xref: /titanic_44/usr/src/uts/common/fs/zfs/dmu_send.c (revision 2d6b5ea734bb47d251c82670646fde46af15fd69)
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 (bp == NULL && zb->zb_object == 0) {
165 		uint64_t span = BP_SPAN(dnp, zb->zb_level);
166 		uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
167 		err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
168 	} else if (bp == NULL) {
169 		uint64_t span = BP_SPAN(dnp, zb->zb_level);
170 		err = dump_free(ba, zb->zb_object, zb->zb_blkid * span, span);
171 	} else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
172 		return (0);
173 	} else if (type == DMU_OT_DNODE) {
174 		dnode_phys_t *blk;
175 		int i;
176 		int blksz = BP_GET_LSIZE(bp);
177 		uint32_t aflags = ARC_WAIT;
178 		arc_buf_t *abuf;
179 
180 		if (arc_read_nolock(NULL, spa, bp,
181 		    arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
182 		    ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
183 			return (EIO);
184 
185 		blk = abuf->b_data;
186 		for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
187 			uint64_t dnobj = (zb->zb_blkid <<
188 			    (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
189 			err = dump_dnode(ba, dnobj, blk+i);
190 			if (err)
191 				break;
192 		}
193 		(void) arc_buf_remove_ref(abuf, &abuf);
194 	} else { /* it's a level-0 block of a regular object */
195 		uint32_t aflags = ARC_WAIT;
196 		arc_buf_t *abuf;
197 		int blksz = BP_GET_LSIZE(bp);
198 
199 		if (arc_read_nolock(NULL, spa, bp,
200 		    arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
201 		    ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
202 			return (EIO);
203 
204 		err = dump_data(ba, type, zb->zb_object, zb->zb_blkid * blksz,
205 		    blksz, abuf->b_data);
206 		(void) arc_buf_remove_ref(abuf, &abuf);
207 	}
208 
209 	ASSERT(err == 0 || err == EINTR);
210 	return (err);
211 }
212 
213 int
214 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
215     vnode_t *vp, offset_t *off)
216 {
217 	dsl_dataset_t *ds = tosnap->os->os_dsl_dataset;
218 	dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL;
219 	dmu_replay_record_t *drr;
220 	struct backuparg ba;
221 	int err;
222 	uint64_t fromtxg = 0;
223 
224 	/* tosnap must be a snapshot */
225 	if (ds->ds_phys->ds_next_snap_obj == 0)
226 		return (EINVAL);
227 
228 	/* fromsnap must be an earlier snapshot from the same fs as tosnap */
229 	if (fromds && (ds->ds_dir != fromds->ds_dir ||
230 	    fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
231 		return (EXDEV);
232 
233 	if (fromorigin) {
234 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
235 
236 		if (fromsnap)
237 			return (EINVAL);
238 
239 		if (dsl_dir_is_clone(ds->ds_dir)) {
240 			rw_enter(&dp->dp_config_rwlock, RW_READER);
241 			err = dsl_dataset_hold_obj(dp,
242 			    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
243 			rw_exit(&dp->dp_config_rwlock);
244 			if (err)
245 				return (err);
246 		} else {
247 			fromorigin = B_FALSE;
248 		}
249 	}
250 
251 
252 	drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
253 	drr->drr_type = DRR_BEGIN;
254 	drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
255 	drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION;
256 	drr->drr_u.drr_begin.drr_creation_time =
257 	    ds->ds_phys->ds_creation_time;
258 	drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type;
259 	if (fromorigin)
260 		drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
261 	drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
262 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
263 		drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
264 
265 	if (fromds)
266 		drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
267 	dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
268 
269 	if (fromds)
270 		fromtxg = fromds->ds_phys->ds_creation_txg;
271 	if (fromorigin)
272 		dsl_dataset_rele(fromds, FTAG);
273 
274 	ba.drr = drr;
275 	ba.vp = vp;
276 	ba.os = tosnap;
277 	ba.off = off;
278 	ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
279 
280 	if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
281 		kmem_free(drr, sizeof (dmu_replay_record_t));
282 		return (ba.err);
283 	}
284 
285 	err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
286 	    backup_cb, &ba);
287 
288 	if (err) {
289 		if (err == EINTR && ba.err)
290 			err = ba.err;
291 		kmem_free(drr, sizeof (dmu_replay_record_t));
292 		return (err);
293 	}
294 
295 	bzero(drr, sizeof (dmu_replay_record_t));
296 	drr->drr_type = DRR_END;
297 	drr->drr_u.drr_end.drr_checksum = ba.zc;
298 
299 	if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
300 		kmem_free(drr, sizeof (dmu_replay_record_t));
301 		return (ba.err);
302 	}
303 
304 	kmem_free(drr, sizeof (dmu_replay_record_t));
305 
306 	return (0);
307 }
308 
309 struct recvbeginsyncarg {
310 	const char *tofs;
311 	const char *tosnap;
312 	dsl_dataset_t *origin;
313 	uint64_t fromguid;
314 	dmu_objset_type_t type;
315 	void *tag;
316 	boolean_t force;
317 	uint64_t dsflags;
318 	char clonelastname[MAXNAMELEN];
319 	dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
320 };
321 
322 static dsl_dataset_t *
323 recv_full_sync_impl(dsl_pool_t *dp, uint64_t dsobj, dmu_objset_type_t type,
324     cred_t *cr, dmu_tx_t *tx)
325 {
326 	dsl_dataset_t *ds;
327 
328 	/* This should always work, since we just created it */
329 	/* XXX - create should return an owned ds */
330 	VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
331 	    DS_MODE_INCONSISTENT, dmu_recv_tag, &ds));
332 
333 	if (type != DMU_OST_NONE) {
334 		(void) dmu_objset_create_impl(dp->dp_spa,
335 		    ds, &ds->ds_phys->ds_bp, type, tx);
336 	}
337 
338 	spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC,
339 	    dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
340 
341 	return (ds);
342 }
343 
344 /* ARGSUSED */
345 static int
346 recv_full_check(void *arg1, void *arg2, dmu_tx_t *tx)
347 {
348 	dsl_dir_t *dd = arg1;
349 	struct recvbeginsyncarg *rbsa = arg2;
350 	objset_t *mos = dd->dd_pool->dp_meta_objset;
351 	uint64_t val;
352 	int err;
353 
354 	err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
355 	    strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
356 
357 	if (err != ENOENT)
358 		return (err ? err : EEXIST);
359 
360 	if (rbsa->origin) {
361 		/* make sure it's a snap in the same pool */
362 		if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
363 			return (EXDEV);
364 		if (rbsa->origin->ds_phys->ds_num_children == 0)
365 			return (EINVAL);
366 		if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
367 			return (ENODEV);
368 	}
369 
370 	return (0);
371 }
372 
373 static void
374 recv_full_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
375 {
376 	dsl_dir_t *dd = arg1;
377 	struct recvbeginsyncarg *rbsa = arg2;
378 	uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
379 	uint64_t dsobj;
380 
381 	dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
382 	    rbsa->origin, flags, cr, tx);
383 
384 	rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
385 	    rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
386 }
387 
388 static int
389 recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
390 {
391 	dsl_dataset_t *ds = arg1;
392 	struct recvbeginsyncarg *rbsa = arg2;
393 	int err;
394 
395 	/* must be a head ds */
396 	if (ds->ds_phys->ds_next_snap_obj != 0)
397 		return (EINVAL);
398 
399 	/* must not be a clone ds */
400 	if (dsl_dir_is_clone(ds->ds_dir))
401 		return (EINVAL);
402 
403 	err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
404 	if (err)
405 		return (err);
406 
407 	if (rbsa->origin) {
408 		/* make sure it's a snap in the same pool */
409 		if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool)
410 			return (EXDEV);
411 		if (rbsa->origin->ds_phys->ds_num_children == 0)
412 			return (EINVAL);
413 		if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
414 			return (ENODEV);
415 	}
416 
417 	return (0);
418 }
419 
420 static void
421 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
422 {
423 	dsl_dataset_t *ds = arg1;
424 	struct recvbeginsyncarg *rbsa = arg2;
425 	dsl_dir_t *dd = ds->ds_dir;
426 	uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
427 	uint64_t dsobj;
428 
429 	/*
430 	 * NB: caller must provide an extra hold on the dsl_dir_t, so it
431 	 * won't go away when dsl_dataset_destroy_sync() closes the
432 	 * dataset.
433 	 */
434 	dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
435 
436 	dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
437 
438 	rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
439 	    rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
440 }
441 
442 /* ARGSUSED */
443 static int
444 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
445 {
446 	dsl_dataset_t *ds = arg1;
447 	struct recvbeginsyncarg *rbsa = arg2;
448 	int err;
449 	uint64_t val;
450 
451 	/* must not have any changes since most recent snapshot */
452 	if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
453 		return (ETXTBSY);
454 
455 	/* must already be a snapshot of this fs */
456 	if (ds->ds_phys->ds_prev_snap_obj == 0)
457 		return (ENODEV);
458 
459 	/* most recent snapshot must match fromguid */
460 	if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
461 		return (ENODEV);
462 
463 	/* temporary clone name must not exist */
464 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
465 	    ds->ds_dir->dd_phys->dd_child_dir_zapobj,
466 	    rbsa->clonelastname, 8, 1, &val);
467 	if (err == 0)
468 		return (EEXIST);
469 	if (err != ENOENT)
470 		return (err);
471 
472 	/* new snapshot name must not exist */
473 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
474 	    ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
475 	if (err == 0)
476 		return (EEXIST);
477 	if (err != ENOENT)
478 		return (err);
479 	return (0);
480 }
481 
482 /* ARGSUSED */
483 static void
484 recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
485 {
486 	dsl_dataset_t *ohds = arg1;
487 	struct recvbeginsyncarg *rbsa = arg2;
488 	dsl_pool_t *dp = ohds->ds_dir->dd_pool;
489 	dsl_dataset_t *ods, *cds;
490 	uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
491 	uint64_t dsobj;
492 
493 	/* create the temporary clone */
494 	VERIFY(0 == dsl_dataset_hold_obj(dp, ohds->ds_phys->ds_prev_snap_obj,
495 	    FTAG, &ods));
496 	dsobj = dsl_dataset_create_sync(ohds->ds_dir,
497 	    rbsa->clonelastname, ods, flags, cr, tx);
498 	dsl_dataset_rele(ods, FTAG);
499 
500 	/* open the temporary clone */
501 	VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
502 	    DS_MODE_INCONSISTENT, dmu_recv_tag, &cds));
503 
504 	/* copy the refquota from the target fs to the clone */
505 	if (ohds->ds_quota > 0)
506 		dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
507 
508 	rbsa->ds = cds;
509 
510 	spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
511 	    dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
512 }
513 
514 /* ARGSUSED */
515 static void
516 recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
517 {
518 	dsl_dataset_t *ds = arg1;
519 
520 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
521 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
522 
523 	spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
524 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
525 	    ds->ds_object);
526 }
527 
528 /*
529  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
530  * succeeds; otherwise we will leak the holds on the datasets.
531  */
532 int
533 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
534     boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
535 {
536 	int err = 0;
537 	boolean_t byteswap;
538 	struct recvbeginsyncarg rbsa;
539 	uint64_t version;
540 	int flags;
541 	dsl_dataset_t *ds;
542 
543 	if (drrb->drr_magic == DMU_BACKUP_MAGIC)
544 		byteswap = FALSE;
545 	else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
546 		byteswap = TRUE;
547 	else
548 		return (EINVAL);
549 
550 	rbsa.tofs = tofs;
551 	rbsa.tosnap = tosnap;
552 	rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL;
553 	rbsa.fromguid = drrb->drr_fromguid;
554 	rbsa.type = drrb->drr_type;
555 	rbsa.tag = FTAG;
556 	rbsa.dsflags = 0;
557 	version = drrb->drr_version;
558 	flags = drrb->drr_flags;
559 
560 	if (byteswap) {
561 		rbsa.type = BSWAP_32(rbsa.type);
562 		rbsa.fromguid = BSWAP_64(rbsa.fromguid);
563 		version = BSWAP_64(version);
564 		flags = BSWAP_32(flags);
565 	}
566 
567 	if (version != DMU_BACKUP_STREAM_VERSION ||
568 	    rbsa.type >= DMU_OST_NUMTYPES ||
569 	    ((flags & DRR_FLAG_CLONE) && origin == NULL))
570 		return (EINVAL);
571 
572 	if (flags & DRR_FLAG_CI_DATA)
573 		rbsa.dsflags = DS_FLAG_CI_DATASET;
574 
575 	bzero(drc, sizeof (dmu_recv_cookie_t));
576 	drc->drc_drrb = drrb;
577 	drc->drc_tosnap = tosnap;
578 	drc->drc_force = force;
579 
580 	/*
581 	 * Process the begin in syncing context.
582 	 */
583 	if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
584 		/* offline incremental receive */
585 		err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds);
586 		if (err)
587 			return (err);
588 
589 		/*
590 		 * Only do the rollback if the most recent snapshot
591 		 * matches the incremental source
592 		 */
593 		if (force) {
594 			if (ds->ds_prev == NULL ||
595 			    ds->ds_prev->ds_phys->ds_guid !=
596 			    rbsa.fromguid) {
597 				dsl_dataset_disown(ds, dmu_recv_tag);
598 				return (ENODEV);
599 			}
600 			(void) dsl_dataset_rollback(ds, DMU_OST_NONE);
601 		}
602 		rbsa.force = B_FALSE;
603 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
604 		    recv_incremental_check,
605 		    recv_offline_incremental_sync, ds, &rbsa, 1);
606 		if (err) {
607 			dsl_dataset_disown(ds, dmu_recv_tag);
608 			return (err);
609 		}
610 		drc->drc_logical_ds = drc->drc_real_ds = ds;
611 	} else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
612 		/* online incremental receive */
613 
614 		/* tmp clone name is: tofs/%tosnap" */
615 		(void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
616 		    "%%%s", tosnap);
617 
618 		/* open the dataset we are logically receiving into */
619 		err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
620 		if (err)
621 			return (err);
622 
623 		rbsa.force = force;
624 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
625 		    recv_incremental_check,
626 		    recv_online_incremental_sync, ds, &rbsa, 5);
627 		if (err) {
628 			dsl_dataset_rele(ds, dmu_recv_tag);
629 			return (err);
630 		}
631 		drc->drc_logical_ds = ds;
632 		drc->drc_real_ds = rbsa.ds;
633 	} else {
634 		/* create new fs -- full backup or clone */
635 		dsl_dir_t *dd = NULL;
636 		const char *tail;
637 
638 		err = dsl_dir_open(tofs, FTAG, &dd, &tail);
639 		if (err)
640 			return (err);
641 		if (tail == NULL) {
642 			if (!force) {
643 				dsl_dir_close(dd, FTAG);
644 				return (EEXIST);
645 			}
646 
647 			rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
648 			err = dsl_dataset_own_obj(dd->dd_pool,
649 			    dd->dd_phys->dd_head_dataset_obj,
650 			    DS_MODE_INCONSISTENT, FTAG, &ds);
651 			rw_exit(&dd->dd_pool->dp_config_rwlock);
652 			if (err) {
653 				dsl_dir_close(dd, FTAG);
654 				return (err);
655 			}
656 
657 			dsl_dataset_make_exclusive(ds, FTAG);
658 			err = dsl_sync_task_do(dd->dd_pool,
659 			    recv_full_existing_check,
660 			    recv_full_existing_sync, ds, &rbsa, 5);
661 			dsl_dataset_disown(ds, FTAG);
662 		} else {
663 			err = dsl_sync_task_do(dd->dd_pool, recv_full_check,
664 			    recv_full_sync, dd, &rbsa, 5);
665 		}
666 		dsl_dir_close(dd, FTAG);
667 		if (err)
668 			return (err);
669 		drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
670 		drc->drc_newfs = B_TRUE;
671 	}
672 
673 	return (0);
674 }
675 
676 struct restorearg {
677 	int err;
678 	int byteswap;
679 	vnode_t *vp;
680 	char *buf;
681 	uint64_t voff;
682 	int bufsize; /* amount of memory allocated for buf */
683 	zio_cksum_t cksum;
684 };
685 
686 static void *
687 restore_read(struct restorearg *ra, int len)
688 {
689 	void *rv;
690 	int done = 0;
691 
692 	/* some things will require 8-byte alignment, so everything must */
693 	ASSERT3U(len % 8, ==, 0);
694 
695 	while (done < len) {
696 		ssize_t resid;
697 
698 		ra->err = vn_rdwr(UIO_READ, ra->vp,
699 		    (caddr_t)ra->buf + done, len - done,
700 		    ra->voff, UIO_SYSSPACE, FAPPEND,
701 		    RLIM64_INFINITY, CRED(), &resid);
702 
703 		if (resid == len - done)
704 			ra->err = EINVAL;
705 		ra->voff += len - done - resid;
706 		done = len - resid;
707 		if (ra->err)
708 			return (NULL);
709 	}
710 
711 	ASSERT3U(done, ==, len);
712 	rv = ra->buf;
713 	if (ra->byteswap)
714 		fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
715 	else
716 		fletcher_4_incremental_native(rv, len, &ra->cksum);
717 	return (rv);
718 }
719 
720 static void
721 backup_byteswap(dmu_replay_record_t *drr)
722 {
723 #define	DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
724 #define	DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
725 	drr->drr_type = BSWAP_32(drr->drr_type);
726 	drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
727 	switch (drr->drr_type) {
728 	case DRR_BEGIN:
729 		DO64(drr_begin.drr_magic);
730 		DO64(drr_begin.drr_version);
731 		DO64(drr_begin.drr_creation_time);
732 		DO32(drr_begin.drr_type);
733 		DO32(drr_begin.drr_flags);
734 		DO64(drr_begin.drr_toguid);
735 		DO64(drr_begin.drr_fromguid);
736 		break;
737 	case DRR_OBJECT:
738 		DO64(drr_object.drr_object);
739 		/* DO64(drr_object.drr_allocation_txg); */
740 		DO32(drr_object.drr_type);
741 		DO32(drr_object.drr_bonustype);
742 		DO32(drr_object.drr_blksz);
743 		DO32(drr_object.drr_bonuslen);
744 		break;
745 	case DRR_FREEOBJECTS:
746 		DO64(drr_freeobjects.drr_firstobj);
747 		DO64(drr_freeobjects.drr_numobjs);
748 		break;
749 	case DRR_WRITE:
750 		DO64(drr_write.drr_object);
751 		DO32(drr_write.drr_type);
752 		DO64(drr_write.drr_offset);
753 		DO64(drr_write.drr_length);
754 		break;
755 	case DRR_FREE:
756 		DO64(drr_free.drr_object);
757 		DO64(drr_free.drr_offset);
758 		DO64(drr_free.drr_length);
759 		break;
760 	case DRR_END:
761 		DO64(drr_end.drr_checksum.zc_word[0]);
762 		DO64(drr_end.drr_checksum.zc_word[1]);
763 		DO64(drr_end.drr_checksum.zc_word[2]);
764 		DO64(drr_end.drr_checksum.zc_word[3]);
765 		break;
766 	}
767 #undef DO64
768 #undef DO32
769 }
770 
771 static int
772 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
773 {
774 	int err;
775 	dmu_tx_t *tx;
776 	void *data = NULL;
777 
778 	err = dmu_object_info(os, drro->drr_object, NULL);
779 
780 	if (err != 0 && err != ENOENT)
781 		return (EINVAL);
782 
783 	if (drro->drr_type == DMU_OT_NONE ||
784 	    drro->drr_type >= DMU_OT_NUMTYPES ||
785 	    drro->drr_bonustype >= DMU_OT_NUMTYPES ||
786 	    drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
787 	    drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
788 	    P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
789 	    drro->drr_blksz < SPA_MINBLOCKSIZE ||
790 	    drro->drr_blksz > SPA_MAXBLOCKSIZE ||
791 	    drro->drr_bonuslen > DN_MAX_BONUSLEN) {
792 		return (EINVAL);
793 	}
794 
795 	if (drro->drr_bonuslen) {
796 		data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
797 		if (ra->err)
798 			return (ra->err);
799 	}
800 
801 	tx = dmu_tx_create(os);
802 
803 	if (err == ENOENT) {
804 		/* currently free, want to be allocated */
805 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
806 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1);
807 		err = dmu_tx_assign(tx, TXG_WAIT);
808 		if (err) {
809 			dmu_tx_abort(tx);
810 			return (err);
811 		}
812 		err = dmu_object_claim(os, drro->drr_object,
813 		    drro->drr_type, drro->drr_blksz,
814 		    drro->drr_bonustype, drro->drr_bonuslen, tx);
815 	} else {
816 		/* currently allocated, want to be allocated */
817 		dmu_tx_hold_bonus(tx, drro->drr_object);
818 		/*
819 		 * We may change blocksize and delete old content,
820 		 * so need to hold_write and hold_free.
821 		 */
822 		dmu_tx_hold_write(tx, drro->drr_object, 0, 1);
823 		dmu_tx_hold_free(tx, drro->drr_object, 0, DMU_OBJECT_END);
824 		err = dmu_tx_assign(tx, TXG_WAIT);
825 		if (err) {
826 			dmu_tx_abort(tx);
827 			return (err);
828 		}
829 
830 		err = dmu_object_reclaim(os, drro->drr_object,
831 		    drro->drr_type, drro->drr_blksz,
832 		    drro->drr_bonustype, drro->drr_bonuslen, tx);
833 	}
834 	if (err) {
835 		dmu_tx_commit(tx);
836 		return (EINVAL);
837 	}
838 
839 	dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
840 	dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
841 
842 	if (data != NULL) {
843 		dmu_buf_t *db;
844 
845 		VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
846 		dmu_buf_will_dirty(db, tx);
847 
848 		ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
849 		bcopy(data, db->db_data, drro->drr_bonuslen);
850 		if (ra->byteswap) {
851 			dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
852 			    drro->drr_bonuslen);
853 		}
854 		dmu_buf_rele(db, FTAG);
855 	}
856 	dmu_tx_commit(tx);
857 	return (0);
858 }
859 
860 /* ARGSUSED */
861 static int
862 restore_freeobjects(struct restorearg *ra, objset_t *os,
863     struct drr_freeobjects *drrfo)
864 {
865 	uint64_t obj;
866 
867 	if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
868 		return (EINVAL);
869 
870 	for (obj = drrfo->drr_firstobj;
871 	    obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
872 	    (void) dmu_object_next(os, &obj, FALSE, 0)) {
873 		int err;
874 
875 		if (dmu_object_info(os, obj, NULL) != 0)
876 			continue;
877 
878 		err = dmu_free_object(os, obj);
879 		if (err)
880 			return (err);
881 	}
882 	return (0);
883 }
884 
885 static int
886 restore_write(struct restorearg *ra, objset_t *os,
887     struct drr_write *drrw)
888 {
889 	dmu_tx_t *tx;
890 	void *data;
891 	int err;
892 
893 	if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
894 	    drrw->drr_type >= DMU_OT_NUMTYPES)
895 		return (EINVAL);
896 
897 	data = restore_read(ra, drrw->drr_length);
898 	if (data == NULL)
899 		return (ra->err);
900 
901 	if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
902 		return (EINVAL);
903 
904 	tx = dmu_tx_create(os);
905 
906 	dmu_tx_hold_write(tx, drrw->drr_object,
907 	    drrw->drr_offset, drrw->drr_length);
908 	err = dmu_tx_assign(tx, TXG_WAIT);
909 	if (err) {
910 		dmu_tx_abort(tx);
911 		return (err);
912 	}
913 	if (ra->byteswap)
914 		dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
915 	dmu_write(os, drrw->drr_object,
916 	    drrw->drr_offset, drrw->drr_length, data, tx);
917 	dmu_tx_commit(tx);
918 	return (0);
919 }
920 
921 /* ARGSUSED */
922 static int
923 restore_free(struct restorearg *ra, objset_t *os,
924     struct drr_free *drrf)
925 {
926 	int err;
927 
928 	if (drrf->drr_length != -1ULL &&
929 	    drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
930 		return (EINVAL);
931 
932 	if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
933 		return (EINVAL);
934 
935 	err = dmu_free_long_range(os, drrf->drr_object,
936 	    drrf->drr_offset, drrf->drr_length);
937 	return (err);
938 }
939 
940 void
941 dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
942 {
943 	if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
944 		/*
945 		 * online incremental or new fs: destroy the fs (which
946 		 * may be a clone) that we created
947 		 */
948 		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
949 		if (drc->drc_real_ds != drc->drc_logical_ds)
950 			dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
951 	} else {
952 		/*
953 		 * offline incremental: rollback to most recent snapshot.
954 		 */
955 		(void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE);
956 		dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag);
957 	}
958 }
959 
960 /*
961  * NB: callers *must* call dmu_recv_end() if this succeeds.
962  */
963 int
964 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp)
965 {
966 	struct restorearg ra = { 0 };
967 	dmu_replay_record_t *drr;
968 	objset_t *os;
969 	zio_cksum_t pcksum;
970 
971 	if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
972 		ra.byteswap = TRUE;
973 
974 	{
975 		/* compute checksum of drr_begin record */
976 		dmu_replay_record_t *drr;
977 		drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
978 
979 		drr->drr_type = DRR_BEGIN;
980 		drr->drr_u.drr_begin = *drc->drc_drrb;
981 		if (ra.byteswap) {
982 			fletcher_4_incremental_byteswap(drr,
983 			    sizeof (dmu_replay_record_t), &ra.cksum);
984 		} else {
985 			fletcher_4_incremental_native(drr,
986 			    sizeof (dmu_replay_record_t), &ra.cksum);
987 		}
988 		kmem_free(drr, sizeof (dmu_replay_record_t));
989 	}
990 
991 	if (ra.byteswap) {
992 		struct drr_begin *drrb = drc->drc_drrb;
993 		drrb->drr_magic = BSWAP_64(drrb->drr_magic);
994 		drrb->drr_version = BSWAP_64(drrb->drr_version);
995 		drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
996 		drrb->drr_type = BSWAP_32(drrb->drr_type);
997 		drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
998 		drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
999 	}
1000 
1001 	ra.vp = vp;
1002 	ra.voff = *voffp;
1003 	ra.bufsize = 1<<20;
1004 	ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1005 
1006 	/* these were verified in dmu_recv_begin */
1007 	ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
1008 	ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1009 
1010 	/*
1011 	 * Open the objset we are modifying.
1012 	 */
1013 	VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0);
1014 
1015 	ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1016 
1017 	/*
1018 	 * Read records and process them.
1019 	 */
1020 	pcksum = ra.cksum;
1021 	while (ra.err == 0 &&
1022 	    NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1023 		if (issig(JUSTLOOKING) && issig(FORREAL)) {
1024 			ra.err = EINTR;
1025 			goto out;
1026 		}
1027 
1028 		if (ra.byteswap)
1029 			backup_byteswap(drr);
1030 
1031 		switch (drr->drr_type) {
1032 		case DRR_OBJECT:
1033 		{
1034 			/*
1035 			 * We need to make a copy of the record header,
1036 			 * because restore_{object,write} may need to
1037 			 * restore_read(), which will invalidate drr.
1038 			 */
1039 			struct drr_object drro = drr->drr_u.drr_object;
1040 			ra.err = restore_object(&ra, os, &drro);
1041 			break;
1042 		}
1043 		case DRR_FREEOBJECTS:
1044 		{
1045 			struct drr_freeobjects drrfo =
1046 			    drr->drr_u.drr_freeobjects;
1047 			ra.err = restore_freeobjects(&ra, os, &drrfo);
1048 			break;
1049 		}
1050 		case DRR_WRITE:
1051 		{
1052 			struct drr_write drrw = drr->drr_u.drr_write;
1053 			ra.err = restore_write(&ra, os, &drrw);
1054 			break;
1055 		}
1056 		case DRR_FREE:
1057 		{
1058 			struct drr_free drrf = drr->drr_u.drr_free;
1059 			ra.err = restore_free(&ra, os, &drrf);
1060 			break;
1061 		}
1062 		case DRR_END:
1063 		{
1064 			struct drr_end drre = drr->drr_u.drr_end;
1065 			/*
1066 			 * We compare against the *previous* checksum
1067 			 * value, because the stored checksum is of
1068 			 * everything before the DRR_END record.
1069 			 */
1070 			if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1071 				ra.err = ECKSUM;
1072 			goto out;
1073 		}
1074 		default:
1075 			ra.err = EINVAL;
1076 			goto out;
1077 		}
1078 		pcksum = ra.cksum;
1079 	}
1080 	ASSERT(ra.err != 0);
1081 
1082 out:
1083 	dmu_objset_close(os);
1084 
1085 	if (ra.err != 0) {
1086 		/*
1087 		 * rollback or destroy what we created, so we don't
1088 		 * leave it in the restoring state.
1089 		 */
1090 		txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1091 		dmu_recv_abort_cleanup(drc);
1092 	}
1093 
1094 	kmem_free(ra.buf, ra.bufsize);
1095 	*voffp = ra.voff;
1096 	return (ra.err);
1097 }
1098 
1099 struct recvendsyncarg {
1100 	char *tosnap;
1101 	uint64_t creation_time;
1102 	uint64_t toguid;
1103 };
1104 
1105 static int
1106 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1107 {
1108 	dsl_dataset_t *ds = arg1;
1109 	struct recvendsyncarg *resa = arg2;
1110 
1111 	return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1112 }
1113 
1114 static void
1115 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1116 {
1117 	dsl_dataset_t *ds = arg1;
1118 	struct recvendsyncarg *resa = arg2;
1119 
1120 	dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
1121 
1122 	/* set snapshot's creation time and guid */
1123 	dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1124 	ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1125 	ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1126 	ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1127 
1128 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1129 	ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1130 }
1131 
1132 int
1133 dmu_recv_end(dmu_recv_cookie_t *drc)
1134 {
1135 	struct recvendsyncarg resa;
1136 	dsl_dataset_t *ds = drc->drc_logical_ds;
1137 	int err;
1138 
1139 	/*
1140 	 * XXX hack; seems the ds is still dirty and
1141 	 * dsl_pool_zil_clean() expects it to have a ds_user_ptr
1142 	 * (and zil), but clone_swap() can close it.
1143 	 */
1144 	txg_wait_synced(ds->ds_dir->dd_pool, 0);
1145 
1146 	if (ds != drc->drc_real_ds) {
1147 		/* we are doing an online recv */
1148 		if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1149 			err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1150 			    drc->drc_force);
1151 			if (err)
1152 				dsl_dataset_disown(ds, dmu_recv_tag);
1153 		} else {
1154 			err = EBUSY;
1155 			dsl_dataset_rele(ds, dmu_recv_tag);
1156 		}
1157 		/* dsl_dataset_destroy() will disown the ds */
1158 		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
1159 		if (err)
1160 			return (err);
1161 	}
1162 
1163 	resa.creation_time = drc->drc_drrb->drr_creation_time;
1164 	resa.toguid = drc->drc_drrb->drr_toguid;
1165 	resa.tosnap = drc->drc_tosnap;
1166 
1167 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1168 	    recv_end_check, recv_end_sync, ds, &resa, 3);
1169 	if (err) {
1170 		if (drc->drc_newfs) {
1171 			ASSERT(ds == drc->drc_real_ds);
1172 			(void) dsl_dataset_destroy(ds, dmu_recv_tag);
1173 			return (err);
1174 		} else {
1175 			(void) dsl_dataset_rollback(ds, DMU_OST_NONE);
1176 		}
1177 	}
1178 
1179 	/* release the hold from dmu_recv_begin */
1180 	dsl_dataset_disown(ds, dmu_recv_tag);
1181 	return (err);
1182 }
1183