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.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011 by Delphix. All rights reserved.
26 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
30 #include <sys/dmu_impl.h>
31 #include <sys/dmu_tx.h>
33 #include <sys/dnode.h>
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_dir.h>
39 #include <sys/dsl_prop.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_synctask.h>
42 #include <sys/spa_impl.h>
43 #include <sys/zfs_ioctl.h>
45 #include <sys/zio_checksum.h>
46 #include <sys/zfs_znode.h>
47 #include <zfs_fletcher.h>
50 #include <sys/zfs_onexit.h>
52 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
53 int zfs_send_corrupt_data
= B_FALSE
;
55 static char *dmu_recv_tag
= "dmu_recv_tag";
57 typedef struct dump_bytes_io
{
58 dmu_sendarg_t
*dbi_dsp
;
64 dump_bytes_strategy(void *arg
)
66 dump_bytes_io_t
*dbi
= (dump_bytes_io_t
*)arg
;
67 dmu_sendarg_t
*dsp
= dbi
->dbi_dsp
;
68 dsl_dataset_t
*ds
= dsp
->dsa_os
->os_dsl_dataset
;
69 ssize_t resid
; /* have to get resid to get detailed errno */
70 ASSERT0(dbi
->dbi_len
% 8);
72 fletcher_4_incremental_native(dbi
->dbi_buf
, dbi
->dbi_len
, &dsp
->dsa_zc
);
73 dsp
->dsa_err
= vn_rdwr(UIO_WRITE
, dsp
->dsa_vp
,
74 (caddr_t
)dbi
->dbi_buf
, dbi
->dbi_len
,
75 0, UIO_SYSSPACE
, FAPPEND
, RLIM64_INFINITY
, CRED(), &resid
);
77 mutex_enter(&ds
->ds_sendstream_lock
);
78 *dsp
->dsa_off
+= dbi
->dbi_len
;
79 mutex_exit(&ds
->ds_sendstream_lock
);
83 dump_bytes(dmu_sendarg_t
*dsp
, void *buf
, int len
)
92 * The vn_rdwr() call is performed in a taskq to ensure that there is
93 * always enough stack space to write safely to the target filesystem.
94 * The ZIO_TYPE_FREE threads are used because there can be a lot of
95 * them and they are used in vdev_file.c for a similar purpose.
97 spa_taskq_dispatch_sync(dmu_objset_spa(dsp
->dsa_os
), ZIO_TYPE_FREE
,
98 ZIO_TASKQ_ISSUE
, dump_bytes_strategy
, &dbi
, TQ_SLEEP
);
100 return (dsp
->dsa_err
);
104 dump_free(dmu_sendarg_t
*dsp
, uint64_t object
, uint64_t offset
,
107 struct drr_free
*drrf
= &(dsp
->dsa_drr
->drr_u
.drr_free
);
109 if (length
!= -1ULL && offset
+ length
< offset
)
113 * If there is a pending op, but it's not PENDING_FREE, push it out,
114 * since free block aggregation can only be done for blocks of the
115 * same type (i.e., DRR_FREE records can only be aggregated with
116 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
117 * aggregated with other DRR_FREEOBJECTS records.
119 if (dsp
->dsa_pending_op
!= PENDING_NONE
&&
120 dsp
->dsa_pending_op
!= PENDING_FREE
) {
121 if (dump_bytes(dsp
, dsp
->dsa_drr
,
122 sizeof (dmu_replay_record_t
)) != 0)
124 dsp
->dsa_pending_op
= PENDING_NONE
;
127 if (dsp
->dsa_pending_op
== PENDING_FREE
) {
129 * There should never be a PENDING_FREE if length is -1
130 * (because dump_dnode is the only place where this
131 * function is called with a -1, and only after flushing
132 * any pending record).
134 ASSERT(length
!= -1ULL);
136 * Check to see whether this free block can be aggregated
139 if (drrf
->drr_object
== object
&& drrf
->drr_offset
+
140 drrf
->drr_length
== offset
) {
141 drrf
->drr_length
+= length
;
144 /* not a continuation. Push out pending record */
145 if (dump_bytes(dsp
, dsp
->dsa_drr
,
146 sizeof (dmu_replay_record_t
)) != 0)
148 dsp
->dsa_pending_op
= PENDING_NONE
;
151 /* create a FREE record and make it pending */
152 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
153 dsp
->dsa_drr
->drr_type
= DRR_FREE
;
154 drrf
->drr_object
= object
;
155 drrf
->drr_offset
= offset
;
156 drrf
->drr_length
= length
;
157 drrf
->drr_toguid
= dsp
->dsa_toguid
;
158 if (length
== -1ULL) {
159 if (dump_bytes(dsp
, dsp
->dsa_drr
,
160 sizeof (dmu_replay_record_t
)) != 0)
163 dsp
->dsa_pending_op
= PENDING_FREE
;
170 dump_data(dmu_sendarg_t
*dsp
, dmu_object_type_t type
,
171 uint64_t object
, uint64_t offset
, int blksz
, const blkptr_t
*bp
, void *data
)
173 struct drr_write
*drrw
= &(dsp
->dsa_drr
->drr_u
.drr_write
);
177 * If there is any kind of pending aggregation (currently either
178 * a grouping of free objects or free blocks), push it out to
179 * the stream, since aggregation can't be done across operations
180 * of different types.
182 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
183 if (dump_bytes(dsp
, dsp
->dsa_drr
,
184 sizeof (dmu_replay_record_t
)) != 0)
186 dsp
->dsa_pending_op
= PENDING_NONE
;
188 /* write a DATA record */
189 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
190 dsp
->dsa_drr
->drr_type
= DRR_WRITE
;
191 drrw
->drr_object
= object
;
192 drrw
->drr_type
= type
;
193 drrw
->drr_offset
= offset
;
194 drrw
->drr_length
= blksz
;
195 drrw
->drr_toguid
= dsp
->dsa_toguid
;
196 drrw
->drr_checksumtype
= BP_GET_CHECKSUM(bp
);
197 if (zio_checksum_table
[drrw
->drr_checksumtype
].ci_dedup
)
198 drrw
->drr_checksumflags
|= DRR_CHECKSUM_DEDUP
;
199 DDK_SET_LSIZE(&drrw
->drr_key
, BP_GET_LSIZE(bp
));
200 DDK_SET_PSIZE(&drrw
->drr_key
, BP_GET_PSIZE(bp
));
201 DDK_SET_COMPRESS(&drrw
->drr_key
, BP_GET_COMPRESS(bp
));
202 drrw
->drr_key
.ddk_cksum
= bp
->blk_cksum
;
204 if (dump_bytes(dsp
, dsp
->dsa_drr
, sizeof (dmu_replay_record_t
)) != 0)
206 if (dump_bytes(dsp
, data
, blksz
) != 0)
212 dump_spill(dmu_sendarg_t
*dsp
, uint64_t object
, int blksz
, void *data
)
214 struct drr_spill
*drrs
= &(dsp
->dsa_drr
->drr_u
.drr_spill
);
216 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
217 if (dump_bytes(dsp
, dsp
->dsa_drr
,
218 sizeof (dmu_replay_record_t
)) != 0)
220 dsp
->dsa_pending_op
= PENDING_NONE
;
223 /* write a SPILL record */
224 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
225 dsp
->dsa_drr
->drr_type
= DRR_SPILL
;
226 drrs
->drr_object
= object
;
227 drrs
->drr_length
= blksz
;
228 drrs
->drr_toguid
= dsp
->dsa_toguid
;
230 if (dump_bytes(dsp
, dsp
->dsa_drr
, sizeof (dmu_replay_record_t
)))
232 if (dump_bytes(dsp
, data
, blksz
))
238 dump_freeobjects(dmu_sendarg_t
*dsp
, uint64_t firstobj
, uint64_t numobjs
)
240 struct drr_freeobjects
*drrfo
= &(dsp
->dsa_drr
->drr_u
.drr_freeobjects
);
243 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
244 * push it out, since free block aggregation can only be done for
245 * blocks of the same type (i.e., DRR_FREE records can only be
246 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
247 * can only be aggregated with other DRR_FREEOBJECTS records.
249 if (dsp
->dsa_pending_op
!= PENDING_NONE
&&
250 dsp
->dsa_pending_op
!= PENDING_FREEOBJECTS
) {
251 if (dump_bytes(dsp
, dsp
->dsa_drr
,
252 sizeof (dmu_replay_record_t
)) != 0)
254 dsp
->dsa_pending_op
= PENDING_NONE
;
256 if (dsp
->dsa_pending_op
== PENDING_FREEOBJECTS
) {
258 * See whether this free object array can be aggregated
261 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
== firstobj
) {
262 drrfo
->drr_numobjs
+= numobjs
;
265 /* can't be aggregated. Push out pending record */
266 if (dump_bytes(dsp
, dsp
->dsa_drr
,
267 sizeof (dmu_replay_record_t
)) != 0)
269 dsp
->dsa_pending_op
= PENDING_NONE
;
273 /* write a FREEOBJECTS record */
274 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
275 dsp
->dsa_drr
->drr_type
= DRR_FREEOBJECTS
;
276 drrfo
->drr_firstobj
= firstobj
;
277 drrfo
->drr_numobjs
= numobjs
;
278 drrfo
->drr_toguid
= dsp
->dsa_toguid
;
280 dsp
->dsa_pending_op
= PENDING_FREEOBJECTS
;
286 dump_dnode(dmu_sendarg_t
*dsp
, uint64_t object
, dnode_phys_t
*dnp
)
288 struct drr_object
*drro
= &(dsp
->dsa_drr
->drr_u
.drr_object
);
290 if (dnp
== NULL
|| dnp
->dn_type
== DMU_OT_NONE
)
291 return (dump_freeobjects(dsp
, object
, 1));
293 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
294 if (dump_bytes(dsp
, dsp
->dsa_drr
,
295 sizeof (dmu_replay_record_t
)) != 0)
297 dsp
->dsa_pending_op
= PENDING_NONE
;
300 /* write an OBJECT record */
301 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
302 dsp
->dsa_drr
->drr_type
= DRR_OBJECT
;
303 drro
->drr_object
= object
;
304 drro
->drr_type
= dnp
->dn_type
;
305 drro
->drr_bonustype
= dnp
->dn_bonustype
;
306 drro
->drr_blksz
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
307 drro
->drr_bonuslen
= dnp
->dn_bonuslen
;
308 drro
->drr_checksumtype
= dnp
->dn_checksum
;
309 drro
->drr_compress
= dnp
->dn_compress
;
310 drro
->drr_toguid
= dsp
->dsa_toguid
;
312 if (dump_bytes(dsp
, dsp
->dsa_drr
, sizeof (dmu_replay_record_t
)) != 0)
315 if (dump_bytes(dsp
, DN_BONUS(dnp
), P2ROUNDUP(dnp
->dn_bonuslen
, 8)) != 0)
318 /* free anything past the end of the file */
319 if (dump_free(dsp
, object
, (dnp
->dn_maxblkid
+ 1) *
320 (dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
), -1ULL))
327 #define BP_SPAN(dnp, level) \
328 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
329 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
333 backup_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
334 const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
336 dmu_sendarg_t
*dsp
= arg
;
337 dmu_object_type_t type
= bp
? BP_GET_TYPE(bp
) : DMU_OT_NONE
;
340 if (issig(JUSTLOOKING
) && issig(FORREAL
))
343 if (zb
->zb_object
!= DMU_META_DNODE_OBJECT
&&
344 DMU_OBJECT_IS_SPECIAL(zb
->zb_object
)) {
346 } else if (bp
== NULL
&& zb
->zb_object
== DMU_META_DNODE_OBJECT
) {
347 uint64_t span
= BP_SPAN(dnp
, zb
->zb_level
);
348 uint64_t dnobj
= (zb
->zb_blkid
* span
) >> DNODE_SHIFT
;
349 err
= dump_freeobjects(dsp
, dnobj
, span
>> DNODE_SHIFT
);
350 } else if (bp
== NULL
) {
351 uint64_t span
= BP_SPAN(dnp
, zb
->zb_level
);
352 err
= dump_free(dsp
, zb
->zb_object
, zb
->zb_blkid
* span
, span
);
353 } else if (zb
->zb_level
> 0 || type
== DMU_OT_OBJSET
) {
355 } else if (type
== DMU_OT_DNODE
) {
358 int blksz
= BP_GET_LSIZE(bp
);
359 uint32_t aflags
= ARC_WAIT
;
362 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
363 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
,
368 for (i
= 0; i
< blksz
>> DNODE_SHIFT
; i
++) {
369 uint64_t dnobj
= (zb
->zb_blkid
<<
370 (DNODE_BLOCK_SHIFT
- DNODE_SHIFT
)) + i
;
371 err
= dump_dnode(dsp
, dnobj
, blk
+i
);
375 (void) arc_buf_remove_ref(abuf
, &abuf
);
376 } else if (type
== DMU_OT_SA
) {
377 uint32_t aflags
= ARC_WAIT
;
379 int blksz
= BP_GET_LSIZE(bp
);
381 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
382 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
,
386 err
= dump_spill(dsp
, zb
->zb_object
, blksz
, abuf
->b_data
);
387 (void) arc_buf_remove_ref(abuf
, &abuf
);
388 } else { /* it's a level-0 block of a regular object */
389 uint32_t aflags
= ARC_WAIT
;
391 int blksz
= BP_GET_LSIZE(bp
);
393 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
394 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
,
396 if (zfs_send_corrupt_data
) {
398 /* Send a block filled with 0x"zfs badd bloc" */
399 abuf
= arc_buf_alloc(spa
, blksz
, &abuf
,
401 for (ptr
= abuf
->b_data
;
402 (char *)ptr
< (char *)abuf
->b_data
+ blksz
;
404 *ptr
= 0x2f5baddb10cULL
;
410 err
= dump_data(dsp
, type
, zb
->zb_object
, zb
->zb_blkid
* blksz
,
411 blksz
, bp
, abuf
->b_data
);
412 (void) arc_buf_remove_ref(abuf
, &abuf
);
415 ASSERT(err
== 0 || err
== EINTR
);
420 dmu_send(objset_t
*tosnap
, objset_t
*fromsnap
, boolean_t fromorigin
,
421 int outfd
, vnode_t
*vp
, offset_t
*off
)
423 dsl_dataset_t
*ds
= tosnap
->os_dsl_dataset
;
424 dsl_dataset_t
*fromds
= fromsnap
? fromsnap
->os_dsl_dataset
: NULL
;
425 dmu_replay_record_t
*drr
;
428 uint64_t fromtxg
= 0;
430 /* tosnap must be a snapshot */
431 if (ds
->ds_phys
->ds_next_snap_obj
== 0)
434 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
435 if (fromds
&& (ds
->ds_dir
!= fromds
->ds_dir
||
436 fromds
->ds_phys
->ds_creation_txg
>= ds
->ds_phys
->ds_creation_txg
))
440 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
445 if (dsl_dir_is_clone(ds
->ds_dir
)) {
446 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
447 err
= dsl_dataset_hold_obj(dp
,
448 ds
->ds_dir
->dd_phys
->dd_origin_obj
, FTAG
, &fromds
);
449 rw_exit(&dp
->dp_config_rwlock
);
453 fromorigin
= B_FALSE
;
458 drr
= kmem_zalloc(sizeof (dmu_replay_record_t
), KM_SLEEP
);
459 drr
->drr_type
= DRR_BEGIN
;
460 drr
->drr_u
.drr_begin
.drr_magic
= DMU_BACKUP_MAGIC
;
461 DMU_SET_STREAM_HDRTYPE(drr
->drr_u
.drr_begin
.drr_versioninfo
,
465 if (dmu_objset_type(tosnap
) == DMU_OST_ZFS
) {
467 if (zfs_get_zplprop(tosnap
, ZFS_PROP_VERSION
, &version
) != 0) {
468 kmem_free(drr
, sizeof (dmu_replay_record_t
));
471 if (version
== ZPL_VERSION_SA
) {
472 DMU_SET_FEATUREFLAGS(
473 drr
->drr_u
.drr_begin
.drr_versioninfo
,
474 DMU_BACKUP_FEATURE_SA_SPILL
);
479 drr
->drr_u
.drr_begin
.drr_creation_time
=
480 ds
->ds_phys
->ds_creation_time
;
481 drr
->drr_u
.drr_begin
.drr_type
= tosnap
->os_phys
->os_type
;
483 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CLONE
;
484 drr
->drr_u
.drr_begin
.drr_toguid
= ds
->ds_phys
->ds_guid
;
485 if (ds
->ds_phys
->ds_flags
& DS_FLAG_CI_DATASET
)
486 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CI_DATA
;
489 drr
->drr_u
.drr_begin
.drr_fromguid
= fromds
->ds_phys
->ds_guid
;
490 dsl_dataset_name(ds
, drr
->drr_u
.drr_begin
.drr_toname
);
493 fromtxg
= fromds
->ds_phys
->ds_creation_txg
;
495 dsl_dataset_rele(fromds
, FTAG
);
497 dsp
= kmem_zalloc(sizeof (dmu_sendarg_t
), KM_SLEEP
);
501 dsp
->dsa_outfd
= outfd
;
502 dsp
->dsa_proc
= curproc
;
503 dsp
->dsa_os
= tosnap
;
505 dsp
->dsa_toguid
= ds
->ds_phys
->ds_guid
;
506 ZIO_SET_CHECKSUM(&dsp
->dsa_zc
, 0, 0, 0, 0);
507 dsp
->dsa_pending_op
= PENDING_NONE
;
509 mutex_enter(&ds
->ds_sendstream_lock
);
510 list_insert_head(&ds
->ds_sendstreams
, dsp
);
511 mutex_exit(&ds
->ds_sendstream_lock
);
513 if (dump_bytes(dsp
, drr
, sizeof (dmu_replay_record_t
)) != 0) {
518 err
= traverse_dataset(ds
, fromtxg
, TRAVERSE_PRE
| TRAVERSE_PREFETCH
,
521 if (dsp
->dsa_pending_op
!= PENDING_NONE
)
522 if (dump_bytes(dsp
, drr
, sizeof (dmu_replay_record_t
)) != 0)
526 if (err
== EINTR
&& dsp
->dsa_err
)
531 bzero(drr
, sizeof (dmu_replay_record_t
));
532 drr
->drr_type
= DRR_END
;
533 drr
->drr_u
.drr_end
.drr_checksum
= dsp
->dsa_zc
;
534 drr
->drr_u
.drr_end
.drr_toguid
= dsp
->dsa_toguid
;
536 if (dump_bytes(dsp
, drr
, sizeof (dmu_replay_record_t
)) != 0) {
542 mutex_enter(&ds
->ds_sendstream_lock
);
543 list_remove(&ds
->ds_sendstreams
, dsp
);
544 mutex_exit(&ds
->ds_sendstream_lock
);
546 kmem_free(drr
, sizeof (dmu_replay_record_t
));
547 kmem_free(dsp
, sizeof (dmu_sendarg_t
));
553 dmu_send_estimate(objset_t
*tosnap
, objset_t
*fromsnap
, boolean_t fromorigin
,
556 dsl_dataset_t
*ds
= tosnap
->os_dsl_dataset
;
557 dsl_dataset_t
*fromds
= fromsnap
? fromsnap
->os_dsl_dataset
: NULL
;
558 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
560 uint64_t size
, recordsize
;
562 /* tosnap must be a snapshot */
563 if (ds
->ds_phys
->ds_next_snap_obj
== 0)
566 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
567 if (fromds
&& (ds
->ds_dir
!= fromds
->ds_dir
||
568 fromds
->ds_phys
->ds_creation_txg
>= ds
->ds_phys
->ds_creation_txg
))
575 if (dsl_dir_is_clone(ds
->ds_dir
)) {
576 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
577 err
= dsl_dataset_hold_obj(dp
,
578 ds
->ds_dir
->dd_phys
->dd_origin_obj
, FTAG
, &fromds
);
579 rw_exit(&dp
->dp_config_rwlock
);
583 fromorigin
= B_FALSE
;
587 /* Get uncompressed size estimate of changed data. */
588 if (fromds
== NULL
) {
589 size
= ds
->ds_phys
->ds_uncompressed_bytes
;
592 err
= dsl_dataset_space_written(fromds
, ds
,
593 &used
, &comp
, &size
);
595 dsl_dataset_rele(fromds
, FTAG
);
601 * Assume that space (both on-disk and in-stream) is dominated by
602 * data. We will adjust for indirect blocks and the copies property,
603 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
607 * Subtract out approximate space used by indirect blocks.
608 * Assume most space is used by data blocks (non-indirect, non-dnode).
609 * Assume all blocks are recordsize. Assume ditto blocks and
610 * internal fragmentation counter out compression.
612 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
613 * block, which we observe in practice.
615 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
616 err
= dsl_prop_get_ds(ds
, "recordsize",
617 sizeof (recordsize
), 1, &recordsize
, NULL
);
618 rw_exit(&dp
->dp_config_rwlock
);
621 size
-= size
/ recordsize
* sizeof (blkptr_t
);
623 /* Add in the space for the record associated with each block. */
624 size
+= size
/ recordsize
* sizeof (dmu_replay_record_t
);
631 struct recvbeginsyncarg
{
634 dsl_dataset_t
*origin
;
636 dmu_objset_type_t type
;
640 char clonelastname
[MAXNAMELEN
];
641 dsl_dataset_t
*ds
; /* the ds to recv into; returned from the syncfunc */
647 recv_new_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
649 dsl_dir_t
*dd
= arg1
;
650 struct recvbeginsyncarg
*rbsa
= arg2
;
651 objset_t
*mos
= dd
->dd_pool
->dp_meta_objset
;
655 err
= zap_lookup(mos
, dd
->dd_phys
->dd_child_dir_zapobj
,
656 strrchr(rbsa
->tofs
, '/') + 1, sizeof (uint64_t), 1, &val
);
659 return (err
? err
: EEXIST
);
662 /* make sure it's a snap in the same pool */
663 if (rbsa
->origin
->ds_dir
->dd_pool
!= dd
->dd_pool
)
665 if (!dsl_dataset_is_snapshot(rbsa
->origin
))
667 if (rbsa
->origin
->ds_phys
->ds_guid
!= rbsa
->fromguid
)
675 recv_new_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
677 dsl_dir_t
*dd
= arg1
;
678 struct recvbeginsyncarg
*rbsa
= arg2
;
679 uint64_t flags
= DS_FLAG_INCONSISTENT
| rbsa
->dsflags
;
682 /* Create and open new dataset. */
683 dsobj
= dsl_dataset_create_sync(dd
, strrchr(rbsa
->tofs
, '/') + 1,
684 rbsa
->origin
, flags
, rbsa
->cr
, tx
);
685 VERIFY(0 == dsl_dataset_own_obj(dd
->dd_pool
, dsobj
,
686 B_TRUE
, dmu_recv_tag
, &rbsa
->ds
));
688 if (rbsa
->origin
== NULL
) {
689 (void) dmu_objset_create_impl(dd
->dd_pool
->dp_spa
,
690 rbsa
->ds
, &rbsa
->ds
->ds_phys
->ds_bp
, rbsa
->type
, tx
);
693 spa_history_log_internal(LOG_DS_REPLAY_FULL_SYNC
,
694 dd
->dd_pool
->dp_spa
, tx
, "dataset = %lld", dsobj
);
699 recv_existing_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
701 dsl_dataset_t
*ds
= arg1
;
702 struct recvbeginsyncarg
*rbsa
= arg2
;
706 /* must not have any changes since most recent snapshot */
707 if (!rbsa
->force
&& dsl_dataset_modified_since_lastsnap(ds
))
710 /* new snapshot name must not exist */
711 err
= zap_lookup(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
712 ds
->ds_phys
->ds_snapnames_zapobj
, rbsa
->tosnap
, 8, 1, &val
);
718 if (rbsa
->fromguid
) {
719 /* if incremental, most recent snapshot must match fromguid */
720 if (ds
->ds_prev
== NULL
)
724 * most recent snapshot must match fromguid, or there are no
725 * changes since the fromguid one
727 if (ds
->ds_prev
->ds_phys
->ds_guid
!= rbsa
->fromguid
) {
728 uint64_t birth
= ds
->ds_prev
->ds_phys
->ds_bp
.blk_birth
;
729 uint64_t obj
= ds
->ds_prev
->ds_phys
->ds_prev_snap_obj
;
732 err
= dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
736 if (snap
->ds_phys
->ds_creation_txg
< birth
) {
737 dsl_dataset_rele(snap
, FTAG
);
740 if (snap
->ds_phys
->ds_guid
== rbsa
->fromguid
) {
741 dsl_dataset_rele(snap
, FTAG
);
744 obj
= snap
->ds_phys
->ds_prev_snap_obj
;
745 dsl_dataset_rele(snap
, FTAG
);
751 /* if full, most recent snapshot must be $ORIGIN */
752 if (ds
->ds_phys
->ds_prev_snap_txg
>= TXG_INITIAL
)
756 /* temporary clone name must not exist */
757 err
= zap_lookup(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
758 ds
->ds_dir
->dd_phys
->dd_child_dir_zapobj
,
759 rbsa
->clonelastname
, 8, 1, &val
);
770 recv_existing_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
772 dsl_dataset_t
*ohds
= arg1
;
773 struct recvbeginsyncarg
*rbsa
= arg2
;
774 dsl_pool_t
*dp
= ohds
->ds_dir
->dd_pool
;
776 uint64_t flags
= DS_FLAG_INCONSISTENT
| rbsa
->dsflags
;
779 /* create and open the temporary clone */
780 dsobj
= dsl_dataset_create_sync(ohds
->ds_dir
, rbsa
->clonelastname
,
781 ohds
->ds_prev
, flags
, rbsa
->cr
, tx
);
782 VERIFY(0 == dsl_dataset_own_obj(dp
, dsobj
, B_TRUE
, dmu_recv_tag
, &cds
));
785 * If we actually created a non-clone, we need to create the
786 * objset in our new dataset.
788 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds
))) {
789 (void) dmu_objset_create_impl(dp
->dp_spa
,
790 cds
, dsl_dataset_get_blkptr(cds
), rbsa
->type
, tx
);
795 spa_history_log_internal(LOG_DS_REPLAY_INC_SYNC
,
796 dp
->dp_spa
, tx
, "dataset = %lld", dsobj
);
800 dmu_recv_verify_features(dsl_dataset_t
*ds
, struct drr_begin
*drrb
)
804 featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
806 /* Verify pool version supports SA if SA_SPILL feature set */
807 return ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
808 (spa_version(dsl_dataset_get_spa(ds
)) < SPA_VERSION_SA
));
812 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
813 * succeeds; otherwise we will leak the holds on the datasets.
816 dmu_recv_begin(char *tofs
, char *tosnap
, char *top_ds
, struct drr_begin
*drrb
,
817 boolean_t force
, objset_t
*origin
, dmu_recv_cookie_t
*drc
)
821 struct recvbeginsyncarg rbsa
= { 0 };
822 uint64_t versioninfo
;
826 if (drrb
->drr_magic
== DMU_BACKUP_MAGIC
)
828 else if (drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
))
834 rbsa
.tosnap
= tosnap
;
835 rbsa
.origin
= origin
? origin
->os_dsl_dataset
: NULL
;
836 rbsa
.fromguid
= drrb
->drr_fromguid
;
837 rbsa
.type
= drrb
->drr_type
;
841 versioninfo
= drrb
->drr_versioninfo
;
842 flags
= drrb
->drr_flags
;
845 rbsa
.type
= BSWAP_32(rbsa
.type
);
846 rbsa
.fromguid
= BSWAP_64(rbsa
.fromguid
);
847 versioninfo
= BSWAP_64(versioninfo
);
848 flags
= BSWAP_32(flags
);
851 if (DMU_GET_STREAM_HDRTYPE(versioninfo
) == DMU_COMPOUNDSTREAM
||
852 rbsa
.type
>= DMU_OST_NUMTYPES
||
853 ((flags
& DRR_FLAG_CLONE
) && origin
== NULL
))
856 if (flags
& DRR_FLAG_CI_DATA
)
857 rbsa
.dsflags
= DS_FLAG_CI_DATASET
;
859 bzero(drc
, sizeof (dmu_recv_cookie_t
));
860 drc
->drc_drrb
= drrb
;
861 drc
->drc_tosnap
= tosnap
;
862 drc
->drc_top_ds
= top_ds
;
863 drc
->drc_force
= force
;
866 * Process the begin in syncing context.
869 /* open the dataset we are logically receiving into */
870 err
= dsl_dataset_hold(tofs
, dmu_recv_tag
, &ds
);
872 if (dmu_recv_verify_features(ds
, drrb
)) {
873 dsl_dataset_rele(ds
, dmu_recv_tag
);
876 /* target fs already exists; recv into temp clone */
878 /* Can't recv a clone into an existing fs */
879 if (flags
& DRR_FLAG_CLONE
) {
880 dsl_dataset_rele(ds
, dmu_recv_tag
);
884 /* must not have an incremental recv already in progress */
885 if (!mutex_tryenter(&ds
->ds_recvlock
)) {
886 dsl_dataset_rele(ds
, dmu_recv_tag
);
890 /* tmp clone name is: tofs/%tosnap" */
891 (void) snprintf(rbsa
.clonelastname
, sizeof (rbsa
.clonelastname
),
894 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
895 recv_existing_check
, recv_existing_sync
, ds
, &rbsa
, 5);
897 mutex_exit(&ds
->ds_recvlock
);
898 dsl_dataset_rele(ds
, dmu_recv_tag
);
901 drc
->drc_logical_ds
= ds
;
902 drc
->drc_real_ds
= rbsa
.ds
;
903 } else if (err
== ENOENT
) {
904 /* target fs does not exist; must be a full backup or clone */
908 * If it's a non-clone incremental, we are missing the
909 * target fs, so fail the recv.
911 if (rbsa
.fromguid
&& !(flags
& DRR_FLAG_CLONE
))
914 /* Open the parent of tofs */
915 cp
= strrchr(tofs
, '/');
917 err
= dsl_dataset_hold(tofs
, FTAG
, &ds
);
922 if (dmu_recv_verify_features(ds
, drrb
)) {
923 dsl_dataset_rele(ds
, FTAG
);
927 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
928 recv_new_check
, recv_new_sync
, ds
->ds_dir
, &rbsa
, 5);
929 dsl_dataset_rele(ds
, FTAG
);
932 drc
->drc_logical_ds
= drc
->drc_real_ds
= rbsa
.ds
;
933 drc
->drc_newfs
= B_TRUE
;
945 int bufsize
; /* amount of memory allocated for buf */
947 avl_tree_t
*guid_to_ds_map
;
950 typedef struct guid_map_entry
{
952 dsl_dataset_t
*gme_ds
;
957 guid_compare(const void *arg1
, const void *arg2
)
959 const guid_map_entry_t
*gmep1
= arg1
;
960 const guid_map_entry_t
*gmep2
= arg2
;
962 if (gmep1
->guid
< gmep2
->guid
)
964 else if (gmep1
->guid
> gmep2
->guid
)
970 free_guid_map_onexit(void *arg
)
972 avl_tree_t
*ca
= arg
;
974 guid_map_entry_t
*gmep
;
976 while ((gmep
= avl_destroy_nodes(ca
, &cookie
)) != NULL
) {
977 dsl_dataset_rele(gmep
->gme_ds
, ca
);
978 kmem_free(gmep
, sizeof (guid_map_entry_t
));
981 kmem_free(ca
, sizeof (avl_tree_t
));
985 restore_read(struct restorearg
*ra
, int len
)
990 /* some things will require 8-byte alignment, so everything must */
996 ra
->err
= vn_rdwr(UIO_READ
, ra
->vp
,
997 (caddr_t
)ra
->buf
+ done
, len
- done
,
998 ra
->voff
, UIO_SYSSPACE
, FAPPEND
,
999 RLIM64_INFINITY
, CRED(), &resid
);
1001 if (resid
== len
- done
)
1003 ra
->voff
+= len
- done
- resid
;
1009 ASSERT3U(done
, ==, len
);
1012 fletcher_4_incremental_byteswap(rv
, len
, &ra
->cksum
);
1014 fletcher_4_incremental_native(rv
, len
, &ra
->cksum
);
1018 noinline
static void
1019 backup_byteswap(dmu_replay_record_t
*drr
)
1021 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1022 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1023 drr
->drr_type
= BSWAP_32(drr
->drr_type
);
1024 drr
->drr_payloadlen
= BSWAP_32(drr
->drr_payloadlen
);
1025 switch (drr
->drr_type
) {
1027 DO64(drr_begin
.drr_magic
);
1028 DO64(drr_begin
.drr_versioninfo
);
1029 DO64(drr_begin
.drr_creation_time
);
1030 DO32(drr_begin
.drr_type
);
1031 DO32(drr_begin
.drr_flags
);
1032 DO64(drr_begin
.drr_toguid
);
1033 DO64(drr_begin
.drr_fromguid
);
1036 DO64(drr_object
.drr_object
);
1037 /* DO64(drr_object.drr_allocation_txg); */
1038 DO32(drr_object
.drr_type
);
1039 DO32(drr_object
.drr_bonustype
);
1040 DO32(drr_object
.drr_blksz
);
1041 DO32(drr_object
.drr_bonuslen
);
1042 DO64(drr_object
.drr_toguid
);
1044 case DRR_FREEOBJECTS
:
1045 DO64(drr_freeobjects
.drr_firstobj
);
1046 DO64(drr_freeobjects
.drr_numobjs
);
1047 DO64(drr_freeobjects
.drr_toguid
);
1050 DO64(drr_write
.drr_object
);
1051 DO32(drr_write
.drr_type
);
1052 DO64(drr_write
.drr_offset
);
1053 DO64(drr_write
.drr_length
);
1054 DO64(drr_write
.drr_toguid
);
1055 DO64(drr_write
.drr_key
.ddk_cksum
.zc_word
[0]);
1056 DO64(drr_write
.drr_key
.ddk_cksum
.zc_word
[1]);
1057 DO64(drr_write
.drr_key
.ddk_cksum
.zc_word
[2]);
1058 DO64(drr_write
.drr_key
.ddk_cksum
.zc_word
[3]);
1059 DO64(drr_write
.drr_key
.ddk_prop
);
1061 case DRR_WRITE_BYREF
:
1062 DO64(drr_write_byref
.drr_object
);
1063 DO64(drr_write_byref
.drr_offset
);
1064 DO64(drr_write_byref
.drr_length
);
1065 DO64(drr_write_byref
.drr_toguid
);
1066 DO64(drr_write_byref
.drr_refguid
);
1067 DO64(drr_write_byref
.drr_refobject
);
1068 DO64(drr_write_byref
.drr_refoffset
);
1069 DO64(drr_write_byref
.drr_key
.ddk_cksum
.zc_word
[0]);
1070 DO64(drr_write_byref
.drr_key
.ddk_cksum
.zc_word
[1]);
1071 DO64(drr_write_byref
.drr_key
.ddk_cksum
.zc_word
[2]);
1072 DO64(drr_write_byref
.drr_key
.ddk_cksum
.zc_word
[3]);
1073 DO64(drr_write_byref
.drr_key
.ddk_prop
);
1076 DO64(drr_free
.drr_object
);
1077 DO64(drr_free
.drr_offset
);
1078 DO64(drr_free
.drr_length
);
1079 DO64(drr_free
.drr_toguid
);
1082 DO64(drr_spill
.drr_object
);
1083 DO64(drr_spill
.drr_length
);
1084 DO64(drr_spill
.drr_toguid
);
1087 DO64(drr_end
.drr_checksum
.zc_word
[0]);
1088 DO64(drr_end
.drr_checksum
.zc_word
[1]);
1089 DO64(drr_end
.drr_checksum
.zc_word
[2]);
1090 DO64(drr_end
.drr_checksum
.zc_word
[3]);
1091 DO64(drr_end
.drr_toguid
);
1101 restore_object(struct restorearg
*ra
, objset_t
*os
, struct drr_object
*drro
)
1107 if (drro
->drr_type
== DMU_OT_NONE
||
1108 !DMU_OT_IS_VALID(drro
->drr_type
) ||
1109 !DMU_OT_IS_VALID(drro
->drr_bonustype
) ||
1110 drro
->drr_checksumtype
>= ZIO_CHECKSUM_FUNCTIONS
||
1111 drro
->drr_compress
>= ZIO_COMPRESS_FUNCTIONS
||
1112 P2PHASE(drro
->drr_blksz
, SPA_MINBLOCKSIZE
) ||
1113 drro
->drr_blksz
< SPA_MINBLOCKSIZE
||
1114 drro
->drr_blksz
> SPA_MAXBLOCKSIZE
||
1115 drro
->drr_bonuslen
> DN_MAX_BONUSLEN
) {
1119 err
= dmu_object_info(os
, drro
->drr_object
, NULL
);
1121 if (err
!= 0 && err
!= ENOENT
)
1124 if (drro
->drr_bonuslen
) {
1125 data
= restore_read(ra
, P2ROUNDUP(drro
->drr_bonuslen
, 8));
1130 if (err
== ENOENT
) {
1131 /* currently free, want to be allocated */
1132 tx
= dmu_tx_create(os
);
1133 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1134 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1139 err
= dmu_object_claim(os
, drro
->drr_object
,
1140 drro
->drr_type
, drro
->drr_blksz
,
1141 drro
->drr_bonustype
, drro
->drr_bonuslen
, tx
);
1144 /* currently allocated, want to be allocated */
1145 err
= dmu_object_reclaim(os
, drro
->drr_object
,
1146 drro
->drr_type
, drro
->drr_blksz
,
1147 drro
->drr_bonustype
, drro
->drr_bonuslen
);
1153 tx
= dmu_tx_create(os
);
1154 dmu_tx_hold_bonus(tx
, drro
->drr_object
);
1155 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1161 dmu_object_set_checksum(os
, drro
->drr_object
, drro
->drr_checksumtype
,
1163 dmu_object_set_compress(os
, drro
->drr_object
, drro
->drr_compress
, tx
);
1168 VERIFY(0 == dmu_bonus_hold(os
, drro
->drr_object
, FTAG
, &db
));
1169 dmu_buf_will_dirty(db
, tx
);
1171 ASSERT3U(db
->db_size
, >=, drro
->drr_bonuslen
);
1172 bcopy(data
, db
->db_data
, drro
->drr_bonuslen
);
1174 dmu_object_byteswap_t byteswap
=
1175 DMU_OT_BYTESWAP(drro
->drr_bonustype
);
1176 dmu_ot_byteswap
[byteswap
].ob_func(db
->db_data
,
1177 drro
->drr_bonuslen
);
1179 dmu_buf_rele(db
, FTAG
);
1187 restore_freeobjects(struct restorearg
*ra
, objset_t
*os
,
1188 struct drr_freeobjects
*drrfo
)
1192 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
< drrfo
->drr_firstobj
)
1195 for (obj
= drrfo
->drr_firstobj
;
1196 obj
< drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
;
1197 (void) dmu_object_next(os
, &obj
, FALSE
, 0)) {
1200 if (dmu_object_info(os
, obj
, NULL
) != 0)
1203 err
= dmu_free_object(os
, obj
);
1211 restore_write(struct restorearg
*ra
, objset_t
*os
,
1212 struct drr_write
*drrw
)
1218 if (drrw
->drr_offset
+ drrw
->drr_length
< drrw
->drr_offset
||
1219 !DMU_OT_IS_VALID(drrw
->drr_type
))
1222 data
= restore_read(ra
, drrw
->drr_length
);
1226 if (dmu_object_info(os
, drrw
->drr_object
, NULL
) != 0)
1229 tx
= dmu_tx_create(os
);
1231 dmu_tx_hold_write(tx
, drrw
->drr_object
,
1232 drrw
->drr_offset
, drrw
->drr_length
);
1233 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1239 dmu_object_byteswap_t byteswap
=
1240 DMU_OT_BYTESWAP(drrw
->drr_type
);
1241 dmu_ot_byteswap
[byteswap
].ob_func(data
, drrw
->drr_length
);
1243 dmu_write(os
, drrw
->drr_object
,
1244 drrw
->drr_offset
, drrw
->drr_length
, data
, tx
);
1250 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1251 * streams to refer to a copy of the data that is already on the
1252 * system because it came in earlier in the stream. This function
1253 * finds the earlier copy of the data, and uses that copy instead of
1254 * data from the stream to fulfill this write.
1257 restore_write_byref(struct restorearg
*ra
, objset_t
*os
,
1258 struct drr_write_byref
*drrwbr
)
1262 guid_map_entry_t gmesrch
;
1263 guid_map_entry_t
*gmep
;
1265 objset_t
*ref_os
= NULL
;
1268 if (drrwbr
->drr_offset
+ drrwbr
->drr_length
< drrwbr
->drr_offset
)
1272 * If the GUID of the referenced dataset is different from the
1273 * GUID of the target dataset, find the referenced dataset.
1275 if (drrwbr
->drr_toguid
!= drrwbr
->drr_refguid
) {
1276 gmesrch
.guid
= drrwbr
->drr_refguid
;
1277 if ((gmep
= avl_find(ra
->guid_to_ds_map
, &gmesrch
,
1281 if (dmu_objset_from_ds(gmep
->gme_ds
, &ref_os
))
1287 err
= dmu_buf_hold(ref_os
, drrwbr
->drr_refobject
,
1288 drrwbr
->drr_refoffset
, FTAG
, &dbp
, DMU_READ_PREFETCH
);
1292 tx
= dmu_tx_create(os
);
1294 dmu_tx_hold_write(tx
, drrwbr
->drr_object
,
1295 drrwbr
->drr_offset
, drrwbr
->drr_length
);
1296 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1301 dmu_write(os
, drrwbr
->drr_object
,
1302 drrwbr
->drr_offset
, drrwbr
->drr_length
, dbp
->db_data
, tx
);
1303 dmu_buf_rele(dbp
, FTAG
);
1309 restore_spill(struct restorearg
*ra
, objset_t
*os
, struct drr_spill
*drrs
)
1313 dmu_buf_t
*db
, *db_spill
;
1316 if (drrs
->drr_length
< SPA_MINBLOCKSIZE
||
1317 drrs
->drr_length
> SPA_MAXBLOCKSIZE
)
1320 data
= restore_read(ra
, drrs
->drr_length
);
1324 if (dmu_object_info(os
, drrs
->drr_object
, NULL
) != 0)
1327 VERIFY(0 == dmu_bonus_hold(os
, drrs
->drr_object
, FTAG
, &db
));
1328 if ((err
= dmu_spill_hold_by_bonus(db
, FTAG
, &db_spill
)) != 0) {
1329 dmu_buf_rele(db
, FTAG
);
1333 tx
= dmu_tx_create(os
);
1335 dmu_tx_hold_spill(tx
, db
->db_object
);
1337 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1339 dmu_buf_rele(db
, FTAG
);
1340 dmu_buf_rele(db_spill
, FTAG
);
1344 dmu_buf_will_dirty(db_spill
, tx
);
1346 if (db_spill
->db_size
< drrs
->drr_length
)
1347 VERIFY(0 == dbuf_spill_set_blksz(db_spill
,
1348 drrs
->drr_length
, tx
));
1349 bcopy(data
, db_spill
->db_data
, drrs
->drr_length
);
1351 dmu_buf_rele(db
, FTAG
);
1352 dmu_buf_rele(db_spill
, FTAG
);
1360 restore_free(struct restorearg
*ra
, objset_t
*os
,
1361 struct drr_free
*drrf
)
1365 if (drrf
->drr_length
!= -1ULL &&
1366 drrf
->drr_offset
+ drrf
->drr_length
< drrf
->drr_offset
)
1369 if (dmu_object_info(os
, drrf
->drr_object
, NULL
) != 0)
1372 err
= dmu_free_long_range(os
, drrf
->drr_object
,
1373 drrf
->drr_offset
, drrf
->drr_length
);
1378 * NB: callers *must* call dmu_recv_end() if this succeeds.
1381 dmu_recv_stream(dmu_recv_cookie_t
*drc
, vnode_t
*vp
, offset_t
*voffp
,
1382 int cleanup_fd
, uint64_t *action_handlep
)
1384 struct restorearg ra
= { 0 };
1385 dmu_replay_record_t
*drr
;
1390 if (drc
->drc_drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
))
1394 /* compute checksum of drr_begin record */
1395 dmu_replay_record_t
*drr
;
1396 drr
= kmem_zalloc(sizeof (dmu_replay_record_t
), KM_SLEEP
);
1398 drr
->drr_type
= DRR_BEGIN
;
1399 drr
->drr_u
.drr_begin
= *drc
->drc_drrb
;
1401 fletcher_4_incremental_byteswap(drr
,
1402 sizeof (dmu_replay_record_t
), &ra
.cksum
);
1404 fletcher_4_incremental_native(drr
,
1405 sizeof (dmu_replay_record_t
), &ra
.cksum
);
1407 kmem_free(drr
, sizeof (dmu_replay_record_t
));
1411 struct drr_begin
*drrb
= drc
->drc_drrb
;
1412 drrb
->drr_magic
= BSWAP_64(drrb
->drr_magic
);
1413 drrb
->drr_versioninfo
= BSWAP_64(drrb
->drr_versioninfo
);
1414 drrb
->drr_creation_time
= BSWAP_64(drrb
->drr_creation_time
);
1415 drrb
->drr_type
= BSWAP_32(drrb
->drr_type
);
1416 drrb
->drr_toguid
= BSWAP_64(drrb
->drr_toguid
);
1417 drrb
->drr_fromguid
= BSWAP_64(drrb
->drr_fromguid
);
1423 ra
.buf
= vmem_alloc(ra
.bufsize
, KM_SLEEP
);
1425 /* these were verified in dmu_recv_begin */
1426 ASSERT(DMU_GET_STREAM_HDRTYPE(drc
->drc_drrb
->drr_versioninfo
) ==
1428 ASSERT(drc
->drc_drrb
->drr_type
< DMU_OST_NUMTYPES
);
1431 * Open the objset we are modifying.
1433 VERIFY(dmu_objset_from_ds(drc
->drc_real_ds
, &os
) == 0);
1435 ASSERT(drc
->drc_real_ds
->ds_phys
->ds_flags
& DS_FLAG_INCONSISTENT
);
1437 featureflags
= DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
);
1439 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1440 if (featureflags
& DMU_BACKUP_FEATURE_DEDUP
) {
1443 if (cleanup_fd
== -1) {
1447 ra
.err
= zfs_onexit_fd_hold(cleanup_fd
, &minor
);
1453 if (*action_handlep
== 0) {
1455 kmem_alloc(sizeof (avl_tree_t
), KM_SLEEP
);
1456 avl_create(ra
.guid_to_ds_map
, guid_compare
,
1457 sizeof (guid_map_entry_t
),
1458 offsetof(guid_map_entry_t
, avlnode
));
1459 ra
.err
= zfs_onexit_add_cb(minor
,
1460 free_guid_map_onexit
, ra
.guid_to_ds_map
,
1465 ra
.err
= zfs_onexit_cb_data(minor
, *action_handlep
,
1466 (void **)&ra
.guid_to_ds_map
);
1471 drc
->drc_guid_to_ds_map
= ra
.guid_to_ds_map
;
1475 * Read records and process them.
1478 while (ra
.err
== 0 &&
1479 NULL
!= (drr
= restore_read(&ra
, sizeof (*drr
)))) {
1480 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
1486 backup_byteswap(drr
);
1488 switch (drr
->drr_type
) {
1492 * We need to make a copy of the record header,
1493 * because restore_{object,write} may need to
1494 * restore_read(), which will invalidate drr.
1496 struct drr_object drro
= drr
->drr_u
.drr_object
;
1497 ra
.err
= restore_object(&ra
, os
, &drro
);
1500 case DRR_FREEOBJECTS
:
1502 struct drr_freeobjects drrfo
=
1503 drr
->drr_u
.drr_freeobjects
;
1504 ra
.err
= restore_freeobjects(&ra
, os
, &drrfo
);
1509 struct drr_write drrw
= drr
->drr_u
.drr_write
;
1510 ra
.err
= restore_write(&ra
, os
, &drrw
);
1513 case DRR_WRITE_BYREF
:
1515 struct drr_write_byref drrwbr
=
1516 drr
->drr_u
.drr_write_byref
;
1517 ra
.err
= restore_write_byref(&ra
, os
, &drrwbr
);
1522 struct drr_free drrf
= drr
->drr_u
.drr_free
;
1523 ra
.err
= restore_free(&ra
, os
, &drrf
);
1528 struct drr_end drre
= drr
->drr_u
.drr_end
;
1530 * We compare against the *previous* checksum
1531 * value, because the stored checksum is of
1532 * everything before the DRR_END record.
1534 if (!ZIO_CHECKSUM_EQUAL(drre
.drr_checksum
, pcksum
))
1540 struct drr_spill drrs
= drr
->drr_u
.drr_spill
;
1541 ra
.err
= restore_spill(&ra
, os
, &drrs
);
1550 ASSERT(ra
.err
!= 0);
1553 if ((featureflags
& DMU_BACKUP_FEATURE_DEDUP
) && (cleanup_fd
!= -1))
1554 zfs_onexit_fd_rele(cleanup_fd
);
1558 * destroy what we created, so we don't leave it in the
1559 * inconsistent restoring state.
1561 txg_wait_synced(drc
->drc_real_ds
->ds_dir
->dd_pool
, 0);
1563 (void) dsl_dataset_destroy(drc
->drc_real_ds
, dmu_recv_tag
,
1565 if (drc
->drc_real_ds
!= drc
->drc_logical_ds
) {
1566 mutex_exit(&drc
->drc_logical_ds
->ds_recvlock
);
1567 dsl_dataset_rele(drc
->drc_logical_ds
, dmu_recv_tag
);
1571 vmem_free(ra
.buf
, ra
.bufsize
);
1576 struct recvendsyncarg
{
1578 uint64_t creation_time
;
1583 recv_end_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
1585 dsl_dataset_t
*ds
= arg1
;
1586 struct recvendsyncarg
*resa
= arg2
;
1588 return (dsl_dataset_snapshot_check(ds
, resa
->tosnap
, tx
));
1592 recv_end_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
1594 dsl_dataset_t
*ds
= arg1
;
1595 struct recvendsyncarg
*resa
= arg2
;
1597 dsl_dataset_snapshot_sync(ds
, resa
->tosnap
, tx
);
1599 /* set snapshot's creation time and guid */
1600 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1601 ds
->ds_prev
->ds_phys
->ds_creation_time
= resa
->creation_time
;
1602 ds
->ds_prev
->ds_phys
->ds_guid
= resa
->toguid
;
1603 ds
->ds_prev
->ds_phys
->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
1605 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1606 ds
->ds_phys
->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
1610 add_ds_to_guidmap(avl_tree_t
*guid_map
, dsl_dataset_t
*ds
)
1612 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1613 uint64_t snapobj
= ds
->ds_phys
->ds_prev_snap_obj
;
1614 dsl_dataset_t
*snapds
;
1615 guid_map_entry_t
*gmep
;
1618 ASSERT(guid_map
!= NULL
);
1620 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
1621 err
= dsl_dataset_hold_obj(dp
, snapobj
, guid_map
, &snapds
);
1623 gmep
= kmem_alloc(sizeof (guid_map_entry_t
), KM_SLEEP
);
1624 gmep
->guid
= snapds
->ds_phys
->ds_guid
;
1625 gmep
->gme_ds
= snapds
;
1626 avl_add(guid_map
, gmep
);
1629 rw_exit(&dp
->dp_config_rwlock
);
1634 dmu_recv_existing_end(dmu_recv_cookie_t
*drc
)
1636 struct recvendsyncarg resa
;
1637 dsl_dataset_t
*ds
= drc
->drc_logical_ds
;
1640 if (dsl_dataset_tryown(ds
, FALSE
, dmu_recv_tag
)) {
1641 err
= dsl_dataset_clone_swap(drc
->drc_real_ds
, ds
,
1646 mutex_exit(&ds
->ds_recvlock
);
1647 dsl_dataset_rele(ds
, dmu_recv_tag
);
1648 (void) dsl_dataset_destroy(drc
->drc_real_ds
, dmu_recv_tag
,
1653 resa
.creation_time
= drc
->drc_drrb
->drr_creation_time
;
1654 resa
.toguid
= drc
->drc_drrb
->drr_toguid
;
1655 resa
.tosnap
= drc
->drc_tosnap
;
1657 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
1658 recv_end_check
, recv_end_sync
, ds
, &resa
, 3);
1661 (void) dsl_dataset_clone_swap(drc
->drc_real_ds
, ds
, B_TRUE
);
1665 mutex_exit(&ds
->ds_recvlock
);
1666 if (err
== 0 && drc
->drc_guid_to_ds_map
!= NULL
)
1667 (void) add_ds_to_guidmap(drc
->drc_guid_to_ds_map
, ds
);
1668 dsl_dataset_disown(ds
, dmu_recv_tag
);
1669 myerr
= dsl_dataset_destroy(drc
->drc_real_ds
, dmu_recv_tag
, B_FALSE
);
1675 dmu_recv_new_end(dmu_recv_cookie_t
*drc
)
1677 struct recvendsyncarg resa
;
1678 dsl_dataset_t
*ds
= drc
->drc_logical_ds
;
1682 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1683 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1686 txg_wait_synced(ds
->ds_dir
->dd_pool
, 0);
1688 resa
.creation_time
= drc
->drc_drrb
->drr_creation_time
;
1689 resa
.toguid
= drc
->drc_drrb
->drr_toguid
;
1690 resa
.tosnap
= drc
->drc_tosnap
;
1692 err
= dsl_sync_task_do(ds
->ds_dir
->dd_pool
,
1693 recv_end_check
, recv_end_sync
, ds
, &resa
, 3);
1695 /* clean up the fs we just recv'd into */
1696 (void) dsl_dataset_destroy(ds
, dmu_recv_tag
, B_FALSE
);
1698 if (drc
->drc_guid_to_ds_map
!= NULL
)
1699 (void) add_ds_to_guidmap(drc
->drc_guid_to_ds_map
, ds
);
1700 /* release the hold from dmu_recv_begin */
1701 dsl_dataset_disown(ds
, dmu_recv_tag
);
1707 dmu_recv_end(dmu_recv_cookie_t
*drc
)
1709 if (drc
->drc_logical_ds
!= drc
->drc_real_ds
)
1710 return (dmu_recv_existing_end(drc
));
1712 return (dmu_recv_new_end(drc
));