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