1 /*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012 by Delphix. All rights reserved.
25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 */
27
28 #include <assert.h>
29 #include <ctype.h>
30 #include <errno.h>
31 #include <libintl.h>
32 #include <stdio.h>
33 #include <stdlib.h>
34 #include <strings.h>
35 #include <unistd.h>
36 #include <stddef.h>
37 #include <fcntl.h>
38 #include <sys/mount.h>
39 #include <pthread.h>
40 #include <umem.h>
41 #include <time.h>
42
43 #include <libzfs.h>
44
45 #include "zfs_namecheck.h"
46 #include "zfs_prop.h"
47 #include "zfs_fletcher.h"
48 #include "libzfs_impl.h"
49 #include <sha2.h>
50 #include <sys/zio_checksum.h>
51 #include <sys/ddt.h>
52
53 /* in libzfs_dataset.c */
54 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
55
56 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t *,
57 int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *);
58
59 static const zio_cksum_t zero_cksum = { 0 };
60
61 typedef struct dedup_arg {
62 int inputfd;
63 int outputfd;
64 libzfs_handle_t *dedup_hdl;
65 } dedup_arg_t;
66
67 typedef struct progress_arg {
68 zfs_handle_t *pa_zhp;
69 int pa_fd;
70 boolean_t pa_parsable;
71 } progress_arg_t;
72
73 typedef struct dataref {
74 uint64_t ref_guid;
75 uint64_t ref_object;
76 uint64_t ref_offset;
77 } dataref_t;
78
79 typedef struct dedup_entry {
80 struct dedup_entry *dde_next;
81 zio_cksum_t dde_chksum;
82 uint64_t dde_prop;
83 dataref_t dde_ref;
84 } dedup_entry_t;
85
86 #define MAX_DDT_PHYSMEM_PERCENT 20
87 #define SMALLEST_POSSIBLE_MAX_DDT_MB 128
88
89 typedef struct dedup_table {
90 dedup_entry_t **dedup_hash_array;
91 umem_cache_t *ddecache;
92 uint64_t max_ddt_size; /* max dedup table size in bytes */
93 uint64_t cur_ddt_size; /* current dedup table size in bytes */
94 uint64_t ddt_count;
95 int numhashbits;
96 boolean_t ddt_full;
97 } dedup_table_t;
98
99 static int
100 high_order_bit(uint64_t n)
101 {
102 int count;
103
104 for (count = 0; n != 0; count++)
105 n >>= 1;
106 return (count);
107 }
108
109 static size_t
110 ssread(void *buf, size_t len, FILE *stream)
111 {
112 size_t outlen;
113
114 if ((outlen = fread(buf, len, 1, stream)) == 0)
115 return (0);
116
117 return (outlen);
118 }
119
120 static void
121 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
122 zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
123 {
124 dedup_entry_t *dde;
125
126 if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
127 if (ddt->ddt_full == B_FALSE) {
128 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
129 "Dedup table full. Deduplication will continue "
130 "with existing table entries"));
131 ddt->ddt_full = B_TRUE;
132 }
133 return;
134 }
135
136 if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
137 != NULL) {
138 assert(*ddepp == NULL);
139 dde->dde_next = NULL;
140 dde->dde_chksum = *cs;
141 dde->dde_prop = prop;
142 dde->dde_ref = *dr;
143 *ddepp = dde;
144 ddt->cur_ddt_size += sizeof (dedup_entry_t);
145 ddt->ddt_count++;
146 }
147 }
148
149 /*
150 * Using the specified dedup table, do a lookup for an entry with
151 * the checksum cs. If found, return the block's reference info
152 * in *dr. Otherwise, insert a new entry in the dedup table, using
153 * the reference information specified by *dr.
154 *
155 * return value: true - entry was found
156 * false - entry was not found
157 */
158 static boolean_t
159 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
160 uint64_t prop, dataref_t *dr)
161 {
162 uint32_t hashcode;
163 dedup_entry_t **ddepp;
164
165 hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
166
167 for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
168 ddepp = &((*ddepp)->dde_next)) {
169 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
170 (*ddepp)->dde_prop == prop) {
171 *dr = (*ddepp)->dde_ref;
172 return (B_TRUE);
173 }
174 }
175 ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
176 return (B_FALSE);
177 }
178
179 static int
180 cksum_and_write(const void *buf, uint64_t len, zio_cksum_t *zc, int outfd)
181 {
182 fletcher_4_incremental_native(buf, len, zc);
183 return (write(outfd, buf, len));
184 }
185
186 /*
187 * This function is started in a separate thread when the dedup option
188 * has been requested. The main send thread determines the list of
189 * snapshots to be included in the send stream and makes the ioctl calls
190 * for each one. But instead of having the ioctl send the output to the
191 * the output fd specified by the caller of zfs_send()), the
192 * ioctl is told to direct the output to a pipe, which is read by the
193 * alternate thread running THIS function. This function does the
194 * dedup'ing by:
195 * 1. building a dedup table (the DDT)
196 * 2. doing checksums on each data block and inserting a record in the DDT
197 * 3. looking for matching checksums, and
198 * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever
199 * a duplicate block is found.
200 * The output of this function then goes to the output fd requested
201 * by the caller of zfs_send().
202 */
203 static void *
204 cksummer(void *arg)
205 {
206 dedup_arg_t *dda = arg;
207 char *buf = malloc(1<<20);
208 dmu_replay_record_t thedrr;
209 dmu_replay_record_t *drr = &thedrr;
210 struct drr_begin *drrb = &thedrr.drr_u.drr_begin;
211 struct drr_end *drre = &thedrr.drr_u.drr_end;
212 struct drr_object *drro = &thedrr.drr_u.drr_object;
213 struct drr_write *drrw = &thedrr.drr_u.drr_write;
214 struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
215 FILE *ofp;
216 int outfd;
217 dmu_replay_record_t wbr_drr = {0};
218 struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref;
219 dedup_table_t ddt;
220 zio_cksum_t stream_cksum;
221 uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
222 uint64_t numbuckets;
223
224 ddt.max_ddt_size =
225 MAX((physmem * MAX_DDT_PHYSMEM_PERCENT)/100,
226 SMALLEST_POSSIBLE_MAX_DDT_MB<<20);
227
228 numbuckets = ddt.max_ddt_size/(sizeof (dedup_entry_t));
229
230 /*
231 * numbuckets must be a power of 2. Increase number to
232 * a power of 2 if necessary.
233 */
234 if (!ISP2(numbuckets))
235 numbuckets = 1 << high_order_bit(numbuckets);
236
237 ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
238 ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
239 NULL, NULL, NULL, NULL, NULL, 0);
240 ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
241 ddt.numhashbits = high_order_bit(numbuckets) - 1;
242 ddt.ddt_full = B_FALSE;
243
244 /* Initialize the write-by-reference block. */
245 wbr_drr.drr_type = DRR_WRITE_BYREF;
246 wbr_drr.drr_payloadlen = 0;
247
248 outfd = dda->outputfd;
249 ofp = fdopen(dda->inputfd, "r");
250 while (ssread(drr, sizeof (dmu_replay_record_t), ofp) != 0) {
251
252 switch (drr->drr_type) {
253 case DRR_BEGIN:
254 {
255 int fflags;
256 ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
257
258 /* set the DEDUP feature flag for this stream */
259 fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
260 fflags |= (DMU_BACKUP_FEATURE_DEDUP |
261 DMU_BACKUP_FEATURE_DEDUPPROPS);
262 DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
263
264 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
265 &stream_cksum, outfd) == -1)
266 goto out;
267 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
268 DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
269 int sz = drr->drr_payloadlen;
270
271 if (sz > 1<<20) {
272 free(buf);
273 buf = malloc(sz);
274 }
275 (void) ssread(buf, sz, ofp);
276 if (ferror(stdin))
277 perror("fread");
278 if (cksum_and_write(buf, sz, &stream_cksum,
279 outfd) == -1)
280 goto out;
281 }
282 break;
283 }
284
285 case DRR_END:
286 {
287 /* use the recalculated checksum */
288 ZIO_SET_CHECKSUM(&drre->drr_checksum,
289 stream_cksum.zc_word[0], stream_cksum.zc_word[1],
290 stream_cksum.zc_word[2], stream_cksum.zc_word[3]);
291 if ((write(outfd, drr,
292 sizeof (dmu_replay_record_t))) == -1)
293 goto out;
294 break;
295 }
296
297 case DRR_OBJECT:
298 {
299 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
300 &stream_cksum, outfd) == -1)
301 goto out;
302 if (drro->drr_bonuslen > 0) {
303 (void) ssread(buf,
304 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
305 ofp);
306 if (cksum_and_write(buf,
307 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
308 &stream_cksum, outfd) == -1)
309 goto out;
310 }
311 break;
312 }
313
314 case DRR_SPILL:
315 {
316 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
317 &stream_cksum, outfd) == -1)
318 goto out;
319 (void) ssread(buf, drrs->drr_length, ofp);
320 if (cksum_and_write(buf, drrs->drr_length,
321 &stream_cksum, outfd) == -1)
322 goto out;
323 break;
324 }
325
326 case DRR_FREEOBJECTS:
327 {
328 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
329 &stream_cksum, outfd) == -1)
330 goto out;
331 break;
332 }
333
334 case DRR_WRITE:
335 {
336 dataref_t dataref;
337
338 (void) ssread(buf, drrw->drr_length, ofp);
339
340 /*
341 * Use the existing checksum if it's dedup-capable,
342 * else calculate a SHA256 checksum for it.
343 */
344
345 if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
346 zero_cksum) ||
347 !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
348 SHA256_CTX ctx;
349 zio_cksum_t tmpsha256;
350
351 SHA256Init(&ctx);
352 SHA256Update(&ctx, buf, drrw->drr_length);
353 SHA256Final(&tmpsha256, &ctx);
354 drrw->drr_key.ddk_cksum.zc_word[0] =
355 BE_64(tmpsha256.zc_word[0]);
356 drrw->drr_key.ddk_cksum.zc_word[1] =
357 BE_64(tmpsha256.zc_word[1]);
358 drrw->drr_key.ddk_cksum.zc_word[2] =
359 BE_64(tmpsha256.zc_word[2]);
360 drrw->drr_key.ddk_cksum.zc_word[3] =
361 BE_64(tmpsha256.zc_word[3]);
362 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
363 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
364 }
365
366 dataref.ref_guid = drrw->drr_toguid;
367 dataref.ref_object = drrw->drr_object;
368 dataref.ref_offset = drrw->drr_offset;
369
370 if (ddt_update(dda->dedup_hdl, &ddt,
371 &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
372 &dataref)) {
373 /* block already present in stream */
374 wbr_drrr->drr_object = drrw->drr_object;
375 wbr_drrr->drr_offset = drrw->drr_offset;
376 wbr_drrr->drr_length = drrw->drr_length;
377 wbr_drrr->drr_toguid = drrw->drr_toguid;
378 wbr_drrr->drr_refguid = dataref.ref_guid;
379 wbr_drrr->drr_refobject =
380 dataref.ref_object;
381 wbr_drrr->drr_refoffset =
382 dataref.ref_offset;
383
384 wbr_drrr->drr_checksumtype =
385 drrw->drr_checksumtype;
386 wbr_drrr->drr_checksumflags =
387 drrw->drr_checksumtype;
388 wbr_drrr->drr_key.ddk_cksum =
389 drrw->drr_key.ddk_cksum;
390 wbr_drrr->drr_key.ddk_prop =
391 drrw->drr_key.ddk_prop;
392
393 if (cksum_and_write(&wbr_drr,
394 sizeof (dmu_replay_record_t), &stream_cksum,
395 outfd) == -1)
396 goto out;
397 } else {
398 /* block not previously seen */
399 if (cksum_and_write(drr,
400 sizeof (dmu_replay_record_t), &stream_cksum,
401 outfd) == -1)
402 goto out;
403 if (cksum_and_write(buf,
404 drrw->drr_length,
405 &stream_cksum, outfd) == -1)
406 goto out;
407 }
408 break;
409 }
410
411 case DRR_FREE:
412 {
413 if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
414 &stream_cksum, outfd) == -1)
415 goto out;
416 break;
417 }
418
419 default:
420 (void) printf("INVALID record type 0x%x\n",
421 drr->drr_type);
422 /* should never happen, so assert */
423 assert(B_FALSE);
424 }
425 }
426 out:
427 umem_cache_destroy(ddt.ddecache);
428 free(ddt.dedup_hash_array);
429 free(buf);
430 (void) fclose(ofp);
431
432 return (NULL);
433 }
434
435 /*
436 * Routines for dealing with the AVL tree of fs-nvlists
437 */
438 typedef struct fsavl_node {
439 avl_node_t fn_node;
440 nvlist_t *fn_nvfs;
441 char *fn_snapname;
442 uint64_t fn_guid;
443 } fsavl_node_t;
444
445 static int
446 fsavl_compare(const void *arg1, const void *arg2)
447 {
448 const fsavl_node_t *fn1 = arg1;
449 const fsavl_node_t *fn2 = arg2;
450
451 if (fn1->fn_guid > fn2->fn_guid)
452 return (+1);
453 else if (fn1->fn_guid < fn2->fn_guid)
454 return (-1);
455 else
456 return (0);
457 }
458
459 /*
460 * Given the GUID of a snapshot, find its containing filesystem and
461 * (optionally) name.
462 */
463 static nvlist_t *
464 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
465 {
466 fsavl_node_t fn_find;
467 fsavl_node_t *fn;
468
469 fn_find.fn_guid = snapguid;
470
471 fn = avl_find(avl, &fn_find, NULL);
472 if (fn) {
473 if (snapname)
474 *snapname = fn->fn_snapname;
475 return (fn->fn_nvfs);
476 }
477 return (NULL);
478 }
479
480 static void
481 fsavl_destroy(avl_tree_t *avl)
482 {
483 fsavl_node_t *fn;
484 void *cookie;
485
486 if (avl == NULL)
487 return;
488
489 cookie = NULL;
490 while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
491 free(fn);
492 avl_destroy(avl);
493 free(avl);
494 }
495
496 /*
497 * Given an nvlist, produce an avl tree of snapshots, ordered by guid
498 */
499 static avl_tree_t *
500 fsavl_create(nvlist_t *fss)
501 {
502 avl_tree_t *fsavl;
503 nvpair_t *fselem = NULL;
504
505 if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
506 return (NULL);
507
508 avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
509 offsetof(fsavl_node_t, fn_node));
510
511 while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
512 nvlist_t *nvfs, *snaps;
513 nvpair_t *snapelem = NULL;
514
515 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
516 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
517
518 while ((snapelem =
519 nvlist_next_nvpair(snaps, snapelem)) != NULL) {
520 fsavl_node_t *fn;
521 uint64_t guid;
522
523 VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
524 if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
525 fsavl_destroy(fsavl);
526 return (NULL);
527 }
528 fn->fn_nvfs = nvfs;
529 fn->fn_snapname = nvpair_name(snapelem);
530 fn->fn_guid = guid;
531
532 /*
533 * Note: if there are multiple snaps with the
534 * same GUID, we ignore all but one.
535 */
536 if (avl_find(fsavl, fn, NULL) == NULL)
537 avl_add(fsavl, fn);
538 else
539 free(fn);
540 }
541 }
542
543 return (fsavl);
544 }
545
546 /*
547 * Routines for dealing with the giant nvlist of fs-nvlists, etc.
548 */
549 typedef struct send_data {
550 uint64_t parent_fromsnap_guid;
551 nvlist_t *parent_snaps;
552 nvlist_t *fss;
553 nvlist_t *snapprops;
554 const char *fromsnap;
555 const char *tosnap;
556 boolean_t recursive;
557
558 /*
559 * The header nvlist is of the following format:
560 * {
561 * "tosnap" -> string
562 * "fromsnap" -> string (if incremental)
563 * "fss" -> {
564 * id -> {
565 *
566 * "name" -> string (full name; for debugging)
567 * "parentfromsnap" -> number (guid of fromsnap in parent)
568 *
569 * "props" -> { name -> value (only if set here) }
570 * "snaps" -> { name (lastname) -> number (guid) }
571 * "snapprops" -> { name (lastname) -> { name -> value } }
572 *
573 * "origin" -> number (guid) (if clone)
574 * "sent" -> boolean (not on-disk)
575 * }
576 * }
577 * }
578 *
579 */
580 } send_data_t;
581
582 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
583
584 static int
585 send_iterate_snap(zfs_handle_t *zhp, void *arg)
586 {
587 send_data_t *sd = arg;
588 uint64_t guid = zhp->zfs_dmustats.dds_guid;
589 char *snapname;
590 nvlist_t *nv;
591
592 snapname = strrchr(zhp->zfs_name, '@')+1;
593
594 VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
595 /*
596 * NB: if there is no fromsnap here (it's a newly created fs in
597 * an incremental replication), we will substitute the tosnap.
598 */
599 if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
600 (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
601 strcmp(snapname, sd->tosnap) == 0)) {
602 sd->parent_fromsnap_guid = guid;
603 }
604
605 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
606 send_iterate_prop(zhp, nv);
607 VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
608 nvlist_free(nv);
609
610 zfs_close(zhp);
611 return (0);
612 }
613
614 static void
615 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
616 {
617 nvpair_t *elem = NULL;
618
619 while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
620 char *propname = nvpair_name(elem);
621 zfs_prop_t prop = zfs_name_to_prop(propname);
622 nvlist_t *propnv;
623
624 if (!zfs_prop_user(propname)) {
625 /*
626 * Realistically, this should never happen. However,
627 * we want the ability to add DSL properties without
628 * needing to make incompatible version changes. We
629 * need to ignore unknown properties to allow older
630 * software to still send datasets containing these
631 * properties, with the unknown properties elided.
632 */
633 if (prop == ZPROP_INVAL)
634 continue;
635
636 if (zfs_prop_readonly(prop))
637 continue;
638 }
639
640 verify(nvpair_value_nvlist(elem, &propnv) == 0);
641 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
642 prop == ZFS_PROP_REFQUOTA ||
643 prop == ZFS_PROP_REFRESERVATION) {
644 char *source;
645 uint64_t value;
646 verify(nvlist_lookup_uint64(propnv,
647 ZPROP_VALUE, &value) == 0);
648 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
649 continue;
650 /*
651 * May have no source before SPA_VERSION_RECVD_PROPS,
652 * but is still modifiable.
653 */
654 if (nvlist_lookup_string(propnv,
655 ZPROP_SOURCE, &source) == 0) {
656 if ((strcmp(source, zhp->zfs_name) != 0) &&
657 (strcmp(source,
658 ZPROP_SOURCE_VAL_RECVD) != 0))
659 continue;
660 }
661 } else {
662 char *source;
663 if (nvlist_lookup_string(propnv,
664 ZPROP_SOURCE, &source) != 0)
665 continue;
666 if ((strcmp(source, zhp->zfs_name) != 0) &&
667 (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
668 continue;
669 }
670
671 if (zfs_prop_user(propname) ||
672 zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
673 char *value;
674 verify(nvlist_lookup_string(propnv,
675 ZPROP_VALUE, &value) == 0);
676 VERIFY(0 == nvlist_add_string(nv, propname, value));
677 } else {
678 uint64_t value;
679 verify(nvlist_lookup_uint64(propnv,
680 ZPROP_VALUE, &value) == 0);
681 VERIFY(0 == nvlist_add_uint64(nv, propname, value));
682 }
683 }
684 }
685
686 /*
687 * recursively generate nvlists describing datasets. See comment
688 * for the data structure send_data_t above for description of contents
689 * of the nvlist.
690 */
691 static int
692 send_iterate_fs(zfs_handle_t *zhp, void *arg)
693 {
694 send_data_t *sd = arg;
695 nvlist_t *nvfs, *nv;
696 int rv = 0;
697 uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
698 uint64_t guid = zhp->zfs_dmustats.dds_guid;
699 char guidstring[64];
700
701 VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
702 VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
703 VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
704 sd->parent_fromsnap_guid));
705
706 if (zhp->zfs_dmustats.dds_origin[0]) {
707 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
708 zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
709 if (origin == NULL)
710 return (-1);
711 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
712 origin->zfs_dmustats.dds_guid));
713 }
714
715 /* iterate over props */
716 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
717 send_iterate_prop(zhp, nv);
718 VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
719 nvlist_free(nv);
720
721 /* iterate over snaps, and set sd->parent_fromsnap_guid */
722 sd->parent_fromsnap_guid = 0;
723 VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
724 VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
725 (void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
726 VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
727 VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
728 nvlist_free(sd->parent_snaps);
729 nvlist_free(sd->snapprops);
730
731 /* add this fs to nvlist */
732 (void) snprintf(guidstring, sizeof (guidstring),
733 "0x%llx", (longlong_t)guid);
734 VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
735 nvlist_free(nvfs);
736
737 /* iterate over children */
738 if (sd->recursive)
739 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
740
741 sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
742
743 zfs_close(zhp);
744 return (rv);
745 }
746
747 static int
748 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
749 const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
750 {
751 zfs_handle_t *zhp;
752 send_data_t sd = { 0 };
753 int error;
754
755 zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
756 if (zhp == NULL)
757 return (EZFS_BADTYPE);
758
759 VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
760 sd.fromsnap = fromsnap;
761 sd.tosnap = tosnap;
762 sd.recursive = recursive;
763
764 if ((error = send_iterate_fs(zhp, &sd)) != 0) {
765 nvlist_free(sd.fss);
766 if (avlp != NULL)
767 *avlp = NULL;
768 *nvlp = NULL;
769 return (error);
770 }
771
772 if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
773 nvlist_free(sd.fss);
774 *nvlp = NULL;
775 return (EZFS_NOMEM);
776 }
777
778 *nvlp = sd.fss;
779 return (0);
780 }
781
782 /*
783 * Routines specific to "zfs send"
784 */
785 typedef struct send_dump_data {
786 /* these are all just the short snapname (the part after the @) */
787 const char *fromsnap;
788 const char *tosnap;
789 char prevsnap[ZFS_MAXNAMELEN];
790 uint64_t prevsnap_obj;
791 boolean_t seenfrom, seento, replicate, doall, fromorigin;
792 boolean_t verbose, dryrun, parsable, progress;
793 int outfd;
794 boolean_t err;
795 nvlist_t *fss;
796 avl_tree_t *fsavl;
797 snapfilter_cb_t *filter_cb;
798 void *filter_cb_arg;
799 nvlist_t *debugnv;
800 char holdtag[ZFS_MAXNAMELEN];
801 int cleanup_fd;
802 uint64_t size;
803 } send_dump_data_t;
804
805 static int
806 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
807 boolean_t fromorigin, uint64_t *sizep)
808 {
809 zfs_cmd_t zc = { 0 };
810 libzfs_handle_t *hdl = zhp->zfs_hdl;
811
812 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
813 assert(fromsnap_obj == 0 || !fromorigin);
814
815 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
816 zc.zc_obj = fromorigin;
817 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
818 zc.zc_fromobj = fromsnap_obj;
819 zc.zc_guid = 1; /* estimate flag */
820
821 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
822 char errbuf[1024];
823 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
824 "warning: cannot estimate space for '%s'"), zhp->zfs_name);
825
826 switch (errno) {
827 case EXDEV:
828 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
829 "not an earlier snapshot from the same fs"));
830 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
831
832 case ENOENT:
833 if (zfs_dataset_exists(hdl, zc.zc_name,
834 ZFS_TYPE_SNAPSHOT)) {
835 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
836 "incremental source (@%s) does not exist"),
837 zc.zc_value);
838 }
839 return (zfs_error(hdl, EZFS_NOENT, errbuf));
840
841 case EDQUOT:
842 case EFBIG:
843 case EIO:
844 case ENOLINK:
845 case ENOSPC:
846 case ENOSTR:
847 case ENXIO:
848 case EPIPE:
849 case ERANGE:
850 case EFAULT:
851 case EROFS:
852 zfs_error_aux(hdl, strerror(errno));
853 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
854
855 default:
856 return (zfs_standard_error(hdl, errno, errbuf));
857 }
858 }
859
860 *sizep = zc.zc_objset_type;
861
862 return (0);
863 }
864
865 /*
866 * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
867 * NULL) to the file descriptor specified by outfd.
868 */
869 static int
870 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
871 boolean_t fromorigin, int outfd, nvlist_t *debugnv)
872 {
873 zfs_cmd_t zc = { 0 };
874 libzfs_handle_t *hdl = zhp->zfs_hdl;
875 nvlist_t *thisdbg;
876
877 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
878 assert(fromsnap_obj == 0 || !fromorigin);
879
880 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
881 zc.zc_cookie = outfd;
882 zc.zc_obj = fromorigin;
883 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
884 zc.zc_fromobj = fromsnap_obj;
885
886 VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
887 if (fromsnap && fromsnap[0] != '\0') {
888 VERIFY(0 == nvlist_add_string(thisdbg,
889 "fromsnap", fromsnap));
890 }
891
892 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
893 char errbuf[1024];
894 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
895 "warning: cannot send '%s'"), zhp->zfs_name);
896
897 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
898 if (debugnv) {
899 VERIFY(0 == nvlist_add_nvlist(debugnv,
900 zhp->zfs_name, thisdbg));
901 }
902 nvlist_free(thisdbg);
903
904 switch (errno) {
905 case EXDEV:
906 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
907 "not an earlier snapshot from the same fs"));
908 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
909
910 case ENOENT:
911 if (zfs_dataset_exists(hdl, zc.zc_name,
912 ZFS_TYPE_SNAPSHOT)) {
913 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
914 "incremental source (@%s) does not exist"),
915 zc.zc_value);
916 }
917 return (zfs_error(hdl, EZFS_NOENT, errbuf));
918
919 case EDQUOT:
920 case EFBIG:
921 case EIO:
922 case ENOLINK:
923 case ENOSPC:
924 case ENOSTR:
925 case ENXIO:
926 case EPIPE:
927 case ERANGE:
928 case EFAULT:
929 case EROFS:
930 zfs_error_aux(hdl, strerror(errno));
931 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
932
933 default:
934 return (zfs_standard_error(hdl, errno, errbuf));
935 }
936 }
937
938 if (debugnv)
939 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
940 nvlist_free(thisdbg);
941
942 return (0);
943 }
944
945 static int
946 hold_for_send(zfs_handle_t *zhp, send_dump_data_t *sdd)
947 {
948 zfs_handle_t *pzhp;
949 int error = 0;
950 char *thissnap;
951
952 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
953
954 if (sdd->dryrun)
955 return (0);
956
957 /*
958 * zfs_send() only opens a cleanup_fd for sends that need it,
959 * e.g. replication and doall.
960 */
961 if (sdd->cleanup_fd == -1)
962 return (0);
963
964 thissnap = strchr(zhp->zfs_name, '@') + 1;
965 *(thissnap - 1) = '\0';
966 pzhp = zfs_open(zhp->zfs_hdl, zhp->zfs_name, ZFS_TYPE_DATASET);
967 *(thissnap - 1) = '@';
968
969 /*
970 * It's OK if the parent no longer exists. The send code will
971 * handle that error.
972 */
973 if (pzhp) {
974 error = zfs_hold(pzhp, thissnap, sdd->holdtag,
975 B_FALSE, B_TRUE, sdd->cleanup_fd);
976 zfs_close(pzhp);
977 }
978
979 return (error);
980 }
981
982 static void *
983 send_progress_thread(void *arg)
984 {
985 progress_arg_t *pa = arg;
986
987 zfs_cmd_t zc = { 0 };
988 zfs_handle_t *zhp = pa->pa_zhp;
989 libzfs_handle_t *hdl = zhp->zfs_hdl;
990 unsigned long long bytes;
991 char buf[16];
992
993 time_t t;
994 struct tm *tm;
995
996 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
997 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
998
999 if (!pa->pa_parsable)
1000 (void) fprintf(stderr, "TIME SENT SNAPSHOT\n");
1001
1002 /*
1003 * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1004 */
1005 for (;;) {
1006 (void) sleep(1);
1007
1008 zc.zc_cookie = pa->pa_fd;
1009 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1010 return ((void *)-1);
1011
1012 (void) time(&t);
1013 tm = localtime(&t);
1014 bytes = zc.zc_cookie;
1015
1016 if (pa->pa_parsable) {
1017 (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1018 tm->tm_hour, tm->tm_min, tm->tm_sec,
1019 bytes, zhp->zfs_name);
1020 } else {
1021 zfs_nicenum(bytes, buf, sizeof (buf));
1022 (void) fprintf(stderr, "%02d:%02d:%02d %5s %s\n",
1023 tm->tm_hour, tm->tm_min, tm->tm_sec,
1024 buf, zhp->zfs_name);
1025 }
1026 }
1027 }
1028
1029 static int
1030 dump_snapshot(zfs_handle_t *zhp, void *arg)
1031 {
1032 send_dump_data_t *sdd = arg;
1033 progress_arg_t pa = { 0 };
1034 pthread_t tid;
1035
1036 char *thissnap;
1037 int err;
1038 boolean_t isfromsnap, istosnap, fromorigin;
1039 boolean_t exclude = B_FALSE;
1040
1041 thissnap = strchr(zhp->zfs_name, '@') + 1;
1042 isfromsnap = (sdd->fromsnap != NULL &&
1043 strcmp(sdd->fromsnap, thissnap) == 0);
1044
1045 if (!sdd->seenfrom && isfromsnap) {
1046 err = hold_for_send(zhp, sdd);
1047 if (err == 0) {
1048 sdd->seenfrom = B_TRUE;
1049 (void) strcpy(sdd->prevsnap, thissnap);
1050 sdd->prevsnap_obj = zfs_prop_get_int(zhp,
1051 ZFS_PROP_OBJSETID);
1052 } else if (err == ENOENT) {
1053 err = 0;
1054 }
1055 zfs_close(zhp);
1056 return (err);
1057 }
1058
1059 if (sdd->seento || !sdd->seenfrom) {
1060 zfs_close(zhp);
1061 return (0);
1062 }
1063
1064 istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1065 if (istosnap)
1066 sdd->seento = B_TRUE;
1067
1068 if (!sdd->doall && !isfromsnap && !istosnap) {
1069 if (sdd->replicate) {
1070 char *snapname;
1071 nvlist_t *snapprops;
1072 /*
1073 * Filter out all intermediate snapshots except origin
1074 * snapshots needed to replicate clones.
1075 */
1076 nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1077 zhp->zfs_dmustats.dds_guid, &snapname);
1078
1079 VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1080 "snapprops", &snapprops));
1081 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1082 thissnap, &snapprops));
1083 exclude = !nvlist_exists(snapprops, "is_clone_origin");
1084 } else {
1085 exclude = B_TRUE;
1086 }
1087 }
1088
1089 /*
1090 * If a filter function exists, call it to determine whether
1091 * this snapshot will be sent.
1092 */
1093 if (exclude || (sdd->filter_cb != NULL &&
1094 sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1095 /*
1096 * This snapshot is filtered out. Don't send it, and don't
1097 * set prevsnap_obj, so it will be as if this snapshot didn't
1098 * exist, and the next accepted snapshot will be sent as
1099 * an incremental from the last accepted one, or as the
1100 * first (and full) snapshot in the case of a replication,
1101 * non-incremental send.
1102 */
1103 zfs_close(zhp);
1104 return (0);
1105 }
1106
1107 err = hold_for_send(zhp, sdd);
1108 if (err) {
1109 if (err == ENOENT)
1110 err = 0;
1111 zfs_close(zhp);
1112 return (err);
1113 }
1114
1115 fromorigin = sdd->prevsnap[0] == '\0' &&
1116 (sdd->fromorigin || sdd->replicate);
1117
1118 if (sdd->verbose) {
1119 uint64_t size;
1120 err = estimate_ioctl(zhp, sdd->prevsnap_obj,
1121 fromorigin, &size);
1122
1123 if (sdd->parsable) {
1124 if (sdd->prevsnap[0] != '\0') {
1125 (void) fprintf(stderr, "incremental\t%s\t%s",
1126 sdd->prevsnap, zhp->zfs_name);
1127 } else {
1128 (void) fprintf(stderr, "full\t%s",
1129 zhp->zfs_name);
1130 }
1131 } else {
1132 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1133 "send from @%s to %s"),
1134 sdd->prevsnap, zhp->zfs_name);
1135 }
1136 if (err == 0) {
1137 if (sdd->parsable) {
1138 (void) fprintf(stderr, "\t%llu\n",
1139 (longlong_t)size);
1140 } else {
1141 char buf[16];
1142 zfs_nicenum(size, buf, sizeof (buf));
1143 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1144 " estimated size is %s\n"), buf);
1145 }
1146 sdd->size += size;
1147 } else {
1148 (void) fprintf(stderr, "\n");
1149 }
1150 }
1151
1152 if (!sdd->dryrun) {
1153 /*
1154 * If progress reporting is requested, spawn a new thread to
1155 * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1156 */
1157 if (sdd->progress) {
1158 pa.pa_zhp = zhp;
1159 pa.pa_fd = sdd->outfd;
1160 pa.pa_parsable = sdd->parsable;
1161
1162 if (err = pthread_create(&tid, NULL,
1163 send_progress_thread, &pa)) {
1164 zfs_close(zhp);
1165 return (err);
1166 }
1167 }
1168
1169 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1170 fromorigin, sdd->outfd, sdd->debugnv);
1171
1172 if (sdd->progress) {
1173 (void) pthread_cancel(tid);
1174 (void) pthread_join(tid, NULL);
1175 }
1176 }
1177
1178 (void) strcpy(sdd->prevsnap, thissnap);
1179 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1180 zfs_close(zhp);
1181 return (err);
1182 }
1183
1184 static int
1185 dump_filesystem(zfs_handle_t *zhp, void *arg)
1186 {
1187 int rv = 0;
1188 send_dump_data_t *sdd = arg;
1189 boolean_t missingfrom = B_FALSE;
1190 zfs_cmd_t zc = { 0 };
1191
1192 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1193 zhp->zfs_name, sdd->tosnap);
1194 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1195 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1196 "WARNING: could not send %s@%s: does not exist\n"),
1197 zhp->zfs_name, sdd->tosnap);
1198 sdd->err = B_TRUE;
1199 return (0);
1200 }
1201
1202 if (sdd->replicate && sdd->fromsnap) {
1203 /*
1204 * If this fs does not have fromsnap, and we're doing
1205 * recursive, we need to send a full stream from the
1206 * beginning (or an incremental from the origin if this
1207 * is a clone). If we're doing non-recursive, then let
1208 * them get the error.
1209 */
1210 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1211 zhp->zfs_name, sdd->fromsnap);
1212 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1213 ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1214 missingfrom = B_TRUE;
1215 }
1216 }
1217
1218 sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1219 sdd->prevsnap_obj = 0;
1220 if (sdd->fromsnap == NULL || missingfrom)
1221 sdd->seenfrom = B_TRUE;
1222
1223 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1224 if (!sdd->seenfrom) {
1225 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1226 "WARNING: could not send %s@%s:\n"
1227 "incremental source (%s@%s) does not exist\n"),
1228 zhp->zfs_name, sdd->tosnap,
1229 zhp->zfs_name, sdd->fromsnap);
1230 sdd->err = B_TRUE;
1231 } else if (!sdd->seento) {
1232 if (sdd->fromsnap) {
1233 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1234 "WARNING: could not send %s@%s:\n"
1235 "incremental source (%s@%s) "
1236 "is not earlier than it\n"),
1237 zhp->zfs_name, sdd->tosnap,
1238 zhp->zfs_name, sdd->fromsnap);
1239 } else {
1240 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1241 "WARNING: "
1242 "could not send %s@%s: does not exist\n"),
1243 zhp->zfs_name, sdd->tosnap);
1244 }
1245 sdd->err = B_TRUE;
1246 }
1247
1248 return (rv);
1249 }
1250
1251 static int
1252 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1253 {
1254 send_dump_data_t *sdd = arg;
1255 nvpair_t *fspair;
1256 boolean_t needagain, progress;
1257
1258 if (!sdd->replicate)
1259 return (dump_filesystem(rzhp, sdd));
1260
1261 /* Mark the clone origin snapshots. */
1262 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1263 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1264 nvlist_t *nvfs;
1265 uint64_t origin_guid = 0;
1266
1267 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1268 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1269 if (origin_guid != 0) {
1270 char *snapname;
1271 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1272 origin_guid, &snapname);
1273 if (origin_nv != NULL) {
1274 nvlist_t *snapprops;
1275 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1276 "snapprops", &snapprops));
1277 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1278 snapname, &snapprops));
1279 VERIFY(0 == nvlist_add_boolean(
1280 snapprops, "is_clone_origin"));
1281 }
1282 }
1283 }
1284 again:
1285 needagain = progress = B_FALSE;
1286 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1287 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1288 nvlist_t *fslist, *parent_nv;
1289 char *fsname;
1290 zfs_handle_t *zhp;
1291 int err;
1292 uint64_t origin_guid = 0;
1293 uint64_t parent_guid = 0;
1294
1295 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1296 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1297 continue;
1298
1299 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1300 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1301 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1302 &parent_guid);
1303
1304 if (parent_guid != 0) {
1305 parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1306 if (!nvlist_exists(parent_nv, "sent")) {
1307 /* parent has not been sent; skip this one */
1308 needagain = B_TRUE;
1309 continue;
1310 }
1311 }
1312
1313 if (origin_guid != 0) {
1314 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1315 origin_guid, NULL);
1316 if (origin_nv != NULL &&
1317 !nvlist_exists(origin_nv, "sent")) {
1318 /*
1319 * origin has not been sent yet;
1320 * skip this clone.
1321 */
1322 needagain = B_TRUE;
1323 continue;
1324 }
1325 }
1326
1327 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1328 if (zhp == NULL)
1329 return (-1);
1330 err = dump_filesystem(zhp, sdd);
1331 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1332 progress = B_TRUE;
1333 zfs_close(zhp);
1334 if (err)
1335 return (err);
1336 }
1337 if (needagain) {
1338 assert(progress);
1339 goto again;
1340 }
1341
1342 /* clean out the sent flags in case we reuse this fss */
1343 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1344 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1345 nvlist_t *fslist;
1346
1347 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1348 (void) nvlist_remove_all(fslist, "sent");
1349 }
1350
1351 return (0);
1352 }
1353
1354 /*
1355 * Generate a send stream for the dataset identified by the argument zhp.
1356 *
1357 * The content of the send stream is the snapshot identified by
1358 * 'tosnap'. Incremental streams are requested in two ways:
1359 * - from the snapshot identified by "fromsnap" (if non-null) or
1360 * - from the origin of the dataset identified by zhp, which must
1361 * be a clone. In this case, "fromsnap" is null and "fromorigin"
1362 * is TRUE.
1363 *
1364 * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1365 * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1366 * if "replicate" is set. If "doall" is set, dump all the intermediate
1367 * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1368 * case too. If "props" is set, send properties.
1369 */
1370 int
1371 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1372 sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1373 void *cb_arg, nvlist_t **debugnvp)
1374 {
1375 char errbuf[1024];
1376 send_dump_data_t sdd = { 0 };
1377 int err = 0;
1378 nvlist_t *fss = NULL;
1379 avl_tree_t *fsavl = NULL;
1380 static uint64_t holdseq;
1381 int spa_version;
1382 pthread_t tid;
1383 int pipefd[2];
1384 dedup_arg_t dda = { 0 };
1385 int featureflags = 0;
1386
1387 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1388 "cannot send '%s'"), zhp->zfs_name);
1389
1390 if (fromsnap && fromsnap[0] == '\0') {
1391 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1392 "zero-length incremental source"));
1393 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1394 }
1395
1396 if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1397 uint64_t version;
1398 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1399 if (version >= ZPL_VERSION_SA) {
1400 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1401 }
1402 }
1403
1404 if (flags->dedup && !flags->dryrun) {
1405 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1406 DMU_BACKUP_FEATURE_DEDUPPROPS);
1407 if (err = pipe(pipefd)) {
1408 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1409 return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1410 errbuf));
1411 }
1412 dda.outputfd = outfd;
1413 dda.inputfd = pipefd[1];
1414 dda.dedup_hdl = zhp->zfs_hdl;
1415 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1416 (void) close(pipefd[0]);
1417 (void) close(pipefd[1]);
1418 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1419 return (zfs_error(zhp->zfs_hdl,
1420 EZFS_THREADCREATEFAILED, errbuf));
1421 }
1422 }
1423
1424 if (flags->replicate || flags->doall || flags->props) {
1425 dmu_replay_record_t drr = { 0 };
1426 char *packbuf = NULL;
1427 size_t buflen = 0;
1428 zio_cksum_t zc = { 0 };
1429
1430 if (flags->replicate || flags->props) {
1431 nvlist_t *hdrnv;
1432
1433 VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1434 if (fromsnap) {
1435 VERIFY(0 == nvlist_add_string(hdrnv,
1436 "fromsnap", fromsnap));
1437 }
1438 VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1439 if (!flags->replicate) {
1440 VERIFY(0 == nvlist_add_boolean(hdrnv,
1441 "not_recursive"));
1442 }
1443
1444 err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1445 fromsnap, tosnap, flags->replicate, &fss, &fsavl);
1446 if (err)
1447 goto err_out;
1448 VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1449 err = nvlist_pack(hdrnv, &packbuf, &buflen,
1450 NV_ENCODE_XDR, 0);
1451 if (debugnvp)
1452 *debugnvp = hdrnv;
1453 else
1454 nvlist_free(hdrnv);
1455 if (err) {
1456 fsavl_destroy(fsavl);
1457 nvlist_free(fss);
1458 goto stderr_out;
1459 }
1460 }
1461
1462 if (!flags->dryrun) {
1463 /* write first begin record */
1464 drr.drr_type = DRR_BEGIN;
1465 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1466 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1467 drr_versioninfo, DMU_COMPOUNDSTREAM);
1468 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1469 drr_versioninfo, featureflags);
1470 (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1471 sizeof (drr.drr_u.drr_begin.drr_toname),
1472 "%s@%s", zhp->zfs_name, tosnap);
1473 drr.drr_payloadlen = buflen;
1474 err = cksum_and_write(&drr, sizeof (drr), &zc, outfd);
1475
1476 /* write header nvlist */
1477 if (err != -1 && packbuf != NULL) {
1478 err = cksum_and_write(packbuf, buflen, &zc,
1479 outfd);
1480 }
1481 free(packbuf);
1482 if (err == -1) {
1483 fsavl_destroy(fsavl);
1484 nvlist_free(fss);
1485 err = errno;
1486 goto stderr_out;
1487 }
1488
1489 /* write end record */
1490 bzero(&drr, sizeof (drr));
1491 drr.drr_type = DRR_END;
1492 drr.drr_u.drr_end.drr_checksum = zc;
1493 err = write(outfd, &drr, sizeof (drr));
1494 if (err == -1) {
1495 fsavl_destroy(fsavl);
1496 nvlist_free(fss);
1497 err = errno;
1498 goto stderr_out;
1499 }
1500
1501 err = 0;
1502 }
1503 }
1504
1505 /* dump each stream */
1506 sdd.fromsnap = fromsnap;
1507 sdd.tosnap = tosnap;
1508 if (flags->dedup)
1509 sdd.outfd = pipefd[0];
1510 else
1511 sdd.outfd = outfd;
1512 sdd.replicate = flags->replicate;
1513 sdd.doall = flags->doall;
1514 sdd.fromorigin = flags->fromorigin;
1515 sdd.fss = fss;
1516 sdd.fsavl = fsavl;
1517 sdd.verbose = flags->verbose;
1518 sdd.parsable = flags->parsable;
1519 sdd.progress = flags->progress;
1520 sdd.dryrun = flags->dryrun;
1521 sdd.filter_cb = filter_func;
1522 sdd.filter_cb_arg = cb_arg;
1523 if (debugnvp)
1524 sdd.debugnv = *debugnvp;
1525
1526 /*
1527 * Some flags require that we place user holds on the datasets that are
1528 * being sent so they don't get destroyed during the send. We can skip
1529 * this step if the pool is imported read-only since the datasets cannot
1530 * be destroyed.
1531 */
1532 if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1533 ZPOOL_PROP_READONLY, NULL) &&
1534 zfs_spa_version(zhp, &spa_version) == 0 &&
1535 spa_version >= SPA_VERSION_USERREFS &&
1536 (flags->doall || flags->replicate)) {
1537 ++holdseq;
1538 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1539 ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1540 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1541 if (sdd.cleanup_fd < 0) {
1542 err = errno;
1543 goto stderr_out;
1544 }
1545 } else {
1546 sdd.cleanup_fd = -1;
1547 }
1548 if (flags->verbose) {
1549 /*
1550 * Do a verbose no-op dry run to get all the verbose output
1551 * before generating any data. Then do a non-verbose real
1552 * run to generate the streams.
1553 */
1554 sdd.dryrun = B_TRUE;
1555 err = dump_filesystems(zhp, &sdd);
1556 sdd.dryrun = flags->dryrun;
1557 sdd.verbose = B_FALSE;
1558 if (flags->parsable) {
1559 (void) fprintf(stderr, "size\t%llu\n",
1560 (longlong_t)sdd.size);
1561 } else {
1562 char buf[16];
1563 zfs_nicenum(sdd.size, buf, sizeof (buf));
1564 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1565 "total estimated size is %s\n"), buf);
1566 }
1567 }
1568 err = dump_filesystems(zhp, &sdd);
1569 fsavl_destroy(fsavl);
1570 nvlist_free(fss);
1571
1572 if (flags->dedup) {
1573 (void) close(pipefd[0]);
1574 (void) pthread_join(tid, NULL);
1575 }
1576
1577 if (sdd.cleanup_fd != -1) {
1578 VERIFY(0 == close(sdd.cleanup_fd));
1579 sdd.cleanup_fd = -1;
1580 }
1581
1582 if (!flags->dryrun && (flags->replicate || flags->doall ||
1583 flags->props)) {
1584 /*
1585 * write final end record. NB: want to do this even if
1586 * there was some error, because it might not be totally
1587 * failed.
1588 */
1589 dmu_replay_record_t drr = { 0 };
1590 drr.drr_type = DRR_END;
1591 if (write(outfd, &drr, sizeof (drr)) == -1) {
1592 return (zfs_standard_error(zhp->zfs_hdl,
1593 errno, errbuf));
1594 }
1595 }
1596
1597 return (err || sdd.err);
1598
1599 stderr_out:
1600 err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1601 err_out:
1602 if (sdd.cleanup_fd != -1)
1603 VERIFY(0 == close(sdd.cleanup_fd));
1604 if (flags->dedup) {
1605 (void) pthread_cancel(tid);
1606 (void) pthread_join(tid, NULL);
1607 (void) close(pipefd[0]);
1608 }
1609 return (err);
1610 }
1611
1612 /*
1613 * Routines specific to "zfs recv"
1614 */
1615
1616 static int
1617 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1618 boolean_t byteswap, zio_cksum_t *zc)
1619 {
1620 char *cp = buf;
1621 int rv;
1622 int len = ilen;
1623
1624 do {
1625 rv = read(fd, cp, len);
1626 cp += rv;
1627 len -= rv;
1628 } while (rv > 0);
1629
1630 if (rv < 0 || len != 0) {
1631 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1632 "failed to read from stream"));
1633 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1634 "cannot receive")));
1635 }
1636
1637 if (zc) {
1638 if (byteswap)
1639 fletcher_4_incremental_byteswap(buf, ilen, zc);
1640 else
1641 fletcher_4_incremental_native(buf, ilen, zc);
1642 }
1643 return (0);
1644 }
1645
1646 static int
1647 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1648 boolean_t byteswap, zio_cksum_t *zc)
1649 {
1650 char *buf;
1651 int err;
1652
1653 buf = zfs_alloc(hdl, len);
1654 if (buf == NULL)
1655 return (ENOMEM);
1656
1657 err = recv_read(hdl, fd, buf, len, byteswap, zc);
1658 if (err != 0) {
1659 free(buf);
1660 return (err);
1661 }
1662
1663 err = nvlist_unpack(buf, len, nvp, 0);
1664 free(buf);
1665 if (err != 0) {
1666 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1667 "stream (malformed nvlist)"));
1668 return (EINVAL);
1669 }
1670 return (0);
1671 }
1672
1673 static int
1674 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1675 int baselen, char *newname, recvflags_t *flags)
1676 {
1677 static int seq;
1678 zfs_cmd_t zc = { 0 };
1679 int err;
1680 prop_changelist_t *clp;
1681 zfs_handle_t *zhp;
1682
1683 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1684 if (zhp == NULL)
1685 return (-1);
1686 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1687 flags->force ? MS_FORCE : 0);
1688 zfs_close(zhp);
1689 if (clp == NULL)
1690 return (-1);
1691 err = changelist_prefix(clp);
1692 if (err)
1693 return (err);
1694
1695 zc.zc_objset_type = DMU_OST_ZFS;
1696 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1697
1698 if (tryname) {
1699 (void) strcpy(newname, tryname);
1700
1701 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1702
1703 if (flags->verbose) {
1704 (void) printf("attempting rename %s to %s\n",
1705 zc.zc_name, zc.zc_value);
1706 }
1707 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1708 if (err == 0)
1709 changelist_rename(clp, name, tryname);
1710 } else {
1711 err = ENOENT;
1712 }
1713
1714 if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
1715 seq++;
1716
1717 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u",
1718 baselen, name, getpid(), seq);
1719 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1720
1721 if (flags->verbose) {
1722 (void) printf("failed - trying rename %s to %s\n",
1723 zc.zc_name, zc.zc_value);
1724 }
1725 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1726 if (err == 0)
1727 changelist_rename(clp, name, newname);
1728 if (err && flags->verbose) {
1729 (void) printf("failed (%u) - "
1730 "will try again on next pass\n", errno);
1731 }
1732 err = EAGAIN;
1733 } else if (flags->verbose) {
1734 if (err == 0)
1735 (void) printf("success\n");
1736 else
1737 (void) printf("failed (%u)\n", errno);
1738 }
1739
1740 (void) changelist_postfix(clp);
1741 changelist_free(clp);
1742
1743 return (err);
1744 }
1745
1746 static int
1747 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1748 char *newname, recvflags_t *flags)
1749 {
1750 zfs_cmd_t zc = { 0 };
1751 int err = 0;
1752 prop_changelist_t *clp;
1753 zfs_handle_t *zhp;
1754 boolean_t defer = B_FALSE;
1755 int spa_version;
1756
1757 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1758 if (zhp == NULL)
1759 return (-1);
1760 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1761 flags->force ? MS_FORCE : 0);
1762 if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1763 zfs_spa_version(zhp, &spa_version) == 0 &&
1764 spa_version >= SPA_VERSION_USERREFS)
1765 defer = B_TRUE;
1766 zfs_close(zhp);
1767 if (clp == NULL)
1768 return (-1);
1769 err = changelist_prefix(clp);
1770 if (err)
1771 return (err);
1772
1773 zc.zc_objset_type = DMU_OST_ZFS;
1774 zc.zc_defer_destroy = defer;
1775 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1776
1777 if (flags->verbose)
1778 (void) printf("attempting destroy %s\n", zc.zc_name);
1779 err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1780 if (err == 0) {
1781 if (flags->verbose)
1782 (void) printf("success\n");
1783 changelist_remove(clp, zc.zc_name);
1784 }
1785
1786 (void) changelist_postfix(clp);
1787 changelist_free(clp);
1788
1789 /*
1790 * Deferred destroy might destroy the snapshot or only mark it to be
1791 * destroyed later, and it returns success in either case.
1792 */
1793 if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1794 ZFS_TYPE_SNAPSHOT))) {
1795 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1796 }
1797
1798 return (err);
1799 }
1800
1801 typedef struct guid_to_name_data {
1802 uint64_t guid;
1803 char *name;
1804 char *skip;
1805 } guid_to_name_data_t;
1806
1807 static int
1808 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1809 {
1810 guid_to_name_data_t *gtnd = arg;
1811 int err;
1812
1813 if (gtnd->skip != NULL &&
1814 strcmp(zhp->zfs_name, gtnd->skip) == 0) {
1815 return (0);
1816 }
1817
1818 if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1819 (void) strcpy(gtnd->name, zhp->zfs_name);
1820 zfs_close(zhp);
1821 return (EEXIST);
1822 }
1823
1824 err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1825 zfs_close(zhp);
1826 return (err);
1827 }
1828
1829 /*
1830 * Attempt to find the local dataset associated with this guid. In the case of
1831 * multiple matches, we attempt to find the "best" match by searching
1832 * progressively larger portions of the hierarchy. This allows one to send a
1833 * tree of datasets individually and guarantee that we will find the source
1834 * guid within that hierarchy, even if there are multiple matches elsewhere.
1835 */
1836 static int
1837 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1838 char *name)
1839 {
1840 /* exhaustive search all local snapshots */
1841 char pname[ZFS_MAXNAMELEN];
1842 guid_to_name_data_t gtnd;
1843 int err = 0;
1844 zfs_handle_t *zhp;
1845 char *cp;
1846
1847 gtnd.guid = guid;
1848 gtnd.name = name;
1849 gtnd.skip = NULL;
1850
1851 (void) strlcpy(pname, parent, sizeof (pname));
1852
1853 /*
1854 * Search progressively larger portions of the hierarchy. This will
1855 * select the "most local" version of the origin snapshot in the case
1856 * that there are multiple matching snapshots in the system.
1857 */
1858 while ((cp = strrchr(pname, '/')) != NULL) {
1859
1860 /* Chop off the last component and open the parent */
1861 *cp = '\0';
1862 zhp = make_dataset_handle(hdl, pname);
1863
1864 if (zhp == NULL)
1865 continue;
1866
1867 err = zfs_iter_children(zhp, guid_to_name_cb, >nd);
1868 zfs_close(zhp);
1869 if (err == EEXIST)
1870 return (0);
1871
1872 /*
1873 * Remember the dataset that we already searched, so we
1874 * skip it next time through.
1875 */
1876 gtnd.skip = pname;
1877 }
1878
1879 return (ENOENT);
1880 }
1881
1882 /*
1883 * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
1884 * guid1 is after guid2.
1885 */
1886 static int
1887 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1888 uint64_t guid1, uint64_t guid2)
1889 {
1890 nvlist_t *nvfs;
1891 char *fsname, *snapname;
1892 char buf[ZFS_MAXNAMELEN];
1893 int rv;
1894 zfs_handle_t *guid1hdl, *guid2hdl;
1895 uint64_t create1, create2;
1896
1897 if (guid2 == 0)
1898 return (0);
1899 if (guid1 == 0)
1900 return (1);
1901
1902 nvfs = fsavl_find(avl, guid1, &snapname);
1903 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1904 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1905 guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1906 if (guid1hdl == NULL)
1907 return (-1);
1908
1909 nvfs = fsavl_find(avl, guid2, &snapname);
1910 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1911 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1912 guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1913 if (guid2hdl == NULL) {
1914 zfs_close(guid1hdl);
1915 return (-1);
1916 }
1917
1918 create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
1919 create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
1920
1921 if (create1 < create2)
1922 rv = -1;
1923 else if (create1 > create2)
1924 rv = +1;
1925 else
1926 rv = 0;
1927
1928 zfs_close(guid1hdl);
1929 zfs_close(guid2hdl);
1930
1931 return (rv);
1932 }
1933
1934 static int
1935 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
1936 recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
1937 nvlist_t *renamed)
1938 {
1939 nvlist_t *local_nv;
1940 avl_tree_t *local_avl;
1941 nvpair_t *fselem, *nextfselem;
1942 char *fromsnap;
1943 char newname[ZFS_MAXNAMELEN];
1944 int error;
1945 boolean_t needagain, progress, recursive;
1946 char *s1, *s2;
1947
1948 VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
1949
1950 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
1951 ENOENT);
1952
1953 if (flags->dryrun)
1954 return (0);
1955
1956 again:
1957 needagain = progress = B_FALSE;
1958
1959 if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
1960 recursive, &local_nv, &local_avl)) != 0)
1961 return (error);
1962
1963 /*
1964 * Process deletes and renames
1965 */
1966 for (fselem = nvlist_next_nvpair(local_nv, NULL);
1967 fselem; fselem = nextfselem) {
1968 nvlist_t *nvfs, *snaps;
1969 nvlist_t *stream_nvfs = NULL;
1970 nvpair_t *snapelem, *nextsnapelem;
1971 uint64_t fromguid = 0;
1972 uint64_t originguid = 0;
1973 uint64_t stream_originguid = 0;
1974 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
1975 char *fsname, *stream_fsname;
1976
1977 nextfselem = nvlist_next_nvpair(local_nv, fselem);
1978
1979 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
1980 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
1981 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1982 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
1983 &parent_fromsnap_guid));
1984 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
1985
1986 /*
1987 * First find the stream's fs, so we can check for
1988 * a different origin (due to "zfs promote")
1989 */
1990 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1991 snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
1992 uint64_t thisguid;
1993
1994 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1995 stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
1996
1997 if (stream_nvfs != NULL)
1998 break;
1999 }
2000
2001 /* check for promote */
2002 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2003 &stream_originguid);
2004 if (stream_nvfs && originguid != stream_originguid) {
2005 switch (created_before(hdl, local_avl,
2006 stream_originguid, originguid)) {
2007 case 1: {
2008 /* promote it! */
2009 zfs_cmd_t zc = { 0 };
2010 nvlist_t *origin_nvfs;
2011 char *origin_fsname;
2012
2013 if (flags->verbose)
2014 (void) printf("promoting %s\n", fsname);
2015
2016 origin_nvfs = fsavl_find(local_avl, originguid,
2017 NULL);
2018 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2019 "name", &origin_fsname));
2020 (void) strlcpy(zc.zc_value, origin_fsname,
2021 sizeof (zc.zc_value));
2022 (void) strlcpy(zc.zc_name, fsname,
2023 sizeof (zc.zc_name));
2024 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2025 if (error == 0)
2026 progress = B_TRUE;
2027 break;
2028 }
2029 default:
2030 break;
2031 case -1:
2032 fsavl_destroy(local_avl);
2033 nvlist_free(local_nv);
2034 return (-1);
2035 }
2036 /*
2037 * We had/have the wrong origin, therefore our
2038 * list of snapshots is wrong. Need to handle
2039 * them on the next pass.
2040 */
2041 needagain = B_TRUE;
2042 continue;
2043 }
2044
2045 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2046 snapelem; snapelem = nextsnapelem) {
2047 uint64_t thisguid;
2048 char *stream_snapname;
2049 nvlist_t *found, *props;
2050
2051 nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2052
2053 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2054 found = fsavl_find(stream_avl, thisguid,
2055 &stream_snapname);
2056
2057 /* check for delete */
2058 if (found == NULL) {
2059 char name[ZFS_MAXNAMELEN];
2060
2061 if (!flags->force)
2062 continue;
2063
2064 (void) snprintf(name, sizeof (name), "%s@%s",
2065 fsname, nvpair_name(snapelem));
2066
2067 error = recv_destroy(hdl, name,
2068 strlen(fsname)+1, newname, flags);
2069 if (error)
2070 needagain = B_TRUE;
2071 else
2072 progress = B_TRUE;
2073 continue;
2074 }
2075
2076 stream_nvfs = found;
2077
2078 if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2079 &props) && 0 == nvlist_lookup_nvlist(props,
2080 stream_snapname, &props)) {
2081 zfs_cmd_t zc = { 0 };
2082
2083 zc.zc_cookie = B_TRUE; /* received */
2084 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2085 "%s@%s", fsname, nvpair_name(snapelem));
2086 if (zcmd_write_src_nvlist(hdl, &zc,
2087 props) == 0) {
2088 (void) zfs_ioctl(hdl,
2089 ZFS_IOC_SET_PROP, &zc);
2090 zcmd_free_nvlists(&zc);
2091 }
2092 }
2093
2094 /* check for different snapname */
2095 if (strcmp(nvpair_name(snapelem),
2096 stream_snapname) != 0) {
2097 char name[ZFS_MAXNAMELEN];
2098 char tryname[ZFS_MAXNAMELEN];
2099
2100 (void) snprintf(name, sizeof (name), "%s@%s",
2101 fsname, nvpair_name(snapelem));
2102 (void) snprintf(tryname, sizeof (name), "%s@%s",
2103 fsname, stream_snapname);
2104
2105 error = recv_rename(hdl, name, tryname,
2106 strlen(fsname)+1, newname, flags);
2107 if (error)
2108 needagain = B_TRUE;
2109 else
2110 progress = B_TRUE;
2111 }
2112
2113 if (strcmp(stream_snapname, fromsnap) == 0)
2114 fromguid = thisguid;
2115 }
2116
2117 /* check for delete */
2118 if (stream_nvfs == NULL) {
2119 if (!flags->force)
2120 continue;
2121
2122 error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2123 newname, flags);
2124 if (error)
2125 needagain = B_TRUE;
2126 else
2127 progress = B_TRUE;
2128 continue;
2129 }
2130
2131 if (fromguid == 0) {
2132 if (flags->verbose) {
2133 (void) printf("local fs %s does not have "
2134 "fromsnap (%s in stream); must have "
2135 "been deleted locally; ignoring\n",
2136 fsname, fromsnap);
2137 }
2138 continue;
2139 }
2140
2141 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2142 "name", &stream_fsname));
2143 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2144 "parentfromsnap", &stream_parent_fromsnap_guid));
2145
2146 s1 = strrchr(fsname, '/');
2147 s2 = strrchr(stream_fsname, '/');
2148
2149 /*
2150 * Check for rename. If the exact receive path is specified, it
2151 * does not count as a rename, but we still need to check the
2152 * datasets beneath it.
2153 */
2154 if ((stream_parent_fromsnap_guid != 0 &&
2155 parent_fromsnap_guid != 0 &&
2156 stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2157 ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2158 (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2159 nvlist_t *parent;
2160 char tryname[ZFS_MAXNAMELEN];
2161
2162 parent = fsavl_find(local_avl,
2163 stream_parent_fromsnap_guid, NULL);
2164 /*
2165 * NB: parent might not be found if we used the
2166 * tosnap for stream_parent_fromsnap_guid,
2167 * because the parent is a newly-created fs;
2168 * we'll be able to rename it after we recv the
2169 * new fs.
2170 */
2171 if (parent != NULL) {
2172 char *pname;
2173
2174 VERIFY(0 == nvlist_lookup_string(parent, "name",
2175 &pname));
2176 (void) snprintf(tryname, sizeof (tryname),
2177 "%s%s", pname, strrchr(stream_fsname, '/'));
2178 } else {
2179 tryname[0] = '\0';
2180 if (flags->verbose) {
2181 (void) printf("local fs %s new parent "
2182 "not found\n", fsname);
2183 }
2184 }
2185
2186 newname[0] = '\0';
2187
2188 error = recv_rename(hdl, fsname, tryname,
2189 strlen(tofs)+1, newname, flags);
2190
2191 if (renamed != NULL && newname[0] != '\0') {
2192 VERIFY(0 == nvlist_add_boolean(renamed,
2193 newname));
2194 }
2195
2196 if (error)
2197 needagain = B_TRUE;
2198 else
2199 progress = B_TRUE;
2200 }
2201 }
2202
2203 fsavl_destroy(local_avl);
2204 nvlist_free(local_nv);
2205
2206 if (needagain && progress) {
2207 /* do another pass to fix up temporary names */
2208 if (flags->verbose)
2209 (void) printf("another pass:\n");
2210 goto again;
2211 }
2212
2213 return (needagain);
2214 }
2215
2216 static int
2217 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2218 recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2219 char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2220 {
2221 nvlist_t *stream_nv = NULL;
2222 avl_tree_t *stream_avl = NULL;
2223 char *fromsnap = NULL;
2224 char *cp;
2225 char tofs[ZFS_MAXNAMELEN];
2226 char sendfs[ZFS_MAXNAMELEN];
2227 char errbuf[1024];
2228 dmu_replay_record_t drre;
2229 int error;
2230 boolean_t anyerr = B_FALSE;
2231 boolean_t softerr = B_FALSE;
2232 boolean_t recursive;
2233
2234 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2235 "cannot receive"));
2236
2237 assert(drr->drr_type == DRR_BEGIN);
2238 assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2239 assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2240 DMU_COMPOUNDSTREAM);
2241
2242 /*
2243 * Read in the nvlist from the stream.
2244 */
2245 if (drr->drr_payloadlen != 0) {
2246 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2247 &stream_nv, flags->byteswap, zc);
2248 if (error) {
2249 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2250 goto out;
2251 }
2252 }
2253
2254 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2255 ENOENT);
2256
2257 if (recursive && strchr(destname, '@')) {
2258 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2259 "cannot specify snapshot name for multi-snapshot stream"));
2260 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2261 goto out;
2262 }
2263
2264 /*
2265 * Read in the end record and verify checksum.
2266 */
2267 if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2268 flags->byteswap, NULL)))
2269 goto out;
2270 if (flags->byteswap) {
2271 drre.drr_type = BSWAP_32(drre.drr_type);
2272 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2273 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2274 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2275 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2276 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2277 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2278 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2279 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2280 }
2281 if (drre.drr_type != DRR_END) {
2282 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2283 goto out;
2284 }
2285 if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2286 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2287 "incorrect header checksum"));
2288 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2289 goto out;
2290 }
2291
2292 (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2293
2294 if (drr->drr_payloadlen != 0) {
2295 nvlist_t *stream_fss;
2296
2297 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2298 &stream_fss));
2299 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2300 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2301 "couldn't allocate avl tree"));
2302 error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2303 goto out;
2304 }
2305
2306 if (fromsnap != NULL) {
2307 nvlist_t *renamed = NULL;
2308 nvpair_t *pair = NULL;
2309
2310 (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2311 if (flags->isprefix) {
2312 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2313 int i;
2314
2315 if (flags->istail) {
2316 cp = strrchr(drrb->drr_toname, '/');
2317 if (cp == NULL) {
2318 (void) strlcat(tofs, "/",
2319 ZFS_MAXNAMELEN);
2320 i = 0;
2321 } else {
2322 i = (cp - drrb->drr_toname);
2323 }
2324 } else {
2325 i = strcspn(drrb->drr_toname, "/@");
2326 }
2327 /* zfs_receive_one() will create_parents() */
2328 (void) strlcat(tofs, &drrb->drr_toname[i],
2329 ZFS_MAXNAMELEN);
2330 *strchr(tofs, '@') = '\0';
2331 }
2332
2333 if (recursive && !flags->dryrun && !flags->nomount) {
2334 VERIFY(0 == nvlist_alloc(&renamed,
2335 NV_UNIQUE_NAME, 0));
2336 }
2337
2338 softerr = recv_incremental_replication(hdl, tofs, flags,
2339 stream_nv, stream_avl, renamed);
2340
2341 /* Unmount renamed filesystems before receiving. */
2342 while ((pair = nvlist_next_nvpair(renamed,
2343 pair)) != NULL) {
2344 zfs_handle_t *zhp;
2345 prop_changelist_t *clp = NULL;
2346
2347 zhp = zfs_open(hdl, nvpair_name(pair),
2348 ZFS_TYPE_FILESYSTEM);
2349 if (zhp != NULL) {
2350 clp = changelist_gather(zhp,
2351 ZFS_PROP_MOUNTPOINT, 0, 0);
2352 zfs_close(zhp);
2353 if (clp != NULL) {
2354 softerr |=
2355 changelist_prefix(clp);
2356 changelist_free(clp);
2357 }
2358 }
2359 }
2360
2361 nvlist_free(renamed);
2362 }
2363 }
2364
2365 /*
2366 * Get the fs specified by the first path in the stream (the top level
2367 * specified by 'zfs send') and pass it to each invocation of
2368 * zfs_receive_one().
2369 */
2370 (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2371 ZFS_MAXNAMELEN);
2372 if ((cp = strchr(sendfs, '@')) != NULL)
2373 *cp = '\0';
2374
2375 /* Finally, receive each contained stream */
2376 do {
2377 /*
2378 * we should figure out if it has a recoverable
2379 * error, in which case do a recv_skip() and drive on.
2380 * Note, if we fail due to already having this guid,
2381 * zfs_receive_one() will take care of it (ie,
2382 * recv_skip() and return 0).
2383 */
2384 error = zfs_receive_impl(hdl, destname, flags, fd,
2385 sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2386 action_handlep);
2387 if (error == ENODATA) {
2388 error = 0;
2389 break;
2390 }
2391 anyerr |= error;
2392 } while (error == 0);
2393
2394 if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2395 /*
2396 * Now that we have the fs's they sent us, try the
2397 * renames again.
2398 */
2399 softerr = recv_incremental_replication(hdl, tofs, flags,
2400 stream_nv, stream_avl, NULL);
2401 }
2402
2403 out:
2404 fsavl_destroy(stream_avl);
2405 if (stream_nv)
2406 nvlist_free(stream_nv);
2407 if (softerr)
2408 error = -2;
2409 if (anyerr)
2410 error = -1;
2411 return (error);
2412 }
2413
2414 static void
2415 trunc_prop_errs(int truncated)
2416 {
2417 ASSERT(truncated != 0);
2418
2419 if (truncated == 1)
2420 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2421 "1 more property could not be set\n"));
2422 else
2423 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2424 "%d more properties could not be set\n"), truncated);
2425 }
2426
2427 static int
2428 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2429 {
2430 dmu_replay_record_t *drr;
2431 void *buf = malloc(1<<20);
2432 char errbuf[1024];
2433
2434 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2435 "cannot receive:"));
2436
2437 /* XXX would be great to use lseek if possible... */
2438 drr = buf;
2439
2440 while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2441 byteswap, NULL) == 0) {
2442 if (byteswap)
2443 drr->drr_type = BSWAP_32(drr->drr_type);
2444
2445 switch (drr->drr_type) {
2446 case DRR_BEGIN:
2447 /* NB: not to be used on v2 stream packages */
2448 if (drr->drr_payloadlen != 0) {
2449 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2450 "invalid substream header"));
2451 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2452 }
2453 break;
2454
2455 case DRR_END:
2456 free(buf);
2457 return (0);
2458
2459 case DRR_OBJECT:
2460 if (byteswap) {
2461 drr->drr_u.drr_object.drr_bonuslen =
2462 BSWAP_32(drr->drr_u.drr_object.
2463 drr_bonuslen);
2464 }
2465 (void) recv_read(hdl, fd, buf,
2466 P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2467 B_FALSE, NULL);
2468 break;
2469
2470 case DRR_WRITE:
2471 if (byteswap) {
2472 drr->drr_u.drr_write.drr_length =
2473 BSWAP_64(drr->drr_u.drr_write.drr_length);
2474 }
2475 (void) recv_read(hdl, fd, buf,
2476 drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2477 break;
2478 case DRR_SPILL:
2479 if (byteswap) {
2480 drr->drr_u.drr_write.drr_length =
2481 BSWAP_64(drr->drr_u.drr_spill.drr_length);
2482 }
2483 (void) recv_read(hdl, fd, buf,
2484 drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2485 break;
2486 case DRR_WRITE_BYREF:
2487 case DRR_FREEOBJECTS:
2488 case DRR_FREE:
2489 break;
2490
2491 default:
2492 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2493 "invalid record type"));
2494 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2495 }
2496 }
2497
2498 free(buf);
2499 return (-1);
2500 }
2501
2502 /*
2503 * Restores a backup of tosnap from the file descriptor specified by infd.
2504 */
2505 static int
2506 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2507 recvflags_t *flags, dmu_replay_record_t *drr,
2508 dmu_replay_record_t *drr_noswap, const char *sendfs,
2509 nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2510 uint64_t *action_handlep)
2511 {
2512 zfs_cmd_t zc = { 0 };
2513 time_t begin_time;
2514 int ioctl_err, ioctl_errno, err;
2515 char *cp;
2516 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2517 char errbuf[1024];
2518 char prop_errbuf[1024];
2519 const char *chopprefix;
2520 boolean_t newfs = B_FALSE;
2521 boolean_t stream_wantsnewfs;
2522 uint64_t parent_snapguid = 0;
2523 prop_changelist_t *clp = NULL;
2524 nvlist_t *snapprops_nvlist = NULL;
2525 zprop_errflags_t prop_errflags;
2526 boolean_t recursive;
2527
2528 begin_time = time(NULL);
2529
2530 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2531 "cannot receive"));
2532
2533 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2534 ENOENT);
2535
2536 if (stream_avl != NULL) {
2537 char *snapname;
2538 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2539 &snapname);
2540 nvlist_t *props;
2541 int ret;
2542
2543 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2544 &parent_snapguid);
2545 err = nvlist_lookup_nvlist(fs, "props", &props);
2546 if (err)
2547 VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2548
2549 if (flags->canmountoff) {
2550 VERIFY(0 == nvlist_add_uint64(props,
2551 zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2552 }
2553 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2554 if (err)
2555 nvlist_free(props);
2556
2557 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2558 VERIFY(0 == nvlist_lookup_nvlist(props,
2559 snapname, &snapprops_nvlist));
2560 }
2561
2562 if (ret != 0)
2563 return (-1);
2564 }
2565
2566 cp = NULL;
2567
2568 /*
2569 * Determine how much of the snapshot name stored in the stream
2570 * we are going to tack on to the name they specified on the
2571 * command line, and how much we are going to chop off.
2572 *
2573 * If they specified a snapshot, chop the entire name stored in
2574 * the stream.
2575 */
2576 if (flags->istail) {
2577 /*
2578 * A filesystem was specified with -e. We want to tack on only
2579 * the tail of the sent snapshot path.
2580 */
2581 if (strchr(tosnap, '@')) {
2582 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2583 "argument - snapshot not allowed with -e"));
2584 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2585 }
2586
2587 chopprefix = strrchr(sendfs, '/');
2588
2589 if (chopprefix == NULL) {
2590 /*
2591 * The tail is the poolname, so we need to
2592 * prepend a path separator.
2593 */
2594 int len = strlen(drrb->drr_toname);
2595 cp = malloc(len + 2);
2596 cp[0] = '/';
2597 (void) strcpy(&cp[1], drrb->drr_toname);
2598 chopprefix = cp;
2599 } else {
2600 chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2601 }
2602 } else if (flags->isprefix) {
2603 /*
2604 * A filesystem was specified with -d. We want to tack on
2605 * everything but the first element of the sent snapshot path
2606 * (all but the pool name).
2607 */
2608 if (strchr(tosnap, '@')) {
2609 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2610 "argument - snapshot not allowed with -d"));
2611 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2612 }
2613
2614 chopprefix = strchr(drrb->drr_toname, '/');
2615 if (chopprefix == NULL)
2616 chopprefix = strchr(drrb->drr_toname, '@');
2617 } else if (strchr(tosnap, '@') == NULL) {
2618 /*
2619 * If a filesystem was specified without -d or -e, we want to
2620 * tack on everything after the fs specified by 'zfs send'.
2621 */
2622 chopprefix = drrb->drr_toname + strlen(sendfs);
2623 } else {
2624 /* A snapshot was specified as an exact path (no -d or -e). */
2625 if (recursive) {
2626 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2627 "cannot specify snapshot name for multi-snapshot "
2628 "stream"));
2629 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2630 }
2631 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2632 }
2633
2634 ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2635 ASSERT(chopprefix > drrb->drr_toname);
2636 ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2637 ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2638 chopprefix[0] == '\0');
2639
2640 /*
2641 * Determine name of destination snapshot, store in zc_value.
2642 */
2643 (void) strcpy(zc.zc_value, tosnap);
2644 (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2645 free(cp);
2646 if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2647 zcmd_free_nvlists(&zc);
2648 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2649 }
2650
2651 /*
2652 * Determine the name of the origin snapshot, store in zc_string.
2653 */
2654 if (drrb->drr_flags & DRR_FLAG_CLONE) {
2655 if (guid_to_name(hdl, zc.zc_value,
2656 drrb->drr_fromguid, zc.zc_string) != 0) {
2657 zcmd_free_nvlists(&zc);
2658 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2659 "local origin for clone %s does not exist"),
2660 zc.zc_value);
2661 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2662 }
2663 if (flags->verbose)
2664 (void) printf("found clone origin %s\n", zc.zc_string);
2665 }
2666
2667 stream_wantsnewfs = (drrb->drr_fromguid == NULL ||
2668 (drrb->drr_flags & DRR_FLAG_CLONE));
2669
2670 if (stream_wantsnewfs) {
2671 /*
2672 * if the parent fs does not exist, look for it based on
2673 * the parent snap GUID
2674 */
2675 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2676 "cannot receive new filesystem stream"));
2677
2678 (void) strcpy(zc.zc_name, zc.zc_value);
2679 cp = strrchr(zc.zc_name, '/');
2680 if (cp)
2681 *cp = '\0';
2682 if (cp &&
2683 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2684 char suffix[ZFS_MAXNAMELEN];
2685 (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2686 if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
2687 zc.zc_value) == 0) {
2688 *strchr(zc.zc_value, '@') = '\0';
2689 (void) strcat(zc.zc_value, suffix);
2690 }
2691 }
2692 } else {
2693 /*
2694 * if the fs does not exist, look for it based on the
2695 * fromsnap GUID
2696 */
2697 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2698 "cannot receive incremental stream"));
2699
2700 (void) strcpy(zc.zc_name, zc.zc_value);
2701 *strchr(zc.zc_name, '@') = '\0';
2702
2703 /*
2704 * If the exact receive path was specified and this is the
2705 * topmost path in the stream, then if the fs does not exist we
2706 * should look no further.
2707 */
2708 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
2709 strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2710 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2711 char snap[ZFS_MAXNAMELEN];
2712 (void) strcpy(snap, strchr(zc.zc_value, '@'));
2713 if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
2714 zc.zc_value) == 0) {
2715 *strchr(zc.zc_value, '@') = '\0';
2716 (void) strcat(zc.zc_value, snap);
2717 }
2718 }
2719 }
2720
2721 (void) strcpy(zc.zc_name, zc.zc_value);
2722 *strchr(zc.zc_name, '@') = '\0';
2723
2724 if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2725 zfs_handle_t *zhp;
2726
2727 /*
2728 * Destination fs exists. Therefore this should either
2729 * be an incremental, or the stream specifies a new fs
2730 * (full stream or clone) and they want us to blow it
2731 * away (and have therefore specified -F and removed any
2732 * snapshots).
2733 */
2734 if (stream_wantsnewfs) {
2735 if (!flags->force) {
2736 zcmd_free_nvlists(&zc);
2737 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2738 "destination '%s' exists\n"
2739 "must specify -F to overwrite it"),
2740 zc.zc_name);
2741 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2742 }
2743 if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2744 &zc) == 0) {
2745 zcmd_free_nvlists(&zc);
2746 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2747 "destination has snapshots (eg. %s)\n"
2748 "must destroy them to overwrite it"),
2749 zc.zc_name);
2750 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2751 }
2752 }
2753
2754 if ((zhp = zfs_open(hdl, zc.zc_name,
2755 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2756 zcmd_free_nvlists(&zc);
2757 return (-1);
2758 }
2759
2760 if (stream_wantsnewfs &&
2761 zhp->zfs_dmustats.dds_origin[0]) {
2762 zcmd_free_nvlists(&zc);
2763 zfs_close(zhp);
2764 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2765 "destination '%s' is a clone\n"
2766 "must destroy it to overwrite it"),
2767 zc.zc_name);
2768 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2769 }
2770
2771 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2772 stream_wantsnewfs) {
2773 /* We can't do online recv in this case */
2774 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2775 if (clp == NULL) {
2776 zfs_close(zhp);
2777 zcmd_free_nvlists(&zc);
2778 return (-1);
2779 }
2780 if (changelist_prefix(clp) != 0) {
2781 changelist_free(clp);
2782 zfs_close(zhp);
2783 zcmd_free_nvlists(&zc);
2784 return (-1);
2785 }
2786 }
2787 zfs_close(zhp);
2788 } else {
2789 /*
2790 * Destination filesystem does not exist. Therefore we better
2791 * be creating a new filesystem (either from a full backup, or
2792 * a clone). It would therefore be invalid if the user
2793 * specified only the pool name (i.e. if the destination name
2794 * contained no slash character).
2795 */
2796 if (!stream_wantsnewfs ||
2797 (cp = strrchr(zc.zc_name, '/')) == NULL) {
2798 zcmd_free_nvlists(&zc);
2799 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2800 "destination '%s' does not exist"), zc.zc_name);
2801 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2802 }
2803
2804 /*
2805 * Trim off the final dataset component so we perform the
2806 * recvbackup ioctl to the filesystems's parent.
2807 */
2808 *cp = '\0';
2809
2810 if (flags->isprefix && !flags->istail && !flags->dryrun &&
2811 create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2812 zcmd_free_nvlists(&zc);
2813 return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2814 }
2815
2816 newfs = B_TRUE;
2817 }
2818
2819 zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2820 zc.zc_cookie = infd;
2821 zc.zc_guid = flags->force;
2822 if (flags->verbose) {
2823 (void) printf("%s %s stream of %s into %s\n",
2824 flags->dryrun ? "would receive" : "receiving",
2825 drrb->drr_fromguid ? "incremental" : "full",
2826 drrb->drr_toname, zc.zc_value);
2827 (void) fflush(stdout);
2828 }
2829
2830 if (flags->dryrun) {
2831 zcmd_free_nvlists(&zc);
2832 return (recv_skip(hdl, infd, flags->byteswap));
2833 }
2834
2835 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2836 zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2837 zc.zc_cleanup_fd = cleanup_fd;
2838 zc.zc_action_handle = *action_handlep;
2839
2840 err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2841 ioctl_errno = errno;
2842 prop_errflags = (zprop_errflags_t)zc.zc_obj;
2843
2844 if (err == 0) {
2845 nvlist_t *prop_errors;
2846 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2847 zc.zc_nvlist_dst_size, &prop_errors, 0));
2848
2849 nvpair_t *prop_err = NULL;
2850
2851 while ((prop_err = nvlist_next_nvpair(prop_errors,
2852 prop_err)) != NULL) {
2853 char tbuf[1024];
2854 zfs_prop_t prop;
2855 int intval;
2856
2857 prop = zfs_name_to_prop(nvpair_name(prop_err));
2858 (void) nvpair_value_int32(prop_err, &intval);
2859 if (strcmp(nvpair_name(prop_err),
2860 ZPROP_N_MORE_ERRORS) == 0) {
2861 trunc_prop_errs(intval);
2862 break;
2863 } else {
2864 (void) snprintf(tbuf, sizeof (tbuf),
2865 dgettext(TEXT_DOMAIN,
2866 "cannot receive %s property on %s"),
2867 nvpair_name(prop_err), zc.zc_name);
2868 zfs_setprop_error(hdl, prop, intval, tbuf);
2869 }
2870 }
2871 nvlist_free(prop_errors);
2872 }
2873
2874 zc.zc_nvlist_dst = 0;
2875 zc.zc_nvlist_dst_size = 0;
2876 zcmd_free_nvlists(&zc);
2877
2878 if (err == 0 && snapprops_nvlist) {
2879 zfs_cmd_t zc2 = { 0 };
2880
2881 (void) strcpy(zc2.zc_name, zc.zc_value);
2882 zc2.zc_cookie = B_TRUE; /* received */
2883 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
2884 (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
2885 zcmd_free_nvlists(&zc2);
2886 }
2887 }
2888
2889 if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
2890 /*
2891 * It may be that this snapshot already exists,
2892 * in which case we want to consume & ignore it
2893 * rather than failing.
2894 */
2895 avl_tree_t *local_avl;
2896 nvlist_t *local_nv, *fs;
2897 cp = strchr(zc.zc_value, '@');
2898
2899 /*
2900 * XXX Do this faster by just iterating over snaps in
2901 * this fs. Also if zc_value does not exist, we will
2902 * get a strange "does not exist" error message.
2903 */
2904 *cp = '\0';
2905 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
2906 &local_nv, &local_avl) == 0) {
2907 *cp = '@';
2908 fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
2909 fsavl_destroy(local_avl);
2910 nvlist_free(local_nv);
2911
2912 if (fs != NULL) {
2913 if (flags->verbose) {
2914 (void) printf("snap %s already exists; "
2915 "ignoring\n", zc.zc_value);
2916 }
2917 err = ioctl_err = recv_skip(hdl, infd,
2918 flags->byteswap);
2919 }
2920 }
2921 *cp = '@';
2922 }
2923
2924 if (ioctl_err != 0) {
2925 switch (ioctl_errno) {
2926 case ENODEV:
2927 cp = strchr(zc.zc_value, '@');
2928 *cp = '\0';
2929 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2930 "most recent snapshot of %s does not\n"
2931 "match incremental source"), zc.zc_value);
2932 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2933 *cp = '@';
2934 break;
2935 case ETXTBSY:
2936 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2937 "destination %s has been modified\n"
2938 "since most recent snapshot"), zc.zc_name);
2939 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2940 break;
2941 case EEXIST:
2942 cp = strchr(zc.zc_value, '@');
2943 if (newfs) {
2944 /* it's the containing fs that exists */
2945 *cp = '\0';
2946 }
2947 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2948 "destination already exists"));
2949 (void) zfs_error_fmt(hdl, EZFS_EXISTS,
2950 dgettext(TEXT_DOMAIN, "cannot restore to %s"),
2951 zc.zc_value);
2952 *cp = '@';
2953 break;
2954 case EINVAL:
2955 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2956 break;
2957 case ECKSUM:
2958 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2959 "invalid stream (checksum mismatch)"));
2960 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2961 break;
2962 case ENOTSUP:
2963 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2964 "pool must be upgraded to receive this stream."));
2965 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
2966 break;
2967 case EDQUOT:
2968 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2969 "destination %s space quota exceeded"), zc.zc_name);
2970 (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
2971 break;
2972 default:
2973 (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
2974 }
2975 }
2976
2977 /*
2978 * Mount the target filesystem (if created). Also mount any
2979 * children of the target filesystem if we did a replication
2980 * receive (indicated by stream_avl being non-NULL).
2981 */
2982 cp = strchr(zc.zc_value, '@');
2983 if (cp && (ioctl_err == 0 || !newfs)) {
2984 zfs_handle_t *h;
2985
2986 *cp = '\0';
2987 h = zfs_open(hdl, zc.zc_value,
2988 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
2989 if (h != NULL) {
2990 if (h->zfs_type == ZFS_TYPE_VOLUME) {
2991 *cp = '@';
2992 } else if (newfs || stream_avl) {
2993 /*
2994 * Track the first/top of hierarchy fs,
2995 * for mounting and sharing later.
2996 */
2997 if (top_zfs && *top_zfs == NULL)
2998 *top_zfs = zfs_strdup(hdl, zc.zc_value);
2999 }
3000 zfs_close(h);
3001 }
3002 *cp = '@';
3003 }
3004
3005 if (clp) {
3006 err |= changelist_postfix(clp);
3007 changelist_free(clp);
3008 }
3009
3010 if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3011 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3012 "failed to clear unreceived properties on %s"),
3013 zc.zc_name);
3014 (void) fprintf(stderr, "\n");
3015 }
3016 if (prop_errflags & ZPROP_ERR_NORESTORE) {
3017 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3018 "failed to restore original properties on %s"),
3019 zc.zc_name);
3020 (void) fprintf(stderr, "\n");
3021 }
3022
3023 if (err || ioctl_err)
3024 return (-1);
3025
3026 *action_handlep = zc.zc_action_handle;
3027
3028 if (flags->verbose) {
3029 char buf1[64];
3030 char buf2[64];
3031 uint64_t bytes = zc.zc_cookie;
3032 time_t delta = time(NULL) - begin_time;
3033 if (delta == 0)
3034 delta = 1;
3035 zfs_nicenum(bytes, buf1, sizeof (buf1));
3036 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3037
3038 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3039 buf1, delta, buf2);
3040 }
3041
3042 return (0);
3043 }
3044
3045 static int
3046 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3047 int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl,
3048 char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
3049 {
3050 int err;
3051 dmu_replay_record_t drr, drr_noswap;
3052 struct drr_begin *drrb = &drr.drr_u.drr_begin;
3053 char errbuf[1024];
3054 zio_cksum_t zcksum = { 0 };
3055 uint64_t featureflags;
3056 int hdrtype;
3057
3058 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3059 "cannot receive"));
3060
3061 if (flags->isprefix &&
3062 !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3063 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3064 "(%s) does not exist"), tosnap);
3065 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3066 }
3067
3068 /* read in the BEGIN record */
3069 if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3070 &zcksum)))
3071 return (err);
3072
3073 if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3074 /* It's the double end record at the end of a package */
3075 return (ENODATA);
3076 }
3077
3078 /* the kernel needs the non-byteswapped begin record */
3079 drr_noswap = drr;
3080
3081 flags->byteswap = B_FALSE;
3082 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3083 /*
3084 * We computed the checksum in the wrong byteorder in
3085 * recv_read() above; do it again correctly.
3086 */
3087 bzero(&zcksum, sizeof (zio_cksum_t));
3088 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3089 flags->byteswap = B_TRUE;
3090
3091 drr.drr_type = BSWAP_32(drr.drr_type);
3092 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3093 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3094 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3095 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3096 drrb->drr_type = BSWAP_32(drrb->drr_type);
3097 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3098 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3099 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3100 }
3101
3102 if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3103 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3104 "stream (bad magic number)"));
3105 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3106 }
3107
3108 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3109 hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3110
3111 if (!DMU_STREAM_SUPPORTED(featureflags) ||
3112 (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3113 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3114 "stream has unsupported feature, feature flags = %lx"),
3115 featureflags);
3116 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3117 }
3118
3119 if (strchr(drrb->drr_toname, '@') == NULL) {
3120 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3121 "stream (bad snapshot name)"));
3122 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3123 }
3124
3125 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3126 char nonpackage_sendfs[ZFS_MAXNAMELEN];
3127 if (sendfs == NULL) {
3128 /*
3129 * We were not called from zfs_receive_package(). Get
3130 * the fs specified by 'zfs send'.
3131 */
3132 char *cp;
3133 (void) strlcpy(nonpackage_sendfs,
3134 drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
3135 if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3136 *cp = '\0';
3137 sendfs = nonpackage_sendfs;
3138 }
3139 return (zfs_receive_one(hdl, infd, tosnap, flags,
3140 &drr, &drr_noswap, sendfs, stream_nv, stream_avl,
3141 top_zfs, cleanup_fd, action_handlep));
3142 } else {
3143 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3144 DMU_COMPOUNDSTREAM);
3145 return (zfs_receive_package(hdl, infd, tosnap, flags,
3146 &drr, &zcksum, top_zfs, cleanup_fd, action_handlep));
3147 }
3148 }
3149
3150 /*
3151 * Restores a backup of tosnap from the file descriptor specified by infd.
3152 * Return 0 on total success, -2 if some things couldn't be
3153 * destroyed/renamed/promoted, -1 if some things couldn't be received.
3154 * (-1 will override -2).
3155 */
3156 int
3157 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3158 int infd, avl_tree_t *stream_avl)
3159 {
3160 char *top_zfs = NULL;
3161 int err;
3162 int cleanup_fd;
3163 uint64_t action_handle = 0;
3164
3165 cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3166 VERIFY(cleanup_fd >= 0);
3167
3168 err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL,
3169 stream_avl, &top_zfs, cleanup_fd, &action_handle);
3170
3171 VERIFY(0 == close(cleanup_fd));
3172
3173 if (err == 0 && !flags->nomount && top_zfs) {
3174 zfs_handle_t *zhp;
3175 prop_changelist_t *clp;
3176
3177 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3178 if (zhp != NULL) {
3179 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3180 CL_GATHER_MOUNT_ALWAYS, 0);
3181 zfs_close(zhp);
3182 if (clp != NULL) {
3183 /* mount and share received datasets */
3184 err = changelist_postfix(clp);
3185 changelist_free(clp);
3186 }
3187 }
3188 if (zhp == NULL || clp == NULL || err)
3189 err = -1;
3190 }
3191 if (top_zfs)
3192 free(top_zfs);
3193
3194 return (err);
3195 }