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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
26 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
27 * Copyright (c) 2011 by Delphix. All rights reserved.
28 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
29 * Copyright (c) 2012, Martin Matuska <mm@FreeBSD.org>. All rights reserved.
33 #include <sys/dmu_impl.h>
34 #include <sys/dmu_tx.h>
36 #include <sys/dnode.h>
37 #include <sys/zfs_context.h>
38 #include <sys/dmu_objset.h>
39 #include <sys/dmu_traverse.h>
40 #include <sys/dsl_dataset.h>
41 #include <sys/dsl_dir.h>
42 #include <sys/dsl_prop.h>
43 #include <sys/dsl_pool.h>
44 #include <sys/dsl_synctask.h>
45 #include <sys/zfs_ioctl.h>
47 #include <sys/zio_checksum.h>
48 #include <sys/zfs_znode.h>
49 #include <zfs_fletcher.h>
52 #include <sys/zfs_onexit.h>
54 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
55 int zfs_send_corrupt_data = B_FALSE;
57 static char *dmu_recv_tag = "dmu_recv_tag";
60 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
62 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
65 ASSERT3U(len % 8, ==, 0);
67 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
73 auio.uio_segflg = UIO_SYSSPACE;
74 auio.uio_rw = UIO_WRITE;
75 auio.uio_offset = (off_t)-1;
76 auio.uio_td = dsp->dsa_td;
78 if (dsp->dsa_fp->f_type == DTYPE_VNODE)
80 dsp->dsa_err = fo_write(dsp->dsa_fp, &auio, dsp->dsa_td->td_ucred, 0,
83 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
84 dsp->dsa_err = EOPNOTSUPP;
86 mutex_enter(&ds->ds_sendstream_lock);
88 mutex_exit(&ds->ds_sendstream_lock);
90 return (dsp->dsa_err);
94 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
97 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
99 if (length != -1ULL && offset + length < offset)
103 * If there is a pending op, but it's not PENDING_FREE, push it out,
104 * since free block aggregation can only be done for blocks of the
105 * same type (i.e., DRR_FREE records can only be aggregated with
106 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
107 * aggregated with other DRR_FREEOBJECTS records.
109 if (dsp->dsa_pending_op != PENDING_NONE &&
110 dsp->dsa_pending_op != PENDING_FREE) {
111 if (dump_bytes(dsp, dsp->dsa_drr,
112 sizeof (dmu_replay_record_t)) != 0)
114 dsp->dsa_pending_op = PENDING_NONE;
117 if (dsp->dsa_pending_op == PENDING_FREE) {
119 * There should never be a PENDING_FREE if length is -1
120 * (because dump_dnode is the only place where this
121 * function is called with a -1, and only after flushing
122 * any pending record).
124 ASSERT(length != -1ULL);
126 * Check to see whether this free block can be aggregated
129 if (drrf->drr_object == object && drrf->drr_offset +
130 drrf->drr_length == offset) {
131 drrf->drr_length += length;
134 /* not a continuation. Push out pending record */
135 if (dump_bytes(dsp, dsp->dsa_drr,
136 sizeof (dmu_replay_record_t)) != 0)
138 dsp->dsa_pending_op = PENDING_NONE;
141 /* create a FREE record and make it pending */
142 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
143 dsp->dsa_drr->drr_type = DRR_FREE;
144 drrf->drr_object = object;
145 drrf->drr_offset = offset;
146 drrf->drr_length = length;
147 drrf->drr_toguid = dsp->dsa_toguid;
148 if (length == -1ULL) {
149 if (dump_bytes(dsp, dsp->dsa_drr,
150 sizeof (dmu_replay_record_t)) != 0)
153 dsp->dsa_pending_op = PENDING_FREE;
160 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
161 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
163 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
167 * If there is any kind of pending aggregation (currently either
168 * a grouping of free objects or free blocks), push it out to
169 * the stream, since aggregation can't be done across operations
170 * of different types.
172 if (dsp->dsa_pending_op != PENDING_NONE) {
173 if (dump_bytes(dsp, dsp->dsa_drr,
174 sizeof (dmu_replay_record_t)) != 0)
176 dsp->dsa_pending_op = PENDING_NONE;
178 /* write a DATA record */
179 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
180 dsp->dsa_drr->drr_type = DRR_WRITE;
181 drrw->drr_object = object;
182 drrw->drr_type = type;
183 drrw->drr_offset = offset;
184 drrw->drr_length = blksz;
185 drrw->drr_toguid = dsp->dsa_toguid;
186 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
187 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
188 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
189 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
190 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
191 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
192 drrw->drr_key.ddk_cksum = bp->blk_cksum;
194 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
196 if (dump_bytes(dsp, data, blksz) != 0)
202 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
204 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
206 if (dsp->dsa_pending_op != PENDING_NONE) {
207 if (dump_bytes(dsp, dsp->dsa_drr,
208 sizeof (dmu_replay_record_t)) != 0)
210 dsp->dsa_pending_op = PENDING_NONE;
213 /* write a SPILL record */
214 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
215 dsp->dsa_drr->drr_type = DRR_SPILL;
216 drrs->drr_object = object;
217 drrs->drr_length = blksz;
218 drrs->drr_toguid = dsp->dsa_toguid;
220 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
222 if (dump_bytes(dsp, data, blksz))
228 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
230 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
233 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
234 * push it out, since free block aggregation can only be done for
235 * blocks of the same type (i.e., DRR_FREE records can only be
236 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
237 * can only be aggregated with other DRR_FREEOBJECTS records.
239 if (dsp->dsa_pending_op != PENDING_NONE &&
240 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
241 if (dump_bytes(dsp, dsp->dsa_drr,
242 sizeof (dmu_replay_record_t)) != 0)
244 dsp->dsa_pending_op = PENDING_NONE;
246 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
248 * See whether this free object array can be aggregated
251 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
252 drrfo->drr_numobjs += numobjs;
255 /* can't be aggregated. Push out pending record */
256 if (dump_bytes(dsp, dsp->dsa_drr,
257 sizeof (dmu_replay_record_t)) != 0)
259 dsp->dsa_pending_op = PENDING_NONE;
263 /* write a FREEOBJECTS record */
264 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
265 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
266 drrfo->drr_firstobj = firstobj;
267 drrfo->drr_numobjs = numobjs;
268 drrfo->drr_toguid = dsp->dsa_toguid;
270 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
276 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
278 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
280 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
281 return (dump_freeobjects(dsp, object, 1));
283 if (dsp->dsa_pending_op != PENDING_NONE) {
284 if (dump_bytes(dsp, dsp->dsa_drr,
285 sizeof (dmu_replay_record_t)) != 0)
287 dsp->dsa_pending_op = PENDING_NONE;
290 /* write an OBJECT record */
291 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
292 dsp->dsa_drr->drr_type = DRR_OBJECT;
293 drro->drr_object = object;
294 drro->drr_type = dnp->dn_type;
295 drro->drr_bonustype = dnp->dn_bonustype;
296 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
297 drro->drr_bonuslen = dnp->dn_bonuslen;
298 drro->drr_checksumtype = dnp->dn_checksum;
299 drro->drr_compress = dnp->dn_compress;
300 drro->drr_toguid = dsp->dsa_toguid;
302 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
305 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
308 /* free anything past the end of the file */
309 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
310 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
317 #define BP_SPAN(dnp, level) \
318 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
319 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
323 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
324 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
326 dmu_sendarg_t *dsp = arg;
327 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
330 if (issig(JUSTLOOKING) && issig(FORREAL))
333 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
334 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
336 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
337 uint64_t span = BP_SPAN(dnp, zb->zb_level);
338 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
339 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
340 } else if (bp == NULL) {
341 uint64_t span = BP_SPAN(dnp, zb->zb_level);
342 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
343 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
345 } else if (type == DMU_OT_DNODE) {
348 int blksz = BP_GET_LSIZE(bp);
349 uint32_t aflags = ARC_WAIT;
352 if (dsl_read(NULL, spa, bp, pbuf,
353 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
354 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
358 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
359 uint64_t dnobj = (zb->zb_blkid <<
360 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
361 err = dump_dnode(dsp, dnobj, blk+i);
365 (void) arc_buf_remove_ref(abuf, &abuf);
366 } else if (type == DMU_OT_SA) {
367 uint32_t aflags = ARC_WAIT;
369 int blksz = BP_GET_LSIZE(bp);
371 if (arc_read_nolock(NULL, spa, bp,
372 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
373 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
376 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
377 (void) arc_buf_remove_ref(abuf, &abuf);
378 } else { /* it's a level-0 block of a regular object */
379 uint32_t aflags = ARC_WAIT;
381 int blksz = BP_GET_LSIZE(bp);
383 if (dsl_read(NULL, spa, bp, pbuf,
384 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
385 ZIO_FLAG_CANFAIL, &aflags, zb) != 0) {
386 if (zfs_send_corrupt_data) {
387 /* Send a block filled with 0x"zfs badd bloc" */
388 abuf = arc_buf_alloc(spa, blksz, &abuf,
391 for (ptr = abuf->b_data;
392 (char *)ptr < (char *)abuf->b_data + blksz;
394 *ptr = 0x2f5baddb10c;
400 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
401 blksz, bp, abuf->b_data);
402 (void) arc_buf_remove_ref(abuf, &abuf);
405 ASSERT(err == 0 || err == EINTR);
410 dmu_send(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
411 int outfd, struct file *fp, offset_t *off)
413 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
414 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
415 dmu_replay_record_t *drr;
418 uint64_t fromtxg = 0;
420 /* tosnap must be a snapshot */
421 if (ds->ds_phys->ds_next_snap_obj == 0)
424 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
425 if (fromds && (ds->ds_dir != fromds->ds_dir ||
426 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
430 dsl_pool_t *dp = ds->ds_dir->dd_pool;
435 if (dsl_dir_is_clone(ds->ds_dir)) {
436 rw_enter(&dp->dp_config_rwlock, RW_READER);
437 err = dsl_dataset_hold_obj(dp,
438 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
439 rw_exit(&dp->dp_config_rwlock);
443 fromorigin = B_FALSE;
448 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
449 drr->drr_type = DRR_BEGIN;
450 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
451 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
455 if (dmu_objset_type(tosnap) == DMU_OST_ZFS) {
457 if (zfs_get_zplprop(tosnap, ZFS_PROP_VERSION, &version) != 0) {
458 kmem_free(drr, sizeof (dmu_replay_record_t));
461 if (version == ZPL_VERSION_SA) {
462 DMU_SET_FEATUREFLAGS(
463 drr->drr_u.drr_begin.drr_versioninfo,
464 DMU_BACKUP_FEATURE_SA_SPILL);
469 drr->drr_u.drr_begin.drr_creation_time =
470 ds->ds_phys->ds_creation_time;
471 drr->drr_u.drr_begin.drr_type = tosnap->os_phys->os_type;
473 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
474 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
475 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
476 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
479 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
480 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
483 fromtxg = fromds->ds_phys->ds_creation_txg;
485 dsl_dataset_rele(fromds, FTAG);
487 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
490 dsp->dsa_outfd = outfd;
491 dsp->dsa_proc = curproc;
492 dsp->dsa_td = curthread;
494 dsp->dsa_os = tosnap;
496 dsp->dsa_toguid = ds->ds_phys->ds_guid;
497 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
498 dsp->dsa_pending_op = PENDING_NONE;
500 mutex_enter(&ds->ds_sendstream_lock);
501 list_insert_head(&ds->ds_sendstreams, dsp);
502 mutex_exit(&ds->ds_sendstream_lock);
504 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
509 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
512 if (dsp->dsa_pending_op != PENDING_NONE)
513 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
517 if (err == EINTR && dsp->dsa_err)
522 bzero(drr, sizeof (dmu_replay_record_t));
523 drr->drr_type = DRR_END;
524 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
525 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
527 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
533 mutex_enter(&ds->ds_sendstream_lock);
534 list_remove(&ds->ds_sendstreams, dsp);
535 mutex_exit(&ds->ds_sendstream_lock);
537 kmem_free(drr, sizeof (dmu_replay_record_t));
538 kmem_free(dsp, sizeof (dmu_sendarg_t));
544 dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
547 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
548 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
549 dsl_pool_t *dp = ds->ds_dir->dd_pool;
553 /* tosnap must be a snapshot */
554 if (ds->ds_phys->ds_next_snap_obj == 0)
557 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
558 if (fromds && (ds->ds_dir != fromds->ds_dir ||
559 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
566 if (dsl_dir_is_clone(ds->ds_dir)) {
567 rw_enter(&dp->dp_config_rwlock, RW_READER);
568 err = dsl_dataset_hold_obj(dp,
569 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
570 rw_exit(&dp->dp_config_rwlock);
574 fromorigin = B_FALSE;
578 /* Get uncompressed size estimate of changed data. */
579 if (fromds == NULL) {
580 size = ds->ds_phys->ds_uncompressed_bytes;
583 err = dsl_dataset_space_written(fromds, ds,
584 &used, &comp, &size);
586 dsl_dataset_rele(fromds, FTAG);
592 * Assume that space (both on-disk and in-stream) is dominated by
593 * data. We will adjust for indirect blocks and the copies property,
594 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
598 * Subtract out approximate space used by indirect blocks.
599 * Assume most space is used by data blocks (non-indirect, non-dnode).
600 * Assume all blocks are recordsize. Assume ditto blocks and
601 * internal fragmentation counter out compression.
603 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
604 * block, which we observe in practice.
607 rw_enter(&dp->dp_config_rwlock, RW_READER);
608 err = dsl_prop_get_ds(ds, "recordsize",
609 sizeof (recordsize), 1, &recordsize, NULL);
610 rw_exit(&dp->dp_config_rwlock);
613 size -= size / recordsize * sizeof (blkptr_t);
615 /* Add in the space for the record associated with each block. */
616 size += size / recordsize * sizeof (dmu_replay_record_t);
623 struct recvbeginsyncarg {
626 dsl_dataset_t *origin;
628 dmu_objset_type_t type;
632 char clonelastname[MAXNAMELEN];
633 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
639 recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx)
641 dsl_dir_t *dd = arg1;
642 struct recvbeginsyncarg *rbsa = arg2;
643 objset_t *mos = dd->dd_pool->dp_meta_objset;
647 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
648 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
651 return (err ? err : EEXIST);
654 /* make sure it's a snap in the same pool */
655 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
657 if (!dsl_dataset_is_snapshot(rbsa->origin))
659 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
667 recv_new_sync(void *arg1, void *arg2, dmu_tx_t *tx)
669 dsl_dir_t *dd = arg1;
670 struct recvbeginsyncarg *rbsa = arg2;
671 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
674 /* Create and open new dataset. */
675 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
676 rbsa->origin, flags, rbsa->cr, tx);
677 VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj,
678 B_TRUE, dmu_recv_tag, &rbsa->ds));
680 if (rbsa->origin == NULL) {
681 (void) dmu_objset_create_impl(dd->dd_pool->dp_spa,
682 rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx);
685 spa_history_log_internal(LOG_DS_REPLAY_FULL_SYNC,
686 dd->dd_pool->dp_spa, tx, "dataset = %lld", dsobj);
691 recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
693 dsl_dataset_t *ds = arg1;
694 struct recvbeginsyncarg *rbsa = arg2;
698 /* must not have any changes since most recent snapshot */
699 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
702 /* new snapshot name must not exist */
703 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
704 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
710 if (rbsa->fromguid) {
711 /* if incremental, most recent snapshot must match fromguid */
712 if (ds->ds_prev == NULL)
716 * most recent snapshot must match fromguid, or there are no
717 * changes since the fromguid one
719 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) {
720 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth;
721 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj;
724 err = dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
728 if (snap->ds_phys->ds_creation_txg < birth) {
729 dsl_dataset_rele(snap, FTAG);
732 if (snap->ds_phys->ds_guid == rbsa->fromguid) {
733 dsl_dataset_rele(snap, FTAG);
736 obj = snap->ds_phys->ds_prev_snap_obj;
737 dsl_dataset_rele(snap, FTAG);
743 /* if full, most recent snapshot must be $ORIGIN */
744 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
748 /* temporary clone name must not exist */
749 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
750 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
751 rbsa->clonelastname, 8, 1, &val);
762 recv_existing_sync(void *arg1, void *arg2, dmu_tx_t *tx)
764 dsl_dataset_t *ohds = arg1;
765 struct recvbeginsyncarg *rbsa = arg2;
766 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
768 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
771 /* create and open the temporary clone */
772 dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname,
773 ohds->ds_prev, flags, rbsa->cr, tx);
774 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, B_TRUE, dmu_recv_tag, &cds));
777 * If we actually created a non-clone, we need to create the
778 * objset in our new dataset.
780 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) {
781 (void) dmu_objset_create_impl(dp->dp_spa,
782 cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx);
787 spa_history_log_internal(LOG_DS_REPLAY_INC_SYNC,
788 dp->dp_spa, tx, "dataset = %lld", dsobj);
792 dmu_recv_verify_features(dsl_dataset_t *ds, struct drr_begin *drrb)
796 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
798 /* Verify pool version supports SA if SA_SPILL feature set */
799 return ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
800 (spa_version(dsl_dataset_get_spa(ds)) < SPA_VERSION_SA));
804 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
805 * succeeds; otherwise we will leak the holds on the datasets.
808 dmu_recv_begin(char *tofs, char *tosnap, char *top_ds, struct drr_begin *drrb,
809 boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
813 struct recvbeginsyncarg rbsa = { 0 };
814 uint64_t versioninfo;
818 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
820 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
826 rbsa.tosnap = tosnap;
827 rbsa.origin = origin ? origin->os_dsl_dataset : NULL;
828 rbsa.fromguid = drrb->drr_fromguid;
829 rbsa.type = drrb->drr_type;
833 versioninfo = drrb->drr_versioninfo;
834 flags = drrb->drr_flags;
837 rbsa.type = BSWAP_32(rbsa.type);
838 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
839 versioninfo = BSWAP_64(versioninfo);
840 flags = BSWAP_32(flags);
843 if (DMU_GET_STREAM_HDRTYPE(versioninfo) == DMU_COMPOUNDSTREAM ||
844 rbsa.type >= DMU_OST_NUMTYPES ||
845 ((flags & DRR_FLAG_CLONE) && origin == NULL))
848 if (flags & DRR_FLAG_CI_DATA)
849 rbsa.dsflags = DS_FLAG_CI_DATASET;
851 bzero(drc, sizeof (dmu_recv_cookie_t));
852 drc->drc_drrb = drrb;
853 drc->drc_tosnap = tosnap;
854 drc->drc_top_ds = top_ds;
855 drc->drc_force = force;
858 * Process the begin in syncing context.
861 /* open the dataset we are logically receiving into */
862 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
864 if (dmu_recv_verify_features(ds, drrb)) {
865 dsl_dataset_rele(ds, dmu_recv_tag);
868 /* target fs already exists; recv into temp clone */
870 /* Can't recv a clone into an existing fs */
871 if (flags & DRR_FLAG_CLONE) {
872 dsl_dataset_rele(ds, dmu_recv_tag);
876 /* must not have an incremental recv already in progress */
877 if (!mutex_tryenter(&ds->ds_recvlock)) {
878 dsl_dataset_rele(ds, dmu_recv_tag);
882 /* tmp clone name is: tofs/%tosnap" */
883 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
886 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
887 recv_existing_check, recv_existing_sync, ds, &rbsa, 5);
889 mutex_exit(&ds->ds_recvlock);
890 dsl_dataset_rele(ds, dmu_recv_tag);
893 drc->drc_logical_ds = ds;
894 drc->drc_real_ds = rbsa.ds;
895 } else if (err == ENOENT) {
896 /* target fs does not exist; must be a full backup or clone */
900 * If it's a non-clone incremental, we are missing the
901 * target fs, so fail the recv.
903 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE))
906 /* Open the parent of tofs */
907 cp = strrchr(tofs, '/');
909 err = dsl_dataset_hold(tofs, FTAG, &ds);
914 if (dmu_recv_verify_features(ds, drrb)) {
915 dsl_dataset_rele(ds, FTAG);
919 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
920 recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5);
921 dsl_dataset_rele(ds, FTAG);
924 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
925 drc->drc_newfs = B_TRUE;
938 int bufsize; /* amount of memory allocated for buf */
940 avl_tree_t *guid_to_ds_map;
943 typedef struct guid_map_entry {
945 dsl_dataset_t *gme_ds;
950 guid_compare(const void *arg1, const void *arg2)
952 const guid_map_entry_t *gmep1 = arg1;
953 const guid_map_entry_t *gmep2 = arg2;
955 if (gmep1->guid < gmep2->guid)
957 else if (gmep1->guid > gmep2->guid)
963 free_guid_map_onexit(void *arg)
965 avl_tree_t *ca = arg;
967 guid_map_entry_t *gmep;
969 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
970 dsl_dataset_rele(gmep->gme_ds, ca);
971 kmem_free(gmep, sizeof (guid_map_entry_t));
974 kmem_free(ca, sizeof (avl_tree_t));
978 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
986 auio.uio_iov = &aiov;
988 auio.uio_resid = len;
989 auio.uio_segflg = UIO_SYSSPACE;
990 auio.uio_rw = UIO_READ;
991 auio.uio_offset = off;
992 auio.uio_td = ra->td;
994 error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
996 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
999 *resid = auio.uio_resid;
1004 restore_read(struct restorearg *ra, int len)
1009 /* some things will require 8-byte alignment, so everything must */
1010 ASSERT3U(len % 8, ==, 0);
1012 while (done < len) {
1015 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
1016 len - done, ra->voff, &resid);
1018 if (resid == len - done)
1020 ra->voff += len - done - resid;
1026 ASSERT3U(done, ==, len);
1029 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1031 fletcher_4_incremental_native(rv, len, &ra->cksum);
1036 backup_byteswap(dmu_replay_record_t *drr)
1038 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1039 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1040 drr->drr_type = BSWAP_32(drr->drr_type);
1041 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1042 switch (drr->drr_type) {
1044 DO64(drr_begin.drr_magic);
1045 DO64(drr_begin.drr_versioninfo);
1046 DO64(drr_begin.drr_creation_time);
1047 DO32(drr_begin.drr_type);
1048 DO32(drr_begin.drr_flags);
1049 DO64(drr_begin.drr_toguid);
1050 DO64(drr_begin.drr_fromguid);
1053 DO64(drr_object.drr_object);
1054 /* DO64(drr_object.drr_allocation_txg); */
1055 DO32(drr_object.drr_type);
1056 DO32(drr_object.drr_bonustype);
1057 DO32(drr_object.drr_blksz);
1058 DO32(drr_object.drr_bonuslen);
1059 DO64(drr_object.drr_toguid);
1061 case DRR_FREEOBJECTS:
1062 DO64(drr_freeobjects.drr_firstobj);
1063 DO64(drr_freeobjects.drr_numobjs);
1064 DO64(drr_freeobjects.drr_toguid);
1067 DO64(drr_write.drr_object);
1068 DO32(drr_write.drr_type);
1069 DO64(drr_write.drr_offset);
1070 DO64(drr_write.drr_length);
1071 DO64(drr_write.drr_toguid);
1072 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1073 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1074 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1075 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1076 DO64(drr_write.drr_key.ddk_prop);
1078 case DRR_WRITE_BYREF:
1079 DO64(drr_write_byref.drr_object);
1080 DO64(drr_write_byref.drr_offset);
1081 DO64(drr_write_byref.drr_length);
1082 DO64(drr_write_byref.drr_toguid);
1083 DO64(drr_write_byref.drr_refguid);
1084 DO64(drr_write_byref.drr_refobject);
1085 DO64(drr_write_byref.drr_refoffset);
1086 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1087 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1088 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1089 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1090 DO64(drr_write_byref.drr_key.ddk_prop);
1093 DO64(drr_free.drr_object);
1094 DO64(drr_free.drr_offset);
1095 DO64(drr_free.drr_length);
1096 DO64(drr_free.drr_toguid);
1099 DO64(drr_spill.drr_object);
1100 DO64(drr_spill.drr_length);
1101 DO64(drr_spill.drr_toguid);
1104 DO64(drr_end.drr_checksum.zc_word[0]);
1105 DO64(drr_end.drr_checksum.zc_word[1]);
1106 DO64(drr_end.drr_checksum.zc_word[2]);
1107 DO64(drr_end.drr_checksum.zc_word[3]);
1108 DO64(drr_end.drr_toguid);
1116 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1122 if (drro->drr_type == DMU_OT_NONE ||
1123 drro->drr_type >= DMU_OT_NUMTYPES ||
1124 drro->drr_bonustype >= DMU_OT_NUMTYPES ||
1125 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1126 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1127 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1128 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1129 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1130 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1134 err = dmu_object_info(os, drro->drr_object, NULL);
1136 if (err != 0 && err != ENOENT)
1139 if (drro->drr_bonuslen) {
1140 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1145 if (err == ENOENT) {
1146 /* currently free, want to be allocated */
1147 tx = dmu_tx_create(os);
1148 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1149 err = dmu_tx_assign(tx, TXG_WAIT);
1154 err = dmu_object_claim(os, drro->drr_object,
1155 drro->drr_type, drro->drr_blksz,
1156 drro->drr_bonustype, drro->drr_bonuslen, tx);
1159 /* currently allocated, want to be allocated */
1160 err = dmu_object_reclaim(os, drro->drr_object,
1161 drro->drr_type, drro->drr_blksz,
1162 drro->drr_bonustype, drro->drr_bonuslen);
1168 tx = dmu_tx_create(os);
1169 dmu_tx_hold_bonus(tx, drro->drr_object);
1170 err = dmu_tx_assign(tx, TXG_WAIT);
1176 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1178 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1183 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1184 dmu_buf_will_dirty(db, tx);
1186 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1187 bcopy(data, db->db_data, drro->drr_bonuslen);
1189 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
1190 drro->drr_bonuslen);
1192 dmu_buf_rele(db, FTAG);
1200 restore_freeobjects(struct restorearg *ra, objset_t *os,
1201 struct drr_freeobjects *drrfo)
1205 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1208 for (obj = drrfo->drr_firstobj;
1209 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1210 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1213 if (dmu_object_info(os, obj, NULL) != 0)
1216 err = dmu_free_object(os, obj);
1224 restore_write(struct restorearg *ra, objset_t *os,
1225 struct drr_write *drrw)
1231 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1232 drrw->drr_type >= DMU_OT_NUMTYPES)
1235 data = restore_read(ra, drrw->drr_length);
1239 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1242 tx = dmu_tx_create(os);
1244 dmu_tx_hold_write(tx, drrw->drr_object,
1245 drrw->drr_offset, drrw->drr_length);
1246 err = dmu_tx_assign(tx, TXG_WAIT);
1252 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
1253 dmu_write(os, drrw->drr_object,
1254 drrw->drr_offset, drrw->drr_length, data, tx);
1260 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1261 * streams to refer to a copy of the data that is already on the
1262 * system because it came in earlier in the stream. This function
1263 * finds the earlier copy of the data, and uses that copy instead of
1264 * data from the stream to fulfill this write.
1267 restore_write_byref(struct restorearg *ra, objset_t *os,
1268 struct drr_write_byref *drrwbr)
1272 guid_map_entry_t gmesrch;
1273 guid_map_entry_t *gmep;
1275 objset_t *ref_os = NULL;
1278 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1282 * If the GUID of the referenced dataset is different from the
1283 * GUID of the target dataset, find the referenced dataset.
1285 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1286 gmesrch.guid = drrwbr->drr_refguid;
1287 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1291 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1297 if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1298 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1301 tx = dmu_tx_create(os);
1303 dmu_tx_hold_write(tx, drrwbr->drr_object,
1304 drrwbr->drr_offset, drrwbr->drr_length);
1305 err = dmu_tx_assign(tx, TXG_WAIT);
1310 dmu_write(os, drrwbr->drr_object,
1311 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1312 dmu_buf_rele(dbp, FTAG);
1318 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1322 dmu_buf_t *db, *db_spill;
1325 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1326 drrs->drr_length > SPA_MAXBLOCKSIZE)
1329 data = restore_read(ra, drrs->drr_length);
1333 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1336 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1337 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1338 dmu_buf_rele(db, FTAG);
1342 tx = dmu_tx_create(os);
1344 dmu_tx_hold_spill(tx, db->db_object);
1346 err = dmu_tx_assign(tx, TXG_WAIT);
1348 dmu_buf_rele(db, FTAG);
1349 dmu_buf_rele(db_spill, FTAG);
1353 dmu_buf_will_dirty(db_spill, tx);
1355 if (db_spill->db_size < drrs->drr_length)
1356 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1357 drrs->drr_length, tx));
1358 bcopy(data, db_spill->db_data, drrs->drr_length);
1360 dmu_buf_rele(db, FTAG);
1361 dmu_buf_rele(db_spill, FTAG);
1369 restore_free(struct restorearg *ra, objset_t *os,
1370 struct drr_free *drrf)
1374 if (drrf->drr_length != -1ULL &&
1375 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1378 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1381 err = dmu_free_long_range(os, drrf->drr_object,
1382 drrf->drr_offset, drrf->drr_length);
1387 * NB: callers *must* call dmu_recv_end() if this succeeds.
1390 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1391 int cleanup_fd, uint64_t *action_handlep)
1393 struct restorearg ra = { 0 };
1394 dmu_replay_record_t *drr;
1399 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1403 /* compute checksum of drr_begin record */
1404 dmu_replay_record_t *drr;
1405 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1407 drr->drr_type = DRR_BEGIN;
1408 drr->drr_u.drr_begin = *drc->drc_drrb;
1410 fletcher_4_incremental_byteswap(drr,
1411 sizeof (dmu_replay_record_t), &ra.cksum);
1413 fletcher_4_incremental_native(drr,
1414 sizeof (dmu_replay_record_t), &ra.cksum);
1416 kmem_free(drr, sizeof (dmu_replay_record_t));
1420 struct drr_begin *drrb = drc->drc_drrb;
1421 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1422 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1423 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1424 drrb->drr_type = BSWAP_32(drrb->drr_type);
1425 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1426 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1433 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1435 /* these were verified in dmu_recv_begin */
1436 ASSERT(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo) ==
1438 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1441 * Open the objset we are modifying.
1443 VERIFY(dmu_objset_from_ds(drc->drc_real_ds, &os) == 0);
1445 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1447 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1449 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1450 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1453 if (cleanup_fd == -1) {
1457 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1463 if (*action_handlep == 0) {
1465 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1466 avl_create(ra.guid_to_ds_map, guid_compare,
1467 sizeof (guid_map_entry_t),
1468 offsetof(guid_map_entry_t, avlnode));
1469 ra.err = zfs_onexit_add_cb(minor,
1470 free_guid_map_onexit, ra.guid_to_ds_map,
1475 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1476 (void **)&ra.guid_to_ds_map);
1481 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1485 * Read records and process them.
1488 while (ra.err == 0 &&
1489 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1490 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1496 backup_byteswap(drr);
1498 switch (drr->drr_type) {
1502 * We need to make a copy of the record header,
1503 * because restore_{object,write} may need to
1504 * restore_read(), which will invalidate drr.
1506 struct drr_object drro = drr->drr_u.drr_object;
1507 ra.err = restore_object(&ra, os, &drro);
1510 case DRR_FREEOBJECTS:
1512 struct drr_freeobjects drrfo =
1513 drr->drr_u.drr_freeobjects;
1514 ra.err = restore_freeobjects(&ra, os, &drrfo);
1519 struct drr_write drrw = drr->drr_u.drr_write;
1520 ra.err = restore_write(&ra, os, &drrw);
1523 case DRR_WRITE_BYREF:
1525 struct drr_write_byref drrwbr =
1526 drr->drr_u.drr_write_byref;
1527 ra.err = restore_write_byref(&ra, os, &drrwbr);
1532 struct drr_free drrf = drr->drr_u.drr_free;
1533 ra.err = restore_free(&ra, os, &drrf);
1538 struct drr_end drre = drr->drr_u.drr_end;
1540 * We compare against the *previous* checksum
1541 * value, because the stored checksum is of
1542 * everything before the DRR_END record.
1544 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1550 struct drr_spill drrs = drr->drr_u.drr_spill;
1551 ra.err = restore_spill(&ra, os, &drrs);
1560 ASSERT(ra.err != 0);
1563 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1564 zfs_onexit_fd_rele(cleanup_fd);
1568 * destroy what we created, so we don't leave it in the
1569 * inconsistent restoring state.
1571 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1573 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1575 if (drc->drc_real_ds != drc->drc_logical_ds) {
1576 mutex_exit(&drc->drc_logical_ds->ds_recvlock);
1577 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
1581 kmem_free(ra.buf, ra.bufsize);
1586 struct recvendsyncarg {
1588 uint64_t creation_time;
1593 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1595 dsl_dataset_t *ds = arg1;
1596 struct recvendsyncarg *resa = arg2;
1598 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1602 recv_end_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1604 dsl_dataset_t *ds = arg1;
1605 struct recvendsyncarg *resa = arg2;
1607 dsl_dataset_snapshot_sync(ds, resa->tosnap, tx);
1609 /* set snapshot's creation time and guid */
1610 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1611 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1612 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1613 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1615 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1616 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1620 add_ds_to_guidmap(avl_tree_t *guid_map, dsl_dataset_t *ds)
1622 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1623 uint64_t snapobj = ds->ds_phys->ds_prev_snap_obj;
1624 dsl_dataset_t *snapds;
1625 guid_map_entry_t *gmep;
1628 ASSERT(guid_map != NULL);
1630 rw_enter(&dp->dp_config_rwlock, RW_READER);
1631 err = dsl_dataset_hold_obj(dp, snapobj, guid_map, &snapds);
1633 gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP);
1634 gmep->guid = snapds->ds_phys->ds_guid;
1635 gmep->gme_ds = snapds;
1636 avl_add(guid_map, gmep);
1639 rw_exit(&dp->dp_config_rwlock);
1644 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1646 struct recvendsyncarg resa;
1647 dsl_dataset_t *ds = drc->drc_logical_ds;
1651 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1652 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1655 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1657 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1658 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1663 mutex_exit(&ds->ds_recvlock);
1664 dsl_dataset_rele(ds, dmu_recv_tag);
1665 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1670 resa.creation_time = drc->drc_drrb->drr_creation_time;
1671 resa.toguid = drc->drc_drrb->drr_toguid;
1672 resa.tosnap = drc->drc_tosnap;
1674 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1675 recv_end_check, recv_end_sync, ds, &resa, 3);
1678 (void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE);
1682 mutex_exit(&ds->ds_recvlock);
1683 if (err == 0 && drc->drc_guid_to_ds_map != NULL)
1684 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1685 dsl_dataset_disown(ds, dmu_recv_tag);
1686 myerr = dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE);
1687 ASSERT3U(myerr, ==, 0);
1692 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1694 struct recvendsyncarg resa;
1695 dsl_dataset_t *ds = drc->drc_logical_ds;
1699 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1700 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1703 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1705 resa.creation_time = drc->drc_drrb->drr_creation_time;
1706 resa.toguid = drc->drc_drrb->drr_toguid;
1707 resa.tosnap = drc->drc_tosnap;
1709 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1710 recv_end_check, recv_end_sync, ds, &resa, 3);
1712 /* clean up the fs we just recv'd into */
1713 (void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE);
1715 if (drc->drc_guid_to_ds_map != NULL)
1716 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1717 /* release the hold from dmu_recv_begin */
1718 dsl_dataset_disown(ds, dmu_recv_tag);
1724 dmu_recv_end(dmu_recv_cookie_t *drc)
1726 if (drc->drc_logical_ds != drc->drc_real_ds)
1727 return (dmu_recv_existing_end(drc));
1729 return (dmu_recv_new_end(drc));