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) 2013 by Delphix. All rights reserved.
  25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2012 Pawel Jakub Dawidek <pawel@dawidek.net>.
  27  * All rights reserved.
  28  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  29  */
  30
  31 #include <assert.h>
  32 #include <ctype.h>
  33 #include <errno.h>
  34 #include <libintl.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <strings.h>
  38 #include <unistd.h>
  39 #include <stddef.h>
  40 #include <fcntl.h>
  41 #include <sys/param.h>
  42 #include <sys/mount.h>
  43 #include <pthread.h>
  44 #include <umem.h>
  45 #include <time.h>
  46
  47 #include <libzfs.h>
  48
  49 #include "zfs_namecheck.h"
  50 #include "zfs_prop.h"
  51 #include "zfs_fletcher.h"
  52 #include "libzfs_impl.h"
  53 #include <sha2.h>
  54 #include <sys/zio_checksum.h>
  55 #include <sys/ddt.h>
  56
  57 #ifdef __FreeBSD__
  58 extern int zfs_ioctl_version;
  59 #endif
  60
  61 /* in libzfs_dataset.c */
  62 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  63 /* We need to use something for ENODATA. */
  64 #define ENODATA EIDRM
  65
  66 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t *,
  67     int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *);
  68
  69 static const zio_cksum_t zero_cksum = { 0 };
  70
  71 typedef struct dedup_arg {
  72         int     inputfd;
  73         int     outputfd;
  74         libzfs_handle_t  *dedup_hdl;
  75 } dedup_arg_t;
  76
  77 typedef struct progress_arg {
  78         zfs_handle_t *pa_zhp;
  79         int pa_fd;
  80         boolean_t pa_parsable;
  81 } progress_arg_t;
  82
  83 typedef struct dataref {
  84         uint64_t ref_guid;
  85         uint64_t ref_object;
  86         uint64_t ref_offset;
  87 } dataref_t;
  88
  89 typedef struct dedup_entry {
  90         struct dedup_entry      *dde_next;
  91         zio_cksum_t dde_chksum;
  92         uint64_t dde_prop;
  93         dataref_t dde_ref;
  94 } dedup_entry_t;
  95
  96 #define MAX_DDT_PHYSMEM_PERCENT         20
  97 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
  98
  99 typedef struct dedup_table {
 100         dedup_entry_t   **dedup_hash_array;
 101         umem_cache_t    *ddecache;
 102         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
 103         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
 104         uint64_t        ddt_count;
 105         int             numhashbits;
 106         boolean_t       ddt_full;
 107 } dedup_table_t;
 108
 109 static int
 110 high_order_bit(uint64_t n)
 111 {
 112         int count;
 113
 114         for (count = 0; n != 0; count++)
 115                 n >>= 1;
 116         return (count);
 117 }
 118
 119 static size_t
 120 ssread(void *buf, size_t len, FILE *stream)
 121 {
 122         size_t outlen;
 123
 124         if ((outlen = fread(buf, len, 1, stream)) == 0)
 125                 return (0);
 126
 127         return (outlen);
 128 }
 129
 130 static void
 131 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 132     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 133 {
 134         dedup_entry_t   *dde;
 135
 136         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 137                 if (ddt->ddt_full == B_FALSE) {
 138                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 139                             "Dedup table full.  Deduplication will continue "
 140                             "with existing table entries"));
 141                         ddt->ddt_full = B_TRUE;
 142                 }
 143                 return;
 144         }
 145
 146         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 147             != NULL) {
 148                 assert(*ddepp == NULL);
 149                 dde->dde_next = NULL;
 150                 dde->dde_chksum = *cs;
 151                 dde->dde_prop = prop;
 152                 dde->dde_ref = *dr;
 153                 *ddepp = dde;
 154                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 155                 ddt->ddt_count++;
 156         }
 157 }
 158
 159 /*
 160  * Using the specified dedup table, do a lookup for an entry with
 161  * the checksum cs.  If found, return the block's reference info
 162  * in *dr. Otherwise, insert a new entry in the dedup table, using
 163  * the reference information specified by *dr.
 164  *
 165  * return value:  true - entry was found
 166  *                false - entry was not found
 167  */
 168 static boolean_t
 169 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 170     uint64_t prop, dataref_t *dr)
 171 {
 172         uint32_t hashcode;
 173         dedup_entry_t **ddepp;
 174
 175         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 176
 177         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 178             ddepp = &((*ddepp)->dde_next)) {
 179                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 180                     (*ddepp)->dde_prop == prop) {
 181                         *dr = (*ddepp)->dde_ref;
 182                         return (B_TRUE);
 183                 }
 184         }
 185         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 186         return (B_FALSE);
 187 }
 188
 189 static int
 190 cksum_and_write(const void *buf, uint64_t len, zio_cksum_t *zc, int outfd)
 191 {
 192         fletcher_4_incremental_native(buf, len, zc);
 193         return (write(outfd, buf, len));
 194 }
 195
 196 /*
 197  * This function is started in a separate thread when the dedup option
 198  * has been requested.  The main send thread determines the list of
 199  * snapshots to be included in the send stream and makes the ioctl calls
 200  * for each one.  But instead of having the ioctl send the output to the
 201  * the output fd specified by the caller of zfs_send()), the
 202  * ioctl is told to direct the output to a pipe, which is read by the
 203  * alternate thread running THIS function.  This function does the
 204  * dedup'ing by:
 205  *  1. building a dedup table (the DDT)
 206  *  2. doing checksums on each data block and inserting a record in the DDT
 207  *  3. looking for matching checksums, and
 208  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 209  *      a duplicate block is found.
 210  * The output of this function then goes to the output fd requested
 211  * by the caller of zfs_send().
 212  */
 213 static void *
 214 cksummer(void *arg)
 215 {
 216         dedup_arg_t *dda = arg;
 217         char *buf = malloc(1<<20);
 218         dmu_replay_record_t thedrr;
 219         dmu_replay_record_t *drr = &thedrr;
 220         struct drr_begin *drrb = &thedrr.drr_u.drr_begin;
 221         struct drr_end *drre = &thedrr.drr_u.drr_end;
 222         struct drr_object *drro = &thedrr.drr_u.drr_object;
 223         struct drr_write *drrw = &thedrr.drr_u.drr_write;
 224         struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
 225         FILE *ofp;
 226         int outfd;
 227         dmu_replay_record_t wbr_drr = {0};
 228         struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref;
 229         dedup_table_t ddt;
 230         zio_cksum_t stream_cksum;
 231         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 232         uint64_t numbuckets;
 233
 234         ddt.max_ddt_size =
 235             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT)/100,
 236             SMALLEST_POSSIBLE_MAX_DDT_MB<<20);
 237
 238         numbuckets = ddt.max_ddt_size/(sizeof (dedup_entry_t));
 239
 240         /*
 241          * numbuckets must be a power of 2.  Increase number to
 242          * a power of 2 if necessary.
 243          */
 244         if (!ISP2(numbuckets))
 245                 numbuckets = 1 << high_order_bit(numbuckets);
 246
 247         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 248         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 249             NULL, NULL, NULL, NULL, NULL, 0);
 250         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 251         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 252         ddt.ddt_full = B_FALSE;
 253
 254         /* Initialize the write-by-reference block. */
 255         wbr_drr.drr_type = DRR_WRITE_BYREF;
 256         wbr_drr.drr_payloadlen = 0;
 257
 258         outfd = dda->outputfd;
 259         ofp = fdopen(dda->inputfd, "r");
 260         while (ssread(drr, sizeof (dmu_replay_record_t), ofp) != 0) {
 261
 262                 switch (drr->drr_type) {
 263                 case DRR_BEGIN:
 264                 {
 265                         int     fflags;
 266                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 267
 268                         /* set the DEDUP feature flag for this stream */
 269                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 270                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 271                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 272                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 273
 274                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 275                             &stream_cksum, outfd) == -1)
 276                                 goto out;
 277                         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 278                             DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
 279                                 int sz = drr->drr_payloadlen;
 280
 281                                 if (sz > 1<<20) {
 282                                         free(buf);
 283                                         buf = malloc(sz);
 284                                 }
 285                                 (void) ssread(buf, sz, ofp);
 286                                 if (ferror(stdin))
 287                                         perror("fread");
 288                                 if (cksum_and_write(buf, sz, &stream_cksum,
 289                                     outfd) == -1)
 290                                         goto out;
 291                         }
 292                         break;
 293                 }
 294
 295                 case DRR_END:
 296                 {
 297                         /* use the recalculated checksum */
 298                         ZIO_SET_CHECKSUM(&drre->drr_checksum,
 299                             stream_cksum.zc_word[0], stream_cksum.zc_word[1],
 300                             stream_cksum.zc_word[2], stream_cksum.zc_word[3]);
 301                         if ((write(outfd, drr,
 302                             sizeof (dmu_replay_record_t))) == -1)
 303                                 goto out;
 304                         break;
 305                 }
 306
 307                 case DRR_OBJECT:
 308                 {
 309                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 310                             &stream_cksum, outfd) == -1)
 311                                 goto out;
 312                         if (drro->drr_bonuslen > 0) {
 313                                 (void) ssread(buf,
 314                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 315                                     ofp);
 316                                 if (cksum_and_write(buf,
 317                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 318                                     &stream_cksum, outfd) == -1)
 319                                         goto out;
 320                         }
 321                         break;
 322                 }
 323
 324                 case DRR_SPILL:
 325                 {
 326                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 327                             &stream_cksum, outfd) == -1)
 328                                 goto out;
 329                         (void) ssread(buf, drrs->drr_length, ofp);
 330                         if (cksum_and_write(buf, drrs->drr_length,
 331                             &stream_cksum, outfd) == -1)
 332                                 goto out;
 333                         break;
 334                 }
 335
 336                 case DRR_FREEOBJECTS:
 337                 {
 338                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 339                             &stream_cksum, outfd) == -1)
 340                                 goto out;
 341                         break;
 342                 }
 343
 344                 case DRR_WRITE:
 345                 {
 346                         dataref_t       dataref;
 347
 348                         (void) ssread(buf, drrw->drr_length, ofp);
 349
 350                         /*
 351                          * Use the existing checksum if it's dedup-capable,
 352                          * else calculate a SHA256 checksum for it.
 353                          */
 354
 355                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 356                             zero_cksum) ||
 357                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 358                                 SHA256_CTX      ctx;
 359                                 zio_cksum_t     tmpsha256;
 360
 361                                 SHA256Init(&ctx);
 362                                 SHA256Update(&ctx, buf, drrw->drr_length);
 363                                 SHA256Final(&tmpsha256, &ctx);
 364                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 365                                     BE_64(tmpsha256.zc_word[0]);
 366                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 367                                     BE_64(tmpsha256.zc_word[1]);
 368                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 369                                     BE_64(tmpsha256.zc_word[2]);
 370                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 371                                     BE_64(tmpsha256.zc_word[3]);
 372                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 373                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 374                         }
 375
 376                         dataref.ref_guid = drrw->drr_toguid;
 377                         dataref.ref_object = drrw->drr_object;
 378                         dataref.ref_offset = drrw->drr_offset;
 379
 380                         if (ddt_update(dda->dedup_hdl, &ddt,
 381                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 382                             &dataref)) {
 383                                 /* block already present in stream */
 384                                 wbr_drrr->drr_object = drrw->drr_object;
 385                                 wbr_drrr->drr_offset = drrw->drr_offset;
 386                                 wbr_drrr->drr_length = drrw->drr_length;
 387                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 388                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 389                                 wbr_drrr->drr_refobject =
 390                                     dataref.ref_object;
 391                                 wbr_drrr->drr_refoffset =
 392                                     dataref.ref_offset;
 393
 394                                 wbr_drrr->drr_checksumtype =
 395                                     drrw->drr_checksumtype;
 396                                 wbr_drrr->drr_checksumflags =
 397                                     drrw->drr_checksumtype;
 398                                 wbr_drrr->drr_key.ddk_cksum =
 399                                     drrw->drr_key.ddk_cksum;
 400                                 wbr_drrr->drr_key.ddk_prop =
 401                                     drrw->drr_key.ddk_prop;
 402
 403                                 if (cksum_and_write(&wbr_drr,
 404                                     sizeof (dmu_replay_record_t), &stream_cksum,
 405                                     outfd) == -1)
 406                                         goto out;
 407                         } else {
 408                                 /* block not previously seen */
 409                                 if (cksum_and_write(drr,
 410                                     sizeof (dmu_replay_record_t), &stream_cksum,
 411                                     outfd) == -1)
 412                                         goto out;
 413                                 if (cksum_and_write(buf,
 414                                     drrw->drr_length,
 415                                     &stream_cksum, outfd) == -1)
 416                                         goto out;
 417                         }
 418                         break;
 419                 }
 420
 421                 case DRR_FREE:
 422                 {
 423                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 424                             &stream_cksum, outfd) == -1)
 425                                 goto out;
 426                         break;
 427                 }
 428
 429                 default:
 430                         (void) printf("INVALID record type 0x%x\n",
 431                             drr->drr_type);
 432                         /* should never happen, so assert */
 433                         assert(B_FALSE);
 434                 }
 435         }
 436 out:
 437         umem_cache_destroy(ddt.ddecache);
 438         free(ddt.dedup_hash_array);
 439         free(buf);
 440         (void) fclose(ofp);
 441
 442         return (NULL);
 443 }
 444
 445 /*
 446  * Routines for dealing with the AVL tree of fs-nvlists
 447  */
 448 typedef struct fsavl_node {
 449         avl_node_t fn_node;
 450         nvlist_t *fn_nvfs;
 451         char *fn_snapname;
 452         uint64_t fn_guid;
 453 } fsavl_node_t;
 454
 455 static int
 456 fsavl_compare(const void *arg1, const void *arg2)
 457 {
 458         const fsavl_node_t *fn1 = arg1;
 459         const fsavl_node_t *fn2 = arg2;
 460
 461         if (fn1->fn_guid > fn2->fn_guid)
 462                 return (+1);
 463         else if (fn1->fn_guid < fn2->fn_guid)
 464                 return (-1);
 465         else
 466                 return (0);
 467 }
 468
 469 /*
 470  * Given the GUID of a snapshot, find its containing filesystem and
 471  * (optionally) name.
 472  */
 473 static nvlist_t *
 474 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 475 {
 476         fsavl_node_t fn_find;
 477         fsavl_node_t *fn;
 478
 479         fn_find.fn_guid = snapguid;
 480
 481         fn = avl_find(avl, &fn_find, NULL);
 482         if (fn) {
 483                 if (snapname)
 484                         *snapname = fn->fn_snapname;
 485                 return (fn->fn_nvfs);
 486         }
 487         return (NULL);
 488 }
 489
 490 static void
 491 fsavl_destroy(avl_tree_t *avl)
 492 {
 493         fsavl_node_t *fn;
 494         void *cookie;
 495
 496         if (avl == NULL)
 497                 return;
 498
 499         cookie = NULL;
 500         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 501                 free(fn);
 502         avl_destroy(avl);
 503         free(avl);
 504 }
 505
 506 /*
 507  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 508  */
 509 static avl_tree_t *
 510 fsavl_create(nvlist_t *fss)
 511 {
 512         avl_tree_t *fsavl;
 513         nvpair_t *fselem = NULL;
 514
 515         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 516                 return (NULL);
 517
 518         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 519             offsetof(fsavl_node_t, fn_node));
 520
 521         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 522                 nvlist_t *nvfs, *snaps;
 523                 nvpair_t *snapelem = NULL;
 524
 525                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 526                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 527
 528                 while ((snapelem =
 529                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 530                         fsavl_node_t *fn;
 531                         uint64_t guid;
 532
 533                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 534                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 535                                 fsavl_destroy(fsavl);
 536                                 return (NULL);
 537                         }
 538                         fn->fn_nvfs = nvfs;
 539                         fn->fn_snapname = nvpair_name(snapelem);
 540                         fn->fn_guid = guid;
 541
 542                         /*
 543                          * Note: if there are multiple snaps with the
 544                          * same GUID, we ignore all but one.
 545                          */
 546                         if (avl_find(fsavl, fn, NULL) == NULL)
 547                                 avl_add(fsavl, fn);
 548                         else
 549                                 free(fn);
 550                 }
 551         }
 552
 553         return (fsavl);
 554 }
 555
 556 /*
 557  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 558  */
 559 typedef struct send_data {
 560         uint64_t parent_fromsnap_guid;
 561         nvlist_t *parent_snaps;
 562         nvlist_t *fss;
 563         nvlist_t *snapprops;
 564         const char *fromsnap;
 565         const char *tosnap;
 566         boolean_t recursive;
 567
 568         /*
 569          * The header nvlist is of the following format:
 570          * {
 571          *   "tosnap" -> string
 572          *   "fromsnap" -> string (if incremental)
 573          *   "fss" -> {
 574          *      id -> {
 575          *
 576          *       "name" -> string (full name; for debugging)
 577          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 578          *
 579          *       "props" -> { name -> value (only if set here) }
 580          *       "snaps" -> { name (lastname) -> number (guid) }
 581          *       "snapprops" -> { name (lastname) -> { name -> value } }
 582          *
 583          *       "origin" -> number (guid) (if clone)
 584          *       "sent" -> boolean (not on-disk)
 585          *      }
 586          *   }
 587          * }
 588          *
 589          */
 590 } send_data_t;
 591
 592 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 593
 594 static int
 595 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 596 {
 597         send_data_t *sd = arg;
 598         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 599         char *snapname;
 600         nvlist_t *nv;
 601
 602         snapname = strrchr(zhp->zfs_name, '@')+1;
 603
 604         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 605         /*
 606          * NB: if there is no fromsnap here (it's a newly created fs in
 607          * an incremental replication), we will substitute the tosnap.
 608          */
 609         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 610             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 611             strcmp(snapname, sd->tosnap) == 0)) {
 612                 sd->parent_fromsnap_guid = guid;
 613         }
 614
 615         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 616         send_iterate_prop(zhp, nv);
 617         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 618         nvlist_free(nv);
 619
 620         zfs_close(zhp);
 621         return (0);
 622 }
 623
 624 static void
 625 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 626 {
 627         nvpair_t *elem = NULL;
 628
 629         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 630                 char *propname = nvpair_name(elem);
 631                 zfs_prop_t prop = zfs_name_to_prop(propname);
 632                 nvlist_t *propnv;
 633
 634                 if (!zfs_prop_user(propname)) {
 635                         /*
 636                          * Realistically, this should never happen.  However,
 637                          * we want the ability to add DSL properties without
 638                          * needing to make incompatible version changes.  We
 639                          * need to ignore unknown properties to allow older
 640                          * software to still send datasets containing these
 641                          * properties, with the unknown properties elided.
 642                          */
 643                         if (prop == ZPROP_INVAL)
 644                                 continue;
 645
 646                         if (zfs_prop_readonly(prop))
 647                                 continue;
 648                 }
 649
 650                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 651                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 652                     prop == ZFS_PROP_REFQUOTA ||
 653                     prop == ZFS_PROP_REFRESERVATION) {
 654                         char *source;
 655                         uint64_t value;
 656                         verify(nvlist_lookup_uint64(propnv,
 657                             ZPROP_VALUE, &value) == 0);
 658                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 659                                 continue;
 660                         /*
 661                          * May have no source before SPA_VERSION_RECVD_PROPS,
 662                          * but is still modifiable.
 663                          */
 664                         if (nvlist_lookup_string(propnv,
 665                             ZPROP_SOURCE, &source) == 0) {
 666                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 667                                     (strcmp(source,
 668                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 669                                         continue;
 670                         }
 671                 } else {
 672                         char *source;
 673                         if (nvlist_lookup_string(propnv,
 674                             ZPROP_SOURCE, &source) != 0)
 675                                 continue;
 676                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 677                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 678                                 continue;
 679                 }
 680
 681                 if (zfs_prop_user(propname) ||
 682                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 683                         char *value;
 684                         verify(nvlist_lookup_string(propnv,
 685                             ZPROP_VALUE, &value) == 0);
 686                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 687                 } else {
 688                         uint64_t value;
 689                         verify(nvlist_lookup_uint64(propnv,
 690                             ZPROP_VALUE, &value) == 0);
 691                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 692                 }
 693         }
 694 }
 695
 696 /*
 697  * recursively generate nvlists describing datasets.  See comment
 698  * for the data structure send_data_t above for description of contents
 699  * of the nvlist.
 700  */
 701 static int
 702 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 703 {
 704         send_data_t *sd = arg;
 705         nvlist_t *nvfs, *nv;
 706         int rv = 0;
 707         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 708         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 709         char guidstring[64];
 710
 711         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 712         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 713         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 714             sd->parent_fromsnap_guid));
 715
 716         if (zhp->zfs_dmustats.dds_origin[0]) {
 717                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 718                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 719                 if (origin == NULL)
 720                         return (-1);
 721                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 722                     origin->zfs_dmustats.dds_guid));
 723         }
 724
 725         /* iterate over props */
 726         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 727         send_iterate_prop(zhp, nv);
 728         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 729         nvlist_free(nv);
 730
 731         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 732         sd->parent_fromsnap_guid = 0;
 733         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 734         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 735         (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd);
 736         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 737         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 738         nvlist_free(sd->parent_snaps);
 739         nvlist_free(sd->snapprops);
 740
 741         /* add this fs to nvlist */
 742         (void) snprintf(guidstring, sizeof (guidstring),
 743             "0x%llx", (longlong_t)guid);
 744         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 745         nvlist_free(nvfs);
 746
 747         /* iterate over children */
 748         if (sd->recursive)
 749                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 750
 751         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 752
 753         zfs_close(zhp);
 754         return (rv);
 755 }
 756
 757 static int
 758 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 759     const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
 760 {
 761         zfs_handle_t *zhp;
 762         send_data_t sd = { 0 };
 763         int error;
 764
 765         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 766         if (zhp == NULL)
 767                 return (EZFS_BADTYPE);
 768
 769         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 770         sd.fromsnap = fromsnap;
 771         sd.tosnap = tosnap;
 772         sd.recursive = recursive;
 773
 774         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 775                 nvlist_free(sd.fss);
 776                 if (avlp != NULL)
 777                         *avlp = NULL;
 778                 *nvlp = NULL;
 779                 return (error);
 780         }
 781
 782         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 783                 nvlist_free(sd.fss);
 784                 *nvlp = NULL;
 785                 return (EZFS_NOMEM);
 786         }
 787
 788         *nvlp = sd.fss;
 789         return (0);
 790 }
 791
 792 /*
 793  * Routines specific to "zfs send"
 794  */
 795 typedef struct send_dump_data {
 796         /* these are all just the short snapname (the part after the @) */
 797         const char *fromsnap;
 798         const char *tosnap;
 799         char prevsnap[ZFS_MAXNAMELEN];
 800         uint64_t prevsnap_obj;
 801         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 802         boolean_t verbose, dryrun, parsable, progress;
 803         int outfd;
 804         boolean_t err;
 805         nvlist_t *fss;
 806         nvlist_t *snapholds;
 807         avl_tree_t *fsavl;
 808         snapfilter_cb_t *filter_cb;
 809         void *filter_cb_arg;
 810         nvlist_t *debugnv;
 811         char holdtag[ZFS_MAXNAMELEN];
 812         int cleanup_fd;
 813         uint64_t size;
 814 } send_dump_data_t;
 815
 816 static int
 817 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
 818     boolean_t fromorigin, uint64_t *sizep)
 819 {
 820         zfs_cmd_t zc = { 0 };
 821         libzfs_handle_t *hdl = zhp->zfs_hdl;
 822
 823         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 824         assert(fromsnap_obj == 0 || !fromorigin);
 825
 826         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 827         zc.zc_obj = fromorigin;
 828         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 829         zc.zc_fromobj = fromsnap_obj;
 830         zc.zc_guid = 1;  /* estimate flag */
 831
 832         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 833                 char errbuf[1024];
 834                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 835                     "warning: cannot estimate space for '%s'"), zhp->zfs_name);
 836
 837                 switch (errno) {
 838                 case EXDEV:
 839                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 840                             "not an earlier snapshot from the same fs"));
 841                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 842
 843                 case ENOENT:
 844                         if (zfs_dataset_exists(hdl, zc.zc_name,
 845                             ZFS_TYPE_SNAPSHOT)) {
 846                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 847                                     "incremental source (@%s) does not exist"),
 848                                     zc.zc_value);
 849                         }
 850                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 851
 852                 case EDQUOT:
 853                 case EFBIG:
 854                 case EIO:
 855                 case ENOLINK:
 856                 case ENOSPC:
 857                 case ENXIO:
 858                 case EPIPE:
 859                 case ERANGE:
 860                 case EFAULT:
 861                 case EROFS:
 862                         zfs_error_aux(hdl, strerror(errno));
 863                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 864
 865                 default:
 866                         return (zfs_standard_error(hdl, errno, errbuf));
 867                 }
 868         }
 869
 870         *sizep = zc.zc_objset_type;
 871
 872         return (0);
 873 }
 874
 875 /*
 876  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 877  * NULL) to the file descriptor specified by outfd.
 878  */
 879 static int
 880 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
 881     boolean_t fromorigin, int outfd, nvlist_t *debugnv)
 882 {
 883         zfs_cmd_t zc = { 0 };
 884         libzfs_handle_t *hdl = zhp->zfs_hdl;
 885         nvlist_t *thisdbg;
 886
 887         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 888         assert(fromsnap_obj == 0 || !fromorigin);
 889
 890         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 891         zc.zc_cookie = outfd;
 892         zc.zc_obj = fromorigin;
 893         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 894         zc.zc_fromobj = fromsnap_obj;
 895
 896         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 897         if (fromsnap && fromsnap[0] != '\0') {
 898                 VERIFY(0 == nvlist_add_string(thisdbg,
 899                     "fromsnap", fromsnap));
 900         }
 901
 902         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 903                 char errbuf[1024];
 904                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 905                     "warning: cannot send '%s'"), zhp->zfs_name);
 906
 907                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
 908                 if (debugnv) {
 909                         VERIFY(0 == nvlist_add_nvlist(debugnv,
 910                             zhp->zfs_name, thisdbg));
 911                 }
 912                 nvlist_free(thisdbg);
 913
 914                 switch (errno) {
 915                 case EXDEV:
 916                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 917                             "not an earlier snapshot from the same fs"));
 918                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 919
 920                 case ENOENT:
 921                         if (zfs_dataset_exists(hdl, zc.zc_name,
 922                             ZFS_TYPE_SNAPSHOT)) {
 923                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 924                                     "incremental source (@%s) does not exist"),
 925                                     zc.zc_value);
 926                         }
 927                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 928
 929                 case EDQUOT:
 930                 case EFBIG:
 931                 case EIO:
 932                 case ENOLINK:
 933                 case ENOSPC:
 934 #ifdef sun
 935                 case ENOSTR:
 936 #endif
 937                 case ENXIO:
 938                 case EPIPE:
 939                 case ERANGE:
 940                 case EFAULT:
 941                 case EROFS:
 942                         zfs_error_aux(hdl, strerror(errno));
 943                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 944
 945                 default:
 946                         return (zfs_standard_error(hdl, errno, errbuf));
 947                 }
 948         }
 949
 950         if (debugnv)
 951                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
 952         nvlist_free(thisdbg);
 953
 954         return (0);
 955 }
 956
 957 static void
 958 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
 959 {
 960         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 961
 962         /*
 963          * zfs_send() only sets snapholds for sends that need them,
 964          * e.g. replication and doall.
 965          */
 966         if (sdd->snapholds == NULL)
 967                 return;
 968
 969         fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
 970 }
 971
 972 static void *
 973 send_progress_thread(void *arg)
 974 {
 975         progress_arg_t *pa = arg;
 976
 977         zfs_cmd_t zc = { 0 };
 978         zfs_handle_t *zhp = pa->pa_zhp;
 979         libzfs_handle_t *hdl = zhp->zfs_hdl;
 980         unsigned long long bytes;
 981         char buf[16];
 982
 983         time_t t;
 984         struct tm *tm;
 985
 986         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 987         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 988
 989         if (!pa->pa_parsable)
 990                 (void) fprintf(stderr, "TIME        SENT   SNAPSHOT\n");
 991
 992         /*
 993          * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
 994          */
 995         for (;;) {
 996                 (void) sleep(1);
 997
 998                 zc.zc_cookie = pa->pa_fd;
 999                 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1000                         return ((void *)-1);
1001
1002                 (void) time(&t);
1003                 tm = localtime(&t);
1004                 bytes = zc.zc_cookie;
1005
1006                 if (pa->pa_parsable) {
1007                         (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1008                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1009                             bytes, zhp->zfs_name);
1010                 } else {
1011                         zfs_nicenum(bytes, buf, sizeof (buf));
1012                         (void) fprintf(stderr, "%02d:%02d:%02d   %5s   %s\n",
1013                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1014                             buf, zhp->zfs_name);
1015                 }
1016         }
1017 }
1018
1019 static int
1020 dump_snapshot(zfs_handle_t *zhp, void *arg)
1021 {
1022         send_dump_data_t *sdd = arg;
1023         progress_arg_t pa = { 0 };
1024         pthread_t tid;
1025         char *thissnap;
1026         int err;
1027         boolean_t isfromsnap, istosnap, fromorigin;
1028         boolean_t exclude = B_FALSE;
1029
1030         err = 0;
1031         thissnap = strchr(zhp->zfs_name, '@') + 1;
1032         isfromsnap = (sdd->fromsnap != NULL &&
1033             strcmp(sdd->fromsnap, thissnap) == 0);
1034
1035         if (!sdd->seenfrom && isfromsnap) {
1036                 gather_holds(zhp, sdd);
1037                 sdd->seenfrom = B_TRUE;
1038                 (void) strcpy(sdd->prevsnap, thissnap);
1039                 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1040                 zfs_close(zhp);
1041                 return (0);
1042         }
1043
1044         if (sdd->seento || !sdd->seenfrom) {
1045                 zfs_close(zhp);
1046                 return (0);
1047         }
1048
1049         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1050         if (istosnap)
1051                 sdd->seento = B_TRUE;
1052
1053         if (!sdd->doall && !isfromsnap && !istosnap) {
1054                 if (sdd->replicate) {
1055                         char *snapname;
1056                         nvlist_t *snapprops;
1057                         /*
1058                          * Filter out all intermediate snapshots except origin
1059                          * snapshots needed to replicate clones.
1060                          */
1061                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1062                             zhp->zfs_dmustats.dds_guid, &snapname);
1063
1064                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1065                             "snapprops", &snapprops));
1066                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1067                             thissnap, &snapprops));
1068                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
1069                 } else {
1070                         exclude = B_TRUE;
1071                 }
1072         }
1073
1074         /*
1075          * If a filter function exists, call it to determine whether
1076          * this snapshot will be sent.
1077          */
1078         if (exclude || (sdd->filter_cb != NULL &&
1079             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1080                 /*
1081                  * This snapshot is filtered out.  Don't send it, and don't
1082                  * set prevsnap_obj, so it will be as if this snapshot didn't
1083                  * exist, and the next accepted snapshot will be sent as
1084                  * an incremental from the last accepted one, or as the
1085                  * first (and full) snapshot in the case of a replication,
1086                  * non-incremental send.
1087                  */
1088                 zfs_close(zhp);
1089                 return (0);
1090         }
1091
1092         gather_holds(zhp, sdd);
1093         fromorigin = sdd->prevsnap[0] == '\0' &&
1094             (sdd->fromorigin || sdd->replicate);
1095
1096         if (sdd->verbose) {
1097                 uint64_t size;
1098                 err = estimate_ioctl(zhp, sdd->prevsnap_obj,
1099                     fromorigin, &size);
1100
1101                 if (sdd->parsable) {
1102                         if (sdd->prevsnap[0] != '\0') {
1103                                 (void) fprintf(stderr, "incremental\t%s\t%s",
1104                                     sdd->prevsnap, zhp->zfs_name);
1105                         } else {
1106                                 (void) fprintf(stderr, "full\t%s",
1107                                     zhp->zfs_name);
1108                         }
1109                 } else {
1110                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1111                             "send from @%s to %s"),
1112                             sdd->prevsnap, zhp->zfs_name);
1113                 }
1114                 if (err == 0) {
1115                         if (sdd->parsable) {
1116                                 (void) fprintf(stderr, "\t%llu\n",
1117                                     (longlong_t)size);
1118                         } else {
1119                                 char buf[16];
1120                                 zfs_nicenum(size, buf, sizeof (buf));
1121                                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1122                                     " estimated size is %s\n"), buf);
1123                         }
1124                         sdd->size += size;
1125                 } else {
1126                         (void) fprintf(stderr, "\n");
1127                 }
1128         }
1129
1130         if (!sdd->dryrun) {
1131                 /*
1132                  * If progress reporting is requested, spawn a new thread to
1133                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1134                  */
1135                 if (sdd->progress) {
1136                         pa.pa_zhp = zhp;
1137                         pa.pa_fd = sdd->outfd;
1138                         pa.pa_parsable = sdd->parsable;
1139
1140                         if (err = pthread_create(&tid, NULL,
1141                             send_progress_thread, &pa)) {
1142                                 zfs_close(zhp);
1143                                 return (err);
1144                         }
1145                 }
1146
1147                 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1148                     fromorigin, sdd->outfd, sdd->debugnv);
1149
1150                 if (sdd->progress) {
1151                         (void) pthread_cancel(tid);
1152                         (void) pthread_join(tid, NULL);
1153                 }
1154         }
1155
1156         (void) strcpy(sdd->prevsnap, thissnap);
1157         sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1158         zfs_close(zhp);
1159         return (err);
1160 }
1161
1162 static int
1163 dump_filesystem(zfs_handle_t *zhp, void *arg)
1164 {
1165         int rv = 0;
1166         send_dump_data_t *sdd = arg;
1167         boolean_t missingfrom = B_FALSE;
1168         zfs_cmd_t zc = { 0 };
1169
1170         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1171             zhp->zfs_name, sdd->tosnap);
1172         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1173                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1174                     "WARNING: could not send %s@%s: does not exist\n"),
1175                     zhp->zfs_name, sdd->tosnap);
1176                 sdd->err = B_TRUE;
1177                 return (0);
1178         }
1179
1180         if (sdd->replicate && sdd->fromsnap) {
1181                 /*
1182                  * If this fs does not have fromsnap, and we're doing
1183                  * recursive, we need to send a full stream from the
1184                  * beginning (or an incremental from the origin if this
1185                  * is a clone).  If we're doing non-recursive, then let
1186                  * them get the error.
1187                  */
1188                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1189                     zhp->zfs_name, sdd->fromsnap);
1190                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1191                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1192                         missingfrom = B_TRUE;
1193                 }
1194         }
1195
1196         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1197         sdd->prevsnap_obj = 0;
1198         if (sdd->fromsnap == NULL || missingfrom)
1199                 sdd->seenfrom = B_TRUE;
1200
1201         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1202         if (!sdd->seenfrom) {
1203                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1204                     "WARNING: could not send %s@%s:\n"
1205                     "incremental source (%s@%s) does not exist\n"),
1206                     zhp->zfs_name, sdd->tosnap,
1207                     zhp->zfs_name, sdd->fromsnap);
1208                 sdd->err = B_TRUE;
1209         } else if (!sdd->seento) {
1210                 if (sdd->fromsnap) {
1211                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1212                             "WARNING: could not send %s@%s:\n"
1213                             "incremental source (%s@%s) "
1214                             "is not earlier than it\n"),
1215                             zhp->zfs_name, sdd->tosnap,
1216                             zhp->zfs_name, sdd->fromsnap);
1217                 } else {
1218                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1219                             "WARNING: "
1220                             "could not send %s@%s: does not exist\n"),
1221                             zhp->zfs_name, sdd->tosnap);
1222                 }
1223                 sdd->err = B_TRUE;
1224         }
1225
1226         return (rv);
1227 }
1228
1229 static int
1230 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1231 {
1232         send_dump_data_t *sdd = arg;
1233         nvpair_t *fspair;
1234         boolean_t needagain, progress;
1235
1236         if (!sdd->replicate)
1237                 return (dump_filesystem(rzhp, sdd));
1238
1239         /* Mark the clone origin snapshots. */
1240         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1241             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1242                 nvlist_t *nvfs;
1243                 uint64_t origin_guid = 0;
1244
1245                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1246                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1247                 if (origin_guid != 0) {
1248                         char *snapname;
1249                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1250                             origin_guid, &snapname);
1251                         if (origin_nv != NULL) {
1252                                 nvlist_t *snapprops;
1253                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1254                                     "snapprops", &snapprops));
1255                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1256                                     snapname, &snapprops));
1257                                 VERIFY(0 == nvlist_add_boolean(
1258                                     snapprops, "is_clone_origin"));
1259                         }
1260                 }
1261         }
1262 again:
1263         needagain = progress = B_FALSE;
1264         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1265             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1266                 nvlist_t *fslist, *parent_nv;
1267                 char *fsname;
1268                 zfs_handle_t *zhp;
1269                 int err;
1270                 uint64_t origin_guid = 0;
1271                 uint64_t parent_guid = 0;
1272
1273                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1274                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1275                         continue;
1276
1277                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1278                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1279                 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1280                     &parent_guid);
1281
1282                 if (parent_guid != 0) {
1283                         parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1284                         if (!nvlist_exists(parent_nv, "sent")) {
1285                                 /* parent has not been sent; skip this one */
1286                                 needagain = B_TRUE;
1287                                 continue;
1288                         }
1289                 }
1290
1291                 if (origin_guid != 0) {
1292                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1293                             origin_guid, NULL);
1294                         if (origin_nv != NULL &&
1295                             !nvlist_exists(origin_nv, "sent")) {
1296                                 /*
1297                                  * origin has not been sent yet;
1298                                  * skip this clone.
1299                                  */
1300                                 needagain = B_TRUE;
1301                                 continue;
1302                         }
1303                 }
1304
1305                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1306                 if (zhp == NULL)
1307                         return (-1);
1308                 err = dump_filesystem(zhp, sdd);
1309                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1310                 progress = B_TRUE;
1311                 zfs_close(zhp);
1312                 if (err)
1313                         return (err);
1314         }
1315         if (needagain) {
1316                 assert(progress);
1317                 goto again;
1318         }
1319
1320         /* clean out the sent flags in case we reuse this fss */
1321         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1322             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1323                 nvlist_t *fslist;
1324
1325                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1326                 (void) nvlist_remove_all(fslist, "sent");
1327         }
1328
1329         return (0);
1330 }
1331
1332 /*
1333  * Generate a send stream for the dataset identified by the argument zhp.
1334  *
1335  * The content of the send stream is the snapshot identified by
1336  * 'tosnap'.  Incremental streams are requested in two ways:
1337  *     - from the snapshot identified by "fromsnap" (if non-null) or
1338  *     - from the origin of the dataset identified by zhp, which must
1339  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1340  *       is TRUE.
1341  *
1342  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1343  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1344  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1345  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1346  * case too. If "props" is set, send properties.
1347  */
1348 int
1349 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1350     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1351     void *cb_arg, nvlist_t **debugnvp)
1352 {
1353         char errbuf[1024];
1354         send_dump_data_t sdd = { 0 };
1355         int err = 0;
1356         nvlist_t *fss = NULL;
1357         avl_tree_t *fsavl = NULL;
1358         static uint64_t holdseq;
1359         int spa_version;
1360         pthread_t tid = 0;
1361         int pipefd[2];
1362         dedup_arg_t dda = { 0 };
1363         int featureflags = 0;
1364
1365         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1366             "cannot send '%s'"), zhp->zfs_name);
1367
1368         if (fromsnap && fromsnap[0] == '\0') {
1369                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1370                     "zero-length incremental source"));
1371                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1372         }
1373
1374         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1375                 uint64_t version;
1376                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1377                 if (version >= ZPL_VERSION_SA) {
1378                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1379                 }
1380         }
1381
1382         if (flags->dedup && !flags->dryrun) {
1383                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1384                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1385                 if (err = pipe(pipefd)) {
1386                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1387                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1388                             errbuf));
1389                 }
1390                 dda.outputfd = outfd;
1391                 dda.inputfd = pipefd[1];
1392                 dda.dedup_hdl = zhp->zfs_hdl;
1393                 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1394                         (void) close(pipefd[0]);
1395                         (void) close(pipefd[1]);
1396                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1397                         return (zfs_error(zhp->zfs_hdl,
1398                             EZFS_THREADCREATEFAILED, errbuf));
1399                 }
1400         }
1401
1402         if (flags->replicate || flags->doall || flags->props) {
1403                 dmu_replay_record_t drr = { 0 };
1404                 char *packbuf = NULL;
1405                 size_t buflen = 0;
1406                 zio_cksum_t zc = { 0 };
1407
1408                 if (flags->replicate || flags->props) {
1409                         nvlist_t *hdrnv;
1410
1411                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1412                         if (fromsnap) {
1413                                 VERIFY(0 == nvlist_add_string(hdrnv,
1414                                     "fromsnap", fromsnap));
1415                         }
1416                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1417                         if (!flags->replicate) {
1418                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1419                                     "not_recursive"));
1420                         }
1421
1422                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1423                             fromsnap, tosnap, flags->replicate, &fss, &fsavl);
1424                         if (err)
1425                                 goto err_out;
1426                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1427                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1428                             NV_ENCODE_XDR, 0);
1429                         if (debugnvp)
1430                                 *debugnvp = hdrnv;
1431                         else
1432                                 nvlist_free(hdrnv);
1433                         if (err)
1434                                 goto stderr_out;
1435                 }
1436
1437                 if (!flags->dryrun) {
1438                         /* write first begin record */
1439                         drr.drr_type = DRR_BEGIN;
1440                         drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1441                         DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1442                             drr_versioninfo, DMU_COMPOUNDSTREAM);
1443                         DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1444                             drr_versioninfo, featureflags);
1445                         (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1446                             sizeof (drr.drr_u.drr_begin.drr_toname),
1447                             "%s@%s", zhp->zfs_name, tosnap);
1448                         drr.drr_payloadlen = buflen;
1449                         err = cksum_and_write(&drr, sizeof (drr), &zc, outfd);
1450
1451                         /* write header nvlist */
1452                         if (err != -1 && packbuf != NULL) {
1453                                 err = cksum_and_write(packbuf, buflen, &zc,
1454                                     outfd);
1455                         }
1456                         free(packbuf);
1457                         if (err == -1) {
1458                                 err = errno;
1459                                 goto stderr_out;
1460                         }
1461
1462                         /* write end record */
1463                         bzero(&drr, sizeof (drr));
1464                         drr.drr_type = DRR_END;
1465                         drr.drr_u.drr_end.drr_checksum = zc;
1466                         err = write(outfd, &drr, sizeof (drr));
1467                         if (err == -1) {
1468                                 err = errno;
1469                                 goto stderr_out;
1470                         }
1471
1472                         err = 0;
1473                 }
1474         }
1475
1476         /* dump each stream */
1477         sdd.fromsnap = fromsnap;
1478         sdd.tosnap = tosnap;
1479         if (tid != 0)
1480                 sdd.outfd = pipefd[0];
1481         else
1482                 sdd.outfd = outfd;
1483         sdd.replicate = flags->replicate;
1484         sdd.doall = flags->doall;
1485         sdd.fromorigin = flags->fromorigin;
1486         sdd.fss = fss;
1487         sdd.fsavl = fsavl;
1488         sdd.verbose = flags->verbose;
1489         sdd.parsable = flags->parsable;
1490         sdd.progress = flags->progress;
1491         sdd.dryrun = flags->dryrun;
1492         sdd.filter_cb = filter_func;
1493         sdd.filter_cb_arg = cb_arg;
1494         if (debugnvp)
1495                 sdd.debugnv = *debugnvp;
1496
1497         /*
1498          * Some flags require that we place user holds on the datasets that are
1499          * being sent so they don't get destroyed during the send. We can skip
1500          * this step if the pool is imported read-only since the datasets cannot
1501          * be destroyed.
1502          */
1503         if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1504             ZPOOL_PROP_READONLY, NULL) &&
1505             zfs_spa_version(zhp, &spa_version) == 0 &&
1506             spa_version >= SPA_VERSION_USERREFS &&
1507             (flags->doall || flags->replicate)) {
1508                 ++holdseq;
1509                 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1510                     ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1511                 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1512                 if (sdd.cleanup_fd < 0) {
1513                         err = errno;
1514                         goto stderr_out;
1515                 }
1516                 sdd.snapholds = fnvlist_alloc();
1517         } else {
1518                 sdd.cleanup_fd = -1;
1519                 sdd.snapholds = NULL;
1520         }
1521         if (flags->verbose || sdd.snapholds != NULL) {
1522                 /*
1523                  * Do a verbose no-op dry run to get all the verbose output
1524                  * or to gather snapshot hold's before generating any data,
1525                  * then do a non-verbose real run to generate the streams.
1526                  */
1527                 sdd.dryrun = B_TRUE;
1528                 err = dump_filesystems(zhp, &sdd);
1529
1530                 if (err != 0)
1531                         goto stderr_out;
1532
1533                 if (flags->verbose) {
1534                         if (flags->parsable) {
1535                                 (void) fprintf(stderr, "size\t%llu\n",
1536                                     (longlong_t)sdd.size);
1537                         } else {
1538                                 char buf[16];
1539                                 zfs_nicenum(sdd.size, buf, sizeof (buf));
1540                                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1541                                     "total estimated size is %s\n"), buf);
1542                         }
1543                 }
1544
1545                 /* Ensure no snaps found is treated as an error. */
1546                 if (!sdd.seento) {
1547                         err = ENOENT;
1548                         goto err_out;
1549                 }
1550
1551                 /* Skip the second run if dryrun was requested. */
1552                 if (flags->dryrun)
1553                         goto err_out;
1554
1555                 if (sdd.snapholds != NULL) {
1556                         err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
1557                         if (err != 0)
1558                                 goto stderr_out;
1559
1560                         fnvlist_free(sdd.snapholds);
1561                         sdd.snapholds = NULL;
1562                 }
1563
1564                 sdd.dryrun = B_FALSE;
1565                 sdd.verbose = B_FALSE;
1566         }
1567
1568         err = dump_filesystems(zhp, &sdd);
1569         fsavl_destroy(fsavl);
1570         nvlist_free(fss);
1571
1572         /* Ensure no snaps found is treated as an error. */
1573         if (err == 0 && !sdd.seento)
1574                 err = ENOENT;
1575
1576         if (tid != 0) {
1577                 if (err != 0)
1578                         (void) pthread_cancel(tid);
1579                 (void) close(pipefd[0]);
1580                 (void) pthread_join(tid, NULL);
1581         }
1582
1583         if (sdd.cleanup_fd != -1) {
1584                 VERIFY(0 == close(sdd.cleanup_fd));
1585                 sdd.cleanup_fd = -1;
1586         }
1587
1588         if (!flags->dryrun && (flags->replicate || flags->doall ||
1589             flags->props)) {
1590                 /*
1591                  * write final end record.  NB: want to do this even if
1592                  * there was some error, because it might not be totally
1593                  * failed.
1594                  */
1595                 dmu_replay_record_t drr = { 0 };
1596                 drr.drr_type = DRR_END;
1597                 if (write(outfd, &drr, sizeof (drr)) == -1) {
1598                         return (zfs_standard_error(zhp->zfs_hdl,
1599                             errno, errbuf));
1600                 }
1601         }
1602
1603         return (err || sdd.err);
1604
1605 stderr_out:
1606         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1607 err_out:
1608         fsavl_destroy(fsavl);
1609         nvlist_free(fss);
1610         fnvlist_free(sdd.snapholds);
1611
1612         if (sdd.cleanup_fd != -1)
1613                 VERIFY(0 == close(sdd.cleanup_fd));
1614         if (tid != 0) {
1615                 (void) pthread_cancel(tid);
1616                 (void) close(pipefd[0]);
1617                 (void) pthread_join(tid, NULL);
1618         }
1619         return (err);
1620 }
1621
1622 int
1623 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd)
1624 {
1625         int err;
1626         libzfs_handle_t *hdl = zhp->zfs_hdl;
1627
1628         char errbuf[1024];
1629         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1630             "warning: cannot send '%s'"), zhp->zfs_name);
1631
1632         err = lzc_send(zhp->zfs_name, from, fd);
1633         if (err != 0) {
1634                 switch (errno) {
1635                 case EXDEV:
1636                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1637                             "not an earlier snapshot from the same fs"));
1638                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
1639
1640                 case ENOENT:
1641                 case ESRCH:
1642                         if (lzc_exists(zhp->zfs_name)) {
1643                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1644                                     "incremental source (%s) does not exist"),
1645                                     from);
1646                         }
1647                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
1648
1649                 case EBUSY:
1650                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1651                             "target is busy; if a filesystem, "
1652                             "it must not be mounted"));
1653                         return (zfs_error(hdl, EZFS_BUSY, errbuf));
1654
1655                 case EDQUOT:
1656                 case EFBIG:
1657                 case EIO:
1658                 case ENOLINK:
1659                 case ENOSPC:
1660 #ifdef illumos
1661                 case ENOSTR:
1662 #endif
1663                 case ENXIO:
1664                 case EPIPE:
1665                 case ERANGE:
1666                 case EFAULT:
1667                 case EROFS:
1668                         zfs_error_aux(hdl, strerror(errno));
1669                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1670
1671                 default:
1672                         return (zfs_standard_error(hdl, errno, errbuf));
1673                 }
1674         }
1675         return (err != 0);
1676 }
1677
1678 /*
1679  * Routines specific to "zfs recv"
1680  */
1681
1682 static int
1683 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1684     boolean_t byteswap, zio_cksum_t *zc)
1685 {
1686         char *cp = buf;
1687         int rv;
1688         int len = ilen;
1689
1690         do {
1691                 rv = read(fd, cp, len);
1692                 cp += rv;
1693                 len -= rv;
1694         } while (rv > 0);
1695
1696         if (rv < 0 || len != 0) {
1697                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1698                     "failed to read from stream"));
1699                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1700                     "cannot receive")));
1701         }
1702
1703         if (zc) {
1704                 if (byteswap)
1705                         fletcher_4_incremental_byteswap(buf, ilen, zc);
1706                 else
1707                         fletcher_4_incremental_native(buf, ilen, zc);
1708         }
1709         return (0);
1710 }
1711
1712 static int
1713 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1714     boolean_t byteswap, zio_cksum_t *zc)
1715 {
1716         char *buf;
1717         int err;
1718
1719         buf = zfs_alloc(hdl, len);
1720         if (buf == NULL)
1721                 return (ENOMEM);
1722
1723         err = recv_read(hdl, fd, buf, len, byteswap, zc);
1724         if (err != 0) {
1725                 free(buf);
1726                 return (err);
1727         }
1728
1729         err = nvlist_unpack(buf, len, nvp, 0);
1730         free(buf);
1731         if (err != 0) {
1732                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1733                     "stream (malformed nvlist)"));
1734                 return (EINVAL);
1735         }
1736         return (0);
1737 }
1738
1739 static int
1740 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1741     int baselen, char *newname, recvflags_t *flags)
1742 {
1743         static int seq;
1744         zfs_cmd_t zc = { 0 };
1745         int err;
1746         prop_changelist_t *clp;
1747         zfs_handle_t *zhp;
1748
1749         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1750         if (zhp == NULL)
1751                 return (-1);
1752         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1753             flags->force ? MS_FORCE : 0);
1754         zfs_close(zhp);
1755         if (clp == NULL)
1756                 return (-1);
1757         err = changelist_prefix(clp);
1758         if (err)
1759                 return (err);
1760
1761         zc.zc_objset_type = DMU_OST_ZFS;
1762         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1763
1764         if (tryname) {
1765                 (void) strcpy(newname, tryname);
1766
1767                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1768
1769                 if (flags->verbose) {
1770                         (void) printf("attempting rename %s to %s\n",
1771                             zc.zc_name, zc.zc_value);
1772                 }
1773                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1774                 if (err == 0)
1775                         changelist_rename(clp, name, tryname);
1776         } else {
1777                 err = ENOENT;
1778         }
1779
1780         if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
1781                 seq++;
1782
1783                 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u",
1784                     baselen, name, getpid(), seq);
1785                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1786
1787                 if (flags->verbose) {
1788                         (void) printf("failed - trying rename %s to %s\n",
1789                             zc.zc_name, zc.zc_value);
1790                 }
1791                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1792                 if (err == 0)
1793                         changelist_rename(clp, name, newname);
1794                 if (err && flags->verbose) {
1795                         (void) printf("failed (%u) - "
1796                             "will try again on next pass\n", errno);
1797                 }
1798                 err = EAGAIN;
1799         } else if (flags->verbose) {
1800                 if (err == 0)
1801                         (void) printf("success\n");
1802                 else
1803                         (void) printf("failed (%u)\n", errno);
1804         }
1805
1806         (void) changelist_postfix(clp);
1807         changelist_free(clp);
1808
1809         return (err);
1810 }
1811
1812 static int
1813 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1814     char *newname, recvflags_t *flags)
1815 {
1816         zfs_cmd_t zc = { 0 };
1817         int err = 0;
1818         prop_changelist_t *clp;
1819         zfs_handle_t *zhp;
1820         boolean_t defer = B_FALSE;
1821         int spa_version;
1822
1823         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1824         if (zhp == NULL)
1825                 return (-1);
1826         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1827             flags->force ? MS_FORCE : 0);
1828         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1829             zfs_spa_version(zhp, &spa_version) == 0 &&
1830             spa_version >= SPA_VERSION_USERREFS)
1831                 defer = B_TRUE;
1832         zfs_close(zhp);
1833         if (clp == NULL)
1834                 return (-1);
1835         err = changelist_prefix(clp);
1836         if (err)
1837                 return (err);
1838
1839         zc.zc_objset_type = DMU_OST_ZFS;
1840         zc.zc_defer_destroy = defer;
1841         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1842
1843         if (flags->verbose)
1844                 (void) printf("attempting destroy %s\n", zc.zc_name);
1845         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1846         if (err == 0) {
1847                 if (flags->verbose)
1848                         (void) printf("success\n");
1849                 changelist_remove(clp, zc.zc_name);
1850         }
1851
1852         (void) changelist_postfix(clp);
1853         changelist_free(clp);
1854
1855         /*
1856          * Deferred destroy might destroy the snapshot or only mark it to be
1857          * destroyed later, and it returns success in either case.
1858          */
1859         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1860             ZFS_TYPE_SNAPSHOT))) {
1861                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1862         }
1863
1864         return (err);
1865 }
1866
1867 typedef struct guid_to_name_data {
1868         uint64_t guid;
1869         char *name;
1870         char *skip;
1871 } guid_to_name_data_t;
1872
1873 static int
1874 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1875 {
1876         guid_to_name_data_t *gtnd = arg;
1877         int err;
1878
1879         if (gtnd->skip != NULL &&
1880             strcmp(zhp->zfs_name, gtnd->skip) == 0) {
1881                 return (0);
1882         }
1883
1884         if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1885                 (void) strcpy(gtnd->name, zhp->zfs_name);
1886                 zfs_close(zhp);
1887                 return (EEXIST);
1888         }
1889
1890         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1891         zfs_close(zhp);
1892         return (err);
1893 }
1894
1895 /*
1896  * Attempt to find the local dataset associated with this guid.  In the case of
1897  * multiple matches, we attempt to find the "best" match by searching
1898  * progressively larger portions of the hierarchy.  This allows one to send a
1899  * tree of datasets individually and guarantee that we will find the source
1900  * guid within that hierarchy, even if there are multiple matches elsewhere.
1901  */
1902 static int
1903 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1904     char *name)
1905 {
1906         /* exhaustive search all local snapshots */
1907         char pname[ZFS_MAXNAMELEN];
1908         guid_to_name_data_t gtnd;
1909         int err = 0;
1910         zfs_handle_t *zhp;
1911         char *cp;
1912
1913         gtnd.guid = guid;
1914         gtnd.name = name;
1915         gtnd.skip = NULL;
1916
1917         (void) strlcpy(pname, parent, sizeof (pname));
1918
1919         /*
1920          * Search progressively larger portions of the hierarchy.  This will
1921          * select the "most local" version of the origin snapshot in the case
1922          * that there are multiple matching snapshots in the system.
1923          */
1924         while ((cp = strrchr(pname, '/')) != NULL) {
1925
1926                 /* Chop off the last component and open the parent */
1927                 *cp = '\0';
1928                 zhp = make_dataset_handle(hdl, pname);
1929
1930                 if (zhp == NULL)
1931                         continue;
1932
1933                 err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1934                 zfs_close(zhp);
1935                 if (err == EEXIST)
1936                         return (0);
1937
1938                 /*
1939                  * Remember the dataset that we already searched, so we
1940                  * skip it next time through.
1941                  */
1942                 gtnd.skip = pname;
1943         }
1944
1945         return (ENOENT);
1946 }
1947
1948 /*
1949  * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
1950  * guid1 is after guid2.
1951  */
1952 static int
1953 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1954     uint64_t guid1, uint64_t guid2)
1955 {
1956         nvlist_t *nvfs;
1957         char *fsname, *snapname;
1958         char buf[ZFS_MAXNAMELEN];
1959         int rv;
1960         zfs_handle_t *guid1hdl, *guid2hdl;
1961         uint64_t create1, create2;
1962
1963         if (guid2 == 0)
1964                 return (0);
1965         if (guid1 == 0)
1966                 return (1);
1967
1968         nvfs = fsavl_find(avl, guid1, &snapname);
1969         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1970         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1971         guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1972         if (guid1hdl == NULL)
1973                 return (-1);
1974
1975         nvfs = fsavl_find(avl, guid2, &snapname);
1976         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1977         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1978         guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1979         if (guid2hdl == NULL) {
1980                 zfs_close(guid1hdl);
1981                 return (-1);
1982         }
1983
1984         create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
1985         create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
1986
1987         if (create1 < create2)
1988                 rv = -1;
1989         else if (create1 > create2)
1990                 rv = +1;
1991         else
1992                 rv = 0;
1993
1994         zfs_close(guid1hdl);
1995         zfs_close(guid2hdl);
1996
1997         return (rv);
1998 }
1999
2000 static int
2001 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2002     recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2003     nvlist_t *renamed)
2004 {
2005         nvlist_t *local_nv, *deleted = NULL;
2006         avl_tree_t *local_avl;
2007         nvpair_t *fselem, *nextfselem;
2008         char *fromsnap;
2009         char newname[ZFS_MAXNAMELEN];
2010         char guidname[32];
2011         int error;
2012         boolean_t needagain, progress, recursive;
2013         char *s1, *s2;
2014
2015         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2016
2017         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2018             ENOENT);
2019
2020         if (flags->dryrun)
2021                 return (0);
2022
2023 again:
2024         needagain = progress = B_FALSE;
2025
2026         VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0));
2027
2028         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2029             recursive, &local_nv, &local_avl)) != 0)
2030                 return (error);
2031
2032         /*
2033          * Process deletes and renames
2034          */
2035         for (fselem = nvlist_next_nvpair(local_nv, NULL);
2036             fselem; fselem = nextfselem) {
2037                 nvlist_t *nvfs, *snaps;
2038                 nvlist_t *stream_nvfs = NULL;
2039                 nvpair_t *snapelem, *nextsnapelem;
2040                 uint64_t fromguid = 0;
2041                 uint64_t originguid = 0;
2042                 uint64_t stream_originguid = 0;
2043                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2044                 char *fsname, *stream_fsname;
2045
2046                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2047
2048                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2049                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2050                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2051                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2052                     &parent_fromsnap_guid));
2053                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2054
2055                 /*
2056                  * First find the stream's fs, so we can check for
2057                  * a different origin (due to "zfs promote")
2058                  */
2059                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2060                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2061                         uint64_t thisguid;
2062
2063                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2064                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2065
2066                         if (stream_nvfs != NULL)
2067                                 break;
2068                 }
2069
2070                 /* check for promote */
2071                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2072                     &stream_originguid);
2073                 if (stream_nvfs && originguid != stream_originguid) {
2074                         switch (created_before(hdl, local_avl,
2075                             stream_originguid, originguid)) {
2076                         case 1: {
2077                                 /* promote it! */
2078                                 zfs_cmd_t zc = { 0 };
2079                                 nvlist_t *origin_nvfs;
2080                                 char *origin_fsname;
2081
2082                                 if (flags->verbose)
2083                                         (void) printf("promoting %s\n", fsname);
2084
2085                                 origin_nvfs = fsavl_find(local_avl, originguid,
2086                                     NULL);
2087                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2088                                     "name", &origin_fsname));
2089                                 (void) strlcpy(zc.zc_value, origin_fsname,
2090                                     sizeof (zc.zc_value));
2091                                 (void) strlcpy(zc.zc_name, fsname,
2092                                     sizeof (zc.zc_name));
2093                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2094                                 if (error == 0)
2095                                         progress = B_TRUE;
2096                                 break;
2097                         }
2098                         default:
2099                                 break;
2100                         case -1:
2101                                 fsavl_destroy(local_avl);
2102                                 nvlist_free(local_nv);
2103                                 return (-1);
2104                         }
2105                         /*
2106                          * We had/have the wrong origin, therefore our
2107                          * list of snapshots is wrong.  Need to handle
2108                          * them on the next pass.
2109                          */
2110                         needagain = B_TRUE;
2111                         continue;
2112                 }
2113
2114                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2115                     snapelem; snapelem = nextsnapelem) {
2116                         uint64_t thisguid;
2117                         char *stream_snapname;
2118                         nvlist_t *found, *props;
2119
2120                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2121
2122                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2123                         found = fsavl_find(stream_avl, thisguid,
2124                             &stream_snapname);
2125
2126                         /* check for delete */
2127                         if (found == NULL) {
2128                                 char name[ZFS_MAXNAMELEN];
2129
2130                                 if (!flags->force)
2131                                         continue;
2132
2133                                 (void) snprintf(name, sizeof (name), "%s@%s",
2134                                     fsname, nvpair_name(snapelem));
2135
2136                                 error = recv_destroy(hdl, name,
2137                                     strlen(fsname)+1, newname, flags);
2138                                 if (error)
2139                                         needagain = B_TRUE;
2140                                 else
2141                                         progress = B_TRUE;
2142                                 sprintf(guidname, "%lu", thisguid);
2143                                 nvlist_add_boolean(deleted, guidname);
2144                                 continue;
2145                         }
2146
2147                         stream_nvfs = found;
2148
2149                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2150                             &props) && 0 == nvlist_lookup_nvlist(props,
2151                             stream_snapname, &props)) {
2152                                 zfs_cmd_t zc = { 0 };
2153
2154                                 zc.zc_cookie = B_TRUE; /* received */
2155                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2156                                     "%s@%s", fsname, nvpair_name(snapelem));
2157                                 if (zcmd_write_src_nvlist(hdl, &zc,
2158                                     props) == 0) {
2159                                         (void) zfs_ioctl(hdl,
2160                                             ZFS_IOC_SET_PROP, &zc);
2161                                         zcmd_free_nvlists(&zc);
2162                                 }
2163                         }
2164
2165                         /* check for different snapname */
2166                         if (strcmp(nvpair_name(snapelem),
2167                             stream_snapname) != 0) {
2168                                 char name[ZFS_MAXNAMELEN];
2169                                 char tryname[ZFS_MAXNAMELEN];
2170
2171                                 (void) snprintf(name, sizeof (name), "%s@%s",
2172                                     fsname, nvpair_name(snapelem));
2173                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
2174                                     fsname, stream_snapname);
2175
2176                                 error = recv_rename(hdl, name, tryname,
2177                                     strlen(fsname)+1, newname, flags);
2178                                 if (error)
2179                                         needagain = B_TRUE;
2180                                 else
2181                                         progress = B_TRUE;
2182                         }
2183
2184                         if (strcmp(stream_snapname, fromsnap) == 0)
2185                                 fromguid = thisguid;
2186                 }
2187
2188                 /* check for delete */
2189                 if (stream_nvfs == NULL) {
2190                         if (!flags->force)
2191                                 continue;
2192
2193                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2194                             newname, flags);
2195                         if (error)
2196                                 needagain = B_TRUE;
2197                         else
2198                                 progress = B_TRUE;
2199                         sprintf(guidname, "%lu", parent_fromsnap_guid);
2200                         nvlist_add_boolean(deleted, guidname);
2201                         continue;
2202                 }
2203
2204                 if (fromguid == 0) {
2205                         if (flags->verbose) {
2206                                 (void) printf("local fs %s does not have "
2207                                     "fromsnap (%s in stream); must have "
2208                                     "been deleted locally; ignoring\n",
2209                                     fsname, fromsnap);
2210                         }
2211                         continue;
2212                 }
2213
2214                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2215                     "name", &stream_fsname));
2216                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2217                     "parentfromsnap", &stream_parent_fromsnap_guid));
2218
2219                 s1 = strrchr(fsname, '/');
2220                 s2 = strrchr(stream_fsname, '/');
2221
2222                 /*
2223                  * Check if we're going to rename based on parent guid change
2224                  * and the current parent guid was also deleted. If it was then
2225                  * rename will fail and is likely unneeded, so avoid this and
2226                  * force an early retry to determine the new
2227                  * parent_fromsnap_guid.
2228                  */
2229                 if (stream_parent_fromsnap_guid != 0 &&
2230                     parent_fromsnap_guid != 0 &&
2231                     stream_parent_fromsnap_guid != parent_fromsnap_guid) {
2232                         sprintf(guidname, "%lu", parent_fromsnap_guid);
2233                         if (nvlist_exists(deleted, guidname)) {
2234                                 progress = B_TRUE;
2235                                 needagain = B_TRUE;
2236                                 goto doagain;
2237                         }
2238                 }
2239
2240                 /*
2241                  * Check for rename. If the exact receive path is specified, it
2242                  * does not count as a rename, but we still need to check the
2243                  * datasets beneath it.
2244                  */
2245                 if ((stream_parent_fromsnap_guid != 0 &&
2246                     parent_fromsnap_guid != 0 &&
2247                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2248                     ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2249                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2250                         nvlist_t *parent;
2251                         char tryname[ZFS_MAXNAMELEN];
2252
2253                         parent = fsavl_find(local_avl,
2254                             stream_parent_fromsnap_guid, NULL);
2255                         /*
2256                          * NB: parent might not be found if we used the
2257                          * tosnap for stream_parent_fromsnap_guid,
2258                          * because the parent is a newly-created fs;
2259                          * we'll be able to rename it after we recv the
2260                          * new fs.
2261                          */
2262                         if (parent != NULL) {
2263                                 char *pname;
2264
2265                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
2266                                     &pname));
2267                                 (void) snprintf(tryname, sizeof (tryname),
2268                                     "%s%s", pname, strrchr(stream_fsname, '/'));
2269                         } else {
2270                                 tryname[0] = '\0';
2271                                 if (flags->verbose) {
2272                                         (void) printf("local fs %s new parent "
2273                                             "not found\n", fsname);
2274                                 }
2275                         }
2276
2277                         newname[0] = '\0';
2278
2279                         error = recv_rename(hdl, fsname, tryname,
2280                             strlen(tofs)+1, newname, flags);
2281
2282                         if (renamed != NULL && newname[0] != '\0') {
2283                                 VERIFY(0 == nvlist_add_boolean(renamed,
2284                                     newname));
2285                         }
2286
2287                         if (error)
2288                                 needagain = B_TRUE;
2289                         else
2290                                 progress = B_TRUE;
2291                 }
2292         }
2293
2294 doagain:
2295         fsavl_destroy(local_avl);
2296         nvlist_free(local_nv);
2297         nvlist_free(deleted);
2298
2299         if (needagain && progress) {
2300                 /* do another pass to fix up temporary names */
2301                 if (flags->verbose)
2302                         (void) printf("another pass:\n");
2303                 goto again;
2304         }
2305
2306         return (needagain);
2307 }
2308
2309 static int
2310 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2311     recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2312     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2313 {
2314         nvlist_t *stream_nv = NULL;
2315         avl_tree_t *stream_avl = NULL;
2316         char *fromsnap = NULL;
2317         char *cp;
2318         char tofs[ZFS_MAXNAMELEN];
2319         char sendfs[ZFS_MAXNAMELEN];
2320         char errbuf[1024];
2321         dmu_replay_record_t drre;
2322         int error;
2323         boolean_t anyerr = B_FALSE;
2324         boolean_t softerr = B_FALSE;
2325         boolean_t recursive;
2326
2327         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2328             "cannot receive"));
2329
2330         assert(drr->drr_type == DRR_BEGIN);
2331         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2332         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2333             DMU_COMPOUNDSTREAM);
2334
2335         /*
2336          * Read in the nvlist from the stream.
2337          */
2338         if (drr->drr_payloadlen != 0) {
2339                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2340                     &stream_nv, flags->byteswap, zc);
2341                 if (error) {
2342                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2343                         goto out;
2344                 }
2345         }
2346
2347         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2348             ENOENT);
2349
2350         if (recursive && strchr(destname, '@')) {
2351                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2352                     "cannot specify snapshot name for multi-snapshot stream"));
2353                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2354                 goto out;
2355         }
2356
2357         /*
2358          * Read in the end record and verify checksum.
2359          */
2360         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2361             flags->byteswap, NULL)))
2362                 goto out;
2363         if (flags->byteswap) {
2364                 drre.drr_type = BSWAP_32(drre.drr_type);
2365                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2366                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2367                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2368                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2369                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2370                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2371                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2372                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2373         }
2374         if (drre.drr_type != DRR_END) {
2375                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2376                 goto out;
2377         }
2378         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2379                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2380                     "incorrect header checksum"));
2381                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2382                 goto out;
2383         }
2384
2385         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2386
2387         if (drr->drr_payloadlen != 0) {
2388                 nvlist_t *stream_fss;
2389
2390                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2391                     &stream_fss));
2392                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2393                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2394                             "couldn't allocate avl tree"));
2395                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2396                         goto out;
2397                 }
2398
2399                 if (fromsnap != NULL) {
2400                         nvlist_t *renamed = NULL;
2401                         nvpair_t *pair = NULL;
2402
2403                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2404                         if (flags->isprefix) {
2405                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2406                                 int i;
2407
2408                                 if (flags->istail) {
2409                                         cp = strrchr(drrb->drr_toname, '/');
2410                                         if (cp == NULL) {
2411                                                 (void) strlcat(tofs, "/",
2412                                                     ZFS_MAXNAMELEN);
2413                                                 i = 0;
2414                                         } else {
2415                                                 i = (cp - drrb->drr_toname);
2416                                         }
2417                                 } else {
2418                                         i = strcspn(drrb->drr_toname, "/@");
2419                                 }
2420                                 /* zfs_receive_one() will create_parents() */
2421                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2422                                     ZFS_MAXNAMELEN);
2423                                 *strchr(tofs, '@') = '\0';
2424                         }
2425
2426                         if (recursive && !flags->dryrun && !flags->nomount) {
2427                                 VERIFY(0 == nvlist_alloc(&renamed,
2428                                     NV_UNIQUE_NAME, 0));
2429                         }
2430
2431                         softerr = recv_incremental_replication(hdl, tofs, flags,
2432                             stream_nv, stream_avl, renamed);
2433
2434                         /* Unmount renamed filesystems before receiving. */
2435                         while ((pair = nvlist_next_nvpair(renamed,
2436                             pair)) != NULL) {
2437                                 zfs_handle_t *zhp;
2438                                 prop_changelist_t *clp = NULL;
2439
2440                                 zhp = zfs_open(hdl, nvpair_name(pair),
2441                                     ZFS_TYPE_FILESYSTEM);
2442                                 if (zhp != NULL) {
2443                                         clp = changelist_gather(zhp,
2444                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2445                                         zfs_close(zhp);
2446                                         if (clp != NULL) {
2447                                                 softerr |=
2448                                                     changelist_prefix(clp);
2449                                                 changelist_free(clp);
2450                                         }
2451                                 }
2452                         }
2453
2454                         nvlist_free(renamed);
2455                 }
2456         }
2457
2458         /*
2459          * Get the fs specified by the first path in the stream (the top level
2460          * specified by 'zfs send') and pass it to each invocation of
2461          * zfs_receive_one().
2462          */
2463         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2464             ZFS_MAXNAMELEN);
2465         if ((cp = strchr(sendfs, '@')) != NULL)
2466                 *cp = '\0';
2467
2468         /* Finally, receive each contained stream */
2469         do {
2470                 /*
2471                  * we should figure out if it has a recoverable
2472                  * error, in which case do a recv_skip() and drive on.
2473                  * Note, if we fail due to already having this guid,
2474                  * zfs_receive_one() will take care of it (ie,
2475                  * recv_skip() and return 0).
2476                  */
2477                 error = zfs_receive_impl(hdl, destname, flags, fd,
2478                     sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2479                     action_handlep);
2480                 if (error == ENODATA) {
2481                         error = 0;
2482                         break;
2483                 }
2484                 anyerr |= error;
2485         } while (error == 0);
2486
2487         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2488                 /*
2489                  * Now that we have the fs's they sent us, try the
2490                  * renames again.
2491                  */
2492                 softerr = recv_incremental_replication(hdl, tofs, flags,
2493                     stream_nv, stream_avl, NULL);
2494         }
2495
2496 out:
2497         fsavl_destroy(stream_avl);
2498         if (stream_nv)
2499                 nvlist_free(stream_nv);
2500         if (softerr)
2501                 error = -2;
2502         if (anyerr)
2503                 error = -1;
2504         return (error);
2505 }
2506
2507 static void
2508 trunc_prop_errs(int truncated)
2509 {
2510         ASSERT(truncated != 0);
2511
2512         if (truncated == 1)
2513                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2514                     "1 more property could not be set\n"));
2515         else
2516                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2517                     "%d more properties could not be set\n"), truncated);
2518 }
2519
2520 static int
2521 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2522 {
2523         dmu_replay_record_t *drr;
2524         void *buf = malloc(1<<20);
2525         char errbuf[1024];
2526
2527         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2528             "cannot receive:"));
2529
2530         /* XXX would be great to use lseek if possible... */
2531         drr = buf;
2532
2533         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2534             byteswap, NULL) == 0) {
2535                 if (byteswap)
2536                         drr->drr_type = BSWAP_32(drr->drr_type);
2537
2538                 switch (drr->drr_type) {
2539                 case DRR_BEGIN:
2540                         /* NB: not to be used on v2 stream packages */
2541                         if (drr->drr_payloadlen != 0) {
2542                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2543                                     "invalid substream header"));
2544                                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2545                         }
2546                         break;
2547
2548                 case DRR_END:
2549                         free(buf);
2550                         return (0);
2551
2552                 case DRR_OBJECT:
2553                         if (byteswap) {
2554                                 drr->drr_u.drr_object.drr_bonuslen =
2555                                     BSWAP_32(drr->drr_u.drr_object.
2556                                     drr_bonuslen);
2557                         }
2558                         (void) recv_read(hdl, fd, buf,
2559                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2560                             B_FALSE, NULL);
2561                         break;
2562
2563                 case DRR_WRITE:
2564                         if (byteswap) {
2565                                 drr->drr_u.drr_write.drr_length =
2566                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
2567                         }
2568                         (void) recv_read(hdl, fd, buf,
2569                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2570                         break;
2571                 case DRR_SPILL:
2572                         if (byteswap) {
2573                                 drr->drr_u.drr_write.drr_length =
2574                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
2575                         }
2576                         (void) recv_read(hdl, fd, buf,
2577                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2578                         break;
2579                 case DRR_WRITE_BYREF:
2580                 case DRR_FREEOBJECTS:
2581                 case DRR_FREE:
2582                         break;
2583
2584                 default:
2585                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2586                             "invalid record type"));
2587                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2588                 }
2589         }
2590
2591         free(buf);
2592         return (-1);
2593 }
2594
2595 /*
2596  * Restores a backup of tosnap from the file descriptor specified by infd.
2597  */
2598 static int
2599 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2600     recvflags_t *flags, dmu_replay_record_t *drr,
2601     dmu_replay_record_t *drr_noswap, const char *sendfs,
2602     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2603     uint64_t *action_handlep)
2604 {
2605         zfs_cmd_t zc = { 0 };
2606         time_t begin_time;
2607         int ioctl_err, ioctl_errno, err;
2608         char *cp;
2609         struct drr_begin *drrb = &drr->drr_u.drr_begin;
2610         char errbuf[1024];
2611         char prop_errbuf[1024];
2612         const char *chopprefix;
2613         boolean_t newfs = B_FALSE;
2614         boolean_t stream_wantsnewfs;
2615         uint64_t parent_snapguid = 0;
2616         prop_changelist_t *clp = NULL;
2617         nvlist_t *snapprops_nvlist = NULL;
2618         zprop_errflags_t prop_errflags;
2619         boolean_t recursive;
2620
2621         begin_time = time(NULL);
2622
2623         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2624             "cannot receive"));
2625
2626         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2627             ENOENT);
2628
2629         if (stream_avl != NULL) {
2630                 char *snapname;
2631                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2632                     &snapname);
2633                 nvlist_t *props;
2634                 int ret;
2635
2636                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2637                     &parent_snapguid);
2638                 err = nvlist_lookup_nvlist(fs, "props", &props);
2639                 if (err)
2640                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2641
2642                 if (flags->canmountoff) {
2643                         VERIFY(0 == nvlist_add_uint64(props,
2644                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2645                 }
2646                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2647                 if (err)
2648                         nvlist_free(props);
2649
2650                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2651                         VERIFY(0 == nvlist_lookup_nvlist(props,
2652                             snapname, &snapprops_nvlist));
2653                 }
2654
2655                 if (ret != 0)
2656                         return (-1);
2657         }
2658
2659         cp = NULL;
2660
2661         /*
2662          * Determine how much of the snapshot name stored in the stream
2663          * we are going to tack on to the name they specified on the
2664          * command line, and how much we are going to chop off.
2665          *
2666          * If they specified a snapshot, chop the entire name stored in
2667          * the stream.
2668          */
2669         if (flags->istail) {
2670                 /*
2671                  * A filesystem was specified with -e. We want to tack on only
2672                  * the tail of the sent snapshot path.
2673                  */
2674                 if (strchr(tosnap, '@')) {
2675                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2676                             "argument - snapshot not allowed with -e"));
2677                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2678                 }
2679
2680                 chopprefix = strrchr(sendfs, '/');
2681
2682                 if (chopprefix == NULL) {
2683                         /*
2684                          * The tail is the poolname, so we need to
2685                          * prepend a path separator.
2686                          */
2687                         int len = strlen(drrb->drr_toname);
2688                         cp = malloc(len + 2);
2689                         cp[0] = '/';
2690                         (void) strcpy(&cp[1], drrb->drr_toname);
2691                         chopprefix = cp;
2692                 } else {
2693                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2694                 }
2695         } else if (flags->isprefix) {
2696                 /*
2697                  * A filesystem was specified with -d. We want to tack on
2698                  * everything but the first element of the sent snapshot path
2699                  * (all but the pool name).
2700                  */
2701                 if (strchr(tosnap, '@')) {
2702                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2703                             "argument - snapshot not allowed with -d"));
2704                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2705                 }
2706
2707                 chopprefix = strchr(drrb->drr_toname, '/');
2708                 if (chopprefix == NULL)
2709                         chopprefix = strchr(drrb->drr_toname, '@');
2710         } else if (strchr(tosnap, '@') == NULL) {
2711                 /*
2712                  * If a filesystem was specified without -d or -e, we want to
2713                  * tack on everything after the fs specified by 'zfs send'.
2714                  */
2715                 chopprefix = drrb->drr_toname + strlen(sendfs);
2716         } else {
2717                 /* A snapshot was specified as an exact path (no -d or -e). */
2718                 if (recursive) {
2719                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2720                             "cannot specify snapshot name for multi-snapshot "
2721                             "stream"));
2722                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2723                 }
2724                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2725         }
2726
2727         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2728         ASSERT(chopprefix > drrb->drr_toname);
2729         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2730         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2731             chopprefix[0] == '\0');
2732
2733         /*
2734          * Determine name of destination snapshot, store in zc_value.
2735          */
2736         (void) strcpy(zc.zc_value, tosnap);
2737         (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2738 #ifdef __FreeBSD__
2739         if (zfs_ioctl_version == ZFS_IOCVER_UNDEF)
2740                 zfs_ioctl_version = get_zfs_ioctl_version();
2741         /*
2742          * For forward compatibility hide tosnap in zc_value
2743          */
2744         if (zfs_ioctl_version < ZFS_IOCVER_LZC)
2745                 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap);
2746 #endif
2747         free(cp);
2748         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2749                 zcmd_free_nvlists(&zc);
2750                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2751         }
2752
2753         /*
2754          * Determine the name of the origin snapshot, store in zc_string.
2755          */
2756         if (drrb->drr_flags & DRR_FLAG_CLONE) {
2757                 if (guid_to_name(hdl, zc.zc_value,
2758                     drrb->drr_fromguid, zc.zc_string) != 0) {
2759                         zcmd_free_nvlists(&zc);
2760                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2761                             "local origin for clone %s does not exist"),
2762                             zc.zc_value);
2763                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2764                 }
2765                 if (flags->verbose)
2766                         (void) printf("found clone origin %s\n", zc.zc_string);
2767         }
2768
2769         stream_wantsnewfs = (drrb->drr_fromguid == 0 ||
2770             (drrb->drr_flags & DRR_FLAG_CLONE));
2771
2772         if (stream_wantsnewfs) {
2773                 /*
2774                  * if the parent fs does not exist, look for it based on
2775                  * the parent snap GUID
2776                  */
2777                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2778                     "cannot receive new filesystem stream"));
2779
2780                 (void) strcpy(zc.zc_name, zc.zc_value);
2781                 cp = strrchr(zc.zc_name, '/');
2782                 if (cp)
2783                         *cp = '\0';
2784                 if (cp &&
2785                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2786                         char suffix[ZFS_MAXNAMELEN];
2787                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2788                         if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
2789                             zc.zc_value) == 0) {
2790                                 *strchr(zc.zc_value, '@') = '\0';
2791                                 (void) strcat(zc.zc_value, suffix);
2792                         }
2793                 }
2794         } else {
2795                 /*
2796                  * if the fs does not exist, look for it based on the
2797                  * fromsnap GUID
2798                  */
2799                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2800                     "cannot receive incremental stream"));
2801
2802                 (void) strcpy(zc.zc_name, zc.zc_value);
2803                 *strchr(zc.zc_name, '@') = '\0';
2804
2805                 /*
2806                  * If the exact receive path was specified and this is the
2807                  * topmost path in the stream, then if the fs does not exist we
2808                  * should look no further.
2809                  */
2810                 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
2811                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2812                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2813                         char snap[ZFS_MAXNAMELEN];
2814                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
2815                         if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
2816                             zc.zc_value) == 0) {
2817                                 *strchr(zc.zc_value, '@') = '\0';
2818                                 (void) strcat(zc.zc_value, snap);
2819                         }
2820                 }
2821         }
2822
2823         (void) strcpy(zc.zc_name, zc.zc_value);
2824         *strchr(zc.zc_name, '@') = '\0';
2825
2826         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2827                 zfs_handle_t *zhp;
2828
2829                 /*
2830                  * Destination fs exists.  Therefore this should either
2831                  * be an incremental, or the stream specifies a new fs
2832                  * (full stream or clone) and they want us to blow it
2833                  * away (and have therefore specified -F and removed any
2834                  * snapshots).
2835                  */
2836                 if (stream_wantsnewfs) {
2837                         if (!flags->force) {
2838                                 zcmd_free_nvlists(&zc);
2839                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2840                                     "destination '%s' exists\n"
2841                                     "must specify -F to overwrite it"),
2842                                     zc.zc_name);
2843                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2844                         }
2845                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2846                             &zc) == 0) {
2847                                 zcmd_free_nvlists(&zc);
2848                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2849                                     "destination has snapshots (eg. %s)\n"
2850                                     "must destroy them to overwrite it"),
2851                                     zc.zc_name);
2852                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2853                         }
2854                 }
2855
2856                 if ((zhp = zfs_open(hdl, zc.zc_name,
2857                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2858                         zcmd_free_nvlists(&zc);
2859                         return (-1);
2860                 }
2861
2862                 if (stream_wantsnewfs &&
2863                     zhp->zfs_dmustats.dds_origin[0]) {
2864                         zcmd_free_nvlists(&zc);
2865                         zfs_close(zhp);
2866                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2867                             "destination '%s' is a clone\n"
2868                             "must destroy it to overwrite it"),
2869                             zc.zc_name);
2870                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2871                 }
2872
2873                 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2874                     stream_wantsnewfs) {
2875                         /* We can't do online recv in this case */
2876                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2877                         if (clp == NULL) {
2878                                 zfs_close(zhp);
2879                                 zcmd_free_nvlists(&zc);
2880                                 return (-1);
2881                         }
2882                         if (changelist_prefix(clp) != 0) {
2883                                 changelist_free(clp);
2884                                 zfs_close(zhp);
2885                                 zcmd_free_nvlists(&zc);
2886                                 return (-1);
2887                         }
2888                 }
2889                 zfs_close(zhp);
2890         } else {
2891                 /*
2892                  * Destination filesystem does not exist.  Therefore we better
2893                  * be creating a new filesystem (either from a full backup, or
2894                  * a clone).  It would therefore be invalid if the user
2895                  * specified only the pool name (i.e. if the destination name
2896                  * contained no slash character).
2897                  */
2898                 if (!stream_wantsnewfs ||
2899                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
2900                         zcmd_free_nvlists(&zc);
2901                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2902                             "destination '%s' does not exist"), zc.zc_name);
2903                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2904                 }
2905
2906                 /*
2907                  * Trim off the final dataset component so we perform the
2908                  * recvbackup ioctl to the filesystems's parent.
2909                  */
2910                 *cp = '\0';
2911
2912                 if (flags->isprefix && !flags->istail && !flags->dryrun &&
2913                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2914                         zcmd_free_nvlists(&zc);
2915                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2916                 }
2917
2918                 newfs = B_TRUE;
2919         }
2920
2921         zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2922         zc.zc_cookie = infd;
2923         zc.zc_guid = flags->force;
2924         if (flags->verbose) {
2925                 (void) printf("%s %s stream of %s into %s\n",
2926                     flags->dryrun ? "would receive" : "receiving",
2927                     drrb->drr_fromguid ? "incremental" : "full",
2928                     drrb->drr_toname, zc.zc_value);
2929                 (void) fflush(stdout);
2930         }
2931
2932         if (flags->dryrun) {
2933                 zcmd_free_nvlists(&zc);
2934                 return (recv_skip(hdl, infd, flags->byteswap));
2935         }
2936
2937         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2938         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2939         zc.zc_cleanup_fd = cleanup_fd;
2940         zc.zc_action_handle = *action_handlep;
2941
2942         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2943         ioctl_errno = errno;
2944         prop_errflags = (zprop_errflags_t)zc.zc_obj;
2945
2946         if (err == 0) {
2947                 nvlist_t *prop_errors;
2948                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2949                     zc.zc_nvlist_dst_size, &prop_errors, 0));
2950
2951                 nvpair_t *prop_err = NULL;
2952
2953                 while ((prop_err = nvlist_next_nvpair(prop_errors,
2954                     prop_err)) != NULL) {
2955                         char tbuf[1024];
2956                         zfs_prop_t prop;
2957                         int intval;
2958
2959                         prop = zfs_name_to_prop(nvpair_name(prop_err));
2960                         (void) nvpair_value_int32(prop_err, &intval);
2961                         if (strcmp(nvpair_name(prop_err),
2962                             ZPROP_N_MORE_ERRORS) == 0) {
2963                                 trunc_prop_errs(intval);
2964                                 break;
2965                         } else {
2966                                 (void) snprintf(tbuf, sizeof (tbuf),
2967                                     dgettext(TEXT_DOMAIN,
2968                                     "cannot receive %s property on %s"),
2969                                     nvpair_name(prop_err), zc.zc_name);
2970                                 zfs_setprop_error(hdl, prop, intval, tbuf);
2971                         }
2972                 }
2973                 nvlist_free(prop_errors);
2974         }
2975
2976         zc.zc_nvlist_dst = 0;
2977         zc.zc_nvlist_dst_size = 0;
2978         zcmd_free_nvlists(&zc);
2979
2980         if (err == 0 && snapprops_nvlist) {
2981                 zfs_cmd_t zc2 = { 0 };
2982
2983                 (void) strcpy(zc2.zc_name, zc.zc_value);
2984                 zc2.zc_cookie = B_TRUE; /* received */
2985                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
2986                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
2987                         zcmd_free_nvlists(&zc2);
2988                 }
2989         }
2990
2991         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
2992                 /*
2993                  * It may be that this snapshot already exists,
2994                  * in which case we want to consume & ignore it
2995                  * rather than failing.
2996                  */
2997                 avl_tree_t *local_avl;
2998                 nvlist_t *local_nv, *fs;
2999                 cp = strchr(zc.zc_value, '@');
3000
3001                 /*
3002                  * XXX Do this faster by just iterating over snaps in
3003                  * this fs.  Also if zc_value does not exist, we will
3004                  * get a strange "does not exist" error message.
3005                  */
3006                 *cp = '\0';
3007                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3008                     &local_nv, &local_avl) == 0) {
3009                         *cp = '@';
3010                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3011                         fsavl_destroy(local_avl);
3012                         nvlist_free(local_nv);
3013
3014                         if (fs != NULL) {
3015                                 if (flags->verbose) {
3016                                         (void) printf("snap %s already exists; "
3017                                             "ignoring\n", zc.zc_value);
3018                                 }
3019                                 err = ioctl_err = recv_skip(hdl, infd,
3020                                     flags->byteswap);
3021                         }
3022                 }
3023                 *cp = '@';
3024         }
3025
3026         if (ioctl_err != 0) {
3027                 switch (ioctl_errno) {
3028                 case ENODEV:
3029                         cp = strchr(zc.zc_value, '@');
3030                         *cp = '\0';
3031                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3032                             "most recent snapshot of %s does not\n"
3033                             "match incremental source"), zc.zc_value);
3034                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3035                         *cp = '@';
3036                         break;
3037                 case ETXTBSY:
3038                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3039                             "destination %s has been modified\n"
3040                             "since most recent snapshot"), zc.zc_name);
3041                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3042                         break;
3043                 case EEXIST:
3044                         cp = strchr(zc.zc_value, '@');
3045                         if (newfs) {
3046                                 /* it's the containing fs that exists */
3047                                 *cp = '\0';
3048                         }
3049                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3050                             "destination already exists"));
3051                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3052                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3053                             zc.zc_value);
3054                         *cp = '@';
3055                         break;
3056                 case EINVAL:
3057                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3058                         break;
3059                 case ECKSUM:
3060                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3061                             "invalid stream (checksum mismatch)"));
3062                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3063                         break;
3064                 case ENOTSUP:
3065                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3066                             "pool must be upgraded to receive this stream."));
3067                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3068                         break;
3069                 case EDQUOT:
3070                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3071                             "destination %s space quota exceeded"), zc.zc_name);
3072                         (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3073                         break;
3074                 default:
3075                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3076                 }
3077         }
3078
3079         /*
3080          * Mount the target filesystem (if created).  Also mount any
3081          * children of the target filesystem if we did a replication
3082          * receive (indicated by stream_avl being non-NULL).
3083          */
3084         cp = strchr(zc.zc_value, '@');
3085         if (cp && (ioctl_err == 0 || !newfs)) {
3086                 zfs_handle_t *h;
3087
3088                 *cp = '\0';
3089                 h = zfs_open(hdl, zc.zc_value,
3090                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3091                 if (h != NULL) {
3092                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
3093                                 *cp = '@';
3094                         } else if (newfs || stream_avl) {
3095                                 /*
3096                                  * Track the first/top of hierarchy fs,
3097                                  * for mounting and sharing later.
3098                                  */
3099                                 if (top_zfs && *top_zfs == NULL)
3100                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
3101                         }
3102                         zfs_close(h);
3103                 }
3104                 *cp = '@';
3105         }
3106
3107         if (clp) {
3108                 if (!flags->nomount)
3109                         err |= changelist_postfix(clp);
3110                 changelist_free(clp);
3111         }
3112
3113         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3114                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3115                     "failed to clear unreceived properties on %s"),
3116                     zc.zc_name);
3117                 (void) fprintf(stderr, "\n");
3118         }
3119         if (prop_errflags & ZPROP_ERR_NORESTORE) {
3120                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3121                     "failed to restore original properties on %s"),
3122                     zc.zc_name);
3123                 (void) fprintf(stderr, "\n");
3124         }
3125
3126         if (err || ioctl_err)
3127                 return (-1);
3128
3129         *action_handlep = zc.zc_action_handle;
3130
3131         if (flags->verbose) {
3132                 char buf1[64];
3133                 char buf2[64];
3134                 uint64_t bytes = zc.zc_cookie;
3135                 time_t delta = time(NULL) - begin_time;
3136                 if (delta == 0)
3137                         delta = 1;
3138                 zfs_nicenum(bytes, buf1, sizeof (buf1));
3139                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3140
3141                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3142                     buf1, delta, buf2);
3143         }
3144
3145         return (0);
3146 }
3147
3148 static int
3149 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3150     int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl,
3151     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
3152 {
3153         int err;
3154         dmu_replay_record_t drr, drr_noswap;
3155         struct drr_begin *drrb = &drr.drr_u.drr_begin;
3156         char errbuf[1024];
3157         zio_cksum_t zcksum = { 0 };
3158         uint64_t featureflags;
3159         int hdrtype;
3160
3161         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3162             "cannot receive"));
3163
3164         if (flags->isprefix &&
3165             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3166                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3167                     "(%s) does not exist"), tosnap);
3168                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3169         }
3170
3171         /* read in the BEGIN record */
3172         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3173             &zcksum)))
3174                 return (err);
3175
3176         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3177                 /* It's the double end record at the end of a package */
3178                 return (ENODATA);
3179         }
3180
3181         /* the kernel needs the non-byteswapped begin record */
3182         drr_noswap = drr;
3183
3184         flags->byteswap = B_FALSE;
3185         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3186                 /*
3187                  * We computed the checksum in the wrong byteorder in
3188                  * recv_read() above; do it again correctly.
3189                  */
3190                 bzero(&zcksum, sizeof (zio_cksum_t));
3191                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3192                 flags->byteswap = B_TRUE;
3193
3194                 drr.drr_type = BSWAP_32(drr.drr_type);
3195                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3196                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3197                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3198                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3199                 drrb->drr_type = BSWAP_32(drrb->drr_type);
3200                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3201                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3202                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3203         }
3204
3205         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3206                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3207                     "stream (bad magic number)"));
3208                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3209         }
3210
3211         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3212         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3213
3214         if (!DMU_STREAM_SUPPORTED(featureflags) ||
3215             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3216                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3217                     "stream has unsupported feature, feature flags = %lx"),
3218                     featureflags);
3219                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3220         }
3221
3222         if (strchr(drrb->drr_toname, '@') == NULL) {
3223                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3224                     "stream (bad snapshot name)"));
3225                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3226         }
3227
3228         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3229                 char nonpackage_sendfs[ZFS_MAXNAMELEN];
3230                 if (sendfs == NULL) {
3231                         /*
3232                          * We were not called from zfs_receive_package(). Get
3233                          * the fs specified by 'zfs send'.
3234                          */
3235                         char *cp;
3236                         (void) strlcpy(nonpackage_sendfs,
3237                             drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
3238                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3239                                 *cp = '\0';
3240                         sendfs = nonpackage_sendfs;
3241                 }
3242                 return (zfs_receive_one(hdl, infd, tosnap, flags,
3243                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl,
3244                     top_zfs, cleanup_fd, action_handlep));
3245         } else {
3246                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3247                     DMU_COMPOUNDSTREAM);
3248                 return (zfs_receive_package(hdl, infd, tosnap, flags,
3249                     &drr, &zcksum, top_zfs, cleanup_fd, action_handlep));
3250         }
3251 }
3252
3253 /*
3254  * Restores a backup of tosnap from the file descriptor specified by infd.
3255  * Return 0 on total success, -2 if some things couldn't be
3256  * destroyed/renamed/promoted, -1 if some things couldn't be received.
3257  * (-1 will override -2).
3258  */
3259 int
3260 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3261     int infd, avl_tree_t *stream_avl)
3262 {
3263         char *top_zfs = NULL;
3264         int err;
3265         int cleanup_fd;
3266         uint64_t action_handle = 0;
3267
3268         cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3269         VERIFY(cleanup_fd >= 0);
3270
3271         err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL,
3272             stream_avl, &top_zfs, cleanup_fd, &action_handle);
3273
3274         VERIFY(0 == close(cleanup_fd));
3275
3276         if (err == 0 && !flags->nomount && top_zfs) {
3277                 zfs_handle_t *zhp;
3278                 prop_changelist_t *clp;
3279
3280                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3281                 if (zhp != NULL) {
3282                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3283                             CL_GATHER_MOUNT_ALWAYS, 0);
3284                         zfs_close(zhp);
3285                         if (clp != NULL) {
3286                                 /* mount and share received datasets */
3287                                 err = changelist_postfix(clp);
3288                                 changelist_free(clp);
3289                         }
3290                 }
3291                 if (zhp == NULL || clp == NULL || err)
3292                         err = -1;
3293         }
3294         if (top_zfs)
3295                 free(top_zfs);
3296
3297         return (err);
3298 }