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
));
227 * If there is a pending op, but it's not PENDING_FREE, push it out,
228 * since free block aggregation can only be done for blocks of the
229 * same type (i.e., DRR_FREE records can only be aggregated with
230 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
231 * aggregated with other DRR_FREEOBJECTS records.
233 if (dsp
->dsa_pending_op
!= PENDING_NONE
&&
234 dsp
->dsa_pending_op
!= PENDING_FREE
) {
235 if (dump_record(dsp
, NULL
, 0) != 0)
236 return (SET_ERROR(EINTR
));
237 dsp
->dsa_pending_op
= PENDING_NONE
;
240 if (dsp
->dsa_pending_op
== PENDING_FREE
) {
242 * There should never be a PENDING_FREE if length is
243 * DMU_OBJECT_END (because dump_dnode is the only place where
244 * this function is called with a DMU_OBJECT_END, and only after
245 * flushing any pending record).
247 ASSERT(length
!= DMU_OBJECT_END
);
249 * Check to see whether this free block can be aggregated
252 if (drrf
->drr_object
== object
&& drrf
->drr_offset
+
253 drrf
->drr_length
== offset
) {
254 if (offset
+ length
< offset
)
255 drrf
->drr_length
= DMU_OBJECT_END
;
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 if (offset
+ length
< offset
)
272 drrf
->drr_length
= DMU_OBJECT_END
;
274 drrf
->drr_length
= length
;
275 drrf
->drr_toguid
= dsp
->dsa_toguid
;
276 if (length
== DMU_OBJECT_END
) {
277 if (dump_record(dsp
, NULL
, 0) != 0)
278 return (SET_ERROR(EINTR
));
280 dsp
->dsa_pending_op
= PENDING_FREE
;
287 dump_write(dmu_sendarg_t
*dsp
, dmu_object_type_t type
, uint64_t object
,
288 uint64_t offset
, int lsize
, int psize
, const blkptr_t
*bp
, void *data
)
290 uint64_t payload_size
;
291 boolean_t raw
= (dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
);
292 struct drr_write
*drrw
= &(dsp
->dsa_drr
->drr_u
.drr_write
);
295 * We send data in increasing object, offset order.
296 * See comment in dump_free() for details.
298 ASSERT(object
> dsp
->dsa_last_data_object
||
299 (object
== dsp
->dsa_last_data_object
&&
300 offset
> dsp
->dsa_last_data_offset
));
301 dsp
->dsa_last_data_object
= object
;
302 dsp
->dsa_last_data_offset
= offset
+ lsize
- 1;
305 * If there is any kind of pending aggregation (currently either
306 * a grouping of free objects or free blocks), push it out to
307 * the stream, since aggregation can't be done across operations
308 * of different types.
310 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
311 if (dump_record(dsp
, NULL
, 0) != 0)
312 return (SET_ERROR(EINTR
));
313 dsp
->dsa_pending_op
= PENDING_NONE
;
315 /* write a WRITE record */
316 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
317 dsp
->dsa_drr
->drr_type
= DRR_WRITE
;
318 drrw
->drr_object
= object
;
319 drrw
->drr_type
= type
;
320 drrw
->drr_offset
= offset
;
321 drrw
->drr_toguid
= dsp
->dsa_toguid
;
322 drrw
->drr_logical_size
= lsize
;
324 /* only set the compression fields if the buf is compressed or raw */
325 if (raw
|| lsize
!= psize
) {
326 ASSERT(!BP_IS_EMBEDDED(bp
));
327 ASSERT3S(psize
, >, 0);
330 ASSERT(BP_IS_PROTECTED(bp
));
333 * This is a raw protected block so we need to pass
334 * along everything the receiving side will need to
335 * interpret this block, including the byteswap, salt,
338 if (BP_SHOULD_BYTESWAP(bp
))
339 drrw
->drr_flags
|= DRR_RAW_BYTESWAP
;
340 zio_crypt_decode_params_bp(bp
, drrw
->drr_salt
,
342 zio_crypt_decode_mac_bp(bp
, drrw
->drr_mac
);
344 /* this is a compressed block */
345 ASSERT(dsp
->dsa_featureflags
&
346 DMU_BACKUP_FEATURE_COMPRESSED
);
347 ASSERT(!BP_SHOULD_BYTESWAP(bp
));
348 ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)));
349 ASSERT3U(BP_GET_COMPRESS(bp
), !=, ZIO_COMPRESS_OFF
);
350 ASSERT3S(lsize
, >=, psize
);
353 /* set fields common to compressed and raw sends */
354 drrw
->drr_compressiontype
= BP_GET_COMPRESS(bp
);
355 drrw
->drr_compressed_size
= psize
;
356 payload_size
= drrw
->drr_compressed_size
;
358 payload_size
= drrw
->drr_logical_size
;
361 if (bp
== NULL
|| BP_IS_EMBEDDED(bp
) || (BP_IS_PROTECTED(bp
) && !raw
)) {
363 * There's no pre-computed checksum for partial-block writes,
364 * embedded BP's, or encrypted BP's that are being sent as
365 * plaintext, so (like fletcher4-checkummed blocks) userland
366 * will have to compute a dedup-capable checksum itself.
368 drrw
->drr_checksumtype
= ZIO_CHECKSUM_OFF
;
370 drrw
->drr_checksumtype
= BP_GET_CHECKSUM(bp
);
371 if (zio_checksum_table
[drrw
->drr_checksumtype
].ci_flags
&
372 ZCHECKSUM_FLAG_DEDUP
)
373 drrw
->drr_flags
|= DRR_CHECKSUM_DEDUP
;
374 DDK_SET_LSIZE(&drrw
->drr_key
, BP_GET_LSIZE(bp
));
375 DDK_SET_PSIZE(&drrw
->drr_key
, BP_GET_PSIZE(bp
));
376 DDK_SET_COMPRESS(&drrw
->drr_key
, BP_GET_COMPRESS(bp
));
377 DDK_SET_CRYPT(&drrw
->drr_key
, BP_IS_PROTECTED(bp
));
378 drrw
->drr_key
.ddk_cksum
= bp
->blk_cksum
;
381 if (dump_record(dsp
, data
, payload_size
) != 0)
382 return (SET_ERROR(EINTR
));
387 dump_write_embedded(dmu_sendarg_t
*dsp
, uint64_t object
, uint64_t offset
,
388 int blksz
, const blkptr_t
*bp
)
390 char buf
[BPE_PAYLOAD_SIZE
];
391 struct drr_write_embedded
*drrw
=
392 &(dsp
->dsa_drr
->drr_u
.drr_write_embedded
);
394 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
395 if (dump_record(dsp
, NULL
, 0) != 0)
396 return (SET_ERROR(EINTR
));
397 dsp
->dsa_pending_op
= PENDING_NONE
;
400 ASSERT(BP_IS_EMBEDDED(bp
));
402 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
403 dsp
->dsa_drr
->drr_type
= DRR_WRITE_EMBEDDED
;
404 drrw
->drr_object
= object
;
405 drrw
->drr_offset
= offset
;
406 drrw
->drr_length
= blksz
;
407 drrw
->drr_toguid
= dsp
->dsa_toguid
;
408 drrw
->drr_compression
= BP_GET_COMPRESS(bp
);
409 drrw
->drr_etype
= BPE_GET_ETYPE(bp
);
410 drrw
->drr_lsize
= BPE_GET_LSIZE(bp
);
411 drrw
->drr_psize
= BPE_GET_PSIZE(bp
);
413 decode_embedded_bp_compressed(bp
, buf
);
415 if (dump_record(dsp
, buf
, P2ROUNDUP(drrw
->drr_psize
, 8)) != 0)
416 return (SET_ERROR(EINTR
));
421 dump_spill(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
, uint64_t object
, void *data
)
423 struct drr_spill
*drrs
= &(dsp
->dsa_drr
->drr_u
.drr_spill
);
424 uint64_t blksz
= BP_GET_LSIZE(bp
);
426 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
427 if (dump_record(dsp
, NULL
, 0) != 0)
428 return (SET_ERROR(EINTR
));
429 dsp
->dsa_pending_op
= PENDING_NONE
;
432 /* write a SPILL record */
433 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
434 dsp
->dsa_drr
->drr_type
= DRR_SPILL
;
435 drrs
->drr_object
= object
;
436 drrs
->drr_length
= blksz
;
437 drrs
->drr_toguid
= dsp
->dsa_toguid
;
439 /* handle raw send fields */
440 if (dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
441 ASSERT(BP_IS_PROTECTED(bp
));
443 if (BP_SHOULD_BYTESWAP(bp
))
444 drrs
->drr_flags
|= DRR_RAW_BYTESWAP
;
445 drrs
->drr_compressiontype
= BP_GET_COMPRESS(bp
);
446 drrs
->drr_compressed_size
= BP_GET_PSIZE(bp
);
447 zio_crypt_decode_params_bp(bp
, drrs
->drr_salt
, drrs
->drr_iv
);
448 zio_crypt_decode_mac_bp(bp
, drrs
->drr_mac
);
451 if (dump_record(dsp
, data
, blksz
) != 0)
452 return (SET_ERROR(EINTR
));
457 dump_freeobjects(dmu_sendarg_t
*dsp
, uint64_t firstobj
, uint64_t numobjs
)
459 struct drr_freeobjects
*drrfo
= &(dsp
->dsa_drr
->drr_u
.drr_freeobjects
);
460 uint64_t maxobj
= DNODES_PER_BLOCK
*
461 (DMU_META_DNODE(dsp
->dsa_os
)->dn_maxblkid
+ 1);
464 * ZoL < 0.7 does not handle large FREEOBJECTS records correctly,
465 * leading to zfs recv never completing. to avoid this issue, don't
466 * send FREEOBJECTS records for object IDs which cannot exist on the
470 if (maxobj
< firstobj
)
473 if (maxobj
< firstobj
+ numobjs
)
474 numobjs
= maxobj
- firstobj
;
478 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
479 * push it out, since free block aggregation can only be done for
480 * blocks of the same type (i.e., DRR_FREE records can only be
481 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
482 * can only be aggregated with other DRR_FREEOBJECTS records.
484 if (dsp
->dsa_pending_op
!= PENDING_NONE
&&
485 dsp
->dsa_pending_op
!= PENDING_FREEOBJECTS
) {
486 if (dump_record(dsp
, NULL
, 0) != 0)
487 return (SET_ERROR(EINTR
));
488 dsp
->dsa_pending_op
= PENDING_NONE
;
490 if (dsp
->dsa_pending_op
== PENDING_FREEOBJECTS
) {
492 * See whether this free object array can be aggregated
495 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
== firstobj
) {
496 drrfo
->drr_numobjs
+= numobjs
;
499 /* can't be aggregated. Push out pending record */
500 if (dump_record(dsp
, NULL
, 0) != 0)
501 return (SET_ERROR(EINTR
));
502 dsp
->dsa_pending_op
= PENDING_NONE
;
506 /* write a FREEOBJECTS record */
507 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
508 dsp
->dsa_drr
->drr_type
= DRR_FREEOBJECTS
;
509 drrfo
->drr_firstobj
= firstobj
;
510 drrfo
->drr_numobjs
= numobjs
;
511 drrfo
->drr_toguid
= dsp
->dsa_toguid
;
513 dsp
->dsa_pending_op
= PENDING_FREEOBJECTS
;
519 dump_dnode(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
, uint64_t object
,
522 struct drr_object
*drro
= &(dsp
->dsa_drr
->drr_u
.drr_object
);
525 if (object
< dsp
->dsa_resume_object
) {
527 * Note: when resuming, we will visit all the dnodes in
528 * the block of dnodes that we are resuming from. In
529 * this case it's unnecessary to send the dnodes prior to
530 * the one we are resuming from. We should be at most one
531 * block's worth of dnodes behind the resume point.
533 ASSERT3U(dsp
->dsa_resume_object
- object
, <,
534 1 << (DNODE_BLOCK_SHIFT
- DNODE_SHIFT
));
538 if (dnp
== NULL
|| dnp
->dn_type
== DMU_OT_NONE
)
539 return (dump_freeobjects(dsp
, object
, 1));
541 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
542 if (dump_record(dsp
, NULL
, 0) != 0)
543 return (SET_ERROR(EINTR
));
544 dsp
->dsa_pending_op
= PENDING_NONE
;
547 /* write an OBJECT record */
548 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
549 dsp
->dsa_drr
->drr_type
= DRR_OBJECT
;
550 drro
->drr_object
= object
;
551 drro
->drr_type
= dnp
->dn_type
;
552 drro
->drr_bonustype
= dnp
->dn_bonustype
;
553 drro
->drr_blksz
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
554 drro
->drr_bonuslen
= dnp
->dn_bonuslen
;
555 drro
->drr_dn_slots
= dnp
->dn_extra_slots
+ 1;
556 drro
->drr_checksumtype
= dnp
->dn_checksum
;
557 drro
->drr_compress
= dnp
->dn_compress
;
558 drro
->drr_toguid
= dsp
->dsa_toguid
;
560 if (!(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
561 drro
->drr_blksz
> SPA_OLD_MAXBLOCKSIZE
)
562 drro
->drr_blksz
= SPA_OLD_MAXBLOCKSIZE
;
564 bonuslen
= P2ROUNDUP(dnp
->dn_bonuslen
, 8);
566 if ((dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
)) {
567 ASSERT(BP_IS_ENCRYPTED(bp
));
569 if (BP_SHOULD_BYTESWAP(bp
))
570 drro
->drr_flags
|= DRR_RAW_BYTESWAP
;
572 /* needed for reconstructing dnp on recv side */
573 drro
->drr_indblkshift
= dnp
->dn_indblkshift
;
574 drro
->drr_nlevels
= dnp
->dn_nlevels
;
575 drro
->drr_nblkptr
= dnp
->dn_nblkptr
;
578 * Since we encrypt the entire bonus area, the (raw) part
579 * beyond the bonuslen is actually nonzero, so we need
583 drro
->drr_raw_bonuslen
= DN_MAX_BONUS_LEN(dnp
);
584 bonuslen
= drro
->drr_raw_bonuslen
;
588 if (dump_record(dsp
, DN_BONUS(dnp
), bonuslen
) != 0)
589 return (SET_ERROR(EINTR
));
591 /* Free anything past the end of the file. */
592 if (dump_free(dsp
, object
, (dnp
->dn_maxblkid
+ 1) *
593 (dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
), DMU_OBJECT_END
) != 0)
594 return (SET_ERROR(EINTR
));
595 if (dsp
->dsa_err
!= 0)
596 return (SET_ERROR(EINTR
));
601 dump_object_range(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
, uint64_t firstobj
,
604 struct drr_object_range
*drror
=
605 &(dsp
->dsa_drr
->drr_u
.drr_object_range
);
607 /* we only use this record type for raw sends */
608 ASSERT(BP_IS_PROTECTED(bp
));
609 ASSERT(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
);
610 ASSERT3U(BP_GET_COMPRESS(bp
), ==, ZIO_COMPRESS_OFF
);
611 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_DNODE
);
612 ASSERT0(BP_GET_LEVEL(bp
));
614 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
615 if (dump_record(dsp
, NULL
, 0) != 0)
616 return (SET_ERROR(EINTR
));
617 dsp
->dsa_pending_op
= PENDING_NONE
;
620 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
621 dsp
->dsa_drr
->drr_type
= DRR_OBJECT_RANGE
;
622 drror
->drr_firstobj
= firstobj
;
623 drror
->drr_numslots
= numslots
;
624 drror
->drr_toguid
= dsp
->dsa_toguid
;
625 if (BP_SHOULD_BYTESWAP(bp
))
626 drror
->drr_flags
|= DRR_RAW_BYTESWAP
;
627 zio_crypt_decode_params_bp(bp
, drror
->drr_salt
, drror
->drr_iv
);
628 zio_crypt_decode_mac_bp(bp
, drror
->drr_mac
);
630 if (dump_record(dsp
, NULL
, 0) != 0)
631 return (SET_ERROR(EINTR
));
636 backup_do_embed(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
)
638 if (!BP_IS_EMBEDDED(bp
))
642 * Compression function must be legacy, or explicitly enabled.
644 if ((BP_GET_COMPRESS(bp
) >= ZIO_COMPRESS_LEGACY_FUNCTIONS
&&
645 !(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_LZ4
)))
649 * Embed type must be explicitly enabled.
651 switch (BPE_GET_ETYPE(bp
)) {
652 case BP_EMBEDDED_TYPE_DATA
:
653 if (dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)
663 * This is the callback function to traverse_dataset that acts as the worker
664 * thread for dmu_send_impl.
668 send_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
669 const zbookmark_phys_t
*zb
, const struct dnode_phys
*dnp
, void *arg
)
671 struct send_thread_arg
*sta
= arg
;
672 struct send_block_record
*record
;
673 uint64_t record_size
;
676 ASSERT(zb
->zb_object
== DMU_META_DNODE_OBJECT
||
677 zb
->zb_object
>= sta
->resume
.zb_object
);
678 ASSERT3P(sta
->ds
, !=, NULL
);
681 return (SET_ERROR(EINTR
));
684 ASSERT3U(zb
->zb_level
, ==, ZB_DNODE_LEVEL
);
686 } else if (zb
->zb_level
< 0) {
690 record
= kmem_zalloc(sizeof (struct send_block_record
), KM_SLEEP
);
691 record
->eos_marker
= B_FALSE
;
694 record
->indblkshift
= dnp
->dn_indblkshift
;
695 record
->datablkszsec
= dnp
->dn_datablkszsec
;
696 record_size
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
697 bqueue_enqueue(&sta
->q
, record
, record_size
);
703 * This function kicks off the traverse_dataset. It also handles setting the
704 * error code of the thread in case something goes wrong, and pushes the End of
705 * Stream record when the traverse_dataset call has finished. If there is no
706 * dataset to traverse, the thread immediately pushes End of Stream marker.
709 send_traverse_thread(void *arg
)
711 struct send_thread_arg
*st_arg
= arg
;
713 struct send_block_record
*data
;
714 fstrans_cookie_t cookie
= spl_fstrans_mark();
716 if (st_arg
->ds
!= NULL
) {
717 err
= traverse_dataset_resume(st_arg
->ds
,
718 st_arg
->fromtxg
, &st_arg
->resume
,
719 st_arg
->flags
, send_cb
, st_arg
);
722 st_arg
->error_code
= err
;
724 data
= kmem_zalloc(sizeof (*data
), KM_SLEEP
);
725 data
->eos_marker
= B_TRUE
;
726 bqueue_enqueue(&st_arg
->q
, data
, 1);
727 spl_fstrans_unmark(cookie
);
732 * This function actually handles figuring out what kind of record needs to be
733 * dumped, reading the data (which has hopefully been prefetched), and calling
734 * the appropriate helper function.
737 do_dump(dmu_sendarg_t
*dsa
, struct send_block_record
*data
)
739 dsl_dataset_t
*ds
= dmu_objset_ds(dsa
->dsa_os
);
740 const blkptr_t
*bp
= &data
->bp
;
741 const zbookmark_phys_t
*zb
= &data
->zb
;
742 uint8_t indblkshift
= data
->indblkshift
;
743 uint16_t dblkszsec
= data
->datablkszsec
;
744 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
745 dmu_object_type_t type
= bp
? BP_GET_TYPE(bp
) : DMU_OT_NONE
;
748 ASSERT3U(zb
->zb_level
, >=, 0);
750 ASSERT(zb
->zb_object
== DMU_META_DNODE_OBJECT
||
751 zb
->zb_object
>= dsa
->dsa_resume_object
);
754 * All bps of an encrypted os should have the encryption bit set.
755 * If this is not true it indicates tampering and we report an error.
757 if (dsa
->dsa_os
->os_encrypted
&&
758 !BP_IS_HOLE(bp
) && !BP_USES_CRYPT(bp
)) {
759 spa_log_error(spa
, zb
);
760 zfs_panic_recover("unencrypted block in encrypted "
761 "object set %llu", ds
->ds_object
);
762 return (SET_ERROR(EIO
));
765 if (zb
->zb_object
!= DMU_META_DNODE_OBJECT
&&
766 DMU_OBJECT_IS_SPECIAL(zb
->zb_object
)) {
768 } else if (BP_IS_HOLE(bp
) &&
769 zb
->zb_object
== DMU_META_DNODE_OBJECT
) {
770 uint64_t span
= BP_SPAN(dblkszsec
, indblkshift
, zb
->zb_level
);
771 uint64_t dnobj
= (zb
->zb_blkid
* span
) >> DNODE_SHIFT
;
772 err
= dump_freeobjects(dsa
, dnobj
, span
>> DNODE_SHIFT
);
773 } else if (BP_IS_HOLE(bp
)) {
774 uint64_t span
= BP_SPAN(dblkszsec
, indblkshift
, zb
->zb_level
);
775 uint64_t offset
= zb
->zb_blkid
* span
;
776 /* Don't dump free records for offsets > DMU_OBJECT_END */
777 if (zb
->zb_blkid
== 0 || span
<= DMU_OBJECT_END
/ zb
->zb_blkid
)
778 err
= dump_free(dsa
, zb
->zb_object
, offset
, span
);
779 } else if (zb
->zb_level
> 0 || type
== DMU_OT_OBJSET
) {
781 } else if (type
== DMU_OT_DNODE
) {
782 int epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
783 arc_flags_t aflags
= ARC_FLAG_WAIT
;
785 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
787 if (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
788 ASSERT(BP_IS_ENCRYPTED(bp
));
789 ASSERT3U(BP_GET_COMPRESS(bp
), ==, ZIO_COMPRESS_OFF
);
790 zioflags
|= ZIO_FLAG_RAW
;
793 ASSERT0(zb
->zb_level
);
795 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
796 ZIO_PRIORITY_ASYNC_READ
, zioflags
, &aflags
, zb
) != 0)
797 return (SET_ERROR(EIO
));
799 dnode_phys_t
*blk
= abuf
->b_data
;
800 uint64_t dnobj
= zb
->zb_blkid
* epb
;
803 * Raw sends require sending encryption parameters for the
804 * block of dnodes. Regular sends do not need to send this
807 if (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
808 ASSERT(arc_is_encrypted(abuf
));
809 err
= dump_object_range(dsa
, bp
, dnobj
, epb
);
813 for (int i
= 0; i
< epb
;
814 i
+= blk
[i
].dn_extra_slots
+ 1) {
815 err
= dump_dnode(dsa
, bp
, dnobj
+ i
, blk
+ i
);
820 arc_buf_destroy(abuf
, &abuf
);
821 } else if (type
== DMU_OT_SA
) {
822 arc_flags_t aflags
= ARC_FLAG_WAIT
;
824 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
826 if (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) {
827 ASSERT(BP_IS_PROTECTED(bp
));
828 zioflags
|= ZIO_FLAG_RAW
;
831 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
832 ZIO_PRIORITY_ASYNC_READ
, zioflags
, &aflags
, zb
) != 0)
833 return (SET_ERROR(EIO
));
835 err
= dump_spill(dsa
, bp
, zb
->zb_object
, abuf
->b_data
);
836 arc_buf_destroy(abuf
, &abuf
);
837 } else if (backup_do_embed(dsa
, bp
)) {
838 /* it's an embedded level-0 block of a regular object */
839 int blksz
= dblkszsec
<< SPA_MINBLOCKSHIFT
;
840 ASSERT0(zb
->zb_level
);
841 err
= dump_write_embedded(dsa
, zb
->zb_object
,
842 zb
->zb_blkid
* blksz
, blksz
, bp
);
844 /* it's a level-0 block of a regular object */
845 arc_flags_t aflags
= ARC_FLAG_WAIT
;
847 int blksz
= dblkszsec
<< SPA_MINBLOCKSHIFT
;
851 * If we have large blocks stored on disk but the send flags
852 * don't allow us to send large blocks, we split the data from
853 * the arc buf into chunks.
855 boolean_t split_large_blocks
= blksz
> SPA_OLD_MAXBLOCKSIZE
&&
856 !(dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
);
859 * Raw sends require that we always get raw data as it exists
860 * on disk, so we assert that we are not splitting blocks here.
862 boolean_t request_raw
=
863 (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0;
866 * We should only request compressed data from the ARC if all
867 * the following are true:
868 * - stream compression was requested
869 * - we aren't splitting large blocks into smaller chunks
870 * - the data won't need to be byteswapped before sending
871 * - this isn't an embedded block
872 * - this isn't metadata (if receiving on a different endian
873 * system it can be byteswapped more easily)
875 boolean_t request_compressed
=
876 (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
) &&
877 !split_large_blocks
&& !BP_SHOULD_BYTESWAP(bp
) &&
878 !BP_IS_EMBEDDED(bp
) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp
));
880 IMPLY(request_raw
, !split_large_blocks
);
881 IMPLY(request_raw
, BP_IS_PROTECTED(bp
));
882 ASSERT0(zb
->zb_level
);
883 ASSERT(zb
->zb_object
> dsa
->dsa_resume_object
||
884 (zb
->zb_object
== dsa
->dsa_resume_object
&&
885 zb
->zb_blkid
* blksz
>= dsa
->dsa_resume_offset
));
887 ASSERT3U(blksz
, ==, BP_GET_LSIZE(bp
));
889 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
891 zioflags
|= ZIO_FLAG_RAW
;
892 else if (request_compressed
)
893 zioflags
|= ZIO_FLAG_RAW_COMPRESS
;
895 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
896 ZIO_PRIORITY_ASYNC_READ
, zioflags
, &aflags
, zb
) != 0) {
897 if (zfs_send_corrupt_data
) {
898 /* Send a block filled with 0x"zfs badd bloc" */
899 abuf
= arc_alloc_buf(spa
, &abuf
, ARC_BUFC_DATA
,
902 for (ptr
= abuf
->b_data
;
903 (char *)ptr
< (char *)abuf
->b_data
+ blksz
;
905 *ptr
= 0x2f5baddb10cULL
;
907 return (SET_ERROR(EIO
));
911 offset
= zb
->zb_blkid
* blksz
;
913 if (split_large_blocks
) {
914 ASSERT0(arc_is_encrypted(abuf
));
915 ASSERT3U(arc_get_compression(abuf
), ==,
917 char *buf
= abuf
->b_data
;
918 while (blksz
> 0 && err
== 0) {
919 int n
= MIN(blksz
, SPA_OLD_MAXBLOCKSIZE
);
920 err
= dump_write(dsa
, type
, zb
->zb_object
,
921 offset
, n
, n
, NULL
, buf
);
927 err
= dump_write(dsa
, type
, zb
->zb_object
, offset
,
928 blksz
, arc_buf_size(abuf
), bp
, abuf
->b_data
);
930 arc_buf_destroy(abuf
, &abuf
);
933 ASSERT(err
== 0 || err
== EINTR
);
938 * Pop the new data off the queue, and free the old data.
940 static struct send_block_record
*
941 get_next_record(bqueue_t
*bq
, struct send_block_record
*data
)
943 struct send_block_record
*tmp
= bqueue_dequeue(bq
);
944 kmem_free(data
, sizeof (*data
));
949 * Actually do the bulk of the work in a zfs send.
951 * Note: Releases dp using the specified tag.
954 dmu_send_impl(void *tag
, dsl_pool_t
*dp
, dsl_dataset_t
*to_ds
,
955 zfs_bookmark_phys_t
*ancestor_zb
, boolean_t is_clone
,
956 boolean_t embedok
, boolean_t large_block_ok
, boolean_t compressok
,
957 boolean_t rawok
, int outfd
, uint64_t resumeobj
, uint64_t resumeoff
,
958 vnode_t
*vp
, offset_t
*off
)
961 dmu_replay_record_t
*drr
;
964 uint64_t fromtxg
= 0;
965 uint64_t featureflags
= 0;
966 struct send_thread_arg to_arg
;
967 void *payload
= NULL
;
968 size_t payload_len
= 0;
969 struct send_block_record
*to_data
;
971 err
= dmu_objset_from_ds(to_ds
, &os
);
973 dsl_pool_rele(dp
, tag
);
978 * If this is a non-raw send of an encrypted ds, we can ensure that
979 * the objset_phys_t is authenticated. This is safe because this is
980 * either a snapshot or we have owned the dataset, ensuring that
981 * it can't be modified.
983 if (!rawok
&& os
->os_encrypted
&&
984 arc_is_unauthenticated(os
->os_phys_buf
)) {
985 err
= arc_untransform(os
->os_phys_buf
, os
->os_spa
,
986 to_ds
->ds_object
, B_FALSE
);
988 dsl_pool_rele(dp
, tag
);
992 ASSERT0(arc_is_unauthenticated(os
->os_phys_buf
));
995 drr
= kmem_zalloc(sizeof (dmu_replay_record_t
), KM_SLEEP
);
996 drr
->drr_type
= DRR_BEGIN
;
997 drr
->drr_u
.drr_begin
.drr_magic
= DMU_BACKUP_MAGIC
;
998 DMU_SET_STREAM_HDRTYPE(drr
->drr_u
.drr_begin
.drr_versioninfo
,
1001 bzero(&to_arg
, sizeof (to_arg
));
1004 if (dmu_objset_type(os
) == DMU_OST_ZFS
) {
1006 if (zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &version
) != 0) {
1007 kmem_free(drr
, sizeof (dmu_replay_record_t
));
1008 dsl_pool_rele(dp
, tag
);
1009 return (SET_ERROR(EINVAL
));
1011 if (version
>= ZPL_VERSION_SA
) {
1012 featureflags
|= DMU_BACKUP_FEATURE_SA_SPILL
;
1017 /* raw sends imply large_block_ok */
1018 if ((large_block_ok
|| rawok
) &&
1019 to_ds
->ds_feature_inuse
[SPA_FEATURE_LARGE_BLOCKS
])
1020 featureflags
|= DMU_BACKUP_FEATURE_LARGE_BLOCKS
;
1021 if (to_ds
->ds_feature_inuse
[SPA_FEATURE_LARGE_DNODE
])
1022 featureflags
|= DMU_BACKUP_FEATURE_LARGE_DNODE
;
1024 /* encrypted datasets will not have embedded blocks */
1025 if ((embedok
|| rawok
) && !os
->os_encrypted
&&
1026 spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
)) {
1027 featureflags
|= DMU_BACKUP_FEATURE_EMBED_DATA
;
1030 /* raw send implies compressok */
1031 if (compressok
|| rawok
)
1032 featureflags
|= DMU_BACKUP_FEATURE_COMPRESSED
;
1033 if (rawok
&& os
->os_encrypted
)
1034 featureflags
|= DMU_BACKUP_FEATURE_RAW
;
1037 (DMU_BACKUP_FEATURE_EMBED_DATA
| DMU_BACKUP_FEATURE_COMPRESSED
|
1038 DMU_BACKUP_FEATURE_RAW
)) != 0 &&
1039 spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
)) {
1040 featureflags
|= DMU_BACKUP_FEATURE_LZ4
;
1043 if (resumeobj
!= 0 || resumeoff
!= 0) {
1044 featureflags
|= DMU_BACKUP_FEATURE_RESUMING
;
1047 DMU_SET_FEATUREFLAGS(drr
->drr_u
.drr_begin
.drr_versioninfo
,
1050 drr
->drr_u
.drr_begin
.drr_creation_time
=
1051 dsl_dataset_phys(to_ds
)->ds_creation_time
;
1052 drr
->drr_u
.drr_begin
.drr_type
= dmu_objset_type(os
);
1054 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CLONE
;
1055 drr
->drr_u
.drr_begin
.drr_toguid
= dsl_dataset_phys(to_ds
)->ds_guid
;
1056 if (dsl_dataset_phys(to_ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
1057 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CI_DATA
;
1058 if (zfs_send_set_freerecords_bit
)
1059 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_FREERECORDS
;
1061 if (ancestor_zb
!= NULL
) {
1062 drr
->drr_u
.drr_begin
.drr_fromguid
=
1063 ancestor_zb
->zbm_guid
;
1064 fromtxg
= ancestor_zb
->zbm_creation_txg
;
1066 dsl_dataset_name(to_ds
, drr
->drr_u
.drr_begin
.drr_toname
);
1067 if (!to_ds
->ds_is_snapshot
) {
1068 (void) strlcat(drr
->drr_u
.drr_begin
.drr_toname
, "@--head--",
1069 sizeof (drr
->drr_u
.drr_begin
.drr_toname
));
1072 dsp
= kmem_zalloc(sizeof (dmu_sendarg_t
), KM_SLEEP
);
1076 dsp
->dsa_outfd
= outfd
;
1077 dsp
->dsa_proc
= curproc
;
1080 dsp
->dsa_toguid
= dsl_dataset_phys(to_ds
)->ds_guid
;
1081 dsp
->dsa_pending_op
= PENDING_NONE
;
1082 dsp
->dsa_featureflags
= featureflags
;
1083 dsp
->dsa_resume_object
= resumeobj
;
1084 dsp
->dsa_resume_offset
= resumeoff
;
1086 mutex_enter(&to_ds
->ds_sendstream_lock
);
1087 list_insert_head(&to_ds
->ds_sendstreams
, dsp
);
1088 mutex_exit(&to_ds
->ds_sendstream_lock
);
1090 dsl_dataset_long_hold(to_ds
, FTAG
);
1091 dsl_pool_rele(dp
, tag
);
1093 /* handle features that require a DRR_BEGIN payload */
1095 (DMU_BACKUP_FEATURE_RESUMING
| DMU_BACKUP_FEATURE_RAW
)) {
1096 nvlist_t
*keynvl
= NULL
;
1097 nvlist_t
*nvl
= fnvlist_alloc();
1099 if (featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
1100 dmu_object_info_t to_doi
;
1101 err
= dmu_object_info(os
, resumeobj
, &to_doi
);
1107 SET_BOOKMARK(&to_arg
.resume
, to_ds
->ds_object
,
1109 resumeoff
/ to_doi
.doi_data_block_size
);
1111 fnvlist_add_uint64(nvl
, "resume_object", resumeobj
);
1112 fnvlist_add_uint64(nvl
, "resume_offset", resumeoff
);
1115 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1116 ASSERT(os
->os_encrypted
);
1118 err
= dsl_crypto_populate_key_nvlist(to_ds
, &keynvl
);
1124 fnvlist_add_nvlist(nvl
, "crypt_keydata", keynvl
);
1127 payload
= fnvlist_pack(nvl
, &payload_len
);
1128 drr
->drr_payloadlen
= payload_len
;
1129 fnvlist_free(keynvl
);
1133 err
= dump_record(dsp
, payload
, payload_len
);
1134 fnvlist_pack_free(payload
, payload_len
);
1140 err
= bqueue_init(&to_arg
.q
, zfs_send_queue_length
,
1141 offsetof(struct send_block_record
, ln
));
1142 to_arg
.error_code
= 0;
1143 to_arg
.cancel
= B_FALSE
;
1145 to_arg
.fromtxg
= fromtxg
;
1146 to_arg
.flags
= TRAVERSE_PRE
| TRAVERSE_PREFETCH
;
1148 to_arg
.flags
|= TRAVERSE_NO_DECRYPT
;
1149 (void) thread_create(NULL
, 0, send_traverse_thread
, &to_arg
, 0, curproc
,
1150 TS_RUN
, minclsyspri
);
1152 to_data
= bqueue_dequeue(&to_arg
.q
);
1154 while (!to_data
->eos_marker
&& err
== 0) {
1155 err
= do_dump(dsp
, to_data
);
1156 to_data
= get_next_record(&to_arg
.q
, to_data
);
1157 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1162 to_arg
.cancel
= B_TRUE
;
1163 while (!to_data
->eos_marker
) {
1164 to_data
= get_next_record(&to_arg
.q
, to_data
);
1167 kmem_free(to_data
, sizeof (*to_data
));
1169 bqueue_destroy(&to_arg
.q
);
1171 if (err
== 0 && to_arg
.error_code
!= 0)
1172 err
= to_arg
.error_code
;
1177 if (dsp
->dsa_pending_op
!= PENDING_NONE
)
1178 if (dump_record(dsp
, NULL
, 0) != 0)
1179 err
= SET_ERROR(EINTR
);
1182 if (err
== EINTR
&& dsp
->dsa_err
!= 0)
1187 bzero(drr
, sizeof (dmu_replay_record_t
));
1188 drr
->drr_type
= DRR_END
;
1189 drr
->drr_u
.drr_end
.drr_checksum
= dsp
->dsa_zc
;
1190 drr
->drr_u
.drr_end
.drr_toguid
= dsp
->dsa_toguid
;
1192 if (dump_record(dsp
, NULL
, 0) != 0)
1195 mutex_enter(&to_ds
->ds_sendstream_lock
);
1196 list_remove(&to_ds
->ds_sendstreams
, dsp
);
1197 mutex_exit(&to_ds
->ds_sendstream_lock
);
1199 VERIFY(err
!= 0 || (dsp
->dsa_sent_begin
&& dsp
->dsa_sent_end
));
1201 kmem_free(drr
, sizeof (dmu_replay_record_t
));
1202 kmem_free(dsp
, sizeof (dmu_sendarg_t
));
1204 dsl_dataset_long_rele(to_ds
, FTAG
);
1210 dmu_send_obj(const char *pool
, uint64_t tosnap
, uint64_t fromsnap
,
1211 boolean_t embedok
, boolean_t large_block_ok
, boolean_t compressok
,
1212 boolean_t rawok
, int outfd
, vnode_t
*vp
, offset_t
*off
)
1216 dsl_dataset_t
*fromds
= NULL
;
1217 ds_hold_flags_t dsflags
= (rawok
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
1220 err
= dsl_pool_hold(pool
, FTAG
, &dp
);
1224 err
= dsl_dataset_hold_obj_flags(dp
, tosnap
, dsflags
, FTAG
, &ds
);
1226 dsl_pool_rele(dp
, FTAG
);
1230 if (fromsnap
!= 0) {
1231 zfs_bookmark_phys_t zb
;
1234 err
= dsl_dataset_hold_obj(dp
, fromsnap
, FTAG
, &fromds
);
1236 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1237 dsl_pool_rele(dp
, FTAG
);
1240 if (!dsl_dataset_is_before(ds
, fromds
, 0))
1241 err
= SET_ERROR(EXDEV
);
1242 zb
.zbm_creation_time
=
1243 dsl_dataset_phys(fromds
)->ds_creation_time
;
1244 zb
.zbm_creation_txg
= dsl_dataset_phys(fromds
)->ds_creation_txg
;
1245 zb
.zbm_guid
= dsl_dataset_phys(fromds
)->ds_guid
;
1246 is_clone
= (fromds
->ds_dir
!= ds
->ds_dir
);
1247 dsl_dataset_rele(fromds
, FTAG
);
1248 err
= dmu_send_impl(FTAG
, dp
, ds
, &zb
, is_clone
,
1249 embedok
, large_block_ok
, compressok
, rawok
, outfd
,
1252 err
= dmu_send_impl(FTAG
, dp
, ds
, NULL
, B_FALSE
,
1253 embedok
, large_block_ok
, compressok
, rawok
, outfd
,
1256 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1261 dmu_send(const char *tosnap
, const char *fromsnap
, boolean_t embedok
,
1262 boolean_t large_block_ok
, boolean_t compressok
, boolean_t rawok
,
1263 int outfd
, uint64_t resumeobj
, uint64_t resumeoff
, vnode_t
*vp
,
1269 ds_hold_flags_t dsflags
= (rawok
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
1270 boolean_t owned
= B_FALSE
;
1272 if (fromsnap
!= NULL
&& strpbrk(fromsnap
, "@#") == NULL
)
1273 return (SET_ERROR(EINVAL
));
1275 err
= dsl_pool_hold(tosnap
, FTAG
, &dp
);
1279 if (strchr(tosnap
, '@') == NULL
&& spa_writeable(dp
->dp_spa
)) {
1281 * We are sending a filesystem or volume. Ensure
1282 * that it doesn't change by owning the dataset.
1284 err
= dsl_dataset_own(dp
, tosnap
, dsflags
, FTAG
, &ds
);
1287 err
= dsl_dataset_hold_flags(dp
, tosnap
, dsflags
, FTAG
, &ds
);
1290 dsl_pool_rele(dp
, FTAG
);
1294 if (fromsnap
!= NULL
) {
1295 zfs_bookmark_phys_t zb
;
1296 boolean_t is_clone
= B_FALSE
;
1297 int fsnamelen
= strchr(tosnap
, '@') - tosnap
;
1300 * If the fromsnap is in a different filesystem, then
1301 * mark the send stream as a clone.
1303 if (strncmp(tosnap
, fromsnap
, fsnamelen
) != 0 ||
1304 (fromsnap
[fsnamelen
] != '@' &&
1305 fromsnap
[fsnamelen
] != '#')) {
1309 if (strchr(fromsnap
, '@')) {
1310 dsl_dataset_t
*fromds
;
1311 err
= dsl_dataset_hold(dp
, fromsnap
, FTAG
, &fromds
);
1313 if (!dsl_dataset_is_before(ds
, fromds
, 0))
1314 err
= SET_ERROR(EXDEV
);
1315 zb
.zbm_creation_time
=
1316 dsl_dataset_phys(fromds
)->ds_creation_time
;
1317 zb
.zbm_creation_txg
=
1318 dsl_dataset_phys(fromds
)->ds_creation_txg
;
1319 zb
.zbm_guid
= dsl_dataset_phys(fromds
)->ds_guid
;
1320 is_clone
= (ds
->ds_dir
!= fromds
->ds_dir
);
1321 dsl_dataset_rele(fromds
, FTAG
);
1324 err
= dsl_bookmark_lookup(dp
, fromsnap
, ds
, &zb
);
1328 dsl_dataset_disown(ds
, dsflags
, FTAG
);
1330 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1332 dsl_pool_rele(dp
, FTAG
);
1335 err
= dmu_send_impl(FTAG
, dp
, ds
, &zb
, is_clone
,
1336 embedok
, large_block_ok
, compressok
, rawok
,
1337 outfd
, resumeobj
, resumeoff
, vp
, off
);
1339 err
= dmu_send_impl(FTAG
, dp
, ds
, NULL
, B_FALSE
,
1340 embedok
, large_block_ok
, compressok
, rawok
,
1341 outfd
, resumeobj
, resumeoff
, vp
, off
);
1344 dsl_dataset_disown(ds
, dsflags
, FTAG
);
1346 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1352 dmu_adjust_send_estimate_for_indirects(dsl_dataset_t
*ds
, uint64_t uncompressed
,
1353 uint64_t compressed
, boolean_t stream_compressed
, uint64_t *sizep
)
1358 * Assume that space (both on-disk and in-stream) is dominated by
1359 * data. We will adjust for indirect blocks and the copies property,
1360 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
1363 uint64_t recordsize
;
1364 uint64_t record_count
;
1366 VERIFY0(dmu_objset_from_ds(ds
, &os
));
1368 /* Assume all (uncompressed) blocks are recordsize. */
1369 if (os
->os_phys
->os_type
== DMU_OST_ZVOL
) {
1370 err
= dsl_prop_get_int_ds(ds
,
1371 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
), &recordsize
);
1373 err
= dsl_prop_get_int_ds(ds
,
1374 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
), &recordsize
);
1378 record_count
= uncompressed
/ recordsize
;
1381 * If we're estimating a send size for a compressed stream, use the
1382 * compressed data size to estimate the stream size. Otherwise, use the
1383 * uncompressed data size.
1385 size
= stream_compressed
? compressed
: uncompressed
;
1388 * Subtract out approximate space used by indirect blocks.
1389 * Assume most space is used by data blocks (non-indirect, non-dnode).
1390 * Assume no ditto blocks or internal fragmentation.
1392 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
1395 size
-= record_count
* sizeof (blkptr_t
);
1397 /* Add in the space for the record associated with each block. */
1398 size
+= record_count
* sizeof (dmu_replay_record_t
);
1406 dmu_send_estimate(dsl_dataset_t
*ds
, dsl_dataset_t
*fromds
,
1407 boolean_t stream_compressed
, uint64_t *sizep
)
1410 uint64_t uncomp
, comp
;
1412 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1414 /* tosnap must be a snapshot */
1415 if (!ds
->ds_is_snapshot
)
1416 return (SET_ERROR(EINVAL
));
1418 /* fromsnap, if provided, must be a snapshot */
1419 if (fromds
!= NULL
&& !fromds
->ds_is_snapshot
)
1420 return (SET_ERROR(EINVAL
));
1423 * fromsnap must be an earlier snapshot from the same fs as tosnap,
1424 * or the origin's fs.
1426 if (fromds
!= NULL
&& !dsl_dataset_is_before(ds
, fromds
, 0))
1427 return (SET_ERROR(EXDEV
));
1429 /* Get compressed and uncompressed size estimates of changed data. */
1430 if (fromds
== NULL
) {
1431 uncomp
= dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1432 comp
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1435 err
= dsl_dataset_space_written(fromds
, ds
,
1436 &used
, &comp
, &uncomp
);
1441 err
= dmu_adjust_send_estimate_for_indirects(ds
, uncomp
, comp
,
1442 stream_compressed
, sizep
);
1444 * Add the size of the BEGIN and END records to the estimate.
1446 *sizep
+= 2 * sizeof (dmu_replay_record_t
);
1450 struct calculate_send_arg
{
1451 uint64_t uncompressed
;
1452 uint64_t compressed
;
1456 * Simple callback used to traverse the blocks of a snapshot and sum their
1457 * uncompressed and compressed sizes.
1461 dmu_calculate_send_traversal(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1462 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1464 struct calculate_send_arg
*space
= arg
;
1465 if (bp
!= NULL
&& !BP_IS_HOLE(bp
)) {
1466 space
->uncompressed
+= BP_GET_UCSIZE(bp
);
1467 space
->compressed
+= BP_GET_PSIZE(bp
);
1473 * Given a desination snapshot and a TXG, calculate the approximate size of a
1474 * send stream sent from that TXG. from_txg may be zero, indicating that the
1475 * whole snapshot will be sent.
1478 dmu_send_estimate_from_txg(dsl_dataset_t
*ds
, uint64_t from_txg
,
1479 boolean_t stream_compressed
, uint64_t *sizep
)
1482 struct calculate_send_arg size
= { 0 };
1484 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1486 /* tosnap must be a snapshot */
1487 if (!dsl_dataset_is_snapshot(ds
))
1488 return (SET_ERROR(EINVAL
));
1490 /* verify that from_txg is before the provided snapshot was taken */
1491 if (from_txg
>= dsl_dataset_phys(ds
)->ds_creation_txg
) {
1492 return (SET_ERROR(EXDEV
));
1495 * traverse the blocks of the snapshot with birth times after
1496 * from_txg, summing their uncompressed size
1498 err
= traverse_dataset(ds
, from_txg
,
1499 TRAVERSE_POST
| TRAVERSE_NO_DECRYPT
,
1500 dmu_calculate_send_traversal
, &size
);
1505 err
= dmu_adjust_send_estimate_for_indirects(ds
, size
.uncompressed
,
1506 size
.compressed
, stream_compressed
, sizep
);
1510 typedef struct dmu_recv_begin_arg
{
1511 const char *drba_origin
;
1512 dmu_recv_cookie_t
*drba_cookie
;
1514 uint64_t drba_snapobj
;
1515 } dmu_recv_begin_arg_t
;
1518 recv_begin_check_existing_impl(dmu_recv_begin_arg_t
*drba
, dsl_dataset_t
*ds
,
1523 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1525 /* temporary clone name must not exist */
1526 error
= zap_lookup(dp
->dp_meta_objset
,
1527 dsl_dir_phys(ds
->ds_dir
)->dd_child_dir_zapobj
, recv_clone_name
,
1529 if (error
!= ENOENT
)
1530 return (error
== 0 ? EBUSY
: error
);
1532 /* new snapshot name must not exist */
1533 error
= zap_lookup(dp
->dp_meta_objset
,
1534 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1535 drba
->drba_cookie
->drc_tosnap
, 8, 1, &val
);
1536 if (error
!= ENOENT
)
1537 return (error
== 0 ? EEXIST
: error
);
1540 * Check snapshot limit before receiving. We'll recheck again at the
1541 * end, but might as well abort before receiving if we're already over
1544 * Note that we do not check the file system limit with
1545 * dsl_dir_fscount_check because the temporary %clones don't count
1546 * against that limit.
1548 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1, ZFS_PROP_SNAPSHOT_LIMIT
,
1549 NULL
, drba
->drba_cred
);
1553 if (fromguid
!= 0) {
1554 dsl_dataset_t
*snap
;
1555 uint64_t obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1557 /* Find snapshot in this dir that matches fromguid. */
1559 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
1562 return (SET_ERROR(ENODEV
));
1563 if (snap
->ds_dir
!= ds
->ds_dir
) {
1564 dsl_dataset_rele(snap
, FTAG
);
1565 return (SET_ERROR(ENODEV
));
1567 if (dsl_dataset_phys(snap
)->ds_guid
== fromguid
)
1569 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
1570 dsl_dataset_rele(snap
, FTAG
);
1573 return (SET_ERROR(ENODEV
));
1575 if (drba
->drba_cookie
->drc_force
) {
1576 drba
->drba_snapobj
= obj
;
1579 * If we are not forcing, there must be no
1580 * changes since fromsnap.
1582 if (dsl_dataset_modified_since_snap(ds
, snap
)) {
1583 dsl_dataset_rele(snap
, FTAG
);
1584 return (SET_ERROR(ETXTBSY
));
1586 drba
->drba_snapobj
= ds
->ds_prev
->ds_object
;
1589 dsl_dataset_rele(snap
, FTAG
);
1591 /* if full, then must be forced */
1592 if (!drba
->drba_cookie
->drc_force
)
1593 return (SET_ERROR(EEXIST
));
1596 * We don't support using zfs recv -F to blow away
1597 * encrypted filesystems. This would require the
1598 * dsl dir to point to the old encryption key and
1599 * the new one at the same time during the receive.
1601 if (ds
->ds_dir
->dd_crypto_obj
!= 0)
1602 return (SET_ERROR(EINVAL
));
1604 drba
->drba_snapobj
= 0;
1612 dmu_recv_begin_check(void *arg
, dmu_tx_t
*tx
)
1614 dmu_recv_begin_arg_t
*drba
= arg
;
1615 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1616 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1617 uint64_t fromguid
= drrb
->drr_fromguid
;
1618 int flags
= drrb
->drr_flags
;
1619 ds_hold_flags_t dsflags
= 0;
1621 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1623 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1625 /* already checked */
1626 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1627 ASSERT(!(featureflags
& DMU_BACKUP_FEATURE_RESUMING
));
1629 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1630 DMU_COMPOUNDSTREAM
||
1631 drrb
->drr_type
>= DMU_OST_NUMTYPES
||
1632 ((flags
& DRR_FLAG_CLONE
) && drba
->drba_origin
== NULL
))
1633 return (SET_ERROR(EINVAL
));
1635 /* Verify pool version supports SA if SA_SPILL feature set */
1636 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
1637 spa_version(dp
->dp_spa
) < SPA_VERSION_SA
)
1638 return (SET_ERROR(ENOTSUP
));
1640 if (drba
->drba_cookie
->drc_resumable
&&
1641 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EXTENSIBLE_DATASET
))
1642 return (SET_ERROR(ENOTSUP
));
1645 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1646 * record to a plain WRITE record, so the pool must have the
1647 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1648 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1650 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
1651 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
))
1652 return (SET_ERROR(ENOTSUP
));
1653 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
1654 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
))
1655 return (SET_ERROR(ENOTSUP
));
1658 * The receiving code doesn't know how to translate large blocks
1659 * to smaller ones, so the pool must have the LARGE_BLOCKS
1660 * feature enabled if the stream has LARGE_BLOCKS. Same with
1663 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
1664 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_BLOCKS
))
1665 return (SET_ERROR(ENOTSUP
));
1666 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_DNODE
) &&
1667 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_DNODE
))
1668 return (SET_ERROR(ENOTSUP
));
1670 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
)) {
1671 /* raw receives require the encryption feature */
1672 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_ENCRYPTION
))
1673 return (SET_ERROR(ENOTSUP
));
1675 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1678 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1680 /* target fs already exists; recv into temp clone */
1682 /* Can't recv a clone into an existing fs */
1683 if (flags
& DRR_FLAG_CLONE
|| drba
->drba_origin
) {
1684 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1685 return (SET_ERROR(EINVAL
));
1688 error
= recv_begin_check_existing_impl(drba
, ds
, fromguid
);
1689 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1690 } else if (error
== ENOENT
) {
1691 /* target fs does not exist; must be a full backup or clone */
1692 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1695 * If it's a non-clone incremental, we are missing the
1696 * target fs, so fail the recv.
1698 if (fromguid
!= 0 && !(flags
& DRR_FLAG_CLONE
||
1700 return (SET_ERROR(ENOENT
));
1703 * If we're receiving a full send as a clone, and it doesn't
1704 * contain all the necessary free records and freeobject
1705 * records, reject it.
1707 if (fromguid
== 0 && drba
->drba_origin
&&
1708 !(flags
& DRR_FLAG_FREERECORDS
))
1709 return (SET_ERROR(EINVAL
));
1711 /* Open the parent of tofs */
1712 ASSERT3U(strlen(tofs
), <, sizeof (buf
));
1713 (void) strlcpy(buf
, tofs
, strrchr(tofs
, '/') - tofs
+ 1);
1714 error
= dsl_dataset_hold_flags(dp
, buf
, dsflags
, FTAG
, &ds
);
1719 * Check filesystem and snapshot limits before receiving. We'll
1720 * recheck snapshot limits again at the end (we create the
1721 * filesystems and increment those counts during begin_sync).
1723 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
1724 ZFS_PROP_FILESYSTEM_LIMIT
, NULL
, drba
->drba_cred
);
1726 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1730 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
1731 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, drba
->drba_cred
);
1733 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1737 if (drba
->drba_origin
!= NULL
) {
1738 dsl_dataset_t
*origin
;
1740 error
= dsl_dataset_hold_flags(dp
, drba
->drba_origin
,
1741 dsflags
, FTAG
, &origin
);
1743 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1746 if (!origin
->ds_is_snapshot
) {
1747 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
1748 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1749 return (SET_ERROR(EINVAL
));
1751 if (dsl_dataset_phys(origin
)->ds_guid
!= fromguid
&&
1753 dsl_dataset_rele_flags(origin
, dsflags
, FTAG
);
1754 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1755 return (SET_ERROR(ENODEV
));
1757 dsl_dataset_rele_flags(origin
,
1760 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1767 dmu_recv_begin_sync(void *arg
, dmu_tx_t
*tx
)
1769 dmu_recv_begin_arg_t
*drba
= arg
;
1770 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1771 objset_t
*mos
= dp
->dp_meta_objset
;
1772 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1773 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1774 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1775 dsl_dataset_t
*ds
, *newds
;
1778 ds_hold_flags_t dsflags
= 0;
1780 uint64_t crflags
= 0;
1781 dsl_crypto_params_t
*dcpp
= NULL
;
1782 dsl_crypto_params_t dcp
= { 0 };
1784 if (drrb
->drr_flags
& DRR_FLAG_CI_DATA
)
1785 crflags
|= DS_FLAG_CI_DATASET
;
1786 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0) {
1787 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1789 dcp
.cp_cmd
= DCP_CMD_RAW_RECV
;
1792 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1794 /* create temporary clone */
1795 dsl_dataset_t
*snap
= NULL
;
1797 if (drba
->drba_snapobj
!= 0) {
1798 VERIFY0(dsl_dataset_hold_obj(dp
,
1799 drba
->drba_snapobj
, FTAG
, &snap
));
1801 /* we use the dcp whenever we are not making a clone */
1805 dsobj
= dsl_dataset_create_sync(ds
->ds_dir
, recv_clone_name
,
1806 snap
, crflags
, drba
->drba_cred
, dcpp
, tx
);
1807 if (drba
->drba_snapobj
!= 0)
1808 dsl_dataset_rele(snap
, FTAG
);
1809 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1813 dsl_dataset_t
*origin
= NULL
;
1815 VERIFY0(dsl_dir_hold(dp
, tofs
, FTAG
, &dd
, &tail
));
1817 if (drba
->drba_origin
!= NULL
) {
1818 VERIFY0(dsl_dataset_hold(dp
, drba
->drba_origin
,
1821 /* we use the dcp whenever we are not making a clone */
1825 /* Create new dataset. */
1826 dsobj
= dsl_dataset_create_sync(dd
, strrchr(tofs
, '/') + 1,
1827 origin
, crflags
, drba
->drba_cred
, dcpp
, tx
);
1829 dsl_dataset_rele(origin
, FTAG
);
1830 dsl_dir_rele(dd
, FTAG
);
1831 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
1833 VERIFY0(dsl_dataset_own_obj(dp
, dsobj
, dsflags
, dmu_recv_tag
, &newds
));
1834 VERIFY0(dmu_objset_from_ds(newds
, &os
));
1836 if (drba
->drba_cookie
->drc_resumable
) {
1837 dsl_dataset_zapify(newds
, tx
);
1838 if (drrb
->drr_fromguid
!= 0) {
1839 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_FROMGUID
,
1840 8, 1, &drrb
->drr_fromguid
, tx
));
1842 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TOGUID
,
1843 8, 1, &drrb
->drr_toguid
, tx
));
1844 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TONAME
,
1845 1, strlen(drrb
->drr_toname
) + 1, drrb
->drr_toname
, tx
));
1848 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OBJECT
,
1850 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OFFSET
,
1852 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_BYTES
,
1854 if (featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) {
1855 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_LARGEBLOCK
,
1858 if (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) {
1859 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_EMBEDOK
,
1862 if (featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
) {
1863 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_COMPRESSOK
,
1866 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1867 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_RAWOK
,
1873 * Usually the os->os_encrypted value is tied to the presence of a
1874 * DSL Crypto Key object in the dd. However, that will not be received
1875 * until dmu_recv_stream(), so we set the value manually for now.
1877 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
1878 os
->os_encrypted
= B_TRUE
;
1879 drba
->drba_cookie
->drc_raw
= B_TRUE
;
1882 dmu_buf_will_dirty(newds
->ds_dbuf
, tx
);
1883 dsl_dataset_phys(newds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
1886 * If we actually created a non-clone, we need to create the objset
1887 * in our new dataset. If this is a raw send we postpone this until
1888 * dmu_recv_stream() so that we can allocate the metadnode with the
1889 * properties from the DRR_BEGIN payload.
1891 rrw_enter(&newds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1892 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds
)) &&
1893 (featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0) {
1894 (void) dmu_objset_create_impl(dp
->dp_spa
,
1895 newds
, dsl_dataset_get_blkptr(newds
), drrb
->drr_type
, tx
);
1897 rrw_exit(&newds
->ds_bp_rwlock
, FTAG
);
1899 drba
->drba_cookie
->drc_ds
= newds
;
1901 spa_history_log_internal_ds(newds
, "receive", tx
, "");
1905 dmu_recv_resume_begin_check(void *arg
, dmu_tx_t
*tx
)
1907 dmu_recv_begin_arg_t
*drba
= arg
;
1908 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1909 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1911 ds_hold_flags_t dsflags
= 0;
1912 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1914 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1916 /* already checked */
1917 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1918 ASSERT(featureflags
& DMU_BACKUP_FEATURE_RESUMING
);
1920 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1921 DMU_COMPOUNDSTREAM
||
1922 drrb
->drr_type
>= DMU_OST_NUMTYPES
)
1923 return (SET_ERROR(EINVAL
));
1925 /* Verify pool version supports SA if SA_SPILL feature set */
1926 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
1927 spa_version(dp
->dp_spa
) < SPA_VERSION_SA
)
1928 return (SET_ERROR(ENOTSUP
));
1931 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1932 * record to a plain WRITE record, so the pool must have the
1933 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1934 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1936 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
1937 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
))
1938 return (SET_ERROR(ENOTSUP
));
1939 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
1940 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
))
1941 return (SET_ERROR(ENOTSUP
));
1944 * The receiving code doesn't know how to translate large blocks
1945 * to smaller ones, so the pool must have the LARGE_BLOCKS
1946 * feature enabled if the stream has LARGE_BLOCKS. Same with
1949 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
1950 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_BLOCKS
))
1951 return (SET_ERROR(ENOTSUP
));
1952 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_DNODE
) &&
1953 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_DNODE
))
1954 return (SET_ERROR(ENOTSUP
));
1956 /* 6 extra bytes for /%recv */
1957 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1958 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
1959 tofs
, recv_clone_name
);
1961 if ((featureflags
& DMU_BACKUP_FEATURE_RAW
) == 0)
1962 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
1964 if (dsl_dataset_hold_flags(dp
, recvname
, dsflags
, FTAG
, &ds
) != 0) {
1965 /* %recv does not exist; continue in tofs */
1966 error
= dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
);
1971 /* check that ds is marked inconsistent */
1972 if (!DS_IS_INCONSISTENT(ds
)) {
1973 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1974 return (SET_ERROR(EINVAL
));
1977 /* check that there is resuming data, and that the toguid matches */
1978 if (!dsl_dataset_is_zapified(ds
)) {
1979 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1980 return (SET_ERROR(EINVAL
));
1983 error
= zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1984 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
);
1985 if (error
!= 0 || drrb
->drr_toguid
!= val
) {
1986 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1987 return (SET_ERROR(EINVAL
));
1991 * Check if the receive is still running. If so, it will be owned.
1992 * Note that nothing else can own the dataset (e.g. after the receive
1993 * fails) because it will be marked inconsistent.
1995 if (dsl_dataset_has_owner(ds
)) {
1996 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
1997 return (SET_ERROR(EBUSY
));
2000 /* There should not be any snapshots of this fs yet. */
2001 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
->ds_dir
== ds
->ds_dir
) {
2002 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
2003 return (SET_ERROR(EINVAL
));
2007 * Note: resume point will be checked when we process the first WRITE
2011 /* check that the origin matches */
2013 (void) zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
2014 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
);
2015 if (drrb
->drr_fromguid
!= val
) {
2016 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
2017 return (SET_ERROR(EINVAL
));
2020 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
2025 dmu_recv_resume_begin_sync(void *arg
, dmu_tx_t
*tx
)
2027 dmu_recv_begin_arg_t
*drba
= arg
;
2028 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2029 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
2030 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
2031 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
2034 ds_hold_flags_t dsflags
= 0;
2036 /* 6 extra bytes for /%recv */
2037 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
2039 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
2040 tofs
, recv_clone_name
);
2042 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
2043 drba
->drba_cookie
->drc_raw
= B_TRUE
;
2045 dsflags
|= DS_HOLD_FLAG_DECRYPT
;
2048 if (dsl_dataset_hold_flags(dp
, recvname
, dsflags
, FTAG
, &ds
) != 0) {
2049 /* %recv does not exist; continue in tofs */
2050 VERIFY0(dsl_dataset_hold_flags(dp
, tofs
, dsflags
, FTAG
, &ds
));
2051 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
2054 /* clear the inconsistent flag so that we can own it */
2055 ASSERT(DS_IS_INCONSISTENT(ds
));
2056 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2057 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
2058 dsobj
= ds
->ds_object
;
2059 dsl_dataset_rele_flags(ds
, dsflags
, FTAG
);
2061 VERIFY0(dsl_dataset_own_obj(dp
, dsobj
, dsflags
, dmu_recv_tag
, &ds
));
2062 VERIFY0(dmu_objset_from_ds(ds
, &os
));
2064 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2065 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
2067 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2068 ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds
)) ||
2069 drba
->drba_cookie
->drc_raw
);
2070 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2072 drba
->drba_cookie
->drc_ds
= ds
;
2074 spa_history_log_internal_ds(ds
, "resume receive", tx
, "");
2078 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
2079 * succeeds; otherwise we will leak the holds on the datasets.
2082 dmu_recv_begin(char *tofs
, char *tosnap
, dmu_replay_record_t
*drr_begin
,
2083 boolean_t force
, boolean_t resumable
, char *origin
, dmu_recv_cookie_t
*drc
)
2085 dmu_recv_begin_arg_t drba
= { 0 };
2087 bzero(drc
, sizeof (dmu_recv_cookie_t
));
2088 drc
->drc_drr_begin
= drr_begin
;
2089 drc
->drc_drrb
= &drr_begin
->drr_u
.drr_begin
;
2090 drc
->drc_tosnap
= tosnap
;
2091 drc
->drc_tofs
= tofs
;
2092 drc
->drc_force
= force
;
2093 drc
->drc_resumable
= resumable
;
2094 drc
->drc_cred
= CRED();
2095 drc
->drc_clone
= (origin
!= NULL
);
2097 if (drc
->drc_drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
)) {
2098 drc
->drc_byteswap
= B_TRUE
;
2099 (void) fletcher_4_incremental_byteswap(drr_begin
,
2100 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
2101 byteswap_record(drr_begin
);
2102 } else if (drc
->drc_drrb
->drr_magic
== DMU_BACKUP_MAGIC
) {
2103 (void) fletcher_4_incremental_native(drr_begin
,
2104 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
2106 return (SET_ERROR(EINVAL
));
2109 drba
.drba_origin
= origin
;
2110 drba
.drba_cookie
= drc
;
2111 drba
.drba_cred
= CRED();
2113 if (DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
) &
2114 DMU_BACKUP_FEATURE_RESUMING
) {
2115 return (dsl_sync_task(tofs
,
2116 dmu_recv_resume_begin_check
, dmu_recv_resume_begin_sync
,
2117 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
));
2119 return (dsl_sync_task(tofs
,
2120 dmu_recv_begin_check
, dmu_recv_begin_sync
,
2121 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
));
2125 struct receive_record_arg
{
2126 dmu_replay_record_t header
;
2127 void *payload
; /* Pointer to a buffer containing the payload */
2129 * If the record is a write, pointer to the arc_buf_t containing the
2134 uint64_t bytes_read
; /* bytes read from stream when record created */
2135 boolean_t eos_marker
; /* Marks the end of the stream */
2139 struct receive_writer_arg
{
2145 * These three args are used to signal to the main thread that we're
2153 /* A map from guid to dataset to help handle dedup'd streams. */
2154 avl_tree_t
*guid_to_ds_map
;
2155 boolean_t resumable
;
2157 uint64_t last_object
;
2158 uint64_t last_offset
;
2159 uint64_t max_object
; /* highest object ID referenced in stream */
2160 uint64_t bytes_read
; /* bytes read when current record created */
2164 list_t list
; /* List of struct receive_objnode. */
2166 * Last object looked up. Used to assert that objects are being looked
2167 * up in ascending order.
2169 uint64_t last_lookup
;
2172 struct receive_objnode
{
2177 struct receive_arg
{
2179 vnode_t
*vp
; /* The vnode to read the stream from */
2180 uint64_t voff
; /* The current offset in the stream */
2181 uint64_t bytes_read
;
2183 * A record that has had its payload read in, but hasn't yet been handed
2184 * off to the worker thread.
2186 struct receive_record_arg
*rrd
;
2187 /* A record that has had its header read in, but not its payload. */
2188 struct receive_record_arg
*next_rrd
;
2190 zio_cksum_t prev_cksum
;
2194 uint64_t featureflags
;
2195 /* Sorted list of objects not to issue prefetches for. */
2196 struct objlist ignore_objlist
;
2199 typedef struct guid_map_entry
{
2202 dsl_dataset_t
*gme_ds
;
2207 guid_compare(const void *arg1
, const void *arg2
)
2209 const guid_map_entry_t
*gmep1
= (const guid_map_entry_t
*)arg1
;
2210 const guid_map_entry_t
*gmep2
= (const guid_map_entry_t
*)arg2
;
2212 return (AVL_CMP(gmep1
->guid
, gmep2
->guid
));
2216 free_guid_map_onexit(void *arg
)
2218 avl_tree_t
*ca
= arg
;
2219 void *cookie
= NULL
;
2220 guid_map_entry_t
*gmep
;
2222 while ((gmep
= avl_destroy_nodes(ca
, &cookie
)) != NULL
) {
2223 dsl_dataset_long_rele(gmep
->gme_ds
, gmep
);
2224 dsl_dataset_rele_flags(gmep
->gme_ds
,
2225 (gmep
->raw
) ? 0 : DS_HOLD_FLAG_DECRYPT
, gmep
);
2226 kmem_free(gmep
, sizeof (guid_map_entry_t
));
2229 kmem_free(ca
, sizeof (avl_tree_t
));
2233 receive_read(struct receive_arg
*ra
, int len
, void *buf
)
2238 * The code doesn't rely on this (lengths being multiples of 8). See
2239 * comment in dump_bytes.
2241 ASSERT(len
% 8 == 0 ||
2242 (ra
->featureflags
& DMU_BACKUP_FEATURE_RAW
) != 0);
2244 while (done
< len
) {
2247 ra
->err
= vn_rdwr(UIO_READ
, ra
->vp
,
2248 (char *)buf
+ done
, len
- done
,
2249 ra
->voff
, UIO_SYSSPACE
, FAPPEND
,
2250 RLIM64_INFINITY
, CRED(), &resid
);
2252 if (resid
== len
- done
) {
2254 * Note: ECKSUM indicates that the receive
2255 * was interrupted and can potentially be resumed.
2257 ra
->err
= SET_ERROR(ECKSUM
);
2259 ra
->voff
+= len
- done
- resid
;
2265 ra
->bytes_read
+= len
;
2267 ASSERT3U(done
, ==, len
);
2271 noinline
static void
2272 byteswap_record(dmu_replay_record_t
*drr
)
2274 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
2275 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
2276 drr
->drr_type
= BSWAP_32(drr
->drr_type
);
2277 drr
->drr_payloadlen
= BSWAP_32(drr
->drr_payloadlen
);
2279 switch (drr
->drr_type
) {
2281 DO64(drr_begin
.drr_magic
);
2282 DO64(drr_begin
.drr_versioninfo
);
2283 DO64(drr_begin
.drr_creation_time
);
2284 DO32(drr_begin
.drr_type
);
2285 DO32(drr_begin
.drr_flags
);
2286 DO64(drr_begin
.drr_toguid
);
2287 DO64(drr_begin
.drr_fromguid
);
2290 DO64(drr_object
.drr_object
);
2291 DO32(drr_object
.drr_type
);
2292 DO32(drr_object
.drr_bonustype
);
2293 DO32(drr_object
.drr_blksz
);
2294 DO32(drr_object
.drr_bonuslen
);
2295 DO32(drr_object
.drr_raw_bonuslen
);
2296 DO64(drr_object
.drr_toguid
);
2298 case DRR_FREEOBJECTS
:
2299 DO64(drr_freeobjects
.drr_firstobj
);
2300 DO64(drr_freeobjects
.drr_numobjs
);
2301 DO64(drr_freeobjects
.drr_toguid
);
2304 DO64(drr_write
.drr_object
);
2305 DO32(drr_write
.drr_type
);
2306 DO64(drr_write
.drr_offset
);
2307 DO64(drr_write
.drr_logical_size
);
2308 DO64(drr_write
.drr_toguid
);
2309 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write
.drr_key
.ddk_cksum
);
2310 DO64(drr_write
.drr_key
.ddk_prop
);
2311 DO64(drr_write
.drr_compressed_size
);
2313 case DRR_WRITE_BYREF
:
2314 DO64(drr_write_byref
.drr_object
);
2315 DO64(drr_write_byref
.drr_offset
);
2316 DO64(drr_write_byref
.drr_length
);
2317 DO64(drr_write_byref
.drr_toguid
);
2318 DO64(drr_write_byref
.drr_refguid
);
2319 DO64(drr_write_byref
.drr_refobject
);
2320 DO64(drr_write_byref
.drr_refoffset
);
2321 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write_byref
.
2323 DO64(drr_write_byref
.drr_key
.ddk_prop
);
2325 case DRR_WRITE_EMBEDDED
:
2326 DO64(drr_write_embedded
.drr_object
);
2327 DO64(drr_write_embedded
.drr_offset
);
2328 DO64(drr_write_embedded
.drr_length
);
2329 DO64(drr_write_embedded
.drr_toguid
);
2330 DO32(drr_write_embedded
.drr_lsize
);
2331 DO32(drr_write_embedded
.drr_psize
);
2334 DO64(drr_free
.drr_object
);
2335 DO64(drr_free
.drr_offset
);
2336 DO64(drr_free
.drr_length
);
2337 DO64(drr_free
.drr_toguid
);
2340 DO64(drr_spill
.drr_object
);
2341 DO64(drr_spill
.drr_length
);
2342 DO64(drr_spill
.drr_toguid
);
2343 DO64(drr_spill
.drr_compressed_size
);
2344 DO32(drr_spill
.drr_type
);
2346 case DRR_OBJECT_RANGE
:
2347 DO64(drr_object_range
.drr_firstobj
);
2348 DO64(drr_object_range
.drr_numslots
);
2349 DO64(drr_object_range
.drr_toguid
);
2352 DO64(drr_end
.drr_toguid
);
2353 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_end
.drr_checksum
);
2359 if (drr
->drr_type
!= DRR_BEGIN
) {
2360 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_checksum
.drr_checksum
);
2367 static inline uint8_t
2368 deduce_nblkptr(dmu_object_type_t bonus_type
, uint64_t bonus_size
)
2370 if (bonus_type
== DMU_OT_SA
) {
2374 ((DN_OLD_MAX_BONUSLEN
-
2375 MIN(DN_OLD_MAX_BONUSLEN
, bonus_size
)) >> SPA_BLKPTRSHIFT
));
2380 save_resume_state(struct receive_writer_arg
*rwa
,
2381 uint64_t object
, uint64_t offset
, dmu_tx_t
*tx
)
2383 int txgoff
= dmu_tx_get_txg(tx
) & TXG_MASK
;
2385 if (!rwa
->resumable
)
2389 * We use ds_resume_bytes[] != 0 to indicate that we need to
2390 * update this on disk, so it must not be 0.
2392 ASSERT(rwa
->bytes_read
!= 0);
2395 * We only resume from write records, which have a valid
2396 * (non-meta-dnode) object number.
2398 ASSERT(object
!= 0);
2401 * For resuming to work correctly, we must receive records in order,
2402 * sorted by object,offset. This is checked by the callers, but
2403 * assert it here for good measure.
2405 ASSERT3U(object
, >=, rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
]);
2406 ASSERT(object
!= rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] ||
2407 offset
>= rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
]);
2408 ASSERT3U(rwa
->bytes_read
, >=,
2409 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
]);
2411 rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] = object
;
2412 rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
] = offset
;
2413 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
] = rwa
->bytes_read
;
2417 receive_object(struct receive_writer_arg
*rwa
, struct drr_object
*drro
,
2420 dmu_object_info_t doi
;
2425 if (drro
->drr_type
== DMU_OT_NONE
||
2426 !DMU_OT_IS_VALID(drro
->drr_type
) ||
2427 !DMU_OT_IS_VALID(drro
->drr_bonustype
) ||
2428 drro
->drr_checksumtype
>= ZIO_CHECKSUM_FUNCTIONS
||
2429 drro
->drr_compress
>= ZIO_COMPRESS_FUNCTIONS
||
2430 P2PHASE(drro
->drr_blksz
, SPA_MINBLOCKSIZE
) ||
2431 drro
->drr_blksz
< SPA_MINBLOCKSIZE
||
2432 drro
->drr_blksz
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)) ||
2433 drro
->drr_bonuslen
>
2434 DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa
->os
))) ||
2435 drro
->drr_dn_slots
>
2436 (spa_maxdnodesize(dmu_objset_spa(rwa
->os
)) >> DNODE_SHIFT
)) {
2437 return (SET_ERROR(EINVAL
));
2441 if (drro
->drr_raw_bonuslen
< drro
->drr_bonuslen
||
2442 drro
->drr_indblkshift
> SPA_MAXBLOCKSHIFT
||
2443 drro
->drr_nlevels
> DN_MAX_LEVELS
||
2444 drro
->drr_nblkptr
> DN_MAX_NBLKPTR
||
2445 DN_SLOTS_TO_BONUSLEN(drro
->drr_dn_slots
) <
2446 drro
->drr_raw_bonuslen
)
2447 return (SET_ERROR(EINVAL
));
2449 if (drro
->drr_flags
!= 0 || drro
->drr_raw_bonuslen
!= 0 ||
2450 drro
->drr_indblkshift
!= 0 || drro
->drr_nlevels
!= 0 ||
2451 drro
->drr_nblkptr
!= 0)
2452 return (SET_ERROR(EINVAL
));
2455 err
= dmu_object_info(rwa
->os
, drro
->drr_object
, &doi
);
2457 if (err
!= 0 && err
!= ENOENT
)
2458 return (SET_ERROR(EINVAL
));
2459 object
= err
== 0 ? drro
->drr_object
: DMU_NEW_OBJECT
;
2461 if (drro
->drr_object
> rwa
->max_object
)
2462 rwa
->max_object
= drro
->drr_object
;
2465 * If we are losing blkptrs or changing the block size this must
2466 * be a new file instance. We must clear out the previous file
2467 * contents before we can change this type of metadata in the dnode.
2468 * Raw receives will also check that the indirect structure of the
2469 * dnode hasn't changed.
2472 uint32_t indblksz
= drro
->drr_indblkshift
?
2473 1ULL << drro
->drr_indblkshift
: 0;
2474 int nblkptr
= deduce_nblkptr(drro
->drr_bonustype
,
2475 drro
->drr_bonuslen
);
2477 /* nblkptr will be bounded by the bonus size and type */
2478 if (rwa
->raw
&& nblkptr
!= drro
->drr_nblkptr
)
2479 return (SET_ERROR(EINVAL
));
2481 if (drro
->drr_blksz
!= doi
.doi_data_block_size
||
2482 nblkptr
< doi
.doi_nblkptr
||
2484 (indblksz
!= doi
.doi_metadata_block_size
||
2485 drro
->drr_nlevels
< doi
.doi_indirection
))) {
2486 err
= dmu_free_long_range(rwa
->os
, drro
->drr_object
,
2489 return (SET_ERROR(EINVAL
));
2493 tx
= dmu_tx_create(rwa
->os
);
2494 dmu_tx_hold_bonus(tx
, object
);
2495 dmu_tx_hold_write(tx
, object
, 0, 0);
2496 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2502 if (object
== DMU_NEW_OBJECT
) {
2503 /* currently free, want to be allocated */
2504 err
= dmu_object_claim_dnsize(rwa
->os
, drro
->drr_object
,
2505 drro
->drr_type
, drro
->drr_blksz
,
2506 drro
->drr_bonustype
, drro
->drr_bonuslen
,
2507 drro
->drr_dn_slots
<< DNODE_SHIFT
, tx
);
2508 } else if (drro
->drr_type
!= doi
.doi_type
||
2509 drro
->drr_blksz
!= doi
.doi_data_block_size
||
2510 drro
->drr_bonustype
!= doi
.doi_bonus_type
||
2511 drro
->drr_bonuslen
!= doi
.doi_bonus_size
) {
2512 /* currently allocated, but with different properties */
2513 err
= dmu_object_reclaim(rwa
->os
, drro
->drr_object
,
2514 drro
->drr_type
, drro
->drr_blksz
,
2515 drro
->drr_bonustype
, drro
->drr_bonuslen
, tx
);
2519 return (SET_ERROR(EINVAL
));
2523 VERIFY0(dmu_object_dirty_raw(rwa
->os
, drro
->drr_object
, tx
));
2525 dmu_object_set_checksum(rwa
->os
, drro
->drr_object
,
2526 drro
->drr_checksumtype
, tx
);
2527 dmu_object_set_compress(rwa
->os
, drro
->drr_object
,
2528 drro
->drr_compress
, tx
);
2530 /* handle more restrictive dnode structuring for raw recvs */
2533 * Set the indirect block shift and nlevels. This will not fail
2534 * because we ensured all of the blocks were free earlier if
2535 * this is a new object.
2537 VERIFY0(dmu_object_set_blocksize(rwa
->os
, drro
->drr_object
,
2538 drro
->drr_blksz
, drro
->drr_indblkshift
, tx
));
2539 VERIFY0(dmu_object_set_nlevels(rwa
->os
, drro
->drr_object
,
2540 drro
->drr_nlevels
, tx
));
2545 uint32_t flags
= DMU_READ_NO_PREFETCH
;
2548 flags
|= DMU_READ_NO_DECRYPT
;
2550 VERIFY0(dmu_bonus_hold_impl(rwa
->os
, drro
->drr_object
,
2552 dmu_buf_will_dirty(db
, tx
);
2554 ASSERT3U(db
->db_size
, >=, drro
->drr_bonuslen
);
2555 bcopy(data
, db
->db_data
, DRR_OBJECT_PAYLOAD_SIZE(drro
));
2558 * Raw bonus buffers have their byteorder determined by the
2559 * DRR_OBJECT_RANGE record.
2561 if (rwa
->byteswap
&& !rwa
->raw
) {
2562 dmu_object_byteswap_t byteswap
=
2563 DMU_OT_BYTESWAP(drro
->drr_bonustype
);
2564 dmu_ot_byteswap
[byteswap
].ob_func(db
->db_data
,
2565 DRR_OBJECT_PAYLOAD_SIZE(drro
));
2567 dmu_buf_rele(db
, FTAG
);
2576 receive_freeobjects(struct receive_writer_arg
*rwa
,
2577 struct drr_freeobjects
*drrfo
)
2582 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
< drrfo
->drr_firstobj
)
2583 return (SET_ERROR(EINVAL
));
2585 for (obj
= drrfo
->drr_firstobj
== 0 ? 1 : drrfo
->drr_firstobj
;
2586 obj
< drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
&& next_err
== 0;
2587 next_err
= dmu_object_next(rwa
->os
, &obj
, FALSE
, 0)) {
2588 dmu_object_info_t doi
;
2591 err
= dmu_object_info(rwa
->os
, obj
, &doi
);
2598 err
= dmu_free_long_object_raw(rwa
->os
, obj
);
2600 err
= dmu_free_long_object(rwa
->os
, obj
);
2605 if (obj
> rwa
->max_object
)
2606 rwa
->max_object
= obj
;
2608 if (next_err
!= ESRCH
)
2614 receive_write(struct receive_writer_arg
*rwa
, struct drr_write
*drrw
,
2621 if (drrw
->drr_offset
+ drrw
->drr_logical_size
< drrw
->drr_offset
||
2622 !DMU_OT_IS_VALID(drrw
->drr_type
))
2623 return (SET_ERROR(EINVAL
));
2626 * For resuming to work, records must be in increasing order
2627 * by (object, offset).
2629 if (drrw
->drr_object
< rwa
->last_object
||
2630 (drrw
->drr_object
== rwa
->last_object
&&
2631 drrw
->drr_offset
< rwa
->last_offset
)) {
2632 return (SET_ERROR(EINVAL
));
2634 rwa
->last_object
= drrw
->drr_object
;
2635 rwa
->last_offset
= drrw
->drr_offset
;
2637 if (rwa
->last_object
> rwa
->max_object
)
2638 rwa
->max_object
= rwa
->last_object
;
2640 if (dmu_object_info(rwa
->os
, drrw
->drr_object
, NULL
) != 0)
2641 return (SET_ERROR(EINVAL
));
2643 tx
= dmu_tx_create(rwa
->os
);
2644 dmu_tx_hold_write(tx
, drrw
->drr_object
,
2645 drrw
->drr_offset
, drrw
->drr_logical_size
);
2646 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2653 VERIFY0(dmu_object_dirty_raw(rwa
->os
, drrw
->drr_object
, tx
));
2655 if (rwa
->byteswap
&& !arc_is_encrypted(abuf
) &&
2656 arc_get_compression(abuf
) == ZIO_COMPRESS_OFF
) {
2657 dmu_object_byteswap_t byteswap
=
2658 DMU_OT_BYTESWAP(drrw
->drr_type
);
2659 dmu_ot_byteswap
[byteswap
].ob_func(abuf
->b_data
,
2660 DRR_WRITE_PAYLOAD_SIZE(drrw
));
2663 VERIFY0(dnode_hold(rwa
->os
, drrw
->drr_object
, FTAG
, &dn
));
2664 dmu_assign_arcbuf_by_dnode(dn
, drrw
->drr_offset
, abuf
, tx
);
2665 dnode_rele(dn
, FTAG
);
2668 * Note: If the receive fails, we want the resume stream to start
2669 * with the same record that we last successfully received (as opposed
2670 * to the next record), so that we can verify that we are
2671 * resuming from the correct location.
2673 save_resume_state(rwa
, drrw
->drr_object
, drrw
->drr_offset
, tx
);
2680 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
2681 * streams to refer to a copy of the data that is already on the
2682 * system because it came in earlier in the stream. This function
2683 * finds the earlier copy of the data, and uses that copy instead of
2684 * data from the stream to fulfill this write.
2687 receive_write_byref(struct receive_writer_arg
*rwa
,
2688 struct drr_write_byref
*drrwbr
)
2692 guid_map_entry_t gmesrch
;
2693 guid_map_entry_t
*gmep
;
2695 objset_t
*ref_os
= NULL
;
2696 int flags
= DMU_READ_PREFETCH
;
2699 if (drrwbr
->drr_offset
+ drrwbr
->drr_length
< drrwbr
->drr_offset
)
2700 return (SET_ERROR(EINVAL
));
2703 * If the GUID of the referenced dataset is different from the
2704 * GUID of the target dataset, find the referenced dataset.
2706 if (drrwbr
->drr_toguid
!= drrwbr
->drr_refguid
) {
2707 gmesrch
.guid
= drrwbr
->drr_refguid
;
2708 if ((gmep
= avl_find(rwa
->guid_to_ds_map
, &gmesrch
,
2710 return (SET_ERROR(EINVAL
));
2712 if (dmu_objset_from_ds(gmep
->gme_ds
, &ref_os
))
2713 return (SET_ERROR(EINVAL
));
2718 if (drrwbr
->drr_object
> rwa
->max_object
)
2719 rwa
->max_object
= drrwbr
->drr_object
;
2722 flags
|= DMU_READ_NO_DECRYPT
;
2724 /* may return either a regular db or an encrypted one */
2725 err
= dmu_buf_hold(ref_os
, drrwbr
->drr_refobject
,
2726 drrwbr
->drr_refoffset
, FTAG
, &dbp
, flags
);
2730 tx
= dmu_tx_create(rwa
->os
);
2732 dmu_tx_hold_write(tx
, drrwbr
->drr_object
,
2733 drrwbr
->drr_offset
, drrwbr
->drr_length
);
2734 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2741 VERIFY0(dmu_object_dirty_raw(rwa
->os
, drrwbr
->drr_object
, tx
));
2742 dmu_copy_from_buf(rwa
->os
, drrwbr
->drr_object
,
2743 drrwbr
->drr_offset
, dbp
, tx
);
2745 dmu_write(rwa
->os
, drrwbr
->drr_object
,
2746 drrwbr
->drr_offset
, drrwbr
->drr_length
, dbp
->db_data
, tx
);
2748 dmu_buf_rele(dbp
, FTAG
);
2750 /* See comment in restore_write. */
2751 save_resume_state(rwa
, drrwbr
->drr_object
, drrwbr
->drr_offset
, tx
);
2757 receive_write_embedded(struct receive_writer_arg
*rwa
,
2758 struct drr_write_embedded
*drrwe
, void *data
)
2763 if (drrwe
->drr_offset
+ drrwe
->drr_length
< drrwe
->drr_offset
)
2764 return (SET_ERROR(EINVAL
));
2766 if (drrwe
->drr_psize
> BPE_PAYLOAD_SIZE
)
2767 return (SET_ERROR(EINVAL
));
2769 if (drrwe
->drr_etype
>= NUM_BP_EMBEDDED_TYPES
)
2770 return (SET_ERROR(EINVAL
));
2771 if (drrwe
->drr_compression
>= ZIO_COMPRESS_FUNCTIONS
)
2772 return (SET_ERROR(EINVAL
));
2774 return (SET_ERROR(EINVAL
));
2776 if (drrwe
->drr_object
> rwa
->max_object
)
2777 rwa
->max_object
= drrwe
->drr_object
;
2779 tx
= dmu_tx_create(rwa
->os
);
2781 dmu_tx_hold_write(tx
, drrwe
->drr_object
,
2782 drrwe
->drr_offset
, drrwe
->drr_length
);
2783 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2789 dmu_write_embedded(rwa
->os
, drrwe
->drr_object
,
2790 drrwe
->drr_offset
, data
, drrwe
->drr_etype
,
2791 drrwe
->drr_compression
, drrwe
->drr_lsize
, drrwe
->drr_psize
,
2792 rwa
->byteswap
^ ZFS_HOST_BYTEORDER
, tx
);
2794 /* See comment in restore_write. */
2795 save_resume_state(rwa
, drrwe
->drr_object
, drrwe
->drr_offset
, tx
);
2801 receive_spill(struct receive_writer_arg
*rwa
, struct drr_spill
*drrs
,
2805 dmu_buf_t
*db
, *db_spill
;
2808 if (drrs
->drr_length
< SPA_MINBLOCKSIZE
||
2809 drrs
->drr_length
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)))
2810 return (SET_ERROR(EINVAL
));
2813 if (!DMU_OT_IS_VALID(drrs
->drr_type
) ||
2814 drrs
->drr_compressiontype
>= ZIO_COMPRESS_FUNCTIONS
||
2815 drrs
->drr_compressed_size
== 0)
2816 return (SET_ERROR(EINVAL
));
2819 if (dmu_object_info(rwa
->os
, drrs
->drr_object
, NULL
) != 0)
2820 return (SET_ERROR(EINVAL
));
2822 if (drrs
->drr_object
> rwa
->max_object
)
2823 rwa
->max_object
= drrs
->drr_object
;
2825 VERIFY0(dmu_bonus_hold(rwa
->os
, drrs
->drr_object
, FTAG
, &db
));
2826 if ((err
= dmu_spill_hold_by_bonus(db
, FTAG
, &db_spill
)) != 0) {
2827 dmu_buf_rele(db
, FTAG
);
2831 tx
= dmu_tx_create(rwa
->os
);
2833 dmu_tx_hold_spill(tx
, db
->db_object
);
2835 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2837 dmu_buf_rele(db
, FTAG
);
2838 dmu_buf_rele(db_spill
, FTAG
);
2842 dmu_buf_will_dirty(db_spill
, tx
);
2844 VERIFY0(dmu_object_dirty_raw(rwa
->os
, drrs
->drr_object
, tx
));
2846 if (db_spill
->db_size
< drrs
->drr_length
)
2847 VERIFY(0 == dbuf_spill_set_blksz(db_spill
,
2848 drrs
->drr_length
, tx
));
2849 dbuf_assign_arcbuf((dmu_buf_impl_t
*)db_spill
, abuf
, tx
);
2851 dmu_buf_rele(db
, FTAG
);
2852 dmu_buf_rele(db_spill
, FTAG
);
2860 receive_free(struct receive_writer_arg
*rwa
, struct drr_free
*drrf
)
2864 if (drrf
->drr_length
!= DMU_OBJECT_END
&&
2865 drrf
->drr_offset
+ drrf
->drr_length
< drrf
->drr_offset
)
2866 return (SET_ERROR(EINVAL
));
2868 if (dmu_object_info(rwa
->os
, drrf
->drr_object
, NULL
) != 0)
2869 return (SET_ERROR(EINVAL
));
2871 if (drrf
->drr_object
> rwa
->max_object
)
2872 rwa
->max_object
= drrf
->drr_object
;
2875 err
= dmu_free_long_range_raw(rwa
->os
, drrf
->drr_object
,
2876 drrf
->drr_offset
, drrf
->drr_length
);
2878 err
= dmu_free_long_range(rwa
->os
, drrf
->drr_object
,
2879 drrf
->drr_offset
, drrf
->drr_length
);
2886 receive_object_range(struct receive_writer_arg
*rwa
,
2887 struct drr_object_range
*drror
)
2891 dnode_t
*mdn
= NULL
;
2892 dmu_buf_t
*db
= NULL
;
2896 * By default, we assume this block is in our native format
2897 * (ZFS_HOST_BYTEORDER). We then take into account whether
2898 * the send stream is byteswapped (rwa->byteswap). Finally,
2899 * we need to byteswap again if this particular block was
2900 * in non-native format on the send side.
2902 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^ rwa
->byteswap
^
2903 !!DRR_IS_RAW_BYTESWAPPED(drror
->drr_flags
);
2906 * Since dnode block sizes are constant, we should not need to worry
2907 * about making sure that the dnode block size is the same on the
2908 * sending and receiving sides for the time being. For non-raw sends,
2909 * this does not matter (and in fact we do not send a DRR_OBJECT_RANGE
2910 * record at all). Raw sends require this record type because the
2911 * encryption parameters are used to protect an entire block of bonus
2912 * buffers. If the size of dnode blocks ever becomes variable,
2913 * handling will need to be added to ensure that dnode block sizes
2914 * match on the sending and receiving side.
2916 if (drror
->drr_numslots
!= DNODES_PER_BLOCK
||
2917 P2PHASE(drror
->drr_firstobj
, DNODES_PER_BLOCK
) != 0 ||
2919 return (SET_ERROR(EINVAL
));
2921 if (drror
->drr_firstobj
> rwa
->max_object
)
2922 rwa
->max_object
= drror
->drr_firstobj
;
2924 offset
= drror
->drr_firstobj
* sizeof (dnode_phys_t
);
2925 mdn
= DMU_META_DNODE(rwa
->os
);
2927 tx
= dmu_tx_create(rwa
->os
);
2928 ret
= dmu_tx_assign(tx
, TXG_WAIT
);
2934 ret
= dmu_buf_hold_by_dnode(mdn
, offset
, FTAG
, &db
,
2935 DMU_READ_PREFETCH
| DMU_READ_NO_DECRYPT
);
2942 * Convert the buffer associated with this range of dnodes to a
2943 * raw buffer. This ensures that it will be written out as a raw
2944 * buffer when we fill in the dnode objects in future records.
2945 * Since we are commiting this tx now, it is technically possible
2946 * for the dnode block to end up on-disk with the incorrect MAC.
2947 * Despite this, the dataset is marked as inconsistent so no other
2948 * code paths (apart from scrubs) will attempt to read this data.
2949 * Scrubs will not be effected by this either since scrubs only
2950 * read raw data and do not attempt to check the MAC.
2952 dmu_convert_to_raw(db
, byteorder
, drror
->drr_salt
, drror
->drr_iv
,
2953 drror
->drr_mac
, tx
);
2954 dmu_buf_rele(db
, FTAG
);
2959 /* used to destroy the drc_ds on error */
2961 dmu_recv_cleanup_ds(dmu_recv_cookie_t
*drc
)
2963 dsl_dataset_t
*ds
= drc
->drc_ds
;
2964 ds_hold_flags_t dsflags
= (drc
->drc_raw
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
2967 * Wait for the txg sync before cleaning up the receive. For
2968 * resumable receives, this ensures that our resume state has
2969 * been written out to disk. For raw receives, this ensures
2970 * that the user accounting code will not attempt to do anything
2971 * after we stopped receiving the dataset.
2973 txg_wait_synced(ds
->ds_dir
->dd_pool
, 0);
2975 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2976 if (drc
->drc_resumable
&& !BP_IS_HOLE(dsl_dataset_get_blkptr(ds
))) {
2977 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2978 dsl_dataset_disown(ds
, dsflags
, dmu_recv_tag
);
2980 char name
[ZFS_MAX_DATASET_NAME_LEN
];
2981 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2982 dsl_dataset_name(ds
, name
);
2983 dsl_dataset_disown(ds
, dsflags
, dmu_recv_tag
);
2984 (void) dsl_destroy_head(name
);
2989 receive_cksum(struct receive_arg
*ra
, int len
, void *buf
)
2992 (void) fletcher_4_incremental_byteswap(buf
, len
, &ra
->cksum
);
2994 (void) fletcher_4_incremental_native(buf
, len
, &ra
->cksum
);
2999 * Read the payload into a buffer of size len, and update the current record's
3001 * Allocate ra->next_rrd and read the next record's header into
3002 * ra->next_rrd->header.
3003 * Verify checksum of payload and next record.
3006 receive_read_payload_and_next_header(struct receive_arg
*ra
, int len
, void *buf
)
3009 zio_cksum_t cksum_orig
;
3010 zio_cksum_t
*cksump
;
3013 ASSERT3U(len
, <=, SPA_MAXBLOCKSIZE
);
3014 err
= receive_read(ra
, len
, buf
);
3017 receive_cksum(ra
, len
, buf
);
3019 /* note: rrd is NULL when reading the begin record's payload */
3020 if (ra
->rrd
!= NULL
) {
3021 ra
->rrd
->payload
= buf
;
3022 ra
->rrd
->payload_size
= len
;
3023 ra
->rrd
->bytes_read
= ra
->bytes_read
;
3027 ra
->prev_cksum
= ra
->cksum
;
3029 ra
->next_rrd
= kmem_zalloc(sizeof (*ra
->next_rrd
), KM_SLEEP
);
3030 err
= receive_read(ra
, sizeof (ra
->next_rrd
->header
),
3031 &ra
->next_rrd
->header
);
3032 ra
->next_rrd
->bytes_read
= ra
->bytes_read
;
3035 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
3036 ra
->next_rrd
= NULL
;
3039 if (ra
->next_rrd
->header
.drr_type
== DRR_BEGIN
) {
3040 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
3041 ra
->next_rrd
= NULL
;
3042 return (SET_ERROR(EINVAL
));
3046 * Note: checksum is of everything up to but not including the
3049 ASSERT3U(offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
3050 ==, sizeof (dmu_replay_record_t
) - sizeof (zio_cksum_t
));
3052 offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
3053 &ra
->next_rrd
->header
);
3055 cksum_orig
= ra
->next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
3056 cksump
= &ra
->next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
3059 byteswap_record(&ra
->next_rrd
->header
);
3061 if ((!ZIO_CHECKSUM_IS_ZERO(cksump
)) &&
3062 !ZIO_CHECKSUM_EQUAL(ra
->cksum
, *cksump
)) {
3063 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
3064 ra
->next_rrd
= NULL
;
3065 return (SET_ERROR(ECKSUM
));
3068 receive_cksum(ra
, sizeof (cksum_orig
), &cksum_orig
);
3074 objlist_create(struct objlist
*list
)
3076 list_create(&list
->list
, sizeof (struct receive_objnode
),
3077 offsetof(struct receive_objnode
, node
));
3078 list
->last_lookup
= 0;
3082 objlist_destroy(struct objlist
*list
)
3084 for (struct receive_objnode
*n
= list_remove_head(&list
->list
);
3085 n
!= NULL
; n
= list_remove_head(&list
->list
)) {
3086 kmem_free(n
, sizeof (*n
));
3088 list_destroy(&list
->list
);
3092 * This function looks through the objlist to see if the specified object number
3093 * is contained in the objlist. In the process, it will remove all object
3094 * numbers in the list that are smaller than the specified object number. Thus,
3095 * any lookup of an object number smaller than a previously looked up object
3096 * number will always return false; therefore, all lookups should be done in
3100 objlist_exists(struct objlist
*list
, uint64_t object
)
3102 struct receive_objnode
*node
= list_head(&list
->list
);
3103 ASSERT3U(object
, >=, list
->last_lookup
);
3104 list
->last_lookup
= object
;
3105 while (node
!= NULL
&& node
->object
< object
) {
3106 VERIFY3P(node
, ==, list_remove_head(&list
->list
));
3107 kmem_free(node
, sizeof (*node
));
3108 node
= list_head(&list
->list
);
3110 return (node
!= NULL
&& node
->object
== object
);
3114 * The objlist is a list of object numbers stored in ascending order. However,
3115 * the insertion of new object numbers does not seek out the correct location to
3116 * store a new object number; instead, it appends it to the list for simplicity.
3117 * Thus, any users must take care to only insert new object numbers in ascending
3121 objlist_insert(struct objlist
*list
, uint64_t object
)
3123 struct receive_objnode
*node
= kmem_zalloc(sizeof (*node
), KM_SLEEP
);
3124 node
->object
= object
;
3127 struct receive_objnode
*last_object
= list_tail(&list
->list
);
3128 uint64_t last_objnum
= (last_object
!= NULL
? last_object
->object
: 0);
3129 ASSERT3U(node
->object
, >, last_objnum
);
3132 list_insert_tail(&list
->list
, node
);
3136 * Issue the prefetch reads for any necessary indirect blocks.
3138 * We use the object ignore list to tell us whether or not to issue prefetches
3139 * for a given object. We do this for both correctness (in case the blocksize
3140 * of an object has changed) and performance (if the object doesn't exist, don't
3141 * needlessly try to issue prefetches). We also trim the list as we go through
3142 * the stream to prevent it from growing to an unbounded size.
3144 * The object numbers within will always be in sorted order, and any write
3145 * records we see will also be in sorted order, but they're not sorted with
3146 * respect to each other (i.e. we can get several object records before
3147 * receiving each object's write records). As a result, once we've reached a
3148 * given object number, we can safely remove any reference to lower object
3149 * numbers in the ignore list. In practice, we receive up to 32 object records
3150 * before receiving write records, so the list can have up to 32 nodes in it.
3154 receive_read_prefetch(struct receive_arg
*ra
,
3155 uint64_t object
, uint64_t offset
, uint64_t length
)
3157 if (!objlist_exists(&ra
->ignore_objlist
, object
)) {
3158 dmu_prefetch(ra
->os
, object
, 1, offset
, length
,
3159 ZIO_PRIORITY_SYNC_READ
);
3164 * Read records off the stream, issuing any necessary prefetches.
3167 receive_read_record(struct receive_arg
*ra
)
3171 switch (ra
->rrd
->header
.drr_type
) {
3174 struct drr_object
*drro
= &ra
->rrd
->header
.drr_u
.drr_object
;
3175 uint32_t size
= DRR_OBJECT_PAYLOAD_SIZE(drro
);
3176 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
3177 dmu_object_info_t doi
;
3179 err
= receive_read_payload_and_next_header(ra
, size
, buf
);
3181 kmem_free(buf
, size
);
3184 err
= dmu_object_info(ra
->os
, drro
->drr_object
, &doi
);
3186 * See receive_read_prefetch for an explanation why we're
3187 * storing this object in the ignore_obj_list.
3189 if (err
== ENOENT
||
3190 (err
== 0 && doi
.doi_data_block_size
!= drro
->drr_blksz
)) {
3191 objlist_insert(&ra
->ignore_objlist
, drro
->drr_object
);
3196 case DRR_FREEOBJECTS
:
3198 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3203 struct drr_write
*drrw
= &ra
->rrd
->header
.drr_u
.drr_write
;
3205 boolean_t is_meta
= DMU_OT_IS_METADATA(drrw
->drr_type
);
3208 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^
3209 !!DRR_IS_RAW_BYTESWAPPED(drrw
->drr_flags
) ^
3212 abuf
= arc_loan_raw_buf(dmu_objset_spa(ra
->os
),
3213 drrw
->drr_object
, byteorder
, drrw
->drr_salt
,
3214 drrw
->drr_iv
, drrw
->drr_mac
, drrw
->drr_type
,
3215 drrw
->drr_compressed_size
, drrw
->drr_logical_size
,
3216 drrw
->drr_compressiontype
);
3217 } else if (DRR_WRITE_COMPRESSED(drrw
)) {
3218 ASSERT3U(drrw
->drr_compressed_size
, >, 0);
3219 ASSERT3U(drrw
->drr_logical_size
, >=,
3220 drrw
->drr_compressed_size
);
3222 abuf
= arc_loan_compressed_buf(
3223 dmu_objset_spa(ra
->os
),
3224 drrw
->drr_compressed_size
, drrw
->drr_logical_size
,
3225 drrw
->drr_compressiontype
);
3227 abuf
= arc_loan_buf(dmu_objset_spa(ra
->os
),
3228 is_meta
, drrw
->drr_logical_size
);
3231 err
= receive_read_payload_and_next_header(ra
,
3232 DRR_WRITE_PAYLOAD_SIZE(drrw
), abuf
->b_data
);
3234 dmu_return_arcbuf(abuf
);
3237 ra
->rrd
->arc_buf
= abuf
;
3238 receive_read_prefetch(ra
, drrw
->drr_object
, drrw
->drr_offset
,
3239 drrw
->drr_logical_size
);
3242 case DRR_WRITE_BYREF
:
3244 struct drr_write_byref
*drrwb
=
3245 &ra
->rrd
->header
.drr_u
.drr_write_byref
;
3246 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3247 receive_read_prefetch(ra
, drrwb
->drr_object
, drrwb
->drr_offset
,
3251 case DRR_WRITE_EMBEDDED
:
3253 struct drr_write_embedded
*drrwe
=
3254 &ra
->rrd
->header
.drr_u
.drr_write_embedded
;
3255 uint32_t size
= P2ROUNDUP(drrwe
->drr_psize
, 8);
3256 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
3258 err
= receive_read_payload_and_next_header(ra
, size
, buf
);
3260 kmem_free(buf
, size
);
3264 receive_read_prefetch(ra
, drrwe
->drr_object
, drrwe
->drr_offset
,
3271 * It might be beneficial to prefetch indirect blocks here, but
3272 * we don't really have the data to decide for sure.
3274 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3279 struct drr_end
*drre
= &ra
->rrd
->header
.drr_u
.drr_end
;
3280 if (!ZIO_CHECKSUM_EQUAL(ra
->prev_cksum
, drre
->drr_checksum
))
3281 return (SET_ERROR(ECKSUM
));
3286 struct drr_spill
*drrs
= &ra
->rrd
->header
.drr_u
.drr_spill
;
3288 int len
= DRR_SPILL_PAYLOAD_SIZE(drrs
);
3290 /* DRR_SPILL records are either raw or uncompressed */
3292 boolean_t byteorder
= ZFS_HOST_BYTEORDER
^
3293 !!DRR_IS_RAW_BYTESWAPPED(drrs
->drr_flags
) ^
3296 abuf
= arc_loan_raw_buf(dmu_objset_spa(ra
->os
),
3297 drrs
->drr_object
, byteorder
, drrs
->drr_salt
,
3298 drrs
->drr_iv
, drrs
->drr_mac
, drrs
->drr_type
,
3299 drrs
->drr_compressed_size
, drrs
->drr_length
,
3300 drrs
->drr_compressiontype
);
3302 abuf
= arc_loan_buf(dmu_objset_spa(ra
->os
),
3303 DMU_OT_IS_METADATA(drrs
->drr_type
),
3307 err
= receive_read_payload_and_next_header(ra
, len
,
3310 dmu_return_arcbuf(abuf
);
3313 ra
->rrd
->arc_buf
= abuf
;
3316 case DRR_OBJECT_RANGE
:
3318 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
3322 return (SET_ERROR(EINVAL
));
3327 dprintf_drr(struct receive_record_arg
*rrd
, int err
)
3329 switch (rrd
->header
.drr_type
) {
3332 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
3333 dprintf("drr_type = OBJECT obj = %llu type = %u "
3334 "bonustype = %u blksz = %u bonuslen = %u cksumtype = %u "
3335 "compress = %u dn_slots = %u err = %d\n",
3336 drro
->drr_object
, drro
->drr_type
, drro
->drr_bonustype
,
3337 drro
->drr_blksz
, drro
->drr_bonuslen
,
3338 drro
->drr_checksumtype
, drro
->drr_compress
,
3339 drro
->drr_dn_slots
, err
);
3342 case DRR_FREEOBJECTS
:
3344 struct drr_freeobjects
*drrfo
=
3345 &rrd
->header
.drr_u
.drr_freeobjects
;
3346 dprintf("drr_type = FREEOBJECTS firstobj = %llu "
3347 "numobjs = %llu err = %d\n",
3348 drrfo
->drr_firstobj
, drrfo
->drr_numobjs
, err
);
3353 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
3354 dprintf("drr_type = WRITE obj = %llu type = %u offset = %llu "
3355 "lsize = %llu cksumtype = %u cksumflags = %u "
3356 "compress = %u psize = %llu err = %d\n",
3357 drrw
->drr_object
, drrw
->drr_type
, drrw
->drr_offset
,
3358 drrw
->drr_logical_size
, drrw
->drr_checksumtype
,
3359 drrw
->drr_flags
, drrw
->drr_compressiontype
,
3360 drrw
->drr_compressed_size
, err
);
3363 case DRR_WRITE_BYREF
:
3365 struct drr_write_byref
*drrwbr
=
3366 &rrd
->header
.drr_u
.drr_write_byref
;
3367 dprintf("drr_type = WRITE_BYREF obj = %llu offset = %llu "
3368 "length = %llu toguid = %llx refguid = %llx "
3369 "refobject = %llu refoffset = %llu cksumtype = %u "
3370 "cksumflags = %u err = %d\n",
3371 drrwbr
->drr_object
, drrwbr
->drr_offset
,
3372 drrwbr
->drr_length
, drrwbr
->drr_toguid
,
3373 drrwbr
->drr_refguid
, drrwbr
->drr_refobject
,
3374 drrwbr
->drr_refoffset
, drrwbr
->drr_checksumtype
,
3375 drrwbr
->drr_flags
, err
);
3378 case DRR_WRITE_EMBEDDED
:
3380 struct drr_write_embedded
*drrwe
=
3381 &rrd
->header
.drr_u
.drr_write_embedded
;
3382 dprintf("drr_type = WRITE_EMBEDDED obj = %llu offset = %llu "
3383 "length = %llu compress = %u etype = %u lsize = %u "
3384 "psize = %u err = %d\n",
3385 drrwe
->drr_object
, drrwe
->drr_offset
, drrwe
->drr_length
,
3386 drrwe
->drr_compression
, drrwe
->drr_etype
,
3387 drrwe
->drr_lsize
, drrwe
->drr_psize
, err
);
3392 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
3393 dprintf("drr_type = FREE obj = %llu offset = %llu "
3394 "length = %lld err = %d\n",
3395 drrf
->drr_object
, drrf
->drr_offset
, drrf
->drr_length
,
3401 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
3402 dprintf("drr_type = SPILL obj = %llu length = %llu "
3403 "err = %d\n", drrs
->drr_object
, drrs
->drr_length
, err
);
3412 * Commit the records to the pool.
3415 receive_process_record(struct receive_writer_arg
*rwa
,
3416 struct receive_record_arg
*rrd
)
3420 /* Processing in order, therefore bytes_read should be increasing. */
3421 ASSERT3U(rrd
->bytes_read
, >=, rwa
->bytes_read
);
3422 rwa
->bytes_read
= rrd
->bytes_read
;
3424 switch (rrd
->header
.drr_type
) {
3427 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
3428 err
= receive_object(rwa
, drro
, rrd
->payload
);
3429 kmem_free(rrd
->payload
, rrd
->payload_size
);
3430 rrd
->payload
= NULL
;
3433 case DRR_FREEOBJECTS
:
3435 struct drr_freeobjects
*drrfo
=
3436 &rrd
->header
.drr_u
.drr_freeobjects
;
3437 err
= receive_freeobjects(rwa
, drrfo
);
3442 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
3443 err
= receive_write(rwa
, drrw
, rrd
->arc_buf
);
3444 /* if receive_write() is successful, it consumes the arc_buf */
3446 dmu_return_arcbuf(rrd
->arc_buf
);
3447 rrd
->arc_buf
= NULL
;
3448 rrd
->payload
= NULL
;
3451 case DRR_WRITE_BYREF
:
3453 struct drr_write_byref
*drrwbr
=
3454 &rrd
->header
.drr_u
.drr_write_byref
;
3455 err
= receive_write_byref(rwa
, drrwbr
);
3458 case DRR_WRITE_EMBEDDED
:
3460 struct drr_write_embedded
*drrwe
=
3461 &rrd
->header
.drr_u
.drr_write_embedded
;
3462 err
= receive_write_embedded(rwa
, drrwe
, rrd
->payload
);
3463 kmem_free(rrd
->payload
, rrd
->payload_size
);
3464 rrd
->payload
= NULL
;
3469 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
3470 err
= receive_free(rwa
, drrf
);
3475 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
3476 err
= receive_spill(rwa
, drrs
, rrd
->arc_buf
);
3477 /* if receive_spill() is successful, it consumes the arc_buf */
3479 dmu_return_arcbuf(rrd
->arc_buf
);
3480 rrd
->arc_buf
= NULL
;
3481 rrd
->payload
= NULL
;
3484 case DRR_OBJECT_RANGE
:
3486 struct drr_object_range
*drror
=
3487 &rrd
->header
.drr_u
.drr_object_range
;
3488 return (receive_object_range(rwa
, drror
));
3491 return (SET_ERROR(EINVAL
));
3495 dprintf_drr(rrd
, err
);
3501 * dmu_recv_stream's worker thread; pull records off the queue, and then call
3502 * receive_process_record When we're done, signal the main thread and exit.
3505 receive_writer_thread(void *arg
)
3507 struct receive_writer_arg
*rwa
= arg
;
3508 struct receive_record_arg
*rrd
;
3509 fstrans_cookie_t cookie
= spl_fstrans_mark();
3511 for (rrd
= bqueue_dequeue(&rwa
->q
); !rrd
->eos_marker
;
3512 rrd
= bqueue_dequeue(&rwa
->q
)) {
3514 * If there's an error, the main thread will stop putting things
3515 * on the queue, but we need to clear everything in it before we
3518 if (rwa
->err
== 0) {
3519 rwa
->err
= receive_process_record(rwa
, rrd
);
3520 } else if (rrd
->arc_buf
!= NULL
) {
3521 dmu_return_arcbuf(rrd
->arc_buf
);
3522 rrd
->arc_buf
= NULL
;
3523 rrd
->payload
= NULL
;
3524 } else if (rrd
->payload
!= NULL
) {
3525 kmem_free(rrd
->payload
, rrd
->payload_size
);
3526 rrd
->payload
= NULL
;
3528 kmem_free(rrd
, sizeof (*rrd
));
3530 kmem_free(rrd
, sizeof (*rrd
));
3531 mutex_enter(&rwa
->mutex
);
3533 cv_signal(&rwa
->cv
);
3534 mutex_exit(&rwa
->mutex
);
3535 spl_fstrans_unmark(cookie
);
3540 resume_check(struct receive_arg
*ra
, nvlist_t
*begin_nvl
)
3543 objset_t
*mos
= dmu_objset_pool(ra
->os
)->dp_meta_objset
;
3544 uint64_t dsobj
= dmu_objset_id(ra
->os
);
3545 uint64_t resume_obj
, resume_off
;
3547 if (nvlist_lookup_uint64(begin_nvl
,
3548 "resume_object", &resume_obj
) != 0 ||
3549 nvlist_lookup_uint64(begin_nvl
,
3550 "resume_offset", &resume_off
) != 0) {
3551 return (SET_ERROR(EINVAL
));
3553 VERIFY0(zap_lookup(mos
, dsobj
,
3554 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
));
3555 if (resume_obj
!= val
)
3556 return (SET_ERROR(EINVAL
));
3557 VERIFY0(zap_lookup(mos
, dsobj
,
3558 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
));
3559 if (resume_off
!= val
)
3560 return (SET_ERROR(EINVAL
));
3566 * Read in the stream's records, one by one, and apply them to the pool. There
3567 * are two threads involved; the thread that calls this function will spin up a
3568 * worker thread, read the records off the stream one by one, and issue
3569 * prefetches for any necessary indirect blocks. It will then push the records
3570 * onto an internal blocking queue. The worker thread will pull the records off
3571 * the queue, and actually write the data into the DMU. This way, the worker
3572 * thread doesn't have to wait for reads to complete, since everything it needs
3573 * (the indirect blocks) will be prefetched.
3575 * NB: callers *must* call dmu_recv_end() if this succeeds.
3578 dmu_recv_stream(dmu_recv_cookie_t
*drc
, vnode_t
*vp
, offset_t
*voffp
,
3579 int cleanup_fd
, uint64_t *action_handlep
)
3582 struct receive_arg
*ra
;
3583 struct receive_writer_arg
*rwa
;
3585 uint32_t payloadlen
;
3587 nvlist_t
*begin_nvl
= NULL
;
3589 ra
= kmem_zalloc(sizeof (*ra
), KM_SLEEP
);
3590 rwa
= kmem_zalloc(sizeof (*rwa
), KM_SLEEP
);
3592 ra
->byteswap
= drc
->drc_byteswap
;
3593 ra
->raw
= drc
->drc_raw
;
3594 ra
->cksum
= drc
->drc_cksum
;
3598 if (dsl_dataset_is_zapified(drc
->drc_ds
)) {
3599 (void) zap_lookup(drc
->drc_ds
->ds_dir
->dd_pool
->dp_meta_objset
,
3600 drc
->drc_ds
->ds_object
, DS_FIELD_RESUME_BYTES
,
3601 sizeof (ra
->bytes_read
), 1, &ra
->bytes_read
);
3604 objlist_create(&ra
->ignore_objlist
);
3606 /* these were verified in dmu_recv_begin */
3607 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc
->drc_drrb
->drr_versioninfo
), ==,
3609 ASSERT3U(drc
->drc_drrb
->drr_type
, <, DMU_OST_NUMTYPES
);
3612 * Open the objset we are modifying.
3614 VERIFY0(dmu_objset_from_ds(drc
->drc_ds
, &ra
->os
));
3616 ASSERT(dsl_dataset_phys(drc
->drc_ds
)->ds_flags
& DS_FLAG_INCONSISTENT
);
3618 featureflags
= DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
);
3619 ra
->featureflags
= featureflags
;
3621 /* embedded data is incompatible with encrypted datasets */
3622 if (ra
->os
->os_encrypted
&&
3623 (featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)) {
3624 err
= SET_ERROR(EINVAL
);
3628 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
3629 if (featureflags
& DMU_BACKUP_FEATURE_DEDUP
) {
3632 if (cleanup_fd
== -1) {
3633 err
= SET_ERROR(EBADF
);
3636 err
= zfs_onexit_fd_hold(cleanup_fd
, &minor
);
3642 if (*action_handlep
== 0) {
3643 rwa
->guid_to_ds_map
=
3644 kmem_alloc(sizeof (avl_tree_t
), KM_SLEEP
);
3645 avl_create(rwa
->guid_to_ds_map
, guid_compare
,
3646 sizeof (guid_map_entry_t
),
3647 offsetof(guid_map_entry_t
, avlnode
));
3648 err
= zfs_onexit_add_cb(minor
,
3649 free_guid_map_onexit
, rwa
->guid_to_ds_map
,
3654 err
= zfs_onexit_cb_data(minor
, *action_handlep
,
3655 (void **)&rwa
->guid_to_ds_map
);
3660 drc
->drc_guid_to_ds_map
= rwa
->guid_to_ds_map
;
3663 payloadlen
= drc
->drc_drr_begin
->drr_payloadlen
;
3665 if (payloadlen
!= 0)
3666 payload
= kmem_alloc(payloadlen
, KM_SLEEP
);
3668 err
= receive_read_payload_and_next_header(ra
, payloadlen
, payload
);
3670 if (payloadlen
!= 0)
3671 kmem_free(payload
, payloadlen
);
3674 if (payloadlen
!= 0) {
3675 err
= nvlist_unpack(payload
, payloadlen
, &begin_nvl
, KM_SLEEP
);
3676 kmem_free(payload
, payloadlen
);
3681 /* handle DSL encryption key payload */
3682 if (featureflags
& DMU_BACKUP_FEATURE_RAW
) {
3683 nvlist_t
*keynvl
= NULL
;
3685 ASSERT(ra
->os
->os_encrypted
);
3686 ASSERT(drc
->drc_raw
);
3688 err
= nvlist_lookup_nvlist(begin_nvl
, "crypt_keydata", &keynvl
);
3692 err
= dsl_crypto_recv_key(spa_name(ra
->os
->os_spa
),
3693 drc
->drc_ds
->ds_object
, drc
->drc_drrb
->drr_type
,
3699 if (featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
3700 err
= resume_check(ra
, begin_nvl
);
3705 (void) bqueue_init(&rwa
->q
, zfs_recv_queue_length
,
3706 offsetof(struct receive_record_arg
, node
));
3707 cv_init(&rwa
->cv
, NULL
, CV_DEFAULT
, NULL
);
3708 mutex_init(&rwa
->mutex
, NULL
, MUTEX_DEFAULT
, NULL
);
3710 rwa
->byteswap
= drc
->drc_byteswap
;
3711 rwa
->resumable
= drc
->drc_resumable
;
3712 rwa
->raw
= drc
->drc_raw
;
3714 (void) thread_create(NULL
, 0, receive_writer_thread
, rwa
, 0, curproc
,
3715 TS_RUN
, minclsyspri
);
3717 * We're reading rwa->err without locks, which is safe since we are the
3718 * only reader, and the worker thread is the only writer. It's ok if we
3719 * miss a write for an iteration or two of the loop, since the writer
3720 * thread will keep freeing records we send it until we send it an eos
3723 * We can leave this loop in 3 ways: First, if rwa->err is
3724 * non-zero. In that case, the writer thread will free the rrd we just
3725 * pushed. Second, if we're interrupted; in that case, either it's the
3726 * first loop and ra->rrd was never allocated, or it's later and ra->rrd
3727 * has been handed off to the writer thread who will free it. Finally,
3728 * if receive_read_record fails or we're at the end of the stream, then
3729 * we free ra->rrd and exit.
3731 while (rwa
->err
== 0) {
3732 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
3733 err
= SET_ERROR(EINTR
);
3737 ASSERT3P(ra
->rrd
, ==, NULL
);
3738 ra
->rrd
= ra
->next_rrd
;
3739 ra
->next_rrd
= NULL
;
3740 /* Allocates and loads header into ra->next_rrd */
3741 err
= receive_read_record(ra
);
3743 if (ra
->rrd
->header
.drr_type
== DRR_END
|| err
!= 0) {
3744 kmem_free(ra
->rrd
, sizeof (*ra
->rrd
));
3749 bqueue_enqueue(&rwa
->q
, ra
->rrd
,
3750 sizeof (struct receive_record_arg
) + ra
->rrd
->payload_size
);
3753 if (ra
->next_rrd
== NULL
)
3754 ra
->next_rrd
= kmem_zalloc(sizeof (*ra
->next_rrd
), KM_SLEEP
);
3755 ra
->next_rrd
->eos_marker
= B_TRUE
;
3756 bqueue_enqueue(&rwa
->q
, ra
->next_rrd
, 1);
3758 mutex_enter(&rwa
->mutex
);
3759 while (!rwa
->done
) {
3760 cv_wait(&rwa
->cv
, &rwa
->mutex
);
3762 mutex_exit(&rwa
->mutex
);
3765 * If we are receiving a full stream as a clone, all object IDs which
3766 * are greater than the maximum ID referenced in the stream are
3767 * by definition unused and must be freed.
3769 if (drc
->drc_clone
&& drc
->drc_drrb
->drr_fromguid
== 0) {
3770 uint64_t obj
= rwa
->max_object
+ 1;
3774 while (next_err
== 0) {
3775 free_err
= dmu_free_long_object(rwa
->os
, obj
);
3776 if (free_err
!= 0 && free_err
!= ENOENT
)
3779 next_err
= dmu_object_next(rwa
->os
, &obj
, FALSE
, 0);
3783 if (free_err
!= 0 && free_err
!= ENOENT
)
3785 else if (next_err
!= ESRCH
)
3790 cv_destroy(&rwa
->cv
);
3791 mutex_destroy(&rwa
->mutex
);
3792 bqueue_destroy(&rwa
->q
);
3797 nvlist_free(begin_nvl
);
3798 if ((featureflags
& DMU_BACKUP_FEATURE_DEDUP
) && (cleanup_fd
!= -1))
3799 zfs_onexit_fd_rele(cleanup_fd
);
3803 * Clean up references. If receive is not resumable,
3804 * destroy what we created, so we don't leave it in
3805 * the inconsistent state.
3807 dmu_recv_cleanup_ds(drc
);
3811 objlist_destroy(&ra
->ignore_objlist
);
3812 kmem_free(ra
, sizeof (*ra
));
3813 kmem_free(rwa
, sizeof (*rwa
));
3818 dmu_recv_end_check(void *arg
, dmu_tx_t
*tx
)
3820 dmu_recv_cookie_t
*drc
= arg
;
3821 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3824 ASSERT3P(drc
->drc_ds
->ds_owner
, ==, dmu_recv_tag
);
3826 if (!drc
->drc_newfs
) {
3827 dsl_dataset_t
*origin_head
;
3829 error
= dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
, &origin_head
);
3832 if (drc
->drc_force
) {
3834 * We will destroy any snapshots in tofs (i.e. before
3835 * origin_head) that are after the origin (which is
3836 * the snap before drc_ds, because drc_ds can not
3837 * have any snaps of its own).
3841 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3843 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3844 dsl_dataset_t
*snap
;
3845 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3849 if (snap
->ds_dir
!= origin_head
->ds_dir
)
3850 error
= SET_ERROR(EINVAL
);
3852 error
= dsl_destroy_snapshot_check_impl(
3855 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3856 dsl_dataset_rele(snap
, FTAG
);
3861 dsl_dataset_rele(origin_head
, FTAG
);
3865 error
= dsl_dataset_clone_swap_check_impl(drc
->drc_ds
,
3866 origin_head
, drc
->drc_force
, drc
->drc_owner
, tx
);
3868 dsl_dataset_rele(origin_head
, FTAG
);
3871 error
= dsl_dataset_snapshot_check_impl(origin_head
,
3872 drc
->drc_tosnap
, tx
, B_TRUE
, 1, drc
->drc_cred
);
3873 dsl_dataset_rele(origin_head
, FTAG
);
3877 error
= dsl_destroy_head_check_impl(drc
->drc_ds
, 1);
3879 error
= dsl_dataset_snapshot_check_impl(drc
->drc_ds
,
3880 drc
->drc_tosnap
, tx
, B_TRUE
, 1, drc
->drc_cred
);
3886 dmu_recv_end_sync(void *arg
, dmu_tx_t
*tx
)
3888 dmu_recv_cookie_t
*drc
= arg
;
3889 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3890 boolean_t encrypted
= drc
->drc_ds
->ds_dir
->dd_crypto_obj
!= 0;
3892 spa_history_log_internal_ds(drc
->drc_ds
, "finish receiving",
3893 tx
, "snap=%s", drc
->drc_tosnap
);
3895 if (!drc
->drc_newfs
) {
3896 dsl_dataset_t
*origin_head
;
3898 VERIFY0(dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
,
3901 if (drc
->drc_force
) {
3903 * Destroy any snapshots of drc_tofs (origin_head)
3904 * after the origin (the snap before drc_ds).
3908 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3910 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3911 dsl_dataset_t
*snap
;
3912 VERIFY0(dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3914 ASSERT3P(snap
->ds_dir
, ==, origin_head
->ds_dir
);
3915 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3916 dsl_destroy_snapshot_sync_impl(snap
,
3918 dsl_dataset_rele(snap
, FTAG
);
3921 VERIFY3P(drc
->drc_ds
->ds_prev
, ==,
3922 origin_head
->ds_prev
);
3924 dsl_dataset_clone_swap_sync_impl(drc
->drc_ds
,
3926 dsl_dataset_snapshot_sync_impl(origin_head
,
3927 drc
->drc_tosnap
, tx
);
3929 /* set snapshot's creation time and guid */
3930 dmu_buf_will_dirty(origin_head
->ds_prev
->ds_dbuf
, tx
);
3931 dsl_dataset_phys(origin_head
->ds_prev
)->ds_creation_time
=
3932 drc
->drc_drrb
->drr_creation_time
;
3933 dsl_dataset_phys(origin_head
->ds_prev
)->ds_guid
=
3934 drc
->drc_drrb
->drr_toguid
;
3935 dsl_dataset_phys(origin_head
->ds_prev
)->ds_flags
&=
3936 ~DS_FLAG_INCONSISTENT
;
3938 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
3939 dsl_dataset_phys(origin_head
)->ds_flags
&=
3940 ~DS_FLAG_INCONSISTENT
;
3942 drc
->drc_newsnapobj
=
3943 dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3945 dsl_dataset_rele(origin_head
, FTAG
);
3946 dsl_destroy_head_sync_impl(drc
->drc_ds
, tx
);
3948 if (drc
->drc_owner
!= NULL
)
3949 VERIFY3P(origin_head
->ds_owner
, ==, drc
->drc_owner
);
3951 dsl_dataset_t
*ds
= drc
->drc_ds
;
3953 dsl_dataset_snapshot_sync_impl(ds
, drc
->drc_tosnap
, tx
);
3955 /* set snapshot's creation time and guid */
3956 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
3957 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_time
=
3958 drc
->drc_drrb
->drr_creation_time
;
3959 dsl_dataset_phys(ds
->ds_prev
)->ds_guid
=
3960 drc
->drc_drrb
->drr_toguid
;
3961 dsl_dataset_phys(ds
->ds_prev
)->ds_flags
&=
3962 ~DS_FLAG_INCONSISTENT
;
3964 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3965 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
3966 if (dsl_dataset_has_resume_receive_state(ds
)) {
3967 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3968 DS_FIELD_RESUME_FROMGUID
, tx
);
3969 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3970 DS_FIELD_RESUME_OBJECT
, tx
);
3971 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3972 DS_FIELD_RESUME_OFFSET
, tx
);
3973 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3974 DS_FIELD_RESUME_BYTES
, tx
);
3975 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3976 DS_FIELD_RESUME_TOGUID
, tx
);
3977 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3978 DS_FIELD_RESUME_TONAME
, tx
);
3980 drc
->drc_newsnapobj
=
3981 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
;
3983 zvol_create_minors(dp
->dp_spa
, drc
->drc_tofs
, B_TRUE
);
3986 * Release the hold from dmu_recv_begin. This must be done before
3987 * we return to open context, so that when we free the dataset's dnode
3988 * we can evict its bonus buffer. Since the dataset may be destroyed
3989 * at this point (and therefore won't have a valid pointer to the spa)
3990 * we release the key mapping manually here while we do have a valid
3991 * pointer, if it exists.
3993 if (!drc
->drc_raw
&& encrypted
) {
3994 (void) spa_keystore_remove_mapping(dmu_tx_pool(tx
)->dp_spa
,
3995 drc
->drc_ds
->ds_object
, drc
->drc_ds
);
3997 dsl_dataset_disown(drc
->drc_ds
, 0, dmu_recv_tag
);
4002 add_ds_to_guidmap(const char *name
, avl_tree_t
*guid_map
, uint64_t snapobj
,
4006 dsl_dataset_t
*snapds
;
4007 guid_map_entry_t
*gmep
;
4008 ds_hold_flags_t dsflags
= (raw
) ? 0 : DS_HOLD_FLAG_DECRYPT
;
4011 ASSERT(guid_map
!= NULL
);
4013 err
= dsl_pool_hold(name
, FTAG
, &dp
);
4016 gmep
= kmem_alloc(sizeof (*gmep
), KM_SLEEP
);
4017 err
= dsl_dataset_hold_obj_flags(dp
, snapobj
, dsflags
, gmep
, &snapds
);
4019 gmep
->guid
= dsl_dataset_phys(snapds
)->ds_guid
;
4021 gmep
->gme_ds
= snapds
;
4022 avl_add(guid_map
, gmep
);
4023 dsl_dataset_long_hold(snapds
, gmep
);
4025 kmem_free(gmep
, sizeof (*gmep
));
4028 dsl_pool_rele(dp
, FTAG
);
4032 static int dmu_recv_end_modified_blocks
= 3;
4035 dmu_recv_existing_end(dmu_recv_cookie_t
*drc
)
4039 * We will be destroying the ds; make sure its origin is unmounted if
4042 char name
[ZFS_MAX_DATASET_NAME_LEN
];
4043 dsl_dataset_name(drc
->drc_ds
, name
);
4044 zfs_destroy_unmount_origin(name
);
4047 return (dsl_sync_task(drc
->drc_tofs
,
4048 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
4049 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
4053 dmu_recv_new_end(dmu_recv_cookie_t
*drc
)
4055 return (dsl_sync_task(drc
->drc_tofs
,
4056 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
4057 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
));
4061 dmu_recv_end(dmu_recv_cookie_t
*drc
, void *owner
)
4065 drc
->drc_owner
= owner
;
4068 error
= dmu_recv_new_end(drc
);
4070 error
= dmu_recv_existing_end(drc
);
4073 dmu_recv_cleanup_ds(drc
);
4074 } else if (drc
->drc_guid_to_ds_map
!= NULL
) {
4075 (void) add_ds_to_guidmap(drc
->drc_tofs
, drc
->drc_guid_to_ds_map
,
4076 drc
->drc_newsnapobj
, drc
->drc_raw
);
4082 * Return TRUE if this objset is currently being received into.
4085 dmu_objset_is_receiving(objset_t
*os
)
4087 return (os
->os_dsl_dataset
!= NULL
&&
4088 os
->os_dsl_dataset
->ds_owner
== dmu_recv_tag
);
4091 #if defined(_KERNEL)
4092 module_param(zfs_send_corrupt_data
, int, 0644);
4093 MODULE_PARM_DESC(zfs_send_corrupt_data
, "Allow sending corrupt data");