rele(ds, FTAG);
 874                                 return (SET_ERROR(ENODEV));
 875                         }
 876                         dsl_dataset_rele(origin, FTAG);
 877                 }
 878                 dsl_dataset_rele(ds, FTAG);
 879                 error = 0;
 880         }
 881         return (error);
 882 }
 883
 884 static void
 885 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
 886 {
 887         dmu_recv_begin_arg_t *drba = arg;
 888         dsl_pool_t *dp = dmu_tx_pool(tx);
 889         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 890         const char *tofs = drba->drba_cookie->drc_tofs;
 891         dsl_dataset_t *ds, *newds;
 892         uint64_t dsobj;
 893         int error;
 894         uint64_t crflags;
 895
 896         crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
 897             DS_FLAG_CI_DATASET : 0;
 898
 899         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 900         if (error == 0) {
 901                 /* create temporary clone */
 902                 dsl_dataset_t *snap = NULL;
 903                 if (drba->drba_snapobj != 0) {
 904                         VERIFY0(dsl_dataset_hold_obj(dp,
 905                             drba->drba_snapobj, FTAG, &snap));
 906                 }
 907                 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
 908                     snap, crflags, drba->drba_cred, tx);
 909                 dsl_dataset_rele(snap, FTAG);
 910                 dsl_dataset_rele(ds, FTAG);
 911         } else {
 912                 dsl_dir_t *dd;
 913                 const char *tail;
 914                 dsl_dataset_t *origin = NULL;
 915
 916                 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
 917
 918                 if (drba->drba_origin != NULL) {
 919                         VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
 920                             FTAG, &origin));
 921                 }
 922
 923                 /* Create new dataset. */
 924                 dsobj = dsl_dataset_create_sync(dd,
 925                     strrchr(tofs, '/') + 1,
 926                     origin, crflags, drba->drba_cred, tx);
 927                 if (origin != NULL)
 928                         dsl_dataset_rele(origin, FTAG);
 929                 dsl_dir_rele(dd, FTAG);
 930                 drba->drba_cookie->drc_newfs = B_TRUE;
 931         }
 932         VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
 933
 934         dmu_buf_will_dirty(newds->ds_dbuf, tx);
 935         newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
 936
 937         /*
 938          * If we actually created a non-clone, we need to create the
 939          * objset in our new dataset.
 940          */
 941         if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
 942                 (void) dmu_objset_create_impl(dp->dp_spa,
 943                     newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
 944         }
 945
 946         drba->drba_cookie->drc_ds = newds;
 947
 948         spa_history_log_internal_ds(newds, "receive", tx, "");
 949 }
 950
 951 /*
 952  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
 953  * succeeds; otherwise we will leak the holds on the datasets.
 954  */
 955 int
 956 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
 957     boolean_t force, char *origin, dmu_recv_cookie_t *drc)
 958 {
 959         dmu_recv_begin_arg_t drba = { 0 };
 960         dmu_replay_record_t *drr;
 961
 962         bzero(drc, sizeof (dmu_recv_cookie_t));
 963         drc->drc_drrb = drrb;
 964         drc->drc_tosnap = tosnap;
 965         drc->drc_tofs = tofs;
 966         drc->drc_force = force;
 967
 968         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
 969                 drc->drc_byteswap = B_TRUE;
 970         else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
 971                 return (SET_ERROR(EINVAL));
 972
 973         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
 974         drr->drr_type = DRR_BEGIN;
 975         drr->drr_u.drr_begin = *drc->drc_drrb;
 976         if (drc->drc_byteswap) {
 977                 fletcher_4_incremental_byteswap(drr,
 978                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
 979         } else {
 980                 fletcher_4_incremental_native(drr,
 981                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
 982         }
 983         kmem_free(drr, sizeof (dmu_replay_record_t));
 984
 985         if (drc->drc_byteswap) {
 986                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
 987                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
 988                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
 989                 drrb->drr_type = BSWAP_32(drrb->drr_type);
 990                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
 991                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
 992         }
 993
 994         drba.drba_origin = origin;
 995         drba.drba_cookie = drc;
 996         drba.drba_cred = CRED();
 997
 998         return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
 999             &drba, 5));
1000 }
1001
1002 struct restorearg {
1003         int err;
1004         boolean_t byteswap;
1005         kthread_t *td;
1006         struct file *fp;
1007         char *buf;
1008         uint64_t voff;
1009         int bufsize; /* amount of memory allocated for buf */
1010         zio_cksum_t cksum;
1011         avl_tree_t *guid_to_ds_map;
1012 };
1013
1014 typedef struct guid_map_entry {
1015         uint64_t        guid;
1016         dsl_dataset_t   *gme_ds;
1017         avl_node_t      avlnode;
1018 } guid_map_entry_t;
1019
1020 static int
1021 guid_compare(const void *arg1, const void *arg2)
1022 {
1023         const guid_map_entry_t *gmep1 = arg1;
1024         const guid_map_entry_t *gmep2 = arg2;
1025
1026         if (gmep1->guid < gmep2->guid)
1027                 return (-1);
1028         else if (gmep1->guid > gmep2->guid)
1029                 return (1);
1030         return (0);
1031 }
1032
1033 static void
1034 free_guid_map_onexit(void *arg)
1035 {
1036         avl_tree_t *ca = arg;
1037         void *cookie = NULL;
1038         guid_map_entry_t *gmep;
1039
1040         while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1041                 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1042                 dsl_dataset_rele(gmep->gme_ds, gmep);
1043                 kmem_free(gmep, sizeof (guid_map_entry_t));
1044         }
1045         avl_destroy(ca);
1046         kmem_free(ca, sizeof (avl_tree_t));
1047 }
1048
1049 static int
1050 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
1051 {
1052         struct uio auio;
1053         struct iovec aiov;
1054         int error;
1055
1056         aiov.iov_base = buf;
1057         aiov.iov_len = len;
1058         auio.uio_iov = &aiov;
1059         auio.uio_iovcnt = 1;
1060         auio.uio_resid = len;
1061         auio.uio_segflg = UIO_SYSSPACE;
1062         auio.uio_rw = UIO_READ;
1063         auio.uio_offset = off;
1064         auio.uio_td = ra->td;
1065 #ifdef _KERNEL
1066         error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
1067 #else
1068         fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
1069         error = EOPNOTSUPP;
1070 #endif
1071         *resid = auio.uio_resid;
1072         return (error);
1073 }
1074
1075 static void *
1076 restore_read(struct restorearg *ra, int len)
1077 {
1078         void *rv;
1079         int done = 0;
1080
1081         /* some things will require 8-byte alignment, so everything must */
1082         ASSERT0(len % 8);
1083
1084         while (done < len) {
1085                 ssize_t resid;
1086
1087                 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
1088                     len - done, ra->voff, &resid);
1089
1090                 if (resid == len - done)
1091                         ra->err = SET_ERROR(EINVAL);
1092                 ra->voff += len - done - resid;
1093                 done = len - resid;
1094                 if (ra->err != 0)
1095                         return (NULL);
1096         }
1097
1098         ASSERT3U(done, ==, len);
1099         rv = ra->buf;
1100         if (ra->byteswap)
1101                 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1102         else
1103                 fletcher_4_incremental_native(rv, len, &ra->cksum);
1104         return (rv);
1105 }
1106
1107 static void
1108 backup_byteswap(dmu_replay_record_t *drr)
1109 {
1110 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1111 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1112         drr->drr_type = BSWAP_32(drr->drr_type);
1113         drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1114         switch (drr->drr_type) {
1115         case DRR_BEGIN:
1116                 DO64(drr_begin.drr_magic);
1117                 DO64(drr_begin.drr_versioninfo);
1118                 DO64(drr_begin.drr_creation_time);
1119                 DO32(drr_begin.drr_type);
1120                 DO32(drr_begin.drr_flags);
1121                 DO64(drr_begin.drr_toguid);
1122                 DO64(drr_begin.drr_fromguid);
1123                 break;
1124         case DRR_OBJECT:
1125                 DO64(drr_object.drr_object);
1126                 /* DO64(drr_object.drr_allocation_txg); */
1127                 DO32(drr_object.drr_type);
1128                 DO32(drr_object.drr_bonustype);
1129                 DO32(drr_object.drr_blksz);
1130                 DO32(drr_object.drr_bonuslen);
1131                 DO64(drr_object.drr_toguid);
1132                 break;
1133         case DRR_FREEOBJECTS:
1134                 DO64(drr_freeobjects.drr_firstobj);
1135                 DO64(drr_freeobjects.drr_numobjs);
1136                 DO64(drr_freeobjects.drr_toguid);
1137                 break;
1138         case DRR_WRITE:
1139                 DO64(drr_write.drr_object);
1140                 DO32(drr_write.drr_type);
1141                 DO64(drr_write.drr_offset);
1142                 DO64(drr_write.drr_length);
1143                 DO64(drr_write.drr_toguid);
1144                 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1145                 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1146                 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1147                 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1148                 DO64(drr_write.drr_key.ddk_prop);
1149                 break;
1150         case DRR_WRITE_BYREF:
1151                 DO64(drr_write_byref.drr_object);
1152                 DO64(drr_write_byref.drr_offset);
1153                 DO64(drr_write_byref.drr_length);
1154                 DO64(drr_write_byref.drr_toguid);
1155                 DO64(drr_write_byref.drr_refguid);
1156                 DO64(drr_write_byref.drr_refobject);
1157                 DO64(drr_write_byref.drr_refoffset);
1158                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1159                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1160                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1161                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1162                 DO64(drr_write_byref.drr_key.ddk_prop);
1163                 break;
1164         case DRR_FREE:
1165                 DO64(drr_free.drr_object);
1166                 DO64(drr_free.drr_offset);
1167                 DO64(drr_free.drr_length);
1168                 DO64(drr_free.drr_toguid);
1169                 break;
1170         case DRR_SPILL:
1171                 DO64(drr_spill.drr_object);
1172                 DO64(drr_spill.drr_length);
1173                 DO64(drr_spill.drr_toguid);
1174                 break;
1175         case DRR_END:
1176                 DO64(drr_end.drr_checksum.zc_word[0]);
1177                 DO64(drr_end.drr_checksum.zc_word[1]);
1178                 DO64(drr_end.drr_checksum.zc_word[2]);
1179                 DO64(drr_end.drr_checksum.zc_word[3]);
1180                 DO64(drr_end.drr_toguid);
1181                 break;
1182         }
1183 #undef DO64
1184 #undef DO32
1185 }
1186
1187 static int
1188 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1189 {
1190         int err;
1191         dmu_tx_t *tx;
1192         void *data = NULL;
1193
1194         if (drro->drr_type == DMU_OT_NONE ||
1195             !DMU_OT_IS_VALID(drro->drr_type) ||
1196             !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1197             drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1198             drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1199             P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1200             drro->drr_blksz < SPA_MINBLOCKSIZE ||
1201             drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1202             drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1203                 return (SET_ERROR(EINVAL));
1204         }
1205
1206         err = dmu_object_info(os, drro->drr_object, NULL);
1207
1208         if (err != 0 && err != ENOENT)
1209                 return (SET_ERROR(EINVAL));
1210
1211         if (drro->drr_bonuslen) {
1212                 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1213                 if (ra->err != 0)
1214                         return (ra->err);
1215         }
1216
1217         if (err == ENOENT) {
1218                 /* currently free, want to be allocated */
1219                 tx = dmu_tx_create(os);
1220                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1221                 err = dmu_tx_assign(tx, TXG_WAIT);
1222                 if (err != 0) {
1223                         dmu_tx_abort(tx);
1224                         return (err);
1225                 }
1226                 err = dmu_object_claim(os, drro->drr_object,
1227                     drro->drr_type, drro->drr_blksz,
1228                     drro->drr_bonustype, drro->drr_bonuslen, tx);
1229                 dmu_tx_commit(tx);
1230         } else {
1231                 /* currently allocated, want to be allocated */
1232                 err = dmu_object_reclaim(os, drro->drr_object,
1233                     drro->drr_type, drro->drr_blksz,
1234                     drro->drr_bonustype, drro->drr_bonuslen);
1235         }
1236         if (err != 0) {
1237                 return (SET_ERROR(EINVAL));
1238         }
1239
1240         tx = dmu_tx_create(os);
1241         dmu_tx_hold_bonus(tx, drro->drr_object);
1242         err = dmu_tx_assign(tx, TXG_WAIT);
1243         if (err != 0) {
1244                 dmu_tx_abort(tx);
1245                 return (err);
1246         }
1247
1248         dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1249             tx);
1250         dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1251
1252         if (data != NULL) {
1253                 dmu_buf_t *db;
1254
1255                 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1256                 dmu_buf_will_dirty(db, tx);
1257
1258                 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1259                 bcopy(data, db->db_data, drro->drr_bonuslen);
1260                 if (ra->byteswap) {
1261                         dmu_object_byteswap_t byteswap =
1262                             DMU_OT_BYTESWAP(drro->drr_bonustype);
1263                         dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1264                             drro->drr_bonuslen);
1265                 }
1266                 dmu_buf_rele(db, FTAG);
1267         }
1268         dmu_tx_commit(tx);
1269         return (0);
1270 }
1271
1272 /* ARGSUSED */
1273 static int
1274 restore_freeobjects(struct restorearg *ra, objset_t *os,
1275     struct drr_freeobjects *drrfo)
1276 {
1277         uint64_t obj;
1278
1279         if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1280                 return (SET_ERROR(EINVAL));
1281
1282         for (obj = drrfo->drr_firstobj;
1283             obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1284             (void) dmu_object_next(os, &obj, FALSE, 0)) {
1285                 int err;
1286
1287                 if (dmu_object_info(os, obj, NULL) != 0)
1288                         continue;
1289
1290                 err = dmu_free_long_object(os, obj);
1291                 if (err != 0)
1292                         return (err);
1293         }
1294         return (0);
1295 }
1296
1297 static int
1298 restore_write(struct restorearg *ra, objset_t *os,
1299     struct drr_write *drrw)
1300 {
1301         dmu_tx_t *tx;
1302         void *data;
1303         int err;
1304
1305         if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1306             !DMU_OT_IS_VALID(drrw->drr_type))
1307                 return (SET_ERROR(EINVAL));
1308
1309         data = restore_read(ra, drrw->drr_length);
1310         if (data == NULL)
1311                 return (ra->err);
1312
1313         if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1314                 return (SET_ERROR(EINVAL));
1315
1316         tx = dmu_tx_create(os);
1317
1318         dmu_tx_hold_write(tx, drrw->drr_object,
1319             drrw->drr_offset, drrw->drr_length);
1320         err = dmu_tx_assign(tx, TXG_WAIT);
1321         if (err != 0) {
1322                 dmu_tx_abort(tx);
1323                 return (err);
1324         }
1325         if (ra->byteswap) {
1326                 dmu_object_byteswap_t byteswap =
1327                     DMU_OT_BYTESWAP(drrw->drr_type);
1328                 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1329         }
1330         dmu_write(os, drrw->drr_object,
1331             drrw->drr_offset, drrw->drr_length, data, tx);
1332         dmu_tx_commit(tx);
1333         return (0);
1334 }
1335
1336 /*
1337  * Handle a DRR_WRITE_BYREF record.  This record is used in dedup'ed
1338  * streams to refer to a copy of the data that is already on the
1339  * system because it came in earlier in the stream.  This function
1340  * finds the earlier copy of the data, and uses that copy instead of
1341  * data from the stream to fulfill this write.
1342  */
1343 static int
1344 restore_write_byref(struct restorearg *ra, objset_t *os,
1345     struct drr_write_byref *drrwbr)
1346 {
1347         dmu_tx_t *tx;
1348         int err;
1349         guid_map_entry_t gmesrch;
1350         guid_map_entry_t *gmep;
1351         avl_index_t     where;
1352         objset_t *ref_os = NULL;
1353         dmu_buf_t *dbp;
1354
1355         if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1356                 return (SET_ERROR(EINVAL));
1357
1358         /*
1359          * If the GUID of the referenced dataset is different from the
1360          * GUID of the target dataset, find the referenced dataset.
1361          */
1362         if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1363                 gmesrch.guid = drrwbr->drr_refguid;
1364                 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1365                     &where)) == NULL) {
1366                         return (SET_ERROR(EINVAL));
1367                 }
1368                 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1369                         return (SET_ERROR(EINVAL));
1370         } else {
1371                 ref_os = os;
1372         }
1373
1374         if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1375             drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1376                 return (err);
1377
1378         tx = dmu_tx_create(os);
1379
1380         dmu_tx_hold_write(tx, drrwbr->drr_object,
1381             drrwbr->drr_offset, drrwbr->drr_length);
1382         err = dmu_tx_assign(tx, TXG_WAIT);
1383         if (err != 0) {
1384                 dmu_tx_abort(tx);
1385                 return (err);
1386         }
1387         dmu_write(os, drrwbr->drr_object,
1388             drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1389         dmu_buf_rele(dbp, FTAG);
1390         dmu_tx_commit(tx);
1391         return (0);
1392 }
1393
1394 static int
1395 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1396 {
1397         dmu_tx_t *tx;
1398         void *data;
1399         dmu_buf_t *db, *db_spill;
1400         int err;
1401
1402         if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1403             drrs->drr_length > SPA_MAXBLOCKSIZE)
1404                 return (SET_ERROR(EINVAL));
1405
1406         data = restore_read(ra, drrs->drr_length);
1407         if (data == NULL)
1408                 return (ra->err);
1409
1410         if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1411                 return (SET_ERROR(EINVAL));
1412
1413         VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1414         if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1415                 dmu_buf_rele(db, FTAG);
1416                 return (err);
1417         }
1418
1419         tx = dmu_tx_create(os);
1420
1421         dmu_tx_hold_spill(tx, db->db_object);
1422
1423         err = dmu_tx_assign(tx, TXG_WAIT);
1424         if (err != 0) {
1425                 dmu_buf_rele(db, FTAG);
1426                 dmu_buf_rele(db_spill, FTAG);
1427                 dmu_tx_abort(tx);
1428                 return (err);
1429         }
1430         dmu_buf_will_dirty(db_spill, tx);
1431
1432         if (db_spill->db_size < drrs->drr_length)
1433                 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1434                     drrs->drr_length, tx));
1435         bcopy(data, db_spill->db_data, drrs->drr_length);
1436
1437         dmu_buf_rele(db, FTAG);
1438         dmu_buf_rele(db_spill, FTAG);
1439
1440         dmu_tx_commit(tx);
1441         return (0);
1442 }
1443
1444 /* ARGSUSED */
1445 static int
1446 restore_free(struct restorearg *ra, objset_t *os,
1447     struct drr_free *drrf)
1448 {
1449         int err;
1450
1451         if (drrf->drr_length != -1ULL &&
1452             drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1453                 return (SET_ERROR(EINVAL));
1454
1455         if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1456                 return (SET_ERROR(EINVAL));
1457
1458         err = dmu_free_long_range(os, drrf->drr_object,
1459             drrf->drr_offset, drrf->drr_length);
1460         return (err);
1461 }
1462
1463 /* used to destroy the drc_ds on error */
1464 static void
1465 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1466 {
1467         char name[MAXNAMELEN];
1468         dsl_dataset_name(drc->drc_ds, name);
1469         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1470         (void) dsl_destroy_head(name);
1471 }
1472
1473 /*
1474  * NB: callers *must* call dmu_recv_end() if this succeeds.
1475  */
1476 int
1477 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1478     int cleanup_fd, uint64_t *action_handlep)
1479 {
1480         struct restorearg ra = { 0 };
1481         dmu_replay_record_t *drr;
1482         objset_t *os;
1483         zio_cksum_t pcksum;
1484         int featureflags;
1485
1486         ra.byteswap = drc->drc_byteswap;
1487         ra.cksum = drc->drc_cksum;
1488         ra.td = curthread;
1489         ra.fp = fp;
1490         ra.voff = *voffp;
1491         ra.bufsize = 1<<20;
1492         ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1493
1494         /* these were verified in dmu_recv_begin */
1495         ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1496             DMU_SUBSTREAM);
1497         ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1498
1499         /*
1500          * Open the objset we are modifying.
1501          */
1502         VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1503
1504         ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1505
1506         featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1507
1508         /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1509         if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1510                 minor_t minor;
1511
1512                 if (cleanup_fd == -1) {
1513                         ra.err = SET_ERROR(EBADF);
1514                         goto out;
1515                 }
1516                 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1517                 if (ra.err != 0) {
1518                         cleanup_fd = -1;
1519                         goto out;
1520                 }
1521
1522                 if (*action_handlep == 0) {
1523                         ra.guid_to_ds_map =
1524                             kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1525                         avl_create(ra.guid_to_ds_map, guid_compare,
1526                             sizeof (guid_map_entry_t),
1527                             offsetof(guid_map_entry_t, avlnode));
1528                         ra.err = zfs_onexit_add_cb(minor,
1529                             free_guid_map_onexit, ra.guid_to_ds_map,
1530                             action_handlep);
1531                         if (ra.err != 0)
1532                                 goto out;
1533                 } else {
1534                         ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1535                             (void **)&ra.guid_to_ds_map);
1536                         if (ra.err != 0)
1537                                 goto out;
1538                 }
1539
1540                 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1541         }
1542
1543         /*
1544          * Read records and process them.
1545          */
1546         pcksum = ra.cksum;
1547         while (ra.err == 0 &&
1548             NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1549                 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1550                         ra.err = SET_ERROR(EINTR);
1551                         goto out;
1552                 }
1553
1554                 if (ra.byteswap)
1555                         backup_byteswap(drr);
1556
1557                 switch (drr->drr_type) {
1558                 case DRR_OBJECT:
1559                 {
1560                         /*
1561                          * We need to make a copy of the record header,
1562                          * because restore_{object,write} may need to
1563                          * restore_read(), which will invalidate drr.
1564                          */
1565                         struct drr_object drro = drr->drr_u.drr_object;
1566                         ra.err = restore_object(&ra, os, &drro);
1567                         break;
1568                 }
1569                 case DRR_FREEOBJECTS:
1570                 {
1571                         struct drr_freeobjects drrfo =
1572                             drr->drr_u.drr_freeobjects;
1573                         ra.err = restore_freeobjects(&ra, os, &drrfo);
1574                         break;
1575                 }
1576                 case DRR_WRITE:
1577                 {
1578                         struct drr_write drrw = drr->drr_u.drr_write;
1579                         ra.err = restore_write(&ra, os, &drrw);
1580                         break;
1581                 }
1582                 case DRR_WRITE_BYREF:
1583                 {
1584                         struct drr_write_byref drrwbr =
1585                             drr->drr_u.drr_write_byref;
1586                         ra.err = restore_write_byref(&ra, os, &drrwbr);
1587                         break;
1588                 }
1589                 case DRR_FREE:
1590                 {
1591                         struct drr_free drrf = drr->drr_u.drr_free;
1592                         ra.err = restore_free(&ra, os, &drrf);
1593                         break;
1594                 }
1595                 case DRR_END:
1596                 {
1597                         struct drr_end drre = drr->drr_u.drr_end;
1598                         /*
1599                          * We compare against the *previous* checksum
1600                          * value, because the stored checksum is of
1601                          * everything before the DRR_END record.
1602                          */
1603                         if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1604                                 ra.err = SET_ERROR(ECKSUM);
1605                         goto out;
1606                 }
1607                 case DRR_SPILL:
1608                 {
1609                         struct drr_spill drrs = drr->drr_u.drr_spill;
1610                         ra.err = restore_spill(&ra, os, &drrs);
1611                         break;
1612                 }
1613                 default:
1614                         ra.err = SET_ERROR(EINVAL);
1615                         goto out;
1616                 }
1617                 pcksum = ra.cksum;
1618         }
1619         ASSERT(ra.err != 0);
1620
1621 out:
1622         if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1623                 zfs_onexit_fd_rele(cleanup_fd);
1624
1625         if (ra.err != 0) {
1626                 /*
1627                  * destroy what we created, so we don't leave it in the
1628                  * inconsistent restoring state.
1629                  */
1630                 dmu_recv_cleanup_ds(drc);
1631         }
1632
1633         kmem_free(ra.buf, ra.bufsize);
1634         *voffp = ra.voff;
1635         return (ra.err);
1636 }
1637
1638 static int
1639 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1640 {
1641         dmu_recv_cookie_t *drc = arg;
1642         dsl_pool_t *dp = dmu_tx_pool(tx);
1643         int error;
1644
1645         ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1646
1647         if (!drc->drc_newfs) {
1648                 dsl_dataset_t *origin_head;
1649
1650                 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1651                 if (error != 0)
1652                         return (error);
1653                 if (drc->drc_force) {
1654                         /*
1655                          * We will destroy any snapshots in tofs (i.e. before
1656                          * origin_head) that are after the origin (which is
1657                          * the snap before drc_ds, because drc_ds can not
1658                          * have any snaps of its own).
1659                          */
1660                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1661                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1662                                 dsl_dataset_t *snap;
1663                                 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1664                                     &snap);
1665                                 if (error != 0)
1666                                         return (error);
1667                                 if (snap->ds_dir != origin_head->ds_dir)
1668                                         error = SET_ERROR(EINVAL);
1669                                 if (error == 0)  {
1670                                         error = dsl_destroy_snapshot_check_impl(
1671                                             snap, B_FALSE);
1672                                 }
1673                                 obj = snap->ds_phys->ds_prev_snap_obj;
1674                                 dsl_dataset_rele(snap, FTAG);
1675                                 if (error != 0)
1676                                         return (error);
1677                         }
1678                 }
1679                 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1680                     origin_head, drc->drc_force, drc->drc_owner, tx);
1681                 if (error != 0) {
1682                         dsl_dataset_rele(origin_head, FTAG);
1683                         return (error);
1684                 }
1685                 error = dsl_dataset_snapshot_check_impl(origin_head,
1686                     drc->drc_tosnap, tx, B_TRUE);
1687                 dsl_dataset_rele(origin_head, FTAG);
1688                 if (error != 0)
1689                         return (error);
1690
1691                 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1692         } else {
1693                 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1694                     drc->drc_tosnap, tx, B_TRUE);
1695         }
1696         return (error);
1697 }
1698
1699 static void
1700 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1701 {
1702         dmu_recv_cookie_t *drc = arg;
1703         dsl_pool_t *dp = dmu_tx_pool(tx);
1704
1705         spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1706             tx, "snap=%s", drc->drc_tosnap);
1707
1708         if (!drc->drc_newfs) {
1709                 dsl_dataset_t *origin_head;
1710
1711                 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1712                     &origin_head));
1713
1714                 if (drc->drc_force) {
1715                         /*
1716                          * Destroy any snapshots of drc_tofs (origin_head)
1717                          * after the origin (the snap before drc_ds).
1718                          */
1719                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1720                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1721                                 dsl_dataset_t *snap;
1722                                 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1723                                     &snap));
1724                                 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1725                                 obj = snap->ds_phys->ds_prev_snap_obj;
1726                                 dsl_destroy_snapshot_sync_impl(snap,
1727                                     B_FALSE, tx);
1728                                 dsl_dataset_rele(snap, FTAG);
1729                         }
1730                 }
1731                 VERIFY3P(drc->drc_ds->ds_prev, ==,
1732                     origin_head->ds_prev);
1733
1734                 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1735                     origin_head, tx);
1736                 dsl_dataset_snapshot_sync_impl(origin_head,
1737                     drc->drc_tosnap, tx);
1738
1739                 /* set snapshot's creation time and guid */
1740                 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1741                 origin_head->ds_prev->ds_phys->ds_creation_time =
1742                     drc->drc_drrb->drr_creation_time;
1743                 origin_head->ds_prev->ds_phys->ds_guid =
1744                     drc->drc_drrb->drr_toguid;
1745                 origin_head->ds_prev->ds_phys->ds_flags &=
1746                     ~DS_FLAG_INCONSISTENT;
1747
1748                 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1749                 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1750
1751                 dsl_dataset_rele(origin_head, FTAG);
1752                 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1753
1754                 if (drc->drc_owner != NULL)
1755                         VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
1756         } else {
1757                 dsl_dataset_t *ds = drc->drc_ds;
1758
1759                 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1760
1761                 /* set snapshot's creation time and guid */
1762                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1763                 ds->ds_prev->ds_phys->ds_creation_time =
1764                     drc->drc_drrb->drr_creation_time;
1765                 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1766                 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1767
1768                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1769                 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1770         }
1771         drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1772         /*
1773          * Release the hold from dmu_recv_begin.  This must be done before
1774          * we return to open context, so that when we free the dataset's dnode,
1775          * we can evict its bonus buffer.
1776          */
1777         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1778         drc->drc_ds = NULL;
1779 }
1780
1781 static int
1782 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1783 {
1784         dsl_pool_t *dp;
1785         dsl_dataset_t *snapds;
1786         guid_map_entry_t *gmep;
1787         int err;
1788
1789         ASSERT(guid_map != NULL);
1790
1791         err = dsl_pool_hold(name, FTAG, &dp);
1792         if (err != 0)
1793                 return (err);
1794         gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
1795         err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
1796         if (err == 0) {
1797                 gmep->guid = snapds->ds_phys->ds_guid;
1798                 gmep->gme_ds = snapds;
1799                 avl_add(guid_map, gmep);
1800                 dsl_dataset_long_hold(snapds, gmep);
1801         } else
1802                 kmem_free(gmep, sizeof (*gmep));
1803
1804         dsl_pool_rele(dp, FTAG);
1805         return (err);
1806 }
1807
1808 static int dmu_recv_end_modified_blocks = 3;
1809
1810 static int
1811 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1812 {
1813         int error;
1814         char name[MAXNAMELEN];
1815
1816 #ifdef _KERNEL
1817         /*
1818          * We will be destroying the ds; make sure its origin is unmounted if
1819          * necessary.
1820          */
1821         dsl_dataset_name(drc->drc_ds, name);
1822         zfs_destroy_unmount_origin(name);
1823 #endif
1824
1825         error = dsl_sync_task(drc->drc_tofs,
1826             dmu_recv_end_check, dmu_recv_end_sync, drc,
1827             dmu_recv_end_modified_blocks);
1828
1829         if (error != 0)
1830                 dmu_recv_cleanup_ds(drc);
1831         return (error);
1832 }
1833
1834 static int
1835 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1836 {
1837         int error;
1838
1839         error = dsl_sync_task(drc->drc_tofs,
1840             dmu_recv_end_check, dmu_recv_end_sync, drc,
1841             dmu_recv_end_modified_blocks);
1842
1843         if (error != 0) {
1844                 dmu_recv_cleanup_ds(drc);
1845         } else if (drc->drc_guid_to_ds_map != NULL) {
1846                 (void) add_ds_to_guidmap(drc->drc_tofs,
1847                     drc->drc_guid_to_ds_map,
1848                     drc->drc_newsnapobj);
1849         }
1850         return (error);
1851 }
1852
1853 int
1854 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
1855 {
1856         drc->drc_owner = owner;
1857
1858         if (drc->drc_newfs)
1859                 return (dmu_recv_new_end(drc));
1860         else
1861                 return (dmu_recv_existing_end(drc));
1862 }
1863
1864 /*
1865  * Return TRUE if this objset is currently being received into.
1866  */
1867 boolean_t
1868 dmu_objset_is_receiving(objset_t *os)
1869 {
1870         return (os->os_dsl_dataset != NULL &&
1871             os->os_dsl_dataset->ds_owner == dmu_recv_tag);
1872 }