cddl/contrib/opensolaris/lib/libzfs/common/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  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2012 Pawel Jakub Dawidek. All rights reserved.
  27  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  28  * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
  29  * Copyright (c) 2014 Integros [integros.com]
  30  */
  31
  32 #include <assert.h>
  33 #include <ctype.h>
  34 #include <errno.h>
  35 #include <libintl.h>
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <strings.h>
  39 #include <unistd.h>
  40 #include <stddef.h>
  41 #include <fcntl.h>
  42 #include <sys/param.h>
  43 #include <sys/mount.h>
  44 #include <pthread.h>
  45 #include <umem.h>
  46 #include <time.h>
  47
  48 #include <libzfs.h>
  49 #include <libzfs_core.h>
  50
  51 #include "zfs_namecheck.h"
  52 #include "zfs_prop.h"
  53 #include "zfs_fletcher.h"
  54 #include "libzfs_impl.h"
  55 #include <zlib.h>
  56 #include <sha2.h>
  57 #include <sys/zio_checksum.h>
  58 #include <sys/ddt.h>
  59
  60 #ifdef __FreeBSD__
  61 extern int zfs_ioctl_version;
  62 #endif
  63
  64 /* in libzfs_dataset.c */
  65 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  66 /* We need to use something for ENODATA. */
  67 #define ENODATA EIDRM
  68
  69 static int zfs_receive_impl(libzfs_handle_t *, const char *, const char *,
  70     recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **, int,
  71     uint64_t *, const char *);
  72 static int guid_to_name(libzfs_handle_t *, const char *,
  73     uint64_t, boolean_t, char *);
  74
  75 static const zio_cksum_t zero_cksum = { 0 };
  76
  77 typedef struct dedup_arg {
  78         int     inputfd;
  79         int     outputfd;
  80         libzfs_handle_t  *dedup_hdl;
  81 } dedup_arg_t;
  82
  83 typedef struct progress_arg {
  84         zfs_handle_t *pa_zhp;
  85         int pa_fd;
  86         boolean_t pa_parsable;
  87 } progress_arg_t;
  88
  89 typedef struct dataref {
  90         uint64_t ref_guid;
  91         uint64_t ref_object;
  92         uint64_t ref_offset;
  93 } dataref_t;
  94
  95 typedef struct dedup_entry {
  96         struct dedup_entry      *dde_next;
  97         zio_cksum_t dde_chksum;
  98         uint64_t dde_prop;
  99         dataref_t dde_ref;
 100 } dedup_entry_t;
 101
 102 #define MAX_DDT_PHYSMEM_PERCENT         20
 103 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
 104
 105 typedef struct dedup_table {
 106         dedup_entry_t   **dedup_hash_array;
 107         umem_cache_t    *ddecache;
 108         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
 109         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
 110         uint64_t        ddt_count;
 111         int             numhashbits;
 112         boolean_t       ddt_full;
 113 } dedup_table_t;
 114
 115 static int
 116 high_order_bit(uint64_t n)
 117 {
 118         int count;
 119
 120         for (count = 0; n != 0; count++)
 121                 n >>= 1;
 122         return (count);
 123 }
 124
 125 static size_t
 126 ssread(void *buf, size_t len, FILE *stream)
 127 {
 128         size_t outlen;
 129
 130         if ((outlen = fread(buf, len, 1, stream)) == 0)
 131                 return (0);
 132
 133         return (outlen);
 134 }
 135
 136 static void
 137 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 138     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 139 {
 140         dedup_entry_t   *dde;
 141
 142         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 143                 if (ddt->ddt_full == B_FALSE) {
 144                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 145                             "Dedup table full.  Deduplication will continue "
 146                             "with existing table entries"));
 147                         ddt->ddt_full = B_TRUE;
 148                 }
 149                 return;
 150         }
 151
 152         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 153             != NULL) {
 154                 assert(*ddepp == NULL);
 155                 dde->dde_next = NULL;
 156                 dde->dde_chksum = *cs;
 157                 dde->dde_prop = prop;
 158                 dde->dde_ref = *dr;
 159                 *ddepp = dde;
 160                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 161                 ddt->ddt_count++;
 162         }
 163 }
 164
 165 /*
 166  * Using the specified dedup table, do a lookup for an entry with
 167  * the checksum cs.  If found, return the block's reference info
 168  * in *dr. Otherwise, insert a new entry in the dedup table, using
 169  * the reference information specified by *dr.
 170  *
 171  * return value:  true - entry was found
 172  *                false - entry was not found
 173  */
 174 static boolean_t
 175 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 176     uint64_t prop, dataref_t *dr)
 177 {
 178         uint32_t hashcode;
 179         dedup_entry_t **ddepp;
 180
 181         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 182
 183         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 184             ddepp = &((*ddepp)->dde_next)) {
 185                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 186                     (*ddepp)->dde_prop == prop) {
 187                         *dr = (*ddepp)->dde_ref;
 188                         return (B_TRUE);
 189                 }
 190         }
 191         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 192         return (B_FALSE);
 193 }
 194
 195 static int
 196 dump_record(dmu_replay_record_t *drr, void *payload, int payload_len,
 197     zio_cksum_t *zc, int outfd)
 198 {
 199         ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
 200             ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t));
 201         fletcher_4_incremental_native(drr,
 202             offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc);
 203         if (drr->drr_type != DRR_BEGIN) {
 204                 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u.
 205                     drr_checksum.drr_checksum));
 206                 drr->drr_u.drr_checksum.drr_checksum = *zc;
 207         }
 208         fletcher_4_incremental_native(&drr->drr_u.drr_checksum.drr_checksum,
 209             sizeof (zio_cksum_t), zc);
 210         if (write(outfd, drr, sizeof (*drr)) == -1)
 211                 return (errno);
 212         if (payload_len != 0) {
 213                 fletcher_4_incremental_native(payload, payload_len, zc);
 214                 if (write(outfd, payload, payload_len) == -1)
 215                         return (errno);
 216         }
 217         return (0);
 218 }
 219
 220 /*
 221  * This function is started in a separate thread when the dedup option
 222  * has been requested.  The main send thread determines the list of
 223  * snapshots to be included in the send stream and makes the ioctl calls
 224  * for each one.  But instead of having the ioctl send the output to the
 225  * the output fd specified by the caller of zfs_send()), the
 226  * ioctl is told to direct the output to a pipe, which is read by the
 227  * alternate thread running THIS function.  This function does the
 228  * dedup'ing by:
 229  *  1. building a dedup table (the DDT)
 230  *  2. doing checksums on each data block and inserting a record in the DDT
 231  *  3. looking for matching checksums, and
 232  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 233  *      a duplicate block is found.
 234  * The output of this function then goes to the output fd requested
 235  * by the caller of zfs_send().
 236  */
 237 static void *
 238 cksummer(void *arg)
 239 {
 240         dedup_arg_t *dda = arg;
 241         char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE);
 242         dmu_replay_record_t thedrr;
 243         dmu_replay_record_t *drr = &thedrr;
 244         FILE *ofp;
 245         int outfd;
 246         dedup_table_t ddt;
 247         zio_cksum_t stream_cksum;
 248         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 249         uint64_t numbuckets;
 250
 251         ddt.max_ddt_size =
 252             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100,
 253             SMALLEST_POSSIBLE_MAX_DDT_MB << 20);
 254
 255         numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t));
 256
 257         /*
 258          * numbuckets must be a power of 2.  Increase number to
 259          * a power of 2 if necessary.
 260          */
 261         if (!ISP2(numbuckets))
 262                 numbuckets = 1 << high_order_bit(numbuckets);
 263
 264         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 265         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 266             NULL, NULL, NULL, NULL, NULL, 0);
 267         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 268         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 269         ddt.ddt_full = B_FALSE;
 270
 271         outfd = dda->outputfd;
 272         ofp = fdopen(dda->inputfd, "r");
 273         while (ssread(drr, sizeof (*drr), ofp) != 0) {
 274
 275                 switch (drr->drr_type) {
 276                 case DRR_BEGIN:
 277                 {
 278                         struct drr_begin *drrb = &drr->drr_u.drr_begin;
 279                         int fflags;
 280                         int sz = 0;
 281                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 282
 283                         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 284
 285                         /* set the DEDUP feature flag for this stream */
 286                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 287                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 288                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 289                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 290
 291                         if (drr->drr_payloadlen != 0) {
 292                                 sz = drr->drr_payloadlen;
 293
 294                                 if (sz > SPA_MAXBLOCKSIZE) {
 295                                         buf = zfs_realloc(dda->dedup_hdl, buf,
 296                                             SPA_MAXBLOCKSIZE, sz);
 297                                 }
 298                                 (void) ssread(buf, sz, ofp);
 299                                 if (ferror(stdin))
 300                                         perror("fread");
 301                         }
 302                         if (dump_record(drr, buf, sz, &stream_cksum,
 303                             outfd) != 0)
 304                                 goto out;
 305                         break;
 306                 }
 307
 308                 case DRR_END:
 309                 {
 310                         struct drr_end *drre = &drr->drr_u.drr_end;
 311                         /* use the recalculated checksum */
 312                         drre->drr_checksum = stream_cksum;
 313                         if (dump_record(drr, NULL, 0, &stream_cksum,
 314                             outfd) != 0)
 315                                 goto out;
 316                         break;
 317                 }
 318
 319                 case DRR_OBJECT:
 320                 {
 321                         struct drr_object *drro = &drr->drr_u.drr_object;
 322                         if (drro->drr_bonuslen > 0) {
 323                                 (void) ssread(buf,
 324                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 325                                     ofp);
 326                         }
 327                         if (dump_record(drr, buf,
 328                             P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 329                             &stream_cksum, outfd) != 0)
 330                                 goto out;
 331                         break;
 332                 }
 333
 334                 case DRR_SPILL:
 335                 {
 336                         struct drr_spill *drrs = &drr->drr_u.drr_spill;
 337                         (void) ssread(buf, drrs->drr_length, ofp);
 338                         if (dump_record(drr, buf, drrs->drr_length,
 339                             &stream_cksum, outfd) != 0)
 340                                 goto out;
 341                         break;
 342                 }
 343
 344                 case DRR_FREEOBJECTS:
 345                 {
 346                         if (dump_record(drr, NULL, 0, &stream_cksum,
 347                             outfd) != 0)
 348                                 goto out;
 349                         break;
 350                 }
 351
 352                 case DRR_WRITE:
 353                 {
 354                         struct drr_write *drrw = &drr->drr_u.drr_write;
 355                         dataref_t       dataref;
 356
 357                         (void) ssread(buf, drrw->drr_length, ofp);
 358
 359                         /*
 360                          * Use the existing checksum if it's dedup-capable,
 361                          * else calculate a SHA256 checksum for it.
 362                          */
 363
 364                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 365                             zero_cksum) ||
 366                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 367                                 SHA256_CTX      ctx;
 368                                 zio_cksum_t     tmpsha256;
 369
 370                                 SHA256Init(&ctx);
 371                                 SHA256Update(&ctx, buf, drrw->drr_length);
 372                                 SHA256Final(&tmpsha256, &ctx);
 373                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 374                                     BE_64(tmpsha256.zc_word[0]);
 375                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 376                                     BE_64(tmpsha256.zc_word[1]);
 377                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 378                                     BE_64(tmpsha256.zc_word[2]);
 379                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 380                                     BE_64(tmpsha256.zc_word[3]);
 381                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 382                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 383                         }
 384
 385                         dataref.ref_guid = drrw->drr_toguid;
 386                         dataref.ref_object = drrw->drr_object;
 387                         dataref.ref_offset = drrw->drr_offset;
 388
 389                         if (ddt_update(dda->dedup_hdl, &ddt,
 390                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 391                             &dataref)) {
 392                                 dmu_replay_record_t wbr_drr = {0};
 393                                 struct drr_write_byref *wbr_drrr =
 394                                     &wbr_drr.drr_u.drr_write_byref;
 395
 396                                 /* block already present in stream */
 397                                 wbr_drr.drr_type = DRR_WRITE_BYREF;
 398
 399                                 wbr_drrr->drr_object = drrw->drr_object;
 400                                 wbr_drrr->drr_offset = drrw->drr_offset;
 401                                 wbr_drrr->drr_length = drrw->drr_length;
 402                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 403                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 404                                 wbr_drrr->drr_refobject =
 405                                     dataref.ref_object;
 406                                 wbr_drrr->drr_refoffset =
 407                                     dataref.ref_offset;
 408
 409                                 wbr_drrr->drr_checksumtype =
 410                                     drrw->drr_checksumtype;
 411                                 wbr_drrr->drr_checksumflags =
 412                                     drrw->drr_checksumtype;
 413                                 wbr_drrr->drr_key.ddk_cksum =
 414                                     drrw->drr_key.ddk_cksum;
 415                                 wbr_drrr->drr_key.ddk_prop =
 416                                     drrw->drr_key.ddk_prop;
 417
 418                                 if (dump_record(&wbr_drr, NULL, 0,
 419                                     &stream_cksum, outfd) != 0)
 420                                         goto out;
 421                         } else {
 422                                 /* block not previously seen */
 423                                 if (dump_record(drr, buf, drrw->drr_length,
 424                                     &stream_cksum, outfd) != 0)
 425                                         goto out;
 426                         }
 427                         break;
 428                 }
 429
 430                 case DRR_WRITE_EMBEDDED:
 431                 {
 432                         struct drr_write_embedded *drrwe =
 433                             &drr->drr_u.drr_write_embedded;
 434                         (void) ssread(buf,
 435                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp);
 436                         if (dump_record(drr, buf,
 437                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8),
 438                             &stream_cksum, outfd) != 0)
 439                                 goto out;
 440                         break;
 441                 }
 442
 443                 case DRR_FREE:
 444                 {
 445                         if (dump_record(drr, NULL, 0, &stream_cksum,
 446                             outfd) != 0)
 447                                 goto out;
 448                         break;
 449                 }
 450
 451                 default:
 452                         (void) fprintf(stderr, "INVALID record type 0x%x\n",
 453                             drr->drr_type);
 454                         /* should never happen, so assert */
 455                         assert(B_FALSE);
 456                 }
 457         }
 458 out:
 459         umem_cache_destroy(ddt.ddecache);
 460         free(ddt.dedup_hash_array);
 461         free(buf);
 462         (void) fclose(ofp);
 463
 464         return (NULL);
 465 }
 466
 467 /*
 468  * Routines for dealing with the AVL tree of fs-nvlists
 469  */
 470 typedef struct fsavl_node {
 471         avl_node_t fn_node;
 472         nvlist_t *fn_nvfs;
 473         char *fn_snapname;
 474         uint64_t fn_guid;
 475 } fsavl_node_t;
 476
 477 static int
 478 fsavl_compare(const void *arg1, const void *arg2)
 479 {
 480         const fsavl_node_t *fn1 = arg1;
 481         const fsavl_node_t *fn2 = arg2;
 482
 483         if (fn1->fn_guid > fn2->fn_guid)
 484                 return (+1);
 485         else if (fn1->fn_guid < fn2->fn_guid)
 486                 return (-1);
 487         else
 488                 return (0);
 489 }
 490
 491 /*
 492  * Given the GUID of a snapshot, find its containing filesystem and
 493  * (optionally) name.
 494  */
 495 static nvlist_t *
 496 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 497 {
 498         fsavl_node_t fn_find;
 499         fsavl_node_t *fn;
 500
 501         fn_find.fn_guid = snapguid;
 502
 503         fn = avl_find(avl, &fn_find, NULL);
 504         if (fn) {
 505                 if (snapname)
 506                         *snapname = fn->fn_snapname;
 507                 return (fn->fn_nvfs);
 508         }
 509         return (NULL);
 510 }
 511
 512 static void
 513 fsavl_destroy(avl_tree_t *avl)
 514 {
 515         fsavl_node_t *fn;
 516         void *cookie;
 517
 518         if (avl == NULL)
 519                 return;
 520
 521         cookie = NULL;
 522         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 523                 free(fn);
 524         avl_destroy(avl);
 525         free(avl);
 526 }
 527
 528 /*
 529  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 530  */
 531 static avl_tree_t *
 532 fsavl_create(nvlist_t *fss)
 533 {
 534         avl_tree_t *fsavl;
 535         nvpair_t *fselem = NULL;
 536
 537         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 538                 return (NULL);
 539
 540         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 541             offsetof(fsavl_node_t, fn_node));
 542
 543         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 544                 nvlist_t *nvfs, *snaps;
 545                 nvpair_t *snapelem = NULL;
 546
 547                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 548                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 549
 550                 while ((snapelem =
 551                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 552                         fsavl_node_t *fn;
 553                         uint64_t guid;
 554
 555                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 556                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 557                                 fsavl_destroy(fsavl);
 558                                 return (NULL);
 559                         }
 560                         fn->fn_nvfs = nvfs;
 561                         fn->fn_snapname = nvpair_name(snapelem);
 562                         fn->fn_guid = guid;
 563
 564                         /*
 565                          * Note: if there are multiple snaps with the
 566                          * same GUID, we ignore all but one.
 567                          */
 568                         if (avl_find(fsavl, fn, NULL) == NULL)
 569                                 avl_add(fsavl, fn);
 570                         else
 571                                 free(fn);
 572                 }
 573         }
 574
 575         return (fsavl);
 576 }
 577
 578 /*
 579  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 580  */
 581 typedef struct send_data {
 582         uint64_t parent_fromsnap_guid;
 583         nvlist_t *parent_snaps;
 584         nvlist_t *fss;
 585         nvlist_t *snapprops;
 586         const char *fromsnap;
 587         const char *tosnap;
 588         boolean_t recursive;
 589
 590         /*
 591          * The header nvlist is of the following format:
 592          * {
 593          *   "tosnap" -> string
 594          *   "fromsnap" -> string (if incremental)
 595          *   "fss" -> {
 596          *      id -> {
 597          *
 598          *       "name" -> string (full name; for debugging)
 599          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 600          *
 601          *       "props" -> { name -> value (only if set here) }
 602          *       "snaps" -> { name (lastname) -> number (guid) }
 603          *       "snapprops" -> { name (lastname) -> { name -> value } }
 604          *
 605          *       "origin" -> number (guid) (if clone)
 606          *       "sent" -> boolean (not on-disk)
 607          *      }
 608          *   }
 609          * }
 610          *
 611          */
 612 } send_data_t;
 613
 614 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 615
 616 static int
 617 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 618 {
 619         send_data_t *sd = arg;
 620         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 621         char *snapname;
 622         nvlist_t *nv;
 623
 624         snapname = strrchr(zhp->zfs_name, '@')+1;
 625
 626         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 627         /*
 628          * NB: if there is no fromsnap here (it's a newly created fs in
 629          * an incremental replication), we will substitute the tosnap.
 630          */
 631         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 632             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 633             strcmp(snapname, sd->tosnap) == 0)) {
 634                 sd->parent_fromsnap_guid = guid;
 635         }
 636
 637         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 638         send_iterate_prop(zhp, nv);
 639         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 640         nvlist_free(nv);
 641
 642         zfs_close(zhp);
 643         return (0);
 644 }
 645
 646 static void
 647 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 648 {
 649         nvpair_t *elem = NULL;
 650
 651         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 652                 char *propname = nvpair_name(elem);
 653                 zfs_prop_t prop = zfs_name_to_prop(propname);
 654                 nvlist_t *propnv;
 655
 656                 if (!zfs_prop_user(propname)) {
 657                         /*
 658                          * Realistically, this should never happen.  However,
 659                          * we want the ability to add DSL properties without
 660                          * needing to make incompatible version changes.  We
 661                          * need to ignore unknown properties to allow older
 662                          * software to still send datasets containing these
 663                          * properties, with the unknown properties elided.
 664                          */
 665                         if (prop == ZPROP_INVAL)
 666                                 continue;
 667
 668                         if (zfs_prop_readonly(prop))
 669                                 continue;
 670                 }
 671
 672                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 673                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 674                     prop == ZFS_PROP_REFQUOTA ||
 675                     prop == ZFS_PROP_REFRESERVATION) {
 676                         char *source;
 677                         uint64_t value;
 678                         verify(nvlist_lookup_uint64(propnv,
 679                             ZPROP_VALUE, &value) == 0);
 680                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 681                                 continue;
 682                         /*
 683                          * May have no source before SPA_VERSION_RECVD_PROPS,
 684                          * but is still modifiable.
 685                          */
 686                         if (nvlist_lookup_string(propnv,
 687                             ZPROP_SOURCE, &source) == 0) {
 688                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 689                                     (strcmp(source,
 690                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 691                                         continue;
 692                         }
 693                 } else {
 694                         char *source;
 695                         if (nvlist_lookup_string(propnv,
 696                             ZPROP_SOURCE, &source) != 0)
 697                                 continue;
 698                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 699                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 700                                 continue;
 701                 }
 702
 703                 if (zfs_prop_user(propname) ||
 704                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 705                         char *value;
 706                         verify(nvlist_lookup_string(propnv,
 707                             ZPROP_VALUE, &value) == 0);
 708                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 709                 } else {
 710                         uint64_t value;
 711                         verify(nvlist_lookup_uint64(propnv,
 712                             ZPROP_VALUE, &value) == 0);
 713                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 714                 }
 715         }
 716 }
 717
 718 /*
 719  * recursively generate nvlists describing datasets.  See comment
 720  * for the data structure send_data_t above for description of contents
 721  * of the nvlist.
 722  */
 723 static int
 724 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 725 {
 726         send_data_t *sd = arg;
 727         nvlist_t *nvfs, *nv;
 728         int rv = 0;
 729         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 730         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 731         char guidstring[64];
 732
 733         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 734         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 735         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 736             sd->parent_fromsnap_guid));
 737
 738         if (zhp->zfs_dmustats.dds_origin[0]) {
 739                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 740                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 741                 if (origin == NULL)
 742                         return (-1);
 743                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 744                     origin->zfs_dmustats.dds_guid));
 745         }
 746
 747         /* iterate over props */
 748         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 749         send_iterate_prop(zhp, nv);
 750         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 751         nvlist_free(nv);
 752
 753         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 754         sd->parent_fromsnap_guid = 0;
 755         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 756         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 757         (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd);
 758         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 759         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 760         nvlist_free(sd->parent_snaps);
 761         nvlist_free(sd->snapprops);
 762
 763         /* add this fs to nvlist */
 764         (void) snprintf(guidstring, sizeof (guidstring),
 765             "0x%llx", (longlong_t)guid);
 766         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 767         nvlist_free(nvfs);
 768
 769         /* iterate over children */
 770         if (sd->recursive)
 771                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 772
 773         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 774
 775         zfs_close(zhp);
 776         return (rv);
 777 }
 778
 779 static int
 780 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 781     const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
 782 {
 783         zfs_handle_t *zhp;
 784         send_data_t sd = { 0 };
 785         int error;
 786
 787         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 788         if (zhp == NULL)
 789                 return (EZFS_BADTYPE);
 790
 791         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 792         sd.fromsnap = fromsnap;
 793         sd.tosnap = tosnap;
 794         sd.recursive = recursive;
 795
 796         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 797                 nvlist_free(sd.fss);
 798                 if (avlp != NULL)
 799                         *avlp = NULL;
 800                 *nvlp = NULL;
 801                 return (error);
 802         }
 803
 804         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 805                 nvlist_free(sd.fss);
 806                 *nvlp = NULL;
 807                 return (EZFS_NOMEM);
 808         }
 809
 810         *nvlp = sd.fss;
 811         return (0);
 812 }
 813
 814 /*
 815  * Routines specific to "zfs send"
 816  */
 817 typedef struct send_dump_data {
 818         /* these are all just the short snapname (the part after the @) */
 819         const char *fromsnap;
 820         const char *tosnap;
 821         char prevsnap[ZFS_MAXNAMELEN];
 822         uint64_t prevsnap_obj;
 823         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 824         boolean_t verbose, dryrun, parsable, progress, embed_data, std_out;
 825         boolean_t large_block;
 826         int outfd;
 827         boolean_t err;
 828         nvlist_t *fss;
 829         nvlist_t *snapholds;
 830         avl_tree_t *fsavl;
 831         snapfilter_cb_t *filter_cb;
 832         void *filter_cb_arg;
 833         nvlist_t *debugnv;
 834         char holdtag[ZFS_MAXNAMELEN];
 835         int cleanup_fd;
 836         uint64_t size;
 837 } send_dump_data_t;
 838
 839 static int
 840 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
 841     boolean_t fromorigin, uint64_t *sizep)
 842 {
 843         zfs_cmd_t zc = { 0 };
 844         libzfs_handle_t *hdl = zhp->zfs_hdl;
 845
 846         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 847         assert(fromsnap_obj == 0 || !fromorigin);
 848
 849         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 850         zc.zc_obj = fromorigin;
 851         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 852         zc.zc_fromobj = fromsnap_obj;
 853         zc.zc_guid = 1;  /* estimate flag */
 854
 855         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 856                 char errbuf[1024];
 857                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 858                     "warning: cannot estimate space for '%s'"), zhp->zfs_name);
 859
 860                 switch (errno) {
 861                 case EXDEV:
 862                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 863                             "not an earlier snapshot from the same fs"));
 864                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 865
 866                 case ENOENT:
 867                         if (zfs_dataset_exists(hdl, zc.zc_name,
 868                             ZFS_TYPE_SNAPSHOT)) {
 869                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 870                                     "incremental source (@%s) does not exist"),
 871                                     zc.zc_value);
 872                         }
 873                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 874
 875                 case EDQUOT:
 876                 case EFBIG:
 877                 case EIO:
 878                 case ENOLINK:
 879                 case ENOSPC:
 880                 case ENXIO:
 881                 case EPIPE:
 882                 case ERANGE:
 883                 case EFAULT:
 884                 case EROFS:
 885                         zfs_error_aux(hdl, strerror(errno));
 886                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 887
 888                 default:
 889                         return (zfs_standard_error(hdl, errno, errbuf));
 890                 }
 891         }
 892
 893         *sizep = zc.zc_objset_type;
 894
 895         return (0);
 896 }
 897
 898 /*
 899  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 900  * NULL) to the file descriptor specified by outfd.
 901  */
 902 static int
 903 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
 904     boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
 905     nvlist_t *debugnv)
 906 {
 907         zfs_cmd_t zc = { 0 };
 908         libzfs_handle_t *hdl = zhp->zfs_hdl;
 909         nvlist_t *thisdbg;
 910
 911         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 912         assert(fromsnap_obj == 0 || !fromorigin);
 913
 914         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 915         zc.zc_cookie = outfd;
 916         zc.zc_obj = fromorigin;
 917         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 918         zc.zc_fromobj = fromsnap_obj;
 919         zc.zc_flags = flags;
 920
 921         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 922         if (fromsnap && fromsnap[0] != '\0') {
 923                 VERIFY(0 == nvlist_add_string(thisdbg,
 924                     "fromsnap", fromsnap));
 925         }
 926
 927         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 928                 char errbuf[1024];
 929                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 930                     "warning: cannot send '%s'"), zhp->zfs_name);
 931
 932                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
 933                 if (debugnv) {
 934                         VERIFY(0 == nvlist_add_nvlist(debugnv,
 935                             zhp->zfs_name, thisdbg));
 936                 }
 937                 nvlist_free(thisdbg);
 938
 939                 switch (errno) {
 940                 case EXDEV:
 941                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 942                             "not an earlier snapshot from the same fs"));
 943                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 944
 945                 case ENOENT:
 946                         if (zfs_dataset_exists(hdl, zc.zc_name,
 947                             ZFS_TYPE_SNAPSHOT)) {
 948                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 949                                     "incremental source (@%s) does not exist"),
 950                                     zc.zc_value);
 951                         }
 952                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 953
 954                 case EDQUOT:
 955                 case EFBIG:
 956                 case EIO:
 957                 case ENOLINK:
 958                 case ENOSPC:
 959 #ifdef illumos
 960                 case ENOSTR:
 961 #endif
 962                 case ENXIO:
 963                 case EPIPE:
 964                 case ERANGE:
 965                 case EFAULT:
 966                 case EROFS:
 967                         zfs_error_aux(hdl, strerror(errno));
 968                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 969
 970                 default:
 971                         return (zfs_standard_error(hdl, errno, errbuf));
 972                 }
 973         }
 974
 975         if (debugnv)
 976                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
 977         nvlist_free(thisdbg);
 978
 979         return (0);
 980 }
 981
 982 static void
 983 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
 984 {
 985         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 986
 987         /*
 988          * zfs_send() only sets snapholds for sends that need them,
 989          * e.g. replication and doall.
 990          */
 991         if (sdd->snapholds == NULL)
 992                 return;
 993
 994         fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
 995 }
 996
 997 static void *
 998 send_progress_thread(void *arg)
 999 {
1000         progress_arg_t *pa = arg;
1001         zfs_cmd_t zc = { 0 };
1002         zfs_handle_t *zhp = pa->pa_zhp;
1003         libzfs_handle_t *hdl = zhp->zfs_hdl;
1004         unsigned long long bytes;
1005         char buf[16];
1006         time_t t;
1007         struct tm *tm;
1008
1009         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1010
1011         if (!pa->pa_parsable)
1012                 (void) fprintf(stderr, "TIME        SENT   SNAPSHOT\n");
1013
1014         /*
1015          * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1016          */
1017         for (;;) {
1018                 (void) sleep(1);
1019
1020                 zc.zc_cookie = pa->pa_fd;
1021                 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1022                         return ((void *)-1);
1023
1024                 (void) time(&t);
1025                 tm = localtime(&t);
1026                 bytes = zc.zc_cookie;
1027
1028                 if (pa->pa_parsable) {
1029                         (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1030                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1031                             bytes, zhp->zfs_name);
1032                 } else {
1033                         zfs_nicenum(bytes, buf, sizeof (buf));
1034                         (void) fprintf(stderr, "%02d:%02d:%02d   %5s   %s\n",
1035                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1036                             buf, zhp->zfs_name);
1037                 }
1038         }
1039 }
1040
1041 static void
1042 send_print_verbose(FILE *fout, const char *tosnap, const char *fromsnap,
1043     uint64_t size, boolean_t parsable)
1044 {
1045         if (parsable) {
1046                 if (fromsnap != NULL) {
1047                         (void) fprintf(fout, "incremental\t%s\t%s",
1048                             fromsnap, tosnap);
1049                 } else {
1050                         (void) fprintf(fout, "full\t%s",
1051                             tosnap);
1052                 }
1053         } else {
1054                 if (fromsnap != NULL) {
1055                         if (strchr(fromsnap, '@') == NULL &&
1056                             strchr(fromsnap, '#') == NULL) {
1057                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1058                                     "send from @%s to %s"),
1059                                     fromsnap, tosnap);
1060                         } else {
1061                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1062                                     "send from %s to %s"),
1063                                     fromsnap, tosnap);
1064                         }
1065                 } else {
1066                         (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1067                             "full send of %s"),
1068                             tosnap);
1069                 }
1070         }
1071
1072         if (size != 0) {
1073                 if (parsable) {
1074                         (void) fprintf(fout, "\t%llu",
1075                             (longlong_t)size);
1076                 } else {
1077                         char buf[16];
1078                         zfs_nicenum(size, buf, sizeof (buf));
1079                         (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1080                             " estimated size is %s"), buf);
1081                 }
1082         }
1083         (void) fprintf(fout, "\n");
1084 }
1085
1086 static int
1087 dump_snapshot(zfs_handle_t *zhp, void *arg)
1088 {
1089         send_dump_data_t *sdd = arg;
1090         progress_arg_t pa = { 0 };
1091         pthread_t tid;
1092         char *thissnap;
1093         int err;
1094         boolean_t isfromsnap, istosnap, fromorigin;
1095         boolean_t exclude = B_FALSE;
1096         FILE *fout = sdd->std_out ? stdout : stderr;
1097
1098         err = 0;
1099         thissnap = strchr(zhp->zfs_name, '@') + 1;
1100         isfromsnap = (sdd->fromsnap != NULL &&
1101             strcmp(sdd->fromsnap, thissnap) == 0);
1102
1103         if (!sdd->seenfrom && isfromsnap) {
1104                 gather_holds(zhp, sdd);
1105                 sdd->seenfrom = B_TRUE;
1106                 (void) strcpy(sdd->prevsnap, thissnap);
1107                 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1108                 zfs_close(zhp);
1109                 return (0);
1110         }
1111
1112         if (sdd->seento || !sdd->seenfrom) {
1113                 zfs_close(zhp);
1114                 return (0);
1115         }
1116
1117         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1118         if (istosnap)
1119                 sdd->seento = B_TRUE;
1120
1121         if (!sdd->doall && !isfromsnap && !istosnap) {
1122                 if (sdd->replicate) {
1123                         char *snapname;
1124                         nvlist_t *snapprops;
1125                         /*
1126                          * Filter out all intermediate snapshots except origin
1127                          * snapshots needed to replicate clones.
1128                          */
1129                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1130                             zhp->zfs_dmustats.dds_guid, &snapname);
1131
1132                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1133                             "snapprops", &snapprops));
1134                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1135                             thissnap, &snapprops));
1136                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
1137                 } else {
1138                         exclude = B_TRUE;
1139                 }
1140         }
1141
1142         /*
1143          * If a filter function exists, call it to determine whether
1144          * this snapshot will be sent.
1145          */
1146         if (exclude || (sdd->filter_cb != NULL &&
1147             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1148                 /*
1149                  * This snapshot is filtered out.  Don't send it, and don't
1150                  * set prevsnap_obj, so it will be as if this snapshot didn't
1151                  * exist, and the next accepted snapshot will be sent as
1152                  * an incremental from the last accepted one, or as the
1153                  * first (and full) snapshot in the case of a replication,
1154                  * non-incremental send.
1155                  */
1156                 zfs_close(zhp);
1157                 return (0);
1158         }
1159
1160         gather_holds(zhp, sdd);
1161         fromorigin = sdd->prevsnap[0] == '\0' &&
1162             (sdd->fromorigin || sdd->replicate);
1163
1164         if (sdd->verbose) {
1165                 uint64_t size = 0;
1166                 (void) estimate_ioctl(zhp, sdd->prevsnap_obj,
1167                     fromorigin, &size);
1168
1169                 send_print_verbose(fout, zhp->zfs_name,
1170                     sdd->prevsnap[0] ? sdd->prevsnap : NULL,
1171                     size, sdd->parsable);
1172                 sdd->size += size;
1173         }
1174
1175         if (!sdd->dryrun) {
1176                 /*
1177                  * If progress reporting is requested, spawn a new thread to
1178                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1179                  */
1180                 if (sdd->progress) {
1181                         pa.pa_zhp = zhp;
1182                         pa.pa_fd = sdd->outfd;
1183                         pa.pa_parsable = sdd->parsable;
1184
1185                         if (err = pthread_create(&tid, NULL,
1186                             send_progress_thread, &pa)) {
1187                                 zfs_close(zhp);
1188                                 return (err);
1189                         }
1190                 }
1191
1192                 enum lzc_send_flags flags = 0;
1193                 if (sdd->large_block)
1194                         flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1195                 if (sdd->embed_data)
1196                         flags |= LZC_SEND_FLAG_EMBED_DATA;
1197
1198                 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1199                     fromorigin, sdd->outfd, flags, sdd->debugnv);
1200
1201                 if (sdd->progress) {
1202                         (void) pthread_cancel(tid);
1203                         (void) pthread_join(tid, NULL);
1204                 }
1205         }
1206
1207         (void) strcpy(sdd->prevsnap, thissnap);
1208         sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1209         zfs_close(zhp);
1210         return (err);
1211 }
1212
1213 static int
1214 dump_filesystem(zfs_handle_t *zhp, void *arg)
1215 {
1216         int rv = 0;
1217         send_dump_data_t *sdd = arg;
1218         boolean_t missingfrom = B_FALSE;
1219         zfs_cmd_t zc = { 0 };
1220
1221         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1222             zhp->zfs_name, sdd->tosnap);
1223         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1224                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1225                     "WARNING: could not send %s@%s: does not exist\n"),
1226                     zhp->zfs_name, sdd->tosnap);
1227                 sdd->err = B_TRUE;
1228                 return (0);
1229         }
1230
1231         if (sdd->replicate && sdd->fromsnap) {
1232                 /*
1233                  * If this fs does not have fromsnap, and we're doing
1234                  * recursive, we need to send a full stream from the
1235                  * beginning (or an incremental from the origin if this
1236                  * is a clone).  If we're doing non-recursive, then let
1237                  * them get the error.
1238                  */
1239                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1240                     zhp->zfs_name, sdd->fromsnap);
1241                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1242                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1243                         missingfrom = B_TRUE;
1244                 }
1245         }
1246
1247         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1248         sdd->prevsnap_obj = 0;
1249         if (sdd->fromsnap == NULL || missingfrom)
1250                 sdd->seenfrom = B_TRUE;
1251
1252         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1253         if (!sdd->seenfrom) {
1254                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1255                     "WARNING: could not send %s@%s:\n"
1256                     "incremental source (%s@%s) does not exist\n"),
1257                     zhp->zfs_name, sdd->tosnap,
1258                     zhp->zfs_name, sdd->fromsnap);
1259                 sdd->err = B_TRUE;
1260         } else if (!sdd->seento) {
1261                 if (sdd->fromsnap) {
1262                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1263                             "WARNING: could not send %s@%s:\n"
1264                             "incremental source (%s@%s) "
1265                             "is not earlier than it\n"),
1266                             zhp->zfs_name, sdd->tosnap,
1267                             zhp->zfs_name, sdd->fromsnap);
1268                 } else {
1269                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1270                             "WARNING: "
1271                             "could not send %s@%s: does not exist\n"),
1272                             zhp->zfs_name, sdd->tosnap);
1273                 }
1274                 sdd->err = B_TRUE;
1275         }
1276
1277         return (rv);
1278 }
1279
1280 static int
1281 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1282 {
1283         send_dump_data_t *sdd = arg;
1284         nvpair_t *fspair;
1285         boolean_t needagain, progress;
1286
1287         if (!sdd->replicate)
1288                 return (dump_filesystem(rzhp, sdd));
1289
1290         /* Mark the clone origin snapshots. */
1291         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1292             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1293                 nvlist_t *nvfs;
1294                 uint64_t origin_guid = 0;
1295
1296                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1297                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1298                 if (origin_guid != 0) {
1299                         char *snapname;
1300                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1301                             origin_guid, &snapname);
1302                         if (origin_nv != NULL) {
1303                                 nvlist_t *snapprops;
1304                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1305                                     "snapprops", &snapprops));
1306                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1307                                     snapname, &snapprops));
1308                                 VERIFY(0 == nvlist_add_boolean(
1309                                     snapprops, "is_clone_origin"));
1310                         }
1311                 }
1312         }
1313 again:
1314         needagain = progress = B_FALSE;
1315         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1316             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1317                 nvlist_t *fslist, *parent_nv;
1318                 char *fsname;
1319                 zfs_handle_t *zhp;
1320                 int err;
1321                 uint64_t origin_guid = 0;
1322                 uint64_t parent_guid = 0;
1323
1324                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1325                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1326                         continue;
1327
1328                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1329                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1330                 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1331                     &parent_guid);
1332
1333                 if (parent_guid != 0) {
1334                         parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1335                         if (!nvlist_exists(parent_nv, "sent")) {
1336                                 /* parent has not been sent; skip this one */
1337                                 needagain = B_TRUE;
1338                                 continue;
1339                         }
1340                 }
1341
1342                 if (origin_guid != 0) {
1343                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1344                             origin_guid, NULL);
1345                         if (origin_nv != NULL &&
1346                             !nvlist_exists(origin_nv, "sent")) {
1347                                 /*
1348                                  * origin has not been sent yet;
1349                                  * skip this clone.
1350                                  */
1351                                 needagain = B_TRUE;
1352                                 continue;
1353                         }
1354                 }
1355
1356                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1357                 if (zhp == NULL)
1358                         return (-1);
1359                 err = dump_filesystem(zhp, sdd);
1360                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1361                 progress = B_TRUE;
1362                 zfs_close(zhp);
1363                 if (err)
1364                         return (err);
1365         }
1366         if (needagain) {
1367                 assert(progress);
1368                 goto again;
1369         }
1370
1371         /* clean out the sent flags in case we reuse this fss */
1372         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1373             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1374                 nvlist_t *fslist;
1375
1376                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1377                 (void) nvlist_remove_all(fslist, "sent");
1378         }
1379
1380         return (0);
1381 }
1382
1383 nvlist_t *
1384 zfs_send_resume_token_to_nvlist(libzfs_handle_t *hdl, const char *token)
1385 {
1386         unsigned int version;
1387         int nread;
1388         unsigned long long checksum, packed_len;
1389
1390         /*
1391          * Decode token header, which is:
1392          *   <token version>-<checksum of payload>-<uncompressed payload length>
1393          * Note that the only supported token version is 1.
1394          */
1395         nread = sscanf(token, "%u-%llx-%llx-",
1396             &version, &checksum, &packed_len);
1397         if (nread != 3) {
1398                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1399                     "resume token is corrupt (invalid format)"));
1400                 return (NULL);
1401         }
1402
1403         if (version != ZFS_SEND_RESUME_TOKEN_VERSION) {
1404                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1405                     "resume token is corrupt (invalid version %u)"),
1406                     version);
1407                 return (NULL);
1408         }
1409
1410         /* convert hexadecimal representation to binary */
1411         token = strrchr(token, '-') + 1;
1412         int len = strlen(token) / 2;
1413         unsigned char *compressed = zfs_alloc(hdl, len);
1414         for (int i = 0; i < len; i++) {
1415                 nread = sscanf(token + i * 2, "%2hhx", compressed + i);
1416                 if (nread != 1) {
1417                         free(compressed);
1418                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1419                             "resume token is corrupt "
1420                             "(payload is not hex-encoded)"));
1421                         return (NULL);
1422                 }
1423         }
1424
1425         /* verify checksum */
1426         zio_cksum_t cksum;
1427         fletcher_4_native(compressed, len, NULL, &cksum);
1428         if (cksum.zc_word[0] != checksum) {
1429                 free(compressed);
1430                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1431                     "resume token is corrupt (incorrect checksum)"));
1432                 return (NULL);
1433         }
1434
1435         /* uncompress */
1436         void *packed = zfs_alloc(hdl, packed_len);
1437         uLongf packed_len_long = packed_len;
1438         if (uncompress(packed, &packed_len_long, compressed, len) != Z_OK ||
1439             packed_len_long != packed_len) {
1440                 free(packed);
1441                 free(compressed);
1442                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1443                     "resume token is corrupt (decompression failed)"));
1444                 return (NULL);
1445         }
1446
1447         /* unpack nvlist */
1448         nvlist_t *nv;
1449         int error = nvlist_unpack(packed, packed_len, &nv, KM_SLEEP);
1450         free(packed);
1451         free(compressed);
1452         if (error != 0) {
1453                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1454                     "resume token is corrupt (nvlist_unpack failed)"));
1455                 return (NULL);
1456         }
1457         return (nv);
1458 }
1459
1460 int
1461 zfs_send_resume(libzfs_handle_t *hdl, sendflags_t *flags, int outfd,
1462     const char *resume_token)
1463 {
1464         char errbuf[1024];
1465         char *toname;
1466         char *fromname = NULL;
1467         uint64_t resumeobj, resumeoff, toguid, fromguid, bytes;
1468         zfs_handle_t *zhp;
1469         int error = 0;
1470         char name[ZFS_MAXNAMELEN];
1471         enum lzc_send_flags lzc_flags = 0;
1472
1473         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1474             "cannot resume send"));
1475
1476         nvlist_t *resume_nvl =
1477             zfs_send_resume_token_to_nvlist(hdl, resume_token);
1478         if (resume_nvl == NULL) {
1479                 /*
1480                  * zfs_error_aux has already been set by
1481                  * zfs_send_resume_token_to_nvlist
1482                  */
1483                 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1484         }
1485         if (flags->verbose) {
1486                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1487                     "resume token contents:\n"));
1488                 nvlist_print(stderr, resume_nvl);
1489         }
1490
1491         if (nvlist_lookup_string(resume_nvl, "toname", &toname) != 0 ||
1492             nvlist_lookup_uint64(resume_nvl, "object", &resumeobj) != 0 ||
1493             nvlist_lookup_uint64(resume_nvl, "offset", &resumeoff) != 0 ||
1494             nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 ||
1495             nvlist_lookup_uint64(resume_nvl, "toguid", &toguid) != 0) {
1496                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1497                     "resume token is corrupt"));
1498                 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1499         }
1500         fromguid = 0;
1501         (void) nvlist_lookup_uint64(resume_nvl, "fromguid", &fromguid);
1502
1503         if (flags->embed_data || nvlist_exists(resume_nvl, "embedok"))
1504                 lzc_flags |= LZC_SEND_FLAG_EMBED_DATA;
1505
1506         if (guid_to_name(hdl, toname, toguid, B_FALSE, name) != 0) {
1507                 if (zfs_dataset_exists(hdl, toname, ZFS_TYPE_DATASET)) {
1508                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1509                             "'%s' is no longer the same snapshot used in "
1510                             "the initial send"), toname);
1511                 } else {
1512                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1513                             "'%s' used in the initial send no longer exists"),
1514                             toname);
1515                 }
1516                 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1517         }
1518         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1519         if (zhp == NULL) {
1520                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1521                     "unable to access '%s'"), name);
1522                 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1523         }
1524
1525         if (fromguid != 0) {
1526                 if (guid_to_name(hdl, toname, fromguid, B_TRUE, name) != 0) {
1527                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1528                             "incremental source %#llx no longer exists"),
1529                             (longlong_t)fromguid);
1530                         return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1531                 }
1532                 fromname = name;
1533         }
1534
1535         if (flags->verbose) {
1536                 uint64_t size = 0;
1537                 error = lzc_send_space(zhp->zfs_name, fromname, &size);
1538                 if (error == 0)
1539                         size = MAX(0, (int64_t)(size - bytes));
1540                 send_print_verbose(stderr, zhp->zfs_name, fromname,
1541                     size, flags->parsable);
1542         }
1543
1544         if (!flags->dryrun) {
1545                 progress_arg_t pa = { 0 };
1546                 pthread_t tid;
1547                 /*
1548                  * If progress reporting is requested, spawn a new thread to
1549                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1550                  */
1551                 if (flags->progress) {
1552                         pa.pa_zhp = zhp;
1553                         pa.pa_fd = outfd;
1554                         pa.pa_parsable = flags->parsable;
1555
1556                         error = pthread_create(&tid, NULL,
1557                             send_progress_thread, &pa);
1558                         if (error != 0) {
1559                                 zfs_close(zhp);
1560                                 return (error);
1561                         }
1562                 }
1563
1564                 error = lzc_send_resume(zhp->zfs_name, fromname, outfd,
1565                     lzc_flags, resumeobj, resumeoff);
1566
1567                 if (flags->progress) {
1568                         (void) pthread_cancel(tid);
1569                         (void) pthread_join(tid, NULL);
1570                 }
1571
1572                 char errbuf[1024];
1573                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1574                     "warning: cannot send '%s'"), zhp->zfs_name);
1575
1576                 zfs_close(zhp);
1577
1578                 switch (error) {
1579                 case 0:
1580                         return (0);
1581                 case EXDEV:
1582                 case ENOENT:
1583                 case EDQUOT:
1584                 case EFBIG:
1585                 case EIO:
1586                 case ENOLINK:
1587                 case ENOSPC:
1588 #ifdef illumos
1589                 case ENOSTR:
1590 #endif
1591                 case ENXIO:
1592                 case EPIPE:
1593                 case ERANGE:
1594                 case EFAULT:
1595                 case EROFS:
1596                         zfs_error_aux(hdl, strerror(errno));
1597                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1598
1599                 default:
1600                         return (zfs_standard_error(hdl, errno, errbuf));
1601                 }
1602         }
1603
1604
1605         zfs_close(zhp);
1606
1607         return (error);
1608 }
1609
1610 /*
1611  * Generate a send stream for the dataset identified by the argument zhp.
1612  *
1613  * The content of the send stream is the snapshot identified by
1614  * 'tosnap'.  Incremental streams are requested in two ways:
1615  *     - from the snapshot identified by "fromsnap" (if non-null) or
1616  *     - from the origin of the dataset identified by zhp, which must
1617  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1618  *       is TRUE.
1619  *
1620  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1621  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1622  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1623  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1624  * case too. If "props" is set, send properties.
1625  */
1626 int
1627 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1628     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1629     void *cb_arg, nvlist_t **debugnvp)
1630 {
1631         char errbuf[1024];
1632         send_dump_data_t sdd = { 0 };
1633         int err = 0;
1634         nvlist_t *fss = NULL;
1635         avl_tree_t *fsavl = NULL;
1636         static uint64_t holdseq;
1637         int spa_version;
1638         pthread_t tid = 0;
1639         int pipefd[2];
1640         dedup_arg_t dda = { 0 };
1641         int featureflags = 0;
1642         FILE *fout;
1643
1644         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1645             "cannot send '%s'"), zhp->zfs_name);
1646
1647         if (fromsnap && fromsnap[0] == '\0') {
1648                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1649                     "zero-length incremental source"));
1650                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1651         }
1652
1653         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1654                 uint64_t version;
1655                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1656                 if (version >= ZPL_VERSION_SA) {
1657                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1658                 }
1659         }
1660
1661         if (flags->dedup && !flags->dryrun) {
1662                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1663                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1664                 if (err = pipe(pipefd)) {
1665                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1666                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1667                             errbuf));
1668                 }
1669                 dda.outputfd = outfd;
1670                 dda.inputfd = pipefd[1];
1671                 dda.dedup_hdl = zhp->zfs_hdl;
1672                 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1673                         (void) close(pipefd[0]);
1674                         (void) close(pipefd[1]);
1675                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1676                         return (zfs_error(zhp->zfs_hdl,
1677                             EZFS_THREADCREATEFAILED, errbuf));
1678                 }
1679         }
1680
1681         if (flags->replicate || flags->doall || flags->props) {
1682                 dmu_replay_record_t drr = { 0 };
1683                 char *packbuf = NULL;
1684                 size_t buflen = 0;
1685                 zio_cksum_t zc = { 0 };
1686
1687                 if (flags->replicate || flags->props) {
1688                         nvlist_t *hdrnv;
1689
1690                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1691                         if (fromsnap) {
1692                                 VERIFY(0 == nvlist_add_string(hdrnv,
1693                                     "fromsnap", fromsnap));
1694                         }
1695                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1696                         if (!flags->replicate) {
1697                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1698                                     "not_recursive"));
1699                         }
1700
1701                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1702                             fromsnap, tosnap, flags->replicate, &fss, &fsavl);
1703                         if (err)
1704                                 goto err_out;
1705                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1706                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1707                             NV_ENCODE_XDR, 0);
1708                         if (debugnvp)
1709                                 *debugnvp = hdrnv;
1710                         else
1711                                 nvlist_free(hdrnv);
1712                         if (err)
1713                                 goto stderr_out;
1714                 }
1715
1716                 if (!flags->dryrun) {
1717                         /* write first begin record */
1718                         drr.drr_type = DRR_BEGIN;
1719                         drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1720                         DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1721                             drr_versioninfo, DMU_COMPOUNDSTREAM);
1722                         DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1723                             drr_versioninfo, featureflags);
1724                         (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1725                             sizeof (drr.drr_u.drr_begin.drr_toname),
1726                             "%s@%s", zhp->zfs_name, tosnap);
1727                         drr.drr_payloadlen = buflen;
1728
1729                         err = dump_record(&drr, packbuf, buflen, &zc, outfd);
1730                         free(packbuf);
1731                         if (err != 0)
1732                                 goto stderr_out;
1733
1734                         /* write end record */
1735                         bzero(&drr, sizeof (drr));
1736                         drr.drr_type = DRR_END;
1737                         drr.drr_u.drr_end.drr_checksum = zc;
1738                         err = write(outfd, &drr, sizeof (drr));
1739                         if (err == -1) {
1740                                 err = errno;
1741                                 goto stderr_out;
1742                         }
1743
1744                         err = 0;
1745                 }
1746         }
1747
1748         /* dump each stream */
1749         sdd.fromsnap = fromsnap;
1750         sdd.tosnap = tosnap;
1751         if (tid != 0)
1752                 sdd.outfd = pipefd[0];
1753         else
1754                 sdd.outfd = outfd;
1755         sdd.replicate = flags->replicate;
1756         sdd.doall = flags->doall;
1757         sdd.fromorigin = flags->fromorigin;
1758         sdd.fss = fss;
1759         sdd.fsavl = fsavl;
1760         sdd.verbose = flags->verbose;
1761         sdd.parsable = flags->parsable;
1762         sdd.progress = flags->progress;
1763         sdd.dryrun = flags->dryrun;
1764         sdd.large_block = flags->largeblock;
1765         sdd.embed_data = flags->embed_data;
1766         sdd.filter_cb = filter_func;
1767         sdd.filter_cb_arg = cb_arg;
1768         if (debugnvp)
1769                 sdd.debugnv = *debugnvp;
1770         if (sdd.verbose && sdd.dryrun)
1771                 sdd.std_out = B_TRUE;
1772         fout = sdd.std_out ? stdout : stderr;
1773
1774         /*
1775          * Some flags require that we place user holds on the datasets that are
1776          * being sent so they don't get destroyed during the send. We can skip
1777          * this step if the pool is imported read-only since the datasets cannot
1778          * be destroyed.
1779          */
1780         if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1781             ZPOOL_PROP_READONLY, NULL) &&
1782             zfs_spa_version(zhp, &spa_version) == 0 &&
1783             spa_version >= SPA_VERSION_USERREFS &&
1784             (flags->doall || flags->replicate)) {
1785                 ++holdseq;
1786                 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1787                     ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1788                 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1789                 if (sdd.cleanup_fd < 0) {
1790                         err = errno;
1791                         goto stderr_out;
1792                 }
1793                 sdd.snapholds = fnvlist_alloc();
1794         } else {
1795                 sdd.cleanup_fd = -1;
1796                 sdd.snapholds = NULL;
1797         }
1798         if (flags->verbose || sdd.snapholds != NULL) {
1799                 /*
1800                  * Do a verbose no-op dry run to get all the verbose output
1801                  * or to gather snapshot hold's before generating any data,
1802                  * then do a non-verbose real run to generate the streams.
1803                  */
1804                 sdd.dryrun = B_TRUE;
1805                 err = dump_filesystems(zhp, &sdd);
1806
1807                 if (err != 0)
1808                         goto stderr_out;
1809
1810                 if (flags->verbose) {
1811                         if (flags->parsable) {
1812                                 (void) fprintf(fout, "size\t%llu\n",
1813                                     (longlong_t)sdd.size);
1814                         } else {
1815                                 char buf[16];
1816                                 zfs_nicenum(sdd.size, buf, sizeof (buf));
1817                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1818                                     "total estimated size is %s\n"), buf);
1819                         }
1820                 }
1821
1822                 /* Ensure no snaps found is treated as an error. */
1823                 if (!sdd.seento) {
1824                         err = ENOENT;
1825                         goto err_out;
1826                 }
1827
1828                 /* Skip the second run if dryrun was requested. */
1829                 if (flags->dryrun)
1830                         goto err_out;
1831
1832                 if (sdd.snapholds != NULL) {
1833                         err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
1834                         if (err != 0)
1835                                 goto stderr_out;
1836
1837                         fnvlist_free(sdd.snapholds);
1838                         sdd.snapholds = NULL;
1839                 }
1840
1841                 sdd.dryrun = B_FALSE;
1842                 sdd.verbose = B_FALSE;
1843         }
1844
1845         err = dump_filesystems(zhp, &sdd);
1846         fsavl_destroy(fsavl);
1847         nvlist_free(fss);
1848
1849         /* Ensure no snaps found is treated as an error. */
1850         if (err == 0 && !sdd.seento)
1851                 err = ENOENT;
1852
1853         if (tid != 0) {
1854                 if (err != 0)
1855                         (void) pthread_cancel(tid);
1856                 (void) close(pipefd[0]);
1857                 (void) pthread_join(tid, NULL);
1858         }
1859
1860         if (sdd.cleanup_fd != -1) {
1861                 VERIFY(0 == close(sdd.cleanup_fd));
1862                 sdd.cleanup_fd = -1;
1863         }
1864
1865         if (!flags->dryrun && (flags->replicate || flags->doall ||
1866             flags->props)) {
1867                 /*
1868                  * write final end record.  NB: want to do this even if
1869                  * there was some error, because it might not be totally
1870                  * failed.
1871                  */
1872                 dmu_replay_record_t drr = { 0 };
1873                 drr.drr_type = DRR_END;
1874                 if (write(outfd, &drr, sizeof (drr)) == -1) {
1875                         return (zfs_standard_error(zhp->zfs_hdl,
1876                             errno, errbuf));
1877                 }
1878         }
1879
1880         return (err || sdd.err);
1881
1882 stderr_out:
1883         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1884 err_out:
1885         fsavl_destroy(fsavl);
1886         nvlist_free(fss);
1887         fnvlist_free(sdd.snapholds);
1888
1889         if (sdd.cleanup_fd != -1)
1890                 VERIFY(0 == close(sdd.cleanup_fd));
1891         if (tid != 0) {
1892                 (void) pthread_cancel(tid);
1893                 (void) close(pipefd[0]);
1894                 (void) pthread_join(tid, NULL);
1895         }
1896         return (err);
1897 }
1898
1899 int
1900 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd,
1901     enum lzc_send_flags flags)
1902 {
1903         int err;
1904         libzfs_handle_t *hdl = zhp->zfs_hdl;
1905
1906         char errbuf[1024];
1907         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1908             "warning: cannot send '%s'"), zhp->zfs_name);
1909
1910         err = lzc_send(zhp->zfs_name, from, fd, flags);
1911         if (err != 0) {
1912                 switch (errno) {
1913                 case EXDEV:
1914                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1915                             "not an earlier snapshot from the same fs"));
1916                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
1917
1918                 case ENOENT:
1919                 case ESRCH:
1920                         if (lzc_exists(zhp->zfs_name)) {
1921                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1922                                     "incremental source (%s) does not exist"),
1923                                     from);
1924                         }
1925                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
1926
1927                 case EBUSY:
1928                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1929                             "target is busy; if a filesystem, "
1930                             "it must not be mounted"));
1931                         return (zfs_error(hdl, EZFS_BUSY, errbuf));
1932
1933                 case EDQUOT:
1934                 case EFBIG:
1935                 case EIO:
1936                 case ENOLINK:
1937                 case ENOSPC:
1938 #ifdef illumos
1939                 case ENOSTR:
1940 #endif
1941                 case ENXIO:
1942                 case EPIPE:
1943                 case ERANGE:
1944                 case EFAULT:
1945                 case EROFS:
1946                         zfs_error_aux(hdl, strerror(errno));
1947                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1948
1949                 default:
1950                         return (zfs_standard_error(hdl, errno, errbuf));
1951                 }
1952         }
1953         return (err != 0);
1954 }
1955
1956 /*
1957  * Routines specific to "zfs recv"
1958  */
1959
1960 static int
1961 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1962     boolean_t byteswap, zio_cksum_t *zc)
1963 {
1964         char *cp = buf;
1965         int rv;
1966         int len = ilen;
1967
1968         assert(ilen <= SPA_MAXBLOCKSIZE);
1969
1970         do {
1971                 rv = read(fd, cp, len);
1972                 cp += rv;
1973                 len -= rv;
1974         } while (rv > 0);
1975
1976         if (rv < 0 || len != 0) {
1977                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1978                     "failed to read from stream"));
1979                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1980                     "cannot receive")));
1981         }
1982
1983         if (zc) {
1984                 if (byteswap)
1985                         fletcher_4_incremental_byteswap(buf, ilen, zc);
1986                 else
1987                         fletcher_4_incremental_native(buf, ilen, zc);
1988         }
1989         return (0);
1990 }
1991
1992 static int
1993 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1994     boolean_t byteswap, zio_cksum_t *zc)
1995 {
1996         char *buf;
1997         int err;
1998
1999         buf = zfs_alloc(hdl, len);
2000         if (buf == NULL)
2001                 return (ENOMEM);
2002
2003         err = recv_read(hdl, fd, buf, len, byteswap, zc);
2004         if (err != 0) {
2005                 free(buf);
2006                 return (err);
2007         }
2008
2009         err = nvlist_unpack(buf, len, nvp, 0);
2010         free(buf);
2011         if (err != 0) {
2012                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2013                     "stream (malformed nvlist)"));
2014                 return (EINVAL);
2015         }
2016         return (0);
2017 }
2018
2019 static int
2020 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
2021     int baselen, char *newname, recvflags_t *flags)
2022 {
2023         static int seq;
2024         zfs_cmd_t zc = { 0 };
2025         int err;
2026         prop_changelist_t *clp;
2027         zfs_handle_t *zhp;
2028
2029         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2030         if (zhp == NULL)
2031                 return (-1);
2032         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2033             flags->force ? MS_FORCE : 0);
2034         zfs_close(zhp);
2035         if (clp == NULL)
2036                 return (-1);
2037         err = changelist_prefix(clp);
2038         if (err)
2039                 return (err);
2040
2041         zc.zc_objset_type = DMU_OST_ZFS;
2042         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
2043
2044         if (tryname) {
2045                 (void) strcpy(newname, tryname);
2046
2047                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
2048
2049                 if (flags->verbose) {
2050                         (void) printf("attempting rename %s to %s\n",
2051                             zc.zc_name, zc.zc_value);
2052                 }
2053                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
2054                 if (err == 0)
2055                         changelist_rename(clp, name, tryname);
2056         } else {
2057                 err = ENOENT;
2058         }
2059
2060         if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
2061                 seq++;
2062
2063                 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u",
2064                     baselen, name, getpid(), seq);
2065                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
2066
2067                 if (flags->verbose) {
2068                         (void) printf("failed - trying rename %s to %s\n",
2069                             zc.zc_name, zc.zc_value);
2070                 }
2071                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
2072                 if (err == 0)
2073                         changelist_rename(clp, name, newname);
2074                 if (err && flags->verbose) {
2075                         (void) printf("failed (%u) - "
2076                             "will try again on next pass\n", errno);
2077                 }
2078                 err = EAGAIN;
2079         } else if (flags->verbose) {
2080                 if (err == 0)
2081                         (void) printf("success\n");
2082                 else
2083                         (void) printf("failed (%u)\n", errno);
2084         }
2085
2086         (void) changelist_postfix(clp);
2087         changelist_free(clp);
2088
2089         return (err);
2090 }
2091
2092 static int
2093 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
2094     char *newname, recvflags_t *flags)
2095 {
2096         zfs_cmd_t zc = { 0 };
2097         int err = 0;
2098         prop_changelist_t *clp;
2099         zfs_handle_t *zhp;
2100         boolean_t defer = B_FALSE;
2101         int spa_version;
2102
2103         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2104         if (zhp == NULL)
2105                 return (-1);
2106         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2107             flags->force ? MS_FORCE : 0);
2108         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
2109             zfs_spa_version(zhp, &spa_version) == 0 &&
2110             spa_version >= SPA_VERSION_USERREFS)
2111                 defer = B_TRUE;
2112         zfs_close(zhp);
2113         if (clp == NULL)
2114                 return (-1);
2115         err = changelist_prefix(clp);
2116         if (err)
2117                 return (err);
2118
2119         zc.zc_objset_type = DMU_OST_ZFS;
2120         zc.zc_defer_destroy = defer;
2121         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
2122
2123         if (flags->verbose)
2124                 (void) printf("attempting destroy %s\n", zc.zc_name);
2125         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
2126         if (err == 0) {
2127                 if (flags->verbose)
2128                         (void) printf("success\n");
2129                 changelist_remove(clp, zc.zc_name);
2130         }
2131
2132         (void) changelist_postfix(clp);
2133         changelist_free(clp);
2134
2135         /*
2136          * Deferred destroy might destroy the snapshot or only mark it to be
2137          * destroyed later, and it returns success in either case.
2138          */
2139         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
2140             ZFS_TYPE_SNAPSHOT))) {
2141                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
2142         }
2143
2144         return (err);
2145 }
2146
2147 typedef struct guid_to_name_data {
2148         uint64_t guid;
2149         boolean_t bookmark_ok;
2150         char *name;
2151         char *skip;
2152 } guid_to_name_data_t;
2153
2154 static int
2155 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
2156 {
2157         guid_to_name_data_t *gtnd = arg;
2158         const char *slash;
2159         int err;
2160
2161         if (gtnd->skip != NULL &&
2162             (slash = strrchr(zhp->zfs_name, '/')) != NULL &&
2163             strcmp(slash + 1, gtnd->skip) == 0) {
2164                 zfs_close(zhp);
2165                 return (0);
2166         }
2167
2168         if (zfs_prop_get_int(zhp, ZFS_PROP_GUID) == gtnd->guid) {
2169                 (void) strcpy(gtnd->name, zhp->zfs_name);
2170                 zfs_close(zhp);
2171                 return (EEXIST);
2172         }
2173
2174         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
2175         if (err != EEXIST && gtnd->bookmark_ok)
2176                 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, gtnd);
2177         zfs_close(zhp);
2178         return (err);
2179 }
2180
2181 /*
2182  * Attempt to find the local dataset associated with this guid.  In the case of
2183  * multiple matches, we attempt to find the "best" match by searching
2184  * progressively larger portions of the hierarchy.  This allows one to send a
2185  * tree of datasets individually and guarantee that we will find the source
2186  * guid within that hierarchy, even if there are multiple matches elsewhere.
2187  */
2188 static int
2189 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
2190     boolean_t bookmark_ok, char *name)
2191 {
2192         char pname[ZFS_MAXNAMELEN];
2193         guid_to_name_data_t gtnd;
2194
2195         gtnd.guid = guid;
2196         gtnd.bookmark_ok = bookmark_ok;
2197         gtnd.name = name;
2198         gtnd.skip = NULL;
2199
2200         /*
2201          * Search progressively larger portions of the hierarchy, starting
2202          * with the filesystem specified by 'parent'.  This will
2203          * select the "most local" version of the origin snapshot in the case
2204          * that there are multiple matching snapshots in the system.
2205          */
2206         (void) strlcpy(pname, parent, sizeof (pname));
2207         char *cp = strrchr(pname, '@');
2208         if (cp == NULL)
2209                 cp = strchr(pname, '\0');
2210         for (; cp != NULL; cp = strrchr(pname, '/')) {
2211                 /* Chop off the last component and open the parent */
2212                 *cp = '\0';
2213                 zfs_handle_t *zhp = make_dataset_handle(hdl, pname);
2214
2215                 if (zhp == NULL)
2216                         continue;
2217                 int err = guid_to_name_cb(zfs_handle_dup(zhp), &gtnd);
2218                 if (err != EEXIST)
2219                         err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
2220                 if (err != EEXIST && bookmark_ok)
2221                         err = zfs_iter_bookmarks(zhp, guid_to_name_cb, &gtnd);
2222                 zfs_close(zhp);
2223                 if (err == EEXIST)
2224                         return (0);
2225
2226                 /*
2227                  * Remember the last portion of the dataset so we skip it next
2228                  * time through (as we've already searched that portion of the
2229                  * hierarchy).
2230                  */
2231                 gtnd.skip = strrchr(pname, '/') + 1;
2232         }
2233
2234         return (ENOENT);
2235 }
2236
2237 /*
2238  * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
2239  * guid1 is after guid2.
2240  */
2241 static int
2242 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
2243     uint64_t guid1, uint64_t guid2)
2244 {
2245         nvlist_t *nvfs;
2246         char *fsname, *snapname;
2247         char buf[ZFS_MAXNAMELEN];
2248         int rv;
2249         zfs_handle_t *guid1hdl, *guid2hdl;
2250         uint64_t create1, create2;
2251
2252         if (guid2 == 0)
2253                 return (0);
2254         if (guid1 == 0)
2255                 return (1);
2256
2257         nvfs = fsavl_find(avl, guid1, &snapname);
2258         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2259         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2260         guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2261         if (guid1hdl == NULL)
2262                 return (-1);
2263
2264         nvfs = fsavl_find(avl, guid2, &snapname);
2265         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2266         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2267         guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2268         if (guid2hdl == NULL) {
2269                 zfs_close(guid1hdl);
2270                 return (-1);
2271         }
2272
2273         create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
2274         create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
2275
2276         if (create1 < create2)
2277                 rv = -1;
2278         else if (create1 > create2)
2279                 rv = +1;
2280         else
2281                 rv = 0;
2282
2283         zfs_close(guid1hdl);
2284         zfs_close(guid2hdl);
2285
2286         return (rv);
2287 }
2288
2289 static int
2290 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2291     recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2292     nvlist_t *renamed)
2293 {
2294         nvlist_t *local_nv, *deleted = NULL;
2295         avl_tree_t *local_avl;
2296         nvpair_t *fselem, *nextfselem;
2297         char *fromsnap;
2298         char newname[ZFS_MAXNAMELEN];
2299         char guidname[32];
2300         int error;
2301         boolean_t needagain, progress, recursive;
2302         char *s1, *s2;
2303
2304         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2305
2306         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2307             ENOENT);
2308
2309         if (flags->dryrun)
2310                 return (0);
2311
2312 again:
2313         needagain = progress = B_FALSE;
2314
2315         VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0));
2316
2317         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2318             recursive, &local_nv, &local_avl)) != 0)
2319                 return (error);
2320
2321         /*
2322          * Process deletes and renames
2323          */
2324         for (fselem = nvlist_next_nvpair(local_nv, NULL);
2325             fselem; fselem = nextfselem) {
2326                 nvlist_t *nvfs, *snaps;
2327                 nvlist_t *stream_nvfs = NULL;
2328                 nvpair_t *snapelem, *nextsnapelem;
2329                 uint64_t fromguid = 0;
2330                 uint64_t originguid = 0;
2331                 uint64_t stream_originguid = 0;
2332                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2333                 char *fsname, *stream_fsname;
2334
2335                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2336
2337                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2338                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2339                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2340                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2341                     &parent_fromsnap_guid));
2342                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2343
2344                 /*
2345                  * First find the stream's fs, so we can check for
2346                  * a different origin (due to "zfs promote")
2347                  */
2348                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2349                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2350                         uint64_t thisguid;
2351
2352                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2353                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2354
2355                         if (stream_nvfs != NULL)
2356                                 break;
2357                 }
2358
2359                 /* check for promote */
2360                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2361                     &stream_originguid);
2362                 if (stream_nvfs && originguid != stream_originguid) {
2363                         switch (created_before(hdl, local_avl,
2364                             stream_originguid, originguid)) {
2365                         case 1: {
2366                                 /* promote it! */
2367                                 zfs_cmd_t zc = { 0 };
2368                                 nvlist_t *origin_nvfs;
2369                                 char *origin_fsname;
2370
2371                                 if (flags->verbose)
2372                                         (void) printf("promoting %s\n", fsname);
2373
2374                                 origin_nvfs = fsavl_find(local_avl, originguid,
2375                                     NULL);
2376                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2377                                     "name", &origin_fsname));
2378                                 (void) strlcpy(zc.zc_value, origin_fsname,
2379                                     sizeof (zc.zc_value));
2380                                 (void) strlcpy(zc.zc_name, fsname,
2381                                     sizeof (zc.zc_name));
2382                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2383                                 if (error == 0)
2384                                         progress = B_TRUE;
2385                                 break;
2386                         }
2387                         default:
2388                                 break;
2389                         case -1:
2390                                 fsavl_destroy(local_avl);
2391                                 nvlist_free(local_nv);
2392                                 return (-1);
2393                         }
2394                         /*
2395                          * We had/have the wrong origin, therefore our
2396                          * list of snapshots is wrong.  Need to handle
2397                          * them on the next pass.
2398                          */
2399                         needagain = B_TRUE;
2400                         continue;
2401                 }
2402
2403                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2404                     snapelem; snapelem = nextsnapelem) {
2405                         uint64_t thisguid;
2406                         char *stream_snapname;
2407                         nvlist_t *found, *props;
2408
2409                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2410
2411                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2412                         found = fsavl_find(stream_avl, thisguid,
2413                             &stream_snapname);
2414
2415                         /* check for delete */
2416                         if (found == NULL) {
2417                                 char name[ZFS_MAXNAMELEN];
2418
2419                                 if (!flags->force)
2420                                         continue;
2421
2422                                 (void) snprintf(name, sizeof (name), "%s@%s",
2423                                     fsname, nvpair_name(snapelem));
2424
2425                                 error = recv_destroy(hdl, name,
2426                                     strlen(fsname)+1, newname, flags);
2427                                 if (error)
2428                                         needagain = B_TRUE;
2429                                 else
2430                                         progress = B_TRUE;
2431                                 sprintf(guidname, "%lu", thisguid);
2432                                 nvlist_add_boolean(deleted, guidname);
2433                                 continue;
2434                         }
2435
2436                         stream_nvfs = found;
2437
2438                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2439                             &props) && 0 == nvlist_lookup_nvlist(props,
2440                             stream_snapname, &props)) {
2441                                 zfs_cmd_t zc = { 0 };
2442
2443                                 zc.zc_cookie = B_TRUE; /* received */
2444                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2445                                     "%s@%s", fsname, nvpair_name(snapelem));
2446                                 if (zcmd_write_src_nvlist(hdl, &zc,
2447                                     props) == 0) {
2448                                         (void) zfs_ioctl(hdl,
2449                                             ZFS_IOC_SET_PROP, &zc);
2450                                         zcmd_free_nvlists(&zc);
2451                                 }
2452                         }
2453
2454                         /* check for different snapname */
2455                         if (strcmp(nvpair_name(snapelem),
2456                             stream_snapname) != 0) {
2457                                 char name[ZFS_MAXNAMELEN];
2458                                 char tryname[ZFS_MAXNAMELEN];
2459
2460                                 (void) snprintf(name, sizeof (name), "%s@%s",
2461                                     fsname, nvpair_name(snapelem));
2462                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
2463                                     fsname, stream_snapname);
2464
2465                                 error = recv_rename(hdl, name, tryname,
2466                                     strlen(fsname)+1, newname, flags);
2467                                 if (error)
2468                                         needagain = B_TRUE;
2469                                 else
2470                                         progress = B_TRUE;
2471                         }
2472
2473                         if (strcmp(stream_snapname, fromsnap) == 0)
2474                                 fromguid = thisguid;
2475                 }
2476
2477                 /* check for delete */
2478                 if (stream_nvfs == NULL) {
2479                         if (!flags->force)
2480                                 continue;
2481
2482                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2483                             newname, flags);
2484                         if (error)
2485                                 needagain = B_TRUE;
2486                         else
2487                                 progress = B_TRUE;
2488                         sprintf(guidname, "%lu", parent_fromsnap_guid);
2489                         nvlist_add_boolean(deleted, guidname);
2490                         continue;
2491                 }
2492
2493                 if (fromguid == 0) {
2494                         if (flags->verbose) {
2495                                 (void) printf("local fs %s does not have "
2496                                     "fromsnap (%s in stream); must have "
2497                                     "been deleted locally; ignoring\n",
2498                                     fsname, fromsnap);
2499                         }
2500                         continue;
2501                 }
2502
2503                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2504                     "name", &stream_fsname));
2505                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2506                     "parentfromsnap", &stream_parent_fromsnap_guid));
2507
2508                 s1 = strrchr(fsname, '/');
2509                 s2 = strrchr(stream_fsname, '/');
2510
2511                 /*
2512                  * Check if we're going to rename based on parent guid change
2513                  * and the current parent guid was also deleted. If it was then
2514                  * rename will fail and is likely unneeded, so avoid this and
2515                  * force an early retry to determine the new
2516                  * parent_fromsnap_guid.
2517                  */
2518                 if (stream_parent_fromsnap_guid != 0 &&
2519                     parent_fromsnap_guid != 0 &&
2520                     stream_parent_fromsnap_guid != parent_fromsnap_guid) {
2521                         sprintf(guidname, "%lu", parent_fromsnap_guid);
2522                         if (nvlist_exists(deleted, guidname)) {
2523                                 progress = B_TRUE;
2524                                 needagain = B_TRUE;
2525                                 goto doagain;
2526                         }
2527                 }
2528
2529                 /*
2530                  * Check for rename. If the exact receive path is specified, it
2531                  * does not count as a rename, but we still need to check the
2532                  * datasets beneath it.
2533                  */
2534                 if ((stream_parent_fromsnap_guid != 0 &&
2535                     parent_fromsnap_guid != 0 &&
2536                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2537                     ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2538                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2539                         nvlist_t *parent;
2540                         char tryname[ZFS_MAXNAMELEN];
2541
2542                         parent = fsavl_find(local_avl,
2543                             stream_parent_fromsnap_guid, NULL);
2544                         /*
2545                          * NB: parent might not be found if we used the
2546                          * tosnap for stream_parent_fromsnap_guid,
2547                          * because the parent is a newly-created fs;
2548                          * we'll be able to rename it after we recv the
2549                          * new fs.
2550                          */
2551                         if (parent != NULL) {
2552                                 char *pname;
2553
2554                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
2555                                     &pname));
2556                                 (void) snprintf(tryname, sizeof (tryname),
2557                                     "%s%s", pname, strrchr(stream_fsname, '/'));
2558                         } else {
2559                                 tryname[0] = '\0';
2560                                 if (flags->verbose) {
2561                                         (void) printf("local fs %s new parent "
2562                                             "not found\n", fsname);
2563                                 }
2564                         }
2565
2566                         newname[0] = '\0';
2567
2568                         error = recv_rename(hdl, fsname, tryname,
2569                             strlen(tofs)+1, newname, flags);
2570
2571                         if (renamed != NULL && newname[0] != '\0') {
2572                                 VERIFY(0 == nvlist_add_boolean(renamed,
2573                                     newname));
2574                         }
2575
2576                         if (error)
2577                                 needagain = B_TRUE;
2578                         else
2579                                 progress = B_TRUE;
2580                 }
2581         }
2582
2583 doagain:
2584         fsavl_destroy(local_avl);
2585         nvlist_free(local_nv);
2586         nvlist_free(deleted);
2587
2588         if (needagain && progress) {
2589                 /* do another pass to fix up temporary names */
2590                 if (flags->verbose)
2591                         (void) printf("another pass:\n");
2592                 goto again;
2593         }
2594
2595         return (needagain);
2596 }
2597
2598 static int
2599 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2600     recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2601     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2602 {
2603         nvlist_t *stream_nv = NULL;
2604         avl_tree_t *stream_avl = NULL;
2605         char *fromsnap = NULL;
2606         char *sendsnap = NULL;
2607         char *cp;
2608         char tofs[ZFS_MAXNAMELEN];
2609         char sendfs[ZFS_MAXNAMELEN];
2610         char errbuf[1024];
2611         dmu_replay_record_t drre;
2612         int error;
2613         boolean_t anyerr = B_FALSE;
2614         boolean_t softerr = B_FALSE;
2615         boolean_t recursive;
2616
2617         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2618             "cannot receive"));
2619
2620         assert(drr->drr_type == DRR_BEGIN);
2621         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2622         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2623             DMU_COMPOUNDSTREAM);
2624
2625         /*
2626          * Read in the nvlist from the stream.
2627          */
2628         if (drr->drr_payloadlen != 0) {
2629                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2630                     &stream_nv, flags->byteswap, zc);
2631                 if (error) {
2632                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2633                         goto out;
2634                 }
2635         }
2636
2637         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2638             ENOENT);
2639
2640         if (recursive && strchr(destname, '@')) {
2641                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2642                     "cannot specify snapshot name for multi-snapshot stream"));
2643                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2644                 goto out;
2645         }
2646
2647         /*
2648          * Read in the end record and verify checksum.
2649          */
2650         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2651             flags->byteswap, NULL)))
2652                 goto out;
2653         if (flags->byteswap) {
2654                 drre.drr_type = BSWAP_32(drre.drr_type);
2655                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2656                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2657                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2658                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2659                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2660                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2661                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2662                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2663         }
2664         if (drre.drr_type != DRR_END) {
2665                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2666                 goto out;
2667         }
2668         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2669                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2670                     "incorrect header checksum"));
2671                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2672                 goto out;
2673         }
2674
2675         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2676
2677         if (drr->drr_payloadlen != 0) {
2678                 nvlist_t *stream_fss;
2679
2680                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2681                     &stream_fss));
2682                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2683                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2684                             "couldn't allocate avl tree"));
2685                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2686                         goto out;
2687                 }
2688
2689                 if (fromsnap != NULL) {
2690                         nvlist_t *renamed = NULL;
2691                         nvpair_t *pair = NULL;
2692
2693                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2694                         if (flags->isprefix) {
2695                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2696                                 int i;
2697
2698                                 if (flags->istail) {
2699                                         cp = strrchr(drrb->drr_toname, '/');
2700                                         if (cp == NULL) {
2701                                                 (void) strlcat(tofs, "/",
2702                                                     ZFS_MAXNAMELEN);
2703                                                 i = 0;
2704                                         } else {
2705                                                 i = (cp - drrb->drr_toname);
2706                                         }
2707                                 } else {
2708                                         i = strcspn(drrb->drr_toname, "/@");
2709                                 }
2710                                 /* zfs_receive_one() will create_parents() */
2711                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2712                                     ZFS_MAXNAMELEN);
2713                                 *strchr(tofs, '@') = '\0';
2714                         }
2715
2716                         if (recursive && !flags->dryrun && !flags->nomount) {
2717                                 VERIFY(0 == nvlist_alloc(&renamed,
2718                                     NV_UNIQUE_NAME, 0));
2719                         }
2720
2721                         softerr = recv_incremental_replication(hdl, tofs, flags,
2722                             stream_nv, stream_avl, renamed);
2723
2724                         /* Unmount renamed filesystems before receiving. */
2725                         while ((pair = nvlist_next_nvpair(renamed,
2726                             pair)) != NULL) {
2727                                 zfs_handle_t *zhp;
2728                                 prop_changelist_t *clp = NULL;
2729
2730                                 zhp = zfs_open(hdl, nvpair_name(pair),
2731                                     ZFS_TYPE_FILESYSTEM);
2732                                 if (zhp != NULL) {
2733                                         clp = changelist_gather(zhp,
2734                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2735                                         zfs_close(zhp);
2736                                         if (clp != NULL) {
2737                                                 softerr |=
2738                                                     changelist_prefix(clp);
2739                                                 changelist_free(clp);
2740                                         }
2741                                 }
2742                         }
2743
2744                         nvlist_free(renamed);
2745                 }
2746         }
2747
2748         /*
2749          * Get the fs specified by the first path in the stream (the top level
2750          * specified by 'zfs send') and pass it to each invocation of
2751          * zfs_receive_one().
2752          */
2753         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2754             ZFS_MAXNAMELEN);
2755         if ((cp = strchr(sendfs, '@')) != NULL) {
2756                 *cp = '\0';
2757                 /*
2758                  * Find the "sendsnap", the final snapshot in a replication
2759                  * stream.  zfs_receive_one() handles certain errors
2760                  * differently, depending on if the contained stream is the
2761                  * last one or not.
2762                  */
2763                 sendsnap = (cp + 1);
2764         }
2765
2766         /* Finally, receive each contained stream */
2767         do {
2768                 /*
2769                  * we should figure out if it has a recoverable
2770                  * error, in which case do a recv_skip() and drive on.
2771                  * Note, if we fail due to already having this guid,
2772                  * zfs_receive_one() will take care of it (ie,
2773                  * recv_skip() and return 0).
2774                  */
2775                 error = zfs_receive_impl(hdl, destname, NULL, flags, fd,
2776                     sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2777                     action_handlep, sendsnap);
2778                 if (error == ENODATA) {
2779                         error = 0;
2780                         break;
2781                 }
2782                 anyerr |= error;
2783         } while (error == 0);
2784
2785         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2786                 /*
2787                  * Now that we have the fs's they sent us, try the
2788                  * renames again.
2789                  */
2790                 softerr = recv_incremental_replication(hdl, tofs, flags,
2791                     stream_nv, stream_avl, NULL);
2792         }
2793
2794 out:
2795         fsavl_destroy(stream_avl);
2796         nvlist_free(stream_nv);
2797         if (softerr)
2798                 error = -2;
2799         if (anyerr)
2800                 error = -1;
2801         return (error);
2802 }
2803
2804 static void
2805 trunc_prop_errs(int truncated)
2806 {
2807         ASSERT(truncated != 0);
2808
2809         if (truncated == 1)
2810                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2811                     "1 more property could not be set\n"));
2812         else
2813                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2814                     "%d more properties could not be set\n"), truncated);
2815 }
2816
2817 static int
2818 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2819 {
2820         dmu_replay_record_t *drr;
2821         void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
2822         char errbuf[1024];
2823
2824         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2825             "cannot receive:"));
2826
2827         /* XXX would be great to use lseek if possible... */
2828         drr = buf;
2829
2830         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2831             byteswap, NULL) == 0) {
2832                 if (byteswap)
2833                         drr->drr_type = BSWAP_32(drr->drr_type);
2834
2835                 switch (drr->drr_type) {
2836                 case DRR_BEGIN:
2837                         if (drr->drr_payloadlen != 0) {
2838                                 (void) recv_read(hdl, fd, buf,
2839                                     drr->drr_payloadlen, B_FALSE, NULL);
2840                         }
2841                         break;
2842
2843                 case DRR_END:
2844                         free(buf);
2845                         return (0);
2846
2847                 case DRR_OBJECT:
2848                         if (byteswap) {
2849                                 drr->drr_u.drr_object.drr_bonuslen =
2850                                     BSWAP_32(drr->drr_u.drr_object.
2851                                     drr_bonuslen);
2852                         }
2853                         (void) recv_read(hdl, fd, buf,
2854                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2855                             B_FALSE, NULL);
2856                         break;
2857
2858                 case DRR_WRITE:
2859                         if (byteswap) {
2860                                 drr->drr_u.drr_write.drr_length =
2861                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
2862                         }
2863                         (void) recv_read(hdl, fd, buf,
2864                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2865                         break;
2866                 case DRR_SPILL:
2867                         if (byteswap) {
2868                                 drr->drr_u.drr_write.drr_length =
2869                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
2870                         }
2871                         (void) recv_read(hdl, fd, buf,
2872                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2873                         break;
2874                 case DRR_WRITE_EMBEDDED:
2875                         if (byteswap) {
2876                                 drr->drr_u.drr_write_embedded.drr_psize =
2877                                     BSWAP_32(drr->drr_u.drr_write_embedded.
2878                                     drr_psize);
2879                         }
2880                         (void) recv_read(hdl, fd, buf,
2881                             P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
2882                             8), B_FALSE, NULL);
2883                         break;
2884                 case DRR_WRITE_BYREF:
2885                 case DRR_FREEOBJECTS:
2886                 case DRR_FREE:
2887                         break;
2888
2889                 default:
2890                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2891                             "invalid record type"));
2892                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2893                 }
2894         }
2895
2896         free(buf);
2897         return (-1);
2898 }
2899
2900 static void
2901 recv_ecksum_set_aux(libzfs_handle_t *hdl, const char *target_snap,
2902     boolean_t resumable)
2903 {
2904         char target_fs[ZFS_MAXNAMELEN];
2905
2906         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2907             "checksum mismatch or incomplete stream"));
2908
2909         if (!resumable)
2910                 return;
2911         (void) strlcpy(target_fs, target_snap, sizeof (target_fs));
2912         *strchr(target_fs, '@') = '\0';
2913         zfs_handle_t *zhp = zfs_open(hdl, target_fs,
2914             ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
2915         if (zhp == NULL)
2916                 return;
2917
2918         char token_buf[ZFS_MAXPROPLEN];
2919         int error = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN,
2920             token_buf, sizeof (token_buf),
2921             NULL, NULL, 0, B_TRUE);
2922         if (error == 0) {
2923                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2924                     "checksum mismatch or incomplete stream.\n"
2925                     "Partially received snapshot is saved.\n"
2926                     "A resuming stream can be generated on the sending "
2927                     "system by running:\n"
2928                     "    zfs send -t %s"),
2929                     token_buf);
2930         }
2931         zfs_close(zhp);
2932 }
2933
2934 /*
2935  * Restores a backup of tosnap from the file descriptor specified by infd.
2936  */
2937 static int
2938 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2939     const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr,
2940     dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv,
2941     avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2942     uint64_t *action_handlep, const char *finalsnap)
2943 {
2944         zfs_cmd_t zc = { 0 };
2945         time_t begin_time;
2946         int ioctl_err, ioctl_errno, err;
2947         char *cp;
2948         struct drr_begin *drrb = &drr->drr_u.drr_begin;
2949         char errbuf[1024];
2950         char prop_errbuf[1024];
2951         const char *chopprefix;
2952         boolean_t newfs = B_FALSE;
2953         boolean_t stream_wantsnewfs;
2954         uint64_t parent_snapguid = 0;
2955         prop_changelist_t *clp = NULL;
2956         nvlist_t *snapprops_nvlist = NULL;
2957         zprop_errflags_t prop_errflags;
2958         boolean_t recursive;
2959         char *snapname = NULL;
2960
2961         begin_time = time(NULL);
2962
2963         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2964             "cannot receive"));
2965
2966         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2967             ENOENT);
2968
2969         if (stream_avl != NULL) {
2970                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2971                     &snapname);
2972                 nvlist_t *props;
2973                 int ret;
2974
2975                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2976                     &parent_snapguid);
2977                 err = nvlist_lookup_nvlist(fs, "props", &props);
2978                 if (err)
2979                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2980
2981                 if (flags->canmountoff) {
2982                         VERIFY(0 == nvlist_add_uint64(props,
2983                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2984                 }
2985                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2986                 if (err)
2987                         nvlist_free(props);
2988
2989                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2990                         VERIFY(0 == nvlist_lookup_nvlist(props,
2991                             snapname, &snapprops_nvlist));
2992                 }
2993
2994                 if (ret != 0)
2995                         return (-1);
2996         }
2997
2998         cp = NULL;
2999
3000         /*
3001          * Determine how much of the snapshot name stored in the stream
3002          * we are going to tack on to the name they specified on the
3003          * command line, and how much we are going to chop off.
3004          *
3005          * If they specified a snapshot, chop the entire name stored in
3006          * the stream.
3007          */
3008         if (flags->istail) {
3009                 /*
3010                  * A filesystem was specified with -e. We want to tack on only
3011                  * the tail of the sent snapshot path.
3012                  */
3013                 if (strchr(tosnap, '@')) {
3014                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3015                             "argument - snapshot not allowed with -e"));
3016                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3017                 }
3018
3019                 chopprefix = strrchr(sendfs, '/');
3020
3021                 if (chopprefix == NULL) {
3022                         /*
3023                          * The tail is the poolname, so we need to
3024                          * prepend a path separator.
3025                          */
3026                         int len = strlen(drrb->drr_toname);
3027                         cp = malloc(len + 2);
3028                         cp[0] = '/';
3029                         (void) strcpy(&cp[1], drrb->drr_toname);
3030                         chopprefix = cp;
3031                 } else {
3032                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
3033                 }
3034         } else if (flags->isprefix) {
3035                 /*
3036                  * A filesystem was specified with -d. We want to tack on
3037                  * everything but the first element of the sent snapshot path
3038                  * (all but the pool name).
3039                  */
3040                 if (strchr(tosnap, '@')) {
3041                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3042                             "argument - snapshot not allowed with -d"));
3043                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3044                 }
3045
3046                 chopprefix = strchr(drrb->drr_toname, '/');
3047                 if (chopprefix == NULL)
3048                         chopprefix = strchr(drrb->drr_toname, '@');
3049         } else if (strchr(tosnap, '@') == NULL) {
3050                 /*
3051                  * If a filesystem was specified without -d or -e, we want to
3052                  * tack on everything after the fs specified by 'zfs send'.
3053                  */
3054                 chopprefix = drrb->drr_toname + strlen(sendfs);
3055         } else {
3056                 /* A snapshot was specified as an exact path (no -d or -e). */
3057                 if (recursive) {
3058                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3059                             "cannot specify snapshot name for multi-snapshot "
3060                             "stream"));
3061                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3062                 }
3063                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
3064         }
3065
3066         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
3067         ASSERT(chopprefix > drrb->drr_toname);
3068         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
3069         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
3070             chopprefix[0] == '\0');
3071
3072         /*
3073          * Determine name of destination snapshot, store in zc_value.
3074          */
3075         (void) strcpy(zc.zc_value, tosnap);
3076         (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
3077 #ifdef __FreeBSD__
3078         if (zfs_ioctl_version == ZFS_IOCVER_UNDEF)
3079                 zfs_ioctl_version = get_zfs_ioctl_version();
3080         /*
3081          * For forward compatibility hide tosnap in zc_value
3082          */
3083         if (zfs_ioctl_version < ZFS_IOCVER_LZC)
3084                 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap);
3085 #endif
3086         free(cp);
3087         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
3088                 zcmd_free_nvlists(&zc);
3089                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3090         }
3091
3092         /*
3093          * Determine the name of the origin snapshot, store in zc_string.
3094          */
3095         if (drrb->drr_flags & DRR_FLAG_CLONE) {
3096                 if (guid_to_name(hdl, zc.zc_value,
3097                     drrb->drr_fromguid, B_FALSE, zc.zc_string) != 0) {
3098                         zcmd_free_nvlists(&zc);
3099                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3100                             "local origin for clone %s does not exist"),
3101                             zc.zc_value);
3102                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
3103                 }
3104                 if (flags->verbose)
3105                         (void) printf("found clone origin %s\n", zc.zc_string);
3106         } else if (originsnap) {
3107                 (void) strncpy(zc.zc_string, originsnap, ZFS_MAXNAMELEN);
3108                 if (flags->verbose)
3109                         (void) printf("using provided clone origin %s\n",
3110                             zc.zc_string);
3111         }
3112
3113         boolean_t resuming = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
3114             DMU_BACKUP_FEATURE_RESUMING;
3115         stream_wantsnewfs = (drrb->drr_fromguid == 0 ||
3116             (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && !resuming;
3117
3118         if (stream_wantsnewfs) {
3119                 /*
3120                  * if the parent fs does not exist, look for it based on
3121                  * the parent snap GUID
3122                  */
3123                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3124                     "cannot receive new filesystem stream"));
3125
3126                 (void) strcpy(zc.zc_name, zc.zc_value);
3127                 cp = strrchr(zc.zc_name, '/');
3128                 if (cp)
3129                         *cp = '\0';
3130                 if (cp &&
3131                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3132                         char suffix[ZFS_MAXNAMELEN];
3133                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
3134                         if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
3135                             B_FALSE, zc.zc_value) == 0) {
3136                                 *strchr(zc.zc_value, '@') = '\0';
3137                                 (void) strcat(zc.zc_value, suffix);
3138                         }
3139                 }
3140         } else {
3141                 /*
3142                  * if the fs does not exist, look for it based on the
3143                  * fromsnap GUID
3144                  */
3145                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3146                     "cannot receive incremental stream"));
3147
3148                 (void) strcpy(zc.zc_name, zc.zc_value);
3149                 *strchr(zc.zc_name, '@') = '\0';
3150
3151                 /*
3152                  * If the exact receive path was specified and this is the
3153                  * topmost path in the stream, then if the fs does not exist we
3154                  * should look no further.
3155                  */
3156                 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
3157                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
3158                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3159                         char snap[ZFS_MAXNAMELEN];
3160                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
3161                         if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
3162                             B_FALSE, zc.zc_value) == 0) {
3163                                 *strchr(zc.zc_value, '@') = '\0';
3164                                 (void) strcat(zc.zc_value, snap);
3165                         }
3166                 }
3167         }
3168
3169         (void) strcpy(zc.zc_name, zc.zc_value);
3170         *strchr(zc.zc_name, '@') = '\0';
3171
3172         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3173                 zfs_handle_t *zhp;
3174
3175                 /*
3176                  * Destination fs exists.  It must be one of these cases:
3177                  *  - an incremental send stream
3178                  *  - the stream specifies a new fs (full stream or clone)
3179                  *    and they want us to blow away the existing fs (and
3180                  *    have therefore specified -F and removed any snapshots)
3181                  *  - we are resuming a failed receive.
3182                  */
3183                 if (stream_wantsnewfs) {
3184                         if (!flags->force) {
3185                                 zcmd_free_nvlists(&zc);
3186                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3187                                     "destination '%s' exists\n"
3188                                     "must specify -F to overwrite it"),
3189                                     zc.zc_name);
3190                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3191                         }
3192                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
3193                             &zc) == 0) {
3194                                 zcmd_free_nvlists(&zc);
3195                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3196                                     "destination has snapshots (eg. %s)\n"
3197                                     "must destroy them to overwrite it"),
3198                                     zc.zc_name);
3199                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3200                         }
3201                 }
3202
3203                 if ((zhp = zfs_open(hdl, zc.zc_name,
3204                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
3205                         zcmd_free_nvlists(&zc);
3206                         return (-1);
3207                 }
3208
3209                 if (stream_wantsnewfs &&
3210                     zhp->zfs_dmustats.dds_origin[0]) {
3211                         zcmd_free_nvlists(&zc);
3212                         zfs_close(zhp);
3213                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3214                             "destination '%s' is a clone\n"
3215                             "must destroy it to overwrite it"),
3216                             zc.zc_name);
3217                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3218                 }
3219
3220                 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
3221                     stream_wantsnewfs) {
3222                         /* We can't do online recv in this case */
3223                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
3224                         if (clp == NULL) {
3225                                 zfs_close(zhp);
3226                                 zcmd_free_nvlists(&zc);
3227                                 return (-1);
3228                         }
3229                         if (changelist_prefix(clp) != 0) {
3230                                 changelist_free(clp);
3231                                 zfs_close(zhp);
3232                                 zcmd_free_nvlists(&zc);
3233                                 return (-1);
3234                         }
3235                 }
3236
3237                 /*
3238                  * If we are resuming a newfs, set newfs here so that we will
3239                  * mount it if the recv succeeds this time.  We can tell
3240                  * that it was a newfs on the first recv because the fs
3241                  * itself will be inconsistent (if the fs existed when we
3242                  * did the first recv, we would have received it into
3243                  * .../%recv).
3244                  */
3245                 if (resuming && zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT))
3246                         newfs = B_TRUE;
3247
3248                 zfs_close(zhp);
3249         } else {
3250                 /*
3251                  * Destination filesystem does not exist.  Therefore we better
3252                  * be creating a new filesystem (either from a full backup, or
3253                  * a clone).  It would therefore be invalid if the user
3254                  * specified only the pool name (i.e. if the destination name
3255                  * contained no slash character).
3256                  */
3257                 if (!stream_wantsnewfs ||
3258                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
3259                         zcmd_free_nvlists(&zc);
3260                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3261                             "destination '%s' does not exist"), zc.zc_name);
3262                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
3263                 }
3264
3265                 /*
3266                  * Trim off the final dataset component so we perform the
3267                  * recvbackup ioctl to the filesystems's parent.
3268                  */
3269                 *cp = '\0';
3270
3271                 if (flags->isprefix && !flags->istail && !flags->dryrun &&
3272                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
3273                         zcmd_free_nvlists(&zc);
3274                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
3275                 }
3276
3277                 newfs = B_TRUE;
3278         }
3279
3280         zc.zc_begin_record = *drr_noswap;
3281         zc.zc_cookie = infd;
3282         zc.zc_guid = flags->force;
3283         zc.zc_resumable = flags->resumable;
3284         if (flags->verbose) {
3285                 (void) printf("%s %s stream of %s into %s\n",
3286                     flags->dryrun ? "would receive" : "receiving",
3287                     drrb->drr_fromguid ? "incremental" : "full",
3288                     drrb->drr_toname, zc.zc_value);
3289                 (void) fflush(stdout);
3290         }
3291
3292         if (flags->dryrun) {
3293                 zcmd_free_nvlists(&zc);
3294                 return (recv_skip(hdl, infd, flags->byteswap));
3295         }
3296
3297         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
3298         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
3299         zc.zc_cleanup_fd = cleanup_fd;
3300         zc.zc_action_handle = *action_handlep;
3301
3302         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
3303         ioctl_errno = errno;
3304         prop_errflags = (zprop_errflags_t)zc.zc_obj;
3305
3306         if (err == 0) {
3307                 nvlist_t *prop_errors;
3308                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
3309                     zc.zc_nvlist_dst_size, &prop_errors, 0));
3310
3311                 nvpair_t *prop_err = NULL;
3312
3313                 while ((prop_err = nvlist_next_nvpair(prop_errors,
3314                     prop_err)) != NULL) {
3315                         char tbuf[1024];
3316                         zfs_prop_t prop;
3317                         int intval;
3318
3319                         prop = zfs_name_to_prop(nvpair_name(prop_err));
3320                         (void) nvpair_value_int32(prop_err, &intval);
3321                         if (strcmp(nvpair_name(prop_err),
3322                             ZPROP_N_MORE_ERRORS) == 0) {
3323                                 trunc_prop_errs(intval);
3324                                 break;
3325                         } else if (snapname == NULL || finalsnap == NULL ||
3326                             strcmp(finalsnap, snapname) == 0 ||
3327                             strcmp(nvpair_name(prop_err),
3328                             zfs_prop_to_name(ZFS_PROP_REFQUOTA)) != 0) {
3329                                 /*
3330                                  * Skip the special case of, for example,
3331                                  * "refquota", errors on intermediate
3332                                  * snapshots leading up to a final one.
3333                                  * That's why we have all of the checks above.
3334                                  *
3335                                  * See zfs_ioctl.c's extract_delay_props() for
3336                                  * a list of props which can fail on
3337                                  * intermediate snapshots, but shouldn't
3338                                  * affect the overall receive.
3339                                  */
3340                                 (void) snprintf(tbuf, sizeof (tbuf),
3341                                     dgettext(TEXT_DOMAIN,
3342                                     "cannot receive %s property on %s"),
3343                                     nvpair_name(prop_err), zc.zc_name);
3344                                 zfs_setprop_error(hdl, prop, intval, tbuf);
3345                         }
3346                 }
3347                 nvlist_free(prop_errors);
3348         }
3349
3350         zc.zc_nvlist_dst = 0;
3351         zc.zc_nvlist_dst_size = 0;
3352         zcmd_free_nvlists(&zc);
3353
3354         if (err == 0 && snapprops_nvlist) {
3355                 zfs_cmd_t zc2 = { 0 };
3356
3357                 (void) strcpy(zc2.zc_name, zc.zc_value);
3358                 zc2.zc_cookie = B_TRUE; /* received */
3359                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
3360                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
3361                         zcmd_free_nvlists(&zc2);
3362                 }
3363         }
3364
3365         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
3366                 /*
3367                  * It may be that this snapshot already exists,
3368                  * in which case we want to consume & ignore it
3369                  * rather than failing.
3370                  */
3371                 avl_tree_t *local_avl;
3372                 nvlist_t *local_nv, *fs;
3373                 cp = strchr(zc.zc_value, '@');
3374
3375                 /*
3376                  * XXX Do this faster by just iterating over snaps in
3377                  * this fs.  Also if zc_value does not exist, we will
3378                  * get a strange "does not exist" error message.
3379                  */
3380                 *cp = '\0';
3381                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3382                     &local_nv, &local_avl) == 0) {
3383                         *cp = '@';
3384                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3385                         fsavl_destroy(local_avl);
3386                         nvlist_free(local_nv);
3387
3388                         if (fs != NULL) {
3389                                 if (flags->verbose) {
3390                                         (void) printf("snap %s already exists; "
3391                                             "ignoring\n", zc.zc_value);
3392                                 }
3393                                 err = ioctl_err = recv_skip(hdl, infd,
3394                                     flags->byteswap);
3395                         }
3396                 }
3397                 *cp = '@';
3398         }
3399
3400         if (ioctl_err != 0) {
3401                 switch (ioctl_errno) {
3402                 case ENODEV:
3403                         cp = strchr(zc.zc_value, '@');
3404                         *cp = '\0';
3405                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3406                             "most recent snapshot of %s does not\n"
3407                             "match incremental source"), zc.zc_value);
3408                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3409                         *cp = '@';
3410                         break;
3411                 case ETXTBSY:
3412                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3413                             "destination %s has been modified\n"
3414                             "since most recent snapshot"), zc.zc_name);
3415                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3416                         break;
3417                 case EEXIST:
3418                         cp = strchr(zc.zc_value, '@');
3419                         if (newfs) {
3420                                 /* it's the containing fs that exists */
3421                                 *cp = '\0';
3422                         }
3423                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3424                             "destination already exists"));
3425                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3426                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3427                             zc.zc_value);
3428                         *cp = '@';
3429                         break;
3430                 case EINVAL:
3431                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3432                         break;
3433                 case ECKSUM:
3434                         recv_ecksum_set_aux(hdl, zc.zc_value, flags->resumable);
3435                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3436                         break;
3437                 case ENOTSUP:
3438                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3439                             "pool must be upgraded to receive this stream."));
3440                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3441                         break;
3442                 case EDQUOT:
3443                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3444                             "destination %s space quota exceeded"), zc.zc_name);
3445                         (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3446                         break;
3447                 default:
3448                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3449                 }
3450         }
3451
3452         /*
3453          * Mount the target filesystem (if created).  Also mount any
3454          * children of the target filesystem if we did a replication
3455          * receive (indicated by stream_avl being non-NULL).
3456          */
3457         cp = strchr(zc.zc_value, '@');
3458         if (cp && (ioctl_err == 0 || !newfs)) {
3459                 zfs_handle_t *h;
3460
3461                 *cp = '\0';
3462                 h = zfs_open(hdl, zc.zc_value,
3463                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3464                 if (h != NULL) {
3465                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
3466                                 *cp = '@';
3467                         } else if (newfs || stream_avl) {
3468                                 /*
3469                                  * Track the first/top of hierarchy fs,
3470                                  * for mounting and sharing later.
3471                                  */
3472                                 if (top_zfs && *top_zfs == NULL)
3473                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
3474                         }
3475                         zfs_close(h);
3476                 }
3477                 *cp = '@';
3478         }
3479
3480         if (clp) {
3481                 if (!flags->nomount)
3482                         err |= changelist_postfix(clp);
3483                 changelist_free(clp);
3484         }
3485
3486         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3487                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3488                     "failed to clear unreceived properties on %s"),
3489                     zc.zc_name);
3490                 (void) fprintf(stderr, "\n");
3491         }
3492         if (prop_errflags & ZPROP_ERR_NORESTORE) {
3493                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3494                     "failed to restore original properties on %s"),
3495                     zc.zc_name);
3496                 (void) fprintf(stderr, "\n");
3497         }
3498
3499         if (err || ioctl_err)
3500                 return (-1);
3501
3502         *action_handlep = zc.zc_action_handle;
3503
3504         if (flags->verbose) {
3505                 char buf1[64];
3506                 char buf2[64];
3507                 uint64_t bytes = zc.zc_cookie;
3508                 time_t delta = time(NULL) - begin_time;
3509                 if (delta == 0)
3510                         delta = 1;
3511                 zfs_nicenum(bytes, buf1, sizeof (buf1));
3512                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3513
3514                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3515                     buf1, delta, buf2);
3516         }
3517
3518         return (0);
3519 }
3520
3521 static int
3522 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap,
3523     const char *originsnap, recvflags_t *flags, int infd, const char *sendfs,
3524     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
3525     uint64_t *action_handlep, const char *finalsnap)
3526 {
3527         int err;
3528         dmu_replay_record_t drr, drr_noswap;
3529         struct drr_begin *drrb = &drr.drr_u.drr_begin;
3530         char errbuf[1024];
3531         zio_cksum_t zcksum = { 0 };
3532         uint64_t featureflags;
3533         int hdrtype;
3534
3535         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3536             "cannot receive"));
3537
3538         if (flags->isprefix &&
3539             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3540                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3541                     "(%s) does not exist"), tosnap);
3542                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3543         }
3544         if (originsnap &&
3545             !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) {
3546                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs "
3547                     "(%s) does not exist"), originsnap);
3548                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3549         }
3550
3551         /* read in the BEGIN record */
3552         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3553             &zcksum)))
3554                 return (err);
3555
3556         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3557                 /* It's the double end record at the end of a package */
3558                 return (ENODATA);
3559         }
3560
3561         /* the kernel needs the non-byteswapped begin record */
3562         drr_noswap = drr;
3563
3564         flags->byteswap = B_FALSE;
3565         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3566                 /*
3567                  * We computed the checksum in the wrong byteorder in
3568                  * recv_read() above; do it again correctly.
3569                  */
3570                 bzero(&zcksum, sizeof (zio_cksum_t));
3571                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3572                 flags->byteswap = B_TRUE;
3573
3574                 drr.drr_type = BSWAP_32(drr.drr_type);
3575                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3576                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3577                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3578                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3579                 drrb->drr_type = BSWAP_32(drrb->drr_type);
3580                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3581                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3582                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3583         }
3584
3585         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3586                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3587                     "stream (bad magic number)"));
3588                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3589         }
3590
3591         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3592         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3593
3594         if (!DMU_STREAM_SUPPORTED(featureflags) ||
3595             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3596                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3597                     "stream has unsupported feature, feature flags = %lx"),
3598                     featureflags);
3599                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3600         }
3601
3602         if (strchr(drrb->drr_toname, '@') == NULL) {
3603                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3604                     "stream (bad snapshot name)"));
3605                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3606         }
3607
3608         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3609                 char nonpackage_sendfs[ZFS_MAXNAMELEN];
3610                 if (sendfs == NULL) {
3611                         /*
3612                          * We were not called from zfs_receive_package(). Get
3613                          * the fs specified by 'zfs send'.
3614                          */
3615                         char *cp;
3616                         (void) strlcpy(nonpackage_sendfs,
3617                             drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
3618                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3619                                 *cp = '\0';
3620                         sendfs = nonpackage_sendfs;
3621                         VERIFY(finalsnap == NULL);
3622                 }
3623                 return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags,
3624                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs,
3625                     cleanup_fd, action_handlep, finalsnap));
3626         } else {
3627                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3628                     DMU_COMPOUNDSTREAM);
3629                 return (zfs_receive_package(hdl, infd, tosnap, flags, &drr,
3630                     &zcksum, top_zfs, cleanup_fd, action_handlep));
3631         }
3632 }
3633
3634 /*
3635  * Restores a backup of tosnap from the file descriptor specified by infd.
3636  * Return 0 on total success, -2 if some things couldn't be
3637  * destroyed/renamed/promoted, -1 if some things couldn't be received.
3638  * (-1 will override -2, if -1 and the resumable flag was specified the
3639  * transfer can be resumed if the sending side supports it).
3640  */
3641 int
3642 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props,
3643     recvflags_t *flags, int infd, avl_tree_t *stream_avl)
3644 {
3645         char *top_zfs = NULL;
3646         int err;
3647         int cleanup_fd;
3648         uint64_t action_handle = 0;
3649         char *originsnap = NULL;
3650         if (props) {
3651                 err = nvlist_lookup_string(props, "origin", &originsnap);
3652                 if (err && err != ENOENT)
3653                         return (err);
3654         }
3655
3656         cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3657         VERIFY(cleanup_fd >= 0);
3658
3659         err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL,
3660             stream_avl, &top_zfs, cleanup_fd, &action_handle, NULL);
3661
3662         VERIFY(0 == close(cleanup_fd));
3663
3664         if (err == 0 && !flags->nomount && top_zfs) {
3665                 zfs_handle_t *zhp;
3666                 prop_changelist_t *clp;
3667
3668                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3669                 if (zhp != NULL) {
3670                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3671                             CL_GATHER_MOUNT_ALWAYS, 0);
3672                         zfs_close(zhp);
3673                         if (clp != NULL) {
3674                                 /* mount and share received datasets */
3675                                 err = changelist_postfix(clp);
3676                                 changelist_free(clp);
3677                         }
3678                 }
3679                 if (zhp == NULL || clp == NULL || err)
3680                         err = -1;
3681         }
3682         if (top_zfs)
3683                 free(top_zfs);
3684
3685         return (err);
3686 }