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 this (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).
118 * Removing the assertions could be rolled into a new feature that uses
119 * data that isn't 8-byte aligned; if the assertions were removed, a
120 * feature flag would have to be added.
123 ASSERT0(dbi
->dbi_len
% 8);
125 dsp
->dsa_err
= vn_rdwr(UIO_WRITE
, dsp
->dsa_vp
,
126 (caddr_t
)dbi
->dbi_buf
, dbi
->dbi_len
,
127 0, UIO_SYSSPACE
, FAPPEND
, RLIM64_INFINITY
, CRED(), &resid
);
129 mutex_enter(&ds
->ds_sendstream_lock
);
130 *dsp
->dsa_off
+= dbi
->dbi_len
;
131 mutex_exit(&ds
->ds_sendstream_lock
);
135 dump_bytes(dmu_sendarg_t
*dsp
, void *buf
, int len
)
143 #if defined(HAVE_LARGE_STACKS)
147 * The vn_rdwr() call is performed in a taskq to ensure that there is
148 * always enough stack space to write safely to the target filesystem.
149 * The ZIO_TYPE_FREE threads are used because there can be a lot of
150 * them and they are used in vdev_file.c for a similar purpose.
152 spa_taskq_dispatch_sync(dmu_objset_spa(dsp
->dsa_os
), ZIO_TYPE_FREE
,
153 ZIO_TASKQ_ISSUE
, dump_bytes_cb
, &dbi
, TQ_SLEEP
);
154 #endif /* HAVE_LARGE_STACKS */
156 return (dsp
->dsa_err
);
160 * For all record types except BEGIN, fill in the checksum (overlaid in
161 * drr_u.drr_checksum.drr_checksum). The checksum verifies everything
162 * up to the start of the checksum itself.
165 dump_record(dmu_sendarg_t
*dsp
, void *payload
, int payload_len
)
167 ASSERT3U(offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
168 ==, sizeof (dmu_replay_record_t
) - sizeof (zio_cksum_t
));
169 fletcher_4_incremental_native(dsp
->dsa_drr
,
170 offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
172 if (dsp
->dsa_drr
->drr_type
!= DRR_BEGIN
) {
173 ASSERT(ZIO_CHECKSUM_IS_ZERO(&dsp
->dsa_drr
->drr_u
.
174 drr_checksum
.drr_checksum
));
175 dsp
->dsa_drr
->drr_u
.drr_checksum
.drr_checksum
= dsp
->dsa_zc
;
177 fletcher_4_incremental_native(&dsp
->dsa_drr
->
178 drr_u
.drr_checksum
.drr_checksum
,
179 sizeof (zio_cksum_t
), &dsp
->dsa_zc
);
180 if (dump_bytes(dsp
, dsp
->dsa_drr
, sizeof (dmu_replay_record_t
)) != 0)
181 return (SET_ERROR(EINTR
));
182 if (payload_len
!= 0) {
183 fletcher_4_incremental_native(payload
, payload_len
,
185 if (dump_bytes(dsp
, payload
, payload_len
) != 0)
186 return (SET_ERROR(EINTR
));
192 * Fill in the drr_free struct, or perform aggregation if the previous record is
193 * also a free record, and the two are adjacent.
195 * Note that we send free records even for a full send, because we want to be
196 * able to receive a full send as a clone, which requires a list of all the free
197 * and freeobject records that were generated on the source.
200 dump_free(dmu_sendarg_t
*dsp
, uint64_t object
, uint64_t offset
,
203 struct drr_free
*drrf
= &(dsp
->dsa_drr
->drr_u
.drr_free
);
206 * When we receive a free record, dbuf_free_range() assumes
207 * that the receiving system doesn't have any dbufs in the range
208 * being freed. This is always true because there is a one-record
209 * constraint: we only send one WRITE record for any given
210 * object,offset. We know that the one-record constraint is
211 * true because we always send data in increasing order by
214 * If the increasing-order constraint ever changes, we should find
215 * another way to assert that the one-record constraint is still
218 ASSERT(object
> dsp
->dsa_last_data_object
||
219 (object
== dsp
->dsa_last_data_object
&&
220 offset
> dsp
->dsa_last_data_offset
));
222 if (length
!= -1ULL && offset
+ length
< offset
)
226 * If there is a pending op, but it's not PENDING_FREE, push it out,
227 * since free block aggregation can only be done for blocks of the
228 * same type (i.e., DRR_FREE records can only be aggregated with
229 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
230 * aggregated with other DRR_FREEOBJECTS records.
232 if (dsp
->dsa_pending_op
!= PENDING_NONE
&&
233 dsp
->dsa_pending_op
!= PENDING_FREE
) {
234 if (dump_record(dsp
, NULL
, 0) != 0)
235 return (SET_ERROR(EINTR
));
236 dsp
->dsa_pending_op
= PENDING_NONE
;
239 if (dsp
->dsa_pending_op
== PENDING_FREE
) {
241 * There should never be a PENDING_FREE if length is -1
242 * (because dump_dnode is the only place where this
243 * function is called with a -1, and only after flushing
244 * any pending record).
246 ASSERT(length
!= -1ULL);
248 * Check to see whether this free block can be aggregated
251 if (drrf
->drr_object
== object
&& drrf
->drr_offset
+
252 drrf
->drr_length
== offset
) {
253 drrf
->drr_length
+= length
;
256 /* not a continuation. Push out pending record */
257 if (dump_record(dsp
, NULL
, 0) != 0)
258 return (SET_ERROR(EINTR
));
259 dsp
->dsa_pending_op
= PENDING_NONE
;
262 /* create a FREE record and make it pending */
263 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
264 dsp
->dsa_drr
->drr_type
= DRR_FREE
;
265 drrf
->drr_object
= object
;
266 drrf
->drr_offset
= offset
;
267 drrf
->drr_length
= length
;
268 drrf
->drr_toguid
= dsp
->dsa_toguid
;
269 if (length
== -1ULL) {
270 if (dump_record(dsp
, NULL
, 0) != 0)
271 return (SET_ERROR(EINTR
));
273 dsp
->dsa_pending_op
= PENDING_FREE
;
280 dump_write(dmu_sendarg_t
*dsp
, dmu_object_type_t type
,
281 uint64_t object
, uint64_t offset
, int lsize
, int psize
, const blkptr_t
*bp
,
284 uint64_t payload_size
;
285 struct drr_write
*drrw
= &(dsp
->dsa_drr
->drr_u
.drr_write
);
288 * We send data in increasing object, offset order.
289 * See comment in dump_free() for details.
291 ASSERT(object
> dsp
->dsa_last_data_object
||
292 (object
== dsp
->dsa_last_data_object
&&
293 offset
> dsp
->dsa_last_data_offset
));
294 dsp
->dsa_last_data_object
= object
;
295 dsp
->dsa_last_data_offset
= offset
+ lsize
- 1;
298 * If there is any kind of pending aggregation (currently either
299 * a grouping of free objects or free blocks), push it out to
300 * the stream, since aggregation can't be done across operations
301 * of different types.
303 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
304 if (dump_record(dsp
, NULL
, 0) != 0)
305 return (SET_ERROR(EINTR
));
306 dsp
->dsa_pending_op
= PENDING_NONE
;
308 /* write a WRITE record */
309 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
310 dsp
->dsa_drr
->drr_type
= DRR_WRITE
;
311 drrw
->drr_object
= object
;
312 drrw
->drr_type
= type
;
313 drrw
->drr_offset
= offset
;
314 drrw
->drr_toguid
= dsp
->dsa_toguid
;
315 drrw
->drr_logical_size
= lsize
;
317 /* only set the compression fields if the buf is compressed */
318 if (lsize
!= psize
) {
319 ASSERT(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
);
320 ASSERT(!BP_IS_EMBEDDED(bp
));
321 ASSERT(!BP_SHOULD_BYTESWAP(bp
));
322 ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)));
323 ASSERT3U(BP_GET_COMPRESS(bp
), !=, ZIO_COMPRESS_OFF
);
324 ASSERT3S(psize
, >, 0);
325 ASSERT3S(lsize
, >=, psize
);
327 drrw
->drr_compressiontype
= BP_GET_COMPRESS(bp
);
328 drrw
->drr_compressed_size
= psize
;
329 payload_size
= drrw
->drr_compressed_size
;
331 payload_size
= drrw
->drr_logical_size
;
334 if (bp
== NULL
|| BP_IS_EMBEDDED(bp
)) {
336 * There's no pre-computed checksum for partial-block
337 * writes or embedded BP's, so (like
338 * fletcher4-checkummed blocks) userland will have to
339 * compute a dedup-capable checksum itself.
341 drrw
->drr_checksumtype
= ZIO_CHECKSUM_OFF
;
343 drrw
->drr_checksumtype
= BP_GET_CHECKSUM(bp
);
344 if (zio_checksum_table
[drrw
->drr_checksumtype
].ci_dedup
)
345 drrw
->drr_checksumflags
|= DRR_CHECKSUM_DEDUP
;
346 DDK_SET_LSIZE(&drrw
->drr_key
, BP_GET_LSIZE(bp
));
347 DDK_SET_PSIZE(&drrw
->drr_key
, BP_GET_PSIZE(bp
));
348 DDK_SET_COMPRESS(&drrw
->drr_key
, BP_GET_COMPRESS(bp
));
349 drrw
->drr_key
.ddk_cksum
= bp
->blk_cksum
;
352 if (dump_record(dsp
, data
, payload_size
) != 0)
353 return (SET_ERROR(EINTR
));
358 dump_write_embedded(dmu_sendarg_t
*dsp
, uint64_t object
, uint64_t offset
,
359 int blksz
, const blkptr_t
*bp
)
361 char buf
[BPE_PAYLOAD_SIZE
];
362 struct drr_write_embedded
*drrw
=
363 &(dsp
->dsa_drr
->drr_u
.drr_write_embedded
);
365 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
366 if (dump_record(dsp
, NULL
, 0) != 0)
368 dsp
->dsa_pending_op
= PENDING_NONE
;
371 ASSERT(BP_IS_EMBEDDED(bp
));
373 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
374 dsp
->dsa_drr
->drr_type
= DRR_WRITE_EMBEDDED
;
375 drrw
->drr_object
= object
;
376 drrw
->drr_offset
= offset
;
377 drrw
->drr_length
= blksz
;
378 drrw
->drr_toguid
= dsp
->dsa_toguid
;
379 drrw
->drr_compression
= BP_GET_COMPRESS(bp
);
380 drrw
->drr_etype
= BPE_GET_ETYPE(bp
);
381 drrw
->drr_lsize
= BPE_GET_LSIZE(bp
);
382 drrw
->drr_psize
= BPE_GET_PSIZE(bp
);
384 decode_embedded_bp_compressed(bp
, buf
);
386 if (dump_record(dsp
, buf
, P2ROUNDUP(drrw
->drr_psize
, 8)) != 0)
392 dump_spill(dmu_sendarg_t
*dsp
, uint64_t object
, int blksz
, void *data
)
394 struct drr_spill
*drrs
= &(dsp
->dsa_drr
->drr_u
.drr_spill
);
396 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
397 if (dump_record(dsp
, NULL
, 0) != 0)
398 return (SET_ERROR(EINTR
));
399 dsp
->dsa_pending_op
= PENDING_NONE
;
402 /* write a SPILL record */
403 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
404 dsp
->dsa_drr
->drr_type
= DRR_SPILL
;
405 drrs
->drr_object
= object
;
406 drrs
->drr_length
= blksz
;
407 drrs
->drr_toguid
= dsp
->dsa_toguid
;
409 if (dump_record(dsp
, data
, blksz
) != 0)
410 return (SET_ERROR(EINTR
));
415 dump_freeobjects(dmu_sendarg_t
*dsp
, uint64_t firstobj
, uint64_t numobjs
)
417 struct drr_freeobjects
*drrfo
= &(dsp
->dsa_drr
->drr_u
.drr_freeobjects
);
420 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
421 * push it out, since free block aggregation can only be done for
422 * blocks of the same type (i.e., DRR_FREE records can only be
423 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
424 * can only be aggregated with other DRR_FREEOBJECTS records.
426 if (dsp
->dsa_pending_op
!= PENDING_NONE
&&
427 dsp
->dsa_pending_op
!= PENDING_FREEOBJECTS
) {
428 if (dump_record(dsp
, NULL
, 0) != 0)
429 return (SET_ERROR(EINTR
));
430 dsp
->dsa_pending_op
= PENDING_NONE
;
432 if (dsp
->dsa_pending_op
== PENDING_FREEOBJECTS
) {
434 * See whether this free object array can be aggregated
437 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
== firstobj
) {
438 drrfo
->drr_numobjs
+= numobjs
;
441 /* can't be aggregated. Push out pending record */
442 if (dump_record(dsp
, NULL
, 0) != 0)
443 return (SET_ERROR(EINTR
));
444 dsp
->dsa_pending_op
= PENDING_NONE
;
448 /* write a FREEOBJECTS record */
449 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
450 dsp
->dsa_drr
->drr_type
= DRR_FREEOBJECTS
;
451 drrfo
->drr_firstobj
= firstobj
;
452 drrfo
->drr_numobjs
= numobjs
;
453 drrfo
->drr_toguid
= dsp
->dsa_toguid
;
455 dsp
->dsa_pending_op
= PENDING_FREEOBJECTS
;
461 dump_dnode(dmu_sendarg_t
*dsp
, uint64_t object
, dnode_phys_t
*dnp
)
463 struct drr_object
*drro
= &(dsp
->dsa_drr
->drr_u
.drr_object
);
465 if (object
< dsp
->dsa_resume_object
) {
467 * Note: when resuming, we will visit all the dnodes in
468 * the block of dnodes that we are resuming from. In
469 * this case it's unnecessary to send the dnodes prior to
470 * the one we are resuming from. We should be at most one
471 * block's worth of dnodes behind the resume point.
473 ASSERT3U(dsp
->dsa_resume_object
- object
, <,
474 1 << (DNODE_BLOCK_SHIFT
- DNODE_SHIFT
));
478 if (dnp
== NULL
|| dnp
->dn_type
== DMU_OT_NONE
)
479 return (dump_freeobjects(dsp
, object
, 1));
481 if (dsp
->dsa_pending_op
!= PENDING_NONE
) {
482 if (dump_record(dsp
, NULL
, 0) != 0)
483 return (SET_ERROR(EINTR
));
484 dsp
->dsa_pending_op
= PENDING_NONE
;
487 /* write an OBJECT record */
488 bzero(dsp
->dsa_drr
, sizeof (dmu_replay_record_t
));
489 dsp
->dsa_drr
->drr_type
= DRR_OBJECT
;
490 drro
->drr_object
= object
;
491 drro
->drr_type
= dnp
->dn_type
;
492 drro
->drr_bonustype
= dnp
->dn_bonustype
;
493 drro
->drr_blksz
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
494 drro
->drr_bonuslen
= dnp
->dn_bonuslen
;
495 drro
->drr_dn_slots
= dnp
->dn_extra_slots
+ 1;
496 drro
->drr_checksumtype
= dnp
->dn_checksum
;
497 drro
->drr_compress
= dnp
->dn_compress
;
498 drro
->drr_toguid
= dsp
->dsa_toguid
;
500 if (!(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
501 drro
->drr_blksz
> SPA_OLD_MAXBLOCKSIZE
)
502 drro
->drr_blksz
= SPA_OLD_MAXBLOCKSIZE
;
504 if (dump_record(dsp
, DN_BONUS(dnp
),
505 P2ROUNDUP(dnp
->dn_bonuslen
, 8)) != 0) {
506 return (SET_ERROR(EINTR
));
509 /* Free anything past the end of the file. */
510 if (dump_free(dsp
, object
, (dnp
->dn_maxblkid
+ 1) *
511 (dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
), -1ULL) != 0)
512 return (SET_ERROR(EINTR
));
513 if (dsp
->dsa_err
!= 0)
514 return (SET_ERROR(EINTR
));
519 backup_do_embed(dmu_sendarg_t
*dsp
, const blkptr_t
*bp
)
521 if (!BP_IS_EMBEDDED(bp
))
525 * Compression function must be legacy, or explicitly enabled.
527 if ((BP_GET_COMPRESS(bp
) >= ZIO_COMPRESS_LEGACY_FUNCTIONS
&&
528 !(dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_LZ4
)))
532 * Embed type must be explicitly enabled.
534 switch (BPE_GET_ETYPE(bp
)) {
535 case BP_EMBEDDED_TYPE_DATA
:
536 if (dsp
->dsa_featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
)
546 * This is the callback function to traverse_dataset that acts as the worker
547 * thread for dmu_send_impl.
551 send_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
552 const zbookmark_phys_t
*zb
, const struct dnode_phys
*dnp
, void *arg
)
554 struct send_thread_arg
*sta
= arg
;
555 struct send_block_record
*record
;
556 uint64_t record_size
;
559 ASSERT(zb
->zb_object
== DMU_META_DNODE_OBJECT
||
560 zb
->zb_object
>= sta
->resume
.zb_object
);
563 return (SET_ERROR(EINTR
));
566 ASSERT3U(zb
->zb_level
, ==, ZB_DNODE_LEVEL
);
568 } else if (zb
->zb_level
< 0) {
572 record
= kmem_zalloc(sizeof (struct send_block_record
), KM_SLEEP
);
573 record
->eos_marker
= B_FALSE
;
576 record
->indblkshift
= dnp
->dn_indblkshift
;
577 record
->datablkszsec
= dnp
->dn_datablkszsec
;
578 record_size
= dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
;
579 bqueue_enqueue(&sta
->q
, record
, record_size
);
585 * This function kicks off the traverse_dataset. It also handles setting the
586 * error code of the thread in case something goes wrong, and pushes the End of
587 * Stream record when the traverse_dataset call has finished. If there is no
588 * dataset to traverse, the thread immediately pushes End of Stream marker.
591 send_traverse_thread(void *arg
)
593 struct send_thread_arg
*st_arg
= arg
;
595 struct send_block_record
*data
;
596 fstrans_cookie_t cookie
= spl_fstrans_mark();
598 if (st_arg
->ds
!= NULL
) {
599 err
= traverse_dataset_resume(st_arg
->ds
,
600 st_arg
->fromtxg
, &st_arg
->resume
,
601 st_arg
->flags
, send_cb
, st_arg
);
604 st_arg
->error_code
= err
;
606 data
= kmem_zalloc(sizeof (*data
), KM_SLEEP
);
607 data
->eos_marker
= B_TRUE
;
608 bqueue_enqueue(&st_arg
->q
, data
, 1);
609 spl_fstrans_unmark(cookie
);
613 * This function actually handles figuring out what kind of record needs to be
614 * dumped, reading the data (which has hopefully been prefetched), and calling
615 * the appropriate helper function.
618 do_dump(dmu_sendarg_t
*dsa
, struct send_block_record
*data
)
620 dsl_dataset_t
*ds
= dmu_objset_ds(dsa
->dsa_os
);
621 const blkptr_t
*bp
= &data
->bp
;
622 const zbookmark_phys_t
*zb
= &data
->zb
;
623 uint8_t indblkshift
= data
->indblkshift
;
624 uint16_t dblkszsec
= data
->datablkszsec
;
625 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
626 dmu_object_type_t type
= bp
? BP_GET_TYPE(bp
) : DMU_OT_NONE
;
630 ASSERT3U(zb
->zb_level
, >=, 0);
632 ASSERT(zb
->zb_object
== DMU_META_DNODE_OBJECT
||
633 zb
->zb_object
>= dsa
->dsa_resume_object
);
635 if (zb
->zb_object
!= DMU_META_DNODE_OBJECT
&&
636 DMU_OBJECT_IS_SPECIAL(zb
->zb_object
)) {
638 } else if (BP_IS_HOLE(bp
) &&
639 zb
->zb_object
== DMU_META_DNODE_OBJECT
) {
640 uint64_t span
= BP_SPAN(dblkszsec
, indblkshift
, zb
->zb_level
);
641 uint64_t dnobj
= (zb
->zb_blkid
* span
) >> DNODE_SHIFT
;
642 err
= dump_freeobjects(dsa
, dnobj
, span
>> DNODE_SHIFT
);
643 } else if (BP_IS_HOLE(bp
)) {
644 uint64_t span
= BP_SPAN(dblkszsec
, indblkshift
, zb
->zb_level
);
645 uint64_t offset
= zb
->zb_blkid
* span
;
646 err
= dump_free(dsa
, zb
->zb_object
, offset
, span
);
647 } else if (zb
->zb_level
> 0 || type
== DMU_OT_OBJSET
) {
649 } else if (type
== DMU_OT_DNODE
) {
651 int epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
652 arc_flags_t aflags
= ARC_FLAG_WAIT
;
656 ASSERT0(zb
->zb_level
);
658 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
659 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
,
661 return (SET_ERROR(EIO
));
664 dnobj
= zb
->zb_blkid
* epb
;
665 for (i
= 0; i
< epb
; i
+= blk
[i
].dn_extra_slots
+ 1) {
666 err
= dump_dnode(dsa
, dnobj
+ i
, blk
+ i
);
670 arc_buf_destroy(abuf
, &abuf
);
671 } else if (type
== DMU_OT_SA
) {
672 arc_flags_t aflags
= ARC_FLAG_WAIT
;
674 int blksz
= BP_GET_LSIZE(bp
);
676 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
677 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
,
679 return (SET_ERROR(EIO
));
681 err
= dump_spill(dsa
, zb
->zb_object
, blksz
, abuf
->b_data
);
682 arc_buf_destroy(abuf
, &abuf
);
683 } else if (backup_do_embed(dsa
, bp
)) {
684 /* it's an embedded level-0 block of a regular object */
685 int blksz
= dblkszsec
<< SPA_MINBLOCKSHIFT
;
686 ASSERT0(zb
->zb_level
);
687 err
= dump_write_embedded(dsa
, zb
->zb_object
,
688 zb
->zb_blkid
* blksz
, blksz
, bp
);
690 /* it's a level-0 block of a regular object */
691 arc_flags_t aflags
= ARC_FLAG_WAIT
;
693 int blksz
= dblkszsec
<< SPA_MINBLOCKSHIFT
;
695 enum zio_flag zioflags
= ZIO_FLAG_CANFAIL
;
698 * If we have large blocks stored on disk but the send flags
699 * don't allow us to send large blocks, we split the data from
700 * the arc buf into chunks.
702 boolean_t split_large_blocks
=
703 data
->datablkszsec
> SPA_OLD_MAXBLOCKSIZE
&&
704 !(dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
);
706 * We should only request compressed data from the ARC if all
707 * the following are true:
708 * - stream compression was requested
709 * - we aren't splitting large blocks into smaller chunks
710 * - the data won't need to be byteswapped before sending
711 * - this isn't an embedded block
712 * - this isn't metadata (if receiving on a different endian
713 * system it can be byteswapped more easily)
715 boolean_t request_compressed
=
716 (dsa
->dsa_featureflags
& DMU_BACKUP_FEATURE_COMPRESSED
) &&
717 !split_large_blocks
&& !BP_SHOULD_BYTESWAP(bp
) &&
718 !BP_IS_EMBEDDED(bp
) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp
));
720 ASSERT0(zb
->zb_level
);
721 ASSERT(zb
->zb_object
> dsa
->dsa_resume_object
||
722 (zb
->zb_object
== dsa
->dsa_resume_object
&&
723 zb
->zb_blkid
* blksz
>= dsa
->dsa_resume_offset
));
725 if (request_compressed
)
726 zioflags
|= ZIO_FLAG_RAW
;
728 if (arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &abuf
,
729 ZIO_PRIORITY_ASYNC_READ
, zioflags
,
731 if (zfs_send_corrupt_data
) {
733 /* Send a block filled with 0x"zfs badd bloc" */
734 abuf
= arc_alloc_buf(spa
, &abuf
, ARC_BUFC_DATA
,
736 for (ptr
= abuf
->b_data
;
737 (char *)ptr
< (char *)abuf
->b_data
+ blksz
;
739 *ptr
= 0x2f5baddb10cULL
;
741 return (SET_ERROR(EIO
));
745 offset
= zb
->zb_blkid
* blksz
;
747 if (split_large_blocks
) {
748 char *buf
= abuf
->b_data
;
749 ASSERT3U(arc_get_compression(abuf
), ==,
751 while (blksz
> 0 && err
== 0) {
752 int n
= MIN(blksz
, SPA_OLD_MAXBLOCKSIZE
);
753 err
= dump_write(dsa
, type
, zb
->zb_object
,
754 offset
, n
, n
, NULL
, buf
);
760 err
= dump_write(dsa
, type
, zb
->zb_object
, offset
,
761 blksz
, arc_buf_size(abuf
), bp
,
764 arc_buf_destroy(abuf
, &abuf
);
767 ASSERT(err
== 0 || err
== EINTR
);
772 * Pop the new data off the queue, and free the old data.
774 static struct send_block_record
*
775 get_next_record(bqueue_t
*bq
, struct send_block_record
*data
)
777 struct send_block_record
*tmp
= bqueue_dequeue(bq
);
778 kmem_free(data
, sizeof (*data
));
783 * Actually do the bulk of the work in a zfs send.
785 * Note: Releases dp using the specified tag.
788 dmu_send_impl(void *tag
, dsl_pool_t
*dp
, dsl_dataset_t
*to_ds
,
789 zfs_bookmark_phys_t
*ancestor_zb
, boolean_t is_clone
,
790 boolean_t embedok
, boolean_t large_block_ok
, boolean_t compressok
,
791 int outfd
, uint64_t resumeobj
, uint64_t resumeoff
,
792 vnode_t
*vp
, offset_t
*off
)
795 dmu_replay_record_t
*drr
;
798 uint64_t fromtxg
= 0;
799 uint64_t featureflags
= 0;
800 struct send_thread_arg to_arg
;
801 void *payload
= NULL
;
802 size_t payload_len
= 0;
803 struct send_block_record
*to_data
;
805 err
= dmu_objset_from_ds(to_ds
, &os
);
807 dsl_pool_rele(dp
, tag
);
811 drr
= kmem_zalloc(sizeof (dmu_replay_record_t
), KM_SLEEP
);
812 drr
->drr_type
= DRR_BEGIN
;
813 drr
->drr_u
.drr_begin
.drr_magic
= DMU_BACKUP_MAGIC
;
814 DMU_SET_STREAM_HDRTYPE(drr
->drr_u
.drr_begin
.drr_versioninfo
,
817 bzero(&to_arg
, sizeof (to_arg
));
820 if (dmu_objset_type(os
) == DMU_OST_ZFS
) {
822 if (zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &version
) != 0) {
823 kmem_free(drr
, sizeof (dmu_replay_record_t
));
824 dsl_pool_rele(dp
, tag
);
825 return (SET_ERROR(EINVAL
));
827 if (version
>= ZPL_VERSION_SA
) {
828 featureflags
|= DMU_BACKUP_FEATURE_SA_SPILL
;
833 if (large_block_ok
&& to_ds
->ds_feature_inuse
[SPA_FEATURE_LARGE_BLOCKS
])
834 featureflags
|= DMU_BACKUP_FEATURE_LARGE_BLOCKS
;
835 if (to_ds
->ds_feature_inuse
[SPA_FEATURE_LARGE_DNODE
])
836 featureflags
|= DMU_BACKUP_FEATURE_LARGE_DNODE
;
838 spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
)) {
839 featureflags
|= DMU_BACKUP_FEATURE_EMBED_DATA
;
842 featureflags
|= DMU_BACKUP_FEATURE_COMPRESSED
;
845 (DMU_BACKUP_FEATURE_EMBED_DATA
| DMU_BACKUP_FEATURE_COMPRESSED
)) !=
846 0 && spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
)) {
847 featureflags
|= DMU_BACKUP_FEATURE_LZ4
;
850 if (resumeobj
!= 0 || resumeoff
!= 0) {
851 featureflags
|= DMU_BACKUP_FEATURE_RESUMING
;
854 DMU_SET_FEATUREFLAGS(drr
->drr_u
.drr_begin
.drr_versioninfo
,
857 drr
->drr_u
.drr_begin
.drr_creation_time
=
858 dsl_dataset_phys(to_ds
)->ds_creation_time
;
859 drr
->drr_u
.drr_begin
.drr_type
= dmu_objset_type(os
);
861 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CLONE
;
862 drr
->drr_u
.drr_begin
.drr_toguid
= dsl_dataset_phys(to_ds
)->ds_guid
;
863 if (dsl_dataset_phys(to_ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
864 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_CI_DATA
;
865 if (zfs_send_set_freerecords_bit
)
866 drr
->drr_u
.drr_begin
.drr_flags
|= DRR_FLAG_FREERECORDS
;
868 if (ancestor_zb
!= NULL
) {
869 drr
->drr_u
.drr_begin
.drr_fromguid
=
870 ancestor_zb
->zbm_guid
;
871 fromtxg
= ancestor_zb
->zbm_creation_txg
;
873 dsl_dataset_name(to_ds
, drr
->drr_u
.drr_begin
.drr_toname
);
874 if (!to_ds
->ds_is_snapshot
) {
875 (void) strlcat(drr
->drr_u
.drr_begin
.drr_toname
, "@--head--",
876 sizeof (drr
->drr_u
.drr_begin
.drr_toname
));
879 dsp
= kmem_zalloc(sizeof (dmu_sendarg_t
), KM_SLEEP
);
883 dsp
->dsa_outfd
= outfd
;
884 dsp
->dsa_proc
= curproc
;
887 dsp
->dsa_toguid
= dsl_dataset_phys(to_ds
)->ds_guid
;
888 dsp
->dsa_pending_op
= PENDING_NONE
;
889 dsp
->dsa_featureflags
= featureflags
;
890 dsp
->dsa_resume_object
= resumeobj
;
891 dsp
->dsa_resume_offset
= resumeoff
;
893 mutex_enter(&to_ds
->ds_sendstream_lock
);
894 list_insert_head(&to_ds
->ds_sendstreams
, dsp
);
895 mutex_exit(&to_ds
->ds_sendstream_lock
);
897 dsl_dataset_long_hold(to_ds
, FTAG
);
898 dsl_pool_rele(dp
, tag
);
900 if (resumeobj
!= 0 || resumeoff
!= 0) {
901 dmu_object_info_t to_doi
;
903 err
= dmu_object_info(os
, resumeobj
, &to_doi
);
906 SET_BOOKMARK(&to_arg
.resume
, to_ds
->ds_object
, resumeobj
, 0,
907 resumeoff
/ to_doi
.doi_data_block_size
);
909 nvl
= fnvlist_alloc();
910 fnvlist_add_uint64(nvl
, "resume_object", resumeobj
);
911 fnvlist_add_uint64(nvl
, "resume_offset", resumeoff
);
912 payload
= fnvlist_pack(nvl
, &payload_len
);
913 drr
->drr_payloadlen
= payload_len
;
917 err
= dump_record(dsp
, payload
, payload_len
);
918 fnvlist_pack_free(payload
, payload_len
);
924 err
= bqueue_init(&to_arg
.q
, zfs_send_queue_length
,
925 offsetof(struct send_block_record
, ln
));
926 to_arg
.error_code
= 0;
927 to_arg
.cancel
= B_FALSE
;
929 to_arg
.fromtxg
= fromtxg
;
930 to_arg
.flags
= TRAVERSE_PRE
| TRAVERSE_PREFETCH
;
931 (void) thread_create(NULL
, 0, send_traverse_thread
, &to_arg
, 0, curproc
,
932 TS_RUN
, minclsyspri
);
934 to_data
= bqueue_dequeue(&to_arg
.q
);
936 while (!to_data
->eos_marker
&& err
== 0) {
937 err
= do_dump(dsp
, to_data
);
938 to_data
= get_next_record(&to_arg
.q
, to_data
);
939 if (issig(JUSTLOOKING
) && issig(FORREAL
))
944 to_arg
.cancel
= B_TRUE
;
945 while (!to_data
->eos_marker
) {
946 to_data
= get_next_record(&to_arg
.q
, to_data
);
949 kmem_free(to_data
, sizeof (*to_data
));
951 bqueue_destroy(&to_arg
.q
);
953 if (err
== 0 && to_arg
.error_code
!= 0)
954 err
= to_arg
.error_code
;
959 if (dsp
->dsa_pending_op
!= PENDING_NONE
)
960 if (dump_record(dsp
, NULL
, 0) != 0)
961 err
= SET_ERROR(EINTR
);
964 if (err
== EINTR
&& dsp
->dsa_err
!= 0)
969 bzero(drr
, sizeof (dmu_replay_record_t
));
970 drr
->drr_type
= DRR_END
;
971 drr
->drr_u
.drr_end
.drr_checksum
= dsp
->dsa_zc
;
972 drr
->drr_u
.drr_end
.drr_toguid
= dsp
->dsa_toguid
;
974 if (dump_record(dsp
, NULL
, 0) != 0)
978 mutex_enter(&to_ds
->ds_sendstream_lock
);
979 list_remove(&to_ds
->ds_sendstreams
, dsp
);
980 mutex_exit(&to_ds
->ds_sendstream_lock
);
982 kmem_free(drr
, sizeof (dmu_replay_record_t
));
983 kmem_free(dsp
, sizeof (dmu_sendarg_t
));
985 dsl_dataset_long_rele(to_ds
, FTAG
);
991 dmu_send_obj(const char *pool
, uint64_t tosnap
, uint64_t fromsnap
,
992 boolean_t embedok
, boolean_t large_block_ok
, boolean_t compressok
,
993 int outfd
, vnode_t
*vp
, offset_t
*off
)
997 dsl_dataset_t
*fromds
= NULL
;
1000 err
= dsl_pool_hold(pool
, FTAG
, &dp
);
1004 err
= dsl_dataset_hold_obj(dp
, tosnap
, FTAG
, &ds
);
1006 dsl_pool_rele(dp
, FTAG
);
1010 if (fromsnap
!= 0) {
1011 zfs_bookmark_phys_t zb
;
1014 err
= dsl_dataset_hold_obj(dp
, fromsnap
, FTAG
, &fromds
);
1016 dsl_dataset_rele(ds
, FTAG
);
1017 dsl_pool_rele(dp
, FTAG
);
1020 if (!dsl_dataset_is_before(ds
, fromds
, 0))
1021 err
= SET_ERROR(EXDEV
);
1022 zb
.zbm_creation_time
=
1023 dsl_dataset_phys(fromds
)->ds_creation_time
;
1024 zb
.zbm_creation_txg
= dsl_dataset_phys(fromds
)->ds_creation_txg
;
1025 zb
.zbm_guid
= dsl_dataset_phys(fromds
)->ds_guid
;
1026 is_clone
= (fromds
->ds_dir
!= ds
->ds_dir
);
1027 dsl_dataset_rele(fromds
, FTAG
);
1028 err
= dmu_send_impl(FTAG
, dp
, ds
, &zb
, is_clone
,
1029 embedok
, large_block_ok
, compressok
, outfd
, 0, 0, vp
, off
);
1031 err
= dmu_send_impl(FTAG
, dp
, ds
, NULL
, B_FALSE
,
1032 embedok
, large_block_ok
, compressok
, outfd
, 0, 0, vp
, off
);
1034 dsl_dataset_rele(ds
, FTAG
);
1039 dmu_send(const char *tosnap
, const char *fromsnap
, boolean_t embedok
,
1040 boolean_t large_block_ok
, boolean_t compressok
, int outfd
,
1041 uint64_t resumeobj
, uint64_t resumeoff
,
1042 vnode_t
*vp
, offset_t
*off
)
1047 boolean_t owned
= B_FALSE
;
1049 if (fromsnap
!= NULL
&& strpbrk(fromsnap
, "@#") == NULL
)
1050 return (SET_ERROR(EINVAL
));
1052 err
= dsl_pool_hold(tosnap
, FTAG
, &dp
);
1056 if (strchr(tosnap
, '@') == NULL
&& spa_writeable(dp
->dp_spa
)) {
1058 * We are sending a filesystem or volume. Ensure
1059 * that it doesn't change by owning the dataset.
1061 err
= dsl_dataset_own(dp
, tosnap
, FTAG
, &ds
);
1064 err
= dsl_dataset_hold(dp
, tosnap
, FTAG
, &ds
);
1067 dsl_pool_rele(dp
, FTAG
);
1071 if (fromsnap
!= NULL
) {
1072 zfs_bookmark_phys_t zb
;
1073 boolean_t is_clone
= B_FALSE
;
1074 int fsnamelen
= strchr(tosnap
, '@') - tosnap
;
1077 * If the fromsnap is in a different filesystem, then
1078 * mark the send stream as a clone.
1080 if (strncmp(tosnap
, fromsnap
, fsnamelen
) != 0 ||
1081 (fromsnap
[fsnamelen
] != '@' &&
1082 fromsnap
[fsnamelen
] != '#')) {
1086 if (strchr(fromsnap
, '@')) {
1087 dsl_dataset_t
*fromds
;
1088 err
= dsl_dataset_hold(dp
, fromsnap
, FTAG
, &fromds
);
1090 if (!dsl_dataset_is_before(ds
, fromds
, 0))
1091 err
= SET_ERROR(EXDEV
);
1092 zb
.zbm_creation_time
=
1093 dsl_dataset_phys(fromds
)->ds_creation_time
;
1094 zb
.zbm_creation_txg
=
1095 dsl_dataset_phys(fromds
)->ds_creation_txg
;
1096 zb
.zbm_guid
= dsl_dataset_phys(fromds
)->ds_guid
;
1097 is_clone
= (ds
->ds_dir
!= fromds
->ds_dir
);
1098 dsl_dataset_rele(fromds
, FTAG
);
1101 err
= dsl_bookmark_lookup(dp
, fromsnap
, ds
, &zb
);
1104 dsl_dataset_rele(ds
, FTAG
);
1105 dsl_pool_rele(dp
, FTAG
);
1108 err
= dmu_send_impl(FTAG
, dp
, ds
, &zb
, is_clone
,
1109 embedok
, large_block_ok
, compressok
,
1110 outfd
, resumeobj
, resumeoff
, vp
, off
);
1112 err
= dmu_send_impl(FTAG
, dp
, ds
, NULL
, B_FALSE
,
1113 embedok
, large_block_ok
, compressok
,
1114 outfd
, resumeobj
, resumeoff
, vp
, off
);
1117 dsl_dataset_disown(ds
, FTAG
);
1119 dsl_dataset_rele(ds
, FTAG
);
1124 dmu_adjust_send_estimate_for_indirects(dsl_dataset_t
*ds
, uint64_t uncompressed
,
1125 uint64_t compressed
, boolean_t stream_compressed
, uint64_t *sizep
)
1130 * Assume that space (both on-disk and in-stream) is dominated by
1131 * data. We will adjust for indirect blocks and the copies property,
1132 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
1135 uint64_t recordsize
;
1136 uint64_t record_count
;
1138 /* Assume all (uncompressed) blocks are recordsize. */
1139 err
= dsl_prop_get_int_ds(ds
, zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
1143 record_count
= uncompressed
/ recordsize
;
1146 * If we're estimating a send size for a compressed stream, use the
1147 * compressed data size to estimate the stream size. Otherwise, use the
1148 * uncompressed data size.
1150 size
= stream_compressed
? compressed
: uncompressed
;
1153 * Subtract out approximate space used by indirect blocks.
1154 * Assume most space is used by data blocks (non-indirect, non-dnode).
1155 * Assume no ditto blocks or internal fragmentation.
1157 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
1160 size
-= record_count
* sizeof (blkptr_t
);
1162 /* Add in the space for the record associated with each block. */
1163 size
+= record_count
* sizeof (dmu_replay_record_t
);
1171 dmu_send_estimate(dsl_dataset_t
*ds
, dsl_dataset_t
*fromds
,
1172 boolean_t stream_compressed
, uint64_t *sizep
)
1175 uint64_t uncomp
, comp
;
1177 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1179 /* tosnap must be a snapshot */
1180 if (!ds
->ds_is_snapshot
)
1181 return (SET_ERROR(EINVAL
));
1183 /* fromsnap, if provided, must be a snapshot */
1184 if (fromds
!= NULL
&& !fromds
->ds_is_snapshot
)
1185 return (SET_ERROR(EINVAL
));
1188 * fromsnap must be an earlier snapshot from the same fs as tosnap,
1189 * or the origin's fs.
1191 if (fromds
!= NULL
&& !dsl_dataset_is_before(ds
, fromds
, 0))
1192 return (SET_ERROR(EXDEV
));
1194 /* Get compressed and uncompressed size estimates of changed data. */
1195 if (fromds
== NULL
) {
1196 uncomp
= dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1197 comp
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1200 err
= dsl_dataset_space_written(fromds
, ds
,
1201 &used
, &comp
, &uncomp
);
1206 err
= dmu_adjust_send_estimate_for_indirects(ds
, uncomp
, comp
,
1207 stream_compressed
, sizep
);
1211 struct calculate_send_arg
{
1212 uint64_t uncompressed
;
1213 uint64_t compressed
;
1217 * Simple callback used to traverse the blocks of a snapshot and sum their
1218 * uncompressed and compressed sizes.
1222 dmu_calculate_send_traversal(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1223 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1225 struct calculate_send_arg
*space
= arg
;
1226 if (bp
!= NULL
&& !BP_IS_HOLE(bp
)) {
1227 space
->uncompressed
+= BP_GET_UCSIZE(bp
);
1228 space
->compressed
+= BP_GET_PSIZE(bp
);
1234 * Given a desination snapshot and a TXG, calculate the approximate size of a
1235 * send stream sent from that TXG. from_txg may be zero, indicating that the
1236 * whole snapshot will be sent.
1239 dmu_send_estimate_from_txg(dsl_dataset_t
*ds
, uint64_t from_txg
,
1240 boolean_t stream_compressed
, uint64_t *sizep
)
1243 struct calculate_send_arg size
= { 0 };
1245 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1247 /* tosnap must be a snapshot */
1248 if (!dsl_dataset_is_snapshot(ds
))
1249 return (SET_ERROR(EINVAL
));
1251 /* verify that from_txg is before the provided snapshot was taken */
1252 if (from_txg
>= dsl_dataset_phys(ds
)->ds_creation_txg
) {
1253 return (SET_ERROR(EXDEV
));
1256 * traverse the blocks of the snapshot with birth times after
1257 * from_txg, summing their uncompressed size
1259 err
= traverse_dataset(ds
, from_txg
, TRAVERSE_POST
,
1260 dmu_calculate_send_traversal
, &size
);
1265 err
= dmu_adjust_send_estimate_for_indirects(ds
, size
.uncompressed
,
1266 size
.compressed
, stream_compressed
, sizep
);
1270 typedef struct dmu_recv_begin_arg
{
1271 const char *drba_origin
;
1272 dmu_recv_cookie_t
*drba_cookie
;
1274 uint64_t drba_snapobj
;
1275 } dmu_recv_begin_arg_t
;
1278 recv_begin_check_existing_impl(dmu_recv_begin_arg_t
*drba
, dsl_dataset_t
*ds
,
1283 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1285 /* temporary clone name must not exist */
1286 error
= zap_lookup(dp
->dp_meta_objset
,
1287 dsl_dir_phys(ds
->ds_dir
)->dd_child_dir_zapobj
, recv_clone_name
,
1289 if (error
!= ENOENT
)
1290 return (error
== 0 ? EBUSY
: error
);
1292 /* new snapshot name must not exist */
1293 error
= zap_lookup(dp
->dp_meta_objset
,
1294 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1295 drba
->drba_cookie
->drc_tosnap
, 8, 1, &val
);
1296 if (error
!= ENOENT
)
1297 return (error
== 0 ? EEXIST
: error
);
1300 * Check snapshot limit before receiving. We'll recheck again at the
1301 * end, but might as well abort before receiving if we're already over
1304 * Note that we do not check the file system limit with
1305 * dsl_dir_fscount_check because the temporary %clones don't count
1306 * against that limit.
1308 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1, ZFS_PROP_SNAPSHOT_LIMIT
,
1309 NULL
, drba
->drba_cred
);
1313 if (fromguid
!= 0) {
1314 dsl_dataset_t
*snap
;
1315 uint64_t obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1317 /* Find snapshot in this dir that matches fromguid. */
1319 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
1322 return (SET_ERROR(ENODEV
));
1323 if (snap
->ds_dir
!= ds
->ds_dir
) {
1324 dsl_dataset_rele(snap
, FTAG
);
1325 return (SET_ERROR(ENODEV
));
1327 if (dsl_dataset_phys(snap
)->ds_guid
== fromguid
)
1329 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
1330 dsl_dataset_rele(snap
, FTAG
);
1333 return (SET_ERROR(ENODEV
));
1335 if (drba
->drba_cookie
->drc_force
) {
1336 drba
->drba_snapobj
= obj
;
1339 * If we are not forcing, there must be no
1340 * changes since fromsnap.
1342 if (dsl_dataset_modified_since_snap(ds
, snap
)) {
1343 dsl_dataset_rele(snap
, FTAG
);
1344 return (SET_ERROR(ETXTBSY
));
1346 drba
->drba_snapobj
= ds
->ds_prev
->ds_object
;
1349 dsl_dataset_rele(snap
, FTAG
);
1351 /* if full, then must be forced */
1352 if (!drba
->drba_cookie
->drc_force
)
1353 return (SET_ERROR(EEXIST
));
1354 /* start from $ORIGIN@$ORIGIN, if supported */
1355 drba
->drba_snapobj
= dp
->dp_origin_snap
!= NULL
?
1356 dp
->dp_origin_snap
->ds_object
: 0;
1364 dmu_recv_begin_check(void *arg
, dmu_tx_t
*tx
)
1366 dmu_recv_begin_arg_t
*drba
= arg
;
1367 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1368 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1369 uint64_t fromguid
= drrb
->drr_fromguid
;
1370 int flags
= drrb
->drr_flags
;
1372 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1374 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1376 /* already checked */
1377 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1378 ASSERT(!(featureflags
& DMU_BACKUP_FEATURE_RESUMING
));
1380 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1381 DMU_COMPOUNDSTREAM
||
1382 drrb
->drr_type
>= DMU_OST_NUMTYPES
||
1383 ((flags
& DRR_FLAG_CLONE
) && drba
->drba_origin
== NULL
))
1384 return (SET_ERROR(EINVAL
));
1386 /* Verify pool version supports SA if SA_SPILL feature set */
1387 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
1388 spa_version(dp
->dp_spa
) < SPA_VERSION_SA
)
1389 return (SET_ERROR(ENOTSUP
));
1391 if (drba
->drba_cookie
->drc_resumable
&&
1392 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EXTENSIBLE_DATASET
))
1393 return (SET_ERROR(ENOTSUP
));
1396 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1397 * record to a plain WRITE record, so the pool must have the
1398 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1399 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1401 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
1402 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
))
1403 return (SET_ERROR(ENOTSUP
));
1404 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
1405 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
))
1406 return (SET_ERROR(ENOTSUP
));
1409 * The receiving code doesn't know how to translate large blocks
1410 * to smaller ones, so the pool must have the LARGE_BLOCKS
1411 * feature enabled if the stream has LARGE_BLOCKS.
1413 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_BLOCKS
) &&
1414 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_BLOCKS
))
1415 return (SET_ERROR(ENOTSUP
));
1418 * The receiving code doesn't know how to translate large dnodes
1419 * to smaller ones, so the pool must have the LARGE_DNODE
1420 * feature enabled if the stream has LARGE_DNODE.
1422 if ((featureflags
& DMU_BACKUP_FEATURE_LARGE_DNODE
) &&
1423 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LARGE_DNODE
))
1424 return (SET_ERROR(ENOTSUP
));
1426 error
= dsl_dataset_hold(dp
, tofs
, FTAG
, &ds
);
1428 /* target fs already exists; recv into temp clone */
1430 /* Can't recv a clone into an existing fs */
1431 if (flags
& DRR_FLAG_CLONE
|| drba
->drba_origin
) {
1432 dsl_dataset_rele(ds
, FTAG
);
1433 return (SET_ERROR(EINVAL
));
1436 error
= recv_begin_check_existing_impl(drba
, ds
, fromguid
);
1437 dsl_dataset_rele(ds
, FTAG
);
1438 } else if (error
== ENOENT
) {
1439 /* target fs does not exist; must be a full backup or clone */
1440 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1443 * If it's a non-clone incremental, we are missing the
1444 * target fs, so fail the recv.
1446 if (fromguid
!= 0 && !(flags
& DRR_FLAG_CLONE
||
1448 return (SET_ERROR(ENOENT
));
1451 * If we're receiving a full send as a clone, and it doesn't
1452 * contain all the necessary free records and freeobject
1453 * records, reject it.
1455 if (fromguid
== 0 && drba
->drba_origin
&&
1456 !(flags
& DRR_FLAG_FREERECORDS
))
1457 return (SET_ERROR(EINVAL
));
1459 /* Open the parent of tofs */
1460 ASSERT3U(strlen(tofs
), <, sizeof (buf
));
1461 (void) strlcpy(buf
, tofs
, strrchr(tofs
, '/') - tofs
+ 1);
1462 error
= dsl_dataset_hold(dp
, buf
, FTAG
, &ds
);
1467 * Check filesystem and snapshot limits before receiving. We'll
1468 * recheck snapshot limits again at the end (we create the
1469 * filesystems and increment those counts during begin_sync).
1471 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
1472 ZFS_PROP_FILESYSTEM_LIMIT
, NULL
, drba
->drba_cred
);
1474 dsl_dataset_rele(ds
, FTAG
);
1478 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, 1,
1479 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, drba
->drba_cred
);
1481 dsl_dataset_rele(ds
, FTAG
);
1485 if (drba
->drba_origin
!= NULL
) {
1486 dsl_dataset_t
*origin
;
1487 error
= dsl_dataset_hold(dp
, drba
->drba_origin
,
1490 dsl_dataset_rele(ds
, FTAG
);
1493 if (!origin
->ds_is_snapshot
) {
1494 dsl_dataset_rele(origin
, FTAG
);
1495 dsl_dataset_rele(ds
, FTAG
);
1496 return (SET_ERROR(EINVAL
));
1498 if (dsl_dataset_phys(origin
)->ds_guid
!= fromguid
&&
1500 dsl_dataset_rele(origin
, FTAG
);
1501 dsl_dataset_rele(ds
, FTAG
);
1502 return (SET_ERROR(ENODEV
));
1504 dsl_dataset_rele(origin
, FTAG
);
1506 dsl_dataset_rele(ds
, FTAG
);
1513 dmu_recv_begin_sync(void *arg
, dmu_tx_t
*tx
)
1515 dmu_recv_begin_arg_t
*drba
= arg
;
1516 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1517 objset_t
*mos
= dp
->dp_meta_objset
;
1518 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1519 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1520 dsl_dataset_t
*ds
, *newds
;
1523 uint64_t crflags
= 0;
1525 if (drrb
->drr_flags
& DRR_FLAG_CI_DATA
)
1526 crflags
|= DS_FLAG_CI_DATASET
;
1528 error
= dsl_dataset_hold(dp
, tofs
, FTAG
, &ds
);
1530 /* create temporary clone */
1531 dsl_dataset_t
*snap
= NULL
;
1532 if (drba
->drba_snapobj
!= 0) {
1533 VERIFY0(dsl_dataset_hold_obj(dp
,
1534 drba
->drba_snapobj
, FTAG
, &snap
));
1536 dsobj
= dsl_dataset_create_sync(ds
->ds_dir
, recv_clone_name
,
1537 snap
, crflags
, drba
->drba_cred
, tx
);
1538 if (drba
->drba_snapobj
!= 0)
1539 dsl_dataset_rele(snap
, FTAG
);
1540 dsl_dataset_rele(ds
, FTAG
);
1544 dsl_dataset_t
*origin
= NULL
;
1546 VERIFY0(dsl_dir_hold(dp
, tofs
, FTAG
, &dd
, &tail
));
1548 if (drba
->drba_origin
!= NULL
) {
1549 VERIFY0(dsl_dataset_hold(dp
, drba
->drba_origin
,
1553 /* Create new dataset. */
1554 dsobj
= dsl_dataset_create_sync(dd
,
1555 strrchr(tofs
, '/') + 1,
1556 origin
, crflags
, drba
->drba_cred
, tx
);
1558 dsl_dataset_rele(origin
, FTAG
);
1559 dsl_dir_rele(dd
, FTAG
);
1560 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
1562 VERIFY0(dsl_dataset_own_obj(dp
, dsobj
, dmu_recv_tag
, &newds
));
1564 if (drba
->drba_cookie
->drc_resumable
) {
1568 dsl_dataset_zapify(newds
, tx
);
1569 if (drrb
->drr_fromguid
!= 0) {
1570 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_FROMGUID
,
1571 8, 1, &drrb
->drr_fromguid
, tx
));
1573 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TOGUID
,
1574 8, 1, &drrb
->drr_toguid
, tx
));
1575 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_TONAME
,
1576 1, strlen(drrb
->drr_toname
) + 1, drrb
->drr_toname
, tx
));
1577 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OBJECT
,
1579 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_OFFSET
,
1581 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_BYTES
,
1583 if (DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
) &
1584 DMU_BACKUP_FEATURE_LARGE_BLOCKS
) {
1585 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_LARGEBLOCK
,
1588 if (DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
) &
1589 DMU_BACKUP_FEATURE_EMBED_DATA
) {
1590 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_EMBEDOK
,
1593 if (DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
) &
1594 DMU_BACKUP_FEATURE_COMPRESSED
) {
1595 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_RESUME_COMPRESSOK
,
1600 dmu_buf_will_dirty(newds
->ds_dbuf
, tx
);
1601 dsl_dataset_phys(newds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
1604 * If we actually created a non-clone, we need to create the
1605 * objset in our new dataset.
1607 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds
))) {
1608 (void) dmu_objset_create_impl(dp
->dp_spa
,
1609 newds
, dsl_dataset_get_blkptr(newds
), drrb
->drr_type
, tx
);
1612 drba
->drba_cookie
->drc_ds
= newds
;
1614 spa_history_log_internal_ds(newds
, "receive", tx
, "");
1618 dmu_recv_resume_begin_check(void *arg
, dmu_tx_t
*tx
)
1620 dmu_recv_begin_arg_t
*drba
= arg
;
1621 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1622 struct drr_begin
*drrb
= drba
->drba_cookie
->drc_drrb
;
1624 uint64_t featureflags
= DMU_GET_FEATUREFLAGS(drrb
->drr_versioninfo
);
1626 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1629 /* 6 extra bytes for /%recv */
1630 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1632 /* already checked */
1633 ASSERT3U(drrb
->drr_magic
, ==, DMU_BACKUP_MAGIC
);
1634 ASSERT(featureflags
& DMU_BACKUP_FEATURE_RESUMING
);
1636 if (DMU_GET_STREAM_HDRTYPE(drrb
->drr_versioninfo
) ==
1637 DMU_COMPOUNDSTREAM
||
1638 drrb
->drr_type
>= DMU_OST_NUMTYPES
)
1639 return (SET_ERROR(EINVAL
));
1641 /* Verify pool version supports SA if SA_SPILL feature set */
1642 if ((featureflags
& DMU_BACKUP_FEATURE_SA_SPILL
) &&
1643 spa_version(dp
->dp_spa
) < SPA_VERSION_SA
)
1644 return (SET_ERROR(ENOTSUP
));
1647 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1648 * record to a plain WRITE record, so the pool must have the
1649 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1650 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1652 if ((featureflags
& DMU_BACKUP_FEATURE_EMBED_DATA
) &&
1653 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_EMBEDDED_DATA
))
1654 return (SET_ERROR(ENOTSUP
));
1655 if ((featureflags
& DMU_BACKUP_FEATURE_LZ4
) &&
1656 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_LZ4_COMPRESS
))
1657 return (SET_ERROR(ENOTSUP
));
1659 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
1660 tofs
, recv_clone_name
);
1662 if (dsl_dataset_hold(dp
, recvname
, FTAG
, &ds
) != 0) {
1663 /* %recv does not exist; continue in tofs */
1664 error
= dsl_dataset_hold(dp
, tofs
, FTAG
, &ds
);
1669 /* check that ds is marked inconsistent */
1670 if (!DS_IS_INCONSISTENT(ds
)) {
1671 dsl_dataset_rele(ds
, FTAG
);
1672 return (SET_ERROR(EINVAL
));
1675 /* check that there is resuming data, and that the toguid matches */
1676 if (!dsl_dataset_is_zapified(ds
)) {
1677 dsl_dataset_rele(ds
, FTAG
);
1678 return (SET_ERROR(EINVAL
));
1680 error
= zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1681 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
);
1682 if (error
!= 0 || drrb
->drr_toguid
!= val
) {
1683 dsl_dataset_rele(ds
, FTAG
);
1684 return (SET_ERROR(EINVAL
));
1688 * Check if the receive is still running. If so, it will be owned.
1689 * Note that nothing else can own the dataset (e.g. after the receive
1690 * fails) because it will be marked inconsistent.
1692 if (dsl_dataset_has_owner(ds
)) {
1693 dsl_dataset_rele(ds
, FTAG
);
1694 return (SET_ERROR(EBUSY
));
1697 /* There should not be any snapshots of this fs yet. */
1698 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
->ds_dir
== ds
->ds_dir
) {
1699 dsl_dataset_rele(ds
, FTAG
);
1700 return (SET_ERROR(EINVAL
));
1704 * Note: resume point will be checked when we process the first WRITE
1708 /* check that the origin matches */
1710 (void) zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1711 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
);
1712 if (drrb
->drr_fromguid
!= val
) {
1713 dsl_dataset_rele(ds
, FTAG
);
1714 return (SET_ERROR(EINVAL
));
1717 dsl_dataset_rele(ds
, FTAG
);
1722 dmu_recv_resume_begin_sync(void *arg
, dmu_tx_t
*tx
)
1724 dmu_recv_begin_arg_t
*drba
= arg
;
1725 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1726 const char *tofs
= drba
->drba_cookie
->drc_tofs
;
1729 /* 6 extra bytes for /%recv */
1730 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1732 (void) snprintf(recvname
, sizeof (recvname
), "%s/%s",
1733 tofs
, recv_clone_name
);
1735 if (dsl_dataset_hold(dp
, recvname
, FTAG
, &ds
) != 0) {
1736 /* %recv does not exist; continue in tofs */
1737 VERIFY0(dsl_dataset_hold(dp
, tofs
, FTAG
, &ds
));
1738 drba
->drba_cookie
->drc_newfs
= B_TRUE
;
1741 /* clear the inconsistent flag so that we can own it */
1742 ASSERT(DS_IS_INCONSISTENT(ds
));
1743 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1744 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
1745 dsobj
= ds
->ds_object
;
1746 dsl_dataset_rele(ds
, FTAG
);
1748 VERIFY0(dsl_dataset_own_obj(dp
, dsobj
, dmu_recv_tag
, &ds
));
1750 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1751 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
1753 ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds
)));
1755 drba
->drba_cookie
->drc_ds
= ds
;
1757 spa_history_log_internal_ds(ds
, "resume receive", tx
, "");
1761 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1762 * succeeds; otherwise we will leak the holds on the datasets.
1765 dmu_recv_begin(char *tofs
, char *tosnap
, dmu_replay_record_t
*drr_begin
,
1766 boolean_t force
, boolean_t resumable
, char *origin
, dmu_recv_cookie_t
*drc
)
1768 dmu_recv_begin_arg_t drba
= { 0 };
1770 bzero(drc
, sizeof (dmu_recv_cookie_t
));
1771 drc
->drc_drr_begin
= drr_begin
;
1772 drc
->drc_drrb
= &drr_begin
->drr_u
.drr_begin
;
1773 drc
->drc_tosnap
= tosnap
;
1774 drc
->drc_tofs
= tofs
;
1775 drc
->drc_force
= force
;
1776 drc
->drc_resumable
= resumable
;
1777 drc
->drc_cred
= CRED();
1779 if (drc
->drc_drrb
->drr_magic
== BSWAP_64(DMU_BACKUP_MAGIC
)) {
1780 drc
->drc_byteswap
= B_TRUE
;
1781 fletcher_4_incremental_byteswap(drr_begin
,
1782 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
1783 byteswap_record(drr_begin
);
1784 } else if (drc
->drc_drrb
->drr_magic
== DMU_BACKUP_MAGIC
) {
1785 fletcher_4_incremental_native(drr_begin
,
1786 sizeof (dmu_replay_record_t
), &drc
->drc_cksum
);
1788 return (SET_ERROR(EINVAL
));
1791 drba
.drba_origin
= origin
;
1792 drba
.drba_cookie
= drc
;
1793 drba
.drba_cred
= CRED();
1795 if (DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
) &
1796 DMU_BACKUP_FEATURE_RESUMING
) {
1797 return (dsl_sync_task(tofs
,
1798 dmu_recv_resume_begin_check
, dmu_recv_resume_begin_sync
,
1799 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
));
1801 return (dsl_sync_task(tofs
,
1802 dmu_recv_begin_check
, dmu_recv_begin_sync
,
1803 &drba
, 5, ZFS_SPACE_CHECK_NORMAL
));
1807 struct receive_record_arg
{
1808 dmu_replay_record_t header
;
1809 void *payload
; /* Pointer to a buffer containing the payload */
1811 * If the record is a write, pointer to the arc_buf_t containing the
1814 arc_buf_t
*write_buf
;
1816 uint64_t bytes_read
; /* bytes read from stream when record created */
1817 boolean_t eos_marker
; /* Marks the end of the stream */
1821 struct receive_writer_arg
{
1827 * These three args are used to signal to the main thread that we're
1835 /* A map from guid to dataset to help handle dedup'd streams. */
1836 avl_tree_t
*guid_to_ds_map
;
1837 boolean_t resumable
;
1838 uint64_t last_object
, last_offset
;
1839 uint64_t bytes_read
; /* bytes read when current record created */
1843 list_t list
; /* List of struct receive_objnode. */
1845 * Last object looked up. Used to assert that objects are being looked
1846 * up in ascending order.
1848 uint64_t last_lookup
;
1851 struct receive_objnode
{
1856 struct receive_arg
{
1858 vnode_t
*vp
; /* The vnode to read the stream from */
1859 uint64_t voff
; /* The current offset in the stream */
1860 uint64_t bytes_read
;
1862 * A record that has had its payload read in, but hasn't yet been handed
1863 * off to the worker thread.
1865 struct receive_record_arg
*rrd
;
1866 /* A record that has had its header read in, but not its payload. */
1867 struct receive_record_arg
*next_rrd
;
1869 zio_cksum_t prev_cksum
;
1872 /* Sorted list of objects not to issue prefetches for. */
1873 struct objlist ignore_objlist
;
1876 typedef struct guid_map_entry
{
1878 dsl_dataset_t
*gme_ds
;
1883 guid_compare(const void *arg1
, const void *arg2
)
1885 const guid_map_entry_t
*gmep1
= (const guid_map_entry_t
*)arg1
;
1886 const guid_map_entry_t
*gmep2
= (const guid_map_entry_t
*)arg2
;
1888 return (AVL_CMP(gmep1
->guid
, gmep2
->guid
));
1892 free_guid_map_onexit(void *arg
)
1894 avl_tree_t
*ca
= arg
;
1895 void *cookie
= NULL
;
1896 guid_map_entry_t
*gmep
;
1898 while ((gmep
= avl_destroy_nodes(ca
, &cookie
)) != NULL
) {
1899 dsl_dataset_long_rele(gmep
->gme_ds
, gmep
);
1900 dsl_dataset_rele(gmep
->gme_ds
, gmep
);
1901 kmem_free(gmep
, sizeof (guid_map_entry_t
));
1904 kmem_free(ca
, sizeof (avl_tree_t
));
1908 receive_read(struct receive_arg
*ra
, int len
, void *buf
)
1913 * The code doesn't rely on this (lengths being multiples of 8). See
1914 * comment in dump_bytes.
1918 while (done
< len
) {
1921 ra
->err
= vn_rdwr(UIO_READ
, ra
->vp
,
1922 (char *)buf
+ done
, len
- done
,
1923 ra
->voff
, UIO_SYSSPACE
, FAPPEND
,
1924 RLIM64_INFINITY
, CRED(), &resid
);
1926 if (resid
== len
- done
) {
1928 * Note: ECKSUM indicates that the receive
1929 * was interrupted and can potentially be resumed.
1931 ra
->err
= SET_ERROR(ECKSUM
);
1933 ra
->voff
+= len
- done
- resid
;
1939 ra
->bytes_read
+= len
;
1941 ASSERT3U(done
, ==, len
);
1945 noinline
static void
1946 byteswap_record(dmu_replay_record_t
*drr
)
1948 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1949 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1950 drr
->drr_type
= BSWAP_32(drr
->drr_type
);
1951 drr
->drr_payloadlen
= BSWAP_32(drr
->drr_payloadlen
);
1953 switch (drr
->drr_type
) {
1955 DO64(drr_begin
.drr_magic
);
1956 DO64(drr_begin
.drr_versioninfo
);
1957 DO64(drr_begin
.drr_creation_time
);
1958 DO32(drr_begin
.drr_type
);
1959 DO32(drr_begin
.drr_flags
);
1960 DO64(drr_begin
.drr_toguid
);
1961 DO64(drr_begin
.drr_fromguid
);
1964 DO64(drr_object
.drr_object
);
1965 DO32(drr_object
.drr_type
);
1966 DO32(drr_object
.drr_bonustype
);
1967 DO32(drr_object
.drr_blksz
);
1968 DO32(drr_object
.drr_bonuslen
);
1969 DO64(drr_object
.drr_toguid
);
1971 case DRR_FREEOBJECTS
:
1972 DO64(drr_freeobjects
.drr_firstobj
);
1973 DO64(drr_freeobjects
.drr_numobjs
);
1974 DO64(drr_freeobjects
.drr_toguid
);
1977 DO64(drr_write
.drr_object
);
1978 DO32(drr_write
.drr_type
);
1979 DO64(drr_write
.drr_offset
);
1980 DO64(drr_write
.drr_logical_size
);
1981 DO64(drr_write
.drr_toguid
);
1982 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write
.drr_key
.ddk_cksum
);
1983 DO64(drr_write
.drr_key
.ddk_prop
);
1984 DO64(drr_write
.drr_compressed_size
);
1986 case DRR_WRITE_BYREF
:
1987 DO64(drr_write_byref
.drr_object
);
1988 DO64(drr_write_byref
.drr_offset
);
1989 DO64(drr_write_byref
.drr_length
);
1990 DO64(drr_write_byref
.drr_toguid
);
1991 DO64(drr_write_byref
.drr_refguid
);
1992 DO64(drr_write_byref
.drr_refobject
);
1993 DO64(drr_write_byref
.drr_refoffset
);
1994 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_write_byref
.
1996 DO64(drr_write_byref
.drr_key
.ddk_prop
);
1998 case DRR_WRITE_EMBEDDED
:
1999 DO64(drr_write_embedded
.drr_object
);
2000 DO64(drr_write_embedded
.drr_offset
);
2001 DO64(drr_write_embedded
.drr_length
);
2002 DO64(drr_write_embedded
.drr_toguid
);
2003 DO32(drr_write_embedded
.drr_lsize
);
2004 DO32(drr_write_embedded
.drr_psize
);
2007 DO64(drr_free
.drr_object
);
2008 DO64(drr_free
.drr_offset
);
2009 DO64(drr_free
.drr_length
);
2010 DO64(drr_free
.drr_toguid
);
2013 DO64(drr_spill
.drr_object
);
2014 DO64(drr_spill
.drr_length
);
2015 DO64(drr_spill
.drr_toguid
);
2018 DO64(drr_end
.drr_toguid
);
2019 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_end
.drr_checksum
);
2025 if (drr
->drr_type
!= DRR_BEGIN
) {
2026 ZIO_CHECKSUM_BSWAP(&drr
->drr_u
.drr_checksum
.drr_checksum
);
2033 static inline uint8_t
2034 deduce_nblkptr(dmu_object_type_t bonus_type
, uint64_t bonus_size
)
2036 if (bonus_type
== DMU_OT_SA
) {
2040 ((DN_OLD_MAX_BONUSLEN
-
2041 MIN(DN_OLD_MAX_BONUSLEN
, bonus_size
)) >> SPA_BLKPTRSHIFT
));
2046 save_resume_state(struct receive_writer_arg
*rwa
,
2047 uint64_t object
, uint64_t offset
, dmu_tx_t
*tx
)
2049 int txgoff
= dmu_tx_get_txg(tx
) & TXG_MASK
;
2051 if (!rwa
->resumable
)
2055 * We use ds_resume_bytes[] != 0 to indicate that we need to
2056 * update this on disk, so it must not be 0.
2058 ASSERT(rwa
->bytes_read
!= 0);
2061 * We only resume from write records, which have a valid
2062 * (non-meta-dnode) object number.
2064 ASSERT(object
!= 0);
2067 * For resuming to work correctly, we must receive records in order,
2068 * sorted by object,offset. This is checked by the callers, but
2069 * assert it here for good measure.
2071 ASSERT3U(object
, >=, rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
]);
2072 ASSERT(object
!= rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] ||
2073 offset
>= rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
]);
2074 ASSERT3U(rwa
->bytes_read
, >=,
2075 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
]);
2077 rwa
->os
->os_dsl_dataset
->ds_resume_object
[txgoff
] = object
;
2078 rwa
->os
->os_dsl_dataset
->ds_resume_offset
[txgoff
] = offset
;
2079 rwa
->os
->os_dsl_dataset
->ds_resume_bytes
[txgoff
] = rwa
->bytes_read
;
2083 receive_object(struct receive_writer_arg
*rwa
, struct drr_object
*drro
,
2086 dmu_object_info_t doi
;
2091 if (drro
->drr_type
== DMU_OT_NONE
||
2092 !DMU_OT_IS_VALID(drro
->drr_type
) ||
2093 !DMU_OT_IS_VALID(drro
->drr_bonustype
) ||
2094 drro
->drr_checksumtype
>= ZIO_CHECKSUM_FUNCTIONS
||
2095 drro
->drr_compress
>= ZIO_COMPRESS_FUNCTIONS
||
2096 P2PHASE(drro
->drr_blksz
, SPA_MINBLOCKSIZE
) ||
2097 drro
->drr_blksz
< SPA_MINBLOCKSIZE
||
2098 drro
->drr_blksz
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)) ||
2099 drro
->drr_bonuslen
>
2100 DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa
->os
)))) {
2101 return (SET_ERROR(EINVAL
));
2104 err
= dmu_object_info(rwa
->os
, drro
->drr_object
, &doi
);
2106 if (err
!= 0 && err
!= ENOENT
)
2107 return (SET_ERROR(EINVAL
));
2108 object
= err
== 0 ? drro
->drr_object
: DMU_NEW_OBJECT
;
2111 * If we are losing blkptrs or changing the block size this must
2112 * be a new file instance. We must clear out the previous file
2113 * contents before we can change this type of metadata in the dnode.
2118 nblkptr
= deduce_nblkptr(drro
->drr_bonustype
,
2119 drro
->drr_bonuslen
);
2121 if (drro
->drr_blksz
!= doi
.doi_data_block_size
||
2122 nblkptr
< doi
.doi_nblkptr
) {
2123 err
= dmu_free_long_range(rwa
->os
, drro
->drr_object
,
2126 return (SET_ERROR(EINVAL
));
2130 tx
= dmu_tx_create(rwa
->os
);
2131 dmu_tx_hold_bonus(tx
, object
);
2132 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2138 if (object
== DMU_NEW_OBJECT
) {
2139 /* currently free, want to be allocated */
2140 err
= dmu_object_claim_dnsize(rwa
->os
, drro
->drr_object
,
2141 drro
->drr_type
, drro
->drr_blksz
,
2142 drro
->drr_bonustype
, drro
->drr_bonuslen
,
2143 drro
->drr_dn_slots
<< DNODE_SHIFT
, tx
);
2144 } else if (drro
->drr_type
!= doi
.doi_type
||
2145 drro
->drr_blksz
!= doi
.doi_data_block_size
||
2146 drro
->drr_bonustype
!= doi
.doi_bonus_type
||
2147 drro
->drr_bonuslen
!= doi
.doi_bonus_size
) {
2148 /* currently allocated, but with different properties */
2149 err
= dmu_object_reclaim(rwa
->os
, drro
->drr_object
,
2150 drro
->drr_type
, drro
->drr_blksz
,
2151 drro
->drr_bonustype
, drro
->drr_bonuslen
, tx
);
2155 return (SET_ERROR(EINVAL
));
2158 dmu_object_set_checksum(rwa
->os
, drro
->drr_object
,
2159 drro
->drr_checksumtype
, tx
);
2160 dmu_object_set_compress(rwa
->os
, drro
->drr_object
,
2161 drro
->drr_compress
, tx
);
2166 VERIFY0(dmu_bonus_hold(rwa
->os
, drro
->drr_object
, FTAG
, &db
));
2167 dmu_buf_will_dirty(db
, tx
);
2169 ASSERT3U(db
->db_size
, >=, drro
->drr_bonuslen
);
2170 bcopy(data
, db
->db_data
, drro
->drr_bonuslen
);
2171 if (rwa
->byteswap
) {
2172 dmu_object_byteswap_t byteswap
=
2173 DMU_OT_BYTESWAP(drro
->drr_bonustype
);
2174 dmu_ot_byteswap
[byteswap
].ob_func(db
->db_data
,
2175 drro
->drr_bonuslen
);
2177 dmu_buf_rele(db
, FTAG
);
2186 receive_freeobjects(struct receive_writer_arg
*rwa
,
2187 struct drr_freeobjects
*drrfo
)
2192 if (drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
< drrfo
->drr_firstobj
)
2193 return (SET_ERROR(EINVAL
));
2195 for (obj
= drrfo
->drr_firstobj
== 0 ? 1 : drrfo
->drr_firstobj
;
2196 obj
< drrfo
->drr_firstobj
+ drrfo
->drr_numobjs
&& next_err
== 0;
2197 next_err
= dmu_object_next(rwa
->os
, &obj
, FALSE
, 0)) {
2198 dmu_object_info_t doi
;
2201 err
= dmu_object_info(rwa
->os
, obj
, &doi
);
2202 if (err
== ENOENT
) {
2205 } else if (err
!= 0) {
2209 err
= dmu_free_long_object(rwa
->os
, obj
);
2213 if (next_err
!= ESRCH
)
2219 receive_write(struct receive_writer_arg
*rwa
, struct drr_write
*drrw
,
2226 if (drrw
->drr_offset
+ drrw
->drr_logical_size
< drrw
->drr_offset
||
2227 !DMU_OT_IS_VALID(drrw
->drr_type
))
2228 return (SET_ERROR(EINVAL
));
2231 * For resuming to work, records must be in increasing order
2232 * by (object, offset).
2234 if (drrw
->drr_object
< rwa
->last_object
||
2235 (drrw
->drr_object
== rwa
->last_object
&&
2236 drrw
->drr_offset
< rwa
->last_offset
)) {
2237 return (SET_ERROR(EINVAL
));
2239 rwa
->last_object
= drrw
->drr_object
;
2240 rwa
->last_offset
= drrw
->drr_offset
;
2242 if (dmu_object_info(rwa
->os
, drrw
->drr_object
, NULL
) != 0)
2243 return (SET_ERROR(EINVAL
));
2245 tx
= dmu_tx_create(rwa
->os
);
2247 dmu_tx_hold_write(tx
, drrw
->drr_object
,
2248 drrw
->drr_offset
, drrw
->drr_logical_size
);
2249 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2254 if (rwa
->byteswap
) {
2255 dmu_object_byteswap_t byteswap
=
2256 DMU_OT_BYTESWAP(drrw
->drr_type
);
2257 dmu_ot_byteswap
[byteswap
].ob_func(abuf
->b_data
,
2258 DRR_WRITE_PAYLOAD_SIZE(drrw
));
2261 /* use the bonus buf to look up the dnode in dmu_assign_arcbuf */
2262 if (dmu_bonus_hold(rwa
->os
, drrw
->drr_object
, FTAG
, &bonus
) != 0)
2263 return (SET_ERROR(EINVAL
));
2264 dmu_assign_arcbuf(bonus
, drrw
->drr_offset
, abuf
, tx
);
2267 * Note: If the receive fails, we want the resume stream to start
2268 * with the same record that we last successfully received (as opposed
2269 * to the next record), so that we can verify that we are
2270 * resuming from the correct location.
2272 save_resume_state(rwa
, drrw
->drr_object
, drrw
->drr_offset
, tx
);
2274 dmu_buf_rele(bonus
, FTAG
);
2280 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
2281 * streams to refer to a copy of the data that is already on the
2282 * system because it came in earlier in the stream. This function
2283 * finds the earlier copy of the data, and uses that copy instead of
2284 * data from the stream to fulfill this write.
2287 receive_write_byref(struct receive_writer_arg
*rwa
,
2288 struct drr_write_byref
*drrwbr
)
2292 guid_map_entry_t gmesrch
;
2293 guid_map_entry_t
*gmep
;
2295 objset_t
*ref_os
= NULL
;
2298 if (drrwbr
->drr_offset
+ drrwbr
->drr_length
< drrwbr
->drr_offset
)
2299 return (SET_ERROR(EINVAL
));
2302 * If the GUID of the referenced dataset is different from the
2303 * GUID of the target dataset, find the referenced dataset.
2305 if (drrwbr
->drr_toguid
!= drrwbr
->drr_refguid
) {
2306 gmesrch
.guid
= drrwbr
->drr_refguid
;
2307 if ((gmep
= avl_find(rwa
->guid_to_ds_map
, &gmesrch
,
2309 return (SET_ERROR(EINVAL
));
2311 if (dmu_objset_from_ds(gmep
->gme_ds
, &ref_os
))
2312 return (SET_ERROR(EINVAL
));
2317 err
= dmu_buf_hold(ref_os
, drrwbr
->drr_refobject
,
2318 drrwbr
->drr_refoffset
, FTAG
, &dbp
, DMU_READ_PREFETCH
);
2322 tx
= dmu_tx_create(rwa
->os
);
2324 dmu_tx_hold_write(tx
, drrwbr
->drr_object
,
2325 drrwbr
->drr_offset
, drrwbr
->drr_length
);
2326 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2331 dmu_write(rwa
->os
, drrwbr
->drr_object
,
2332 drrwbr
->drr_offset
, drrwbr
->drr_length
, dbp
->db_data
, tx
);
2333 dmu_buf_rele(dbp
, FTAG
);
2335 /* See comment in restore_write. */
2336 save_resume_state(rwa
, drrwbr
->drr_object
, drrwbr
->drr_offset
, tx
);
2342 receive_write_embedded(struct receive_writer_arg
*rwa
,
2343 struct drr_write_embedded
*drrwe
, void *data
)
2348 if (drrwe
->drr_offset
+ drrwe
->drr_length
< drrwe
->drr_offset
)
2351 if (drrwe
->drr_psize
> BPE_PAYLOAD_SIZE
)
2354 if (drrwe
->drr_etype
>= NUM_BP_EMBEDDED_TYPES
)
2356 if (drrwe
->drr_compression
>= ZIO_COMPRESS_FUNCTIONS
)
2359 tx
= dmu_tx_create(rwa
->os
);
2361 dmu_tx_hold_write(tx
, drrwe
->drr_object
,
2362 drrwe
->drr_offset
, drrwe
->drr_length
);
2363 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2369 dmu_write_embedded(rwa
->os
, drrwe
->drr_object
,
2370 drrwe
->drr_offset
, data
, drrwe
->drr_etype
,
2371 drrwe
->drr_compression
, drrwe
->drr_lsize
, drrwe
->drr_psize
,
2372 rwa
->byteswap
^ ZFS_HOST_BYTEORDER
, tx
);
2374 /* See comment in restore_write. */
2375 save_resume_state(rwa
, drrwe
->drr_object
, drrwe
->drr_offset
, tx
);
2381 receive_spill(struct receive_writer_arg
*rwa
, struct drr_spill
*drrs
,
2385 dmu_buf_t
*db
, *db_spill
;
2388 if (drrs
->drr_length
< SPA_MINBLOCKSIZE
||
2389 drrs
->drr_length
> spa_maxblocksize(dmu_objset_spa(rwa
->os
)))
2390 return (SET_ERROR(EINVAL
));
2392 if (dmu_object_info(rwa
->os
, drrs
->drr_object
, NULL
) != 0)
2393 return (SET_ERROR(EINVAL
));
2395 VERIFY0(dmu_bonus_hold(rwa
->os
, drrs
->drr_object
, FTAG
, &db
));
2396 if ((err
= dmu_spill_hold_by_bonus(db
, FTAG
, &db_spill
)) != 0) {
2397 dmu_buf_rele(db
, FTAG
);
2401 tx
= dmu_tx_create(rwa
->os
);
2403 dmu_tx_hold_spill(tx
, db
->db_object
);
2405 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2407 dmu_buf_rele(db
, FTAG
);
2408 dmu_buf_rele(db_spill
, FTAG
);
2412 dmu_buf_will_dirty(db_spill
, tx
);
2414 if (db_spill
->db_size
< drrs
->drr_length
)
2415 VERIFY(0 == dbuf_spill_set_blksz(db_spill
,
2416 drrs
->drr_length
, tx
));
2417 bcopy(data
, db_spill
->db_data
, drrs
->drr_length
);
2419 dmu_buf_rele(db
, FTAG
);
2420 dmu_buf_rele(db_spill
, FTAG
);
2428 receive_free(struct receive_writer_arg
*rwa
, struct drr_free
*drrf
)
2432 if (drrf
->drr_length
!= -1ULL &&
2433 drrf
->drr_offset
+ drrf
->drr_length
< drrf
->drr_offset
)
2434 return (SET_ERROR(EINVAL
));
2436 if (dmu_object_info(rwa
->os
, drrf
->drr_object
, NULL
) != 0)
2437 return (SET_ERROR(EINVAL
));
2439 err
= dmu_free_long_range(rwa
->os
, drrf
->drr_object
,
2440 drrf
->drr_offset
, drrf
->drr_length
);
2445 /* used to destroy the drc_ds on error */
2447 dmu_recv_cleanup_ds(dmu_recv_cookie_t
*drc
)
2449 if (drc
->drc_resumable
) {
2450 /* wait for our resume state to be written to disk */
2451 txg_wait_synced(drc
->drc_ds
->ds_dir
->dd_pool
, 0);
2452 dsl_dataset_disown(drc
->drc_ds
, dmu_recv_tag
);
2454 char name
[ZFS_MAX_DATASET_NAME_LEN
];
2455 dsl_dataset_name(drc
->drc_ds
, name
);
2456 dsl_dataset_disown(drc
->drc_ds
, dmu_recv_tag
);
2457 (void) dsl_destroy_head(name
);
2462 receive_cksum(struct receive_arg
*ra
, int len
, void *buf
)
2465 fletcher_4_incremental_byteswap(buf
, len
, &ra
->cksum
);
2467 fletcher_4_incremental_native(buf
, len
, &ra
->cksum
);
2472 * Read the payload into a buffer of size len, and update the current record's
2474 * Allocate ra->next_rrd and read the next record's header into
2475 * ra->next_rrd->header.
2476 * Verify checksum of payload and next record.
2479 receive_read_payload_and_next_header(struct receive_arg
*ra
, int len
, void *buf
)
2482 zio_cksum_t cksum_orig
;
2483 zio_cksum_t
*cksump
;
2486 ASSERT3U(len
, <=, SPA_MAXBLOCKSIZE
);
2487 err
= receive_read(ra
, len
, buf
);
2490 receive_cksum(ra
, len
, buf
);
2492 /* note: rrd is NULL when reading the begin record's payload */
2493 if (ra
->rrd
!= NULL
) {
2494 ra
->rrd
->payload
= buf
;
2495 ra
->rrd
->payload_size
= len
;
2496 ra
->rrd
->bytes_read
= ra
->bytes_read
;
2500 ra
->prev_cksum
= ra
->cksum
;
2502 ra
->next_rrd
= kmem_zalloc(sizeof (*ra
->next_rrd
), KM_SLEEP
);
2503 err
= receive_read(ra
, sizeof (ra
->next_rrd
->header
),
2504 &ra
->next_rrd
->header
);
2505 ra
->next_rrd
->bytes_read
= ra
->bytes_read
;
2507 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
2508 ra
->next_rrd
= NULL
;
2511 if (ra
->next_rrd
->header
.drr_type
== DRR_BEGIN
) {
2512 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
2513 ra
->next_rrd
= NULL
;
2514 return (SET_ERROR(EINVAL
));
2518 * Note: checksum is of everything up to but not including the
2521 ASSERT3U(offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2522 ==, sizeof (dmu_replay_record_t
) - sizeof (zio_cksum_t
));
2524 offsetof(dmu_replay_record_t
, drr_u
.drr_checksum
.drr_checksum
),
2525 &ra
->next_rrd
->header
);
2527 cksum_orig
= ra
->next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
2528 cksump
= &ra
->next_rrd
->header
.drr_u
.drr_checksum
.drr_checksum
;
2531 byteswap_record(&ra
->next_rrd
->header
);
2533 if ((!ZIO_CHECKSUM_IS_ZERO(cksump
)) &&
2534 !ZIO_CHECKSUM_EQUAL(ra
->cksum
, *cksump
)) {
2535 kmem_free(ra
->next_rrd
, sizeof (*ra
->next_rrd
));
2536 ra
->next_rrd
= NULL
;
2537 return (SET_ERROR(ECKSUM
));
2540 receive_cksum(ra
, sizeof (cksum_orig
), &cksum_orig
);
2546 objlist_create(struct objlist
*list
)
2548 list_create(&list
->list
, sizeof (struct receive_objnode
),
2549 offsetof(struct receive_objnode
, node
));
2550 list
->last_lookup
= 0;
2554 objlist_destroy(struct objlist
*list
)
2556 struct receive_objnode
*n
;
2558 for (n
= list_remove_head(&list
->list
);
2559 n
!= NULL
; n
= list_remove_head(&list
->list
)) {
2560 kmem_free(n
, sizeof (*n
));
2562 list_destroy(&list
->list
);
2566 * This function looks through the objlist to see if the specified object number
2567 * is contained in the objlist. In the process, it will remove all object
2568 * numbers in the list that are smaller than the specified object number. Thus,
2569 * any lookup of an object number smaller than a previously looked up object
2570 * number will always return false; therefore, all lookups should be done in
2574 objlist_exists(struct objlist
*list
, uint64_t object
)
2576 struct receive_objnode
*node
= list_head(&list
->list
);
2577 ASSERT3U(object
, >=, list
->last_lookup
);
2578 list
->last_lookup
= object
;
2579 while (node
!= NULL
&& node
->object
< object
) {
2580 VERIFY3P(node
, ==, list_remove_head(&list
->list
));
2581 kmem_free(node
, sizeof (*node
));
2582 node
= list_head(&list
->list
);
2584 return (node
!= NULL
&& node
->object
== object
);
2588 * The objlist is a list of object numbers stored in ascending order. However,
2589 * the insertion of new object numbers does not seek out the correct location to
2590 * store a new object number; instead, it appends it to the list for simplicity.
2591 * Thus, any users must take care to only insert new object numbers in ascending
2595 objlist_insert(struct objlist
*list
, uint64_t object
)
2597 struct receive_objnode
*node
= kmem_zalloc(sizeof (*node
), KM_SLEEP
);
2598 node
->object
= object
;
2601 struct receive_objnode
*last_object
= list_tail(&list
->list
);
2602 uint64_t last_objnum
= (last_object
!= NULL
? last_object
->object
: 0);
2603 ASSERT3U(node
->object
, >, last_objnum
);
2606 list_insert_tail(&list
->list
, node
);
2610 * Issue the prefetch reads for any necessary indirect blocks.
2612 * We use the object ignore list to tell us whether or not to issue prefetches
2613 * for a given object. We do this for both correctness (in case the blocksize
2614 * of an object has changed) and performance (if the object doesn't exist, don't
2615 * needlessly try to issue prefetches). We also trim the list as we go through
2616 * the stream to prevent it from growing to an unbounded size.
2618 * The object numbers within will always be in sorted order, and any write
2619 * records we see will also be in sorted order, but they're not sorted with
2620 * respect to each other (i.e. we can get several object records before
2621 * receiving each object's write records). As a result, once we've reached a
2622 * given object number, we can safely remove any reference to lower object
2623 * numbers in the ignore list. In practice, we receive up to 32 object records
2624 * before receiving write records, so the list can have up to 32 nodes in it.
2628 receive_read_prefetch(struct receive_arg
*ra
,
2629 uint64_t object
, uint64_t offset
, uint64_t length
)
2631 if (!objlist_exists(&ra
->ignore_objlist
, object
)) {
2632 dmu_prefetch(ra
->os
, object
, 1, offset
, length
,
2633 ZIO_PRIORITY_SYNC_READ
);
2638 * Read records off the stream, issuing any necessary prefetches.
2641 receive_read_record(struct receive_arg
*ra
)
2645 switch (ra
->rrd
->header
.drr_type
) {
2648 struct drr_object
*drro
= &ra
->rrd
->header
.drr_u
.drr_object
;
2649 uint32_t size
= P2ROUNDUP(drro
->drr_bonuslen
, 8);
2650 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
2651 dmu_object_info_t doi
;
2652 err
= receive_read_payload_and_next_header(ra
, size
, buf
);
2654 kmem_free(buf
, size
);
2657 err
= dmu_object_info(ra
->os
, drro
->drr_object
, &doi
);
2659 * See receive_read_prefetch for an explanation why we're
2660 * storing this object in the ignore_obj_list.
2662 if (err
== ENOENT
||
2663 (err
== 0 && doi
.doi_data_block_size
!= drro
->drr_blksz
)) {
2664 objlist_insert(&ra
->ignore_objlist
, drro
->drr_object
);
2669 case DRR_FREEOBJECTS
:
2671 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
2676 struct drr_write
*drrw
= &ra
->rrd
->header
.drr_u
.drr_write
;
2678 boolean_t is_meta
= DMU_OT_IS_METADATA(drrw
->drr_type
);
2679 if (DRR_WRITE_COMPRESSED(drrw
)) {
2680 ASSERT3U(drrw
->drr_compressed_size
, >, 0);
2681 ASSERT3U(drrw
->drr_logical_size
, >=,
2682 drrw
->drr_compressed_size
);
2684 abuf
= arc_loan_compressed_buf(
2685 dmu_objset_spa(ra
->os
),
2686 drrw
->drr_compressed_size
, drrw
->drr_logical_size
,
2687 drrw
->drr_compressiontype
);
2689 abuf
= arc_loan_buf(dmu_objset_spa(ra
->os
),
2690 is_meta
, drrw
->drr_logical_size
);
2693 err
= receive_read_payload_and_next_header(ra
,
2694 DRR_WRITE_PAYLOAD_SIZE(drrw
), abuf
->b_data
);
2696 dmu_return_arcbuf(abuf
);
2699 ra
->rrd
->write_buf
= abuf
;
2700 receive_read_prefetch(ra
, drrw
->drr_object
, drrw
->drr_offset
,
2701 drrw
->drr_logical_size
);
2704 case DRR_WRITE_BYREF
:
2706 struct drr_write_byref
*drrwb
=
2707 &ra
->rrd
->header
.drr_u
.drr_write_byref
;
2708 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
2709 receive_read_prefetch(ra
, drrwb
->drr_object
, drrwb
->drr_offset
,
2713 case DRR_WRITE_EMBEDDED
:
2715 struct drr_write_embedded
*drrwe
=
2716 &ra
->rrd
->header
.drr_u
.drr_write_embedded
;
2717 uint32_t size
= P2ROUNDUP(drrwe
->drr_psize
, 8);
2718 void *buf
= kmem_zalloc(size
, KM_SLEEP
);
2720 err
= receive_read_payload_and_next_header(ra
, size
, buf
);
2722 kmem_free(buf
, size
);
2726 receive_read_prefetch(ra
, drrwe
->drr_object
, drrwe
->drr_offset
,
2733 * It might be beneficial to prefetch indirect blocks here, but
2734 * we don't really have the data to decide for sure.
2736 err
= receive_read_payload_and_next_header(ra
, 0, NULL
);
2741 struct drr_end
*drre
= &ra
->rrd
->header
.drr_u
.drr_end
;
2742 if (!ZIO_CHECKSUM_EQUAL(ra
->prev_cksum
, drre
->drr_checksum
))
2743 return (SET_ERROR(ECKSUM
));
2748 struct drr_spill
*drrs
= &ra
->rrd
->header
.drr_u
.drr_spill
;
2749 void *buf
= kmem_zalloc(drrs
->drr_length
, KM_SLEEP
);
2750 err
= receive_read_payload_and_next_header(ra
, drrs
->drr_length
,
2753 kmem_free(buf
, drrs
->drr_length
);
2757 return (SET_ERROR(EINVAL
));
2762 * Commit the records to the pool.
2765 receive_process_record(struct receive_writer_arg
*rwa
,
2766 struct receive_record_arg
*rrd
)
2770 /* Processing in order, therefore bytes_read should be increasing. */
2771 ASSERT3U(rrd
->bytes_read
, >=, rwa
->bytes_read
);
2772 rwa
->bytes_read
= rrd
->bytes_read
;
2774 switch (rrd
->header
.drr_type
) {
2777 struct drr_object
*drro
= &rrd
->header
.drr_u
.drr_object
;
2778 err
= receive_object(rwa
, drro
, rrd
->payload
);
2779 kmem_free(rrd
->payload
, rrd
->payload_size
);
2780 rrd
->payload
= NULL
;
2783 case DRR_FREEOBJECTS
:
2785 struct drr_freeobjects
*drrfo
=
2786 &rrd
->header
.drr_u
.drr_freeobjects
;
2787 return (receive_freeobjects(rwa
, drrfo
));
2791 struct drr_write
*drrw
= &rrd
->header
.drr_u
.drr_write
;
2792 err
= receive_write(rwa
, drrw
, rrd
->write_buf
);
2793 /* if receive_write() is successful, it consumes the arc_buf */
2795 dmu_return_arcbuf(rrd
->write_buf
);
2796 rrd
->write_buf
= NULL
;
2797 rrd
->payload
= NULL
;
2800 case DRR_WRITE_BYREF
:
2802 struct drr_write_byref
*drrwbr
=
2803 &rrd
->header
.drr_u
.drr_write_byref
;
2804 return (receive_write_byref(rwa
, drrwbr
));
2806 case DRR_WRITE_EMBEDDED
:
2808 struct drr_write_embedded
*drrwe
=
2809 &rrd
->header
.drr_u
.drr_write_embedded
;
2810 err
= receive_write_embedded(rwa
, drrwe
, rrd
->payload
);
2811 kmem_free(rrd
->payload
, rrd
->payload_size
);
2812 rrd
->payload
= NULL
;
2817 struct drr_free
*drrf
= &rrd
->header
.drr_u
.drr_free
;
2818 return (receive_free(rwa
, drrf
));
2822 struct drr_spill
*drrs
= &rrd
->header
.drr_u
.drr_spill
;
2823 err
= receive_spill(rwa
, drrs
, rrd
->payload
);
2824 kmem_free(rrd
->payload
, rrd
->payload_size
);
2825 rrd
->payload
= NULL
;
2829 return (SET_ERROR(EINVAL
));
2834 * dmu_recv_stream's worker thread; pull records off the queue, and then call
2835 * receive_process_record When we're done, signal the main thread and exit.
2838 receive_writer_thread(void *arg
)
2840 struct receive_writer_arg
*rwa
= arg
;
2841 struct receive_record_arg
*rrd
;
2842 fstrans_cookie_t cookie
= spl_fstrans_mark();
2844 for (rrd
= bqueue_dequeue(&rwa
->q
); !rrd
->eos_marker
;
2845 rrd
= bqueue_dequeue(&rwa
->q
)) {
2847 * If there's an error, the main thread will stop putting things
2848 * on the queue, but we need to clear everything in it before we
2851 if (rwa
->err
== 0) {
2852 rwa
->err
= receive_process_record(rwa
, rrd
);
2853 } else if (rrd
->write_buf
!= NULL
) {
2854 dmu_return_arcbuf(rrd
->write_buf
);
2855 rrd
->write_buf
= NULL
;
2856 rrd
->payload
= NULL
;
2857 } else if (rrd
->payload
!= NULL
) {
2858 kmem_free(rrd
->payload
, rrd
->payload_size
);
2859 rrd
->payload
= NULL
;
2861 kmem_free(rrd
, sizeof (*rrd
));
2863 kmem_free(rrd
, sizeof (*rrd
));
2864 mutex_enter(&rwa
->mutex
);
2866 cv_signal(&rwa
->cv
);
2867 mutex_exit(&rwa
->mutex
);
2868 spl_fstrans_unmark(cookie
);
2872 resume_check(struct receive_arg
*ra
, nvlist_t
*begin_nvl
)
2875 objset_t
*mos
= dmu_objset_pool(ra
->os
)->dp_meta_objset
;
2876 uint64_t dsobj
= dmu_objset_id(ra
->os
);
2877 uint64_t resume_obj
, resume_off
;
2879 if (nvlist_lookup_uint64(begin_nvl
,
2880 "resume_object", &resume_obj
) != 0 ||
2881 nvlist_lookup_uint64(begin_nvl
,
2882 "resume_offset", &resume_off
) != 0) {
2883 return (SET_ERROR(EINVAL
));
2885 VERIFY0(zap_lookup(mos
, dsobj
,
2886 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
));
2887 if (resume_obj
!= val
)
2888 return (SET_ERROR(EINVAL
));
2889 VERIFY0(zap_lookup(mos
, dsobj
,
2890 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
));
2891 if (resume_off
!= val
)
2892 return (SET_ERROR(EINVAL
));
2898 * Read in the stream's records, one by one, and apply them to the pool. There
2899 * are two threads involved; the thread that calls this function will spin up a
2900 * worker thread, read the records off the stream one by one, and issue
2901 * prefetches for any necessary indirect blocks. It will then push the records
2902 * onto an internal blocking queue. The worker thread will pull the records off
2903 * the queue, and actually write the data into the DMU. This way, the worker
2904 * thread doesn't have to wait for reads to complete, since everything it needs
2905 * (the indirect blocks) will be prefetched.
2907 * NB: callers *must* call dmu_recv_end() if this succeeds.
2910 dmu_recv_stream(dmu_recv_cookie_t
*drc
, vnode_t
*vp
, offset_t
*voffp
,
2911 int cleanup_fd
, uint64_t *action_handlep
)
2914 struct receive_arg
*ra
;
2915 struct receive_writer_arg
*rwa
;
2917 uint32_t payloadlen
;
2919 nvlist_t
*begin_nvl
= NULL
;
2921 ra
= kmem_zalloc(sizeof (*ra
), KM_SLEEP
);
2922 rwa
= kmem_zalloc(sizeof (*rwa
), KM_SLEEP
);
2924 ra
->byteswap
= drc
->drc_byteswap
;
2925 ra
->cksum
= drc
->drc_cksum
;
2929 if (dsl_dataset_is_zapified(drc
->drc_ds
)) {
2930 (void) zap_lookup(drc
->drc_ds
->ds_dir
->dd_pool
->dp_meta_objset
,
2931 drc
->drc_ds
->ds_object
, DS_FIELD_RESUME_BYTES
,
2932 sizeof (ra
->bytes_read
), 1, &ra
->bytes_read
);
2935 objlist_create(&ra
->ignore_objlist
);
2937 /* these were verified in dmu_recv_begin */
2938 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc
->drc_drrb
->drr_versioninfo
), ==,
2940 ASSERT3U(drc
->drc_drrb
->drr_type
, <, DMU_OST_NUMTYPES
);
2943 * Open the objset we are modifying.
2945 VERIFY0(dmu_objset_from_ds(drc
->drc_ds
, &ra
->os
));
2947 ASSERT(dsl_dataset_phys(drc
->drc_ds
)->ds_flags
& DS_FLAG_INCONSISTENT
);
2949 featureflags
= DMU_GET_FEATUREFLAGS(drc
->drc_drrb
->drr_versioninfo
);
2951 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
2952 if (featureflags
& DMU_BACKUP_FEATURE_DEDUP
) {
2955 if (cleanup_fd
== -1) {
2956 ra
->err
= SET_ERROR(EBADF
);
2959 ra
->err
= zfs_onexit_fd_hold(cleanup_fd
, &minor
);
2965 if (*action_handlep
== 0) {
2966 rwa
->guid_to_ds_map
=
2967 kmem_alloc(sizeof (avl_tree_t
), KM_SLEEP
);
2968 avl_create(rwa
->guid_to_ds_map
, guid_compare
,
2969 sizeof (guid_map_entry_t
),
2970 offsetof(guid_map_entry_t
, avlnode
));
2971 err
= zfs_onexit_add_cb(minor
,
2972 free_guid_map_onexit
, rwa
->guid_to_ds_map
,
2977 err
= zfs_onexit_cb_data(minor
, *action_handlep
,
2978 (void **)&rwa
->guid_to_ds_map
);
2983 drc
->drc_guid_to_ds_map
= rwa
->guid_to_ds_map
;
2986 payloadlen
= drc
->drc_drr_begin
->drr_payloadlen
;
2988 if (payloadlen
!= 0)
2989 payload
= kmem_alloc(payloadlen
, KM_SLEEP
);
2991 err
= receive_read_payload_and_next_header(ra
, payloadlen
, payload
);
2993 if (payloadlen
!= 0)
2994 kmem_free(payload
, payloadlen
);
2997 if (payloadlen
!= 0) {
2998 err
= nvlist_unpack(payload
, payloadlen
, &begin_nvl
, KM_SLEEP
);
2999 kmem_free(payload
, payloadlen
);
3004 if (featureflags
& DMU_BACKUP_FEATURE_RESUMING
) {
3005 err
= resume_check(ra
, begin_nvl
);
3010 (void) bqueue_init(&rwa
->q
, zfs_recv_queue_length
,
3011 offsetof(struct receive_record_arg
, node
));
3012 cv_init(&rwa
->cv
, NULL
, CV_DEFAULT
, NULL
);
3013 mutex_init(&rwa
->mutex
, NULL
, MUTEX_DEFAULT
, NULL
);
3015 rwa
->byteswap
= drc
->drc_byteswap
;
3016 rwa
->resumable
= drc
->drc_resumable
;
3018 (void) thread_create(NULL
, 0, receive_writer_thread
, rwa
, 0, curproc
,
3019 TS_RUN
, minclsyspri
);
3021 * We're reading rwa->err without locks, which is safe since we are the
3022 * only reader, and the worker thread is the only writer. It's ok if we
3023 * miss a write for an iteration or two of the loop, since the writer
3024 * thread will keep freeing records we send it until we send it an eos
3027 * We can leave this loop in 3 ways: First, if rwa->err is
3028 * non-zero. In that case, the writer thread will free the rrd we just
3029 * pushed. Second, if we're interrupted; in that case, either it's the
3030 * first loop and ra->rrd was never allocated, or it's later and ra->rrd
3031 * has been handed off to the writer thread who will free it. Finally,
3032 * if receive_read_record fails or we're at the end of the stream, then
3033 * we free ra->rrd and exit.
3035 while (rwa
->err
== 0) {
3036 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
3037 err
= SET_ERROR(EINTR
);
3041 ASSERT3P(ra
->rrd
, ==, NULL
);
3042 ra
->rrd
= ra
->next_rrd
;
3043 ra
->next_rrd
= NULL
;
3044 /* Allocates and loads header into ra->next_rrd */
3045 err
= receive_read_record(ra
);
3047 if (ra
->rrd
->header
.drr_type
== DRR_END
|| err
!= 0) {
3048 kmem_free(ra
->rrd
, sizeof (*ra
->rrd
));
3053 bqueue_enqueue(&rwa
->q
, ra
->rrd
,
3054 sizeof (struct receive_record_arg
) + ra
->rrd
->payload_size
);
3057 if (ra
->next_rrd
== NULL
)
3058 ra
->next_rrd
= kmem_zalloc(sizeof (*ra
->next_rrd
), KM_SLEEP
);
3059 ra
->next_rrd
->eos_marker
= B_TRUE
;
3060 bqueue_enqueue(&rwa
->q
, ra
->next_rrd
, 1);
3062 mutex_enter(&rwa
->mutex
);
3063 while (!rwa
->done
) {
3064 cv_wait(&rwa
->cv
, &rwa
->mutex
);
3066 mutex_exit(&rwa
->mutex
);
3068 cv_destroy(&rwa
->cv
);
3069 mutex_destroy(&rwa
->mutex
);
3070 bqueue_destroy(&rwa
->q
);
3075 nvlist_free(begin_nvl
);
3076 if ((featureflags
& DMU_BACKUP_FEATURE_DEDUP
) && (cleanup_fd
!= -1))
3077 zfs_onexit_fd_rele(cleanup_fd
);
3081 * Clean up references. If receive is not resumable,
3082 * destroy what we created, so we don't leave it in
3083 * the inconsistent state.
3085 dmu_recv_cleanup_ds(drc
);
3089 objlist_destroy(&ra
->ignore_objlist
);
3090 kmem_free(ra
, sizeof (*ra
));
3091 kmem_free(rwa
, sizeof (*rwa
));
3096 dmu_recv_end_check(void *arg
, dmu_tx_t
*tx
)
3098 dmu_recv_cookie_t
*drc
= arg
;
3099 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3102 ASSERT3P(drc
->drc_ds
->ds_owner
, ==, dmu_recv_tag
);
3104 if (!drc
->drc_newfs
) {
3105 dsl_dataset_t
*origin_head
;
3107 error
= dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
, &origin_head
);
3110 if (drc
->drc_force
) {
3112 * We will destroy any snapshots in tofs (i.e. before
3113 * origin_head) that are after the origin (which is
3114 * the snap before drc_ds, because drc_ds can not
3115 * have any snaps of its own).
3119 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3121 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3122 dsl_dataset_t
*snap
;
3123 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3127 if (snap
->ds_dir
!= origin_head
->ds_dir
)
3128 error
= SET_ERROR(EINVAL
);
3130 error
= dsl_destroy_snapshot_check_impl(
3133 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3134 dsl_dataset_rele(snap
, FTAG
);
3139 dsl_dataset_rele(origin_head
, FTAG
);
3143 error
= dsl_dataset_clone_swap_check_impl(drc
->drc_ds
,
3144 origin_head
, drc
->drc_force
, drc
->drc_owner
, tx
);
3146 dsl_dataset_rele(origin_head
, FTAG
);
3149 error
= dsl_dataset_snapshot_check_impl(origin_head
,
3150 drc
->drc_tosnap
, tx
, B_TRUE
, 1, drc
->drc_cred
);
3151 dsl_dataset_rele(origin_head
, FTAG
);
3155 error
= dsl_destroy_head_check_impl(drc
->drc_ds
, 1);
3157 error
= dsl_dataset_snapshot_check_impl(drc
->drc_ds
,
3158 drc
->drc_tosnap
, tx
, B_TRUE
, 1, drc
->drc_cred
);
3164 dmu_recv_end_sync(void *arg
, dmu_tx_t
*tx
)
3166 dmu_recv_cookie_t
*drc
= arg
;
3167 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3169 spa_history_log_internal_ds(drc
->drc_ds
, "finish receiving",
3170 tx
, "snap=%s", drc
->drc_tosnap
);
3172 if (!drc
->drc_newfs
) {
3173 dsl_dataset_t
*origin_head
;
3175 VERIFY0(dsl_dataset_hold(dp
, drc
->drc_tofs
, FTAG
,
3178 if (drc
->drc_force
) {
3180 * Destroy any snapshots of drc_tofs (origin_head)
3181 * after the origin (the snap before drc_ds).
3185 obj
= dsl_dataset_phys(origin_head
)->ds_prev_snap_obj
;
3187 dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
) {
3188 dsl_dataset_t
*snap
;
3189 VERIFY0(dsl_dataset_hold_obj(dp
, obj
, FTAG
,
3191 ASSERT3P(snap
->ds_dir
, ==, origin_head
->ds_dir
);
3192 obj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3193 dsl_destroy_snapshot_sync_impl(snap
,
3195 dsl_dataset_rele(snap
, FTAG
);
3198 VERIFY3P(drc
->drc_ds
->ds_prev
, ==,
3199 origin_head
->ds_prev
);
3201 dsl_dataset_clone_swap_sync_impl(drc
->drc_ds
,
3203 dsl_dataset_snapshot_sync_impl(origin_head
,
3204 drc
->drc_tosnap
, tx
);
3206 /* set snapshot's creation time and guid */
3207 dmu_buf_will_dirty(origin_head
->ds_prev
->ds_dbuf
, tx
);
3208 dsl_dataset_phys(origin_head
->ds_prev
)->ds_creation_time
=
3209 drc
->drc_drrb
->drr_creation_time
;
3210 dsl_dataset_phys(origin_head
->ds_prev
)->ds_guid
=
3211 drc
->drc_drrb
->drr_toguid
;
3212 dsl_dataset_phys(origin_head
->ds_prev
)->ds_flags
&=
3213 ~DS_FLAG_INCONSISTENT
;
3215 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
3216 dsl_dataset_phys(origin_head
)->ds_flags
&=
3217 ~DS_FLAG_INCONSISTENT
;
3219 dsl_dataset_rele(origin_head
, FTAG
);
3220 dsl_destroy_head_sync_impl(drc
->drc_ds
, tx
);
3222 if (drc
->drc_owner
!= NULL
)
3223 VERIFY3P(origin_head
->ds_owner
, ==, drc
->drc_owner
);
3225 dsl_dataset_t
*ds
= drc
->drc_ds
;
3227 dsl_dataset_snapshot_sync_impl(ds
, drc
->drc_tosnap
, tx
);
3229 /* set snapshot's creation time and guid */
3230 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
3231 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_time
=
3232 drc
->drc_drrb
->drr_creation_time
;
3233 dsl_dataset_phys(ds
->ds_prev
)->ds_guid
=
3234 drc
->drc_drrb
->drr_toguid
;
3235 dsl_dataset_phys(ds
->ds_prev
)->ds_flags
&=
3236 ~DS_FLAG_INCONSISTENT
;
3238 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3239 dsl_dataset_phys(ds
)->ds_flags
&= ~DS_FLAG_INCONSISTENT
;
3240 if (dsl_dataset_has_resume_receive_state(ds
)) {
3241 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3242 DS_FIELD_RESUME_FROMGUID
, tx
);
3243 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3244 DS_FIELD_RESUME_OBJECT
, tx
);
3245 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3246 DS_FIELD_RESUME_OFFSET
, tx
);
3247 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3248 DS_FIELD_RESUME_BYTES
, tx
);
3249 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3250 DS_FIELD_RESUME_TOGUID
, tx
);
3251 (void) zap_remove(dp
->dp_meta_objset
, ds
->ds_object
,
3252 DS_FIELD_RESUME_TONAME
, tx
);
3255 drc
->drc_newsnapobj
= dsl_dataset_phys(drc
->drc_ds
)->ds_prev_snap_obj
;
3256 zvol_create_minors(dp
->dp_spa
, drc
->drc_tofs
, B_TRUE
);
3258 * Release the hold from dmu_recv_begin. This must be done before
3259 * we return to open context, so that when we free the dataset's dnode,
3260 * we can evict its bonus buffer.
3262 dsl_dataset_disown(drc
->drc_ds
, dmu_recv_tag
);
3267 add_ds_to_guidmap(const char *name
, avl_tree_t
*guid_map
, uint64_t snapobj
)
3270 dsl_dataset_t
*snapds
;
3271 guid_map_entry_t
*gmep
;
3274 ASSERT(guid_map
!= NULL
);
3276 err
= dsl_pool_hold(name
, FTAG
, &dp
);
3279 gmep
= kmem_alloc(sizeof (*gmep
), KM_SLEEP
);
3280 err
= dsl_dataset_hold_obj(dp
, snapobj
, gmep
, &snapds
);
3282 gmep
->guid
= dsl_dataset_phys(snapds
)->ds_guid
;
3283 gmep
->gme_ds
= snapds
;
3284 avl_add(guid_map
, gmep
);
3285 dsl_dataset_long_hold(snapds
, gmep
);
3287 kmem_free(gmep
, sizeof (*gmep
));
3290 dsl_pool_rele(dp
, FTAG
);
3294 static int dmu_recv_end_modified_blocks
= 3;
3297 dmu_recv_existing_end(dmu_recv_cookie_t
*drc
)
3303 * We will be destroying the ds; make sure its origin is unmounted if
3306 char name
[ZFS_MAX_DATASET_NAME_LEN
];
3307 dsl_dataset_name(drc
->drc_ds
, name
);
3308 zfs_destroy_unmount_origin(name
);
3311 error
= dsl_sync_task(drc
->drc_tofs
,
3312 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3313 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
);
3316 dmu_recv_cleanup_ds(drc
);
3321 dmu_recv_new_end(dmu_recv_cookie_t
*drc
)
3325 error
= dsl_sync_task(drc
->drc_tofs
,
3326 dmu_recv_end_check
, dmu_recv_end_sync
, drc
,
3327 dmu_recv_end_modified_blocks
, ZFS_SPACE_CHECK_NORMAL
);
3330 dmu_recv_cleanup_ds(drc
);
3331 } else if (drc
->drc_guid_to_ds_map
!= NULL
) {
3332 (void) add_ds_to_guidmap(drc
->drc_tofs
,
3333 drc
->drc_guid_to_ds_map
,
3334 drc
->drc_newsnapobj
);
3340 dmu_recv_end(dmu_recv_cookie_t
*drc
, void *owner
)
3342 drc
->drc_owner
= owner
;
3345 return (dmu_recv_new_end(drc
));
3347 return (dmu_recv_existing_end(drc
));
3351 * Return TRUE if this objset is currently being received into.
3354 dmu_objset_is_receiving(objset_t
*os
)
3356 return (os
->os_dsl_dataset
!= NULL
&&
3357 os
->os_dsl_dataset
->ds_owner
== dmu_recv_tag
);
3360 #if defined(_KERNEL)
3361 module_param(zfs_send_corrupt_data
, int, 0644);
3362 MODULE_PARM_DESC(zfs_send_corrupt_data
, "Allow sending corrupt data");