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) 2011, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2012 Pawel Jakub Dawidek. All rights reserved.
27 * Copyright (c) 2013 Steven Hartland. All rights reserved.
28 * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
31 */
32
33 #include <assert.h>
34 #include <ctype.h>
35 #include <errno.h>
36 #include <libintl.h>
37 #include <stdio.h>
38 #include <stdlib.h>
39 #include <strings.h>
40 #include <unistd.h>
41 #include <stddef.h>
42 #include <fcntl.h>
43 #include <sys/mount.h>
44 #include <pthread.h>
45 #include <umem.h>
46 #include <time.h>
47
48 #include <libzfs.h>
49 #include <libzfs_core.h>
50
51 #include "zfs_namecheck.h"
52 #include "zfs_prop.h"
53 #include "zfs_fletcher.h"
54 #include "libzfs_impl.h"
55 #include <zlib.h>
56 #include <sha2.h>
57 #include <sys/zio_checksum.h>
58 #include <sys/ddt.h>
59
60 /* in libzfs_dataset.c */
61 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
62
63 static int zfs_receive_impl(libzfs_handle_t *, const char *, const char *,
64 recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **, int,
65 uint64_t *, const char *);
66 static int guid_to_name(libzfs_handle_t *, const char *,
67 uint64_t, boolean_t, char *);
68
69 static const zio_cksum_t zero_cksum = { 0 };
70
71 typedef struct dedup_arg {
72 int inputfd;
73 int outputfd;
74 libzfs_handle_t *dedup_hdl;
75 } dedup_arg_t;
76
77 typedef struct progress_arg {
78 zfs_handle_t *pa_zhp;
79 int pa_fd;
80 boolean_t pa_parsable;
81 } progress_arg_t;
82
83 typedef struct dataref {
84 uint64_t ref_guid;
85 uint64_t ref_object;
86 uint64_t ref_offset;
87 } dataref_t;
88
89 typedef struct dedup_entry {
90 struct dedup_entry *dde_next;
91 zio_cksum_t dde_chksum;
92 uint64_t dde_prop;
93 dataref_t dde_ref;
94 } dedup_entry_t;
95
96 #define MAX_DDT_PHYSMEM_PERCENT 20
97 #define SMALLEST_POSSIBLE_MAX_DDT_MB 128
98
99 typedef struct dedup_table {
100 dedup_entry_t **dedup_hash_array;
101 umem_cache_t *ddecache;
102 uint64_t max_ddt_size; /* max dedup table size in bytes */
103 uint64_t cur_ddt_size; /* current dedup table size in bytes */
104 uint64_t ddt_count;
105 int numhashbits;
106 boolean_t ddt_full;
107 } dedup_table_t;
108
109 static int
110 high_order_bit(uint64_t n)
111 {
112 int count;
113
114 for (count = 0; n != 0; count++)
115 n >>= 1;
116 return (count);
117 }
118
119 static size_t
120 ssread(void *buf, size_t len, FILE *stream)
121 {
122 size_t outlen;
123
124 if ((outlen = fread(buf, len, 1, stream)) == 0)
125 return (0);
126
127 return (outlen);
128 }
129
130 static void
131 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
132 zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
133 {
134 dedup_entry_t *dde;
135
136 if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
137 if (ddt->ddt_full == B_FALSE) {
138 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
139 "Dedup table full. Deduplication will continue "
140 "with existing table entries"));
141 ddt->ddt_full = B_TRUE;
142 }
143 return;
144 }
145
146 if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
147 != NULL) {
148 assert(*ddepp == NULL);
149 dde->dde_next = NULL;
150 dde->dde_chksum = *cs;
151 dde->dde_prop = prop;
152 dde->dde_ref = *dr;
153 *ddepp = dde;
154 ddt->cur_ddt_size += sizeof (dedup_entry_t);
155 ddt->ddt_count++;
156 }
157 }
158
159 /*
160 * Using the specified dedup table, do a lookup for an entry with
161 * the checksum cs. If found, return the block's reference info
162 * in *dr. Otherwise, insert a new entry in the dedup table, using
163 * the reference information specified by *dr.
164 *
165 * return value: true - entry was found
166 * false - entry was not found
167 */
168 static boolean_t
169 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
170 uint64_t prop, dataref_t *dr)
171 {
172 uint32_t hashcode;
173 dedup_entry_t **ddepp;
174
175 hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
176
177 for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
178 ddepp = &((*ddepp)->dde_next)) {
179 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
180 (*ddepp)->dde_prop == prop) {
181 *dr = (*ddepp)->dde_ref;
182 return (B_TRUE);
183 }
184 }
185 ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
186 return (B_FALSE);
187 }
188
189 static int
190 dump_record(dmu_replay_record_t *drr, void *payload, int payload_len,
191 zio_cksum_t *zc, int outfd)
192 {
193 ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
194 ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t));
195 (void) fletcher_4_incremental_native(drr,
196 offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc);
197 if (drr->drr_type != DRR_BEGIN) {
198 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u.
199 drr_checksum.drr_checksum));
200 drr->drr_u.drr_checksum.drr_checksum = *zc;
201 }
202 (void) fletcher_4_incremental_native(
203 &drr->drr_u.drr_checksum.drr_checksum, sizeof (zio_cksum_t), zc);
204 if (write(outfd, drr, sizeof (*drr)) == -1)
205 return (errno);
206 if (payload_len != 0) {
207 (void) fletcher_4_incremental_native(payload, payload_len, zc);
208 if (write(outfd, payload, payload_len) == -1)
209 return (errno);
210 }
211 return (0);
212 }
213
214 /*
215 * This function is started in a separate thread when the dedup option
216 * has been requested. The main send thread determines the list of
217 * snapshots to be included in the send stream and makes the ioctl calls
218 * for each one. But instead of having the ioctl send the output to the
219 * the output fd specified by the caller of zfs_send()), the
220 * ioctl is told to direct the output to a pipe, which is read by the
221 * alternate thread running THIS function. This function does the
222 * dedup'ing by:
223 * 1. building a dedup table (the DDT)
224 * 2. doing checksums on each data block and inserting a record in the DDT
225 * 3. looking for matching checksums, and
226 * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever
227 * a duplicate block is found.
228 * The output of this function then goes to the output fd requested
229 * by the caller of zfs_send().
230 */
231 static void *
232 cksummer(void *arg)
233 {
234 dedup_arg_t *dda = arg;
235 char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE);
236 dmu_replay_record_t thedrr;
237 dmu_replay_record_t *drr = &thedrr;
238 FILE *ofp;
239 int outfd;
240 dedup_table_t ddt;
241 zio_cksum_t stream_cksum;
242 uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
243 uint64_t numbuckets;
244
245 ddt.max_ddt_size =
246 MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100,
247 SMALLEST_POSSIBLE_MAX_DDT_MB << 20);
248
249 numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t));
250
251 /*
252 * numbuckets must be a power of 2. Increase number to
253 * a power of 2 if necessary.
254 */
255 if (!ISP2(numbuckets))
256 numbuckets = 1 << high_order_bit(numbuckets);
257
258 ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
259 ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
260 NULL, NULL, NULL, NULL, NULL, 0);
261 ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
262 ddt.numhashbits = high_order_bit(numbuckets) - 1;
263 ddt.ddt_full = B_FALSE;
264
265 outfd = dda->outputfd;
266 ofp = fdopen(dda->inputfd, "r");
267 while (ssread(drr, sizeof (*drr), ofp) != 0) {
268
269 /*
270 * kernel filled in checksum, we are going to write same
271 * record, but need to regenerate checksum.
272 */
273 if (drr->drr_type != DRR_BEGIN) {
274 bzero(&drr->drr_u.drr_checksum.drr_checksum,
275 sizeof (drr->drr_u.drr_checksum.drr_checksum));
276 }
277
278 switch (drr->drr_type) {
279 case DRR_BEGIN:
280 {
281 struct drr_begin *drrb = &drr->drr_u.drr_begin;
282 int fflags;
283 int sz = 0;
284 ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
285
286 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
287
288 /* set the DEDUP feature flag for this stream */
289 fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
290 fflags |= (DMU_BACKUP_FEATURE_DEDUP |
291 DMU_BACKUP_FEATURE_DEDUPPROPS);
292 DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
293
294 if (drr->drr_payloadlen != 0) {
295 sz = drr->drr_payloadlen;
296
297 if (sz > SPA_MAXBLOCKSIZE) {
298 buf = zfs_realloc(dda->dedup_hdl, buf,
299 SPA_MAXBLOCKSIZE, sz);
300 }
301 (void) ssread(buf, sz, ofp);
302 if (ferror(stdin))
303 perror("fread");
304 }
305 if (dump_record(drr, buf, sz, &stream_cksum,
306 outfd) != 0)
307 goto out;
308 break;
309 }
310
311 case DRR_END:
312 {
313 struct drr_end *drre = &drr->drr_u.drr_end;
314 /* use the recalculated checksum */
315 drre->drr_checksum = stream_cksum;
316 if (dump_record(drr, NULL, 0, &stream_cksum,
317 outfd) != 0)
318 goto out;
319 break;
320 }
321
322 case DRR_OBJECT:
323 {
324 struct drr_object *drro = &drr->drr_u.drr_object;
325 if (drro->drr_bonuslen > 0) {
326 (void) ssread(buf,
327 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
328 ofp);
329 }
330 if (dump_record(drr, buf,
331 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
332 &stream_cksum, outfd) != 0)
333 goto out;
334 break;
335 }
336
337 case DRR_SPILL:
338 {
339 struct drr_spill *drrs = &drr->drr_u.drr_spill;
340 (void) ssread(buf, drrs->drr_length, ofp);
341 if (dump_record(drr, buf, drrs->drr_length,
342 &stream_cksum, outfd) != 0)
343 goto out;
344 break;
345 }
346
347 case DRR_FREEOBJECTS:
348 {
349 if (dump_record(drr, NULL, 0, &stream_cksum,
350 outfd) != 0)
351 goto out;
352 break;
353 }
354
355 case DRR_WRITE:
356 {
357 struct drr_write *drrw = &drr->drr_u.drr_write;
358 dataref_t dataref;
359 uint64_t payload_size;
360
361 payload_size = DRR_WRITE_PAYLOAD_SIZE(drrw);
362 (void) ssread(buf, payload_size, ofp);
363
364 /*
365 * Use the existing checksum if it's dedup-capable,
366 * else calculate a SHA256 checksum for it.
367 */
368
369 if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
370 zero_cksum) ||
371 !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
372 SHA256_CTX ctx;
373 zio_cksum_t tmpsha256;
374
375 SHA256Init(&ctx);
376 SHA256Update(&ctx, buf, payload_size);
377 SHA256Final(&tmpsha256, &ctx);
378 drrw->drr_key.ddk_cksum.zc_word[0] =
379 BE_64(tmpsha256.zc_word[0]);
380 drrw->drr_key.ddk_cksum.zc_word[1] =
381 BE_64(tmpsha256.zc_word[1]);
382 drrw->drr_key.ddk_cksum.zc_word[2] =
383 BE_64(tmpsha256.zc_word[2]);
384 drrw->drr_key.ddk_cksum.zc_word[3] =
385 BE_64(tmpsha256.zc_word[3]);
386 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
387 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
388 }
389
390 dataref.ref_guid = drrw->drr_toguid;
391 dataref.ref_object = drrw->drr_object;
392 dataref.ref_offset = drrw->drr_offset;
393
394 if (ddt_update(dda->dedup_hdl, &ddt,
395 &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
396 &dataref)) {
397 dmu_replay_record_t wbr_drr = {0};
398 struct drr_write_byref *wbr_drrr =
399 &wbr_drr.drr_u.drr_write_byref;
400
401 /* block already present in stream */
402 wbr_drr.drr_type = DRR_WRITE_BYREF;
403
404 wbr_drrr->drr_object = drrw->drr_object;
405 wbr_drrr->drr_offset = drrw->drr_offset;
406 wbr_drrr->drr_length = drrw->drr_logical_size;
407 wbr_drrr->drr_toguid = drrw->drr_toguid;
408 wbr_drrr->drr_refguid = dataref.ref_guid;
409 wbr_drrr->drr_refobject =
410 dataref.ref_object;
411 wbr_drrr->drr_refoffset =
412 dataref.ref_offset;
413
414 wbr_drrr->drr_checksumtype =
415 drrw->drr_checksumtype;
416 wbr_drrr->drr_checksumflags =
417 drrw->drr_checksumtype;
418 wbr_drrr->drr_key.ddk_cksum =
419 drrw->drr_key.ddk_cksum;
420 wbr_drrr->drr_key.ddk_prop =
421 drrw->drr_key.ddk_prop;
422
423 if (dump_record(&wbr_drr, NULL, 0,
424 &stream_cksum, outfd) != 0)
425 goto out;
426 } else {
427 /* block not previously seen */
428 if (dump_record(drr, buf, payload_size,
429 &stream_cksum, outfd) != 0)
430 goto out;
431 }
432 break;
433 }
434
435 case DRR_WRITE_EMBEDDED:
436 {
437 struct drr_write_embedded *drrwe =
438 &drr->drr_u.drr_write_embedded;
439 (void) ssread(buf,
440 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp);
441 if (dump_record(drr, buf,
442 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8),
443 &stream_cksum, outfd) != 0)
444 goto out;
445 break;
446 }
447
448 case DRR_FREE:
449 {
450 if (dump_record(drr, NULL, 0, &stream_cksum,
451 outfd) != 0)
452 goto out;
453 break;
454 }
455
456 default:
457 (void) fprintf(stderr, "INVALID record type 0x%x\n",
458 drr->drr_type);
459 /* should never happen, so assert */
460 assert(B_FALSE);
461 }
462 }
463 out:
464 umem_cache_destroy(ddt.ddecache);
465 free(ddt.dedup_hash_array);
466 free(buf);
467 (void) fclose(ofp);
468
469 return (NULL);
470 }
471
472 /*
473 * Routines for dealing with the AVL tree of fs-nvlists
474 */
475 typedef struct fsavl_node {
476 avl_node_t fn_node;
477 nvlist_t *fn_nvfs;
478 char *fn_snapname;
479 uint64_t fn_guid;
480 } fsavl_node_t;
481
482 static int
483 fsavl_compare(const void *arg1, const void *arg2)
484 {
485 const fsavl_node_t *fn1 = arg1;
486 const fsavl_node_t *fn2 = arg2;
487
488 if (fn1->fn_guid > fn2->fn_guid)
489 return (+1);
490 else if (fn1->fn_guid < fn2->fn_guid)
491 return (-1);
492 else
493 return (0);
494 }
495
496 /*
497 * Given the GUID of a snapshot, find its containing filesystem and
498 * (optionally) name.
499 */
500 static nvlist_t *
501 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
502 {
503 fsavl_node_t fn_find;
504 fsavl_node_t *fn;
505
506 fn_find.fn_guid = snapguid;
507
508 fn = avl_find(avl, &fn_find, NULL);
509 if (fn) {
510 if (snapname)
511 *snapname = fn->fn_snapname;
512 return (fn->fn_nvfs);
513 }
514 return (NULL);
515 }
516
517 static void
518 fsavl_destroy(avl_tree_t *avl)
519 {
520 fsavl_node_t *fn;
521 void *cookie;
522
523 if (avl == NULL)
524 return;
525
526 cookie = NULL;
527 while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
528 free(fn);
529 avl_destroy(avl);
530 free(avl);
531 }
532
533 /*
534 * Given an nvlist, produce an avl tree of snapshots, ordered by guid
535 */
536 static avl_tree_t *
537 fsavl_create(nvlist_t *fss)
538 {
539 avl_tree_t *fsavl;
540 nvpair_t *fselem = NULL;
541
542 if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
543 return (NULL);
544
545 avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
546 offsetof(fsavl_node_t, fn_node));
547
548 while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
549 nvlist_t *nvfs, *snaps;
550 nvpair_t *snapelem = NULL;
551
552 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
553 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
554
555 while ((snapelem =
556 nvlist_next_nvpair(snaps, snapelem)) != NULL) {
557 fsavl_node_t *fn;
558 uint64_t guid;
559
560 VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
561 if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
562 fsavl_destroy(fsavl);
563 return (NULL);
564 }
565 fn->fn_nvfs = nvfs;
566 fn->fn_snapname = nvpair_name(snapelem);
567 fn->fn_guid = guid;
568
569 /*
570 * Note: if there are multiple snaps with the
571 * same GUID, we ignore all but one.
572 */
573 if (avl_find(fsavl, fn, NULL) == NULL)
574 avl_add(fsavl, fn);
575 else
576 free(fn);
577 }
578 }
579
580 return (fsavl);
581 }
582
583 /*
584 * Routines for dealing with the giant nvlist of fs-nvlists, etc.
585 */
586 typedef struct send_data {
587 /*
588 * assigned inside every recursive call,
589 * restored from *_save on return:
590 *
591 * guid of fromsnap snapshot in parent dataset
592 * txg of fromsnap snapshot in current dataset
593 * txg of tosnap snapshot in current dataset
594 */
595
596 uint64_t parent_fromsnap_guid;
597 uint64_t fromsnap_txg;
598 uint64_t tosnap_txg;
599
600 /* the nvlists get accumulated during depth-first traversal */
601 nvlist_t *parent_snaps;
602 nvlist_t *fss;
603 nvlist_t *snapprops;
604
605 /* send-receive configuration, does not change during traversal */
606 const char *fsname;
607 const char *fromsnap;
608 const char *tosnap;
609 boolean_t recursive;
610 boolean_t verbose;
611
612 /*
613 * The header nvlist is of the following format:
614 * {
615 * "tosnap" -> string
616 * "fromsnap" -> string (if incremental)
617 * "fss" -> {
618 * id -> {
619 *
620 * "name" -> string (full name; for debugging)
621 * "parentfromsnap" -> number (guid of fromsnap in parent)
622 *
623 * "props" -> { name -> value (only if set here) }
624 * "snaps" -> { name (lastname) -> number (guid) }
625 * "snapprops" -> { name (lastname) -> { name -> value } }
626 *
627 * "origin" -> number (guid) (if clone)
628 * "sent" -> boolean (not on-disk)
629 * }
630 * }
631 * }
632 *
633 */
634 } send_data_t;
635
636 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
637
638 static int
639 send_iterate_snap(zfs_handle_t *zhp, void *arg)
640 {
641 send_data_t *sd = arg;
642 uint64_t guid = zhp->zfs_dmustats.dds_guid;
643 uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
644 char *snapname;
645 nvlist_t *nv;
646
647 snapname = strrchr(zhp->zfs_name, '@')+1;
648
649 if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
650 if (sd->verbose) {
651 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
652 "skipping snapshot %s because it was created "
653 "after the destination snapshot (%s)\n"),
654 zhp->zfs_name, sd->tosnap);
655 }
656 zfs_close(zhp);
657 return (0);
658 }
659
660 VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
661 /*
662 * NB: if there is no fromsnap here (it's a newly created fs in
663 * an incremental replication), we will substitute the tosnap.
664 */
665 if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
666 (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
667 strcmp(snapname, sd->tosnap) == 0)) {
668 sd->parent_fromsnap_guid = guid;
669 }
670
671 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
672 send_iterate_prop(zhp, nv);
673 VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
674 nvlist_free(nv);
675
676 zfs_close(zhp);
677 return (0);
678 }
679
680 static void
681 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
682 {
683 nvpair_t *elem = NULL;
684
685 while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
686 char *propname = nvpair_name(elem);
687 zfs_prop_t prop = zfs_name_to_prop(propname);
688 nvlist_t *propnv;
689
690 if (!zfs_prop_user(propname)) {
691 /*
692 * Realistically, this should never happen. However,
693 * we want the ability to add DSL properties without
694 * needing to make incompatible version changes. We
695 * need to ignore unknown properties to allow older
696 * software to still send datasets containing these
697 * properties, with the unknown properties elided.
698 */
699 if (prop == ZPROP_INVAL)
700 continue;
701
702 if (zfs_prop_readonly(prop))
703 continue;
704 }
705
706 verify(nvpair_value_nvlist(elem, &propnv) == 0);
707 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
708 prop == ZFS_PROP_REFQUOTA ||
709 prop == ZFS_PROP_REFRESERVATION) {
710 char *source;
711 uint64_t value;
712 verify(nvlist_lookup_uint64(propnv,
713 ZPROP_VALUE, &value) == 0);
714 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
715 continue;
716 /*
717 * May have no source before SPA_VERSION_RECVD_PROPS,
718 * but is still modifiable.
719 */
720 if (nvlist_lookup_string(propnv,
721 ZPROP_SOURCE, &source) == 0) {
722 if ((strcmp(source, zhp->zfs_name) != 0) &&
723 (strcmp(source,
724 ZPROP_SOURCE_VAL_RECVD) != 0))
725 continue;
726 }
727 } else {
728 char *source;
729 if (nvlist_lookup_string(propnv,
730 ZPROP_SOURCE, &source) != 0)
731 continue;
732 if ((strcmp(source, zhp->zfs_name) != 0) &&
733 (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
734 continue;
735 }
736
737 if (zfs_prop_user(propname) ||
738 zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
739 char *value;
740 verify(nvlist_lookup_string(propnv,
741 ZPROP_VALUE, &value) == 0);
742 VERIFY(0 == nvlist_add_string(nv, propname, value));
743 } else {
744 uint64_t value;
745 verify(nvlist_lookup_uint64(propnv,
746 ZPROP_VALUE, &value) == 0);
747 VERIFY(0 == nvlist_add_uint64(nv, propname, value));
748 }
749 }
750 }
751
752 /*
753 * returns snapshot creation txg
754 * and returns 0 if the snapshot does not exist
755 */
756 static uint64_t
757 get_snap_txg(libzfs_handle_t *hdl, const char *fs, const char *snap)
758 {
759 char name[ZFS_MAX_DATASET_NAME_LEN];
760 uint64_t txg = 0;
761
762 if (fs == NULL || fs[0] == '\0' || snap == NULL || snap[0] == '\0')
763 return (txg);
764
765 (void) snprintf(name, sizeof (name), "%s@%s", fs, snap);
766 if (zfs_dataset_exists(hdl, name, ZFS_TYPE_SNAPSHOT)) {
767 zfs_handle_t *zhp = zfs_open(hdl, name, ZFS_TYPE_SNAPSHOT);
768 if (zhp != NULL) {
769 txg = zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG);
770 zfs_close(zhp);
771 }
772 }
773
774 return (txg);
775 }
776
777 /*
778 * recursively generate nvlists describing datasets. See comment
779 * for the data structure send_data_t above for description of contents
780 * of the nvlist.
781 */
782 static int
783 send_iterate_fs(zfs_handle_t *zhp, void *arg)
784 {
785 send_data_t *sd = arg;
786 nvlist_t *nvfs, *nv;
787 int rv = 0;
788 uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
789 uint64_t fromsnap_txg_save = sd->fromsnap_txg;
790 uint64_t tosnap_txg_save = sd->tosnap_txg;
791 uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
792 uint64_t guid = zhp->zfs_dmustats.dds_guid;
793 uint64_t fromsnap_txg, tosnap_txg;
794 char guidstring[64];
795
796 fromsnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->fromsnap);
797 if (fromsnap_txg != 0)
798 sd->fromsnap_txg = fromsnap_txg;
799
800 tosnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->tosnap);
801 if (tosnap_txg != 0)
802 sd->tosnap_txg = tosnap_txg;
803
804 /*
805 * on the send side, if the current dataset does not have tosnap,
806 * perform two additional checks:
807 *
808 * - skip sending the current dataset if it was created later than
809 * the parent tosnap
810 * - return error if the current dataset was created earlier than
811 * the parent tosnap
812 */
813 if (sd->tosnap != NULL && tosnap_txg == 0) {
814 if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
815 if (sd->verbose) {
816 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
817 "skipping dataset %s: snapshot %s does "
818 "not exist\n"), zhp->zfs_name, sd->tosnap);
819 }
820 } else {
821 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
822 "cannot send %s@%s%s: snapshot %s@%s does not "
823 "exist\n"), sd->fsname, sd->tosnap, sd->recursive ?
824 dgettext(TEXT_DOMAIN, " recursively") : "",
825 zhp->zfs_name, sd->tosnap);
826 rv = -1;
827 }
828 goto out;
829 }
830
831 VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
832 VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
833 VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
834 sd->parent_fromsnap_guid));
835
836 if (zhp->zfs_dmustats.dds_origin[0]) {
837 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
838 zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
839 if (origin == NULL) {
840 rv = -1;
841 goto out;
842 }
843 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
844 origin->zfs_dmustats.dds_guid));
845 }
846
847 /* iterate over props */
848 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
849 send_iterate_prop(zhp, nv);
850 VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
851 nvlist_free(nv);
852
853 /* iterate over snaps, and set sd->parent_fromsnap_guid */
854 sd->parent_fromsnap_guid = 0;
855 VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
856 VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
857 (void) zfs_iter_snapshots(zhp, B_FALSE, send_iterate_snap, sd);
858 VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
859 VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
860 nvlist_free(sd->parent_snaps);
861 nvlist_free(sd->snapprops);
862
863 /* add this fs to nvlist */
864 (void) snprintf(guidstring, sizeof (guidstring),
865 "0x%llx", (longlong_t)guid);
866 VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
867 nvlist_free(nvfs);
868
869 /* iterate over children */
870 if (sd->recursive)
871 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
872
873 out:
874 sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
875 sd->fromsnap_txg = fromsnap_txg_save;
876 sd->tosnap_txg = tosnap_txg_save;
877
878 zfs_close(zhp);
879 return (rv);
880 }
881
882 static int
883 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
884 const char *tosnap, boolean_t recursive, boolean_t verbose,
885 nvlist_t **nvlp, avl_tree_t **avlp)
886 {
887 zfs_handle_t *zhp;
888 send_data_t sd = { 0 };
889 int error;
890
891 zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
892 if (zhp == NULL)
893 return (EZFS_BADTYPE);
894
895 VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
896 sd.fsname = fsname;
897 sd.fromsnap = fromsnap;
898 sd.tosnap = tosnap;
899 sd.recursive = recursive;
900 sd.verbose = verbose;
901
902 if ((error = send_iterate_fs(zhp, &sd)) != 0) {
903 nvlist_free(sd.fss);
904 if (avlp != NULL)
905 *avlp = NULL;
906 *nvlp = NULL;
907 return (error);
908 }
909
910 if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
911 nvlist_free(sd.fss);
912 *nvlp = NULL;
913 return (EZFS_NOMEM);
914 }
915
916 *nvlp = sd.fss;
917 return (0);
918 }
919
920 /*
921 * Routines specific to "zfs send"
922 */
923 typedef struct send_dump_data {
924 /* these are all just the short snapname (the part after the @) */
925 const char *fromsnap;
926 const char *tosnap;
927 char prevsnap[ZFS_MAX_DATASET_NAME_LEN];
928 uint64_t prevsnap_obj;
929 boolean_t seenfrom, seento, replicate, doall, fromorigin;
930 boolean_t verbose, dryrun, parsable, progress, embed_data, std_out;
931 boolean_t large_block, compress;
932 int outfd;
933 boolean_t err;
934 nvlist_t *fss;
935 nvlist_t *snapholds;
936 avl_tree_t *fsavl;
937 snapfilter_cb_t *filter_cb;
938 void *filter_cb_arg;
939 nvlist_t *debugnv;
940 char holdtag[ZFS_MAX_DATASET_NAME_LEN];
941 int cleanup_fd;
942 uint64_t size;
943 } send_dump_data_t;
944
945 static int
946 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
947 boolean_t fromorigin, enum lzc_send_flags flags, uint64_t *sizep)
948 {
949 zfs_cmd_t zc = { 0 };
950 libzfs_handle_t *hdl = zhp->zfs_hdl;
951
952 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
953 assert(fromsnap_obj == 0 || !fromorigin);
954
955 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
956 zc.zc_obj = fromorigin;
957 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
958 zc.zc_fromobj = fromsnap_obj;
959 zc.zc_guid = 1; /* estimate flag */
960 zc.zc_flags = flags;
961
962 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
963 char errbuf[1024];
964 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
965 "warning: cannot estimate space for '%s'"), zhp->zfs_name);
966
967 switch (errno) {
968 case EXDEV:
969 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
970 "not an earlier snapshot from the same fs"));
971 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
972
973 case ENOENT:
974 if (zfs_dataset_exists(hdl, zc.zc_name,
975 ZFS_TYPE_SNAPSHOT)) {
976 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
977 "incremental source (@%s) does not exist"),
978 zc.zc_value);
979 }
980 return (zfs_error(hdl, EZFS_NOENT, errbuf));
981
982 case EDQUOT:
983 case EFBIG:
984 case EIO:
985 case ENOLINK:
986 case ENOSPC:
987 case ENOSTR:
988 case ENXIO:
989 case EPIPE:
990 case ERANGE:
991 case EFAULT:
992 case EROFS:
993 zfs_error_aux(hdl, strerror(errno));
994 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
995
996 default:
997 return (zfs_standard_error(hdl, errno, errbuf));
998 }
999 }
1000
1001 *sizep = zc.zc_objset_type;
1002
1003 return (0);
1004 }
1005
1006 /*
1007 * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
1008 * NULL) to the file descriptor specified by outfd.
1009 */
1010 static int
1011 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
1012 boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
1013 nvlist_t *debugnv)
1014 {
1015 zfs_cmd_t zc = { 0 };
1016 libzfs_handle_t *hdl = zhp->zfs_hdl;
1017 nvlist_t *thisdbg;
1018
1019 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
1020 assert(fromsnap_obj == 0 || !fromorigin);
1021
1022 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1023 zc.zc_cookie = outfd;
1024 zc.zc_obj = fromorigin;
1025 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1026 zc.zc_fromobj = fromsnap_obj;
1027 zc.zc_flags = flags;
1028
1029 VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
1030 if (fromsnap && fromsnap[0] != '\0') {
1031 VERIFY(0 == nvlist_add_string(thisdbg,
1032 "fromsnap", fromsnap));
1033 }
1034
1035 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
1036 char errbuf[1024];
1037 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1038 "warning: cannot send '%s'"), zhp->zfs_name);
1039
1040 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
1041 if (debugnv) {
1042 VERIFY(0 == nvlist_add_nvlist(debugnv,
1043 zhp->zfs_name, thisdbg));
1044 }
1045 nvlist_free(thisdbg);
1046
1047 switch (errno) {
1048 case EXDEV:
1049 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1050 "not an earlier snapshot from the same fs"));
1051 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
1052
1053 case ENOENT:
1054 if (zfs_dataset_exists(hdl, zc.zc_name,
1055 ZFS_TYPE_SNAPSHOT)) {
1056 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1057 "incremental source (@%s) does not exist"),
1058 zc.zc_value);
1059 }
1060 return (zfs_error(hdl, EZFS_NOENT, errbuf));
1061
1062 case EDQUOT:
1063 case EFBIG:
1064 case EIO:
1065 case ENOLINK:
1066 case ENOSPC:
1067 case ENOSTR:
1068 case ENXIO:
1069 case EPIPE:
1070 case ERANGE:
1071 case EFAULT:
1072 case EROFS:
1073 zfs_error_aux(hdl, strerror(errno));
1074 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1075
1076 default:
1077 return (zfs_standard_error(hdl, errno, errbuf));
1078 }
1079 }
1080
1081 if (debugnv)
1082 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
1083 nvlist_free(thisdbg);
1084
1085 return (0);
1086 }
1087
1088 static void
1089 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
1090 {
1091 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
1092
1093 /*
1094 * zfs_send() only sets snapholds for sends that need them,
1095 * e.g. replication and doall.
1096 */
1097 if (sdd->snapholds == NULL)
1098 return;
1099
1100 fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
1101 }
1102
1103 static void *
1104 send_progress_thread(void *arg)
1105 {
1106 progress_arg_t *pa = arg;
1107 zfs_cmd_t zc = { 0 };
1108 zfs_handle_t *zhp = pa->pa_zhp;
1109 libzfs_handle_t *hdl = zhp->zfs_hdl;
1110 unsigned long long bytes;
1111 char buf[16];
1112 time_t t;
1113 struct tm *tm;
1114
1115 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1116
1117 if (!pa->pa_parsable)
1118 (void) fprintf(stderr, "TIME SENT SNAPSHOT\n");
1119
1120 /*
1121 * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1122 */
1123 for (;;) {
1124 (void) sleep(1);
1125
1126 zc.zc_cookie = pa->pa_fd;
1127 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1128 return ((void *)-1);
1129
1130 (void) time(&t);
1131 tm = localtime(&t);
1132 bytes = zc.zc_cookie;
1133
1134 if (pa->pa_parsable) {
1135 (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1136 tm->tm_hour, tm->tm_min, tm->tm_sec,
1137 bytes, zhp->zfs_name);
1138 } else {
1139 zfs_nicenum(bytes, buf, sizeof (buf));
1140 (void) fprintf(stderr, "%02d:%02d:%02d %5s %s\n",
1141 tm->tm_hour, tm->tm_min, tm->tm_sec,
1142 buf, zhp->zfs_name);
1143 }
1144 }
1145 }
1146
1147 static void
1148 send_print_verbose(FILE *fout, const char *tosnap, const char *fromsnap,
1149 uint64_t size, boolean_t parsable)
1150 {
1151 if (parsable) {
1152 if (fromsnap != NULL) {
1153 (void) fprintf(fout, "incremental\t%s\t%s",
1154 fromsnap, tosnap);
1155 } else {
1156 (void) fprintf(fout, "full\t%s",
1157 tosnap);
1158 }
1159 } else {
1160 if (fromsnap != NULL) {
1161 if (strchr(fromsnap, '@') == NULL &&
1162 strchr(fromsnap, '#') == NULL) {
1163 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1164 "send from @%s to %s"),
1165 fromsnap, tosnap);
1166 } else {
1167 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1168 "send from %s to %s"),
1169 fromsnap, tosnap);
1170 }
1171 } else {
1172 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1173 "full send of %s"),
1174 tosnap);
1175 }
1176 }
1177
1178 if (size != 0) {
1179 if (parsable) {
1180 (void) fprintf(fout, "\t%llu",
1181 (longlong_t)size);
1182 } else {
1183 char buf[16];
1184 zfs_nicenum(size, buf, sizeof (buf));
1185 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1186 " estimated size is %s"), buf);
1187 }
1188 }
1189 (void) fprintf(fout, "\n");
1190 }
1191
1192 static int
1193 dump_snapshot(zfs_handle_t *zhp, void *arg)
1194 {
1195 send_dump_data_t *sdd = arg;
1196 progress_arg_t pa = { 0 };
1197 pthread_t tid;
1198 char *thissnap;
1199 enum lzc_send_flags flags = 0;
1200 int err;
1201 boolean_t isfromsnap, istosnap, fromorigin;
1202 boolean_t exclude = B_FALSE;
1203 FILE *fout = sdd->std_out ? stdout : stderr;
1204
1205 err = 0;
1206 thissnap = strchr(zhp->zfs_name, '@') + 1;
1207 isfromsnap = (sdd->fromsnap != NULL &&
1208 strcmp(sdd->fromsnap, thissnap) == 0);
1209
1210 if (!sdd->seenfrom && isfromsnap) {
1211 gather_holds(zhp, sdd);
1212 sdd->seenfrom = B_TRUE;
1213 (void) strcpy(sdd->prevsnap, thissnap);
1214 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1215 zfs_close(zhp);
1216 return (0);
1217 }
1218
1219 if (sdd->seento || !sdd->seenfrom) {
1220 zfs_close(zhp);
1221 return (0);
1222 }
1223
1224 istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1225 if (istosnap)
1226 sdd->seento = B_TRUE;
1227
1228 if (sdd->large_block)
1229 flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1230 if (sdd->embed_data)
1231 flags |= LZC_SEND_FLAG_EMBED_DATA;
1232 if (sdd->compress)
1233 flags |= LZC_SEND_FLAG_COMPRESS;
1234
1235 if (!sdd->doall && !isfromsnap && !istosnap) {
1236 if (sdd->replicate) {
1237 char *snapname;
1238 nvlist_t *snapprops;
1239 /*
1240 * Filter out all intermediate snapshots except origin
1241 * snapshots needed to replicate clones.
1242 */
1243 nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1244 zhp->zfs_dmustats.dds_guid, &snapname);
1245
1246 VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1247 "snapprops", &snapprops));
1248 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1249 thissnap, &snapprops));
1250 exclude = !nvlist_exists(snapprops, "is_clone_origin");
1251 } else {
1252 exclude = B_TRUE;
1253 }
1254 }
1255
1256 /*
1257 * If a filter function exists, call it to determine whether
1258 * this snapshot will be sent.
1259 */
1260 if (exclude || (sdd->filter_cb != NULL &&
1261 sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1262 /*
1263 * This snapshot is filtered out. Don't send it, and don't
1264 * set prevsnap_obj, so it will be as if this snapshot didn't
1265 * exist, and the next accepted snapshot will be sent as
1266 * an incremental from the last accepted one, or as the
1267 * first (and full) snapshot in the case of a replication,
1268 * non-incremental send.
1269 */
1270 zfs_close(zhp);
1271 return (0);
1272 }
1273
1274 gather_holds(zhp, sdd);
1275 fromorigin = sdd->prevsnap[0] == '\0' &&
1276 (sdd->fromorigin || sdd->replicate);
1277
1278 if (sdd->verbose) {
1279 uint64_t size = 0;
1280 (void) estimate_ioctl(zhp, sdd->prevsnap_obj,
1281 fromorigin, flags, &size);
1282
1283 send_print_verbose(fout, zhp->zfs_name,
1284 sdd->prevsnap[0] ? sdd->prevsnap : NULL,
1285 size, sdd->parsable);
1286 sdd->size += size;
1287 }
1288
1289 if (!sdd->dryrun) {
1290 /*
1291 * If progress reporting is requested, spawn a new thread to
1292 * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1293 */
1294 if (sdd->progress) {
1295 pa.pa_zhp = zhp;
1296 pa.pa_fd = sdd->outfd;
1297 pa.pa_parsable = sdd->parsable;
1298
1299 if ((err = pthread_create(&tid, NULL,
1300 send_progress_thread, &pa)) != 0) {
1301 zfs_close(zhp);
1302 return (err);
1303 }
1304 }
1305
1306 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1307 fromorigin, sdd->outfd, flags, sdd->debugnv);
1308
1309 if (sdd->progress) {
1310 (void) pthread_cancel(tid);
1311 (void) pthread_join(tid, NULL);
1312 }
1313 }
1314
1315 (void) strcpy(sdd->prevsnap, thissnap);
1316 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1317 zfs_close(zhp);
1318 return (err);
1319 }
1320
1321 static int
1322 dump_filesystem(zfs_handle_t *zhp, void *arg)
1323 {
1324 int rv = 0;
1325 send_dump_data_t *sdd = arg;
1326 boolean_t missingfrom = B_FALSE;
1327 zfs_cmd_t zc = { 0 };
1328
1329 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1330 zhp->zfs_name, sdd->tosnap);
1331 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1332 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1333 "WARNING: could not send %s@%s: does not exist\n"),
1334 zhp->zfs_name, sdd->tosnap);
1335 sdd->err = B_TRUE;
1336 return (0);
1337 }
1338
1339 if (sdd->replicate && sdd->fromsnap) {
1340 /*
1341 * If this fs does not have fromsnap, and we're doing
1342 * recursive, we need to send a full stream from the
1343 * beginning (or an incremental from the origin if this
1344 * is a clone). If we're doing non-recursive, then let
1345 * them get the error.
1346 */
1347 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1348 zhp->zfs_name, sdd->fromsnap);
1349 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1350 ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1351 missingfrom = B_TRUE;
1352 }
1353 }
1354
1355 sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1356 sdd->prevsnap_obj = 0;
1357 if (sdd->fromsnap == NULL || missingfrom)
1358 sdd->seenfrom = B_TRUE;
1359
1360 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1361 if (!sdd->seenfrom) {
1362 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1363 "WARNING: could not send %s@%s:\n"
1364 "incremental source (%s@%s) does not exist\n"),
1365 zhp->zfs_name, sdd->tosnap,
1366 zhp->zfs_name, sdd->fromsnap);
1367 sdd->err = B_TRUE;
1368 } else if (!sdd->seento) {
1369 if (sdd->fromsnap) {
1370 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1371 "WARNING: could not send %s@%s:\n"
1372 "incremental source (%s@%s) "
1373 "is not earlier than it\n"),
1374 zhp->zfs_name, sdd->tosnap,
1375 zhp->zfs_name, sdd->fromsnap);
1376 } else {
1377 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1378 "WARNING: "
1379 "could not send %s@%s: does not exist\n"),
1380 zhp->zfs_name, sdd->tosnap);
1381 }
1382 sdd->err = B_TRUE;
1383 }
1384
1385 return (rv);
1386 }
1387
1388 static int
1389 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1390 {
1391 send_dump_data_t *sdd = arg;
1392 nvpair_t *fspair;
1393 boolean_t needagain, progress;
1394
1395 if (!sdd->replicate)
1396 return (dump_filesystem(rzhp, sdd));
1397
1398 /* Mark the clone origin snapshots. */
1399 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1400 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1401 nvlist_t *nvfs;
1402 uint64_t origin_guid = 0;
1403
1404 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1405 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1406 if (origin_guid != 0) {
1407 char *snapname;
1408 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1409 origin_guid, &snapname);
1410 if (origin_nv != NULL) {
1411 nvlist_t *snapprops;
1412 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1413 "snapprops", &snapprops));
1414 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1415 snapname, &snapprops));
1416 VERIFY(0 == nvlist_add_boolean(
1417 snapprops, "is_clone_origin"));
1418 }
1419 }
1420 }
1421 again:
1422 needagain = progress = B_FALSE;
1423 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1424 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1425 nvlist_t *fslist, *parent_nv;
1426 char *fsname;
1427 zfs_handle_t *zhp;
1428 int err;
1429 uint64_t origin_guid = 0;
1430 uint64_t parent_guid = 0;
1431
1432 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1433 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1434 continue;
1435
1436 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1437 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1438 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1439 &parent_guid);
1440
1441 if (parent_guid != 0) {
1442 parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1443 if (!nvlist_exists(parent_nv, "sent")) {
1444 /* parent has not been sent; skip this one */
1445 needagain = B_TRUE;
1446 continue;
1447 }
1448 }
1449
1450 if (origin_guid != 0) {
1451 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1452 origin_guid, NULL);
1453 if (origin_nv != NULL &&
1454 !nvlist_exists(origin_nv, "sent")) {
1455 /*
1456 * origin has not been sent yet;
1457 * skip this clone.
1458 */
1459 needagain = B_TRUE;
1460 continue;
1461 }
1462 }
1463
1464 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1465 if (zhp == NULL)
1466 return (-1);
1467 err = dump_filesystem(zhp, sdd);
1468 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1469 progress = B_TRUE;
1470 zfs_close(zhp);
1471 if (err)
1472 return (err);
1473 }
1474 if (needagain) {
1475 assert(progress);
1476 goto again;
1477 }
1478
1479 /* clean out the sent flags in case we reuse this fss */
1480 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1481 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1482 nvlist_t *fslist;
1483
1484 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1485 (void) nvlist_remove_all(fslist, "sent");
1486 }
1487
1488 return (0);
1489 }
1490
1491 nvlist_t *
1492 zfs_send_resume_token_to_nvlist(libzfs_handle_t *hdl, const char *token)
1493 {
1494 unsigned int version;
1495 int nread;
1496 unsigned long long checksum, packed_len;
1497
1498 /*
1499 * Decode token header, which is:
1500 * <token version>-<checksum of payload>-<uncompressed payload length>
1501 * Note that the only supported token version is 1.
1502 */
1503 nread = sscanf(token, "%u-%llx-%llx-",
1504 &version, &checksum, &packed_len);
1505 if (nread != 3) {
1506 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1507 "resume token is corrupt (invalid format)"));
1508 return (NULL);
1509 }
1510
1511 if (version != ZFS_SEND_RESUME_TOKEN_VERSION) {
1512 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1513 "resume token is corrupt (invalid version %u)"),
1514 version);
1515 return (NULL);
1516 }
1517
1518 /* convert hexadecimal representation to binary */
1519 token = strrchr(token, '-') + 1;
1520 int len = strlen(token) / 2;
1521 unsigned char *compressed = zfs_alloc(hdl, len);
1522 for (int i = 0; i < len; i++) {
1523 nread = sscanf(token + i * 2, "%2hhx", compressed + i);
1524 if (nread != 1) {
1525 free(compressed);
1526 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1527 "resume token is corrupt "
1528 "(payload is not hex-encoded)"));
1529 return (NULL);
1530 }
1531 }
1532
1533 /* verify checksum */
1534 zio_cksum_t cksum;
1535 fletcher_4_native(compressed, len, NULL, &cksum);
1536 if (cksum.zc_word[0] != checksum) {
1537 free(compressed);
1538 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1539 "resume token is corrupt (incorrect checksum)"));
1540 return (NULL);
1541 }
1542
1543 /* uncompress */
1544 void *packed = zfs_alloc(hdl, packed_len);
1545 uLongf packed_len_long = packed_len;
1546 if (uncompress(packed, &packed_len_long, compressed, len) != Z_OK ||
1547 packed_len_long != packed_len) {
1548 free(packed);
1549 free(compressed);
1550 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1551 "resume token is corrupt (decompression failed)"));
1552 return (NULL);
1553 }
1554
1555 /* unpack nvlist */
1556 nvlist_t *nv;
1557 int error = nvlist_unpack(packed, packed_len, &nv, KM_SLEEP);
1558 free(packed);
1559 free(compressed);
1560 if (error != 0) {
1561 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1562 "resume token is corrupt (nvlist_unpack failed)"));
1563 return (NULL);
1564 }
1565 return (nv);
1566 }
1567
1568 int
1569 zfs_send_resume(libzfs_handle_t *hdl, sendflags_t *flags, int outfd,
1570 const char *resume_token)
1571 {
1572 char errbuf[1024];
1573 char *toname;
1574 char *fromname = NULL;
1575 uint64_t resumeobj, resumeoff, toguid, fromguid, bytes;
1576 zfs_handle_t *zhp;
1577 int error = 0;
1578 char name[ZFS_MAX_DATASET_NAME_LEN];
1579 enum lzc_send_flags lzc_flags = 0;
1580 FILE *fout = (flags->verbose && flags->dryrun) ? stdout : stderr;
1581
1582 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1583 "cannot resume send"));
1584
1585 nvlist_t *resume_nvl =
1586 zfs_send_resume_token_to_nvlist(hdl, resume_token);
1587 if (resume_nvl == NULL) {
1588 /*
1589 * zfs_error_aux has already been set by
1590 * zfs_send_resume_token_to_nvlist
1591 */
1592 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1593 }
1594 if (flags->verbose) {
1595 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1596 "resume token contents:\n"));
1597 nvlist_print(fout, resume_nvl);
1598 }
1599
1600 if (nvlist_lookup_string(resume_nvl, "toname", &toname) != 0 ||
1601 nvlist_lookup_uint64(resume_nvl, "object", &resumeobj) != 0 ||
1602 nvlist_lookup_uint64(resume_nvl, "offset", &resumeoff) != 0 ||
1603 nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 ||
1604 nvlist_lookup_uint64(resume_nvl, "toguid", &toguid) != 0) {
1605 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1606 "resume token is corrupt"));
1607 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1608 }
1609 fromguid = 0;
1610 (void) nvlist_lookup_uint64(resume_nvl, "fromguid", &fromguid);
1611
1612 if (flags->largeblock || nvlist_exists(resume_nvl, "largeblockok"))
1613 lzc_flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1614 if (flags->embed_data || nvlist_exists(resume_nvl, "embedok"))
1615 lzc_flags |= LZC_SEND_FLAG_EMBED_DATA;
1616 if (flags->compress || nvlist_exists(resume_nvl, "compressok"))
1617 lzc_flags |= LZC_SEND_FLAG_COMPRESS;
1618
1619 if (guid_to_name(hdl, toname, toguid, B_FALSE, name) != 0) {
1620 if (zfs_dataset_exists(hdl, toname, ZFS_TYPE_DATASET)) {
1621 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1622 "'%s' is no longer the same snapshot used in "
1623 "the initial send"), toname);
1624 } else {
1625 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1626 "'%s' used in the initial send no longer exists"),
1627 toname);
1628 }
1629 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1630 }
1631 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1632 if (zhp == NULL) {
1633 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1634 "unable to access '%s'"), name);
1635 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1636 }
1637
1638 if (fromguid != 0) {
1639 if (guid_to_name(hdl, toname, fromguid, B_TRUE, name) != 0) {
1640 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1641 "incremental source %#llx no longer exists"),
1642 (longlong_t)fromguid);
1643 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1644 }
1645 fromname = name;
1646 }
1647
1648 if (flags->verbose) {
1649 uint64_t size = 0;
1650 error = lzc_send_space(zhp->zfs_name, fromname,
1651 lzc_flags, &size);
1652 if (error == 0)
1653 size = MAX(0, (int64_t)(size - bytes));
1654 send_print_verbose(fout, zhp->zfs_name, fromname,
1655 size, flags->parsable);
1656 }
1657
1658 if (!flags->dryrun) {
1659 progress_arg_t pa = { 0 };
1660 pthread_t tid;
1661 /*
1662 * If progress reporting is requested, spawn a new thread to
1663 * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1664 */
1665 if (flags->progress) {
1666 pa.pa_zhp = zhp;
1667 pa.pa_fd = outfd;
1668 pa.pa_parsable = flags->parsable;
1669
1670 error = pthread_create(&tid, NULL,
1671 send_progress_thread, &pa);
1672 if (error != 0) {
1673 zfs_close(zhp);
1674 return (error);
1675 }
1676 }
1677
1678 error = lzc_send_resume(zhp->zfs_name, fromname, outfd,
1679 lzc_flags, resumeobj, resumeoff);
1680
1681 if (flags->progress) {
1682 (void) pthread_cancel(tid);
1683 (void) pthread_join(tid, NULL);
1684 }
1685
1686 char errbuf[1024];
1687 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1688 "warning: cannot send '%s'"), zhp->zfs_name);
1689
1690 zfs_close(zhp);
1691
1692 switch (error) {
1693 case 0:
1694 return (0);
1695 case EXDEV:
1696 case ENOENT:
1697 case EDQUOT:
1698 case EFBIG:
1699 case EIO:
1700 case ENOLINK:
1701 case ENOSPC:
1702 case ENOSTR:
1703 case ENXIO:
1704 case EPIPE:
1705 case ERANGE:
1706 case EFAULT:
1707 case EROFS:
1708 zfs_error_aux(hdl, strerror(errno));
1709 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1710
1711 default:
1712 return (zfs_standard_error(hdl, errno, errbuf));
1713 }
1714 }
1715
1716
1717 zfs_close(zhp);
1718
1719 return (error);
1720 }
1721
1722 /*
1723 * Generate a send stream for the dataset identified by the argument zhp.
1724 *
1725 * The content of the send stream is the snapshot identified by
1726 * 'tosnap'. Incremental streams are requested in two ways:
1727 * - from the snapshot identified by "fromsnap" (if non-null) or
1728 * - from the origin of the dataset identified by zhp, which must
1729 * be a clone. In this case, "fromsnap" is null and "fromorigin"
1730 * is TRUE.
1731 *
1732 * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1733 * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1734 * if "replicate" is set. If "doall" is set, dump all the intermediate
1735 * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1736 * case too. If "props" is set, send properties.
1737 */
1738 int
1739 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1740 sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1741 void *cb_arg, nvlist_t **debugnvp)
1742 {
1743 char errbuf[1024];
1744 send_dump_data_t sdd = { 0 };
1745 int err = 0;
1746 nvlist_t *fss = NULL;
1747 avl_tree_t *fsavl = NULL;
1748 static uint64_t holdseq;
1749 int spa_version;
1750 pthread_t tid = 0;
1751 int pipefd[2];
1752 dedup_arg_t dda = { 0 };
1753 int featureflags = 0;
1754 FILE *fout;
1755
1756 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1757 "cannot send '%s'"), zhp->zfs_name);
1758
1759 if (fromsnap && fromsnap[0] == '\0') {
1760 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1761 "zero-length incremental source"));
1762 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1763 }
1764
1765 if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1766 uint64_t version;
1767 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1768 if (version >= ZPL_VERSION_SA) {
1769 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1770 }
1771 }
1772
1773 if (flags->dedup && !flags->dryrun) {
1774 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1775 DMU_BACKUP_FEATURE_DEDUPPROPS);
1776 if ((err = pipe(pipefd)) != 0) {
1777 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1778 return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1779 errbuf));
1780 }
1781 dda.outputfd = outfd;
1782 dda.inputfd = pipefd[1];
1783 dda.dedup_hdl = zhp->zfs_hdl;
1784 if ((err = pthread_create(&tid, NULL, cksummer, &dda)) != 0) {
1785 (void) close(pipefd[0]);
1786 (void) close(pipefd[1]);
1787 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1788 return (zfs_error(zhp->zfs_hdl,
1789 EZFS_THREADCREATEFAILED, errbuf));
1790 }
1791 }
1792
1793 if (flags->replicate || flags->doall || flags->props) {
1794 dmu_replay_record_t drr = { 0 };
1795 char *packbuf = NULL;
1796 size_t buflen = 0;
1797 zio_cksum_t zc = { 0 };
1798
1799 if (flags->replicate || flags->props) {
1800 nvlist_t *hdrnv;
1801
1802 VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1803 if (fromsnap) {
1804 VERIFY(0 == nvlist_add_string(hdrnv,
1805 "fromsnap", fromsnap));
1806 }
1807 VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1808 if (!flags->replicate) {
1809 VERIFY(0 == nvlist_add_boolean(hdrnv,
1810 "not_recursive"));
1811 }
1812
1813 err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1814 fromsnap, tosnap, flags->replicate, flags->verbose,
1815 &fss, &fsavl);
1816 if (err)
1817 goto err_out;
1818 VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1819 err = nvlist_pack(hdrnv, &packbuf, &buflen,
1820 NV_ENCODE_XDR, 0);
1821 if (debugnvp)
1822 *debugnvp = hdrnv;
1823 else
1824 nvlist_free(hdrnv);
1825 if (err)
1826 goto stderr_out;
1827 }
1828
1829 if (!flags->dryrun) {
1830 /* write first begin record */
1831 drr.drr_type = DRR_BEGIN;
1832 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1833 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1834 drr_versioninfo, DMU_COMPOUNDSTREAM);
1835 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1836 drr_versioninfo, featureflags);
1837 (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1838 sizeof (drr.drr_u.drr_begin.drr_toname),
1839 "%s@%s", zhp->zfs_name, tosnap);
1840 drr.drr_payloadlen = buflen;
1841
1842 err = dump_record(&drr, packbuf, buflen, &zc, outfd);
1843 free(packbuf);
1844 if (err != 0)
1845 goto stderr_out;
1846
1847 /* write end record */
1848 bzero(&drr, sizeof (drr));
1849 drr.drr_type = DRR_END;
1850 drr.drr_u.drr_end.drr_checksum = zc;
1851 err = write(outfd, &drr, sizeof (drr));
1852 if (err == -1) {
1853 err = errno;
1854 goto stderr_out;
1855 }
1856
1857 err = 0;
1858 }
1859 }
1860
1861 /* dump each stream */
1862 sdd.fromsnap = fromsnap;
1863 sdd.tosnap = tosnap;
1864 if (tid != 0)
1865 sdd.outfd = pipefd[0];
1866 else
1867 sdd.outfd = outfd;
1868 sdd.replicate = flags->replicate;
1869 sdd.doall = flags->doall;
1870 sdd.fromorigin = flags->fromorigin;
1871 sdd.fss = fss;
1872 sdd.fsavl = fsavl;
1873 sdd.verbose = flags->verbose;
1874 sdd.parsable = flags->parsable;
1875 sdd.progress = flags->progress;
1876 sdd.dryrun = flags->dryrun;
1877 sdd.large_block = flags->largeblock;
1878 sdd.embed_data = flags->embed_data;
1879 sdd.compress = flags->compress;
1880 sdd.filter_cb = filter_func;
1881 sdd.filter_cb_arg = cb_arg;
1882 if (debugnvp)
1883 sdd.debugnv = *debugnvp;
1884 if (sdd.verbose && sdd.dryrun)
1885 sdd.std_out = B_TRUE;
1886 fout = sdd.std_out ? stdout : stderr;
1887
1888 /*
1889 * Some flags require that we place user holds on the datasets that are
1890 * being sent so they don't get destroyed during the send. We can skip
1891 * this step if the pool is imported read-only since the datasets cannot
1892 * be destroyed.
1893 */
1894 if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1895 ZPOOL_PROP_READONLY, NULL) &&
1896 zfs_spa_version(zhp, &spa_version) == 0 &&
1897 spa_version >= SPA_VERSION_USERREFS &&
1898 (flags->doall || flags->replicate)) {
1899 ++holdseq;
1900 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1901 ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1902 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1903 if (sdd.cleanup_fd < 0) {
1904 err = errno;
1905 goto stderr_out;
1906 }
1907 sdd.snapholds = fnvlist_alloc();
1908 } else {
1909 sdd.cleanup_fd = -1;
1910 sdd.snapholds = NULL;
1911 }
1912 if (flags->verbose || sdd.snapholds != NULL) {
1913 /*
1914 * Do a verbose no-op dry run to get all the verbose output
1915 * or to gather snapshot hold's before generating any data,
1916 * then do a non-verbose real run to generate the streams.
1917 */
1918 sdd.dryrun = B_TRUE;
1919 err = dump_filesystems(zhp, &sdd);
1920
1921 if (err != 0)
1922 goto stderr_out;
1923
1924 if (flags->verbose) {
1925 if (flags->parsable) {
1926 (void) fprintf(fout, "size\t%llu\n",
1927 (longlong_t)sdd.size);
1928 } else {
1929 char buf[16];
1930 zfs_nicenum(sdd.size, buf, sizeof (buf));
1931 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1932 "total estimated size is %s\n"), buf);
1933 }
1934 }
1935
1936 /* Ensure no snaps found is treated as an error. */
1937 if (!sdd.seento) {
1938 err = ENOENT;
1939 goto err_out;
1940 }
1941
1942 /* Skip the second run if dryrun was requested. */
1943 if (flags->dryrun)
1944 goto err_out;
1945
1946 if (sdd.snapholds != NULL) {
1947 err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
1948 if (err != 0)
1949 goto stderr_out;
1950
1951 fnvlist_free(sdd.snapholds);
1952 sdd.snapholds = NULL;
1953 }
1954
1955 sdd.dryrun = B_FALSE;
1956 sdd.verbose = B_FALSE;
1957 }
1958
1959 err = dump_filesystems(zhp, &sdd);
1960 fsavl_destroy(fsavl);
1961 nvlist_free(fss);
1962
1963 /* Ensure no snaps found is treated as an error. */
1964 if (err == 0 && !sdd.seento)
1965 err = ENOENT;
1966
1967 if (tid != 0) {
1968 if (err != 0)
1969 (void) pthread_cancel(tid);
1970 (void) close(pipefd[0]);
1971 (void) pthread_join(tid, NULL);
1972 }
1973
1974 if (sdd.cleanup_fd != -1) {
1975 VERIFY(0 == close(sdd.cleanup_fd));
1976 sdd.cleanup_fd = -1;
1977 }
1978
1979 if (!flags->dryrun && (flags->replicate || flags->doall ||
1980 flags->props)) {
1981 /*
1982 * write final end record. NB: want to do this even if
1983 * there was some error, because it might not be totally
1984 * failed.
1985 */
1986 dmu_replay_record_t drr = { 0 };
1987 drr.drr_type = DRR_END;
1988 if (write(outfd, &drr, sizeof (drr)) == -1) {
1989 return (zfs_standard_error(zhp->zfs_hdl,
1990 errno, errbuf));
1991 }
1992 }
1993
1994 return (err || sdd.err);
1995
1996 stderr_out:
1997 err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1998 err_out:
1999 fsavl_destroy(fsavl);
2000 nvlist_free(fss);
2001 fnvlist_free(sdd.snapholds);
2002
2003 if (sdd.cleanup_fd != -1)
2004 VERIFY(0 == close(sdd.cleanup_fd));
2005 if (tid != 0) {
2006 (void) pthread_cancel(tid);
2007 (void) close(pipefd[0]);
2008 (void) pthread_join(tid, NULL);
2009 }
2010 return (err);
2011 }
2012
2013 int
2014 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd,
2015 enum lzc_send_flags flags)
2016 {
2017 int err;
2018 libzfs_handle_t *hdl = zhp->zfs_hdl;
2019
2020 char errbuf[1024];
2021 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2022 "warning: cannot send '%s'"), zhp->zfs_name);
2023
2024 err = lzc_send(zhp->zfs_name, from, fd, flags);
2025 if (err != 0) {
2026 switch (errno) {
2027 case EXDEV:
2028 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2029 "not an earlier snapshot from the same fs"));
2030 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
2031
2032 case ENOENT:
2033 case ESRCH:
2034 if (lzc_exists(zhp->zfs_name)) {
2035 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2036 "incremental source (%s) does not exist"),
2037 from);
2038 }
2039 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2040
2041 case EBUSY:
2042 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2043 "target is busy; if a filesystem, "
2044 "it must not be mounted"));
2045 return (zfs_error(hdl, EZFS_BUSY, errbuf));
2046
2047 case EDQUOT:
2048 case EFBIG:
2049 case EIO:
2050 case ENOLINK:
2051 case ENOSPC:
2052 case ENOSTR:
2053 case ENXIO:
2054 case EPIPE:
2055 case ERANGE:
2056 case EFAULT:
2057 case EROFS:
2058 zfs_error_aux(hdl, strerror(errno));
2059 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
2060
2061 default:
2062 return (zfs_standard_error(hdl, errno, errbuf));
2063 }
2064 }
2065 return (err != 0);
2066 }
2067
2068 /*
2069 * Routines specific to "zfs recv"
2070 */
2071
2072 static int
2073 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
2074 boolean_t byteswap, zio_cksum_t *zc)
2075 {
2076 char *cp = buf;
2077 int rv;
2078 int len = ilen;
2079
2080 assert(ilen <= SPA_MAXBLOCKSIZE);
2081
2082 do {
2083 rv = read(fd, cp, len);
2084 cp += rv;
2085 len -= rv;
2086 } while (rv > 0);
2087
2088 if (rv < 0 || len != 0) {
2089 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2090 "failed to read from stream"));
2091 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
2092 "cannot receive")));
2093 }
2094
2095 if (zc) {
2096 if (byteswap)
2097 (void) fletcher_4_incremental_byteswap(buf, ilen, zc);
2098 else
2099 (void) fletcher_4_incremental_native(buf, ilen, zc);
2100 }
2101 return (0);
2102 }
2103
2104 static int
2105 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
2106 boolean_t byteswap, zio_cksum_t *zc)
2107 {
2108 char *buf;
2109 int err;
2110
2111 buf = zfs_alloc(hdl, len);
2112 if (buf == NULL)
2113 return (ENOMEM);
2114
2115 err = recv_read(hdl, fd, buf, len, byteswap, zc);
2116 if (err != 0) {
2117 free(buf);
2118 return (err);
2119 }
2120
2121 err = nvlist_unpack(buf, len, nvp, 0);
2122 free(buf);
2123 if (err != 0) {
2124 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2125 "stream (malformed nvlist)"));
2126 return (EINVAL);
2127 }
2128 return (0);
2129 }
2130
2131 static int
2132 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
2133 int baselen, char *newname, recvflags_t *flags)
2134 {
2135 static int seq;
2136 zfs_cmd_t zc = { 0 };
2137 int err;
2138 prop_changelist_t *clp;
2139 zfs_handle_t *zhp;
2140
2141 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2142 if (zhp == NULL)
2143 return (-1);
2144 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2145 flags->force ? MS_FORCE : 0);
2146 zfs_close(zhp);
2147 if (clp == NULL)
2148 return (-1);
2149 err = changelist_prefix(clp);
2150 if (err)
2151 return (err);
2152
2153 zc.zc_objset_type = DMU_OST_ZFS;
2154 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
2155
2156 if (tryname) {
2157 (void) strcpy(newname, tryname);
2158
2159 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
2160
2161 if (flags->verbose) {
2162 (void) printf("attempting rename %s to %s\n",
2163 zc.zc_name, zc.zc_value);
2164 }
2165 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
2166 if (err == 0)
2167 changelist_rename(clp, name, tryname);
2168 } else {
2169 err = ENOENT;
2170 }
2171
2172 if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
2173 seq++;
2174
2175 (void) snprintf(newname, ZFS_MAX_DATASET_NAME_LEN,
2176 "%.*srecv-%u-%u", baselen, name, getpid(), seq);
2177 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
2178
2179 if (flags->verbose) {
2180 (void) printf("failed - trying rename %s to %s\n",
2181 zc.zc_name, zc.zc_value);
2182 }
2183 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
2184 if (err == 0)
2185 changelist_rename(clp, name, newname);
2186 if (err && flags->verbose) {
2187 (void) printf("failed (%u) - "
2188 "will try again on next pass\n", errno);
2189 }
2190 err = EAGAIN;
2191 } else if (flags->verbose) {
2192 if (err == 0)
2193 (void) printf("success\n");
2194 else
2195 (void) printf("failed (%u)\n", errno);
2196 }
2197
2198 (void) changelist_postfix(clp);
2199 changelist_free(clp);
2200
2201 return (err);
2202 }
2203
2204 static int
2205 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
2206 char *newname, recvflags_t *flags)
2207 {
2208 zfs_cmd_t zc = { 0 };
2209 int err = 0;
2210 prop_changelist_t *clp;
2211 zfs_handle_t *zhp;
2212 boolean_t defer = B_FALSE;
2213 int spa_version;
2214
2215 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2216 if (zhp == NULL)
2217 return (-1);
2218 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2219 flags->force ? MS_FORCE : 0);
2220 if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
2221 zfs_spa_version(zhp, &spa_version) == 0 &&
2222 spa_version >= SPA_VERSION_USERREFS)
2223 defer = B_TRUE;
2224 zfs_close(zhp);
2225 if (clp == NULL)
2226 return (-1);
2227 err = changelist_prefix(clp);
2228 if (err)
2229 return (err);
2230
2231 zc.zc_objset_type = DMU_OST_ZFS;
2232 zc.zc_defer_destroy = defer;
2233 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
2234
2235 if (flags->verbose)
2236 (void) printf("attempting destroy %s\n", zc.zc_name);
2237 err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
2238 if (err == 0) {
2239 if (flags->verbose)
2240 (void) printf("success\n");
2241 changelist_remove(clp, zc.zc_name);
2242 }
2243
2244 (void) changelist_postfix(clp);
2245 changelist_free(clp);
2246
2247 /*
2248 * Deferred destroy might destroy the snapshot or only mark it to be
2249 * destroyed later, and it returns success in either case.
2250 */
2251 if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
2252 ZFS_TYPE_SNAPSHOT))) {
2253 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
2254 }
2255
2256 return (err);
2257 }
2258
2259 typedef struct guid_to_name_data {
2260 uint64_t guid;
2261 boolean_t bookmark_ok;
2262 char *name;
2263 char *skip;
2264 } guid_to_name_data_t;
2265
2266 static int
2267 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
2268 {
2269 guid_to_name_data_t *gtnd = arg;
2270 const char *slash;
2271 int err;
2272
2273 if (gtnd->skip != NULL &&
2274 (slash = strrchr(zhp->zfs_name, '/')) != NULL &&
2275 strcmp(slash + 1, gtnd->skip) == 0) {
2276 zfs_close(zhp);
2277 return (0);
2278 }
2279
2280 if (zfs_prop_get_int(zhp, ZFS_PROP_GUID) == gtnd->guid) {
2281 (void) strcpy(gtnd->name, zhp->zfs_name);
2282 zfs_close(zhp);
2283 return (EEXIST);
2284 }
2285
2286 err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
2287 if (err != EEXIST && gtnd->bookmark_ok)
2288 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, gtnd);
2289 zfs_close(zhp);
2290 return (err);
2291 }
2292
2293 /*
2294 * Attempt to find the local dataset associated with this guid. In the case of
2295 * multiple matches, we attempt to find the "best" match by searching
2296 * progressively larger portions of the hierarchy. This allows one to send a
2297 * tree of datasets individually and guarantee that we will find the source
2298 * guid within that hierarchy, even if there are multiple matches elsewhere.
2299 */
2300 static int
2301 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
2302 boolean_t bookmark_ok, char *name)
2303 {
2304 char pname[ZFS_MAX_DATASET_NAME_LEN];
2305 guid_to_name_data_t gtnd;
2306
2307 gtnd.guid = guid;
2308 gtnd.bookmark_ok = bookmark_ok;
2309 gtnd.name = name;
2310 gtnd.skip = NULL;
2311
2312 /*
2313 * Search progressively larger portions of the hierarchy, starting
2314 * with the filesystem specified by 'parent'. This will
2315 * select the "most local" version of the origin snapshot in the case
2316 * that there are multiple matching snapshots in the system.
2317 */
2318 (void) strlcpy(pname, parent, sizeof (pname));
2319 char *cp = strrchr(pname, '@');
2320 if (cp == NULL)
2321 cp = strchr(pname, '\0');
2322 for (; cp != NULL; cp = strrchr(pname, '/')) {
2323 /* Chop off the last component and open the parent */
2324 *cp = '\0';
2325 zfs_handle_t *zhp = make_dataset_handle(hdl, pname);
2326
2327 if (zhp == NULL)
2328 continue;
2329 int err = guid_to_name_cb(zfs_handle_dup(zhp), >nd);
2330 if (err != EEXIST)
2331 err = zfs_iter_children(zhp, guid_to_name_cb, >nd);
2332 if (err != EEXIST && bookmark_ok)
2333 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, >nd);
2334 zfs_close(zhp);
2335 if (err == EEXIST)
2336 return (0);
2337
2338 /*
2339 * Remember the last portion of the dataset so we skip it next
2340 * time through (as we've already searched that portion of the
2341 * hierarchy).
2342 */
2343 gtnd.skip = strrchr(pname, '/') + 1;
2344 }
2345
2346 return (ENOENT);
2347 }
2348
2349 /*
2350 * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
2351 * guid1 is after guid2.
2352 */
2353 static int
2354 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
2355 uint64_t guid1, uint64_t guid2)
2356 {
2357 nvlist_t *nvfs;
2358 char *fsname, *snapname;
2359 char buf[ZFS_MAX_DATASET_NAME_LEN];
2360 int rv;
2361 zfs_handle_t *guid1hdl, *guid2hdl;
2362 uint64_t create1, create2;
2363
2364 if (guid2 == 0)
2365 return (0);
2366 if (guid1 == 0)
2367 return (1);
2368
2369 nvfs = fsavl_find(avl, guid1, &snapname);
2370 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2371 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2372 guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2373 if (guid1hdl == NULL)
2374 return (-1);
2375
2376 nvfs = fsavl_find(avl, guid2, &snapname);
2377 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2378 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2379 guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2380 if (guid2hdl == NULL) {
2381 zfs_close(guid1hdl);
2382 return (-1);
2383 }
2384
2385 create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
2386 create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
2387
2388 if (create1 < create2)
2389 rv = -1;
2390 else if (create1 > create2)
2391 rv = +1;
2392 else
2393 rv = 0;
2394
2395 zfs_close(guid1hdl);
2396 zfs_close(guid2hdl);
2397
2398 return (rv);
2399 }
2400
2401 static int
2402 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2403 recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2404 nvlist_t *renamed)
2405 {
2406 nvlist_t *local_nv;
2407 avl_tree_t *local_avl;
2408 nvpair_t *fselem, *nextfselem;
2409 char *fromsnap;
2410 char newname[ZFS_MAX_DATASET_NAME_LEN];
2411 int error;
2412 boolean_t needagain, progress, recursive;
2413 char *s1, *s2;
2414
2415 VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2416
2417 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2418 ENOENT);
2419
2420 if (flags->dryrun)
2421 return (0);
2422
2423 again:
2424 needagain = progress = B_FALSE;
2425
2426 if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2427 recursive, B_FALSE, &local_nv, &local_avl)) != 0)
2428 return (error);
2429
2430 /*
2431 * Process deletes and renames
2432 */
2433 for (fselem = nvlist_next_nvpair(local_nv, NULL);
2434 fselem; fselem = nextfselem) {
2435 nvlist_t *nvfs, *snaps;
2436 nvlist_t *stream_nvfs = NULL;
2437 nvpair_t *snapelem, *nextsnapelem;
2438 uint64_t fromguid = 0;
2439 uint64_t originguid = 0;
2440 uint64_t stream_originguid = 0;
2441 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2442 char *fsname, *stream_fsname;
2443
2444 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2445
2446 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2447 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2448 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2449 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2450 &parent_fromsnap_guid));
2451 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2452
2453 /*
2454 * First find the stream's fs, so we can check for
2455 * a different origin (due to "zfs promote")
2456 */
2457 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2458 snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2459 uint64_t thisguid;
2460
2461 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2462 stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2463
2464 if (stream_nvfs != NULL)
2465 break;
2466 }
2467
2468 /* check for promote */
2469 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2470 &stream_originguid);
2471 if (stream_nvfs && originguid != stream_originguid) {
2472 switch (created_before(hdl, local_avl,
2473 stream_originguid, originguid)) {
2474 case 1: {
2475 /* promote it! */
2476 zfs_cmd_t zc = { 0 };
2477 nvlist_t *origin_nvfs;
2478 char *origin_fsname;
2479
2480 if (flags->verbose)
2481 (void) printf("promoting %s\n", fsname);
2482
2483 origin_nvfs = fsavl_find(local_avl, originguid,
2484 NULL);
2485 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2486 "name", &origin_fsname));
2487 (void) strlcpy(zc.zc_value, origin_fsname,
2488 sizeof (zc.zc_value));
2489 (void) strlcpy(zc.zc_name, fsname,
2490 sizeof (zc.zc_name));
2491 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2492 if (error == 0)
2493 progress = B_TRUE;
2494 break;
2495 }
2496 default:
2497 break;
2498 case -1:
2499 fsavl_destroy(local_avl);
2500 nvlist_free(local_nv);
2501 return (-1);
2502 }
2503 /*
2504 * We had/have the wrong origin, therefore our
2505 * list of snapshots is wrong. Need to handle
2506 * them on the next pass.
2507 */
2508 needagain = B_TRUE;
2509 continue;
2510 }
2511
2512 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2513 snapelem; snapelem = nextsnapelem) {
2514 uint64_t thisguid;
2515 char *stream_snapname;
2516 nvlist_t *found, *props;
2517
2518 nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2519
2520 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2521 found = fsavl_find(stream_avl, thisguid,
2522 &stream_snapname);
2523
2524 /* check for delete */
2525 if (found == NULL) {
2526 char name[ZFS_MAX_DATASET_NAME_LEN];
2527
2528 if (!flags->force)
2529 continue;
2530
2531 (void) snprintf(name, sizeof (name), "%s@%s",
2532 fsname, nvpair_name(snapelem));
2533
2534 error = recv_destroy(hdl, name,
2535 strlen(fsname)+1, newname, flags);
2536 if (error)
2537 needagain = B_TRUE;
2538 else
2539 progress = B_TRUE;
2540 continue;
2541 }
2542
2543 stream_nvfs = found;
2544
2545 if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2546 &props) && 0 == nvlist_lookup_nvlist(props,
2547 stream_snapname, &props)) {
2548 zfs_cmd_t zc = { 0 };
2549
2550 zc.zc_cookie = B_TRUE; /* received */
2551 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2552 "%s@%s", fsname, nvpair_name(snapelem));
2553 if (zcmd_write_src_nvlist(hdl, &zc,
2554 props) == 0) {
2555 (void) zfs_ioctl(hdl,
2556 ZFS_IOC_SET_PROP, &zc);
2557 zcmd_free_nvlists(&zc);
2558 }
2559 }
2560
2561 /* check for different snapname */
2562 if (strcmp(nvpair_name(snapelem),
2563 stream_snapname) != 0) {
2564 char name[ZFS_MAX_DATASET_NAME_LEN];
2565 char tryname[ZFS_MAX_DATASET_NAME_LEN];
2566
2567 (void) snprintf(name, sizeof (name), "%s@%s",
2568 fsname, nvpair_name(snapelem));
2569 (void) snprintf(tryname, sizeof (name), "%s@%s",
2570 fsname, stream_snapname);
2571
2572 error = recv_rename(hdl, name, tryname,
2573 strlen(fsname)+1, newname, flags);
2574 if (error)
2575 needagain = B_TRUE;
2576 else
2577 progress = B_TRUE;
2578 }
2579
2580 if (strcmp(stream_snapname, fromsnap) == 0)
2581 fromguid = thisguid;
2582 }
2583
2584 /* check for delete */
2585 if (stream_nvfs == NULL) {
2586 if (!flags->force)
2587 continue;
2588
2589 error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2590 newname, flags);
2591 if (error)
2592 needagain = B_TRUE;
2593 else
2594 progress = B_TRUE;
2595 continue;
2596 }
2597
2598 if (fromguid == 0) {
2599 if (flags->verbose) {
2600 (void) printf("local fs %s does not have "
2601 "fromsnap (%s in stream); must have "
2602 "been deleted locally; ignoring\n",
2603 fsname, fromsnap);
2604 }
2605 continue;
2606 }
2607
2608 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2609 "name", &stream_fsname));
2610 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2611 "parentfromsnap", &stream_parent_fromsnap_guid));
2612
2613 s1 = strrchr(fsname, '/');
2614 s2 = strrchr(stream_fsname, '/');
2615
2616 /*
2617 * Check for rename. If the exact receive path is specified, it
2618 * does not count as a rename, but we still need to check the
2619 * datasets beneath it.
2620 */
2621 if ((stream_parent_fromsnap_guid != 0 &&
2622 parent_fromsnap_guid != 0 &&
2623 stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2624 ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2625 (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2626 nvlist_t *parent;
2627 char tryname[ZFS_MAX_DATASET_NAME_LEN];
2628
2629 parent = fsavl_find(local_avl,
2630 stream_parent_fromsnap_guid, NULL);
2631 /*
2632 * NB: parent might not be found if we used the
2633 * tosnap for stream_parent_fromsnap_guid,
2634 * because the parent is a newly-created fs;
2635 * we'll be able to rename it after we recv the
2636 * new fs.
2637 */
2638 if (parent != NULL) {
2639 char *pname;
2640
2641 VERIFY(0 == nvlist_lookup_string(parent, "name",
2642 &pname));
2643 (void) snprintf(tryname, sizeof (tryname),
2644 "%s%s", pname, strrchr(stream_fsname, '/'));
2645 } else {
2646 tryname[0] = '\0';
2647 if (flags->verbose) {
2648 (void) printf("local fs %s new parent "
2649 "not found\n", fsname);
2650 }
2651 }
2652
2653 newname[0] = '\0';
2654
2655 error = recv_rename(hdl, fsname, tryname,
2656 strlen(tofs)+1, newname, flags);
2657
2658 if (renamed != NULL && newname[0] != '\0') {
2659 VERIFY(0 == nvlist_add_boolean(renamed,
2660 newname));
2661 }
2662
2663 if (error)
2664 needagain = B_TRUE;
2665 else
2666 progress = B_TRUE;
2667 }
2668 }
2669
2670 fsavl_destroy(local_avl);
2671 nvlist_free(local_nv);
2672
2673 if (needagain && progress) {
2674 /* do another pass to fix up temporary names */
2675 if (flags->verbose)
2676 (void) printf("another pass:\n");
2677 goto again;
2678 }
2679
2680 return (needagain);
2681 }
2682
2683 static int
2684 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2685 recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2686 char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2687 {
2688 nvlist_t *stream_nv = NULL;
2689 avl_tree_t *stream_avl = NULL;
2690 char *fromsnap = NULL;
2691 char *sendsnap = NULL;
2692 char *cp;
2693 char tofs[ZFS_MAX_DATASET_NAME_LEN];
2694 char sendfs[ZFS_MAX_DATASET_NAME_LEN];
2695 char errbuf[1024];
2696 dmu_replay_record_t drre;
2697 int error;
2698 boolean_t anyerr = B_FALSE;
2699 boolean_t softerr = B_FALSE;
2700 boolean_t recursive;
2701
2702 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2703 "cannot receive"));
2704
2705 assert(drr->drr_type == DRR_BEGIN);
2706 assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2707 assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2708 DMU_COMPOUNDSTREAM);
2709
2710 /*
2711 * Read in the nvlist from the stream.
2712 */
2713 if (drr->drr_payloadlen != 0) {
2714 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2715 &stream_nv, flags->byteswap, zc);
2716 if (error) {
2717 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2718 goto out;
2719 }
2720 }
2721
2722 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2723 ENOENT);
2724
2725 if (recursive && strchr(destname, '@')) {
2726 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2727 "cannot specify snapshot name for multi-snapshot stream"));
2728 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2729 goto out;
2730 }
2731
2732 /*
2733 * Read in the end record and verify checksum.
2734 */
2735 if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2736 flags->byteswap, NULL)))
2737 goto out;
2738 if (flags->byteswap) {
2739 drre.drr_type = BSWAP_32(drre.drr_type);
2740 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2741 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2742 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2743 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2744 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2745 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2746 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2747 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2748 }
2749 if (drre.drr_type != DRR_END) {
2750 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2751 goto out;
2752 }
2753 if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2754 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2755 "incorrect header checksum"));
2756 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2757 goto out;
2758 }
2759
2760 (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2761
2762 if (drr->drr_payloadlen != 0) {
2763 nvlist_t *stream_fss;
2764
2765 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2766 &stream_fss));
2767 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2768 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2769 "couldn't allocate avl tree"));
2770 error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2771 goto out;
2772 }
2773
2774 if (fromsnap != NULL && recursive) {
2775 nvlist_t *renamed = NULL;
2776 nvpair_t *pair = NULL;
2777
2778 (void) strlcpy(tofs, destname, sizeof (tofs));
2779 if (flags->isprefix) {
2780 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2781 int i;
2782
2783 if (flags->istail) {
2784 cp = strrchr(drrb->drr_toname, '/');
2785 if (cp == NULL) {
2786 (void) strlcat(tofs, "/",
2787 sizeof (tofs));
2788 i = 0;
2789 } else {
2790 i = (cp - drrb->drr_toname);
2791 }
2792 } else {
2793 i = strcspn(drrb->drr_toname, "/@");
2794 }
2795 /* zfs_receive_one() will create_parents() */
2796 (void) strlcat(tofs, &drrb->drr_toname[i],
2797 sizeof (tofs));
2798 *strchr(tofs, '@') = '\0';
2799 }
2800
2801 if (!flags->dryrun && !flags->nomount) {
2802 VERIFY(0 == nvlist_alloc(&renamed,
2803 NV_UNIQUE_NAME, 0));
2804 }
2805
2806 softerr = recv_incremental_replication(hdl, tofs, flags,
2807 stream_nv, stream_avl, renamed);
2808
2809 /* Unmount renamed filesystems before receiving. */
2810 while ((pair = nvlist_next_nvpair(renamed,
2811 pair)) != NULL) {
2812 zfs_handle_t *zhp;
2813 prop_changelist_t *clp = NULL;
2814
2815 zhp = zfs_open(hdl, nvpair_name(pair),
2816 ZFS_TYPE_FILESYSTEM);
2817 if (zhp != NULL) {
2818 clp = changelist_gather(zhp,
2819 ZFS_PROP_MOUNTPOINT, 0, 0);
2820 zfs_close(zhp);
2821 if (clp != NULL) {
2822 softerr |=
2823 changelist_prefix(clp);
2824 changelist_free(clp);
2825 }
2826 }
2827 }
2828
2829 nvlist_free(renamed);
2830 }
2831 }
2832
2833 /*
2834 * Get the fs specified by the first path in the stream (the top level
2835 * specified by 'zfs send') and pass it to each invocation of
2836 * zfs_receive_one().
2837 */
2838 (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2839 sizeof (sendfs));
2840 if ((cp = strchr(sendfs, '@')) != NULL) {
2841 *cp = '\0';
2842 /*
2843 * Find the "sendsnap", the final snapshot in a replication
2844 * stream. zfs_receive_one() handles certain errors
2845 * differently, depending on if the contained stream is the
2846 * last one or not.
2847 */
2848 sendsnap = (cp + 1);
2849 }
2850
2851 /* Finally, receive each contained stream */
2852 do {
2853 /*
2854 * we should figure out if it has a recoverable
2855 * error, in which case do a recv_skip() and drive on.
2856 * Note, if we fail due to already having this guid,
2857 * zfs_receive_one() will take care of it (ie,
2858 * recv_skip() and return 0).
2859 */
2860 error = zfs_receive_impl(hdl, destname, NULL, flags, fd,
2861 sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2862 action_handlep, sendsnap);
2863 if (error == ENODATA) {
2864 error = 0;
2865 break;
2866 }
2867 anyerr |= error;
2868 } while (error == 0);
2869
2870 if (drr->drr_payloadlen != 0 && recursive && fromsnap != NULL) {
2871 /*
2872 * Now that we have the fs's they sent us, try the
2873 * renames again.
2874 */
2875 softerr = recv_incremental_replication(hdl, tofs, flags,
2876 stream_nv, stream_avl, NULL);
2877 }
2878
2879 out:
2880 fsavl_destroy(stream_avl);
2881 nvlist_free(stream_nv);
2882 if (softerr)
2883 error = -2;
2884 if (anyerr)
2885 error = -1;
2886 return (error);
2887 }
2888
2889 static void
2890 trunc_prop_errs(int truncated)
2891 {
2892 ASSERT(truncated != 0);
2893
2894 if (truncated == 1)
2895 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2896 "1 more property could not be set\n"));
2897 else
2898 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2899 "%d more properties could not be set\n"), truncated);
2900 }
2901
2902 static int
2903 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2904 {
2905 dmu_replay_record_t *drr;
2906 void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
2907 char errbuf[1024];
2908
2909 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2910 "cannot receive:"));
2911
2912 /* XXX would be great to use lseek if possible... */
2913 drr = buf;
2914
2915 while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2916 byteswap, NULL) == 0) {
2917 if (byteswap)
2918 drr->drr_type = BSWAP_32(drr->drr_type);
2919
2920 switch (drr->drr_type) {
2921 case DRR_BEGIN:
2922 if (drr->drr_payloadlen != 0) {
2923 (void) recv_read(hdl, fd, buf,
2924 drr->drr_payloadlen, B_FALSE, NULL);
2925 }
2926 break;
2927
2928 case DRR_END:
2929 free(buf);
2930 return (0);
2931
2932 case DRR_OBJECT:
2933 if (byteswap) {
2934 drr->drr_u.drr_object.drr_bonuslen =
2935 BSWAP_32(drr->drr_u.drr_object.
2936 drr_bonuslen);
2937 }
2938 (void) recv_read(hdl, fd, buf,
2939 P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2940 B_FALSE, NULL);
2941 break;
2942
2943 case DRR_WRITE:
2944 if (byteswap) {
2945 drr->drr_u.drr_write.drr_logical_size =
2946 BSWAP_64(
2947 drr->drr_u.drr_write.drr_logical_size);
2948 drr->drr_u.drr_write.drr_compressed_size =
2949 BSWAP_64(
2950 drr->drr_u.drr_write.drr_compressed_size);
2951 }
2952 uint64_t payload_size =
2953 DRR_WRITE_PAYLOAD_SIZE(&drr->drr_u.drr_write);
2954 (void) recv_read(hdl, fd, buf,
2955 payload_size, B_FALSE, NULL);
2956 break;
2957 case DRR_SPILL:
2958 if (byteswap) {
2959 drr->drr_u.drr_spill.drr_length =
2960 BSWAP_64(drr->drr_u.drr_spill.drr_length);
2961 }
2962 (void) recv_read(hdl, fd, buf,
2963 drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2964 break;
2965 case DRR_WRITE_EMBEDDED:
2966 if (byteswap) {
2967 drr->drr_u.drr_write_embedded.drr_psize =
2968 BSWAP_32(drr->drr_u.drr_write_embedded.
2969 drr_psize);
2970 }
2971 (void) recv_read(hdl, fd, buf,
2972 P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
2973 8), B_FALSE, NULL);
2974 break;
2975 case DRR_WRITE_BYREF:
2976 case DRR_FREEOBJECTS:
2977 case DRR_FREE:
2978 break;
2979
2980 default:
2981 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2982 "invalid record type"));
2983 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2984 }
2985 }
2986
2987 free(buf);
2988 return (-1);
2989 }
2990
2991 static void
2992 recv_ecksum_set_aux(libzfs_handle_t *hdl, const char *target_snap,
2993 boolean_t resumable)
2994 {
2995 char target_fs[ZFS_MAX_DATASET_NAME_LEN];
2996
2997 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2998 "checksum mismatch or incomplete stream"));
2999
3000 if (!resumable)
3001 return;
3002 (void) strlcpy(target_fs, target_snap, sizeof (target_fs));
3003 *strchr(target_fs, '@') = '\0';
3004 zfs_handle_t *zhp = zfs_open(hdl, target_fs,
3005 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3006 if (zhp == NULL)
3007 return;
3008
3009 char token_buf[ZFS_MAXPROPLEN];
3010 int error = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN,
3011 token_buf, sizeof (token_buf),
3012 NULL, NULL, 0, B_TRUE);
3013 if (error == 0) {
3014 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3015 "checksum mismatch or incomplete stream.\n"
3016 "Partially received snapshot is saved.\n"
3017 "A resuming stream can be generated on the sending "
3018 "system by running:\n"
3019 " zfs send -t %s"),
3020 token_buf);
3021 }
3022 zfs_close(zhp);
3023 }
3024
3025 /*
3026 * Restores a backup of tosnap from the file descriptor specified by infd.
3027 */
3028 static int
3029 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
3030 const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr,
3031 dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv,
3032 avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
3033 uint64_t *action_handlep, const char *finalsnap)
3034 {
3035 zfs_cmd_t zc = { 0 };
3036 time_t begin_time;
3037 int ioctl_err, ioctl_errno, err;
3038 char *cp;
3039 struct drr_begin *drrb = &drr->drr_u.drr_begin;
3040 char errbuf[1024];
3041 char prop_errbuf[1024];
3042 const char *chopprefix;
3043 boolean_t newfs = B_FALSE;
3044 boolean_t stream_wantsnewfs;
3045 uint64_t parent_snapguid = 0;
3046 prop_changelist_t *clp = NULL;
3047 nvlist_t *snapprops_nvlist = NULL;
3048 zprop_errflags_t prop_errflags;
3049 boolean_t recursive;
3050 char *snapname = NULL;
3051
3052 begin_time = time(NULL);
3053
3054 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3055 "cannot receive"));
3056
3057 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
3058 ENOENT);
3059
3060 if (stream_avl != NULL) {
3061 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
3062 &snapname);
3063 nvlist_t *props;
3064 int ret;
3065
3066 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
3067 &parent_snapguid);
3068 err = nvlist_lookup_nvlist(fs, "props", &props);
3069 if (err)
3070 VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
3071
3072 if (flags->canmountoff) {
3073 VERIFY(0 == nvlist_add_uint64(props,
3074 zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
3075 }
3076 ret = zcmd_write_src_nvlist(hdl, &zc, props);
3077 if (err)
3078 nvlist_free(props);
3079
3080 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
3081 VERIFY(0 == nvlist_lookup_nvlist(props,
3082 snapname, &snapprops_nvlist));
3083 }
3084
3085 if (ret != 0)
3086 return (-1);
3087 }
3088
3089 cp = NULL;
3090
3091 /*
3092 * Determine how much of the snapshot name stored in the stream
3093 * we are going to tack on to the name they specified on the
3094 * command line, and how much we are going to chop off.
3095 *
3096 * If they specified a snapshot, chop the entire name stored in
3097 * the stream.
3098 */
3099 if (flags->istail) {
3100 /*
3101 * A filesystem was specified with -e. We want to tack on only
3102 * the tail of the sent snapshot path.
3103 */
3104 if (strchr(tosnap, '@')) {
3105 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3106 "argument - snapshot not allowed with -e"));
3107 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3108 }
3109
3110 chopprefix = strrchr(sendfs, '/');
3111
3112 if (chopprefix == NULL) {
3113 /*
3114 * The tail is the poolname, so we need to
3115 * prepend a path separator.
3116 */
3117 int len = strlen(drrb->drr_toname);
3118 cp = malloc(len + 2);
3119 cp[0] = '/';
3120 (void) strcpy(&cp[1], drrb->drr_toname);
3121 chopprefix = cp;
3122 } else {
3123 chopprefix = drrb->drr_toname + (chopprefix - sendfs);
3124 }
3125 } else if (flags->isprefix) {
3126 /*
3127 * A filesystem was specified with -d. We want to tack on
3128 * everything but the first element of the sent snapshot path
3129 * (all but the pool name).
3130 */
3131 if (strchr(tosnap, '@')) {
3132 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3133 "argument - snapshot not allowed with -d"));
3134 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3135 }
3136
3137 chopprefix = strchr(drrb->drr_toname, '/');
3138 if (chopprefix == NULL)
3139 chopprefix = strchr(drrb->drr_toname, '@');
3140 } else if (strchr(tosnap, '@') == NULL) {
3141 /*
3142 * If a filesystem was specified without -d or -e, we want to
3143 * tack on everything after the fs specified by 'zfs send'.
3144 */
3145 chopprefix = drrb->drr_toname + strlen(sendfs);
3146 } else {
3147 /* A snapshot was specified as an exact path (no -d or -e). */
3148 if (recursive) {
3149 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3150 "cannot specify snapshot name for multi-snapshot "
3151 "stream"));
3152 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3153 }
3154 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
3155 }
3156
3157 ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
3158 ASSERT(chopprefix > drrb->drr_toname);
3159 ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
3160 ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
3161 chopprefix[0] == '\0');
3162
3163 /*
3164 * Determine name of destination snapshot, store in zc_value.
3165 */
3166 (void) strcpy(zc.zc_value, tosnap);
3167 (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
3168 free(cp);
3169 if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
3170 zcmd_free_nvlists(&zc);
3171 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3172 }
3173
3174 /*
3175 * Determine the name of the origin snapshot, store in zc_string.
3176 */
3177 if (originsnap) {
3178 (void) strncpy(zc.zc_string, originsnap, sizeof (zc.zc_string));
3179 if (flags->verbose)
3180 (void) printf("using provided clone origin %s\n",
3181 zc.zc_string);
3182 } else if (drrb->drr_flags & DRR_FLAG_CLONE) {
3183 if (guid_to_name(hdl, zc.zc_value,
3184 drrb->drr_fromguid, B_FALSE, zc.zc_string) != 0) {
3185 zcmd_free_nvlists(&zc);
3186 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3187 "local origin for clone %s does not exist"),
3188 zc.zc_value);
3189 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3190 }
3191 if (flags->verbose)
3192 (void) printf("found clone origin %s\n", zc.zc_string);
3193 }
3194
3195 boolean_t resuming = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
3196 DMU_BACKUP_FEATURE_RESUMING;
3197 stream_wantsnewfs = (drrb->drr_fromguid == NULL ||
3198 (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && !resuming;
3199
3200 if (stream_wantsnewfs) {
3201 /*
3202 * if the parent fs does not exist, look for it based on
3203 * the parent snap GUID
3204 */
3205 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3206 "cannot receive new filesystem stream"));
3207
3208 (void) strcpy(zc.zc_name, zc.zc_value);
3209 cp = strrchr(zc.zc_name, '/');
3210 if (cp)
3211 *cp = '\0';
3212 if (cp &&
3213 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3214 char suffix[ZFS_MAX_DATASET_NAME_LEN];
3215 (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
3216 if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
3217 B_FALSE, zc.zc_value) == 0) {
3218 *strchr(zc.zc_value, '@') = '\0';
3219 (void) strcat(zc.zc_value, suffix);
3220 }
3221 }
3222 } else {
3223 /*
3224 * if the fs does not exist, look for it based on the
3225 * fromsnap GUID
3226 */
3227 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3228 "cannot receive incremental stream"));
3229
3230 (void) strcpy(zc.zc_name, zc.zc_value);
3231 *strchr(zc.zc_name, '@') = '\0';
3232
3233 /*
3234 * If the exact receive path was specified and this is the
3235 * topmost path in the stream, then if the fs does not exist we
3236 * should look no further.
3237 */
3238 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
3239 strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
3240 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3241 char snap[ZFS_MAX_DATASET_NAME_LEN];
3242 (void) strcpy(snap, strchr(zc.zc_value, '@'));
3243 if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
3244 B_FALSE, zc.zc_value) == 0) {
3245 *strchr(zc.zc_value, '@') = '\0';
3246 (void) strcat(zc.zc_value, snap);
3247 }
3248 }
3249 }
3250
3251 (void) strcpy(zc.zc_name, zc.zc_value);
3252 *strchr(zc.zc_name, '@') = '\0';
3253
3254 if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3255 zfs_handle_t *zhp;
3256
3257 /*
3258 * Destination fs exists. It must be one of these cases:
3259 * - an incremental send stream
3260 * - the stream specifies a new fs (full stream or clone)
3261 * and they want us to blow away the existing fs (and
3262 * have therefore specified -F and removed any snapshots)
3263 * - we are resuming a failed receive.
3264 */
3265 if (stream_wantsnewfs) {
3266 if (!flags->force) {
3267 zcmd_free_nvlists(&zc);
3268 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3269 "destination '%s' exists\n"
3270 "must specify -F to overwrite it"),
3271 zc.zc_name);
3272 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3273 }
3274 if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
3275 &zc) == 0) {
3276 zcmd_free_nvlists(&zc);
3277 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3278 "destination has snapshots (eg. %s)\n"
3279 "must destroy them to overwrite it"),
3280 zc.zc_name);
3281 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3282 }
3283 }
3284
3285 if ((zhp = zfs_open(hdl, zc.zc_name,
3286 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
3287 zcmd_free_nvlists(&zc);
3288 return (-1);
3289 }
3290
3291 if (stream_wantsnewfs &&
3292 zhp->zfs_dmustats.dds_origin[0]) {
3293 zcmd_free_nvlists(&zc);
3294 zfs_close(zhp);
3295 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3296 "destination '%s' is a clone\n"
3297 "must destroy it to overwrite it"),
3298 zc.zc_name);
3299 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3300 }
3301
3302 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
3303 stream_wantsnewfs) {
3304 /* We can't do online recv in this case */
3305 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
3306 if (clp == NULL) {
3307 zfs_close(zhp);
3308 zcmd_free_nvlists(&zc);
3309 return (-1);
3310 }
3311 if (changelist_prefix(clp) != 0) {
3312 changelist_free(clp);
3313 zfs_close(zhp);
3314 zcmd_free_nvlists(&zc);
3315 return (-1);
3316 }
3317 }
3318
3319 /*
3320 * If we are resuming a newfs, set newfs here so that we will
3321 * mount it if the recv succeeds this time. We can tell
3322 * that it was a newfs on the first recv because the fs
3323 * itself will be inconsistent (if the fs existed when we
3324 * did the first recv, we would have received it into
3325 * .../%recv).
3326 */
3327 if (resuming && zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT))
3328 newfs = B_TRUE;
3329
3330 zfs_close(zhp);
3331 } else {
3332 /*
3333 * Destination filesystem does not exist. Therefore we better
3334 * be creating a new filesystem (either from a full backup, or
3335 * a clone). It would therefore be invalid if the user
3336 * specified only the pool name (i.e. if the destination name
3337 * contained no slash character).
3338 */
3339 if (!stream_wantsnewfs ||
3340 (cp = strrchr(zc.zc_name, '/')) == NULL) {
3341 zcmd_free_nvlists(&zc);
3342 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3343 "destination '%s' does not exist"), zc.zc_name);
3344 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3345 }
3346
3347 /*
3348 * Trim off the final dataset component so we perform the
3349 * recvbackup ioctl to the filesystems's parent.
3350 */
3351 *cp = '\0';
3352
3353 if (flags->isprefix && !flags->istail && !flags->dryrun &&
3354 create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
3355 zcmd_free_nvlists(&zc);
3356 return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
3357 }
3358
3359 newfs = B_TRUE;
3360 }
3361
3362 zc.zc_begin_record = *drr_noswap;
3363 zc.zc_cookie = infd;
3364 zc.zc_guid = flags->force;
3365 zc.zc_resumable = flags->resumable;
3366 if (flags->verbose) {
3367 (void) printf("%s %s stream of %s into %s\n",
3368 flags->dryrun ? "would receive" : "receiving",
3369 drrb->drr_fromguid ? "incremental" : "full",
3370 drrb->drr_toname, zc.zc_value);
3371 (void) fflush(stdout);
3372 }
3373
3374 if (flags->dryrun) {
3375 zcmd_free_nvlists(&zc);
3376 return (recv_skip(hdl, infd, flags->byteswap));
3377 }
3378
3379 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
3380 zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
3381 zc.zc_cleanup_fd = cleanup_fd;
3382 zc.zc_action_handle = *action_handlep;
3383
3384 err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
3385 ioctl_errno = errno;
3386 prop_errflags = (zprop_errflags_t)zc.zc_obj;
3387
3388 if (err == 0) {
3389 nvlist_t *prop_errors;
3390 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
3391 zc.zc_nvlist_dst_size, &prop_errors, 0));
3392
3393 nvpair_t *prop_err = NULL;
3394
3395 while ((prop_err = nvlist_next_nvpair(prop_errors,
3396 prop_err)) != NULL) {
3397 char tbuf[1024];
3398 zfs_prop_t prop;
3399 int intval;
3400
3401 prop = zfs_name_to_prop(nvpair_name(prop_err));
3402 (void) nvpair_value_int32(prop_err, &intval);
3403 if (strcmp(nvpair_name(prop_err),
3404 ZPROP_N_MORE_ERRORS) == 0) {
3405 trunc_prop_errs(intval);
3406 break;
3407 } else if (snapname == NULL || finalsnap == NULL ||
3408 strcmp(finalsnap, snapname) == 0 ||
3409 strcmp(nvpair_name(prop_err),
3410 zfs_prop_to_name(ZFS_PROP_REFQUOTA)) != 0) {
3411 /*
3412 * Skip the special case of, for example,
3413 * "refquota", errors on intermediate
3414 * snapshots leading up to a final one.
3415 * That's why we have all of the checks above.
3416 *
3417 * See zfs_ioctl.c's extract_delay_props() for
3418 * a list of props which can fail on
3419 * intermediate snapshots, but shouldn't
3420 * affect the overall receive.
3421 */
3422 (void) snprintf(tbuf, sizeof (tbuf),
3423 dgettext(TEXT_DOMAIN,
3424 "cannot receive %s property on %s"),
3425 nvpair_name(prop_err), zc.zc_name);
3426 zfs_setprop_error(hdl, prop, intval, tbuf);
3427 }
3428 }
3429 nvlist_free(prop_errors);
3430 }
3431
3432 zc.zc_nvlist_dst = 0;
3433 zc.zc_nvlist_dst_size = 0;
3434 zcmd_free_nvlists(&zc);
3435
3436 if (err == 0 && snapprops_nvlist) {
3437 zfs_cmd_t zc2 = { 0 };
3438
3439 (void) strcpy(zc2.zc_name, zc.zc_value);
3440 zc2.zc_cookie = B_TRUE; /* received */
3441 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
3442 (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
3443 zcmd_free_nvlists(&zc2);
3444 }
3445 }
3446
3447 if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
3448 /*
3449 * It may be that this snapshot already exists,
3450 * in which case we want to consume & ignore it
3451 * rather than failing.
3452 */
3453 avl_tree_t *local_avl;
3454 nvlist_t *local_nv, *fs;
3455 cp = strchr(zc.zc_value, '@');
3456
3457 /*
3458 * XXX Do this faster by just iterating over snaps in
3459 * this fs. Also if zc_value does not exist, we will
3460 * get a strange "does not exist" error message.
3461 */
3462 *cp = '\0';
3463 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3464 B_FALSE, &local_nv, &local_avl) == 0) {
3465 *cp = '@';
3466 fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3467 fsavl_destroy(local_avl);
3468 nvlist_free(local_nv);
3469
3470 if (fs != NULL) {
3471 if (flags->verbose) {
3472 (void) printf("snap %s already exists; "
3473 "ignoring\n", zc.zc_value);
3474 }
3475 err = ioctl_err = recv_skip(hdl, infd,
3476 flags->byteswap);
3477 }
3478 }
3479 *cp = '@';
3480 }
3481
3482 if (ioctl_err != 0) {
3483 switch (ioctl_errno) {
3484 case ENODEV:
3485 cp = strchr(zc.zc_value, '@');
3486 *cp = '\0';
3487 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3488 "most recent snapshot of %s does not\n"
3489 "match incremental source"), zc.zc_value);
3490 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3491 *cp = '@';
3492 break;
3493 case ETXTBSY:
3494 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3495 "destination %s has been modified\n"
3496 "since most recent snapshot"), zc.zc_name);
3497 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3498 break;
3499 case EEXIST:
3500 cp = strchr(zc.zc_value, '@');
3501 if (newfs) {
3502 /* it's the containing fs that exists */
3503 *cp = '\0';
3504 }
3505 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3506 "destination already exists"));
3507 (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3508 dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3509 zc.zc_value);
3510 *cp = '@';
3511 break;
3512 case EINVAL:
3513 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3514 break;
3515 case ECKSUM:
3516 recv_ecksum_set_aux(hdl, zc.zc_value, flags->resumable);
3517 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3518 break;
3519 case ENOTSUP:
3520 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3521 "pool must be upgraded to receive this stream."));
3522 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3523 break;
3524 case EDQUOT:
3525 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3526 "destination %s space quota exceeded"), zc.zc_name);
3527 (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3528 break;
3529 default:
3530 (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3531 }
3532 }
3533
3534 /*
3535 * Mount the target filesystem (if created). Also mount any
3536 * children of the target filesystem if we did a replication
3537 * receive (indicated by stream_avl being non-NULL).
3538 */
3539 cp = strchr(zc.zc_value, '@');
3540 if (cp && (ioctl_err == 0 || !newfs)) {
3541 zfs_handle_t *h;
3542
3543 *cp = '\0';
3544 h = zfs_open(hdl, zc.zc_value,
3545 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3546 if (h != NULL) {
3547 if (h->zfs_type == ZFS_TYPE_VOLUME) {
3548 *cp = '@';
3549 } else if (newfs || stream_avl) {
3550 /*
3551 * Track the first/top of hierarchy fs,
3552 * for mounting and sharing later.
3553 */
3554 if (top_zfs && *top_zfs == NULL)
3555 *top_zfs = zfs_strdup(hdl, zc.zc_value);
3556 }
3557 zfs_close(h);
3558 }
3559 *cp = '@';
3560 }
3561
3562 if (clp) {
3563 if (!flags->nomount)
3564 err |= changelist_postfix(clp);
3565 changelist_free(clp);
3566 }
3567
3568 if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3569 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3570 "failed to clear unreceived properties on %s"),
3571 zc.zc_name);
3572 (void) fprintf(stderr, "\n");
3573 }
3574 if (prop_errflags & ZPROP_ERR_NORESTORE) {
3575 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3576 "failed to restore original properties on %s"),
3577 zc.zc_name);
3578 (void) fprintf(stderr, "\n");
3579 }
3580
3581 if (err || ioctl_err)
3582 return (-1);
3583
3584 *action_handlep = zc.zc_action_handle;
3585
3586 if (flags->verbose) {
3587 char buf1[64];
3588 char buf2[64];
3589 uint64_t bytes = zc.zc_cookie;
3590 time_t delta = time(NULL) - begin_time;
3591 if (delta == 0)
3592 delta = 1;
3593 zfs_nicenum(bytes, buf1, sizeof (buf1));
3594 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3595
3596 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3597 buf1, delta, buf2);
3598 }
3599
3600 return (0);
3601 }
3602
3603 static int
3604 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap,
3605 const char *originsnap, recvflags_t *flags, int infd, const char *sendfs,
3606 nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
3607 uint64_t *action_handlep, const char *finalsnap)
3608 {
3609 int err;
3610 dmu_replay_record_t drr, drr_noswap;
3611 struct drr_begin *drrb = &drr.drr_u.drr_begin;
3612 char errbuf[1024];
3613 zio_cksum_t zcksum = { 0 };
3614 uint64_t featureflags;
3615 int hdrtype;
3616
3617 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3618 "cannot receive"));
3619
3620 if (flags->isprefix &&
3621 !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3622 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3623 "(%s) does not exist"), tosnap);
3624 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3625 }
3626 if (originsnap &&
3627 !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) {
3628 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs "
3629 "(%s) does not exist"), originsnap);
3630 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3631 }
3632
3633 /* read in the BEGIN record */
3634 if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3635 &zcksum)))
3636 return (err);
3637
3638 if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3639 /* It's the double end record at the end of a package */
3640 return (ENODATA);
3641 }
3642
3643 /* the kernel needs the non-byteswapped begin record */
3644 drr_noswap = drr;
3645
3646 flags->byteswap = B_FALSE;
3647 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3648 /*
3649 * We computed the checksum in the wrong byteorder in
3650 * recv_read() above; do it again correctly.
3651 */
3652 bzero(&zcksum, sizeof (zio_cksum_t));
3653 (void) fletcher_4_incremental_byteswap(&drr,
3654 sizeof (drr), &zcksum);
3655 flags->byteswap = B_TRUE;
3656
3657 drr.drr_type = BSWAP_32(drr.drr_type);
3658 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3659 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3660 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3661 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3662 drrb->drr_type = BSWAP_32(drrb->drr_type);
3663 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3664 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3665 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3666 }
3667
3668 if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3669 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3670 "stream (bad magic number)"));
3671 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3672 }
3673
3674 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3675 hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3676
3677 if (!DMU_STREAM_SUPPORTED(featureflags) ||
3678 (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3679 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3680 "stream has unsupported feature, feature flags = %lx"),
3681 featureflags);
3682 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3683 }
3684
3685 if (strchr(drrb->drr_toname, '@') == NULL) {
3686 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3687 "stream (bad snapshot name)"));
3688 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3689 }
3690
3691 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3692 char nonpackage_sendfs[ZFS_MAX_DATASET_NAME_LEN];
3693 if (sendfs == NULL) {
3694 /*
3695 * We were not called from zfs_receive_package(). Get
3696 * the fs specified by 'zfs send'.
3697 */
3698 char *cp;
3699 (void) strlcpy(nonpackage_sendfs,
3700 drr.drr_u.drr_begin.drr_toname,
3701 sizeof (nonpackage_sendfs));
3702 if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3703 *cp = '\0';
3704 sendfs = nonpackage_sendfs;
3705 VERIFY(finalsnap == NULL);
3706 }
3707 return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags,
3708 &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs,
3709 cleanup_fd, action_handlep, finalsnap));
3710 } else {
3711 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3712 DMU_COMPOUNDSTREAM);
3713 return (zfs_receive_package(hdl, infd, tosnap, flags, &drr,
3714 &zcksum, top_zfs, cleanup_fd, action_handlep));
3715 }
3716 }
3717
3718 /*
3719 * Restores a backup of tosnap from the file descriptor specified by infd.
3720 * Return 0 on total success, -2 if some things couldn't be
3721 * destroyed/renamed/promoted, -1 if some things couldn't be received.
3722 * (-1 will override -2, if -1 and the resumable flag was specified the
3723 * transfer can be resumed if the sending side supports it).
3724 */
3725 int
3726 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props,
3727 recvflags_t *flags, int infd, avl_tree_t *stream_avl)
3728 {
3729 char *top_zfs = NULL;
3730 int err;
3731 int cleanup_fd;
3732 uint64_t action_handle = 0;
3733 char *originsnap = NULL;
3734 if (props) {
3735 err = nvlist_lookup_string(props, "origin", &originsnap);
3736 if (err && err != ENOENT)
3737 return (err);
3738 }
3739
3740 cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3741 VERIFY(cleanup_fd >= 0);
3742
3743 err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL,
3744 stream_avl, &top_zfs, cleanup_fd, &action_handle, NULL);
3745
3746 VERIFY(0 == close(cleanup_fd));
3747
3748 if (err == 0 && !flags->nomount && top_zfs) {
3749 zfs_handle_t *zhp;
3750 prop_changelist_t *clp;
3751
3752 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3753 if (zhp != NULL) {
3754 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3755 CL_GATHER_MOUNT_ALWAYS, 0);
3756 zfs_close(zhp);
3757 if (clp != NULL) {
3758 /* mount and share received datasets */
3759 err = changelist_postfix(clp);
3760 changelist_free(clp);
3761 }
3762 }
3763 if (zhp == NULL || clp == NULL || err)
3764 err = -1;
3765 }
3766 if (top_zfs)
3767 free(top_zfs);
3768
3769 return (err);
3770 }