lib/libzfs/libzfs_sendrecv.c

   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  */
  25
  26 #include <assert.h>
  27 #include <ctype.h>
  28 #include <errno.h>
  29 #include <libintl.h>
  30 #include <stdio.h>
  31 #include <stdlib.h>
  32 #include <strings.h>
  33 #include <unistd.h>
  34 #include <stddef.h>
  35 #include <fcntl.h>
  36 #include <sys/mount.h>
  37 #include <sys/mntent.h>
  38 #include <sys/mnttab.h>
  39 #include <sys/avl.h>
  40 #include <sys/debug.h>
  41 #include <stddef.h>
  42 #include <pthread.h>
  43 #include <umem.h>
  44
  45 #include <libzfs.h>
  46
  47 #include "zfs_namecheck.h"
  48 #include "zfs_prop.h"
  49 #include "zfs_fletcher.h"
  50 #include "libzfs_impl.h"
  51 #include <sys/zio_checksum.h>
  52 #include <sys/ddt.h>
  53 #include <sys/socket.h>
  54
  55 /* in libzfs_dataset.c */
  56 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  57
  58 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t,
  59     int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *);
  60
  61 static const zio_cksum_t zero_cksum = { { 0 } };
  62
  63 typedef struct dedup_arg {
  64         int     inputfd;
  65         int     outputfd;
  66         libzfs_handle_t  *dedup_hdl;
  67 } dedup_arg_t;
  68
  69 typedef struct dataref {
  70         uint64_t ref_guid;
  71         uint64_t ref_object;
  72         uint64_t ref_offset;
  73 } dataref_t;
  74
  75 typedef struct dedup_entry {
  76         struct dedup_entry      *dde_next;
  77         zio_cksum_t dde_chksum;
  78         uint64_t dde_prop;
  79         dataref_t dde_ref;
  80 } dedup_entry_t;
  81
  82 #define MAX_DDT_PHYSMEM_PERCENT         20
  83 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
  84
  85 typedef struct dedup_table {
  86         dedup_entry_t   **dedup_hash_array;
  87         umem_cache_t    *ddecache;
  88         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
  89         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
  90         uint64_t        ddt_count;
  91         int             numhashbits;
  92         boolean_t       ddt_full;
  93 } dedup_table_t;
  94
  95 static int
  96 high_order_bit(uint64_t n)
  97 {
  98         int count;
  99
 100         for (count = 0; n != 0; count++)
 101                 n >>= 1;
 102         return (count);
 103 }
 104
 105 static size_t
 106 ssread(void *buf, size_t len, FILE *stream)
 107 {
 108         size_t outlen;
 109
 110         if ((outlen = fread(buf, len, 1, stream)) == 0)
 111                 return (0);
 112
 113         return (outlen);
 114 }
 115
 116 static void
 117 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 118     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 119 {
 120         dedup_entry_t   *dde;
 121
 122         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 123                 if (ddt->ddt_full == B_FALSE) {
 124                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 125                             "Dedup table full.  Deduplication will continue "
 126                             "with existing table entries"));
 127                         ddt->ddt_full = B_TRUE;
 128                 }
 129                 return;
 130         }
 131
 132         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 133             != NULL) {
 134                 assert(*ddepp == NULL);
 135                 dde->dde_next = NULL;
 136                 dde->dde_chksum = *cs;
 137                 dde->dde_prop = prop;
 138                 dde->dde_ref = *dr;
 139                 *ddepp = dde;
 140                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 141                 ddt->ddt_count++;
 142         }
 143 }
 144
 145 /*
 146  * Using the specified dedup table, do a lookup for an entry with
 147  * the checksum cs.  If found, return the block's reference info
 148  * in *dr. Otherwise, insert a new entry in the dedup table, using
 149  * the reference information specified by *dr.
 150  *
 151  * return value:  true - entry was found
 152  *                false - entry was not found
 153  */
 154 static boolean_t
 155 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 156     uint64_t prop, dataref_t *dr)
 157 {
 158         uint32_t hashcode;
 159         dedup_entry_t **ddepp;
 160
 161         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 162
 163         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 164             ddepp = &((*ddepp)->dde_next)) {
 165                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 166                     (*ddepp)->dde_prop == prop) {
 167                         *dr = (*ddepp)->dde_ref;
 168                         return (B_TRUE);
 169                 }
 170         }
 171         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 172         return (B_FALSE);
 173 }
 174
 175 static int
 176 cksum_and_write(const void *buf, uint64_t len, zio_cksum_t *zc, int outfd)
 177 {
 178         fletcher_4_incremental_native(buf, len, zc);
 179         return (write(outfd, buf, len));
 180 }
 181
 182 /*
 183  * This function is started in a separate thread when the dedup option
 184  * has been requested.  The main send thread determines the list of
 185  * snapshots to be included in the send stream and makes the ioctl calls
 186  * for each one.  But instead of having the ioctl send the output to the
 187  * the output fd specified by the caller of zfs_send()), the
 188  * ioctl is told to direct the output to a pipe, which is read by the
 189  * alternate thread running THIS function.  This function does the
 190  * dedup'ing by:
 191  *  1. building a dedup table (the DDT)
 192  *  2. doing checksums on each data block and inserting a record in the DDT
 193  *  3. looking for matching checksums, and
 194  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 195  *      a duplicate block is found.
 196  * The output of this function then goes to the output fd requested
 197  * by the caller of zfs_send().
 198  */
 199 static void *
 200 cksummer(void *arg)
 201 {
 202         dedup_arg_t *dda = arg;
 203         char *buf = malloc(1<<20);
 204         dmu_replay_record_t thedrr;
 205         dmu_replay_record_t *drr = &thedrr;
 206         struct drr_begin *drrb = &thedrr.drr_u.drr_begin;
 207         struct drr_end *drre = &thedrr.drr_u.drr_end;
 208         struct drr_object *drro = &thedrr.drr_u.drr_object;
 209         struct drr_write *drrw = &thedrr.drr_u.drr_write;
 210         struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
 211         FILE *ofp;
 212         int outfd;
 213         dmu_replay_record_t wbr_drr = {0};
 214         struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref;
 215         dedup_table_t ddt;
 216         zio_cksum_t stream_cksum;
 217         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 218         uint64_t numbuckets;
 219
 220         ddt.max_ddt_size =
 221             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT)/100,
 222             SMALLEST_POSSIBLE_MAX_DDT_MB<<20);
 223
 224         numbuckets = ddt.max_ddt_size/(sizeof (dedup_entry_t));
 225
 226         /*
 227          * numbuckets must be a power of 2.  Increase number to
 228          * a power of 2 if necessary.
 229          */
 230         if (!ISP2(numbuckets))
 231                 numbuckets = 1 << high_order_bit(numbuckets);
 232
 233         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 234         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 235             NULL, NULL, NULL, NULL, NULL, 0);
 236         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 237         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 238         ddt.ddt_full = B_FALSE;
 239
 240         /* Initialize the write-by-reference block. */
 241         wbr_drr.drr_type = DRR_WRITE_BYREF;
 242         wbr_drr.drr_payloadlen = 0;
 243
 244         outfd = dda->outputfd;
 245         ofp = fdopen(dda->inputfd, "r");
 246         while (ssread(drr, sizeof (dmu_replay_record_t), ofp) != 0) {
 247
 248                 switch (drr->drr_type) {
 249                 case DRR_BEGIN:
 250                 {
 251                         int     fflags;
 252                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 253
 254                         /* set the DEDUP feature flag for this stream */
 255                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 256                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 257                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 258                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 259
 260                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 261                             &stream_cksum, outfd) == -1)
 262                                 goto out;
 263                         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 264                             DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
 265                                 int sz = drr->drr_payloadlen;
 266
 267                                 if (sz > 1<<20) {
 268                                         free(buf);
 269                                         buf = malloc(sz);
 270                                 }
 271                                 (void) ssread(buf, sz, ofp);
 272                                 if (ferror(stdin))
 273                                         perror("fread");
 274                                 if (cksum_and_write(buf, sz, &stream_cksum,
 275                                     outfd) == -1)
 276                                         goto out;
 277                         }
 278                         break;
 279                 }
 280
 281                 case DRR_END:
 282                 {
 283                         /* use the recalculated checksum */
 284                         ZIO_SET_CHECKSUM(&drre->drr_checksum,
 285                             stream_cksum.zc_word[0], stream_cksum.zc_word[1],
 286                             stream_cksum.zc_word[2], stream_cksum.zc_word[3]);
 287                         if ((write(outfd, drr,
 288                             sizeof (dmu_replay_record_t))) == -1)
 289                                 goto out;
 290                         break;
 291                 }
 292
 293                 case DRR_OBJECT:
 294                 {
 295                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 296                             &stream_cksum, outfd) == -1)
 297                                 goto out;
 298                         if (drro->drr_bonuslen > 0) {
 299                                 (void) ssread(buf,
 300                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 301                                     ofp);
 302                                 if (cksum_and_write(buf,
 303                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 304                                     &stream_cksum, outfd) == -1)
 305                                         goto out;
 306                         }
 307                         break;
 308                 }
 309
 310                 case DRR_SPILL:
 311                 {
 312                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 313                             &stream_cksum, outfd) == -1)
 314                                 goto out;
 315                         (void) ssread(buf, drrs->drr_length, ofp);
 316                         if (cksum_and_write(buf, drrs->drr_length,
 317                             &stream_cksum, outfd) == -1)
 318                                 goto out;
 319                         break;
 320                 }
 321
 322                 case DRR_FREEOBJECTS:
 323                 {
 324                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 325                             &stream_cksum, outfd) == -1)
 326                                 goto out;
 327                         break;
 328                 }
 329
 330                 case DRR_WRITE:
 331                 {
 332                         dataref_t       dataref;
 333
 334                         (void) ssread(buf, drrw->drr_length, ofp);
 335
 336                         /*
 337                          * Use the existing checksum if it's dedup-capable,
 338                          * else calculate a SHA256 checksum for it.
 339                          */
 340
 341                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 342                             zero_cksum) ||
 343                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 344                                 zio_cksum_t tmpsha256;
 345
 346                                 zio_checksum_SHA256(buf,
 347                                     drrw->drr_length, &tmpsha256);
 348
 349                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 350                                     BE_64(tmpsha256.zc_word[0]);
 351                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 352                                     BE_64(tmpsha256.zc_word[1]);
 353                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 354                                     BE_64(tmpsha256.zc_word[2]);
 355                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 356                                     BE_64(tmpsha256.zc_word[3]);
 357                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 358                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 359                         }
 360
 361                         dataref.ref_guid = drrw->drr_toguid;
 362                         dataref.ref_object = drrw->drr_object;
 363                         dataref.ref_offset = drrw->drr_offset;
 364
 365                         if (ddt_update(dda->dedup_hdl, &ddt,
 366                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 367                             &dataref)) {
 368                                 /* block already present in stream */
 369                                 wbr_drrr->drr_object = drrw->drr_object;
 370                                 wbr_drrr->drr_offset = drrw->drr_offset;
 371                                 wbr_drrr->drr_length = drrw->drr_length;
 372                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 373                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 374                                 wbr_drrr->drr_refobject =
 375                                     dataref.ref_object;
 376                                 wbr_drrr->drr_refoffset =
 377                                     dataref.ref_offset;
 378
 379                                 wbr_drrr->drr_checksumtype =
 380                                     drrw->drr_checksumtype;
 381                                 wbr_drrr->drr_checksumflags =
 382                                     drrw->drr_checksumtype;
 383                                 wbr_drrr->drr_key.ddk_cksum =
 384                                     drrw->drr_key.ddk_cksum;
 385                                 wbr_drrr->drr_key.ddk_prop =
 386                                     drrw->drr_key.ddk_prop;
 387
 388                                 if (cksum_and_write(&wbr_drr,
 389                                     sizeof (dmu_replay_record_t), &stream_cksum,
 390                                     outfd) == -1)
 391                                         goto out;
 392                         } else {
 393                                 /* block not previously seen */
 394                                 if (cksum_and_write(drr,
 395                                     sizeof (dmu_replay_record_t), &stream_cksum,
 396                                     outfd) == -1)
 397                                         goto out;
 398                                 if (cksum_and_write(buf,
 399                                     drrw->drr_length,
 400                                     &stream_cksum, outfd) == -1)
 401                                         goto out;
 402                         }
 403                         break;
 404                 }
 405
 406                 case DRR_FREE:
 407                 {
 408                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 409                             &stream_cksum, outfd) == -1)
 410                                 goto out;
 411                         break;
 412                 }
 413
 414                 default:
 415                         (void) printf("INVALID record type 0x%x\n",
 416                             drr->drr_type);
 417                         /* should never happen, so assert */
 418                         assert(B_FALSE);
 419                 }
 420         }
 421 out:
 422         umem_cache_destroy(ddt.ddecache);
 423         free(ddt.dedup_hash_array);
 424         free(buf);
 425         (void) fclose(ofp);
 426
 427         return (NULL);
 428 }
 429
 430 /*
 431  * Routines for dealing with the AVL tree of fs-nvlists
 432  */
 433 typedef struct fsavl_node {
 434         avl_node_t fn_node;
 435         nvlist_t *fn_nvfs;
 436         char *fn_snapname;
 437         uint64_t fn_guid;
 438 } fsavl_node_t;
 439
 440 static int
 441 fsavl_compare(const void *arg1, const void *arg2)
 442 {
 443         const fsavl_node_t *fn1 = arg1;
 444         const fsavl_node_t *fn2 = arg2;
 445
 446         if (fn1->fn_guid > fn2->fn_guid)
 447                 return (+1);
 448         else if (fn1->fn_guid < fn2->fn_guid)
 449                 return (-1);
 450         else
 451                 return (0);
 452 }
 453
 454 /*
 455  * Given the GUID of a snapshot, find its containing filesystem and
 456  * (optionally) name.
 457  */
 458 static nvlist_t *
 459 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 460 {
 461         fsavl_node_t fn_find;
 462         fsavl_node_t *fn;
 463
 464         fn_find.fn_guid = snapguid;
 465
 466         fn = avl_find(avl, &fn_find, NULL);
 467         if (fn) {
 468                 if (snapname)
 469                         *snapname = fn->fn_snapname;
 470                 return (fn->fn_nvfs);
 471         }
 472         return (NULL);
 473 }
 474
 475 static void
 476 fsavl_destroy(avl_tree_t *avl)
 477 {
 478         fsavl_node_t *fn;
 479         void *cookie;
 480
 481         if (avl == NULL)
 482                 return;
 483
 484         cookie = NULL;
 485         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 486                 free(fn);
 487         avl_destroy(avl);
 488         free(avl);
 489 }
 490
 491 /*
 492  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 493  */
 494 static avl_tree_t *
 495 fsavl_create(nvlist_t *fss)
 496 {
 497         avl_tree_t *fsavl;
 498         nvpair_t *fselem = NULL;
 499
 500         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 501                 return (NULL);
 502
 503         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 504             offsetof(fsavl_node_t, fn_node));
 505
 506         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 507                 nvlist_t *nvfs, *snaps;
 508                 nvpair_t *snapelem = NULL;
 509
 510                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 511                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 512
 513                 while ((snapelem =
 514                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 515                         fsavl_node_t *fn;
 516                         uint64_t guid;
 517
 518                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 519                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 520                                 fsavl_destroy(fsavl);
 521                                 return (NULL);
 522                         }
 523                         fn->fn_nvfs = nvfs;
 524                         fn->fn_snapname = nvpair_name(snapelem);
 525                         fn->fn_guid = guid;
 526
 527                         /*
 528                          * Note: if there are multiple snaps with the
 529                          * same GUID, we ignore all but one.
 530                          */
 531                         if (avl_find(fsavl, fn, NULL) == NULL)
 532                                 avl_add(fsavl, fn);
 533                         else
 534                                 free(fn);
 535                 }
 536         }
 537
 538         return (fsavl);
 539 }
 540
 541 /*
 542  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 543  */
 544 typedef struct send_data {
 545         uint64_t parent_fromsnap_guid;
 546         nvlist_t *parent_snaps;
 547         nvlist_t *fss;
 548         nvlist_t *snapprops;
 549         const char *fromsnap;
 550         const char *tosnap;
 551         boolean_t recursive;
 552
 553         /*
 554          * The header nvlist is of the following format:
 555          * {
 556          *   "tosnap" -> string
 557          *   "fromsnap" -> string (if incremental)
 558          *   "fss" -> {
 559          *      id -> {
 560          *
 561          *       "name" -> string (full name; for debugging)
 562          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 563          *
 564          *       "props" -> { name -> value (only if set here) }
 565          *       "snaps" -> { name (lastname) -> number (guid) }
 566          *       "snapprops" -> { name (lastname) -> { name -> value } }
 567          *
 568          *       "origin" -> number (guid) (if clone)
 569          *       "sent" -> boolean (not on-disk)
 570          *      }
 571          *   }
 572          * }
 573          *
 574          */
 575 } send_data_t;
 576
 577 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 578
 579 static int
 580 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 581 {
 582         send_data_t *sd = arg;
 583         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 584         char *snapname;
 585         nvlist_t *nv;
 586
 587         snapname = strrchr(zhp->zfs_name, '@')+1;
 588
 589         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 590         /*
 591          * NB: if there is no fromsnap here (it's a newly created fs in
 592          * an incremental replication), we will substitute the tosnap.
 593          */
 594         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 595             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 596             strcmp(snapname, sd->tosnap) == 0)) {
 597                 sd->parent_fromsnap_guid = guid;
 598         }
 599
 600         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 601         send_iterate_prop(zhp, nv);
 602         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 603         nvlist_free(nv);
 604
 605         zfs_close(zhp);
 606         return (0);
 607 }
 608
 609 static void
 610 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 611 {
 612         nvpair_t *elem = NULL;
 613
 614         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 615                 char *propname = nvpair_name(elem);
 616                 zfs_prop_t prop = zfs_name_to_prop(propname);
 617                 nvlist_t *propnv;
 618
 619                 if (!zfs_prop_user(propname)) {
 620                         /*
 621                          * Realistically, this should never happen.  However,
 622                          * we want the ability to add DSL properties without
 623                          * needing to make incompatible version changes.  We
 624                          * need to ignore unknown properties to allow older
 625                          * software to still send datasets containing these
 626                          * properties, with the unknown properties elided.
 627                          */
 628                         if (prop == ZPROP_INVAL)
 629                                 continue;
 630
 631                         if (zfs_prop_readonly(prop))
 632                                 continue;
 633                 }
 634
 635                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 636                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 637                     prop == ZFS_PROP_REFQUOTA ||
 638                     prop == ZFS_PROP_REFRESERVATION) {
 639                         char *source;
 640                         uint64_t value;
 641                         verify(nvlist_lookup_uint64(propnv,
 642                             ZPROP_VALUE, &value) == 0);
 643                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 644                                 continue;
 645                         /*
 646                          * May have no source before SPA_VERSION_RECVD_PROPS,
 647                          * but is still modifiable.
 648                          */
 649                         if (nvlist_lookup_string(propnv,
 650                             ZPROP_SOURCE, &source) == 0) {
 651                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 652                                     (strcmp(source,
 653                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 654                                         continue;
 655                         }
 656                 } else {
 657                         char *source;
 658                         if (nvlist_lookup_string(propnv,
 659                             ZPROP_SOURCE, &source) != 0)
 660                                 continue;
 661                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 662                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 663                                 continue;
 664                 }
 665
 666                 if (zfs_prop_user(propname) ||
 667                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 668                         char *value;
 669                         verify(nvlist_lookup_string(propnv,
 670                             ZPROP_VALUE, &value) == 0);
 671                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 672                 } else {
 673                         uint64_t value;
 674                         verify(nvlist_lookup_uint64(propnv,
 675                             ZPROP_VALUE, &value) == 0);
 676                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 677                 }
 678         }
 679 }
 680
 681 /*
 682  * recursively generate nvlists describing datasets.  See comment
 683  * for the data structure send_data_t above for description of contents
 684  * of the nvlist.
 685  */
 686 static int
 687 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 688 {
 689         send_data_t *sd = arg;
 690         nvlist_t *nvfs, *nv;
 691         int rv = 0;
 692         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 693         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 694         char guidstring[64];
 695
 696         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 697         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 698         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 699             sd->parent_fromsnap_guid));
 700
 701         if (zhp->zfs_dmustats.dds_origin[0]) {
 702                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 703                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 704                 if (origin == NULL)
 705                         return (-1);
 706                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 707                     origin->zfs_dmustats.dds_guid));
 708         }
 709
 710         /* iterate over props */
 711         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 712         send_iterate_prop(zhp, nv);
 713         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 714         nvlist_free(nv);
 715
 716         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 717         sd->parent_fromsnap_guid = 0;
 718         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 719         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 720         (void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
 721         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 722         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 723         nvlist_free(sd->parent_snaps);
 724         nvlist_free(sd->snapprops);
 725
 726         /* add this fs to nvlist */
 727         (void) snprintf(guidstring, sizeof (guidstring),
 728             "0x%llx", (longlong_t)guid);
 729         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 730         nvlist_free(nvfs);
 731
 732         /* iterate over children */
 733         if (sd->recursive)
 734                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 735
 736         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 737
 738         zfs_close(zhp);
 739         return (rv);
 740 }
 741
 742 static int
 743 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 744     const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
 745 {
 746         zfs_handle_t *zhp;
 747         send_data_t sd = { 0 };
 748         int error;
 749
 750         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 751         if (zhp == NULL)
 752                 return (EZFS_BADTYPE);
 753
 754         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 755         sd.fromsnap = fromsnap;
 756         sd.tosnap = tosnap;
 757         sd.recursive = recursive;
 758
 759         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 760                 nvlist_free(sd.fss);
 761                 if (avlp != NULL)
 762                         *avlp = NULL;
 763                 *nvlp = NULL;
 764                 return (error);
 765         }
 766
 767         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 768                 nvlist_free(sd.fss);
 769                 *nvlp = NULL;
 770                 return (EZFS_NOMEM);
 771         }
 772
 773         *nvlp = sd.fss;
 774         return (0);
 775 }
 776
 777 /*
 778  * Routines for dealing with the sorted snapshot functionality
 779  */
 780 typedef struct zfs_node {
 781         zfs_handle_t    *zn_handle;
 782         avl_node_t      zn_avlnode;
 783 } zfs_node_t;
 784
 785 static int
 786 zfs_sort_snaps(zfs_handle_t *zhp, void *data)
 787 {
 788         avl_tree_t *avl = data;
 789         zfs_node_t *node;
 790         zfs_node_t search;
 791
 792         search.zn_handle = zhp;
 793         node = avl_find(avl, &search, NULL);
 794         if (node) {
 795                 /*
 796                  * If this snapshot was renamed while we were creating the
 797                  * AVL tree, it's possible that we already inserted it under
 798                  * its old name. Remove the old handle before adding the new
 799                  * one.
 800                  */
 801                 zfs_close(node->zn_handle);
 802                 avl_remove(avl, node);
 803                 free(node);
 804         }
 805
 806         node = zfs_alloc(zhp->zfs_hdl, sizeof (zfs_node_t));
 807         node->zn_handle = zhp;
 808         avl_add(avl, node);
 809
 810         return (0);
 811 }
 812
 813 static int
 814 zfs_snapshot_compare(const void *larg, const void *rarg)
 815 {
 816         zfs_handle_t *l = ((zfs_node_t *)larg)->zn_handle;
 817         zfs_handle_t *r = ((zfs_node_t *)rarg)->zn_handle;
 818         uint64_t lcreate, rcreate;
 819
 820         /*
 821          * Sort them according to creation time.  We use the hidden
 822          * CREATETXG property to get an absolute ordering of snapshots.
 823          */
 824         lcreate = zfs_prop_get_int(l, ZFS_PROP_CREATETXG);
 825         rcreate = zfs_prop_get_int(r, ZFS_PROP_CREATETXG);
 826
 827         if (lcreate < rcreate)
 828                 return (-1);
 829         else if (lcreate > rcreate)
 830                 return (+1);
 831         else
 832                 return (0);
 833 }
 834
 835 int
 836 zfs_iter_snapshots_sorted(zfs_handle_t *zhp, zfs_iter_f callback, void *data)
 837 {
 838         int ret = 0;
 839         zfs_node_t *node;
 840         avl_tree_t avl;
 841         void *cookie = NULL;
 842
 843         avl_create(&avl, zfs_snapshot_compare,
 844             sizeof (zfs_node_t), offsetof(zfs_node_t, zn_avlnode));
 845
 846         ret = zfs_iter_snapshots(zhp, zfs_sort_snaps, &avl);
 847
 848         for (node = avl_first(&avl); node != NULL; node = AVL_NEXT(&avl, node))
 849                 ret |= callback(node->zn_handle, data);
 850
 851         while ((node = avl_destroy_nodes(&avl, &cookie)) != NULL)
 852                 free(node);
 853
 854         avl_destroy(&avl);
 855
 856         return (ret);
 857 }
 858
 859 /*
 860  * Routines specific to "zfs send"
 861  */
 862 typedef struct send_dump_data {
 863         /* these are all just the short snapname (the part after the @) */
 864         const char *fromsnap;
 865         const char *tosnap;
 866         char prevsnap[ZFS_MAXNAMELEN];
 867         uint64_t prevsnap_obj;
 868         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 869         boolean_t verbose;
 870         int outfd;
 871         boolean_t err;
 872         nvlist_t *fss;
 873         avl_tree_t *fsavl;
 874         snapfilter_cb_t *filter_cb;
 875         void *filter_cb_arg;
 876         nvlist_t *debugnv;
 877         char holdtag[ZFS_MAXNAMELEN];
 878         int cleanup_fd;
 879 } send_dump_data_t;
 880
 881 /*
 882  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 883  * NULL) to the file descriptor specified by outfd.
 884  */
 885 static int
 886 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
 887     boolean_t fromorigin, int outfd, nvlist_t *debugnv)
 888 {
 889         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
 890         libzfs_handle_t *hdl = zhp->zfs_hdl;
 891         nvlist_t *thisdbg;
 892
 893         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 894         assert(fromsnap_obj == 0 || !fromorigin);
 895
 896         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 897         zc.zc_cookie = outfd;
 898         zc.zc_obj = fromorigin;
 899         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 900         zc.zc_fromobj = fromsnap_obj;
 901
 902         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 903         if (fromsnap && fromsnap[0] != '\0') {
 904                 VERIFY(0 == nvlist_add_string(thisdbg,
 905                     "fromsnap", fromsnap));
 906         }
 907
 908         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_SEND, &zc) != 0) {
 909                 char errbuf[1024];
 910                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 911                     "warning: cannot send '%s'"), zhp->zfs_name);
 912
 913                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
 914                 if (debugnv) {
 915                         VERIFY(0 == nvlist_add_nvlist(debugnv,
 916                             zhp->zfs_name, thisdbg));
 917                 }
 918                 nvlist_free(thisdbg);
 919
 920                 switch (errno) {
 921
 922                 case EXDEV:
 923                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 924                             "not an earlier snapshot from the same fs"));
 925                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 926
 927                 case ENOENT:
 928                         if (zfs_dataset_exists(hdl, zc.zc_name,
 929                             ZFS_TYPE_SNAPSHOT)) {
 930                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 931                                     "incremental source (@%s) does not exist"),
 932                                     zc.zc_value);
 933                         }
 934                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 935
 936                 case EDQUOT:
 937                 case EFBIG:
 938                 case EIO:
 939                 case ENOLINK:
 940                 case ENOSPC:
 941                 case ENOSTR:
 942                 case ENXIO:
 943                 case EPIPE:
 944                 case ERANGE:
 945                 case EFAULT:
 946                 case EROFS:
 947                         zfs_error_aux(hdl, strerror(errno));
 948                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 949
 950                 default:
 951                         return (zfs_standard_error(hdl, errno, errbuf));
 952                 }
 953         }
 954
 955         if (debugnv)
 956                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
 957         nvlist_free(thisdbg);
 958
 959         return (0);
 960 }
 961
 962 static int
 963 hold_for_send(zfs_handle_t *zhp, send_dump_data_t *sdd)
 964 {
 965         zfs_handle_t *pzhp;
 966         int error = 0;
 967         char *thissnap;
 968
 969         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 970
 971         /*
 972          * zfs_send() only opens a cleanup_fd for sends that need it,
 973          * e.g. replication and doall.
 974          */
 975         if (sdd->cleanup_fd == -1)
 976                 return (0);
 977
 978         thissnap = strchr(zhp->zfs_name, '@') + 1;
 979         *(thissnap - 1) = '\0';
 980         pzhp = zfs_open(zhp->zfs_hdl, zhp->zfs_name, ZFS_TYPE_DATASET);
 981         *(thissnap - 1) = '@';
 982
 983         /*
 984          * It's OK if the parent no longer exists.  The send code will
 985          * handle that error.
 986          */
 987         if (pzhp) {
 988                 error = zfs_hold(pzhp, thissnap, sdd->holdtag,
 989                     B_FALSE, B_TRUE, B_TRUE, sdd->cleanup_fd,
 990                     zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID),
 991                     zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG));
 992                 zfs_close(pzhp);
 993         }
 994
 995         return (error);
 996 }
 997
 998 static int
 999 dump_snapshot(zfs_handle_t *zhp, void *arg)
1000 {
1001         send_dump_data_t *sdd = arg;
1002         char *thissnap;
1003         int err;
1004         boolean_t isfromsnap, istosnap;
1005         boolean_t exclude = B_FALSE;
1006
1007         thissnap = strchr(zhp->zfs_name, '@') + 1;
1008         isfromsnap = (sdd->fromsnap != NULL &&
1009             strcmp(sdd->fromsnap, thissnap) == 0);
1010
1011         if (!sdd->seenfrom && isfromsnap) {
1012                 err = hold_for_send(zhp, sdd);
1013                 if (err == 0) {
1014                         sdd->seenfrom = B_TRUE;
1015                         (void) strcpy(sdd->prevsnap, thissnap);
1016                         sdd->prevsnap_obj = zfs_prop_get_int(zhp,
1017                             ZFS_PROP_OBJSETID);
1018                 } else if (err == ENOENT) {
1019                         err = 0;
1020                 }
1021                 zfs_close(zhp);
1022                 return (err);
1023         }
1024
1025         if (sdd->seento || !sdd->seenfrom) {
1026                 zfs_close(zhp);
1027                 return (0);
1028         }
1029
1030         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1031         if (istosnap)
1032                 sdd->seento = B_TRUE;
1033
1034         if (!sdd->doall && !isfromsnap && !istosnap) {
1035                 if (sdd->replicate) {
1036                         char *snapname;
1037                         nvlist_t *snapprops;
1038                         /*
1039                          * Filter out all intermediate snapshots except origin
1040                          * snapshots needed to replicate clones.
1041                          */
1042                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1043                             zhp->zfs_dmustats.dds_guid, &snapname);
1044
1045                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1046                             "snapprops", &snapprops));
1047                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1048                             thissnap, &snapprops));
1049                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
1050                 } else {
1051                         exclude = B_TRUE;
1052                 }
1053         }
1054
1055         /*
1056          * If a filter function exists, call it to determine whether
1057          * this snapshot will be sent.
1058          */
1059         if (exclude || (sdd->filter_cb != NULL &&
1060             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1061                 /*
1062                  * This snapshot is filtered out.  Don't send it, and don't
1063                  * set prevsnap_obj, so it will be as if this snapshot didn't
1064                  * exist, and the next accepted snapshot will be sent as
1065                  * an incremental from the last accepted one, or as the
1066                  * first (and full) snapshot in the case of a replication,
1067                  * non-incremental send.
1068                  */
1069                 zfs_close(zhp);
1070                 return (0);
1071         }
1072
1073         err = hold_for_send(zhp, sdd);
1074         if (err) {
1075                 if (err == ENOENT)
1076                         err = 0;
1077                 zfs_close(zhp);
1078                 return (err);
1079         }
1080
1081         /* send it */
1082         if (sdd->verbose) {
1083                 (void) fprintf(stderr, "sending from @%s to %s\n",
1084                     sdd->prevsnap, zhp->zfs_name);
1085         }
1086
1087         err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1088             sdd->prevsnap[0] == '\0' && (sdd->fromorigin || sdd->replicate),
1089             sdd->outfd, sdd->debugnv);
1090
1091         (void) strcpy(sdd->prevsnap, thissnap);
1092         sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1093         zfs_close(zhp);
1094         return (err);
1095 }
1096
1097 static int
1098 dump_filesystem(zfs_handle_t *zhp, void *arg)
1099 {
1100         int rv = 0;
1101         send_dump_data_t *sdd = arg;
1102         boolean_t missingfrom = B_FALSE;
1103         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1104
1105         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1106             zhp->zfs_name, sdd->tosnap);
1107         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1108                 (void) fprintf(stderr, "WARNING: "
1109                     "could not send %s@%s: does not exist\n",
1110                     zhp->zfs_name, sdd->tosnap);
1111                 sdd->err = B_TRUE;
1112                 return (0);
1113         }
1114
1115         if (sdd->replicate && sdd->fromsnap) {
1116                 /*
1117                  * If this fs does not have fromsnap, and we're doing
1118                  * recursive, we need to send a full stream from the
1119                  * beginning (or an incremental from the origin if this
1120                  * is a clone).  If we're doing non-recursive, then let
1121                  * them get the error.
1122                  */
1123                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1124                     zhp->zfs_name, sdd->fromsnap);
1125                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1126                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1127                         missingfrom = B_TRUE;
1128                 }
1129         }
1130
1131         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1132         sdd->prevsnap_obj = 0;
1133         if (sdd->fromsnap == NULL || missingfrom)
1134                 sdd->seenfrom = B_TRUE;
1135
1136         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1137         if (!sdd->seenfrom) {
1138                 (void) fprintf(stderr,
1139                     "WARNING: could not send %s@%s:\n"
1140                     "incremental source (%s@%s) does not exist\n",
1141                     zhp->zfs_name, sdd->tosnap,
1142                     zhp->zfs_name, sdd->fromsnap);
1143                 sdd->err = B_TRUE;
1144         } else if (!sdd->seento) {
1145                 if (sdd->fromsnap) {
1146                         (void) fprintf(stderr,
1147                             "WARNING: could not send %s@%s:\n"
1148                             "incremental source (%s@%s) "
1149                             "is not earlier than it\n",
1150                             zhp->zfs_name, sdd->tosnap,
1151                             zhp->zfs_name, sdd->fromsnap);
1152                 } else {
1153                         (void) fprintf(stderr, "WARNING: "
1154                             "could not send %s@%s: does not exist\n",
1155                             zhp->zfs_name, sdd->tosnap);
1156                 }
1157                 sdd->err = B_TRUE;
1158         }
1159
1160         return (rv);
1161 }
1162
1163 static int
1164 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1165 {
1166         send_dump_data_t *sdd = arg;
1167         nvpair_t *fspair;
1168         boolean_t needagain, progress;
1169
1170         if (!sdd->replicate)
1171                 return (dump_filesystem(rzhp, sdd));
1172
1173         /* Mark the clone origin snapshots. */
1174         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1175             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1176                 nvlist_t *nvfs;
1177                 uint64_t origin_guid = 0;
1178
1179                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1180                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1181                 if (origin_guid != 0) {
1182                         char *snapname;
1183                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1184                             origin_guid, &snapname);
1185                         if (origin_nv != NULL) {
1186                                 nvlist_t *snapprops;
1187                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1188                                     "snapprops", &snapprops));
1189                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1190                                     snapname, &snapprops));
1191                                 VERIFY(0 == nvlist_add_boolean(
1192                                     snapprops, "is_clone_origin"));
1193                         }
1194                 }
1195         }
1196 again:
1197         needagain = progress = B_FALSE;
1198         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1199             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1200                 nvlist_t *fslist;
1201                 char *fsname;
1202                 zfs_handle_t *zhp;
1203                 int err;
1204                 uint64_t origin_guid = 0;
1205
1206                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1207                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1208                         continue;
1209
1210                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1211                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1212
1213                 if (origin_guid != 0) {
1214                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1215                             origin_guid, NULL);
1216                         if (origin_nv != NULL &&
1217                             nvlist_lookup_boolean(origin_nv,
1218                             "sent") == ENOENT) {
1219                                 /*
1220                                  * origin has not been sent yet;
1221                                  * skip this clone.
1222                                  */
1223                                 needagain = B_TRUE;
1224                                 continue;
1225                         }
1226                 }
1227
1228                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1229                 if (zhp == NULL)
1230                         return (-1);
1231                 err = dump_filesystem(zhp, sdd);
1232                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1233                 progress = B_TRUE;
1234                 zfs_close(zhp);
1235                 if (err)
1236                         return (err);
1237         }
1238         if (needagain) {
1239                 assert(progress);
1240                 goto again;
1241         }
1242         return (0);
1243 }
1244
1245 /*
1246  * Generate a send stream for the dataset identified by the argument zhp.
1247  *
1248  * The content of the send stream is the snapshot identified by
1249  * 'tosnap'.  Incremental streams are requested in two ways:
1250  *     - from the snapshot identified by "fromsnap" (if non-null) or
1251  *     - from the origin of the dataset identified by zhp, which must
1252  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1253  *       is TRUE.
1254  *
1255  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1256  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1257  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1258  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1259  * case too. If "props" is set, send properties.
1260  */
1261 int
1262 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1263     sendflags_t flags, int outfd, snapfilter_cb_t filter_func,
1264     void *cb_arg, nvlist_t **debugnvp)
1265 {
1266         char errbuf[1024];
1267         send_dump_data_t sdd = { 0 };
1268         int err;
1269         nvlist_t *fss = NULL;
1270         avl_tree_t *fsavl = NULL;
1271         static uint64_t holdseq;
1272         int spa_version;
1273         boolean_t holdsnaps = B_FALSE;
1274         pthread_t tid;
1275         int pipefd[2];
1276         dedup_arg_t dda = { 0 };
1277         int featureflags = 0;
1278
1279         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1280             "cannot send '%s'"), zhp->zfs_name);
1281
1282         if (fromsnap && fromsnap[0] == '\0') {
1283                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1284                     "zero-length incremental source"));
1285                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1286         }
1287
1288         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1289                 uint64_t version;
1290                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1291                 if (version >= ZPL_VERSION_SA) {
1292                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1293                 }
1294         }
1295
1296         if (zfs_spa_version(zhp, &spa_version) == 0 &&
1297             spa_version >= SPA_VERSION_USERREFS &&
1298             (flags.doall || flags.replicate))
1299                 holdsnaps = B_TRUE;
1300
1301         if (flags.dedup) {
1302                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1303                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1304                 if ((err = socketpair(AF_UNIX, SOCK_STREAM, 0, pipefd))) {
1305                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1306                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1307                             errbuf));
1308                 }
1309                 dda.outputfd = outfd;
1310                 dda.inputfd = pipefd[1];
1311                 dda.dedup_hdl = zhp->zfs_hdl;
1312                 if ((err = pthread_create(&tid, NULL, cksummer, &dda))) {
1313                         (void) close(pipefd[0]);
1314                         (void) close(pipefd[1]);
1315                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1316                         return (zfs_error(zhp->zfs_hdl,
1317                             EZFS_THREADCREATEFAILED, errbuf));
1318                 }
1319         }
1320
1321         if (flags.replicate || flags.doall || flags.props) {
1322                 dmu_replay_record_t drr = { 0 };
1323                 char *packbuf = NULL;
1324                 size_t buflen = 0;
1325                 zio_cksum_t zc = { { 0 } };
1326
1327                 if (flags.replicate || flags.props) {
1328                         nvlist_t *hdrnv;
1329
1330                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1331                         if (fromsnap) {
1332                                 VERIFY(0 == nvlist_add_string(hdrnv,
1333                                     "fromsnap", fromsnap));
1334                         }
1335                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1336                         if (!flags.replicate) {
1337                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1338                                     "not_recursive"));
1339                         }
1340
1341                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1342                             fromsnap, tosnap, flags.replicate, &fss, &fsavl);
1343                         if (err)
1344                                 goto err_out;
1345                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1346                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1347                             NV_ENCODE_XDR, 0);
1348                         if (debugnvp)
1349                                 *debugnvp = hdrnv;
1350                         else
1351                                 nvlist_free(hdrnv);
1352                         if (err) {
1353                                 fsavl_destroy(fsavl);
1354                                 nvlist_free(fss);
1355                                 goto stderr_out;
1356                         }
1357                 }
1358
1359                 /* write first begin record */
1360                 drr.drr_type = DRR_BEGIN;
1361                 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1362                 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.drr_versioninfo,
1363                     DMU_COMPOUNDSTREAM);
1364                 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.drr_versioninfo,
1365                     featureflags);
1366                 (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1367                     sizeof (drr.drr_u.drr_begin.drr_toname),
1368                     "%s@%s", zhp->zfs_name, tosnap);
1369                 drr.drr_payloadlen = buflen;
1370                 err = cksum_and_write(&drr, sizeof (drr), &zc, outfd);
1371
1372                 /* write header nvlist */
1373                 if (err != -1 && packbuf != NULL) {
1374                         err = cksum_and_write(packbuf, buflen, &zc, outfd);
1375                 }
1376                 free(packbuf);
1377                 if (err == -1) {
1378                         fsavl_destroy(fsavl);
1379                         nvlist_free(fss);
1380                         err = errno;
1381                         goto stderr_out;
1382                 }
1383
1384                 /* write end record */
1385                 if (err != -1) {
1386                         bzero(&drr, sizeof (drr));
1387                         drr.drr_type = DRR_END;
1388                         drr.drr_u.drr_end.drr_checksum = zc;
1389                         err = write(outfd, &drr, sizeof (drr));
1390                         if (err == -1) {
1391                                 fsavl_destroy(fsavl);
1392                                 nvlist_free(fss);
1393                                 err = errno;
1394                                 goto stderr_out;
1395                         }
1396                 }
1397         }
1398
1399         /* dump each stream */
1400         sdd.fromsnap = fromsnap;
1401         sdd.tosnap = tosnap;
1402         if (flags.dedup)
1403                 sdd.outfd = pipefd[0];
1404         else
1405                 sdd.outfd = outfd;
1406         sdd.replicate = flags.replicate;
1407         sdd.doall = flags.doall;
1408         sdd.fromorigin = flags.fromorigin;
1409         sdd.fss = fss;
1410         sdd.fsavl = fsavl;
1411         sdd.verbose = flags.verbose;
1412         sdd.filter_cb = filter_func;
1413         sdd.filter_cb_arg = cb_arg;
1414         if (debugnvp)
1415                 sdd.debugnv = *debugnvp;
1416         if (holdsnaps) {
1417                 ++holdseq;
1418                 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1419                     ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1420                 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR);
1421                 if (sdd.cleanup_fd < 0) {
1422                         err = errno;
1423                         goto stderr_out;
1424                 }
1425         } else {
1426                 sdd.cleanup_fd = -1;
1427         }
1428         err = dump_filesystems(zhp, &sdd);
1429         fsavl_destroy(fsavl);
1430         nvlist_free(fss);
1431
1432         if (flags.dedup) {
1433                 (void) close(pipefd[0]);
1434                 (void) pthread_join(tid, NULL);
1435         }
1436
1437         if (sdd.cleanup_fd != -1) {
1438                 VERIFY(0 == close(sdd.cleanup_fd));
1439                 sdd.cleanup_fd = -1;
1440         }
1441
1442         if (flags.replicate || flags.doall || flags.props) {
1443                 /*
1444                  * write final end record.  NB: want to do this even if
1445                  * there was some error, because it might not be totally
1446                  * failed.
1447                  */
1448                 dmu_replay_record_t drr = { 0 };
1449                 drr.drr_type = DRR_END;
1450                 if (write(outfd, &drr, sizeof (drr)) == -1) {
1451                         return (zfs_standard_error(zhp->zfs_hdl,
1452                             errno, errbuf));
1453                 }
1454         }
1455
1456         return (err || sdd.err);
1457
1458 stderr_out:
1459         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1460 err_out:
1461         if (sdd.cleanup_fd != -1)
1462                 VERIFY(0 == close(sdd.cleanup_fd));
1463         if (flags.dedup) {
1464                 (void) pthread_cancel(tid);
1465                 (void) pthread_join(tid, NULL);
1466                 (void) close(pipefd[0]);
1467         }
1468         return (err);
1469 }
1470
1471 /*
1472  * Routines specific to "zfs recv"
1473  */
1474
1475 static int
1476 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1477     boolean_t byteswap, zio_cksum_t *zc)
1478 {
1479         char *cp = buf;
1480         int rv;
1481         int len = ilen;
1482
1483         do {
1484                 rv = read(fd, cp, len);
1485                 cp += rv;
1486                 len -= rv;
1487         } while (rv > 0);
1488
1489         if (rv < 0 || len != 0) {
1490                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1491                     "failed to read from stream"));
1492                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1493                     "cannot receive")));
1494         }
1495
1496         if (zc) {
1497                 if (byteswap)
1498                         fletcher_4_incremental_byteswap(buf, ilen, zc);
1499                 else
1500                         fletcher_4_incremental_native(buf, ilen, zc);
1501         }
1502         return (0);
1503 }
1504
1505 static int
1506 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1507     boolean_t byteswap, zio_cksum_t *zc)
1508 {
1509         char *buf;
1510         int err;
1511
1512         buf = zfs_alloc(hdl, len);
1513         if (buf == NULL)
1514                 return (ENOMEM);
1515
1516         err = recv_read(hdl, fd, buf, len, byteswap, zc);
1517         if (err != 0) {
1518                 free(buf);
1519                 return (err);
1520         }
1521
1522         err = nvlist_unpack(buf, len, nvp, 0);
1523         free(buf);
1524         if (err != 0) {
1525                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1526                     "stream (malformed nvlist)"));
1527                 return (EINVAL);
1528         }
1529         return (0);
1530 }
1531
1532 static int
1533 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1534     int baselen, char *newname, recvflags_t flags)
1535 {
1536         static int seq;
1537         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1538         int err;
1539         prop_changelist_t *clp;
1540         zfs_handle_t *zhp;
1541
1542         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1543         if (zhp == NULL)
1544                 return (-1);
1545         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1546             flags.force ? MS_FORCE : 0);
1547         zfs_close(zhp);
1548         if (clp == NULL)
1549                 return (-1);
1550         err = changelist_prefix(clp);
1551         if (err)
1552                 return (err);
1553
1554         zc.zc_objset_type = DMU_OST_ZFS;
1555         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1556
1557         if (tryname) {
1558                 (void) strcpy(newname, tryname);
1559
1560                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1561
1562                 if (flags.verbose) {
1563                         (void) printf("attempting rename %s to %s\n",
1564                             zc.zc_name, zc.zc_value);
1565                 }
1566                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1567                 if (err == 0)
1568                         changelist_rename(clp, name, tryname);
1569         } else {
1570                 err = ENOENT;
1571         }
1572
1573         if (err != 0 && strncmp(name+baselen, "recv-", 5) != 0) {
1574                 seq++;
1575
1576                 (void) strncpy(newname, name, baselen);
1577                 (void) snprintf(newname+baselen, ZFS_MAXNAMELEN-baselen,
1578                     "recv-%ld-%u", (long) getpid(), seq);
1579                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1580
1581                 if (flags.verbose) {
1582                         (void) printf("failed - trying rename %s to %s\n",
1583                             zc.zc_name, zc.zc_value);
1584                 }
1585                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1586                 if (err == 0)
1587                         changelist_rename(clp, name, newname);
1588                 if (err && flags.verbose) {
1589                         (void) printf("failed (%u) - "
1590                             "will try again on next pass\n", errno);
1591                 }
1592                 err = EAGAIN;
1593         } else if (flags.verbose) {
1594                 if (err == 0)
1595                         (void) printf("success\n");
1596                 else
1597                         (void) printf("failed (%u)\n", errno);
1598         }
1599
1600         (void) changelist_postfix(clp);
1601         changelist_free(clp);
1602
1603         return (err);
1604 }
1605
1606 static int
1607 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1608     char *newname, recvflags_t flags)
1609 {
1610         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1611         int err = 0;
1612         prop_changelist_t *clp;
1613         zfs_handle_t *zhp;
1614         boolean_t defer = B_FALSE;
1615         int spa_version;
1616
1617         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1618         if (zhp == NULL)
1619                 return (-1);
1620         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1621             flags.force ? MS_FORCE : 0);
1622         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1623             zfs_spa_version(zhp, &spa_version) == 0 &&
1624             spa_version >= SPA_VERSION_USERREFS)
1625                 defer = B_TRUE;
1626         zfs_close(zhp);
1627         if (clp == NULL)
1628                 return (-1);
1629         err = changelist_prefix(clp);
1630         if (err)
1631                 return (err);
1632
1633         zc.zc_objset_type = DMU_OST_ZFS;
1634         zc.zc_defer_destroy = defer;
1635         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1636
1637         if (flags.verbose)
1638                 (void) printf("attempting destroy %s\n", zc.zc_name);
1639         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1640         if (err == 0) {
1641                 if (flags.verbose)
1642                         (void) printf("success\n");
1643                 changelist_remove(clp, zc.zc_name);
1644         }
1645
1646         (void) changelist_postfix(clp);
1647         changelist_free(clp);
1648
1649         /*
1650          * Deferred destroy might destroy the snapshot or only mark it to be
1651          * destroyed later, and it returns success in either case.
1652          */
1653         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1654             ZFS_TYPE_SNAPSHOT))) {
1655                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1656         }
1657
1658         return (err);
1659 }
1660
1661 typedef struct guid_to_name_data {
1662         uint64_t guid;
1663         char *name;
1664 } guid_to_name_data_t;
1665
1666 static int
1667 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1668 {
1669         guid_to_name_data_t *gtnd = arg;
1670         int err;
1671
1672         if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1673                 (void) strcpy(gtnd->name, zhp->zfs_name);
1674                 zfs_close(zhp);
1675                 return (EEXIST);
1676         }
1677         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1678         zfs_close(zhp);
1679         return (err);
1680 }
1681
1682 static int
1683 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1684     char *name)
1685 {
1686         /* exhaustive search all local snapshots */
1687         guid_to_name_data_t gtnd;
1688         int err = 0;
1689         zfs_handle_t *zhp;
1690         char *cp;
1691
1692         gtnd.guid = guid;
1693         gtnd.name = name;
1694
1695         if (strchr(parent, '@') == NULL) {
1696                 zhp = make_dataset_handle(hdl, parent);
1697                 if (zhp != NULL) {
1698                         err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1699                         zfs_close(zhp);
1700                         if (err == EEXIST)
1701                                 return (0);
1702                 }
1703         }
1704
1705         cp = strchr(parent, '/');
1706         if (cp)
1707                 *cp = '\0';
1708         zhp = make_dataset_handle(hdl, parent);
1709         if (cp)
1710                 *cp = '/';
1711
1712         if (zhp) {
1713                 err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1714                 zfs_close(zhp);
1715         }
1716
1717         return (err == EEXIST ? 0 : ENOENT);
1718
1719 }
1720
1721 /*
1722  * Return true if dataset guid1 is created before guid2.
1723  */
1724 static int
1725 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1726     uint64_t guid1, uint64_t guid2)
1727 {
1728         nvlist_t *nvfs;
1729         char *fsname, *snapname;
1730         char buf[ZFS_MAXNAMELEN];
1731         int rv;
1732         zfs_node_t zn1, zn2;
1733
1734         if (guid2 == 0)
1735                 return (0);
1736         if (guid1 == 0)
1737                 return (1);
1738
1739         nvfs = fsavl_find(avl, guid1, &snapname);
1740         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1741         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1742         zn1.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1743         if (zn1.zn_handle == NULL)
1744                 return (-1);
1745
1746         nvfs = fsavl_find(avl, guid2, &snapname);
1747         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1748         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1749         zn2.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1750         if (zn2.zn_handle == NULL) {
1751                 zfs_close(zn2.zn_handle);
1752                 return (-1);
1753         }
1754
1755         rv = (zfs_snapshot_compare(&zn1, &zn2) == -1);
1756
1757         zfs_close(zn1.zn_handle);
1758         zfs_close(zn2.zn_handle);
1759
1760         return (rv);
1761 }
1762
1763 static int
1764 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
1765     recvflags_t flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
1766     nvlist_t *renamed)
1767 {
1768         nvlist_t *local_nv;
1769         avl_tree_t *local_avl;
1770         nvpair_t *fselem, *nextfselem;
1771         char *fromsnap;
1772         char newname[ZFS_MAXNAMELEN];
1773         int error;
1774         boolean_t needagain, progress, recursive;
1775         char *s1, *s2;
1776
1777         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
1778
1779         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
1780             ENOENT);
1781
1782         if (flags.dryrun)
1783                 return (0);
1784
1785 again:
1786         needagain = progress = B_FALSE;
1787
1788         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
1789             recursive, &local_nv, &local_avl)) != 0)
1790                 return (error);
1791
1792         /*
1793          * Process deletes and renames
1794          */
1795         for (fselem = nvlist_next_nvpair(local_nv, NULL);
1796             fselem; fselem = nextfselem) {
1797                 nvlist_t *nvfs, *snaps;
1798                 nvlist_t *stream_nvfs = NULL;
1799                 nvpair_t *snapelem, *nextsnapelem;
1800                 uint64_t fromguid = 0;
1801                 uint64_t originguid = 0;
1802                 uint64_t stream_originguid = 0;
1803                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
1804                 char *fsname, *stream_fsname;
1805
1806                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
1807
1808                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
1809                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
1810                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1811                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
1812                     &parent_fromsnap_guid));
1813                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
1814
1815                 /*
1816                  * First find the stream's fs, so we can check for
1817                  * a different origin (due to "zfs promote")
1818                  */
1819                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1820                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
1821                         uint64_t thisguid;
1822
1823                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1824                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
1825
1826                         if (stream_nvfs != NULL)
1827                                 break;
1828                 }
1829
1830                 /* check for promote */
1831                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
1832                     &stream_originguid);
1833                 if (stream_nvfs && originguid != stream_originguid) {
1834                         switch (created_before(hdl, local_avl,
1835                             stream_originguid, originguid)) {
1836                         case 1: {
1837                                 /* promote it! */
1838                                 zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1839                                 nvlist_t *origin_nvfs;
1840                                 char *origin_fsname;
1841
1842                                 if (flags.verbose)
1843                                         (void) printf("promoting %s\n", fsname);
1844
1845                                 origin_nvfs = fsavl_find(local_avl, originguid,
1846                                     NULL);
1847                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
1848                                     "name", &origin_fsname));
1849                                 (void) strlcpy(zc.zc_value, origin_fsname,
1850                                     sizeof (zc.zc_value));
1851                                 (void) strlcpy(zc.zc_name, fsname,
1852                                     sizeof (zc.zc_name));
1853                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
1854                                 if (error == 0)
1855                                         progress = B_TRUE;
1856                                 break;
1857                         }
1858                         default:
1859                                 break;
1860                         case -1:
1861                                 fsavl_destroy(local_avl);
1862                                 nvlist_free(local_nv);
1863                                 return (-1);
1864                         }
1865                         /*
1866                          * We had/have the wrong origin, therefore our
1867                          * list of snapshots is wrong.  Need to handle
1868                          * them on the next pass.
1869                          */
1870                         needagain = B_TRUE;
1871                         continue;
1872                 }
1873
1874                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1875                     snapelem; snapelem = nextsnapelem) {
1876                         uint64_t thisguid;
1877                         char *stream_snapname;
1878                         nvlist_t *found, *props;
1879
1880                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
1881
1882                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1883                         found = fsavl_find(stream_avl, thisguid,
1884                             &stream_snapname);
1885
1886                         /* check for delete */
1887                         if (found == NULL) {
1888                                 char name[ZFS_MAXNAMELEN];
1889
1890                                 if (!flags.force)
1891                                         continue;
1892
1893                                 (void) snprintf(name, sizeof (name), "%s@%s",
1894                                     fsname, nvpair_name(snapelem));
1895
1896                                 error = recv_destroy(hdl, name,
1897                                     strlen(fsname)+1, newname, flags);
1898                                 if (error)
1899                                         needagain = B_TRUE;
1900                                 else
1901                                         progress = B_TRUE;
1902                                 continue;
1903                         }
1904
1905                         stream_nvfs = found;
1906
1907                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
1908                             &props) && 0 == nvlist_lookup_nvlist(props,
1909                             stream_snapname, &props)) {
1910                                 zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1911
1912                                 zc.zc_cookie = B_TRUE; /* received */
1913                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
1914                                     "%s@%s", fsname, nvpair_name(snapelem));
1915                                 if (zcmd_write_src_nvlist(hdl, &zc,
1916                                     props) == 0) {
1917                                         (void) zfs_ioctl(hdl,
1918                                             ZFS_IOC_SET_PROP, &zc);
1919                                         zcmd_free_nvlists(&zc);
1920                                 }
1921                         }
1922
1923                         /* check for different snapname */
1924                         if (strcmp(nvpair_name(snapelem),
1925                             stream_snapname) != 0) {
1926                                 char name[ZFS_MAXNAMELEN];
1927                                 char tryname[ZFS_MAXNAMELEN];
1928
1929                                 (void) snprintf(name, sizeof (name), "%s@%s",
1930                                     fsname, nvpair_name(snapelem));
1931                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
1932                                     fsname, stream_snapname);
1933
1934                                 error = recv_rename(hdl, name, tryname,
1935                                     strlen(fsname)+1, newname, flags);
1936                                 if (error)
1937                                         needagain = B_TRUE;
1938                                 else
1939                                         progress = B_TRUE;
1940                         }
1941
1942                         if (strcmp(stream_snapname, fromsnap) == 0)
1943                                 fromguid = thisguid;
1944                 }
1945
1946                 /* check for delete */
1947                 if (stream_nvfs == NULL) {
1948                         if (!flags.force)
1949                                 continue;
1950
1951                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
1952                             newname, flags);
1953                         if (error)
1954                                 needagain = B_TRUE;
1955                         else
1956                                 progress = B_TRUE;
1957                         continue;
1958                 }
1959
1960                 if (fromguid == 0) {
1961                         if (flags.verbose) {
1962                                 (void) printf("local fs %s does not have "
1963                                     "fromsnap (%s in stream); must have "
1964                                     "been deleted locally; ignoring\n",
1965                                     fsname, fromsnap);
1966                         }
1967                         continue;
1968                 }
1969
1970                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
1971                     "name", &stream_fsname));
1972                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
1973                     "parentfromsnap", &stream_parent_fromsnap_guid));
1974
1975                 s1 = strrchr(fsname, '/');
1976                 s2 = strrchr(stream_fsname, '/');
1977
1978                 /*
1979                  * Check for rename. If the exact receive path is specified, it
1980                  * does not count as a rename, but we still need to check the
1981                  * datasets beneath it.
1982                  */
1983                 if ((stream_parent_fromsnap_guid != 0 &&
1984                     parent_fromsnap_guid != 0 &&
1985                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
1986                     ((flags.isprefix || strcmp(tofs, fsname) != 0) &&
1987                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
1988                         nvlist_t *parent;
1989                         char tryname[ZFS_MAXNAMELEN];
1990
1991                         parent = fsavl_find(local_avl,
1992                             stream_parent_fromsnap_guid, NULL);
1993                         /*
1994                          * NB: parent might not be found if we used the
1995                          * tosnap for stream_parent_fromsnap_guid,
1996                          * because the parent is a newly-created fs;
1997                          * we'll be able to rename it after we recv the
1998                          * new fs.
1999                          */
2000                         if (parent != NULL) {
2001                                 char *pname;
2002
2003                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
2004                                     &pname));
2005                                 (void) snprintf(tryname, sizeof (tryname),
2006                                     "%s%s", pname, strrchr(stream_fsname, '/'));
2007                         } else {
2008                                 tryname[0] = '\0';
2009                                 if (flags.verbose) {
2010                                         (void) printf("local fs %s new parent "
2011                                             "not found\n", fsname);
2012                                 }
2013                         }
2014
2015                         newname[0] = '\0';
2016
2017                         error = recv_rename(hdl, fsname, tryname,
2018                             strlen(tofs)+1, newname, flags);
2019
2020                         if (renamed != NULL && newname[0] != '\0') {
2021                                 VERIFY(0 == nvlist_add_boolean(renamed,
2022                                     newname));
2023                         }
2024
2025                         if (error)
2026                                 needagain = B_TRUE;
2027                         else
2028                                 progress = B_TRUE;
2029                 }
2030         }
2031
2032         fsavl_destroy(local_avl);
2033         nvlist_free(local_nv);
2034
2035         if (needagain && progress) {
2036                 /* do another pass to fix up temporary names */
2037                 if (flags.verbose)
2038                         (void) printf("another pass:\n");
2039                 goto again;
2040         }
2041
2042         return (needagain);
2043 }
2044
2045 static int
2046 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2047     recvflags_t flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2048     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2049 {
2050         nvlist_t *stream_nv = NULL;
2051         avl_tree_t *stream_avl = NULL;
2052         char *fromsnap = NULL;
2053         char *cp;
2054         char tofs[ZFS_MAXNAMELEN];
2055         char sendfs[ZFS_MAXNAMELEN];
2056         char errbuf[1024];
2057         dmu_replay_record_t drre;
2058         int error;
2059         boolean_t anyerr = B_FALSE;
2060         boolean_t softerr = B_FALSE;
2061         boolean_t recursive;
2062
2063         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2064             "cannot receive"));
2065
2066         assert(drr->drr_type == DRR_BEGIN);
2067         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2068         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2069             DMU_COMPOUNDSTREAM);
2070
2071         /*
2072          * Read in the nvlist from the stream.
2073          */
2074         if (drr->drr_payloadlen != 0) {
2075                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2076                     &stream_nv, flags.byteswap, zc);
2077                 if (error) {
2078                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2079                         goto out;
2080                 }
2081         }
2082
2083         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2084             ENOENT);
2085
2086         if (recursive && strchr(destname, '@')) {
2087                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2088                     "cannot specify snapshot name for multi-snapshot stream"));
2089                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2090                 goto out;
2091         }
2092
2093         /*
2094          * Read in the end record and verify checksum.
2095          */
2096         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2097             flags.byteswap, NULL)))
2098                 goto out;
2099         if (flags.byteswap) {
2100                 drre.drr_type = BSWAP_32(drre.drr_type);
2101                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2102                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2103                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2104                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2105                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2106                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2107                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2108                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2109         }
2110         if (drre.drr_type != DRR_END) {
2111                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2112                 goto out;
2113         }
2114         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2115                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2116                     "incorrect header checksum"));
2117                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2118                 goto out;
2119         }
2120
2121         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2122
2123         if (drr->drr_payloadlen != 0) {
2124                 nvlist_t *stream_fss;
2125
2126                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2127                     &stream_fss));
2128                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2129                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2130                             "couldn't allocate avl tree"));
2131                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2132                         goto out;
2133                 }
2134
2135                 if (fromsnap != NULL) {
2136                         nvlist_t *renamed = NULL;
2137                         nvpair_t *pair = NULL;
2138
2139                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2140                         if (flags.isprefix) {
2141                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2142                                 int i;
2143
2144                                 if (flags.istail) {
2145                                         cp = strrchr(drrb->drr_toname, '/');
2146                                         if (cp == NULL) {
2147                                                 (void) strlcat(tofs, "/",
2148                                                     ZFS_MAXNAMELEN);
2149                                                 i = 0;
2150                                         } else {
2151                                                 i = (cp - drrb->drr_toname);
2152                                         }
2153                                 } else {
2154                                         i = strcspn(drrb->drr_toname, "/@");
2155                                 }
2156                                 /* zfs_receive_one() will create_parents() */
2157                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2158                                     ZFS_MAXNAMELEN);
2159                                 *strchr(tofs, '@') = '\0';
2160                         }
2161
2162                         if (recursive && !flags.dryrun && !flags.nomount) {
2163                                 VERIFY(0 == nvlist_alloc(&renamed,
2164                                     NV_UNIQUE_NAME, 0));
2165                         }
2166
2167                         softerr = recv_incremental_replication(hdl, tofs, flags,
2168                             stream_nv, stream_avl, renamed);
2169
2170                         /* Unmount renamed filesystems before receiving. */
2171                         while ((pair = nvlist_next_nvpair(renamed,
2172                             pair)) != NULL) {
2173                                 zfs_handle_t *zhp;
2174                                 prop_changelist_t *clp = NULL;
2175
2176                                 zhp = zfs_open(hdl, nvpair_name(pair),
2177                                     ZFS_TYPE_FILESYSTEM);
2178                                 if (zhp != NULL) {
2179                                         clp = changelist_gather(zhp,
2180                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2181                                         zfs_close(zhp);
2182                                         if (clp != NULL) {
2183                                                 softerr |=
2184                                                     changelist_prefix(clp);
2185                                                 changelist_free(clp);
2186                                         }
2187                                 }
2188                         }
2189
2190                         nvlist_free(renamed);
2191                 }
2192         }
2193
2194         /*
2195          * Get the fs specified by the first path in the stream (the top level
2196          * specified by 'zfs send') and pass it to each invocation of
2197          * zfs_receive_one().
2198          */
2199         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2200             ZFS_MAXNAMELEN);
2201         if ((cp = strchr(sendfs, '@')) != NULL)
2202                 *cp = '\0';
2203
2204         /* Finally, receive each contained stream */
2205         do {
2206                 /*
2207                  * we should figure out if it has a recoverable
2208                  * error, in which case do a recv_skip() and drive on.
2209                  * Note, if we fail due to already having this guid,
2210                  * zfs_receive_one() will take care of it (ie,
2211                  * recv_skip() and return 0).
2212                  */
2213                 error = zfs_receive_impl(hdl, destname, flags, fd,
2214                     sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2215                     action_handlep);
2216                 if (error == ENODATA) {
2217                         error = 0;
2218                         break;
2219                 }
2220                 anyerr |= error;
2221         } while (error == 0);
2222
2223         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2224                 /*
2225                  * Now that we have the fs's they sent us, try the
2226                  * renames again.
2227                  */
2228                 softerr = recv_incremental_replication(hdl, tofs, flags,
2229                     stream_nv, stream_avl, NULL);
2230         }
2231
2232 out:
2233         fsavl_destroy(stream_avl);
2234         if (stream_nv)
2235                 nvlist_free(stream_nv);
2236         if (softerr)
2237                 error = -2;
2238         if (anyerr)
2239                 error = -1;
2240         return (error);
2241 }
2242
2243 static void
2244 trunc_prop_errs(int truncated)
2245 {
2246         ASSERT(truncated != 0);
2247
2248         if (truncated == 1)
2249                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2250                     "1 more property could not be set\n"));
2251         else
2252                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2253                     "%d more properties could not be set\n"), truncated);
2254 }
2255
2256 static int
2257 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2258 {
2259         dmu_replay_record_t *drr;
2260         void *buf = malloc(1<<20);
2261         char errbuf[1024];
2262
2263         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2264             "cannot receive:"));
2265
2266         /* XXX would be great to use lseek if possible... */
2267         drr = buf;
2268
2269         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2270             byteswap, NULL) == 0) {
2271                 if (byteswap)
2272                         drr->drr_type = BSWAP_32(drr->drr_type);
2273
2274                 switch (drr->drr_type) {
2275                 case DRR_BEGIN:
2276                         /* NB: not to be used on v2 stream packages */
2277                         if (drr->drr_payloadlen != 0) {
2278                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2279                                     "invalid substream header"));
2280                                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2281                         }
2282                         break;
2283
2284                 case DRR_END:
2285                         free(buf);
2286                         return (0);
2287
2288                 case DRR_OBJECT:
2289                         if (byteswap) {
2290                                 drr->drr_u.drr_object.drr_bonuslen =
2291                                     BSWAP_32(drr->drr_u.drr_object.
2292                                     drr_bonuslen);
2293                         }
2294                         (void) recv_read(hdl, fd, buf,
2295                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2296                             B_FALSE, NULL);
2297                         break;
2298
2299                 case DRR_WRITE:
2300                         if (byteswap) {
2301                                 drr->drr_u.drr_write.drr_length =
2302                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
2303                         }
2304                         (void) recv_read(hdl, fd, buf,
2305                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2306                         break;
2307                 case DRR_SPILL:
2308                         if (byteswap) {
2309                                 drr->drr_u.drr_write.drr_length =
2310                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
2311                         }
2312                         (void) recv_read(hdl, fd, buf,
2313                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2314                         break;
2315                 case DRR_WRITE_BYREF:
2316                 case DRR_FREEOBJECTS:
2317                 case DRR_FREE:
2318                         break;
2319
2320                 default:
2321                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2322                             "invalid record type"));
2323                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2324                 }
2325         }
2326
2327         free(buf);
2328         return (-1);
2329 }
2330
2331 /*
2332  * Restores a backup of tosnap from the file descriptor specified by infd.
2333  */
2334 static int
2335 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2336     recvflags_t flags, dmu_replay_record_t *drr,
2337     dmu_replay_record_t *drr_noswap, const char *sendfs,
2338     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2339     uint64_t *action_handlep)
2340 {
2341         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2342         time_t begin_time;
2343         int ioctl_err, ioctl_errno, err;
2344         char *cp;
2345         struct drr_begin *drrb = &drr->drr_u.drr_begin;
2346         char errbuf[1024];
2347         char prop_errbuf[1024];
2348         const char *chopprefix;
2349         boolean_t newfs = B_FALSE;
2350         boolean_t stream_wantsnewfs;
2351         uint64_t parent_snapguid = 0;
2352         prop_changelist_t *clp = NULL;
2353         nvlist_t *snapprops_nvlist = NULL;
2354         zprop_errflags_t prop_errflags;
2355         boolean_t recursive;
2356
2357         begin_time = time(NULL);
2358
2359         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2360             "cannot receive"));
2361
2362         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2363             ENOENT);
2364
2365         if (stream_avl != NULL) {
2366                 char *snapname;
2367                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2368                     &snapname);
2369                 nvlist_t *props;
2370                 int ret;
2371
2372                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2373                     &parent_snapguid);
2374                 err = nvlist_lookup_nvlist(fs, "props", &props);
2375                 if (err)
2376                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2377
2378                 if (flags.canmountoff) {
2379                         VERIFY(0 == nvlist_add_uint64(props,
2380                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2381                 }
2382                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2383                 if (err)
2384                         nvlist_free(props);
2385
2386                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2387                         VERIFY(0 == nvlist_lookup_nvlist(props,
2388                             snapname, &snapprops_nvlist));
2389                 }
2390
2391                 if (ret != 0)
2392                         return (-1);
2393         }
2394
2395         cp = NULL;
2396
2397         /*
2398          * Determine how much of the snapshot name stored in the stream
2399          * we are going to tack on to the name they specified on the
2400          * command line, and how much we are going to chop off.
2401          *
2402          * If they specified a snapshot, chop the entire name stored in
2403          * the stream.
2404          */
2405         if (flags.istail) {
2406                 /*
2407                  * A filesystem was specified with -e. We want to tack on only
2408                  * the tail of the sent snapshot path.
2409                  */
2410                 if (strchr(tosnap, '@')) {
2411                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2412                             "argument - snapshot not allowed with -e"));
2413                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2414                 }
2415
2416                 chopprefix = strrchr(sendfs, '/');
2417
2418                 if (chopprefix == NULL) {
2419                         /*
2420                          * The tail is the poolname, so we need to
2421                          * prepend a path separator.
2422                          */
2423                         int len = strlen(drrb->drr_toname);
2424                         cp = malloc(len + 2);
2425                         cp[0] = '/';
2426                         (void) strcpy(&cp[1], drrb->drr_toname);
2427                         chopprefix = cp;
2428                 } else {
2429                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2430                 }
2431         } else if (flags.isprefix) {
2432                 /*
2433                  * A filesystem was specified with -d. We want to tack on
2434                  * everything but the first element of the sent snapshot path
2435                  * (all but the pool name).
2436                  */
2437                 if (strchr(tosnap, '@')) {
2438                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2439                             "argument - snapshot not allowed with -d"));
2440                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2441                 }
2442
2443                 chopprefix = strchr(drrb->drr_toname, '/');
2444                 if (chopprefix == NULL)
2445                         chopprefix = strchr(drrb->drr_toname, '@');
2446         } else if (strchr(tosnap, '@') == NULL) {
2447                 /*
2448                  * If a filesystem was specified without -d or -e, we want to
2449                  * tack on everything after the fs specified by 'zfs send'.
2450                  */
2451                 chopprefix = drrb->drr_toname + strlen(sendfs);
2452         } else {
2453                 /* A snapshot was specified as an exact path (no -d or -e). */
2454                 if (recursive) {
2455                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2456                             "cannot specify snapshot name for multi-snapshot "
2457                             "stream"));
2458                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2459                 }
2460                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2461         }
2462
2463         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2464         ASSERT(chopprefix > drrb->drr_toname);
2465         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2466         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2467             chopprefix[0] == '\0');
2468
2469         /*
2470          * Determine name of destination snapshot, store in zc_value.
2471          */
2472         (void) strcpy(zc.zc_top_ds, tosnap);
2473         (void) strcpy(zc.zc_value, tosnap);
2474         (void) strlcat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2475         free(cp);
2476         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2477                 zcmd_free_nvlists(&zc);
2478                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2479         }
2480
2481         /*
2482          * Determine the name of the origin snapshot, store in zc_string.
2483          */
2484         if (drrb->drr_flags & DRR_FLAG_CLONE) {
2485                 if (guid_to_name(hdl, tosnap,
2486                     drrb->drr_fromguid, zc.zc_string) != 0) {
2487                         zcmd_free_nvlists(&zc);
2488                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2489                             "local origin for clone %s does not exist"),
2490                             zc.zc_value);
2491                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2492                 }
2493                 if (flags.verbose)
2494                         (void) printf("found clone origin %s\n", zc.zc_string);
2495         }
2496
2497         stream_wantsnewfs = (drrb->drr_fromguid == 0 ||
2498             (drrb->drr_flags & DRR_FLAG_CLONE));
2499
2500         if (stream_wantsnewfs) {
2501                 /*
2502                  * if the parent fs does not exist, look for it based on
2503                  * the parent snap GUID
2504                  */
2505                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2506                     "cannot receive new filesystem stream"));
2507
2508                 (void) strcpy(zc.zc_name, zc.zc_value);
2509                 cp = strrchr(zc.zc_name, '/');
2510                 if (cp)
2511                         *cp = '\0';
2512                 if (cp &&
2513                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2514                         char suffix[ZFS_MAXNAMELEN];
2515                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2516                         if (guid_to_name(hdl, tosnap, parent_snapguid,
2517                             zc.zc_value) == 0) {
2518                                 *strchr(zc.zc_value, '@') = '\0';
2519                                 (void) strcat(zc.zc_value, suffix);
2520                         }
2521                 }
2522         } else {
2523                 /*
2524                  * if the fs does not exist, look for it based on the
2525                  * fromsnap GUID
2526                  */
2527                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2528                     "cannot receive incremental stream"));
2529
2530                 (void) strcpy(zc.zc_name, zc.zc_value);
2531                 *strchr(zc.zc_name, '@') = '\0';
2532
2533                 /*
2534                  * If the exact receive path was specified and this is the
2535                  * topmost path in the stream, then if the fs does not exist we
2536                  * should look no further.
2537                  */
2538                 if ((flags.isprefix || (*(chopprefix = drrb->drr_toname +
2539                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2540                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2541                         char snap[ZFS_MAXNAMELEN];
2542                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
2543                         if (guid_to_name(hdl, tosnap, drrb->drr_fromguid,
2544                             zc.zc_value) == 0) {
2545                                 *strchr(zc.zc_value, '@') = '\0';
2546                                 (void) strcat(zc.zc_value, snap);
2547                         }
2548                 }
2549         }
2550
2551         (void) strcpy(zc.zc_name, zc.zc_value);
2552         *strchr(zc.zc_name, '@') = '\0';
2553
2554         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2555                 zfs_handle_t *zhp;
2556
2557                 /*
2558                  * Destination fs exists.  Therefore this should either
2559                  * be an incremental, or the stream specifies a new fs
2560                  * (full stream or clone) and they want us to blow it
2561                  * away (and have therefore specified -F and removed any
2562                  * snapshots).
2563                  */
2564                 if (stream_wantsnewfs) {
2565                         if (!flags.force) {
2566                                 zcmd_free_nvlists(&zc);
2567                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2568                                     "destination '%s' exists\n"
2569                                     "must specify -F to overwrite it"),
2570                                     zc.zc_name);
2571                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2572                         }
2573                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2574                             &zc) == 0) {
2575                                 zcmd_free_nvlists(&zc);
2576                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2577                                     "destination has snapshots (eg. %s)\n"
2578                                     "must destroy them to overwrite it"),
2579                                     zc.zc_name);
2580                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2581                         }
2582                 }
2583
2584                 if ((zhp = zfs_open(hdl, zc.zc_name,
2585                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2586                         zcmd_free_nvlists(&zc);
2587                         return (-1);
2588                 }
2589
2590                 if (stream_wantsnewfs &&
2591                     zhp->zfs_dmustats.dds_origin[0]) {
2592                         zcmd_free_nvlists(&zc);
2593                         zfs_close(zhp);
2594                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2595                             "destination '%s' is a clone\n"
2596                             "must destroy it to overwrite it"),
2597                             zc.zc_name);
2598                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2599                 }
2600
2601                 if (!flags.dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2602                     stream_wantsnewfs) {
2603                         /* We can't do online recv in this case */
2604                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2605                         if (clp == NULL) {
2606                                 zfs_close(zhp);
2607                                 zcmd_free_nvlists(&zc);
2608                                 return (-1);
2609                         }
2610                         if (changelist_prefix(clp) != 0) {
2611                                 changelist_free(clp);
2612                                 zfs_close(zhp);
2613                                 zcmd_free_nvlists(&zc);
2614                                 return (-1);
2615                         }
2616                 }
2617                 if (!flags.dryrun && zhp->zfs_type == ZFS_TYPE_VOLUME &&
2618                     zvol_remove_link(hdl, zhp->zfs_name) != 0) {
2619                         zfs_close(zhp);
2620                         zcmd_free_nvlists(&zc);
2621                         return (-1);
2622                 }
2623                 zfs_close(zhp);
2624         } else {
2625                 /*
2626                  * Destination filesystem does not exist.  Therefore we better
2627                  * be creating a new filesystem (either from a full backup, or
2628                  * a clone).  It would therefore be invalid if the user
2629                  * specified only the pool name (i.e. if the destination name
2630                  * contained no slash character).
2631                  */
2632                 if (!stream_wantsnewfs ||
2633                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
2634                         zcmd_free_nvlists(&zc);
2635                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2636                             "destination '%s' does not exist"), zc.zc_name);
2637                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2638                 }
2639
2640                 /*
2641                  * Trim off the final dataset component so we perform the
2642                  * recvbackup ioctl to the filesystems's parent.
2643                  */
2644                 *cp = '\0';
2645
2646                 if (flags.isprefix && !flags.istail && !flags.dryrun &&
2647                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2648                         zcmd_free_nvlists(&zc);
2649                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2650                 }
2651
2652                 newfs = B_TRUE;
2653         }
2654
2655         zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2656         zc.zc_cookie = infd;
2657         zc.zc_guid = flags.force;
2658         if (flags.verbose) {
2659                 (void) printf("%s %s stream of %s into %s\n",
2660                     flags.dryrun ? "would receive" : "receiving",
2661                     drrb->drr_fromguid ? "incremental" : "full",
2662                     drrb->drr_toname, zc.zc_value);
2663                 (void) fflush(stdout);
2664         }
2665
2666         if (flags.dryrun) {
2667                 zcmd_free_nvlists(&zc);
2668                 return (recv_skip(hdl, infd, flags.byteswap));
2669         }
2670
2671         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2672         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2673         zc.zc_cleanup_fd = cleanup_fd;
2674         zc.zc_action_handle = *action_handlep;
2675
2676         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2677         ioctl_errno = errno;
2678         prop_errflags = (zprop_errflags_t)zc.zc_obj;
2679
2680         if (err == 0) {
2681                 nvlist_t *prop_errors;
2682                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2683                     zc.zc_nvlist_dst_size, &prop_errors, 0));
2684
2685                 nvpair_t *prop_err = NULL;
2686
2687                 while ((prop_err = nvlist_next_nvpair(prop_errors,
2688                     prop_err)) != NULL) {
2689                         char tbuf[1024];
2690                         zfs_prop_t prop;
2691                         int intval;
2692
2693                         prop = zfs_name_to_prop(nvpair_name(prop_err));
2694                         (void) nvpair_value_int32(prop_err, &intval);
2695                         if (strcmp(nvpair_name(prop_err),
2696                             ZPROP_N_MORE_ERRORS) == 0) {
2697                                 trunc_prop_errs(intval);
2698                                 break;
2699                         } else {
2700                                 (void) snprintf(tbuf, sizeof (tbuf),
2701                                     dgettext(TEXT_DOMAIN,
2702                                     "cannot receive %s property on %s"),
2703                                     nvpair_name(prop_err), zc.zc_name);
2704                                 zfs_setprop_error(hdl, prop, intval, tbuf);
2705                         }
2706                 }
2707                 nvlist_free(prop_errors);
2708         }
2709
2710         zc.zc_nvlist_dst = 0;
2711         zc.zc_nvlist_dst_size = 0;
2712         zcmd_free_nvlists(&zc);
2713
2714         if (err == 0 && snapprops_nvlist) {
2715                 zfs_cmd_t zc2 = { "\0", "\0", "\0", "\0", 0 };
2716
2717                 (void) strcpy(zc2.zc_name, zc.zc_value);
2718                 zc2.zc_cookie = B_TRUE; /* received */
2719                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
2720                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
2721                         zcmd_free_nvlists(&zc2);
2722                 }
2723         }
2724
2725         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
2726                 /*
2727                  * It may be that this snapshot already exists,
2728                  * in which case we want to consume & ignore it
2729                  * rather than failing.
2730                  */
2731                 avl_tree_t *local_avl;
2732                 nvlist_t *local_nv, *fs;
2733                 cp = strchr(zc.zc_value, '@');
2734
2735                 /*
2736                  * XXX Do this faster by just iterating over snaps in
2737                  * this fs.  Also if zc_value does not exist, we will
2738                  * get a strange "does not exist" error message.
2739                  */
2740                 *cp = '\0';
2741                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
2742                     &local_nv, &local_avl) == 0) {
2743                         *cp = '@';
2744                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
2745                         fsavl_destroy(local_avl);
2746                         nvlist_free(local_nv);
2747
2748                         if (fs != NULL) {
2749                                 if (flags.verbose) {
2750                                         (void) printf("snap %s already exists; "
2751                                             "ignoring\n", zc.zc_value);
2752                                 }
2753                                 err = ioctl_err = recv_skip(hdl, infd,
2754                                     flags.byteswap);
2755                         }
2756                 }
2757                 *cp = '@';
2758         }
2759
2760         if (ioctl_err != 0) {
2761                 switch (ioctl_errno) {
2762                 case ENODEV:
2763                         cp = strchr(zc.zc_value, '@');
2764                         *cp = '\0';
2765                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2766                             "most recent snapshot of %s does not\n"
2767                             "match incremental source"), zc.zc_value);
2768                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2769                         *cp = '@';
2770                         break;
2771                 case ETXTBSY:
2772                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2773                             "destination %s has been modified\n"
2774                             "since most recent snapshot"), zc.zc_name);
2775                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2776                         break;
2777                 case EEXIST:
2778                         cp = strchr(zc.zc_value, '@');
2779                         if (newfs) {
2780                                 /* it's the containing fs that exists */
2781                                 *cp = '\0';
2782                         }
2783                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2784                             "destination already exists"));
2785                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
2786                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
2787                             zc.zc_value);
2788                         *cp = '@';
2789                         break;
2790                 case EINVAL:
2791                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2792                         break;
2793                 case ECKSUM:
2794                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2795                             "invalid stream (checksum mismatch)"));
2796                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2797                         break;
2798                 case ENOTSUP:
2799                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2800                             "pool must be upgraded to receive this stream."));
2801                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
2802                         break;
2803                 case EDQUOT:
2804                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2805                             "destination %s space quota exceeded"), zc.zc_name);
2806                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2807                         break;
2808                 default:
2809                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
2810                 }
2811         }
2812
2813         /*
2814          * Mount the target filesystem (if created).  Also mount any
2815          * children of the target filesystem if we did a replication
2816          * receive (indicated by stream_avl being non-NULL).
2817          */
2818         cp = strchr(zc.zc_value, '@');
2819         if (cp && (ioctl_err == 0 || !newfs)) {
2820                 zfs_handle_t *h;
2821
2822                 *cp = '\0';
2823                 h = zfs_open(hdl, zc.zc_value,
2824                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
2825                 if (h != NULL) {
2826                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
2827                                 *cp = '@';
2828                                 err = zvol_create_link(hdl, h->zfs_name);
2829                                 if (err == 0 && ioctl_err == 0)
2830                                         err = zvol_create_link(hdl,
2831                                             zc.zc_value);
2832                         } else if (newfs || stream_avl) {
2833                                 /*
2834                                  * Track the first/top of hierarchy fs,
2835                                  * for mounting and sharing later.
2836                                  */
2837                                 if (top_zfs && *top_zfs == NULL)
2838                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
2839                         }
2840                         zfs_close(h);
2841                 }
2842                 *cp = '@';
2843         }
2844
2845         if (clp) {
2846                 err |= changelist_postfix(clp);
2847                 changelist_free(clp);
2848         }
2849
2850         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
2851                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
2852                     "failed to clear unreceived properties on %s"),
2853                     zc.zc_name);
2854                 (void) fprintf(stderr, "\n");
2855         }
2856         if (prop_errflags & ZPROP_ERR_NORESTORE) {
2857                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
2858                     "failed to restore original properties on %s"),
2859                     zc.zc_name);
2860                 (void) fprintf(stderr, "\n");
2861         }
2862
2863         if (err || ioctl_err)
2864                 return (-1);
2865
2866         *action_handlep = zc.zc_action_handle;
2867
2868         if (flags.verbose) {
2869                 char buf1[64];
2870                 char buf2[64];
2871                 uint64_t bytes = zc.zc_cookie;
2872                 time_t delta = time(NULL) - begin_time;
2873                 if (delta == 0)
2874                         delta = 1;
2875                 zfs_nicenum(bytes, buf1, sizeof (buf1));
2876                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
2877
2878                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
2879                     buf1, delta, buf2);
2880         }
2881
2882         return (0);
2883 }
2884
2885 static int
2886 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
2887     int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2888     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2889 {
2890         int err;
2891         dmu_replay_record_t drr, drr_noswap;
2892         struct drr_begin *drrb = &drr.drr_u.drr_begin;
2893         char errbuf[1024];
2894         zio_cksum_t zcksum = { { 0 } };
2895         uint64_t featureflags;
2896         int hdrtype;
2897
2898         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2899             "cannot receive"));
2900
2901         if (flags.isprefix &&
2902             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
2903                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
2904                     "(%s) does not exist"), tosnap);
2905                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2906         }
2907
2908         /* read in the BEGIN record */
2909         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
2910             &zcksum)))
2911                 return (err);
2912
2913         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
2914                 /* It's the double end record at the end of a package */
2915                 return (ENODATA);
2916         }
2917
2918         /* the kernel needs the non-byteswapped begin record */
2919         drr_noswap = drr;
2920
2921         flags.byteswap = B_FALSE;
2922         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
2923                 /*
2924                  * We computed the checksum in the wrong byteorder in
2925                  * recv_read() above; do it again correctly.
2926                  */
2927                 bzero(&zcksum, sizeof (zio_cksum_t));
2928                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
2929                 flags.byteswap = B_TRUE;
2930
2931                 drr.drr_type = BSWAP_32(drr.drr_type);
2932                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
2933                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
2934                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
2935                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
2936                 drrb->drr_type = BSWAP_32(drrb->drr_type);
2937                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
2938                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
2939                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
2940         }
2941
2942         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
2943                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2944                     "stream (bad magic number)"));
2945                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2946         }
2947
2948         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
2949         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
2950
2951         if (!DMU_STREAM_SUPPORTED(featureflags) ||
2952             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
2953                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2954                     "stream has unsupported feature, feature flags = %lx"),
2955                     featureflags);
2956                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2957         }
2958
2959         if (strchr(drrb->drr_toname, '@') == NULL) {
2960                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2961                     "stream (bad snapshot name)"));
2962                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2963         }
2964
2965         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
2966                 char nonpackage_sendfs[ZFS_MAXNAMELEN];
2967                 if (sendfs == NULL) {
2968                         /*
2969                          * We were not called from zfs_receive_package(). Get
2970                          * the fs specified by 'zfs send'.
2971                          */
2972                         char *cp;
2973                         (void) strlcpy(nonpackage_sendfs,
2974                             drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
2975                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
2976                                 *cp = '\0';
2977                         sendfs = nonpackage_sendfs;
2978                 }
2979                 return (zfs_receive_one(hdl, infd, tosnap, flags,
2980                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl,
2981                     top_zfs, cleanup_fd, action_handlep));
2982         } else {
2983                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
2984                     DMU_COMPOUNDSTREAM);
2985                 return (zfs_receive_package(hdl, infd, tosnap, flags,
2986                     &drr, &zcksum, top_zfs, cleanup_fd, action_handlep));
2987         }
2988 }
2989
2990 /*
2991  * Restores a backup of tosnap from the file descriptor specified by infd.
2992  * Return 0 on total success, -2 if some things couldn't be
2993  * destroyed/renamed/promoted, -1 if some things couldn't be received.
2994  * (-1 will override -2).
2995  */
2996 int
2997 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
2998     int infd, avl_tree_t *stream_avl)
2999 {
3000         char *top_zfs = NULL;
3001         int err;
3002         int cleanup_fd;
3003         uint64_t action_handle = 0;
3004
3005         cleanup_fd = open(ZFS_DEV, O_RDWR);
3006         VERIFY(cleanup_fd >= 0);
3007
3008         err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL,
3009             stream_avl, &top_zfs, cleanup_fd, &action_handle);
3010
3011         VERIFY(0 == close(cleanup_fd));
3012
3013         if (err == 0 && !flags.nomount && top_zfs) {
3014                 zfs_handle_t *zhp;
3015                 prop_changelist_t *clp;
3016
3017                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3018                 if (zhp != NULL) {
3019                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3020                             CL_GATHER_MOUNT_ALWAYS, 0);
3021                         zfs_close(zhp);
3022                         if (clp != NULL) {
3023                                 /* mount and share received datasets */
3024                                 err = changelist_postfix(clp);
3025                                 changelist_free(clp);
3026                         }
3027                 }
3028                 if (zhp == NULL || clp == NULL || err)
3029                         err = -1;
3030         }
3031         if (top_zfs)
3032                 free(top_zfs);
3033
3034         return (err);
3035 }