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.
29 #include <sys/dmu_impl.h>
30 #include <sys/dmu_tx.h>
32 #include <sys/dnode.h>
33 #include <sys/zfs_context.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dsl_dataset.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dsl_prop.h>
39 #include <sys/dsl_pool.h>
40 #include <sys/dsl_synctask.h>
41 #include <sys/zfs_ioctl.h>
43 #include <sys/zio_checksum.h>
44 #include <sys/zfs_znode.h>
45 #include <zfs_fletcher.h>
48 #include <sys/zfs_onexit.h>
50 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
51 int zfs_send_corrupt_data = B_FALSE;
53 static char *dmu_recv_tag = "dmu_recv_tag";
56 * The list of data whose inclusion in a send stream can be pending from
57 * one call to backup_cb to another. Multiple calls to dump_free() and
58 * dump_freeobjects() can be aggregated into a single DRR_FREE or
59 * DRR_FREEOBJECTS replay record.
68 dmu_replay_record_t *drr;
79 dump_bytes(struct backuparg *ba, void *buf, int len)
81 ssize_t resid; /* have to get resid to get detailed errno */
82 ASSERT3U(len % 8, ==, 0);
84 fletcher_4_incremental_native(buf, len, &ba->zc);
85 ba->err = vn_rdwr(UIO_WRITE, ba->vp,
87 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
93 dump_free(struct backuparg *ba, uint64_t object, uint64_t offset,
96 struct drr_free *drrf = &(ba->drr->drr_u.drr_free);
99 * If there is a pending op, but it's not PENDING_FREE, push it out,
100 * since free block aggregation can only be done for blocks of the
101 * same type (i.e., DRR_FREE records can only be aggregated with
102 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
103 * aggregated with other DRR_FREEOBJECTS records.
105 if (ba->pending_op != PENDING_NONE && ba->pending_op != PENDING_FREE) {
106 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)) != 0)
108 ba->pending_op = PENDING_NONE;
111 if (ba->pending_op == PENDING_FREE) {
113 * There should never be a PENDING_FREE if length is -1
114 * (because dump_dnode is the only place where this
115 * function is called with a -1, and only after flushing
116 * any pending record).
118 ASSERT(length != -1ULL);
120 * Check to see whether this free block can be aggregated
123 if (drrf->drr_object == object && drrf->drr_offset +
124 drrf->drr_length == offset) {
125 drrf->drr_length += length;
128 /* not a continuation. Push out pending record */
129 if (dump_bytes(ba, ba->drr,
130 sizeof (dmu_replay_record_t)) != 0)
132 ba->pending_op = PENDING_NONE;
135 /* create a FREE record and make it pending */
136 bzero(ba->drr, sizeof (dmu_replay_record_t));
137 ba->drr->drr_type = DRR_FREE;
138 drrf->drr_object = object;
139 drrf->drr_offset = offset;
140 drrf->drr_length = length;
141 drrf->drr_toguid = ba->toguid;
142 if (length == -1ULL) {
143 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)) != 0)
146 ba->pending_op = PENDING_FREE;
153 dump_data(struct backuparg *ba, dmu_object_type_t type,
154 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
156 struct drr_write *drrw = &(ba->drr->drr_u.drr_write);
160 * If there is any kind of pending aggregation (currently either
161 * a grouping of free objects or free blocks), push it out to
162 * the stream, since aggregation can't be done across operations
163 * of different types.
165 if (ba->pending_op != PENDING_NONE) {
166 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)) != 0)
168 ba->pending_op = PENDING_NONE;
170 /* write a DATA record */
171 bzero(ba->drr, sizeof (dmu_replay_record_t));
172 ba->drr->drr_type = DRR_WRITE;
173 drrw->drr_object = object;
174 drrw->drr_type = type;
175 drrw->drr_offset = offset;
176 drrw->drr_length = blksz;
177 drrw->drr_toguid = ba->toguid;
178 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
179 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
180 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
181 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
182 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
183 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
184 drrw->drr_key.ddk_cksum = bp->blk_cksum;
186 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)) != 0)
188 if (dump_bytes(ba, data, blksz) != 0)
194 dump_spill(struct backuparg *ba, uint64_t object, int blksz, void *data)
196 struct drr_spill *drrs = &(ba->drr->drr_u.drr_spill);
198 if (ba->pending_op != PENDING_NONE) {
199 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)) != 0)
201 ba->pending_op = PENDING_NONE;
204 /* write a SPILL record */
205 bzero(ba->drr, sizeof (dmu_replay_record_t));
206 ba->drr->drr_type = DRR_SPILL;
207 drrs->drr_object = object;
208 drrs->drr_length = blksz;
209 drrs->drr_toguid = ba->toguid;
211 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
213 if (dump_bytes(ba, data, blksz))
219 dump_freeobjects(struct backuparg *ba, uint64_t firstobj, uint64_t numobjs)
221 struct drr_freeobjects *drrfo = &(ba->drr->drr_u.drr_freeobjects);
224 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
225 * push it out, since free block aggregation can only be done for
226 * blocks of the same type (i.e., DRR_FREE records can only be
227 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
228 * can only be aggregated with other DRR_FREEOBJECTS records.
230 if (ba->pending_op != PENDING_NONE &&
231 ba->pending_op != PENDING_FREEOBJECTS) {
232 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)) != 0)
234 ba->pending_op = PENDING_NONE;
236 if (ba->pending_op == PENDING_FREEOBJECTS) {
238 * See whether this free object array can be aggregated
241 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
242 drrfo->drr_numobjs += numobjs;
245 /* can't be aggregated. Push out pending record */
246 if (dump_bytes(ba, ba->drr,
247 sizeof (dmu_replay_record_t)) != 0)
249 ba->pending_op = PENDING_NONE;
253 /* write a FREEOBJECTS record */
254 bzero(ba->drr, sizeof (dmu_replay_record_t));
255 ba->drr->drr_type = DRR_FREEOBJECTS;
256 drrfo->drr_firstobj = firstobj;
257 drrfo->drr_numobjs = numobjs;
258 drrfo->drr_toguid = ba->toguid;
260 ba->pending_op = PENDING_FREEOBJECTS;
266 dump_dnode(struct backuparg *ba, uint64_t object, dnode_phys_t *dnp)
268 struct drr_object *drro = &(ba->drr->drr_u.drr_object);
270 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
271 return (dump_freeobjects(ba, object, 1));
273 if (ba->pending_op != PENDING_NONE) {
274 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)) != 0)
276 ba->pending_op = PENDING_NONE;
279 /* write an OBJECT record */
280 bzero(ba->drr, sizeof (dmu_replay_record_t));
281 ba->drr->drr_type = DRR_OBJECT;
282 drro->drr_object = object;
283 drro->drr_type = dnp->dn_type;
284 drro->drr_bonustype = dnp->dn_bonustype;
285 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
286 drro->drr_bonuslen = dnp->dn_bonuslen;
287 drro->drr_checksumtype = dnp->dn_checksum;
288 drro->drr_compress = dnp->dn_compress;
289 drro->drr_toguid = ba->toguid;
291 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)) != 0)
294 if (dump_bytes(ba, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
297 /* free anything past the end of the file */
298 if (dump_free(ba, object, (dnp->dn_maxblkid + 1) *
299 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
306 #define BP_SPAN(dnp, level) \
307 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
308 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
312 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
313 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
315 struct backuparg *ba = arg;
316 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
319 if (issig(JUSTLOOKING) && issig(FORREAL))
322 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
323 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
325 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
326 uint64_t span = BP_SPAN(dnp, zb->zb_level);
327 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
328 err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
329 } else if (bp == NULL) {
330 uint64_t span = BP_SPAN(dnp, zb->zb_level);
331 err = dump_free(ba, zb->zb_object, zb->zb_blkid * span, span);
332 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
334 } else if (type == DMU_OT_DNODE) {
337 int blksz = BP_GET_LSIZE(bp);
338 uint32_t aflags = ARC_WAIT;
341 if (dsl_read(NULL, spa, bp, pbuf,
342 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
343 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
347 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
348 uint64_t dnobj = (zb->zb_blkid <<
349 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
350 err = dump_dnode(ba, dnobj, blk+i);
354 (void) arc_buf_remove_ref(abuf, &abuf);
355 } else if (type == DMU_OT_SA) {
356 uint32_t aflags = ARC_WAIT;
358 int blksz = BP_GET_LSIZE(bp);
360 if (arc_read_nolock(NULL, spa, bp,
361 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
362 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
365 err = dump_spill(ba, zb->zb_object, blksz, abuf->b_data);
366 (void) arc_buf_remove_ref(abuf, &abuf);
367 } else { /* it's a level-0 block of a regular object */
368 uint32_t aflags = ARC_WAIT;
370 int blksz = BP_GET_LSIZE(bp);
372 if (dsl_read(NULL, spa, bp, pbuf,
373 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
374 ZIO_FLAG_CANFAIL, &aflags, zb) != 0) {
375 if (zfs_send_corrupt_data) {
377 /* Send a block filled with 0x"zfs badd bloc" */
378 abuf = arc_buf_alloc(spa, blksz, &abuf,
380 for (ptr = abuf->b_data;
381 (char *)ptr < (char *)abuf->b_data + blksz;
383 *ptr = 0x2f5baddb10c;
389 err = dump_data(ba, type, zb->zb_object, zb->zb_blkid * blksz,
390 blksz, bp, abuf->b_data);
391 (void) arc_buf_remove_ref(abuf, &abuf);
394 ASSERT(err == 0 || err == EINTR);
399 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
400 vnode_t *vp, offset_t *off)
402 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
403 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
404 dmu_replay_record_t *drr;
407 uint64_t fromtxg = 0;
409 /* tosnap must be a snapshot */
410 if (ds->ds_phys->ds_next_snap_obj == 0)
413 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
414 if (fromds && (ds->ds_dir != fromds->ds_dir ||
415 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
419 dsl_pool_t *dp = ds->ds_dir->dd_pool;
424 if (dsl_dir_is_clone(ds->ds_dir)) {
425 rw_enter(&dp->dp_config_rwlock, RW_READER);
426 err = dsl_dataset_hold_obj(dp,
427 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
428 rw_exit(&dp->dp_config_rwlock);
432 fromorigin = B_FALSE;
437 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
438 drr->drr_type = DRR_BEGIN;
439 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
440 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
444 if (dmu_objset_type(tosnap) == DMU_OST_ZFS) {
446 if (zfs_get_zplprop(tosnap, ZFS_PROP_VERSION, &version) != 0)
448 if (version == ZPL_VERSION_SA) {
449 DMU_SET_FEATUREFLAGS(
450 drr->drr_u.drr_begin.drr_versioninfo,
451 DMU_BACKUP_FEATURE_SA_SPILL);
456 drr->drr_u.drr_begin.drr_creation_time =
457 ds->ds_phys->ds_creation_time;
458 drr->drr_u.drr_begin.drr_type = tosnap->os_phys->os_type;
460 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
461 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
462 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
463 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
466 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
467 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
470 fromtxg = fromds->ds_phys->ds_creation_txg;
472 dsl_dataset_rele(fromds, FTAG);
478 ba.toguid = ds->ds_phys->ds_guid;
479 ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
480 ba.pending_op = PENDING_NONE;
482 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t)) != 0) {
483 kmem_free(drr, sizeof (dmu_replay_record_t));
487 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
490 if (ba.pending_op != PENDING_NONE)
491 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t)) != 0)
495 if (err == EINTR && ba.err)
497 kmem_free(drr, sizeof (dmu_replay_record_t));
501 bzero(drr, sizeof (dmu_replay_record_t));
502 drr->drr_type = DRR_END;
503 drr->drr_u.drr_end.drr_checksum = ba.zc;
504 drr->drr_u.drr_end.drr_toguid = ba.toguid;
506 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t)) != 0) {
507 kmem_free(drr, sizeof (dmu_replay_record_t));
511 kmem_free(drr, sizeof (dmu_replay_record_t));
517 dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
520 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
521 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
522 dsl_pool_t *dp = ds->ds_dir->dd_pool;
524 uint64_t size, recordsize;
526 /* tosnap must be a snapshot */
527 if (ds->ds_phys->ds_next_snap_obj == 0)
530 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
531 if (fromds && (ds->ds_dir != fromds->ds_dir ||
532 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
539 if (dsl_dir_is_clone(ds->ds_dir)) {
540 rw_enter(&dp->dp_config_rwlock, RW_READER);
541 err = dsl_dataset_hold_obj(dp,
542 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
543 rw_exit(&dp->dp_config_rwlock);
547 fromorigin = B_FALSE;
551 /* Get uncompressed size estimate of changed data. */
552 if (fromds == NULL) {
553 size = ds->ds_phys->ds_uncompressed_bytes;
556 err = dsl_dataset_space_written(fromds, ds,
557 &used, &comp, &size);
559 dsl_dataset_rele(fromds, FTAG);
565 * Assume that space (both on-disk and in-stream) is dominated by
566 * data. We will adjust for indirect blocks and the copies property,
567 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
571 * Subtract out approximate space used by indirect blocks.
572 * Assume most space is used by data blocks (non-indirect, non-dnode).
573 * Assume all blocks are recordsize. Assume ditto blocks and
574 * internal fragmentation counter out compression.
576 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
577 * block, which we observe in practice.
579 rw_enter(&dp->dp_config_rwlock, RW_READER);
580 err = dsl_prop_get_ds(ds, "recordsize",
581 sizeof (recordsize), 1, &recordsize, NULL);
582 rw_exit(&dp->dp_config_rwlock);
585 size -= size / recordsize * sizeof (blkptr_t);
587 /* Add in the space for the record associated with each block. */
588 size += size / recordsize * sizeof (dmu_replay_record_t);
595 struct recvbeginsyncarg {
598 dsl_dataset_t *origin;
600 dmu_objset_type_t type;
604 char clonelastname[MAXNAMELEN];
605 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
611 recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx)
613 dsl_dir_t *dd = arg1;
614 struct recvbeginsyncarg *rbsa = arg2;
615 objset_t *mos = dd->dd_pool->dp_meta_objset;
619 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
620 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
623 return (err ? err : EEXIST);
626 /* make sure it's a snap in the same pool */
627 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
629 if (!dsl_dataset_is_snapshot(rbsa->origin))
631 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
639 recv_new_sync(void *arg1, void *arg2, dmu_tx_t *tx)
641 dsl_dir_t *dd = arg1;
642 struct recvbeginsyncarg *rbsa = arg2;
643 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
646 /* Create and open new dataset. */
647 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
648 rbsa->origin, flags, rbsa->cr, tx);
649 VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj,
650 B_TRUE, dmu_recv_tag, &rbsa->ds));
652 if (rbsa->origin == NULL) {
653 (void) dmu_objset_create_impl(dd->dd_pool->dp_spa,
654 rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx);
657 spa_history_log_internal(LOG_DS_REPLAY_FULL_SYNC,
658 dd->dd_pool->dp_spa, tx, "dataset = %lld", dsobj);
663 recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
665 dsl_dataset_t *ds = arg1;
666 struct recvbeginsyncarg *rbsa = arg2;
670 /* must not have any changes since most recent snapshot */
671 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
674 /* new snapshot name must not exist */
675 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
676 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
682 if (rbsa->fromguid) {
683 /* if incremental, most recent snapshot must match fromguid */
684 if (ds->ds_prev == NULL)
688 * most recent snapshot must match fromguid, or there are no
689 * changes since the fromguid one
691 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) {
692 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth;
693 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj;
696 err = dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
700 if (snap->ds_phys->ds_creation_txg < birth) {
701 dsl_dataset_rele(snap, FTAG);
704 if (snap->ds_phys->ds_guid == rbsa->fromguid) {
705 dsl_dataset_rele(snap, FTAG);
708 obj = snap->ds_phys->ds_prev_snap_obj;
709 dsl_dataset_rele(snap, FTAG);
715 /* if full, most recent snapshot must be $ORIGIN */
716 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
720 /* temporary clone name must not exist */
721 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
722 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
723 rbsa->clonelastname, 8, 1, &val);
734 recv_existing_sync(void *arg1, void *arg2, dmu_tx_t *tx)
736 dsl_dataset_t *ohds = arg1;
737 struct recvbeginsyncarg *rbsa = arg2;
738 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
740 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
743 /* create and open the temporary clone */
744 dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname,
745 ohds->ds_prev, flags, rbsa->cr, tx);
746 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, B_TRUE, dmu_recv_tag, &cds));
749 * If we actually created a non-clone, we need to create the
750 * objset in our new dataset.
752 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) {
753 (void) dmu_objset_create_impl(dp->dp_spa,
754 cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx);
759 spa_history_log_internal(LOG_DS_REPLAY_INC_SYNC,
760 dp->dp_spa, tx, "dataset = %lld", dsobj);
764 dmu_recv_verify_features(dsl_dataset_t *ds, struct drr_begin *drrb)
768 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
770 /* Verify pool version supports SA if SA_SPILL feature set */
771 return ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
772 (spa_version(dsl_dataset_get_spa(ds)) < SPA_VERSION_SA));
776 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
777 * succeeds; otherwise we will leak the holds on the datasets.
780 dmu_recv_begin(char *tofs, char *tosnap, char *top_ds, struct drr_begin *drrb,
781 boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
785 struct recvbeginsyncarg rbsa = { 0 };
786 uint64_t versioninfo;
790 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
792 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
798 rbsa.tosnap = tosnap;
799 rbsa.origin = origin ? origin->os_dsl_dataset : NULL;
800 rbsa.fromguid = drrb->drr_fromguid;
801 rbsa.type = drrb->drr_type;
805 versioninfo = drrb->drr_versioninfo;
806 flags = drrb->drr_flags;
809 rbsa.type = BSWAP_32(rbsa.type);
810 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
811 versioninfo = BSWAP_64(versioninfo);
812 flags = BSWAP_32(flags);
815 if (DMU_GET_STREAM_HDRTYPE(versioninfo) == DMU_COMPOUNDSTREAM ||
816 rbsa.type >= DMU_OST_NUMTYPES ||
817 ((flags & DRR_FLAG_CLONE) && origin == NULL))
820 if (flags & DRR_FLAG_CI_DATA)
821 rbsa.dsflags = DS_FLAG_CI_DATASET;
823 bzero(drc, sizeof (dmu_recv_cookie_t));
824 drc->drc_drrb = drrb;
825 drc->drc_tosnap = tosnap;
826 drc->drc_top_ds = top_ds;
827 drc->drc_force = force;
830 * Process the begin in syncing context.
833 /* open the dataset we are logically receiving into */
834 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
836 if (dmu_recv_verify_features(ds, drrb)) {
837 dsl_dataset_rele(ds, dmu_recv_tag);
840 /* target fs already exists; recv into temp clone */
842 /* Can't recv a clone into an existing fs */
843 if (flags & DRR_FLAG_CLONE) {
844 dsl_dataset_rele(ds, dmu_recv_tag);
848 /* must not have an incremental recv already in progress */
849 if (!mutex_tryenter(&ds->ds_recvlock)) {
850 dsl_dataset_rele(ds, dmu_recv_tag);
854 /* tmp clone name is: tofs/%tosnap" */
855 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
858 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
859 recv_existing_check, recv_existing_sync, ds, &rbsa, 5);
861 mutex_exit(&ds->ds_recvlock);
862 dsl_dataset_rele(ds, dmu_recv_tag);
865 drc->drc_logical_ds = ds;
866 drc->drc_real_ds = rbsa.ds;
867 } else if (err == ENOENT) {
868 /* target fs does not exist; must be a full backup or clone */
872 * If it's a non-clone incremental, we are missing the
873 * target fs, so fail the recv.
875 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE))
878 /* Open the parent of tofs */
879 cp = strrchr(tofs, '/');
881 err = dsl_dataset_hold(tofs, FTAG, &ds);
886 if (dmu_recv_verify_features(ds, drrb)) {
887 dsl_dataset_rele(ds, FTAG);
891 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
892 recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5);
893 dsl_dataset_rele(ds, FTAG);
896 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
897 drc->drc_newfs = B_TRUE;
909 int bufsize; /* amount of memory allocated for buf */
911 avl_tree_t *guid_to_ds_map;
914 typedef struct guid_map_entry {
916 dsl_dataset_t *gme_ds;
921 guid_compare(const void *arg1, const void *arg2)
923 const guid_map_entry_t *gmep1 = arg1;
924 const guid_map_entry_t *gmep2 = arg2;
926 if (gmep1->guid < gmep2->guid)
928 else if (gmep1->guid > gmep2->guid)
934 free_guid_map_onexit(void *arg)
936 avl_tree_t *ca = arg;
938 guid_map_entry_t *gmep;
940 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
941 dsl_dataset_rele(gmep->gme_ds, ca);
942 kmem_free(gmep, sizeof (guid_map_entry_t));
945 kmem_free(ca, sizeof (avl_tree_t));
949 restore_read(struct restorearg *ra, int len)
954 /* some things will require 8-byte alignment, so everything must */
955 ASSERT3U(len % 8, ==, 0);
960 ra->err = vn_rdwr(UIO_READ, ra->vp,
961 (caddr_t)ra->buf + done, len - done,
962 ra->voff, UIO_SYSSPACE, FAPPEND,
963 RLIM64_INFINITY, CRED(), &resid);
965 if (resid == len - done)
967 ra->voff += len - done - resid;
973 ASSERT3U(done, ==, len);
976 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
978 fletcher_4_incremental_native(rv, len, &ra->cksum);
983 backup_byteswap(dmu_replay_record_t *drr)
985 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
986 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
987 drr->drr_type = BSWAP_32(drr->drr_type);
988 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
989 switch (drr->drr_type) {
991 DO64(drr_begin.drr_magic);
992 DO64(drr_begin.drr_versioninfo);
993 DO64(drr_begin.drr_creation_time);
994 DO32(drr_begin.drr_type);
995 DO32(drr_begin.drr_flags);
996 DO64(drr_begin.drr_toguid);
997 DO64(drr_begin.drr_fromguid);
1000 DO64(drr_object.drr_object);
1001 /* DO64(drr_object.drr_allocation_txg); */
1002 DO32(drr_object.drr_type);
1003 DO32(drr_object.drr_bonustype);
1004 DO32(drr_object.drr_blksz);
1005 DO32(drr_object.drr_bonuslen);
1006 DO64(drr_object.drr_toguid);
1008 case DRR_FREEOBJECTS:
1009 DO64(drr_freeobjects.drr_firstobj);
1010 DO64(drr_freeobjects.drr_numobjs);
1011 DO64(drr_freeobjects.drr_toguid);
1014 DO64(drr_write.drr_object);
1015 DO32(drr_write.drr_type);
1016 DO64(drr_write.drr_offset);
1017 DO64(drr_write.drr_length);
1018 DO64(drr_write.drr_toguid);
1019 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1020 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1021 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1022 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1023 DO64(drr_write.drr_key.ddk_prop);
1025 case DRR_WRITE_BYREF:
1026 DO64(drr_write_byref.drr_object);
1027 DO64(drr_write_byref.drr_offset);
1028 DO64(drr_write_byref.drr_length);
1029 DO64(drr_write_byref.drr_toguid);
1030 DO64(drr_write_byref.drr_refguid);
1031 DO64(drr_write_byref.drr_refobject);
1032 DO64(drr_write_byref.drr_refoffset);
1033 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1034 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1035 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1036 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1037 DO64(drr_write_byref.drr_key.ddk_prop);
1040 DO64(drr_free.drr_object);
1041 DO64(drr_free.drr_offset);
1042 DO64(drr_free.drr_length);
1043 DO64(drr_free.drr_toguid);
1046 DO64(drr_spill.drr_object);
1047 DO64(drr_spill.drr_length);
1048 DO64(drr_spill.drr_toguid);
1051 DO64(drr_end.drr_checksum.zc_word[0]);
1052 DO64(drr_end.drr_checksum.zc_word[1]);
1053 DO64(drr_end.drr_checksum.zc_word[2]);
1054 DO64(drr_end.drr_checksum.zc_word[3]);
1055 DO64(drr_end.drr_toguid);
1065 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1071 if (drro->drr_type == DMU_OT_NONE ||
1072 drro->drr_type >= DMU_OT_NUMTYPES ||
1073 drro->drr_bonustype >= DMU_OT_NUMTYPES ||
1074 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1075 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1076 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1077 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1078 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1079 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1083 err = dmu_object_info(os, drro->drr_object, NULL);
1085 if (err != 0 && err != ENOENT)
1088 if (drro->drr_bonuslen) {
1089 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1094 if (err == ENOENT) {
1095 /* currently free, want to be allocated */
1096 tx = dmu_tx_create(os);
1097 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1098 err = dmu_tx_assign(tx, TXG_WAIT);
1103 err = dmu_object_claim(os, drro->drr_object,
1104 drro->drr_type, drro->drr_blksz,
1105 drro->drr_bonustype, drro->drr_bonuslen, tx);
1108 /* currently allocated, want to be allocated */
1109 err = dmu_object_reclaim(os, drro->drr_object,
1110 drro->drr_type, drro->drr_blksz,
1111 drro->drr_bonustype, drro->drr_bonuslen);
1117 tx = dmu_tx_create(os);
1118 dmu_tx_hold_bonus(tx, drro->drr_object);
1119 err = dmu_tx_assign(tx, TXG_WAIT);
1125 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1127 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1132 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1133 dmu_buf_will_dirty(db, tx);
1135 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1136 bcopy(data, db->db_data, drro->drr_bonuslen);
1138 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
1139 drro->drr_bonuslen);
1141 dmu_buf_rele(db, FTAG);
1149 restore_freeobjects(struct restorearg *ra, objset_t *os,
1150 struct drr_freeobjects *drrfo)
1154 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1157 for (obj = drrfo->drr_firstobj;
1158 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1159 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1162 if (dmu_object_info(os, obj, NULL) != 0)
1165 err = dmu_free_object(os, obj);
1173 restore_write(struct restorearg *ra, objset_t *os,
1174 struct drr_write *drrw)
1180 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1181 drrw->drr_type >= DMU_OT_NUMTYPES)
1184 data = restore_read(ra, drrw->drr_length);
1188 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1191 tx = dmu_tx_create(os);
1193 dmu_tx_hold_write(tx, drrw->drr_object,
1194 drrw->drr_offset, drrw->drr_length);
1195 err = dmu_tx_assign(tx, TXG_WAIT);
1201 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
1202 dmu_write(os, drrw->drr_object,
1203 drrw->drr_offset, drrw->drr_length, data, tx);
1209 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1210 * streams to refer to a copy of the data that is already on the
1211 * system because it came in earlier in the stream. This function
1212 * finds the earlier copy of the data, and uses that copy instead of
1213 * data from the stream to fulfill this write.
1216 restore_write_byref(struct restorearg *ra, objset_t *os,
1217 struct drr_write_byref *drrwbr)
1221 guid_map_entry_t gmesrch;
1222 guid_map_entry_t *gmep;
1224 objset_t *ref_os = NULL;
1227 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1231 * If the GUID of the referenced dataset is different from the
1232 * GUID of the target dataset, find the referenced dataset.
1234 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1235 gmesrch.guid = drrwbr->drr_refguid;
1236 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1240 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1246 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1247 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1251 tx = dmu_tx_create(os);
1253 dmu_tx_hold_write(tx, drrwbr->drr_object,
1254 drrwbr->drr_offset, drrwbr->drr_length);
1255 err = dmu_tx_assign(tx, TXG_WAIT);
1260 dmu_write(os, drrwbr->drr_object,
1261 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1262 dmu_buf_rele(dbp, FTAG);
1268 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1272 dmu_buf_t *db, *db_spill;
1275 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1276 drrs->drr_length > SPA_MAXBLOCKSIZE)
1279 data = restore_read(ra, drrs->drr_length);
1283 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1286 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1287 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1288 dmu_buf_rele(db, FTAG);
1292 tx = dmu_tx_create(os);
1294 dmu_tx_hold_spill(tx, db->db_object);
1296 err = dmu_tx_assign(tx, TXG_WAIT);
1298 dmu_buf_rele(db, FTAG);
1299 dmu_buf_rele(db_spill, FTAG);
1303 dmu_buf_will_dirty(db_spill, tx);
1305 if (db_spill->db_size < drrs->drr_length)
1306 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1307 drrs->drr_length, tx));
1308 bcopy(data, db_spill->db_data, drrs->drr_length);
1310 dmu_buf_rele(db, FTAG);
1311 dmu_buf_rele(db_spill, FTAG);
1319 restore_free(struct restorearg *ra, objset_t *os,
1320 struct drr_free *drrf)
1324 if (drrf->drr_length != -1ULL &&
1325 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1328 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1331 err = dmu_free_long_range(os, drrf->drr_object,
1332 drrf->drr_offset, drrf->drr_length);
1337 * NB: callers *must* call dmu_recv_end() if this succeeds.
1340 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1341 int cleanup_fd, uint64_t *action_handlep)
1343 struct restorearg ra = { 0 };
1344 dmu_replay_record_t *drr;
1349 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1353 /* compute checksum of drr_begin record */
1354 dmu_replay_record_t *drr;
1355 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1357 drr->drr_type = DRR_BEGIN;
1358 drr->drr_u.drr_begin = *drc->drc_drrb;
1360 fletcher_4_incremental_byteswap(drr,
1361 sizeof (dmu_replay_record_t), &ra.cksum);
1363 fletcher_4_incremental_native(drr,
1364 sizeof (dmu_replay_record_t), &ra.cksum);
1366 kmem_free(drr, sizeof (dmu_replay_record_t));
1370 struct drr_begin *drrb = drc->drc_drrb;
1371 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1372 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1373 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1374 drrb->drr_type = BSWAP_32(drrb->drr_type);
1375 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1376 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1382 ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1384 /* these were verified in dmu_recv_begin */
1385 ASSERT(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo) ==
1387 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1390 * Open the objset we are modifying.
1392 VERIFY(dmu_objset_from_ds(drc->drc_real_ds, &os) == 0);
1394 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1396 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1398 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1399 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1402 if (cleanup_fd == -1) {
1406 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1412 if (*action_handlep == 0) {
1414 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1415 avl_create(ra.guid_to_ds_map, guid_compare,
1416 sizeof (guid_map_entry_t),
1417 offsetof(guid_map_entry_t, avlnode));
1418 ra.err = zfs_onexit_add_cb(minor,
1419 free_guid_map_onexit, ra.guid_to_ds_map,
1424 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1425 (void **)&ra.guid_to_ds_map);
1430 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1434 * Read records and process them.
1437 while (ra.err == 0 &&
1438 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1439 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1445 backup_byteswap(drr);
1447 switch (drr->drr_type) {
1451 * We need to make a copy of the record header,
1452 * because restore_{object,write} may need to
1453 * restore_read(), which will invalidate drr.
1455 struct drr_object drro = drr->drr_u.drr_object;
1456 ra.err = restore_object(&ra, os, &drro);
1459 case DRR_FREEOBJECTS:
1461 struct drr_freeobjects drrfo =
1462 drr->drr_u.drr_freeobjects;
1463 ra.err = restore_freeobjects(&ra, os, &drrfo);
1468 struct drr_write drrw = drr->drr_u.drr_write;
1469 ra.err = restore_write(&ra, os, &drrw);
1472 case DRR_WRITE_BYREF:
1474 struct drr_write_byref drrwbr =
1475 drr->drr_u.drr_write_byref;
1476 ra.err = restore_write_byref(&ra, os, &drrwbr);
1481 struct drr_free drrf = drr->drr_u.drr_free;
1482 ra.err = restore_free(&ra, os, &drrf);
1487 struct drr_end drre = drr->drr_u.drr_end;
1489 * We compare against the *previous* checksum
1490 * value, because the stored checksum is of
1491 * everything before the DRR_END record.
1493 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1499 struct drr_spill drrs = drr->drr_u.drr_spill;
1500 ra.err = restore_spill(&ra, os, &drrs);
1509 ASSERT(ra.err != 0);
1512 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1513 zfs_onexit_fd_rele(cleanup_fd);
1517 * destroy what we created, so we don't leave it in the
1518 * inconsistent restoring state.
1520 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1522 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1524 if (drc->drc_real_ds != drc->drc_logical_ds) {
1525 mutex_exit(&drc->drc_logical_ds->ds_recvlock);
1526 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
1530 vmem_free(ra.buf, ra.bufsize);
1535 struct recvendsyncarg {
1537 uint64_t creation_time;
1542 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1544 dsl_dataset_t *ds = arg1;
1545 struct recvendsyncarg *resa = arg2;
1547 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1551 recv_end_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1553 dsl_dataset_t *ds = arg1;
1554 struct recvendsyncarg *resa = arg2;
1556 dsl_dataset_snapshot_sync(ds, resa->tosnap, tx);
1558 /* set snapshot's creation time and guid */
1559 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1560 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1561 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1562 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1564 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1565 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1569 add_ds_to_guidmap(avl_tree_t *guid_map, dsl_dataset_t *ds)
1571 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1572 uint64_t snapobj = ds->ds_phys->ds_prev_snap_obj;
1573 dsl_dataset_t *snapds;
1574 guid_map_entry_t *gmep;
1577 ASSERT(guid_map != NULL);
1579 rw_enter(&dp->dp_config_rwlock, RW_READER);
1580 err = dsl_dataset_hold_obj(dp, snapobj, guid_map, &snapds);
1582 gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP);
1583 gmep->guid = snapds->ds_phys->ds_guid;
1584 gmep->gme_ds = snapds;
1585 avl_add(guid_map, gmep);
1588 rw_exit(&dp->dp_config_rwlock);
1593 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1595 struct recvendsyncarg resa;
1596 dsl_dataset_t *ds = drc->drc_logical_ds;
1600 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1601 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1604 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1606 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1607 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1612 mutex_exit(&ds->ds_recvlock);
1613 dsl_dataset_rele(ds, dmu_recv_tag);
1614 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1619 resa.creation_time = drc->drc_drrb->drr_creation_time;
1620 resa.toguid = drc->drc_drrb->drr_toguid;
1621 resa.tosnap = drc->drc_tosnap;
1623 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1624 recv_end_check, recv_end_sync, ds, &resa, 3);
1627 (void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE);
1631 mutex_exit(&ds->ds_recvlock);
1632 if (err == 0 && drc->drc_guid_to_ds_map != NULL)
1633 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1634 dsl_dataset_disown(ds, dmu_recv_tag);
1635 myerr = dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE);
1636 ASSERT3U(myerr, ==, 0);
1641 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1643 struct recvendsyncarg resa;
1644 dsl_dataset_t *ds = drc->drc_logical_ds;
1648 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1649 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1652 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1654 resa.creation_time = drc->drc_drrb->drr_creation_time;
1655 resa.toguid = drc->drc_drrb->drr_toguid;
1656 resa.tosnap = drc->drc_tosnap;
1658 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1659 recv_end_check, recv_end_sync, ds, &resa, 3);
1661 /* clean up the fs we just recv'd into */
1662 (void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE);
1664 if (drc->drc_guid_to_ds_map != NULL)
1665 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1666 /* release the hold from dmu_recv_begin */
1667 dsl_dataset_disown(ds, dmu_recv_tag);
1673 dmu_recv_end(dmu_recv_cookie_t *drc)
1675 if (drc->drc_logical_ds != drc->drc_real_ds)
1676 return (dmu_recv_existing_end(drc));
1678 return (dmu_recv_new_end(drc));