4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright 2014 HybridCluster. All rights reserved.
27 * Copyright 2016 RackTop Systems.
28 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
32 #include <sys/dmu_impl.h>
33 #include <sys/dmu_tx.h>
35 #include <sys/dnode.h>
36 #include <sys/zfs_context.h>
37 #include <sys/dmu_objset.h>
38 #include <sys/dmu_traverse.h>
39 #include <sys/dsl_dataset.h>
40 #include <sys/dsl_dir.h>
41 #include <sys/dsl_prop.h>
42 #include <sys/dsl_pool.h>
43 #include <sys/dsl_synctask.h>
44 #include <sys/spa_impl.h>
45 #include <sys/zfs_ioctl.h>
47 #include <sys/zio_checksum.h>
48 #include <sys/zfs_znode.h>
49 #include <zfs_fletcher.h>
52 #include <sys/zfs_onexit.h>
53 #include <sys/dmu_send.h>
54 #include <sys/dsl_destroy.h>
55 #include <sys/blkptr.h>
56 #include <sys/dsl_bookmark.h>
57 #include <sys/zfeature.h>
58 #include <sys/bqueue.h>
60 #include <sys/policy.h>
62 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
63 int zfs_send_corrupt_data
= B_FALSE
;
64 int zfs_send_queue_length
= 16 * 1024 * 1024;
65 int zfs_recv_queue_length
= 16 * 1024 * 1024;
66 /* Set this tunable to FALSE to disable setting of DRR_FLAG_FREERECORDS */
67 int zfs_send_set_freerecords_bit
= B_TRUE
;
69 static char *dmu_recv_tag
= "dmu_recv_tag";
70 const char *recv_clone_name
= "%recv";
72 #define BP_SPAN(datablkszsec, indblkshift, level) \
73 (((uint64_t)datablkszsec) << (SPA_MINBLOCKSHIFT + \
74 (level) * (indblkshift - SPA_BLKPTRSHIFT)))
76 static void byteswap_record(dmu_replay_record_t
*drr
);
78 struct send_thread_arg
{
80 dsl_dataset_t
*ds
; /* Dataset to traverse */
81 uint64_t fromtxg
; /* Traverse from this txg */
82 int flags
; /* flags to pass to traverse_dataset */
85 zbookmark_phys_t resume
;
88 struct send_block_record
{
89 boolean_t eos_marker
; /* Marks the end of the stream */
93 uint16_t datablkszsec
;
97 typedef struct dump_bytes_io
{
98 dmu_sendarg_t
*dbi_dsp
;
104 dump_bytes_cb(void *arg
)
106 dump_bytes_io_t
*dbi
= (dump_bytes_io_t
*)arg
;
107 dmu_sendarg_t
*dsp
= dbi
->dbi_dsp
;
108 dsl_dataset_t
*ds
= dmu_objset_ds(dsp
->dsa_os
);
109 ssize_t resid
; /* have to get resid to get detailed errno */
112 * The code does not rely on len being a multiple of 8. We keep
113 * this assertion because of the corresponding assertion in
114 * receive_read(). Keeping this assertion ensures that we do not
115 * inadvertently break backwards compatibility (causing the assertion
116 * in receive_read() to trigger on old software). Newer feature flags
117 * (such as raw send) may break this assertion since they were
118 * introduced after the requirement was made obsolete.
121 ASSERT(dbi
->dbi_len
% 8 == 0 ||
122 (dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0);
124 dsp
->dsa_err
= vn_rdwr(UIO_WRITE
, dsp
->dsa_vp
,
125 (caddr_t
)dbi
->dbi_buf
, dbi
->dbi_len
,
126 0, UIO_SYSSPACE
, FAPPEND
, RLIM64_INFINITY
, CRED(), &resid
);
128 mutex_enter(&ds
->ds_sendstream_lock
);
129 *dsp
->dsa_off
+= dbi
->dbi_len
;
130 mutex_exit(&ds
->ds_sendstream_lock
);
134 dump_bytes(dmu_sendarg_t
*dsp
, void *buf
, int len
)
142 #if defined(HAVE_LARGE_STACKS)
146 * The vn_rdwr() call is performed in a taskq to ensure that there is
147 * always enough stack space to write safely to the target filesystem.
148 * The ZIO_TYPE_FREE threads are used because there can be a lot of
149 * them and they are used in vdev_file.c for a similar purpose.
151 spa_taskq_dispatch_sync(dmu_objset_spa(dsp
->dsa_os
), ZIO_TYPE_FREE
,
152 ZIO_TASKQ_ISSUE
, dump_bytes_cb
, &dbi
, TQ_SLEEP
);
153 #endif /* HAVE_LARGE_STACKS */
155 return (dsp
->dsa_err
);
159 * For all record types except BEGIN, fill in the checksum (overlaid in
160 * drr_u.drr_checksum.drr_checksum). The checksum verifies everything
161 * up to the start of the checksum itself.
164 dump_record(dmu_sendarg_t
*dsp
, void *payload
, int payload_len
)
166 ASSERT3U(offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
167 ==, sizeof (dmu_replay_record_t
) - sizeof (zio_cksum_t
));
168 (void) fletcher_4_incremental_native(dsp
->dsa_drr
,
169 offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
171 if (dsp
->dsa_drr
->drr_type
== DRR_BEGIN
) {
172 dsp
->dsa_sent_begin
= B_TRUE
;
174 ASSERT(ZIO_CHECKSUM_IS_ZERO(&dsp
->dsa_drr
->drr_u
.
175 drr_checksum
.drr_checksum
));
176 dsp
->dsa_drr
->drr_u
.drr_checksum
.drr_checksum
= dsp
->dsa_zc
;
178 if (dsp
->dsa_drr
->drr_type
== DRR_END
) {
179 dsp
->dsa_sent_end
= B_TRUE
;
181 (void) fletcher_4_incremental_native(&dsp
->dsa_drr
->
182 drr_u
.drr_checksum
.drr_checksum
,
183 sizeof (zio_cksum_t
), &dsp
->dsa_zc
);
184 if (dump_bytes(dsp
, dsp
->dsa_drr
, sizeof (dmu_replay_record_t
)) != 0)
185 return (SET_ERROR(EINTR
));
186 if (payload_len
!= 0) {
187 (void) fletcher_4_incremental_native(payload
, payload_len
,
189 if (dump_bytes(dsp
, payload
, payload_len
) != 0)
190 return (SET_ERROR(EINTR
));
196 * Fill in the drr_free struct, or perform aggregation if the previous record is
197 * also a free record, and the two are adjacent.
199 * Note that we send free records even for a full send, because we want to be
200 * able to receive a full send as a clone, which requires a list of all the free
201 * and freeobject records that were generated on the source.
204 dump_free(dmu_sendarg_t
*dsp
, uint64_t object
, uint64_t offset
,
207 struct drr_free
*drrf
= &(dsp
->dsa_drr
->drr_u
.drr_free
);
210 * When we receive a free record, dbuf_free_range() assumes
211 * that the receiving system doesn't have any dbufs in the range
212 * being freed. This is always true because there is a one-record
213 * constraint: we only send one WRITE record for any given
214 * object,offset. We know that the one-record constraint is
215 * true because we always send data in increasing order by
218 * If the increasing-order constraint ever changes, we should find
219 * another way to assert that the one-record constraint is still
222 ASSERT(object
> dsp
->dsa_last_data_object
||
223 (object
== dsp
->dsa_last_data_object
&&
224 offset
> dsp
->dsa_last_data_offset
));
226 if (length
!= -1ULL && offset
+ length
< offset
)
230 * If there is a pending op, but it's not PENDING_FREE, push it out,
231 * since free block aggregation can only be done for blocks of the
232 * same type (i.e., DRR_FREE records can only be aggregated with
233 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
234 * aggregated with other DRR_FREEOBJECTS records.
236 if (dsp
->dsa_pending_op
!= PENDING_NONE
&&
237 dsp
->dsa_pending_op
!= PENDING_FREE
) {
238 if (dump_record(dsp
, NULL
, 0) != 0)
239 return (SET_ERROR(EINTR
));
240 dsp
->dsa_pending_op
= PENDING_NONE
;
243 if (dsp
->dsa_pending_op
== PENDING_FREE
) {
245 * There should never be a PENDING_FREE if length is -1
246 * (because dump_dnode is the only place where this
247 * function is called with a -1, and only after flushing
248 * any pending record).
250 ASSERT(length
!= -1ULL);
252 * Check to see whether this free block can be aggregated
255 if (drrf
->drr_object
== object
&& drrf
->drr_offset
+
256 drrf
->drr_length
== offset
) {
257 drrf
->drr_length
+= length
;
260 /* not a continuation. Push out pending record */
261 if (dump_record(dsp
, NULL
, 0) != 0)
262 return (SET_ERROR(EINTR
));
263 dsp
->dsa_pending_op
= PENDING_NONE
;
266 /* create a FREE record and make it pending */
267 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
268 dsp
->dsa_drr
->drr_type
= DRR_FREE
;
269 drrf
->drr_object
= object
;
270 drrf
->drr_offset
= offset
;
271 drrf
->drr_length
= length
;
272 drrf
->drr_toguid
= dsp
->dsa_toguid
;
273 if (length
== -1ULL) {
274 if (dump_record(dsp
, NULL
, 0) != 0)
275 return (SET_ERROR(EINTR
));
277 dsp
->dsa_pending_op
= PENDING_FREE
;
284 dump_write(dmu_sendarg_t
*dsp
, dmu_object_type_t type
, uint64_t object
,
285 uint64_t offset
, int lsize
, int psize
, const blkptr_t
*bp
, void *data
)
287 uint64_t payload_size
;
288 boolean_t raw
= (dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
);
289 struct drr_write
*drrw
= &(dsp
->dsa_drr
->drr_u
.drr_write
);
292 * We send data in increasing object, offset order.
293 * See comment in dump_free() for details.
295 ASSERT(object
> dsp
->dsa_last_data_object
||
296 (object
== dsp
->dsa_last_data_object
&&
297 offset
> dsp
->dsa_last_data_offset
));
298 dsp
->dsa_last_data_object
= object
;
299 dsp
->dsa_last_data_offset
= offset
+ lsize
- 1;
302 * If there is any kind of pending aggregation (currently either
303 * a grouping of free objects or free blocks), push it out to
304 * the stream, since aggregation can't be done across operations
305 * of different types.
307 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
308 if (dump_record(dsp
, NULL
, 0) != 0)
309 return (SET_ERROR(EINTR
));
310 dsp
->dsa_pending_op
= PENDING_NONE
;
312 /* write a WRITE record */
313 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
314 dsp
->dsa_drr
->drr_type
= DRR_WRITE
;
315 drrw
->drr_object
= object
;
316 drrw
->drr_type
= type
;
317 drrw
->drr_offset
= offset
;
318 drrw
->drr_toguid
= dsp
->dsa_toguid
;
319 drrw
->drr_logical_size
= lsize
;
321 /* only set the compression fields if the buf is compressed or raw */
322 if (raw
|| lsize
!= psize
) {
323 ASSERT(!BP_IS_EMBEDDED(bp
));
324 ASSERT3S(psize
, >, 0);
327 ASSERT(BP_IS_PROTECTED(bp
));
330 * This is a raw protected block so we need to pass
331 * along everything the receiving side will need to
332 * interpret this block, including the byteswap, salt,
335 if (BP_SHOULD_BYTESWAP(bp
))
336 drrw
->drr_flags
|= DRR_RAW_BYTESWAP
;
337 zio_crypt_decode_params_bp(bp
, drrw
->drr_salt
,
339 zio_crypt_decode_mac_bp(bp
, drrw
->drr_mac
);
341 /* this is a compressed block */
342 ASSERT(dsp
->dsa_featureflags
&
343 DMU_BACKUP_FEATURE_COMPRESSED
);
344 ASSERT(!BP_SHOULD_BYTESWAP(bp
));
345 ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)));
346 ASSERT3U(BP_GET_COMPRESS(bp
), !=, ZIO_COMPRESS_OFF
);
347 ASSERT3S(lsize
, >=, psize
);
350 /* set fields common to compressed and raw sends */
351 drrw
->drr_compressiontype
= BP_GET_COMPRESS(bp
);
352 drrw
->drr_compressed_size
= psize
;
353 payload_size
= drrw
->drr_compressed_size
;
355 payload_size
= drrw
->drr_logical_size
;
358 if (bp
== NULL
|| BP_IS_EMBEDDED(bp
) || (BP_IS_PROTECTED(bp
) && !raw
)) {
360 * There's no pre-computed checksum for partial-block writes,
361 * embedded BP's, or encrypted BP's that are being sent as
362 * plaintext, so (like fletcher4-checkummed blocks) userland
363 * will have to compute a dedup-capable checksum itself.
365 drrw
->drr_checksumtype
= ZIO_CHECKSUM_OFF
;
367 drrw
->drr_checksumtype
= BP_GET_CHECKSUM(bp
);
368 if (zio_checksum_table
[drrw
->drr_checksumtype
].ci_flags
&
369 ZCHECKSUM_FLAG_DEDUP
)
370 drrw
->drr_flags
|= DRR_CHECKSUM_DEDUP
;
371 DDK_SET_LSIZE(&drrw
->drr_key
, BP_GET_LSIZE(bp
));
372 DDK_SET_PSIZE(&drrw
->drr_key
, BP_GET_PSIZE(bp
));
373 DDK_SET_COMPRESS(&drrw
->drr_key
, BP_GET_COMPRESS(bp
));
374 DDK_SET_CRYPT(&drrw
->drr_key
, BP_IS_PROTECTED(bp
));
375 drrw
->drr_key
.ddk_cksum
= bp
->blk_cksum
;
378 if (dump_record(dsp
, data
, payload_size
) != 0)
379 return (SET_ERROR(EINTR
));
384 dump_write_embedded(dmu_sendarg_t
*dsp
, uint64_t object
, uint64_t offset
,
385 int blksz
, const blkptr_t
*bp
)
387 char buf
[BPE_PAYLOAD_SIZE
];
388 struct drr_write_embedded
*drrw
=
389 &(dsp
->dsa_drr
->drr_u
.drr_write_embedded
);
391 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
392 if (dump_record(dsp
, NULL
, 0) != 0)
393 return (SET_ERROR(EINTR
));
394 dsp
->dsa_pending_op
= PENDING_NONE
;
397 ASSERT(BP_IS_EMBEDDED(bp
));
399 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
400 dsp
->dsa_drr
->drr_type
= DRR_WRITE_EMBEDDED
;
401 drrw
->drr_object
= object
;
402 drrw
->drr_offset
= offset
;
403 drrw
->drr_length
= blksz
;
404 drrw
->drr_toguid
= dsp
->dsa_toguid
;
405 drrw
->drr_compression
= BP_GET_COMPRESS(bp
);
406 drrw
->drr_etype
= BPE_GET_ETYPE(bp
);
407 drrw
->drr_lsize
= BPE_GET_LSIZE(bp
);
408 drrw
->drr_psize
= BPE_GET_PSIZE(bp
);
410 decode_embedded_bp_compressed(bp
, buf
);
412 if (dump_record(dsp
, buf
, P2ROUNDUP(drrw
->drr_psize
, 8)) != 0)
413 return (SET_ERROR(EINTR
));
418 dump_spill(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
, uint64_t object
, void *data
)
420 struct drr_spill
*drrs
= &(dsp
->dsa_drr
->drr_u
.drr_spill
);
421 uint64_t blksz
= BP_GET_LSIZE(bp
);
423 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
424 if (dump_record(dsp
, NULL
, 0) != 0)
425 return (SET_ERROR(EINTR
));
426 dsp
->dsa_pending_op
= PENDING_NONE
;
429 /* write a SPILL record */
430 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
431 dsp
->dsa_drr
->drr_type
= DRR_SPILL
;
432 drrs
->drr_object
= object
;
433 drrs
->drr_length
= blksz
;
434 drrs
->drr_toguid
= dsp
->dsa_toguid
;
436 /* handle raw send fields */
437 if (dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
438 ASSERT(BP_IS_PROTECTED(bp
));
440 if (BP_SHOULD_BYTESWAP(bp
))
441 drrs
->drr_flags
|= DRR_RAW_BYTESWAP
;
442 drrs
->drr_compressiontype
= BP_GET_COMPRESS(bp
);
443 drrs
->drr_compressed_size
= BP_GET_PSIZE(bp
);
444 zio_crypt_decode_params_bp(bp
, drrs
->drr_salt
, drrs
->drr_iv
);
445 zio_crypt_decode_mac_bp(bp
, drrs
->drr_mac
);
448 if (dump_record(dsp
, data
, blksz
) != 0)
449 return (SET_ERROR(EINTR
));
454 dump_freeobjects(dmu_sendarg_t
*dsp
, uint64_t firstobj
, uint64_t numobjs
)
456 struct drr_freeobjects
*drrfo
= &(dsp
->dsa_drr
->drr_u
.drr_freeobjects
);
459 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
460 * push it out, since free block aggregation can only be done for
461 * blocks of the same type (i.e., DRR_FREE records can only be
462 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
463 * can only be aggregated with other DRR_FREEOBJECTS records.
465 if (dsp
->dsa_pending_op
!= PENDING_NONE
&&
466 dsp
->dsa_pending_op
!= PENDING_FREEOBJECTS
) {
467 if (dump_record(dsp
, NULL
, 0) != 0)
468 return (SET_ERROR(EINTR
));
469 dsp
->dsa_pending_op
= PENDING_NONE
;
471 if (dsp
->dsa_pending_op
== PENDING_FREEOBJECTS
) {
473 * See whether this free object array can be aggregated
476 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
== firstobj
) {
477 drrfo
->drr_numobjs
+= numobjs
;
480 /* can't be aggregated. Push out pending record */
481 if (dump_record(dsp
, NULL
, 0) != 0)
482 return (SET_ERROR(EINTR
));
483 dsp
->dsa_pending_op
= PENDING_NONE
;
487 /* write a FREEOBJECTS record */
488 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
489 dsp
->dsa_drr
->drr_type
= DRR_FREEOBJECTS
;
490 drrfo
->drr_firstobj
= firstobj
;
491 drrfo
->drr_numobjs
= numobjs
;
492 drrfo
->drr_toguid
= dsp
->dsa_toguid
;
494 dsp
->dsa_pending_op
= PENDING_FREEOBJECTS
;
500 dump_dnode(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
, uint64_t object
,
503 struct drr_object
*drro
= &(dsp
->dsa_drr
->drr_u
.drr_object
);
504 int bonuslen
= P2ROUNDUP(dnp
->dn_bonuslen
, 8);
506 if (object
< dsp
->dsa_resume_object
) {
508 * Note: when resuming, we will visit all the dnodes in
509 * the block of dnodes that we are resuming from. In
510 * this case it's unnecessary to send the dnodes prior to
511 * the one we are resuming from. We should be at most one
512 * block's worth of dnodes behind the resume point.
514 ASSERT3U(dsp
->dsa_resume_object
- object
, <,
515 1 << (DNODE_BLOCK_SHIFT
- DNODE_SHIFT
));
519 if (dnp
== NULL
|| dnp
->dn_type
== DMU_OT_NONE
)
520 return (dump_freeobjects(dsp
, object
, 1));
522 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
523 if (dump_record(dsp
, NULL
, 0) != 0)
524 return (SET_ERROR(EINTR
));
525 dsp
->dsa_pending_op
= PENDING_NONE
;
528 /* write an OBJECT record */
529 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
530 dsp
->dsa_drr
->drr_type
= DRR_OBJECT
;
531 drro
->drr_object
= object
;
532 drro
->drr_type
= dnp
->dn_type
;
533 drro
->drr_bonustype
= dnp
->dn_bonustype
;
534 drro
->drr_blksz
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
535 drro
->drr_bonuslen
= dnp
->dn_bonuslen
;
536 drro
->drr_dn_slots
= dnp
->dn_extra_slots
+ 1;
537 drro
->drr_checksumtype
= dnp
->dn_checksum
;
538 drro
->drr_compress
= dnp
->dn_compress
;
539 drro
->drr_toguid
= dsp
->dsa_toguid
;
541 if (!(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
542 drro
->drr_blksz
> SPA_OLD_MAXBLOCKSIZE
)
543 drro
->drr_blksz
= SPA_OLD_MAXBLOCKSIZE
;
545 if ((dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
)) {
546 ASSERT(BP_IS_ENCRYPTED(bp
));
548 if (BP_SHOULD_BYTESWAP(bp
))
549 drro
->drr_flags
|= DRR_RAW_BYTESWAP
;
551 /* needed for reconstructing dnp on recv side */
552 drro
->drr_indblkshift
= dnp
->dn_indblkshift
;
553 drro
->drr_nlevels
= dnp
->dn_nlevels
;
554 drro
->drr_nblkptr
= dnp
->dn_nblkptr
;
557 * Since we encrypt the entire bonus area, the (raw) part
558 * beyond the the bonuslen is actually nonzero, so we need
562 drro
->drr_raw_bonuslen
= DN_MAX_BONUS_LEN(dnp
);
563 bonuslen
= drro
->drr_raw_bonuslen
;
567 if (dump_record(dsp
, DN_BONUS(dnp
), bonuslen
) != 0)
568 return (SET_ERROR(EINTR
));
570 /* Free anything past the end of the file. */
571 if (dump_free(dsp
, object
, (dnp
->dn_maxblkid
+ 1) *
572 (dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
), -1ULL) != 0)
573 return (SET_ERROR(EINTR
));
574 if (dsp
->dsa_err
!= 0)
575 return (SET_ERROR(EINTR
));
580 dump_object_range(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
, uint64_t firstobj
,
583 struct drr_object_range
*drror
=
584 &(dsp
->dsa_drr
->drr_u
.drr_object_range
);
586 /* we only use this record type for raw sends */
587 ASSERT(BP_IS_PROTECTED(bp
));
588 ASSERT(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
);
589 ASSERT3U(BP_GET_COMPRESS(bp
), ==, ZIO_COMPRESS_OFF
);
590 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_DNODE
);
591 ASSERT0(BP_GET_LEVEL(bp
));
593 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
594 if (dump_record(dsp
, NULL
, 0) != 0)
595 return (SET_ERROR(EINTR
));
596 dsp
->dsa_pending_op
= PENDING_NONE
;
599 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
600 dsp
->dsa_drr
->drr_type
= DRR_OBJECT_RANGE
;
601 drror
->drr_firstobj
= firstobj
;
602 drror
->drr_numslots
= numslots
;
603 drror
->drr_toguid
= dsp
->dsa_toguid
;
604 if (BP_SHOULD_BYTESWAP(bp
))
605 drror
->drr_flags
|= DRR_RAW_BYTESWAP
;
606 zio_crypt_decode_params_bp(bp
, drror
->drr_salt
, drror
->drr_iv
);
607 zio_crypt_decode_mac_bp(bp
, drror
->drr_mac
);
609 if (dump_record(dsp
, NULL
, 0) != 0)
610 return (SET_ERROR(EINTR
));
615 backup_do_embed(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
)
617 if (!BP_IS_EMBEDDED(bp
))
621 * Compression function must be legacy, or explicitly enabled.
623 if ((BP_GET_COMPRESS(bp
) >= ZIO_COMPRESS_LEGACY_FUNCTIONS
&&
624 !(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_LZ4
)))
628 * Embed type must be explicitly enabled.
630 switch (BPE_GET_ETYPE(bp
)) {
631 case BP_EMBEDDED_TYPE_DATA
:
632 if (dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)
642 * This is the callback function to traverse_dataset that acts as the worker
643 * thread for dmu_send_impl.
647 send_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
648 const zbookmark_phys_t
*zb
, const struct dnode_phys
*dnp
, void *arg
)
650 struct send_thread_arg
*sta
= arg
;
651 struct send_block_record
*record
;
652 uint64_t record_size
;
655 ASSERT(zb
->zb_object
== DMU_META_DNODE_OBJECT
||
656 zb
->zb_object
>= sta
->resume
.zb_object
);
657 ASSERT3P(sta
->ds
, !=, NULL
);
660 return (SET_ERROR(EINTR
));
663 ASSERT3U(zb
->zb_level
, ==, ZB_DNODE_LEVEL
);
665 } else if (zb
->zb_level
< 0) {
669 record
= kmem_zalloc(sizeof (struct send_block_record
), KM_SLEEP
);
670 record
->eos_marker
= B_FALSE
;
673 record
->indblkshift
= dnp
->dn_indblkshift
;
674 record
->datablkszsec
= dnp
->dn_datablkszsec
;
675 record_size
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
676 bqueue_enqueue(&sta
->q
, record
, record_size
);
682 * This function kicks off the traverse_dataset. It also handles setting the
683 * error code of the thread in case something goes wrong, and pushes the End of
684 * Stream record when the traverse_dataset call has finished. If there is no
685 * dataset to traverse, the thread immediately pushes End of Stream marker.
688 send_traverse_thread(void *arg
)
690 struct send_thread_arg
*st_arg
= arg
;
692 struct send_block_record
*data
;
693 fstrans_cookie_t cookie
= spl_fstrans_mark();
695 if (st_arg
->ds
!= NULL
) {
696 err
= traverse_dataset_resume(st_arg
->ds
,
697 st_arg
->fromtxg
, &st_arg
->resume
,
698 st_arg
->flags
, send_cb
, st_arg
);
701 st_arg
->error_code
= err
;
703 data
= kmem_zalloc(sizeof (*data
), KM_SLEEP
);
704 data
->eos_marker
= B_TRUE
;
705 bqueue_enqueue(&st_arg
->q
, data
, 1);
706 spl_fstrans_unmark(cookie
);
711 * This function actually handles figuring out what kind of record needs to be
712 * dumped, reading the data (which has hopefully been prefetched), and calling
713 * the appropriate helper function.
716 do_dump(dmu_sendarg_t
*dsa
, struct send_block_record
*data
)
718 dsl_dataset_t
*ds
= dmu_objset_ds(dsa
->dsa_os
);
719 const blkptr_t
*bp
= &data
->bp
;
720 const zbookmark_phys_t
*zb
= &data
->zb
;
721 uint8_t indblkshift
= data
->indblkshift
;
722 uint16_t dblkszsec
= data
->datablkszsec
;
723 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
724 dmu_object_type_t type
= bp
? BP_GET_TYPE(bp
) : DMU_OT_NONE
;
728 ASSERT3U(zb
->zb_level
, >=, 0);
730 ASSERT(zb
->zb_object
== DMU_META_DNODE_OBJECT
||
731 zb
->zb_object
>= dsa
->dsa_resume_object
);
734 * All bps of an encrypted os should have the encryption bit set.
735 * If this is not true it indicates tampering and we report an error.
737 if (dsa
->dsa_os
->os_encrypted
&&
738 !BP_IS_HOLE(bp
) && !BP_USES_CRYPT(bp
)) {
739 spa_log_error(spa
, zb
);
740 zfs_panic_recover("unencrypted block in encrypted "
741 "object set %llu", ds
->ds_object
);
742 return (SET_ERROR(EIO
));
745 if (zb
->zb_object
!= DMU_META_DNODE_OBJECT
&&
746 DMU_OBJECT_IS_SPECIAL(zb
->zb_object
)) {
748 } else if (BP_IS_HOLE(bp
) &&
749 zb
->zb_object
== DMU_META_DNODE_OBJECT
) {
750 uint64_t span
= BP_SPAN(dblkszsec
, indblkshift
, zb
->zb_level
);
751 uint64_t dnobj
= (zb
->zb_blkid
* span
) >> DNODE_SHIFT
;
752 err
= dump_freeobjects(dsa
, dnobj
, span
>> DNODE_SHIFT
);
753 } else if (BP_IS_HOLE(bp
)) {
754 uint64_t span
= BP_SPAN(dblkszsec
, indblkshift
, zb
->zb_level
);
755 uint64_t offset
= zb
->zb_blkid
* span
;
756 err
= dump_free(dsa
, zb
->zb_object
, offset
, span
);
757 } else if (zb
->zb_level
> 0 || type
== DMU_OT_OBJSET
) {
759 } else if (type
== DMU_OT_DNODE
) {
761 int epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
762 arc_flags_t aflags
= ARC_FLAG_WAIT
;
764 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
767 if (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
768 ASSERT(BP_IS_ENCRYPTED(bp
));
769 ASSERT3U(BP_GET_COMPRESS(bp
), ==, ZIO_COMPRESS_OFF
);
770 zioflags
|= ZIO_FLAG_RAW
;
773 ASSERT0(zb
->zb_level
);
775 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
776 ZIO_PRIORITY_ASYNC_READ
, zioflags
, &aflags
, zb
) != 0)
777 return (SET_ERROR(EIO
));
780 dnobj
= zb
->zb_blkid
* epb
;
783 * Raw sends require sending encryption parameters for the
784 * block of dnodes. Regular sends do not need to send this
787 if (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
788 ASSERT(arc_is_encrypted(abuf
));
789 err
= dump_object_range(dsa
, bp
, dnobj
, epb
);
793 for (i
= 0; i
< epb
; i
+= blk
[i
].dn_extra_slots
+ 1) {
794 err
= dump_dnode(dsa
, bp
, dnobj
+ i
, blk
+ i
);
799 arc_buf_destroy(abuf
, &abuf
);
800 } else if (type
== DMU_OT_SA
) {
801 arc_flags_t aflags
= ARC_FLAG_WAIT
;
803 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
805 if (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
806 ASSERT(BP_IS_PROTECTED(bp
));
807 zioflags
|= ZIO_FLAG_RAW
;
810 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
811 ZIO_PRIORITY_ASYNC_READ
, zioflags
, &aflags
, zb
) != 0)
812 return (SET_ERROR(EIO
));
814 err
= dump_spill(dsa
, bp
, zb
->zb_object
, abuf
->b_data
);
815 arc_buf_destroy(abuf
, &abuf
);
816 } else if (backup_do_embed(dsa
, bp
)) {
817 /* it's an embedded level-0 block of a regular object */
818 int blksz
= dblkszsec
<< SPA_MINBLOCKSHIFT
;
819 ASSERT0(zb
->zb_level
);
820 err
= dump_write_embedded(dsa
, zb
->zb_object
,
821 zb
->zb_blkid
* blksz
, blksz
, bp
);
823 /* it's a level-0 block of a regular object */
824 arc_flags_t aflags
= ARC_FLAG_WAIT
;
826 int blksz
= dblkszsec
<< SPA_MINBLOCKSHIFT
;
830 * If we have large blocks stored on disk but the send flags
831 * don't allow us to send large blocks, we split the data from
832 * the arc buf into chunks.
834 boolean_t split_large_blocks
= blksz
> SPA_OLD_MAXBLOCKSIZE
&&
835 !(dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
);
838 * Raw sends require that we always get raw data as it exists
839 * on disk, so we assert that we are not splitting blocks here.
841 boolean_t request_raw
=
842 (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0;
845 * We should only request compressed data from the ARC if all
846 * the following are true:
847 * - stream compression was requested
848 * - we aren't splitting large blocks into smaller chunks
849 * - the data won't need to be byteswapped before sending
850 * - this isn't an embedded block
851 * - this isn't metadata (if receiving on a different endian
852 * system it can be byteswapped more easily)
854 boolean_t request_compressed
=
855 (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
) &&
856 !split_large_blocks
&& !BP_SHOULD_BYTESWAP(bp
) &&
857 !BP_IS_EMBEDDED(bp
) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp
));
859 IMPLY(request_raw
, !split_large_blocks
);
860 IMPLY(request_raw
, BP_IS_PROTECTED(bp
));
861 ASSERT0(zb
->zb_level
);
862 ASSERT(zb
->zb_object
> dsa
->dsa_resume_object
||
863 (zb
->zb_object
== dsa
->dsa_resume_object
&&
864 zb
->zb_blkid
* blksz
>= dsa
->dsa_resume_offset
));
866 ASSERT3U(blksz
, ==, BP_GET_LSIZE(bp
));
868 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
870 zioflags
|= ZIO_FLAG_RAW
;
871 else if (request_compressed
)
872 zioflags
|= ZIO_FLAG_RAW_COMPRESS
;
874 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
875 ZIO_PRIORITY_ASYNC_READ
, zioflags
, &aflags
, zb
) != 0) {
876 if (zfs_send_corrupt_data
) {
877 /* Send a block filled with 0x"zfs badd bloc" */
878 abuf
= arc_alloc_buf(spa
, &abuf
, ARC_BUFC_DATA
,
881 for (ptr
= abuf
->b_data
;
882 (char *)ptr
< (char *)abuf
->b_data
+ blksz
;
884 *ptr
= 0x2f5baddb10cULL
;
886 return (SET_ERROR(EIO
));
890 offset
= zb
->zb_blkid
* blksz
;
892 if (split_large_blocks
) {
893 ASSERT0(arc_is_encrypted(abuf
));
894 ASSERT3U(arc_get_compression(abuf
), ==,
896 char *buf
= abuf
->b_data
;
897 while (blksz
> 0 && err
== 0) {
898 int n
= MIN(blksz
, SPA_OLD_MAXBLOCKSIZE
);
899 err
= dump_write(dsa
, type
, zb
->zb_object
,
900 offset
, n
, n
, NULL
, buf
);
906 err
= dump_write(dsa
, type
, zb
->zb_object
, offset
,
907 blksz
, arc_buf_size(abuf
), bp
, abuf
->b_data
);
909 arc_buf_destroy(abuf
, &abuf
);
912 ASSERT(err
== 0 || err
== EINTR
);
917 * Pop the new data off the queue, and free the old data.
919 static struct send_block_record
*
920 get_next_record(bqueue_t
*bq
, struct send_block_record
*data
)
922 struct send_block_record
*tmp
= bqueue_dequeue(bq
);
923 kmem_free(data
, sizeof (*data
));
928 * Actually do the bulk of the work in a zfs send.
930 * Note: Releases dp using the specified tag.
933 dmu_send_impl(void *tag
, dsl_pool_t
*dp
, dsl_dataset_t
*to_ds
,
934 zfs_bookmark_phys_t
*ancestor_zb
, boolean_t is_clone
,
935 boolean_t embedok
, boolean_t large_block_ok
, boolean_t compressok
,
936 boolean_t rawok
, int outfd
, uint64_t resumeobj
, uint64_t resumeoff
,
937 vnode_t
*vp
, offset_t
*off
)
940 dmu_replay_record_t
*drr
;
943 uint64_t fromtxg
= 0;
944 uint64_t featureflags
= 0;
945 struct send_thread_arg to_arg
;
946 void *payload
= NULL
;
947 size_t payload_len
= 0;
948 struct send_block_record
*to_data
;
950 err
= dmu_objset_from_ds(to_ds
, &os
);
952 dsl_pool_rele(dp
, tag
);
957 * If this is a non-raw send of an encrypted ds, we can ensure that
958 * the objset_phys_t is authenticated. This is safe because this is
959 * either a snapshot or we have owned the dataset, ensuring that
960 * it can't be modified.
962 if (!rawok
&& os
->os_encrypted
&&
963 arc_is_unauthenticated(os
->os_phys_buf
)) {
964 err
= arc_untransform(os
->os_phys_buf
, os
->os_spa
,
965 to_ds
->ds_object
, B_FALSE
);
967 dsl_pool_rele(dp
, tag
);
971 ASSERT0(arc_is_unauthenticated(os
->os_phys_buf
));
974 drr
= kmem_zalloc(sizeof (dmu_replay_record_t
), KM_SLEEP
);
975 drr
->drr_type
= DRR_BEGIN
;
976 drr
->drr_u
.drr_begin
.drr_magic
= DMU_BACKUP_MAGIC
;
977 DMU_SET_STREAM_HDRTYPE(drr
->drr_u
.drr_begin
.drr_versioninfo
,
980 bzero(&to_arg
, sizeof (to_arg
));
983 if (dmu_objset_type(os
) == DMU_OST_ZFS
) {
985 if (zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &version
) != 0) {
986 kmem_free(drr
, sizeof (dmu_replay_record_t
));
987 dsl_pool_rele(dp
, tag
);
988 return (SET_ERROR(EINVAL
));
990 if (version
>= ZPL_VERSION_SA
) {
991 featureflags
|= DMU_BACKUP_FEATURE_SA_SPILL
;
996 /* raw sends imply large_block_ok */
997 if ((large_block_ok
|| rawok
) &&
998 to_ds
->ds_feature_inuse
[SPA_FEATURE_LARGE_BLOCKS
])
999 featureflags
|= DMU_BACKUP_FEATURE_LARGE_BLOCKS
;
1000 if (to_ds
->ds_feature_inuse
[SPA_FEATURE_LARGE_DNODE
])
1001 featureflags
|= DMU_BACKUP_FEATURE_LARGE_DNODE
;
1003 /* encrypted datasets will not have embedded blocks */
1004 if ((embedok
|| rawok
) && !os
->os_encrypted
&&
1005 spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
)) {
1006 featureflags
|= DMU_BACKUP_FEATURE_EMBED_DATA
;
1009 /* raw send implies compressok */
1010 if (compressok
|| rawok
)
1011 featureflags
|= DMU_BACKUP_FEATURE_COMPRESSED
;
1012 if (rawok
&& os
->os_encrypted
)
1013 featureflags
|= DMU_BACKUP_FEATURE_RAW
;
1016 (DMU_BACKUP_FEATURE_EMBED_DATA
| DMU_BACKUP_FEATURE_COMPRESSED
|
1017 DMU_BACKUP_FEATURE_RAW
)) != 0 &&
1018 spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
)) {
1019 featureflags
|= DMU_BACKUP_FEATURE_LZ4
;
1022 if (resumeobj
!= 0 || resumeoff
!= 0) {
1023 featureflags
|= DMU_BACKUP_FEATURE_RESUMING
;
1026 DMU_SET_FEATUREFLAGS(drr
->drr_u
.drr_begin
.drr_versioninfo
,
1029 drr
->drr_u
.drr_begin
.drr_creation_time
=
1030 dsl_dataset_phys(to_ds
)->ds_creation_time
;
1031 drr
->drr_u
.drr_begin
.drr_type
= dmu_objset_type(os
);
1033 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CLONE
;
1034 drr
->drr_u
.drr_begin
.drr_toguid
= dsl_dataset_phys(to_ds
)->ds_guid
;
1035 if (dsl_dataset_phys(to_ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
1036 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CI_DATA
;
1037 if (zfs_send_set_freerecords_bit
)
1038 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_FREERECORDS
;
1040 if (ancestor_zb
!= NULL
) {
1041 drr
->drr_u
.drr_begin
.drr_fromguid
=
1042 ancestor_zb
->zbm_guid
;
1043 fromtxg
= ancestor_zb
->zbm_creation_txg
;
1045 dsl_dataset_name(to_ds
, drr
->drr_u
.drr_begin
.drr_toname
);
1046 if (!to_ds
->ds_is_snapshot
) {
1047 (void) strlcat(drr
->drr_u
.drr_begin
.drr_toname
, "@--head--",
1048 sizeof (drr
->drr_u
.drr_begin
.drr_toname
));
1051 dsp
= kmem_zalloc(sizeof (dmu_sendarg_t
), KM_SLEEP
);
1055 dsp
->dsa_outfd
= outfd
;
1056 dsp
->dsa_proc
= curproc
;
1059 dsp
->dsa_toguid
= dsl_dataset_phys(to_ds
)->ds_guid
;
1060 dsp
->dsa_pending_op
= PENDING_NONE
;
1061 dsp
->dsa_featureflags
= featureflags
;
1062 dsp
->dsa_resume_object
= resumeobj
;
1063 dsp
->dsa_resume_offset
= resumeoff
;
1065 mutex_enter(&to_ds
->ds_sendstream_lock
);
1066 list_insert_head(&to_ds
->ds_sendstreams
, dsp
);
1067 mutex_exit(&to_ds
->ds_sendstream_lock
);
1069 dsl_dataset_long_hold(to_ds
, FTAG
);
1070 dsl_pool_rele(dp
, tag
);
1072 /* handle features that require a DRR_BEGIN payload */
1074 (DMU_BACKUP_FEATURE_RESUMING
| DMU_BACKUP_FEATURE_RAW
)) {
1075 nvlist_t
*keynvl
= NULL
;
1076 nvlist_t
*nvl
= fnvlist_alloc();
1078 if (featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
1079 dmu_object_info_t to_doi
;
1080 err
= dmu_object_info(os
, resumeobj
, &to_doi
);
1086 SET_BOOKMARK(&to_arg
.resume
, to_ds
->ds_object
,
1088 resumeoff
/ to_doi
.doi_data_block_size
);
1090 fnvlist_add_uint64(nvl
, "resume_object", resumeobj
);
1091 fnvlist_add_uint64(nvl
, "resume_offset", resumeoff
);
1094 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1095 ASSERT(os
->os_encrypted
);
1097 err
= dsl_crypto_populate_key_nvlist(to_ds
, &keynvl
);
1103 fnvlist_add_nvlist(nvl
, "crypt_keydata", keynvl
);
1106 payload
= fnvlist_pack(nvl
, &payload_len
);
1107 drr
->drr_payloadlen
= payload_len
;
1108 fnvlist_free(keynvl
);
1112 err
= dump_record(dsp
, payload
, payload_len
);
1113 fnvlist_pack_free(payload
, payload_len
);
1119 err
= bqueue_init(&to_arg
.q
, zfs_send_queue_length
,
1120 offsetof(struct send_block_record
, ln
));
1121 to_arg
.error_code
= 0;
1122 to_arg
.cancel
= B_FALSE
;
1124 to_arg
.fromtxg
= fromtxg
;
1125 to_arg
.flags
= TRAVERSE_PRE
| TRAVERSE_PREFETCH
;
1127 to_arg
.flags
|= TRAVERSE_NO_DECRYPT
;
1128 (void) thread_create(NULL
, 0, send_traverse_thread
, &to_arg
, 0, curproc
,
1129 TS_RUN
, minclsyspri
);
1131 to_data
= bqueue_dequeue(&to_arg
.q
);
1133 while (!to_data
->eos_marker
&& err
== 0) {
1134 err
= do_dump(dsp
, to_data
);
1135 to_data
= get_next_record(&to_arg
.q
, to_data
);
1136 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1141 to_arg
.cancel
= B_TRUE
;
1142 while (!to_data
->eos_marker
) {
1143 to_data
= get_next_record(&to_arg
.q
, to_data
);
1146 kmem_free(to_data
, sizeof (*to_data
));
1148 bqueue_destroy(&to_arg
.q
);
1150 if (err
== 0 && to_arg
.error_code
!= 0)
1151 err
= to_arg
.error_code
;
1156 if (dsp
->dsa_pending_op
!= PENDING_NONE
)
1157 if (dump_record(dsp
, NULL
, 0) != 0)
1158 err
= SET_ERROR(EINTR
);
1161 if (err
== EINTR
&& dsp
->dsa_err
!= 0)
1166 bzero(drr
, sizeof (dmu_replay_record_t
));
1167 drr
->drr_type
= DRR_END
;
1168 drr
->drr_u
.drr_end
.drr_checksum
= dsp
->dsa_zc
;
1169 drr
->drr_u
.drr_end
.drr_toguid
= dsp
->dsa_toguid
;
1171 if (dump_record(dsp
, NULL
, 0) != 0)
1174 mutex_enter(&to_ds
->ds_sendstream_lock
);
1175 list_remove(&to_ds
->ds_sendstreams
, dsp
);
1176 mutex_exit(&to_ds
->ds_sendstream_lock
);
1178 VERIFY(err
!= 0 || (dsp
->dsa_sent_begin
&& dsp
->dsa_sent_end
));
1180 kmem_free(drr
, sizeof (dmu_replay_record_t
));
1181 kmem_free(dsp
, sizeof (dmu_sendarg_t
));
1183 dsl_dataset_long_rele(to_ds
, FTAG
);
1189 dmu_send_obj(const char *pool
, uint64_t tosnap
, uint64_t fromsnap
,
1190 boolean_t embedok
, boolean_t large_block_ok
, boolean_t compressok
,
1191 boolean_t rawok
, int outfd
, vnode_t
*vp
, offset_t
*off
)
1195 dsl_dataset_t
*fromds
= NULL
;
1196 ds_hold_flags_t dsflags
= (rawok
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
1199 err
= dsl_pool_hold(pool
, FTAG
, &dp
);
1203 err
= dsl_dataset_hold_obj_flags(dp
, tosnap
, dsflags
, FTAG
, &ds
);
1205 dsl_pool_rele(dp
, FTAG
);
1209 if (fromsnap
!= 0) {
1210 zfs_bookmark_phys_t zb
;
1213 err
= dsl_dataset_hold_obj(dp
, fromsnap
, FTAG
, &fromds
);
1215 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1216 dsl_pool_rele(dp
, FTAG
);
1219 if (!dsl_dataset_is_before(ds
, fromds
, 0))
1220 err
= SET_ERROR(EXDEV
);
1221 zb
.zbm_creation_time
=
1222 dsl_dataset_phys(fromds
)->ds_creation_time
;
1223 zb
.zbm_creation_txg
= dsl_dataset_phys(fromds
)->ds_creation_txg
;
1224 zb
.zbm_guid
= dsl_dataset_phys(fromds
)->ds_guid
;
1225 is_clone
= (fromds
->ds_dir
!= ds
->ds_dir
);
1226 dsl_dataset_rele(fromds
, FTAG
);
1227 err
= dmu_send_impl(FTAG
, dp
, ds
, &zb
, is_clone
,
1228 embedok
, large_block_ok
, compressok
, rawok
, outfd
,
1231 err
= dmu_send_impl(FTAG
, dp
, ds
, NULL
, B_FALSE
,
1232 embedok
, large_block_ok
, compressok
, rawok
, outfd
,
1235 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1240 dmu_send(const char *tosnap
, const char *fromsnap
, boolean_t embedok
,
1241 boolean_t large_block_ok
, boolean_t compressok
, boolean_t rawok
,
1242 int outfd
, uint64_t resumeobj
, uint64_t resumeoff
, vnode_t
*vp
,
1248 ds_hold_flags_t dsflags
= (rawok
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
1249 boolean_t owned
= B_FALSE
;
1251 if (fromsnap
!= NULL
&& strpbrk(fromsnap
, "@#") == NULL
)
1252 return (SET_ERROR(EINVAL
));
1254 err
= dsl_pool_hold(tosnap
, FTAG
, &dp
);
1258 if (strchr(tosnap
, '@') == NULL
&& spa_writeable(dp
->dp_spa
)) {
1260 * We are sending a filesystem or volume. Ensure
1261 * that it doesn't change by owning the dataset.
1263 err
= dsl_dataset_own(dp
, tosnap
, dsflags
, FTAG
, &ds
);
1266 err
= dsl_dataset_hold_flags(dp
, tosnap
, dsflags
, FTAG
, &ds
);
1269 dsl_pool_rele(dp
, FTAG
);
1273 if (fromsnap
!= NULL
) {
1274 zfs_bookmark_phys_t zb
;
1275 boolean_t is_clone
= B_FALSE
;
1276 int fsnamelen
= strchr(tosnap
, '@') - tosnap
;
1279 * If the fromsnap is in a different filesystem, then
1280 * mark the send stream as a clone.
1282 if (strncmp(tosnap
, fromsnap
, fsnamelen
) != 0 ||
1283 (fromsnap
[fsnamelen
] != '@' &&
1284 fromsnap
[fsnamelen
] != '#')) {
1288 if (strchr(fromsnap
, '@')) {
1289 dsl_dataset_t
*fromds
;
1290 err
= dsl_dataset_hold(dp
, fromsnap
, FTAG
, &fromds
);
1292 if (!dsl_dataset_is_before(ds
, fromds
, 0))
1293 err
= SET_ERROR(EXDEV
);
1294 zb
.zbm_creation_time
=
1295 dsl_dataset_phys(fromds
)->ds_creation_time
;
1296 zb
.zbm_creation_txg
=
1297 dsl_dataset_phys(fromds
)->ds_creation_txg
;
1298 zb
.zbm_guid
= dsl_dataset_phys(fromds
)->ds_guid
;
1299 is_clone
= (ds
->ds_dir
!= fromds
->ds_dir
);
1300 dsl_dataset_rele(fromds
, FTAG
);
1303 err
= dsl_bookmark_lookup(dp
, fromsnap
, ds
, &zb
);
1307 dsl_dataset_disown(ds
, dsflags
, FTAG
);
1309 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1311 dsl_pool_rele(dp
, FTAG
);
1314 err
= dmu_send_impl(FTAG
, dp
, ds
, &zb
, is_clone
,
1315 embedok
, large_block_ok
, compressok
, rawok
,
1316 outfd
, resumeobj
, resumeoff
, vp
, off
);
1318 err
= dmu_send_impl(FTAG
, dp
, ds
, NULL
, B_FALSE
,
1319 embedok
, large_block_ok
, compressok
, rawok
,
1320 outfd
, resumeobj
, resumeoff
, vp
, off
);
1323 dsl_dataset_disown(ds
, dsflags
, FTAG
);
1325 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1331 dmu_adjust_send_estimate_for_indirects(dsl_dataset_t
*ds
, uint64_t uncompressed
,
1332 uint64_t compressed
, boolean_t stream_compressed
, uint64_t *sizep
)
1337 * Assume that space (both on-disk and in-stream) is dominated by
1338 * data. We will adjust for indirect blocks and the copies property,
1339 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
1342 uint64_t recordsize
;
1343 uint64_t record_count
;
1345 VERIFY0(dmu_objset_from_ds(ds
, &os
));
1347 /* Assume all (uncompressed) blocks are recordsize. */
1348 if (os
->os_phys
->os_type
== DMU_OST_ZVOL
) {
1349 err
= dsl_prop_get_int_ds(ds
,
1350 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
), &recordsize
);
1352 err
= dsl_prop_get_int_ds(ds
,
1353 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
), &recordsize
);
1357 record_count
= uncompressed
/ recordsize
;
1360 * If we're estimating a send size for a compressed stream, use the
1361 * compressed data size to estimate the stream size. Otherwise, use the
1362 * uncompressed data size.
1364 size
= stream_compressed
? compressed
: uncompressed
;
1367 * Subtract out approximate space used by indirect blocks.
1368 * Assume most space is used by data blocks (non-indirect, non-dnode).
1369 * Assume no ditto blocks or internal fragmentation.
1371 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
1374 size
-= record_count
* sizeof (blkptr_t
);
1376 /* Add in the space for the record associated with each block. */
1377 size
+= record_count
* sizeof (dmu_replay_record_t
);
1385 dmu_send_estimate(dsl_dataset_t
*ds
, dsl_dataset_t
*fromds
,
1386 boolean_t stream_compressed
, uint64_t *sizep
)
1389 uint64_t uncomp
, comp
;
1391 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1393 /* tosnap must be a snapshot */
1394 if (!ds
->ds_is_snapshot
)
1395 return (SET_ERROR(EINVAL
));
1397 /* fromsnap, if provided, must be a snapshot */
1398 if (fromds
!= NULL
&& !fromds
->ds_is_snapshot
)
1399 return (SET_ERROR(EINVAL
));
1402 * fromsnap must be an earlier snapshot from the same fs as tosnap,
1403 * or the origin's fs.
1405 if (fromds
!= NULL
&& !dsl_dataset_is_before(ds
, fromds
, 0))
1406 return (SET_ERROR(EXDEV
));
1408 /* Get compressed and uncompressed size estimates of changed data. */
1409 if (fromds
== NULL
) {
1410 uncomp
= dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1411 comp
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1414 err
= dsl_dataset_space_written(fromds
, ds
,
1415 &used
, &comp
, &uncomp
);
1420 err
= dmu_adjust_send_estimate_for_indirects(ds
, uncomp
, comp
,
1421 stream_compressed
, sizep
);
1423 * Add the size of the BEGIN and END records to the estimate.
1425 *sizep
+= 2 * sizeof (dmu_replay_record_t
);
1429 struct calculate_send_arg
{
1430 uint64_t uncompressed
;
1431 uint64_t compressed
;
1435 * Simple callback used to traverse the blocks of a snapshot and sum their
1436 * uncompressed and compressed sizes.
1440 dmu_calculate_send_traversal(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1441 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1443 struct calculate_send_arg
*space
= arg
;
1444 if (bp
!= NULL
&& !BP_IS_HOLE(bp
)) {
1445 space
->uncompressed
+= BP_GET_UCSIZE(bp
);
1446 space
->compressed
+= BP_GET_PSIZE(bp
);
1452 * Given a desination snapshot and a TXG, calculate the approximate size of a
1453 * send stream sent from that TXG. from_txg may be zero, indicating that the
1454 * whole snapshot will be sent.
1457 dmu_send_estimate_from_txg(dsl_dataset_t
*ds
, uint64_t from_txg
,
1458 boolean_t stream_compressed
, uint64_t *sizep
)
1461 struct calculate_send_arg size
= { 0 };
1463 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1465 /* tosnap must be a snapshot */
1466 if (!dsl_dataset_is_snapshot(ds
))
1467 return (SET_ERROR(EINVAL
));
1469 /* verify that from_txg is before the provided snapshot was taken */
1470 if (from_txg
>= dsl_dataset_phys(ds
)->ds_creation_txg
) {
1471 return (SET_ERROR(EXDEV
));
1474 * traverse the blocks of the snapshot with birth times after
1475 * from_txg, summing their uncompressed size
1477 err
= traverse_dataset(ds
, from_txg
,
1478 TRAVERSE_POST
| TRAVERSE_NO_DECRYPT
,
1479 dmu_calculate_send_traversal
, &size
);
1484 err
= dmu_adjust_send_estimate_for_indirects(ds
, size
.uncompressed
,
1485 size
.compressed
, stream_compressed
, sizep
);
1489 typedef struct dmu_recv_begin_arg
{
1490 const char *drba_origin
;
1491 dmu_recv_cookie_t
*drba_cookie
;
1493 uint64_t drba_snapobj
;
1494 } dmu_recv_begin_arg_t
;
1497 recv_begin_check_existing_impl(dmu_recv_begin_arg_t
*drba
, dsl_dataset_t
*ds
,
1502 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1504 /* temporary clone name must not exist */
1505 error
= zap_lookup(dp
->dp_meta_objset
,
1506 dsl_dir_phys(ds
->ds_dir
)->dd_child_dir_zapobj
, recv_clone_name
,
1508 if (error
!= ENOENT
)
1509 return (error
== 0 ? EBUSY
: error
);
1511 /* new snapshot name must not exist */
1512 error
= zap_lookup(dp
->dp_meta_objset
,
1513 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1514 drba
->drba_cookie
->drc_tosnap
, 8, 1, &val
);
1515 if (error
!= ENOENT
)
1516 return (error
== 0 ? EEXIST
: error
);
1519 * Check snapshot limit before receiving. We'll recheck again at the
1520 * end, but might as well abort before receiving if we're already over
1523 * Note that we do not check the file system limit with
1524 * dsl_dir_fscount_check because the temporary %clones don't count
1525 * against that limit.
1527 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1, ZFS_PROP_SNAPSHOT_LIMIT
,
1528 NULL
, drba
->drba_cred
);
1532 if (fromguid
!= 0) {
1533 dsl_dataset_t
*snap
;
1534 uint64_t obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1536 /* Find snapshot in this dir that matches fromguid. */
1538 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
1541 return (SET_ERROR(ENODEV
));
1542 if (snap
->ds_dir
!= ds
->ds_dir
) {
1543 dsl_dataset_rele(snap
, FTAG
);
1544 return (SET_ERROR(ENODEV
));
1546 if (dsl_dataset_phys(snap
)->ds_guid
== fromguid
)
1548 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
1549 dsl_dataset_rele(snap
, FTAG
);
1552 return (SET_ERROR(ENODEV
));
1554 if (drba
->drba_cookie
->drc_force
) {
1555 drba
->drba_snapobj
= obj
;
1558 * If we are not forcing, there must be no
1559 * changes since fromsnap.
1561 if (dsl_dataset_modified_since_snap(ds
, snap
)) {
1562 dsl_dataset_rele(snap
, FTAG
);
1563 return (SET_ERROR(ETXTBSY
));
1565 drba
->drba_snapobj
= ds
->ds_prev
->ds_object
;
1568 dsl_dataset_rele(snap
, FTAG
);
1570 /* if full, then must be forced */
1571 if (!drba
->drba_cookie
->drc_force
)
1572 return (SET_ERROR(EEXIST
));
1575 * We don't support using zfs recv -F to blow away
1576 * encrypted filesystems. This would require the
1577 * dsl dir to point to the old encryption key and
1578 * the new one at the same time during the receive.
1580 if (ds
->ds_dir
->dd_crypto_obj
!= 0)
1581 return (SET_ERROR(EINVAL
));
1583 drba
->drba_snapobj
= 0;
1591 dmu_recv_begin_check(void *arg
, dmu_tx_t
*tx
)
1593 dmu_recv_begin_arg_t
*drba
= arg
;
1594 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1595 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1596 uint64_t fromguid
= drrb
->drr_fromguid
;
1597 int flags
= drrb
->drr_flags
;
1598 ds_hold_flags_t dsflags
= 0;
1600 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1602 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1604 /* already checked */
1605 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1606 ASSERT(!(featureflags
& DMU_BACKUP_FEATURE_RESUMING
));
1608 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1609 DMU_COMPOUNDSTREAM
||
1610 drrb
->drr_type
>= DMU_OST_NUMTYPES
||
1611 ((flags
& DRR_FLAG_CLONE
) && drba
->drba_origin
== NULL
))
1612 return (SET_ERROR(EINVAL
));
1614 /* Verify pool version supports SA if SA_SPILL feature set */
1615 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
1616 spa_version(dp
->dp_spa
) < SPA_VERSION_SA
)
1617 return (SET_ERROR(ENOTSUP
));
1619 if (drba
->drba_cookie
->drc_resumable
&&
1620 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EXTENSIBLE_DATASET
))
1621 return (SET_ERROR(ENOTSUP
));
1624 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1625 * record to a plain WRITE record, so the pool must have the
1626 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1627 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1629 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
1630 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
))
1631 return (SET_ERROR(ENOTSUP
));
1632 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
1633 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
))
1634 return (SET_ERROR(ENOTSUP
));
1637 * The receiving code doesn't know how to translate large blocks
1638 * to smaller ones, so the pool must have the LARGE_BLOCKS
1639 * feature enabled if the stream has LARGE_BLOCKS. Same with
1642 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
1643 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_BLOCKS
))
1644 return (SET_ERROR(ENOTSUP
));
1645 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_DNODE
) &&
1646 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_DNODE
))
1647 return (SET_ERROR(ENOTSUP
));
1649 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
)) {
1650 /* raw receives require the encryption feature */
1651 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_ENCRYPTION
))
1652 return (SET_ERROR(ENOTSUP
));
1654 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1657 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1659 /* target fs already exists; recv into temp clone */
1661 /* Can't recv a clone into an existing fs */
1662 if (flags
& DRR_FLAG_CLONE
|| drba
->drba_origin
) {
1663 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1664 return (SET_ERROR(EINVAL
));
1667 error
= recv_begin_check_existing_impl(drba
, ds
, fromguid
);
1668 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1669 } else if (error
== ENOENT
) {
1670 /* target fs does not exist; must be a full backup or clone */
1671 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1674 * If it's a non-clone incremental, we are missing the
1675 * target fs, so fail the recv.
1677 if (fromguid
!= 0 && !(flags
& DRR_FLAG_CLONE
||
1679 return (SET_ERROR(ENOENT
));
1682 * If we're receiving a full send as a clone, and it doesn't
1683 * contain all the necessary free records and freeobject
1684 * records, reject it.
1686 if (fromguid
== 0 && drba
->drba_origin
&&
1687 !(flags
& DRR_FLAG_FREERECORDS
))
1688 return (SET_ERROR(EINVAL
));
1690 /* Open the parent of tofs */
1691 ASSERT3U(strlen(tofs
), <, sizeof (buf
));
1692 (void) strlcpy(buf
, tofs
, strrchr(tofs
, '/') - tofs
+ 1);
1693 error
= dsl_dataset_hold_flags(dp
, buf
, dsflags
, FTAG
, &ds
);
1698 * Check filesystem and snapshot limits before receiving. We'll
1699 * recheck snapshot limits again at the end (we create the
1700 * filesystems and increment those counts during begin_sync).
1702 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
1703 ZFS_PROP_FILESYSTEM_LIMIT
, NULL
, drba
->drba_cred
);
1705 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1709 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
1710 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, drba
->drba_cred
);
1712 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1716 if (drba
->drba_origin
!= NULL
) {
1717 dsl_dataset_t
*origin
;
1719 error
= dsl_dataset_hold_flags(dp
, drba
->drba_origin
,
1720 dsflags
, FTAG
, &origin
);
1722 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1725 if (!origin
->ds_is_snapshot
) {
1726 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
1727 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1728 return (SET_ERROR(EINVAL
));
1730 if (dsl_dataset_phys(origin
)->ds_guid
!= fromguid
&&
1732 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
1733 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1734 return (SET_ERROR(ENODEV
));
1736 dsl_dataset_rele_flags(origin
,
1739 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1746 dmu_recv_begin_sync(void *arg
, dmu_tx_t
*tx
)
1748 dmu_recv_begin_arg_t
*drba
= arg
;
1749 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1750 objset_t
*mos
= dp
->dp_meta_objset
;
1751 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1752 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1753 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1754 dsl_dataset_t
*ds
, *newds
;
1757 ds_hold_flags_t dsflags
= 0;
1759 uint64_t crflags
= 0;
1760 dsl_crypto_params_t
*dcpp
= NULL
;
1761 dsl_crypto_params_t dcp
= { 0 };
1763 if (drrb
->drr_flags
& DRR_FLAG_CI_DATA
)
1764 crflags
|= DS_FLAG_CI_DATASET
;
1765 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0) {
1766 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1768 dcp
.cp_cmd
= DCP_CMD_RAW_RECV
;
1771 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1773 /* create temporary clone */
1774 dsl_dataset_t
*snap
= NULL
;
1776 if (drba
->drba_snapobj
!= 0) {
1777 VERIFY0(dsl_dataset_hold_obj(dp
,
1778 drba
->drba_snapobj
, FTAG
, &snap
));
1780 /* we use the dcp whenever we are not making a clone */
1784 dsobj
= dsl_dataset_create_sync(ds
->ds_dir
, recv_clone_name
,
1785 snap
, crflags
, drba
->drba_cred
, dcpp
, tx
);
1786 if (drba
->drba_snapobj
!= 0)
1787 dsl_dataset_rele(snap
, FTAG
);
1788 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1792 dsl_dataset_t
*origin
= NULL
;
1794 VERIFY0(dsl_dir_hold(dp
, tofs
, FTAG
, &dd
, &tail
));
1796 if (drba
->drba_origin
!= NULL
) {
1797 VERIFY0(dsl_dataset_hold(dp
, drba
->drba_origin
,
1800 /* we use the dcp whenever we are not making a clone */
1804 /* Create new dataset. */
1805 dsobj
= dsl_dataset_create_sync(dd
, strrchr(tofs
, '/') + 1,
1806 origin
, crflags
, drba
->drba_cred
, dcpp
, tx
);
1808 dsl_dataset_rele(origin
, FTAG
);
1809 dsl_dir_rele(dd
, FTAG
);
1810 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
1812 VERIFY0(dsl_dataset_own_obj(dp
, dsobj
, dsflags
, dmu_recv_tag
, &newds
));
1813 VERIFY0(dmu_objset_from_ds(newds
, &os
));
1815 if (drba
->drba_cookie
->drc_resumable
) {
1819 dsl_dataset_zapify(newds
, tx
);
1820 if (drrb
->drr_fromguid
!= 0) {
1821 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_FROMGUID
,
1822 8, 1, &drrb
->drr_fromguid
, tx
));
1824 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TOGUID
,
1825 8, 1, &drrb
->drr_toguid
, tx
));
1826 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TONAME
,
1827 1, strlen(drrb
->drr_toname
) + 1, drrb
->drr_toname
, tx
));
1828 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OBJECT
,
1830 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OFFSET
,
1832 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_BYTES
,
1834 if (featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) {
1835 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_LARGEBLOCK
,
1838 if (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) {
1839 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_EMBEDOK
,
1842 if (featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
) {
1843 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_COMPRESSOK
,
1846 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1847 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_RAWOK
,
1853 * Usually the os->os_encrypted value is tied to the presence of a
1854 * DSL Crypto Key object in the dd. However, that will not be received
1855 * until dmu_recv_stream(), so we set the value manually for now.
1857 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1858 os
->os_encrypted
= B_TRUE
;
1859 drba
->drba_cookie
->drc_raw
= B_TRUE
;
1862 dmu_buf_will_dirty(newds
->ds_dbuf
, tx
);
1863 dsl_dataset_phys(newds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
1866 * If we actually created a non-clone, we need to create the objset
1867 * in our new dataset. If this is a raw send we postpone this until
1868 * dmu_recv_stream() so that we can allocate the metadnode with the
1869 * properties from the DRR_BEGIN payload.
1871 rrw_enter(&newds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1872 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds
)) &&
1873 (featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0) {
1874 (void) dmu_objset_create_impl(dp
->dp_spa
,
1875 newds
, dsl_dataset_get_blkptr(newds
), drrb
->drr_type
, tx
);
1877 rrw_exit(&newds
->ds_bp_rwlock
, FTAG
);
1879 drba
->drba_cookie
->drc_ds
= newds
;
1881 spa_history_log_internal_ds(newds
, "receive", tx
, "");
1885 dmu_recv_resume_begin_check(void *arg
, dmu_tx_t
*tx
)
1887 dmu_recv_begin_arg_t
*drba
= arg
;
1888 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1889 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1891 ds_hold_flags_t dsflags
= 0;
1892 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1894 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1897 /* 6 extra bytes for /%recv */
1898 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1900 /* already checked */
1901 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1902 ASSERT(featureflags
& DMU_BACKUP_FEATURE_RESUMING
);
1904 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1905 DMU_COMPOUNDSTREAM
||
1906 drrb
->drr_type
>= DMU_OST_NUMTYPES
)
1907 return (SET_ERROR(EINVAL
));
1909 /* Verify pool version supports SA if SA_SPILL feature set */
1910 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
1911 spa_version(dp
->dp_spa
) < SPA_VERSION_SA
)
1912 return (SET_ERROR(ENOTSUP
));
1915 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1916 * record to a plain WRITE record, so the pool must have the
1917 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1918 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1920 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
1921 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
))
1922 return (SET_ERROR(ENOTSUP
));
1923 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
1924 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
))
1925 return (SET_ERROR(ENOTSUP
));
1928 * The receiving code doesn't know how to translate large blocks
1929 * to smaller ones, so the pool must have the LARGE_BLOCKS
1930 * feature enabled if the stream has LARGE_BLOCKS. Same with
1933 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
1934 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_BLOCKS
))
1935 return (SET_ERROR(ENOTSUP
));
1936 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_DNODE
) &&
1937 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_DNODE
))
1938 return (SET_ERROR(ENOTSUP
));
1940 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
1941 tofs
, recv_clone_name
);
1943 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0)
1944 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1946 if (dsl_dataset_hold_flags(dp
, recvname
, dsflags
, FTAG
, &ds
) != 0) {
1947 /* %recv does not exist; continue in tofs */
1948 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1953 /* check that ds is marked inconsistent */
1954 if (!DS_IS_INCONSISTENT(ds
)) {
1955 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1956 return (SET_ERROR(EINVAL
));
1959 /* check that there is resuming data, and that the toguid matches */
1960 if (!dsl_dataset_is_zapified(ds
)) {
1961 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1962 return (SET_ERROR(EINVAL
));
1964 error
= zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1965 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
);
1966 if (error
!= 0 || drrb
->drr_toguid
!= val
) {
1967 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1968 return (SET_ERROR(EINVAL
));
1972 * Check if the receive is still running. If so, it will be owned.
1973 * Note that nothing else can own the dataset (e.g. after the receive
1974 * fails) because it will be marked inconsistent.
1976 if (dsl_dataset_has_owner(ds
)) {
1977 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1978 return (SET_ERROR(EBUSY
));
1981 /* There should not be any snapshots of this fs yet. */
1982 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
->ds_dir
== ds
->ds_dir
) {
1983 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1984 return (SET_ERROR(EINVAL
));
1988 * Note: resume point will be checked when we process the first WRITE
1992 /* check that the origin matches */
1994 (void) zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1995 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
);
1996 if (drrb
->drr_fromguid
!= val
) {
1997 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1998 return (SET_ERROR(EINVAL
));
2001 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
2006 dmu_recv_resume_begin_sync(void *arg
, dmu_tx_t
*tx
)
2008 dmu_recv_begin_arg_t
*drba
= arg
;
2009 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2010 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
2011 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
2012 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
2015 ds_hold_flags_t dsflags
= 0;
2017 /* 6 extra bytes for /%recv */
2018 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
2020 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
2021 tofs
, recv_clone_name
);
2023 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
2024 drba
->drba_cookie
->drc_raw
= B_TRUE
;
2026 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
2029 if (dsl_dataset_hold_flags(dp
, recvname
, dsflags
, FTAG
, &ds
) != 0) {
2030 /* %recv does not exist; continue in tofs */
2031 VERIFY0(dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
));
2032 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
2035 /* clear the inconsistent flag so that we can own it */
2036 ASSERT(DS_IS_INCONSISTENT(ds
));
2037 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2038 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
2039 dsobj
= ds
->ds_object
;
2040 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
2042 VERIFY0(dsl_dataset_own_obj(dp
, dsobj
, dsflags
, dmu_recv_tag
, &ds
));
2043 VERIFY0(dmu_objset_from_ds(ds
, &os
));
2045 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2046 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
2048 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2049 ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds
)));
2050 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2052 drba
->drba_cookie
->drc_ds
= ds
;
2054 spa_history_log_internal_ds(ds
, "resume receive", tx
, "");
2058 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
2059 * succeeds; otherwise we will leak the holds on the datasets.
2062 dmu_recv_begin(char *tofs
, char *tosnap
, dmu_replay_record_t
*drr_begin
,
2063 boolean_t force
, boolean_t resumable
, char *origin
, dmu_recv_cookie_t
*drc
)
2065 dmu_recv_begin_arg_t drba
= { 0 };
2067 bzero(drc
, sizeof (dmu_recv_cookie_t
));
2068 drc
->drc_drr_begin
= drr_begin
;
2069 drc
->drc_drrb
= &drr_begin
->drr_u
.drr_begin
;
2070 drc
->drc_tosnap
= tosnap
;
2071 drc
->drc_tofs
= tofs
;
2072 drc
->drc_force
= force
;
2073 drc
->drc_resumable
= resumable
;
2074 drc
->drc_cred
= CRED();
2076 if (drc
->drc_drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
)) {
2077 drc
->drc_byteswap
= B_TRUE
;
2078 (void) fletcher_4_incremental_byteswap(drr_begin
,
2079 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
2080 byteswap_record(drr_begin
);
2081 } else if (drc
->drc_drrb
->drr_magic
== DMU_BACKUP_MAGIC
) {
2082 (void) fletcher_4_incremental_native(drr_begin
,
2083 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
2085 return (SET_ERROR(EINVAL
));
2088 drba
.drba_origin
= origin
;
2089 drba
.drba_cookie
= drc
;
2090 drba
.drba_cred
= CRED();
2092 if (DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
) &
2093 DMU_BACKUP_FEATURE_RESUMING
) {
2094 return (dsl_sync_task(tofs
,
2095 dmu_recv_resume_begin_check
, dmu_recv_resume_begin_sync
,
2096 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
));
2098 return (dsl_sync_task(tofs
,
2099 dmu_recv_begin_check
, dmu_recv_begin_sync
,
2100 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
));
2104 struct receive_record_arg
{
2105 dmu_replay_record_t header
;
2106 void *payload
; /* Pointer to a buffer containing the payload */
2108 * If the record is a write, pointer to the arc_buf_t containing the
2113 uint64_t bytes_read
; /* bytes read from stream when record created */
2114 boolean_t eos_marker
; /* Marks the end of the stream */
2118 struct receive_writer_arg
{
2124 * These three args are used to signal to the main thread that we're
2132 /* A map from guid to dataset to help handle dedup'd streams. */
2133 avl_tree_t
*guid_to_ds_map
;
2134 boolean_t resumable
;
2136 uint64_t last_object
, last_offset
;
2137 uint64_t bytes_read
; /* bytes read when current record created */
2141 list_t list
; /* List of struct receive_objnode. */
2143 * Last object looked up. Used to assert that objects are being looked
2144 * up in ascending order.
2146 uint64_t last_lookup
;
2149 struct receive_objnode
{
2154 struct receive_arg
{
2156 vnode_t
*vp
; /* The vnode to read the stream from */
2157 uint64_t voff
; /* The current offset in the stream */
2158 uint64_t bytes_read
;
2160 * A record that has had its payload read in, but hasn't yet been handed
2161 * off to the worker thread.
2163 struct receive_record_arg
*rrd
;
2164 /* A record that has had its header read in, but not its payload. */
2165 struct receive_record_arg
*next_rrd
;
2167 zio_cksum_t prev_cksum
;
2171 uint64_t featureflags
;
2172 /* Sorted list of objects not to issue prefetches for. */
2173 struct objlist ignore_objlist
;
2176 typedef struct guid_map_entry
{
2179 dsl_dataset_t
*gme_ds
;
2184 guid_compare(const void *arg1
, const void *arg2
)
2186 const guid_map_entry_t
*gmep1
= (const guid_map_entry_t
*)arg1
;
2187 const guid_map_entry_t
*gmep2
= (const guid_map_entry_t
*)arg2
;
2189 return (AVL_CMP(gmep1
->guid
, gmep2
->guid
));
2193 free_guid_map_onexit(void *arg
)
2195 avl_tree_t
*ca
= arg
;
2196 void *cookie
= NULL
;
2197 guid_map_entry_t
*gmep
;
2199 while ((gmep
= avl_destroy_nodes(ca
, &cookie
)) != NULL
) {
2200 dsl_dataset_long_rele(gmep
->gme_ds
, gmep
);
2201 dsl_dataset_rele_flags(gmep
->gme_ds
,
2202 (gmep
->raw
) ? 0 : DS_HOLD_FLAG_DECRYPT
, gmep
);
2203 kmem_free(gmep
, sizeof (guid_map_entry_t
));
2206 kmem_free(ca
, sizeof (avl_tree_t
));
2210 receive_read(struct receive_arg
*ra
, int len
, void *buf
)
2215 * The code doesn't rely on this (lengths being multiples of 8). See
2216 * comment in dump_bytes.
2218 ASSERT(len
% 8 == 0 ||
2219 (ra
->featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0);
2221 while (done
< len
) {
2224 ra
->err
= vn_rdwr(UIO_READ
, ra
->vp
,
2225 (char *)buf
+ done
, len
- done
,
2226 ra
->voff
, UIO_SYSSPACE
, FAPPEND
,
2227 RLIM64_INFINITY
, CRED(), &resid
);
2229 if (resid
== len
- done
) {
2231 * Note: ECKSUM indicates that the receive
2232 * was interrupted and can potentially be resumed.
2234 ra
->err
= SET_ERROR(ECKSUM
);
2236 ra
->voff
+= len
- done
- resid
;
2242 ra
->bytes_read
+= len
;
2244 ASSERT3U(done
, ==, len
);
2248 noinline
static void
2249 byteswap_record(dmu_replay_record_t
*drr
)
2251 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
2252 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
2253 drr
->drr_type
= BSWAP_32(drr
->drr_type
);
2254 drr
->drr_payloadlen
= BSWAP_32(drr
->drr_payloadlen
);
2256 switch (drr
->drr_type
) {
2258 DO64(drr_begin
.drr_magic
);
2259 DO64(drr_begin
.drr_versioninfo
);
2260 DO64(drr_begin
.drr_creation_time
);
2261 DO32(drr_begin
.drr_type
);
2262 DO32(drr_begin
.drr_flags
);
2263 DO64(drr_begin
.drr_toguid
);
2264 DO64(drr_begin
.drr_fromguid
);
2267 DO64(drr_object
.drr_object
);
2268 DO32(drr_object
.drr_type
);
2269 DO32(drr_object
.drr_bonustype
);
2270 DO32(drr_object
.drr_blksz
);
2271 DO32(drr_object
.drr_bonuslen
);
2272 DO32(drr_object
.drr_raw_bonuslen
);
2273 DO64(drr_object
.drr_toguid
);
2275 case DRR_FREEOBJECTS
:
2276 DO64(drr_freeobjects
.drr_firstobj
);
2277 DO64(drr_freeobjects
.drr_numobjs
);
2278 DO64(drr_freeobjects
.drr_toguid
);
2281 DO64(drr_write
.drr_object
);
2282 DO32(drr_write
.drr_type
);
2283 DO64(drr_write
.drr_offset
);
2284 DO64(drr_write
.drr_logical_size
);
2285 DO64(drr_write
.drr_toguid
);
2286 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write
.drr_key
.ddk_cksum
);
2287 DO64(drr_write
.drr_key
.ddk_prop
);
2288 DO64(drr_write
.drr_compressed_size
);
2290 case DRR_WRITE_BYREF
:
2291 DO64(drr_write_byref
.drr_object
);
2292 DO64(drr_write_byref
.drr_offset
);
2293 DO64(drr_write_byref
.drr_length
);
2294 DO64(drr_write_byref
.drr_toguid
);
2295 DO64(drr_write_byref
.drr_refguid
);
2296 DO64(drr_write_byref
.drr_refobject
);
2297 DO64(drr_write_byref
.drr_refoffset
);
2298 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write_byref
.
2300 DO64(drr_write_byref
.drr_key
.ddk_prop
);
2302 case DRR_WRITE_EMBEDDED
:
2303 DO64(drr_write_embedded
.drr_object
);
2304 DO64(drr_write_embedded
.drr_offset
);
2305 DO64(drr_write_embedded
.drr_length
);
2306 DO64(drr_write_embedded
.drr_toguid
);
2307 DO32(drr_write_embedded
.drr_lsize
);
2308 DO32(drr_write_embedded
.drr_psize
);
2311 DO64(drr_free
.drr_object
);
2312 DO64(drr_free
.drr_offset
);
2313 DO64(drr_free
.drr_length
);
2314 DO64(drr_free
.drr_toguid
);
2317 DO64(drr_spill
.drr_object
);
2318 DO64(drr_spill
.drr_length
);
2319 DO64(drr_spill
.drr_toguid
);
2320 DO64(drr_spill
.drr_compressed_size
);
2321 DO32(drr_spill
.drr_type
);
2323 case DRR_OBJECT_RANGE
:
2324 DO64(drr_object_range
.drr_firstobj
);
2325 DO64(drr_object_range
.drr_numslots
);
2326 DO64(drr_object_range
.drr_toguid
);
2329 DO64(drr_end
.drr_toguid
);
2330 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_end
.drr_checksum
);
2336 if (drr
->drr_type
!= DRR_BEGIN
) {
2337 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_checksum
.drr_checksum
);
2344 static inline uint8_t
2345 deduce_nblkptr(dmu_object_type_t bonus_type
, uint64_t bonus_size
)
2347 if (bonus_type
== DMU_OT_SA
) {
2351 ((DN_OLD_MAX_BONUSLEN
-
2352 MIN(DN_OLD_MAX_BONUSLEN
, bonus_size
)) >> SPA_BLKPTRSHIFT
));
2357 save_resume_state(struct receive_writer_arg
*rwa
,
2358 uint64_t object
, uint64_t offset
, dmu_tx_t
*tx
)
2360 int txgoff
= dmu_tx_get_txg(tx
) & TXG_MASK
;
2362 if (!rwa
->resumable
)
2366 * We use ds_resume_bytes[] != 0 to indicate that we need to
2367 * update this on disk, so it must not be 0.
2369 ASSERT(rwa
->bytes_read
!= 0);
2372 * We only resume from write records, which have a valid
2373 * (non-meta-dnode) object number.
2375 ASSERT(object
!= 0);
2378 * For resuming to work correctly, we must receive records in order,
2379 * sorted by object,offset. This is checked by the callers, but
2380 * assert it here for good measure.
2382 ASSERT3U(object
, >=, rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
]);
2383 ASSERT(object
!= rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] ||
2384 offset
>= rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
]);
2385 ASSERT3U(rwa
->bytes_read
, >=,
2386 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
]);
2388 rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] = object
;
2389 rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
] = offset
;
2390 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
] = rwa
->bytes_read
;
2394 receive_object(struct receive_writer_arg
*rwa
, struct drr_object
*drro
,
2397 dmu_object_info_t doi
;
2402 if (drro
->drr_type
== DMU_OT_NONE
||
2403 !DMU_OT_IS_VALID(drro
->drr_type
) ||
2404 !DMU_OT_IS_VALID(drro
->drr_bonustype
) ||
2405 drro
->drr_checksumtype
>= ZIO_CHECKSUM_FUNCTIONS
||
2406 drro
->drr_compress
>= ZIO_COMPRESS_FUNCTIONS
||
2407 P2PHASE(drro
->drr_blksz
, SPA_MINBLOCKSIZE
) ||
2408 drro
->drr_blksz
< SPA_MINBLOCKSIZE
||
2409 drro
->drr_blksz
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)) ||
2410 drro
->drr_bonuslen
>
2411 DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa
->os
))) ||
2412 drro
->drr_dn_slots
>
2413 (spa_maxdnodesize(dmu_objset_spa(rwa
->os
)) >> DNODE_SHIFT
)) {
2414 return (SET_ERROR(EINVAL
));
2418 if (drro
->drr_raw_bonuslen
< drro
->drr_bonuslen
||
2419 drro
->drr_indblkshift
> SPA_MAXBLOCKSHIFT
||
2420 drro
->drr_nlevels
> DN_MAX_LEVELS
||
2421 drro
->drr_nblkptr
> DN_MAX_NBLKPTR
||
2422 DN_SLOTS_TO_BONUSLEN(drro
->drr_dn_slots
) <
2423 drro
->drr_raw_bonuslen
)
2424 return (SET_ERROR(EINVAL
));
2426 if (drro
->drr_flags
!= 0 || drro
->drr_raw_bonuslen
!= 0 ||
2427 drro
->drr_indblkshift
!= 0 || drro
->drr_nlevels
!= 0 ||
2428 drro
->drr_nblkptr
!= 0)
2429 return (SET_ERROR(EINVAL
));
2432 err
= dmu_object_info(rwa
->os
, drro
->drr_object
, &doi
);
2434 if (err
!= 0 && err
!= ENOENT
)
2435 return (SET_ERROR(EINVAL
));
2436 object
= err
== 0 ? drro
->drr_object
: DMU_NEW_OBJECT
;
2439 * If we are losing blkptrs or changing the block size this must
2440 * be a new file instance. We must clear out the previous file
2441 * contents before we can change this type of metadata in the dnode.
2442 * Raw receives will also check that the indirect structure of the
2443 * dnode hasn't changed.
2446 uint32_t indblksz
= drro
->drr_indblkshift
?
2447 1ULL << drro
->drr_indblkshift
: 0;
2448 int nblkptr
= deduce_nblkptr(drro
->drr_bonustype
,
2449 drro
->drr_bonuslen
);
2451 /* nblkptr will be bounded by the bonus size and type */
2452 if (rwa
->raw
&& nblkptr
!= drro
->drr_nblkptr
)
2453 return (SET_ERROR(EINVAL
));
2455 if (drro
->drr_blksz
!= doi
.doi_data_block_size
||
2456 nblkptr
< doi
.doi_nblkptr
||
2458 (indblksz
!= doi
.doi_metadata_block_size
||
2459 drro
->drr_nlevels
< doi
.doi_indirection
))) {
2460 err
= dmu_free_long_range(rwa
->os
, drro
->drr_object
,
2463 return (SET_ERROR(EINVAL
));
2467 tx
= dmu_tx_create(rwa
->os
);
2468 dmu_tx_hold_bonus(tx
, object
);
2469 dmu_tx_hold_write(tx
, object
, 0, 0);
2470 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2476 if (object
== DMU_NEW_OBJECT
) {
2477 /* currently free, want to be allocated */
2478 err
= dmu_object_claim_dnsize(rwa
->os
, drro
->drr_object
,
2479 drro
->drr_type
, drro
->drr_blksz
,
2480 drro
->drr_bonustype
, drro
->drr_bonuslen
,
2481 drro
->drr_dn_slots
<< DNODE_SHIFT
, tx
);
2482 } else if (drro
->drr_type
!= doi
.doi_type
||
2483 drro
->drr_blksz
!= doi
.doi_data_block_size
||
2484 drro
->drr_bonustype
!= doi
.doi_bonus_type
||
2485 drro
->drr_bonuslen
!= doi
.doi_bonus_size
) {
2486 /* currently allocated, but with different properties */
2487 err
= dmu_object_reclaim(rwa
->os
, drro
->drr_object
,
2488 drro
->drr_type
, drro
->drr_blksz
,
2489 drro
->drr_bonustype
, drro
->drr_bonuslen
, tx
);
2493 return (SET_ERROR(EINVAL
));
2497 VERIFY0(dmu_object_dirty_raw(rwa
->os
, drro
->drr_object
, tx
));
2499 dmu_object_set_checksum(rwa
->os
, drro
->drr_object
,
2500 drro
->drr_checksumtype
, tx
);
2501 dmu_object_set_compress(rwa
->os
, drro
->drr_object
,
2502 drro
->drr_compress
, tx
);
2504 /* handle more restrictive dnode structuring for raw recvs */
2507 * Set the indirect block shift and nlevels. This will not fail
2508 * because we ensured all of the blocks were free earlier if
2509 * this is a new object.
2511 VERIFY0(dmu_object_set_blocksize(rwa
->os
, drro
->drr_object
,
2512 drro
->drr_blksz
, drro
->drr_indblkshift
, tx
));
2513 VERIFY0(dmu_object_set_nlevels(rwa
->os
, drro
->drr_object
,
2514 drro
->drr_nlevels
, tx
));
2519 uint32_t flags
= DMU_READ_NO_PREFETCH
;
2522 flags
|= DMU_READ_NO_DECRYPT
;
2524 VERIFY0(dmu_bonus_hold_impl(rwa
->os
, drro
->drr_object
,
2526 dmu_buf_will_dirty(db
, tx
);
2528 ASSERT3U(db
->db_size
, >=, drro
->drr_bonuslen
);
2529 bcopy(data
, db
->db_data
, DRR_OBJECT_PAYLOAD_SIZE(drro
));
2532 * Raw bonus buffers have their byteorder determined by the
2533 * DRR_OBJECT_RANGE record.
2535 if (rwa
->byteswap
&& !rwa
->raw
) {
2536 dmu_object_byteswap_t byteswap
=
2537 DMU_OT_BYTESWAP(drro
->drr_bonustype
);
2538 dmu_ot_byteswap
[byteswap
].ob_func(db
->db_data
,
2539 DRR_OBJECT_PAYLOAD_SIZE(drro
));
2541 dmu_buf_rele(db
, FTAG
);
2550 receive_freeobjects(struct receive_writer_arg
*rwa
,
2551 struct drr_freeobjects
*drrfo
)
2556 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
< drrfo
->drr_firstobj
)
2557 return (SET_ERROR(EINVAL
));
2559 for (obj
= drrfo
->drr_firstobj
== 0 ? 1 : drrfo
->drr_firstobj
;
2560 obj
< drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
&& next_err
== 0;
2561 next_err
= dmu_object_next(rwa
->os
, &obj
, FALSE
, 0)) {
2562 dmu_object_info_t doi
;
2565 err
= dmu_object_info(rwa
->os
, obj
, &doi
);
2566 if (err
== ENOENT
) {
2569 } else if (err
!= 0) {
2573 err
= dmu_free_long_object(rwa
->os
, obj
);
2577 if (next_err
!= ESRCH
)
2583 receive_write(struct receive_writer_arg
*rwa
, struct drr_write
*drrw
,
2590 if (drrw
->drr_offset
+ drrw
->drr_logical_size
< drrw
->drr_offset
||
2591 !DMU_OT_IS_VALID(drrw
->drr_type
))
2592 return (SET_ERROR(EINVAL
));
2595 * For resuming to work, records must be in increasing order
2596 * by (object, offset).
2598 if (drrw
->drr_object
< rwa
->last_object
||
2599 (drrw
->drr_object
== rwa
->last_object
&&
2600 drrw
->drr_offset
< rwa
->last_offset
)) {
2601 return (SET_ERROR(EINVAL
));
2603 rwa
->last_object
= drrw
->drr_object
;
2604 rwa
->last_offset
= drrw
->drr_offset
;
2606 if (dmu_object_info(rwa
->os
, drrw
->drr_object
, NULL
) != 0)
2607 return (SET_ERROR(EINVAL
));
2609 tx
= dmu_tx_create(rwa
->os
);
2611 dmu_tx_hold_write(tx
, drrw
->drr_object
,
2612 drrw
->drr_offset
, drrw
->drr_logical_size
);
2613 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2620 VERIFY0(dmu_object_dirty_raw(rwa
->os
, drrw
->drr_object
, tx
));
2622 if (rwa
->byteswap
&& !arc_is_encrypted(abuf
) &&
2623 arc_get_compression(abuf
) == ZIO_COMPRESS_OFF
) {
2624 dmu_object_byteswap_t byteswap
=
2625 DMU_OT_BYTESWAP(drrw
->drr_type
);
2626 dmu_ot_byteswap
[byteswap
].ob_func(abuf
->b_data
,
2627 DRR_WRITE_PAYLOAD_SIZE(drrw
));
2630 /* use the bonus buf to look up the dnode in dmu_assign_arcbuf */
2631 if (dmu_bonus_hold(rwa
->os
, drrw
->drr_object
, FTAG
, &bonus
) != 0)
2632 return (SET_ERROR(EINVAL
));
2633 dmu_assign_arcbuf(bonus
, drrw
->drr_offset
, abuf
, tx
);
2636 * Note: If the receive fails, we want the resume stream to start
2637 * with the same record that we last successfully received (as opposed
2638 * to the next record), so that we can verify that we are
2639 * resuming from the correct location.
2641 save_resume_state(rwa
, drrw
->drr_object
, drrw
->drr_offset
, tx
);
2643 dmu_buf_rele(bonus
, FTAG
);
2649 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
2650 * streams to refer to a copy of the data that is already on the
2651 * system because it came in earlier in the stream. This function
2652 * finds the earlier copy of the data, and uses that copy instead of
2653 * data from the stream to fulfill this write.
2656 receive_write_byref(struct receive_writer_arg
*rwa
,
2657 struct drr_write_byref
*drrwbr
)
2661 guid_map_entry_t gmesrch
;
2662 guid_map_entry_t
*gmep
;
2664 objset_t
*ref_os
= NULL
;
2665 int flags
= DMU_READ_PREFETCH
;
2668 if (drrwbr
->drr_offset
+ drrwbr
->drr_length
< drrwbr
->drr_offset
)
2669 return (SET_ERROR(EINVAL
));
2672 * If the GUID of the referenced dataset is different from the
2673 * GUID of the target dataset, find the referenced dataset.
2675 if (drrwbr
->drr_toguid
!= drrwbr
->drr_refguid
) {
2676 gmesrch
.guid
= drrwbr
->drr_refguid
;
2677 if ((gmep
= avl_find(rwa
->guid_to_ds_map
, &gmesrch
,
2679 return (SET_ERROR(EINVAL
));
2681 if (dmu_objset_from_ds(gmep
->gme_ds
, &ref_os
))
2682 return (SET_ERROR(EINVAL
));
2688 flags
|= DMU_READ_NO_DECRYPT
;
2690 /* may return either a regular db or an encrypted one */
2691 err
= dmu_buf_hold(ref_os
, drrwbr
->drr_refobject
,
2692 drrwbr
->drr_refoffset
, FTAG
, &dbp
, flags
);
2696 tx
= dmu_tx_create(rwa
->os
);
2698 dmu_tx_hold_write(tx
, drrwbr
->drr_object
,
2699 drrwbr
->drr_offset
, drrwbr
->drr_length
);
2700 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2707 VERIFY0(dmu_object_dirty_raw(rwa
->os
, drrwbr
->drr_object
, tx
));
2708 dmu_copy_from_buf(rwa
->os
, drrwbr
->drr_object
,
2709 drrwbr
->drr_offset
, dbp
, tx
);
2711 dmu_write(rwa
->os
, drrwbr
->drr_object
,
2712 drrwbr
->drr_offset
, drrwbr
->drr_length
, dbp
->db_data
, tx
);
2714 dmu_buf_rele(dbp
, FTAG
);
2716 /* See comment in restore_write. */
2717 save_resume_state(rwa
, drrwbr
->drr_object
, drrwbr
->drr_offset
, tx
);
2723 receive_write_embedded(struct receive_writer_arg
*rwa
,
2724 struct drr_write_embedded
*drrwe
, void *data
)
2729 if (drrwe
->drr_offset
+ drrwe
->drr_length
< drrwe
->drr_offset
)
2730 return (SET_ERROR(EINVAL
));
2732 if (drrwe
->drr_psize
> BPE_PAYLOAD_SIZE
)
2733 return (SET_ERROR(EINVAL
));
2735 if (drrwe
->drr_etype
>= NUM_BP_EMBEDDED_TYPES
)
2736 return (SET_ERROR(EINVAL
));
2737 if (drrwe
->drr_compression
>= ZIO_COMPRESS_FUNCTIONS
)
2738 return (SET_ERROR(EINVAL
));
2740 tx
= dmu_tx_create(rwa
->os
);
2742 dmu_tx_hold_write(tx
, drrwe
->drr_object
,
2743 drrwe
->drr_offset
, drrwe
->drr_length
);
2744 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2750 dmu_write_embedded(rwa
->os
, drrwe
->drr_object
,
2751 drrwe
->drr_offset
, data
, drrwe
->drr_etype
,
2752 drrwe
->drr_compression
, drrwe
->drr_lsize
, drrwe
->drr_psize
,
2753 rwa
->byteswap
^ ZFS_HOST_BYTEORDER
, tx
);
2755 /* See comment in restore_write. */
2756 save_resume_state(rwa
, drrwe
->drr_object
, drrwe
->drr_offset
, tx
);
2762 receive_spill(struct receive_writer_arg
*rwa
, struct drr_spill
*drrs
,
2766 dmu_buf_t
*db
, *db_spill
;
2769 if (drrs
->drr_length
< SPA_MINBLOCKSIZE
||
2770 drrs
->drr_length
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)))
2771 return (SET_ERROR(EINVAL
));
2774 if (!DMU_OT_IS_VALID(drrs
->drr_type
) ||
2775 drrs
->drr_compressiontype
>= ZIO_COMPRESS_FUNCTIONS
||
2776 drrs
->drr_compressed_size
== 0)
2777 return (SET_ERROR(EINVAL
));
2780 if (dmu_object_info(rwa
->os
, drrs
->drr_object
, NULL
) != 0)
2781 return (SET_ERROR(EINVAL
));
2783 VERIFY0(dmu_bonus_hold(rwa
->os
, drrs
->drr_object
, FTAG
, &db
));
2784 if ((err
= dmu_spill_hold_by_bonus(db
, FTAG
, &db_spill
)) != 0) {
2785 dmu_buf_rele(db
, FTAG
);
2789 tx
= dmu_tx_create(rwa
->os
);
2791 dmu_tx_hold_spill(tx
, db
->db_object
);
2793 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2795 dmu_buf_rele(db
, FTAG
);
2796 dmu_buf_rele(db_spill
, FTAG
);
2800 dmu_buf_will_dirty(db_spill
, tx
);
2802 VERIFY0(dmu_object_dirty_raw(rwa
->os
, drrs
->drr_object
, tx
));
2804 if (db_spill
->db_size
< drrs
->drr_length
)
2805 VERIFY(0 == dbuf_spill_set_blksz(db_spill
,
2806 drrs
->drr_length
, tx
));
2807 dmu_assign_arcbuf_impl(db_spill
, abuf
, tx
);
2809 dmu_buf_rele(db
, FTAG
);
2810 dmu_buf_rele(db_spill
, FTAG
);
2818 receive_free(struct receive_writer_arg
*rwa
, struct drr_free
*drrf
)
2822 if (drrf
->drr_length
!= -1ULL &&
2823 drrf
->drr_offset
+ drrf
->drr_length
< drrf
->drr_offset
)
2824 return (SET_ERROR(EINVAL
));
2826 if (dmu_object_info(rwa
->os
, drrf
->drr_object
, NULL
) != 0)
2827 return (SET_ERROR(EINVAL
));
2829 err
= dmu_free_long_range(rwa
->os
, drrf
->drr_object
,
2830 drrf
->drr_offset
, drrf
->drr_length
);
2836 receive_object_range(struct receive_writer_arg
*rwa
,
2837 struct drr_object_range
*drror
)
2841 dnode_t
*mdn
= NULL
;
2842 dmu_buf_t
*db
= NULL
;
2846 * By default, we assume this block is in our native format
2847 * (ZFS_HOST_BYTEORDER). We then take into account whether
2848 * the send stream is byteswapped (rwa->byteswap). Finally,
2849 * we need to byteswap again if this particular block was
2850 * in non-native format on the send side.
2852 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^ rwa
->byteswap
^
2853 !!DRR_IS_RAW_BYTESWAPPED(drror
->drr_flags
);
2856 * Since dnode block sizes are constant, we should not need to worry
2857 * about making sure that the dnode block size is the same on the
2858 * sending and receiving sides for the time being. For non-raw sends,
2859 * this does not matter (and in fact we do not send a DRR_OBJECT_RANGE
2860 * record at all). Raw sends require this record type because the
2861 * encryption parameters are used to protect an entire block of bonus
2862 * buffers. If the size of dnode blocks ever becomes variable,
2863 * handling will need to be added to ensure that dnode block sizes
2864 * match on the sending and receiving side.
2866 if (drror
->drr_numslots
!= DNODES_PER_BLOCK
||
2867 P2PHASE(drror
->drr_firstobj
, DNODES_PER_BLOCK
) != 0 ||
2869 return (SET_ERROR(EINVAL
));
2871 offset
= drror
->drr_firstobj
* sizeof (dnode_phys_t
);
2872 mdn
= DMU_META_DNODE(rwa
->os
);
2874 tx
= dmu_tx_create(rwa
->os
);
2875 ret
= dmu_tx_assign(tx
, TXG_WAIT
);
2881 ret
= dmu_buf_hold_by_dnode(mdn
, offset
, FTAG
, &db
,
2882 DMU_READ_PREFETCH
| DMU_READ_NO_DECRYPT
);
2889 * Convert the buffer associated with this range of dnodes to a
2890 * raw buffer. This ensures that it will be written out as a raw
2891 * buffer when we fill in the dnode objects in future records.
2892 * Since we are commiting this tx now, it is technically possible
2893 * for the dnode block to end up on-disk with the incorrect MAC.
2894 * Despite this, the dataset is marked as inconsistent so no other
2895 * code paths (apart from scrubs) will attempt to read this data.
2896 * Scrubs will not be effected by this either since scrubs only
2897 * read raw data and do not attempt to check the MAC.
2899 dmu_convert_to_raw(db
, byteorder
, drror
->drr_salt
, drror
->drr_iv
,
2900 drror
->drr_mac
, tx
);
2901 dmu_buf_rele(db
, FTAG
);
2906 /* used to destroy the drc_ds on error */
2908 dmu_recv_cleanup_ds(dmu_recv_cookie_t
*drc
)
2910 ds_hold_flags_t dsflags
= (drc
->drc_raw
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
2913 * Wait for the txg sync before cleaning up the receive. For
2914 * resumable receives, this ensures that our resume state has
2915 * been written out to disk. For raw receives, this ensures
2916 * that the user accounting code will not attempt to do anything
2917 * after we stopped receiving the dataset.
2919 txg_wait_synced(drc
->drc_ds
->ds_dir
->dd_pool
, 0);
2921 if (drc
->drc_resumable
) {
2922 dsl_dataset_disown(drc
->drc_ds
, dsflags
, dmu_recv_tag
);
2924 char name
[ZFS_MAX_DATASET_NAME_LEN
];
2925 dsl_dataset_name(drc
->drc_ds
, name
);
2926 dsl_dataset_disown(drc
->drc_ds
, dsflags
, dmu_recv_tag
);
2927 (void) dsl_destroy_head(name
);
2932 receive_cksum(struct receive_arg
*ra
, int len
, void *buf
)
2935 (void) fletcher_4_incremental_byteswap(buf
, len
, &ra
->cksum
);
2937 (void) fletcher_4_incremental_native(buf
, len
, &ra
->cksum
);
2942 * Read the payload into a buffer of size len, and update the current record's
2944 * Allocate ra->next_rrd and read the next record's header into
2945 * ra->next_rrd->header.
2946 * Verify checksum of payload and next record.
2949 receive_read_payload_and_next_header(struct receive_arg
*ra
, int len
, void *buf
)
2952 zio_cksum_t cksum_orig
;
2953 zio_cksum_t
*cksump
;
2956 ASSERT3U(len
, <=, SPA_MAXBLOCKSIZE
);
2957 err
= receive_read(ra
, len
, buf
);
2960 receive_cksum(ra
, len
, buf
);
2962 /* note: rrd is NULL when reading the begin record's payload */
2963 if (ra
->rrd
!= NULL
) {
2964 ra
->rrd
->payload
= buf
;
2965 ra
->rrd
->payload_size
= len
;
2966 ra
->rrd
->bytes_read
= ra
->bytes_read
;
2970 ra
->prev_cksum
= ra
->cksum
;
2972 ra
->next_rrd
= kmem_zalloc(sizeof (*ra
->next_rrd
), KM_SLEEP
);
2973 err
= receive_read(ra
, sizeof (ra
->next_rrd
->header
),
2974 &ra
->next_rrd
->header
);
2975 ra
->next_rrd
->bytes_read
= ra
->bytes_read
;
2978 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
2979 ra
->next_rrd
= NULL
;
2982 if (ra
->next_rrd
->header
.drr_type
== DRR_BEGIN
) {
2983 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
2984 ra
->next_rrd
= NULL
;
2985 return (SET_ERROR(EINVAL
));
2989 * Note: checksum is of everything up to but not including the
2992 ASSERT3U(offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2993 ==, sizeof (dmu_replay_record_t
) - sizeof (zio_cksum_t
));
2995 offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2996 &ra
->next_rrd
->header
);
2998 cksum_orig
= ra
->next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
2999 cksump
= &ra
->next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
3002 byteswap_record(&ra
->next_rrd
->header
);
3004 if ((!ZIO_CHECKSUM_IS_ZERO(cksump
)) &&
3005 !ZIO_CHECKSUM_EQUAL(ra
->cksum
, *cksump
)) {
3006 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
3007 ra
->next_rrd
= NULL
;
3008 return (SET_ERROR(ECKSUM
));
3011 receive_cksum(ra
, sizeof (cksum_orig
), &cksum_orig
);
3017 objlist_create(struct objlist
*list
)
3019 list_create(&list
->list
, sizeof (struct receive_objnode
),
3020 offsetof(struct receive_objnode
, node
));
3021 list
->last_lookup
= 0;
3025 objlist_destroy(struct objlist
*list
)
3027 struct receive_objnode
*n
;
3029 for (n
= list_remove_head(&list
->list
);
3030 n
!= NULL
; n
= list_remove_head(&list
->list
)) {
3031 kmem_free(n
, sizeof (*n
));
3033 list_destroy(&list
->list
);
3037 * This function looks through the objlist to see if the specified object number
3038 * is contained in the objlist. In the process, it will remove all object
3039 * numbers in the list that are smaller than the specified object number. Thus,
3040 * any lookup of an object number smaller than a previously looked up object
3041 * number will always return false; therefore, all lookups should be done in
3045 objlist_exists(struct objlist
*list
, uint64_t object
)
3047 struct receive_objnode
*node
= list_head(&list
->list
);
3048 ASSERT3U(object
, >=, list
->last_lookup
);
3049 list
->last_lookup
= object
;
3050 while (node
!= NULL
&& node
->object
< object
) {
3051 VERIFY3P(node
, ==, list_remove_head(&list
->list
));
3052 kmem_free(node
, sizeof (*node
));
3053 node
= list_head(&list
->list
);
3055 return (node
!= NULL
&& node
->object
== object
);
3059 * The objlist is a list of object numbers stored in ascending order. However,
3060 * the insertion of new object numbers does not seek out the correct location to
3061 * store a new object number; instead, it appends it to the list for simplicity.
3062 * Thus, any users must take care to only insert new object numbers in ascending
3066 objlist_insert(struct objlist
*list
, uint64_t object
)
3068 struct receive_objnode
*node
= kmem_zalloc(sizeof (*node
), KM_SLEEP
);
3069 node
->object
= object
;
3072 struct receive_objnode
*last_object
= list_tail(&list
->list
);
3073 uint64_t last_objnum
= (last_object
!= NULL
? last_object
->object
: 0);
3074 ASSERT3U(node
->object
, >, last_objnum
);
3077 list_insert_tail(&list
->list
, node
);
3081 * Issue the prefetch reads for any necessary indirect blocks.
3083 * We use the object ignore list to tell us whether or not to issue prefetches
3084 * for a given object. We do this for both correctness (in case the blocksize
3085 * of an object has changed) and performance (if the object doesn't exist, don't
3086 * needlessly try to issue prefetches). We also trim the list as we go through
3087 * the stream to prevent it from growing to an unbounded size.
3089 * The object numbers within will always be in sorted order, and any write
3090 * records we see will also be in sorted order, but they're not sorted with
3091 * respect to each other (i.e. we can get several object records before
3092 * receiving each object's write records). As a result, once we've reached a
3093 * given object number, we can safely remove any reference to lower object
3094 * numbers in the ignore list. In practice, we receive up to 32 object records
3095 * before receiving write records, so the list can have up to 32 nodes in it.
3099 receive_read_prefetch(struct receive_arg
*ra
,
3100 uint64_t object
, uint64_t offset
, uint64_t length
)
3102 if (!objlist_exists(&ra
->ignore_objlist
, object
)) {
3103 dmu_prefetch(ra
->os
, object
, 1, offset
, length
,
3104 ZIO_PRIORITY_SYNC_READ
);
3109 * Read records off the stream, issuing any necessary prefetches.
3112 receive_read_record(struct receive_arg
*ra
)
3116 switch (ra
->rrd
->header
.drr_type
) {
3119 struct drr_object
*drro
= &ra
->rrd
->header
.drr_u
.drr_object
;
3120 uint32_t size
= DRR_OBJECT_PAYLOAD_SIZE(drro
);
3121 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
3122 dmu_object_info_t doi
;
3124 err
= receive_read_payload_and_next_header(ra
, size
, buf
);
3126 kmem_free(buf
, size
);
3129 err
= dmu_object_info(ra
->os
, drro
->drr_object
, &doi
);
3131 * See receive_read_prefetch for an explanation why we're
3132 * storing this object in the ignore_obj_list.
3134 if (err
== ENOENT
||
3135 (err
== 0 && doi
.doi_data_block_size
!= drro
->drr_blksz
)) {
3136 objlist_insert(&ra
->ignore_objlist
, drro
->drr_object
);
3141 case DRR_FREEOBJECTS
:
3143 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3148 struct drr_write
*drrw
= &ra
->rrd
->header
.drr_u
.drr_write
;
3150 boolean_t is_meta
= DMU_OT_IS_METADATA(drrw
->drr_type
);
3153 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^
3154 !!DRR_IS_RAW_BYTESWAPPED(drrw
->drr_flags
) ^
3157 abuf
= arc_loan_raw_buf(dmu_objset_spa(ra
->os
),
3158 drrw
->drr_object
, byteorder
, drrw
->drr_salt
,
3159 drrw
->drr_iv
, drrw
->drr_mac
, drrw
->drr_type
,
3160 drrw
->drr_compressed_size
, drrw
->drr_logical_size
,
3161 drrw
->drr_compressiontype
);
3162 } else if (DRR_WRITE_COMPRESSED(drrw
)) {
3163 ASSERT3U(drrw
->drr_compressed_size
, >, 0);
3164 ASSERT3U(drrw
->drr_logical_size
, >=,
3165 drrw
->drr_compressed_size
);
3167 abuf
= arc_loan_compressed_buf(
3168 dmu_objset_spa(ra
->os
),
3169 drrw
->drr_compressed_size
, drrw
->drr_logical_size
,
3170 drrw
->drr_compressiontype
);
3172 abuf
= arc_loan_buf(dmu_objset_spa(ra
->os
),
3173 is_meta
, drrw
->drr_logical_size
);
3176 err
= receive_read_payload_and_next_header(ra
,
3177 DRR_WRITE_PAYLOAD_SIZE(drrw
), abuf
->b_data
);
3179 dmu_return_arcbuf(abuf
);
3182 ra
->rrd
->arc_buf
= abuf
;
3183 receive_read_prefetch(ra
, drrw
->drr_object
, drrw
->drr_offset
,
3184 drrw
->drr_logical_size
);
3187 case DRR_WRITE_BYREF
:
3189 struct drr_write_byref
*drrwb
=
3190 &ra
->rrd
->header
.drr_u
.drr_write_byref
;
3191 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3192 receive_read_prefetch(ra
, drrwb
->drr_object
, drrwb
->drr_offset
,
3196 case DRR_WRITE_EMBEDDED
:
3198 struct drr_write_embedded
*drrwe
=
3199 &ra
->rrd
->header
.drr_u
.drr_write_embedded
;
3200 uint32_t size
= P2ROUNDUP(drrwe
->drr_psize
, 8);
3201 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
3203 err
= receive_read_payload_and_next_header(ra
, size
, buf
);
3205 kmem_free(buf
, size
);
3209 receive_read_prefetch(ra
, drrwe
->drr_object
, drrwe
->drr_offset
,
3216 * It might be beneficial to prefetch indirect blocks here, but
3217 * we don't really have the data to decide for sure.
3219 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3224 struct drr_end
*drre
= &ra
->rrd
->header
.drr_u
.drr_end
;
3225 if (!ZIO_CHECKSUM_EQUAL(ra
->prev_cksum
, drre
->drr_checksum
))
3226 return (SET_ERROR(ECKSUM
));
3231 struct drr_spill
*drrs
= &ra
->rrd
->header
.drr_u
.drr_spill
;
3233 int len
= DRR_SPILL_PAYLOAD_SIZE(drrs
);
3235 /* DRR_SPILL records are either raw or uncompressed */
3237 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^
3238 !!DRR_IS_RAW_BYTESWAPPED(drrs
->drr_flags
) ^
3241 abuf
= arc_loan_raw_buf(dmu_objset_spa(ra
->os
),
3242 drrs
->drr_object
, byteorder
, drrs
->drr_salt
,
3243 drrs
->drr_iv
, drrs
->drr_mac
, drrs
->drr_type
,
3244 drrs
->drr_compressed_size
, drrs
->drr_length
,
3245 drrs
->drr_compressiontype
);
3247 abuf
= arc_loan_buf(dmu_objset_spa(ra
->os
),
3248 DMU_OT_IS_METADATA(drrs
->drr_type
),
3252 err
= receive_read_payload_and_next_header(ra
, len
,
3255 dmu_return_arcbuf(abuf
);
3258 ra
->rrd
->arc_buf
= abuf
;
3261 case DRR_OBJECT_RANGE
:
3263 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3267 return (SET_ERROR(EINVAL
));
3272 dprintf_drr(struct receive_record_arg
*rrd
, int err
)
3274 switch (rrd
->header
.drr_type
) {
3277 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
3278 dprintf("drr_type = OBJECT obj = %llu type = %u "
3279 "bonustype = %u blksz = %u bonuslen = %u cksumtype = %u "
3280 "compress = %u dn_slots = %u err = %d\n",
3281 drro
->drr_object
, drro
->drr_type
, drro
->drr_bonustype
,
3282 drro
->drr_blksz
, drro
->drr_bonuslen
,
3283 drro
->drr_checksumtype
, drro
->drr_compress
,
3284 drro
->drr_dn_slots
, err
);
3287 case DRR_FREEOBJECTS
:
3289 struct drr_freeobjects
*drrfo
=
3290 &rrd
->header
.drr_u
.drr_freeobjects
;
3291 dprintf("drr_type = FREEOBJECTS firstobj = %llu "
3292 "numobjs = %llu err = %d\n",
3293 drrfo
->drr_firstobj
, drrfo
->drr_numobjs
, err
);
3298 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
3299 dprintf("drr_type = WRITE obj = %llu type = %u offset = %llu "
3300 "lsize = %llu cksumtype = %u cksumflags = %u "
3301 "compress = %u psize = %llu err = %d\n",
3302 drrw
->drr_object
, drrw
->drr_type
, drrw
->drr_offset
,
3303 drrw
->drr_logical_size
, drrw
->drr_checksumtype
,
3304 drrw
->drr_flags
, drrw
->drr_compressiontype
,
3305 drrw
->drr_compressed_size
, err
);
3308 case DRR_WRITE_BYREF
:
3310 struct drr_write_byref
*drrwbr
=
3311 &rrd
->header
.drr_u
.drr_write_byref
;
3312 dprintf("drr_type = WRITE_BYREF obj = %llu offset = %llu "
3313 "length = %llu toguid = %llx refguid = %llx "
3314 "refobject = %llu refoffset = %llu cksumtype = %u "
3315 "cksumflags = %u err = %d\n",
3316 drrwbr
->drr_object
, drrwbr
->drr_offset
,
3317 drrwbr
->drr_length
, drrwbr
->drr_toguid
,
3318 drrwbr
->drr_refguid
, drrwbr
->drr_refobject
,
3319 drrwbr
->drr_refoffset
, drrwbr
->drr_checksumtype
,
3320 drrwbr
->drr_flags
, err
);
3323 case DRR_WRITE_EMBEDDED
:
3325 struct drr_write_embedded
*drrwe
=
3326 &rrd
->header
.drr_u
.drr_write_embedded
;
3327 dprintf("drr_type = WRITE_EMBEDDED obj = %llu offset = %llu "
3328 "length = %llu compress = %u etype = %u lsize = %u "
3329 "psize = %u err = %d\n",
3330 drrwe
->drr_object
, drrwe
->drr_offset
, drrwe
->drr_length
,
3331 drrwe
->drr_compression
, drrwe
->drr_etype
,
3332 drrwe
->drr_lsize
, drrwe
->drr_psize
, err
);
3337 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
3338 dprintf("drr_type = FREE obj = %llu offset = %llu "
3339 "length = %lld err = %d\n",
3340 drrf
->drr_object
, drrf
->drr_offset
, drrf
->drr_length
,
3346 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
3347 dprintf("drr_type = SPILL obj = %llu length = %llu "
3348 "err = %d\n", drrs
->drr_object
, drrs
->drr_length
, err
);
3357 * Commit the records to the pool.
3360 receive_process_record(struct receive_writer_arg
*rwa
,
3361 struct receive_record_arg
*rrd
)
3365 /* Processing in order, therefore bytes_read should be increasing. */
3366 ASSERT3U(rrd
->bytes_read
, >=, rwa
->bytes_read
);
3367 rwa
->bytes_read
= rrd
->bytes_read
;
3369 switch (rrd
->header
.drr_type
) {
3372 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
3373 err
= receive_object(rwa
, drro
, rrd
->payload
);
3374 kmem_free(rrd
->payload
, rrd
->payload_size
);
3375 rrd
->payload
= NULL
;
3378 case DRR_FREEOBJECTS
:
3380 struct drr_freeobjects
*drrfo
=
3381 &rrd
->header
.drr_u
.drr_freeobjects
;
3382 err
= receive_freeobjects(rwa
, drrfo
);
3387 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
3388 err
= receive_write(rwa
, drrw
, rrd
->arc_buf
);
3389 /* if receive_write() is successful, it consumes the arc_buf */
3391 dmu_return_arcbuf(rrd
->arc_buf
);
3392 rrd
->arc_buf
= NULL
;
3393 rrd
->payload
= NULL
;
3396 case DRR_WRITE_BYREF
:
3398 struct drr_write_byref
*drrwbr
=
3399 &rrd
->header
.drr_u
.drr_write_byref
;
3400 err
= receive_write_byref(rwa
, drrwbr
);
3403 case DRR_WRITE_EMBEDDED
:
3405 struct drr_write_embedded
*drrwe
=
3406 &rrd
->header
.drr_u
.drr_write_embedded
;
3407 err
= receive_write_embedded(rwa
, drrwe
, rrd
->payload
);
3408 kmem_free(rrd
->payload
, rrd
->payload_size
);
3409 rrd
->payload
= NULL
;
3414 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
3415 err
= receive_free(rwa
, drrf
);
3420 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
3421 err
= receive_spill(rwa
, drrs
, rrd
->arc_buf
);
3422 /* if receive_spill() is successful, it consumes the arc_buf */
3424 dmu_return_arcbuf(rrd
->arc_buf
);
3425 rrd
->arc_buf
= NULL
;
3426 rrd
->payload
= NULL
;
3429 case DRR_OBJECT_RANGE
:
3431 struct drr_object_range
*drror
=
3432 &rrd
->header
.drr_u
.drr_object_range
;
3433 return (receive_object_range(rwa
, drror
));
3436 return (SET_ERROR(EINVAL
));
3440 dprintf_drr(rrd
, err
);
3446 * dmu_recv_stream's worker thread; pull records off the queue, and then call
3447 * receive_process_record When we're done, signal the main thread and exit.
3450 receive_writer_thread(void *arg
)
3452 struct receive_writer_arg
*rwa
= arg
;
3453 struct receive_record_arg
*rrd
;
3454 fstrans_cookie_t cookie
= spl_fstrans_mark();
3456 for (rrd
= bqueue_dequeue(&rwa
->q
); !rrd
->eos_marker
;
3457 rrd
= bqueue_dequeue(&rwa
->q
)) {
3459 * If there's an error, the main thread will stop putting things
3460 * on the queue, but we need to clear everything in it before we
3463 if (rwa
->err
== 0) {
3464 rwa
->err
= receive_process_record(rwa
, rrd
);
3465 } else if (rrd
->arc_buf
!= NULL
) {
3466 dmu_return_arcbuf(rrd
->arc_buf
);
3467 rrd
->arc_buf
= NULL
;
3468 rrd
->payload
= NULL
;
3469 } else if (rrd
->payload
!= NULL
) {
3470 kmem_free(rrd
->payload
, rrd
->payload_size
);
3471 rrd
->payload
= NULL
;
3473 kmem_free(rrd
, sizeof (*rrd
));
3475 kmem_free(rrd
, sizeof (*rrd
));
3476 mutex_enter(&rwa
->mutex
);
3478 cv_signal(&rwa
->cv
);
3479 mutex_exit(&rwa
->mutex
);
3480 spl_fstrans_unmark(cookie
);
3485 resume_check(struct receive_arg
*ra
, nvlist_t
*begin_nvl
)
3488 objset_t
*mos
= dmu_objset_pool(ra
->os
)->dp_meta_objset
;
3489 uint64_t dsobj
= dmu_objset_id(ra
->os
);
3490 uint64_t resume_obj
, resume_off
;
3492 if (nvlist_lookup_uint64(begin_nvl
,
3493 "resume_object", &resume_obj
) != 0 ||
3494 nvlist_lookup_uint64(begin_nvl
,
3495 "resume_offset", &resume_off
) != 0) {
3496 return (SET_ERROR(EINVAL
));
3498 VERIFY0(zap_lookup(mos
, dsobj
,
3499 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
));
3500 if (resume_obj
!= val
)
3501 return (SET_ERROR(EINVAL
));
3502 VERIFY0(zap_lookup(mos
, dsobj
,
3503 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
));
3504 if (resume_off
!= val
)
3505 return (SET_ERROR(EINVAL
));
3511 * Read in the stream's records, one by one, and apply them to the pool. There
3512 * are two threads involved; the thread that calls this function will spin up a
3513 * worker thread, read the records off the stream one by one, and issue
3514 * prefetches for any necessary indirect blocks. It will then push the records
3515 * onto an internal blocking queue. The worker thread will pull the records off
3516 * the queue, and actually write the data into the DMU. This way, the worker
3517 * thread doesn't have to wait for reads to complete, since everything it needs
3518 * (the indirect blocks) will be prefetched.
3520 * NB: callers *must* call dmu_recv_end() if this succeeds.
3523 dmu_recv_stream(dmu_recv_cookie_t
*drc
, vnode_t
*vp
, offset_t
*voffp
,
3524 int cleanup_fd
, uint64_t *action_handlep
)
3527 struct receive_arg
*ra
;
3528 struct receive_writer_arg
*rwa
;
3530 uint32_t payloadlen
;
3532 nvlist_t
*begin_nvl
= NULL
;
3534 ra
= kmem_zalloc(sizeof (*ra
), KM_SLEEP
);
3535 rwa
= kmem_zalloc(sizeof (*rwa
), KM_SLEEP
);
3537 ra
->byteswap
= drc
->drc_byteswap
;
3538 ra
->raw
= drc
->drc_raw
;
3539 ra
->cksum
= drc
->drc_cksum
;
3543 if (dsl_dataset_is_zapified(drc
->drc_ds
)) {
3544 (void) zap_lookup(drc
->drc_ds
->ds_dir
->dd_pool
->dp_meta_objset
,
3545 drc
->drc_ds
->ds_object
, DS_FIELD_RESUME_BYTES
,
3546 sizeof (ra
->bytes_read
), 1, &ra
->bytes_read
);
3549 objlist_create(&ra
->ignore_objlist
);
3551 /* these were verified in dmu_recv_begin */
3552 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc
->drc_drrb
->drr_versioninfo
), ==,
3554 ASSERT3U(drc
->drc_drrb
->drr_type
, <, DMU_OST_NUMTYPES
);
3557 * Open the objset we are modifying.
3559 VERIFY0(dmu_objset_from_ds(drc
->drc_ds
, &ra
->os
));
3561 ASSERT(dsl_dataset_phys(drc
->drc_ds
)->ds_flags
& DS_FLAG_INCONSISTENT
);
3563 featureflags
= DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
);
3564 ra
->featureflags
= featureflags
;
3566 /* embedded data is incompatible with encrypted datasets */
3567 if (ra
->os
->os_encrypted
&&
3568 (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)) {
3569 err
= SET_ERROR(EINVAL
);
3573 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
3574 if (featureflags
& DMU_BACKUP_FEATURE_DEDUP
) {
3577 if (cleanup_fd
== -1) {
3578 err
= SET_ERROR(EBADF
);
3581 err
= zfs_onexit_fd_hold(cleanup_fd
, &minor
);
3587 if (*action_handlep
== 0) {
3588 rwa
->guid_to_ds_map
=
3589 kmem_alloc(sizeof (avl_tree_t
), KM_SLEEP
);
3590 avl_create(rwa
->guid_to_ds_map
, guid_compare
,
3591 sizeof (guid_map_entry_t
),
3592 offsetof(guid_map_entry_t
, avlnode
));
3593 err
= zfs_onexit_add_cb(minor
,
3594 free_guid_map_onexit
, rwa
->guid_to_ds_map
,
3599 err
= zfs_onexit_cb_data(minor
, *action_handlep
,
3600 (void **)&rwa
->guid_to_ds_map
);
3605 drc
->drc_guid_to_ds_map
= rwa
->guid_to_ds_map
;
3608 payloadlen
= drc
->drc_drr_begin
->drr_payloadlen
;
3610 if (payloadlen
!= 0)
3611 payload
= kmem_alloc(payloadlen
, KM_SLEEP
);
3613 err
= receive_read_payload_and_next_header(ra
, payloadlen
, payload
);
3615 if (payloadlen
!= 0)
3616 kmem_free(payload
, payloadlen
);
3619 if (payloadlen
!= 0) {
3620 err
= nvlist_unpack(payload
, payloadlen
, &begin_nvl
, KM_SLEEP
);
3621 kmem_free(payload
, payloadlen
);
3626 /* handle DSL encryption key payload */
3627 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
3628 nvlist_t
*keynvl
= NULL
;
3630 ASSERT(ra
->os
->os_encrypted
);
3631 ASSERT(drc
->drc_raw
);
3633 err
= nvlist_lookup_nvlist(begin_nvl
, "crypt_keydata", &keynvl
);
3637 err
= dsl_crypto_recv_key(spa_name(ra
->os
->os_spa
),
3638 drc
->drc_ds
->ds_object
, drc
->drc_drrb
->drr_type
,
3644 if (featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
3645 err
= resume_check(ra
, begin_nvl
);
3650 (void) bqueue_init(&rwa
->q
, zfs_recv_queue_length
,
3651 offsetof(struct receive_record_arg
, node
));
3652 cv_init(&rwa
->cv
, NULL
, CV_DEFAULT
, NULL
);
3653 mutex_init(&rwa
->mutex
, NULL
, MUTEX_DEFAULT
, NULL
);
3655 rwa
->byteswap
= drc
->drc_byteswap
;
3656 rwa
->resumable
= drc
->drc_resumable
;
3657 rwa
->raw
= drc
->drc_raw
;
3659 (void) thread_create(NULL
, 0, receive_writer_thread
, rwa
, 0, curproc
,
3660 TS_RUN
, minclsyspri
);
3662 * We're reading rwa->err without locks, which is safe since we are the
3663 * only reader, and the worker thread is the only writer. It's ok if we
3664 * miss a write for an iteration or two of the loop, since the writer
3665 * thread will keep freeing records we send it until we send it an eos
3668 * We can leave this loop in 3 ways: First, if rwa->err is
3669 * non-zero. In that case, the writer thread will free the rrd we just
3670 * pushed. Second, if we're interrupted; in that case, either it's the
3671 * first loop and ra->rrd was never allocated, or it's later and ra->rrd
3672 * has been handed off to the writer thread who will free it. Finally,
3673 * if receive_read_record fails or we're at the end of the stream, then
3674 * we free ra->rrd and exit.
3676 while (rwa
->err
== 0) {
3677 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
3678 err
= SET_ERROR(EINTR
);
3682 ASSERT3P(ra
->rrd
, ==, NULL
);
3683 ra
->rrd
= ra
->next_rrd
;
3684 ra
->next_rrd
= NULL
;
3685 /* Allocates and loads header into ra->next_rrd */
3686 err
= receive_read_record(ra
);
3688 if (ra
->rrd
->header
.drr_type
== DRR_END
|| err
!= 0) {
3689 kmem_free(ra
->rrd
, sizeof (*ra
->rrd
));
3694 bqueue_enqueue(&rwa
->q
, ra
->rrd
,
3695 sizeof (struct receive_record_arg
) + ra
->rrd
->payload_size
);
3698 if (ra
->next_rrd
== NULL
)
3699 ra
->next_rrd
= kmem_zalloc(sizeof (*ra
->next_rrd
), KM_SLEEP
);
3700 ra
->next_rrd
->eos_marker
= B_TRUE
;
3701 bqueue_enqueue(&rwa
->q
, ra
->next_rrd
, 1);
3703 mutex_enter(&rwa
->mutex
);
3704 while (!rwa
->done
) {
3705 cv_wait(&rwa
->cv
, &rwa
->mutex
);
3707 mutex_exit(&rwa
->mutex
);
3709 cv_destroy(&rwa
->cv
);
3710 mutex_destroy(&rwa
->mutex
);
3711 bqueue_destroy(&rwa
->q
);
3716 nvlist_free(begin_nvl
);
3717 if ((featureflags
& DMU_BACKUP_FEATURE_DEDUP
) && (cleanup_fd
!= -1))
3718 zfs_onexit_fd_rele(cleanup_fd
);
3722 * Clean up references. If receive is not resumable,
3723 * destroy what we created, so we don't leave it in
3724 * the inconsistent state.
3726 dmu_recv_cleanup_ds(drc
);
3730 objlist_destroy(&ra
->ignore_objlist
);
3731 kmem_free(ra
, sizeof (*ra
));
3732 kmem_free(rwa
, sizeof (*rwa
));
3737 dmu_recv_end_check(void *arg
, dmu_tx_t
*tx
)
3739 dmu_recv_cookie_t
*drc
= arg
;
3740 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3743 ASSERT3P(drc
->drc_ds
->ds_owner
, ==, dmu_recv_tag
);
3745 if (!drc
->drc_newfs
) {
3746 dsl_dataset_t
*origin_head
;
3748 error
= dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
, &origin_head
);
3751 if (drc
->drc_force
) {
3753 * We will destroy any snapshots in tofs (i.e. before
3754 * origin_head) that are after the origin (which is
3755 * the snap before drc_ds, because drc_ds can not
3756 * have any snaps of its own).
3760 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3762 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3763 dsl_dataset_t
*snap
;
3764 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3768 if (snap
->ds_dir
!= origin_head
->ds_dir
)
3769 error
= SET_ERROR(EINVAL
);
3771 error
= dsl_destroy_snapshot_check_impl(
3774 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3775 dsl_dataset_rele(snap
, FTAG
);
3780 dsl_dataset_rele(origin_head
, FTAG
);
3784 error
= dsl_dataset_clone_swap_check_impl(drc
->drc_ds
,
3785 origin_head
, drc
->drc_force
, drc
->drc_owner
, tx
);
3787 dsl_dataset_rele(origin_head
, FTAG
);
3790 error
= dsl_dataset_snapshot_check_impl(origin_head
,
3791 drc
->drc_tosnap
, tx
, B_TRUE
, 1, drc
->drc_cred
);
3792 dsl_dataset_rele(origin_head
, FTAG
);
3796 error
= dsl_destroy_head_check_impl(drc
->drc_ds
, 1);
3798 error
= dsl_dataset_snapshot_check_impl(drc
->drc_ds
,
3799 drc
->drc_tosnap
, tx
, B_TRUE
, 1, drc
->drc_cred
);
3805 dmu_recv_end_sync(void *arg
, dmu_tx_t
*tx
)
3807 dmu_recv_cookie_t
*drc
= arg
;
3808 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3809 boolean_t encrypted
= drc
->drc_ds
->ds_dir
->dd_crypto_obj
!= 0;
3811 spa_history_log_internal_ds(drc
->drc_ds
, "finish receiving",
3812 tx
, "snap=%s", drc
->drc_tosnap
);
3814 if (!drc
->drc_newfs
) {
3815 dsl_dataset_t
*origin_head
;
3817 VERIFY0(dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
,
3820 if (drc
->drc_force
) {
3822 * Destroy any snapshots of drc_tofs (origin_head)
3823 * after the origin (the snap before drc_ds).
3827 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3829 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3830 dsl_dataset_t
*snap
;
3831 VERIFY0(dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3833 ASSERT3P(snap
->ds_dir
, ==, origin_head
->ds_dir
);
3834 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3835 dsl_destroy_snapshot_sync_impl(snap
,
3837 dsl_dataset_rele(snap
, FTAG
);
3840 VERIFY3P(drc
->drc_ds
->ds_prev
, ==,
3841 origin_head
->ds_prev
);
3843 dsl_dataset_clone_swap_sync_impl(drc
->drc_ds
,
3845 dsl_dataset_snapshot_sync_impl(origin_head
,
3846 drc
->drc_tosnap
, tx
);
3848 /* set snapshot's creation time and guid */
3849 dmu_buf_will_dirty(origin_head
->ds_prev
->ds_dbuf
, tx
);
3850 dsl_dataset_phys(origin_head
->ds_prev
)->ds_creation_time
=
3851 drc
->drc_drrb
->drr_creation_time
;
3852 dsl_dataset_phys(origin_head
->ds_prev
)->ds_guid
=
3853 drc
->drc_drrb
->drr_toguid
;
3854 dsl_dataset_phys(origin_head
->ds_prev
)->ds_flags
&=
3855 ~DS_FLAG_INCONSISTENT
;
3857 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
3858 dsl_dataset_phys(origin_head
)->ds_flags
&=
3859 ~DS_FLAG_INCONSISTENT
;
3861 drc
->drc_newsnapobj
=
3862 dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3864 dsl_dataset_rele(origin_head
, FTAG
);
3865 dsl_destroy_head_sync_impl(drc
->drc_ds
, tx
);
3867 if (drc
->drc_owner
!= NULL
)
3868 VERIFY3P(origin_head
->ds_owner
, ==, drc
->drc_owner
);
3870 dsl_dataset_t
*ds
= drc
->drc_ds
;
3872 dsl_dataset_snapshot_sync_impl(ds
, drc
->drc_tosnap
, tx
);
3874 /* set snapshot's creation time and guid */
3875 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
3876 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_time
=
3877 drc
->drc_drrb
->drr_creation_time
;
3878 dsl_dataset_phys(ds
->ds_prev
)->ds_guid
=
3879 drc
->drc_drrb
->drr_toguid
;
3880 dsl_dataset_phys(ds
->ds_prev
)->ds_flags
&=
3881 ~DS_FLAG_INCONSISTENT
;
3883 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3884 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
3885 if (dsl_dataset_has_resume_receive_state(ds
)) {
3886 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3887 DS_FIELD_RESUME_FROMGUID
, tx
);
3888 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3889 DS_FIELD_RESUME_OBJECT
, tx
);
3890 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3891 DS_FIELD_RESUME_OFFSET
, tx
);
3892 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3893 DS_FIELD_RESUME_BYTES
, tx
);
3894 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3895 DS_FIELD_RESUME_TOGUID
, tx
);
3896 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3897 DS_FIELD_RESUME_TONAME
, tx
);
3899 drc
->drc_newsnapobj
=
3900 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
;
3902 zvol_create_minors(dp
->dp_spa
, drc
->drc_tofs
, B_TRUE
);
3905 * Release the hold from dmu_recv_begin. This must be done before
3906 * we return to open context, so that when we free the dataset's dnode
3907 * we can evict its bonus buffer. Since the dataset may be destroyed
3908 * at this point (and therefore won't have a valid pointer to the spa)
3909 * we release the key mapping manually here while we do have a valid
3910 * pointer, if it exists.
3912 if (!drc
->drc_raw
&& encrypted
) {
3913 (void) spa_keystore_remove_mapping(dmu_tx_pool(tx
)->dp_spa
,
3914 drc
->drc_ds
->ds_object
, drc
->drc_ds
);
3916 dsl_dataset_disown(drc
->drc_ds
, 0, dmu_recv_tag
);
3921 add_ds_to_guidmap(const char *name
, avl_tree_t
*guid_map
, uint64_t snapobj
,
3925 dsl_dataset_t
*snapds
;
3926 guid_map_entry_t
*gmep
;
3927 ds_hold_flags_t dsflags
= (raw
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
3930 ASSERT(guid_map
!= NULL
);
3932 err
= dsl_pool_hold(name
, FTAG
, &dp
);
3935 gmep
= kmem_alloc(sizeof (*gmep
), KM_SLEEP
);
3936 err
= dsl_dataset_hold_obj_flags(dp
, snapobj
, dsflags
, gmep
, &snapds
);
3938 gmep
->guid
= dsl_dataset_phys(snapds
)->ds_guid
;
3940 gmep
->gme_ds
= snapds
;
3941 avl_add(guid_map
, gmep
);
3942 dsl_dataset_long_hold(snapds
, gmep
);
3944 kmem_free(gmep
, sizeof (*gmep
));
3947 dsl_pool_rele(dp
, FTAG
);
3951 static int dmu_recv_end_modified_blocks
= 3;
3954 dmu_recv_existing_end(dmu_recv_cookie_t
*drc
)
3958 * We will be destroying the ds; make sure its origin is unmounted if
3961 char name
[ZFS_MAX_DATASET_NAME_LEN
];
3962 dsl_dataset_name(drc
->drc_ds
, name
);
3963 zfs_destroy_unmount_origin(name
);
3966 return (dsl_sync_task(drc
->drc_tofs
,
3967 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3968 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
3972 dmu_recv_new_end(dmu_recv_cookie_t
*drc
)
3974 return (dsl_sync_task(drc
->drc_tofs
,
3975 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3976 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
3980 dmu_recv_end(dmu_recv_cookie_t
*drc
, void *owner
)
3984 drc
->drc_owner
= owner
;
3987 error
= dmu_recv_new_end(drc
);
3989 error
= dmu_recv_existing_end(drc
);
3992 dmu_recv_cleanup_ds(drc
);
3993 } else if (drc
->drc_guid_to_ds_map
!= NULL
) {
3994 (void) add_ds_to_guidmap(drc
->drc_tofs
, drc
->drc_guid_to_ds_map
,
3995 drc
->drc_newsnapobj
, drc
->drc_raw
);
4001 * Return TRUE if this objset is currently being received into.
4004 dmu_objset_is_receiving(objset_t
*os
)
4006 return (os
->os_dsl_dataset
!= NULL
&&
4007 os
->os_dsl_dataset
->ds_owner
== dmu_recv_tag
);
4010 #if defined(_KERNEL)
4011 module_param(zfs_send_corrupt_data
, int, 0644);
4012 MODULE_PARM_DESC(zfs_send_corrupt_data
, "Allow sending corrupt data");