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