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";
85 static int receive_read_payload_and_next_header(dmu_recv_cookie_t
*ra
, int len
,
88 struct receive_record_arg
{
89 dmu_replay_record_t header
;
90 void *payload
; /* Pointer to a buffer containing the payload */
92 * If the record is a WRITE or SPILL, pointer to the abd containing the
97 uint64_t bytes_read
; /* bytes read from stream when record created */
98 boolean_t eos_marker
; /* Marks the end of the stream */
102 struct receive_writer_arg
{
108 * These three members are used to signal to the main thread when
119 boolean_t raw
; /* DMU_BACKUP_FEATURE_RAW set */
120 boolean_t spill
; /* DRR_FLAG_SPILL_BLOCK set */
121 boolean_t full
; /* this is a full send stream */
122 uint64_t last_object
;
123 uint64_t last_offset
;
124 uint64_t max_object
; /* highest object ID referenced in stream */
125 uint64_t bytes_read
; /* bytes read when current record created */
129 /* Encryption parameters for the last received DRR_OBJECT_RANGE */
130 boolean_t or_crypt_params_present
;
131 uint64_t or_firstobj
;
132 uint64_t or_numslots
;
133 uint8_t or_salt
[ZIO_DATA_SALT_LEN
];
134 uint8_t or_iv
[ZIO_DATA_IV_LEN
];
135 uint8_t or_mac
[ZIO_DATA_MAC_LEN
];
136 boolean_t or_byteorder
;
139 /* Keep track of DRR_FREEOBJECTS right after DRR_OBJECT_RANGE */
140 or_need_sync_t or_need_sync
;
143 typedef struct dmu_recv_begin_arg
{
144 const char *drba_origin
;
145 dmu_recv_cookie_t
*drba_cookie
;
148 dsl_crypto_params_t
*drba_dcp
;
149 } dmu_recv_begin_arg_t
;
152 byteswap_record(dmu_replay_record_t
*drr
)
154 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
155 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
156 drr
->drr_type
= BSWAP_32(drr
->drr_type
);
157 drr
->drr_payloadlen
= BSWAP_32(drr
->drr_payloadlen
);
159 switch (drr
->drr_type
) {
161 DO64(drr_begin
.drr_magic
);
162 DO64(drr_begin
.drr_versioninfo
);
163 DO64(drr_begin
.drr_creation_time
);
164 DO32(drr_begin
.drr_type
);
165 DO32(drr_begin
.drr_flags
);
166 DO64(drr_begin
.drr_toguid
);
167 DO64(drr_begin
.drr_fromguid
);
170 DO64(drr_object
.drr_object
);
171 DO32(drr_object
.drr_type
);
172 DO32(drr_object
.drr_bonustype
);
173 DO32(drr_object
.drr_blksz
);
174 DO32(drr_object
.drr_bonuslen
);
175 DO32(drr_object
.drr_raw_bonuslen
);
176 DO64(drr_object
.drr_toguid
);
177 DO64(drr_object
.drr_maxblkid
);
179 case DRR_FREEOBJECTS
:
180 DO64(drr_freeobjects
.drr_firstobj
);
181 DO64(drr_freeobjects
.drr_numobjs
);
182 DO64(drr_freeobjects
.drr_toguid
);
185 DO64(drr_write
.drr_object
);
186 DO32(drr_write
.drr_type
);
187 DO64(drr_write
.drr_offset
);
188 DO64(drr_write
.drr_logical_size
);
189 DO64(drr_write
.drr_toguid
);
190 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write
.drr_key
.ddk_cksum
);
191 DO64(drr_write
.drr_key
.ddk_prop
);
192 DO64(drr_write
.drr_compressed_size
);
194 case DRR_WRITE_EMBEDDED
:
195 DO64(drr_write_embedded
.drr_object
);
196 DO64(drr_write_embedded
.drr_offset
);
197 DO64(drr_write_embedded
.drr_length
);
198 DO64(drr_write_embedded
.drr_toguid
);
199 DO32(drr_write_embedded
.drr_lsize
);
200 DO32(drr_write_embedded
.drr_psize
);
203 DO64(drr_free
.drr_object
);
204 DO64(drr_free
.drr_offset
);
205 DO64(drr_free
.drr_length
);
206 DO64(drr_free
.drr_toguid
);
209 DO64(drr_spill
.drr_object
);
210 DO64(drr_spill
.drr_length
);
211 DO64(drr_spill
.drr_toguid
);
212 DO64(drr_spill
.drr_compressed_size
);
213 DO32(drr_spill
.drr_type
);
215 case DRR_OBJECT_RANGE
:
216 DO64(drr_object_range
.drr_firstobj
);
217 DO64(drr_object_range
.drr_numslots
);
218 DO64(drr_object_range
.drr_toguid
);
221 DO64(drr_redact
.drr_object
);
222 DO64(drr_redact
.drr_offset
);
223 DO64(drr_redact
.drr_length
);
224 DO64(drr_redact
.drr_toguid
);
227 DO64(drr_end
.drr_toguid
);
228 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_end
.drr_checksum
);
234 if (drr
->drr_type
!= DRR_BEGIN
) {
235 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_checksum
.drr_checksum
);
243 redact_snaps_contains(uint64_t *snaps
, uint64_t num_snaps
, uint64_t guid
)
245 for (int i
= 0; i
< num_snaps
; i
++) {
246 if (snaps
[i
] == guid
)
253 * Check that the new stream we're trying to receive is redacted with respect to
254 * a subset of the snapshots that the origin was redacted with respect to. For
255 * the reasons behind this, see the man page on redacted zfs sends and receives.
258 compatible_redact_snaps(uint64_t *origin_snaps
, uint64_t origin_num_snaps
,
259 uint64_t *redact_snaps
, uint64_t num_redact_snaps
)
262 * Short circuit the comparison; if we are redacted with respect to
263 * more snapshots than the origin, we can't be redacted with respect
266 if (num_redact_snaps
> origin_num_snaps
) {
270 for (int i
= 0; i
< num_redact_snaps
; i
++) {
271 if (!redact_snaps_contains(origin_snaps
, origin_num_snaps
,
280 redact_check(dmu_recv_begin_arg_t
*drba
, dsl_dataset_t
*origin
)
282 uint64_t *origin_snaps
;
283 uint64_t origin_num_snaps
;
284 dmu_recv_cookie_t
*drc
= drba
->drba_cookie
;
285 struct drr_begin
*drrb
= drc
->drc_drrb
;
286 int featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
288 boolean_t ret
= B_TRUE
;
289 uint64_t *redact_snaps
;
290 uint_t numredactsnaps
;
293 * If this is a full send stream, we're safe no matter what.
295 if (drrb
->drr_fromguid
== 0)
298 VERIFY(dsl_dataset_get_uint64_array_feature(origin
,
299 SPA_FEATURE_REDACTED_DATASETS
, &origin_num_snaps
, &origin_snaps
));
301 if (nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
302 BEGINNV_REDACT_FROM_SNAPS
, &redact_snaps
, &numredactsnaps
) ==
305 * If the send stream was sent from the redaction bookmark or
306 * the redacted version of the dataset, then we're safe. Verify
307 * that this is from the a compatible redaction bookmark or
310 if (!compatible_redact_snaps(origin_snaps
, origin_num_snaps
,
311 redact_snaps
, numredactsnaps
)) {
314 } else if (featureflags
& DMU_BACKUP_FEATURE_REDACTED
) {
316 * If the stream is redacted, it must be redacted with respect
317 * to a subset of what the origin is redacted with respect to.
318 * See case number 2 in the zfs man page section on redacted zfs
321 err
= nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
322 BEGINNV_REDACT_SNAPS
, &redact_snaps
, &numredactsnaps
);
324 if (err
!= 0 || !compatible_redact_snaps(origin_snaps
,
325 origin_num_snaps
, redact_snaps
, numredactsnaps
)) {
328 } else if (!redact_snaps_contains(origin_snaps
, origin_num_snaps
,
331 * If the stream isn't redacted but the origin is, this must be
332 * one of the snapshots the origin is redacted with respect to.
333 * See case number 1 in the zfs man page section on redacted zfs
345 * If we previously received a stream with --large-block, we don't support
346 * receiving an incremental on top of it without --large-block. This avoids
347 * forcing a read-modify-write or trying to re-aggregate a string of WRITE
351 recv_check_large_blocks(dsl_dataset_t
*ds
, uint64_t featureflags
)
353 if (dsl_dataset_feature_is_active(ds
, SPA_FEATURE_LARGE_BLOCKS
) &&
354 !(featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
))
355 return (SET_ERROR(ZFS_ERR_STREAM_LARGE_BLOCK_MISMATCH
));
360 recv_begin_check_existing_impl(dmu_recv_begin_arg_t
*drba
, dsl_dataset_t
*ds
,
361 uint64_t fromguid
, uint64_t featureflags
)
367 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
368 boolean_t encrypted
= ds
->ds_dir
->dd_crypto_obj
!= 0;
369 boolean_t raw
= (featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0;
370 boolean_t embed
= (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) != 0;
372 /* Temporary clone name must not exist. */
373 error
= zap_lookup(dp
->dp_meta_objset
,
374 dsl_dir_phys(ds
->ds_dir
)->dd_child_dir_zapobj
, recv_clone_name
,
377 return (error
== 0 ? SET_ERROR(EBUSY
) : error
);
379 /* Resume state must not be set. */
380 if (dsl_dataset_has_resume_receive_state(ds
))
381 return (SET_ERROR(EBUSY
));
383 /* New snapshot name must not exist if we're not healing it. */
384 error
= zap_lookup(dp
->dp_meta_objset
,
385 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
386 drba
->drba_cookie
->drc_tosnap
, 8, 1, &obj
);
387 if (drba
->drba_cookie
->drc_heal
) {
390 } else if (error
!= ENOENT
) {
391 return (error
== 0 ? SET_ERROR(EEXIST
) : error
);
394 /* Must not have children if receiving a ZVOL. */
395 error
= zap_count(dp
->dp_meta_objset
,
396 dsl_dir_phys(ds
->ds_dir
)->dd_child_dir_zapobj
, &children
);
399 if (drba
->drba_cookie
->drc_drrb
->drr_type
!= DMU_OST_ZFS
&&
401 return (SET_ERROR(ZFS_ERR_WRONG_PARENT
));
404 * Check snapshot limit before receiving. We'll recheck again at the
405 * end, but might as well abort before receiving if we're already over
408 * Note that we do not check the file system limit with
409 * dsl_dir_fscount_check because the temporary %clones don't count
410 * against that limit.
412 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1, ZFS_PROP_SNAPSHOT_LIMIT
,
413 NULL
, drba
->drba_cred
, drba
->drba_proc
);
417 if (drba
->drba_cookie
->drc_heal
) {
418 /* Encryption is incompatible with embedded data. */
419 if (encrypted
&& embed
)
420 return (SET_ERROR(EINVAL
));
422 /* Healing is not supported when in 'force' mode. */
423 if (drba
->drba_cookie
->drc_force
)
424 return (SET_ERROR(EINVAL
));
426 /* Must have keys loaded if doing encrypted non-raw recv. */
427 if (encrypted
&& !raw
) {
428 if (spa_keystore_lookup_key(dp
->dp_spa
, ds
->ds_object
,
430 return (SET_ERROR(EACCES
));
433 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &snap
);
438 * When not doing best effort corrective recv healing can only
439 * be done if the send stream is for the same snapshot as the
440 * one we are trying to heal.
442 if (zfs_recv_best_effort_corrective
== 0 &&
443 drba
->drba_cookie
->drc_drrb
->drr_toguid
!=
444 dsl_dataset_phys(snap
)->ds_guid
) {
445 dsl_dataset_rele(snap
, FTAG
);
446 return (SET_ERROR(ENOTSUP
));
448 dsl_dataset_rele(snap
, FTAG
);
449 } else if (fromguid
!= 0) {
450 /* Sanity check the incremental recv */
451 uint64_t obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
453 /* Can't perform a raw receive on top of a non-raw receive */
454 if (!encrypted
&& raw
)
455 return (SET_ERROR(EINVAL
));
457 /* Encryption is incompatible with embedded data */
458 if (encrypted
&& embed
)
459 return (SET_ERROR(EINVAL
));
461 /* Find snapshot in this dir that matches fromguid. */
463 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
466 return (SET_ERROR(ENODEV
));
467 if (snap
->ds_dir
!= ds
->ds_dir
) {
468 dsl_dataset_rele(snap
, FTAG
);
469 return (SET_ERROR(ENODEV
));
471 if (dsl_dataset_phys(snap
)->ds_guid
== fromguid
)
473 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
474 dsl_dataset_rele(snap
, FTAG
);
477 return (SET_ERROR(ENODEV
));
479 if (drba
->drba_cookie
->drc_force
) {
480 drba
->drba_cookie
->drc_fromsnapobj
= obj
;
483 * If we are not forcing, there must be no
484 * changes since fromsnap. Raw sends have an
485 * additional constraint that requires that
486 * no "noop" snapshots exist between fromsnap
487 * and tosnap for the IVset checking code to
490 if (dsl_dataset_modified_since_snap(ds
, snap
) ||
492 dsl_dataset_phys(ds
)->ds_prev_snap_obj
!=
494 dsl_dataset_rele(snap
, FTAG
);
495 return (SET_ERROR(ETXTBSY
));
497 drba
->drba_cookie
->drc_fromsnapobj
=
498 ds
->ds_prev
->ds_object
;
501 if (dsl_dataset_feature_is_active(snap
,
502 SPA_FEATURE_REDACTED_DATASETS
) && !redact_check(drba
,
504 dsl_dataset_rele(snap
, FTAG
);
505 return (SET_ERROR(EINVAL
));
508 error
= recv_check_large_blocks(snap
, featureflags
);
510 dsl_dataset_rele(snap
, FTAG
);
514 dsl_dataset_rele(snap
, FTAG
);
516 /* If full and not healing then must be forced. */
517 if (!drba
->drba_cookie
->drc_force
)
518 return (SET_ERROR(EEXIST
));
521 * We don't support using zfs recv -F to blow away
522 * encrypted filesystems. This would require the
523 * dsl dir to point to the old encryption key and
524 * the new one at the same time during the receive.
526 if ((!encrypted
&& raw
) || encrypted
)
527 return (SET_ERROR(EINVAL
));
530 * Perform the same encryption checks we would if
531 * we were creating a new dataset from scratch.
534 boolean_t will_encrypt
;
536 error
= dmu_objset_create_crypt_check(
537 ds
->ds_dir
->dd_parent
, drba
->drba_dcp
,
542 if (will_encrypt
&& embed
)
543 return (SET_ERROR(EINVAL
));
551 * Check that any feature flags used in the data stream we're receiving are
552 * supported by the pool we are receiving into.
554 * Note that some of the features we explicitly check here have additional
555 * (implicit) features they depend on, but those dependencies are enforced
556 * through the zfeature_register() calls declaring the features that we
560 recv_begin_check_feature_flags_impl(uint64_t featureflags
, spa_t
*spa
)
563 * Check if there are any unsupported feature flags.
565 if (!DMU_STREAM_SUPPORTED(featureflags
)) {
566 return (SET_ERROR(ZFS_ERR_UNKNOWN_SEND_STREAM_FEATURE
));
569 /* Verify pool version supports SA if SA_SPILL feature set */
570 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
571 spa_version(spa
) < SPA_VERSION_SA
)
572 return (SET_ERROR(ENOTSUP
));
575 * LZ4 compressed, ZSTD compressed, embedded, mooched, large blocks,
576 * and large_dnodes in the stream can only be used if those pool
577 * features are enabled because we don't attempt to decompress /
578 * un-embed / un-mooch / split up the blocks / dnodes during the
581 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
582 !spa_feature_is_enabled(spa
, SPA_FEATURE_LZ4_COMPRESS
))
583 return (SET_ERROR(ENOTSUP
));
584 if ((featureflags
& DMU_BACKUP_FEATURE_ZSTD
) &&
585 !spa_feature_is_enabled(spa
, SPA_FEATURE_ZSTD_COMPRESS
))
586 return (SET_ERROR(ENOTSUP
));
587 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
588 !spa_feature_is_enabled(spa
, SPA_FEATURE_EMBEDDED_DATA
))
589 return (SET_ERROR(ENOTSUP
));
590 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
591 !spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
))
592 return (SET_ERROR(ENOTSUP
));
593 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_DNODE
) &&
594 !spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
))
595 return (SET_ERROR(ENOTSUP
));
598 * Receiving redacted streams requires that redacted datasets are
601 if ((featureflags
& DMU_BACKUP_FEATURE_REDACTED
) &&
602 !spa_feature_is_enabled(spa
, SPA_FEATURE_REDACTED_DATASETS
))
603 return (SET_ERROR(ENOTSUP
));
609 dmu_recv_begin_check(void *arg
, dmu_tx_t
*tx
)
611 dmu_recv_begin_arg_t
*drba
= arg
;
612 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
613 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
614 uint64_t fromguid
= drrb
->drr_fromguid
;
615 int flags
= drrb
->drr_flags
;
616 ds_hold_flags_t dsflags
= DS_HOLD_FLAG_NONE
;
618 uint64_t featureflags
= drba
->drba_cookie
->drc_featureflags
;
620 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
622 /* already checked */
623 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
624 ASSERT(!(featureflags
& DMU_BACKUP_FEATURE_RESUMING
));
626 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
627 DMU_COMPOUNDSTREAM
||
628 drrb
->drr_type
>= DMU_OST_NUMTYPES
||
629 ((flags
& DRR_FLAG_CLONE
) && drba
->drba_origin
== NULL
))
630 return (SET_ERROR(EINVAL
));
632 error
= recv_begin_check_feature_flags_impl(featureflags
, dp
->dp_spa
);
636 /* Resumable receives require extensible datasets */
637 if (drba
->drba_cookie
->drc_resumable
&&
638 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EXTENSIBLE_DATASET
))
639 return (SET_ERROR(ENOTSUP
));
641 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
642 /* raw receives require the encryption feature */
643 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_ENCRYPTION
))
644 return (SET_ERROR(ENOTSUP
));
646 /* embedded data is incompatible with encryption and raw recv */
647 if (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)
648 return (SET_ERROR(EINVAL
));
650 /* raw receives require spill block allocation flag */
651 if (!(flags
& DRR_FLAG_SPILL_BLOCK
))
652 return (SET_ERROR(ZFS_ERR_SPILL_BLOCK_FLAG_MISSING
));
655 * We support unencrypted datasets below encrypted ones now,
656 * so add the DS_HOLD_FLAG_DECRYPT flag only if we are dealing
657 * with a dataset we may encrypt.
659 if (drba
->drba_dcp
== NULL
||
660 drba
->drba_dcp
->cp_crypt
!= ZIO_CRYPT_OFF
) {
661 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
665 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
667 /* target fs already exists; recv into temp clone */
669 /* Can't recv a clone into an existing fs */
670 if (flags
& DRR_FLAG_CLONE
|| drba
->drba_origin
) {
671 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
672 return (SET_ERROR(EINVAL
));
675 error
= recv_begin_check_existing_impl(drba
, ds
, fromguid
,
677 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
678 } else if (error
== ENOENT
) {
679 /* target fs does not exist; must be a full backup or clone */
680 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
683 /* healing recv must be done "into" an existing snapshot */
684 if (drba
->drba_cookie
->drc_heal
== B_TRUE
)
685 return (SET_ERROR(ENOTSUP
));
688 * If it's a non-clone incremental, we are missing the
689 * target fs, so fail the recv.
691 if (fromguid
!= 0 && !((flags
& DRR_FLAG_CLONE
) ||
693 return (SET_ERROR(ENOENT
));
696 * If we're receiving a full send as a clone, and it doesn't
697 * contain all the necessary free records and freeobject
698 * records, reject it.
700 if (fromguid
== 0 && drba
->drba_origin
!= NULL
&&
701 !(flags
& DRR_FLAG_FREERECORDS
))
702 return (SET_ERROR(EINVAL
));
704 /* Open the parent of tofs */
705 ASSERT3U(strlen(tofs
), <, sizeof (buf
));
706 (void) strlcpy(buf
, tofs
, strrchr(tofs
, '/') - tofs
+ 1);
707 error
= dsl_dataset_hold(dp
, buf
, FTAG
, &ds
);
711 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0 &&
712 drba
->drba_origin
== NULL
) {
713 boolean_t will_encrypt
;
716 * Check that we aren't breaking any encryption rules
717 * and that we have all the parameters we need to
718 * create an encrypted dataset if necessary. If we are
719 * making an encrypted dataset the stream can't have
722 error
= dmu_objset_create_crypt_check(ds
->ds_dir
,
723 drba
->drba_dcp
, &will_encrypt
);
725 dsl_dataset_rele(ds
, FTAG
);
730 (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)) {
731 dsl_dataset_rele(ds
, FTAG
);
732 return (SET_ERROR(EINVAL
));
737 * Check filesystem and snapshot limits before receiving. We'll
738 * recheck snapshot limits again at the end (we create the
739 * filesystems and increment those counts during begin_sync).
741 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
742 ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
743 drba
->drba_cred
, drba
->drba_proc
);
745 dsl_dataset_rele(ds
, FTAG
);
749 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
750 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
,
751 drba
->drba_cred
, drba
->drba_proc
);
753 dsl_dataset_rele(ds
, FTAG
);
757 /* can't recv below anything but filesystems (eg. no ZVOLs) */
758 error
= dmu_objset_from_ds(ds
, &os
);
760 dsl_dataset_rele(ds
, FTAG
);
763 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
764 dsl_dataset_rele(ds
, FTAG
);
765 return (SET_ERROR(ZFS_ERR_WRONG_PARENT
));
768 if (drba
->drba_origin
!= NULL
) {
769 dsl_dataset_t
*origin
;
770 error
= dsl_dataset_hold_flags(dp
, drba
->drba_origin
,
771 dsflags
, FTAG
, &origin
);
773 dsl_dataset_rele(ds
, FTAG
);
776 if (!origin
->ds_is_snapshot
) {
777 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
778 dsl_dataset_rele(ds
, FTAG
);
779 return (SET_ERROR(EINVAL
));
781 if (dsl_dataset_phys(origin
)->ds_guid
!= fromguid
&&
783 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
784 dsl_dataset_rele(ds
, FTAG
);
785 return (SET_ERROR(ENODEV
));
788 if (origin
->ds_dir
->dd_crypto_obj
!= 0 &&
789 (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)) {
790 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
791 dsl_dataset_rele(ds
, FTAG
);
792 return (SET_ERROR(EINVAL
));
796 * If the origin is redacted we need to verify that this
797 * send stream can safely be received on top of the
800 if (dsl_dataset_feature_is_active(origin
,
801 SPA_FEATURE_REDACTED_DATASETS
)) {
802 if (!redact_check(drba
, origin
)) {
803 dsl_dataset_rele_flags(origin
, dsflags
,
805 dsl_dataset_rele_flags(ds
, dsflags
,
807 return (SET_ERROR(EINVAL
));
811 error
= recv_check_large_blocks(ds
, featureflags
);
813 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
814 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
818 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
821 dsl_dataset_rele(ds
, FTAG
);
828 dmu_recv_begin_sync(void *arg
, dmu_tx_t
*tx
)
830 dmu_recv_begin_arg_t
*drba
= arg
;
831 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
832 objset_t
*mos
= dp
->dp_meta_objset
;
833 dmu_recv_cookie_t
*drc
= drba
->drba_cookie
;
834 struct drr_begin
*drrb
= drc
->drc_drrb
;
835 const char *tofs
= drc
->drc_tofs
;
836 uint64_t featureflags
= drc
->drc_featureflags
;
837 dsl_dataset_t
*ds
, *newds
;
840 ds_hold_flags_t dsflags
= DS_HOLD_FLAG_NONE
;
842 uint64_t crflags
= 0;
843 dsl_crypto_params_t dummy_dcp
= { 0 };
844 dsl_crypto_params_t
*dcp
= drba
->drba_dcp
;
846 if (drrb
->drr_flags
& DRR_FLAG_CI_DATA
)
847 crflags
|= DS_FLAG_CI_DATASET
;
849 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0)
850 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
853 * Raw, non-incremental recvs always use a dummy dcp with
854 * the raw cmd set. Raw incremental recvs do not use a dcp
855 * since the encryption parameters are already set in stone.
857 if (dcp
== NULL
&& drrb
->drr_fromguid
== 0 &&
858 drba
->drba_origin
== NULL
) {
859 ASSERT3P(dcp
, ==, NULL
);
862 if (featureflags
& DMU_BACKUP_FEATURE_RAW
)
863 dcp
->cp_cmd
= DCP_CMD_RAW_RECV
;
866 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
868 /* Create temporary clone unless we're doing corrective recv */
869 dsl_dataset_t
*snap
= NULL
;
871 if (drba
->drba_cookie
->drc_fromsnapobj
!= 0) {
872 VERIFY0(dsl_dataset_hold_obj(dp
,
873 drba
->drba_cookie
->drc_fromsnapobj
, FTAG
, &snap
));
874 ASSERT3P(dcp
, ==, NULL
);
877 /* When healing we want to use the provided snapshot */
878 VERIFY0(dsl_dataset_snap_lookup(ds
, drc
->drc_tosnap
,
881 dsobj
= dsl_dataset_create_sync(ds
->ds_dir
,
882 recv_clone_name
, snap
, crflags
, drba
->drba_cred
,
885 if (drba
->drba_cookie
->drc_fromsnapobj
!= 0)
886 dsl_dataset_rele(snap
, FTAG
);
887 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
891 dsl_dataset_t
*origin
= NULL
;
893 VERIFY0(dsl_dir_hold(dp
, tofs
, FTAG
, &dd
, &tail
));
895 if (drba
->drba_origin
!= NULL
) {
896 VERIFY0(dsl_dataset_hold(dp
, drba
->drba_origin
,
898 ASSERT3P(dcp
, ==, NULL
);
901 /* Create new dataset. */
902 dsobj
= dsl_dataset_create_sync(dd
, strrchr(tofs
, '/') + 1,
903 origin
, crflags
, drba
->drba_cred
, dcp
, tx
);
905 dsl_dataset_rele(origin
, FTAG
);
906 dsl_dir_rele(dd
, FTAG
);
907 drc
->drc_newfs
= B_TRUE
;
909 VERIFY0(dsl_dataset_own_obj_force(dp
, dsobj
, dsflags
, dmu_recv_tag
,
911 if (dsl_dataset_feature_is_active(newds
,
912 SPA_FEATURE_REDACTED_DATASETS
)) {
914 * If the origin dataset is redacted, the child will be redacted
915 * when we create it. We clear the new dataset's
916 * redaction info; if it should be redacted, we'll fill
917 * in its information later.
919 dsl_dataset_deactivate_feature(newds
,
920 SPA_FEATURE_REDACTED_DATASETS
, tx
);
922 VERIFY0(dmu_objset_from_ds(newds
, &os
));
924 if (drc
->drc_resumable
) {
925 dsl_dataset_zapify(newds
, tx
);
926 if (drrb
->drr_fromguid
!= 0) {
927 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_FROMGUID
,
928 8, 1, &drrb
->drr_fromguid
, tx
));
930 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TOGUID
,
931 8, 1, &drrb
->drr_toguid
, tx
));
932 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TONAME
,
933 1, strlen(drrb
->drr_toname
) + 1, drrb
->drr_toname
, tx
));
936 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OBJECT
,
938 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OFFSET
,
940 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_BYTES
,
942 if (featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) {
943 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_LARGEBLOCK
,
946 if (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) {
947 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_EMBEDOK
,
950 if (featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
) {
951 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_COMPRESSOK
,
954 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
955 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_RAWOK
,
959 uint64_t *redact_snaps
;
960 uint_t numredactsnaps
;
961 if (nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
962 BEGINNV_REDACT_FROM_SNAPS
, &redact_snaps
,
963 &numredactsnaps
) == 0) {
964 VERIFY0(zap_add(mos
, dsobj
,
965 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS
,
966 sizeof (*redact_snaps
), numredactsnaps
,
972 * Usually the os->os_encrypted value is tied to the presence of a
973 * DSL Crypto Key object in the dd. However, that will not be received
974 * until dmu_recv_stream(), so we set the value manually for now.
976 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
977 os
->os_encrypted
= B_TRUE
;
978 drba
->drba_cookie
->drc_raw
= B_TRUE
;
981 if (featureflags
& DMU_BACKUP_FEATURE_REDACTED
) {
982 uint64_t *redact_snaps
;
983 uint_t numredactsnaps
;
984 VERIFY0(nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
985 BEGINNV_REDACT_SNAPS
, &redact_snaps
, &numredactsnaps
));
986 dsl_dataset_activate_redaction(newds
, redact_snaps
,
990 dmu_buf_will_dirty(newds
->ds_dbuf
, tx
);
991 dsl_dataset_phys(newds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
994 * If we actually created a non-clone, we need to create the objset
995 * in our new dataset. If this is a raw send we postpone this until
996 * dmu_recv_stream() so that we can allocate the metadnode with the
997 * properties from the DRR_BEGIN payload.
999 rrw_enter(&newds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1000 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds
)) &&
1001 (featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0 &&
1003 (void) dmu_objset_create_impl(dp
->dp_spa
,
1004 newds
, dsl_dataset_get_blkptr(newds
), drrb
->drr_type
, tx
);
1006 rrw_exit(&newds
->ds_bp_rwlock
, FTAG
);
1008 drba
->drba_cookie
->drc_ds
= newds
;
1009 drba
->drba_cookie
->drc_os
= os
;
1011 spa_history_log_internal_ds(newds
, "receive", tx
, " ");
1015 dmu_recv_resume_begin_check(void *arg
, dmu_tx_t
*tx
)
1017 dmu_recv_begin_arg_t
*drba
= arg
;
1018 dmu_recv_cookie_t
*drc
= drba
->drba_cookie
;
1019 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1020 struct drr_begin
*drrb
= drc
->drc_drrb
;
1022 ds_hold_flags_t dsflags
= DS_HOLD_FLAG_NONE
;
1024 const char *tofs
= drc
->drc_tofs
;
1026 /* already checked */
1027 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1028 ASSERT(drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RESUMING
);
1030 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1031 DMU_COMPOUNDSTREAM
||
1032 drrb
->drr_type
>= DMU_OST_NUMTYPES
)
1033 return (SET_ERROR(EINVAL
));
1036 * This is mostly a sanity check since we should have already done these
1037 * checks during a previous attempt to receive the data.
1039 error
= recv_begin_check_feature_flags_impl(drc
->drc_featureflags
,
1044 /* 6 extra bytes for /%recv */
1045 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1047 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
1048 tofs
, recv_clone_name
);
1050 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1051 /* raw receives require spill block allocation flag */
1052 if (!(drrb
->drr_flags
& DRR_FLAG_SPILL_BLOCK
))
1053 return (SET_ERROR(ZFS_ERR_SPILL_BLOCK_FLAG_MISSING
));
1055 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1058 boolean_t recvexist
= B_TRUE
;
1059 if (dsl_dataset_hold_flags(dp
, recvname
, dsflags
, FTAG
, &ds
) != 0) {
1060 /* %recv does not exist; continue in tofs */
1061 recvexist
= B_FALSE
;
1062 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1068 * Resume of full/newfs recv on existing dataset should be done with
1071 if (recvexist
&& drrb
->drr_fromguid
== 0 && !drc
->drc_force
) {
1072 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1073 return (SET_ERROR(ZFS_ERR_RESUME_EXISTS
));
1076 /* check that ds is marked inconsistent */
1077 if (!DS_IS_INCONSISTENT(ds
)) {
1078 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1079 return (SET_ERROR(EINVAL
));
1082 /* check that there is resuming data, and that the toguid matches */
1083 if (!dsl_dataset_is_zapified(ds
)) {
1084 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1085 return (SET_ERROR(EINVAL
));
1088 error
= zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1089 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
);
1090 if (error
!= 0 || drrb
->drr_toguid
!= val
) {
1091 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1092 return (SET_ERROR(EINVAL
));
1096 * Check if the receive is still running. If so, it will be owned.
1097 * Note that nothing else can own the dataset (e.g. after the receive
1098 * fails) because it will be marked inconsistent.
1100 if (dsl_dataset_has_owner(ds
)) {
1101 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1102 return (SET_ERROR(EBUSY
));
1105 /* There should not be any snapshots of this fs yet. */
1106 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
->ds_dir
== ds
->ds_dir
) {
1107 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1108 return (SET_ERROR(EINVAL
));
1112 * Note: resume point will be checked when we process the first WRITE
1116 /* check that the origin matches */
1118 (void) zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1119 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
);
1120 if (drrb
->drr_fromguid
!= val
) {
1121 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1122 return (SET_ERROR(EINVAL
));
1125 if (ds
->ds_prev
!= NULL
&& drrb
->drr_fromguid
!= 0)
1126 drc
->drc_fromsnapobj
= ds
->ds_prev
->ds_object
;
1129 * If we're resuming, and the send is redacted, then the original send
1130 * must have been redacted, and must have been redacted with respect to
1131 * the same snapshots.
1133 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_REDACTED
) {
1134 uint64_t num_ds_redact_snaps
;
1135 uint64_t *ds_redact_snaps
;
1137 uint_t num_stream_redact_snaps
;
1138 uint64_t *stream_redact_snaps
;
1140 if (nvlist_lookup_uint64_array(drc
->drc_begin_nvl
,
1141 BEGINNV_REDACT_SNAPS
, &stream_redact_snaps
,
1142 &num_stream_redact_snaps
) != 0) {
1143 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1144 return (SET_ERROR(EINVAL
));
1147 if (!dsl_dataset_get_uint64_array_feature(ds
,
1148 SPA_FEATURE_REDACTED_DATASETS
, &num_ds_redact_snaps
,
1149 &ds_redact_snaps
)) {
1150 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1151 return (SET_ERROR(EINVAL
));
1154 for (int i
= 0; i
< num_ds_redact_snaps
; i
++) {
1155 if (!redact_snaps_contains(ds_redact_snaps
,
1156 num_ds_redact_snaps
, stream_redact_snaps
[i
])) {
1157 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1158 return (SET_ERROR(EINVAL
));
1163 error
= recv_check_large_blocks(ds
, drc
->drc_featureflags
);
1165 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1169 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1174 dmu_recv_resume_begin_sync(void *arg
, dmu_tx_t
*tx
)
1176 dmu_recv_begin_arg_t
*drba
= arg
;
1177 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1178 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1179 uint64_t featureflags
= drba
->drba_cookie
->drc_featureflags
;
1181 ds_hold_flags_t dsflags
= DS_HOLD_FLAG_NONE
;
1182 /* 6 extra bytes for /%recv */
1183 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1185 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s", tofs
,
1188 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1189 drba
->drba_cookie
->drc_raw
= B_TRUE
;
1191 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1194 if (dsl_dataset_own_force(dp
, recvname
, dsflags
, dmu_recv_tag
, &ds
)
1196 /* %recv does not exist; continue in tofs */
1197 VERIFY0(dsl_dataset_own_force(dp
, tofs
, dsflags
, dmu_recv_tag
,
1199 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
1202 ASSERT(DS_IS_INCONSISTENT(ds
));
1203 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1204 ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds
)) ||
1205 drba
->drba_cookie
->drc_raw
);
1206 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
1208 drba
->drba_cookie
->drc_ds
= ds
;
1209 VERIFY0(dmu_objset_from_ds(ds
, &drba
->drba_cookie
->drc_os
));
1210 drba
->drba_cookie
->drc_should_save
= B_TRUE
;
1212 spa_history_log_internal_ds(ds
, "resume receive", tx
, " ");
1216 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1217 * succeeds; otherwise we will leak the holds on the datasets.
1220 dmu_recv_begin(const char *tofs
, const char *tosnap
,
1221 dmu_replay_record_t
*drr_begin
, boolean_t force
, boolean_t heal
,
1222 boolean_t resumable
, nvlist_t
*localprops
, nvlist_t
*hidden_args
,
1223 const char *origin
, dmu_recv_cookie_t
*drc
, zfs_file_t
*fp
,
1226 dmu_recv_begin_arg_t drba
= { 0 };
1229 memset(drc
, 0, sizeof (dmu_recv_cookie_t
));
1230 drc
->drc_drr_begin
= drr_begin
;
1231 drc
->drc_drrb
= &drr_begin
->drr_u
.drr_begin
;
1232 drc
->drc_tosnap
= tosnap
;
1233 drc
->drc_tofs
= tofs
;
1234 drc
->drc_force
= force
;
1235 drc
->drc_heal
= heal
;
1236 drc
->drc_resumable
= resumable
;
1237 drc
->drc_cred
= CRED();
1238 drc
->drc_proc
= curproc
;
1239 drc
->drc_clone
= (origin
!= NULL
);
1241 if (drc
->drc_drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
)) {
1242 drc
->drc_byteswap
= B_TRUE
;
1243 (void) fletcher_4_incremental_byteswap(drr_begin
,
1244 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
1245 byteswap_record(drr_begin
);
1246 } else if (drc
->drc_drrb
->drr_magic
== DMU_BACKUP_MAGIC
) {
1247 (void) fletcher_4_incremental_native(drr_begin
,
1248 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
1250 return (SET_ERROR(EINVAL
));
1254 drc
->drc_voff
= *voffp
;
1255 drc
->drc_featureflags
=
1256 DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
);
1258 uint32_t payloadlen
= drc
->drc_drr_begin
->drr_payloadlen
;
1261 * Since OpenZFS 2.0.0, we have enforced a 64MB limit in userspace
1262 * configurable via ZFS_SENDRECV_MAX_NVLIST. We enforce 256MB as a hard
1263 * upper limit. Systems with less than 1GB of RAM will see a lower
1264 * limit from `arc_all_memory() / 4`.
1266 if (payloadlen
> (MIN((1U << 28), arc_all_memory() / 4)))
1270 if (payloadlen
!= 0) {
1271 void *payload
= vmem_alloc(payloadlen
, KM_SLEEP
);
1273 * For compatibility with recursive send streams, we don't do
1274 * this here if the stream could be part of a package. Instead,
1275 * we'll do it in dmu_recv_stream. If we pull the next header
1276 * too early, and it's the END record, we break the `recv_skip`
1280 err
= receive_read_payload_and_next_header(drc
, payloadlen
,
1283 vmem_free(payload
, payloadlen
);
1286 err
= nvlist_unpack(payload
, payloadlen
, &drc
->drc_begin_nvl
,
1288 vmem_free(payload
, payloadlen
);
1290 kmem_free(drc
->drc_next_rrd
,
1291 sizeof (*drc
->drc_next_rrd
));
1296 if (drc
->drc_drrb
->drr_flags
& DRR_FLAG_SPILL_BLOCK
)
1297 drc
->drc_spill
= B_TRUE
;
1299 drba
.drba_origin
= origin
;
1300 drba
.drba_cookie
= drc
;
1301 drba
.drba_cred
= CRED();
1302 drba
.drba_proc
= curproc
;
1304 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
1305 err
= dsl_sync_task(tofs
,
1306 dmu_recv_resume_begin_check
, dmu_recv_resume_begin_sync
,
1307 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
);
1310 * For non-raw, non-incremental, non-resuming receives the
1311 * user can specify encryption parameters on the command line
1312 * with "zfs recv -o". For these receives we create a dcp and
1313 * pass it to the sync task. Creating the dcp will implicitly
1314 * remove the encryption params from the localprops nvlist,
1315 * which avoids errors when trying to set these normally
1316 * read-only properties. Any other kind of receive that
1317 * attempts to set these properties will fail as a result.
1319 if ((DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
) &
1320 DMU_BACKUP_FEATURE_RAW
) == 0 &&
1321 origin
== NULL
&& drc
->drc_drrb
->drr_fromguid
== 0) {
1322 err
= dsl_crypto_params_create_nvlist(DCP_CMD_NONE
,
1323 localprops
, hidden_args
, &drba
.drba_dcp
);
1327 err
= dsl_sync_task(tofs
,
1328 dmu_recv_begin_check
, dmu_recv_begin_sync
,
1329 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
);
1330 dsl_crypto_params_free(drba
.drba_dcp
, !!err
);
1335 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
1336 nvlist_free(drc
->drc_begin_nvl
);
1342 * Holds data need for corrective recv callback
1344 typedef struct cr_cb_data
{
1346 zbookmark_phys_t zb
;
1351 corrective_read_done(zio_t
*zio
)
1353 cr_cb_data_t
*data
= zio
->io_private
;
1354 /* Corruption corrected; update error log if needed */
1355 if (zio
->io_error
== 0)
1356 spa_remove_error(data
->spa
, &data
->zb
, &zio
->io_bp
->blk_birth
);
1357 kmem_free(data
, sizeof (cr_cb_data_t
));
1358 abd_free(zio
->io_abd
);
1362 * zio_rewrite the data pointed to by bp with the data from the rrd's abd.
1365 do_corrective_recv(struct receive_writer_arg
*rwa
, struct drr_write
*drrw
,
1366 struct receive_record_arg
*rrd
, blkptr_t
*bp
)
1370 zbookmark_phys_t zb
;
1372 abd_t
*abd
= rrd
->abd
;
1373 zio_cksum_t bp_cksum
= bp
->blk_cksum
;
1374 zio_flag_t flags
= ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_DONT_RETRY
|
1378 flags
|= ZIO_FLAG_RAW
;
1380 err
= dnode_hold(rwa
->os
, drrw
->drr_object
, FTAG
, &dn
);
1383 SET_BOOKMARK(&zb
, dmu_objset_id(rwa
->os
), drrw
->drr_object
, 0,
1384 dbuf_whichblock(dn
, 0, drrw
->drr_offset
));
1385 dnode_rele(dn
, FTAG
);
1387 if (!rwa
->raw
&& DRR_WRITE_COMPRESSED(drrw
)) {
1388 /* Decompress the stream data */
1389 abd_t
*dabd
= abd_alloc_linear(
1390 drrw
->drr_logical_size
, B_FALSE
);
1391 err
= zio_decompress_data(drrw
->drr_compressiontype
,
1392 abd
, abd_to_buf(dabd
), abd_get_size(abd
),
1393 abd_get_size(dabd
), NULL
);
1399 /* Swap in the newly decompressed data into the abd */
1404 if (!rwa
->raw
&& BP_GET_COMPRESS(bp
) != ZIO_COMPRESS_OFF
) {
1405 /* Recompress the data */
1406 abd_t
*cabd
= abd_alloc_linear(BP_GET_PSIZE(bp
),
1408 void *buf
= abd_to_buf(cabd
);
1409 uint64_t csize
= zio_compress_data(BP_GET_COMPRESS(bp
),
1410 abd
, &buf
, abd_get_size(abd
),
1411 rwa
->os
->os_complevel
);
1412 abd_zero_off(cabd
, csize
, BP_GET_PSIZE(bp
) - csize
);
1413 /* Swap in newly compressed data into the abd */
1416 flags
|= ZIO_FLAG_RAW_COMPRESS
;
1420 * The stream is not encrypted but the data on-disk is.
1421 * We need to re-encrypt the buf using the same
1422 * encryption type, salt, iv, and mac that was used to encrypt
1423 * the block previosly.
1425 if (!rwa
->raw
&& BP_USES_CRYPT(bp
)) {
1427 dsl_crypto_key_t
*dck
= NULL
;
1428 uint8_t salt
[ZIO_DATA_SALT_LEN
];
1429 uint8_t iv
[ZIO_DATA_IV_LEN
];
1430 uint8_t mac
[ZIO_DATA_MAC_LEN
];
1431 boolean_t no_crypt
= B_FALSE
;
1432 dsl_pool_t
*dp
= dmu_objset_pool(rwa
->os
);
1433 abd_t
*eabd
= abd_alloc_linear(BP_GET_PSIZE(bp
), B_FALSE
);
1435 zio_crypt_decode_params_bp(bp
, salt
, iv
);
1436 zio_crypt_decode_mac_bp(bp
, mac
);
1438 dsl_pool_config_enter(dp
, FTAG
);
1439 err
= dsl_dataset_hold_flags(dp
, rwa
->tofs
,
1440 DS_HOLD_FLAG_DECRYPT
, FTAG
, &ds
);
1442 dsl_pool_config_exit(dp
, FTAG
);
1444 return (SET_ERROR(EACCES
));
1447 /* Look up the key from the spa's keystore */
1448 err
= spa_keystore_lookup_key(rwa
->os
->os_spa
,
1449 zb
.zb_objset
, FTAG
, &dck
);
1451 dsl_dataset_rele_flags(ds
, DS_HOLD_FLAG_DECRYPT
,
1453 dsl_pool_config_exit(dp
, FTAG
);
1455 return (SET_ERROR(EACCES
));
1458 err
= zio_do_crypt_abd(B_TRUE
, &dck
->dck_key
,
1459 BP_GET_TYPE(bp
), BP_SHOULD_BYTESWAP(bp
), salt
, iv
,
1460 mac
, abd_get_size(abd
), abd
, eabd
, &no_crypt
);
1462 spa_keystore_dsl_key_rele(rwa
->os
->os_spa
, dck
, FTAG
);
1463 dsl_dataset_rele_flags(ds
, DS_HOLD_FLAG_DECRYPT
, FTAG
);
1464 dsl_pool_config_exit(dp
, FTAG
);
1471 /* Swap in the newly encrypted data into the abd */
1476 * We want to prevent zio_rewrite() from trying to
1477 * encrypt the data again
1479 flags
|= ZIO_FLAG_RAW_ENCRYPT
;
1483 io
= zio_rewrite(NULL
, rwa
->os
->os_spa
, bp
->blk_birth
, bp
, abd
,
1484 BP_GET_PSIZE(bp
), NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
, flags
, &zb
);
1486 ASSERT(abd_get_size(abd
) == BP_GET_LSIZE(bp
) ||
1487 abd_get_size(abd
) == BP_GET_PSIZE(bp
));
1489 /* compute new bp checksum value and make sure it matches the old one */
1490 zio_checksum_compute(io
, BP_GET_CHECKSUM(bp
), abd
, abd_get_size(abd
));
1491 if (!ZIO_CHECKSUM_EQUAL(bp_cksum
, io
->io_bp
->blk_cksum
)) {
1493 if (zfs_recv_best_effort_corrective
!= 0)
1495 return (SET_ERROR(ECKSUM
));
1498 /* Correct the corruption in place */
1501 cr_cb_data_t
*cb_data
=
1502 kmem_alloc(sizeof (cr_cb_data_t
), KM_SLEEP
);
1503 cb_data
->spa
= rwa
->os
->os_spa
;
1504 cb_data
->size
= drrw
->drr_logical_size
;
1506 /* Test if healing worked by re-reading the bp */
1507 err
= zio_wait(zio_read(rwa
->heal_pio
, rwa
->os
->os_spa
, bp
,
1508 abd_alloc_for_io(drrw
->drr_logical_size
, B_FALSE
),
1509 drrw
->drr_logical_size
, corrective_read_done
,
1510 cb_data
, ZIO_PRIORITY_ASYNC_READ
, flags
, NULL
));
1512 if (err
!= 0 && zfs_recv_best_effort_corrective
!= 0)
1519 receive_read(dmu_recv_cookie_t
*drc
, int len
, void *buf
)
1524 * The code doesn't rely on this (lengths being multiples of 8). See
1525 * comment in dump_bytes.
1527 ASSERT(len
% 8 == 0 ||
1528 (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0);
1530 while (done
< len
) {
1531 ssize_t resid
= len
- done
;
1532 zfs_file_t
*fp
= drc
->drc_fp
;
1533 int err
= zfs_file_read(fp
, (char *)buf
+ done
,
1534 len
- done
, &resid
);
1535 if (err
== 0 && resid
== len
- done
) {
1537 * Note: ECKSUM or ZFS_ERR_STREAM_TRUNCATED indicates
1538 * that the receive was interrupted and can
1539 * potentially be resumed.
1541 err
= SET_ERROR(ZFS_ERR_STREAM_TRUNCATED
);
1543 drc
->drc_voff
+= len
- done
- resid
;
1549 drc
->drc_bytes_read
+= len
;
1551 ASSERT3U(done
, ==, len
);
1555 static inline uint8_t
1556 deduce_nblkptr(dmu_object_type_t bonus_type
, uint64_t bonus_size
)
1558 if (bonus_type
== DMU_OT_SA
) {
1562 ((DN_OLD_MAX_BONUSLEN
-
1563 MIN(DN_OLD_MAX_BONUSLEN
, bonus_size
)) >> SPA_BLKPTRSHIFT
));
1568 save_resume_state(struct receive_writer_arg
*rwa
,
1569 uint64_t object
, uint64_t offset
, dmu_tx_t
*tx
)
1571 int txgoff
= dmu_tx_get_txg(tx
) & TXG_MASK
;
1573 if (!rwa
->resumable
)
1577 * We use ds_resume_bytes[] != 0 to indicate that we need to
1578 * update this on disk, so it must not be 0.
1580 ASSERT(rwa
->bytes_read
!= 0);
1583 * We only resume from write records, which have a valid
1584 * (non-meta-dnode) object number.
1586 ASSERT(object
!= 0);
1589 * For resuming to work correctly, we must receive records in order,
1590 * sorted by object,offset. This is checked by the callers, but
1591 * assert it here for good measure.
1593 ASSERT3U(object
, >=, rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
]);
1594 ASSERT(object
!= rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] ||
1595 offset
>= rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
]);
1596 ASSERT3U(rwa
->bytes_read
, >=,
1597 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
]);
1599 rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] = object
;
1600 rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
] = offset
;
1601 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
] = rwa
->bytes_read
;
1605 receive_object_is_same_generation(objset_t
*os
, uint64_t object
,
1606 dmu_object_type_t old_bonus_type
, dmu_object_type_t new_bonus_type
,
1607 const void *new_bonus
, boolean_t
*samegenp
)
1609 zfs_file_info_t zoi
;
1612 dmu_buf_t
*old_bonus_dbuf
;
1613 err
= dmu_bonus_hold(os
, object
, FTAG
, &old_bonus_dbuf
);
1616 err
= dmu_get_file_info(os
, old_bonus_type
, old_bonus_dbuf
->db_data
,
1618 dmu_buf_rele(old_bonus_dbuf
, FTAG
);
1621 uint64_t old_gen
= zoi
.zfi_generation
;
1623 err
= dmu_get_file_info(os
, new_bonus_type
, new_bonus
, &zoi
);
1626 uint64_t new_gen
= zoi
.zfi_generation
;
1628 *samegenp
= (old_gen
== new_gen
);
1633 receive_handle_existing_object(const struct receive_writer_arg
*rwa
,
1634 const struct drr_object
*drro
, const dmu_object_info_t
*doi
,
1635 const void *bonus_data
,
1636 uint64_t *object_to_hold
, uint32_t *new_blksz
)
1638 uint32_t indblksz
= drro
->drr_indblkshift
?
1639 1ULL << drro
->drr_indblkshift
: 0;
1640 int nblkptr
= deduce_nblkptr(drro
->drr_bonustype
,
1641 drro
->drr_bonuslen
);
1642 uint8_t dn_slots
= drro
->drr_dn_slots
!= 0 ?
1643 drro
->drr_dn_slots
: DNODE_MIN_SLOTS
;
1644 boolean_t do_free_range
= B_FALSE
;
1647 *object_to_hold
= drro
->drr_object
;
1649 /* nblkptr should be bounded by the bonus size and type */
1650 if (rwa
->raw
&& nblkptr
!= drro
->drr_nblkptr
)
1651 return (SET_ERROR(EINVAL
));
1654 * After the previous send stream, the sending system may
1655 * have freed this object, and then happened to re-allocate
1656 * this object number in a later txg. In this case, we are
1657 * receiving a different logical file, and the block size may
1658 * appear to be different. i.e. we may have a different
1659 * block size for this object than what the send stream says.
1660 * In this case we need to remove the object's contents,
1661 * so that its structure can be changed and then its contents
1662 * entirely replaced by subsequent WRITE records.
1664 * If this is a -L (--large-block) incremental stream, and
1665 * the previous stream was not -L, the block size may appear
1666 * to increase. i.e. we may have a smaller block size for
1667 * this object than what the send stream says. In this case
1668 * we need to keep the object's contents and block size
1669 * intact, so that we don't lose parts of the object's
1670 * contents that are not changed by this incremental send
1673 * We can distinguish between the two above cases by using
1674 * the ZPL's generation number (see
1675 * receive_object_is_same_generation()). However, we only
1676 * want to rely on the generation number when absolutely
1677 * necessary, because with raw receives, the generation is
1678 * encrypted. We also want to minimize dependence on the
1679 * ZPL, so that other types of datasets can also be received
1680 * (e.g. ZVOLs, although note that ZVOLS currently do not
1681 * reallocate their objects or change their structure).
1682 * Therefore, we check a number of different cases where we
1683 * know it is safe to discard the object's contents, before
1684 * using the ZPL's generation number to make the above
1687 if (drro
->drr_blksz
!= doi
->doi_data_block_size
) {
1690 * RAW streams always have large blocks, so
1691 * we are sure that the data is not needed
1692 * due to changing --large-block to be on.
1693 * Which is fortunate since the bonus buffer
1694 * (which contains the ZPL generation) is
1695 * encrypted, and the key might not be
1698 do_free_range
= B_TRUE
;
1699 } else if (rwa
->full
) {
1701 * This is a full send stream, so it always
1702 * replaces what we have. Even if the
1703 * generation numbers happen to match, this
1704 * can not actually be the same logical file.
1705 * This is relevant when receiving a full
1708 do_free_range
= B_TRUE
;
1709 } else if (drro
->drr_type
!=
1710 DMU_OT_PLAIN_FILE_CONTENTS
||
1711 doi
->doi_type
!= DMU_OT_PLAIN_FILE_CONTENTS
) {
1713 * PLAIN_FILE_CONTENTS are the only type of
1714 * objects that have ever been stored with
1715 * large blocks, so we don't need the special
1716 * logic below. ZAP blocks can shrink (when
1717 * there's only one block), so we don't want
1718 * to hit the error below about block size
1721 do_free_range
= B_TRUE
;
1722 } else if (doi
->doi_max_offset
<=
1723 doi
->doi_data_block_size
) {
1725 * There is only one block. We can free it,
1726 * because its contents will be replaced by a
1727 * WRITE record. This can not be the no-L ->
1728 * -L case, because the no-L case would have
1729 * resulted in multiple blocks. If we
1730 * supported -L -> no-L, it would not be safe
1731 * to free the file's contents. Fortunately,
1732 * that is not allowed (see
1733 * recv_check_large_blocks()).
1735 do_free_range
= B_TRUE
;
1737 boolean_t is_same_gen
;
1738 err
= receive_object_is_same_generation(rwa
->os
,
1739 drro
->drr_object
, doi
->doi_bonus_type
,
1740 drro
->drr_bonustype
, bonus_data
, &is_same_gen
);
1742 return (SET_ERROR(EINVAL
));
1746 * This is the same logical file, and
1747 * the block size must be increasing.
1748 * It could only decrease if
1749 * --large-block was changed to be
1750 * off, which is checked in
1751 * recv_check_large_blocks().
1753 if (drro
->drr_blksz
<=
1754 doi
->doi_data_block_size
)
1755 return (SET_ERROR(EINVAL
));
1757 * We keep the existing blocksize and
1761 doi
->doi_data_block_size
;
1763 do_free_range
= B_TRUE
;
1768 /* nblkptr can only decrease if the object was reallocated */
1769 if (nblkptr
< doi
->doi_nblkptr
)
1770 do_free_range
= B_TRUE
;
1772 /* number of slots can only change on reallocation */
1773 if (dn_slots
!= doi
->doi_dnodesize
>> DNODE_SHIFT
)
1774 do_free_range
= B_TRUE
;
1777 * For raw sends we also check a few other fields to
1778 * ensure we are preserving the objset structure exactly
1779 * as it was on the receive side:
1780 * - A changed indirect block size
1781 * - A smaller nlevels
1784 if (indblksz
!= doi
->doi_metadata_block_size
)
1785 do_free_range
= B_TRUE
;
1786 if (drro
->drr_nlevels
< doi
->doi_indirection
)
1787 do_free_range
= B_TRUE
;
1790 if (do_free_range
) {
1791 err
= dmu_free_long_range(rwa
->os
, drro
->drr_object
,
1794 return (SET_ERROR(EINVAL
));
1798 * The dmu does not currently support decreasing nlevels or changing
1799 * indirect block size if there is already one, same as changing the
1800 * number of of dnode slots on an object. For non-raw sends this
1801 * does not matter and the new object can just use the previous one's
1802 * parameters. For raw sends, however, the structure of the received
1803 * dnode (including indirects and dnode slots) must match that of the
1804 * send side. Therefore, instead of using dmu_object_reclaim(), we
1805 * must free the object completely and call dmu_object_claim_dnsize()
1808 if ((rwa
->raw
&& ((doi
->doi_indirection
> 1 &&
1809 indblksz
!= doi
->doi_metadata_block_size
) ||
1810 drro
->drr_nlevels
< doi
->doi_indirection
)) ||
1811 dn_slots
!= doi
->doi_dnodesize
>> DNODE_SHIFT
) {
1812 err
= dmu_free_long_object(rwa
->os
, drro
->drr_object
);
1814 return (SET_ERROR(EINVAL
));
1816 txg_wait_synced(dmu_objset_pool(rwa
->os
), 0);
1817 *object_to_hold
= DMU_NEW_OBJECT
;
1821 * For raw receives, free everything beyond the new incoming
1822 * maxblkid. Normally this would be done with a DRR_FREE
1823 * record that would come after this DRR_OBJECT record is
1824 * processed. However, for raw receives we manually set the
1825 * maxblkid from the drr_maxblkid and so we must first free
1826 * everything above that blkid to ensure the DMU is always
1827 * consistent with itself. We will never free the first block
1828 * of the object here because a maxblkid of 0 could indicate
1829 * an object with a single block or one with no blocks. This
1830 * free may be skipped when dmu_free_long_range() was called
1831 * above since it covers the entire object's contents.
1833 if (rwa
->raw
&& *object_to_hold
!= DMU_NEW_OBJECT
&& !do_free_range
) {
1834 err
= dmu_free_long_range(rwa
->os
, drro
->drr_object
,
1835 (drro
->drr_maxblkid
+ 1) * doi
->doi_data_block_size
,
1838 return (SET_ERROR(EINVAL
));
1844 receive_object(struct receive_writer_arg
*rwa
, struct drr_object
*drro
,
1847 dmu_object_info_t doi
;
1850 uint32_t new_blksz
= drro
->drr_blksz
;
1851 uint8_t dn_slots
= drro
->drr_dn_slots
!= 0 ?
1852 drro
->drr_dn_slots
: DNODE_MIN_SLOTS
;
1854 if (drro
->drr_type
== DMU_OT_NONE
||
1855 !DMU_OT_IS_VALID(drro
->drr_type
) ||
1856 !DMU_OT_IS_VALID(drro
->drr_bonustype
) ||
1857 drro
->drr_checksumtype
>= ZIO_CHECKSUM_FUNCTIONS
||
1858 drro
->drr_compress
>= ZIO_COMPRESS_FUNCTIONS
||
1859 P2PHASE(drro
->drr_blksz
, SPA_MINBLOCKSIZE
) ||
1860 drro
->drr_blksz
< SPA_MINBLOCKSIZE
||
1861 drro
->drr_blksz
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)) ||
1862 drro
->drr_bonuslen
>
1863 DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa
->os
))) ||
1865 (spa_maxdnodesize(dmu_objset_spa(rwa
->os
)) >> DNODE_SHIFT
)) {
1866 return (SET_ERROR(EINVAL
));
1871 * We should have received a DRR_OBJECT_RANGE record
1872 * containing this block and stored it in rwa.
1874 if (drro
->drr_object
< rwa
->or_firstobj
||
1875 drro
->drr_object
>= rwa
->or_firstobj
+ rwa
->or_numslots
||
1876 drro
->drr_raw_bonuslen
< drro
->drr_bonuslen
||
1877 drro
->drr_indblkshift
> SPA_MAXBLOCKSHIFT
||
1878 drro
->drr_nlevels
> DN_MAX_LEVELS
||
1879 drro
->drr_nblkptr
> DN_MAX_NBLKPTR
||
1880 DN_SLOTS_TO_BONUSLEN(dn_slots
) <
1881 drro
->drr_raw_bonuslen
)
1882 return (SET_ERROR(EINVAL
));
1885 * The DRR_OBJECT_SPILL flag is valid when the DRR_BEGIN
1886 * record indicates this by setting DRR_FLAG_SPILL_BLOCK.
1888 if (((drro
->drr_flags
& ~(DRR_OBJECT_SPILL
))) ||
1889 (!rwa
->spill
&& DRR_OBJECT_HAS_SPILL(drro
->drr_flags
))) {
1890 return (SET_ERROR(EINVAL
));
1893 if (drro
->drr_raw_bonuslen
!= 0 || drro
->drr_nblkptr
!= 0 ||
1894 drro
->drr_indblkshift
!= 0 || drro
->drr_nlevels
!= 0) {
1895 return (SET_ERROR(EINVAL
));
1899 err
= dmu_object_info(rwa
->os
, drro
->drr_object
, &doi
);
1901 if (err
!= 0 && err
!= ENOENT
&& err
!= EEXIST
)
1902 return (SET_ERROR(EINVAL
));
1904 if (drro
->drr_object
> rwa
->max_object
)
1905 rwa
->max_object
= drro
->drr_object
;
1908 * If we are losing blkptrs or changing the block size this must
1909 * be a new file instance. We must clear out the previous file
1910 * contents before we can change this type of metadata in the dnode.
1911 * Raw receives will also check that the indirect structure of the
1912 * dnode hasn't changed.
1914 uint64_t object_to_hold
;
1916 err
= receive_handle_existing_object(rwa
, drro
, &doi
, data
,
1917 &object_to_hold
, &new_blksz
);
1920 } else if (err
== EEXIST
) {
1922 * The object requested is currently an interior slot of a
1923 * multi-slot dnode. This will be resolved when the next txg
1924 * is synced out, since the send stream will have told us
1925 * to free this slot when we freed the associated dnode
1926 * earlier in the stream.
1928 txg_wait_synced(dmu_objset_pool(rwa
->os
), 0);
1930 if (dmu_object_info(rwa
->os
, drro
->drr_object
, NULL
) != ENOENT
)
1931 return (SET_ERROR(EINVAL
));
1933 /* object was freed and we are about to allocate a new one */
1934 object_to_hold
= DMU_NEW_OBJECT
;
1937 * If the only record in this range so far was DRR_FREEOBJECTS
1938 * with at least one actually freed object, it's possible that
1939 * the block will now be converted to a hole. We need to wait
1940 * for the txg to sync to prevent races.
1942 if (rwa
->or_need_sync
== ORNS_YES
)
1943 txg_wait_synced(dmu_objset_pool(rwa
->os
), 0);
1945 /* object is free and we are about to allocate a new one */
1946 object_to_hold
= DMU_NEW_OBJECT
;
1949 /* Only relevant for the first object in the range */
1950 rwa
->or_need_sync
= ORNS_NO
;
1953 * If this is a multi-slot dnode there is a chance that this
1954 * object will expand into a slot that is already used by
1955 * another object from the previous snapshot. We must free
1956 * these objects before we attempt to allocate the new dnode.
1959 boolean_t need_sync
= B_FALSE
;
1961 for (uint64_t slot
= drro
->drr_object
+ 1;
1962 slot
< drro
->drr_object
+ dn_slots
;
1964 dmu_object_info_t slot_doi
;
1966 err
= dmu_object_info(rwa
->os
, slot
, &slot_doi
);
1967 if (err
== ENOENT
|| err
== EEXIST
)
1972 err
= dmu_free_long_object(rwa
->os
, slot
);
1980 txg_wait_synced(dmu_objset_pool(rwa
->os
), 0);
1983 tx
= dmu_tx_create(rwa
->os
);
1984 dmu_tx_hold_bonus(tx
, object_to_hold
);
1985 dmu_tx_hold_write(tx
, object_to_hold
, 0, 0);
1986 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1992 if (object_to_hold
== DMU_NEW_OBJECT
) {
1993 /* Currently free, wants to be allocated */
1994 err
= dmu_object_claim_dnsize(rwa
->os
, drro
->drr_object
,
1995 drro
->drr_type
, new_blksz
,
1996 drro
->drr_bonustype
, drro
->drr_bonuslen
,
1997 dn_slots
<< DNODE_SHIFT
, tx
);
1998 } else if (drro
->drr_type
!= doi
.doi_type
||
1999 new_blksz
!= doi
.doi_data_block_size
||
2000 drro
->drr_bonustype
!= doi
.doi_bonus_type
||
2001 drro
->drr_bonuslen
!= doi
.doi_bonus_size
) {
2002 /* Currently allocated, but with different properties */
2003 err
= dmu_object_reclaim_dnsize(rwa
->os
, drro
->drr_object
,
2004 drro
->drr_type
, new_blksz
,
2005 drro
->drr_bonustype
, drro
->drr_bonuslen
,
2006 dn_slots
<< DNODE_SHIFT
, rwa
->spill
?
2007 DRR_OBJECT_HAS_SPILL(drro
->drr_flags
) : B_FALSE
, tx
);
2008 } else if (rwa
->spill
&& !DRR_OBJECT_HAS_SPILL(drro
->drr_flags
)) {
2010 * Currently allocated, the existing version of this object
2011 * may reference a spill block that is no longer allocated
2012 * at the source and needs to be freed.
2014 err
= dmu_object_rm_spill(rwa
->os
, drro
->drr_object
, tx
);
2019 return (SET_ERROR(EINVAL
));
2022 if (rwa
->or_crypt_params_present
) {
2024 * Set the crypt params for the buffer associated with this
2025 * range of dnodes. This causes the blkptr_t to have the
2026 * same crypt params (byteorder, salt, iv, mac) as on the
2029 * Since we are committing this tx now, it is possible for
2030 * the dnode block to end up on-disk with the incorrect MAC,
2031 * if subsequent objects in this block are received in a
2032 * different txg. However, since the dataset is marked as
2033 * inconsistent, no code paths will do a non-raw read (or
2034 * decrypt the block / verify the MAC). The receive code and
2035 * scrub code can safely do raw reads and verify the
2036 * checksum. They don't need to verify the MAC.
2038 dmu_buf_t
*db
= NULL
;
2039 uint64_t offset
= rwa
->or_firstobj
* DNODE_MIN_SIZE
;
2041 err
= dmu_buf_hold_by_dnode(DMU_META_DNODE(rwa
->os
),
2042 offset
, FTAG
, &db
, DMU_READ_PREFETCH
| DMU_READ_NO_DECRYPT
);
2045 return (SET_ERROR(EINVAL
));
2048 dmu_buf_set_crypt_params(db
, rwa
->or_byteorder
,
2049 rwa
->or_salt
, rwa
->or_iv
, rwa
->or_mac
, tx
);
2051 dmu_buf_rele(db
, FTAG
);
2053 rwa
->or_crypt_params_present
= B_FALSE
;
2056 dmu_object_set_checksum(rwa
->os
, drro
->drr_object
,
2057 drro
->drr_checksumtype
, tx
);
2058 dmu_object_set_compress(rwa
->os
, drro
->drr_object
,
2059 drro
->drr_compress
, tx
);
2061 /* handle more restrictive dnode structuring for raw recvs */
2064 * Set the indirect block size, block shift, nlevels.
2065 * This will not fail because we ensured all of the
2066 * blocks were freed earlier if this is a new object.
2067 * For non-new objects block size and indirect block
2068 * shift cannot change and nlevels can only increase.
2070 ASSERT3U(new_blksz
, ==, drro
->drr_blksz
);
2071 VERIFY0(dmu_object_set_blocksize(rwa
->os
, drro
->drr_object
,
2072 drro
->drr_blksz
, drro
->drr_indblkshift
, tx
));
2073 VERIFY0(dmu_object_set_nlevels(rwa
->os
, drro
->drr_object
,
2074 drro
->drr_nlevels
, tx
));
2077 * Set the maxblkid. This will always succeed because
2078 * we freed all blocks beyond the new maxblkid above.
2080 VERIFY0(dmu_object_set_maxblkid(rwa
->os
, drro
->drr_object
,
2081 drro
->drr_maxblkid
, tx
));
2087 uint32_t flags
= DMU_READ_NO_PREFETCH
;
2090 flags
|= DMU_READ_NO_DECRYPT
;
2092 VERIFY0(dnode_hold(rwa
->os
, drro
->drr_object
, FTAG
, &dn
));
2093 VERIFY0(dmu_bonus_hold_by_dnode(dn
, FTAG
, &db
, flags
));
2095 dmu_buf_will_dirty(db
, tx
);
2097 ASSERT3U(db
->db_size
, >=, drro
->drr_bonuslen
);
2098 memcpy(db
->db_data
, data
, DRR_OBJECT_PAYLOAD_SIZE(drro
));
2101 * Raw bonus buffers have their byteorder determined by the
2102 * DRR_OBJECT_RANGE record.
2104 if (rwa
->byteswap
&& !rwa
->raw
) {
2105 dmu_object_byteswap_t byteswap
=
2106 DMU_OT_BYTESWAP(drro
->drr_bonustype
);
2107 dmu_ot_byteswap
[byteswap
].ob_func(db
->db_data
,
2108 DRR_OBJECT_PAYLOAD_SIZE(drro
));
2110 dmu_buf_rele(db
, FTAG
);
2111 dnode_rele(dn
, FTAG
);
2119 receive_freeobjects(struct receive_writer_arg
*rwa
,
2120 struct drr_freeobjects
*drrfo
)
2125 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
< drrfo
->drr_firstobj
)
2126 return (SET_ERROR(EINVAL
));
2128 for (obj
= drrfo
->drr_firstobj
== 0 ? 1 : drrfo
->drr_firstobj
;
2129 obj
< drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
&&
2130 obj
< DN_MAX_OBJECT
&& next_err
== 0;
2131 next_err
= dmu_object_next(rwa
->os
, &obj
, FALSE
, 0)) {
2132 dmu_object_info_t doi
;
2135 err
= dmu_object_info(rwa
->os
, obj
, &doi
);
2141 err
= dmu_free_long_object(rwa
->os
, obj
);
2146 if (rwa
->or_need_sync
== ORNS_MAYBE
)
2147 rwa
->or_need_sync
= ORNS_YES
;
2149 if (next_err
!= ESRCH
)
2155 * Note: if this fails, the caller will clean up any records left on the
2156 * rwa->write_batch list.
2159 flush_write_batch_impl(struct receive_writer_arg
*rwa
)
2164 if (dnode_hold(rwa
->os
, rwa
->last_object
, FTAG
, &dn
) != 0)
2165 return (SET_ERROR(EINVAL
));
2167 struct receive_record_arg
*last_rrd
= list_tail(&rwa
->write_batch
);
2168 struct drr_write
*last_drrw
= &last_rrd
->header
.drr_u
.drr_write
;
2170 struct receive_record_arg
*first_rrd
= list_head(&rwa
->write_batch
);
2171 struct drr_write
*first_drrw
= &first_rrd
->header
.drr_u
.drr_write
;
2173 ASSERT3U(rwa
->last_object
, ==, last_drrw
->drr_object
);
2174 ASSERT3U(rwa
->last_offset
, ==, last_drrw
->drr_offset
);
2176 dmu_tx_t
*tx
= dmu_tx_create(rwa
->os
);
2177 dmu_tx_hold_write_by_dnode(tx
, dn
, first_drrw
->drr_offset
,
2178 last_drrw
->drr_offset
- first_drrw
->drr_offset
+
2179 last_drrw
->drr_logical_size
);
2180 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2183 dnode_rele(dn
, FTAG
);
2187 struct receive_record_arg
*rrd
;
2188 while ((rrd
= list_head(&rwa
->write_batch
)) != NULL
) {
2189 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
2190 abd_t
*abd
= rrd
->abd
;
2192 ASSERT3U(drrw
->drr_object
, ==, rwa
->last_object
);
2194 if (drrw
->drr_logical_size
!= dn
->dn_datablksz
) {
2196 * The WRITE record is larger than the object's block
2197 * size. We must be receiving an incremental
2198 * large-block stream into a dataset that previously did
2199 * a non-large-block receive. Lightweight writes must
2200 * be exactly one block, so we need to decompress the
2201 * data (if compressed) and do a normal dmu_write().
2203 ASSERT3U(drrw
->drr_logical_size
, >, dn
->dn_datablksz
);
2204 if (DRR_WRITE_COMPRESSED(drrw
)) {
2206 abd_alloc_linear(drrw
->drr_logical_size
,
2209 err
= zio_decompress_data(
2210 drrw
->drr_compressiontype
,
2211 abd
, abd_to_buf(decomp_abd
),
2213 abd_get_size(decomp_abd
), NULL
);
2216 dmu_write_by_dnode(dn
,
2218 drrw
->drr_logical_size
,
2219 abd_to_buf(decomp_abd
), tx
);
2221 abd_free(decomp_abd
);
2223 dmu_write_by_dnode(dn
,
2225 drrw
->drr_logical_size
,
2226 abd_to_buf(abd
), tx
);
2231 zio_prop_t zp
= {0};
2232 dmu_write_policy(rwa
->os
, dn
, 0, 0, &zp
);
2234 zio_flag_t zio_flags
= 0;
2237 zp
.zp_encrypt
= B_TRUE
;
2238 zp
.zp_compress
= drrw
->drr_compressiontype
;
2239 zp
.zp_byteorder
= ZFS_HOST_BYTEORDER
^
2240 !!DRR_IS_RAW_BYTESWAPPED(drrw
->drr_flags
) ^
2242 memcpy(zp
.zp_salt
, drrw
->drr_salt
,
2244 memcpy(zp
.zp_iv
, drrw
->drr_iv
,
2246 memcpy(zp
.zp_mac
, drrw
->drr_mac
,
2248 if (DMU_OT_IS_ENCRYPTED(zp
.zp_type
)) {
2249 zp
.zp_nopwrite
= B_FALSE
;
2250 zp
.zp_copies
= MIN(zp
.zp_copies
,
2251 SPA_DVAS_PER_BP
- 1);
2253 zio_flags
|= ZIO_FLAG_RAW
;
2254 } else if (DRR_WRITE_COMPRESSED(drrw
)) {
2255 ASSERT3U(drrw
->drr_compressed_size
, >, 0);
2256 ASSERT3U(drrw
->drr_logical_size
, >=,
2257 drrw
->drr_compressed_size
);
2258 zp
.zp_compress
= drrw
->drr_compressiontype
;
2259 zio_flags
|= ZIO_FLAG_RAW_COMPRESS
;
2260 } else if (rwa
->byteswap
) {
2262 * Note: compressed blocks never need to be
2263 * byteswapped, because WRITE records for
2264 * metadata blocks are never compressed. The
2265 * exception is raw streams, which are written
2266 * in the original byteorder, and the byteorder
2267 * bit is preserved in the BP by setting
2268 * zp_byteorder above.
2270 dmu_object_byteswap_t byteswap
=
2271 DMU_OT_BYTESWAP(drrw
->drr_type
);
2272 dmu_ot_byteswap
[byteswap
].ob_func(
2274 DRR_WRITE_PAYLOAD_SIZE(drrw
));
2278 * Since this data can't be read until the receive
2279 * completes, we can do a "lightweight" write for
2280 * improved performance.
2282 err
= dmu_lightweight_write_by_dnode(dn
,
2283 drrw
->drr_offset
, abd
, &zp
, zio_flags
, tx
);
2288 * This rrd is left on the list, so the caller will
2289 * free it (and the abd).
2295 * Note: If the receive fails, we want the resume stream to
2296 * start with the same record that we last successfully
2297 * received (as opposed to the next record), so that we can
2298 * verify that we are resuming from the correct location.
2300 save_resume_state(rwa
, drrw
->drr_object
, drrw
->drr_offset
, tx
);
2302 list_remove(&rwa
->write_batch
, rrd
);
2303 kmem_free(rrd
, sizeof (*rrd
));
2307 dnode_rele(dn
, FTAG
);
2312 flush_write_batch(struct receive_writer_arg
*rwa
)
2314 if (list_is_empty(&rwa
->write_batch
))
2318 err
= flush_write_batch_impl(rwa
);
2320 struct receive_record_arg
*rrd
;
2321 while ((rrd
= list_remove_head(&rwa
->write_batch
)) != NULL
) {
2323 kmem_free(rrd
, sizeof (*rrd
));
2326 ASSERT(list_is_empty(&rwa
->write_batch
));
2331 receive_process_write_record(struct receive_writer_arg
*rwa
,
2332 struct receive_record_arg
*rrd
)
2336 ASSERT3U(rrd
->header
.drr_type
, ==, DRR_WRITE
);
2337 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
2339 if (drrw
->drr_offset
+ drrw
->drr_logical_size
< drrw
->drr_offset
||
2340 !DMU_OT_IS_VALID(drrw
->drr_type
))
2341 return (SET_ERROR(EINVAL
));
2347 int flags
= DB_RF_CANFAIL
;
2350 flags
|= DB_RF_NO_DECRYPT
;
2352 if (rwa
->byteswap
) {
2353 dmu_object_byteswap_t byteswap
=
2354 DMU_OT_BYTESWAP(drrw
->drr_type
);
2355 dmu_ot_byteswap
[byteswap
].ob_func(abd_to_buf(rrd
->abd
),
2356 DRR_WRITE_PAYLOAD_SIZE(drrw
));
2359 err
= dmu_buf_hold_noread(rwa
->os
, drrw
->drr_object
,
2360 drrw
->drr_offset
, FTAG
, &dbp
);
2364 /* Try to read the object to see if it needs healing */
2365 err
= dbuf_read((dmu_buf_impl_t
*)dbp
, NULL
, flags
);
2367 * We only try to heal when dbuf_read() returns a ECKSUMs.
2368 * Other errors (even EIO) get returned to caller.
2369 * EIO indicates that the device is not present/accessible,
2370 * so writing to it will likely fail.
2371 * If the block is healthy, we don't want to overwrite it
2374 if (err
!= ECKSUM
) {
2375 dmu_buf_rele(dbp
, FTAG
);
2378 dn
= dmu_buf_dnode_enter(dbp
);
2379 /* Make sure the on-disk block and recv record sizes match */
2380 if (drrw
->drr_logical_size
!=
2381 dn
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
) {
2383 dmu_buf_dnode_exit(dbp
);
2384 dmu_buf_rele(dbp
, FTAG
);
2387 /* Get the block pointer for the corrupted block */
2388 bp
= dmu_buf_get_blkptr(dbp
);
2389 err
= do_corrective_recv(rwa
, drrw
, rrd
, bp
);
2390 dmu_buf_dnode_exit(dbp
);
2391 dmu_buf_rele(dbp
, FTAG
);
2396 * For resuming to work, records must be in increasing order
2397 * by (object, offset).
2399 if (drrw
->drr_object
< rwa
->last_object
||
2400 (drrw
->drr_object
== rwa
->last_object
&&
2401 drrw
->drr_offset
< rwa
->last_offset
)) {
2402 return (SET_ERROR(EINVAL
));
2405 struct receive_record_arg
*first_rrd
= list_head(&rwa
->write_batch
);
2406 struct drr_write
*first_drrw
= &first_rrd
->header
.drr_u
.drr_write
;
2407 uint64_t batch_size
=
2408 MIN(zfs_recv_write_batch_size
, DMU_MAX_ACCESS
/ 2);
2409 if (first_rrd
!= NULL
&&
2410 (drrw
->drr_object
!= first_drrw
->drr_object
||
2411 drrw
->drr_offset
>= first_drrw
->drr_offset
+ batch_size
)) {
2412 err
= flush_write_batch(rwa
);
2417 rwa
->last_object
= drrw
->drr_object
;
2418 rwa
->last_offset
= drrw
->drr_offset
;
2420 if (rwa
->last_object
> rwa
->max_object
)
2421 rwa
->max_object
= rwa
->last_object
;
2423 list_insert_tail(&rwa
->write_batch
, rrd
);
2425 * Return EAGAIN to indicate that we will use this rrd again,
2426 * so the caller should not free it
2432 receive_write_embedded(struct receive_writer_arg
*rwa
,
2433 struct drr_write_embedded
*drrwe
, void *data
)
2438 if (drrwe
->drr_offset
+ drrwe
->drr_length
< drrwe
->drr_offset
)
2439 return (SET_ERROR(EINVAL
));
2441 if (drrwe
->drr_psize
> BPE_PAYLOAD_SIZE
)
2442 return (SET_ERROR(EINVAL
));
2444 if (drrwe
->drr_etype
>= NUM_BP_EMBEDDED_TYPES
)
2445 return (SET_ERROR(EINVAL
));
2446 if (drrwe
->drr_compression
>= ZIO_COMPRESS_FUNCTIONS
)
2447 return (SET_ERROR(EINVAL
));
2449 return (SET_ERROR(EINVAL
));
2451 if (drrwe
->drr_object
> rwa
->max_object
)
2452 rwa
->max_object
= drrwe
->drr_object
;
2454 tx
= dmu_tx_create(rwa
->os
);
2456 dmu_tx_hold_write(tx
, drrwe
->drr_object
,
2457 drrwe
->drr_offset
, drrwe
->drr_length
);
2458 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2464 dmu_write_embedded(rwa
->os
, drrwe
->drr_object
,
2465 drrwe
->drr_offset
, data
, drrwe
->drr_etype
,
2466 drrwe
->drr_compression
, drrwe
->drr_lsize
, drrwe
->drr_psize
,
2467 rwa
->byteswap
^ ZFS_HOST_BYTEORDER
, tx
);
2469 /* See comment in restore_write. */
2470 save_resume_state(rwa
, drrwe
->drr_object
, drrwe
->drr_offset
, tx
);
2476 receive_spill(struct receive_writer_arg
*rwa
, struct drr_spill
*drrs
,
2479 dmu_buf_t
*db
, *db_spill
;
2482 if (drrs
->drr_length
< SPA_MINBLOCKSIZE
||
2483 drrs
->drr_length
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)))
2484 return (SET_ERROR(EINVAL
));
2487 * This is an unmodified spill block which was added to the stream
2488 * to resolve an issue with incorrectly removing spill blocks. It
2489 * should be ignored by current versions of the code which support
2490 * the DRR_FLAG_SPILL_BLOCK flag.
2492 if (rwa
->spill
&& DRR_SPILL_IS_UNMODIFIED(drrs
->drr_flags
)) {
2498 if (!DMU_OT_IS_VALID(drrs
->drr_type
) ||
2499 drrs
->drr_compressiontype
>= ZIO_COMPRESS_FUNCTIONS
||
2500 drrs
->drr_compressed_size
== 0)
2501 return (SET_ERROR(EINVAL
));
2504 if (dmu_object_info(rwa
->os
, drrs
->drr_object
, NULL
) != 0)
2505 return (SET_ERROR(EINVAL
));
2507 if (drrs
->drr_object
> rwa
->max_object
)
2508 rwa
->max_object
= drrs
->drr_object
;
2510 VERIFY0(dmu_bonus_hold(rwa
->os
, drrs
->drr_object
, FTAG
, &db
));
2511 if ((err
= dmu_spill_hold_by_bonus(db
, DMU_READ_NO_DECRYPT
, FTAG
,
2513 dmu_buf_rele(db
, FTAG
);
2517 dmu_tx_t
*tx
= dmu_tx_create(rwa
->os
);
2519 dmu_tx_hold_spill(tx
, db
->db_object
);
2521 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2523 dmu_buf_rele(db
, FTAG
);
2524 dmu_buf_rele(db_spill
, FTAG
);
2530 * Spill blocks may both grow and shrink. When a change in size
2531 * occurs any existing dbuf must be updated to match the logical
2532 * size of the provided arc_buf_t.
2534 if (db_spill
->db_size
!= drrs
->drr_length
) {
2535 dmu_buf_will_fill(db_spill
, tx
, B_FALSE
);
2536 VERIFY0(dbuf_spill_set_blksz(db_spill
,
2537 drrs
->drr_length
, tx
));
2542 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^
2543 !!DRR_IS_RAW_BYTESWAPPED(drrs
->drr_flags
) ^
2546 abuf
= arc_loan_raw_buf(dmu_objset_spa(rwa
->os
),
2547 drrs
->drr_object
, byteorder
, drrs
->drr_salt
,
2548 drrs
->drr_iv
, drrs
->drr_mac
, drrs
->drr_type
,
2549 drrs
->drr_compressed_size
, drrs
->drr_length
,
2550 drrs
->drr_compressiontype
, 0);
2552 abuf
= arc_loan_buf(dmu_objset_spa(rwa
->os
),
2553 DMU_OT_IS_METADATA(drrs
->drr_type
),
2555 if (rwa
->byteswap
) {
2556 dmu_object_byteswap_t byteswap
=
2557 DMU_OT_BYTESWAP(drrs
->drr_type
);
2558 dmu_ot_byteswap
[byteswap
].ob_func(abd_to_buf(abd
),
2559 DRR_SPILL_PAYLOAD_SIZE(drrs
));
2563 memcpy(abuf
->b_data
, abd_to_buf(abd
), DRR_SPILL_PAYLOAD_SIZE(drrs
));
2565 dbuf_assign_arcbuf((dmu_buf_impl_t
*)db_spill
, abuf
, tx
);
2567 dmu_buf_rele(db
, FTAG
);
2568 dmu_buf_rele(db_spill
, FTAG
);
2575 receive_free(struct receive_writer_arg
*rwa
, struct drr_free
*drrf
)
2579 if (drrf
->drr_length
!= -1ULL &&
2580 drrf
->drr_offset
+ drrf
->drr_length
< drrf
->drr_offset
)
2581 return (SET_ERROR(EINVAL
));
2583 if (dmu_object_info(rwa
->os
, drrf
->drr_object
, NULL
) != 0)
2584 return (SET_ERROR(EINVAL
));
2586 if (drrf
->drr_object
> rwa
->max_object
)
2587 rwa
->max_object
= drrf
->drr_object
;
2589 err
= dmu_free_long_range(rwa
->os
, drrf
->drr_object
,
2590 drrf
->drr_offset
, drrf
->drr_length
);
2596 receive_object_range(struct receive_writer_arg
*rwa
,
2597 struct drr_object_range
*drror
)
2600 * By default, we assume this block is in our native format
2601 * (ZFS_HOST_BYTEORDER). We then take into account whether
2602 * the send stream is byteswapped (rwa->byteswap). Finally,
2603 * we need to byteswap again if this particular block was
2604 * in non-native format on the send side.
2606 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^ rwa
->byteswap
^
2607 !!DRR_IS_RAW_BYTESWAPPED(drror
->drr_flags
);
2610 * Since dnode block sizes are constant, we should not need to worry
2611 * about making sure that the dnode block size is the same on the
2612 * sending and receiving sides for the time being. For non-raw sends,
2613 * this does not matter (and in fact we do not send a DRR_OBJECT_RANGE
2614 * record at all). Raw sends require this record type because the
2615 * encryption parameters are used to protect an entire block of bonus
2616 * buffers. If the size of dnode blocks ever becomes variable,
2617 * handling will need to be added to ensure that dnode block sizes
2618 * match on the sending and receiving side.
2620 if (drror
->drr_numslots
!= DNODES_PER_BLOCK
||
2621 P2PHASE(drror
->drr_firstobj
, DNODES_PER_BLOCK
) != 0 ||
2623 return (SET_ERROR(EINVAL
));
2625 if (drror
->drr_firstobj
> rwa
->max_object
)
2626 rwa
->max_object
= drror
->drr_firstobj
;
2629 * The DRR_OBJECT_RANGE handling must be deferred to receive_object()
2630 * so that the block of dnodes is not written out when it's empty,
2631 * and converted to a HOLE BP.
2633 rwa
->or_crypt_params_present
= B_TRUE
;
2634 rwa
->or_firstobj
= drror
->drr_firstobj
;
2635 rwa
->or_numslots
= drror
->drr_numslots
;
2636 memcpy(rwa
->or_salt
, drror
->drr_salt
, ZIO_DATA_SALT_LEN
);
2637 memcpy(rwa
->or_iv
, drror
->drr_iv
, ZIO_DATA_IV_LEN
);
2638 memcpy(rwa
->or_mac
, drror
->drr_mac
, ZIO_DATA_MAC_LEN
);
2639 rwa
->or_byteorder
= byteorder
;
2641 rwa
->or_need_sync
= ORNS_MAYBE
;
2647 * Until we have the ability to redact large ranges of data efficiently, we
2648 * process these records as frees.
2651 receive_redact(struct receive_writer_arg
*rwa
, struct drr_redact
*drrr
)
2653 struct drr_free drrf
= {0};
2654 drrf
.drr_length
= drrr
->drr_length
;
2655 drrf
.drr_object
= drrr
->drr_object
;
2656 drrf
.drr_offset
= drrr
->drr_offset
;
2657 drrf
.drr_toguid
= drrr
->drr_toguid
;
2658 return (receive_free(rwa
, &drrf
));
2661 /* used to destroy the drc_ds on error */
2663 dmu_recv_cleanup_ds(dmu_recv_cookie_t
*drc
)
2665 dsl_dataset_t
*ds
= drc
->drc_ds
;
2666 ds_hold_flags_t dsflags
;
2668 dsflags
= (drc
->drc_raw
) ? DS_HOLD_FLAG_NONE
: DS_HOLD_FLAG_DECRYPT
;
2670 * Wait for the txg sync before cleaning up the receive. For
2671 * resumable receives, this ensures that our resume state has
2672 * been written out to disk. For raw receives, this ensures
2673 * that the user accounting code will not attempt to do anything
2674 * after we stopped receiving the dataset.
2676 txg_wait_synced(ds
->ds_dir
->dd_pool
, 0);
2677 ds
->ds_objset
->os_raw_receive
= B_FALSE
;
2679 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2680 if (drc
->drc_resumable
&& drc
->drc_should_save
&&
2681 !BP_IS_HOLE(dsl_dataset_get_blkptr(ds
))) {
2682 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2683 dsl_dataset_disown(ds
, dsflags
, dmu_recv_tag
);
2685 char name
[ZFS_MAX_DATASET_NAME_LEN
];
2686 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2687 dsl_dataset_name(ds
, name
);
2688 dsl_dataset_disown(ds
, dsflags
, dmu_recv_tag
);
2690 (void) dsl_destroy_head(name
);
2695 receive_cksum(dmu_recv_cookie_t
*drc
, int len
, void *buf
)
2697 if (drc
->drc_byteswap
) {
2698 (void) fletcher_4_incremental_byteswap(buf
, len
,
2701 (void) fletcher_4_incremental_native(buf
, len
, &drc
->drc_cksum
);
2706 * Read the payload into a buffer of size len, and update the current record's
2708 * Allocate drc->drc_next_rrd and read the next record's header into
2709 * drc->drc_next_rrd->header.
2710 * Verify checksum of payload and next record.
2713 receive_read_payload_and_next_header(dmu_recv_cookie_t
*drc
, int len
, void *buf
)
2718 ASSERT3U(len
, <=, SPA_MAXBLOCKSIZE
);
2719 err
= receive_read(drc
, len
, buf
);
2722 receive_cksum(drc
, len
, buf
);
2724 /* note: rrd is NULL when reading the begin record's payload */
2725 if (drc
->drc_rrd
!= NULL
) {
2726 drc
->drc_rrd
->payload
= buf
;
2727 drc
->drc_rrd
->payload_size
= len
;
2728 drc
->drc_rrd
->bytes_read
= drc
->drc_bytes_read
;
2731 ASSERT3P(buf
, ==, NULL
);
2734 drc
->drc_prev_cksum
= drc
->drc_cksum
;
2736 drc
->drc_next_rrd
= kmem_zalloc(sizeof (*drc
->drc_next_rrd
), KM_SLEEP
);
2737 err
= receive_read(drc
, sizeof (drc
->drc_next_rrd
->header
),
2738 &drc
->drc_next_rrd
->header
);
2739 drc
->drc_next_rrd
->bytes_read
= drc
->drc_bytes_read
;
2742 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
2743 drc
->drc_next_rrd
= NULL
;
2746 if (drc
->drc_next_rrd
->header
.drr_type
== DRR_BEGIN
) {
2747 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
2748 drc
->drc_next_rrd
= NULL
;
2749 return (SET_ERROR(EINVAL
));
2753 * Note: checksum is of everything up to but not including the
2756 ASSERT3U(offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2757 ==, sizeof (dmu_replay_record_t
) - sizeof (zio_cksum_t
));
2759 offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2760 &drc
->drc_next_rrd
->header
);
2762 zio_cksum_t cksum_orig
=
2763 drc
->drc_next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
2764 zio_cksum_t
*cksump
=
2765 &drc
->drc_next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
2767 if (drc
->drc_byteswap
)
2768 byteswap_record(&drc
->drc_next_rrd
->header
);
2770 if ((!ZIO_CHECKSUM_IS_ZERO(cksump
)) &&
2771 !ZIO_CHECKSUM_EQUAL(drc
->drc_cksum
, *cksump
)) {
2772 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
2773 drc
->drc_next_rrd
= NULL
;
2774 return (SET_ERROR(ECKSUM
));
2777 receive_cksum(drc
, sizeof (cksum_orig
), &cksum_orig
);
2783 * Issue the prefetch reads for any necessary indirect blocks.
2785 * We use the object ignore list to tell us whether or not to issue prefetches
2786 * for a given object. We do this for both correctness (in case the blocksize
2787 * of an object has changed) and performance (if the object doesn't exist, don't
2788 * needlessly try to issue prefetches). We also trim the list as we go through
2789 * the stream to prevent it from growing to an unbounded size.
2791 * The object numbers within will always be in sorted order, and any write
2792 * records we see will also be in sorted order, but they're not sorted with
2793 * respect to each other (i.e. we can get several object records before
2794 * receiving each object's write records). As a result, once we've reached a
2795 * given object number, we can safely remove any reference to lower object
2796 * numbers in the ignore list. In practice, we receive up to 32 object records
2797 * before receiving write records, so the list can have up to 32 nodes in it.
2800 receive_read_prefetch(dmu_recv_cookie_t
*drc
, uint64_t object
, uint64_t offset
,
2803 if (!objlist_exists(drc
->drc_ignore_objlist
, object
)) {
2804 dmu_prefetch(drc
->drc_os
, object
, 1, offset
, length
,
2805 ZIO_PRIORITY_SYNC_READ
);
2810 * Read records off the stream, issuing any necessary prefetches.
2813 receive_read_record(dmu_recv_cookie_t
*drc
)
2817 switch (drc
->drc_rrd
->header
.drr_type
) {
2820 struct drr_object
*drro
=
2821 &drc
->drc_rrd
->header
.drr_u
.drr_object
;
2822 uint32_t size
= DRR_OBJECT_PAYLOAD_SIZE(drro
);
2824 dmu_object_info_t doi
;
2827 buf
= kmem_zalloc(size
, KM_SLEEP
);
2829 err
= receive_read_payload_and_next_header(drc
, size
, buf
);
2831 kmem_free(buf
, size
);
2834 err
= dmu_object_info(drc
->drc_os
, drro
->drr_object
, &doi
);
2836 * See receive_read_prefetch for an explanation why we're
2837 * storing this object in the ignore_obj_list.
2839 if (err
== ENOENT
|| err
== EEXIST
||
2840 (err
== 0 && doi
.doi_data_block_size
!= drro
->drr_blksz
)) {
2841 objlist_insert(drc
->drc_ignore_objlist
,
2847 case DRR_FREEOBJECTS
:
2849 err
= receive_read_payload_and_next_header(drc
, 0, NULL
);
2854 struct drr_write
*drrw
= &drc
->drc_rrd
->header
.drr_u
.drr_write
;
2855 int size
= DRR_WRITE_PAYLOAD_SIZE(drrw
);
2856 abd_t
*abd
= abd_alloc_linear(size
, B_FALSE
);
2857 err
= receive_read_payload_and_next_header(drc
, size
,
2863 drc
->drc_rrd
->abd
= abd
;
2864 receive_read_prefetch(drc
, drrw
->drr_object
, drrw
->drr_offset
,
2865 drrw
->drr_logical_size
);
2868 case DRR_WRITE_EMBEDDED
:
2870 struct drr_write_embedded
*drrwe
=
2871 &drc
->drc_rrd
->header
.drr_u
.drr_write_embedded
;
2872 uint32_t size
= P2ROUNDUP(drrwe
->drr_psize
, 8);
2873 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
2875 err
= receive_read_payload_and_next_header(drc
, size
, buf
);
2877 kmem_free(buf
, size
);
2881 receive_read_prefetch(drc
, drrwe
->drr_object
, drrwe
->drr_offset
,
2889 * It might be beneficial to prefetch indirect blocks here, but
2890 * we don't really have the data to decide for sure.
2892 err
= receive_read_payload_and_next_header(drc
, 0, NULL
);
2897 struct drr_end
*drre
= &drc
->drc_rrd
->header
.drr_u
.drr_end
;
2898 if (!ZIO_CHECKSUM_EQUAL(drc
->drc_prev_cksum
,
2899 drre
->drr_checksum
))
2900 return (SET_ERROR(ECKSUM
));
2905 struct drr_spill
*drrs
= &drc
->drc_rrd
->header
.drr_u
.drr_spill
;
2906 int size
= DRR_SPILL_PAYLOAD_SIZE(drrs
);
2907 abd_t
*abd
= abd_alloc_linear(size
, B_FALSE
);
2908 err
= receive_read_payload_and_next_header(drc
, size
,
2913 drc
->drc_rrd
->abd
= abd
;
2916 case DRR_OBJECT_RANGE
:
2918 err
= receive_read_payload_and_next_header(drc
, 0, NULL
);
2923 return (SET_ERROR(EINVAL
));
2930 dprintf_drr(struct receive_record_arg
*rrd
, int err
)
2933 switch (rrd
->header
.drr_type
) {
2936 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
2937 dprintf("drr_type = OBJECT obj = %llu type = %u "
2938 "bonustype = %u blksz = %u bonuslen = %u cksumtype = %u "
2939 "compress = %u dn_slots = %u err = %d\n",
2940 (u_longlong_t
)drro
->drr_object
, drro
->drr_type
,
2941 drro
->drr_bonustype
, drro
->drr_blksz
, drro
->drr_bonuslen
,
2942 drro
->drr_checksumtype
, drro
->drr_compress
,
2943 drro
->drr_dn_slots
, err
);
2946 case DRR_FREEOBJECTS
:
2948 struct drr_freeobjects
*drrfo
=
2949 &rrd
->header
.drr_u
.drr_freeobjects
;
2950 dprintf("drr_type = FREEOBJECTS firstobj = %llu "
2951 "numobjs = %llu err = %d\n",
2952 (u_longlong_t
)drrfo
->drr_firstobj
,
2953 (u_longlong_t
)drrfo
->drr_numobjs
, err
);
2958 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
2959 dprintf("drr_type = WRITE obj = %llu type = %u offset = %llu "
2960 "lsize = %llu cksumtype = %u flags = %u "
2961 "compress = %u psize = %llu err = %d\n",
2962 (u_longlong_t
)drrw
->drr_object
, drrw
->drr_type
,
2963 (u_longlong_t
)drrw
->drr_offset
,
2964 (u_longlong_t
)drrw
->drr_logical_size
,
2965 drrw
->drr_checksumtype
, drrw
->drr_flags
,
2966 drrw
->drr_compressiontype
,
2967 (u_longlong_t
)drrw
->drr_compressed_size
, err
);
2970 case DRR_WRITE_BYREF
:
2972 struct drr_write_byref
*drrwbr
=
2973 &rrd
->header
.drr_u
.drr_write_byref
;
2974 dprintf("drr_type = WRITE_BYREF obj = %llu offset = %llu "
2975 "length = %llu toguid = %llx refguid = %llx "
2976 "refobject = %llu refoffset = %llu cksumtype = %u "
2977 "flags = %u err = %d\n",
2978 (u_longlong_t
)drrwbr
->drr_object
,
2979 (u_longlong_t
)drrwbr
->drr_offset
,
2980 (u_longlong_t
)drrwbr
->drr_length
,
2981 (u_longlong_t
)drrwbr
->drr_toguid
,
2982 (u_longlong_t
)drrwbr
->drr_refguid
,
2983 (u_longlong_t
)drrwbr
->drr_refobject
,
2984 (u_longlong_t
)drrwbr
->drr_refoffset
,
2985 drrwbr
->drr_checksumtype
, drrwbr
->drr_flags
, err
);
2988 case DRR_WRITE_EMBEDDED
:
2990 struct drr_write_embedded
*drrwe
=
2991 &rrd
->header
.drr_u
.drr_write_embedded
;
2992 dprintf("drr_type = WRITE_EMBEDDED obj = %llu offset = %llu "
2993 "length = %llu compress = %u etype = %u lsize = %u "
2994 "psize = %u err = %d\n",
2995 (u_longlong_t
)drrwe
->drr_object
,
2996 (u_longlong_t
)drrwe
->drr_offset
,
2997 (u_longlong_t
)drrwe
->drr_length
,
2998 drrwe
->drr_compression
, drrwe
->drr_etype
,
2999 drrwe
->drr_lsize
, drrwe
->drr_psize
, err
);
3004 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
3005 dprintf("drr_type = FREE obj = %llu offset = %llu "
3006 "length = %lld err = %d\n",
3007 (u_longlong_t
)drrf
->drr_object
,
3008 (u_longlong_t
)drrf
->drr_offset
,
3009 (longlong_t
)drrf
->drr_length
,
3015 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
3016 dprintf("drr_type = SPILL obj = %llu length = %llu "
3017 "err = %d\n", (u_longlong_t
)drrs
->drr_object
,
3018 (u_longlong_t
)drrs
->drr_length
, err
);
3021 case DRR_OBJECT_RANGE
:
3023 struct drr_object_range
*drror
=
3024 &rrd
->header
.drr_u
.drr_object_range
;
3025 dprintf("drr_type = OBJECT_RANGE firstobj = %llu "
3026 "numslots = %llu flags = %u err = %d\n",
3027 (u_longlong_t
)drror
->drr_firstobj
,
3028 (u_longlong_t
)drror
->drr_numslots
,
3029 drror
->drr_flags
, err
);
3039 * Commit the records to the pool.
3042 receive_process_record(struct receive_writer_arg
*rwa
,
3043 struct receive_record_arg
*rrd
)
3047 /* Processing in order, therefore bytes_read should be increasing. */
3048 ASSERT3U(rrd
->bytes_read
, >=, rwa
->bytes_read
);
3049 rwa
->bytes_read
= rrd
->bytes_read
;
3051 /* We can only heal write records; other ones get ignored */
3052 if (rwa
->heal
&& rrd
->header
.drr_type
!= DRR_WRITE
) {
3053 if (rrd
->abd
!= NULL
) {
3056 } else if (rrd
->payload
!= NULL
) {
3057 kmem_free(rrd
->payload
, rrd
->payload_size
);
3058 rrd
->payload
= NULL
;
3063 if (!rwa
->heal
&& rrd
->header
.drr_type
!= DRR_WRITE
) {
3064 err
= flush_write_batch(rwa
);
3066 if (rrd
->abd
!= NULL
) {
3069 rrd
->payload
= NULL
;
3070 } else if (rrd
->payload
!= NULL
) {
3071 kmem_free(rrd
->payload
, rrd
->payload_size
);
3072 rrd
->payload
= NULL
;
3079 switch (rrd
->header
.drr_type
) {
3082 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
3083 err
= receive_object(rwa
, drro
, rrd
->payload
);
3084 kmem_free(rrd
->payload
, rrd
->payload_size
);
3085 rrd
->payload
= NULL
;
3088 case DRR_FREEOBJECTS
:
3090 struct drr_freeobjects
*drrfo
=
3091 &rrd
->header
.drr_u
.drr_freeobjects
;
3092 err
= receive_freeobjects(rwa
, drrfo
);
3097 err
= receive_process_write_record(rwa
, rrd
);
3100 * If healing - always free the abd after processing
3104 } else if (err
!= EAGAIN
) {
3106 * On success, a non-healing
3107 * receive_process_write_record() returns
3108 * EAGAIN to indicate that we do not want to free
3109 * the rrd or arc_buf.
3117 case DRR_WRITE_EMBEDDED
:
3119 struct drr_write_embedded
*drrwe
=
3120 &rrd
->header
.drr_u
.drr_write_embedded
;
3121 err
= receive_write_embedded(rwa
, drrwe
, rrd
->payload
);
3122 kmem_free(rrd
->payload
, rrd
->payload_size
);
3123 rrd
->payload
= NULL
;
3128 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
3129 err
= receive_free(rwa
, drrf
);
3134 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
3135 err
= receive_spill(rwa
, drrs
, rrd
->abd
);
3139 rrd
->payload
= NULL
;
3142 case DRR_OBJECT_RANGE
:
3144 struct drr_object_range
*drror
=
3145 &rrd
->header
.drr_u
.drr_object_range
;
3146 err
= receive_object_range(rwa
, drror
);
3151 struct drr_redact
*drrr
= &rrd
->header
.drr_u
.drr_redact
;
3152 err
= receive_redact(rwa
, drrr
);
3156 err
= (SET_ERROR(EINVAL
));
3160 dprintf_drr(rrd
, err
);
3166 * dmu_recv_stream's worker thread; pull records off the queue, and then call
3167 * receive_process_record When we're done, signal the main thread and exit.
3169 static __attribute__((noreturn
)) void
3170 receive_writer_thread(void *arg
)
3172 struct receive_writer_arg
*rwa
= arg
;
3173 struct receive_record_arg
*rrd
;
3174 fstrans_cookie_t cookie
= spl_fstrans_mark();
3176 for (rrd
= bqueue_dequeue(&rwa
->q
); !rrd
->eos_marker
;
3177 rrd
= bqueue_dequeue(&rwa
->q
)) {
3179 * If there's an error, the main thread will stop putting things
3180 * on the queue, but we need to clear everything in it before we
3184 if (rwa
->err
== 0) {
3185 err
= receive_process_record(rwa
, rrd
);
3186 } else if (rrd
->abd
!= NULL
) {
3189 rrd
->payload
= NULL
;
3190 } else if (rrd
->payload
!= NULL
) {
3191 kmem_free(rrd
->payload
, rrd
->payload_size
);
3192 rrd
->payload
= NULL
;
3195 * EAGAIN indicates that this record has been saved (on
3196 * raw->write_batch), and will be used again, so we don't
3198 * When healing data we always need to free the record.
3200 if (err
!= EAGAIN
|| rwa
->heal
) {
3203 kmem_free(rrd
, sizeof (*rrd
));
3206 kmem_free(rrd
, sizeof (*rrd
));
3209 zio_wait(rwa
->heal_pio
);
3211 int err
= flush_write_batch(rwa
);
3215 mutex_enter(&rwa
->mutex
);
3217 cv_signal(&rwa
->cv
);
3218 mutex_exit(&rwa
->mutex
);
3219 spl_fstrans_unmark(cookie
);
3224 resume_check(dmu_recv_cookie_t
*drc
, nvlist_t
*begin_nvl
)
3227 objset_t
*mos
= dmu_objset_pool(drc
->drc_os
)->dp_meta_objset
;
3228 uint64_t dsobj
= dmu_objset_id(drc
->drc_os
);
3229 uint64_t resume_obj
, resume_off
;
3231 if (nvlist_lookup_uint64(begin_nvl
,
3232 "resume_object", &resume_obj
) != 0 ||
3233 nvlist_lookup_uint64(begin_nvl
,
3234 "resume_offset", &resume_off
) != 0) {
3235 return (SET_ERROR(EINVAL
));
3237 VERIFY0(zap_lookup(mos
, dsobj
,
3238 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
));
3239 if (resume_obj
!= val
)
3240 return (SET_ERROR(EINVAL
));
3241 VERIFY0(zap_lookup(mos
, dsobj
,
3242 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
));
3243 if (resume_off
!= val
)
3244 return (SET_ERROR(EINVAL
));
3250 * Read in the stream's records, one by one, and apply them to the pool. There
3251 * are two threads involved; the thread that calls this function will spin up a
3252 * worker thread, read the records off the stream one by one, and issue
3253 * prefetches for any necessary indirect blocks. It will then push the records
3254 * onto an internal blocking queue. The worker thread will pull the records off
3255 * the queue, and actually write the data into the DMU. This way, the worker
3256 * thread doesn't have to wait for reads to complete, since everything it needs
3257 * (the indirect blocks) will be prefetched.
3259 * NB: callers *must* call dmu_recv_end() if this succeeds.
3262 dmu_recv_stream(dmu_recv_cookie_t
*drc
, offset_t
*voffp
)
3265 struct receive_writer_arg
*rwa
= kmem_zalloc(sizeof (*rwa
), KM_SLEEP
);
3267 if (dsl_dataset_has_resume_receive_state(drc
->drc_ds
)) {
3269 (void) zap_lookup(drc
->drc_ds
->ds_dir
->dd_pool
->dp_meta_objset
,
3270 drc
->drc_ds
->ds_object
, DS_FIELD_RESUME_BYTES
,
3271 sizeof (bytes
), 1, &bytes
);
3272 drc
->drc_bytes_read
+= bytes
;
3275 drc
->drc_ignore_objlist
= objlist_create();
3277 /* these were verified in dmu_recv_begin */
3278 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc
->drc_drrb
->drr_versioninfo
), ==,
3280 ASSERT3U(drc
->drc_drrb
->drr_type
, <, DMU_OST_NUMTYPES
);
3282 ASSERT(dsl_dataset_phys(drc
->drc_ds
)->ds_flags
& DS_FLAG_INCONSISTENT
);
3283 ASSERT0(drc
->drc_os
->os_encrypted
&&
3284 (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
));
3286 /* handle DSL encryption key payload */
3287 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
3288 nvlist_t
*keynvl
= NULL
;
3290 ASSERT(drc
->drc_os
->os_encrypted
);
3291 ASSERT(drc
->drc_raw
);
3293 err
= nvlist_lookup_nvlist(drc
->drc_begin_nvl
, "crypt_keydata",
3298 if (!drc
->drc_heal
) {
3300 * If this is a new dataset we set the key immediately.
3301 * Otherwise we don't want to change the key until we
3302 * are sure the rest of the receive succeeded so we
3303 * stash the keynvl away until then.
3305 err
= dsl_crypto_recv_raw(spa_name(drc
->drc_os
->os_spa
),
3306 drc
->drc_ds
->ds_object
, drc
->drc_fromsnapobj
,
3307 drc
->drc_drrb
->drr_type
, keynvl
, drc
->drc_newfs
);
3312 /* see comment in dmu_recv_end_sync() */
3313 drc
->drc_ivset_guid
= 0;
3314 (void) nvlist_lookup_uint64(keynvl
, "to_ivset_guid",
3315 &drc
->drc_ivset_guid
);
3317 if (!drc
->drc_newfs
)
3318 drc
->drc_keynvl
= fnvlist_dup(keynvl
);
3321 if (drc
->drc_featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
3322 err
= resume_check(drc
, drc
->drc_begin_nvl
);
3328 * For compatibility with recursive send streams, we do this here,
3329 * rather than in dmu_recv_begin. If we pull the next header too
3330 * early, and it's the END record, we break the `recv_skip` logic.
3332 if (drc
->drc_drr_begin
->drr_payloadlen
== 0) {
3333 err
= receive_read_payload_and_next_header(drc
, 0, NULL
);
3339 * If we failed before this point we will clean up any new resume
3340 * state that was created. Now that we've gotten past the initial
3341 * checks we are ok to retain that resume state.
3343 drc
->drc_should_save
= B_TRUE
;
3345 (void) bqueue_init(&rwa
->q
, zfs_recv_queue_ff
,
3346 MAX(zfs_recv_queue_length
, 2 * zfs_max_recordsize
),
3347 offsetof(struct receive_record_arg
, node
));
3348 cv_init(&rwa
->cv
, NULL
, CV_DEFAULT
, NULL
);
3349 mutex_init(&rwa
->mutex
, NULL
, MUTEX_DEFAULT
, NULL
);
3350 rwa
->os
= drc
->drc_os
;
3351 rwa
->byteswap
= drc
->drc_byteswap
;
3352 rwa
->heal
= drc
->drc_heal
;
3353 rwa
->tofs
= drc
->drc_tofs
;
3354 rwa
->resumable
= drc
->drc_resumable
;
3355 rwa
->raw
= drc
->drc_raw
;
3356 rwa
->spill
= drc
->drc_spill
;
3357 rwa
->full
= (drc
->drc_drr_begin
->drr_u
.drr_begin
.drr_fromguid
== 0);
3358 rwa
->os
->os_raw_receive
= drc
->drc_raw
;
3359 if (drc
->drc_heal
) {
3360 rwa
->heal_pio
= zio_root(drc
->drc_os
->os_spa
, NULL
, NULL
,
3361 ZIO_FLAG_GODFATHER
);
3363 list_create(&rwa
->write_batch
, sizeof (struct receive_record_arg
),
3364 offsetof(struct receive_record_arg
, node
.bqn_node
));
3366 (void) thread_create(NULL
, 0, receive_writer_thread
, rwa
, 0, curproc
,
3367 TS_RUN
, minclsyspri
);
3369 * We're reading rwa->err without locks, which is safe since we are the
3370 * only reader, and the worker thread is the only writer. It's ok if we
3371 * miss a write for an iteration or two of the loop, since the writer
3372 * thread will keep freeing records we send it until we send it an eos
3375 * We can leave this loop in 3 ways: First, if rwa->err is
3376 * non-zero. In that case, the writer thread will free the rrd we just
3377 * pushed. Second, if we're interrupted; in that case, either it's the
3378 * first loop and drc->drc_rrd was never allocated, or it's later, and
3379 * drc->drc_rrd has been handed off to the writer thread who will free
3380 * it. Finally, if receive_read_record fails or we're at the end of the
3381 * stream, then we free drc->drc_rrd and exit.
3383 while (rwa
->err
== 0) {
3384 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
3385 err
= SET_ERROR(EINTR
);
3389 ASSERT3P(drc
->drc_rrd
, ==, NULL
);
3390 drc
->drc_rrd
= drc
->drc_next_rrd
;
3391 drc
->drc_next_rrd
= NULL
;
3392 /* Allocates and loads header into drc->drc_next_rrd */
3393 err
= receive_read_record(drc
);
3395 if (drc
->drc_rrd
->header
.drr_type
== DRR_END
|| err
!= 0) {
3396 kmem_free(drc
->drc_rrd
, sizeof (*drc
->drc_rrd
));
3397 drc
->drc_rrd
= NULL
;
3401 bqueue_enqueue(&rwa
->q
, drc
->drc_rrd
,
3402 sizeof (struct receive_record_arg
) +
3403 drc
->drc_rrd
->payload_size
);
3404 drc
->drc_rrd
= NULL
;
3407 ASSERT3P(drc
->drc_rrd
, ==, NULL
);
3408 drc
->drc_rrd
= kmem_zalloc(sizeof (*drc
->drc_rrd
), KM_SLEEP
);
3409 drc
->drc_rrd
->eos_marker
= B_TRUE
;
3410 bqueue_enqueue_flush(&rwa
->q
, drc
->drc_rrd
, 1);
3412 mutex_enter(&rwa
->mutex
);
3413 while (!rwa
->done
) {
3415 * We need to use cv_wait_sig() so that any process that may
3416 * be sleeping here can still fork.
3418 (void) cv_wait_sig(&rwa
->cv
, &rwa
->mutex
);
3420 mutex_exit(&rwa
->mutex
);
3423 * If we are receiving a full stream as a clone, all object IDs which
3424 * are greater than the maximum ID referenced in the stream are
3425 * by definition unused and must be freed.
3427 if (drc
->drc_clone
&& drc
->drc_drrb
->drr_fromguid
== 0) {
3428 uint64_t obj
= rwa
->max_object
+ 1;
3432 while (next_err
== 0) {
3433 free_err
= dmu_free_long_object(rwa
->os
, obj
);
3434 if (free_err
!= 0 && free_err
!= ENOENT
)
3437 next_err
= dmu_object_next(rwa
->os
, &obj
, FALSE
, 0);
3441 if (free_err
!= 0 && free_err
!= ENOENT
)
3443 else if (next_err
!= ESRCH
)
3448 cv_destroy(&rwa
->cv
);
3449 mutex_destroy(&rwa
->mutex
);
3450 bqueue_destroy(&rwa
->q
);
3451 list_destroy(&rwa
->write_batch
);
3457 * If we hit an error before we started the receive_writer_thread
3458 * we need to clean up the next_rrd we create by processing the
3461 if (drc
->drc_next_rrd
!= NULL
)
3462 kmem_free(drc
->drc_next_rrd
, sizeof (*drc
->drc_next_rrd
));
3465 * The objset will be invalidated by dmu_recv_end() when we do
3466 * dsl_dataset_clone_swap_sync_impl().
3470 kmem_free(rwa
, sizeof (*rwa
));
3471 nvlist_free(drc
->drc_begin_nvl
);
3475 * Clean up references. If receive is not resumable,
3476 * destroy what we created, so we don't leave it in
3477 * the inconsistent state.
3479 dmu_recv_cleanup_ds(drc
);
3480 nvlist_free(drc
->drc_keynvl
);
3483 objlist_destroy(drc
->drc_ignore_objlist
);
3484 drc
->drc_ignore_objlist
= NULL
;
3485 *voffp
= drc
->drc_voff
;
3490 dmu_recv_end_check(void *arg
, dmu_tx_t
*tx
)
3492 dmu_recv_cookie_t
*drc
= arg
;
3493 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3496 ASSERT3P(drc
->drc_ds
->ds_owner
, ==, dmu_recv_tag
);
3498 if (drc
->drc_heal
) {
3500 } else if (!drc
->drc_newfs
) {
3501 dsl_dataset_t
*origin_head
;
3503 error
= dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
, &origin_head
);
3506 if (drc
->drc_force
) {
3508 * We will destroy any snapshots in tofs (i.e. before
3509 * origin_head) that are after the origin (which is
3510 * the snap before drc_ds, because drc_ds can not
3511 * have any snaps of its own).
3515 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3517 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3518 dsl_dataset_t
*snap
;
3519 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3523 if (snap
->ds_dir
!= origin_head
->ds_dir
)
3524 error
= SET_ERROR(EINVAL
);
3526 error
= dsl_destroy_snapshot_check_impl(
3529 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3530 dsl_dataset_rele(snap
, FTAG
);
3535 dsl_dataset_rele(origin_head
, FTAG
);
3539 if (drc
->drc_keynvl
!= NULL
) {
3540 error
= dsl_crypto_recv_raw_key_check(drc
->drc_ds
,
3541 drc
->drc_keynvl
, tx
);
3543 dsl_dataset_rele(origin_head
, FTAG
);
3548 error
= dsl_dataset_clone_swap_check_impl(drc
->drc_ds
,
3549 origin_head
, drc
->drc_force
, drc
->drc_owner
, tx
);
3551 dsl_dataset_rele(origin_head
, FTAG
);
3554 error
= dsl_dataset_snapshot_check_impl(origin_head
,
3555 drc
->drc_tosnap
, tx
, B_TRUE
, 1,
3556 drc
->drc_cred
, drc
->drc_proc
);
3557 dsl_dataset_rele(origin_head
, FTAG
);
3561 error
= dsl_destroy_head_check_impl(drc
->drc_ds
, 1);
3563 error
= dsl_dataset_snapshot_check_impl(drc
->drc_ds
,
3564 drc
->drc_tosnap
, tx
, B_TRUE
, 1,
3565 drc
->drc_cred
, drc
->drc_proc
);
3571 dmu_recv_end_sync(void *arg
, dmu_tx_t
*tx
)
3573 dmu_recv_cookie_t
*drc
= arg
;
3574 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3575 boolean_t encrypted
= drc
->drc_ds
->ds_dir
->dd_crypto_obj
!= 0;
3576 uint64_t newsnapobj
= 0;
3578 spa_history_log_internal_ds(drc
->drc_ds
, "finish receiving",
3579 tx
, "snap=%s", drc
->drc_tosnap
);
3580 drc
->drc_ds
->ds_objset
->os_raw_receive
= B_FALSE
;
3582 if (drc
->drc_heal
) {
3583 if (drc
->drc_keynvl
!= NULL
) {
3584 nvlist_free(drc
->drc_keynvl
);
3585 drc
->drc_keynvl
= NULL
;
3587 } else if (!drc
->drc_newfs
) {
3588 dsl_dataset_t
*origin_head
;
3590 VERIFY0(dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
,
3593 if (drc
->drc_force
) {
3595 * Destroy any snapshots of drc_tofs (origin_head)
3596 * after the origin (the snap before drc_ds).
3600 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3602 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3603 dsl_dataset_t
*snap
;
3604 VERIFY0(dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3606 ASSERT3P(snap
->ds_dir
, ==, origin_head
->ds_dir
);
3607 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3608 dsl_destroy_snapshot_sync_impl(snap
,
3610 dsl_dataset_rele(snap
, FTAG
);
3613 if (drc
->drc_keynvl
!= NULL
) {
3614 dsl_crypto_recv_raw_key_sync(drc
->drc_ds
,
3615 drc
->drc_keynvl
, tx
);
3616 nvlist_free(drc
->drc_keynvl
);
3617 drc
->drc_keynvl
= NULL
;
3620 VERIFY3P(drc
->drc_ds
->ds_prev
, ==,
3621 origin_head
->ds_prev
);
3623 dsl_dataset_clone_swap_sync_impl(drc
->drc_ds
,
3626 * The objset was evicted by dsl_dataset_clone_swap_sync_impl,
3627 * so drc_os is no longer valid.
3631 dsl_dataset_snapshot_sync_impl(origin_head
,
3632 drc
->drc_tosnap
, tx
);
3634 /* set snapshot's creation time and guid */
3635 dmu_buf_will_dirty(origin_head
->ds_prev
->ds_dbuf
, tx
);
3636 dsl_dataset_phys(origin_head
->ds_prev
)->ds_creation_time
=
3637 drc
->drc_drrb
->drr_creation_time
;
3638 dsl_dataset_phys(origin_head
->ds_prev
)->ds_guid
=
3639 drc
->drc_drrb
->drr_toguid
;
3640 dsl_dataset_phys(origin_head
->ds_prev
)->ds_flags
&=
3641 ~DS_FLAG_INCONSISTENT
;
3643 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
3644 dsl_dataset_phys(origin_head
)->ds_flags
&=
3645 ~DS_FLAG_INCONSISTENT
;
3648 dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3650 dsl_dataset_rele(origin_head
, FTAG
);
3651 dsl_destroy_head_sync_impl(drc
->drc_ds
, tx
);
3653 if (drc
->drc_owner
!= NULL
)
3654 VERIFY3P(origin_head
->ds_owner
, ==, drc
->drc_owner
);
3656 dsl_dataset_t
*ds
= drc
->drc_ds
;
3658 dsl_dataset_snapshot_sync_impl(ds
, drc
->drc_tosnap
, tx
);
3660 /* set snapshot's creation time and guid */
3661 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
3662 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_time
=
3663 drc
->drc_drrb
->drr_creation_time
;
3664 dsl_dataset_phys(ds
->ds_prev
)->ds_guid
=
3665 drc
->drc_drrb
->drr_toguid
;
3666 dsl_dataset_phys(ds
->ds_prev
)->ds_flags
&=
3667 ~DS_FLAG_INCONSISTENT
;
3669 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3670 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
3671 if (dsl_dataset_has_resume_receive_state(ds
)) {
3672 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3673 DS_FIELD_RESUME_FROMGUID
, tx
);
3674 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3675 DS_FIELD_RESUME_OBJECT
, tx
);
3676 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3677 DS_FIELD_RESUME_OFFSET
, tx
);
3678 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3679 DS_FIELD_RESUME_BYTES
, tx
);
3680 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3681 DS_FIELD_RESUME_TOGUID
, tx
);
3682 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3683 DS_FIELD_RESUME_TONAME
, tx
);
3684 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3685 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS
, tx
);
3688 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
;
3692 * If this is a raw receive, the crypt_keydata nvlist will include
3693 * a to_ivset_guid for us to set on the new snapshot. This value
3694 * will override the value generated by the snapshot code. However,
3695 * this value may not be present, because older implementations of
3696 * the raw send code did not include this value, and we are still
3697 * allowed to receive them if the zfs_disable_ivset_guid_check
3698 * tunable is set, in which case we will leave the newly-generated
3701 if (!drc
->drc_heal
&& drc
->drc_raw
&& drc
->drc_ivset_guid
!= 0) {
3702 dmu_object_zapify(dp
->dp_meta_objset
, newsnapobj
,
3703 DMU_OT_DSL_DATASET
, tx
);
3704 VERIFY0(zap_update(dp
->dp_meta_objset
, newsnapobj
,
3705 DS_FIELD_IVSET_GUID
, sizeof (uint64_t), 1,
3706 &drc
->drc_ivset_guid
, tx
));
3710 * Release the hold from dmu_recv_begin. This must be done before
3711 * we return to open context, so that when we free the dataset's dnode
3712 * we can evict its bonus buffer. Since the dataset may be destroyed
3713 * at this point (and therefore won't have a valid pointer to the spa)
3714 * we release the key mapping manually here while we do have a valid
3715 * pointer, if it exists.
3717 if (!drc
->drc_raw
&& encrypted
) {
3718 (void) spa_keystore_remove_mapping(dmu_tx_pool(tx
)->dp_spa
,
3719 drc
->drc_ds
->ds_object
, drc
->drc_ds
);
3721 dsl_dataset_disown(drc
->drc_ds
, 0, dmu_recv_tag
);
3725 static int dmu_recv_end_modified_blocks
= 3;
3728 dmu_recv_existing_end(dmu_recv_cookie_t
*drc
)
3732 * We will be destroying the ds; make sure its origin is unmounted if
3735 char name
[ZFS_MAX_DATASET_NAME_LEN
];
3736 dsl_dataset_name(drc
->drc_ds
, name
);
3737 zfs_destroy_unmount_origin(name
);
3740 return (dsl_sync_task(drc
->drc_tofs
,
3741 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3742 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
3746 dmu_recv_new_end(dmu_recv_cookie_t
*drc
)
3748 return (dsl_sync_task(drc
->drc_tofs
,
3749 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3750 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
3754 dmu_recv_end(dmu_recv_cookie_t
*drc
, void *owner
)
3758 drc
->drc_owner
= owner
;
3761 error
= dmu_recv_new_end(drc
);
3763 error
= dmu_recv_existing_end(drc
);
3766 dmu_recv_cleanup_ds(drc
);
3767 nvlist_free(drc
->drc_keynvl
);
3768 } else if (!drc
->drc_heal
) {
3769 if (drc
->drc_newfs
) {
3770 zvol_create_minor(drc
->drc_tofs
);
3772 char *snapname
= kmem_asprintf("%s@%s",
3773 drc
->drc_tofs
, drc
->drc_tosnap
);
3774 zvol_create_minor(snapname
);
3775 kmem_strfree(snapname
);
3781 * Return TRUE if this objset is currently being received into.
3784 dmu_objset_is_receiving(objset_t
*os
)
3786 return (os
->os_dsl_dataset
!= NULL
&&
3787 os
->os_dsl_dataset
->ds_owner
== dmu_recv_tag
);
3790 ZFS_MODULE_PARAM(zfs_recv
, zfs_recv_
, queue_length
, UINT
, ZMOD_RW
,
3791 "Maximum receive queue length");
3793 ZFS_MODULE_PARAM(zfs_recv
, zfs_recv_
, queue_ff
, UINT
, ZMOD_RW
,
3794 "Receive queue fill fraction");
3796 ZFS_MODULE_PARAM(zfs_recv
, zfs_recv_
, write_batch_size
, UINT
, ZMOD_RW
,
3797 "Maximum amount of writes to batch into one transaction");
3799 ZFS_MODULE_PARAM(zfs_recv
, zfs_recv_
, best_effort_corrective
, INT
, ZMOD_RW
,
3800 "Ignore errors during corrective receive");