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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2012 Pawel Jakub Dawidek. All rights reserved.
  27  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  28  * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
  29  * Copyright (c) 2014 Integros [integros.com]
  30  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
  31  */
  32
  33 #include <assert.h>
  34 #include <ctype.h>
  35 #include <errno.h>
  36 #include <libintl.h>
  37 #include <stdio.h>
  38 #include <stdlib.h>
  39 #include <strings.h>
  40 #include <unistd.h>
  41 #include <stddef.h>
  42 #include <fcntl.h>
  43 #include <sys/param.h>
  44 #include <sys/mount.h>
  45 #include <pthread.h>
  46 #include <umem.h>
  47 #include <time.h>
  48
  49 #include <libzfs.h>
  50 #include <libzfs_core.h>
  51
  52 #include "zfs_namecheck.h"
  53 #include "zfs_prop.h"
  54 #include "zfs_fletcher.h"
  55 #include "libzfs_impl.h"
  56 #include <zlib.h>
  57 #include <sha2.h>
  58 #include <sys/zio_checksum.h>
  59 #include <sys/ddt.h>
  60
  61 #ifdef __FreeBSD__
  62 extern int zfs_ioctl_version;
  63 #endif
  64
  65 /* in libzfs_dataset.c */
  66 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  67 /* We need to use something for ENODATA. */
  68 #define ENODATA EIDRM
  69
  70 static int zfs_receive_impl(libzfs_handle_t *, const char *, const char *,
  71     recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **, int,
  72     uint64_t *, const char *);
  73 static int guid_to_name(libzfs_handle_t *, const char *,
  74     uint64_t, boolean_t, char *);
  75
  76 static const zio_cksum_t zero_cksum = { 0 };
  77
  78 typedef struct dedup_arg {
  79         int     inputfd;
  80         int     outputfd;
  81         libzfs_handle_t  *dedup_hdl;
  82 } dedup_arg_t;
  83
  84 typedef struct progress_arg {
  85         zfs_handle_t *pa_zhp;
  86         int pa_fd;
  87         boolean_t pa_parsable;
  88         boolean_t pa_astitle;
  89         uint64_t pa_size;
  90 } progress_arg_t;
  91
  92 typedef struct dataref {
  93         uint64_t ref_guid;
  94         uint64_t ref_object;
  95         uint64_t ref_offset;
  96 } dataref_t;
  97
  98 typedef struct dedup_entry {
  99         struct dedup_entry      *dde_next;
 100         zio_cksum_t dde_chksum;
 101         uint64_t dde_prop;
 102         dataref_t dde_ref;
 103 } dedup_entry_t;
 104
 105 #define MAX_DDT_PHYSMEM_PERCENT         20
 106 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
 107
 108 typedef struct dedup_table {
 109         dedup_entry_t   **dedup_hash_array;
 110         umem_cache_t    *ddecache;
 111         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
 112         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
 113         uint64_t        ddt_count;
 114         int             numhashbits;
 115         boolean_t       ddt_full;
 116 } dedup_table_t;
 117
 118 static int
 119 high_order_bit(uint64_t n)
 120 {
 121         int count;
 122
 123         for (count = 0; n != 0; count++)
 124                 n >>= 1;
 125         return (count);
 126 }
 127
 128 static size_t
 129 ssread(void *buf, size_t len, FILE *stream)
 130 {
 131         size_t outlen;
 132
 133         if ((outlen = fread(buf, len, 1, stream)) == 0)
 134                 return (0);
 135
 136         return (outlen);
 137 }
 138
 139 static void
 140 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 141     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 142 {
 143         dedup_entry_t   *dde;
 144
 145         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 146                 if (ddt->ddt_full == B_FALSE) {
 147                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 148                             "Dedup table full.  Deduplication will continue "
 149                             "with existing table entries"));
 150                         ddt->ddt_full = B_TRUE;
 151                 }
 152                 return;
 153         }
 154
 155         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 156             != NULL) {
 157                 assert(*ddepp == NULL);
 158                 dde->dde_next = NULL;
 159                 dde->dde_chksum = *cs;
 160                 dde->dde_prop = prop;
 161                 dde->dde_ref = *dr;
 162                 *ddepp = dde;
 163                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 164                 ddt->ddt_count++;
 165         }
 166 }
 167
 168 /*
 169  * Using the specified dedup table, do a lookup for an entry with
 170  * the checksum cs.  If found, return the block's reference info
 171  * in *dr. Otherwise, insert a new entry in the dedup table, using
 172  * the reference information specified by *dr.
 173  *
 174  * return value:  true - entry was found
 175  *                false - entry was not found
 176  */
 177 static boolean_t
 178 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 179     uint64_t prop, dataref_t *dr)
 180 {
 181         uint32_t hashcode;
 182         dedup_entry_t **ddepp;
 183
 184         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 185
 186         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 187             ddepp = &((*ddepp)->dde_next)) {
 188                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 189                     (*ddepp)->dde_prop == prop) {
 190                         *dr = (*ddepp)->dde_ref;
 191                         return (B_TRUE);
 192                 }
 193         }
 194         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 195         return (B_FALSE);
 196 }
 197
 198 static int
 199 dump_record(dmu_replay_record_t *drr, void *payload, int payload_len,
 200     zio_cksum_t *zc, int outfd)
 201 {
 202         ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
 203             ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t));
 204         (void) fletcher_4_incremental_native(drr,
 205             offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc);
 206         if (drr->drr_type != DRR_BEGIN) {
 207                 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u.
 208                     drr_checksum.drr_checksum));
 209                 drr->drr_u.drr_checksum.drr_checksum = *zc;
 210         }
 211         (void) fletcher_4_incremental_native(
 212             &drr->drr_u.drr_checksum.drr_checksum, sizeof (zio_cksum_t), zc);
 213         if (write(outfd, drr, sizeof (*drr)) == -1)
 214                 return (errno);
 215         if (payload_len != 0) {
 216                 (void) fletcher_4_incremental_native(payload, payload_len, zc);
 217                 if (write(outfd, payload, payload_len) == -1)
 218                         return (errno);
 219         }
 220         return (0);
 221 }
 222
 223 /*
 224  * This function is started in a separate thread when the dedup option
 225  * has been requested.  The main send thread determines the list of
 226  * snapshots to be included in the send stream and makes the ioctl calls
 227  * for each one.  But instead of having the ioctl send the output to the
 228  * the output fd specified by the caller of zfs_send()), the
 229  * ioctl is told to direct the output to a pipe, which is read by the
 230  * alternate thread running THIS function.  This function does the
 231  * dedup'ing by:
 232  *  1. building a dedup table (the DDT)
 233  *  2. doing checksums on each data block and inserting a record in the DDT
 234  *  3. looking for matching checksums, and
 235  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 236  *      a duplicate block is found.
 237  * The output of this function then goes to the output fd requested
 238  * by the caller of zfs_send().
 239  */
 240 static void *
 241 cksummer(void *arg)
 242 {
 243         dedup_arg_t *dda = arg;
 244         char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE);
 245         dmu_replay_record_t thedrr;
 246         dmu_replay_record_t *drr = &thedrr;
 247         FILE *ofp;
 248         int outfd;
 249         dedup_table_t ddt;
 250         zio_cksum_t stream_cksum;
 251         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 252         uint64_t numbuckets;
 253
 254         ddt.max_ddt_size =
 255             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100,
 256             SMALLEST_POSSIBLE_MAX_DDT_MB << 20);
 257
 258         numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t));
 259
 260         /*
 261          * numbuckets must be a power of 2.  Increase number to
 262          * a power of 2 if necessary.
 263          */
 264         if (!ISP2(numbuckets))
 265                 numbuckets = 1 << high_order_bit(numbuckets);
 266
 267         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 268         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 269             NULL, NULL, NULL, NULL, NULL, 0);
 270         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 271         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 272         ddt.ddt_full = B_FALSE;
 273
 274         outfd = dda->outputfd;
 275         ofp = fdopen(dda->inputfd, "r");
 276         while (ssread(drr, sizeof (*drr), ofp) != 0) {
 277
 278                 /*
 279                  * kernel filled in checksum, we are going to write same
 280                  * record, but need to regenerate checksum.
 281                  */
 282                 if (drr->drr_type != DRR_BEGIN) {
 283                         bzero(&drr->drr_u.drr_checksum.drr_checksum,
 284                             sizeof (drr->drr_u.drr_checksum.drr_checksum));
 285                 }
 286
 287                 switch (drr->drr_type) {
 288                 case DRR_BEGIN:
 289                 {
 290                         struct drr_begin *drrb = &drr->drr_u.drr_begin;
 291                         int fflags;
 292                         int sz = 0;
 293                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 294
 295                         ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
 296
 297                         /* set the DEDUP feature flag for this stream */
 298                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 299                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 300                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 301                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 302
 303                         if (drr->drr_payloadlen != 0) {
 304                                 sz = drr->drr_payloadlen;
 305
 306                                 if (sz > SPA_MAXBLOCKSIZE) {
 307                                         buf = zfs_realloc(dda->dedup_hdl, buf,
 308                                             SPA_MAXBLOCKSIZE, sz);
 309                                 }
 310                                 (void) ssread(buf, sz, ofp);
 311                                 if (ferror(stdin))
 312                                         perror("fread");
 313                         }
 314                         if (dump_record(drr, buf, sz, &stream_cksum,
 315                             outfd) != 0)
 316                                 goto out;
 317                         break;
 318                 }
 319
 320                 case DRR_END:
 321                 {
 322                         struct drr_end *drre = &drr->drr_u.drr_end;
 323                         /* use the recalculated checksum */
 324                         drre->drr_checksum = stream_cksum;
 325                         if (dump_record(drr, NULL, 0, &stream_cksum,
 326                             outfd) != 0)
 327                                 goto out;
 328                         break;
 329                 }
 330
 331                 case DRR_OBJECT:
 332                 {
 333                         struct drr_object *drro = &drr->drr_u.drr_object;
 334                         if (drro->drr_bonuslen > 0) {
 335                                 (void) ssread(buf,
 336                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 337                                     ofp);
 338                         }
 339                         if (dump_record(drr, buf,
 340                             P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 341                             &stream_cksum, outfd) != 0)
 342                                 goto out;
 343                         break;
 344                 }
 345
 346                 case DRR_SPILL:
 347                 {
 348                         struct drr_spill *drrs = &drr->drr_u.drr_spill;
 349                         (void) ssread(buf, drrs->drr_length, ofp);
 350                         if (dump_record(drr, buf, drrs->drr_length,
 351                             &stream_cksum, outfd) != 0)
 352                                 goto out;
 353                         break;
 354                 }
 355
 356                 case DRR_FREEOBJECTS:
 357                 {
 358                         if (dump_record(drr, NULL, 0, &stream_cksum,
 359                             outfd) != 0)
 360                                 goto out;
 361                         break;
 362                 }
 363
 364                 case DRR_WRITE:
 365                 {
 366                         struct drr_write *drrw = &drr->drr_u.drr_write;
 367                         dataref_t       dataref;
 368                         uint64_t        payload_size;
 369
 370                         payload_size = DRR_WRITE_PAYLOAD_SIZE(drrw);
 371                         (void) ssread(buf, payload_size, ofp);
 372
 373                         /*
 374                          * Use the existing checksum if it's dedup-capable,
 375                          * else calculate a SHA256 checksum for it.
 376                          */
 377
 378                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 379                             zero_cksum) ||
 380                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 381                                 SHA256_CTX      ctx;
 382                                 zio_cksum_t     tmpsha256;
 383
 384                                 SHA256Init(&ctx);
 385                                 SHA256Update(&ctx, buf, payload_size);
 386                                 SHA256Final(&tmpsha256, &ctx);
 387                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 388                                     BE_64(tmpsha256.zc_word[0]);
 389                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 390                                     BE_64(tmpsha256.zc_word[1]);
 391                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 392                                     BE_64(tmpsha256.zc_word[2]);
 393                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 394                                     BE_64(tmpsha256.zc_word[3]);
 395                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 396                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 397                         }
 398
 399                         dataref.ref_guid = drrw->drr_toguid;
 400                         dataref.ref_object = drrw->drr_object;
 401                         dataref.ref_offset = drrw->drr_offset;
 402
 403                         if (ddt_update(dda->dedup_hdl, &ddt,
 404                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 405                             &dataref)) {
 406                                 dmu_replay_record_t wbr_drr = {0};
 407                                 struct drr_write_byref *wbr_drrr =
 408                                     &wbr_drr.drr_u.drr_write_byref;
 409
 410                                 /* block already present in stream */
 411                                 wbr_drr.drr_type = DRR_WRITE_BYREF;
 412
 413                                 wbr_drrr->drr_object = drrw->drr_object;
 414                                 wbr_drrr->drr_offset = drrw->drr_offset;
 415                                 wbr_drrr->drr_length = drrw->drr_logical_size;
 416                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 417                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 418                                 wbr_drrr->drr_refobject =
 419                                     dataref.ref_object;
 420                                 wbr_drrr->drr_refoffset =
 421                                     dataref.ref_offset;
 422
 423                                 wbr_drrr->drr_checksumtype =
 424                                     drrw->drr_checksumtype;
 425                                 wbr_drrr->drr_checksumflags =
 426                                     drrw->drr_checksumtype;
 427                                 wbr_drrr->drr_key.ddk_cksum =
 428                                     drrw->drr_key.ddk_cksum;
 429                                 wbr_drrr->drr_key.ddk_prop =
 430                                     drrw->drr_key.ddk_prop;
 431
 432                                 if (dump_record(&wbr_drr, NULL, 0,
 433                                     &stream_cksum, outfd) != 0)
 434                                         goto out;
 435                         } else {
 436                                 /* block not previously seen */
 437                                 if (dump_record(drr, buf, payload_size,
 438                                     &stream_cksum, outfd) != 0)
 439                                         goto out;
 440                         }
 441                         break;
 442                 }
 443
 444                 case DRR_WRITE_EMBEDDED:
 445                 {
 446                         struct drr_write_embedded *drrwe =
 447                             &drr->drr_u.drr_write_embedded;
 448                         (void) ssread(buf,
 449                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp);
 450                         if (dump_record(drr, buf,
 451                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8),
 452                             &stream_cksum, outfd) != 0)
 453                                 goto out;
 454                         break;
 455                 }
 456
 457                 case DRR_FREE:
 458                 {
 459                         if (dump_record(drr, NULL, 0, &stream_cksum,
 460                             outfd) != 0)
 461                                 goto out;
 462                         break;
 463                 }
 464
 465                 default:
 466                         (void) fprintf(stderr, "INVALID record type 0x%x\n",
 467                             drr->drr_type);
 468                         /* should never happen, so assert */
 469                         assert(B_FALSE);
 470                 }
 471         }
 472 out:
 473         umem_cache_destroy(ddt.ddecache);
 474         free(ddt.dedup_hash_array);
 475         free(buf);
 476         (void) fclose(ofp);
 477
 478         return (NULL);
 479 }
 480
 481 /*
 482  * Routines for dealing with the AVL tree of fs-nvlists
 483  */
 484 typedef struct fsavl_node {
 485         avl_node_t fn_node;
 486         nvlist_t *fn_nvfs;
 487         char *fn_snapname;
 488         uint64_t fn_guid;
 489 } fsavl_node_t;
 490
 491 static int
 492 fsavl_compare(const void *arg1, const void *arg2)
 493 {
 494         const fsavl_node_t *fn1 = (const fsavl_node_t *)arg1;
 495         const fsavl_node_t *fn2 = (const fsavl_node_t *)arg2;
 496
 497         return (AVL_CMP(fn1->fn_guid, fn2->fn_guid));
 498 }
 499
 500 /*
 501  * Given the GUID of a snapshot, find its containing filesystem and
 502  * (optionally) name.
 503  */
 504 static nvlist_t *
 505 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 506 {
 507         fsavl_node_t fn_find;
 508         fsavl_node_t *fn;
 509
 510         fn_find.fn_guid = snapguid;
 511
 512         fn = avl_find(avl, &fn_find, NULL);
 513         if (fn) {
 514                 if (snapname)
 515                         *snapname = fn->fn_snapname;
 516                 return (fn->fn_nvfs);
 517         }
 518         return (NULL);
 519 }
 520
 521 static void
 522 fsavl_destroy(avl_tree_t *avl)
 523 {
 524         fsavl_node_t *fn;
 525         void *cookie;
 526
 527         if (avl == NULL)
 528                 return;
 529
 530         cookie = NULL;
 531         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 532                 free(fn);
 533         avl_destroy(avl);
 534         free(avl);
 535 }
 536
 537 /*
 538  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 539  */
 540 static avl_tree_t *
 541 fsavl_create(nvlist_t *fss)
 542 {
 543         avl_tree_t *fsavl;
 544         nvpair_t *fselem = NULL;
 545
 546         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 547                 return (NULL);
 548
 549         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 550             offsetof(fsavl_node_t, fn_node));
 551
 552         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 553                 nvlist_t *nvfs, *snaps;
 554                 nvpair_t *snapelem = NULL;
 555
 556                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 557                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 558
 559                 while ((snapelem =
 560                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 561                         fsavl_node_t *fn;
 562                         uint64_t guid;
 563
 564                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 565                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 566                                 fsavl_destroy(fsavl);
 567                                 return (NULL);
 568                         }
 569                         fn->fn_nvfs = nvfs;
 570                         fn->fn_snapname = nvpair_name(snapelem);
 571                         fn->fn_guid = guid;
 572
 573                         /*
 574                          * Note: if there are multiple snaps with the
 575                          * same GUID, we ignore all but one.
 576                          */
 577                         if (avl_find(fsavl, fn, NULL) == NULL)
 578                                 avl_add(fsavl, fn);
 579                         else
 580                                 free(fn);
 581                 }
 582         }
 583
 584         return (fsavl);
 585 }
 586
 587 /*
 588  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 589  */
 590 typedef struct send_data {
 591         /*
 592          * assigned inside every recursive call,
 593          * restored from *_save on return:
 594          *
 595          * guid of fromsnap snapshot in parent dataset
 596          * txg of fromsnap snapshot in current dataset
 597          * txg of tosnap snapshot in current dataset
 598          */
 599
 600         uint64_t parent_fromsnap_guid;
 601         uint64_t fromsnap_txg;
 602         uint64_t tosnap_txg;
 603
 604         /* the nvlists get accumulated during depth-first traversal */
 605         nvlist_t *parent_snaps;
 606         nvlist_t *fss;
 607         nvlist_t *snapprops;
 608
 609         /* send-receive configuration, does not change during traversal */
 610         const char *fsname;
 611         const char *fromsnap;
 612         const char *tosnap;
 613         boolean_t recursive;
 614         boolean_t verbose;
 615
 616         /*
 617          * The header nvlist is of the following format:
 618          * {
 619          *   "tosnap" -> string
 620          *   "fromsnap" -> string (if incremental)
 621          *   "fss" -> {
 622          *      id -> {
 623          *
 624          *       "name" -> string (full name; for debugging)
 625          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 626          *
 627          *       "props" -> { name -> value (only if set here) }
 628          *       "snaps" -> { name (lastname) -> number (guid) }
 629          *       "snapprops" -> { name (lastname) -> { name -> value } }
 630          *
 631          *       "origin" -> number (guid) (if clone)
 632          *       "sent" -> boolean (not on-disk)
 633          *      }
 634          *   }
 635          * }
 636          *
 637          */
 638 } send_data_t;
 639
 640 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 641
 642 static int
 643 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 644 {
 645         send_data_t *sd = arg;
 646         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 647         uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
 648         char *snapname;
 649         nvlist_t *nv;
 650
 651         snapname = strrchr(zhp->zfs_name, '@')+1;
 652
 653         if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
 654                 if (sd->verbose) {
 655                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
 656                             "skipping snapshot %s because it was created "
 657                             "after the destination snapshot (%s)\n"),
 658                             zhp->zfs_name, sd->tosnap);
 659                 }
 660                 zfs_close(zhp);
 661                 return (0);
 662         }
 663
 664         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 665         /*
 666          * NB: if there is no fromsnap here (it's a newly created fs in
 667          * an incremental replication), we will substitute the tosnap.
 668          */
 669         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 670             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 671             strcmp(snapname, sd->tosnap) == 0)) {
 672                 sd->parent_fromsnap_guid = guid;
 673         }
 674
 675         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 676         send_iterate_prop(zhp, nv);
 677         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 678         nvlist_free(nv);
 679
 680         zfs_close(zhp);
 681         return (0);
 682 }
 683
 684 static void
 685 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 686 {
 687         nvpair_t *elem = NULL;
 688
 689         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 690                 char *propname = nvpair_name(elem);
 691                 zfs_prop_t prop = zfs_name_to_prop(propname);
 692                 nvlist_t *propnv;
 693
 694                 if (!zfs_prop_user(propname)) {
 695                         /*
 696                          * Realistically, this should never happen.  However,
 697                          * we want the ability to add DSL properties without
 698                          * needing to make incompatible version changes.  We
 699                          * need to ignore unknown properties to allow older
 700                          * software to still send datasets containing these
 701                          * properties, with the unknown properties elided.
 702                          */
 703                         if (prop == ZPROP_INVAL)
 704                                 continue;
 705
 706                         if (zfs_prop_readonly(prop))
 707                                 continue;
 708                 }
 709
 710                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 711                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 712                     prop == ZFS_PROP_REFQUOTA ||
 713                     prop == ZFS_PROP_REFRESERVATION) {
 714                         char *source;
 715                         uint64_t value;
 716                         verify(nvlist_lookup_uint64(propnv,
 717                             ZPROP_VALUE, &value) == 0);
 718                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 719                                 continue;
 720                         /*
 721                          * May have no source before SPA_VERSION_RECVD_PROPS,
 722                          * but is still modifiable.
 723                          */
 724                         if (nvlist_lookup_string(propnv,
 725                             ZPROP_SOURCE, &source) == 0) {
 726                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 727                                     (strcmp(source,
 728                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 729                                         continue;
 730                         }
 731                 } else {
 732                         char *source;
 733                         if (nvlist_lookup_string(propnv,
 734                             ZPROP_SOURCE, &source) != 0)
 735                                 continue;
 736                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 737                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 738                                 continue;
 739                 }
 740
 741                 if (zfs_prop_user(propname) ||
 742                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 743                         char *value;
 744                         verify(nvlist_lookup_string(propnv,
 745                             ZPROP_VALUE, &value) == 0);
 746                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 747                 } else {
 748                         uint64_t value;
 749                         verify(nvlist_lookup_uint64(propnv,
 750                             ZPROP_VALUE, &value) == 0);
 751                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 752                 }
 753         }
 754 }
 755
 756 /*
 757  * returns snapshot creation txg
 758  * and returns 0 if the snapshot does not exist
 759  */
 760 static uint64_t
 761 get_snap_txg(libzfs_handle_t *hdl, const char *fs, const char *snap)
 762 {
 763         char name[ZFS_MAX_DATASET_NAME_LEN];
 764         uint64_t txg = 0;
 765
 766         if (fs == NULL || fs[0] == '\0' || snap == NULL || snap[0] == '\0')
 767                 return (txg);
 768
 769         (void) snprintf(name, sizeof (name), "%s@%s", fs, snap);
 770         if (zfs_dataset_exists(hdl, name, ZFS_TYPE_SNAPSHOT)) {
 771                 zfs_handle_t *zhp = zfs_open(hdl, name, ZFS_TYPE_SNAPSHOT);
 772                 if (zhp != NULL) {
 773                         txg = zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG);
 774                         zfs_close(zhp);
 775                 }
 776         }
 777
 778         return (txg);
 779 }
 780
 781 /*
 782  * recursively generate nvlists describing datasets.  See comment
 783  * for the data structure send_data_t above for description of contents
 784  * of the nvlist.
 785  */
 786 static int
 787 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 788 {
 789         send_data_t *sd = arg;
 790         nvlist_t *nvfs, *nv;
 791         int rv = 0;
 792         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 793         uint64_t fromsnap_txg_save = sd->fromsnap_txg;
 794         uint64_t tosnap_txg_save = sd->tosnap_txg;
 795         uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
 796         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 797         uint64_t fromsnap_txg, tosnap_txg;
 798         char guidstring[64];
 799
 800         fromsnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->fromsnap);
 801         if (fromsnap_txg != 0)
 802                 sd->fromsnap_txg = fromsnap_txg;
 803
 804         tosnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->tosnap);
 805         if (tosnap_txg != 0)
 806                 sd->tosnap_txg = tosnap_txg;
 807
 808         /*
 809          * on the send side, if the current dataset does not have tosnap,
 810          * perform two additional checks:
 811          *
 812          * - skip sending the current dataset if it was created later than
 813          *   the parent tosnap
 814          * - return error if the current dataset was created earlier than
 815          *   the parent tosnap
 816          */
 817         if (sd->tosnap != NULL && tosnap_txg == 0) {
 818                 if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
 819                         if (sd->verbose) {
 820                                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
 821                                     "skipping dataset %s: snapshot %s does "
 822                                     "not exist\n"), zhp->zfs_name, sd->tosnap);
 823                         }
 824                 } else {
 825                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
 826                             "cannot send %s@%s%s: snapshot %s@%s does not "
 827                             "exist\n"), sd->fsname, sd->tosnap, sd->recursive ?
 828                             dgettext(TEXT_DOMAIN, " recursively") : "",
 829                             zhp->zfs_name, sd->tosnap);
 830                         rv = -1;
 831                 }
 832                 goto out;
 833         }
 834
 835         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 836         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 837         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 838             sd->parent_fromsnap_guid));
 839
 840         if (zhp->zfs_dmustats.dds_origin[0]) {
 841                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 842                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 843                 if (origin == NULL) {
 844                         rv = -1;
 845                         goto out;
 846                 }
 847                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 848                     origin->zfs_dmustats.dds_guid));
 849         }
 850
 851         /* iterate over props */
 852         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 853         send_iterate_prop(zhp, nv);
 854         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 855         nvlist_free(nv);
 856
 857         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 858         sd->parent_fromsnap_guid = 0;
 859         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 860         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 861         (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd);
 862         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 863         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 864         nvlist_free(sd->parent_snaps);
 865         nvlist_free(sd->snapprops);
 866
 867         /* add this fs to nvlist */
 868         (void) snprintf(guidstring, sizeof (guidstring),
 869             "0x%llx", (longlong_t)guid);
 870         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 871         nvlist_free(nvfs);
 872
 873         /* iterate over children */
 874         if (sd->recursive)
 875                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 876
 877 out:
 878         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 879         sd->fromsnap_txg = fromsnap_txg_save;
 880         sd->tosnap_txg = tosnap_txg_save;
 881
 882         zfs_close(zhp);
 883         return (rv);
 884 }
 885
 886 static int
 887 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 888     const char *tosnap, boolean_t recursive, boolean_t verbose,
 889     nvlist_t **nvlp, avl_tree_t **avlp)
 890 {
 891         zfs_handle_t *zhp;
 892         send_data_t sd = { 0 };
 893         int error;
 894
 895         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 896         if (zhp == NULL)
 897                 return (EZFS_BADTYPE);
 898
 899         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 900         sd.fsname = fsname;
 901         sd.fromsnap = fromsnap;
 902         sd.tosnap = tosnap;
 903         sd.recursive = recursive;
 904         sd.verbose = verbose;
 905
 906         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 907                 nvlist_free(sd.fss);
 908                 if (avlp != NULL)
 909                         *avlp = NULL;
 910                 *nvlp = NULL;
 911                 return (error);
 912         }
 913
 914         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 915                 nvlist_free(sd.fss);
 916                 *nvlp = NULL;
 917                 return (EZFS_NOMEM);
 918         }
 919
 920         *nvlp = sd.fss;
 921         return (0);
 922 }
 923
 924 /*
 925  * Routines specific to "zfs send"
 926  */
 927 typedef struct send_dump_data {
 928         /* these are all just the short snapname (the part after the @) */
 929         const char *fromsnap;
 930         const char *tosnap;
 931         char prevsnap[ZFS_MAX_DATASET_NAME_LEN];
 932         uint64_t prevsnap_obj;
 933         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 934         boolean_t verbose, dryrun, parsable, progress, embed_data, std_out;
 935         boolean_t progressastitle;
 936         boolean_t large_block, compress;
 937         int outfd;
 938         boolean_t err;
 939         nvlist_t *fss;
 940         nvlist_t *snapholds;
 941         avl_tree_t *fsavl;
 942         snapfilter_cb_t *filter_cb;
 943         void *filter_cb_arg;
 944         nvlist_t *debugnv;
 945         char holdtag[ZFS_MAX_DATASET_NAME_LEN];
 946         int cleanup_fd;
 947         uint64_t size;
 948 } send_dump_data_t;
 949
 950 static int
 951 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
 952     boolean_t fromorigin, enum lzc_send_flags flags, uint64_t *sizep)
 953 {
 954         zfs_cmd_t zc = { 0 };
 955         libzfs_handle_t *hdl = zhp->zfs_hdl;
 956
 957         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 958         assert(fromsnap_obj == 0 || !fromorigin);
 959
 960         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 961         zc.zc_obj = fromorigin;
 962         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 963         zc.zc_fromobj = fromsnap_obj;
 964         zc.zc_guid = 1;  /* estimate flag */
 965         zc.zc_flags = flags;
 966
 967         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 968                 char errbuf[1024];
 969                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 970                     "warning: cannot estimate space for '%s'"), zhp->zfs_name);
 971
 972                 switch (errno) {
 973                 case EXDEV:
 974                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 975                             "not an earlier snapshot from the same fs"));
 976                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 977
 978                 case ENOENT:
 979                         if (zfs_dataset_exists(hdl, zc.zc_name,
 980                             ZFS_TYPE_SNAPSHOT)) {
 981                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 982                                     "incremental source (@%s) does not exist"),
 983                                     zc.zc_value);
 984                         }
 985                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 986
 987                 case EDQUOT:
 988                 case EFBIG:
 989                 case EIO:
 990                 case ENOLINK:
 991                 case ENOSPC:
 992                 case ENXIO:
 993                 case EPIPE:
 994                 case ERANGE:
 995                 case EFAULT:
 996                 case EROFS:
 997                         zfs_error_aux(hdl, strerror(errno));
 998                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 999
1000                 default:
1001                         return (zfs_standard_error(hdl, errno, errbuf));
1002                 }
1003         }
1004
1005         *sizep = zc.zc_objset_type;
1006
1007         return (0);
1008 }
1009
1010 /*
1011  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
1012  * NULL) to the file descriptor specified by outfd.
1013  */
1014 static int
1015 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
1016     boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
1017     nvlist_t *debugnv)
1018 {
1019         zfs_cmd_t zc = { 0 };
1020         libzfs_handle_t *hdl = zhp->zfs_hdl;
1021         nvlist_t *thisdbg;
1022
1023         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
1024         assert(fromsnap_obj == 0 || !fromorigin);
1025
1026         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1027         zc.zc_cookie = outfd;
1028         zc.zc_obj = fromorigin;
1029         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1030         zc.zc_fromobj = fromsnap_obj;
1031         zc.zc_flags = flags;
1032
1033         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
1034         if (fromsnap && fromsnap[0] != '\0') {
1035                 VERIFY(0 == nvlist_add_string(thisdbg,
1036                     "fromsnap", fromsnap));
1037         }
1038
1039         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
1040                 char errbuf[1024];
1041                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1042                     "warning: cannot send '%s'"), zhp->zfs_name);
1043
1044                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
1045                 if (debugnv) {
1046                         VERIFY(0 == nvlist_add_nvlist(debugnv,
1047                             zhp->zfs_name, thisdbg));
1048                 }
1049                 nvlist_free(thisdbg);
1050
1051                 switch (errno) {
1052                 case EXDEV:
1053                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1054                             "not an earlier snapshot from the same fs"));
1055                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
1056
1057                 case ENOENT:
1058                         if (zfs_dataset_exists(hdl, zc.zc_name,
1059                             ZFS_TYPE_SNAPSHOT)) {
1060                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1061                                     "incremental source (@%s) does not exist"),
1062                                     zc.zc_value);
1063                         }
1064                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
1065
1066                 case EDQUOT:
1067                 case EFBIG:
1068                 case EIO:
1069                 case ENOLINK:
1070                 case ENOSPC:
1071 #ifdef illumos
1072                 case ENOSTR:
1073 #endif
1074                 case ENXIO:
1075                 case EPIPE:
1076                 case ERANGE:
1077                 case EFAULT:
1078                 case EROFS:
1079                         zfs_error_aux(hdl, strerror(errno));
1080                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1081
1082                 default:
1083                         return (zfs_standard_error(hdl, errno, errbuf));
1084                 }
1085         }
1086
1087         if (debugnv)
1088                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
1089         nvlist_free(thisdbg);
1090
1091         return (0);
1092 }
1093
1094 static void
1095 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
1096 {
1097         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
1098
1099         /*
1100          * zfs_send() only sets snapholds for sends that need them,
1101          * e.g. replication and doall.
1102          */
1103         if (sdd->snapholds == NULL)
1104                 return;
1105
1106         fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
1107 }
1108
1109 static void *
1110 send_progress_thread(void *arg)
1111 {
1112         progress_arg_t *pa = arg;
1113         zfs_cmd_t zc = { 0 };
1114         zfs_handle_t *zhp = pa->pa_zhp;
1115         libzfs_handle_t *hdl = zhp->zfs_hdl;
1116         unsigned long long bytes, total;
1117         char buf[16];
1118         time_t t;
1119         struct tm *tm;
1120
1121         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1122
1123         if (!pa->pa_parsable && !pa->pa_astitle)
1124                 (void) fprintf(stderr, "TIME        SENT   SNAPSHOT\n");
1125
1126         /*
1127          * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1128          */
1129         for (;;) {
1130                 (void) sleep(1);
1131
1132                 zc.zc_cookie = pa->pa_fd;
1133                 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1134                         return ((void *)-1);
1135
1136                 (void) time(&t);
1137                 tm = localtime(&t);
1138                 bytes = zc.zc_cookie;
1139
1140                 if (pa->pa_astitle) {
1141                         int pct;
1142                         if (pa->pa_size > bytes)
1143                                 pct = 100 * bytes / pa->pa_size;
1144                         else
1145                                 pct = 100;
1146
1147                         setproctitle("sending %s (%d%%: %llu/%llu)",
1148                             zhp->zfs_name, pct, bytes, pa->pa_size);
1149                 } else if (pa->pa_parsable) {
1150                         (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1151                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1152                             bytes, zhp->zfs_name);
1153                 } else {
1154                         zfs_nicenum(bytes, buf, sizeof (buf));
1155                         (void) fprintf(stderr, "%02d:%02d:%02d   %5s   %s\n",
1156                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1157                             buf, zhp->zfs_name);
1158                 }
1159         }
1160 }
1161
1162 static void
1163 send_print_verbose(FILE *fout, const char *tosnap, const char *fromsnap,
1164     uint64_t size, boolean_t parsable)
1165 {
1166         if (parsable) {
1167                 if (fromsnap != NULL) {
1168                         (void) fprintf(fout, "incremental\t%s\t%s",
1169                             fromsnap, tosnap);
1170                 } else {
1171                         (void) fprintf(fout, "full\t%s",
1172                             tosnap);
1173                 }
1174         } else {
1175                 if (fromsnap != NULL) {
1176                         if (strchr(fromsnap, '@') == NULL &&
1177                             strchr(fromsnap, '#') == NULL) {
1178                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1179                                     "send from @%s to %s"),
1180                                     fromsnap, tosnap);
1181                         } else {
1182                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1183                                     "send from %s to %s"),
1184                                     fromsnap, tosnap);
1185                         }
1186                 } else {
1187                         (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1188                             "full send of %s"),
1189                             tosnap);
1190                 }
1191         }
1192
1193         if (parsable) {
1194                 (void) fprintf(fout, "\t%llu",
1195                     (longlong_t)size);
1196         } else if (size != 0) {
1197                 char buf[16];
1198                 zfs_nicenum(size, buf, sizeof (buf));
1199                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1200                     " estimated size is %s"), buf);
1201         }
1202         (void) fprintf(fout, "\n");
1203 }
1204
1205 static int
1206 dump_snapshot(zfs_handle_t *zhp, void *arg)
1207 {
1208         send_dump_data_t *sdd = arg;
1209         progress_arg_t pa = { 0 };
1210         pthread_t tid;
1211         char *thissnap;
1212         enum lzc_send_flags flags = 0;
1213         int err;
1214         boolean_t isfromsnap, istosnap, fromorigin;
1215         boolean_t exclude = B_FALSE;
1216         FILE *fout = sdd->std_out ? stdout : stderr;
1217         uint64_t size = 0;
1218
1219         err = 0;
1220         thissnap = strchr(zhp->zfs_name, '@') + 1;
1221         isfromsnap = (sdd->fromsnap != NULL &&
1222             strcmp(sdd->fromsnap, thissnap) == 0);
1223
1224         if (!sdd->seenfrom && isfromsnap) {
1225                 gather_holds(zhp, sdd);
1226                 sdd->seenfrom = B_TRUE;
1227                 (void) strcpy(sdd->prevsnap, thissnap);
1228                 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1229                 zfs_close(zhp);
1230                 return (0);
1231         }
1232
1233         if (sdd->seento || !sdd->seenfrom) {
1234                 zfs_close(zhp);
1235                 return (0);
1236         }
1237
1238         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1239         if (istosnap)
1240                 sdd->seento = B_TRUE;
1241
1242         if (sdd->large_block)
1243                 flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1244         if (sdd->embed_data)
1245                 flags |= LZC_SEND_FLAG_EMBED_DATA;
1246         if (sdd->compress)
1247                 flags |= LZC_SEND_FLAG_COMPRESS;
1248
1249         if (!sdd->doall && !isfromsnap && !istosnap) {
1250                 if (sdd->replicate) {
1251                         char *snapname;
1252                         nvlist_t *snapprops;
1253                         /*
1254                          * Filter out all intermediate snapshots except origin
1255                          * snapshots needed to replicate clones.
1256                          */
1257                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1258                             zhp->zfs_dmustats.dds_guid, &snapname);
1259
1260                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1261                             "snapprops", &snapprops));
1262                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1263                             thissnap, &snapprops));
1264                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
1265                 } else {
1266                         exclude = B_TRUE;
1267                 }
1268         }
1269
1270         /*
1271          * If a filter function exists, call it to determine whether
1272          * this snapshot will be sent.
1273          */
1274         if (exclude || (sdd->filter_cb != NULL &&
1275             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1276                 /*
1277                  * This snapshot is filtered out.  Don't send it, and don't
1278                  * set prevsnap_obj, so it will be as if this snapshot didn't
1279                  * exist, and the next accepted snapshot will be sent as
1280                  * an incremental from the last accepted one, or as the
1281                  * first (and full) snapshot in the case of a replication,
1282                  * non-incremental send.
1283                  */
1284                 zfs_close(zhp);
1285                 return (0);
1286         }
1287
1288         gather_holds(zhp, sdd);
1289         fromorigin = sdd->prevsnap[0] == '\0' &&
1290             (sdd->fromorigin || sdd->replicate);
1291
1292         if (sdd->verbose || sdd->progress) {
1293                 (void) estimate_ioctl(zhp, sdd->prevsnap_obj,
1294                     fromorigin, flags, &size);
1295                 sdd->size += size;
1296
1297                 send_print_verbose(fout, zhp->zfs_name,
1298                     sdd->prevsnap[0] ? sdd->prevsnap : NULL,
1299                     size, sdd->parsable);
1300         }
1301
1302         if (!sdd->dryrun) {
1303                 /*
1304                  * If progress reporting is requested, spawn a new thread to
1305                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1306                  */
1307                 if (sdd->progress) {
1308                         pa.pa_zhp = zhp;
1309                         pa.pa_fd = sdd->outfd;
1310                         pa.pa_parsable = sdd->parsable;
1311                         pa.pa_size = sdd->size;
1312                         pa.pa_astitle = sdd->progressastitle;
1313
1314                         if ((err = pthread_create(&tid, NULL,
1315                             send_progress_thread, &pa)) != 0) {
1316                                 zfs_close(zhp);
1317                                 return (err);
1318                         }
1319                 }
1320
1321                 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1322                     fromorigin, sdd->outfd, flags, sdd->debugnv);
1323
1324                 if (sdd->progress) {
1325                         (void) pthread_cancel(tid);
1326                         (void) pthread_join(tid, NULL);
1327                 }
1328         }
1329
1330         (void) strcpy(sdd->prevsnap, thissnap);
1331         sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1332         zfs_close(zhp);
1333         return (err);
1334 }
1335
1336 static int
1337 dump_filesystem(zfs_handle_t *zhp, void *arg)
1338 {
1339         int rv = 0;
1340         send_dump_data_t *sdd = arg;
1341         boolean_t missingfrom = B_FALSE;
1342         zfs_cmd_t zc = { 0 };
1343
1344         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1345             zhp->zfs_name, sdd->tosnap);
1346         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1347                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1348                     "WARNING: could not send %s@%s: does not exist\n"),
1349                     zhp->zfs_name, sdd->tosnap);
1350                 sdd->err = B_TRUE;
1351                 return (0);
1352         }
1353
1354         if (sdd->replicate && sdd->fromsnap) {
1355                 /*
1356                  * If this fs does not have fromsnap, and we're doing
1357                  * recursive, we need to send a full stream from the
1358                  * beginning (or an incremental from the origin if this
1359                  * is a clone).  If we're doing non-recursive, then let
1360                  * them get the error.
1361                  */
1362                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1363                     zhp->zfs_name, sdd->fromsnap);
1364                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1365                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1366                         missingfrom = B_TRUE;
1367                 }
1368         }
1369
1370         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1371         sdd->prevsnap_obj = 0;
1372         if (sdd->fromsnap == NULL || missingfrom)
1373                 sdd->seenfrom = B_TRUE;
1374
1375         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1376         if (!sdd->seenfrom) {
1377                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1378                     "WARNING: could not send %s@%s:\n"
1379                     "incremental source (%s@%s) does not exist\n"),
1380                     zhp->zfs_name, sdd->tosnap,
1381                     zhp->zfs_name, sdd->fromsnap);
1382                 sdd->err = B_TRUE;
1383         } else if (!sdd->seento) {
1384                 if (sdd->fromsnap) {
1385                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1386                             "WARNING: could not send %s@%s:\n"
1387                             "incremental source (%s@%s) "
1388                             "is not earlier than it\n"),
1389                             zhp->zfs_name, sdd->tosnap,
1390                             zhp->zfs_name, sdd->fromsnap);
1391                 } else {
1392                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1393                             "WARNING: "
1394                             "could not send %s@%s: does not exist\n"),
1395                             zhp->zfs_name, sdd->tosnap);
1396                 }
1397                 sdd->err = B_TRUE;
1398         }
1399
1400         return (rv);
1401 }
1402
1403 static int
1404 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1405 {
1406         send_dump_data_t *sdd = arg;
1407         nvpair_t *fspair;
1408         boolean_t needagain, progress;
1409
1410         if (!sdd->replicate)
1411                 return (dump_filesystem(rzhp, sdd));
1412
1413         /* Mark the clone origin snapshots. */
1414         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1415             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1416                 nvlist_t *nvfs;
1417                 uint64_t origin_guid = 0;
1418
1419                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1420                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1421                 if (origin_guid != 0) {
1422                         char *snapname;
1423                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1424                             origin_guid, &snapname);
1425                         if (origin_nv != NULL) {
1426                                 nvlist_t *snapprops;
1427                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1428                                     "snapprops", &snapprops));
1429                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1430                                     snapname, &snapprops));
1431                                 VERIFY(0 == nvlist_add_boolean(
1432                                     snapprops, "is_clone_origin"));
1433                         }
1434                 }
1435         }
1436 again:
1437         needagain = progress = B_FALSE;
1438         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1439             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1440                 nvlist_t *fslist, *parent_nv;
1441                 char *fsname;
1442                 zfs_handle_t *zhp;
1443                 int err;
1444                 uint64_t origin_guid = 0;
1445                 uint64_t parent_guid = 0;
1446
1447                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1448                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1449                         continue;
1450
1451                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1452                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1453                 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1454                     &parent_guid);
1455
1456                 if (parent_guid != 0) {
1457                         parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1458                         if (!nvlist_exists(parent_nv, "sent")) {
1459                                 /* parent has not been sent; skip this one */
1460                                 needagain = B_TRUE;
1461                                 continue;
1462                         }
1463                 }
1464
1465                 if (origin_guid != 0) {
1466                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1467                             origin_guid, NULL);
1468                         if (origin_nv != NULL &&
1469                             !nvlist_exists(origin_nv, "sent")) {
1470                                 /*
1471                                  * origin has not been sent yet;
1472                                  * skip this clone.
1473                                  */
1474                                 needagain = B_TRUE;
1475                                 continue;
1476                         }
1477                 }
1478
1479                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1480                 if (zhp == NULL)
1481                         return (-1);
1482                 err = dump_filesystem(zhp, sdd);
1483                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1484                 progress = B_TRUE;
1485                 zfs_close(zhp);
1486                 if (err)
1487                         return (err);
1488         }
1489         if (needagain) {
1490                 assert(progress);
1491                 goto again;
1492         }
1493
1494         /* clean out the sent flags in case we reuse this fss */
1495         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1496             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1497                 nvlist_t *fslist;
1498
1499                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1500                 (void) nvlist_remove_all(fslist, "sent");
1501         }
1502
1503         return (0);
1504 }
1505
1506 nvlist_t *
1507 zfs_send_resume_token_to_nvlist(libzfs_handle_t *hdl, const char *token)
1508 {
1509         unsigned int version;
1510         int nread;
1511         unsigned long long checksum, packed_len;
1512
1513         /*
1514          * Decode token header, which is:
1515          *   <token version>-<checksum of payload>-<uncompressed payload length>
1516          * Note that the only supported token version is 1.
1517          */
1518         nread = sscanf(token, "%u-%llx-%llx-",
1519             &version, &checksum, &packed_len);
1520         if (nread != 3) {
1521                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1522                     "resume token is corrupt (invalid format)"));
1523                 return (NULL);
1524         }
1525
1526         if (version != ZFS_SEND_RESUME_TOKEN_VERSION) {
1527                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1528                     "resume token is corrupt (invalid version %u)"),
1529                     version);
1530                 return (NULL);
1531         }
1532
1533         /* convert hexadecimal representation to binary */
1534         token = strrchr(token, '-') + 1;
1535         int len = strlen(token) / 2;
1536         unsigned char *compressed = zfs_alloc(hdl, len);
1537         for (int i = 0; i < len; i++) {
1538                 nread = sscanf(token + i * 2, "%2hhx", compressed + i);
1539                 if (nread != 1) {
1540                         free(compressed);
1541                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1542                             "resume token is corrupt "
1543                             "(payload is not hex-encoded)"));
1544                         return (NULL);
1545                 }
1546         }
1547
1548         /* verify checksum */
1549         zio_cksum_t cksum;
1550         fletcher_4_native(compressed, len, NULL, &cksum);
1551         if (cksum.zc_word[0] != checksum) {
1552                 free(compressed);
1553                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1554                     "resume token is corrupt (incorrect checksum)"));
1555                 return (NULL);
1556         }
1557
1558         /* uncompress */
1559         void *packed = zfs_alloc(hdl, packed_len);
1560         uLongf packed_len_long = packed_len;
1561         if (uncompress(packed, &packed_len_long, compressed, len) != Z_OK ||
1562             packed_len_long != packed_len) {
1563                 free(packed);
1564                 free(compressed);
1565                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1566                     "resume token is corrupt (decompression failed)"));
1567                 return (NULL);
1568         }
1569
1570         /* unpack nvlist */
1571         nvlist_t *nv;
1572         int error = nvlist_unpack(packed, packed_len, &nv, KM_SLEEP);
1573         free(packed);
1574         free(compressed);
1575         if (error != 0) {
1576                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1577                     "resume token is corrupt (nvlist_unpack failed)"));
1578                 return (NULL);
1579         }
1580         return (nv);
1581 }
1582
1583 int
1584 zfs_send_resume(libzfs_handle_t *hdl, sendflags_t *flags, int outfd,
1585     const char *resume_token)
1586 {
1587         char errbuf[1024];
1588         char *toname;
1589         char *fromname = NULL;
1590         uint64_t resumeobj, resumeoff, toguid, fromguid, bytes;
1591         zfs_handle_t *zhp;
1592         int error = 0;
1593         char name[ZFS_MAX_DATASET_NAME_LEN];
1594         enum lzc_send_flags lzc_flags = 0;
1595         uint64_t size = 0;
1596         FILE *fout = (flags->verbose && flags->dryrun) ? stdout : stderr;
1597
1598         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1599             "cannot resume send"));
1600
1601         nvlist_t *resume_nvl =
1602             zfs_send_resume_token_to_nvlist(hdl, resume_token);
1603         if (resume_nvl == NULL) {
1604                 /*
1605                  * zfs_error_aux has already been set by
1606                  * zfs_send_resume_token_to_nvlist
1607                  */
1608                 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1609         }
1610         if (flags->verbose) {
1611                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1612                     "resume token contents:\n"));
1613                 nvlist_print(fout, resume_nvl);
1614         }
1615
1616         if (nvlist_lookup_string(resume_nvl, "toname", &toname) != 0 ||
1617             nvlist_lookup_uint64(resume_nvl, "object", &resumeobj) != 0 ||
1618             nvlist_lookup_uint64(resume_nvl, "offset", &resumeoff) != 0 ||
1619             nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 ||
1620             nvlist_lookup_uint64(resume_nvl, "toguid", &toguid) != 0) {
1621                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1622                     "resume token is corrupt"));
1623                 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1624         }
1625         fromguid = 0;
1626         (void) nvlist_lookup_uint64(resume_nvl, "fromguid", &fromguid);
1627
1628         if (flags->largeblock || nvlist_exists(resume_nvl, "largeblockok"))
1629                 lzc_flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1630         if (flags->embed_data || nvlist_exists(resume_nvl, "embedok"))
1631                 lzc_flags |= LZC_SEND_FLAG_EMBED_DATA;
1632         if (flags->compress || nvlist_exists(resume_nvl, "compressok"))
1633                 lzc_flags |= LZC_SEND_FLAG_COMPRESS;
1634
1635         if (guid_to_name(hdl, toname, toguid, B_FALSE, name) != 0) {
1636                 if (zfs_dataset_exists(hdl, toname, ZFS_TYPE_DATASET)) {
1637                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1638                             "'%s' is no longer the same snapshot used in "
1639                             "the initial send"), toname);
1640                 } else {
1641                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1642                             "'%s' used in the initial send no longer exists"),
1643                             toname);
1644                 }
1645                 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1646         }
1647         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1648         if (zhp == NULL) {
1649                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1650                     "unable to access '%s'"), name);
1651                 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1652         }
1653
1654         if (fromguid != 0) {
1655                 if (guid_to_name(hdl, toname, fromguid, B_TRUE, name) != 0) {
1656                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1657                             "incremental source %#llx no longer exists"),
1658                             (longlong_t)fromguid);
1659                         return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1660                 }
1661                 fromname = name;
1662         }
1663
1664         if (flags->progress || flags->verbose) {
1665                 error = lzc_send_space(zhp->zfs_name, fromname,
1666                     lzc_flags, &size);
1667                 if (error == 0)
1668                         size = MAX(0, (int64_t)(size - bytes));
1669         }
1670         if (flags->verbose) {
1671                 send_print_verbose(fout, zhp->zfs_name, fromname,
1672                     size, flags->parsable);
1673         }
1674
1675         if (!flags->dryrun) {
1676                 progress_arg_t pa = { 0 };
1677                 pthread_t tid;
1678                 /*
1679                  * If progress reporting is requested, spawn a new thread to
1680                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1681                  */
1682                 if (flags->progress) {
1683                         pa.pa_zhp = zhp;
1684                         pa.pa_fd = outfd;
1685                         pa.pa_parsable = flags->parsable;
1686                         pa.pa_size = size;
1687                         pa.pa_astitle = flags->progressastitle;
1688
1689                         error = pthread_create(&tid, NULL,
1690                             send_progress_thread, &pa);
1691                         if (error != 0) {
1692                                 zfs_close(zhp);
1693                                 return (error);
1694                         }
1695                 }
1696
1697                 error = lzc_send_resume(zhp->zfs_name, fromname, outfd,
1698                     lzc_flags, resumeobj, resumeoff);
1699
1700                 if (flags->progress) {
1701                         (void) pthread_cancel(tid);
1702                         (void) pthread_join(tid, NULL);
1703                 }
1704
1705                 char errbuf[1024];
1706                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1707                     "warning: cannot send '%s'"), zhp->zfs_name);
1708
1709                 zfs_close(zhp);
1710
1711                 switch (error) {
1712                 case 0:
1713                         return (0);
1714                 case EXDEV:
1715                 case ENOENT:
1716                 case EDQUOT:
1717                 case EFBIG:
1718                 case EIO:
1719                 case ENOLINK:
1720                 case ENOSPC:
1721 #ifdef illumos
1722                 case ENOSTR:
1723 #endif
1724                 case ENXIO:
1725                 case EPIPE:
1726                 case ERANGE:
1727                 case EFAULT:
1728                 case EROFS:
1729                         zfs_error_aux(hdl, strerror(errno));
1730                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1731
1732                 default:
1733                         return (zfs_standard_error(hdl, errno, errbuf));
1734                 }
1735         }
1736
1737
1738         zfs_close(zhp);
1739
1740         return (error);
1741 }
1742
1743 /*
1744  * Generate a send stream for the dataset identified by the argument zhp.
1745  *
1746  * The content of the send stream is the snapshot identified by
1747  * 'tosnap'.  Incremental streams are requested in two ways:
1748  *     - from the snapshot identified by "fromsnap" (if non-null) or
1749  *     - from the origin of the dataset identified by zhp, which must
1750  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1751  *       is TRUE.
1752  *
1753  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1754  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1755  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1756  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1757  * case too. If "props" is set, send properties.
1758  */
1759 int
1760 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1761     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1762     void *cb_arg, nvlist_t **debugnvp)
1763 {
1764         char errbuf[1024];
1765         send_dump_data_t sdd = { 0 };
1766         int err = 0;
1767         nvlist_t *fss = NULL;
1768         avl_tree_t *fsavl = NULL;
1769         static uint64_t holdseq;
1770         int spa_version;
1771         pthread_t tid = 0;
1772         int pipefd[2];
1773         dedup_arg_t dda = { 0 };
1774         int featureflags = 0;
1775         FILE *fout;
1776
1777         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1778             "cannot send '%s'"), zhp->zfs_name);
1779
1780         if (fromsnap && fromsnap[0] == '\0') {
1781                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1782                     "zero-length incremental source"));
1783                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1784         }
1785
1786         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1787                 uint64_t version;
1788                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1789                 if (version >= ZPL_VERSION_SA) {
1790                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1791                 }
1792         }
1793
1794         if (flags->dedup && !flags->dryrun) {
1795                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1796                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1797                 if ((err = pipe(pipefd)) != 0) {
1798                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1799                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1800                             errbuf));
1801                 }
1802                 dda.outputfd = outfd;
1803                 dda.inputfd = pipefd[1];
1804                 dda.dedup_hdl = zhp->zfs_hdl;
1805                 if ((err = pthread_create(&tid, NULL, cksummer, &dda)) != 0) {
1806                         (void) close(pipefd[0]);
1807                         (void) close(pipefd[1]);
1808                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1809                         return (zfs_error(zhp->zfs_hdl,
1810                             EZFS_THREADCREATEFAILED, errbuf));
1811                 }
1812         }
1813
1814         if (flags->replicate || flags->doall || flags->props) {
1815                 dmu_replay_record_t drr = { 0 };
1816                 char *packbuf = NULL;
1817                 size_t buflen = 0;
1818                 zio_cksum_t zc = { 0 };
1819
1820                 if (flags->replicate || flags->props) {
1821                         nvlist_t *hdrnv;
1822
1823                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1824                         if (fromsnap) {
1825                                 VERIFY(0 == nvlist_add_string(hdrnv,
1826                                     "fromsnap", fromsnap));
1827                         }
1828                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1829                         if (!flags->replicate) {
1830                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1831                                     "not_recursive"));
1832                         }
1833
1834                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1835                             fromsnap, tosnap, flags->replicate, flags->verbose,
1836                             &fss, &fsavl);
1837                         if (err)
1838                                 goto err_out;
1839                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1840                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1841                             NV_ENCODE_XDR, 0);
1842                         if (debugnvp)
1843                                 *debugnvp = hdrnv;
1844                         else
1845                                 nvlist_free(hdrnv);
1846                         if (err)
1847                                 goto stderr_out;
1848                 }
1849
1850                 if (!flags->dryrun) {
1851                         /* write first begin record */
1852                         drr.drr_type = DRR_BEGIN;
1853                         drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1854                         DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1855                             drr_versioninfo, DMU_COMPOUNDSTREAM);
1856                         DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1857                             drr_versioninfo, featureflags);
1858                         (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1859                             sizeof (drr.drr_u.drr_begin.drr_toname),
1860                             "%s@%s", zhp->zfs_name, tosnap);
1861                         drr.drr_payloadlen = buflen;
1862
1863                         err = dump_record(&drr, packbuf, buflen, &zc, outfd);
1864                         free(packbuf);
1865                         if (err != 0)
1866                                 goto stderr_out;
1867
1868                         /* write end record */
1869                         bzero(&drr, sizeof (drr));
1870                         drr.drr_type = DRR_END;
1871                         drr.drr_u.drr_end.drr_checksum = zc;
1872                         err = write(outfd, &drr, sizeof (drr));
1873                         if (err == -1) {
1874                                 err = errno;
1875                                 goto stderr_out;
1876                         }
1877
1878                         err = 0;
1879                 }
1880         }
1881
1882         /* dump each stream */
1883         sdd.fromsnap = fromsnap;
1884         sdd.tosnap = tosnap;
1885         if (tid != 0)
1886                 sdd.outfd = pipefd[0];
1887         else
1888                 sdd.outfd = outfd;
1889         sdd.replicate = flags->replicate;
1890         sdd.doall = flags->doall;
1891         sdd.fromorigin = flags->fromorigin;
1892         sdd.fss = fss;
1893         sdd.fsavl = fsavl;
1894         sdd.verbose = flags->verbose;
1895         sdd.parsable = flags->parsable;
1896         sdd.progress = flags->progress;
1897         sdd.progressastitle = flags->progressastitle;
1898         sdd.dryrun = flags->dryrun;
1899         sdd.large_block = flags->largeblock;
1900         sdd.embed_data = flags->embed_data;
1901         sdd.compress = flags->compress;
1902         sdd.filter_cb = filter_func;
1903         sdd.filter_cb_arg = cb_arg;
1904         if (debugnvp)
1905                 sdd.debugnv = *debugnvp;
1906         if (sdd.verbose && sdd.dryrun)
1907                 sdd.std_out = B_TRUE;
1908         fout = sdd.std_out ? stdout : stderr;
1909
1910         /*
1911          * Some flags require that we place user holds on the datasets that are
1912          * being sent so they don't get destroyed during the send. We can skip
1913          * this step if the pool is imported read-only since the datasets cannot
1914          * be destroyed.
1915          */
1916         if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1917             ZPOOL_PROP_READONLY, NULL) &&
1918             zfs_spa_version(zhp, &spa_version) == 0 &&
1919             spa_version >= SPA_VERSION_USERREFS &&
1920             (flags->doall || flags->replicate)) {
1921                 ++holdseq;
1922                 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1923                     ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1924                 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1925                 if (sdd.cleanup_fd < 0) {
1926                         err = errno;
1927                         goto stderr_out;
1928                 }
1929                 sdd.snapholds = fnvlist_alloc();
1930         } else {
1931                 sdd.cleanup_fd = -1;
1932                 sdd.snapholds = NULL;
1933         }
1934         if (flags->progress || flags->verbose || sdd.snapholds != NULL) {
1935                 /*
1936                  * Do a verbose no-op dry run to get all the verbose output
1937                  * or to gather snapshot hold's before generating any data,
1938                  * then do a non-verbose real run to generate the streams.
1939                  */
1940                 sdd.dryrun = B_TRUE;
1941                 err = dump_filesystems(zhp, &sdd);
1942
1943                 if (err != 0)
1944                         goto stderr_out;
1945
1946                 if (flags->verbose) {
1947                         if (flags->parsable) {
1948                                 (void) fprintf(fout, "size\t%llu\n",
1949                                     (longlong_t)sdd.size);
1950                         } else {
1951                                 char buf[16];
1952                                 zfs_nicenum(sdd.size, buf, sizeof (buf));
1953                                 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1954                                     "total estimated size is %s\n"), buf);
1955                         }
1956                 }
1957
1958                 /* Ensure no snaps found is treated as an error. */
1959                 if (!sdd.seento) {
1960                         err = ENOENT;
1961                         goto err_out;
1962                 }
1963
1964                 /* Skip the second run if dryrun was requested. */
1965                 if (flags->dryrun)
1966                         goto err_out;
1967
1968                 if (sdd.snapholds != NULL) {
1969                         err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
1970                         if (err != 0)
1971                                 goto stderr_out;
1972
1973                         fnvlist_free(sdd.snapholds);
1974                         sdd.snapholds = NULL;
1975                 }
1976
1977                 sdd.dryrun = B_FALSE;
1978                 sdd.verbose = B_FALSE;
1979         }
1980
1981         err = dump_filesystems(zhp, &sdd);
1982         fsavl_destroy(fsavl);
1983         nvlist_free(fss);
1984
1985         /* Ensure no snaps found is treated as an error. */
1986         if (err == 0 && !sdd.seento)
1987                 err = ENOENT;
1988
1989         if (tid != 0) {
1990                 if (err != 0)
1991                         (void) pthread_cancel(tid);
1992                 (void) close(pipefd[0]);
1993                 (void) pthread_join(tid, NULL);
1994         }
1995
1996         if (sdd.cleanup_fd != -1) {
1997                 VERIFY(0 == close(sdd.cleanup_fd));
1998                 sdd.cleanup_fd = -1;
1999         }
2000
2001         if (!flags->dryrun && (flags->replicate || flags->doall ||
2002             flags->props)) {
2003                 /*
2004                  * write final end record.  NB: want to do this even if
2005                  * there was some error, because it might not be totally
2006                  * failed.
2007                  */
2008                 dmu_replay_record_t drr = { 0 };
2009                 drr.drr_type = DRR_END;
2010                 if (write(outfd, &drr, sizeof (drr)) == -1) {
2011                         return (zfs_standard_error(zhp->zfs_hdl,
2012                             errno, errbuf));
2013                 }
2014         }
2015
2016         return (err || sdd.err);
2017
2018 stderr_out:
2019         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
2020 err_out:
2021         fsavl_destroy(fsavl);
2022         nvlist_free(fss);
2023         fnvlist_free(sdd.snapholds);
2024
2025         if (sdd.cleanup_fd != -1)
2026                 VERIFY(0 == close(sdd.cleanup_fd));
2027         if (tid != 0) {
2028                 (void) pthread_cancel(tid);
2029                 (void) close(pipefd[0]);
2030                 (void) pthread_join(tid, NULL);
2031         }
2032         return (err);
2033 }
2034
2035 int
2036 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd, sendflags_t flags)
2037 {
2038         int err = 0;
2039         libzfs_handle_t *hdl = zhp->zfs_hdl;
2040         enum lzc_send_flags lzc_flags = 0;
2041         FILE *fout = (flags.verbose && flags.dryrun) ? stdout : stderr;
2042         char errbuf[1024];
2043
2044         if (flags.largeblock)
2045                 lzc_flags |= LZC_SEND_FLAG_LARGE_BLOCK;
2046         if (flags.embed_data)
2047                 lzc_flags |= LZC_SEND_FLAG_EMBED_DATA;
2048         if (flags.compress)
2049                 lzc_flags |= LZC_SEND_FLAG_COMPRESS;
2050
2051         if (flags.verbose) {
2052                 uint64_t size = 0;
2053                 err = lzc_send_space(zhp->zfs_name, from, lzc_flags, &size);
2054                 if (err == 0) {
2055                         send_print_verbose(fout, zhp->zfs_name, from, size,
2056                             flags.parsable);
2057                 } else {
2058                         (void) fprintf(stderr, "Cannot estimate send size: "
2059                             "%s\n", strerror(errno));
2060                 }
2061         }
2062
2063         if (flags.dryrun)
2064                 return (err);
2065
2066         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2067             "warning: cannot send '%s'"), zhp->zfs_name);
2068
2069         err = lzc_send(zhp->zfs_name, from, fd, lzc_flags);
2070         if (err != 0) {
2071                 switch (errno) {
2072                 case EXDEV:
2073                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2074                             "not an earlier snapshot from the same fs"));
2075                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
2076
2077                 case ENOENT:
2078                 case ESRCH:
2079                         if (lzc_exists(zhp->zfs_name)) {
2080                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2081                                     "incremental source (%s) does not exist"),
2082                                     from);
2083                         }
2084                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2085
2086                 case EBUSY:
2087                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2088                             "target is busy; if a filesystem, "
2089                             "it must not be mounted"));
2090                         return (zfs_error(hdl, EZFS_BUSY, errbuf));
2091
2092                 case EDQUOT:
2093                 case EFBIG:
2094                 case EIO:
2095                 case ENOLINK:
2096                 case ENOSPC:
2097 #ifdef illumos
2098                 case ENOSTR:
2099 #endif
2100                 case ENXIO:
2101                 case EPIPE:
2102                 case ERANGE:
2103                 case EFAULT:
2104                 case EROFS:
2105                         zfs_error_aux(hdl, strerror(errno));
2106                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
2107
2108                 default:
2109                         return (zfs_standard_error(hdl, errno, errbuf));
2110                 }
2111         }
2112         return (err != 0);
2113 }
2114
2115 /*
2116  * Routines specific to "zfs recv"
2117  */
2118
2119 static int
2120 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
2121     boolean_t byteswap, zio_cksum_t *zc)
2122 {
2123         char *cp = buf;
2124         int rv;
2125         int len = ilen;
2126
2127         assert(ilen <= SPA_MAXBLOCKSIZE);
2128
2129         do {
2130                 rv = read(fd, cp, len);
2131                 cp += rv;
2132                 len -= rv;
2133         } while (rv > 0);
2134
2135         if (rv < 0 || len != 0) {
2136                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2137                     "failed to read from stream"));
2138                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
2139                     "cannot receive")));
2140         }
2141
2142         if (zc) {
2143                 if (byteswap)
2144                         (void) fletcher_4_incremental_byteswap(buf, ilen, zc);
2145                 else
2146                         (void) fletcher_4_incremental_native(buf, ilen, zc);
2147         }
2148         return (0);
2149 }
2150
2151 static int
2152 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
2153     boolean_t byteswap, zio_cksum_t *zc)
2154 {
2155         char *buf;
2156         int err;
2157
2158         buf = zfs_alloc(hdl, len);
2159         if (buf == NULL)
2160                 return (ENOMEM);
2161
2162         err = recv_read(hdl, fd, buf, len, byteswap, zc);
2163         if (err != 0) {
2164                 free(buf);
2165                 return (err);
2166         }
2167
2168         err = nvlist_unpack(buf, len, nvp, 0);
2169         free(buf);
2170         if (err != 0) {
2171                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2172                     "stream (malformed nvlist)"));
2173                 return (EINVAL);
2174         }
2175         return (0);
2176 }
2177
2178 static int
2179 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
2180     int baselen, char *newname, recvflags_t *flags)
2181 {
2182         static int seq;
2183         int err;
2184         prop_changelist_t *clp;
2185         zfs_handle_t *zhp;
2186
2187         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2188         if (zhp == NULL)
2189                 return (-1);
2190         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2191             flags->force ? MS_FORCE : 0);
2192         zfs_close(zhp);
2193         if (clp == NULL)
2194                 return (-1);
2195         err = changelist_prefix(clp);
2196         if (err)
2197                 return (err);
2198
2199         if (tryname) {
2200                 (void) strcpy(newname, tryname);
2201                 if (flags->verbose) {
2202                         (void) printf("attempting rename %s to %s\n",
2203                             name, newname);
2204                 }
2205                 err = lzc_rename(name, newname);
2206                 if (err == 0)
2207                         changelist_rename(clp, name, tryname);
2208         } else {
2209                 err = ENOENT;
2210         }
2211
2212         if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
2213                 seq++;
2214
2215                 (void) snprintf(newname, ZFS_MAX_DATASET_NAME_LEN,
2216                     "%.*srecv-%u-%u", baselen, name, getpid(), seq);
2217                 if (flags->verbose) {
2218                         (void) printf("failed - trying rename %s to %s\n",
2219                             name, newname);
2220                 }
2221                 err = lzc_rename(name, newname);
2222                 if (err == 0)
2223                         changelist_rename(clp, name, newname);
2224                 if (err && flags->verbose) {
2225                         (void) printf("failed (%u) - "
2226                             "will try again on next pass\n", errno);
2227                 }
2228                 err = EAGAIN;
2229         } else if (flags->verbose) {
2230                 if (err == 0)
2231                         (void) printf("success\n");
2232                 else
2233                         (void) printf("failed (%u)\n", errno);
2234         }
2235
2236         (void) changelist_postfix(clp);
2237         changelist_free(clp);
2238
2239         return (err);
2240 }
2241
2242 static int
2243 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
2244     char *newname, recvflags_t *flags)
2245 {
2246         int err = 0;
2247         prop_changelist_t *clp;
2248         zfs_handle_t *zhp;
2249         boolean_t defer = B_FALSE;
2250         int spa_version;
2251
2252         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2253         if (zhp == NULL)
2254                 return (-1);
2255         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2256             flags->force ? MS_FORCE : 0);
2257         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
2258             zfs_spa_version(zhp, &spa_version) == 0 &&
2259             spa_version >= SPA_VERSION_USERREFS)
2260                 defer = B_TRUE;
2261         zfs_close(zhp);
2262         if (clp == NULL)
2263                 return (-1);
2264         err = changelist_prefix(clp);
2265         if (err)
2266                 return (err);
2267
2268         if (flags->verbose)
2269                 (void) printf("attempting destroy %s\n", name);
2270         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) {
2271                 nvlist_t *nv = fnvlist_alloc();
2272                 fnvlist_add_boolean(nv, name);
2273                 err = lzc_destroy_snaps(nv, defer, NULL);
2274                 fnvlist_free(nv);
2275         } else {
2276                 err = lzc_destroy(name);
2277         }
2278         if (err == 0) {
2279                 if (flags->verbose)
2280                         (void) printf("success\n");
2281                 changelist_remove(clp, name);
2282         }
2283
2284         (void) changelist_postfix(clp);
2285         changelist_free(clp);
2286
2287         /*
2288          * Deferred destroy might destroy the snapshot or only mark it to be
2289          * destroyed later, and it returns success in either case.
2290          */
2291         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
2292             ZFS_TYPE_SNAPSHOT))) {
2293                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
2294         }
2295
2296         return (err);
2297 }
2298
2299 typedef struct guid_to_name_data {
2300         uint64_t guid;
2301         boolean_t bookmark_ok;
2302         char *name;
2303         char *skip;
2304 } guid_to_name_data_t;
2305
2306 static int
2307 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
2308 {
2309         guid_to_name_data_t *gtnd = arg;
2310         const char *slash;
2311         int err;
2312
2313         if (gtnd->skip != NULL &&
2314             (slash = strrchr(zhp->zfs_name, '/')) != NULL &&
2315             strcmp(slash + 1, gtnd->skip) == 0) {
2316                 zfs_close(zhp);
2317                 return (0);
2318         }
2319
2320         if (zfs_prop_get_int(zhp, ZFS_PROP_GUID) == gtnd->guid) {
2321                 (void) strcpy(gtnd->name, zhp->zfs_name);
2322                 zfs_close(zhp);
2323                 return (EEXIST);
2324         }
2325
2326         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
2327         if (err != EEXIST && gtnd->bookmark_ok)
2328                 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, gtnd);
2329         zfs_close(zhp);
2330         return (err);
2331 }
2332
2333 /*
2334  * Attempt to find the local dataset associated with this guid.  In the case of
2335  * multiple matches, we attempt to find the "best" match by searching
2336  * progressively larger portions of the hierarchy.  This allows one to send a
2337  * tree of datasets individually and guarantee that we will find the source
2338  * guid within that hierarchy, even if there are multiple matches elsewhere.
2339  */
2340 static int
2341 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
2342     boolean_t bookmark_ok, char *name)
2343 {
2344         char pname[ZFS_MAX_DATASET_NAME_LEN];
2345         guid_to_name_data_t gtnd;
2346
2347         gtnd.guid = guid;
2348         gtnd.bookmark_ok = bookmark_ok;
2349         gtnd.name = name;
2350         gtnd.skip = NULL;
2351
2352         /*
2353          * Search progressively larger portions of the hierarchy, starting
2354          * with the filesystem specified by 'parent'.  This will
2355          * select the "most local" version of the origin snapshot in the case
2356          * that there are multiple matching snapshots in the system.
2357          */
2358         (void) strlcpy(pname, parent, sizeof (pname));
2359         char *cp = strrchr(pname, '@');
2360         if (cp == NULL)
2361                 cp = strchr(pname, '\0');
2362         for (; cp != NULL; cp = strrchr(pname, '/')) {
2363                 /* Chop off the last component and open the parent */
2364                 *cp = '\0';
2365                 zfs_handle_t *zhp = make_dataset_handle(hdl, pname);
2366
2367                 if (zhp == NULL)
2368                         continue;
2369                 int err = guid_to_name_cb(zfs_handle_dup(zhp), &gtnd);
2370                 if (err != EEXIST)
2371                         err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
2372                 if (err != EEXIST && bookmark_ok)
2373                         err = zfs_iter_bookmarks(zhp, guid_to_name_cb, &gtnd);
2374                 zfs_close(zhp);
2375                 if (err == EEXIST)
2376                         return (0);
2377
2378                 /*
2379                  * Remember the last portion of the dataset so we skip it next
2380                  * time through (as we've already searched that portion of the
2381                  * hierarchy).
2382                  */
2383                 gtnd.skip = strrchr(pname, '/') + 1;
2384         }
2385
2386         return (ENOENT);
2387 }
2388
2389 /*
2390  * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
2391  * guid1 is after guid2.
2392  */
2393 static int
2394 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
2395     uint64_t guid1, uint64_t guid2)
2396 {
2397         nvlist_t *nvfs;
2398         char *fsname, *snapname;
2399         char buf[ZFS_MAX_DATASET_NAME_LEN];
2400         int rv;
2401         zfs_handle_t *guid1hdl, *guid2hdl;
2402         uint64_t create1, create2;
2403
2404         if (guid2 == 0)
2405                 return (0);
2406         if (guid1 == 0)
2407                 return (1);
2408
2409         nvfs = fsavl_find(avl, guid1, &snapname);
2410         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2411         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2412         guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2413         if (guid1hdl == NULL)
2414                 return (-1);
2415
2416         nvfs = fsavl_find(avl, guid2, &snapname);
2417         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2418         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2419         guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2420         if (guid2hdl == NULL) {
2421                 zfs_close(guid1hdl);
2422                 return (-1);
2423         }
2424
2425         create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
2426         create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
2427
2428         if (create1 < create2)
2429                 rv = -1;
2430         else if (create1 > create2)
2431                 rv = +1;
2432         else
2433                 rv = 0;
2434
2435         zfs_close(guid1hdl);
2436         zfs_close(guid2hdl);
2437
2438         return (rv);
2439 }
2440
2441 static int
2442 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2443     recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2444     nvlist_t *renamed)
2445 {
2446         nvlist_t *local_nv, *deleted = NULL;
2447         avl_tree_t *local_avl;
2448         nvpair_t *fselem, *nextfselem;
2449         char *fromsnap;
2450         char newname[ZFS_MAX_DATASET_NAME_LEN];
2451         char guidname[32];
2452         int error;
2453         boolean_t needagain, progress, recursive;
2454         char *s1, *s2;
2455
2456         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2457
2458         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2459             ENOENT);
2460
2461         if (flags->dryrun)
2462                 return (0);
2463
2464 again:
2465         needagain = progress = B_FALSE;
2466
2467         VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0));
2468
2469         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2470             recursive, B_FALSE, &local_nv, &local_avl)) != 0)
2471                 return (error);
2472
2473         /*
2474          * Process deletes and renames
2475          */
2476         for (fselem = nvlist_next_nvpair(local_nv, NULL);
2477             fselem; fselem = nextfselem) {
2478                 nvlist_t *nvfs, *snaps;
2479                 nvlist_t *stream_nvfs = NULL;
2480                 nvpair_t *snapelem, *nextsnapelem;
2481                 uint64_t fromguid = 0;
2482                 uint64_t originguid = 0;
2483                 uint64_t stream_originguid = 0;
2484                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2485                 char *fsname, *stream_fsname;
2486
2487                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2488
2489                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2490                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2491                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2492                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2493                     &parent_fromsnap_guid));
2494                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2495
2496                 /*
2497                  * First find the stream's fs, so we can check for
2498                  * a different origin (due to "zfs promote")
2499                  */
2500                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2501                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2502                         uint64_t thisguid;
2503
2504                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2505                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2506
2507                         if (stream_nvfs != NULL)
2508                                 break;
2509                 }
2510
2511                 /* check for promote */
2512                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2513                     &stream_originguid);
2514                 if (stream_nvfs && originguid != stream_originguid) {
2515                         switch (created_before(hdl, local_avl,
2516                             stream_originguid, originguid)) {
2517                         case 1: {
2518                                 /* promote it! */
2519                                 zfs_cmd_t zc = { 0 };
2520                                 nvlist_t *origin_nvfs;
2521                                 char *origin_fsname;
2522
2523                                 if (flags->verbose)
2524                                         (void) printf("promoting %s\n", fsname);
2525
2526                                 origin_nvfs = fsavl_find(local_avl, originguid,
2527                                     NULL);
2528                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2529                                     "name", &origin_fsname));
2530                                 (void) strlcpy(zc.zc_value, origin_fsname,
2531                                     sizeof (zc.zc_value));
2532                                 (void) strlcpy(zc.zc_name, fsname,
2533                                     sizeof (zc.zc_name));
2534                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2535                                 if (error == 0)
2536                                         progress = B_TRUE;
2537                                 break;
2538                         }
2539                         default:
2540                                 break;
2541                         case -1:
2542                                 fsavl_destroy(local_avl);
2543                                 nvlist_free(local_nv);
2544                                 return (-1);
2545                         }
2546                         /*
2547                          * We had/have the wrong origin, therefore our
2548                          * list of snapshots is wrong.  Need to handle
2549                          * them on the next pass.
2550                          */
2551                         needagain = B_TRUE;
2552                         continue;
2553                 }
2554
2555                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2556                     snapelem; snapelem = nextsnapelem) {
2557                         uint64_t thisguid;
2558                         char *stream_snapname;
2559                         nvlist_t *found, *props;
2560
2561                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2562
2563                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2564                         found = fsavl_find(stream_avl, thisguid,
2565                             &stream_snapname);
2566
2567                         /* check for delete */
2568                         if (found == NULL) {
2569                                 char name[ZFS_MAX_DATASET_NAME_LEN];
2570
2571                                 if (!flags->force)
2572                                         continue;
2573
2574                                 (void) snprintf(name, sizeof (name), "%s@%s",
2575                                     fsname, nvpair_name(snapelem));
2576
2577                                 error = recv_destroy(hdl, name,
2578                                     strlen(fsname)+1, newname, flags);
2579                                 if (error)
2580                                         needagain = B_TRUE;
2581                                 else
2582                                         progress = B_TRUE;
2583                                 sprintf(guidname, "%" PRIu64, thisguid);
2584                                 nvlist_add_boolean(deleted, guidname);
2585                                 continue;
2586                         }
2587
2588                         stream_nvfs = found;
2589
2590                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2591                             &props) && 0 == nvlist_lookup_nvlist(props,
2592                             stream_snapname, &props)) {
2593                                 zfs_cmd_t zc = { 0 };
2594
2595                                 zc.zc_cookie = B_TRUE; /* received */
2596                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2597                                     "%s@%s", fsname, nvpair_name(snapelem));
2598                                 if (zcmd_write_src_nvlist(hdl, &zc,
2599                                     props) == 0) {
2600                                         (void) zfs_ioctl(hdl,
2601                                             ZFS_IOC_SET_PROP, &zc);
2602                                         zcmd_free_nvlists(&zc);
2603                                 }
2604                         }
2605
2606                         /* check for different snapname */
2607                         if (strcmp(nvpair_name(snapelem),
2608                             stream_snapname) != 0) {
2609                                 char name[ZFS_MAX_DATASET_NAME_LEN];
2610                                 char tryname[ZFS_MAX_DATASET_NAME_LEN];
2611
2612                                 (void) snprintf(name, sizeof (name), "%s@%s",
2613                                     fsname, nvpair_name(snapelem));
2614                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
2615                                     fsname, stream_snapname);
2616
2617                                 error = recv_rename(hdl, name, tryname,
2618                                     strlen(fsname)+1, newname, flags);
2619                                 if (error)
2620                                         needagain = B_TRUE;
2621                                 else
2622                                         progress = B_TRUE;
2623                         }
2624
2625                         if (strcmp(stream_snapname, fromsnap) == 0)
2626                                 fromguid = thisguid;
2627                 }
2628
2629                 /* check for delete */
2630                 if (stream_nvfs == NULL) {
2631                         if (!flags->force)
2632                                 continue;
2633
2634                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2635                             newname, flags);
2636                         if (error)
2637                                 needagain = B_TRUE;
2638                         else
2639                                 progress = B_TRUE;
2640                         sprintf(guidname, "%" PRIu64, parent_fromsnap_guid);
2641                         nvlist_add_boolean(deleted, guidname);
2642                         continue;
2643                 }
2644
2645                 if (fromguid == 0) {
2646                         if (flags->verbose) {
2647                                 (void) printf("local fs %s does not have "
2648                                     "fromsnap (%s in stream); must have "
2649                                     "been deleted locally; ignoring\n",
2650                                     fsname, fromsnap);
2651                         }
2652                         continue;
2653                 }
2654
2655                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2656                     "name", &stream_fsname));
2657                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2658                     "parentfromsnap", &stream_parent_fromsnap_guid));
2659
2660                 s1 = strrchr(fsname, '/');
2661                 s2 = strrchr(stream_fsname, '/');
2662
2663                 /*
2664                  * Check if we're going to rename based on parent guid change
2665                  * and the current parent guid was also deleted. If it was then
2666                  * rename will fail and is likely unneeded, so avoid this and
2667                  * force an early retry to determine the new
2668                  * parent_fromsnap_guid.
2669                  */
2670                 if (stream_parent_fromsnap_guid != 0 &&
2671                     parent_fromsnap_guid != 0 &&
2672                     stream_parent_fromsnap_guid != parent_fromsnap_guid) {
2673                         sprintf(guidname, "%" PRIu64, parent_fromsnap_guid);
2674                         if (nvlist_exists(deleted, guidname)) {
2675                                 progress = B_TRUE;
2676                                 needagain = B_TRUE;
2677                                 goto doagain;
2678                         }
2679                 }
2680
2681                 /*
2682                  * Check for rename. If the exact receive path is specified, it
2683                  * does not count as a rename, but we still need to check the
2684                  * datasets beneath it.
2685                  */
2686                 if ((stream_parent_fromsnap_guid != 0 &&
2687                     parent_fromsnap_guid != 0 &&
2688                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2689                     ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2690                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2691                         nvlist_t *parent;
2692                         char tryname[ZFS_MAX_DATASET_NAME_LEN];
2693
2694                         parent = fsavl_find(local_avl,
2695                             stream_parent_fromsnap_guid, NULL);
2696                         /*
2697                          * NB: parent might not be found if we used the
2698                          * tosnap for stream_parent_fromsnap_guid,
2699                          * because the parent is a newly-created fs;
2700                          * we'll be able to rename it after we recv the
2701                          * new fs.
2702                          */
2703                         if (parent != NULL) {
2704                                 char *pname;
2705
2706                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
2707                                     &pname));
2708                                 (void) snprintf(tryname, sizeof (tryname),
2709                                     "%s%s", pname, strrchr(stream_fsname, '/'));
2710                         } else {
2711                                 tryname[0] = '\0';
2712                                 if (flags->verbose) {
2713                                         (void) printf("local fs %s new parent "
2714                                             "not found\n", fsname);
2715                                 }
2716                         }
2717
2718                         newname[0] = '\0';
2719
2720                         error = recv_rename(hdl, fsname, tryname,
2721                             strlen(tofs)+1, newname, flags);
2722
2723                         if (renamed != NULL && newname[0] != '\0') {
2724                                 VERIFY(0 == nvlist_add_boolean(renamed,
2725                                     newname));
2726                         }
2727
2728                         if (error)
2729                                 needagain = B_TRUE;
2730                         else
2731                                 progress = B_TRUE;
2732                 }
2733         }
2734
2735 doagain:
2736         fsavl_destroy(local_avl);
2737         nvlist_free(local_nv);
2738         nvlist_free(deleted);
2739
2740         if (needagain && progress) {
2741                 /* do another pass to fix up temporary names */
2742                 if (flags->verbose)
2743                         (void) printf("another pass:\n");
2744                 goto again;
2745         }
2746
2747         return (needagain);
2748 }
2749
2750 static int
2751 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2752     recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2753     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2754 {
2755         nvlist_t *stream_nv = NULL;
2756         avl_tree_t *stream_avl = NULL;
2757         char *fromsnap = NULL;
2758         char *sendsnap = NULL;
2759         char *cp;
2760         char tofs[ZFS_MAX_DATASET_NAME_LEN];
2761         char sendfs[ZFS_MAX_DATASET_NAME_LEN];
2762         char errbuf[1024];
2763         dmu_replay_record_t drre;
2764         int error;
2765         boolean_t anyerr = B_FALSE;
2766         boolean_t softerr = B_FALSE;
2767         boolean_t recursive;
2768
2769         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2770             "cannot receive"));
2771
2772         assert(drr->drr_type == DRR_BEGIN);
2773         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2774         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2775             DMU_COMPOUNDSTREAM);
2776
2777         /*
2778          * Read in the nvlist from the stream.
2779          */
2780         if (drr->drr_payloadlen != 0) {
2781                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2782                     &stream_nv, flags->byteswap, zc);
2783                 if (error) {
2784                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2785                         goto out;
2786                 }
2787         }
2788
2789         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2790             ENOENT);
2791
2792         if (recursive && strchr(destname, '@')) {
2793                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2794                     "cannot specify snapshot name for multi-snapshot stream"));
2795                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2796                 goto out;
2797         }
2798
2799         /*
2800          * Read in the end record and verify checksum.
2801          */
2802         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2803             flags->byteswap, NULL)))
2804                 goto out;
2805         if (flags->byteswap) {
2806                 drre.drr_type = BSWAP_32(drre.drr_type);
2807                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2808                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2809                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2810                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2811                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2812                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2813                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2814                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2815         }
2816         if (drre.drr_type != DRR_END) {
2817                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2818                 goto out;
2819         }
2820         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2821                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2822                     "incorrect header checksum"));
2823                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2824                 goto out;
2825         }
2826
2827         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2828
2829         if (drr->drr_payloadlen != 0) {
2830                 nvlist_t *stream_fss;
2831
2832                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2833                     &stream_fss));
2834                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2835                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2836                             "couldn't allocate avl tree"));
2837                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2838                         goto out;
2839                 }
2840
2841                 if (fromsnap != NULL && recursive) {
2842                         nvlist_t *renamed = NULL;
2843                         nvpair_t *pair = NULL;
2844
2845                         (void) strlcpy(tofs, destname, sizeof (tofs));
2846                         if (flags->isprefix) {
2847                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2848                                 int i;
2849
2850                                 if (flags->istail) {
2851                                         cp = strrchr(drrb->drr_toname, '/');
2852                                         if (cp == NULL) {
2853                                                 (void) strlcat(tofs, "/",
2854                                                     sizeof (tofs));
2855                                                 i = 0;
2856                                         } else {
2857                                                 i = (cp - drrb->drr_toname);
2858                                         }
2859                                 } else {
2860                                         i = strcspn(drrb->drr_toname, "/@");
2861                                 }
2862                                 /* zfs_receive_one() will create_parents() */
2863                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2864                                     sizeof (tofs));
2865                                 *strchr(tofs, '@') = '\0';
2866                         }
2867
2868                         if (!flags->dryrun && !flags->nomount) {
2869                                 VERIFY(0 == nvlist_alloc(&renamed,
2870                                     NV_UNIQUE_NAME, 0));
2871                         }
2872
2873                         softerr = recv_incremental_replication(hdl, tofs, flags,
2874                             stream_nv, stream_avl, renamed);
2875
2876                         /* Unmount renamed filesystems before receiving. */
2877                         while ((pair = nvlist_next_nvpair(renamed,
2878                             pair)) != NULL) {
2879                                 zfs_handle_t *zhp;
2880                                 prop_changelist_t *clp = NULL;
2881
2882                                 zhp = zfs_open(hdl, nvpair_name(pair),
2883                                     ZFS_TYPE_FILESYSTEM);
2884                                 if (zhp != NULL) {
2885                                         clp = changelist_gather(zhp,
2886                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2887                                         zfs_close(zhp);
2888                                         if (clp != NULL) {
2889                                                 softerr |=
2890                                                     changelist_prefix(clp);
2891                                                 changelist_free(clp);
2892                                         }
2893                                 }
2894                         }
2895
2896                         nvlist_free(renamed);
2897                 }
2898         }
2899
2900         /*
2901          * Get the fs specified by the first path in the stream (the top level
2902          * specified by 'zfs send') and pass it to each invocation of
2903          * zfs_receive_one().
2904          */
2905         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2906             sizeof (sendfs));
2907         if ((cp = strchr(sendfs, '@')) != NULL) {
2908                 *cp = '\0';
2909                 /*
2910                  * Find the "sendsnap", the final snapshot in a replication
2911                  * stream.  zfs_receive_one() handles certain errors
2912                  * differently, depending on if the contained stream is the
2913                  * last one or not.
2914                  */
2915                 sendsnap = (cp + 1);
2916         }
2917
2918         /* Finally, receive each contained stream */
2919         do {
2920                 /*
2921                  * we should figure out if it has a recoverable
2922                  * error, in which case do a recv_skip() and drive on.
2923                  * Note, if we fail due to already having this guid,
2924                  * zfs_receive_one() will take care of it (ie,
2925                  * recv_skip() and return 0).
2926                  */
2927                 error = zfs_receive_impl(hdl, destname, NULL, flags, fd,
2928                     sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2929                     action_handlep, sendsnap);
2930                 if (error == ENODATA) {
2931                         error = 0;
2932                         break;
2933                 }
2934                 anyerr |= error;
2935         } while (error == 0);
2936
2937         if (drr->drr_payloadlen != 0 && recursive && fromsnap != NULL) {
2938                 /*
2939                  * Now that we have the fs's they sent us, try the
2940                  * renames again.
2941                  */
2942                 softerr = recv_incremental_replication(hdl, tofs, flags,
2943                     stream_nv, stream_avl, NULL);
2944         }
2945
2946 out:
2947         fsavl_destroy(stream_avl);
2948         nvlist_free(stream_nv);
2949         if (softerr)
2950                 error = -2;
2951         if (anyerr)
2952                 error = -1;
2953         return (error);
2954 }
2955
2956 static void
2957 trunc_prop_errs(int truncated)
2958 {
2959         ASSERT(truncated != 0);
2960
2961         if (truncated == 1)
2962                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2963                     "1 more property could not be set\n"));
2964         else
2965                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2966                     "%d more properties could not be set\n"), truncated);
2967 }
2968
2969 static int
2970 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2971 {
2972         dmu_replay_record_t *drr;
2973         void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
2974         char errbuf[1024];
2975
2976         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2977             "cannot receive:"));
2978
2979         /* XXX would be great to use lseek if possible... */
2980         drr = buf;
2981
2982         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2983             byteswap, NULL) == 0) {
2984                 if (byteswap)
2985                         drr->drr_type = BSWAP_32(drr->drr_type);
2986
2987                 switch (drr->drr_type) {
2988                 case DRR_BEGIN:
2989                         if (drr->drr_payloadlen != 0) {
2990                                 (void) recv_read(hdl, fd, buf,
2991                                     drr->drr_payloadlen, B_FALSE, NULL);
2992                         }
2993                         break;
2994
2995                 case DRR_END:
2996                         free(buf);
2997                         return (0);
2998
2999                 case DRR_OBJECT:
3000                         if (byteswap) {
3001                                 drr->drr_u.drr_object.drr_bonuslen =
3002                                     BSWAP_32(drr->drr_u.drr_object.
3003                                     drr_bonuslen);
3004                         }
3005                         (void) recv_read(hdl, fd, buf,
3006                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
3007                             B_FALSE, NULL);
3008                         break;
3009
3010                 case DRR_WRITE:
3011                         if (byteswap) {
3012                                 drr->drr_u.drr_write.drr_logical_size =
3013                                     BSWAP_64(
3014                                     drr->drr_u.drr_write.drr_logical_size);
3015                                 drr->drr_u.drr_write.drr_compressed_size =
3016                                     BSWAP_64(
3017                                     drr->drr_u.drr_write.drr_compressed_size);
3018                         }
3019                         uint64_t payload_size =
3020                             DRR_WRITE_PAYLOAD_SIZE(&drr->drr_u.drr_write);
3021                         (void) recv_read(hdl, fd, buf,
3022                             payload_size, B_FALSE, NULL);
3023                         break;
3024                 case DRR_SPILL:
3025                         if (byteswap) {
3026                                 drr->drr_u.drr_spill.drr_length =
3027                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
3028                         }
3029                         (void) recv_read(hdl, fd, buf,
3030                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
3031                         break;
3032                 case DRR_WRITE_EMBEDDED:
3033                         if (byteswap) {
3034                                 drr->drr_u.drr_write_embedded.drr_psize =
3035                                     BSWAP_32(drr->drr_u.drr_write_embedded.
3036                                     drr_psize);
3037                         }
3038                         (void) recv_read(hdl, fd, buf,
3039                             P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
3040                             8), B_FALSE, NULL);
3041                         break;
3042                 case DRR_WRITE_BYREF:
3043                 case DRR_FREEOBJECTS:
3044                 case DRR_FREE:
3045                         break;
3046
3047                 default:
3048                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3049                             "invalid record type"));
3050                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3051                 }
3052         }
3053
3054         free(buf);
3055         return (-1);
3056 }
3057
3058 static void
3059 recv_ecksum_set_aux(libzfs_handle_t *hdl, const char *target_snap,
3060     boolean_t resumable)
3061 {
3062         char target_fs[ZFS_MAX_DATASET_NAME_LEN];
3063
3064         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3065             "checksum mismatch or incomplete stream"));
3066
3067         if (!resumable)
3068                 return;
3069         (void) strlcpy(target_fs, target_snap, sizeof (target_fs));
3070         *strchr(target_fs, '@') = '\0';
3071         zfs_handle_t *zhp = zfs_open(hdl, target_fs,
3072             ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3073         if (zhp == NULL)
3074                 return;
3075
3076         char token_buf[ZFS_MAXPROPLEN];
3077         int error = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN,
3078             token_buf, sizeof (token_buf),
3079             NULL, NULL, 0, B_TRUE);
3080         if (error == 0) {
3081                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3082                     "checksum mismatch or incomplete stream.\n"
3083                     "Partially received snapshot is saved.\n"
3084                     "A resuming stream can be generated on the sending "
3085                     "system by running:\n"
3086                     "    zfs send -t %s"),
3087                     token_buf);
3088         }
3089         zfs_close(zhp);
3090 }
3091
3092 /*
3093  * Restores a backup of tosnap from the file descriptor specified by infd.
3094  */
3095 static int
3096 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
3097     const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr,
3098     dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv,
3099     avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
3100     uint64_t *action_handlep, const char *finalsnap)
3101 {
3102         zfs_cmd_t zc = { 0 };
3103         time_t begin_time;
3104         int ioctl_err, ioctl_errno, err;
3105         char *cp;
3106         struct drr_begin *drrb = &drr->drr_u.drr_begin;
3107         char errbuf[1024];
3108         char prop_errbuf[1024];
3109         const char *chopprefix;
3110         boolean_t newfs = B_FALSE;
3111         boolean_t stream_wantsnewfs;
3112         uint64_t parent_snapguid = 0;
3113         prop_changelist_t *clp = NULL;
3114         nvlist_t *snapprops_nvlist = NULL;
3115         zprop_errflags_t prop_errflags;
3116         boolean_t recursive;
3117         char *snapname = NULL;
3118
3119         begin_time = time(NULL);
3120
3121         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3122             "cannot receive"));
3123
3124         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
3125             ENOENT);
3126
3127         if (stream_avl != NULL) {
3128                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
3129                     &snapname);
3130                 nvlist_t *props;
3131                 int ret;
3132
3133                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
3134                     &parent_snapguid);
3135                 err = nvlist_lookup_nvlist(fs, "props", &props);
3136                 if (err)
3137                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
3138
3139                 if (flags->canmountoff) {
3140                         VERIFY(0 == nvlist_add_uint64(props,
3141                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
3142                 }
3143                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
3144                 if (err)
3145                         nvlist_free(props);
3146
3147                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
3148                         VERIFY(0 == nvlist_lookup_nvlist(props,
3149                             snapname, &snapprops_nvlist));
3150                 }
3151
3152                 if (ret != 0)
3153                         return (-1);
3154         }
3155
3156         cp = NULL;
3157
3158         /*
3159          * Determine how much of the snapshot name stored in the stream
3160          * we are going to tack on to the name they specified on the
3161          * command line, and how much we are going to chop off.
3162          *
3163          * If they specified a snapshot, chop the entire name stored in
3164          * the stream.
3165          */
3166         if (flags->istail) {
3167                 /*
3168                  * A filesystem was specified with -e. We want to tack on only
3169                  * the tail of the sent snapshot path.
3170                  */
3171                 if (strchr(tosnap, '@')) {
3172                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3173                             "argument - snapshot not allowed with -e"));
3174                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3175                 }
3176
3177                 chopprefix = strrchr(sendfs, '/');
3178
3179                 if (chopprefix == NULL) {
3180                         /*
3181                          * The tail is the poolname, so we need to
3182                          * prepend a path separator.
3183                          */
3184                         int len = strlen(drrb->drr_toname);
3185                         cp = malloc(len + 2);
3186                         cp[0] = '/';
3187                         (void) strcpy(&cp[1], drrb->drr_toname);
3188                         chopprefix = cp;
3189                 } else {
3190                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
3191                 }
3192         } else if (flags->isprefix) {
3193                 /*
3194                  * A filesystem was specified with -d. We want to tack on
3195                  * everything but the first element of the sent snapshot path
3196                  * (all but the pool name).
3197                  */
3198                 if (strchr(tosnap, '@')) {
3199                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3200                             "argument - snapshot not allowed with -d"));
3201                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3202                 }
3203
3204                 chopprefix = strchr(drrb->drr_toname, '/');
3205                 if (chopprefix == NULL)
3206                         chopprefix = strchr(drrb->drr_toname, '@');
3207         } else if (strchr(tosnap, '@') == NULL) {
3208                 /*
3209                  * If a filesystem was specified without -d or -e, we want to
3210                  * tack on everything after the fs specified by 'zfs send'.
3211                  */
3212                 chopprefix = drrb->drr_toname + strlen(sendfs);
3213         } else {
3214                 /* A snapshot was specified as an exact path (no -d or -e). */
3215                 if (recursive) {
3216                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3217                             "cannot specify snapshot name for multi-snapshot "
3218                             "stream"));
3219                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3220                 }
3221                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
3222         }
3223
3224         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
3225         ASSERT(chopprefix > drrb->drr_toname);
3226         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
3227         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
3228             chopprefix[0] == '\0');
3229
3230         /*
3231          * Determine name of destination snapshot, store in zc_value.
3232          */
3233         (void) strcpy(zc.zc_value, tosnap);
3234         (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
3235 #ifdef __FreeBSD__
3236         if (zfs_ioctl_version == ZFS_IOCVER_UNDEF)
3237                 zfs_ioctl_version = get_zfs_ioctl_version();
3238         /*
3239          * For forward compatibility hide tosnap in zc_value
3240          */
3241         if (zfs_ioctl_version < ZFS_IOCVER_LZC)
3242                 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap);
3243 #endif
3244         free(cp);
3245         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
3246                 zcmd_free_nvlists(&zc);
3247                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3248         }
3249
3250         /*
3251          * Determine the name of the origin snapshot, store in zc_string.
3252          */
3253         if (originsnap) {
3254                 (void) strncpy(zc.zc_string, originsnap, sizeof (zc.zc_string));
3255                 if (flags->verbose)
3256                         (void) printf("using provided clone origin %s\n",
3257                             zc.zc_string);
3258         } else if (drrb->drr_flags & DRR_FLAG_CLONE) {
3259                 if (guid_to_name(hdl, zc.zc_value,
3260                     drrb->drr_fromguid, B_FALSE, zc.zc_string) != 0) {
3261                         zcmd_free_nvlists(&zc);
3262                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3263                             "local origin for clone %s does not exist"),
3264                             zc.zc_value);
3265                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
3266                 }
3267                 if (flags->verbose)
3268                         (void) printf("found clone origin %s\n", zc.zc_string);
3269         }
3270
3271         boolean_t resuming = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
3272             DMU_BACKUP_FEATURE_RESUMING;
3273         stream_wantsnewfs = (drrb->drr_fromguid == 0 ||
3274             (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && !resuming;
3275
3276         if (stream_wantsnewfs) {
3277                 /*
3278                  * if the parent fs does not exist, look for it based on
3279                  * the parent snap GUID
3280                  */
3281                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3282                     "cannot receive new filesystem stream"));
3283
3284                 (void) strcpy(zc.zc_name, zc.zc_value);
3285                 cp = strrchr(zc.zc_name, '/');
3286                 if (cp)
3287                         *cp = '\0';
3288                 if (cp &&
3289                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3290                         char suffix[ZFS_MAX_DATASET_NAME_LEN];
3291                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
3292                         if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
3293                             B_FALSE, zc.zc_value) == 0) {
3294                                 *strchr(zc.zc_value, '@') = '\0';
3295                                 (void) strcat(zc.zc_value, suffix);
3296                         }
3297                 }
3298         } else {
3299                 /*
3300                  * If the fs does not exist, look for it based on the
3301                  * fromsnap GUID.
3302                  */
3303                 if (resuming) {
3304                         (void) snprintf(errbuf, sizeof (errbuf),
3305                             dgettext(TEXT_DOMAIN,
3306                             "cannot receive resume stream"));
3307                 } else {
3308                         (void) snprintf(errbuf, sizeof (errbuf),
3309                             dgettext(TEXT_DOMAIN,
3310                             "cannot receive incremental stream"));
3311                 }
3312
3313                 (void) strcpy(zc.zc_name, zc.zc_value);
3314                 *strchr(zc.zc_name, '@') = '\0';
3315
3316                 /*
3317                  * If the exact receive path was specified and this is the
3318                  * topmost path in the stream, then if the fs does not exist we
3319                  * should look no further.
3320                  */
3321                 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
3322                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
3323                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3324                         char snap[ZFS_MAX_DATASET_NAME_LEN];
3325                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
3326                         if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
3327                             B_FALSE, zc.zc_value) == 0) {
3328                                 *strchr(zc.zc_value, '@') = '\0';
3329                                 (void) strcat(zc.zc_value, snap);
3330                         }
3331                 }
3332         }
3333
3334         (void) strcpy(zc.zc_name, zc.zc_value);
3335         *strchr(zc.zc_name, '@') = '\0';
3336
3337         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3338                 zfs_handle_t *zhp;
3339
3340                 /*
3341                  * Destination fs exists.  It must be one of these cases:
3342                  *  - an incremental send stream
3343                  *  - the stream specifies a new fs (full stream or clone)
3344                  *    and they want us to blow away the existing fs (and
3345                  *    have therefore specified -F and removed any snapshots)
3346                  *  - we are resuming a failed receive.
3347                  */
3348                 if (stream_wantsnewfs) {
3349                         if (!flags->force) {
3350                                 zcmd_free_nvlists(&zc);
3351                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3352                                     "destination '%s' exists\n"
3353                                     "must specify -F to overwrite it"),
3354                                     zc.zc_name);
3355                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3356                         }
3357                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
3358                             &zc) == 0) {
3359                                 zcmd_free_nvlists(&zc);
3360                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3361                                     "destination has snapshots (eg. %s)\n"
3362                                     "must destroy them to overwrite it"),
3363                                     zc.zc_name);
3364                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3365                         }
3366                 }
3367
3368                 if ((zhp = zfs_open(hdl, zc.zc_name,
3369                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
3370                         zcmd_free_nvlists(&zc);
3371                         return (-1);
3372                 }
3373
3374                 if (stream_wantsnewfs &&
3375                     zhp->zfs_dmustats.dds_origin[0]) {
3376                         zcmd_free_nvlists(&zc);
3377                         zfs_close(zhp);
3378                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3379                             "destination '%s' is a clone\n"
3380                             "must destroy it to overwrite it"),
3381                             zc.zc_name);
3382                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3383                 }
3384
3385                 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
3386                     stream_wantsnewfs) {
3387                         /* We can't do online recv in this case */
3388                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
3389                         if (clp == NULL) {
3390                                 zfs_close(zhp);
3391                                 zcmd_free_nvlists(&zc);
3392                                 return (-1);
3393                         }
3394                         if (changelist_prefix(clp) != 0) {
3395                                 changelist_free(clp);
3396                                 zfs_close(zhp);
3397                                 zcmd_free_nvlists(&zc);
3398                                 return (-1);
3399                         }
3400                 }
3401
3402                 /*
3403                  * If we are resuming a newfs, set newfs here so that we will
3404                  * mount it if the recv succeeds this time.  We can tell
3405                  * that it was a newfs on the first recv because the fs
3406                  * itself will be inconsistent (if the fs existed when we
3407                  * did the first recv, we would have received it into
3408                  * .../%recv).
3409                  */
3410                 if (resuming && zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT))
3411                         newfs = B_TRUE;
3412
3413                 zfs_close(zhp);
3414         } else {
3415                 /*
3416                  * Destination filesystem does not exist.  Therefore we better
3417                  * be creating a new filesystem (either from a full backup, or
3418                  * a clone).  It would therefore be invalid if the user
3419                  * specified only the pool name (i.e. if the destination name
3420                  * contained no slash character).
3421                  */
3422                 if (!stream_wantsnewfs ||
3423                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
3424                         zcmd_free_nvlists(&zc);
3425                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3426                             "destination '%s' does not exist"), zc.zc_name);
3427                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
3428                 }
3429
3430                 /*
3431                  * Trim off the final dataset component so we perform the
3432                  * recvbackup ioctl to the filesystems's parent.
3433                  */
3434                 *cp = '\0';
3435
3436                 if (flags->isprefix && !flags->istail && !flags->dryrun &&
3437                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
3438                         zcmd_free_nvlists(&zc);
3439                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
3440                 }
3441
3442                 newfs = B_TRUE;
3443         }
3444
3445         zc.zc_begin_record = *drr_noswap;
3446         zc.zc_cookie = infd;
3447         zc.zc_guid = flags->force;
3448         zc.zc_resumable = flags->resumable;
3449         if (flags->verbose) {
3450                 (void) printf("%s %s stream of %s into %s\n",
3451                     flags->dryrun ? "would receive" : "receiving",
3452                     drrb->drr_fromguid ? "incremental" : "full",
3453                     drrb->drr_toname, zc.zc_value);
3454                 (void) fflush(stdout);
3455         }
3456
3457         if (flags->dryrun) {
3458                 zcmd_free_nvlists(&zc);
3459                 return (recv_skip(hdl, infd, flags->byteswap));
3460         }
3461
3462         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
3463         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
3464         zc.zc_cleanup_fd = cleanup_fd;
3465         zc.zc_action_handle = *action_handlep;
3466
3467         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
3468         ioctl_errno = errno;
3469         prop_errflags = (zprop_errflags_t)zc.zc_obj;
3470
3471         if (err == 0) {
3472                 nvlist_t *prop_errors;
3473                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
3474                     zc.zc_nvlist_dst_size, &prop_errors, 0));
3475
3476                 nvpair_t *prop_err = NULL;
3477
3478                 while ((prop_err = nvlist_next_nvpair(prop_errors,
3479                     prop_err)) != NULL) {
3480                         char tbuf[1024];
3481                         zfs_prop_t prop;
3482                         int intval;
3483
3484                         prop = zfs_name_to_prop(nvpair_name(prop_err));
3485                         (void) nvpair_value_int32(prop_err, &intval);
3486                         if (strcmp(nvpair_name(prop_err),
3487                             ZPROP_N_MORE_ERRORS) == 0) {
3488                                 trunc_prop_errs(intval);
3489                                 break;
3490                         } else if (snapname == NULL || finalsnap == NULL ||
3491                             strcmp(finalsnap, snapname) == 0 ||
3492                             strcmp(nvpair_name(prop_err),
3493                             zfs_prop_to_name(ZFS_PROP_REFQUOTA)) != 0) {
3494                                 /*
3495                                  * Skip the special case of, for example,
3496                                  * "refquota", errors on intermediate
3497                                  * snapshots leading up to a final one.
3498                                  * That's why we have all of the checks above.
3499                                  *
3500                                  * See zfs_ioctl.c's extract_delay_props() for
3501                                  * a list of props which can fail on
3502                                  * intermediate snapshots, but shouldn't
3503                                  * affect the overall receive.
3504                                  */
3505                                 (void) snprintf(tbuf, sizeof (tbuf),
3506                                     dgettext(TEXT_DOMAIN,
3507                                     "cannot receive %s property on %s"),
3508                                     nvpair_name(prop_err), zc.zc_name);
3509                                 zfs_setprop_error(hdl, prop, intval, tbuf);
3510                         }
3511                 }
3512                 nvlist_free(prop_errors);
3513         }
3514
3515         zc.zc_nvlist_dst = 0;
3516         zc.zc_nvlist_dst_size = 0;
3517         zcmd_free_nvlists(&zc);
3518
3519         if (err == 0 && snapprops_nvlist) {
3520                 zfs_cmd_t zc2 = { 0 };
3521
3522                 (void) strcpy(zc2.zc_name, zc.zc_value);
3523                 zc2.zc_cookie = B_TRUE; /* received */
3524                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
3525                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
3526                         zcmd_free_nvlists(&zc2);
3527                 }
3528         }
3529
3530         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
3531                 /*
3532                  * It may be that this snapshot already exists,
3533                  * in which case we want to consume & ignore it
3534                  * rather than failing.
3535                  */
3536                 avl_tree_t *local_avl;
3537                 nvlist_t *local_nv, *fs;
3538                 cp = strchr(zc.zc_value, '@');
3539
3540                 /*
3541                  * XXX Do this faster by just iterating over snaps in
3542                  * this fs.  Also if zc_value does not exist, we will
3543                  * get a strange "does not exist" error message.
3544                  */
3545                 *cp = '\0';
3546                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3547                     B_FALSE, &local_nv, &local_avl) == 0) {
3548                         *cp = '@';
3549                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3550                         fsavl_destroy(local_avl);
3551                         nvlist_free(local_nv);
3552
3553                         if (fs != NULL) {
3554                                 if (flags->verbose) {
3555                                         (void) printf("snap %s already exists; "
3556                                             "ignoring\n", zc.zc_value);
3557                                 }
3558                                 err = ioctl_err = recv_skip(hdl, infd,
3559                                     flags->byteswap);
3560                         }
3561                 }
3562                 *cp = '@';
3563         }
3564
3565         if (ioctl_err != 0) {
3566                 switch (ioctl_errno) {
3567                 case ENODEV:
3568                         cp = strchr(zc.zc_value, '@');
3569                         *cp = '\0';
3570                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3571                             "most recent snapshot of %s does not\n"
3572                             "match incremental source"), zc.zc_value);
3573                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3574                         *cp = '@';
3575                         break;
3576                 case ETXTBSY:
3577                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3578                             "destination %s has been modified\n"
3579                             "since most recent snapshot"), zc.zc_name);
3580                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3581                         break;
3582                 case EEXIST:
3583                         cp = strchr(zc.zc_value, '@');
3584                         if (newfs) {
3585                                 /* it's the containing fs that exists */
3586                                 *cp = '\0';
3587                         }
3588                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3589                             "destination already exists"));
3590                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3591                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3592                             zc.zc_value);
3593                         *cp = '@';
3594                         break;
3595                 case EINVAL:
3596                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3597                         break;
3598                 case ECKSUM:
3599                         recv_ecksum_set_aux(hdl, zc.zc_value, flags->resumable);
3600                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3601                         break;
3602                 case ENOTSUP:
3603                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3604                             "pool must be upgraded to receive this stream."));
3605                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3606                         break;
3607                 case EDQUOT:
3608                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3609                             "destination %s space quota exceeded"), zc.zc_name);
3610                         (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3611                         break;
3612                 default:
3613                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3614                 }
3615         }
3616
3617         /*
3618          * Mount the target filesystem (if created).  Also mount any
3619          * children of the target filesystem if we did a replication
3620          * receive (indicated by stream_avl being non-NULL).
3621          */
3622         cp = strchr(zc.zc_value, '@');
3623         if (cp && (ioctl_err == 0 || !newfs)) {
3624                 zfs_handle_t *h;
3625
3626                 *cp = '\0';
3627                 h = zfs_open(hdl, zc.zc_value,
3628                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3629                 if (h != NULL) {
3630                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
3631                                 *cp = '@';
3632                         } else if (newfs || stream_avl) {
3633                                 /*
3634                                  * Track the first/top of hierarchy fs,
3635                                  * for mounting and sharing later.
3636                                  */
3637                                 if (top_zfs && *top_zfs == NULL)
3638                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
3639                         }
3640                         zfs_close(h);
3641                 }
3642                 *cp = '@';
3643         }
3644
3645         if (clp) {
3646                 if (!flags->nomount)
3647                         err |= changelist_postfix(clp);
3648                 changelist_free(clp);
3649         }
3650
3651         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3652                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3653                     "failed to clear unreceived properties on %s"),
3654                     zc.zc_name);
3655                 (void) fprintf(stderr, "\n");
3656         }
3657         if (prop_errflags & ZPROP_ERR_NORESTORE) {
3658                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3659                     "failed to restore original properties on %s"),
3660                     zc.zc_name);
3661                 (void) fprintf(stderr, "\n");
3662         }
3663
3664         if (err || ioctl_err)
3665                 return (-1);
3666
3667         *action_handlep = zc.zc_action_handle;
3668
3669         if (flags->verbose) {
3670                 char buf1[64];
3671                 char buf2[64];
3672                 uint64_t bytes = zc.zc_cookie;
3673                 time_t delta = time(NULL) - begin_time;
3674                 if (delta == 0)
3675                         delta = 1;
3676                 zfs_nicenum(bytes, buf1, sizeof (buf1));
3677                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3678
3679                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3680                     buf1, delta, buf2);
3681         }
3682
3683         return (0);
3684 }
3685
3686 static int
3687 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap,
3688     const char *originsnap, recvflags_t *flags, int infd, const char *sendfs,
3689     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
3690     uint64_t *action_handlep, const char *finalsnap)
3691 {
3692         int err;
3693         dmu_replay_record_t drr, drr_noswap;
3694         struct drr_begin *drrb = &drr.drr_u.drr_begin;
3695         char errbuf[1024];
3696         zio_cksum_t zcksum = { 0 };
3697         uint64_t featureflags;
3698         int hdrtype;
3699
3700         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3701             "cannot receive"));
3702
3703         if (flags->isprefix &&
3704             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3705                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3706                     "(%s) does not exist"), tosnap);
3707                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3708         }
3709         if (originsnap &&
3710             !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) {
3711                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs "
3712                     "(%s) does not exist"), originsnap);
3713                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3714         }
3715
3716         /* read in the BEGIN record */
3717         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3718             &zcksum)))
3719                 return (err);
3720
3721         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3722                 /* It's the double end record at the end of a package */
3723                 return (ENODATA);
3724         }
3725
3726         /* the kernel needs the non-byteswapped begin record */
3727         drr_noswap = drr;
3728
3729         flags->byteswap = B_FALSE;
3730         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3731                 /*
3732                  * We computed the checksum in the wrong byteorder in
3733                  * recv_read() above; do it again correctly.
3734                  */
3735                 bzero(&zcksum, sizeof (zio_cksum_t));
3736                 (void) fletcher_4_incremental_byteswap(&drr,
3737                     sizeof (drr), &zcksum);
3738                 flags->byteswap = B_TRUE;
3739
3740                 drr.drr_type = BSWAP_32(drr.drr_type);
3741                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3742                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3743                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3744                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3745                 drrb->drr_type = BSWAP_32(drrb->drr_type);
3746                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3747                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3748                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3749         }
3750
3751         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3752                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3753                     "stream (bad magic number)"));
3754                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3755         }
3756
3757         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3758         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3759
3760         if (!DMU_STREAM_SUPPORTED(featureflags) ||
3761             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3762                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3763                     "stream has unsupported feature, feature flags = %lx"),
3764                     featureflags);
3765                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3766         }
3767
3768         if (strchr(drrb->drr_toname, '@') == NULL) {
3769                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3770                     "stream (bad snapshot name)"));
3771                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3772         }
3773
3774         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3775                 char nonpackage_sendfs[ZFS_MAX_DATASET_NAME_LEN];
3776                 if (sendfs == NULL) {
3777                         /*
3778                          * We were not called from zfs_receive_package(). Get
3779                          * the fs specified by 'zfs send'.
3780                          */
3781                         char *cp;
3782                         (void) strlcpy(nonpackage_sendfs,
3783                             drr.drr_u.drr_begin.drr_toname,
3784                             sizeof (nonpackage_sendfs));
3785                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3786                                 *cp = '\0';
3787                         sendfs = nonpackage_sendfs;
3788                         VERIFY(finalsnap == NULL);
3789                 }
3790                 return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags,
3791                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs,
3792                     cleanup_fd, action_handlep, finalsnap));
3793         } else {
3794                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3795                     DMU_COMPOUNDSTREAM);
3796                 return (zfs_receive_package(hdl, infd, tosnap, flags, &drr,
3797                     &zcksum, top_zfs, cleanup_fd, action_handlep));
3798         }
3799 }
3800
3801 /*
3802  * Restores a backup of tosnap from the file descriptor specified by infd.
3803  * Return 0 on total success, -2 if some things couldn't be
3804  * destroyed/renamed/promoted, -1 if some things couldn't be received.
3805  * (-1 will override -2, if -1 and the resumable flag was specified the
3806  * transfer can be resumed if the sending side supports it).
3807  */
3808 int
3809 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props,
3810     recvflags_t *flags, int infd, avl_tree_t *stream_avl)
3811 {
3812         char *top_zfs = NULL;
3813         int err;
3814         int cleanup_fd;
3815         uint64_t action_handle = 0;
3816         char *originsnap = NULL;
3817         if (props) {
3818                 err = nvlist_lookup_string(props, "origin", &originsnap);
3819                 if (err && err != ENOENT)
3820                         return (err);
3821         }
3822
3823         cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3824         VERIFY(cleanup_fd >= 0);
3825
3826         err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL,
3827             stream_avl, &top_zfs, cleanup_fd, &action_handle, NULL);
3828
3829         VERIFY(0 == close(cleanup_fd));
3830
3831         if (err == 0 && !flags->nomount && top_zfs) {
3832                 zfs_handle_t *zhp;
3833                 prop_changelist_t *clp;
3834
3835                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3836                 if (zhp != NULL) {
3837                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3838                             CL_GATHER_MOUNT_ALWAYS, 0);
3839                         zfs_close(zhp);
3840                         if (clp != NULL) {
3841                                 /* mount and share received datasets */
3842                                 err = changelist_postfix(clp);
3843                                 changelist_free(clp);
3844                         }
3845                 }
3846                 if (zhp == NULL || clp == NULL || err)
3847                         err = -1;
3848         }
3849         if (top_zfs)
3850                 free(top_zfs);
3851
3852         return (err);
3853 }