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) 2011, 2014 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/blkptr.h>
  53 #include <sys/dsl_bookmark.h>
  54 #include <sys/zfeature.h>
  55
  56 #ifdef __FreeBSD__
  57 #undef dump_write
  58 #define dump_write dmu_dump_write
  59 #endif
  60
  61 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
  62 int zfs_send_corrupt_data = B_FALSE;
  63
  64 static char *dmu_recv_tag = "dmu_recv_tag";
  65 static const char *recv_clone_name = "%recv";
  66
  67 static int
  68 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
  69 {
  70         dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
  71         struct uio auio;
  72         struct iovec aiov;
  73         ASSERT0(len % 8);
  74
  75         fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
  76         aiov.iov_base = buf;
  77         aiov.iov_len = len;
  78         auio.uio_iov = &aiov;
  79         auio.uio_iovcnt = 1;
  80         auio.uio_resid = len;
  81         auio.uio_segflg = UIO_SYSSPACE;
  82         auio.uio_rw = UIO_WRITE;
  83         auio.uio_offset = (off_t)-1;
  84         auio.uio_td = dsp->dsa_td;
  85 #ifdef _KERNEL
  86         if (dsp->dsa_fp->f_type == DTYPE_VNODE)
  87                 bwillwrite();
  88         dsp->dsa_err = fo_write(dsp->dsa_fp, &auio, dsp->dsa_td->td_ucred, 0,
  89             dsp->dsa_td);
  90 #else
  91         fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
  92         dsp->dsa_err = EOPNOTSUPP;
  93 #endif
  94         mutex_enter(&ds->ds_sendstream_lock);
  95         *dsp->dsa_off += len;
  96         mutex_exit(&ds->ds_sendstream_lock);
  97
  98         return (dsp->dsa_err);
  99 }
 100
 101 static int
 102 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
 103     uint64_t length)
 104 {
 105         struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
 106
 107         /*
 108          * When we receive a free record, dbuf_free_range() assumes
 109          * that the receiving system doesn't have any dbufs in the range
 110          * being freed.  This is always true because there is a one-record
 111          * constraint: we only send one WRITE record for any given
 112          * object+offset.  We know that the one-record constraint is
 113          * true because we always send data in increasing order by
 114          * object,offset.
 115          *
 116          * If the increasing-order constraint ever changes, we should find
 117          * another way to assert that the one-record constraint is still
 118          * satisfied.
 119          */
 120         ASSERT(object > dsp->dsa_last_data_object ||
 121             (object == dsp->dsa_last_data_object &&
 122             offset > dsp->dsa_last_data_offset));
 123
 124         /*
 125          * If we are doing a non-incremental send, then there can't
 126          * be any data in the dataset we're receiving into.  Therefore
 127          * a free record would simply be a no-op.  Save space by not
 128          * sending it to begin with.
 129          */
 130         if (!dsp->dsa_incremental)
 131                 return (0);
 132
 133         if (length != -1ULL && offset + length < offset)
 134                 length = -1ULL;
 135
 136         /*
 137          * If there is a pending op, but it's not PENDING_FREE, push it out,
 138          * since free block aggregation can only be done for blocks of the
 139          * same type (i.e., DRR_FREE records can only be aggregated with
 140          * other DRR_FREE records.  DRR_FREEOBJECTS records can only be
 141          * aggregated with other DRR_FREEOBJECTS records.
 142          */
 143         if (dsp->dsa_pending_op != PENDING_NONE &&
 144             dsp->dsa_pending_op != PENDING_FREE) {
 145                 if (dump_bytes(dsp, dsp->dsa_drr,
 146                     sizeof (dmu_replay_record_t)) != 0)
 147                         return (SET_ERROR(EINTR));
 148                 dsp->dsa_pending_op = PENDING_NONE;
 149         }
 150
 151         if (dsp->dsa_pending_op == PENDING_FREE) {
 152                 /*
 153                  * There should never be a PENDING_FREE if length is -1
 154                  * (because dump_dnode is the only place where this
 155                  * function is called with a -1, and only after flushing
 156                  * any pending record).
 157                  */
 158                 ASSERT(length != -1ULL);
 159                 /*
 160                  * Check to see whether this free block can be aggregated
 161                  * with pending one.
 162                  */
 163                 if (drrf->drr_object == object && drrf->drr_offset +
 164                     drrf->drr_length == offset) {
 165                         drrf->drr_length += length;
 166                         return (0);
 167                 } else {
 168                         /* not a continuation.  Push out pending record */
 169                         if (dump_bytes(dsp, dsp->dsa_drr,
 170                             sizeof (dmu_replay_record_t)) != 0)
 171                                 return (SET_ERROR(EINTR));
 172                         dsp->dsa_pending_op = PENDING_NONE;
 173                 }
 174         }
 175         /* create a FREE record and make it pending */
 176         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 177         dsp->dsa_drr->drr_type = DRR_FREE;
 178         drrf->drr_object = object;
 179         drrf->drr_offset = offset;
 180         drrf->drr_length = length;
 181         drrf->drr_toguid = dsp->dsa_toguid;
 182         if (length == -1ULL) {
 183                 if (dump_bytes(dsp, dsp->dsa_drr,
 184                     sizeof (dmu_replay_record_t)) != 0)
 185                         return (SET_ERROR(EINTR));
 186         } else {
 187                 dsp->dsa_pending_op = PENDING_FREE;
 188         }
 189
 190         return (0);
 191 }
 192
 193 static int
 194 dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type,
 195     uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
 196 {
 197         struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
 198
 199         /*
 200          * We send data in increasing object, offset order.
 201          * See comment in dump_free() for details.
 202          */
 203         ASSERT(object > dsp->dsa_last_data_object ||
 204             (object == dsp->dsa_last_data_object &&
 205             offset > dsp->dsa_last_data_offset));
 206         dsp->dsa_last_data_object = object;
 207         dsp->dsa_last_data_offset = offset + blksz - 1;
 208
 209         /*
 210          * If there is any kind of pending aggregation (currently either
 211          * a grouping of free objects or free blocks), push it out to
 212          * the stream, since aggregation can't be done across operations
 213          * of different types.
 214          */
 215         if (dsp->dsa_pending_op != PENDING_NONE) {
 216                 if (dump_bytes(dsp, dsp->dsa_drr,
 217                     sizeof (dmu_replay_record_t)) != 0)
 218                         return (SET_ERROR(EINTR));
 219                 dsp->dsa_pending_op = PENDING_NONE;
 220         }
 221         /* write a DATA record */
 222         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 223         dsp->dsa_drr->drr_type = DRR_WRITE;
 224         drrw->drr_object = object;
 225         drrw->drr_type = type;
 226         drrw->drr_offset = offset;
 227         drrw->drr_length = blksz;
 228         drrw->drr_toguid = dsp->dsa_toguid;
 229         if (BP_IS_EMBEDDED(bp)) {
 230                 /*
 231                  * There's no pre-computed checksum of embedded BP's, so
 232                  * (like fletcher4-checkummed blocks) userland will have
 233                  * to compute a dedup-capable checksum itself.
 234                  */
 235                 drrw->drr_checksumtype = ZIO_CHECKSUM_OFF;
 236         } else {
 237                 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
 238                 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
 239                         drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
 240                 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
 241                 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
 242                 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
 243                 drrw->drr_key.ddk_cksum = bp->blk_cksum;
 244         }
 245
 246         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 247                 return (SET_ERROR(EINTR));
 248         if (dump_bytes(dsp, data, blksz) != 0)
 249                 return (SET_ERROR(EINTR));
 250         return (0);
 251 }
 252
 253 static int
 254 dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
 255     int blksz, const blkptr_t *bp)
 256 {
 257         char buf[BPE_PAYLOAD_SIZE];
 258         struct drr_write_embedded *drrw =
 259             &(dsp->dsa_drr->drr_u.drr_write_embedded);
 260
 261         if (dsp->dsa_pending_op != PENDING_NONE) {
 262                 if (dump_bytes(dsp, dsp->dsa_drr,
 263                     sizeof (dmu_replay_record_t)) != 0)
 264                         return (EINTR);
 265                 dsp->dsa_pending_op = PENDING_NONE;
 266         }
 267
 268         ASSERT(BP_IS_EMBEDDED(bp));
 269
 270         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 271         dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED;
 272         drrw->drr_object = object;
 273         drrw->drr_offset = offset;
 274         drrw->drr_length = blksz;
 275         drrw->drr_toguid = dsp->dsa_toguid;
 276         drrw->drr_compression = BP_GET_COMPRESS(bp);
 277         drrw->drr_etype = BPE_GET_ETYPE(bp);
 278         drrw->drr_lsize = BPE_GET_LSIZE(bp);
 279         drrw->drr_psize = BPE_GET_PSIZE(bp);
 280
 281         decode_embedded_bp_compressed(bp, buf);
 282
 283         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 284                 return (EINTR);
 285         if (dump_bytes(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0)
 286                 return (EINTR);
 287         return (0);
 288 }
 289
 290 static int
 291 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
 292 {
 293         struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
 294
 295         if (dsp->dsa_pending_op != PENDING_NONE) {
 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         /* write a SPILL record */
 303         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 304         dsp->dsa_drr->drr_type = DRR_SPILL;
 305         drrs->drr_object = object;
 306         drrs->drr_length = blksz;
 307         drrs->drr_toguid = dsp->dsa_toguid;
 308
 309         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
 310                 return (SET_ERROR(EINTR));
 311         if (dump_bytes(dsp, data, blksz))
 312                 return (SET_ERROR(EINTR));
 313         return (0);
 314 }
 315
 316 static int
 317 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
 318 {
 319         struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
 320
 321         /* See comment in dump_free(). */
 322         if (!dsp->dsa_incremental)
 323                 return (0);
 324
 325         /*
 326          * If there is a pending op, but it's not PENDING_FREEOBJECTS,
 327          * push it out, since free block aggregation can only be done for
 328          * blocks of the same type (i.e., DRR_FREE records can only be
 329          * aggregated with other DRR_FREE records.  DRR_FREEOBJECTS records
 330          * can only be aggregated with other DRR_FREEOBJECTS records.
 331          */
 332         if (dsp->dsa_pending_op != PENDING_NONE &&
 333             dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
 334                 if (dump_bytes(dsp, dsp->dsa_drr,
 335                     sizeof (dmu_replay_record_t)) != 0)
 336                         return (SET_ERROR(EINTR));
 337                 dsp->dsa_pending_op = PENDING_NONE;
 338         }
 339         if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
 340                 /*
 341                  * See whether this free object array can be aggregated
 342                  * with pending one
 343                  */
 344                 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
 345                         drrfo->drr_numobjs += numobjs;
 346                         return (0);
 347                 } else {
 348                         /* can't be aggregated.  Push out pending record */
 349                         if (dump_bytes(dsp, dsp->dsa_drr,
 350                             sizeof (dmu_replay_record_t)) != 0)
 351                                 return (SET_ERROR(EINTR));
 352                         dsp->dsa_pending_op = PENDING_NONE;
 353                 }
 354         }
 355
 356         /* write a FREEOBJECTS record */
 357         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 358         dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
 359         drrfo->drr_firstobj = firstobj;
 360         drrfo->drr_numobjs = numobjs;
 361         drrfo->drr_toguid = dsp->dsa_toguid;
 362
 363         dsp->dsa_pending_op = PENDING_FREEOBJECTS;
 364
 365         return (0);
 366 }
 367
 368 static int
 369 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
 370 {
 371         struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
 372
 373         if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
 374                 return (dump_freeobjects(dsp, object, 1));
 375
 376         if (dsp->dsa_pending_op != PENDING_NONE) {
 377                 if (dump_bytes(dsp, dsp->dsa_drr,
 378                     sizeof (dmu_replay_record_t)) != 0)
 379                         return (SET_ERROR(EINTR));
 380                 dsp->dsa_pending_op = PENDING_NONE;
 381         }
 382
 383         /* write an OBJECT record */
 384         bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
 385         dsp->dsa_drr->drr_type = DRR_OBJECT;
 386         drro->drr_object = object;
 387         drro->drr_type = dnp->dn_type;
 388         drro->drr_bonustype = dnp->dn_bonustype;
 389         drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
 390         drro->drr_bonuslen = dnp->dn_bonuslen;
 391         drro->drr_checksumtype = dnp->dn_checksum;
 392         drro->drr_compress = dnp->dn_compress;
 393         drro->drr_toguid = dsp->dsa_toguid;
 394
 395         if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 396                 return (SET_ERROR(EINTR));
 397
 398         if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
 399                 return (SET_ERROR(EINTR));
 400
 401         /* Free anything past the end of the file. */
 402         if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
 403             (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
 404                 return (SET_ERROR(EINTR));
 405         if (dsp->dsa_err != 0)
 406                 return (SET_ERROR(EINTR));
 407         return (0);
 408 }
 409
 410 static boolean_t
 411 backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp)
 412 {
 413         if (!BP_IS_EMBEDDED(bp))
 414                 return (B_FALSE);
 415
 416         /*
 417          * Compression function must be legacy, or explicitly enabled.
 418          */
 419         if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS &&
 420             !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4)))
 421                 return (B_FALSE);
 422
 423         /*
 424          * Embed type must be explicitly enabled.
 425          */
 426         switch (BPE_GET_ETYPE(bp)) {
 427         case BP_EMBEDDED_TYPE_DATA:
 428                 if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)
 429                         return (B_TRUE);
 430                 break;
 431         default:
 432                 return (B_FALSE);
 433         }
 434         return (B_FALSE);
 435 }
 436
 437 #define BP_SPAN(dnp, level) \
 438         (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
 439         (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
 440
 441 /* ARGSUSED */
 442 static int
 443 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 444     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
 445 {
 446         dmu_sendarg_t *dsp = arg;
 447         dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
 448         int err = 0;
 449
 450         if (issig(JUSTLOOKING) && issig(FORREAL))
 451                 return (SET_ERROR(EINTR));
 452
 453         if (zb->zb_object != DMU_META_DNODE_OBJECT &&
 454             DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
 455                 return (0);
 456         } else if (zb->zb_level == ZB_ZIL_LEVEL) {
 457                 /*
 458                  * If we are sending a non-snapshot (which is allowed on
 459                  * read-only pools), it may have a ZIL, which must be ignored.
 460                  */
 461                 return (0);
 462         } else if (BP_IS_HOLE(bp) &&
 463             zb->zb_object == DMU_META_DNODE_OBJECT) {
 464                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 465                 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
 466                 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
 467         } else if (BP_IS_HOLE(bp)) {
 468                 uint64_t span = BP_SPAN(dnp, zb->zb_level);
 469                 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
 470         } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
 471                 return (0);
 472         } else if (type == DMU_OT_DNODE) {
 473                 dnode_phys_t *blk;
 474                 int i;
 475                 int blksz = BP_GET_LSIZE(bp);
 476                 uint32_t aflags = ARC_WAIT;
 477                 arc_buf_t *abuf;
 478
 479                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 480                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 481                     &aflags, zb) != 0)
 482                         return (SET_ERROR(EIO));
 483
 484                 blk = abuf->b_data;
 485                 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
 486                         uint64_t dnobj = (zb->zb_blkid <<
 487                             (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
 488                         err = dump_dnode(dsp, dnobj, blk+i);
 489                         if (err != 0)
 490                                 break;
 491                 }
 492                 (void) arc_buf_remove_ref(abuf, &abuf);
 493         } else if (type == DMU_OT_SA) {
 494                 uint32_t aflags = ARC_WAIT;
 495                 arc_buf_t *abuf;
 496                 int blksz = BP_GET_LSIZE(bp);
 497
 498                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 499                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 500                     &aflags, zb) != 0)
 501                         return (SET_ERROR(EIO));
 502
 503                 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
 504                 (void) arc_buf_remove_ref(abuf, &abuf);
 505         } else if (backup_do_embed(dsp, bp)) {
 506                 /* it's an embedded level-0 block of a regular object */
 507                 int blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
 508                 err = dump_write_embedded(dsp, zb->zb_object,
 509                     zb->zb_blkid * blksz, blksz, bp);
 510         } else { /* it's a level-0 block of a regular object */
 511                 uint32_t aflags = ARC_WAIT;
 512                 arc_buf_t *abuf;
 513                 int blksz = BP_GET_LSIZE(bp);
 514
 515                 ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
 516                 ASSERT0(zb->zb_level);
 517                 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 518                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 519                     &aflags, zb) != 0) {
 520                         if (zfs_send_corrupt_data) {
 521                                 /* Send a block filled with 0x"zfs badd bloc" */
 522                                 abuf = arc_buf_alloc(spa, blksz, &abuf,
 523                                     ARC_BUFC_DATA);
 524                                 uint64_t *ptr;
 525                                 for (ptr = abuf->b_data;
 526                                     (char *)ptr < (char *)abuf->b_data + blksz;
 527                                     ptr++)
 528                                         *ptr = 0x2f5baddb10c;
 529                         } else {
 530                                 return (SET_ERROR(EIO));
 531                         }
 532                 }
 533
 534                 err = dump_write(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
 535                     blksz, bp, abuf->b_data);
 536                 (void) arc_buf_remove_ref(abuf, &abuf);
 537         }
 538
 539         ASSERT(err == 0 || err == EINTR);
 540         return (err);
 541 }
 542
 543 /*
 544  * Releases dp using the specified tag.
 545  */
 546 static int
 547 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 548     zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok,
 549 #ifdef illumos
 550     int outfd, vnode_t *vp, offset_t *off)
 551 #else
 552     int outfd, struct file *fp, offset_t *off)
 553 #endif
 554 {
 555         objset_t *os;
 556         dmu_replay_record_t *drr;
 557         dmu_sendarg_t *dsp;
 558         int err;
 559         uint64_t fromtxg = 0;
 560         uint64_t featureflags = 0;
 561
 562         err = dmu_objset_from_ds(ds, &os);
 563         if (err != 0) {
 564                 dsl_pool_rele(dp, tag);
 565                 return (err);
 566         }
 567
 568         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
 569         drr->drr_type = DRR_BEGIN;
 570         drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
 571         DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
 572             DMU_SUBSTREAM);
 573
 574 #ifdef _KERNEL
 575         if (dmu_objset_type(os) == DMU_OST_ZFS) {
 576                 uint64_t version;
 577                 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
 578                         kmem_free(drr, sizeof (dmu_replay_record_t));
 579                         dsl_pool_rele(dp, tag);
 580                         return (SET_ERROR(EINVAL));
 581                 }
 582                 if (version >= ZPL_VERSION_SA) {
 583                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
 584                 }
 585         }
 586 #endif
 587
 588         if (embedok &&
 589             spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) {
 590                 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA;
 591                 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
 592                         featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4;
 593         } else {
 594                 embedok = B_FALSE;
 595         }
 596
 597         DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo,
 598             featureflags);
 599
 600         drr->drr_u.drr_begin.drr_creation_time =
 601             ds->ds_phys->ds_creation_time;
 602         drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
 603         if (is_clone)
 604                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
 605         drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
 606         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 607                 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
 608
 609         if (fromzb != NULL) {
 610                 drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid;
 611                 fromtxg = fromzb->zbm_creation_txg;
 612         }
 613         dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
 614         if (!dsl_dataset_is_snapshot(ds)) {
 615                 (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
 616                     sizeof (drr->drr_u.drr_begin.drr_toname));
 617         }
 618
 619         dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
 620
 621         dsp->dsa_drr = drr;
 622         dsp->dsa_outfd = outfd;
 623         dsp->dsa_proc = curproc;
 624         dsp->dsa_td = curthread;
 625         dsp->dsa_fp = fp;
 626         dsp->dsa_os = os;
 627         dsp->dsa_off = off;
 628         dsp->dsa_toguid = ds->ds_phys->ds_guid;
 629         ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
 630         dsp->dsa_pending_op = PENDING_NONE;
 631         dsp->dsa_incremental = (fromzb != NULL);
 632         dsp->dsa_featureflags = featureflags;
 633
 634         mutex_enter(&ds->ds_sendstream_lock);
 635         list_insert_head(&ds->ds_sendstreams, dsp);
 636         mutex_exit(&ds->ds_sendstream_lock);
 637
 638         dsl_dataset_long_hold(ds, FTAG);
 639         dsl_pool_rele(dp, tag);
 640
 641         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 642                 err = dsp->dsa_err;
 643                 goto out;
 644         }
 645
 646         err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
 647             backup_cb, dsp);
 648
 649         if (dsp->dsa_pending_op != PENDING_NONE)
 650                 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
 651                         err = SET_ERROR(EINTR);
 652
 653         if (err != 0) {
 654                 if (err == EINTR && dsp->dsa_err != 0)
 655                         err = dsp->dsa_err;
 656                 goto out;
 657         }
 658
 659         bzero(drr, sizeof (dmu_replay_record_t));
 660         drr->drr_type = DRR_END;
 661         drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
 662         drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
 663
 664         if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
 665                 err = dsp->dsa_err;
 666                 goto out;
 667         }
 668
 669 out:
 670         mutex_enter(&ds->ds_sendstream_lock);
 671         list_remove(&ds->ds_sendstreams, dsp);
 672         mutex_exit(&ds->ds_sendstream_lock);
 673
 674         kmem_free(drr, sizeof (dmu_replay_record_t));
 675         kmem_free(dsp, sizeof (dmu_sendarg_t));
 676
 677         dsl_dataset_long_rele(ds, FTAG);
 678
 679         return (err);
 680 }
 681
 682 int
 683 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
 684 #ifdef illumos
 685     boolean_t embedok, int outfd, vnode_t *vp, offset_t *off)
 686 #else
 687     boolean_t embedok, int outfd, struct file *fp, offset_t *off)
 688 #endif
 689 {
 690         dsl_pool_t *dp;
 691         dsl_dataset_t *ds;
 692         dsl_dataset_t *fromds = NULL;
 693         int err;
 694
 695         err = dsl_pool_hold(pool, FTAG, &dp);
 696         if (err != 0)
 697                 return (err);
 698
 699         err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
 700         if (err != 0) {
 701                 dsl_pool_rele(dp, FTAG);
 702                 return (err);
 703         }
 704
 705         if (fromsnap != 0) {
 706                 zfs_bookmark_phys_t zb;
 707                 boolean_t is_clone;
 708
 709                 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
 710                 if (err != 0) {
 711                         dsl_dataset_rele(ds, FTAG);
 712                         dsl_pool_rele(dp, FTAG);
 713                         return (err);
 714                 }
 715                 if (!dsl_dataset_is_before(ds, fromds, 0))
 716                         err = SET_ERROR(EXDEV);
 717                 zb.zbm_creation_time = fromds->ds_phys->ds_creation_time;
 718                 zb.zbm_creation_txg = fromds->ds_phys->ds_creation_txg;
 719                 zb.zbm_guid = fromds->ds_phys->ds_guid;
 720                 is_clone = (fromds->ds_dir != ds->ds_dir);
 721                 dsl_dataset_rele(fromds, FTAG);
 722                 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok,
 723                     outfd, fp, off);
 724         } else {
 725                 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok,
 726                     outfd, fp, off);
 727         }
 728         dsl_dataset_rele(ds, FTAG);
 729         return (err);
 730 }
 731
 732 int
 733 dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok,
 734 #ifdef illumos
 735     int outfd, vnode_t *vp, offset_t *off)
 736 #else
 737     int outfd, struct file *fp, offset_t *off)
 738 #endif
 739 {
 740         dsl_pool_t *dp;
 741         dsl_dataset_t *ds;
 742         int err;
 743         boolean_t owned = B_FALSE;
 744
 745         if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
 746                 return (SET_ERROR(EINVAL));
 747
 748         err = dsl_pool_hold(tosnap, FTAG, &dp);
 749         if (err != 0)
 750                 return (err);
 751
 752         if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) {
 753                 /*
 754                  * We are sending a filesystem or volume.  Ensure
 755                  * that it doesn't change by owning the dataset.
 756                  */
 757                 err = dsl_dataset_own(dp, tosnap, FTAG, &ds);
 758                 owned = B_TRUE;
 759         } else {
 760                 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
 761         }
 762         if (err != 0) {
 763                 dsl_pool_rele(dp, FTAG);
 764                 return (err);
 765         }
 766
 767         if (fromsnap != NULL) {
 768                 zfs_bookmark_phys_t zb;
 769                 boolean_t is_clone = B_FALSE;
 770                 int fsnamelen = strchr(tosnap, '@') - tosnap;
 771
 772                 /*
 773                  * If the fromsnap is in a different filesystem, then
 774                  * mark the send stream as a clone.
 775                  */
 776                 if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
 777                     (fromsnap[fsnamelen] != '@' &&
 778                     fromsnap[fsnamelen] != '#')) {
 779                         is_clone = B_TRUE;
 780                 }
 781
 782                 if (strchr(fromsnap, '@')) {
 783                         dsl_dataset_t *fromds;
 784                         err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
 785                         if (err == 0) {
 786                                 if (!dsl_dataset_is_before(ds, fromds, 0))
 787                                         err = SET_ERROR(EXDEV);
 788                                 zb.zbm_creation_time =
 789                                     fromds->ds_phys->ds_creation_time;
 790                                 zb.zbm_creation_txg =
 791                                     fromds->ds_phys->ds_creation_txg;
 792                                 zb.zbm_guid = fromds->ds_phys->ds_guid;
 793                                 is_clone = (ds->ds_dir != fromds->ds_dir);
 794                                 dsl_dataset_rele(fromds, FTAG);
 795                         }
 796                 } else {
 797                         err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb);
 798                 }
 799                 if (err != 0) {
 800                         dsl_dataset_rele(ds, FTAG);
 801                         dsl_pool_rele(dp, FTAG);
 802                         return (err);
 803                 }
 804                 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok,
 805                     outfd, fp, off);
 806         } else {
 807                 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok,
 808                     outfd, fp, off);
 809         }
 810         if (owned)
 811                 dsl_dataset_disown(ds, FTAG);
 812         else
 813                 dsl_dataset_rele(ds, FTAG);
 814         return (err);
 815 }
 816
 817 int
 818 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
 819 {
 820         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 821         int err;
 822         uint64_t size;
 823
 824         ASSERT(dsl_pool_config_held(dp));
 825
 826         /* tosnap must be a snapshot */
 827         if (!dsl_dataset_is_snapshot(ds))
 828                 return (SET_ERROR(EINVAL));
 829
 830         /*
 831          * fromsnap must be an earlier snapshot from the same fs as tosnap,
 832          * or the origin's fs.
 833          */
 834         if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
 835                 return (SET_ERROR(EXDEV));
 836
 837         /* Get uncompressed size estimate of changed data. */
 838         if (fromds == NULL) {
 839                 size = ds->ds_phys->ds_uncompressed_bytes;
 840         } else {
 841                 uint64_t used, comp;
 842                 err = dsl_dataset_space_written(fromds, ds,
 843                     &used, &comp, &size);
 844                 if (err != 0)
 845                         return (err);
 846         }
 847
 848         /*
 849          * Assume that space (both on-disk and in-stream) is dominated by
 850          * data.  We will adjust for indirect blocks and the copies property,
 851          * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
 852          */
 853
 854         /*
 855          * Subtract out approximate space used by indirect blocks.
 856          * Assume most space is used by data blocks (non-indirect, non-dnode).
 857          * Assume all blocks are recordsize.  Assume ditto blocks and
 858          * internal fragmentation counter out compression.
 859          *
 860          * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
 861          * block, which we observe in practice.
 862          */
 863         uint64_t recordsize;
 864         err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
 865         if (err != 0)
 866                 return (err);
 867         size -= size / recordsize * sizeof (blkptr_t);
 868
 869         /* Add in the space for the record associated with each block. */
 870         size += size / recordsize * sizeof (dmu_replay_record_t);
 871
 872         *sizep = size;
 873
 874         return (0);
 875 }
 876
 877 typedef struct dmu_recv_begin_arg {
 878         const char *drba_origin;
 879         dmu_recv_cookie_t *drba_cookie;
 880         cred_t *drba_cred;
 881         uint64_t drba_snapobj;
 882 } dmu_recv_begin_arg_t;
 883
 884 static int
 885 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
 886     uint64_t fromguid)
 887 {
 888         uint64_t val;
 889         int error;
 890         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 891
 892         /* temporary clone name must not exist */
 893         error = zap_lookup(dp->dp_meta_objset,
 894             ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
 895             8, 1, &val);
 896         if (error != ENOENT)
 897                 return (error == 0 ? EBUSY : error);
 898
 899         /* new snapshot name must not exist */
 900         error = zap_lookup(dp->dp_meta_objset,
 901             ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
 902             8, 1, &val);
 903         if (error != ENOENT)
 904                 return (error == 0 ? EEXIST : error);
 905
 906         /*
 907          * Check snapshot limit before receiving. We'll recheck again at the
 908          * end, but might as well abort before receiving if we're already over
 909          * the limit.
 910          *
 911          * Note that we do not check the file system limit with
 912          * dsl_dir_fscount_check because the temporary %clones don't count
 913          * against that limit.
 914          */
 915         error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT,
 916             NULL, drba->drba_cred);
 917         if (error != 0)
 918                 return (error);
 919
 920         if (fromguid != 0) {
 921                 dsl_dataset_t *snap;
 922                 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
 923
 924                 /* Find snapshot in this dir that matches fromguid. */
 925                 while (obj != 0) {
 926                         error = dsl_dataset_hold_obj(dp, obj, FTAG,
 927                             &snap);
 928                         if (error != 0)
 929                                 return (SET_ERROR(ENODEV));
 930                         if (snap->ds_dir != ds->ds_dir) {
 931                                 dsl_dataset_rele(snap, FTAG);
 932                                 return (SET_ERROR(ENODEV));
 933                         }
 934                         if (snap->ds_phys->ds_guid == fromguid)
 935                                 break;
 936                         obj = snap->ds_phys->ds_prev_snap_obj;
 937                         dsl_dataset_rele(snap, FTAG);
 938                 }
 939                 if (obj == 0)
 940                         return (SET_ERROR(ENODEV));
 941
 942                 if (drba->drba_cookie->drc_force) {
 943                         drba->drba_snapobj = obj;
 944                 } else {
 945                         /*
 946                          * If we are not forcing, there must be no
 947                          * changes since fromsnap.
 948                          */
 949                         if (dsl_dataset_modified_since_snap(ds, snap)) {
 950                                 dsl_dataset_rele(snap, FTAG);
 951                                 return (SET_ERROR(ETXTBSY));
 952                         }
 953                         drba->drba_snapobj = ds->ds_prev->ds_object;
 954                 }
 955
 956                 dsl_dataset_rele(snap, FTAG);
 957         } else {
 958                 /* if full, most recent snapshot must be $ORIGIN */
 959                 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
 960                         return (SET_ERROR(ENODEV));
 961                 drba->drba_snapobj = ds->ds_phys->ds_prev_snap_obj;
 962         }
 963
 964         return (0);
 965
 966 }
 967
 968 static int
 969 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 970 {
 971         dmu_recv_begin_arg_t *drba = arg;
 972         dsl_pool_t *dp = dmu_tx_pool(tx);
 973         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 974         uint64_t fromguid = drrb->drr_fromguid;
 975         int flags = drrb->drr_flags;
 976         int error;
 977         uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 978         dsl_dataset_t *ds;
 979         const char *tofs = drba->drba_cookie->drc_tofs;
 980
 981         /* already checked */
 982         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 983
 984         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 985             DMU_COMPOUNDSTREAM ||
 986             drrb->drr_type >= DMU_OST_NUMTYPES ||
 987             ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
 988                 return (SET_ERROR(EINVAL));
 989
 990         /* Verify pool version supports SA if SA_SPILL feature set */
 991         if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
 992             spa_version(dp->dp_spa) < SPA_VERSION_SA)
 993                 return (SET_ERROR(ENOTSUP));
 994
 995         /*
 996          * The receiving code doesn't know how to translate a WRITE_EMBEDDED
 997          * record to a plan WRITE record, so the pool must have the
 998          * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
 999          * records.  Same with WRITE_EMBEDDED records that use LZ4 compression.
1000          */
1001         if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) &&
1002             !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA))
1003                 return (SET_ERROR(ENOTSUP));
1004         if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) &&
1005             !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
1006                 return (SET_ERROR(ENOTSUP));
1007
1008         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1009         if (error == 0) {
1010                 /* target fs already exists; recv into temp clone */
1011
1012                 /* Can't recv a clone into an existing fs */
1013                 if (flags & DRR_FLAG_CLONE) {
1014                         dsl_dataset_rele(ds, FTAG);
1015                         return (SET_ERROR(EINVAL));
1016                 }
1017
1018                 error = recv_begin_check_existing_impl(drba, ds, fromguid);
1019                 dsl_dataset_rele(ds, FTAG);
1020         } else if (error == ENOENT) {
1021                 /* target fs does not exist; must be a full backup or clone */
1022                 char buf[MAXNAMELEN];
1023
1024                 /*
1025                  * If it's a non-clone incremental, we are missing the
1026                  * target fs, so fail the recv.
1027                  */
1028                 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
1029                         return (SET_ERROR(ENOENT));
1030
1031                 /* Open the parent of tofs */
1032                 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
1033                 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
1034                 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
1035                 if (error != 0)
1036                         return (error);
1037
1038                 /*
1039                  * Check filesystem and snapshot limits before receiving. We'll
1040                  * recheck snapshot limits again at the end (we create the
1041                  * filesystems and increment those counts during begin_sync).
1042                  */
1043                 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1044                     ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred);
1045                 if (error != 0) {
1046                         dsl_dataset_rele(ds, FTAG);
1047                         return (error);
1048                 }
1049
1050                 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1051                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred);
1052                 if (error != 0) {
1053                         dsl_dataset_rele(ds, FTAG);
1054                         return (error);
1055                 }
1056
1057                 if (drba->drba_origin != NULL) {
1058                         dsl_dataset_t *origin;
1059                         error = dsl_dataset_hold(dp, drba->drba_origin,
1060                             FTAG, &origin);
1061                         if (error != 0) {
1062                                 dsl_dataset_rele(ds, FTAG);
1063                                 return (error);
1064                         }
1065                         if (!dsl_dataset_is_snapshot(origin)) {
1066                                 dsl_dataset_rele(origin, FTAG);
1067                                 dsl_dataset_rele(ds, FTAG);
1068                                 return (SET_ERROR(EINVAL));
1069                         }
1070                         if (origin->ds_phys->ds_guid != fromguid) {
1071                                 dsl_dataset_rele(origin, FTAG);
1072                                 dsl_dataset_rele(ds, FTAG);
1073                                 return (SET_ERROR(ENODEV));
1074                         }
1075                         dsl_dataset_rele(origin, FTAG);
1076                 }
1077                 dsl_dataset_rele(ds, FTAG);
1078                 error = 0;
1079         }
1080         return (error);
1081 }
1082
1083 static void
1084 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
1085 {
1086         dmu_recv_begin_arg_t *drba = arg;
1087         dsl_pool_t *dp = dmu_tx_pool(tx);
1088         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1089         const char *tofs = drba->drba_cookie->drc_tofs;
1090         dsl_dataset_t *ds, *newds;
1091         uint64_t dsobj;
1092         int error;
1093         uint64_t crflags;
1094
1095         crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
1096             DS_FLAG_CI_DATASET : 0;
1097
1098         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1099         if (error == 0) {
1100                 /* create temporary clone */
1101                 dsl_dataset_t *snap = NULL;
1102                 if (drba->drba_snapobj != 0) {
1103                         VERIFY0(dsl_dataset_hold_obj(dp,
1104                             drba->drba_snapobj, FTAG, &snap));
1105                 }
1106                 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
1107                     snap, crflags, drba->drba_cred, tx);
1108                 dsl_dataset_rele(snap, FTAG);
1109                 dsl_dataset_rele(ds, FTAG);
1110         } else {
1111                 dsl_dir_t *dd;
1112                 const char *tail;
1113                 dsl_dataset_t *origin = NULL;
1114
1115                 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1116
1117                 if (drba->drba_origin != NULL) {
1118                         VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1119                             FTAG, &origin));
1120                 }
1121
1122                 /* Create new dataset. */
1123                 dsobj = dsl_dataset_create_sync(dd,
1124                     strrchr(tofs, '/') + 1,
1125                     origin, crflags, drba->drba_cred, tx);
1126                 if (origin != NULL)
1127                         dsl_dataset_rele(origin, FTAG);
1128                 dsl_dir_rele(dd, FTAG);
1129                 drba->drba_cookie->drc_newfs = B_TRUE;
1130         }
1131         VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1132
1133         dmu_buf_will_dirty(newds->ds_dbuf, tx);
1134         newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1135
1136         /*
1137          * If we actually created a non-clone, we need to create the
1138          * objset in our new dataset.
1139          */
1140         if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1141                 (void) dmu_objset_create_impl(dp->dp_spa,
1142                     newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1143         }
1144
1145         drba->drba_cookie->drc_ds = newds;
1146
1147         spa_history_log_internal_ds(newds, "receive", tx, "");
1148 }
1149
1150 /*
1151  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1152  * succeeds; otherwise we will leak the holds on the datasets.
1153  */
1154 int
1155 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1156     boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1157 {
1158         dmu_recv_begin_arg_t drba = { 0 };
1159         dmu_replay_record_t *drr;
1160
1161         bzero(drc, sizeof (dmu_recv_cookie_t));
1162         drc->drc_drrb = drrb;
1163         drc->drc_tosnap = tosnap;
1164         drc->drc_tofs = tofs;
1165         drc->drc_force = force;
1166         drc->drc_cred = CRED();
1167
1168         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1169                 drc->drc_byteswap = B_TRUE;
1170         else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1171                 return (SET_ERROR(EINVAL));
1172
1173         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1174         drr->drr_type = DRR_BEGIN;
1175         drr->drr_u.drr_begin = *drc->drc_drrb;
1176         if (drc->drc_byteswap) {
1177                 fletcher_4_incremental_byteswap(drr,
1178                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1179         } else {
1180                 fletcher_4_incremental_native(drr,
1181                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1182         }
1183         kmem_free(drr, sizeof (dmu_replay_record_t));
1184
1185         if (drc->drc_byteswap) {
1186                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1187                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1188                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1189                 drrb->drr_type = BSWAP_32(drrb->drr_type);
1190                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1191                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1192         }
1193
1194         drba.drba_origin = origin;
1195         drba.drba_cookie = drc;
1196         drba.drba_cred = CRED();
1197
1198         return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1199             &drba, 5, ZFS_SPACE_CHECK_NORMAL));
1200 }
1201
1202 struct restorearg {
1203         int err;
1204         boolean_t byteswap;
1205         kthread_t *td;
1206         struct file *fp;
1207         char *buf;
1208         uint64_t voff;
1209         int bufsize; /* amount of memory allocated for buf */
1210         zio_cksum_t cksum;
1211         avl_tree_t *guid_to_ds_map;
1212 };
1213
1214 typedef struct guid_map_entry {
1215         uint64_t        guid;
1216         dsl_dataset_t   *gme_ds;
1217         avl_node_t      avlnode;
1218 } guid_map_entry_t;
1219
1220 static int
1221 guid_compare(const void *arg1, const void *arg2)
1222 {
1223         const guid_map_entry_t *gmep1 = arg1;
1224         const guid_map_entry_t *gmep2 = arg2;
1225
1226         if (gmep1->guid < gmep2->guid)
1227                 return (-1);
1228         else if (gmep1->guid > gmep2->guid)
1229                 return (1);
1230         return (0);
1231 }
1232
1233 static void
1234 free_guid_map_onexit(void *arg)
1235 {
1236         avl_tree_t *ca = arg;
1237         void *cookie = NULL;
1238         guid_map_entry_t *gmep;
1239
1240         while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1241                 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1242                 dsl_dataset_rele(gmep->gme_ds, gmep);
1243                 kmem_free(gmep, sizeof (guid_map_entry_t));
1244         }
1245         avl_destroy(ca);
1246         kmem_free(ca, sizeof (avl_tree_t));
1247 }
1248
1249 static int
1250 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
1251 {
1252         struct uio auio;
1253         struct iovec aiov;
1254         int error;
1255
1256         aiov.iov_base = buf;
1257         aiov.iov_len = len;
1258         auio.uio_iov = &aiov;
1259         auio.uio_iovcnt = 1;
1260         auio.uio_resid = len;
1261         auio.uio_segflg = UIO_SYSSPACE;
1262         auio.uio_rw = UIO_READ;
1263         auio.uio_offset = off;
1264         auio.uio_td = ra->td;
1265 #ifdef _KERNEL
1266         error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
1267 #else
1268         fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
1269         error = EOPNOTSUPP;
1270 #endif
1271         *resid = auio.uio_resid;
1272         return (error);
1273 }
1274
1275 static void *
1276 restore_read(struct restorearg *ra, int len)
1277 {
1278         void *rv;
1279         int done = 0;
1280
1281         /* some things will require 8-byte alignment, so everything must */
1282         ASSERT0(len % 8);
1283
1284         while (done < len) {
1285                 ssize_t resid;
1286
1287                 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
1288                     len - done, ra->voff, &resid);
1289
1290                 if (resid == len - done)
1291                         ra->err = SET_ERROR(EINVAL);
1292                 ra->voff += len - done - resid;
1293                 done = len - resid;
1294                 if (ra->err != 0)
1295                         return (NULL);
1296         }
1297
1298         ASSERT3U(done, ==, len);
1299         rv = ra->buf;
1300         if (ra->byteswap)
1301                 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1302         else
1303                 fletcher_4_incremental_native(rv, len, &ra->cksum);
1304         return (rv);
1305 }
1306
1307 static void
1308 backup_byteswap(dmu_replay_record_t *drr)
1309 {
1310 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1311 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1312         drr->drr_type = BSWAP_32(drr->drr_type);
1313         drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1314         switch (drr->drr_type) {
1315         case DRR_BEGIN:
1316                 DO64(drr_begin.drr_magic);
1317                 DO64(drr_begin.drr_versioninfo);
1318                 DO64(drr_begin.drr_creation_time);
1319                 DO32(drr_begin.drr_type);
1320                 DO32(drr_begin.drr_flags);
1321                 DO64(drr_begin.drr_toguid);
1322                 DO64(drr_begin.drr_fromguid);
1323                 break;
1324         case DRR_OBJECT:
1325                 DO64(drr_object.drr_object);
1326                 DO32(drr_object.drr_type);
1327                 DO32(drr_object.drr_bonustype);
1328                 DO32(drr_object.drr_blksz);
1329                 DO32(drr_object.drr_bonuslen);
1330                 DO64(drr_object.drr_toguid);
1331                 break;
1332         case DRR_FREEOBJECTS:
1333                 DO64(drr_freeobjects.drr_firstobj);
1334                 DO64(drr_freeobjects.drr_numobjs);
1335                 DO64(drr_freeobjects.drr_toguid);
1336                 break;
1337         case DRR_WRITE:
1338                 DO64(drr_write.drr_object);
1339                 DO32(drr_write.drr_type);
1340                 DO64(drr_write.drr_offset);
1341                 DO64(drr_write.drr_length);
1342                 DO64(drr_write.drr_toguid);
1343                 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1344                 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1345                 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1346                 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1347                 DO64(drr_write.drr_key.ddk_prop);
1348                 break;
1349         case DRR_WRITE_BYREF:
1350                 DO64(drr_write_byref.drr_object);
1351                 DO64(drr_write_byref.drr_offset);
1352                 DO64(drr_write_byref.drr_length);
1353                 DO64(drr_write_byref.drr_toguid);
1354                 DO64(drr_write_byref.drr_refguid);
1355                 DO64(drr_write_byref.drr_refobject);
1356                 DO64(drr_write_byref.drr_refoffset);
1357                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1358                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1359                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1360                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1361                 DO64(drr_write_byref.drr_key.ddk_prop);
1362                 break;
1363         case DRR_WRITE_EMBEDDED:
1364                 DO64(drr_write_embedded.drr_object);
1365                 DO64(drr_write_embedded.drr_offset);
1366                 DO64(drr_write_embedded.drr_length);
1367                 DO64(drr_write_embedded.drr_toguid);
1368                 DO32(drr_write_embedded.drr_lsize);
1369                 DO32(drr_write_embedded.drr_psize);
1370                 break;
1371         case DRR_FREE:
1372                 DO64(drr_free.drr_object);
1373                 DO64(drr_free.drr_offset);
1374                 DO64(drr_free.drr_length);
1375                 DO64(drr_free.drr_toguid);
1376                 break;
1377         case DRR_SPILL:
1378                 DO64(drr_spill.drr_object);
1379                 DO64(drr_spill.drr_length);
1380                 DO64(drr_spill.drr_toguid);
1381                 break;
1382         case DRR_END:
1383                 DO64(drr_end.drr_checksum.zc_word[0]);
1384                 DO64(drr_end.drr_checksum.zc_word[1]);
1385                 DO64(drr_end.drr_checksum.zc_word[2]);
1386                 DO64(drr_end.drr_checksum.zc_word[3]);
1387                 DO64(drr_end.drr_toguid);
1388                 break;
1389         }
1390 #undef DO64
1391 #undef DO32
1392 }
1393
1394 static int
1395 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1396 {
1397         int err;
1398         dmu_tx_t *tx;
1399         void *data = NULL;
1400
1401         if (drro->drr_type == DMU_OT_NONE ||
1402             !DMU_OT_IS_VALID(drro->drr_type) ||
1403             !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1404             drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1405             drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1406             P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1407             drro->drr_blksz < SPA_MINBLOCKSIZE ||
1408             drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1409             drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1410                 return (SET_ERROR(EINVAL));
1411         }
1412
1413         err = dmu_object_info(os, drro->drr_object, NULL);
1414
1415         if (err != 0 && err != ENOENT)
1416                 return (SET_ERROR(EINVAL));
1417
1418         if (drro->drr_bonuslen) {
1419                 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1420                 if (ra->err != 0)
1421                         return (ra->err);
1422         }
1423
1424         if (err == ENOENT) {
1425                 /* currently free, want to be allocated */
1426                 tx = dmu_tx_create(os);
1427                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1428                 err = dmu_tx_assign(tx, TXG_WAIT);
1429                 if (err != 0) {
1430                         dmu_tx_abort(tx);
1431                         return (err);
1432                 }
1433                 err = dmu_object_claim(os, drro->drr_object,
1434                     drro->drr_type, drro->drr_blksz,
1435                     drro->drr_bonustype, drro->drr_bonuslen, tx);
1436                 dmu_tx_commit(tx);
1437         } else {
1438                 /* currently allocated, want to be allocated */
1439                 err = dmu_object_reclaim(os, drro->drr_object,
1440                     drro->drr_type, drro->drr_blksz,
1441                     drro->drr_bonustype, drro->drr_bonuslen);
1442         }
1443         if (err != 0) {
1444                 return (SET_ERROR(EINVAL));
1445         }
1446
1447         tx = dmu_tx_create(os);
1448         dmu_tx_hold_bonus(tx, drro->drr_object);
1449         err = dmu_tx_assign(tx, TXG_WAIT);
1450         if (err != 0) {
1451                 dmu_tx_abort(tx);
1452                 return (err);
1453         }
1454
1455         dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1456             tx);
1457         dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1458
1459         if (data != NULL) {
1460                 dmu_buf_t *db;
1461
1462                 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1463                 dmu_buf_will_dirty(db, tx);
1464
1465                 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1466                 bcopy(data, db->db_data, drro->drr_bonuslen);
1467                 if (ra->byteswap) {
1468                         dmu_object_byteswap_t byteswap =
1469                             DMU_OT_BYTESWAP(drro->drr_bonustype);
1470                         dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1471                             drro->drr_bonuslen);
1472                 }
1473                 dmu_buf_rele(db, FTAG);
1474         }
1475         dmu_tx_commit(tx);
1476         return (0);
1477 }
1478
1479 /* ARGSUSED */
1480 static int
1481 restore_freeobjects(struct restorearg *ra, objset_t *os,
1482     struct drr_freeobjects *drrfo)
1483 {
1484         uint64_t obj;
1485
1486         if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1487                 return (SET_ERROR(EINVAL));
1488
1489         for (obj = drrfo->drr_firstobj;
1490             obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1491             (void) dmu_object_next(os, &obj, FALSE, 0)) {
1492                 int err;
1493
1494                 if (dmu_object_info(os, obj, NULL) != 0)
1495                         continue;
1496
1497                 err = dmu_free_long_object(os, obj);
1498                 if (err != 0)
1499                         return (err);
1500         }
1501         return (0);
1502 }
1503
1504 static int
1505 restore_write(struct restorearg *ra, objset_t *os,
1506     struct drr_write *drrw)
1507 {
1508         dmu_tx_t *tx;
1509         void *data;
1510         int err;
1511
1512         if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1513             !DMU_OT_IS_VALID(drrw->drr_type))
1514                 return (SET_ERROR(EINVAL));
1515
1516         data = restore_read(ra, drrw->drr_length);
1517         if (data == NULL)
1518                 return (ra->err);
1519
1520         if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1521                 return (SET_ERROR(EINVAL));
1522
1523         tx = dmu_tx_create(os);
1524
1525         dmu_tx_hold_write(tx, drrw->drr_object,
1526             drrw->drr_offset, drrw->drr_length);
1527         err = dmu_tx_assign(tx, TXG_WAIT);
1528         if (err != 0) {
1529                 dmu_tx_abort(tx);
1530                 return (err);
1531         }
1532         if (ra->byteswap) {
1533                 dmu_object_byteswap_t byteswap =
1534                     DMU_OT_BYTESWAP(drrw->drr_type);
1535                 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1536         }
1537         dmu_write(os, drrw->drr_object,
1538             drrw->drr_offset, drrw->drr_length, data, tx);
1539         dmu_tx_commit(tx);
1540         return (0);
1541 }
1542
1543 /*
1544  * Handle a DRR_WRITE_BYREF record.  This record is used in dedup'ed
1545  * streams to refer to a copy of the data that is already on the
1546  * system because it came in earlier in the stream.  This function
1547  * finds the earlier copy of the data, and uses that copy instead of
1548  * data from the stream to fulfill this write.
1549  */
1550 static int
1551 restore_write_byref(struct restorearg *ra, objset_t *os,
1552     struct drr_write_byref *drrwbr)
1553 {
1554         dmu_tx_t *tx;
1555         int err;
1556         guid_map_entry_t gmesrch;
1557         guid_map_entry_t *gmep;
1558         avl_index_t where;
1559         objset_t *ref_os = NULL;
1560         dmu_buf_t *dbp;
1561
1562         if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1563                 return (SET_ERROR(EINVAL));
1564
1565         /*
1566          * If the GUID of the referenced dataset is different from the
1567          * GUID of the target dataset, find the referenced dataset.
1568          */
1569         if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1570                 gmesrch.guid = drrwbr->drr_refguid;
1571                 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1572                     &where)) == NULL) {
1573                         return (SET_ERROR(EINVAL));
1574                 }
1575                 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1576                         return (SET_ERROR(EINVAL));
1577         } else {
1578                 ref_os = os;
1579         }
1580
1581         err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1582             drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1583         if (err != 0)
1584                 return (err);
1585
1586         tx = dmu_tx_create(os);
1587
1588         dmu_tx_hold_write(tx, drrwbr->drr_object,
1589             drrwbr->drr_offset, drrwbr->drr_length);
1590         err = dmu_tx_assign(tx, TXG_WAIT);
1591         if (err != 0) {
1592                 dmu_tx_abort(tx);
1593                 return (err);
1594         }
1595         dmu_write(os, drrwbr->drr_object,
1596             drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1597         dmu_buf_rele(dbp, FTAG);
1598         dmu_tx_commit(tx);
1599         return (0);
1600 }
1601
1602 static int
1603 restore_write_embedded(struct restorearg *ra, objset_t *os,
1604     struct drr_write_embedded *drrwnp)
1605 {
1606         dmu_tx_t *tx;
1607         int err;
1608         void *data;
1609
1610         if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset)
1611                 return (EINVAL);
1612
1613         if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE)
1614                 return (EINVAL);
1615
1616         if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES)
1617                 return (EINVAL);
1618         if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS)
1619                 return (EINVAL);
1620
1621         data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8));
1622         if (data == NULL)
1623                 return (ra->err);
1624
1625         tx = dmu_tx_create(os);
1626
1627         dmu_tx_hold_write(tx, drrwnp->drr_object,
1628             drrwnp->drr_offset, drrwnp->drr_length);
1629         err = dmu_tx_assign(tx, TXG_WAIT);
1630         if (err != 0) {
1631                 dmu_tx_abort(tx);
1632                 return (err);
1633         }
1634
1635         dmu_write_embedded(os, drrwnp->drr_object,
1636             drrwnp->drr_offset, data, drrwnp->drr_etype,
1637             drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize,
1638             ra->byteswap ^ ZFS_HOST_BYTEORDER, tx);
1639
1640         dmu_tx_commit(tx);
1641         return (0);
1642 }
1643
1644 static int
1645 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1646 {
1647         dmu_tx_t *tx;
1648         void *data;
1649         dmu_buf_t *db, *db_spill;
1650         int err;
1651
1652         if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1653             drrs->drr_length > SPA_MAXBLOCKSIZE)
1654                 return (SET_ERROR(EINVAL));
1655
1656         data = restore_read(ra, drrs->drr_length);
1657         if (data == NULL)
1658                 return (ra->err);
1659
1660         if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1661                 return (SET_ERROR(EINVAL));
1662
1663         VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1664         if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1665                 dmu_buf_rele(db, FTAG);
1666                 return (err);
1667         }
1668
1669         tx = dmu_tx_create(os);
1670
1671         dmu_tx_hold_spill(tx, db->db_object);
1672
1673         err = dmu_tx_assign(tx, TXG_WAIT);
1674         if (err != 0) {
1675                 dmu_buf_rele(db, FTAG);
1676                 dmu_buf_rele(db_spill, FTAG);
1677                 dmu_tx_abort(tx);
1678                 return (err);
1679         }
1680         dmu_buf_will_dirty(db_spill, tx);
1681
1682         if (db_spill->db_size < drrs->drr_length)
1683                 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1684                     drrs->drr_length, tx));
1685         bcopy(data, db_spill->db_data, drrs->drr_length);
1686
1687         dmu_buf_rele(db, FTAG);
1688         dmu_buf_rele(db_spill, FTAG);
1689
1690         dmu_tx_commit(tx);
1691         return (0);
1692 }
1693
1694 /* ARGSUSED */
1695 static int
1696 restore_free(struct restorearg *ra, objset_t *os,
1697     struct drr_free *drrf)
1698 {
1699         int err;
1700
1701         if (drrf->drr_length != -1ULL &&
1702             drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1703                 return (SET_ERROR(EINVAL));
1704
1705         if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1706                 return (SET_ERROR(EINVAL));
1707
1708         err = dmu_free_long_range(os, drrf->drr_object,
1709             drrf->drr_offset, drrf->drr_length);
1710         return (err);
1711 }
1712
1713 /* used to destroy the drc_ds on error */
1714 static void
1715 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1716 {
1717         char name[MAXNAMELEN];
1718         dsl_dataset_name(drc->drc_ds, name);
1719         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1720         (void) dsl_destroy_head(name);
1721 }
1722
1723 /*
1724  * NB: callers *must* call dmu_recv_end() if this succeeds.
1725  */
1726 int
1727 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1728     int cleanup_fd, uint64_t *action_handlep)
1729 {
1730         struct restorearg ra = { 0 };
1731         dmu_replay_record_t *drr;
1732         objset_t *os;
1733         zio_cksum_t pcksum;
1734         int featureflags;
1735
1736         ra.byteswap = drc->drc_byteswap;
1737         ra.cksum = drc->drc_cksum;
1738         ra.td = curthread;
1739         ra.fp = fp;
1740         ra.voff = *voffp;
1741         ra.bufsize = 1<<20;
1742         ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1743
1744         /* these were verified in dmu_recv_begin */
1745         ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1746             DMU_SUBSTREAM);
1747         ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1748
1749         /*
1750          * Open the objset we are modifying.
1751          */
1752         VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1753
1754         ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1755
1756         featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1757
1758         /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1759         if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1760                 minor_t minor;
1761
1762                 if (cleanup_fd == -1) {
1763                         ra.err = SET_ERROR(EBADF);
1764                         goto out;
1765                 }
1766                 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1767                 if (ra.err != 0) {
1768                         cleanup_fd = -1;
1769                         goto out;
1770                 }
1771
1772                 if (*action_handlep == 0) {
1773                         ra.guid_to_ds_map =
1774                             kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1775                         avl_create(ra.guid_to_ds_map, guid_compare,
1776                             sizeof (guid_map_entry_t),
1777                             offsetof(guid_map_entry_t, avlnode));
1778                         ra.err = zfs_onexit_add_cb(minor,
1779                             free_guid_map_onexit, ra.guid_to_ds_map,
1780                             action_handlep);
1781                         if (ra.err != 0)
1782                                 goto out;
1783                 } else {
1784                         ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1785                             (void **)&ra.guid_to_ds_map);
1786                         if (ra.err != 0)
1787                                 goto out;
1788                 }
1789
1790                 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1791         }
1792
1793         /*
1794          * Read records and process them.
1795          */
1796         pcksum = ra.cksum;
1797         while (ra.err == 0 &&
1798             NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1799                 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1800                         ra.err = SET_ERROR(EINTR);
1801                         goto out;
1802                 }
1803
1804                 if (ra.byteswap)
1805                         backup_byteswap(drr);
1806
1807                 switch (drr->drr_type) {
1808                 case DRR_OBJECT:
1809                 {
1810                         /*
1811                          * We need to make a copy of the record header,
1812                          * because restore_{object,write} may need to
1813                          * restore_read(), which will invalidate drr.
1814                          */
1815                         struct drr_object drro = drr->drr_u.drr_object;
1816                         ra.err = restore_object(&ra, os, &drro);
1817                         break;
1818                 }
1819                 case DRR_FREEOBJECTS:
1820                 {
1821                         struct drr_freeobjects drrfo =
1822                             drr->drr_u.drr_freeobjects;
1823                         ra.err = restore_freeobjects(&ra, os, &drrfo);
1824                         break;
1825                 }
1826                 case DRR_WRITE:
1827                 {
1828                         struct drr_write drrw = drr->drr_u.drr_write;
1829                         ra.err = restore_write(&ra, os, &drrw);
1830                         break;
1831                 }
1832                 case DRR_WRITE_BYREF:
1833                 {
1834                         struct drr_write_byref drrwbr =
1835                             drr->drr_u.drr_write_byref;
1836                         ra.err = restore_write_byref(&ra, os, &drrwbr);
1837                         break;
1838                 }
1839                 case DRR_WRITE_EMBEDDED:
1840                 {
1841                         struct drr_write_embedded drrwe =
1842                             drr->drr_u.drr_write_embedded;
1843                         ra.err = restore_write_embedded(&ra, os, &drrwe);
1844                         break;
1845                 }
1846                 case DRR_FREE:
1847                 {
1848                         struct drr_free drrf = drr->drr_u.drr_free;
1849                         ra.err = restore_free(&ra, os, &drrf);
1850                         break;
1851                 }
1852                 case DRR_END:
1853                 {
1854                         struct drr_end drre = drr->drr_u.drr_end;
1855                         /*
1856                          * We compare against the *previous* checksum
1857                          * value, because the stored checksum is of
1858                          * everything before the DRR_END record.
1859                          */
1860                         if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1861                                 ra.err = SET_ERROR(ECKSUM);
1862                         goto out;
1863                 }
1864                 case DRR_SPILL:
1865                 {
1866                         struct drr_spill drrs = drr->drr_u.drr_spill;
1867                         ra.err = restore_spill(&ra, os, &drrs);
1868                         break;
1869                 }
1870                 default:
1871                         ra.err = SET_ERROR(EINVAL);
1872                         goto out;
1873                 }
1874                 pcksum = ra.cksum;
1875         }
1876         ASSERT(ra.err != 0);
1877
1878 out:
1879         if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1880                 zfs_onexit_fd_rele(cleanup_fd);
1881
1882         if (ra.err != 0) {
1883                 /*
1884                  * destroy what we created, so we don't leave it in the
1885                  * inconsistent restoring state.
1886                  */
1887                 dmu_recv_cleanup_ds(drc);
1888         }
1889
1890         kmem_free(ra.buf, ra.bufsize);
1891         *voffp = ra.voff;
1892         return (ra.err);
1893 }
1894
1895 static int
1896 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1897 {
1898         dmu_recv_cookie_t *drc = arg;
1899         dsl_pool_t *dp = dmu_tx_pool(tx);
1900         int error;
1901
1902         ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1903
1904         if (!drc->drc_newfs) {
1905                 dsl_dataset_t *origin_head;
1906
1907                 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1908                 if (error != 0)
1909                         return (error);
1910                 if (drc->drc_force) {
1911                         /*
1912                          * We will destroy any snapshots in tofs (i.e. before
1913                          * origin_head) that are after the origin (which is
1914                          * the snap before drc_ds, because drc_ds can not
1915                          * have any snaps of its own).
1916                          */
1917                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1918                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1919                                 dsl_dataset_t *snap;
1920                                 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1921                                     &snap);
1922                                 if (error != 0)
1923                                         return (error);
1924                                 if (snap->ds_dir != origin_head->ds_dir)
1925                                         error = SET_ERROR(EINVAL);
1926                                 if (error == 0)  {
1927                                         error = dsl_destroy_snapshot_check_impl(
1928                                             snap, B_FALSE);
1929                                 }
1930                                 obj = snap->ds_phys->ds_prev_snap_obj;
1931                                 dsl_dataset_rele(snap, FTAG);
1932                                 if (error != 0)
1933                                         return (error);
1934                         }
1935                 }
1936                 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1937                     origin_head, drc->drc_force, drc->drc_owner, tx);
1938                 if (error != 0) {
1939                         dsl_dataset_rele(origin_head, FTAG);
1940                         return (error);
1941                 }
1942                 error = dsl_dataset_snapshot_check_impl(origin_head,
1943                     drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
1944                 dsl_dataset_rele(origin_head, FTAG);
1945                 if (error != 0)
1946                         return (error);
1947
1948                 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1949         } else {
1950                 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1951                     drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
1952         }
1953         return (error);
1954 }
1955
1956 static void
1957 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1958 {
1959         dmu_recv_cookie_t *drc = arg;
1960         dsl_pool_t *dp = dmu_tx_pool(tx);
1961
1962         spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1963             tx, "snap=%s", drc->drc_tosnap);
1964
1965         if (!drc->drc_newfs) {
1966                 dsl_dataset_t *origin_head;
1967
1968                 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1969                     &origin_head));
1970
1971                 if (drc->drc_force) {
1972                         /*
1973                          * Destroy any snapshots of drc_tofs (origin_head)
1974                          * after the origin (the snap before drc_ds).
1975                          */
1976                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1977                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1978                                 dsl_dataset_t *snap;
1979                                 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1980                                     &snap));
1981                                 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1982                                 obj = snap->ds_phys->ds_prev_snap_obj;
1983                                 dsl_destroy_snapshot_sync_impl(snap,
1984                                     B_FALSE, tx);
1985                                 dsl_dataset_rele(snap, FTAG);
1986                         }
1987                 }
1988                 VERIFY3P(drc->drc_ds->ds_prev, ==,
1989                     origin_head->ds_prev);
1990
1991                 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1992                     origin_head, tx);
1993                 dsl_dataset_snapshot_sync_impl(origin_head,
1994                     drc->drc_tosnap, tx);
1995
1996                 /* set snapshot's creation time and guid */
1997                 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1998                 origin_head->ds_prev->ds_phys->ds_creation_time =
1999                     drc->drc_drrb->drr_creation_time;
2000                 origin_head->ds_prev->ds_phys->ds_guid =
2001                     drc->drc_drrb->drr_toguid;
2002                 origin_head->ds_prev->ds_phys->ds_flags &=
2003                     ~DS_FLAG_INCONSISTENT;
2004
2005                 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2006                 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
2007
2008                 dsl_dataset_rele(origin_head, FTAG);
2009                 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
2010
2011                 if (drc->drc_owner != NULL)
2012                         VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
2013         } else {
2014                 dsl_dataset_t *ds = drc->drc_ds;
2015
2016                 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
2017
2018                 /* set snapshot's creation time and guid */
2019                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2020                 ds->ds_prev->ds_phys->ds_creation_time =
2021                     drc->drc_drrb->drr_creation_time;
2022                 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
2023                 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
2024
2025                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2026                 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
2027         }
2028         drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
2029         /*
2030          * Release the hold from dmu_recv_begin.  This must be done before
2031          * we return to open context, so that when we free the dataset's dnode,
2032          * we can evict its bonus buffer.
2033          */
2034         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
2035         drc->drc_ds = NULL;
2036 }
2037
2038 static int
2039 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
2040 {
2041         dsl_pool_t *dp;
2042         dsl_dataset_t *snapds;
2043         guid_map_entry_t *gmep;
2044         int err;
2045
2046         ASSERT(guid_map != NULL);
2047
2048         err = dsl_pool_hold(name, FTAG, &dp);
2049         if (err != 0)
2050                 return (err);
2051         gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
2052         err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
2053         if (err == 0) {
2054                 gmep->guid = snapds->ds_phys->ds_guid;
2055                 gmep->gme_ds = snapds;
2056                 avl_add(guid_map, gmep);
2057                 dsl_dataset_long_hold(snapds, gmep);
2058         } else
2059                 kmem_free(gmep, sizeof (*gmep));
2060
2061         dsl_pool_rele(dp, FTAG);
2062         return (err);
2063 }
2064
2065 static int dmu_recv_end_modified_blocks = 3;
2066
2067 static int
2068 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
2069 {
2070         int error;
2071         char name[MAXNAMELEN];
2072
2073 #ifdef _KERNEL
2074         /*
2075          * We will be destroying the ds; make sure its origin is unmounted if
2076          * necessary.
2077          */
2078         dsl_dataset_name(drc->drc_ds, name);
2079         zfs_destroy_unmount_origin(name);
2080 #endif
2081
2082         error = dsl_sync_task(drc->drc_tofs,
2083             dmu_recv_end_check, dmu_recv_end_sync, drc,
2084             dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2085
2086         if (error != 0)
2087                 dmu_recv_cleanup_ds(drc);
2088         return (error);
2089 }
2090
2091 static int
2092 dmu_recv_new_end(dmu_recv_cookie_t *drc)
2093 {
2094         int error;
2095
2096         error = dsl_sync_task(drc->drc_tofs,
2097             dmu_recv_end_check, dmu_recv_end_sync, drc,
2098             dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2099
2100         if (error != 0) {
2101                 dmu_recv_cleanup_ds(drc);
2102         } else if (drc->drc_guid_to_ds_map != NULL) {
2103                 (void) add_ds_to_guidmap(drc->drc_tofs,
2104                     drc->drc_guid_to_ds_map,
2105                     drc->drc_newsnapobj);
2106         }
2107         return (error);
2108 }
2109
2110 int
2111 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
2112 {
2113         drc->drc_owner = owner;
2114
2115         if (drc->drc_newfs)
2116                 return (dmu_recv_new_end(drc));
2117         else
2118                 return (dmu_recv_existing_end(drc));
2119 }
2120
2121 /*
2122  * Return TRUE if this objset is currently being received into.
2123  */
2124 boolean_t
2125 dmu_objset_is_receiving(objset_t *os)
2126 {
2127         return (os->os_dsl_dataset != NULL &&
2128             os->os_dsl_dataset->ds_owner == dmu_recv_tag);
2129 }