xref: /freebsd/sys/contrib/openzfs/lib/libzfs/libzfs_sendrecv.c (revision f374ba41f55c1a127303d92d830dd58eef2f5243)
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 https://opensource.org/licenses/CDDL-1.0.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26  * Copyright (c) 2012 Pawel Jakub Dawidek <pawel@dawidek.net>.
27  * All rights reserved
28  * Copyright (c) 2013 Steven Hartland. All rights reserved.
29  * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
30  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
31  * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
32  * Copyright (c) 2019 Datto Inc.
33  */
34 
35 #include <assert.h>
36 #include <ctype.h>
37 #include <errno.h>
38 #include <libintl.h>
39 #include <stdio.h>
40 #include <stdlib.h>
41 #include <string.h>
42 #include <unistd.h>
43 #include <stddef.h>
44 #include <fcntl.h>
45 #include <sys/mount.h>
46 #include <sys/mntent.h>
47 #include <sys/mnttab.h>
48 #include <sys/avl.h>
49 #include <sys/debug.h>
50 #include <sys/stat.h>
51 #include <pthread.h>
52 #include <umem.h>
53 #include <time.h>
54 
55 #include <libzfs.h>
56 #include <libzfs_core.h>
57 #include <libzutil.h>
58 
59 #include "zfs_namecheck.h"
60 #include "zfs_prop.h"
61 #include "zfs_fletcher.h"
62 #include "libzfs_impl.h"
63 #include <cityhash.h>
64 #include <zlib.h>
65 #include <sys/zio_checksum.h>
66 #include <sys/dsl_crypt.h>
67 #include <sys/ddt.h>
68 #include <sys/socket.h>
69 #include <sys/sha2.h>
70 
71 static int zfs_receive_impl(libzfs_handle_t *, const char *, const char *,
72     recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **,
73     const char *, nvlist_t *);
74 static int guid_to_name_redact_snaps(libzfs_handle_t *hdl, const char *parent,
75     uint64_t guid, boolean_t bookmark_ok, uint64_t *redact_snap_guids,
76     uint64_t num_redact_snaps, char *name);
77 static int guid_to_name(libzfs_handle_t *, const char *,
78     uint64_t, boolean_t, char *);
79 
80 typedef struct progress_arg {
81 	zfs_handle_t *pa_zhp;
82 	int pa_fd;
83 	boolean_t pa_parsable;
84 	boolean_t pa_estimate;
85 	int pa_verbosity;
86 	boolean_t pa_astitle;
87 	uint64_t pa_size;
88 } progress_arg_t;
89 
90 static int
91 dump_record(dmu_replay_record_t *drr, void *payload, size_t payload_len,
92     zio_cksum_t *zc, int outfd)
93 {
94 	ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
95 	    ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t));
96 	fletcher_4_incremental_native(drr,
97 	    offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc);
98 	if (drr->drr_type != DRR_BEGIN) {
99 		ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u.
100 		    drr_checksum.drr_checksum));
101 		drr->drr_u.drr_checksum.drr_checksum = *zc;
102 	}
103 	fletcher_4_incremental_native(&drr->drr_u.drr_checksum.drr_checksum,
104 	    sizeof (zio_cksum_t), zc);
105 	if (write(outfd, drr, sizeof (*drr)) == -1)
106 		return (errno);
107 	if (payload_len != 0) {
108 		fletcher_4_incremental_native(payload, payload_len, zc);
109 		if (write(outfd, payload, payload_len) == -1)
110 			return (errno);
111 	}
112 	return (0);
113 }
114 
115 /*
116  * Routines for dealing with the AVL tree of fs-nvlists
117  */
118 typedef struct fsavl_node {
119 	avl_node_t fn_node;
120 	nvlist_t *fn_nvfs;
121 	char *fn_snapname;
122 	uint64_t fn_guid;
123 } fsavl_node_t;
124 
125 static int
126 fsavl_compare(const void *arg1, const void *arg2)
127 {
128 	const fsavl_node_t *fn1 = (const fsavl_node_t *)arg1;
129 	const fsavl_node_t *fn2 = (const fsavl_node_t *)arg2;
130 
131 	return (TREE_CMP(fn1->fn_guid, fn2->fn_guid));
132 }
133 
134 /*
135  * Given the GUID of a snapshot, find its containing filesystem and
136  * (optionally) name.
137  */
138 static nvlist_t *
139 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
140 {
141 	fsavl_node_t fn_find;
142 	fsavl_node_t *fn;
143 
144 	fn_find.fn_guid = snapguid;
145 
146 	fn = avl_find(avl, &fn_find, NULL);
147 	if (fn) {
148 		if (snapname)
149 			*snapname = fn->fn_snapname;
150 		return (fn->fn_nvfs);
151 	}
152 	return (NULL);
153 }
154 
155 static void
156 fsavl_destroy(avl_tree_t *avl)
157 {
158 	fsavl_node_t *fn;
159 	void *cookie;
160 
161 	if (avl == NULL)
162 		return;
163 
164 	cookie = NULL;
165 	while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
166 		free(fn);
167 	avl_destroy(avl);
168 	free(avl);
169 }
170 
171 /*
172  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
173  */
174 static avl_tree_t *
175 fsavl_create(nvlist_t *fss)
176 {
177 	avl_tree_t *fsavl;
178 	nvpair_t *fselem = NULL;
179 
180 	if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
181 		return (NULL);
182 
183 	avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
184 	    offsetof(fsavl_node_t, fn_node));
185 
186 	while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
187 		nvlist_t *nvfs, *snaps;
188 		nvpair_t *snapelem = NULL;
189 
190 		nvfs = fnvpair_value_nvlist(fselem);
191 		snaps = fnvlist_lookup_nvlist(nvfs, "snaps");
192 
193 		while ((snapelem =
194 		    nvlist_next_nvpair(snaps, snapelem)) != NULL) {
195 			fsavl_node_t *fn;
196 
197 			if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
198 				fsavl_destroy(fsavl);
199 				return (NULL);
200 			}
201 			fn->fn_nvfs = nvfs;
202 			fn->fn_snapname = nvpair_name(snapelem);
203 			fn->fn_guid = fnvpair_value_uint64(snapelem);
204 
205 			/*
206 			 * Note: if there are multiple snaps with the
207 			 * same GUID, we ignore all but one.
208 			 */
209 			avl_index_t where = 0;
210 			if (avl_find(fsavl, fn, &where) == NULL)
211 				avl_insert(fsavl, fn, where);
212 			else
213 				free(fn);
214 		}
215 	}
216 
217 	return (fsavl);
218 }
219 
220 /*
221  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
222  */
223 typedef struct send_data {
224 	/*
225 	 * assigned inside every recursive call,
226 	 * restored from *_save on return:
227 	 *
228 	 * guid of fromsnap snapshot in parent dataset
229 	 * txg of fromsnap snapshot in current dataset
230 	 * txg of tosnap snapshot in current dataset
231 	 */
232 
233 	uint64_t parent_fromsnap_guid;
234 	uint64_t fromsnap_txg;
235 	uint64_t tosnap_txg;
236 
237 	/* the nvlists get accumulated during depth-first traversal */
238 	nvlist_t *parent_snaps;
239 	nvlist_t *fss;
240 	nvlist_t *snapprops;
241 	nvlist_t *snapholds;	/* user holds */
242 
243 	/* send-receive configuration, does not change during traversal */
244 	const char *fsname;
245 	const char *fromsnap;
246 	const char *tosnap;
247 	boolean_t recursive;
248 	boolean_t raw;
249 	boolean_t doall;
250 	boolean_t replicate;
251 	boolean_t skipmissing;
252 	boolean_t verbose;
253 	boolean_t backup;
254 	boolean_t seenfrom;
255 	boolean_t seento;
256 	boolean_t holds;	/* were holds requested with send -h */
257 	boolean_t props;
258 
259 	/*
260 	 * The header nvlist is of the following format:
261 	 * {
262 	 *   "tosnap" -> string
263 	 *   "fromsnap" -> string (if incremental)
264 	 *   "fss" -> {
265 	 *	id -> {
266 	 *
267 	 *	 "name" -> string (full name; for debugging)
268 	 *	 "parentfromsnap" -> number (guid of fromsnap in parent)
269 	 *
270 	 *	 "props" -> { name -> value (only if set here) }
271 	 *	 "snaps" -> { name (lastname) -> number (guid) }
272 	 *	 "snapprops" -> { name (lastname) -> { name -> value } }
273 	 *	 "snapholds" -> { name (lastname) -> { holdname -> crtime } }
274 	 *
275 	 *	 "origin" -> number (guid) (if clone)
276 	 *	 "is_encroot" -> boolean
277 	 *	 "sent" -> boolean (not on-disk)
278 	 *	}
279 	 *   }
280 	 * }
281 	 *
282 	 */
283 } send_data_t;
284 
285 static void
286 send_iterate_prop(zfs_handle_t *zhp, boolean_t received_only, nvlist_t *nv);
287 
288 /*
289  * Collect guid, valid props, optionally holds, etc. of a snapshot.
290  * This interface is intended for use as a zfs_iter_snapshots_sorted visitor.
291  */
292 static int
293 send_iterate_snap(zfs_handle_t *zhp, void *arg)
294 {
295 	send_data_t *sd = arg;
296 	uint64_t guid = zhp->zfs_dmustats.dds_guid;
297 	uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
298 	boolean_t isfromsnap, istosnap, istosnapwithnofrom;
299 	char *snapname;
300 	const char *from = sd->fromsnap;
301 	const char *to = sd->tosnap;
302 
303 	snapname = strrchr(zhp->zfs_name, '@');
304 	assert(snapname != NULL);
305 	++snapname;
306 
307 	isfromsnap = (from != NULL && strcmp(from, snapname) == 0);
308 	istosnap = (to != NULL && strcmp(to, snapname) == 0);
309 	istosnapwithnofrom = (istosnap && from == NULL);
310 
311 	if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
312 		if (sd->verbose) {
313 			(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
314 			    "skipping snapshot %s because it was created "
315 			    "after the destination snapshot (%s)\n"),
316 			    zhp->zfs_name, to);
317 		}
318 		zfs_close(zhp);
319 		return (0);
320 	}
321 
322 	fnvlist_add_uint64(sd->parent_snaps, snapname, guid);
323 
324 	/*
325 	 * NB: if there is no fromsnap here (it's a newly created fs in
326 	 * an incremental replication), we will substitute the tosnap.
327 	 */
328 	if (isfromsnap || (sd->parent_fromsnap_guid == 0 && istosnap))
329 		sd->parent_fromsnap_guid = guid;
330 
331 	if (!sd->recursive) {
332 		/*
333 		 * To allow a doall stream to work properly
334 		 * with a NULL fromsnap
335 		 */
336 		if (sd->doall && from == NULL && !sd->seenfrom)
337 			sd->seenfrom = B_TRUE;
338 
339 		if (!sd->seenfrom && isfromsnap) {
340 			sd->seenfrom = B_TRUE;
341 			zfs_close(zhp);
342 			return (0);
343 		}
344 
345 		if ((sd->seento || !sd->seenfrom) && !istosnapwithnofrom) {
346 			zfs_close(zhp);
347 			return (0);
348 		}
349 
350 		if (istosnap)
351 			sd->seento = B_TRUE;
352 	}
353 
354 	nvlist_t *nv = fnvlist_alloc();
355 	send_iterate_prop(zhp, sd->backup, nv);
356 	fnvlist_add_nvlist(sd->snapprops, snapname, nv);
357 	fnvlist_free(nv);
358 
359 	if (sd->holds) {
360 		nvlist_t *holds;
361 		if (lzc_get_holds(zhp->zfs_name, &holds) == 0) {
362 			fnvlist_add_nvlist(sd->snapholds, snapname, holds);
363 			fnvlist_free(holds);
364 		}
365 	}
366 
367 	zfs_close(zhp);
368 	return (0);
369 }
370 
371 /*
372  * Collect all valid props from the handle snap into an nvlist.
373  */
374 static void
375 send_iterate_prop(zfs_handle_t *zhp, boolean_t received_only, nvlist_t *nv)
376 {
377 	nvlist_t *props;
378 
379 	if (received_only)
380 		props = zfs_get_recvd_props(zhp);
381 	else
382 		props = zhp->zfs_props;
383 
384 	nvpair_t *elem = NULL;
385 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
386 		char *propname = nvpair_name(elem);
387 		zfs_prop_t prop = zfs_name_to_prop(propname);
388 
389 		if (!zfs_prop_user(propname)) {
390 			/*
391 			 * Realistically, this should never happen.  However,
392 			 * we want the ability to add DSL properties without
393 			 * needing to make incompatible version changes.  We
394 			 * need to ignore unknown properties to allow older
395 			 * software to still send datasets containing these
396 			 * properties, with the unknown properties elided.
397 			 */
398 			if (prop == ZPROP_INVAL)
399 				continue;
400 
401 			if (zfs_prop_readonly(prop))
402 				continue;
403 		}
404 
405 		nvlist_t *propnv = fnvpair_value_nvlist(elem);
406 
407 		boolean_t isspacelimit = (prop == ZFS_PROP_QUOTA ||
408 		    prop == ZFS_PROP_RESERVATION ||
409 		    prop == ZFS_PROP_REFQUOTA ||
410 		    prop == ZFS_PROP_REFRESERVATION);
411 		if (isspacelimit && zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
412 			continue;
413 
414 		char *source;
415 		if (nvlist_lookup_string(propnv, ZPROP_SOURCE, &source) == 0) {
416 			if (strcmp(source, zhp->zfs_name) != 0 &&
417 			    strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0)
418 				continue;
419 		} else {
420 			/*
421 			 * May have no source before SPA_VERSION_RECVD_PROPS,
422 			 * but is still modifiable.
423 			 */
424 			if (!isspacelimit)
425 				continue;
426 		}
427 
428 		if (zfs_prop_user(propname) ||
429 		    zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
430 			char *value;
431 			value = fnvlist_lookup_string(propnv, ZPROP_VALUE);
432 			fnvlist_add_string(nv, propname, value);
433 		} else {
434 			uint64_t value;
435 			value = fnvlist_lookup_uint64(propnv, ZPROP_VALUE);
436 			fnvlist_add_uint64(nv, propname, value);
437 		}
438 	}
439 }
440 
441 /*
442  * returns snapshot guid
443  * and returns 0 if the snapshot does not exist
444  */
445 static uint64_t
446 get_snap_guid(libzfs_handle_t *hdl, const char *fs, const char *snap)
447 {
448 	char name[MAXPATHLEN + 1];
449 	uint64_t guid = 0;
450 
451 	if (fs == NULL || fs[0] == '\0' || snap == NULL || snap[0] == '\0')
452 		return (guid);
453 
454 	(void) snprintf(name, sizeof (name), "%s@%s", fs, snap);
455 	zfs_handle_t *zhp = zfs_open(hdl, name, ZFS_TYPE_SNAPSHOT);
456 	if (zhp != NULL) {
457 		guid = zfs_prop_get_int(zhp, ZFS_PROP_GUID);
458 		zfs_close(zhp);
459 	}
460 
461 	return (guid);
462 }
463 
464 /*
465  * returns snapshot creation txg
466  * and returns 0 if the snapshot does not exist
467  */
468 static uint64_t
469 get_snap_txg(libzfs_handle_t *hdl, const char *fs, const char *snap)
470 {
471 	char name[ZFS_MAX_DATASET_NAME_LEN];
472 	uint64_t txg = 0;
473 
474 	if (fs == NULL || fs[0] == '\0' || snap == NULL || snap[0] == '\0')
475 		return (txg);
476 
477 	(void) snprintf(name, sizeof (name), "%s@%s", fs, snap);
478 	if (zfs_dataset_exists(hdl, name, ZFS_TYPE_SNAPSHOT)) {
479 		zfs_handle_t *zhp = zfs_open(hdl, name, ZFS_TYPE_SNAPSHOT);
480 		if (zhp != NULL) {
481 			txg = zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG);
482 			zfs_close(zhp);
483 		}
484 	}
485 
486 	return (txg);
487 }
488 
489 /*
490  * Recursively generate nvlists describing datasets.  See comment
491  * for the data structure send_data_t above for description of contents
492  * of the nvlist.
493  */
494 static int
495 send_iterate_fs(zfs_handle_t *zhp, void *arg)
496 {
497 	send_data_t *sd = arg;
498 	nvlist_t *nvfs = NULL, *nv = NULL;
499 	int rv = 0;
500 	uint64_t min_txg = 0, max_txg = 0;
501 	uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
502 	uint64_t guid = zhp->zfs_dmustats.dds_guid;
503 	uint64_t fromsnap_txg, tosnap_txg;
504 	char guidstring[64];
505 
506 	/* These fields are restored on return from a recursive call. */
507 	uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
508 	uint64_t fromsnap_txg_save = sd->fromsnap_txg;
509 	uint64_t tosnap_txg_save = sd->tosnap_txg;
510 
511 	fromsnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->fromsnap);
512 	if (fromsnap_txg != 0)
513 		sd->fromsnap_txg = fromsnap_txg;
514 
515 	tosnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->tosnap);
516 	if (tosnap_txg != 0)
517 		sd->tosnap_txg = tosnap_txg;
518 
519 	/*
520 	 * On the send side, if the current dataset does not have tosnap,
521 	 * perform two additional checks:
522 	 *
523 	 * - Skip sending the current dataset if it was created later than
524 	 *   the parent tosnap.
525 	 * - Return error if the current dataset was created earlier than
526 	 *   the parent tosnap, unless --skip-missing specified. Then
527 	 *   just print a warning.
528 	 */
529 	if (sd->tosnap != NULL && tosnap_txg == 0) {
530 		if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
531 			if (sd->verbose) {
532 				(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
533 				    "skipping dataset %s: snapshot %s does "
534 				    "not exist\n"), zhp->zfs_name, sd->tosnap);
535 			}
536 		} else if (sd->skipmissing) {
537 			(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
538 			    "WARNING: skipping dataset %s and its children:"
539 			    " snapshot %s does not exist\n"),
540 			    zhp->zfs_name, sd->tosnap);
541 		} else {
542 			(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
543 			    "cannot send %s@%s%s: snapshot %s@%s does not "
544 			    "exist\n"), sd->fsname, sd->tosnap, sd->recursive ?
545 			    dgettext(TEXT_DOMAIN, " recursively") : "",
546 			    zhp->zfs_name, sd->tosnap);
547 			rv = EZFS_NOENT;
548 		}
549 		goto out;
550 	}
551 
552 	nvfs = fnvlist_alloc();
553 	fnvlist_add_string(nvfs, "name", zhp->zfs_name);
554 	fnvlist_add_uint64(nvfs, "parentfromsnap", sd->parent_fromsnap_guid);
555 
556 	if (zhp->zfs_dmustats.dds_origin[0] != '\0') {
557 		zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
558 		    zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
559 		if (origin == NULL) {
560 			rv = -1;
561 			goto out;
562 		}
563 		fnvlist_add_uint64(nvfs, "origin",
564 		    origin->zfs_dmustats.dds_guid);
565 		zfs_close(origin);
566 	}
567 
568 	/* Iterate over props. */
569 	if (sd->props || sd->backup || sd->recursive) {
570 		nv = fnvlist_alloc();
571 		send_iterate_prop(zhp, sd->backup, nv);
572 		fnvlist_add_nvlist(nvfs, "props", nv);
573 	}
574 	if (zfs_prop_get_int(zhp, ZFS_PROP_ENCRYPTION) != ZIO_CRYPT_OFF) {
575 		boolean_t encroot;
576 
577 		/* Determine if this dataset is an encryption root. */
578 		if (zfs_crypto_get_encryption_root(zhp, &encroot, NULL) != 0) {
579 			rv = -1;
580 			goto out;
581 		}
582 
583 		if (encroot)
584 			fnvlist_add_boolean(nvfs, "is_encroot");
585 
586 		/*
587 		 * Encrypted datasets can only be sent with properties if
588 		 * the raw flag is specified because the receive side doesn't
589 		 * currently have a mechanism for recursively asking the user
590 		 * for new encryption parameters.
591 		 */
592 		if (!sd->raw) {
593 			(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
594 			    "cannot send %s@%s: encrypted dataset %s may not "
595 			    "be sent with properties without the raw flag\n"),
596 			    sd->fsname, sd->tosnap, zhp->zfs_name);
597 			rv = -1;
598 			goto out;
599 		}
600 
601 	}
602 
603 	/*
604 	 * Iterate over snaps, and set sd->parent_fromsnap_guid.
605 	 *
606 	 * If this is a "doall" send, a replicate send or we're just trying
607 	 * to gather a list of previous snapshots, iterate through all the
608 	 * snaps in the txg range. Otherwise just look at the one we're
609 	 * interested in.
610 	 */
611 	sd->parent_fromsnap_guid = 0;
612 	sd->parent_snaps = fnvlist_alloc();
613 	sd->snapprops = fnvlist_alloc();
614 	if (sd->holds)
615 		sd->snapholds = fnvlist_alloc();
616 	if (sd->doall || sd->replicate || sd->tosnap == NULL) {
617 		if (!sd->replicate && fromsnap_txg != 0)
618 			min_txg = fromsnap_txg;
619 		if (!sd->replicate && tosnap_txg != 0)
620 			max_txg = tosnap_txg;
621 		(void) zfs_iter_snapshots_sorted(zhp, 0, send_iterate_snap, sd,
622 		    min_txg, max_txg);
623 	} else {
624 		char snapname[MAXPATHLEN] = { 0 };
625 		zfs_handle_t *snap;
626 
627 		(void) snprintf(snapname, sizeof (snapname), "%s@%s",
628 		    zhp->zfs_name, sd->tosnap);
629 		if (sd->fromsnap != NULL)
630 			sd->seenfrom = B_TRUE;
631 		snap = zfs_open(zhp->zfs_hdl, snapname, ZFS_TYPE_SNAPSHOT);
632 		if (snap != NULL)
633 			(void) send_iterate_snap(snap, sd);
634 	}
635 
636 	fnvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps);
637 	fnvlist_free(sd->parent_snaps);
638 	fnvlist_add_nvlist(nvfs, "snapprops", sd->snapprops);
639 	fnvlist_free(sd->snapprops);
640 	if (sd->holds) {
641 		fnvlist_add_nvlist(nvfs, "snapholds", sd->snapholds);
642 		fnvlist_free(sd->snapholds);
643 	}
644 
645 	/* Do not allow the size of the properties list to exceed the limit */
646 	if ((fnvlist_size(nvfs) + fnvlist_size(sd->fss)) >
647 	    zhp->zfs_hdl->libzfs_max_nvlist) {
648 		(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
649 		    "warning: cannot send %s@%s: the size of the list of "
650 		    "snapshots and properties is too large to be received "
651 		    "successfully.\n"
652 		    "Select a smaller number of snapshots to send.\n"),
653 		    zhp->zfs_name, sd->tosnap);
654 		rv = EZFS_NOSPC;
655 		goto out;
656 	}
657 	/* Add this fs to nvlist. */
658 	(void) snprintf(guidstring, sizeof (guidstring),
659 	    "0x%llx", (longlong_t)guid);
660 	fnvlist_add_nvlist(sd->fss, guidstring, nvfs);
661 
662 	/* Iterate over children. */
663 	if (sd->recursive)
664 		rv = zfs_iter_filesystems(zhp, 0, send_iterate_fs, sd);
665 
666 out:
667 	/* Restore saved fields. */
668 	sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
669 	sd->fromsnap_txg = fromsnap_txg_save;
670 	sd->tosnap_txg = tosnap_txg_save;
671 
672 	fnvlist_free(nv);
673 	fnvlist_free(nvfs);
674 
675 	zfs_close(zhp);
676 	return (rv);
677 }
678 
679 static int
680 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
681     const char *tosnap, boolean_t recursive, boolean_t raw, boolean_t doall,
682     boolean_t replicate, boolean_t skipmissing, boolean_t verbose,
683     boolean_t backup, boolean_t holds, boolean_t props, nvlist_t **nvlp,
684     avl_tree_t **avlp)
685 {
686 	zfs_handle_t *zhp;
687 	send_data_t sd = { 0 };
688 	int error;
689 
690 	zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
691 	if (zhp == NULL)
692 		return (EZFS_BADTYPE);
693 
694 	sd.fss = fnvlist_alloc();
695 	sd.fsname = fsname;
696 	sd.fromsnap = fromsnap;
697 	sd.tosnap = tosnap;
698 	sd.recursive = recursive;
699 	sd.raw = raw;
700 	sd.doall = doall;
701 	sd.replicate = replicate;
702 	sd.skipmissing = skipmissing;
703 	sd.verbose = verbose;
704 	sd.backup = backup;
705 	sd.holds = holds;
706 	sd.props = props;
707 
708 	if ((error = send_iterate_fs(zhp, &sd)) != 0) {
709 		fnvlist_free(sd.fss);
710 		if (avlp != NULL)
711 			*avlp = NULL;
712 		*nvlp = NULL;
713 		return (error);
714 	}
715 
716 	if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
717 		fnvlist_free(sd.fss);
718 		*nvlp = NULL;
719 		return (EZFS_NOMEM);
720 	}
721 
722 	*nvlp = sd.fss;
723 	return (0);
724 }
725 
726 /*
727  * Routines specific to "zfs send"
728  */
729 typedef struct send_dump_data {
730 	/* these are all just the short snapname (the part after the @) */
731 	const char *fromsnap;
732 	const char *tosnap;
733 	char prevsnap[ZFS_MAX_DATASET_NAME_LEN];
734 	uint64_t prevsnap_obj;
735 	boolean_t seenfrom, seento, replicate, doall, fromorigin;
736 	boolean_t dryrun, parsable, progress, embed_data, std_out;
737 	boolean_t large_block, compress, raw, holds;
738 	boolean_t progressastitle;
739 	int outfd;
740 	boolean_t err;
741 	nvlist_t *fss;
742 	nvlist_t *snapholds;
743 	avl_tree_t *fsavl;
744 	snapfilter_cb_t *filter_cb;
745 	void *filter_cb_arg;
746 	nvlist_t *debugnv;
747 	char holdtag[ZFS_MAX_DATASET_NAME_LEN];
748 	int cleanup_fd;
749 	int verbosity;
750 	uint64_t size;
751 } send_dump_data_t;
752 
753 static int
754 zfs_send_space(zfs_handle_t *zhp, const char *snapname, const char *from,
755     enum lzc_send_flags flags, uint64_t *spacep)
756 {
757 	assert(snapname != NULL);
758 
759 	int error = lzc_send_space(snapname, from, flags, spacep);
760 	if (error == 0)
761 		return (0);
762 
763 	char errbuf[ERRBUFLEN];
764 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
765 	    "warning: cannot estimate space for '%s'"), snapname);
766 
767 	libzfs_handle_t *hdl = zhp->zfs_hdl;
768 	switch (error) {
769 	case EXDEV:
770 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
771 		    "not an earlier snapshot from the same fs"));
772 		return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
773 
774 	case ENOENT:
775 		if (zfs_dataset_exists(hdl, snapname,
776 		    ZFS_TYPE_SNAPSHOT)) {
777 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
778 			    "incremental source (%s) does not exist"),
779 			    snapname);
780 		}
781 		return (zfs_error(hdl, EZFS_NOENT, errbuf));
782 
783 	case EDQUOT:
784 	case EFBIG:
785 	case EIO:
786 	case ENOLINK:
787 	case ENOSPC:
788 	case ENOSTR:
789 	case ENXIO:
790 	case EPIPE:
791 	case ERANGE:
792 	case EFAULT:
793 	case EROFS:
794 	case EINVAL:
795 		zfs_error_aux(hdl, "%s", strerror(error));
796 		return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
797 
798 	default:
799 		return (zfs_standard_error(hdl, error, errbuf));
800 	}
801 }
802 
803 /*
804  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
805  * NULL) to the file descriptor specified by outfd.
806  */
807 static int
808 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
809     boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
810     nvlist_t *debugnv)
811 {
812 	zfs_cmd_t zc = {"\0"};
813 	libzfs_handle_t *hdl = zhp->zfs_hdl;
814 	nvlist_t *thisdbg;
815 
816 	assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
817 	assert(fromsnap_obj == 0 || !fromorigin);
818 
819 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
820 	zc.zc_cookie = outfd;
821 	zc.zc_obj = fromorigin;
822 	zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
823 	zc.zc_fromobj = fromsnap_obj;
824 	zc.zc_flags = flags;
825 
826 	if (debugnv != NULL) {
827 		thisdbg = fnvlist_alloc();
828 		if (fromsnap != NULL && fromsnap[0] != '\0')
829 			fnvlist_add_string(thisdbg, "fromsnap", fromsnap);
830 	}
831 
832 	if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
833 		char errbuf[ERRBUFLEN];
834 		int error = errno;
835 
836 		(void) snprintf(errbuf, sizeof (errbuf), "%s '%s'",
837 		    dgettext(TEXT_DOMAIN, "warning: cannot send"),
838 		    zhp->zfs_name);
839 
840 		if (debugnv != NULL) {
841 			fnvlist_add_uint64(thisdbg, "error", error);
842 			fnvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg);
843 			fnvlist_free(thisdbg);
844 		}
845 
846 		switch (error) {
847 		case EXDEV:
848 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
849 			    "not an earlier snapshot from the same fs"));
850 			return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
851 
852 		case EACCES:
853 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
854 			    "source key must be loaded"));
855 			return (zfs_error(hdl, EZFS_CRYPTOFAILED, errbuf));
856 
857 		case ENOENT:
858 			if (zfs_dataset_exists(hdl, zc.zc_name,
859 			    ZFS_TYPE_SNAPSHOT)) {
860 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
861 				    "incremental source (@%s) does not exist"),
862 				    zc.zc_value);
863 			}
864 			return (zfs_error(hdl, EZFS_NOENT, errbuf));
865 
866 		case EDQUOT:
867 		case EFBIG:
868 		case EIO:
869 		case ENOLINK:
870 		case ENOSPC:
871 		case ENOSTR:
872 		case ENXIO:
873 		case EPIPE:
874 		case ERANGE:
875 		case EFAULT:
876 		case EROFS:
877 		case EINVAL:
878 			zfs_error_aux(hdl, "%s", strerror(errno));
879 			return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
880 
881 		default:
882 			return (zfs_standard_error(hdl, errno, errbuf));
883 		}
884 	}
885 
886 	if (debugnv != NULL) {
887 		fnvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg);
888 		fnvlist_free(thisdbg);
889 	}
890 
891 	return (0);
892 }
893 
894 static void
895 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
896 {
897 	assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
898 
899 	/*
900 	 * zfs_send() only sets snapholds for sends that need them,
901 	 * e.g. replication and doall.
902 	 */
903 	if (sdd->snapholds == NULL)
904 		return;
905 
906 	fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
907 }
908 
909 int
910 zfs_send_progress(zfs_handle_t *zhp, int fd, uint64_t *bytes_written,
911     uint64_t *blocks_visited)
912 {
913 	zfs_cmd_t zc = {"\0"};
914 
915 	if (bytes_written != NULL)
916 		*bytes_written = 0;
917 	if (blocks_visited != NULL)
918 		*blocks_visited = 0;
919 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
920 	zc.zc_cookie = fd;
921 	if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
922 		return (errno);
923 	if (bytes_written != NULL)
924 		*bytes_written = zc.zc_cookie;
925 	if (blocks_visited != NULL)
926 		*blocks_visited = zc.zc_objset_type;
927 	return (0);
928 }
929 
930 static void *
931 send_progress_thread(void *arg)
932 {
933 	progress_arg_t *pa = arg;
934 	zfs_handle_t *zhp = pa->pa_zhp;
935 	uint64_t bytes;
936 	uint64_t blocks;
937 	uint64_t total = pa->pa_size / 100;
938 	char buf[16];
939 	time_t t;
940 	struct tm tm;
941 	int err;
942 
943 	if (!pa->pa_parsable && pa->pa_verbosity != 0) {
944 		(void) fprintf(stderr,
945 		    "TIME       %s   %sSNAPSHOT %s\n",
946 		    pa->pa_estimate ? "BYTES" : " SENT",
947 		    pa->pa_verbosity >= 2 ? "   BLOCKS    " : "",
948 		    zhp->zfs_name);
949 	}
950 
951 	/*
952 	 * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
953 	 */
954 	for (;;) {
955 		(void) sleep(1);
956 		if ((err = zfs_send_progress(zhp, pa->pa_fd, &bytes,
957 		    &blocks)) != 0) {
958 			if (err == EINTR || err == ENOENT)
959 				return ((void *)0);
960 			return ((void *)(uintptr_t)err);
961 		}
962 
963 		(void) time(&t);
964 		localtime_r(&t, &tm);
965 
966 		if (pa->pa_astitle) {
967 			char buf_bytes[16];
968 			char buf_size[16];
969 			int pct;
970 			zfs_nicenum(bytes, buf_bytes, sizeof (buf_bytes));
971 			zfs_nicenum(pa->pa_size, buf_size, sizeof (buf_size));
972 			pct = (total > 0) ? bytes / total : 100;
973 			zfs_setproctitle("sending %s (%d%%: %s/%s)",
974 			    zhp->zfs_name, MIN(pct, 100), buf_bytes, buf_size);
975 		}
976 
977 		if (pa->pa_verbosity >= 2 && pa->pa_parsable) {
978 			(void) fprintf(stderr,
979 			    "%02d:%02d:%02d\t%llu\t%llu\t%s\n",
980 			    tm.tm_hour, tm.tm_min, tm.tm_sec,
981 			    (u_longlong_t)bytes, (u_longlong_t)blocks,
982 			    zhp->zfs_name);
983 		} else if (pa->pa_verbosity >= 2) {
984 			zfs_nicenum(bytes, buf, sizeof (buf));
985 			(void) fprintf(stderr,
986 			    "%02d:%02d:%02d   %5s    %8llu    %s\n",
987 			    tm.tm_hour, tm.tm_min, tm.tm_sec,
988 			    buf, (u_longlong_t)blocks, zhp->zfs_name);
989 		} else if (pa->pa_parsable) {
990 			(void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
991 			    tm.tm_hour, tm.tm_min, tm.tm_sec,
992 			    (u_longlong_t)bytes, zhp->zfs_name);
993 		} else if (pa->pa_verbosity != 0) {
994 			zfs_nicebytes(bytes, buf, sizeof (buf));
995 			(void) fprintf(stderr, "%02d:%02d:%02d   %5s   %s\n",
996 			    tm.tm_hour, tm.tm_min, tm.tm_sec,
997 			    buf, zhp->zfs_name);
998 		}
999 	}
1000 }
1001 
1002 static boolean_t
1003 send_progress_thread_exit(libzfs_handle_t *hdl, pthread_t ptid)
1004 {
1005 	void *status = NULL;
1006 	(void) pthread_cancel(ptid);
1007 	(void) pthread_join(ptid, &status);
1008 	int error = (int)(uintptr_t)status;
1009 	if (error != 0 && status != PTHREAD_CANCELED)
1010 		return (zfs_standard_error(hdl, error,
1011 		    dgettext(TEXT_DOMAIN, "progress thread exited nonzero")));
1012 	else
1013 		return (B_FALSE);
1014 }
1015 
1016 static void
1017 send_print_verbose(FILE *fout, const char *tosnap, const char *fromsnap,
1018     uint64_t size, boolean_t parsable)
1019 {
1020 	if (parsable) {
1021 		if (fromsnap != NULL) {
1022 			(void) fprintf(fout, dgettext(TEXT_DOMAIN,
1023 			    "incremental\t%s\t%s"), fromsnap, tosnap);
1024 		} else {
1025 /*
1026  * Workaround for GCC 12+ with UBSan enabled deficencies.
1027  *
1028  * GCC 12+ invoked with -fsanitize=undefined incorrectly reports the code
1029  * below as violating -Wformat-overflow.
1030  */
1031 #if defined(__GNUC__) && !defined(__clang__) && \
1032 	defined(ZFS_UBSAN_ENABLED) && defined(HAVE_FORMAT_OVERFLOW)
1033 #pragma GCC diagnostic push
1034 #pragma GCC diagnostic ignored "-Wformat-overflow"
1035 #endif
1036 			(void) fprintf(fout, dgettext(TEXT_DOMAIN,
1037 			    "full\t%s"), tosnap);
1038 #if defined(__GNUC__) && !defined(__clang__) && \
1039 	defined(ZFS_UBSAN_ENABLED) && defined(HAVE_FORMAT_OVERFLOW)
1040 #pragma GCC diagnostic pop
1041 #endif
1042 		}
1043 		(void) fprintf(fout, "\t%llu", (longlong_t)size);
1044 	} else {
1045 		if (fromsnap != NULL) {
1046 			if (strchr(fromsnap, '@') == NULL &&
1047 			    strchr(fromsnap, '#') == NULL) {
1048 				(void) fprintf(fout, dgettext(TEXT_DOMAIN,
1049 				    "send from @%s to %s"), fromsnap, tosnap);
1050 			} else {
1051 				(void) fprintf(fout, dgettext(TEXT_DOMAIN,
1052 				    "send from %s to %s"), fromsnap, tosnap);
1053 			}
1054 		} else {
1055 			(void) fprintf(fout, dgettext(TEXT_DOMAIN,
1056 			    "full send of %s"), tosnap);
1057 		}
1058 		if (size != 0) {
1059 			char buf[16];
1060 			zfs_nicebytes(size, buf, sizeof (buf));
1061 /*
1062  * Workaround for GCC 12+ with UBSan enabled deficencies.
1063  *
1064  * GCC 12+ invoked with -fsanitize=undefined incorrectly reports the code
1065  * below as violating -Wformat-overflow.
1066  */
1067 #if defined(__GNUC__) && !defined(__clang__) && \
1068 	defined(ZFS_UBSAN_ENABLED) && defined(HAVE_FORMAT_OVERFLOW)
1069 #pragma GCC diagnostic push
1070 #pragma GCC diagnostic ignored "-Wformat-overflow"
1071 #endif
1072 			(void) fprintf(fout, dgettext(TEXT_DOMAIN,
1073 			    " estimated size is %s"), buf);
1074 #if defined(__GNUC__) && !defined(__clang__) && \
1075 	defined(ZFS_UBSAN_ENABLED) && defined(HAVE_FORMAT_OVERFLOW)
1076 #pragma GCC diagnostic pop
1077 #endif
1078 		}
1079 	}
1080 	(void) fprintf(fout, "\n");
1081 }
1082 
1083 /*
1084  * Send a single filesystem snapshot, updating the send dump data.
1085  * This interface is intended for use as a zfs_iter_snapshots_sorted visitor.
1086  */
1087 static int
1088 dump_snapshot(zfs_handle_t *zhp, void *arg)
1089 {
1090 	send_dump_data_t *sdd = arg;
1091 	progress_arg_t pa = { 0 };
1092 	pthread_t tid;
1093 	char *thissnap;
1094 	enum lzc_send_flags flags = 0;
1095 	int err;
1096 	boolean_t isfromsnap, istosnap, fromorigin;
1097 	boolean_t exclude = B_FALSE;
1098 	FILE *fout = sdd->std_out ? stdout : stderr;
1099 
1100 	err = 0;
1101 	thissnap = strchr(zhp->zfs_name, '@') + 1;
1102 	isfromsnap = (sdd->fromsnap != NULL &&
1103 	    strcmp(sdd->fromsnap, thissnap) == 0);
1104 
1105 	if (!sdd->seenfrom && isfromsnap) {
1106 		gather_holds(zhp, sdd);
1107 		sdd->seenfrom = B_TRUE;
1108 		(void) strlcpy(sdd->prevsnap, thissnap, sizeof (sdd->prevsnap));
1109 		sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1110 		zfs_close(zhp);
1111 		return (0);
1112 	}
1113 
1114 	if (sdd->seento || !sdd->seenfrom) {
1115 		zfs_close(zhp);
1116 		return (0);
1117 	}
1118 
1119 	istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1120 	if (istosnap)
1121 		sdd->seento = B_TRUE;
1122 
1123 	if (sdd->large_block)
1124 		flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1125 	if (sdd->embed_data)
1126 		flags |= LZC_SEND_FLAG_EMBED_DATA;
1127 	if (sdd->compress)
1128 		flags |= LZC_SEND_FLAG_COMPRESS;
1129 	if (sdd->raw)
1130 		flags |= LZC_SEND_FLAG_RAW;
1131 
1132 	if (!sdd->doall && !isfromsnap && !istosnap) {
1133 		if (sdd->replicate) {
1134 			char *snapname;
1135 			nvlist_t *snapprops;
1136 			/*
1137 			 * Filter out all intermediate snapshots except origin
1138 			 * snapshots needed to replicate clones.
1139 			 */
1140 			nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1141 			    zhp->zfs_dmustats.dds_guid, &snapname);
1142 
1143 			if (nvfs != NULL) {
1144 				snapprops = fnvlist_lookup_nvlist(nvfs,
1145 				    "snapprops");
1146 				snapprops = fnvlist_lookup_nvlist(snapprops,
1147 				    thissnap);
1148 				exclude = !nvlist_exists(snapprops,
1149 				    "is_clone_origin");
1150 			}
1151 		} else {
1152 			exclude = B_TRUE;
1153 		}
1154 	}
1155 
1156 	/*
1157 	 * If a filter function exists, call it to determine whether
1158 	 * this snapshot will be sent.
1159 	 */
1160 	if (exclude || (sdd->filter_cb != NULL &&
1161 	    sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1162 		/*
1163 		 * This snapshot is filtered out.  Don't send it, and don't
1164 		 * set prevsnap_obj, so it will be as if this snapshot didn't
1165 		 * exist, and the next accepted snapshot will be sent as
1166 		 * an incremental from the last accepted one, or as the
1167 		 * first (and full) snapshot in the case of a replication,
1168 		 * non-incremental send.
1169 		 */
1170 		zfs_close(zhp);
1171 		return (0);
1172 	}
1173 
1174 	gather_holds(zhp, sdd);
1175 	fromorigin = sdd->prevsnap[0] == '\0' &&
1176 	    (sdd->fromorigin || sdd->replicate);
1177 
1178 	if (sdd->verbosity != 0) {
1179 		uint64_t size = 0;
1180 		char fromds[ZFS_MAX_DATASET_NAME_LEN];
1181 
1182 		if (sdd->prevsnap[0] != '\0') {
1183 			(void) strlcpy(fromds, zhp->zfs_name, sizeof (fromds));
1184 			*(strchr(fromds, '@') + 1) = '\0';
1185 			(void) strlcat(fromds, sdd->prevsnap, sizeof (fromds));
1186 		}
1187 		if (zfs_send_space(zhp, zhp->zfs_name,
1188 		    sdd->prevsnap[0] ? fromds : NULL, flags, &size) == 0) {
1189 			send_print_verbose(fout, zhp->zfs_name,
1190 			    sdd->prevsnap[0] ? sdd->prevsnap : NULL,
1191 			    size, sdd->parsable);
1192 			sdd->size += size;
1193 		}
1194 	}
1195 
1196 	if (!sdd->dryrun) {
1197 		/*
1198 		 * If progress reporting is requested, spawn a new thread to
1199 		 * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1200 		 */
1201 		if (sdd->progress || sdd->progressastitle) {
1202 			pa.pa_zhp = zhp;
1203 			pa.pa_fd = sdd->outfd;
1204 			pa.pa_parsable = sdd->parsable;
1205 			pa.pa_estimate = B_FALSE;
1206 			pa.pa_verbosity = sdd->verbosity;
1207 			pa.pa_size = sdd->size;
1208 			pa.pa_astitle = sdd->progressastitle;
1209 
1210 			if ((err = pthread_create(&tid, NULL,
1211 			    send_progress_thread, &pa)) != 0) {
1212 				zfs_close(zhp);
1213 				return (err);
1214 			}
1215 		}
1216 
1217 		err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1218 		    fromorigin, sdd->outfd, flags, sdd->debugnv);
1219 
1220 		if ((sdd->progress || sdd->progressastitle) &&
1221 		    send_progress_thread_exit(zhp->zfs_hdl, tid))
1222 			return (-1);
1223 	}
1224 
1225 	(void) strlcpy(sdd->prevsnap, thissnap, sizeof (sdd->prevsnap));
1226 	sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1227 	zfs_close(zhp);
1228 	return (err);
1229 }
1230 
1231 /*
1232  * Send all snapshots for a filesystem, updating the send dump data.
1233  */
1234 static int
1235 dump_filesystem(zfs_handle_t *zhp, send_dump_data_t *sdd)
1236 {
1237 	int rv = 0;
1238 	boolean_t missingfrom = B_FALSE;
1239 	zfs_cmd_t zc = {"\0"};
1240 	uint64_t min_txg = 0, max_txg = 0;
1241 
1242 	/*
1243 	 * Make sure the tosnap exists.
1244 	 */
1245 	(void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1246 	    zhp->zfs_name, sdd->tosnap);
1247 	if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1248 		(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1249 		    "WARNING: could not send %s@%s: does not exist\n"),
1250 		    zhp->zfs_name, sdd->tosnap);
1251 		sdd->err = B_TRUE;
1252 		return (0);
1253 	}
1254 
1255 	/*
1256 	 * If this fs does not have fromsnap, and we're doing
1257 	 * recursive, we need to send a full stream from the
1258 	 * beginning (or an incremental from the origin if this
1259 	 * is a clone).  If we're doing non-recursive, then let
1260 	 * them get the error.
1261 	 */
1262 	if (sdd->replicate && sdd->fromsnap) {
1263 		/*
1264 		 * Make sure the fromsnap exists.
1265 		 */
1266 		(void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1267 		    zhp->zfs_name, sdd->fromsnap);
1268 		if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_OBJSET_STATS, &zc) != 0)
1269 			missingfrom = B_TRUE;
1270 	}
1271 
1272 	sdd->seenfrom = sdd->seento = B_FALSE;
1273 	sdd->prevsnap[0] = '\0';
1274 	sdd->prevsnap_obj = 0;
1275 	if (sdd->fromsnap == NULL || missingfrom)
1276 		sdd->seenfrom = B_TRUE;
1277 
1278 	/*
1279 	 * Iterate through all snapshots and process the ones we will be
1280 	 * sending. If we only have a "from" and "to" snapshot to deal
1281 	 * with, we can avoid iterating through all the other snapshots.
1282 	 */
1283 	if (sdd->doall || sdd->replicate || sdd->tosnap == NULL) {
1284 		if (!sdd->replicate) {
1285 			if (sdd->fromsnap != NULL) {
1286 				min_txg = get_snap_txg(zhp->zfs_hdl,
1287 				    zhp->zfs_name, sdd->fromsnap);
1288 			}
1289 			if (sdd->tosnap != NULL) {
1290 				max_txg = get_snap_txg(zhp->zfs_hdl,
1291 				    zhp->zfs_name, sdd->tosnap);
1292 			}
1293 		}
1294 		rv = zfs_iter_snapshots_sorted(zhp, 0, dump_snapshot, sdd,
1295 		    min_txg, max_txg);
1296 	} else {
1297 		char snapname[MAXPATHLEN] = { 0 };
1298 		zfs_handle_t *snap;
1299 
1300 		/* Dump fromsnap. */
1301 		if (!sdd->seenfrom) {
1302 			(void) snprintf(snapname, sizeof (snapname),
1303 			    "%s@%s", zhp->zfs_name, sdd->fromsnap);
1304 			snap = zfs_open(zhp->zfs_hdl, snapname,
1305 			    ZFS_TYPE_SNAPSHOT);
1306 			if (snap != NULL)
1307 				rv = dump_snapshot(snap, sdd);
1308 			else
1309 				rv = errno;
1310 		}
1311 
1312 		/* Dump tosnap. */
1313 		if (rv == 0) {
1314 			(void) snprintf(snapname, sizeof (snapname),
1315 			    "%s@%s", zhp->zfs_name, sdd->tosnap);
1316 			snap = zfs_open(zhp->zfs_hdl, snapname,
1317 			    ZFS_TYPE_SNAPSHOT);
1318 			if (snap != NULL)
1319 				rv = dump_snapshot(snap, sdd);
1320 			else
1321 				rv = errno;
1322 		}
1323 	}
1324 
1325 	if (!sdd->seenfrom) {
1326 		(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1327 		    "WARNING: could not send %s@%s:\n"
1328 		    "incremental source (%s@%s) does not exist\n"),
1329 		    zhp->zfs_name, sdd->tosnap,
1330 		    zhp->zfs_name, sdd->fromsnap);
1331 		sdd->err = B_TRUE;
1332 	} else if (!sdd->seento) {
1333 		if (sdd->fromsnap) {
1334 			(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1335 			    "WARNING: could not send %s@%s:\n"
1336 			    "incremental source (%s@%s) "
1337 			    "is not earlier than it\n"),
1338 			    zhp->zfs_name, sdd->tosnap,
1339 			    zhp->zfs_name, sdd->fromsnap);
1340 		} else {
1341 			(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1342 			    "WARNING: "
1343 			    "could not send %s@%s: does not exist\n"),
1344 			    zhp->zfs_name, sdd->tosnap);
1345 		}
1346 		sdd->err = B_TRUE;
1347 	}
1348 
1349 	return (rv);
1350 }
1351 
1352 /*
1353  * Send all snapshots for all filesystems in sdd.
1354  */
1355 static int
1356 dump_filesystems(zfs_handle_t *rzhp, send_dump_data_t *sdd)
1357 {
1358 	nvpair_t *fspair;
1359 	boolean_t needagain, progress;
1360 
1361 	if (!sdd->replicate)
1362 		return (dump_filesystem(rzhp, sdd));
1363 
1364 	/* Mark the clone origin snapshots. */
1365 	for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1366 	    fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1367 		nvlist_t *nvfs;
1368 		uint64_t origin_guid = 0;
1369 
1370 		nvfs = fnvpair_value_nvlist(fspair);
1371 		(void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1372 		if (origin_guid != 0) {
1373 			char *snapname;
1374 			nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1375 			    origin_guid, &snapname);
1376 			if (origin_nv != NULL) {
1377 				nvlist_t *snapprops;
1378 				snapprops = fnvlist_lookup_nvlist(origin_nv,
1379 				    "snapprops");
1380 				snapprops = fnvlist_lookup_nvlist(snapprops,
1381 				    snapname);
1382 				fnvlist_add_boolean(snapprops,
1383 				    "is_clone_origin");
1384 			}
1385 		}
1386 	}
1387 again:
1388 	needagain = progress = B_FALSE;
1389 	for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1390 	    fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1391 		nvlist_t *fslist, *parent_nv;
1392 		char *fsname;
1393 		zfs_handle_t *zhp;
1394 		int err;
1395 		uint64_t origin_guid = 0;
1396 		uint64_t parent_guid = 0;
1397 
1398 		fslist = fnvpair_value_nvlist(fspair);
1399 		if (nvlist_lookup_boolean(fslist, "sent") == 0)
1400 			continue;
1401 
1402 		fsname = fnvlist_lookup_string(fslist, "name");
1403 		(void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1404 		(void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1405 		    &parent_guid);
1406 
1407 		if (parent_guid != 0) {
1408 			parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1409 			if (!nvlist_exists(parent_nv, "sent")) {
1410 				/* Parent has not been sent; skip this one. */
1411 				needagain = B_TRUE;
1412 				continue;
1413 			}
1414 		}
1415 
1416 		if (origin_guid != 0) {
1417 			nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1418 			    origin_guid, NULL);
1419 			if (origin_nv != NULL &&
1420 			    !nvlist_exists(origin_nv, "sent")) {
1421 				/*
1422 				 * Origin has not been sent yet;
1423 				 * skip this clone.
1424 				 */
1425 				needagain = B_TRUE;
1426 				continue;
1427 			}
1428 		}
1429 
1430 		zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1431 		if (zhp == NULL)
1432 			return (-1);
1433 		err = dump_filesystem(zhp, sdd);
1434 		fnvlist_add_boolean(fslist, "sent");
1435 		progress = B_TRUE;
1436 		zfs_close(zhp);
1437 		if (err)
1438 			return (err);
1439 	}
1440 	if (needagain) {
1441 		assert(progress);
1442 		goto again;
1443 	}
1444 
1445 	/* Clean out the sent flags in case we reuse this fss. */
1446 	for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1447 	    fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1448 		nvlist_t *fslist;
1449 
1450 		fslist = fnvpair_value_nvlist(fspair);
1451 		(void) nvlist_remove_all(fslist, "sent");
1452 	}
1453 
1454 	return (0);
1455 }
1456 
1457 nvlist_t *
1458 zfs_send_resume_token_to_nvlist(libzfs_handle_t *hdl, const char *token)
1459 {
1460 	unsigned int version;
1461 	int nread, i;
1462 	unsigned long long checksum, packed_len;
1463 
1464 	/*
1465 	 * Decode token header, which is:
1466 	 *   <token version>-<checksum of payload>-<uncompressed payload length>
1467 	 * Note that the only supported token version is 1.
1468 	 */
1469 	nread = sscanf(token, "%u-%llx-%llx-",
1470 	    &version, &checksum, &packed_len);
1471 	if (nread != 3) {
1472 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1473 		    "resume token is corrupt (invalid format)"));
1474 		return (NULL);
1475 	}
1476 
1477 	if (version != ZFS_SEND_RESUME_TOKEN_VERSION) {
1478 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1479 		    "resume token is corrupt (invalid version %u)"),
1480 		    version);
1481 		return (NULL);
1482 	}
1483 
1484 	/* Convert hexadecimal representation to binary. */
1485 	token = strrchr(token, '-') + 1;
1486 	int len = strlen(token) / 2;
1487 	unsigned char *compressed = zfs_alloc(hdl, len);
1488 	for (i = 0; i < len; i++) {
1489 		nread = sscanf(token + i * 2, "%2hhx", compressed + i);
1490 		if (nread != 1) {
1491 			free(compressed);
1492 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1493 			    "resume token is corrupt "
1494 			    "(payload is not hex-encoded)"));
1495 			return (NULL);
1496 		}
1497 	}
1498 
1499 	/* Verify checksum. */
1500 	zio_cksum_t cksum;
1501 	fletcher_4_native_varsize(compressed, len, &cksum);
1502 	if (cksum.zc_word[0] != checksum) {
1503 		free(compressed);
1504 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1505 		    "resume token is corrupt (incorrect checksum)"));
1506 		return (NULL);
1507 	}
1508 
1509 	/* Uncompress. */
1510 	void *packed = zfs_alloc(hdl, packed_len);
1511 	uLongf packed_len_long = packed_len;
1512 	if (uncompress(packed, &packed_len_long, compressed, len) != Z_OK ||
1513 	    packed_len_long != packed_len) {
1514 		free(packed);
1515 		free(compressed);
1516 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1517 		    "resume token is corrupt (decompression failed)"));
1518 		return (NULL);
1519 	}
1520 
1521 	/* Unpack nvlist. */
1522 	nvlist_t *nv;
1523 	int error = nvlist_unpack(packed, packed_len, &nv, KM_SLEEP);
1524 	free(packed);
1525 	free(compressed);
1526 	if (error != 0) {
1527 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1528 		    "resume token is corrupt (nvlist_unpack failed)"));
1529 		return (NULL);
1530 	}
1531 	return (nv);
1532 }
1533 
1534 static enum lzc_send_flags
1535 lzc_flags_from_sendflags(const sendflags_t *flags)
1536 {
1537 	enum lzc_send_flags lzc_flags = 0;
1538 
1539 	if (flags->largeblock)
1540 		lzc_flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1541 	if (flags->embed_data)
1542 		lzc_flags |= LZC_SEND_FLAG_EMBED_DATA;
1543 	if (flags->compress)
1544 		lzc_flags |= LZC_SEND_FLAG_COMPRESS;
1545 	if (flags->raw)
1546 		lzc_flags |= LZC_SEND_FLAG_RAW;
1547 	if (flags->saved)
1548 		lzc_flags |= LZC_SEND_FLAG_SAVED;
1549 
1550 	return (lzc_flags);
1551 }
1552 
1553 static int
1554 estimate_size(zfs_handle_t *zhp, const char *from, int fd, sendflags_t *flags,
1555     uint64_t resumeobj, uint64_t resumeoff, uint64_t bytes,
1556     const char *redactbook, char *errbuf, uint64_t *sizep)
1557 {
1558 	uint64_t size;
1559 	FILE *fout = flags->dryrun ? stdout : stderr;
1560 	progress_arg_t pa = { 0 };
1561 	int err = 0;
1562 	pthread_t ptid;
1563 
1564 	if (flags->progress || flags->progressastitle) {
1565 		pa.pa_zhp = zhp;
1566 		pa.pa_fd = fd;
1567 		pa.pa_parsable = flags->parsable;
1568 		pa.pa_estimate = B_TRUE;
1569 		pa.pa_verbosity = flags->verbosity;
1570 
1571 		err = pthread_create(&ptid, NULL,
1572 		    send_progress_thread, &pa);
1573 		if (err != 0) {
1574 			zfs_error_aux(zhp->zfs_hdl, "%s", strerror(errno));
1575 			return (zfs_error(zhp->zfs_hdl,
1576 			    EZFS_THREADCREATEFAILED, errbuf));
1577 		}
1578 	}
1579 
1580 	err = lzc_send_space_resume_redacted(zhp->zfs_name, from,
1581 	    lzc_flags_from_sendflags(flags), resumeobj, resumeoff, bytes,
1582 	    redactbook, fd, &size);
1583 	*sizep = size;
1584 
1585 	if ((flags->progress || flags->progressastitle) &&
1586 	    send_progress_thread_exit(zhp->zfs_hdl, ptid))
1587 		return (-1);
1588 
1589 	if (!flags->progress && !flags->parsable)
1590 		return (err);
1591 
1592 	if (err != 0) {
1593 		zfs_error_aux(zhp->zfs_hdl, "%s", strerror(err));
1594 		return (zfs_error(zhp->zfs_hdl, EZFS_BADBACKUP,
1595 		    errbuf));
1596 	}
1597 	send_print_verbose(fout, zhp->zfs_name, from, size,
1598 	    flags->parsable);
1599 
1600 	if (flags->parsable) {
1601 		(void) fprintf(fout, "size\t%llu\n", (longlong_t)size);
1602 	} else {
1603 		char buf[16];
1604 		zfs_nicenum(size, buf, sizeof (buf));
1605 		(void) fprintf(fout, dgettext(TEXT_DOMAIN,
1606 		    "total estimated size is %s\n"), buf);
1607 	}
1608 	return (0);
1609 }
1610 
1611 static boolean_t
1612 redact_snaps_contains(const uint64_t *snaps, uint64_t num_snaps, uint64_t guid)
1613 {
1614 	for (int i = 0; i < num_snaps; i++) {
1615 		if (snaps[i] == guid)
1616 			return (B_TRUE);
1617 	}
1618 	return (B_FALSE);
1619 }
1620 
1621 static boolean_t
1622 redact_snaps_equal(const uint64_t *snaps1, uint64_t num_snaps1,
1623     const uint64_t *snaps2, uint64_t num_snaps2)
1624 {
1625 	if (num_snaps1 != num_snaps2)
1626 		return (B_FALSE);
1627 	for (int i = 0; i < num_snaps1; i++) {
1628 		if (!redact_snaps_contains(snaps2, num_snaps2, snaps1[i]))
1629 			return (B_FALSE);
1630 	}
1631 	return (B_TRUE);
1632 }
1633 
1634 static int
1635 get_bookmarks(const char *path, nvlist_t **bmarksp)
1636 {
1637 	nvlist_t *props = fnvlist_alloc();
1638 	int error;
1639 
1640 	fnvlist_add_boolean(props, "redact_complete");
1641 	fnvlist_add_boolean(props, zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS));
1642 	error = lzc_get_bookmarks(path, props, bmarksp);
1643 	fnvlist_free(props);
1644 	return (error);
1645 }
1646 
1647 static nvpair_t *
1648 find_redact_pair(nvlist_t *bmarks, const uint64_t *redact_snap_guids,
1649     int num_redact_snaps)
1650 {
1651 	nvpair_t *pair;
1652 
1653 	for (pair = nvlist_next_nvpair(bmarks, NULL); pair;
1654 	    pair = nvlist_next_nvpair(bmarks, pair)) {
1655 
1656 		nvlist_t *bmark = fnvpair_value_nvlist(pair);
1657 		nvlist_t *vallist = fnvlist_lookup_nvlist(bmark,
1658 		    zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS));
1659 		uint_t len = 0;
1660 		uint64_t *bmarksnaps = fnvlist_lookup_uint64_array(vallist,
1661 		    ZPROP_VALUE, &len);
1662 		if (redact_snaps_equal(redact_snap_guids,
1663 		    num_redact_snaps, bmarksnaps, len)) {
1664 			break;
1665 		}
1666 	}
1667 	return (pair);
1668 }
1669 
1670 static boolean_t
1671 get_redact_complete(nvpair_t *pair)
1672 {
1673 	nvlist_t *bmark = fnvpair_value_nvlist(pair);
1674 	nvlist_t *vallist = fnvlist_lookup_nvlist(bmark, "redact_complete");
1675 	boolean_t complete = fnvlist_lookup_boolean_value(vallist,
1676 	    ZPROP_VALUE);
1677 
1678 	return (complete);
1679 }
1680 
1681 /*
1682  * Check that the list of redaction snapshots in the bookmark matches the send
1683  * we're resuming, and return whether or not it's complete.
1684  *
1685  * Note that the caller needs to free the contents of *bookname with free() if
1686  * this function returns successfully.
1687  */
1688 static int
1689 find_redact_book(libzfs_handle_t *hdl, const char *path,
1690     const uint64_t *redact_snap_guids, int num_redact_snaps,
1691     char **bookname)
1692 {
1693 	char errbuf[ERRBUFLEN];
1694 	nvlist_t *bmarks;
1695 
1696 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1697 	    "cannot resume send"));
1698 
1699 	int error = get_bookmarks(path, &bmarks);
1700 	if (error != 0) {
1701 		if (error == ESRCH) {
1702 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1703 			    "nonexistent redaction bookmark provided"));
1704 		} else if (error == ENOENT) {
1705 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1706 			    "dataset to be sent no longer exists"));
1707 		} else {
1708 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1709 			    "unknown error: %s"), strerror(error));
1710 		}
1711 		return (zfs_error(hdl, EZFS_BADPROP, errbuf));
1712 	}
1713 	nvpair_t *pair = find_redact_pair(bmarks, redact_snap_guids,
1714 	    num_redact_snaps);
1715 	if (pair == NULL)  {
1716 		fnvlist_free(bmarks);
1717 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1718 		    "no appropriate redaction bookmark exists"));
1719 		return (zfs_error(hdl, EZFS_BADPROP, errbuf));
1720 	}
1721 	boolean_t complete = get_redact_complete(pair);
1722 	if (!complete) {
1723 		fnvlist_free(bmarks);
1724 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1725 		    "incomplete redaction bookmark provided"));
1726 		return (zfs_error(hdl, EZFS_BADPROP, errbuf));
1727 	}
1728 	*bookname = strndup(nvpair_name(pair), ZFS_MAX_DATASET_NAME_LEN);
1729 	ASSERT3P(*bookname, !=, NULL);
1730 	fnvlist_free(bmarks);
1731 	return (0);
1732 }
1733 
1734 static enum lzc_send_flags
1735 lzc_flags_from_resume_nvl(nvlist_t *resume_nvl)
1736 {
1737 	enum lzc_send_flags lzc_flags = 0;
1738 
1739 	if (nvlist_exists(resume_nvl, "largeblockok"))
1740 		lzc_flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1741 	if (nvlist_exists(resume_nvl, "embedok"))
1742 		lzc_flags |= LZC_SEND_FLAG_EMBED_DATA;
1743 	if (nvlist_exists(resume_nvl, "compressok"))
1744 		lzc_flags |= LZC_SEND_FLAG_COMPRESS;
1745 	if (nvlist_exists(resume_nvl, "rawok"))
1746 		lzc_flags |= LZC_SEND_FLAG_RAW;
1747 	if (nvlist_exists(resume_nvl, "savedok"))
1748 		lzc_flags |= LZC_SEND_FLAG_SAVED;
1749 
1750 	return (lzc_flags);
1751 }
1752 
1753 static int
1754 zfs_send_resume_impl_cb_impl(libzfs_handle_t *hdl, sendflags_t *flags,
1755     int outfd, nvlist_t *resume_nvl)
1756 {
1757 	char errbuf[ERRBUFLEN];
1758 	char *toname;
1759 	char *fromname = NULL;
1760 	uint64_t resumeobj, resumeoff, toguid, fromguid, bytes;
1761 	zfs_handle_t *zhp;
1762 	int error = 0;
1763 	char name[ZFS_MAX_DATASET_NAME_LEN];
1764 	FILE *fout = (flags->verbosity > 0 && flags->dryrun) ? stdout : stderr;
1765 	uint64_t *redact_snap_guids = NULL;
1766 	int num_redact_snaps = 0;
1767 	char *redact_book = NULL;
1768 	uint64_t size = 0;
1769 
1770 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1771 	    "cannot resume send"));
1772 
1773 	if (flags->verbosity != 0) {
1774 		(void) fprintf(fout, dgettext(TEXT_DOMAIN,
1775 		    "resume token contents:\n"));
1776 		nvlist_print(fout, resume_nvl);
1777 	}
1778 
1779 	if (nvlist_lookup_string(resume_nvl, "toname", &toname) != 0 ||
1780 	    nvlist_lookup_uint64(resume_nvl, "object", &resumeobj) != 0 ||
1781 	    nvlist_lookup_uint64(resume_nvl, "offset", &resumeoff) != 0 ||
1782 	    nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 ||
1783 	    nvlist_lookup_uint64(resume_nvl, "toguid", &toguid) != 0) {
1784 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1785 		    "resume token is corrupt"));
1786 		return (zfs_error(hdl, EZFS_FAULT, errbuf));
1787 	}
1788 	fromguid = 0;
1789 	(void) nvlist_lookup_uint64(resume_nvl, "fromguid", &fromguid);
1790 
1791 	if (flags->saved) {
1792 		(void) strlcpy(name, toname, sizeof (name));
1793 	} else {
1794 		error = guid_to_name(hdl, toname, toguid, B_FALSE, name);
1795 		if (error != 0) {
1796 			if (zfs_dataset_exists(hdl, toname, ZFS_TYPE_DATASET)) {
1797 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1798 				    "'%s' is no longer the same snapshot "
1799 				    "used in the initial send"), toname);
1800 			} else {
1801 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1802 				    "'%s' used in the initial send no "
1803 				    "longer exists"), toname);
1804 			}
1805 			return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1806 		}
1807 	}
1808 
1809 	zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1810 	if (zhp == NULL) {
1811 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1812 		    "unable to access '%s'"), name);
1813 		return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1814 	}
1815 
1816 	if (nvlist_lookup_uint64_array(resume_nvl, "book_redact_snaps",
1817 	    &redact_snap_guids, (uint_t *)&num_redact_snaps) != 0) {
1818 		num_redact_snaps = -1;
1819 	}
1820 
1821 	if (fromguid != 0) {
1822 		if (guid_to_name_redact_snaps(hdl, toname, fromguid, B_TRUE,
1823 		    redact_snap_guids, num_redact_snaps, name) != 0) {
1824 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1825 			    "incremental source %#llx no longer exists"),
1826 			    (longlong_t)fromguid);
1827 			return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1828 		}
1829 		fromname = name;
1830 	}
1831 
1832 	redact_snap_guids = NULL;
1833 
1834 	if (nvlist_lookup_uint64_array(resume_nvl,
1835 	    zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS), &redact_snap_guids,
1836 	    (uint_t *)&num_redact_snaps) == 0) {
1837 		char path[ZFS_MAX_DATASET_NAME_LEN];
1838 
1839 		(void) strlcpy(path, toname, sizeof (path));
1840 		char *at = strchr(path, '@');
1841 		ASSERT3P(at, !=, NULL);
1842 
1843 		*at = '\0';
1844 
1845 		if ((error = find_redact_book(hdl, path, redact_snap_guids,
1846 		    num_redact_snaps, &redact_book)) != 0) {
1847 			return (error);
1848 		}
1849 	}
1850 
1851 	enum lzc_send_flags lzc_flags = lzc_flags_from_sendflags(flags) |
1852 	    lzc_flags_from_resume_nvl(resume_nvl);
1853 
1854 	if (flags->verbosity != 0 || flags->progressastitle) {
1855 		/*
1856 		 * Some of these may have come from the resume token, set them
1857 		 * here for size estimate purposes.
1858 		 */
1859 		sendflags_t tmpflags = *flags;
1860 		if (lzc_flags & LZC_SEND_FLAG_LARGE_BLOCK)
1861 			tmpflags.largeblock = B_TRUE;
1862 		if (lzc_flags & LZC_SEND_FLAG_COMPRESS)
1863 			tmpflags.compress = B_TRUE;
1864 		if (lzc_flags & LZC_SEND_FLAG_EMBED_DATA)
1865 			tmpflags.embed_data = B_TRUE;
1866 		if (lzc_flags & LZC_SEND_FLAG_RAW)
1867 			tmpflags.raw = B_TRUE;
1868 		if (lzc_flags & LZC_SEND_FLAG_SAVED)
1869 			tmpflags.saved = B_TRUE;
1870 		error = estimate_size(zhp, fromname, outfd, &tmpflags,
1871 		    resumeobj, resumeoff, bytes, redact_book, errbuf, &size);
1872 	}
1873 
1874 	if (!flags->dryrun) {
1875 		progress_arg_t pa = { 0 };
1876 		pthread_t tid;
1877 		/*
1878 		 * If progress reporting is requested, spawn a new thread to
1879 		 * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1880 		 */
1881 		if (flags->progress || flags->progressastitle) {
1882 			pa.pa_zhp = zhp;
1883 			pa.pa_fd = outfd;
1884 			pa.pa_parsable = flags->parsable;
1885 			pa.pa_estimate = B_FALSE;
1886 			pa.pa_verbosity = flags->verbosity;
1887 			pa.pa_size = size;
1888 			pa.pa_astitle = flags->progressastitle;
1889 
1890 			error = pthread_create(&tid, NULL,
1891 			    send_progress_thread, &pa);
1892 			if (error != 0) {
1893 				if (redact_book != NULL)
1894 					free(redact_book);
1895 				zfs_close(zhp);
1896 				return (error);
1897 			}
1898 		}
1899 
1900 		error = lzc_send_resume_redacted(zhp->zfs_name, fromname, outfd,
1901 		    lzc_flags, resumeobj, resumeoff, redact_book);
1902 		if (redact_book != NULL)
1903 			free(redact_book);
1904 
1905 		if ((flags->progressastitle || flags->progress) &&
1906 		    send_progress_thread_exit(hdl, tid)) {
1907 			zfs_close(zhp);
1908 			return (-1);
1909 		}
1910 
1911 		char errbuf[ERRBUFLEN];
1912 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1913 		    "warning: cannot send '%s'"), zhp->zfs_name);
1914 
1915 		zfs_close(zhp);
1916 
1917 		switch (error) {
1918 		case 0:
1919 			return (0);
1920 		case EACCES:
1921 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1922 			    "source key must be loaded"));
1923 			return (zfs_error(hdl, EZFS_CRYPTOFAILED, errbuf));
1924 		case ESRCH:
1925 			if (lzc_exists(zhp->zfs_name)) {
1926 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1927 				    "incremental source could not be found"));
1928 			}
1929 			return (zfs_error(hdl, EZFS_NOENT, errbuf));
1930 
1931 		case EXDEV:
1932 		case ENOENT:
1933 		case EDQUOT:
1934 		case EFBIG:
1935 		case EIO:
1936 		case ENOLINK:
1937 		case ENOSPC:
1938 		case ENOSTR:
1939 		case ENXIO:
1940 		case EPIPE:
1941 		case ERANGE:
1942 		case EFAULT:
1943 		case EROFS:
1944 			zfs_error_aux(hdl, "%s", strerror(errno));
1945 			return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1946 
1947 		default:
1948 			return (zfs_standard_error(hdl, errno, errbuf));
1949 		}
1950 	} else {
1951 		if (redact_book != NULL)
1952 			free(redact_book);
1953 	}
1954 
1955 	zfs_close(zhp);
1956 
1957 	return (error);
1958 }
1959 
1960 struct zfs_send_resume_impl {
1961 	libzfs_handle_t *hdl;
1962 	sendflags_t *flags;
1963 	nvlist_t *resume_nvl;
1964 };
1965 
1966 static int
1967 zfs_send_resume_impl_cb(int outfd, void *arg)
1968 {
1969 	struct zfs_send_resume_impl *zsri = arg;
1970 	return (zfs_send_resume_impl_cb_impl(zsri->hdl, zsri->flags, outfd,
1971 	    zsri->resume_nvl));
1972 }
1973 
1974 static int
1975 zfs_send_resume_impl(libzfs_handle_t *hdl, sendflags_t *flags, int outfd,
1976     nvlist_t *resume_nvl)
1977 {
1978 	struct zfs_send_resume_impl zsri = {
1979 		.hdl = hdl,
1980 		.flags = flags,
1981 		.resume_nvl = resume_nvl,
1982 	};
1983 	return (lzc_send_wrapper(zfs_send_resume_impl_cb, outfd, &zsri));
1984 }
1985 
1986 int
1987 zfs_send_resume(libzfs_handle_t *hdl, sendflags_t *flags, int outfd,
1988     const char *resume_token)
1989 {
1990 	int ret;
1991 	char errbuf[ERRBUFLEN];
1992 	nvlist_t *resume_nvl;
1993 
1994 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1995 	    "cannot resume send"));
1996 
1997 	resume_nvl = zfs_send_resume_token_to_nvlist(hdl, resume_token);
1998 	if (resume_nvl == NULL) {
1999 		/*
2000 		 * zfs_error_aux has already been set by
2001 		 * zfs_send_resume_token_to_nvlist()
2002 		 */
2003 		return (zfs_error(hdl, EZFS_FAULT, errbuf));
2004 	}
2005 
2006 	ret = zfs_send_resume_impl(hdl, flags, outfd, resume_nvl);
2007 	fnvlist_free(resume_nvl);
2008 
2009 	return (ret);
2010 }
2011 
2012 int
2013 zfs_send_saved(zfs_handle_t *zhp, sendflags_t *flags, int outfd,
2014     const char *resume_token)
2015 {
2016 	int ret;
2017 	libzfs_handle_t *hdl = zhp->zfs_hdl;
2018 	nvlist_t *saved_nvl = NULL, *resume_nvl = NULL;
2019 	uint64_t saved_guid = 0, resume_guid = 0;
2020 	uint64_t obj = 0, off = 0, bytes = 0;
2021 	char token_buf[ZFS_MAXPROPLEN];
2022 	char errbuf[ERRBUFLEN];
2023 
2024 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2025 	    "saved send failed"));
2026 
2027 	ret = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN,
2028 	    token_buf, sizeof (token_buf), NULL, NULL, 0, B_TRUE);
2029 	if (ret != 0)
2030 		goto out;
2031 
2032 	saved_nvl = zfs_send_resume_token_to_nvlist(hdl, token_buf);
2033 	if (saved_nvl == NULL) {
2034 		/*
2035 		 * zfs_error_aux has already been set by
2036 		 * zfs_send_resume_token_to_nvlist()
2037 		 */
2038 		ret = zfs_error(hdl, EZFS_FAULT, errbuf);
2039 		goto out;
2040 	}
2041 
2042 	/*
2043 	 * If a resume token is provided we use the object and offset
2044 	 * from that instead of the default, which starts from the
2045 	 * beginning.
2046 	 */
2047 	if (resume_token != NULL) {
2048 		resume_nvl = zfs_send_resume_token_to_nvlist(hdl,
2049 		    resume_token);
2050 		if (resume_nvl == NULL) {
2051 			ret = zfs_error(hdl, EZFS_FAULT, errbuf);
2052 			goto out;
2053 		}
2054 
2055 		if (nvlist_lookup_uint64(resume_nvl, "object", &obj) != 0 ||
2056 		    nvlist_lookup_uint64(resume_nvl, "offset", &off) != 0 ||
2057 		    nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 ||
2058 		    nvlist_lookup_uint64(resume_nvl, "toguid",
2059 		    &resume_guid) != 0) {
2060 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2061 			    "provided resume token is corrupt"));
2062 			ret = zfs_error(hdl, EZFS_FAULT, errbuf);
2063 			goto out;
2064 		}
2065 
2066 		if (nvlist_lookup_uint64(saved_nvl, "toguid",
2067 		    &saved_guid)) {
2068 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2069 			    "dataset's resume token is corrupt"));
2070 			ret = zfs_error(hdl, EZFS_FAULT, errbuf);
2071 			goto out;
2072 		}
2073 
2074 		if (resume_guid != saved_guid) {
2075 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2076 			    "provided resume token does not match dataset"));
2077 			ret = zfs_error(hdl, EZFS_BADBACKUP, errbuf);
2078 			goto out;
2079 		}
2080 	}
2081 
2082 	(void) nvlist_remove_all(saved_nvl, "object");
2083 	fnvlist_add_uint64(saved_nvl, "object", obj);
2084 
2085 	(void) nvlist_remove_all(saved_nvl, "offset");
2086 	fnvlist_add_uint64(saved_nvl, "offset", off);
2087 
2088 	(void) nvlist_remove_all(saved_nvl, "bytes");
2089 	fnvlist_add_uint64(saved_nvl, "bytes", bytes);
2090 
2091 	(void) nvlist_remove_all(saved_nvl, "toname");
2092 	fnvlist_add_string(saved_nvl, "toname", zhp->zfs_name);
2093 
2094 	ret = zfs_send_resume_impl(hdl, flags, outfd, saved_nvl);
2095 
2096 out:
2097 	fnvlist_free(saved_nvl);
2098 	fnvlist_free(resume_nvl);
2099 	return (ret);
2100 }
2101 
2102 /*
2103  * This function informs the target system that the recursive send is complete.
2104  * The record is also expected in the case of a send -p.
2105  */
2106 static int
2107 send_conclusion_record(int fd, zio_cksum_t *zc)
2108 {
2109 	dmu_replay_record_t drr = { 0 };
2110 	drr.drr_type = DRR_END;
2111 	if (zc != NULL)
2112 		drr.drr_u.drr_end.drr_checksum = *zc;
2113 	if (write(fd, &drr, sizeof (drr)) == -1) {
2114 		return (errno);
2115 	}
2116 	return (0);
2117 }
2118 
2119 /*
2120  * This function is responsible for sending the records that contain the
2121  * necessary information for the target system's libzfs to be able to set the
2122  * properties of the filesystem being received, or to be able to prepare for
2123  * a recursive receive.
2124  *
2125  * The "zhp" argument is the handle of the snapshot we are sending
2126  * (the "tosnap").  The "from" argument is the short snapshot name (the part
2127  * after the @) of the incremental source.
2128  */
2129 static int
2130 send_prelim_records(zfs_handle_t *zhp, const char *from, int fd,
2131     boolean_t gather_props, boolean_t recursive, boolean_t verbose,
2132     boolean_t dryrun, boolean_t raw, boolean_t replicate, boolean_t skipmissing,
2133     boolean_t backup, boolean_t holds, boolean_t props, boolean_t doall,
2134     nvlist_t **fssp, avl_tree_t **fsavlp)
2135 {
2136 	int err = 0;
2137 	char *packbuf = NULL;
2138 	size_t buflen = 0;
2139 	zio_cksum_t zc = { {0} };
2140 	int featureflags = 0;
2141 	/* name of filesystem/volume that contains snapshot we are sending */
2142 	char tofs[ZFS_MAX_DATASET_NAME_LEN];
2143 	/* short name of snap we are sending */
2144 	const char *tosnap = "";
2145 
2146 	char errbuf[ERRBUFLEN];
2147 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2148 	    "warning: cannot send '%s'"), zhp->zfs_name);
2149 	if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM && zfs_prop_get_int(zhp,
2150 	    ZFS_PROP_VERSION) >= ZPL_VERSION_SA) {
2151 		featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
2152 	}
2153 
2154 	if (holds)
2155 		featureflags |= DMU_BACKUP_FEATURE_HOLDS;
2156 
2157 	(void) strlcpy(tofs, zhp->zfs_name, ZFS_MAX_DATASET_NAME_LEN);
2158 	char *at = strchr(tofs, '@');
2159 	if (at != NULL) {
2160 		*at = '\0';
2161 		tosnap = at + 1;
2162 	}
2163 
2164 	if (gather_props) {
2165 		nvlist_t *hdrnv = fnvlist_alloc();
2166 		nvlist_t *fss = NULL;
2167 
2168 		if (from != NULL)
2169 			fnvlist_add_string(hdrnv, "fromsnap", from);
2170 		fnvlist_add_string(hdrnv, "tosnap", tosnap);
2171 		if (!recursive)
2172 			fnvlist_add_boolean(hdrnv, "not_recursive");
2173 
2174 		if (raw) {
2175 			fnvlist_add_boolean(hdrnv, "raw");
2176 		}
2177 
2178 		if (gather_nvlist(zhp->zfs_hdl, tofs,
2179 		    from, tosnap, recursive, raw, doall, replicate, skipmissing,
2180 		    verbose, backup, holds, props, &fss, fsavlp) != 0) {
2181 			return (zfs_error(zhp->zfs_hdl, EZFS_BADBACKUP,
2182 			    errbuf));
2183 		}
2184 		/*
2185 		 * Do not allow the size of the properties list to exceed
2186 		 * the limit
2187 		 */
2188 		if ((fnvlist_size(fss) + fnvlist_size(hdrnv)) >
2189 		    zhp->zfs_hdl->libzfs_max_nvlist) {
2190 			(void) snprintf(errbuf, sizeof (errbuf),
2191 			    dgettext(TEXT_DOMAIN, "warning: cannot send '%s': "
2192 			    "the size of the list of snapshots and properties "
2193 			    "is too large to be received successfully.\n"
2194 			    "Select a smaller number of snapshots to send.\n"),
2195 			    zhp->zfs_name);
2196 			return (zfs_error(zhp->zfs_hdl, EZFS_NOSPC,
2197 			    errbuf));
2198 		}
2199 		fnvlist_add_nvlist(hdrnv, "fss", fss);
2200 		VERIFY0(nvlist_pack(hdrnv, &packbuf, &buflen, NV_ENCODE_XDR,
2201 		    0));
2202 		if (fssp != NULL) {
2203 			*fssp = fss;
2204 		} else {
2205 			fnvlist_free(fss);
2206 		}
2207 		fnvlist_free(hdrnv);
2208 	}
2209 
2210 	if (!dryrun) {
2211 		dmu_replay_record_t drr = { 0 };
2212 		/* write first begin record */
2213 		drr.drr_type = DRR_BEGIN;
2214 		drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
2215 		DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
2216 		    drr_versioninfo, DMU_COMPOUNDSTREAM);
2217 		DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
2218 		    drr_versioninfo, featureflags);
2219 		if (snprintf(drr.drr_u.drr_begin.drr_toname,
2220 		    sizeof (drr.drr_u.drr_begin.drr_toname), "%s@%s", tofs,
2221 		    tosnap) >= sizeof (drr.drr_u.drr_begin.drr_toname)) {
2222 			return (zfs_error(zhp->zfs_hdl, EZFS_BADBACKUP,
2223 			    errbuf));
2224 		}
2225 		drr.drr_payloadlen = buflen;
2226 
2227 		err = dump_record(&drr, packbuf, buflen, &zc, fd);
2228 		free(packbuf);
2229 		if (err != 0) {
2230 			zfs_error_aux(zhp->zfs_hdl, "%s", strerror(err));
2231 			return (zfs_error(zhp->zfs_hdl, EZFS_BADBACKUP,
2232 			    errbuf));
2233 		}
2234 		err = send_conclusion_record(fd, &zc);
2235 		if (err != 0) {
2236 			zfs_error_aux(zhp->zfs_hdl, "%s", strerror(err));
2237 			return (zfs_error(zhp->zfs_hdl, EZFS_BADBACKUP,
2238 			    errbuf));
2239 		}
2240 	}
2241 	return (0);
2242 }
2243 
2244 /*
2245  * Generate a send stream.  The "zhp" argument is the filesystem/volume
2246  * that contains the snapshot to send.  The "fromsnap" argument is the
2247  * short name (the part after the '@') of the snapshot that is the
2248  * incremental source to send from (if non-NULL).  The "tosnap" argument
2249  * is the short name of the snapshot to send.
2250  *
2251  * The content of the send stream is the snapshot identified by
2252  * 'tosnap'.  Incremental streams are requested in two ways:
2253  *     - from the snapshot identified by "fromsnap" (if non-null) or
2254  *     - from the origin of the dataset identified by zhp, which must
2255  *	 be a clone.  In this case, "fromsnap" is null and "fromorigin"
2256  *	 is TRUE.
2257  *
2258  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
2259  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
2260  * if "replicate" is set.  If "doall" is set, dump all the intermediate
2261  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
2262  * case too. If "props" is set, send properties.
2263  *
2264  * Pre-wrapped (cf. lzc_send_wrapper()).
2265  */
2266 static int
2267 zfs_send_cb_impl(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
2268     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
2269     void *cb_arg, nvlist_t **debugnvp)
2270 {
2271 	char errbuf[ERRBUFLEN];
2272 	send_dump_data_t sdd = { 0 };
2273 	int err = 0;
2274 	nvlist_t *fss = NULL;
2275 	avl_tree_t *fsavl = NULL;
2276 	static uint64_t holdseq;
2277 	int spa_version;
2278 	FILE *fout;
2279 
2280 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2281 	    "cannot send '%s'"), zhp->zfs_name);
2282 
2283 	if (fromsnap && fromsnap[0] == '\0') {
2284 		zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
2285 		    "zero-length incremental source"));
2286 		return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
2287 	}
2288 
2289 	if (fromsnap) {
2290 		char full_fromsnap_name[ZFS_MAX_DATASET_NAME_LEN];
2291 		if (snprintf(full_fromsnap_name, sizeof (full_fromsnap_name),
2292 		    "%s@%s", zhp->zfs_name, fromsnap) >=
2293 		    sizeof (full_fromsnap_name)) {
2294 			err = EINVAL;
2295 			goto stderr_out;
2296 		}
2297 		zfs_handle_t *fromsnapn = zfs_open(zhp->zfs_hdl,
2298 		    full_fromsnap_name, ZFS_TYPE_SNAPSHOT);
2299 		if (fromsnapn == NULL) {
2300 			err = -1;
2301 			goto err_out;
2302 		}
2303 		zfs_close(fromsnapn);
2304 	}
2305 
2306 	if (flags->replicate || flags->doall || flags->props ||
2307 	    flags->holds || flags->backup) {
2308 		char full_tosnap_name[ZFS_MAX_DATASET_NAME_LEN];
2309 		if (snprintf(full_tosnap_name, sizeof (full_tosnap_name),
2310 		    "%s@%s", zhp->zfs_name, tosnap) >=
2311 		    sizeof (full_tosnap_name)) {
2312 			err = EINVAL;
2313 			goto stderr_out;
2314 		}
2315 		zfs_handle_t *tosnap = zfs_open(zhp->zfs_hdl,
2316 		    full_tosnap_name, ZFS_TYPE_SNAPSHOT);
2317 		if (tosnap == NULL) {
2318 			err = -1;
2319 			goto err_out;
2320 		}
2321 		err = send_prelim_records(tosnap, fromsnap, outfd,
2322 		    flags->replicate || flags->props || flags->holds,
2323 		    flags->replicate, flags->verbosity > 0, flags->dryrun,
2324 		    flags->raw, flags->replicate, flags->skipmissing,
2325 		    flags->backup, flags->holds, flags->props, flags->doall,
2326 		    &fss, &fsavl);
2327 		zfs_close(tosnap);
2328 		if (err != 0)
2329 			goto err_out;
2330 	}
2331 
2332 	/* dump each stream */
2333 	sdd.fromsnap = fromsnap;
2334 	sdd.tosnap = tosnap;
2335 	sdd.outfd = outfd;
2336 	sdd.replicate = flags->replicate;
2337 	sdd.doall = flags->doall;
2338 	sdd.fromorigin = flags->fromorigin;
2339 	sdd.fss = fss;
2340 	sdd.fsavl = fsavl;
2341 	sdd.verbosity = flags->verbosity;
2342 	sdd.parsable = flags->parsable;
2343 	sdd.progress = flags->progress;
2344 	sdd.progressastitle = flags->progressastitle;
2345 	sdd.dryrun = flags->dryrun;
2346 	sdd.large_block = flags->largeblock;
2347 	sdd.embed_data = flags->embed_data;
2348 	sdd.compress = flags->compress;
2349 	sdd.raw = flags->raw;
2350 	sdd.holds = flags->holds;
2351 	sdd.filter_cb = filter_func;
2352 	sdd.filter_cb_arg = cb_arg;
2353 	if (debugnvp)
2354 		sdd.debugnv = *debugnvp;
2355 	if (sdd.verbosity != 0 && sdd.dryrun)
2356 		sdd.std_out = B_TRUE;
2357 	fout = sdd.std_out ? stdout : stderr;
2358 
2359 	/*
2360 	 * Some flags require that we place user holds on the datasets that are
2361 	 * being sent so they don't get destroyed during the send. We can skip
2362 	 * this step if the pool is imported read-only since the datasets cannot
2363 	 * be destroyed.
2364 	 */
2365 	if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
2366 	    ZPOOL_PROP_READONLY, NULL) &&
2367 	    zfs_spa_version(zhp, &spa_version) == 0 &&
2368 	    spa_version >= SPA_VERSION_USERREFS &&
2369 	    (flags->doall || flags->replicate)) {
2370 		++holdseq;
2371 		(void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
2372 		    ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
2373 		sdd.cleanup_fd = open(ZFS_DEV, O_RDWR | O_CLOEXEC);
2374 		if (sdd.cleanup_fd < 0) {
2375 			err = errno;
2376 			goto stderr_out;
2377 		}
2378 		sdd.snapholds = fnvlist_alloc();
2379 	} else {
2380 		sdd.cleanup_fd = -1;
2381 		sdd.snapholds = NULL;
2382 	}
2383 
2384 	if (flags->verbosity != 0 || sdd.snapholds != NULL) {
2385 		/*
2386 		 * Do a verbose no-op dry run to get all the verbose output
2387 		 * or to gather snapshot hold's before generating any data,
2388 		 * then do a non-verbose real run to generate the streams.
2389 		 */
2390 		sdd.dryrun = B_TRUE;
2391 		err = dump_filesystems(zhp, &sdd);
2392 
2393 		if (err != 0)
2394 			goto stderr_out;
2395 
2396 		if (flags->verbosity != 0) {
2397 			if (flags->parsable) {
2398 				(void) fprintf(fout, "size\t%llu\n",
2399 				    (longlong_t)sdd.size);
2400 			} else {
2401 				char buf[16];
2402 				zfs_nicebytes(sdd.size, buf, sizeof (buf));
2403 				(void) fprintf(fout, dgettext(TEXT_DOMAIN,
2404 				    "total estimated size is %s\n"), buf);
2405 			}
2406 		}
2407 
2408 		/* Ensure no snaps found is treated as an error. */
2409 		if (!sdd.seento) {
2410 			err = ENOENT;
2411 			goto err_out;
2412 		}
2413 
2414 		/* Skip the second run if dryrun was requested. */
2415 		if (flags->dryrun)
2416 			goto err_out;
2417 
2418 		if (sdd.snapholds != NULL) {
2419 			err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
2420 			if (err != 0)
2421 				goto stderr_out;
2422 
2423 			fnvlist_free(sdd.snapholds);
2424 			sdd.snapholds = NULL;
2425 		}
2426 
2427 		sdd.dryrun = B_FALSE;
2428 		sdd.verbosity = 0;
2429 	}
2430 
2431 	err = dump_filesystems(zhp, &sdd);
2432 	fsavl_destroy(fsavl);
2433 	fnvlist_free(fss);
2434 
2435 	/* Ensure no snaps found is treated as an error. */
2436 	if (err == 0 && !sdd.seento)
2437 		err = ENOENT;
2438 
2439 	if (sdd.cleanup_fd != -1) {
2440 		VERIFY(0 == close(sdd.cleanup_fd));
2441 		sdd.cleanup_fd = -1;
2442 	}
2443 
2444 	if (!flags->dryrun && (flags->replicate || flags->doall ||
2445 	    flags->props || flags->backup || flags->holds)) {
2446 		/*
2447 		 * write final end record.  NB: want to do this even if
2448 		 * there was some error, because it might not be totally
2449 		 * failed.
2450 		 */
2451 		int err2 = send_conclusion_record(outfd, NULL);
2452 		if (err2 != 0)
2453 			return (zfs_standard_error(zhp->zfs_hdl, err2, errbuf));
2454 	}
2455 
2456 	return (err || sdd.err);
2457 
2458 stderr_out:
2459 	err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
2460 err_out:
2461 	fsavl_destroy(fsavl);
2462 	fnvlist_free(fss);
2463 	fnvlist_free(sdd.snapholds);
2464 
2465 	if (sdd.cleanup_fd != -1)
2466 		VERIFY(0 == close(sdd.cleanup_fd));
2467 	return (err);
2468 }
2469 
2470 struct zfs_send {
2471 	zfs_handle_t *zhp;
2472 	const char *fromsnap;
2473 	const char *tosnap;
2474 	sendflags_t *flags;
2475 	snapfilter_cb_t *filter_func;
2476 	void *cb_arg;
2477 	nvlist_t **debugnvp;
2478 };
2479 
2480 static int
2481 zfs_send_cb(int outfd, void *arg)
2482 {
2483 	struct zfs_send *zs = arg;
2484 	return (zfs_send_cb_impl(zs->zhp, zs->fromsnap, zs->tosnap, zs->flags,
2485 	    outfd, zs->filter_func, zs->cb_arg, zs->debugnvp));
2486 }
2487 
2488 int
2489 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
2490     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
2491     void *cb_arg, nvlist_t **debugnvp)
2492 {
2493 	struct zfs_send arg = {
2494 		.zhp = zhp,
2495 		.fromsnap = fromsnap,
2496 		.tosnap = tosnap,
2497 		.flags = flags,
2498 		.filter_func = filter_func,
2499 		.cb_arg = cb_arg,
2500 		.debugnvp = debugnvp,
2501 	};
2502 	return (lzc_send_wrapper(zfs_send_cb, outfd, &arg));
2503 }
2504 
2505 
2506 static zfs_handle_t *
2507 name_to_dir_handle(libzfs_handle_t *hdl, const char *snapname)
2508 {
2509 	char dirname[ZFS_MAX_DATASET_NAME_LEN];
2510 	(void) strlcpy(dirname, snapname, ZFS_MAX_DATASET_NAME_LEN);
2511 	char *c = strchr(dirname, '@');
2512 	if (c != NULL)
2513 		*c = '\0';
2514 	return (zfs_open(hdl, dirname, ZFS_TYPE_DATASET));
2515 }
2516 
2517 /*
2518  * Returns B_TRUE if earlier is an earlier snapshot in later's timeline; either
2519  * an earlier snapshot in the same filesystem, or a snapshot before later's
2520  * origin, or it's origin's origin, etc.
2521  */
2522 static boolean_t
2523 snapshot_is_before(zfs_handle_t *earlier, zfs_handle_t *later)
2524 {
2525 	boolean_t ret;
2526 	uint64_t later_txg =
2527 	    (later->zfs_type == ZFS_TYPE_FILESYSTEM ||
2528 	    later->zfs_type == ZFS_TYPE_VOLUME ?
2529 	    UINT64_MAX : zfs_prop_get_int(later, ZFS_PROP_CREATETXG));
2530 	uint64_t earlier_txg = zfs_prop_get_int(earlier, ZFS_PROP_CREATETXG);
2531 
2532 	if (earlier_txg >= later_txg)
2533 		return (B_FALSE);
2534 
2535 	zfs_handle_t *earlier_dir = name_to_dir_handle(earlier->zfs_hdl,
2536 	    earlier->zfs_name);
2537 	zfs_handle_t *later_dir = name_to_dir_handle(later->zfs_hdl,
2538 	    later->zfs_name);
2539 
2540 	if (strcmp(earlier_dir->zfs_name, later_dir->zfs_name) == 0) {
2541 		zfs_close(earlier_dir);
2542 		zfs_close(later_dir);
2543 		return (B_TRUE);
2544 	}
2545 
2546 	char clonename[ZFS_MAX_DATASET_NAME_LEN];
2547 	if (zfs_prop_get(later_dir, ZFS_PROP_ORIGIN, clonename,
2548 	    ZFS_MAX_DATASET_NAME_LEN, NULL, NULL, 0, B_TRUE) != 0) {
2549 		zfs_close(earlier_dir);
2550 		zfs_close(later_dir);
2551 		return (B_FALSE);
2552 	}
2553 
2554 	zfs_handle_t *origin = zfs_open(earlier->zfs_hdl, clonename,
2555 	    ZFS_TYPE_DATASET);
2556 	uint64_t origin_txg = zfs_prop_get_int(origin, ZFS_PROP_CREATETXG);
2557 
2558 	/*
2559 	 * If "earlier" is exactly the origin, then
2560 	 * snapshot_is_before(earlier, origin) will return false (because
2561 	 * they're the same).
2562 	 */
2563 	if (origin_txg == earlier_txg &&
2564 	    strcmp(origin->zfs_name, earlier->zfs_name) == 0) {
2565 		zfs_close(earlier_dir);
2566 		zfs_close(later_dir);
2567 		zfs_close(origin);
2568 		return (B_TRUE);
2569 	}
2570 	zfs_close(earlier_dir);
2571 	zfs_close(later_dir);
2572 
2573 	ret = snapshot_is_before(earlier, origin);
2574 	zfs_close(origin);
2575 	return (ret);
2576 }
2577 
2578 /*
2579  * The "zhp" argument is the handle of the dataset to send (typically a
2580  * snapshot).  The "from" argument is the full name of the snapshot or
2581  * bookmark that is the incremental source.
2582  *
2583  * Pre-wrapped (cf. lzc_send_wrapper()).
2584  */
2585 static int
2586 zfs_send_one_cb_impl(zfs_handle_t *zhp, const char *from, int fd,
2587     sendflags_t *flags, const char *redactbook)
2588 {
2589 	int err;
2590 	libzfs_handle_t *hdl = zhp->zfs_hdl;
2591 	char *name = zhp->zfs_name;
2592 	pthread_t ptid;
2593 	progress_arg_t pa = { 0 };
2594 	uint64_t size = 0;
2595 
2596 	char errbuf[ERRBUFLEN];
2597 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2598 	    "warning: cannot send '%s'"), name);
2599 
2600 	if (from != NULL && strchr(from, '@')) {
2601 		zfs_handle_t *from_zhp = zfs_open(hdl, from,
2602 		    ZFS_TYPE_DATASET);
2603 		if (from_zhp == NULL)
2604 			return (-1);
2605 		if (!snapshot_is_before(from_zhp, zhp)) {
2606 			zfs_close(from_zhp);
2607 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2608 			    "not an earlier snapshot from the same fs"));
2609 			return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
2610 		}
2611 		zfs_close(from_zhp);
2612 	}
2613 
2614 	if (redactbook != NULL) {
2615 		char bookname[ZFS_MAX_DATASET_NAME_LEN];
2616 		nvlist_t *redact_snaps;
2617 		zfs_handle_t *book_zhp;
2618 		char *at, *pound;
2619 		int dsnamelen;
2620 
2621 		pound = strchr(redactbook, '#');
2622 		if (pound != NULL)
2623 			redactbook = pound + 1;
2624 		at = strchr(name, '@');
2625 		if (at == NULL) {
2626 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2627 			    "cannot do a redacted send to a filesystem"));
2628 			return (zfs_error(hdl, EZFS_BADTYPE, errbuf));
2629 		}
2630 		dsnamelen = at - name;
2631 		if (snprintf(bookname, sizeof (bookname), "%.*s#%s",
2632 		    dsnamelen, name, redactbook)
2633 		    >= sizeof (bookname)) {
2634 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2635 			    "invalid bookmark name"));
2636 			return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2637 		}
2638 		book_zhp = zfs_open(hdl, bookname, ZFS_TYPE_BOOKMARK);
2639 		if (book_zhp == NULL)
2640 			return (-1);
2641 		if (nvlist_lookup_nvlist(book_zhp->zfs_props,
2642 		    zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS),
2643 		    &redact_snaps) != 0 || redact_snaps == NULL) {
2644 			zfs_close(book_zhp);
2645 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2646 			    "not a redaction bookmark"));
2647 			return (zfs_error(hdl, EZFS_BADTYPE, errbuf));
2648 		}
2649 		zfs_close(book_zhp);
2650 	}
2651 
2652 	/*
2653 	 * Send fs properties
2654 	 */
2655 	if (flags->props || flags->holds || flags->backup) {
2656 		/*
2657 		 * Note: the header generated by send_prelim_records()
2658 		 * assumes that the incremental source is in the same
2659 		 * filesystem/volume as the target (which is a requirement
2660 		 * when doing "zfs send -R").  But that isn't always the
2661 		 * case here (e.g. send from snap in origin, or send from
2662 		 * bookmark).  We pass from=NULL, which will omit this
2663 		 * information from the prelim records; it isn't used
2664 		 * when receiving this type of stream.
2665 		 */
2666 		err = send_prelim_records(zhp, NULL, fd, B_TRUE, B_FALSE,
2667 		    flags->verbosity > 0, flags->dryrun, flags->raw,
2668 		    flags->replicate, B_FALSE, flags->backup, flags->holds,
2669 		    flags->props, flags->doall, NULL, NULL);
2670 		if (err != 0)
2671 			return (err);
2672 	}
2673 
2674 	/*
2675 	 * Perform size estimate if verbose was specified.
2676 	 */
2677 	if (flags->verbosity != 0 || flags->progressastitle) {
2678 		err = estimate_size(zhp, from, fd, flags, 0, 0, 0, redactbook,
2679 		    errbuf, &size);
2680 		if (err != 0)
2681 			return (err);
2682 	}
2683 
2684 	if (flags->dryrun)
2685 		return (0);
2686 
2687 	/*
2688 	 * If progress reporting is requested, spawn a new thread to poll
2689 	 * ZFS_IOC_SEND_PROGRESS at a regular interval.
2690 	 */
2691 	if (flags->progress || flags->progressastitle) {
2692 		pa.pa_zhp = zhp;
2693 		pa.pa_fd = fd;
2694 		pa.pa_parsable = flags->parsable;
2695 		pa.pa_estimate = B_FALSE;
2696 		pa.pa_verbosity = flags->verbosity;
2697 		pa.pa_size = size;
2698 		pa.pa_astitle = flags->progressastitle;
2699 
2700 		err = pthread_create(&ptid, NULL,
2701 		    send_progress_thread, &pa);
2702 		if (err != 0) {
2703 			zfs_error_aux(zhp->zfs_hdl, "%s", strerror(errno));
2704 			return (zfs_error(zhp->zfs_hdl,
2705 			    EZFS_THREADCREATEFAILED, errbuf));
2706 		}
2707 	}
2708 
2709 	err = lzc_send_redacted(name, from, fd,
2710 	    lzc_flags_from_sendflags(flags), redactbook);
2711 
2712 	if ((flags->progress || flags->progressastitle) &&
2713 	    send_progress_thread_exit(hdl, ptid))
2714 			return (-1);
2715 
2716 	if (err == 0 && (flags->props || flags->holds || flags->backup)) {
2717 		/* Write the final end record. */
2718 		err = send_conclusion_record(fd, NULL);
2719 		if (err != 0)
2720 			return (zfs_standard_error(hdl, err, errbuf));
2721 	}
2722 	if (err != 0) {
2723 		switch (errno) {
2724 		case EXDEV:
2725 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2726 			    "not an earlier snapshot from the same fs"));
2727 			return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
2728 
2729 		case ENOENT:
2730 		case ESRCH:
2731 			if (lzc_exists(name)) {
2732 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2733 				    "incremental source (%s) does not exist"),
2734 				    from);
2735 			}
2736 			return (zfs_error(hdl, EZFS_NOENT, errbuf));
2737 
2738 		case EACCES:
2739 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2740 			    "dataset key must be loaded"));
2741 			return (zfs_error(hdl, EZFS_CRYPTOFAILED, errbuf));
2742 
2743 		case EBUSY:
2744 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2745 			    "target is busy; if a filesystem, "
2746 			    "it must not be mounted"));
2747 			return (zfs_error(hdl, EZFS_BUSY, errbuf));
2748 
2749 		case EDQUOT:
2750 		case EFAULT:
2751 		case EFBIG:
2752 		case EINVAL:
2753 		case EIO:
2754 		case ENOLINK:
2755 		case ENOSPC:
2756 		case ENOSTR:
2757 		case ENXIO:
2758 		case EPIPE:
2759 		case ERANGE:
2760 		case EROFS:
2761 			zfs_error_aux(hdl, "%s", strerror(errno));
2762 			return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
2763 
2764 		default:
2765 			return (zfs_standard_error(hdl, errno, errbuf));
2766 		}
2767 	}
2768 	return (err != 0);
2769 }
2770 
2771 struct zfs_send_one {
2772 	zfs_handle_t *zhp;
2773 	const char *from;
2774 	sendflags_t *flags;
2775 	const char *redactbook;
2776 };
2777 
2778 static int
2779 zfs_send_one_cb(int fd, void *arg)
2780 {
2781 	struct zfs_send_one *zso = arg;
2782 	return (zfs_send_one_cb_impl(zso->zhp, zso->from, fd, zso->flags,
2783 	    zso->redactbook));
2784 }
2785 
2786 int
2787 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd, sendflags_t *flags,
2788     const char *redactbook)
2789 {
2790 	struct zfs_send_one zso = {
2791 		.zhp = zhp,
2792 		.from = from,
2793 		.flags = flags,
2794 		.redactbook = redactbook,
2795 	};
2796 	return (lzc_send_wrapper(zfs_send_one_cb, fd, &zso));
2797 }
2798 
2799 /*
2800  * Routines specific to "zfs recv"
2801  */
2802 
2803 static int
2804 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
2805     boolean_t byteswap, zio_cksum_t *zc)
2806 {
2807 	char *cp = buf;
2808 	int rv;
2809 	int len = ilen;
2810 
2811 	do {
2812 		rv = read(fd, cp, len);
2813 		cp += rv;
2814 		len -= rv;
2815 	} while (rv > 0);
2816 
2817 	if (rv < 0 || len != 0) {
2818 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2819 		    "failed to read from stream"));
2820 		return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
2821 		    "cannot receive")));
2822 	}
2823 
2824 	if (zc) {
2825 		if (byteswap)
2826 			fletcher_4_incremental_byteswap(buf, ilen, zc);
2827 		else
2828 			fletcher_4_incremental_native(buf, ilen, zc);
2829 	}
2830 	return (0);
2831 }
2832 
2833 static int
2834 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
2835     boolean_t byteswap, zio_cksum_t *zc)
2836 {
2837 	char *buf;
2838 	int err;
2839 
2840 	buf = zfs_alloc(hdl, len);
2841 
2842 	if (len > hdl->libzfs_max_nvlist) {
2843 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "nvlist too large"));
2844 		free(buf);
2845 		return (ENOMEM);
2846 	}
2847 
2848 	err = recv_read(hdl, fd, buf, len, byteswap, zc);
2849 	if (err != 0) {
2850 		free(buf);
2851 		return (err);
2852 	}
2853 
2854 	err = nvlist_unpack(buf, len, nvp, 0);
2855 	free(buf);
2856 	if (err != 0) {
2857 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2858 		    "stream (malformed nvlist)"));
2859 		return (EINVAL);
2860 	}
2861 	return (0);
2862 }
2863 
2864 /*
2865  * Returns the grand origin (origin of origin of origin...) of a given handle.
2866  * If this dataset is not a clone, it simply returns a copy of the original
2867  * handle.
2868  */
2869 static zfs_handle_t *
2870 recv_open_grand_origin(zfs_handle_t *zhp)
2871 {
2872 	char origin[ZFS_MAX_DATASET_NAME_LEN];
2873 	zprop_source_t src;
2874 	zfs_handle_t *ozhp = zfs_handle_dup(zhp);
2875 
2876 	while (ozhp != NULL) {
2877 		if (zfs_prop_get(ozhp, ZFS_PROP_ORIGIN, origin,
2878 		    sizeof (origin), &src, NULL, 0, B_FALSE) != 0)
2879 			break;
2880 
2881 		(void) zfs_close(ozhp);
2882 		ozhp = zfs_open(zhp->zfs_hdl, origin, ZFS_TYPE_FILESYSTEM);
2883 	}
2884 
2885 	return (ozhp);
2886 }
2887 
2888 static int
2889 recv_rename_impl(zfs_handle_t *zhp, const char *name, const char *newname)
2890 {
2891 	int err;
2892 	zfs_handle_t *ozhp = NULL;
2893 
2894 	/*
2895 	 * Attempt to rename the dataset. If it fails with EACCES we have
2896 	 * attempted to rename the dataset outside of its encryption root.
2897 	 * Force the dataset to become an encryption root and try again.
2898 	 */
2899 	err = lzc_rename(name, newname);
2900 	if (err == EACCES) {
2901 		ozhp = recv_open_grand_origin(zhp);
2902 		if (ozhp == NULL) {
2903 			err = ENOENT;
2904 			goto out;
2905 		}
2906 
2907 		err = lzc_change_key(ozhp->zfs_name, DCP_CMD_FORCE_NEW_KEY,
2908 		    NULL, NULL, 0);
2909 		if (err != 0)
2910 			goto out;
2911 
2912 		err = lzc_rename(name, newname);
2913 	}
2914 
2915 out:
2916 	if (ozhp != NULL)
2917 		zfs_close(ozhp);
2918 	return (err);
2919 }
2920 
2921 static int
2922 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
2923     int baselen, char *newname, recvflags_t *flags)
2924 {
2925 	static int seq;
2926 	int err;
2927 	prop_changelist_t *clp = NULL;
2928 	zfs_handle_t *zhp = NULL;
2929 
2930 	zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2931 	if (zhp == NULL) {
2932 		err = -1;
2933 		goto out;
2934 	}
2935 	clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2936 	    flags->force ? MS_FORCE : 0);
2937 	if (clp == NULL) {
2938 		err = -1;
2939 		goto out;
2940 	}
2941 	err = changelist_prefix(clp);
2942 	if (err)
2943 		goto out;
2944 
2945 	if (tryname) {
2946 		(void) strlcpy(newname, tryname, ZFS_MAX_DATASET_NAME_LEN);
2947 		if (flags->verbose) {
2948 			(void) printf("attempting rename %s to %s\n",
2949 			    name, newname);
2950 		}
2951 		err = recv_rename_impl(zhp, name, newname);
2952 		if (err == 0)
2953 			changelist_rename(clp, name, tryname);
2954 	} else {
2955 		err = ENOENT;
2956 	}
2957 
2958 	if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
2959 		seq++;
2960 
2961 		(void) snprintf(newname, ZFS_MAX_DATASET_NAME_LEN,
2962 		    "%.*srecv-%u-%u", baselen, name, getpid(), seq);
2963 
2964 		if (flags->verbose) {
2965 			(void) printf("failed - trying rename %s to %s\n",
2966 			    name, newname);
2967 		}
2968 		err = recv_rename_impl(zhp, name, newname);
2969 		if (err == 0)
2970 			changelist_rename(clp, name, newname);
2971 		if (err && flags->verbose) {
2972 			(void) printf("failed (%u) - "
2973 			    "will try again on next pass\n", errno);
2974 		}
2975 		err = EAGAIN;
2976 	} else if (flags->verbose) {
2977 		if (err == 0)
2978 			(void) printf("success\n");
2979 		else
2980 			(void) printf("failed (%u)\n", errno);
2981 	}
2982 
2983 	(void) changelist_postfix(clp);
2984 
2985 out:
2986 	if (clp != NULL)
2987 		changelist_free(clp);
2988 	if (zhp != NULL)
2989 		zfs_close(zhp);
2990 
2991 	return (err);
2992 }
2993 
2994 static int
2995 recv_promote(libzfs_handle_t *hdl, const char *fsname,
2996     const char *origin_fsname, recvflags_t *flags)
2997 {
2998 	int err;
2999 	zfs_cmd_t zc = {"\0"};
3000 	zfs_handle_t *zhp = NULL, *ozhp = NULL;
3001 
3002 	if (flags->verbose)
3003 		(void) printf("promoting %s\n", fsname);
3004 
3005 	(void) strlcpy(zc.zc_value, origin_fsname, sizeof (zc.zc_value));
3006 	(void) strlcpy(zc.zc_name, fsname, sizeof (zc.zc_name));
3007 
3008 	/*
3009 	 * Attempt to promote the dataset. If it fails with EACCES the
3010 	 * promotion would cause this dataset to leave its encryption root.
3011 	 * Force the origin to become an encryption root and try again.
3012 	 */
3013 	err = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
3014 	if (err == EACCES) {
3015 		zhp = zfs_open(hdl, fsname, ZFS_TYPE_DATASET);
3016 		if (zhp == NULL) {
3017 			err = -1;
3018 			goto out;
3019 		}
3020 
3021 		ozhp = recv_open_grand_origin(zhp);
3022 		if (ozhp == NULL) {
3023 			err = -1;
3024 			goto out;
3025 		}
3026 
3027 		err = lzc_change_key(ozhp->zfs_name, DCP_CMD_FORCE_NEW_KEY,
3028 		    NULL, NULL, 0);
3029 		if (err != 0)
3030 			goto out;
3031 
3032 		err = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
3033 	}
3034 
3035 out:
3036 	if (zhp != NULL)
3037 		zfs_close(zhp);
3038 	if (ozhp != NULL)
3039 		zfs_close(ozhp);
3040 
3041 	return (err);
3042 }
3043 
3044 static int
3045 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
3046     char *newname, recvflags_t *flags)
3047 {
3048 	int err = 0;
3049 	prop_changelist_t *clp;
3050 	zfs_handle_t *zhp;
3051 	boolean_t defer = B_FALSE;
3052 	int spa_version;
3053 
3054 	zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
3055 	if (zhp == NULL)
3056 		return (-1);
3057 	zfs_type_t type = zfs_get_type(zhp);
3058 	if (type == ZFS_TYPE_SNAPSHOT &&
3059 	    zfs_spa_version(zhp, &spa_version) == 0 &&
3060 	    spa_version >= SPA_VERSION_USERREFS)
3061 		defer = B_TRUE;
3062 	clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
3063 	    flags->force ? MS_FORCE : 0);
3064 	zfs_close(zhp);
3065 	if (clp == NULL)
3066 		return (-1);
3067 
3068 	err = changelist_prefix(clp);
3069 	if (err)
3070 		return (err);
3071 
3072 	if (flags->verbose)
3073 		(void) printf("attempting destroy %s\n", name);
3074 	if (type == ZFS_TYPE_SNAPSHOT) {
3075 		nvlist_t *nv = fnvlist_alloc();
3076 		fnvlist_add_boolean(nv, name);
3077 		err = lzc_destroy_snaps(nv, defer, NULL);
3078 		fnvlist_free(nv);
3079 	} else {
3080 		err = lzc_destroy(name);
3081 	}
3082 	if (err == 0) {
3083 		if (flags->verbose)
3084 			(void) printf("success\n");
3085 		changelist_remove(clp, name);
3086 	}
3087 
3088 	(void) changelist_postfix(clp);
3089 	changelist_free(clp);
3090 
3091 	/*
3092 	 * Deferred destroy might destroy the snapshot or only mark it to be
3093 	 * destroyed later, and it returns success in either case.
3094 	 */
3095 	if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
3096 	    ZFS_TYPE_SNAPSHOT))) {
3097 		err = recv_rename(hdl, name, NULL, baselen, newname, flags);
3098 	}
3099 
3100 	return (err);
3101 }
3102 
3103 typedef struct guid_to_name_data {
3104 	uint64_t guid;
3105 	boolean_t bookmark_ok;
3106 	char *name;
3107 	char *skip;
3108 	uint64_t *redact_snap_guids;
3109 	uint64_t num_redact_snaps;
3110 } guid_to_name_data_t;
3111 
3112 static boolean_t
3113 redact_snaps_match(zfs_handle_t *zhp, guid_to_name_data_t *gtnd)
3114 {
3115 	uint64_t *bmark_snaps;
3116 	uint_t bmark_num_snaps;
3117 	nvlist_t *nvl;
3118 	if (zhp->zfs_type != ZFS_TYPE_BOOKMARK)
3119 		return (B_FALSE);
3120 
3121 	nvl = fnvlist_lookup_nvlist(zhp->zfs_props,
3122 	    zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS));
3123 	bmark_snaps = fnvlist_lookup_uint64_array(nvl, ZPROP_VALUE,
3124 	    &bmark_num_snaps);
3125 	if (bmark_num_snaps != gtnd->num_redact_snaps)
3126 		return (B_FALSE);
3127 	int i = 0;
3128 	for (; i < bmark_num_snaps; i++) {
3129 		int j = 0;
3130 		for (; j < bmark_num_snaps; j++) {
3131 			if (bmark_snaps[i] == gtnd->redact_snap_guids[j])
3132 				break;
3133 		}
3134 		if (j == bmark_num_snaps)
3135 			break;
3136 	}
3137 	return (i == bmark_num_snaps);
3138 }
3139 
3140 static int
3141 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
3142 {
3143 	guid_to_name_data_t *gtnd = arg;
3144 	const char *slash;
3145 	int err;
3146 
3147 	if (gtnd->skip != NULL &&
3148 	    (slash = strrchr(zhp->zfs_name, '/')) != NULL &&
3149 	    strcmp(slash + 1, gtnd->skip) == 0) {
3150 		zfs_close(zhp);
3151 		return (0);
3152 	}
3153 
3154 	if (zfs_prop_get_int(zhp, ZFS_PROP_GUID) == gtnd->guid &&
3155 	    (gtnd->num_redact_snaps == -1 || redact_snaps_match(zhp, gtnd))) {
3156 		(void) strcpy(gtnd->name, zhp->zfs_name);
3157 		zfs_close(zhp);
3158 		return (EEXIST);
3159 	}
3160 
3161 	err = zfs_iter_children(zhp, 0, guid_to_name_cb, gtnd);
3162 	if (err != EEXIST && gtnd->bookmark_ok)
3163 		err = zfs_iter_bookmarks(zhp, 0, guid_to_name_cb, gtnd);
3164 	zfs_close(zhp);
3165 	return (err);
3166 }
3167 
3168 /*
3169  * Attempt to find the local dataset associated with this guid.  In the case of
3170  * multiple matches, we attempt to find the "best" match by searching
3171  * progressively larger portions of the hierarchy.  This allows one to send a
3172  * tree of datasets individually and guarantee that we will find the source
3173  * guid within that hierarchy, even if there are multiple matches elsewhere.
3174  *
3175  * If num_redact_snaps is not -1, we attempt to find a redaction bookmark with
3176  * the specified number of redaction snapshots.  If num_redact_snaps isn't 0 or
3177  * -1, then redact_snap_guids will be an array of the guids of the snapshots the
3178  * redaction bookmark was created with.  If num_redact_snaps is -1, then we will
3179  * attempt to find a snapshot or bookmark (if bookmark_ok is passed) with the
3180  * given guid.  Note that a redaction bookmark can be returned if
3181  * num_redact_snaps == -1.
3182  */
3183 static int
3184 guid_to_name_redact_snaps(libzfs_handle_t *hdl, const char *parent,
3185     uint64_t guid, boolean_t bookmark_ok, uint64_t *redact_snap_guids,
3186     uint64_t num_redact_snaps, char *name)
3187 {
3188 	char pname[ZFS_MAX_DATASET_NAME_LEN];
3189 	guid_to_name_data_t gtnd;
3190 
3191 	gtnd.guid = guid;
3192 	gtnd.bookmark_ok = bookmark_ok;
3193 	gtnd.name = name;
3194 	gtnd.skip = NULL;
3195 	gtnd.redact_snap_guids = redact_snap_guids;
3196 	gtnd.num_redact_snaps = num_redact_snaps;
3197 
3198 	/*
3199 	 * Search progressively larger portions of the hierarchy, starting
3200 	 * with the filesystem specified by 'parent'.  This will
3201 	 * select the "most local" version of the origin snapshot in the case
3202 	 * that there are multiple matching snapshots in the system.
3203 	 */
3204 	(void) strlcpy(pname, parent, sizeof (pname));
3205 	char *cp = strrchr(pname, '@');
3206 	if (cp == NULL)
3207 		cp = strchr(pname, '\0');
3208 	for (; cp != NULL; cp = strrchr(pname, '/')) {
3209 		/* Chop off the last component and open the parent */
3210 		*cp = '\0';
3211 		zfs_handle_t *zhp = make_dataset_handle(hdl, pname);
3212 
3213 		if (zhp == NULL)
3214 			continue;
3215 		int err = guid_to_name_cb(zfs_handle_dup(zhp), &gtnd);
3216 		if (err != EEXIST)
3217 			err = zfs_iter_children(zhp, 0, guid_to_name_cb, &gtnd);
3218 		if (err != EEXIST && bookmark_ok)
3219 			err = zfs_iter_bookmarks(zhp, 0, guid_to_name_cb,
3220 			    &gtnd);
3221 		zfs_close(zhp);
3222 		if (err == EEXIST)
3223 			return (0);
3224 
3225 		/*
3226 		 * Remember the last portion of the dataset so we skip it next
3227 		 * time through (as we've already searched that portion of the
3228 		 * hierarchy).
3229 		 */
3230 		gtnd.skip = strrchr(pname, '/') + 1;
3231 	}
3232 
3233 	return (ENOENT);
3234 }
3235 
3236 static int
3237 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
3238     boolean_t bookmark_ok, char *name)
3239 {
3240 	return (guid_to_name_redact_snaps(hdl, parent, guid, bookmark_ok, NULL,
3241 	    -1, name));
3242 }
3243 
3244 /*
3245  * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
3246  * guid1 is after guid2.
3247  */
3248 static int
3249 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
3250     uint64_t guid1, uint64_t guid2)
3251 {
3252 	nvlist_t *nvfs;
3253 	char *fsname = NULL, *snapname = NULL;
3254 	char buf[ZFS_MAX_DATASET_NAME_LEN];
3255 	int rv;
3256 	zfs_handle_t *guid1hdl, *guid2hdl;
3257 	uint64_t create1, create2;
3258 
3259 	if (guid2 == 0)
3260 		return (0);
3261 	if (guid1 == 0)
3262 		return (1);
3263 
3264 	nvfs = fsavl_find(avl, guid1, &snapname);
3265 	fsname = fnvlist_lookup_string(nvfs, "name");
3266 	(void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
3267 	guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
3268 	if (guid1hdl == NULL)
3269 		return (-1);
3270 
3271 	nvfs = fsavl_find(avl, guid2, &snapname);
3272 	fsname = fnvlist_lookup_string(nvfs, "name");
3273 	(void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
3274 	guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
3275 	if (guid2hdl == NULL) {
3276 		zfs_close(guid1hdl);
3277 		return (-1);
3278 	}
3279 
3280 	create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
3281 	create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
3282 
3283 	if (create1 < create2)
3284 		rv = -1;
3285 	else if (create1 > create2)
3286 		rv = +1;
3287 	else
3288 		rv = 0;
3289 
3290 	zfs_close(guid1hdl);
3291 	zfs_close(guid2hdl);
3292 
3293 	return (rv);
3294 }
3295 
3296 /*
3297  * This function reestablishes the hierarchy of encryption roots after a
3298  * recursive incremental receive has completed. This must be done after the
3299  * second call to recv_incremental_replication() has renamed and promoted all
3300  * sent datasets to their final locations in the dataset hierarchy.
3301  */
3302 static int
3303 recv_fix_encryption_hierarchy(libzfs_handle_t *hdl, const char *top_zfs,
3304     nvlist_t *stream_nv)
3305 {
3306 	int err;
3307 	nvpair_t *fselem = NULL;
3308 	nvlist_t *stream_fss;
3309 
3310 	stream_fss = fnvlist_lookup_nvlist(stream_nv, "fss");
3311 
3312 	while ((fselem = nvlist_next_nvpair(stream_fss, fselem)) != NULL) {
3313 		zfs_handle_t *zhp = NULL;
3314 		uint64_t crypt;
3315 		nvlist_t *snaps, *props, *stream_nvfs = NULL;
3316 		nvpair_t *snapel = NULL;
3317 		boolean_t is_encroot, is_clone, stream_encroot;
3318 		char *cp;
3319 		char *stream_keylocation = NULL;
3320 		char keylocation[MAXNAMELEN];
3321 		char fsname[ZFS_MAX_DATASET_NAME_LEN];
3322 
3323 		keylocation[0] = '\0';
3324 		stream_nvfs = fnvpair_value_nvlist(fselem);
3325 		snaps = fnvlist_lookup_nvlist(stream_nvfs, "snaps");
3326 		props = fnvlist_lookup_nvlist(stream_nvfs, "props");
3327 		stream_encroot = nvlist_exists(stream_nvfs, "is_encroot");
3328 
3329 		/* find a snapshot from the stream that exists locally */
3330 		err = ENOENT;
3331 		while ((snapel = nvlist_next_nvpair(snaps, snapel)) != NULL) {
3332 			uint64_t guid;
3333 
3334 			guid = fnvpair_value_uint64(snapel);
3335 			err = guid_to_name(hdl, top_zfs, guid, B_FALSE,
3336 			    fsname);
3337 			if (err == 0)
3338 				break;
3339 		}
3340 
3341 		if (err != 0)
3342 			continue;
3343 
3344 		cp = strchr(fsname, '@');
3345 		if (cp != NULL)
3346 			*cp = '\0';
3347 
3348 		zhp = zfs_open(hdl, fsname, ZFS_TYPE_DATASET);
3349 		if (zhp == NULL) {
3350 			err = ENOENT;
3351 			goto error;
3352 		}
3353 
3354 		crypt = zfs_prop_get_int(zhp, ZFS_PROP_ENCRYPTION);
3355 		is_clone = zhp->zfs_dmustats.dds_origin[0] != '\0';
3356 		(void) zfs_crypto_get_encryption_root(zhp, &is_encroot, NULL);
3357 
3358 		/* we don't need to do anything for unencrypted datasets */
3359 		if (crypt == ZIO_CRYPT_OFF) {
3360 			zfs_close(zhp);
3361 			continue;
3362 		}
3363 
3364 		/*
3365 		 * If the dataset is flagged as an encryption root, was not
3366 		 * received as a clone and is not currently an encryption root,
3367 		 * force it to become one. Fixup the keylocation if necessary.
3368 		 */
3369 		if (stream_encroot) {
3370 			if (!is_clone && !is_encroot) {
3371 				err = lzc_change_key(fsname,
3372 				    DCP_CMD_FORCE_NEW_KEY, NULL, NULL, 0);
3373 				if (err != 0) {
3374 					zfs_close(zhp);
3375 					goto error;
3376 				}
3377 			}
3378 
3379 			stream_keylocation = fnvlist_lookup_string(props,
3380 			    zfs_prop_to_name(ZFS_PROP_KEYLOCATION));
3381 
3382 			/*
3383 			 * Refresh the properties in case the call to
3384 			 * lzc_change_key() changed the value.
3385 			 */
3386 			zfs_refresh_properties(zhp);
3387 			err = zfs_prop_get(zhp, ZFS_PROP_KEYLOCATION,
3388 			    keylocation, sizeof (keylocation), NULL, NULL,
3389 			    0, B_TRUE);
3390 			if (err != 0) {
3391 				zfs_close(zhp);
3392 				goto error;
3393 			}
3394 
3395 			if (strcmp(keylocation, stream_keylocation) != 0) {
3396 				err = zfs_prop_set(zhp,
3397 				    zfs_prop_to_name(ZFS_PROP_KEYLOCATION),
3398 				    stream_keylocation);
3399 				if (err != 0) {
3400 					zfs_close(zhp);
3401 					goto error;
3402 				}
3403 			}
3404 		}
3405 
3406 		/*
3407 		 * If the dataset is not flagged as an encryption root and is
3408 		 * currently an encryption root, force it to inherit from its
3409 		 * parent. The root of a raw send should never be
3410 		 * force-inherited.
3411 		 */
3412 		if (!stream_encroot && is_encroot &&
3413 		    strcmp(top_zfs, fsname) != 0) {
3414 			err = lzc_change_key(fsname, DCP_CMD_FORCE_INHERIT,
3415 			    NULL, NULL, 0);
3416 			if (err != 0) {
3417 				zfs_close(zhp);
3418 				goto error;
3419 			}
3420 		}
3421 
3422 		zfs_close(zhp);
3423 	}
3424 
3425 	return (0);
3426 
3427 error:
3428 	return (err);
3429 }
3430 
3431 static int
3432 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
3433     recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
3434     nvlist_t *renamed)
3435 {
3436 	nvlist_t *local_nv, *deleted = NULL;
3437 	avl_tree_t *local_avl;
3438 	nvpair_t *fselem, *nextfselem;
3439 	char *fromsnap;
3440 	char newname[ZFS_MAX_DATASET_NAME_LEN];
3441 	char guidname[32];
3442 	int error;
3443 	boolean_t needagain, progress, recursive;
3444 	char *s1, *s2;
3445 
3446 	fromsnap = fnvlist_lookup_string(stream_nv, "fromsnap");
3447 
3448 	recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
3449 	    ENOENT);
3450 
3451 	if (flags->dryrun)
3452 		return (0);
3453 
3454 again:
3455 	needagain = progress = B_FALSE;
3456 
3457 	deleted = fnvlist_alloc();
3458 
3459 	if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
3460 	    recursive, B_TRUE, B_FALSE, recursive, B_FALSE, B_FALSE, B_FALSE,
3461 	    B_FALSE, B_TRUE, &local_nv, &local_avl)) != 0)
3462 		return (error);
3463 
3464 	/*
3465 	 * Process deletes and renames
3466 	 */
3467 	for (fselem = nvlist_next_nvpair(local_nv, NULL);
3468 	    fselem; fselem = nextfselem) {
3469 		nvlist_t *nvfs, *snaps;
3470 		nvlist_t *stream_nvfs = NULL;
3471 		nvpair_t *snapelem, *nextsnapelem;
3472 		uint64_t fromguid = 0;
3473 		uint64_t originguid = 0;
3474 		uint64_t stream_originguid = 0;
3475 		uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
3476 		char *fsname, *stream_fsname;
3477 
3478 		nextfselem = nvlist_next_nvpair(local_nv, fselem);
3479 
3480 		nvfs = fnvpair_value_nvlist(fselem);
3481 		snaps = fnvlist_lookup_nvlist(nvfs, "snaps");
3482 		fsname = fnvlist_lookup_string(nvfs, "name");
3483 		parent_fromsnap_guid = fnvlist_lookup_uint64(nvfs,
3484 		    "parentfromsnap");
3485 		(void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
3486 
3487 		/*
3488 		 * First find the stream's fs, so we can check for
3489 		 * a different origin (due to "zfs promote")
3490 		 */
3491 		for (snapelem = nvlist_next_nvpair(snaps, NULL);
3492 		    snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
3493 			uint64_t thisguid;
3494 
3495 			thisguid = fnvpair_value_uint64(snapelem);
3496 			stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
3497 
3498 			if (stream_nvfs != NULL)
3499 				break;
3500 		}
3501 
3502 		/* check for promote */
3503 		(void) nvlist_lookup_uint64(stream_nvfs, "origin",
3504 		    &stream_originguid);
3505 		if (stream_nvfs && originguid != stream_originguid) {
3506 			switch (created_before(hdl, local_avl,
3507 			    stream_originguid, originguid)) {
3508 			case 1: {
3509 				/* promote it! */
3510 				nvlist_t *origin_nvfs;
3511 				char *origin_fsname;
3512 
3513 				origin_nvfs = fsavl_find(local_avl, originguid,
3514 				    NULL);
3515 				origin_fsname = fnvlist_lookup_string(
3516 				    origin_nvfs, "name");
3517 				error = recv_promote(hdl, fsname, origin_fsname,
3518 				    flags);
3519 				if (error == 0)
3520 					progress = B_TRUE;
3521 				break;
3522 			}
3523 			default:
3524 				break;
3525 			case -1:
3526 				fsavl_destroy(local_avl);
3527 				fnvlist_free(local_nv);
3528 				return (-1);
3529 			}
3530 			/*
3531 			 * We had/have the wrong origin, therefore our
3532 			 * list of snapshots is wrong.  Need to handle
3533 			 * them on the next pass.
3534 			 */
3535 			needagain = B_TRUE;
3536 			continue;
3537 		}
3538 
3539 		for (snapelem = nvlist_next_nvpair(snaps, NULL);
3540 		    snapelem; snapelem = nextsnapelem) {
3541 			uint64_t thisguid;
3542 			char *stream_snapname;
3543 			nvlist_t *found, *props;
3544 
3545 			nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
3546 
3547 			thisguid = fnvpair_value_uint64(snapelem);
3548 			found = fsavl_find(stream_avl, thisguid,
3549 			    &stream_snapname);
3550 
3551 			/* check for delete */
3552 			if (found == NULL) {
3553 				char name[ZFS_MAX_DATASET_NAME_LEN];
3554 
3555 				if (!flags->force)
3556 					continue;
3557 
3558 				(void) snprintf(name, sizeof (name), "%s@%s",
3559 				    fsname, nvpair_name(snapelem));
3560 
3561 				error = recv_destroy(hdl, name,
3562 				    strlen(fsname)+1, newname, flags);
3563 				if (error)
3564 					needagain = B_TRUE;
3565 				else
3566 					progress = B_TRUE;
3567 				sprintf(guidname, "%llu",
3568 				    (u_longlong_t)thisguid);
3569 				nvlist_add_boolean(deleted, guidname);
3570 				continue;
3571 			}
3572 
3573 			stream_nvfs = found;
3574 
3575 			if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
3576 			    &props) && 0 == nvlist_lookup_nvlist(props,
3577 			    stream_snapname, &props)) {
3578 				zfs_cmd_t zc = {"\0"};
3579 
3580 				zc.zc_cookie = B_TRUE; /* received */
3581 				(void) snprintf(zc.zc_name, sizeof (zc.zc_name),
3582 				    "%s@%s", fsname, nvpair_name(snapelem));
3583 				zcmd_write_src_nvlist(hdl, &zc, props);
3584 				(void) zfs_ioctl(hdl,
3585 				    ZFS_IOC_SET_PROP, &zc);
3586 				zcmd_free_nvlists(&zc);
3587 			}
3588 
3589 			/* check for different snapname */
3590 			if (strcmp(nvpair_name(snapelem),
3591 			    stream_snapname) != 0) {
3592 				char name[ZFS_MAX_DATASET_NAME_LEN];
3593 				char tryname[ZFS_MAX_DATASET_NAME_LEN];
3594 
3595 				(void) snprintf(name, sizeof (name), "%s@%s",
3596 				    fsname, nvpair_name(snapelem));
3597 				(void) snprintf(tryname, sizeof (name), "%s@%s",
3598 				    fsname, stream_snapname);
3599 
3600 				error = recv_rename(hdl, name, tryname,
3601 				    strlen(fsname)+1, newname, flags);
3602 				if (error)
3603 					needagain = B_TRUE;
3604 				else
3605 					progress = B_TRUE;
3606 			}
3607 
3608 			if (strcmp(stream_snapname, fromsnap) == 0)
3609 				fromguid = thisguid;
3610 		}
3611 
3612 		/* check for delete */
3613 		if (stream_nvfs == NULL) {
3614 			if (!flags->force)
3615 				continue;
3616 
3617 			error = recv_destroy(hdl, fsname, strlen(tofs)+1,
3618 			    newname, flags);
3619 			if (error)
3620 				needagain = B_TRUE;
3621 			else
3622 				progress = B_TRUE;
3623 			sprintf(guidname, "%llu",
3624 			    (u_longlong_t)parent_fromsnap_guid);
3625 			nvlist_add_boolean(deleted, guidname);
3626 			continue;
3627 		}
3628 
3629 		if (fromguid == 0) {
3630 			if (flags->verbose) {
3631 				(void) printf("local fs %s does not have "
3632 				    "fromsnap (%s in stream); must have "
3633 				    "been deleted locally; ignoring\n",
3634 				    fsname, fromsnap);
3635 			}
3636 			continue;
3637 		}
3638 
3639 		stream_fsname = fnvlist_lookup_string(stream_nvfs, "name");
3640 		stream_parent_fromsnap_guid = fnvlist_lookup_uint64(
3641 		    stream_nvfs, "parentfromsnap");
3642 
3643 		s1 = strrchr(fsname, '/');
3644 		s2 = strrchr(stream_fsname, '/');
3645 
3646 		/*
3647 		 * Check if we're going to rename based on parent guid change
3648 		 * and the current parent guid was also deleted. If it was then
3649 		 * rename will fail and is likely unneeded, so avoid this and
3650 		 * force an early retry to determine the new
3651 		 * parent_fromsnap_guid.
3652 		 */
3653 		if (stream_parent_fromsnap_guid != 0 &&
3654 		    parent_fromsnap_guid != 0 &&
3655 		    stream_parent_fromsnap_guid != parent_fromsnap_guid) {
3656 			sprintf(guidname, "%llu",
3657 			    (u_longlong_t)parent_fromsnap_guid);
3658 			if (nvlist_exists(deleted, guidname)) {
3659 				progress = B_TRUE;
3660 				needagain = B_TRUE;
3661 				goto doagain;
3662 			}
3663 		}
3664 
3665 		/*
3666 		 * Check for rename. If the exact receive path is specified, it
3667 		 * does not count as a rename, but we still need to check the
3668 		 * datasets beneath it.
3669 		 */
3670 		if ((stream_parent_fromsnap_guid != 0 &&
3671 		    parent_fromsnap_guid != 0 &&
3672 		    stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
3673 		    ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
3674 		    (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
3675 			nvlist_t *parent;
3676 			char tryname[ZFS_MAX_DATASET_NAME_LEN];
3677 
3678 			parent = fsavl_find(local_avl,
3679 			    stream_parent_fromsnap_guid, NULL);
3680 			/*
3681 			 * NB: parent might not be found if we used the
3682 			 * tosnap for stream_parent_fromsnap_guid,
3683 			 * because the parent is a newly-created fs;
3684 			 * we'll be able to rename it after we recv the
3685 			 * new fs.
3686 			 */
3687 			if (parent != NULL) {
3688 				char *pname;
3689 
3690 				pname = fnvlist_lookup_string(parent, "name");
3691 				(void) snprintf(tryname, sizeof (tryname),
3692 				    "%s%s", pname, strrchr(stream_fsname, '/'));
3693 			} else {
3694 				tryname[0] = '\0';
3695 				if (flags->verbose) {
3696 					(void) printf("local fs %s new parent "
3697 					    "not found\n", fsname);
3698 				}
3699 			}
3700 
3701 			newname[0] = '\0';
3702 
3703 			error = recv_rename(hdl, fsname, tryname,
3704 			    strlen(tofs)+1, newname, flags);
3705 
3706 			if (renamed != NULL && newname[0] != '\0') {
3707 				fnvlist_add_boolean(renamed, newname);
3708 			}
3709 
3710 			if (error)
3711 				needagain = B_TRUE;
3712 			else
3713 				progress = B_TRUE;
3714 		}
3715 	}
3716 
3717 doagain:
3718 	fsavl_destroy(local_avl);
3719 	fnvlist_free(local_nv);
3720 	fnvlist_free(deleted);
3721 
3722 	if (needagain && progress) {
3723 		/* do another pass to fix up temporary names */
3724 		if (flags->verbose)
3725 			(void) printf("another pass:\n");
3726 		goto again;
3727 	}
3728 
3729 	return (needagain || error != 0);
3730 }
3731 
3732 static int
3733 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
3734     recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
3735     char **top_zfs, nvlist_t *cmdprops)
3736 {
3737 	nvlist_t *stream_nv = NULL;
3738 	avl_tree_t *stream_avl = NULL;
3739 	char *fromsnap = NULL;
3740 	char *sendsnap = NULL;
3741 	char *cp;
3742 	char tofs[ZFS_MAX_DATASET_NAME_LEN];
3743 	char sendfs[ZFS_MAX_DATASET_NAME_LEN];
3744 	char errbuf[ERRBUFLEN];
3745 	dmu_replay_record_t drre;
3746 	int error;
3747 	boolean_t anyerr = B_FALSE;
3748 	boolean_t softerr = B_FALSE;
3749 	boolean_t recursive, raw;
3750 
3751 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3752 	    "cannot receive"));
3753 
3754 	assert(drr->drr_type == DRR_BEGIN);
3755 	assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
3756 	assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
3757 	    DMU_COMPOUNDSTREAM);
3758 
3759 	/*
3760 	 * Read in the nvlist from the stream.
3761 	 */
3762 	if (drr->drr_payloadlen != 0) {
3763 		error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
3764 		    &stream_nv, flags->byteswap, zc);
3765 		if (error) {
3766 			error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3767 			goto out;
3768 		}
3769 	}
3770 
3771 	recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
3772 	    ENOENT);
3773 	raw = (nvlist_lookup_boolean(stream_nv, "raw") == 0);
3774 
3775 	if (recursive && strchr(destname, '@')) {
3776 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3777 		    "cannot specify snapshot name for multi-snapshot stream"));
3778 		error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3779 		goto out;
3780 	}
3781 
3782 	/*
3783 	 * Read in the end record and verify checksum.
3784 	 */
3785 	if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
3786 	    flags->byteswap, NULL)))
3787 		goto out;
3788 	if (flags->byteswap) {
3789 		drre.drr_type = BSWAP_32(drre.drr_type);
3790 		drre.drr_u.drr_end.drr_checksum.zc_word[0] =
3791 		    BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
3792 		drre.drr_u.drr_end.drr_checksum.zc_word[1] =
3793 		    BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
3794 		drre.drr_u.drr_end.drr_checksum.zc_word[2] =
3795 		    BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
3796 		drre.drr_u.drr_end.drr_checksum.zc_word[3] =
3797 		    BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
3798 	}
3799 	if (drre.drr_type != DRR_END) {
3800 		error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3801 		goto out;
3802 	}
3803 	if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
3804 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3805 		    "incorrect header checksum"));
3806 		error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3807 		goto out;
3808 	}
3809 
3810 	(void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
3811 
3812 	if (drr->drr_payloadlen != 0) {
3813 		nvlist_t *stream_fss;
3814 
3815 		stream_fss = fnvlist_lookup_nvlist(stream_nv, "fss");
3816 		if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
3817 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3818 			    "couldn't allocate avl tree"));
3819 			error = zfs_error(hdl, EZFS_NOMEM, errbuf);
3820 			goto out;
3821 		}
3822 
3823 		if (fromsnap != NULL && recursive) {
3824 			nvlist_t *renamed = NULL;
3825 			nvpair_t *pair = NULL;
3826 
3827 			(void) strlcpy(tofs, destname, sizeof (tofs));
3828 			if (flags->isprefix) {
3829 				struct drr_begin *drrb = &drr->drr_u.drr_begin;
3830 				int i;
3831 
3832 				if (flags->istail) {
3833 					cp = strrchr(drrb->drr_toname, '/');
3834 					if (cp == NULL) {
3835 						(void) strlcat(tofs, "/",
3836 						    sizeof (tofs));
3837 						i = 0;
3838 					} else {
3839 						i = (cp - drrb->drr_toname);
3840 					}
3841 				} else {
3842 					i = strcspn(drrb->drr_toname, "/@");
3843 				}
3844 				/* zfs_receive_one() will create_parents() */
3845 				(void) strlcat(tofs, &drrb->drr_toname[i],
3846 				    sizeof (tofs));
3847 				*strchr(tofs, '@') = '\0';
3848 			}
3849 
3850 			if (!flags->dryrun && !flags->nomount) {
3851 				renamed = fnvlist_alloc();
3852 			}
3853 
3854 			softerr = recv_incremental_replication(hdl, tofs, flags,
3855 			    stream_nv, stream_avl, renamed);
3856 
3857 			/* Unmount renamed filesystems before receiving. */
3858 			while ((pair = nvlist_next_nvpair(renamed,
3859 			    pair)) != NULL) {
3860 				zfs_handle_t *zhp;
3861 				prop_changelist_t *clp = NULL;
3862 
3863 				zhp = zfs_open(hdl, nvpair_name(pair),
3864 				    ZFS_TYPE_FILESYSTEM);
3865 				if (zhp != NULL) {
3866 					clp = changelist_gather(zhp,
3867 					    ZFS_PROP_MOUNTPOINT, 0,
3868 					    flags->forceunmount ? MS_FORCE : 0);
3869 					zfs_close(zhp);
3870 					if (clp != NULL) {
3871 						softerr |=
3872 						    changelist_prefix(clp);
3873 						changelist_free(clp);
3874 					}
3875 				}
3876 			}
3877 
3878 			fnvlist_free(renamed);
3879 		}
3880 	}
3881 
3882 	/*
3883 	 * Get the fs specified by the first path in the stream (the top level
3884 	 * specified by 'zfs send') and pass it to each invocation of
3885 	 * zfs_receive_one().
3886 	 */
3887 	(void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
3888 	    sizeof (sendfs));
3889 	if ((cp = strchr(sendfs, '@')) != NULL) {
3890 		*cp = '\0';
3891 		/*
3892 		 * Find the "sendsnap", the final snapshot in a replication
3893 		 * stream.  zfs_receive_one() handles certain errors
3894 		 * differently, depending on if the contained stream is the
3895 		 * last one or not.
3896 		 */
3897 		sendsnap = (cp + 1);
3898 	}
3899 
3900 	/* Finally, receive each contained stream */
3901 	do {
3902 		/*
3903 		 * we should figure out if it has a recoverable
3904 		 * error, in which case do a recv_skip() and drive on.
3905 		 * Note, if we fail due to already having this guid,
3906 		 * zfs_receive_one() will take care of it (ie,
3907 		 * recv_skip() and return 0).
3908 		 */
3909 		error = zfs_receive_impl(hdl, destname, NULL, flags, fd,
3910 		    sendfs, stream_nv, stream_avl, top_zfs, sendsnap, cmdprops);
3911 		if (error == ENODATA) {
3912 			error = 0;
3913 			break;
3914 		}
3915 		anyerr |= error;
3916 	} while (error == 0);
3917 
3918 	if (drr->drr_payloadlen != 0 && recursive && fromsnap != NULL) {
3919 		/*
3920 		 * Now that we have the fs's they sent us, try the
3921 		 * renames again.
3922 		 */
3923 		softerr = recv_incremental_replication(hdl, tofs, flags,
3924 		    stream_nv, stream_avl, NULL);
3925 	}
3926 
3927 	if (raw && softerr == 0 && *top_zfs != NULL) {
3928 		softerr = recv_fix_encryption_hierarchy(hdl, *top_zfs,
3929 		    stream_nv);
3930 	}
3931 
3932 out:
3933 	fsavl_destroy(stream_avl);
3934 	fnvlist_free(stream_nv);
3935 	if (softerr)
3936 		error = -2;
3937 	if (anyerr)
3938 		error = -1;
3939 	return (error);
3940 }
3941 
3942 static void
3943 trunc_prop_errs(int truncated)
3944 {
3945 	ASSERT(truncated != 0);
3946 
3947 	if (truncated == 1)
3948 		(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
3949 		    "1 more property could not be set\n"));
3950 	else
3951 		(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
3952 		    "%d more properties could not be set\n"), truncated);
3953 }
3954 
3955 static int
3956 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
3957 {
3958 	dmu_replay_record_t *drr;
3959 	void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
3960 	uint64_t payload_size;
3961 	char errbuf[ERRBUFLEN];
3962 
3963 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3964 	    "cannot receive"));
3965 
3966 	/* XXX would be great to use lseek if possible... */
3967 	drr = buf;
3968 
3969 	while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
3970 	    byteswap, NULL) == 0) {
3971 		if (byteswap)
3972 			drr->drr_type = BSWAP_32(drr->drr_type);
3973 
3974 		switch (drr->drr_type) {
3975 		case DRR_BEGIN:
3976 			if (drr->drr_payloadlen != 0) {
3977 				(void) recv_read(hdl, fd, buf,
3978 				    drr->drr_payloadlen, B_FALSE, NULL);
3979 			}
3980 			break;
3981 
3982 		case DRR_END:
3983 			free(buf);
3984 			return (0);
3985 
3986 		case DRR_OBJECT:
3987 			if (byteswap) {
3988 				drr->drr_u.drr_object.drr_bonuslen =
3989 				    BSWAP_32(drr->drr_u.drr_object.
3990 				    drr_bonuslen);
3991 				drr->drr_u.drr_object.drr_raw_bonuslen =
3992 				    BSWAP_32(drr->drr_u.drr_object.
3993 				    drr_raw_bonuslen);
3994 			}
3995 
3996 			payload_size =
3997 			    DRR_OBJECT_PAYLOAD_SIZE(&drr->drr_u.drr_object);
3998 			(void) recv_read(hdl, fd, buf, payload_size,
3999 			    B_FALSE, NULL);
4000 			break;
4001 
4002 		case DRR_WRITE:
4003 			if (byteswap) {
4004 				drr->drr_u.drr_write.drr_logical_size =
4005 				    BSWAP_64(
4006 				    drr->drr_u.drr_write.drr_logical_size);
4007 				drr->drr_u.drr_write.drr_compressed_size =
4008 				    BSWAP_64(
4009 				    drr->drr_u.drr_write.drr_compressed_size);
4010 			}
4011 			payload_size =
4012 			    DRR_WRITE_PAYLOAD_SIZE(&drr->drr_u.drr_write);
4013 			assert(payload_size <= SPA_MAXBLOCKSIZE);
4014 			(void) recv_read(hdl, fd, buf,
4015 			    payload_size, B_FALSE, NULL);
4016 			break;
4017 		case DRR_SPILL:
4018 			if (byteswap) {
4019 				drr->drr_u.drr_spill.drr_length =
4020 				    BSWAP_64(drr->drr_u.drr_spill.drr_length);
4021 				drr->drr_u.drr_spill.drr_compressed_size =
4022 				    BSWAP_64(drr->drr_u.drr_spill.
4023 				    drr_compressed_size);
4024 			}
4025 
4026 			payload_size =
4027 			    DRR_SPILL_PAYLOAD_SIZE(&drr->drr_u.drr_spill);
4028 			(void) recv_read(hdl, fd, buf, payload_size,
4029 			    B_FALSE, NULL);
4030 			break;
4031 		case DRR_WRITE_EMBEDDED:
4032 			if (byteswap) {
4033 				drr->drr_u.drr_write_embedded.drr_psize =
4034 				    BSWAP_32(drr->drr_u.drr_write_embedded.
4035 				    drr_psize);
4036 			}
4037 			(void) recv_read(hdl, fd, buf,
4038 			    P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
4039 			    8), B_FALSE, NULL);
4040 			break;
4041 		case DRR_OBJECT_RANGE:
4042 		case DRR_WRITE_BYREF:
4043 		case DRR_FREEOBJECTS:
4044 		case DRR_FREE:
4045 			break;
4046 
4047 		default:
4048 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4049 			    "invalid record type"));
4050 			free(buf);
4051 			return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
4052 		}
4053 	}
4054 
4055 	free(buf);
4056 	return (-1);
4057 }
4058 
4059 static void
4060 recv_ecksum_set_aux(libzfs_handle_t *hdl, const char *target_snap,
4061     boolean_t resumable, boolean_t checksum)
4062 {
4063 	char target_fs[ZFS_MAX_DATASET_NAME_LEN];
4064 
4065 	zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, (checksum ?
4066 	    "checksum mismatch" : "incomplete stream")));
4067 
4068 	if (!resumable)
4069 		return;
4070 	(void) strlcpy(target_fs, target_snap, sizeof (target_fs));
4071 	*strchr(target_fs, '@') = '\0';
4072 	zfs_handle_t *zhp = zfs_open(hdl, target_fs,
4073 	    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
4074 	if (zhp == NULL)
4075 		return;
4076 
4077 	char token_buf[ZFS_MAXPROPLEN];
4078 	int error = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN,
4079 	    token_buf, sizeof (token_buf),
4080 	    NULL, NULL, 0, B_TRUE);
4081 	if (error == 0) {
4082 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4083 		    "checksum mismatch or incomplete stream.\n"
4084 		    "Partially received snapshot is saved.\n"
4085 		    "A resuming stream can be generated on the sending "
4086 		    "system by running:\n"
4087 		    "    zfs send -t %s"),
4088 		    token_buf);
4089 	}
4090 	zfs_close(zhp);
4091 }
4092 
4093 /*
4094  * Prepare a new nvlist of properties that are to override (-o) or be excluded
4095  * (-x) from the received dataset
4096  * recvprops: received properties from the send stream
4097  * cmdprops: raw input properties from command line
4098  * origprops: properties, both locally-set and received, currently set on the
4099  *            target dataset if it exists, NULL otherwise.
4100  * oxprops: valid output override (-o) and excluded (-x) properties
4101  */
4102 static int
4103 zfs_setup_cmdline_props(libzfs_handle_t *hdl, zfs_type_t type,
4104     char *fsname, boolean_t zoned, boolean_t recursive, boolean_t newfs,
4105     boolean_t raw, boolean_t toplevel, nvlist_t *recvprops, nvlist_t *cmdprops,
4106     nvlist_t *origprops, nvlist_t **oxprops, uint8_t **wkeydata_out,
4107     uint_t *wkeylen_out, const char *errbuf)
4108 {
4109 	nvpair_t *nvp;
4110 	nvlist_t *oprops, *voprops;
4111 	zfs_handle_t *zhp = NULL;
4112 	zpool_handle_t *zpool_hdl = NULL;
4113 	char *cp;
4114 	int ret = 0;
4115 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
4116 
4117 	if (nvlist_empty(cmdprops))
4118 		return (0); /* No properties to override or exclude */
4119 
4120 	*oxprops = fnvlist_alloc();
4121 	oprops = fnvlist_alloc();
4122 
4123 	strlcpy(namebuf, fsname, ZFS_MAX_DATASET_NAME_LEN);
4124 
4125 	/*
4126 	 * Get our dataset handle. The target dataset may not exist yet.
4127 	 */
4128 	if (zfs_dataset_exists(hdl, namebuf, ZFS_TYPE_DATASET)) {
4129 		zhp = zfs_open(hdl, namebuf, ZFS_TYPE_DATASET);
4130 		if (zhp == NULL) {
4131 			ret = -1;
4132 			goto error;
4133 		}
4134 	}
4135 
4136 	/* open the zpool handle */
4137 	cp = strchr(namebuf, '/');
4138 	if (cp != NULL)
4139 		*cp = '\0';
4140 	zpool_hdl = zpool_open(hdl, namebuf);
4141 	if (zpool_hdl == NULL) {
4142 		ret = -1;
4143 		goto error;
4144 	}
4145 
4146 	/* restore namebuf to match fsname for later use */
4147 	if (cp != NULL)
4148 		*cp = '/';
4149 
4150 	/*
4151 	 * first iteration: process excluded (-x) properties now and gather
4152 	 * added (-o) properties to be later processed by zfs_valid_proplist()
4153 	 */
4154 	nvp = NULL;
4155 	while ((nvp = nvlist_next_nvpair(cmdprops, nvp)) != NULL) {
4156 		const char *name = nvpair_name(nvp);
4157 		zfs_prop_t prop = zfs_name_to_prop(name);
4158 
4159 		/*
4160 		 * It turns out, if we don't normalize "aliased" names
4161 		 * e.g. compress= against the "real" names (e.g. compression)
4162 		 * here, then setting/excluding them does not work as
4163 		 * intended.
4164 		 *
4165 		 * But since user-defined properties wouldn't have a valid
4166 		 * mapping here, we do this conditional dance.
4167 		 */
4168 		const char *newname = name;
4169 		if (prop >= ZFS_PROP_TYPE)
4170 			newname = zfs_prop_to_name(prop);
4171 
4172 		/* "origin" is processed separately, don't handle it here */
4173 		if (prop == ZFS_PROP_ORIGIN)
4174 			continue;
4175 
4176 		/* raw streams can't override encryption properties */
4177 		if ((zfs_prop_encryption_key_param(prop) ||
4178 		    prop == ZFS_PROP_ENCRYPTION) && raw) {
4179 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4180 			    "encryption property '%s' cannot "
4181 			    "be set or excluded for raw streams."), name);
4182 			ret = zfs_error(hdl, EZFS_BADPROP, errbuf);
4183 			goto error;
4184 		}
4185 
4186 		/*
4187 		 * For plain replicated send, we can ignore encryption
4188 		 * properties other than first stream
4189 		 */
4190 		if ((zfs_prop_encryption_key_param(prop) || prop ==
4191 		    ZFS_PROP_ENCRYPTION) && !newfs && recursive && !raw) {
4192 			continue;
4193 		}
4194 
4195 		/* incremental streams can only exclude encryption properties */
4196 		if ((zfs_prop_encryption_key_param(prop) ||
4197 		    prop == ZFS_PROP_ENCRYPTION) && !newfs &&
4198 		    nvpair_type(nvp) != DATA_TYPE_BOOLEAN) {
4199 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4200 			    "encryption property '%s' cannot "
4201 			    "be set for incremental streams."), name);
4202 			ret = zfs_error(hdl, EZFS_BADPROP, errbuf);
4203 			goto error;
4204 		}
4205 
4206 		switch (nvpair_type(nvp)) {
4207 		case DATA_TYPE_BOOLEAN: /* -x property */
4208 			/*
4209 			 * DATA_TYPE_BOOLEAN is the way we're asked to "exclude"
4210 			 * a property: this is done by forcing an explicit
4211 			 * inherit on the destination so the effective value is
4212 			 * not the one we received from the send stream.
4213 			 */
4214 			if (!zfs_prop_valid_for_type(prop, type, B_FALSE) &&
4215 			    !zfs_prop_user(name)) {
4216 				(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
4217 				    "Warning: %s: property '%s' does not "
4218 				    "apply to datasets of this type\n"),
4219 				    fsname, name);
4220 				continue;
4221 			}
4222 			/*
4223 			 * We do this only if the property is not already
4224 			 * locally-set, in which case its value will take
4225 			 * priority over the received anyway.
4226 			 */
4227 			if (nvlist_exists(origprops, newname)) {
4228 				nvlist_t *attrs;
4229 				char *source = NULL;
4230 
4231 				attrs = fnvlist_lookup_nvlist(origprops,
4232 				    newname);
4233 				if (nvlist_lookup_string(attrs,
4234 				    ZPROP_SOURCE, &source) == 0 &&
4235 				    strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0)
4236 					continue;
4237 			}
4238 			/*
4239 			 * We can't force an explicit inherit on non-inheritable
4240 			 * properties: if we're asked to exclude this kind of
4241 			 * values we remove them from "recvprops" input nvlist.
4242 			 */
4243 			if (!zfs_prop_user(name) && /* can be inherited too */
4244 			    !zfs_prop_inheritable(prop) &&
4245 			    nvlist_exists(recvprops, newname))
4246 				fnvlist_remove(recvprops, newname);
4247 			else
4248 				fnvlist_add_boolean(*oxprops, newname);
4249 			break;
4250 		case DATA_TYPE_STRING: /* -o property=value */
4251 			/*
4252 			 * we're trying to override a property that does not
4253 			 * make sense for this type of dataset, but we don't
4254 			 * want to fail if the receive is recursive: this comes
4255 			 * in handy when the send stream contains, for
4256 			 * instance, a child ZVOL and we're trying to receive
4257 			 * it with "-o atime=on"
4258 			 */
4259 			if (!zfs_prop_valid_for_type(prop, type, B_FALSE) &&
4260 			    !zfs_prop_user(name)) {
4261 				if (recursive)
4262 					continue;
4263 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4264 				    "property '%s' does not apply to datasets "
4265 				    "of this type"), name);
4266 				ret = zfs_error(hdl, EZFS_BADPROP, errbuf);
4267 				goto error;
4268 			}
4269 			fnvlist_add_string(oprops, newname,
4270 			    fnvpair_value_string(nvp));
4271 			break;
4272 		default:
4273 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4274 			    "property '%s' must be a string or boolean"), name);
4275 			ret = zfs_error(hdl, EZFS_BADPROP, errbuf);
4276 			goto error;
4277 		}
4278 	}
4279 
4280 	if (toplevel) {
4281 		/* convert override strings properties to native */
4282 		if ((voprops = zfs_valid_proplist(hdl, ZFS_TYPE_DATASET,
4283 		    oprops, zoned, zhp, zpool_hdl, B_FALSE, errbuf)) == NULL) {
4284 			ret = zfs_error(hdl, EZFS_BADPROP, errbuf);
4285 			goto error;
4286 		}
4287 
4288 		/*
4289 		 * zfs_crypto_create() requires the parent name. Get it
4290 		 * by truncating the fsname copy stored in namebuf.
4291 		 */
4292 		cp = strrchr(namebuf, '/');
4293 		if (cp != NULL)
4294 			*cp = '\0';
4295 
4296 		if (!raw && !(!newfs && recursive) &&
4297 		    zfs_crypto_create(hdl, namebuf, voprops, NULL,
4298 		    B_FALSE, wkeydata_out, wkeylen_out) != 0) {
4299 			fnvlist_free(voprops);
4300 			ret = zfs_error(hdl, EZFS_CRYPTOFAILED, errbuf);
4301 			goto error;
4302 		}
4303 
4304 		/* second pass: process "-o" properties */
4305 		fnvlist_merge(*oxprops, voprops);
4306 		fnvlist_free(voprops);
4307 	} else {
4308 		/* override props on child dataset are inherited */
4309 		nvp = NULL;
4310 		while ((nvp = nvlist_next_nvpair(oprops, nvp)) != NULL) {
4311 			const char *name = nvpair_name(nvp);
4312 			fnvlist_add_boolean(*oxprops, name);
4313 		}
4314 	}
4315 
4316 error:
4317 	if (zhp != NULL)
4318 		zfs_close(zhp);
4319 	if (zpool_hdl != NULL)
4320 		zpool_close(zpool_hdl);
4321 	fnvlist_free(oprops);
4322 	return (ret);
4323 }
4324 
4325 /*
4326  * Restores a backup of tosnap from the file descriptor specified by infd.
4327  */
4328 static int
4329 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
4330     const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr,
4331     dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv,
4332     avl_tree_t *stream_avl, char **top_zfs,
4333     const char *finalsnap, nvlist_t *cmdprops)
4334 {
4335 	struct timespec begin_time;
4336 	int ioctl_err, ioctl_errno, err;
4337 	char *cp;
4338 	struct drr_begin *drrb = &drr->drr_u.drr_begin;
4339 	char errbuf[ERRBUFLEN];
4340 	const char *chopprefix;
4341 	boolean_t newfs = B_FALSE;
4342 	boolean_t stream_wantsnewfs, stream_resumingnewfs;
4343 	boolean_t newprops = B_FALSE;
4344 	uint64_t read_bytes = 0;
4345 	uint64_t errflags = 0;
4346 	uint64_t parent_snapguid = 0;
4347 	prop_changelist_t *clp = NULL;
4348 	nvlist_t *snapprops_nvlist = NULL;
4349 	nvlist_t *snapholds_nvlist = NULL;
4350 	zprop_errflags_t prop_errflags;
4351 	nvlist_t *prop_errors = NULL;
4352 	boolean_t recursive;
4353 	char *snapname = NULL;
4354 	char destsnap[MAXPATHLEN * 2];
4355 	char origin[MAXNAMELEN] = {0};
4356 	char name[MAXPATHLEN];
4357 	char tmp_keylocation[MAXNAMELEN] = {0};
4358 	nvlist_t *rcvprops = NULL; /* props received from the send stream */
4359 	nvlist_t *oxprops = NULL; /* override (-o) and exclude (-x) props */
4360 	nvlist_t *origprops = NULL; /* original props (if destination exists) */
4361 	zfs_type_t type = ZFS_TYPE_INVALID;
4362 	boolean_t toplevel = B_FALSE;
4363 	boolean_t zoned = B_FALSE;
4364 	boolean_t hastoken = B_FALSE;
4365 	boolean_t redacted;
4366 	uint8_t *wkeydata = NULL;
4367 	uint_t wkeylen = 0;
4368 
4369 #ifndef CLOCK_MONOTONIC_RAW
4370 #define	CLOCK_MONOTONIC_RAW CLOCK_MONOTONIC
4371 #endif
4372 	clock_gettime(CLOCK_MONOTONIC_RAW, &begin_time);
4373 
4374 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4375 	    "cannot receive"));
4376 
4377 	recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
4378 	    ENOENT);
4379 
4380 	/* Did the user request holds be skipped via zfs recv -k? */
4381 	boolean_t holds = flags->holds && !flags->skipholds;
4382 
4383 	if (stream_avl != NULL) {
4384 		char *keylocation = NULL;
4385 		nvlist_t *lookup = NULL;
4386 		nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
4387 		    &snapname);
4388 
4389 		(void) nvlist_lookup_uint64(fs, "parentfromsnap",
4390 		    &parent_snapguid);
4391 		err = nvlist_lookup_nvlist(fs, "props", &rcvprops);
4392 		if (err) {
4393 			rcvprops = fnvlist_alloc();
4394 			newprops = B_TRUE;
4395 		}
4396 
4397 		/*
4398 		 * The keylocation property may only be set on encryption roots,
4399 		 * but this dataset might not become an encryption root until
4400 		 * recv_fix_encryption_hierarchy() is called. That function
4401 		 * will fixup the keylocation anyway, so we temporarily unset
4402 		 * the keylocation for now to avoid any errors from the receive
4403 		 * ioctl.
4404 		 */
4405 		err = nvlist_lookup_string(rcvprops,
4406 		    zfs_prop_to_name(ZFS_PROP_KEYLOCATION), &keylocation);
4407 		if (err == 0) {
4408 			strlcpy(tmp_keylocation, keylocation, MAXNAMELEN);
4409 			(void) nvlist_remove_all(rcvprops,
4410 			    zfs_prop_to_name(ZFS_PROP_KEYLOCATION));
4411 		}
4412 
4413 		if (flags->canmountoff) {
4414 			fnvlist_add_uint64(rcvprops,
4415 			    zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0);
4416 		} else if (newprops) {	/* nothing in rcvprops, eliminate it */
4417 			fnvlist_free(rcvprops);
4418 			rcvprops = NULL;
4419 			newprops = B_FALSE;
4420 		}
4421 		if (0 == nvlist_lookup_nvlist(fs, "snapprops", &lookup)) {
4422 			snapprops_nvlist = fnvlist_lookup_nvlist(lookup,
4423 			    snapname);
4424 		}
4425 		if (holds) {
4426 			if (0 == nvlist_lookup_nvlist(fs, "snapholds",
4427 			    &lookup)) {
4428 				snapholds_nvlist = fnvlist_lookup_nvlist(
4429 				    lookup, snapname);
4430 			}
4431 		}
4432 	}
4433 
4434 	cp = NULL;
4435 
4436 	/*
4437 	 * Determine how much of the snapshot name stored in the stream
4438 	 * we are going to tack on to the name they specified on the
4439 	 * command line, and how much we are going to chop off.
4440 	 *
4441 	 * If they specified a snapshot, chop the entire name stored in
4442 	 * the stream.
4443 	 */
4444 	if (flags->istail) {
4445 		/*
4446 		 * A filesystem was specified with -e. We want to tack on only
4447 		 * the tail of the sent snapshot path.
4448 		 */
4449 		if (strchr(tosnap, '@')) {
4450 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
4451 			    "argument - snapshot not allowed with -e"));
4452 			err = zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4453 			goto out;
4454 		}
4455 
4456 		chopprefix = strrchr(sendfs, '/');
4457 
4458 		if (chopprefix == NULL) {
4459 			/*
4460 			 * The tail is the poolname, so we need to
4461 			 * prepend a path separator.
4462 			 */
4463 			int len = strlen(drrb->drr_toname);
4464 			cp = umem_alloc(len + 2, UMEM_NOFAIL);
4465 			cp[0] = '/';
4466 			(void) strcpy(&cp[1], drrb->drr_toname);
4467 			chopprefix = cp;
4468 		} else {
4469 			chopprefix = drrb->drr_toname + (chopprefix - sendfs);
4470 		}
4471 	} else if (flags->isprefix) {
4472 		/*
4473 		 * A filesystem was specified with -d. We want to tack on
4474 		 * everything but the first element of the sent snapshot path
4475 		 * (all but the pool name).
4476 		 */
4477 		if (strchr(tosnap, '@')) {
4478 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
4479 			    "argument - snapshot not allowed with -d"));
4480 			err = zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4481 			goto out;
4482 		}
4483 
4484 		chopprefix = strchr(drrb->drr_toname, '/');
4485 		if (chopprefix == NULL)
4486 			chopprefix = strchr(drrb->drr_toname, '@');
4487 	} else if (strchr(tosnap, '@') == NULL) {
4488 		/*
4489 		 * If a filesystem was specified without -d or -e, we want to
4490 		 * tack on everything after the fs specified by 'zfs send'.
4491 		 */
4492 		chopprefix = drrb->drr_toname + strlen(sendfs);
4493 	} else {
4494 		/* A snapshot was specified as an exact path (no -d or -e). */
4495 		if (recursive) {
4496 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4497 			    "cannot specify snapshot name for multi-snapshot "
4498 			    "stream"));
4499 			err = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
4500 			goto out;
4501 		}
4502 		chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
4503 	}
4504 
4505 	ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
4506 	ASSERT(chopprefix > drrb->drr_toname || strchr(sendfs, '/') == NULL);
4507 	ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname) ||
4508 	    strchr(sendfs, '/') == NULL);
4509 	ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
4510 	    chopprefix[0] == '\0');
4511 
4512 	/*
4513 	 * Determine name of destination snapshot.
4514 	 */
4515 	(void) strlcpy(destsnap, tosnap, sizeof (destsnap));
4516 	(void) strlcat(destsnap, chopprefix, sizeof (destsnap));
4517 	if (cp != NULL)
4518 		umem_free(cp, strlen(cp) + 1);
4519 	if (!zfs_name_valid(destsnap, ZFS_TYPE_SNAPSHOT)) {
4520 		err = zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4521 		goto out;
4522 	}
4523 
4524 	/*
4525 	 * Determine the name of the origin snapshot.
4526 	 */
4527 	if (originsnap) {
4528 		(void) strlcpy(origin, originsnap, sizeof (origin));
4529 		if (flags->verbose)
4530 			(void) printf("using provided clone origin %s\n",
4531 			    origin);
4532 	} else if (drrb->drr_flags & DRR_FLAG_CLONE) {
4533 		if (guid_to_name(hdl, destsnap,
4534 		    drrb->drr_fromguid, B_FALSE, origin) != 0) {
4535 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4536 			    "local origin for clone %s does not exist"),
4537 			    destsnap);
4538 			err = zfs_error(hdl, EZFS_NOENT, errbuf);
4539 			goto out;
4540 		}
4541 		if (flags->verbose)
4542 			(void) printf("found clone origin %s\n", origin);
4543 	}
4544 
4545 	if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
4546 	    DMU_BACKUP_FEATURE_DEDUP)) {
4547 		(void) fprintf(stderr,
4548 		    gettext("ERROR: \"zfs receive\" no longer supports "
4549 		    "deduplicated send streams.  Use\n"
4550 		    "the \"zstream redup\" command to convert this stream "
4551 		    "to a regular,\n"
4552 		    "non-deduplicated stream.\n"));
4553 		err = zfs_error(hdl, EZFS_NOTSUP, errbuf);
4554 		goto out;
4555 	}
4556 
4557 	boolean_t resuming = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
4558 	    DMU_BACKUP_FEATURE_RESUMING;
4559 	boolean_t raw = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
4560 	    DMU_BACKUP_FEATURE_RAW;
4561 	boolean_t embedded = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
4562 	    DMU_BACKUP_FEATURE_EMBED_DATA;
4563 	stream_wantsnewfs = (drrb->drr_fromguid == 0 ||
4564 	    (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && !resuming;
4565 	stream_resumingnewfs = (drrb->drr_fromguid == 0 ||
4566 	    (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && resuming;
4567 
4568 	if (stream_wantsnewfs) {
4569 		/*
4570 		 * if the parent fs does not exist, look for it based on
4571 		 * the parent snap GUID
4572 		 */
4573 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4574 		    "cannot receive new filesystem stream"));
4575 
4576 		(void) strlcpy(name, destsnap, sizeof (name));
4577 		cp = strrchr(name, '/');
4578 		if (cp)
4579 			*cp = '\0';
4580 		if (cp &&
4581 		    !zfs_dataset_exists(hdl, name, ZFS_TYPE_DATASET)) {
4582 			char suffix[ZFS_MAX_DATASET_NAME_LEN];
4583 			(void) strlcpy(suffix, strrchr(destsnap, '/'),
4584 			    sizeof (suffix));
4585 			if (guid_to_name(hdl, name, parent_snapguid,
4586 			    B_FALSE, destsnap) == 0) {
4587 				*strchr(destsnap, '@') = '\0';
4588 				(void) strlcat(destsnap, suffix,
4589 				    sizeof (destsnap) - strlen(destsnap));
4590 			}
4591 		}
4592 	} else {
4593 		/*
4594 		 * If the fs does not exist, look for it based on the
4595 		 * fromsnap GUID.
4596 		 */
4597 		if (resuming) {
4598 			(void) snprintf(errbuf, sizeof (errbuf),
4599 			    dgettext(TEXT_DOMAIN,
4600 			    "cannot receive resume stream"));
4601 		} else {
4602 			(void) snprintf(errbuf, sizeof (errbuf),
4603 			    dgettext(TEXT_DOMAIN,
4604 			    "cannot receive incremental stream"));
4605 		}
4606 
4607 		(void) strlcpy(name, destsnap, sizeof (name));
4608 		*strchr(name, '@') = '\0';
4609 
4610 		/*
4611 		 * If the exact receive path was specified and this is the
4612 		 * topmost path in the stream, then if the fs does not exist we
4613 		 * should look no further.
4614 		 */
4615 		if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
4616 		    strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
4617 		    !zfs_dataset_exists(hdl, name, ZFS_TYPE_DATASET)) {
4618 			char snap[ZFS_MAX_DATASET_NAME_LEN];
4619 			(void) strlcpy(snap, strchr(destsnap, '@'),
4620 			    sizeof (snap));
4621 			if (guid_to_name(hdl, name, drrb->drr_fromguid,
4622 			    B_FALSE, destsnap) == 0) {
4623 				*strchr(destsnap, '@') = '\0';
4624 				(void) strlcat(destsnap, snap,
4625 				    sizeof (destsnap) - strlen(destsnap));
4626 			}
4627 		}
4628 	}
4629 
4630 	(void) strlcpy(name, destsnap, sizeof (name));
4631 	*strchr(name, '@') = '\0';
4632 
4633 	redacted = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
4634 	    DMU_BACKUP_FEATURE_REDACTED;
4635 
4636 	if (flags->heal) {
4637 		if (flags->isprefix || flags->istail || flags->force ||
4638 		    flags->canmountoff || flags->resumable || flags->nomount ||
4639 		    flags->skipholds) {
4640 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4641 			    "corrective recv can not be used when combined with"
4642 			    " this flag"));
4643 			err = zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4644 			goto out;
4645 		}
4646 		uint64_t guid =
4647 		    get_snap_guid(hdl, name, strchr(destsnap, '@') + 1);
4648 		if (guid == 0) {
4649 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4650 			    "corrective recv must specify an existing snapshot"
4651 			    " to heal"));
4652 			err = zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4653 			goto out;
4654 		} else if (guid != drrb->drr_toguid) {
4655 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4656 			    "local snapshot doesn't match the snapshot"
4657 			    " in the provided stream"));
4658 			err = zfs_error(hdl, EZFS_WRONG_PARENT, errbuf);
4659 			goto out;
4660 		}
4661 	} else if (zfs_dataset_exists(hdl, name, ZFS_TYPE_DATASET)) {
4662 		zfs_cmd_t zc = {"\0"};
4663 		zfs_handle_t *zhp = NULL;
4664 		boolean_t encrypted;
4665 
4666 		(void) strcpy(zc.zc_name, name);
4667 
4668 		/*
4669 		 * Destination fs exists.  It must be one of these cases:
4670 		 *  - an incremental send stream
4671 		 *  - the stream specifies a new fs (full stream or clone)
4672 		 *    and they want us to blow away the existing fs (and
4673 		 *    have therefore specified -F and removed any snapshots)
4674 		 *  - we are resuming a failed receive.
4675 		 */
4676 		if (stream_wantsnewfs) {
4677 			boolean_t is_volume = drrb->drr_type == DMU_OST_ZVOL;
4678 			if (!flags->force) {
4679 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4680 				    "destination '%s' exists\n"
4681 				    "must specify -F to overwrite it"), name);
4682 				err = zfs_error(hdl, EZFS_EXISTS, errbuf);
4683 				goto out;
4684 			}
4685 			if (zfs_ioctl(hdl, ZFS_IOC_SNAPSHOT_LIST_NEXT,
4686 			    &zc) == 0) {
4687 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4688 				    "destination has snapshots (eg. %s)\n"
4689 				    "must destroy them to overwrite it"),
4690 				    zc.zc_name);
4691 				err = zfs_error(hdl, EZFS_EXISTS, errbuf);
4692 				goto out;
4693 			}
4694 			if (is_volume && strrchr(name, '/') == NULL) {
4695 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4696 				    "destination %s is the root dataset\n"
4697 				    "cannot overwrite with a ZVOL"),
4698 				    name);
4699 				err = zfs_error(hdl, EZFS_EXISTS, errbuf);
4700 				goto out;
4701 			}
4702 			if (is_volume &&
4703 			    zfs_ioctl(hdl, ZFS_IOC_DATASET_LIST_NEXT,
4704 			    &zc) == 0) {
4705 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4706 				    "destination has children (eg. %s)\n"
4707 				    "cannot overwrite with a ZVOL"),
4708 				    zc.zc_name);
4709 				err = zfs_error(hdl, EZFS_WRONG_PARENT, errbuf);
4710 				goto out;
4711 			}
4712 		}
4713 
4714 		if ((zhp = zfs_open(hdl, name,
4715 		    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
4716 			err = -1;
4717 			goto out;
4718 		}
4719 
4720 		/*
4721 		 * When receiving full/newfs on existing dataset, then it
4722 		 * should be done with "-F" flag. Its enforced for initial
4723 		 * receive in previous checks in this function.
4724 		 * Similarly, on resuming full/newfs recv on existing dataset,
4725 		 * it should be done with "-F" flag.
4726 		 *
4727 		 * When dataset doesn't exist, then full/newfs recv is done on
4728 		 * newly created dataset and it's marked INCONSISTENT. But
4729 		 * When receiving on existing dataset, recv is first done on
4730 		 * %recv and its marked INCONSISTENT. Existing dataset is not
4731 		 * marked INCONSISTENT.
4732 		 * Resume of full/newfs receive with dataset not INCONSISTENT
4733 		 * indicates that its resuming newfs on existing dataset. So,
4734 		 * enforce "-F" flag in this case.
4735 		 */
4736 		if (stream_resumingnewfs &&
4737 		    !zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT) &&
4738 		    !flags->force) {
4739 			zfs_close(zhp);
4740 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4741 			    "Resuming recv on existing destination '%s'\n"
4742 			    "must specify -F to overwrite it"), name);
4743 			err = zfs_error(hdl, EZFS_RESUME_EXISTS, errbuf);
4744 			goto out;
4745 		}
4746 
4747 		if (stream_wantsnewfs &&
4748 		    zhp->zfs_dmustats.dds_origin[0]) {
4749 			zfs_close(zhp);
4750 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4751 			    "destination '%s' is a clone\n"
4752 			    "must destroy it to overwrite it"), name);
4753 			err = zfs_error(hdl, EZFS_EXISTS, errbuf);
4754 			goto out;
4755 		}
4756 
4757 		/*
4758 		 * Raw sends can not be performed as an incremental on top
4759 		 * of existing unencrypted datasets. zfs recv -F can't be
4760 		 * used to blow away an existing encrypted filesystem. This
4761 		 * is because it would require the dsl dir to point to the
4762 		 * new key (or lack of a key) and the old key at the same
4763 		 * time. The -F flag may still be used for deleting
4764 		 * intermediate snapshots that would otherwise prevent the
4765 		 * receive from working.
4766 		 */
4767 		encrypted = zfs_prop_get_int(zhp, ZFS_PROP_ENCRYPTION) !=
4768 		    ZIO_CRYPT_OFF;
4769 		if (!stream_wantsnewfs && !encrypted && raw) {
4770 			zfs_close(zhp);
4771 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4772 			    "cannot perform raw receive on top of "
4773 			    "existing unencrypted dataset"));
4774 			err = zfs_error(hdl, EZFS_BADRESTORE, errbuf);
4775 			goto out;
4776 		}
4777 
4778 		if (stream_wantsnewfs && flags->force &&
4779 		    ((raw && !encrypted) || encrypted)) {
4780 			zfs_close(zhp);
4781 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4782 			    "zfs receive -F cannot be used to destroy an "
4783 			    "encrypted filesystem or overwrite an "
4784 			    "unencrypted one with an encrypted one"));
4785 			err = zfs_error(hdl, EZFS_BADRESTORE, errbuf);
4786 			goto out;
4787 		}
4788 
4789 		if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
4790 		    (stream_wantsnewfs || stream_resumingnewfs)) {
4791 			/* We can't do online recv in this case */
4792 			clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
4793 			    flags->forceunmount ? MS_FORCE : 0);
4794 			if (clp == NULL) {
4795 				zfs_close(zhp);
4796 				err = -1;
4797 				goto out;
4798 			}
4799 			if (changelist_prefix(clp) != 0) {
4800 				changelist_free(clp);
4801 				zfs_close(zhp);
4802 				err = -1;
4803 				goto out;
4804 			}
4805 		}
4806 
4807 		/*
4808 		 * If we are resuming a newfs, set newfs here so that we will
4809 		 * mount it if the recv succeeds this time.  We can tell
4810 		 * that it was a newfs on the first recv because the fs
4811 		 * itself will be inconsistent (if the fs existed when we
4812 		 * did the first recv, we would have received it into
4813 		 * .../%recv).
4814 		 */
4815 		if (resuming && zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT))
4816 			newfs = B_TRUE;
4817 
4818 		/* we want to know if we're zoned when validating -o|-x props */
4819 		zoned = zfs_prop_get_int(zhp, ZFS_PROP_ZONED);
4820 
4821 		/* may need this info later, get it now we have zhp around */
4822 		if (zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN, NULL, 0,
4823 		    NULL, NULL, 0, B_TRUE) == 0)
4824 			hastoken = B_TRUE;
4825 
4826 		/* gather existing properties on destination */
4827 		origprops = fnvlist_alloc();
4828 		fnvlist_merge(origprops, zhp->zfs_props);
4829 		fnvlist_merge(origprops, zhp->zfs_user_props);
4830 
4831 		zfs_close(zhp);
4832 	} else {
4833 		zfs_handle_t *zhp;
4834 
4835 		/*
4836 		 * Destination filesystem does not exist.  Therefore we better
4837 		 * be creating a new filesystem (either from a full backup, or
4838 		 * a clone).  It would therefore be invalid if the user
4839 		 * specified only the pool name (i.e. if the destination name
4840 		 * contained no slash character).
4841 		 */
4842 		cp = strrchr(name, '/');
4843 
4844 		if (!stream_wantsnewfs || cp == NULL) {
4845 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4846 			    "destination '%s' does not exist"), name);
4847 			err = zfs_error(hdl, EZFS_NOENT, errbuf);
4848 			goto out;
4849 		}
4850 
4851 		/*
4852 		 * Trim off the final dataset component so we perform the
4853 		 * recvbackup ioctl to the filesystems's parent.
4854 		 */
4855 		*cp = '\0';
4856 
4857 		if (flags->isprefix && !flags->istail && !flags->dryrun &&
4858 		    create_parents(hdl, destsnap, strlen(tosnap)) != 0) {
4859 			err = zfs_error(hdl, EZFS_BADRESTORE, errbuf);
4860 			goto out;
4861 		}
4862 
4863 		/* validate parent */
4864 		zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
4865 		if (zhp == NULL) {
4866 			err = zfs_error(hdl, EZFS_BADRESTORE, errbuf);
4867 			goto out;
4868 		}
4869 		if (zfs_get_type(zhp) != ZFS_TYPE_FILESYSTEM) {
4870 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4871 			    "parent '%s' is not a filesystem"), name);
4872 			err = zfs_error(hdl, EZFS_WRONG_PARENT, errbuf);
4873 			zfs_close(zhp);
4874 			goto out;
4875 		}
4876 
4877 		zfs_close(zhp);
4878 
4879 		newfs = B_TRUE;
4880 		*cp = '/';
4881 	}
4882 
4883 	if (flags->verbose) {
4884 		(void) printf("%s %s%s stream of %s into %s\n",
4885 		    flags->dryrun ? "would receive" : "receiving",
4886 		    flags->heal ? " corrective" : "",
4887 		    drrb->drr_fromguid ? "incremental" : "full",
4888 		    drrb->drr_toname, destsnap);
4889 		(void) fflush(stdout);
4890 	}
4891 
4892 	/*
4893 	 * If this is the top-level dataset, record it so we can use it
4894 	 * for recursive operations later.
4895 	 */
4896 	if (top_zfs != NULL &&
4897 	    (*top_zfs == NULL || strcmp(*top_zfs, name) == 0)) {
4898 		toplevel = B_TRUE;
4899 		if (*top_zfs == NULL)
4900 			*top_zfs = zfs_strdup(hdl, name);
4901 	}
4902 
4903 	if (drrb->drr_type == DMU_OST_ZVOL) {
4904 		type = ZFS_TYPE_VOLUME;
4905 	} else if (drrb->drr_type == DMU_OST_ZFS) {
4906 		type = ZFS_TYPE_FILESYSTEM;
4907 	} else {
4908 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4909 		    "invalid record type: 0x%d"), drrb->drr_type);
4910 		err = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
4911 		goto out;
4912 	}
4913 	if ((err = zfs_setup_cmdline_props(hdl, type, name, zoned, recursive,
4914 	    stream_wantsnewfs, raw, toplevel, rcvprops, cmdprops, origprops,
4915 	    &oxprops, &wkeydata, &wkeylen, errbuf)) != 0)
4916 		goto out;
4917 
4918 	/*
4919 	 * When sending with properties (zfs send -p), the encryption property
4920 	 * is not included because it is a SETONCE property and therefore
4921 	 * treated as read only. However, we are always able to determine its
4922 	 * value because raw sends will include it in the DRR_BDEGIN payload
4923 	 * and non-raw sends with properties are not allowed for encrypted
4924 	 * datasets. Therefore, if this is a non-raw properties stream, we can
4925 	 * infer that the value should be ZIO_CRYPT_OFF and manually add that
4926 	 * to the received properties.
4927 	 */
4928 	if (stream_wantsnewfs && !raw && rcvprops != NULL &&
4929 	    !nvlist_exists(cmdprops, zfs_prop_to_name(ZFS_PROP_ENCRYPTION))) {
4930 		if (oxprops == NULL)
4931 			oxprops = fnvlist_alloc();
4932 		fnvlist_add_uint64(oxprops,
4933 		    zfs_prop_to_name(ZFS_PROP_ENCRYPTION), ZIO_CRYPT_OFF);
4934 	}
4935 
4936 	if (flags->dryrun) {
4937 		void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
4938 
4939 		/*
4940 		 * We have read the DRR_BEGIN record, but we have
4941 		 * not yet read the payload. For non-dryrun sends
4942 		 * this will be done by the kernel, so we must
4943 		 * emulate that here, before attempting to read
4944 		 * more records.
4945 		 */
4946 		err = recv_read(hdl, infd, buf, drr->drr_payloadlen,
4947 		    flags->byteswap, NULL);
4948 		free(buf);
4949 		if (err != 0)
4950 			goto out;
4951 
4952 		err = recv_skip(hdl, infd, flags->byteswap);
4953 		goto out;
4954 	}
4955 
4956 	if (flags->heal) {
4957 		err = ioctl_err = lzc_receive_with_heal(destsnap, rcvprops,
4958 		    oxprops, wkeydata, wkeylen, origin, flags->force,
4959 		    flags->heal, flags->resumable, raw, infd, drr_noswap, -1,
4960 		    &read_bytes, &errflags, NULL, &prop_errors);
4961 	} else {
4962 		err = ioctl_err = lzc_receive_with_cmdprops(destsnap, rcvprops,
4963 		    oxprops, wkeydata, wkeylen, origin, flags->force,
4964 		    flags->resumable, raw, infd, drr_noswap, -1, &read_bytes,
4965 		    &errflags, NULL, &prop_errors);
4966 	}
4967 	ioctl_errno = ioctl_err;
4968 	prop_errflags = errflags;
4969 
4970 	if (err == 0) {
4971 		nvpair_t *prop_err = NULL;
4972 
4973 		while ((prop_err = nvlist_next_nvpair(prop_errors,
4974 		    prop_err)) != NULL) {
4975 			char tbuf[1024];
4976 			zfs_prop_t prop;
4977 			int intval;
4978 
4979 			prop = zfs_name_to_prop(nvpair_name(prop_err));
4980 			(void) nvpair_value_int32(prop_err, &intval);
4981 			if (strcmp(nvpair_name(prop_err),
4982 			    ZPROP_N_MORE_ERRORS) == 0) {
4983 				trunc_prop_errs(intval);
4984 				break;
4985 			} else if (snapname == NULL || finalsnap == NULL ||
4986 			    strcmp(finalsnap, snapname) == 0 ||
4987 			    strcmp(nvpair_name(prop_err),
4988 			    zfs_prop_to_name(ZFS_PROP_REFQUOTA)) != 0) {
4989 				/*
4990 				 * Skip the special case of, for example,
4991 				 * "refquota", errors on intermediate
4992 				 * snapshots leading up to a final one.
4993 				 * That's why we have all of the checks above.
4994 				 *
4995 				 * See zfs_ioctl.c's extract_delay_props() for
4996 				 * a list of props which can fail on
4997 				 * intermediate snapshots, but shouldn't
4998 				 * affect the overall receive.
4999 				 */
5000 				(void) snprintf(tbuf, sizeof (tbuf),
5001 				    dgettext(TEXT_DOMAIN,
5002 				    "cannot receive %s property on %s"),
5003 				    nvpair_name(prop_err), name);
5004 				zfs_setprop_error(hdl, prop, intval, tbuf);
5005 			}
5006 		}
5007 	}
5008 
5009 	if (err == 0 && snapprops_nvlist) {
5010 		zfs_cmd_t zc = {"\0"};
5011 
5012 		(void) strlcpy(zc.zc_name, destsnap, sizeof (zc.zc_name));
5013 		zc.zc_cookie = B_TRUE; /* received */
5014 		zcmd_write_src_nvlist(hdl, &zc, snapprops_nvlist);
5015 		(void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc);
5016 		zcmd_free_nvlists(&zc);
5017 	}
5018 	if (err == 0 && snapholds_nvlist) {
5019 		nvpair_t *pair;
5020 		nvlist_t *holds, *errors = NULL;
5021 		int cleanup_fd = -1;
5022 
5023 		VERIFY(0 == nvlist_alloc(&holds, 0, KM_SLEEP));
5024 		for (pair = nvlist_next_nvpair(snapholds_nvlist, NULL);
5025 		    pair != NULL;
5026 		    pair = nvlist_next_nvpair(snapholds_nvlist, pair)) {
5027 			fnvlist_add_string(holds, destsnap, nvpair_name(pair));
5028 		}
5029 		(void) lzc_hold(holds, cleanup_fd, &errors);
5030 		fnvlist_free(snapholds_nvlist);
5031 		fnvlist_free(holds);
5032 	}
5033 
5034 	if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
5035 		/*
5036 		 * It may be that this snapshot already exists,
5037 		 * in which case we want to consume & ignore it
5038 		 * rather than failing.
5039 		 */
5040 		avl_tree_t *local_avl;
5041 		nvlist_t *local_nv, *fs;
5042 		cp = strchr(destsnap, '@');
5043 
5044 		/*
5045 		 * XXX Do this faster by just iterating over snaps in
5046 		 * this fs.  Also if zc_value does not exist, we will
5047 		 * get a strange "does not exist" error message.
5048 		 */
5049 		*cp = '\0';
5050 		if (gather_nvlist(hdl, destsnap, NULL, NULL, B_FALSE, B_TRUE,
5051 		    B_FALSE, B_FALSE, B_FALSE, B_FALSE, B_FALSE, B_FALSE,
5052 		    B_TRUE, &local_nv, &local_avl) == 0) {
5053 			*cp = '@';
5054 			fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
5055 			fsavl_destroy(local_avl);
5056 			fnvlist_free(local_nv);
5057 
5058 			if (fs != NULL) {
5059 				if (flags->verbose) {
5060 					(void) printf("snap %s already exists; "
5061 					    "ignoring\n", destsnap);
5062 				}
5063 				err = ioctl_err = recv_skip(hdl, infd,
5064 				    flags->byteswap);
5065 			}
5066 		}
5067 		*cp = '@';
5068 	}
5069 
5070 	if (ioctl_err != 0) {
5071 		switch (ioctl_errno) {
5072 		case ENODEV:
5073 			cp = strchr(destsnap, '@');
5074 			*cp = '\0';
5075 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5076 			    "most recent snapshot of %s does not\n"
5077 			    "match incremental source"), destsnap);
5078 			(void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
5079 			*cp = '@';
5080 			break;
5081 		case ETXTBSY:
5082 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5083 			    "destination %s has been modified\n"
5084 			    "since most recent snapshot"), name);
5085 			(void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
5086 			break;
5087 		case EACCES:
5088 			if (flags->heal) {
5089 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5090 				    "key must be loaded to do a non-raw "
5091 				    "corrective recv on an encrypted "
5092 				    "dataset."));
5093 			} else if (raw && stream_wantsnewfs) {
5094 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5095 				    "failed to create encryption key"));
5096 			} else if (raw && !stream_wantsnewfs) {
5097 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5098 				    "encryption key does not match "
5099 				    "existing key"));
5100 			} else {
5101 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5102 				    "inherited key must be loaded"));
5103 			}
5104 			(void) zfs_error(hdl, EZFS_CRYPTOFAILED, errbuf);
5105 			break;
5106 		case EEXIST:
5107 			cp = strchr(destsnap, '@');
5108 			if (newfs) {
5109 				/* it's the containing fs that exists */
5110 				*cp = '\0';
5111 			}
5112 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5113 			    "destination already exists"));
5114 			(void) zfs_error_fmt(hdl, EZFS_EXISTS,
5115 			    dgettext(TEXT_DOMAIN, "cannot restore to %s"),
5116 			    destsnap);
5117 			*cp = '@';
5118 			break;
5119 		case EINVAL:
5120 			if (embedded && !raw) {
5121 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5122 				    "incompatible embedded data stream "
5123 				    "feature with encrypted receive."));
5124 			} else if (flags->resumable) {
5125 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5126 				    "kernel modules must be upgraded to "
5127 				    "receive this stream."));
5128 			}
5129 			(void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
5130 			break;
5131 		case ECKSUM:
5132 		case ZFS_ERR_STREAM_TRUNCATED:
5133 			if (flags->heal)
5134 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5135 				    "corrective receive was not able to "
5136 				    "reconstruct the data needed for "
5137 				    "healing."));
5138 			else
5139 				recv_ecksum_set_aux(hdl, destsnap,
5140 				    flags->resumable, ioctl_err == ECKSUM);
5141 			(void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
5142 			break;
5143 		case ZFS_ERR_STREAM_LARGE_BLOCK_MISMATCH:
5144 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5145 			    "incremental send stream requires -L "
5146 			    "(--large-block), to match previous receive."));
5147 			(void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
5148 			break;
5149 		case ENOTSUP:
5150 			if (flags->heal)
5151 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5152 				    "stream is not compatible with the "
5153 				    "data in the pool."));
5154 			else
5155 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5156 				    "pool must be upgraded to receive this "
5157 				    "stream."));
5158 			(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
5159 			break;
5160 		case EDQUOT:
5161 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5162 			    "destination %s space quota exceeded."), name);
5163 			(void) zfs_error(hdl, EZFS_NOSPC, errbuf);
5164 			break;
5165 		case ZFS_ERR_FROM_IVSET_GUID_MISSING:
5166 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5167 			    "IV set guid missing. See errata %u at "
5168 			    "https://openzfs.github.io/openzfs-docs/msg/"
5169 			    "ZFS-8000-ER."),
5170 			    ZPOOL_ERRATA_ZOL_8308_ENCRYPTION);
5171 			(void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
5172 			break;
5173 		case ZFS_ERR_FROM_IVSET_GUID_MISMATCH:
5174 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5175 			    "IV set guid mismatch. See the 'zfs receive' "
5176 			    "man page section\n discussing the limitations "
5177 			    "of raw encrypted send streams."));
5178 			(void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
5179 			break;
5180 		case ZFS_ERR_SPILL_BLOCK_FLAG_MISSING:
5181 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5182 			    "Spill block flag missing for raw send.\n"
5183 			    "The zfs software on the sending system must "
5184 			    "be updated."));
5185 			(void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
5186 			break;
5187 		case ZFS_ERR_RESUME_EXISTS:
5188 			cp = strchr(destsnap, '@');
5189 			if (newfs) {
5190 				/* it's the containing fs that exists */
5191 				*cp = '\0';
5192 			}
5193 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5194 			    "Resuming recv on existing dataset without force"));
5195 			(void) zfs_error_fmt(hdl, EZFS_RESUME_EXISTS,
5196 			    dgettext(TEXT_DOMAIN, "cannot resume recv %s"),
5197 			    destsnap);
5198 			*cp = '@';
5199 			break;
5200 		case E2BIG:
5201 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5202 			    "zfs receive required kernel memory allocation "
5203 			    "larger than the system can support. Please file "
5204 			    "an issue at the OpenZFS issue tracker:\n"
5205 			    "https://github.com/openzfs/zfs/issues/new"));
5206 			(void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
5207 			break;
5208 		case EBUSY:
5209 			if (hastoken) {
5210 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5211 				    "destination %s contains "
5212 				    "partially-complete state from "
5213 				    "\"zfs receive -s\"."), name);
5214 				(void) zfs_error(hdl, EZFS_BUSY, errbuf);
5215 				break;
5216 			}
5217 			zfs_fallthrough;
5218 		default:
5219 			(void) zfs_standard_error(hdl, ioctl_errno, errbuf);
5220 		}
5221 	}
5222 
5223 	/*
5224 	 * Mount the target filesystem (if created).  Also mount any
5225 	 * children of the target filesystem if we did a replication
5226 	 * receive (indicated by stream_avl being non-NULL).
5227 	 */
5228 	if (clp) {
5229 		if (!flags->nomount)
5230 			err |= changelist_postfix(clp);
5231 		changelist_free(clp);
5232 	}
5233 
5234 	if ((newfs || stream_avl) && type == ZFS_TYPE_FILESYSTEM && !redacted)
5235 		flags->domount = B_TRUE;
5236 
5237 	if (prop_errflags & ZPROP_ERR_NOCLEAR) {
5238 		(void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
5239 		    "failed to clear unreceived properties on %s"), name);
5240 		(void) fprintf(stderr, "\n");
5241 	}
5242 	if (prop_errflags & ZPROP_ERR_NORESTORE) {
5243 		(void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
5244 		    "failed to restore original properties on %s"), name);
5245 		(void) fprintf(stderr, "\n");
5246 	}
5247 
5248 	if (err || ioctl_err) {
5249 		err = -1;
5250 		goto out;
5251 	}
5252 
5253 	if (flags->verbose) {
5254 		char buf1[64];
5255 		char buf2[64];
5256 		uint64_t bytes = read_bytes;
5257 		struct timespec delta;
5258 		clock_gettime(CLOCK_MONOTONIC_RAW, &delta);
5259 		if (begin_time.tv_nsec > delta.tv_nsec) {
5260 			delta.tv_nsec =
5261 			    1000000000 + delta.tv_nsec - begin_time.tv_nsec;
5262 			delta.tv_sec -= 1;
5263 		} else
5264 			delta.tv_nsec -= begin_time.tv_nsec;
5265 		delta.tv_sec -= begin_time.tv_sec;
5266 		if (delta.tv_sec == 0 && delta.tv_nsec == 0)
5267 			delta.tv_nsec = 1;
5268 		double delta_f = delta.tv_sec + (delta.tv_nsec / 1e9);
5269 		zfs_nicebytes(bytes, buf1, sizeof (buf1));
5270 		zfs_nicebytes(bytes / delta_f, buf2, sizeof (buf2));
5271 
5272 		(void) printf("received %s stream in %.2f seconds (%s/sec)\n",
5273 		    buf1, delta_f, buf2);
5274 	}
5275 
5276 	err = 0;
5277 out:
5278 	if (prop_errors != NULL)
5279 		fnvlist_free(prop_errors);
5280 
5281 	if (tmp_keylocation[0] != '\0') {
5282 		fnvlist_add_string(rcvprops,
5283 		    zfs_prop_to_name(ZFS_PROP_KEYLOCATION), tmp_keylocation);
5284 	}
5285 
5286 	if (newprops)
5287 		fnvlist_free(rcvprops);
5288 
5289 	fnvlist_free(oxprops);
5290 	fnvlist_free(origprops);
5291 
5292 	return (err);
5293 }
5294 
5295 /*
5296  * Check properties we were asked to override (both -o|-x)
5297  */
5298 static boolean_t
5299 zfs_receive_checkprops(libzfs_handle_t *hdl, nvlist_t *props,
5300     const char *errbuf)
5301 {
5302 	nvpair_t *nvp = NULL;
5303 	zfs_prop_t prop;
5304 	const char *name;
5305 
5306 	while ((nvp = nvlist_next_nvpair(props, nvp)) != NULL) {
5307 		name = nvpair_name(nvp);
5308 		prop = zfs_name_to_prop(name);
5309 
5310 		if (prop == ZPROP_USERPROP) {
5311 			if (!zfs_prop_user(name)) {
5312 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5313 				    "%s: invalid property '%s'"), errbuf, name);
5314 				return (B_FALSE);
5315 			}
5316 			continue;
5317 		}
5318 		/*
5319 		 * "origin" is readonly but is used to receive datasets as
5320 		 * clones so we don't raise an error here
5321 		 */
5322 		if (prop == ZFS_PROP_ORIGIN)
5323 			continue;
5324 
5325 		/* encryption params have their own verification later */
5326 		if (prop == ZFS_PROP_ENCRYPTION ||
5327 		    zfs_prop_encryption_key_param(prop))
5328 			continue;
5329 
5330 		/*
5331 		 * cannot override readonly, set-once and other specific
5332 		 * settable properties
5333 		 */
5334 		if (zfs_prop_readonly(prop) || prop == ZFS_PROP_VERSION ||
5335 		    prop == ZFS_PROP_VOLSIZE) {
5336 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5337 			    "%s: invalid property '%s'"), errbuf, name);
5338 			return (B_FALSE);
5339 		}
5340 	}
5341 
5342 	return (B_TRUE);
5343 }
5344 
5345 static int
5346 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap,
5347     const char *originsnap, recvflags_t *flags, int infd, const char *sendfs,
5348     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs,
5349     const char *finalsnap, nvlist_t *cmdprops)
5350 {
5351 	int err;
5352 	dmu_replay_record_t drr, drr_noswap;
5353 	struct drr_begin *drrb = &drr.drr_u.drr_begin;
5354 	char errbuf[ERRBUFLEN];
5355 	zio_cksum_t zcksum = { { 0 } };
5356 	uint64_t featureflags;
5357 	int hdrtype;
5358 
5359 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
5360 	    "cannot receive"));
5361 
5362 	/* check cmdline props, raise an error if they cannot be received */
5363 	if (!zfs_receive_checkprops(hdl, cmdprops, errbuf))
5364 		return (zfs_error(hdl, EZFS_BADPROP, errbuf));
5365 
5366 	if (flags->isprefix &&
5367 	    !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
5368 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
5369 		    "(%s) does not exist"), tosnap);
5370 		return (zfs_error(hdl, EZFS_NOENT, errbuf));
5371 	}
5372 	if (originsnap &&
5373 	    !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) {
5374 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs "
5375 		    "(%s) does not exist"), originsnap);
5376 		return (zfs_error(hdl, EZFS_NOENT, errbuf));
5377 	}
5378 
5379 	/* read in the BEGIN record */
5380 	if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
5381 	    &zcksum)))
5382 		return (err);
5383 
5384 	if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
5385 		/* It's the double end record at the end of a package */
5386 		return (ENODATA);
5387 	}
5388 
5389 	/* the kernel needs the non-byteswapped begin record */
5390 	drr_noswap = drr;
5391 
5392 	flags->byteswap = B_FALSE;
5393 	if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
5394 		/*
5395 		 * We computed the checksum in the wrong byteorder in
5396 		 * recv_read() above; do it again correctly.
5397 		 */
5398 		memset(&zcksum, 0, sizeof (zio_cksum_t));
5399 		fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
5400 		flags->byteswap = B_TRUE;
5401 
5402 		drr.drr_type = BSWAP_32(drr.drr_type);
5403 		drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
5404 		drrb->drr_magic = BSWAP_64(drrb->drr_magic);
5405 		drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
5406 		drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
5407 		drrb->drr_type = BSWAP_32(drrb->drr_type);
5408 		drrb->drr_flags = BSWAP_32(drrb->drr_flags);
5409 		drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
5410 		drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
5411 	}
5412 
5413 	if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
5414 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
5415 		    "stream (bad magic number)"));
5416 		return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
5417 	}
5418 
5419 	featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
5420 	hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
5421 
5422 	if (!DMU_STREAM_SUPPORTED(featureflags) ||
5423 	    (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
5424 		/*
5425 		 * Let's be explicit about this one, since rather than
5426 		 * being a new feature we can't know, it's an old
5427 		 * feature we dropped.
5428 		 */
5429 		if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
5430 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5431 			    "stream has deprecated feature: dedup, try "
5432 			    "'zstream redup [send in a file] | zfs recv "
5433 			    "[...]'"));
5434 		} else {
5435 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
5436 			    "stream has unsupported feature, feature flags = "
5437 			    "%llx (unknown flags = %llx)"),
5438 			    (u_longlong_t)featureflags,
5439 			    (u_longlong_t)((featureflags) &
5440 			    ~DMU_BACKUP_FEATURE_MASK));
5441 		}
5442 		return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
5443 	}
5444 
5445 	/* Holds feature is set once in the compound stream header. */
5446 	if (featureflags & DMU_BACKUP_FEATURE_HOLDS)
5447 		flags->holds = B_TRUE;
5448 
5449 	if (strchr(drrb->drr_toname, '@') == NULL) {
5450 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
5451 		    "stream (bad snapshot name)"));
5452 		return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
5453 	}
5454 
5455 	if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
5456 		char nonpackage_sendfs[ZFS_MAX_DATASET_NAME_LEN];
5457 		if (sendfs == NULL) {
5458 			/*
5459 			 * We were not called from zfs_receive_package(). Get
5460 			 * the fs specified by 'zfs send'.
5461 			 */
5462 			char *cp;
5463 			(void) strlcpy(nonpackage_sendfs,
5464 			    drr.drr_u.drr_begin.drr_toname,
5465 			    sizeof (nonpackage_sendfs));
5466 			if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
5467 				*cp = '\0';
5468 			sendfs = nonpackage_sendfs;
5469 			VERIFY(finalsnap == NULL);
5470 		}
5471 		return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags,
5472 		    &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs,
5473 		    finalsnap, cmdprops));
5474 	} else {
5475 		assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
5476 		    DMU_COMPOUNDSTREAM);
5477 		return (zfs_receive_package(hdl, infd, tosnap, flags, &drr,
5478 		    &zcksum, top_zfs, cmdprops));
5479 	}
5480 }
5481 
5482 /*
5483  * Restores a backup of tosnap from the file descriptor specified by infd.
5484  * Return 0 on total success, -2 if some things couldn't be
5485  * destroyed/renamed/promoted, -1 if some things couldn't be received.
5486  * (-1 will override -2, if -1 and the resumable flag was specified the
5487  * transfer can be resumed if the sending side supports it).
5488  */
5489 int
5490 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props,
5491     recvflags_t *flags, int infd, avl_tree_t *stream_avl)
5492 {
5493 	char *top_zfs = NULL;
5494 	int err;
5495 	struct stat sb;
5496 	char *originsnap = NULL;
5497 
5498 	/*
5499 	 * The only way fstat can fail is if we do not have a valid file
5500 	 * descriptor.
5501 	 */
5502 	if (fstat(infd, &sb) == -1) {
5503 		perror("fstat");
5504 		return (-2);
5505 	}
5506 
5507 	if (props) {
5508 		err = nvlist_lookup_string(props, "origin", &originsnap);
5509 		if (err && err != ENOENT)
5510 			return (err);
5511 	}
5512 
5513 	err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL,
5514 	    stream_avl, &top_zfs, NULL, props);
5515 
5516 	if (err == 0 && !flags->nomount && flags->domount && top_zfs) {
5517 		zfs_handle_t *zhp = NULL;
5518 		prop_changelist_t *clp = NULL;
5519 
5520 		zhp = zfs_open(hdl, top_zfs,
5521 		    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
5522 		if (zhp == NULL) {
5523 			err = -1;
5524 			goto out;
5525 		} else {
5526 			if (zhp->zfs_type == ZFS_TYPE_VOLUME) {
5527 				zfs_close(zhp);
5528 				goto out;
5529 			}
5530 
5531 			clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
5532 			    CL_GATHER_MOUNT_ALWAYS,
5533 			    flags->forceunmount ? MS_FORCE : 0);
5534 			zfs_close(zhp);
5535 			if (clp == NULL) {
5536 				err = -1;
5537 				goto out;
5538 			}
5539 
5540 			/* mount and share received datasets */
5541 			err = changelist_postfix(clp);
5542 			changelist_free(clp);
5543 			if (err != 0)
5544 				err = -1;
5545 		}
5546 	}
5547 
5548 out:
5549 	if (top_zfs)
5550 		free(top_zfs);
5551 
5552 	return (err);
5553 }
5554