xref: /titanic_44/usr/src/uts/common/fs/zfs/dmu_send.c (revision 75ce41a57ff334bd8fe2cb9ed51eea835892f944)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/dmu.h>
27 #include <sys/dmu_impl.h>
28 #include <sys/dmu_tx.h>
29 #include <sys/dbuf.h>
30 #include <sys/dnode.h>
31 #include <sys/zfs_context.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/zfs_ioctl.h>
39 #include <sys/zap.h>
40 #include <sys/zio_checksum.h>
41 
42 static char *dmu_recv_tag = "dmu_recv_tag";
43 
44 struct backuparg {
45 	dmu_replay_record_t *drr;
46 	vnode_t *vp;
47 	offset_t *off;
48 	objset_t *os;
49 	zio_cksum_t zc;
50 	int err;
51 };
52 
53 static int
54 dump_bytes(struct backuparg *ba, void *buf, int len)
55 {
56 	ssize_t resid; /* have to get resid to get detailed errno */
57 	ASSERT3U(len % 8, ==, 0);
58 
59 	fletcher_4_incremental_native(buf, len, &ba->zc);
60 	ba->err = vn_rdwr(UIO_WRITE, ba->vp,
61 	    (caddr_t)buf, len,
62 	    0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
63 	*ba->off += len;
64 	return (ba->err);
65 }
66 
67 static int
68 dump_free(struct backuparg *ba, uint64_t object, uint64_t offset,
69     uint64_t length)
70 {
71 	/* write a FREE record */
72 	bzero(ba->drr, sizeof (dmu_replay_record_t));
73 	ba->drr->drr_type = DRR_FREE;
74 	ba->drr->drr_u.drr_free.drr_object = object;
75 	ba->drr->drr_u.drr_free.drr_offset = offset;
76 	ba->drr->drr_u.drr_free.drr_length = length;
77 
78 	if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
79 		return (EINTR);
80 	return (0);
81 }
82 
83 static int
84 dump_data(struct backuparg *ba, dmu_object_type_t type,
85     uint64_t object, uint64_t offset, int blksz, void *data)
86 {
87 	/* write a DATA record */
88 	bzero(ba->drr, sizeof (dmu_replay_record_t));
89 	ba->drr->drr_type = DRR_WRITE;
90 	ba->drr->drr_u.drr_write.drr_object = object;
91 	ba->drr->drr_u.drr_write.drr_type = type;
92 	ba->drr->drr_u.drr_write.drr_offset = offset;
93 	ba->drr->drr_u.drr_write.drr_length = blksz;
94 
95 	if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
96 		return (EINTR);
97 	if (dump_bytes(ba, data, blksz))
98 		return (EINTR);
99 	return (0);
100 }
101 
102 static int
103 dump_freeobjects(struct backuparg *ba, uint64_t firstobj, uint64_t numobjs)
104 {
105 	/* write a FREEOBJECTS record */
106 	bzero(ba->drr, sizeof (dmu_replay_record_t));
107 	ba->drr->drr_type = DRR_FREEOBJECTS;
108 	ba->drr->drr_u.drr_freeobjects.drr_firstobj = firstobj;
109 	ba->drr->drr_u.drr_freeobjects.drr_numobjs = numobjs;
110 
111 	if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
112 		return (EINTR);
113 	return (0);
114 }
115 
116 static int
117 dump_dnode(struct backuparg *ba, uint64_t object, dnode_phys_t *dnp)
118 {
119 	if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
120 		return (dump_freeobjects(ba, object, 1));
121 
122 	/* write an OBJECT record */
123 	bzero(ba->drr, sizeof (dmu_replay_record_t));
124 	ba->drr->drr_type = DRR_OBJECT;
125 	ba->drr->drr_u.drr_object.drr_object = object;
126 	ba->drr->drr_u.drr_object.drr_type = dnp->dn_type;
127 	ba->drr->drr_u.drr_object.drr_bonustype = dnp->dn_bonustype;
128 	ba->drr->drr_u.drr_object.drr_blksz =
129 	    dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
130 	ba->drr->drr_u.drr_object.drr_bonuslen = dnp->dn_bonuslen;
131 	ba->drr->drr_u.drr_object.drr_checksum = dnp->dn_checksum;
132 	ba->drr->drr_u.drr_object.drr_compress = dnp->dn_compress;
133 
134 	if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
135 		return (EINTR);
136 
137 	if (dump_bytes(ba, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)))
138 		return (EINTR);
139 
140 	/* free anything past the end of the file */
141 	if (dump_free(ba, object, (dnp->dn_maxblkid + 1) *
142 	    (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
143 		return (EINTR);
144 	if (ba->err)
145 		return (EINTR);
146 	return (0);
147 }
148 
149 #define	BP_SPAN(dnp, level) \
150 	(((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
151 	(level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
152 
153 static int
154 backup_cb(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
155     const dnode_phys_t *dnp, void *arg)
156 {
157 	struct backuparg *ba = arg;
158 	dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
159 	int err = 0;
160 
161 	if (issig(JUSTLOOKING) && issig(FORREAL))
162 		return (EINTR);
163 
164 	if (zb->zb_object != 0 && DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
165 		return (0);
166 	} else if (bp == NULL && zb->zb_object == 0) {
167 		uint64_t span = BP_SPAN(dnp, zb->zb_level);
168 		uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
169 		err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
170 	} else if (bp == NULL) {
171 		uint64_t span = BP_SPAN(dnp, zb->zb_level);
172 		err = dump_free(ba, zb->zb_object, zb->zb_blkid * span, span);
173 	} else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
174 		return (0);
175 	} else if (type == DMU_OT_DNODE) {
176 		dnode_phys_t *blk;
177 		int i;
178 		int blksz = BP_GET_LSIZE(bp);
179 		uint32_t aflags = ARC_WAIT;
180 		arc_buf_t *abuf;
181 
182 		if (arc_read_nolock(NULL, spa, bp,
183 		    arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
184 		    ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
185 			return (EIO);
186 
187 		blk = abuf->b_data;
188 		for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
189 			uint64_t dnobj = (zb->zb_blkid <<
190 			    (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
191 			err = dump_dnode(ba, dnobj, blk+i);
192 			if (err)
193 				break;
194 		}
195 		(void) arc_buf_remove_ref(abuf, &abuf);
196 	} else { /* it's a level-0 block of a regular object */
197 		uint32_t aflags = ARC_WAIT;
198 		arc_buf_t *abuf;
199 		int blksz = BP_GET_LSIZE(bp);
200 
201 		if (arc_read_nolock(NULL, spa, bp,
202 		    arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
203 		    ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
204 			return (EIO);
205 
206 		err = dump_data(ba, type, zb->zb_object, zb->zb_blkid * blksz,
207 		    blksz, abuf->b_data);
208 		(void) arc_buf_remove_ref(abuf, &abuf);
209 	}
210 
211 	ASSERT(err == 0 || err == EINTR);
212 	return (err);
213 }
214 
215 int
216 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
217     vnode_t *vp, offset_t *off)
218 {
219 	dsl_dataset_t *ds = tosnap->os_dsl_dataset;
220 	dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
221 	dmu_replay_record_t *drr;
222 	struct backuparg ba;
223 	int err;
224 	uint64_t fromtxg = 0;
225 
226 	/* tosnap must be a snapshot */
227 	if (ds->ds_phys->ds_next_snap_obj == 0)
228 		return (EINVAL);
229 
230 	/* fromsnap must be an earlier snapshot from the same fs as tosnap */
231 	if (fromds && (ds->ds_dir != fromds->ds_dir ||
232 	    fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
233 		return (EXDEV);
234 
235 	if (fromorigin) {
236 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
237 
238 		if (fromsnap)
239 			return (EINVAL);
240 
241 		if (dsl_dir_is_clone(ds->ds_dir)) {
242 			rw_enter(&dp->dp_config_rwlock, RW_READER);
243 			err = dsl_dataset_hold_obj(dp,
244 			    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
245 			rw_exit(&dp->dp_config_rwlock);
246 			if (err)
247 				return (err);
248 		} else {
249 			fromorigin = B_FALSE;
250 		}
251 	}
252 
253 
254 	drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
255 	drr->drr_type = DRR_BEGIN;
256 	drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
257 	drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION;
258 	drr->drr_u.drr_begin.drr_creation_time =
259 	    ds->ds_phys->ds_creation_time;
260 	drr->drr_u.drr_begin.drr_type = tosnap->os_phys->os_type;
261 	if (fromorigin)
262 		drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
263 	drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
264 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
265 		drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
266 
267 	if (fromds)
268 		drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
269 	dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
270 
271 	if (fromds)
272 		fromtxg = fromds->ds_phys->ds_creation_txg;
273 	if (fromorigin)
274 		dsl_dataset_rele(fromds, FTAG);
275 
276 	ba.drr = drr;
277 	ba.vp = vp;
278 	ba.os = tosnap;
279 	ba.off = off;
280 	ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
281 
282 	if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
283 		kmem_free(drr, sizeof (dmu_replay_record_t));
284 		return (ba.err);
285 	}
286 
287 	err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
288 	    backup_cb, &ba);
289 
290 	if (err) {
291 		if (err == EINTR && ba.err)
292 			err = ba.err;
293 		kmem_free(drr, sizeof (dmu_replay_record_t));
294 		return (err);
295 	}
296 
297 	bzero(drr, sizeof (dmu_replay_record_t));
298 	drr->drr_type = DRR_END;
299 	drr->drr_u.drr_end.drr_checksum = ba.zc;
300 
301 	if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
302 		kmem_free(drr, sizeof (dmu_replay_record_t));
303 		return (ba.err);
304 	}
305 
306 	kmem_free(drr, sizeof (dmu_replay_record_t));
307 
308 	return (0);
309 }
310 
311 struct recvbeginsyncarg {
312 	const char *tofs;
313 	const char *tosnap;
314 	dsl_dataset_t *origin;
315 	uint64_t fromguid;
316 	dmu_objset_type_t type;
317 	void *tag;
318 	boolean_t force;
319 	uint64_t dsflags;
320 	char clonelastname[MAXNAMELEN];
321 	dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
322 };
323 
324 /* ARGSUSED */
325 static int
326 recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx)
327 {
328 	dsl_dir_t *dd = arg1;
329 	struct recvbeginsyncarg *rbsa = arg2;
330 	objset_t *mos = dd->dd_pool->dp_meta_objset;
331 	uint64_t val;
332 	int err;
333 
334 	err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
335 	    strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
336 
337 	if (err != ENOENT)
338 		return (err ? err : EEXIST);
339 
340 	if (rbsa->origin) {
341 		/* make sure it's a snap in the same pool */
342 		if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
343 			return (EXDEV);
344 		if (!dsl_dataset_is_snapshot(rbsa->origin))
345 			return (EINVAL);
346 		if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
347 			return (ENODEV);
348 	}
349 
350 	return (0);
351 }
352 
353 static void
354 recv_new_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
355 {
356 	dsl_dir_t *dd = arg1;
357 	struct recvbeginsyncarg *rbsa = arg2;
358 	uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
359 	uint64_t dsobj;
360 
361 	/* Create and open new dataset. */
362 	dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
363 	    rbsa->origin, flags, cr, tx);
364 	VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj,
365 	    B_TRUE, dmu_recv_tag, &rbsa->ds));
366 
367 	if (rbsa->origin == NULL) {
368 		(void) dmu_objset_create_impl(dd->dd_pool->dp_spa,
369 		    rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx);
370 	}
371 
372 	spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC,
373 	    dd->dd_pool->dp_spa, tx, cr, "dataset = %lld", dsobj);
374 }
375 
376 /* ARGSUSED */
377 static int
378 recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
379 {
380 	dsl_dataset_t *ds = arg1;
381 	struct recvbeginsyncarg *rbsa = arg2;
382 	int err;
383 	uint64_t val;
384 
385 	/* must not have any changes since most recent snapshot */
386 	if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
387 		return (ETXTBSY);
388 
389 	if (rbsa->fromguid) {
390 		/* if incremental, most recent snapshot must match fromguid */
391 		if (ds->ds_prev == NULL)
392 			return (ENODEV);
393 		if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
394 			return (ENODEV);
395 	} else {
396 		/* if full, most recent snapshot must be $ORIGIN */
397 		if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
398 			return (ENODEV);
399 	}
400 
401 	/* temporary clone name must not exist */
402 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
403 	    ds->ds_dir->dd_phys->dd_child_dir_zapobj,
404 	    rbsa->clonelastname, 8, 1, &val);
405 	if (err == 0)
406 		return (EEXIST);
407 	if (err != ENOENT)
408 		return (err);
409 
410 	/* new snapshot name must not exist */
411 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
412 	    ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
413 	if (err == 0)
414 		return (EEXIST);
415 	if (err != ENOENT)
416 		return (err);
417 	return (0);
418 }
419 
420 /* ARGSUSED */
421 static void
422 recv_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
423 {
424 	dsl_dataset_t *ohds = arg1;
425 	struct recvbeginsyncarg *rbsa = arg2;
426 	dsl_pool_t *dp = ohds->ds_dir->dd_pool;
427 	dsl_dataset_t *cds;
428 	uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
429 	uint64_t dsobj;
430 
431 	/* create and open the temporary clone */
432 	dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname,
433 	    ohds->ds_prev, flags, cr, tx);
434 	VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, B_TRUE, dmu_recv_tag, &cds));
435 
436 	/*
437 	 * If we actually created a non-clone, we need to create the
438 	 * objset in our new dataset.
439 	 */
440 	if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) {
441 		(void) dmu_objset_create_impl(dp->dp_spa,
442 		    cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx);
443 	}
444 
445 	/* copy the refquota from the target fs to the clone */
446 	if (ohds->ds_quota > 0)
447 		dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
448 
449 	rbsa->ds = cds;
450 
451 	spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
452 	    dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
453 }
454 
455 /*
456  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
457  * succeeds; otherwise we will leak the holds on the datasets.
458  */
459 int
460 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
461     boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
462 {
463 	int err = 0;
464 	boolean_t byteswap;
465 	struct recvbeginsyncarg rbsa = { 0 };
466 	uint64_t version;
467 	int flags;
468 	dsl_dataset_t *ds;
469 
470 	if (drrb->drr_magic == DMU_BACKUP_MAGIC)
471 		byteswap = FALSE;
472 	else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
473 		byteswap = TRUE;
474 	else
475 		return (EINVAL);
476 
477 	rbsa.tofs = tofs;
478 	rbsa.tosnap = tosnap;
479 	rbsa.origin = origin ? origin->os_dsl_dataset : NULL;
480 	rbsa.fromguid = drrb->drr_fromguid;
481 	rbsa.type = drrb->drr_type;
482 	rbsa.tag = FTAG;
483 	rbsa.dsflags = 0;
484 	version = drrb->drr_version;
485 	flags = drrb->drr_flags;
486 
487 	if (byteswap) {
488 		rbsa.type = BSWAP_32(rbsa.type);
489 		rbsa.fromguid = BSWAP_64(rbsa.fromguid);
490 		version = BSWAP_64(version);
491 		flags = BSWAP_32(flags);
492 	}
493 
494 	if (version != DMU_BACKUP_STREAM_VERSION ||
495 	    rbsa.type >= DMU_OST_NUMTYPES ||
496 	    ((flags & DRR_FLAG_CLONE) && origin == NULL))
497 		return (EINVAL);
498 
499 	if (flags & DRR_FLAG_CI_DATA)
500 		rbsa.dsflags = DS_FLAG_CI_DATASET;
501 
502 	bzero(drc, sizeof (dmu_recv_cookie_t));
503 	drc->drc_drrb = drrb;
504 	drc->drc_tosnap = tosnap;
505 	drc->drc_force = force;
506 
507 	/*
508 	 * Process the begin in syncing context.
509 	 */
510 
511 	/* open the dataset we are logically receiving into */
512 	err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
513 	if (err == 0) {
514 		/* target fs already exists; recv into temp clone */
515 
516 		/* Can't recv a clone into an existing fs */
517 		if (flags & DRR_FLAG_CLONE) {
518 			dsl_dataset_rele(ds, dmu_recv_tag);
519 			return (EINVAL);
520 		}
521 
522 		/* must not have an incremental recv already in progress */
523 		if (!mutex_tryenter(&ds->ds_recvlock)) {
524 			dsl_dataset_rele(ds, dmu_recv_tag);
525 			return (EBUSY);
526 		}
527 
528 		/* tmp clone name is: tofs/%tosnap" */
529 		(void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
530 		    "%%%s", tosnap);
531 		rbsa.force = force;
532 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
533 		    recv_existing_check, recv_existing_sync, ds, &rbsa, 5);
534 		if (err) {
535 			mutex_exit(&ds->ds_recvlock);
536 			dsl_dataset_rele(ds, dmu_recv_tag);
537 			return (err);
538 		}
539 		drc->drc_logical_ds = ds;
540 		drc->drc_real_ds = rbsa.ds;
541 	} else if (err == ENOENT) {
542 		/* target fs does not exist; must be a full backup or clone */
543 		dsl_dataset_t *parent;
544 		char *cp;
545 
546 		/*
547 		 * If it's a non-clone incremental, we are missing the
548 		 * target fs, so fail the recv.
549 		 */
550 		if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE))
551 			return (ENOENT);
552 
553 		/* Open the parent of tofs */
554 		cp = strrchr(tofs, '/');
555 		*cp = '\0';
556 		err = dsl_dataset_hold(tofs, FTAG, &parent);
557 		*cp = '/';
558 		if (err)
559 			return (err);
560 
561 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
562 		    recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5);
563 		dsl_dataset_rele(parent, FTAG);
564 		if (err)
565 			return (err);
566 		drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
567 		drc->drc_newfs = B_TRUE;
568 	}
569 
570 	return (err);
571 }
572 
573 struct restorearg {
574 	int err;
575 	int byteswap;
576 	vnode_t *vp;
577 	char *buf;
578 	uint64_t voff;
579 	int bufsize; /* amount of memory allocated for buf */
580 	zio_cksum_t cksum;
581 };
582 
583 static void *
584 restore_read(struct restorearg *ra, int len)
585 {
586 	void *rv;
587 	int done = 0;
588 
589 	/* some things will require 8-byte alignment, so everything must */
590 	ASSERT3U(len % 8, ==, 0);
591 
592 	while (done < len) {
593 		ssize_t resid;
594 
595 		ra->err = vn_rdwr(UIO_READ, ra->vp,
596 		    (caddr_t)ra->buf + done, len - done,
597 		    ra->voff, UIO_SYSSPACE, FAPPEND,
598 		    RLIM64_INFINITY, CRED(), &resid);
599 
600 		if (resid == len - done)
601 			ra->err = EINVAL;
602 		ra->voff += len - done - resid;
603 		done = len - resid;
604 		if (ra->err)
605 			return (NULL);
606 	}
607 
608 	ASSERT3U(done, ==, len);
609 	rv = ra->buf;
610 	if (ra->byteswap)
611 		fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
612 	else
613 		fletcher_4_incremental_native(rv, len, &ra->cksum);
614 	return (rv);
615 }
616 
617 static void
618 backup_byteswap(dmu_replay_record_t *drr)
619 {
620 #define	DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
621 #define	DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
622 	drr->drr_type = BSWAP_32(drr->drr_type);
623 	drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
624 	switch (drr->drr_type) {
625 	case DRR_BEGIN:
626 		DO64(drr_begin.drr_magic);
627 		DO64(drr_begin.drr_version);
628 		DO64(drr_begin.drr_creation_time);
629 		DO32(drr_begin.drr_type);
630 		DO32(drr_begin.drr_flags);
631 		DO64(drr_begin.drr_toguid);
632 		DO64(drr_begin.drr_fromguid);
633 		break;
634 	case DRR_OBJECT:
635 		DO64(drr_object.drr_object);
636 		/* DO64(drr_object.drr_allocation_txg); */
637 		DO32(drr_object.drr_type);
638 		DO32(drr_object.drr_bonustype);
639 		DO32(drr_object.drr_blksz);
640 		DO32(drr_object.drr_bonuslen);
641 		break;
642 	case DRR_FREEOBJECTS:
643 		DO64(drr_freeobjects.drr_firstobj);
644 		DO64(drr_freeobjects.drr_numobjs);
645 		break;
646 	case DRR_WRITE:
647 		DO64(drr_write.drr_object);
648 		DO32(drr_write.drr_type);
649 		DO64(drr_write.drr_offset);
650 		DO64(drr_write.drr_length);
651 		break;
652 	case DRR_FREE:
653 		DO64(drr_free.drr_object);
654 		DO64(drr_free.drr_offset);
655 		DO64(drr_free.drr_length);
656 		break;
657 	case DRR_END:
658 		DO64(drr_end.drr_checksum.zc_word[0]);
659 		DO64(drr_end.drr_checksum.zc_word[1]);
660 		DO64(drr_end.drr_checksum.zc_word[2]);
661 		DO64(drr_end.drr_checksum.zc_word[3]);
662 		break;
663 	}
664 #undef DO64
665 #undef DO32
666 }
667 
668 static int
669 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
670 {
671 	int err;
672 	dmu_tx_t *tx;
673 	void *data = NULL;
674 
675 	if (drro->drr_type == DMU_OT_NONE ||
676 	    drro->drr_type >= DMU_OT_NUMTYPES ||
677 	    drro->drr_bonustype >= DMU_OT_NUMTYPES ||
678 	    drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
679 	    drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
680 	    P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
681 	    drro->drr_blksz < SPA_MINBLOCKSIZE ||
682 	    drro->drr_blksz > SPA_MAXBLOCKSIZE ||
683 	    drro->drr_bonuslen > DN_MAX_BONUSLEN) {
684 		return (EINVAL);
685 	}
686 
687 	err = dmu_object_info(os, drro->drr_object, NULL);
688 
689 	if (err != 0 && err != ENOENT)
690 		return (EINVAL);
691 
692 	if (drro->drr_bonuslen) {
693 		data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
694 		if (ra->err)
695 			return (ra->err);
696 	}
697 
698 	if (err == ENOENT) {
699 		/* currently free, want to be allocated */
700 		tx = dmu_tx_create(os);
701 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
702 		err = dmu_tx_assign(tx, TXG_WAIT);
703 		if (err) {
704 			dmu_tx_abort(tx);
705 			return (err);
706 		}
707 		err = dmu_object_claim(os, drro->drr_object,
708 		    drro->drr_type, drro->drr_blksz,
709 		    drro->drr_bonustype, drro->drr_bonuslen, tx);
710 		dmu_tx_commit(tx);
711 	} else {
712 		/* currently allocated, want to be allocated */
713 		err = dmu_object_reclaim(os, drro->drr_object,
714 		    drro->drr_type, drro->drr_blksz,
715 		    drro->drr_bonustype, drro->drr_bonuslen);
716 	}
717 	if (err)
718 		return (EINVAL);
719 
720 	tx = dmu_tx_create(os);
721 	dmu_tx_hold_bonus(tx, drro->drr_object);
722 	err = dmu_tx_assign(tx, TXG_WAIT);
723 	if (err) {
724 		dmu_tx_abort(tx);
725 		return (err);
726 	}
727 
728 	dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
729 	dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
730 
731 	if (data != NULL) {
732 		dmu_buf_t *db;
733 
734 		VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
735 		dmu_buf_will_dirty(db, tx);
736 
737 		ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
738 		bcopy(data, db->db_data, drro->drr_bonuslen);
739 		if (ra->byteswap) {
740 			dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
741 			    drro->drr_bonuslen);
742 		}
743 		dmu_buf_rele(db, FTAG);
744 	}
745 	dmu_tx_commit(tx);
746 	return (0);
747 }
748 
749 /* ARGSUSED */
750 static int
751 restore_freeobjects(struct restorearg *ra, objset_t *os,
752     struct drr_freeobjects *drrfo)
753 {
754 	uint64_t obj;
755 
756 	if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
757 		return (EINVAL);
758 
759 	for (obj = drrfo->drr_firstobj;
760 	    obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
761 	    (void) dmu_object_next(os, &obj, FALSE, 0)) {
762 		int err;
763 
764 		if (dmu_object_info(os, obj, NULL) != 0)
765 			continue;
766 
767 		err = dmu_free_object(os, obj);
768 		if (err)
769 			return (err);
770 	}
771 	return (0);
772 }
773 
774 static int
775 restore_write(struct restorearg *ra, objset_t *os,
776     struct drr_write *drrw)
777 {
778 	dmu_tx_t *tx;
779 	void *data;
780 	int err;
781 
782 	if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
783 	    drrw->drr_type >= DMU_OT_NUMTYPES)
784 		return (EINVAL);
785 
786 	data = restore_read(ra, drrw->drr_length);
787 	if (data == NULL)
788 		return (ra->err);
789 
790 	if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
791 		return (EINVAL);
792 
793 	tx = dmu_tx_create(os);
794 
795 	dmu_tx_hold_write(tx, drrw->drr_object,
796 	    drrw->drr_offset, drrw->drr_length);
797 	err = dmu_tx_assign(tx, TXG_WAIT);
798 	if (err) {
799 		dmu_tx_abort(tx);
800 		return (err);
801 	}
802 	if (ra->byteswap)
803 		dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
804 	dmu_write(os, drrw->drr_object,
805 	    drrw->drr_offset, drrw->drr_length, data, tx);
806 	dmu_tx_commit(tx);
807 	return (0);
808 }
809 
810 /* ARGSUSED */
811 static int
812 restore_free(struct restorearg *ra, objset_t *os,
813     struct drr_free *drrf)
814 {
815 	int err;
816 
817 	if (drrf->drr_length != -1ULL &&
818 	    drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
819 		return (EINVAL);
820 
821 	if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
822 		return (EINVAL);
823 
824 	err = dmu_free_long_range(os, drrf->drr_object,
825 	    drrf->drr_offset, drrf->drr_length);
826 	return (err);
827 }
828 
829 /*
830  * NB: callers *must* call dmu_recv_end() if this succeeds.
831  */
832 int
833 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp)
834 {
835 	struct restorearg ra = { 0 };
836 	dmu_replay_record_t *drr;
837 	objset_t *os;
838 	zio_cksum_t pcksum;
839 
840 	if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
841 		ra.byteswap = TRUE;
842 
843 	{
844 		/* compute checksum of drr_begin record */
845 		dmu_replay_record_t *drr;
846 		drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
847 
848 		drr->drr_type = DRR_BEGIN;
849 		drr->drr_u.drr_begin = *drc->drc_drrb;
850 		if (ra.byteswap) {
851 			fletcher_4_incremental_byteswap(drr,
852 			    sizeof (dmu_replay_record_t), &ra.cksum);
853 		} else {
854 			fletcher_4_incremental_native(drr,
855 			    sizeof (dmu_replay_record_t), &ra.cksum);
856 		}
857 		kmem_free(drr, sizeof (dmu_replay_record_t));
858 	}
859 
860 	if (ra.byteswap) {
861 		struct drr_begin *drrb = drc->drc_drrb;
862 		drrb->drr_magic = BSWAP_64(drrb->drr_magic);
863 		drrb->drr_version = BSWAP_64(drrb->drr_version);
864 		drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
865 		drrb->drr_type = BSWAP_32(drrb->drr_type);
866 		drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
867 		drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
868 	}
869 
870 	ra.vp = vp;
871 	ra.voff = *voffp;
872 	ra.bufsize = 1<<20;
873 	ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
874 
875 	/* these were verified in dmu_recv_begin */
876 	ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
877 	ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
878 
879 	/*
880 	 * Open the objset we are modifying.
881 	 */
882 	VERIFY(dmu_objset_from_ds(drc->drc_real_ds, &os) == 0);
883 
884 	ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
885 
886 	/*
887 	 * Read records and process them.
888 	 */
889 	pcksum = ra.cksum;
890 	while (ra.err == 0 &&
891 	    NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
892 		if (issig(JUSTLOOKING) && issig(FORREAL)) {
893 			ra.err = EINTR;
894 			goto out;
895 		}
896 
897 		if (ra.byteswap)
898 			backup_byteswap(drr);
899 
900 		switch (drr->drr_type) {
901 		case DRR_OBJECT:
902 		{
903 			/*
904 			 * We need to make a copy of the record header,
905 			 * because restore_{object,write} may need to
906 			 * restore_read(), which will invalidate drr.
907 			 */
908 			struct drr_object drro = drr->drr_u.drr_object;
909 			ra.err = restore_object(&ra, os, &drro);
910 			break;
911 		}
912 		case DRR_FREEOBJECTS:
913 		{
914 			struct drr_freeobjects drrfo =
915 			    drr->drr_u.drr_freeobjects;
916 			ra.err = restore_freeobjects(&ra, os, &drrfo);
917 			break;
918 		}
919 		case DRR_WRITE:
920 		{
921 			struct drr_write drrw = drr->drr_u.drr_write;
922 			ra.err = restore_write(&ra, os, &drrw);
923 			break;
924 		}
925 		case DRR_FREE:
926 		{
927 			struct drr_free drrf = drr->drr_u.drr_free;
928 			ra.err = restore_free(&ra, os, &drrf);
929 			break;
930 		}
931 		case DRR_END:
932 		{
933 			struct drr_end drre = drr->drr_u.drr_end;
934 			/*
935 			 * We compare against the *previous* checksum
936 			 * value, because the stored checksum is of
937 			 * everything before the DRR_END record.
938 			 */
939 			if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
940 				ra.err = ECKSUM;
941 			goto out;
942 		}
943 		default:
944 			ra.err = EINVAL;
945 			goto out;
946 		}
947 		pcksum = ra.cksum;
948 	}
949 	ASSERT(ra.err != 0);
950 
951 out:
952 	if (ra.err != 0) {
953 		/*
954 		 * destroy what we created, so we don't leave it in the
955 		 * inconsistent restoring state.
956 		 */
957 		txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
958 
959 		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
960 		    B_FALSE);
961 		if (drc->drc_real_ds != drc->drc_logical_ds) {
962 			mutex_exit(&drc->drc_logical_ds->ds_recvlock);
963 			dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
964 		}
965 	}
966 
967 	kmem_free(ra.buf, ra.bufsize);
968 	*voffp = ra.voff;
969 	return (ra.err);
970 }
971 
972 struct recvendsyncarg {
973 	char *tosnap;
974 	uint64_t creation_time;
975 	uint64_t toguid;
976 };
977 
978 static int
979 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
980 {
981 	dsl_dataset_t *ds = arg1;
982 	struct recvendsyncarg *resa = arg2;
983 
984 	return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
985 }
986 
987 static void
988 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
989 {
990 	dsl_dataset_t *ds = arg1;
991 	struct recvendsyncarg *resa = arg2;
992 
993 	dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
994 
995 	/* set snapshot's creation time and guid */
996 	dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
997 	ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
998 	ds->ds_prev->ds_phys->ds_guid = resa->toguid;
999 	ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1000 
1001 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1002 	ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1003 }
1004 
1005 static int
1006 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1007 {
1008 	struct recvendsyncarg resa;
1009 	dsl_dataset_t *ds = drc->drc_logical_ds;
1010 	int err;
1011 
1012 	/*
1013 	 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1014 	 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1015 	 * can close it.
1016 	 */
1017 	txg_wait_synced(ds->ds_dir->dd_pool, 0);
1018 
1019 	if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1020 		err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1021 		    drc->drc_force);
1022 		if (err)
1023 			goto out;
1024 	} else {
1025 		mutex_exit(&ds->ds_recvlock);
1026 		dsl_dataset_rele(ds, dmu_recv_tag);
1027 		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1028 		    B_FALSE);
1029 		return (EBUSY);
1030 	}
1031 
1032 	resa.creation_time = drc->drc_drrb->drr_creation_time;
1033 	resa.toguid = drc->drc_drrb->drr_toguid;
1034 	resa.tosnap = drc->drc_tosnap;
1035 
1036 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1037 	    recv_end_check, recv_end_sync, ds, &resa, 3);
1038 	if (err) {
1039 		/* swap back */
1040 		(void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE);
1041 	}
1042 
1043 out:
1044 	mutex_exit(&ds->ds_recvlock);
1045 	dsl_dataset_disown(ds, dmu_recv_tag);
1046 	(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE);
1047 	return (err);
1048 }
1049 
1050 static int
1051 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1052 {
1053 	struct recvendsyncarg resa;
1054 	dsl_dataset_t *ds = drc->drc_logical_ds;
1055 	int err;
1056 
1057 	/*
1058 	 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1059 	 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1060 	 * can close it.
1061 	 */
1062 	txg_wait_synced(ds->ds_dir->dd_pool, 0);
1063 
1064 	resa.creation_time = drc->drc_drrb->drr_creation_time;
1065 	resa.toguid = drc->drc_drrb->drr_toguid;
1066 	resa.tosnap = drc->drc_tosnap;
1067 
1068 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1069 	    recv_end_check, recv_end_sync, ds, &resa, 3);
1070 	if (err) {
1071 		/* clean up the fs we just recv'd into */
1072 		(void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE);
1073 	} else {
1074 		/* release the hold from dmu_recv_begin */
1075 		dsl_dataset_disown(ds, dmu_recv_tag);
1076 	}
1077 	return (err);
1078 }
1079 
1080 int
1081 dmu_recv_end(dmu_recv_cookie_t *drc)
1082 {
1083 	if (drc->drc_logical_ds != drc->drc_real_ds)
1084 		return (dmu_recv_existing_end(drc));
1085 	else
1086 		return (dmu_recv_new_end(drc));
1087 }
1088