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 (bp == NULL && zb->zb_object == 0) {
184 uint64_t span = BP_SPAN(dnp, zb->zb_level);
185 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
186 err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
187 } else if (bp == NULL) {
188 uint64_t span = BP_SPAN(dnp, zb->zb_level);
189 err = dump_free(ba, zb->zb_object, zb->zb_blkid * span, span);
190 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
192 } else if (type == DMU_OT_DNODE) {
195 int blksz = BP_GET_LSIZE(bp);
196 uint32_t aflags = ARC_WAIT;
199 if (arc_read_nolock(NULL, spa, bp,
200 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
201 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
205 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
206 uint64_t dnobj = (zb->zb_blkid <<
207 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
208 err = dump_dnode(ba, dnobj, blk+i);
212 (void) arc_buf_remove_ref(abuf, &abuf);
213 } else { /* it's a level-0 block of a regular object */
214 uint32_t aflags = ARC_WAIT;
216 int blksz = BP_GET_LSIZE(bp);
218 if (arc_read_nolock(NULL, spa, bp,
219 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
220 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
223 err = dump_data(ba, type, zb->zb_object, zb->zb_blkid * blksz,
224 blksz, abuf->b_data);
225 (void) arc_buf_remove_ref(abuf, &abuf);
228 ASSERT(err == 0 || err == EINTR);
233 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
234 struct file *fp, offset_t *off)
236 dsl_dataset_t *ds = tosnap->os->os_dsl_dataset;
237 dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL;
238 dmu_replay_record_t *drr;
241 uint64_t fromtxg = 0;
243 /* tosnap must be a snapshot */
244 if (ds->ds_phys->ds_next_snap_obj == 0)
247 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
248 if (fromds && (ds->ds_dir != fromds->ds_dir ||
249 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
253 dsl_pool_t *dp = ds->ds_dir->dd_pool;
258 if (dsl_dir_is_clone(ds->ds_dir)) {
259 rw_enter(&dp->dp_config_rwlock, RW_READER);
260 err = dsl_dataset_hold_obj(dp,
261 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
262 rw_exit(&dp->dp_config_rwlock);
266 fromorigin = B_FALSE;
271 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
272 drr->drr_type = DRR_BEGIN;
273 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
274 drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION;
275 drr->drr_u.drr_begin.drr_creation_time =
276 ds->ds_phys->ds_creation_time;
277 drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type;
279 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
280 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
281 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
282 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
285 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
286 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
289 fromtxg = fromds->ds_phys->ds_creation_txg;
291 dsl_dataset_rele(fromds, FTAG);
298 ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
300 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
301 kmem_free(drr, sizeof (dmu_replay_record_t));
305 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
309 if (err == EINTR && ba.err)
311 kmem_free(drr, sizeof (dmu_replay_record_t));
315 bzero(drr, sizeof (dmu_replay_record_t));
316 drr->drr_type = DRR_END;
317 drr->drr_u.drr_end.drr_checksum = ba.zc;
319 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
320 kmem_free(drr, sizeof (dmu_replay_record_t));
324 kmem_free(drr, sizeof (dmu_replay_record_t));
329 struct recvbeginsyncarg {
332 dsl_dataset_t *origin;
334 dmu_objset_type_t type;
338 char clonelastname[MAXNAMELEN];
339 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
342 static dsl_dataset_t *
343 recv_full_sync_impl(dsl_pool_t *dp, uint64_t dsobj, dmu_objset_type_t type,
344 cred_t *cr, dmu_tx_t *tx)
348 /* This should always work, since we just created it */
349 /* XXX - create should return an owned ds */
350 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
351 DS_MODE_INCONSISTENT, dmu_recv_tag, &ds));
353 if (type != DMU_OST_NONE) {
354 (void) dmu_objset_create_impl(dp->dp_spa,
355 ds, &ds->ds_phys->ds_bp, type, tx);
358 spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC,
359 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
366 recv_full_check(void *arg1, void *arg2, dmu_tx_t *tx)
368 dsl_dir_t *dd = arg1;
369 struct recvbeginsyncarg *rbsa = arg2;
370 objset_t *mos = dd->dd_pool->dp_meta_objset;
374 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
375 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
378 return (err ? err : EEXIST);
381 /* make sure it's a snap in the same pool */
382 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
384 if (rbsa->origin->ds_phys->ds_num_children == 0)
386 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
394 recv_full_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
396 dsl_dir_t *dd = arg1;
397 struct recvbeginsyncarg *rbsa = arg2;
398 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
401 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
402 rbsa->origin, flags, cr, tx);
404 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
405 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
409 recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
411 dsl_dataset_t *ds = arg1;
412 struct recvbeginsyncarg *rbsa = arg2;
415 /* must be a head ds */
416 if (ds->ds_phys->ds_next_snap_obj != 0)
419 /* must not be a clone ds */
420 if (dsl_dir_is_clone(ds->ds_dir))
423 err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
428 /* make sure it's a snap in the same pool */
429 if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool)
431 if (rbsa->origin->ds_phys->ds_num_children == 0)
433 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
441 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
443 dsl_dataset_t *ds = arg1;
444 struct recvbeginsyncarg *rbsa = arg2;
445 dsl_dir_t *dd = ds->ds_dir;
446 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
450 * NB: caller must provide an extra hold on the dsl_dir_t, so it
451 * won't go away when dsl_dataset_destroy_sync() closes the
454 dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
456 dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
458 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
459 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
464 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
466 dsl_dataset_t *ds = arg1;
467 struct recvbeginsyncarg *rbsa = arg2;
471 /* must not have any changes since most recent snapshot */
472 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
475 /* must already be a snapshot of this fs */
476 if (ds->ds_phys->ds_prev_snap_obj == 0)
479 /* most recent snapshot must match fromguid */
480 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
483 /* temporary clone name must not exist */
484 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
485 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
486 rbsa->clonelastname, 8, 1, &val);
492 /* new snapshot name must not exist */
493 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
494 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
504 recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
506 dsl_dataset_t *ohds = arg1;
507 struct recvbeginsyncarg *rbsa = arg2;
508 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
509 dsl_dataset_t *ods, *cds;
510 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
513 /* create the temporary clone */
514 VERIFY(0 == dsl_dataset_hold_obj(dp, ohds->ds_phys->ds_prev_snap_obj,
516 dsobj = dsl_dataset_create_sync(ohds->ds_dir,
517 rbsa->clonelastname, ods, flags, cr, tx);
518 dsl_dataset_rele(ods, FTAG);
520 /* open the temporary clone */
521 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
522 DS_MODE_INCONSISTENT, dmu_recv_tag, &cds));
524 /* copy the refquota from the target fs to the clone */
525 if (ohds->ds_quota > 0)
526 dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
530 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
531 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
536 recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
538 dsl_dataset_t *ds = arg1;
540 dmu_buf_will_dirty(ds->ds_dbuf, tx);
541 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
543 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
544 ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
549 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
550 * succeeds; otherwise we will leak the holds on the datasets.
553 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
554 boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
558 struct recvbeginsyncarg rbsa;
563 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
565 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
571 rbsa.tosnap = tosnap;
572 rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL;
573 rbsa.fromguid = drrb->drr_fromguid;
574 rbsa.type = drrb->drr_type;
577 version = drrb->drr_version;
578 flags = drrb->drr_flags;
581 rbsa.type = BSWAP_32(rbsa.type);
582 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
583 version = BSWAP_64(version);
584 flags = BSWAP_32(flags);
587 if (version != DMU_BACKUP_STREAM_VERSION ||
588 rbsa.type >= DMU_OST_NUMTYPES ||
589 ((flags & DRR_FLAG_CLONE) && origin == NULL))
592 if (flags & DRR_FLAG_CI_DATA)
593 rbsa.dsflags = DS_FLAG_CI_DATASET;
595 bzero(drc, sizeof (dmu_recv_cookie_t));
596 drc->drc_drrb = drrb;
597 drc->drc_tosnap = tosnap;
598 drc->drc_force = force;
601 * Process the begin in syncing context.
603 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
604 /* offline incremental receive */
605 err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds);
610 * Only do the rollback if the most recent snapshot
611 * matches the incremental source
614 if (ds->ds_prev == NULL ||
615 ds->ds_prev->ds_phys->ds_guid !=
617 dsl_dataset_disown(ds, dmu_recv_tag);
620 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
622 rbsa.force = B_FALSE;
623 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
624 recv_incremental_check,
625 recv_offline_incremental_sync, ds, &rbsa, 1);
627 dsl_dataset_disown(ds, dmu_recv_tag);
630 drc->drc_logical_ds = drc->drc_real_ds = ds;
631 } else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
632 /* online incremental receive */
634 /* tmp clone name is: tofs/%tosnap" */
635 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
638 /* open the dataset we are logically receiving into */
639 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
644 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
645 recv_incremental_check,
646 recv_online_incremental_sync, ds, &rbsa, 5);
648 dsl_dataset_rele(ds, dmu_recv_tag);
651 drc->drc_logical_ds = ds;
652 drc->drc_real_ds = rbsa.ds;
654 /* create new fs -- full backup or clone */
655 dsl_dir_t *dd = NULL;
658 err = dsl_dir_open(tofs, FTAG, &dd, &tail);
663 dsl_dir_close(dd, FTAG);
667 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
668 err = dsl_dataset_own_obj(dd->dd_pool,
669 dd->dd_phys->dd_head_dataset_obj,
670 DS_MODE_INCONSISTENT, FTAG, &ds);
671 rw_exit(&dd->dd_pool->dp_config_rwlock);
673 dsl_dir_close(dd, FTAG);
677 dsl_dataset_make_exclusive(ds, FTAG);
678 err = dsl_sync_task_do(dd->dd_pool,
679 recv_full_existing_check,
680 recv_full_existing_sync, ds, &rbsa, 5);
681 dsl_dataset_disown(ds, FTAG);
683 err = dsl_sync_task_do(dd->dd_pool, recv_full_check,
684 recv_full_sync, dd, &rbsa, 5);
686 dsl_dir_close(dd, FTAG);
689 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
690 drc->drc_newfs = B_TRUE;
703 int bufsize; /* amount of memory allocated for buf */
708 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, int *resid)
716 auio.uio_iov = &aiov;
718 auio.uio_resid = len;
719 auio.uio_segflg = UIO_SYSSPACE;
720 auio.uio_rw = UIO_READ;
721 auio.uio_offset = off;
722 auio.uio_td = ra->td;
724 error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
726 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
729 *resid = auio.uio_resid;
734 restore_read(struct restorearg *ra, int len)
739 /* some things will require 8-byte alignment, so everything must */
740 ASSERT3U(len % 8, ==, 0);
745 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
746 len - done, ra->voff, &resid);
748 if (resid == len - done)
750 ra->voff += len - done - resid;
756 ASSERT3U(done, ==, len);
759 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
761 fletcher_4_incremental_native(rv, len, &ra->cksum);
766 backup_byteswap(dmu_replay_record_t *drr)
768 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
769 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
770 drr->drr_type = BSWAP_32(drr->drr_type);
771 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
772 switch (drr->drr_type) {
774 DO64(drr_begin.drr_magic);
775 DO64(drr_begin.drr_version);
776 DO64(drr_begin.drr_creation_time);
777 DO32(drr_begin.drr_type);
778 DO32(drr_begin.drr_flags);
779 DO64(drr_begin.drr_toguid);
780 DO64(drr_begin.drr_fromguid);
783 DO64(drr_object.drr_object);
784 /* DO64(drr_object.drr_allocation_txg); */
785 DO32(drr_object.drr_type);
786 DO32(drr_object.drr_bonustype);
787 DO32(drr_object.drr_blksz);
788 DO32(drr_object.drr_bonuslen);
790 case DRR_FREEOBJECTS:
791 DO64(drr_freeobjects.drr_firstobj);
792 DO64(drr_freeobjects.drr_numobjs);
795 DO64(drr_write.drr_object);
796 DO32(drr_write.drr_type);
797 DO64(drr_write.drr_offset);
798 DO64(drr_write.drr_length);
801 DO64(drr_free.drr_object);
802 DO64(drr_free.drr_offset);
803 DO64(drr_free.drr_length);
806 DO64(drr_end.drr_checksum.zc_word[0]);
807 DO64(drr_end.drr_checksum.zc_word[1]);
808 DO64(drr_end.drr_checksum.zc_word[2]);
809 DO64(drr_end.drr_checksum.zc_word[3]);
817 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
823 if (drro->drr_type == DMU_OT_NONE ||
824 drro->drr_type >= DMU_OT_NUMTYPES ||
825 drro->drr_bonustype >= DMU_OT_NUMTYPES ||
826 drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
827 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
828 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
829 drro->drr_blksz < SPA_MINBLOCKSIZE ||
830 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
831 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
835 err = dmu_object_info(os, drro->drr_object, NULL);
837 if (err != 0 && err != ENOENT)
840 if (drro->drr_bonuslen) {
841 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
847 /* currently free, want to be allocated */
848 tx = dmu_tx_create(os);
849 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
850 err = dmu_tx_assign(tx, TXG_WAIT);
855 err = dmu_object_claim(os, drro->drr_object,
856 drro->drr_type, drro->drr_blksz,
857 drro->drr_bonustype, drro->drr_bonuslen, tx);
860 /* currently allocated, want to be allocated */
861 err = dmu_object_reclaim(os, drro->drr_object,
862 drro->drr_type, drro->drr_blksz,
863 drro->drr_bonustype, drro->drr_bonuslen);
868 tx = dmu_tx_create(os);
869 dmu_tx_hold_bonus(tx, drro->drr_object);
870 err = dmu_tx_assign(tx, TXG_WAIT);
876 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
877 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
882 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
883 dmu_buf_will_dirty(db, tx);
885 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
886 bcopy(data, db->db_data, drro->drr_bonuslen);
888 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
891 dmu_buf_rele(db, FTAG);
899 restore_freeobjects(struct restorearg *ra, objset_t *os,
900 struct drr_freeobjects *drrfo)
904 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
907 for (obj = drrfo->drr_firstobj;
908 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
909 (void) dmu_object_next(os, &obj, FALSE, 0)) {
912 if (dmu_object_info(os, obj, NULL) != 0)
915 err = dmu_free_object(os, obj);
923 restore_write(struct restorearg *ra, objset_t *os,
924 struct drr_write *drrw)
930 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
931 drrw->drr_type >= DMU_OT_NUMTYPES)
934 data = restore_read(ra, drrw->drr_length);
938 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
941 tx = dmu_tx_create(os);
943 dmu_tx_hold_write(tx, drrw->drr_object,
944 drrw->drr_offset, drrw->drr_length);
945 err = dmu_tx_assign(tx, TXG_WAIT);
951 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
952 dmu_write(os, drrw->drr_object,
953 drrw->drr_offset, drrw->drr_length, data, tx);
960 restore_free(struct restorearg *ra, objset_t *os,
961 struct drr_free *drrf)
965 if (drrf->drr_length != -1ULL &&
966 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
969 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
972 err = dmu_free_long_range(os, drrf->drr_object,
973 drrf->drr_offset, drrf->drr_length);
978 dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
980 if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
982 * online incremental or new fs: destroy the fs (which
983 * may be a clone) that we created
985 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
986 if (drc->drc_real_ds != drc->drc_logical_ds)
987 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
990 * offline incremental: rollback to most recent snapshot.
992 (void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE);
993 dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag);
998 * NB: callers *must* call dmu_recv_end() if this succeeds.
1001 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp)
1003 kthread_t *td = curthread;
1004 struct restorearg ra = { 0 };
1005 dmu_replay_record_t *drr;
1009 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1013 /* compute checksum of drr_begin record */
1014 dmu_replay_record_t *drr;
1015 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1017 drr->drr_type = DRR_BEGIN;
1018 drr->drr_u.drr_begin = *drc->drc_drrb;
1020 fletcher_4_incremental_byteswap(drr,
1021 sizeof (dmu_replay_record_t), &ra.cksum);
1023 fletcher_4_incremental_native(drr,
1024 sizeof (dmu_replay_record_t), &ra.cksum);
1026 kmem_free(drr, sizeof (dmu_replay_record_t));
1030 struct drr_begin *drrb = drc->drc_drrb;
1031 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1032 drrb->drr_version = BSWAP_64(drrb->drr_version);
1033 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1034 drrb->drr_type = BSWAP_32(drrb->drr_type);
1035 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1036 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1043 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1045 /* these were verified in dmu_recv_begin */
1046 ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
1047 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1050 * Open the objset we are modifying.
1052 VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0);
1054 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1057 * Read records and process them.
1060 while (ra.err == 0 &&
1061 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1062 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1068 backup_byteswap(drr);
1070 switch (drr->drr_type) {
1074 * We need to make a copy of the record header,
1075 * because restore_{object,write} may need to
1076 * restore_read(), which will invalidate drr.
1078 struct drr_object drro = drr->drr_u.drr_object;
1079 ra.err = restore_object(&ra, os, &drro);
1082 case DRR_FREEOBJECTS:
1084 struct drr_freeobjects drrfo =
1085 drr->drr_u.drr_freeobjects;
1086 ra.err = restore_freeobjects(&ra, os, &drrfo);
1091 struct drr_write drrw = drr->drr_u.drr_write;
1092 ra.err = restore_write(&ra, os, &drrw);
1097 struct drr_free drrf = drr->drr_u.drr_free;
1098 ra.err = restore_free(&ra, os, &drrf);
1103 struct drr_end drre = drr->drr_u.drr_end;
1105 * We compare against the *previous* checksum
1106 * value, because the stored checksum is of
1107 * everything before the DRR_END record.
1109 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1119 ASSERT(ra.err != 0);
1122 dmu_objset_close(os);
1126 * rollback or destroy what we created, so we don't
1127 * leave it in the restoring state.
1129 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1130 dmu_recv_abort_cleanup(drc);
1133 kmem_free(ra.buf, ra.bufsize);
1138 struct recvendsyncarg {
1140 uint64_t creation_time;
1145 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1147 dsl_dataset_t *ds = arg1;
1148 struct recvendsyncarg *resa = arg2;
1150 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1154 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1156 dsl_dataset_t *ds = arg1;
1157 struct recvendsyncarg *resa = arg2;
1159 dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
1161 /* set snapshot's creation time and guid */
1162 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1163 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1164 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1165 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1167 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1168 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1172 dmu_recv_end(dmu_recv_cookie_t *drc)
1174 struct recvendsyncarg resa;
1175 dsl_dataset_t *ds = drc->drc_logical_ds;
1179 * XXX hack; seems the ds is still dirty and
1180 * dsl_pool_zil_clean() expects it to have a ds_user_ptr
1181 * (and zil), but clone_swap() can close it.
1183 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1185 if (ds != drc->drc_real_ds) {
1186 /* we are doing an online recv */
1187 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1188 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1191 dsl_dataset_disown(ds, dmu_recv_tag);
1194 dsl_dataset_rele(ds, dmu_recv_tag);
1196 /* dsl_dataset_destroy() will disown the ds */
1197 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
1202 resa.creation_time = drc->drc_drrb->drr_creation_time;
1203 resa.toguid = drc->drc_drrb->drr_toguid;
1204 resa.tosnap = drc->drc_tosnap;
1206 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1207 recv_end_check, recv_end_sync, ds, &resa, 3);
1209 if (drc->drc_newfs) {
1210 ASSERT(ds == drc->drc_real_ds);
1211 (void) dsl_dataset_destroy(ds, dmu_recv_tag);
1214 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
1218 /* release the hold from dmu_recv_begin */
1219 dsl_dataset_disown(ds, dmu_recv_tag);