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;
55 dump_bytes(struct backuparg *ba, void *buf, int len)
60 ASSERT3U(len % 8, ==, 0);
62 fletcher_4_incremental_native(buf, len, &ba->zc);
69 auio.uio_segflg = UIO_SYSSPACE;
70 auio.uio_rw = UIO_WRITE;
71 auio.uio_offset = (off_t)-1;
74 if (ba->fp->f_type == DTYPE_VNODE)
76 ba->err = fo_write(ba->fp, &auio, ba->td->td_ucred, 0, ba->td);
78 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
87 dump_free(struct backuparg *ba, uint64_t object, uint64_t offset,
90 /* write a FREE record */
91 bzero(ba->drr, sizeof (dmu_replay_record_t));
92 ba->drr->drr_type = DRR_FREE;
93 ba->drr->drr_u.drr_free.drr_object = object;
94 ba->drr->drr_u.drr_free.drr_offset = offset;
95 ba->drr->drr_u.drr_free.drr_length = length;
97 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
103 dump_data(struct backuparg *ba, dmu_object_type_t type,
104 uint64_t object, uint64_t offset, int blksz, void *data)
106 /* write a DATA record */
107 bzero(ba->drr, sizeof (dmu_replay_record_t));
108 ba->drr->drr_type = DRR_WRITE;
109 ba->drr->drr_u.drr_write.drr_object = object;
110 ba->drr->drr_u.drr_write.drr_type = type;
111 ba->drr->drr_u.drr_write.drr_offset = offset;
112 ba->drr->drr_u.drr_write.drr_length = blksz;
114 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
116 if (dump_bytes(ba, data, blksz))
122 dump_freeobjects(struct backuparg *ba, uint64_t firstobj, uint64_t numobjs)
124 /* write a FREEOBJECTS record */
125 bzero(ba->drr, sizeof (dmu_replay_record_t));
126 ba->drr->drr_type = DRR_FREEOBJECTS;
127 ba->drr->drr_u.drr_freeobjects.drr_firstobj = firstobj;
128 ba->drr->drr_u.drr_freeobjects.drr_numobjs = numobjs;
130 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
136 dump_dnode(struct backuparg *ba, uint64_t object, dnode_phys_t *dnp)
138 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
139 return (dump_freeobjects(ba, object, 1));
141 /* write an OBJECT record */
142 bzero(ba->drr, sizeof (dmu_replay_record_t));
143 ba->drr->drr_type = DRR_OBJECT;
144 ba->drr->drr_u.drr_object.drr_object = object;
145 ba->drr->drr_u.drr_object.drr_type = dnp->dn_type;
146 ba->drr->drr_u.drr_object.drr_bonustype = dnp->dn_bonustype;
147 ba->drr->drr_u.drr_object.drr_blksz =
148 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
149 ba->drr->drr_u.drr_object.drr_bonuslen = dnp->dn_bonuslen;
150 ba->drr->drr_u.drr_object.drr_checksum = dnp->dn_checksum;
151 ba->drr->drr_u.drr_object.drr_compress = dnp->dn_compress;
153 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
156 if (dump_bytes(ba, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)))
159 /* free anything past the end of the file */
160 if (dump_free(ba, object, (dnp->dn_maxblkid + 1) *
161 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
168 #define BP_SPAN(dnp, level) \
169 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
170 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
173 backup_cb(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
174 const dnode_phys_t *dnp, void *arg)
176 struct backuparg *ba = arg;
177 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
180 if (issig(JUSTLOOKING) && issig(FORREAL))
183 if (zb->zb_object != 0 && DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
185 } else if (bp == NULL && zb->zb_object == 0) {
186 uint64_t span = BP_SPAN(dnp, zb->zb_level);
187 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
188 err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
189 } else if (bp == NULL) {
190 uint64_t span = BP_SPAN(dnp, zb->zb_level);
191 err = dump_free(ba, zb->zb_object, zb->zb_blkid * span, span);
192 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
194 } else if (type == DMU_OT_DNODE) {
197 int blksz = BP_GET_LSIZE(bp);
198 uint32_t aflags = ARC_WAIT;
201 if (arc_read_nolock(NULL, spa, bp,
202 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
203 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
207 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
208 uint64_t dnobj = (zb->zb_blkid <<
209 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
210 err = dump_dnode(ba, dnobj, blk+i);
214 (void) arc_buf_remove_ref(abuf, &abuf);
215 } else { /* it's a level-0 block of a regular object */
216 uint32_t aflags = ARC_WAIT;
218 int blksz = BP_GET_LSIZE(bp);
220 if (arc_read_nolock(NULL, spa, bp,
221 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
222 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
225 err = dump_data(ba, type, zb->zb_object, zb->zb_blkid * blksz,
226 blksz, abuf->b_data);
227 (void) arc_buf_remove_ref(abuf, &abuf);
230 ASSERT(err == 0 || err == EINTR);
235 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
236 struct file *fp, offset_t *off)
238 dsl_dataset_t *ds = tosnap->os->os_dsl_dataset;
239 dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL;
240 dmu_replay_record_t *drr;
243 uint64_t fromtxg = 0;
245 /* tosnap must be a snapshot */
246 if (ds->ds_phys->ds_next_snap_obj == 0)
249 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
250 if (fromds && (ds->ds_dir != fromds->ds_dir ||
251 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
255 dsl_pool_t *dp = ds->ds_dir->dd_pool;
260 if (dsl_dir_is_clone(ds->ds_dir)) {
261 rw_enter(&dp->dp_config_rwlock, RW_READER);
262 err = dsl_dataset_hold_obj(dp,
263 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
264 rw_exit(&dp->dp_config_rwlock);
268 fromorigin = B_FALSE;
273 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
274 drr->drr_type = DRR_BEGIN;
275 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
276 drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION;
277 drr->drr_u.drr_begin.drr_creation_time =
278 ds->ds_phys->ds_creation_time;
279 drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type;
281 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
282 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
283 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
284 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
287 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
288 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
291 fromtxg = fromds->ds_phys->ds_creation_txg;
293 dsl_dataset_rele(fromds, FTAG);
300 ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
302 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
303 kmem_free(drr, sizeof (dmu_replay_record_t));
307 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
311 if (err == EINTR && ba.err)
313 kmem_free(drr, sizeof (dmu_replay_record_t));
317 bzero(drr, sizeof (dmu_replay_record_t));
318 drr->drr_type = DRR_END;
319 drr->drr_u.drr_end.drr_checksum = ba.zc;
321 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
322 kmem_free(drr, sizeof (dmu_replay_record_t));
326 kmem_free(drr, sizeof (dmu_replay_record_t));
331 struct recvbeginsyncarg {
334 dsl_dataset_t *origin;
336 dmu_objset_type_t type;
340 char clonelastname[MAXNAMELEN];
341 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
344 static dsl_dataset_t *
345 recv_full_sync_impl(dsl_pool_t *dp, uint64_t dsobj, dmu_objset_type_t type,
346 cred_t *cr, dmu_tx_t *tx)
350 /* This should always work, since we just created it */
351 /* XXX - create should return an owned ds */
352 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
353 DS_MODE_INCONSISTENT, dmu_recv_tag, &ds));
355 if (type != DMU_OST_NONE) {
356 (void) dmu_objset_create_impl(dp->dp_spa,
357 ds, &ds->ds_phys->ds_bp, type, tx);
360 spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC,
361 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
368 recv_full_check(void *arg1, void *arg2, dmu_tx_t *tx)
370 dsl_dir_t *dd = arg1;
371 struct recvbeginsyncarg *rbsa = arg2;
372 objset_t *mos = dd->dd_pool->dp_meta_objset;
376 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
377 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
380 return (err ? err : EEXIST);
383 /* make sure it's a snap in the same pool */
384 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
386 if (rbsa->origin->ds_phys->ds_num_children == 0)
388 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
396 recv_full_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
398 dsl_dir_t *dd = arg1;
399 struct recvbeginsyncarg *rbsa = arg2;
400 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
403 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
404 rbsa->origin, flags, cr, tx);
406 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
407 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
411 recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
413 dsl_dataset_t *ds = arg1;
414 struct recvbeginsyncarg *rbsa = arg2;
417 /* must be a head ds */
418 if (ds->ds_phys->ds_next_snap_obj != 0)
421 /* must not be a clone ds */
422 if (dsl_dir_is_clone(ds->ds_dir))
425 err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
430 /* make sure it's a snap in the same pool */
431 if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool)
433 if (rbsa->origin->ds_phys->ds_num_children == 0)
435 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
443 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
445 dsl_dataset_t *ds = arg1;
446 struct recvbeginsyncarg *rbsa = arg2;
447 dsl_dir_t *dd = ds->ds_dir;
448 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
452 * NB: caller must provide an extra hold on the dsl_dir_t, so it
453 * won't go away when dsl_dataset_destroy_sync() closes the
456 dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
458 dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
460 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
461 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
466 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
468 dsl_dataset_t *ds = arg1;
469 struct recvbeginsyncarg *rbsa = arg2;
473 /* must not have any changes since most recent snapshot */
474 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
477 /* must already be a snapshot of this fs */
478 if (ds->ds_phys->ds_prev_snap_obj == 0)
481 /* most recent snapshot must match fromguid */
482 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
485 /* temporary clone name must not exist */
486 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
487 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
488 rbsa->clonelastname, 8, 1, &val);
494 /* new snapshot name must not exist */
495 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
496 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
506 recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
508 dsl_dataset_t *ohds = arg1;
509 struct recvbeginsyncarg *rbsa = arg2;
510 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
511 dsl_dataset_t *ods, *cds;
512 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
515 /* create the temporary clone */
516 VERIFY(0 == dsl_dataset_hold_obj(dp, ohds->ds_phys->ds_prev_snap_obj,
518 dsobj = dsl_dataset_create_sync(ohds->ds_dir,
519 rbsa->clonelastname, ods, flags, cr, tx);
520 dsl_dataset_rele(ods, FTAG);
522 /* open the temporary clone */
523 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
524 DS_MODE_INCONSISTENT, dmu_recv_tag, &cds));
526 /* copy the refquota from the target fs to the clone */
527 if (ohds->ds_quota > 0)
528 dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
532 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
533 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
538 recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
540 dsl_dataset_t *ds = arg1;
542 dmu_buf_will_dirty(ds->ds_dbuf, tx);
543 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
545 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
546 ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
551 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
552 * succeeds; otherwise we will leak the holds on the datasets.
555 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
556 boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
560 struct recvbeginsyncarg rbsa;
565 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
567 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
573 rbsa.tosnap = tosnap;
574 rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL;
575 rbsa.fromguid = drrb->drr_fromguid;
576 rbsa.type = drrb->drr_type;
579 version = drrb->drr_version;
580 flags = drrb->drr_flags;
583 rbsa.type = BSWAP_32(rbsa.type);
584 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
585 version = BSWAP_64(version);
586 flags = BSWAP_32(flags);
589 if (version != DMU_BACKUP_STREAM_VERSION ||
590 rbsa.type >= DMU_OST_NUMTYPES ||
591 ((flags & DRR_FLAG_CLONE) && origin == NULL))
594 if (flags & DRR_FLAG_CI_DATA)
595 rbsa.dsflags = DS_FLAG_CI_DATASET;
597 bzero(drc, sizeof (dmu_recv_cookie_t));
598 drc->drc_drrb = drrb;
599 drc->drc_tosnap = tosnap;
600 drc->drc_force = force;
603 * Process the begin in syncing context.
605 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
606 /* offline incremental receive */
607 err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds);
612 * Only do the rollback if the most recent snapshot
613 * matches the incremental source
616 if (ds->ds_prev == NULL ||
617 ds->ds_prev->ds_phys->ds_guid !=
619 dsl_dataset_disown(ds, dmu_recv_tag);
622 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
624 rbsa.force = B_FALSE;
625 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
626 recv_incremental_check,
627 recv_offline_incremental_sync, ds, &rbsa, 1);
629 dsl_dataset_disown(ds, dmu_recv_tag);
632 drc->drc_logical_ds = drc->drc_real_ds = ds;
633 } else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
634 /* online incremental receive */
636 /* tmp clone name is: tofs/%tosnap" */
637 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
640 /* open the dataset we are logically receiving into */
641 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
646 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
647 recv_incremental_check,
648 recv_online_incremental_sync, ds, &rbsa, 5);
650 dsl_dataset_rele(ds, dmu_recv_tag);
653 drc->drc_logical_ds = ds;
654 drc->drc_real_ds = rbsa.ds;
656 /* create new fs -- full backup or clone */
657 dsl_dir_t *dd = NULL;
660 err = dsl_dir_open(tofs, FTAG, &dd, &tail);
665 dsl_dir_close(dd, FTAG);
669 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
670 err = dsl_dataset_own_obj(dd->dd_pool,
671 dd->dd_phys->dd_head_dataset_obj,
672 DS_MODE_INCONSISTENT, FTAG, &ds);
673 rw_exit(&dd->dd_pool->dp_config_rwlock);
675 dsl_dir_close(dd, FTAG);
679 dsl_dataset_make_exclusive(ds, FTAG);
680 err = dsl_sync_task_do(dd->dd_pool,
681 recv_full_existing_check,
682 recv_full_existing_sync, ds, &rbsa, 5);
683 dsl_dataset_disown(ds, FTAG);
685 err = dsl_sync_task_do(dd->dd_pool, recv_full_check,
686 recv_full_sync, dd, &rbsa, 5);
688 dsl_dir_close(dd, FTAG);
691 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
692 drc->drc_newfs = B_TRUE;
705 int bufsize; /* amount of memory allocated for buf */
710 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, int *resid)
718 auio.uio_iov = &aiov;
720 auio.uio_resid = len;
721 auio.uio_segflg = UIO_SYSSPACE;
722 auio.uio_rw = UIO_READ;
723 auio.uio_offset = off;
724 auio.uio_td = ra->td;
726 error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
728 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
731 *resid = auio.uio_resid;
736 restore_read(struct restorearg *ra, int len)
741 /* some things will require 8-byte alignment, so everything must */
742 ASSERT3U(len % 8, ==, 0);
747 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
748 len - done, ra->voff, &resid);
750 if (resid == len - done)
752 ra->voff += len - done - resid;
758 ASSERT3U(done, ==, len);
761 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
763 fletcher_4_incremental_native(rv, len, &ra->cksum);
768 backup_byteswap(dmu_replay_record_t *drr)
770 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
771 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
772 drr->drr_type = BSWAP_32(drr->drr_type);
773 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
774 switch (drr->drr_type) {
776 DO64(drr_begin.drr_magic);
777 DO64(drr_begin.drr_version);
778 DO64(drr_begin.drr_creation_time);
779 DO32(drr_begin.drr_type);
780 DO32(drr_begin.drr_flags);
781 DO64(drr_begin.drr_toguid);
782 DO64(drr_begin.drr_fromguid);
785 DO64(drr_object.drr_object);
786 /* DO64(drr_object.drr_allocation_txg); */
787 DO32(drr_object.drr_type);
788 DO32(drr_object.drr_bonustype);
789 DO32(drr_object.drr_blksz);
790 DO32(drr_object.drr_bonuslen);
792 case DRR_FREEOBJECTS:
793 DO64(drr_freeobjects.drr_firstobj);
794 DO64(drr_freeobjects.drr_numobjs);
797 DO64(drr_write.drr_object);
798 DO32(drr_write.drr_type);
799 DO64(drr_write.drr_offset);
800 DO64(drr_write.drr_length);
803 DO64(drr_free.drr_object);
804 DO64(drr_free.drr_offset);
805 DO64(drr_free.drr_length);
808 DO64(drr_end.drr_checksum.zc_word[0]);
809 DO64(drr_end.drr_checksum.zc_word[1]);
810 DO64(drr_end.drr_checksum.zc_word[2]);
811 DO64(drr_end.drr_checksum.zc_word[3]);
819 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
825 if (drro->drr_type == DMU_OT_NONE ||
826 drro->drr_type >= DMU_OT_NUMTYPES ||
827 drro->drr_bonustype >= DMU_OT_NUMTYPES ||
828 drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
829 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
830 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
831 drro->drr_blksz < SPA_MINBLOCKSIZE ||
832 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
833 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
837 err = dmu_object_info(os, drro->drr_object, NULL);
839 if (err != 0 && err != ENOENT)
842 if (drro->drr_bonuslen) {
843 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
849 /* currently free, want to be allocated */
850 tx = dmu_tx_create(os);
851 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
852 err = dmu_tx_assign(tx, TXG_WAIT);
857 err = dmu_object_claim(os, drro->drr_object,
858 drro->drr_type, drro->drr_blksz,
859 drro->drr_bonustype, drro->drr_bonuslen, tx);
862 /* currently allocated, want to be allocated */
863 err = dmu_object_reclaim(os, drro->drr_object,
864 drro->drr_type, drro->drr_blksz,
865 drro->drr_bonustype, drro->drr_bonuslen);
870 tx = dmu_tx_create(os);
871 dmu_tx_hold_bonus(tx, drro->drr_object);
872 err = dmu_tx_assign(tx, TXG_WAIT);
878 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
879 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
884 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
885 dmu_buf_will_dirty(db, tx);
887 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
888 bcopy(data, db->db_data, drro->drr_bonuslen);
890 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
893 dmu_buf_rele(db, FTAG);
901 restore_freeobjects(struct restorearg *ra, objset_t *os,
902 struct drr_freeobjects *drrfo)
906 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
909 for (obj = drrfo->drr_firstobj;
910 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
911 (void) dmu_object_next(os, &obj, FALSE, 0)) {
914 if (dmu_object_info(os, obj, NULL) != 0)
917 err = dmu_free_object(os, obj);
925 restore_write(struct restorearg *ra, objset_t *os,
926 struct drr_write *drrw)
932 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
933 drrw->drr_type >= DMU_OT_NUMTYPES)
936 data = restore_read(ra, drrw->drr_length);
940 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
943 tx = dmu_tx_create(os);
945 dmu_tx_hold_write(tx, drrw->drr_object,
946 drrw->drr_offset, drrw->drr_length);
947 err = dmu_tx_assign(tx, TXG_WAIT);
953 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
954 dmu_write(os, drrw->drr_object,
955 drrw->drr_offset, drrw->drr_length, data, tx);
962 restore_free(struct restorearg *ra, objset_t *os,
963 struct drr_free *drrf)
967 if (drrf->drr_length != -1ULL &&
968 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
971 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
974 err = dmu_free_long_range(os, drrf->drr_object,
975 drrf->drr_offset, drrf->drr_length);
980 dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
982 if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
984 * online incremental or new fs: destroy the fs (which
985 * may be a clone) that we created
987 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
988 if (drc->drc_real_ds != drc->drc_logical_ds)
989 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
992 * offline incremental: rollback to most recent snapshot.
994 (void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE);
995 dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag);
1000 * NB: callers *must* call dmu_recv_end() if this succeeds.
1003 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp)
1005 kthread_t *td = curthread;
1006 struct restorearg ra = { 0 };
1007 dmu_replay_record_t *drr;
1011 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1015 /* compute checksum of drr_begin record */
1016 dmu_replay_record_t *drr;
1017 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1019 drr->drr_type = DRR_BEGIN;
1020 drr->drr_u.drr_begin = *drc->drc_drrb;
1022 fletcher_4_incremental_byteswap(drr,
1023 sizeof (dmu_replay_record_t), &ra.cksum);
1025 fletcher_4_incremental_native(drr,
1026 sizeof (dmu_replay_record_t), &ra.cksum);
1028 kmem_free(drr, sizeof (dmu_replay_record_t));
1032 struct drr_begin *drrb = drc->drc_drrb;
1033 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1034 drrb->drr_version = BSWAP_64(drrb->drr_version);
1035 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1036 drrb->drr_type = BSWAP_32(drrb->drr_type);
1037 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1038 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1045 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1047 /* these were verified in dmu_recv_begin */
1048 ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
1049 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1052 * Open the objset we are modifying.
1054 VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0);
1056 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1059 * Read records and process them.
1062 while (ra.err == 0 &&
1063 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1064 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1070 backup_byteswap(drr);
1072 switch (drr->drr_type) {
1076 * We need to make a copy of the record header,
1077 * because restore_{object,write} may need to
1078 * restore_read(), which will invalidate drr.
1080 struct drr_object drro = drr->drr_u.drr_object;
1081 ra.err = restore_object(&ra, os, &drro);
1084 case DRR_FREEOBJECTS:
1086 struct drr_freeobjects drrfo =
1087 drr->drr_u.drr_freeobjects;
1088 ra.err = restore_freeobjects(&ra, os, &drrfo);
1093 struct drr_write drrw = drr->drr_u.drr_write;
1094 ra.err = restore_write(&ra, os, &drrw);
1099 struct drr_free drrf = drr->drr_u.drr_free;
1100 ra.err = restore_free(&ra, os, &drrf);
1105 struct drr_end drre = drr->drr_u.drr_end;
1107 * We compare against the *previous* checksum
1108 * value, because the stored checksum is of
1109 * everything before the DRR_END record.
1111 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1121 ASSERT(ra.err != 0);
1124 dmu_objset_close(os);
1128 * rollback or destroy what we created, so we don't
1129 * leave it in the restoring state.
1131 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1132 dmu_recv_abort_cleanup(drc);
1135 kmem_free(ra.buf, ra.bufsize);
1140 struct recvendsyncarg {
1142 uint64_t creation_time;
1147 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1149 dsl_dataset_t *ds = arg1;
1150 struct recvendsyncarg *resa = arg2;
1152 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1156 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1158 dsl_dataset_t *ds = arg1;
1159 struct recvendsyncarg *resa = arg2;
1161 dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
1163 /* set snapshot's creation time and guid */
1164 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1165 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1166 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1167 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1169 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1170 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1174 dmu_recv_end(dmu_recv_cookie_t *drc)
1176 struct recvendsyncarg resa;
1177 dsl_dataset_t *ds = drc->drc_logical_ds;
1181 * XXX hack; seems the ds is still dirty and
1182 * dsl_pool_zil_clean() expects it to have a ds_user_ptr
1183 * (and zil), but clone_swap() can close it.
1185 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1187 if (ds != drc->drc_real_ds) {
1188 /* we are doing an online recv */
1189 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1190 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1193 dsl_dataset_disown(ds, dmu_recv_tag);
1196 dsl_dataset_rele(ds, dmu_recv_tag);
1198 /* dsl_dataset_destroy() will disown the ds */
1199 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
1204 resa.creation_time = drc->drc_drrb->drr_creation_time;
1205 resa.toguid = drc->drc_drrb->drr_toguid;
1206 resa.tosnap = drc->drc_tosnap;
1208 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1209 recv_end_check, recv_end_sync, ds, &resa, 3);
1211 if (drc->drc_newfs) {
1212 ASSERT(ds == drc->drc_real_ds);
1213 (void) dsl_dataset_destroy(ds, dmu_recv_tag);
1216 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
1220 /* release the hold from dmu_recv_begin */
1221 dsl_dataset_disown(ds, dmu_recv_tag);