Home | History | Annotate | Download | only in common
      1 /*
      2  * CDDL HEADER START
      3  *
      4  * The contents of this file are subject to the terms of the
      5  * Common Development and Distribution License (the "License").
      6  * You may not use this file except in compliance with the License.
      7  *
      8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
      9  * or http://www.opensolaris.org/os/licensing.
     10  * See the License for the specific language governing permissions
     11  * and limitations under the License.
     12  *
     13  * When distributing Covered Code, include this CDDL HEADER in each
     14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
     15  * If applicable, add the following below this CDDL HEADER, with the
     16  * fields enclosed by brackets "[]" replaced with your own identifying
     17  * information: Portions Copyright [yyyy] [name of copyright owner]
     18  *
     19  * CDDL HEADER END
     20  */
     21 
     22 /*
     23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
     24  * Use is subject to license terms.
     25  */
     26 
     27 #include <assert.h>
     28 #include <ctype.h>
     29 #include <errno.h>
     30 #include <libintl.h>
     31 #include <stdio.h>
     32 #include <stdlib.h>
     33 #include <strings.h>
     34 #include <unistd.h>
     35 #include <stddef.h>
     36 #include <fcntl.h>
     37 #include <sys/mount.h>
     38 
     39 #include <libzfs.h>
     40 
     41 #include "zfs_namecheck.h"
     42 #include "zfs_prop.h"
     43 #include "zfs_fletcher.h"
     44 #include "libzfs_impl.h"
     45 
     46 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t,
     47     int, avl_tree_t *, char **);
     48 
     49 /*
     50  * Routines for dealing with the AVL tree of fs-nvlists
     51  */
     52 typedef struct fsavl_node {
     53 	avl_node_t fn_node;
     54 	nvlist_t *fn_nvfs;
     55 	char *fn_snapname;
     56 	uint64_t fn_guid;
     57 } fsavl_node_t;
     58 
     59 static int
     60 fsavl_compare(const void *arg1, const void *arg2)
     61 {
     62 	const fsavl_node_t *fn1 = arg1;
     63 	const fsavl_node_t *fn2 = arg2;
     64 
     65 	if (fn1->fn_guid > fn2->fn_guid)
     66 		return (+1);
     67 	else if (fn1->fn_guid < fn2->fn_guid)
     68 		return (-1);
     69 	else
     70 		return (0);
     71 }
     72 
     73 /*
     74  * Given the GUID of a snapshot, find its containing filesystem and
     75  * (optionally) name.
     76  */
     77 static nvlist_t *
     78 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
     79 {
     80 	fsavl_node_t fn_find;
     81 	fsavl_node_t *fn;
     82 
     83 	fn_find.fn_guid = snapguid;
     84 
     85 	fn = avl_find(avl, &fn_find, NULL);
     86 	if (fn) {
     87 		if (snapname)
     88 			*snapname = fn->fn_snapname;
     89 		return (fn->fn_nvfs);
     90 	}
     91 	return (NULL);
     92 }
     93 
     94 static void
     95 fsavl_destroy(avl_tree_t *avl)
     96 {
     97 	fsavl_node_t *fn;
     98 	void *cookie;
     99 
    100 	if (avl == NULL)
    101 		return;
    102 
    103 	cookie = NULL;
    104 	while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
    105 		free(fn);
    106 	avl_destroy(avl);
    107 	free(avl);
    108 }
    109 
    110 /*
    111  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
    112  */
    113 static avl_tree_t *
    114 fsavl_create(nvlist_t *fss)
    115 {
    116 	avl_tree_t *fsavl;
    117 	nvpair_t *fselem = NULL;
    118 
    119 	if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
    120 		return (NULL);
    121 
    122 	avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
    123 	    offsetof(fsavl_node_t, fn_node));
    124 
    125 	while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
    126 		nvlist_t *nvfs, *snaps;
    127 		nvpair_t *snapelem = NULL;
    128 
    129 		VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
    130 		VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
    131 
    132 		while ((snapelem =
    133 		    nvlist_next_nvpair(snaps, snapelem)) != NULL) {
    134 			fsavl_node_t *fn;
    135 			uint64_t guid;
    136 
    137 			VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
    138 			if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
    139 				fsavl_destroy(fsavl);
    140 				return (NULL);
    141 			}
    142 			fn->fn_nvfs = nvfs;
    143 			fn->fn_snapname = nvpair_name(snapelem);
    144 			fn->fn_guid = guid;
    145 
    146 			/*
    147 			 * Note: if there are multiple snaps with the
    148 			 * same GUID, we ignore all but one.
    149 			 */
    150 			if (avl_find(fsavl, fn, NULL) == NULL)
    151 				avl_add(fsavl, fn);
    152 			else
    153 				free(fn);
    154 		}
    155 	}
    156 
    157 	return (fsavl);
    158 }
    159 
    160 /*
    161  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
    162  */
    163 typedef struct send_data {
    164 	uint64_t parent_fromsnap_guid;
    165 	nvlist_t *parent_snaps;
    166 	nvlist_t *fss;
    167 	nvlist_t *snapprops;
    168 	const char *fromsnap;
    169 	const char *tosnap;
    170 
    171 	/*
    172 	 * The header nvlist is of the following format:
    173 	 * {
    174 	 *   "tosnap" -> string
    175 	 *   "fromsnap" -> string (if incremental)
    176 	 *   "fss" -> {
    177 	 *	id -> {
    178 	 *
    179 	 *	 "name" -> string (full name; for debugging)
    180 	 *	 "parentfromsnap" -> number (guid of fromsnap in parent)
    181 	 *
    182 	 *	 "props" -> { name -> value (only if set here) }
    183 	 *	 "snaps" -> { name (lastname) -> number (guid) }
    184 	 *	 "snapprops" -> { name (lastname) -> { name -> value } }
    185 	 *
    186 	 *	 "origin" -> number (guid) (if clone)
    187 	 *	 "sent" -> boolean (not on-disk)
    188 	 *	}
    189 	 *   }
    190 	 * }
    191 	 *
    192 	 */
    193 } send_data_t;
    194 
    195 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
    196 
    197 static int
    198 send_iterate_snap(zfs_handle_t *zhp, void *arg)
    199 {
    200 	send_data_t *sd = arg;
    201 	uint64_t guid = zhp->zfs_dmustats.dds_guid;
    202 	char *snapname;
    203 	nvlist_t *nv;
    204 
    205 	snapname = strrchr(zhp->zfs_name, '@')+1;
    206 
    207 	VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
    208 	/*
    209 	 * NB: if there is no fromsnap here (it's a newly created fs in
    210 	 * an incremental replication), we will substitute the tosnap.
    211 	 */
    212 	if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
    213 	    (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
    214 	    strcmp(snapname, sd->tosnap) == 0)) {
    215 		sd->parent_fromsnap_guid = guid;
    216 	}
    217 
    218 	VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
    219 	send_iterate_prop(zhp, nv);
    220 	VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
    221 	nvlist_free(nv);
    222 
    223 	zfs_close(zhp);
    224 	return (0);
    225 }
    226 
    227 static void
    228 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
    229 {
    230 	nvpair_t *elem = NULL;
    231 
    232 	while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
    233 		char *propname = nvpair_name(elem);
    234 		zfs_prop_t prop = zfs_name_to_prop(propname);
    235 		nvlist_t *propnv;
    236 
    237 		if (!zfs_prop_user(propname)) {
    238 			/*
    239 			 * Realistically, this should never happen.  However,
    240 			 * we want the ability to add DSL properties without
    241 			 * needing to make incompatible version changes.  We
    242 			 * need to ignore unknown properties to allow older
    243 			 * software to still send datasets containing these
    244 			 * properties, with the unknown properties elided.
    245 			 */
    246 			if (prop == ZPROP_INVAL)
    247 				continue;
    248 
    249 			if (zfs_prop_readonly(prop))
    250 				continue;
    251 		}
    252 
    253 		verify(nvpair_value_nvlist(elem, &propnv) == 0);
    254 		if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
    255 		    prop == ZFS_PROP_REFQUOTA ||
    256 		    prop == ZFS_PROP_REFRESERVATION) {
    257 			/* these guys are modifyable, but have no source */
    258 			uint64_t value;
    259 			verify(nvlist_lookup_uint64(propnv,
    260 			    ZPROP_VALUE, &value) == 0);
    261 			if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
    262 				continue;
    263 		} else {
    264 			char *source;
    265 			if (nvlist_lookup_string(propnv,
    266 			    ZPROP_SOURCE, &source) != 0)
    267 				continue;
    268 			if (strcmp(source, zhp->zfs_name) != 0)
    269 				continue;
    270 		}
    271 
    272 		if (zfs_prop_user(propname) ||
    273 		    zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
    274 			char *value;
    275 			verify(nvlist_lookup_string(propnv,
    276 			    ZPROP_VALUE, &value) == 0);
    277 			VERIFY(0 == nvlist_add_string(nv, propname, value));
    278 		} else {
    279 			uint64_t value;
    280 			verify(nvlist_lookup_uint64(propnv,
    281 			    ZPROP_VALUE, &value) == 0);
    282 			VERIFY(0 == nvlist_add_uint64(nv, propname, value));
    283 		}
    284 	}
    285 }
    286 
    287 /*
    288  * recursively generate nvlists describing datasets.  See comment
    289  * for the data structure send_data_t above for description of contents
    290  * of the nvlist.
    291  */
    292 static int
    293 send_iterate_fs(zfs_handle_t *zhp, void *arg)
    294 {
    295 	send_data_t *sd = arg;
    296 	nvlist_t *nvfs, *nv;
    297 	int rv;
    298 	uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
    299 	uint64_t guid = zhp->zfs_dmustats.dds_guid;
    300 	char guidstring[64];
    301 
    302 	VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
    303 	VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
    304 	VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
    305 	    sd->parent_fromsnap_guid));
    306 
    307 	if (zhp->zfs_dmustats.dds_origin[0]) {
    308 		zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
    309 		    zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
    310 		if (origin == NULL)
    311 			return (-1);
    312 		VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
    313 		    origin->zfs_dmustats.dds_guid));
    314 	}
    315 
    316 	/* iterate over props */
    317 	VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
    318 	send_iterate_prop(zhp, nv);
    319 	VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
    320 	nvlist_free(nv);
    321 
    322 	/* iterate over snaps, and set sd->parent_fromsnap_guid */
    323 	sd->parent_fromsnap_guid = 0;
    324 	VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
    325 	VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
    326 	(void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
    327 	VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
    328 	VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
    329 	nvlist_free(sd->parent_snaps);
    330 	nvlist_free(sd->snapprops);
    331 
    332 	/* add this fs to nvlist */
    333 	(void) snprintf(guidstring, sizeof (guidstring),
    334 	    "0x%llx", (longlong_t)guid);
    335 	VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
    336 	nvlist_free(nvfs);
    337 
    338 	/* iterate over children */
    339 	rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
    340 
    341 	sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
    342 
    343 	zfs_close(zhp);
    344 	return (rv);
    345 }
    346 
    347 static int
    348 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
    349     const char *tosnap, nvlist_t **nvlp, avl_tree_t **avlp)
    350 {
    351 	zfs_handle_t *zhp;
    352 	send_data_t sd = { 0 };
    353 	int error;
    354 
    355 	zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
    356 	if (zhp == NULL)
    357 		return (EZFS_BADTYPE);
    358 
    359 	VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
    360 	sd.fromsnap = fromsnap;
    361 	sd.tosnap = tosnap;
    362 
    363 	if ((error = send_iterate_fs(zhp, &sd)) != 0) {
    364 		nvlist_free(sd.fss);
    365 		if (avlp != NULL)
    366 			*avlp = NULL;
    367 		*nvlp = NULL;
    368 		return (error);
    369 	}
    370 
    371 	if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
    372 		nvlist_free(sd.fss);
    373 		*nvlp = NULL;
    374 		return (EZFS_NOMEM);
    375 	}
    376 
    377 	*nvlp = sd.fss;
    378 	return (0);
    379 }
    380 
    381 /*
    382  * Routines for dealing with the sorted snapshot functionality
    383  */
    384 typedef struct zfs_node {
    385 	zfs_handle_t	*zn_handle;
    386 	avl_node_t	zn_avlnode;
    387 } zfs_node_t;
    388 
    389 static int
    390 zfs_sort_snaps(zfs_handle_t *zhp, void *data)
    391 {
    392 	avl_tree_t *avl = data;
    393 	zfs_node_t *node = zfs_alloc(zhp->zfs_hdl, sizeof (zfs_node_t));
    394 
    395 	node->zn_handle = zhp;
    396 	avl_add(avl, node);
    397 	return (0);
    398 }
    399 
    400 /* ARGSUSED */
    401 static int
    402 zfs_snapshot_compare(const void *larg, const void *rarg)
    403 {
    404 	zfs_handle_t *l = ((zfs_node_t *)larg)->zn_handle;
    405 	zfs_handle_t *r = ((zfs_node_t *)rarg)->zn_handle;
    406 	uint64_t lcreate, rcreate;
    407 
    408 	/*
    409 	 * Sort them according to creation time.  We use the hidden
    410 	 * CREATETXG property to get an absolute ordering of snapshots.
    411 	 */
    412 	lcreate = zfs_prop_get_int(l, ZFS_PROP_CREATETXG);
    413 	rcreate = zfs_prop_get_int(r, ZFS_PROP_CREATETXG);
    414 
    415 	if (lcreate < rcreate)
    416 		return (-1);
    417 	else if (lcreate > rcreate)
    418 		return (+1);
    419 	else
    420 		return (0);
    421 }
    422 
    423 int
    424 zfs_iter_snapshots_sorted(zfs_handle_t *zhp, zfs_iter_f callback, void *data)
    425 {
    426 	int ret = 0;
    427 	zfs_node_t *node;
    428 	avl_tree_t avl;
    429 	void *cookie = NULL;
    430 
    431 	avl_create(&avl, zfs_snapshot_compare,
    432 	    sizeof (zfs_node_t), offsetof(zfs_node_t, zn_avlnode));
    433 
    434 	ret = zfs_iter_snapshots(zhp, zfs_sort_snaps, &avl);
    435 
    436 	for (node = avl_first(&avl); node != NULL; node = AVL_NEXT(&avl, node))
    437 		ret |= callback(node->zn_handle, data);
    438 
    439 	while ((node = avl_destroy_nodes(&avl, &cookie)) != NULL)
    440 		free(node);
    441 
    442 	avl_destroy(&avl);
    443 
    444 	return (ret);
    445 }
    446 
    447 /*
    448  * Routines specific to "zfs send"
    449  */
    450 typedef struct send_dump_data {
    451 	/* these are all just the short snapname (the part after the @) */
    452 	const char *fromsnap;
    453 	const char *tosnap;
    454 	char lastsnap[ZFS_MAXNAMELEN];
    455 	boolean_t seenfrom, seento, replicate, doall, fromorigin;
    456 	boolean_t verbose;
    457 	int outfd;
    458 	boolean_t err;
    459 	nvlist_t *fss;
    460 	avl_tree_t *fsavl;
    461 } send_dump_data_t;
    462 
    463 /*
    464  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
    465  * NULL) to the file descriptor specified by outfd.
    466  */
    467 static int
    468 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, boolean_t fromorigin,
    469     int outfd)
    470 {
    471 	zfs_cmd_t zc = { 0 };
    472 	libzfs_handle_t *hdl = zhp->zfs_hdl;
    473 
    474 	assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
    475 	assert(fromsnap == NULL || fromsnap[0] == '\0' || !fromorigin);
    476 
    477 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
    478 	if (fromsnap)
    479 		(void) strlcpy(zc.zc_value, fromsnap, sizeof (zc.zc_value));
    480 	zc.zc_cookie = outfd;
    481 	zc.zc_obj = fromorigin;
    482 
    483 	if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_SEND, &zc) != 0) {
    484 		char errbuf[1024];
    485 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
    486 		    "warning: cannot send '%s'"), zhp->zfs_name);
    487 
    488 		switch (errno) {
    489 
    490 		case EXDEV:
    491 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
    492 			    "not an earlier snapshot from the same fs"));
    493 			return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
    494 
    495 		case ENOENT:
    496 			if (zfs_dataset_exists(hdl, zc.zc_name,
    497 			    ZFS_TYPE_SNAPSHOT)) {
    498 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
    499 				    "incremental source (@%s) does not exist"),
    500 				    zc.zc_value);
    501 			}
    502 			return (zfs_error(hdl, EZFS_NOENT, errbuf));
    503 
    504 		case EDQUOT:
    505 		case EFBIG:
    506 		case EIO:
    507 		case ENOLINK:
    508 		case ENOSPC:
    509 		case ENOSTR:
    510 		case ENXIO:
    511 		case EPIPE:
    512 		case ERANGE:
    513 		case EFAULT:
    514 		case EROFS:
    515 			zfs_error_aux(hdl, strerror(errno));
    516 			return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
    517 
    518 		default:
    519 			return (zfs_standard_error(hdl, errno, errbuf));
    520 		}
    521 	}
    522 
    523 	return (0);
    524 }
    525 
    526 static int
    527 dump_snapshot(zfs_handle_t *zhp, void *arg)
    528 {
    529 	send_dump_data_t *sdd = arg;
    530 	const char *thissnap;
    531 	int err;
    532 
    533 	thissnap = strchr(zhp->zfs_name, '@') + 1;
    534 
    535 	if (sdd->fromsnap && !sdd->seenfrom &&
    536 	    strcmp(sdd->fromsnap, thissnap) == 0) {
    537 		sdd->seenfrom = B_TRUE;
    538 		(void) strcpy(sdd->lastsnap, thissnap);
    539 		zfs_close(zhp);
    540 		return (0);
    541 	}
    542 
    543 	if (sdd->seento || !sdd->seenfrom) {
    544 		zfs_close(zhp);
    545 		return (0);
    546 	}
    547 
    548 	/* send it */
    549 	if (sdd->verbose) {
    550 		(void) fprintf(stderr, "sending from @%s to %s\n",
    551 		    sdd->lastsnap, zhp->zfs_name);
    552 	}
    553 
    554 	err = dump_ioctl(zhp, sdd->lastsnap,
    555 	    sdd->lastsnap[0] == '\0' && (sdd->fromorigin || sdd->replicate),
    556 	    sdd->outfd);
    557 
    558 	if (!sdd->seento && strcmp(sdd->tosnap, thissnap) == 0)
    559 		sdd->seento = B_TRUE;
    560 
    561 	(void) strcpy(sdd->lastsnap, thissnap);
    562 	zfs_close(zhp);
    563 	return (err);
    564 }
    565 
    566 static int
    567 dump_filesystem(zfs_handle_t *zhp, void *arg)
    568 {
    569 	int rv = 0;
    570 	send_dump_data_t *sdd = arg;
    571 	boolean_t missingfrom = B_FALSE;
    572 	zfs_cmd_t zc = { 0 };
    573 
    574 	(void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
    575 	    zhp->zfs_name, sdd->tosnap);
    576 	if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
    577 		(void) fprintf(stderr, "WARNING: "
    578 		    "could not send %s@%s: does not exist\n",
    579 		    zhp->zfs_name, sdd->tosnap);
    580 		sdd->err = B_TRUE;
    581 		return (0);
    582 	}
    583 
    584 	if (sdd->replicate && sdd->fromsnap) {
    585 		/*
    586 		 * If this fs does not have fromsnap, and we're doing
    587 		 * recursive, we need to send a full stream from the
    588 		 * beginning (or an incremental from the origin if this
    589 		 * is a clone).  If we're doing non-recursive, then let
    590 		 * them get the error.
    591 		 */
    592 		(void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
    593 		    zhp->zfs_name, sdd->fromsnap);
    594 		if (ioctl(zhp->zfs_hdl->libzfs_fd,
    595 		    ZFS_IOC_OBJSET_STATS, &zc) != 0) {
    596 			missingfrom = B_TRUE;
    597 		}
    598 	}
    599 
    600 	if (sdd->doall) {
    601 		sdd->seenfrom = sdd->seento = sdd->lastsnap[0] = 0;
    602 		if (sdd->fromsnap == NULL || missingfrom)
    603 			sdd->seenfrom = B_TRUE;
    604 
    605 		rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
    606 		if (!sdd->seenfrom) {
    607 			(void) fprintf(stderr,
    608 			    "WARNING: could not send %s@%s:\n"
    609 			    "incremental source (%s@%s) does not exist\n",
    610 			    zhp->zfs_name, sdd->tosnap,
    611 			    zhp->zfs_name, sdd->fromsnap);
    612 			sdd->err = B_TRUE;
    613 		} else if (!sdd->seento) {
    614 			if (sdd->fromsnap) {
    615 				(void) fprintf(stderr,
    616 				    "WARNING: could not send %s@%s:\n"
    617 				    "incremental source (%s@%s) "
    618 				    "is not earlier than it\n",
    619 				    zhp->zfs_name, sdd->tosnap,
    620 				    zhp->zfs_name, sdd->fromsnap);
    621 			} else {
    622 				(void) fprintf(stderr, "WARNING: "
    623 				    "could not send %s@%s: does not exist\n",
    624 				    zhp->zfs_name, sdd->tosnap);
    625 			}
    626 			sdd->err = B_TRUE;
    627 		}
    628 	} else {
    629 		zfs_handle_t *snapzhp;
    630 		char snapname[ZFS_MAXNAMELEN];
    631 
    632 		(void) snprintf(snapname, sizeof (snapname), "%s@%s",
    633 		    zfs_get_name(zhp), sdd->tosnap);
    634 		snapzhp = zfs_open(zhp->zfs_hdl, snapname, ZFS_TYPE_SNAPSHOT);
    635 		if (snapzhp == NULL) {
    636 			rv = -1;
    637 		} else {
    638 			rv = dump_ioctl(snapzhp,
    639 			    missingfrom ? NULL : sdd->fromsnap,
    640 			    sdd->fromorigin || missingfrom,
    641 			    sdd->outfd);
    642 			sdd->seento = B_TRUE;
    643 			zfs_close(snapzhp);
    644 		}
    645 	}
    646 
    647 	return (rv);
    648 }
    649 
    650 static int
    651 dump_filesystems(zfs_handle_t *rzhp, void *arg)
    652 {
    653 	send_dump_data_t *sdd = arg;
    654 	nvpair_t *fspair;
    655 	boolean_t needagain, progress;
    656 
    657 	if (!sdd->replicate)
    658 		return (dump_filesystem(rzhp, sdd));
    659 
    660 again:
    661 	needagain = progress = B_FALSE;
    662 	for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
    663 	    fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
    664 		nvlist_t *fslist;
    665 		char *fsname;
    666 		zfs_handle_t *zhp;
    667 		int err;
    668 		uint64_t origin_guid = 0;
    669 		nvlist_t *origin_nv;
    670 
    671 		VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
    672 		if (nvlist_lookup_boolean(fslist, "sent") == 0)
    673 			continue;
    674 
    675 		VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
    676 		(void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
    677 
    678 		origin_nv = fsavl_find(sdd->fsavl, origin_guid, NULL);
    679 		if (origin_nv &&
    680 		    nvlist_lookup_boolean(origin_nv, "sent") == ENOENT) {
    681 			/*
    682 			 * origin has not been sent yet;
    683 			 * skip this clone.
    684 			 */
    685 			needagain = B_TRUE;
    686 			continue;
    687 		}
    688 
    689 		zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
    690 		if (zhp == NULL)
    691 			return (-1);
    692 		err = dump_filesystem(zhp, sdd);
    693 		VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
    694 		progress = B_TRUE;
    695 		zfs_close(zhp);
    696 		if (err)
    697 			return (err);
    698 	}
    699 	if (needagain) {
    700 		assert(progress);
    701 		goto again;
    702 	}
    703 	return (0);
    704 }
    705 
    706 /*
    707  * Generate a send stream for the dataset identified by the argument zhp.
    708  *
    709  * The content of the send stream is the snapshot identified by
    710  * 'tosnap'.  Incremental streams are requested in two ways:
    711  *     - from the snapshot identified by "fromsnap" (if non-null) or
    712  *     - from the origin of the dataset identified by zhp, which must
    713  *	 be a clone.  In this case, "fromsnap" is null and "fromorigin"
    714  *	 is TRUE.
    715  *
    716  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
    717  * uses a special header (with a version field of DMU_BACKUP_HEADER_VERSION)
    718  * if "replicate" is set.  If "doall" is set, dump all the intermediate
    719  * snapshots. The DMU_BACKUP_HEADER_VERSION header is used in the "doall"
    720  * case too.
    721  */
    722 int
    723 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
    724     boolean_t replicate, boolean_t doall, boolean_t fromorigin,
    725     boolean_t verbose, int outfd)
    726 {
    727 	char errbuf[1024];
    728 	send_dump_data_t sdd = { 0 };
    729 	int err;
    730 	nvlist_t *fss = NULL;
    731 	avl_tree_t *fsavl = NULL;
    732 	char holdtag[128];
    733 	static uint64_t holdseq;
    734 	int spa_version;
    735 	boolean_t holdsnaps = B_FALSE;
    736 
    737 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
    738 	    "cannot send '%s'"), zhp->zfs_name);
    739 
    740 	if (fromsnap && fromsnap[0] == '\0') {
    741 		zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
    742 		    "zero-length incremental source"));
    743 		return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
    744 	}
    745 
    746 	if (zfs_spa_version(zhp, &spa_version) == 0 &&
    747 	    spa_version >= SPA_VERSION_USERREFS)
    748 		holdsnaps = B_TRUE;
    749 
    750 	if (replicate || doall) {
    751 		dmu_replay_record_t drr = { 0 };
    752 		char *packbuf = NULL;
    753 		size_t buflen = 0;
    754 		zio_cksum_t zc = { 0 };
    755 
    756 		assert(fromsnap || doall);
    757 
    758 		if (holdsnaps) {
    759 			(void) snprintf(holdtag, sizeof (holdtag),
    760 			    ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
    761 			++holdseq;
    762 			err = zfs_hold_range(zhp, fromsnap, tosnap,
    763 			    holdtag, B_TRUE);
    764 			if (err)
    765 				return (err);
    766 		}
    767 		if (replicate) {
    768 			nvlist_t *hdrnv;
    769 
    770 			VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
    771 			if (fromsnap) {
    772 				VERIFY(0 == nvlist_add_string(hdrnv,
    773 				    "fromsnap", fromsnap));
    774 			}
    775 			VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
    776 
    777 			err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
    778 			    fromsnap, tosnap, &fss, &fsavl);
    779 			if (err) {
    780 				if (holdsnaps) {
    781 					(void) zfs_release_range(zhp, fromsnap,
    782 					    tosnap, holdtag);
    783 				}
    784 				return (err);
    785 			}
    786 			VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
    787 			err = nvlist_pack(hdrnv, &packbuf, &buflen,
    788 			    NV_ENCODE_XDR, 0);
    789 			nvlist_free(hdrnv);
    790 			if (err) {
    791 				fsavl_destroy(fsavl);
    792 				nvlist_free(fss);
    793 				if (holdsnaps) {
    794 					(void) zfs_release_range(zhp, fromsnap,
    795 					    tosnap, holdtag);
    796 				}
    797 				return (zfs_standard_error(zhp->zfs_hdl,
    798 				    err, errbuf));
    799 			}
    800 		}
    801 
    802 		/* write first begin record */
    803 		drr.drr_type = DRR_BEGIN;
    804 		drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
    805 		drr.drr_u.drr_begin.drr_version = DMU_BACKUP_HEADER_VERSION;
    806 		(void) snprintf(drr.drr_u.drr_begin.drr_toname,
    807 		    sizeof (drr.drr_u.drr_begin.drr_toname),
    808 		    "%s@%s", zhp->zfs_name, tosnap);
    809 		drr.drr_payloadlen = buflen;
    810 		fletcher_4_incremental_native(&drr, sizeof (drr), &zc);
    811 		err = write(outfd, &drr, sizeof (drr));
    812 
    813 		/* write header nvlist */
    814 		if (err != -1) {
    815 			fletcher_4_incremental_native(packbuf, buflen, &zc);
    816 			err = write(outfd, packbuf, buflen);
    817 		}
    818 		free(packbuf);
    819 		if (err == -1) {
    820 			fsavl_destroy(fsavl);
    821 			nvlist_free(fss);
    822 			if (holdsnaps) {
    823 				(void) zfs_release_range(zhp, fromsnap, tosnap,
    824 				    holdtag);
    825 			}
    826 			return (zfs_standard_error(zhp->zfs_hdl,
    827 			    errno, errbuf));
    828 		}
    829 
    830 		/* write end record */
    831 		if (err != -1) {
    832 			bzero(&drr, sizeof (drr));
    833 			drr.drr_type = DRR_END;
    834 			drr.drr_u.drr_end.drr_checksum = zc;
    835 			err = write(outfd, &drr, sizeof (drr));
    836 			if (err == -1) {
    837 				fsavl_destroy(fsavl);
    838 				nvlist_free(fss);
    839 				if (holdsnaps) {
    840 					(void) zfs_release_range(zhp, fromsnap,
    841 					    tosnap, holdtag);
    842 				}
    843 				return (zfs_standard_error(zhp->zfs_hdl,
    844 				    errno, errbuf));
    845 			}
    846 		}
    847 	}
    848 
    849 	/* dump each stream */
    850 	sdd.fromsnap = fromsnap;
    851 	sdd.tosnap = tosnap;
    852 	sdd.outfd = outfd;
    853 	sdd.replicate = replicate;
    854 	sdd.doall = doall;
    855 	sdd.fromorigin = fromorigin;
    856 	sdd.fss = fss;
    857 	sdd.fsavl = fsavl;
    858 	sdd.verbose = verbose;
    859 	err = dump_filesystems(zhp, &sdd);
    860 	fsavl_destroy(fsavl);
    861 	nvlist_free(fss);
    862 
    863 	if (replicate || doall) {
    864 		/*
    865 		 * write final end record.  NB: want to do this even if
    866 		 * there was some error, because it might not be totally
    867 		 * failed.
    868 		 */
    869 		dmu_replay_record_t drr = { 0 };
    870 		drr.drr_type = DRR_END;
    871 		if (holdsnaps) {
    872 			(void) zfs_release_range(zhp, fromsnap, tosnap,
    873 			    holdtag);
    874 		}
    875 		if (write(outfd, &drr, sizeof (drr)) == -1) {
    876 			return (zfs_standard_error(zhp->zfs_hdl,
    877 			    errno, errbuf));
    878 		}
    879 	}
    880 
    881 	return (err || sdd.err);
    882 }
    883 
    884 /*
    885  * Routines specific to "zfs recv"
    886  */
    887 
    888 static int
    889 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
    890     boolean_t byteswap, zio_cksum_t *zc)
    891 {
    892 	char *cp = buf;
    893 	int rv;
    894 	int len = ilen;
    895 
    896 	do {
    897 		rv = read(fd, cp, len);
    898 		cp += rv;
    899 		len -= rv;
    900 	} while (rv > 0);
    901 
    902 	if (rv < 0 || len != 0) {
    903 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
    904 		    "failed to read from stream"));
    905 		return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
    906 		    "cannot receive")));
    907 	}
    908 
    909 	if (zc) {
    910 		if (byteswap)
    911 			fletcher_4_incremental_byteswap(buf, ilen, zc);
    912 		else
    913 			fletcher_4_incremental_native(buf, ilen, zc);
    914 	}
    915 	return (0);
    916 }
    917 
    918 static int
    919 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
    920     boolean_t byteswap, zio_cksum_t *zc)
    921 {
    922 	char *buf;
    923 	int err;
    924 
    925 	buf = zfs_alloc(hdl, len);
    926 	if (buf == NULL)
    927 		return (ENOMEM);
    928 
    929 	err = recv_read(hdl, fd, buf, len, byteswap, zc);
    930 	if (err != 0) {
    931 		free(buf);
    932 		return (err);
    933 	}
    934 
    935 	err = nvlist_unpack(buf, len, nvp, 0);
    936 	free(buf);
    937 	if (err != 0) {
    938 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
    939 		    "stream (malformed nvlist)"));
    940 		return (EINVAL);
    941 	}
    942 	return (0);
    943 }
    944 
    945 static int
    946 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
    947     int baselen, char *newname, recvflags_t flags)
    948 {
    949 	static int seq;
    950 	zfs_cmd_t zc = { 0 };
    951 	int err;
    952 	prop_changelist_t *clp;
    953 	zfs_handle_t *zhp;
    954 
    955 	zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
    956 	if (zhp == NULL)
    957 		return (-1);
    958 	clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
    959 	    flags.force ? MS_FORCE : 0);
    960 	zfs_close(zhp);
    961 	if (clp == NULL)
    962 		return (-1);
    963 	err = changelist_prefix(clp);
    964 	if (err)
    965 		return (err);
    966 
    967 	zc.zc_objset_type = DMU_OST_ZFS;
    968 	(void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
    969 
    970 	if (tryname) {
    971 		(void) strcpy(newname, tryname);
    972 
    973 		(void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
    974 
    975 		if (flags.verbose) {
    976 			(void) printf("attempting rename %s to %s\n",
    977 			    zc.zc_name, zc.zc_value);
    978 		}
    979 		err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
    980 		if (err == 0)
    981 			changelist_rename(clp, name, tryname);
    982 	} else {
    983 		err = ENOENT;
    984 	}
    985 
    986 	if (err != 0 && strncmp(name+baselen, "recv-", 5) != 0) {
    987 		seq++;
    988 
    989 		(void) strncpy(newname, name, baselen);
    990 		(void) snprintf(newname+baselen, ZFS_MAXNAMELEN-baselen,
    991 		    "recv-%u-%u", getpid(), seq);
    992 		(void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
    993 
    994 		if (flags.verbose) {
    995 			(void) printf("failed - trying rename %s to %s\n",
    996 			    zc.zc_name, zc.zc_value);
    997 		}
    998 		err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
    999 		if (err == 0)
   1000 			changelist_rename(clp, name, newname);
   1001 		if (err && flags.verbose) {
   1002 			(void) printf("failed (%u) - "
   1003 			    "will try again on next pass\n", errno);
   1004 		}
   1005 		err = EAGAIN;
   1006 	} else if (flags.verbose) {
   1007 		if (err == 0)
   1008 			(void) printf("success\n");
   1009 		else
   1010 			(void) printf("failed (%u)\n", errno);
   1011 	}
   1012 
   1013 	(void) changelist_postfix(clp);
   1014 	changelist_free(clp);
   1015 
   1016 	return (err);
   1017 }
   1018 
   1019 static int
   1020 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
   1021     char *newname, recvflags_t flags)
   1022 {
   1023 	zfs_cmd_t zc = { 0 };
   1024 	int err = 0;
   1025 	prop_changelist_t *clp;
   1026 	zfs_handle_t *zhp;
   1027 	boolean_t defer = B_FALSE;
   1028 	int spa_version;
   1029 
   1030 	zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
   1031 	if (zhp == NULL)
   1032 		return (-1);
   1033 	clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
   1034 	    flags.force ? MS_FORCE : 0);
   1035 	if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
   1036 	    zfs_spa_version(zhp, &spa_version) == 0 &&
   1037 	    spa_version >= SPA_VERSION_USERREFS)
   1038 		defer = B_TRUE;
   1039 	zfs_close(zhp);
   1040 	if (clp == NULL)
   1041 		return (-1);
   1042 	err = changelist_prefix(clp);
   1043 	if (err)
   1044 		return (err);
   1045 
   1046 	zc.zc_objset_type = DMU_OST_ZFS;
   1047 	zc.zc_defer_destroy = defer;
   1048 	(void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
   1049 
   1050 	if (flags.verbose)
   1051 		(void) printf("attempting destroy %s\n", zc.zc_name);
   1052 	err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
   1053 	if (err == 0) {
   1054 		if (flags.verbose)
   1055 			(void) printf("success\n");
   1056 		changelist_remove(clp, zc.zc_name);
   1057 	}
   1058 
   1059 	(void) changelist_postfix(clp);
   1060 	changelist_free(clp);
   1061 
   1062 	/*
   1063 	 * Deferred destroy should always succeed. Since we can't tell
   1064 	 * if it destroyed the dataset or just marked it for deferred
   1065 	 * destroy, always do the rename just in case.
   1066 	 */
   1067 	if (err != 0 || defer)
   1068 		err = recv_rename(hdl, name, NULL, baselen, newname, flags);
   1069 
   1070 	return (err);
   1071 }
   1072 
   1073 typedef struct guid_to_name_data {
   1074 	uint64_t guid;
   1075 	char *name;
   1076 } guid_to_name_data_t;
   1077 
   1078 static int
   1079 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
   1080 {
   1081 	guid_to_name_data_t *gtnd = arg;
   1082 	int err;
   1083 
   1084 	if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
   1085 		(void) strcpy(gtnd->name, zhp->zfs_name);
   1086 		return (EEXIST);
   1087 	}
   1088 	err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
   1089 	zfs_close(zhp);
   1090 	return (err);
   1091 }
   1092 
   1093 static int
   1094 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
   1095     char *name)
   1096 {
   1097 	/* exhaustive search all local snapshots */
   1098 	guid_to_name_data_t gtnd;
   1099 	int err = 0;
   1100 	zfs_handle_t *zhp;
   1101 	char *cp;
   1102 
   1103 	gtnd.guid = guid;
   1104 	gtnd.name = name;
   1105 
   1106 	if (strchr(parent, '@') == NULL) {
   1107 		zhp = make_dataset_handle(hdl, parent);
   1108 		if (zhp != NULL) {
   1109 			err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
   1110 			zfs_close(zhp);
   1111 			if (err == EEXIST)
   1112 				return (0);
   1113 		}
   1114 	}
   1115 
   1116 	cp = strchr(parent, '/');
   1117 	if (cp)
   1118 		*cp = '\0';
   1119 	zhp = make_dataset_handle(hdl, parent);
   1120 	if (cp)
   1121 		*cp = '/';
   1122 
   1123 	if (zhp) {
   1124 		err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
   1125 		zfs_close(zhp);
   1126 	}
   1127 
   1128 	return (err == EEXIST ? 0 : ENOENT);
   1129 
   1130 }
   1131 
   1132 /*
   1133  * Return true if dataset guid1 is created before guid2.
   1134  */
   1135 static int
   1136 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
   1137     uint64_t guid1, uint64_t guid2)
   1138 {
   1139 	nvlist_t *nvfs;
   1140 	char *fsname, *snapname;
   1141 	char buf[ZFS_MAXNAMELEN];
   1142 	int rv;
   1143 	zfs_node_t zn1, zn2;
   1144 
   1145 	if (guid2 == 0)
   1146 		return (0);
   1147 	if (guid1 == 0)
   1148 		return (1);
   1149 
   1150 	nvfs = fsavl_find(avl, guid1, &snapname);
   1151 	VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
   1152 	(void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
   1153 	zn1.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
   1154 	if (zn1.zn_handle == NULL)
   1155 		return (-1);
   1156 
   1157 	nvfs = fsavl_find(avl, guid2, &snapname);
   1158 	VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
   1159 	(void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
   1160 	zn2.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
   1161 	if (zn2.zn_handle == NULL) {
   1162 		zfs_close(zn2.zn_handle);
   1163 		return (-1);
   1164 	}
   1165 
   1166 	rv = (zfs_snapshot_compare(&zn1, &zn2) == -1);
   1167 
   1168 	zfs_close(zn1.zn_handle);
   1169 	zfs_close(zn2.zn_handle);
   1170 
   1171 	return (rv);
   1172 }
   1173 
   1174 static int
   1175 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
   1176     recvflags_t flags, nvlist_t *stream_nv, avl_tree_t *stream_avl)
   1177 {
   1178 	nvlist_t *local_nv;
   1179 	avl_tree_t *local_avl;
   1180 	nvpair_t *fselem, *nextfselem;
   1181 	char *tosnap, *fromsnap;
   1182 	char newname[ZFS_MAXNAMELEN];
   1183 	int error;
   1184 	boolean_t needagain, progress;
   1185 	char *s1, *s2;
   1186 
   1187 	VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
   1188 	VERIFY(0 == nvlist_lookup_string(stream_nv, "tosnap", &tosnap));
   1189 
   1190 	if (flags.dryrun)
   1191 		return (0);
   1192 
   1193 again:
   1194 	needagain = progress = B_FALSE;
   1195 
   1196 	if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
   1197 	    &local_nv, &local_avl)) != 0)
   1198 		return (error);
   1199 
   1200 	/*
   1201 	 * Process deletes and renames
   1202 	 */
   1203 	for (fselem = nvlist_next_nvpair(local_nv, NULL);
   1204 	    fselem; fselem = nextfselem) {
   1205 		nvlist_t *nvfs, *snaps;
   1206 		nvlist_t *stream_nvfs = NULL;
   1207 		nvpair_t *snapelem, *nextsnapelem;
   1208 		uint64_t fromguid = 0;
   1209 		uint64_t originguid = 0;
   1210 		uint64_t stream_originguid = 0;
   1211 		uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
   1212 		char *fsname, *stream_fsname;
   1213 
   1214 		nextfselem = nvlist_next_nvpair(local_nv, fselem);
   1215 
   1216 		VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
   1217 		VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
   1218 		VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
   1219 		VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
   1220 		    &parent_fromsnap_guid));
   1221 		(void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
   1222 
   1223 		/*
   1224 		 * First find the stream's fs, so we can check for
   1225 		 * a different origin (due to "zfs promote")
   1226 		 */
   1227 		for (snapelem = nvlist_next_nvpair(snaps, NULL);
   1228 		    snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
   1229 			uint64_t thisguid;
   1230 
   1231 			VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
   1232 			stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
   1233 
   1234 			if (stream_nvfs != NULL)
   1235 				break;
   1236 		}
   1237 
   1238 		/* check for promote */
   1239 		(void) nvlist_lookup_uint64(stream_nvfs, "origin",
   1240 		    &stream_originguid);
   1241 		if (stream_nvfs && originguid != stream_originguid) {
   1242 			switch (created_before(hdl, local_avl,
   1243 			    stream_originguid, originguid)) {
   1244 			case 1: {
   1245 				/* promote it! */
   1246 				zfs_cmd_t zc = { 0 };
   1247 				nvlist_t *origin_nvfs;
   1248 				char *origin_fsname;
   1249 
   1250 				if (flags.verbose)
   1251 					(void) printf("promoting %s\n", fsname);
   1252 
   1253 				origin_nvfs = fsavl_find(local_avl, originguid,
   1254 				    NULL);
   1255 				VERIFY(0 == nvlist_lookup_string(origin_nvfs,
   1256 				    "name", &origin_fsname));
   1257 				(void) strlcpy(zc.zc_value, origin_fsname,
   1258 				    sizeof (zc.zc_value));
   1259 				(void) strlcpy(zc.zc_name, fsname,
   1260 				    sizeof (zc.zc_name));
   1261 				error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
   1262 				if (error == 0)
   1263 					progress = B_TRUE;
   1264 				break;
   1265 			}
   1266 			default:
   1267 				break;
   1268 			case -1:
   1269 				fsavl_destroy(local_avl);
   1270 				nvlist_free(local_nv);
   1271 				return (-1);
   1272 			}
   1273 			/*
   1274 			 * We had/have the wrong origin, therefore our
   1275 			 * list of snapshots is wrong.  Need to handle
   1276 			 * them on the next pass.
   1277 			 */
   1278 			needagain = B_TRUE;
   1279 			continue;
   1280 		}
   1281 
   1282 		for (snapelem = nvlist_next_nvpair(snaps, NULL);
   1283 		    snapelem; snapelem = nextsnapelem) {
   1284 			uint64_t thisguid;
   1285 			char *stream_snapname;
   1286 			nvlist_t *found, *props;
   1287 
   1288 			nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
   1289 
   1290 			VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
   1291 			found = fsavl_find(stream_avl, thisguid,
   1292 			    &stream_snapname);
   1293 
   1294 			/* check for delete */
   1295 			if (found == NULL) {
   1296 				char name[ZFS_MAXNAMELEN];
   1297 
   1298 				if (!flags.force)
   1299 					continue;
   1300 
   1301 				(void) snprintf(name, sizeof (name), "%s@%s",
   1302 				    fsname, nvpair_name(snapelem));
   1303 
   1304 				error = recv_destroy(hdl, name,
   1305 				    strlen(fsname)+1, newname, flags);
   1306 				if (error)
   1307 					needagain = B_TRUE;
   1308 				else
   1309 					progress = B_TRUE;
   1310 				continue;
   1311 			}
   1312 
   1313 			stream_nvfs = found;
   1314 
   1315 			if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
   1316 			    &props) && 0 == nvlist_lookup_nvlist(props,
   1317 			    stream_snapname, &props)) {
   1318 				zfs_cmd_t zc = { 0 };
   1319 
   1320 				zc.zc_cookie = B_TRUE; /* clear current props */
   1321 				(void) snprintf(zc.zc_name, sizeof (zc.zc_name),
   1322 				    "%s@%s", fsname, nvpair_name(snapelem));
   1323 				if (zcmd_write_src_nvlist(hdl, &zc,
   1324 				    props) == 0) {
   1325 					(void) zfs_ioctl(hdl,
   1326 					    ZFS_IOC_SET_PROP, &zc);
   1327 					zcmd_free_nvlists(&zc);
   1328 				}
   1329 			}
   1330 
   1331 			/* check for different snapname */
   1332 			if (strcmp(nvpair_name(snapelem),
   1333 			    stream_snapname) != 0) {
   1334 				char name[ZFS_MAXNAMELEN];
   1335 				char tryname[ZFS_MAXNAMELEN];
   1336 
   1337 				(void) snprintf(name, sizeof (name), "%s@%s",
   1338 				    fsname, nvpair_name(snapelem));
   1339 				(void) snprintf(tryname, sizeof (name), "%s@%s",
   1340 				    fsname, stream_snapname);
   1341 
   1342 				error = recv_rename(hdl, name, tryname,
   1343 				    strlen(fsname)+1, newname, flags);
   1344 				if (error)
   1345 					needagain = B_TRUE;
   1346 				else
   1347 					progress = B_TRUE;
   1348 			}
   1349 
   1350 			if (strcmp(stream_snapname, fromsnap) == 0)
   1351 				fromguid = thisguid;
   1352 		}
   1353 
   1354 		/* check for delete */
   1355 		if (stream_nvfs == NULL) {
   1356 			if (!flags.force)
   1357 				continue;
   1358 
   1359 			error = recv_destroy(hdl, fsname, strlen(tofs)+1,
   1360 			    newname, flags);
   1361 			if (error)
   1362 				needagain = B_TRUE;
   1363 			else
   1364 				progress = B_TRUE;
   1365 			continue;
   1366 		}
   1367 
   1368 		if (fromguid == 0 && flags.verbose) {
   1369 			(void) printf("local fs %s does not have fromsnap "
   1370 			    "(%s in stream); must have been deleted locally; "
   1371 			    "ignoring\n", fsname, fromsnap);
   1372 			continue;
   1373 		}
   1374 
   1375 		VERIFY(0 == nvlist_lookup_string(stream_nvfs,
   1376 		    "name", &stream_fsname));
   1377 		VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
   1378 		    "parentfromsnap", &stream_parent_fromsnap_guid));
   1379 
   1380 		s1 = strrchr(fsname, '/');
   1381 		s2 = strrchr(stream_fsname, '/');
   1382 
   1383 		/* check for rename */
   1384 		if ((stream_parent_fromsnap_guid != 0 &&
   1385 		    stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
   1386 		    ((s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
   1387 			nvlist_t *parent;
   1388 			char tryname[ZFS_MAXNAMELEN];
   1389 
   1390 			parent = fsavl_find(local_avl,
   1391 			    stream_parent_fromsnap_guid, NULL);
   1392 			/*
   1393 			 * NB: parent might not be found if we used the
   1394 			 * tosnap for stream_parent_fromsnap_guid,
   1395 			 * because the parent is a newly-created fs;
   1396 			 * we'll be able to rename it after we recv the
   1397 			 * new fs.
   1398 			 */
   1399 			if (parent != NULL) {
   1400 				char *pname;
   1401 
   1402 				VERIFY(0 == nvlist_lookup_string(parent, "name",
   1403 				    &pname));
   1404 				(void) snprintf(tryname, sizeof (tryname),
   1405 				    "%s%s", pname, strrchr(stream_fsname, '/'));
   1406 			} else {
   1407 				tryname[0] = '\0';
   1408 				if (flags.verbose) {
   1409 					(void) printf("local fs %s new parent "
   1410 					    "not found\n", fsname);
   1411 				}
   1412 			}
   1413 
   1414 			error = recv_rename(hdl, fsname, tryname,
   1415 			    strlen(tofs)+1, newname, flags);
   1416 			if (error)
   1417 				needagain = B_TRUE;
   1418 			else
   1419 				progress = B_TRUE;
   1420 		}
   1421 	}
   1422 
   1423 	fsavl_destroy(local_avl);
   1424 	nvlist_free(local_nv);
   1425 
   1426 	if (needagain && progress) {
   1427 		/* do another pass to fix up temporary names */
   1428 		if (flags.verbose)
   1429 			(void) printf("another pass:\n");
   1430 		goto again;
   1431 	}
   1432 
   1433 	return (needagain);
   1434 }
   1435 
   1436 static int
   1437 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
   1438     recvflags_t flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
   1439     char **top_zfs)
   1440 {
   1441 	nvlist_t *stream_nv = NULL;
   1442 	avl_tree_t *stream_avl = NULL;
   1443 	char *fromsnap = NULL;
   1444 	char tofs[ZFS_MAXNAMELEN];
   1445 	char errbuf[1024];
   1446 	dmu_replay_record_t drre;
   1447 	int error;
   1448 	boolean_t anyerr = B_FALSE;
   1449 	boolean_t softerr = B_FALSE;
   1450 
   1451 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
   1452 	    "cannot receive"));
   1453 
   1454 	if (strchr(destname, '@')) {
   1455 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1456 		    "can not specify snapshot name for multi-snapshot stream"));
   1457 		return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
   1458 	}
   1459 
   1460 	assert(drr->drr_type == DRR_BEGIN);
   1461 	assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
   1462 	assert(drr->drr_u.drr_begin.drr_version == DMU_BACKUP_HEADER_VERSION);
   1463 
   1464 	/*
   1465 	 * Read in the nvlist from the stream.
   1466 	 */
   1467 	if (drr->drr_payloadlen != 0) {
   1468 		if (!flags.isprefix) {
   1469 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1470 			    "must use -d to receive replication "
   1471 			    "(send -R) stream"));
   1472 			return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
   1473 		}
   1474 
   1475 		error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
   1476 		    &stream_nv, flags.byteswap, zc);
   1477 		if (error) {
   1478 			error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
   1479 			goto out;
   1480 		}
   1481 	}
   1482 
   1483 	/*
   1484 	 * Read in the end record and verify checksum.
   1485 	 */
   1486 	if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
   1487 	    flags.byteswap, NULL)))
   1488 		goto out;
   1489 	if (flags.byteswap) {
   1490 		drre.drr_type = BSWAP_32(drre.drr_type);
   1491 		drre.drr_u.drr_end.drr_checksum.zc_word[0] =
   1492 		    BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
   1493 		drre.drr_u.drr_end.drr_checksum.zc_word[1] =
   1494 		    BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
   1495 		drre.drr_u.drr_end.drr_checksum.zc_word[2] =
   1496 		    BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
   1497 		drre.drr_u.drr_end.drr_checksum.zc_word[3] =
   1498 		    BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
   1499 	}
   1500 	if (drre.drr_type != DRR_END) {
   1501 		error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
   1502 		goto out;
   1503 	}
   1504 	if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
   1505 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1506 		    "incorrect header checksum"));
   1507 		error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
   1508 		goto out;
   1509 	}
   1510 
   1511 	(void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
   1512 
   1513 	if (drr->drr_payloadlen != 0) {
   1514 		nvlist_t *stream_fss;
   1515 
   1516 		VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
   1517 		    &stream_fss));
   1518 		if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
   1519 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1520 			    "couldn't allocate avl tree"));
   1521 			error = zfs_error(hdl, EZFS_NOMEM, errbuf);
   1522 			goto out;
   1523 		}
   1524 
   1525 		if (fromsnap != NULL) {
   1526 			(void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
   1527 			if (flags.isprefix) {
   1528 				int i = strcspn(drr->drr_u.drr_begin.drr_toname,
   1529 				    "/@");
   1530 				/* zfs_receive_one() will create_parents() */
   1531 				(void) strlcat(tofs,
   1532 				    &drr->drr_u.drr_begin.drr_toname[i],
   1533 				    ZFS_MAXNAMELEN);
   1534 				*strchr(tofs, '@') = '\0';
   1535 			}
   1536 			softerr = recv_incremental_replication(hdl, tofs,
   1537 			    flags, stream_nv, stream_avl);
   1538 		}
   1539 	}
   1540 
   1541 
   1542 	/* Finally, receive each contained stream */
   1543 	do {
   1544 		/*
   1545 		 * we should figure out if it has a recoverable
   1546 		 * error, in which case do a recv_skip() and drive on.
   1547 		 * Note, if we fail due to already having this guid,
   1548 		 * zfs_receive_one() will take care of it (ie,
   1549 		 * recv_skip() and return 0).
   1550 		 */
   1551 		error = zfs_receive_impl(hdl, destname, flags, fd,
   1552 		    stream_avl, top_zfs);
   1553 		if (error == ENODATA) {
   1554 			error = 0;
   1555 			break;
   1556 		}
   1557 		anyerr |= error;
   1558 	} while (error == 0);
   1559 
   1560 	if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
   1561 		/*
   1562 		 * Now that we have the fs's they sent us, try the
   1563 		 * renames again.
   1564 		 */
   1565 		softerr = recv_incremental_replication(hdl, tofs, flags,
   1566 		    stream_nv, stream_avl);
   1567 	}
   1568 
   1569 out:
   1570 	fsavl_destroy(stream_avl);
   1571 	if (stream_nv)
   1572 		nvlist_free(stream_nv);
   1573 	if (softerr)
   1574 		error = -2;
   1575 	if (anyerr)
   1576 		error = -1;
   1577 	return (error);
   1578 }
   1579 
   1580 static int
   1581 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
   1582 {
   1583 	dmu_replay_record_t *drr;
   1584 	void *buf = malloc(1<<20);
   1585 
   1586 	/* XXX would be great to use lseek if possible... */
   1587 	drr = buf;
   1588 
   1589 	while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
   1590 	    byteswap, NULL) == 0) {
   1591 		if (byteswap)
   1592 			drr->drr_type = BSWAP_32(drr->drr_type);
   1593 
   1594 		switch (drr->drr_type) {
   1595 		case DRR_BEGIN:
   1596 			/* NB: not to be used on v2 stream packages */
   1597 			assert(drr->drr_payloadlen == 0);
   1598 			break;
   1599 
   1600 		case DRR_END:
   1601 			free(buf);
   1602 			return (0);
   1603 
   1604 		case DRR_OBJECT:
   1605 			if (byteswap) {
   1606 				drr->drr_u.drr_object.drr_bonuslen =
   1607 				    BSWAP_32(drr->drr_u.drr_object.
   1608 				    drr_bonuslen);
   1609 			}
   1610 			(void) recv_read(hdl, fd, buf,
   1611 			    P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
   1612 			    B_FALSE, NULL);
   1613 			break;
   1614 
   1615 		case DRR_WRITE:
   1616 			if (byteswap) {
   1617 				drr->drr_u.drr_write.drr_length =
   1618 				    BSWAP_64(drr->drr_u.drr_write.drr_length);
   1619 			}
   1620 			(void) recv_read(hdl, fd, buf,
   1621 			    drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
   1622 			break;
   1623 
   1624 		case DRR_FREEOBJECTS:
   1625 		case DRR_FREE:
   1626 			break;
   1627 
   1628 		default:
   1629 			assert(!"invalid record type");
   1630 		}
   1631 	}
   1632 
   1633 	free(buf);
   1634 	return (-1);
   1635 }
   1636 
   1637 /*
   1638  * Restores a backup of tosnap from the file descriptor specified by infd.
   1639  */
   1640 static int
   1641 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
   1642     recvflags_t flags, dmu_replay_record_t *drr,
   1643     dmu_replay_record_t *drr_noswap, avl_tree_t *stream_avl,
   1644     char **top_zfs)
   1645 {
   1646 	zfs_cmd_t zc = { 0 };
   1647 	time_t begin_time;
   1648 	int ioctl_err, ioctl_errno, err, choplen;
   1649 	char *cp;
   1650 	struct drr_begin *drrb = &drr->drr_u.drr_begin;
   1651 	char errbuf[1024];
   1652 	char chopprefix[ZFS_MAXNAMELEN];
   1653 	boolean_t newfs = B_FALSE;
   1654 	boolean_t stream_wantsnewfs;
   1655 	uint64_t parent_snapguid = 0;
   1656 	prop_changelist_t *clp = NULL;
   1657 	nvlist_t *snapprops_nvlist = NULL;
   1658 
   1659 	begin_time = time(NULL);
   1660 
   1661 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
   1662 	    "cannot receive"));
   1663 
   1664 	if (stream_avl != NULL) {
   1665 		char *snapname;
   1666 		nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
   1667 		    &snapname);
   1668 		nvlist_t *props;
   1669 		int ret;
   1670 
   1671 		(void) nvlist_lookup_uint64(fs, "parentfromsnap",
   1672 		    &parent_snapguid);
   1673 		err = nvlist_lookup_nvlist(fs, "props", &props);
   1674 		if (err)
   1675 			VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
   1676 
   1677 		if (flags.canmountoff) {
   1678 			VERIFY(0 == nvlist_add_uint64(props,
   1679 			    zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
   1680 		}
   1681 		ret = zcmd_write_src_nvlist(hdl, &zc, props);
   1682 		if (err)
   1683 			nvlist_free(props);
   1684 
   1685 		if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
   1686 			VERIFY(0 == nvlist_lookup_nvlist(props,
   1687 			    snapname, &snapprops_nvlist));
   1688 		}
   1689 
   1690 		if (ret != 0)
   1691 			return (-1);
   1692 	}
   1693 
   1694 	/*
   1695 	 * Determine how much of the snapshot name stored in the stream
   1696 	 * we are going to tack on to the name they specified on the
   1697 	 * command line, and how much we are going to chop off.
   1698 	 *
   1699 	 * If they specified a snapshot, chop the entire name stored in
   1700 	 * the stream.
   1701 	 */
   1702 	(void) strcpy(chopprefix, drrb->drr_toname);
   1703 	if (flags.isprefix) {
   1704 		/*
   1705 		 * They specified a fs with -d, we want to tack on
   1706 		 * everything but the pool name stored in the stream
   1707 		 */
   1708 		if (strchr(tosnap, '@')) {
   1709 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
   1710 			    "argument - snapshot not allowed with -d"));
   1711 			return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
   1712 		}
   1713 		cp = strchr(chopprefix, '/');
   1714 		if (cp == NULL)
   1715 			cp = strchr(chopprefix, '@');
   1716 		*cp = '\0';
   1717 	} else if (strchr(tosnap, '@') == NULL) {
   1718 		/*
   1719 		 * If they specified a filesystem without -d, we want to
   1720 		 * tack on everything after the fs specified in the
   1721 		 * first name from the stream.
   1722 		 */
   1723 		cp = strchr(chopprefix, '@');
   1724 		*cp = '\0';
   1725 	}
   1726 	choplen = strlen(chopprefix);
   1727 
   1728 	/*
   1729 	 * Determine name of destination snapshot, store in zc_value.
   1730 	 */
   1731 	(void) strcpy(zc.zc_value, tosnap);
   1732 	(void) strncat(zc.zc_value, drrb->drr_toname+choplen,
   1733 	    sizeof (zc.zc_value));
   1734 	if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
   1735 		zcmd_free_nvlists(&zc);
   1736 		return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
   1737 	}
   1738 
   1739 	/*
   1740 	 * Determine the name of the origin snapshot, store in zc_string.
   1741 	 */
   1742 	if (drrb->drr_flags & DRR_FLAG_CLONE) {
   1743 		if (guid_to_name(hdl, tosnap,
   1744 		    drrb->drr_fromguid, zc.zc_string) != 0) {
   1745 			zcmd_free_nvlists(&zc);
   1746 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1747 			    "local origin for clone %s does not exist"),
   1748 			    zc.zc_value);
   1749 			return (zfs_error(hdl, EZFS_NOENT, errbuf));
   1750 		}
   1751 		if (flags.verbose)
   1752 			(void) printf("found clone origin %s\n", zc.zc_string);
   1753 	}
   1754 
   1755 	stream_wantsnewfs = (drrb->drr_fromguid == NULL ||
   1756 	    (drrb->drr_flags & DRR_FLAG_CLONE));
   1757 
   1758 	if (stream_wantsnewfs) {
   1759 		/*
   1760 		 * if the parent fs does not exist, look for it based on
   1761 		 * the parent snap GUID
   1762 		 */
   1763 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
   1764 		    "cannot receive new filesystem stream"));
   1765 
   1766 		(void) strcpy(zc.zc_name, zc.zc_value);
   1767 		cp = strrchr(zc.zc_name, '/');
   1768 		if (cp)
   1769 			*cp = '\0';
   1770 		if (cp &&
   1771 		    !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
   1772 			char suffix[ZFS_MAXNAMELEN];
   1773 			(void) strcpy(suffix, strrchr(zc.zc_value, '/'));
   1774 			if (guid_to_name(hdl, tosnap, parent_snapguid,
   1775 			    zc.zc_value) == 0) {
   1776 				*strchr(zc.zc_value, '@') = '\0';
   1777 				(void) strcat(zc.zc_value, suffix);
   1778 			}
   1779 		}
   1780 	} else {
   1781 		/*
   1782 		 * if the fs does not exist, look for it based on the
   1783 		 * fromsnap GUID
   1784 		 */
   1785 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
   1786 		    "cannot receive incremental stream"));
   1787 
   1788 		(void) strcpy(zc.zc_name, zc.zc_value);
   1789 		*strchr(zc.zc_name, '@') = '\0';
   1790 
   1791 		if (!zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
   1792 			char snap[ZFS_MAXNAMELEN];
   1793 			(void) strcpy(snap, strchr(zc.zc_value, '@'));
   1794 			if (guid_to_name(hdl, tosnap, drrb->drr_fromguid,
   1795 			    zc.zc_value) == 0) {
   1796 				*strchr(zc.zc_value, '@') = '\0';
   1797 				(void) strcat(zc.zc_value, snap);
   1798 			}
   1799 		}
   1800 	}
   1801 
   1802 	(void) strcpy(zc.zc_name, zc.zc_value);
   1803 	*strchr(zc.zc_name, '@') = '\0';
   1804 
   1805 	if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
   1806 		zfs_handle_t *zhp;
   1807 		/*
   1808 		 * Destination fs exists.  Therefore this should either
   1809 		 * be an incremental, or the stream specifies a new fs
   1810 		 * (full stream or clone) and they want us to blow it
   1811 		 * away (and have therefore specified -F and removed any
   1812 		 * snapshots).
   1813 		 */
   1814 
   1815 		if (stream_wantsnewfs) {
   1816 			if (!flags.force) {
   1817 				zcmd_free_nvlists(&zc);
   1818 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1819 				    "destination '%s' exists\n"
   1820 				    "must specify -F to overwrite it"),
   1821 				    zc.zc_name);
   1822 				return (zfs_error(hdl, EZFS_EXISTS, errbuf));
   1823 			}
   1824 			if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
   1825 			    &zc) == 0) {
   1826 				zcmd_free_nvlists(&zc);
   1827 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1828 				    "destination has snapshots (eg. %s)\n"
   1829 				    "must destroy them to overwrite it"),
   1830 				    zc.zc_name);
   1831 				return (zfs_error(hdl, EZFS_EXISTS, errbuf));
   1832 			}
   1833 		}
   1834 
   1835 		if ((zhp = zfs_open(hdl, zc.zc_name,
   1836 		    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
   1837 			zcmd_free_nvlists(&zc);
   1838 			return (-1);
   1839 		}
   1840 
   1841 		if (stream_wantsnewfs &&
   1842 		    zhp->zfs_dmustats.dds_origin[0]) {
   1843 			zcmd_free_nvlists(&zc);
   1844 			zfs_close(zhp);
   1845 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1846 			    "destination '%s' is a clone\n"
   1847 			    "must destroy it to overwrite it"),
   1848 			    zc.zc_name);
   1849 			return (zfs_error(hdl, EZFS_EXISTS, errbuf));
   1850 		}
   1851 
   1852 		if (!flags.dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
   1853 		    stream_wantsnewfs) {
   1854 			/* We can't do online recv in this case */
   1855 			clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
   1856 			if (clp == NULL) {
   1857 				zfs_close(zhp);
   1858 				zcmd_free_nvlists(&zc);
   1859 				return (-1);
   1860 			}
   1861 			if (changelist_prefix(clp) != 0) {
   1862 				changelist_free(clp);
   1863 				zfs_close(zhp);
   1864 				zcmd_free_nvlists(&zc);
   1865 				return (-1);
   1866 			}
   1867 		}
   1868 		zfs_close(zhp);
   1869 	} else {
   1870 		/*
   1871 		 * Destination filesystem does not exist.  Therefore we better
   1872 		 * be creating a new filesystem (either from a full backup, or
   1873 		 * a clone).  It would therefore be invalid if the user
   1874 		 * specified only the pool name (i.e. if the destination name
   1875 		 * contained no slash character).
   1876 		 */
   1877 		if (!stream_wantsnewfs ||
   1878 		    (cp = strrchr(zc.zc_name, '/')) == NULL) {
   1879 			zcmd_free_nvlists(&zc);
   1880 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1881 			    "destination '%s' does not exist"), zc.zc_name);
   1882 			return (zfs_error(hdl, EZFS_NOENT, errbuf));
   1883 		}
   1884 
   1885 		/*
   1886 		 * Trim off the final dataset component so we perform the
   1887 		 * recvbackup ioctl to the filesystems's parent.
   1888 		 */
   1889 		*cp = '\0';
   1890 
   1891 		if (flags.isprefix && !flags.dryrun &&
   1892 		    create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
   1893 			zcmd_free_nvlists(&zc);
   1894 			return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
   1895 		}
   1896 
   1897 		newfs = B_TRUE;
   1898 	}
   1899 
   1900 	zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
   1901 	zc.zc_cookie = infd;
   1902 	zc.zc_guid = flags.force;
   1903 	if (flags.verbose) {
   1904 		(void) printf("%s %s stream of %s into %s\n",
   1905 		    flags.dryrun ? "would receive" : "receiving",
   1906 		    drrb->drr_fromguid ? "incremental" : "full",
   1907 		    drrb->drr_toname, zc.zc_value);
   1908 		(void) fflush(stdout);
   1909 	}
   1910 
   1911 	if (flags.dryrun) {
   1912 		zcmd_free_nvlists(&zc);
   1913 		return (recv_skip(hdl, infd, flags.byteswap));
   1914 	}
   1915 
   1916 	err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
   1917 	ioctl_errno = errno;
   1918 	zcmd_free_nvlists(&zc);
   1919 
   1920 	if (err == 0 && snapprops_nvlist) {
   1921 		zfs_cmd_t zc2 = { 0 };
   1922 
   1923 		(void) strcpy(zc2.zc_name, zc.zc_value);
   1924 		if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
   1925 			(void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
   1926 			zcmd_free_nvlists(&zc2);
   1927 		}
   1928 	}
   1929 
   1930 	if (err && (ioctl_errno == ENOENT || ioctl_errno == ENODEV)) {
   1931 		/*
   1932 		 * It may be that this snapshot already exists,
   1933 		 * in which case we want to consume & ignore it
   1934 		 * rather than failing.
   1935 		 */
   1936 		avl_tree_t *local_avl;
   1937 		nvlist_t *local_nv, *fs;
   1938 		char *cp = strchr(zc.zc_value, '@');
   1939 
   1940 		/*
   1941 		 * XXX Do this faster by just iterating over snaps in
   1942 		 * this fs.  Also if zc_value does not exist, we will
   1943 		 * get a strange "does not exist" error message.
   1944 		 */
   1945 		*cp = '\0';
   1946 		if (gather_nvlist(hdl, zc.zc_value, NULL, NULL,
   1947 		    &local_nv, &local_avl) == 0) {
   1948 			*cp = '@';
   1949 			fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
   1950 			fsavl_destroy(local_avl);
   1951 			nvlist_free(local_nv);
   1952 
   1953 			if (fs != NULL) {
   1954 				if (flags.verbose) {
   1955 					(void) printf("snap %s already exists; "
   1956 					    "ignoring\n", zc.zc_value);
   1957 				}
   1958 				ioctl_err = recv_skip(hdl, infd,
   1959 				    flags.byteswap);
   1960 			}
   1961 		}
   1962 		*cp = '@';
   1963 	}
   1964 
   1965 
   1966 	if (ioctl_err != 0) {
   1967 		switch (ioctl_errno) {
   1968 		case ENODEV:
   1969 			cp = strchr(zc.zc_value, '@');
   1970 			*cp = '\0';
   1971 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1972 			    "most recent snapshot of %s does not\n"
   1973 			    "match incremental source"), zc.zc_value);
   1974 			(void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
   1975 			*cp = '@';
   1976 			break;
   1977 		case ETXTBSY:
   1978 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1979 			    "destination %s has been modified\n"
   1980 			    "since most recent snapshot"), zc.zc_name);
   1981 			(void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
   1982 			break;
   1983 		case EEXIST:
   1984 			cp = strchr(zc.zc_value, '@');
   1985 			if (newfs) {
   1986 				/* it's the containing fs that exists */
   1987 				*cp = '\0';
   1988 			}
   1989 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   1990 			    "destination already exists"));
   1991 			(void) zfs_error_fmt(hdl, EZFS_EXISTS,
   1992 			    dgettext(TEXT_DOMAIN, "cannot restore to %s"),
   1993 			    zc.zc_value);
   1994 			*cp = '@';
   1995 			break;
   1996 		case EINVAL:
   1997 			(void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
   1998 			break;
   1999 		case ECKSUM:
   2000 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   2001 			    "invalid stream (checksum mismatch)"));
   2002 			(void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
   2003 			break;
   2004 		default:
   2005 			(void) zfs_standard_error(hdl, ioctl_errno, errbuf);
   2006 		}
   2007 	}
   2008 
   2009 	/*
   2010 	 * Mount the target filesystem (if created).  Also mount any
   2011 	 * children of the target filesystem if we did a replication
   2012 	 * receive (indicated by stream_avl being non-NULL).
   2013 	 */
   2014 	cp = strchr(zc.zc_value, '@');
   2015 	if (cp && (ioctl_err == 0 || !newfs)) {
   2016 		zfs_handle_t *h;
   2017 
   2018 		*cp = '\0';
   2019 		h = zfs_open(hdl, zc.zc_value,
   2020 		    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
   2021 		if (h != NULL) {
   2022 			if (h->zfs_type == ZFS_TYPE_VOLUME) {
   2023 				*cp = '@';
   2024 			} else if (newfs || stream_avl) {
   2025 				/*
   2026 				 * Track the first/top of hierarchy fs,
   2027 				 * for mounting and sharing later.
   2028 				 */
   2029 				if (top_zfs && *top_zfs == NULL)
   2030 					*top_zfs = zfs_strdup(hdl, zc.zc_value);
   2031 			}
   2032 			zfs_close(h);
   2033 		}
   2034 		*cp = '@';
   2035 	}
   2036 
   2037 	if (clp) {
   2038 		err |= changelist_postfix(clp);
   2039 		changelist_free(clp);
   2040 	}
   2041 
   2042 	if (err || ioctl_err)
   2043 		return (-1);
   2044 
   2045 	if (flags.verbose) {
   2046 		char buf1[64];
   2047 		char buf2[64];
   2048 		uint64_t bytes = zc.zc_cookie;
   2049 		time_t delta = time(NULL) - begin_time;
   2050 		if (delta == 0)
   2051 			delta = 1;
   2052 		zfs_nicenum(bytes, buf1, sizeof (buf1));
   2053 		zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
   2054 
   2055 		(void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
   2056 		    buf1, delta, buf2);
   2057 	}
   2058 
   2059 	return (0);
   2060 }
   2061 
   2062 static int
   2063 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
   2064     int infd, avl_tree_t *stream_avl, char **top_zfs)
   2065 {
   2066 	int err;
   2067 	dmu_replay_record_t drr, drr_noswap;
   2068 	struct drr_begin *drrb = &drr.drr_u.drr_begin;
   2069 	char errbuf[1024];
   2070 	zio_cksum_t zcksum = { 0 };
   2071 
   2072 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
   2073 	    "cannot receive"));
   2074 
   2075 	if (flags.isprefix &&
   2076 	    !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
   2077 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
   2078 		    "(%s) does not exist"), tosnap);
   2079 		return (zfs_error(hdl, EZFS_NOENT, errbuf));
   2080 	}
   2081 
   2082 	/* read in the BEGIN record */
   2083 	if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
   2084 	    &zcksum)))
   2085 		return (err);
   2086 
   2087 	if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
   2088 		/* It's the double end record at the end of a package */
   2089 		return (ENODATA);
   2090 	}
   2091 
   2092 	/* the kernel needs the non-byteswapped begin record */
   2093 	drr_noswap = drr;
   2094 
   2095 	flags.byteswap = B_FALSE;
   2096 	if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
   2097 		/*
   2098 		 * We computed the checksum in the wrong byteorder in
   2099 		 * recv_read() above; do it again correctly.
   2100 		 */
   2101 		bzero(&zcksum, sizeof (zio_cksum_t));
   2102 		fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
   2103 		flags.byteswap = B_TRUE;
   2104 
   2105 		drr.drr_type = BSWAP_32(drr.drr_type);
   2106 		drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
   2107 		drrb->drr_magic = BSWAP_64(drrb->drr_magic);
   2108 		drrb->drr_version = BSWAP_64(drrb->drr_version);
   2109 		drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
   2110 		drrb->drr_type = BSWAP_32(drrb->drr_type);
   2111 		drrb->drr_flags = BSWAP_32(drrb->drr_flags);
   2112 		drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
   2113 		drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
   2114 	}
   2115 
   2116 	if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
   2117 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
   2118 		    "stream (bad magic number)"));
   2119 		return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
   2120 	}
   2121 
   2122 	if (strchr(drrb->drr_toname, '@') == NULL) {
   2123 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
   2124 		    "stream (bad snapshot name)"));
   2125 		return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
   2126 	}
   2127 
   2128 	if (drrb->drr_version == DMU_BACKUP_STREAM_VERSION) {
   2129 		return (zfs_receive_one(hdl, infd, tosnap, flags,
   2130 		    &drr, &drr_noswap, stream_avl, top_zfs));
   2131 	} else if (drrb->drr_version == DMU_BACKUP_HEADER_VERSION) {
   2132 		return (zfs_receive_package(hdl, infd, tosnap, flags,
   2133 		    &drr, &zcksum, top_zfs));
   2134 	} else {
   2135 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
   2136 		    "stream is unsupported version %llu"),
   2137 		    drrb->drr_version);
   2138 		return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
   2139 	}
   2140 }
   2141 
   2142 /*
   2143  * Restores a backup of tosnap from the file descriptor specified by infd.
   2144  * Return 0 on total success, -2 if some things couldn't be
   2145  * destroyed/renamed/promoted, -1 if some things couldn't be received.
   2146  * (-1 will override -2).
   2147  */
   2148 int
   2149 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
   2150     int infd, avl_tree_t *stream_avl)
   2151 {
   2152 	char *top_zfs = NULL;
   2153 	int err;
   2154 
   2155 	err = zfs_receive_impl(hdl, tosnap, flags, infd, stream_avl, &top_zfs);
   2156 
   2157 	if (err == 0 && !flags.nomount && top_zfs) {
   2158 		zfs_handle_t *zhp;
   2159 		prop_changelist_t *clp;
   2160 
   2161 		zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
   2162 		if (zhp != NULL) {
   2163 			clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
   2164 			    CL_GATHER_MOUNT_ALWAYS, 0);
   2165 			zfs_close(zhp);
   2166 			if (clp != NULL) {
   2167 				/* mount and share received datasets */
   2168 				err = changelist_postfix(clp);
   2169 				changelist_free(clp);
   2170 			}
   2171 		}
   2172 		if (zhp == NULL || clp == NULL || err)
   2173 			err = -1;
   2174 	}
   2175 	if (top_zfs)
   2176 		free(top_zfs);
   2177 
   2178 	return (err);
   2179 }
   2180