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 (c) 2011 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011 by Delphix. All rights reserved.
26 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
30 #include <sys/dmu_impl.h>
31 #include <sys/dmu_tx.h>
33 #include <sys/dnode.h>
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_dir.h>
39 #include <sys/dsl_prop.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_synctask.h>
42 #include <sys/zfs_ioctl.h>
44 #include <sys/zio_checksum.h>
45 #include <sys/zfs_znode.h>
46 #include <zfs_fletcher.h>
49 #include <sys/zfs_onexit.h>
51 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
52 int zfs_send_corrupt_data = B_FALSE;
54 static char *dmu_recv_tag = "dmu_recv_tag";
57 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
59 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
60 ssize_t resid; /* have to get resid to get detailed errno */
61 ASSERT3U(len % 8, ==, 0);
63 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
64 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
66 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
68 mutex_enter(&ds->ds_sendstream_lock);
70 mutex_exit(&ds->ds_sendstream_lock);
72 return (dsp->dsa_err);
76 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
79 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
81 if (length != -1ULL && offset + length < offset)
85 * If there is a pending op, but it's not PENDING_FREE, push it out,
86 * since free block aggregation can only be done for blocks of the
87 * same type (i.e., DRR_FREE records can only be aggregated with
88 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
89 * aggregated with other DRR_FREEOBJECTS records.
91 if (dsp->dsa_pending_op != PENDING_NONE &&
92 dsp->dsa_pending_op != PENDING_FREE) {
93 if (dump_bytes(dsp, dsp->dsa_drr,
94 sizeof (dmu_replay_record_t)) != 0)
96 dsp->dsa_pending_op = PENDING_NONE;
99 if (dsp->dsa_pending_op == PENDING_FREE) {
101 * There should never be a PENDING_FREE if length is -1
102 * (because dump_dnode is the only place where this
103 * function is called with a -1, and only after flushing
104 * any pending record).
106 ASSERT(length != -1ULL);
108 * Check to see whether this free block can be aggregated
111 if (drrf->drr_object == object && drrf->drr_offset +
112 drrf->drr_length == offset) {
113 drrf->drr_length += length;
116 /* not a continuation. Push out pending record */
117 if (dump_bytes(dsp, dsp->dsa_drr,
118 sizeof (dmu_replay_record_t)) != 0)
120 dsp->dsa_pending_op = PENDING_NONE;
123 /* create a FREE record and make it pending */
124 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
125 dsp->dsa_drr->drr_type = DRR_FREE;
126 drrf->drr_object = object;
127 drrf->drr_offset = offset;
128 drrf->drr_length = length;
129 drrf->drr_toguid = dsp->dsa_toguid;
130 if (length == -1ULL) {
131 if (dump_bytes(dsp, dsp->dsa_drr,
132 sizeof (dmu_replay_record_t)) != 0)
135 dsp->dsa_pending_op = PENDING_FREE;
142 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
143 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
145 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
149 * If there is any kind of pending aggregation (currently either
150 * a grouping of free objects or free blocks), push it out to
151 * the stream, since aggregation can't be done across operations
152 * of different types.
154 if (dsp->dsa_pending_op != PENDING_NONE) {
155 if (dump_bytes(dsp, dsp->dsa_drr,
156 sizeof (dmu_replay_record_t)) != 0)
158 dsp->dsa_pending_op = PENDING_NONE;
160 /* write a DATA record */
161 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
162 dsp->dsa_drr->drr_type = DRR_WRITE;
163 drrw->drr_object = object;
164 drrw->drr_type = type;
165 drrw->drr_offset = offset;
166 drrw->drr_length = blksz;
167 drrw->drr_toguid = dsp->dsa_toguid;
168 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
169 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
170 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
171 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
172 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
173 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
174 drrw->drr_key.ddk_cksum = bp->blk_cksum;
176 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
178 if (dump_bytes(dsp, data, blksz) != 0)
184 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
186 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
188 if (dsp->dsa_pending_op != PENDING_NONE) {
189 if (dump_bytes(dsp, dsp->dsa_drr,
190 sizeof (dmu_replay_record_t)) != 0)
192 dsp->dsa_pending_op = PENDING_NONE;
195 /* write a SPILL record */
196 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
197 dsp->dsa_drr->drr_type = DRR_SPILL;
198 drrs->drr_object = object;
199 drrs->drr_length = blksz;
200 drrs->drr_toguid = dsp->dsa_toguid;
202 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
204 if (dump_bytes(dsp, data, blksz))
210 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
212 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
215 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
216 * push it out, since free block aggregation can only be done for
217 * blocks of the same type (i.e., DRR_FREE records can only be
218 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
219 * can only be aggregated with other DRR_FREEOBJECTS records.
221 if (dsp->dsa_pending_op != PENDING_NONE &&
222 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
223 if (dump_bytes(dsp, dsp->dsa_drr,
224 sizeof (dmu_replay_record_t)) != 0)
226 dsp->dsa_pending_op = PENDING_NONE;
228 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
230 * See whether this free object array can be aggregated
233 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
234 drrfo->drr_numobjs += numobjs;
237 /* can't be aggregated. Push out pending record */
238 if (dump_bytes(dsp, dsp->dsa_drr,
239 sizeof (dmu_replay_record_t)) != 0)
241 dsp->dsa_pending_op = PENDING_NONE;
245 /* write a FREEOBJECTS record */
246 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
247 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
248 drrfo->drr_firstobj = firstobj;
249 drrfo->drr_numobjs = numobjs;
250 drrfo->drr_toguid = dsp->dsa_toguid;
252 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
258 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
260 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
262 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
263 return (dump_freeobjects(dsp, object, 1));
265 if (dsp->dsa_pending_op != PENDING_NONE) {
266 if (dump_bytes(dsp, dsp->dsa_drr,
267 sizeof (dmu_replay_record_t)) != 0)
269 dsp->dsa_pending_op = PENDING_NONE;
272 /* write an OBJECT record */
273 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
274 dsp->dsa_drr->drr_type = DRR_OBJECT;
275 drro->drr_object = object;
276 drro->drr_type = dnp->dn_type;
277 drro->drr_bonustype = dnp->dn_bonustype;
278 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
279 drro->drr_bonuslen = dnp->dn_bonuslen;
280 drro->drr_checksumtype = dnp->dn_checksum;
281 drro->drr_compress = dnp->dn_compress;
282 drro->drr_toguid = dsp->dsa_toguid;
284 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
287 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
290 /* free anything past the end of the file */
291 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
292 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
299 #define BP_SPAN(dnp, level) \
300 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
301 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
305 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
306 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
308 dmu_sendarg_t *dsp = arg;
309 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
312 if (issig(JUSTLOOKING) && issig(FORREAL))
315 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
316 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
318 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
319 uint64_t span = BP_SPAN(dnp, zb->zb_level);
320 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
321 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
322 } else if (bp == NULL) {
323 uint64_t span = BP_SPAN(dnp, zb->zb_level);
324 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
325 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
327 } else if (type == DMU_OT_DNODE) {
330 int blksz = BP_GET_LSIZE(bp);
331 uint32_t aflags = ARC_WAIT;
334 if (dsl_read(NULL, spa, bp, pbuf,
335 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
336 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
340 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
341 uint64_t dnobj = (zb->zb_blkid <<
342 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
343 err = dump_dnode(dsp, dnobj, blk+i);
347 (void) arc_buf_remove_ref(abuf, &abuf);
348 } else if (type == DMU_OT_SA) {
349 uint32_t aflags = ARC_WAIT;
351 int blksz = BP_GET_LSIZE(bp);
353 if (arc_read_nolock(NULL, spa, bp,
354 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
355 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
358 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
359 (void) arc_buf_remove_ref(abuf, &abuf);
360 } else { /* it's a level-0 block of a regular object */
361 uint32_t aflags = ARC_WAIT;
363 int blksz = BP_GET_LSIZE(bp);
365 if (dsl_read(NULL, spa, bp, pbuf,
366 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
367 ZIO_FLAG_CANFAIL, &aflags, zb) != 0) {
368 if (zfs_send_corrupt_data) {
370 /* Send a block filled with 0x"zfs badd bloc" */
371 abuf = arc_buf_alloc(spa, blksz, &abuf,
373 for (ptr = abuf->b_data;
374 (char *)ptr < (char *)abuf->b_data + blksz;
376 *ptr = 0x2f5baddb10c;
382 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
383 blksz, bp, abuf->b_data);
384 (void) arc_buf_remove_ref(abuf, &abuf);
387 ASSERT(err == 0 || err == EINTR);
392 dmu_send(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
393 int outfd, vnode_t *vp, offset_t *off)
395 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
396 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
397 dmu_replay_record_t *drr;
400 uint64_t fromtxg = 0;
402 /* tosnap must be a snapshot */
403 if (ds->ds_phys->ds_next_snap_obj == 0)
406 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
407 if (fromds && (ds->ds_dir != fromds->ds_dir ||
408 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
412 dsl_pool_t *dp = ds->ds_dir->dd_pool;
417 if (dsl_dir_is_clone(ds->ds_dir)) {
418 rw_enter(&dp->dp_config_rwlock, RW_READER);
419 err = dsl_dataset_hold_obj(dp,
420 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
421 rw_exit(&dp->dp_config_rwlock);
425 fromorigin = B_FALSE;
430 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
431 drr->drr_type = DRR_BEGIN;
432 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
433 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
437 if (dmu_objset_type(tosnap) == DMU_OST_ZFS) {
439 if (zfs_get_zplprop(tosnap, ZFS_PROP_VERSION, &version) != 0) {
440 kmem_free(drr, sizeof (dmu_replay_record_t));
443 if (version == ZPL_VERSION_SA) {
444 DMU_SET_FEATUREFLAGS(
445 drr->drr_u.drr_begin.drr_versioninfo,
446 DMU_BACKUP_FEATURE_SA_SPILL);
451 drr->drr_u.drr_begin.drr_creation_time =
452 ds->ds_phys->ds_creation_time;
453 drr->drr_u.drr_begin.drr_type = tosnap->os_phys->os_type;
455 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
456 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
457 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
458 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
461 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
462 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
465 fromtxg = fromds->ds_phys->ds_creation_txg;
467 dsl_dataset_rele(fromds, FTAG);
469 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
473 dsp->dsa_outfd = outfd;
474 dsp->dsa_proc = curproc;
475 dsp->dsa_os = tosnap;
477 dsp->dsa_toguid = ds->ds_phys->ds_guid;
478 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
479 dsp->dsa_pending_op = PENDING_NONE;
481 mutex_enter(&ds->ds_sendstream_lock);
482 list_insert_head(&ds->ds_sendstreams, dsp);
483 mutex_exit(&ds->ds_sendstream_lock);
485 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
490 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
493 if (dsp->dsa_pending_op != PENDING_NONE)
494 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
498 if (err == EINTR && dsp->dsa_err)
503 bzero(drr, sizeof (dmu_replay_record_t));
504 drr->drr_type = DRR_END;
505 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
506 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
508 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
514 mutex_enter(&ds->ds_sendstream_lock);
515 list_remove(&ds->ds_sendstreams, dsp);
516 mutex_exit(&ds->ds_sendstream_lock);
518 kmem_free(drr, sizeof (dmu_replay_record_t));
519 kmem_free(dsp, sizeof (dmu_sendarg_t));
525 dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
528 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
529 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
530 dsl_pool_t *dp = ds->ds_dir->dd_pool;
532 uint64_t size, recordsize;
534 /* tosnap must be a snapshot */
535 if (ds->ds_phys->ds_next_snap_obj == 0)
538 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
539 if (fromds && (ds->ds_dir != fromds->ds_dir ||
540 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
547 if (dsl_dir_is_clone(ds->ds_dir)) {
548 rw_enter(&dp->dp_config_rwlock, RW_READER);
549 err = dsl_dataset_hold_obj(dp,
550 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
551 rw_exit(&dp->dp_config_rwlock);
555 fromorigin = B_FALSE;
559 /* Get uncompressed size estimate of changed data. */
560 if (fromds == NULL) {
561 size = ds->ds_phys->ds_uncompressed_bytes;
564 err = dsl_dataset_space_written(fromds, ds,
565 &used, &comp, &size);
567 dsl_dataset_rele(fromds, FTAG);
573 * Assume that space (both on-disk and in-stream) is dominated by
574 * data. We will adjust for indirect blocks and the copies property,
575 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
579 * Subtract out approximate space used by indirect blocks.
580 * Assume most space is used by data blocks (non-indirect, non-dnode).
581 * Assume all blocks are recordsize. Assume ditto blocks and
582 * internal fragmentation counter out compression.
584 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
585 * block, which we observe in practice.
587 rw_enter(&dp->dp_config_rwlock, RW_READER);
588 err = dsl_prop_get_ds(ds, "recordsize",
589 sizeof (recordsize), 1, &recordsize, NULL);
590 rw_exit(&dp->dp_config_rwlock);
593 size -= size / recordsize * sizeof (blkptr_t);
595 /* Add in the space for the record associated with each block. */
596 size += size / recordsize * sizeof (dmu_replay_record_t);
603 struct recvbeginsyncarg {
606 dsl_dataset_t *origin;
608 dmu_objset_type_t type;
612 char clonelastname[MAXNAMELEN];
613 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
619 recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx)
621 dsl_dir_t *dd = arg1;
622 struct recvbeginsyncarg *rbsa = arg2;
623 objset_t *mos = dd->dd_pool->dp_meta_objset;
627 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
628 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
631 return (err ? err : EEXIST);
634 /* make sure it's a snap in the same pool */
635 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
637 if (!dsl_dataset_is_snapshot(rbsa->origin))
639 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
647 recv_new_sync(void *arg1, void *arg2, dmu_tx_t *tx)
649 dsl_dir_t *dd = arg1;
650 struct recvbeginsyncarg *rbsa = arg2;
651 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
654 /* Create and open new dataset. */
655 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
656 rbsa->origin, flags, rbsa->cr, tx);
657 VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj,
658 B_TRUE, dmu_recv_tag, &rbsa->ds));
660 if (rbsa->origin == NULL) {
661 (void) dmu_objset_create_impl(dd->dd_pool->dp_spa,
662 rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx);
665 spa_history_log_internal(LOG_DS_REPLAY_FULL_SYNC,
666 dd->dd_pool->dp_spa, tx, "dataset = %lld", dsobj);
671 recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
673 dsl_dataset_t *ds = arg1;
674 struct recvbeginsyncarg *rbsa = arg2;
678 /* must not have any changes since most recent snapshot */
679 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
682 /* new snapshot name must not exist */
683 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
684 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
690 if (rbsa->fromguid) {
691 /* if incremental, most recent snapshot must match fromguid */
692 if (ds->ds_prev == NULL)
696 * most recent snapshot must match fromguid, or there are no
697 * changes since the fromguid one
699 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) {
700 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth;
701 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj;
704 err = dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
708 if (snap->ds_phys->ds_creation_txg < birth) {
709 dsl_dataset_rele(snap, FTAG);
712 if (snap->ds_phys->ds_guid == rbsa->fromguid) {
713 dsl_dataset_rele(snap, FTAG);
716 obj = snap->ds_phys->ds_prev_snap_obj;
717 dsl_dataset_rele(snap, FTAG);
723 /* if full, most recent snapshot must be $ORIGIN */
724 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
728 /* temporary clone name must not exist */
729 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
730 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
731 rbsa->clonelastname, 8, 1, &val);
742 recv_existing_sync(void *arg1, void *arg2, dmu_tx_t *tx)
744 dsl_dataset_t *ohds = arg1;
745 struct recvbeginsyncarg *rbsa = arg2;
746 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
748 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
751 /* create and open the temporary clone */
752 dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname,
753 ohds->ds_prev, flags, rbsa->cr, tx);
754 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, B_TRUE, dmu_recv_tag, &cds));
757 * If we actually created a non-clone, we need to create the
758 * objset in our new dataset.
760 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) {
761 (void) dmu_objset_create_impl(dp->dp_spa,
762 cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx);
767 spa_history_log_internal(LOG_DS_REPLAY_INC_SYNC,
768 dp->dp_spa, tx, "dataset = %lld", dsobj);
772 dmu_recv_verify_features(dsl_dataset_t *ds, struct drr_begin *drrb)
776 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
778 /* Verify pool version supports SA if SA_SPILL feature set */
779 return ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
780 (spa_version(dsl_dataset_get_spa(ds)) < SPA_VERSION_SA));
784 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
785 * succeeds; otherwise we will leak the holds on the datasets.
788 dmu_recv_begin(char *tofs, char *tosnap, char *top_ds, struct drr_begin *drrb,
789 boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
793 struct recvbeginsyncarg rbsa = { 0 };
794 uint64_t versioninfo;
798 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
800 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
806 rbsa.tosnap = tosnap;
807 rbsa.origin = origin ? origin->os_dsl_dataset : NULL;
808 rbsa.fromguid = drrb->drr_fromguid;
809 rbsa.type = drrb->drr_type;
813 versioninfo = drrb->drr_versioninfo;
814 flags = drrb->drr_flags;
817 rbsa.type = BSWAP_32(rbsa.type);
818 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
819 versioninfo = BSWAP_64(versioninfo);
820 flags = BSWAP_32(flags);
823 if (DMU_GET_STREAM_HDRTYPE(versioninfo) == DMU_COMPOUNDSTREAM ||
824 rbsa.type >= DMU_OST_NUMTYPES ||
825 ((flags & DRR_FLAG_CLONE) && origin == NULL))
828 if (flags & DRR_FLAG_CI_DATA)
829 rbsa.dsflags = DS_FLAG_CI_DATASET;
831 bzero(drc, sizeof (dmu_recv_cookie_t));
832 drc->drc_drrb = drrb;
833 drc->drc_tosnap = tosnap;
834 drc->drc_top_ds = top_ds;
835 drc->drc_force = force;
838 * Process the begin in syncing context.
841 /* open the dataset we are logically receiving into */
842 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
844 if (dmu_recv_verify_features(ds, drrb)) {
845 dsl_dataset_rele(ds, dmu_recv_tag);
848 /* target fs already exists; recv into temp clone */
850 /* Can't recv a clone into an existing fs */
851 if (flags & DRR_FLAG_CLONE) {
852 dsl_dataset_rele(ds, dmu_recv_tag);
856 /* must not have an incremental recv already in progress */
857 if (!mutex_tryenter(&ds->ds_recvlock)) {
858 dsl_dataset_rele(ds, dmu_recv_tag);
862 /* tmp clone name is: tofs/%tosnap" */
863 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
866 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
867 recv_existing_check, recv_existing_sync, ds, &rbsa, 5);
869 mutex_exit(&ds->ds_recvlock);
870 dsl_dataset_rele(ds, dmu_recv_tag);
873 drc->drc_logical_ds = ds;
874 drc->drc_real_ds = rbsa.ds;
875 } else if (err == ENOENT) {
876 /* target fs does not exist; must be a full backup or clone */
880 * If it's a non-clone incremental, we are missing the
881 * target fs, so fail the recv.
883 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE))
886 /* Open the parent of tofs */
887 cp = strrchr(tofs, '/');
889 err = dsl_dataset_hold(tofs, FTAG, &ds);
894 if (dmu_recv_verify_features(ds, drrb)) {
895 dsl_dataset_rele(ds, FTAG);
899 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
900 recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5);
901 dsl_dataset_rele(ds, FTAG);
904 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
905 drc->drc_newfs = B_TRUE;
917 int bufsize; /* amount of memory allocated for buf */
919 avl_tree_t *guid_to_ds_map;
922 typedef struct guid_map_entry {
924 dsl_dataset_t *gme_ds;
929 guid_compare(const void *arg1, const void *arg2)
931 const guid_map_entry_t *gmep1 = arg1;
932 const guid_map_entry_t *gmep2 = arg2;
934 if (gmep1->guid < gmep2->guid)
936 else if (gmep1->guid > gmep2->guid)
942 free_guid_map_onexit(void *arg)
944 avl_tree_t *ca = arg;
946 guid_map_entry_t *gmep;
948 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
949 dsl_dataset_rele(gmep->gme_ds, ca);
950 kmem_free(gmep, sizeof (guid_map_entry_t));
953 kmem_free(ca, sizeof (avl_tree_t));
957 restore_read(struct restorearg *ra, int len)
962 /* some things will require 8-byte alignment, so everything must */
963 ASSERT3U(len % 8, ==, 0);
968 ra->err = vn_rdwr(UIO_READ, ra->vp,
969 (caddr_t)ra->buf + done, len - done,
970 ra->voff, UIO_SYSSPACE, FAPPEND,
971 RLIM64_INFINITY, CRED(), &resid);
973 if (resid == len - done)
975 ra->voff += len - done - resid;
981 ASSERT3U(done, ==, len);
984 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
986 fletcher_4_incremental_native(rv, len, &ra->cksum);
991 backup_byteswap(dmu_replay_record_t *drr)
993 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
994 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
995 drr->drr_type = BSWAP_32(drr->drr_type);
996 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
997 switch (drr->drr_type) {
999 DO64(drr_begin.drr_magic);
1000 DO64(drr_begin.drr_versioninfo);
1001 DO64(drr_begin.drr_creation_time);
1002 DO32(drr_begin.drr_type);
1003 DO32(drr_begin.drr_flags);
1004 DO64(drr_begin.drr_toguid);
1005 DO64(drr_begin.drr_fromguid);
1008 DO64(drr_object.drr_object);
1009 /* DO64(drr_object.drr_allocation_txg); */
1010 DO32(drr_object.drr_type);
1011 DO32(drr_object.drr_bonustype);
1012 DO32(drr_object.drr_blksz);
1013 DO32(drr_object.drr_bonuslen);
1014 DO64(drr_object.drr_toguid);
1016 case DRR_FREEOBJECTS:
1017 DO64(drr_freeobjects.drr_firstobj);
1018 DO64(drr_freeobjects.drr_numobjs);
1019 DO64(drr_freeobjects.drr_toguid);
1022 DO64(drr_write.drr_object);
1023 DO32(drr_write.drr_type);
1024 DO64(drr_write.drr_offset);
1025 DO64(drr_write.drr_length);
1026 DO64(drr_write.drr_toguid);
1027 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1028 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1029 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1030 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1031 DO64(drr_write.drr_key.ddk_prop);
1033 case DRR_WRITE_BYREF:
1034 DO64(drr_write_byref.drr_object);
1035 DO64(drr_write_byref.drr_offset);
1036 DO64(drr_write_byref.drr_length);
1037 DO64(drr_write_byref.drr_toguid);
1038 DO64(drr_write_byref.drr_refguid);
1039 DO64(drr_write_byref.drr_refobject);
1040 DO64(drr_write_byref.drr_refoffset);
1041 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1042 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1043 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1044 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1045 DO64(drr_write_byref.drr_key.ddk_prop);
1048 DO64(drr_free.drr_object);
1049 DO64(drr_free.drr_offset);
1050 DO64(drr_free.drr_length);
1051 DO64(drr_free.drr_toguid);
1054 DO64(drr_spill.drr_object);
1055 DO64(drr_spill.drr_length);
1056 DO64(drr_spill.drr_toguid);
1059 DO64(drr_end.drr_checksum.zc_word[0]);
1060 DO64(drr_end.drr_checksum.zc_word[1]);
1061 DO64(drr_end.drr_checksum.zc_word[2]);
1062 DO64(drr_end.drr_checksum.zc_word[3]);
1063 DO64(drr_end.drr_toguid);
1073 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1079 if (drro->drr_type == DMU_OT_NONE ||
1080 drro->drr_type >= DMU_OT_NUMTYPES ||
1081 drro->drr_bonustype >= DMU_OT_NUMTYPES ||
1082 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1083 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1084 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1085 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1086 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1087 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1091 err = dmu_object_info(os, drro->drr_object, NULL);
1093 if (err != 0 && err != ENOENT)
1096 if (drro->drr_bonuslen) {
1097 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1102 if (err == ENOENT) {
1103 /* currently free, want to be allocated */
1104 tx = dmu_tx_create(os);
1105 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1106 err = dmu_tx_assign(tx, TXG_WAIT);
1111 err = dmu_object_claim(os, drro->drr_object,
1112 drro->drr_type, drro->drr_blksz,
1113 drro->drr_bonustype, drro->drr_bonuslen, tx);
1116 /* currently allocated, want to be allocated */
1117 err = dmu_object_reclaim(os, drro->drr_object,
1118 drro->drr_type, drro->drr_blksz,
1119 drro->drr_bonustype, drro->drr_bonuslen);
1125 tx = dmu_tx_create(os);
1126 dmu_tx_hold_bonus(tx, drro->drr_object);
1127 err = dmu_tx_assign(tx, TXG_WAIT);
1133 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1135 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1140 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1141 dmu_buf_will_dirty(db, tx);
1143 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1144 bcopy(data, db->db_data, drro->drr_bonuslen);
1146 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
1147 drro->drr_bonuslen);
1149 dmu_buf_rele(db, FTAG);
1157 restore_freeobjects(struct restorearg *ra, objset_t *os,
1158 struct drr_freeobjects *drrfo)
1162 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1165 for (obj = drrfo->drr_firstobj;
1166 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1167 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1170 if (dmu_object_info(os, obj, NULL) != 0)
1173 err = dmu_free_object(os, obj);
1181 restore_write(struct restorearg *ra, objset_t *os,
1182 struct drr_write *drrw)
1188 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1189 drrw->drr_type >= DMU_OT_NUMTYPES)
1192 data = restore_read(ra, drrw->drr_length);
1196 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1199 tx = dmu_tx_create(os);
1201 dmu_tx_hold_write(tx, drrw->drr_object,
1202 drrw->drr_offset, drrw->drr_length);
1203 err = dmu_tx_assign(tx, TXG_WAIT);
1209 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
1210 dmu_write(os, drrw->drr_object,
1211 drrw->drr_offset, drrw->drr_length, data, tx);
1217 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1218 * streams to refer to a copy of the data that is already on the
1219 * system because it came in earlier in the stream. This function
1220 * finds the earlier copy of the data, and uses that copy instead of
1221 * data from the stream to fulfill this write.
1224 restore_write_byref(struct restorearg *ra, objset_t *os,
1225 struct drr_write_byref *drrwbr)
1229 guid_map_entry_t gmesrch;
1230 guid_map_entry_t *gmep;
1232 objset_t *ref_os = NULL;
1235 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1239 * If the GUID of the referenced dataset is different from the
1240 * GUID of the target dataset, find the referenced dataset.
1242 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1243 gmesrch.guid = drrwbr->drr_refguid;
1244 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1248 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1254 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1255 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1259 tx = dmu_tx_create(os);
1261 dmu_tx_hold_write(tx, drrwbr->drr_object,
1262 drrwbr->drr_offset, drrwbr->drr_length);
1263 err = dmu_tx_assign(tx, TXG_WAIT);
1268 dmu_write(os, drrwbr->drr_object,
1269 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1270 dmu_buf_rele(dbp, FTAG);
1276 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1280 dmu_buf_t *db, *db_spill;
1283 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1284 drrs->drr_length > SPA_MAXBLOCKSIZE)
1287 data = restore_read(ra, drrs->drr_length);
1291 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1294 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1295 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1296 dmu_buf_rele(db, FTAG);
1300 tx = dmu_tx_create(os);
1302 dmu_tx_hold_spill(tx, db->db_object);
1304 err = dmu_tx_assign(tx, TXG_WAIT);
1306 dmu_buf_rele(db, FTAG);
1307 dmu_buf_rele(db_spill, FTAG);
1311 dmu_buf_will_dirty(db_spill, tx);
1313 if (db_spill->db_size < drrs->drr_length)
1314 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1315 drrs->drr_length, tx));
1316 bcopy(data, db_spill->db_data, drrs->drr_length);
1318 dmu_buf_rele(db, FTAG);
1319 dmu_buf_rele(db_spill, FTAG);
1327 restore_free(struct restorearg *ra, objset_t *os,
1328 struct drr_free *drrf)
1332 if (drrf->drr_length != -1ULL &&
1333 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1336 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1339 err = dmu_free_long_range(os, drrf->drr_object,
1340 drrf->drr_offset, drrf->drr_length);
1345 * NB: callers *must* call dmu_recv_end() if this succeeds.
1348 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1349 int cleanup_fd, uint64_t *action_handlep)
1351 struct restorearg ra = { 0 };
1352 dmu_replay_record_t *drr;
1357 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1361 /* compute checksum of drr_begin record */
1362 dmu_replay_record_t *drr;
1363 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1365 drr->drr_type = DRR_BEGIN;
1366 drr->drr_u.drr_begin = *drc->drc_drrb;
1368 fletcher_4_incremental_byteswap(drr,
1369 sizeof (dmu_replay_record_t), &ra.cksum);
1371 fletcher_4_incremental_native(drr,
1372 sizeof (dmu_replay_record_t), &ra.cksum);
1374 kmem_free(drr, sizeof (dmu_replay_record_t));
1378 struct drr_begin *drrb = drc->drc_drrb;
1379 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1380 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1381 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1382 drrb->drr_type = BSWAP_32(drrb->drr_type);
1383 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1384 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1390 ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1392 /* these were verified in dmu_recv_begin */
1393 ASSERT(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo) ==
1395 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1398 * Open the objset we are modifying.
1400 VERIFY(dmu_objset_from_ds(drc->drc_real_ds, &os) == 0);
1402 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1404 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1406 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1407 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1410 if (cleanup_fd == -1) {
1414 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1420 if (*action_handlep == 0) {
1422 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1423 avl_create(ra.guid_to_ds_map, guid_compare,
1424 sizeof (guid_map_entry_t),
1425 offsetof(guid_map_entry_t, avlnode));
1426 ra.err = zfs_onexit_add_cb(minor,
1427 free_guid_map_onexit, ra.guid_to_ds_map,
1432 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1433 (void **)&ra.guid_to_ds_map);
1438 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1442 * Read records and process them.
1445 while (ra.err == 0 &&
1446 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1447 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1453 backup_byteswap(drr);
1455 switch (drr->drr_type) {
1459 * We need to make a copy of the record header,
1460 * because restore_{object,write} may need to
1461 * restore_read(), which will invalidate drr.
1463 struct drr_object drro = drr->drr_u.drr_object;
1464 ra.err = restore_object(&ra, os, &drro);
1467 case DRR_FREEOBJECTS:
1469 struct drr_freeobjects drrfo =
1470 drr->drr_u.drr_freeobjects;
1471 ra.err = restore_freeobjects(&ra, os, &drrfo);
1476 struct drr_write drrw = drr->drr_u.drr_write;
1477 ra.err = restore_write(&ra, os, &drrw);
1480 case DRR_WRITE_BYREF:
1482 struct drr_write_byref drrwbr =
1483 drr->drr_u.drr_write_byref;
1484 ra.err = restore_write_byref(&ra, os, &drrwbr);
1489 struct drr_free drrf = drr->drr_u.drr_free;
1490 ra.err = restore_free(&ra, os, &drrf);
1495 struct drr_end drre = drr->drr_u.drr_end;
1497 * We compare against the *previous* checksum
1498 * value, because the stored checksum is of
1499 * everything before the DRR_END record.
1501 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1507 struct drr_spill drrs = drr->drr_u.drr_spill;
1508 ra.err = restore_spill(&ra, os, &drrs);
1517 ASSERT(ra.err != 0);
1520 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1521 zfs_onexit_fd_rele(cleanup_fd);
1525 * destroy what we created, so we don't leave it in the
1526 * inconsistent restoring state.
1528 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1530 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1532 if (drc->drc_real_ds != drc->drc_logical_ds) {
1533 mutex_exit(&drc->drc_logical_ds->ds_recvlock);
1534 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
1538 vmem_free(ra.buf, ra.bufsize);
1543 struct recvendsyncarg {
1545 uint64_t creation_time;
1550 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1552 dsl_dataset_t *ds = arg1;
1553 struct recvendsyncarg *resa = arg2;
1555 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1559 recv_end_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1561 dsl_dataset_t *ds = arg1;
1562 struct recvendsyncarg *resa = arg2;
1564 dsl_dataset_snapshot_sync(ds, resa->tosnap, tx);
1566 /* set snapshot's creation time and guid */
1567 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1568 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1569 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1570 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1572 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1573 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1577 add_ds_to_guidmap(avl_tree_t *guid_map, dsl_dataset_t *ds)
1579 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1580 uint64_t snapobj = ds->ds_phys->ds_prev_snap_obj;
1581 dsl_dataset_t *snapds;
1582 guid_map_entry_t *gmep;
1585 ASSERT(guid_map != NULL);
1587 rw_enter(&dp->dp_config_rwlock, RW_READER);
1588 err = dsl_dataset_hold_obj(dp, snapobj, guid_map, &snapds);
1590 gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP);
1591 gmep->guid = snapds->ds_phys->ds_guid;
1592 gmep->gme_ds = snapds;
1593 avl_add(guid_map, gmep);
1596 rw_exit(&dp->dp_config_rwlock);
1601 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1603 struct recvendsyncarg resa;
1604 dsl_dataset_t *ds = drc->drc_logical_ds;
1608 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1609 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1612 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1614 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1615 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1620 mutex_exit(&ds->ds_recvlock);
1621 dsl_dataset_rele(ds, dmu_recv_tag);
1622 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1627 resa.creation_time = drc->drc_drrb->drr_creation_time;
1628 resa.toguid = drc->drc_drrb->drr_toguid;
1629 resa.tosnap = drc->drc_tosnap;
1631 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1632 recv_end_check, recv_end_sync, ds, &resa, 3);
1635 (void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE);
1639 mutex_exit(&ds->ds_recvlock);
1640 if (err == 0 && drc->drc_guid_to_ds_map != NULL)
1641 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1642 dsl_dataset_disown(ds, dmu_recv_tag);
1643 myerr = dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE);
1644 ASSERT3U(myerr, ==, 0);
1649 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1651 struct recvendsyncarg resa;
1652 dsl_dataset_t *ds = drc->drc_logical_ds;
1656 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1657 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1660 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1662 resa.creation_time = drc->drc_drrb->drr_creation_time;
1663 resa.toguid = drc->drc_drrb->drr_toguid;
1664 resa.tosnap = drc->drc_tosnap;
1666 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1667 recv_end_check, recv_end_sync, ds, &resa, 3);
1669 /* clean up the fs we just recv'd into */
1670 (void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE);
1672 if (drc->drc_guid_to_ds_map != NULL)
1673 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1674 /* release the hold from dmu_recv_begin */
1675 dsl_dataset_disown(ds, dmu_recv_tag);
1681 dmu_recv_end(dmu_recv_cookie_t *drc)
1683 if (drc->drc_logical_ds != drc->drc_real_ds)
1684 return (dmu_recv_existing_end(drc));
1686 return (dmu_recv_new_end(drc));