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, 2014 by Delphix. All rights reserved.
  25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  27  */
  28 
  29 #include <assert.h>
  30 #include <ctype.h>
  31 #include <errno.h>
  32 #include <libintl.h>
  33 #include <stdio.h>
  34 #include <stdlib.h>
  35 #include <strings.h>
  36 #include <unistd.h>
  37 #include <stddef.h>
  38 #include <fcntl.h>
  39 #include <sys/mount.h>
  40 #include <pthread.h>
  41 #include <umem.h>
  42 #include <time.h>
  43 
  44 #include <libzfs.h>
  45 #include <libzfs_core.h>
  46 
  47 #include "zfs_namecheck.h"
  48 #include "zfs_prop.h"
  49 #include "zfs_fletcher.h"
  50 #include "libzfs_impl.h"
  51 #include <sha2.h>
  52 #include <sys/zio_checksum.h>
  53 #include <sys/ddt.h>
  54 
  55 /* in libzfs_dataset.c */
  56 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  57 
  58 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t *,
  59     int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *);
  60 
  61 static const zio_cksum_t zero_cksum = { 0 };
  62 
  63 typedef struct dedup_arg {
  64         int     inputfd;
  65         int     outputfd;
  66         libzfs_handle_t  *dedup_hdl;
  67 } dedup_arg_t;
  68 
  69 typedef struct progress_arg {
  70         zfs_handle_t *pa_zhp;
  71         int pa_fd;
  72         boolean_t pa_parsable;
  73 } progress_arg_t;
  74 
  75 typedef struct dataref {
  76         uint64_t ref_guid;
  77         uint64_t ref_object;
  78         uint64_t ref_offset;
  79 } dataref_t;
  80 
  81 typedef struct dedup_entry {
  82         struct dedup_entry      *dde_next;
  83         zio_cksum_t dde_chksum;
  84         uint64_t dde_prop;
  85         dataref_t dde_ref;
  86 } dedup_entry_t;
  87 
  88 #define MAX_DDT_PHYSMEM_PERCENT         20
  89 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
  90 
  91 typedef struct dedup_table {
  92         dedup_entry_t   **dedup_hash_array;
  93         umem_cache_t    *ddecache;
  94         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
  95         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
  96         uint64_t        ddt_count;
  97         int             numhashbits;
  98         boolean_t       ddt_full;
  99 } dedup_table_t;
 100 
 101 static int
 102 high_order_bit(uint64_t n)
 103 {
 104         int count;
 105 
 106         for (count = 0; n != 0; count++)
 107                 n >>= 1;
 108         return (count);
 109 }
 110 
 111 static size_t
 112 ssread(void *buf, size_t len, FILE *stream)
 113 {
 114         size_t outlen;
 115 
 116         if ((outlen = fread(buf, len, 1, stream)) == 0)
 117                 return (0);
 118 
 119         return (outlen);
 120 }
 121 
 122 static void
 123 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 124     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 125 {
 126         dedup_entry_t   *dde;
 127 
 128         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 129                 if (ddt->ddt_full == B_FALSE) {
 130                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 131                             "Dedup table full.  Deduplication will continue "
 132                             "with existing table entries"));
 133                         ddt->ddt_full = B_TRUE;
 134                 }
 135                 return;
 136         }
 137 
 138         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 139             != NULL) {
 140                 assert(*ddepp == NULL);
 141                 dde->dde_next = NULL;
 142                 dde->dde_chksum = *cs;
 143                 dde->dde_prop = prop;
 144                 dde->dde_ref = *dr;
 145                 *ddepp = dde;
 146                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 147                 ddt->ddt_count++;
 148         }
 149 }
 150 
 151 /*
 152  * Using the specified dedup table, do a lookup for an entry with
 153  * the checksum cs.  If found, return the block's reference info
 154  * in *dr. Otherwise, insert a new entry in the dedup table, using
 155  * the reference information specified by *dr.
 156  *
 157  * return value:  true - entry was found
 158  *                false - entry was not found
 159  */
 160 static boolean_t
 161 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 162     uint64_t prop, dataref_t *dr)
 163 {
 164         uint32_t hashcode;
 165         dedup_entry_t **ddepp;
 166 
 167         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 168 
 169         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 170             ddepp = &((*ddepp)->dde_next)) {
 171                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 172                     (*ddepp)->dde_prop == prop) {
 173                         *dr = (*ddepp)->dde_ref;
 174                         return (B_TRUE);
 175                 }
 176         }
 177         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 178         return (B_FALSE);
 179 }
 180 
 181 static int
 182 dump_record(dmu_replay_record_t *drr, void *payload, int payload_len,
 183     zio_cksum_t *zc, int outfd)
 184 {
 185         ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
 186             ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t));
 187         fletcher_4_incremental_native(drr,
 188             offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc);
 189         if (drr->drr_type != DRR_BEGIN) {
 190                 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u.
 191                     drr_checksum.drr_checksum));
 192                 drr->drr_u.drr_checksum.drr_checksum = *zc;
 193         }
 194         fletcher_4_incremental_native(&drr->drr_u.drr_checksum.drr_checksum,
 195             sizeof (zio_cksum_t), zc);
 196         if (write(outfd, drr, sizeof (*drr)) == -1)
 197                 return (errno);
 198         if (payload_len != 0) {
 199                 fletcher_4_incremental_native(payload, payload_len, zc);
 200                 if (write(outfd, payload, payload_len) == -1)
 201                         return (errno);
 202         }
 203         return (0);
 204 }
 205 
 206 /*
 207  * This function is started in a separate thread when the dedup option
 208  * has been requested.  The main send thread determines the list of
 209  * snapshots to be included in the send stream and makes the ioctl calls
 210  * for each one.  But instead of having the ioctl send the output to the
 211  * the output fd specified by the caller of zfs_send()), the
 212  * ioctl is told to direct the output to a pipe, which is read by the
 213  * alternate thread running THIS function.  This function does the
 214  * dedup'ing by:
 215  *  1. building a dedup table (the DDT)
 216  *  2. doing checksums on each data block and inserting a record in the DDT
 217  *  3. looking for matching checksums, and
 218  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 219  *      a duplicate block is found.
 220  * The output of this function then goes to the output fd requested
 221  * by the caller of zfs_send().
 222  */
 223 static void *
 224 cksummer(void *arg)
 225 {
 226         dedup_arg_t *dda = arg;
 227         char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE);
 228         dmu_replay_record_t thedrr;
 229         dmu_replay_record_t *drr = &thedrr;
 230         FILE *ofp;
 231         int outfd;
 232         dedup_table_t ddt;
 233         zio_cksum_t stream_cksum;
 234         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 235         uint64_t numbuckets;
 236 
 237         ddt.max_ddt_size =
 238             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100,
 239             SMALLEST_POSSIBLE_MAX_DDT_MB << 20);
 240 
 241         numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t));
 242 
 243         /*
 244          * numbuckets must be a power of 2.  Increase number to
 245          * a power of 2 if necessary.
 246          */
 247         if (!ISP2(numbuckets))
 248                 numbuckets = 1 << high_order_bit(numbuckets);
 249 
 250         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 251         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 252             NULL, NULL, NULL, NULL, NULL, 0);
 253         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 254         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 255         ddt.ddt_full = B_FALSE;
 256 
 257         outfd = dda->outputfd;
 258         ofp = fdopen(dda->inputfd, "r");
 259         while (ssread(drr, sizeof (*drr), ofp) != 0) {
 260 
 261                 switch (drr->drr_type) {
 262                 case DRR_BEGIN:
 263                 {
 264                         struct drr_begin *drrb = &drr->drr_u.drr_begin;
 265                         int fflags;
 266                         int sz = 0;
 267                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 268 
 269                         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 270 
 271                         /* set the DEDUP feature flag for this stream */
 272                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 273                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 274                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 275                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 276 
 277                         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 278                             DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
 279                                 sz = drr->drr_payloadlen;
 280 
 281                                 if (sz > SPA_MAXBLOCKSIZE) {
 282                                         buf = zfs_realloc(dda->dedup_hdl, buf,
 283                                             SPA_MAXBLOCKSIZE, sz);
 284                                 }
 285                                 (void) ssread(buf, sz, ofp);
 286                                 if (ferror(stdin))
 287                                         perror("fread");
 288                         }
 289                         if (dump_record(drr, buf, sz, &stream_cksum,
 290                             outfd) != 0)
 291                                 goto out;
 292                         break;
 293                 }
 294 
 295                 case DRR_END:
 296                 {
 297                         struct drr_end *drre = &drr->drr_u.drr_end;
 298                         /* use the recalculated checksum */
 299                         drre->drr_checksum = stream_cksum;
 300                         if (dump_record(drr, NULL, 0, &stream_cksum,
 301                             outfd) != 0)
 302                                 goto out;
 303                         break;
 304                 }
 305 
 306                 case DRR_OBJECT:
 307                 {
 308                         struct drr_object *drro = &drr->drr_u.drr_object;
 309                         if (drro->drr_bonuslen > 0) {
 310                                 (void) ssread(buf,
 311                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 312                                     ofp);
 313                         }
 314                         if (dump_record(drr, buf,
 315                             P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 316                             &stream_cksum, outfd) != 0)
 317                                 goto out;
 318                         break;
 319                 }
 320 
 321                 case DRR_SPILL:
 322                 {
 323                         struct drr_spill *drrs = &drr->drr_u.drr_spill;
 324                         (void) ssread(buf, drrs->drr_length, ofp);
 325                         if (dump_record(drr, buf, drrs->drr_length,
 326                             &stream_cksum, outfd) != 0)
 327                                 goto out;
 328                         break;
 329                 }
 330 
 331                 case DRR_FREEOBJECTS:
 332                 {
 333                         if (dump_record(drr, NULL, 0, &stream_cksum,
 334                             outfd) != 0)
 335                                 goto out;
 336                         break;
 337                 }
 338 
 339                 case DRR_WRITE:
 340                 {
 341                         struct drr_write *drrw = &drr->drr_u.drr_write;
 342                         dataref_t       dataref;
 343 
 344                         (void) ssread(buf, drrw->drr_length, ofp);
 345 
 346                         /*
 347                          * Use the existing checksum if it's dedup-capable,
 348                          * else calculate a SHA256 checksum for it.
 349                          */
 350 
 351                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 352                             zero_cksum) ||
 353                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 354                                 SHA256_CTX      ctx;
 355                                 zio_cksum_t     tmpsha256;
 356 
 357                                 SHA256Init(&ctx);
 358                                 SHA256Update(&ctx, buf, drrw->drr_length);
 359                                 SHA256Final(&tmpsha256, &ctx);
 360                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 361                                     BE_64(tmpsha256.zc_word[0]);
 362                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 363                                     BE_64(tmpsha256.zc_word[1]);
 364                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 365                                     BE_64(tmpsha256.zc_word[2]);
 366                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 367                                     BE_64(tmpsha256.zc_word[3]);
 368                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 369                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 370                         }
 371 
 372                         dataref.ref_guid = drrw->drr_toguid;
 373                         dataref.ref_object = drrw->drr_object;
 374                         dataref.ref_offset = drrw->drr_offset;
 375 
 376                         if (ddt_update(dda->dedup_hdl, &ddt,
 377                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 378                             &dataref)) {
 379                                 dmu_replay_record_t wbr_drr = {0};
 380                                 struct drr_write_byref *wbr_drrr =
 381                                     &wbr_drr.drr_u.drr_write_byref;
 382 
 383                                 /* block already present in stream */
 384                                 wbr_drr.drr_type = DRR_WRITE_BYREF;
 385 
 386                                 wbr_drrr->drr_object = drrw->drr_object;
 387                                 wbr_drrr->drr_offset = drrw->drr_offset;
 388                                 wbr_drrr->drr_length = drrw->drr_length;
 389                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 390                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 391                                 wbr_drrr->drr_refobject =
 392                                     dataref.ref_object;
 393                                 wbr_drrr->drr_refoffset =
 394                                     dataref.ref_offset;
 395 
 396                                 wbr_drrr->drr_checksumtype =
 397                                     drrw->drr_checksumtype;
 398                                 wbr_drrr->drr_checksumflags =
 399                                     drrw->drr_checksumtype;
 400                                 wbr_drrr->drr_key.ddk_cksum =
 401                                     drrw->drr_key.ddk_cksum;
 402                                 wbr_drrr->drr_key.ddk_prop =
 403                                     drrw->drr_key.ddk_prop;
 404 
 405                                 if (dump_record(&wbr_drr, NULL, 0,
 406                                     &stream_cksum, outfd) != 0)
 407                                         goto out;
 408                         } else {
 409                                 /* block not previously seen */
 410                                 if (dump_record(drr, buf, drrw->drr_length,
 411                                     &stream_cksum, outfd) != 0)
 412                                         goto out;
 413                         }
 414                         break;
 415                 }
 416 
 417                 case DRR_WRITE_EMBEDDED:
 418                 {
 419                         struct drr_write_embedded *drrwe =
 420                             &drr->drr_u.drr_write_embedded;
 421                         (void) ssread(buf,
 422                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp);
 423                         if (dump_record(drr, buf,
 424                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8),
 425                             &stream_cksum, outfd) != 0)
 426                                 goto out;
 427                         break;
 428                 }
 429 
 430                 case DRR_FREE:
 431                 {
 432                         if (dump_record(drr, NULL, 0, &stream_cksum,
 433                             outfd) != 0)
 434                                 goto out;
 435                         break;
 436                 }
 437 
 438                 default:
 439                         (void) fprintf(stderr, "INVALID record type 0x%x\n",
 440                             drr->drr_type);
 441                         /* should never happen, so assert */
 442                         assert(B_FALSE);
 443                 }
 444         }
 445 out:
 446         umem_cache_destroy(ddt.ddecache);
 447         free(ddt.dedup_hash_array);
 448         free(buf);
 449         (void) fclose(ofp);
 450 
 451         return (NULL);
 452 }
 453 
 454 /*
 455  * Routines for dealing with the AVL tree of fs-nvlists
 456  */
 457 typedef struct fsavl_node {
 458         avl_node_t fn_node;
 459         nvlist_t *fn_nvfs;
 460         char *fn_snapname;
 461         uint64_t fn_guid;
 462 } fsavl_node_t;
 463 
 464 static int
 465 fsavl_compare(const void *arg1, const void *arg2)
 466 {
 467         const fsavl_node_t *fn1 = arg1;
 468         const fsavl_node_t *fn2 = arg2;
 469 
 470         if (fn1->fn_guid > fn2->fn_guid)
 471                 return (+1);
 472         else if (fn1->fn_guid < fn2->fn_guid)
 473                 return (-1);
 474         else
 475                 return (0);
 476 }
 477 
 478 /*
 479  * Given the GUID of a snapshot, find its containing filesystem and
 480  * (optionally) name.
 481  */
 482 static nvlist_t *
 483 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 484 {
 485         fsavl_node_t fn_find;
 486         fsavl_node_t *fn;
 487 
 488         fn_find.fn_guid = snapguid;
 489 
 490         fn = avl_find(avl, &fn_find, NULL);
 491         if (fn) {
 492                 if (snapname)
 493                         *snapname = fn->fn_snapname;
 494                 return (fn->fn_nvfs);
 495         }
 496         return (NULL);
 497 }
 498 
 499 static void
 500 fsavl_destroy(avl_tree_t *avl)
 501 {
 502         fsavl_node_t *fn;
 503         void *cookie;
 504 
 505         if (avl == NULL)
 506                 return;
 507 
 508         cookie = NULL;
 509         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 510                 free(fn);
 511         avl_destroy(avl);
 512         free(avl);
 513 }
 514 
 515 /*
 516  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 517  */
 518 static avl_tree_t *
 519 fsavl_create(nvlist_t *fss)
 520 {
 521         avl_tree_t *fsavl;
 522         nvpair_t *fselem = NULL;
 523 
 524         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 525                 return (NULL);
 526 
 527         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 528             offsetof(fsavl_node_t, fn_node));
 529 
 530         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 531                 nvlist_t *nvfs, *snaps;
 532                 nvpair_t *snapelem = NULL;
 533 
 534                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 535                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 536 
 537                 while ((snapelem =
 538                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 539                         fsavl_node_t *fn;
 540                         uint64_t guid;
 541 
 542                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 543                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 544                                 fsavl_destroy(fsavl);
 545                                 return (NULL);
 546                         }
 547                         fn->fn_nvfs = nvfs;
 548                         fn->fn_snapname = nvpair_name(snapelem);
 549                         fn->fn_guid = guid;
 550 
 551                         /*
 552                          * Note: if there are multiple snaps with the
 553                          * same GUID, we ignore all but one.
 554                          */
 555                         if (avl_find(fsavl, fn, NULL) == NULL)
 556                                 avl_add(fsavl, fn);
 557                         else
 558                                 free(fn);
 559                 }
 560         }
 561 
 562         return (fsavl);
 563 }
 564 
 565 /*
 566  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 567  */
 568 typedef struct send_data {
 569         uint64_t parent_fromsnap_guid;
 570         nvlist_t *parent_snaps;
 571         nvlist_t *fss;
 572         nvlist_t *snapprops;
 573         const char *fromsnap;
 574         const char *tosnap;
 575         boolean_t recursive;
 576 
 577         /*
 578          * The header nvlist is of the following format:
 579          * {
 580          *   "tosnap" -> string
 581          *   "fromsnap" -> string (if incremental)
 582          *   "fss" -> {
 583          *      id -> {
 584          *
 585          *       "name" -> string (full name; for debugging)
 586          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 587          *
 588          *       "props" -> { name -> value (only if set here) }
 589          *       "snaps" -> { name (lastname) -> number (guid) }
 590          *       "snapprops" -> { name (lastname) -> { name -> value } }
 591          *
 592          *       "origin" -> number (guid) (if clone)
 593          *       "sent" -> boolean (not on-disk)
 594          *      }
 595          *   }
 596          * }
 597          *
 598          */
 599 } send_data_t;
 600 
 601 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 602 
 603 static int
 604 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 605 {
 606         send_data_t *sd = arg;
 607         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 608         char *snapname;
 609         nvlist_t *nv;
 610 
 611         snapname = strrchr(zhp->zfs_name, '@')+1;
 612 
 613         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 614         /*
 615          * NB: if there is no fromsnap here (it's a newly created fs in
 616          * an incremental replication), we will substitute the tosnap.
 617          */
 618         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 619             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 620             strcmp(snapname, sd->tosnap) == 0)) {
 621                 sd->parent_fromsnap_guid = guid;
 622         }
 623 
 624         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 625         send_iterate_prop(zhp, nv);
 626         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 627         nvlist_free(nv);
 628 
 629         zfs_close(zhp);
 630         return (0);
 631 }
 632 
 633 static void
 634 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 635 {
 636         nvpair_t *elem = NULL;
 637 
 638         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 639                 char *propname = nvpair_name(elem);
 640                 zfs_prop_t prop = zfs_name_to_prop(propname);
 641                 nvlist_t *propnv;
 642 
 643                 if (!zfs_prop_user(propname)) {
 644                         /*
 645                          * Realistically, this should never happen.  However,
 646                          * we want the ability to add DSL properties without
 647                          * needing to make incompatible version changes.  We
 648                          * need to ignore unknown properties to allow older
 649                          * software to still send datasets containing these
 650                          * properties, with the unknown properties elided.
 651                          */
 652                         if (prop == ZPROP_INVAL)
 653                                 continue;
 654 
 655                         if (zfs_prop_readonly(prop))
 656                                 continue;
 657                 }
 658 
 659                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 660                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 661                     prop == ZFS_PROP_REFQUOTA ||
 662                     prop == ZFS_PROP_REFRESERVATION) {
 663                         char *source;
 664                         uint64_t value;
 665                         verify(nvlist_lookup_uint64(propnv,
 666                             ZPROP_VALUE, &value) == 0);
 667                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 668                                 continue;
 669                         /*
 670                          * May have no source before SPA_VERSION_RECVD_PROPS,
 671                          * but is still modifiable.
 672                          */
 673                         if (nvlist_lookup_string(propnv,
 674                             ZPROP_SOURCE, &source) == 0) {
 675                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 676                                     (strcmp(source,
 677                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 678                                         continue;
 679                         }
 680                 } else {
 681                         char *source;
 682                         if (nvlist_lookup_string(propnv,
 683                             ZPROP_SOURCE, &source) != 0)
 684                                 continue;
 685                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 686                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 687                                 continue;
 688                 }
 689 
 690                 if (zfs_prop_user(propname) ||
 691                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 692                         char *value;
 693                         verify(nvlist_lookup_string(propnv,
 694                             ZPROP_VALUE, &value) == 0);
 695                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 696                 } else {
 697                         uint64_t value;
 698                         verify(nvlist_lookup_uint64(propnv,
 699                             ZPROP_VALUE, &value) == 0);
 700                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 701                 }
 702         }
 703 }
 704 
 705 /*
 706  * recursively generate nvlists describing datasets.  See comment
 707  * for the data structure send_data_t above for description of contents
 708  * of the nvlist.
 709  */
 710 static int
 711 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 712 {
 713         send_data_t *sd = arg;
 714         nvlist_t *nvfs, *nv;
 715         int rv = 0;
 716         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 717         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 718         char guidstring[64];
 719 
 720         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 721         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 722         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 723             sd->parent_fromsnap_guid));
 724 
 725         if (zhp->zfs_dmustats.dds_origin[0]) {
 726                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 727                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 728                 if (origin == NULL)
 729                         return (-1);
 730                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 731                     origin->zfs_dmustats.dds_guid));
 732         }
 733 
 734         /* iterate over props */
 735         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 736         send_iterate_prop(zhp, nv);
 737         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 738         nvlist_free(nv);
 739 
 740         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 741         sd->parent_fromsnap_guid = 0;
 742         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 743         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 744         (void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
 745         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 746         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 747         nvlist_free(sd->parent_snaps);
 748         nvlist_free(sd->snapprops);
 749 
 750         /* add this fs to nvlist */
 751         (void) snprintf(guidstring, sizeof (guidstring),
 752             "0x%llx", (longlong_t)guid);
 753         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 754         nvlist_free(nvfs);
 755 
 756         /* iterate over children */
 757         if (sd->recursive)
 758                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 759 
 760         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 761 
 762         zfs_close(zhp);
 763         return (rv);
 764 }
 765 
 766 static int
 767 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 768     const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
 769 {
 770         zfs_handle_t *zhp;
 771         send_data_t sd = { 0 };
 772         int error;
 773 
 774         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 775         if (zhp == NULL)
 776                 return (EZFS_BADTYPE);
 777 
 778         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 779         sd.fromsnap = fromsnap;
 780         sd.tosnap = tosnap;
 781         sd.recursive = recursive;
 782 
 783         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 784                 nvlist_free(sd.fss);
 785                 if (avlp != NULL)
 786                         *avlp = NULL;
 787                 *nvlp = NULL;
 788                 return (error);
 789         }
 790 
 791         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 792                 nvlist_free(sd.fss);
 793                 *nvlp = NULL;
 794                 return (EZFS_NOMEM);
 795         }
 796 
 797         *nvlp = sd.fss;
 798         return (0);
 799 }
 800 
 801 /*
 802  * Routines specific to "zfs send"
 803  */
 804 typedef struct send_dump_data {
 805         /* these are all just the short snapname (the part after the @) */
 806         const char *fromsnap;
 807         const char *tosnap;
 808         char prevsnap[ZFS_MAXNAMELEN];
 809         uint64_t prevsnap_obj;
 810         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 811         boolean_t verbose, dryrun, parsable, progress, embed_data, std_out;
 812         boolean_t large_block;
 813         int outfd;
 814         boolean_t err;
 815         nvlist_t *fss;
 816         nvlist_t *snapholds;
 817         avl_tree_t *fsavl;
 818         snapfilter_cb_t *filter_cb;
 819         void *filter_cb_arg;
 820         nvlist_t *debugnv;
 821         char holdtag[ZFS_MAXNAMELEN];
 822         int cleanup_fd;
 823         uint64_t size;
 824 } send_dump_data_t;
 825 
 826 static int
 827 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
 828     boolean_t fromorigin, uint64_t *sizep)
 829 {
 830         zfs_cmd_t zc = { 0 };
 831         libzfs_handle_t *hdl = zhp->zfs_hdl;
 832 
 833         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 834         assert(fromsnap_obj == 0 || !fromorigin);
 835 
 836         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 837         zc.zc_obj = fromorigin;
 838         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 839         zc.zc_fromobj = fromsnap_obj;
 840         zc.zc_guid = 1;  /* estimate flag */
 841 
 842         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 843                 char errbuf[1024];
 844                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 845                     "warning: cannot estimate space for '%s'"), zhp->zfs_name);
 846 
 847                 switch (errno) {
 848                 case EXDEV:
 849                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 850                             "not an earlier snapshot from the same fs"));
 851                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 852 
 853                 case ENOENT:
 854                         if (zfs_dataset_exists(hdl, zc.zc_name,
 855                             ZFS_TYPE_SNAPSHOT)) {
 856                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 857                                     "incremental source (@%s) does not exist"),
 858                                     zc.zc_value);
 859                         }
 860                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 861 
 862                 case EDQUOT:
 863                 case EFBIG:
 864                 case EIO:
 865                 case ENOLINK:
 866                 case ENOSPC:
 867                 case ENOSTR:
 868                 case ENXIO:
 869                 case EPIPE:
 870                 case ERANGE:
 871                 case EFAULT:
 872                 case EROFS:
 873                         zfs_error_aux(hdl, strerror(errno));
 874                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 875 
 876                 default:
 877                         return (zfs_standard_error(hdl, errno, errbuf));
 878                 }
 879         }
 880 
 881         *sizep = zc.zc_objset_type;
 882 
 883         return (0);
 884 }
 885 
 886 /*
 887  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 888  * NULL) to the file descriptor specified by outfd.
 889  */
 890 static int
 891 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
 892     boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
 893     nvlist_t *debugnv)
 894 {
 895         zfs_cmd_t zc = { 0 };
 896         libzfs_handle_t *hdl = zhp->zfs_hdl;
 897         nvlist_t *thisdbg;
 898 
 899         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 900         assert(fromsnap_obj == 0 || !fromorigin);
 901 
 902         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 903         zc.zc_cookie = outfd;
 904         zc.zc_obj = fromorigin;
 905         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 906         zc.zc_fromobj = fromsnap_obj;
 907         zc.zc_flags = flags;
 908 
 909         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 910         if (fromsnap && fromsnap[0] != '\0') {
 911                 VERIFY(0 == nvlist_add_string(thisdbg,
 912                     "fromsnap", fromsnap));
 913         }
 914 
 915         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 916                 char errbuf[1024];
 917                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 918                     "warning: cannot send '%s'"), zhp->zfs_name);
 919 
 920                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
 921                 if (debugnv) {
 922                         VERIFY(0 == nvlist_add_nvlist(debugnv,
 923                             zhp->zfs_name, thisdbg));
 924                 }
 925                 nvlist_free(thisdbg);
 926 
 927                 switch (errno) {
 928                 case EXDEV:
 929                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 930                             "not an earlier snapshot from the same fs"));
 931                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 932 
 933                 case ENOENT:
 934                         if (zfs_dataset_exists(hdl, zc.zc_name,
 935                             ZFS_TYPE_SNAPSHOT)) {
 936                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 937                                     "incremental source (@%s) does not exist"),
 938                                     zc.zc_value);
 939                         }
 940                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 941 
 942                 case EDQUOT:
 943                 case EFBIG:
 944                 case EIO:
 945                 case ENOLINK:
 946                 case ENOSPC:
 947                 case ENOSTR:
 948                 case ENXIO:
 949                 case EPIPE:
 950                 case ERANGE:
 951                 case EFAULT:
 952                 case EROFS:
 953                         zfs_error_aux(hdl, strerror(errno));
 954                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 955 
 956                 default:
 957                         return (zfs_standard_error(hdl, errno, errbuf));
 958                 }
 959         }
 960 
 961         if (debugnv)
 962                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
 963         nvlist_free(thisdbg);
 964 
 965         return (0);
 966 }
 967 
 968 static void
 969 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
 970 {
 971         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 972 
 973         /*
 974          * zfs_send() only sets snapholds for sends that need them,
 975          * e.g. replication and doall.
 976          */
 977         if (sdd->snapholds == NULL)
 978                 return;
 979 
 980         fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
 981 }
 982 
 983 static void *
 984 send_progress_thread(void *arg)
 985 {
 986         progress_arg_t *pa = arg;
 987 
 988         zfs_cmd_t zc = { 0 };
 989         zfs_handle_t *zhp = pa->pa_zhp;
 990         libzfs_handle_t *hdl = zhp->zfs_hdl;
 991         unsigned long long bytes;
 992         char buf[16];
 993 
 994         time_t t;
 995         struct tm *tm;
 996 
 997         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 998         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 999 
1000         if (!pa->pa_parsable)
1001                 (void) fprintf(stderr, "TIME        SENT   SNAPSHOT\n");
1002 
1003         /*
1004          * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1005          */
1006         for (;;) {
1007                 (void) sleep(1);
1008 
1009                 zc.zc_cookie = pa->pa_fd;
1010                 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1011                         return ((void *)-1);
1012 
1013                 (void) time(&t);
1014                 tm = localtime(&t);
1015                 bytes = zc.zc_cookie;
1016 
1017                 if (pa->pa_parsable) {
1018                         (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1019                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1020                             bytes, zhp->zfs_name);
1021                 } else {
1022                         zfs_nicenum(bytes, buf, sizeof (buf));
1023                         (void) fprintf(stderr, "%02d:%02d:%02d   %5s   %s\n",
1024                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1025                             buf, zhp->zfs_name);
1026                 }
1027         }
1028 }
1029 
1030 static int
1031 dump_snapshot(zfs_handle_t *zhp, void *arg)
1032 {
1033         send_dump_data_t *sdd = arg;
1034         progress_arg_t pa = { 0 };
1035         pthread_t tid;
1036         char *thissnap;
1037         int err;
1038         boolean_t isfromsnap, istosnap, fromorigin;
1039         boolean_t exclude = B_FALSE;
1040         FILE *fout = sdd->std_out ? stdout : stderr;
1041 
1042         err = 0;
1043         thissnap = strchr(zhp->zfs_name, '@') + 1;
1044         isfromsnap = (sdd->fromsnap != NULL &&
1045             strcmp(sdd->fromsnap, thissnap) == 0);
1046 
1047         if (!sdd->seenfrom && isfromsnap) {
1048                 gather_holds(zhp, sdd);
1049                 sdd->seenfrom = B_TRUE;
1050                 (void) strcpy(sdd->prevsnap, thissnap);
1051                 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1052                 zfs_close(zhp);
1053                 return (0);
1054         }
1055 
1056         if (sdd->seento || !sdd->seenfrom) {
1057                 zfs_close(zhp);
1058                 return (0);
1059         }
1060 
1061         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1062         if (istosnap)
1063                 sdd->seento = B_TRUE;
1064 
1065         if (!sdd->doall && !isfromsnap && !istosnap) {
1066                 if (sdd->replicate) {
1067                         char *snapname;
1068                         nvlist_t *snapprops;
1069                         /*
1070                          * Filter out all intermediate snapshots except origin
1071                          * snapshots needed to replicate clones.
1072                          */
1073                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1074                             zhp->zfs_dmustats.dds_guid, &snapname);
1075 
1076                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1077                             "snapprops", &snapprops));
1078                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1079                             thissnap, &snapprops));
1080                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
1081                 } else {
1082                         exclude = B_TRUE;
1083                 }
1084         }
1085 
1086         /*
1087          * If a filter function exists, call it to determine whether
1088          * this snapshot will be sent.
1089          */
1090         if (exclude || (sdd->filter_cb != NULL &&
1091             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1092                 /*
1093                  * This snapshot is filtered out.  Don't send it, and don't
1094                  * set prevsnap_obj, so it will be as if this snapshot didn't
1095                  * exist, and the next accepted snapshot will be sent as
1096                  * an incremental from the last accepted one, or as the
1097                  * first (and full) snapshot in the case of a replication,
1098                  * non-incremental send.
1099                  */
1100                 zfs_close(zhp);
1101                 return (0);
1102         }
1103 
1104         gather_holds(zhp, sdd);
1105         fromorigin = sdd->prevsnap[0] == '\0' &&
1106             (sdd->fromorigin || sdd->replicate);
1107 
1108         if (sdd->verbose) {
1109                 uint64_t size;
1110                 err = estimate_ioctl(zhp, sdd->prevsnap_obj,
1111                     fromorigin, &size);
1112 
1113                 if (sdd->parsable) {
1114                         if (sdd->prevsnap[0] != '\0') {
1115                                 (void) fprintf(fout, "incremental\t%s\t%s",
1116                                     sdd->prevsnap, zhp->zfs_name);
1117                         } else {
1118                                 (void) fprintf(fout, "full\t%s",
1119                                     zhp->zfs_name);
1120                         }
1121                 } else {
1122                         (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1123                             "send from @%s to %s"),
1124                             sdd->prevsnap, zhp->zfs_name);
1125                 }
1126                 if (err == 0) {
1127                         if (sdd->parsable) {
1128                                 (void) fprintf(fout, "\t%llu\n",
1129                                     (longlong_t)size);
1130                         } else {
1131                                 char buf[16];
1132                                 zfs_nicenum(size, buf, sizeof (buf));
1133                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1134                                     " estimated size is %s\n"), buf);
1135                         }
1136                         sdd->size += size;
1137                 } else {
1138                         (void) fprintf(fout, "\n");
1139                 }
1140         }
1141 
1142         if (!sdd->dryrun) {
1143                 /*
1144                  * If progress reporting is requested, spawn a new thread to
1145                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1146                  */
1147                 if (sdd->progress) {
1148                         pa.pa_zhp = zhp;
1149                         pa.pa_fd = sdd->outfd;
1150                         pa.pa_parsable = sdd->parsable;
1151 
1152                         if (err = pthread_create(&tid, NULL,
1153                             send_progress_thread, &pa)) {
1154                                 zfs_close(zhp);
1155                                 return (err);
1156                         }
1157                 }
1158 
1159                 enum lzc_send_flags flags = 0;
1160                 if (sdd->large_block)
1161                         flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1162                 if (sdd->embed_data)
1163                         flags |= LZC_SEND_FLAG_EMBED_DATA;
1164 
1165                 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1166                     fromorigin, sdd->outfd, flags, sdd->debugnv);
1167 
1168                 if (sdd->progress) {
1169                         (void) pthread_cancel(tid);
1170                         (void) pthread_join(tid, NULL);
1171                 }
1172         }
1173 
1174         (void) strcpy(sdd->prevsnap, thissnap);
1175         sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1176         zfs_close(zhp);
1177         return (err);
1178 }
1179 
1180 static int
1181 dump_filesystem(zfs_handle_t *zhp, void *arg)
1182 {
1183         int rv = 0;
1184         send_dump_data_t *sdd = arg;
1185         boolean_t missingfrom = B_FALSE;
1186         zfs_cmd_t zc = { 0 };
1187 
1188         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1189             zhp->zfs_name, sdd->tosnap);
1190         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1191                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1192                     "WARNING: could not send %s@%s: does not exist\n"),
1193                     zhp->zfs_name, sdd->tosnap);
1194                 sdd->err = B_TRUE;
1195                 return (0);
1196         }
1197 
1198         if (sdd->replicate && sdd->fromsnap) {
1199                 /*
1200                  * If this fs does not have fromsnap, and we're doing
1201                  * recursive, we need to send a full stream from the
1202                  * beginning (or an incremental from the origin if this
1203                  * is a clone).  If we're doing non-recursive, then let
1204                  * them get the error.
1205                  */
1206                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1207                     zhp->zfs_name, sdd->fromsnap);
1208                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1209                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1210                         missingfrom = B_TRUE;
1211                 }
1212         }
1213 
1214         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1215         sdd->prevsnap_obj = 0;
1216         if (sdd->fromsnap == NULL || missingfrom)
1217                 sdd->seenfrom = B_TRUE;
1218 
1219         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1220         if (!sdd->seenfrom) {
1221                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1222                     "WARNING: could not send %s@%s:\n"
1223                     "incremental source (%s@%s) does not exist\n"),
1224                     zhp->zfs_name, sdd->tosnap,
1225                     zhp->zfs_name, sdd->fromsnap);
1226                 sdd->err = B_TRUE;
1227         } else if (!sdd->seento) {
1228                 if (sdd->fromsnap) {
1229                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1230                             "WARNING: could not send %s@%s:\n"
1231                             "incremental source (%s@%s) "
1232                             "is not earlier than it\n"),
1233                             zhp->zfs_name, sdd->tosnap,
1234                             zhp->zfs_name, sdd->fromsnap);
1235                 } else {
1236                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1237                             "WARNING: "
1238                             "could not send %s@%s: does not exist\n"),
1239                             zhp->zfs_name, sdd->tosnap);
1240                 }
1241                 sdd->err = B_TRUE;
1242         }
1243 
1244         return (rv);
1245 }
1246 
1247 static int
1248 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1249 {
1250         send_dump_data_t *sdd = arg;
1251         nvpair_t *fspair;
1252         boolean_t needagain, progress;
1253 
1254         if (!sdd->replicate)
1255                 return (dump_filesystem(rzhp, sdd));
1256 
1257         /* Mark the clone origin snapshots. */
1258         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1259             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1260                 nvlist_t *nvfs;
1261                 uint64_t origin_guid = 0;
1262 
1263                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1264                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1265                 if (origin_guid != 0) {
1266                         char *snapname;
1267                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1268                             origin_guid, &snapname);
1269                         if (origin_nv != NULL) {
1270                                 nvlist_t *snapprops;
1271                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1272                                     "snapprops", &snapprops));
1273                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1274                                     snapname, &snapprops));
1275                                 VERIFY(0 == nvlist_add_boolean(
1276                                     snapprops, "is_clone_origin"));
1277                         }
1278                 }
1279         }
1280 again:
1281         needagain = progress = B_FALSE;
1282         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1283             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1284                 nvlist_t *fslist, *parent_nv;
1285                 char *fsname;
1286                 zfs_handle_t *zhp;
1287                 int err;
1288                 uint64_t origin_guid = 0;
1289                 uint64_t parent_guid = 0;
1290 
1291                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1292                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1293                         continue;
1294 
1295                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1296                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1297                 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1298                     &parent_guid);
1299 
1300                 if (parent_guid != 0) {
1301                         parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1302                         if (!nvlist_exists(parent_nv, "sent")) {
1303                                 /* parent has not been sent; skip this one */
1304                                 needagain = B_TRUE;
1305                                 continue;
1306                         }
1307                 }
1308 
1309                 if (origin_guid != 0) {
1310                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1311                             origin_guid, NULL);
1312                         if (origin_nv != NULL &&
1313                             !nvlist_exists(origin_nv, "sent")) {
1314                                 /*
1315                                  * origin has not been sent yet;
1316                                  * skip this clone.
1317                                  */
1318                                 needagain = B_TRUE;
1319                                 continue;
1320                         }
1321                 }
1322 
1323                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1324                 if (zhp == NULL)
1325                         return (-1);
1326                 err = dump_filesystem(zhp, sdd);
1327                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1328                 progress = B_TRUE;
1329                 zfs_close(zhp);
1330                 if (err)
1331                         return (err);
1332         }
1333         if (needagain) {
1334                 assert(progress);
1335                 goto again;
1336         }
1337 
1338         /* clean out the sent flags in case we reuse this fss */
1339         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1340             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1341                 nvlist_t *fslist;
1342 
1343                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1344                 (void) nvlist_remove_all(fslist, "sent");
1345         }
1346 
1347         return (0);
1348 }
1349 
1350 /*
1351  * Generate a send stream for the dataset identified by the argument zhp.
1352  *
1353  * The content of the send stream is the snapshot identified by
1354  * 'tosnap'.  Incremental streams are requested in two ways:
1355  *     - from the snapshot identified by "fromsnap" (if non-null) or
1356  *     - from the origin of the dataset identified by zhp, which must
1357  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1358  *       is TRUE.
1359  *
1360  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1361  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1362  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1363  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1364  * case too. If "props" is set, send properties.
1365  */
1366 int
1367 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1368     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1369     void *cb_arg, nvlist_t **debugnvp)
1370 {
1371         char errbuf[1024];
1372         send_dump_data_t sdd = { 0 };
1373         int err = 0;
1374         nvlist_t *fss = NULL;
1375         avl_tree_t *fsavl = NULL;
1376         static uint64_t holdseq;
1377         int spa_version;
1378         pthread_t tid = 0;
1379         int pipefd[2];
1380         dedup_arg_t dda = { 0 };
1381         int featureflags = 0;
1382         FILE *fout;
1383 
1384         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1385             "cannot send '%s'"), zhp->zfs_name);
1386 
1387         if (fromsnap && fromsnap[0] == '\0') {
1388                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1389                     "zero-length incremental source"));
1390                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1391         }
1392 
1393         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1394                 uint64_t version;
1395                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1396                 if (version >= ZPL_VERSION_SA) {
1397                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1398                 }
1399         }
1400 
1401         if (flags->dedup && !flags->dryrun) {
1402                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1403                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1404                 if (err = pipe(pipefd)) {
1405                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1406                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1407                             errbuf));
1408                 }
1409                 dda.outputfd = outfd;
1410                 dda.inputfd = pipefd[1];
1411                 dda.dedup_hdl = zhp->zfs_hdl;
1412                 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1413                         (void) close(pipefd[0]);
1414                         (void) close(pipefd[1]);
1415                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1416                         return (zfs_error(zhp->zfs_hdl,
1417                             EZFS_THREADCREATEFAILED, errbuf));
1418                 }
1419         }
1420 
1421         if (flags->replicate || flags->doall || flags->props) {
1422                 dmu_replay_record_t drr = { 0 };
1423                 char *packbuf = NULL;
1424                 size_t buflen = 0;
1425                 zio_cksum_t zc = { 0 };
1426 
1427                 if (flags->replicate || flags->props) {
1428                         nvlist_t *hdrnv;
1429 
1430                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1431                         if (fromsnap) {
1432                                 VERIFY(0 == nvlist_add_string(hdrnv,
1433                                     "fromsnap", fromsnap));
1434                         }
1435                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1436                         if (!flags->replicate) {
1437                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1438                                     "not_recursive"));
1439                         }
1440 
1441                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1442                             fromsnap, tosnap, flags->replicate, &fss, &fsavl);
1443                         if (err)
1444                                 goto err_out;
1445                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1446                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1447                             NV_ENCODE_XDR, 0);
1448                         if (debugnvp)
1449                                 *debugnvp = hdrnv;
1450                         else
1451                                 nvlist_free(hdrnv);
1452                         if (err)
1453                                 goto stderr_out;
1454                 }
1455 
1456                 if (!flags->dryrun) {
1457                         /* write first begin record */
1458                         drr.drr_type = DRR_BEGIN;
1459                         drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1460                         DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1461                             drr_versioninfo, DMU_COMPOUNDSTREAM);
1462                         DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1463                             drr_versioninfo, featureflags);
1464                         (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1465                             sizeof (drr.drr_u.drr_begin.drr_toname),
1466                             "%s@%s", zhp->zfs_name, tosnap);
1467                         drr.drr_payloadlen = buflen;
1468 
1469                         err = dump_record(&drr, packbuf, buflen, &zc, outfd);
1470                         free(packbuf);
1471                         if (err != 0)
1472                                 goto stderr_out;
1473 
1474                         /* write end record */
1475                         bzero(&drr, sizeof (drr));
1476                         drr.drr_type = DRR_END;
1477                         drr.drr_u.drr_end.drr_checksum = zc;
1478                         err = write(outfd, &drr, sizeof (drr));
1479                         if (err == -1) {
1480                                 err = errno;
1481                                 goto stderr_out;
1482                         }
1483 
1484                         err = 0;
1485                 }
1486         }
1487 
1488         /* dump each stream */
1489         sdd.fromsnap = fromsnap;
1490         sdd.tosnap = tosnap;
1491         if (tid != 0)
1492                 sdd.outfd = pipefd[0];
1493         else
1494                 sdd.outfd = outfd;
1495         sdd.replicate = flags->replicate;
1496         sdd.doall = flags->doall;
1497         sdd.fromorigin = flags->fromorigin;
1498         sdd.fss = fss;
1499         sdd.fsavl = fsavl;
1500         sdd.verbose = flags->verbose;
1501         sdd.parsable = flags->parsable;
1502         sdd.progress = flags->progress;
1503         sdd.dryrun = flags->dryrun;
1504         sdd.large_block = flags->largeblock;
1505         sdd.embed_data = flags->embed_data;
1506         sdd.filter_cb = filter_func;
1507         sdd.filter_cb_arg = cb_arg;
1508         if (debugnvp)
1509                 sdd.debugnv = *debugnvp;
1510         if (sdd.verbose && sdd.dryrun)
1511                 sdd.std_out = B_TRUE;
1512         fout = sdd.std_out ? stdout : stderr;
1513 
1514         /*
1515          * Some flags require that we place user holds on the datasets that are
1516          * being sent so they don't get destroyed during the send. We can skip
1517          * this step if the pool is imported read-only since the datasets cannot
1518          * be destroyed.
1519          */
1520         if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1521             ZPOOL_PROP_READONLY, NULL) &&
1522             zfs_spa_version(zhp, &spa_version) == 0 &&
1523             spa_version >= SPA_VERSION_USERREFS &&
1524             (flags->doall || flags->replicate)) {
1525                 ++holdseq;
1526                 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1527                     ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1528                 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1529                 if (sdd.cleanup_fd < 0) {
1530                         err = errno;
1531                         goto stderr_out;
1532                 }
1533                 sdd.snapholds = fnvlist_alloc();
1534         } else {
1535                 sdd.cleanup_fd = -1;
1536                 sdd.snapholds = NULL;
1537         }
1538         if (flags->verbose || sdd.snapholds != NULL) {
1539                 /*
1540                  * Do a verbose no-op dry run to get all the verbose output
1541                  * or to gather snapshot hold's before generating any data,
1542                  * then do a non-verbose real run to generate the streams.
1543                  */
1544                 sdd.dryrun = B_TRUE;
1545                 err = dump_filesystems(zhp, &sdd);
1546 
1547                 if (err != 0)
1548                         goto stderr_out;
1549 
1550                 if (flags->verbose) {
1551                         if (flags->parsable) {
1552                                 (void) fprintf(fout, "size\t%llu\n",
1553                                     (longlong_t)sdd.size);
1554                         } else {
1555                                 char buf[16];
1556                                 zfs_nicenum(sdd.size, buf, sizeof (buf));
1557                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1558                                     "total estimated size is %s\n"), buf);
1559                         }
1560                 }
1561 
1562                 /* Ensure no snaps found is treated as an error. */
1563                 if (!sdd.seento) {
1564                         err = ENOENT;
1565                         goto err_out;
1566                 }
1567 
1568                 /* Skip the second run if dryrun was requested. */
1569                 if (flags->dryrun)
1570                         goto err_out;
1571 
1572                 if (sdd.snapholds != NULL) {
1573                         err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
1574                         if (err != 0)
1575                                 goto stderr_out;
1576 
1577                         fnvlist_free(sdd.snapholds);
1578                         sdd.snapholds = NULL;
1579                 }
1580 
1581                 sdd.dryrun = B_FALSE;
1582                 sdd.verbose = B_FALSE;
1583         }
1584 
1585         err = dump_filesystems(zhp, &sdd);
1586         fsavl_destroy(fsavl);
1587         nvlist_free(fss);
1588 
1589         /* Ensure no snaps found is treated as an error. */
1590         if (err == 0 && !sdd.seento)
1591                 err = ENOENT;
1592 
1593         if (tid != 0) {
1594                 if (err != 0)
1595                         (void) pthread_cancel(tid);
1596                 (void) close(pipefd[0]);
1597                 (void) pthread_join(tid, NULL);
1598         }
1599 
1600         if (sdd.cleanup_fd != -1) {
1601                 VERIFY(0 == close(sdd.cleanup_fd));
1602                 sdd.cleanup_fd = -1;
1603         }
1604 
1605         if (!flags->dryrun && (flags->replicate || flags->doall ||
1606             flags->props)) {
1607                 /*
1608                  * write final end record.  NB: want to do this even if
1609                  * there was some error, because it might not be totally
1610                  * failed.
1611                  */
1612                 dmu_replay_record_t drr = { 0 };
1613                 drr.drr_type = DRR_END;
1614                 if (write(outfd, &drr, sizeof (drr)) == -1) {
1615                         return (zfs_standard_error(zhp->zfs_hdl,
1616                             errno, errbuf));
1617                 }
1618         }
1619 
1620         return (err || sdd.err);
1621 
1622 stderr_out:
1623         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1624 err_out:
1625         fsavl_destroy(fsavl);
1626         nvlist_free(fss);
1627         fnvlist_free(sdd.snapholds);
1628 
1629         if (sdd.cleanup_fd != -1)
1630                 VERIFY(0 == close(sdd.cleanup_fd));
1631         if (tid != 0) {
1632                 (void) pthread_cancel(tid);
1633                 (void) close(pipefd[0]);
1634                 (void) pthread_join(tid, NULL);
1635         }
1636         return (err);
1637 }
1638 
1639 int
1640 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd,
1641     enum lzc_send_flags flags)
1642 {
1643         int err;
1644         libzfs_handle_t *hdl = zhp->zfs_hdl;
1645 
1646         char errbuf[1024];
1647         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1648             "warning: cannot send '%s'"), zhp->zfs_name);
1649 
1650         err = lzc_send(zhp->zfs_name, from, fd, flags);
1651         if (err != 0) {
1652                 switch (errno) {
1653                 case EXDEV:
1654                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1655                             "not an earlier snapshot from the same fs"));
1656                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
1657 
1658                 case ENOENT:
1659                 case ESRCH:
1660                         if (lzc_exists(zhp->zfs_name)) {
1661                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1662                                     "incremental source (%s) does not exist"),
1663                                     from);
1664                         }
1665                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
1666 
1667                 case EBUSY:
1668                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1669                             "target is busy; if a filesystem, "
1670                             "it must not be mounted"));
1671                         return (zfs_error(hdl, EZFS_BUSY, errbuf));
1672 
1673                 case EDQUOT:
1674                 case EFBIG:
1675                 case EIO:
1676                 case ENOLINK:
1677                 case ENOSPC:
1678                 case ENOSTR:
1679                 case ENXIO:
1680                 case EPIPE:
1681                 case ERANGE:
1682                 case EFAULT:
1683                 case EROFS:
1684                         zfs_error_aux(hdl, strerror(errno));
1685                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1686 
1687                 default:
1688                         return (zfs_standard_error(hdl, errno, errbuf));
1689                 }
1690         }
1691         return (err != 0);
1692 }
1693 
1694 /*
1695  * Routines specific to "zfs recv"
1696  */
1697 
1698 static int
1699 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1700     boolean_t byteswap, zio_cksum_t *zc)
1701 {
1702         char *cp = buf;
1703         int rv;
1704         int len = ilen;
1705 
1706         assert(ilen <= SPA_MAXBLOCKSIZE);
1707 
1708         do {
1709                 rv = read(fd, cp, len);
1710                 cp += rv;
1711                 len -= rv;
1712         } while (rv > 0);
1713 
1714         if (rv < 0 || len != 0) {
1715                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1716                     "failed to read from stream"));
1717                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1718                     "cannot receive")));
1719         }
1720 
1721         if (zc) {
1722                 if (byteswap)
1723                         fletcher_4_incremental_byteswap(buf, ilen, zc);
1724                 else
1725                         fletcher_4_incremental_native(buf, ilen, zc);
1726         }
1727         return (0);
1728 }
1729 
1730 static int
1731 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1732     boolean_t byteswap, zio_cksum_t *zc)
1733 {
1734         char *buf;
1735         int err;
1736 
1737         buf = zfs_alloc(hdl, len);
1738         if (buf == NULL)
1739                 return (ENOMEM);
1740 
1741         err = recv_read(hdl, fd, buf, len, byteswap, zc);
1742         if (err != 0) {
1743                 free(buf);
1744                 return (err);
1745         }
1746 
1747         err = nvlist_unpack(buf, len, nvp, 0);
1748         free(buf);
1749         if (err != 0) {
1750                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1751                     "stream (malformed nvlist)"));
1752                 return (EINVAL);
1753         }
1754         return (0);
1755 }
1756 
1757 static int
1758 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1759     int baselen, char *newname, recvflags_t *flags)
1760 {
1761         static int seq;
1762         zfs_cmd_t zc = { 0 };
1763         int err;
1764         prop_changelist_t *clp;
1765         zfs_handle_t *zhp;
1766 
1767         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1768         if (zhp == NULL)
1769                 return (-1);
1770         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1771             flags->force ? MS_FORCE : 0);
1772         zfs_close(zhp);
1773         if (clp == NULL)
1774                 return (-1);
1775         err = changelist_prefix(clp);
1776         if (err)
1777                 return (err);
1778 
1779         zc.zc_objset_type = DMU_OST_ZFS;
1780         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1781 
1782         if (tryname) {
1783                 (void) strcpy(newname, tryname);
1784 
1785                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1786 
1787                 if (flags->verbose) {
1788                         (void) printf("attempting rename %s to %s\n",
1789                             zc.zc_name, zc.zc_value);
1790                 }
1791                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1792                 if (err == 0)
1793                         changelist_rename(clp, name, tryname);
1794         } else {
1795                 err = ENOENT;
1796         }
1797 
1798         if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
1799                 seq++;
1800 
1801                 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u",
1802                     baselen, name, getpid(), seq);
1803                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1804 
1805                 if (flags->verbose) {
1806                         (void) printf("failed - trying rename %s to %s\n",
1807                             zc.zc_name, zc.zc_value);
1808                 }
1809                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1810                 if (err == 0)
1811                         changelist_rename(clp, name, newname);
1812                 if (err && flags->verbose) {
1813                         (void) printf("failed (%u) - "
1814                             "will try again on next pass\n", errno);
1815                 }
1816                 err = EAGAIN;
1817         } else if (flags->verbose) {
1818                 if (err == 0)
1819                         (void) printf("success\n");
1820                 else
1821                         (void) printf("failed (%u)\n", errno);
1822         }
1823 
1824         (void) changelist_postfix(clp);
1825         changelist_free(clp);
1826 
1827         return (err);
1828 }
1829 
1830 static int
1831 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1832     char *newname, recvflags_t *flags)
1833 {
1834         zfs_cmd_t zc = { 0 };
1835         int err = 0;
1836         prop_changelist_t *clp;
1837         zfs_handle_t *zhp;
1838         boolean_t defer = B_FALSE;
1839         int spa_version;
1840 
1841         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1842         if (zhp == NULL)
1843                 return (-1);
1844         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1845             flags->force ? MS_FORCE : 0);
1846         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1847             zfs_spa_version(zhp, &spa_version) == 0 &&
1848             spa_version >= SPA_VERSION_USERREFS)
1849                 defer = B_TRUE;
1850         zfs_close(zhp);
1851         if (clp == NULL)
1852                 return (-1);
1853         err = changelist_prefix(clp);
1854         if (err)
1855                 return (err);
1856 
1857         zc.zc_objset_type = DMU_OST_ZFS;
1858         zc.zc_defer_destroy = defer;
1859         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1860 
1861         if (flags->verbose)
1862                 (void) printf("attempting destroy %s\n", zc.zc_name);
1863         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1864         if (err == 0) {
1865                 if (flags->verbose)
1866                         (void) printf("success\n");
1867                 changelist_remove(clp, zc.zc_name);
1868         }
1869 
1870         (void) changelist_postfix(clp);
1871         changelist_free(clp);
1872 
1873         /*
1874          * Deferred destroy might destroy the snapshot or only mark it to be
1875          * destroyed later, and it returns success in either case.
1876          */
1877         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1878             ZFS_TYPE_SNAPSHOT))) {
1879                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1880         }
1881 
1882         return (err);
1883 }
1884 
1885 typedef struct guid_to_name_data {
1886         uint64_t guid;
1887         char *name;
1888         char *skip;
1889 } guid_to_name_data_t;
1890 
1891 static int
1892 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1893 {
1894         guid_to_name_data_t *gtnd = arg;
1895         int err;
1896 
1897         if (gtnd->skip != NULL &&
1898             strcmp(zhp->zfs_name, gtnd->skip) == 0) {
1899                 return (0);
1900         }
1901 
1902         if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1903                 (void) strcpy(gtnd->name, zhp->zfs_name);
1904                 zfs_close(zhp);
1905                 return (EEXIST);
1906         }
1907 
1908         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1909         zfs_close(zhp);
1910         return (err);
1911 }
1912 
1913 /*
1914  * Attempt to find the local dataset associated with this guid.  In the case of
1915  * multiple matches, we attempt to find the "best" match by searching
1916  * progressively larger portions of the hierarchy.  This allows one to send a
1917  * tree of datasets individually and guarantee that we will find the source
1918  * guid within that hierarchy, even if there are multiple matches elsewhere.
1919  */
1920 static int
1921 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1922     char *name)
1923 {
1924         /* exhaustive search all local snapshots */
1925         char pname[ZFS_MAXNAMELEN];
1926         guid_to_name_data_t gtnd;
1927         int err = 0;
1928         zfs_handle_t *zhp;
1929         char *cp;
1930 
1931         gtnd.guid = guid;
1932         gtnd.name = name;
1933         gtnd.skip = NULL;
1934 
1935         (void) strlcpy(pname, parent, sizeof (pname));
1936 
1937         /*
1938          * Search progressively larger portions of the hierarchy.  This will
1939          * select the "most local" version of the origin snapshot in the case
1940          * that there are multiple matching snapshots in the system.
1941          */
1942         while ((cp = strrchr(pname, '/')) != NULL) {
1943 
1944                 /* Chop off the last component and open the parent */
1945                 *cp = '\0';
1946                 zhp = make_dataset_handle(hdl, pname);
1947 
1948                 if (zhp == NULL)
1949                         continue;
1950 
1951                 err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1952                 zfs_close(zhp);
1953                 if (err == EEXIST)
1954                         return (0);
1955 
1956                 /*
1957                  * Remember the dataset that we already searched, so we
1958                  * skip it next time through.
1959                  */
1960                 gtnd.skip = pname;
1961         }
1962 
1963         return (ENOENT);
1964 }
1965 
1966 /*
1967  * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
1968  * guid1 is after guid2.
1969  */
1970 static int
1971 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1972     uint64_t guid1, uint64_t guid2)
1973 {
1974         nvlist_t *nvfs;
1975         char *fsname, *snapname;
1976         char buf[ZFS_MAXNAMELEN];
1977         int rv;
1978         zfs_handle_t *guid1hdl, *guid2hdl;
1979         uint64_t create1, create2;
1980 
1981         if (guid2 == 0)
1982                 return (0);
1983         if (guid1 == 0)
1984                 return (1);
1985 
1986         nvfs = fsavl_find(avl, guid1, &snapname);
1987         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1988         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1989         guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1990         if (guid1hdl == NULL)
1991                 return (-1);
1992 
1993         nvfs = fsavl_find(avl, guid2, &snapname);
1994         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1995         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1996         guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1997         if (guid2hdl == NULL) {
1998                 zfs_close(guid1hdl);
1999                 return (-1);
2000         }
2001 
2002         create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
2003         create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
2004 
2005         if (create1 < create2)
2006                 rv = -1;
2007         else if (create1 > create2)
2008                 rv = +1;
2009         else
2010                 rv = 0;
2011 
2012         zfs_close(guid1hdl);
2013         zfs_close(guid2hdl);
2014 
2015         return (rv);
2016 }
2017 
2018 static int
2019 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2020     recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2021     nvlist_t *renamed)
2022 {
2023         nvlist_t *local_nv;
2024         avl_tree_t *local_avl;
2025         nvpair_t *fselem, *nextfselem;
2026         char *fromsnap;
2027         char newname[ZFS_MAXNAMELEN];
2028         int error;
2029         boolean_t needagain, progress, recursive;
2030         char *s1, *s2;
2031 
2032         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2033 
2034         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2035             ENOENT);
2036 
2037         if (flags->dryrun)
2038                 return (0);
2039 
2040 again:
2041         needagain = progress = B_FALSE;
2042 
2043         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2044             recursive, &local_nv, &local_avl)) != 0)
2045                 return (error);
2046 
2047         /*
2048          * Process deletes and renames
2049          */
2050         for (fselem = nvlist_next_nvpair(local_nv, NULL);
2051             fselem; fselem = nextfselem) {
2052                 nvlist_t *nvfs, *snaps;
2053                 nvlist_t *stream_nvfs = NULL;
2054                 nvpair_t *snapelem, *nextsnapelem;
2055                 uint64_t fromguid = 0;
2056                 uint64_t originguid = 0;
2057                 uint64_t stream_originguid = 0;
2058                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2059                 char *fsname, *stream_fsname;
2060 
2061                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2062 
2063                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2064                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2065                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2066                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2067                     &parent_fromsnap_guid));
2068                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2069 
2070                 /*
2071                  * First find the stream's fs, so we can check for
2072                  * a different origin (due to "zfs promote")
2073                  */
2074                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2075                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2076                         uint64_t thisguid;
2077 
2078                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2079                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2080 
2081                         if (stream_nvfs != NULL)
2082                                 break;
2083                 }
2084 
2085                 /* check for promote */
2086                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2087                     &stream_originguid);
2088                 if (stream_nvfs && originguid != stream_originguid) {
2089                         switch (created_before(hdl, local_avl,
2090                             stream_originguid, originguid)) {
2091                         case 1: {
2092                                 /* promote it! */
2093                                 zfs_cmd_t zc = { 0 };
2094                                 nvlist_t *origin_nvfs;
2095                                 char *origin_fsname;
2096 
2097                                 if (flags->verbose)
2098                                         (void) printf("promoting %s\n", fsname);
2099 
2100                                 origin_nvfs = fsavl_find(local_avl, originguid,
2101                                     NULL);
2102                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2103                                     "name", &origin_fsname));
2104                                 (void) strlcpy(zc.zc_value, origin_fsname,
2105                                     sizeof (zc.zc_value));
2106                                 (void) strlcpy(zc.zc_name, fsname,
2107                                     sizeof (zc.zc_name));
2108                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2109                                 if (error == 0)
2110                                         progress = B_TRUE;
2111                                 break;
2112                         }
2113                         default:
2114                                 break;
2115                         case -1:
2116                                 fsavl_destroy(local_avl);
2117                                 nvlist_free(local_nv);
2118                                 return (-1);
2119                         }
2120                         /*
2121                          * We had/have the wrong origin, therefore our
2122                          * list of snapshots is wrong.  Need to handle
2123                          * them on the next pass.
2124                          */
2125                         needagain = B_TRUE;
2126                         continue;
2127                 }
2128 
2129                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2130                     snapelem; snapelem = nextsnapelem) {
2131                         uint64_t thisguid;
2132                         char *stream_snapname;
2133                         nvlist_t *found, *props;
2134 
2135                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2136 
2137                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2138                         found = fsavl_find(stream_avl, thisguid,
2139                             &stream_snapname);
2140 
2141                         /* check for delete */
2142                         if (found == NULL) {
2143                                 char name[ZFS_MAXNAMELEN];
2144 
2145                                 if (!flags->force)
2146                                         continue;
2147 
2148                                 (void) snprintf(name, sizeof (name), "%s@%s",
2149                                     fsname, nvpair_name(snapelem));
2150 
2151                                 error = recv_destroy(hdl, name,
2152                                     strlen(fsname)+1, newname, flags);
2153                                 if (error)
2154                                         needagain = B_TRUE;
2155                                 else
2156                                         progress = B_TRUE;
2157                                 continue;
2158                         }
2159 
2160                         stream_nvfs = found;
2161 
2162                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2163                             &props) && 0 == nvlist_lookup_nvlist(props,
2164                             stream_snapname, &props)) {
2165                                 zfs_cmd_t zc = { 0 };
2166 
2167                                 zc.zc_cookie = B_TRUE; /* received */
2168                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2169                                     "%s@%s", fsname, nvpair_name(snapelem));
2170                                 if (zcmd_write_src_nvlist(hdl, &zc,
2171                                     props) == 0) {
2172                                         (void) zfs_ioctl(hdl,
2173                                             ZFS_IOC_SET_PROP, &zc);
2174                                         zcmd_free_nvlists(&zc);
2175                                 }
2176                         }
2177 
2178                         /* check for different snapname */
2179                         if (strcmp(nvpair_name(snapelem),
2180                             stream_snapname) != 0) {
2181                                 char name[ZFS_MAXNAMELEN];
2182                                 char tryname[ZFS_MAXNAMELEN];
2183 
2184                                 (void) snprintf(name, sizeof (name), "%s@%s",
2185                                     fsname, nvpair_name(snapelem));
2186                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
2187                                     fsname, stream_snapname);
2188 
2189                                 error = recv_rename(hdl, name, tryname,
2190                                     strlen(fsname)+1, newname, flags);
2191                                 if (error)
2192                                         needagain = B_TRUE;
2193                                 else
2194                                         progress = B_TRUE;
2195                         }
2196 
2197                         if (strcmp(stream_snapname, fromsnap) == 0)
2198                                 fromguid = thisguid;
2199                 }
2200 
2201                 /* check for delete */
2202                 if (stream_nvfs == NULL) {
2203                         if (!flags->force)
2204                                 continue;
2205 
2206                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2207                             newname, flags);
2208                         if (error)
2209                                 needagain = B_TRUE;
2210                         else
2211                                 progress = B_TRUE;
2212                         continue;
2213                 }
2214 
2215                 if (fromguid == 0) {
2216                         if (flags->verbose) {
2217                                 (void) printf("local fs %s does not have "
2218                                     "fromsnap (%s in stream); must have "
2219                                     "been deleted locally; ignoring\n",
2220                                     fsname, fromsnap);
2221                         }
2222                         continue;
2223                 }
2224 
2225                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2226                     "name", &stream_fsname));
2227                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2228                     "parentfromsnap", &stream_parent_fromsnap_guid));
2229 
2230                 s1 = strrchr(fsname, '/');
2231                 s2 = strrchr(stream_fsname, '/');
2232 
2233                 /*
2234                  * Check for rename. If the exact receive path is specified, it
2235                  * does not count as a rename, but we still need to check the
2236                  * datasets beneath it.
2237                  */
2238                 if ((stream_parent_fromsnap_guid != 0 &&
2239                     parent_fromsnap_guid != 0 &&
2240                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2241                     ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2242                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2243                         nvlist_t *parent;
2244                         char tryname[ZFS_MAXNAMELEN];
2245 
2246                         parent = fsavl_find(local_avl,
2247                             stream_parent_fromsnap_guid, NULL);
2248                         /*
2249                          * NB: parent might not be found if we used the
2250                          * tosnap for stream_parent_fromsnap_guid,
2251                          * because the parent is a newly-created fs;
2252                          * we'll be able to rename it after we recv the
2253                          * new fs.
2254                          */
2255                         if (parent != NULL) {
2256                                 char *pname;
2257 
2258                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
2259                                     &pname));
2260                                 (void) snprintf(tryname, sizeof (tryname),
2261                                     "%s%s", pname, strrchr(stream_fsname, '/'));
2262                         } else {
2263                                 tryname[0] = '\0';
2264                                 if (flags->verbose) {
2265                                         (void) printf("local fs %s new parent "
2266                                             "not found\n", fsname);
2267                                 }
2268                         }
2269 
2270                         newname[0] = '\0';
2271 
2272                         error = recv_rename(hdl, fsname, tryname,
2273                             strlen(tofs)+1, newname, flags);
2274 
2275                         if (renamed != NULL && newname[0] != '\0') {
2276                                 VERIFY(0 == nvlist_add_boolean(renamed,
2277                                     newname));
2278                         }
2279 
2280                         if (error)
2281                                 needagain = B_TRUE;
2282                         else
2283                                 progress = B_TRUE;
2284                 }
2285         }
2286 
2287         fsavl_destroy(local_avl);
2288         nvlist_free(local_nv);
2289 
2290         if (needagain && progress) {
2291                 /* do another pass to fix up temporary names */
2292                 if (flags->verbose)
2293                         (void) printf("another pass:\n");
2294                 goto again;
2295         }
2296 
2297         return (needagain);
2298 }
2299 
2300 static int
2301 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2302     recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2303     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2304 {
2305         nvlist_t *stream_nv = NULL;
2306         avl_tree_t *stream_avl = NULL;
2307         char *fromsnap = NULL;
2308         char *cp;
2309         char tofs[ZFS_MAXNAMELEN];
2310         char sendfs[ZFS_MAXNAMELEN];
2311         char errbuf[1024];
2312         dmu_replay_record_t drre;
2313         int error;
2314         boolean_t anyerr = B_FALSE;
2315         boolean_t softerr = B_FALSE;
2316         boolean_t recursive;
2317 
2318         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2319             "cannot receive"));
2320 
2321         assert(drr->drr_type == DRR_BEGIN);
2322         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2323         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2324             DMU_COMPOUNDSTREAM);
2325 
2326         /*
2327          * Read in the nvlist from the stream.
2328          */
2329         if (drr->drr_payloadlen != 0) {
2330                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2331                     &stream_nv, flags->byteswap, zc);
2332                 if (error) {
2333                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2334                         goto out;
2335                 }
2336         }
2337 
2338         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2339             ENOENT);
2340 
2341         if (recursive && strchr(destname, '@')) {
2342                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2343                     "cannot specify snapshot name for multi-snapshot stream"));
2344                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2345                 goto out;
2346         }
2347 
2348         /*
2349          * Read in the end record and verify checksum.
2350          */
2351         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2352             flags->byteswap, NULL)))
2353                 goto out;
2354         if (flags->byteswap) {
2355                 drre.drr_type = BSWAP_32(drre.drr_type);
2356                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2357                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2358                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2359                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2360                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2361                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2362                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2363                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2364         }
2365         if (drre.drr_type != DRR_END) {
2366                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2367                 goto out;
2368         }
2369         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2370                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2371                     "incorrect header checksum"));
2372                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2373                 goto out;
2374         }
2375 
2376         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2377 
2378         if (drr->drr_payloadlen != 0) {
2379                 nvlist_t *stream_fss;
2380 
2381                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2382                     &stream_fss));
2383                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2384                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2385                             "couldn't allocate avl tree"));
2386                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2387                         goto out;
2388                 }
2389 
2390                 if (fromsnap != NULL) {
2391                         nvlist_t *renamed = NULL;
2392                         nvpair_t *pair = NULL;
2393 
2394                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2395                         if (flags->isprefix) {
2396                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2397                                 int i;
2398 
2399                                 if (flags->istail) {
2400                                         cp = strrchr(drrb->drr_toname, '/');
2401                                         if (cp == NULL) {
2402                                                 (void) strlcat(tofs, "/",
2403                                                     ZFS_MAXNAMELEN);
2404                                                 i = 0;
2405                                         } else {
2406                                                 i = (cp - drrb->drr_toname);
2407                                         }
2408                                 } else {
2409                                         i = strcspn(drrb->drr_toname, "/@");
2410                                 }
2411                                 /* zfs_receive_one() will create_parents() */
2412                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2413                                     ZFS_MAXNAMELEN);
2414                                 *strchr(tofs, '@') = '\0';
2415                         }
2416 
2417                         if (recursive && !flags->dryrun && !flags->nomount) {
2418                                 VERIFY(0 == nvlist_alloc(&renamed,
2419                                     NV_UNIQUE_NAME, 0));
2420                         }
2421 
2422                         softerr = recv_incremental_replication(hdl, tofs, flags,
2423                             stream_nv, stream_avl, renamed);
2424 
2425                         /* Unmount renamed filesystems before receiving. */
2426                         while ((pair = nvlist_next_nvpair(renamed,
2427                             pair)) != NULL) {
2428                                 zfs_handle_t *zhp;
2429                                 prop_changelist_t *clp = NULL;
2430 
2431                                 zhp = zfs_open(hdl, nvpair_name(pair),
2432                                     ZFS_TYPE_FILESYSTEM);
2433                                 if (zhp != NULL) {
2434                                         clp = changelist_gather(zhp,
2435                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2436                                         zfs_close(zhp);
2437                                         if (clp != NULL) {
2438                                                 softerr |=
2439                                                     changelist_prefix(clp);
2440                                                 changelist_free(clp);
2441                                         }
2442                                 }
2443                         }
2444 
2445                         nvlist_free(renamed);
2446                 }
2447         }
2448 
2449         /*
2450          * Get the fs specified by the first path in the stream (the top level
2451          * specified by 'zfs send') and pass it to each invocation of
2452          * zfs_receive_one().
2453          */
2454         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2455             ZFS_MAXNAMELEN);
2456         if ((cp = strchr(sendfs, '@')) != NULL)
2457                 *cp = '\0';
2458 
2459         /* Finally, receive each contained stream */
2460         do {
2461                 /*
2462                  * we should figure out if it has a recoverable
2463                  * error, in which case do a recv_skip() and drive on.
2464                  * Note, if we fail due to already having this guid,
2465                  * zfs_receive_one() will take care of it (ie,
2466                  * recv_skip() and return 0).
2467                  */
2468                 error = zfs_receive_impl(hdl, destname, flags, fd,
2469                     sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2470                     action_handlep);
2471                 if (error == ENODATA) {
2472                         error = 0;
2473                         break;
2474                 }
2475                 anyerr |= error;
2476         } while (error == 0);
2477 
2478         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2479                 /*
2480                  * Now that we have the fs's they sent us, try the
2481                  * renames again.
2482                  */
2483                 softerr = recv_incremental_replication(hdl, tofs, flags,
2484                     stream_nv, stream_avl, NULL);
2485         }
2486 
2487 out:
2488         fsavl_destroy(stream_avl);
2489         if (stream_nv)
2490                 nvlist_free(stream_nv);
2491         if (softerr)
2492                 error = -2;
2493         if (anyerr)
2494                 error = -1;
2495         return (error);
2496 }
2497 
2498 static void
2499 trunc_prop_errs(int truncated)
2500 {
2501         ASSERT(truncated != 0);
2502 
2503         if (truncated == 1)
2504                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2505                     "1 more property could not be set\n"));
2506         else
2507                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2508                     "%d more properties could not be set\n"), truncated);
2509 }
2510 
2511 static int
2512 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2513 {
2514         dmu_replay_record_t *drr;
2515         void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
2516         char errbuf[1024];
2517 
2518         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2519             "cannot receive:"));
2520 
2521         /* XXX would be great to use lseek if possible... */
2522         drr = buf;
2523 
2524         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2525             byteswap, NULL) == 0) {
2526                 if (byteswap)
2527                         drr->drr_type = BSWAP_32(drr->drr_type);
2528 
2529                 switch (drr->drr_type) {
2530                 case DRR_BEGIN:
2531                         /* NB: not to be used on v2 stream packages */
2532                         if (drr->drr_payloadlen != 0) {
2533                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2534                                     "invalid substream header"));
2535                                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2536                         }
2537                         break;
2538 
2539                 case DRR_END:
2540                         free(buf);
2541                         return (0);
2542 
2543                 case DRR_OBJECT:
2544                         if (byteswap) {
2545                                 drr->drr_u.drr_object.drr_bonuslen =
2546                                     BSWAP_32(drr->drr_u.drr_object.
2547                                     drr_bonuslen);
2548                         }
2549                         (void) recv_read(hdl, fd, buf,
2550                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2551                             B_FALSE, NULL);
2552                         break;
2553 
2554                 case DRR_WRITE:
2555                         if (byteswap) {
2556                                 drr->drr_u.drr_write.drr_length =
2557                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
2558                         }
2559                         (void) recv_read(hdl, fd, buf,
2560                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2561                         break;
2562                 case DRR_SPILL:
2563                         if (byteswap) {
2564                                 drr->drr_u.drr_write.drr_length =
2565                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
2566                         }
2567                         (void) recv_read(hdl, fd, buf,
2568                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2569                         break;
2570                 case DRR_WRITE_EMBEDDED:
2571                         if (byteswap) {
2572                                 drr->drr_u.drr_write_embedded.drr_psize =
2573                                     BSWAP_32(drr->drr_u.drr_write_embedded.
2574                                     drr_psize);
2575                         }
2576                         (void) recv_read(hdl, fd, buf,
2577                             P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
2578                             8), B_FALSE, NULL);
2579                         break;
2580                 case DRR_WRITE_BYREF:
2581                 case DRR_FREEOBJECTS:
2582                 case DRR_FREE:
2583                         break;
2584 
2585                 default:
2586                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2587                             "invalid record type"));
2588                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2589                 }
2590         }
2591 
2592         free(buf);
2593         return (-1);
2594 }
2595 
2596 /*
2597  * Restores a backup of tosnap from the file descriptor specified by infd.
2598  */
2599 static int
2600 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2601     recvflags_t *flags, dmu_replay_record_t *drr,
2602     dmu_replay_record_t *drr_noswap, const char *sendfs,
2603     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2604     uint64_t *action_handlep)
2605 {
2606         zfs_cmd_t zc = { 0 };
2607         time_t begin_time;
2608         int ioctl_err, ioctl_errno, err;
2609         char *cp;
2610         struct drr_begin *drrb = &drr->drr_u.drr_begin;
2611         char errbuf[1024];
2612         char prop_errbuf[1024];
2613         const char *chopprefix;
2614         boolean_t newfs = B_FALSE;
2615         boolean_t stream_wantsnewfs;
2616         uint64_t parent_snapguid = 0;
2617         prop_changelist_t *clp = NULL;
2618         nvlist_t *snapprops_nvlist = NULL;
2619         zprop_errflags_t prop_errflags;
2620         boolean_t recursive;
2621 
2622         begin_time = time(NULL);
2623 
2624         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2625             "cannot receive"));
2626 
2627         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2628             ENOENT);
2629 
2630         if (stream_avl != NULL) {
2631                 char *snapname;
2632                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2633                     &snapname);
2634                 nvlist_t *props;
2635                 int ret;
2636 
2637                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2638                     &parent_snapguid);
2639                 err = nvlist_lookup_nvlist(fs, "props", &props);
2640                 if (err)
2641                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2642 
2643                 if (flags->canmountoff) {
2644                         VERIFY(0 == nvlist_add_uint64(props,
2645                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2646                 }
2647                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2648                 if (err)
2649                         nvlist_free(props);
2650 
2651                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2652                         VERIFY(0 == nvlist_lookup_nvlist(props,
2653                             snapname, &snapprops_nvlist));
2654                 }
2655 
2656                 if (ret != 0)
2657                         return (-1);
2658         }
2659 
2660         cp = NULL;
2661 
2662         /*
2663          * Determine how much of the snapshot name stored in the stream
2664          * we are going to tack on to the name they specified on the
2665          * command line, and how much we are going to chop off.
2666          *
2667          * If they specified a snapshot, chop the entire name stored in
2668          * the stream.
2669          */
2670         if (flags->istail) {
2671                 /*
2672                  * A filesystem was specified with -e. We want to tack on only
2673                  * the tail of the sent snapshot path.
2674                  */
2675                 if (strchr(tosnap, '@')) {
2676                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2677                             "argument - snapshot not allowed with -e"));
2678                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2679                 }
2680 
2681                 chopprefix = strrchr(sendfs, '/');
2682 
2683                 if (chopprefix == NULL) {
2684                         /*
2685                          * The tail is the poolname, so we need to
2686                          * prepend a path separator.
2687                          */
2688                         int len = strlen(drrb->drr_toname);
2689                         cp = malloc(len + 2);
2690                         cp[0] = '/';
2691                         (void) strcpy(&cp[1], drrb->drr_toname);
2692                         chopprefix = cp;
2693                 } else {
2694                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2695                 }
2696         } else if (flags->isprefix) {
2697                 /*
2698                  * A filesystem was specified with -d. We want to tack on
2699                  * everything but the first element of the sent snapshot path
2700                  * (all but the pool name).
2701                  */
2702                 if (strchr(tosnap, '@')) {
2703                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2704                             "argument - snapshot not allowed with -d"));
2705                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2706                 }
2707 
2708                 chopprefix = strchr(drrb->drr_toname, '/');
2709                 if (chopprefix == NULL)
2710                         chopprefix = strchr(drrb->drr_toname, '@');
2711         } else if (strchr(tosnap, '@') == NULL) {
2712                 /*
2713                  * If a filesystem was specified without -d or -e, we want to
2714                  * tack on everything after the fs specified by 'zfs send'.
2715                  */
2716                 chopprefix = drrb->drr_toname + strlen(sendfs);
2717         } else {
2718                 /* A snapshot was specified as an exact path (no -d or -e). */
2719                 if (recursive) {
2720                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2721                             "cannot specify snapshot name for multi-snapshot "
2722                             "stream"));
2723                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2724                 }
2725                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2726         }
2727 
2728         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2729         ASSERT(chopprefix > drrb->drr_toname);
2730         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2731         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2732             chopprefix[0] == '\0');
2733 
2734         /*
2735          * Determine name of destination snapshot, store in zc_value.
2736          */
2737         (void) strcpy(zc.zc_value, tosnap);
2738         (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2739         free(cp);
2740         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2741                 zcmd_free_nvlists(&zc);
2742                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2743         }
2744 
2745         /*
2746          * Determine the name of the origin snapshot, store in zc_string.
2747          */
2748         if (drrb->drr_flags & DRR_FLAG_CLONE) {
2749                 if (guid_to_name(hdl, zc.zc_value,
2750                     drrb->drr_fromguid, zc.zc_string) != 0) {
2751                         zcmd_free_nvlists(&zc);
2752                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2753                             "local origin for clone %s does not exist"),
2754                             zc.zc_value);
2755                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2756                 }
2757                 if (flags->verbose)
2758                         (void) printf("found clone origin %s\n", zc.zc_string);
2759         }
2760 
2761         stream_wantsnewfs = (drrb->drr_fromguid == NULL ||
2762             (drrb->drr_flags & DRR_FLAG_CLONE));
2763 
2764         if (stream_wantsnewfs) {
2765                 /*
2766                  * if the parent fs does not exist, look for it based on
2767                  * the parent snap GUID
2768                  */
2769                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2770                     "cannot receive new filesystem stream"));
2771 
2772                 (void) strcpy(zc.zc_name, zc.zc_value);
2773                 cp = strrchr(zc.zc_name, '/');
2774                 if (cp)
2775                         *cp = '\0';
2776                 if (cp &&
2777                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2778                         char suffix[ZFS_MAXNAMELEN];
2779                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2780                         if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
2781                             zc.zc_value) == 0) {
2782                                 *strchr(zc.zc_value, '@') = '\0';
2783                                 (void) strcat(zc.zc_value, suffix);
2784                         }
2785                 }
2786         } else {
2787                 /*
2788                  * if the fs does not exist, look for it based on the
2789                  * fromsnap GUID
2790                  */
2791                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2792                     "cannot receive incremental stream"));
2793 
2794                 (void) strcpy(zc.zc_name, zc.zc_value);
2795                 *strchr(zc.zc_name, '@') = '\0';
2796 
2797                 /*
2798                  * If the exact receive path was specified and this is the
2799                  * topmost path in the stream, then if the fs does not exist we
2800                  * should look no further.
2801                  */
2802                 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
2803                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2804                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2805                         char snap[ZFS_MAXNAMELEN];
2806                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
2807                         if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
2808                             zc.zc_value) == 0) {
2809                                 *strchr(zc.zc_value, '@') = '\0';
2810                                 (void) strcat(zc.zc_value, snap);
2811                         }
2812                 }
2813         }
2814 
2815         (void) strcpy(zc.zc_name, zc.zc_value);
2816         *strchr(zc.zc_name, '@') = '\0';
2817 
2818         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2819                 zfs_handle_t *zhp;
2820 
2821                 /*
2822                  * Destination fs exists.  Therefore this should either
2823                  * be an incremental, or the stream specifies a new fs
2824                  * (full stream or clone) and they want us to blow it
2825                  * away (and have therefore specified -F and removed any
2826                  * snapshots).
2827                  */
2828                 if (stream_wantsnewfs) {
2829                         if (!flags->force) {
2830                                 zcmd_free_nvlists(&zc);
2831                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2832                                     "destination '%s' exists\n"
2833                                     "must specify -F to overwrite it"),
2834                                     zc.zc_name);
2835                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2836                         }
2837                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2838                             &zc) == 0) {
2839                                 zcmd_free_nvlists(&zc);
2840                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2841                                     "destination has snapshots (eg. %s)\n"
2842                                     "must destroy them to overwrite it"),
2843                                     zc.zc_name);
2844                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2845                         }
2846                 }
2847 
2848                 if ((zhp = zfs_open(hdl, zc.zc_name,
2849                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2850                         zcmd_free_nvlists(&zc);
2851                         return (-1);
2852                 }
2853 
2854                 if (stream_wantsnewfs &&
2855                     zhp->zfs_dmustats.dds_origin[0]) {
2856                         zcmd_free_nvlists(&zc);
2857                         zfs_close(zhp);
2858                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2859                             "destination '%s' is a clone\n"
2860                             "must destroy it to overwrite it"),
2861                             zc.zc_name);
2862                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2863                 }
2864 
2865                 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2866                     stream_wantsnewfs) {
2867                         /* We can't do online recv in this case */
2868                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2869                         if (clp == NULL) {
2870                                 zfs_close(zhp);
2871                                 zcmd_free_nvlists(&zc);
2872                                 return (-1);
2873                         }
2874                         if (changelist_prefix(clp) != 0) {
2875                                 changelist_free(clp);
2876                                 zfs_close(zhp);
2877                                 zcmd_free_nvlists(&zc);
2878                                 return (-1);
2879                         }
2880                 }
2881                 zfs_close(zhp);
2882         } else {
2883                 /*
2884                  * Destination filesystem does not exist.  Therefore we better
2885                  * be creating a new filesystem (either from a full backup, or
2886                  * a clone).  It would therefore be invalid if the user
2887                  * specified only the pool name (i.e. if the destination name
2888                  * contained no slash character).
2889                  */
2890                 if (!stream_wantsnewfs ||
2891                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
2892                         zcmd_free_nvlists(&zc);
2893                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2894                             "destination '%s' does not exist"), zc.zc_name);
2895                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2896                 }
2897 
2898                 /*
2899                  * Trim off the final dataset component so we perform the
2900                  * recvbackup ioctl to the filesystems's parent.
2901                  */
2902                 *cp = '\0';
2903 
2904                 if (flags->isprefix && !flags->istail && !flags->dryrun &&
2905                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2906                         zcmd_free_nvlists(&zc);
2907                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2908                 }
2909 
2910                 newfs = B_TRUE;
2911         }
2912 
2913         zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2914         zc.zc_cookie = infd;
2915         zc.zc_guid = flags->force;
2916         if (flags->verbose) {
2917                 (void) printf("%s %s stream of %s into %s\n",
2918                     flags->dryrun ? "would receive" : "receiving",
2919                     drrb->drr_fromguid ? "incremental" : "full",
2920                     drrb->drr_toname, zc.zc_value);
2921                 (void) fflush(stdout);
2922         }
2923 
2924         if (flags->dryrun) {
2925                 zcmd_free_nvlists(&zc);
2926                 return (recv_skip(hdl, infd, flags->byteswap));
2927         }
2928 
2929         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2930         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2931         zc.zc_cleanup_fd = cleanup_fd;
2932         zc.zc_action_handle = *action_handlep;
2933 
2934         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2935         ioctl_errno = errno;
2936         prop_errflags = (zprop_errflags_t)zc.zc_obj;
2937 
2938         if (err == 0) {
2939                 nvlist_t *prop_errors;
2940                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2941                     zc.zc_nvlist_dst_size, &prop_errors, 0));
2942 
2943                 nvpair_t *prop_err = NULL;
2944 
2945                 while ((prop_err = nvlist_next_nvpair(prop_errors,
2946                     prop_err)) != NULL) {
2947                         char tbuf[1024];
2948                         zfs_prop_t prop;
2949                         int intval;
2950 
2951                         prop = zfs_name_to_prop(nvpair_name(prop_err));
2952                         (void) nvpair_value_int32(prop_err, &intval);
2953                         if (strcmp(nvpair_name(prop_err),
2954                             ZPROP_N_MORE_ERRORS) == 0) {
2955                                 trunc_prop_errs(intval);
2956                                 break;
2957                         } else {
2958                                 (void) snprintf(tbuf, sizeof (tbuf),
2959                                     dgettext(TEXT_DOMAIN,
2960                                     "cannot receive %s property on %s"),
2961                                     nvpair_name(prop_err), zc.zc_name);
2962                                 zfs_setprop_error(hdl, prop, intval, tbuf);
2963                         }
2964                 }
2965                 nvlist_free(prop_errors);
2966         }
2967 
2968         zc.zc_nvlist_dst = 0;
2969         zc.zc_nvlist_dst_size = 0;
2970         zcmd_free_nvlists(&zc);
2971 
2972         if (err == 0 && snapprops_nvlist) {
2973                 zfs_cmd_t zc2 = { 0 };
2974 
2975                 (void) strcpy(zc2.zc_name, zc.zc_value);
2976                 zc2.zc_cookie = B_TRUE; /* received */
2977                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
2978                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
2979                         zcmd_free_nvlists(&zc2);
2980                 }
2981         }
2982 
2983         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
2984                 /*
2985                  * It may be that this snapshot already exists,
2986                  * in which case we want to consume & ignore it
2987                  * rather than failing.
2988                  */
2989                 avl_tree_t *local_avl;
2990                 nvlist_t *local_nv, *fs;
2991                 cp = strchr(zc.zc_value, '@');
2992 
2993                 /*
2994                  * XXX Do this faster by just iterating over snaps in
2995                  * this fs.  Also if zc_value does not exist, we will
2996                  * get a strange "does not exist" error message.
2997                  */
2998                 *cp = '\0';
2999                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3000                     &local_nv, &local_avl) == 0) {
3001                         *cp = '@';
3002                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3003                         fsavl_destroy(local_avl);
3004                         nvlist_free(local_nv);
3005 
3006                         if (fs != NULL) {
3007                                 if (flags->verbose) {
3008                                         (void) printf("snap %s already exists; "
3009                                             "ignoring\n", zc.zc_value);
3010                                 }
3011                                 err = ioctl_err = recv_skip(hdl, infd,
3012                                     flags->byteswap);
3013                         }
3014                 }
3015                 *cp = '@';
3016         }
3017 
3018         if (ioctl_err != 0) {
3019                 switch (ioctl_errno) {
3020                 case ENODEV:
3021                         cp = strchr(zc.zc_value, '@');
3022                         *cp = '\0';
3023                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3024                             "most recent snapshot of %s does not\n"
3025                             "match incremental source"), zc.zc_value);
3026                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3027                         *cp = '@';
3028                         break;
3029                 case ETXTBSY:
3030                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3031                             "destination %s has been modified\n"
3032                             "since most recent snapshot"), zc.zc_name);
3033                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3034                         break;
3035                 case EEXIST:
3036                         cp = strchr(zc.zc_value, '@');
3037                         if (newfs) {
3038                                 /* it's the containing fs that exists */
3039                                 *cp = '\0';
3040                         }
3041                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3042                             "destination already exists"));
3043                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3044                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3045                             zc.zc_value);
3046                         *cp = '@';
3047                         break;
3048                 case EINVAL:
3049                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3050                         break;
3051                 case ECKSUM:
3052                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3053                             "invalid stream (checksum mismatch)"));
3054                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3055                         break;
3056                 case ENOTSUP:
3057                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3058                             "pool must be upgraded to receive this stream."));
3059                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3060                         break;
3061                 case EDQUOT:
3062                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3063                             "destination %s space quota exceeded"), zc.zc_name);
3064                         (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3065                         break;
3066                 default:
3067                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3068                 }
3069         }
3070 
3071         /*
3072          * Mount the target filesystem (if created).  Also mount any
3073          * children of the target filesystem if we did a replication
3074          * receive (indicated by stream_avl being non-NULL).
3075          */
3076         cp = strchr(zc.zc_value, '@');
3077         if (cp && (ioctl_err == 0 || !newfs)) {
3078                 zfs_handle_t *h;
3079 
3080                 *cp = '\0';
3081                 h = zfs_open(hdl, zc.zc_value,
3082                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3083                 if (h != NULL) {
3084                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
3085                                 *cp = '@';
3086                         } else if (newfs || stream_avl) {
3087                                 /*
3088                                  * Track the first/top of hierarchy fs,
3089                                  * for mounting and sharing later.
3090                                  */
3091                                 if (top_zfs && *top_zfs == NULL)
3092                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
3093                         }
3094                         zfs_close(h);
3095                 }
3096                 *cp = '@';
3097         }
3098 
3099         if (clp) {
3100                 err |= changelist_postfix(clp);
3101                 changelist_free(clp);
3102         }
3103 
3104         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3105                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3106                     "failed to clear unreceived properties on %s"),
3107                     zc.zc_name);
3108                 (void) fprintf(stderr, "\n");
3109         }
3110         if (prop_errflags & ZPROP_ERR_NORESTORE) {
3111                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3112                     "failed to restore original properties on %s"),
3113                     zc.zc_name);
3114                 (void) fprintf(stderr, "\n");
3115         }
3116 
3117         if (err || ioctl_err)
3118                 return (-1);
3119 
3120         *action_handlep = zc.zc_action_handle;
3121 
3122         if (flags->verbose) {
3123                 char buf1[64];
3124                 char buf2[64];
3125                 uint64_t bytes = zc.zc_cookie;
3126                 time_t delta = time(NULL) - begin_time;
3127                 if (delta == 0)
3128                         delta = 1;
3129                 zfs_nicenum(bytes, buf1, sizeof (buf1));
3130                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3131 
3132                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3133                     buf1, delta, buf2);
3134         }
3135 
3136         return (0);
3137 }
3138 
3139 static int
3140 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3141     int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl,
3142     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
3143 {
3144         int err;
3145         dmu_replay_record_t drr, drr_noswap;
3146         struct drr_begin *drrb = &drr.drr_u.drr_begin;
3147         char errbuf[1024];
3148         zio_cksum_t zcksum = { 0 };
3149         uint64_t featureflags;
3150         int hdrtype;
3151 
3152         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3153             "cannot receive"));
3154 
3155         if (flags->isprefix &&
3156             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3157                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3158                     "(%s) does not exist"), tosnap);
3159                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3160         }
3161 
3162         /* read in the BEGIN record */
3163         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3164             &zcksum)))
3165                 return (err);
3166 
3167         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3168                 /* It's the double end record at the end of a package */
3169                 return (ENODATA);
3170         }
3171 
3172         /* the kernel needs the non-byteswapped begin record */
3173         drr_noswap = drr;
3174 
3175         flags->byteswap = B_FALSE;
3176         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3177                 /*
3178                  * We computed the checksum in the wrong byteorder in
3179                  * recv_read() above; do it again correctly.
3180                  */
3181                 bzero(&zcksum, sizeof (zio_cksum_t));
3182                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3183                 flags->byteswap = B_TRUE;
3184 
3185                 drr.drr_type = BSWAP_32(drr.drr_type);
3186                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3187                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3188                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3189                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3190                 drrb->drr_type = BSWAP_32(drrb->drr_type);
3191                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3192                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3193                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3194         }
3195 
3196         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3197                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3198                     "stream (bad magic number)"));
3199                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3200         }
3201 
3202         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3203         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3204 
3205         if (!DMU_STREAM_SUPPORTED(featureflags) ||
3206             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3207                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3208                     "stream has unsupported feature, feature flags = %lx"),
3209                     featureflags);
3210                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3211         }
3212 
3213         if (strchr(drrb->drr_toname, '@') == NULL) {
3214                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3215                     "stream (bad snapshot name)"));
3216                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3217         }
3218 
3219         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3220                 char nonpackage_sendfs[ZFS_MAXNAMELEN];
3221                 if (sendfs == NULL) {
3222                         /*
3223                          * We were not called from zfs_receive_package(). Get
3224                          * the fs specified by 'zfs send'.
3225                          */
3226                         char *cp;
3227                         (void) strlcpy(nonpackage_sendfs,
3228                             drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
3229                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3230                                 *cp = '\0';
3231                         sendfs = nonpackage_sendfs;
3232                 }
3233                 return (zfs_receive_one(hdl, infd, tosnap, flags,
3234                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl,
3235                     top_zfs, cleanup_fd, action_handlep));
3236         } else {
3237                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3238                     DMU_COMPOUNDSTREAM);
3239                 return (zfs_receive_package(hdl, infd, tosnap, flags,
3240                     &drr, &zcksum, top_zfs, cleanup_fd, action_handlep));
3241         }
3242 }
3243 
3244 /*
3245  * Restores a backup of tosnap from the file descriptor specified by infd.
3246  * Return 0 on total success, -2 if some things couldn't be
3247  * destroyed/renamed/promoted, -1 if some things couldn't be received.
3248  * (-1 will override -2).
3249  */
3250 int
3251 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3252     int infd, avl_tree_t *stream_avl)
3253 {
3254         char *top_zfs = NULL;
3255         int err;
3256         int cleanup_fd;
3257         uint64_t action_handle = 0;
3258 
3259         cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3260         VERIFY(cleanup_fd >= 0);
3261 
3262         err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL,
3263             stream_avl, &top_zfs, cleanup_fd, &action_handle);
3264 
3265         VERIFY(0 == close(cleanup_fd));
3266 
3267         if (err == 0 && !flags->nomount && top_zfs) {
3268                 zfs_handle_t *zhp;
3269                 prop_changelist_t *clp;
3270 
3271                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3272                 if (zhp != NULL) {
3273                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3274                             CL_GATHER_MOUNT_ALWAYS, 0);
3275                         zfs_close(zhp);
3276                         if (clp != NULL) {
3277                                 /* mount and share received datasets */
3278                                 err = changelist_postfix(clp);
3279                                 changelist_free(clp);
3280                         }
3281                 }
3282                 if (zhp == NULL || clp == NULL || err)
3283                         err = -1;
3284         }
3285         if (top_zfs)
3286                 free(top_zfs);
3287 
3288         return (err);
3289 }