1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012 by Delphix. All rights reserved.
25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
27 */
28
29 #include <assert.h>
30 #include <ctype.h>
31 #include <errno.h>
32 #include <libintl.h>
33 #include <stdio.h>
34 #include <stdlib.h>
35 #include <strings.h>
36 #include <unistd.h>
37 #include <stddef.h>
38 #include <fcntl.h>
39 #include <sys/mount.h>
40 #include <pthread.h>
41 #include <umem.h>
42 #include <time.h>
43
44 #include <libzfs.h>
45
46 #include "zfs_namecheck.h"
47 #include "zfs_prop.h"
48 #include "zfs_fletcher.h"
49 #include "libzfs_impl.h"
50 #include <sha2.h>
51 #include <sys/zio_checksum.h>
52 #include <sys/ddt.h>
53
54 /* in libzfs_dataset.c */
55 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
56
57 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t *,
58 int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *);
59
60 static const zio_cksum_t zero_cksum = { 0 };
61
62 typedef struct dedup_arg {
63 int inputfd;
64 int outputfd;
65 uint64_t dedup_data_sz;
66 boolean_t sendsize;
67 libzfs_handle_t *dedup_hdl;
68 } dedup_arg_t;
69
70 typedef struct progress_arg {
71 zfs_handle_t *pa_zhp;
72 int pa_fd;
73 boolean_t pa_parsable;
74 } progress_arg_t;
75
76 typedef struct dataref {
77 uint64_t ref_guid;
78 uint64_t ref_object;
79 uint64_t ref_offset;
80 } dataref_t;
81
82 typedef struct dedup_entry {
83 struct dedup_entry *dde_next;
84 zio_cksum_t dde_chksum;
85 uint64_t dde_prop;
86 dataref_t dde_ref;
87 } dedup_entry_t;
88
89 #define MAX_DDT_PHYSMEM_PERCENT 20
90 #define SMALLEST_POSSIBLE_MAX_DDT_MB 128
91
92 typedef struct dedup_table {
93 dedup_entry_t **dedup_hash_array;
94 umem_cache_t *ddecache;
95 uint64_t max_ddt_size; /* max dedup table size in bytes */
96 uint64_t cur_ddt_size; /* current dedup table size in bytes */
97 uint64_t ddt_count;
98 int numhashbits;
99 boolean_t ddt_full;
100 } dedup_table_t;
101
102 static int
103 high_order_bit(uint64_t n)
104 {
105 int count;
106
107 for (count = 0; n != 0; count++)
108 n >>= 1;
109 return (count);
110 }
111
112 static size_t
113 ssread(void *buf, size_t len, FILE *stream)
114 {
115 size_t outlen;
116
117 if ((outlen = fread(buf, len, 1, stream)) == 0)
118 return (0);
119
120 return (outlen);
121 }
122
123 static void
124 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
125 zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
126 {
127 dedup_entry_t *dde;
128
129 if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
130 if (ddt->ddt_full == B_FALSE) {
131 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
132 "Dedup table full. Deduplication will continue "
133 "with existing table entries"));
134 ddt->ddt_full = B_TRUE;
135 }
136 return;
137 }
138
139 if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
140 != NULL) {
141 assert(*ddepp == NULL);
142 dde->dde_next = NULL;
143 dde->dde_chksum = *cs;
144 dde->dde_prop = prop;
145 dde->dde_ref = *dr;
146 *ddepp = dde;
147 ddt->cur_ddt_size += sizeof (dedup_entry_t);
148 ddt->ddt_count++;
149 }
150 }
151
152 /*
153 * Using the specified dedup table, do a lookup for an entry with
154 * the checksum cs. If found, return the block's reference info
155 * in *dr. Otherwise, insert a new entry in the dedup table, using
156 * the reference information specified by *dr.
157 *
158 * return value: true - entry was found
159 * false - entry was not found
160 */
161 static boolean_t
162 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
163 uint64_t prop, dataref_t *dr)
164 {
165 uint32_t hashcode;
166 dedup_entry_t **ddepp;
167
168 hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
169
170 for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
171 ddepp = &((*ddepp)->dde_next)) {
172 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
173 (*ddepp)->dde_prop == prop) {
174 *dr = (*ddepp)->dde_ref;
175 return (B_TRUE);
176 }
177 }
178 ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
179 return (B_FALSE);
180 }
181
182 static int
183 cksum_and_write(const void *buf, uint64_t len, zio_cksum_t *zc, int outfd)
184 {
185 fletcher_4_incremental_native(buf, len, zc);
186 return (write(outfd, buf, len));
187 }
188
189 /*
190 * the function used by the cksummer thread that needs to know
191 * about the sendsize flag
192 */
193 static int
194 dedup_cksum_and_write(dedup_arg_t *dda, const void *buf, uint64_t len,
195 zio_cksum_t *zc, int outfd)
196 {
197 int ret = len;
198
199 dda->dedup_data_sz += len;
200 fletcher_4_incremental_native(buf, len, zc);
201 if (!dda->sendsize)
202 ret = (write(outfd, buf, len));
203
204 return (ret);
205 }
206
207 /*
208 * This function is started in a separate thread when the dedup option
209 * has been requested. The main send thread determines the list of
210 * snapshots to be included in the send stream and makes the ioctl calls
211 * for each one. But instead of having the ioctl send the output to the
212 * the output fd specified by the caller of zfs_send()), the
213 * ioctl is told to direct the output to a pipe, which is read by the
214 * alternate thread running THIS function. This function does the
215 * dedup'ing by:
216 * 1. building a dedup table (the DDT)
217 * 2. doing checksums on each data block and inserting a record in the DDT
218 * 3. looking for matching checksums, and
219 * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever
220 * a duplicate block is found.
221 * The output of this function then goes to the output fd requested
222 * by the caller of zfs_send().
223 */
224 static void *
225 cksummer(void *arg)
226 {
227 dedup_arg_t *dda = arg;
228 char *buf = malloc(1<<20);
229 dmu_replay_record_t thedrr;
230 dmu_replay_record_t *drr = &thedrr;
231 struct drr_begin *drrb = &thedrr.drr_u.drr_begin;
232 struct drr_end *drre = &thedrr.drr_u.drr_end;
233 struct drr_object *drro = &thedrr.drr_u.drr_object;
234 struct drr_write *drrw = &thedrr.drr_u.drr_write;
235 struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
236 FILE *ofp;
237 int outfd;
238 dmu_replay_record_t wbr_drr = {0};
239 struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref;
240 dedup_table_t ddt;
241 zio_cksum_t stream_cksum;
242 uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
243 uint64_t numbuckets;
244
245 ddt.max_ddt_size =
246 MAX((physmem * MAX_DDT_PHYSMEM_PERCENT)/100,
247 SMALLEST_POSSIBLE_MAX_DDT_MB<<20);
248
249 numbuckets = ddt.max_ddt_size/(sizeof (dedup_entry_t));
250
251 /*
252 * numbuckets must be a power of 2. Increase number to
253 * a power of 2 if necessary.
254 */
255 if (!ISP2(numbuckets))
256 numbuckets = 1 << high_order_bit(numbuckets);
257
258 ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
259 ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
260 NULL, NULL, NULL, NULL, NULL, 0);
261 ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
262 ddt.numhashbits = high_order_bit(numbuckets) - 1;
263 ddt.ddt_full = B_FALSE;
264
265 /* Initialize the write-by-reference block. */
266 wbr_drr.drr_type = DRR_WRITE_BYREF;
267 wbr_drr.drr_payloadlen = 0;
268
269 outfd = dda->outputfd;
270 ofp = fdopen(dda->inputfd, "r");
271 while (ssread(drr, sizeof (dmu_replay_record_t), ofp) != 0) {
272
273 switch (drr->drr_type) {
274 case DRR_BEGIN:
275 {
276 int fflags;
277 ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
278
279 /* set the DEDUP feature flag for this stream */
280 fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
281 fflags |= (DMU_BACKUP_FEATURE_DEDUP |
282 DMU_BACKUP_FEATURE_DEDUPPROPS);
283 DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
284
285 if (dedup_cksum_and_write(dda, drr,
286 sizeof (dmu_replay_record_t),
287 &stream_cksum, outfd) == -1)
288 goto out;
289 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
290 DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
291 int sz = drr->drr_payloadlen;
292
293 if (sz > 1<<20) {
294 free(buf);
295 buf = malloc(sz);
296 }
297 (void) ssread(buf, sz, ofp);
298 if (ferror(stdin))
299 perror("fread");
300 if (dedup_cksum_and_write(dda, buf, sz,
301 &stream_cksum, outfd) == -1)
302 goto out;
303 }
304 break;
305 }
306
307 case DRR_END:
308 {
309 /* use the recalculated checksum */
310 ZIO_SET_CHECKSUM(&drre->drr_checksum,
311 stream_cksum.zc_word[0], stream_cksum.zc_word[1],
312 stream_cksum.zc_word[2], stream_cksum.zc_word[3]);
313 if ((write(outfd, drr,
314 sizeof (dmu_replay_record_t))) == -1)
315 goto out;
316 dda->dedup_data_sz += sizeof (dmu_replay_record_t);
317 break;
318 }
319
320 case DRR_OBJECT:
321 {
322 if (dedup_cksum_and_write(dda, drr,
323 sizeof (dmu_replay_record_t),
324 &stream_cksum, outfd) == -1)
325 goto out;
326 if (drro->drr_bonuslen > 0) {
327 (void) ssread(buf,
328 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
329 ofp);
330 if (dedup_cksum_and_write(dda, buf,
331 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
332 &stream_cksum, outfd) == -1)
333 goto out;
334 }
335 break;
336 }
337
338 case DRR_SPILL:
339 {
340 if (dedup_cksum_and_write(dda, drr,
341 sizeof (dmu_replay_record_t),
342 &stream_cksum, outfd) == -1)
343 goto out;
344 (void) ssread(buf, drrs->drr_length, ofp);
345 if (dedup_cksum_and_write(dda, buf, drrs->drr_length,
346 &stream_cksum, outfd) == -1)
347 goto out;
348 break;
349 }
350
351 case DRR_FREEOBJECTS:
352 {
353 if (dedup_cksum_and_write(dda, drr,
354 sizeof (dmu_replay_record_t),
355 &stream_cksum, outfd) == -1)
356 goto out;
357 break;
358 }
359
360 case DRR_WRITE:
361 {
362 dataref_t dataref;
363
364 (void) ssread(buf, drrw->drr_length, ofp);
365
366 /*
367 * Use the existing checksum if it's dedup-capable,
368 * else calculate a SHA256 checksum for it.
369 */
370
371 if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
372 zero_cksum) ||
373 !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
374 SHA256_CTX ctx;
375 zio_cksum_t tmpsha256;
376
377 SHA256Init(&ctx);
378 SHA256Update(&ctx, buf, drrw->drr_length);
379 SHA256Final(&tmpsha256, &ctx);
380 drrw->drr_key.ddk_cksum.zc_word[0] =
381 BE_64(tmpsha256.zc_word[0]);
382 drrw->drr_key.ddk_cksum.zc_word[1] =
383 BE_64(tmpsha256.zc_word[1]);
384 drrw->drr_key.ddk_cksum.zc_word[2] =
385 BE_64(tmpsha256.zc_word[2]);
386 drrw->drr_key.ddk_cksum.zc_word[3] =
387 BE_64(tmpsha256.zc_word[3]);
388 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
389 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
390 }
391
392 dataref.ref_guid = drrw->drr_toguid;
393 dataref.ref_object = drrw->drr_object;
394 dataref.ref_offset = drrw->drr_offset;
395
396 if (ddt_update(dda->dedup_hdl, &ddt,
397 &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
398 &dataref)) {
399 /* block already present in stream */
400 wbr_drrr->drr_object = drrw->drr_object;
401 wbr_drrr->drr_offset = drrw->drr_offset;
402 wbr_drrr->drr_length = drrw->drr_length;
403 wbr_drrr->drr_toguid = drrw->drr_toguid;
404 wbr_drrr->drr_refguid = dataref.ref_guid;
405 wbr_drrr->drr_refobject =
406 dataref.ref_object;
407 wbr_drrr->drr_refoffset =
408 dataref.ref_offset;
409
410 wbr_drrr->drr_checksumtype =
411 drrw->drr_checksumtype;
412 wbr_drrr->drr_checksumflags =
413 drrw->drr_checksumtype;
414 wbr_drrr->drr_key.ddk_cksum =
415 drrw->drr_key.ddk_cksum;
416 wbr_drrr->drr_key.ddk_prop =
417 drrw->drr_key.ddk_prop;
418
419 if (dedup_cksum_and_write(dda, &wbr_drr,
420 sizeof (dmu_replay_record_t), &stream_cksum,
421 outfd) == -1)
422 goto out;
423 } else {
424 /* block not previously seen */
425 if (dedup_cksum_and_write(dda, drr,
426 sizeof (dmu_replay_record_t), &stream_cksum,
427 outfd) == -1)
428 goto out;
429 if (dedup_cksum_and_write(dda, buf,
430 drrw->drr_length,
431 &stream_cksum, outfd) == -1)
432 goto out;
433 }
434 break;
435 }
436
437 case DRR_FREE:
438 {
439 if (dedup_cksum_and_write(dda, drr,
440 sizeof (dmu_replay_record_t),
441 &stream_cksum, outfd) == -1)
442 goto out;
443 break;
444 }
445
446 default:
447 (void) printf("INVALID record type 0x%x\n",
448 drr->drr_type);
449 /* should never happen, so assert */
450 assert(B_FALSE);
451 }
452 }
453 out:
454 umem_cache_destroy(ddt.ddecache);
455 free(ddt.dedup_hash_array);
456 free(buf);
457 (void) fclose(ofp);
458
459 return (NULL);
460 }
461
462 /*
463 * Routines for dealing with the AVL tree of fs-nvlists
464 */
465 typedef struct fsavl_node {
466 avl_node_t fn_node;
467 nvlist_t *fn_nvfs;
468 char *fn_snapname;
469 uint64_t fn_guid;
470 } fsavl_node_t;
471
472 static int
473 fsavl_compare(const void *arg1, const void *arg2)
474 {
475 const fsavl_node_t *fn1 = arg1;
476 const fsavl_node_t *fn2 = arg2;
477
478 if (fn1->fn_guid > fn2->fn_guid)
479 return (+1);
480 else if (fn1->fn_guid < fn2->fn_guid)
481 return (-1);
482 else
483 return (0);
484 }
485
486 /*
487 * Given the GUID of a snapshot, find its containing filesystem and
488 * (optionally) name.
489 */
490 static nvlist_t *
491 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
492 {
493 fsavl_node_t fn_find;
494 fsavl_node_t *fn;
495
496 fn_find.fn_guid = snapguid;
497
498 fn = avl_find(avl, &fn_find, NULL);
499 if (fn) {
500 if (snapname)
501 *snapname = fn->fn_snapname;
502 return (fn->fn_nvfs);
503 }
504 return (NULL);
505 }
506
507 static void
508 fsavl_destroy(avl_tree_t *avl)
509 {
510 fsavl_node_t *fn;
511 void *cookie;
512
513 if (avl == NULL)
514 return;
515
516 cookie = NULL;
517 while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
518 free(fn);
519 avl_destroy(avl);
520 free(avl);
521 }
522
523 /*
524 * Given an nvlist, produce an avl tree of snapshots, ordered by guid
525 */
526 static avl_tree_t *
527 fsavl_create(nvlist_t *fss)
528 {
529 avl_tree_t *fsavl;
530 nvpair_t *fselem = NULL;
531
532 if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
533 return (NULL);
534
535 avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
536 offsetof(fsavl_node_t, fn_node));
537
538 while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
539 nvlist_t *nvfs, *snaps;
540 nvpair_t *snapelem = NULL;
541
542 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
543 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
544
545 while ((snapelem =
546 nvlist_next_nvpair(snaps, snapelem)) != NULL) {
547 fsavl_node_t *fn;
548 uint64_t guid;
549
550 VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
551 if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
552 fsavl_destroy(fsavl);
553 return (NULL);
554 }
555 fn->fn_nvfs = nvfs;
556 fn->fn_snapname = nvpair_name(snapelem);
557 fn->fn_guid = guid;
558
559 /*
560 * Note: if there are multiple snaps with the
561 * same GUID, we ignore all but one.
562 */
563 if (avl_find(fsavl, fn, NULL) == NULL)
564 avl_add(fsavl, fn);
565 else
566 free(fn);
567 }
568 }
569
570 return (fsavl);
571 }
572
573 /*
574 * Routines for dealing with the giant nvlist of fs-nvlists, etc.
575 */
576 typedef struct send_data {
577 uint64_t parent_fromsnap_guid;
578 nvlist_t *parent_snaps;
579 nvlist_t *fss;
580 nvlist_t *snapprops;
581 const char *fromsnap;
582 const char *tosnap;
583 boolean_t recursive;
584
585 /*
586 * The header nvlist is of the following format:
587 * {
588 * "tosnap" -> string
589 * "fromsnap" -> string (if incremental)
590 * "fss" -> {
591 * id -> {
592 *
593 * "name" -> string (full name; for debugging)
594 * "parentfromsnap" -> number (guid of fromsnap in parent)
595 *
596 * "props" -> { name -> value (only if set here) }
597 * "snaps" -> { name (lastname) -> number (guid) }
598 * "snapprops" -> { name (lastname) -> { name -> value } }
599 *
600 * "origin" -> number (guid) (if clone)
601 * "sent" -> boolean (not on-disk)
602 * }
603 * }
604 * }
605 *
606 */
607 } send_data_t;
608
609 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
610
611 static int
612 send_iterate_snap(zfs_handle_t *zhp, void *arg)
613 {
614 send_data_t *sd = arg;
615 uint64_t guid = zhp->zfs_dmustats.dds_guid;
616 char *snapname;
617 nvlist_t *nv;
618
619 snapname = strrchr(zhp->zfs_name, '@')+1;
620
621 VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
622 /*
623 * NB: if there is no fromsnap here (it's a newly created fs in
624 * an incremental replication), we will substitute the tosnap.
625 */
626 if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
627 (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
628 strcmp(snapname, sd->tosnap) == 0)) {
629 sd->parent_fromsnap_guid = guid;
630 }
631
632 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
633 send_iterate_prop(zhp, nv);
634 VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
635 nvlist_free(nv);
636
637 zfs_close(zhp);
638 return (0);
639 }
640
641 static void
642 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
643 {
644 nvpair_t *elem = NULL;
645
646 while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
647 char *propname = nvpair_name(elem);
648 zfs_prop_t prop = zfs_name_to_prop(propname);
649 nvlist_t *propnv;
650
651 if (!zfs_prop_user(propname)) {
652 /*
653 * Realistically, this should never happen. However,
654 * we want the ability to add DSL properties without
655 * needing to make incompatible version changes. We
656 * need to ignore unknown properties to allow older
657 * software to still send datasets containing these
658 * properties, with the unknown properties elided.
659 */
660 if (prop == ZPROP_INVAL)
661 continue;
662
663 if (zfs_prop_readonly(prop))
664 continue;
665 }
666
667 verify(nvpair_value_nvlist(elem, &propnv) == 0);
668 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
669 prop == ZFS_PROP_REFQUOTA ||
670 prop == ZFS_PROP_REFRESERVATION) {
671 char *source;
672 uint64_t value;
673 verify(nvlist_lookup_uint64(propnv,
674 ZPROP_VALUE, &value) == 0);
675 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
676 continue;
677 /*
678 * May have no source before SPA_VERSION_RECVD_PROPS,
679 * but is still modifiable.
680 */
681 if (nvlist_lookup_string(propnv,
682 ZPROP_SOURCE, &source) == 0) {
683 if ((strcmp(source, zhp->zfs_name) != 0) &&
684 (strcmp(source,
685 ZPROP_SOURCE_VAL_RECVD) != 0))
686 continue;
687 }
688 } else {
689 char *source;
690 if (nvlist_lookup_string(propnv,
691 ZPROP_SOURCE, &source) != 0)
692 continue;
693 if ((strcmp(source, zhp->zfs_name) != 0) &&
694 (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
695 continue;
696 }
697
698 if (zfs_prop_user(propname) ||
699 zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
700 char *value;
701 verify(nvlist_lookup_string(propnv,
702 ZPROP_VALUE, &value) == 0);
703 VERIFY(0 == nvlist_add_string(nv, propname, value));
704 } else {
705 uint64_t value;
706 verify(nvlist_lookup_uint64(propnv,
707 ZPROP_VALUE, &value) == 0);
708 VERIFY(0 == nvlist_add_uint64(nv, propname, value));
709 }
710 }
711 }
712
713 /*
714 * recursively generate nvlists describing datasets. See comment
715 * for the data structure send_data_t above for description of contents
716 * of the nvlist.
717 */
718 static int
719 send_iterate_fs(zfs_handle_t *zhp, void *arg)
720 {
721 send_data_t *sd = arg;
722 nvlist_t *nvfs, *nv;
723 int rv = 0;
724 uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
725 uint64_t guid = zhp->zfs_dmustats.dds_guid;
726 char guidstring[64];
727
728 VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
729 VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
730 VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
731 sd->parent_fromsnap_guid));
732
733 if (zhp->zfs_dmustats.dds_origin[0]) {
734 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
735 zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
736 if (origin == NULL)
737 return (-1);
738 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
739 origin->zfs_dmustats.dds_guid));
740 }
741
742 /* iterate over props */
743 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
744 send_iterate_prop(zhp, nv);
745 VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
746 nvlist_free(nv);
747
748 /* iterate over snaps, and set sd->parent_fromsnap_guid */
749 sd->parent_fromsnap_guid = 0;
750 VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
751 VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
752 (void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
753 VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
754 VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
755 nvlist_free(sd->parent_snaps);
756 nvlist_free(sd->snapprops);
757
758 /* add this fs to nvlist */
759 (void) snprintf(guidstring, sizeof (guidstring),
760 "0x%llx", (longlong_t)guid);
761 VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
762 nvlist_free(nvfs);
763
764 /* iterate over children */
765 if (sd->recursive)
766 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
767
768 sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
769
770 zfs_close(zhp);
771 return (rv);
772 }
773
774 static int
775 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
776 const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
777 {
778 zfs_handle_t *zhp;
779 send_data_t sd = { 0 };
780 int error;
781
782 zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
783 if (zhp == NULL)
784 return (EZFS_BADTYPE);
785
786 VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
787 sd.fromsnap = fromsnap;
788 sd.tosnap = tosnap;
789 sd.recursive = recursive;
790
791 if ((error = send_iterate_fs(zhp, &sd)) != 0) {
792 nvlist_free(sd.fss);
793 if (avlp != NULL)
794 *avlp = NULL;
795 *nvlp = NULL;
796 return (error);
797 }
798
799 if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
800 nvlist_free(sd.fss);
801 *nvlp = NULL;
802 return (EZFS_NOMEM);
803 }
804
805 *nvlp = sd.fss;
806 return (0);
807 }
808
809 /*
810 * Routines specific to "zfs send"
811 */
812 typedef struct send_dump_data {
813 /* these are all just the short snapname (the part after the @) */
814 const char *fromsnap;
815 const char *tosnap;
816 char prevsnap[ZFS_MAXNAMELEN];
817 uint64_t prevsnap_obj;
818 boolean_t seenfrom, seento, replicate, doall, fromorigin;
819 boolean_t verbose, dryrun, dedup, parsable, progress;
820 boolean_t sendsize;
821 uint32_t hdr_send_sz;
822 uint64_t send_sz;
823 int outfd;
824 boolean_t err;
825 nvlist_t *fss;
826 avl_tree_t *fsavl;
827 snapfilter_cb_t *filter_cb;
828 void *filter_cb_arg;
829 nvlist_t *debugnv;
830 char holdtag[ZFS_MAXNAMELEN];
831 int cleanup_fd;
832 uint64_t size;
833 } send_dump_data_t;
834
835 static int
836 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
837 boolean_t fromorigin, uint64_t *sizep)
838 {
839 zfs_cmd_t zc = { 0 };
840 libzfs_handle_t *hdl = zhp->zfs_hdl;
841
842 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
843 assert(fromsnap_obj == 0 || !fromorigin);
844
845 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
846 zc.zc_obj = fromorigin;
847 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
848 zc.zc_fromobj = fromsnap_obj;
849 zc.zc_guid = 1; /* estimate flag */
850
851 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
852 char errbuf[1024];
853 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
854 "warning: cannot estimate space for '%s'"), zhp->zfs_name);
855
856 switch (errno) {
857 case EXDEV:
858 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
859 "not an earlier snapshot from the same fs"));
860 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
861
862 case ENOENT:
863 if (zfs_dataset_exists(hdl, zc.zc_name,
864 ZFS_TYPE_SNAPSHOT)) {
865 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
866 "incremental source (@%s) does not exist"),
867 zc.zc_value);
868 }
869 return (zfs_error(hdl, EZFS_NOENT, errbuf));
870
871 case EDQUOT:
872 case EFBIG:
873 case EIO:
874 case ENOLINK:
875 case ENOSPC:
876 case ENOSTR:
877 case ENXIO:
878 case EPIPE:
879 case ERANGE:
880 case EFAULT:
881 case EROFS:
882 zfs_error_aux(hdl, strerror(errno));
883 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
884
885 default:
886 return (zfs_standard_error(hdl, errno, errbuf));
887 }
888 }
889
890 *sizep = zc.zc_objset_type;
891
892 return (0);
893 }
894
895 /*
896 * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
897 * NULL) to the file descriptor specified by outfd.
898 */
899 static int
900 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
901 boolean_t fromorigin, int outfd, nvlist_t *debugnv,
902 boolean_t sendsize, uint64_t *sendcounter)
903 {
904 zfs_cmd_t zc = { 0 };
905 libzfs_handle_t *hdl = zhp->zfs_hdl;
906 nvlist_t *thisdbg;
907
908 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
909 assert(fromsnap_obj == 0 || !fromorigin);
910
911 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
912 zc.zc_cookie = outfd;
913 zc.zc_obj = fromorigin;
914 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
915 zc.zc_fromobj = fromsnap_obj;
916 zc.zc_sendsize = sendsize;
917 zc.zc_sendcounter = 0;
918
919 VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
920 if (fromsnap && fromsnap[0] != '\0') {
921 VERIFY(0 == nvlist_add_string(thisdbg,
922 "fromsnap", fromsnap));
923 }
924
925 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
926 char errbuf[1024];
927 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
928 "warning: cannot send '%s'"), zhp->zfs_name);
929
930 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
931 if (debugnv) {
932 VERIFY(0 == nvlist_add_nvlist(debugnv,
933 zhp->zfs_name, thisdbg));
934 }
935 nvlist_free(thisdbg);
936
937 switch (errno) {
938 case EXDEV:
939 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
940 "not an earlier snapshot from the same fs"));
941 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
942
943 case ENOENT:
944 if (zfs_dataset_exists(hdl, zc.zc_name,
945 ZFS_TYPE_SNAPSHOT)) {
946 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
947 "incremental source (@%s) does not exist"),
948 zc.zc_value);
949 }
950 return (zfs_error(hdl, EZFS_NOENT, errbuf));
951
952 case EDQUOT:
953 case EFBIG:
954 case EIO:
955 case ENOLINK:
956 case ENOSPC:
957 case ENOSTR:
958 case ENXIO:
959 case EPIPE:
960 case ERANGE:
961 case EFAULT:
962 case EROFS:
963 zfs_error_aux(hdl, strerror(errno));
964 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
965
966 default:
967 return (zfs_standard_error(hdl, errno, errbuf));
968 }
969 }
970
971 *sendcounter = (uint64_t)zc.zc_sendcounter;
972 if (debugnv)
973 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
974 nvlist_free(thisdbg);
975
976 return (0);
977 }
978
979 static int
980 hold_for_send(zfs_handle_t *zhp, send_dump_data_t *sdd)
981 {
982 zfs_handle_t *pzhp;
983 int error = 0;
984 char *thissnap;
985
986 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
987
988 if (sdd->dryrun)
989 return (0);
990
991 /*
992 * zfs_send() only opens a cleanup_fd for sends that need it,
993 * e.g. replication and doall.
994 */
995 if (sdd->cleanup_fd == -1)
996 return (0);
997
998 thissnap = strchr(zhp->zfs_name, '@') + 1;
999 *(thissnap - 1) = '\0';
1000 pzhp = zfs_open(zhp->zfs_hdl, zhp->zfs_name, ZFS_TYPE_DATASET);
1001 *(thissnap - 1) = '@';
1002
1003 /*
1004 * It's OK if the parent no longer exists. The send code will
1005 * handle that error.
1006 */
1007 if (pzhp) {
1008 error = zfs_hold(pzhp, thissnap, sdd->holdtag,
1009 B_FALSE, B_TRUE, B_TRUE, sdd->cleanup_fd,
1010 zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID),
1011 zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG));
1012 zfs_close(pzhp);
1013 }
1014
1015 return (error);
1016 }
1017
1018 static void *
1019 send_progress_thread(void *arg)
1020 {
1021 progress_arg_t *pa = arg;
1022
1023 zfs_cmd_t zc = { 0 };
1024 zfs_handle_t *zhp = pa->pa_zhp;
1025 libzfs_handle_t *hdl = zhp->zfs_hdl;
1026 unsigned long long bytes;
1027 char buf[16];
1028
1029 time_t t;
1030 struct tm *tm;
1031
1032 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
1033 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1034
1035 if (!pa->pa_parsable)
1036 (void) fprintf(stderr, "TIME SENT SNAPSHOT\n");
1037
1038 /*
1039 * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1040 */
1041 for (;;) {
1042 (void) sleep(1);
1043
1044 zc.zc_cookie = pa->pa_fd;
1045 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1046 return ((void *)-1);
1047
1048 (void) time(&t);
1049 tm = localtime(&t);
1050 bytes = zc.zc_cookie;
1051
1052 if (pa->pa_parsable) {
1053 (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1054 tm->tm_hour, tm->tm_min, tm->tm_sec,
1055 bytes, zhp->zfs_name);
1056 } else {
1057 zfs_nicenum(bytes, buf, sizeof (buf));
1058 (void) fprintf(stderr, "%02d:%02d:%02d %5s %s\n",
1059 tm->tm_hour, tm->tm_min, tm->tm_sec,
1060 buf, zhp->zfs_name);
1061 }
1062 }
1063 }
1064
1065 static int
1066 dump_snapshot(zfs_handle_t *zhp, void *arg)
1067 {
1068 send_dump_data_t *sdd = arg;
1069 progress_arg_t pa = { 0 };
1070 pthread_t tid;
1071
1072 char *thissnap;
1073 int err;
1074 boolean_t isfromsnap, istosnap, fromorigin;
1075 boolean_t exclude = B_FALSE;
1076
1077 thissnap = strchr(zhp->zfs_name, '@') + 1;
1078 isfromsnap = (sdd->fromsnap != NULL &&
1079 strcmp(sdd->fromsnap, thissnap) == 0);
1080
1081 if (!sdd->seenfrom && isfromsnap) {
1082 err = hold_for_send(zhp, sdd);
1083 if (err == 0) {
1084 sdd->seenfrom = B_TRUE;
1085 (void) strcpy(sdd->prevsnap, thissnap);
1086 sdd->prevsnap_obj = zfs_prop_get_int(zhp,
1087 ZFS_PROP_OBJSETID);
1088 } else if (err == ENOENT) {
1089 err = 0;
1090 }
1091 zfs_close(zhp);
1092 return (err);
1093 }
1094
1095 if (sdd->seento || !sdd->seenfrom) {
1096 zfs_close(zhp);
1097 return (0);
1098 }
1099
1100 istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1101 if (istosnap)
1102 sdd->seento = B_TRUE;
1103
1104 if (!sdd->doall && !isfromsnap && !istosnap) {
1105 if (sdd->replicate) {
1106 char *snapname;
1107 nvlist_t *snapprops;
1108 /*
1109 * Filter out all intermediate snapshots except origin
1110 * snapshots needed to replicate clones.
1111 */
1112 nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1113 zhp->zfs_dmustats.dds_guid, &snapname);
1114
1115 VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1116 "snapprops", &snapprops));
1117 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1118 thissnap, &snapprops));
1119 exclude = !nvlist_exists(snapprops, "is_clone_origin");
1120 } else {
1121 exclude = B_TRUE;
1122 }
1123 }
1124
1125 /*
1126 * If a filter function exists, call it to determine whether
1127 * this snapshot will be sent.
1128 */
1129 if (exclude || (sdd->filter_cb != NULL &&
1130 sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1131 /*
1132 * This snapshot is filtered out. Don't send it, and don't
1133 * set prevsnap_obj, so it will be as if this snapshot didn't
1134 * exist, and the next accepted snapshot will be sent as
1135 * an incremental from the last accepted one, or as the
1136 * first (and full) snapshot in the case of a replication,
1137 * non-incremental send.
1138 */
1139 zfs_close(zhp);
1140 return (0);
1141 }
1142
1143 err = hold_for_send(zhp, sdd);
1144 if (err) {
1145 if (err == ENOENT)
1146 err = 0;
1147 zfs_close(zhp);
1148 return (err);
1149 }
1150
1151 fromorigin = sdd->prevsnap[0] == '\0' &&
1152 (sdd->fromorigin || sdd->replicate);
1153
1154 /* print out to-from and approximate size in verbose mode */
1155 if (sdd->verbose) {
1156 /* print preamble */
1157 if (sdd->parsable) {
1158 if (sdd->prevsnap[0] != '\0') {
1159 (void) fprintf(stderr, "incremental\t%s\t%s",
1160 sdd->prevsnap, zhp->zfs_name);
1161 } else {
1162 (void) fprintf(stderr, "full\t%s",
1163 zhp->zfs_name);
1164 }
1165 } else {
1166 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1167 "send from @%s to %s"),
1168 sdd->prevsnap, zhp->zfs_name);
1169 }
1170
1171 if (sdd->sendsize) {
1172 /*
1173 * we are going to print out the exact stream size info,
1174 * so skip the estimate
1175 */
1176 (void) fprintf(stderr, "\n");
1177 } else {
1178 /*
1179 * provide stream size estimate otherwise
1180 */
1181 uint64_t size;
1182 err = estimate_ioctl(zhp, sdd->prevsnap_obj,
1183 fromorigin, &size);
1184
1185 if (err == 0) {
1186 if (sdd->parsable) {
1187 (void) fprintf(stderr, "\t%llu\n",
1188 (longlong_t)size);
1189 } else {
1190 char buf[16];
1191 zfs_nicenum(size, buf, sizeof (buf));
1192 (void) fprintf(stderr,
1193 dgettext(TEXT_DOMAIN,
1194 " estimated size is %s\n"),
1195 buf);
1196 }
1197 sdd->size += size;
1198 } else {
1199 /* could not estimate */
1200 (void) fprintf(stderr, "\n");
1201 }
1202 }
1203 }
1204
1205 if (!sdd->dryrun) {
1206 uint64_t sendcounter = 0;
1207 boolean_t track_progress = (sdd->progress && !sdd->sendsize);
1208 boolean_t sendsize = B_FALSE;
1209 /*
1210 * If progress reporting is requested, spawn a new thread to
1211 * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1212 */
1213 if (track_progress) {
1214 pa.pa_zhp = zhp;
1215 pa.pa_fd = sdd->outfd;
1216 pa.pa_parsable = sdd->parsable;
1217
1218 if (err = pthread_create(&tid, NULL,
1219 send_progress_thread, &pa)) {
1220 zfs_close(zhp);
1221 return (err);
1222 }
1223 }
1224
1225
1226 /*
1227 * We need to reset the sendsize flag being sent to
1228 * kernel if sdd->dedup is set. With dedup, the file
1229 * descriptor sent to kernel is one end of the pipe,
1230 * and we would want the data back in the pipe for
1231 * cksummer() to calculate the exact size of the dedup-ed
1232 * stream. So reset the sendsize flag such that
1233 * kernel writes to the pipe.
1234 */
1235
1236 sendsize = sdd->dedup ? B_FALSE : sdd->sendsize;
1237
1238 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1239 fromorigin, sdd->outfd, sdd->debugnv,
1240 sendsize, &sendcounter);
1241
1242 sdd->send_sz += sendcounter;
1243
1244 if (track_progress) {
1245 (void) pthread_cancel(tid);
1246 (void) pthread_join(tid, NULL);
1247 }
1248 }
1249
1250 (void) strcpy(sdd->prevsnap, thissnap);
1251 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1252 zfs_close(zhp);
1253 return (err);
1254 }
1255
1256 static int
1257 dump_filesystem(zfs_handle_t *zhp, void *arg)
1258 {
1259 int rv = 0;
1260 send_dump_data_t *sdd = arg;
1261 boolean_t missingfrom = B_FALSE;
1262 zfs_cmd_t zc = { 0 };
1263
1264 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1265 zhp->zfs_name, sdd->tosnap);
1266 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1267 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1268 "WARNING: could not send %s@%s: does not exist\n"),
1269 zhp->zfs_name, sdd->tosnap);
1270 sdd->err = B_TRUE;
1271 return (0);
1272 }
1273
1274 if (sdd->replicate && sdd->fromsnap) {
1275 /*
1276 * If this fs does not have fromsnap, and we're doing
1277 * recursive, we need to send a full stream from the
1278 * beginning (or an incremental from the origin if this
1279 * is a clone). If we're doing non-recursive, then let
1280 * them get the error.
1281 */
1282 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1283 zhp->zfs_name, sdd->fromsnap);
1284 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1285 ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1286 missingfrom = B_TRUE;
1287 }
1288 }
1289
1290 sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1291 sdd->prevsnap_obj = 0;
1292 if (sdd->fromsnap == NULL || missingfrom)
1293 sdd->seenfrom = B_TRUE;
1294
1295 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1296 if (!sdd->seenfrom) {
1297 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1298 "WARNING: could not send %s@%s:\n"
1299 "incremental source (%s@%s) does not exist\n"),
1300 zhp->zfs_name, sdd->tosnap,
1301 zhp->zfs_name, sdd->fromsnap);
1302 sdd->err = B_TRUE;
1303 } else if (!sdd->seento) {
1304 if (sdd->fromsnap) {
1305 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1306 "WARNING: could not send %s@%s:\n"
1307 "incremental source (%s@%s) "
1308 "is not earlier than it\n"),
1309 zhp->zfs_name, sdd->tosnap,
1310 zhp->zfs_name, sdd->fromsnap);
1311 } else {
1312 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1313 "WARNING: "
1314 "could not send %s@%s: does not exist\n"),
1315 zhp->zfs_name, sdd->tosnap);
1316 }
1317 sdd->err = B_TRUE;
1318 }
1319
1320 return (rv);
1321 }
1322
1323 static int
1324 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1325 {
1326 send_dump_data_t *sdd = arg;
1327 nvpair_t *fspair;
1328 boolean_t needagain, progress;
1329
1330 if (!sdd->replicate)
1331 return (dump_filesystem(rzhp, sdd));
1332
1333 /* Mark the clone origin snapshots. */
1334 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1335 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1336 nvlist_t *nvfs;
1337 uint64_t origin_guid = 0;
1338
1339 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1340 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1341 if (origin_guid != 0) {
1342 char *snapname;
1343 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1344 origin_guid, &snapname);
1345 if (origin_nv != NULL) {
1346 nvlist_t *snapprops;
1347 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1348 "snapprops", &snapprops));
1349 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1350 snapname, &snapprops));
1351 VERIFY(0 == nvlist_add_boolean(
1352 snapprops, "is_clone_origin"));
1353 }
1354 }
1355 }
1356 again:
1357 needagain = progress = B_FALSE;
1358 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1359 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1360 nvlist_t *fslist, *parent_nv;
1361 char *fsname;
1362 zfs_handle_t *zhp;
1363 int err;
1364 uint64_t origin_guid = 0;
1365 uint64_t parent_guid = 0;
1366
1367 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1368 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1369 continue;
1370
1371 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1372 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1373 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1374 &parent_guid);
1375
1376 if (parent_guid != 0) {
1377 parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1378 if (!nvlist_exists(parent_nv, "sent")) {
1379 /* parent has not been sent; skip this one */
1380 needagain = B_TRUE;
1381 continue;
1382 }
1383 }
1384
1385 if (origin_guid != 0) {
1386 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1387 origin_guid, NULL);
1388 if (origin_nv != NULL &&
1389 !nvlist_exists(origin_nv, "sent")) {
1390 /*
1391 * origin has not been sent yet;
1392 * skip this clone.
1393 */
1394 needagain = B_TRUE;
1395 continue;
1396 }
1397 }
1398
1399 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1400 if (zhp == NULL)
1401 return (-1);
1402 err = dump_filesystem(zhp, sdd);
1403 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1404 progress = B_TRUE;
1405 zfs_close(zhp);
1406 if (err)
1407 return (err);
1408 }
1409 if (needagain) {
1410 assert(progress);
1411 goto again;
1412 }
1413
1414 /* clean out the sent flags in case we reuse this fss */
1415 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1416 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1417 nvlist_t *fslist;
1418
1419 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1420 (void) nvlist_remove_all(fslist, "sent");
1421 }
1422
1423 return (0);
1424 }
1425
1426 /*
1427 * Generate a send stream for the dataset identified by the argument zhp.
1428 *
1429 * The content of the send stream is the snapshot identified by
1430 * 'tosnap'. Incremental streams are requested in two ways:
1431 * - from the snapshot identified by "fromsnap" (if non-null) or
1432 * - from the origin of the dataset identified by zhp, which must
1433 * be a clone. In this case, "fromsnap" is null and "fromorigin"
1434 * is TRUE.
1435 *
1436 * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1437 * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1438 * if "replicate" is set. If "doall" is set, dump all the intermediate
1439 * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1440 * case too. If "props" is set, send properties.
1441 */
1442 int
1443 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1444 sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1445 void *cb_arg, nvlist_t **debugnvp)
1446 {
1447 char errbuf[1024];
1448 send_dump_data_t sdd = { 0 };
1449 int err = 0;
1450 nvlist_t *fss = NULL;
1451 avl_tree_t *fsavl = NULL;
1452 static uint64_t holdseq;
1453 int spa_version;
1454 pthread_t tid;
1455 int pipefd[2];
1456 dedup_arg_t dda = { 0 };
1457 int featureflags = 0;
1458
1459 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1460 "cannot send '%s'"), zhp->zfs_name);
1461
1462 if (fromsnap && fromsnap[0] == '\0') {
1463 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1464 "zero-length incremental source"));
1465 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1466 }
1467
1468 if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1469 uint64_t version;
1470 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1471 if (version >= ZPL_VERSION_SA) {
1472 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1473 }
1474 }
1475
1476 if (flags->dedup && !flags->dryrun) {
1477 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1478 DMU_BACKUP_FEATURE_DEDUPPROPS);
1479 if (err = pipe(pipefd)) {
1480 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1481 return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1482 errbuf));
1483 }
1484 dda.outputfd = outfd;
1485 dda.inputfd = pipefd[1];
1486 dda.dedup_hdl = zhp->zfs_hdl;
1487 dda.sendsize = flags->sendsize;
1488 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1489 (void) close(pipefd[0]);
1490 (void) close(pipefd[1]);
1491 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1492 return (zfs_error(zhp->zfs_hdl,
1493 EZFS_THREADCREATEFAILED, errbuf));
1494 }
1495 }
1496
1497 if (flags->replicate || flags->doall || flags->props) {
1498 dmu_replay_record_t drr = { 0 };
1499 char *packbuf = NULL;
1500 size_t buflen = 0;
1501 zio_cksum_t zc = { 0 };
1502
1503 if (flags->replicate || flags->props) {
1504 nvlist_t *hdrnv;
1505
1506 VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1507 if (fromsnap) {
1508 VERIFY(0 == nvlist_add_string(hdrnv,
1509 "fromsnap", fromsnap));
1510 }
1511 VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1512 if (!flags->replicate) {
1513 VERIFY(0 == nvlist_add_boolean(hdrnv,
1514 "not_recursive"));
1515 }
1516
1517 err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1518 fromsnap, tosnap, flags->replicate, &fss, &fsavl);
1519 if (err)
1520 goto err_out;
1521 VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1522 err = nvlist_pack(hdrnv, &packbuf, &buflen,
1523 NV_ENCODE_XDR, 0);
1524 if (debugnvp)
1525 *debugnvp = hdrnv;
1526 else
1527 nvlist_free(hdrnv);
1528 if (err) {
1529 fsavl_destroy(fsavl);
1530 nvlist_free(fss);
1531 goto stderr_out;
1532 }
1533 }
1534
1535 if (!flags->dryrun) {
1536 /* write first begin record */
1537 drr.drr_type = DRR_BEGIN;
1538 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1539 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1540 drr_versioninfo, DMU_COMPOUNDSTREAM);
1541 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1542 drr_versioninfo, featureflags);
1543 (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1544 sizeof (drr.drr_u.drr_begin.drr_toname),
1545 "%s@%s", zhp->zfs_name, tosnap);
1546 drr.drr_payloadlen = buflen;
1547 err = cksum_and_write(&drr, sizeof (drr), &zc, outfd);
1548 sdd.hdr_send_sz += sizeof (drr);
1549
1550 /* write header nvlist */
1551 if (err != -1 && packbuf != NULL) {
1552 err = cksum_and_write(packbuf, buflen, &zc,
1553 outfd);
1554 sdd.hdr_send_sz += buflen;
1555 }
1556 free(packbuf);
1557 if (err == -1) {
1558 fsavl_destroy(fsavl);
1559 nvlist_free(fss);
1560 err = errno;
1561 goto stderr_out;
1562 }
1563
1564 /* write end record */
1565 bzero(&drr, sizeof (drr));
1566 drr.drr_type = DRR_END;
1567 drr.drr_u.drr_end.drr_checksum = zc;
1568 err = write(outfd, &drr, sizeof (drr));
1569 sdd.hdr_send_sz += sizeof (drr);
1570 if (err == -1) {
1571 fsavl_destroy(fsavl);
1572 nvlist_free(fss);
1573 err = errno;
1574 goto stderr_out;
1575 }
1576
1577 err = 0;
1578 }
1579 }
1580
1581 /* dump each stream */
1582 sdd.fromsnap = fromsnap;
1583 sdd.tosnap = tosnap;
1584 if (flags->dedup)
1585 sdd.outfd = pipefd[0];
1586 else
1587 sdd.outfd = outfd;
1588 sdd.replicate = flags->replicate;
1589 sdd.doall = flags->doall;
1590 sdd.fromorigin = flags->fromorigin;
1591 sdd.fss = fss;
1592 sdd.fsavl = fsavl;
1593 sdd.verbose = flags->verbose;
1594 sdd.dedup = flags->dedup;
1595 sdd.sendsize = flags->sendsize;
1596 sdd.parsable = flags->parsable;
1597 sdd.progress = flags->progress;
1598 sdd.dryrun = flags->dryrun;
1599 sdd.filter_cb = filter_func;
1600 sdd.filter_cb_arg = cb_arg;
1601 if (debugnvp)
1602 sdd.debugnv = *debugnvp;
1603
1604 /*
1605 * Some flags require that we place user holds on the datasets that are
1606 * being sent so they don't get destroyed during the send. We can skip
1607 * this step if the pool is imported read-only since the datasets cannot
1608 * be destroyed.
1609 */
1610 if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1611 ZPOOL_PROP_READONLY, NULL) &&
1612 zfs_spa_version(zhp, &spa_version) == 0 &&
1613 spa_version >= SPA_VERSION_USERREFS &&
1614 (flags->doall || flags->replicate)) {
1615 ++holdseq;
1616 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1617 ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1618 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1619 if (sdd.cleanup_fd < 0) {
1620 err = errno;
1621 goto stderr_out;
1622 }
1623 } else {
1624 sdd.cleanup_fd = -1;
1625 }
1626 if (flags->verbose && !flags->sendsize) {
1627 /*
1628 * Do a verbose no-op dry run to get all the verbose output
1629 * before generating any data. Then do a non-verbose real
1630 * run to generate the streams.
1631 */
1632 sdd.dryrun = B_TRUE;
1633 err = dump_filesystems(zhp, &sdd);
1634 sdd.dryrun = flags->dryrun;
1635 sdd.verbose = B_FALSE;
1636 if (flags->parsable) {
1637 (void) fprintf(stderr, "size\t%llu\n",
1638 (longlong_t)sdd.size);
1639 } else {
1640 char buf[16];
1641 zfs_nicenum(sdd.size, buf, sizeof (buf));
1642 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1643 "total estimated size is %s\n"), buf);
1644 }
1645 }
1646 err = dump_filesystems(zhp, &sdd);
1647 fsavl_destroy(fsavl);
1648 nvlist_free(fss);
1649
1650 if (flags->dedup) {
1651 (void) close(pipefd[0]);
1652 (void) pthread_join(tid, NULL);
1653 sdd.send_sz = dda.dedup_data_sz;
1654 }
1655
1656 if (sdd.cleanup_fd != -1) {
1657 VERIFY(0 == close(sdd.cleanup_fd));
1658 sdd.cleanup_fd = -1;
1659 }
1660
1661 if (!flags->dryrun && (flags->replicate || flags->doall ||
1662 flags->props)) {
1663 /*
1664 * write final end record. NB: want to do this even if
1665 * there was some error, because it might not be totally
1666 * failed.
1667 */
1668 dmu_replay_record_t drr = { 0 };
1669 drr.drr_type = DRR_END;
1670 if (write(outfd, &drr, sizeof (drr)) == -1) {
1671 return (zfs_standard_error(zhp->zfs_hdl,
1672 errno, errbuf));
1673 }
1674 sdd.hdr_send_sz += sizeof (drr);
1675 }
1676
1677 if (flags->sendsize) {
1678 if (flags->verbose) {
1679 fprintf(stderr, "Send stream header size (bytes): "
1680 "%u\n", sdd.hdr_send_sz);
1681 fprintf(stderr, "Send stream data size (bytes): "
1682 "%llu\n", sdd.send_sz);
1683 fprintf(stderr, "Total send stream size (bytes): "
1684 "%llu\n", sdd.send_sz + (uint64_t)sdd.hdr_send_sz);
1685 } else {
1686 fprintf(stderr, "Total send stream size (bytes): "
1687 "%llu\n", sdd.send_sz + (uint64_t)sdd.hdr_send_sz);
1688 }
1689 }
1690
1691 return (err || sdd.err);
1692
1693 stderr_out:
1694 err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1695 err_out:
1696 if (sdd.cleanup_fd != -1)
1697 VERIFY(0 == close(sdd.cleanup_fd));
1698 if (flags->dedup) {
1699 (void) pthread_cancel(tid);
1700 (void) pthread_join(tid, NULL);
1701 (void) close(pipefd[0]);
1702 }
1703 return (err);
1704 }
1705
1706 /*
1707 * Routines specific to "zfs recv"
1708 */
1709
1710 static int
1711 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1712 boolean_t byteswap, zio_cksum_t *zc)
1713 {
1714 char *cp = buf;
1715 int rv;
1716 int len = ilen;
1717
1718 do {
1719 rv = read(fd, cp, len);
1720 cp += rv;
1721 len -= rv;
1722 } while (rv > 0);
1723
1724 if (rv < 0 || len != 0) {
1725 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1726 "failed to read from stream"));
1727 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1728 "cannot receive")));
1729 }
1730
1731 if (zc) {
1732 if (byteswap)
1733 fletcher_4_incremental_byteswap(buf, ilen, zc);
1734 else
1735 fletcher_4_incremental_native(buf, ilen, zc);
1736 }
1737 return (0);
1738 }
1739
1740 static int
1741 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1742 boolean_t byteswap, zio_cksum_t *zc)
1743 {
1744 char *buf;
1745 int err;
1746
1747 buf = zfs_alloc(hdl, len);
1748 if (buf == NULL)
1749 return (ENOMEM);
1750
1751 err = recv_read(hdl, fd, buf, len, byteswap, zc);
1752 if (err != 0) {
1753 free(buf);
1754 return (err);
1755 }
1756
1757 err = nvlist_unpack(buf, len, nvp, 0);
1758 free(buf);
1759 if (err != 0) {
1760 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1761 "stream (malformed nvlist)"));
1762 return (EINVAL);
1763 }
1764 return (0);
1765 }
1766
1767 static int
1768 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1769 int baselen, char *newname, recvflags_t *flags)
1770 {
1771 static int seq;
1772 zfs_cmd_t zc = { 0 };
1773 int err;
1774 prop_changelist_t *clp;
1775 zfs_handle_t *zhp;
1776
1777 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1778 if (zhp == NULL)
1779 return (-1);
1780 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1781 flags->force ? MS_FORCE : 0);
1782 zfs_close(zhp);
1783 if (clp == NULL)
1784 return (-1);
1785 err = changelist_prefix(clp);
1786 if (err)
1787 return (err);
1788
1789 zc.zc_objset_type = DMU_OST_ZFS;
1790 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1791
1792 if (tryname) {
1793 (void) strcpy(newname, tryname);
1794
1795 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1796
1797 if (flags->verbose) {
1798 (void) printf("attempting rename %s to %s\n",
1799 zc.zc_name, zc.zc_value);
1800 }
1801 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1802 if (err == 0)
1803 changelist_rename(clp, name, tryname);
1804 } else {
1805 err = ENOENT;
1806 }
1807
1808 if (err != 0 && strncmp(name+baselen, "recv-", 5) != 0) {
1809 seq++;
1810
1811 (void) strncpy(newname, name, baselen);
1812 (void) snprintf(newname+baselen, ZFS_MAXNAMELEN-baselen,
1813 "recv-%u-%u", getpid(), seq);
1814 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1815
1816 if (flags->verbose) {
1817 (void) printf("failed - trying rename %s to %s\n",
1818 zc.zc_name, zc.zc_value);
1819 }
1820 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1821 if (err == 0)
1822 changelist_rename(clp, name, newname);
1823 if (err && flags->verbose) {
1824 (void) printf("failed (%u) - "
1825 "will try again on next pass\n", errno);
1826 }
1827 err = EAGAIN;
1828 } else if (flags->verbose) {
1829 if (err == 0)
1830 (void) printf("success\n");
1831 else
1832 (void) printf("failed (%u)\n", errno);
1833 }
1834
1835 (void) changelist_postfix(clp);
1836 changelist_free(clp);
1837
1838 return (err);
1839 }
1840
1841 static int
1842 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1843 char *newname, recvflags_t *flags)
1844 {
1845 zfs_cmd_t zc = { 0 };
1846 int err = 0;
1847 prop_changelist_t *clp;
1848 zfs_handle_t *zhp;
1849 boolean_t defer = B_FALSE;
1850 int spa_version;
1851
1852 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1853 if (zhp == NULL)
1854 return (-1);
1855 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1856 flags->force ? MS_FORCE : 0);
1857 if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1858 zfs_spa_version(zhp, &spa_version) == 0 &&
1859 spa_version >= SPA_VERSION_USERREFS)
1860 defer = B_TRUE;
1861 zfs_close(zhp);
1862 if (clp == NULL)
1863 return (-1);
1864 err = changelist_prefix(clp);
1865 if (err)
1866 return (err);
1867
1868 zc.zc_objset_type = DMU_OST_ZFS;
1869 zc.zc_defer_destroy = defer;
1870 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1871
1872 if (flags->verbose)
1873 (void) printf("attempting destroy %s\n", zc.zc_name);
1874 err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1875 if (err == 0) {
1876 if (flags->verbose)
1877 (void) printf("success\n");
1878 changelist_remove(clp, zc.zc_name);
1879 }
1880
1881 (void) changelist_postfix(clp);
1882 changelist_free(clp);
1883
1884 /*
1885 * Deferred destroy might destroy the snapshot or only mark it to be
1886 * destroyed later, and it returns success in either case.
1887 */
1888 if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1889 ZFS_TYPE_SNAPSHOT))) {
1890 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1891 }
1892
1893 return (err);
1894 }
1895
1896 typedef struct guid_to_name_data {
1897 uint64_t guid;
1898 char *name;
1899 char *skip;
1900 } guid_to_name_data_t;
1901
1902 static int
1903 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1904 {
1905 guid_to_name_data_t *gtnd = arg;
1906 int err;
1907
1908 if (gtnd->skip != NULL &&
1909 strcmp(zhp->zfs_name, gtnd->skip) == 0) {
1910 return (0);
1911 }
1912
1913 if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1914 (void) strcpy(gtnd->name, zhp->zfs_name);
1915 zfs_close(zhp);
1916 return (EEXIST);
1917 }
1918
1919 err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1920 zfs_close(zhp);
1921 return (err);
1922 }
1923
1924 /*
1925 * Attempt to find the local dataset associated with this guid. In the case of
1926 * multiple matches, we attempt to find the "best" match by searching
1927 * progressively larger portions of the hierarchy. This allows one to send a
1928 * tree of datasets individually and guarantee that we will find the source
1929 * guid within that hierarchy, even if there are multiple matches elsewhere.
1930 */
1931 static int
1932 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1933 char *name)
1934 {
1935 /* exhaustive search all local snapshots */
1936 char pname[ZFS_MAXNAMELEN];
1937 guid_to_name_data_t gtnd;
1938 int err = 0;
1939 zfs_handle_t *zhp;
1940 char *cp;
1941
1942 gtnd.guid = guid;
1943 gtnd.name = name;
1944 gtnd.skip = NULL;
1945
1946 (void) strlcpy(pname, parent, sizeof (pname));
1947
1948 /*
1949 * Search progressively larger portions of the hierarchy. This will
1950 * select the "most local" version of the origin snapshot in the case
1951 * that there are multiple matching snapshots in the system.
1952 */
1953 while ((cp = strrchr(pname, '/')) != NULL) {
1954
1955 /* Chop off the last component and open the parent */
1956 *cp = '\0';
1957 zhp = make_dataset_handle(hdl, pname);
1958
1959 if (zhp == NULL)
1960 continue;
1961
1962 err = zfs_iter_children(zhp, guid_to_name_cb, >nd);
1963 zfs_close(zhp);
1964 if (err == EEXIST)
1965 return (0);
1966
1967 /*
1968 * Remember the dataset that we already searched, so we
1969 * skip it next time through.
1970 */
1971 gtnd.skip = pname;
1972 }
1973
1974 return (ENOENT);
1975 }
1976
1977 /*
1978 * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
1979 * guid1 is after guid2.
1980 */
1981 static int
1982 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1983 uint64_t guid1, uint64_t guid2)
1984 {
1985 nvlist_t *nvfs;
1986 char *fsname, *snapname;
1987 char buf[ZFS_MAXNAMELEN];
1988 int rv;
1989 zfs_handle_t *guid1hdl, *guid2hdl;
1990 uint64_t create1, create2;
1991
1992 if (guid2 == 0)
1993 return (0);
1994 if (guid1 == 0)
1995 return (1);
1996
1997 nvfs = fsavl_find(avl, guid1, &snapname);
1998 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1999 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2000 guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2001 if (guid1hdl == NULL)
2002 return (-1);
2003
2004 nvfs = fsavl_find(avl, guid2, &snapname);
2005 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2006 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2007 guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2008 if (guid2hdl == NULL) {
2009 zfs_close(guid1hdl);
2010 return (-1);
2011 }
2012
2013 create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
2014 create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
2015
2016 if (create1 < create2)
2017 rv = -1;
2018 else if (create1 > create2)
2019 rv = +1;
2020 else
2021 rv = 0;
2022
2023 zfs_close(guid1hdl);
2024 zfs_close(guid2hdl);
2025
2026 return (rv);
2027 }
2028
2029 static int
2030 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2031 recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2032 nvlist_t *renamed)
2033 {
2034 nvlist_t *local_nv;
2035 avl_tree_t *local_avl;
2036 nvpair_t *fselem, *nextfselem;
2037 char *fromsnap;
2038 char newname[ZFS_MAXNAMELEN];
2039 int error;
2040 boolean_t needagain, progress, recursive;
2041 char *s1, *s2;
2042
2043 VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2044
2045 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2046 ENOENT);
2047
2048 if (flags->dryrun)
2049 return (0);
2050
2051 again:
2052 needagain = progress = B_FALSE;
2053
2054 if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2055 recursive, &local_nv, &local_avl)) != 0)
2056 return (error);
2057
2058 /*
2059 * Process deletes and renames
2060 */
2061 for (fselem = nvlist_next_nvpair(local_nv, NULL);
2062 fselem; fselem = nextfselem) {
2063 nvlist_t *nvfs, *snaps;
2064 nvlist_t *stream_nvfs = NULL;
2065 nvpair_t *snapelem, *nextsnapelem;
2066 uint64_t fromguid = 0;
2067 uint64_t originguid = 0;
2068 uint64_t stream_originguid = 0;
2069 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2070 char *fsname, *stream_fsname;
2071
2072 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2073
2074 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2075 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2076 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2077 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2078 &parent_fromsnap_guid));
2079 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2080
2081 /*
2082 * First find the stream's fs, so we can check for
2083 * a different origin (due to "zfs promote")
2084 */
2085 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2086 snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2087 uint64_t thisguid;
2088
2089 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2090 stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2091
2092 if (stream_nvfs != NULL)
2093 break;
2094 }
2095
2096 /* check for promote */
2097 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2098 &stream_originguid);
2099 if (stream_nvfs && originguid != stream_originguid) {
2100 switch (created_before(hdl, local_avl,
2101 stream_originguid, originguid)) {
2102 case 1: {
2103 /* promote it! */
2104 zfs_cmd_t zc = { 0 };
2105 nvlist_t *origin_nvfs;
2106 char *origin_fsname;
2107
2108 if (flags->verbose)
2109 (void) printf("promoting %s\n", fsname);
2110
2111 origin_nvfs = fsavl_find(local_avl, originguid,
2112 NULL);
2113 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2114 "name", &origin_fsname));
2115 (void) strlcpy(zc.zc_value, origin_fsname,
2116 sizeof (zc.zc_value));
2117 (void) strlcpy(zc.zc_name, fsname,
2118 sizeof (zc.zc_name));
2119 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2120 if (error == 0)
2121 progress = B_TRUE;
2122 break;
2123 }
2124 default:
2125 break;
2126 case -1:
2127 fsavl_destroy(local_avl);
2128 nvlist_free(local_nv);
2129 return (-1);
2130 }
2131 /*
2132 * We had/have the wrong origin, therefore our
2133 * list of snapshots is wrong. Need to handle
2134 * them on the next pass.
2135 */
2136 needagain = B_TRUE;
2137 continue;
2138 }
2139
2140 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2141 snapelem; snapelem = nextsnapelem) {
2142 uint64_t thisguid;
2143 char *stream_snapname;
2144 nvlist_t *found, *props;
2145
2146 nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2147
2148 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2149 found = fsavl_find(stream_avl, thisguid,
2150 &stream_snapname);
2151
2152 /* check for delete */
2153 if (found == NULL) {
2154 char name[ZFS_MAXNAMELEN];
2155
2156 if (!flags->force)
2157 continue;
2158
2159 (void) snprintf(name, sizeof (name), "%s@%s",
2160 fsname, nvpair_name(snapelem));
2161
2162 error = recv_destroy(hdl, name,
2163 strlen(fsname)+1, newname, flags);
2164 if (error)
2165 needagain = B_TRUE;
2166 else
2167 progress = B_TRUE;
2168 continue;
2169 }
2170
2171 stream_nvfs = found;
2172
2173 if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2174 &props) && 0 == nvlist_lookup_nvlist(props,
2175 stream_snapname, &props)) {
2176 zfs_cmd_t zc = { 0 };
2177
2178 zc.zc_cookie = B_TRUE; /* received */
2179 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2180 "%s@%s", fsname, nvpair_name(snapelem));
2181 if (zcmd_write_src_nvlist(hdl, &zc,
2182 props) == 0) {
2183 (void) zfs_ioctl(hdl,
2184 ZFS_IOC_SET_PROP, &zc);
2185 zcmd_free_nvlists(&zc);
2186 }
2187 }
2188
2189 /* check for different snapname */
2190 if (strcmp(nvpair_name(snapelem),
2191 stream_snapname) != 0) {
2192 char name[ZFS_MAXNAMELEN];
2193 char tryname[ZFS_MAXNAMELEN];
2194
2195 (void) snprintf(name, sizeof (name), "%s@%s",
2196 fsname, nvpair_name(snapelem));
2197 (void) snprintf(tryname, sizeof (name), "%s@%s",
2198 fsname, stream_snapname);
2199
2200 error = recv_rename(hdl, name, tryname,
2201 strlen(fsname)+1, newname, flags);
2202 if (error)
2203 needagain = B_TRUE;
2204 else
2205 progress = B_TRUE;
2206 }
2207
2208 if (strcmp(stream_snapname, fromsnap) == 0)
2209 fromguid = thisguid;
2210 }
2211
2212 /* check for delete */
2213 if (stream_nvfs == NULL) {
2214 if (!flags->force)
2215 continue;
2216
2217 error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2218 newname, flags);
2219 if (error)
2220 needagain = B_TRUE;
2221 else
2222 progress = B_TRUE;
2223 continue;
2224 }
2225
2226 if (fromguid == 0) {
2227 if (flags->verbose) {
2228 (void) printf("local fs %s does not have "
2229 "fromsnap (%s in stream); must have "
2230 "been deleted locally; ignoring\n",
2231 fsname, fromsnap);
2232 }
2233 continue;
2234 }
2235
2236 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2237 "name", &stream_fsname));
2238 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2239 "parentfromsnap", &stream_parent_fromsnap_guid));
2240
2241 s1 = strrchr(fsname, '/');
2242 s2 = strrchr(stream_fsname, '/');
2243
2244 /*
2245 * Check for rename. If the exact receive path is specified, it
2246 * does not count as a rename, but we still need to check the
2247 * datasets beneath it.
2248 */
2249 if ((stream_parent_fromsnap_guid != 0 &&
2250 parent_fromsnap_guid != 0 &&
2251 stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2252 ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2253 (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2254 nvlist_t *parent;
2255 char tryname[ZFS_MAXNAMELEN];
2256
2257 parent = fsavl_find(local_avl,
2258 stream_parent_fromsnap_guid, NULL);
2259 /*
2260 * NB: parent might not be found if we used the
2261 * tosnap for stream_parent_fromsnap_guid,
2262 * because the parent is a newly-created fs;
2263 * we'll be able to rename it after we recv the
2264 * new fs.
2265 */
2266 if (parent != NULL) {
2267 char *pname;
2268
2269 VERIFY(0 == nvlist_lookup_string(parent, "name",
2270 &pname));
2271 (void) snprintf(tryname, sizeof (tryname),
2272 "%s%s", pname, strrchr(stream_fsname, '/'));
2273 } else {
2274 tryname[0] = '\0';
2275 if (flags->verbose) {
2276 (void) printf("local fs %s new parent "
2277 "not found\n", fsname);
2278 }
2279 }
2280
2281 newname[0] = '\0';
2282
2283 error = recv_rename(hdl, fsname, tryname,
2284 strlen(tofs)+1, newname, flags);
2285
2286 if (renamed != NULL && newname[0] != '\0') {
2287 VERIFY(0 == nvlist_add_boolean(renamed,
2288 newname));
2289 }
2290
2291 if (error)
2292 needagain = B_TRUE;
2293 else
2294 progress = B_TRUE;
2295 }
2296 }
2297
2298 fsavl_destroy(local_avl);
2299 nvlist_free(local_nv);
2300
2301 if (needagain && progress) {
2302 /* do another pass to fix up temporary names */
2303 if (flags->verbose)
2304 (void) printf("another pass:\n");
2305 goto again;
2306 }
2307
2308 return (needagain);
2309 }
2310
2311 static int
2312 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2313 recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2314 char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2315 {
2316 nvlist_t *stream_nv = NULL;
2317 avl_tree_t *stream_avl = NULL;
2318 char *fromsnap = NULL;
2319 char *cp;
2320 char tofs[ZFS_MAXNAMELEN];
2321 char sendfs[ZFS_MAXNAMELEN];
2322 char errbuf[1024];
2323 dmu_replay_record_t drre;
2324 int error;
2325 boolean_t anyerr = B_FALSE;
2326 boolean_t softerr = B_FALSE;
2327 boolean_t recursive;
2328
2329 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2330 "cannot receive"));
2331
2332 assert(drr->drr_type == DRR_BEGIN);
2333 assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2334 assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2335 DMU_COMPOUNDSTREAM);
2336
2337 /*
2338 * Read in the nvlist from the stream.
2339 */
2340 if (drr->drr_payloadlen != 0) {
2341 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2342 &stream_nv, flags->byteswap, zc);
2343 if (error) {
2344 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2345 goto out;
2346 }
2347 }
2348
2349 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2350 ENOENT);
2351
2352 if (recursive && strchr(destname, '@')) {
2353 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2354 "cannot specify snapshot name for multi-snapshot stream"));
2355 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2356 goto out;
2357 }
2358
2359 /*
2360 * Read in the end record and verify checksum.
2361 */
2362 if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2363 flags->byteswap, NULL)))
2364 goto out;
2365 if (flags->byteswap) {
2366 drre.drr_type = BSWAP_32(drre.drr_type);
2367 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2368 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2369 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2370 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2371 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2372 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2373 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2374 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2375 }
2376 if (drre.drr_type != DRR_END) {
2377 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2378 goto out;
2379 }
2380 if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2381 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2382 "incorrect header checksum"));
2383 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2384 goto out;
2385 }
2386
2387 (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2388
2389 if (drr->drr_payloadlen != 0) {
2390 nvlist_t *stream_fss;
2391
2392 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2393 &stream_fss));
2394 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2395 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2396 "couldn't allocate avl tree"));
2397 error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2398 goto out;
2399 }
2400
2401 if (fromsnap != NULL) {
2402 nvlist_t *renamed = NULL;
2403 nvpair_t *pair = NULL;
2404
2405 (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2406 if (flags->isprefix) {
2407 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2408 int i;
2409
2410 if (flags->istail) {
2411 cp = strrchr(drrb->drr_toname, '/');
2412 if (cp == NULL) {
2413 (void) strlcat(tofs, "/",
2414 ZFS_MAXNAMELEN);
2415 i = 0;
2416 } else {
2417 i = (cp - drrb->drr_toname);
2418 }
2419 } else {
2420 i = strcspn(drrb->drr_toname, "/@");
2421 }
2422 /* zfs_receive_one() will create_parents() */
2423 (void) strlcat(tofs, &drrb->drr_toname[i],
2424 ZFS_MAXNAMELEN);
2425 *strchr(tofs, '@') = '\0';
2426 }
2427
2428 if (recursive && !flags->dryrun && !flags->nomount) {
2429 VERIFY(0 == nvlist_alloc(&renamed,
2430 NV_UNIQUE_NAME, 0));
2431 }
2432
2433 softerr = recv_incremental_replication(hdl, tofs, flags,
2434 stream_nv, stream_avl, renamed);
2435
2436 /* Unmount renamed filesystems before receiving. */
2437 while ((pair = nvlist_next_nvpair(renamed,
2438 pair)) != NULL) {
2439 zfs_handle_t *zhp;
2440 prop_changelist_t *clp = NULL;
2441
2442 zhp = zfs_open(hdl, nvpair_name(pair),
2443 ZFS_TYPE_FILESYSTEM);
2444 if (zhp != NULL) {
2445 clp = changelist_gather(zhp,
2446 ZFS_PROP_MOUNTPOINT, 0, 0);
2447 zfs_close(zhp);
2448 if (clp != NULL) {
2449 softerr |=
2450 changelist_prefix(clp);
2451 changelist_free(clp);
2452 }
2453 }
2454 }
2455
2456 nvlist_free(renamed);
2457 }
2458 }
2459
2460 /*
2461 * Get the fs specified by the first path in the stream (the top level
2462 * specified by 'zfs send') and pass it to each invocation of
2463 * zfs_receive_one().
2464 */
2465 (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2466 ZFS_MAXNAMELEN);
2467 if ((cp = strchr(sendfs, '@')) != NULL)
2468 *cp = '\0';
2469
2470 /* Finally, receive each contained stream */
2471 do {
2472 /*
2473 * we should figure out if it has a recoverable
2474 * error, in which case do a recv_skip() and drive on.
2475 * Note, if we fail due to already having this guid,
2476 * zfs_receive_one() will take care of it (ie,
2477 * recv_skip() and return 0).
2478 */
2479 error = zfs_receive_impl(hdl, destname, flags, fd,
2480 sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2481 action_handlep);
2482 if (error == ENODATA) {
2483 error = 0;
2484 break;
2485 }
2486 anyerr |= error;
2487 } while (error == 0);
2488
2489 if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2490 /*
2491 * Now that we have the fs's they sent us, try the
2492 * renames again.
2493 */
2494 softerr = recv_incremental_replication(hdl, tofs, flags,
2495 stream_nv, stream_avl, NULL);
2496 }
2497
2498 out:
2499 fsavl_destroy(stream_avl);
2500 if (stream_nv)
2501 nvlist_free(stream_nv);
2502 if (softerr)
2503 error = -2;
2504 if (anyerr)
2505 error = -1;
2506 return (error);
2507 }
2508
2509 static void
2510 trunc_prop_errs(int truncated)
2511 {
2512 ASSERT(truncated != 0);
2513
2514 if (truncated == 1)
2515 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2516 "1 more property could not be set\n"));
2517 else
2518 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2519 "%d more properties could not be set\n"), truncated);
2520 }
2521
2522 static int
2523 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2524 {
2525 dmu_replay_record_t *drr;
2526 void *buf = malloc(1<<20);
2527 char errbuf[1024];
2528
2529 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2530 "cannot receive:"));
2531
2532 /* XXX would be great to use lseek if possible... */
2533 drr = buf;
2534
2535 while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2536 byteswap, NULL) == 0) {
2537 if (byteswap)
2538 drr->drr_type = BSWAP_32(drr->drr_type);
2539
2540 switch (drr->drr_type) {
2541 case DRR_BEGIN:
2542 /* NB: not to be used on v2 stream packages */
2543 if (drr->drr_payloadlen != 0) {
2544 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2545 "invalid substream header"));
2546 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2547 }
2548 break;
2549
2550 case DRR_END:
2551 free(buf);
2552 return (0);
2553
2554 case DRR_OBJECT:
2555 if (byteswap) {
2556 drr->drr_u.drr_object.drr_bonuslen =
2557 BSWAP_32(drr->drr_u.drr_object.
2558 drr_bonuslen);
2559 }
2560 (void) recv_read(hdl, fd, buf,
2561 P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2562 B_FALSE, NULL);
2563 break;
2564
2565 case DRR_WRITE:
2566 if (byteswap) {
2567 drr->drr_u.drr_write.drr_length =
2568 BSWAP_64(drr->drr_u.drr_write.drr_length);
2569 }
2570 (void) recv_read(hdl, fd, buf,
2571 drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2572 break;
2573 case DRR_SPILL:
2574 if (byteswap) {
2575 drr->drr_u.drr_write.drr_length =
2576 BSWAP_64(drr->drr_u.drr_spill.drr_length);
2577 }
2578 (void) recv_read(hdl, fd, buf,
2579 drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2580 break;
2581 case DRR_WRITE_BYREF:
2582 case DRR_FREEOBJECTS:
2583 case DRR_FREE:
2584 break;
2585
2586 default:
2587 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2588 "invalid record type"));
2589 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2590 }
2591 }
2592
2593 free(buf);
2594 return (-1);
2595 }
2596
2597 /*
2598 * Restores a backup of tosnap from the file descriptor specified by infd.
2599 */
2600 static int
2601 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2602 recvflags_t *flags, dmu_replay_record_t *drr,
2603 dmu_replay_record_t *drr_noswap, const char *sendfs,
2604 nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2605 uint64_t *action_handlep)
2606 {
2607 zfs_cmd_t zc = { 0 };
2608 time_t begin_time;
2609 int ioctl_err, ioctl_errno, err;
2610 char *cp;
2611 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2612 char errbuf[1024];
2613 char prop_errbuf[1024];
2614 const char *chopprefix;
2615 boolean_t newfs = B_FALSE;
2616 boolean_t stream_wantsnewfs;
2617 uint64_t parent_snapguid = 0;
2618 prop_changelist_t *clp = NULL;
2619 nvlist_t *snapprops_nvlist = NULL;
2620 zprop_errflags_t prop_errflags;
2621 boolean_t recursive;
2622
2623 begin_time = time(NULL);
2624
2625 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2626 "cannot receive"));
2627
2628 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2629 ENOENT);
2630
2631 if (stream_avl != NULL) {
2632 char *snapname;
2633 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2634 &snapname);
2635 nvlist_t *props;
2636 int ret;
2637
2638 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2639 &parent_snapguid);
2640 err = nvlist_lookup_nvlist(fs, "props", &props);
2641 if (err)
2642 VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2643
2644 if (flags->canmountoff) {
2645 VERIFY(0 == nvlist_add_uint64(props,
2646 zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2647 }
2648 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2649 if (err)
2650 nvlist_free(props);
2651
2652 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2653 VERIFY(0 == nvlist_lookup_nvlist(props,
2654 snapname, &snapprops_nvlist));
2655 }
2656
2657 if (ret != 0)
2658 return (-1);
2659 }
2660
2661 cp = NULL;
2662
2663 /*
2664 * Determine how much of the snapshot name stored in the stream
2665 * we are going to tack on to the name they specified on the
2666 * command line, and how much we are going to chop off.
2667 *
2668 * If they specified a snapshot, chop the entire name stored in
2669 * the stream.
2670 */
2671 if (flags->istail) {
2672 /*
2673 * A filesystem was specified with -e. We want to tack on only
2674 * the tail of the sent snapshot path.
2675 */
2676 if (strchr(tosnap, '@')) {
2677 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2678 "argument - snapshot not allowed with -e"));
2679 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2680 }
2681
2682 chopprefix = strrchr(sendfs, '/');
2683
2684 if (chopprefix == NULL) {
2685 /*
2686 * The tail is the poolname, so we need to
2687 * prepend a path separator.
2688 */
2689 int len = strlen(drrb->drr_toname);
2690 cp = malloc(len + 2);
2691 cp[0] = '/';
2692 (void) strcpy(&cp[1], drrb->drr_toname);
2693 chopprefix = cp;
2694 } else {
2695 chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2696 }
2697 } else if (flags->isprefix) {
2698 /*
2699 * A filesystem was specified with -d. We want to tack on
2700 * everything but the first element of the sent snapshot path
2701 * (all but the pool name).
2702 */
2703 if (strchr(tosnap, '@')) {
2704 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2705 "argument - snapshot not allowed with -d"));
2706 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2707 }
2708
2709 chopprefix = strchr(drrb->drr_toname, '/');
2710 if (chopprefix == NULL)
2711 chopprefix = strchr(drrb->drr_toname, '@');
2712 } else if (strchr(tosnap, '@') == NULL) {
2713 /*
2714 * If a filesystem was specified without -d or -e, we want to
2715 * tack on everything after the fs specified by 'zfs send'.
2716 */
2717 chopprefix = drrb->drr_toname + strlen(sendfs);
2718 } else {
2719 /* A snapshot was specified as an exact path (no -d or -e). */
2720 if (recursive) {
2721 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2722 "cannot specify snapshot name for multi-snapshot "
2723 "stream"));
2724 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2725 }
2726 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2727 }
2728
2729 ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2730 ASSERT(chopprefix > drrb->drr_toname);
2731 ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2732 ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2733 chopprefix[0] == '\0');
2734
2735 /*
2736 * Determine name of destination snapshot, store in zc_value.
2737 */
2738 (void) strcpy(zc.zc_top_ds, tosnap);
2739 (void) strcpy(zc.zc_value, tosnap);
2740 (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2741 free(cp);
2742 if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2743 zcmd_free_nvlists(&zc);
2744 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2745 }
2746
2747 /*
2748 * Determine the name of the origin snapshot, store in zc_string.
2749 */
2750 if (drrb->drr_flags & DRR_FLAG_CLONE) {
2751 if (guid_to_name(hdl, zc.zc_value,
2752 drrb->drr_fromguid, zc.zc_string) != 0) {
2753 zcmd_free_nvlists(&zc);
2754 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2755 "local origin for clone %s does not exist"),
2756 zc.zc_value);
2757 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2758 }
2759 if (flags->verbose)
2760 (void) printf("found clone origin %s\n", zc.zc_string);
2761 }
2762
2763 stream_wantsnewfs = (drrb->drr_fromguid == NULL ||
2764 (drrb->drr_flags & DRR_FLAG_CLONE));
2765
2766 if (stream_wantsnewfs) {
2767 /*
2768 * if the parent fs does not exist, look for it based on
2769 * the parent snap GUID
2770 */
2771 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2772 "cannot receive new filesystem stream"));
2773
2774 (void) strcpy(zc.zc_name, zc.zc_value);
2775 cp = strrchr(zc.zc_name, '/');
2776 if (cp)
2777 *cp = '\0';
2778 if (cp &&
2779 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2780 char suffix[ZFS_MAXNAMELEN];
2781 (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2782 if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
2783 zc.zc_value) == 0) {
2784 *strchr(zc.zc_value, '@') = '\0';
2785 (void) strcat(zc.zc_value, suffix);
2786 }
2787 }
2788 } else {
2789 /*
2790 * if the fs does not exist, look for it based on the
2791 * fromsnap GUID
2792 */
2793 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2794 "cannot receive incremental stream"));
2795
2796 (void) strcpy(zc.zc_name, zc.zc_value);
2797 *strchr(zc.zc_name, '@') = '\0';
2798
2799 /*
2800 * If the exact receive path was specified and this is the
2801 * topmost path in the stream, then if the fs does not exist we
2802 * should look no further.
2803 */
2804 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
2805 strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2806 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2807 char snap[ZFS_MAXNAMELEN];
2808 (void) strcpy(snap, strchr(zc.zc_value, '@'));
2809 if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
2810 zc.zc_value) == 0) {
2811 *strchr(zc.zc_value, '@') = '\0';
2812 (void) strcat(zc.zc_value, snap);
2813 }
2814 }
2815 }
2816
2817 (void) strcpy(zc.zc_name, zc.zc_value);
2818 *strchr(zc.zc_name, '@') = '\0';
2819
2820 if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2821 zfs_handle_t *zhp;
2822
2823 /*
2824 * Destination fs exists. Therefore this should either
2825 * be an incremental, or the stream specifies a new fs
2826 * (full stream or clone) and they want us to blow it
2827 * away (and have therefore specified -F and removed any
2828 * snapshots).
2829 */
2830 if (stream_wantsnewfs) {
2831 if (!flags->force) {
2832 zcmd_free_nvlists(&zc);
2833 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2834 "destination '%s' exists\n"
2835 "must specify -F to overwrite it"),
2836 zc.zc_name);
2837 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2838 }
2839 if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2840 &zc) == 0) {
2841 zcmd_free_nvlists(&zc);
2842 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2843 "destination has snapshots (eg. %s)\n"
2844 "must destroy them to overwrite it"),
2845 zc.zc_name);
2846 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2847 }
2848 }
2849
2850 if ((zhp = zfs_open(hdl, zc.zc_name,
2851 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2852 zcmd_free_nvlists(&zc);
2853 return (-1);
2854 }
2855
2856 if (stream_wantsnewfs &&
2857 zhp->zfs_dmustats.dds_origin[0]) {
2858 zcmd_free_nvlists(&zc);
2859 zfs_close(zhp);
2860 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2861 "destination '%s' is a clone\n"
2862 "must destroy it to overwrite it"),
2863 zc.zc_name);
2864 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2865 }
2866
2867 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2868 stream_wantsnewfs) {
2869 /* We can't do online recv in this case */
2870 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2871 if (clp == NULL) {
2872 zfs_close(zhp);
2873 zcmd_free_nvlists(&zc);
2874 return (-1);
2875 }
2876 if (changelist_prefix(clp) != 0) {
2877 changelist_free(clp);
2878 zfs_close(zhp);
2879 zcmd_free_nvlists(&zc);
2880 return (-1);
2881 }
2882 }
2883 zfs_close(zhp);
2884 } else {
2885 /*
2886 * Destination filesystem does not exist. Therefore we better
2887 * be creating a new filesystem (either from a full backup, or
2888 * a clone). It would therefore be invalid if the user
2889 * specified only the pool name (i.e. if the destination name
2890 * contained no slash character).
2891 */
2892 if (!stream_wantsnewfs ||
2893 (cp = strrchr(zc.zc_name, '/')) == NULL) {
2894 zcmd_free_nvlists(&zc);
2895 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2896 "destination '%s' does not exist"), zc.zc_name);
2897 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2898 }
2899
2900 /*
2901 * Trim off the final dataset component so we perform the
2902 * recvbackup ioctl to the filesystems's parent.
2903 */
2904 *cp = '\0';
2905
2906 if (flags->isprefix && !flags->istail && !flags->dryrun &&
2907 create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2908 zcmd_free_nvlists(&zc);
2909 return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2910 }
2911
2912 newfs = B_TRUE;
2913 }
2914
2915 zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2916 zc.zc_cookie = infd;
2917 zc.zc_guid = flags->force;
2918 if (flags->verbose) {
2919 (void) printf("%s %s stream of %s into %s\n",
2920 flags->dryrun ? "would receive" : "receiving",
2921 drrb->drr_fromguid ? "incremental" : "full",
2922 drrb->drr_toname, zc.zc_value);
2923 (void) fflush(stdout);
2924 }
2925
2926 if (flags->dryrun) {
2927 zcmd_free_nvlists(&zc);
2928 return (recv_skip(hdl, infd, flags->byteswap));
2929 }
2930
2931 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2932 zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2933 zc.zc_cleanup_fd = cleanup_fd;
2934 zc.zc_action_handle = *action_handlep;
2935
2936 err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2937 ioctl_errno = errno;
2938 prop_errflags = (zprop_errflags_t)zc.zc_obj;
2939
2940 if (err == 0) {
2941 nvlist_t *prop_errors;
2942 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2943 zc.zc_nvlist_dst_size, &prop_errors, 0));
2944
2945 nvpair_t *prop_err = NULL;
2946
2947 while ((prop_err = nvlist_next_nvpair(prop_errors,
2948 prop_err)) != NULL) {
2949 char tbuf[1024];
2950 zfs_prop_t prop;
2951 int intval;
2952
2953 prop = zfs_name_to_prop(nvpair_name(prop_err));
2954 (void) nvpair_value_int32(prop_err, &intval);
2955 if (strcmp(nvpair_name(prop_err),
2956 ZPROP_N_MORE_ERRORS) == 0) {
2957 trunc_prop_errs(intval);
2958 break;
2959 } else {
2960 (void) snprintf(tbuf, sizeof (tbuf),
2961 dgettext(TEXT_DOMAIN,
2962 "cannot receive %s property on %s"),
2963 nvpair_name(prop_err), zc.zc_name);
2964 zfs_setprop_error(hdl, prop, intval, tbuf);
2965 }
2966 }
2967 nvlist_free(prop_errors);
2968 }
2969
2970 zc.zc_nvlist_dst = 0;
2971 zc.zc_nvlist_dst_size = 0;
2972 zcmd_free_nvlists(&zc);
2973
2974 if (err == 0 && snapprops_nvlist) {
2975 zfs_cmd_t zc2 = { 0 };
2976
2977 (void) strcpy(zc2.zc_name, zc.zc_value);
2978 zc2.zc_cookie = B_TRUE; /* received */
2979 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
2980 (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
2981 zcmd_free_nvlists(&zc2);
2982 }
2983 }
2984
2985 if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
2986 /*
2987 * It may be that this snapshot already exists,
2988 * in which case we want to consume & ignore it
2989 * rather than failing.
2990 */
2991 avl_tree_t *local_avl;
2992 nvlist_t *local_nv, *fs;
2993 cp = strchr(zc.zc_value, '@');
2994
2995 /*
2996 * XXX Do this faster by just iterating over snaps in
2997 * this fs. Also if zc_value does not exist, we will
2998 * get a strange "does not exist" error message.
2999 */
3000 *cp = '\0';
3001 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3002 &local_nv, &local_avl) == 0) {
3003 *cp = '@';
3004 fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3005 fsavl_destroy(local_avl);
3006 nvlist_free(local_nv);
3007
3008 if (fs != NULL) {
3009 if (flags->verbose) {
3010 (void) printf("snap %s already exists; "
3011 "ignoring\n", zc.zc_value);
3012 }
3013 err = ioctl_err = recv_skip(hdl, infd,
3014 flags->byteswap);
3015 }
3016 }
3017 *cp = '@';
3018 }
3019
3020 if (ioctl_err != 0) {
3021 switch (ioctl_errno) {
3022 case ENODEV:
3023 cp = strchr(zc.zc_value, '@');
3024 *cp = '\0';
3025 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3026 "most recent snapshot of %s does not\n"
3027 "match incremental source"), zc.zc_value);
3028 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3029 *cp = '@';
3030 break;
3031 case ETXTBSY:
3032 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3033 "destination %s has been modified\n"
3034 "since most recent snapshot"), zc.zc_name);
3035 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3036 break;
3037 case EEXIST:
3038 cp = strchr(zc.zc_value, '@');
3039 if (newfs) {
3040 /* it's the containing fs that exists */
3041 *cp = '\0';
3042 }
3043 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3044 "destination already exists"));
3045 (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3046 dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3047 zc.zc_value);
3048 *cp = '@';
3049 break;
3050 case EINVAL:
3051 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3052 break;
3053 case ECKSUM:
3054 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3055 "invalid stream (checksum mismatch)"));
3056 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3057 break;
3058 case ENOTSUP:
3059 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3060 "pool must be upgraded to receive this stream."));
3061 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3062 break;
3063 case EDQUOT:
3064 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3065 "destination %s space quota exceeded"), zc.zc_name);
3066 (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3067 break;
3068 default:
3069 (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3070 }
3071 }
3072
3073 /*
3074 * Mount the target filesystem (if created). Also mount any
3075 * children of the target filesystem if we did a replication
3076 * receive (indicated by stream_avl being non-NULL).
3077 */
3078 cp = strchr(zc.zc_value, '@');
3079 if (cp && (ioctl_err == 0 || !newfs)) {
3080 zfs_handle_t *h;
3081
3082 *cp = '\0';
3083 h = zfs_open(hdl, zc.zc_value,
3084 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3085 if (h != NULL) {
3086 if (h->zfs_type == ZFS_TYPE_VOLUME) {
3087 *cp = '@';
3088 } else if (newfs || stream_avl) {
3089 /*
3090 * Track the first/top of hierarchy fs,
3091 * for mounting and sharing later.
3092 */
3093 if (top_zfs && *top_zfs == NULL)
3094 *top_zfs = zfs_strdup(hdl, zc.zc_value);
3095 }
3096 zfs_close(h);
3097 }
3098 *cp = '@';
3099 }
3100
3101 if (clp) {
3102 err |= changelist_postfix(clp);
3103 changelist_free(clp);
3104 }
3105
3106 if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3107 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3108 "failed to clear unreceived properties on %s"),
3109 zc.zc_name);
3110 (void) fprintf(stderr, "\n");
3111 }
3112 if (prop_errflags & ZPROP_ERR_NORESTORE) {
3113 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3114 "failed to restore original properties on %s"),
3115 zc.zc_name);
3116 (void) fprintf(stderr, "\n");
3117 }
3118
3119 if (err || ioctl_err)
3120 return (-1);
3121
3122 *action_handlep = zc.zc_action_handle;
3123
3124 if (flags->verbose) {
3125 char buf1[64];
3126 char buf2[64];
3127 uint64_t bytes = zc.zc_cookie;
3128 time_t delta = time(NULL) - begin_time;
3129 if (delta == 0)
3130 delta = 1;
3131 zfs_nicenum(bytes, buf1, sizeof (buf1));
3132 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3133
3134 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3135 buf1, delta, buf2);
3136 }
3137
3138 return (0);
3139 }
3140
3141 static int
3142 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3143 int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl,
3144 char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
3145 {
3146 int err;
3147 dmu_replay_record_t drr, drr_noswap;
3148 struct drr_begin *drrb = &drr.drr_u.drr_begin;
3149 char errbuf[1024];
3150 zio_cksum_t zcksum = { 0 };
3151 uint64_t featureflags;
3152 int hdrtype;
3153
3154 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3155 "cannot receive"));
3156
3157 if (flags->isprefix &&
3158 !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3159 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3160 "(%s) does not exist"), tosnap);
3161 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3162 }
3163
3164 /* read in the BEGIN record */
3165 if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3166 &zcksum)))
3167 return (err);
3168
3169 if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3170 /* It's the double end record at the end of a package */
3171 return (ENODATA);
3172 }
3173
3174 /* the kernel needs the non-byteswapped begin record */
3175 drr_noswap = drr;
3176
3177 flags->byteswap = B_FALSE;
3178 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3179 /*
3180 * We computed the checksum in the wrong byteorder in
3181 * recv_read() above; do it again correctly.
3182 */
3183 bzero(&zcksum, sizeof (zio_cksum_t));
3184 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3185 flags->byteswap = B_TRUE;
3186
3187 drr.drr_type = BSWAP_32(drr.drr_type);
3188 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3189 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3190 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3191 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3192 drrb->drr_type = BSWAP_32(drrb->drr_type);
3193 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3194 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3195 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3196 }
3197
3198 if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3199 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3200 "stream (bad magic number)"));
3201 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3202 }
3203
3204 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3205 hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3206
3207 if (!DMU_STREAM_SUPPORTED(featureflags) ||
3208 (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3209 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3210 "stream has unsupported feature, feature flags = %lx"),
3211 featureflags);
3212 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3213 }
3214
3215 if (strchr(drrb->drr_toname, '@') == NULL) {
3216 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3217 "stream (bad snapshot name)"));
3218 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3219 }
3220
3221 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3222 char nonpackage_sendfs[ZFS_MAXNAMELEN];
3223 if (sendfs == NULL) {
3224 /*
3225 * We were not called from zfs_receive_package(). Get
3226 * the fs specified by 'zfs send'.
3227 */
3228 char *cp;
3229 (void) strlcpy(nonpackage_sendfs,
3230 drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
3231 if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3232 *cp = '\0';
3233 sendfs = nonpackage_sendfs;
3234 }
3235 return (zfs_receive_one(hdl, infd, tosnap, flags,
3236 &drr, &drr_noswap, sendfs, stream_nv, stream_avl,
3237 top_zfs, cleanup_fd, action_handlep));
3238 } else {
3239 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3240 DMU_COMPOUNDSTREAM);
3241 return (zfs_receive_package(hdl, infd, tosnap, flags,
3242 &drr, &zcksum, top_zfs, cleanup_fd, action_handlep));
3243 }
3244 }
3245
3246 /*
3247 * Restores a backup of tosnap from the file descriptor specified by infd.
3248 * Return 0 on total success, -2 if some things couldn't be
3249 * destroyed/renamed/promoted, -1 if some things couldn't be received.
3250 * (-1 will override -2).
3251 */
3252 int
3253 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3254 int infd, avl_tree_t *stream_avl)
3255 {
3256 char *top_zfs = NULL;
3257 int err;
3258 int cleanup_fd;
3259 uint64_t action_handle = 0;
3260
3261 cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3262 VERIFY(cleanup_fd >= 0);
3263
3264 err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL,
3265 stream_avl, &top_zfs, cleanup_fd, &action_handle);
3266
3267 VERIFY(0 == close(cleanup_fd));
3268
3269 if (err == 0 && !flags->nomount && top_zfs) {
3270 zfs_handle_t *zhp;
3271 prop_changelist_t *clp;
3272
3273 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3274 if (zhp != NULL) {
3275 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3276 CL_GATHER_MOUNT_ALWAYS, 0);
3277 zfs_close(zhp);
3278 if (clp != NULL) {
3279 /* mount and share received datasets */
3280 err = changelist_postfix(clp);
3281 changelist_free(clp);
3282 }
3283 }
3284 if (zhp == NULL || clp == NULL || err)
3285 err = -1;
3286 }
3287 if (top_zfs)
3288 free(top_zfs);
3289
3290 return (err);
3291 }