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) 2014, 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         /*
 810          * Check snapshot limit before receiving. We'll recheck again at the
 811          * end, but might as well abort before receiving if we're already over
 812          * the limit.
 813          *
 814          * Note that we do not check the file system limit with
 815          * dsl_dir_fscount_check because the temporary %clones don't count
 816          * against that limit.
 817          */
 818         error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT,
 819             NULL, drba->drba_cred);
 820         if (error != 0)
 821                 return (error);
 822
 823         if (fromguid != 0) {
 824                 dsl_dataset_t *snap;
 825                 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
 826
 827                 /* Find snapshot in this dir that matches fromguid. */
 828                 while (obj != 0) {
 829                         error = dsl_dataset_hold_obj(dp, obj, FTAG,
 830                             &snap);
 831                         if (error != 0)
 832                                 return (SET_ERROR(ENODEV));
 833                         if (snap->ds_dir != ds->ds_dir) {
 834                                 dsl_dataset_rele(snap, FTAG);
 835                                 return (SET_ERROR(ENODEV));
 836                         }
 837                         if (snap->ds_phys->ds_guid == fromguid)
 838                                 break;
 839                         obj = snap->ds_phys->ds_prev_snap_obj;
 840                         dsl_dataset_rele(snap, FTAG);
 841                 }
 842                 if (obj == 0)
 843                         return (SET_ERROR(ENODEV));
 844
 845                 if (drba->drba_cookie->drc_force) {
 846                         drba->drba_snapobj = obj;
 847                 } else {
 848                         /*
 849                          * If we are not forcing, there must be no
 850                          * changes since fromsnap.
 851                          */
 852                         if (dsl_dataset_modified_since_snap(ds, snap)) {
 853                                 dsl_dataset_rele(snap, FTAG);
 854                                 return (SET_ERROR(ETXTBSY));
 855                         }
 856                         drba->drba_snapobj = ds->ds_prev->ds_object;
 857                 }
 858
 859                 dsl_dataset_rele(snap, FTAG);
 860         } else {
 861                 /* if full, most recent snapshot must be $ORIGIN */
 862                 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
 863                         return (SET_ERROR(ENODEV));
 864                 drba->drba_snapobj = ds->ds_phys->ds_prev_snap_obj;
 865         }
 866
 867         return (0);
 868
 869 }
 870
 871 static int
 872 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 873 {
 874         dmu_recv_begin_arg_t *drba = arg;
 875         dsl_pool_t *dp = dmu_tx_pool(tx);
 876         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 877         uint64_t fromguid = drrb->drr_fromguid;
 878         int flags = drrb->drr_flags;
 879         int error;
 880         dsl_dataset_t *ds;
 881         const char *tofs = drba->drba_cookie->drc_tofs;
 882
 883         /* already checked */
 884         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 885
 886         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 887             DMU_COMPOUNDSTREAM ||
 888             drrb->drr_type >= DMU_OST_NUMTYPES ||
 889             ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
 890                 return (SET_ERROR(EINVAL));
 891
 892         /* Verify pool version supports SA if SA_SPILL feature set */
 893         if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
 894             DMU_BACKUP_FEATURE_SA_SPILL) &&
 895             spa_version(dp->dp_spa) < SPA_VERSION_SA) {
 896                 return (SET_ERROR(ENOTSUP));
 897         }
 898
 899         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 900         if (error == 0) {
 901                 /* target fs already exists; recv into temp clone */
 902
 903                 /* Can't recv a clone into an existing fs */
 904                 if (flags & DRR_FLAG_CLONE) {
 905                         dsl_dataset_rele(ds, FTAG);
 906                         return (SET_ERROR(EINVAL));
 907                 }
 908
 909                 error = recv_begin_check_existing_impl(drba, ds, fromguid);
 910                 dsl_dataset_rele(ds, FTAG);
 911         } else if (error == ENOENT) {
 912                 /* target fs does not exist; must be a full backup or clone */
 913                 char buf[MAXNAMELEN];
 914
 915                 /*
 916                  * If it's a non-clone incremental, we are missing the
 917                  * target fs, so fail the recv.
 918                  */
 919                 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
 920                         return (SET_ERROR(ENOENT));
 921
 922                 /* Open the parent of tofs */
 923                 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
 924                 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
 925                 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
 926                 if (error != 0)
 927                         return (error);
 928
 929                 /*
 930                  * Check filesystem and snapshot limits before receiving. We'll
 931                  * recheck snapshot limits again at the end (we create the
 932                  * filesystems and increment those counts during begin_sync).
 933                  */
 934                 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
 935                     ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred);
 936                 if (error != 0) {
 937                         dsl_dataset_rele(ds, FTAG);
 938                         return (error);
 939                 }
 940
 941                 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
 942                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred);
 943                 if (error != 0) {
 944                         dsl_dataset_rele(ds, FTAG);
 945                         return (error);
 946                 }
 947
 948                 if (drba->drba_origin != NULL) {
 949                         dsl_dataset_t *origin;
 950                         error = dsl_dataset_hold(dp, drba->drba_origin,
 951                             FTAG, &origin);
 952                         if (error != 0) {
 953                                 dsl_dataset_rele(ds, FTAG);
 954                                 return (error);
 955                         }
 956                         if (!dsl_dataset_is_snapshot(origin)) {
 957                                 dsl_dataset_rele(origin, FTAG);
 958                                 dsl_dataset_rele(ds, FTAG);
 959                                 return (SET_ERROR(EINVAL));
 960                         }
 961                         if (origin->ds_phys->ds_guid != fromguid) {
 962                                 dsl_dataset_rele(origin, FTAG);
 963                                 dsl_dataset_rele(ds, FTAG);
 964                                 return (SET_ERROR(ENODEV));
 965                         }
 966                         dsl_dataset_rele(origin, FTAG);
 967                 }
 968                 dsl_dataset_rele(ds, FTAG);
 969                 error = 0;
 970         }
 971         return (error);
 972 }
 973
 974 static void
 975 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
 976 {
 977         dmu_recv_begin_arg_t *drba = arg;
 978         dsl_pool_t *dp = dmu_tx_pool(tx);
 979         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 980         const char *tofs = drba->drba_cookie->drc_tofs;
 981         dsl_dataset_t *ds, *newds;
 982         uint64_t dsobj;
 983         int error;
 984         uint64_t crflags;
 985
 986         crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
 987             DS_FLAG_CI_DATASET : 0;
 988
 989         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 990         if (error == 0) {
 991                 /* create temporary clone */
 992                 dsl_dataset_t *snap = NULL;
 993                 if (drba->drba_snapobj != 0) {
 994                         VERIFY0(dsl_dataset_hold_obj(dp,
 995                             drba->drba_snapobj, FTAG, &snap));
 996                 }
 997                 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
 998                     snap, crflags, drba->drba_cred, tx);
 999                 dsl_dataset_rele(snap, FTAG);
1000                 dsl_dataset_rele(ds, FTAG);
1001         } else {
1002                 dsl_dir_t *dd;
1003                 const char *tail;
1004                 dsl_dataset_t *origin = NULL;
1005
1006                 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1007
1008                 if (drba->drba_origin != NULL) {
1009                         VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1010                             FTAG, &origin));
1011                 }
1012
1013                 /* Create new dataset. */
1014                 dsobj = dsl_dataset_create_sync(dd,
1015                     strrchr(tofs, '/') + 1,
1016                     origin, crflags, drba->drba_cred, tx);
1017                 if (origin != NULL)
1018                         dsl_dataset_rele(origin, FTAG);
1019                 dsl_dir_rele(dd, FTAG);
1020                 drba->drba_cookie->drc_newfs = B_TRUE;
1021         }
1022         VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1023
1024         dmu_buf_will_dirty(newds->ds_dbuf, tx);
1025         newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1026
1027         /*
1028          * If we actually created a non-clone, we need to create the
1029          * objset in our new dataset.
1030          */
1031         if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1032                 (void) dmu_objset_create_impl(dp->dp_spa,
1033                     newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1034         }
1035
1036         drba->drba_cookie->drc_ds = newds;
1037
1038         spa_history_log_internal_ds(newds, "receive", tx, "");
1039 }
1040
1041 /*
1042  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1043  * succeeds; otherwise we will leak the holds on the datasets.
1044  */
1045 int
1046 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1047     boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1048 {
1049         dmu_recv_begin_arg_t drba = { 0 };
1050         dmu_replay_record_t *drr;
1051
1052         bzero(drc, sizeof (dmu_recv_cookie_t));
1053         drc->drc_drrb = drrb;
1054         drc->drc_tosnap = tosnap;
1055         drc->drc_tofs = tofs;
1056         drc->drc_force = force;
1057         drc->drc_cred = CRED();
1058
1059         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1060                 drc->drc_byteswap = B_TRUE;
1061         else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1062                 return (SET_ERROR(EINVAL));
1063
1064         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1065         drr->drr_type = DRR_BEGIN;
1066         drr->drr_u.drr_begin = *drc->drc_drrb;
1067         if (drc->drc_byteswap) {
1068                 fletcher_4_incremental_byteswap(drr,
1069                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1070         } else {
1071                 fletcher_4_incremental_native(drr,
1072                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1073         }
1074         kmem_free(drr, sizeof (dmu_replay_record_t));
1075
1076         if (drc->drc_byteswap) {
1077                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1078                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1079                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1080                 drrb->drr_type = BSWAP_32(drrb->drr_type);
1081                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1082                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1083         }
1084
1085         drba.drba_origin = origin;
1086         drba.drba_cookie = drc;
1087         drba.drba_cred = CRED();
1088
1089         return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1090             &drba, 5));
1091 }
1092
1093 struct restorearg {
1094         int err;
1095         boolean_t byteswap;
1096         kthread_t *td;
1097         struct file *fp;
1098         char *buf;
1099         uint64_t voff;
1100         int bufsize; /* amount of memory allocated for buf */
1101         zio_cksum_t cksum;
1102         avl_tree_t *guid_to_ds_map;
1103 };
1104
1105 typedef struct guid_map_entry {
1106         uint64_t        guid;
1107         dsl_dataset_t   *gme_ds;
1108         avl_node_t      avlnode;
1109 } guid_map_entry_t;
1110
1111 static int
1112 guid_compare(const void *arg1, const void *arg2)
1113 {
1114         const guid_map_entry_t *gmep1 = arg1;
1115         const guid_map_entry_t *gmep2 = arg2;
1116
1117         if (gmep1->guid < gmep2->guid)
1118                 return (-1);
1119         else if (gmep1->guid > gmep2->guid)
1120                 return (1);
1121         return (0);
1122 }
1123
1124 static void
1125 free_guid_map_onexit(void *arg)
1126 {
1127         avl_tree_t *ca = arg;
1128         void *cookie = NULL;
1129         guid_map_entry_t *gmep;
1130
1131         while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1132                 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1133                 dsl_dataset_rele(gmep->gme_ds, gmep);
1134                 kmem_free(gmep, sizeof (guid_map_entry_t));
1135         }
1136         avl_destroy(ca);
1137         kmem_free(ca, sizeof (avl_tree_t));
1138 }
1139
1140 static int
1141 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
1142 {
1143         struct uio auio;
1144         struct iovec aiov;
1145         int error;
1146
1147         aiov.iov_base = buf;
1148         aiov.iov_len = len;
1149         auio.uio_iov = &aiov;
1150         auio.uio_iovcnt = 1;
1151         auio.uio_resid = len;
1152         auio.uio_segflg = UIO_SYSSPACE;
1153         auio.uio_rw = UIO_READ;
1154         auio.uio_offset = off;
1155         auio.uio_td = ra->td;
1156 #ifdef _KERNEL
1157         error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
1158 #else
1159         fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
1160         error = EOPNOTSUPP;
1161 #endif
1162         *resid = auio.uio_resid;
1163         return (error);
1164 }
1165
1166 static void *
1167 restore_read(struct restorearg *ra, int len)
1168 {
1169         void *rv;
1170         int done = 0;
1171
1172         /* some things will require 8-byte alignment, so everything must */
1173         ASSERT0(len % 8);
1174
1175         while (done < len) {
1176                 ssize_t resid;
1177
1178                 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
1179                     len - done, ra->voff, &resid);
1180
1181                 if (resid == len - done)
1182                         ra->err = SET_ERROR(EINVAL);
1183                 ra->voff += len - done - resid;
1184                 done = len - resid;
1185                 if (ra->err != 0)
1186                         return (NULL);
1187         }
1188
1189         ASSERT3U(done, ==, len);
1190         rv = ra->buf;
1191         if (ra->byteswap)
1192                 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1193         else
1194                 fletcher_4_incremental_native(rv, len, &ra->cksum);
1195         return (rv);
1196 }
1197
1198 static void
1199 backup_byteswap(dmu_replay_record_t *drr)
1200 {
1201 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1202 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1203         drr->drr_type = BSWAP_32(drr->drr_type);
1204         drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1205         switch (drr->drr_type) {
1206         case DRR_BEGIN:
1207                 DO64(drr_begin.drr_magic);
1208                 DO64(drr_begin.drr_versioninfo);
1209                 DO64(drr_begin.drr_creation_time);
1210                 DO32(drr_begin.drr_type);
1211                 DO32(drr_begin.drr_flags);
1212                 DO64(drr_begin.drr_toguid);
1213                 DO64(drr_begin.drr_fromguid);
1214                 break;
1215         case DRR_OBJECT:
1216                 DO64(drr_object.drr_object);
1217                 /* DO64(drr_object.drr_allocation_txg); */
1218                 DO32(drr_object.drr_type);
1219                 DO32(drr_object.drr_bonustype);
1220                 DO32(drr_object.drr_blksz);
1221                 DO32(drr_object.drr_bonuslen);
1222                 DO64(drr_object.drr_toguid);
1223                 break;
1224         case DRR_FREEOBJECTS:
1225                 DO64(drr_freeobjects.drr_firstobj);
1226                 DO64(drr_freeobjects.drr_numobjs);
1227                 DO64(drr_freeobjects.drr_toguid);
1228                 break;
1229         case DRR_WRITE:
1230                 DO64(drr_write.drr_object);
1231                 DO32(drr_write.drr_type);
1232                 DO64(drr_write.drr_offset);
1233                 DO64(drr_write.drr_length);
1234                 DO64(drr_write.drr_toguid);
1235                 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1236                 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1237                 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1238                 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1239                 DO64(drr_write.drr_key.ddk_prop);
1240                 break;
1241         case DRR_WRITE_BYREF:
1242                 DO64(drr_write_byref.drr_object);
1243                 DO64(drr_write_byref.drr_offset);
1244                 DO64(drr_write_byref.drr_length);
1245                 DO64(drr_write_byref.drr_toguid);
1246                 DO64(drr_write_byref.drr_refguid);
1247                 DO64(drr_write_byref.drr_refobject);
1248                 DO64(drr_write_byref.drr_refoffset);
1249                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1250                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1251                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1252                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1253                 DO64(drr_write_byref.drr_key.ddk_prop);
1254                 break;
1255         case DRR_FREE:
1256                 DO64(drr_free.drr_object);
1257                 DO64(drr_free.drr_offset);
1258                 DO64(drr_free.drr_length);
1259                 DO64(drr_free.drr_toguid);
1260                 break;
1261         case DRR_SPILL:
1262                 DO64(drr_spill.drr_object);
1263                 DO64(drr_spill.drr_length);
1264                 DO64(drr_spill.drr_toguid);
1265                 break;
1266         case DRR_END:
1267                 DO64(drr_end.drr_checksum.zc_word[0]);
1268                 DO64(drr_end.drr_checksum.zc_word[1]);
1269                 DO64(drr_end.drr_checksum.zc_word[2]);
1270                 DO64(drr_end.drr_checksum.zc_word[3]);
1271                 DO64(drr_end.drr_toguid);
1272                 break;
1273         }
1274 #undef DO64
1275 #undef DO32
1276 }
1277
1278 static int
1279 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1280 {
1281         int err;
1282         dmu_tx_t *tx;
1283         void *data = NULL;
1284
1285         if (drro->drr_type == DMU_OT_NONE ||
1286             !DMU_OT_IS_VALID(drro->drr_type) ||
1287             !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1288             drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1289             drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1290             P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1291             drro->drr_blksz < SPA_MINBLOCKSIZE ||
1292             drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1293             drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1294                 return (SET_ERROR(EINVAL));
1295         }
1296
1297         err = dmu_object_info(os, drro->drr_object, NULL);
1298
1299         if (err != 0 && err != ENOENT)
1300                 return (SET_ERROR(EINVAL));
1301
1302         if (drro->drr_bonuslen) {
1303                 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1304                 if (ra->err != 0)
1305                         return (ra->err);
1306         }
1307
1308         if (err == ENOENT) {
1309                 /* currently free, want to be allocated */
1310                 tx = dmu_tx_create(os);
1311                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1312                 err = dmu_tx_assign(tx, TXG_WAIT);
1313                 if (err != 0) {
1314                         dmu_tx_abort(tx);
1315                         return (err);
1316                 }
1317                 err = dmu_object_claim(os, drro->drr_object,
1318                     drro->drr_type, drro->drr_blksz,
1319                     drro->drr_bonustype, drro->drr_bonuslen, tx);
1320                 dmu_tx_commit(tx);
1321         } else {
1322                 /* currently allocated, want to be allocated */
1323                 err = dmu_object_reclaim(os, drro->drr_object,
1324                     drro->drr_type, drro->drr_blksz,
1325                     drro->drr_bonustype, drro->drr_bonuslen);
1326         }
1327         if (err != 0) {
1328                 return (SET_ERROR(EINVAL));
1329         }
1330
1331         tx = dmu_tx_create(os);
1332         dmu_tx_hold_bonus(tx, drro->drr_object);
1333         err = dmu_tx_assign(tx, TXG_WAIT);
1334         if (err != 0) {
1335                 dmu_tx_abort(tx);
1336                 return (err);
1337         }
1338
1339         dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1340             tx);
1341         dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1342
1343         if (data != NULL) {
1344                 dmu_buf_t *db;
1345
1346                 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1347                 dmu_buf_will_dirty(db, tx);
1348
1349                 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1350                 bcopy(data, db->db_data, drro->drr_bonuslen);
1351                 if (ra->byteswap) {
1352                         dmu_object_byteswap_t byteswap =
1353                             DMU_OT_BYTESWAP(drro->drr_bonustype);
1354                         dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1355                             drro->drr_bonuslen);
1356                 }
1357                 dmu_buf_rele(db, FTAG);
1358         }
1359         dmu_tx_commit(tx);
1360         return (0);
1361 }
1362
1363 /* ARGSUSED */
1364 static int
1365 restore_freeobjects(struct restorearg *ra, objset_t *os,
1366     struct drr_freeobjects *drrfo)
1367 {
1368         uint64_t obj;
1369
1370         if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1371                 return (SET_ERROR(EINVAL));
1372
1373         for (obj = drrfo->drr_firstobj;
1374             obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1375             (void) dmu_object_next(os, &obj, FALSE, 0)) {
1376                 int err;
1377
1378                 if (dmu_object_info(os, obj, NULL) != 0)
1379                         continue;
1380
1381                 err = dmu_free_long_object(os, obj);
1382                 if (err != 0)
1383                         return (err);
1384         }
1385         return (0);
1386 }
1387
1388 static int
1389 restore_write(struct restorearg *ra, objset_t *os,
1390     struct drr_write *drrw)
1391 {
1392         dmu_tx_t *tx;
1393         void *data;
1394         int err;
1395
1396         if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1397             !DMU_OT_IS_VALID(drrw->drr_type))
1398                 return (SET_ERROR(EINVAL));
1399
1400         data = restore_read(ra, drrw->drr_length);
1401         if (data == NULL)
1402                 return (ra->err);
1403
1404         if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1405                 return (SET_ERROR(EINVAL));
1406
1407         tx = dmu_tx_create(os);
1408
1409         dmu_tx_hold_write(tx, drrw->drr_object,
1410             drrw->drr_offset, drrw->drr_length);
1411         err = dmu_tx_assign(tx, TXG_WAIT);
1412         if (err != 0) {
1413                 dmu_tx_abort(tx);
1414                 return (err);
1415         }
1416         if (ra->byteswap) {
1417                 dmu_object_byteswap_t byteswap =
1418                     DMU_OT_BYTESWAP(drrw->drr_type);
1419                 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1420         }
1421         dmu_write(os, drrw->drr_object,
1422             drrw->drr_offset, drrw->drr_length, data, tx);
1423         dmu_tx_commit(tx);
1424         return (0);
1425 }
1426
1427 /*
1428  * Handle a DRR_WRITE_BYREF record.  This record is used in dedup'ed
1429  * streams to refer to a copy of the data that is already on the
1430  * system because it came in earlier in the stream.  This function
1431  * finds the earlier copy of the data, and uses that copy instead of
1432  * data from the stream to fulfill this write.
1433  */
1434 static int
1435 restore_write_byref(struct restorearg *ra, objset_t *os,
1436     struct drr_write_byref *drrwbr)
1437 {
1438         dmu_tx_t *tx;
1439         int err;
1440         guid_map_entry_t gmesrch;
1441         guid_map_entry_t *gmep;
1442         avl_index_t     where;
1443         objset_t *ref_os = NULL;
1444         dmu_buf_t *dbp;
1445
1446         if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1447                 return (SET_ERROR(EINVAL));
1448
1449         /*
1450          * If the GUID of the referenced dataset is different from the
1451          * GUID of the target dataset, find the referenced dataset.
1452          */
1453         if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1454                 gmesrch.guid = drrwbr->drr_refguid;
1455                 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1456                     &where)) == NULL) {
1457                         return (SET_ERROR(EINVAL));
1458                 }
1459                 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1460                         return (SET_ERROR(EINVAL));
1461         } else {
1462                 ref_os = os;
1463         }
1464
1465         if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1466             drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1467                 return (err);
1468
1469         tx = dmu_tx_create(os);
1470
1471         dmu_tx_hold_write(tx, drrwbr->drr_object,
1472             drrwbr->drr_offset, drrwbr->drr_length);
1473         err = dmu_tx_assign(tx, TXG_WAIT);
1474         if (err != 0) {
1475                 dmu_tx_abort(tx);
1476                 return (err);
1477         }
1478         dmu_write(os, drrwbr->drr_object,
1479             drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1480         dmu_buf_rele(dbp, FTAG);
1481         dmu_tx_commit(tx);
1482         return (0);
1483 }
1484
1485 static int
1486 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1487 {
1488         dmu_tx_t *tx;
1489         void *data;
1490         dmu_buf_t *db, *db_spill;
1491         int err;
1492
1493         if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1494             drrs->drr_length > SPA_MAXBLOCKSIZE)
1495                 return (SET_ERROR(EINVAL));
1496
1497         data = restore_read(ra, drrs->drr_length);
1498         if (data == NULL)
1499                 return (ra->err);
1500
1501         if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1502                 return (SET_ERROR(EINVAL));
1503
1504         VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1505         if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1506                 dmu_buf_rele(db, FTAG);
1507                 return (err);
1508         }
1509
1510         tx = dmu_tx_create(os);
1511
1512         dmu_tx_hold_spill(tx, db->db_object);
1513
1514         err = dmu_tx_assign(tx, TXG_WAIT);
1515         if (err != 0) {
1516                 dmu_buf_rele(db, FTAG);
1517                 dmu_buf_rele(db_spill, FTAG);
1518                 dmu_tx_abort(tx);
1519                 return (err);
1520         }
1521         dmu_buf_will_dirty(db_spill, tx);
1522
1523         if (db_spill->db_size < drrs->drr_length)
1524                 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1525                     drrs->drr_length, tx));
1526         bcopy(data, db_spill->db_data, drrs->drr_length);
1527
1528         dmu_buf_rele(db, FTAG);
1529         dmu_buf_rele(db_spill, FTAG);
1530
1531         dmu_tx_commit(tx);
1532         return (0);
1533 }
1534
1535 /* ARGSUSED */
1536 static int
1537 restore_free(struct restorearg *ra, objset_t *os,
1538     struct drr_free *drrf)
1539 {
1540         int err;
1541
1542         if (drrf->drr_length != -1ULL &&
1543             drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1544                 return (SET_ERROR(EINVAL));
1545
1546         if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1547                 return (SET_ERROR(EINVAL));
1548
1549         err = dmu_free_long_range(os, drrf->drr_object,
1550             drrf->drr_offset, drrf->drr_length);
1551         return (err);
1552 }
1553
1554 /* used to destroy the drc_ds on error */
1555 static void
1556 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1557 {
1558         char name[MAXNAMELEN];
1559         dsl_dataset_name(drc->drc_ds, name);
1560         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1561         (void) dsl_destroy_head(name);
1562 }
1563
1564 /*
1565  * NB: callers *must* call dmu_recv_end() if this succeeds.
1566  */
1567 int
1568 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1569     int cleanup_fd, uint64_t *action_handlep)
1570 {
1571         struct restorearg ra = { 0 };
1572         dmu_replay_record_t *drr;
1573         objset_t *os;
1574         zio_cksum_t pcksum;
1575         int featureflags;
1576
1577         ra.byteswap = drc->drc_byteswap;
1578         ra.cksum = drc->drc_cksum;
1579         ra.td = curthread;
1580         ra.fp = fp;
1581         ra.voff = *voffp;
1582         ra.bufsize = 1<<20;
1583         ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1584
1585         /* these were verified in dmu_recv_begin */
1586         ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1587             DMU_SUBSTREAM);
1588         ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1589
1590         /*
1591          * Open the objset we are modifying.
1592          */
1593         VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1594
1595         ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1596
1597         featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1598
1599         /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1600         if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1601                 minor_t minor;
1602
1603                 if (cleanup_fd == -1) {
1604                         ra.err = SET_ERROR(EBADF);
1605                         goto out;
1606                 }
1607                 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1608                 if (ra.err != 0) {
1609                         cleanup_fd = -1;
1610                         goto out;
1611                 }
1612
1613                 if (*action_handlep == 0) {
1614                         ra.guid_to_ds_map =
1615                             kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1616                         avl_create(ra.guid_to_ds_map, guid_compare,
1617                             sizeof (guid_map_entry_t),
1618                             offsetof(guid_map_entry_t, avlnode));
1619                         ra.err = zfs_onexit_add_cb(minor,
1620                             free_guid_map_onexit, ra.guid_to_ds_map,
1621                             action_handlep);
1622                         if (ra.err != 0)
1623                                 goto out;
1624                 } else {
1625                         ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1626                             (void **)&ra.guid_to_ds_map);
1627                         if (ra.err != 0)
1628                                 goto out;
1629                 }
1630
1631                 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1632         }
1633
1634         /*
1635          * Read records and process them.
1636          */
1637         pcksum = ra.cksum;
1638         while (ra.err == 0 &&
1639             NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1640                 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1641                         ra.err = SET_ERROR(EINTR);
1642                         goto out;
1643                 }
1644
1645                 if (ra.byteswap)
1646                         backup_byteswap(drr);
1647
1648                 switch (drr->drr_type) {
1649                 case DRR_OBJECT:
1650                 {
1651                         /*
1652                          * We need to make a copy of the record header,
1653                          * because restore_{object,write} may need to
1654                          * restore_read(), which will invalidate drr.
1655                          */
1656                         struct drr_object drro = drr->drr_u.drr_object;
1657                         ra.err = restore_object(&ra, os, &drro);
1658                         break;
1659                 }
1660                 case DRR_FREEOBJECTS:
1661                 {
1662                         struct drr_freeobjects drrfo =
1663                             drr->drr_u.drr_freeobjects;
1664                         ra.err = restore_freeobjects(&ra, os, &drrfo);
1665                         break;
1666                 }
1667                 case DRR_WRITE:
1668                 {
1669                         struct drr_write drrw = drr->drr_u.drr_write;
1670                         ra.err = restore_write(&ra, os, &drrw);
1671                         break;
1672                 }
1673                 case DRR_WRITE_BYREF:
1674                 {
1675                         struct drr_write_byref drrwbr =
1676                             drr->drr_u.drr_write_byref;
1677                         ra.err = restore_write_byref(&ra, os, &drrwbr);
1678                         break;
1679                 }
1680                 case DRR_FREE:
1681                 {
1682                         struct drr_free drrf = drr->drr_u.drr_free;
1683                         ra.err = restore_free(&ra, os, &drrf);
1684                         break;
1685                 }
1686                 case DRR_END:
1687                 {
1688                         struct drr_end drre = drr->drr_u.drr_end;
1689                         /*
1690                          * We compare against the *previous* checksum
1691                          * value, because the stored checksum is of
1692                          * everything before the DRR_END record.
1693                          */
1694                         if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1695                                 ra.err = SET_ERROR(ECKSUM);
1696                         goto out;
1697                 }
1698                 case DRR_SPILL:
1699                 {
1700                         struct drr_spill drrs = drr->drr_u.drr_spill;
1701                         ra.err = restore_spill(&ra, os, &drrs);
1702                         break;
1703                 }
1704                 default:
1705                         ra.err = SET_ERROR(EINVAL);
1706                         goto out;
1707                 }
1708                 pcksum = ra.cksum;
1709         }
1710         ASSERT(ra.err != 0);
1711
1712 out:
1713         if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1714                 zfs_onexit_fd_rele(cleanup_fd);
1715
1716         if (ra.err != 0) {
1717                 /*
1718                  * destroy what we created, so we don't leave it in the
1719                  * inconsistent restoring state.
1720                  */
1721                 dmu_recv_cleanup_ds(drc);
1722         }
1723
1724         kmem_free(ra.buf, ra.bufsize);
1725         *voffp = ra.voff;
1726         return (ra.err);
1727 }
1728
1729 static int
1730 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1731 {
1732         dmu_recv_cookie_t *drc = arg;
1733         dsl_pool_t *dp = dmu_tx_pool(tx);
1734         int error;
1735
1736         ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1737
1738         if (!drc->drc_newfs) {
1739                 dsl_dataset_t *origin_head;
1740
1741                 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1742                 if (error != 0)
1743                         return (error);
1744                 if (drc->drc_force) {
1745                         /*
1746                          * We will destroy any snapshots in tofs (i.e. before
1747                          * origin_head) that are after the origin (which is
1748                          * the snap before drc_ds, because drc_ds can not
1749                          * have any snaps of its own).
1750                          */
1751                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1752                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1753                                 dsl_dataset_t *snap;
1754                                 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1755                                     &snap);
1756                                 if (error != 0)
1757                                         return (error);
1758                                 if (snap->ds_dir != origin_head->ds_dir)
1759                                         error = SET_ERROR(EINVAL);
1760                                 if (error == 0)  {
1761                                         error = dsl_destroy_snapshot_check_impl(
1762                                             snap, B_FALSE);
1763                                 }
1764                                 obj = snap->ds_phys->ds_prev_snap_obj;
1765                                 dsl_dataset_rele(snap, FTAG);
1766                                 if (error != 0)
1767                                         return (error);
1768                         }
1769                 }
1770                 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1771                     origin_head, drc->drc_force, drc->drc_owner, tx);
1772                 if (error != 0) {
1773                         dsl_dataset_rele(origin_head, FTAG);
1774                         return (error);
1775                 }
1776                 error = dsl_dataset_snapshot_check_impl(origin_head,
1777                     drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
1778                 dsl_dataset_rele(origin_head, FTAG);
1779                 if (error != 0)
1780                         return (error);
1781
1782                 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1783         } else {
1784                 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1785                     drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
1786         }
1787         return (error);
1788 }
1789
1790 static void
1791 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1792 {
1793         dmu_recv_cookie_t *drc = arg;
1794         dsl_pool_t *dp = dmu_tx_pool(tx);
1795
1796         spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1797             tx, "snap=%s", drc->drc_tosnap);
1798
1799         if (!drc->drc_newfs) {
1800                 dsl_dataset_t *origin_head;
1801
1802                 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1803                     &origin_head));
1804
1805                 if (drc->drc_force) {
1806                         /*
1807                          * Destroy any snapshots of drc_tofs (origin_head)
1808                          * after the origin (the snap before drc_ds).
1809                          */
1810                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1811                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1812                                 dsl_dataset_t *snap;
1813                                 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1814                                     &snap));
1815                                 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1816                                 obj = snap->ds_phys->ds_prev_snap_obj;
1817                                 dsl_destroy_snapshot_sync_impl(snap,
1818                                     B_FALSE, tx);
1819                                 dsl_dataset_rele(snap, FTAG);
1820                         }
1821                 }
1822                 VERIFY3P(drc->drc_ds->ds_prev, ==,
1823                     origin_head->ds_prev);
1824
1825                 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1826                     origin_head, tx);
1827                 dsl_dataset_snapshot_sync_impl(origin_head,
1828                     drc->drc_tosnap, tx);
1829
1830                 /* set snapshot's creation time and guid */
1831                 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1832                 origin_head->ds_prev->ds_phys->ds_creation_time =
1833                     drc->drc_drrb->drr_creation_time;
1834                 origin_head->ds_prev->ds_phys->ds_guid =
1835                     drc->drc_drrb->drr_toguid;
1836                 origin_head->ds_prev->ds_phys->ds_flags &=
1837                     ~DS_FLAG_INCONSISTENT;
1838
1839                 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1840                 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1841
1842                 dsl_dataset_rele(origin_head, FTAG);
1843                 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1844
1845                 if (drc->drc_owner != NULL)
1846                         VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
1847         } else {
1848                 dsl_dataset_t *ds = drc->drc_ds;
1849
1850                 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1851
1852                 /* set snapshot's creation time and guid */
1853                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1854                 ds->ds_prev->ds_phys->ds_creation_time =
1855                     drc->drc_drrb->drr_creation_time;
1856                 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1857                 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1858
1859                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1860                 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1861         }
1862         drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1863         /*
1864          * Release the hold from dmu_recv_begin.  This must be done before
1865          * we return to open context, so that when we free the dataset's dnode,
1866          * we can evict its bonus buffer.
1867          */
1868         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1869         drc->drc_ds = NULL;
1870 }
1871
1872 static int
1873 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1874 {
1875         dsl_pool_t *dp;
1876         dsl_dataset_t *snapds;
1877         guid_map_entry_t *gmep;
1878         int err;
1879
1880         ASSERT(guid_map != NULL);
1881
1882         err = dsl_pool_hold(name, FTAG, &dp);
1883         if (err != 0)
1884                 return (err);
1885         gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
1886         err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
1887         if (err == 0) {
1888                 gmep->guid = snapds->ds_phys->ds_guid;
1889                 gmep->gme_ds = snapds;
1890                 avl_add(guid_map, gmep);
1891                 dsl_dataset_long_hold(snapds, gmep);
1892         } else
1893                 kmem_free(gmep, sizeof (*gmep));
1894
1895         dsl_pool_rele(dp, FTAG);
1896         return (err);
1897 }
1898
1899 static int dmu_recv_end_modified_blocks = 3;
1900
1901 static int
1902 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1903 {
1904         int error;
1905         char name[MAXNAMELEN];
1906
1907 #ifdef _KERNEL
1908         /*
1909          * We will be destroying the ds; make sure its origin is unmounted if
1910          * necessary.
1911          */
1912         dsl_dataset_name(drc->drc_ds, name);
1913         zfs_destroy_unmount_origin(name);
1914 #endif
1915
1916         error = dsl_sync_task(drc->drc_tofs,
1917             dmu_recv_end_check, dmu_recv_end_sync, drc,
1918             dmu_recv_end_modified_blocks);
1919
1920         if (error != 0)
1921                 dmu_recv_cleanup_ds(drc);
1922         return (error);
1923 }
1924
1925 static int
1926 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1927 {
1928         int error;
1929
1930         error = dsl_sync_task(drc->drc_tofs,
1931             dmu_recv_end_check, dmu_recv_end_sync, drc,
1932             dmu_recv_end_modified_blocks);
1933
1934         if (error != 0) {
1935                 dmu_recv_cleanup_ds(drc);
1936         } else if (drc->drc_guid_to_ds_map != NULL) {
1937                 (void) add_ds_to_guidmap(drc->drc_tofs,
1938                     drc->drc_guid_to_ds_map,
1939                     drc->drc_newsnapobj);
1940         }
1941         return (error);
1942 }
1943
1944 int
1945 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
1946 {
1947         drc->drc_owner = owner;
1948
1949         if (drc->drc_newfs)
1950                 return (dmu_recv_new_end(drc));
1951         else
1952                 return (dmu_recv_existing_end(drc));
1953 }
1954
1955 /*
1956  * Return TRUE if this objset is currently being received into.
1957  */
1958 boolean_t
1959 dmu_objset_is_receiving(objset_t *os)
1960 {
1961         return (os->os_dsl_dataset != NULL &&
1962             os->os_dsl_dataset->ds_owner == dmu_recv_tag);
1963 }