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 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
27 #include <sys/dmu_impl.h>
28 #include <sys/dmu_tx.h>
30 #include <sys/dnode.h>
31 #include <sys/zfs_context.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/zfs_ioctl.h>
40 #include <sys/zio_checksum.h>
42 static char *dmu_recv_tag = "dmu_recv_tag";
45 dmu_replay_record_t *drr;
54 dump_bytes(struct backuparg *ba, void *buf, int len)
56 ssize_t resid; /* have to get resid to get detailed errno */
57 ASSERT3U(len % 8, ==, 0);
59 fletcher_4_incremental_native(buf, len, &ba->zc);
60 ba->err = vn_rdwr(UIO_WRITE, ba->vp,
62 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
68 dump_free(struct backuparg *ba, uint64_t object, uint64_t offset,
71 /* write a FREE record */
72 bzero(ba->drr, sizeof (dmu_replay_record_t));
73 ba->drr->drr_type = DRR_FREE;
74 ba->drr->drr_u.drr_free.drr_object = object;
75 ba->drr->drr_u.drr_free.drr_offset = offset;
76 ba->drr->drr_u.drr_free.drr_length = length;
78 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
84 dump_data(struct backuparg *ba, dmu_object_type_t type,
85 uint64_t object, uint64_t offset, int blksz, void *data)
87 /* write a DATA record */
88 bzero(ba->drr, sizeof (dmu_replay_record_t));
89 ba->drr->drr_type = DRR_WRITE;
90 ba->drr->drr_u.drr_write.drr_object = object;
91 ba->drr->drr_u.drr_write.drr_type = type;
92 ba->drr->drr_u.drr_write.drr_offset = offset;
93 ba->drr->drr_u.drr_write.drr_length = blksz;
95 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
97 if (dump_bytes(ba, data, blksz))
103 dump_freeobjects(struct backuparg *ba, uint64_t firstobj, uint64_t numobjs)
105 /* write a FREEOBJECTS record */
106 bzero(ba->drr, sizeof (dmu_replay_record_t));
107 ba->drr->drr_type = DRR_FREEOBJECTS;
108 ba->drr->drr_u.drr_freeobjects.drr_firstobj = firstobj;
109 ba->drr->drr_u.drr_freeobjects.drr_numobjs = numobjs;
111 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
117 dump_dnode(struct backuparg *ba, uint64_t object, dnode_phys_t *dnp)
119 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
120 return (dump_freeobjects(ba, object, 1));
122 /* write an OBJECT record */
123 bzero(ba->drr, sizeof (dmu_replay_record_t));
124 ba->drr->drr_type = DRR_OBJECT;
125 ba->drr->drr_u.drr_object.drr_object = object;
126 ba->drr->drr_u.drr_object.drr_type = dnp->dn_type;
127 ba->drr->drr_u.drr_object.drr_bonustype = dnp->dn_bonustype;
128 ba->drr->drr_u.drr_object.drr_blksz =
129 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
130 ba->drr->drr_u.drr_object.drr_bonuslen = dnp->dn_bonuslen;
131 ba->drr->drr_u.drr_object.drr_checksum = dnp->dn_checksum;
132 ba->drr->drr_u.drr_object.drr_compress = dnp->dn_compress;
134 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
137 if (dump_bytes(ba, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)))
140 /* free anything past the end of the file */
141 if (dump_free(ba, object, (dnp->dn_maxblkid + 1) *
142 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
149 #define BP_SPAN(dnp, level) \
150 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
151 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
154 backup_cb(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
155 const dnode_phys_t *dnp, void *arg)
157 struct backuparg *ba = arg;
158 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
161 if (issig(JUSTLOOKING) && issig(FORREAL))
164 if (bp == NULL && zb->zb_object == 0) {
165 uint64_t span = BP_SPAN(dnp, zb->zb_level);
166 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
167 err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
168 } else if (bp == NULL) {
169 uint64_t span = BP_SPAN(dnp, zb->zb_level);
170 err = dump_free(ba, zb->zb_object, zb->zb_blkid * span, span);
171 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
173 } else if (type == DMU_OT_DNODE) {
176 int blksz = BP_GET_LSIZE(bp);
177 uint32_t aflags = ARC_WAIT;
180 if (arc_read_nolock(NULL, spa, bp,
181 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
182 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
186 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
187 uint64_t dnobj = (zb->zb_blkid <<
188 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
189 err = dump_dnode(ba, dnobj, blk+i);
193 (void) arc_buf_remove_ref(abuf, &abuf);
194 } else { /* it's a level-0 block of a regular object */
195 uint32_t aflags = ARC_WAIT;
197 int blksz = BP_GET_LSIZE(bp);
199 if (arc_read_nolock(NULL, spa, bp,
200 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
201 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
204 err = dump_data(ba, type, zb->zb_object, zb->zb_blkid * blksz,
205 blksz, abuf->b_data);
206 (void) arc_buf_remove_ref(abuf, &abuf);
209 ASSERT(err == 0 || err == EINTR);
214 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
215 vnode_t *vp, offset_t *off)
217 dsl_dataset_t *ds = tosnap->os->os_dsl_dataset;
218 dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL;
219 dmu_replay_record_t *drr;
222 uint64_t fromtxg = 0;
224 /* tosnap must be a snapshot */
225 if (ds->ds_phys->ds_next_snap_obj == 0)
228 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
229 if (fromds && (ds->ds_dir != fromds->ds_dir ||
230 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
234 dsl_pool_t *dp = ds->ds_dir->dd_pool;
239 if (dsl_dir_is_clone(ds->ds_dir)) {
240 rw_enter(&dp->dp_config_rwlock, RW_READER);
241 err = dsl_dataset_hold_obj(dp,
242 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
243 rw_exit(&dp->dp_config_rwlock);
247 fromorigin = B_FALSE;
252 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
253 drr->drr_type = DRR_BEGIN;
254 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
255 drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION;
256 drr->drr_u.drr_begin.drr_creation_time =
257 ds->ds_phys->ds_creation_time;
258 drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type;
260 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
261 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
262 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
263 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
266 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
267 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
270 fromtxg = fromds->ds_phys->ds_creation_txg;
272 dsl_dataset_rele(fromds, FTAG);
278 ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
280 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
281 kmem_free(drr, sizeof (dmu_replay_record_t));
285 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
289 if (err == EINTR && ba.err)
291 kmem_free(drr, sizeof (dmu_replay_record_t));
295 bzero(drr, sizeof (dmu_replay_record_t));
296 drr->drr_type = DRR_END;
297 drr->drr_u.drr_end.drr_checksum = ba.zc;
299 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
300 kmem_free(drr, sizeof (dmu_replay_record_t));
304 kmem_free(drr, sizeof (dmu_replay_record_t));
309 struct recvbeginsyncarg {
312 dsl_dataset_t *origin;
314 dmu_objset_type_t type;
318 char clonelastname[MAXNAMELEN];
319 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
322 static dsl_dataset_t *
323 recv_full_sync_impl(dsl_pool_t *dp, uint64_t dsobj, dmu_objset_type_t type,
324 cred_t *cr, dmu_tx_t *tx)
328 /* This should always work, since we just created it */
329 /* XXX - create should return an owned ds */
330 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
331 DS_MODE_INCONSISTENT, dmu_recv_tag, &ds));
333 if (type != DMU_OST_NONE) {
334 (void) dmu_objset_create_impl(dp->dp_spa,
335 ds, &ds->ds_phys->ds_bp, type, tx);
338 spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC,
339 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
346 recv_full_check(void *arg1, void *arg2, dmu_tx_t *tx)
348 dsl_dir_t *dd = arg1;
349 struct recvbeginsyncarg *rbsa = arg2;
350 objset_t *mos = dd->dd_pool->dp_meta_objset;
354 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
355 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
358 return (err ? err : EEXIST);
361 /* make sure it's a snap in the same pool */
362 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
364 if (rbsa->origin->ds_phys->ds_num_children == 0)
366 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
374 recv_full_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
376 dsl_dir_t *dd = arg1;
377 struct recvbeginsyncarg *rbsa = arg2;
378 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
381 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
382 rbsa->origin, flags, cr, tx);
384 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
385 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
389 recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
391 dsl_dataset_t *ds = arg1;
392 struct recvbeginsyncarg *rbsa = arg2;
395 /* must be a head ds */
396 if (ds->ds_phys->ds_next_snap_obj != 0)
399 /* must not be a clone ds */
400 if (dsl_dir_is_clone(ds->ds_dir))
403 err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
408 /* make sure it's a snap in the same pool */
409 if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool)
411 if (rbsa->origin->ds_phys->ds_num_children == 0)
413 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
421 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
423 dsl_dataset_t *ds = arg1;
424 struct recvbeginsyncarg *rbsa = arg2;
425 dsl_dir_t *dd = ds->ds_dir;
426 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
430 * NB: caller must provide an extra hold on the dsl_dir_t, so it
431 * won't go away when dsl_dataset_destroy_sync() closes the
434 dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
436 dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
438 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
439 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
444 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
446 dsl_dataset_t *ds = arg1;
447 struct recvbeginsyncarg *rbsa = arg2;
451 /* must not have any changes since most recent snapshot */
452 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
455 /* must already be a snapshot of this fs */
456 if (ds->ds_phys->ds_prev_snap_obj == 0)
459 /* most recent snapshot must match fromguid */
460 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
463 /* temporary clone name must not exist */
464 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
465 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
466 rbsa->clonelastname, 8, 1, &val);
472 /* new snapshot name must not exist */
473 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
474 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
484 recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
486 dsl_dataset_t *ohds = arg1;
487 struct recvbeginsyncarg *rbsa = arg2;
488 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
489 dsl_dataset_t *ods, *cds;
490 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
493 /* create the temporary clone */
494 VERIFY(0 == dsl_dataset_hold_obj(dp, ohds->ds_phys->ds_prev_snap_obj,
496 dsobj = dsl_dataset_create_sync(ohds->ds_dir,
497 rbsa->clonelastname, ods, flags, cr, tx);
498 dsl_dataset_rele(ods, FTAG);
500 /* open the temporary clone */
501 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
502 DS_MODE_INCONSISTENT, dmu_recv_tag, &cds));
504 /* copy the refquota from the target fs to the clone */
505 if (ohds->ds_quota > 0)
506 dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
510 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
511 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
516 recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
518 dsl_dataset_t *ds = arg1;
520 dmu_buf_will_dirty(ds->ds_dbuf, tx);
521 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
523 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
524 ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
529 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
530 * succeeds; otherwise we will leak the holds on the datasets.
533 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
534 boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
538 struct recvbeginsyncarg rbsa;
543 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
545 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
551 rbsa.tosnap = tosnap;
552 rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL;
553 rbsa.fromguid = drrb->drr_fromguid;
554 rbsa.type = drrb->drr_type;
557 version = drrb->drr_version;
558 flags = drrb->drr_flags;
561 rbsa.type = BSWAP_32(rbsa.type);
562 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
563 version = BSWAP_64(version);
564 flags = BSWAP_32(flags);
567 if (version != DMU_BACKUP_STREAM_VERSION ||
568 rbsa.type >= DMU_OST_NUMTYPES ||
569 ((flags & DRR_FLAG_CLONE) && origin == NULL))
572 if (flags & DRR_FLAG_CI_DATA)
573 rbsa.dsflags = DS_FLAG_CI_DATASET;
575 bzero(drc, sizeof (dmu_recv_cookie_t));
576 drc->drc_drrb = drrb;
577 drc->drc_tosnap = tosnap;
578 drc->drc_force = force;
581 * Process the begin in syncing context.
583 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
584 /* offline incremental receive */
585 err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds);
590 * Only do the rollback if the most recent snapshot
591 * matches the incremental source
594 if (ds->ds_prev == NULL ||
595 ds->ds_prev->ds_phys->ds_guid !=
597 dsl_dataset_disown(ds, dmu_recv_tag);
600 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
602 rbsa.force = B_FALSE;
603 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
604 recv_incremental_check,
605 recv_offline_incremental_sync, ds, &rbsa, 1);
607 dsl_dataset_disown(ds, dmu_recv_tag);
610 drc->drc_logical_ds = drc->drc_real_ds = ds;
611 } else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
612 /* online incremental receive */
614 /* tmp clone name is: tofs/%tosnap" */
615 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
618 /* open the dataset we are logically receiving into */
619 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
624 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
625 recv_incremental_check,
626 recv_online_incremental_sync, ds, &rbsa, 5);
628 dsl_dataset_rele(ds, dmu_recv_tag);
631 drc->drc_logical_ds = ds;
632 drc->drc_real_ds = rbsa.ds;
634 /* create new fs -- full backup or clone */
635 dsl_dir_t *dd = NULL;
638 err = dsl_dir_open(tofs, FTAG, &dd, &tail);
643 dsl_dir_close(dd, FTAG);
647 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
648 err = dsl_dataset_own_obj(dd->dd_pool,
649 dd->dd_phys->dd_head_dataset_obj,
650 DS_MODE_INCONSISTENT, FTAG, &ds);
651 rw_exit(&dd->dd_pool->dp_config_rwlock);
653 dsl_dir_close(dd, FTAG);
657 dsl_dataset_make_exclusive(ds, FTAG);
658 err = dsl_sync_task_do(dd->dd_pool,
659 recv_full_existing_check,
660 recv_full_existing_sync, ds, &rbsa, 5);
661 dsl_dataset_disown(ds, FTAG);
663 err = dsl_sync_task_do(dd->dd_pool, recv_full_check,
664 recv_full_sync, dd, &rbsa, 5);
666 dsl_dir_close(dd, FTAG);
669 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
670 drc->drc_newfs = B_TRUE;
682 int bufsize; /* amount of memory allocated for buf */
687 restore_read(struct restorearg *ra, int len)
692 /* some things will require 8-byte alignment, so everything must */
693 ASSERT3U(len % 8, ==, 0);
698 ra->err = vn_rdwr(UIO_READ, ra->vp,
699 (caddr_t)ra->buf + done, len - done,
700 ra->voff, UIO_SYSSPACE, FAPPEND,
701 RLIM64_INFINITY, CRED(), &resid);
703 if (resid == len - done)
705 ra->voff += len - done - resid;
711 ASSERT3U(done, ==, len);
714 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
716 fletcher_4_incremental_native(rv, len, &ra->cksum);
721 backup_byteswap(dmu_replay_record_t *drr)
723 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
724 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
725 drr->drr_type = BSWAP_32(drr->drr_type);
726 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
727 switch (drr->drr_type) {
729 DO64(drr_begin.drr_magic);
730 DO64(drr_begin.drr_version);
731 DO64(drr_begin.drr_creation_time);
732 DO32(drr_begin.drr_type);
733 DO32(drr_begin.drr_flags);
734 DO64(drr_begin.drr_toguid);
735 DO64(drr_begin.drr_fromguid);
738 DO64(drr_object.drr_object);
739 /* DO64(drr_object.drr_allocation_txg); */
740 DO32(drr_object.drr_type);
741 DO32(drr_object.drr_bonustype);
742 DO32(drr_object.drr_blksz);
743 DO32(drr_object.drr_bonuslen);
745 case DRR_FREEOBJECTS:
746 DO64(drr_freeobjects.drr_firstobj);
747 DO64(drr_freeobjects.drr_numobjs);
750 DO64(drr_write.drr_object);
751 DO32(drr_write.drr_type);
752 DO64(drr_write.drr_offset);
753 DO64(drr_write.drr_length);
756 DO64(drr_free.drr_object);
757 DO64(drr_free.drr_offset);
758 DO64(drr_free.drr_length);
761 DO64(drr_end.drr_checksum.zc_word[0]);
762 DO64(drr_end.drr_checksum.zc_word[1]);
763 DO64(drr_end.drr_checksum.zc_word[2]);
764 DO64(drr_end.drr_checksum.zc_word[3]);
772 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
778 err = dmu_object_info(os, drro->drr_object, NULL);
780 if (err != 0 && err != ENOENT)
783 if (drro->drr_type == DMU_OT_NONE ||
784 drro->drr_type >= DMU_OT_NUMTYPES ||
785 drro->drr_bonustype >= DMU_OT_NUMTYPES ||
786 drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
787 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
788 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
789 drro->drr_blksz < SPA_MINBLOCKSIZE ||
790 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
791 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
795 if (drro->drr_bonuslen) {
796 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
801 tx = dmu_tx_create(os);
804 /* currently free, want to be allocated */
805 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
806 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1);
807 err = dmu_tx_assign(tx, TXG_WAIT);
812 err = dmu_object_claim(os, drro->drr_object,
813 drro->drr_type, drro->drr_blksz,
814 drro->drr_bonustype, drro->drr_bonuslen, tx);
816 /* currently allocated, want to be allocated */
817 dmu_tx_hold_bonus(tx, drro->drr_object);
819 * We may change blocksize, so need to
822 dmu_tx_hold_write(tx, drro->drr_object, 0, 1);
823 err = dmu_tx_assign(tx, TXG_WAIT);
829 err = dmu_object_reclaim(os, drro->drr_object,
830 drro->drr_type, drro->drr_blksz,
831 drro->drr_bonustype, drro->drr_bonuslen, tx);
838 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
839 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
844 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
845 dmu_buf_will_dirty(db, tx);
847 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
848 bcopy(data, db->db_data, drro->drr_bonuslen);
850 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
853 dmu_buf_rele(db, FTAG);
861 restore_freeobjects(struct restorearg *ra, objset_t *os,
862 struct drr_freeobjects *drrfo)
866 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
869 for (obj = drrfo->drr_firstobj;
870 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
871 (void) dmu_object_next(os, &obj, FALSE, 0)) {
874 if (dmu_object_info(os, obj, NULL) != 0)
877 err = dmu_free_object(os, obj);
885 restore_write(struct restorearg *ra, objset_t *os,
886 struct drr_write *drrw)
892 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
893 drrw->drr_type >= DMU_OT_NUMTYPES)
896 data = restore_read(ra, drrw->drr_length);
900 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
903 tx = dmu_tx_create(os);
905 dmu_tx_hold_write(tx, drrw->drr_object,
906 drrw->drr_offset, drrw->drr_length);
907 err = dmu_tx_assign(tx, TXG_WAIT);
913 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
914 dmu_write(os, drrw->drr_object,
915 drrw->drr_offset, drrw->drr_length, data, tx);
922 restore_free(struct restorearg *ra, objset_t *os,
923 struct drr_free *drrf)
927 if (drrf->drr_length != -1ULL &&
928 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
931 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
934 err = dmu_free_long_range(os, drrf->drr_object,
935 drrf->drr_offset, drrf->drr_length);
940 dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
942 if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
944 * online incremental or new fs: destroy the fs (which
945 * may be a clone) that we created
947 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
948 if (drc->drc_real_ds != drc->drc_logical_ds)
949 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
952 * offline incremental: rollback to most recent snapshot.
954 (void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE);
955 dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag);
960 * NB: callers *must* call dmu_recv_end() if this succeeds.
963 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp)
965 struct restorearg ra = { 0 };
966 dmu_replay_record_t *drr;
970 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
974 /* compute checksum of drr_begin record */
975 dmu_replay_record_t *drr;
976 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
978 drr->drr_type = DRR_BEGIN;
979 drr->drr_u.drr_begin = *drc->drc_drrb;
981 fletcher_4_incremental_byteswap(drr,
982 sizeof (dmu_replay_record_t), &ra.cksum);
984 fletcher_4_incremental_native(drr,
985 sizeof (dmu_replay_record_t), &ra.cksum);
987 kmem_free(drr, sizeof (dmu_replay_record_t));
991 struct drr_begin *drrb = drc->drc_drrb;
992 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
993 drrb->drr_version = BSWAP_64(drrb->drr_version);
994 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
995 drrb->drr_type = BSWAP_32(drrb->drr_type);
996 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
997 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1003 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1005 /* these were verified in dmu_recv_begin */
1006 ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
1007 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1010 * Open the objset we are modifying.
1012 VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0);
1014 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1017 * Read records and process them.
1020 while (ra.err == 0 &&
1021 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1022 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1028 backup_byteswap(drr);
1030 switch (drr->drr_type) {
1034 * We need to make a copy of the record header,
1035 * because restore_{object,write} may need to
1036 * restore_read(), which will invalidate drr.
1038 struct drr_object drro = drr->drr_u.drr_object;
1039 ra.err = restore_object(&ra, os, &drro);
1042 case DRR_FREEOBJECTS:
1044 struct drr_freeobjects drrfo =
1045 drr->drr_u.drr_freeobjects;
1046 ra.err = restore_freeobjects(&ra, os, &drrfo);
1051 struct drr_write drrw = drr->drr_u.drr_write;
1052 ra.err = restore_write(&ra, os, &drrw);
1057 struct drr_free drrf = drr->drr_u.drr_free;
1058 ra.err = restore_free(&ra, os, &drrf);
1063 struct drr_end drre = drr->drr_u.drr_end;
1065 * We compare against the *previous* checksum
1066 * value, because the stored checksum is of
1067 * everything before the DRR_END record.
1069 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1079 ASSERT(ra.err != 0);
1082 dmu_objset_close(os);
1086 * rollback or destroy what we created, so we don't
1087 * leave it in the restoring state.
1089 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1090 dmu_recv_abort_cleanup(drc);
1093 kmem_free(ra.buf, ra.bufsize);
1098 struct recvendsyncarg {
1100 uint64_t creation_time;
1105 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1107 dsl_dataset_t *ds = arg1;
1108 struct recvendsyncarg *resa = arg2;
1110 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1114 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1116 dsl_dataset_t *ds = arg1;
1117 struct recvendsyncarg *resa = arg2;
1119 dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
1121 /* set snapshot's creation time and guid */
1122 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1123 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1124 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1125 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1127 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1128 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1132 dmu_recv_end(dmu_recv_cookie_t *drc)
1134 struct recvendsyncarg resa;
1135 dsl_dataset_t *ds = drc->drc_logical_ds;
1139 * XXX hack; seems the ds is still dirty and
1140 * dsl_pool_zil_clean() expects it to have a ds_user_ptr
1141 * (and zil), but clone_swap() can close it.
1143 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1145 if (ds != drc->drc_real_ds) {
1146 /* we are doing an online recv */
1147 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1148 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1151 dsl_dataset_disown(ds, dmu_recv_tag);
1154 dsl_dataset_rele(ds, dmu_recv_tag);
1156 /* dsl_dataset_destroy() will disown the ds */
1157 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
1162 resa.creation_time = drc->drc_drrb->drr_creation_time;
1163 resa.toguid = drc->drc_drrb->drr_toguid;
1164 resa.tosnap = drc->drc_tosnap;
1166 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1167 recv_end_check, recv_end_sync, ds, &resa, 3);
1169 if (drc->drc_newfs) {
1170 ASSERT(ds == drc->drc_real_ds);
1171 (void) dsl_dataset_destroy(ds, dmu_recv_tag);
1174 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
1178 /* release the hold from dmu_recv_begin */
1179 dsl_dataset_disown(ds, dmu_recv_tag);