1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2012 by Delphix. All rights reserved.
  25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  26  */
  27 
  28 #include <assert.h>
  29 #include <ctype.h>
  30 #include <errno.h>
  31 #include <libintl.h>
  32 #include <stdio.h>
  33 #include <stdlib.h>
  34 #include <strings.h>
  35 #include <unistd.h>
  36 #include <stddef.h>
  37 #include <fcntl.h>
  38 #include <sys/mount.h>
  39 #include <pthread.h>
  40 #include <umem.h>
  41 #include <time.h>
  42 
  43 #include <libzfs.h>
  44 
  45 #include "zfs_namecheck.h"
  46 #include "zfs_prop.h"
  47 #include "zfs_fletcher.h"
  48 #include "libzfs_impl.h"
  49 #include <sha2.h>
  50 #include <sys/zio_checksum.h>
  51 #include <sys/ddt.h>
  52 
  53 /* in libzfs_dataset.c */
  54 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  55 
  56 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t *,
  57     int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *);
  58 
  59 static const zio_cksum_t zero_cksum = { 0 };
  60 
  61 typedef struct dedup_arg {
  62         int     inputfd;
  63         int     outputfd;
  64         libzfs_handle_t  *dedup_hdl;
  65 } dedup_arg_t;
  66 
  67 typedef struct progress_arg {
  68         zfs_handle_t *pa_zhp;
  69         int pa_fd;
  70         boolean_t pa_parsable;
  71 } progress_arg_t;
  72 
  73 typedef struct dataref {
  74         uint64_t ref_guid;
  75         uint64_t ref_object;
  76         uint64_t ref_offset;
  77 } dataref_t;
  78 
  79 typedef struct dedup_entry {
  80         struct dedup_entry      *dde_next;
  81         zio_cksum_t dde_chksum;
  82         uint64_t dde_prop;
  83         dataref_t dde_ref;
  84 } dedup_entry_t;
  85 
  86 #define MAX_DDT_PHYSMEM_PERCENT         20
  87 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
  88 
  89 typedef struct dedup_table {
  90         dedup_entry_t   **dedup_hash_array;
  91         umem_cache_t    *ddecache;
  92         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
  93         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
  94         uint64_t        ddt_count;
  95         int             numhashbits;
  96         boolean_t       ddt_full;
  97 } dedup_table_t;
  98 
  99 static int
 100 high_order_bit(uint64_t n)
 101 {
 102         int count;
 103 
 104         for (count = 0; n != 0; count++)
 105                 n >>= 1;
 106         return (count);
 107 }
 108 
 109 static size_t
 110 ssread(void *buf, size_t len, FILE *stream)
 111 {
 112         size_t outlen;
 113 
 114         if ((outlen = fread(buf, len, 1, stream)) == 0)
 115                 return (0);
 116 
 117         return (outlen);
 118 }
 119 
 120 static void
 121 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 122     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 123 {
 124         dedup_entry_t   *dde;
 125 
 126         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 127                 if (ddt->ddt_full == B_FALSE) {
 128                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 129                             "Dedup table full.  Deduplication will continue "
 130                             "with existing table entries"));
 131                         ddt->ddt_full = B_TRUE;
 132                 }
 133                 return;
 134         }
 135 
 136         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 137             != NULL) {
 138                 assert(*ddepp == NULL);
 139                 dde->dde_next = NULL;
 140                 dde->dde_chksum = *cs;
 141                 dde->dde_prop = prop;
 142                 dde->dde_ref = *dr;
 143                 *ddepp = dde;
 144                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 145                 ddt->ddt_count++;
 146         }
 147 }
 148 
 149 /*
 150  * Using the specified dedup table, do a lookup for an entry with
 151  * the checksum cs.  If found, return the block's reference info
 152  * in *dr. Otherwise, insert a new entry in the dedup table, using
 153  * the reference information specified by *dr.
 154  *
 155  * return value:  true - entry was found
 156  *                false - entry was not found
 157  */
 158 static boolean_t
 159 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 160     uint64_t prop, dataref_t *dr)
 161 {
 162         uint32_t hashcode;
 163         dedup_entry_t **ddepp;
 164 
 165         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 166 
 167         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 168             ddepp = &((*ddepp)->dde_next)) {
 169                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 170                     (*ddepp)->dde_prop == prop) {
 171                         *dr = (*ddepp)->dde_ref;
 172                         return (B_TRUE);
 173                 }
 174         }
 175         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 176         return (B_FALSE);
 177 }
 178 
 179 static int
 180 cksum_and_write(const void *buf, uint64_t len, zio_cksum_t *zc, int outfd)
 181 {
 182         fletcher_4_incremental_native(buf, len, zc);
 183         return (write(outfd, buf, len));
 184 }
 185 
 186 /*
 187  * This function is started in a separate thread when the dedup option
 188  * has been requested.  The main send thread determines the list of
 189  * snapshots to be included in the send stream and makes the ioctl calls
 190  * for each one.  But instead of having the ioctl send the output to the
 191  * the output fd specified by the caller of zfs_send()), the
 192  * ioctl is told to direct the output to a pipe, which is read by the
 193  * alternate thread running THIS function.  This function does the
 194  * dedup'ing by:
 195  *  1. building a dedup table (the DDT)
 196  *  2. doing checksums on each data block and inserting a record in the DDT
 197  *  3. looking for matching checksums, and
 198  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 199  *      a duplicate block is found.
 200  * The output of this function then goes to the output fd requested
 201  * by the caller of zfs_send().
 202  */
 203 static void *
 204 cksummer(void *arg)
 205 {
 206         dedup_arg_t *dda = arg;
 207         char *buf = malloc(1<<20);
 208         dmu_replay_record_t thedrr;
 209         dmu_replay_record_t *drr = &thedrr;
 210         struct drr_begin *drrb = &thedrr.drr_u.drr_begin;
 211         struct drr_end *drre = &thedrr.drr_u.drr_end;
 212         struct drr_object *drro = &thedrr.drr_u.drr_object;
 213         struct drr_write *drrw = &thedrr.drr_u.drr_write;
 214         struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
 215         FILE *ofp;
 216         int outfd;
 217         dmu_replay_record_t wbr_drr = {0};
 218         struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref;
 219         dedup_table_t ddt;
 220         zio_cksum_t stream_cksum;
 221         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 222         uint64_t numbuckets;
 223 
 224         ddt.max_ddt_size =
 225             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT)/100,
 226             SMALLEST_POSSIBLE_MAX_DDT_MB<<20);
 227 
 228         numbuckets = ddt.max_ddt_size/(sizeof (dedup_entry_t));
 229 
 230         /*
 231          * numbuckets must be a power of 2.  Increase number to
 232          * a power of 2 if necessary.
 233          */
 234         if (!ISP2(numbuckets))
 235                 numbuckets = 1 << high_order_bit(numbuckets);
 236 
 237         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 238         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 239             NULL, NULL, NULL, NULL, NULL, 0);
 240         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 241         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 242         ddt.ddt_full = B_FALSE;
 243 
 244         /* Initialize the write-by-reference block. */
 245         wbr_drr.drr_type = DRR_WRITE_BYREF;
 246         wbr_drr.drr_payloadlen = 0;
 247 
 248         outfd = dda->outputfd;
 249         ofp = fdopen(dda->inputfd, "r");
 250         while (ssread(drr, sizeof (dmu_replay_record_t), ofp) != 0) {
 251 
 252                 switch (drr->drr_type) {
 253                 case DRR_BEGIN:
 254                 {
 255                         int     fflags;
 256                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 257 
 258                         /* set the DEDUP feature flag for this stream */
 259                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 260                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 261                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 262                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 263 
 264                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 265                             &stream_cksum, outfd) == -1)
 266                                 goto out;
 267                         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 268                             DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
 269                                 int sz = drr->drr_payloadlen;
 270 
 271                                 if (sz > 1<<20) {
 272                                         free(buf);
 273                                         buf = malloc(sz);
 274                                 }
 275                                 (void) ssread(buf, sz, ofp);
 276                                 if (ferror(stdin))
 277                                         perror("fread");
 278                                 if (cksum_and_write(buf, sz, &stream_cksum,
 279                                     outfd) == -1)
 280                                         goto out;
 281                         }
 282                         break;
 283                 }
 284 
 285                 case DRR_END:
 286                 {
 287                         /* use the recalculated checksum */
 288                         ZIO_SET_CHECKSUM(&drre->drr_checksum,
 289                             stream_cksum.zc_word[0], stream_cksum.zc_word[1],
 290                             stream_cksum.zc_word[2], stream_cksum.zc_word[3]);
 291                         if ((write(outfd, drr,
 292                             sizeof (dmu_replay_record_t))) == -1)
 293                                 goto out;
 294                         break;
 295                 }
 296 
 297                 case DRR_OBJECT:
 298                 {
 299                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 300                             &stream_cksum, outfd) == -1)
 301                                 goto out;
 302                         if (drro->drr_bonuslen > 0) {
 303                                 (void) ssread(buf,
 304                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 305                                     ofp);
 306                                 if (cksum_and_write(buf,
 307                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 308                                     &stream_cksum, outfd) == -1)
 309                                         goto out;
 310                         }
 311                         break;
 312                 }
 313 
 314                 case DRR_SPILL:
 315                 {
 316                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 317                             &stream_cksum, outfd) == -1)
 318                                 goto out;
 319                         (void) ssread(buf, drrs->drr_length, ofp);
 320                         if (cksum_and_write(buf, drrs->drr_length,
 321                             &stream_cksum, outfd) == -1)
 322                                 goto out;
 323                         break;
 324                 }
 325 
 326                 case DRR_FREEOBJECTS:
 327                 {
 328                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 329                             &stream_cksum, outfd) == -1)
 330                                 goto out;
 331                         break;
 332                 }
 333 
 334                 case DRR_WRITE:
 335                 {
 336                         dataref_t       dataref;
 337 
 338                         (void) ssread(buf, drrw->drr_length, ofp);
 339 
 340                         /*
 341                          * Use the existing checksum if it's dedup-capable,
 342                          * else calculate a SHA256 checksum for it.
 343                          */
 344 
 345                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 346                             zero_cksum) ||
 347                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 348                                 SHA256_CTX      ctx;
 349                                 zio_cksum_t     tmpsha256;
 350 
 351                                 SHA256Init(&ctx);
 352                                 SHA256Update(&ctx, buf, drrw->drr_length);
 353                                 SHA256Final(&tmpsha256, &ctx);
 354                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 355                                     BE_64(tmpsha256.zc_word[0]);
 356                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 357                                     BE_64(tmpsha256.zc_word[1]);
 358                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 359                                     BE_64(tmpsha256.zc_word[2]);
 360                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 361                                     BE_64(tmpsha256.zc_word[3]);
 362                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 363                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 364                         }
 365 
 366                         dataref.ref_guid = drrw->drr_toguid;
 367                         dataref.ref_object = drrw->drr_object;
 368                         dataref.ref_offset = drrw->drr_offset;
 369 
 370                         if (ddt_update(dda->dedup_hdl, &ddt,
 371                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 372                             &dataref)) {
 373                                 /* block already present in stream */
 374                                 wbr_drrr->drr_object = drrw->drr_object;
 375                                 wbr_drrr->drr_offset = drrw->drr_offset;
 376                                 wbr_drrr->drr_length = drrw->drr_length;
 377                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 378                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 379                                 wbr_drrr->drr_refobject =
 380                                     dataref.ref_object;
 381                                 wbr_drrr->drr_refoffset =
 382                                     dataref.ref_offset;
 383 
 384                                 wbr_drrr->drr_checksumtype =
 385                                     drrw->drr_checksumtype;
 386                                 wbr_drrr->drr_checksumflags =
 387                                     drrw->drr_checksumtype;
 388                                 wbr_drrr->drr_key.ddk_cksum =
 389                                     drrw->drr_key.ddk_cksum;
 390                                 wbr_drrr->drr_key.ddk_prop =
 391                                     drrw->drr_key.ddk_prop;
 392 
 393                                 if (cksum_and_write(&wbr_drr,
 394                                     sizeof (dmu_replay_record_t), &stream_cksum,
 395                                     outfd) == -1)
 396                                         goto out;
 397                         } else {
 398                                 /* block not previously seen */
 399                                 if (cksum_and_write(drr,
 400                                     sizeof (dmu_replay_record_t), &stream_cksum,
 401                                     outfd) == -1)
 402                                         goto out;
 403                                 if (cksum_and_write(buf,
 404                                     drrw->drr_length,
 405                                     &stream_cksum, outfd) == -1)
 406                                         goto out;
 407                         }
 408                         break;
 409                 }
 410 
 411                 case DRR_FREE:
 412                 {
 413                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 414                             &stream_cksum, outfd) == -1)
 415                                 goto out;
 416                         break;
 417                 }
 418 
 419                 default:
 420                         (void) printf("INVALID record type 0x%x\n",
 421                             drr->drr_type);
 422                         /* should never happen, so assert */
 423                         assert(B_FALSE);
 424                 }
 425         }
 426 out:
 427         umem_cache_destroy(ddt.ddecache);
 428         free(ddt.dedup_hash_array);
 429         free(buf);
 430         (void) fclose(ofp);
 431 
 432         return (NULL);
 433 }
 434 
 435 /*
 436  * Routines for dealing with the AVL tree of fs-nvlists
 437  */
 438 typedef struct fsavl_node {
 439         avl_node_t fn_node;
 440         nvlist_t *fn_nvfs;
 441         char *fn_snapname;
 442         uint64_t fn_guid;
 443 } fsavl_node_t;
 444 
 445 static int
 446 fsavl_compare(const void *arg1, const void *arg2)
 447 {
 448         const fsavl_node_t *fn1 = arg1;
 449         const fsavl_node_t *fn2 = arg2;
 450 
 451         if (fn1->fn_guid > fn2->fn_guid)
 452                 return (+1);
 453         else if (fn1->fn_guid < fn2->fn_guid)
 454                 return (-1);
 455         else
 456                 return (0);
 457 }
 458 
 459 /*
 460  * Given the GUID of a snapshot, find its containing filesystem and
 461  * (optionally) name.
 462  */
 463 static nvlist_t *
 464 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 465 {
 466         fsavl_node_t fn_find;
 467         fsavl_node_t *fn;
 468 
 469         fn_find.fn_guid = snapguid;
 470 
 471         fn = avl_find(avl, &fn_find, NULL);
 472         if (fn) {
 473                 if (snapname)
 474                         *snapname = fn->fn_snapname;
 475                 return (fn->fn_nvfs);
 476         }
 477         return (NULL);
 478 }
 479 
 480 static void
 481 fsavl_destroy(avl_tree_t *avl)
 482 {
 483         fsavl_node_t *fn;
 484         void *cookie;
 485 
 486         if (avl == NULL)
 487                 return;
 488 
 489         cookie = NULL;
 490         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 491                 free(fn);
 492         avl_destroy(avl);
 493         free(avl);
 494 }
 495 
 496 /*
 497  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 498  */
 499 static avl_tree_t *
 500 fsavl_create(nvlist_t *fss)
 501 {
 502         avl_tree_t *fsavl;
 503         nvpair_t *fselem = NULL;
 504 
 505         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 506                 return (NULL);
 507 
 508         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 509             offsetof(fsavl_node_t, fn_node));
 510 
 511         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 512                 nvlist_t *nvfs, *snaps;
 513                 nvpair_t *snapelem = NULL;
 514 
 515                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 516                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 517 
 518                 while ((snapelem =
 519                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 520                         fsavl_node_t *fn;
 521                         uint64_t guid;
 522 
 523                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 524                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 525                                 fsavl_destroy(fsavl);
 526                                 return (NULL);
 527                         }
 528                         fn->fn_nvfs = nvfs;
 529                         fn->fn_snapname = nvpair_name(snapelem);
 530                         fn->fn_guid = guid;
 531 
 532                         /*
 533                          * Note: if there are multiple snaps with the
 534                          * same GUID, we ignore all but one.
 535                          */
 536                         if (avl_find(fsavl, fn, NULL) == NULL)
 537                                 avl_add(fsavl, fn);
 538                         else
 539                                 free(fn);
 540                 }
 541         }
 542 
 543         return (fsavl);
 544 }
 545 
 546 /*
 547  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 548  */
 549 typedef struct send_data {
 550         uint64_t parent_fromsnap_guid;
 551         nvlist_t *parent_snaps;
 552         nvlist_t *fss;
 553         nvlist_t *snapprops;
 554         const char *fromsnap;
 555         const char *tosnap;
 556         boolean_t recursive;
 557 
 558         /*
 559          * The header nvlist is of the following format:
 560          * {
 561          *   "tosnap" -> string
 562          *   "fromsnap" -> string (if incremental)
 563          *   "fss" -> {
 564          *      id -> {
 565          *
 566          *       "name" -> string (full name; for debugging)
 567          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 568          *
 569          *       "props" -> { name -> value (only if set here) }
 570          *       "snaps" -> { name (lastname) -> number (guid) }
 571          *       "snapprops" -> { name (lastname) -> { name -> value } }
 572          *
 573          *       "origin" -> number (guid) (if clone)
 574          *       "sent" -> boolean (not on-disk)
 575          *      }
 576          *   }
 577          * }
 578          *
 579          */
 580 } send_data_t;
 581 
 582 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 583 
 584 static int
 585 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 586 {
 587         send_data_t *sd = arg;
 588         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 589         char *snapname;
 590         nvlist_t *nv;
 591 
 592         snapname = strrchr(zhp->zfs_name, '@')+1;
 593 
 594         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 595         /*
 596          * NB: if there is no fromsnap here (it's a newly created fs in
 597          * an incremental replication), we will substitute the tosnap.
 598          */
 599         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 600             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 601             strcmp(snapname, sd->tosnap) == 0)) {
 602                 sd->parent_fromsnap_guid = guid;
 603         }
 604 
 605         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 606         send_iterate_prop(zhp, nv);
 607         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 608         nvlist_free(nv);
 609 
 610         zfs_close(zhp);
 611         return (0);
 612 }
 613 
 614 static void
 615 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 616 {
 617         nvpair_t *elem = NULL;
 618 
 619         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 620                 char *propname = nvpair_name(elem);
 621                 zfs_prop_t prop = zfs_name_to_prop(propname);
 622                 nvlist_t *propnv;
 623 
 624                 if (!zfs_prop_user(propname)) {
 625                         /*
 626                          * Realistically, this should never happen.  However,
 627                          * we want the ability to add DSL properties without
 628                          * needing to make incompatible version changes.  We
 629                          * need to ignore unknown properties to allow older
 630                          * software to still send datasets containing these
 631                          * properties, with the unknown properties elided.
 632                          */
 633                         if (prop == ZPROP_INVAL)
 634                                 continue;
 635 
 636                         if (zfs_prop_readonly(prop))
 637                                 continue;
 638                 }
 639 
 640                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 641                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 642                     prop == ZFS_PROP_REFQUOTA ||
 643                     prop == ZFS_PROP_REFRESERVATION) {
 644                         char *source;
 645                         uint64_t value;
 646                         verify(nvlist_lookup_uint64(propnv,
 647                             ZPROP_VALUE, &value) == 0);
 648                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 649                                 continue;
 650                         /*
 651                          * May have no source before SPA_VERSION_RECVD_PROPS,
 652                          * but is still modifiable.
 653                          */
 654                         if (nvlist_lookup_string(propnv,
 655                             ZPROP_SOURCE, &source) == 0) {
 656                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 657                                     (strcmp(source,
 658                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 659                                         continue;
 660                         }
 661                 } else {
 662                         char *source;
 663                         if (nvlist_lookup_string(propnv,
 664                             ZPROP_SOURCE, &source) != 0)
 665                                 continue;
 666                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 667                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 668                                 continue;
 669                 }
 670 
 671                 if (zfs_prop_user(propname) ||
 672                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 673                         char *value;
 674                         verify(nvlist_lookup_string(propnv,
 675                             ZPROP_VALUE, &value) == 0);
 676                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 677                 } else {
 678                         uint64_t value;
 679                         verify(nvlist_lookup_uint64(propnv,
 680                             ZPROP_VALUE, &value) == 0);
 681                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 682                 }
 683         }
 684 }
 685 
 686 /*
 687  * recursively generate nvlists describing datasets.  See comment
 688  * for the data structure send_data_t above for description of contents
 689  * of the nvlist.
 690  */
 691 static int
 692 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 693 {
 694         send_data_t *sd = arg;
 695         nvlist_t *nvfs, *nv;
 696         int rv = 0;
 697         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 698         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 699         char guidstring[64];
 700 
 701         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 702         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 703         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 704             sd->parent_fromsnap_guid));
 705 
 706         if (zhp->zfs_dmustats.dds_origin[0]) {
 707                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 708                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 709                 if (origin == NULL)
 710                         return (-1);
 711                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 712                     origin->zfs_dmustats.dds_guid));
 713         }
 714 
 715         /* iterate over props */
 716         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 717         send_iterate_prop(zhp, nv);
 718         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 719         nvlist_free(nv);
 720 
 721         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 722         sd->parent_fromsnap_guid = 0;
 723         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 724         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 725         (void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
 726         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 727         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 728         nvlist_free(sd->parent_snaps);
 729         nvlist_free(sd->snapprops);
 730 
 731         /* add this fs to nvlist */
 732         (void) snprintf(guidstring, sizeof (guidstring),
 733             "0x%llx", (longlong_t)guid);
 734         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 735         nvlist_free(nvfs);
 736 
 737         /* iterate over children */
 738         if (sd->recursive)
 739                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 740 
 741         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 742 
 743         zfs_close(zhp);
 744         return (rv);
 745 }
 746 
 747 static int
 748 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 749     const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
 750 {
 751         zfs_handle_t *zhp;
 752         send_data_t sd = { 0 };
 753         int error;
 754 
 755         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 756         if (zhp == NULL)
 757                 return (EZFS_BADTYPE);
 758 
 759         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 760         sd.fromsnap = fromsnap;
 761         sd.tosnap = tosnap;
 762         sd.recursive = recursive;
 763 
 764         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 765                 nvlist_free(sd.fss);
 766                 if (avlp != NULL)
 767                         *avlp = NULL;
 768                 *nvlp = NULL;
 769                 return (error);
 770         }
 771 
 772         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 773                 nvlist_free(sd.fss);
 774                 *nvlp = NULL;
 775                 return (EZFS_NOMEM);
 776         }
 777 
 778         *nvlp = sd.fss;
 779         return (0);
 780 }
 781 
 782 /*
 783  * Routines specific to "zfs send"
 784  */
 785 typedef struct send_dump_data {
 786         /* these are all just the short snapname (the part after the @) */
 787         const char *fromsnap;
 788         const char *tosnap;
 789         char prevsnap[ZFS_MAXNAMELEN];
 790         uint64_t prevsnap_obj;
 791         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 792         boolean_t verbose, dryrun, parsable, progress;
 793         int outfd;
 794         boolean_t err;
 795         nvlist_t *fss;
 796         avl_tree_t *fsavl;
 797         snapfilter_cb_t *filter_cb;
 798         void *filter_cb_arg;
 799         nvlist_t *debugnv;
 800         char holdtag[ZFS_MAXNAMELEN];
 801         int cleanup_fd;
 802         uint64_t size;
 803 } send_dump_data_t;
 804 
 805 static int
 806 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
 807     boolean_t fromorigin, uint64_t *sizep)
 808 {
 809         zfs_cmd_t zc = { 0 };
 810         libzfs_handle_t *hdl = zhp->zfs_hdl;
 811 
 812         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 813         assert(fromsnap_obj == 0 || !fromorigin);
 814 
 815         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 816         zc.zc_obj = fromorigin;
 817         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 818         zc.zc_fromobj = fromsnap_obj;
 819         zc.zc_guid = 1;  /* estimate flag */
 820 
 821         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 822                 char errbuf[1024];
 823                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 824                     "warning: cannot estimate space for '%s'"), zhp->zfs_name);
 825 
 826                 switch (errno) {
 827                 case EXDEV:
 828                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 829                             "not an earlier snapshot from the same fs"));
 830                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 831 
 832                 case ENOENT:
 833                         if (zfs_dataset_exists(hdl, zc.zc_name,
 834                             ZFS_TYPE_SNAPSHOT)) {
 835                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 836                                     "incremental source (@%s) does not exist"),
 837                                     zc.zc_value);
 838                         }
 839                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 840 
 841                 case EDQUOT:
 842                 case EFBIG:
 843                 case EIO:
 844                 case ENOLINK:
 845                 case ENOSPC:
 846                 case ENOSTR:
 847                 case ENXIO:
 848                 case EPIPE:
 849                 case ERANGE:
 850                 case EFAULT:
 851                 case EROFS:
 852                         zfs_error_aux(hdl, strerror(errno));
 853                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 854 
 855                 default:
 856                         return (zfs_standard_error(hdl, errno, errbuf));
 857                 }
 858         }
 859 
 860         *sizep = zc.zc_objset_type;
 861 
 862         return (0);
 863 }
 864 
 865 /*
 866  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 867  * NULL) to the file descriptor specified by outfd.
 868  */
 869 static int
 870 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
 871     boolean_t fromorigin, int outfd, nvlist_t *debugnv)
 872 {
 873         zfs_cmd_t zc = { 0 };
 874         libzfs_handle_t *hdl = zhp->zfs_hdl;
 875         nvlist_t *thisdbg;
 876 
 877         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 878         assert(fromsnap_obj == 0 || !fromorigin);
 879 
 880         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 881         zc.zc_cookie = outfd;
 882         zc.zc_obj = fromorigin;
 883         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 884         zc.zc_fromobj = fromsnap_obj;
 885 
 886         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 887         if (fromsnap && fromsnap[0] != '\0') {
 888                 VERIFY(0 == nvlist_add_string(thisdbg,
 889                     "fromsnap", fromsnap));
 890         }
 891 
 892         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 893                 char errbuf[1024];
 894                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 895                     "warning: cannot send '%s'"), zhp->zfs_name);
 896 
 897                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
 898                 if (debugnv) {
 899                         VERIFY(0 == nvlist_add_nvlist(debugnv,
 900                             zhp->zfs_name, thisdbg));
 901                 }
 902                 nvlist_free(thisdbg);
 903 
 904                 switch (errno) {
 905                 case EXDEV:
 906                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 907                             "not an earlier snapshot from the same fs"));
 908                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 909 
 910                 case ENOENT:
 911                         if (zfs_dataset_exists(hdl, zc.zc_name,
 912                             ZFS_TYPE_SNAPSHOT)) {
 913                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 914                                     "incremental source (@%s) does not exist"),
 915                                     zc.zc_value);
 916                         }
 917                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 918 
 919                 case EDQUOT:
 920                 case EFBIG:
 921                 case EIO:
 922                 case ENOLINK:
 923                 case ENOSPC:
 924                 case ENOSTR:
 925                 case ENXIO:
 926                 case EPIPE:
 927                 case ERANGE:
 928                 case EFAULT:
 929                 case EROFS:
 930                         zfs_error_aux(hdl, strerror(errno));
 931                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 932 
 933                 default:
 934                         return (zfs_standard_error(hdl, errno, errbuf));
 935                 }
 936         }
 937 
 938         if (debugnv)
 939                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
 940         nvlist_free(thisdbg);
 941 
 942         return (0);
 943 }
 944 
 945 static int
 946 hold_for_send(zfs_handle_t *zhp, send_dump_data_t *sdd)
 947 {
 948         zfs_handle_t *pzhp;
 949         int error = 0;
 950         char *thissnap;
 951 
 952         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 953 
 954         if (sdd->dryrun)
 955                 return (0);
 956 
 957         /*
 958          * zfs_send() only opens a cleanup_fd for sends that need it,
 959          * e.g. replication and doall.
 960          */
 961         if (sdd->cleanup_fd == -1)
 962                 return (0);
 963 
 964         thissnap = strchr(zhp->zfs_name, '@') + 1;
 965         *(thissnap - 1) = '\0';
 966         pzhp = zfs_open(zhp->zfs_hdl, zhp->zfs_name, ZFS_TYPE_DATASET);
 967         *(thissnap - 1) = '@';
 968 
 969         /*
 970          * It's OK if the parent no longer exists.  The send code will
 971          * handle that error.
 972          */
 973         if (pzhp) {
 974                 error = zfs_hold(pzhp, thissnap, sdd->holdtag,
 975                     B_FALSE, B_TRUE, B_TRUE, sdd->cleanup_fd,
 976                     zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID),
 977                     zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG));
 978                 zfs_close(pzhp);
 979         }
 980 
 981         return (error);
 982 }
 983 
 984 static void *
 985 send_progress_thread(void *arg)
 986 {
 987         progress_arg_t *pa = arg;
 988 
 989         zfs_cmd_t zc = { 0 };
 990         zfs_handle_t *zhp = pa->pa_zhp;
 991         libzfs_handle_t *hdl = zhp->zfs_hdl;
 992         unsigned long long bytes;
 993         char buf[16];
 994 
 995         time_t t;
 996         struct tm *tm;
 997 
 998         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 999         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1000 
1001         if (!pa->pa_parsable)
1002                 (void) fprintf(stderr, "TIME        SENT   SNAPSHOT\n");
1003 
1004         /*
1005          * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1006          */
1007         for (;;) {
1008                 (void) sleep(1);
1009 
1010                 zc.zc_cookie = pa->pa_fd;
1011                 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1012                         return ((void *)-1);
1013 
1014                 (void) time(&t);
1015                 tm = localtime(&t);
1016                 bytes = zc.zc_cookie;
1017 
1018                 if (pa->pa_parsable) {
1019                         (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1020                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1021                             bytes, zhp->zfs_name);
1022                 } else {
1023                         zfs_nicenum(bytes, buf, sizeof (buf));
1024                         (void) fprintf(stderr, "%02d:%02d:%02d   %5s   %s\n",
1025                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1026                             buf, zhp->zfs_name);
1027                 }
1028         }
1029 }
1030 
1031 static int
1032 dump_snapshot(zfs_handle_t *zhp, void *arg)
1033 {
1034         send_dump_data_t *sdd = arg;
1035         progress_arg_t pa = { 0 };
1036         pthread_t tid;
1037 
1038         char *thissnap;
1039         int err;
1040         boolean_t isfromsnap, istosnap, fromorigin;
1041         boolean_t exclude = B_FALSE;
1042 
1043         thissnap = strchr(zhp->zfs_name, '@') + 1;
1044         isfromsnap = (sdd->fromsnap != NULL &&
1045             strcmp(sdd->fromsnap, thissnap) == 0);
1046 
1047         if (!sdd->seenfrom && isfromsnap) {
1048                 err = hold_for_send(zhp, sdd);
1049                 if (err == 0) {
1050                         sdd->seenfrom = B_TRUE;
1051                         (void) strcpy(sdd->prevsnap, thissnap);
1052                         sdd->prevsnap_obj = zfs_prop_get_int(zhp,
1053                             ZFS_PROP_OBJSETID);
1054                 } else if (err == ENOENT) {
1055                         err = 0;
1056                 }
1057                 zfs_close(zhp);
1058                 return (err);
1059         }
1060 
1061         if (sdd->seento || !sdd->seenfrom) {
1062                 zfs_close(zhp);
1063                 return (0);
1064         }
1065 
1066         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1067         if (istosnap)
1068                 sdd->seento = B_TRUE;
1069 
1070         if (!sdd->doall && !isfromsnap && !istosnap) {
1071                 if (sdd->replicate) {
1072                         char *snapname;
1073                         nvlist_t *snapprops;
1074                         /*
1075                          * Filter out all intermediate snapshots except origin
1076                          * snapshots needed to replicate clones.
1077                          */
1078                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1079                             zhp->zfs_dmustats.dds_guid, &snapname);
1080 
1081                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1082                             "snapprops", &snapprops));
1083                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1084                             thissnap, &snapprops));
1085                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
1086                 } else {
1087                         exclude = B_TRUE;
1088                 }
1089         }
1090 
1091         /*
1092          * If a filter function exists, call it to determine whether
1093          * this snapshot will be sent.
1094          */
1095         if (exclude || (sdd->filter_cb != NULL &&
1096             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1097                 /*
1098                  * This snapshot is filtered out.  Don't send it, and don't
1099                  * set prevsnap_obj, so it will be as if this snapshot didn't
1100                  * exist, and the next accepted snapshot will be sent as
1101                  * an incremental from the last accepted one, or as the
1102                  * first (and full) snapshot in the case of a replication,
1103                  * non-incremental send.
1104                  */
1105                 zfs_close(zhp);
1106                 return (0);
1107         }
1108 
1109         err = hold_for_send(zhp, sdd);
1110         if (err) {
1111                 if (err == ENOENT)
1112                         err = 0;
1113                 zfs_close(zhp);
1114                 return (err);
1115         }
1116 
1117         fromorigin = sdd->prevsnap[0] == '\0' &&
1118             (sdd->fromorigin || sdd->replicate);
1119 
1120         if (sdd->verbose) {
1121                 uint64_t size;
1122                 err = estimate_ioctl(zhp, sdd->prevsnap_obj,
1123                     fromorigin, &size);
1124 
1125                 if (sdd->parsable) {
1126                         if (sdd->prevsnap[0] != '\0') {
1127                                 (void) fprintf(stderr, "incremental\t%s\t%s",
1128                                     sdd->prevsnap, zhp->zfs_name);
1129                         } else {
1130                                 (void) fprintf(stderr, "full\t%s",
1131                                     zhp->zfs_name);
1132                         }
1133                 } else {
1134                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1135                             "send from @%s to %s"),
1136                             sdd->prevsnap, zhp->zfs_name);
1137                 }
1138                 if (err == 0) {
1139                         if (sdd->parsable) {
1140                                 (void) fprintf(stderr, "\t%llu\n",
1141                                     (longlong_t)size);
1142                         } else {
1143                                 char buf[16];
1144                                 zfs_nicenum(size, buf, sizeof (buf));
1145                                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1146                                     " estimated size is %s\n"), buf);
1147                         }
1148                         sdd->size += size;
1149                 } else {
1150                         (void) fprintf(stderr, "\n");
1151                 }
1152         }
1153 
1154         if (!sdd->dryrun) {
1155                 /*
1156                  * If progress reporting is requested, spawn a new thread to
1157                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1158                  */
1159                 if (sdd->progress) {
1160                         pa.pa_zhp = zhp;
1161                         pa.pa_fd = sdd->outfd;
1162                         pa.pa_parsable = sdd->parsable;
1163 
1164                         if (err = pthread_create(&tid, NULL,
1165                             send_progress_thread, &pa)) {
1166                                 zfs_close(zhp);
1167                                 return (err);
1168                         }
1169                 }
1170 
1171                 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1172                     fromorigin, sdd->outfd, sdd->debugnv);
1173 
1174                 if (sdd->progress) {
1175                         (void) pthread_cancel(tid);
1176                         (void) pthread_join(tid, NULL);
1177                 }
1178         }
1179 
1180         (void) strcpy(sdd->prevsnap, thissnap);
1181         sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1182         zfs_close(zhp);
1183         return (err);
1184 }
1185 
1186 static int
1187 dump_filesystem(zfs_handle_t *zhp, void *arg)
1188 {
1189         int rv = 0;
1190         send_dump_data_t *sdd = arg;
1191         boolean_t missingfrom = B_FALSE;
1192         zfs_cmd_t zc = { 0 };
1193 
1194         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1195             zhp->zfs_name, sdd->tosnap);
1196         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1197                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1198                     "WARNING: could not send %s@%s: does not exist\n"),
1199                     zhp->zfs_name, sdd->tosnap);
1200                 sdd->err = B_TRUE;
1201                 return (0);
1202         }
1203 
1204         if (sdd->replicate && sdd->fromsnap) {
1205                 /*
1206                  * If this fs does not have fromsnap, and we're doing
1207                  * recursive, we need to send a full stream from the
1208                  * beginning (or an incremental from the origin if this
1209                  * is a clone).  If we're doing non-recursive, then let
1210                  * them get the error.
1211                  */
1212                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1213                     zhp->zfs_name, sdd->fromsnap);
1214                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1215                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1216                         missingfrom = B_TRUE;
1217                 }
1218         }
1219 
1220         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1221         sdd->prevsnap_obj = 0;
1222         if (sdd->fromsnap == NULL || missingfrom)
1223                 sdd->seenfrom = B_TRUE;
1224 
1225         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1226         if (!sdd->seenfrom) {
1227                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1228                     "WARNING: could not send %s@%s:\n"
1229                     "incremental source (%s@%s) does not exist\n"),
1230                     zhp->zfs_name, sdd->tosnap,
1231                     zhp->zfs_name, sdd->fromsnap);
1232                 sdd->err = B_TRUE;
1233         } else if (!sdd->seento) {
1234                 if (sdd->fromsnap) {
1235                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1236                             "WARNING: could not send %s@%s:\n"
1237                             "incremental source (%s@%s) "
1238                             "is not earlier than it\n"),
1239                             zhp->zfs_name, sdd->tosnap,
1240                             zhp->zfs_name, sdd->fromsnap);
1241                 } else {
1242                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1243                             "WARNING: "
1244                             "could not send %s@%s: does not exist\n"),
1245                             zhp->zfs_name, sdd->tosnap);
1246                 }
1247                 sdd->err = B_TRUE;
1248         }
1249 
1250         return (rv);
1251 }
1252 
1253 static int
1254 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1255 {
1256         send_dump_data_t *sdd = arg;
1257         nvpair_t *fspair;
1258         boolean_t needagain, progress;
1259 
1260         if (!sdd->replicate)
1261                 return (dump_filesystem(rzhp, sdd));
1262 
1263         /* Mark the clone origin snapshots. */
1264         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1265             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1266                 nvlist_t *nvfs;
1267                 uint64_t origin_guid = 0;
1268 
1269                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1270                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1271                 if (origin_guid != 0) {
1272                         char *snapname;
1273                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1274                             origin_guid, &snapname);
1275                         if (origin_nv != NULL) {
1276                                 nvlist_t *snapprops;
1277                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1278                                     "snapprops", &snapprops));
1279                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1280                                     snapname, &snapprops));
1281                                 VERIFY(0 == nvlist_add_boolean(
1282                                     snapprops, "is_clone_origin"));
1283                         }
1284                 }
1285         }
1286 again:
1287         needagain = progress = B_FALSE;
1288         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1289             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1290                 nvlist_t *fslist, *parent_nv;
1291                 char *fsname;
1292                 zfs_handle_t *zhp;
1293                 int err;
1294                 uint64_t origin_guid = 0;
1295                 uint64_t parent_guid = 0;
1296 
1297                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1298                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1299                         continue;
1300 
1301                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1302                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1303                 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1304                     &parent_guid);
1305 
1306                 if (parent_guid != 0) {
1307                         parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1308                         if (!nvlist_exists(parent_nv, "sent")) {
1309                                 /* parent has not been sent; skip this one */
1310                                 needagain = B_TRUE;
1311                                 continue;
1312                         }
1313                 }
1314 
1315                 if (origin_guid != 0) {
1316                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1317                             origin_guid, NULL);
1318                         if (origin_nv != NULL &&
1319                             !nvlist_exists(origin_nv, "sent")) {
1320                                 /*
1321                                  * origin has not been sent yet;
1322                                  * skip this clone.
1323                                  */
1324                                 needagain = B_TRUE;
1325                                 continue;
1326                         }
1327                 }
1328 
1329                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1330                 if (zhp == NULL)
1331                         return (-1);
1332                 err = dump_filesystem(zhp, sdd);
1333                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1334                 progress = B_TRUE;
1335                 zfs_close(zhp);
1336                 if (err)
1337                         return (err);
1338         }
1339         if (needagain) {
1340                 assert(progress);
1341                 goto again;
1342         }
1343 
1344         /* clean out the sent flags in case we reuse this fss */
1345         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1346             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1347                 nvlist_t *fslist;
1348 
1349                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1350                 (void) nvlist_remove_all(fslist, "sent");
1351         }
1352 
1353         return (0);
1354 }
1355 
1356 /*
1357  * Generate a send stream for the dataset identified by the argument zhp.
1358  *
1359  * The content of the send stream is the snapshot identified by
1360  * 'tosnap'.  Incremental streams are requested in two ways:
1361  *     - from the snapshot identified by "fromsnap" (if non-null) or
1362  *     - from the origin of the dataset identified by zhp, which must
1363  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1364  *       is TRUE.
1365  *
1366  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1367  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1368  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1369  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1370  * case too. If "props" is set, send properties.
1371  */
1372 int
1373 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1374     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1375     void *cb_arg, nvlist_t **debugnvp)
1376 {
1377         char errbuf[1024];
1378         send_dump_data_t sdd = { 0 };
1379         int err = 0;
1380         nvlist_t *fss = NULL;
1381         avl_tree_t *fsavl = NULL;
1382         static uint64_t holdseq;
1383         int spa_version;
1384         pthread_t tid;
1385         int pipefd[2];
1386         dedup_arg_t dda = { 0 };
1387         int featureflags = 0;
1388 
1389         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1390             "cannot send '%s'"), zhp->zfs_name);
1391 
1392         if (fromsnap && fromsnap[0] == '\0') {
1393                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1394                     "zero-length incremental source"));
1395                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1396         }
1397 
1398         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1399                 uint64_t version;
1400                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1401                 if (version >= ZPL_VERSION_SA) {
1402                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1403                 }
1404         }
1405 
1406         if (flags->dedup && !flags->dryrun) {
1407                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1408                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1409                 if (err = pipe(pipefd)) {
1410                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1411                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1412                             errbuf));
1413                 }
1414                 dda.outputfd = outfd;
1415                 dda.inputfd = pipefd[1];
1416                 dda.dedup_hdl = zhp->zfs_hdl;
1417                 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1418                         (void) close(pipefd[0]);
1419                         (void) close(pipefd[1]);
1420                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1421                         return (zfs_error(zhp->zfs_hdl,
1422                             EZFS_THREADCREATEFAILED, errbuf));
1423                 }
1424         }
1425 
1426         if (flags->replicate || flags->doall || flags->props) {
1427                 dmu_replay_record_t drr = { 0 };
1428                 char *packbuf = NULL;
1429                 size_t buflen = 0;
1430                 zio_cksum_t zc = { 0 };
1431 
1432                 if (flags->replicate || flags->props) {
1433                         nvlist_t *hdrnv;
1434 
1435                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1436                         if (fromsnap) {
1437                                 VERIFY(0 == nvlist_add_string(hdrnv,
1438                                     "fromsnap", fromsnap));
1439                         }
1440                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1441                         if (!flags->replicate) {
1442                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1443                                     "not_recursive"));
1444                         }
1445 
1446                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1447                             fromsnap, tosnap, flags->replicate, &fss, &fsavl);
1448                         if (err)
1449                                 goto err_out;
1450                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1451                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1452                             NV_ENCODE_XDR, 0);
1453                         if (debugnvp)
1454                                 *debugnvp = hdrnv;
1455                         else
1456                                 nvlist_free(hdrnv);
1457                         if (err) {
1458                                 fsavl_destroy(fsavl);
1459                                 nvlist_free(fss);
1460                                 goto stderr_out;
1461                         }
1462                 }
1463 
1464                 if (!flags->dryrun) {
1465                         /* write first begin record */
1466                         drr.drr_type = DRR_BEGIN;
1467                         drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1468                         DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1469                             drr_versioninfo, DMU_COMPOUNDSTREAM);
1470                         DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1471                             drr_versioninfo, featureflags);
1472                         (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1473                             sizeof (drr.drr_u.drr_begin.drr_toname),
1474                             "%s@%s", zhp->zfs_name, tosnap);
1475                         drr.drr_payloadlen = buflen;
1476                         err = cksum_and_write(&drr, sizeof (drr), &zc, outfd);
1477 
1478                         /* write header nvlist */
1479                         if (err != -1 && packbuf != NULL) {
1480                                 err = cksum_and_write(packbuf, buflen, &zc,
1481                                     outfd);
1482                         }
1483                         free(packbuf);
1484                         if (err == -1) {
1485                                 fsavl_destroy(fsavl);
1486                                 nvlist_free(fss);
1487                                 err = errno;
1488                                 goto stderr_out;
1489                         }
1490 
1491                         /* write end record */
1492                         bzero(&drr, sizeof (drr));
1493                         drr.drr_type = DRR_END;
1494                         drr.drr_u.drr_end.drr_checksum = zc;
1495                         err = write(outfd, &drr, sizeof (drr));
1496                         if (err == -1) {
1497                                 fsavl_destroy(fsavl);
1498                                 nvlist_free(fss);
1499                                 err = errno;
1500                                 goto stderr_out;
1501                         }
1502 
1503                         err = 0;
1504                 }
1505         }
1506 
1507         /* dump each stream */
1508         sdd.fromsnap = fromsnap;
1509         sdd.tosnap = tosnap;
1510         if (flags->dedup)
1511                 sdd.outfd = pipefd[0];
1512         else
1513                 sdd.outfd = outfd;
1514         sdd.replicate = flags->replicate;
1515         sdd.doall = flags->doall;
1516         sdd.fromorigin = flags->fromorigin;
1517         sdd.fss = fss;
1518         sdd.fsavl = fsavl;
1519         sdd.verbose = flags->verbose;
1520         sdd.parsable = flags->parsable;
1521         sdd.progress = flags->progress;
1522         sdd.dryrun = flags->dryrun;
1523         sdd.filter_cb = filter_func;
1524         sdd.filter_cb_arg = cb_arg;
1525         if (debugnvp)
1526                 sdd.debugnv = *debugnvp;
1527 
1528         /*
1529          * Some flags require that we place user holds on the datasets that are
1530          * being sent so they don't get destroyed during the send. We can skip
1531          * this step if the pool is imported read-only since the datasets cannot
1532          * be destroyed.
1533          */
1534         if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1535             ZPOOL_PROP_READONLY, NULL) &&
1536             zfs_spa_version(zhp, &spa_version) == 0 &&
1537             spa_version >= SPA_VERSION_USERREFS &&
1538             (flags->doall || flags->replicate)) {
1539                 ++holdseq;
1540                 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1541                     ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1542                 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1543                 if (sdd.cleanup_fd < 0) {
1544                         err = errno;
1545                         goto stderr_out;
1546                 }
1547         } else {
1548                 sdd.cleanup_fd = -1;
1549         }
1550         if (flags->verbose) {
1551                 /*
1552                  * Do a verbose no-op dry run to get all the verbose output
1553                  * before generating any data.  Then do a non-verbose real
1554                  * run to generate the streams.
1555                  */
1556                 sdd.dryrun = B_TRUE;
1557                 err = dump_filesystems(zhp, &sdd);
1558                 sdd.dryrun = flags->dryrun;
1559                 sdd.verbose = B_FALSE;
1560                 if (flags->parsable) {
1561                         (void) fprintf(stderr, "size\t%llu\n",
1562                             (longlong_t)sdd.size);
1563                 } else {
1564                         char buf[16];
1565                         zfs_nicenum(sdd.size, buf, sizeof (buf));
1566                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1567                             "total estimated size is %s\n"), buf);
1568                 }
1569         }
1570         err = dump_filesystems(zhp, &sdd);
1571         fsavl_destroy(fsavl);
1572         nvlist_free(fss);
1573 
1574         if (flags->dedup) {
1575                 (void) close(pipefd[0]);
1576                 (void) pthread_join(tid, NULL);
1577         }
1578 
1579         if (sdd.cleanup_fd != -1) {
1580                 VERIFY(0 == close(sdd.cleanup_fd));
1581                 sdd.cleanup_fd = -1;
1582         }
1583 
1584         if (!flags->dryrun && (flags->replicate || flags->doall ||
1585             flags->props)) {
1586                 /*
1587                  * write final end record.  NB: want to do this even if
1588                  * there was some error, because it might not be totally
1589                  * failed.
1590                  */
1591                 dmu_replay_record_t drr = { 0 };
1592                 drr.drr_type = DRR_END;
1593                 if (write(outfd, &drr, sizeof (drr)) == -1) {
1594                         return (zfs_standard_error(zhp->zfs_hdl,
1595                             errno, errbuf));
1596                 }
1597         }
1598 
1599         return (err || sdd.err);
1600 
1601 stderr_out:
1602         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1603 err_out:
1604         if (sdd.cleanup_fd != -1)
1605                 VERIFY(0 == close(sdd.cleanup_fd));
1606         if (flags->dedup) {
1607                 (void) pthread_cancel(tid);
1608                 (void) pthread_join(tid, NULL);
1609                 (void) close(pipefd[0]);
1610         }
1611         return (err);
1612 }
1613 
1614 /*
1615  * Routines specific to "zfs recv"
1616  */
1617 
1618 static int
1619 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1620     boolean_t byteswap, zio_cksum_t *zc)
1621 {
1622         char *cp = buf;
1623         int rv;
1624         int len = ilen;
1625 
1626         do {
1627                 rv = read(fd, cp, len);
1628                 cp += rv;
1629                 len -= rv;
1630         } while (rv > 0);
1631 
1632         if (rv < 0 || len != 0) {
1633                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1634                     "failed to read from stream"));
1635                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1636                     "cannot receive")));
1637         }
1638 
1639         if (zc) {
1640                 if (byteswap)
1641                         fletcher_4_incremental_byteswap(buf, ilen, zc);
1642                 else
1643                         fletcher_4_incremental_native(buf, ilen, zc);
1644         }
1645         return (0);
1646 }
1647 
1648 static int
1649 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1650     boolean_t byteswap, zio_cksum_t *zc)
1651 {
1652         char *buf;
1653         int err;
1654 
1655         buf = zfs_alloc(hdl, len);
1656         if (buf == NULL)
1657                 return (ENOMEM);
1658 
1659         err = recv_read(hdl, fd, buf, len, byteswap, zc);
1660         if (err != 0) {
1661                 free(buf);
1662                 return (err);
1663         }
1664 
1665         err = nvlist_unpack(buf, len, nvp, 0);
1666         free(buf);
1667         if (err != 0) {
1668                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1669                     "stream (malformed nvlist)"));
1670                 return (EINVAL);
1671         }
1672         return (0);
1673 }
1674 
1675 static int
1676 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1677     int baselen, char *newname, recvflags_t *flags)
1678 {
1679         static int seq;
1680         zfs_cmd_t zc = { 0 };
1681         int err;
1682         prop_changelist_t *clp;
1683         zfs_handle_t *zhp;
1684 
1685         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1686         if (zhp == NULL)
1687                 return (-1);
1688         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1689             flags->force ? MS_FORCE : 0);
1690         zfs_close(zhp);
1691         if (clp == NULL)
1692                 return (-1);
1693         err = changelist_prefix(clp);
1694         if (err)
1695                 return (err);
1696 
1697         zc.zc_objset_type = DMU_OST_ZFS;
1698         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1699 
1700         if (tryname) {
1701                 (void) strcpy(newname, tryname);
1702 
1703                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1704 
1705                 if (flags->verbose) {
1706                         (void) printf("attempting rename %s to %s\n",
1707                             zc.zc_name, zc.zc_value);
1708                 }
1709                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1710                 if (err == 0)
1711                         changelist_rename(clp, name, tryname);
1712         } else {
1713                 err = ENOENT;
1714         }
1715 
1716         if (err != 0 && strncmp(name+baselen, "recv-", 5) != 0) {
1717                 seq++;
1718 
1719                 (void) strncpy(newname, name, baselen);
1720                 (void) snprintf(newname+baselen, ZFS_MAXNAMELEN-baselen,
1721                     "recv-%u-%u", getpid(), seq);
1722                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1723 
1724                 if (flags->verbose) {
1725                         (void) printf("failed - trying rename %s to %s\n",
1726                             zc.zc_name, zc.zc_value);
1727                 }
1728                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1729                 if (err == 0)
1730                         changelist_rename(clp, name, newname);
1731                 if (err && flags->verbose) {
1732                         (void) printf("failed (%u) - "
1733                             "will try again on next pass\n", errno);
1734                 }
1735                 err = EAGAIN;
1736         } else if (flags->verbose) {
1737                 if (err == 0)
1738                         (void) printf("success\n");
1739                 else
1740                         (void) printf("failed (%u)\n", errno);
1741         }
1742 
1743         (void) changelist_postfix(clp);
1744         changelist_free(clp);
1745 
1746         return (err);
1747 }
1748 
1749 static int
1750 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1751     char *newname, recvflags_t *flags)
1752 {
1753         zfs_cmd_t zc = { 0 };
1754         int err = 0;
1755         prop_changelist_t *clp;
1756         zfs_handle_t *zhp;
1757         boolean_t defer = B_FALSE;
1758         int spa_version;
1759 
1760         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1761         if (zhp == NULL)
1762                 return (-1);
1763         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1764             flags->force ? MS_FORCE : 0);
1765         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1766             zfs_spa_version(zhp, &spa_version) == 0 &&
1767             spa_version >= SPA_VERSION_USERREFS)
1768                 defer = B_TRUE;
1769         zfs_close(zhp);
1770         if (clp == NULL)
1771                 return (-1);
1772         err = changelist_prefix(clp);
1773         if (err)
1774                 return (err);
1775 
1776         zc.zc_objset_type = DMU_OST_ZFS;
1777         zc.zc_defer_destroy = defer;
1778         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1779 
1780         if (flags->verbose)
1781                 (void) printf("attempting destroy %s\n", zc.zc_name);
1782         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1783         if (err == 0) {
1784                 if (flags->verbose)
1785                         (void) printf("success\n");
1786                 changelist_remove(clp, zc.zc_name);
1787         }
1788 
1789         (void) changelist_postfix(clp);
1790         changelist_free(clp);
1791 
1792         /*
1793          * Deferred destroy might destroy the snapshot or only mark it to be
1794          * destroyed later, and it returns success in either case.
1795          */
1796         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1797             ZFS_TYPE_SNAPSHOT))) {
1798                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1799         }
1800 
1801         return (err);
1802 }
1803 
1804 typedef struct guid_to_name_data {
1805         uint64_t guid;
1806         char *name;
1807         char *skip;
1808 } guid_to_name_data_t;
1809 
1810 static int
1811 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1812 {
1813         guid_to_name_data_t *gtnd = arg;
1814         int err;
1815 
1816         if (gtnd->skip != NULL &&
1817             strcmp(zhp->zfs_name, gtnd->skip) == 0) {
1818                 return (0);
1819         }
1820 
1821         if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1822                 (void) strcpy(gtnd->name, zhp->zfs_name);
1823                 zfs_close(zhp);
1824                 return (EEXIST);
1825         }
1826 
1827         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1828         zfs_close(zhp);
1829         return (err);
1830 }
1831 
1832 /*
1833  * Attempt to find the local dataset associated with this guid.  In the case of
1834  * multiple matches, we attempt to find the "best" match by searching
1835  * progressively larger portions of the hierarchy.  This allows one to send a
1836  * tree of datasets individually and guarantee that we will find the source
1837  * guid within that hierarchy, even if there are multiple matches elsewhere.
1838  */
1839 static int
1840 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1841     char *name)
1842 {
1843         /* exhaustive search all local snapshots */
1844         char pname[ZFS_MAXNAMELEN];
1845         guid_to_name_data_t gtnd;
1846         int err = 0;
1847         zfs_handle_t *zhp;
1848         char *cp;
1849 
1850         gtnd.guid = guid;
1851         gtnd.name = name;
1852         gtnd.skip = NULL;
1853 
1854         (void) strlcpy(pname, parent, sizeof (pname));
1855 
1856         /*
1857          * Search progressively larger portions of the hierarchy.  This will
1858          * select the "most local" version of the origin snapshot in the case
1859          * that there are multiple matching snapshots in the system.
1860          */
1861         while ((cp = strrchr(pname, '/')) != NULL) {
1862 
1863                 /* Chop off the last component and open the parent */
1864                 *cp = '\0';
1865                 zhp = make_dataset_handle(hdl, pname);
1866 
1867                 if (zhp == NULL)
1868                         continue;
1869 
1870                 err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1871                 zfs_close(zhp);
1872                 if (err == EEXIST)
1873                         return (0);
1874 
1875                 /*
1876                  * Remember the dataset that we already searched, so we
1877                  * skip it next time through.
1878                  */
1879                 gtnd.skip = pname;
1880         }
1881 
1882         return (ENOENT);
1883 }
1884 
1885 /*
1886  * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
1887  * guid1 is after guid2.
1888  */
1889 static int
1890 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1891     uint64_t guid1, uint64_t guid2)
1892 {
1893         nvlist_t *nvfs;
1894         char *fsname, *snapname;
1895         char buf[ZFS_MAXNAMELEN];
1896         int rv;
1897         zfs_handle_t *guid1hdl, *guid2hdl;
1898         uint64_t create1, create2;
1899 
1900         if (guid2 == 0)
1901                 return (0);
1902         if (guid1 == 0)
1903                 return (1);
1904 
1905         nvfs = fsavl_find(avl, guid1, &snapname);
1906         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1907         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1908         guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1909         if (guid1hdl == NULL)
1910                 return (-1);
1911 
1912         nvfs = fsavl_find(avl, guid2, &snapname);
1913         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1914         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1915         guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1916         if (guid2hdl == NULL) {
1917                 zfs_close(guid1hdl);
1918                 return (-1);
1919         }
1920 
1921         create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
1922         create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
1923 
1924         if (create1 < create2)
1925                 rv = -1;
1926         else if (create1 > create2)
1927                 rv = +1;
1928         else
1929                 rv = 0;
1930 
1931         zfs_close(guid1hdl);
1932         zfs_close(guid2hdl);
1933 
1934         return (rv);
1935 }
1936 
1937 static int
1938 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
1939     recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
1940     nvlist_t *renamed)
1941 {
1942         nvlist_t *local_nv;
1943         avl_tree_t *local_avl;
1944         nvpair_t *fselem, *nextfselem;
1945         char *fromsnap;
1946         char newname[ZFS_MAXNAMELEN];
1947         int error;
1948         boolean_t needagain, progress, recursive;
1949         char *s1, *s2;
1950 
1951         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
1952 
1953         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
1954             ENOENT);
1955 
1956         if (flags->dryrun)
1957                 return (0);
1958 
1959 again:
1960         needagain = progress = B_FALSE;
1961 
1962         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
1963             recursive, &local_nv, &local_avl)) != 0)
1964                 return (error);
1965 
1966         /*
1967          * Process deletes and renames
1968          */
1969         for (fselem = nvlist_next_nvpair(local_nv, NULL);
1970             fselem; fselem = nextfselem) {
1971                 nvlist_t *nvfs, *snaps;
1972                 nvlist_t *stream_nvfs = NULL;
1973                 nvpair_t *snapelem, *nextsnapelem;
1974                 uint64_t fromguid = 0;
1975                 uint64_t originguid = 0;
1976                 uint64_t stream_originguid = 0;
1977                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
1978                 char *fsname, *stream_fsname;
1979 
1980                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
1981 
1982                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
1983                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
1984                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1985                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
1986                     &parent_fromsnap_guid));
1987                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
1988 
1989                 /*
1990                  * First find the stream's fs, so we can check for
1991                  * a different origin (due to "zfs promote")
1992                  */
1993                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1994                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
1995                         uint64_t thisguid;
1996 
1997                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1998                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
1999 
2000                         if (stream_nvfs != NULL)
2001                                 break;
2002                 }
2003 
2004                 /* check for promote */
2005                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2006                     &stream_originguid);
2007                 if (stream_nvfs && originguid != stream_originguid) {
2008                         switch (created_before(hdl, local_avl,
2009                             stream_originguid, originguid)) {
2010                         case 1: {
2011                                 /* promote it! */
2012                                 zfs_cmd_t zc = { 0 };
2013                                 nvlist_t *origin_nvfs;
2014                                 char *origin_fsname;
2015 
2016                                 if (flags->verbose)
2017                                         (void) printf("promoting %s\n", fsname);
2018 
2019                                 origin_nvfs = fsavl_find(local_avl, originguid,
2020                                     NULL);
2021                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2022                                     "name", &origin_fsname));
2023                                 (void) strlcpy(zc.zc_value, origin_fsname,
2024                                     sizeof (zc.zc_value));
2025                                 (void) strlcpy(zc.zc_name, fsname,
2026                                     sizeof (zc.zc_name));
2027                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2028                                 if (error == 0)
2029                                         progress = B_TRUE;
2030                                 break;
2031                         }
2032                         default:
2033                                 break;
2034                         case -1:
2035                                 fsavl_destroy(local_avl);
2036                                 nvlist_free(local_nv);
2037                                 return (-1);
2038                         }
2039                         /*
2040                          * We had/have the wrong origin, therefore our
2041                          * list of snapshots is wrong.  Need to handle
2042                          * them on the next pass.
2043                          */
2044                         needagain = B_TRUE;
2045                         continue;
2046                 }
2047 
2048                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2049                     snapelem; snapelem = nextsnapelem) {
2050                         uint64_t thisguid;
2051                         char *stream_snapname;
2052                         nvlist_t *found, *props;
2053 
2054                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2055 
2056                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2057                         found = fsavl_find(stream_avl, thisguid,
2058                             &stream_snapname);
2059 
2060                         /* check for delete */
2061                         if (found == NULL) {
2062                                 char name[ZFS_MAXNAMELEN];
2063 
2064                                 if (!flags->force)
2065                                         continue;
2066 
2067                                 (void) snprintf(name, sizeof (name), "%s@%s",
2068                                     fsname, nvpair_name(snapelem));
2069 
2070                                 error = recv_destroy(hdl, name,
2071                                     strlen(fsname)+1, newname, flags);
2072                                 if (error)
2073                                         needagain = B_TRUE;
2074                                 else
2075                                         progress = B_TRUE;
2076                                 continue;
2077                         }
2078 
2079                         stream_nvfs = found;
2080 
2081                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2082                             &props) && 0 == nvlist_lookup_nvlist(props,
2083                             stream_snapname, &props)) {
2084                                 zfs_cmd_t zc = { 0 };
2085 
2086                                 zc.zc_cookie = B_TRUE; /* received */
2087                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2088                                     "%s@%s", fsname, nvpair_name(snapelem));
2089                                 if (zcmd_write_src_nvlist(hdl, &zc,
2090                                     props) == 0) {
2091                                         (void) zfs_ioctl(hdl,
2092                                             ZFS_IOC_SET_PROP, &zc);
2093                                         zcmd_free_nvlists(&zc);
2094                                 }
2095                         }
2096 
2097                         /* check for different snapname */
2098                         if (strcmp(nvpair_name(snapelem),
2099                             stream_snapname) != 0) {
2100                                 char name[ZFS_MAXNAMELEN];
2101                                 char tryname[ZFS_MAXNAMELEN];
2102 
2103                                 (void) snprintf(name, sizeof (name), "%s@%s",
2104                                     fsname, nvpair_name(snapelem));
2105                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
2106                                     fsname, stream_snapname);
2107 
2108                                 error = recv_rename(hdl, name, tryname,
2109                                     strlen(fsname)+1, newname, flags);
2110                                 if (error)
2111                                         needagain = B_TRUE;
2112                                 else
2113                                         progress = B_TRUE;
2114                         }
2115 
2116                         if (strcmp(stream_snapname, fromsnap) == 0)
2117                                 fromguid = thisguid;
2118                 }
2119 
2120                 /* check for delete */
2121                 if (stream_nvfs == NULL) {
2122                         if (!flags->force)
2123                                 continue;
2124 
2125                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2126                             newname, flags);
2127                         if (error)
2128                                 needagain = B_TRUE;
2129                         else
2130                                 progress = B_TRUE;
2131                         continue;
2132                 }
2133 
2134                 if (fromguid == 0) {
2135                         if (flags->verbose) {
2136                                 (void) printf("local fs %s does not have "
2137                                     "fromsnap (%s in stream); must have "
2138                                     "been deleted locally; ignoring\n",
2139                                     fsname, fromsnap);
2140                         }
2141                         continue;
2142                 }
2143 
2144                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2145                     "name", &stream_fsname));
2146                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2147                     "parentfromsnap", &stream_parent_fromsnap_guid));
2148 
2149                 s1 = strrchr(fsname, '/');
2150                 s2 = strrchr(stream_fsname, '/');
2151 
2152                 /*
2153                  * Check for rename. If the exact receive path is specified, it
2154                  * does not count as a rename, but we still need to check the
2155                  * datasets beneath it.
2156                  */
2157                 if ((stream_parent_fromsnap_guid != 0 &&
2158                     parent_fromsnap_guid != 0 &&
2159                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2160                     ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2161                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2162                         nvlist_t *parent;
2163                         char tryname[ZFS_MAXNAMELEN];
2164 
2165                         parent = fsavl_find(local_avl,
2166                             stream_parent_fromsnap_guid, NULL);
2167                         /*
2168                          * NB: parent might not be found if we used the
2169                          * tosnap for stream_parent_fromsnap_guid,
2170                          * because the parent is a newly-created fs;
2171                          * we'll be able to rename it after we recv the
2172                          * new fs.
2173                          */
2174                         if (parent != NULL) {
2175                                 char *pname;
2176 
2177                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
2178                                     &pname));
2179                                 (void) snprintf(tryname, sizeof (tryname),
2180                                     "%s%s", pname, strrchr(stream_fsname, '/'));
2181                         } else {
2182                                 tryname[0] = '\0';
2183                                 if (flags->verbose) {
2184                                         (void) printf("local fs %s new parent "
2185                                             "not found\n", fsname);
2186                                 }
2187                         }
2188 
2189                         newname[0] = '\0';
2190 
2191                         error = recv_rename(hdl, fsname, tryname,
2192                             strlen(tofs)+1, newname, flags);
2193 
2194                         if (renamed != NULL && newname[0] != '\0') {
2195                                 VERIFY(0 == nvlist_add_boolean(renamed,
2196                                     newname));
2197                         }
2198 
2199                         if (error)
2200                                 needagain = B_TRUE;
2201                         else
2202                                 progress = B_TRUE;
2203                 }
2204         }
2205 
2206         fsavl_destroy(local_avl);
2207         nvlist_free(local_nv);
2208 
2209         if (needagain && progress) {
2210                 /* do another pass to fix up temporary names */
2211                 if (flags->verbose)
2212                         (void) printf("another pass:\n");
2213                 goto again;
2214         }
2215 
2216         return (needagain);
2217 }
2218 
2219 static int
2220 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2221     recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2222     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2223 {
2224         nvlist_t *stream_nv = NULL;
2225         avl_tree_t *stream_avl = NULL;
2226         char *fromsnap = NULL;
2227         char *cp;
2228         char tofs[ZFS_MAXNAMELEN];
2229         char sendfs[ZFS_MAXNAMELEN];
2230         char errbuf[1024];
2231         dmu_replay_record_t drre;
2232         int error;
2233         boolean_t anyerr = B_FALSE;
2234         boolean_t softerr = B_FALSE;
2235         boolean_t recursive;
2236 
2237         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2238             "cannot receive"));
2239 
2240         assert(drr->drr_type == DRR_BEGIN);
2241         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2242         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2243             DMU_COMPOUNDSTREAM);
2244 
2245         /*
2246          * Read in the nvlist from the stream.
2247          */
2248         if (drr->drr_payloadlen != 0) {
2249                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2250                     &stream_nv, flags->byteswap, zc);
2251                 if (error) {
2252                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2253                         goto out;
2254                 }
2255         }
2256 
2257         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2258             ENOENT);
2259 
2260         if (recursive && strchr(destname, '@')) {
2261                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2262                     "cannot specify snapshot name for multi-snapshot stream"));
2263                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2264                 goto out;
2265         }
2266 
2267         /*
2268          * Read in the end record and verify checksum.
2269          */
2270         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2271             flags->byteswap, NULL)))
2272                 goto out;
2273         if (flags->byteswap) {
2274                 drre.drr_type = BSWAP_32(drre.drr_type);
2275                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2276                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2277                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2278                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2279                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2280                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2281                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2282                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2283         }
2284         if (drre.drr_type != DRR_END) {
2285                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2286                 goto out;
2287         }
2288         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2289                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2290                     "incorrect header checksum"));
2291                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2292                 goto out;
2293         }
2294 
2295         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2296 
2297         if (drr->drr_payloadlen != 0) {
2298                 nvlist_t *stream_fss;
2299 
2300                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2301                     &stream_fss));
2302                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2303                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2304                             "couldn't allocate avl tree"));
2305                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2306                         goto out;
2307                 }
2308 
2309                 if (fromsnap != NULL) {
2310                         nvlist_t *renamed = NULL;
2311                         nvpair_t *pair = NULL;
2312 
2313                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2314                         if (flags->isprefix) {
2315                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2316                                 int i;
2317 
2318                                 if (flags->istail) {
2319                                         cp = strrchr(drrb->drr_toname, '/');
2320                                         if (cp == NULL) {
2321                                                 (void) strlcat(tofs, "/",
2322                                                     ZFS_MAXNAMELEN);
2323                                                 i = 0;
2324                                         } else {
2325                                                 i = (cp - drrb->drr_toname);
2326                                         }
2327                                 } else {
2328                                         i = strcspn(drrb->drr_toname, "/@");
2329                                 }
2330                                 /* zfs_receive_one() will create_parents() */
2331                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2332                                     ZFS_MAXNAMELEN);
2333                                 *strchr(tofs, '@') = '\0';
2334                         }
2335 
2336                         if (recursive && !flags->dryrun && !flags->nomount) {
2337                                 VERIFY(0 == nvlist_alloc(&renamed,
2338                                     NV_UNIQUE_NAME, 0));
2339                         }
2340 
2341                         softerr = recv_incremental_replication(hdl, tofs, flags,
2342                             stream_nv, stream_avl, renamed);
2343 
2344                         /* Unmount renamed filesystems before receiving. */
2345                         while ((pair = nvlist_next_nvpair(renamed,
2346                             pair)) != NULL) {
2347                                 zfs_handle_t *zhp;
2348                                 prop_changelist_t *clp = NULL;
2349 
2350                                 zhp = zfs_open(hdl, nvpair_name(pair),
2351                                     ZFS_TYPE_FILESYSTEM);
2352                                 if (zhp != NULL) {
2353                                         clp = changelist_gather(zhp,
2354                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2355                                         zfs_close(zhp);
2356                                         if (clp != NULL) {
2357                                                 softerr |=
2358                                                     changelist_prefix(clp);
2359                                                 changelist_free(clp);
2360                                         }
2361                                 }
2362                         }
2363 
2364                         nvlist_free(renamed);
2365                 }
2366         }
2367 
2368         /*
2369          * Get the fs specified by the first path in the stream (the top level
2370          * specified by 'zfs send') and pass it to each invocation of
2371          * zfs_receive_one().
2372          */
2373         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2374             ZFS_MAXNAMELEN);
2375         if ((cp = strchr(sendfs, '@')) != NULL)
2376                 *cp = '\0';
2377 
2378         /* Finally, receive each contained stream */
2379         do {
2380                 /*
2381                  * we should figure out if it has a recoverable
2382                  * error, in which case do a recv_skip() and drive on.
2383                  * Note, if we fail due to already having this guid,
2384                  * zfs_receive_one() will take care of it (ie,
2385                  * recv_skip() and return 0).
2386                  */
2387                 error = zfs_receive_impl(hdl, destname, flags, fd,
2388                     sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2389                     action_handlep);
2390                 if (error == ENODATA) {
2391                         error = 0;
2392                         break;
2393                 }
2394                 anyerr |= error;
2395         } while (error == 0);
2396 
2397         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2398                 /*
2399                  * Now that we have the fs's they sent us, try the
2400                  * renames again.
2401                  */
2402                 softerr = recv_incremental_replication(hdl, tofs, flags,
2403                     stream_nv, stream_avl, NULL);
2404         }
2405 
2406 out:
2407         fsavl_destroy(stream_avl);
2408         if (stream_nv)
2409                 nvlist_free(stream_nv);
2410         if (softerr)
2411                 error = -2;
2412         if (anyerr)
2413                 error = -1;
2414         return (error);
2415 }
2416 
2417 static void
2418 trunc_prop_errs(int truncated)
2419 {
2420         ASSERT(truncated != 0);
2421 
2422         if (truncated == 1)
2423                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2424                     "1 more property could not be set\n"));
2425         else
2426                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2427                     "%d more properties could not be set\n"), truncated);
2428 }
2429 
2430 static int
2431 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2432 {
2433         dmu_replay_record_t *drr;
2434         void *buf = malloc(1<<20);
2435         char errbuf[1024];
2436 
2437         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2438             "cannot receive:"));
2439 
2440         /* XXX would be great to use lseek if possible... */
2441         drr = buf;
2442 
2443         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2444             byteswap, NULL) == 0) {
2445                 if (byteswap)
2446                         drr->drr_type = BSWAP_32(drr->drr_type);
2447 
2448                 switch (drr->drr_type) {
2449                 case DRR_BEGIN:
2450                         /* NB: not to be used on v2 stream packages */
2451                         if (drr->drr_payloadlen != 0) {
2452                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2453                                     "invalid substream header"));
2454                                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2455                         }
2456                         break;
2457 
2458                 case DRR_END:
2459                         free(buf);
2460                         return (0);
2461 
2462                 case DRR_OBJECT:
2463                         if (byteswap) {
2464                                 drr->drr_u.drr_object.drr_bonuslen =
2465                                     BSWAP_32(drr->drr_u.drr_object.
2466                                     drr_bonuslen);
2467                         }
2468                         (void) recv_read(hdl, fd, buf,
2469                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2470                             B_FALSE, NULL);
2471                         break;
2472 
2473                 case DRR_WRITE:
2474                         if (byteswap) {
2475                                 drr->drr_u.drr_write.drr_length =
2476                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
2477                         }
2478                         (void) recv_read(hdl, fd, buf,
2479                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2480                         break;
2481                 case DRR_SPILL:
2482                         if (byteswap) {
2483                                 drr->drr_u.drr_write.drr_length =
2484                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
2485                         }
2486                         (void) recv_read(hdl, fd, buf,
2487                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2488                         break;
2489                 case DRR_WRITE_BYREF:
2490                 case DRR_FREEOBJECTS:
2491                 case DRR_FREE:
2492                         break;
2493 
2494                 default:
2495                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2496                             "invalid record type"));
2497                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2498                 }
2499         }
2500 
2501         free(buf);
2502         return (-1);
2503 }
2504 
2505 /*
2506  * Restores a backup of tosnap from the file descriptor specified by infd.
2507  */
2508 static int
2509 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2510     recvflags_t *flags, dmu_replay_record_t *drr,
2511     dmu_replay_record_t *drr_noswap, const char *sendfs,
2512     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2513     uint64_t *action_handlep)
2514 {
2515         zfs_cmd_t zc = { 0 };
2516         time_t begin_time;
2517         int ioctl_err, ioctl_errno, err;
2518         char *cp;
2519         struct drr_begin *drrb = &drr->drr_u.drr_begin;
2520         char errbuf[1024];
2521         char prop_errbuf[1024];
2522         const char *chopprefix;
2523         boolean_t newfs = B_FALSE;
2524         boolean_t stream_wantsnewfs;
2525         uint64_t parent_snapguid = 0;
2526         prop_changelist_t *clp = NULL;
2527         nvlist_t *snapprops_nvlist = NULL;
2528         zprop_errflags_t prop_errflags;
2529         boolean_t recursive;
2530 
2531         begin_time = time(NULL);
2532 
2533         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2534             "cannot receive"));
2535 
2536         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2537             ENOENT);
2538 
2539         if (stream_avl != NULL) {
2540                 char *snapname;
2541                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2542                     &snapname);
2543                 nvlist_t *props;
2544                 int ret;
2545 
2546                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2547                     &parent_snapguid);
2548                 err = nvlist_lookup_nvlist(fs, "props", &props);
2549                 if (err)
2550                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2551 
2552                 if (flags->canmountoff) {
2553                         VERIFY(0 == nvlist_add_uint64(props,
2554                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2555                 }
2556                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2557                 if (err)
2558                         nvlist_free(props);
2559 
2560                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2561                         VERIFY(0 == nvlist_lookup_nvlist(props,
2562                             snapname, &snapprops_nvlist));
2563                 }
2564 
2565                 if (ret != 0)
2566                         return (-1);
2567         }
2568 
2569         cp = NULL;
2570 
2571         /*
2572          * Determine how much of the snapshot name stored in the stream
2573          * we are going to tack on to the name they specified on the
2574          * command line, and how much we are going to chop off.
2575          *
2576          * If they specified a snapshot, chop the entire name stored in
2577          * the stream.
2578          */
2579         if (flags->istail) {
2580                 /*
2581                  * A filesystem was specified with -e. We want to tack on only
2582                  * the tail of the sent snapshot path.
2583                  */
2584                 if (strchr(tosnap, '@')) {
2585                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2586                             "argument - snapshot not allowed with -e"));
2587                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2588                 }
2589 
2590                 chopprefix = strrchr(sendfs, '/');
2591 
2592                 if (chopprefix == NULL) {
2593                         /*
2594                          * The tail is the poolname, so we need to
2595                          * prepend a path separator.
2596                          */
2597                         int len = strlen(drrb->drr_toname);
2598                         cp = malloc(len + 2);
2599                         cp[0] = '/';
2600                         (void) strcpy(&cp[1], drrb->drr_toname);
2601                         chopprefix = cp;
2602                 } else {
2603                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2604                 }
2605         } else if (flags->isprefix) {
2606                 /*
2607                  * A filesystem was specified with -d. We want to tack on
2608                  * everything but the first element of the sent snapshot path
2609                  * (all but the pool name).
2610                  */
2611                 if (strchr(tosnap, '@')) {
2612                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2613                             "argument - snapshot not allowed with -d"));
2614                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2615                 }
2616 
2617                 chopprefix = strchr(drrb->drr_toname, '/');
2618                 if (chopprefix == NULL)
2619                         chopprefix = strchr(drrb->drr_toname, '@');
2620         } else if (strchr(tosnap, '@') == NULL) {
2621                 /*
2622                  * If a filesystem was specified without -d or -e, we want to
2623                  * tack on everything after the fs specified by 'zfs send'.
2624                  */
2625                 chopprefix = drrb->drr_toname + strlen(sendfs);
2626         } else {
2627                 /* A snapshot was specified as an exact path (no -d or -e). */
2628                 if (recursive) {
2629                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2630                             "cannot specify snapshot name for multi-snapshot "
2631                             "stream"));
2632                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2633                 }
2634                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2635         }
2636 
2637         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2638         ASSERT(chopprefix > drrb->drr_toname);
2639         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2640         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2641             chopprefix[0] == '\0');
2642 
2643         /*
2644          * Determine name of destination snapshot, store in zc_value.
2645          */
2646         (void) strcpy(zc.zc_top_ds, tosnap);
2647         (void) strcpy(zc.zc_value, tosnap);
2648         (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2649         free(cp);
2650         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2651                 zcmd_free_nvlists(&zc);
2652                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2653         }
2654 
2655         /*
2656          * Determine the name of the origin snapshot, store in zc_string.
2657          */
2658         if (drrb->drr_flags & DRR_FLAG_CLONE) {
2659                 if (guid_to_name(hdl, zc.zc_value,
2660                     drrb->drr_fromguid, zc.zc_string) != 0) {
2661                         zcmd_free_nvlists(&zc);
2662                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2663                             "local origin for clone %s does not exist"),
2664                             zc.zc_value);
2665                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2666                 }
2667                 if (flags->verbose)
2668                         (void) printf("found clone origin %s\n", zc.zc_string);
2669         }
2670 
2671         stream_wantsnewfs = (drrb->drr_fromguid == NULL ||
2672             (drrb->drr_flags & DRR_FLAG_CLONE));
2673 
2674         if (stream_wantsnewfs) {
2675                 /*
2676                  * if the parent fs does not exist, look for it based on
2677                  * the parent snap GUID
2678                  */
2679                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2680                     "cannot receive new filesystem stream"));
2681 
2682                 (void) strcpy(zc.zc_name, zc.zc_value);
2683                 cp = strrchr(zc.zc_name, '/');
2684                 if (cp)
2685                         *cp = '\0';
2686                 if (cp &&
2687                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2688                         char suffix[ZFS_MAXNAMELEN];
2689                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2690                         if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
2691                             zc.zc_value) == 0) {
2692                                 *strchr(zc.zc_value, '@') = '\0';
2693                                 (void) strcat(zc.zc_value, suffix);
2694                         }
2695                 }
2696         } else {
2697                 /*
2698                  * if the fs does not exist, look for it based on the
2699                  * fromsnap GUID
2700                  */
2701                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2702                     "cannot receive incremental stream"));
2703 
2704                 (void) strcpy(zc.zc_name, zc.zc_value);
2705                 *strchr(zc.zc_name, '@') = '\0';
2706 
2707                 /*
2708                  * If the exact receive path was specified and this is the
2709                  * topmost path in the stream, then if the fs does not exist we
2710                  * should look no further.
2711                  */
2712                 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
2713                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2714                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2715                         char snap[ZFS_MAXNAMELEN];
2716                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
2717                         if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
2718                             zc.zc_value) == 0) {
2719                                 *strchr(zc.zc_value, '@') = '\0';
2720                                 (void) strcat(zc.zc_value, snap);
2721                         }
2722                 }
2723         }
2724 
2725         (void) strcpy(zc.zc_name, zc.zc_value);
2726         *strchr(zc.zc_name, '@') = '\0';
2727 
2728         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2729                 zfs_handle_t *zhp;
2730 
2731                 /*
2732                  * Destination fs exists.  Therefore this should either
2733                  * be an incremental, or the stream specifies a new fs
2734                  * (full stream or clone) and they want us to blow it
2735                  * away (and have therefore specified -F and removed any
2736                  * snapshots).
2737                  */
2738                 if (stream_wantsnewfs) {
2739                         if (!flags->force) {
2740                                 zcmd_free_nvlists(&zc);
2741                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2742                                     "destination '%s' exists\n"
2743                                     "must specify -F to overwrite it"),
2744                                     zc.zc_name);
2745                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2746                         }
2747                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2748                             &zc) == 0) {
2749                                 zcmd_free_nvlists(&zc);
2750                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2751                                     "destination has snapshots (eg. %s)\n"
2752                                     "must destroy them to overwrite it"),
2753                                     zc.zc_name);
2754                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2755                         }
2756                 }
2757 
2758                 if ((zhp = zfs_open(hdl, zc.zc_name,
2759                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2760                         zcmd_free_nvlists(&zc);
2761                         return (-1);
2762                 }
2763 
2764                 if (stream_wantsnewfs &&
2765                     zhp->zfs_dmustats.dds_origin[0]) {
2766                         zcmd_free_nvlists(&zc);
2767                         zfs_close(zhp);
2768                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2769                             "destination '%s' is a clone\n"
2770                             "must destroy it to overwrite it"),
2771                             zc.zc_name);
2772                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2773                 }
2774 
2775                 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2776                     stream_wantsnewfs) {
2777                         /* We can't do online recv in this case */
2778                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2779                         if (clp == NULL) {
2780                                 zfs_close(zhp);
2781                                 zcmd_free_nvlists(&zc);
2782                                 return (-1);
2783                         }
2784                         if (changelist_prefix(clp) != 0) {
2785                                 changelist_free(clp);
2786                                 zfs_close(zhp);
2787                                 zcmd_free_nvlists(&zc);
2788                                 return (-1);
2789                         }
2790                 }
2791                 zfs_close(zhp);
2792         } else {
2793                 /*
2794                  * Destination filesystem does not exist.  Therefore we better
2795                  * be creating a new filesystem (either from a full backup, or
2796                  * a clone).  It would therefore be invalid if the user
2797                  * specified only the pool name (i.e. if the destination name
2798                  * contained no slash character).
2799                  */
2800                 if (!stream_wantsnewfs ||
2801                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
2802                         zcmd_free_nvlists(&zc);
2803                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2804                             "destination '%s' does not exist"), zc.zc_name);
2805                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2806                 }
2807 
2808                 /*
2809                  * Trim off the final dataset component so we perform the
2810                  * recvbackup ioctl to the filesystems's parent.
2811                  */
2812                 *cp = '\0';
2813 
2814                 if (flags->isprefix && !flags->istail && !flags->dryrun &&
2815                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2816                         zcmd_free_nvlists(&zc);
2817                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2818                 }
2819 
2820                 newfs = B_TRUE;
2821         }
2822 
2823         zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2824         zc.zc_cookie = infd;
2825         zc.zc_guid = flags->force;
2826         if (flags->verbose) {
2827                 (void) printf("%s %s stream of %s into %s\n",
2828                     flags->dryrun ? "would receive" : "receiving",
2829                     drrb->drr_fromguid ? "incremental" : "full",
2830                     drrb->drr_toname, zc.zc_value);
2831                 (void) fflush(stdout);
2832         }
2833 
2834         if (flags->dryrun) {
2835                 zcmd_free_nvlists(&zc);
2836                 return (recv_skip(hdl, infd, flags->byteswap));
2837         }
2838 
2839         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2840         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2841         zc.zc_cleanup_fd = cleanup_fd;
2842         zc.zc_action_handle = *action_handlep;
2843 
2844         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2845         ioctl_errno = errno;
2846         prop_errflags = (zprop_errflags_t)zc.zc_obj;
2847 
2848         if (err == 0) {
2849                 nvlist_t *prop_errors;
2850                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2851                     zc.zc_nvlist_dst_size, &prop_errors, 0));
2852 
2853                 nvpair_t *prop_err = NULL;
2854 
2855                 while ((prop_err = nvlist_next_nvpair(prop_errors,
2856                     prop_err)) != NULL) {
2857                         char tbuf[1024];
2858                         zfs_prop_t prop;
2859                         int intval;
2860 
2861                         prop = zfs_name_to_prop(nvpair_name(prop_err));
2862                         (void) nvpair_value_int32(prop_err, &intval);
2863                         if (strcmp(nvpair_name(prop_err),
2864                             ZPROP_N_MORE_ERRORS) == 0) {
2865                                 trunc_prop_errs(intval);
2866                                 break;
2867                         } else {
2868                                 (void) snprintf(tbuf, sizeof (tbuf),
2869                                     dgettext(TEXT_DOMAIN,
2870                                     "cannot receive %s property on %s"),
2871                                     nvpair_name(prop_err), zc.zc_name);
2872                                 zfs_setprop_error(hdl, prop, intval, tbuf);
2873                         }
2874                 }
2875                 nvlist_free(prop_errors);
2876         }
2877 
2878         zc.zc_nvlist_dst = 0;
2879         zc.zc_nvlist_dst_size = 0;
2880         zcmd_free_nvlists(&zc);
2881 
2882         if (err == 0 && snapprops_nvlist) {
2883                 zfs_cmd_t zc2 = { 0 };
2884 
2885                 (void) strcpy(zc2.zc_name, zc.zc_value);
2886                 zc2.zc_cookie = B_TRUE; /* received */
2887                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
2888                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
2889                         zcmd_free_nvlists(&zc2);
2890                 }
2891         }
2892 
2893         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
2894                 /*
2895                  * It may be that this snapshot already exists,
2896                  * in which case we want to consume & ignore it
2897                  * rather than failing.
2898                  */
2899                 avl_tree_t *local_avl;
2900                 nvlist_t *local_nv, *fs;
2901                 cp = strchr(zc.zc_value, '@');
2902 
2903                 /*
2904                  * XXX Do this faster by just iterating over snaps in
2905                  * this fs.  Also if zc_value does not exist, we will
2906                  * get a strange "does not exist" error message.
2907                  */
2908                 *cp = '\0';
2909                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
2910                     &local_nv, &local_avl) == 0) {
2911                         *cp = '@';
2912                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
2913                         fsavl_destroy(local_avl);
2914                         nvlist_free(local_nv);
2915 
2916                         if (fs != NULL) {
2917                                 if (flags->verbose) {
2918                                         (void) printf("snap %s already exists; "
2919                                             "ignoring\n", zc.zc_value);
2920                                 }
2921                                 err = ioctl_err = recv_skip(hdl, infd,
2922                                     flags->byteswap);
2923                         }
2924                 }
2925                 *cp = '@';
2926         }
2927 
2928         if (ioctl_err != 0) {
2929                 switch (ioctl_errno) {
2930                 case ENODEV:
2931                         cp = strchr(zc.zc_value, '@');
2932                         *cp = '\0';
2933                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2934                             "most recent snapshot of %s does not\n"
2935                             "match incremental source"), zc.zc_value);
2936                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2937                         *cp = '@';
2938                         break;
2939                 case ETXTBSY:
2940                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2941                             "destination %s has been modified\n"
2942                             "since most recent snapshot"), zc.zc_name);
2943                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2944                         break;
2945                 case EEXIST:
2946                         cp = strchr(zc.zc_value, '@');
2947                         if (newfs) {
2948                                 /* it's the containing fs that exists */
2949                                 *cp = '\0';
2950                         }
2951                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2952                             "destination already exists"));
2953                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
2954                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
2955                             zc.zc_value);
2956                         *cp = '@';
2957                         break;
2958                 case EINVAL:
2959                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2960                         break;
2961                 case ECKSUM:
2962                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2963                             "invalid stream (checksum mismatch)"));
2964                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2965                         break;
2966                 case ENOTSUP:
2967                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2968                             "pool must be upgraded to receive this stream."));
2969                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
2970                         break;
2971                 case EDQUOT:
2972                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2973                             "destination %s space quota exceeded"), zc.zc_name);
2974                         (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
2975                         break;
2976                 default:
2977                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
2978                 }
2979         }
2980 
2981         /*
2982          * Mount the target filesystem (if created).  Also mount any
2983          * children of the target filesystem if we did a replication
2984          * receive (indicated by stream_avl being non-NULL).
2985          */
2986         cp = strchr(zc.zc_value, '@');
2987         if (cp && (ioctl_err == 0 || !newfs)) {
2988                 zfs_handle_t *h;
2989 
2990                 *cp = '\0';
2991                 h = zfs_open(hdl, zc.zc_value,
2992                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
2993                 if (h != NULL) {
2994                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
2995                                 *cp = '@';
2996                         } else if (newfs || stream_avl) {
2997                                 /*
2998                                  * Track the first/top of hierarchy fs,
2999                                  * for mounting and sharing later.
3000                                  */
3001                                 if (top_zfs && *top_zfs == NULL)
3002                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
3003                         }
3004                         zfs_close(h);
3005                 }
3006                 *cp = '@';
3007         }
3008 
3009         if (clp) {
3010                 err |= changelist_postfix(clp);
3011                 changelist_free(clp);
3012         }
3013 
3014         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3015                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3016                     "failed to clear unreceived properties on %s"),
3017                     zc.zc_name);
3018                 (void) fprintf(stderr, "\n");
3019         }
3020         if (prop_errflags & ZPROP_ERR_NORESTORE) {
3021                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3022                     "failed to restore original properties on %s"),
3023                     zc.zc_name);
3024                 (void) fprintf(stderr, "\n");
3025         }
3026 
3027         if (err || ioctl_err)
3028                 return (-1);
3029 
3030         *action_handlep = zc.zc_action_handle;
3031 
3032         if (flags->verbose) {
3033                 char buf1[64];
3034                 char buf2[64];
3035                 uint64_t bytes = zc.zc_cookie;
3036                 time_t delta = time(NULL) - begin_time;
3037                 if (delta == 0)
3038                         delta = 1;
3039                 zfs_nicenum(bytes, buf1, sizeof (buf1));
3040                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3041 
3042                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3043                     buf1, delta, buf2);
3044         }
3045 
3046         return (0);
3047 }
3048 
3049 static int
3050 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3051     int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl,
3052     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
3053 {
3054         int err;
3055         dmu_replay_record_t drr, drr_noswap;
3056         struct drr_begin *drrb = &drr.drr_u.drr_begin;
3057         char errbuf[1024];
3058         zio_cksum_t zcksum = { 0 };
3059         uint64_t featureflags;
3060         int hdrtype;
3061 
3062         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3063             "cannot receive"));
3064 
3065         if (flags->isprefix &&
3066             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3067                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3068                     "(%s) does not exist"), tosnap);
3069                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3070         }
3071 
3072         /* read in the BEGIN record */
3073         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3074             &zcksum)))
3075                 return (err);
3076 
3077         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3078                 /* It's the double end record at the end of a package */
3079                 return (ENODATA);
3080         }
3081 
3082         /* the kernel needs the non-byteswapped begin record */
3083         drr_noswap = drr;
3084 
3085         flags->byteswap = B_FALSE;
3086         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3087                 /*
3088                  * We computed the checksum in the wrong byteorder in
3089                  * recv_read() above; do it again correctly.
3090                  */
3091                 bzero(&zcksum, sizeof (zio_cksum_t));
3092                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3093                 flags->byteswap = B_TRUE;
3094 
3095                 drr.drr_type = BSWAP_32(drr.drr_type);
3096                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3097                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3098                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3099                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3100                 drrb->drr_type = BSWAP_32(drrb->drr_type);
3101                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3102                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3103                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3104         }
3105 
3106         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3107                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3108                     "stream (bad magic number)"));
3109                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3110         }
3111 
3112         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3113         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3114 
3115         if (!DMU_STREAM_SUPPORTED(featureflags) ||
3116             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3117                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3118                     "stream has unsupported feature, feature flags = %lx"),
3119                     featureflags);
3120                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3121         }
3122 
3123         if (strchr(drrb->drr_toname, '@') == NULL) {
3124                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3125                     "stream (bad snapshot name)"));
3126                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3127         }
3128 
3129         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3130                 char nonpackage_sendfs[ZFS_MAXNAMELEN];
3131                 if (sendfs == NULL) {
3132                         /*
3133                          * We were not called from zfs_receive_package(). Get
3134                          * the fs specified by 'zfs send'.
3135                          */
3136                         char *cp;
3137                         (void) strlcpy(nonpackage_sendfs,
3138                             drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
3139                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3140                                 *cp = '\0';
3141                         sendfs = nonpackage_sendfs;
3142                 }
3143                 return (zfs_receive_one(hdl, infd, tosnap, flags,
3144                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl,
3145                     top_zfs, cleanup_fd, action_handlep));
3146         } else {
3147                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3148                     DMU_COMPOUNDSTREAM);
3149                 return (zfs_receive_package(hdl, infd, tosnap, flags,
3150                     &drr, &zcksum, top_zfs, cleanup_fd, action_handlep));
3151         }
3152 }
3153 
3154 /*
3155  * Restores a backup of tosnap from the file descriptor specified by infd.
3156  * Return 0 on total success, -2 if some things couldn't be
3157  * destroyed/renamed/promoted, -1 if some things couldn't be received.
3158  * (-1 will override -2).
3159  */
3160 int
3161 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3162     int infd, avl_tree_t *stream_avl)
3163 {
3164         char *top_zfs = NULL;
3165         int err;
3166         int cleanup_fd;
3167         uint64_t action_handle = 0;
3168 
3169         cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3170         VERIFY(cleanup_fd >= 0);
3171 
3172         err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL,
3173             stream_avl, &top_zfs, cleanup_fd, &action_handle);
3174 
3175         VERIFY(0 == close(cleanup_fd));
3176 
3177         if (err == 0 && !flags->nomount && top_zfs) {
3178                 zfs_handle_t *zhp;
3179                 prop_changelist_t *clp;
3180 
3181                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3182                 if (zhp != NULL) {
3183                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3184                             CL_GATHER_MOUNT_ALWAYS, 0);
3185                         zfs_close(zhp);
3186                         if (clp != NULL) {
3187                                 /* mount and share received datasets */
3188                                 err = changelist_postfix(clp);
3189                                 changelist_free(clp);
3190                         }
3191                 }
3192                 if (zhp == NULL || clp == NULL || err)
3193                         err = -1;
3194         }
3195         if (top_zfs)
3196                 free(top_zfs);
3197 
3198         return (err);
3199 }