xref: /titanic_44/usr/src/uts/common/fs/zfs/dmu_send.c (revision b6c3f7863936abeae522e48a13887dddeb691a45)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/dmu.h>
29 #include <sys/dmu_impl.h>
30 #include <sys/dmu_tx.h>
31 #include <sys/dbuf.h>
32 #include <sys/dnode.h>
33 #include <sys/zfs_context.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dsl_dataset.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dsl_pool.h>
39 #include <sys/dsl_synctask.h>
40 #include <sys/zfs_ioctl.h>
41 #include <sys/zap.h>
42 #include <sys/zio_checksum.h>
43 
44 static char *dmu_recv_tag = "dmu_recv_tag";
45 
46 struct backuparg {
47 	dmu_replay_record_t *drr;
48 	vnode_t *vp;
49 	offset_t *off;
50 	objset_t *os;
51 	zio_cksum_t zc;
52 	int err;
53 };
54 
55 static int
56 dump_bytes(struct backuparg *ba, void *buf, int len)
57 {
58 	ssize_t resid; /* have to get resid to get detailed errno */
59 	ASSERT3U(len % 8, ==, 0);
60 
61 	fletcher_4_incremental_native(buf, len, &ba->zc);
62 	ba->err = vn_rdwr(UIO_WRITE, ba->vp,
63 	    (caddr_t)buf, len,
64 	    0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
65 	*ba->off += len;
66 	return (ba->err);
67 }
68 
69 static int
70 dump_free(struct backuparg *ba, uint64_t object, uint64_t offset,
71     uint64_t length)
72 {
73 	/* write a FREE record */
74 	bzero(ba->drr, sizeof (dmu_replay_record_t));
75 	ba->drr->drr_type = DRR_FREE;
76 	ba->drr->drr_u.drr_free.drr_object = object;
77 	ba->drr->drr_u.drr_free.drr_offset = offset;
78 	ba->drr->drr_u.drr_free.drr_length = length;
79 
80 	if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
81 		return (EINTR);
82 	return (0);
83 }
84 
85 static int
86 dump_data(struct backuparg *ba, dmu_object_type_t type,
87     uint64_t object, uint64_t offset, int blksz, void *data)
88 {
89 	/* write a DATA record */
90 	bzero(ba->drr, sizeof (dmu_replay_record_t));
91 	ba->drr->drr_type = DRR_WRITE;
92 	ba->drr->drr_u.drr_write.drr_object = object;
93 	ba->drr->drr_u.drr_write.drr_type = type;
94 	ba->drr->drr_u.drr_write.drr_offset = offset;
95 	ba->drr->drr_u.drr_write.drr_length = blksz;
96 
97 	if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
98 		return (EINTR);
99 	if (dump_bytes(ba, data, blksz))
100 		return (EINTR);
101 	return (0);
102 }
103 
104 static int
105 dump_freeobjects(struct backuparg *ba, uint64_t firstobj, uint64_t numobjs)
106 {
107 	/* write a FREEOBJECTS record */
108 	bzero(ba->drr, sizeof (dmu_replay_record_t));
109 	ba->drr->drr_type = DRR_FREEOBJECTS;
110 	ba->drr->drr_u.drr_freeobjects.drr_firstobj = firstobj;
111 	ba->drr->drr_u.drr_freeobjects.drr_numobjs = numobjs;
112 
113 	if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
114 		return (EINTR);
115 	return (0);
116 }
117 
118 static int
119 dump_dnode(struct backuparg *ba, uint64_t object, dnode_phys_t *dnp)
120 {
121 	if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
122 		return (dump_freeobjects(ba, object, 1));
123 
124 	/* write an OBJECT record */
125 	bzero(ba->drr, sizeof (dmu_replay_record_t));
126 	ba->drr->drr_type = DRR_OBJECT;
127 	ba->drr->drr_u.drr_object.drr_object = object;
128 	ba->drr->drr_u.drr_object.drr_type = dnp->dn_type;
129 	ba->drr->drr_u.drr_object.drr_bonustype = dnp->dn_bonustype;
130 	ba->drr->drr_u.drr_object.drr_blksz =
131 	    dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
132 	ba->drr->drr_u.drr_object.drr_bonuslen = dnp->dn_bonuslen;
133 	ba->drr->drr_u.drr_object.drr_checksum = dnp->dn_checksum;
134 	ba->drr->drr_u.drr_object.drr_compress = dnp->dn_compress;
135 
136 	if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
137 		return (EINTR);
138 
139 	if (dump_bytes(ba, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)))
140 		return (EINTR);
141 
142 	/* free anything past the end of the file */
143 	if (dump_free(ba, object, (dnp->dn_maxblkid + 1) *
144 	    (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
145 		return (EINTR);
146 	if (ba->err)
147 		return (EINTR);
148 	return (0);
149 }
150 
151 #define	BP_SPAN(dnp, level) \
152 	(((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
153 	(level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
154 
155 static int
156 backup_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
157 {
158 	struct backuparg *ba = arg;
159 	uint64_t object = bc->bc_bookmark.zb_object;
160 	int level = bc->bc_bookmark.zb_level;
161 	uint64_t blkid = bc->bc_bookmark.zb_blkid;
162 	blkptr_t *bp = bc->bc_blkptr.blk_birth ? &bc->bc_blkptr : NULL;
163 	dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
164 	void *data = bc->bc_data;
165 	int err = 0;
166 
167 	if (issig(JUSTLOOKING) && issig(FORREAL))
168 		return (EINTR);
169 
170 	ASSERT(data || bp == NULL);
171 
172 	if (bp == NULL && object == 0) {
173 		uint64_t span = BP_SPAN(bc->bc_dnode, level);
174 		uint64_t dnobj = (blkid * span) >> DNODE_SHIFT;
175 		err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
176 	} else if (bp == NULL) {
177 		uint64_t span = BP_SPAN(bc->bc_dnode, level);
178 		err = dump_free(ba, object, blkid * span, span);
179 	} else if (data && level == 0 && type == DMU_OT_DNODE) {
180 		dnode_phys_t *blk = data;
181 		int i;
182 		int blksz = BP_GET_LSIZE(bp);
183 
184 		for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
185 			uint64_t dnobj =
186 			    (blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
187 			err = dump_dnode(ba, dnobj, blk+i);
188 			if (err)
189 				break;
190 		}
191 	} else if (level == 0 &&
192 	    type != DMU_OT_DNODE && type != DMU_OT_OBJSET) {
193 		int blksz = BP_GET_LSIZE(bp);
194 		if (data == NULL) {
195 			uint32_t aflags = ARC_WAIT;
196 			arc_buf_t *abuf;
197 			zbookmark_t zb;
198 
199 			zb.zb_objset = ba->os->os->os_dsl_dataset->ds_object;
200 			zb.zb_object = object;
201 			zb.zb_level = level;
202 			zb.zb_blkid = blkid;
203 			(void) arc_read(NULL, spa, bp,
204 			    dmu_ot[type].ot_byteswap, arc_getbuf_func, &abuf,
205 			    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_MUSTSUCCEED,
206 			    &aflags, &zb);
207 
208 			if (abuf) {
209 				err = dump_data(ba, type, object, blkid * blksz,
210 				    blksz, abuf->b_data);
211 				(void) arc_buf_remove_ref(abuf, &abuf);
212 			}
213 		} else {
214 			err = dump_data(ba, type, object, blkid * blksz,
215 			    blksz, data);
216 		}
217 	}
218 
219 	ASSERT(err == 0 || err == EINTR);
220 	return (err);
221 }
222 
223 int
224 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
225     vnode_t *vp, offset_t *off)
226 {
227 	dsl_dataset_t *ds = tosnap->os->os_dsl_dataset;
228 	dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL;
229 	dmu_replay_record_t *drr;
230 	struct backuparg ba;
231 	int err;
232 	uint64_t fromtxg = 0;
233 
234 	/* tosnap must be a snapshot */
235 	if (ds->ds_phys->ds_next_snap_obj == 0)
236 		return (EINVAL);
237 
238 	/* fromsnap must be an earlier snapshot from the same fs as tosnap */
239 	if (fromds && (ds->ds_dir != fromds->ds_dir ||
240 	    fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
241 		return (EXDEV);
242 
243 	if (fromorigin) {
244 		if (fromsnap)
245 			return (EINVAL);
246 
247 		if (ds->ds_dir->dd_phys->dd_origin_obj != NULL) {
248 			dsl_pool_t *dp = ds->ds_dir->dd_pool;
249 			rw_enter(&dp->dp_config_rwlock, RW_READER);
250 			err = dsl_dataset_open_obj(dp,
251 			    ds->ds_dir->dd_phys->dd_origin_obj, NULL,
252 			    DS_MODE_NONE, FTAG, &fromds);
253 			rw_exit(&dp->dp_config_rwlock);
254 			if (err)
255 				return (err);
256 		} else {
257 			fromorigin = B_FALSE;
258 		}
259 	}
260 
261 
262 	drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
263 	drr->drr_type = DRR_BEGIN;
264 	drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
265 	drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION;
266 	drr->drr_u.drr_begin.drr_creation_time =
267 	    ds->ds_phys->ds_creation_time;
268 	drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type;
269 	if (fromorigin)
270 		drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
271 	drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
272 	if (fromds)
273 		drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
274 	dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
275 
276 	if (fromds)
277 		fromtxg = fromds->ds_phys->ds_creation_txg;
278 	if (fromorigin)
279 		dsl_dataset_close(fromds, DS_MODE_NONE, FTAG);
280 
281 	ba.drr = drr;
282 	ba.vp = vp;
283 	ba.os = tosnap;
284 	ba.off = off;
285 	ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
286 
287 	if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
288 		kmem_free(drr, sizeof (dmu_replay_record_t));
289 		return (ba.err);
290 	}
291 
292 	err = traverse_dsl_dataset(ds, fromtxg,
293 	    ADVANCE_PRE | ADVANCE_HOLES | ADVANCE_DATA | ADVANCE_NOLOCK,
294 	    backup_cb, &ba);
295 
296 	if (err) {
297 		if (err == EINTR && ba.err)
298 			err = ba.err;
299 		kmem_free(drr, sizeof (dmu_replay_record_t));
300 		return (err);
301 	}
302 
303 	bzero(drr, sizeof (dmu_replay_record_t));
304 	drr->drr_type = DRR_END;
305 	drr->drr_u.drr_end.drr_checksum = ba.zc;
306 
307 	if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
308 		kmem_free(drr, sizeof (dmu_replay_record_t));
309 		return (ba.err);
310 	}
311 
312 	kmem_free(drr, sizeof (dmu_replay_record_t));
313 
314 	return (0);
315 }
316 
317 struct recvbeginsyncarg {
318 	const char *tofs;
319 	const char *tosnap;
320 	dsl_dataset_t *origin;
321 	uint64_t fromguid;
322 	dmu_objset_type_t type;
323 	void *tag;
324 	boolean_t force;
325 	char clonelastname[MAXNAMELEN];
326 	dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
327 };
328 
329 static dsl_dataset_t *
330 recv_full_sync_impl(dsl_pool_t *dp, uint64_t dsobj, dmu_objset_type_t type,
331     cred_t *cr, dmu_tx_t *tx)
332 {
333 	dsl_dataset_t *ds;
334 
335 	VERIFY(0 == dsl_dataset_open_obj(dp, dsobj, NULL,
336 	    DS_MODE_EXCLUSIVE, dmu_recv_tag, &ds));
337 
338 	if (type != DMU_OST_NONE) {
339 		(void) dmu_objset_create_impl(dp->dp_spa,
340 		    ds, &ds->ds_phys->ds_bp, type, tx);
341 	}
342 
343 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
344 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
345 
346 	spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC,
347 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
348 	    ds->ds_phys->ds_dir_obj);
349 
350 	return (ds);
351 }
352 
353 /* ARGSUSED */
354 static int
355 recv_full_check(void *arg1, void *arg2, dmu_tx_t *tx)
356 {
357 	dsl_dir_t *dd = arg1;
358 	struct recvbeginsyncarg *rbsa = arg2;
359 	objset_t *mos = dd->dd_pool->dp_meta_objset;
360 	uint64_t val;
361 	int err;
362 
363 	err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
364 	    strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
365 
366 	if (err != ENOENT)
367 		return (err ? err : EEXIST);
368 
369 	if (rbsa->origin) {
370 		/* make sure it's a snap in the same pool */
371 		if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
372 			return (EXDEV);
373 		if (rbsa->origin->ds_phys->ds_num_children == 0)
374 			return (EINVAL);
375 		if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
376 			return (ENODEV);
377 	}
378 
379 	return (0);
380 }
381 
382 static void
383 recv_full_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
384 {
385 	dsl_dir_t *dd = arg1;
386 	struct recvbeginsyncarg *rbsa = arg2;
387 	uint64_t dsobj;
388 
389 	dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
390 	    rbsa->origin, cr, tx);
391 
392 	rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
393 	    rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
394 }
395 
396 static int
397 recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
398 {
399 	dsl_dataset_t *ds = arg1;
400 	struct recvbeginsyncarg *rbsa = arg2;
401 	int err;
402 
403 	/* must be a head ds */
404 	if (ds->ds_phys->ds_next_snap_obj != 0)
405 		return (EINVAL);
406 
407 	/* must not be a clone ds */
408 	if (ds->ds_prev != NULL)
409 		return (EINVAL);
410 
411 	err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
412 	if (err)
413 		return (err);
414 
415 	if (rbsa->origin) {
416 		/* make sure it's a snap in the same pool */
417 		if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool)
418 			return (EXDEV);
419 		if (rbsa->origin->ds_phys->ds_num_children == 0)
420 			return (EINVAL);
421 		if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
422 			return (ENODEV);
423 	}
424 
425 	return (0);
426 }
427 
428 static void
429 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
430 {
431 	dsl_dataset_t *ds = arg1;
432 	struct recvbeginsyncarg *rbsa = arg2;
433 	dsl_dir_t *dd = ds->ds_dir;
434 	uint64_t dsobj;
435 
436 	/*
437 	 * NB: caller must provide an extra hold on the dsl_dir_t, so it
438 	 * won't go away when dsl_dataset_destroy_sync() closes the
439 	 * dataset.
440 	 */
441 	dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
442 
443 	dsobj = dsl_dataset_create_sync_impl(dd, rbsa->origin, tx);
444 
445 	rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
446 	    rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
447 }
448 
449 /* ARGSUSED */
450 static int
451 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
452 {
453 	dsl_dataset_t *ds = arg1;
454 	struct recvbeginsyncarg *rbsa = arg2;
455 	int err;
456 	uint64_t val;
457 
458 	/* must not have any changes since most recent snapshot */
459 	if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
460 		return (ETXTBSY);
461 
462 	/* must already be a snapshot of this fs */
463 	if (ds->ds_phys->ds_prev_snap_obj == 0)
464 		return (ENODEV);
465 
466 	/* most recent snapshot must match fromguid */
467 	if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
468 		return (ENODEV);
469 
470 	/* temporary clone name must not exist */
471 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
472 	    ds->ds_dir->dd_phys->dd_child_dir_zapobj,
473 	    rbsa->clonelastname, 8, 1, &val);
474 	if (err == 0)
475 		return (EEXIST);
476 	if (err != ENOENT)
477 		return (err);
478 
479 	/* new snapshot name must not exist */
480 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
481 	    ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
482 	if (err == 0)
483 		return (EEXIST);
484 	if (err != ENOENT)
485 		return (err);
486 	return (0);
487 }
488 
489 /* ARGSUSED */
490 static void
491 recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
492 {
493 	dsl_dataset_t *ohds = arg1;
494 	struct recvbeginsyncarg *rbsa = arg2;
495 	dsl_pool_t *dp = ohds->ds_dir->dd_pool;
496 	dsl_dataset_t *ods, *cds;
497 	uint64_t dsobj;
498 
499 	/* create the temporary clone */
500 	VERIFY(0 == dsl_dataset_open_obj(dp, ohds->ds_phys->ds_prev_snap_obj,
501 	    NULL, DS_MODE_STANDARD, FTAG, &ods));
502 	dsobj = dsl_dataset_create_sync(ohds->ds_dir,
503 	    rbsa->clonelastname, ods, cr, tx);
504 	dsl_dataset_close(ods, DS_MODE_STANDARD, FTAG);
505 
506 	/* open the temporary clone */
507 	VERIFY(0 == dsl_dataset_open_obj(dp, dsobj, NULL,
508 	    DS_MODE_EXCLUSIVE, dmu_recv_tag, &cds));
509 
510 	/* copy the refquota from the target fs to the clone */
511 	if (ohds->ds_quota > 0)
512 		dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
513 
514 	dmu_buf_will_dirty(cds->ds_dbuf, tx);
515 	cds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
516 
517 	rbsa->ds = cds;
518 
519 	spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
520 	    dp->dp_spa, tx, cr, "dataset = %lld",
521 	    cds->ds_phys->ds_dir_obj);
522 }
523 
524 /* ARGSUSED */
525 static void
526 recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
527 {
528 	dsl_dataset_t *ds = arg1;
529 
530 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
531 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
532 
533 	spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
534 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
535 	    ds->ds_phys->ds_dir_obj);
536 }
537 
538 /*
539  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
540  * succeeds; otherwise we will leak the holds on the datasets.
541  */
542 int
543 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
544     boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
545 {
546 	int err = 0;
547 	boolean_t byteswap;
548 	struct recvbeginsyncarg rbsa;
549 	uint64_t version;
550 	int flags;
551 	dsl_dataset_t *ds;
552 
553 	if (drrb->drr_magic == DMU_BACKUP_MAGIC)
554 		byteswap = FALSE;
555 	else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
556 		byteswap = TRUE;
557 	else
558 		return (EINVAL);
559 
560 	rbsa.tofs = tofs;
561 	rbsa.tosnap = tosnap;
562 	rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL;
563 	rbsa.fromguid = drrb->drr_fromguid;
564 	rbsa.type = drrb->drr_type;
565 	rbsa.tag = FTAG;
566 	version = drrb->drr_version;
567 	flags = drrb->drr_flags;
568 
569 	if (byteswap) {
570 		rbsa.type = BSWAP_32(rbsa.type);
571 		rbsa.fromguid = BSWAP_64(rbsa.fromguid);
572 		version = BSWAP_64(version);
573 		flags = BSWAP_32(flags);
574 	}
575 
576 	if (version != DMU_BACKUP_STREAM_VERSION ||
577 	    rbsa.type >= DMU_OST_NUMTYPES ||
578 	    ((flags & DRR_FLAG_CLONE) && origin == NULL))
579 		return (EINVAL);
580 
581 	bzero(drc, sizeof (dmu_recv_cookie_t));
582 	drc->drc_drrb = drrb;
583 	drc->drc_tosnap = tosnap;
584 	drc->drc_force = force;
585 
586 	/*
587 	 * Process the begin in syncing context.
588 	 */
589 	if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
590 		/* offline incremental receive */
591 		err = dsl_dataset_open(tofs,
592 		    DS_MODE_EXCLUSIVE, dmu_recv_tag, &ds);
593 		if (err)
594 			return (err);
595 
596 		/*
597 		 * Only do the rollback if the most recent snapshot
598 		 * matches the incremental source
599 		 */
600 		if (force) {
601 			if (ds->ds_prev == NULL ||
602 			    ds->ds_prev->ds_phys->ds_guid !=
603 			    rbsa.fromguid) {
604 				dsl_dataset_close(ds, DS_MODE_EXCLUSIVE,
605 				    dmu_recv_tag);
606 				return (ENODEV);
607 			}
608 			(void) dsl_dataset_rollback(ds, DMU_OST_NONE);
609 		}
610 		rbsa.force = B_FALSE;
611 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
612 		    recv_incremental_check,
613 		    recv_offline_incremental_sync,
614 		    ds, &rbsa, 1);
615 		if (err) {
616 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, dmu_recv_tag);
617 			return (err);
618 		}
619 		drc->drc_logical_ds = drc->drc_real_ds = ds;
620 	} else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
621 		/* online incremental receive */
622 
623 		/* tmp clone name is: tofs/%tosnap" */
624 		(void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
625 		    "%%%s", tosnap);
626 
627 		/* open the dataset we are logically receiving into */
628 		err = dsl_dataset_open(tofs,
629 		    DS_MODE_STANDARD, dmu_recv_tag, &ds);
630 		if (err)
631 			return (err);
632 
633 		rbsa.force = force;
634 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
635 		    recv_incremental_check,
636 		    recv_online_incremental_sync, ds, &rbsa, 5);
637 		if (err) {
638 			dsl_dataset_close(ds, DS_MODE_STANDARD, dmu_recv_tag);
639 			return (err);
640 		}
641 		drc->drc_logical_ds = ds;
642 		drc->drc_real_ds = rbsa.ds;
643 	} else {
644 		/* create new fs -- full backup or clone */
645 		dsl_dir_t *dd = NULL;
646 		const char *tail;
647 
648 		err = dsl_dir_open(tofs, FTAG, &dd, &tail);
649 		if (err)
650 			return (err);
651 		if (tail == NULL) {
652 			if (!force) {
653 				dsl_dir_close(dd, FTAG);
654 				return (EEXIST);
655 			}
656 
657 			rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
658 			err = dsl_dataset_open_obj(dd->dd_pool,
659 			    dd->dd_phys->dd_head_dataset_obj, NULL,
660 			    DS_MODE_EXCLUSIVE | DS_MODE_INCONSISTENT,
661 			    FTAG, &ds);
662 			rw_exit(&dd->dd_pool->dp_config_rwlock);
663 			if (err) {
664 				dsl_dir_close(dd, FTAG);
665 				return (err);
666 			}
667 
668 			err = dsl_sync_task_do(dd->dd_pool,
669 			    recv_full_existing_check,
670 			    recv_full_existing_sync, ds, &rbsa, 5);
671 			/* if successful, sync task closes the ds for us */
672 			if (err)
673 				dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
674 		} else {
675 			err = dsl_sync_task_do(dd->dd_pool, recv_full_check,
676 			    recv_full_sync, dd, &rbsa, 5);
677 			if (err)
678 				return (err);
679 		}
680 		dsl_dir_close(dd, FTAG);
681 		if (err)
682 			return (err);
683 		drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
684 		drc->drc_newfs = B_TRUE;
685 	}
686 
687 	/* downgrade our hold on the ds from EXCLUSIVE to PRIMARY */
688 	dsl_dataset_downgrade(drc->drc_real_ds,
689 	    DS_MODE_EXCLUSIVE, DS_MODE_PRIMARY);
690 
691 	return (0);
692 }
693 
694 struct restorearg {
695 	int err;
696 	int byteswap;
697 	vnode_t *vp;
698 	char *buf;
699 	uint64_t voff;
700 	int bufsize; /* amount of memory allocated for buf */
701 	zio_cksum_t cksum;
702 };
703 
704 static void *
705 restore_read(struct restorearg *ra, int len)
706 {
707 	void *rv;
708 	int done = 0;
709 
710 	/* some things will require 8-byte alignment, so everything must */
711 	ASSERT3U(len % 8, ==, 0);
712 
713 	while (done < len) {
714 		ssize_t resid;
715 
716 		ra->err = vn_rdwr(UIO_READ, ra->vp,
717 		    (caddr_t)ra->buf + done, len - done,
718 		    ra->voff, UIO_SYSSPACE, FAPPEND,
719 		    RLIM64_INFINITY, CRED(), &resid);
720 
721 		if (resid == len - done)
722 			ra->err = EINVAL;
723 		ra->voff += len - done - resid;
724 		done = len - resid;
725 		if (ra->err)
726 			return (NULL);
727 	}
728 
729 	ASSERT3U(done, ==, len);
730 	rv = ra->buf;
731 	if (ra->byteswap)
732 		fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
733 	else
734 		fletcher_4_incremental_native(rv, len, &ra->cksum);
735 	return (rv);
736 }
737 
738 static void
739 backup_byteswap(dmu_replay_record_t *drr)
740 {
741 #define	DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
742 #define	DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
743 	drr->drr_type = BSWAP_32(drr->drr_type);
744 	drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
745 	switch (drr->drr_type) {
746 	case DRR_BEGIN:
747 		DO64(drr_begin.drr_magic);
748 		DO64(drr_begin.drr_version);
749 		DO64(drr_begin.drr_creation_time);
750 		DO32(drr_begin.drr_type);
751 		DO32(drr_begin.drr_flags);
752 		DO64(drr_begin.drr_toguid);
753 		DO64(drr_begin.drr_fromguid);
754 		break;
755 	case DRR_OBJECT:
756 		DO64(drr_object.drr_object);
757 		/* DO64(drr_object.drr_allocation_txg); */
758 		DO32(drr_object.drr_type);
759 		DO32(drr_object.drr_bonustype);
760 		DO32(drr_object.drr_blksz);
761 		DO32(drr_object.drr_bonuslen);
762 		break;
763 	case DRR_FREEOBJECTS:
764 		DO64(drr_freeobjects.drr_firstobj);
765 		DO64(drr_freeobjects.drr_numobjs);
766 		break;
767 	case DRR_WRITE:
768 		DO64(drr_write.drr_object);
769 		DO32(drr_write.drr_type);
770 		DO64(drr_write.drr_offset);
771 		DO64(drr_write.drr_length);
772 		break;
773 	case DRR_FREE:
774 		DO64(drr_free.drr_object);
775 		DO64(drr_free.drr_offset);
776 		DO64(drr_free.drr_length);
777 		break;
778 	case DRR_END:
779 		DO64(drr_end.drr_checksum.zc_word[0]);
780 		DO64(drr_end.drr_checksum.zc_word[1]);
781 		DO64(drr_end.drr_checksum.zc_word[2]);
782 		DO64(drr_end.drr_checksum.zc_word[3]);
783 		break;
784 	}
785 #undef DO64
786 #undef DO32
787 }
788 
789 static int
790 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
791 {
792 	int err;
793 	dmu_tx_t *tx;
794 
795 	err = dmu_object_info(os, drro->drr_object, NULL);
796 
797 	if (err != 0 && err != ENOENT)
798 		return (EINVAL);
799 
800 	if (drro->drr_type == DMU_OT_NONE ||
801 	    drro->drr_type >= DMU_OT_NUMTYPES ||
802 	    drro->drr_bonustype >= DMU_OT_NUMTYPES ||
803 	    drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
804 	    drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
805 	    P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
806 	    drro->drr_blksz < SPA_MINBLOCKSIZE ||
807 	    drro->drr_blksz > SPA_MAXBLOCKSIZE ||
808 	    drro->drr_bonuslen > DN_MAX_BONUSLEN) {
809 		return (EINVAL);
810 	}
811 
812 	tx = dmu_tx_create(os);
813 
814 	if (err == ENOENT) {
815 		/* currently free, want to be allocated */
816 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
817 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1);
818 		err = dmu_tx_assign(tx, TXG_WAIT);
819 		if (err) {
820 			dmu_tx_abort(tx);
821 			return (err);
822 		}
823 		err = dmu_object_claim(os, drro->drr_object,
824 		    drro->drr_type, drro->drr_blksz,
825 		    drro->drr_bonustype, drro->drr_bonuslen, tx);
826 	} else {
827 		/* currently allocated, want to be allocated */
828 		dmu_tx_hold_bonus(tx, drro->drr_object);
829 		/*
830 		 * We may change blocksize, so need to
831 		 * hold_write
832 		 */
833 		dmu_tx_hold_write(tx, drro->drr_object, 0, 1);
834 		err = dmu_tx_assign(tx, TXG_WAIT);
835 		if (err) {
836 			dmu_tx_abort(tx);
837 			return (err);
838 		}
839 
840 		err = dmu_object_reclaim(os, drro->drr_object,
841 		    drro->drr_type, drro->drr_blksz,
842 		    drro->drr_bonustype, drro->drr_bonuslen, tx);
843 	}
844 	if (err) {
845 		dmu_tx_commit(tx);
846 		return (EINVAL);
847 	}
848 
849 	dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
850 	dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
851 
852 	if (drro->drr_bonuslen) {
853 		dmu_buf_t *db;
854 		void *data;
855 		VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
856 		dmu_buf_will_dirty(db, tx);
857 
858 		ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
859 		data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
860 		if (data == NULL) {
861 			dmu_tx_commit(tx);
862 			return (ra->err);
863 		}
864 		bcopy(data, db->db_data, drro->drr_bonuslen);
865 		if (ra->byteswap) {
866 			dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
867 			    drro->drr_bonuslen);
868 		}
869 		dmu_buf_rele(db, FTAG);
870 	}
871 	dmu_tx_commit(tx);
872 	return (0);
873 }
874 
875 /* ARGSUSED */
876 static int
877 restore_freeobjects(struct restorearg *ra, objset_t *os,
878     struct drr_freeobjects *drrfo)
879 {
880 	uint64_t obj;
881 
882 	if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
883 		return (EINVAL);
884 
885 	for (obj = drrfo->drr_firstobj;
886 	    obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
887 	    (void) dmu_object_next(os, &obj, FALSE, 0)) {
888 		dmu_tx_t *tx;
889 		int err;
890 
891 		if (dmu_object_info(os, obj, NULL) != 0)
892 			continue;
893 
894 		tx = dmu_tx_create(os);
895 		dmu_tx_hold_bonus(tx, obj);
896 		err = dmu_tx_assign(tx, TXG_WAIT);
897 		if (err) {
898 			dmu_tx_abort(tx);
899 			return (err);
900 		}
901 		err = dmu_object_free(os, obj, tx);
902 		dmu_tx_commit(tx);
903 		if (err && err != ENOENT)
904 			return (EINVAL);
905 	}
906 	return (0);
907 }
908 
909 static int
910 restore_write(struct restorearg *ra, objset_t *os,
911     struct drr_write *drrw)
912 {
913 	dmu_tx_t *tx;
914 	void *data;
915 	int err;
916 
917 	if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
918 	    drrw->drr_type >= DMU_OT_NUMTYPES)
919 		return (EINVAL);
920 
921 	data = restore_read(ra, drrw->drr_length);
922 	if (data == NULL)
923 		return (ra->err);
924 
925 	if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
926 		return (EINVAL);
927 
928 	tx = dmu_tx_create(os);
929 
930 	dmu_tx_hold_write(tx, drrw->drr_object,
931 	    drrw->drr_offset, drrw->drr_length);
932 	err = dmu_tx_assign(tx, TXG_WAIT);
933 	if (err) {
934 		dmu_tx_abort(tx);
935 		return (err);
936 	}
937 	if (ra->byteswap)
938 		dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
939 	dmu_write(os, drrw->drr_object,
940 	    drrw->drr_offset, drrw->drr_length, data, tx);
941 	dmu_tx_commit(tx);
942 	return (0);
943 }
944 
945 /* ARGSUSED */
946 static int
947 restore_free(struct restorearg *ra, objset_t *os,
948     struct drr_free *drrf)
949 {
950 	dmu_tx_t *tx;
951 	int err;
952 
953 	if (drrf->drr_length != -1ULL &&
954 	    drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
955 		return (EINVAL);
956 
957 	if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
958 		return (EINVAL);
959 
960 	tx = dmu_tx_create(os);
961 
962 	dmu_tx_hold_free(tx, drrf->drr_object,
963 	    drrf->drr_offset, drrf->drr_length);
964 	err = dmu_tx_assign(tx, TXG_WAIT);
965 	if (err) {
966 		dmu_tx_abort(tx);
967 		return (err);
968 	}
969 	err = dmu_free_range(os, drrf->drr_object,
970 	    drrf->drr_offset, drrf->drr_length, tx);
971 	dmu_tx_commit(tx);
972 	return (err);
973 }
974 
975 void
976 dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
977 {
978 	if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
979 		/*
980 		 * online incremental or new fs: destroy the fs (which
981 		 * may be a clone) that we created
982 		 */
983 		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
984 		if (drc->drc_real_ds != drc->drc_logical_ds) {
985 			dsl_dataset_close(drc->drc_logical_ds,
986 			    DS_MODE_STANDARD, dmu_recv_tag);
987 		}
988 	} else {
989 		/*
990 		 * offline incremental: rollback to most recent snapshot.
991 		 */
992 		int lmode = DS_MODE_PRIMARY;
993 		if (dsl_dataset_tryupgrade(drc->drc_real_ds,
994 		    DS_MODE_PRIMARY, DS_MODE_EXCLUSIVE)) {
995 			lmode = DS_MODE_EXCLUSIVE;
996 			(void) dsl_dataset_rollback(drc->drc_real_ds,
997 			    DMU_OST_NONE);
998 		}
999 		dsl_dataset_close(drc->drc_real_ds, lmode, FTAG);
1000 	}
1001 }
1002 
1003 /*
1004  * NB: callers *must* call dmu_recv_end() if this succeeds.
1005  */
1006 int
1007 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp)
1008 {
1009 	struct restorearg ra = { 0 };
1010 	dmu_replay_record_t *drr;
1011 	objset_t *os;
1012 	zio_cksum_t pcksum;
1013 
1014 	if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1015 		ra.byteswap = TRUE;
1016 
1017 	{
1018 		/* compute checksum of drr_begin record */
1019 		dmu_replay_record_t *drr;
1020 		drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1021 
1022 		drr->drr_type = DRR_BEGIN;
1023 		drr->drr_u.drr_begin = *drc->drc_drrb;
1024 		if (ra.byteswap) {
1025 			fletcher_4_incremental_byteswap(drr,
1026 			    sizeof (dmu_replay_record_t), &ra.cksum);
1027 		} else {
1028 			fletcher_4_incremental_native(drr,
1029 			    sizeof (dmu_replay_record_t), &ra.cksum);
1030 		}
1031 		kmem_free(drr, sizeof (dmu_replay_record_t));
1032 	}
1033 
1034 	if (ra.byteswap) {
1035 		struct drr_begin *drrb = drc->drc_drrb;
1036 		drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1037 		drrb->drr_version = BSWAP_64(drrb->drr_version);
1038 		drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1039 		drrb->drr_type = BSWAP_32(drrb->drr_type);
1040 		drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1041 		drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1042 	}
1043 
1044 	ra.vp = vp;
1045 	ra.voff = *voffp;
1046 	ra.bufsize = 1<<20;
1047 	ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1048 
1049 	/* these were verified in dmu_recv_begin */
1050 	ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
1051 	ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1052 
1053 	/*
1054 	 * Open the objset we are modifying.
1055 	 */
1056 	VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0);
1057 
1058 	ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1059 
1060 	/*
1061 	 * Read records and process them.
1062 	 */
1063 	pcksum = ra.cksum;
1064 	while (ra.err == 0 &&
1065 	    NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1066 		if (issig(JUSTLOOKING) && issig(FORREAL)) {
1067 			ra.err = EINTR;
1068 			goto out;
1069 		}
1070 
1071 		if (ra.byteswap)
1072 			backup_byteswap(drr);
1073 
1074 		switch (drr->drr_type) {
1075 		case DRR_OBJECT:
1076 		{
1077 			/*
1078 			 * We need to make a copy of the record header,
1079 			 * because restore_{object,write} may need to
1080 			 * restore_read(), which will invalidate drr.
1081 			 */
1082 			struct drr_object drro = drr->drr_u.drr_object;
1083 			ra.err = restore_object(&ra, os, &drro);
1084 			break;
1085 		}
1086 		case DRR_FREEOBJECTS:
1087 		{
1088 			struct drr_freeobjects drrfo =
1089 			    drr->drr_u.drr_freeobjects;
1090 			ra.err = restore_freeobjects(&ra, os, &drrfo);
1091 			break;
1092 		}
1093 		case DRR_WRITE:
1094 		{
1095 			struct drr_write drrw = drr->drr_u.drr_write;
1096 			ra.err = restore_write(&ra, os, &drrw);
1097 			break;
1098 		}
1099 		case DRR_FREE:
1100 		{
1101 			struct drr_free drrf = drr->drr_u.drr_free;
1102 			ra.err = restore_free(&ra, os, &drrf);
1103 			break;
1104 		}
1105 		case DRR_END:
1106 		{
1107 			struct drr_end drre = drr->drr_u.drr_end;
1108 			/*
1109 			 * We compare against the *previous* checksum
1110 			 * value, because the stored checksum is of
1111 			 * everything before the DRR_END record.
1112 			 */
1113 			if (drre.drr_checksum.zc_word[0] != 0 &&
1114 			    !ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum)) {
1115 				ra.err = ECKSUM;
1116 				goto out;
1117 			}
1118 			goto out;
1119 		}
1120 		default:
1121 			ra.err = EINVAL;
1122 			goto out;
1123 		}
1124 		pcksum = ra.cksum;
1125 	}
1126 
1127 out:
1128 	dmu_objset_close(os);
1129 
1130 	if (ra.err != 0) {
1131 		/*
1132 		 * rollback or destroy what we created, so we don't
1133 		 * leave it in the restoring state.
1134 		 */
1135 		txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1136 		dmu_recv_abort_cleanup(drc);
1137 	}
1138 
1139 	kmem_free(ra.buf, ra.bufsize);
1140 	*voffp = ra.voff;
1141 	return (ra.err);
1142 }
1143 
1144 struct recvendsyncarg {
1145 	char *tosnap;
1146 	uint64_t creation_time;
1147 	uint64_t toguid;
1148 };
1149 
1150 static int
1151 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1152 {
1153 	dsl_dataset_t *ds = arg1;
1154 	struct recvendsyncarg *resa = arg2;
1155 
1156 	return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1157 }
1158 
1159 static void
1160 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1161 {
1162 	dsl_dataset_t *ds = arg1;
1163 	struct recvendsyncarg *resa = arg2;
1164 
1165 	dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
1166 
1167 	/* set snapshot's creation time and guid */
1168 	dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1169 	ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1170 	ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1171 	ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1172 
1173 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1174 	ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1175 }
1176 
1177 int
1178 dmu_recv_end(dmu_recv_cookie_t *drc)
1179 {
1180 	int err = 0;
1181 	int lmode;
1182 
1183 	/*
1184 	 * XXX hack; seems the ds is still dirty and
1185 	 * dsl_pool_zil_clean() expects it to have a ds_user_ptr (and
1186 	 * zil), but clone_swap() can close it.
1187 	 */
1188 	txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1189 
1190 	if (dsl_dataset_tryupgrade(drc->drc_real_ds,
1191 	    DS_MODE_PRIMARY, DS_MODE_EXCLUSIVE)) {
1192 		lmode = DS_MODE_EXCLUSIVE;
1193 	} else {
1194 		dmu_recv_abort_cleanup(drc);
1195 		return (EBUSY);
1196 	}
1197 
1198 	if (drc->drc_logical_ds != drc->drc_real_ds) {
1199 		if (err == 0 && dsl_dataset_tryupgrade(drc->drc_logical_ds,
1200 		    DS_MODE_STANDARD, DS_MODE_EXCLUSIVE)) {
1201 			lmode = DS_MODE_EXCLUSIVE;
1202 			err = dsl_dataset_clone_swap(drc->drc_real_ds,
1203 			    drc->drc_logical_ds, drc->drc_force);
1204 		} else {
1205 			lmode = DS_MODE_STANDARD;
1206 			err = EBUSY;
1207 		}
1208 	}
1209 
1210 	if (err == 0) {
1211 		struct recvendsyncarg resa;
1212 
1213 		resa.creation_time = drc->drc_drrb->drr_creation_time;
1214 		resa.toguid = drc->drc_drrb->drr_toguid;
1215 		resa.tosnap = drc->drc_tosnap;
1216 
1217 		err = dsl_sync_task_do(drc->drc_real_ds->ds_dir->dd_pool,
1218 		    recv_end_check, recv_end_sync,
1219 		    drc->drc_logical_ds, &resa, 3);
1220 		if (err) {
1221 			if (drc->drc_newfs) {
1222 				ASSERT(drc->drc_logical_ds == drc->drc_real_ds);
1223 				(void) dsl_dataset_destroy(drc->drc_real_ds,
1224 				    dmu_recv_tag);
1225 				return (err);
1226 			} else {
1227 				(void) dsl_dataset_rollback(drc->drc_logical_ds,
1228 				    DMU_OST_NONE);
1229 			}
1230 		}
1231 	}
1232 
1233 	if (drc->drc_logical_ds != drc->drc_real_ds) {
1234 		/* dsl_dataset_destroy() will close the ds */
1235 		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
1236 	}
1237 	/* close the hold from dmu_recv_begin */
1238 	dsl_dataset_close(drc->drc_logical_ds, lmode, dmu_recv_tag);
1239 	return (err);
1240 }
1241