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 2009 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 (zb->zb_object != 0 && DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
166 } else if (bp == NULL && zb->zb_object == 0) {
167 uint64_t span = BP_SPAN(dnp, zb->zb_level);
168 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
169 err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
170 } else if (bp == NULL) {
171 uint64_t span = BP_SPAN(dnp, zb->zb_level);
172 err = dump_free(ba, zb->zb_object, zb->zb_blkid * span, span);
173 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
175 } else if (type == DMU_OT_DNODE) {
178 int blksz = BP_GET_LSIZE(bp);
179 uint32_t aflags = ARC_WAIT;
182 if (arc_read_nolock(NULL, spa, bp,
183 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
184 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
188 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
189 uint64_t dnobj = (zb->zb_blkid <<
190 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
191 err = dump_dnode(ba, dnobj, blk+i);
195 (void) arc_buf_remove_ref(abuf, &abuf);
196 } else { /* it's a level-0 block of a regular object */
197 uint32_t aflags = ARC_WAIT;
199 int blksz = BP_GET_LSIZE(bp);
201 if (arc_read_nolock(NULL, spa, bp,
202 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
203 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
206 err = dump_data(ba, type, zb->zb_object, zb->zb_blkid * blksz,
207 blksz, abuf->b_data);
208 (void) arc_buf_remove_ref(abuf, &abuf);
211 ASSERT(err == 0 || err == EINTR);
216 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
217 vnode_t *vp, offset_t *off)
219 dsl_dataset_t *ds = tosnap->os->os_dsl_dataset;
220 dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL;
221 dmu_replay_record_t *drr;
224 uint64_t fromtxg = 0;
226 /* tosnap must be a snapshot */
227 if (ds->ds_phys->ds_next_snap_obj == 0)
230 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
231 if (fromds && (ds->ds_dir != fromds->ds_dir ||
232 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
236 dsl_pool_t *dp = ds->ds_dir->dd_pool;
241 if (dsl_dir_is_clone(ds->ds_dir)) {
242 rw_enter(&dp->dp_config_rwlock, RW_READER);
243 err = dsl_dataset_hold_obj(dp,
244 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
245 rw_exit(&dp->dp_config_rwlock);
249 fromorigin = B_FALSE;
254 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
255 drr->drr_type = DRR_BEGIN;
256 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
257 drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION;
258 drr->drr_u.drr_begin.drr_creation_time =
259 ds->ds_phys->ds_creation_time;
260 drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type;
262 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
263 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
264 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
265 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
268 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
269 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
272 fromtxg = fromds->ds_phys->ds_creation_txg;
274 dsl_dataset_rele(fromds, FTAG);
280 ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
282 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
283 kmem_free(drr, sizeof (dmu_replay_record_t));
287 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
291 if (err == EINTR && ba.err)
293 kmem_free(drr, sizeof (dmu_replay_record_t));
297 bzero(drr, sizeof (dmu_replay_record_t));
298 drr->drr_type = DRR_END;
299 drr->drr_u.drr_end.drr_checksum = ba.zc;
301 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
302 kmem_free(drr, sizeof (dmu_replay_record_t));
306 kmem_free(drr, sizeof (dmu_replay_record_t));
311 struct recvbeginsyncarg {
314 dsl_dataset_t *origin;
316 dmu_objset_type_t type;
320 char clonelastname[MAXNAMELEN];
321 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
324 static dsl_dataset_t *
325 recv_full_sync_impl(dsl_pool_t *dp, uint64_t dsobj, dmu_objset_type_t type,
326 cred_t *cr, dmu_tx_t *tx)
330 /* This should always work, since we just created it */
331 /* XXX - create should return an owned ds */
332 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
333 DS_MODE_INCONSISTENT, dmu_recv_tag, &ds));
335 if (type != DMU_OST_NONE) {
336 (void) dmu_objset_create_impl(dp->dp_spa,
337 ds, &ds->ds_phys->ds_bp, type, tx);
340 spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC,
341 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
348 recv_full_check(void *arg1, void *arg2, dmu_tx_t *tx)
350 dsl_dir_t *dd = arg1;
351 struct recvbeginsyncarg *rbsa = arg2;
352 objset_t *mos = dd->dd_pool->dp_meta_objset;
356 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
357 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
360 return (err ? err : EEXIST);
363 /* make sure it's a snap in the same pool */
364 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
366 if (rbsa->origin->ds_phys->ds_num_children == 0)
368 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
376 recv_full_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
378 dsl_dir_t *dd = arg1;
379 struct recvbeginsyncarg *rbsa = arg2;
380 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
383 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
384 rbsa->origin, flags, cr, tx);
386 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
387 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
391 recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
393 dsl_dataset_t *ds = arg1;
394 struct recvbeginsyncarg *rbsa = arg2;
397 /* must be a head ds */
398 if (ds->ds_phys->ds_next_snap_obj != 0)
401 /* must not be a clone ds */
402 if (dsl_dir_is_clone(ds->ds_dir))
405 err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
410 /* make sure it's a snap in the same pool */
411 if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool)
413 if (rbsa->origin->ds_phys->ds_num_children == 0)
415 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
423 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
425 dsl_dataset_t *ds = arg1;
426 struct recvbeginsyncarg *rbsa = arg2;
427 dsl_dir_t *dd = ds->ds_dir;
428 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
432 * NB: caller must provide an extra hold on the dsl_dir_t, so it
433 * won't go away when dsl_dataset_destroy_sync() closes the
436 dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
438 dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
440 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
441 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
446 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
448 dsl_dataset_t *ds = arg1;
449 struct recvbeginsyncarg *rbsa = arg2;
453 /* must not have any changes since most recent snapshot */
454 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
457 /* must already be a snapshot of this fs */
458 if (ds->ds_phys->ds_prev_snap_obj == 0)
461 /* most recent snapshot must match fromguid */
462 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
465 /* temporary clone name must not exist */
466 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
467 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
468 rbsa->clonelastname, 8, 1, &val);
474 /* new snapshot name must not exist */
475 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
476 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
486 recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
488 dsl_dataset_t *ohds = arg1;
489 struct recvbeginsyncarg *rbsa = arg2;
490 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
491 dsl_dataset_t *ods, *cds;
492 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
495 /* create the temporary clone */
496 VERIFY(0 == dsl_dataset_hold_obj(dp, ohds->ds_phys->ds_prev_snap_obj,
498 dsobj = dsl_dataset_create_sync(ohds->ds_dir,
499 rbsa->clonelastname, ods, flags, cr, tx);
500 dsl_dataset_rele(ods, FTAG);
502 /* open the temporary clone */
503 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
504 DS_MODE_INCONSISTENT, dmu_recv_tag, &cds));
506 /* copy the refquota from the target fs to the clone */
507 if (ohds->ds_quota > 0)
508 dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
512 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
513 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
518 recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
520 dsl_dataset_t *ds = arg1;
522 dmu_buf_will_dirty(ds->ds_dbuf, tx);
523 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
525 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
526 ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
531 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
532 * succeeds; otherwise we will leak the holds on the datasets.
535 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
536 boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
540 struct recvbeginsyncarg rbsa;
545 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
547 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
553 rbsa.tosnap = tosnap;
554 rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL;
555 rbsa.fromguid = drrb->drr_fromguid;
556 rbsa.type = drrb->drr_type;
559 version = drrb->drr_version;
560 flags = drrb->drr_flags;
563 rbsa.type = BSWAP_32(rbsa.type);
564 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
565 version = BSWAP_64(version);
566 flags = BSWAP_32(flags);
569 if (version != DMU_BACKUP_STREAM_VERSION ||
570 rbsa.type >= DMU_OST_NUMTYPES ||
571 ((flags & DRR_FLAG_CLONE) && origin == NULL))
574 if (flags & DRR_FLAG_CI_DATA)
575 rbsa.dsflags = DS_FLAG_CI_DATASET;
577 bzero(drc, sizeof (dmu_recv_cookie_t));
578 drc->drc_drrb = drrb;
579 drc->drc_tosnap = tosnap;
580 drc->drc_force = force;
583 * Process the begin in syncing context.
585 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
586 /* offline incremental receive */
587 err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds);
592 * Only do the rollback if the most recent snapshot
593 * matches the incremental source
596 if (ds->ds_prev == NULL ||
597 ds->ds_prev->ds_phys->ds_guid !=
599 dsl_dataset_disown(ds, dmu_recv_tag);
602 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
604 rbsa.force = B_FALSE;
605 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
606 recv_incremental_check,
607 recv_offline_incremental_sync, ds, &rbsa, 1);
609 dsl_dataset_disown(ds, dmu_recv_tag);
612 drc->drc_logical_ds = drc->drc_real_ds = ds;
613 } else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
614 /* online incremental receive */
616 /* tmp clone name is: tofs/%tosnap" */
617 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
620 /* open the dataset we are logically receiving into */
621 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
626 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
627 recv_incremental_check,
628 recv_online_incremental_sync, ds, &rbsa, 5);
630 dsl_dataset_rele(ds, dmu_recv_tag);
633 drc->drc_logical_ds = ds;
634 drc->drc_real_ds = rbsa.ds;
636 /* create new fs -- full backup or clone */
637 dsl_dir_t *dd = NULL;
640 err = dsl_dir_open(tofs, FTAG, &dd, &tail);
645 dsl_dir_close(dd, FTAG);
649 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
650 err = dsl_dataset_own_obj(dd->dd_pool,
651 dd->dd_phys->dd_head_dataset_obj,
652 DS_MODE_INCONSISTENT, FTAG, &ds);
653 rw_exit(&dd->dd_pool->dp_config_rwlock);
655 dsl_dir_close(dd, FTAG);
659 dsl_dataset_make_exclusive(ds, FTAG);
660 err = dsl_sync_task_do(dd->dd_pool,
661 recv_full_existing_check,
662 recv_full_existing_sync, ds, &rbsa, 5);
663 dsl_dataset_disown(ds, FTAG);
665 err = dsl_sync_task_do(dd->dd_pool, recv_full_check,
666 recv_full_sync, dd, &rbsa, 5);
668 dsl_dir_close(dd, FTAG);
671 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
672 drc->drc_newfs = B_TRUE;
684 int bufsize; /* amount of memory allocated for buf */
689 restore_read(struct restorearg *ra, int len)
694 /* some things will require 8-byte alignment, so everything must */
695 ASSERT3U(len % 8, ==, 0);
700 ra->err = vn_rdwr(UIO_READ, ra->vp,
701 (caddr_t)ra->buf + done, len - done,
702 ra->voff, UIO_SYSSPACE, FAPPEND,
703 RLIM64_INFINITY, CRED(), &resid);
705 if (resid == len - done)
707 ra->voff += len - done - resid;
713 ASSERT3U(done, ==, len);
716 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
718 fletcher_4_incremental_native(rv, len, &ra->cksum);
723 backup_byteswap(dmu_replay_record_t *drr)
725 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
726 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
727 drr->drr_type = BSWAP_32(drr->drr_type);
728 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
729 switch (drr->drr_type) {
731 DO64(drr_begin.drr_magic);
732 DO64(drr_begin.drr_version);
733 DO64(drr_begin.drr_creation_time);
734 DO32(drr_begin.drr_type);
735 DO32(drr_begin.drr_flags);
736 DO64(drr_begin.drr_toguid);
737 DO64(drr_begin.drr_fromguid);
740 DO64(drr_object.drr_object);
741 /* DO64(drr_object.drr_allocation_txg); */
742 DO32(drr_object.drr_type);
743 DO32(drr_object.drr_bonustype);
744 DO32(drr_object.drr_blksz);
745 DO32(drr_object.drr_bonuslen);
747 case DRR_FREEOBJECTS:
748 DO64(drr_freeobjects.drr_firstobj);
749 DO64(drr_freeobjects.drr_numobjs);
752 DO64(drr_write.drr_object);
753 DO32(drr_write.drr_type);
754 DO64(drr_write.drr_offset);
755 DO64(drr_write.drr_length);
758 DO64(drr_free.drr_object);
759 DO64(drr_free.drr_offset);
760 DO64(drr_free.drr_length);
763 DO64(drr_end.drr_checksum.zc_word[0]);
764 DO64(drr_end.drr_checksum.zc_word[1]);
765 DO64(drr_end.drr_checksum.zc_word[2]);
766 DO64(drr_end.drr_checksum.zc_word[3]);
774 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
780 if (drro->drr_type == DMU_OT_NONE ||
781 drro->drr_type >= DMU_OT_NUMTYPES ||
782 drro->drr_bonustype >= DMU_OT_NUMTYPES ||
783 drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
784 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
785 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
786 drro->drr_blksz < SPA_MINBLOCKSIZE ||
787 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
788 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
792 err = dmu_object_info(os, drro->drr_object, NULL);
794 if (err != 0 && err != ENOENT)
797 if (drro->drr_bonuslen) {
798 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
804 /* currently free, want to be allocated */
805 tx = dmu_tx_create(os);
806 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
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);
817 /* currently allocated, want to be allocated */
818 err = dmu_object_reclaim(os, drro->drr_object,
819 drro->drr_type, drro->drr_blksz,
820 drro->drr_bonustype, drro->drr_bonuslen);
825 tx = dmu_tx_create(os);
826 dmu_tx_hold_bonus(tx, drro->drr_object);
827 err = dmu_tx_assign(tx, TXG_WAIT);
833 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
834 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
839 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
840 dmu_buf_will_dirty(db, tx);
842 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
843 bcopy(data, db->db_data, drro->drr_bonuslen);
845 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
848 dmu_buf_rele(db, FTAG);
856 restore_freeobjects(struct restorearg *ra, objset_t *os,
857 struct drr_freeobjects *drrfo)
861 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
864 for (obj = drrfo->drr_firstobj;
865 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
866 (void) dmu_object_next(os, &obj, FALSE, 0)) {
869 if (dmu_object_info(os, obj, NULL) != 0)
872 err = dmu_free_object(os, obj);
880 restore_write(struct restorearg *ra, objset_t *os,
881 struct drr_write *drrw)
887 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
888 drrw->drr_type >= DMU_OT_NUMTYPES)
891 data = restore_read(ra, drrw->drr_length);
895 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
898 tx = dmu_tx_create(os);
900 dmu_tx_hold_write(tx, drrw->drr_object,
901 drrw->drr_offset, drrw->drr_length);
902 err = dmu_tx_assign(tx, TXG_WAIT);
908 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
909 dmu_write(os, drrw->drr_object,
910 drrw->drr_offset, drrw->drr_length, data, tx);
917 restore_free(struct restorearg *ra, objset_t *os,
918 struct drr_free *drrf)
922 if (drrf->drr_length != -1ULL &&
923 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
926 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
929 err = dmu_free_long_range(os, drrf->drr_object,
930 drrf->drr_offset, drrf->drr_length);
935 dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
937 if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
939 * online incremental or new fs: destroy the fs (which
940 * may be a clone) that we created
942 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
943 if (drc->drc_real_ds != drc->drc_logical_ds)
944 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
947 * offline incremental: rollback to most recent snapshot.
949 (void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE);
950 dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag);
955 * NB: callers *must* call dmu_recv_end() if this succeeds.
958 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp)
960 struct restorearg ra = { 0 };
961 dmu_replay_record_t *drr;
965 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
969 /* compute checksum of drr_begin record */
970 dmu_replay_record_t *drr;
971 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
973 drr->drr_type = DRR_BEGIN;
974 drr->drr_u.drr_begin = *drc->drc_drrb;
976 fletcher_4_incremental_byteswap(drr,
977 sizeof (dmu_replay_record_t), &ra.cksum);
979 fletcher_4_incremental_native(drr,
980 sizeof (dmu_replay_record_t), &ra.cksum);
982 kmem_free(drr, sizeof (dmu_replay_record_t));
986 struct drr_begin *drrb = drc->drc_drrb;
987 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
988 drrb->drr_version = BSWAP_64(drrb->drr_version);
989 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
990 drrb->drr_type = BSWAP_32(drrb->drr_type);
991 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
992 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
998 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1000 /* these were verified in dmu_recv_begin */
1001 ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
1002 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1005 * Open the objset we are modifying.
1007 VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0);
1009 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1012 * Read records and process them.
1015 while (ra.err == 0 &&
1016 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1017 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1023 backup_byteswap(drr);
1025 switch (drr->drr_type) {
1029 * We need to make a copy of the record header,
1030 * because restore_{object,write} may need to
1031 * restore_read(), which will invalidate drr.
1033 struct drr_object drro = drr->drr_u.drr_object;
1034 ra.err = restore_object(&ra, os, &drro);
1037 case DRR_FREEOBJECTS:
1039 struct drr_freeobjects drrfo =
1040 drr->drr_u.drr_freeobjects;
1041 ra.err = restore_freeobjects(&ra, os, &drrfo);
1046 struct drr_write drrw = drr->drr_u.drr_write;
1047 ra.err = restore_write(&ra, os, &drrw);
1052 struct drr_free drrf = drr->drr_u.drr_free;
1053 ra.err = restore_free(&ra, os, &drrf);
1058 struct drr_end drre = drr->drr_u.drr_end;
1060 * We compare against the *previous* checksum
1061 * value, because the stored checksum is of
1062 * everything before the DRR_END record.
1064 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1074 ASSERT(ra.err != 0);
1077 dmu_objset_close(os);
1081 * rollback or destroy what we created, so we don't
1082 * leave it in the restoring state.
1084 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1085 dmu_recv_abort_cleanup(drc);
1088 kmem_free(ra.buf, ra.bufsize);
1093 struct recvendsyncarg {
1095 uint64_t creation_time;
1100 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1102 dsl_dataset_t *ds = arg1;
1103 struct recvendsyncarg *resa = arg2;
1105 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1109 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1111 dsl_dataset_t *ds = arg1;
1112 struct recvendsyncarg *resa = arg2;
1114 dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
1116 /* set snapshot's creation time and guid */
1117 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1118 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1119 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1120 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1122 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1123 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1127 dmu_recv_end(dmu_recv_cookie_t *drc)
1129 struct recvendsyncarg resa;
1130 dsl_dataset_t *ds = drc->drc_logical_ds;
1134 * XXX hack; seems the ds is still dirty and
1135 * dsl_pool_zil_clean() expects it to have a ds_user_ptr
1136 * (and zil), but clone_swap() can close it.
1138 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1140 if (ds != drc->drc_real_ds) {
1141 /* we are doing an online recv */
1142 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1143 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1146 dsl_dataset_disown(ds, dmu_recv_tag);
1149 dsl_dataset_rele(ds, dmu_recv_tag);
1151 /* dsl_dataset_destroy() will disown the ds */
1152 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
1157 resa.creation_time = drc->drc_drrb->drr_creation_time;
1158 resa.toguid = drc->drc_drrb->drr_toguid;
1159 resa.tosnap = drc->drc_tosnap;
1161 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1162 recv_end_check, recv_end_sync, ds, &resa, 3);
1164 if (drc->drc_newfs) {
1165 ASSERT(ds == drc->drc_real_ds);
1166 (void) dsl_dataset_destroy(ds, dmu_recv_tag);
1169 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
1173 /* release the hold from dmu_recv_begin */
1174 dsl_dataset_disown(ds, dmu_recv_tag);