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.
26 #pragma ident "%Z%%M% %I% %E% SMI"
29 #include <sys/dmu_impl.h>
30 #include <sys/dmu_tx.h>
32 #include <sys/dnode.h>
33 #include <sys/zfs_context.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dsl_dataset.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dsl_pool.h>
39 #include <sys/dsl_synctask.h>
40 #include <sys/zfs_ioctl.h>
42 #include <sys/zio_checksum.h>
44 static char *dmu_recv_tag = "dmu_recv_tag";
47 dmu_replay_record_t *drr;
57 dump_bytes(struct backuparg *ba, void *buf, int len)
62 ASSERT3U(len % 8, ==, 0);
64 fletcher_4_incremental_native(buf, len, &ba->zc);
71 auio.uio_segflg = UIO_SYSSPACE;
72 auio.uio_rw = UIO_WRITE;
73 auio.uio_offset = (off_t)-1;
76 if (ba->fp->f_type == DTYPE_VNODE)
78 ba->err = fo_write(ba->fp, &auio, ba->td->td_ucred, 0, ba->td);
80 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
89 dump_free(struct backuparg *ba, uint64_t object, uint64_t offset,
92 /* write a FREE record */
93 bzero(ba->drr, sizeof (dmu_replay_record_t));
94 ba->drr->drr_type = DRR_FREE;
95 ba->drr->drr_u.drr_free.drr_object = object;
96 ba->drr->drr_u.drr_free.drr_offset = offset;
97 ba->drr->drr_u.drr_free.drr_length = length;
99 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
105 dump_data(struct backuparg *ba, dmu_object_type_t type,
106 uint64_t object, uint64_t offset, int blksz, void *data)
108 /* write a DATA record */
109 bzero(ba->drr, sizeof (dmu_replay_record_t));
110 ba->drr->drr_type = DRR_WRITE;
111 ba->drr->drr_u.drr_write.drr_object = object;
112 ba->drr->drr_u.drr_write.drr_type = type;
113 ba->drr->drr_u.drr_write.drr_offset = offset;
114 ba->drr->drr_u.drr_write.drr_length = blksz;
116 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
118 if (dump_bytes(ba, data, blksz))
124 dump_freeobjects(struct backuparg *ba, uint64_t firstobj, uint64_t numobjs)
126 /* write a FREEOBJECTS record */
127 bzero(ba->drr, sizeof (dmu_replay_record_t));
128 ba->drr->drr_type = DRR_FREEOBJECTS;
129 ba->drr->drr_u.drr_freeobjects.drr_firstobj = firstobj;
130 ba->drr->drr_u.drr_freeobjects.drr_numobjs = numobjs;
132 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
138 dump_dnode(struct backuparg *ba, uint64_t object, dnode_phys_t *dnp)
140 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
141 return (dump_freeobjects(ba, object, 1));
143 /* write an OBJECT record */
144 bzero(ba->drr, sizeof (dmu_replay_record_t));
145 ba->drr->drr_type = DRR_OBJECT;
146 ba->drr->drr_u.drr_object.drr_object = object;
147 ba->drr->drr_u.drr_object.drr_type = dnp->dn_type;
148 ba->drr->drr_u.drr_object.drr_bonustype = dnp->dn_bonustype;
149 ba->drr->drr_u.drr_object.drr_blksz =
150 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
151 ba->drr->drr_u.drr_object.drr_bonuslen = dnp->dn_bonuslen;
152 ba->drr->drr_u.drr_object.drr_checksum = dnp->dn_checksum;
153 ba->drr->drr_u.drr_object.drr_compress = dnp->dn_compress;
155 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t)))
158 if (dump_bytes(ba, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)))
161 /* free anything past the end of the file */
162 if (dump_free(ba, object, (dnp->dn_maxblkid + 1) *
163 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
170 #define BP_SPAN(dnp, level) \
171 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
172 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
175 backup_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
177 struct backuparg *ba = arg;
178 uint64_t object = bc->bc_bookmark.zb_object;
179 int level = bc->bc_bookmark.zb_level;
180 uint64_t blkid = bc->bc_bookmark.zb_blkid;
181 blkptr_t *bp = bc->bc_blkptr.blk_birth ? &bc->bc_blkptr : NULL;
182 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
183 void *data = bc->bc_data;
186 if (issig(JUSTLOOKING) && issig(FORREAL))
189 ASSERT(data || bp == NULL);
191 if (bp == NULL && object == 0) {
192 uint64_t span = BP_SPAN(bc->bc_dnode, level);
193 uint64_t dnobj = (blkid * span) >> DNODE_SHIFT;
194 err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT);
195 } else if (bp == NULL) {
196 uint64_t span = BP_SPAN(bc->bc_dnode, level);
197 err = dump_free(ba, object, blkid * span, span);
198 } else if (data && level == 0 && type == DMU_OT_DNODE) {
199 dnode_phys_t *blk = data;
201 int blksz = BP_GET_LSIZE(bp);
203 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
205 (blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
206 err = dump_dnode(ba, dnobj, blk+i);
210 } else if (level == 0 &&
211 type != DMU_OT_DNODE && type != DMU_OT_OBJSET) {
212 int blksz = BP_GET_LSIZE(bp);
214 uint32_t aflags = ARC_WAIT;
218 zb.zb_objset = ba->os->os->os_dsl_dataset->ds_object;
219 zb.zb_object = object;
222 (void) arc_read_nolock(NULL, spa, bp,
223 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
224 ZIO_FLAG_MUSTSUCCEED, &aflags, &zb);
227 err = dump_data(ba, type, object, blkid * blksz,
228 blksz, abuf->b_data);
229 (void) arc_buf_remove_ref(abuf, &abuf);
232 err = dump_data(ba, type, object, blkid * blksz,
237 ASSERT(err == 0 || err == EINTR);
242 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
243 struct file *fp, offset_t *off)
245 dsl_dataset_t *ds = tosnap->os->os_dsl_dataset;
246 dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL;
247 dmu_replay_record_t *drr;
250 uint64_t fromtxg = 0;
252 /* tosnap must be a snapshot */
253 if (ds->ds_phys->ds_next_snap_obj == 0)
256 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
257 if (fromds && (ds->ds_dir != fromds->ds_dir ||
258 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
262 dsl_pool_t *dp = ds->ds_dir->dd_pool;
267 if (dsl_dir_is_clone(ds->ds_dir)) {
268 rw_enter(&dp->dp_config_rwlock, RW_READER);
269 err = dsl_dataset_hold_obj(dp,
270 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
271 rw_exit(&dp->dp_config_rwlock);
275 fromorigin = B_FALSE;
280 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
281 drr->drr_type = DRR_BEGIN;
282 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
283 drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION;
284 drr->drr_u.drr_begin.drr_creation_time =
285 ds->ds_phys->ds_creation_time;
286 drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type;
288 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
289 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
290 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
291 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
294 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
295 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
298 fromtxg = fromds->ds_phys->ds_creation_txg;
300 dsl_dataset_rele(fromds, FTAG);
307 ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0);
309 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
310 kmem_free(drr, sizeof (dmu_replay_record_t));
314 err = traverse_dsl_dataset(ds, fromtxg,
315 ADVANCE_PRE | ADVANCE_HOLES | ADVANCE_DATA | ADVANCE_NOLOCK,
319 if (err == EINTR && ba.err)
321 kmem_free(drr, sizeof (dmu_replay_record_t));
325 bzero(drr, sizeof (dmu_replay_record_t));
326 drr->drr_type = DRR_END;
327 drr->drr_u.drr_end.drr_checksum = ba.zc;
329 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) {
330 kmem_free(drr, sizeof (dmu_replay_record_t));
334 kmem_free(drr, sizeof (dmu_replay_record_t));
339 struct recvbeginsyncarg {
342 dsl_dataset_t *origin;
344 dmu_objset_type_t type;
348 char clonelastname[MAXNAMELEN];
349 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
352 static dsl_dataset_t *
353 recv_full_sync_impl(dsl_pool_t *dp, uint64_t dsobj, dmu_objset_type_t type,
354 cred_t *cr, dmu_tx_t *tx)
358 /* This should always work, since we just created it */
359 /* XXX - create should return an owned ds */
360 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
361 DS_MODE_INCONSISTENT, dmu_recv_tag, &ds));
363 if (type != DMU_OST_NONE) {
364 (void) dmu_objset_create_impl(dp->dp_spa,
365 ds, &ds->ds_phys->ds_bp, type, tx);
368 spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC,
369 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
376 recv_full_check(void *arg1, void *arg2, dmu_tx_t *tx)
378 dsl_dir_t *dd = arg1;
379 struct recvbeginsyncarg *rbsa = arg2;
380 objset_t *mos = dd->dd_pool->dp_meta_objset;
384 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
385 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
388 return (err ? err : EEXIST);
391 /* make sure it's a snap in the same pool */
392 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
394 if (rbsa->origin->ds_phys->ds_num_children == 0)
396 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
404 recv_full_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
406 dsl_dir_t *dd = arg1;
407 struct recvbeginsyncarg *rbsa = arg2;
408 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
411 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
412 rbsa->origin, flags, cr, tx);
414 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
415 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
419 recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
421 dsl_dataset_t *ds = arg1;
422 struct recvbeginsyncarg *rbsa = arg2;
425 /* must be a head ds */
426 if (ds->ds_phys->ds_next_snap_obj != 0)
429 /* must not be a clone ds */
430 if (dsl_dir_is_clone(ds->ds_dir))
433 err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
438 /* make sure it's a snap in the same pool */
439 if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool)
441 if (rbsa->origin->ds_phys->ds_num_children == 0)
443 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
451 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
453 dsl_dataset_t *ds = arg1;
454 struct recvbeginsyncarg *rbsa = arg2;
455 dsl_dir_t *dd = ds->ds_dir;
456 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
460 * NB: caller must provide an extra hold on the dsl_dir_t, so it
461 * won't go away when dsl_dataset_destroy_sync() closes the
464 dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
466 dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
468 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj,
469 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx);
474 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
476 dsl_dataset_t *ds = arg1;
477 struct recvbeginsyncarg *rbsa = arg2;
481 /* must not have any changes since most recent snapshot */
482 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
485 /* must already be a snapshot of this fs */
486 if (ds->ds_phys->ds_prev_snap_obj == 0)
489 /* most recent snapshot must match fromguid */
490 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid)
493 /* temporary clone name must not exist */
494 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
495 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
496 rbsa->clonelastname, 8, 1, &val);
502 /* new snapshot name must not exist */
503 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
504 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
514 recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
516 dsl_dataset_t *ohds = arg1;
517 struct recvbeginsyncarg *rbsa = arg2;
518 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
519 dsl_dataset_t *ods, *cds;
520 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
523 /* create the temporary clone */
524 VERIFY(0 == dsl_dataset_hold_obj(dp, ohds->ds_phys->ds_prev_snap_obj,
526 dsobj = dsl_dataset_create_sync(ohds->ds_dir,
527 rbsa->clonelastname, ods, flags, cr, tx);
528 dsl_dataset_rele(ods, FTAG);
530 /* open the temporary clone */
531 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj,
532 DS_MODE_INCONSISTENT, dmu_recv_tag, &cds));
534 /* copy the refquota from the target fs to the clone */
535 if (ohds->ds_quota > 0)
536 dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx);
540 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
541 dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
546 recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
548 dsl_dataset_t *ds = arg1;
550 dmu_buf_will_dirty(ds->ds_dbuf, tx);
551 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
553 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
554 ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
559 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
560 * succeeds; otherwise we will leak the holds on the datasets.
563 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
564 boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
568 struct recvbeginsyncarg rbsa;
573 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
575 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
581 rbsa.tosnap = tosnap;
582 rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL;
583 rbsa.fromguid = drrb->drr_fromguid;
584 rbsa.type = drrb->drr_type;
587 version = drrb->drr_version;
588 flags = drrb->drr_flags;
591 rbsa.type = BSWAP_32(rbsa.type);
592 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
593 version = BSWAP_64(version);
594 flags = BSWAP_32(flags);
597 if (version != DMU_BACKUP_STREAM_VERSION ||
598 rbsa.type >= DMU_OST_NUMTYPES ||
599 ((flags & DRR_FLAG_CLONE) && origin == NULL))
602 if (flags & DRR_FLAG_CI_DATA)
603 rbsa.dsflags = DS_FLAG_CI_DATASET;
605 bzero(drc, sizeof (dmu_recv_cookie_t));
606 drc->drc_drrb = drrb;
607 drc->drc_tosnap = tosnap;
608 drc->drc_force = force;
611 * Process the begin in syncing context.
613 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
614 /* offline incremental receive */
615 err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds);
620 * Only do the rollback if the most recent snapshot
621 * matches the incremental source
624 if (ds->ds_prev == NULL ||
625 ds->ds_prev->ds_phys->ds_guid !=
627 dsl_dataset_disown(ds, dmu_recv_tag);
630 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
632 rbsa.force = B_FALSE;
633 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
634 recv_incremental_check,
635 recv_offline_incremental_sync, ds, &rbsa, 1);
637 dsl_dataset_disown(ds, dmu_recv_tag);
640 drc->drc_logical_ds = drc->drc_real_ds = ds;
641 } else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
642 /* online incremental receive */
644 /* tmp clone name is: tofs/%tosnap" */
645 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
648 /* open the dataset we are logically receiving into */
649 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
654 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
655 recv_incremental_check,
656 recv_online_incremental_sync, ds, &rbsa, 5);
658 dsl_dataset_rele(ds, dmu_recv_tag);
661 drc->drc_logical_ds = ds;
662 drc->drc_real_ds = rbsa.ds;
664 /* create new fs -- full backup or clone */
665 dsl_dir_t *dd = NULL;
668 err = dsl_dir_open(tofs, FTAG, &dd, &tail);
673 dsl_dir_close(dd, FTAG);
677 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
678 err = dsl_dataset_own_obj(dd->dd_pool,
679 dd->dd_phys->dd_head_dataset_obj,
680 DS_MODE_INCONSISTENT, FTAG, &ds);
681 rw_exit(&dd->dd_pool->dp_config_rwlock);
683 dsl_dir_close(dd, FTAG);
687 dsl_dataset_make_exclusive(ds, FTAG);
688 err = dsl_sync_task_do(dd->dd_pool,
689 recv_full_existing_check,
690 recv_full_existing_sync, ds, &rbsa, 5);
691 dsl_dataset_disown(ds, FTAG);
693 err = dsl_sync_task_do(dd->dd_pool, recv_full_check,
694 recv_full_sync, dd, &rbsa, 5);
696 dsl_dir_close(dd, FTAG);
699 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
700 drc->drc_newfs = B_TRUE;
713 int bufsize; /* amount of memory allocated for buf */
718 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, int *resid)
726 auio.uio_iov = &aiov;
728 auio.uio_resid = len;
729 auio.uio_segflg = UIO_SYSSPACE;
730 auio.uio_rw = UIO_READ;
731 auio.uio_offset = off;
732 auio.uio_td = ra->td;
734 error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
736 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
739 *resid = auio.uio_resid;
744 restore_read(struct restorearg *ra, int len)
749 /* some things will require 8-byte alignment, so everything must */
750 ASSERT3U(len % 8, ==, 0);
755 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
756 len - done, ra->voff, &resid);
758 if (resid == len - done)
760 ra->voff += len - done - resid;
766 ASSERT3U(done, ==, len);
769 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
771 fletcher_4_incremental_native(rv, len, &ra->cksum);
776 backup_byteswap(dmu_replay_record_t *drr)
778 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
779 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
780 drr->drr_type = BSWAP_32(drr->drr_type);
781 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
782 switch (drr->drr_type) {
784 DO64(drr_begin.drr_magic);
785 DO64(drr_begin.drr_version);
786 DO64(drr_begin.drr_creation_time);
787 DO32(drr_begin.drr_type);
788 DO32(drr_begin.drr_flags);
789 DO64(drr_begin.drr_toguid);
790 DO64(drr_begin.drr_fromguid);
793 DO64(drr_object.drr_object);
794 /* DO64(drr_object.drr_allocation_txg); */
795 DO32(drr_object.drr_type);
796 DO32(drr_object.drr_bonustype);
797 DO32(drr_object.drr_blksz);
798 DO32(drr_object.drr_bonuslen);
800 case DRR_FREEOBJECTS:
801 DO64(drr_freeobjects.drr_firstobj);
802 DO64(drr_freeobjects.drr_numobjs);
805 DO64(drr_write.drr_object);
806 DO32(drr_write.drr_type);
807 DO64(drr_write.drr_offset);
808 DO64(drr_write.drr_length);
811 DO64(drr_free.drr_object);
812 DO64(drr_free.drr_offset);
813 DO64(drr_free.drr_length);
816 DO64(drr_end.drr_checksum.zc_word[0]);
817 DO64(drr_end.drr_checksum.zc_word[1]);
818 DO64(drr_end.drr_checksum.zc_word[2]);
819 DO64(drr_end.drr_checksum.zc_word[3]);
827 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
832 err = dmu_object_info(os, drro->drr_object, NULL);
834 if (err != 0 && err != ENOENT)
837 if (drro->drr_type == DMU_OT_NONE ||
838 drro->drr_type >= DMU_OT_NUMTYPES ||
839 drro->drr_bonustype >= DMU_OT_NUMTYPES ||
840 drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS ||
841 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
842 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
843 drro->drr_blksz < SPA_MINBLOCKSIZE ||
844 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
845 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
849 tx = dmu_tx_create(os);
852 /* currently free, want to be allocated */
853 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
854 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1);
855 err = dmu_tx_assign(tx, TXG_WAIT);
860 err = dmu_object_claim(os, drro->drr_object,
861 drro->drr_type, drro->drr_blksz,
862 drro->drr_bonustype, drro->drr_bonuslen, tx);
864 /* currently allocated, want to be allocated */
865 dmu_tx_hold_bonus(tx, drro->drr_object);
867 * We may change blocksize and delete old content,
868 * so need to hold_write and hold_free.
870 dmu_tx_hold_write(tx, drro->drr_object, 0, 1);
871 dmu_tx_hold_free(tx, drro->drr_object, 0, DMU_OBJECT_END);
872 err = dmu_tx_assign(tx, TXG_WAIT);
878 err = dmu_object_reclaim(os, drro->drr_object,
879 drro->drr_type, drro->drr_blksz,
880 drro->drr_bonustype, drro->drr_bonuslen, tx);
887 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx);
888 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
890 if (drro->drr_bonuslen) {
893 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
894 dmu_buf_will_dirty(db, tx);
896 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
897 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
902 bcopy(data, db->db_data, drro->drr_bonuslen);
904 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
907 dmu_buf_rele(db, FTAG);
915 restore_freeobjects(struct restorearg *ra, objset_t *os,
916 struct drr_freeobjects *drrfo)
920 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
923 for (obj = drrfo->drr_firstobj;
924 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
925 (void) dmu_object_next(os, &obj, FALSE, 0)) {
928 if (dmu_object_info(os, obj, NULL) != 0)
931 err = dmu_free_object(os, obj);
939 restore_write(struct restorearg *ra, objset_t *os,
940 struct drr_write *drrw)
946 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
947 drrw->drr_type >= DMU_OT_NUMTYPES)
950 data = restore_read(ra, drrw->drr_length);
954 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
957 tx = dmu_tx_create(os);
959 dmu_tx_hold_write(tx, drrw->drr_object,
960 drrw->drr_offset, drrw->drr_length);
961 err = dmu_tx_assign(tx, TXG_WAIT);
967 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
968 dmu_write(os, drrw->drr_object,
969 drrw->drr_offset, drrw->drr_length, data, tx);
976 restore_free(struct restorearg *ra, objset_t *os,
977 struct drr_free *drrf)
981 if (drrf->drr_length != -1ULL &&
982 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
985 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
988 err = dmu_free_long_range(os, drrf->drr_object,
989 drrf->drr_offset, drrf->drr_length);
994 dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
996 if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
998 * online incremental or new fs: destroy the fs (which
999 * may be a clone) that we created
1001 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
1002 if (drc->drc_real_ds != drc->drc_logical_ds)
1003 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
1006 * offline incremental: rollback to most recent snapshot.
1008 (void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE);
1009 dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag);
1014 * NB: callers *must* call dmu_recv_end() if this succeeds.
1017 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp)
1019 kthread_t *td = curthread;
1020 struct restorearg ra = { 0 };
1021 dmu_replay_record_t *drr;
1025 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1029 /* compute checksum of drr_begin record */
1030 dmu_replay_record_t *drr;
1031 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1033 drr->drr_type = DRR_BEGIN;
1034 drr->drr_u.drr_begin = *drc->drc_drrb;
1036 fletcher_4_incremental_byteswap(drr,
1037 sizeof (dmu_replay_record_t), &ra.cksum);
1039 fletcher_4_incremental_native(drr,
1040 sizeof (dmu_replay_record_t), &ra.cksum);
1042 kmem_free(drr, sizeof (dmu_replay_record_t));
1046 struct drr_begin *drrb = drc->drc_drrb;
1047 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1048 drrb->drr_version = BSWAP_64(drrb->drr_version);
1049 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1050 drrb->drr_type = BSWAP_32(drrb->drr_type);
1051 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1052 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1059 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1061 /* these were verified in dmu_recv_begin */
1062 ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION);
1063 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1066 * Open the objset we are modifying.
1068 VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0);
1070 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1073 * Read records and process them.
1076 while (ra.err == 0 &&
1077 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1078 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1084 backup_byteswap(drr);
1086 switch (drr->drr_type) {
1090 * We need to make a copy of the record header,
1091 * because restore_{object,write} may need to
1092 * restore_read(), which will invalidate drr.
1094 struct drr_object drro = drr->drr_u.drr_object;
1095 ra.err = restore_object(&ra, os, &drro);
1098 case DRR_FREEOBJECTS:
1100 struct drr_freeobjects drrfo =
1101 drr->drr_u.drr_freeobjects;
1102 ra.err = restore_freeobjects(&ra, os, &drrfo);
1107 struct drr_write drrw = drr->drr_u.drr_write;
1108 ra.err = restore_write(&ra, os, &drrw);
1113 struct drr_free drrf = drr->drr_u.drr_free;
1114 ra.err = restore_free(&ra, os, &drrf);
1119 struct drr_end drre = drr->drr_u.drr_end;
1121 * We compare against the *previous* checksum
1122 * value, because the stored checksum is of
1123 * everything before the DRR_END record.
1125 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1135 ASSERT(ra.err != 0);
1138 dmu_objset_close(os);
1142 * rollback or destroy what we created, so we don't
1143 * leave it in the restoring state.
1145 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1146 dmu_recv_abort_cleanup(drc);
1149 kmem_free(ra.buf, ra.bufsize);
1154 struct recvendsyncarg {
1156 uint64_t creation_time;
1161 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1163 dsl_dataset_t *ds = arg1;
1164 struct recvendsyncarg *resa = arg2;
1166 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1170 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1172 dsl_dataset_t *ds = arg1;
1173 struct recvendsyncarg *resa = arg2;
1175 dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx);
1177 /* set snapshot's creation time and guid */
1178 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1179 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1180 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1181 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1183 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1184 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1188 dmu_recv_end(dmu_recv_cookie_t *drc)
1190 struct recvendsyncarg resa;
1191 dsl_dataset_t *ds = drc->drc_logical_ds;
1195 * XXX hack; seems the ds is still dirty and
1196 * dsl_pool_zil_clean() expects it to have a ds_user_ptr
1197 * (and zil), but clone_swap() can close it.
1199 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1201 if (ds != drc->drc_real_ds) {
1202 /* we are doing an online recv */
1203 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1204 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1207 dsl_dataset_disown(ds, dmu_recv_tag);
1210 dsl_dataset_rele(ds, dmu_recv_tag);
1212 /* dsl_dataset_destroy() will disown the ds */
1213 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
1218 resa.creation_time = drc->drc_drrb->drr_creation_time;
1219 resa.toguid = drc->drc_drrb->drr_toguid;
1220 resa.tosnap = drc->drc_tosnap;
1222 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1223 recv_end_check, recv_end_sync, ds, &resa, 3);
1225 if (drc->drc_newfs) {
1226 ASSERT(ds == drc->drc_real_ds);
1227 (void) dsl_dataset_destroy(ds, dmu_recv_tag);
1230 (void) dsl_dataset_rollback(ds, DMU_OST_NONE);
1234 /* release the hold from dmu_recv_begin */
1235 dsl_dataset_disown(ds, dmu_recv_tag);