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