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 #include <sys/dsl_bookmark.h>
  53
  54 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
  55 int zfs_send_corrupt_data = B_FALSE;
  56
  57 static char *dmu_recv_tag = "dmu_recv_tag";
  58 static const char *recv_clone_name = "%recv";
  59
  60 static int
  61 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
  62 {
  63         dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
  64         struct uio auio;
  65         struct iovec aiov;
  66         ASSERT0(len % 8);
  67
  68         fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
  69         aiov.iov_base = buf;
  70         aiov.iov_len = len;
  71         auio.uio_iov = &aiov;
  72         auio.uio_iovcnt = 1;
  73         auio.uio_resid = len;
  74         auio.uio_segflg = UIO_SYSSPACE;
  75         auio.uio_rw = UIO_WRITE;
  76         auio.uio_offset = (off_t)-1;
  77         auio.uio_td = dsp->dsa_td;
  78 #ifdef _KERNEL
  79         if (dsp->dsa_fp->f_type == DTYPE_VNODE)
  80                 bwillwrite();
  81         dsp->dsa_err = fo_write(dsp->dsa_fp, &auio, dsp->dsa_td->td_ucred, 0,
  82             dsp->dsa_td);
  83 #else
  84         fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
  85         dsp->dsa_err = EOPNOTSUPP;
  86 #endif
  87         mutex_enter(&ds->ds_sendstream_lock);
  88         *dsp->dsa_off += len;
  89         mutex_exit(&ds->ds_sendstream_lock);
  90
  91         return (dsp->dsa_err);
  92 }
  93
  94 static int
  95 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
  96     uint64_t length)
  97 {
  98         struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
  99
 100         /*
 101          * When we receive a free record, dbuf_free_range() assumes
 102          * that the receiving system doesn't have any dbufs in the range
 103          * being freed.  This is always true because there is a one-record
 104          * constraint: we only send one WRITE record for any given
 105          * object+offset.  We know that the one-record constraint is
 106          * true because we always send data in increasing order by
 107          * object,offset.
 108          *
 109          * If the increasing-order constraint ever changes, we should find
 110          * another way to assert that the one-record constraint is still
 111          * satisfied.
 112          */
 113         ASSERT(object > dsp->dsa_last_data_object ||
 114             (object == dsp->dsa_last_data_object &&
 115             offset > dsp->dsa_last_data_offset));
 116
 117         /*
 118          * If we are doing a non-incremental send, then there can't
 119          * be any data in the dataset we're receiving into.  Therefore
 120          * a free record would simply be a no-op.  Save space by not
 121          * sending it to begin with.
 122          */
 123         if (!dsp->dsa_incremental)
 124                 return (0);
 125
 126         if (length != -1ULL && offset + length < offset)
 127                 length = -1ULL;
 128
 129         /*
 130          * If there is a pending op, but it's not PENDING_FREE, push it out,
 131          * since free block aggregation can only be done for blocks of the
 132          * same type (i.e., DRR_FREE records can only be aggregated with
 133          * other DRR_FREE records.  DRR_FREEOBJECTS records can only be
 134          * aggregated with other DRR_FREEOBJECTS records.
 135          */
 136         if (dsp->dsa_pending_op != PENDING_NONE &&
 137             dsp->dsa_pending_op != PENDING_FREE) {
 138                 if (dump_bytes(dsp, dsp->dsa_drr,
 139                     sizeof (dmu_replay_record_t)) != 0)
 140                         return (SET_ERROR(EINTR));
 141                 dsp->dsa_pending_op = PENDING_NONE;
 142         }
 143
 144         if (dsp->dsa_pending_op == PENDING_FREE) {
 145                 /*
 146                  * There should never be a PENDING_FREE if length is -1
 147                  * (because dump_dnode is the only place where this
 148                  * function is called with a -1, and only after flushing
 149                  * any pending record).
 150                  */
 151                 ASSERT(length != -1ULL);
 152                 /*
 153                  * Check to see whether this free block can be aggregated
 154                  * with pending one.
 155                  */
 156                 if (drrf->drr_object == object && drrf->drr_offset +
 157                     drrf->drr_length == offset) {
 158                         drrf->drr_length += length;
 159                         return (0);
 160                 } else {
 161                         /* not a continuation.  Push out pending record */
 162                         if (dump_bytes(dsp, dsp->dsa_drr,
 163                             sizeof (dmu_replay_record_t)) != 0)
 164                                 return (SET_ERROR(EINTR));
 165                         dsp->dsa_pending_op = PENDING_NONE;
 166                 }
 167         }
 168         /* create a FREE record and make it pending */
 169         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 170         dsp->dsa_drr->drr_type = DRR_FREE;
 171         drrf->drr_object = object;
 172         drrf->drr_offset = offset;
 173         drrf->drr_length = length;
 174         drrf->drr_toguid = dsp->dsa_toguid;
 175         if (length == -1ULL) {
 176                 if (dump_bytes(dsp, dsp->dsa_drr,
 177                     sizeof (dmu_replay_record_t)) != 0)
 178                         return (SET_ERROR(EINTR));
 179         } else {
 180                 dsp->dsa_pending_op = PENDING_FREE;
 181         }
 182
 183         return (0);
 184 }
 185
 186 static int
 187 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
 188     uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
 189 {
 190         struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
 191
 192         /*
 193          * We send data in increasing object, offset order.
 194          * See comment in dump_free() for details.
 195          */
 196         ASSERT(object > dsp->dsa_last_data_object ||
 197             (object == dsp->dsa_last_data_object &&
 198             offset > dsp->dsa_last_data_offset));
 199         dsp->dsa_last_data_object = object;
 200         dsp->dsa_last_data_offset = offset + blksz - 1;
 201
 202         /*
 203          * If there is any kind of pending aggregation (currently either
 204          * a grouping of free objects or free blocks), push it out to
 205          * the stream, since aggregation can't be done across operations
 206          * of different types.
 207          */
 208         if (dsp->dsa_pending_op != PENDING_NONE) {
 209                 if (dump_bytes(dsp, dsp->dsa_drr,
 210                     sizeof (dmu_replay_record_t)) != 0)
 211                         return (SET_ERROR(EINTR));
 212                 dsp->dsa_pending_op = PENDING_NONE;
 213         }
 214         /* write a DATA record */
 215         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 216         dsp->dsa_drr->drr_type = DRR_WRITE;
 217         drrw->drr_object = object;
 218         drrw->drr_type = type;
 219         drrw->drr_offset = offset;
 220         drrw->drr_length = blksz;
 221         drrw->drr_toguid = dsp->dsa_toguid;
 222         drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
 223         if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
 224                 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
 225         DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
 226         DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
 227         DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
 228         drrw->drr_key.ddk_cksum = bp->blk_cksum;
 229
 230         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 231                 return (SET_ERROR(EINTR));
 232         if (dump_bytes(dsp, data, blksz) != 0)
 233                 return (SET_ERROR(EINTR));
 234         return (0);
 235 }
 236
 237 static int
 238 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
 239 {
 240         struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
 241
 242         if (dsp->dsa_pending_op != PENDING_NONE) {
 243                 if (dump_bytes(dsp, dsp->dsa_drr,
 244                     sizeof (dmu_replay_record_t)) != 0)
 245                         return (SET_ERROR(EINTR));
 246                 dsp->dsa_pending_op = PENDING_NONE;
 247         }
 248
 249         /* write a SPILL record */
 250         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 251         dsp->dsa_drr->drr_type = DRR_SPILL;
 252         drrs->drr_object = object;
 253         drrs->drr_length = blksz;
 254         drrs->drr_toguid = dsp->dsa_toguid;
 255
 256         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
 257                 return (SET_ERROR(EINTR));
 258         if (dump_bytes(dsp, data, blksz))
 259                 return (SET_ERROR(EINTR));
 260         return (0);
 261 }
 262
 263 static int
 264 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
 265 {
 266         struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
 267
 268         /* See comment in dump_free(). */
 269         if (!dsp->dsa_incremental)
 270                 return (0);
 271
 272         /*
 273          * If there is a pending op, but it's not PENDING_FREEOBJECTS,
 274          * push it out, since free block aggregation can only be done for
 275          * blocks of the same type (i.e., DRR_FREE records can only be
 276          * aggregated with other DRR_FREE records.  DRR_FREEOBJECTS records
 277          * can only be aggregated with other DRR_FREEOBJECTS records.
 278          */
 279         if (dsp->dsa_pending_op != PENDING_NONE &&
 280             dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
 281                 if (dump_bytes(dsp, dsp->dsa_drr,
 282                     sizeof (dmu_replay_record_t)) != 0)
 283                         return (SET_ERROR(EINTR));
 284                 dsp->dsa_pending_op = PENDING_NONE;
 285         }
 286         if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
 287                 /*
 288                  * See whether this free object array can be aggregated
 289                  * with pending one
 290                  */
 291                 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
 292                         drrfo->drr_numobjs += numobjs;
 293                         return (0);
 294                 } else {
 295                         /* can't be aggregated.  Push out pending record */
 296                         if (dump_bytes(dsp, dsp->dsa_drr,
 297                             sizeof (dmu_replay_record_t)) != 0)
 298                                 return (SET_ERROR(EINTR));
 299                         dsp->dsa_pending_op = PENDING_NONE;
 300                 }
 301         }
 302
 303         /* write a FREEOBJECTS record */
 304         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 305         dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
 306         drrfo->drr_firstobj = firstobj;
 307         drrfo->drr_numobjs = numobjs;
 308         drrfo->drr_toguid = dsp->dsa_toguid;
 309
 310         dsp->dsa_pending_op = PENDING_FREEOBJECTS;
 311
 312         return (0);
 313 }
 314
 315 static int
 316 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
 317 {
 318         struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
 319
 320         if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
 321                 return (dump_freeobjects(dsp, object, 1));
 322
 323         if (dsp->dsa_pending_op != PENDING_NONE) {
 324                 if (dump_bytes(dsp, dsp->dsa_drr,
 325                     sizeof (dmu_replay_record_t)) != 0)
 326                         return (SET_ERROR(EINTR));
 327                 dsp->dsa_pending_op = PENDING_NONE;
 328         }
 329
 330         /* write an OBJECT record */
 331         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 332         dsp->dsa_drr->drr_type = DRR_OBJECT;
 333         drro->drr_object = object;
 334         drro->drr_type = dnp->dn_type;
 335         drro->drr_bonustype = dnp->dn_bonustype;
 336         drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
 337         drro->drr_bonuslen = dnp->dn_bonuslen;
 338         drro->drr_checksumtype = dnp->dn_checksum;
 339         drro->drr_compress = dnp->dn_compress;
 340         drro->drr_toguid = dsp->dsa_toguid;
 341
 342         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 343                 return (SET_ERROR(EINTR));
 344
 345         if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
 346                 return (SET_ERROR(EINTR));
 347
 348         /* Free anything past the end of the file. */
 349         if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
 350             (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
 351                 return (SET_ERROR(EINTR));
 352         if (dsp->dsa_err != 0)
 353                 return (SET_ERROR(EINTR));
 354         return (0);
 355 }
 356
 357 #define BP_SPAN(dnp, level) \
 358         (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
 359         (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
 360
 361 /* ARGSUSED */
 362 static int
 363 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 364     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
 365 {
 366         dmu_sendarg_t *dsp = arg;
 367         dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
 368         int err = 0;
 369
 370         if (issig(JUSTLOOKING) && issig(FORREAL))
 371                 return (SET_ERROR(EINTR));
 372
 373         if (zb->zb_object != DMU_META_DNODE_OBJECT &&
 374             DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
 375                 return (0);
 376         } else if (zb->zb_level == ZB_ZIL_LEVEL) {
 377                 /*
 378                  * If we are sending a non-snapshot (which is allowed on
 379                  * read-only pools), it may have a ZIL, which must be ignored.
 380                  */
 381                 return (0);
 382         } else if (BP_IS_HOLE(bp) &&
 383             zb->zb_object == DMU_META_DNODE_OBJECT) {
 384                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 385                 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
 386                 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
 387         } else if (BP_IS_HOLE(bp)) {
 388                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 389                 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
 390         } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
 391                 return (0);
 392         } else if (type == DMU_OT_DNODE) {
 393                 dnode_phys_t *blk;
 394                 int i;
 395                 int blksz = BP_GET_LSIZE(bp);
 396                 uint32_t aflags = ARC_WAIT;
 397                 arc_buf_t *abuf;
 398
 399                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 400                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 401                     &aflags, zb) != 0)
 402                         return (SET_ERROR(EIO));
 403
 404                 blk = abuf->b_data;
 405                 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
 406                         uint64_t dnobj = (zb->zb_blkid <<
 407                             (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
 408                         err = dump_dnode(dsp, dnobj, blk+i);
 409                         if (err != 0)
 410                                 break;
 411                 }
 412                 (void) arc_buf_remove_ref(abuf, &abuf);
 413         } else if (type == DMU_OT_SA) {
 414                 uint32_t aflags = ARC_WAIT;
 415                 arc_buf_t *abuf;
 416                 int blksz = BP_GET_LSIZE(bp);
 417
 418                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 419                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 420                     &aflags, zb) != 0)
 421                         return (SET_ERROR(EIO));
 422
 423                 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
 424                 (void) arc_buf_remove_ref(abuf, &abuf);
 425         } else { /* it's a level-0 block of a regular object */
 426                 uint32_t aflags = ARC_WAIT;
 427                 arc_buf_t *abuf;
 428                 int blksz = BP_GET_LSIZE(bp);
 429
 430                 ASSERT0(zb->zb_level);
 431                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 432                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 433                     &aflags, zb) != 0) {
 434                         if (zfs_send_corrupt_data) {
 435                                 /* Send a block filled with 0x"zfs badd bloc" */
 436                                 abuf = arc_buf_alloc(spa, blksz, &abuf,
 437                                     ARC_BUFC_DATA);
 438                                 uint64_t *ptr;
 439                                 for (ptr = abuf->b_data;
 440                                     (char *)ptr < (char *)abuf->b_data + blksz;
 441                                     ptr++)
 442                                         *ptr = 0x2f5baddb10c;
 443                         } else {
 444                                 return (SET_ERROR(EIO));
 445                         }
 446                 }
 447
 448                 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
 449                     blksz, bp, abuf->b_data);
 450                 (void) arc_buf_remove_ref(abuf, &abuf);
 451         }
 452
 453         ASSERT(err == 0 || err == EINTR);
 454         return (err);
 455 }
 456
 457 /*
 458  * Releases dp using the specified tag.
 459  */
 460 static int
 461 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 462 #ifdef illumos
 463     zfs_bookmark_phys_t *fromzb, boolean_t is_clone, int outfd,
 464     vnode_t *vp, offset_t *off)
 465 #else
 466     zfs_bookmark_phys_t *fromzb, boolean_t is_clone, int outfd,
 467     struct file *fp, offset_t *off)
 468 #endif
 469 {
 470         objset_t *os;
 471         dmu_replay_record_t *drr;
 472         dmu_sendarg_t *dsp;
 473         int err;
 474         uint64_t fromtxg = 0;
 475
 476         err = dmu_objset_from_ds(ds, &os);
 477         if (err != 0) {
 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                         dsl_pool_rele(dp, tag);
 494                         return (SET_ERROR(EINVAL));
 495                 }
 496                 if (version >= ZPL_VERSION_SA) {
 497                         DMU_SET_FEATUREFLAGS(
 498                             drr->drr_u.drr_begin.drr_versioninfo,
 499                             DMU_BACKUP_FEATURE_SA_SPILL);
 500                 }
 501         }
 502 #endif
 503
 504         drr->drr_u.drr_begin.drr_creation_time =
 505             ds->ds_phys->ds_creation_time;
 506         drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
 507         if (is_clone)
 508                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
 509         drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
 510         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 511                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
 512
 513         if (fromzb != NULL) {
 514                 drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid;
 515                 fromtxg = fromzb->zbm_creation_txg;
 516         }
 517         dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
 518         if (!dsl_dataset_is_snapshot(ds)) {
 519                 (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
 520                     sizeof (drr->drr_u.drr_begin.drr_toname));
 521         }
 522
 523         dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
 524
 525         dsp->dsa_drr = drr;
 526         dsp->dsa_outfd = outfd;
 527         dsp->dsa_proc = curproc;
 528         dsp->dsa_td = curthread;
 529         dsp->dsa_fp = fp;
 530         dsp->dsa_os = os;
 531         dsp->dsa_off = off;
 532         dsp->dsa_toguid = ds->ds_phys->ds_guid;
 533         ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
 534         dsp->dsa_pending_op = PENDING_NONE;
 535         dsp->dsa_incremental = (fromzb != NULL);
 536
 537         mutex_enter(&ds->ds_sendstream_lock);
 538         list_insert_head(&ds->ds_sendstreams, dsp);
 539         mutex_exit(&ds->ds_sendstream_lock);
 540
 541         dsl_dataset_long_hold(ds, FTAG);
 542         dsl_pool_rele(dp, tag);
 543
 544         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 545                 err = dsp->dsa_err;
 546                 goto out;
 547         }
 548
 549         err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
 550             backup_cb, dsp);
 551
 552         if (dsp->dsa_pending_op != PENDING_NONE)
 553                 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
 554                         err = SET_ERROR(EINTR);
 555
 556         if (err != 0) {
 557                 if (err == EINTR && dsp->dsa_err != 0)
 558                         err = dsp->dsa_err;
 559                 goto out;
 560         }
 561
 562         bzero(drr, sizeof (dmu_replay_record_t));
 563         drr->drr_type = DRR_END;
 564         drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
 565         drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
 566
 567         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 568                 err = dsp->dsa_err;
 569                 goto out;
 570         }
 571
 572 out:
 573         mutex_enter(&ds->ds_sendstream_lock);
 574         list_remove(&ds->ds_sendstreams, dsp);
 575         mutex_exit(&ds->ds_sendstream_lock);
 576
 577         kmem_free(drr, sizeof (dmu_replay_record_t));
 578         kmem_free(dsp, sizeof (dmu_sendarg_t));
 579
 580         dsl_dataset_long_rele(ds, FTAG);
 581
 582         return (err);
 583 }
 584
 585 int
 586 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
 587 #ifdef illumos
 588     int outfd, vnode_t *vp, offset_t *off)
 589 #else
 590     int outfd, struct file *fp, offset_t *off)
 591 #endif
 592 {
 593         dsl_pool_t *dp;
 594         dsl_dataset_t *ds;
 595         dsl_dataset_t *fromds = NULL;
 596         int err;
 597
 598         err = dsl_pool_hold(pool, FTAG, &dp);
 599         if (err != 0)
 600                 return (err);
 601
 602         err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
 603         if (err != 0) {
 604                 dsl_pool_rele(dp, FTAG);
 605                 return (err);
 606         }
 607
 608         if (fromsnap != 0) {
 609                 zfs_bookmark_phys_t zb;
 610                 boolean_t is_clone;
 611
 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                 if (!dsl_dataset_is_before(ds, fromds, 0))
 619                         err = SET_ERROR(EXDEV);
 620                 zb.zbm_creation_time = fromds->ds_phys->ds_creation_time;
 621                 zb.zbm_creation_txg = fromds->ds_phys->ds_creation_txg;
 622                 zb.zbm_guid = fromds->ds_phys->ds_guid;
 623                 is_clone = (fromds->ds_dir != ds->ds_dir);
 624                 dsl_dataset_rele(fromds, FTAG);
 625                 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
 626                     outfd, fp, off);
 627         } else {
 628                 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
 629                     outfd, fp, off);
 630         }
 631         dsl_dataset_rele(ds, FTAG);
 632         return (err);
 633 }
 634
 635 int
 636 dmu_send(const char *tosnap, const char *fromsnap,
 637 #ifdef illumos
 638     int outfd, vnode_t *vp, offset_t *off)
 639 #else
 640     int outfd, struct file *fp, offset_t *off)
 641 #endif
 642 {
 643         dsl_pool_t *dp;
 644         dsl_dataset_t *ds;
 645         int err;
 646         boolean_t owned = B_FALSE;
 647
 648         if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
 649                 return (SET_ERROR(EINVAL));
 650
 651         err = dsl_pool_hold(tosnap, FTAG, &dp);
 652         if (err != 0)
 653                 return (err);
 654
 655         if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) {
 656                 /*
 657                  * We are sending a filesystem or volume.  Ensure
 658                  * that it doesn't change by owning the dataset.
 659                  */
 660                 err = dsl_dataset_own(dp, tosnap, FTAG, &ds);
 661                 owned = B_TRUE;
 662         } else {
 663                 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
 664         }
 665         if (err != 0) {
 666                 dsl_pool_rele(dp, FTAG);
 667                 return (err);
 668         }
 669
 670         if (fromsnap != NULL) {
 671                 zfs_bookmark_phys_t zb;
 672                 boolean_t is_clone = B_FALSE;
 673                 int fsnamelen = strchr(tosnap, '@') - tosnap;
 674
 675                 /*
 676                  * If the fromsnap is in a different filesystem, then
 677                  * mark the send stream as a clone.
 678                  */
 679                 if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
 680                     (fromsnap[fsnamelen] != '@' &&
 681                     fromsnap[fsnamelen] != '#')) {
 682                         is_clone = B_TRUE;
 683                 }
 684
 685                 if (strchr(fromsnap, '@')) {
 686                         dsl_dataset_t *fromds;
 687                         err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
 688                         if (err == 0) {
 689                                 if (!dsl_dataset_is_before(ds, fromds, 0))
 690                                         err = SET_ERROR(EXDEV);
 691                                 zb.zbm_creation_time =
 692                                     fromds->ds_phys->ds_creation_time;
 693                                 zb.zbm_creation_txg =
 694                                     fromds->ds_phys->ds_creation_txg;
 695                                 zb.zbm_guid = fromds->ds_phys->ds_guid;
 696                                 is_clone = (ds->ds_dir != fromds->ds_dir);
 697                                 dsl_dataset_rele(fromds, FTAG);
 698                         }
 699                 } else {
 700                         err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb);
 701                 }
 702                 if (err != 0) {
 703                         dsl_dataset_rele(ds, FTAG);
 704                         dsl_pool_rele(dp, FTAG);
 705                         return (err);
 706                 }
 707                 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
 708                     outfd, fp, off);
 709         } else {
 710                 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
 711                     outfd, fp, off);
 712         }
 713         if (owned)
 714                 dsl_dataset_disown(ds, FTAG);
 715         else
 716                 dsl_dataset_rele(ds, FTAG);
 717         return (err);
 718 }
 719
 720 int
 721 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
 722 {
 723         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 724         int err;
 725         uint64_t size;
 726
 727         ASSERT(dsl_pool_config_held(dp));
 728
 729         /* tosnap must be a snapshot */
 730         if (!dsl_dataset_is_snapshot(ds))
 731                 return (SET_ERROR(EINVAL));
 732
 733         /*
 734          * fromsnap must be an earlier snapshot from the same fs as tosnap,
 735          * or the origin's fs.
 736          */
 737         if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
 738                 return (SET_ERROR(EXDEV));
 739
 740         /* Get uncompressed size estimate of changed data. */
 741         if (fromds == NULL) {
 742                 size = ds->ds_phys->ds_uncompressed_bytes;
 743         } else {
 744                 uint64_t used, comp;
 745                 err = dsl_dataset_space_written(fromds, ds,
 746                     &used, &comp, &size);
 747                 if (err != 0)
 748                         return (err);
 749         }
 750
 751         /*
 752          * Assume that space (both on-disk and in-stream) is dominated by
 753          * data.  We will adjust for indirect blocks and the copies property,
 754          * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
 755          */
 756
 757         /*
 758          * Subtract out approximate space used by indirect blocks.
 759          * Assume most space is used by data blocks (non-indirect, non-dnode).
 760          * Assume all blocks are recordsize.  Assume ditto blocks and
 761          * internal fragmentation counter out compression.
 762          *
 763          * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
 764          * block, which we observe in practice.
 765          */
 766         uint64_t recordsize;
 767         err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
 768         if (err != 0)
 769                 return (err);
 770         size -= size / recordsize * sizeof (blkptr_t);
 771
 772         /* Add in the space for the record associated with each block. */
 773         size += size / recordsize * sizeof (dmu_replay_record_t);
 774
 775         *sizep = size;
 776
 777         return (0);
 778 }
 779
 780 typedef struct dmu_recv_begin_arg {
 781         const char *drba_origin;
 782         dmu_recv_cookie_t *drba_cookie;
 783         cred_t *drba_cred;
 784         uint64_t drba_snapobj;
 785 } dmu_recv_begin_arg_t;
 786
 787 static int
 788 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
 789     uint64_t fromguid)
 790 {
 791         uint64_t val;
 792         int error;
 793         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 794
 795         /* temporary clone name must not exist */
 796         error = zap_lookup(dp->dp_meta_objset,
 797             ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
 798             8, 1, &val);
 799         if (error != ENOENT)
 800                 return (error == 0 ? EBUSY : error);
 801
 802         /* new snapshot name must not exist */
 803         error = zap_lookup(dp->dp_meta_objset,
 804             ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
 805             8, 1, &val);
 806         if (error != ENOENT)
 807                 return (error == 0 ? EEXIST : error);
 808
 809         if (fromguid != 0) {
 810                 dsl_dataset_t *snap;
 811                 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
 812
 813                 /* Find snapshot in this dir that matches fromguid. */
 814                 while (obj != 0) {
 815                         error = dsl_dataset_hold_obj(dp, obj, FTAG,
 816                             &snap);
 817                         if (error != 0)
 818                                 return (SET_ERROR(ENODEV));
 819                         if (snap->ds_dir != ds->ds_dir) {
 820                                 dsl_dataset_rele(snap, FTAG);
 821                                 return (SET_ERROR(ENODEV));
 822                         }
 823                         if (snap->ds_phys->ds_guid == fromguid)
 824                                 break;
 825                         obj = snap->ds_phys->ds_prev_snap_obj;
 826                         dsl_dataset_rele(snap, FTAG);
 827                 }
 828                 if (obj == 0)
 829                         return (SET_ERROR(ENODEV));
 830
 831                 if (drba->drba_cookie->drc_force) {
 832                         drba->drba_snapobj = obj;
 833                 } else {
 834                         /*
 835                          * If we are not forcing, there must be no
 836                          * changes since fromsnap.
 837                          */
 838                         if (dsl_dataset_modified_since_snap(ds, snap)) {
 839                                 dsl_dataset_rele(snap, FTAG);
 840                                 return (SET_ERROR(ETXTBSY));
 841                         }
 842                         drba->drba_snapobj = ds->ds_prev->ds_object;
 843                 }
 844
 845                 dsl_dataset_rele(snap, FTAG);
 846         } else {
 847                 /* if full, most recent snapshot must be $ORIGIN */
 848                 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
 849                         return (SET_ERROR(ENODEV));
 850                 drba->drba_snapobj = ds->ds_phys->ds_prev_snap_obj;
 851         }
 852
 853         return (0);
 854
 855 }
 856
 857 static int
 858 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 859 {
 860         dmu_recv_begin_arg_t *drba = arg;
 861         dsl_pool_t *dp = dmu_tx_pool(tx);
 862         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 863         uint64_t fromguid = drrb->drr_fromguid;
 864         int flags = drrb->drr_flags;
 865         int error;
 866         dsl_dataset_t *ds;
 867         const char *tofs = drba->drba_cookie->drc_tofs;
 868
 869         /* already checked */
 870         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 871
 872         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 873             DMU_COMPOUNDSTREAM ||
 874             drrb->drr_type >= DMU_OST_NUMTYPES ||
 875             ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
 876                 return (SET_ERROR(EINVAL));
 877
 878         /* Verify pool version supports SA if SA_SPILL feature set */
 879         if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
 880             DMU_BACKUP_FEATURE_SA_SPILL) &&
 881             spa_version(dp->dp_spa) < SPA_VERSION_SA) {
 882                 return (SET_ERROR(ENOTSUP));
 883         }
 884
 885         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 886         if (error == 0) {
 887                 /* target fs already exists; recv into temp clone */
 888
 889                 /* Can't recv a clone into an existing fs */
 890                 if (flags & DRR_FLAG_CLONE) {
 891                         dsl_dataset_rele(ds, FTAG);
 892                         return (SET_ERROR(EINVAL));
 893                 }
 894
 895                 error = recv_begin_check_existing_impl(drba, ds, fromguid);
 896                 dsl_dataset_rele(ds, FTAG);
 897         } else if (error == ENOENT) {
 898                 /* target fs does not exist; must be a full backup or clone */
 899                 char buf[MAXNAMELEN];
 900
 901                 /*
 902                  * If it's a non-clone incremental, we are missing the
 903                  * target fs, so fail the recv.
 904                  */
 905                 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
 906                         return (SET_ERROR(ENOENT));
 907
 908                 /* Open the parent of tofs */
 909                 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
 910                 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
 911                 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
 912                 if (error != 0)
 913                         return (error);
 914
 915                 if (drba->drba_origin != NULL) {
 916                         dsl_dataset_t *origin;
 917                         error = dsl_dataset_hold(dp, drba->drba_origin,
 918                             FTAG, &origin);
 919                         if (error != 0) {
 920                                 dsl_dataset_rele(ds, FTAG);
 921                                 return (error);
 922                         }
 923                         if (!dsl_dataset_is_snapshot(origin)) {
 924                                 dsl_dataset_rele(origin, FTAG);
 925                                 dsl_dataset_rele(ds, FTAG);
 926                                 return (SET_ERROR(EINVAL));
 927                         }
 928                         if (origin->ds_phys->ds_guid != fromguid) {
 929                                 dsl_dataset_rele(origin, FTAG);
 930                                 dsl_dataset_rele(ds, FTAG);
 931                                 return (SET_ERROR(ENODEV));
 932                         }
 933                         dsl_dataset_rele(origin, FTAG);
 934                 }
 935                 dsl_dataset_rele(ds, FTAG);
 936                 error = 0;
 937         }
 938         return (error);
 939 }
 940
 941 static void
 942 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
 943 {
 944         dmu_recv_begin_arg_t *drba = arg;
 945         dsl_pool_t *dp = dmu_tx_pool(tx);
 946         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 947         const char *tofs = drba->drba_cookie->drc_tofs;
 948         dsl_dataset_t *ds, *newds;
 949         uint64_t dsobj;
 950         int error;
 951         uint64_t crflags;
 952
 953         crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
 954             DS_FLAG_CI_DATASET : 0;
 955
 956         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 957         if (error == 0) {
 958                 /* create temporary clone */
 959                 dsl_dataset_t *snap = NULL;
 960                 if (drba->drba_snapobj != 0) {
 961                         VERIFY0(dsl_dataset_hold_obj(dp,
 962                             drba->drba_snapobj, FTAG, &snap));
 963                 }
 964                 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
 965                     snap, crflags, drba->drba_cred, tx);
 966                 dsl_dataset_rele(snap, FTAG);
 967                 dsl_dataset_rele(ds, FTAG);
 968         } else {
 969                 dsl_dir_t *dd;
 970                 const char *tail;
 971                 dsl_dataset_t *origin = NULL;
 972
 973                 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
 974
 975                 if (drba->drba_origin != NULL) {
 976                         VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
 977                             FTAG, &origin));
 978                 }
 979
 980                 /* Create new dataset. */
 981                 dsobj = dsl_dataset_create_sync(dd,
 982                     strrchr(tofs, '/') + 1,
 983                     origin, crflags, drba->drba_cred, tx);
 984                 if (origin != NULL)
 985                         dsl_dataset_rele(origin, FTAG);
 986                 dsl_dir_rele(dd, FTAG);
 987                 drba->drba_cookie->drc_newfs = B_TRUE;
 988         }
 989         VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
 990
 991         dmu_buf_will_dirty(newds->ds_dbuf, tx);
 992         newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
 993
 994         /*
 995          * If we actually created a non-clone, we need to create the
 996          * objset in our new dataset.
 997          */
 998         if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
 999                 (void) dmu_objset_create_impl(dp->dp_spa,
1000                     newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1001         }
1002
1003         drba->drba_cookie->drc_ds = newds;
1004
1005         spa_history_log_internal_ds(newds, "receive", tx, "");
1006 }
1007
1008 /*
1009  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1010  * succeeds; otherwise we will leak the holds on the datasets.
1011  */
1012 int
1013 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1014     boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1015 {
1016         dmu_recv_begin_arg_t drba = { 0 };
1017         dmu_replay_record_t *drr;
1018
1019         bzero(drc, sizeof (dmu_recv_cookie_t));
1020         drc->drc_drrb = drrb;
1021         drc->drc_tosnap = tosnap;
1022         drc->drc_tofs = tofs;
1023         drc->drc_force = force;
1024
1025         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1026                 drc->drc_byteswap = B_TRUE;
1027         else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1028                 return (SET_ERROR(EINVAL));
1029
1030         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1031         drr->drr_type = DRR_BEGIN;
1032         drr->drr_u.drr_begin = *drc->drc_drrb;
1033         if (drc->drc_byteswap) {
1034                 fletcher_4_incremental_byteswap(drr,
1035                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1036         } else {
1037                 fletcher_4_incremental_native(drr,
1038                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1039         }
1040         kmem_free(drr, sizeof (dmu_replay_record_t));
1041
1042         if (drc->drc_byteswap) {
1043                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1044                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1045                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1046                 drrb->drr_type = BSWAP_32(drrb->drr_type);
1047                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1048                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1049         }
1050
1051         drba.drba_origin = origin;
1052         drba.drba_cookie = drc;
1053         drba.drba_cred = CRED();
1054
1055         return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1056             &drba, 5));
1057 }
1058
1059 struct restorearg {
1060         int err;
1061         boolean_t byteswap;
1062         kthread_t *td;
1063         struct file *fp;
1064         char *buf;
1065         uint64_t voff;
1066         int bufsize; /* amount of memory allocated for buf */
1067         zio_cksum_t cksum;
1068         avl_tree_t *guid_to_ds_map;
1069 };
1070
1071 typedef struct guid_map_entry {
1072         uint64_t        guid;
1073         dsl_dataset_t   *gme_ds;
1074         avl_node_t      avlnode;
1075 } guid_map_entry_t;
1076
1077 static int
1078 guid_compare(const void *arg1, const void *arg2)
1079 {
1080         const guid_map_entry_t *gmep1 = arg1;
1081         const guid_map_entry_t *gmep2 = arg2;
1082
1083         if (gmep1->guid < gmep2->guid)
1084                 return (-1);
1085         else if (gmep1->guid > gmep2->guid)
1086                 return (1);
1087         return (0);
1088 }
1089
1090 static void
1091 free_guid_map_onexit(void *arg)
1092 {
1093         avl_tree_t *ca = arg;
1094         void *cookie = NULL;
1095         guid_map_entry_t *gmep;
1096
1097         while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1098                 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1099                 dsl_dataset_rele(gmep->gme_ds, gmep);
1100                 kmem_free(gmep, sizeof (guid_map_entry_t));
1101         }
1102         avl_destroy(ca);
1103         kmem_free(ca, sizeof (avl_tree_t));
1104 }
1105
1106 static int
1107 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
1108 {
1109         struct uio auio;
1110         struct iovec aiov;
1111         int error;
1112
1113         aiov.iov_base = buf;
1114         aiov.iov_len = len;
1115         auio.uio_iov = &aiov;
1116         auio.uio_iovcnt = 1;
1117         auio.uio_resid = len;
1118         auio.uio_segflg = UIO_SYSSPACE;
1119         auio.uio_rw = UIO_READ;
1120         auio.uio_offset = off;
1121         auio.uio_td = ra->td;
1122 #ifdef _KERNEL
1123         error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
1124 #else
1125         fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
1126         error = EOPNOTSUPP;
1127 #endif
1128         *resid = auio.uio_resid;
1129         return (error);
1130 }
1131
1132 static void *
1133 restore_read(struct restorearg *ra, int len)
1134 {
1135         void *rv;
1136         int done = 0;
1137
1138         /* some things will require 8-byte alignment, so everything must */
1139         ASSERT0(len % 8);
1140
1141         while (done < len) {
1142                 ssize_t resid;
1143
1144                 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
1145                     len - done, ra->voff, &resid);
1146
1147                 if (resid == len - done)
1148                         ra->err = SET_ERROR(EINVAL);
1149                 ra->voff += len - done - resid;
1150                 done = len - resid;
1151                 if (ra->err != 0)
1152                         return (NULL);
1153         }
1154
1155         ASSERT3U(done, ==, len);
1156         rv = ra->buf;
1157         if (ra->byteswap)
1158                 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1159         else
1160                 fletcher_4_incremental_native(rv, len, &ra->cksum);
1161         return (rv);
1162 }
1163
1164 static void
1165 backup_byteswap(dmu_replay_record_t *drr)
1166 {
1167 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1168 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1169         drr->drr_type = BSWAP_32(drr->drr_type);
1170         drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1171         switch (drr->drr_type) {
1172         case DRR_BEGIN:
1173                 DO64(drr_begin.drr_magic);
1174                 DO64(drr_begin.drr_versioninfo);
1175                 DO64(drr_begin.drr_creation_time);
1176                 DO32(drr_begin.drr_type);
1177                 DO32(drr_begin.drr_flags);
1178                 DO64(drr_begin.drr_toguid);
1179                 DO64(drr_begin.drr_fromguid);
1180                 break;
1181         case DRR_OBJECT:
1182                 DO64(drr_object.drr_object);
1183                 /* DO64(drr_object.drr_allocation_txg); */
1184                 DO32(drr_object.drr_type);
1185                 DO32(drr_object.drr_bonustype);
1186                 DO32(drr_object.drr_blksz);
1187                 DO32(drr_object.drr_bonuslen);
1188                 DO64(drr_object.drr_toguid);
1189                 break;
1190         case DRR_FREEOBJECTS:
1191                 DO64(drr_freeobjects.drr_firstobj);
1192                 DO64(drr_freeobjects.drr_numobjs);
1193                 DO64(drr_freeobjects.drr_toguid);
1194                 break;
1195         case DRR_WRITE:
1196                 DO64(drr_write.drr_object);
1197                 DO32(drr_write.drr_type);
1198                 DO64(drr_write.drr_offset);
1199                 DO64(drr_write.drr_length);
1200                 DO64(drr_write.drr_toguid);
1201                 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1202                 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1203                 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1204                 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1205                 DO64(drr_write.drr_key.ddk_prop);
1206                 break;
1207         case DRR_WRITE_BYREF:
1208                 DO64(drr_write_byref.drr_object);
1209                 DO64(drr_write_byref.drr_offset);
1210                 DO64(drr_write_byref.drr_length);
1211                 DO64(drr_write_byref.drr_toguid);
1212                 DO64(drr_write_byref.drr_refguid);
1213                 DO64(drr_write_byref.drr_refobject);
1214                 DO64(drr_write_byref.drr_refoffset);
1215                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1216                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1217                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1218                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1219                 DO64(drr_write_byref.drr_key.ddk_prop);
1220                 break;
1221         case DRR_FREE:
1222                 DO64(drr_free.drr_object);
1223                 DO64(drr_free.drr_offset);
1224                 DO64(drr_free.drr_length);
1225                 DO64(drr_free.drr_toguid);
1226                 break;
1227         case DRR_SPILL:
1228                 DO64(drr_spill.drr_object);
1229                 DO64(drr_spill.drr_length);
1230                 DO64(drr_spill.drr_toguid);
1231                 break;
1232         case DRR_END:
1233                 DO64(drr_end.drr_checksum.zc_word[0]);
1234                 DO64(drr_end.drr_checksum.zc_word[1]);
1235                 DO64(drr_end.drr_checksum.zc_word[2]);
1236                 DO64(drr_end.drr_checksum.zc_word[3]);
1237                 DO64(drr_end.drr_toguid);
1238                 break;
1239         }
1240 #undef DO64
1241 #undef DO32
1242 }
1243
1244 static int
1245 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1246 {
1247         int err;
1248         dmu_tx_t *tx;
1249         void *data = NULL;
1250
1251         if (drro->drr_type == DMU_OT_NONE ||
1252             !DMU_OT_IS_VALID(drro->drr_type) ||
1253             !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1254             drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1255             drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1256             P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1257             drro->drr_blksz < SPA_MINBLOCKSIZE ||
1258             drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1259             drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1260                 return (SET_ERROR(EINVAL));
1261         }
1262
1263         err = dmu_object_info(os, drro->drr_object, NULL);
1264
1265         if (err != 0 && err != ENOENT)
1266                 return (SET_ERROR(EINVAL));
1267
1268         if (drro->drr_bonuslen) {
1269                 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1270                 if (ra->err != 0)
1271                         return (ra->err);
1272         }
1273
1274         if (err == ENOENT) {
1275                 /* currently free, want to be allocated */
1276                 tx = dmu_tx_create(os);
1277                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1278                 err = dmu_tx_assign(tx, TXG_WAIT);
1279                 if (err != 0) {
1280                         dmu_tx_abort(tx);
1281                         return (err);
1282                 }
1283                 err = dmu_object_claim(os, drro->drr_object,
1284                     drro->drr_type, drro->drr_blksz,
1285                     drro->drr_bonustype, drro->drr_bonuslen, tx);
1286                 dmu_tx_commit(tx);
1287         } else {
1288                 /* currently allocated, want to be allocated */
1289                 err = dmu_object_reclaim(os, drro->drr_object,
1290                     drro->drr_type, drro->drr_blksz,
1291                     drro->drr_bonustype, drro->drr_bonuslen);
1292         }
1293         if (err != 0) {
1294                 return (SET_ERROR(EINVAL));
1295         }
1296
1297         tx = dmu_tx_create(os);
1298         dmu_tx_hold_bonus(tx, drro->drr_object);
1299         err = dmu_tx_assign(tx, TXG_WAIT);
1300         if (err != 0) {
1301                 dmu_tx_abort(tx);
1302                 return (err);
1303         }
1304
1305         dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1306             tx);
1307         dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1308
1309         if (data != NULL) {
1310                 dmu_buf_t *db;
1311
1312                 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1313                 dmu_buf_will_dirty(db, tx);
1314
1315                 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1316                 bcopy(data, db->db_data, drro->drr_bonuslen);
1317                 if (ra->byteswap) {
1318                         dmu_object_byteswap_t byteswap =
1319                             DMU_OT_BYTESWAP(drro->drr_bonustype);
1320                         dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1321                             drro->drr_bonuslen);
1322                 }
1323                 dmu_buf_rele(db, FTAG);
1324         }
1325         dmu_tx_commit(tx);
1326         return (0);
1327 }
1328
1329 /* ARGSUSED */
1330 static int
1331 restore_freeobjects(struct restorearg *ra, objset_t *os,
1332     struct drr_freeobjects *drrfo)
1333 {
1334         uint64_t obj;
1335
1336         if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1337                 return (SET_ERROR(EINVAL));
1338
1339         for (obj = drrfo->drr_firstobj;
1340             obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1341             (void) dmu_object_next(os, &obj, FALSE, 0)) {
1342                 int err;
1343
1344                 if (dmu_object_info(os, obj, NULL) != 0)
1345                         continue;
1346
1347                 err = dmu_free_long_object(os, obj);
1348                 if (err != 0)
1349                         return (err);
1350         }
1351         return (0);
1352 }
1353
1354 static int
1355 restore_write(struct restorearg *ra, objset_t *os,
1356     struct drr_write *drrw)
1357 {
1358         dmu_tx_t *tx;
1359         void *data;
1360         int err;
1361
1362         if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1363             !DMU_OT_IS_VALID(drrw->drr_type))
1364                 return (SET_ERROR(EINVAL));
1365
1366         data = restore_read(ra, drrw->drr_length);
1367         if (data == NULL)
1368                 return (ra->err);
1369
1370         if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1371                 return (SET_ERROR(EINVAL));
1372
1373         tx = dmu_tx_create(os);
1374
1375         dmu_tx_hold_write(tx, drrw->drr_object,
1376             drrw->drr_offset, drrw->drr_length);
1377         err = dmu_tx_assign(tx, TXG_WAIT);
1378         if (err != 0) {
1379                 dmu_tx_abort(tx);
1380                 return (err);
1381         }
1382         if (ra->byteswap) {
1383                 dmu_object_byteswap_t byteswap =
1384                     DMU_OT_BYTESWAP(drrw->drr_type);
1385                 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1386         }
1387         dmu_write(os, drrw->drr_object,
1388             drrw->drr_offset, drrw->drr_length, data, tx);
1389         dmu_tx_commit(tx);
1390         return (0);
1391 }
1392
1393 /*
1394  * Handle a DRR_WRITE_BYREF record.  This record is used in dedup'ed
1395  * streams to refer to a copy of the data that is already on the
1396  * system because it came in earlier in the stream.  This function
1397  * finds the earlier copy of the data, and uses that copy instead of
1398  * data from the stream to fulfill this write.
1399  */
1400 static int
1401 restore_write_byref(struct restorearg *ra, objset_t *os,
1402     struct drr_write_byref *drrwbr)
1403 {
1404         dmu_tx_t *tx;
1405         int err;
1406         guid_map_entry_t gmesrch;
1407         guid_map_entry_t *gmep;
1408         avl_index_t     where;
1409         objset_t *ref_os = NULL;
1410         dmu_buf_t *dbp;
1411
1412         if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1413                 return (SET_ERROR(EINVAL));
1414
1415         /*
1416          * If the GUID of the referenced dataset is different from the
1417          * GUID of the target dataset, find the referenced dataset.
1418          */
1419         if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1420                 gmesrch.guid = drrwbr->drr_refguid;
1421                 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1422                     &where)) == NULL) {
1423                         return (SET_ERROR(EINVAL));
1424                 }
1425                 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1426                         return (SET_ERROR(EINVAL));
1427         } else {
1428                 ref_os = os;
1429         }
1430
1431         if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1432             drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1433                 return (err);
1434
1435         tx = dmu_tx_create(os);
1436
1437         dmu_tx_hold_write(tx, drrwbr->drr_object,
1438             drrwbr->drr_offset, drrwbr->drr_length);
1439         err = dmu_tx_assign(tx, TXG_WAIT);
1440         if (err != 0) {
1441                 dmu_tx_abort(tx);
1442                 return (err);
1443         }
1444         dmu_write(os, drrwbr->drr_object,
1445             drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1446         dmu_buf_rele(dbp, FTAG);
1447         dmu_tx_commit(tx);
1448         return (0);
1449 }
1450
1451 static int
1452 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1453 {
1454         dmu_tx_t *tx;
1455         void *data;
1456         dmu_buf_t *db, *db_spill;
1457         int err;
1458
1459         if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1460             drrs->drr_length > SPA_MAXBLOCKSIZE)
1461                 return (SET_ERROR(EINVAL));
1462
1463         data = restore_read(ra, drrs->drr_length);
1464         if (data == NULL)
1465                 return (ra->err);
1466
1467         if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1468                 return (SET_ERROR(EINVAL));
1469
1470         VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1471         if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1472                 dmu_buf_rele(db, FTAG);
1473                 return (err);
1474         }
1475
1476         tx = dmu_tx_create(os);
1477
1478         dmu_tx_hold_spill(tx, db->db_object);
1479
1480         err = dmu_tx_assign(tx, TXG_WAIT);
1481         if (err != 0) {
1482                 dmu_buf_rele(db, FTAG);
1483                 dmu_buf_rele(db_spill, FTAG);
1484                 dmu_tx_abort(tx);
1485                 return (err);
1486         }
1487         dmu_buf_will_dirty(db_spill, tx);
1488
1489         if (db_spill->db_size < drrs->drr_length)
1490                 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1491                     drrs->drr_length, tx));
1492         bcopy(data, db_spill->db_data, drrs->drr_length);
1493
1494         dmu_buf_rele(db, FTAG);
1495         dmu_buf_rele(db_spill, FTAG);
1496
1497         dmu_tx_commit(tx);
1498         return (0);
1499 }
1500
1501 /* ARGSUSED */
1502 static int
1503 restore_free(struct restorearg *ra, objset_t *os,
1504     struct drr_free *drrf)
1505 {
1506         int err;
1507
1508         if (drrf->drr_length != -1ULL &&
1509             drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1510                 return (SET_ERROR(EINVAL));
1511
1512         if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1513                 return (SET_ERROR(EINVAL));
1514
1515         err = dmu_free_long_range(os, drrf->drr_object,
1516             drrf->drr_offset, drrf->drr_length);
1517         return (err);
1518 }
1519
1520 /* used to destroy the drc_ds on error */
1521 static void
1522 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1523 {
1524         char name[MAXNAMELEN];
1525         dsl_dataset_name(drc->drc_ds, name);
1526         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1527         (void) dsl_destroy_head(name);
1528 }
1529
1530 /*
1531  * NB: callers *must* call dmu_recv_end() if this succeeds.
1532  */
1533 int
1534 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1535     int cleanup_fd, uint64_t *action_handlep)
1536 {
1537         struct restorearg ra = { 0 };
1538         dmu_replay_record_t *drr;
1539         objset_t *os;
1540         zio_cksum_t pcksum;
1541         int featureflags;
1542
1543         ra.byteswap = drc->drc_byteswap;
1544         ra.cksum = drc->drc_cksum;
1545         ra.td = curthread;
1546         ra.fp = fp;
1547         ra.voff = *voffp;
1548         ra.bufsize = 1<<20;
1549         ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1550
1551         /* these were verified in dmu_recv_begin */
1552         ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1553             DMU_SUBSTREAM);
1554         ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1555
1556         /*
1557          * Open the objset we are modifying.
1558          */
1559         VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1560
1561         ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1562
1563         featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1564
1565         /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1566         if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1567                 minor_t minor;
1568
1569                 if (cleanup_fd == -1) {
1570                         ra.err = SET_ERROR(EBADF);
1571                         goto out;
1572                 }
1573                 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1574                 if (ra.err != 0) {
1575                         cleanup_fd = -1;
1576                         goto out;
1577                 }
1578
1579                 if (*action_handlep == 0) {
1580                         ra.guid_to_ds_map =
1581                             kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1582                         avl_create(ra.guid_to_ds_map, guid_compare,
1583                             sizeof (guid_map_entry_t),
1584                             offsetof(guid_map_entry_t, avlnode));
1585                         ra.err = zfs_onexit_add_cb(minor,
1586                             free_guid_map_onexit, ra.guid_to_ds_map,
1587                             action_handlep);
1588                         if (ra.err != 0)
1589                                 goto out;
1590                 } else {
1591                         ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1592                             (void **)&ra.guid_to_ds_map);
1593                         if (ra.err != 0)
1594                                 goto out;
1595                 }
1596
1597                 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1598         }
1599
1600         /*
1601          * Read records and process them.
1602          */
1603         pcksum = ra.cksum;
1604         while (ra.err == 0 &&
1605             NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1606                 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1607                         ra.err = SET_ERROR(EINTR);
1608                         goto out;
1609                 }
1610
1611                 if (ra.byteswap)
1612                         backup_byteswap(drr);
1613
1614                 switch (drr->drr_type) {
1615                 case DRR_OBJECT:
1616                 {
1617                         /*
1618                          * We need to make a copy of the record header,
1619                          * because restore_{object,write} may need to
1620                          * restore_read(), which will invalidate drr.
1621                          */
1622                         struct drr_object drro = drr->drr_u.drr_object;
1623                         ra.err = restore_object(&ra, os, &drro);
1624                         break;
1625                 }
1626                 case DRR_FREEOBJECTS:
1627                 {
1628                         struct drr_freeobjects drrfo =
1629                             drr->drr_u.drr_freeobjects;
1630                         ra.err = restore_freeobjects(&ra, os, &drrfo);
1631                         break;
1632                 }
1633                 case DRR_WRITE:
1634                 {
1635                         struct drr_write drrw = drr->drr_u.drr_write;
1636                         ra.err = restore_write(&ra, os, &drrw);
1637                         break;
1638                 }
1639                 case DRR_WRITE_BYREF:
1640                 {
1641                         struct drr_write_byref drrwbr =
1642                             drr->drr_u.drr_write_byref;
1643                         ra.err = restore_write_byref(&ra, os, &drrwbr);
1644                         break;
1645                 }
1646                 case DRR_FREE:
1647                 {
1648                         struct drr_free drrf = drr->drr_u.drr_free;
1649                         ra.err = restore_free(&ra, os, &drrf);
1650                         break;
1651                 }
1652                 case DRR_END:
1653                 {
1654                         struct drr_end drre = drr->drr_u.drr_end;
1655                         /*
1656                          * We compare against the *previous* checksum
1657                          * value, because the stored checksum is of
1658                          * everything before the DRR_END record.
1659                          */
1660                         if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1661                                 ra.err = SET_ERROR(ECKSUM);
1662                         goto out;
1663                 }
1664                 case DRR_SPILL:
1665                 {
1666                         struct drr_spill drrs = drr->drr_u.drr_spill;
1667                         ra.err = restore_spill(&ra, os, &drrs);
1668                         break;
1669                 }
1670                 default:
1671                         ra.err = SET_ERROR(EINVAL);
1672                         goto out;
1673                 }
1674                 pcksum = ra.cksum;
1675         }
1676         ASSERT(ra.err != 0);
1677
1678 out:
1679         if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1680                 zfs_onexit_fd_rele(cleanup_fd);
1681
1682         if (ra.err != 0) {
1683                 /*
1684                  * destroy what we created, so we don't leave it in the
1685                  * inconsistent restoring state.
1686                  */
1687                 dmu_recv_cleanup_ds(drc);
1688         }
1689
1690         kmem_free(ra.buf, ra.bufsize);
1691         *voffp = ra.voff;
1692         return (ra.err);
1693 }
1694
1695 static int
1696 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1697 {
1698         dmu_recv_cookie_t *drc = arg;
1699         dsl_pool_t *dp = dmu_tx_pool(tx);
1700         int error;
1701
1702         ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1703
1704         if (!drc->drc_newfs) {
1705                 dsl_dataset_t *origin_head;
1706
1707                 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1708                 if (error != 0)
1709                         return (error);
1710                 if (drc->drc_force) {
1711                         /*
1712                          * We will destroy any snapshots in tofs (i.e. before
1713                          * origin_head) that are after the origin (which is
1714                          * the snap before drc_ds, because drc_ds can not
1715                          * have any snaps of its own).
1716                          */
1717                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1718                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1719                                 dsl_dataset_t *snap;
1720                                 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1721                                     &snap);
1722                                 if (error != 0)
1723                                         return (error);
1724                                 if (snap->ds_dir != origin_head->ds_dir)
1725                                         error = SET_ERROR(EINVAL);
1726                                 if (error == 0)  {
1727                                         error = dsl_destroy_snapshot_check_impl(
1728                                             snap, B_FALSE);
1729                                 }
1730                                 obj = snap->ds_phys->ds_prev_snap_obj;
1731                                 dsl_dataset_rele(snap, FTAG);
1732                                 if (error != 0)
1733                                         return (error);
1734                         }
1735                 }
1736                 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1737                     origin_head, drc->drc_force, drc->drc_owner, tx);
1738                 if (error != 0) {
1739                         dsl_dataset_rele(origin_head, FTAG);
1740                         return (error);
1741                 }
1742                 error = dsl_dataset_snapshot_check_impl(origin_head,
1743                     drc->drc_tosnap, tx, B_TRUE);
1744                 dsl_dataset_rele(origin_head, FTAG);
1745                 if (error != 0)
1746                         return (error);
1747
1748                 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1749         } else {
1750                 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1751                     drc->drc_tosnap, tx, B_TRUE);
1752         }
1753         return (error);
1754 }
1755
1756 static void
1757 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1758 {
1759         dmu_recv_cookie_t *drc = arg;
1760         dsl_pool_t *dp = dmu_tx_pool(tx);
1761
1762         spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1763             tx, "snap=%s", drc->drc_tosnap);
1764
1765         if (!drc->drc_newfs) {
1766                 dsl_dataset_t *origin_head;
1767
1768                 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1769                     &origin_head));
1770
1771                 if (drc->drc_force) {
1772                         /*
1773                          * Destroy any snapshots of drc_tofs (origin_head)
1774                          * after the origin (the snap before drc_ds).
1775                          */
1776                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1777                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1778                                 dsl_dataset_t *snap;
1779                                 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1780                                     &snap));
1781                                 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1782                                 obj = snap->ds_phys->ds_prev_snap_obj;
1783                                 dsl_destroy_snapshot_sync_impl(snap,
1784                                     B_FALSE, tx);
1785                                 dsl_dataset_rele(snap, FTAG);
1786                         }
1787                 }
1788                 VERIFY3P(drc->drc_ds->ds_prev, ==,
1789                     origin_head->ds_prev);
1790
1791                 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1792                     origin_head, tx);
1793                 dsl_dataset_snapshot_sync_impl(origin_head,
1794                     drc->drc_tosnap, tx);
1795
1796                 /* set snapshot's creation time and guid */
1797                 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1798                 origin_head->ds_prev->ds_phys->ds_creation_time =
1799                     drc->drc_drrb->drr_creation_time;
1800                 origin_head->ds_prev->ds_phys->ds_guid =
1801                     drc->drc_drrb->drr_toguid;
1802                 origin_head->ds_prev->ds_phys->ds_flags &=
1803                     ~DS_FLAG_INCONSISTENT;
1804
1805                 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1806                 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1807
1808                 dsl_dataset_rele(origin_head, FTAG);
1809                 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1810
1811                 if (drc->drc_owner != NULL)
1812                         VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
1813         } else {
1814                 dsl_dataset_t *ds = drc->drc_ds;
1815
1816                 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1817
1818                 /* set snapshot's creation time and guid */
1819                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1820                 ds->ds_prev->ds_phys->ds_creation_time =
1821                     drc->drc_drrb->drr_creation_time;
1822                 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1823                 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1824
1825                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1826                 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1827         }
1828         drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1829         /*
1830          * Release the hold from dmu_recv_begin.  This must be done before
1831          * we return to open context, so that when we free the dataset's dnode,
1832          * we can evict its bonus buffer.
1833          */
1834         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1835         drc->drc_ds = NULL;
1836 }
1837
1838 static int
1839 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1840 {
1841         dsl_pool_t *dp;
1842         dsl_dataset_t *snapds;
1843         guid_map_entry_t *gmep;
1844         int err;
1845
1846         ASSERT(guid_map != NULL);
1847
1848         err = dsl_pool_hold(name, FTAG, &dp);
1849         if (err != 0)
1850                 return (err);
1851         gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
1852         err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
1853         if (err == 0) {
1854                 gmep->guid = snapds->ds_phys->ds_guid;
1855                 gmep->gme_ds = snapds;
1856                 avl_add(guid_map, gmep);
1857                 dsl_dataset_long_hold(snapds, gmep);
1858         } else
1859                 kmem_free(gmep, sizeof (*gmep));
1860
1861         dsl_pool_rele(dp, FTAG);
1862         return (err);
1863 }
1864
1865 static int dmu_recv_end_modified_blocks = 3;
1866
1867 static int
1868 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1869 {
1870         int error;
1871         char name[MAXNAMELEN];
1872
1873 #ifdef _KERNEL
1874         /*
1875          * We will be destroying the ds; make sure its origin is unmounted if
1876          * necessary.
1877          */
1878         dsl_dataset_name(drc->drc_ds, name);
1879         zfs_destroy_unmount_origin(name);
1880 #endif
1881
1882         error = dsl_sync_task(drc->drc_tofs,
1883             dmu_recv_end_check, dmu_recv_end_sync, drc,
1884             dmu_recv_end_modified_blocks);
1885
1886         if (error != 0)
1887                 dmu_recv_cleanup_ds(drc);
1888         return (error);
1889 }
1890
1891 static int
1892 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1893 {
1894         int error;
1895
1896         error = dsl_sync_task(drc->drc_tofs,
1897             dmu_recv_end_check, dmu_recv_end_sync, drc,
1898             dmu_recv_end_modified_blocks);
1899
1900         if (error != 0) {
1901                 dmu_recv_cleanup_ds(drc);
1902         } else if (drc->drc_guid_to_ds_map != NULL) {
1903                 (void) add_ds_to_guidmap(drc->drc_tofs,
1904                     drc->drc_guid_to_ds_map,
1905                     drc->drc_newsnapobj);
1906         }
1907         return (error);
1908 }
1909
1910 int
1911 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
1912 {
1913         drc->drc_owner = owner;
1914
1915         if (drc->drc_newfs)
1916                 return (dmu_recv_new_end(drc));
1917         else
1918                 return (dmu_recv_existing_end(drc));
1919 }
1920
1921 /*
1922  * Return TRUE if this objset is currently being received into.
1923  */
1924 boolean_t
1925 dmu_objset_is_receiving(objset_t *os)
1926 {
1927         return (os->os_dsl_dataset != NULL &&
1928             os->os_dsl_dataset->ds_owner == dmu_recv_tag);
1929 }