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         /* fromsnap, if provided, must be a snapshot */
 734         if (fromds != NULL && !dsl_dataset_is_snapshot(fromds))
 735                 return (SET_ERROR(EINVAL));
 736
 737         /*
 738          * fromsnap must be an earlier snapshot from the same fs as tosnap,
 739          * or the origin's fs.
 740          */
 741         if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
 742                 return (SET_ERROR(EXDEV));
 743
 744         /* Get uncompressed size estimate of changed data. */
 745         if (fromds == NULL) {
 746                 size = ds->ds_phys->ds_uncompressed_bytes;
 747         } else {
 748                 uint64_t used, comp;
 749                 err = dsl_dataset_space_written(fromds, ds,
 750                     &used, &comp, &size);
 751                 if (err != 0)
 752                         return (err);
 753         }
 754
 755         /*
 756          * Assume that space (both on-disk and in-stream) is dominated by
 757          * data.  We will adjust for indirect blocks and the copies property,
 758          * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
 759          */
 760
 761         /*
 762          * Subtract out approximate space used by indirect blocks.
 763          * Assume most space is used by data blocks (non-indirect, non-dnode).
 764          * Assume all blocks are recordsize.  Assume ditto blocks and
 765          * internal fragmentation counter out compression.
 766          *
 767          * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
 768          * block, which we observe in practice.
 769          */
 770         uint64_t recordsize;
 771         err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
 772         if (err != 0)
 773                 return (err);
 774         size -= size / recordsize * sizeof (blkptr_t);
 775
 776         /* Add in the space for the record associated with each block. */
 777         size += size / recordsize * sizeof (dmu_replay_record_t);
 778
 779         *sizep = size;
 780
 781         return (0);
 782 }
 783
 784 typedef struct dmu_recv_begin_arg {
 785         const char *drba_origin;
 786         dmu_recv_cookie_t *drba_cookie;
 787         cred_t *drba_cred;
 788         uint64_t drba_snapobj;
 789 } dmu_recv_begin_arg_t;
 790
 791 static int
 792 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
 793     uint64_t fromguid)
 794 {
 795         uint64_t val;
 796         int error;
 797         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 798
 799         /* temporary clone name must not exist */
 800         error = zap_lookup(dp->dp_meta_objset,
 801             ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
 802             8, 1, &val);
 803         if (error != ENOENT)
 804                 return (error == 0 ? EBUSY : error);
 805
 806         /* new snapshot name must not exist */
 807         error = zap_lookup(dp->dp_meta_objset,
 808             ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
 809             8, 1, &val);
 810         if (error != ENOENT)
 811                 return (error == 0 ? EEXIST : error);
 812
 813         /*
 814          * Check snapshot limit before receiving. We'll recheck again at the
 815          * end, but might as well abort before receiving if we're already over
 816          * the limit.
 817          *
 818          * Note that we do not check the file system limit with
 819          * dsl_dir_fscount_check because the temporary %clones don't count
 820          * against that limit.
 821          */
 822         error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT,
 823             NULL, drba->drba_cred);
 824         if (error != 0)
 825                 return (error);
 826
 827         if (fromguid != 0) {
 828                 dsl_dataset_t *snap;
 829                 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
 830
 831                 /* Find snapshot in this dir that matches fromguid. */
 832                 while (obj != 0) {
 833                         error = dsl_dataset_hold_obj(dp, obj, FTAG,
 834                             &snap);
 835                         if (error != 0)
 836                                 return (SET_ERROR(ENODEV));
 837                         if (snap->ds_dir != ds->ds_dir) {
 838                                 dsl_dataset_rele(snap, FTAG);
 839                                 return (SET_ERROR(ENODEV));
 840                         }
 841                         if (snap->ds_phys->ds_guid == fromguid)
 842                                 break;
 843                         obj = snap->ds_phys->ds_prev_snap_obj;
 844                         dsl_dataset_rele(snap, FTAG);
 845                 }
 846                 if (obj == 0)
 847                         return (SET_ERROR(ENODEV));
 848
 849                 if (drba->drba_cookie->drc_force) {
 850                         drba->drba_snapobj = obj;
 851                 } else {
 852                         /*
 853                          * If we are not forcing, there must be no
 854                          * changes since fromsnap.
 855                          */
 856                         if (dsl_dataset_modified_since_snap(ds, snap)) {
 857                                 dsl_dataset_rele(snap, FTAG);
 858                                 return (SET_ERROR(ETXTBSY));
 859                         }
 860                         drba->drba_snapobj = ds->ds_prev->ds_object;
 861                 }
 862
 863                 dsl_dataset_rele(snap, FTAG);
 864         } else {
 865                 /* if full, then must be forced */
 866                 if (!drba->drba_cookie->drc_force)
 867                         return (SET_ERROR(EEXIST));
 868                 /* start from $ORIGIN@$ORIGIN, if supported */
 869                 drba->drba_snapobj = dp->dp_origin_snap != NULL ?
 870                     dp->dp_origin_snap->ds_object : 0;
 871         }
 872
 873         return (0);
 874
 875 }
 876
 877 static int
 878 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 879 {
 880         dmu_recv_begin_arg_t *drba = arg;
 881         dsl_pool_t *dp = dmu_tx_pool(tx);
 882         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 883         uint64_t fromguid = drrb->drr_fromguid;
 884         int flags = drrb->drr_flags;
 885         int error;
 886         dsl_dataset_t *ds;
 887         const char *tofs = drba->drba_cookie->drc_tofs;
 888
 889         /* already checked */
 890         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 891
 892         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 893             DMU_COMPOUNDSTREAM ||
 894             drrb->drr_type >= DMU_OST_NUMTYPES ||
 895             ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
 896                 return (SET_ERROR(EINVAL));
 897
 898         /* Verify pool version supports SA if SA_SPILL feature set */
 899         if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
 900             DMU_BACKUP_FEATURE_SA_SPILL) &&
 901             spa_version(dp->dp_spa) < SPA_VERSION_SA) {
 902                 return (SET_ERROR(ENOTSUP));
 903         }
 904
 905         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 906         if (error == 0) {
 907                 /* target fs already exists; recv into temp clone */
 908
 909                 /* Can't recv a clone into an existing fs */
 910                 if (flags & DRR_FLAG_CLONE) {
 911                         dsl_dataset_rele(ds, FTAG);
 912                         return (SET_ERROR(EINVAL));
 913                 }
 914
 915                 error = recv_begin_check_existing_impl(drba, ds, fromguid);
 916                 dsl_dataset_rele(ds, FTAG);
 917         } else if (error == ENOENT) {
 918                 /* target fs does not exist; must be a full backup or clone */
 919                 char buf[MAXNAMELEN];
 920
 921                 /*
 922                  * If it's a non-clone incremental, we are missing the
 923                  * target fs, so fail the recv.
 924                  */
 925                 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
 926                         return (SET_ERROR(ENOENT));
 927
 928                 /* Open the parent of tofs */
 929                 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
 930                 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
 931                 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
 932                 if (error != 0)
 933                         return (error);
 934
 935                 /*
 936                  * Check filesystem and snapshot limits before receiving. We'll
 937                  * recheck snapshot limits again at the end (we create the
 938                  * filesystems and increment those counts during begin_sync).
 939                  */
 940                 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
 941                     ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred);
 942                 if (error != 0) {
 943                         dsl_dataset_rele(ds, FTAG);
 944                         return (error);
 945                 }
 946
 947                 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
 948                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred);
 949                 if (error != 0) {
 950                         dsl_dataset_rele(ds, FTAG);
 951                         return (error);
 952                 }
 953
 954                 if (drba->drba_origin != NULL) {
 955                         dsl_dataset_t *origin;
 956                         error = dsl_dataset_hold(dp, drba->drba_origin,
 957                             FTAG, &origin);
 958                         if (error != 0) {
 959                                 dsl_dataset_rele(ds, FTAG);
 960                                 return (error);
 961                         }
 962                         if (!dsl_dataset_is_snapshot(origin)) {
 963                                 dsl_dataset_rele(origin, FTAG);
 964                                 dsl_dataset_rele(ds, FTAG);
 965                                 return (SET_ERROR(EINVAL));
 966                         }
 967                         if (origin->ds_phys->ds_guid != fromguid) {
 968                                 dsl_dataset_rele(origin, FTAG);
 969                                 dsl_dataset_rele(ds, FTAG);
 970                                 return (SET_ERROR(ENODEV));
 971                         }
 972                         dsl_dataset_rele(origin, FTAG);
 973                 }
 974                 dsl_dataset_rele(ds, FTAG);
 975                 error = 0;
 976         }
 977         return (error);
 978 }
 979
 980 static void
 981 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
 982 {
 983         dmu_recv_begin_arg_t *drba = arg;
 984         dsl_pool_t *dp = dmu_tx_pool(tx);
 985         struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
 986         const char *tofs = drba->drba_cookie->drc_tofs;
 987         dsl_dataset_t *ds, *newds;
 988         uint64_t dsobj;
 989         int error;
 990         uint64_t crflags;
 991
 992         crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
 993             DS_FLAG_CI_DATASET : 0;
 994
 995         error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 996         if (error == 0) {
 997                 /* create temporary clone */
 998                 dsl_dataset_t *snap = NULL;
 999                 if (drba->drba_snapobj != 0) {
1000                         VERIFY0(dsl_dataset_hold_obj(dp,
1001                             drba->drba_snapobj, FTAG, &snap));
1002                 }
1003                 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
1004                     snap, crflags, drba->drba_cred, tx);
1005                 if (drba->drba_snapobj != 0)
1006                         dsl_dataset_rele(snap, FTAG);
1007                 dsl_dataset_rele(ds, FTAG);
1008         } else {
1009                 dsl_dir_t *dd;
1010                 const char *tail;
1011                 dsl_dataset_t *origin = NULL;
1012
1013                 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1014
1015                 if (drba->drba_origin != NULL) {
1016                         VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1017                             FTAG, &origin));
1018                 }
1019
1020                 /* Create new dataset. */
1021                 dsobj = dsl_dataset_create_sync(dd,
1022                     strrchr(tofs, '/') + 1,
1023                     origin, crflags, drba->drba_cred, tx);
1024                 if (origin != NULL)
1025                         dsl_dataset_rele(origin, FTAG);
1026                 dsl_dir_rele(dd, FTAG);
1027                 drba->drba_cookie->drc_newfs = B_TRUE;
1028         }
1029         VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1030
1031         dmu_buf_will_dirty(newds->ds_dbuf, tx);
1032         newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1033
1034         /*
1035          * If we actually created a non-clone, we need to create the
1036          * objset in our new dataset.
1037          */
1038         if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1039                 (void) dmu_objset_create_impl(dp->dp_spa,
1040                     newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1041         }
1042
1043         drba->drba_cookie->drc_ds = newds;
1044
1045         spa_history_log_internal_ds(newds, "receive", tx, "");
1046 }
1047
1048 /*
1049  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1050  * succeeds; otherwise we will leak the holds on the datasets.
1051  */
1052 int
1053 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1054     boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1055 {
1056         dmu_recv_begin_arg_t drba = { 0 };
1057         dmu_replay_record_t *drr;
1058
1059         bzero(drc, sizeof (dmu_recv_cookie_t));
1060         drc->drc_drrb = drrb;
1061         drc->drc_tosnap = tosnap;
1062         drc->drc_tofs = tofs;
1063         drc->drc_force = force;
1064         drc->drc_cred = CRED();
1065
1066         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1067                 drc->drc_byteswap = B_TRUE;
1068         else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1069                 return (SET_ERROR(EINVAL));
1070
1071         drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1072         drr->drr_type = DRR_BEGIN;
1073         drr->drr_u.drr_begin = *drc->drc_drrb;
1074         if (drc->drc_byteswap) {
1075                 fletcher_4_incremental_byteswap(drr,
1076                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1077         } else {
1078                 fletcher_4_incremental_native(drr,
1079                     sizeof (dmu_replay_record_t), &drc->drc_cksum);
1080         }
1081         kmem_free(drr, sizeof (dmu_replay_record_t));
1082
1083         if (drc->drc_byteswap) {
1084                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1085                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1086                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1087                 drrb->drr_type = BSWAP_32(drrb->drr_type);
1088                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1089                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1090         }
1091
1092         drba.drba_origin = origin;
1093         drba.drba_cookie = drc;
1094         drba.drba_cred = CRED();
1095
1096         return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1097             &drba, 5));
1098 }
1099
1100 struct restorearg {
1101         int err;
1102         boolean_t byteswap;
1103         kthread_t *td;
1104         struct file *fp;
1105         char *buf;
1106         uint64_t voff;
1107         int bufsize; /* amount of memory allocated for buf */
1108         zio_cksum_t cksum;
1109         avl_tree_t *guid_to_ds_map;
1110 };
1111
1112 typedef struct guid_map_entry {
1113         uint64_t        guid;
1114         dsl_dataset_t   *gme_ds;
1115         avl_node_t      avlnode;
1116 } guid_map_entry_t;
1117
1118 static int
1119 guid_compare(const void *arg1, const void *arg2)
1120 {
1121         const guid_map_entry_t *gmep1 = arg1;
1122         const guid_map_entry_t *gmep2 = arg2;
1123
1124         if (gmep1->guid < gmep2->guid)
1125                 return (-1);
1126         else if (gmep1->guid > gmep2->guid)
1127                 return (1);
1128         return (0);
1129 }
1130
1131 static void
1132 free_guid_map_onexit(void *arg)
1133 {
1134         avl_tree_t *ca = arg;
1135         void *cookie = NULL;
1136         guid_map_entry_t *gmep;
1137
1138         while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1139                 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1140                 dsl_dataset_rele(gmep->gme_ds, gmep);
1141                 kmem_free(gmep, sizeof (guid_map_entry_t));
1142         }
1143         avl_destroy(ca);
1144         kmem_free(ca, sizeof (avl_tree_t));
1145 }
1146
1147 static int
1148 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
1149 {
1150         struct uio auio;
1151         struct iovec aiov;
1152         int error;
1153
1154         aiov.iov_base = buf;
1155         aiov.iov_len = len;
1156         auio.uio_iov = &aiov;
1157         auio.uio_iovcnt = 1;
1158         auio.uio_resid = len;
1159         auio.uio_segflg = UIO_SYSSPACE;
1160         auio.uio_rw = UIO_READ;
1161         auio.uio_offset = off;
1162         auio.uio_td = ra->td;
1163 #ifdef _KERNEL
1164         error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
1165 #else
1166         fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
1167         error = EOPNOTSUPP;
1168 #endif
1169         *resid = auio.uio_resid;
1170         return (error);
1171 }
1172
1173 static void *
1174 restore_read(struct restorearg *ra, int len)
1175 {
1176         void *rv;
1177         int done = 0;
1178
1179         /* some things will require 8-byte alignment, so everything must */
1180         ASSERT0(len % 8);
1181
1182         while (done < len) {
1183                 ssize_t resid;
1184
1185                 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
1186                     len - done, ra->voff, &resid);
1187
1188                 if (resid == len - done)
1189                         ra->err = SET_ERROR(EINVAL);
1190                 ra->voff += len - done - resid;
1191                 done = len - resid;
1192                 if (ra->err != 0)
1193                         return (NULL);
1194         }
1195
1196         ASSERT3U(done, ==, len);
1197         rv = ra->buf;
1198         if (ra->byteswap)
1199                 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1200         else
1201                 fletcher_4_incremental_native(rv, len, &ra->cksum);
1202         return (rv);
1203 }
1204
1205 static void
1206 backup_byteswap(dmu_replay_record_t *drr)
1207 {
1208 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1209 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1210         drr->drr_type = BSWAP_32(drr->drr_type);
1211         drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1212         switch (drr->drr_type) {
1213         case DRR_BEGIN:
1214                 DO64(drr_begin.drr_magic);
1215                 DO64(drr_begin.drr_versioninfo);
1216                 DO64(drr_begin.drr_creation_time);
1217                 DO32(drr_begin.drr_type);
1218                 DO32(drr_begin.drr_flags);
1219                 DO64(drr_begin.drr_toguid);
1220                 DO64(drr_begin.drr_fromguid);
1221                 break;
1222         case DRR_OBJECT:
1223                 DO64(drr_object.drr_object);
1224                 /* DO64(drr_object.drr_allocation_txg); */
1225                 DO32(drr_object.drr_type);
1226                 DO32(drr_object.drr_bonustype);
1227                 DO32(drr_object.drr_blksz);
1228                 DO32(drr_object.drr_bonuslen);
1229                 DO64(drr_object.drr_toguid);
1230                 break;
1231         case DRR_FREEOBJECTS:
1232                 DO64(drr_freeobjects.drr_firstobj);
1233                 DO64(drr_freeobjects.drr_numobjs);
1234                 DO64(drr_freeobjects.drr_toguid);
1235                 break;
1236         case DRR_WRITE:
1237                 DO64(drr_write.drr_object);
1238                 DO32(drr_write.drr_type);
1239                 DO64(drr_write.drr_offset);
1240                 DO64(drr_write.drr_length);
1241                 DO64(drr_write.drr_toguid);
1242                 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1243                 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1244                 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1245                 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1246                 DO64(drr_write.drr_key.ddk_prop);
1247                 break;
1248         case DRR_WRITE_BYREF:
1249                 DO64(drr_write_byref.drr_object);
1250                 DO64(drr_write_byref.drr_offset);
1251                 DO64(drr_write_byref.drr_length);
1252                 DO64(drr_write_byref.drr_toguid);
1253                 DO64(drr_write_byref.drr_refguid);
1254                 DO64(drr_write_byref.drr_refobject);
1255                 DO64(drr_write_byref.drr_refoffset);
1256                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1257                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1258                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1259                 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1260                 DO64(drr_write_byref.drr_key.ddk_prop);
1261                 break;
1262         case DRR_FREE:
1263                 DO64(drr_free.drr_object);
1264                 DO64(drr_free.drr_offset);
1265                 DO64(drr_free.drr_length);
1266                 DO64(drr_free.drr_toguid);
1267                 break;
1268         case DRR_SPILL:
1269                 DO64(drr_spill.drr_object);
1270                 DO64(drr_spill.drr_length);
1271                 DO64(drr_spill.drr_toguid);
1272                 break;
1273         case DRR_END:
1274                 DO64(drr_end.drr_checksum.zc_word[0]);
1275                 DO64(drr_end.drr_checksum.zc_word[1]);
1276                 DO64(drr_end.drr_checksum.zc_word[2]);
1277                 DO64(drr_end.drr_checksum.zc_word[3]);
1278                 DO64(drr_end.drr_toguid);
1279                 break;
1280         }
1281 #undef DO64
1282 #undef DO32
1283 }
1284
1285 static int
1286 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1287 {
1288         int err;
1289         dmu_tx_t *tx;
1290         void *data = NULL;
1291
1292         if (drro->drr_type == DMU_OT_NONE ||
1293             !DMU_OT_IS_VALID(drro->drr_type) ||
1294             !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1295             drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1296             drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1297             P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1298             drro->drr_blksz < SPA_MINBLOCKSIZE ||
1299             drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1300             drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1301                 return (SET_ERROR(EINVAL));
1302         }
1303
1304         err = dmu_object_info(os, drro->drr_object, NULL);
1305
1306         if (err != 0 && err != ENOENT)
1307                 return (SET_ERROR(EINVAL));
1308
1309         if (drro->drr_bonuslen) {
1310                 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1311                 if (ra->err != 0)
1312                         return (ra->err);
1313         }
1314
1315         if (err == ENOENT) {
1316                 /* currently free, want to be allocated */
1317                 tx = dmu_tx_create(os);
1318                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1319                 err = dmu_tx_assign(tx, TXG_WAIT);
1320                 if (err != 0) {
1321                         dmu_tx_abort(tx);
1322                         return (err);
1323                 }
1324                 err = dmu_object_claim(os, drro->drr_object,
1325                     drro->drr_type, drro->drr_blksz,
1326                     drro->drr_bonustype, drro->drr_bonuslen, tx);
1327                 dmu_tx_commit(tx);
1328         } else {
1329                 /* currently allocated, want to be allocated */
1330                 err = dmu_object_reclaim(os, drro->drr_object,
1331                     drro->drr_type, drro->drr_blksz,
1332                     drro->drr_bonustype, drro->drr_bonuslen);
1333         }
1334         if (err != 0) {
1335                 return (SET_ERROR(EINVAL));
1336         }
1337
1338         tx = dmu_tx_create(os);
1339         dmu_tx_hold_bonus(tx, drro->drr_object);
1340         err = dmu_tx_assign(tx, TXG_WAIT);
1341         if (err != 0) {
1342                 dmu_tx_abort(tx);
1343                 return (err);
1344         }
1345
1346         dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1347             tx);
1348         dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1349
1350         if (data != NULL) {
1351                 dmu_buf_t *db;
1352
1353                 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1354                 dmu_buf_will_dirty(db, tx);
1355
1356                 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1357                 bcopy(data, db->db_data, drro->drr_bonuslen);
1358                 if (ra->byteswap) {
1359                         dmu_object_byteswap_t byteswap =
1360                             DMU_OT_BYTESWAP(drro->drr_bonustype);
1361                         dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1362                             drro->drr_bonuslen);
1363                 }
1364                 dmu_buf_rele(db, FTAG);
1365         }
1366         dmu_tx_commit(tx);
1367         return (0);
1368 }
1369
1370 /* ARGSUSED */
1371 static int
1372 restore_freeobjects(struct restorearg *ra, objset_t *os,
1373     struct drr_freeobjects *drrfo)
1374 {
1375         uint64_t obj;
1376
1377         if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1378                 return (SET_ERROR(EINVAL));
1379
1380         for (obj = drrfo->drr_firstobj;
1381             obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1382             (void) dmu_object_next(os, &obj, FALSE, 0)) {
1383                 int err;
1384
1385                 if (dmu_object_info(os, obj, NULL) != 0)
1386                         continue;
1387
1388                 err = dmu_free_long_object(os, obj);
1389                 if (err != 0)
1390                         return (err);
1391         }
1392         return (0);
1393 }
1394
1395 static int
1396 restore_write(struct restorearg *ra, objset_t *os,
1397     struct drr_write *drrw)
1398 {
1399         dmu_tx_t *tx;
1400         void *data;
1401         int err;
1402
1403         if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1404             !DMU_OT_IS_VALID(drrw->drr_type))
1405                 return (SET_ERROR(EINVAL));
1406
1407         data = restore_read(ra, drrw->drr_length);
1408         if (data == NULL)
1409                 return (ra->err);
1410
1411         if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1412                 return (SET_ERROR(EINVAL));
1413
1414         tx = dmu_tx_create(os);
1415
1416         dmu_tx_hold_write(tx, drrw->drr_object,
1417             drrw->drr_offset, drrw->drr_length);
1418         err = dmu_tx_assign(tx, TXG_WAIT);
1419         if (err != 0) {
1420                 dmu_tx_abort(tx);
1421                 return (err);
1422         }
1423         if (ra->byteswap) {
1424                 dmu_object_byteswap_t byteswap =
1425                     DMU_OT_BYTESWAP(drrw->drr_type);
1426                 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1427         }
1428         dmu_write(os, drrw->drr_object,
1429             drrw->drr_offset, drrw->drr_length, data, tx);
1430         dmu_tx_commit(tx);
1431         return (0);
1432 }
1433
1434 /*
1435  * Handle a DRR_WRITE_BYREF record.  This record is used in dedup'ed
1436  * streams to refer to a copy of the data that is already on the
1437  * system because it came in earlier in the stream.  This function
1438  * finds the earlier copy of the data, and uses that copy instead of
1439  * data from the stream to fulfill this write.
1440  */
1441 static int
1442 restore_write_byref(struct restorearg *ra, objset_t *os,
1443     struct drr_write_byref *drrwbr)
1444 {
1445         dmu_tx_t *tx;
1446         int err;
1447         guid_map_entry_t gmesrch;
1448         guid_map_entry_t *gmep;
1449         avl_index_t     where;
1450         objset_t *ref_os = NULL;
1451         dmu_buf_t *dbp;
1452
1453         if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1454                 return (SET_ERROR(EINVAL));
1455
1456         /*
1457          * If the GUID of the referenced dataset is different from the
1458          * GUID of the target dataset, find the referenced dataset.
1459          */
1460         if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1461                 gmesrch.guid = drrwbr->drr_refguid;
1462                 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1463                     &where)) == NULL) {
1464                         return (SET_ERROR(EINVAL));
1465                 }
1466                 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1467                         return (SET_ERROR(EINVAL));
1468         } else {
1469                 ref_os = os;
1470         }
1471
1472         if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1473             drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1474                 return (err);
1475
1476         tx = dmu_tx_create(os);
1477
1478         dmu_tx_hold_write(tx, drrwbr->drr_object,
1479             drrwbr->drr_offset, drrwbr->drr_length);
1480         err = dmu_tx_assign(tx, TXG_WAIT);
1481         if (err != 0) {
1482                 dmu_tx_abort(tx);
1483                 return (err);
1484         }
1485         dmu_write(os, drrwbr->drr_object,
1486             drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1487         dmu_buf_rele(dbp, FTAG);
1488         dmu_tx_commit(tx);
1489         return (0);
1490 }
1491
1492 static int
1493 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1494 {
1495         dmu_tx_t *tx;
1496         void *data;
1497         dmu_buf_t *db, *db_spill;
1498         int err;
1499
1500         if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1501             drrs->drr_length > SPA_MAXBLOCKSIZE)
1502                 return (SET_ERROR(EINVAL));
1503
1504         data = restore_read(ra, drrs->drr_length);
1505         if (data == NULL)
1506                 return (ra->err);
1507
1508         if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1509                 return (SET_ERROR(EINVAL));
1510
1511         VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1512         if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1513                 dmu_buf_rele(db, FTAG);
1514                 return (err);
1515         }
1516
1517         tx = dmu_tx_create(os);
1518
1519         dmu_tx_hold_spill(tx, db->db_object);
1520
1521         err = dmu_tx_assign(tx, TXG_WAIT);
1522         if (err != 0) {
1523                 dmu_buf_rele(db, FTAG);
1524                 dmu_buf_rele(db_spill, FTAG);
1525                 dmu_tx_abort(tx);
1526                 return (err);
1527         }
1528         dmu_buf_will_dirty(db_spill, tx);
1529
1530         if (db_spill->db_size < drrs->drr_length)
1531                 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1532                     drrs->drr_length, tx));
1533         bcopy(data, db_spill->db_data, drrs->drr_length);
1534
1535         dmu_buf_rele(db, FTAG);
1536         dmu_buf_rele(db_spill, FTAG);
1537
1538         dmu_tx_commit(tx);
1539         return (0);
1540 }
1541
1542 /* ARGSUSED */
1543 static int
1544 restore_free(struct restorearg *ra, objset_t *os,
1545     struct drr_free *drrf)
1546 {
1547         int err;
1548
1549         if (drrf->drr_length != -1ULL &&
1550             drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1551                 return (SET_ERROR(EINVAL));
1552
1553         if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1554                 return (SET_ERROR(EINVAL));
1555
1556         err = dmu_free_long_range(os, drrf->drr_object,
1557             drrf->drr_offset, drrf->drr_length);
1558         return (err);
1559 }
1560
1561 /* used to destroy the drc_ds on error */
1562 static void
1563 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1564 {
1565         char name[MAXNAMELEN];
1566         dsl_dataset_name(drc->drc_ds, name);
1567         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1568         (void) dsl_destroy_head(name);
1569 }
1570
1571 /*
1572  * NB: callers *must* call dmu_recv_end() if this succeeds.
1573  */
1574 int
1575 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1576     int cleanup_fd, uint64_t *action_handlep)
1577 {
1578         struct restorearg ra = { 0 };
1579         dmu_replay_record_t *drr;
1580         objset_t *os;
1581         zio_cksum_t pcksum;
1582         int featureflags;
1583
1584         ra.byteswap = drc->drc_byteswap;
1585         ra.cksum = drc->drc_cksum;
1586         ra.td = curthread;
1587         ra.fp = fp;
1588         ra.voff = *voffp;
1589         ra.bufsize = 1<<20;
1590         ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1591
1592         /* these were verified in dmu_recv_begin */
1593         ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1594             DMU_SUBSTREAM);
1595         ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1596
1597         /*
1598          * Open the objset we are modifying.
1599          */
1600         VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1601
1602         ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1603
1604         featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1605
1606         /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1607         if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1608                 minor_t minor;
1609
1610                 if (cleanup_fd == -1) {
1611                         ra.err = SET_ERROR(EBADF);
1612                         goto out;
1613                 }
1614                 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1615                 if (ra.err != 0) {
1616                         cleanup_fd = -1;
1617                         goto out;
1618                 }
1619
1620                 if (*action_handlep == 0) {
1621                         ra.guid_to_ds_map =
1622                             kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1623                         avl_create(ra.guid_to_ds_map, guid_compare,
1624                             sizeof (guid_map_entry_t),
1625                             offsetof(guid_map_entry_t, avlnode));
1626                         ra.err = zfs_onexit_add_cb(minor,
1627                             free_guid_map_onexit, ra.guid_to_ds_map,
1628                             action_handlep);
1629                         if (ra.err != 0)
1630                                 goto out;
1631                 } else {
1632                         ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1633                             (void **)&ra.guid_to_ds_map);
1634                         if (ra.err != 0)
1635                                 goto out;
1636                 }
1637
1638                 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1639         }
1640
1641         /*
1642          * Read records and process them.
1643          */
1644         pcksum = ra.cksum;
1645         while (ra.err == 0 &&
1646             NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1647                 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1648                         ra.err = SET_ERROR(EINTR);
1649                         goto out;
1650                 }
1651
1652                 if (ra.byteswap)
1653                         backup_byteswap(drr);
1654
1655                 switch (drr->drr_type) {
1656                 case DRR_OBJECT:
1657                 {
1658                         /*
1659                          * We need to make a copy of the record header,
1660                          * because restore_{object,write} may need to
1661                          * restore_read(), which will invalidate drr.
1662                          */
1663                         struct drr_object drro = drr->drr_u.drr_object;
1664                         ra.err = restore_object(&ra, os, &drro);
1665                         break;
1666                 }
1667                 case DRR_FREEOBJECTS:
1668                 {
1669                         struct drr_freeobjects drrfo =
1670                             drr->drr_u.drr_freeobjects;
1671                         ra.err = restore_freeobjects(&ra, os, &drrfo);
1672                         break;
1673                 }
1674                 case DRR_WRITE:
1675                 {
1676                         struct drr_write drrw = drr->drr_u.drr_write;
1677                         ra.err = restore_write(&ra, os, &drrw);
1678                         break;
1679                 }
1680                 case DRR_WRITE_BYREF:
1681                 {
1682                         struct drr_write_byref drrwbr =
1683                             drr->drr_u.drr_write_byref;
1684                         ra.err = restore_write_byref(&ra, os, &drrwbr);
1685                         break;
1686                 }
1687                 case DRR_FREE:
1688                 {
1689                         struct drr_free drrf = drr->drr_u.drr_free;
1690                         ra.err = restore_free(&ra, os, &drrf);
1691                         break;
1692                 }
1693                 case DRR_END:
1694                 {
1695                         struct drr_end drre = drr->drr_u.drr_end;
1696                         /*
1697                          * We compare against the *previous* checksum
1698                          * value, because the stored checksum is of
1699                          * everything before the DRR_END record.
1700                          */
1701                         if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1702                                 ra.err = SET_ERROR(ECKSUM);
1703                         goto out;
1704                 }
1705                 case DRR_SPILL:
1706                 {
1707                         struct drr_spill drrs = drr->drr_u.drr_spill;
1708                         ra.err = restore_spill(&ra, os, &drrs);
1709                         break;
1710                 }
1711                 default:
1712                         ra.err = SET_ERROR(EINVAL);
1713                         goto out;
1714                 }
1715                 pcksum = ra.cksum;
1716         }
1717         ASSERT(ra.err != 0);
1718
1719 out:
1720         if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1721                 zfs_onexit_fd_rele(cleanup_fd);
1722
1723         if (ra.err != 0) {
1724                 /*
1725                  * destroy what we created, so we don't leave it in the
1726                  * inconsistent restoring state.
1727                  */
1728                 dmu_recv_cleanup_ds(drc);
1729         }
1730
1731         kmem_free(ra.buf, ra.bufsize);
1732         *voffp = ra.voff;
1733         return (ra.err);
1734 }
1735
1736 static int
1737 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1738 {
1739         dmu_recv_cookie_t *drc = arg;
1740         dsl_pool_t *dp = dmu_tx_pool(tx);
1741         int error;
1742
1743         ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1744
1745         if (!drc->drc_newfs) {
1746                 dsl_dataset_t *origin_head;
1747
1748                 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1749                 if (error != 0)
1750                         return (error);
1751                 if (drc->drc_force) {
1752                         /*
1753                          * We will destroy any snapshots in tofs (i.e. before
1754                          * origin_head) that are after the origin (which is
1755                          * the snap before drc_ds, because drc_ds can not
1756                          * have any snaps of its own).
1757                          */
1758                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1759                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1760                                 dsl_dataset_t *snap;
1761                                 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1762                                     &snap);
1763                                 if (error != 0)
1764                                         break;
1765                                 if (snap->ds_dir != origin_head->ds_dir)
1766                                         error = SET_ERROR(EINVAL);
1767                                 if (error == 0)  {
1768                                         error = dsl_destroy_snapshot_check_impl(
1769                                             snap, B_FALSE);
1770                                 }
1771                                 obj = snap->ds_phys->ds_prev_snap_obj;
1772                                 dsl_dataset_rele(snap, FTAG);
1773                                 if (error != 0)
1774                                         break;
1775                         }
1776                         if (error != 0) {
1777                                 dsl_dataset_rele(origin_head, FTAG);
1778                                 return (error);
1779                         }
1780                 }
1781                 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1782                     origin_head, drc->drc_force, drc->drc_owner, tx);
1783                 if (error != 0) {
1784                         dsl_dataset_rele(origin_head, FTAG);
1785                         return (error);
1786                 }
1787                 error = dsl_dataset_snapshot_check_impl(origin_head,
1788                     drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
1789                 dsl_dataset_rele(origin_head, FTAG);
1790                 if (error != 0)
1791                         return (error);
1792
1793                 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1794         } else {
1795                 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1796                     drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
1797         }
1798         return (error);
1799 }
1800
1801 static void
1802 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1803 {
1804         dmu_recv_cookie_t *drc = arg;
1805         dsl_pool_t *dp = dmu_tx_pool(tx);
1806
1807         spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1808             tx, "snap=%s", drc->drc_tosnap);
1809
1810         if (!drc->drc_newfs) {
1811                 dsl_dataset_t *origin_head;
1812
1813                 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1814                     &origin_head));
1815
1816                 if (drc->drc_force) {
1817                         /*
1818                          * Destroy any snapshots of drc_tofs (origin_head)
1819                          * after the origin (the snap before drc_ds).
1820                          */
1821                         uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1822                         while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1823                                 dsl_dataset_t *snap;
1824                                 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1825                                     &snap));
1826                                 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1827                                 obj = snap->ds_phys->ds_prev_snap_obj;
1828                                 dsl_destroy_snapshot_sync_impl(snap,
1829                                     B_FALSE, tx);
1830                                 dsl_dataset_rele(snap, FTAG);
1831                         }
1832                 }
1833                 VERIFY3P(drc->drc_ds->ds_prev, ==,
1834                     origin_head->ds_prev);
1835
1836                 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1837                     origin_head, tx);
1838                 dsl_dataset_snapshot_sync_impl(origin_head,
1839                     drc->drc_tosnap, tx);
1840
1841                 /* set snapshot's creation time and guid */
1842                 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1843                 origin_head->ds_prev->ds_phys->ds_creation_time =
1844                     drc->drc_drrb->drr_creation_time;
1845                 origin_head->ds_prev->ds_phys->ds_guid =
1846                     drc->drc_drrb->drr_toguid;
1847                 origin_head->ds_prev->ds_phys->ds_flags &=
1848                     ~DS_FLAG_INCONSISTENT;
1849
1850                 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1851                 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1852
1853                 dsl_dataset_rele(origin_head, FTAG);
1854                 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1855
1856                 if (drc->drc_owner != NULL)
1857                         VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
1858         } else {
1859                 dsl_dataset_t *ds = drc->drc_ds;
1860
1861                 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1862
1863                 /* set snapshot's creation time and guid */
1864                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1865                 ds->ds_prev->ds_phys->ds_creation_time =
1866                     drc->drc_drrb->drr_creation_time;
1867                 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1868                 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1869
1870                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1871                 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1872         }
1873         drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1874         /*
1875          * Release the hold from dmu_recv_begin.  This must be done before
1876          * we return to open context, so that when we free the dataset's dnode,
1877          * we can evict its bonus buffer.
1878          */
1879         dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1880         drc->drc_ds = NULL;
1881 }
1882
1883 static int
1884 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1885 {
1886         dsl_pool_t *dp;
1887         dsl_dataset_t *snapds;
1888         guid_map_entry_t *gmep;
1889         int err;
1890
1891         ASSERT(guid_map != NULL);
1892
1893         err = dsl_pool_hold(name, FTAG, &dp);
1894         if (err != 0)
1895                 return (err);
1896         gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
1897         err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
1898         if (err == 0) {
1899                 gmep->guid = snapds->ds_phys->ds_guid;
1900                 gmep->gme_ds = snapds;
1901                 avl_add(guid_map, gmep);
1902                 dsl_dataset_long_hold(snapds, gmep);
1903         } else
1904                 kmem_free(gmep, sizeof (*gmep));
1905
1906         dsl_pool_rele(dp, FTAG);
1907         return (err);
1908 }
1909
1910 static int dmu_recv_end_modified_blocks = 3;
1911
1912 static int
1913 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1914 {
1915         int error;
1916         char name[MAXNAMELEN];
1917
1918 #ifdef _KERNEL
1919         /*
1920          * We will be destroying the ds; make sure its origin is unmounted if
1921          * necessary.
1922          */
1923         dsl_dataset_name(drc->drc_ds, name);
1924         zfs_destroy_unmount_origin(name);
1925 #endif
1926
1927         error = dsl_sync_task(drc->drc_tofs,
1928             dmu_recv_end_check, dmu_recv_end_sync, drc,
1929             dmu_recv_end_modified_blocks);
1930
1931         if (error != 0)
1932                 dmu_recv_cleanup_ds(drc);
1933         return (error);
1934 }
1935
1936 static int
1937 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1938 {
1939         int error;
1940
1941         error = dsl_sync_task(drc->drc_tofs,
1942             dmu_recv_end_check, dmu_recv_end_sync, drc,
1943             dmu_recv_end_modified_blocks);
1944
1945         if (error != 0) {
1946                 dmu_recv_cleanup_ds(drc);
1947         } else if (drc->drc_guid_to_ds_map != NULL) {
1948                 (void) add_ds_to_guidmap(drc->drc_tofs,
1949                     drc->drc_guid_to_ds_map,
1950                     drc->drc_newsnapobj);
1951         }
1952         return (error);
1953 }
1954
1955 int
1956 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
1957 {
1958         drc->drc_owner = owner;
1959
1960         if (drc->drc_newfs)
1961                 return (dmu_recv_new_end(drc));
1962         else
1963                 return (dmu_recv_existing_end(drc));
1964 }
1965
1966 /*
1967  * Return TRUE if this objset is currently being received into.
1968  */
1969 boolean_t
1970 dmu_objset_is_receiving(objset_t *os)
1971 {
1972         return (os->os_dsl_dataset != NULL &&
1973             os->os_dsl_dataset->ds_owner == dmu_recv_tag);
1974 }