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 https://opensource.org/licenses/CDDL-1.0.
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 2011 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright 2014 HybridCluster. All rights reserved.
27 * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
28 * Copyright (c) 2019, Klara Inc.
29 * Copyright (c) 2019, Allan Jude
30 * Copyright (c) 2019 Datto Inc.
31 * Copyright (c) 2022 Axcient.
35 #include <sys/spa_impl.h>
37 #include <sys/dmu_impl.h>
38 #include <sys/dmu_send.h>
39 #include <sys/dmu_recv.h>
40 #include <sys/dmu_tx.h>
42 #include <sys/dnode.h>
43 #include <sys/zfs_context.h>
44 #include <sys/dmu_objset.h>
45 #include <sys/dmu_traverse.h>
46 #include <sys/dsl_dataset.h>
47 #include <sys/dsl_dir.h>
48 #include <sys/dsl_prop.h>
49 #include <sys/dsl_pool.h>
50 #include <sys/dsl_synctask.h>
51 #include <sys/zfs_ioctl.h>
54 #include <sys/zio_checksum.h>
55 #include <sys/zfs_znode.h>
56 #include <zfs_fletcher.h>
59 #include <sys/zfs_onexit.h>
60 #include <sys/dsl_destroy.h>
61 #include <sys/blkptr.h>
62 #include <sys/dsl_bookmark.h>
63 #include <sys/zfeature.h>
64 #include <sys/bqueue.h>
65 #include <sys/objlist.h>
67 #include <sys/zfs_vfsops.h>
69 #include <sys/zfs_file.h>
71 static uint_t zfs_recv_queue_length
= SPA_MAXBLOCKSIZE
;
72 static uint_t zfs_recv_queue_ff
= 20;
73 static uint_t zfs_recv_write_batch_size
= 1024 * 1024;
74 static int zfs_recv_best_effort_corrective
= 0;
76 static const void *const dmu_recv_tag
= "dmu_recv_tag";
77 const char *const recv_clone_name
= "%recv";
79 static int receive_read_payload_and_next_header(dmu_recv_cookie_t
*ra
, int len
,
82 struct receive_record_arg
{
83 dmu_replay_record_t header
;
84 void *payload
; /* Pointer to a buffer containing the payload */
86 * If the record is a WRITE or SPILL, pointer to the abd containing the
91 uint64_t bytes_read
; /* bytes read from stream when record created */
92 boolean_t eos_marker
; /* Marks the end of the stream */
96 struct receive_writer_arg
{
102 * These three members are used to signal to the main thread when
113 boolean_t raw
; /* DMU_BACKUP_FEATURE_RAW set */
114 boolean_t spill
; /* DRR_FLAG_SPILL_BLOCK set */
115 boolean_t full
; /* this is a full send stream */
116 uint64_t last_object
;
117 uint64_t last_offset
;
118 uint64_t max_object
; /* highest object ID referenced in stream */
119 uint64_t bytes_read
; /* bytes read when current record created */
123 /* Encryption parameters for the last received DRR_OBJECT_RANGE */
124 boolean_t or_crypt_params_present
;
125 uint64_t or_firstobj
;
126 uint64_t or_numslots
;
127 uint8_t or_salt
[ZIO_DATA_SALT_LEN
];
128 uint8_t or_iv
[ZIO_DATA_IV_LEN
];
129 uint8_t or_mac
[ZIO_DATA_MAC_LEN
];
130 boolean_t or_byteorder
;
134 typedef struct dmu_recv_begin_arg
{
135 const char *drba_origin
;
136 dmu_recv_cookie_t
*drba_cookie
;
139 dsl_crypto_params_t
*drba_dcp
;
140 } dmu_recv_begin_arg_t
;
143 byteswap_record(dmu_replay_record_t
*drr
)
145 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
146 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
147 drr
->drr_type
= BSWAP_32(drr
->drr_type
);
148 drr
->drr_payloadlen
= BSWAP_32(drr
->drr_payloadlen
);
150 switch (drr
->drr_type
) {
152 DO64(drr_begin
.drr_magic
);
153 DO64(drr_begin
.drr_versioninfo
);
154 DO64(drr_begin
.drr_creation_time
);
155 DO32(drr_begin
.drr_type
);
156 DO32(drr_begin
.drr_flags
);
157 DO64(drr_begin
.drr_toguid
);
158 DO64(drr_begin
.drr_fromguid
);
161 DO64(drr_object
.drr_object
);
162 DO32(drr_object
.drr_type
);
163 DO32(drr_object
.drr_bonustype
);
164 DO32(drr_object
.drr_blksz
);
165 DO32(drr_object
.drr_bonuslen
);
166 DO32(drr_object
.drr_raw_bonuslen
);
167 DO64(drr_object
.drr_toguid
);
168 DO64(drr_object
.drr_maxblkid
);
170 case DRR_FREEOBJECTS
:
171 DO64(drr_freeobjects
.drr_firstobj
);
172 DO64(drr_freeobjects
.drr_numobjs
);
173 DO64(drr_freeobjects
.drr_toguid
);
176 DO64(drr_write
.drr_object
);
177 DO32(drr_write
.drr_type
);
178 DO64(drr_write
.drr_offset
);
179 DO64(drr_write
.drr_logical_size
);
180 DO64(drr_write
.drr_toguid
);
181 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write
.drr_key
.ddk_cksum
);
182 DO64(drr_write
.drr_key
.ddk_prop
);
183 DO64(drr_write
.drr_compressed_size
);
185 case DRR_WRITE_EMBEDDED
:
186 DO64(drr_write_embedded
.drr_object
);
187 DO64(drr_write_embedded
.drr_offset
);
188 DO64(drr_write_embedded
.drr_length
);
189 DO64(drr_write_embedded
.drr_toguid
);
190 DO32(drr_write_embedded
.drr_lsize
);
191 DO32(drr_write_embedded
.drr_psize
);
194 DO64(drr_free
.drr_object
);
195 DO64(drr_free
.drr_offset
);
196 DO64(drr_free
.drr_length
);
197 DO64(drr_free
.drr_toguid
);
200 DO64(drr_spill
.drr_object
);
201 DO64(drr_spill
.drr_length
);
202 DO64(drr_spill
.drr_toguid
);
203 DO64(drr_spill
.drr_compressed_size
);
204 DO32(drr_spill
.drr_type
);
206 case DRR_OBJECT_RANGE
:
207 DO64(drr_object_range
.drr_firstobj
);
208 DO64(drr_object_range
.drr_numslots
);
209 DO64(drr_object_range
.drr_toguid
);
212 DO64(drr_redact
.drr_object
);
213 DO64(drr_redact
.drr_offset
);
214 DO64(drr_redact
.drr_length
);
215 DO64(drr_redact
.drr_toguid
);
218 DO64(drr_end
.drr_toguid
);
219 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_end
.drr_checksum
);
225 if (drr
->drr_type
!= DRR_BEGIN
) {
226 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_checksum
.drr_checksum
);
234 redact_snaps_contains(uint64_t *snaps
, uint64_t num_snaps
, uint64_t guid
)
236 for (int i
= 0; i
< num_snaps
; i
++) {
237 if (snaps
[i
] == guid
)
244 * Check that the new stream we're trying to receive is redacted with respect to
245 * a subset of the snapshots that the origin was redacted with respect to. For
246 * the reasons behind this, see the man page on redacted zfs sends and receives.
249 compatible_redact_snaps(uint64_t *origin_snaps
, uint64_t origin_num_snaps
,
250 uint64_t *redact_snaps
, uint64_t num_redact_snaps
)
253 * Short circuit the comparison; if we are redacted with respect to
254 * more snapshots than the origin, we can't be redacted with respect
257 if (num_redact_snaps
> origin_num_snaps
) {
261 for (int i
= 0; i
< num_redact_snaps
; i
++) {
262 if (!redact_snaps_contains(origin_snaps
, origin_num_snaps
,
271 redact_check(dmu_recv_begin_arg_t
*drba
, dsl_dataset_t
*origin
)
273 uint64_t *origin_snaps
;
274 uint64_t origin_num_snaps
;
275 dmu_recv_cookie_t
*drc
= drba
->drba_cookie
;
276 struct drr_begin
*drrb
= drc
->drc_drrb
;
277 int featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
279 boolean_t ret
= B_TRUE
;
280 uint64_t *redact_snaps
;
281 uint_t numredactsnaps
;
284 * If this is a full send stream, we're safe no matter what.
286 if (drrb
->drr_fromguid
== 0)
289 VERIFY(dsl_dataset_get_uint64_array_feature(origin
,
290 SPA_FEATURE_REDACTED_DATASETS
, &origin_num_snaps
, &origin_snaps
));
292 if (nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
293 BEGINNV_REDACT_FROM_SNAPS
, &redact_snaps
, &numredactsnaps
) ==
296 * If the send stream was sent from the redaction bookmark or
297 * the redacted version of the dataset, then we're safe. Verify
298 * that this is from the a compatible redaction bookmark or
301 if (!compatible_redact_snaps(origin_snaps
, origin_num_snaps
,
302 redact_snaps
, numredactsnaps
)) {
305 } else if (featureflags
& DMU_BACKUP_FEATURE_REDACTED
) {
307 * If the stream is redacted, it must be redacted with respect
308 * to a subset of what the origin is redacted with respect to.
309 * See case number 2 in the zfs man page section on redacted zfs
312 err
= nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
313 BEGINNV_REDACT_SNAPS
, &redact_snaps
, &numredactsnaps
);
315 if (err
!= 0 || !compatible_redact_snaps(origin_snaps
,
316 origin_num_snaps
, redact_snaps
, numredactsnaps
)) {
319 } else if (!redact_snaps_contains(origin_snaps
, origin_num_snaps
,
322 * If the stream isn't redacted but the origin is, this must be
323 * one of the snapshots the origin is redacted with respect to.
324 * See case number 1 in the zfs man page section on redacted zfs
336 * If we previously received a stream with --large-block, we don't support
337 * receiving an incremental on top of it without --large-block. This avoids
338 * forcing a read-modify-write or trying to re-aggregate a string of WRITE
342 recv_check_large_blocks(dsl_dataset_t
*ds
, uint64_t featureflags
)
344 if (dsl_dataset_feature_is_active(ds
, SPA_FEATURE_LARGE_BLOCKS
) &&
345 !(featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
))
346 return (SET_ERROR(ZFS_ERR_STREAM_LARGE_BLOCK_MISMATCH
));
351 recv_begin_check_existing_impl(dmu_recv_begin_arg_t
*drba
, dsl_dataset_t
*ds
,
352 uint64_t fromguid
, uint64_t featureflags
)
358 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
359 boolean_t encrypted
= ds
->ds_dir
->dd_crypto_obj
!= 0;
360 boolean_t raw
= (featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0;
361 boolean_t embed
= (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) != 0;
363 /* Temporary clone name must not exist. */
364 error
= zap_lookup(dp
->dp_meta_objset
,
365 dsl_dir_phys(ds
->ds_dir
)->dd_child_dir_zapobj
, recv_clone_name
,
368 return (error
== 0 ? SET_ERROR(EBUSY
) : error
);
370 /* Resume state must not be set. */
371 if (dsl_dataset_has_resume_receive_state(ds
))
372 return (SET_ERROR(EBUSY
));
374 /* New snapshot name must not exist if we're not healing it. */
375 error
= zap_lookup(dp
->dp_meta_objset
,
376 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
377 drba
->drba_cookie
->drc_tosnap
, 8, 1, &obj
);
378 if (drba
->drba_cookie
->drc_heal
) {
381 } else if (error
!= ENOENT
) {
382 return (error
== 0 ? SET_ERROR(EEXIST
) : error
);
385 /* Must not have children if receiving a ZVOL. */
386 error
= zap_count(dp
->dp_meta_objset
,
387 dsl_dir_phys(ds
->ds_dir
)->dd_child_dir_zapobj
, &children
);
390 if (drba
->drba_cookie
->drc_drrb
->drr_type
!= DMU_OST_ZFS
&&
392 return (SET_ERROR(ZFS_ERR_WRONG_PARENT
));
395 * Check snapshot limit before receiving. We'll recheck again at the
396 * end, but might as well abort before receiving if we're already over
399 * Note that we do not check the file system limit with
400 * dsl_dir_fscount_check because the temporary %clones don't count
401 * against that limit.
403 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1, ZFS_PROP_SNAPSHOT_LIMIT
,
404 NULL
, drba
->drba_cred
, drba
->drba_proc
);
408 if (drba
->drba_cookie
->drc_heal
) {
409 /* Encryption is incompatible with embedded data. */
410 if (encrypted
&& embed
)
411 return (SET_ERROR(EINVAL
));
413 /* Healing is not supported when in 'force' mode. */
414 if (drba
->drba_cookie
->drc_force
)
415 return (SET_ERROR(EINVAL
));
417 /* Must have keys loaded if doing encrypted non-raw recv. */
418 if (encrypted
&& !raw
) {
419 if (spa_keystore_lookup_key(dp
->dp_spa
, ds
->ds_object
,
421 return (SET_ERROR(EACCES
));
424 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &snap
);
429 * When not doing best effort corrective recv healing can only
430 * be done if the send stream is for the same snapshot as the
431 * one we are trying to heal.
433 if (zfs_recv_best_effort_corrective
== 0 &&
434 drba
->drba_cookie
->drc_drrb
->drr_toguid
!=
435 dsl_dataset_phys(snap
)->ds_guid
) {
436 dsl_dataset_rele(snap
, FTAG
);
437 return (SET_ERROR(ENOTSUP
));
439 dsl_dataset_rele(snap
, FTAG
);
440 } else if (fromguid
!= 0) {
441 /* Sanity check the incremental recv */
442 uint64_t obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
444 /* Can't perform a raw receive on top of a non-raw receive */
445 if (!encrypted
&& raw
)
446 return (SET_ERROR(EINVAL
));
448 /* Encryption is incompatible with embedded data */
449 if (encrypted
&& embed
)
450 return (SET_ERROR(EINVAL
));
452 /* Find snapshot in this dir that matches fromguid. */
454 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
457 return (SET_ERROR(ENODEV
));
458 if (snap
->ds_dir
!= ds
->ds_dir
) {
459 dsl_dataset_rele(snap
, FTAG
);
460 return (SET_ERROR(ENODEV
));
462 if (dsl_dataset_phys(snap
)->ds_guid
== fromguid
)
464 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
465 dsl_dataset_rele(snap
, FTAG
);
468 return (SET_ERROR(ENODEV
));
470 if (drba
->drba_cookie
->drc_force
) {
471 drba
->drba_cookie
->drc_fromsnapobj
= obj
;
474 * If we are not forcing, there must be no
475 * changes since fromsnap. Raw sends have an
476 * additional constraint that requires that
477 * no "noop" snapshots exist between fromsnap
478 * and tosnap for the IVset checking code to
481 if (dsl_dataset_modified_since_snap(ds
, snap
) ||
483 dsl_dataset_phys(ds
)->ds_prev_snap_obj
!=
485 dsl_dataset_rele(snap
, FTAG
);
486 return (SET_ERROR(ETXTBSY
));
488 drba
->drba_cookie
->drc_fromsnapobj
=
489 ds
->ds_prev
->ds_object
;
492 if (dsl_dataset_feature_is_active(snap
,
493 SPA_FEATURE_REDACTED_DATASETS
) && !redact_check(drba
,
495 dsl_dataset_rele(snap
, FTAG
);
496 return (SET_ERROR(EINVAL
));
499 error
= recv_check_large_blocks(snap
, featureflags
);
501 dsl_dataset_rele(snap
, FTAG
);
505 dsl_dataset_rele(snap
, FTAG
);
507 /* If full and not healing then must be forced. */
508 if (!drba
->drba_cookie
->drc_force
)
509 return (SET_ERROR(EEXIST
));
512 * We don't support using zfs recv -F to blow away
513 * encrypted filesystems. This would require the
514 * dsl dir to point to the old encryption key and
515 * the new one at the same time during the receive.
517 if ((!encrypted
&& raw
) || encrypted
)
518 return (SET_ERROR(EINVAL
));
521 * Perform the same encryption checks we would if
522 * we were creating a new dataset from scratch.
525 boolean_t will_encrypt
;
527 error
= dmu_objset_create_crypt_check(
528 ds
->ds_dir
->dd_parent
, drba
->drba_dcp
,
533 if (will_encrypt
&& embed
)
534 return (SET_ERROR(EINVAL
));
542 * Check that any feature flags used in the data stream we're receiving are
543 * supported by the pool we are receiving into.
545 * Note that some of the features we explicitly check here have additional
546 * (implicit) features they depend on, but those dependencies are enforced
547 * through the zfeature_register() calls declaring the features that we
551 recv_begin_check_feature_flags_impl(uint64_t featureflags
, spa_t
*spa
)
554 * Check if there are any unsupported feature flags.
556 if (!DMU_STREAM_SUPPORTED(featureflags
)) {
557 return (SET_ERROR(ZFS_ERR_UNKNOWN_SEND_STREAM_FEATURE
));
560 /* Verify pool version supports SA if SA_SPILL feature set */
561 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
562 spa_version(spa
) < SPA_VERSION_SA
)
563 return (SET_ERROR(ENOTSUP
));
566 * LZ4 compressed, ZSTD compressed, embedded, mooched, large blocks,
567 * and large_dnodes in the stream can only be used if those pool
568 * features are enabled because we don't attempt to decompress /
569 * un-embed / un-mooch / split up the blocks / dnodes during the
572 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
573 !spa_feature_is_enabled(spa
, SPA_FEATURE_LZ4_COMPRESS
))
574 return (SET_ERROR(ENOTSUP
));
575 if ((featureflags
& DMU_BACKUP_FEATURE_ZSTD
) &&
576 !spa_feature_is_enabled(spa
, SPA_FEATURE_ZSTD_COMPRESS
))
577 return (SET_ERROR(ENOTSUP
));
578 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
579 !spa_feature_is_enabled(spa
, SPA_FEATURE_EMBEDDED_DATA
))
580 return (SET_ERROR(ENOTSUP
));
581 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
582 !spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
))
583 return (SET_ERROR(ENOTSUP
));
584 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_DNODE
) &&
585 !spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
))
586 return (SET_ERROR(ENOTSUP
));
589 * Receiving redacted streams requires that redacted datasets are
592 if ((featureflags
& DMU_BACKUP_FEATURE_REDACTED
) &&
593 !spa_feature_is_enabled(spa
, SPA_FEATURE_REDACTED_DATASETS
))
594 return (SET_ERROR(ENOTSUP
));
600 dmu_recv_begin_check(void *arg
, dmu_tx_t
*tx
)
602 dmu_recv_begin_arg_t
*drba
= arg
;
603 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
604 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
605 uint64_t fromguid
= drrb
->drr_fromguid
;
606 int flags
= drrb
->drr_flags
;
607 ds_hold_flags_t dsflags
= DS_HOLD_FLAG_NONE
;
609 uint64_t featureflags
= drba
->drba_cookie
->drc_featureflags
;
611 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
613 /* already checked */
614 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
615 ASSERT(!(featureflags
& DMU_BACKUP_FEATURE_RESUMING
));
617 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
618 DMU_COMPOUNDSTREAM
||
619 drrb
->drr_type
>= DMU_OST_NUMTYPES
||
620 ((flags
& DRR_FLAG_CLONE
) && drba
->drba_origin
== NULL
))
621 return (SET_ERROR(EINVAL
));
623 error
= recv_begin_check_feature_flags_impl(featureflags
, dp
->dp_spa
);
627 /* Resumable receives require extensible datasets */
628 if (drba
->drba_cookie
->drc_resumable
&&
629 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EXTENSIBLE_DATASET
))
630 return (SET_ERROR(ENOTSUP
));
632 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
633 /* raw receives require the encryption feature */
634 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_ENCRYPTION
))
635 return (SET_ERROR(ENOTSUP
));
637 /* embedded data is incompatible with encryption and raw recv */
638 if (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)
639 return (SET_ERROR(EINVAL
));
641 /* raw receives require spill block allocation flag */
642 if (!(flags
& DRR_FLAG_SPILL_BLOCK
))
643 return (SET_ERROR(ZFS_ERR_SPILL_BLOCK_FLAG_MISSING
));
646 * We support unencrypted datasets below encrypted ones now,
647 * so add the DS_HOLD_FLAG_DECRYPT flag only if we are dealing
648 * with a dataset we may encrypt.
650 if (drba
->drba_dcp
== NULL
||
651 drba
->drba_dcp
->cp_crypt
!= ZIO_CRYPT_OFF
) {
652 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
656 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
658 /* target fs already exists; recv into temp clone */
660 /* Can't recv a clone into an existing fs */
661 if (flags
& DRR_FLAG_CLONE
|| drba
->drba_origin
) {
662 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
663 return (SET_ERROR(EINVAL
));
666 error
= recv_begin_check_existing_impl(drba
, ds
, fromguid
,
668 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
669 } else if (error
== ENOENT
) {
670 /* target fs does not exist; must be a full backup or clone */
671 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
674 /* healing recv must be done "into" an existing snapshot */
675 if (drba
->drba_cookie
->drc_heal
== B_TRUE
)
676 return (SET_ERROR(ENOTSUP
));
679 * If it's a non-clone incremental, we are missing the
680 * target fs, so fail the recv.
682 if (fromguid
!= 0 && !((flags
& DRR_FLAG_CLONE
) ||
684 return (SET_ERROR(ENOENT
));
687 * If we're receiving a full send as a clone, and it doesn't
688 * contain all the necessary free records and freeobject
689 * records, reject it.
691 if (fromguid
== 0 && drba
->drba_origin
!= NULL
&&
692 !(flags
& DRR_FLAG_FREERECORDS
))
693 return (SET_ERROR(EINVAL
));
695 /* Open the parent of tofs */
696 ASSERT3U(strlen(tofs
), <, sizeof (buf
));
697 (void) strlcpy(buf
, tofs
, strrchr(tofs
, '/') - tofs
+ 1);
698 error
= dsl_dataset_hold(dp
, buf
, FTAG
, &ds
);
702 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0 &&
703 drba
->drba_origin
== NULL
) {
704 boolean_t will_encrypt
;
707 * Check that we aren't breaking any encryption rules
708 * and that we have all the parameters we need to
709 * create an encrypted dataset if necessary. If we are
710 * making an encrypted dataset the stream can't have
713 error
= dmu_objset_create_crypt_check(ds
->ds_dir
,
714 drba
->drba_dcp
, &will_encrypt
);
716 dsl_dataset_rele(ds
, FTAG
);
721 (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)) {
722 dsl_dataset_rele(ds
, FTAG
);
723 return (SET_ERROR(EINVAL
));
728 * Check filesystem and snapshot limits before receiving. We'll
729 * recheck snapshot limits again at the end (we create the
730 * filesystems and increment those counts during begin_sync).
732 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
733 ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
734 drba
->drba_cred
, drba
->drba_proc
);
736 dsl_dataset_rele(ds
, FTAG
);
740 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
741 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
,
742 drba
->drba_cred
, drba
->drba_proc
);
744 dsl_dataset_rele(ds
, FTAG
);
748 /* can't recv below anything but filesystems (eg. no ZVOLs) */
749 error
= dmu_objset_from_ds(ds
, &os
);
751 dsl_dataset_rele(ds
, FTAG
);
754 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
755 dsl_dataset_rele(ds
, FTAG
);
756 return (SET_ERROR(ZFS_ERR_WRONG_PARENT
));
759 if (drba
->drba_origin
!= NULL
) {
760 dsl_dataset_t
*origin
;
761 error
= dsl_dataset_hold_flags(dp
, drba
->drba_origin
,
762 dsflags
, FTAG
, &origin
);
764 dsl_dataset_rele(ds
, FTAG
);
767 if (!origin
->ds_is_snapshot
) {
768 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
769 dsl_dataset_rele(ds
, FTAG
);
770 return (SET_ERROR(EINVAL
));
772 if (dsl_dataset_phys(origin
)->ds_guid
!= fromguid
&&
774 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
775 dsl_dataset_rele(ds
, FTAG
);
776 return (SET_ERROR(ENODEV
));
779 if (origin
->ds_dir
->dd_crypto_obj
!= 0 &&
780 (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)) {
781 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
782 dsl_dataset_rele(ds
, FTAG
);
783 return (SET_ERROR(EINVAL
));
787 * If the origin is redacted we need to verify that this
788 * send stream can safely be received on top of the
791 if (dsl_dataset_feature_is_active(origin
,
792 SPA_FEATURE_REDACTED_DATASETS
)) {
793 if (!redact_check(drba
, origin
)) {
794 dsl_dataset_rele_flags(origin
, dsflags
,
796 dsl_dataset_rele_flags(ds
, dsflags
,
798 return (SET_ERROR(EINVAL
));
802 error
= recv_check_large_blocks(ds
, featureflags
);
804 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
805 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
809 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
812 dsl_dataset_rele(ds
, FTAG
);
819 dmu_recv_begin_sync(void *arg
, dmu_tx_t
*tx
)
821 dmu_recv_begin_arg_t
*drba
= arg
;
822 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
823 objset_t
*mos
= dp
->dp_meta_objset
;
824 dmu_recv_cookie_t
*drc
= drba
->drba_cookie
;
825 struct drr_begin
*drrb
= drc
->drc_drrb
;
826 const char *tofs
= drc
->drc_tofs
;
827 uint64_t featureflags
= drc
->drc_featureflags
;
828 dsl_dataset_t
*ds
, *newds
;
831 ds_hold_flags_t dsflags
= DS_HOLD_FLAG_NONE
;
833 uint64_t crflags
= 0;
834 dsl_crypto_params_t dummy_dcp
= { 0 };
835 dsl_crypto_params_t
*dcp
= drba
->drba_dcp
;
837 if (drrb
->drr_flags
& DRR_FLAG_CI_DATA
)
838 crflags
|= DS_FLAG_CI_DATASET
;
840 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0)
841 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
844 * Raw, non-incremental recvs always use a dummy dcp with
845 * the raw cmd set. Raw incremental recvs do not use a dcp
846 * since the encryption parameters are already set in stone.
848 if (dcp
== NULL
&& drrb
->drr_fromguid
== 0 &&
849 drba
->drba_origin
== NULL
) {
850 ASSERT3P(dcp
, ==, NULL
);
853 if (featureflags
& DMU_BACKUP_FEATURE_RAW
)
854 dcp
->cp_cmd
= DCP_CMD_RAW_RECV
;
857 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
859 /* Create temporary clone unless we're doing corrective recv */
860 dsl_dataset_t
*snap
= NULL
;
862 if (drba
->drba_cookie
->drc_fromsnapobj
!= 0) {
863 VERIFY0(dsl_dataset_hold_obj(dp
,
864 drba
->drba_cookie
->drc_fromsnapobj
, FTAG
, &snap
));
865 ASSERT3P(dcp
, ==, NULL
);
868 /* When healing we want to use the provided snapshot */
869 VERIFY0(dsl_dataset_snap_lookup(ds
, drc
->drc_tosnap
,
872 dsobj
= dsl_dataset_create_sync(ds
->ds_dir
,
873 recv_clone_name
, snap
, crflags
, drba
->drba_cred
,
876 if (drba
->drba_cookie
->drc_fromsnapobj
!= 0)
877 dsl_dataset_rele(snap
, FTAG
);
878 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
882 dsl_dataset_t
*origin
= NULL
;
884 VERIFY0(dsl_dir_hold(dp
, tofs
, FTAG
, &dd
, &tail
));
886 if (drba
->drba_origin
!= NULL
) {
887 VERIFY0(dsl_dataset_hold(dp
, drba
->drba_origin
,
889 ASSERT3P(dcp
, ==, NULL
);
892 /* Create new dataset. */
893 dsobj
= dsl_dataset_create_sync(dd
, strrchr(tofs
, '/') + 1,
894 origin
, crflags
, drba
->drba_cred
, dcp
, tx
);
896 dsl_dataset_rele(origin
, FTAG
);
897 dsl_dir_rele(dd
, FTAG
);
898 drc
->drc_newfs
= B_TRUE
;
900 VERIFY0(dsl_dataset_own_obj_force(dp
, dsobj
, dsflags
, dmu_recv_tag
,
902 if (dsl_dataset_feature_is_active(newds
,
903 SPA_FEATURE_REDACTED_DATASETS
)) {
905 * If the origin dataset is redacted, the child will be redacted
906 * when we create it. We clear the new dataset's
907 * redaction info; if it should be redacted, we'll fill
908 * in its information later.
910 dsl_dataset_deactivate_feature(newds
,
911 SPA_FEATURE_REDACTED_DATASETS
, tx
);
913 VERIFY0(dmu_objset_from_ds(newds
, &os
));
915 if (drc
->drc_resumable
) {
916 dsl_dataset_zapify(newds
, tx
);
917 if (drrb
->drr_fromguid
!= 0) {
918 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_FROMGUID
,
919 8, 1, &drrb
->drr_fromguid
, tx
));
921 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TOGUID
,
922 8, 1, &drrb
->drr_toguid
, tx
));
923 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TONAME
,
924 1, strlen(drrb
->drr_toname
) + 1, drrb
->drr_toname
, tx
));
927 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OBJECT
,
929 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OFFSET
,
931 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_BYTES
,
933 if (featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) {
934 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_LARGEBLOCK
,
937 if (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) {
938 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_EMBEDOK
,
941 if (featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
) {
942 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_COMPRESSOK
,
945 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
946 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_RAWOK
,
950 uint64_t *redact_snaps
;
951 uint_t numredactsnaps
;
952 if (nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
953 BEGINNV_REDACT_FROM_SNAPS
, &redact_snaps
,
954 &numredactsnaps
) == 0) {
955 VERIFY0(zap_add(mos
, dsobj
,
956 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS
,
957 sizeof (*redact_snaps
), numredactsnaps
,
963 * Usually the os->os_encrypted value is tied to the presence of a
964 * DSL Crypto Key object in the dd. However, that will not be received
965 * until dmu_recv_stream(), so we set the value manually for now.
967 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
968 os
->os_encrypted
= B_TRUE
;
969 drba
->drba_cookie
->drc_raw
= B_TRUE
;
972 if (featureflags
& DMU_BACKUP_FEATURE_REDACTED
) {
973 uint64_t *redact_snaps
;
974 uint_t numredactsnaps
;
975 VERIFY0(nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
976 BEGINNV_REDACT_SNAPS
, &redact_snaps
, &numredactsnaps
));
977 dsl_dataset_activate_redaction(newds
, redact_snaps
,
981 dmu_buf_will_dirty(newds
->ds_dbuf
, tx
);
982 dsl_dataset_phys(newds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
985 * If we actually created a non-clone, we need to create the objset
986 * in our new dataset. If this is a raw send we postpone this until
987 * dmu_recv_stream() so that we can allocate the metadnode with the
988 * properties from the DRR_BEGIN payload.
990 rrw_enter(&newds
->ds_bp_rwlock
, RW_READER
, FTAG
);
991 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds
)) &&
992 (featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0 &&
994 (void) dmu_objset_create_impl(dp
->dp_spa
,
995 newds
, dsl_dataset_get_blkptr(newds
), drrb
->drr_type
, tx
);
997 rrw_exit(&newds
->ds_bp_rwlock
, FTAG
);
999 drba
->drba_cookie
->drc_ds
= newds
;
1000 drba
->drba_cookie
->drc_os
= os
;
1002 spa_history_log_internal_ds(newds
, "receive", tx
, " ");
1006 dmu_recv_resume_begin_check(void *arg
, dmu_tx_t
*tx
)
1008 dmu_recv_begin_arg_t
*drba
= arg
;
1009 dmu_recv_cookie_t
*drc
= drba
->drba_cookie
;
1010 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1011 struct drr_begin
*drrb
= drc
->drc_drrb
;
1013 ds_hold_flags_t dsflags
= DS_HOLD_FLAG_NONE
;
1015 const char *tofs
= drc
->drc_tofs
;
1017 /* already checked */
1018 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1019 ASSERT(drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RESUMING
);
1021 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1022 DMU_COMPOUNDSTREAM
||
1023 drrb
->drr_type
>= DMU_OST_NUMTYPES
)
1024 return (SET_ERROR(EINVAL
));
1027 * This is mostly a sanity check since we should have already done these
1028 * checks during a previous attempt to receive the data.
1030 error
= recv_begin_check_feature_flags_impl(drc
->drc_featureflags
,
1035 /* 6 extra bytes for /%recv */
1036 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1038 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
1039 tofs
, recv_clone_name
);
1041 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1042 /* raw receives require spill block allocation flag */
1043 if (!(drrb
->drr_flags
& DRR_FLAG_SPILL_BLOCK
))
1044 return (SET_ERROR(ZFS_ERR_SPILL_BLOCK_FLAG_MISSING
));
1046 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1049 boolean_t recvexist
= B_TRUE
;
1050 if (dsl_dataset_hold_flags(dp
, recvname
, dsflags
, FTAG
, &ds
) != 0) {
1051 /* %recv does not exist; continue in tofs */
1052 recvexist
= B_FALSE
;
1053 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1059 * Resume of full/newfs recv on existing dataset should be done with
1062 if (recvexist
&& drrb
->drr_fromguid
== 0 && !drc
->drc_force
) {
1063 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1064 return (SET_ERROR(ZFS_ERR_RESUME_EXISTS
));
1067 /* check that ds is marked inconsistent */
1068 if (!DS_IS_INCONSISTENT(ds
)) {
1069 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1070 return (SET_ERROR(EINVAL
));
1073 /* check that there is resuming data, and that the toguid matches */
1074 if (!dsl_dataset_is_zapified(ds
)) {
1075 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1076 return (SET_ERROR(EINVAL
));
1079 error
= zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1080 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
);
1081 if (error
!= 0 || drrb
->drr_toguid
!= val
) {
1082 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1083 return (SET_ERROR(EINVAL
));
1087 * Check if the receive is still running. If so, it will be owned.
1088 * Note that nothing else can own the dataset (e.g. after the receive
1089 * fails) because it will be marked inconsistent.
1091 if (dsl_dataset_has_owner(ds
)) {
1092 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1093 return (SET_ERROR(EBUSY
));
1096 /* There should not be any snapshots of this fs yet. */
1097 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
->ds_dir
== ds
->ds_dir
) {
1098 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1099 return (SET_ERROR(EINVAL
));
1103 * Note: resume point will be checked when we process the first WRITE
1107 /* check that the origin matches */
1109 (void) zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1110 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
);
1111 if (drrb
->drr_fromguid
!= val
) {
1112 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1113 return (SET_ERROR(EINVAL
));
1116 if (ds
->ds_prev
!= NULL
&& drrb
->drr_fromguid
!= 0)
1117 drc
->drc_fromsnapobj
= ds
->ds_prev
->ds_object
;
1120 * If we're resuming, and the send is redacted, then the original send
1121 * must have been redacted, and must have been redacted with respect to
1122 * the same snapshots.
1124 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_REDACTED
) {
1125 uint64_t num_ds_redact_snaps
;
1126 uint64_t *ds_redact_snaps
;
1128 uint_t num_stream_redact_snaps
;
1129 uint64_t *stream_redact_snaps
;
1131 if (nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
1132 BEGINNV_REDACT_SNAPS
, &stream_redact_snaps
,
1133 &num_stream_redact_snaps
) != 0) {
1134 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1135 return (SET_ERROR(EINVAL
));
1138 if (!dsl_dataset_get_uint64_array_feature(ds
,
1139 SPA_FEATURE_REDACTED_DATASETS
, &num_ds_redact_snaps
,
1140 &ds_redact_snaps
)) {
1141 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1142 return (SET_ERROR(EINVAL
));
1145 for (int i
= 0; i
< num_ds_redact_snaps
; i
++) {
1146 if (!redact_snaps_contains(ds_redact_snaps
,
1147 num_ds_redact_snaps
, stream_redact_snaps
[i
])) {
1148 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1149 return (SET_ERROR(EINVAL
));
1154 error
= recv_check_large_blocks(ds
, drc
->drc_featureflags
);
1156 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1160 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1165 dmu_recv_resume_begin_sync(void *arg
, dmu_tx_t
*tx
)
1167 dmu_recv_begin_arg_t
*drba
= arg
;
1168 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1169 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1170 uint64_t featureflags
= drba
->drba_cookie
->drc_featureflags
;
1172 ds_hold_flags_t dsflags
= DS_HOLD_FLAG_NONE
;
1173 /* 6 extra bytes for /%recv */
1174 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1176 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s", tofs
,
1179 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1180 drba
->drba_cookie
->drc_raw
= B_TRUE
;
1182 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1185 if (dsl_dataset_own_force(dp
, recvname
, dsflags
, dmu_recv_tag
, &ds
)
1187 /* %recv does not exist; continue in tofs */
1188 VERIFY0(dsl_dataset_own_force(dp
, tofs
, dsflags
, dmu_recv_tag
,
1190 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
1193 ASSERT(DS_IS_INCONSISTENT(ds
));
1194 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1195 ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds
)) ||
1196 drba
->drba_cookie
->drc_raw
);
1197 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
1199 drba
->drba_cookie
->drc_ds
= ds
;
1200 VERIFY0(dmu_objset_from_ds(ds
, &drba
->drba_cookie
->drc_os
));
1201 drba
->drba_cookie
->drc_should_save
= B_TRUE
;
1203 spa_history_log_internal_ds(ds
, "resume receive", tx
, " ");
1207 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1208 * succeeds; otherwise we will leak the holds on the datasets.
1211 dmu_recv_begin(char *tofs
, char *tosnap
, dmu_replay_record_t
*drr_begin
,
1212 boolean_t force
, boolean_t heal
, boolean_t resumable
, nvlist_t
*localprops
,
1213 nvlist_t
*hidden_args
, char *origin
, dmu_recv_cookie_t
*drc
,
1214 zfs_file_t
*fp
, offset_t
*voffp
)
1216 dmu_recv_begin_arg_t drba
= { 0 };
1219 memset(drc
, 0, sizeof (dmu_recv_cookie_t
));
1220 drc
->drc_drr_begin
= drr_begin
;
1221 drc
->drc_drrb
= &drr_begin
->drr_u
.drr_begin
;
1222 drc
->drc_tosnap
= tosnap
;
1223 drc
->drc_tofs
= tofs
;
1224 drc
->drc_force
= force
;
1225 drc
->drc_heal
= heal
;
1226 drc
->drc_resumable
= resumable
;
1227 drc
->drc_cred
= CRED();
1228 drc
->drc_proc
= curproc
;
1229 drc
->drc_clone
= (origin
!= NULL
);
1231 if (drc
->drc_drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
)) {
1232 drc
->drc_byteswap
= B_TRUE
;
1233 (void) fletcher_4_incremental_byteswap(drr_begin
,
1234 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
1235 byteswap_record(drr_begin
);
1236 } else if (drc
->drc_drrb
->drr_magic
== DMU_BACKUP_MAGIC
) {
1237 (void) fletcher_4_incremental_native(drr_begin
,
1238 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
1240 return (SET_ERROR(EINVAL
));
1244 drc
->drc_voff
= *voffp
;
1245 drc
->drc_featureflags
=
1246 DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
);
1248 uint32_t payloadlen
= drc
->drc_drr_begin
->drr_payloadlen
;
1249 void *payload
= NULL
;
1252 * Since OpenZFS 2.0.0, we have enforced a 64MB limit in userspace
1253 * configurable via ZFS_SENDRECV_MAX_NVLIST. We enforce 256MB as a hard
1254 * upper limit. Systems with less than 1GB of RAM will see a lower
1255 * limit from `arc_all_memory() / 4`.
1257 if (payloadlen
> (MIN((1U << 28), arc_all_memory() / 4)))
1260 if (payloadlen
!= 0)
1261 payload
= vmem_alloc(payloadlen
, KM_SLEEP
);
1263 err
= receive_read_payload_and_next_header(drc
, payloadlen
,
1266 vmem_free(payload
, payloadlen
);
1269 if (payloadlen
!= 0) {
1270 err
= nvlist_unpack(payload
, payloadlen
, &drc
->drc_begin_nvl
,
1272 vmem_free(payload
, payloadlen
);
1274 kmem_free(drc
->drc_next_rrd
,
1275 sizeof (*drc
->drc_next_rrd
));
1280 if (drc
->drc_drrb
->drr_flags
& DRR_FLAG_SPILL_BLOCK
)
1281 drc
->drc_spill
= B_TRUE
;
1283 drba
.drba_origin
= origin
;
1284 drba
.drba_cookie
= drc
;
1285 drba
.drba_cred
= CRED();
1286 drba
.drba_proc
= curproc
;
1288 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
1289 err
= dsl_sync_task(tofs
,
1290 dmu_recv_resume_begin_check
, dmu_recv_resume_begin_sync
,
1291 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
);
1294 * For non-raw, non-incremental, non-resuming receives the
1295 * user can specify encryption parameters on the command line
1296 * with "zfs recv -o". For these receives we create a dcp and
1297 * pass it to the sync task. Creating the dcp will implicitly
1298 * remove the encryption params from the localprops nvlist,
1299 * which avoids errors when trying to set these normally
1300 * read-only properties. Any other kind of receive that
1301 * attempts to set these properties will fail as a result.
1303 if ((DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
) &
1304 DMU_BACKUP_FEATURE_RAW
) == 0 &&
1305 origin
== NULL
&& drc
->drc_drrb
->drr_fromguid
== 0) {
1306 err
= dsl_crypto_params_create_nvlist(DCP_CMD_NONE
,
1307 localprops
, hidden_args
, &drba
.drba_dcp
);
1311 err
= dsl_sync_task(tofs
,
1312 dmu_recv_begin_check
, dmu_recv_begin_sync
,
1313 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
);
1314 dsl_crypto_params_free(drba
.drba_dcp
, !!err
);
1319 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
1320 nvlist_free(drc
->drc_begin_nvl
);
1326 * Holds data need for corrective recv callback
1328 typedef struct cr_cb_data
{
1330 zbookmark_phys_t zb
;
1335 corrective_read_done(zio_t
*zio
)
1337 cr_cb_data_t
*data
= zio
->io_private
;
1338 /* Corruption corrected; update error log if needed */
1339 if (zio
->io_error
== 0)
1340 spa_remove_error(data
->spa
, &data
->zb
);
1341 kmem_free(data
, sizeof (cr_cb_data_t
));
1342 abd_free(zio
->io_abd
);
1346 * zio_rewrite the data pointed to by bp with the data from the rrd's abd.
1349 do_corrective_recv(struct receive_writer_arg
*rwa
, struct drr_write
*drrw
,
1350 struct receive_record_arg
*rrd
, blkptr_t
*bp
)
1354 zbookmark_phys_t zb
;
1356 abd_t
*abd
= rrd
->abd
;
1357 zio_cksum_t bp_cksum
= bp
->blk_cksum
;
1358 zio_flag_t flags
= ZIO_FLAG_SPECULATIVE
|
1359 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_RETRY
| ZIO_FLAG_CANFAIL
;
1362 flags
|= ZIO_FLAG_RAW
;
1364 err
= dnode_hold(rwa
->os
, drrw
->drr_object
, FTAG
, &dn
);
1367 SET_BOOKMARK(&zb
, dmu_objset_id(rwa
->os
), drrw
->drr_object
, 0,
1368 dbuf_whichblock(dn
, 0, drrw
->drr_offset
));
1369 dnode_rele(dn
, FTAG
);
1371 if (!rwa
->raw
&& DRR_WRITE_COMPRESSED(drrw
)) {
1372 /* Decompress the stream data */
1373 abd_t
*dabd
= abd_alloc_linear(
1374 drrw
->drr_logical_size
, B_FALSE
);
1375 err
= zio_decompress_data(drrw
->drr_compressiontype
,
1376 abd
, abd_to_buf(dabd
), abd_get_size(abd
),
1377 abd_get_size(dabd
), NULL
);
1383 /* Swap in the newly decompressed data into the abd */
1388 if (!rwa
->raw
&& BP_GET_COMPRESS(bp
) != ZIO_COMPRESS_OFF
) {
1389 /* Recompress the data */
1390 abd_t
*cabd
= abd_alloc_linear(BP_GET_PSIZE(bp
),
1392 uint64_t csize
= zio_compress_data(BP_GET_COMPRESS(bp
),
1393 abd
, abd_to_buf(cabd
), abd_get_size(abd
),
1394 rwa
->os
->os_complevel
);
1395 abd_zero_off(cabd
, csize
, BP_GET_PSIZE(bp
) - csize
);
1396 /* Swap in newly compressed data into the abd */
1399 flags
|= ZIO_FLAG_RAW_COMPRESS
;
1403 * The stream is not encrypted but the data on-disk is.
1404 * We need to re-encrypt the buf using the same
1405 * encryption type, salt, iv, and mac that was used to encrypt
1406 * the block previosly.
1408 if (!rwa
->raw
&& BP_USES_CRYPT(bp
)) {
1410 dsl_crypto_key_t
*dck
= NULL
;
1411 uint8_t salt
[ZIO_DATA_SALT_LEN
];
1412 uint8_t iv
[ZIO_DATA_IV_LEN
];
1413 uint8_t mac
[ZIO_DATA_MAC_LEN
];
1414 boolean_t no_crypt
= B_FALSE
;
1415 dsl_pool_t
*dp
= dmu_objset_pool(rwa
->os
);
1416 abd_t
*eabd
= abd_alloc_linear(BP_GET_PSIZE(bp
), B_FALSE
);
1418 zio_crypt_decode_params_bp(bp
, salt
, iv
);
1419 zio_crypt_decode_mac_bp(bp
, mac
);
1421 dsl_pool_config_enter(dp
, FTAG
);
1422 err
= dsl_dataset_hold_flags(dp
, rwa
->tofs
,
1423 DS_HOLD_FLAG_DECRYPT
, FTAG
, &ds
);
1425 dsl_pool_config_exit(dp
, FTAG
);
1427 return (SET_ERROR(EACCES
));
1430 /* Look up the key from the spa's keystore */
1431 err
= spa_keystore_lookup_key(rwa
->os
->os_spa
,
1432 zb
.zb_objset
, FTAG
, &dck
);
1434 dsl_dataset_rele_flags(ds
, DS_HOLD_FLAG_DECRYPT
,
1436 dsl_pool_config_exit(dp
, FTAG
);
1438 return (SET_ERROR(EACCES
));
1441 err
= zio_do_crypt_abd(B_TRUE
, &dck
->dck_key
,
1442 BP_GET_TYPE(bp
), BP_SHOULD_BYTESWAP(bp
), salt
, iv
,
1443 mac
, abd_get_size(abd
), abd
, eabd
, &no_crypt
);
1445 spa_keystore_dsl_key_rele(rwa
->os
->os_spa
, dck
, FTAG
);
1446 dsl_dataset_rele_flags(ds
, DS_HOLD_FLAG_DECRYPT
, FTAG
);
1447 dsl_pool_config_exit(dp
, FTAG
);
1454 /* Swap in the newly encrypted data into the abd */
1459 * We want to prevent zio_rewrite() from trying to
1460 * encrypt the data again
1462 flags
|= ZIO_FLAG_RAW_ENCRYPT
;
1466 io
= zio_rewrite(NULL
, rwa
->os
->os_spa
, bp
->blk_birth
, bp
, abd
,
1467 BP_GET_PSIZE(bp
), NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
, flags
, &zb
);
1469 ASSERT(abd_get_size(abd
) == BP_GET_LSIZE(bp
) ||
1470 abd_get_size(abd
) == BP_GET_PSIZE(bp
));
1472 /* compute new bp checksum value and make sure it matches the old one */
1473 zio_checksum_compute(io
, BP_GET_CHECKSUM(bp
), abd
, abd_get_size(abd
));
1474 if (!ZIO_CHECKSUM_EQUAL(bp_cksum
, io
->io_bp
->blk_cksum
)) {
1476 if (zfs_recv_best_effort_corrective
!= 0)
1478 return (SET_ERROR(ECKSUM
));
1481 /* Correct the corruption in place */
1484 cr_cb_data_t
*cb_data
=
1485 kmem_alloc(sizeof (cr_cb_data_t
), KM_SLEEP
);
1486 cb_data
->spa
= rwa
->os
->os_spa
;
1487 cb_data
->size
= drrw
->drr_logical_size
;
1489 /* Test if healing worked by re-reading the bp */
1490 err
= zio_wait(zio_read(rwa
->heal_pio
, rwa
->os
->os_spa
, bp
,
1491 abd_alloc_for_io(drrw
->drr_logical_size
, B_FALSE
),
1492 drrw
->drr_logical_size
, corrective_read_done
,
1493 cb_data
, ZIO_PRIORITY_ASYNC_READ
, flags
, NULL
));
1495 if (err
!= 0 && zfs_recv_best_effort_corrective
!= 0)
1502 receive_read(dmu_recv_cookie_t
*drc
, int len
, void *buf
)
1507 * The code doesn't rely on this (lengths being multiples of 8). See
1508 * comment in dump_bytes.
1510 ASSERT(len
% 8 == 0 ||
1511 (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0);
1513 while (done
< len
) {
1514 ssize_t resid
= len
- done
;
1515 zfs_file_t
*fp
= drc
->drc_fp
;
1516 int err
= zfs_file_read(fp
, (char *)buf
+ done
,
1517 len
- done
, &resid
);
1518 if (err
== 0 && resid
== len
- done
) {
1520 * Note: ECKSUM or ZFS_ERR_STREAM_TRUNCATED indicates
1521 * that the receive was interrupted and can
1522 * potentially be resumed.
1524 err
= SET_ERROR(ZFS_ERR_STREAM_TRUNCATED
);
1526 drc
->drc_voff
+= len
- done
- resid
;
1532 drc
->drc_bytes_read
+= len
;
1534 ASSERT3U(done
, ==, len
);
1538 static inline uint8_t
1539 deduce_nblkptr(dmu_object_type_t bonus_type
, uint64_t bonus_size
)
1541 if (bonus_type
== DMU_OT_SA
) {
1545 ((DN_OLD_MAX_BONUSLEN
-
1546 MIN(DN_OLD_MAX_BONUSLEN
, bonus_size
)) >> SPA_BLKPTRSHIFT
));
1551 save_resume_state(struct receive_writer_arg
*rwa
,
1552 uint64_t object
, uint64_t offset
, dmu_tx_t
*tx
)
1554 int txgoff
= dmu_tx_get_txg(tx
) & TXG_MASK
;
1556 if (!rwa
->resumable
)
1560 * We use ds_resume_bytes[] != 0 to indicate that we need to
1561 * update this on disk, so it must not be 0.
1563 ASSERT(rwa
->bytes_read
!= 0);
1566 * We only resume from write records, which have a valid
1567 * (non-meta-dnode) object number.
1569 ASSERT(object
!= 0);
1572 * For resuming to work correctly, we must receive records in order,
1573 * sorted by object,offset. This is checked by the callers, but
1574 * assert it here for good measure.
1576 ASSERT3U(object
, >=, rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
]);
1577 ASSERT(object
!= rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] ||
1578 offset
>= rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
]);
1579 ASSERT3U(rwa
->bytes_read
, >=,
1580 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
]);
1582 rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] = object
;
1583 rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
] = offset
;
1584 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
] = rwa
->bytes_read
;
1588 receive_object_is_same_generation(objset_t
*os
, uint64_t object
,
1589 dmu_object_type_t old_bonus_type
, dmu_object_type_t new_bonus_type
,
1590 const void *new_bonus
, boolean_t
*samegenp
)
1592 zfs_file_info_t zoi
;
1595 dmu_buf_t
*old_bonus_dbuf
;
1596 err
= dmu_bonus_hold(os
, object
, FTAG
, &old_bonus_dbuf
);
1599 err
= dmu_get_file_info(os
, old_bonus_type
, old_bonus_dbuf
->db_data
,
1601 dmu_buf_rele(old_bonus_dbuf
, FTAG
);
1604 uint64_t old_gen
= zoi
.zfi_generation
;
1606 err
= dmu_get_file_info(os
, new_bonus_type
, new_bonus
, &zoi
);
1609 uint64_t new_gen
= zoi
.zfi_generation
;
1611 *samegenp
= (old_gen
== new_gen
);
1616 receive_handle_existing_object(const struct receive_writer_arg
*rwa
,
1617 const struct drr_object
*drro
, const dmu_object_info_t
*doi
,
1618 const void *bonus_data
,
1619 uint64_t *object_to_hold
, uint32_t *new_blksz
)
1621 uint32_t indblksz
= drro
->drr_indblkshift
?
1622 1ULL << drro
->drr_indblkshift
: 0;
1623 int nblkptr
= deduce_nblkptr(drro
->drr_bonustype
,
1624 drro
->drr_bonuslen
);
1625 uint8_t dn_slots
= drro
->drr_dn_slots
!= 0 ?
1626 drro
->drr_dn_slots
: DNODE_MIN_SLOTS
;
1627 boolean_t do_free_range
= B_FALSE
;
1630 *object_to_hold
= drro
->drr_object
;
1632 /* nblkptr should be bounded by the bonus size and type */
1633 if (rwa
->raw
&& nblkptr
!= drro
->drr_nblkptr
)
1634 return (SET_ERROR(EINVAL
));
1637 * After the previous send stream, the sending system may
1638 * have freed this object, and then happened to re-allocate
1639 * this object number in a later txg. In this case, we are
1640 * receiving a different logical file, and the block size may
1641 * appear to be different. i.e. we may have a different
1642 * block size for this object than what the send stream says.
1643 * In this case we need to remove the object's contents,
1644 * so that its structure can be changed and then its contents
1645 * entirely replaced by subsequent WRITE records.
1647 * If this is a -L (--large-block) incremental stream, and
1648 * the previous stream was not -L, the block size may appear
1649 * to increase. i.e. we may have a smaller block size for
1650 * this object than what the send stream says. In this case
1651 * we need to keep the object's contents and block size
1652 * intact, so that we don't lose parts of the object's
1653 * contents that are not changed by this incremental send
1656 * We can distinguish between the two above cases by using
1657 * the ZPL's generation number (see
1658 * receive_object_is_same_generation()). However, we only
1659 * want to rely on the generation number when absolutely
1660 * necessary, because with raw receives, the generation is
1661 * encrypted. We also want to minimize dependence on the
1662 * ZPL, so that other types of datasets can also be received
1663 * (e.g. ZVOLs, although note that ZVOLS currently do not
1664 * reallocate their objects or change their structure).
1665 * Therefore, we check a number of different cases where we
1666 * know it is safe to discard the object's contents, before
1667 * using the ZPL's generation number to make the above
1670 if (drro
->drr_blksz
!= doi
->doi_data_block_size
) {
1673 * RAW streams always have large blocks, so
1674 * we are sure that the data is not needed
1675 * due to changing --large-block to be on.
1676 * Which is fortunate since the bonus buffer
1677 * (which contains the ZPL generation) is
1678 * encrypted, and the key might not be
1681 do_free_range
= B_TRUE
;
1682 } else if (rwa
->full
) {
1684 * This is a full send stream, so it always
1685 * replaces what we have. Even if the
1686 * generation numbers happen to match, this
1687 * can not actually be the same logical file.
1688 * This is relevant when receiving a full
1691 do_free_range
= B_TRUE
;
1692 } else if (drro
->drr_type
!=
1693 DMU_OT_PLAIN_FILE_CONTENTS
||
1694 doi
->doi_type
!= DMU_OT_PLAIN_FILE_CONTENTS
) {
1696 * PLAIN_FILE_CONTENTS are the only type of
1697 * objects that have ever been stored with
1698 * large blocks, so we don't need the special
1699 * logic below. ZAP blocks can shrink (when
1700 * there's only one block), so we don't want
1701 * to hit the error below about block size
1704 do_free_range
= B_TRUE
;
1705 } else if (doi
->doi_max_offset
<=
1706 doi
->doi_data_block_size
) {
1708 * There is only one block. We can free it,
1709 * because its contents will be replaced by a
1710 * WRITE record. This can not be the no-L ->
1711 * -L case, because the no-L case would have
1712 * resulted in multiple blocks. If we
1713 * supported -L -> no-L, it would not be safe
1714 * to free the file's contents. Fortunately,
1715 * that is not allowed (see
1716 * recv_check_large_blocks()).
1718 do_free_range
= B_TRUE
;
1720 boolean_t is_same_gen
;
1721 err
= receive_object_is_same_generation(rwa
->os
,
1722 drro
->drr_object
, doi
->doi_bonus_type
,
1723 drro
->drr_bonustype
, bonus_data
, &is_same_gen
);
1725 return (SET_ERROR(EINVAL
));
1729 * This is the same logical file, and
1730 * the block size must be increasing.
1731 * It could only decrease if
1732 * --large-block was changed to be
1733 * off, which is checked in
1734 * recv_check_large_blocks().
1736 if (drro
->drr_blksz
<=
1737 doi
->doi_data_block_size
)
1738 return (SET_ERROR(EINVAL
));
1740 * We keep the existing blocksize and
1744 doi
->doi_data_block_size
;
1746 do_free_range
= B_TRUE
;
1751 /* nblkptr can only decrease if the object was reallocated */
1752 if (nblkptr
< doi
->doi_nblkptr
)
1753 do_free_range
= B_TRUE
;
1755 /* number of slots can only change on reallocation */
1756 if (dn_slots
!= doi
->doi_dnodesize
>> DNODE_SHIFT
)
1757 do_free_range
= B_TRUE
;
1760 * For raw sends we also check a few other fields to
1761 * ensure we are preserving the objset structure exactly
1762 * as it was on the receive side:
1763 * - A changed indirect block size
1764 * - A smaller nlevels
1767 if (indblksz
!= doi
->doi_metadata_block_size
)
1768 do_free_range
= B_TRUE
;
1769 if (drro
->drr_nlevels
< doi
->doi_indirection
)
1770 do_free_range
= B_TRUE
;
1773 if (do_free_range
) {
1774 err
= dmu_free_long_range(rwa
->os
, drro
->drr_object
,
1777 return (SET_ERROR(EINVAL
));
1781 * The dmu does not currently support decreasing nlevels
1782 * or changing the number of dnode slots on an object. For
1783 * non-raw sends, this does not matter and the new object
1784 * can just use the previous one's nlevels. For raw sends,
1785 * however, the structure of the received dnode (including
1786 * nlevels and dnode slots) must match that of the send
1787 * side. Therefore, instead of using dmu_object_reclaim(),
1788 * we must free the object completely and call
1789 * dmu_object_claim_dnsize() instead.
1791 if ((rwa
->raw
&& drro
->drr_nlevels
< doi
->doi_indirection
) ||
1792 dn_slots
!= doi
->doi_dnodesize
>> DNODE_SHIFT
) {
1793 err
= dmu_free_long_object(rwa
->os
, drro
->drr_object
);
1795 return (SET_ERROR(EINVAL
));
1797 txg_wait_synced(dmu_objset_pool(rwa
->os
), 0);
1798 *object_to_hold
= DMU_NEW_OBJECT
;
1802 * For raw receives, free everything beyond the new incoming
1803 * maxblkid. Normally this would be done with a DRR_FREE
1804 * record that would come after this DRR_OBJECT record is
1805 * processed. However, for raw receives we manually set the
1806 * maxblkid from the drr_maxblkid and so we must first free
1807 * everything above that blkid to ensure the DMU is always
1808 * consistent with itself. We will never free the first block
1809 * of the object here because a maxblkid of 0 could indicate
1810 * an object with a single block or one with no blocks. This
1811 * free may be skipped when dmu_free_long_range() was called
1812 * above since it covers the entire object's contents.
1814 if (rwa
->raw
&& *object_to_hold
!= DMU_NEW_OBJECT
&& !do_free_range
) {
1815 err
= dmu_free_long_range(rwa
->os
, drro
->drr_object
,
1816 (drro
->drr_maxblkid
+ 1) * doi
->doi_data_block_size
,
1819 return (SET_ERROR(EINVAL
));
1825 receive_object(struct receive_writer_arg
*rwa
, struct drr_object
*drro
,
1828 dmu_object_info_t doi
;
1831 uint32_t new_blksz
= drro
->drr_blksz
;
1832 uint8_t dn_slots
= drro
->drr_dn_slots
!= 0 ?
1833 drro
->drr_dn_slots
: DNODE_MIN_SLOTS
;
1835 if (drro
->drr_type
== DMU_OT_NONE
||
1836 !DMU_OT_IS_VALID(drro
->drr_type
) ||
1837 !DMU_OT_IS_VALID(drro
->drr_bonustype
) ||
1838 drro
->drr_checksumtype
>= ZIO_CHECKSUM_FUNCTIONS
||
1839 drro
->drr_compress
>= ZIO_COMPRESS_FUNCTIONS
||
1840 P2PHASE(drro
->drr_blksz
, SPA_MINBLOCKSIZE
) ||
1841 drro
->drr_blksz
< SPA_MINBLOCKSIZE
||
1842 drro
->drr_blksz
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)) ||
1843 drro
->drr_bonuslen
>
1844 DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa
->os
))) ||
1846 (spa_maxdnodesize(dmu_objset_spa(rwa
->os
)) >> DNODE_SHIFT
)) {
1847 return (SET_ERROR(EINVAL
));
1852 * We should have received a DRR_OBJECT_RANGE record
1853 * containing this block and stored it in rwa.
1855 if (drro
->drr_object
< rwa
->or_firstobj
||
1856 drro
->drr_object
>= rwa
->or_firstobj
+ rwa
->or_numslots
||
1857 drro
->drr_raw_bonuslen
< drro
->drr_bonuslen
||
1858 drro
->drr_indblkshift
> SPA_MAXBLOCKSHIFT
||
1859 drro
->drr_nlevels
> DN_MAX_LEVELS
||
1860 drro
->drr_nblkptr
> DN_MAX_NBLKPTR
||
1861 DN_SLOTS_TO_BONUSLEN(dn_slots
) <
1862 drro
->drr_raw_bonuslen
)
1863 return (SET_ERROR(EINVAL
));
1866 * The DRR_OBJECT_SPILL flag is valid when the DRR_BEGIN
1867 * record indicates this by setting DRR_FLAG_SPILL_BLOCK.
1869 if (((drro
->drr_flags
& ~(DRR_OBJECT_SPILL
))) ||
1870 (!rwa
->spill
&& DRR_OBJECT_HAS_SPILL(drro
->drr_flags
))) {
1871 return (SET_ERROR(EINVAL
));
1874 if (drro
->drr_raw_bonuslen
!= 0 || drro
->drr_nblkptr
!= 0 ||
1875 drro
->drr_indblkshift
!= 0 || drro
->drr_nlevels
!= 0) {
1876 return (SET_ERROR(EINVAL
));
1880 err
= dmu_object_info(rwa
->os
, drro
->drr_object
, &doi
);
1882 if (err
!= 0 && err
!= ENOENT
&& err
!= EEXIST
)
1883 return (SET_ERROR(EINVAL
));
1885 if (drro
->drr_object
> rwa
->max_object
)
1886 rwa
->max_object
= drro
->drr_object
;
1889 * If we are losing blkptrs or changing the block size this must
1890 * be a new file instance. We must clear out the previous file
1891 * contents before we can change this type of metadata in the dnode.
1892 * Raw receives will also check that the indirect structure of the
1893 * dnode hasn't changed.
1895 uint64_t object_to_hold
;
1897 err
= receive_handle_existing_object(rwa
, drro
, &doi
, data
,
1898 &object_to_hold
, &new_blksz
);
1901 } else if (err
== EEXIST
) {
1903 * The object requested is currently an interior slot of a
1904 * multi-slot dnode. This will be resolved when the next txg
1905 * is synced out, since the send stream will have told us
1906 * to free this slot when we freed the associated dnode
1907 * earlier in the stream.
1909 txg_wait_synced(dmu_objset_pool(rwa
->os
), 0);
1911 if (dmu_object_info(rwa
->os
, drro
->drr_object
, NULL
) != ENOENT
)
1912 return (SET_ERROR(EINVAL
));
1914 /* object was freed and we are about to allocate a new one */
1915 object_to_hold
= DMU_NEW_OBJECT
;
1917 /* object is free and we are about to allocate a new one */
1918 object_to_hold
= DMU_NEW_OBJECT
;
1922 * If this is a multi-slot dnode there is a chance that this
1923 * object will expand into a slot that is already used by
1924 * another object from the previous snapshot. We must free
1925 * these objects before we attempt to allocate the new dnode.
1928 boolean_t need_sync
= B_FALSE
;
1930 for (uint64_t slot
= drro
->drr_object
+ 1;
1931 slot
< drro
->drr_object
+ dn_slots
;
1933 dmu_object_info_t slot_doi
;
1935 err
= dmu_object_info(rwa
->os
, slot
, &slot_doi
);
1936 if (err
== ENOENT
|| err
== EEXIST
)
1941 err
= dmu_free_long_object(rwa
->os
, slot
);
1949 txg_wait_synced(dmu_objset_pool(rwa
->os
), 0);
1952 tx
= dmu_tx_create(rwa
->os
);
1953 dmu_tx_hold_bonus(tx
, object_to_hold
);
1954 dmu_tx_hold_write(tx
, object_to_hold
, 0, 0);
1955 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1961 if (object_to_hold
== DMU_NEW_OBJECT
) {
1962 /* Currently free, wants to be allocated */
1963 err
= dmu_object_claim_dnsize(rwa
->os
, drro
->drr_object
,
1964 drro
->drr_type
, new_blksz
,
1965 drro
->drr_bonustype
, drro
->drr_bonuslen
,
1966 dn_slots
<< DNODE_SHIFT
, tx
);
1967 } else if (drro
->drr_type
!= doi
.doi_type
||
1968 new_blksz
!= doi
.doi_data_block_size
||
1969 drro
->drr_bonustype
!= doi
.doi_bonus_type
||
1970 drro
->drr_bonuslen
!= doi
.doi_bonus_size
) {
1971 /* Currently allocated, but with different properties */
1972 err
= dmu_object_reclaim_dnsize(rwa
->os
, drro
->drr_object
,
1973 drro
->drr_type
, new_blksz
,
1974 drro
->drr_bonustype
, drro
->drr_bonuslen
,
1975 dn_slots
<< DNODE_SHIFT
, rwa
->spill
?
1976 DRR_OBJECT_HAS_SPILL(drro
->drr_flags
) : B_FALSE
, tx
);
1977 } else if (rwa
->spill
&& !DRR_OBJECT_HAS_SPILL(drro
->drr_flags
)) {
1979 * Currently allocated, the existing version of this object
1980 * may reference a spill block that is no longer allocated
1981 * at the source and needs to be freed.
1983 err
= dmu_object_rm_spill(rwa
->os
, drro
->drr_object
, tx
);
1988 return (SET_ERROR(EINVAL
));
1991 if (rwa
->or_crypt_params_present
) {
1993 * Set the crypt params for the buffer associated with this
1994 * range of dnodes. This causes the blkptr_t to have the
1995 * same crypt params (byteorder, salt, iv, mac) as on the
1998 * Since we are committing this tx now, it is possible for
1999 * the dnode block to end up on-disk with the incorrect MAC,
2000 * if subsequent objects in this block are received in a
2001 * different txg. However, since the dataset is marked as
2002 * inconsistent, no code paths will do a non-raw read (or
2003 * decrypt the block / verify the MAC). The receive code and
2004 * scrub code can safely do raw reads and verify the
2005 * checksum. They don't need to verify the MAC.
2007 dmu_buf_t
*db
= NULL
;
2008 uint64_t offset
= rwa
->or_firstobj
* DNODE_MIN_SIZE
;
2010 err
= dmu_buf_hold_by_dnode(DMU_META_DNODE(rwa
->os
),
2011 offset
, FTAG
, &db
, DMU_READ_PREFETCH
| DMU_READ_NO_DECRYPT
);
2014 return (SET_ERROR(EINVAL
));
2017 dmu_buf_set_crypt_params(db
, rwa
->or_byteorder
,
2018 rwa
->or_salt
, rwa
->or_iv
, rwa
->or_mac
, tx
);
2020 dmu_buf_rele(db
, FTAG
);
2022 rwa
->or_crypt_params_present
= B_FALSE
;
2025 dmu_object_set_checksum(rwa
->os
, drro
->drr_object
,
2026 drro
->drr_checksumtype
, tx
);
2027 dmu_object_set_compress(rwa
->os
, drro
->drr_object
,
2028 drro
->drr_compress
, tx
);
2030 /* handle more restrictive dnode structuring for raw recvs */
2033 * Set the indirect block size, block shift, nlevels.
2034 * This will not fail because we ensured all of the
2035 * blocks were freed earlier if this is a new object.
2036 * For non-new objects block size and indirect block
2037 * shift cannot change and nlevels can only increase.
2039 ASSERT3U(new_blksz
, ==, drro
->drr_blksz
);
2040 VERIFY0(dmu_object_set_blocksize(rwa
->os
, drro
->drr_object
,
2041 drro
->drr_blksz
, drro
->drr_indblkshift
, tx
));
2042 VERIFY0(dmu_object_set_nlevels(rwa
->os
, drro
->drr_object
,
2043 drro
->drr_nlevels
, tx
));
2046 * Set the maxblkid. This will always succeed because
2047 * we freed all blocks beyond the new maxblkid above.
2049 VERIFY0(dmu_object_set_maxblkid(rwa
->os
, drro
->drr_object
,
2050 drro
->drr_maxblkid
, tx
));
2056 uint32_t flags
= DMU_READ_NO_PREFETCH
;
2059 flags
|= DMU_READ_NO_DECRYPT
;
2061 VERIFY0(dnode_hold(rwa
->os
, drro
->drr_object
, FTAG
, &dn
));
2062 VERIFY0(dmu_bonus_hold_by_dnode(dn
, FTAG
, &db
, flags
));
2064 dmu_buf_will_dirty(db
, tx
);
2066 ASSERT3U(db
->db_size
, >=, drro
->drr_bonuslen
);
2067 memcpy(db
->db_data
, data
, DRR_OBJECT_PAYLOAD_SIZE(drro
));
2070 * Raw bonus buffers have their byteorder determined by the
2071 * DRR_OBJECT_RANGE record.
2073 if (rwa
->byteswap
&& !rwa
->raw
) {
2074 dmu_object_byteswap_t byteswap
=
2075 DMU_OT_BYTESWAP(drro
->drr_bonustype
);
2076 dmu_ot_byteswap
[byteswap
].ob_func(db
->db_data
,
2077 DRR_OBJECT_PAYLOAD_SIZE(drro
));
2079 dmu_buf_rele(db
, FTAG
);
2080 dnode_rele(dn
, FTAG
);
2088 receive_freeobjects(struct receive_writer_arg
*rwa
,
2089 struct drr_freeobjects
*drrfo
)
2094 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
< drrfo
->drr_firstobj
)
2095 return (SET_ERROR(EINVAL
));
2097 for (obj
= drrfo
->drr_firstobj
== 0 ? 1 : drrfo
->drr_firstobj
;
2098 obj
< drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
&&
2099 obj
< DN_MAX_OBJECT
&& next_err
== 0;
2100 next_err
= dmu_object_next(rwa
->os
, &obj
, FALSE
, 0)) {
2101 dmu_object_info_t doi
;
2104 err
= dmu_object_info(rwa
->os
, obj
, &doi
);
2110 err
= dmu_free_long_object(rwa
->os
, obj
);
2115 if (next_err
!= ESRCH
)
2121 * Note: if this fails, the caller will clean up any records left on the
2122 * rwa->write_batch list.
2125 flush_write_batch_impl(struct receive_writer_arg
*rwa
)
2130 if (dnode_hold(rwa
->os
, rwa
->last_object
, FTAG
, &dn
) != 0)
2131 return (SET_ERROR(EINVAL
));
2133 struct receive_record_arg
*last_rrd
= list_tail(&rwa
->write_batch
);
2134 struct drr_write
*last_drrw
= &last_rrd
->header
.drr_u
.drr_write
;
2136 struct receive_record_arg
*first_rrd
= list_head(&rwa
->write_batch
);
2137 struct drr_write
*first_drrw
= &first_rrd
->header
.drr_u
.drr_write
;
2139 ASSERT3U(rwa
->last_object
, ==, last_drrw
->drr_object
);
2140 ASSERT3U(rwa
->last_offset
, ==, last_drrw
->drr_offset
);
2142 dmu_tx_t
*tx
= dmu_tx_create(rwa
->os
);
2143 dmu_tx_hold_write_by_dnode(tx
, dn
, first_drrw
->drr_offset
,
2144 last_drrw
->drr_offset
- first_drrw
->drr_offset
+
2145 last_drrw
->drr_logical_size
);
2146 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2149 dnode_rele(dn
, FTAG
);
2153 struct receive_record_arg
*rrd
;
2154 while ((rrd
= list_head(&rwa
->write_batch
)) != NULL
) {
2155 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
2156 abd_t
*abd
= rrd
->abd
;
2158 ASSERT3U(drrw
->drr_object
, ==, rwa
->last_object
);
2160 if (drrw
->drr_logical_size
!= dn
->dn_datablksz
) {
2162 * The WRITE record is larger than the object's block
2163 * size. We must be receiving an incremental
2164 * large-block stream into a dataset that previously did
2165 * a non-large-block receive. Lightweight writes must
2166 * be exactly one block, so we need to decompress the
2167 * data (if compressed) and do a normal dmu_write().
2169 ASSERT3U(drrw
->drr_logical_size
, >, dn
->dn_datablksz
);
2170 if (DRR_WRITE_COMPRESSED(drrw
)) {
2172 abd_alloc_linear(drrw
->drr_logical_size
,
2175 err
= zio_decompress_data(
2176 drrw
->drr_compressiontype
,
2177 abd
, abd_to_buf(decomp_abd
),
2179 abd_get_size(decomp_abd
), NULL
);
2182 dmu_write_by_dnode(dn
,
2184 drrw
->drr_logical_size
,
2185 abd_to_buf(decomp_abd
), tx
);
2187 abd_free(decomp_abd
);
2189 dmu_write_by_dnode(dn
,
2191 drrw
->drr_logical_size
,
2192 abd_to_buf(abd
), tx
);
2198 dmu_write_policy(rwa
->os
, dn
, 0, 0, &zp
);
2200 zio_flag_t zio_flags
= 0;
2203 zp
.zp_encrypt
= B_TRUE
;
2204 zp
.zp_compress
= drrw
->drr_compressiontype
;
2205 zp
.zp_byteorder
= ZFS_HOST_BYTEORDER
^
2206 !!DRR_IS_RAW_BYTESWAPPED(drrw
->drr_flags
) ^
2208 memcpy(zp
.zp_salt
, drrw
->drr_salt
,
2210 memcpy(zp
.zp_iv
, drrw
->drr_iv
,
2212 memcpy(zp
.zp_mac
, drrw
->drr_mac
,
2214 if (DMU_OT_IS_ENCRYPTED(zp
.zp_type
)) {
2215 zp
.zp_nopwrite
= B_FALSE
;
2216 zp
.zp_copies
= MIN(zp
.zp_copies
,
2217 SPA_DVAS_PER_BP
- 1);
2219 zio_flags
|= ZIO_FLAG_RAW
;
2220 } else if (DRR_WRITE_COMPRESSED(drrw
)) {
2221 ASSERT3U(drrw
->drr_compressed_size
, >, 0);
2222 ASSERT3U(drrw
->drr_logical_size
, >=,
2223 drrw
->drr_compressed_size
);
2224 zp
.zp_compress
= drrw
->drr_compressiontype
;
2225 zio_flags
|= ZIO_FLAG_RAW_COMPRESS
;
2226 } else if (rwa
->byteswap
) {
2228 * Note: compressed blocks never need to be
2229 * byteswapped, because WRITE records for
2230 * metadata blocks are never compressed. The
2231 * exception is raw streams, which are written
2232 * in the original byteorder, and the byteorder
2233 * bit is preserved in the BP by setting
2234 * zp_byteorder above.
2236 dmu_object_byteswap_t byteswap
=
2237 DMU_OT_BYTESWAP(drrw
->drr_type
);
2238 dmu_ot_byteswap
[byteswap
].ob_func(
2240 DRR_WRITE_PAYLOAD_SIZE(drrw
));
2244 * Since this data can't be read until the receive
2245 * completes, we can do a "lightweight" write for
2246 * improved performance.
2248 err
= dmu_lightweight_write_by_dnode(dn
,
2249 drrw
->drr_offset
, abd
, &zp
, zio_flags
, tx
);
2254 * This rrd is left on the list, so the caller will
2255 * free it (and the abd).
2261 * Note: If the receive fails, we want the resume stream to
2262 * start with the same record that we last successfully
2263 * received (as opposed to the next record), so that we can
2264 * verify that we are resuming from the correct location.
2266 save_resume_state(rwa
, drrw
->drr_object
, drrw
->drr_offset
, tx
);
2268 list_remove(&rwa
->write_batch
, rrd
);
2269 kmem_free(rrd
, sizeof (*rrd
));
2273 dnode_rele(dn
, FTAG
);
2278 flush_write_batch(struct receive_writer_arg
*rwa
)
2280 if (list_is_empty(&rwa
->write_batch
))
2284 err
= flush_write_batch_impl(rwa
);
2286 struct receive_record_arg
*rrd
;
2287 while ((rrd
= list_remove_head(&rwa
->write_batch
)) != NULL
) {
2289 kmem_free(rrd
, sizeof (*rrd
));
2292 ASSERT(list_is_empty(&rwa
->write_batch
));
2297 receive_process_write_record(struct receive_writer_arg
*rwa
,
2298 struct receive_record_arg
*rrd
)
2302 ASSERT3U(rrd
->header
.drr_type
, ==, DRR_WRITE
);
2303 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
2305 if (drrw
->drr_offset
+ drrw
->drr_logical_size
< drrw
->drr_offset
||
2306 !DMU_OT_IS_VALID(drrw
->drr_type
))
2307 return (SET_ERROR(EINVAL
));
2313 int flags
= DB_RF_CANFAIL
;
2316 flags
|= DB_RF_NO_DECRYPT
;
2318 if (rwa
->byteswap
) {
2319 dmu_object_byteswap_t byteswap
=
2320 DMU_OT_BYTESWAP(drrw
->drr_type
);
2321 dmu_ot_byteswap
[byteswap
].ob_func(abd_to_buf(rrd
->abd
),
2322 DRR_WRITE_PAYLOAD_SIZE(drrw
));
2325 err
= dmu_buf_hold_noread(rwa
->os
, drrw
->drr_object
,
2326 drrw
->drr_offset
, FTAG
, &dbp
);
2330 /* Try to read the object to see if it needs healing */
2331 err
= dbuf_read((dmu_buf_impl_t
*)dbp
, NULL
, flags
);
2333 * We only try to heal when dbuf_read() returns a ECKSUMs.
2334 * Other errors (even EIO) get returned to caller.
2335 * EIO indicates that the device is not present/accessible,
2336 * so writing to it will likely fail.
2337 * If the block is healthy, we don't want to overwrite it
2340 if (err
!= ECKSUM
) {
2341 dmu_buf_rele(dbp
, FTAG
);
2344 dn
= dmu_buf_dnode_enter(dbp
);
2345 /* Make sure the on-disk block and recv record sizes match */
2346 if (drrw
->drr_logical_size
!=
2347 dn
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
) {
2349 dmu_buf_dnode_exit(dbp
);
2350 dmu_buf_rele(dbp
, FTAG
);
2353 /* Get the block pointer for the corrupted block */
2354 bp
= dmu_buf_get_blkptr(dbp
);
2355 err
= do_corrective_recv(rwa
, drrw
, rrd
, bp
);
2356 dmu_buf_dnode_exit(dbp
);
2357 dmu_buf_rele(dbp
, FTAG
);
2362 * For resuming to work, records must be in increasing order
2363 * by (object, offset).
2365 if (drrw
->drr_object
< rwa
->last_object
||
2366 (drrw
->drr_object
== rwa
->last_object
&&
2367 drrw
->drr_offset
< rwa
->last_offset
)) {
2368 return (SET_ERROR(EINVAL
));
2371 struct receive_record_arg
*first_rrd
= list_head(&rwa
->write_batch
);
2372 struct drr_write
*first_drrw
= &first_rrd
->header
.drr_u
.drr_write
;
2373 uint64_t batch_size
=
2374 MIN(zfs_recv_write_batch_size
, DMU_MAX_ACCESS
/ 2);
2375 if (first_rrd
!= NULL
&&
2376 (drrw
->drr_object
!= first_drrw
->drr_object
||
2377 drrw
->drr_offset
>= first_drrw
->drr_offset
+ batch_size
)) {
2378 err
= flush_write_batch(rwa
);
2383 rwa
->last_object
= drrw
->drr_object
;
2384 rwa
->last_offset
= drrw
->drr_offset
;
2386 if (rwa
->last_object
> rwa
->max_object
)
2387 rwa
->max_object
= rwa
->last_object
;
2389 list_insert_tail(&rwa
->write_batch
, rrd
);
2391 * Return EAGAIN to indicate that we will use this rrd again,
2392 * so the caller should not free it
2398 receive_write_embedded(struct receive_writer_arg
*rwa
,
2399 struct drr_write_embedded
*drrwe
, void *data
)
2404 if (drrwe
->drr_offset
+ drrwe
->drr_length
< drrwe
->drr_offset
)
2405 return (SET_ERROR(EINVAL
));
2407 if (drrwe
->drr_psize
> BPE_PAYLOAD_SIZE
)
2408 return (SET_ERROR(EINVAL
));
2410 if (drrwe
->drr_etype
>= NUM_BP_EMBEDDED_TYPES
)
2411 return (SET_ERROR(EINVAL
));
2412 if (drrwe
->drr_compression
>= ZIO_COMPRESS_FUNCTIONS
)
2413 return (SET_ERROR(EINVAL
));
2415 return (SET_ERROR(EINVAL
));
2417 if (drrwe
->drr_object
> rwa
->max_object
)
2418 rwa
->max_object
= drrwe
->drr_object
;
2420 tx
= dmu_tx_create(rwa
->os
);
2422 dmu_tx_hold_write(tx
, drrwe
->drr_object
,
2423 drrwe
->drr_offset
, drrwe
->drr_length
);
2424 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2430 dmu_write_embedded(rwa
->os
, drrwe
->drr_object
,
2431 drrwe
->drr_offset
, data
, drrwe
->drr_etype
,
2432 drrwe
->drr_compression
, drrwe
->drr_lsize
, drrwe
->drr_psize
,
2433 rwa
->byteswap
^ ZFS_HOST_BYTEORDER
, tx
);
2435 /* See comment in restore_write. */
2436 save_resume_state(rwa
, drrwe
->drr_object
, drrwe
->drr_offset
, tx
);
2442 receive_spill(struct receive_writer_arg
*rwa
, struct drr_spill
*drrs
,
2445 dmu_buf_t
*db
, *db_spill
;
2448 if (drrs
->drr_length
< SPA_MINBLOCKSIZE
||
2449 drrs
->drr_length
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)))
2450 return (SET_ERROR(EINVAL
));
2453 * This is an unmodified spill block which was added to the stream
2454 * to resolve an issue with incorrectly removing spill blocks. It
2455 * should be ignored by current versions of the code which support
2456 * the DRR_FLAG_SPILL_BLOCK flag.
2458 if (rwa
->spill
&& DRR_SPILL_IS_UNMODIFIED(drrs
->drr_flags
)) {
2464 if (!DMU_OT_IS_VALID(drrs
->drr_type
) ||
2465 drrs
->drr_compressiontype
>= ZIO_COMPRESS_FUNCTIONS
||
2466 drrs
->drr_compressed_size
== 0)
2467 return (SET_ERROR(EINVAL
));
2470 if (dmu_object_info(rwa
->os
, drrs
->drr_object
, NULL
) != 0)
2471 return (SET_ERROR(EINVAL
));
2473 if (drrs
->drr_object
> rwa
->max_object
)
2474 rwa
->max_object
= drrs
->drr_object
;
2476 VERIFY0(dmu_bonus_hold(rwa
->os
, drrs
->drr_object
, FTAG
, &db
));
2477 if ((err
= dmu_spill_hold_by_bonus(db
, DMU_READ_NO_DECRYPT
, FTAG
,
2479 dmu_buf_rele(db
, FTAG
);
2483 dmu_tx_t
*tx
= dmu_tx_create(rwa
->os
);
2485 dmu_tx_hold_spill(tx
, db
->db_object
);
2487 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2489 dmu_buf_rele(db
, FTAG
);
2490 dmu_buf_rele(db_spill
, FTAG
);
2496 * Spill blocks may both grow and shrink. When a change in size
2497 * occurs any existing dbuf must be updated to match the logical
2498 * size of the provided arc_buf_t.
2500 if (db_spill
->db_size
!= drrs
->drr_length
) {
2501 dmu_buf_will_fill(db_spill
, tx
);
2502 VERIFY0(dbuf_spill_set_blksz(db_spill
,
2503 drrs
->drr_length
, tx
));
2508 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^
2509 !!DRR_IS_RAW_BYTESWAPPED(drrs
->drr_flags
) ^
2512 abuf
= arc_loan_raw_buf(dmu_objset_spa(rwa
->os
),
2513 drrs
->drr_object
, byteorder
, drrs
->drr_salt
,
2514 drrs
->drr_iv
, drrs
->drr_mac
, drrs
->drr_type
,
2515 drrs
->drr_compressed_size
, drrs
->drr_length
,
2516 drrs
->drr_compressiontype
, 0);
2518 abuf
= arc_loan_buf(dmu_objset_spa(rwa
->os
),
2519 DMU_OT_IS_METADATA(drrs
->drr_type
),
2521 if (rwa
->byteswap
) {
2522 dmu_object_byteswap_t byteswap
=
2523 DMU_OT_BYTESWAP(drrs
->drr_type
);
2524 dmu_ot_byteswap
[byteswap
].ob_func(abd_to_buf(abd
),
2525 DRR_SPILL_PAYLOAD_SIZE(drrs
));
2529 memcpy(abuf
->b_data
, abd_to_buf(abd
), DRR_SPILL_PAYLOAD_SIZE(drrs
));
2531 dbuf_assign_arcbuf((dmu_buf_impl_t
*)db_spill
, abuf
, tx
);
2533 dmu_buf_rele(db
, FTAG
);
2534 dmu_buf_rele(db_spill
, FTAG
);
2541 receive_free(struct receive_writer_arg
*rwa
, struct drr_free
*drrf
)
2545 if (drrf
->drr_length
!= -1ULL &&
2546 drrf
->drr_offset
+ drrf
->drr_length
< drrf
->drr_offset
)
2547 return (SET_ERROR(EINVAL
));
2549 if (dmu_object_info(rwa
->os
, drrf
->drr_object
, NULL
) != 0)
2550 return (SET_ERROR(EINVAL
));
2552 if (drrf
->drr_object
> rwa
->max_object
)
2553 rwa
->max_object
= drrf
->drr_object
;
2555 err
= dmu_free_long_range(rwa
->os
, drrf
->drr_object
,
2556 drrf
->drr_offset
, drrf
->drr_length
);
2562 receive_object_range(struct receive_writer_arg
*rwa
,
2563 struct drr_object_range
*drror
)
2566 * By default, we assume this block is in our native format
2567 * (ZFS_HOST_BYTEORDER). We then take into account whether
2568 * the send stream is byteswapped (rwa->byteswap). Finally,
2569 * we need to byteswap again if this particular block was
2570 * in non-native format on the send side.
2572 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^ rwa
->byteswap
^
2573 !!DRR_IS_RAW_BYTESWAPPED(drror
->drr_flags
);
2576 * Since dnode block sizes are constant, we should not need to worry
2577 * about making sure that the dnode block size is the same on the
2578 * sending and receiving sides for the time being. For non-raw sends,
2579 * this does not matter (and in fact we do not send a DRR_OBJECT_RANGE
2580 * record at all). Raw sends require this record type because the
2581 * encryption parameters are used to protect an entire block of bonus
2582 * buffers. If the size of dnode blocks ever becomes variable,
2583 * handling will need to be added to ensure that dnode block sizes
2584 * match on the sending and receiving side.
2586 if (drror
->drr_numslots
!= DNODES_PER_BLOCK
||
2587 P2PHASE(drror
->drr_firstobj
, DNODES_PER_BLOCK
) != 0 ||
2589 return (SET_ERROR(EINVAL
));
2591 if (drror
->drr_firstobj
> rwa
->max_object
)
2592 rwa
->max_object
= drror
->drr_firstobj
;
2595 * The DRR_OBJECT_RANGE handling must be deferred to receive_object()
2596 * so that the block of dnodes is not written out when it's empty,
2597 * and converted to a HOLE BP.
2599 rwa
->or_crypt_params_present
= B_TRUE
;
2600 rwa
->or_firstobj
= drror
->drr_firstobj
;
2601 rwa
->or_numslots
= drror
->drr_numslots
;
2602 memcpy(rwa
->or_salt
, drror
->drr_salt
, ZIO_DATA_SALT_LEN
);
2603 memcpy(rwa
->or_iv
, drror
->drr_iv
, ZIO_DATA_IV_LEN
);
2604 memcpy(rwa
->or_mac
, drror
->drr_mac
, ZIO_DATA_MAC_LEN
);
2605 rwa
->or_byteorder
= byteorder
;
2611 * Until we have the ability to redact large ranges of data efficiently, we
2612 * process these records as frees.
2615 receive_redact(struct receive_writer_arg
*rwa
, struct drr_redact
*drrr
)
2617 struct drr_free drrf
= {0};
2618 drrf
.drr_length
= drrr
->drr_length
;
2619 drrf
.drr_object
= drrr
->drr_object
;
2620 drrf
.drr_offset
= drrr
->drr_offset
;
2621 drrf
.drr_toguid
= drrr
->drr_toguid
;
2622 return (receive_free(rwa
, &drrf
));
2625 /* used to destroy the drc_ds on error */
2627 dmu_recv_cleanup_ds(dmu_recv_cookie_t
*drc
)
2629 dsl_dataset_t
*ds
= drc
->drc_ds
;
2630 ds_hold_flags_t dsflags
;
2632 dsflags
= (drc
->drc_raw
) ? DS_HOLD_FLAG_NONE
: DS_HOLD_FLAG_DECRYPT
;
2634 * Wait for the txg sync before cleaning up the receive. For
2635 * resumable receives, this ensures that our resume state has
2636 * been written out to disk. For raw receives, this ensures
2637 * that the user accounting code will not attempt to do anything
2638 * after we stopped receiving the dataset.
2640 txg_wait_synced(ds
->ds_dir
->dd_pool
, 0);
2641 ds
->ds_objset
->os_raw_receive
= B_FALSE
;
2643 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2644 if (drc
->drc_resumable
&& drc
->drc_should_save
&&
2645 !BP_IS_HOLE(dsl_dataset_get_blkptr(ds
))) {
2646 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2647 dsl_dataset_disown(ds
, dsflags
, dmu_recv_tag
);
2649 char name
[ZFS_MAX_DATASET_NAME_LEN
];
2650 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2651 dsl_dataset_name(ds
, name
);
2652 dsl_dataset_disown(ds
, dsflags
, dmu_recv_tag
);
2654 (void) dsl_destroy_head(name
);
2659 receive_cksum(dmu_recv_cookie_t
*drc
, int len
, void *buf
)
2661 if (drc
->drc_byteswap
) {
2662 (void) fletcher_4_incremental_byteswap(buf
, len
,
2665 (void) fletcher_4_incremental_native(buf
, len
, &drc
->drc_cksum
);
2670 * Read the payload into a buffer of size len, and update the current record's
2672 * Allocate drc->drc_next_rrd and read the next record's header into
2673 * drc->drc_next_rrd->header.
2674 * Verify checksum of payload and next record.
2677 receive_read_payload_and_next_header(dmu_recv_cookie_t
*drc
, int len
, void *buf
)
2682 ASSERT3U(len
, <=, SPA_MAXBLOCKSIZE
);
2683 err
= receive_read(drc
, len
, buf
);
2686 receive_cksum(drc
, len
, buf
);
2688 /* note: rrd is NULL when reading the begin record's payload */
2689 if (drc
->drc_rrd
!= NULL
) {
2690 drc
->drc_rrd
->payload
= buf
;
2691 drc
->drc_rrd
->payload_size
= len
;
2692 drc
->drc_rrd
->bytes_read
= drc
->drc_bytes_read
;
2695 ASSERT3P(buf
, ==, NULL
);
2698 drc
->drc_prev_cksum
= drc
->drc_cksum
;
2700 drc
->drc_next_rrd
= kmem_zalloc(sizeof (*drc
->drc_next_rrd
), KM_SLEEP
);
2701 err
= receive_read(drc
, sizeof (drc
->drc_next_rrd
->header
),
2702 &drc
->drc_next_rrd
->header
);
2703 drc
->drc_next_rrd
->bytes_read
= drc
->drc_bytes_read
;
2706 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
2707 drc
->drc_next_rrd
= NULL
;
2710 if (drc
->drc_next_rrd
->header
.drr_type
== DRR_BEGIN
) {
2711 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
2712 drc
->drc_next_rrd
= NULL
;
2713 return (SET_ERROR(EINVAL
));
2717 * Note: checksum is of everything up to but not including the
2720 ASSERT3U(offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2721 ==, sizeof (dmu_replay_record_t
) - sizeof (zio_cksum_t
));
2723 offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2724 &drc
->drc_next_rrd
->header
);
2726 zio_cksum_t cksum_orig
=
2727 drc
->drc_next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
2728 zio_cksum_t
*cksump
=
2729 &drc
->drc_next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
2731 if (drc
->drc_byteswap
)
2732 byteswap_record(&drc
->drc_next_rrd
->header
);
2734 if ((!ZIO_CHECKSUM_IS_ZERO(cksump
)) &&
2735 !ZIO_CHECKSUM_EQUAL(drc
->drc_cksum
, *cksump
)) {
2736 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
2737 drc
->drc_next_rrd
= NULL
;
2738 return (SET_ERROR(ECKSUM
));
2741 receive_cksum(drc
, sizeof (cksum_orig
), &cksum_orig
);
2747 * Issue the prefetch reads for any necessary indirect blocks.
2749 * We use the object ignore list to tell us whether or not to issue prefetches
2750 * for a given object. We do this for both correctness (in case the blocksize
2751 * of an object has changed) and performance (if the object doesn't exist, don't
2752 * needlessly try to issue prefetches). We also trim the list as we go through
2753 * the stream to prevent it from growing to an unbounded size.
2755 * The object numbers within will always be in sorted order, and any write
2756 * records we see will also be in sorted order, but they're not sorted with
2757 * respect to each other (i.e. we can get several object records before
2758 * receiving each object's write records). As a result, once we've reached a
2759 * given object number, we can safely remove any reference to lower object
2760 * numbers in the ignore list. In practice, we receive up to 32 object records
2761 * before receiving write records, so the list can have up to 32 nodes in it.
2764 receive_read_prefetch(dmu_recv_cookie_t
*drc
, uint64_t object
, uint64_t offset
,
2767 if (!objlist_exists(drc
->drc_ignore_objlist
, object
)) {
2768 dmu_prefetch(drc
->drc_os
, object
, 1, offset
, length
,
2769 ZIO_PRIORITY_SYNC_READ
);
2774 * Read records off the stream, issuing any necessary prefetches.
2777 receive_read_record(dmu_recv_cookie_t
*drc
)
2781 switch (drc
->drc_rrd
->header
.drr_type
) {
2784 struct drr_object
*drro
=
2785 &drc
->drc_rrd
->header
.drr_u
.drr_object
;
2786 uint32_t size
= DRR_OBJECT_PAYLOAD_SIZE(drro
);
2788 dmu_object_info_t doi
;
2791 buf
= kmem_zalloc(size
, KM_SLEEP
);
2793 err
= receive_read_payload_and_next_header(drc
, size
, buf
);
2795 kmem_free(buf
, size
);
2798 err
= dmu_object_info(drc
->drc_os
, drro
->drr_object
, &doi
);
2800 * See receive_read_prefetch for an explanation why we're
2801 * storing this object in the ignore_obj_list.
2803 if (err
== ENOENT
|| err
== EEXIST
||
2804 (err
== 0 && doi
.doi_data_block_size
!= drro
->drr_blksz
)) {
2805 objlist_insert(drc
->drc_ignore_objlist
,
2811 case DRR_FREEOBJECTS
:
2813 err
= receive_read_payload_and_next_header(drc
, 0, NULL
);
2818 struct drr_write
*drrw
= &drc
->drc_rrd
->header
.drr_u
.drr_write
;
2819 int size
= DRR_WRITE_PAYLOAD_SIZE(drrw
);
2820 abd_t
*abd
= abd_alloc_linear(size
, B_FALSE
);
2821 err
= receive_read_payload_and_next_header(drc
, size
,
2827 drc
->drc_rrd
->abd
= abd
;
2828 receive_read_prefetch(drc
, drrw
->drr_object
, drrw
->drr_offset
,
2829 drrw
->drr_logical_size
);
2832 case DRR_WRITE_EMBEDDED
:
2834 struct drr_write_embedded
*drrwe
=
2835 &drc
->drc_rrd
->header
.drr_u
.drr_write_embedded
;
2836 uint32_t size
= P2ROUNDUP(drrwe
->drr_psize
, 8);
2837 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
2839 err
= receive_read_payload_and_next_header(drc
, size
, buf
);
2841 kmem_free(buf
, size
);
2845 receive_read_prefetch(drc
, drrwe
->drr_object
, drrwe
->drr_offset
,
2853 * It might be beneficial to prefetch indirect blocks here, but
2854 * we don't really have the data to decide for sure.
2856 err
= receive_read_payload_and_next_header(drc
, 0, NULL
);
2861 struct drr_end
*drre
= &drc
->drc_rrd
->header
.drr_u
.drr_end
;
2862 if (!ZIO_CHECKSUM_EQUAL(drc
->drc_prev_cksum
,
2863 drre
->drr_checksum
))
2864 return (SET_ERROR(ECKSUM
));
2869 struct drr_spill
*drrs
= &drc
->drc_rrd
->header
.drr_u
.drr_spill
;
2870 int size
= DRR_SPILL_PAYLOAD_SIZE(drrs
);
2871 abd_t
*abd
= abd_alloc_linear(size
, B_FALSE
);
2872 err
= receive_read_payload_and_next_header(drc
, size
,
2877 drc
->drc_rrd
->abd
= abd
;
2880 case DRR_OBJECT_RANGE
:
2882 err
= receive_read_payload_and_next_header(drc
, 0, NULL
);
2887 return (SET_ERROR(EINVAL
));
2894 dprintf_drr(struct receive_record_arg
*rrd
, int err
)
2897 switch (rrd
->header
.drr_type
) {
2900 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
2901 dprintf("drr_type = OBJECT obj = %llu type = %u "
2902 "bonustype = %u blksz = %u bonuslen = %u cksumtype = %u "
2903 "compress = %u dn_slots = %u err = %d\n",
2904 (u_longlong_t
)drro
->drr_object
, drro
->drr_type
,
2905 drro
->drr_bonustype
, drro
->drr_blksz
, drro
->drr_bonuslen
,
2906 drro
->drr_checksumtype
, drro
->drr_compress
,
2907 drro
->drr_dn_slots
, err
);
2910 case DRR_FREEOBJECTS
:
2912 struct drr_freeobjects
*drrfo
=
2913 &rrd
->header
.drr_u
.drr_freeobjects
;
2914 dprintf("drr_type = FREEOBJECTS firstobj = %llu "
2915 "numobjs = %llu err = %d\n",
2916 (u_longlong_t
)drrfo
->drr_firstobj
,
2917 (u_longlong_t
)drrfo
->drr_numobjs
, err
);
2922 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
2923 dprintf("drr_type = WRITE obj = %llu type = %u offset = %llu "
2924 "lsize = %llu cksumtype = %u flags = %u "
2925 "compress = %u psize = %llu err = %d\n",
2926 (u_longlong_t
)drrw
->drr_object
, drrw
->drr_type
,
2927 (u_longlong_t
)drrw
->drr_offset
,
2928 (u_longlong_t
)drrw
->drr_logical_size
,
2929 drrw
->drr_checksumtype
, drrw
->drr_flags
,
2930 drrw
->drr_compressiontype
,
2931 (u_longlong_t
)drrw
->drr_compressed_size
, err
);
2934 case DRR_WRITE_BYREF
:
2936 struct drr_write_byref
*drrwbr
=
2937 &rrd
->header
.drr_u
.drr_write_byref
;
2938 dprintf("drr_type = WRITE_BYREF obj = %llu offset = %llu "
2939 "length = %llu toguid = %llx refguid = %llx "
2940 "refobject = %llu refoffset = %llu cksumtype = %u "
2941 "flags = %u err = %d\n",
2942 (u_longlong_t
)drrwbr
->drr_object
,
2943 (u_longlong_t
)drrwbr
->drr_offset
,
2944 (u_longlong_t
)drrwbr
->drr_length
,
2945 (u_longlong_t
)drrwbr
->drr_toguid
,
2946 (u_longlong_t
)drrwbr
->drr_refguid
,
2947 (u_longlong_t
)drrwbr
->drr_refobject
,
2948 (u_longlong_t
)drrwbr
->drr_refoffset
,
2949 drrwbr
->drr_checksumtype
, drrwbr
->drr_flags
, err
);
2952 case DRR_WRITE_EMBEDDED
:
2954 struct drr_write_embedded
*drrwe
=
2955 &rrd
->header
.drr_u
.drr_write_embedded
;
2956 dprintf("drr_type = WRITE_EMBEDDED obj = %llu offset = %llu "
2957 "length = %llu compress = %u etype = %u lsize = %u "
2958 "psize = %u err = %d\n",
2959 (u_longlong_t
)drrwe
->drr_object
,
2960 (u_longlong_t
)drrwe
->drr_offset
,
2961 (u_longlong_t
)drrwe
->drr_length
,
2962 drrwe
->drr_compression
, drrwe
->drr_etype
,
2963 drrwe
->drr_lsize
, drrwe
->drr_psize
, err
);
2968 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
2969 dprintf("drr_type = FREE obj = %llu offset = %llu "
2970 "length = %lld err = %d\n",
2971 (u_longlong_t
)drrf
->drr_object
,
2972 (u_longlong_t
)drrf
->drr_offset
,
2973 (longlong_t
)drrf
->drr_length
,
2979 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
2980 dprintf("drr_type = SPILL obj = %llu length = %llu "
2981 "err = %d\n", (u_longlong_t
)drrs
->drr_object
,
2982 (u_longlong_t
)drrs
->drr_length
, err
);
2985 case DRR_OBJECT_RANGE
:
2987 struct drr_object_range
*drror
=
2988 &rrd
->header
.drr_u
.drr_object_range
;
2989 dprintf("drr_type = OBJECT_RANGE firstobj = %llu "
2990 "numslots = %llu flags = %u err = %d\n",
2991 (u_longlong_t
)drror
->drr_firstobj
,
2992 (u_longlong_t
)drror
->drr_numslots
,
2993 drror
->drr_flags
, err
);
3003 * Commit the records to the pool.
3006 receive_process_record(struct receive_writer_arg
*rwa
,
3007 struct receive_record_arg
*rrd
)
3011 /* Processing in order, therefore bytes_read should be increasing. */
3012 ASSERT3U(rrd
->bytes_read
, >=, rwa
->bytes_read
);
3013 rwa
->bytes_read
= rrd
->bytes_read
;
3015 /* We can only heal write records; other ones get ignored */
3016 if (rwa
->heal
&& rrd
->header
.drr_type
!= DRR_WRITE
) {
3017 if (rrd
->abd
!= NULL
) {
3020 } else if (rrd
->payload
!= NULL
) {
3021 kmem_free(rrd
->payload
, rrd
->payload_size
);
3022 rrd
->payload
= NULL
;
3027 if (!rwa
->heal
&& rrd
->header
.drr_type
!= DRR_WRITE
) {
3028 err
= flush_write_batch(rwa
);
3030 if (rrd
->abd
!= NULL
) {
3033 rrd
->payload
= NULL
;
3034 } else if (rrd
->payload
!= NULL
) {
3035 kmem_free(rrd
->payload
, rrd
->payload_size
);
3036 rrd
->payload
= NULL
;
3043 switch (rrd
->header
.drr_type
) {
3046 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
3047 err
= receive_object(rwa
, drro
, rrd
->payload
);
3048 kmem_free(rrd
->payload
, rrd
->payload_size
);
3049 rrd
->payload
= NULL
;
3052 case DRR_FREEOBJECTS
:
3054 struct drr_freeobjects
*drrfo
=
3055 &rrd
->header
.drr_u
.drr_freeobjects
;
3056 err
= receive_freeobjects(rwa
, drrfo
);
3061 err
= receive_process_write_record(rwa
, rrd
);
3064 * If healing - always free the abd after processing
3068 } else if (err
!= EAGAIN
) {
3070 * On success, a non-healing
3071 * receive_process_write_record() returns
3072 * EAGAIN to indicate that we do not want to free
3073 * the rrd or arc_buf.
3081 case DRR_WRITE_EMBEDDED
:
3083 struct drr_write_embedded
*drrwe
=
3084 &rrd
->header
.drr_u
.drr_write_embedded
;
3085 err
= receive_write_embedded(rwa
, drrwe
, rrd
->payload
);
3086 kmem_free(rrd
->payload
, rrd
->payload_size
);
3087 rrd
->payload
= NULL
;
3092 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
3093 err
= receive_free(rwa
, drrf
);
3098 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
3099 err
= receive_spill(rwa
, drrs
, rrd
->abd
);
3103 rrd
->payload
= NULL
;
3106 case DRR_OBJECT_RANGE
:
3108 struct drr_object_range
*drror
=
3109 &rrd
->header
.drr_u
.drr_object_range
;
3110 err
= receive_object_range(rwa
, drror
);
3115 struct drr_redact
*drrr
= &rrd
->header
.drr_u
.drr_redact
;
3116 err
= receive_redact(rwa
, drrr
);
3120 err
= (SET_ERROR(EINVAL
));
3124 dprintf_drr(rrd
, err
);
3130 * dmu_recv_stream's worker thread; pull records off the queue, and then call
3131 * receive_process_record When we're done, signal the main thread and exit.
3133 static __attribute__((noreturn
)) void
3134 receive_writer_thread(void *arg
)
3136 struct receive_writer_arg
*rwa
= arg
;
3137 struct receive_record_arg
*rrd
;
3138 fstrans_cookie_t cookie
= spl_fstrans_mark();
3140 for (rrd
= bqueue_dequeue(&rwa
->q
); !rrd
->eos_marker
;
3141 rrd
= bqueue_dequeue(&rwa
->q
)) {
3143 * If there's an error, the main thread will stop putting things
3144 * on the queue, but we need to clear everything in it before we
3148 if (rwa
->err
== 0) {
3149 err
= receive_process_record(rwa
, rrd
);
3150 } else if (rrd
->abd
!= NULL
) {
3153 rrd
->payload
= NULL
;
3154 } else if (rrd
->payload
!= NULL
) {
3155 kmem_free(rrd
->payload
, rrd
->payload_size
);
3156 rrd
->payload
= NULL
;
3159 * EAGAIN indicates that this record has been saved (on
3160 * raw->write_batch), and will be used again, so we don't
3162 * When healing data we always need to free the record.
3164 if (err
!= EAGAIN
|| rwa
->heal
) {
3167 kmem_free(rrd
, sizeof (*rrd
));
3170 kmem_free(rrd
, sizeof (*rrd
));
3173 zio_wait(rwa
->heal_pio
);
3175 int err
= flush_write_batch(rwa
);
3179 mutex_enter(&rwa
->mutex
);
3181 cv_signal(&rwa
->cv
);
3182 mutex_exit(&rwa
->mutex
);
3183 spl_fstrans_unmark(cookie
);
3188 resume_check(dmu_recv_cookie_t
*drc
, nvlist_t
*begin_nvl
)
3191 objset_t
*mos
= dmu_objset_pool(drc
->drc_os
)->dp_meta_objset
;
3192 uint64_t dsobj
= dmu_objset_id(drc
->drc_os
);
3193 uint64_t resume_obj
, resume_off
;
3195 if (nvlist_lookup_uint64(begin_nvl
,
3196 "resume_object", &resume_obj
) != 0 ||
3197 nvlist_lookup_uint64(begin_nvl
,
3198 "resume_offset", &resume_off
) != 0) {
3199 return (SET_ERROR(EINVAL
));
3201 VERIFY0(zap_lookup(mos
, dsobj
,
3202 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
));
3203 if (resume_obj
!= val
)
3204 return (SET_ERROR(EINVAL
));
3205 VERIFY0(zap_lookup(mos
, dsobj
,
3206 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
));
3207 if (resume_off
!= val
)
3208 return (SET_ERROR(EINVAL
));
3214 * Read in the stream's records, one by one, and apply them to the pool. There
3215 * are two threads involved; the thread that calls this function will spin up a
3216 * worker thread, read the records off the stream one by one, and issue
3217 * prefetches for any necessary indirect blocks. It will then push the records
3218 * onto an internal blocking queue. The worker thread will pull the records off
3219 * the queue, and actually write the data into the DMU. This way, the worker
3220 * thread doesn't have to wait for reads to complete, since everything it needs
3221 * (the indirect blocks) will be prefetched.
3223 * NB: callers *must* call dmu_recv_end() if this succeeds.
3226 dmu_recv_stream(dmu_recv_cookie_t
*drc
, offset_t
*voffp
)
3229 struct receive_writer_arg
*rwa
= kmem_zalloc(sizeof (*rwa
), KM_SLEEP
);
3231 if (dsl_dataset_has_resume_receive_state(drc
->drc_ds
)) {
3233 (void) zap_lookup(drc
->drc_ds
->ds_dir
->dd_pool
->dp_meta_objset
,
3234 drc
->drc_ds
->ds_object
, DS_FIELD_RESUME_BYTES
,
3235 sizeof (bytes
), 1, &bytes
);
3236 drc
->drc_bytes_read
+= bytes
;
3239 drc
->drc_ignore_objlist
= objlist_create();
3241 /* these were verified in dmu_recv_begin */
3242 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc
->drc_drrb
->drr_versioninfo
), ==,
3244 ASSERT3U(drc
->drc_drrb
->drr_type
, <, DMU_OST_NUMTYPES
);
3246 ASSERT(dsl_dataset_phys(drc
->drc_ds
)->ds_flags
& DS_FLAG_INCONSISTENT
);
3247 ASSERT0(drc
->drc_os
->os_encrypted
&&
3248 (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
));
3250 /* handle DSL encryption key payload */
3251 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
3252 nvlist_t
*keynvl
= NULL
;
3254 ASSERT(drc
->drc_os
->os_encrypted
);
3255 ASSERT(drc
->drc_raw
);
3257 err
= nvlist_lookup_nvlist(drc
->drc_begin_nvl
, "crypt_keydata",
3262 if (!drc
->drc_heal
) {
3264 * If this is a new dataset we set the key immediately.
3265 * Otherwise we don't want to change the key until we
3266 * are sure the rest of the receive succeeded so we
3267 * stash the keynvl away until then.
3269 err
= dsl_crypto_recv_raw(spa_name(drc
->drc_os
->os_spa
),
3270 drc
->drc_ds
->ds_object
, drc
->drc_fromsnapobj
,
3271 drc
->drc_drrb
->drr_type
, keynvl
, drc
->drc_newfs
);
3276 /* see comment in dmu_recv_end_sync() */
3277 drc
->drc_ivset_guid
= 0;
3278 (void) nvlist_lookup_uint64(keynvl
, "to_ivset_guid",
3279 &drc
->drc_ivset_guid
);
3281 if (!drc
->drc_newfs
)
3282 drc
->drc_keynvl
= fnvlist_dup(keynvl
);
3285 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
3286 err
= resume_check(drc
, drc
->drc_begin_nvl
);
3292 * If we failed before this point we will clean up any new resume
3293 * state that was created. Now that we've gotten past the initial
3294 * checks we are ok to retain that resume state.
3296 drc
->drc_should_save
= B_TRUE
;
3298 (void) bqueue_init(&rwa
->q
, zfs_recv_queue_ff
,
3299 MAX(zfs_recv_queue_length
, 2 * zfs_max_recordsize
),
3300 offsetof(struct receive_record_arg
, node
));
3301 cv_init(&rwa
->cv
, NULL
, CV_DEFAULT
, NULL
);
3302 mutex_init(&rwa
->mutex
, NULL
, MUTEX_DEFAULT
, NULL
);
3303 rwa
->os
= drc
->drc_os
;
3304 rwa
->byteswap
= drc
->drc_byteswap
;
3305 rwa
->heal
= drc
->drc_heal
;
3306 rwa
->tofs
= drc
->drc_tofs
;
3307 rwa
->resumable
= drc
->drc_resumable
;
3308 rwa
->raw
= drc
->drc_raw
;
3309 rwa
->spill
= drc
->drc_spill
;
3310 rwa
->full
= (drc
->drc_drr_begin
->drr_u
.drr_begin
.drr_fromguid
== 0);
3311 rwa
->os
->os_raw_receive
= drc
->drc_raw
;
3312 if (drc
->drc_heal
) {
3313 rwa
->heal_pio
= zio_root(drc
->drc_os
->os_spa
, NULL
, NULL
,
3314 ZIO_FLAG_GODFATHER
);
3316 list_create(&rwa
->write_batch
, sizeof (struct receive_record_arg
),
3317 offsetof(struct receive_record_arg
, node
.bqn_node
));
3319 (void) thread_create(NULL
, 0, receive_writer_thread
, rwa
, 0, curproc
,
3320 TS_RUN
, minclsyspri
);
3322 * We're reading rwa->err without locks, which is safe since we are the
3323 * only reader, and the worker thread is the only writer. It's ok if we
3324 * miss a write for an iteration or two of the loop, since the writer
3325 * thread will keep freeing records we send it until we send it an eos
3328 * We can leave this loop in 3 ways: First, if rwa->err is
3329 * non-zero. In that case, the writer thread will free the rrd we just
3330 * pushed. Second, if we're interrupted; in that case, either it's the
3331 * first loop and drc->drc_rrd was never allocated, or it's later, and
3332 * drc->drc_rrd has been handed off to the writer thread who will free
3333 * it. Finally, if receive_read_record fails or we're at the end of the
3334 * stream, then we free drc->drc_rrd and exit.
3336 while (rwa
->err
== 0) {
3337 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
3338 err
= SET_ERROR(EINTR
);
3342 ASSERT3P(drc
->drc_rrd
, ==, NULL
);
3343 drc
->drc_rrd
= drc
->drc_next_rrd
;
3344 drc
->drc_next_rrd
= NULL
;
3345 /* Allocates and loads header into drc->drc_next_rrd */
3346 err
= receive_read_record(drc
);
3348 if (drc
->drc_rrd
->header
.drr_type
== DRR_END
|| err
!= 0) {
3349 kmem_free(drc
->drc_rrd
, sizeof (*drc
->drc_rrd
));
3350 drc
->drc_rrd
= NULL
;
3354 bqueue_enqueue(&rwa
->q
, drc
->drc_rrd
,
3355 sizeof (struct receive_record_arg
) +
3356 drc
->drc_rrd
->payload_size
);
3357 drc
->drc_rrd
= NULL
;
3360 ASSERT3P(drc
->drc_rrd
, ==, NULL
);
3361 drc
->drc_rrd
= kmem_zalloc(sizeof (*drc
->drc_rrd
), KM_SLEEP
);
3362 drc
->drc_rrd
->eos_marker
= B_TRUE
;
3363 bqueue_enqueue_flush(&rwa
->q
, drc
->drc_rrd
, 1);
3365 mutex_enter(&rwa
->mutex
);
3366 while (!rwa
->done
) {
3368 * We need to use cv_wait_sig() so that any process that may
3369 * be sleeping here can still fork.
3371 (void) cv_wait_sig(&rwa
->cv
, &rwa
->mutex
);
3373 mutex_exit(&rwa
->mutex
);
3376 * If we are receiving a full stream as a clone, all object IDs which
3377 * are greater than the maximum ID referenced in the stream are
3378 * by definition unused and must be freed.
3380 if (drc
->drc_clone
&& drc
->drc_drrb
->drr_fromguid
== 0) {
3381 uint64_t obj
= rwa
->max_object
+ 1;
3385 while (next_err
== 0) {
3386 free_err
= dmu_free_long_object(rwa
->os
, obj
);
3387 if (free_err
!= 0 && free_err
!= ENOENT
)
3390 next_err
= dmu_object_next(rwa
->os
, &obj
, FALSE
, 0);
3394 if (free_err
!= 0 && free_err
!= ENOENT
)
3396 else if (next_err
!= ESRCH
)
3401 cv_destroy(&rwa
->cv
);
3402 mutex_destroy(&rwa
->mutex
);
3403 bqueue_destroy(&rwa
->q
);
3404 list_destroy(&rwa
->write_batch
);
3410 * If we hit an error before we started the receive_writer_thread
3411 * we need to clean up the next_rrd we create by processing the
3414 if (drc
->drc_next_rrd
!= NULL
)
3415 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
3418 * The objset will be invalidated by dmu_recv_end() when we do
3419 * dsl_dataset_clone_swap_sync_impl().
3423 kmem_free(rwa
, sizeof (*rwa
));
3424 nvlist_free(drc
->drc_begin_nvl
);
3428 * Clean up references. If receive is not resumable,
3429 * destroy what we created, so we don't leave it in
3430 * the inconsistent state.
3432 dmu_recv_cleanup_ds(drc
);
3433 nvlist_free(drc
->drc_keynvl
);
3436 objlist_destroy(drc
->drc_ignore_objlist
);
3437 drc
->drc_ignore_objlist
= NULL
;
3438 *voffp
= drc
->drc_voff
;
3443 dmu_recv_end_check(void *arg
, dmu_tx_t
*tx
)
3445 dmu_recv_cookie_t
*drc
= arg
;
3446 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3449 ASSERT3P(drc
->drc_ds
->ds_owner
, ==, dmu_recv_tag
);
3451 if (drc
->drc_heal
) {
3453 } else if (!drc
->drc_newfs
) {
3454 dsl_dataset_t
*origin_head
;
3456 error
= dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
, &origin_head
);
3459 if (drc
->drc_force
) {
3461 * We will destroy any snapshots in tofs (i.e. before
3462 * origin_head) that are after the origin (which is
3463 * the snap before drc_ds, because drc_ds can not
3464 * have any snaps of its own).
3468 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3470 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3471 dsl_dataset_t
*snap
;
3472 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3476 if (snap
->ds_dir
!= origin_head
->ds_dir
)
3477 error
= SET_ERROR(EINVAL
);
3479 error
= dsl_destroy_snapshot_check_impl(
3482 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3483 dsl_dataset_rele(snap
, FTAG
);
3488 dsl_dataset_rele(origin_head
, FTAG
);
3492 if (drc
->drc_keynvl
!= NULL
) {
3493 error
= dsl_crypto_recv_raw_key_check(drc
->drc_ds
,
3494 drc
->drc_keynvl
, tx
);
3496 dsl_dataset_rele(origin_head
, FTAG
);
3501 error
= dsl_dataset_clone_swap_check_impl(drc
->drc_ds
,
3502 origin_head
, drc
->drc_force
, drc
->drc_owner
, tx
);
3504 dsl_dataset_rele(origin_head
, FTAG
);
3507 error
= dsl_dataset_snapshot_check_impl(origin_head
,
3508 drc
->drc_tosnap
, tx
, B_TRUE
, 1,
3509 drc
->drc_cred
, drc
->drc_proc
);
3510 dsl_dataset_rele(origin_head
, FTAG
);
3514 error
= dsl_destroy_head_check_impl(drc
->drc_ds
, 1);
3516 error
= dsl_dataset_snapshot_check_impl(drc
->drc_ds
,
3517 drc
->drc_tosnap
, tx
, B_TRUE
, 1,
3518 drc
->drc_cred
, drc
->drc_proc
);
3524 dmu_recv_end_sync(void *arg
, dmu_tx_t
*tx
)
3526 dmu_recv_cookie_t
*drc
= arg
;
3527 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3528 boolean_t encrypted
= drc
->drc_ds
->ds_dir
->dd_crypto_obj
!= 0;
3529 uint64_t newsnapobj
= 0;
3531 spa_history_log_internal_ds(drc
->drc_ds
, "finish receiving",
3532 tx
, "snap=%s", drc
->drc_tosnap
);
3533 drc
->drc_ds
->ds_objset
->os_raw_receive
= B_FALSE
;
3535 if (drc
->drc_heal
) {
3536 if (drc
->drc_keynvl
!= NULL
) {
3537 nvlist_free(drc
->drc_keynvl
);
3538 drc
->drc_keynvl
= NULL
;
3540 } else if (!drc
->drc_newfs
) {
3541 dsl_dataset_t
*origin_head
;
3543 VERIFY0(dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
,
3546 if (drc
->drc_force
) {
3548 * Destroy any snapshots of drc_tofs (origin_head)
3549 * after the origin (the snap before drc_ds).
3553 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3555 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3556 dsl_dataset_t
*snap
;
3557 VERIFY0(dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3559 ASSERT3P(snap
->ds_dir
, ==, origin_head
->ds_dir
);
3560 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3561 dsl_destroy_snapshot_sync_impl(snap
,
3563 dsl_dataset_rele(snap
, FTAG
);
3566 if (drc
->drc_keynvl
!= NULL
) {
3567 dsl_crypto_recv_raw_key_sync(drc
->drc_ds
,
3568 drc
->drc_keynvl
, tx
);
3569 nvlist_free(drc
->drc_keynvl
);
3570 drc
->drc_keynvl
= NULL
;
3573 VERIFY3P(drc
->drc_ds
->ds_prev
, ==,
3574 origin_head
->ds_prev
);
3576 dsl_dataset_clone_swap_sync_impl(drc
->drc_ds
,
3579 * The objset was evicted by dsl_dataset_clone_swap_sync_impl,
3580 * so drc_os is no longer valid.
3584 dsl_dataset_snapshot_sync_impl(origin_head
,
3585 drc
->drc_tosnap
, tx
);
3587 /* set snapshot's creation time and guid */
3588 dmu_buf_will_dirty(origin_head
->ds_prev
->ds_dbuf
, tx
);
3589 dsl_dataset_phys(origin_head
->ds_prev
)->ds_creation_time
=
3590 drc
->drc_drrb
->drr_creation_time
;
3591 dsl_dataset_phys(origin_head
->ds_prev
)->ds_guid
=
3592 drc
->drc_drrb
->drr_toguid
;
3593 dsl_dataset_phys(origin_head
->ds_prev
)->ds_flags
&=
3594 ~DS_FLAG_INCONSISTENT
;
3596 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
3597 dsl_dataset_phys(origin_head
)->ds_flags
&=
3598 ~DS_FLAG_INCONSISTENT
;
3601 dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3603 dsl_dataset_rele(origin_head
, FTAG
);
3604 dsl_destroy_head_sync_impl(drc
->drc_ds
, tx
);
3606 if (drc
->drc_owner
!= NULL
)
3607 VERIFY3P(origin_head
->ds_owner
, ==, drc
->drc_owner
);
3609 dsl_dataset_t
*ds
= drc
->drc_ds
;
3611 dsl_dataset_snapshot_sync_impl(ds
, drc
->drc_tosnap
, tx
);
3613 /* set snapshot's creation time and guid */
3614 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
3615 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_time
=
3616 drc
->drc_drrb
->drr_creation_time
;
3617 dsl_dataset_phys(ds
->ds_prev
)->ds_guid
=
3618 drc
->drc_drrb
->drr_toguid
;
3619 dsl_dataset_phys(ds
->ds_prev
)->ds_flags
&=
3620 ~DS_FLAG_INCONSISTENT
;
3622 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3623 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
3624 if (dsl_dataset_has_resume_receive_state(ds
)) {
3625 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3626 DS_FIELD_RESUME_FROMGUID
, tx
);
3627 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3628 DS_FIELD_RESUME_OBJECT
, tx
);
3629 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3630 DS_FIELD_RESUME_OFFSET
, tx
);
3631 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3632 DS_FIELD_RESUME_BYTES
, tx
);
3633 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3634 DS_FIELD_RESUME_TOGUID
, tx
);
3635 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3636 DS_FIELD_RESUME_TONAME
, tx
);
3637 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3638 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS
, tx
);
3641 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
;
3645 * If this is a raw receive, the crypt_keydata nvlist will include
3646 * a to_ivset_guid for us to set on the new snapshot. This value
3647 * will override the value generated by the snapshot code. However,
3648 * this value may not be present, because older implementations of
3649 * the raw send code did not include this value, and we are still
3650 * allowed to receive them if the zfs_disable_ivset_guid_check
3651 * tunable is set, in which case we will leave the newly-generated
3654 if (!drc
->drc_heal
&& drc
->drc_raw
&& drc
->drc_ivset_guid
!= 0) {
3655 dmu_object_zapify(dp
->dp_meta_objset
, newsnapobj
,
3656 DMU_OT_DSL_DATASET
, tx
);
3657 VERIFY0(zap_update(dp
->dp_meta_objset
, newsnapobj
,
3658 DS_FIELD_IVSET_GUID
, sizeof (uint64_t), 1,
3659 &drc
->drc_ivset_guid
, tx
));
3663 * Release the hold from dmu_recv_begin. This must be done before
3664 * we return to open context, so that when we free the dataset's dnode
3665 * we can evict its bonus buffer. Since the dataset may be destroyed
3666 * at this point (and therefore won't have a valid pointer to the spa)
3667 * we release the key mapping manually here while we do have a valid
3668 * pointer, if it exists.
3670 if (!drc
->drc_raw
&& encrypted
) {
3671 (void) spa_keystore_remove_mapping(dmu_tx_pool(tx
)->dp_spa
,
3672 drc
->drc_ds
->ds_object
, drc
->drc_ds
);
3674 dsl_dataset_disown(drc
->drc_ds
, 0, dmu_recv_tag
);
3678 static int dmu_recv_end_modified_blocks
= 3;
3681 dmu_recv_existing_end(dmu_recv_cookie_t
*drc
)
3685 * We will be destroying the ds; make sure its origin is unmounted if
3688 char name
[ZFS_MAX_DATASET_NAME_LEN
];
3689 dsl_dataset_name(drc
->drc_ds
, name
);
3690 zfs_destroy_unmount_origin(name
);
3693 return (dsl_sync_task(drc
->drc_tofs
,
3694 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3695 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
3699 dmu_recv_new_end(dmu_recv_cookie_t
*drc
)
3701 return (dsl_sync_task(drc
->drc_tofs
,
3702 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3703 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
3707 dmu_recv_end(dmu_recv_cookie_t
*drc
, void *owner
)
3711 drc
->drc_owner
= owner
;
3714 error
= dmu_recv_new_end(drc
);
3716 error
= dmu_recv_existing_end(drc
);
3719 dmu_recv_cleanup_ds(drc
);
3720 nvlist_free(drc
->drc_keynvl
);
3721 } else if (!drc
->drc_heal
) {
3722 if (drc
->drc_newfs
) {
3723 zvol_create_minor(drc
->drc_tofs
);
3725 char *snapname
= kmem_asprintf("%s@%s",
3726 drc
->drc_tofs
, drc
->drc_tosnap
);
3727 zvol_create_minor(snapname
);
3728 kmem_strfree(snapname
);
3734 * Return TRUE if this objset is currently being received into.
3737 dmu_objset_is_receiving(objset_t
*os
)
3739 return (os
->os_dsl_dataset
!= NULL
&&
3740 os
->os_dsl_dataset
->ds_owner
== dmu_recv_tag
);
3743 ZFS_MODULE_PARAM(zfs_recv
, zfs_recv_
, queue_length
, UINT
, ZMOD_RW
,
3744 "Maximum receive queue length");
3746 ZFS_MODULE_PARAM(zfs_recv
, zfs_recv_
, queue_ff
, UINT
, ZMOD_RW
,
3747 "Receive queue fill fraction");
3749 ZFS_MODULE_PARAM(zfs_recv
, zfs_recv_
, write_batch_size
, UINT
, ZMOD_RW
,
3750 "Maximum amount of writes to batch into one transaction");
3752 ZFS_MODULE_PARAM(zfs_recv
, zfs_recv_
, best_effort_corrective
, INT
, ZMOD_RW
,
3753 "Ignore errors during corrective receive");