1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  27  * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
  28  * Copyright (c) 2014 Integros [integros.com]
  29  */
  30 
  31 #include <assert.h>
  32 #include <ctype.h>
  33 #include <errno.h>
  34 #include <libintl.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <strings.h>
  38 #include <unistd.h>
  39 #include <stddef.h>
  40 #include <fcntl.h>
  41 #include <sys/mount.h>
  42 #include <pthread.h>
  43 #include <umem.h>
  44 #include <time.h>
  45 
  46 #include <libzfs.h>
  47 #include <libzfs_core.h>
  48 
  49 #include "zfs_namecheck.h"
  50 #include "zfs_prop.h"
  51 #include "zfs_fletcher.h"
  52 #include "libzfs_impl.h"
  53 #include <zlib.h>
  54 #include <sha2.h>
  55 #include <sys/zio_checksum.h>
  56 #include <sys/ddt.h>
  57 
  58 /* in libzfs_dataset.c */
  59 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  60 
  61 static int zfs_receive_impl(libzfs_handle_t *, const char *, const char *,
  62     recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **, int,
  63     uint64_t *, const char *);
  64 static int guid_to_name(libzfs_handle_t *, const char *,
  65     uint64_t, boolean_t, char *);
  66 
  67 static const zio_cksum_t zero_cksum = { 0 };
  68 
  69 typedef struct dedup_arg {
  70         int     inputfd;
  71         int     outputfd;
  72         libzfs_handle_t  *dedup_hdl;
  73 } dedup_arg_t;
  74 
  75 typedef struct progress_arg {
  76         zfs_handle_t *pa_zhp;
  77         int pa_fd;
  78         boolean_t pa_parsable;
  79 } progress_arg_t;
  80 
  81 typedef struct dataref {
  82         uint64_t ref_guid;
  83         uint64_t ref_object;
  84         uint64_t ref_offset;
  85 } dataref_t;
  86 
  87 typedef struct dedup_entry {
  88         struct dedup_entry      *dde_next;
  89         zio_cksum_t dde_chksum;
  90         uint64_t dde_prop;
  91         dataref_t dde_ref;
  92 } dedup_entry_t;
  93 
  94 #define MAX_DDT_PHYSMEM_PERCENT         20
  95 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
  96 
  97 typedef struct dedup_table {
  98         dedup_entry_t   **dedup_hash_array;
  99         umem_cache_t    *ddecache;
 100         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
 101         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
 102         uint64_t        ddt_count;
 103         int             numhashbits;
 104         boolean_t       ddt_full;
 105 } dedup_table_t;
 106 
 107 static int
 108 high_order_bit(uint64_t n)
 109 {
 110         int count;
 111 
 112         for (count = 0; n != 0; count++)
 113                 n >>= 1;
 114         return (count);
 115 }
 116 
 117 static size_t
 118 ssread(void *buf, size_t len, FILE *stream)
 119 {
 120         size_t outlen;
 121 
 122         if ((outlen = fread(buf, len, 1, stream)) == 0)
 123                 return (0);
 124 
 125         return (outlen);
 126 }
 127 
 128 static void
 129 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 130     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 131 {
 132         dedup_entry_t   *dde;
 133 
 134         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 135                 if (ddt->ddt_full == B_FALSE) {
 136                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 137                             "Dedup table full.  Deduplication will continue "
 138                             "with existing table entries"));
 139                         ddt->ddt_full = B_TRUE;
 140                 }
 141                 return;
 142         }
 143 
 144         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 145             != NULL) {
 146                 assert(*ddepp == NULL);
 147                 dde->dde_next = NULL;
 148                 dde->dde_chksum = *cs;
 149                 dde->dde_prop = prop;
 150                 dde->dde_ref = *dr;
 151                 *ddepp = dde;
 152                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 153                 ddt->ddt_count++;
 154         }
 155 }
 156 
 157 /*
 158  * Using the specified dedup table, do a lookup for an entry with
 159  * the checksum cs.  If found, return the block's reference info
 160  * in *dr. Otherwise, insert a new entry in the dedup table, using
 161  * the reference information specified by *dr.
 162  *
 163  * return value:  true - entry was found
 164  *                false - entry was not found
 165  */
 166 static boolean_t
 167 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 168     uint64_t prop, dataref_t *dr)
 169 {
 170         uint32_t hashcode;
 171         dedup_entry_t **ddepp;
 172 
 173         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 174 
 175         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 176             ddepp = &((*ddepp)->dde_next)) {
 177                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 178                     (*ddepp)->dde_prop == prop) {
 179                         *dr = (*ddepp)->dde_ref;
 180                         return (B_TRUE);
 181                 }
 182         }
 183         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 184         return (B_FALSE);
 185 }
 186 
 187 static int
 188 dump_record(dmu_replay_record_t *drr, void *payload, int payload_len,
 189     zio_cksum_t *zc, int outfd)
 190 {
 191         ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
 192             ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t));
 193         fletcher_4_incremental_native(drr,
 194             offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc);
 195         if (drr->drr_type != DRR_BEGIN) {
 196                 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u.
 197                     drr_checksum.drr_checksum));
 198                 drr->drr_u.drr_checksum.drr_checksum = *zc;
 199         }
 200         fletcher_4_incremental_native(&drr->drr_u.drr_checksum.drr_checksum,
 201             sizeof (zio_cksum_t), zc);
 202         if (write(outfd, drr, sizeof (*drr)) == -1)
 203                 return (errno);
 204         if (payload_len != 0) {
 205                 fletcher_4_incremental_native(payload, payload_len, zc);
 206                 if (write(outfd, payload, payload_len) == -1)
 207                         return (errno);
 208         }
 209         return (0);
 210 }
 211 
 212 /*
 213  * This function is started in a separate thread when the dedup option
 214  * has been requested.  The main send thread determines the list of
 215  * snapshots to be included in the send stream and makes the ioctl calls
 216  * for each one.  But instead of having the ioctl send the output to the
 217  * the output fd specified by the caller of zfs_send()), the
 218  * ioctl is told to direct the output to a pipe, which is read by the
 219  * alternate thread running THIS function.  This function does the
 220  * dedup'ing by:
 221  *  1. building a dedup table (the DDT)
 222  *  2. doing checksums on each data block and inserting a record in the DDT
 223  *  3. looking for matching checksums, and
 224  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 225  *      a duplicate block is found.
 226  * The output of this function then goes to the output fd requested
 227  * by the caller of zfs_send().
 228  */
 229 static void *
 230 cksummer(void *arg)
 231 {
 232         dedup_arg_t *dda = arg;
 233         char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE);
 234         dmu_replay_record_t thedrr;
 235         dmu_replay_record_t *drr = &thedrr;
 236         FILE *ofp;
 237         int outfd;
 238         dedup_table_t ddt;
 239         zio_cksum_t stream_cksum;
 240         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 241         uint64_t numbuckets;
 242 
 243         ddt.max_ddt_size =
 244             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100,
 245             SMALLEST_POSSIBLE_MAX_DDT_MB << 20);
 246 
 247         numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t));
 248 
 249         /*
 250          * numbuckets must be a power of 2.  Increase number to
 251          * a power of 2 if necessary.
 252          */
 253         if (!ISP2(numbuckets))
 254                 numbuckets = 1 << high_order_bit(numbuckets);
 255 
 256         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 257         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 258             NULL, NULL, NULL, NULL, NULL, 0);
 259         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 260         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 261         ddt.ddt_full = B_FALSE;
 262 
 263         outfd = dda->outputfd;
 264         ofp = fdopen(dda->inputfd, "r");
 265         while (ssread(drr, sizeof (*drr), ofp) != 0) {
 266 
 267                 switch (drr->drr_type) {
 268                 case DRR_BEGIN:
 269                 {
 270                         struct drr_begin *drrb = &drr->drr_u.drr_begin;
 271                         int fflags;
 272                         int sz = 0;
 273                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 274 
 275                         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 276 
 277                         /* set the DEDUP feature flag for this stream */
 278                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 279                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 280                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 281                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 282 
 283                         if (drr->drr_payloadlen != 0) {
 284                                 sz = drr->drr_payloadlen;
 285 
 286                                 if (sz > SPA_MAXBLOCKSIZE) {
 287                                         buf = zfs_realloc(dda->dedup_hdl, buf,
 288                                             SPA_MAXBLOCKSIZE, sz);
 289                                 }
 290                                 (void) ssread(buf, sz, ofp);
 291                                 if (ferror(stdin))
 292                                         perror("fread");
 293                         }
 294                         if (dump_record(drr, buf, sz, &stream_cksum,
 295                             outfd) != 0)
 296                                 goto out;
 297                         break;
 298                 }
 299 
 300                 case DRR_END:
 301                 {
 302                         struct drr_end *drre = &drr->drr_u.drr_end;
 303                         /* use the recalculated checksum */
 304                         drre->drr_checksum = stream_cksum;
 305                         if (dump_record(drr, NULL, 0, &stream_cksum,
 306                             outfd) != 0)
 307                                 goto out;
 308                         break;
 309                 }
 310 
 311                 case DRR_OBJECT:
 312                 {
 313                         struct drr_object *drro = &drr->drr_u.drr_object;
 314                         if (drro->drr_bonuslen > 0) {
 315                                 (void) ssread(buf,
 316                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 317                                     ofp);
 318                         }
 319                         if (dump_record(drr, buf,
 320                             P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 321                             &stream_cksum, outfd) != 0)
 322                                 goto out;
 323                         break;
 324                 }
 325 
 326                 case DRR_SPILL:
 327                 {
 328                         struct drr_spill *drrs = &drr->drr_u.drr_spill;
 329                         (void) ssread(buf, drrs->drr_length, ofp);
 330                         if (dump_record(drr, buf, drrs->drr_length,
 331                             &stream_cksum, outfd) != 0)
 332                                 goto out;
 333                         break;
 334                 }
 335 
 336                 case DRR_FREEOBJECTS:
 337                 {
 338                         if (dump_record(drr, NULL, 0, &stream_cksum,
 339                             outfd) != 0)
 340                                 goto out;
 341                         break;
 342                 }
 343 
 344                 case DRR_WRITE:
 345                 {
 346                         struct drr_write *drrw = &drr->drr_u.drr_write;
 347                         dataref_t       dataref;
 348 
 349                         (void) ssread(buf, drrw->drr_length, ofp);
 350 
 351                         /*
 352                          * Use the existing checksum if it's dedup-capable,
 353                          * else calculate a SHA256 checksum for it.
 354                          */
 355 
 356                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 357                             zero_cksum) ||
 358                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 359                                 SHA256_CTX      ctx;
 360                                 zio_cksum_t     tmpsha256;
 361 
 362                                 SHA256Init(&ctx);
 363                                 SHA256Update(&ctx, buf, drrw->drr_length);
 364                                 SHA256Final(&tmpsha256, &ctx);
 365                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 366                                     BE_64(tmpsha256.zc_word[0]);
 367                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 368                                     BE_64(tmpsha256.zc_word[1]);
 369                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 370                                     BE_64(tmpsha256.zc_word[2]);
 371                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 372                                     BE_64(tmpsha256.zc_word[3]);
 373                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 374                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 375                         }
 376 
 377                         dataref.ref_guid = drrw->drr_toguid;
 378                         dataref.ref_object = drrw->drr_object;
 379                         dataref.ref_offset = drrw->drr_offset;
 380 
 381                         if (ddt_update(dda->dedup_hdl, &ddt,
 382                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 383                             &dataref)) {
 384                                 dmu_replay_record_t wbr_drr = {0};
 385                                 struct drr_write_byref *wbr_drrr =
 386                                     &wbr_drr.drr_u.drr_write_byref;
 387 
 388                                 /* block already present in stream */
 389                                 wbr_drr.drr_type = DRR_WRITE_BYREF;
 390 
 391                                 wbr_drrr->drr_object = drrw->drr_object;
 392                                 wbr_drrr->drr_offset = drrw->drr_offset;
 393                                 wbr_drrr->drr_length = drrw->drr_length;
 394                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 395                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 396                                 wbr_drrr->drr_refobject =
 397                                     dataref.ref_object;
 398                                 wbr_drrr->drr_refoffset =
 399                                     dataref.ref_offset;
 400 
 401                                 wbr_drrr->drr_checksumtype =
 402                                     drrw->drr_checksumtype;
 403                                 wbr_drrr->drr_checksumflags =
 404                                     drrw->drr_checksumtype;
 405                                 wbr_drrr->drr_key.ddk_cksum =
 406                                     drrw->drr_key.ddk_cksum;
 407                                 wbr_drrr->drr_key.ddk_prop =
 408                                     drrw->drr_key.ddk_prop;
 409 
 410                                 if (dump_record(&wbr_drr, NULL, 0,
 411                                     &stream_cksum, outfd) != 0)
 412                                         goto out;
 413                         } else {
 414                                 /* block not previously seen */
 415                                 if (dump_record(drr, buf, drrw->drr_length,
 416                                     &stream_cksum, outfd) != 0)
 417                                         goto out;
 418                         }
 419                         break;
 420                 }
 421 
 422                 case DRR_WRITE_EMBEDDED:
 423                 {
 424                         struct drr_write_embedded *drrwe =
 425                             &drr->drr_u.drr_write_embedded;
 426                         (void) ssread(buf,
 427                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp);
 428                         if (dump_record(drr, buf,
 429                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8),
 430                             &stream_cksum, outfd) != 0)
 431                                 goto out;
 432                         break;
 433                 }
 434 
 435                 case DRR_FREE:
 436                 {
 437                         if (dump_record(drr, NULL, 0, &stream_cksum,
 438                             outfd) != 0)
 439                                 goto out;
 440                         break;
 441                 }
 442 
 443                 default:
 444                         (void) fprintf(stderr, "INVALID record type 0x%x\n",
 445                             drr->drr_type);
 446                         /* should never happen, so assert */
 447                         assert(B_FALSE);
 448                 }
 449         }
 450 out:
 451         umem_cache_destroy(ddt.ddecache);
 452         free(ddt.dedup_hash_array);
 453         free(buf);
 454         (void) fclose(ofp);
 455 
 456         return (NULL);
 457 }
 458 
 459 /*
 460  * Routines for dealing with the AVL tree of fs-nvlists
 461  */
 462 typedef struct fsavl_node {
 463         avl_node_t fn_node;
 464         nvlist_t *fn_nvfs;
 465         char *fn_snapname;
 466         uint64_t fn_guid;
 467 } fsavl_node_t;
 468 
 469 static int
 470 fsavl_compare(const void *arg1, const void *arg2)
 471 {
 472         const fsavl_node_t *fn1 = arg1;
 473         const fsavl_node_t *fn2 = arg2;
 474 
 475         if (fn1->fn_guid > fn2->fn_guid)
 476                 return (+1);
 477         else if (fn1->fn_guid < fn2->fn_guid)
 478                 return (-1);
 479         else
 480                 return (0);
 481 }
 482 
 483 /*
 484  * Given the GUID of a snapshot, find its containing filesystem and
 485  * (optionally) name.
 486  */
 487 static nvlist_t *
 488 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 489 {
 490         fsavl_node_t fn_find;
 491         fsavl_node_t *fn;
 492 
 493         fn_find.fn_guid = snapguid;
 494 
 495         fn = avl_find(avl, &fn_find, NULL);
 496         if (fn) {
 497                 if (snapname)
 498                         *snapname = fn->fn_snapname;
 499                 return (fn->fn_nvfs);
 500         }
 501         return (NULL);
 502 }
 503 
 504 static void
 505 fsavl_destroy(avl_tree_t *avl)
 506 {
 507         fsavl_node_t *fn;
 508         void *cookie;
 509 
 510         if (avl == NULL)
 511                 return;
 512 
 513         cookie = NULL;
 514         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 515                 free(fn);
 516         avl_destroy(avl);
 517         free(avl);
 518 }
 519 
 520 /*
 521  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 522  */
 523 static avl_tree_t *
 524 fsavl_create(nvlist_t *fss)
 525 {
 526         avl_tree_t *fsavl;
 527         nvpair_t *fselem = NULL;
 528 
 529         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 530                 return (NULL);
 531 
 532         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 533             offsetof(fsavl_node_t, fn_node));
 534 
 535         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 536                 nvlist_t *nvfs, *snaps;
 537                 nvpair_t *snapelem = NULL;
 538 
 539                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 540                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 541 
 542                 while ((snapelem =
 543                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 544                         fsavl_node_t *fn;
 545                         uint64_t guid;
 546 
 547                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 548                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 549                                 fsavl_destroy(fsavl);
 550                                 return (NULL);
 551                         }
 552                         fn->fn_nvfs = nvfs;
 553                         fn->fn_snapname = nvpair_name(snapelem);
 554                         fn->fn_guid = guid;
 555 
 556                         /*
 557                          * Note: if there are multiple snaps with the
 558                          * same GUID, we ignore all but one.
 559                          */
 560                         if (avl_find(fsavl, fn, NULL) == NULL)
 561                                 avl_add(fsavl, fn);
 562                         else
 563                                 free(fn);
 564                 }
 565         }
 566 
 567         return (fsavl);
 568 }
 569 
 570 /*
 571  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 572  */
 573 typedef struct send_data {
 574         uint64_t parent_fromsnap_guid;
 575         nvlist_t *parent_snaps;
 576         nvlist_t *fss;
 577         nvlist_t *snapprops;
 578         const char *fromsnap;
 579         const char *tosnap;
 580         boolean_t recursive;
 581 
 582         /*
 583          * The header nvlist is of the following format:
 584          * {
 585          *   "tosnap" -> string
 586          *   "fromsnap" -> string (if incremental)
 587          *   "fss" -> {
 588          *      id -> {
 589          *
 590          *       "name" -> string (full name; for debugging)
 591          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 592          *
 593          *       "props" -> { name -> value (only if set here) }
 594          *       "snaps" -> { name (lastname) -> number (guid) }
 595          *       "snapprops" -> { name (lastname) -> { name -> value } }
 596          *
 597          *       "origin" -> number (guid) (if clone)
 598          *       "sent" -> boolean (not on-disk)
 599          *      }
 600          *   }
 601          * }
 602          *
 603          */
 604 } send_data_t;
 605 
 606 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 607 
 608 static int
 609 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 610 {
 611         send_data_t *sd = arg;
 612         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 613         char *snapname;
 614         nvlist_t *nv;
 615 
 616         snapname = strrchr(zhp->zfs_name, '@')+1;
 617 
 618         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 619         /*
 620          * NB: if there is no fromsnap here (it's a newly created fs in
 621          * an incremental replication), we will substitute the tosnap.
 622          */
 623         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 624             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 625             strcmp(snapname, sd->tosnap) == 0)) {
 626                 sd->parent_fromsnap_guid = guid;
 627         }
 628 
 629         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 630         send_iterate_prop(zhp, nv);
 631         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 632         nvlist_free(nv);
 633 
 634         zfs_close(zhp);
 635         return (0);
 636 }
 637 
 638 static void
 639 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 640 {
 641         nvpair_t *elem = NULL;
 642 
 643         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 644                 char *propname = nvpair_name(elem);
 645                 zfs_prop_t prop = zfs_name_to_prop(propname);
 646                 nvlist_t *propnv;
 647 
 648                 if (!zfs_prop_user(propname)) {
 649                         /*
 650                          * Realistically, this should never happen.  However,
 651                          * we want the ability to add DSL properties without
 652                          * needing to make incompatible version changes.  We
 653                          * need to ignore unknown properties to allow older
 654                          * software to still send datasets containing these
 655                          * properties, with the unknown properties elided.
 656                          */
 657                         if (prop == ZPROP_INVAL)
 658                                 continue;
 659 
 660                         if (zfs_prop_readonly(prop))
 661                                 continue;
 662                 }
 663 
 664                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 665                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 666                     prop == ZFS_PROP_REFQUOTA ||
 667                     prop == ZFS_PROP_REFRESERVATION) {
 668                         char *source;
 669                         uint64_t value;
 670                         verify(nvlist_lookup_uint64(propnv,
 671                             ZPROP_VALUE, &value) == 0);
 672                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 673                                 continue;
 674                         /*
 675                          * May have no source before SPA_VERSION_RECVD_PROPS,
 676                          * but is still modifiable.
 677                          */
 678                         if (nvlist_lookup_string(propnv,
 679                             ZPROP_SOURCE, &source) == 0) {
 680                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 681                                     (strcmp(source,
 682                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 683                                         continue;
 684                         }
 685                 } else {
 686                         char *source;
 687                         if (nvlist_lookup_string(propnv,
 688                             ZPROP_SOURCE, &source) != 0)
 689                                 continue;
 690                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 691                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 692                                 continue;
 693                 }
 694 
 695                 if (zfs_prop_user(propname) ||
 696                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 697                         char *value;
 698                         verify(nvlist_lookup_string(propnv,
 699                             ZPROP_VALUE, &value) == 0);
 700                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 701                 } else {
 702                         uint64_t value;
 703                         verify(nvlist_lookup_uint64(propnv,
 704                             ZPROP_VALUE, &value) == 0);
 705                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 706                 }
 707         }
 708 }
 709 
 710 /*
 711  * recursively generate nvlists describing datasets.  See comment
 712  * for the data structure send_data_t above for description of contents
 713  * of the nvlist.
 714  */
 715 static int
 716 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 717 {
 718         send_data_t *sd = arg;
 719         nvlist_t *nvfs, *nv;
 720         int rv = 0;
 721         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 722         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 723         char guidstring[64];
 724 
 725         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 726         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 727         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 728             sd->parent_fromsnap_guid));
 729 
 730         if (zhp->zfs_dmustats.dds_origin[0]) {
 731                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 732                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 733                 if (origin == NULL)
 734                         return (-1);
 735                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 736                     origin->zfs_dmustats.dds_guid));
 737         }
 738 
 739         /* iterate over props */
 740         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 741         send_iterate_prop(zhp, nv);
 742         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 743         nvlist_free(nv);
 744 
 745         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 746         sd->parent_fromsnap_guid = 0;
 747         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 748         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 749         (void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
 750         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 751         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 752         nvlist_free(sd->parent_snaps);
 753         nvlist_free(sd->snapprops);
 754 
 755         /* add this fs to nvlist */
 756         (void) snprintf(guidstring, sizeof (guidstring),
 757             "0x%llx", (longlong_t)guid);
 758         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 759         nvlist_free(nvfs);
 760 
 761         /* iterate over children */
 762         if (sd->recursive)
 763                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 764 
 765         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 766 
 767         zfs_close(zhp);
 768         return (rv);
 769 }
 770 
 771 static int
 772 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 773     const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
 774 {
 775         zfs_handle_t *zhp;
 776         send_data_t sd = { 0 };
 777         int error;
 778 
 779         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 780         if (zhp == NULL)
 781                 return (EZFS_BADTYPE);
 782 
 783         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 784         sd.fromsnap = fromsnap;
 785         sd.tosnap = tosnap;
 786         sd.recursive = recursive;
 787 
 788         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 789                 nvlist_free(sd.fss);
 790                 if (avlp != NULL)
 791                         *avlp = NULL;
 792                 *nvlp = NULL;
 793                 return (error);
 794         }
 795 
 796         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 797                 nvlist_free(sd.fss);
 798                 *nvlp = NULL;
 799                 return (EZFS_NOMEM);
 800         }
 801 
 802         *nvlp = sd.fss;
 803         return (0);
 804 }
 805 
 806 /*
 807  * Routines specific to "zfs send"
 808  */
 809 typedef struct send_dump_data {
 810         /* these are all just the short snapname (the part after the @) */
 811         const char *fromsnap;
 812         const char *tosnap;
 813         char prevsnap[ZFS_MAXNAMELEN];
 814         uint64_t prevsnap_obj;
 815         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 816         boolean_t verbose, dryrun, parsable, progress, embed_data, std_out;
 817         boolean_t large_block;
 818         int outfd;
 819         boolean_t err;
 820         nvlist_t *fss;
 821         nvlist_t *snapholds;
 822         avl_tree_t *fsavl;
 823         snapfilter_cb_t *filter_cb;
 824         void *filter_cb_arg;
 825         nvlist_t *debugnv;
 826         char holdtag[ZFS_MAXNAMELEN];
 827         int cleanup_fd;
 828         uint64_t size;
 829 } send_dump_data_t;
 830 
 831 static int
 832 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
 833     boolean_t fromorigin, uint64_t *sizep)
 834 {
 835         zfs_cmd_t zc = { 0 };
 836         libzfs_handle_t *hdl = zhp->zfs_hdl;
 837 
 838         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 839         assert(fromsnap_obj == 0 || !fromorigin);
 840 
 841         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 842         zc.zc_obj = fromorigin;
 843         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 844         zc.zc_fromobj = fromsnap_obj;
 845         zc.zc_guid = 1;  /* estimate flag */
 846 
 847         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 848                 char errbuf[1024];
 849                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 850                     "warning: cannot estimate space for '%s'"), zhp->zfs_name);
 851 
 852                 switch (errno) {
 853                 case EXDEV:
 854                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 855                             "not an earlier snapshot from the same fs"));
 856                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 857 
 858                 case ENOENT:
 859                         if (zfs_dataset_exists(hdl, zc.zc_name,
 860                             ZFS_TYPE_SNAPSHOT)) {
 861                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 862                                     "incremental source (@%s) does not exist"),
 863                                     zc.zc_value);
 864                         }
 865                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 866 
 867                 case EDQUOT:
 868                 case EFBIG:
 869                 case EIO:
 870                 case ENOLINK:
 871                 case ENOSPC:
 872                 case ENOSTR:
 873                 case ENXIO:
 874                 case EPIPE:
 875                 case ERANGE:
 876                 case EFAULT:
 877                 case EROFS:
 878                         zfs_error_aux(hdl, 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         *sizep = zc.zc_objset_type;
 887 
 888         return (0);
 889 }
 890 
 891 /*
 892  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 893  * NULL) to the file descriptor specified by outfd.
 894  */
 895 static int
 896 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
 897     boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
 898     nvlist_t *debugnv)
 899 {
 900         zfs_cmd_t zc = { 0 };
 901         libzfs_handle_t *hdl = zhp->zfs_hdl;
 902         nvlist_t *thisdbg;
 903 
 904         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 905         assert(fromsnap_obj == 0 || !fromorigin);
 906 
 907         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 908         zc.zc_cookie = outfd;
 909         zc.zc_obj = fromorigin;
 910         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 911         zc.zc_fromobj = fromsnap_obj;
 912         zc.zc_flags = flags;
 913 
 914         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 915         if (fromsnap && fromsnap[0] != '\0') {
 916                 VERIFY(0 == nvlist_add_string(thisdbg,
 917                     "fromsnap", fromsnap));
 918         }
 919 
 920         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 921                 char errbuf[1024];
 922                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 923                     "warning: cannot send '%s'"), zhp->zfs_name);
 924 
 925                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
 926                 if (debugnv) {
 927                         VERIFY(0 == nvlist_add_nvlist(debugnv,
 928                             zhp->zfs_name, thisdbg));
 929                 }
 930                 nvlist_free(thisdbg);
 931 
 932                 switch (errno) {
 933                 case EXDEV:
 934                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 935                             "not an earlier snapshot from the same fs"));
 936                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 937 
 938                 case ENOENT:
 939                         if (zfs_dataset_exists(hdl, zc.zc_name,
 940                             ZFS_TYPE_SNAPSHOT)) {
 941                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 942                                     "incremental source (@%s) does not exist"),
 943                                     zc.zc_value);
 944                         }
 945                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 946 
 947                 case EDQUOT:
 948                 case EFBIG:
 949                 case EIO:
 950                 case ENOLINK:
 951                 case ENOSPC:
 952                 case ENOSTR:
 953                 case ENXIO:
 954                 case EPIPE:
 955                 case ERANGE:
 956                 case EFAULT:
 957                 case EROFS:
 958                         zfs_error_aux(hdl, strerror(errno));
 959                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 960 
 961                 default:
 962                         return (zfs_standard_error(hdl, errno, errbuf));
 963                 }
 964         }
 965 
 966         if (debugnv)
 967                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
 968         nvlist_free(thisdbg);
 969 
 970         return (0);
 971 }
 972 
 973 static void
 974 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
 975 {
 976         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 977 
 978         /*
 979          * zfs_send() only sets snapholds for sends that need them,
 980          * e.g. replication and doall.
 981          */
 982         if (sdd->snapholds == NULL)
 983                 return;
 984 
 985         fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
 986 }
 987 
 988 static void *
 989 send_progress_thread(void *arg)
 990 {
 991         progress_arg_t *pa = arg;
 992         zfs_cmd_t zc = { 0 };
 993         zfs_handle_t *zhp = pa->pa_zhp;
 994         libzfs_handle_t *hdl = zhp->zfs_hdl;
 995         unsigned long long bytes;
 996         char buf[16];
 997         time_t t;
 998         struct tm *tm;
 999 
1000         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1001 
1002         if (!pa->pa_parsable)
1003                 (void) fprintf(stderr, "TIME        SENT   SNAPSHOT\n");
1004 
1005         /*
1006          * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1007          */
1008         for (;;) {
1009                 (void) sleep(1);
1010 
1011                 zc.zc_cookie = pa->pa_fd;
1012                 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1013                         return ((void *)-1);
1014 
1015                 (void) time(&t);
1016                 tm = localtime(&t);
1017                 bytes = zc.zc_cookie;
1018 
1019                 if (pa->pa_parsable) {
1020                         (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1021                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1022                             bytes, zhp->zfs_name);
1023                 } else {
1024                         zfs_nicenum(bytes, buf, sizeof (buf));
1025                         (void) fprintf(stderr, "%02d:%02d:%02d   %5s   %s\n",
1026                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1027                             buf, zhp->zfs_name);
1028                 }
1029         }
1030 }
1031 
1032 static void
1033 send_print_verbose(FILE *fout, const char *tosnap, const char *fromsnap,
1034     uint64_t size, boolean_t parsable)
1035 {
1036         if (parsable) {
1037                 if (fromsnap != NULL) {
1038                         (void) fprintf(fout, "incremental\t%s\t%s",
1039                             fromsnap, tosnap);
1040                 } else {
1041                         (void) fprintf(fout, "full\t%s",
1042                             tosnap);
1043                 }
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"),
1050                                     fromsnap, tosnap);
1051                         } else {
1052                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1053                                     "send from %s to %s"),
1054                                     fromsnap, tosnap);
1055                         }
1056                 } else {
1057                         (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1058                             "full send of %s"),
1059                             tosnap);
1060                 }
1061         }
1062 
1063         if (size != 0) {
1064                 if (parsable) {
1065                         (void) fprintf(fout, "\t%llu",
1066                             (longlong_t)size);
1067                 } else {
1068                         char buf[16];
1069                         zfs_nicenum(size, buf, sizeof (buf));
1070                         (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1071                             " estimated size is %s"), buf);
1072                 }
1073         }
1074         (void) fprintf(fout, "\n");
1075 }
1076 
1077 static int
1078 dump_snapshot(zfs_handle_t *zhp, void *arg)
1079 {
1080         send_dump_data_t *sdd = arg;
1081         progress_arg_t pa = { 0 };
1082         pthread_t tid;
1083         char *thissnap;
1084         int err;
1085         boolean_t isfromsnap, istosnap, fromorigin;
1086         boolean_t exclude = B_FALSE;
1087         FILE *fout = sdd->std_out ? stdout : stderr;
1088 
1089         err = 0;
1090         thissnap = strchr(zhp->zfs_name, '@') + 1;
1091         isfromsnap = (sdd->fromsnap != NULL &&
1092             strcmp(sdd->fromsnap, thissnap) == 0);
1093 
1094         if (!sdd->seenfrom && isfromsnap) {
1095                 gather_holds(zhp, sdd);
1096                 sdd->seenfrom = B_TRUE;
1097                 (void) strcpy(sdd->prevsnap, thissnap);
1098                 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1099                 zfs_close(zhp);
1100                 return (0);
1101         }
1102 
1103         if (sdd->seento || !sdd->seenfrom) {
1104                 zfs_close(zhp);
1105                 return (0);
1106         }
1107 
1108         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1109         if (istosnap)
1110                 sdd->seento = B_TRUE;
1111 
1112         if (!sdd->doall && !isfromsnap && !istosnap) {
1113                 if (sdd->replicate) {
1114                         char *snapname;
1115                         nvlist_t *snapprops;
1116                         /*
1117                          * Filter out all intermediate snapshots except origin
1118                          * snapshots needed to replicate clones.
1119                          */
1120                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1121                             zhp->zfs_dmustats.dds_guid, &snapname);
1122 
1123                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1124                             "snapprops", &snapprops));
1125                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1126                             thissnap, &snapprops));
1127                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
1128                 } else {
1129                         exclude = B_TRUE;
1130                 }
1131         }
1132 
1133         /*
1134          * If a filter function exists, call it to determine whether
1135          * this snapshot will be sent.
1136          */
1137         if (exclude || (sdd->filter_cb != NULL &&
1138             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1139                 /*
1140                  * This snapshot is filtered out.  Don't send it, and don't
1141                  * set prevsnap_obj, so it will be as if this snapshot didn't
1142                  * exist, and the next accepted snapshot will be sent as
1143                  * an incremental from the last accepted one, or as the
1144                  * first (and full) snapshot in the case of a replication,
1145                  * non-incremental send.
1146                  */
1147                 zfs_close(zhp);
1148                 return (0);
1149         }
1150 
1151         gather_holds(zhp, sdd);
1152         fromorigin = sdd->prevsnap[0] == '\0' &&
1153             (sdd->fromorigin || sdd->replicate);
1154 
1155         if (sdd->verbose) {
1156                 uint64_t size = 0;
1157                 (void) estimate_ioctl(zhp, sdd->prevsnap_obj,
1158                     fromorigin, &size);
1159 
1160                 send_print_verbose(fout, zhp->zfs_name,
1161                     sdd->prevsnap[0] ? sdd->prevsnap : NULL,
1162                     size, sdd->parsable);
1163                 sdd->size += size;
1164         }
1165 
1166         if (!sdd->dryrun) {
1167                 /*
1168                  * If progress reporting is requested, spawn a new thread to
1169                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1170                  */
1171                 if (sdd->progress) {
1172                         pa.pa_zhp = zhp;
1173                         pa.pa_fd = sdd->outfd;
1174                         pa.pa_parsable = sdd->parsable;
1175 
1176                         if (err = pthread_create(&tid, NULL,
1177                             send_progress_thread, &pa)) {
1178                                 zfs_close(zhp);
1179                                 return (err);
1180                         }
1181                 }
1182 
1183                 enum lzc_send_flags flags = 0;
1184                 if (sdd->large_block)
1185                         flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1186                 if (sdd->embed_data)
1187                         flags |= LZC_SEND_FLAG_EMBED_DATA;
1188 
1189                 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1190                     fromorigin, sdd->outfd, flags, sdd->debugnv);
1191 
1192                 if (sdd->progress) {
1193                         (void) pthread_cancel(tid);
1194                         (void) pthread_join(tid, NULL);
1195                 }
1196         }
1197 
1198         (void) strcpy(sdd->prevsnap, thissnap);
1199         sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1200         zfs_close(zhp);
1201         return (err);
1202 }
1203 
1204 static int
1205 dump_filesystem(zfs_handle_t *zhp, void *arg)
1206 {
1207         int rv = 0;
1208         send_dump_data_t *sdd = arg;
1209         boolean_t missingfrom = B_FALSE;
1210         zfs_cmd_t zc = { 0 };
1211 
1212         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1213             zhp->zfs_name, sdd->tosnap);
1214         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1215                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1216                     "WARNING: could not send %s@%s: does not exist\n"),
1217                     zhp->zfs_name, sdd->tosnap);
1218                 sdd->err = B_TRUE;
1219                 return (0);
1220         }
1221 
1222         if (sdd->replicate && sdd->fromsnap) {
1223                 /*
1224                  * If this fs does not have fromsnap, and we're doing
1225                  * recursive, we need to send a full stream from the
1226                  * beginning (or an incremental from the origin if this
1227                  * is a clone).  If we're doing non-recursive, then let
1228                  * them get the error.
1229                  */
1230                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1231                     zhp->zfs_name, sdd->fromsnap);
1232                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1233                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1234                         missingfrom = B_TRUE;
1235                 }
1236         }
1237 
1238         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1239         sdd->prevsnap_obj = 0;
1240         if (sdd->fromsnap == NULL || missingfrom)
1241                 sdd->seenfrom = B_TRUE;
1242 
1243         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1244         if (!sdd->seenfrom) {
1245                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1246                     "WARNING: could not send %s@%s:\n"
1247                     "incremental source (%s@%s) does not exist\n"),
1248                     zhp->zfs_name, sdd->tosnap,
1249                     zhp->zfs_name, sdd->fromsnap);
1250                 sdd->err = B_TRUE;
1251         } else if (!sdd->seento) {
1252                 if (sdd->fromsnap) {
1253                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1254                             "WARNING: could not send %s@%s:\n"
1255                             "incremental source (%s@%s) "
1256                             "is not earlier than it\n"),
1257                             zhp->zfs_name, sdd->tosnap,
1258                             zhp->zfs_name, sdd->fromsnap);
1259                 } else {
1260                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1261                             "WARNING: "
1262                             "could not send %s@%s: does not exist\n"),
1263                             zhp->zfs_name, sdd->tosnap);
1264                 }
1265                 sdd->err = B_TRUE;
1266         }
1267 
1268         return (rv);
1269 }
1270 
1271 static int
1272 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1273 {
1274         send_dump_data_t *sdd = arg;
1275         nvpair_t *fspair;
1276         boolean_t needagain, progress;
1277 
1278         if (!sdd->replicate)
1279                 return (dump_filesystem(rzhp, sdd));
1280 
1281         /* Mark the clone origin snapshots. */
1282         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1283             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1284                 nvlist_t *nvfs;
1285                 uint64_t origin_guid = 0;
1286 
1287                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1288                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1289                 if (origin_guid != 0) {
1290                         char *snapname;
1291                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1292                             origin_guid, &snapname);
1293                         if (origin_nv != NULL) {
1294                                 nvlist_t *snapprops;
1295                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1296                                     "snapprops", &snapprops));
1297                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1298                                     snapname, &snapprops));
1299                                 VERIFY(0 == nvlist_add_boolean(
1300                                     snapprops, "is_clone_origin"));
1301                         }
1302                 }
1303         }
1304 again:
1305         needagain = progress = B_FALSE;
1306         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1307             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1308                 nvlist_t *fslist, *parent_nv;
1309                 char *fsname;
1310                 zfs_handle_t *zhp;
1311                 int err;
1312                 uint64_t origin_guid = 0;
1313                 uint64_t parent_guid = 0;
1314 
1315                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1316                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1317                         continue;
1318 
1319                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1320                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1321                 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1322                     &parent_guid);
1323 
1324                 if (parent_guid != 0) {
1325                         parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1326                         if (!nvlist_exists(parent_nv, "sent")) {
1327                                 /* parent has not been sent; skip this one */
1328                                 needagain = B_TRUE;
1329                                 continue;
1330                         }
1331                 }
1332 
1333                 if (origin_guid != 0) {
1334                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1335                             origin_guid, NULL);
1336                         if (origin_nv != NULL &&
1337                             !nvlist_exists(origin_nv, "sent")) {
1338                                 /*
1339                                  * origin has not been sent yet;
1340                                  * skip this clone.
1341                                  */
1342                                 needagain = B_TRUE;
1343                                 continue;
1344                         }
1345                 }
1346 
1347                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1348                 if (zhp == NULL)
1349                         return (-1);
1350                 err = dump_filesystem(zhp, sdd);
1351                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1352                 progress = B_TRUE;
1353                 zfs_close(zhp);
1354                 if (err)
1355                         return (err);
1356         }
1357         if (needagain) {
1358                 assert(progress);
1359                 goto again;
1360         }
1361 
1362         /* clean out the sent flags in case we reuse this fss */
1363         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1364             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1365                 nvlist_t *fslist;
1366 
1367                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1368                 (void) nvlist_remove_all(fslist, "sent");
1369         }
1370 
1371         return (0);
1372 }
1373 
1374 nvlist_t *
1375 zfs_send_resume_token_to_nvlist(libzfs_handle_t *hdl, const char *token)
1376 {
1377         unsigned int version;
1378         int nread;
1379         unsigned long long checksum, packed_len;
1380 
1381         /*
1382          * Decode token header, which is:
1383          *   <token version>-<checksum of payload>-<uncompressed payload length>
1384          * Note that the only supported token version is 1.
1385          */
1386         nread = sscanf(token, "%u-%llx-%llx-",
1387             &version, &checksum, &packed_len);
1388         if (nread != 3) {
1389                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1390                     "resume token is corrupt (invalid format)"));
1391                 return (NULL);
1392         }
1393 
1394         if (version != ZFS_SEND_RESUME_TOKEN_VERSION) {
1395                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1396                     "resume token is corrupt (invalid version %u)"),
1397                     version);
1398                 return (NULL);
1399         }
1400 
1401         /* convert hexadecimal representation to binary */
1402         token = strrchr(token, '-') + 1;
1403         int len = strlen(token) / 2;
1404         unsigned char *compressed = zfs_alloc(hdl, len);
1405         for (int i = 0; i < len; i++) {
1406                 nread = sscanf(token + i * 2, "%2hhx", compressed + i);
1407                 if (nread != 1) {
1408                         free(compressed);
1409                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1410                             "resume token is corrupt "
1411                             "(payload is not hex-encoded)"));
1412                         return (NULL);
1413                 }
1414         }
1415 
1416         /* verify checksum */
1417         zio_cksum_t cksum;
1418         fletcher_4_native(compressed, len, NULL, &cksum);
1419         if (cksum.zc_word[0] != checksum) {
1420                 free(compressed);
1421                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1422                     "resume token is corrupt (incorrect checksum)"));
1423                 return (NULL);
1424         }
1425 
1426         /* uncompress */
1427         void *packed = zfs_alloc(hdl, packed_len);
1428         uLongf packed_len_long = packed_len;
1429         if (uncompress(packed, &packed_len_long, compressed, len) != Z_OK ||
1430             packed_len_long != packed_len) {
1431                 free(packed);
1432                 free(compressed);
1433                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1434                     "resume token is corrupt (decompression failed)"));
1435                 return (NULL);
1436         }
1437 
1438         /* unpack nvlist */
1439         nvlist_t *nv;
1440         int error = nvlist_unpack(packed, packed_len, &nv, KM_SLEEP);
1441         free(packed);
1442         free(compressed);
1443         if (error != 0) {
1444                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1445                     "resume token is corrupt (nvlist_unpack failed)"));
1446                 return (NULL);
1447         }
1448         return (nv);
1449 }
1450 
1451 int
1452 zfs_send_resume(libzfs_handle_t *hdl, sendflags_t *flags, int outfd,
1453     const char *resume_token)
1454 {
1455         char errbuf[1024];
1456         char *toname;
1457         char *fromname = NULL;
1458         uint64_t resumeobj, resumeoff, toguid, fromguid, bytes;
1459         zfs_handle_t *zhp;
1460         int error = 0;
1461         char name[ZFS_MAXNAMELEN];
1462         enum lzc_send_flags lzc_flags = 0;
1463 
1464         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1465             "cannot resume send"));
1466 
1467         nvlist_t *resume_nvl =
1468             zfs_send_resume_token_to_nvlist(hdl, resume_token);
1469         if (resume_nvl == NULL) {
1470                 /*
1471                  * zfs_error_aux has already been set by
1472                  * zfs_send_resume_token_to_nvlist
1473                  */
1474                 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1475         }
1476         if (flags->verbose) {
1477                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1478                     "resume token contents:\n"));
1479                 nvlist_print(stderr, resume_nvl);
1480         }
1481 
1482         if (nvlist_lookup_string(resume_nvl, "toname", &toname) != 0 ||
1483             nvlist_lookup_uint64(resume_nvl, "object", &resumeobj) != 0 ||
1484             nvlist_lookup_uint64(resume_nvl, "offset", &resumeoff) != 0 ||
1485             nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 ||
1486             nvlist_lookup_uint64(resume_nvl, "toguid", &toguid) != 0) {
1487                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1488                     "resume token is corrupt"));
1489                 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1490         }
1491         fromguid = 0;
1492         (void) nvlist_lookup_uint64(resume_nvl, "fromguid", &fromguid);
1493 
1494         if (flags->embed_data || nvlist_exists(resume_nvl, "embedok"))
1495                 lzc_flags |= LZC_SEND_FLAG_EMBED_DATA;
1496 
1497         if (guid_to_name(hdl, toname, toguid, B_FALSE, name) != 0) {
1498                 if (zfs_dataset_exists(hdl, toname, ZFS_TYPE_DATASET)) {
1499                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1500                             "'%s' is no longer the same snapshot used in "
1501                             "the initial send"), toname);
1502                 } else {
1503                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1504                             "'%s' used in the initial send no longer exists"),
1505                             toname);
1506                 }
1507                 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1508         }
1509         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1510         if (zhp == NULL) {
1511                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1512                     "unable to access '%s'"), name);
1513                 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1514         }
1515 
1516         if (fromguid != 0) {
1517                 if (guid_to_name(hdl, toname, fromguid, B_TRUE, name) != 0) {
1518                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1519                             "incremental source %#llx no longer exists"),
1520                             (longlong_t)fromguid);
1521                         return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1522                 }
1523                 fromname = name;
1524         }
1525 
1526         if (flags->verbose) {
1527                 uint64_t size = 0;
1528                 error = lzc_send_space(zhp->zfs_name, fromname, &size);
1529                 if (error == 0)
1530                         size = MAX(0, (int64_t)(size - bytes));
1531                 send_print_verbose(stderr, zhp->zfs_name, fromname,
1532                     size, flags->parsable);
1533         }
1534 
1535         if (!flags->dryrun) {
1536                 progress_arg_t pa = { 0 };
1537                 pthread_t tid;
1538                 /*
1539                  * If progress reporting is requested, spawn a new thread to
1540                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1541                  */
1542                 if (flags->progress) {
1543                         pa.pa_zhp = zhp;
1544                         pa.pa_fd = outfd;
1545                         pa.pa_parsable = flags->parsable;
1546 
1547                         error = pthread_create(&tid, NULL,
1548                             send_progress_thread, &pa);
1549                         if (error != 0) {
1550                                 zfs_close(zhp);
1551                                 return (error);
1552                         }
1553                 }
1554 
1555                 error = lzc_send_resume(zhp->zfs_name, fromname, outfd,
1556                     lzc_flags, resumeobj, resumeoff);
1557 
1558                 if (flags->progress) {
1559                         (void) pthread_cancel(tid);
1560                         (void) pthread_join(tid, NULL);
1561                 }
1562 
1563                 char errbuf[1024];
1564                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1565                     "warning: cannot send '%s'"), zhp->zfs_name);
1566 
1567                 zfs_close(zhp);
1568 
1569                 switch (error) {
1570                 case 0:
1571                         return (0);
1572                 case EXDEV:
1573                 case ENOENT:
1574                 case EDQUOT:
1575                 case EFBIG:
1576                 case EIO:
1577                 case ENOLINK:
1578                 case ENOSPC:
1579                 case ENOSTR:
1580                 case ENXIO:
1581                 case EPIPE:
1582                 case ERANGE:
1583                 case EFAULT:
1584                 case EROFS:
1585                         zfs_error_aux(hdl, strerror(errno));
1586                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1587 
1588                 default:
1589                         return (zfs_standard_error(hdl, errno, errbuf));
1590                 }
1591         }
1592 
1593 
1594         zfs_close(zhp);
1595 
1596         return (error);
1597 }
1598 
1599 /*
1600  * Generate a send stream for the dataset identified by the argument zhp.
1601  *
1602  * The content of the send stream is the snapshot identified by
1603  * 'tosnap'.  Incremental streams are requested in two ways:
1604  *     - from the snapshot identified by "fromsnap" (if non-null) or
1605  *     - from the origin of the dataset identified by zhp, which must
1606  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1607  *       is TRUE.
1608  *
1609  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1610  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1611  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1612  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1613  * case too. If "props" is set, send properties.
1614  */
1615 int
1616 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1617     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1618     void *cb_arg, nvlist_t **debugnvp)
1619 {
1620         char errbuf[1024];
1621         send_dump_data_t sdd = { 0 };
1622         int err = 0;
1623         nvlist_t *fss = NULL;
1624         avl_tree_t *fsavl = NULL;
1625         static uint64_t holdseq;
1626         int spa_version;
1627         pthread_t tid = 0;
1628         int pipefd[2];
1629         dedup_arg_t dda = { 0 };
1630         int featureflags = 0;
1631         FILE *fout;
1632 
1633         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1634             "cannot send '%s'"), zhp->zfs_name);
1635 
1636         if (fromsnap && fromsnap[0] == '\0') {
1637                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1638                     "zero-length incremental source"));
1639                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1640         }
1641 
1642         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1643                 uint64_t version;
1644                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1645                 if (version >= ZPL_VERSION_SA) {
1646                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1647                 }
1648         }
1649 
1650         if (flags->dedup && !flags->dryrun) {
1651                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1652                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1653                 if (err = pipe(pipefd)) {
1654                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1655                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1656                             errbuf));
1657                 }
1658                 dda.outputfd = outfd;
1659                 dda.inputfd = pipefd[1];
1660                 dda.dedup_hdl = zhp->zfs_hdl;
1661                 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1662                         (void) close(pipefd[0]);
1663                         (void) close(pipefd[1]);
1664                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1665                         return (zfs_error(zhp->zfs_hdl,
1666                             EZFS_THREADCREATEFAILED, errbuf));
1667                 }
1668         }
1669 
1670         if (flags->replicate || flags->doall || flags->props) {
1671                 dmu_replay_record_t drr = { 0 };
1672                 char *packbuf = NULL;
1673                 size_t buflen = 0;
1674                 zio_cksum_t zc = { 0 };
1675 
1676                 if (flags->replicate || flags->props) {
1677                         nvlist_t *hdrnv;
1678 
1679                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1680                         if (fromsnap) {
1681                                 VERIFY(0 == nvlist_add_string(hdrnv,
1682                                     "fromsnap", fromsnap));
1683                         }
1684                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1685                         if (!flags->replicate) {
1686                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1687                                     "not_recursive"));
1688                         }
1689 
1690                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1691                             fromsnap, tosnap, flags->replicate, &fss, &fsavl);
1692                         if (err)
1693                                 goto err_out;
1694                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1695                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1696                             NV_ENCODE_XDR, 0);
1697                         if (debugnvp)
1698                                 *debugnvp = hdrnv;
1699                         else
1700                                 nvlist_free(hdrnv);
1701                         if (err)
1702                                 goto stderr_out;
1703                 }
1704 
1705                 if (!flags->dryrun) {
1706                         /* write first begin record */
1707                         drr.drr_type = DRR_BEGIN;
1708                         drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1709                         DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1710                             drr_versioninfo, DMU_COMPOUNDSTREAM);
1711                         DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1712                             drr_versioninfo, featureflags);
1713                         (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1714                             sizeof (drr.drr_u.drr_begin.drr_toname),
1715                             "%s@%s", zhp->zfs_name, tosnap);
1716                         drr.drr_payloadlen = buflen;
1717 
1718                         err = dump_record(&drr, packbuf, buflen, &zc, outfd);
1719                         free(packbuf);
1720                         if (err != 0)
1721                                 goto stderr_out;
1722 
1723                         /* write end record */
1724                         bzero(&drr, sizeof (drr));
1725                         drr.drr_type = DRR_END;
1726                         drr.drr_u.drr_end.drr_checksum = zc;
1727                         err = write(outfd, &drr, sizeof (drr));
1728                         if (err == -1) {
1729                                 err = errno;
1730                                 goto stderr_out;
1731                         }
1732 
1733                         err = 0;
1734                 }
1735         }
1736 
1737         /* dump each stream */
1738         sdd.fromsnap = fromsnap;
1739         sdd.tosnap = tosnap;
1740         if (tid != 0)
1741                 sdd.outfd = pipefd[0];
1742         else
1743                 sdd.outfd = outfd;
1744         sdd.replicate = flags->replicate;
1745         sdd.doall = flags->doall;
1746         sdd.fromorigin = flags->fromorigin;
1747         sdd.fss = fss;
1748         sdd.fsavl = fsavl;
1749         sdd.verbose = flags->verbose;
1750         sdd.parsable = flags->parsable;
1751         sdd.progress = flags->progress;
1752         sdd.dryrun = flags->dryrun;
1753         sdd.large_block = flags->largeblock;
1754         sdd.embed_data = flags->embed_data;
1755         sdd.filter_cb = filter_func;
1756         sdd.filter_cb_arg = cb_arg;
1757         if (debugnvp)
1758                 sdd.debugnv = *debugnvp;
1759         if (sdd.verbose && sdd.dryrun)
1760                 sdd.std_out = B_TRUE;
1761         fout = sdd.std_out ? stdout : stderr;
1762 
1763         /*
1764          * Some flags require that we place user holds on the datasets that are
1765          * being sent so they don't get destroyed during the send. We can skip
1766          * this step if the pool is imported read-only since the datasets cannot
1767          * be destroyed.
1768          */
1769         if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1770             ZPOOL_PROP_READONLY, NULL) &&
1771             zfs_spa_version(zhp, &spa_version) == 0 &&
1772             spa_version >= SPA_VERSION_USERREFS &&
1773             (flags->doall || flags->replicate)) {
1774                 ++holdseq;
1775                 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1776                     ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1777                 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1778                 if (sdd.cleanup_fd < 0) {
1779                         err = errno;
1780                         goto stderr_out;
1781                 }
1782                 sdd.snapholds = fnvlist_alloc();
1783         } else {
1784                 sdd.cleanup_fd = -1;
1785                 sdd.snapholds = NULL;
1786         }
1787         if (flags->verbose || sdd.snapholds != NULL) {
1788                 /*
1789                  * Do a verbose no-op dry run to get all the verbose output
1790                  * or to gather snapshot hold's before generating any data,
1791                  * then do a non-verbose real run to generate the streams.
1792                  */
1793                 sdd.dryrun = B_TRUE;
1794                 err = dump_filesystems(zhp, &sdd);
1795 
1796                 if (err != 0)
1797                         goto stderr_out;
1798 
1799                 if (flags->verbose) {
1800                         if (flags->parsable) {
1801                                 (void) fprintf(fout, "size\t%llu\n",
1802                                     (longlong_t)sdd.size);
1803                         } else {
1804                                 char buf[16];
1805                                 zfs_nicenum(sdd.size, buf, sizeof (buf));
1806                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1807                                     "total estimated size is %s\n"), buf);
1808                         }
1809                 }
1810 
1811                 /* Ensure no snaps found is treated as an error. */
1812                 if (!sdd.seento) {
1813                         err = ENOENT;
1814                         goto err_out;
1815                 }
1816 
1817                 /* Skip the second run if dryrun was requested. */
1818                 if (flags->dryrun)
1819                         goto err_out;
1820 
1821                 if (sdd.snapholds != NULL) {
1822                         err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
1823                         if (err != 0)
1824                                 goto stderr_out;
1825 
1826                         fnvlist_free(sdd.snapholds);
1827                         sdd.snapholds = NULL;
1828                 }
1829 
1830                 sdd.dryrun = B_FALSE;
1831                 sdd.verbose = B_FALSE;
1832         }
1833 
1834         err = dump_filesystems(zhp, &sdd);
1835         fsavl_destroy(fsavl);
1836         nvlist_free(fss);
1837 
1838         /* Ensure no snaps found is treated as an error. */
1839         if (err == 0 && !sdd.seento)
1840                 err = ENOENT;
1841 
1842         if (tid != 0) {
1843                 if (err != 0)
1844                         (void) pthread_cancel(tid);
1845                 (void) close(pipefd[0]);
1846                 (void) pthread_join(tid, NULL);
1847         }
1848 
1849         if (sdd.cleanup_fd != -1) {
1850                 VERIFY(0 == close(sdd.cleanup_fd));
1851                 sdd.cleanup_fd = -1;
1852         }
1853 
1854         if (!flags->dryrun && (flags->replicate || flags->doall ||
1855             flags->props)) {
1856                 /*
1857                  * write final end record.  NB: want to do this even if
1858                  * there was some error, because it might not be totally
1859                  * failed.
1860                  */
1861                 dmu_replay_record_t drr = { 0 };
1862                 drr.drr_type = DRR_END;
1863                 if (write(outfd, &drr, sizeof (drr)) == -1) {
1864                         return (zfs_standard_error(zhp->zfs_hdl,
1865                             errno, errbuf));
1866                 }
1867         }
1868 
1869         return (err || sdd.err);
1870 
1871 stderr_out:
1872         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1873 err_out:
1874         fsavl_destroy(fsavl);
1875         nvlist_free(fss);
1876         fnvlist_free(sdd.snapholds);
1877 
1878         if (sdd.cleanup_fd != -1)
1879                 VERIFY(0 == close(sdd.cleanup_fd));
1880         if (tid != 0) {
1881                 (void) pthread_cancel(tid);
1882                 (void) close(pipefd[0]);
1883                 (void) pthread_join(tid, NULL);
1884         }
1885         return (err);
1886 }
1887 
1888 int
1889 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd,
1890     enum lzc_send_flags flags)
1891 {
1892         int err;
1893         libzfs_handle_t *hdl = zhp->zfs_hdl;
1894 
1895         char errbuf[1024];
1896         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1897             "warning: cannot send '%s'"), zhp->zfs_name);
1898 
1899         err = lzc_send(zhp->zfs_name, from, fd, flags);
1900         if (err != 0) {
1901                 switch (errno) {
1902                 case EXDEV:
1903                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1904                             "not an earlier snapshot from the same fs"));
1905                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
1906 
1907                 case ENOENT:
1908                 case ESRCH:
1909                         if (lzc_exists(zhp->zfs_name)) {
1910                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1911                                     "incremental source (%s) does not exist"),
1912                                     from);
1913                         }
1914                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
1915 
1916                 case EBUSY:
1917                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1918                             "target is busy; if a filesystem, "
1919                             "it must not be mounted"));
1920                         return (zfs_error(hdl, EZFS_BUSY, errbuf));
1921 
1922                 case EDQUOT:
1923                 case EFBIG:
1924                 case EIO:
1925                 case ENOLINK:
1926                 case ENOSPC:
1927                 case ENOSTR:
1928                 case ENXIO:
1929                 case EPIPE:
1930                 case ERANGE:
1931                 case EFAULT:
1932                 case EROFS:
1933                         zfs_error_aux(hdl, strerror(errno));
1934                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1935 
1936                 default:
1937                         return (zfs_standard_error(hdl, errno, errbuf));
1938                 }
1939         }
1940         return (err != 0);
1941 }
1942 
1943 /*
1944  * Routines specific to "zfs recv"
1945  */
1946 
1947 static int
1948 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1949     boolean_t byteswap, zio_cksum_t *zc)
1950 {
1951         char *cp = buf;
1952         int rv;
1953         int len = ilen;
1954 
1955         assert(ilen <= SPA_MAXBLOCKSIZE);
1956 
1957         do {
1958                 rv = read(fd, cp, len);
1959                 cp += rv;
1960                 len -= rv;
1961         } while (rv > 0);
1962 
1963         if (rv < 0 || len != 0) {
1964                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1965                     "failed to read from stream"));
1966                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1967                     "cannot receive")));
1968         }
1969 
1970         if (zc) {
1971                 if (byteswap)
1972                         fletcher_4_incremental_byteswap(buf, ilen, zc);
1973                 else
1974                         fletcher_4_incremental_native(buf, ilen, zc);
1975         }
1976         return (0);
1977 }
1978 
1979 static int
1980 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1981     boolean_t byteswap, zio_cksum_t *zc)
1982 {
1983         char *buf;
1984         int err;
1985 
1986         buf = zfs_alloc(hdl, len);
1987         if (buf == NULL)
1988                 return (ENOMEM);
1989 
1990         err = recv_read(hdl, fd, buf, len, byteswap, zc);
1991         if (err != 0) {
1992                 free(buf);
1993                 return (err);
1994         }
1995 
1996         err = nvlist_unpack(buf, len, nvp, 0);
1997         free(buf);
1998         if (err != 0) {
1999                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2000                     "stream (malformed nvlist)"));
2001                 return (EINVAL);
2002         }
2003         return (0);
2004 }
2005 
2006 static int
2007 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
2008     int baselen, char *newname, recvflags_t *flags)
2009 {
2010         static int seq;
2011         zfs_cmd_t zc = { 0 };
2012         int err;
2013         prop_changelist_t *clp;
2014         zfs_handle_t *zhp;
2015 
2016         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2017         if (zhp == NULL)
2018                 return (-1);
2019         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2020             flags->force ? MS_FORCE : 0);
2021         zfs_close(zhp);
2022         if (clp == NULL)
2023                 return (-1);
2024         err = changelist_prefix(clp);
2025         if (err)
2026                 return (err);
2027 
2028         zc.zc_objset_type = DMU_OST_ZFS;
2029         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
2030 
2031         if (tryname) {
2032                 (void) strcpy(newname, tryname);
2033 
2034                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
2035 
2036                 if (flags->verbose) {
2037                         (void) printf("attempting rename %s to %s\n",
2038                             zc.zc_name, zc.zc_value);
2039                 }
2040                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
2041                 if (err == 0)
2042                         changelist_rename(clp, name, tryname);
2043         } else {
2044                 err = ENOENT;
2045         }
2046 
2047         if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
2048                 seq++;
2049 
2050                 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u",
2051                     baselen, name, getpid(), seq);
2052                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
2053 
2054                 if (flags->verbose) {
2055                         (void) printf("failed - trying rename %s to %s\n",
2056                             zc.zc_name, zc.zc_value);
2057                 }
2058                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
2059                 if (err == 0)
2060                         changelist_rename(clp, name, newname);
2061                 if (err && flags->verbose) {
2062                         (void) printf("failed (%u) - "
2063                             "will try again on next pass\n", errno);
2064                 }
2065                 err = EAGAIN;
2066         } else if (flags->verbose) {
2067                 if (err == 0)
2068                         (void) printf("success\n");
2069                 else
2070                         (void) printf("failed (%u)\n", errno);
2071         }
2072 
2073         (void) changelist_postfix(clp);
2074         changelist_free(clp);
2075 
2076         return (err);
2077 }
2078 
2079 static int
2080 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
2081     char *newname, recvflags_t *flags)
2082 {
2083         zfs_cmd_t zc = { 0 };
2084         int err = 0;
2085         prop_changelist_t *clp;
2086         zfs_handle_t *zhp;
2087         boolean_t defer = B_FALSE;
2088         int spa_version;
2089 
2090         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2091         if (zhp == NULL)
2092                 return (-1);
2093         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2094             flags->force ? MS_FORCE : 0);
2095         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
2096             zfs_spa_version(zhp, &spa_version) == 0 &&
2097             spa_version >= SPA_VERSION_USERREFS)
2098                 defer = B_TRUE;
2099         zfs_close(zhp);
2100         if (clp == NULL)
2101                 return (-1);
2102         err = changelist_prefix(clp);
2103         if (err)
2104                 return (err);
2105 
2106         zc.zc_objset_type = DMU_OST_ZFS;
2107         zc.zc_defer_destroy = defer;
2108         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
2109 
2110         if (flags->verbose)
2111                 (void) printf("attempting destroy %s\n", zc.zc_name);
2112         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
2113         if (err == 0) {
2114                 if (flags->verbose)
2115                         (void) printf("success\n");
2116                 changelist_remove(clp, zc.zc_name);
2117         }
2118 
2119         (void) changelist_postfix(clp);
2120         changelist_free(clp);
2121 
2122         /*
2123          * Deferred destroy might destroy the snapshot or only mark it to be
2124          * destroyed later, and it returns success in either case.
2125          */
2126         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
2127             ZFS_TYPE_SNAPSHOT))) {
2128                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
2129         }
2130 
2131         return (err);
2132 }
2133 
2134 typedef struct guid_to_name_data {
2135         uint64_t guid;
2136         boolean_t bookmark_ok;
2137         char *name;
2138         char *skip;
2139 } guid_to_name_data_t;
2140 
2141 static int
2142 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
2143 {
2144         guid_to_name_data_t *gtnd = arg;
2145         const char *slash;
2146         int err;
2147 
2148         if (gtnd->skip != NULL &&
2149             (slash = strrchr(zhp->zfs_name, '/')) != NULL &&
2150             strcmp(slash + 1, gtnd->skip) == 0) {
2151                 zfs_close(zhp);
2152                 return (0);
2153         }
2154 
2155         if (zfs_prop_get_int(zhp, ZFS_PROP_GUID) == gtnd->guid) {
2156                 (void) strcpy(gtnd->name, zhp->zfs_name);
2157                 zfs_close(zhp);
2158                 return (EEXIST);
2159         }
2160 
2161         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
2162         if (err != EEXIST && gtnd->bookmark_ok)
2163                 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, gtnd);
2164         zfs_close(zhp);
2165         return (err);
2166 }
2167 
2168 /*
2169  * Attempt to find the local dataset associated with this guid.  In the case of
2170  * multiple matches, we attempt to find the "best" match by searching
2171  * progressively larger portions of the hierarchy.  This allows one to send a
2172  * tree of datasets individually and guarantee that we will find the source
2173  * guid within that hierarchy, even if there are multiple matches elsewhere.
2174  */
2175 static int
2176 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
2177     boolean_t bookmark_ok, char *name)
2178 {
2179         char pname[ZFS_MAXNAMELEN];
2180         guid_to_name_data_t gtnd;
2181 
2182         gtnd.guid = guid;
2183         gtnd.bookmark_ok = bookmark_ok;
2184         gtnd.name = name;
2185         gtnd.skip = NULL;
2186 
2187         /*
2188          * Search progressively larger portions of the hierarchy, starting
2189          * with the filesystem specified by 'parent'.  This will
2190          * select the "most local" version of the origin snapshot in the case
2191          * that there are multiple matching snapshots in the system.
2192          */
2193         (void) strlcpy(pname, parent, sizeof (pname));
2194         char *cp = strrchr(pname, '@');
2195         if (cp == NULL)
2196                 cp = strchr(pname, '\0');
2197         for (; cp != NULL; cp = strrchr(pname, '/')) {
2198                 /* Chop off the last component and open the parent */
2199                 *cp = '\0';
2200                 zfs_handle_t *zhp = make_dataset_handle(hdl, pname);
2201 
2202                 if (zhp == NULL)
2203                         continue;
2204                 int err = guid_to_name_cb(zfs_handle_dup(zhp), &gtnd);
2205                 if (err != EEXIST)
2206                         err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
2207                 if (err != EEXIST && bookmark_ok)
2208                         err = zfs_iter_bookmarks(zhp, guid_to_name_cb, &gtnd);
2209                 zfs_close(zhp);
2210                 if (err == EEXIST)
2211                         return (0);
2212 
2213                 /*
2214                  * Remember the last portion of the dataset so we skip it next
2215                  * time through (as we've already searched that portion of the
2216                  * hierarchy).
2217                  */
2218                 gtnd.skip = strrchr(pname, '/') + 1;
2219         }
2220 
2221         return (ENOENT);
2222 }
2223 
2224 /*
2225  * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
2226  * guid1 is after guid2.
2227  */
2228 static int
2229 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
2230     uint64_t guid1, uint64_t guid2)
2231 {
2232         nvlist_t *nvfs;
2233         char *fsname, *snapname;
2234         char buf[ZFS_MAXNAMELEN];
2235         int rv;
2236         zfs_handle_t *guid1hdl, *guid2hdl;
2237         uint64_t create1, create2;
2238 
2239         if (guid2 == 0)
2240                 return (0);
2241         if (guid1 == 0)
2242                 return (1);
2243 
2244         nvfs = fsavl_find(avl, guid1, &snapname);
2245         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2246         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2247         guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2248         if (guid1hdl == NULL)
2249                 return (-1);
2250 
2251         nvfs = fsavl_find(avl, guid2, &snapname);
2252         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2253         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2254         guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2255         if (guid2hdl == NULL) {
2256                 zfs_close(guid1hdl);
2257                 return (-1);
2258         }
2259 
2260         create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
2261         create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
2262 
2263         if (create1 < create2)
2264                 rv = -1;
2265         else if (create1 > create2)
2266                 rv = +1;
2267         else
2268                 rv = 0;
2269 
2270         zfs_close(guid1hdl);
2271         zfs_close(guid2hdl);
2272 
2273         return (rv);
2274 }
2275 
2276 static int
2277 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2278     recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2279     nvlist_t *renamed)
2280 {
2281         nvlist_t *local_nv;
2282         avl_tree_t *local_avl;
2283         nvpair_t *fselem, *nextfselem;
2284         char *fromsnap;
2285         char newname[ZFS_MAXNAMELEN];
2286         int error;
2287         boolean_t needagain, progress, recursive;
2288         char *s1, *s2;
2289 
2290         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2291 
2292         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2293             ENOENT);
2294 
2295         if (flags->dryrun)
2296                 return (0);
2297 
2298 again:
2299         needagain = progress = B_FALSE;
2300 
2301         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2302             recursive, &local_nv, &local_avl)) != 0)
2303                 return (error);
2304 
2305         /*
2306          * Process deletes and renames
2307          */
2308         for (fselem = nvlist_next_nvpair(local_nv, NULL);
2309             fselem; fselem = nextfselem) {
2310                 nvlist_t *nvfs, *snaps;
2311                 nvlist_t *stream_nvfs = NULL;
2312                 nvpair_t *snapelem, *nextsnapelem;
2313                 uint64_t fromguid = 0;
2314                 uint64_t originguid = 0;
2315                 uint64_t stream_originguid = 0;
2316                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2317                 char *fsname, *stream_fsname;
2318 
2319                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2320 
2321                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2322                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2323                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2324                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2325                     &parent_fromsnap_guid));
2326                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2327 
2328                 /*
2329                  * First find the stream's fs, so we can check for
2330                  * a different origin (due to "zfs promote")
2331                  */
2332                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2333                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2334                         uint64_t thisguid;
2335 
2336                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2337                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2338 
2339                         if (stream_nvfs != NULL)
2340                                 break;
2341                 }
2342 
2343                 /* check for promote */
2344                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2345                     &stream_originguid);
2346                 if (stream_nvfs && originguid != stream_originguid) {
2347                         switch (created_before(hdl, local_avl,
2348                             stream_originguid, originguid)) {
2349                         case 1: {
2350                                 /* promote it! */
2351                                 zfs_cmd_t zc = { 0 };
2352                                 nvlist_t *origin_nvfs;
2353                                 char *origin_fsname;
2354 
2355                                 if (flags->verbose)
2356                                         (void) printf("promoting %s\n", fsname);
2357 
2358                                 origin_nvfs = fsavl_find(local_avl, originguid,
2359                                     NULL);
2360                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2361                                     "name", &origin_fsname));
2362                                 (void) strlcpy(zc.zc_value, origin_fsname,
2363                                     sizeof (zc.zc_value));
2364                                 (void) strlcpy(zc.zc_name, fsname,
2365                                     sizeof (zc.zc_name));
2366                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2367                                 if (error == 0)
2368                                         progress = B_TRUE;
2369                                 break;
2370                         }
2371                         default:
2372                                 break;
2373                         case -1:
2374                                 fsavl_destroy(local_avl);
2375                                 nvlist_free(local_nv);
2376                                 return (-1);
2377                         }
2378                         /*
2379                          * We had/have the wrong origin, therefore our
2380                          * list of snapshots is wrong.  Need to handle
2381                          * them on the next pass.
2382                          */
2383                         needagain = B_TRUE;
2384                         continue;
2385                 }
2386 
2387                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2388                     snapelem; snapelem = nextsnapelem) {
2389                         uint64_t thisguid;
2390                         char *stream_snapname;
2391                         nvlist_t *found, *props;
2392 
2393                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2394 
2395                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2396                         found = fsavl_find(stream_avl, thisguid,
2397                             &stream_snapname);
2398 
2399                         /* check for delete */
2400                         if (found == NULL) {
2401                                 char name[ZFS_MAXNAMELEN];
2402 
2403                                 if (!flags->force)
2404                                         continue;
2405 
2406                                 (void) snprintf(name, sizeof (name), "%s@%s",
2407                                     fsname, nvpair_name(snapelem));
2408 
2409                                 error = recv_destroy(hdl, name,
2410                                     strlen(fsname)+1, newname, flags);
2411                                 if (error)
2412                                         needagain = B_TRUE;
2413                                 else
2414                                         progress = B_TRUE;
2415                                 continue;
2416                         }
2417 
2418                         stream_nvfs = found;
2419 
2420                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2421                             &props) && 0 == nvlist_lookup_nvlist(props,
2422                             stream_snapname, &props)) {
2423                                 zfs_cmd_t zc = { 0 };
2424 
2425                                 zc.zc_cookie = B_TRUE; /* received */
2426                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2427                                     "%s@%s", fsname, nvpair_name(snapelem));
2428                                 if (zcmd_write_src_nvlist(hdl, &zc,
2429                                     props) == 0) {
2430                                         (void) zfs_ioctl(hdl,
2431                                             ZFS_IOC_SET_PROP, &zc);
2432                                         zcmd_free_nvlists(&zc);
2433                                 }
2434                         }
2435 
2436                         /* check for different snapname */
2437                         if (strcmp(nvpair_name(snapelem),
2438                             stream_snapname) != 0) {
2439                                 char name[ZFS_MAXNAMELEN];
2440                                 char tryname[ZFS_MAXNAMELEN];
2441 
2442                                 (void) snprintf(name, sizeof (name), "%s@%s",
2443                                     fsname, nvpair_name(snapelem));
2444                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
2445                                     fsname, stream_snapname);
2446 
2447                                 error = recv_rename(hdl, name, tryname,
2448                                     strlen(fsname)+1, newname, flags);
2449                                 if (error)
2450                                         needagain = B_TRUE;
2451                                 else
2452                                         progress = B_TRUE;
2453                         }
2454 
2455                         if (strcmp(stream_snapname, fromsnap) == 0)
2456                                 fromguid = thisguid;
2457                 }
2458 
2459                 /* check for delete */
2460                 if (stream_nvfs == NULL) {
2461                         if (!flags->force)
2462                                 continue;
2463 
2464                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2465                             newname, flags);
2466                         if (error)
2467                                 needagain = B_TRUE;
2468                         else
2469                                 progress = B_TRUE;
2470                         continue;
2471                 }
2472 
2473                 if (fromguid == 0) {
2474                         if (flags->verbose) {
2475                                 (void) printf("local fs %s does not have "
2476                                     "fromsnap (%s in stream); must have "
2477                                     "been deleted locally; ignoring\n",
2478                                     fsname, fromsnap);
2479                         }
2480                         continue;
2481                 }
2482 
2483                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2484                     "name", &stream_fsname));
2485                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2486                     "parentfromsnap", &stream_parent_fromsnap_guid));
2487 
2488                 s1 = strrchr(fsname, '/');
2489                 s2 = strrchr(stream_fsname, '/');
2490 
2491                 /*
2492                  * Check for rename. If the exact receive path is specified, it
2493                  * does not count as a rename, but we still need to check the
2494                  * datasets beneath it.
2495                  */
2496                 if ((stream_parent_fromsnap_guid != 0 &&
2497                     parent_fromsnap_guid != 0 &&
2498                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2499                     ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2500                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2501                         nvlist_t *parent;
2502                         char tryname[ZFS_MAXNAMELEN];
2503 
2504                         parent = fsavl_find(local_avl,
2505                             stream_parent_fromsnap_guid, NULL);
2506                         /*
2507                          * NB: parent might not be found if we used the
2508                          * tosnap for stream_parent_fromsnap_guid,
2509                          * because the parent is a newly-created fs;
2510                          * we'll be able to rename it after we recv the
2511                          * new fs.
2512                          */
2513                         if (parent != NULL) {
2514                                 char *pname;
2515 
2516                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
2517                                     &pname));
2518                                 (void) snprintf(tryname, sizeof (tryname),
2519                                     "%s%s", pname, strrchr(stream_fsname, '/'));
2520                         } else {
2521                                 tryname[0] = '\0';
2522                                 if (flags->verbose) {
2523                                         (void) printf("local fs %s new parent "
2524                                             "not found\n", fsname);
2525                                 }
2526                         }
2527 
2528                         newname[0] = '\0';
2529 
2530                         error = recv_rename(hdl, fsname, tryname,
2531                             strlen(tofs)+1, newname, flags);
2532 
2533                         if (renamed != NULL && newname[0] != '\0') {
2534                                 VERIFY(0 == nvlist_add_boolean(renamed,
2535                                     newname));
2536                         }
2537 
2538                         if (error)
2539                                 needagain = B_TRUE;
2540                         else
2541                                 progress = B_TRUE;
2542                 }
2543         }
2544 
2545         fsavl_destroy(local_avl);
2546         nvlist_free(local_nv);
2547 
2548         if (needagain && progress) {
2549                 /* do another pass to fix up temporary names */
2550                 if (flags->verbose)
2551                         (void) printf("another pass:\n");
2552                 goto again;
2553         }
2554 
2555         return (needagain);
2556 }
2557 
2558 static int
2559 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2560     recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2561     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2562 {
2563         nvlist_t *stream_nv = NULL;
2564         avl_tree_t *stream_avl = NULL;
2565         char *fromsnap = NULL;
2566         char *sendsnap = NULL;
2567         char *cp;
2568         char tofs[ZFS_MAXNAMELEN];
2569         char sendfs[ZFS_MAXNAMELEN];
2570         char errbuf[1024];
2571         dmu_replay_record_t drre;
2572         int error;
2573         boolean_t anyerr = B_FALSE;
2574         boolean_t softerr = B_FALSE;
2575         boolean_t recursive;
2576 
2577         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2578             "cannot receive"));
2579 
2580         assert(drr->drr_type == DRR_BEGIN);
2581         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2582         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2583             DMU_COMPOUNDSTREAM);
2584 
2585         /*
2586          * Read in the nvlist from the stream.
2587          */
2588         if (drr->drr_payloadlen != 0) {
2589                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2590                     &stream_nv, flags->byteswap, zc);
2591                 if (error) {
2592                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2593                         goto out;
2594                 }
2595         }
2596 
2597         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2598             ENOENT);
2599 
2600         if (recursive && strchr(destname, '@')) {
2601                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2602                     "cannot specify snapshot name for multi-snapshot stream"));
2603                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2604                 goto out;
2605         }
2606 
2607         /*
2608          * Read in the end record and verify checksum.
2609          */
2610         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2611             flags->byteswap, NULL)))
2612                 goto out;
2613         if (flags->byteswap) {
2614                 drre.drr_type = BSWAP_32(drre.drr_type);
2615                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2616                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2617                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2618                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2619                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2620                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2621                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2622                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2623         }
2624         if (drre.drr_type != DRR_END) {
2625                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2626                 goto out;
2627         }
2628         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2629                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2630                     "incorrect header checksum"));
2631                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2632                 goto out;
2633         }
2634 
2635         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2636 
2637         if (drr->drr_payloadlen != 0) {
2638                 nvlist_t *stream_fss;
2639 
2640                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2641                     &stream_fss));
2642                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2643                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2644                             "couldn't allocate avl tree"));
2645                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2646                         goto out;
2647                 }
2648 
2649                 if (fromsnap != NULL) {
2650                         nvlist_t *renamed = NULL;
2651                         nvpair_t *pair = NULL;
2652 
2653                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2654                         if (flags->isprefix) {
2655                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2656                                 int i;
2657 
2658                                 if (flags->istail) {
2659                                         cp = strrchr(drrb->drr_toname, '/');
2660                                         if (cp == NULL) {
2661                                                 (void) strlcat(tofs, "/",
2662                                                     ZFS_MAXNAMELEN);
2663                                                 i = 0;
2664                                         } else {
2665                                                 i = (cp - drrb->drr_toname);
2666                                         }
2667                                 } else {
2668                                         i = strcspn(drrb->drr_toname, "/@");
2669                                 }
2670                                 /* zfs_receive_one() will create_parents() */
2671                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2672                                     ZFS_MAXNAMELEN);
2673                                 *strchr(tofs, '@') = '\0';
2674                         }
2675 
2676                         if (recursive && !flags->dryrun && !flags->nomount) {
2677                                 VERIFY(0 == nvlist_alloc(&renamed,
2678                                     NV_UNIQUE_NAME, 0));
2679                         }
2680 
2681                         softerr = recv_incremental_replication(hdl, tofs, flags,
2682                             stream_nv, stream_avl, renamed);
2683 
2684                         /* Unmount renamed filesystems before receiving. */
2685                         while ((pair = nvlist_next_nvpair(renamed,
2686                             pair)) != NULL) {
2687                                 zfs_handle_t *zhp;
2688                                 prop_changelist_t *clp = NULL;
2689 
2690                                 zhp = zfs_open(hdl, nvpair_name(pair),
2691                                     ZFS_TYPE_FILESYSTEM);
2692                                 if (zhp != NULL) {
2693                                         clp = changelist_gather(zhp,
2694                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2695                                         zfs_close(zhp);
2696                                         if (clp != NULL) {
2697                                                 softerr |=
2698                                                     changelist_prefix(clp);
2699                                                 changelist_free(clp);
2700                                         }
2701                                 }
2702                         }
2703 
2704                         nvlist_free(renamed);
2705                 }
2706         }
2707 
2708         /*
2709          * Get the fs specified by the first path in the stream (the top level
2710          * specified by 'zfs send') and pass it to each invocation of
2711          * zfs_receive_one().
2712          */
2713         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2714             ZFS_MAXNAMELEN);
2715         if ((cp = strchr(sendfs, '@')) != NULL) {
2716                 *cp = '\0';
2717                 /*
2718                  * Find the "sendsnap", the final snapshot in a replication
2719                  * stream.  zfs_receive_one() handles certain errors
2720                  * differently, depending on if the contained stream is the
2721                  * last one or not.
2722                  */
2723                 sendsnap = (cp + 1);
2724         }
2725 
2726         /* Finally, receive each contained stream */
2727         do {
2728                 /*
2729                  * we should figure out if it has a recoverable
2730                  * error, in which case do a recv_skip() and drive on.
2731                  * Note, if we fail due to already having this guid,
2732                  * zfs_receive_one() will take care of it (ie,
2733                  * recv_skip() and return 0).
2734                  */
2735                 error = zfs_receive_impl(hdl, destname, NULL, flags, fd,
2736                     sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2737                     action_handlep, sendsnap);
2738                 if (error == ENODATA) {
2739                         error = 0;
2740                         break;
2741                 }
2742                 anyerr |= error;
2743         } while (error == 0);
2744 
2745         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2746                 /*
2747                  * Now that we have the fs's they sent us, try the
2748                  * renames again.
2749                  */
2750                 softerr = recv_incremental_replication(hdl, tofs, flags,
2751                     stream_nv, stream_avl, NULL);
2752         }
2753 
2754 out:
2755         fsavl_destroy(stream_avl);
2756         if (stream_nv)
2757                 nvlist_free(stream_nv);
2758         if (softerr)
2759                 error = -2;
2760         if (anyerr)
2761                 error = -1;
2762         return (error);
2763 }
2764 
2765 static void
2766 trunc_prop_errs(int truncated)
2767 {
2768         ASSERT(truncated != 0);
2769 
2770         if (truncated == 1)
2771                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2772                     "1 more property could not be set\n"));
2773         else
2774                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2775                     "%d more properties could not be set\n"), truncated);
2776 }
2777 
2778 static int
2779 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2780 {
2781         dmu_replay_record_t *drr;
2782         void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
2783         char errbuf[1024];
2784 
2785         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2786             "cannot receive:"));
2787 
2788         /* XXX would be great to use lseek if possible... */
2789         drr = buf;
2790 
2791         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2792             byteswap, NULL) == 0) {
2793                 if (byteswap)
2794                         drr->drr_type = BSWAP_32(drr->drr_type);
2795 
2796                 switch (drr->drr_type) {
2797                 case DRR_BEGIN:
2798                         if (drr->drr_payloadlen != 0) {
2799                                 (void) recv_read(hdl, fd, buf,
2800                                     drr->drr_payloadlen, B_FALSE, NULL);
2801                         }
2802                         break;
2803 
2804                 case DRR_END:
2805                         free(buf);
2806                         return (0);
2807 
2808                 case DRR_OBJECT:
2809                         if (byteswap) {
2810                                 drr->drr_u.drr_object.drr_bonuslen =
2811                                     BSWAP_32(drr->drr_u.drr_object.
2812                                     drr_bonuslen);
2813                         }
2814                         (void) recv_read(hdl, fd, buf,
2815                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2816                             B_FALSE, NULL);
2817                         break;
2818 
2819                 case DRR_WRITE:
2820                         if (byteswap) {
2821                                 drr->drr_u.drr_write.drr_length =
2822                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
2823                         }
2824                         (void) recv_read(hdl, fd, buf,
2825                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2826                         break;
2827                 case DRR_SPILL:
2828                         if (byteswap) {
2829                                 drr->drr_u.drr_write.drr_length =
2830                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
2831                         }
2832                         (void) recv_read(hdl, fd, buf,
2833                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2834                         break;
2835                 case DRR_WRITE_EMBEDDED:
2836                         if (byteswap) {
2837                                 drr->drr_u.drr_write_embedded.drr_psize =
2838                                     BSWAP_32(drr->drr_u.drr_write_embedded.
2839                                     drr_psize);
2840                         }
2841                         (void) recv_read(hdl, fd, buf,
2842                             P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
2843                             8), B_FALSE, NULL);
2844                         break;
2845                 case DRR_WRITE_BYREF:
2846                 case DRR_FREEOBJECTS:
2847                 case DRR_FREE:
2848                         break;
2849 
2850                 default:
2851                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2852                             "invalid record type"));
2853                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2854                 }
2855         }
2856 
2857         free(buf);
2858         return (-1);
2859 }
2860 
2861 static void
2862 recv_ecksum_set_aux(libzfs_handle_t *hdl, const char *target_snap,
2863     boolean_t resumable)
2864 {
2865         char target_fs[ZFS_MAXNAMELEN];
2866 
2867         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2868             "checksum mismatch or incomplete stream"));
2869 
2870         if (!resumable)
2871                 return;
2872         (void) strlcpy(target_fs, target_snap, sizeof (target_fs));
2873         *strchr(target_fs, '@') = '\0';
2874         zfs_handle_t *zhp = zfs_open(hdl, target_fs,
2875             ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
2876         if (zhp == NULL)
2877                 return;
2878 
2879         char token_buf[ZFS_MAXPROPLEN];
2880         int error = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN,
2881             token_buf, sizeof (token_buf),
2882             NULL, NULL, 0, B_TRUE);
2883         if (error == 0) {
2884                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2885                     "checksum mismatch or incomplete stream.\n"
2886                     "Partially received snapshot is saved.\n"
2887                     "A resuming stream can be generated on the sending "
2888                     "system by running:\n"
2889                     "    zfs send -t %s"),
2890                     token_buf);
2891         }
2892         zfs_close(zhp);
2893 }
2894 
2895 /*
2896  * Restores a backup of tosnap from the file descriptor specified by infd.
2897  */
2898 static int
2899 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2900     const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr,
2901     dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv,
2902     avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2903     uint64_t *action_handlep, const char *finalsnap)
2904 {
2905         zfs_cmd_t zc = { 0 };
2906         time_t begin_time;
2907         int ioctl_err, ioctl_errno, err;
2908         char *cp;
2909         struct drr_begin *drrb = &drr->drr_u.drr_begin;
2910         char errbuf[1024];
2911         char prop_errbuf[1024];
2912         const char *chopprefix;
2913         boolean_t newfs = B_FALSE;
2914         boolean_t stream_wantsnewfs;
2915         uint64_t parent_snapguid = 0;
2916         prop_changelist_t *clp = NULL;
2917         nvlist_t *snapprops_nvlist = NULL;
2918         zprop_errflags_t prop_errflags;
2919         boolean_t recursive;
2920         char *snapname = NULL;
2921 
2922         begin_time = time(NULL);
2923 
2924         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2925             "cannot receive"));
2926 
2927         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2928             ENOENT);
2929 
2930         if (stream_avl != NULL) {
2931                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2932                     &snapname);
2933                 nvlist_t *props;
2934                 int ret;
2935 
2936                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2937                     &parent_snapguid);
2938                 err = nvlist_lookup_nvlist(fs, "props", &props);
2939                 if (err)
2940                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2941 
2942                 if (flags->canmountoff) {
2943                         VERIFY(0 == nvlist_add_uint64(props,
2944                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2945                 }
2946                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2947                 if (err)
2948                         nvlist_free(props);
2949 
2950                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2951                         VERIFY(0 == nvlist_lookup_nvlist(props,
2952                             snapname, &snapprops_nvlist));
2953                 }
2954 
2955                 if (ret != 0)
2956                         return (-1);
2957         }
2958 
2959         cp = NULL;
2960 
2961         /*
2962          * Determine how much of the snapshot name stored in the stream
2963          * we are going to tack on to the name they specified on the
2964          * command line, and how much we are going to chop off.
2965          *
2966          * If they specified a snapshot, chop the entire name stored in
2967          * the stream.
2968          */
2969         if (flags->istail) {
2970                 /*
2971                  * A filesystem was specified with -e. We want to tack on only
2972                  * the tail of the sent snapshot path.
2973                  */
2974                 if (strchr(tosnap, '@')) {
2975                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2976                             "argument - snapshot not allowed with -e"));
2977                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2978                 }
2979 
2980                 chopprefix = strrchr(sendfs, '/');
2981 
2982                 if (chopprefix == NULL) {
2983                         /*
2984                          * The tail is the poolname, so we need to
2985                          * prepend a path separator.
2986                          */
2987                         int len = strlen(drrb->drr_toname);
2988                         cp = malloc(len + 2);
2989                         cp[0] = '/';
2990                         (void) strcpy(&cp[1], drrb->drr_toname);
2991                         chopprefix = cp;
2992                 } else {
2993                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2994                 }
2995         } else if (flags->isprefix) {
2996                 /*
2997                  * A filesystem was specified with -d. We want to tack on
2998                  * everything but the first element of the sent snapshot path
2999                  * (all but the pool name).
3000                  */
3001                 if (strchr(tosnap, '@')) {
3002                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3003                             "argument - snapshot not allowed with -d"));
3004                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3005                 }
3006 
3007                 chopprefix = strchr(drrb->drr_toname, '/');
3008                 if (chopprefix == NULL)
3009                         chopprefix = strchr(drrb->drr_toname, '@');
3010         } else if (strchr(tosnap, '@') == NULL) {
3011                 /*
3012                  * If a filesystem was specified without -d or -e, we want to
3013                  * tack on everything after the fs specified by 'zfs send'.
3014                  */
3015                 chopprefix = drrb->drr_toname + strlen(sendfs);
3016         } else {
3017                 /* A snapshot was specified as an exact path (no -d or -e). */
3018                 if (recursive) {
3019                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3020                             "cannot specify snapshot name for multi-snapshot "
3021                             "stream"));
3022                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3023                 }
3024                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
3025         }
3026 
3027         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
3028         ASSERT(chopprefix > drrb->drr_toname);
3029         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
3030         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
3031             chopprefix[0] == '\0');
3032 
3033         /*
3034          * Determine name of destination snapshot, store in zc_value.
3035          */
3036         (void) strcpy(zc.zc_value, tosnap);
3037         (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
3038         free(cp);
3039         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
3040                 zcmd_free_nvlists(&zc);
3041                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3042         }
3043 
3044         /*
3045          * Determine the name of the origin snapshot, store in zc_string.
3046          */
3047         if (drrb->drr_flags & DRR_FLAG_CLONE) {
3048                 if (guid_to_name(hdl, zc.zc_value,
3049                     drrb->drr_fromguid, B_FALSE, zc.zc_string) != 0) {
3050                         zcmd_free_nvlists(&zc);
3051                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3052                             "local origin for clone %s does not exist"),
3053                             zc.zc_value);
3054                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
3055                 }
3056                 if (flags->verbose)
3057                         (void) printf("found clone origin %s\n", zc.zc_string);
3058         } else if (originsnap) {
3059                 (void) strncpy(zc.zc_string, originsnap, ZFS_MAXNAMELEN);
3060                 if (flags->verbose)
3061                         (void) printf("using provided clone origin %s\n",
3062                             zc.zc_string);
3063         }
3064 
3065         boolean_t resuming = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
3066             DMU_BACKUP_FEATURE_RESUMING;
3067         stream_wantsnewfs = (drrb->drr_fromguid == NULL ||
3068             (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && !resuming;
3069 
3070         if (stream_wantsnewfs) {
3071                 /*
3072                  * if the parent fs does not exist, look for it based on
3073                  * the parent snap GUID
3074                  */
3075                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3076                     "cannot receive new filesystem stream"));
3077 
3078                 (void) strcpy(zc.zc_name, zc.zc_value);
3079                 cp = strrchr(zc.zc_name, '/');
3080                 if (cp)
3081                         *cp = '\0';
3082                 if (cp &&
3083                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3084                         char suffix[ZFS_MAXNAMELEN];
3085                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
3086                         if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
3087                             B_FALSE, zc.zc_value) == 0) {
3088                                 *strchr(zc.zc_value, '@') = '\0';
3089                                 (void) strcat(zc.zc_value, suffix);
3090                         }
3091                 }
3092         } else {
3093                 /*
3094                  * if the fs does not exist, look for it based on the
3095                  * fromsnap GUID
3096                  */
3097                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3098                     "cannot receive incremental stream"));
3099 
3100                 (void) strcpy(zc.zc_name, zc.zc_value);
3101                 *strchr(zc.zc_name, '@') = '\0';
3102 
3103                 /*
3104                  * If the exact receive path was specified and this is the
3105                  * topmost path in the stream, then if the fs does not exist we
3106                  * should look no further.
3107                  */
3108                 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
3109                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
3110                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3111                         char snap[ZFS_MAXNAMELEN];
3112                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
3113                         if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
3114                             B_FALSE, zc.zc_value) == 0) {
3115                                 *strchr(zc.zc_value, '@') = '\0';
3116                                 (void) strcat(zc.zc_value, snap);
3117                         }
3118                 }
3119         }
3120 
3121         (void) strcpy(zc.zc_name, zc.zc_value);
3122         *strchr(zc.zc_name, '@') = '\0';
3123 
3124         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3125                 zfs_handle_t *zhp;
3126 
3127                 /*
3128                  * Destination fs exists.  It must be one of these cases:
3129                  *  - an incremental send stream
3130                  *  - the stream specifies a new fs (full stream or clone)
3131                  *    and they want us to blow away the existing fs (and
3132                  *    have therefore specified -F and removed any snapshots)
3133                  *  - we are resuming a failed receive.
3134                  */
3135                 if (stream_wantsnewfs) {
3136                         if (!flags->force) {
3137                                 zcmd_free_nvlists(&zc);
3138                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3139                                     "destination '%s' exists\n"
3140                                     "must specify -F to overwrite it"),
3141                                     zc.zc_name);
3142                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3143                         }
3144                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
3145                             &zc) == 0) {
3146                                 zcmd_free_nvlists(&zc);
3147                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3148                                     "destination has snapshots (eg. %s)\n"
3149                                     "must destroy them to overwrite it"),
3150                                     zc.zc_name);
3151                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3152                         }
3153                 }
3154 
3155                 if ((zhp = zfs_open(hdl, zc.zc_name,
3156                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
3157                         zcmd_free_nvlists(&zc);
3158                         return (-1);
3159                 }
3160 
3161                 if (stream_wantsnewfs &&
3162                     zhp->zfs_dmustats.dds_origin[0]) {
3163                         zcmd_free_nvlists(&zc);
3164                         zfs_close(zhp);
3165                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3166                             "destination '%s' is a clone\n"
3167                             "must destroy it to overwrite it"),
3168                             zc.zc_name);
3169                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3170                 }
3171 
3172                 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
3173                     stream_wantsnewfs) {
3174                         /* We can't do online recv in this case */
3175                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
3176                         if (clp == NULL) {
3177                                 zfs_close(zhp);
3178                                 zcmd_free_nvlists(&zc);
3179                                 return (-1);
3180                         }
3181                         if (changelist_prefix(clp) != 0) {
3182                                 changelist_free(clp);
3183                                 zfs_close(zhp);
3184                                 zcmd_free_nvlists(&zc);
3185                                 return (-1);
3186                         }
3187                 }
3188 
3189                 /*
3190                  * If we are resuming a newfs, set newfs here so that we will
3191                  * mount it if the recv succeeds this time.  We can tell
3192                  * that it was a newfs on the first recv because the fs
3193                  * itself will be inconsistent (if the fs existed when we
3194                  * did the first recv, we would have received it into
3195                  * .../%recv).
3196                  */
3197                 if (resuming && zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT))
3198                         newfs = B_TRUE;
3199 
3200                 zfs_close(zhp);
3201         } else {
3202                 /*
3203                  * Destination filesystem does not exist.  Therefore we better
3204                  * be creating a new filesystem (either from a full backup, or
3205                  * a clone).  It would therefore be invalid if the user
3206                  * specified only the pool name (i.e. if the destination name
3207                  * contained no slash character).
3208                  */
3209                 if (!stream_wantsnewfs ||
3210                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
3211                         zcmd_free_nvlists(&zc);
3212                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3213                             "destination '%s' does not exist"), zc.zc_name);
3214                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
3215                 }
3216 
3217                 /*
3218                  * Trim off the final dataset component so we perform the
3219                  * recvbackup ioctl to the filesystems's parent.
3220                  */
3221                 *cp = '\0';
3222 
3223                 if (flags->isprefix && !flags->istail && !flags->dryrun &&
3224                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
3225                         zcmd_free_nvlists(&zc);
3226                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
3227                 }
3228 
3229                 newfs = B_TRUE;
3230         }
3231 
3232         zc.zc_begin_record = *drr_noswap;
3233         zc.zc_cookie = infd;
3234         zc.zc_guid = flags->force;
3235         zc.zc_resumable = flags->resumable;
3236         if (flags->verbose) {
3237                 (void) printf("%s %s stream of %s into %s\n",
3238                     flags->dryrun ? "would receive" : "receiving",
3239                     drrb->drr_fromguid ? "incremental" : "full",
3240                     drrb->drr_toname, zc.zc_value);
3241                 (void) fflush(stdout);
3242         }
3243 
3244         if (flags->dryrun) {
3245                 zcmd_free_nvlists(&zc);
3246                 return (recv_skip(hdl, infd, flags->byteswap));
3247         }
3248 
3249         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
3250         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
3251         zc.zc_cleanup_fd = cleanup_fd;
3252         zc.zc_action_handle = *action_handlep;
3253 
3254         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
3255         ioctl_errno = errno;
3256         prop_errflags = (zprop_errflags_t)zc.zc_obj;
3257 
3258         if (err == 0) {
3259                 nvlist_t *prop_errors;
3260                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
3261                     zc.zc_nvlist_dst_size, &prop_errors, 0));
3262 
3263                 nvpair_t *prop_err = NULL;
3264 
3265                 while ((prop_err = nvlist_next_nvpair(prop_errors,
3266                     prop_err)) != NULL) {
3267                         char tbuf[1024];
3268                         zfs_prop_t prop;
3269                         int intval;
3270 
3271                         prop = zfs_name_to_prop(nvpair_name(prop_err));
3272                         (void) nvpair_value_int32(prop_err, &intval);
3273                         if (strcmp(nvpair_name(prop_err),
3274                             ZPROP_N_MORE_ERRORS) == 0) {
3275                                 trunc_prop_errs(intval);
3276                                 break;
3277                         } else if (snapname == NULL || finalsnap == NULL ||
3278                             strcmp(finalsnap, snapname) == 0 ||
3279                             strcmp(nvpair_name(prop_err),
3280                             zfs_prop_to_name(ZFS_PROP_REFQUOTA)) != 0) {
3281                                 /*
3282                                  * Skip the special case of, for example,
3283                                  * "refquota", errors on intermediate
3284                                  * snapshots leading up to a final one.
3285                                  * That's why we have all of the checks above.
3286                                  *
3287                                  * See zfs_ioctl.c's extract_delay_props() for
3288                                  * a list of props which can fail on
3289                                  * intermediate snapshots, but shouldn't
3290                                  * affect the overall receive.
3291                                  */
3292                                 (void) snprintf(tbuf, sizeof (tbuf),
3293                                     dgettext(TEXT_DOMAIN,
3294                                     "cannot receive %s property on %s"),
3295                                     nvpair_name(prop_err), zc.zc_name);
3296                                 zfs_setprop_error(hdl, prop, intval, tbuf);
3297                         }
3298                 }
3299                 nvlist_free(prop_errors);
3300         }
3301 
3302         zc.zc_nvlist_dst = 0;
3303         zc.zc_nvlist_dst_size = 0;
3304         zcmd_free_nvlists(&zc);
3305 
3306         if (err == 0 && snapprops_nvlist) {
3307                 zfs_cmd_t zc2 = { 0 };
3308 
3309                 (void) strcpy(zc2.zc_name, zc.zc_value);
3310                 zc2.zc_cookie = B_TRUE; /* received */
3311                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
3312                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
3313                         zcmd_free_nvlists(&zc2);
3314                 }
3315         }
3316 
3317         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
3318                 /*
3319                  * It may be that this snapshot already exists,
3320                  * in which case we want to consume & ignore it
3321                  * rather than failing.
3322                  */
3323                 avl_tree_t *local_avl;
3324                 nvlist_t *local_nv, *fs;
3325                 cp = strchr(zc.zc_value, '@');
3326 
3327                 /*
3328                  * XXX Do this faster by just iterating over snaps in
3329                  * this fs.  Also if zc_value does not exist, we will
3330                  * get a strange "does not exist" error message.
3331                  */
3332                 *cp = '\0';
3333                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3334                     &local_nv, &local_avl) == 0) {
3335                         *cp = '@';
3336                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3337                         fsavl_destroy(local_avl);
3338                         nvlist_free(local_nv);
3339 
3340                         if (fs != NULL) {
3341                                 if (flags->verbose) {
3342                                         (void) printf("snap %s already exists; "
3343                                             "ignoring\n", zc.zc_value);
3344                                 }
3345                                 err = ioctl_err = recv_skip(hdl, infd,
3346                                     flags->byteswap);
3347                         }
3348                 }
3349                 *cp = '@';
3350         }
3351 
3352         if (ioctl_err != 0) {
3353                 switch (ioctl_errno) {
3354                 case ENODEV:
3355                         cp = strchr(zc.zc_value, '@');
3356                         *cp = '\0';
3357                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3358                             "most recent snapshot of %s does not\n"
3359                             "match incremental source"), zc.zc_value);
3360                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3361                         *cp = '@';
3362                         break;
3363                 case ETXTBSY:
3364                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3365                             "destination %s has been modified\n"
3366                             "since most recent snapshot"), zc.zc_name);
3367                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3368                         break;
3369                 case EEXIST:
3370                         cp = strchr(zc.zc_value, '@');
3371                         if (newfs) {
3372                                 /* it's the containing fs that exists */
3373                                 *cp = '\0';
3374                         }
3375                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3376                             "destination already exists"));
3377                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3378                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3379                             zc.zc_value);
3380                         *cp = '@';
3381                         break;
3382                 case EINVAL:
3383                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3384                         break;
3385                 case ECKSUM:
3386                         recv_ecksum_set_aux(hdl, zc.zc_value, flags->resumable);
3387                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3388                         break;
3389                 case ENOTSUP:
3390                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3391                             "pool must be upgraded to receive this stream."));
3392                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3393                         break;
3394                 case EDQUOT:
3395                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3396                             "destination %s space quota exceeded"), zc.zc_name);
3397                         (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3398                         break;
3399                 default:
3400                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3401                 }
3402         }
3403 
3404         /*
3405          * Mount the target filesystem (if created).  Also mount any
3406          * children of the target filesystem if we did a replication
3407          * receive (indicated by stream_avl being non-NULL).
3408          */
3409         cp = strchr(zc.zc_value, '@');
3410         if (cp && (ioctl_err == 0 || !newfs)) {
3411                 zfs_handle_t *h;
3412 
3413                 *cp = '\0';
3414                 h = zfs_open(hdl, zc.zc_value,
3415                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3416                 if (h != NULL) {
3417                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
3418                                 *cp = '@';
3419                         } else if (newfs || stream_avl) {
3420                                 /*
3421                                  * Track the first/top of hierarchy fs,
3422                                  * for mounting and sharing later.
3423                                  */
3424                                 if (top_zfs && *top_zfs == NULL)
3425                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
3426                         }
3427                         zfs_close(h);
3428                 }
3429                 *cp = '@';
3430         }
3431 
3432         if (clp) {
3433                 err |= changelist_postfix(clp);
3434                 changelist_free(clp);
3435         }
3436 
3437         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3438                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3439                     "failed to clear unreceived properties on %s"),
3440                     zc.zc_name);
3441                 (void) fprintf(stderr, "\n");
3442         }
3443         if (prop_errflags & ZPROP_ERR_NORESTORE) {
3444                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3445                     "failed to restore original properties on %s"),
3446                     zc.zc_name);
3447                 (void) fprintf(stderr, "\n");
3448         }
3449 
3450         if (err || ioctl_err)
3451                 return (-1);
3452 
3453         *action_handlep = zc.zc_action_handle;
3454 
3455         if (flags->verbose) {
3456                 char buf1[64];
3457                 char buf2[64];
3458                 uint64_t bytes = zc.zc_cookie;
3459                 time_t delta = time(NULL) - begin_time;
3460                 if (delta == 0)
3461                         delta = 1;
3462                 zfs_nicenum(bytes, buf1, sizeof (buf1));
3463                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3464 
3465                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3466                     buf1, delta, buf2);
3467         }
3468 
3469         return (0);
3470 }
3471 
3472 static int
3473 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap,
3474     const char *originsnap, recvflags_t *flags, int infd, const char *sendfs,
3475     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
3476     uint64_t *action_handlep, const char *finalsnap)
3477 {
3478         int err;
3479         dmu_replay_record_t drr, drr_noswap;
3480         struct drr_begin *drrb = &drr.drr_u.drr_begin;
3481         char errbuf[1024];
3482         zio_cksum_t zcksum = { 0 };
3483         uint64_t featureflags;
3484         int hdrtype;
3485 
3486         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3487             "cannot receive"));
3488 
3489         if (flags->isprefix &&
3490             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3491                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3492                     "(%s) does not exist"), tosnap);
3493                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3494         }
3495         if (originsnap &&
3496             !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) {
3497                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs "
3498                     "(%s) does not exist"), originsnap);
3499                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3500         }
3501 
3502         /* read in the BEGIN record */
3503         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3504             &zcksum)))
3505                 return (err);
3506 
3507         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3508                 /* It's the double end record at the end of a package */
3509                 return (ENODATA);
3510         }
3511 
3512         /* the kernel needs the non-byteswapped begin record */
3513         drr_noswap = drr;
3514 
3515         flags->byteswap = B_FALSE;
3516         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3517                 /*
3518                  * We computed the checksum in the wrong byteorder in
3519                  * recv_read() above; do it again correctly.
3520                  */
3521                 bzero(&zcksum, sizeof (zio_cksum_t));
3522                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3523                 flags->byteswap = B_TRUE;
3524 
3525                 drr.drr_type = BSWAP_32(drr.drr_type);
3526                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3527                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3528                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3529                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3530                 drrb->drr_type = BSWAP_32(drrb->drr_type);
3531                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3532                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3533                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3534         }
3535 
3536         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3537                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3538                     "stream (bad magic number)"));
3539                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3540         }
3541 
3542         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3543         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3544 
3545         if (!DMU_STREAM_SUPPORTED(featureflags) ||
3546             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3547                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3548                     "stream has unsupported feature, feature flags = %lx"),
3549                     featureflags);
3550                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3551         }
3552 
3553         if (strchr(drrb->drr_toname, '@') == NULL) {
3554                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3555                     "stream (bad snapshot name)"));
3556                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3557         }
3558 
3559         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3560                 char nonpackage_sendfs[ZFS_MAXNAMELEN];
3561                 if (sendfs == NULL) {
3562                         /*
3563                          * We were not called from zfs_receive_package(). Get
3564                          * the fs specified by 'zfs send'.
3565                          */
3566                         char *cp;
3567                         (void) strlcpy(nonpackage_sendfs,
3568                             drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
3569                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3570                                 *cp = '\0';
3571                         sendfs = nonpackage_sendfs;
3572                         VERIFY(finalsnap == NULL);
3573                 }
3574                 return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags,
3575                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs,
3576                     cleanup_fd, action_handlep, finalsnap));
3577         } else {
3578                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3579                     DMU_COMPOUNDSTREAM);
3580                 return (zfs_receive_package(hdl, infd, tosnap, flags, &drr,
3581                     &zcksum, top_zfs, cleanup_fd, action_handlep));
3582         }
3583 }
3584 
3585 /*
3586  * Restores a backup of tosnap from the file descriptor specified by infd.
3587  * Return 0 on total success, -2 if some things couldn't be
3588  * destroyed/renamed/promoted, -1 if some things couldn't be received.
3589  * (-1 will override -2, if -1 and the resumable flag was specified the
3590  * transfer can be resumed if the sending side supports it).
3591  */
3592 int
3593 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props,
3594     recvflags_t *flags, int infd, avl_tree_t *stream_avl)
3595 {
3596         char *top_zfs = NULL;
3597         int err;
3598         int cleanup_fd;
3599         uint64_t action_handle = 0;
3600         char *originsnap = NULL;
3601         if (props) {
3602                 err = nvlist_lookup_string(props, "origin", &originsnap);
3603                 if (err && err != ENOENT)
3604                         return (err);
3605         }
3606 
3607         cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3608         VERIFY(cleanup_fd >= 0);
3609 
3610         err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL,
3611             stream_avl, &top_zfs, cleanup_fd, &action_handle, NULL);
3612 
3613         VERIFY(0 == close(cleanup_fd));
3614 
3615         if (err == 0 && !flags->nomount && top_zfs) {
3616                 zfs_handle_t *zhp;
3617                 prop_changelist_t *clp;
3618 
3619                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3620                 if (zhp != NULL) {
3621                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3622                             CL_GATHER_MOUNT_ALWAYS, 0);
3623                         zfs_close(zhp);
3624                         if (clp != NULL) {
3625                                 /* mount and share received datasets */
3626                                 err = changelist_postfix(clp);
3627                                 changelist_free(clp);
3628                         }
3629                 }
3630                 if (zhp == NULL || clp == NULL || err)
3631                         err = -1;
3632         }
3633         if (top_zfs)
3634                 free(top_zfs);
3635 
3636         return (err);
3637 }