cddl/contrib/opensolaris/lib/libzfs/common/libzfs_sendrecv.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2012 Pawel Jakub Dawidek <pawel@dawidek.net>.
  27  * All rights reserved.
  28  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  29  */
  30
  31 #include <assert.h>
  32 #include <ctype.h>
  33 #include <errno.h>
  34 #include <libintl.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <strings.h>
  38 #include <unistd.h>
  39 #include <stddef.h>
  40 #include <fcntl.h>
  41 #include <sys/param.h>
  42 #include <sys/mount.h>
  43 #include <pthread.h>
  44 #include <umem.h>
  45 #include <time.h>
  46
  47 #include <libzfs.h>
  48 #include <libzfs_core.h>
  49
  50 #include "zfs_namecheck.h"
  51 #include "zfs_prop.h"
  52 #include "zfs_fletcher.h"
  53 #include "libzfs_impl.h"
  54 #include <sha2.h>
  55 #include <sys/zio_checksum.h>
  56 #include <sys/ddt.h>
  57
  58 #ifdef __FreeBSD__
  59 extern int zfs_ioctl_version;
  60 #endif
  61
  62 /* in libzfs_dataset.c */
  63 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  64 /* We need to use something for ENODATA. */
  65 #define ENODATA EIDRM
  66
  67 static int zfs_receive_impl(libzfs_handle_t *, const char *, const char *,
  68     recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **, int,
  69     uint64_t *);
  70
  71 static const zio_cksum_t zero_cksum = { 0 };
  72
  73 typedef struct dedup_arg {
  74         int     inputfd;
  75         int     outputfd;
  76         libzfs_handle_t  *dedup_hdl;
  77 } dedup_arg_t;
  78
  79 typedef struct progress_arg {
  80         zfs_handle_t *pa_zhp;
  81         int pa_fd;
  82         boolean_t pa_parsable;
  83 } progress_arg_t;
  84
  85 typedef struct dataref {
  86         uint64_t ref_guid;
  87         uint64_t ref_object;
  88         uint64_t ref_offset;
  89 } dataref_t;
  90
  91 typedef struct dedup_entry {
  92         struct dedup_entry      *dde_next;
  93         zio_cksum_t dde_chksum;
  94         uint64_t dde_prop;
  95         dataref_t dde_ref;
  96 } dedup_entry_t;
  97
  98 #define MAX_DDT_PHYSMEM_PERCENT         20
  99 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
 100
 101 typedef struct dedup_table {
 102         dedup_entry_t   **dedup_hash_array;
 103         umem_cache_t    *ddecache;
 104         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
 105         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
 106         uint64_t        ddt_count;
 107         int             numhashbits;
 108         boolean_t       ddt_full;
 109 } dedup_table_t;
 110
 111 static int
 112 high_order_bit(uint64_t n)
 113 {
 114         int count;
 115
 116         for (count = 0; n != 0; count++)
 117                 n >>= 1;
 118         return (count);
 119 }
 120
 121 static size_t
 122 ssread(void *buf, size_t len, FILE *stream)
 123 {
 124         size_t outlen;
 125
 126         if ((outlen = fread(buf, len, 1, stream)) == 0)
 127                 return (0);
 128
 129         return (outlen);
 130 }
 131
 132 static void
 133 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 134     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 135 {
 136         dedup_entry_t   *dde;
 137
 138         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 139                 if (ddt->ddt_full == B_FALSE) {
 140                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 141                             "Dedup table full.  Deduplication will continue "
 142                             "with existing table entries"));
 143                         ddt->ddt_full = B_TRUE;
 144                 }
 145                 return;
 146         }
 147
 148         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 149             != NULL) {
 150                 assert(*ddepp == NULL);
 151                 dde->dde_next = NULL;
 152                 dde->dde_chksum = *cs;
 153                 dde->dde_prop = prop;
 154                 dde->dde_ref = *dr;
 155                 *ddepp = dde;
 156                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 157                 ddt->ddt_count++;
 158         }
 159 }
 160
 161 /*
 162  * Using the specified dedup table, do a lookup for an entry with
 163  * the checksum cs.  If found, return the block's reference info
 164  * in *dr. Otherwise, insert a new entry in the dedup table, using
 165  * the reference information specified by *dr.
 166  *
 167  * return value:  true - entry was found
 168  *                false - entry was not found
 169  */
 170 static boolean_t
 171 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 172     uint64_t prop, dataref_t *dr)
 173 {
 174         uint32_t hashcode;
 175         dedup_entry_t **ddepp;
 176
 177         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 178
 179         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 180             ddepp = &((*ddepp)->dde_next)) {
 181                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 182                     (*ddepp)->dde_prop == prop) {
 183                         *dr = (*ddepp)->dde_ref;
 184                         return (B_TRUE);
 185                 }
 186         }
 187         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 188         return (B_FALSE);
 189 }
 190
 191 static int
 192 dump_record(dmu_replay_record_t *drr, void *payload, int payload_len,
 193     zio_cksum_t *zc, int outfd)
 194 {
 195         ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
 196             ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t));
 197         fletcher_4_incremental_native(drr,
 198             offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc);
 199         if (drr->drr_type != DRR_BEGIN) {
 200                 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u.
 201                     drr_checksum.drr_checksum));
 202                 drr->drr_u.drr_checksum.drr_checksum = *zc;
 203         }
 204         fletcher_4_incremental_native(&drr->drr_u.drr_checksum.drr_checksum,
 205             sizeof (zio_cksum_t), zc);
 206         if (write(outfd, drr, sizeof (*drr)) == -1)
 207                 return (errno);
 208         if (payload_len != 0) {
 209                 fletcher_4_incremental_native(payload, payload_len, zc);
 210                 if (write(outfd, payload, payload_len) == -1)
 211                         return (errno);
 212         }
 213         return (0);
 214 }
 215
 216 /*
 217  * This function is started in a separate thread when the dedup option
 218  * has been requested.  The main send thread determines the list of
 219  * snapshots to be included in the send stream and makes the ioctl calls
 220  * for each one.  But instead of having the ioctl send the output to the
 221  * the output fd specified by the caller of zfs_send()), the
 222  * ioctl is told to direct the output to a pipe, which is read by the
 223  * alternate thread running THIS function.  This function does the
 224  * dedup'ing by:
 225  *  1. building a dedup table (the DDT)
 226  *  2. doing checksums on each data block and inserting a record in the DDT
 227  *  3. looking for matching checksums, and
 228  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 229  *      a duplicate block is found.
 230  * The output of this function then goes to the output fd requested
 231  * by the caller of zfs_send().
 232  */
 233 static void *
 234 cksummer(void *arg)
 235 {
 236         dedup_arg_t *dda = arg;
 237         char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE);
 238         dmu_replay_record_t thedrr;
 239         dmu_replay_record_t *drr = &thedrr;
 240         FILE *ofp;
 241         int outfd;
 242         dedup_table_t ddt;
 243         zio_cksum_t stream_cksum;
 244         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 245         uint64_t numbuckets;
 246
 247         ddt.max_ddt_size =
 248             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100,
 249             SMALLEST_POSSIBLE_MAX_DDT_MB << 20);
 250
 251         numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t));
 252
 253         /*
 254          * numbuckets must be a power of 2.  Increase number to
 255          * a power of 2 if necessary.
 256          */
 257         if (!ISP2(numbuckets))
 258                 numbuckets = 1 << high_order_bit(numbuckets);
 259
 260         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 261         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 262             NULL, NULL, NULL, NULL, NULL, 0);
 263         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 264         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 265         ddt.ddt_full = B_FALSE;
 266
 267         outfd = dda->outputfd;
 268         ofp = fdopen(dda->inputfd, "r");
 269         while (ssread(drr, sizeof (*drr), ofp) != 0) {
 270
 271                 switch (drr->drr_type) {
 272                 case DRR_BEGIN:
 273                 {
 274                         struct drr_begin *drrb = &drr->drr_u.drr_begin;
 275                         int fflags;
 276                         int sz = 0;
 277                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 278
 279                         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 280
 281                         /* set the DEDUP feature flag for this stream */
 282                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 283                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 284                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 285                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 286
 287                         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 288                             DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
 289                                 sz = drr->drr_payloadlen;
 290
 291                                 if (sz > SPA_MAXBLOCKSIZE) {
 292                                         buf = zfs_realloc(dda->dedup_hdl, buf,
 293                                             SPA_MAXBLOCKSIZE, sz);
 294                                 }
 295                                 (void) ssread(buf, sz, ofp);
 296                                 if (ferror(stdin))
 297                                         perror("fread");
 298                         }
 299                         if (dump_record(drr, buf, sz, &stream_cksum,
 300                             outfd) != 0)
 301                                 goto out;
 302                         break;
 303                 }
 304
 305                 case DRR_END:
 306                 {
 307                         struct drr_end *drre = &drr->drr_u.drr_end;
 308                         /* use the recalculated checksum */
 309                         drre->drr_checksum = stream_cksum;
 310                         if (dump_record(drr, NULL, 0, &stream_cksum,
 311                             outfd) != 0)
 312                                 goto out;
 313                         break;
 314                 }
 315
 316                 case DRR_OBJECT:
 317                 {
 318                         struct drr_object *drro = &drr->drr_u.drr_object;
 319                         if (drro->drr_bonuslen > 0) {
 320                                 (void) ssread(buf,
 321                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 322                                     ofp);
 323                         }
 324                         if (dump_record(drr, buf,
 325                             P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 326                             &stream_cksum, outfd) != 0)
 327                                 goto out;
 328                         break;
 329                 }
 330
 331                 case DRR_SPILL:
 332                 {
 333                         struct drr_spill *drrs = &drr->drr_u.drr_spill;
 334                         (void) ssread(buf, drrs->drr_length, ofp);
 335                         if (dump_record(drr, buf, drrs->drr_length,
 336                             &stream_cksum, outfd) != 0)
 337                                 goto out;
 338                         break;
 339                 }
 340
 341                 case DRR_FREEOBJECTS:
 342                 {
 343                         if (dump_record(drr, NULL, 0, &stream_cksum,
 344                             outfd) != 0)
 345                                 goto out;
 346                         break;
 347                 }
 348
 349                 case DRR_WRITE:
 350                 {
 351                         struct drr_write *drrw = &drr->drr_u.drr_write;
 352                         dataref_t       dataref;
 353
 354                         (void) ssread(buf, drrw->drr_length, ofp);
 355
 356                         /*
 357                          * Use the existing checksum if it's dedup-capable,
 358                          * else calculate a SHA256 checksum for it.
 359                          */
 360
 361                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 362                             zero_cksum) ||
 363                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 364                                 SHA256_CTX      ctx;
 365                                 zio_cksum_t     tmpsha256;
 366
 367                                 SHA256Init(&ctx);
 368                                 SHA256Update(&ctx, buf, drrw->drr_length);
 369                                 SHA256Final(&tmpsha256, &ctx);
 370                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 371                                     BE_64(tmpsha256.zc_word[0]);
 372                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 373                                     BE_64(tmpsha256.zc_word[1]);
 374                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 375                                     BE_64(tmpsha256.zc_word[2]);
 376                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 377                                     BE_64(tmpsha256.zc_word[3]);
 378                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 379                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 380                         }
 381
 382                         dataref.ref_guid = drrw->drr_toguid;
 383                         dataref.ref_object = drrw->drr_object;
 384                         dataref.ref_offset = drrw->drr_offset;
 385
 386                         if (ddt_update(dda->dedup_hdl, &ddt,
 387                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 388                             &dataref)) {
 389                                 dmu_replay_record_t wbr_drr = {0};
 390                                 struct drr_write_byref *wbr_drrr =
 391                                     &wbr_drr.drr_u.drr_write_byref;
 392
 393                                 /* block already present in stream */
 394                                 wbr_drr.drr_type = DRR_WRITE_BYREF;
 395
 396                                 wbr_drrr->drr_object = drrw->drr_object;
 397                                 wbr_drrr->drr_offset = drrw->drr_offset;
 398                                 wbr_drrr->drr_length = drrw->drr_length;
 399                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 400                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 401                                 wbr_drrr->drr_refobject =
 402                                     dataref.ref_object;
 403                                 wbr_drrr->drr_refoffset =
 404                                     dataref.ref_offset;
 405
 406                                 wbr_drrr->drr_checksumtype =
 407                                     drrw->drr_checksumtype;
 408                                 wbr_drrr->drr_checksumflags =
 409                                     drrw->drr_checksumtype;
 410                                 wbr_drrr->drr_key.ddk_cksum =
 411                                     drrw->drr_key.ddk_cksum;
 412                                 wbr_drrr->drr_key.ddk_prop =
 413                                     drrw->drr_key.ddk_prop;
 414
 415                                 if (dump_record(&wbr_drr, NULL, 0,
 416                                     &stream_cksum, outfd) != 0)
 417                                         goto out;
 418                         } else {
 419                                 /* block not previously seen */
 420                                 if (dump_record(drr, buf, drrw->drr_length,
 421                                     &stream_cksum, outfd) != 0)
 422                                         goto out;
 423                         }
 424                         break;
 425                 }
 426
 427                 case DRR_WRITE_EMBEDDED:
 428                 {
 429                         struct drr_write_embedded *drrwe =
 430                             &drr->drr_u.drr_write_embedded;
 431                         (void) ssread(buf,
 432                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp);
 433                         if (dump_record(drr, buf,
 434                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8),
 435                             &stream_cksum, outfd) != 0)
 436                                 goto out;
 437                         break;
 438                 }
 439
 440                 case DRR_FREE:
 441                 {
 442                         if (dump_record(drr, NULL, 0, &stream_cksum,
 443                             outfd) != 0)
 444                                 goto out;
 445                         break;
 446                 }
 447
 448                 default:
 449                         (void) fprintf(stderr, "INVALID record type 0x%x\n",
 450                             drr->drr_type);
 451                         /* should never happen, so assert */
 452                         assert(B_FALSE);
 453                 }
 454         }
 455 out:
 456         umem_cache_destroy(ddt.ddecache);
 457         free(ddt.dedup_hash_array);
 458         free(buf);
 459         (void) fclose(ofp);
 460
 461         return (NULL);
 462 }
 463
 464 /*
 465  * Routines for dealing with the AVL tree of fs-nvlists
 466  */
 467 typedef struct fsavl_node {
 468         avl_node_t fn_node;
 469         nvlist_t *fn_nvfs;
 470         char *fn_snapname;
 471         uint64_t fn_guid;
 472 } fsavl_node_t;
 473
 474 static int
 475 fsavl_compare(const void *arg1, const void *arg2)
 476 {
 477         const fsavl_node_t *fn1 = arg1;
 478         const fsavl_node_t *fn2 = arg2;
 479
 480         if (fn1->fn_guid > fn2->fn_guid)
 481                 return (+1);
 482         else if (fn1->fn_guid < fn2->fn_guid)
 483                 return (-1);
 484         else
 485                 return (0);
 486 }
 487
 488 /*
 489  * Given the GUID of a snapshot, find its containing filesystem and
 490  * (optionally) name.
 491  */
 492 static nvlist_t *
 493 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 494 {
 495         fsavl_node_t fn_find;
 496         fsavl_node_t *fn;
 497
 498         fn_find.fn_guid = snapguid;
 499
 500         fn = avl_find(avl, &fn_find, NULL);
 501         if (fn) {
 502                 if (snapname)
 503                         *snapname = fn->fn_snapname;
 504                 return (fn->fn_nvfs);
 505         }
 506         return (NULL);
 507 }
 508
 509 static void
 510 fsavl_destroy(avl_tree_t *avl)
 511 {
 512         fsavl_node_t *fn;
 513         void *cookie;
 514
 515         if (avl == NULL)
 516                 return;
 517
 518         cookie = NULL;
 519         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 520                 free(fn);
 521         avl_destroy(avl);
 522         free(avl);
 523 }
 524
 525 /*
 526  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 527  */
 528 static avl_tree_t *
 529 fsavl_create(nvlist_t *fss)
 530 {
 531         avl_tree_t *fsavl;
 532         nvpair_t *fselem = NULL;
 533
 534         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 535                 return (NULL);
 536
 537         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 538             offsetof(fsavl_node_t, fn_node));
 539
 540         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 541                 nvlist_t *nvfs, *snaps;
 542                 nvpair_t *snapelem = NULL;
 543
 544                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 545                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 546
 547                 while ((snapelem =
 548                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 549                         fsavl_node_t *fn;
 550                         uint64_t guid;
 551
 552                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 553                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 554                                 fsavl_destroy(fsavl);
 555                                 return (NULL);
 556                         }
 557                         fn->fn_nvfs = nvfs;
 558                         fn->fn_snapname = nvpair_name(snapelem);
 559                         fn->fn_guid = guid;
 560
 561                         /*
 562                          * Note: if there are multiple snaps with the
 563                          * same GUID, we ignore all but one.
 564                          */
 565                         if (avl_find(fsavl, fn, NULL) == NULL)
 566                                 avl_add(fsavl, fn);
 567                         else
 568                                 free(fn);
 569                 }
 570         }
 571
 572         return (fsavl);
 573 }
 574
 575 /*
 576  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 577  */
 578 typedef struct send_data {
 579         uint64_t parent_fromsnap_guid;
 580         nvlist_t *parent_snaps;
 581         nvlist_t *fss;
 582         nvlist_t *snapprops;
 583         const char *fromsnap;
 584         const char *tosnap;
 585         boolean_t recursive;
 586
 587         /*
 588          * The header nvlist is of the following format:
 589          * {
 590          *   "tosnap" -> string
 591          *   "fromsnap" -> string (if incremental)
 592          *   "fss" -> {
 593          *      id -> {
 594          *
 595          *       "name" -> string (full name; for debugging)
 596          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 597          *
 598          *       "props" -> { name -> value (only if set here) }
 599          *       "snaps" -> { name (lastname) -> number (guid) }
 600          *       "snapprops" -> { name (lastname) -> { name -> value } }
 601          *
 602          *       "origin" -> number (guid) (if clone)
 603          *       "sent" -> boolean (not on-disk)
 604          *      }
 605          *   }
 606          * }
 607          *
 608          */
 609 } send_data_t;
 610
 611 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 612
 613 static int
 614 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 615 {
 616         send_data_t *sd = arg;
 617         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 618         char *snapname;
 619         nvlist_t *nv;
 620
 621         snapname = strrchr(zhp->zfs_name, '@')+1;
 622
 623         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 624         /*
 625          * NB: if there is no fromsnap here (it's a newly created fs in
 626          * an incremental replication), we will substitute the tosnap.
 627          */
 628         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 629             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 630             strcmp(snapname, sd->tosnap) == 0)) {
 631                 sd->parent_fromsnap_guid = guid;
 632         }
 633
 634         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 635         send_iterate_prop(zhp, nv);
 636         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 637         nvlist_free(nv);
 638
 639         zfs_close(zhp);
 640         return (0);
 641 }
 642
 643 static void
 644 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 645 {
 646         nvpair_t *elem = NULL;
 647
 648         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 649                 char *propname = nvpair_name(elem);
 650                 zfs_prop_t prop = zfs_name_to_prop(propname);
 651                 nvlist_t *propnv;
 652
 653                 if (!zfs_prop_user(propname)) {
 654                         /*
 655                          * Realistically, this should never happen.  However,
 656                          * we want the ability to add DSL properties without
 657                          * needing to make incompatible version changes.  We
 658                          * need to ignore unknown properties to allow older
 659                          * software to still send datasets containing these
 660                          * properties, with the unknown properties elided.
 661                          */
 662                         if (prop == ZPROP_INVAL)
 663                                 continue;
 664
 665                         if (zfs_prop_readonly(prop))
 666                                 continue;
 667                 }
 668
 669                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 670                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 671                     prop == ZFS_PROP_REFQUOTA ||
 672                     prop == ZFS_PROP_REFRESERVATION) {
 673                         char *source;
 674                         uint64_t value;
 675                         verify(nvlist_lookup_uint64(propnv,
 676                             ZPROP_VALUE, &value) == 0);
 677                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 678                                 continue;
 679                         /*
 680                          * May have no source before SPA_VERSION_RECVD_PROPS,
 681                          * but is still modifiable.
 682                          */
 683                         if (nvlist_lookup_string(propnv,
 684                             ZPROP_SOURCE, &source) == 0) {
 685                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 686                                     (strcmp(source,
 687                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 688                                         continue;
 689                         }
 690                 } else {
 691                         char *source;
 692                         if (nvlist_lookup_string(propnv,
 693                             ZPROP_SOURCE, &source) != 0)
 694                                 continue;
 695                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 696                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 697                                 continue;
 698                 }
 699
 700                 if (zfs_prop_user(propname) ||
 701                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 702                         char *value;
 703                         verify(nvlist_lookup_string(propnv,
 704                             ZPROP_VALUE, &value) == 0);
 705                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 706                 } else {
 707                         uint64_t value;
 708                         verify(nvlist_lookup_uint64(propnv,
 709                             ZPROP_VALUE, &value) == 0);
 710                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 711                 }
 712         }
 713 }
 714
 715 /*
 716  * recursively generate nvlists describing datasets.  See comment
 717  * for the data structure send_data_t above for description of contents
 718  * of the nvlist.
 719  */
 720 static int
 721 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 722 {
 723         send_data_t *sd = arg;
 724         nvlist_t *nvfs, *nv;
 725         int rv = 0;
 726         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 727         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 728         char guidstring[64];
 729
 730         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 731         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 732         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 733             sd->parent_fromsnap_guid));
 734
 735         if (zhp->zfs_dmustats.dds_origin[0]) {
 736                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 737                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 738                 if (origin == NULL)
 739                         return (-1);
 740                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 741                     origin->zfs_dmustats.dds_guid));
 742         }
 743
 744         /* iterate over props */
 745         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 746         send_iterate_prop(zhp, nv);
 747         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 748         nvlist_free(nv);
 749
 750         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 751         sd->parent_fromsnap_guid = 0;
 752         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 753         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 754         (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd);
 755         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 756         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 757         nvlist_free(sd->parent_snaps);
 758         nvlist_free(sd->snapprops);
 759
 760         /* add this fs to nvlist */
 761         (void) snprintf(guidstring, sizeof (guidstring),
 762             "0x%llx", (longlong_t)guid);
 763         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 764         nvlist_free(nvfs);
 765
 766         /* iterate over children */
 767         if (sd->recursive)
 768                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 769
 770         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 771
 772         zfs_close(zhp);
 773         return (rv);
 774 }
 775
 776 static int
 777 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 778     const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
 779 {
 780         zfs_handle_t *zhp;
 781         send_data_t sd = { 0 };
 782         int error;
 783
 784         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 785         if (zhp == NULL)
 786                 return (EZFS_BADTYPE);
 787
 788         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 789         sd.fromsnap = fromsnap;
 790         sd.tosnap = tosnap;
 791         sd.recursive = recursive;
 792
 793         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 794                 nvlist_free(sd.fss);
 795                 if (avlp != NULL)
 796                         *avlp = NULL;
 797                 *nvlp = NULL;
 798                 return (error);
 799         }
 800
 801         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 802                 nvlist_free(sd.fss);
 803                 *nvlp = NULL;
 804                 return (EZFS_NOMEM);
 805         }
 806
 807         *nvlp = sd.fss;
 808         return (0);
 809 }
 810
 811 /*
 812  * Routines specific to "zfs send"
 813  */
 814 typedef struct send_dump_data {
 815         /* these are all just the short snapname (the part after the @) */
 816         const char *fromsnap;
 817         const char *tosnap;
 818         char prevsnap[ZFS_MAXNAMELEN];
 819         uint64_t prevsnap_obj;
 820         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 821         boolean_t verbose, dryrun, parsable, progress, embed_data, large_block;
 822         int outfd;
 823         boolean_t err;
 824         nvlist_t *fss;
 825         nvlist_t *snapholds;
 826         avl_tree_t *fsavl;
 827         snapfilter_cb_t *filter_cb;
 828         void *filter_cb_arg;
 829         nvlist_t *debugnv;
 830         char holdtag[ZFS_MAXNAMELEN];
 831         int cleanup_fd;
 832         uint64_t size;
 833 } send_dump_data_t;
 834
 835 static int
 836 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
 837     boolean_t fromorigin, uint64_t *sizep)
 838 {
 839         zfs_cmd_t zc = { 0 };
 840         libzfs_handle_t *hdl = zhp->zfs_hdl;
 841
 842         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 843         assert(fromsnap_obj == 0 || !fromorigin);
 844
 845         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 846         zc.zc_obj = fromorigin;
 847         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 848         zc.zc_fromobj = fromsnap_obj;
 849         zc.zc_guid = 1;  /* estimate flag */
 850
 851         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 852                 char errbuf[1024];
 853                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 854                     "warning: cannot estimate space for '%s'"), zhp->zfs_name);
 855
 856                 switch (errno) {
 857                 case EXDEV:
 858                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 859                             "not an earlier snapshot from the same fs"));
 860                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 861
 862                 case ENOENT:
 863                         if (zfs_dataset_exists(hdl, zc.zc_name,
 864                             ZFS_TYPE_SNAPSHOT)) {
 865                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 866                                     "incremental source (@%s) does not exist"),
 867                                     zc.zc_value);
 868                         }
 869                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 870
 871                 case EDQUOT:
 872                 case EFBIG:
 873                 case EIO:
 874                 case ENOLINK:
 875                 case ENOSPC:
 876                 case ENXIO:
 877                 case EPIPE:
 878                 case ERANGE:
 879                 case EFAULT:
 880                 case EROFS:
 881                         zfs_error_aux(hdl, strerror(errno));
 882                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 883
 884                 default:
 885                         return (zfs_standard_error(hdl, errno, errbuf));
 886                 }
 887         }
 888
 889         *sizep = zc.zc_objset_type;
 890
 891         return (0);
 892 }
 893
 894 /*
 895  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 896  * NULL) to the file descriptor specified by outfd.
 897  */
 898 static int
 899 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
 900     boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
 901     nvlist_t *debugnv)
 902 {
 903         zfs_cmd_t zc = { 0 };
 904         libzfs_handle_t *hdl = zhp->zfs_hdl;
 905         nvlist_t *thisdbg;
 906
 907         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 908         assert(fromsnap_obj == 0 || !fromorigin);
 909
 910         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 911         zc.zc_cookie = outfd;
 912         zc.zc_obj = fromorigin;
 913         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 914         zc.zc_fromobj = fromsnap_obj;
 915         zc.zc_flags = flags;
 916
 917         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 918         if (fromsnap && fromsnap[0] != '\0') {
 919                 VERIFY(0 == nvlist_add_string(thisdbg,
 920                     "fromsnap", fromsnap));
 921         }
 922
 923         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 924                 char errbuf[1024];
 925                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 926                     "warning: cannot send '%s'"), zhp->zfs_name);
 927
 928                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
 929                 if (debugnv) {
 930                         VERIFY(0 == nvlist_add_nvlist(debugnv,
 931                             zhp->zfs_name, thisdbg));
 932                 }
 933                 nvlist_free(thisdbg);
 934
 935                 switch (errno) {
 936                 case EXDEV:
 937                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 938                             "not an earlier snapshot from the same fs"));
 939                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 940
 941                 case ENOENT:
 942                         if (zfs_dataset_exists(hdl, zc.zc_name,
 943                             ZFS_TYPE_SNAPSHOT)) {
 944                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 945                                     "incremental source (@%s) does not exist"),
 946                                     zc.zc_value);
 947                         }
 948                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 949
 950                 case EDQUOT:
 951                 case EFBIG:
 952                 case EIO:
 953                 case ENOLINK:
 954                 case ENOSPC:
 955 #ifdef sun
 956                 case ENOSTR:
 957 #endif
 958                 case ENXIO:
 959                 case EPIPE:
 960                 case ERANGE:
 961                 case EFAULT:
 962                 case EROFS:
 963                         zfs_error_aux(hdl, strerror(errno));
 964                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 965
 966                 default:
 967                         return (zfs_standard_error(hdl, errno, errbuf));
 968                 }
 969         }
 970
 971         if (debugnv)
 972                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
 973         nvlist_free(thisdbg);
 974
 975         return (0);
 976 }
 977
 978 static void
 979 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
 980 {
 981         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 982
 983         /*
 984          * zfs_send() only sets snapholds for sends that need them,
 985          * e.g. replication and doall.
 986          */
 987         if (sdd->snapholds == NULL)
 988                 return;
 989
 990         fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
 991 }
 992
 993 static void *
 994 send_progress_thread(void *arg)
 995 {
 996         progress_arg_t *pa = arg;
 997
 998         zfs_cmd_t zc = { 0 };
 999         zfs_handle_t *zhp = pa->pa_zhp;
1000         libzfs_handle_t *hdl = zhp->zfs_hdl;
1001         unsigned long long bytes;
1002         char buf[16];
1003
1004         time_t t;
1005         struct tm *tm;
1006
1007         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
1008         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1009
1010         if (!pa->pa_parsable)
1011                 (void) fprintf(stderr, "TIME        SENT   SNAPSHOT\n");
1012
1013         /*
1014          * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1015          */
1016         for (;;) {
1017                 (void) sleep(1);
1018
1019                 zc.zc_cookie = pa->pa_fd;
1020                 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1021                         return ((void *)-1);
1022
1023                 (void) time(&t);
1024                 tm = localtime(&t);
1025                 bytes = zc.zc_cookie;
1026
1027                 if (pa->pa_parsable) {
1028                         (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1029                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1030                             bytes, zhp->zfs_name);
1031                 } else {
1032                         zfs_nicenum(bytes, buf, sizeof (buf));
1033                         (void) fprintf(stderr, "%02d:%02d:%02d   %5s   %s\n",
1034                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1035                             buf, zhp->zfs_name);
1036                 }
1037         }
1038 }
1039
1040 static int
1041 dump_snapshot(zfs_handle_t *zhp, void *arg)
1042 {
1043         send_dump_data_t *sdd = arg;
1044         progress_arg_t pa = { 0 };
1045         pthread_t tid;
1046         char *thissnap;
1047         int err;
1048         boolean_t isfromsnap, istosnap, fromorigin;
1049         boolean_t exclude = B_FALSE;
1050
1051         err = 0;
1052         thissnap = strchr(zhp->zfs_name, '@') + 1;
1053         isfromsnap = (sdd->fromsnap != NULL &&
1054             strcmp(sdd->fromsnap, thissnap) == 0);
1055
1056         if (!sdd->seenfrom && isfromsnap) {
1057                 gather_holds(zhp, sdd);
1058                 sdd->seenfrom = B_TRUE;
1059                 (void) strcpy(sdd->prevsnap, thissnap);
1060                 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1061                 zfs_close(zhp);
1062                 return (0);
1063         }
1064
1065         if (sdd->seento || !sdd->seenfrom) {
1066                 zfs_close(zhp);
1067                 return (0);
1068         }
1069
1070         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1071         if (istosnap)
1072                 sdd->seento = B_TRUE;
1073
1074         if (!sdd->doall && !isfromsnap && !istosnap) {
1075                 if (sdd->replicate) {
1076                         char *snapname;
1077                         nvlist_t *snapprops;
1078                         /*
1079                          * Filter out all intermediate snapshots except origin
1080                          * snapshots needed to replicate clones.
1081                          */
1082                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1083                             zhp->zfs_dmustats.dds_guid, &snapname);
1084
1085                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1086                             "snapprops", &snapprops));
1087                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1088                             thissnap, &snapprops));
1089                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
1090                 } else {
1091                         exclude = B_TRUE;
1092                 }
1093         }
1094
1095         /*
1096          * If a filter function exists, call it to determine whether
1097          * this snapshot will be sent.
1098          */
1099         if (exclude || (sdd->filter_cb != NULL &&
1100             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1101                 /*
1102                  * This snapshot is filtered out.  Don't send it, and don't
1103                  * set prevsnap_obj, so it will be as if this snapshot didn't
1104                  * exist, and the next accepted snapshot will be sent as
1105                  * an incremental from the last accepted one, or as the
1106                  * first (and full) snapshot in the case of a replication,
1107                  * non-incremental send.
1108                  */
1109                 zfs_close(zhp);
1110                 return (0);
1111         }
1112
1113         gather_holds(zhp, sdd);
1114         fromorigin = sdd->prevsnap[0] == '\0' &&
1115             (sdd->fromorigin || sdd->replicate);
1116
1117         if (sdd->verbose) {
1118                 uint64_t size;
1119                 err = estimate_ioctl(zhp, sdd->prevsnap_obj,
1120                     fromorigin, &size);
1121
1122                 if (sdd->parsable) {
1123                         if (sdd->prevsnap[0] != '\0') {
1124                                 (void) fprintf(stderr, "incremental\t%s\t%s",
1125                                     sdd->prevsnap, zhp->zfs_name);
1126                         } else {
1127                                 (void) fprintf(stderr, "full\t%s",
1128                                     zhp->zfs_name);
1129                         }
1130                 } else {
1131                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1132                             "send from @%s to %s"),
1133                             sdd->prevsnap, zhp->zfs_name);
1134                 }
1135                 if (err == 0) {
1136                         if (sdd->parsable) {
1137                                 (void) fprintf(stderr, "\t%llu\n",
1138                                     (longlong_t)size);
1139                         } else {
1140                                 char buf[16];
1141                                 zfs_nicenum(size, buf, sizeof (buf));
1142                                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1143                                     " estimated size is %s\n"), buf);
1144                         }
1145                         sdd->size += size;
1146                 } else {
1147                         (void) fprintf(stderr, "\n");
1148                 }
1149         }
1150
1151         if (!sdd->dryrun) {
1152                 /*
1153                  * If progress reporting is requested, spawn a new thread to
1154                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1155                  */
1156                 if (sdd->progress) {
1157                         pa.pa_zhp = zhp;
1158                         pa.pa_fd = sdd->outfd;
1159                         pa.pa_parsable = sdd->parsable;
1160
1161                         if (err = pthread_create(&tid, NULL,
1162                             send_progress_thread, &pa)) {
1163                                 zfs_close(zhp);
1164                                 return (err);
1165                         }
1166                 }
1167
1168                 enum lzc_send_flags flags = 0;
1169                 if (sdd->large_block)
1170                         flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1171                 if (sdd->embed_data)
1172                         flags |= LZC_SEND_FLAG_EMBED_DATA;
1173
1174                 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1175                     fromorigin, sdd->outfd, flags, sdd->debugnv);
1176
1177                 if (sdd->progress) {
1178                         (void) pthread_cancel(tid);
1179                         (void) pthread_join(tid, NULL);
1180                 }
1181         }
1182
1183         (void) strcpy(sdd->prevsnap, thissnap);
1184         sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1185         zfs_close(zhp);
1186         return (err);
1187 }
1188
1189 static int
1190 dump_filesystem(zfs_handle_t *zhp, void *arg)
1191 {
1192         int rv = 0;
1193         send_dump_data_t *sdd = arg;
1194         boolean_t missingfrom = B_FALSE;
1195         zfs_cmd_t zc = { 0 };
1196
1197         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1198             zhp->zfs_name, sdd->tosnap);
1199         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1200                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1201                     "WARNING: could not send %s@%s: does not exist\n"),
1202                     zhp->zfs_name, sdd->tosnap);
1203                 sdd->err = B_TRUE;
1204                 return (0);
1205         }
1206
1207         if (sdd->replicate && sdd->fromsnap) {
1208                 /*
1209                  * If this fs does not have fromsnap, and we're doing
1210                  * recursive, we need to send a full stream from the
1211                  * beginning (or an incremental from the origin if this
1212                  * is a clone).  If we're doing non-recursive, then let
1213                  * them get the error.
1214                  */
1215                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1216                     zhp->zfs_name, sdd->fromsnap);
1217                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1218                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1219                         missingfrom = B_TRUE;
1220                 }
1221         }
1222
1223         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1224         sdd->prevsnap_obj = 0;
1225         if (sdd->fromsnap == NULL || missingfrom)
1226                 sdd->seenfrom = B_TRUE;
1227
1228         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1229         if (!sdd->seenfrom) {
1230                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1231                     "WARNING: could not send %s@%s:\n"
1232                     "incremental source (%s@%s) does not exist\n"),
1233                     zhp->zfs_name, sdd->tosnap,
1234                     zhp->zfs_name, sdd->fromsnap);
1235                 sdd->err = B_TRUE;
1236         } else if (!sdd->seento) {
1237                 if (sdd->fromsnap) {
1238                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1239                             "WARNING: could not send %s@%s:\n"
1240                             "incremental source (%s@%s) "
1241                             "is not earlier than it\n"),
1242                             zhp->zfs_name, sdd->tosnap,
1243                             zhp->zfs_name, sdd->fromsnap);
1244                 } else {
1245                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1246                             "WARNING: "
1247                             "could not send %s@%s: does not exist\n"),
1248                             zhp->zfs_name, sdd->tosnap);
1249                 }
1250                 sdd->err = B_TRUE;
1251         }
1252
1253         return (rv);
1254 }
1255
1256 static int
1257 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1258 {
1259         send_dump_data_t *sdd = arg;
1260         nvpair_t *fspair;
1261         boolean_t needagain, progress;
1262
1263         if (!sdd->replicate)
1264                 return (dump_filesystem(rzhp, sdd));
1265
1266         /* Mark the clone origin snapshots. */
1267         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1268             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1269                 nvlist_t *nvfs;
1270                 uint64_t origin_guid = 0;
1271
1272                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1273                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1274                 if (origin_guid != 0) {
1275                         char *snapname;
1276                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1277                             origin_guid, &snapname);
1278                         if (origin_nv != NULL) {
1279                                 nvlist_t *snapprops;
1280                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1281                                     "snapprops", &snapprops));
1282                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1283                                     snapname, &snapprops));
1284                                 VERIFY(0 == nvlist_add_boolean(
1285                                     snapprops, "is_clone_origin"));
1286                         }
1287                 }
1288         }
1289 again:
1290         needagain = progress = B_FALSE;
1291         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1292             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1293                 nvlist_t *fslist, *parent_nv;
1294                 char *fsname;
1295                 zfs_handle_t *zhp;
1296                 int err;
1297                 uint64_t origin_guid = 0;
1298                 uint64_t parent_guid = 0;
1299
1300                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1301                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1302                         continue;
1303
1304                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1305                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1306                 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1307                     &parent_guid);
1308
1309                 if (parent_guid != 0) {
1310                         parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1311                         if (!nvlist_exists(parent_nv, "sent")) {
1312                                 /* parent has not been sent; skip this one */
1313                                 needagain = B_TRUE;
1314                                 continue;
1315                         }
1316                 }
1317
1318                 if (origin_guid != 0) {
1319                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1320                             origin_guid, NULL);
1321                         if (origin_nv != NULL &&
1322                             !nvlist_exists(origin_nv, "sent")) {
1323                                 /*
1324                                  * origin has not been sent yet;
1325                                  * skip this clone.
1326                                  */
1327                                 needagain = B_TRUE;
1328                                 continue;
1329                         }
1330                 }
1331
1332                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1333                 if (zhp == NULL)
1334                         return (-1);
1335                 err = dump_filesystem(zhp, sdd);
1336                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1337                 progress = B_TRUE;
1338                 zfs_close(zhp);
1339                 if (err)
1340                         return (err);
1341         }
1342         if (needagain) {
1343                 assert(progress);
1344                 goto again;
1345         }
1346
1347         /* clean out the sent flags in case we reuse this fss */
1348         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1349             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1350                 nvlist_t *fslist;
1351
1352                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1353                 (void) nvlist_remove_all(fslist, "sent");
1354         }
1355
1356         return (0);
1357 }
1358
1359 /*
1360  * Generate a send stream for the dataset identified by the argument zhp.
1361  *
1362  * The content of the send stream is the snapshot identified by
1363  * 'tosnap'.  Incremental streams are requested in two ways:
1364  *     - from the snapshot identified by "fromsnap" (if non-null) or
1365  *     - from the origin of the dataset identified by zhp, which must
1366  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1367  *       is TRUE.
1368  *
1369  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1370  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1371  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1372  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1373  * case too. If "props" is set, send properties.
1374  */
1375 int
1376 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1377     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1378     void *cb_arg, nvlist_t **debugnvp)
1379 {
1380         char errbuf[1024];
1381         send_dump_data_t sdd = { 0 };
1382         int err = 0;
1383         nvlist_t *fss = NULL;
1384         avl_tree_t *fsavl = NULL;
1385         static uint64_t holdseq;
1386         int spa_version;
1387         pthread_t tid = 0;
1388         int pipefd[2];
1389         dedup_arg_t dda = { 0 };
1390         int featureflags = 0;
1391
1392         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1393             "cannot send '%s'"), zhp->zfs_name);
1394
1395         if (fromsnap && fromsnap[0] == '\0') {
1396                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1397                     "zero-length incremental source"));
1398                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1399         }
1400
1401         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1402                 uint64_t version;
1403                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1404                 if (version >= ZPL_VERSION_SA) {
1405                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1406                 }
1407         }
1408
1409         if (flags->dedup && !flags->dryrun) {
1410                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1411                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1412                 if (err = pipe(pipefd)) {
1413                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1414                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1415                             errbuf));
1416                 }
1417                 dda.outputfd = outfd;
1418                 dda.inputfd = pipefd[1];
1419                 dda.dedup_hdl = zhp->zfs_hdl;
1420                 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1421                         (void) close(pipefd[0]);
1422                         (void) close(pipefd[1]);
1423                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1424                         return (zfs_error(zhp->zfs_hdl,
1425                             EZFS_THREADCREATEFAILED, errbuf));
1426                 }
1427         }
1428
1429         if (flags->replicate || flags->doall || flags->props) {
1430                 dmu_replay_record_t drr = { 0 };
1431                 char *packbuf = NULL;
1432                 size_t buflen = 0;
1433                 zio_cksum_t zc = { 0 };
1434
1435                 if (flags->replicate || flags->props) {
1436                         nvlist_t *hdrnv;
1437
1438                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1439                         if (fromsnap) {
1440                                 VERIFY(0 == nvlist_add_string(hdrnv,
1441                                     "fromsnap", fromsnap));
1442                         }
1443                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1444                         if (!flags->replicate) {
1445                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1446                                     "not_recursive"));
1447                         }
1448
1449                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1450                             fromsnap, tosnap, flags->replicate, &fss, &fsavl);
1451                         if (err)
1452                                 goto err_out;
1453                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1454                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1455                             NV_ENCODE_XDR, 0);
1456                         if (debugnvp)
1457                                 *debugnvp = hdrnv;
1458                         else
1459                                 nvlist_free(hdrnv);
1460                         if (err)
1461                                 goto stderr_out;
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
1477                         err = dump_record(&drr, packbuf, buflen, &zc, outfd);
1478                         free(packbuf);
1479                         if (err != 0)
1480                                 goto stderr_out;
1481
1482                         /* write end record */
1483                         bzero(&drr, sizeof (drr));
1484                         drr.drr_type = DRR_END;
1485                         drr.drr_u.drr_end.drr_checksum = zc;
1486                         err = write(outfd, &drr, sizeof (drr));
1487                         if (err == -1) {
1488                                 err = errno;
1489                                 goto stderr_out;
1490                         }
1491
1492                         err = 0;
1493                 }
1494         }
1495
1496         /* dump each stream */
1497         sdd.fromsnap = fromsnap;
1498         sdd.tosnap = tosnap;
1499         if (tid != 0)
1500                 sdd.outfd = pipefd[0];
1501         else
1502                 sdd.outfd = outfd;
1503         sdd.replicate = flags->replicate;
1504         sdd.doall = flags->doall;
1505         sdd.fromorigin = flags->fromorigin;
1506         sdd.fss = fss;
1507         sdd.fsavl = fsavl;
1508         sdd.verbose = flags->verbose;
1509         sdd.parsable = flags->parsable;
1510         sdd.progress = flags->progress;
1511         sdd.dryrun = flags->dryrun;
1512         sdd.large_block = flags->largeblock;
1513         sdd.embed_data = flags->embed_data;
1514         sdd.filter_cb = filter_func;
1515         sdd.filter_cb_arg = cb_arg;
1516         if (debugnvp)
1517                 sdd.debugnv = *debugnvp;
1518
1519         /*
1520          * Some flags require that we place user holds on the datasets that are
1521          * being sent so they don't get destroyed during the send. We can skip
1522          * this step if the pool is imported read-only since the datasets cannot
1523          * be destroyed.
1524          */
1525         if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1526             ZPOOL_PROP_READONLY, NULL) &&
1527             zfs_spa_version(zhp, &spa_version) == 0 &&
1528             spa_version >= SPA_VERSION_USERREFS &&
1529             (flags->doall || flags->replicate)) {
1530                 ++holdseq;
1531                 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1532                     ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1533                 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1534                 if (sdd.cleanup_fd < 0) {
1535                         err = errno;
1536                         goto stderr_out;
1537                 }
1538                 sdd.snapholds = fnvlist_alloc();
1539         } else {
1540                 sdd.cleanup_fd = -1;
1541                 sdd.snapholds = NULL;
1542         }
1543         if (flags->verbose || sdd.snapholds != NULL) {
1544                 /*
1545                  * Do a verbose no-op dry run to get all the verbose output
1546                  * or to gather snapshot hold's before generating any data,
1547                  * then do a non-verbose real run to generate the streams.
1548                  */
1549                 sdd.dryrun = B_TRUE;
1550                 err = dump_filesystems(zhp, &sdd);
1551
1552                 if (err != 0)
1553                         goto stderr_out;
1554
1555                 if (flags->verbose) {
1556                         if (flags->parsable) {
1557                                 (void) fprintf(stderr, "size\t%llu\n",
1558                                     (longlong_t)sdd.size);
1559                         } else {
1560                                 char buf[16];
1561                                 zfs_nicenum(sdd.size, buf, sizeof (buf));
1562                                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1563                                     "total estimated size is %s\n"), buf);
1564                         }
1565                 }
1566
1567                 /* Ensure no snaps found is treated as an error. */
1568                 if (!sdd.seento) {
1569                         err = ENOENT;
1570                         goto err_out;
1571                 }
1572
1573                 /* Skip the second run if dryrun was requested. */
1574                 if (flags->dryrun)
1575                         goto err_out;
1576
1577                 if (sdd.snapholds != NULL) {
1578                         err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
1579                         if (err != 0)
1580                                 goto stderr_out;
1581
1582                         fnvlist_free(sdd.snapholds);
1583                         sdd.snapholds = NULL;
1584                 }
1585
1586                 sdd.dryrun = B_FALSE;
1587                 sdd.verbose = B_FALSE;
1588         }
1589
1590         err = dump_filesystems(zhp, &sdd);
1591         fsavl_destroy(fsavl);
1592         nvlist_free(fss);
1593
1594         /* Ensure no snaps found is treated as an error. */
1595         if (err == 0 && !sdd.seento)
1596                 err = ENOENT;
1597
1598         if (tid != 0) {
1599                 if (err != 0)
1600                         (void) pthread_cancel(tid);
1601                 (void) close(pipefd[0]);
1602                 (void) pthread_join(tid, NULL);
1603         }
1604
1605         if (sdd.cleanup_fd != -1) {
1606                 VERIFY(0 == close(sdd.cleanup_fd));
1607                 sdd.cleanup_fd = -1;
1608         }
1609
1610         if (!flags->dryrun && (flags->replicate || flags->doall ||
1611             flags->props)) {
1612                 /*
1613                  * write final end record.  NB: want to do this even if
1614                  * there was some error, because it might not be totally
1615                  * failed.
1616                  */
1617                 dmu_replay_record_t drr = { 0 };
1618                 drr.drr_type = DRR_END;
1619                 if (write(outfd, &drr, sizeof (drr)) == -1) {
1620                         return (zfs_standard_error(zhp->zfs_hdl,
1621                             errno, errbuf));
1622                 }
1623         }
1624
1625         return (err || sdd.err);
1626
1627 stderr_out:
1628         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1629 err_out:
1630         fsavl_destroy(fsavl);
1631         nvlist_free(fss);
1632         fnvlist_free(sdd.snapholds);
1633
1634         if (sdd.cleanup_fd != -1)
1635                 VERIFY(0 == close(sdd.cleanup_fd));
1636         if (tid != 0) {
1637                 (void) pthread_cancel(tid);
1638                 (void) close(pipefd[0]);
1639                 (void) pthread_join(tid, NULL);
1640         }
1641         return (err);
1642 }
1643
1644 int
1645 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd,
1646     enum lzc_send_flags flags)
1647 {
1648         int err;
1649         libzfs_handle_t *hdl = zhp->zfs_hdl;
1650
1651         char errbuf[1024];
1652         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1653             "warning: cannot send '%s'"), zhp->zfs_name);
1654
1655         err = lzc_send(zhp->zfs_name, from, fd, flags);
1656         if (err != 0) {
1657                 switch (errno) {
1658                 case EXDEV:
1659                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1660                             "not an earlier snapshot from the same fs"));
1661                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
1662
1663                 case ENOENT:
1664                 case ESRCH:
1665                         if (lzc_exists(zhp->zfs_name)) {
1666                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1667                                     "incremental source (%s) does not exist"),
1668                                     from);
1669                         }
1670                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
1671
1672                 case EBUSY:
1673                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1674                             "target is busy; if a filesystem, "
1675                             "it must not be mounted"));
1676                         return (zfs_error(hdl, EZFS_BUSY, errbuf));
1677
1678                 case EDQUOT:
1679                 case EFBIG:
1680                 case EIO:
1681                 case ENOLINK:
1682                 case ENOSPC:
1683 #ifdef illumos
1684                 case ENOSTR:
1685 #endif
1686                 case ENXIO:
1687                 case EPIPE:
1688                 case ERANGE:
1689                 case EFAULT:
1690                 case EROFS:
1691                         zfs_error_aux(hdl, strerror(errno));
1692                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1693
1694                 default:
1695                         return (zfs_standard_error(hdl, errno, errbuf));
1696                 }
1697         }
1698         return (err != 0);
1699 }
1700
1701 /*
1702  * Routines specific to "zfs recv"
1703  */
1704
1705 static int
1706 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1707     boolean_t byteswap, zio_cksum_t *zc)
1708 {
1709         char *cp = buf;
1710         int rv;
1711         int len = ilen;
1712
1713         assert(ilen <= SPA_MAXBLOCKSIZE);
1714
1715         do {
1716                 rv = read(fd, cp, len);
1717                 cp += rv;
1718                 len -= rv;
1719         } while (rv > 0);
1720
1721         if (rv < 0 || len != 0) {
1722                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1723                     "failed to read from stream"));
1724                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1725                     "cannot receive")));
1726         }
1727
1728         if (zc) {
1729                 if (byteswap)
1730                         fletcher_4_incremental_byteswap(buf, ilen, zc);
1731                 else
1732                         fletcher_4_incremental_native(buf, ilen, zc);
1733         }
1734         return (0);
1735 }
1736
1737 static int
1738 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1739     boolean_t byteswap, zio_cksum_t *zc)
1740 {
1741         char *buf;
1742         int err;
1743
1744         buf = zfs_alloc(hdl, len);
1745         if (buf == NULL)
1746                 return (ENOMEM);
1747
1748         err = recv_read(hdl, fd, buf, len, byteswap, zc);
1749         if (err != 0) {
1750                 free(buf);
1751                 return (err);
1752         }
1753
1754         err = nvlist_unpack(buf, len, nvp, 0);
1755         free(buf);
1756         if (err != 0) {
1757                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1758                     "stream (malformed nvlist)"));
1759                 return (EINVAL);
1760         }
1761         return (0);
1762 }
1763
1764 static int
1765 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1766     int baselen, char *newname, recvflags_t *flags)
1767 {
1768         static int seq;
1769         zfs_cmd_t zc = { 0 };
1770         int err;
1771         prop_changelist_t *clp;
1772         zfs_handle_t *zhp;
1773
1774         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1775         if (zhp == NULL)
1776                 return (-1);
1777         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1778             flags->force ? MS_FORCE : 0);
1779         zfs_close(zhp);
1780         if (clp == NULL)
1781                 return (-1);
1782         err = changelist_prefix(clp);
1783         if (err)
1784                 return (err);
1785
1786         zc.zc_objset_type = DMU_OST_ZFS;
1787         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1788
1789         if (tryname) {
1790                 (void) strcpy(newname, tryname);
1791
1792                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1793
1794                 if (flags->verbose) {
1795                         (void) printf("attempting rename %s to %s\n",
1796                             zc.zc_name, zc.zc_value);
1797                 }
1798                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1799                 if (err == 0)
1800                         changelist_rename(clp, name, tryname);
1801         } else {
1802                 err = ENOENT;
1803         }
1804
1805         if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
1806                 seq++;
1807
1808                 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u",
1809                     baselen, name, getpid(), seq);
1810                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1811
1812                 if (flags->verbose) {
1813                         (void) printf("failed - trying rename %s to %s\n",
1814                             zc.zc_name, zc.zc_value);
1815                 }
1816                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1817                 if (err == 0)
1818                         changelist_rename(clp, name, newname);
1819                 if (err && flags->verbose) {
1820                         (void) printf("failed (%u) - "
1821                             "will try again on next pass\n", errno);
1822                 }
1823                 err = EAGAIN;
1824         } else if (flags->verbose) {
1825                 if (err == 0)
1826                         (void) printf("success\n");
1827                 else
1828                         (void) printf("failed (%u)\n", errno);
1829         }
1830
1831         (void) changelist_postfix(clp);
1832         changelist_free(clp);
1833
1834         return (err);
1835 }
1836
1837 static int
1838 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1839     char *newname, recvflags_t *flags)
1840 {
1841         zfs_cmd_t zc = { 0 };
1842         int err = 0;
1843         prop_changelist_t *clp;
1844         zfs_handle_t *zhp;
1845         boolean_t defer = B_FALSE;
1846         int spa_version;
1847
1848         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1849         if (zhp == NULL)
1850                 return (-1);
1851         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1852             flags->force ? MS_FORCE : 0);
1853         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1854             zfs_spa_version(zhp, &spa_version) == 0 &&
1855             spa_version >= SPA_VERSION_USERREFS)
1856                 defer = B_TRUE;
1857         zfs_close(zhp);
1858         if (clp == NULL)
1859                 return (-1);
1860         err = changelist_prefix(clp);
1861         if (err)
1862                 return (err);
1863
1864         zc.zc_objset_type = DMU_OST_ZFS;
1865         zc.zc_defer_destroy = defer;
1866         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1867
1868         if (flags->verbose)
1869                 (void) printf("attempting destroy %s\n", zc.zc_name);
1870         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1871         if (err == 0) {
1872                 if (flags->verbose)
1873                         (void) printf("success\n");
1874                 changelist_remove(clp, zc.zc_name);
1875         }
1876
1877         (void) changelist_postfix(clp);
1878         changelist_free(clp);
1879
1880         /*
1881          * Deferred destroy might destroy the snapshot or only mark it to be
1882          * destroyed later, and it returns success in either case.
1883          */
1884         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1885             ZFS_TYPE_SNAPSHOT))) {
1886                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1887         }
1888
1889         return (err);
1890 }
1891
1892 typedef struct guid_to_name_data {
1893         uint64_t guid;
1894         char *name;
1895         char *skip;
1896 } guid_to_name_data_t;
1897
1898 static int
1899 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1900 {
1901         guid_to_name_data_t *gtnd = arg;
1902         int err;
1903
1904         if (gtnd->skip != NULL &&
1905             strcmp(zhp->zfs_name, gtnd->skip) == 0) {
1906                 return (0);
1907         }
1908
1909         if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1910                 (void) strcpy(gtnd->name, zhp->zfs_name);
1911                 zfs_close(zhp);
1912                 return (EEXIST);
1913         }
1914
1915         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1916         zfs_close(zhp);
1917         return (err);
1918 }
1919
1920 /*
1921  * Attempt to find the local dataset associated with this guid.  In the case of
1922  * multiple matches, we attempt to find the "best" match by searching
1923  * progressively larger portions of the hierarchy.  This allows one to send a
1924  * tree of datasets individually and guarantee that we will find the source
1925  * guid within that hierarchy, even if there are multiple matches elsewhere.
1926  */
1927 static int
1928 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1929     char *name)
1930 {
1931         /* exhaustive search all local snapshots */
1932         char pname[ZFS_MAXNAMELEN];
1933         guid_to_name_data_t gtnd;
1934         int err = 0;
1935         zfs_handle_t *zhp;
1936         char *cp;
1937
1938         gtnd.guid = guid;
1939         gtnd.name = name;
1940         gtnd.skip = NULL;
1941
1942         (void) strlcpy(pname, parent, sizeof (pname));
1943
1944         /*
1945          * Search progressively larger portions of the hierarchy.  This will
1946          * select the "most local" version of the origin snapshot in the case
1947          * that there are multiple matching snapshots in the system.
1948          */
1949         while ((cp = strrchr(pname, '/')) != NULL) {
1950
1951                 /* Chop off the last component and open the parent */
1952                 *cp = '\0';
1953                 zhp = make_dataset_handle(hdl, pname);
1954
1955                 if (zhp == NULL)
1956                         continue;
1957
1958                 err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1959                 zfs_close(zhp);
1960                 if (err == EEXIST)
1961                         return (0);
1962
1963                 /*
1964                  * Remember the dataset that we already searched, so we
1965                  * skip it next time through.
1966                  */
1967                 gtnd.skip = pname;
1968         }
1969
1970         return (ENOENT);
1971 }
1972
1973 /*
1974  * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
1975  * guid1 is after guid2.
1976  */
1977 static int
1978 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1979     uint64_t guid1, uint64_t guid2)
1980 {
1981         nvlist_t *nvfs;
1982         char *fsname, *snapname;
1983         char buf[ZFS_MAXNAMELEN];
1984         int rv;
1985         zfs_handle_t *guid1hdl, *guid2hdl;
1986         uint64_t create1, create2;
1987
1988         if (guid2 == 0)
1989                 return (0);
1990         if (guid1 == 0)
1991                 return (1);
1992
1993         nvfs = fsavl_find(avl, guid1, &snapname);
1994         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1995         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1996         guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1997         if (guid1hdl == NULL)
1998                 return (-1);
1999
2000         nvfs = fsavl_find(avl, guid2, &snapname);
2001         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2002         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2003         guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2004         if (guid2hdl == NULL) {
2005                 zfs_close(guid1hdl);
2006                 return (-1);
2007         }
2008
2009         create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
2010         create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
2011
2012         if (create1 < create2)
2013                 rv = -1;
2014         else if (create1 > create2)
2015                 rv = +1;
2016         else
2017                 rv = 0;
2018
2019         zfs_close(guid1hdl);
2020         zfs_close(guid2hdl);
2021
2022         return (rv);
2023 }
2024
2025 static int
2026 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2027     recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2028     nvlist_t *renamed)
2029 {
2030         nvlist_t *local_nv, *deleted = NULL;
2031         avl_tree_t *local_avl;
2032         nvpair_t *fselem, *nextfselem;
2033         char *fromsnap;
2034         char newname[ZFS_MAXNAMELEN];
2035         char guidname[32];
2036         int error;
2037         boolean_t needagain, progress, recursive;
2038         char *s1, *s2;
2039
2040         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2041
2042         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2043             ENOENT);
2044
2045         if (flags->dryrun)
2046                 return (0);
2047
2048 again:
2049         needagain = progress = B_FALSE;
2050
2051         VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0));
2052
2053         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2054             recursive, &local_nv, &local_avl)) != 0)
2055                 return (error);
2056
2057         /*
2058          * Process deletes and renames
2059          */
2060         for (fselem = nvlist_next_nvpair(local_nv, NULL);
2061             fselem; fselem = nextfselem) {
2062                 nvlist_t *nvfs, *snaps;
2063                 nvlist_t *stream_nvfs = NULL;
2064                 nvpair_t *snapelem, *nextsnapelem;
2065                 uint64_t fromguid = 0;
2066                 uint64_t originguid = 0;
2067                 uint64_t stream_originguid = 0;
2068                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2069                 char *fsname, *stream_fsname;
2070
2071                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2072
2073                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2074                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2075                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2076                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2077                     &parent_fromsnap_guid));
2078                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2079
2080                 /*
2081                  * First find the stream's fs, so we can check for
2082                  * a different origin (due to "zfs promote")
2083                  */
2084                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2085                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2086                         uint64_t thisguid;
2087
2088                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2089                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2090
2091                         if (stream_nvfs != NULL)
2092                                 break;
2093                 }
2094
2095                 /* check for promote */
2096                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2097                     &stream_originguid);
2098                 if (stream_nvfs && originguid != stream_originguid) {
2099                         switch (created_before(hdl, local_avl,
2100                             stream_originguid, originguid)) {
2101                         case 1: {
2102                                 /* promote it! */
2103                                 zfs_cmd_t zc = { 0 };
2104                                 nvlist_t *origin_nvfs;
2105                                 char *origin_fsname;
2106
2107                                 if (flags->verbose)
2108                                         (void) printf("promoting %s\n", fsname);
2109
2110                                 origin_nvfs = fsavl_find(local_avl, originguid,
2111                                     NULL);
2112                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2113                                     "name", &origin_fsname));
2114                                 (void) strlcpy(zc.zc_value, origin_fsname,
2115                                     sizeof (zc.zc_value));
2116                                 (void) strlcpy(zc.zc_name, fsname,
2117                                     sizeof (zc.zc_name));
2118                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2119                                 if (error == 0)
2120                                         progress = B_TRUE;
2121                                 break;
2122                         }
2123                         default:
2124                                 break;
2125                         case -1:
2126                                 fsavl_destroy(local_avl);
2127                                 nvlist_free(local_nv);
2128                                 return (-1);
2129                         }
2130                         /*
2131                          * We had/have the wrong origin, therefore our
2132                          * list of snapshots is wrong.  Need to handle
2133                          * them on the next pass.
2134                          */
2135                         needagain = B_TRUE;
2136                         continue;
2137                 }
2138
2139                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2140                     snapelem; snapelem = nextsnapelem) {
2141                         uint64_t thisguid;
2142                         char *stream_snapname;
2143                         nvlist_t *found, *props;
2144
2145                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2146
2147                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2148                         found = fsavl_find(stream_avl, thisguid,
2149                             &stream_snapname);
2150
2151                         /* check for delete */
2152                         if (found == NULL) {
2153                                 char name[ZFS_MAXNAMELEN];
2154
2155                                 if (!flags->force)
2156                                         continue;
2157
2158                                 (void) snprintf(name, sizeof (name), "%s@%s",
2159                                     fsname, nvpair_name(snapelem));
2160
2161                                 error = recv_destroy(hdl, name,
2162                                     strlen(fsname)+1, newname, flags);
2163                                 if (error)
2164                                         needagain = B_TRUE;
2165                                 else
2166                                         progress = B_TRUE;
2167                                 sprintf(guidname, "%lu", thisguid);
2168                                 nvlist_add_boolean(deleted, guidname);
2169                                 continue;
2170                         }
2171
2172                         stream_nvfs = found;
2173
2174                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2175                             &props) && 0 == nvlist_lookup_nvlist(props,
2176                             stream_snapname, &props)) {
2177                                 zfs_cmd_t zc = { 0 };
2178
2179                                 zc.zc_cookie = B_TRUE; /* received */
2180                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2181                                     "%s@%s", fsname, nvpair_name(snapelem));
2182                                 if (zcmd_write_src_nvlist(hdl, &zc,
2183                                     props) == 0) {
2184                                         (void) zfs_ioctl(hdl,
2185                                             ZFS_IOC_SET_PROP, &zc);
2186                                         zcmd_free_nvlists(&zc);
2187                                 }
2188                         }
2189
2190                         /* check for different snapname */
2191                         if (strcmp(nvpair_name(snapelem),
2192                             stream_snapname) != 0) {
2193                                 char name[ZFS_MAXNAMELEN];
2194                                 char tryname[ZFS_MAXNAMELEN];
2195
2196                                 (void) snprintf(name, sizeof (name), "%s@%s",
2197                                     fsname, nvpair_name(snapelem));
2198                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
2199                                     fsname, stream_snapname);
2200
2201                                 error = recv_rename(hdl, name, tryname,
2202                                     strlen(fsname)+1, newname, flags);
2203                                 if (error)
2204                                         needagain = B_TRUE;
2205                                 else
2206                                         progress = B_TRUE;
2207                         }
2208
2209                         if (strcmp(stream_snapname, fromsnap) == 0)
2210                                 fromguid = thisguid;
2211                 }
2212
2213                 /* check for delete */
2214                 if (stream_nvfs == NULL) {
2215                         if (!flags->force)
2216                                 continue;
2217
2218                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2219                             newname, flags);
2220                         if (error)
2221                                 needagain = B_TRUE;
2222                         else
2223                                 progress = B_TRUE;
2224                         sprintf(guidname, "%lu", parent_fromsnap_guid);
2225                         nvlist_add_boolean(deleted, guidname);
2226                         continue;
2227                 }
2228
2229                 if (fromguid == 0) {
2230                         if (flags->verbose) {
2231                                 (void) printf("local fs %s does not have "
2232                                     "fromsnap (%s in stream); must have "
2233                                     "been deleted locally; ignoring\n",
2234                                     fsname, fromsnap);
2235                         }
2236                         continue;
2237                 }
2238
2239                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2240                     "name", &stream_fsname));
2241                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2242                     "parentfromsnap", &stream_parent_fromsnap_guid));
2243
2244                 s1 = strrchr(fsname, '/');
2245                 s2 = strrchr(stream_fsname, '/');
2246
2247                 /*
2248                  * Check if we're going to rename based on parent guid change
2249                  * and the current parent guid was also deleted. If it was then
2250                  * rename will fail and is likely unneeded, so avoid this and
2251                  * force an early retry to determine the new
2252                  * parent_fromsnap_guid.
2253                  */
2254                 if (stream_parent_fromsnap_guid != 0 &&
2255                     parent_fromsnap_guid != 0 &&
2256                     stream_parent_fromsnap_guid != parent_fromsnap_guid) {
2257                         sprintf(guidname, "%lu", parent_fromsnap_guid);
2258                         if (nvlist_exists(deleted, guidname)) {
2259                                 progress = B_TRUE;
2260                                 needagain = B_TRUE;
2261                                 goto doagain;
2262                         }
2263                 }
2264
2265                 /*
2266                  * Check for rename. If the exact receive path is specified, it
2267                  * does not count as a rename, but we still need to check the
2268                  * datasets beneath it.
2269                  */
2270                 if ((stream_parent_fromsnap_guid != 0 &&
2271                     parent_fromsnap_guid != 0 &&
2272                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2273                     ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2274                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2275                         nvlist_t *parent;
2276                         char tryname[ZFS_MAXNAMELEN];
2277
2278                         parent = fsavl_find(local_avl,
2279                             stream_parent_fromsnap_guid, NULL);
2280                         /*
2281                          * NB: parent might not be found if we used the
2282                          * tosnap for stream_parent_fromsnap_guid,
2283                          * because the parent is a newly-created fs;
2284                          * we'll be able to rename it after we recv the
2285                          * new fs.
2286                          */
2287                         if (parent != NULL) {
2288                                 char *pname;
2289
2290                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
2291                                     &pname));
2292                                 (void) snprintf(tryname, sizeof (tryname),
2293                                     "%s%s", pname, strrchr(stream_fsname, '/'));
2294                         } else {
2295                                 tryname[0] = '\0';
2296                                 if (flags->verbose) {
2297                                         (void) printf("local fs %s new parent "
2298                                             "not found\n", fsname);
2299                                 }
2300                         }
2301
2302                         newname[0] = '\0';
2303
2304                         error = recv_rename(hdl, fsname, tryname,
2305                             strlen(tofs)+1, newname, flags);
2306
2307                         if (renamed != NULL && newname[0] != '\0') {
2308                                 VERIFY(0 == nvlist_add_boolean(renamed,
2309                                     newname));
2310                         }
2311
2312                         if (error)
2313                                 needagain = B_TRUE;
2314                         else
2315                                 progress = B_TRUE;
2316                 }
2317         }
2318
2319 doagain:
2320         fsavl_destroy(local_avl);
2321         nvlist_free(local_nv);
2322         nvlist_free(deleted);
2323
2324         if (needagain && progress) {
2325                 /* do another pass to fix up temporary names */
2326                 if (flags->verbose)
2327                         (void) printf("another pass:\n");
2328                 goto again;
2329         }
2330
2331         return (needagain);
2332 }
2333
2334 static int
2335 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2336     recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2337     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2338 {
2339         nvlist_t *stream_nv = NULL;
2340         avl_tree_t *stream_avl = NULL;
2341         char *fromsnap = NULL;
2342         char *cp;
2343         char tofs[ZFS_MAXNAMELEN];
2344         char sendfs[ZFS_MAXNAMELEN];
2345         char errbuf[1024];
2346         dmu_replay_record_t drre;
2347         int error;
2348         boolean_t anyerr = B_FALSE;
2349         boolean_t softerr = B_FALSE;
2350         boolean_t recursive;
2351
2352         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2353             "cannot receive"));
2354
2355         assert(drr->drr_type == DRR_BEGIN);
2356         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2357         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2358             DMU_COMPOUNDSTREAM);
2359
2360         /*
2361          * Read in the nvlist from the stream.
2362          */
2363         if (drr->drr_payloadlen != 0) {
2364                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2365                     &stream_nv, flags->byteswap, zc);
2366                 if (error) {
2367                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2368                         goto out;
2369                 }
2370         }
2371
2372         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2373             ENOENT);
2374
2375         if (recursive && strchr(destname, '@')) {
2376                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2377                     "cannot specify snapshot name for multi-snapshot stream"));
2378                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2379                 goto out;
2380         }
2381
2382         /*
2383          * Read in the end record and verify checksum.
2384          */
2385         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2386             flags->byteswap, NULL)))
2387                 goto out;
2388         if (flags->byteswap) {
2389                 drre.drr_type = BSWAP_32(drre.drr_type);
2390                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2391                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2392                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2393                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2394                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2395                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2396                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2397                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2398         }
2399         if (drre.drr_type != DRR_END) {
2400                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2401                 goto out;
2402         }
2403         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2404                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2405                     "incorrect header checksum"));
2406                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2407                 goto out;
2408         }
2409
2410         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2411
2412         if (drr->drr_payloadlen != 0) {
2413                 nvlist_t *stream_fss;
2414
2415                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2416                     &stream_fss));
2417                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2418                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2419                             "couldn't allocate avl tree"));
2420                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2421                         goto out;
2422                 }
2423
2424                 if (fromsnap != NULL) {
2425                         nvlist_t *renamed = NULL;
2426                         nvpair_t *pair = NULL;
2427
2428                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2429                         if (flags->isprefix) {
2430                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2431                                 int i;
2432
2433                                 if (flags->istail) {
2434                                         cp = strrchr(drrb->drr_toname, '/');
2435                                         if (cp == NULL) {
2436                                                 (void) strlcat(tofs, "/",
2437                                                     ZFS_MAXNAMELEN);
2438                                                 i = 0;
2439                                         } else {
2440                                                 i = (cp - drrb->drr_toname);
2441                                         }
2442                                 } else {
2443                                         i = strcspn(drrb->drr_toname, "/@");
2444                                 }
2445                                 /* zfs_receive_one() will create_parents() */
2446                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2447                                     ZFS_MAXNAMELEN);
2448                                 *strchr(tofs, '@') = '\0';
2449                         }
2450
2451                         if (recursive && !flags->dryrun && !flags->nomount) {
2452                                 VERIFY(0 == nvlist_alloc(&renamed,
2453                                     NV_UNIQUE_NAME, 0));
2454                         }
2455
2456                         softerr = recv_incremental_replication(hdl, tofs, flags,
2457                             stream_nv, stream_avl, renamed);
2458
2459                         /* Unmount renamed filesystems before receiving. */
2460                         while ((pair = nvlist_next_nvpair(renamed,
2461                             pair)) != NULL) {
2462                                 zfs_handle_t *zhp;
2463                                 prop_changelist_t *clp = NULL;
2464
2465                                 zhp = zfs_open(hdl, nvpair_name(pair),
2466                                     ZFS_TYPE_FILESYSTEM);
2467                                 if (zhp != NULL) {
2468                                         clp = changelist_gather(zhp,
2469                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2470                                         zfs_close(zhp);
2471                                         if (clp != NULL) {
2472                                                 softerr |=
2473                                                     changelist_prefix(clp);
2474                                                 changelist_free(clp);
2475                                         }
2476                                 }
2477                         }
2478
2479                         nvlist_free(renamed);
2480                 }
2481         }
2482
2483         /*
2484          * Get the fs specified by the first path in the stream (the top level
2485          * specified by 'zfs send') and pass it to each invocation of
2486          * zfs_receive_one().
2487          */
2488         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2489             ZFS_MAXNAMELEN);
2490         if ((cp = strchr(sendfs, '@')) != NULL)
2491                 *cp = '\0';
2492
2493         /* Finally, receive each contained stream */
2494         do {
2495                 /*
2496                  * we should figure out if it has a recoverable
2497                  * error, in which case do a recv_skip() and drive on.
2498                  * Note, if we fail due to already having this guid,
2499                  * zfs_receive_one() will take care of it (ie,
2500                  * recv_skip() and return 0).
2501                  */
2502                 error = zfs_receive_impl(hdl, destname, NULL, flags, fd,
2503                     sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2504                     action_handlep);
2505                 if (error == ENODATA) {
2506                         error = 0;
2507                         break;
2508                 }
2509                 anyerr |= error;
2510         } while (error == 0);
2511
2512         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2513                 /*
2514                  * Now that we have the fs's they sent us, try the
2515                  * renames again.
2516                  */
2517                 softerr = recv_incremental_replication(hdl, tofs, flags,
2518                     stream_nv, stream_avl, NULL);
2519         }
2520
2521 out:
2522         fsavl_destroy(stream_avl);
2523         if (stream_nv)
2524                 nvlist_free(stream_nv);
2525         if (softerr)
2526                 error = -2;
2527         if (anyerr)
2528                 error = -1;
2529         return (error);
2530 }
2531
2532 static void
2533 trunc_prop_errs(int truncated)
2534 {
2535         ASSERT(truncated != 0);
2536
2537         if (truncated == 1)
2538                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2539                     "1 more property could not be set\n"));
2540         else
2541                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2542                     "%d more properties could not be set\n"), truncated);
2543 }
2544
2545 static int
2546 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2547 {
2548         dmu_replay_record_t *drr;
2549         void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
2550         char errbuf[1024];
2551
2552         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2553             "cannot receive:"));
2554
2555         /* XXX would be great to use lseek if possible... */
2556         drr = buf;
2557
2558         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2559             byteswap, NULL) == 0) {
2560                 if (byteswap)
2561                         drr->drr_type = BSWAP_32(drr->drr_type);
2562
2563                 switch (drr->drr_type) {
2564                 case DRR_BEGIN:
2565                         /* NB: not to be used on v2 stream packages */
2566                         if (drr->drr_payloadlen != 0) {
2567                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2568                                     "invalid substream header"));
2569                                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2570                         }
2571                         break;
2572
2573                 case DRR_END:
2574                         free(buf);
2575                         return (0);
2576
2577                 case DRR_OBJECT:
2578                         if (byteswap) {
2579                                 drr->drr_u.drr_object.drr_bonuslen =
2580                                     BSWAP_32(drr->drr_u.drr_object.
2581                                     drr_bonuslen);
2582                         }
2583                         (void) recv_read(hdl, fd, buf,
2584                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2585                             B_FALSE, NULL);
2586                         break;
2587
2588                 case DRR_WRITE:
2589                         if (byteswap) {
2590                                 drr->drr_u.drr_write.drr_length =
2591                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
2592                         }
2593                         (void) recv_read(hdl, fd, buf,
2594                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2595                         break;
2596                 case DRR_SPILL:
2597                         if (byteswap) {
2598                                 drr->drr_u.drr_write.drr_length =
2599                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
2600                         }
2601                         (void) recv_read(hdl, fd, buf,
2602                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2603                         break;
2604                 case DRR_WRITE_EMBEDDED:
2605                         if (byteswap) {
2606                                 drr->drr_u.drr_write_embedded.drr_psize =
2607                                     BSWAP_32(drr->drr_u.drr_write_embedded.
2608                                     drr_psize);
2609                         }
2610                         (void) recv_read(hdl, fd, buf,
2611                             P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
2612                             8), B_FALSE, NULL);
2613                         break;
2614                 case DRR_WRITE_BYREF:
2615                 case DRR_FREEOBJECTS:
2616                 case DRR_FREE:
2617                         break;
2618
2619                 default:
2620                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2621                             "invalid record type"));
2622                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2623                 }
2624         }
2625
2626         free(buf);
2627         return (-1);
2628 }
2629
2630 /*
2631  * Restores a backup of tosnap from the file descriptor specified by infd.
2632  */
2633 static int
2634 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2635     const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr,
2636     dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv,
2637     avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2638     uint64_t *action_handlep)
2639 {
2640         zfs_cmd_t zc = { 0 };
2641         time_t begin_time;
2642         int ioctl_err, ioctl_errno, err;
2643         char *cp;
2644         struct drr_begin *drrb = &drr->drr_u.drr_begin;
2645         char errbuf[1024];
2646         char prop_errbuf[1024];
2647         const char *chopprefix;
2648         boolean_t newfs = B_FALSE;
2649         boolean_t stream_wantsnewfs;
2650         uint64_t parent_snapguid = 0;
2651         prop_changelist_t *clp = NULL;
2652         nvlist_t *snapprops_nvlist = NULL;
2653         zprop_errflags_t prop_errflags;
2654         boolean_t recursive;
2655
2656         begin_time = time(NULL);
2657
2658         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2659             "cannot receive"));
2660
2661         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2662             ENOENT);
2663
2664         if (stream_avl != NULL) {
2665                 char *snapname;
2666                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2667                     &snapname);
2668                 nvlist_t *props;
2669                 int ret;
2670
2671                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2672                     &parent_snapguid);
2673                 err = nvlist_lookup_nvlist(fs, "props", &props);
2674                 if (err)
2675                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2676
2677                 if (flags->canmountoff) {
2678                         VERIFY(0 == nvlist_add_uint64(props,
2679                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2680                 }
2681                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2682                 if (err)
2683                         nvlist_free(props);
2684
2685                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2686                         VERIFY(0 == nvlist_lookup_nvlist(props,
2687                             snapname, &snapprops_nvlist));
2688                 }
2689
2690                 if (ret != 0)
2691                         return (-1);
2692         }
2693
2694         cp = NULL;
2695
2696         /*
2697          * Determine how much of the snapshot name stored in the stream
2698          * we are going to tack on to the name they specified on the
2699          * command line, and how much we are going to chop off.
2700          *
2701          * If they specified a snapshot, chop the entire name stored in
2702          * the stream.
2703          */
2704         if (flags->istail) {
2705                 /*
2706                  * A filesystem was specified with -e. We want to tack on only
2707                  * the tail of the sent snapshot path.
2708                  */
2709                 if (strchr(tosnap, '@')) {
2710                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2711                             "argument - snapshot not allowed with -e"));
2712                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2713                 }
2714
2715                 chopprefix = strrchr(sendfs, '/');
2716
2717                 if (chopprefix == NULL) {
2718                         /*
2719                          * The tail is the poolname, so we need to
2720                          * prepend a path separator.
2721                          */
2722                         int len = strlen(drrb->drr_toname);
2723                         cp = malloc(len + 2);
2724                         cp[0] = '/';
2725                         (void) strcpy(&cp[1], drrb->drr_toname);
2726                         chopprefix = cp;
2727                 } else {
2728                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2729                 }
2730         } else if (flags->isprefix) {
2731                 /*
2732                  * A filesystem was specified with -d. We want to tack on
2733                  * everything but the first element of the sent snapshot path
2734                  * (all but the pool name).
2735                  */
2736                 if (strchr(tosnap, '@')) {
2737                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2738                             "argument - snapshot not allowed with -d"));
2739                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2740                 }
2741
2742                 chopprefix = strchr(drrb->drr_toname, '/');
2743                 if (chopprefix == NULL)
2744                         chopprefix = strchr(drrb->drr_toname, '@');
2745         } else if (strchr(tosnap, '@') == NULL) {
2746                 /*
2747                  * If a filesystem was specified without -d or -e, we want to
2748                  * tack on everything after the fs specified by 'zfs send'.
2749                  */
2750                 chopprefix = drrb->drr_toname + strlen(sendfs);
2751         } else {
2752                 /* A snapshot was specified as an exact path (no -d or -e). */
2753                 if (recursive) {
2754                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2755                             "cannot specify snapshot name for multi-snapshot "
2756                             "stream"));
2757                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2758                 }
2759                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2760         }
2761
2762         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2763         ASSERT(chopprefix > drrb->drr_toname);
2764         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2765         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2766             chopprefix[0] == '\0');
2767
2768         /*
2769          * Determine name of destination snapshot, store in zc_value.
2770          */
2771         (void) strcpy(zc.zc_value, tosnap);
2772         (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2773 #ifdef __FreeBSD__
2774         if (zfs_ioctl_version == ZFS_IOCVER_UNDEF)
2775                 zfs_ioctl_version = get_zfs_ioctl_version();
2776         /*
2777          * For forward compatibility hide tosnap in zc_value
2778          */
2779         if (zfs_ioctl_version < ZFS_IOCVER_LZC)
2780                 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap);
2781 #endif
2782         free(cp);
2783         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2784                 zcmd_free_nvlists(&zc);
2785                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2786         }
2787
2788         /*
2789          * Determine the name of the origin snapshot, store in zc_string.
2790          */
2791         if (drrb->drr_flags & DRR_FLAG_CLONE) {
2792                 if (guid_to_name(hdl, zc.zc_value,
2793                     drrb->drr_fromguid, zc.zc_string) != 0) {
2794                         zcmd_free_nvlists(&zc);
2795                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2796                             "local origin for clone %s does not exist"),
2797                             zc.zc_value);
2798                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2799                 }
2800                 if (flags->verbose)
2801                         (void) printf("found clone origin %s\n", zc.zc_string);
2802         } else if (originsnap) {
2803                 (void) strncpy(zc.zc_string, originsnap, ZFS_MAXNAMELEN);
2804                 if (flags->verbose)
2805                         (void) printf("using provided clone origin %s\n",
2806                             zc.zc_string);
2807         }
2808
2809         stream_wantsnewfs = (drrb->drr_fromguid == 0 ||
2810             (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap);
2811
2812         if (stream_wantsnewfs) {
2813                 /*
2814                  * if the parent fs does not exist, look for it based on
2815                  * the parent snap GUID
2816                  */
2817                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2818                     "cannot receive new filesystem stream"));
2819
2820                 (void) strcpy(zc.zc_name, zc.zc_value);
2821                 cp = strrchr(zc.zc_name, '/');
2822                 if (cp)
2823                         *cp = '\0';
2824                 if (cp &&
2825                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2826                         char suffix[ZFS_MAXNAMELEN];
2827                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2828                         if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
2829                             zc.zc_value) == 0) {
2830                                 *strchr(zc.zc_value, '@') = '\0';
2831                                 (void) strcat(zc.zc_value, suffix);
2832                         }
2833                 }
2834         } else {
2835                 /*
2836                  * if the fs does not exist, look for it based on the
2837                  * fromsnap GUID
2838                  */
2839                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2840                     "cannot receive incremental stream"));
2841
2842                 (void) strcpy(zc.zc_name, zc.zc_value);
2843                 *strchr(zc.zc_name, '@') = '\0';
2844
2845                 /*
2846                  * If the exact receive path was specified and this is the
2847                  * topmost path in the stream, then if the fs does not exist we
2848                  * should look no further.
2849                  */
2850                 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
2851                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2852                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2853                         char snap[ZFS_MAXNAMELEN];
2854                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
2855                         if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
2856                             zc.zc_value) == 0) {
2857                                 *strchr(zc.zc_value, '@') = '\0';
2858                                 (void) strcat(zc.zc_value, snap);
2859                         }
2860                 }
2861         }
2862
2863         (void) strcpy(zc.zc_name, zc.zc_value);
2864         *strchr(zc.zc_name, '@') = '\0';
2865
2866         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2867                 zfs_handle_t *zhp;
2868
2869                 /*
2870                  * Destination fs exists.  Therefore this should either
2871                  * be an incremental, or the stream specifies a new fs
2872                  * (full stream or clone) and they want us to blow it
2873                  * away (and have therefore specified -F and removed any
2874                  * snapshots).
2875                  */
2876                 if (stream_wantsnewfs) {
2877                         if (!flags->force) {
2878                                 zcmd_free_nvlists(&zc);
2879                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2880                                     "destination '%s' exists\n"
2881                                     "must specify -F to overwrite it"),
2882                                     zc.zc_name);
2883                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2884                         }
2885                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2886                             &zc) == 0) {
2887                                 zcmd_free_nvlists(&zc);
2888                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2889                                     "destination has snapshots (eg. %s)\n"
2890                                     "must destroy them to overwrite it"),
2891                                     zc.zc_name);
2892                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2893                         }
2894                 }
2895
2896                 if ((zhp = zfs_open(hdl, zc.zc_name,
2897                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2898                         zcmd_free_nvlists(&zc);
2899                         return (-1);
2900                 }
2901
2902                 if (stream_wantsnewfs &&
2903                     zhp->zfs_dmustats.dds_origin[0]) {
2904                         zcmd_free_nvlists(&zc);
2905                         zfs_close(zhp);
2906                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2907                             "destination '%s' is a clone\n"
2908                             "must destroy it to overwrite it"),
2909                             zc.zc_name);
2910                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2911                 }
2912
2913                 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2914                     stream_wantsnewfs) {
2915                         /* We can't do online recv in this case */
2916                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2917                         if (clp == NULL) {
2918                                 zfs_close(zhp);
2919                                 zcmd_free_nvlists(&zc);
2920                                 return (-1);
2921                         }
2922                         if (changelist_prefix(clp) != 0) {
2923                                 changelist_free(clp);
2924                                 zfs_close(zhp);
2925                                 zcmd_free_nvlists(&zc);
2926                                 return (-1);
2927                         }
2928                 }
2929                 zfs_close(zhp);
2930         } else {
2931                 /*
2932                  * Destination filesystem does not exist.  Therefore we better
2933                  * be creating a new filesystem (either from a full backup, or
2934                  * a clone).  It would therefore be invalid if the user
2935                  * specified only the pool name (i.e. if the destination name
2936                  * contained no slash character).
2937                  */
2938                 if (!stream_wantsnewfs ||
2939                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
2940                         zcmd_free_nvlists(&zc);
2941                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2942                             "destination '%s' does not exist"), zc.zc_name);
2943                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2944                 }
2945
2946                 /*
2947                  * Trim off the final dataset component so we perform the
2948                  * recvbackup ioctl to the filesystems's parent.
2949                  */
2950                 *cp = '\0';
2951
2952                 if (flags->isprefix && !flags->istail && !flags->dryrun &&
2953                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2954                         zcmd_free_nvlists(&zc);
2955                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2956                 }
2957
2958                 newfs = B_TRUE;
2959         }
2960
2961         zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2962         zc.zc_cookie = infd;
2963         zc.zc_guid = flags->force;
2964         if (flags->verbose) {
2965                 (void) printf("%s %s stream of %s into %s\n",
2966                     flags->dryrun ? "would receive" : "receiving",
2967                     drrb->drr_fromguid ? "incremental" : "full",
2968                     drrb->drr_toname, zc.zc_value);
2969                 (void) fflush(stdout);
2970         }
2971
2972         if (flags->dryrun) {
2973                 zcmd_free_nvlists(&zc);
2974                 return (recv_skip(hdl, infd, flags->byteswap));
2975         }
2976
2977         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2978         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2979         zc.zc_cleanup_fd = cleanup_fd;
2980         zc.zc_action_handle = *action_handlep;
2981
2982         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2983         ioctl_errno = errno;
2984         prop_errflags = (zprop_errflags_t)zc.zc_obj;
2985
2986         if (err == 0) {
2987                 nvlist_t *prop_errors;
2988                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2989                     zc.zc_nvlist_dst_size, &prop_errors, 0));
2990
2991                 nvpair_t *prop_err = NULL;
2992
2993                 while ((prop_err = nvlist_next_nvpair(prop_errors,
2994                     prop_err)) != NULL) {
2995                         char tbuf[1024];
2996                         zfs_prop_t prop;
2997                         int intval;
2998
2999                         prop = zfs_name_to_prop(nvpair_name(prop_err));
3000                         (void) nvpair_value_int32(prop_err, &intval);
3001                         if (strcmp(nvpair_name(prop_err),
3002                             ZPROP_N_MORE_ERRORS) == 0) {
3003                                 trunc_prop_errs(intval);
3004                                 break;
3005                         } else {
3006                                 (void) snprintf(tbuf, sizeof (tbuf),
3007                                     dgettext(TEXT_DOMAIN,
3008                                     "cannot receive %s property on %s"),
3009                                     nvpair_name(prop_err), zc.zc_name);
3010                                 zfs_setprop_error(hdl, prop, intval, tbuf);
3011                         }
3012                 }
3013                 nvlist_free(prop_errors);
3014         }
3015
3016         zc.zc_nvlist_dst = 0;
3017         zc.zc_nvlist_dst_size = 0;
3018         zcmd_free_nvlists(&zc);
3019
3020         if (err == 0 && snapprops_nvlist) {
3021                 zfs_cmd_t zc2 = { 0 };
3022
3023                 (void) strcpy(zc2.zc_name, zc.zc_value);
3024                 zc2.zc_cookie = B_TRUE; /* received */
3025                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
3026                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
3027                         zcmd_free_nvlists(&zc2);
3028                 }
3029         }
3030
3031         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
3032                 /*
3033                  * It may be that this snapshot already exists,
3034                  * in which case we want to consume & ignore it
3035                  * rather than failing.
3036                  */
3037                 avl_tree_t *local_avl;
3038                 nvlist_t *local_nv, *fs;
3039                 cp = strchr(zc.zc_value, '@');
3040
3041                 /*
3042                  * XXX Do this faster by just iterating over snaps in
3043                  * this fs.  Also if zc_value does not exist, we will
3044                  * get a strange "does not exist" error message.
3045                  */
3046                 *cp = '\0';
3047                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3048                     &local_nv, &local_avl) == 0) {
3049                         *cp = '@';
3050                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3051                         fsavl_destroy(local_avl);
3052                         nvlist_free(local_nv);
3053
3054                         if (fs != NULL) {
3055                                 if (flags->verbose) {
3056                                         (void) printf("snap %s already exists; "
3057                                             "ignoring\n", zc.zc_value);
3058                                 }
3059                                 err = ioctl_err = recv_skip(hdl, infd,
3060                                     flags->byteswap);
3061                         }
3062                 }
3063                 *cp = '@';
3064         }
3065
3066         if (ioctl_err != 0) {
3067                 switch (ioctl_errno) {
3068                 case ENODEV:
3069                         cp = strchr(zc.zc_value, '@');
3070                         *cp = '\0';
3071                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3072                             "most recent snapshot of %s does not\n"
3073                             "match incremental source"), zc.zc_value);
3074                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3075                         *cp = '@';
3076                         break;
3077                 case ETXTBSY:
3078                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3079                             "destination %s has been modified\n"
3080                             "since most recent snapshot"), zc.zc_name);
3081                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3082                         break;
3083                 case EEXIST:
3084                         cp = strchr(zc.zc_value, '@');
3085                         if (newfs) {
3086                                 /* it's the containing fs that exists */
3087                                 *cp = '\0';
3088                         }
3089                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3090                             "destination already exists"));
3091                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3092                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3093                             zc.zc_value);
3094                         *cp = '@';
3095                         break;
3096                 case EINVAL:
3097                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3098                         break;
3099                 case ECKSUM:
3100                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3101                             "invalid stream (checksum mismatch)"));
3102                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3103                         break;
3104                 case ENOTSUP:
3105                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3106                             "pool must be upgraded to receive this stream."));
3107                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3108                         break;
3109                 case EDQUOT:
3110                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3111                             "destination %s space quota exceeded"), zc.zc_name);
3112                         (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3113                         break;
3114                 default:
3115                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3116                 }
3117         }
3118
3119         /*
3120          * Mount the target filesystem (if created).  Also mount any
3121          * children of the target filesystem if we did a replication
3122          * receive (indicated by stream_avl being non-NULL).
3123          */
3124         cp = strchr(zc.zc_value, '@');
3125         if (cp && (ioctl_err == 0 || !newfs)) {
3126                 zfs_handle_t *h;
3127
3128                 *cp = '\0';
3129                 h = zfs_open(hdl, zc.zc_value,
3130                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3131                 if (h != NULL) {
3132                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
3133                                 *cp = '@';
3134                         } else if (newfs || stream_avl) {
3135                                 /*
3136                                  * Track the first/top of hierarchy fs,
3137                                  * for mounting and sharing later.
3138                                  */
3139                                 if (top_zfs && *top_zfs == NULL)
3140                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
3141                         }
3142                         zfs_close(h);
3143                 }
3144                 *cp = '@';
3145         }
3146
3147         if (clp) {
3148                 err |= changelist_postfix(clp);
3149                 changelist_free(clp);
3150         }
3151
3152         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3153                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3154                     "failed to clear unreceived properties on %s"),
3155                     zc.zc_name);
3156                 (void) fprintf(stderr, "\n");
3157         }
3158         if (prop_errflags & ZPROP_ERR_NORESTORE) {
3159                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3160                     "failed to restore original properties on %s"),
3161                     zc.zc_name);
3162                 (void) fprintf(stderr, "\n");
3163         }
3164
3165         if (err || ioctl_err)
3166                 return (-1);
3167
3168         *action_handlep = zc.zc_action_handle;
3169
3170         if (flags->verbose) {
3171                 char buf1[64];
3172                 char buf2[64];
3173                 uint64_t bytes = zc.zc_cookie;
3174                 time_t delta = time(NULL) - begin_time;
3175                 if (delta == 0)
3176                         delta = 1;
3177                 zfs_nicenum(bytes, buf1, sizeof (buf1));
3178                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3179
3180                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3181                     buf1, delta, buf2);
3182         }
3183
3184         return (0);
3185 }
3186
3187 static int
3188 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap,
3189     const char *originsnap, recvflags_t *flags, int infd, const char *sendfs,
3190     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
3191     uint64_t *action_handlep)
3192 {
3193         int err;
3194         dmu_replay_record_t drr, drr_noswap;
3195         struct drr_begin *drrb = &drr.drr_u.drr_begin;
3196         char errbuf[1024];
3197         zio_cksum_t zcksum = { 0 };
3198         uint64_t featureflags;
3199         int hdrtype;
3200
3201         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3202             "cannot receive"));
3203
3204         if (flags->isprefix &&
3205             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3206                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3207                     "(%s) does not exist"), tosnap);
3208                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3209         }
3210         if (originsnap &&
3211             !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) {
3212                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs "
3213                     "(%s) does not exist"), originsnap);
3214                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3215         }
3216
3217         /* read in the BEGIN record */
3218         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3219             &zcksum)))
3220                 return (err);
3221
3222         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3223                 /* It's the double end record at the end of a package */
3224                 return (ENODATA);
3225         }
3226
3227         /* the kernel needs the non-byteswapped begin record */
3228         drr_noswap = drr;
3229
3230         flags->byteswap = B_FALSE;
3231         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3232                 /*
3233                  * We computed the checksum in the wrong byteorder in
3234                  * recv_read() above; do it again correctly.
3235                  */
3236                 bzero(&zcksum, sizeof (zio_cksum_t));
3237                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3238                 flags->byteswap = B_TRUE;
3239
3240                 drr.drr_type = BSWAP_32(drr.drr_type);
3241                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3242                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3243                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3244                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3245                 drrb->drr_type = BSWAP_32(drrb->drr_type);
3246                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3247                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3248                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3249         }
3250
3251         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3252                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3253                     "stream (bad magic number)"));
3254                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3255         }
3256
3257         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3258         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3259
3260         if (!DMU_STREAM_SUPPORTED(featureflags) ||
3261             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3262                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3263                     "stream has unsupported feature, feature flags = %lx"),
3264                     featureflags);
3265                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3266         }
3267
3268         if (strchr(drrb->drr_toname, '@') == NULL) {
3269                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3270                     "stream (bad snapshot name)"));
3271                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3272         }
3273
3274         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3275                 char nonpackage_sendfs[ZFS_MAXNAMELEN];
3276                 if (sendfs == NULL) {
3277                         /*
3278                          * We were not called from zfs_receive_package(). Get
3279                          * the fs specified by 'zfs send'.
3280                          */
3281                         char *cp;
3282                         (void) strlcpy(nonpackage_sendfs,
3283                             drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
3284                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3285                                 *cp = '\0';
3286                         sendfs = nonpackage_sendfs;
3287                 }
3288                 return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags,
3289                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs,
3290                     cleanup_fd, action_handlep));
3291         } else {
3292                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3293                     DMU_COMPOUNDSTREAM);
3294                 return (zfs_receive_package(hdl, infd, tosnap, flags, &drr,
3295                     &zcksum, top_zfs, cleanup_fd, action_handlep));
3296         }
3297 }
3298
3299 /*
3300  * Restores a backup of tosnap from the file descriptor specified by infd.
3301  * Return 0 on total success, -2 if some things couldn't be
3302  * destroyed/renamed/promoted, -1 if some things couldn't be received.
3303  * (-1 will override -2).
3304  */
3305 int
3306 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props,
3307     recvflags_t *flags, int infd, avl_tree_t *stream_avl)
3308 {
3309         char *top_zfs = NULL;
3310         int err;
3311         int cleanup_fd;
3312         uint64_t action_handle = 0;
3313         char *originsnap = NULL;
3314         if (props) {
3315                 err = nvlist_lookup_string(props, "origin", &originsnap);
3316                 if (err && err != ENOENT)
3317                         return (err);
3318         }
3319
3320         cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3321         VERIFY(cleanup_fd >= 0);
3322
3323         err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL,
3324             stream_avl, &top_zfs, cleanup_fd, &action_handle);
3325
3326         VERIFY(0 == close(cleanup_fd));
3327
3328         if (err == 0 && !flags->nomount && top_zfs) {
3329                 zfs_handle_t *zhp;
3330                 prop_changelist_t *clp;
3331
3332                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3333                 if (zhp != NULL) {
3334                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3335                             CL_GATHER_MOUNT_ALWAYS, 0);
3336                         zfs_close(zhp);
3337                         if (clp != NULL) {
3338                                 /* mount and share received datasets */
3339                                 err = changelist_postfix(clp);
3340                                 changelist_free(clp);
3341                         }
3342                 }
3343                 if (zhp == NULL || clp == NULL || err)
3344                         err = -1;
3345         }
3346         if (top_zfs)
3347                 free(top_zfs);
3348
3349         return (err);
3350 }