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 set the encrypted
331 * flag. We need to pass along everything the receiving
332 * side will need to interpret this block, including the
333 * byteswap, salt, IV, and MAC.
335 drrw
->drr_flags
|= DRR_RAW_ENCRYPTED
;
336 if (BP_SHOULD_BYTESWAP(bp
))
337 drrw
->drr_flags
|= DRR_RAW_BYTESWAP
;
338 zio_crypt_decode_params_bp(bp
, drrw
->drr_salt
,
340 zio_crypt_decode_mac_bp(bp
, drrw
->drr_mac
);
342 /* this is a compressed block */
343 ASSERT(dsp
->dsa_featureflags
&
344 DMU_BACKUP_FEATURE_COMPRESSED
);
345 ASSERT(!BP_SHOULD_BYTESWAP(bp
));
346 ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)));
347 ASSERT3U(BP_GET_COMPRESS(bp
), !=, ZIO_COMPRESS_OFF
);
348 ASSERT3S(lsize
, >=, psize
);
351 /* set fields common to compressed and raw sends */
352 drrw
->drr_compressiontype
= BP_GET_COMPRESS(bp
);
353 drrw
->drr_compressed_size
= psize
;
354 payload_size
= drrw
->drr_compressed_size
;
356 payload_size
= drrw
->drr_logical_size
;
359 if (bp
== NULL
|| BP_IS_EMBEDDED(bp
) || (BP_IS_PROTECTED(bp
) && !raw
)) {
361 * There's no pre-computed checksum for partial-block writes,
362 * embedded BP's, or encrypted BP's that are being sent as
363 * plaintext, so (like fletcher4-checkummed blocks) userland
364 * will have to compute a dedup-capable checksum itself.
366 drrw
->drr_checksumtype
= ZIO_CHECKSUM_OFF
;
368 drrw
->drr_checksumtype
= BP_GET_CHECKSUM(bp
);
369 if (zio_checksum_table
[drrw
->drr_checksumtype
].ci_flags
&
370 ZCHECKSUM_FLAG_DEDUP
)
371 drrw
->drr_flags
|= DRR_CHECKSUM_DEDUP
;
372 DDK_SET_LSIZE(&drrw
->drr_key
, BP_GET_LSIZE(bp
));
373 DDK_SET_PSIZE(&drrw
->drr_key
, BP_GET_PSIZE(bp
));
374 DDK_SET_COMPRESS(&drrw
->drr_key
, BP_GET_COMPRESS(bp
));
375 DDK_SET_CRYPT(&drrw
->drr_key
, BP_IS_PROTECTED(bp
));
376 drrw
->drr_key
.ddk_cksum
= bp
->blk_cksum
;
379 if (dump_record(dsp
, data
, payload_size
) != 0)
380 return (SET_ERROR(EINTR
));
385 dump_write_embedded(dmu_sendarg_t
*dsp
, uint64_t object
, uint64_t offset
,
386 int blksz
, const blkptr_t
*bp
)
388 char buf
[BPE_PAYLOAD_SIZE
];
389 struct drr_write_embedded
*drrw
=
390 &(dsp
->dsa_drr
->drr_u
.drr_write_embedded
);
392 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
393 if (dump_record(dsp
, NULL
, 0) != 0)
394 return (SET_ERROR(EINTR
));
395 dsp
->dsa_pending_op
= PENDING_NONE
;
398 ASSERT(BP_IS_EMBEDDED(bp
));
400 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
401 dsp
->dsa_drr
->drr_type
= DRR_WRITE_EMBEDDED
;
402 drrw
->drr_object
= object
;
403 drrw
->drr_offset
= offset
;
404 drrw
->drr_length
= blksz
;
405 drrw
->drr_toguid
= dsp
->dsa_toguid
;
406 drrw
->drr_compression
= BP_GET_COMPRESS(bp
);
407 drrw
->drr_etype
= BPE_GET_ETYPE(bp
);
408 drrw
->drr_lsize
= BPE_GET_LSIZE(bp
);
409 drrw
->drr_psize
= BPE_GET_PSIZE(bp
);
411 decode_embedded_bp_compressed(bp
, buf
);
413 if (dump_record(dsp
, buf
, P2ROUNDUP(drrw
->drr_psize
, 8)) != 0)
414 return (SET_ERROR(EINTR
));
419 dump_spill(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
, uint64_t object
, void *data
)
421 struct drr_spill
*drrs
= &(dsp
->dsa_drr
->drr_u
.drr_spill
);
422 uint64_t blksz
= BP_GET_LSIZE(bp
);
424 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
425 if (dump_record(dsp
, NULL
, 0) != 0)
426 return (SET_ERROR(EINTR
));
427 dsp
->dsa_pending_op
= PENDING_NONE
;
430 /* write a SPILL record */
431 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
432 dsp
->dsa_drr
->drr_type
= DRR_SPILL
;
433 drrs
->drr_object
= object
;
434 drrs
->drr_length
= blksz
;
435 drrs
->drr_toguid
= dsp
->dsa_toguid
;
437 /* handle raw send fields */
438 if ((dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0 &&
439 BP_IS_PROTECTED(bp
)) {
440 drrs
->drr_flags
|= DRR_RAW_ENCRYPTED
;
441 if (BP_SHOULD_BYTESWAP(bp
))
442 drrs
->drr_flags
|= DRR_RAW_BYTESWAP
;
443 drrs
->drr_compressiontype
= BP_GET_COMPRESS(bp
);
444 drrs
->drr_compressed_size
= BP_GET_PSIZE(bp
);
445 zio_crypt_decode_params_bp(bp
, drrs
->drr_salt
, drrs
->drr_iv
);
446 zio_crypt_decode_mac_bp(bp
, drrs
->drr_mac
);
449 if (dump_record(dsp
, data
, blksz
) != 0)
450 return (SET_ERROR(EINTR
));
455 dump_freeobjects(dmu_sendarg_t
*dsp
, uint64_t firstobj
, uint64_t numobjs
)
457 struct drr_freeobjects
*drrfo
= &(dsp
->dsa_drr
->drr_u
.drr_freeobjects
);
460 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
461 * push it out, since free block aggregation can only be done for
462 * blocks of the same type (i.e., DRR_FREE records can only be
463 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
464 * can only be aggregated with other DRR_FREEOBJECTS records.
466 if (dsp
->dsa_pending_op
!= PENDING_NONE
&&
467 dsp
->dsa_pending_op
!= PENDING_FREEOBJECTS
) {
468 if (dump_record(dsp
, NULL
, 0) != 0)
469 return (SET_ERROR(EINTR
));
470 dsp
->dsa_pending_op
= PENDING_NONE
;
472 if (dsp
->dsa_pending_op
== PENDING_FREEOBJECTS
) {
474 * See whether this free object array can be aggregated
477 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
== firstobj
) {
478 drrfo
->drr_numobjs
+= numobjs
;
481 /* can't be aggregated. Push out pending record */
482 if (dump_record(dsp
, NULL
, 0) != 0)
483 return (SET_ERROR(EINTR
));
484 dsp
->dsa_pending_op
= PENDING_NONE
;
488 /* write a FREEOBJECTS record */
489 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
490 dsp
->dsa_drr
->drr_type
= DRR_FREEOBJECTS
;
491 drrfo
->drr_firstobj
= firstobj
;
492 drrfo
->drr_numobjs
= numobjs
;
493 drrfo
->drr_toguid
= dsp
->dsa_toguid
;
495 dsp
->dsa_pending_op
= PENDING_FREEOBJECTS
;
501 dump_dnode(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
, uint64_t object
,
504 struct drr_object
*drro
= &(dsp
->dsa_drr
->drr_u
.drr_object
);
505 int bonuslen
= P2ROUNDUP(dnp
->dn_bonuslen
, 8);
507 if (object
< dsp
->dsa_resume_object
) {
509 * Note: when resuming, we will visit all the dnodes in
510 * the block of dnodes that we are resuming from. In
511 * this case it's unnecessary to send the dnodes prior to
512 * the one we are resuming from. We should be at most one
513 * block's worth of dnodes behind the resume point.
515 ASSERT3U(dsp
->dsa_resume_object
- object
, <,
516 1 << (DNODE_BLOCK_SHIFT
- DNODE_SHIFT
));
520 if (dnp
== NULL
|| dnp
->dn_type
== DMU_OT_NONE
)
521 return (dump_freeobjects(dsp
, object
, 1));
523 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
524 if (dump_record(dsp
, NULL
, 0) != 0)
525 return (SET_ERROR(EINTR
));
526 dsp
->dsa_pending_op
= PENDING_NONE
;
529 /* write an OBJECT record */
530 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
531 dsp
->dsa_drr
->drr_type
= DRR_OBJECT
;
532 drro
->drr_object
= object
;
533 drro
->drr_type
= dnp
->dn_type
;
534 drro
->drr_bonustype
= dnp
->dn_bonustype
;
535 drro
->drr_blksz
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
536 drro
->drr_bonuslen
= dnp
->dn_bonuslen
;
537 drro
->drr_dn_slots
= dnp
->dn_extra_slots
+ 1;
538 drro
->drr_checksumtype
= dnp
->dn_checksum
;
539 drro
->drr_compress
= dnp
->dn_compress
;
540 drro
->drr_toguid
= dsp
->dsa_toguid
;
542 if (!(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
543 drro
->drr_blksz
> SPA_OLD_MAXBLOCKSIZE
)
544 drro
->drr_blksz
= SPA_OLD_MAXBLOCKSIZE
;
546 if ((dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) &&
547 BP_IS_PROTECTED(bp
)) {
548 drro
->drr_flags
|= DRR_RAW_ENCRYPTED
;
549 if (BP_SHOULD_BYTESWAP(bp
))
550 drro
->drr_flags
|= DRR_RAW_BYTESWAP
;
552 /* needed for reconstructing dnp on recv side */
553 drro
->drr_indblkshift
= dnp
->dn_indblkshift
;
554 drro
->drr_nlevels
= dnp
->dn_nlevels
;
555 drro
->drr_nblkptr
= dnp
->dn_nblkptr
;
558 * Since we encrypt the entire bonus area, the (raw) part
559 * beyond the the bonuslen is actually nonzero, so we need
563 drro
->drr_raw_bonuslen
= DN_MAX_BONUS_LEN(dnp
);
564 bonuslen
= drro
->drr_raw_bonuslen
;
568 if (dump_record(dsp
, DN_BONUS(dnp
), bonuslen
) != 0)
569 return (SET_ERROR(EINTR
));
571 /* Free anything past the end of the file. */
572 if (dump_free(dsp
, object
, (dnp
->dn_maxblkid
+ 1) *
573 (dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
), -1ULL) != 0)
574 return (SET_ERROR(EINTR
));
575 if (dsp
->dsa_err
!= 0)
576 return (SET_ERROR(EINTR
));
581 dump_object_range(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
, uint64_t firstobj
,
584 struct drr_object_range
*drror
=
585 &(dsp
->dsa_drr
->drr_u
.drr_object_range
);
587 /* we only use this record type for raw sends */
588 ASSERT(BP_IS_PROTECTED(bp
));
589 ASSERT(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
);
590 ASSERT3U(BP_GET_COMPRESS(bp
), ==, ZIO_COMPRESS_OFF
);
591 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_DNODE
);
592 ASSERT0(BP_GET_LEVEL(bp
));
594 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
595 if (dump_record(dsp
, NULL
, 0) != 0)
596 return (SET_ERROR(EINTR
));
597 dsp
->dsa_pending_op
= PENDING_NONE
;
600 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
601 dsp
->dsa_drr
->drr_type
= DRR_OBJECT_RANGE
;
602 drror
->drr_firstobj
= firstobj
;
603 drror
->drr_numslots
= numslots
;
604 drror
->drr_toguid
= dsp
->dsa_toguid
;
605 drror
->drr_flags
|= DRR_RAW_ENCRYPTED
;
606 if (BP_SHOULD_BYTESWAP(bp
))
607 drror
->drr_flags
|= DRR_RAW_BYTESWAP
;
608 zio_crypt_decode_params_bp(bp
, drror
->drr_salt
, drror
->drr_iv
);
609 zio_crypt_decode_mac_bp(bp
, drror
->drr_mac
);
611 if (dump_record(dsp
, NULL
, 0) != 0)
612 return (SET_ERROR(EINTR
));
617 backup_do_embed(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
)
619 if (!BP_IS_EMBEDDED(bp
))
623 * Compression function must be legacy, or explicitly enabled.
625 if ((BP_GET_COMPRESS(bp
) >= ZIO_COMPRESS_LEGACY_FUNCTIONS
&&
626 !(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_LZ4
)))
630 * Embed type must be explicitly enabled.
632 switch (BPE_GET_ETYPE(bp
)) {
633 case BP_EMBEDDED_TYPE_DATA
:
634 if (dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)
644 * This is the callback function to traverse_dataset that acts as the worker
645 * thread for dmu_send_impl.
649 send_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
650 const zbookmark_phys_t
*zb
, const struct dnode_phys
*dnp
, void *arg
)
652 struct send_thread_arg
*sta
= arg
;
653 struct send_block_record
*record
;
654 uint64_t record_size
;
657 ASSERT(zb
->zb_object
== DMU_META_DNODE_OBJECT
||
658 zb
->zb_object
>= sta
->resume
.zb_object
);
659 ASSERT3P(sta
->ds
, !=, NULL
);
662 return (SET_ERROR(EINTR
));
665 ASSERT3U(zb
->zb_level
, ==, ZB_DNODE_LEVEL
);
667 } else if (zb
->zb_level
< 0) {
671 record
= kmem_zalloc(sizeof (struct send_block_record
), KM_SLEEP
);
672 record
->eos_marker
= B_FALSE
;
675 record
->indblkshift
= dnp
->dn_indblkshift
;
676 record
->datablkszsec
= dnp
->dn_datablkszsec
;
677 record_size
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
678 bqueue_enqueue(&sta
->q
, record
, record_size
);
684 * This function kicks off the traverse_dataset. It also handles setting the
685 * error code of the thread in case something goes wrong, and pushes the End of
686 * Stream record when the traverse_dataset call has finished. If there is no
687 * dataset to traverse, the thread immediately pushes End of Stream marker.
690 send_traverse_thread(void *arg
)
692 struct send_thread_arg
*st_arg
= arg
;
694 struct send_block_record
*data
;
695 fstrans_cookie_t cookie
= spl_fstrans_mark();
697 if (st_arg
->ds
!= NULL
) {
698 err
= traverse_dataset_resume(st_arg
->ds
,
699 st_arg
->fromtxg
, &st_arg
->resume
,
700 st_arg
->flags
, send_cb
, st_arg
);
703 st_arg
->error_code
= err
;
705 data
= kmem_zalloc(sizeof (*data
), KM_SLEEP
);
706 data
->eos_marker
= B_TRUE
;
707 bqueue_enqueue(&st_arg
->q
, data
, 1);
708 spl_fstrans_unmark(cookie
);
713 * This function actually handles figuring out what kind of record needs to be
714 * dumped, reading the data (which has hopefully been prefetched), and calling
715 * the appropriate helper function.
718 do_dump(dmu_sendarg_t
*dsa
, struct send_block_record
*data
)
720 dsl_dataset_t
*ds
= dmu_objset_ds(dsa
->dsa_os
);
721 const blkptr_t
*bp
= &data
->bp
;
722 const zbookmark_phys_t
*zb
= &data
->zb
;
723 uint8_t indblkshift
= data
->indblkshift
;
724 uint16_t dblkszsec
= data
->datablkszsec
;
725 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
726 dmu_object_type_t type
= bp
? BP_GET_TYPE(bp
) : DMU_OT_NONE
;
730 ASSERT3U(zb
->zb_level
, >=, 0);
732 ASSERT(zb
->zb_object
== DMU_META_DNODE_OBJECT
||
733 zb
->zb_object
>= dsa
->dsa_resume_object
);
736 * All bps of an encrypted os should have the encryption bit set.
737 * If this is not true it indicates tampering and we report an error.
739 if (dsa
->dsa_os
->os_encrypted
&&
740 !BP_IS_HOLE(bp
) && !BP_USES_CRYPT(bp
)) {
741 spa_log_error(spa
, zb
);
742 zfs_panic_recover("unencrypted block in encrypted "
743 "object set %llu", ds
->ds_object
);
744 return (SET_ERROR(EIO
));
747 if (zb
->zb_object
!= DMU_META_DNODE_OBJECT
&&
748 DMU_OBJECT_IS_SPECIAL(zb
->zb_object
)) {
750 } else if (BP_IS_HOLE(bp
) &&
751 zb
->zb_object
== DMU_META_DNODE_OBJECT
) {
752 uint64_t span
= BP_SPAN(dblkszsec
, indblkshift
, zb
->zb_level
);
753 uint64_t dnobj
= (zb
->zb_blkid
* span
) >> DNODE_SHIFT
;
754 err
= dump_freeobjects(dsa
, dnobj
, span
>> DNODE_SHIFT
);
755 } else if (BP_IS_HOLE(bp
)) {
756 uint64_t span
= BP_SPAN(dblkszsec
, indblkshift
, zb
->zb_level
);
757 uint64_t offset
= zb
->zb_blkid
* span
;
758 err
= dump_free(dsa
, zb
->zb_object
, offset
, span
);
759 } else if (zb
->zb_level
> 0 || type
== DMU_OT_OBJSET
) {
761 } else if (type
== DMU_OT_DNODE
) {
763 int epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
764 arc_flags_t aflags
= ARC_FLAG_WAIT
;
766 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
769 if (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
770 ASSERT(BP_IS_ENCRYPTED(bp
));
771 ASSERT3U(BP_GET_COMPRESS(bp
), ==, ZIO_COMPRESS_OFF
);
772 zioflags
|= ZIO_FLAG_RAW
;
775 ASSERT0(zb
->zb_level
);
777 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
778 ZIO_PRIORITY_ASYNC_READ
, zioflags
, &aflags
, zb
) != 0)
779 return (SET_ERROR(EIO
));
782 dnobj
= zb
->zb_blkid
* epb
;
785 * Raw sends require sending encryption parameters for the
786 * block of dnodes. Regular sends do not need to send this
789 if (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
790 ASSERT(arc_is_encrypted(abuf
));
791 err
= dump_object_range(dsa
, bp
, dnobj
, epb
);
795 for (i
= 0; i
< epb
; i
+= blk
[i
].dn_extra_slots
+ 1) {
796 err
= dump_dnode(dsa
, bp
, dnobj
+ i
, blk
+ i
);
801 arc_buf_destroy(abuf
, &abuf
);
802 } else if (type
== DMU_OT_SA
) {
803 arc_flags_t aflags
= ARC_FLAG_WAIT
;
805 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
807 if (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
808 ASSERT(BP_IS_PROTECTED(bp
));
809 zioflags
|= ZIO_FLAG_RAW
;
812 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
813 ZIO_PRIORITY_ASYNC_READ
, zioflags
, &aflags
, zb
) != 0)
814 return (SET_ERROR(EIO
));
816 err
= dump_spill(dsa
, bp
, zb
->zb_object
, abuf
->b_data
);
817 arc_buf_destroy(abuf
, &abuf
);
818 } else if (backup_do_embed(dsa
, bp
)) {
819 /* it's an embedded level-0 block of a regular object */
820 int blksz
= dblkszsec
<< SPA_MINBLOCKSHIFT
;
821 ASSERT0(zb
->zb_level
);
822 err
= dump_write_embedded(dsa
, zb
->zb_object
,
823 zb
->zb_blkid
* blksz
, blksz
, bp
);
825 /* it's a level-0 block of a regular object */
826 arc_flags_t aflags
= ARC_FLAG_WAIT
;
828 int blksz
= dblkszsec
<< SPA_MINBLOCKSHIFT
;
832 * If we have large blocks stored on disk but the send flags
833 * don't allow us to send large blocks, we split the data from
834 * the arc buf into chunks.
836 boolean_t split_large_blocks
= blksz
> SPA_OLD_MAXBLOCKSIZE
&&
837 !(dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
);
840 * Raw sends require that we always get raw data as it exists
841 * on disk, so we assert that we are not splitting blocks here.
843 boolean_t request_raw
=
844 (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0;
847 * We should only request compressed data from the ARC if all
848 * the following are true:
849 * - stream compression was requested
850 * - we aren't splitting large blocks into smaller chunks
851 * - the data won't need to be byteswapped before sending
852 * - this isn't an embedded block
853 * - this isn't metadata (if receiving on a different endian
854 * system it can be byteswapped more easily)
856 boolean_t request_compressed
=
857 (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
) &&
858 !split_large_blocks
&& !BP_SHOULD_BYTESWAP(bp
) &&
859 !BP_IS_EMBEDDED(bp
) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp
));
861 IMPLY(request_raw
, !split_large_blocks
);
862 IMPLY(request_raw
, BP_IS_PROTECTED(bp
));
863 ASSERT0(zb
->zb_level
);
864 ASSERT(zb
->zb_object
> dsa
->dsa_resume_object
||
865 (zb
->zb_object
== dsa
->dsa_resume_object
&&
866 zb
->zb_blkid
* blksz
>= dsa
->dsa_resume_offset
));
868 ASSERT3U(blksz
, ==, BP_GET_LSIZE(bp
));
870 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
872 zioflags
|= ZIO_FLAG_RAW
;
873 else if (request_compressed
)
874 zioflags
|= ZIO_FLAG_RAW_COMPRESS
;
876 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
877 ZIO_PRIORITY_ASYNC_READ
, zioflags
, &aflags
, zb
) != 0) {
878 if (zfs_send_corrupt_data
) {
879 /* Send a block filled with 0x"zfs badd bloc" */
880 abuf
= arc_alloc_buf(spa
, &abuf
, ARC_BUFC_DATA
,
883 for (ptr
= abuf
->b_data
;
884 (char *)ptr
< (char *)abuf
->b_data
+ blksz
;
886 *ptr
= 0x2f5baddb10cULL
;
888 return (SET_ERROR(EIO
));
892 offset
= zb
->zb_blkid
* blksz
;
894 if (split_large_blocks
) {
895 ASSERT0(arc_is_encrypted(abuf
));
896 ASSERT3U(arc_get_compression(abuf
), ==,
898 char *buf
= abuf
->b_data
;
899 while (blksz
> 0 && err
== 0) {
900 int n
= MIN(blksz
, SPA_OLD_MAXBLOCKSIZE
);
901 err
= dump_write(dsa
, type
, zb
->zb_object
,
902 offset
, n
, n
, NULL
, buf
);
908 err
= dump_write(dsa
, type
, zb
->zb_object
, offset
,
909 blksz
, arc_buf_size(abuf
), bp
, abuf
->b_data
);
911 arc_buf_destroy(abuf
, &abuf
);
914 ASSERT(err
== 0 || err
== EINTR
);
919 * Pop the new data off the queue, and free the old data.
921 static struct send_block_record
*
922 get_next_record(bqueue_t
*bq
, struct send_block_record
*data
)
924 struct send_block_record
*tmp
= bqueue_dequeue(bq
);
925 kmem_free(data
, sizeof (*data
));
930 * Actually do the bulk of the work in a zfs send.
932 * Note: Releases dp using the specified tag.
935 dmu_send_impl(void *tag
, dsl_pool_t
*dp
, dsl_dataset_t
*to_ds
,
936 zfs_bookmark_phys_t
*ancestor_zb
, boolean_t is_clone
,
937 boolean_t embedok
, boolean_t large_block_ok
, boolean_t compressok
,
938 boolean_t rawok
, int outfd
, uint64_t resumeobj
, uint64_t resumeoff
,
939 vnode_t
*vp
, offset_t
*off
)
942 dmu_replay_record_t
*drr
;
945 uint64_t fromtxg
= 0;
946 uint64_t featureflags
= 0;
947 struct send_thread_arg to_arg
;
948 void *payload
= NULL
;
949 size_t payload_len
= 0;
950 struct send_block_record
*to_data
;
952 err
= dmu_objset_from_ds(to_ds
, &os
);
954 dsl_pool_rele(dp
, tag
);
959 * If this is a non-raw send of an encrypted ds, we can ensure that
960 * the objset_phys_t is authenticated. This is safe because this is
961 * either a snapshot or we have owned the dataset, ensuring that
962 * it can't be modified.
964 if (!rawok
&& os
->os_encrypted
&&
965 arc_is_unauthenticated(os
->os_phys_buf
)) {
966 err
= arc_untransform(os
->os_phys_buf
, os
->os_spa
,
967 to_ds
->ds_object
, B_FALSE
);
969 dsl_pool_rele(dp
, tag
);
973 ASSERT0(arc_is_unauthenticated(os
->os_phys_buf
));
976 drr
= kmem_zalloc(sizeof (dmu_replay_record_t
), KM_SLEEP
);
977 drr
->drr_type
= DRR_BEGIN
;
978 drr
->drr_u
.drr_begin
.drr_magic
= DMU_BACKUP_MAGIC
;
979 DMU_SET_STREAM_HDRTYPE(drr
->drr_u
.drr_begin
.drr_versioninfo
,
982 bzero(&to_arg
, sizeof (to_arg
));
985 if (dmu_objset_type(os
) == DMU_OST_ZFS
) {
987 if (zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &version
) != 0) {
988 kmem_free(drr
, sizeof (dmu_replay_record_t
));
989 dsl_pool_rele(dp
, tag
);
990 return (SET_ERROR(EINVAL
));
992 if (version
>= ZPL_VERSION_SA
) {
993 featureflags
|= DMU_BACKUP_FEATURE_SA_SPILL
;
998 /* raw sends imply large_block_ok */
999 if ((large_block_ok
|| rawok
) &&
1000 to_ds
->ds_feature_inuse
[SPA_FEATURE_LARGE_BLOCKS
])
1001 featureflags
|= DMU_BACKUP_FEATURE_LARGE_BLOCKS
;
1002 if (to_ds
->ds_feature_inuse
[SPA_FEATURE_LARGE_DNODE
])
1003 featureflags
|= DMU_BACKUP_FEATURE_LARGE_DNODE
;
1005 /* encrypted datasets will not have embedded blocks */
1006 if ((embedok
|| rawok
) && !os
->os_encrypted
&&
1007 spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
)) {
1008 featureflags
|= DMU_BACKUP_FEATURE_EMBED_DATA
;
1011 /* raw send implies compressok */
1012 if (compressok
|| rawok
)
1013 featureflags
|= DMU_BACKUP_FEATURE_COMPRESSED
;
1014 if (rawok
&& os
->os_encrypted
)
1015 featureflags
|= DMU_BACKUP_FEATURE_RAW
;
1018 (DMU_BACKUP_FEATURE_EMBED_DATA
| DMU_BACKUP_FEATURE_COMPRESSED
|
1019 DMU_BACKUP_FEATURE_RAW
)) != 0 &&
1020 spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
)) {
1021 featureflags
|= DMU_BACKUP_FEATURE_LZ4
;
1024 if (resumeobj
!= 0 || resumeoff
!= 0) {
1025 featureflags
|= DMU_BACKUP_FEATURE_RESUMING
;
1028 DMU_SET_FEATUREFLAGS(drr
->drr_u
.drr_begin
.drr_versioninfo
,
1031 drr
->drr_u
.drr_begin
.drr_creation_time
=
1032 dsl_dataset_phys(to_ds
)->ds_creation_time
;
1033 drr
->drr_u
.drr_begin
.drr_type
= dmu_objset_type(os
);
1035 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CLONE
;
1036 drr
->drr_u
.drr_begin
.drr_toguid
= dsl_dataset_phys(to_ds
)->ds_guid
;
1037 if (dsl_dataset_phys(to_ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
1038 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CI_DATA
;
1039 if (zfs_send_set_freerecords_bit
)
1040 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_FREERECORDS
;
1042 if (ancestor_zb
!= NULL
) {
1043 drr
->drr_u
.drr_begin
.drr_fromguid
=
1044 ancestor_zb
->zbm_guid
;
1045 fromtxg
= ancestor_zb
->zbm_creation_txg
;
1047 dsl_dataset_name(to_ds
, drr
->drr_u
.drr_begin
.drr_toname
);
1048 if (!to_ds
->ds_is_snapshot
) {
1049 (void) strlcat(drr
->drr_u
.drr_begin
.drr_toname
, "@--head--",
1050 sizeof (drr
->drr_u
.drr_begin
.drr_toname
));
1053 dsp
= kmem_zalloc(sizeof (dmu_sendarg_t
), KM_SLEEP
);
1057 dsp
->dsa_outfd
= outfd
;
1058 dsp
->dsa_proc
= curproc
;
1061 dsp
->dsa_toguid
= dsl_dataset_phys(to_ds
)->ds_guid
;
1062 dsp
->dsa_pending_op
= PENDING_NONE
;
1063 dsp
->dsa_featureflags
= featureflags
;
1064 dsp
->dsa_resume_object
= resumeobj
;
1065 dsp
->dsa_resume_offset
= resumeoff
;
1067 mutex_enter(&to_ds
->ds_sendstream_lock
);
1068 list_insert_head(&to_ds
->ds_sendstreams
, dsp
);
1069 mutex_exit(&to_ds
->ds_sendstream_lock
);
1071 dsl_dataset_long_hold(to_ds
, FTAG
);
1072 dsl_pool_rele(dp
, tag
);
1074 /* handle features that require a DRR_BEGIN payload */
1076 (DMU_BACKUP_FEATURE_RESUMING
| DMU_BACKUP_FEATURE_RAW
)) {
1077 nvlist_t
*keynvl
= NULL
;
1078 nvlist_t
*nvl
= fnvlist_alloc();
1080 if (featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
1081 dmu_object_info_t to_doi
;
1082 err
= dmu_object_info(os
, resumeobj
, &to_doi
);
1088 SET_BOOKMARK(&to_arg
.resume
, to_ds
->ds_object
,
1090 resumeoff
/ to_doi
.doi_data_block_size
);
1092 fnvlist_add_uint64(nvl
, "resume_object", resumeobj
);
1093 fnvlist_add_uint64(nvl
, "resume_offset", resumeoff
);
1096 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1097 ASSERT(os
->os_encrypted
);
1099 err
= dsl_crypto_populate_key_nvlist(to_ds
, &keynvl
);
1105 fnvlist_add_nvlist(nvl
, "crypt_keydata", keynvl
);
1108 payload
= fnvlist_pack(nvl
, &payload_len
);
1109 drr
->drr_payloadlen
= payload_len
;
1110 fnvlist_free(keynvl
);
1114 err
= dump_record(dsp
, payload
, payload_len
);
1115 fnvlist_pack_free(payload
, payload_len
);
1121 err
= bqueue_init(&to_arg
.q
, zfs_send_queue_length
,
1122 offsetof(struct send_block_record
, ln
));
1123 to_arg
.error_code
= 0;
1124 to_arg
.cancel
= B_FALSE
;
1126 to_arg
.fromtxg
= fromtxg
;
1127 to_arg
.flags
= TRAVERSE_PRE
| TRAVERSE_PREFETCH
;
1129 to_arg
.flags
|= TRAVERSE_NO_DECRYPT
;
1130 (void) thread_create(NULL
, 0, send_traverse_thread
, &to_arg
, 0, curproc
,
1131 TS_RUN
, minclsyspri
);
1133 to_data
= bqueue_dequeue(&to_arg
.q
);
1135 while (!to_data
->eos_marker
&& err
== 0) {
1136 err
= do_dump(dsp
, to_data
);
1137 to_data
= get_next_record(&to_arg
.q
, to_data
);
1138 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1143 to_arg
.cancel
= B_TRUE
;
1144 while (!to_data
->eos_marker
) {
1145 to_data
= get_next_record(&to_arg
.q
, to_data
);
1148 kmem_free(to_data
, sizeof (*to_data
));
1150 bqueue_destroy(&to_arg
.q
);
1152 if (err
== 0 && to_arg
.error_code
!= 0)
1153 err
= to_arg
.error_code
;
1158 if (dsp
->dsa_pending_op
!= PENDING_NONE
)
1159 if (dump_record(dsp
, NULL
, 0) != 0)
1160 err
= SET_ERROR(EINTR
);
1163 if (err
== EINTR
&& dsp
->dsa_err
!= 0)
1168 bzero(drr
, sizeof (dmu_replay_record_t
));
1169 drr
->drr_type
= DRR_END
;
1170 drr
->drr_u
.drr_end
.drr_checksum
= dsp
->dsa_zc
;
1171 drr
->drr_u
.drr_end
.drr_toguid
= dsp
->dsa_toguid
;
1173 if (dump_record(dsp
, NULL
, 0) != 0)
1176 mutex_enter(&to_ds
->ds_sendstream_lock
);
1177 list_remove(&to_ds
->ds_sendstreams
, dsp
);
1178 mutex_exit(&to_ds
->ds_sendstream_lock
);
1180 VERIFY(err
!= 0 || (dsp
->dsa_sent_begin
&& dsp
->dsa_sent_end
));
1182 kmem_free(drr
, sizeof (dmu_replay_record_t
));
1183 kmem_free(dsp
, sizeof (dmu_sendarg_t
));
1185 dsl_dataset_long_rele(to_ds
, FTAG
);
1191 dmu_send_obj(const char *pool
, uint64_t tosnap
, uint64_t fromsnap
,
1192 boolean_t embedok
, boolean_t large_block_ok
, boolean_t compressok
,
1193 boolean_t rawok
, int outfd
, vnode_t
*vp
, offset_t
*off
)
1197 dsl_dataset_t
*fromds
= NULL
;
1198 ds_hold_flags_t dsflags
= (rawok
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
1201 err
= dsl_pool_hold(pool
, FTAG
, &dp
);
1205 err
= dsl_dataset_hold_obj_flags(dp
, tosnap
, dsflags
, FTAG
, &ds
);
1207 dsl_pool_rele(dp
, FTAG
);
1211 if (fromsnap
!= 0) {
1212 zfs_bookmark_phys_t zb
;
1215 err
= dsl_dataset_hold_obj(dp
, fromsnap
, FTAG
, &fromds
);
1217 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1218 dsl_pool_rele(dp
, FTAG
);
1221 if (!dsl_dataset_is_before(ds
, fromds
, 0))
1222 err
= SET_ERROR(EXDEV
);
1223 zb
.zbm_creation_time
=
1224 dsl_dataset_phys(fromds
)->ds_creation_time
;
1225 zb
.zbm_creation_txg
= dsl_dataset_phys(fromds
)->ds_creation_txg
;
1226 zb
.zbm_guid
= dsl_dataset_phys(fromds
)->ds_guid
;
1227 is_clone
= (fromds
->ds_dir
!= ds
->ds_dir
);
1228 dsl_dataset_rele(fromds
, FTAG
);
1229 err
= dmu_send_impl(FTAG
, dp
, ds
, &zb
, is_clone
,
1230 embedok
, large_block_ok
, compressok
, rawok
, outfd
,
1233 err
= dmu_send_impl(FTAG
, dp
, ds
, NULL
, B_FALSE
,
1234 embedok
, large_block_ok
, compressok
, rawok
, outfd
,
1237 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1242 dmu_send(const char *tosnap
, const char *fromsnap
, boolean_t embedok
,
1243 boolean_t large_block_ok
, boolean_t compressok
, boolean_t rawok
,
1244 int outfd
, uint64_t resumeobj
, uint64_t resumeoff
, vnode_t
*vp
,
1250 ds_hold_flags_t dsflags
= (rawok
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
1251 boolean_t owned
= B_FALSE
;
1253 if (fromsnap
!= NULL
&& strpbrk(fromsnap
, "@#") == NULL
)
1254 return (SET_ERROR(EINVAL
));
1256 err
= dsl_pool_hold(tosnap
, FTAG
, &dp
);
1260 if (strchr(tosnap
, '@') == NULL
&& spa_writeable(dp
->dp_spa
)) {
1262 * We are sending a filesystem or volume. Ensure
1263 * that it doesn't change by owning the dataset.
1265 err
= dsl_dataset_own(dp
, tosnap
, dsflags
, FTAG
, &ds
);
1268 err
= dsl_dataset_hold_flags(dp
, tosnap
, dsflags
, FTAG
, &ds
);
1271 dsl_pool_rele(dp
, FTAG
);
1275 if (fromsnap
!= NULL
) {
1276 zfs_bookmark_phys_t zb
;
1277 boolean_t is_clone
= B_FALSE
;
1278 int fsnamelen
= strchr(tosnap
, '@') - tosnap
;
1281 * If the fromsnap is in a different filesystem, then
1282 * mark the send stream as a clone.
1284 if (strncmp(tosnap
, fromsnap
, fsnamelen
) != 0 ||
1285 (fromsnap
[fsnamelen
] != '@' &&
1286 fromsnap
[fsnamelen
] != '#')) {
1290 if (strchr(fromsnap
, '@')) {
1291 dsl_dataset_t
*fromds
;
1292 err
= dsl_dataset_hold(dp
, fromsnap
, FTAG
, &fromds
);
1294 if (!dsl_dataset_is_before(ds
, fromds
, 0))
1295 err
= SET_ERROR(EXDEV
);
1296 zb
.zbm_creation_time
=
1297 dsl_dataset_phys(fromds
)->ds_creation_time
;
1298 zb
.zbm_creation_txg
=
1299 dsl_dataset_phys(fromds
)->ds_creation_txg
;
1300 zb
.zbm_guid
= dsl_dataset_phys(fromds
)->ds_guid
;
1301 is_clone
= (ds
->ds_dir
!= fromds
->ds_dir
);
1302 dsl_dataset_rele(fromds
, FTAG
);
1305 err
= dsl_bookmark_lookup(dp
, fromsnap
, ds
, &zb
);
1309 dsl_dataset_disown(ds
, dsflags
, FTAG
);
1311 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1313 dsl_pool_rele(dp
, FTAG
);
1316 err
= dmu_send_impl(FTAG
, dp
, ds
, &zb
, is_clone
,
1317 embedok
, large_block_ok
, compressok
, rawok
,
1318 outfd
, resumeobj
, resumeoff
, vp
, off
);
1320 err
= dmu_send_impl(FTAG
, dp
, ds
, NULL
, B_FALSE
,
1321 embedok
, large_block_ok
, compressok
, rawok
,
1322 outfd
, resumeobj
, resumeoff
, vp
, off
);
1325 dsl_dataset_disown(ds
, dsflags
, FTAG
);
1327 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1333 dmu_adjust_send_estimate_for_indirects(dsl_dataset_t
*ds
, uint64_t uncompressed
,
1334 uint64_t compressed
, boolean_t stream_compressed
, uint64_t *sizep
)
1339 * Assume that space (both on-disk and in-stream) is dominated by
1340 * data. We will adjust for indirect blocks and the copies property,
1341 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
1344 uint64_t recordsize
;
1345 uint64_t record_count
;
1347 VERIFY0(dmu_objset_from_ds(ds
, &os
));
1349 /* Assume all (uncompressed) blocks are recordsize. */
1350 if (os
->os_phys
->os_type
== DMU_OST_ZVOL
) {
1351 err
= dsl_prop_get_int_ds(ds
,
1352 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
), &recordsize
);
1354 err
= dsl_prop_get_int_ds(ds
,
1355 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
), &recordsize
);
1359 record_count
= uncompressed
/ recordsize
;
1362 * If we're estimating a send size for a compressed stream, use the
1363 * compressed data size to estimate the stream size. Otherwise, use the
1364 * uncompressed data size.
1366 size
= stream_compressed
? compressed
: uncompressed
;
1369 * Subtract out approximate space used by indirect blocks.
1370 * Assume most space is used by data blocks (non-indirect, non-dnode).
1371 * Assume no ditto blocks or internal fragmentation.
1373 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
1376 size
-= record_count
* sizeof (blkptr_t
);
1378 /* Add in the space for the record associated with each block. */
1379 size
+= record_count
* sizeof (dmu_replay_record_t
);
1387 dmu_send_estimate(dsl_dataset_t
*ds
, dsl_dataset_t
*fromds
,
1388 boolean_t stream_compressed
, uint64_t *sizep
)
1391 uint64_t uncomp
, comp
;
1393 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1395 /* tosnap must be a snapshot */
1396 if (!ds
->ds_is_snapshot
)
1397 return (SET_ERROR(EINVAL
));
1399 /* fromsnap, if provided, must be a snapshot */
1400 if (fromds
!= NULL
&& !fromds
->ds_is_snapshot
)
1401 return (SET_ERROR(EINVAL
));
1404 * fromsnap must be an earlier snapshot from the same fs as tosnap,
1405 * or the origin's fs.
1407 if (fromds
!= NULL
&& !dsl_dataset_is_before(ds
, fromds
, 0))
1408 return (SET_ERROR(EXDEV
));
1410 /* Get compressed and uncompressed size estimates of changed data. */
1411 if (fromds
== NULL
) {
1412 uncomp
= dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1413 comp
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1416 err
= dsl_dataset_space_written(fromds
, ds
,
1417 &used
, &comp
, &uncomp
);
1422 err
= dmu_adjust_send_estimate_for_indirects(ds
, uncomp
, comp
,
1423 stream_compressed
, sizep
);
1425 * Add the size of the BEGIN and END records to the estimate.
1427 *sizep
+= 2 * sizeof (dmu_replay_record_t
);
1431 struct calculate_send_arg
{
1432 uint64_t uncompressed
;
1433 uint64_t compressed
;
1437 * Simple callback used to traverse the blocks of a snapshot and sum their
1438 * uncompressed and compressed sizes.
1442 dmu_calculate_send_traversal(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1443 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1445 struct calculate_send_arg
*space
= arg
;
1446 if (bp
!= NULL
&& !BP_IS_HOLE(bp
)) {
1447 space
->uncompressed
+= BP_GET_UCSIZE(bp
);
1448 space
->compressed
+= BP_GET_PSIZE(bp
);
1454 * Given a desination snapshot and a TXG, calculate the approximate size of a
1455 * send stream sent from that TXG. from_txg may be zero, indicating that the
1456 * whole snapshot will be sent.
1459 dmu_send_estimate_from_txg(dsl_dataset_t
*ds
, uint64_t from_txg
,
1460 boolean_t stream_compressed
, uint64_t *sizep
)
1463 struct calculate_send_arg size
= { 0 };
1465 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1467 /* tosnap must be a snapshot */
1468 if (!dsl_dataset_is_snapshot(ds
))
1469 return (SET_ERROR(EINVAL
));
1471 /* verify that from_txg is before the provided snapshot was taken */
1472 if (from_txg
>= dsl_dataset_phys(ds
)->ds_creation_txg
) {
1473 return (SET_ERROR(EXDEV
));
1476 * traverse the blocks of the snapshot with birth times after
1477 * from_txg, summing their uncompressed size
1479 err
= traverse_dataset(ds
, from_txg
,
1480 TRAVERSE_POST
| TRAVERSE_NO_DECRYPT
,
1481 dmu_calculate_send_traversal
, &size
);
1486 err
= dmu_adjust_send_estimate_for_indirects(ds
, size
.uncompressed
,
1487 size
.compressed
, stream_compressed
, sizep
);
1491 typedef struct dmu_recv_begin_arg
{
1492 const char *drba_origin
;
1493 dmu_recv_cookie_t
*drba_cookie
;
1495 uint64_t drba_snapobj
;
1496 } dmu_recv_begin_arg_t
;
1499 recv_begin_check_existing_impl(dmu_recv_begin_arg_t
*drba
, dsl_dataset_t
*ds
,
1504 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1506 /* temporary clone name must not exist */
1507 error
= zap_lookup(dp
->dp_meta_objset
,
1508 dsl_dir_phys(ds
->ds_dir
)->dd_child_dir_zapobj
, recv_clone_name
,
1510 if (error
!= ENOENT
)
1511 return (error
== 0 ? EBUSY
: error
);
1513 /* new snapshot name must not exist */
1514 error
= zap_lookup(dp
->dp_meta_objset
,
1515 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1516 drba
->drba_cookie
->drc_tosnap
, 8, 1, &val
);
1517 if (error
!= ENOENT
)
1518 return (error
== 0 ? EEXIST
: error
);
1521 * Check snapshot limit before receiving. We'll recheck again at the
1522 * end, but might as well abort before receiving if we're already over
1525 * Note that we do not check the file system limit with
1526 * dsl_dir_fscount_check because the temporary %clones don't count
1527 * against that limit.
1529 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1, ZFS_PROP_SNAPSHOT_LIMIT
,
1530 NULL
, drba
->drba_cred
);
1534 if (fromguid
!= 0) {
1535 dsl_dataset_t
*snap
;
1536 uint64_t obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1538 /* Find snapshot in this dir that matches fromguid. */
1540 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
1543 return (SET_ERROR(ENODEV
));
1544 if (snap
->ds_dir
!= ds
->ds_dir
) {
1545 dsl_dataset_rele(snap
, FTAG
);
1546 return (SET_ERROR(ENODEV
));
1548 if (dsl_dataset_phys(snap
)->ds_guid
== fromguid
)
1550 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
1551 dsl_dataset_rele(snap
, FTAG
);
1554 return (SET_ERROR(ENODEV
));
1556 if (drba
->drba_cookie
->drc_force
) {
1557 drba
->drba_snapobj
= obj
;
1560 * If we are not forcing, there must be no
1561 * changes since fromsnap.
1563 if (dsl_dataset_modified_since_snap(ds
, snap
)) {
1564 dsl_dataset_rele(snap
, FTAG
);
1565 return (SET_ERROR(ETXTBSY
));
1567 drba
->drba_snapobj
= ds
->ds_prev
->ds_object
;
1570 dsl_dataset_rele(snap
, FTAG
);
1572 /* if full, then must be forced */
1573 if (!drba
->drba_cookie
->drc_force
)
1574 return (SET_ERROR(EEXIST
));
1577 * We don't support using zfs recv -F to blow away
1578 * encrypted filesystems. This would require the
1579 * dsl dir to point to the old encryption key and
1580 * the new one at the same time during the receive.
1582 if (ds
->ds_dir
->dd_crypto_obj
!= 0)
1583 return (SET_ERROR(EINVAL
));
1585 drba
->drba_snapobj
= 0;
1593 dmu_recv_begin_check(void *arg
, dmu_tx_t
*tx
)
1595 dmu_recv_begin_arg_t
*drba
= arg
;
1596 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1597 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1598 uint64_t fromguid
= drrb
->drr_fromguid
;
1599 int flags
= drrb
->drr_flags
;
1600 ds_hold_flags_t dsflags
= 0;
1602 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1604 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1606 /* already checked */
1607 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1608 ASSERT(!(featureflags
& DMU_BACKUP_FEATURE_RESUMING
));
1610 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1611 DMU_COMPOUNDSTREAM
||
1612 drrb
->drr_type
>= DMU_OST_NUMTYPES
||
1613 ((flags
& DRR_FLAG_CLONE
) && drba
->drba_origin
== NULL
))
1614 return (SET_ERROR(EINVAL
));
1616 /* Verify pool version supports SA if SA_SPILL feature set */
1617 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
1618 spa_version(dp
->dp_spa
) < SPA_VERSION_SA
)
1619 return (SET_ERROR(ENOTSUP
));
1621 if (drba
->drba_cookie
->drc_resumable
&&
1622 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EXTENSIBLE_DATASET
))
1623 return (SET_ERROR(ENOTSUP
));
1626 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1627 * record to a plain WRITE record, so the pool must have the
1628 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1629 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1631 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
1632 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
))
1633 return (SET_ERROR(ENOTSUP
));
1634 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
1635 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
))
1636 return (SET_ERROR(ENOTSUP
));
1639 * The receiving code doesn't know how to translate large blocks
1640 * to smaller ones, so the pool must have the LARGE_BLOCKS
1641 * feature enabled if the stream has LARGE_BLOCKS. Same with
1644 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
1645 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_BLOCKS
))
1646 return (SET_ERROR(ENOTSUP
));
1647 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_DNODE
) &&
1648 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_DNODE
))
1649 return (SET_ERROR(ENOTSUP
));
1651 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
)) {
1652 /* raw receives require the encryption feature */
1653 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_ENCRYPTION
))
1654 return (SET_ERROR(ENOTSUP
));
1656 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1659 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1661 /* target fs already exists; recv into temp clone */
1663 /* Can't recv a clone into an existing fs */
1664 if (flags
& DRR_FLAG_CLONE
|| drba
->drba_origin
) {
1665 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1666 return (SET_ERROR(EINVAL
));
1669 error
= recv_begin_check_existing_impl(drba
, ds
, fromguid
);
1670 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1671 } else if (error
== ENOENT
) {
1672 /* target fs does not exist; must be a full backup or clone */
1673 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1676 * If it's a non-clone incremental, we are missing the
1677 * target fs, so fail the recv.
1679 if (fromguid
!= 0 && !(flags
& DRR_FLAG_CLONE
||
1681 return (SET_ERROR(ENOENT
));
1684 * If we're receiving a full send as a clone, and it doesn't
1685 * contain all the necessary free records and freeobject
1686 * records, reject it.
1688 if (fromguid
== 0 && drba
->drba_origin
&&
1689 !(flags
& DRR_FLAG_FREERECORDS
))
1690 return (SET_ERROR(EINVAL
));
1692 /* Open the parent of tofs */
1693 ASSERT3U(strlen(tofs
), <, sizeof (buf
));
1694 (void) strlcpy(buf
, tofs
, strrchr(tofs
, '/') - tofs
+ 1);
1695 error
= dsl_dataset_hold_flags(dp
, buf
, dsflags
, FTAG
, &ds
);
1700 * Check filesystem and snapshot limits before receiving. We'll
1701 * recheck snapshot limits again at the end (we create the
1702 * filesystems and increment those counts during begin_sync).
1704 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
1705 ZFS_PROP_FILESYSTEM_LIMIT
, NULL
, drba
->drba_cred
);
1707 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1711 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
1712 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, drba
->drba_cred
);
1714 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1718 if (drba
->drba_origin
!= NULL
) {
1719 dsl_dataset_t
*origin
;
1721 error
= dsl_dataset_hold_flags(dp
, drba
->drba_origin
,
1722 dsflags
, FTAG
, &origin
);
1724 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1727 if (!origin
->ds_is_snapshot
) {
1728 dsl_dataset_rele_flags(origin
,
1729 DS_HOLD_FLAG_DECRYPT
, FTAG
);
1730 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1731 return (SET_ERROR(EINVAL
));
1733 if (dsl_dataset_phys(origin
)->ds_guid
!= fromguid
&&
1735 dsl_dataset_rele_flags(origin
,
1736 DS_HOLD_FLAG_DECRYPT
, FTAG
);
1737 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1738 return (SET_ERROR(ENODEV
));
1740 dsl_dataset_rele_flags(origin
,
1743 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1750 dmu_recv_begin_sync(void *arg
, dmu_tx_t
*tx
)
1752 dmu_recv_begin_arg_t
*drba
= arg
;
1753 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1754 objset_t
*mos
= dp
->dp_meta_objset
;
1755 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1756 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1757 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1758 dsl_dataset_t
*ds
, *newds
;
1761 ds_hold_flags_t dsflags
= 0;
1763 uint64_t crflags
= 0;
1764 dsl_crypto_params_t
*dcpp
= NULL
;
1765 dsl_crypto_params_t dcp
= { 0 };
1767 if (drrb
->drr_flags
& DRR_FLAG_CI_DATA
)
1768 crflags
|= DS_FLAG_CI_DATASET
;
1769 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0) {
1770 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1772 dcp
.cp_cmd
= DCP_CMD_RAW_RECV
;
1775 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1777 /* create temporary clone */
1778 dsl_dataset_t
*snap
= NULL
;
1780 if (drba
->drba_snapobj
!= 0) {
1781 VERIFY0(dsl_dataset_hold_obj(dp
,
1782 drba
->drba_snapobj
, FTAG
, &snap
));
1784 /* we use the dcp whenever we are not making a clone */
1788 dsobj
= dsl_dataset_create_sync(ds
->ds_dir
, recv_clone_name
,
1789 snap
, crflags
, drba
->drba_cred
, dcpp
, tx
);
1790 if (drba
->drba_snapobj
!= 0)
1791 dsl_dataset_rele(snap
, FTAG
);
1792 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1796 dsl_dataset_t
*origin
= NULL
;
1798 VERIFY0(dsl_dir_hold(dp
, tofs
, FTAG
, &dd
, &tail
));
1800 if (drba
->drba_origin
!= NULL
) {
1801 VERIFY0(dsl_dataset_hold(dp
, drba
->drba_origin
,
1804 /* we use the dcp whenever we are not making a clone */
1808 /* Create new dataset. */
1809 dsobj
= dsl_dataset_create_sync(dd
, strrchr(tofs
, '/') + 1,
1810 origin
, crflags
, drba
->drba_cred
, dcpp
, tx
);
1812 dsl_dataset_rele(origin
, FTAG
);
1813 dsl_dir_rele(dd
, FTAG
);
1814 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
1816 VERIFY0(dsl_dataset_own_obj(dp
, dsobj
, dsflags
, dmu_recv_tag
, &newds
));
1817 VERIFY0(dmu_objset_from_ds(newds
, &os
));
1819 if (drba
->drba_cookie
->drc_resumable
) {
1823 dsl_dataset_zapify(newds
, tx
);
1824 if (drrb
->drr_fromguid
!= 0) {
1825 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_FROMGUID
,
1826 8, 1, &drrb
->drr_fromguid
, tx
));
1828 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TOGUID
,
1829 8, 1, &drrb
->drr_toguid
, tx
));
1830 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TONAME
,
1831 1, strlen(drrb
->drr_toname
) + 1, drrb
->drr_toname
, tx
));
1832 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OBJECT
,
1834 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OFFSET
,
1836 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_BYTES
,
1838 if (featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) {
1839 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_LARGEBLOCK
,
1842 if (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) {
1843 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_EMBEDOK
,
1846 if (featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
) {
1847 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_COMPRESSOK
,
1850 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1851 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_RAWOK
,
1857 * Usually the os->os_encrypted value is tied to the presence of a
1858 * DSL Crypto Key object in the dd. However, that will not be received
1859 * until dmu_recv_stream(), so we set the value manually for now.
1861 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1862 os
->os_encrypted
= B_TRUE
;
1863 drba
->drba_cookie
->drc_raw
= B_TRUE
;
1866 dmu_buf_will_dirty(newds
->ds_dbuf
, tx
);
1867 dsl_dataset_phys(newds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
1870 * If we actually created a non-clone, we need to create the objset
1871 * in our new dataset. If this is a raw send we postpone this until
1872 * dmu_recv_stream() so that we can allocate the metadnode with the
1873 * properties from the DRR_BEGIN payload.
1875 rrw_enter(&newds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1876 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds
)) &&
1877 (featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0) {
1878 (void) dmu_objset_create_impl(dp
->dp_spa
,
1879 newds
, dsl_dataset_get_blkptr(newds
), drrb
->drr_type
, tx
);
1881 rrw_exit(&newds
->ds_bp_rwlock
, FTAG
);
1883 drba
->drba_cookie
->drc_ds
= newds
;
1885 spa_history_log_internal_ds(newds
, "receive", tx
, "");
1889 dmu_recv_resume_begin_check(void *arg
, dmu_tx_t
*tx
)
1891 dmu_recv_begin_arg_t
*drba
= arg
;
1892 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1893 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1895 ds_hold_flags_t dsflags
= 0;
1896 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1898 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1901 /* 6 extra bytes for /%recv */
1902 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1904 /* already checked */
1905 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1906 ASSERT(featureflags
& DMU_BACKUP_FEATURE_RESUMING
);
1908 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1909 DMU_COMPOUNDSTREAM
||
1910 drrb
->drr_type
>= DMU_OST_NUMTYPES
)
1911 return (SET_ERROR(EINVAL
));
1913 /* Verify pool version supports SA if SA_SPILL feature set */
1914 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
1915 spa_version(dp
->dp_spa
) < SPA_VERSION_SA
)
1916 return (SET_ERROR(ENOTSUP
));
1919 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1920 * record to a plain WRITE record, so the pool must have the
1921 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1922 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1924 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
1925 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
))
1926 return (SET_ERROR(ENOTSUP
));
1927 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
1928 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
))
1929 return (SET_ERROR(ENOTSUP
));
1932 * The receiving code doesn't know how to translate large blocks
1933 * to smaller ones, so the pool must have the LARGE_BLOCKS
1934 * feature enabled if the stream has LARGE_BLOCKS. Same with
1937 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
1938 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_BLOCKS
))
1939 return (SET_ERROR(ENOTSUP
));
1940 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_DNODE
) &&
1941 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_DNODE
))
1942 return (SET_ERROR(ENOTSUP
));
1944 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
1945 tofs
, recv_clone_name
);
1947 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0)
1948 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1950 if (dsl_dataset_hold_flags(dp
, recvname
, dsflags
, FTAG
, &ds
) != 0) {
1951 /* %recv does not exist; continue in tofs */
1952 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1957 /* check that ds is marked inconsistent */
1958 if (!DS_IS_INCONSISTENT(ds
)) {
1959 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1960 return (SET_ERROR(EINVAL
));
1963 /* check that there is resuming data, and that the toguid matches */
1964 if (!dsl_dataset_is_zapified(ds
)) {
1965 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1966 return (SET_ERROR(EINVAL
));
1968 error
= zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1969 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
);
1970 if (error
!= 0 || drrb
->drr_toguid
!= val
) {
1971 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1972 return (SET_ERROR(EINVAL
));
1976 * Check if the receive is still running. If so, it will be owned.
1977 * Note that nothing else can own the dataset (e.g. after the receive
1978 * fails) because it will be marked inconsistent.
1980 if (dsl_dataset_has_owner(ds
)) {
1981 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1982 return (SET_ERROR(EBUSY
));
1985 /* There should not be any snapshots of this fs yet. */
1986 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
->ds_dir
== ds
->ds_dir
) {
1987 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1988 return (SET_ERROR(EINVAL
));
1992 * Note: resume point will be checked when we process the first WRITE
1996 /* check that the origin matches */
1998 (void) zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1999 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
);
2000 if (drrb
->drr_fromguid
!= val
) {
2001 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
2002 return (SET_ERROR(EINVAL
));
2005 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
2010 dmu_recv_resume_begin_sync(void *arg
, dmu_tx_t
*tx
)
2012 dmu_recv_begin_arg_t
*drba
= arg
;
2013 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2014 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
2015 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
2016 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
2019 ds_hold_flags_t dsflags
= 0;
2021 /* 6 extra bytes for /%recv */
2022 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
2024 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
2025 tofs
, recv_clone_name
);
2027 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
2028 drba
->drba_cookie
->drc_raw
= B_TRUE
;
2030 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
2033 if (dsl_dataset_hold_flags(dp
, recvname
, dsflags
, FTAG
, &ds
) != 0) {
2034 /* %recv does not exist; continue in tofs */
2035 VERIFY0(dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
));
2036 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
2039 /* clear the inconsistent flag so that we can own it */
2040 ASSERT(DS_IS_INCONSISTENT(ds
));
2041 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2042 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
2043 dsobj
= ds
->ds_object
;
2044 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
2046 VERIFY0(dsl_dataset_own_obj(dp
, dsobj
, dsflags
, dmu_recv_tag
, &ds
));
2047 VERIFY0(dmu_objset_from_ds(ds
, &os
));
2049 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2050 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
2052 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2053 ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds
)));
2054 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2056 drba
->drba_cookie
->drc_ds
= ds
;
2058 spa_history_log_internal_ds(ds
, "resume receive", tx
, "");
2062 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
2063 * succeeds; otherwise we will leak the holds on the datasets.
2066 dmu_recv_begin(char *tofs
, char *tosnap
, dmu_replay_record_t
*drr_begin
,
2067 boolean_t force
, boolean_t resumable
, char *origin
, dmu_recv_cookie_t
*drc
)
2069 dmu_recv_begin_arg_t drba
= { 0 };
2071 bzero(drc
, sizeof (dmu_recv_cookie_t
));
2072 drc
->drc_drr_begin
= drr_begin
;
2073 drc
->drc_drrb
= &drr_begin
->drr_u
.drr_begin
;
2074 drc
->drc_tosnap
= tosnap
;
2075 drc
->drc_tofs
= tofs
;
2076 drc
->drc_force
= force
;
2077 drc
->drc_resumable
= resumable
;
2078 drc
->drc_cred
= CRED();
2080 if (drc
->drc_drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
)) {
2081 drc
->drc_byteswap
= B_TRUE
;
2082 (void) fletcher_4_incremental_byteswap(drr_begin
,
2083 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
2084 byteswap_record(drr_begin
);
2085 } else if (drc
->drc_drrb
->drr_magic
== DMU_BACKUP_MAGIC
) {
2086 (void) fletcher_4_incremental_native(drr_begin
,
2087 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
2089 return (SET_ERROR(EINVAL
));
2092 drba
.drba_origin
= origin
;
2093 drba
.drba_cookie
= drc
;
2094 drba
.drba_cred
= CRED();
2096 if (DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
) &
2097 DMU_BACKUP_FEATURE_RESUMING
) {
2098 return (dsl_sync_task(tofs
,
2099 dmu_recv_resume_begin_check
, dmu_recv_resume_begin_sync
,
2100 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
));
2102 return (dsl_sync_task(tofs
,
2103 dmu_recv_begin_check
, dmu_recv_begin_sync
,
2104 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
));
2108 struct receive_record_arg
{
2109 dmu_replay_record_t header
;
2110 void *payload
; /* Pointer to a buffer containing the payload */
2112 * If the record is a write, pointer to the arc_buf_t containing the
2117 uint64_t bytes_read
; /* bytes read from stream when record created */
2118 boolean_t eos_marker
; /* Marks the end of the stream */
2122 struct receive_writer_arg
{
2128 * These three args are used to signal to the main thread that we're
2136 /* A map from guid to dataset to help handle dedup'd streams. */
2137 avl_tree_t
*guid_to_ds_map
;
2138 boolean_t resumable
;
2139 uint64_t last_object
, last_offset
;
2140 uint64_t bytes_read
; /* bytes read when current record created */
2144 list_t list
; /* List of struct receive_objnode. */
2146 * Last object looked up. Used to assert that objects are being looked
2147 * up in ascending order.
2149 uint64_t last_lookup
;
2152 struct receive_objnode
{
2157 struct receive_arg
{
2159 vnode_t
*vp
; /* The vnode to read the stream from */
2160 uint64_t voff
; /* The current offset in the stream */
2161 uint64_t bytes_read
;
2163 * A record that has had its payload read in, but hasn't yet been handed
2164 * off to the worker thread.
2166 struct receive_record_arg
*rrd
;
2167 /* A record that has had its header read in, but not its payload. */
2168 struct receive_record_arg
*next_rrd
;
2170 zio_cksum_t prev_cksum
;
2173 uint64_t featureflags
;
2174 /* Sorted list of objects not to issue prefetches for. */
2175 struct objlist ignore_objlist
;
2178 typedef struct guid_map_entry
{
2181 dsl_dataset_t
*gme_ds
;
2186 guid_compare(const void *arg1
, const void *arg2
)
2188 const guid_map_entry_t
*gmep1
= (const guid_map_entry_t
*)arg1
;
2189 const guid_map_entry_t
*gmep2
= (const guid_map_entry_t
*)arg2
;
2191 return (AVL_CMP(gmep1
->guid
, gmep2
->guid
));
2195 free_guid_map_onexit(void *arg
)
2197 avl_tree_t
*ca
= arg
;
2198 void *cookie
= NULL
;
2199 guid_map_entry_t
*gmep
;
2201 while ((gmep
= avl_destroy_nodes(ca
, &cookie
)) != NULL
) {
2202 dsl_dataset_long_rele(gmep
->gme_ds
, gmep
);
2203 dsl_dataset_rele_flags(gmep
->gme_ds
,
2204 (gmep
->raw
) ? 0 : DS_HOLD_FLAG_DECRYPT
, gmep
);
2205 kmem_free(gmep
, sizeof (guid_map_entry_t
));
2208 kmem_free(ca
, sizeof (avl_tree_t
));
2212 receive_read(struct receive_arg
*ra
, int len
, void *buf
)
2217 * The code doesn't rely on this (lengths being multiples of 8). See
2218 * comment in dump_bytes.
2220 ASSERT(len
% 8 == 0 ||
2221 (ra
->featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0);
2223 while (done
< len
) {
2226 ra
->err
= vn_rdwr(UIO_READ
, ra
->vp
,
2227 (char *)buf
+ done
, len
- done
,
2228 ra
->voff
, UIO_SYSSPACE
, FAPPEND
,
2229 RLIM64_INFINITY
, CRED(), &resid
);
2231 if (resid
== len
- done
) {
2233 * Note: ECKSUM indicates that the receive
2234 * was interrupted and can potentially be resumed.
2236 ra
->err
= SET_ERROR(ECKSUM
);
2238 ra
->voff
+= len
- done
- resid
;
2244 ra
->bytes_read
+= len
;
2246 ASSERT3U(done
, ==, len
);
2250 noinline
static void
2251 byteswap_record(dmu_replay_record_t
*drr
)
2253 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
2254 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
2255 drr
->drr_type
= BSWAP_32(drr
->drr_type
);
2256 drr
->drr_payloadlen
= BSWAP_32(drr
->drr_payloadlen
);
2258 switch (drr
->drr_type
) {
2260 DO64(drr_begin
.drr_magic
);
2261 DO64(drr_begin
.drr_versioninfo
);
2262 DO64(drr_begin
.drr_creation_time
);
2263 DO32(drr_begin
.drr_type
);
2264 DO32(drr_begin
.drr_flags
);
2265 DO64(drr_begin
.drr_toguid
);
2266 DO64(drr_begin
.drr_fromguid
);
2269 DO64(drr_object
.drr_object
);
2270 DO32(drr_object
.drr_type
);
2271 DO32(drr_object
.drr_bonustype
);
2272 DO32(drr_object
.drr_blksz
);
2273 DO32(drr_object
.drr_bonuslen
);
2274 DO32(drr_object
.drr_raw_bonuslen
);
2275 DO64(drr_object
.drr_toguid
);
2277 case DRR_FREEOBJECTS
:
2278 DO64(drr_freeobjects
.drr_firstobj
);
2279 DO64(drr_freeobjects
.drr_numobjs
);
2280 DO64(drr_freeobjects
.drr_toguid
);
2283 DO64(drr_write
.drr_object
);
2284 DO32(drr_write
.drr_type
);
2285 DO64(drr_write
.drr_offset
);
2286 DO64(drr_write
.drr_logical_size
);
2287 DO64(drr_write
.drr_toguid
);
2288 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write
.drr_key
.ddk_cksum
);
2289 DO64(drr_write
.drr_key
.ddk_prop
);
2290 DO64(drr_write
.drr_compressed_size
);
2292 case DRR_WRITE_BYREF
:
2293 DO64(drr_write_byref
.drr_object
);
2294 DO64(drr_write_byref
.drr_offset
);
2295 DO64(drr_write_byref
.drr_length
);
2296 DO64(drr_write_byref
.drr_toguid
);
2297 DO64(drr_write_byref
.drr_refguid
);
2298 DO64(drr_write_byref
.drr_refobject
);
2299 DO64(drr_write_byref
.drr_refoffset
);
2300 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write_byref
.
2302 DO64(drr_write_byref
.drr_key
.ddk_prop
);
2304 case DRR_WRITE_EMBEDDED
:
2305 DO64(drr_write_embedded
.drr_object
);
2306 DO64(drr_write_embedded
.drr_offset
);
2307 DO64(drr_write_embedded
.drr_length
);
2308 DO64(drr_write_embedded
.drr_toguid
);
2309 DO32(drr_write_embedded
.drr_lsize
);
2310 DO32(drr_write_embedded
.drr_psize
);
2313 DO64(drr_free
.drr_object
);
2314 DO64(drr_free
.drr_offset
);
2315 DO64(drr_free
.drr_length
);
2316 DO64(drr_free
.drr_toguid
);
2319 DO64(drr_spill
.drr_object
);
2320 DO64(drr_spill
.drr_length
);
2321 DO64(drr_spill
.drr_toguid
);
2322 DO64(drr_spill
.drr_compressed_size
);
2323 DO32(drr_spill
.drr_type
);
2325 case DRR_OBJECT_RANGE
:
2326 DO64(drr_object_range
.drr_firstobj
);
2327 DO64(drr_object_range
.drr_numslots
);
2328 DO64(drr_object_range
.drr_toguid
);
2331 DO64(drr_end
.drr_toguid
);
2332 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_end
.drr_checksum
);
2338 if (drr
->drr_type
!= DRR_BEGIN
) {
2339 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_checksum
.drr_checksum
);
2346 static inline uint8_t
2347 deduce_nblkptr(dmu_object_type_t bonus_type
, uint64_t bonus_size
)
2349 if (bonus_type
== DMU_OT_SA
) {
2353 ((DN_OLD_MAX_BONUSLEN
-
2354 MIN(DN_OLD_MAX_BONUSLEN
, bonus_size
)) >> SPA_BLKPTRSHIFT
));
2359 save_resume_state(struct receive_writer_arg
*rwa
,
2360 uint64_t object
, uint64_t offset
, dmu_tx_t
*tx
)
2362 int txgoff
= dmu_tx_get_txg(tx
) & TXG_MASK
;
2364 if (!rwa
->resumable
)
2368 * We use ds_resume_bytes[] != 0 to indicate that we need to
2369 * update this on disk, so it must not be 0.
2371 ASSERT(rwa
->bytes_read
!= 0);
2374 * We only resume from write records, which have a valid
2375 * (non-meta-dnode) object number.
2377 ASSERT(object
!= 0);
2380 * For resuming to work correctly, we must receive records in order,
2381 * sorted by object,offset. This is checked by the callers, but
2382 * assert it here for good measure.
2384 ASSERT3U(object
, >=, rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
]);
2385 ASSERT(object
!= rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] ||
2386 offset
>= rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
]);
2387 ASSERT3U(rwa
->bytes_read
, >=,
2388 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
]);
2390 rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] = object
;
2391 rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
] = offset
;
2392 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
] = rwa
->bytes_read
;
2396 receive_object(struct receive_writer_arg
*rwa
, struct drr_object
*drro
,
2399 dmu_object_info_t doi
;
2404 if (drro
->drr_type
== DMU_OT_NONE
||
2405 !DMU_OT_IS_VALID(drro
->drr_type
) ||
2406 !DMU_OT_IS_VALID(drro
->drr_bonustype
) ||
2407 drro
->drr_checksumtype
>= ZIO_CHECKSUM_FUNCTIONS
||
2408 drro
->drr_compress
>= ZIO_COMPRESS_FUNCTIONS
||
2409 P2PHASE(drro
->drr_blksz
, SPA_MINBLOCKSIZE
) ||
2410 drro
->drr_blksz
< SPA_MINBLOCKSIZE
||
2411 drro
->drr_blksz
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)) ||
2412 drro
->drr_bonuslen
>
2413 DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa
->os
))) ||
2414 drro
->drr_dn_slots
>
2415 (spa_maxdnodesize(dmu_objset_spa(rwa
->os
)) >> DNODE_SHIFT
)) {
2416 return (SET_ERROR(EINVAL
));
2419 if (DRR_IS_RAW_ENCRYPTED(drro
->drr_flags
)) {
2420 if (drro
->drr_raw_bonuslen
< drro
->drr_bonuslen
||
2421 drro
->drr_indblkshift
> SPA_MAXBLOCKSHIFT
||
2422 drro
->drr_nlevels
> DN_MAX_LEVELS
||
2423 drro
->drr_nblkptr
> DN_MAX_NBLKPTR
||
2424 DN_SLOTS_TO_BONUSLEN(drro
->drr_dn_slots
) <
2425 drro
->drr_raw_bonuslen
)
2426 return (SET_ERROR(EINVAL
));
2428 if (drro
->drr_flags
!= 0 || drro
->drr_raw_bonuslen
!= 0 ||
2429 drro
->drr_indblkshift
!= 0 || drro
->drr_nlevels
!= 0 ||
2430 drro
->drr_nblkptr
!= 0)
2431 return (SET_ERROR(EINVAL
));
2434 err
= dmu_object_info(rwa
->os
, drro
->drr_object
, &doi
);
2436 if (err
!= 0 && err
!= ENOENT
)
2437 return (SET_ERROR(EINVAL
));
2438 object
= err
== 0 ? drro
->drr_object
: DMU_NEW_OBJECT
;
2441 * If we are losing blkptrs or changing the block size this must
2442 * be a new file instance. We must clear out the previous file
2443 * contents before we can change this type of metadata in the dnode.
2444 * Raw receives will also check that the indirect structure of the
2445 * dnode hasn't changed.
2448 uint32_t indblksz
= drro
->drr_indblkshift
?
2449 1ULL << drro
->drr_indblkshift
: 0;
2450 int nblkptr
= deduce_nblkptr(drro
->drr_bonustype
,
2451 drro
->drr_bonuslen
);
2453 /* nblkptr will be bounded by the bonus size and type */
2454 if (DRR_IS_RAW_ENCRYPTED(drro
->drr_flags
) &&
2455 nblkptr
!= drro
->drr_nblkptr
)
2456 return (SET_ERROR(EINVAL
));
2458 if (drro
->drr_blksz
!= doi
.doi_data_block_size
||
2459 nblkptr
< doi
.doi_nblkptr
||
2460 (DRR_IS_RAW_ENCRYPTED(drro
->drr_flags
) &&
2461 (indblksz
!= doi
.doi_metadata_block_size
||
2462 drro
->drr_nlevels
< doi
.doi_indirection
))) {
2463 err
= dmu_free_long_range(rwa
->os
, drro
->drr_object
,
2466 return (SET_ERROR(EINVAL
));
2470 tx
= dmu_tx_create(rwa
->os
);
2471 dmu_tx_hold_bonus(tx
, object
);
2472 dmu_tx_hold_write(tx
, object
, 0, 0);
2473 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2479 if (object
== DMU_NEW_OBJECT
) {
2480 /* currently free, want to be allocated */
2481 err
= dmu_object_claim_dnsize(rwa
->os
, drro
->drr_object
,
2482 drro
->drr_type
, drro
->drr_blksz
,
2483 drro
->drr_bonustype
, drro
->drr_bonuslen
,
2484 drro
->drr_dn_slots
<< DNODE_SHIFT
, tx
);
2485 } else if (drro
->drr_type
!= doi
.doi_type
||
2486 drro
->drr_blksz
!= doi
.doi_data_block_size
||
2487 drro
->drr_bonustype
!= doi
.doi_bonus_type
||
2488 drro
->drr_bonuslen
!= doi
.doi_bonus_size
) {
2489 /* currently allocated, but with different properties */
2490 err
= dmu_object_reclaim(rwa
->os
, drro
->drr_object
,
2491 drro
->drr_type
, drro
->drr_blksz
,
2492 drro
->drr_bonustype
, drro
->drr_bonuslen
, tx
);
2496 return (SET_ERROR(EINVAL
));
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 */
2505 if (DRR_IS_RAW_ENCRYPTED(drro
->drr_flags
)) {
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
;
2521 if (DRR_IS_RAW_ENCRYPTED(drro
->drr_flags
))
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
&& !DRR_IS_RAW_ENCRYPTED(drro
->drr_flags
)) {
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
);
2618 if (rwa
->byteswap
&& !arc_is_encrypted(abuf
) &&
2619 arc_get_compression(abuf
) == ZIO_COMPRESS_OFF
) {
2620 dmu_object_byteswap_t byteswap
=
2621 DMU_OT_BYTESWAP(drrw
->drr_type
);
2622 dmu_ot_byteswap
[byteswap
].ob_func(abuf
->b_data
,
2623 DRR_WRITE_PAYLOAD_SIZE(drrw
));
2626 /* use the bonus buf to look up the dnode in dmu_assign_arcbuf */
2627 if (dmu_bonus_hold(rwa
->os
, drrw
->drr_object
, FTAG
, &bonus
) != 0)
2628 return (SET_ERROR(EINVAL
));
2629 dmu_assign_arcbuf(bonus
, drrw
->drr_offset
, abuf
, tx
);
2632 * Note: If the receive fails, we want the resume stream to start
2633 * with the same record that we last successfully received (as opposed
2634 * to the next record), so that we can verify that we are
2635 * resuming from the correct location.
2637 save_resume_state(rwa
, drrw
->drr_object
, drrw
->drr_offset
, tx
);
2639 dmu_buf_rele(bonus
, FTAG
);
2645 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
2646 * streams to refer to a copy of the data that is already on the
2647 * system because it came in earlier in the stream. This function
2648 * finds the earlier copy of the data, and uses that copy instead of
2649 * data from the stream to fulfill this write.
2652 receive_write_byref(struct receive_writer_arg
*rwa
,
2653 struct drr_write_byref
*drrwbr
)
2657 guid_map_entry_t gmesrch
;
2658 guid_map_entry_t
*gmep
;
2660 objset_t
*ref_os
= NULL
;
2661 int flags
= DMU_READ_PREFETCH
;
2664 if (drrwbr
->drr_offset
+ drrwbr
->drr_length
< drrwbr
->drr_offset
)
2665 return (SET_ERROR(EINVAL
));
2668 * If the GUID of the referenced dataset is different from the
2669 * GUID of the target dataset, find the referenced dataset.
2671 if (drrwbr
->drr_toguid
!= drrwbr
->drr_refguid
) {
2672 gmesrch
.guid
= drrwbr
->drr_refguid
;
2673 if ((gmep
= avl_find(rwa
->guid_to_ds_map
, &gmesrch
,
2675 return (SET_ERROR(EINVAL
));
2677 if (dmu_objset_from_ds(gmep
->gme_ds
, &ref_os
))
2678 return (SET_ERROR(EINVAL
));
2683 if (DRR_IS_RAW_ENCRYPTED(drrwbr
->drr_flags
)) {
2684 flags
|= DMU_READ_NO_DECRYPT
;
2687 /* may return either a regular db or an encrypted one */
2688 err
= dmu_buf_hold(ref_os
, drrwbr
->drr_refobject
,
2689 drrwbr
->drr_refoffset
, FTAG
, &dbp
, flags
);
2693 tx
= dmu_tx_create(rwa
->os
);
2695 dmu_tx_hold_write(tx
, drrwbr
->drr_object
,
2696 drrwbr
->drr_offset
, drrwbr
->drr_length
);
2697 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2703 if (DRR_IS_RAW_ENCRYPTED(drrwbr
->drr_flags
)) {
2704 dmu_copy_from_buf(rwa
->os
, drrwbr
->drr_object
,
2705 drrwbr
->drr_offset
, dbp
, tx
);
2707 dmu_write(rwa
->os
, drrwbr
->drr_object
,
2708 drrwbr
->drr_offset
, drrwbr
->drr_length
, dbp
->db_data
, tx
);
2710 dmu_buf_rele(dbp
, FTAG
);
2712 /* See comment in restore_write. */
2713 save_resume_state(rwa
, drrwbr
->drr_object
, drrwbr
->drr_offset
, tx
);
2719 receive_write_embedded(struct receive_writer_arg
*rwa
,
2720 struct drr_write_embedded
*drrwe
, void *data
)
2725 if (drrwe
->drr_offset
+ drrwe
->drr_length
< drrwe
->drr_offset
)
2726 return (SET_ERROR(EINVAL
));
2728 if (drrwe
->drr_psize
> BPE_PAYLOAD_SIZE
)
2729 return (SET_ERROR(EINVAL
));
2731 if (drrwe
->drr_etype
>= NUM_BP_EMBEDDED_TYPES
)
2732 return (SET_ERROR(EINVAL
));
2733 if (drrwe
->drr_compression
>= ZIO_COMPRESS_FUNCTIONS
)
2734 return (SET_ERROR(EINVAL
));
2736 tx
= dmu_tx_create(rwa
->os
);
2738 dmu_tx_hold_write(tx
, drrwe
->drr_object
,
2739 drrwe
->drr_offset
, drrwe
->drr_length
);
2740 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2746 dmu_write_embedded(rwa
->os
, drrwe
->drr_object
,
2747 drrwe
->drr_offset
, data
, drrwe
->drr_etype
,
2748 drrwe
->drr_compression
, drrwe
->drr_lsize
, drrwe
->drr_psize
,
2749 rwa
->byteswap
^ ZFS_HOST_BYTEORDER
, tx
);
2751 /* See comment in restore_write. */
2752 save_resume_state(rwa
, drrwe
->drr_object
, drrwe
->drr_offset
, tx
);
2758 receive_spill(struct receive_writer_arg
*rwa
, struct drr_spill
*drrs
,
2762 dmu_buf_t
*db
, *db_spill
;
2765 if (drrs
->drr_length
< SPA_MINBLOCKSIZE
||
2766 drrs
->drr_length
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)))
2767 return (SET_ERROR(EINVAL
));
2769 if (DRR_IS_RAW_ENCRYPTED(drrs
->drr_flags
)) {
2770 if (!DMU_OT_IS_VALID(drrs
->drr_type
) ||
2771 drrs
->drr_compressiontype
>= ZIO_COMPRESS_FUNCTIONS
||
2772 drrs
->drr_compressed_size
== 0)
2773 return (SET_ERROR(EINVAL
));
2776 if (dmu_object_info(rwa
->os
, drrs
->drr_object
, NULL
) != 0)
2777 return (SET_ERROR(EINVAL
));
2779 VERIFY0(dmu_bonus_hold(rwa
->os
, drrs
->drr_object
, FTAG
, &db
));
2780 if ((err
= dmu_spill_hold_by_bonus(db
, FTAG
, &db_spill
)) != 0) {
2781 dmu_buf_rele(db
, FTAG
);
2785 tx
= dmu_tx_create(rwa
->os
);
2787 dmu_tx_hold_spill(tx
, db
->db_object
);
2789 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2791 dmu_buf_rele(db
, FTAG
);
2792 dmu_buf_rele(db_spill
, FTAG
);
2796 dmu_buf_will_dirty(db_spill
, tx
);
2798 if (db_spill
->db_size
< drrs
->drr_length
)
2799 VERIFY(0 == dbuf_spill_set_blksz(db_spill
,
2800 drrs
->drr_length
, tx
));
2801 dmu_assign_arcbuf_impl(db_spill
, abuf
, tx
);
2803 dmu_buf_rele(db
, FTAG
);
2804 dmu_buf_rele(db_spill
, FTAG
);
2812 receive_free(struct receive_writer_arg
*rwa
, struct drr_free
*drrf
)
2816 if (drrf
->drr_length
!= -1ULL &&
2817 drrf
->drr_offset
+ drrf
->drr_length
< drrf
->drr_offset
)
2818 return (SET_ERROR(EINVAL
));
2820 if (dmu_object_info(rwa
->os
, drrf
->drr_object
, NULL
) != 0)
2821 return (SET_ERROR(EINVAL
));
2823 err
= dmu_free_long_range(rwa
->os
, drrf
->drr_object
,
2824 drrf
->drr_offset
, drrf
->drr_length
);
2830 receive_object_range(struct receive_writer_arg
*rwa
,
2831 struct drr_object_range
*drror
)
2835 dnode_t
*mdn
= NULL
;
2836 dmu_buf_t
*db
= NULL
;
2840 * By default, we assume this block is in our native format
2841 * (ZFS_HOST_BYTEORDER). We then take into account whether
2842 * the send stream is byteswapped (rwa->byteswap). Finally,
2843 * we need to byteswap again if this particular block was
2844 * in non-native format on the send side.
2846 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^ rwa
->byteswap
^
2847 !!DRR_IS_RAW_BYTESWAPPED(drror
->drr_flags
);
2850 * Since dnode block sizes are constant, we should not need to worry
2851 * about making sure that the dnode block size is the same on the
2852 * sending and receiving sides for the time being. For non-raw sends,
2853 * this does not matter (and in fact we do not send a DRR_OBJECT_RANGE
2854 * record at all). Raw sends require this record type because the
2855 * encryption parameters are used to protect an entire block of bonus
2856 * buffers. If the size of dnode blocks ever becomes variable,
2857 * handling will need to be added to ensure that dnode block sizes
2858 * match on the sending and receiving side.
2860 if (drror
->drr_numslots
!= DNODES_PER_BLOCK
||
2861 P2PHASE(drror
->drr_firstobj
, DNODES_PER_BLOCK
) != 0 ||
2862 !DRR_IS_RAW_ENCRYPTED(drror
->drr_flags
))
2863 return (SET_ERROR(EINVAL
));
2865 offset
= drror
->drr_firstobj
* sizeof (dnode_phys_t
);
2866 mdn
= DMU_META_DNODE(rwa
->os
);
2868 tx
= dmu_tx_create(rwa
->os
);
2869 ret
= dmu_tx_assign(tx
, TXG_WAIT
);
2875 ret
= dmu_buf_hold_by_dnode(mdn
, offset
, FTAG
, &db
,
2876 DMU_READ_PREFETCH
| DMU_READ_NO_DECRYPT
);
2883 * Convert the buffer associated with this range of dnodes to a
2884 * raw buffer. This ensures that it will be written out as a raw
2885 * buffer when we fill in the dnode objects in future records.
2886 * Since we are commiting this tx now, it is technically possible
2887 * for the dnode block to end up on-disk with the incorrect MAC.
2888 * Despite this, the dataset is marked as inconsistent so no other
2889 * code paths (apart from scrubs) will attempt to read this data.
2890 * Scrubs will not be effected by this either since scrubs only
2891 * read raw data and do not attempt to check the MAC.
2893 dmu_convert_to_raw(db
, byteorder
, drror
->drr_salt
, drror
->drr_iv
,
2894 drror
->drr_mac
, tx
);
2895 dmu_buf_rele(db
, FTAG
);
2900 /* used to destroy the drc_ds on error */
2902 dmu_recv_cleanup_ds(dmu_recv_cookie_t
*drc
)
2904 ds_hold_flags_t dsflags
= (drc
->drc_raw
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
2907 * Wait for the txg sync before cleaning up the receive. For
2908 * resumable receives, this ensures that our resume state has
2909 * been written out to disk. For raw receives, this ensures
2910 * that the user accounting code will not attempt to do anything
2911 * after we stopped receiving the dataset.
2913 txg_wait_synced(drc
->drc_ds
->ds_dir
->dd_pool
, 0);
2915 if (drc
->drc_resumable
) {
2916 dsl_dataset_disown(drc
->drc_ds
, dsflags
, dmu_recv_tag
);
2918 char name
[ZFS_MAX_DATASET_NAME_LEN
];
2919 dsl_dataset_name(drc
->drc_ds
, name
);
2920 dsl_dataset_disown(drc
->drc_ds
, dsflags
, dmu_recv_tag
);
2921 (void) dsl_destroy_head(name
);
2926 receive_cksum(struct receive_arg
*ra
, int len
, void *buf
)
2929 (void) fletcher_4_incremental_byteswap(buf
, len
, &ra
->cksum
);
2931 (void) fletcher_4_incremental_native(buf
, len
, &ra
->cksum
);
2936 * Read the payload into a buffer of size len, and update the current record's
2938 * Allocate ra->next_rrd and read the next record's header into
2939 * ra->next_rrd->header.
2940 * Verify checksum of payload and next record.
2943 receive_read_payload_and_next_header(struct receive_arg
*ra
, int len
, void *buf
)
2946 zio_cksum_t cksum_orig
;
2947 zio_cksum_t
*cksump
;
2950 ASSERT3U(len
, <=, SPA_MAXBLOCKSIZE
);
2951 err
= receive_read(ra
, len
, buf
);
2954 receive_cksum(ra
, len
, buf
);
2956 /* note: rrd is NULL when reading the begin record's payload */
2957 if (ra
->rrd
!= NULL
) {
2958 ra
->rrd
->payload
= buf
;
2959 ra
->rrd
->payload_size
= len
;
2960 ra
->rrd
->bytes_read
= ra
->bytes_read
;
2964 ra
->prev_cksum
= ra
->cksum
;
2966 ra
->next_rrd
= kmem_zalloc(sizeof (*ra
->next_rrd
), KM_SLEEP
);
2967 err
= receive_read(ra
, sizeof (ra
->next_rrd
->header
),
2968 &ra
->next_rrd
->header
);
2969 ra
->next_rrd
->bytes_read
= ra
->bytes_read
;
2972 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
2973 ra
->next_rrd
= NULL
;
2976 if (ra
->next_rrd
->header
.drr_type
== DRR_BEGIN
) {
2977 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
2978 ra
->next_rrd
= NULL
;
2979 return (SET_ERROR(EINVAL
));
2983 * Note: checksum is of everything up to but not including the
2986 ASSERT3U(offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2987 ==, sizeof (dmu_replay_record_t
) - sizeof (zio_cksum_t
));
2989 offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2990 &ra
->next_rrd
->header
);
2992 cksum_orig
= ra
->next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
2993 cksump
= &ra
->next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
2996 byteswap_record(&ra
->next_rrd
->header
);
2998 if ((!ZIO_CHECKSUM_IS_ZERO(cksump
)) &&
2999 !ZIO_CHECKSUM_EQUAL(ra
->cksum
, *cksump
)) {
3000 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
3001 ra
->next_rrd
= NULL
;
3002 return (SET_ERROR(ECKSUM
));
3005 receive_cksum(ra
, sizeof (cksum_orig
), &cksum_orig
);
3011 objlist_create(struct objlist
*list
)
3013 list_create(&list
->list
, sizeof (struct receive_objnode
),
3014 offsetof(struct receive_objnode
, node
));
3015 list
->last_lookup
= 0;
3019 objlist_destroy(struct objlist
*list
)
3021 struct receive_objnode
*n
;
3023 for (n
= list_remove_head(&list
->list
);
3024 n
!= NULL
; n
= list_remove_head(&list
->list
)) {
3025 kmem_free(n
, sizeof (*n
));
3027 list_destroy(&list
->list
);
3031 * This function looks through the objlist to see if the specified object number
3032 * is contained in the objlist. In the process, it will remove all object
3033 * numbers in the list that are smaller than the specified object number. Thus,
3034 * any lookup of an object number smaller than a previously looked up object
3035 * number will always return false; therefore, all lookups should be done in
3039 objlist_exists(struct objlist
*list
, uint64_t object
)
3041 struct receive_objnode
*node
= list_head(&list
->list
);
3042 ASSERT3U(object
, >=, list
->last_lookup
);
3043 list
->last_lookup
= object
;
3044 while (node
!= NULL
&& node
->object
< object
) {
3045 VERIFY3P(node
, ==, list_remove_head(&list
->list
));
3046 kmem_free(node
, sizeof (*node
));
3047 node
= list_head(&list
->list
);
3049 return (node
!= NULL
&& node
->object
== object
);
3053 * The objlist is a list of object numbers stored in ascending order. However,
3054 * the insertion of new object numbers does not seek out the correct location to
3055 * store a new object number; instead, it appends it to the list for simplicity.
3056 * Thus, any users must take care to only insert new object numbers in ascending
3060 objlist_insert(struct objlist
*list
, uint64_t object
)
3062 struct receive_objnode
*node
= kmem_zalloc(sizeof (*node
), KM_SLEEP
);
3063 node
->object
= object
;
3066 struct receive_objnode
*last_object
= list_tail(&list
->list
);
3067 uint64_t last_objnum
= (last_object
!= NULL
? last_object
->object
: 0);
3068 ASSERT3U(node
->object
, >, last_objnum
);
3071 list_insert_tail(&list
->list
, node
);
3075 * Issue the prefetch reads for any necessary indirect blocks.
3077 * We use the object ignore list to tell us whether or not to issue prefetches
3078 * for a given object. We do this for both correctness (in case the blocksize
3079 * of an object has changed) and performance (if the object doesn't exist, don't
3080 * needlessly try to issue prefetches). We also trim the list as we go through
3081 * the stream to prevent it from growing to an unbounded size.
3083 * The object numbers within will always be in sorted order, and any write
3084 * records we see will also be in sorted order, but they're not sorted with
3085 * respect to each other (i.e. we can get several object records before
3086 * receiving each object's write records). As a result, once we've reached a
3087 * given object number, we can safely remove any reference to lower object
3088 * numbers in the ignore list. In practice, we receive up to 32 object records
3089 * before receiving write records, so the list can have up to 32 nodes in it.
3093 receive_read_prefetch(struct receive_arg
*ra
,
3094 uint64_t object
, uint64_t offset
, uint64_t length
)
3096 if (!objlist_exists(&ra
->ignore_objlist
, object
)) {
3097 dmu_prefetch(ra
->os
, object
, 1, offset
, length
,
3098 ZIO_PRIORITY_SYNC_READ
);
3103 * Read records off the stream, issuing any necessary prefetches.
3106 receive_read_record(struct receive_arg
*ra
)
3110 switch (ra
->rrd
->header
.drr_type
) {
3113 struct drr_object
*drro
= &ra
->rrd
->header
.drr_u
.drr_object
;
3114 uint32_t size
= DRR_OBJECT_PAYLOAD_SIZE(drro
);
3115 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
3116 dmu_object_info_t doi
;
3118 err
= receive_read_payload_and_next_header(ra
, size
, buf
);
3120 kmem_free(buf
, size
);
3123 err
= dmu_object_info(ra
->os
, drro
->drr_object
, &doi
);
3125 * See receive_read_prefetch for an explanation why we're
3126 * storing this object in the ignore_obj_list.
3128 if (err
== ENOENT
||
3129 (err
== 0 && doi
.doi_data_block_size
!= drro
->drr_blksz
)) {
3130 objlist_insert(&ra
->ignore_objlist
, drro
->drr_object
);
3135 case DRR_FREEOBJECTS
:
3137 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3142 struct drr_write
*drrw
= &ra
->rrd
->header
.drr_u
.drr_write
;
3144 boolean_t is_meta
= DMU_OT_IS_METADATA(drrw
->drr_type
);
3146 if (DRR_IS_RAW_ENCRYPTED(drrw
->drr_flags
)) {
3147 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^
3148 !!DRR_IS_RAW_BYTESWAPPED(drrw
->drr_flags
) ^
3151 abuf
= arc_loan_raw_buf(dmu_objset_spa(ra
->os
),
3152 drrw
->drr_object
, byteorder
, drrw
->drr_salt
,
3153 drrw
->drr_iv
, drrw
->drr_mac
, drrw
->drr_type
,
3154 drrw
->drr_compressed_size
, drrw
->drr_logical_size
,
3155 drrw
->drr_compressiontype
);
3156 } else if (DRR_WRITE_COMPRESSED(drrw
)) {
3157 ASSERT3U(drrw
->drr_compressed_size
, >, 0);
3158 ASSERT3U(drrw
->drr_logical_size
, >=,
3159 drrw
->drr_compressed_size
);
3161 abuf
= arc_loan_compressed_buf(
3162 dmu_objset_spa(ra
->os
),
3163 drrw
->drr_compressed_size
, drrw
->drr_logical_size
,
3164 drrw
->drr_compressiontype
);
3166 abuf
= arc_loan_buf(dmu_objset_spa(ra
->os
),
3167 is_meta
, drrw
->drr_logical_size
);
3170 err
= receive_read_payload_and_next_header(ra
,
3171 DRR_WRITE_PAYLOAD_SIZE(drrw
), abuf
->b_data
);
3173 dmu_return_arcbuf(abuf
);
3176 ra
->rrd
->arc_buf
= abuf
;
3177 receive_read_prefetch(ra
, drrw
->drr_object
, drrw
->drr_offset
,
3178 drrw
->drr_logical_size
);
3181 case DRR_WRITE_BYREF
:
3183 struct drr_write_byref
*drrwb
=
3184 &ra
->rrd
->header
.drr_u
.drr_write_byref
;
3185 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3186 receive_read_prefetch(ra
, drrwb
->drr_object
, drrwb
->drr_offset
,
3190 case DRR_WRITE_EMBEDDED
:
3192 struct drr_write_embedded
*drrwe
=
3193 &ra
->rrd
->header
.drr_u
.drr_write_embedded
;
3194 uint32_t size
= P2ROUNDUP(drrwe
->drr_psize
, 8);
3195 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
3197 err
= receive_read_payload_and_next_header(ra
, size
, buf
);
3199 kmem_free(buf
, size
);
3203 receive_read_prefetch(ra
, drrwe
->drr_object
, drrwe
->drr_offset
,
3210 * It might be beneficial to prefetch indirect blocks here, but
3211 * we don't really have the data to decide for sure.
3213 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3218 struct drr_end
*drre
= &ra
->rrd
->header
.drr_u
.drr_end
;
3219 if (!ZIO_CHECKSUM_EQUAL(ra
->prev_cksum
, drre
->drr_checksum
))
3220 return (SET_ERROR(ECKSUM
));
3225 struct drr_spill
*drrs
= &ra
->rrd
->header
.drr_u
.drr_spill
;
3227 int len
= DRR_SPILL_PAYLOAD_SIZE(drrs
);
3229 /* DRR_SPILL records are either raw or uncompressed */
3230 if (DRR_IS_RAW_ENCRYPTED(drrs
->drr_flags
)) {
3231 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^
3232 !!DRR_IS_RAW_BYTESWAPPED(drrs
->drr_flags
) ^
3235 abuf
= arc_loan_raw_buf(dmu_objset_spa(ra
->os
),
3236 drrs
->drr_object
, byteorder
, drrs
->drr_salt
,
3237 drrs
->drr_iv
, drrs
->drr_mac
, drrs
->drr_type
,
3238 drrs
->drr_compressed_size
, drrs
->drr_length
,
3239 drrs
->drr_compressiontype
);
3241 abuf
= arc_loan_buf(dmu_objset_spa(ra
->os
),
3242 DMU_OT_IS_METADATA(drrs
->drr_type
),
3246 err
= receive_read_payload_and_next_header(ra
, len
,
3249 dmu_return_arcbuf(abuf
);
3252 ra
->rrd
->arc_buf
= abuf
;
3255 case DRR_OBJECT_RANGE
:
3257 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3261 return (SET_ERROR(EINVAL
));
3266 dprintf_drr(struct receive_record_arg
*rrd
, int err
)
3268 switch (rrd
->header
.drr_type
) {
3271 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
3272 dprintf("drr_type = OBJECT obj = %llu type = %u "
3273 "bonustype = %u blksz = %u bonuslen = %u cksumtype = %u "
3274 "compress = %u dn_slots = %u err = %d\n",
3275 drro
->drr_object
, drro
->drr_type
, drro
->drr_bonustype
,
3276 drro
->drr_blksz
, drro
->drr_bonuslen
,
3277 drro
->drr_checksumtype
, drro
->drr_compress
,
3278 drro
->drr_dn_slots
, err
);
3281 case DRR_FREEOBJECTS
:
3283 struct drr_freeobjects
*drrfo
=
3284 &rrd
->header
.drr_u
.drr_freeobjects
;
3285 dprintf("drr_type = FREEOBJECTS firstobj = %llu "
3286 "numobjs = %llu err = %d\n",
3287 drrfo
->drr_firstobj
, drrfo
->drr_numobjs
, err
);
3292 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
3293 dprintf("drr_type = WRITE obj = %llu type = %u offset = %llu "
3294 "lsize = %llu cksumtype = %u cksumflags = %u "
3295 "compress = %u psize = %llu err = %d\n",
3296 drrw
->drr_object
, drrw
->drr_type
, drrw
->drr_offset
,
3297 drrw
->drr_logical_size
, drrw
->drr_checksumtype
,
3298 drrw
->drr_flags
, drrw
->drr_compressiontype
,
3299 drrw
->drr_compressed_size
, err
);
3302 case DRR_WRITE_BYREF
:
3304 struct drr_write_byref
*drrwbr
=
3305 &rrd
->header
.drr_u
.drr_write_byref
;
3306 dprintf("drr_type = WRITE_BYREF obj = %llu offset = %llu "
3307 "length = %llu toguid = %llx refguid = %llx "
3308 "refobject = %llu refoffset = %llu cksumtype = %u "
3309 "cksumflags = %u err = %d\n",
3310 drrwbr
->drr_object
, drrwbr
->drr_offset
,
3311 drrwbr
->drr_length
, drrwbr
->drr_toguid
,
3312 drrwbr
->drr_refguid
, drrwbr
->drr_refobject
,
3313 drrwbr
->drr_refoffset
, drrwbr
->drr_checksumtype
,
3314 drrwbr
->drr_flags
, err
);
3317 case DRR_WRITE_EMBEDDED
:
3319 struct drr_write_embedded
*drrwe
=
3320 &rrd
->header
.drr_u
.drr_write_embedded
;
3321 dprintf("drr_type = WRITE_EMBEDDED obj = %llu offset = %llu "
3322 "length = %llu compress = %u etype = %u lsize = %u "
3323 "psize = %u err = %d\n",
3324 drrwe
->drr_object
, drrwe
->drr_offset
, drrwe
->drr_length
,
3325 drrwe
->drr_compression
, drrwe
->drr_etype
,
3326 drrwe
->drr_lsize
, drrwe
->drr_psize
, err
);
3331 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
3332 dprintf("drr_type = FREE obj = %llu offset = %llu "
3333 "length = %lld err = %d\n",
3334 drrf
->drr_object
, drrf
->drr_offset
, drrf
->drr_length
,
3340 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
3341 dprintf("drr_type = SPILL obj = %llu length = %llu "
3342 "err = %d\n", drrs
->drr_object
, drrs
->drr_length
, err
);
3351 * Commit the records to the pool.
3354 receive_process_record(struct receive_writer_arg
*rwa
,
3355 struct receive_record_arg
*rrd
)
3359 /* Processing in order, therefore bytes_read should be increasing. */
3360 ASSERT3U(rrd
->bytes_read
, >=, rwa
->bytes_read
);
3361 rwa
->bytes_read
= rrd
->bytes_read
;
3363 switch (rrd
->header
.drr_type
) {
3366 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
3367 err
= receive_object(rwa
, drro
, rrd
->payload
);
3368 kmem_free(rrd
->payload
, rrd
->payload_size
);
3369 rrd
->payload
= NULL
;
3372 case DRR_FREEOBJECTS
:
3374 struct drr_freeobjects
*drrfo
=
3375 &rrd
->header
.drr_u
.drr_freeobjects
;
3376 err
= receive_freeobjects(rwa
, drrfo
);
3381 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
3382 err
= receive_write(rwa
, drrw
, rrd
->arc_buf
);
3383 /* if receive_write() is successful, it consumes the arc_buf */
3385 dmu_return_arcbuf(rrd
->arc_buf
);
3386 rrd
->arc_buf
= NULL
;
3387 rrd
->payload
= NULL
;
3390 case DRR_WRITE_BYREF
:
3392 struct drr_write_byref
*drrwbr
=
3393 &rrd
->header
.drr_u
.drr_write_byref
;
3394 err
= receive_write_byref(rwa
, drrwbr
);
3397 case DRR_WRITE_EMBEDDED
:
3399 struct drr_write_embedded
*drrwe
=
3400 &rrd
->header
.drr_u
.drr_write_embedded
;
3401 err
= receive_write_embedded(rwa
, drrwe
, rrd
->payload
);
3402 kmem_free(rrd
->payload
, rrd
->payload_size
);
3403 rrd
->payload
= NULL
;
3408 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
3409 err
= receive_free(rwa
, drrf
);
3414 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
3415 err
= receive_spill(rwa
, drrs
, rrd
->arc_buf
);
3416 /* if receive_spill() is successful, it consumes the arc_buf */
3418 dmu_return_arcbuf(rrd
->arc_buf
);
3419 rrd
->arc_buf
= NULL
;
3420 rrd
->payload
= NULL
;
3423 case DRR_OBJECT_RANGE
:
3425 struct drr_object_range
*drror
=
3426 &rrd
->header
.drr_u
.drr_object_range
;
3427 return (receive_object_range(rwa
, drror
));
3430 return (SET_ERROR(EINVAL
));
3434 dprintf_drr(rrd
, err
);
3440 * dmu_recv_stream's worker thread; pull records off the queue, and then call
3441 * receive_process_record When we're done, signal the main thread and exit.
3444 receive_writer_thread(void *arg
)
3446 struct receive_writer_arg
*rwa
= arg
;
3447 struct receive_record_arg
*rrd
;
3448 fstrans_cookie_t cookie
= spl_fstrans_mark();
3450 for (rrd
= bqueue_dequeue(&rwa
->q
); !rrd
->eos_marker
;
3451 rrd
= bqueue_dequeue(&rwa
->q
)) {
3453 * If there's an error, the main thread will stop putting things
3454 * on the queue, but we need to clear everything in it before we
3457 if (rwa
->err
== 0) {
3458 rwa
->err
= receive_process_record(rwa
, rrd
);
3459 } else if (rrd
->arc_buf
!= NULL
) {
3460 dmu_return_arcbuf(rrd
->arc_buf
);
3461 rrd
->arc_buf
= NULL
;
3462 rrd
->payload
= NULL
;
3463 } else if (rrd
->payload
!= NULL
) {
3464 kmem_free(rrd
->payload
, rrd
->payload_size
);
3465 rrd
->payload
= NULL
;
3467 kmem_free(rrd
, sizeof (*rrd
));
3469 kmem_free(rrd
, sizeof (*rrd
));
3470 mutex_enter(&rwa
->mutex
);
3472 cv_signal(&rwa
->cv
);
3473 mutex_exit(&rwa
->mutex
);
3474 spl_fstrans_unmark(cookie
);
3479 resume_check(struct receive_arg
*ra
, nvlist_t
*begin_nvl
)
3482 objset_t
*mos
= dmu_objset_pool(ra
->os
)->dp_meta_objset
;
3483 uint64_t dsobj
= dmu_objset_id(ra
->os
);
3484 uint64_t resume_obj
, resume_off
;
3486 if (nvlist_lookup_uint64(begin_nvl
,
3487 "resume_object", &resume_obj
) != 0 ||
3488 nvlist_lookup_uint64(begin_nvl
,
3489 "resume_offset", &resume_off
) != 0) {
3490 return (SET_ERROR(EINVAL
));
3492 VERIFY0(zap_lookup(mos
, dsobj
,
3493 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
));
3494 if (resume_obj
!= val
)
3495 return (SET_ERROR(EINVAL
));
3496 VERIFY0(zap_lookup(mos
, dsobj
,
3497 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
));
3498 if (resume_off
!= val
)
3499 return (SET_ERROR(EINVAL
));
3505 * Read in the stream's records, one by one, and apply them to the pool. There
3506 * are two threads involved; the thread that calls this function will spin up a
3507 * worker thread, read the records off the stream one by one, and issue
3508 * prefetches for any necessary indirect blocks. It will then push the records
3509 * onto an internal blocking queue. The worker thread will pull the records off
3510 * the queue, and actually write the data into the DMU. This way, the worker
3511 * thread doesn't have to wait for reads to complete, since everything it needs
3512 * (the indirect blocks) will be prefetched.
3514 * NB: callers *must* call dmu_recv_end() if this succeeds.
3517 dmu_recv_stream(dmu_recv_cookie_t
*drc
, vnode_t
*vp
, offset_t
*voffp
,
3518 int cleanup_fd
, uint64_t *action_handlep
)
3521 struct receive_arg
*ra
;
3522 struct receive_writer_arg
*rwa
;
3524 uint32_t payloadlen
;
3526 nvlist_t
*begin_nvl
= NULL
;
3528 ra
= kmem_zalloc(sizeof (*ra
), KM_SLEEP
);
3529 rwa
= kmem_zalloc(sizeof (*rwa
), KM_SLEEP
);
3531 ra
->byteswap
= drc
->drc_byteswap
;
3532 ra
->cksum
= drc
->drc_cksum
;
3536 if (dsl_dataset_is_zapified(drc
->drc_ds
)) {
3537 (void) zap_lookup(drc
->drc_ds
->ds_dir
->dd_pool
->dp_meta_objset
,
3538 drc
->drc_ds
->ds_object
, DS_FIELD_RESUME_BYTES
,
3539 sizeof (ra
->bytes_read
), 1, &ra
->bytes_read
);
3542 objlist_create(&ra
->ignore_objlist
);
3544 /* these were verified in dmu_recv_begin */
3545 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc
->drc_drrb
->drr_versioninfo
), ==,
3547 ASSERT3U(drc
->drc_drrb
->drr_type
, <, DMU_OST_NUMTYPES
);
3550 * Open the objset we are modifying.
3552 VERIFY0(dmu_objset_from_ds(drc
->drc_ds
, &ra
->os
));
3554 ASSERT(dsl_dataset_phys(drc
->drc_ds
)->ds_flags
& DS_FLAG_INCONSISTENT
);
3556 featureflags
= DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
);
3557 ra
->featureflags
= featureflags
;
3559 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
3560 if (featureflags
& DMU_BACKUP_FEATURE_DEDUP
) {
3563 if (cleanup_fd
== -1) {
3564 ra
->err
= SET_ERROR(EBADF
);
3567 ra
->err
= zfs_onexit_fd_hold(cleanup_fd
, &minor
);
3573 if (*action_handlep
== 0) {
3574 rwa
->guid_to_ds_map
=
3575 kmem_alloc(sizeof (avl_tree_t
), KM_SLEEP
);
3576 avl_create(rwa
->guid_to_ds_map
, guid_compare
,
3577 sizeof (guid_map_entry_t
),
3578 offsetof(guid_map_entry_t
, avlnode
));
3579 err
= zfs_onexit_add_cb(minor
,
3580 free_guid_map_onexit
, rwa
->guid_to_ds_map
,
3585 err
= zfs_onexit_cb_data(minor
, *action_handlep
,
3586 (void **)&rwa
->guid_to_ds_map
);
3591 drc
->drc_guid_to_ds_map
= rwa
->guid_to_ds_map
;
3594 payloadlen
= drc
->drc_drr_begin
->drr_payloadlen
;
3596 if (payloadlen
!= 0)
3597 payload
= kmem_alloc(payloadlen
, KM_SLEEP
);
3599 err
= receive_read_payload_and_next_header(ra
, payloadlen
, payload
);
3601 if (payloadlen
!= 0)
3602 kmem_free(payload
, payloadlen
);
3605 if (payloadlen
!= 0) {
3606 err
= nvlist_unpack(payload
, payloadlen
, &begin_nvl
, KM_SLEEP
);
3607 kmem_free(payload
, payloadlen
);
3612 /* handle DSL encryption key payload */
3613 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
3614 nvlist_t
*keynvl
= NULL
;
3616 ASSERT(ra
->os
->os_encrypted
);
3617 ASSERT(drc
->drc_raw
);
3619 err
= nvlist_lookup_nvlist(begin_nvl
, "crypt_keydata", &keynvl
);
3623 err
= dsl_crypto_recv_key(spa_name(ra
->os
->os_spa
),
3624 drc
->drc_ds
->ds_object
, drc
->drc_drrb
->drr_type
,
3630 if (featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
3631 err
= resume_check(ra
, begin_nvl
);
3636 (void) bqueue_init(&rwa
->q
, zfs_recv_queue_length
,
3637 offsetof(struct receive_record_arg
, node
));
3638 cv_init(&rwa
->cv
, NULL
, CV_DEFAULT
, NULL
);
3639 mutex_init(&rwa
->mutex
, NULL
, MUTEX_DEFAULT
, NULL
);
3641 rwa
->byteswap
= drc
->drc_byteswap
;
3642 rwa
->resumable
= drc
->drc_resumable
;
3644 (void) thread_create(NULL
, 0, receive_writer_thread
, rwa
, 0, curproc
,
3645 TS_RUN
, minclsyspri
);
3647 * We're reading rwa->err without locks, which is safe since we are the
3648 * only reader, and the worker thread is the only writer. It's ok if we
3649 * miss a write for an iteration or two of the loop, since the writer
3650 * thread will keep freeing records we send it until we send it an eos
3653 * We can leave this loop in 3 ways: First, if rwa->err is
3654 * non-zero. In that case, the writer thread will free the rrd we just
3655 * pushed. Second, if we're interrupted; in that case, either it's the
3656 * first loop and ra->rrd was never allocated, or it's later and ra->rrd
3657 * has been handed off to the writer thread who will free it. Finally,
3658 * if receive_read_record fails or we're at the end of the stream, then
3659 * we free ra->rrd and exit.
3661 while (rwa
->err
== 0) {
3662 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
3663 err
= SET_ERROR(EINTR
);
3667 ASSERT3P(ra
->rrd
, ==, NULL
);
3668 ra
->rrd
= ra
->next_rrd
;
3669 ra
->next_rrd
= NULL
;
3670 /* Allocates and loads header into ra->next_rrd */
3671 err
= receive_read_record(ra
);
3673 if (ra
->rrd
->header
.drr_type
== DRR_END
|| err
!= 0) {
3674 kmem_free(ra
->rrd
, sizeof (*ra
->rrd
));
3679 bqueue_enqueue(&rwa
->q
, ra
->rrd
,
3680 sizeof (struct receive_record_arg
) + ra
->rrd
->payload_size
);
3683 if (ra
->next_rrd
== NULL
)
3684 ra
->next_rrd
= kmem_zalloc(sizeof (*ra
->next_rrd
), KM_SLEEP
);
3685 ra
->next_rrd
->eos_marker
= B_TRUE
;
3686 bqueue_enqueue(&rwa
->q
, ra
->next_rrd
, 1);
3688 mutex_enter(&rwa
->mutex
);
3689 while (!rwa
->done
) {
3690 cv_wait(&rwa
->cv
, &rwa
->mutex
);
3692 mutex_exit(&rwa
->mutex
);
3694 cv_destroy(&rwa
->cv
);
3695 mutex_destroy(&rwa
->mutex
);
3696 bqueue_destroy(&rwa
->q
);
3701 nvlist_free(begin_nvl
);
3702 if ((featureflags
& DMU_BACKUP_FEATURE_DEDUP
) && (cleanup_fd
!= -1))
3703 zfs_onexit_fd_rele(cleanup_fd
);
3707 * Clean up references. If receive is not resumable,
3708 * destroy what we created, so we don't leave it in
3709 * the inconsistent state.
3711 dmu_recv_cleanup_ds(drc
);
3715 objlist_destroy(&ra
->ignore_objlist
);
3716 kmem_free(ra
, sizeof (*ra
));
3717 kmem_free(rwa
, sizeof (*rwa
));
3722 dmu_recv_end_check(void *arg
, dmu_tx_t
*tx
)
3724 dmu_recv_cookie_t
*drc
= arg
;
3725 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3728 ASSERT3P(drc
->drc_ds
->ds_owner
, ==, dmu_recv_tag
);
3730 if (!drc
->drc_newfs
) {
3731 dsl_dataset_t
*origin_head
;
3733 error
= dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
, &origin_head
);
3736 if (drc
->drc_force
) {
3738 * We will destroy any snapshots in tofs (i.e. before
3739 * origin_head) that are after the origin (which is
3740 * the snap before drc_ds, because drc_ds can not
3741 * have any snaps of its own).
3745 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3747 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3748 dsl_dataset_t
*snap
;
3749 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3753 if (snap
->ds_dir
!= origin_head
->ds_dir
)
3754 error
= SET_ERROR(EINVAL
);
3756 error
= dsl_destroy_snapshot_check_impl(
3759 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3760 dsl_dataset_rele(snap
, FTAG
);
3765 dsl_dataset_rele(origin_head
, FTAG
);
3769 error
= dsl_dataset_clone_swap_check_impl(drc
->drc_ds
,
3770 origin_head
, drc
->drc_force
, drc
->drc_owner
, tx
);
3772 dsl_dataset_rele(origin_head
, FTAG
);
3775 error
= dsl_dataset_snapshot_check_impl(origin_head
,
3776 drc
->drc_tosnap
, tx
, B_TRUE
, 1, drc
->drc_cred
);
3777 dsl_dataset_rele(origin_head
, FTAG
);
3781 error
= dsl_destroy_head_check_impl(drc
->drc_ds
, 1);
3783 error
= dsl_dataset_snapshot_check_impl(drc
->drc_ds
,
3784 drc
->drc_tosnap
, tx
, B_TRUE
, 1, drc
->drc_cred
);
3790 dmu_recv_end_sync(void *arg
, dmu_tx_t
*tx
)
3792 dmu_recv_cookie_t
*drc
= arg
;
3793 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3794 boolean_t encrypted
= drc
->drc_ds
->ds_dir
->dd_crypto_obj
!= 0;
3796 spa_history_log_internal_ds(drc
->drc_ds
, "finish receiving",
3797 tx
, "snap=%s", drc
->drc_tosnap
);
3799 if (!drc
->drc_newfs
) {
3800 dsl_dataset_t
*origin_head
;
3802 VERIFY0(dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
,
3805 if (drc
->drc_force
) {
3807 * Destroy any snapshots of drc_tofs (origin_head)
3808 * after the origin (the snap before drc_ds).
3812 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3814 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3815 dsl_dataset_t
*snap
;
3816 VERIFY0(dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3818 ASSERT3P(snap
->ds_dir
, ==, origin_head
->ds_dir
);
3819 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3820 dsl_destroy_snapshot_sync_impl(snap
,
3822 dsl_dataset_rele(snap
, FTAG
);
3825 VERIFY3P(drc
->drc_ds
->ds_prev
, ==,
3826 origin_head
->ds_prev
);
3828 dsl_dataset_clone_swap_sync_impl(drc
->drc_ds
,
3830 dsl_dataset_snapshot_sync_impl(origin_head
,
3831 drc
->drc_tosnap
, tx
);
3833 /* set snapshot's creation time and guid */
3834 dmu_buf_will_dirty(origin_head
->ds_prev
->ds_dbuf
, tx
);
3835 dsl_dataset_phys(origin_head
->ds_prev
)->ds_creation_time
=
3836 drc
->drc_drrb
->drr_creation_time
;
3837 dsl_dataset_phys(origin_head
->ds_prev
)->ds_guid
=
3838 drc
->drc_drrb
->drr_toguid
;
3839 dsl_dataset_phys(origin_head
->ds_prev
)->ds_flags
&=
3840 ~DS_FLAG_INCONSISTENT
;
3842 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
3843 dsl_dataset_phys(origin_head
)->ds_flags
&=
3844 ~DS_FLAG_INCONSISTENT
;
3846 drc
->drc_newsnapobj
=
3847 dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3849 dsl_dataset_rele(origin_head
, FTAG
);
3850 dsl_destroy_head_sync_impl(drc
->drc_ds
, tx
);
3852 if (drc
->drc_owner
!= NULL
)
3853 VERIFY3P(origin_head
->ds_owner
, ==, drc
->drc_owner
);
3855 dsl_dataset_t
*ds
= drc
->drc_ds
;
3857 dsl_dataset_snapshot_sync_impl(ds
, drc
->drc_tosnap
, tx
);
3859 /* set snapshot's creation time and guid */
3860 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
3861 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_time
=
3862 drc
->drc_drrb
->drr_creation_time
;
3863 dsl_dataset_phys(ds
->ds_prev
)->ds_guid
=
3864 drc
->drc_drrb
->drr_toguid
;
3865 dsl_dataset_phys(ds
->ds_prev
)->ds_flags
&=
3866 ~DS_FLAG_INCONSISTENT
;
3868 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3869 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
3870 if (dsl_dataset_has_resume_receive_state(ds
)) {
3871 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3872 DS_FIELD_RESUME_FROMGUID
, tx
);
3873 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3874 DS_FIELD_RESUME_OBJECT
, tx
);
3875 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3876 DS_FIELD_RESUME_OFFSET
, tx
);
3877 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3878 DS_FIELD_RESUME_BYTES
, tx
);
3879 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3880 DS_FIELD_RESUME_TOGUID
, tx
);
3881 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3882 DS_FIELD_RESUME_TONAME
, tx
);
3884 drc
->drc_newsnapobj
=
3885 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
;
3887 zvol_create_minors(dp
->dp_spa
, drc
->drc_tofs
, B_TRUE
);
3890 * Release the hold from dmu_recv_begin. This must be done before
3891 * we return to open context, so that when we free the dataset's dnode
3892 * we can evict its bonus buffer. Since the dataset may be destroyed
3893 * at this point (and therefore won't have a valid pointer to the spa)
3894 * we release the key mapping manually here while we do have a valid
3895 * pointer, if it exists.
3897 if (!drc
->drc_raw
&& encrypted
) {
3898 (void) spa_keystore_remove_mapping(dmu_tx_pool(tx
)->dp_spa
,
3899 drc
->drc_ds
->ds_object
, drc
->drc_ds
);
3901 dsl_dataset_disown(drc
->drc_ds
, 0, dmu_recv_tag
);
3906 add_ds_to_guidmap(const char *name
, avl_tree_t
*guid_map
, uint64_t snapobj
,
3910 dsl_dataset_t
*snapds
;
3911 guid_map_entry_t
*gmep
;
3912 ds_hold_flags_t dsflags
= (raw
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
3915 ASSERT(guid_map
!= NULL
);
3917 err
= dsl_pool_hold(name
, FTAG
, &dp
);
3920 gmep
= kmem_alloc(sizeof (*gmep
), KM_SLEEP
);
3921 err
= dsl_dataset_hold_obj_flags(dp
, snapobj
, dsflags
, gmep
, &snapds
);
3923 gmep
->guid
= dsl_dataset_phys(snapds
)->ds_guid
;
3925 gmep
->gme_ds
= snapds
;
3926 avl_add(guid_map
, gmep
);
3927 dsl_dataset_long_hold(snapds
, gmep
);
3929 kmem_free(gmep
, sizeof (*gmep
));
3932 dsl_pool_rele(dp
, FTAG
);
3936 static int dmu_recv_end_modified_blocks
= 3;
3939 dmu_recv_existing_end(dmu_recv_cookie_t
*drc
)
3943 * We will be destroying the ds; make sure its origin is unmounted if
3946 char name
[ZFS_MAX_DATASET_NAME_LEN
];
3947 dsl_dataset_name(drc
->drc_ds
, name
);
3948 zfs_destroy_unmount_origin(name
);
3951 return (dsl_sync_task(drc
->drc_tofs
,
3952 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3953 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
3957 dmu_recv_new_end(dmu_recv_cookie_t
*drc
)
3959 return (dsl_sync_task(drc
->drc_tofs
,
3960 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3961 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
3965 dmu_recv_end(dmu_recv_cookie_t
*drc
, void *owner
)
3969 drc
->drc_owner
= owner
;
3972 error
= dmu_recv_new_end(drc
);
3974 error
= dmu_recv_existing_end(drc
);
3977 dmu_recv_cleanup_ds(drc
);
3978 } else if (drc
->drc_guid_to_ds_map
!= NULL
) {
3979 (void) add_ds_to_guidmap(drc
->drc_tofs
, drc
->drc_guid_to_ds_map
,
3980 drc
->drc_newsnapobj
, drc
->drc_raw
);
3986 * Return TRUE if this objset is currently being received into.
3989 dmu_objset_is_receiving(objset_t
*os
)
3991 return (os
->os_dsl_dataset
!= NULL
&&
3992 os
->os_dsl_dataset
->ds_owner
== dmu_recv_tag
);
3995 #if defined(_KERNEL)
3996 module_param(zfs_send_corrupt_data
, int, 0644);
3997 MODULE_PARM_DESC(zfs_send_corrupt_data
, "Allow sending corrupt data");