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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
26 #include <sys/dmu_impl.h>
27 #include <sys/dmu_tx.h>
29 #include <sys/dnode.h>
30 #include <sys/zfs_context.h>
31 #include <sys/dmu_objset.h>
32 #include <sys/dmu_traverse.h>
33 #include <sys/dsl_dataset.h>
34 #include <sys/dsl_dir.h>
35 #include <sys/dsl_prop.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/zfs_ioctl.h>
40 #include <sys/zio_checksum.h>
41 #include <sys/zfs_znode.h>
42 #include <zfs_fletcher.h>
45 #include <sys/zfs_onexit.h>
47 static char *dmu_recv_tag
= "dmu_recv_tag";
50 * The list of data whose inclusion in a send stream can be pending from
51 * one call to backup_cb to another. Multiple calls to dump_free() and
52 * dump_freeobjects() can be aggregated into a single DRR_FREE or
53 * DRR_FREEOBJECTS replay record.
62 dmu_replay_record_t
*drr
;
73 dump_bytes(struct backuparg
*ba
, void *buf
, int len
)
75 ssize_t resid
; /* have to get resid to get detailed errno */
76 ASSERT3U(len
% 8, ==, 0);
78 fletcher_4_incremental_native(buf
, len
, &ba
->zc
);
79 ba
->err
= vn_rdwr(UIO_WRITE
, ba
->vp
,
81 0, UIO_SYSSPACE
, FAPPEND
, RLIM64_INFINITY
, CRED(), &resid
);
87 dump_free(struct backuparg
*ba
, uint64_t object
, uint64_t offset
,
90 struct drr_free
*drrf
= &(ba
->drr
->drr_u
.drr_free
);
93 * If there is a pending op, but it's not PENDING_FREE, push it out,
94 * since free block aggregation can only be done for blocks of the
95 * same type (i.e., DRR_FREE records can only be aggregated with
96 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
97 * aggregated with other DRR_FREEOBJECTS records.
99 if (ba
->pending_op
!= PENDING_NONE
&& ba
->pending_op
!= PENDING_FREE
) {
100 if (dump_bytes(ba
, ba
->drr
, sizeof (dmu_replay_record_t
)) != 0)
102 ba
->pending_op
= PENDING_NONE
;
105 if (ba
->pending_op
== PENDING_FREE
) {
107 * There should never be a PENDING_FREE if length is -1
108 * (because dump_dnode is the only place where this
109 * function is called with a -1, and only after flushing
110 * any pending record).
112 ASSERT(length
!= -1ULL);
114 * Check to see whether this free block can be aggregated
117 if (drrf
->drr_object
== object
&& drrf
->drr_offset
+
118 drrf
->drr_length
== offset
) {
119 drrf
->drr_length
+= length
;
122 /* not a continuation. Push out pending record */
123 if (dump_bytes(ba
, ba
->drr
,
124 sizeof (dmu_replay_record_t
)) != 0)
126 ba
->pending_op
= PENDING_NONE
;
129 /* create a FREE record and make it pending */
130 bzero(ba
->drr
, sizeof (dmu_replay_record_t
));
131 ba
->drr
->drr_type
= DRR_FREE
;
132 drrf
->drr_object
= object
;
133 drrf
->drr_offset
= offset
;
134 drrf
->drr_length
= length
;
135 drrf
->drr_toguid
= ba
->toguid
;
136 if (length
== -1ULL) {
137 if (dump_bytes(ba
, ba
->drr
, sizeof (dmu_replay_record_t
)) != 0)
140 ba
->pending_op
= PENDING_FREE
;
147 dump_data(struct backuparg
*ba
, dmu_object_type_t type
,
148 uint64_t object
, uint64_t offset
, int blksz
, const blkptr_t
*bp
, void *data
)
150 struct drr_write
*drrw
= &(ba
->drr
->drr_u
.drr_write
);
154 * If there is any kind of pending aggregation (currently either
155 * a grouping of free objects or free blocks), push it out to
156 * the stream, since aggregation can't be done across operations
157 * of different types.
159 if (ba
->pending_op
!= PENDING_NONE
) {
160 if (dump_bytes(ba
, ba
->drr
, sizeof (dmu_replay_record_t
)) != 0)
162 ba
->pending_op
= PENDING_NONE
;
164 /* write a DATA record */
165 bzero(ba
->drr
, sizeof (dmu_replay_record_t
));
166 ba
->drr
->drr_type
= DRR_WRITE
;
167 drrw
->drr_object
= object
;
168 drrw
->drr_type
= type
;
169 drrw
->drr_offset
= offset
;
170 drrw
->drr_length
= blksz
;
171 drrw
->drr_toguid
= ba
->toguid
;
172 drrw
->drr_checksumtype
= BP_GET_CHECKSUM(bp
);
173 if (zio_checksum_table
[drrw
->drr_checksumtype
].ci_dedup
)
174 drrw
->drr_checksumflags
|= DRR_CHECKSUM_DEDUP
;
175 DDK_SET_LSIZE(&drrw
->drr_key
, BP_GET_LSIZE(bp
));
176 DDK_SET_PSIZE(&drrw
->drr_key
, BP_GET_PSIZE(bp
));
177 DDK_SET_COMPRESS(&drrw
->drr_key
, BP_GET_COMPRESS(bp
));
178 drrw
->drr_key
.ddk_cksum
= bp
->blk_cksum
;
180 if (dump_bytes(ba
, ba
->drr
, sizeof (dmu_replay_record_t
)) != 0)
182 if (dump_bytes(ba
, data
, blksz
) != 0)
188 dump_spill(struct backuparg
*ba
, uint64_t object
, int blksz
, void *data
)
190 struct drr_spill
*drrs
= &(ba
->drr
->drr_u
.drr_spill
);
192 if (ba
->pending_op
!= PENDING_NONE
) {
193 if (dump_bytes(ba
, ba
->drr
, sizeof (dmu_replay_record_t
)) != 0)
195 ba
->pending_op
= PENDING_NONE
;
198 /* write a SPILL record */
199 bzero(ba
->drr
, sizeof (dmu_replay_record_t
));
200 ba
->drr
->drr_type
= DRR_SPILL
;
201 drrs
->drr_object
= object
;
202 drrs
->drr_length
= blksz
;
203 drrs
->drr_toguid
= ba
->toguid
;
205 if (dump_bytes(ba
, ba
->drr
, sizeof (dmu_replay_record_t
)))
207 if (dump_bytes(ba
, data
, blksz
))
213 dump_freeobjects(struct backuparg
*ba
, uint64_t firstobj
, uint64_t numobjs
)
215 struct drr_freeobjects
*drrfo
= &(ba
->drr
->drr_u
.drr_freeobjects
);
218 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
219 * push it out, since free block aggregation can only be done for
220 * blocks of the same type (i.e., DRR_FREE records can only be
221 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
222 * can only be aggregated with other DRR_FREEOBJECTS records.
224 if (ba
->pending_op
!= PENDING_NONE
&&
225 ba
->pending_op
!= PENDING_FREEOBJECTS
) {
226 if (dump_bytes(ba
, ba
->drr
, sizeof (dmu_replay_record_t
)) != 0)
228 ba
->pending_op
= PENDING_NONE
;
230 if (ba
->pending_op
== PENDING_FREEOBJECTS
) {
232 * See whether this free object array can be aggregated
235 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
== firstobj
) {
236 drrfo
->drr_numobjs
+= numobjs
;
239 /* can't be aggregated. Push out pending record */
240 if (dump_bytes(ba
, ba
->drr
,
241 sizeof (dmu_replay_record_t
)) != 0)
243 ba
->pending_op
= PENDING_NONE
;
247 /* write a FREEOBJECTS record */
248 bzero(ba
->drr
, sizeof (dmu_replay_record_t
));
249 ba
->drr
->drr_type
= DRR_FREEOBJECTS
;
250 drrfo
->drr_firstobj
= firstobj
;
251 drrfo
->drr_numobjs
= numobjs
;
252 drrfo
->drr_toguid
= ba
->toguid
;
254 ba
->pending_op
= PENDING_FREEOBJECTS
;
260 dump_dnode(struct backuparg
*ba
, uint64_t object
, dnode_phys_t
*dnp
)
262 struct drr_object
*drro
= &(ba
->drr
->drr_u
.drr_object
);
264 if (dnp
== NULL
|| dnp
->dn_type
== DMU_OT_NONE
)
265 return (dump_freeobjects(ba
, object
, 1));
267 if (ba
->pending_op
!= PENDING_NONE
) {
268 if (dump_bytes(ba
, ba
->drr
, sizeof (dmu_replay_record_t
)) != 0)
270 ba
->pending_op
= PENDING_NONE
;
273 /* write an OBJECT record */
274 bzero(ba
->drr
, sizeof (dmu_replay_record_t
));
275 ba
->drr
->drr_type
= DRR_OBJECT
;
276 drro
->drr_object
= object
;
277 drro
->drr_type
= dnp
->dn_type
;
278 drro
->drr_bonustype
= dnp
->dn_bonustype
;
279 drro
->drr_blksz
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
280 drro
->drr_bonuslen
= dnp
->dn_bonuslen
;
281 drro
->drr_checksumtype
= dnp
->dn_checksum
;
282 drro
->drr_compress
= dnp
->dn_compress
;
283 drro
->drr_toguid
= ba
->toguid
;
285 if (dump_bytes(ba
, ba
->drr
, sizeof (dmu_replay_record_t
)) != 0)
288 if (dump_bytes(ba
, DN_BONUS(dnp
), P2ROUNDUP(dnp
->dn_bonuslen
, 8)) != 0)
291 /* free anything past the end of the file */
292 if (dump_free(ba
, object
, (dnp
->dn_maxblkid
+ 1) *
293 (dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
), -1ULL))
300 #define BP_SPAN(dnp, level) \
301 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
302 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
306 backup_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
, arc_buf_t
*pbuf
,
307 const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
309 struct backuparg
*ba
= arg
;
310 dmu_object_type_t type
= bp
? BP_GET_TYPE(bp
) : DMU_OT_NONE
;
313 if (issig(JUSTLOOKING
) && issig(FORREAL
))
316 if (zb
->zb_object
!= DMU_META_DNODE_OBJECT
&&
317 DMU_OBJECT_IS_SPECIAL(zb
->zb_object
)) {
319 } else if (bp
== NULL
&& zb
->zb_object
== DMU_META_DNODE_OBJECT
) {
320 uint64_t span
= BP_SPAN(dnp
, zb
->zb_level
);
321 uint64_t dnobj
= (zb
->zb_blkid
* span
) >> DNODE_SHIFT
;
322 err
= dump_freeobjects(ba
, dnobj
, span
>> DNODE_SHIFT
);
323 } else if (bp
== NULL
) {
324 uint64_t span
= BP_SPAN(dnp
, zb
->zb_level
);
325 err
= dump_free(ba
, zb
->zb_object
, zb
->zb_blkid
* span
, span
);
326 } else if (zb
->zb_level
> 0 || type
== DMU_OT_OBJSET
) {
328 } else if (type
== DMU_OT_DNODE
) {
331 int blksz
= BP_GET_LSIZE(bp
);
332 uint32_t aflags
= ARC_WAIT
;
335 if (dsl_read(NULL
, spa
, bp
, pbuf
,
336 arc_getbuf_func
, &abuf
, ZIO_PRIORITY_ASYNC_READ
,
337 ZIO_FLAG_CANFAIL
, &aflags
, zb
) != 0)
341 for (i
= 0; i
< blksz
>> DNODE_SHIFT
; i
++) {
342 uint64_t dnobj
= (zb
->zb_blkid
<<
343 (DNODE_BLOCK_SHIFT
- DNODE_SHIFT
)) + i
;
344 err
= dump_dnode(ba
, dnobj
, blk
+i
);
348 (void) arc_buf_remove_ref(abuf
, &abuf
);
349 } else if (type
== DMU_OT_SA
) {
350 uint32_t aflags
= ARC_WAIT
;
352 int blksz
= BP_GET_LSIZE(bp
);
354 if (arc_read_nolock(NULL
, spa
, bp
,
355 arc_getbuf_func
, &abuf
, ZIO_PRIORITY_ASYNC_READ
,
356 ZIO_FLAG_CANFAIL
, &aflags
, zb
) != 0)
359 err
= dump_spill(ba
, zb
->zb_object
, blksz
, abuf
->b_data
);
360 (void) arc_buf_remove_ref(abuf
, &abuf
);
361 } else { /* it's a level-0 block of a regular object */
362 uint32_t aflags
= ARC_WAIT
;
364 int blksz
= BP_GET_LSIZE(bp
);
366 if (dsl_read(NULL
, spa
, bp
, pbuf
,
367 arc_getbuf_func
, &abuf
, ZIO_PRIORITY_ASYNC_READ
,
368 ZIO_FLAG_CANFAIL
, &aflags
, zb
) != 0)
371 err
= dump_data(ba
, type
, zb
->zb_object
, zb
->zb_blkid
* blksz
,
372 blksz
, bp
, abuf
->b_data
);
373 (void) arc_buf_remove_ref(abuf
, &abuf
);
376 ASSERT(err
== 0 || err
== EINTR
);
381 dmu_sendbackup(objset_t
*tosnap
, objset_t
*fromsnap
, boolean_t fromorigin
,
382 vnode_t
*vp
, offset_t
*off
)
384 dsl_dataset_t
*ds
= tosnap
->os_dsl_dataset
;
385 dsl_dataset_t
*fromds
= fromsnap
? fromsnap
->os_dsl_dataset
: NULL
;
386 dmu_replay_record_t
*drr
;
389 uint64_t fromtxg
= 0;
391 /* tosnap must be a snapshot */
392 if (ds
->ds_phys
->ds_next_snap_obj
== 0)
395 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
396 if (fromds
&& (ds
->ds_dir
!= fromds
->ds_dir
||
397 fromds
->ds_phys
->ds_creation_txg
>= ds
->ds_phys
->ds_creation_txg
))
401 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
406 if (dsl_dir_is_clone(ds
->ds_dir
)) {
407 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
408 err
= dsl_dataset_hold_obj(dp
,
409 ds
->ds_dir
->dd_phys
->dd_origin_obj
, FTAG
, &fromds
);
410 rw_exit(&dp
->dp_config_rwlock
);
414 fromorigin
= B_FALSE
;
419 drr
= kmem_zalloc(sizeof (dmu_replay_record_t
), KM_SLEEP
);
420 drr
->drr_type
= DRR_BEGIN
;
421 drr
->drr_u
.drr_begin
.drr_magic
= DMU_BACKUP_MAGIC
;
422 DMU_SET_STREAM_HDRTYPE(drr
->drr_u
.drr_begin
.drr_versioninfo
,
426 if (dmu_objset_type(tosnap
) == DMU_OST_ZFS
) {
428 if (zfs_get_zplprop(tosnap
, ZFS_PROP_VERSION
, &version
) != 0)
430 if (version
== ZPL_VERSION_SA
) {
431 DMU_SET_FEATUREFLAGS(
432 drr
->drr_u
.drr_begin
.drr_versioninfo
,
433 DMU_BACKUP_FEATURE_SA_SPILL
);
438 drr
->drr_u
.drr_begin
.drr_creation_time
=
439 ds
->ds_phys
->ds_creation_time
;
440 drr
->drr_u
.drr_begin
.drr_type
= tosnap
->os_phys
->os_type
;
442 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CLONE
;
443 drr
->drr_u
.drr_begin
.drr_toguid
= ds
->ds_phys
->ds_guid
;
444 if (ds
->ds_phys
->ds_flags
& DS_FLAG_CI_DATASET
)
445 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CI_DATA
;
448 drr
->drr_u
.drr_begin
.drr_fromguid
= fromds
->ds_phys
->ds_guid
;
449 dsl_dataset_name(ds
, drr
->drr_u
.drr_begin
.drr_toname
);
452 fromtxg
= fromds
->ds_phys
->ds_creation_txg
;
454 dsl_dataset_rele(fromds
, FTAG
);
460 ba
.toguid
= ds
->ds_phys
->ds_guid
;
461 ZIO_SET_CHECKSUM(&ba
.zc
, 0, 0, 0, 0);
462 ba
.pending_op
= PENDING_NONE
;
464 if (dump_bytes(&ba
, drr
, sizeof (dmu_replay_record_t
)) != 0) {
465 kmem_free(drr
, sizeof (dmu_replay_record_t
));
469 err
= traverse_dataset(ds
, fromtxg
, TRAVERSE_PRE
| TRAVERSE_PREFETCH
,
472 if (ba
.pending_op
!= PENDING_NONE
)
473 if (dump_bytes(&ba
, drr
, sizeof (dmu_replay_record_t
)) != 0)
477 if (err
== EINTR
&& ba
.err
)
479 kmem_free(drr
, sizeof (dmu_replay_record_t
));
483 bzero(drr
, sizeof (dmu_replay_record_t
));
484 drr
->drr_type
= DRR_END
;
485 drr
->drr_u
.drr_end
.drr_checksum
= ba
.zc
;
486 drr
->drr_u
.drr_end
.drr_toguid
= ba
.toguid
;
488 if (dump_bytes(&ba
, drr
, sizeof (dmu_replay_record_t
)) != 0) {
489 kmem_free(drr
, sizeof (dmu_replay_record_t
));
493 kmem_free(drr
, sizeof (dmu_replay_record_t
));
498 struct recvbeginsyncarg
{
501 dsl_dataset_t
*origin
;
503 dmu_objset_type_t type
;
507 char clonelastname
[MAXNAMELEN
];
508 dsl_dataset_t
*ds
; /* the ds to recv into; returned from the syncfunc */
514 recv_new_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
516 dsl_dir_t
*dd
= arg1
;
517 struct recvbeginsyncarg
*rbsa
= arg2
;
518 objset_t
*mos
= dd
->dd_pool
->dp_meta_objset
;
522 err
= zap_lookup(mos
, dd
->dd_phys
->dd_child_dir_zapobj
,
523 strrchr(rbsa
->tofs
, '/') + 1, sizeof (uint64_t), 1, &val
);
526 return (err
? err
: EEXIST
);
529 /* make sure it's a snap in the same pool */
530 if (rbsa
->origin
->ds_dir
->dd_pool
!= dd
->dd_pool
)
532 if (!dsl_dataset_is_snapshot(rbsa
->origin
))
534 if (rbsa
->origin
->ds_phys
->ds_guid
!= rbsa
->fromguid
)
542 recv_new_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
544 dsl_dir_t
*dd
= arg1
;
545 struct recvbeginsyncarg
*rbsa
= arg2
;
546 uint64_t flags
= DS_FLAG_INCONSISTENT
| rbsa
->dsflags
;
549 /* Create and open new dataset. */
550 dsobj
= dsl_dataset_create_sync(dd
, strrchr(rbsa
->tofs
, '/') + 1,
551 rbsa
->origin
, flags
, rbsa
->cr
, tx
);
552 VERIFY(0 == dsl_dataset_own_obj(dd
->dd_pool
, dsobj
,
553 B_TRUE
, dmu_recv_tag
, &rbsa
->ds
));
555 if (rbsa
->origin
== NULL
) {
556 (void) dmu_objset_create_impl(dd
->dd_pool
->dp_spa
,
557 rbsa
->ds
, &rbsa
->ds
->ds_phys
->ds_bp
, rbsa
->type
, tx
);
560 spa_history_log_internal(LOG_DS_REPLAY_FULL_SYNC
,
561 dd
->dd_pool
->dp_spa
, tx
, "dataset = %lld", dsobj
);
566 recv_existing_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
568 dsl_dataset_t
*ds
= arg1
;
569 struct recvbeginsyncarg
*rbsa
= arg2
;
573 /* must not have any changes since most recent snapshot */
574 if (!rbsa
->force
&& dsl_dataset_modified_since_lastsnap(ds
))
577 /* new snapshot name must not exist */
578 err
= zap_lookup(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
579 ds
->ds_phys
->ds_snapnames_zapobj
, rbsa
->tosnap
, 8, 1, &val
);
585 if (rbsa
->fromguid
) {
586 /* if incremental, most recent snapshot must match fromguid */
587 if (ds
->ds_prev
== NULL
)
591 * most recent snapshot must match fromguid, or there are no
592 * changes since the fromguid one
594 if (ds
->ds_prev
->ds_phys
->ds_guid
!= rbsa
->fromguid
) {
595 uint64_t birth
= ds
->ds_prev
->ds_phys
->ds_bp
.blk_birth
;
596 uint64_t obj
= ds
->ds_prev
->ds_phys
->ds_prev_snap_obj
;
599 err
= dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
603 if (snap
->ds_phys
->ds_creation_txg
< birth
) {
604 dsl_dataset_rele(snap
, FTAG
);
607 if (snap
->ds_phys
->ds_guid
== rbsa
->fromguid
) {
608 dsl_dataset_rele(snap
, FTAG
);
611 obj
= snap
->ds_phys
->ds_prev_snap_obj
;
612 dsl_dataset_rele(snap
, FTAG
);
618 /* if full, most recent snapshot must be $ORIGIN */
619 if (ds
->ds_phys
->ds_prev_snap_txg
>= TXG_INITIAL
)
623 /* temporary clone name must not exist */
624 err
= zap_lookup(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
625 ds
->ds_dir
->dd_phys
->dd_child_dir_zapobj
,
626 rbsa
->clonelastname
, 8, 1, &val
);
637 recv_existing_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
639 dsl_dataset_t
*ohds
= arg1
;
640 struct recvbeginsyncarg
*rbsa
= arg2
;
641 dsl_pool_t
*dp
= ohds
->ds_dir
->dd_pool
;
643 uint64_t flags
= DS_FLAG_INCONSISTENT
| rbsa
->dsflags
;
646 /* create and open the temporary clone */
647 dsobj
= dsl_dataset_create_sync(ohds
->ds_dir
, rbsa
->clonelastname
,
648 ohds
->ds_prev
, flags
, rbsa
->cr
, tx
);
649 VERIFY(0 == dsl_dataset_own_obj(dp
, dsobj
, B_TRUE
, dmu_recv_tag
, &cds
));
652 * If we actually created a non-clone, we need to create the
653 * objset in our new dataset.
655 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds
))) {
656 (void) dmu_objset_create_impl(dp
->dp_spa
,
657 cds
, dsl_dataset_get_blkptr(cds
), rbsa
->type
, tx
);
662 spa_history_log_internal(LOG_DS_REPLAY_INC_SYNC
,
663 dp
->dp_spa
, tx
, "dataset = %lld", dsobj
);
667 dmu_recv_verify_features(dsl_dataset_t
*ds
, struct drr_begin
*drrb
)
671 featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
673 /* Verify pool version supports SA if SA_SPILL feature set */
674 return ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
675 (spa_version(dsl_dataset_get_spa(ds
)) < SPA_VERSION_SA
));
679 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
680 * succeeds; otherwise we will leak the holds on the datasets.
683 dmu_recv_begin(char *tofs
, char *tosnap
, char *top_ds
, struct drr_begin
*drrb
,
684 boolean_t force
, objset_t
*origin
, dmu_recv_cookie_t
*drc
)
688 struct recvbeginsyncarg rbsa
= { 0 };
689 uint64_t versioninfo
;
693 if (drrb
->drr_magic
== DMU_BACKUP_MAGIC
)
695 else if (drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
))
701 rbsa
.tosnap
= tosnap
;
702 rbsa
.origin
= origin
? origin
->os_dsl_dataset
: NULL
;
703 rbsa
.fromguid
= drrb
->drr_fromguid
;
704 rbsa
.type
= drrb
->drr_type
;
708 versioninfo
= drrb
->drr_versioninfo
;
709 flags
= drrb
->drr_flags
;
712 rbsa
.type
= BSWAP_32(rbsa
.type
);
713 rbsa
.fromguid
= BSWAP_64(rbsa
.fromguid
);
714 versioninfo
= BSWAP_64(versioninfo
);
715 flags
= BSWAP_32(flags
);
718 if (DMU_GET_STREAM_HDRTYPE(versioninfo
) == DMU_COMPOUNDSTREAM
||
719 rbsa
.type
>= DMU_OST_NUMTYPES
||
720 ((flags
& DRR_FLAG_CLONE
) && origin
== NULL
))
723 if (flags
& DRR_FLAG_CI_DATA
)
724 rbsa
.dsflags
= DS_FLAG_CI_DATASET
;
726 bzero(drc
, sizeof (dmu_recv_cookie_t
));
727 drc
->drc_drrb
= drrb
;
728 drc
->drc_tosnap
= tosnap
;
729 drc
->drc_top_ds
= top_ds
;
730 drc
->drc_force
= force
;
733 * Process the begin in syncing context.
736 /* open the dataset we are logically receiving into */
737 err
= dsl_dataset_hold(tofs
, dmu_recv_tag
, &ds
);
739 if (dmu_recv_verify_features(ds
, drrb
)) {
740 dsl_dataset_rele(ds
, dmu_recv_tag
);
743 /* target fs already exists; recv into temp clone */
745 /* Can't recv a clone into an existing fs */
746 if (flags
& DRR_FLAG_CLONE
) {
747 dsl_dataset_rele(ds
, dmu_recv_tag
);
751 /* must not have an incremental recv already in progress */
752 if (!mutex_tryenter(&ds
->ds_recvlock
)) {
753 dsl_dataset_rele(ds
, dmu_recv_tag
);
757 /* tmp clone name is: tofs/%tosnap" */
758 (void) snprintf(rbsa
.clonelastname
, sizeof (rbsa
.clonelastname
),
761 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
762 recv_existing_check
, recv_existing_sync
, ds
, &rbsa
, 5);
764 mutex_exit(&ds
->ds_recvlock
);
765 dsl_dataset_rele(ds
, dmu_recv_tag
);
768 drc
->drc_logical_ds
= ds
;
769 drc
->drc_real_ds
= rbsa
.ds
;
770 } else if (err
== ENOENT
) {
771 /* target fs does not exist; must be a full backup or clone */
775 * If it's a non-clone incremental, we are missing the
776 * target fs, so fail the recv.
778 if (rbsa
.fromguid
&& !(flags
& DRR_FLAG_CLONE
))
781 /* Open the parent of tofs */
782 cp
= strrchr(tofs
, '/');
784 err
= dsl_dataset_hold(tofs
, FTAG
, &ds
);
789 if (dmu_recv_verify_features(ds
, drrb
)) {
790 dsl_dataset_rele(ds
, FTAG
);
794 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
795 recv_new_check
, recv_new_sync
, ds
->ds_dir
, &rbsa
, 5);
796 dsl_dataset_rele(ds
, FTAG
);
799 drc
->drc_logical_ds
= drc
->drc_real_ds
= rbsa
.ds
;
800 drc
->drc_newfs
= B_TRUE
;
812 int bufsize
; /* amount of memory allocated for buf */
814 avl_tree_t
*guid_to_ds_map
;
817 typedef struct guid_map_entry
{
819 dsl_dataset_t
*gme_ds
;
824 guid_compare(const void *arg1
, const void *arg2
)
826 const guid_map_entry_t
*gmep1
= arg1
;
827 const guid_map_entry_t
*gmep2
= arg2
;
829 if (gmep1
->guid
< gmep2
->guid
)
831 else if (gmep1
->guid
> gmep2
->guid
)
837 * This function is a callback used by dmu_objset_find() (which
838 * enumerates the object sets) to build an avl tree that maps guids
839 * to datasets. The resulting table is used when processing DRR_WRITE_BYREF
840 * send stream records. These records, which are used in dedup'ed
841 * streams, do not contain data themselves, but refer to a copy
842 * of the data block that has already been written because it was
843 * earlier in the stream. That previous copy is identified by the
844 * guid of the dataset with the referenced data.
847 find_ds_by_guid(const char *name
, void *arg
)
849 avl_tree_t
*guid_map
= arg
;
850 dsl_dataset_t
*ds
, *snapds
;
851 guid_map_entry_t
*gmep
;
854 uint64_t lastobj
, firstobj
;
856 if (dsl_dataset_hold(name
, FTAG
, &ds
) != 0)
859 dp
= ds
->ds_dir
->dd_pool
;
860 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
861 firstobj
= ds
->ds_dir
->dd_phys
->dd_origin_obj
;
862 lastobj
= ds
->ds_phys
->ds_prev_snap_obj
;
864 while (lastobj
!= firstobj
) {
865 err
= dsl_dataset_hold_obj(dp
, lastobj
, guid_map
, &snapds
);
868 * Skip this snapshot and move on. It's not
869 * clear why this would ever happen, but the
870 * remainder of the snapshot streadm can be
873 rw_exit(&dp
->dp_config_rwlock
);
874 dsl_dataset_rele(ds
, FTAG
);
878 gmep
= kmem_alloc(sizeof (guid_map_entry_t
), KM_SLEEP
);
879 gmep
->guid
= snapds
->ds_phys
->ds_guid
;
880 gmep
->gme_ds
= snapds
;
881 avl_add(guid_map
, gmep
);
882 lastobj
= snapds
->ds_phys
->ds_prev_snap_obj
;
885 rw_exit(&dp
->dp_config_rwlock
);
886 dsl_dataset_rele(ds
, FTAG
);
892 free_guid_map_onexit(void *arg
)
894 avl_tree_t
*ca
= arg
;
896 guid_map_entry_t
*gmep
;
898 while ((gmep
= avl_destroy_nodes(ca
, &cookie
)) != NULL
) {
899 dsl_dataset_rele(gmep
->gme_ds
, ca
);
900 kmem_free(gmep
, sizeof (guid_map_entry_t
));
903 kmem_free(ca
, sizeof (avl_tree_t
));
907 restore_read(struct restorearg
*ra
, int len
)
912 /* some things will require 8-byte alignment, so everything must */
913 ASSERT3U(len
% 8, ==, 0);
918 ra
->err
= vn_rdwr(UIO_READ
, ra
->vp
,
919 (caddr_t
)ra
->buf
+ done
, len
- done
,
920 ra
->voff
, UIO_SYSSPACE
, FAPPEND
,
921 RLIM64_INFINITY
, CRED(), &resid
);
923 if (resid
== len
- done
)
925 ra
->voff
+= len
- done
- resid
;
931 ASSERT3U(done
, ==, len
);
934 fletcher_4_incremental_byteswap(rv
, len
, &ra
->cksum
);
936 fletcher_4_incremental_native(rv
, len
, &ra
->cksum
);
941 backup_byteswap(dmu_replay_record_t
*drr
)
943 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
944 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
945 drr
->drr_type
= BSWAP_32(drr
->drr_type
);
946 drr
->drr_payloadlen
= BSWAP_32(drr
->drr_payloadlen
);
947 switch (drr
->drr_type
) {
949 DO64(drr_begin
.drr_magic
);
950 DO64(drr_begin
.drr_versioninfo
);
951 DO64(drr_begin
.drr_creation_time
);
952 DO32(drr_begin
.drr_type
);
953 DO32(drr_begin
.drr_flags
);
954 DO64(drr_begin
.drr_toguid
);
955 DO64(drr_begin
.drr_fromguid
);
958 DO64(drr_object
.drr_object
);
959 /* DO64(drr_object.drr_allocation_txg); */
960 DO32(drr_object
.drr_type
);
961 DO32(drr_object
.drr_bonustype
);
962 DO32(drr_object
.drr_blksz
);
963 DO32(drr_object
.drr_bonuslen
);
964 DO64(drr_object
.drr_toguid
);
966 case DRR_FREEOBJECTS
:
967 DO64(drr_freeobjects
.drr_firstobj
);
968 DO64(drr_freeobjects
.drr_numobjs
);
969 DO64(drr_freeobjects
.drr_toguid
);
972 DO64(drr_write
.drr_object
);
973 DO32(drr_write
.drr_type
);
974 DO64(drr_write
.drr_offset
);
975 DO64(drr_write
.drr_length
);
976 DO64(drr_write
.drr_toguid
);
977 DO64(drr_write
.drr_key
.ddk_cksum
.zc_word
[0]);
978 DO64(drr_write
.drr_key
.ddk_cksum
.zc_word
[1]);
979 DO64(drr_write
.drr_key
.ddk_cksum
.zc_word
[2]);
980 DO64(drr_write
.drr_key
.ddk_cksum
.zc_word
[3]);
981 DO64(drr_write
.drr_key
.ddk_prop
);
983 case DRR_WRITE_BYREF
:
984 DO64(drr_write_byref
.drr_object
);
985 DO64(drr_write_byref
.drr_offset
);
986 DO64(drr_write_byref
.drr_length
);
987 DO64(drr_write_byref
.drr_toguid
);
988 DO64(drr_write_byref
.drr_refguid
);
989 DO64(drr_write_byref
.drr_refobject
);
990 DO64(drr_write_byref
.drr_refoffset
);
991 DO64(drr_write_byref
.drr_key
.ddk_cksum
.zc_word
[0]);
992 DO64(drr_write_byref
.drr_key
.ddk_cksum
.zc_word
[1]);
993 DO64(drr_write_byref
.drr_key
.ddk_cksum
.zc_word
[2]);
994 DO64(drr_write_byref
.drr_key
.ddk_cksum
.zc_word
[3]);
995 DO64(drr_write_byref
.drr_key
.ddk_prop
);
998 DO64(drr_free
.drr_object
);
999 DO64(drr_free
.drr_offset
);
1000 DO64(drr_free
.drr_length
);
1001 DO64(drr_free
.drr_toguid
);
1004 DO64(drr_spill
.drr_object
);
1005 DO64(drr_spill
.drr_length
);
1006 DO64(drr_spill
.drr_toguid
);
1009 DO64(drr_end
.drr_checksum
.zc_word
[0]);
1010 DO64(drr_end
.drr_checksum
.zc_word
[1]);
1011 DO64(drr_end
.drr_checksum
.zc_word
[2]);
1012 DO64(drr_end
.drr_checksum
.zc_word
[3]);
1013 DO64(drr_end
.drr_toguid
);
1021 restore_object(struct restorearg
*ra
, objset_t
*os
, struct drr_object
*drro
)
1027 if (drro
->drr_type
== DMU_OT_NONE
||
1028 drro
->drr_type
>= DMU_OT_NUMTYPES
||
1029 drro
->drr_bonustype
>= DMU_OT_NUMTYPES
||
1030 drro
->drr_checksumtype
>= ZIO_CHECKSUM_FUNCTIONS
||
1031 drro
->drr_compress
>= ZIO_COMPRESS_FUNCTIONS
||
1032 P2PHASE(drro
->drr_blksz
, SPA_MINBLOCKSIZE
) ||
1033 drro
->drr_blksz
< SPA_MINBLOCKSIZE
||
1034 drro
->drr_blksz
> SPA_MAXBLOCKSIZE
||
1035 drro
->drr_bonuslen
> DN_MAX_BONUSLEN
) {
1039 err
= dmu_object_info(os
, drro
->drr_object
, NULL
);
1041 if (err
!= 0 && err
!= ENOENT
)
1044 if (drro
->drr_bonuslen
) {
1045 data
= restore_read(ra
, P2ROUNDUP(drro
->drr_bonuslen
, 8));
1050 if (err
== ENOENT
) {
1051 /* currently free, want to be allocated */
1052 tx
= dmu_tx_create(os
);
1053 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1054 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1059 err
= dmu_object_claim(os
, drro
->drr_object
,
1060 drro
->drr_type
, drro
->drr_blksz
,
1061 drro
->drr_bonustype
, drro
->drr_bonuslen
, tx
);
1064 /* currently allocated, want to be allocated */
1065 err
= dmu_object_reclaim(os
, drro
->drr_object
,
1066 drro
->drr_type
, drro
->drr_blksz
,
1067 drro
->drr_bonustype
, drro
->drr_bonuslen
);
1073 tx
= dmu_tx_create(os
);
1074 dmu_tx_hold_bonus(tx
, drro
->drr_object
);
1075 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1081 dmu_object_set_checksum(os
, drro
->drr_object
, drro
->drr_checksumtype
,
1083 dmu_object_set_compress(os
, drro
->drr_object
, drro
->drr_compress
, tx
);
1088 VERIFY(0 == dmu_bonus_hold(os
, drro
->drr_object
, FTAG
, &db
));
1089 dmu_buf_will_dirty(db
, tx
);
1091 ASSERT3U(db
->db_size
, >=, drro
->drr_bonuslen
);
1092 bcopy(data
, db
->db_data
, drro
->drr_bonuslen
);
1094 dmu_ot
[drro
->drr_bonustype
].ot_byteswap(db
->db_data
,
1095 drro
->drr_bonuslen
);
1097 dmu_buf_rele(db
, FTAG
);
1105 restore_freeobjects(struct restorearg
*ra
, objset_t
*os
,
1106 struct drr_freeobjects
*drrfo
)
1110 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
< drrfo
->drr_firstobj
)
1113 for (obj
= drrfo
->drr_firstobj
;
1114 obj
< drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
;
1115 (void) dmu_object_next(os
, &obj
, FALSE
, 0)) {
1118 if (dmu_object_info(os
, obj
, NULL
) != 0)
1121 err
= dmu_free_object(os
, obj
);
1129 restore_write(struct restorearg
*ra
, objset_t
*os
,
1130 struct drr_write
*drrw
)
1136 if (drrw
->drr_offset
+ drrw
->drr_length
< drrw
->drr_offset
||
1137 drrw
->drr_type
>= DMU_OT_NUMTYPES
)
1140 data
= restore_read(ra
, drrw
->drr_length
);
1144 if (dmu_object_info(os
, drrw
->drr_object
, NULL
) != 0)
1147 tx
= dmu_tx_create(os
);
1149 dmu_tx_hold_write(tx
, drrw
->drr_object
,
1150 drrw
->drr_offset
, drrw
->drr_length
);
1151 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1157 dmu_ot
[drrw
->drr_type
].ot_byteswap(data
, drrw
->drr_length
);
1158 dmu_write(os
, drrw
->drr_object
,
1159 drrw
->drr_offset
, drrw
->drr_length
, data
, tx
);
1165 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1166 * streams to refer to a copy of the data that is already on the
1167 * system because it came in earlier in the stream. This function
1168 * finds the earlier copy of the data, and uses that copy instead of
1169 * data from the stream to fulfill this write.
1172 restore_write_byref(struct restorearg
*ra
, objset_t
*os
,
1173 struct drr_write_byref
*drrwbr
)
1177 guid_map_entry_t gmesrch
;
1178 guid_map_entry_t
*gmep
;
1180 objset_t
*ref_os
= NULL
;
1183 if (drrwbr
->drr_offset
+ drrwbr
->drr_length
< drrwbr
->drr_offset
)
1187 * If the GUID of the referenced dataset is different from the
1188 * GUID of the target dataset, find the referenced dataset.
1190 if (drrwbr
->drr_toguid
!= drrwbr
->drr_refguid
) {
1191 gmesrch
.guid
= drrwbr
->drr_refguid
;
1192 if ((gmep
= avl_find(ra
->guid_to_ds_map
, &gmesrch
,
1196 if (dmu_objset_from_ds(gmep
->gme_ds
, &ref_os
))
1202 if (err
= dmu_buf_hold(ref_os
, drrwbr
->drr_refobject
,
1203 drrwbr
->drr_refoffset
, FTAG
, &dbp
, DMU_READ_PREFETCH
))
1206 tx
= dmu_tx_create(os
);
1208 dmu_tx_hold_write(tx
, drrwbr
->drr_object
,
1209 drrwbr
->drr_offset
, drrwbr
->drr_length
);
1210 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1215 dmu_write(os
, drrwbr
->drr_object
,
1216 drrwbr
->drr_offset
, drrwbr
->drr_length
, dbp
->db_data
, tx
);
1217 dmu_buf_rele(dbp
, FTAG
);
1223 restore_spill(struct restorearg
*ra
, objset_t
*os
, struct drr_spill
*drrs
)
1227 dmu_buf_t
*db
, *db_spill
;
1230 if (drrs
->drr_length
< SPA_MINBLOCKSIZE
||
1231 drrs
->drr_length
> SPA_MAXBLOCKSIZE
)
1234 data
= restore_read(ra
, drrs
->drr_length
);
1238 if (dmu_object_info(os
, drrs
->drr_object
, NULL
) != 0)
1241 VERIFY(0 == dmu_bonus_hold(os
, drrs
->drr_object
, FTAG
, &db
));
1242 if ((err
= dmu_spill_hold_by_bonus(db
, FTAG
, &db_spill
)) != 0) {
1243 dmu_buf_rele(db
, FTAG
);
1247 tx
= dmu_tx_create(os
);
1249 dmu_tx_hold_spill(tx
, db
->db_object
);
1251 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1253 dmu_buf_rele(db
, FTAG
);
1254 dmu_buf_rele(db_spill
, FTAG
);
1258 dmu_buf_will_dirty(db_spill
, tx
);
1260 if (db_spill
->db_size
< drrs
->drr_length
)
1261 VERIFY(0 == dbuf_spill_set_blksz(db_spill
,
1262 drrs
->drr_length
, tx
));
1263 bcopy(data
, db_spill
->db_data
, drrs
->drr_length
);
1265 dmu_buf_rele(db
, FTAG
);
1266 dmu_buf_rele(db_spill
, FTAG
);
1274 restore_free(struct restorearg
*ra
, objset_t
*os
,
1275 struct drr_free
*drrf
)
1279 if (drrf
->drr_length
!= -1ULL &&
1280 drrf
->drr_offset
+ drrf
->drr_length
< drrf
->drr_offset
)
1283 if (dmu_object_info(os
, drrf
->drr_object
, NULL
) != 0)
1286 err
= dmu_free_long_range(os
, drrf
->drr_object
,
1287 drrf
->drr_offset
, drrf
->drr_length
);
1292 * NB: callers *must* call dmu_recv_end() if this succeeds.
1295 dmu_recv_stream(dmu_recv_cookie_t
*drc
, vnode_t
*vp
, offset_t
*voffp
,
1296 int cleanup_fd
, uint64_t *action_handlep
)
1298 struct restorearg ra
= { 0 };
1299 dmu_replay_record_t
*drr
;
1304 if (drc
->drc_drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
))
1308 /* compute checksum of drr_begin record */
1309 dmu_replay_record_t
*drr
;
1310 drr
= kmem_zalloc(sizeof (dmu_replay_record_t
), KM_SLEEP
);
1312 drr
->drr_type
= DRR_BEGIN
;
1313 drr
->drr_u
.drr_begin
= *drc
->drc_drrb
;
1315 fletcher_4_incremental_byteswap(drr
,
1316 sizeof (dmu_replay_record_t
), &ra
.cksum
);
1318 fletcher_4_incremental_native(drr
,
1319 sizeof (dmu_replay_record_t
), &ra
.cksum
);
1321 kmem_free(drr
, sizeof (dmu_replay_record_t
));
1325 struct drr_begin
*drrb
= drc
->drc_drrb
;
1326 drrb
->drr_magic
= BSWAP_64(drrb
->drr_magic
);
1327 drrb
->drr_versioninfo
= BSWAP_64(drrb
->drr_versioninfo
);
1328 drrb
->drr_creation_time
= BSWAP_64(drrb
->drr_creation_time
);
1329 drrb
->drr_type
= BSWAP_32(drrb
->drr_type
);
1330 drrb
->drr_toguid
= BSWAP_64(drrb
->drr_toguid
);
1331 drrb
->drr_fromguid
= BSWAP_64(drrb
->drr_fromguid
);
1337 ra
.buf
= kmem_alloc(ra
.bufsize
, KM_SLEEP
);
1339 /* these were verified in dmu_recv_begin */
1340 ASSERT(DMU_GET_STREAM_HDRTYPE(drc
->drc_drrb
->drr_versioninfo
) ==
1342 ASSERT(drc
->drc_drrb
->drr_type
< DMU_OST_NUMTYPES
);
1345 * Open the objset we are modifying.
1347 VERIFY(dmu_objset_from_ds(drc
->drc_real_ds
, &os
) == 0);
1349 ASSERT(drc
->drc_real_ds
->ds_phys
->ds_flags
& DS_FLAG_INCONSISTENT
);
1351 featureflags
= DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
);
1353 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1354 if (featureflags
& DMU_BACKUP_FEATURE_DEDUP
) {
1357 if (cleanup_fd
== -1) {
1361 ra
.err
= zfs_onexit_fd_hold(cleanup_fd
, &minor
);
1367 if (*action_handlep
== 0) {
1369 kmem_alloc(sizeof (avl_tree_t
), KM_SLEEP
);
1370 avl_create(ra
.guid_to_ds_map
, guid_compare
,
1371 sizeof (guid_map_entry_t
),
1372 offsetof(guid_map_entry_t
, avlnode
));
1373 (void) dmu_objset_find(drc
->drc_top_ds
, find_ds_by_guid
,
1374 (void *)ra
.guid_to_ds_map
,
1376 ra
.err
= zfs_onexit_add_cb(minor
,
1377 free_guid_map_onexit
, ra
.guid_to_ds_map
,
1382 ra
.err
= zfs_onexit_cb_data(minor
, *action_handlep
,
1383 (void **)&ra
.guid_to_ds_map
);
1390 * Read records and process them.
1393 while (ra
.err
== 0 &&
1394 NULL
!= (drr
= restore_read(&ra
, sizeof (*drr
)))) {
1395 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
1401 backup_byteswap(drr
);
1403 switch (drr
->drr_type
) {
1407 * We need to make a copy of the record header,
1408 * because restore_{object,write} may need to
1409 * restore_read(), which will invalidate drr.
1411 struct drr_object drro
= drr
->drr_u
.drr_object
;
1412 ra
.err
= restore_object(&ra
, os
, &drro
);
1415 case DRR_FREEOBJECTS
:
1417 struct drr_freeobjects drrfo
=
1418 drr
->drr_u
.drr_freeobjects
;
1419 ra
.err
= restore_freeobjects(&ra
, os
, &drrfo
);
1424 struct drr_write drrw
= drr
->drr_u
.drr_write
;
1425 ra
.err
= restore_write(&ra
, os
, &drrw
);
1428 case DRR_WRITE_BYREF
:
1430 struct drr_write_byref drrwbr
=
1431 drr
->drr_u
.drr_write_byref
;
1432 ra
.err
= restore_write_byref(&ra
, os
, &drrwbr
);
1437 struct drr_free drrf
= drr
->drr_u
.drr_free
;
1438 ra
.err
= restore_free(&ra
, os
, &drrf
);
1443 struct drr_end drre
= drr
->drr_u
.drr_end
;
1445 * We compare against the *previous* checksum
1446 * value, because the stored checksum is of
1447 * everything before the DRR_END record.
1449 if (!ZIO_CHECKSUM_EQUAL(drre
.drr_checksum
, pcksum
))
1455 struct drr_spill drrs
= drr
->drr_u
.drr_spill
;
1456 ra
.err
= restore_spill(&ra
, os
, &drrs
);
1465 ASSERT(ra
.err
!= 0);
1468 if ((featureflags
& DMU_BACKUP_FEATURE_DEDUP
) && (cleanup_fd
!= -1))
1469 zfs_onexit_fd_rele(cleanup_fd
);
1473 * destroy what we created, so we don't leave it in the
1474 * inconsistent restoring state.
1476 txg_wait_synced(drc
->drc_real_ds
->ds_dir
->dd_pool
, 0);
1478 (void) dsl_dataset_destroy(drc
->drc_real_ds
, dmu_recv_tag
,
1480 if (drc
->drc_real_ds
!= drc
->drc_logical_ds
) {
1481 mutex_exit(&drc
->drc_logical_ds
->ds_recvlock
);
1482 dsl_dataset_rele(drc
->drc_logical_ds
, dmu_recv_tag
);
1486 kmem_free(ra
.buf
, ra
.bufsize
);
1491 struct recvendsyncarg
{
1493 uint64_t creation_time
;
1498 recv_end_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
1500 dsl_dataset_t
*ds
= arg1
;
1501 struct recvendsyncarg
*resa
= arg2
;
1503 return (dsl_dataset_snapshot_check(ds
, resa
->tosnap
, tx
));
1507 recv_end_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
1509 dsl_dataset_t
*ds
= arg1
;
1510 struct recvendsyncarg
*resa
= arg2
;
1512 dsl_dataset_snapshot_sync(ds
, resa
->tosnap
, tx
);
1514 /* set snapshot's creation time and guid */
1515 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1516 ds
->ds_prev
->ds_phys
->ds_creation_time
= resa
->creation_time
;
1517 ds
->ds_prev
->ds_phys
->ds_guid
= resa
->toguid
;
1518 ds
->ds_prev
->ds_phys
->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
1520 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1521 ds
->ds_phys
->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
1525 dmu_recv_existing_end(dmu_recv_cookie_t
*drc
)
1527 struct recvendsyncarg resa
;
1528 dsl_dataset_t
*ds
= drc
->drc_logical_ds
;
1532 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1533 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1536 txg_wait_synced(ds
->ds_dir
->dd_pool
, 0);
1538 if (dsl_dataset_tryown(ds
, FALSE
, dmu_recv_tag
)) {
1539 err
= dsl_dataset_clone_swap(drc
->drc_real_ds
, ds
,
1544 mutex_exit(&ds
->ds_recvlock
);
1545 dsl_dataset_rele(ds
, dmu_recv_tag
);
1546 (void) dsl_dataset_destroy(drc
->drc_real_ds
, dmu_recv_tag
,
1551 resa
.creation_time
= drc
->drc_drrb
->drr_creation_time
;
1552 resa
.toguid
= drc
->drc_drrb
->drr_toguid
;
1553 resa
.tosnap
= drc
->drc_tosnap
;
1555 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
1556 recv_end_check
, recv_end_sync
, ds
, &resa
, 3);
1559 (void) dsl_dataset_clone_swap(drc
->drc_real_ds
, ds
, B_TRUE
);
1563 mutex_exit(&ds
->ds_recvlock
);
1564 dsl_dataset_disown(ds
, dmu_recv_tag
);
1565 (void) dsl_dataset_destroy(drc
->drc_real_ds
, dmu_recv_tag
, B_FALSE
);
1570 dmu_recv_new_end(dmu_recv_cookie_t
*drc
)
1572 struct recvendsyncarg resa
;
1573 dsl_dataset_t
*ds
= drc
->drc_logical_ds
;
1577 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1578 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1581 txg_wait_synced(ds
->ds_dir
->dd_pool
, 0);
1583 resa
.creation_time
= drc
->drc_drrb
->drr_creation_time
;
1584 resa
.toguid
= drc
->drc_drrb
->drr_toguid
;
1585 resa
.tosnap
= drc
->drc_tosnap
;
1587 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
1588 recv_end_check
, recv_end_sync
, ds
, &resa
, 3);
1590 /* clean up the fs we just recv'd into */
1591 (void) dsl_dataset_destroy(ds
, dmu_recv_tag
, B_FALSE
);
1593 /* release the hold from dmu_recv_begin */
1594 dsl_dataset_disown(ds
, dmu_recv_tag
);
1600 dmu_recv_end(dmu_recv_cookie_t
*drc
)
1602 if (drc
->drc_logical_ds
!= drc
->drc_real_ds
)
1603 return (dmu_recv_existing_end(drc
));
1605 return (dmu_recv_new_end(drc
));