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) 2014, 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);
810 * Check snapshot limit before receiving. We'll recheck again at the
811 * end, but might as well abort before receiving if we're already over
814 * Note that we do not check the file system limit with
815 * dsl_dir_fscount_check because the temporary %clones don't count
816 * against that limit.
818 error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT,
819 NULL, drba->drba_cred);
825 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
827 /* Find snapshot in this dir that matches fromguid. */
829 error = dsl_dataset_hold_obj(dp, obj, FTAG,
832 return (SET_ERROR(ENODEV));
833 if (snap->ds_dir != ds->ds_dir) {
834 dsl_dataset_rele(snap, FTAG);
835 return (SET_ERROR(ENODEV));
837 if (snap->ds_phys->ds_guid == fromguid)
839 obj = snap->ds_phys->ds_prev_snap_obj;
840 dsl_dataset_rele(snap, FTAG);
843 return (SET_ERROR(ENODEV));
845 if (drba->drba_cookie->drc_force) {
846 drba->drba_snapobj = obj;
849 * If we are not forcing, there must be no
850 * changes since fromsnap.
852 if (dsl_dataset_modified_since_snap(ds, snap)) {
853 dsl_dataset_rele(snap, FTAG);
854 return (SET_ERROR(ETXTBSY));
856 drba->drba_snapobj = ds->ds_prev->ds_object;
859 dsl_dataset_rele(snap, FTAG);
861 /* if full, most recent snapshot must be $ORIGIN */
862 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
863 return (SET_ERROR(ENODEV));
864 drba->drba_snapobj = ds->ds_phys->ds_prev_snap_obj;
872 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
874 dmu_recv_begin_arg_t *drba = arg;
875 dsl_pool_t *dp = dmu_tx_pool(tx);
876 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
877 uint64_t fromguid = drrb->drr_fromguid;
878 int flags = drrb->drr_flags;
881 const char *tofs = drba->drba_cookie->drc_tofs;
883 /* already checked */
884 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
886 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
887 DMU_COMPOUNDSTREAM ||
888 drrb->drr_type >= DMU_OST_NUMTYPES ||
889 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
890 return (SET_ERROR(EINVAL));
892 /* Verify pool version supports SA if SA_SPILL feature set */
893 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
894 DMU_BACKUP_FEATURE_SA_SPILL) &&
895 spa_version(dp->dp_spa) < SPA_VERSION_SA) {
896 return (SET_ERROR(ENOTSUP));
899 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
901 /* target fs already exists; recv into temp clone */
903 /* Can't recv a clone into an existing fs */
904 if (flags & DRR_FLAG_CLONE) {
905 dsl_dataset_rele(ds, FTAG);
906 return (SET_ERROR(EINVAL));
909 error = recv_begin_check_existing_impl(drba, ds, fromguid);
910 dsl_dataset_rele(ds, FTAG);
911 } else if (error == ENOENT) {
912 /* target fs does not exist; must be a full backup or clone */
913 char buf[MAXNAMELEN];
916 * If it's a non-clone incremental, we are missing the
917 * target fs, so fail the recv.
919 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
920 return (SET_ERROR(ENOENT));
922 /* Open the parent of tofs */
923 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
924 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
925 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
930 * Check filesystem and snapshot limits before receiving. We'll
931 * recheck snapshot limits again at the end (we create the
932 * filesystems and increment those counts during begin_sync).
934 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
935 ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred);
937 dsl_dataset_rele(ds, FTAG);
941 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
942 ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred);
944 dsl_dataset_rele(ds, FTAG);
948 if (drba->drba_origin != NULL) {
949 dsl_dataset_t *origin;
950 error = dsl_dataset_hold(dp, drba->drba_origin,
953 dsl_dataset_rele(ds, FTAG);
956 if (!dsl_dataset_is_snapshot(origin)) {
957 dsl_dataset_rele(origin, FTAG);
958 dsl_dataset_rele(ds, FTAG);
959 return (SET_ERROR(EINVAL));
961 if (origin->ds_phys->ds_guid != fromguid) {
962 dsl_dataset_rele(origin, FTAG);
963 dsl_dataset_rele(ds, FTAG);
964 return (SET_ERROR(ENODEV));
966 dsl_dataset_rele(origin, FTAG);
968 dsl_dataset_rele(ds, FTAG);
975 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
977 dmu_recv_begin_arg_t *drba = arg;
978 dsl_pool_t *dp = dmu_tx_pool(tx);
979 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
980 const char *tofs = drba->drba_cookie->drc_tofs;
981 dsl_dataset_t *ds, *newds;
986 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
987 DS_FLAG_CI_DATASET : 0;
989 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
991 /* create temporary clone */
992 dsl_dataset_t *snap = NULL;
993 if (drba->drba_snapobj != 0) {
994 VERIFY0(dsl_dataset_hold_obj(dp,
995 drba->drba_snapobj, FTAG, &snap));
997 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
998 snap, crflags, drba->drba_cred, tx);
999 dsl_dataset_rele(snap, FTAG);
1000 dsl_dataset_rele(ds, FTAG);
1004 dsl_dataset_t *origin = NULL;
1006 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1008 if (drba->drba_origin != NULL) {
1009 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1013 /* Create new dataset. */
1014 dsobj = dsl_dataset_create_sync(dd,
1015 strrchr(tofs, '/') + 1,
1016 origin, crflags, drba->drba_cred, tx);
1018 dsl_dataset_rele(origin, FTAG);
1019 dsl_dir_rele(dd, FTAG);
1020 drba->drba_cookie->drc_newfs = B_TRUE;
1022 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1024 dmu_buf_will_dirty(newds->ds_dbuf, tx);
1025 newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1028 * If we actually created a non-clone, we need to create the
1029 * objset in our new dataset.
1031 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1032 (void) dmu_objset_create_impl(dp->dp_spa,
1033 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1036 drba->drba_cookie->drc_ds = newds;
1038 spa_history_log_internal_ds(newds, "receive", tx, "");
1042 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1043 * succeeds; otherwise we will leak the holds on the datasets.
1046 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1047 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1049 dmu_recv_begin_arg_t drba = { 0 };
1050 dmu_replay_record_t *drr;
1052 bzero(drc, sizeof (dmu_recv_cookie_t));
1053 drc->drc_drrb = drrb;
1054 drc->drc_tosnap = tosnap;
1055 drc->drc_tofs = tofs;
1056 drc->drc_force = force;
1057 drc->drc_cred = CRED();
1059 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1060 drc->drc_byteswap = B_TRUE;
1061 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1062 return (SET_ERROR(EINVAL));
1064 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1065 drr->drr_type = DRR_BEGIN;
1066 drr->drr_u.drr_begin = *drc->drc_drrb;
1067 if (drc->drc_byteswap) {
1068 fletcher_4_incremental_byteswap(drr,
1069 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1071 fletcher_4_incremental_native(drr,
1072 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1074 kmem_free(drr, sizeof (dmu_replay_record_t));
1076 if (drc->drc_byteswap) {
1077 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1078 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1079 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1080 drrb->drr_type = BSWAP_32(drrb->drr_type);
1081 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1082 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1085 drba.drba_origin = origin;
1086 drba.drba_cookie = drc;
1087 drba.drba_cred = CRED();
1089 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1100 int bufsize; /* amount of memory allocated for buf */
1102 avl_tree_t *guid_to_ds_map;
1105 typedef struct guid_map_entry {
1107 dsl_dataset_t *gme_ds;
1112 guid_compare(const void *arg1, const void *arg2)
1114 const guid_map_entry_t *gmep1 = arg1;
1115 const guid_map_entry_t *gmep2 = arg2;
1117 if (gmep1->guid < gmep2->guid)
1119 else if (gmep1->guid > gmep2->guid)
1125 free_guid_map_onexit(void *arg)
1127 avl_tree_t *ca = arg;
1128 void *cookie = NULL;
1129 guid_map_entry_t *gmep;
1131 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1132 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1133 dsl_dataset_rele(gmep->gme_ds, gmep);
1134 kmem_free(gmep, sizeof (guid_map_entry_t));
1137 kmem_free(ca, sizeof (avl_tree_t));
1141 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
1147 aiov.iov_base = buf;
1149 auio.uio_iov = &aiov;
1150 auio.uio_iovcnt = 1;
1151 auio.uio_resid = len;
1152 auio.uio_segflg = UIO_SYSSPACE;
1153 auio.uio_rw = UIO_READ;
1154 auio.uio_offset = off;
1155 auio.uio_td = ra->td;
1157 error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
1159 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
1162 *resid = auio.uio_resid;
1167 restore_read(struct restorearg *ra, int len)
1172 /* some things will require 8-byte alignment, so everything must */
1175 while (done < len) {
1178 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done,
1179 len - done, ra->voff, &resid);
1181 if (resid == len - done)
1182 ra->err = SET_ERROR(EINVAL);
1183 ra->voff += len - done - resid;
1189 ASSERT3U(done, ==, len);
1192 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1194 fletcher_4_incremental_native(rv, len, &ra->cksum);
1199 backup_byteswap(dmu_replay_record_t *drr)
1201 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1202 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1203 drr->drr_type = BSWAP_32(drr->drr_type);
1204 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1205 switch (drr->drr_type) {
1207 DO64(drr_begin.drr_magic);
1208 DO64(drr_begin.drr_versioninfo);
1209 DO64(drr_begin.drr_creation_time);
1210 DO32(drr_begin.drr_type);
1211 DO32(drr_begin.drr_flags);
1212 DO64(drr_begin.drr_toguid);
1213 DO64(drr_begin.drr_fromguid);
1216 DO64(drr_object.drr_object);
1217 /* DO64(drr_object.drr_allocation_txg); */
1218 DO32(drr_object.drr_type);
1219 DO32(drr_object.drr_bonustype);
1220 DO32(drr_object.drr_blksz);
1221 DO32(drr_object.drr_bonuslen);
1222 DO64(drr_object.drr_toguid);
1224 case DRR_FREEOBJECTS:
1225 DO64(drr_freeobjects.drr_firstobj);
1226 DO64(drr_freeobjects.drr_numobjs);
1227 DO64(drr_freeobjects.drr_toguid);
1230 DO64(drr_write.drr_object);
1231 DO32(drr_write.drr_type);
1232 DO64(drr_write.drr_offset);
1233 DO64(drr_write.drr_length);
1234 DO64(drr_write.drr_toguid);
1235 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1236 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1237 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1238 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1239 DO64(drr_write.drr_key.ddk_prop);
1241 case DRR_WRITE_BYREF:
1242 DO64(drr_write_byref.drr_object);
1243 DO64(drr_write_byref.drr_offset);
1244 DO64(drr_write_byref.drr_length);
1245 DO64(drr_write_byref.drr_toguid);
1246 DO64(drr_write_byref.drr_refguid);
1247 DO64(drr_write_byref.drr_refobject);
1248 DO64(drr_write_byref.drr_refoffset);
1249 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1250 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1251 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1252 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1253 DO64(drr_write_byref.drr_key.ddk_prop);
1256 DO64(drr_free.drr_object);
1257 DO64(drr_free.drr_offset);
1258 DO64(drr_free.drr_length);
1259 DO64(drr_free.drr_toguid);
1262 DO64(drr_spill.drr_object);
1263 DO64(drr_spill.drr_length);
1264 DO64(drr_spill.drr_toguid);
1267 DO64(drr_end.drr_checksum.zc_word[0]);
1268 DO64(drr_end.drr_checksum.zc_word[1]);
1269 DO64(drr_end.drr_checksum.zc_word[2]);
1270 DO64(drr_end.drr_checksum.zc_word[3]);
1271 DO64(drr_end.drr_toguid);
1279 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1285 if (drro->drr_type == DMU_OT_NONE ||
1286 !DMU_OT_IS_VALID(drro->drr_type) ||
1287 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1288 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1289 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1290 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1291 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1292 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1293 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1294 return (SET_ERROR(EINVAL));
1297 err = dmu_object_info(os, drro->drr_object, NULL);
1299 if (err != 0 && err != ENOENT)
1300 return (SET_ERROR(EINVAL));
1302 if (drro->drr_bonuslen) {
1303 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1308 if (err == ENOENT) {
1309 /* currently free, want to be allocated */
1310 tx = dmu_tx_create(os);
1311 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1312 err = dmu_tx_assign(tx, TXG_WAIT);
1317 err = dmu_object_claim(os, drro->drr_object,
1318 drro->drr_type, drro->drr_blksz,
1319 drro->drr_bonustype, drro->drr_bonuslen, tx);
1322 /* currently allocated, want to be allocated */
1323 err = dmu_object_reclaim(os, drro->drr_object,
1324 drro->drr_type, drro->drr_blksz,
1325 drro->drr_bonustype, drro->drr_bonuslen);
1328 return (SET_ERROR(EINVAL));
1331 tx = dmu_tx_create(os);
1332 dmu_tx_hold_bonus(tx, drro->drr_object);
1333 err = dmu_tx_assign(tx, TXG_WAIT);
1339 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1341 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1346 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1347 dmu_buf_will_dirty(db, tx);
1349 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1350 bcopy(data, db->db_data, drro->drr_bonuslen);
1352 dmu_object_byteswap_t byteswap =
1353 DMU_OT_BYTESWAP(drro->drr_bonustype);
1354 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1355 drro->drr_bonuslen);
1357 dmu_buf_rele(db, FTAG);
1365 restore_freeobjects(struct restorearg *ra, objset_t *os,
1366 struct drr_freeobjects *drrfo)
1370 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1371 return (SET_ERROR(EINVAL));
1373 for (obj = drrfo->drr_firstobj;
1374 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1375 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1378 if (dmu_object_info(os, obj, NULL) != 0)
1381 err = dmu_free_long_object(os, obj);
1389 restore_write(struct restorearg *ra, objset_t *os,
1390 struct drr_write *drrw)
1396 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1397 !DMU_OT_IS_VALID(drrw->drr_type))
1398 return (SET_ERROR(EINVAL));
1400 data = restore_read(ra, drrw->drr_length);
1404 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1405 return (SET_ERROR(EINVAL));
1407 tx = dmu_tx_create(os);
1409 dmu_tx_hold_write(tx, drrw->drr_object,
1410 drrw->drr_offset, drrw->drr_length);
1411 err = dmu_tx_assign(tx, TXG_WAIT);
1417 dmu_object_byteswap_t byteswap =
1418 DMU_OT_BYTESWAP(drrw->drr_type);
1419 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1421 dmu_write(os, drrw->drr_object,
1422 drrw->drr_offset, drrw->drr_length, data, tx);
1428 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1429 * streams to refer to a copy of the data that is already on the
1430 * system because it came in earlier in the stream. This function
1431 * finds the earlier copy of the data, and uses that copy instead of
1432 * data from the stream to fulfill this write.
1435 restore_write_byref(struct restorearg *ra, objset_t *os,
1436 struct drr_write_byref *drrwbr)
1440 guid_map_entry_t gmesrch;
1441 guid_map_entry_t *gmep;
1443 objset_t *ref_os = NULL;
1446 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1447 return (SET_ERROR(EINVAL));
1450 * If the GUID of the referenced dataset is different from the
1451 * GUID of the target dataset, find the referenced dataset.
1453 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1454 gmesrch.guid = drrwbr->drr_refguid;
1455 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1457 return (SET_ERROR(EINVAL));
1459 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1460 return (SET_ERROR(EINVAL));
1465 if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1466 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1469 tx = dmu_tx_create(os);
1471 dmu_tx_hold_write(tx, drrwbr->drr_object,
1472 drrwbr->drr_offset, drrwbr->drr_length);
1473 err = dmu_tx_assign(tx, TXG_WAIT);
1478 dmu_write(os, drrwbr->drr_object,
1479 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1480 dmu_buf_rele(dbp, FTAG);
1486 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1490 dmu_buf_t *db, *db_spill;
1493 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1494 drrs->drr_length > SPA_MAXBLOCKSIZE)
1495 return (SET_ERROR(EINVAL));
1497 data = restore_read(ra, drrs->drr_length);
1501 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1502 return (SET_ERROR(EINVAL));
1504 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1505 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1506 dmu_buf_rele(db, FTAG);
1510 tx = dmu_tx_create(os);
1512 dmu_tx_hold_spill(tx, db->db_object);
1514 err = dmu_tx_assign(tx, TXG_WAIT);
1516 dmu_buf_rele(db, FTAG);
1517 dmu_buf_rele(db_spill, FTAG);
1521 dmu_buf_will_dirty(db_spill, tx);
1523 if (db_spill->db_size < drrs->drr_length)
1524 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1525 drrs->drr_length, tx));
1526 bcopy(data, db_spill->db_data, drrs->drr_length);
1528 dmu_buf_rele(db, FTAG);
1529 dmu_buf_rele(db_spill, FTAG);
1537 restore_free(struct restorearg *ra, objset_t *os,
1538 struct drr_free *drrf)
1542 if (drrf->drr_length != -1ULL &&
1543 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1544 return (SET_ERROR(EINVAL));
1546 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1547 return (SET_ERROR(EINVAL));
1549 err = dmu_free_long_range(os, drrf->drr_object,
1550 drrf->drr_offset, drrf->drr_length);
1554 /* used to destroy the drc_ds on error */
1556 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1558 char name[MAXNAMELEN];
1559 dsl_dataset_name(drc->drc_ds, name);
1560 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1561 (void) dsl_destroy_head(name);
1565 * NB: callers *must* call dmu_recv_end() if this succeeds.
1568 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1569 int cleanup_fd, uint64_t *action_handlep)
1571 struct restorearg ra = { 0 };
1572 dmu_replay_record_t *drr;
1577 ra.byteswap = drc->drc_byteswap;
1578 ra.cksum = drc->drc_cksum;
1583 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1585 /* these were verified in dmu_recv_begin */
1586 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1588 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1591 * Open the objset we are modifying.
1593 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1595 ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1597 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1599 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1600 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1603 if (cleanup_fd == -1) {
1604 ra.err = SET_ERROR(EBADF);
1607 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1613 if (*action_handlep == 0) {
1615 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1616 avl_create(ra.guid_to_ds_map, guid_compare,
1617 sizeof (guid_map_entry_t),
1618 offsetof(guid_map_entry_t, avlnode));
1619 ra.err = zfs_onexit_add_cb(minor,
1620 free_guid_map_onexit, ra.guid_to_ds_map,
1625 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1626 (void **)&ra.guid_to_ds_map);
1631 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1635 * Read records and process them.
1638 while (ra.err == 0 &&
1639 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1640 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1641 ra.err = SET_ERROR(EINTR);
1646 backup_byteswap(drr);
1648 switch (drr->drr_type) {
1652 * We need to make a copy of the record header,
1653 * because restore_{object,write} may need to
1654 * restore_read(), which will invalidate drr.
1656 struct drr_object drro = drr->drr_u.drr_object;
1657 ra.err = restore_object(&ra, os, &drro);
1660 case DRR_FREEOBJECTS:
1662 struct drr_freeobjects drrfo =
1663 drr->drr_u.drr_freeobjects;
1664 ra.err = restore_freeobjects(&ra, os, &drrfo);
1669 struct drr_write drrw = drr->drr_u.drr_write;
1670 ra.err = restore_write(&ra, os, &drrw);
1673 case DRR_WRITE_BYREF:
1675 struct drr_write_byref drrwbr =
1676 drr->drr_u.drr_write_byref;
1677 ra.err = restore_write_byref(&ra, os, &drrwbr);
1682 struct drr_free drrf = drr->drr_u.drr_free;
1683 ra.err = restore_free(&ra, os, &drrf);
1688 struct drr_end drre = drr->drr_u.drr_end;
1690 * We compare against the *previous* checksum
1691 * value, because the stored checksum is of
1692 * everything before the DRR_END record.
1694 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1695 ra.err = SET_ERROR(ECKSUM);
1700 struct drr_spill drrs = drr->drr_u.drr_spill;
1701 ra.err = restore_spill(&ra, os, &drrs);
1705 ra.err = SET_ERROR(EINVAL);
1710 ASSERT(ra.err != 0);
1713 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1714 zfs_onexit_fd_rele(cleanup_fd);
1718 * destroy what we created, so we don't leave it in the
1719 * inconsistent restoring state.
1721 dmu_recv_cleanup_ds(drc);
1724 kmem_free(ra.buf, ra.bufsize);
1730 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1732 dmu_recv_cookie_t *drc = arg;
1733 dsl_pool_t *dp = dmu_tx_pool(tx);
1736 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1738 if (!drc->drc_newfs) {
1739 dsl_dataset_t *origin_head;
1741 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1744 if (drc->drc_force) {
1746 * We will destroy any snapshots in tofs (i.e. before
1747 * origin_head) that are after the origin (which is
1748 * the snap before drc_ds, because drc_ds can not
1749 * have any snaps of its own).
1751 uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1752 while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1753 dsl_dataset_t *snap;
1754 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1758 if (snap->ds_dir != origin_head->ds_dir)
1759 error = SET_ERROR(EINVAL);
1761 error = dsl_destroy_snapshot_check_impl(
1764 obj = snap->ds_phys->ds_prev_snap_obj;
1765 dsl_dataset_rele(snap, FTAG);
1770 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1771 origin_head, drc->drc_force, drc->drc_owner, tx);
1773 dsl_dataset_rele(origin_head, FTAG);
1776 error = dsl_dataset_snapshot_check_impl(origin_head,
1777 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
1778 dsl_dataset_rele(origin_head, FTAG);
1782 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1784 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1785 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
1791 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1793 dmu_recv_cookie_t *drc = arg;
1794 dsl_pool_t *dp = dmu_tx_pool(tx);
1796 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1797 tx, "snap=%s", drc->drc_tosnap);
1799 if (!drc->drc_newfs) {
1800 dsl_dataset_t *origin_head;
1802 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1805 if (drc->drc_force) {
1807 * Destroy any snapshots of drc_tofs (origin_head)
1808 * after the origin (the snap before drc_ds).
1810 uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1811 while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1812 dsl_dataset_t *snap;
1813 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1815 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1816 obj = snap->ds_phys->ds_prev_snap_obj;
1817 dsl_destroy_snapshot_sync_impl(snap,
1819 dsl_dataset_rele(snap, FTAG);
1822 VERIFY3P(drc->drc_ds->ds_prev, ==,
1823 origin_head->ds_prev);
1825 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1827 dsl_dataset_snapshot_sync_impl(origin_head,
1828 drc->drc_tosnap, tx);
1830 /* set snapshot's creation time and guid */
1831 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1832 origin_head->ds_prev->ds_phys->ds_creation_time =
1833 drc->drc_drrb->drr_creation_time;
1834 origin_head->ds_prev->ds_phys->ds_guid =
1835 drc->drc_drrb->drr_toguid;
1836 origin_head->ds_prev->ds_phys->ds_flags &=
1837 ~DS_FLAG_INCONSISTENT;
1839 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1840 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1842 dsl_dataset_rele(origin_head, FTAG);
1843 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1845 if (drc->drc_owner != NULL)
1846 VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
1848 dsl_dataset_t *ds = drc->drc_ds;
1850 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1852 /* set snapshot's creation time and guid */
1853 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1854 ds->ds_prev->ds_phys->ds_creation_time =
1855 drc->drc_drrb->drr_creation_time;
1856 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1857 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1859 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1860 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1862 drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1864 * Release the hold from dmu_recv_begin. This must be done before
1865 * we return to open context, so that when we free the dataset's dnode,
1866 * we can evict its bonus buffer.
1868 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1873 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1876 dsl_dataset_t *snapds;
1877 guid_map_entry_t *gmep;
1880 ASSERT(guid_map != NULL);
1882 err = dsl_pool_hold(name, FTAG, &dp);
1885 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
1886 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
1888 gmep->guid = snapds->ds_phys->ds_guid;
1889 gmep->gme_ds = snapds;
1890 avl_add(guid_map, gmep);
1891 dsl_dataset_long_hold(snapds, gmep);
1893 kmem_free(gmep, sizeof (*gmep));
1895 dsl_pool_rele(dp, FTAG);
1899 static int dmu_recv_end_modified_blocks = 3;
1902 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1905 char name[MAXNAMELEN];
1909 * We will be destroying the ds; make sure its origin is unmounted if
1912 dsl_dataset_name(drc->drc_ds, name);
1913 zfs_destroy_unmount_origin(name);
1916 error = dsl_sync_task(drc->drc_tofs,
1917 dmu_recv_end_check, dmu_recv_end_sync, drc,
1918 dmu_recv_end_modified_blocks);
1921 dmu_recv_cleanup_ds(drc);
1926 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1930 error = dsl_sync_task(drc->drc_tofs,
1931 dmu_recv_end_check, dmu_recv_end_sync, drc,
1932 dmu_recv_end_modified_blocks);
1935 dmu_recv_cleanup_ds(drc);
1936 } else if (drc->drc_guid_to_ds_map != NULL) {
1937 (void) add_ds_to_guidmap(drc->drc_tofs,
1938 drc->drc_guid_to_ds_map,
1939 drc->drc_newsnapobj);
1945 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
1947 drc->drc_owner = owner;
1950 return (dmu_recv_new_end(drc));
1952 return (dmu_recv_existing_end(drc));
1956 * Return TRUE if this objset is currently being received into.
1959 dmu_objset_is_receiving(objset_t *os)
1961 return (os->os_dsl_dataset != NULL &&
1962 os->os_dsl_dataset->ds_owner == dmu_recv_tag);