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 2011 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2012, Martin Matuska <mm@FreeBSD.org>. All rights reserved.
30 #include <sys/dmu_impl.h>
31 #include <sys/dmu_tx.h>
33 #include <sys/dnode.h>
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_dir.h>
39 #include <sys/dsl_prop.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_synctask.h>
42 #include <sys/zfs_ioctl.h>
44 #include <sys/zio_checksum.h>
45 #include <sys/zfs_znode.h>
46 #include <zfs_fletcher.h>
49 #include <sys/zfs_onexit.h>
50 #include <sys/dmu_send.h>
51 #include <sys/dsl_destroy.h>
52 #include <sys/dsl_bookmark.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";
58 static const char *recv_clone_name = "%recv";
61 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
63 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
68 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
74 auio.uio_segflg = UIO_SYSSPACE;
75 auio.uio_rw = UIO_WRITE;
76 auio.uio_offset = (off_t)-1;
77 auio.uio_td = dsp->dsa_td;
79 if (dsp->dsa_fp->f_type == DTYPE_VNODE)
81 dsp->dsa_err = fo_write(dsp->dsa_fp, &auio, dsp->dsa_td->td_ucred, 0,
84 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
85 dsp->dsa_err = EOPNOTSUPP;
87 mutex_enter(&ds->ds_sendstream_lock);
89 mutex_exit(&ds->ds_sendstream_lock);
91 return (dsp->dsa_err);
95 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
98 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
101 * When we receive a free record, dbuf_free_range() assumes
102 * that the receiving system doesn't have any dbufs in the range
103 * being freed. This is always true because there is a one-record
104 * constraint: we only send one WRITE record for any given
105 * object+offset. We know that the one-record constraint is
106 * true because we always send data in increasing order by
109 * If the increasing-order constraint ever changes, we should find
110 * another way to assert that the one-record constraint is still
113 ASSERT(object > dsp->dsa_last_data_object ||
114 (object == dsp->dsa_last_data_object &&
115 offset > dsp->dsa_last_data_offset));
118 * If we are doing a non-incremental send, then there can't
119 * be any data in the dataset we're receiving into. Therefore
120 * a free record would simply be a no-op. Save space by not
121 * sending it to begin with.
123 if (!dsp->dsa_incremental)
126 if (length != -1ULL && offset + length < offset)
130 * If there is a pending op, but it's not PENDING_FREE, push it out,
131 * since free block aggregation can only be done for blocks of the
132 * same type (i.e., DRR_FREE records can only be aggregated with
133 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
134 * aggregated with other DRR_FREEOBJECTS records.
136 if (dsp->dsa_pending_op != PENDING_NONE &&
137 dsp->dsa_pending_op != PENDING_FREE) {
138 if (dump_bytes(dsp, dsp->dsa_drr,
139 sizeof (dmu_replay_record_t)) != 0)
140 return (SET_ERROR(EINTR));
141 dsp->dsa_pending_op = PENDING_NONE;
144 if (dsp->dsa_pending_op == PENDING_FREE) {
146 * There should never be a PENDING_FREE if length is -1
147 * (because dump_dnode is the only place where this
148 * function is called with a -1, and only after flushing
149 * any pending record).
151 ASSERT(length != -1ULL);
153 * Check to see whether this free block can be aggregated
156 if (drrf->drr_object == object && drrf->drr_offset +
157 drrf->drr_length == offset) {
158 drrf->drr_length += length;
161 /* not a continuation. Push out pending record */
162 if (dump_bytes(dsp, dsp->dsa_drr,
163 sizeof (dmu_replay_record_t)) != 0)
164 return (SET_ERROR(EINTR));
165 dsp->dsa_pending_op = PENDING_NONE;
168 /* create a FREE record and make it pending */
169 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
170 dsp->dsa_drr->drr_type = DRR_FREE;
171 drrf->drr_object = object;
172 drrf->drr_offset = offset;
173 drrf->drr_length = length;
174 drrf->drr_toguid = dsp->dsa_toguid;
175 if (length == -1ULL) {
176 if (dump_bytes(dsp, dsp->dsa_drr,
177 sizeof (dmu_replay_record_t)) != 0)
178 return (SET_ERROR(EINTR));
180 dsp->dsa_pending_op = PENDING_FREE;
187 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
188 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
190 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
193 * We send data in increasing object, offset order.
194 * See comment in dump_free() for details.
196 ASSERT(object > dsp->dsa_last_data_object ||
197 (object == dsp->dsa_last_data_object &&
198 offset > dsp->dsa_last_data_offset));
199 dsp->dsa_last_data_object = object;
200 dsp->dsa_last_data_offset = offset + blksz - 1;
203 * If there is any kind of pending aggregation (currently either
204 * a grouping of free objects or free blocks), push it out to
205 * the stream, since aggregation can't be done across operations
206 * of different types.
208 if (dsp->dsa_pending_op != PENDING_NONE) {
209 if (dump_bytes(dsp, dsp->dsa_drr,
210 sizeof (dmu_replay_record_t)) != 0)
211 return (SET_ERROR(EINTR));
212 dsp->dsa_pending_op = PENDING_NONE;
214 /* write a DATA record */
215 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
216 dsp->dsa_drr->drr_type = DRR_WRITE;
217 drrw->drr_object = object;
218 drrw->drr_type = type;
219 drrw->drr_offset = offset;
220 drrw->drr_length = blksz;
221 drrw->drr_toguid = dsp->dsa_toguid;
222 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
223 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
224 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
225 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
226 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
227 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
228 drrw->drr_key.ddk_cksum = bp->blk_cksum;
230 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
231 return (SET_ERROR(EINTR));
232 if (dump_bytes(dsp, data, blksz) != 0)
233 return (SET_ERROR(EINTR));
238 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
240 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
242 if (dsp->dsa_pending_op != PENDING_NONE) {
243 if (dump_bytes(dsp, dsp->dsa_drr,
244 sizeof (dmu_replay_record_t)) != 0)
245 return (SET_ERROR(EINTR));
246 dsp->dsa_pending_op = PENDING_NONE;
249 /* write a SPILL record */
250 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
251 dsp->dsa_drr->drr_type = DRR_SPILL;
252 drrs->drr_object = object;
253 drrs->drr_length = blksz;
254 drrs->drr_toguid = dsp->dsa_toguid;
256 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
257 return (SET_ERROR(EINTR));
258 if (dump_bytes(dsp, data, blksz))
259 return (SET_ERROR(EINTR));
264 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
266 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
268 /* See comment in dump_free(). */
269 if (!dsp->dsa_incremental)
273 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
274 * push it out, since free block aggregation can only be done for
275 * blocks of the same type (i.e., DRR_FREE records can only be
276 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
277 * can only be aggregated with other DRR_FREEOBJECTS records.
279 if (dsp->dsa_pending_op != PENDING_NONE &&
280 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
281 if (dump_bytes(dsp, dsp->dsa_drr,
282 sizeof (dmu_replay_record_t)) != 0)
283 return (SET_ERROR(EINTR));
284 dsp->dsa_pending_op = PENDING_NONE;
286 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
288 * See whether this free object array can be aggregated
291 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
292 drrfo->drr_numobjs += numobjs;
295 /* can't be aggregated. Push out pending record */
296 if (dump_bytes(dsp, dsp->dsa_drr,
297 sizeof (dmu_replay_record_t)) != 0)
298 return (SET_ERROR(EINTR));
299 dsp->dsa_pending_op = PENDING_NONE;
303 /* write a FREEOBJECTS record */
304 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
305 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
306 drrfo->drr_firstobj = firstobj;
307 drrfo->drr_numobjs = numobjs;
308 drrfo->drr_toguid = dsp->dsa_toguid;
310 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
316 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
318 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
320 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
321 return (dump_freeobjects(dsp, object, 1));
323 if (dsp->dsa_pending_op != PENDING_NONE) {
324 if (dump_bytes(dsp, dsp->dsa_drr,
325 sizeof (dmu_replay_record_t)) != 0)
326 return (SET_ERROR(EINTR));
327 dsp->dsa_pending_op = PENDING_NONE;
330 /* write an OBJECT record */
331 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
332 dsp->dsa_drr->drr_type = DRR_OBJECT;
333 drro->drr_object = object;
334 drro->drr_type = dnp->dn_type;
335 drro->drr_bonustype = dnp->dn_bonustype;
336 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
337 drro->drr_bonuslen = dnp->dn_bonuslen;
338 drro->drr_checksumtype = dnp->dn_checksum;
339 drro->drr_compress = dnp->dn_compress;
340 drro->drr_toguid = dsp->dsa_toguid;
342 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
343 return (SET_ERROR(EINTR));
345 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
346 return (SET_ERROR(EINTR));
348 /* Free anything past the end of the file. */
349 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
350 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
351 return (SET_ERROR(EINTR));
352 if (dsp->dsa_err != 0)
353 return (SET_ERROR(EINTR));
357 #define BP_SPAN(dnp, level) \
358 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
359 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
363 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
364 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
366 dmu_sendarg_t *dsp = arg;
367 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
370 if (issig(JUSTLOOKING) && issig(FORREAL))
371 return (SET_ERROR(EINTR));
373 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
374 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
376 } else if (zb->zb_level == ZB_ZIL_LEVEL) {
378 * If we are sending a non-snapshot (which is allowed on
379 * read-only pools), it may have a ZIL, which must be ignored.
382 } else if (BP_IS_HOLE(bp) &&
383 zb->zb_object == DMU_META_DNODE_OBJECT) {
384 uint64_t span = BP_SPAN(dnp, zb->zb_level);
385 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
386 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
387 } else if (BP_IS_HOLE(bp)) {
388 uint64_t span = BP_SPAN(dnp, zb->zb_level);
389 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
390 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
392 } else if (type == DMU_OT_DNODE) {
395 int blksz = BP_GET_LSIZE(bp);
396 uint32_t aflags = ARC_WAIT;
399 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
400 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
402 return (SET_ERROR(EIO));
405 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
406 uint64_t dnobj = (zb->zb_blkid <<
407 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
408 err = dump_dnode(dsp, dnobj, blk+i);
412 (void) arc_buf_remove_ref(abuf, &abuf);
413 } else if (type == DMU_OT_SA) {
414 uint32_t aflags = ARC_WAIT;
416 int blksz = BP_GET_LSIZE(bp);
418 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
419 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
421 return (SET_ERROR(EIO));
423 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
424 (void) arc_buf_remove_ref(abuf, &abuf);
425 } else { /* it's a level-0 block of a regular object */
426 uint32_t aflags = ARC_WAIT;
428 int blksz = BP_GET_LSIZE(bp);
430 ASSERT0(zb->zb_level);
431 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
432 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
434 if (zfs_send_corrupt_data) {
435 /* Send a block filled with 0x"zfs badd bloc" */
436 abuf = arc_buf_alloc(spa, blksz, &abuf,
439 for (ptr = abuf->b_data;
440 (char *)ptr < (char *)abuf->b_data + blksz;
442 *ptr = 0x2f5baddb10c;
444 return (SET_ERROR(EIO));
448 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
449 blksz, bp, abuf->b_data);
450 (void) arc_buf_remove_ref(abuf, &abuf);
453 ASSERT(err == 0 || err == EINTR);
458 * Releases dp using the specified tag.
461 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
463 zfs_bookmark_phys_t *fromzb, boolean_t is_clone, int outfd,
464 vnode_t *vp, offset_t *off)
466 zfs_bookmark_phys_t *fromzb, boolean_t is_clone, int outfd,
467 struct file *fp, offset_t *off)
471 dmu_replay_record_t *drr;
474 uint64_t fromtxg = 0;
476 err = dmu_objset_from_ds(ds, &os);
478 dsl_pool_rele(dp, tag);
482 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
483 drr->drr_type = DRR_BEGIN;
484 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
485 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
489 if (dmu_objset_type(os) == DMU_OST_ZFS) {
491 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
492 kmem_free(drr, sizeof (dmu_replay_record_t));
493 dsl_pool_rele(dp, tag);
494 return (SET_ERROR(EINVAL));
496 if (version >= ZPL_VERSION_SA) {
497 DMU_SET_FEATUREFLAGS(
498 drr->drr_u.drr_begin.drr_versioninfo,
499 DMU_BACKUP_FEATURE_SA_SPILL);
504 drr->drr_u.drr_begin.drr_creation_time =
505 ds->ds_phys->ds_creation_time;
506 drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
508 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
509 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
510 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
511 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
513 if (fromzb != NULL) {
514 drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid;
515 fromtxg = fromzb->zbm_creation_txg;
517 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
518 if (!dsl_dataset_is_snapshot(ds)) {
519 (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
520 sizeof (drr->drr_u.drr_begin.drr_toname));
523 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
526 dsp->dsa_outfd = outfd;
527 dsp->dsa_proc = curproc;
528 dsp->dsa_td = curthread;
532 dsp->dsa_toguid = ds->ds_phys->ds_guid;
533 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
534 dsp->dsa_pending_op = PENDING_NONE;
535 dsp->dsa_incremental = (fromzb != NULL);
537 mutex_enter(&ds->ds_sendstream_lock);
538 list_insert_head(&ds->ds_sendstreams, dsp);
539 mutex_exit(&ds->ds_sendstream_lock);
541 dsl_dataset_long_hold(ds, FTAG);
542 dsl_pool_rele(dp, tag);
544 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
549 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
552 if (dsp->dsa_pending_op != PENDING_NONE)
553 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
554 err = SET_ERROR(EINTR);
557 if (err == EINTR && dsp->dsa_err != 0)
562 bzero(drr, sizeof (dmu_replay_record_t));
563 drr->drr_type = DRR_END;
564 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
565 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
567 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
573 mutex_enter(&ds->ds_sendstream_lock);
574 list_remove(&ds->ds_sendstreams, dsp);
575 mutex_exit(&ds->ds_sendstream_lock);
577 kmem_free(drr, sizeof (dmu_replay_record_t));
578 kmem_free(dsp, sizeof (dmu_sendarg_t));
580 dsl_dataset_long_rele(ds, FTAG);
586 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
588 int outfd, vnode_t *vp, offset_t *off)
590 int outfd, struct file *fp, offset_t *off)
595 dsl_dataset_t *fromds = NULL;
598 err = dsl_pool_hold(pool, FTAG, &dp);
602 err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
604 dsl_pool_rele(dp, FTAG);
609 zfs_bookmark_phys_t zb;
612 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
614 dsl_dataset_rele(ds, FTAG);
615 dsl_pool_rele(dp, FTAG);
618 if (!dsl_dataset_is_before(ds, fromds, 0))
619 err = SET_ERROR(EXDEV);
620 zb.zbm_creation_time = fromds->ds_phys->ds_creation_time;
621 zb.zbm_creation_txg = fromds->ds_phys->ds_creation_txg;
622 zb.zbm_guid = fromds->ds_phys->ds_guid;
623 is_clone = (fromds->ds_dir != ds->ds_dir);
624 dsl_dataset_rele(fromds, FTAG);
625 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
628 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
631 dsl_dataset_rele(ds, FTAG);
636 dmu_send(const char *tosnap, const char *fromsnap,
638 int outfd, vnode_t *vp, offset_t *off)
640 int outfd, struct file *fp, offset_t *off)
646 boolean_t owned = B_FALSE;
648 if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
649 return (SET_ERROR(EINVAL));
651 err = dsl_pool_hold(tosnap, FTAG, &dp);
655 if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) {
657 * We are sending a filesystem or volume. Ensure
658 * that it doesn't change by owning the dataset.
660 err = dsl_dataset_own(dp, tosnap, FTAG, &ds);
663 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
666 dsl_pool_rele(dp, FTAG);
670 if (fromsnap != NULL) {
671 zfs_bookmark_phys_t zb;
672 boolean_t is_clone = B_FALSE;
673 int fsnamelen = strchr(tosnap, '@') - tosnap;
676 * If the fromsnap is in a different filesystem, then
677 * mark the send stream as a clone.
679 if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
680 (fromsnap[fsnamelen] != '@' &&
681 fromsnap[fsnamelen] != '#')) {
685 if (strchr(fromsnap, '@')) {
686 dsl_dataset_t *fromds;
687 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
689 if (!dsl_dataset_is_before(ds, fromds, 0))
690 err = SET_ERROR(EXDEV);
691 zb.zbm_creation_time =
692 fromds->ds_phys->ds_creation_time;
693 zb.zbm_creation_txg =
694 fromds->ds_phys->ds_creation_txg;
695 zb.zbm_guid = fromds->ds_phys->ds_guid;
696 is_clone = (ds->ds_dir != fromds->ds_dir);
697 dsl_dataset_rele(fromds, FTAG);
700 err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb);
703 dsl_dataset_rele(ds, FTAG);
704 dsl_pool_rele(dp, FTAG);
707 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
710 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
714 dsl_dataset_disown(ds, FTAG);
716 dsl_dataset_rele(ds, FTAG);
721 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
723 dsl_pool_t *dp = ds->ds_dir->dd_pool;
727 ASSERT(dsl_pool_config_held(dp));
729 /* tosnap must be a snapshot */
730 if (!dsl_dataset_is_snapshot(ds))
731 return (SET_ERROR(EINVAL));
734 * fromsnap must be an earlier snapshot from the same fs as tosnap,
735 * or the origin's fs.
737 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
738 return (SET_ERROR(EXDEV));
740 /* Get uncompressed size estimate of changed data. */
741 if (fromds == NULL) {
742 size = ds->ds_phys->ds_uncompressed_bytes;
745 err = dsl_dataset_space_written(fromds, ds,
746 &used, &comp, &size);
752 * Assume that space (both on-disk and in-stream) is dominated by
753 * data. We will adjust for indirect blocks and the copies property,
754 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
758 * Subtract out approximate space used by indirect blocks.
759 * Assume most space is used by data blocks (non-indirect, non-dnode).
760 * Assume all blocks are recordsize. Assume ditto blocks and
761 * internal fragmentation counter out compression.
763 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
764 * block, which we observe in practice.
767 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
770 size -= size / recordsize * sizeof (blkptr_t);
772 /* Add in the space for the record associated with each block. */
773 size += size / recordsize * sizeof (dmu_replay_record_t);
780 typedef struct dmu_recv_begin_arg {
781 const char *drba_origin;
782 dmu_recv_cookie_t *drba_cookie;
784 uint64_t drba_snapobj;
785 } dmu_recv_begin_arg_t;
788 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
793 dsl_pool_t *dp = ds->ds_dir->dd_pool;
795 /* temporary clone name must not exist */
796 error = zap_lookup(dp->dp_meta_objset,
797 ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
800 return (error == 0 ? EBUSY : error);
802 /* new snapshot name must not exist */
803 error = zap_lookup(dp->dp_meta_objset,
804 ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
807 return (error == 0 ? EEXIST : error);
811 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
813 /* Find snapshot in this dir that matches fromguid. */
815 error = dsl_dataset_hold_obj(dp, obj, FTAG,
818 return (SET_ERROR(ENODEV));
819 if (snap->ds_dir != ds->ds_dir) {
820 dsl_dataset_rele(snap, FTAG);
821 return (SET_ERROR(ENODEV));
823 if (snap->ds_phys->ds_guid == fromguid)
825 obj = snap->ds_phys->ds_prev_snap_obj;
826 dsl_dataset_rele(snap, FTAG);
829 return (SET_ERROR(ENODEV));
831 if (drba->drba_cookie->drc_force) {
832 drba->drba_snapobj = obj;
835 * If we are not forcing, there must be no
836 * changes since fromsnap.
838 if (dsl_dataset_modified_since_snap(ds, snap)) {
839 dsl_dataset_rele(snap, FTAG);
840 return (SET_ERROR(ETXTBSY));
842 drba->drba_snapobj = ds->ds_prev->ds_object;
845 dsl_dataset_rele(snap, FTAG);
847 /* if full, most recent snapshot must be $ORIGIN */
848 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
849 return (SET_ERROR(ENODEV));
850 drba->drba_snapobj = ds->ds_phys->ds_prev_snap_obj;
858 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
860 dmu_recv_begin_arg_t *drba = arg;
861 dsl_pool_t *dp = dmu_tx_pool(tx);
862 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
863 uint64_t fromguid = drrb->drr_fromguid;
864 int flags = drrb->drr_flags;
867 const char *tofs = drba->drba_cookie->drc_tofs;
869 /* already checked */
870 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
872 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
873 DMU_COMPOUNDSTREAM ||
874 drrb->drr_type >= DMU_OST_NUMTYPES ||
875 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
876 return (SET_ERROR(EINVAL));
878 /* Verify pool version supports SA if SA_SPILL feature set */
879 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
880 DMU_BACKUP_FEATURE_SA_SPILL) &&
881 spa_version(dp->dp_spa) < SPA_VERSION_SA) {
882 return (SET_ERROR(ENOTSUP));
885 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
887 /* target fs already exists; recv into temp clone */
889 /* Can't recv a clone into an existing fs */
890 if (flags & DRR_FLAG_CLONE) {
891 dsl_dataset_rele(ds, FTAG);
892 return (SET_ERROR(EINVAL));
895 error = recv_begin_check_existing_impl(drba, ds, fromguid);
896 dsl_dataset_rele(ds, FTAG);
897 } else if (error == ENOENT) {
898 /* target fs does not exist; must be a full backup or clone */
899 char buf[MAXNAMELEN];
902 * If it's a non-clone incremental, we are missing the
903 * target fs, so fail the recv.
905 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
906 return (SET_ERROR(ENOENT));
908 /* Open the parent of tofs */
909 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
910 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
911 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
915 if (drba->drba_origin != NULL) {
916 dsl_dataset_t *origin;
917 error = dsl_dataset_hold(dp, drba->drba_origin,
920 dsl_dataset_rele(ds, FTAG);
923 if (!dsl_dataset_is_snapshot(origin)) {
924 dsl_dataset_rele(origin, FTAG);
925 dsl_dataset_rele(ds, FTAG);
926 return (SET_ERROR(EINVAL));
928 if (origin->ds_phys->ds_guid != fromguid) {
929 dsl_dataset_rele(origin, FTAG);
930 dsl_dataset_rele(ds, FTAG);
931 return (SET_ERROR(ENODEV));
933 dsl_dataset_rele(origin, FTAG);
935 dsl_dataset_rele(ds, FTAG);
942 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
944 dmu_recv_begin_arg_t *drba = arg;
945 dsl_pool_t *dp = dmu_tx_pool(tx);
946 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
947 const char *tofs = drba->drba_cookie->drc_tofs;
948 dsl_dataset_t *ds, *newds;
953 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
954 DS_FLAG_CI_DATASET : 0;
956 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
958 /* create temporary clone */
959 dsl_dataset_t *snap = NULL;
960 if (drba->drba_snapobj != 0) {
961 VERIFY0(dsl_dataset_hold_obj(dp,
962 drba->drba_snapobj, FTAG, &snap));
964 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
965 snap, crflags, drba->drba_cred, tx);
966 dsl_dataset_rele(snap, FTAG);
967 dsl_dataset_rele(ds, FTAG);
971 dsl_dataset_t *origin = NULL;
973 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
975 if (drba->drba_origin != NULL) {
976 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
980 /* Create new dataset. */
981 dsobj = dsl_dataset_create_sync(dd,
982 strrchr(tofs, '/') + 1,
983 origin, crflags, drba->drba_cred, tx);
985 dsl_dataset_rele(origin, FTAG);
986 dsl_dir_rele(dd, FTAG);
987 drba->drba_cookie->drc_newfs = B_TRUE;
989 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
991 dmu_buf_will_dirty(newds->ds_dbuf, tx);
992 newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
995 * If we actually created a non-clone, we need to create the
996 * objset in our new dataset.
998 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
999 (void) dmu_objset_create_impl(dp->dp_spa,
1000 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1003 drba->drba_cookie->drc_ds = newds;
1005 spa_history_log_internal_ds(newds, "receive", tx, "");
1009 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1010 * succeeds; otherwise we will leak the holds on the datasets.
1013 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1014 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1016 dmu_recv_begin_arg_t drba = { 0 };
1017 dmu_replay_record_t *drr;
1019 bzero(drc, sizeof (dmu_recv_cookie_t));
1020 drc->drc_drrb = drrb;
1021 drc->drc_tosnap = tosnap;
1022 drc->drc_tofs = tofs;
1023 drc->drc_force = force;
1025 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1026 drc->drc_byteswap = B_TRUE;
1027 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1028 return (SET_ERROR(EINVAL));
1030 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1031 drr->drr_type = DRR_BEGIN;
1032 drr->drr_u.drr_begin = *drc->drc_drrb;
1033 if (drc->drc_byteswap) {
1034 fletcher_4_incremental_byteswap(drr,
1035 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1037 fletcher_4_incremental_native(drr,
1038 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1040 kmem_free(drr, sizeof (dmu_replay_record_t));
1042 if (drc->drc_byteswap) {
1043 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1044 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1045 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1046 drrb->drr_type = BSWAP_32(drrb->drr_type);
1047 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1048 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1051 drba.drba_origin = origin;
1052 drba.drba_cookie = drc;
1053 drba.drba_cred = CRED();
1055 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1066 int bufsize; /* amount of memory allocated for buf */
1068 avl_tree_t *guid_to_ds_map;
1071 typedef struct guid_map_entry {
1073 dsl_dataset_t *gme_ds;
1078 guid_compare(const void *arg1, const void *arg2)
1080 const guid_map_entry_t *gmep1 = arg1;
1081 const guid_map_entry_t *gmep2 = arg2;
1083 if (gmep1->guid < gmep2->guid)
1085 else if (gmep1->guid > gmep2->guid)
1091 free_guid_map_onexit(void *arg)
1093 avl_tree_t *ca = arg;
1094 void *cookie = NULL;
1095 guid_map_entry_t *gmep;
1097 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1098 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1099 dsl_dataset_rele(gmep->gme_ds, gmep);
1100 kmem_free(gmep, sizeof (guid_map_entry_t));
1103 kmem_free(ca, sizeof (avl_tree_t));
1107 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
1113 aiov.iov_base = buf;
1115 auio.uio_iov = &aiov;
1116 auio.uio_iovcnt = 1;
1117 auio.uio_resid = len;
1118 auio.uio_segflg = UIO_SYSSPACE;
1119 auio.uio_rw = UIO_READ;
1120 auio.uio_offset = off;
1121 auio.uio_td = ra->td;
1123 error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
1125 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
1128 *resid = auio.uio_resid;
1133 restore_read(struct restorearg *ra, int len)
1138 /* some things will require 8-byte alignment, so everything must */
1141 while (done < len) {
1144 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
1145 len - done, ra->voff, &resid);
1147 if (resid == len - done)
1148 ra->err = SET_ERROR(EINVAL);
1149 ra->voff += len - done - resid;
1155 ASSERT3U(done, ==, len);
1158 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1160 fletcher_4_incremental_native(rv, len, &ra->cksum);
1165 backup_byteswap(dmu_replay_record_t *drr)
1167 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1168 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1169 drr->drr_type = BSWAP_32(drr->drr_type);
1170 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1171 switch (drr->drr_type) {
1173 DO64(drr_begin.drr_magic);
1174 DO64(drr_begin.drr_versioninfo);
1175 DO64(drr_begin.drr_creation_time);
1176 DO32(drr_begin.drr_type);
1177 DO32(drr_begin.drr_flags);
1178 DO64(drr_begin.drr_toguid);
1179 DO64(drr_begin.drr_fromguid);
1182 DO64(drr_object.drr_object);
1183 /* DO64(drr_object.drr_allocation_txg); */
1184 DO32(drr_object.drr_type);
1185 DO32(drr_object.drr_bonustype);
1186 DO32(drr_object.drr_blksz);
1187 DO32(drr_object.drr_bonuslen);
1188 DO64(drr_object.drr_toguid);
1190 case DRR_FREEOBJECTS:
1191 DO64(drr_freeobjects.drr_firstobj);
1192 DO64(drr_freeobjects.drr_numobjs);
1193 DO64(drr_freeobjects.drr_toguid);
1196 DO64(drr_write.drr_object);
1197 DO32(drr_write.drr_type);
1198 DO64(drr_write.drr_offset);
1199 DO64(drr_write.drr_length);
1200 DO64(drr_write.drr_toguid);
1201 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1202 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1203 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1204 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1205 DO64(drr_write.drr_key.ddk_prop);
1207 case DRR_WRITE_BYREF:
1208 DO64(drr_write_byref.drr_object);
1209 DO64(drr_write_byref.drr_offset);
1210 DO64(drr_write_byref.drr_length);
1211 DO64(drr_write_byref.drr_toguid);
1212 DO64(drr_write_byref.drr_refguid);
1213 DO64(drr_write_byref.drr_refobject);
1214 DO64(drr_write_byref.drr_refoffset);
1215 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1216 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1217 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1218 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1219 DO64(drr_write_byref.drr_key.ddk_prop);
1222 DO64(drr_free.drr_object);
1223 DO64(drr_free.drr_offset);
1224 DO64(drr_free.drr_length);
1225 DO64(drr_free.drr_toguid);
1228 DO64(drr_spill.drr_object);
1229 DO64(drr_spill.drr_length);
1230 DO64(drr_spill.drr_toguid);
1233 DO64(drr_end.drr_checksum.zc_word[0]);
1234 DO64(drr_end.drr_checksum.zc_word[1]);
1235 DO64(drr_end.drr_checksum.zc_word[2]);
1236 DO64(drr_end.drr_checksum.zc_word[3]);
1237 DO64(drr_end.drr_toguid);
1245 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1251 if (drro->drr_type == DMU_OT_NONE ||
1252 !DMU_OT_IS_VALID(drro->drr_type) ||
1253 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1254 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1255 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1256 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1257 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1258 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1259 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1260 return (SET_ERROR(EINVAL));
1263 err = dmu_object_info(os, drro->drr_object, NULL);
1265 if (err != 0 && err != ENOENT)
1266 return (SET_ERROR(EINVAL));
1268 if (drro->drr_bonuslen) {
1269 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1274 if (err == ENOENT) {
1275 /* currently free, want to be allocated */
1276 tx = dmu_tx_create(os);
1277 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1278 err = dmu_tx_assign(tx, TXG_WAIT);
1283 err = dmu_object_claim(os, drro->drr_object,
1284 drro->drr_type, drro->drr_blksz,
1285 drro->drr_bonustype, drro->drr_bonuslen, tx);
1288 /* currently allocated, want to be allocated */
1289 err = dmu_object_reclaim(os, drro->drr_object,
1290 drro->drr_type, drro->drr_blksz,
1291 drro->drr_bonustype, drro->drr_bonuslen);
1294 return (SET_ERROR(EINVAL));
1297 tx = dmu_tx_create(os);
1298 dmu_tx_hold_bonus(tx, drro->drr_object);
1299 err = dmu_tx_assign(tx, TXG_WAIT);
1305 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1307 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1312 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1313 dmu_buf_will_dirty(db, tx);
1315 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1316 bcopy(data, db->db_data, drro->drr_bonuslen);
1318 dmu_object_byteswap_t byteswap =
1319 DMU_OT_BYTESWAP(drro->drr_bonustype);
1320 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1321 drro->drr_bonuslen);
1323 dmu_buf_rele(db, FTAG);
1331 restore_freeobjects(struct restorearg *ra, objset_t *os,
1332 struct drr_freeobjects *drrfo)
1336 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1337 return (SET_ERROR(EINVAL));
1339 for (obj = drrfo->drr_firstobj;
1340 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1341 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1344 if (dmu_object_info(os, obj, NULL) != 0)
1347 err = dmu_free_long_object(os, obj);
1355 restore_write(struct restorearg *ra, objset_t *os,
1356 struct drr_write *drrw)
1362 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1363 !DMU_OT_IS_VALID(drrw->drr_type))
1364 return (SET_ERROR(EINVAL));
1366 data = restore_read(ra, drrw->drr_length);
1370 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1371 return (SET_ERROR(EINVAL));
1373 tx = dmu_tx_create(os);
1375 dmu_tx_hold_write(tx, drrw->drr_object,
1376 drrw->drr_offset, drrw->drr_length);
1377 err = dmu_tx_assign(tx, TXG_WAIT);
1383 dmu_object_byteswap_t byteswap =
1384 DMU_OT_BYTESWAP(drrw->drr_type);
1385 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1387 dmu_write(os, drrw->drr_object,
1388 drrw->drr_offset, drrw->drr_length, data, tx);
1394 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1395 * streams to refer to a copy of the data that is already on the
1396 * system because it came in earlier in the stream. This function
1397 * finds the earlier copy of the data, and uses that copy instead of
1398 * data from the stream to fulfill this write.
1401 restore_write_byref(struct restorearg *ra, objset_t *os,
1402 struct drr_write_byref *drrwbr)
1406 guid_map_entry_t gmesrch;
1407 guid_map_entry_t *gmep;
1409 objset_t *ref_os = NULL;
1412 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1413 return (SET_ERROR(EINVAL));
1416 * If the GUID of the referenced dataset is different from the
1417 * GUID of the target dataset, find the referenced dataset.
1419 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1420 gmesrch.guid = drrwbr->drr_refguid;
1421 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1423 return (SET_ERROR(EINVAL));
1425 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1426 return (SET_ERROR(EINVAL));
1431 if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1432 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1435 tx = dmu_tx_create(os);
1437 dmu_tx_hold_write(tx, drrwbr->drr_object,
1438 drrwbr->drr_offset, drrwbr->drr_length);
1439 err = dmu_tx_assign(tx, TXG_WAIT);
1444 dmu_write(os, drrwbr->drr_object,
1445 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1446 dmu_buf_rele(dbp, FTAG);
1452 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1456 dmu_buf_t *db, *db_spill;
1459 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1460 drrs->drr_length > SPA_MAXBLOCKSIZE)
1461 return (SET_ERROR(EINVAL));
1463 data = restore_read(ra, drrs->drr_length);
1467 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1468 return (SET_ERROR(EINVAL));
1470 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1471 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1472 dmu_buf_rele(db, FTAG);
1476 tx = dmu_tx_create(os);
1478 dmu_tx_hold_spill(tx, db->db_object);
1480 err = dmu_tx_assign(tx, TXG_WAIT);
1482 dmu_buf_rele(db, FTAG);
1483 dmu_buf_rele(db_spill, FTAG);
1487 dmu_buf_will_dirty(db_spill, tx);
1489 if (db_spill->db_size < drrs->drr_length)
1490 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1491 drrs->drr_length, tx));
1492 bcopy(data, db_spill->db_data, drrs->drr_length);
1494 dmu_buf_rele(db, FTAG);
1495 dmu_buf_rele(db_spill, FTAG);
1503 restore_free(struct restorearg *ra, objset_t *os,
1504 struct drr_free *drrf)
1508 if (drrf->drr_length != -1ULL &&
1509 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1510 return (SET_ERROR(EINVAL));
1512 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1513 return (SET_ERROR(EINVAL));
1515 err = dmu_free_long_range(os, drrf->drr_object,
1516 drrf->drr_offset, drrf->drr_length);
1520 /* used to destroy the drc_ds on error */
1522 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1524 char name[MAXNAMELEN];
1525 dsl_dataset_name(drc->drc_ds, name);
1526 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1527 (void) dsl_destroy_head(name);
1531 * NB: callers *must* call dmu_recv_end() if this succeeds.
1534 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1535 int cleanup_fd, uint64_t *action_handlep)
1537 struct restorearg ra = { 0 };
1538 dmu_replay_record_t *drr;
1543 ra.byteswap = drc->drc_byteswap;
1544 ra.cksum = drc->drc_cksum;
1549 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1551 /* these were verified in dmu_recv_begin */
1552 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1554 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1557 * Open the objset we are modifying.
1559 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1561 ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1563 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1565 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1566 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1569 if (cleanup_fd == -1) {
1570 ra.err = SET_ERROR(EBADF);
1573 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1579 if (*action_handlep == 0) {
1581 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1582 avl_create(ra.guid_to_ds_map, guid_compare,
1583 sizeof (guid_map_entry_t),
1584 offsetof(guid_map_entry_t, avlnode));
1585 ra.err = zfs_onexit_add_cb(minor,
1586 free_guid_map_onexit, ra.guid_to_ds_map,
1591 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1592 (void **)&ra.guid_to_ds_map);
1597 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1601 * Read records and process them.
1604 while (ra.err == 0 &&
1605 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1606 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1607 ra.err = SET_ERROR(EINTR);
1612 backup_byteswap(drr);
1614 switch (drr->drr_type) {
1618 * We need to make a copy of the record header,
1619 * because restore_{object,write} may need to
1620 * restore_read(), which will invalidate drr.
1622 struct drr_object drro = drr->drr_u.drr_object;
1623 ra.err = restore_object(&ra, os, &drro);
1626 case DRR_FREEOBJECTS:
1628 struct drr_freeobjects drrfo =
1629 drr->drr_u.drr_freeobjects;
1630 ra.err = restore_freeobjects(&ra, os, &drrfo);
1635 struct drr_write drrw = drr->drr_u.drr_write;
1636 ra.err = restore_write(&ra, os, &drrw);
1639 case DRR_WRITE_BYREF:
1641 struct drr_write_byref drrwbr =
1642 drr->drr_u.drr_write_byref;
1643 ra.err = restore_write_byref(&ra, os, &drrwbr);
1648 struct drr_free drrf = drr->drr_u.drr_free;
1649 ra.err = restore_free(&ra, os, &drrf);
1654 struct drr_end drre = drr->drr_u.drr_end;
1656 * We compare against the *previous* checksum
1657 * value, because the stored checksum is of
1658 * everything before the DRR_END record.
1660 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1661 ra.err = SET_ERROR(ECKSUM);
1666 struct drr_spill drrs = drr->drr_u.drr_spill;
1667 ra.err = restore_spill(&ra, os, &drrs);
1671 ra.err = SET_ERROR(EINVAL);
1676 ASSERT(ra.err != 0);
1679 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1680 zfs_onexit_fd_rele(cleanup_fd);
1684 * destroy what we created, so we don't leave it in the
1685 * inconsistent restoring state.
1687 dmu_recv_cleanup_ds(drc);
1690 kmem_free(ra.buf, ra.bufsize);
1696 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1698 dmu_recv_cookie_t *drc = arg;
1699 dsl_pool_t *dp = dmu_tx_pool(tx);
1702 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1704 if (!drc->drc_newfs) {
1705 dsl_dataset_t *origin_head;
1707 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1710 if (drc->drc_force) {
1712 * We will destroy any snapshots in tofs (i.e. before
1713 * origin_head) that are after the origin (which is
1714 * the snap before drc_ds, because drc_ds can not
1715 * have any snaps of its own).
1717 uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1718 while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1719 dsl_dataset_t *snap;
1720 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1724 if (snap->ds_dir != origin_head->ds_dir)
1725 error = SET_ERROR(EINVAL);
1727 error = dsl_destroy_snapshot_check_impl(
1730 obj = snap->ds_phys->ds_prev_snap_obj;
1731 dsl_dataset_rele(snap, FTAG);
1736 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1737 origin_head, drc->drc_force, drc->drc_owner, tx);
1739 dsl_dataset_rele(origin_head, FTAG);
1742 error = dsl_dataset_snapshot_check_impl(origin_head,
1743 drc->drc_tosnap, tx, B_TRUE);
1744 dsl_dataset_rele(origin_head, FTAG);
1748 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1750 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1751 drc->drc_tosnap, tx, B_TRUE);
1757 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1759 dmu_recv_cookie_t *drc = arg;
1760 dsl_pool_t *dp = dmu_tx_pool(tx);
1762 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1763 tx, "snap=%s", drc->drc_tosnap);
1765 if (!drc->drc_newfs) {
1766 dsl_dataset_t *origin_head;
1768 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1771 if (drc->drc_force) {
1773 * Destroy any snapshots of drc_tofs (origin_head)
1774 * after the origin (the snap before drc_ds).
1776 uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1777 while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1778 dsl_dataset_t *snap;
1779 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1781 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1782 obj = snap->ds_phys->ds_prev_snap_obj;
1783 dsl_destroy_snapshot_sync_impl(snap,
1785 dsl_dataset_rele(snap, FTAG);
1788 VERIFY3P(drc->drc_ds->ds_prev, ==,
1789 origin_head->ds_prev);
1791 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1793 dsl_dataset_snapshot_sync_impl(origin_head,
1794 drc->drc_tosnap, tx);
1796 /* set snapshot's creation time and guid */
1797 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1798 origin_head->ds_prev->ds_phys->ds_creation_time =
1799 drc->drc_drrb->drr_creation_time;
1800 origin_head->ds_prev->ds_phys->ds_guid =
1801 drc->drc_drrb->drr_toguid;
1802 origin_head->ds_prev->ds_phys->ds_flags &=
1803 ~DS_FLAG_INCONSISTENT;
1805 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1806 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1808 dsl_dataset_rele(origin_head, FTAG);
1809 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1811 if (drc->drc_owner != NULL)
1812 VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
1814 dsl_dataset_t *ds = drc->drc_ds;
1816 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1818 /* set snapshot's creation time and guid */
1819 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1820 ds->ds_prev->ds_phys->ds_creation_time =
1821 drc->drc_drrb->drr_creation_time;
1822 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1823 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1825 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1826 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1828 drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1830 * Release the hold from dmu_recv_begin. This must be done before
1831 * we return to open context, so that when we free the dataset's dnode,
1832 * we can evict its bonus buffer.
1834 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1839 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1842 dsl_dataset_t *snapds;
1843 guid_map_entry_t *gmep;
1846 ASSERT(guid_map != NULL);
1848 err = dsl_pool_hold(name, FTAG, &dp);
1851 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
1852 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
1854 gmep->guid = snapds->ds_phys->ds_guid;
1855 gmep->gme_ds = snapds;
1856 avl_add(guid_map, gmep);
1857 dsl_dataset_long_hold(snapds, gmep);
1859 kmem_free(gmep, sizeof (*gmep));
1861 dsl_pool_rele(dp, FTAG);
1865 static int dmu_recv_end_modified_blocks = 3;
1868 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1871 char name[MAXNAMELEN];
1875 * We will be destroying the ds; make sure its origin is unmounted if
1878 dsl_dataset_name(drc->drc_ds, name);
1879 zfs_destroy_unmount_origin(name);
1882 error = dsl_sync_task(drc->drc_tofs,
1883 dmu_recv_end_check, dmu_recv_end_sync, drc,
1884 dmu_recv_end_modified_blocks);
1887 dmu_recv_cleanup_ds(drc);
1892 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1896 error = dsl_sync_task(drc->drc_tofs,
1897 dmu_recv_end_check, dmu_recv_end_sync, drc,
1898 dmu_recv_end_modified_blocks);
1901 dmu_recv_cleanup_ds(drc);
1902 } else if (drc->drc_guid_to_ds_map != NULL) {
1903 (void) add_ds_to_guidmap(drc->drc_tofs,
1904 drc->drc_guid_to_ds_map,
1905 drc->drc_newsnapobj);
1911 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
1913 drc->drc_owner = owner;
1916 return (dmu_recv_new_end(drc));
1918 return (dmu_recv_existing_end(drc));
1922 * Return TRUE if this objset is currently being received into.
1925 dmu_objset_is_receiving(objset_t *os)
1927 return (os->os_dsl_dataset != NULL &&
1928 os->os_dsl_dataset->ds_owner == dmu_recv_tag);