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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
36 #include <sys/mount.h>
42 #include "zfs_namecheck.h"
44 #include "zfs_fletcher.h"
45 #include "libzfs_impl.h"
46 #include <sys/zio_checksum.h>
49 /* in libzfs_dataset.c */
50 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
52 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t,
53 int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *);
55 static const zio_cksum_t zero_cksum = { { 0 } };
57 typedef struct dedup_arg {
60 libzfs_handle_t *dedup_hdl;
63 typedef struct dataref {
69 typedef struct dedup_entry {
70 struct dedup_entry *dde_next;
71 zio_cksum_t dde_chksum;
76 #define MAX_DDT_PHYSMEM_PERCENT 20
77 #define SMALLEST_POSSIBLE_MAX_DDT_MB 128
79 typedef struct dedup_table {
80 dedup_entry_t **dedup_hash_array;
81 umem_cache_t *ddecache;
82 uint64_t max_ddt_size; /* max dedup table size in bytes */
83 uint64_t cur_ddt_size; /* current dedup table size in bytes */
90 high_order_bit(uint64_t n)
94 for (count = 0; n != 0; count++)
100 ssread(void *buf, size_t len, FILE *stream)
104 if ((outlen = fread(buf, len, 1, stream)) == 0)
111 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
112 zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
116 if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
117 if (ddt->ddt_full == B_FALSE) {
118 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
119 "Dedup table full. Deduplication will continue "
120 "with existing table entries"));
121 ddt->ddt_full = B_TRUE;
126 if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
128 assert(*ddepp == NULL);
129 dde->dde_next = NULL;
130 dde->dde_chksum = *cs;
131 dde->dde_prop = prop;
134 ddt->cur_ddt_size += sizeof (dedup_entry_t);
140 * Using the specified dedup table, do a lookup for an entry with
141 * the checksum cs. If found, return the block's reference info
142 * in *dr. Otherwise, insert a new entry in the dedup table, using
143 * the reference information specified by *dr.
145 * return value: true - entry was found
146 * false - entry was not found
149 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
150 uint64_t prop, dataref_t *dr)
153 dedup_entry_t **ddepp;
155 hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
157 for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
158 ddepp = &((*ddepp)->dde_next)) {
159 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
160 (*ddepp)->dde_prop == prop) {
161 *dr = (*ddepp)->dde_ref;
165 ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
170 cksum_and_write(const void *buf, uint64_t len, zio_cksum_t *zc, int outfd)
172 fletcher_4_incremental_native(buf, len, zc);
173 return (write(outfd, buf, len));
177 * This function is started in a separate thread when the dedup option
178 * has been requested. The main send thread determines the list of
179 * snapshots to be included in the send stream and makes the ioctl calls
180 * for each one. But instead of having the ioctl send the output to the
181 * the output fd specified by the caller of zfs_send()), the
182 * ioctl is told to direct the output to a pipe, which is read by the
183 * alternate thread running THIS function. This function does the
185 * 1. building a dedup table (the DDT)
186 * 2. doing checksums on each data block and inserting a record in the DDT
187 * 3. looking for matching checksums, and
188 * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever
189 * a duplicate block is found.
190 * The output of this function then goes to the output fd requested
191 * by the caller of zfs_send().
196 dedup_arg_t *dda = arg;
197 char *buf = malloc(1<<20);
198 dmu_replay_record_t thedrr;
199 dmu_replay_record_t *drr = &thedrr;
200 struct drr_begin *drrb = &thedrr.drr_u.drr_begin;
201 struct drr_end *drre = &thedrr.drr_u.drr_end;
202 struct drr_object *drro = &thedrr.drr_u.drr_object;
203 struct drr_write *drrw = &thedrr.drr_u.drr_write;
204 struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
207 dmu_replay_record_t wbr_drr = {0};
208 struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref;
210 zio_cksum_t stream_cksum;
211 uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
215 MAX((physmem * MAX_DDT_PHYSMEM_PERCENT)/100,
216 SMALLEST_POSSIBLE_MAX_DDT_MB<<20);
218 numbuckets = ddt.max_ddt_size/(sizeof (dedup_entry_t));
221 * numbuckets must be a power of 2. Increase number to
222 * a power of 2 if necessary.
224 if (!ISP2(numbuckets))
225 numbuckets = 1 << high_order_bit(numbuckets);
227 ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
228 ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
229 NULL, NULL, NULL, NULL, NULL, 0);
230 ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
231 ddt.numhashbits = high_order_bit(numbuckets) - 1;
232 ddt.ddt_full = B_FALSE;
234 /* Initialize the write-by-reference block. */
235 wbr_drr.drr_type = DRR_WRITE_BYREF;
236 wbr_drr.drr_payloadlen = 0;
238 outfd = dda->outputfd;
239 ofp = fdopen(dda->inputfd, "r");
240 while (ssread(drr, sizeof (dmu_replay_record_t), ofp) != 0) {
242 switch (drr->drr_type) {
246 ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
248 /* set the DEDUP feature flag for this stream */
249 fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
250 fflags |= (DMU_BACKUP_FEATURE_DEDUP |
251 DMU_BACKUP_FEATURE_DEDUPPROPS);
252 DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
254 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
255 &stream_cksum, outfd) == -1)
257 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
258 DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
259 int sz = drr->drr_payloadlen;
265 (void) ssread(buf, sz, ofp);
268 if (cksum_and_write(buf, sz, &stream_cksum,
277 /* use the recalculated checksum */
278 ZIO_SET_CHECKSUM(&drre->drr_checksum,
279 stream_cksum.zc_word[0], stream_cksum.zc_word[1],
280 stream_cksum.zc_word[2], stream_cksum.zc_word[3]);
281 if ((write(outfd, drr,
282 sizeof (dmu_replay_record_t))) == -1)
289 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
290 &stream_cksum, outfd) == -1)
292 if (drro->drr_bonuslen > 0) {
294 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
296 if (cksum_and_write(buf,
297 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
298 &stream_cksum, outfd) == -1)
306 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
307 &stream_cksum, outfd) == -1)
309 (void) ssread(buf, drrs->drr_length, ofp);
310 if (cksum_and_write(buf, drrs->drr_length,
311 &stream_cksum, outfd) == -1)
316 case DRR_FREEOBJECTS:
318 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
319 &stream_cksum, outfd) == -1)
328 (void) ssread(buf, drrw->drr_length, ofp);
331 * Use the existing checksum if it's dedup-capable,
332 * else calculate a SHA256 checksum for it.
335 if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
337 !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
338 zio_cksum_t tmpsha256;
340 zio_checksum_SHA256(buf,
341 drrw->drr_length, &tmpsha256);
343 drrw->drr_key.ddk_cksum.zc_word[0] =
344 BE_64(tmpsha256.zc_word[0]);
345 drrw->drr_key.ddk_cksum.zc_word[1] =
346 BE_64(tmpsha256.zc_word[1]);
347 drrw->drr_key.ddk_cksum.zc_word[2] =
348 BE_64(tmpsha256.zc_word[2]);
349 drrw->drr_key.ddk_cksum.zc_word[3] =
350 BE_64(tmpsha256.zc_word[3]);
351 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
352 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
355 dataref.ref_guid = drrw->drr_toguid;
356 dataref.ref_object = drrw->drr_object;
357 dataref.ref_offset = drrw->drr_offset;
359 if (ddt_update(dda->dedup_hdl, &ddt,
360 &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
362 /* block already present in stream */
363 wbr_drrr->drr_object = drrw->drr_object;
364 wbr_drrr->drr_offset = drrw->drr_offset;
365 wbr_drrr->drr_length = drrw->drr_length;
366 wbr_drrr->drr_toguid = drrw->drr_toguid;
367 wbr_drrr->drr_refguid = dataref.ref_guid;
368 wbr_drrr->drr_refobject =
370 wbr_drrr->drr_refoffset =
373 wbr_drrr->drr_checksumtype =
374 drrw->drr_checksumtype;
375 wbr_drrr->drr_checksumflags =
376 drrw->drr_checksumtype;
377 wbr_drrr->drr_key.ddk_cksum =
378 drrw->drr_key.ddk_cksum;
379 wbr_drrr->drr_key.ddk_prop =
380 drrw->drr_key.ddk_prop;
382 if (cksum_and_write(&wbr_drr,
383 sizeof (dmu_replay_record_t), &stream_cksum,
387 /* block not previously seen */
388 if (cksum_and_write(drr,
389 sizeof (dmu_replay_record_t), &stream_cksum,
392 if (cksum_and_write(buf,
394 &stream_cksum, outfd) == -1)
402 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
403 &stream_cksum, outfd) == -1)
409 (void) printf("INVALID record type 0x%x\n",
411 /* should never happen, so assert */
416 umem_cache_destroy(ddt.ddecache);
417 free(ddt.dedup_hash_array);
425 * Routines for dealing with the AVL tree of fs-nvlists
427 typedef struct fsavl_node {
435 fsavl_compare(const void *arg1, const void *arg2)
437 const fsavl_node_t *fn1 = arg1;
438 const fsavl_node_t *fn2 = arg2;
440 if (fn1->fn_guid > fn2->fn_guid)
442 else if (fn1->fn_guid < fn2->fn_guid)
449 * Given the GUID of a snapshot, find its containing filesystem and
453 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
455 fsavl_node_t fn_find;
458 fn_find.fn_guid = snapguid;
460 fn = avl_find(avl, &fn_find, NULL);
463 *snapname = fn->fn_snapname;
464 return (fn->fn_nvfs);
470 fsavl_destroy(avl_tree_t *avl)
479 while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
486 * Given an nvlist, produce an avl tree of snapshots, ordered by guid
489 fsavl_create(nvlist_t *fss)
492 nvpair_t *fselem = NULL;
494 if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
497 avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
498 offsetof(fsavl_node_t, fn_node));
500 while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
501 nvlist_t *nvfs, *snaps;
502 nvpair_t *snapelem = NULL;
504 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
505 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
508 nvlist_next_nvpair(snaps, snapelem)) != NULL) {
512 VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
513 if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
514 fsavl_destroy(fsavl);
518 fn->fn_snapname = nvpair_name(snapelem);
522 * Note: if there are multiple snaps with the
523 * same GUID, we ignore all but one.
525 if (avl_find(fsavl, fn, NULL) == NULL)
536 * Routines for dealing with the giant nvlist of fs-nvlists, etc.
538 typedef struct send_data {
539 uint64_t parent_fromsnap_guid;
540 nvlist_t *parent_snaps;
543 const char *fromsnap;
548 * The header nvlist is of the following format:
551 * "fromsnap" -> string (if incremental)
555 * "name" -> string (full name; for debugging)
556 * "parentfromsnap" -> number (guid of fromsnap in parent)
558 * "props" -> { name -> value (only if set here) }
559 * "snaps" -> { name (lastname) -> number (guid) }
560 * "snapprops" -> { name (lastname) -> { name -> value } }
562 * "origin" -> number (guid) (if clone)
563 * "sent" -> boolean (not on-disk)
571 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
574 send_iterate_snap(zfs_handle_t *zhp, void *arg)
576 send_data_t *sd = arg;
577 uint64_t guid = zhp->zfs_dmustats.dds_guid;
581 snapname = strrchr(zhp->zfs_name, '@')+1;
583 VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
585 * NB: if there is no fromsnap here (it's a newly created fs in
586 * an incremental replication), we will substitute the tosnap.
588 if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
589 (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
590 strcmp(snapname, sd->tosnap) == 0)) {
591 sd->parent_fromsnap_guid = guid;
594 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
595 send_iterate_prop(zhp, nv);
596 VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
604 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
606 nvpair_t *elem = NULL;
608 while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
609 char *propname = nvpair_name(elem);
610 zfs_prop_t prop = zfs_name_to_prop(propname);
613 if (!zfs_prop_user(propname)) {
615 * Realistically, this should never happen. However,
616 * we want the ability to add DSL properties without
617 * needing to make incompatible version changes. We
618 * need to ignore unknown properties to allow older
619 * software to still send datasets containing these
620 * properties, with the unknown properties elided.
622 if (prop == ZPROP_INVAL)
625 if (zfs_prop_readonly(prop))
629 verify(nvpair_value_nvlist(elem, &propnv) == 0);
630 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
631 prop == ZFS_PROP_REFQUOTA ||
632 prop == ZFS_PROP_REFRESERVATION) {
635 verify(nvlist_lookup_uint64(propnv,
636 ZPROP_VALUE, &value) == 0);
637 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
640 * May have no source before SPA_VERSION_RECVD_PROPS,
641 * but is still modifiable.
643 if (nvlist_lookup_string(propnv,
644 ZPROP_SOURCE, &source) == 0) {
645 if ((strcmp(source, zhp->zfs_name) != 0) &&
647 ZPROP_SOURCE_VAL_RECVD) != 0))
652 if (nvlist_lookup_string(propnv,
653 ZPROP_SOURCE, &source) != 0)
655 if ((strcmp(source, zhp->zfs_name) != 0) &&
656 (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
660 if (zfs_prop_user(propname) ||
661 zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
663 verify(nvlist_lookup_string(propnv,
664 ZPROP_VALUE, &value) == 0);
665 VERIFY(0 == nvlist_add_string(nv, propname, value));
668 verify(nvlist_lookup_uint64(propnv,
669 ZPROP_VALUE, &value) == 0);
670 VERIFY(0 == nvlist_add_uint64(nv, propname, value));
676 * recursively generate nvlists describing datasets. See comment
677 * for the data structure send_data_t above for description of contents
681 send_iterate_fs(zfs_handle_t *zhp, void *arg)
683 send_data_t *sd = arg;
686 uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
687 uint64_t guid = zhp->zfs_dmustats.dds_guid;
690 VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
691 VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
692 VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
693 sd->parent_fromsnap_guid));
695 if (zhp->zfs_dmustats.dds_origin[0]) {
696 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
697 zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
700 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
701 origin->zfs_dmustats.dds_guid));
704 /* iterate over props */
705 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
706 send_iterate_prop(zhp, nv);
707 VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
710 /* iterate over snaps, and set sd->parent_fromsnap_guid */
711 sd->parent_fromsnap_guid = 0;
712 VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
713 VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
714 (void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
715 VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
716 VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
717 nvlist_free(sd->parent_snaps);
718 nvlist_free(sd->snapprops);
720 /* add this fs to nvlist */
721 (void) snprintf(guidstring, sizeof (guidstring),
722 "0x%llx", (longlong_t)guid);
723 VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
726 /* iterate over children */
728 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
730 sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
737 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
738 const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
741 send_data_t sd = { 0 };
744 zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
746 return (EZFS_BADTYPE);
748 VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
749 sd.fromsnap = fromsnap;
751 sd.recursive = recursive;
753 if ((error = send_iterate_fs(zhp, &sd)) != 0) {
761 if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
772 * Routines for dealing with the sorted snapshot functionality
774 typedef struct zfs_node {
775 zfs_handle_t *zn_handle;
776 avl_node_t zn_avlnode;
780 zfs_sort_snaps(zfs_handle_t *zhp, void *data)
782 avl_tree_t *avl = data;
786 search.zn_handle = zhp;
787 node = avl_find(avl, &search, NULL);
790 * If this snapshot was renamed while we were creating the
791 * AVL tree, it's possible that we already inserted it under
792 * its old name. Remove the old handle before adding the new
795 zfs_close(node->zn_handle);
796 avl_remove(avl, node);
800 node = zfs_alloc(zhp->zfs_hdl, sizeof (zfs_node_t));
801 node->zn_handle = zhp;
808 zfs_snapshot_compare(const void *larg, const void *rarg)
810 zfs_handle_t *l = ((zfs_node_t *)larg)->zn_handle;
811 zfs_handle_t *r = ((zfs_node_t *)rarg)->zn_handle;
812 uint64_t lcreate, rcreate;
815 * Sort them according to creation time. We use the hidden
816 * CREATETXG property to get an absolute ordering of snapshots.
818 lcreate = zfs_prop_get_int(l, ZFS_PROP_CREATETXG);
819 rcreate = zfs_prop_get_int(r, ZFS_PROP_CREATETXG);
821 if (lcreate < rcreate)
823 else if (lcreate > rcreate)
830 zfs_iter_snapshots_sorted(zfs_handle_t *zhp, zfs_iter_f callback, void *data)
837 avl_create(&avl, zfs_snapshot_compare,
838 sizeof (zfs_node_t), offsetof(zfs_node_t, zn_avlnode));
840 ret = zfs_iter_snapshots(zhp, zfs_sort_snaps, &avl);
842 for (node = avl_first(&avl); node != NULL; node = AVL_NEXT(&avl, node))
843 ret |= callback(node->zn_handle, data);
845 while ((node = avl_destroy_nodes(&avl, &cookie)) != NULL)
854 * Routines specific to "zfs send"
856 typedef struct send_dump_data {
857 /* these are all just the short snapname (the part after the @) */
858 const char *fromsnap;
860 char prevsnap[ZFS_MAXNAMELEN];
861 uint64_t prevsnap_obj;
862 boolean_t seenfrom, seento, replicate, doall, fromorigin;
868 snapfilter_cb_t *filter_cb;
871 char holdtag[ZFS_MAXNAMELEN];
876 * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
877 * NULL) to the file descriptor specified by outfd.
880 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
881 boolean_t fromorigin, int outfd, nvlist_t *debugnv)
883 zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
884 libzfs_handle_t *hdl = zhp->zfs_hdl;
887 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
888 assert(fromsnap_obj == 0 || !fromorigin);
890 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
891 zc.zc_cookie = outfd;
892 zc.zc_obj = fromorigin;
893 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
894 zc.zc_fromobj = fromsnap_obj;
896 VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
897 if (fromsnap && fromsnap[0] != '\0') {
898 VERIFY(0 == nvlist_add_string(thisdbg,
899 "fromsnap", fromsnap));
902 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_SEND, &zc) != 0) {
904 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
905 "warning: cannot send '%s'"), zhp->zfs_name);
907 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
909 VERIFY(0 == nvlist_add_nvlist(debugnv,
910 zhp->zfs_name, thisdbg));
912 nvlist_free(thisdbg);
917 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
918 "not an earlier snapshot from the same fs"));
919 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
922 if (zfs_dataset_exists(hdl, zc.zc_name,
923 ZFS_TYPE_SNAPSHOT)) {
924 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
925 "incremental source (@%s) does not exist"),
928 return (zfs_error(hdl, EZFS_NOENT, errbuf));
941 zfs_error_aux(hdl, strerror(errno));
942 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
945 return (zfs_standard_error(hdl, errno, errbuf));
950 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
951 nvlist_free(thisdbg);
957 hold_for_send(zfs_handle_t *zhp, send_dump_data_t *sdd)
963 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
966 * zfs_send() only opens a cleanup_fd for sends that need it,
967 * e.g. replication and doall.
969 if (sdd->cleanup_fd == -1)
972 thissnap = strchr(zhp->zfs_name, '@') + 1;
973 *(thissnap - 1) = '\0';
974 pzhp = zfs_open(zhp->zfs_hdl, zhp->zfs_name, ZFS_TYPE_DATASET);
975 *(thissnap - 1) = '@';
978 * It's OK if the parent no longer exists. The send code will
982 error = zfs_hold(pzhp, thissnap, sdd->holdtag,
983 B_FALSE, B_TRUE, B_TRUE, sdd->cleanup_fd,
984 zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID),
985 zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG));
993 dump_snapshot(zfs_handle_t *zhp, void *arg)
995 send_dump_data_t *sdd = arg;
998 boolean_t isfromsnap, istosnap;
999 boolean_t exclude = B_FALSE;
1001 thissnap = strchr(zhp->zfs_name, '@') + 1;
1002 isfromsnap = (sdd->fromsnap != NULL &&
1003 strcmp(sdd->fromsnap, thissnap) == 0);
1005 if (!sdd->seenfrom && isfromsnap) {
1006 err = hold_for_send(zhp, sdd);
1008 sdd->seenfrom = B_TRUE;
1009 (void) strcpy(sdd->prevsnap, thissnap);
1010 sdd->prevsnap_obj = zfs_prop_get_int(zhp,
1012 } else if (err == ENOENT) {
1019 if (sdd->seento || !sdd->seenfrom) {
1024 istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1026 sdd->seento = B_TRUE;
1028 if (!sdd->doall && !isfromsnap && !istosnap) {
1029 if (sdd->replicate) {
1031 nvlist_t *snapprops;
1033 * Filter out all intermediate snapshots except origin
1034 * snapshots needed to replicate clones.
1036 nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1037 zhp->zfs_dmustats.dds_guid, &snapname);
1039 VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1040 "snapprops", &snapprops));
1041 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1042 thissnap, &snapprops));
1043 exclude = !nvlist_exists(snapprops, "is_clone_origin");
1050 * If a filter function exists, call it to determine whether
1051 * this snapshot will be sent.
1053 if (exclude || (sdd->filter_cb != NULL &&
1054 sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1056 * This snapshot is filtered out. Don't send it, and don't
1057 * set prevsnap_obj, so it will be as if this snapshot didn't
1058 * exist, and the next accepted snapshot will be sent as
1059 * an incremental from the last accepted one, or as the
1060 * first (and full) snapshot in the case of a replication,
1061 * non-incremental send.
1067 err = hold_for_send(zhp, sdd);
1077 (void) fprintf(stderr, "sending from @%s to %s\n",
1078 sdd->prevsnap, zhp->zfs_name);
1081 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1082 sdd->prevsnap[0] == '\0' && (sdd->fromorigin || sdd->replicate),
1083 sdd->outfd, sdd->debugnv);
1085 (void) strcpy(sdd->prevsnap, thissnap);
1086 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1092 dump_filesystem(zfs_handle_t *zhp, void *arg)
1095 send_dump_data_t *sdd = arg;
1096 boolean_t missingfrom = B_FALSE;
1097 zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1099 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1100 zhp->zfs_name, sdd->tosnap);
1101 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1102 (void) fprintf(stderr, "WARNING: "
1103 "could not send %s@%s: does not exist\n",
1104 zhp->zfs_name, sdd->tosnap);
1109 if (sdd->replicate && sdd->fromsnap) {
1111 * If this fs does not have fromsnap, and we're doing
1112 * recursive, we need to send a full stream from the
1113 * beginning (or an incremental from the origin if this
1114 * is a clone). If we're doing non-recursive, then let
1115 * them get the error.
1117 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1118 zhp->zfs_name, sdd->fromsnap);
1119 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1120 ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1121 missingfrom = B_TRUE;
1125 sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1126 sdd->prevsnap_obj = 0;
1127 if (sdd->fromsnap == NULL || missingfrom)
1128 sdd->seenfrom = B_TRUE;
1130 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1131 if (!sdd->seenfrom) {
1132 (void) fprintf(stderr,
1133 "WARNING: could not send %s@%s:\n"
1134 "incremental source (%s@%s) does not exist\n",
1135 zhp->zfs_name, sdd->tosnap,
1136 zhp->zfs_name, sdd->fromsnap);
1138 } else if (!sdd->seento) {
1139 if (sdd->fromsnap) {
1140 (void) fprintf(stderr,
1141 "WARNING: could not send %s@%s:\n"
1142 "incremental source (%s@%s) "
1143 "is not earlier than it\n",
1144 zhp->zfs_name, sdd->tosnap,
1145 zhp->zfs_name, sdd->fromsnap);
1147 (void) fprintf(stderr, "WARNING: "
1148 "could not send %s@%s: does not exist\n",
1149 zhp->zfs_name, sdd->tosnap);
1158 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1160 send_dump_data_t *sdd = arg;
1162 boolean_t needagain, progress;
1164 if (!sdd->replicate)
1165 return (dump_filesystem(rzhp, sdd));
1167 /* Mark the clone origin snapshots. */
1168 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1169 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1171 uint64_t origin_guid = 0;
1173 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1174 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1175 if (origin_guid != 0) {
1177 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1178 origin_guid, &snapname);
1179 if (origin_nv != NULL) {
1180 nvlist_t *snapprops;
1181 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1182 "snapprops", &snapprops));
1183 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1184 snapname, &snapprops));
1185 VERIFY(0 == nvlist_add_boolean(
1186 snapprops, "is_clone_origin"));
1191 needagain = progress = B_FALSE;
1192 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1193 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1198 uint64_t origin_guid = 0;
1200 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1201 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1204 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1205 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1207 if (origin_guid != 0) {
1208 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1210 if (origin_nv != NULL &&
1211 nvlist_lookup_boolean(origin_nv,
1212 "sent") == ENOENT) {
1214 * origin has not been sent yet;
1222 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1225 err = dump_filesystem(zhp, sdd);
1226 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1240 * Generate a send stream for the dataset identified by the argument zhp.
1242 * The content of the send stream is the snapshot identified by
1243 * 'tosnap'. Incremental streams are requested in two ways:
1244 * - from the snapshot identified by "fromsnap" (if non-null) or
1245 * - from the origin of the dataset identified by zhp, which must
1246 * be a clone. In this case, "fromsnap" is null and "fromorigin"
1249 * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1250 * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1251 * if "replicate" is set. If "doall" is set, dump all the intermediate
1252 * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1253 * case too. If "props" is set, send properties.
1256 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1257 sendflags_t flags, int outfd, snapfilter_cb_t filter_func,
1258 void *cb_arg, nvlist_t **debugnvp)
1261 send_dump_data_t sdd = { 0 };
1263 nvlist_t *fss = NULL;
1264 avl_tree_t *fsavl = NULL;
1265 static uint64_t holdseq;
1267 boolean_t holdsnaps = B_FALSE;
1270 dedup_arg_t dda = { 0 };
1271 int featureflags = 0;
1273 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1274 "cannot send '%s'"), zhp->zfs_name);
1276 if (fromsnap && fromsnap[0] == '\0') {
1277 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1278 "zero-length incremental source"));
1279 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1282 if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1284 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1285 if (version >= ZPL_VERSION_SA) {
1286 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1290 if (zfs_spa_version(zhp, &spa_version) == 0 &&
1291 spa_version >= SPA_VERSION_USERREFS &&
1292 (flags.doall || flags.replicate))
1296 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1297 DMU_BACKUP_FEATURE_DEDUPPROPS);
1298 if ((err = pipe(pipefd))) {
1299 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1300 return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1303 dda.outputfd = outfd;
1304 dda.inputfd = pipefd[1];
1305 dda.dedup_hdl = zhp->zfs_hdl;
1306 if ((err = pthread_create(&tid, NULL, cksummer, &dda))) {
1307 (void) close(pipefd[0]);
1308 (void) close(pipefd[1]);
1309 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1310 return (zfs_error(zhp->zfs_hdl,
1311 EZFS_THREADCREATEFAILED, errbuf));
1315 if (flags.replicate || flags.doall || flags.props) {
1316 dmu_replay_record_t drr = { 0 };
1317 char *packbuf = NULL;
1319 zio_cksum_t zc = { { 0 } };
1321 if (flags.replicate || flags.props) {
1324 VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1326 VERIFY(0 == nvlist_add_string(hdrnv,
1327 "fromsnap", fromsnap));
1329 VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1330 if (!flags.replicate) {
1331 VERIFY(0 == nvlist_add_boolean(hdrnv,
1335 err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1336 fromsnap, tosnap, flags.replicate, &fss, &fsavl);
1339 VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1340 err = nvlist_pack(hdrnv, &packbuf, &buflen,
1347 fsavl_destroy(fsavl);
1353 /* write first begin record */
1354 drr.drr_type = DRR_BEGIN;
1355 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1356 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.drr_versioninfo,
1357 DMU_COMPOUNDSTREAM);
1358 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.drr_versioninfo,
1360 (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1361 sizeof (drr.drr_u.drr_begin.drr_toname),
1362 "%s@%s", zhp->zfs_name, tosnap);
1363 drr.drr_payloadlen = buflen;
1364 err = cksum_and_write(&drr, sizeof (drr), &zc, outfd);
1366 /* write header nvlist */
1367 if (err != -1 && packbuf != NULL) {
1368 err = cksum_and_write(packbuf, buflen, &zc, outfd);
1372 fsavl_destroy(fsavl);
1378 /* write end record */
1380 bzero(&drr, sizeof (drr));
1381 drr.drr_type = DRR_END;
1382 drr.drr_u.drr_end.drr_checksum = zc;
1383 err = write(outfd, &drr, sizeof (drr));
1385 fsavl_destroy(fsavl);
1393 /* dump each stream */
1394 sdd.fromsnap = fromsnap;
1395 sdd.tosnap = tosnap;
1397 sdd.outfd = pipefd[0];
1400 sdd.replicate = flags.replicate;
1401 sdd.doall = flags.doall;
1402 sdd.fromorigin = flags.fromorigin;
1405 sdd.verbose = flags.verbose;
1406 sdd.filter_cb = filter_func;
1407 sdd.filter_cb_arg = cb_arg;
1409 sdd.debugnv = *debugnvp;
1412 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1413 ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1414 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR);
1415 if (sdd.cleanup_fd < 0) {
1420 sdd.cleanup_fd = -1;
1422 err = dump_filesystems(zhp, &sdd);
1423 fsavl_destroy(fsavl);
1427 (void) close(pipefd[0]);
1428 (void) pthread_join(tid, NULL);
1431 if (sdd.cleanup_fd != -1) {
1432 VERIFY(0 == close(sdd.cleanup_fd));
1433 sdd.cleanup_fd = -1;
1436 if (flags.replicate || flags.doall || flags.props) {
1438 * write final end record. NB: want to do this even if
1439 * there was some error, because it might not be totally
1442 dmu_replay_record_t drr = { 0 };
1443 drr.drr_type = DRR_END;
1444 if (write(outfd, &drr, sizeof (drr)) == -1) {
1445 return (zfs_standard_error(zhp->zfs_hdl,
1450 return (err || sdd.err);
1453 err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1455 if (sdd.cleanup_fd != -1)
1456 VERIFY(0 == close(sdd.cleanup_fd));
1458 (void) pthread_cancel(tid);
1459 (void) pthread_join(tid, NULL);
1460 (void) close(pipefd[0]);
1466 * Routines specific to "zfs recv"
1470 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1471 boolean_t byteswap, zio_cksum_t *zc)
1478 rv = read(fd, cp, len);
1483 if (rv < 0 || len != 0) {
1484 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1485 "failed to read from stream"));
1486 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1487 "cannot receive")));
1492 fletcher_4_incremental_byteswap(buf, ilen, zc);
1494 fletcher_4_incremental_native(buf, ilen, zc);
1500 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1501 boolean_t byteswap, zio_cksum_t *zc)
1506 buf = zfs_alloc(hdl, len);
1510 err = recv_read(hdl, fd, buf, len, byteswap, zc);
1516 err = nvlist_unpack(buf, len, nvp, 0);
1519 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1520 "stream (malformed nvlist)"));
1527 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1528 int baselen, char *newname, recvflags_t flags)
1531 zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1533 prop_changelist_t *clp;
1536 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1539 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1540 flags.force ? MS_FORCE : 0);
1544 err = changelist_prefix(clp);
1548 zc.zc_objset_type = DMU_OST_ZFS;
1549 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1552 (void) strcpy(newname, tryname);
1554 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1556 if (flags.verbose) {
1557 (void) printf("attempting rename %s to %s\n",
1558 zc.zc_name, zc.zc_value);
1560 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1562 changelist_rename(clp, name, tryname);
1567 if (err != 0 && strncmp(name+baselen, "recv-", 5) != 0) {
1570 (void) strncpy(newname, name, baselen);
1571 (void) snprintf(newname+baselen, ZFS_MAXNAMELEN-baselen,
1572 "recv-%ld-%u", (long) getpid(), seq);
1573 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1575 if (flags.verbose) {
1576 (void) printf("failed - trying rename %s to %s\n",
1577 zc.zc_name, zc.zc_value);
1579 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1581 changelist_rename(clp, name, newname);
1582 if (err && flags.verbose) {
1583 (void) printf("failed (%u) - "
1584 "will try again on next pass\n", errno);
1587 } else if (flags.verbose) {
1589 (void) printf("success\n");
1591 (void) printf("failed (%u)\n", errno);
1594 (void) changelist_postfix(clp);
1595 changelist_free(clp);
1601 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1602 char *newname, recvflags_t flags)
1604 zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1606 prop_changelist_t *clp;
1608 boolean_t defer = B_FALSE;
1611 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1614 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1615 flags.force ? MS_FORCE : 0);
1616 if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1617 zfs_spa_version(zhp, &spa_version) == 0 &&
1618 spa_version >= SPA_VERSION_USERREFS)
1623 err = changelist_prefix(clp);
1627 zc.zc_objset_type = DMU_OST_ZFS;
1628 zc.zc_defer_destroy = defer;
1629 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1632 (void) printf("attempting destroy %s\n", zc.zc_name);
1633 err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1636 (void) printf("success\n");
1637 changelist_remove(clp, zc.zc_name);
1640 (void) changelist_postfix(clp);
1641 changelist_free(clp);
1644 * Deferred destroy might destroy the snapshot or only mark it to be
1645 * destroyed later, and it returns success in either case.
1647 if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1648 ZFS_TYPE_SNAPSHOT))) {
1649 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1655 typedef struct guid_to_name_data {
1658 } guid_to_name_data_t;
1661 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1663 guid_to_name_data_t *gtnd = arg;
1666 if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1667 (void) strcpy(gtnd->name, zhp->zfs_name);
1671 err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1677 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1680 /* exhaustive search all local snapshots */
1681 guid_to_name_data_t gtnd;
1689 if (strchr(parent, '@') == NULL) {
1690 zhp = make_dataset_handle(hdl, parent);
1692 err = zfs_iter_children(zhp, guid_to_name_cb, >nd);
1699 cp = strchr(parent, '/');
1702 zhp = make_dataset_handle(hdl, parent);
1707 err = zfs_iter_children(zhp, guid_to_name_cb, >nd);
1711 return (err == EEXIST ? 0 : ENOENT);
1716 * Return true if dataset guid1 is created before guid2.
1719 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1720 uint64_t guid1, uint64_t guid2)
1723 char *fsname, *snapname;
1724 char buf[ZFS_MAXNAMELEN];
1726 zfs_node_t zn1, zn2;
1733 nvfs = fsavl_find(avl, guid1, &snapname);
1734 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1735 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1736 zn1.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1737 if (zn1.zn_handle == NULL)
1740 nvfs = fsavl_find(avl, guid2, &snapname);
1741 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1742 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1743 zn2.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1744 if (zn2.zn_handle == NULL) {
1745 zfs_close(zn2.zn_handle);
1749 rv = (zfs_snapshot_compare(&zn1, &zn2) == -1);
1751 zfs_close(zn1.zn_handle);
1752 zfs_close(zn2.zn_handle);
1758 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
1759 recvflags_t flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
1763 avl_tree_t *local_avl;
1764 nvpair_t *fselem, *nextfselem;
1766 char newname[ZFS_MAXNAMELEN];
1768 boolean_t needagain, progress, recursive;
1771 VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
1773 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
1780 needagain = progress = B_FALSE;
1782 if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
1783 recursive, &local_nv, &local_avl)) != 0)
1787 * Process deletes and renames
1789 for (fselem = nvlist_next_nvpair(local_nv, NULL);
1790 fselem; fselem = nextfselem) {
1791 nvlist_t *nvfs, *snaps;
1792 nvlist_t *stream_nvfs = NULL;
1793 nvpair_t *snapelem, *nextsnapelem;
1794 uint64_t fromguid = 0;
1795 uint64_t originguid = 0;
1796 uint64_t stream_originguid = 0;
1797 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
1798 char *fsname, *stream_fsname;
1800 nextfselem = nvlist_next_nvpair(local_nv, fselem);
1802 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
1803 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
1804 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1805 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
1806 &parent_fromsnap_guid));
1807 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
1810 * First find the stream's fs, so we can check for
1811 * a different origin (due to "zfs promote")
1813 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1814 snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
1817 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1818 stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
1820 if (stream_nvfs != NULL)
1824 /* check for promote */
1825 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
1826 &stream_originguid);
1827 if (stream_nvfs && originguid != stream_originguid) {
1828 switch (created_before(hdl, local_avl,
1829 stream_originguid, originguid)) {
1832 zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1833 nvlist_t *origin_nvfs;
1834 char *origin_fsname;
1837 (void) printf("promoting %s\n", fsname);
1839 origin_nvfs = fsavl_find(local_avl, originguid,
1841 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
1842 "name", &origin_fsname));
1843 (void) strlcpy(zc.zc_value, origin_fsname,
1844 sizeof (zc.zc_value));
1845 (void) strlcpy(zc.zc_name, fsname,
1846 sizeof (zc.zc_name));
1847 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
1855 fsavl_destroy(local_avl);
1856 nvlist_free(local_nv);
1860 * We had/have the wrong origin, therefore our
1861 * list of snapshots is wrong. Need to handle
1862 * them on the next pass.
1868 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1869 snapelem; snapelem = nextsnapelem) {
1871 char *stream_snapname;
1872 nvlist_t *found, *props;
1874 nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
1876 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1877 found = fsavl_find(stream_avl, thisguid,
1880 /* check for delete */
1881 if (found == NULL) {
1882 char name[ZFS_MAXNAMELEN];
1887 (void) snprintf(name, sizeof (name), "%s@%s",
1888 fsname, nvpair_name(snapelem));
1890 error = recv_destroy(hdl, name,
1891 strlen(fsname)+1, newname, flags);
1899 stream_nvfs = found;
1901 if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
1902 &props) && 0 == nvlist_lookup_nvlist(props,
1903 stream_snapname, &props)) {
1904 zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1906 zc.zc_cookie = B_TRUE; /* received */
1907 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
1908 "%s@%s", fsname, nvpair_name(snapelem));
1909 if (zcmd_write_src_nvlist(hdl, &zc,
1911 (void) zfs_ioctl(hdl,
1912 ZFS_IOC_SET_PROP, &zc);
1913 zcmd_free_nvlists(&zc);
1917 /* check for different snapname */
1918 if (strcmp(nvpair_name(snapelem),
1919 stream_snapname) != 0) {
1920 char name[ZFS_MAXNAMELEN];
1921 char tryname[ZFS_MAXNAMELEN];
1923 (void) snprintf(name, sizeof (name), "%s@%s",
1924 fsname, nvpair_name(snapelem));
1925 (void) snprintf(tryname, sizeof (name), "%s@%s",
1926 fsname, stream_snapname);
1928 error = recv_rename(hdl, name, tryname,
1929 strlen(fsname)+1, newname, flags);
1936 if (strcmp(stream_snapname, fromsnap) == 0)
1937 fromguid = thisguid;
1940 /* check for delete */
1941 if (stream_nvfs == NULL) {
1945 error = recv_destroy(hdl, fsname, strlen(tofs)+1,
1954 if (fromguid == 0) {
1955 if (flags.verbose) {
1956 (void) printf("local fs %s does not have "
1957 "fromsnap (%s in stream); must have "
1958 "been deleted locally; ignoring\n",
1964 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
1965 "name", &stream_fsname));
1966 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
1967 "parentfromsnap", &stream_parent_fromsnap_guid));
1969 s1 = strrchr(fsname, '/');
1970 s2 = strrchr(stream_fsname, '/');
1973 * Check for rename. If the exact receive path is specified, it
1974 * does not count as a rename, but we still need to check the
1975 * datasets beneath it.
1977 if ((stream_parent_fromsnap_guid != 0 &&
1978 parent_fromsnap_guid != 0 &&
1979 stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
1980 ((flags.isprefix || strcmp(tofs, fsname) != 0) &&
1981 (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
1983 char tryname[ZFS_MAXNAMELEN];
1985 parent = fsavl_find(local_avl,
1986 stream_parent_fromsnap_guid, NULL);
1988 * NB: parent might not be found if we used the
1989 * tosnap for stream_parent_fromsnap_guid,
1990 * because the parent is a newly-created fs;
1991 * we'll be able to rename it after we recv the
1994 if (parent != NULL) {
1997 VERIFY(0 == nvlist_lookup_string(parent, "name",
1999 (void) snprintf(tryname, sizeof (tryname),
2000 "%s%s", pname, strrchr(stream_fsname, '/'));
2003 if (flags.verbose) {
2004 (void) printf("local fs %s new parent "
2005 "not found\n", fsname);
2011 error = recv_rename(hdl, fsname, tryname,
2012 strlen(tofs)+1, newname, flags);
2014 if (renamed != NULL && newname[0] != '\0') {
2015 VERIFY(0 == nvlist_add_boolean(renamed,
2026 fsavl_destroy(local_avl);
2027 nvlist_free(local_nv);
2029 if (needagain && progress) {
2030 /* do another pass to fix up temporary names */
2032 (void) printf("another pass:\n");
2040 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2041 recvflags_t flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2042 char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2044 nvlist_t *stream_nv = NULL;
2045 avl_tree_t *stream_avl = NULL;
2046 char *fromsnap = NULL;
2048 char tofs[ZFS_MAXNAMELEN];
2049 char sendfs[ZFS_MAXNAMELEN];
2051 dmu_replay_record_t drre;
2053 boolean_t anyerr = B_FALSE;
2054 boolean_t softerr = B_FALSE;
2055 boolean_t recursive;
2057 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2060 assert(drr->drr_type == DRR_BEGIN);
2061 assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2062 assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2063 DMU_COMPOUNDSTREAM);
2066 * Read in the nvlist from the stream.
2068 if (drr->drr_payloadlen != 0) {
2069 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2070 &stream_nv, flags.byteswap, zc);
2072 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2077 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2080 if (recursive && strchr(destname, '@')) {
2081 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2082 "cannot specify snapshot name for multi-snapshot stream"));
2083 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2088 * Read in the end record and verify checksum.
2090 if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2091 flags.byteswap, NULL)))
2093 if (flags.byteswap) {
2094 drre.drr_type = BSWAP_32(drre.drr_type);
2095 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2096 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2097 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2098 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2099 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2100 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2101 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2102 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2104 if (drre.drr_type != DRR_END) {
2105 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2108 if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2109 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2110 "incorrect header checksum"));
2111 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2115 (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2117 if (drr->drr_payloadlen != 0) {
2118 nvlist_t *stream_fss;
2120 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2122 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2123 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2124 "couldn't allocate avl tree"));
2125 error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2129 if (fromsnap != NULL) {
2130 nvlist_t *renamed = NULL;
2131 nvpair_t *pair = NULL;
2133 (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2134 if (flags.isprefix) {
2135 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2139 cp = strrchr(drrb->drr_toname, '/');
2141 (void) strlcat(tofs, "/",
2145 i = (cp - drrb->drr_toname);
2148 i = strcspn(drrb->drr_toname, "/@");
2150 /* zfs_receive_one() will create_parents() */
2151 (void) strlcat(tofs, &drrb->drr_toname[i],
2153 *strchr(tofs, '@') = '\0';
2156 if (recursive && !flags.dryrun && !flags.nomount) {
2157 VERIFY(0 == nvlist_alloc(&renamed,
2158 NV_UNIQUE_NAME, 0));
2161 softerr = recv_incremental_replication(hdl, tofs, flags,
2162 stream_nv, stream_avl, renamed);
2164 /* Unmount renamed filesystems before receiving. */
2165 while ((pair = nvlist_next_nvpair(renamed,
2168 prop_changelist_t *clp = NULL;
2170 zhp = zfs_open(hdl, nvpair_name(pair),
2171 ZFS_TYPE_FILESYSTEM);
2173 clp = changelist_gather(zhp,
2174 ZFS_PROP_MOUNTPOINT, 0, 0);
2178 changelist_prefix(clp);
2179 changelist_free(clp);
2184 nvlist_free(renamed);
2189 * Get the fs specified by the first path in the stream (the top level
2190 * specified by 'zfs send') and pass it to each invocation of
2191 * zfs_receive_one().
2193 (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2195 if ((cp = strchr(sendfs, '@')) != NULL)
2198 /* Finally, receive each contained stream */
2201 * we should figure out if it has a recoverable
2202 * error, in which case do a recv_skip() and drive on.
2203 * Note, if we fail due to already having this guid,
2204 * zfs_receive_one() will take care of it (ie,
2205 * recv_skip() and return 0).
2207 error = zfs_receive_impl(hdl, destname, flags, fd,
2208 sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2210 if (error == ENODATA) {
2215 } while (error == 0);
2217 if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2219 * Now that we have the fs's they sent us, try the
2222 softerr = recv_incremental_replication(hdl, tofs, flags,
2223 stream_nv, stream_avl, NULL);
2227 fsavl_destroy(stream_avl);
2229 nvlist_free(stream_nv);
2238 trunc_prop_errs(int truncated)
2240 ASSERT(truncated != 0);
2243 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2244 "1 more property could not be set\n"));
2246 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2247 "%d more properties could not be set\n"), truncated);
2251 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2253 dmu_replay_record_t *drr;
2254 void *buf = malloc(1<<20);
2257 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2258 "cannot receive:"));
2260 /* XXX would be great to use lseek if possible... */
2263 while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2264 byteswap, NULL) == 0) {
2266 drr->drr_type = BSWAP_32(drr->drr_type);
2268 switch (drr->drr_type) {
2270 /* NB: not to be used on v2 stream packages */
2271 if (drr->drr_payloadlen != 0) {
2272 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2273 "invalid substream header"));
2274 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2284 drr->drr_u.drr_object.drr_bonuslen =
2285 BSWAP_32(drr->drr_u.drr_object.
2288 (void) recv_read(hdl, fd, buf,
2289 P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2295 drr->drr_u.drr_write.drr_length =
2296 BSWAP_64(drr->drr_u.drr_write.drr_length);
2298 (void) recv_read(hdl, fd, buf,
2299 drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2303 drr->drr_u.drr_write.drr_length =
2304 BSWAP_64(drr->drr_u.drr_spill.drr_length);
2306 (void) recv_read(hdl, fd, buf,
2307 drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2309 case DRR_WRITE_BYREF:
2310 case DRR_FREEOBJECTS:
2315 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2316 "invalid record type"));
2317 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2326 * Restores a backup of tosnap from the file descriptor specified by infd.
2329 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2330 recvflags_t flags, dmu_replay_record_t *drr,
2331 dmu_replay_record_t *drr_noswap, const char *sendfs,
2332 nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2333 uint64_t *action_handlep)
2335 zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2337 int ioctl_err, ioctl_errno, err;
2339 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2341 char prop_errbuf[1024];
2342 const char *chopprefix;
2343 boolean_t newfs = B_FALSE;
2344 boolean_t stream_wantsnewfs;
2345 uint64_t parent_snapguid = 0;
2346 prop_changelist_t *clp = NULL;
2347 nvlist_t *snapprops_nvlist = NULL;
2348 zprop_errflags_t prop_errflags;
2349 boolean_t recursive;
2351 begin_time = time(NULL);
2353 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2356 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2359 if (stream_avl != NULL) {
2361 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2366 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2368 err = nvlist_lookup_nvlist(fs, "props", &props);
2370 VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2372 if (flags.canmountoff) {
2373 VERIFY(0 == nvlist_add_uint64(props,
2374 zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2376 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2380 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2381 VERIFY(0 == nvlist_lookup_nvlist(props,
2382 snapname, &snapprops_nvlist));
2392 * Determine how much of the snapshot name stored in the stream
2393 * we are going to tack on to the name they specified on the
2394 * command line, and how much we are going to chop off.
2396 * If they specified a snapshot, chop the entire name stored in
2401 * A filesystem was specified with -e. We want to tack on only
2402 * the tail of the sent snapshot path.
2404 if (strchr(tosnap, '@')) {
2405 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2406 "argument - snapshot not allowed with -e"));
2407 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2410 chopprefix = strrchr(sendfs, '/');
2412 if (chopprefix == NULL) {
2414 * The tail is the poolname, so we need to
2415 * prepend a path separator.
2417 int len = strlen(drrb->drr_toname);
2418 cp = malloc(len + 2);
2420 (void) strcpy(&cp[1], drrb->drr_toname);
2423 chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2425 } else if (flags.isprefix) {
2427 * A filesystem was specified with -d. We want to tack on
2428 * everything but the first element of the sent snapshot path
2429 * (all but the pool name).
2431 if (strchr(tosnap, '@')) {
2432 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2433 "argument - snapshot not allowed with -d"));
2434 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2437 chopprefix = strchr(drrb->drr_toname, '/');
2438 if (chopprefix == NULL)
2439 chopprefix = strchr(drrb->drr_toname, '@');
2440 } else if (strchr(tosnap, '@') == NULL) {
2442 * If a filesystem was specified without -d or -e, we want to
2443 * tack on everything after the fs specified by 'zfs send'.
2445 chopprefix = drrb->drr_toname + strlen(sendfs);
2447 /* A snapshot was specified as an exact path (no -d or -e). */
2449 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2450 "cannot specify snapshot name for multi-snapshot "
2452 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2454 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2457 ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2458 ASSERT(chopprefix > drrb->drr_toname);
2459 ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2460 ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2461 chopprefix[0] == '\0');
2464 * Determine name of destination snapshot, store in zc_value.
2466 (void) strcpy(zc.zc_top_ds, tosnap);
2467 (void) strcpy(zc.zc_value, tosnap);
2468 (void) strlcat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2470 if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2471 zcmd_free_nvlists(&zc);
2472 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2476 * Determine the name of the origin snapshot, store in zc_string.
2478 if (drrb->drr_flags & DRR_FLAG_CLONE) {
2479 if (guid_to_name(hdl, tosnap,
2480 drrb->drr_fromguid, zc.zc_string) != 0) {
2481 zcmd_free_nvlists(&zc);
2482 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2483 "local origin for clone %s does not exist"),
2485 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2488 (void) printf("found clone origin %s\n", zc.zc_string);
2491 stream_wantsnewfs = (drrb->drr_fromguid == 0 ||
2492 (drrb->drr_flags & DRR_FLAG_CLONE));
2494 if (stream_wantsnewfs) {
2496 * if the parent fs does not exist, look for it based on
2497 * the parent snap GUID
2499 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2500 "cannot receive new filesystem stream"));
2502 (void) strcpy(zc.zc_name, zc.zc_value);
2503 cp = strrchr(zc.zc_name, '/');
2507 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2508 char suffix[ZFS_MAXNAMELEN];
2509 (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2510 if (guid_to_name(hdl, tosnap, parent_snapguid,
2511 zc.zc_value) == 0) {
2512 *strchr(zc.zc_value, '@') = '\0';
2513 (void) strcat(zc.zc_value, suffix);
2518 * if the fs does not exist, look for it based on the
2521 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2522 "cannot receive incremental stream"));
2524 (void) strcpy(zc.zc_name, zc.zc_value);
2525 *strchr(zc.zc_name, '@') = '\0';
2528 * If the exact receive path was specified and this is the
2529 * topmost path in the stream, then if the fs does not exist we
2530 * should look no further.
2532 if ((flags.isprefix || (*(chopprefix = drrb->drr_toname +
2533 strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2534 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2535 char snap[ZFS_MAXNAMELEN];
2536 (void) strcpy(snap, strchr(zc.zc_value, '@'));
2537 if (guid_to_name(hdl, tosnap, drrb->drr_fromguid,
2538 zc.zc_value) == 0) {
2539 *strchr(zc.zc_value, '@') = '\0';
2540 (void) strcat(zc.zc_value, snap);
2545 (void) strcpy(zc.zc_name, zc.zc_value);
2546 *strchr(zc.zc_name, '@') = '\0';
2548 if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2552 * Destination fs exists. Therefore this should either
2553 * be an incremental, or the stream specifies a new fs
2554 * (full stream or clone) and they want us to blow it
2555 * away (and have therefore specified -F and removed any
2558 if (stream_wantsnewfs) {
2560 zcmd_free_nvlists(&zc);
2561 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2562 "destination '%s' exists\n"
2563 "must specify -F to overwrite it"),
2565 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2567 if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2569 zcmd_free_nvlists(&zc);
2570 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2571 "destination has snapshots (eg. %s)\n"
2572 "must destroy them to overwrite it"),
2574 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2578 if ((zhp = zfs_open(hdl, zc.zc_name,
2579 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2580 zcmd_free_nvlists(&zc);
2584 if (stream_wantsnewfs &&
2585 zhp->zfs_dmustats.dds_origin[0]) {
2586 zcmd_free_nvlists(&zc);
2588 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2589 "destination '%s' is a clone\n"
2590 "must destroy it to overwrite it"),
2592 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2595 if (!flags.dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2596 stream_wantsnewfs) {
2597 /* We can't do online recv in this case */
2598 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2601 zcmd_free_nvlists(&zc);
2604 if (changelist_prefix(clp) != 0) {
2605 changelist_free(clp);
2607 zcmd_free_nvlists(&zc);
2614 * Destination filesystem does not exist. Therefore we better
2615 * be creating a new filesystem (either from a full backup, or
2616 * a clone). It would therefore be invalid if the user
2617 * specified only the pool name (i.e. if the destination name
2618 * contained no slash character).
2620 if (!stream_wantsnewfs ||
2621 (cp = strrchr(zc.zc_name, '/')) == NULL) {
2622 zcmd_free_nvlists(&zc);
2623 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2624 "destination '%s' does not exist"), zc.zc_name);
2625 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2629 * Trim off the final dataset component so we perform the
2630 * recvbackup ioctl to the filesystems's parent.
2634 if (flags.isprefix && !flags.istail && !flags.dryrun &&
2635 create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2636 zcmd_free_nvlists(&zc);
2637 return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2643 zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2644 zc.zc_cookie = infd;
2645 zc.zc_guid = flags.force;
2646 if (flags.verbose) {
2647 (void) printf("%s %s stream of %s into %s\n",
2648 flags.dryrun ? "would receive" : "receiving",
2649 drrb->drr_fromguid ? "incremental" : "full",
2650 drrb->drr_toname, zc.zc_value);
2651 (void) fflush(stdout);
2655 zcmd_free_nvlists(&zc);
2656 return (recv_skip(hdl, infd, flags.byteswap));
2659 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2660 zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2661 zc.zc_cleanup_fd = cleanup_fd;
2662 zc.zc_action_handle = *action_handlep;
2664 err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2665 ioctl_errno = errno;
2666 prop_errflags = (zprop_errflags_t)zc.zc_obj;
2669 nvlist_t *prop_errors;
2670 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2671 zc.zc_nvlist_dst_size, &prop_errors, 0));
2673 nvpair_t *prop_err = NULL;
2675 while ((prop_err = nvlist_next_nvpair(prop_errors,
2676 prop_err)) != NULL) {
2681 prop = zfs_name_to_prop(nvpair_name(prop_err));
2682 (void) nvpair_value_int32(prop_err, &intval);
2683 if (strcmp(nvpair_name(prop_err),
2684 ZPROP_N_MORE_ERRORS) == 0) {
2685 trunc_prop_errs(intval);
2688 (void) snprintf(tbuf, sizeof (tbuf),
2689 dgettext(TEXT_DOMAIN,
2690 "cannot receive %s property on %s"),
2691 nvpair_name(prop_err), zc.zc_name);
2692 zfs_setprop_error(hdl, prop, intval, tbuf);
2695 nvlist_free(prop_errors);
2698 zc.zc_nvlist_dst = 0;
2699 zc.zc_nvlist_dst_size = 0;
2700 zcmd_free_nvlists(&zc);
2702 if (err == 0 && snapprops_nvlist) {
2703 zfs_cmd_t zc2 = { "\0", "\0", "\0", "\0", 0 };
2705 (void) strcpy(zc2.zc_name, zc.zc_value);
2706 zc2.zc_cookie = B_TRUE; /* received */
2707 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
2708 (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
2709 zcmd_free_nvlists(&zc2);
2713 if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
2715 * It may be that this snapshot already exists,
2716 * in which case we want to consume & ignore it
2717 * rather than failing.
2719 avl_tree_t *local_avl;
2720 nvlist_t *local_nv, *fs;
2721 cp = strchr(zc.zc_value, '@');
2724 * XXX Do this faster by just iterating over snaps in
2725 * this fs. Also if zc_value does not exist, we will
2726 * get a strange "does not exist" error message.
2729 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
2730 &local_nv, &local_avl) == 0) {
2732 fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
2733 fsavl_destroy(local_avl);
2734 nvlist_free(local_nv);
2737 if (flags.verbose) {
2738 (void) printf("snap %s already exists; "
2739 "ignoring\n", zc.zc_value);
2741 err = ioctl_err = recv_skip(hdl, infd,
2748 if (ioctl_err != 0) {
2749 switch (ioctl_errno) {
2751 cp = strchr(zc.zc_value, '@');
2753 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2754 "most recent snapshot of %s does not\n"
2755 "match incremental source"), zc.zc_value);
2756 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2760 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2761 "destination %s has been modified\n"
2762 "since most recent snapshot"), zc.zc_name);
2763 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2766 cp = strchr(zc.zc_value, '@');
2768 /* it's the containing fs that exists */
2771 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2772 "destination already exists"));
2773 (void) zfs_error_fmt(hdl, EZFS_EXISTS,
2774 dgettext(TEXT_DOMAIN, "cannot restore to %s"),
2779 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2782 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2783 "invalid stream (checksum mismatch)"));
2784 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2787 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2788 "pool must be upgraded to receive this stream."));
2789 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
2792 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2793 "destination %s space quota exceeded"), zc.zc_name);
2794 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2797 (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
2802 * Mount the target filesystem (if created). Also mount any
2803 * children of the target filesystem if we did a replication
2804 * receive (indicated by stream_avl being non-NULL).
2806 cp = strchr(zc.zc_value, '@');
2807 if (cp && (ioctl_err == 0 || !newfs)) {
2811 h = zfs_open(hdl, zc.zc_value,
2812 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
2814 if (h->zfs_type == ZFS_TYPE_VOLUME) {
2816 } else if (newfs || stream_avl) {
2818 * Track the first/top of hierarchy fs,
2819 * for mounting and sharing later.
2821 if (top_zfs && *top_zfs == NULL)
2822 *top_zfs = zfs_strdup(hdl, zc.zc_value);
2830 err |= changelist_postfix(clp);
2831 changelist_free(clp);
2834 if (prop_errflags & ZPROP_ERR_NOCLEAR) {
2835 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
2836 "failed to clear unreceived properties on %s"),
2838 (void) fprintf(stderr, "\n");
2840 if (prop_errflags & ZPROP_ERR_NORESTORE) {
2841 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
2842 "failed to restore original properties on %s"),
2844 (void) fprintf(stderr, "\n");
2847 if (err || ioctl_err)
2850 *action_handlep = zc.zc_action_handle;
2852 if (flags.verbose) {
2855 uint64_t bytes = zc.zc_cookie;
2856 time_t delta = time(NULL) - begin_time;
2859 zfs_nicenum(bytes, buf1, sizeof (buf1));
2860 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
2862 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
2870 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
2871 int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2872 char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2875 dmu_replay_record_t drr, drr_noswap;
2876 struct drr_begin *drrb = &drr.drr_u.drr_begin;
2878 zio_cksum_t zcksum = { { 0 } };
2879 uint64_t featureflags;
2882 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2885 if (flags.isprefix &&
2886 !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
2887 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
2888 "(%s) does not exist"), tosnap);
2889 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2892 /* read in the BEGIN record */
2893 if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
2897 if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
2898 /* It's the double end record at the end of a package */
2902 /* the kernel needs the non-byteswapped begin record */
2905 flags.byteswap = B_FALSE;
2906 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
2908 * We computed the checksum in the wrong byteorder in
2909 * recv_read() above; do it again correctly.
2911 bzero(&zcksum, sizeof (zio_cksum_t));
2912 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
2913 flags.byteswap = B_TRUE;
2915 drr.drr_type = BSWAP_32(drr.drr_type);
2916 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
2917 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
2918 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
2919 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
2920 drrb->drr_type = BSWAP_32(drrb->drr_type);
2921 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
2922 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
2923 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
2926 if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
2927 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2928 "stream (bad magic number)"));
2929 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2932 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
2933 hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
2935 if (!DMU_STREAM_SUPPORTED(featureflags) ||
2936 (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
2937 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2938 "stream has unsupported feature, feature flags = %lx"),
2940 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2943 if (strchr(drrb->drr_toname, '@') == NULL) {
2944 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2945 "stream (bad snapshot name)"));
2946 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2949 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
2950 char nonpackage_sendfs[ZFS_MAXNAMELEN];
2951 if (sendfs == NULL) {
2953 * We were not called from zfs_receive_package(). Get
2954 * the fs specified by 'zfs send'.
2957 (void) strlcpy(nonpackage_sendfs,
2958 drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
2959 if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
2961 sendfs = nonpackage_sendfs;
2963 return (zfs_receive_one(hdl, infd, tosnap, flags,
2964 &drr, &drr_noswap, sendfs, stream_nv, stream_avl,
2965 top_zfs, cleanup_fd, action_handlep));
2967 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
2968 DMU_COMPOUNDSTREAM);
2969 return (zfs_receive_package(hdl, infd, tosnap, flags,
2970 &drr, &zcksum, top_zfs, cleanup_fd, action_handlep));
2975 * Restores a backup of tosnap from the file descriptor specified by infd.
2976 * Return 0 on total success, -2 if some things couldn't be
2977 * destroyed/renamed/promoted, -1 if some things couldn't be received.
2978 * (-1 will override -2).
2981 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
2982 int infd, avl_tree_t *stream_avl)
2984 char *top_zfs = NULL;
2987 uint64_t action_handle = 0;
2989 cleanup_fd = open(ZFS_DEV, O_RDWR);
2990 VERIFY(cleanup_fd >= 0);
2992 err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL,
2993 stream_avl, &top_zfs, cleanup_fd, &action_handle);
2995 VERIFY(0 == close(cleanup_fd));
2997 if (err == 0 && !flags.nomount && top_zfs) {
2999 prop_changelist_t *clp;
3001 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3003 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3004 CL_GATHER_MOUNT_ALWAYS, 0);
3007 /* mount and share received datasets */
3008 err = changelist_postfix(clp);
3009 changelist_free(clp);
3012 if (zhp == NULL || clp == NULL || err)