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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2012 Pawel Jakub Dawidek <pawel@dawidek.net>.
  27  * All rights reserved.
  28  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  29  */
  30
  31 #include <assert.h>
  32 #include <ctype.h>
  33 #include <errno.h>
  34 #include <libintl.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <strings.h>
  38 #include <unistd.h>
  39 #include <stddef.h>
  40 #include <fcntl.h>
  41 #include <sys/param.h>
  42 #include <sys/mount.h>
  43 #include <pthread.h>
  44 #include <umem.h>
  45 #include <time.h>
  46
  47 #include <libzfs.h>
  48 #include <libzfs_core.h>
  49
  50 #include "zfs_namecheck.h"
  51 #include "zfs_prop.h"
  52 #include "zfs_fletcher.h"
  53 #include "libzfs_impl.h"
  54 #include <sha2.h>
  55 #include <sys/zio_checksum.h>
  56 #include <sys/ddt.h>
  57
  58 #ifdef __FreeBSD__
  59 extern int zfs_ioctl_version;
  60 #endif
  61
  62 /* in libzfs_dataset.c */
  63 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  64 /* We need to use something for ENODATA. */
  65 #define ENODATA EIDRM
  66
  67 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t *,
  68     int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *);
  69
  70 static const zio_cksum_t zero_cksum = { 0 };
  71
  72 typedef struct dedup_arg {
  73         int     inputfd;
  74         int     outputfd;
  75         libzfs_handle_t  *dedup_hdl;
  76 } dedup_arg_t;
  77
  78 typedef struct progress_arg {
  79         zfs_handle_t *pa_zhp;
  80         int pa_fd;
  81         boolean_t pa_parsable;
  82 } progress_arg_t;
  83
  84 typedef struct dataref {
  85         uint64_t ref_guid;
  86         uint64_t ref_object;
  87         uint64_t ref_offset;
  88 } dataref_t;
  89
  90 typedef struct dedup_entry {
  91         struct dedup_entry      *dde_next;
  92         zio_cksum_t dde_chksum;
  93         uint64_t dde_prop;
  94         dataref_t dde_ref;
  95 } dedup_entry_t;
  96
  97 #define MAX_DDT_PHYSMEM_PERCENT         20
  98 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
  99
 100 typedef struct dedup_table {
 101         dedup_entry_t   **dedup_hash_array;
 102         umem_cache_t    *ddecache;
 103         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
 104         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
 105         uint64_t        ddt_count;
 106         int             numhashbits;
 107         boolean_t       ddt_full;
 108 } dedup_table_t;
 109
 110 static int
 111 high_order_bit(uint64_t n)
 112 {
 113         int count;
 114
 115         for (count = 0; n != 0; count++)
 116                 n >>= 1;
 117         return (count);
 118 }
 119
 120 static size_t
 121 ssread(void *buf, size_t len, FILE *stream)
 122 {
 123         size_t outlen;
 124
 125         if ((outlen = fread(buf, len, 1, stream)) == 0)
 126                 return (0);
 127
 128         return (outlen);
 129 }
 130
 131 static void
 132 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 133     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 134 {
 135         dedup_entry_t   *dde;
 136
 137         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 138                 if (ddt->ddt_full == B_FALSE) {
 139                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 140                             "Dedup table full.  Deduplication will continue "
 141                             "with existing table entries"));
 142                         ddt->ddt_full = B_TRUE;
 143                 }
 144                 return;
 145         }
 146
 147         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 148             != NULL) {
 149                 assert(*ddepp == NULL);
 150                 dde->dde_next = NULL;
 151                 dde->dde_chksum = *cs;
 152                 dde->dde_prop = prop;
 153                 dde->dde_ref = *dr;
 154                 *ddepp = dde;
 155                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 156                 ddt->ddt_count++;
 157         }
 158 }
 159
 160 /*
 161  * Using the specified dedup table, do a lookup for an entry with
 162  * the checksum cs.  If found, return the block's reference info
 163  * in *dr. Otherwise, insert a new entry in the dedup table, using
 164  * the reference information specified by *dr.
 165  *
 166  * return value:  true - entry was found
 167  *                false - entry was not found
 168  */
 169 static boolean_t
 170 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 171     uint64_t prop, dataref_t *dr)
 172 {
 173         uint32_t hashcode;
 174         dedup_entry_t **ddepp;
 175
 176         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 177
 178         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 179             ddepp = &((*ddepp)->dde_next)) {
 180                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 181                     (*ddepp)->dde_prop == prop) {
 182                         *dr = (*ddepp)->dde_ref;
 183                         return (B_TRUE);
 184                 }
 185         }
 186         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 187         return (B_FALSE);
 188 }
 189
 190 static int
 191 cksum_and_write(const void *buf, uint64_t len, zio_cksum_t *zc, int outfd)
 192 {
 193         fletcher_4_incremental_native(buf, len, zc);
 194         return (write(outfd, buf, len));
 195 }
 196
 197 /*
 198  * This function is started in a separate thread when the dedup option
 199  * has been requested.  The main send thread determines the list of
 200  * snapshots to be included in the send stream and makes the ioctl calls
 201  * for each one.  But instead of having the ioctl send the output to the
 202  * the output fd specified by the caller of zfs_send()), the
 203  * ioctl is told to direct the output to a pipe, which is read by the
 204  * alternate thread running THIS function.  This function does the
 205  * dedup'ing by:
 206  *  1. building a dedup table (the DDT)
 207  *  2. doing checksums on each data block and inserting a record in the DDT
 208  *  3. looking for matching checksums, and
 209  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 210  *      a duplicate block is found.
 211  * The output of this function then goes to the output fd requested
 212  * by the caller of zfs_send().
 213  */
 214 static void *
 215 cksummer(void *arg)
 216 {
 217         dedup_arg_t *dda = arg;
 218         char *buf = malloc(1<<20);
 219         dmu_replay_record_t thedrr;
 220         dmu_replay_record_t *drr = &thedrr;
 221         struct drr_begin *drrb = &thedrr.drr_u.drr_begin;
 222         struct drr_end *drre = &thedrr.drr_u.drr_end;
 223         struct drr_object *drro = &thedrr.drr_u.drr_object;
 224         struct drr_write *drrw = &thedrr.drr_u.drr_write;
 225         struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
 226         struct drr_write_embedded *drrwe = &thedrr.drr_u.drr_write_embedded;
 227         FILE *ofp;
 228         int outfd;
 229         dmu_replay_record_t wbr_drr = {0};
 230         struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref;
 231         dedup_table_t ddt;
 232         zio_cksum_t stream_cksum;
 233         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 234         uint64_t numbuckets;
 235
 236         ddt.max_ddt_size =
 237             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT)/100,
 238             SMALLEST_POSSIBLE_MAX_DDT_MB<<20);
 239
 240         numbuckets = ddt.max_ddt_size/(sizeof (dedup_entry_t));
 241
 242         /*
 243          * numbuckets must be a power of 2.  Increase number to
 244          * a power of 2 if necessary.
 245          */
 246         if (!ISP2(numbuckets))
 247                 numbuckets = 1 << high_order_bit(numbuckets);
 248
 249         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 250         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 251             NULL, NULL, NULL, NULL, NULL, 0);
 252         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 253         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 254         ddt.ddt_full = B_FALSE;
 255
 256         /* Initialize the write-by-reference block. */
 257         wbr_drr.drr_type = DRR_WRITE_BYREF;
 258         wbr_drr.drr_payloadlen = 0;
 259
 260         outfd = dda->outputfd;
 261         ofp = fdopen(dda->inputfd, "r");
 262         while (ssread(drr, sizeof (dmu_replay_record_t), ofp) != 0) {
 263
 264                 switch (drr->drr_type) {
 265                 case DRR_BEGIN:
 266                 {
 267                         int     fflags;
 268                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 269
 270                         /* set the DEDUP feature flag for this stream */
 271                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 272                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 273                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 274                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 275
 276                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 277                             &stream_cksum, outfd) == -1)
 278                                 goto out;
 279                         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 280                             DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
 281                                 int sz = drr->drr_payloadlen;
 282
 283                                 if (sz > 1<<20) {
 284                                         free(buf);
 285                                         buf = malloc(sz);
 286                                 }
 287                                 (void) ssread(buf, sz, ofp);
 288                                 if (ferror(stdin))
 289                                         perror("fread");
 290                                 if (cksum_and_write(buf, sz, &stream_cksum,
 291                                     outfd) == -1)
 292                                         goto out;
 293                         }
 294                         break;
 295                 }
 296
 297                 case DRR_END:
 298                 {
 299                         /* use the recalculated checksum */
 300                         ZIO_SET_CHECKSUM(&drre->drr_checksum,
 301                             stream_cksum.zc_word[0], stream_cksum.zc_word[1],
 302                             stream_cksum.zc_word[2], stream_cksum.zc_word[3]);
 303                         if ((write(outfd, drr,
 304                             sizeof (dmu_replay_record_t))) == -1)
 305                                 goto out;
 306                         break;
 307                 }
 308
 309                 case DRR_OBJECT:
 310                 {
 311                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 312                             &stream_cksum, outfd) == -1)
 313                                 goto out;
 314                         if (drro->drr_bonuslen > 0) {
 315                                 (void) ssread(buf,
 316                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 317                                     ofp);
 318                                 if (cksum_and_write(buf,
 319                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 320                                     &stream_cksum, outfd) == -1)
 321                                         goto out;
 322                         }
 323                         break;
 324                 }
 325
 326                 case DRR_SPILL:
 327                 {
 328                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 329                             &stream_cksum, outfd) == -1)
 330                                 goto out;
 331                         (void) ssread(buf, drrs->drr_length, ofp);
 332                         if (cksum_and_write(buf, drrs->drr_length,
 333                             &stream_cksum, outfd) == -1)
 334                                 goto out;
 335                         break;
 336                 }
 337
 338                 case DRR_FREEOBJECTS:
 339                 {
 340                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 341                             &stream_cksum, outfd) == -1)
 342                                 goto out;
 343                         break;
 344                 }
 345
 346                 case DRR_WRITE:
 347                 {
 348                         dataref_t       dataref;
 349
 350                         (void) ssread(buf, drrw->drr_length, ofp);
 351
 352                         /*
 353                          * Use the existing checksum if it's dedup-capable,
 354                          * else calculate a SHA256 checksum for it.
 355                          */
 356
 357                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 358                             zero_cksum) ||
 359                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 360                                 SHA256_CTX      ctx;
 361                                 zio_cksum_t     tmpsha256;
 362
 363                                 SHA256Init(&ctx);
 364                                 SHA256Update(&ctx, buf, drrw->drr_length);
 365                                 SHA256Final(&tmpsha256, &ctx);
 366                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 367                                     BE_64(tmpsha256.zc_word[0]);
 368                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 369                                     BE_64(tmpsha256.zc_word[1]);
 370                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 371                                     BE_64(tmpsha256.zc_word[2]);
 372                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 373                                     BE_64(tmpsha256.zc_word[3]);
 374                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 375                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 376                         }
 377
 378                         dataref.ref_guid = drrw->drr_toguid;
 379                         dataref.ref_object = drrw->drr_object;
 380                         dataref.ref_offset = drrw->drr_offset;
 381
 382                         if (ddt_update(dda->dedup_hdl, &ddt,
 383                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 384                             &dataref)) {
 385                                 /* block already present in stream */
 386                                 wbr_drrr->drr_object = drrw->drr_object;
 387                                 wbr_drrr->drr_offset = drrw->drr_offset;
 388                                 wbr_drrr->drr_length = drrw->drr_length;
 389                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 390                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 391                                 wbr_drrr->drr_refobject =
 392                                     dataref.ref_object;
 393                                 wbr_drrr->drr_refoffset =
 394                                     dataref.ref_offset;
 395
 396                                 wbr_drrr->drr_checksumtype =
 397                                     drrw->drr_checksumtype;
 398                                 wbr_drrr->drr_checksumflags =
 399                                     drrw->drr_checksumtype;
 400                                 wbr_drrr->drr_key.ddk_cksum =
 401                                     drrw->drr_key.ddk_cksum;
 402                                 wbr_drrr->drr_key.ddk_prop =
 403                                     drrw->drr_key.ddk_prop;
 404
 405                                 if (cksum_and_write(&wbr_drr,
 406                                     sizeof (dmu_replay_record_t), &stream_cksum,
 407                                     outfd) == -1)
 408                                         goto out;
 409                         } else {
 410                                 /* block not previously seen */
 411                                 if (cksum_and_write(drr,
 412                                     sizeof (dmu_replay_record_t), &stream_cksum,
 413                                     outfd) == -1)
 414                                         goto out;
 415                                 if (cksum_and_write(buf,
 416                                     drrw->drr_length,
 417                                     &stream_cksum, outfd) == -1)
 418                                         goto out;
 419                         }
 420                         break;
 421                 }
 422
 423                 case DRR_WRITE_EMBEDDED:
 424                 {
 425                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 426                             &stream_cksum, outfd) == -1)
 427                                 goto out;
 428                         (void) ssread(buf,
 429                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp);
 430                         if (cksum_and_write(buf,
 431                             P2ROUNDUP((uint64_t)drrwe->drr_psize, 8),
 432                             &stream_cksum, outfd) == -1)
 433                                 goto out;
 434                         break;
 435                 }
 436
 437                 case DRR_FREE:
 438                 {
 439                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 440                             &stream_cksum, outfd) == -1)
 441                                 goto out;
 442                         break;
 443                 }
 444
 445                 default:
 446                         (void) printf("INVALID record type 0x%x\n",
 447                             drr->drr_type);
 448                         /* should never happen, so assert */
 449                         assert(B_FALSE);
 450                 }
 451         }
 452 out:
 453         umem_cache_destroy(ddt.ddecache);
 454         free(ddt.dedup_hash_array);
 455         free(buf);
 456         (void) fclose(ofp);
 457
 458         return (NULL);
 459 }
 460
 461 /*
 462  * Routines for dealing with the AVL tree of fs-nvlists
 463  */
 464 typedef struct fsavl_node {
 465         avl_node_t fn_node;
 466         nvlist_t *fn_nvfs;
 467         char *fn_snapname;
 468         uint64_t fn_guid;
 469 } fsavl_node_t;
 470
 471 static int
 472 fsavl_compare(const void *arg1, const void *arg2)
 473 {
 474         const fsavl_node_t *fn1 = arg1;
 475         const fsavl_node_t *fn2 = arg2;
 476
 477         if (fn1->fn_guid > fn2->fn_guid)
 478                 return (+1);
 479         else if (fn1->fn_guid < fn2->fn_guid)
 480                 return (-1);
 481         else
 482                 return (0);
 483 }
 484
 485 /*
 486  * Given the GUID of a snapshot, find its containing filesystem and
 487  * (optionally) name.
 488  */
 489 static nvlist_t *
 490 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 491 {
 492         fsavl_node_t fn_find;
 493         fsavl_node_t *fn;
 494
 495         fn_find.fn_guid = snapguid;
 496
 497         fn = avl_find(avl, &fn_find, NULL);
 498         if (fn) {
 499                 if (snapname)
 500                         *snapname = fn->fn_snapname;
 501                 return (fn->fn_nvfs);
 502         }
 503         return (NULL);
 504 }
 505
 506 static void
 507 fsavl_destroy(avl_tree_t *avl)
 508 {
 509         fsavl_node_t *fn;
 510         void *cookie;
 511
 512         if (avl == NULL)
 513                 return;
 514
 515         cookie = NULL;
 516         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 517                 free(fn);
 518         avl_destroy(avl);
 519         free(avl);
 520 }
 521
 522 /*
 523  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 524  */
 525 static avl_tree_t *
 526 fsavl_create(nvlist_t *fss)
 527 {
 528         avl_tree_t *fsavl;
 529         nvpair_t *fselem = NULL;
 530
 531         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 532                 return (NULL);
 533
 534         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 535             offsetof(fsavl_node_t, fn_node));
 536
 537         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 538                 nvlist_t *nvfs, *snaps;
 539                 nvpair_t *snapelem = NULL;
 540
 541                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 542                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 543
 544                 while ((snapelem =
 545                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 546                         fsavl_node_t *fn;
 547                         uint64_t guid;
 548
 549                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 550                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 551                                 fsavl_destroy(fsavl);
 552                                 return (NULL);
 553                         }
 554                         fn->fn_nvfs = nvfs;
 555                         fn->fn_snapname = nvpair_name(snapelem);
 556                         fn->fn_guid = guid;
 557
 558                         /*
 559                          * Note: if there are multiple snaps with the
 560                          * same GUID, we ignore all but one.
 561                          */
 562                         if (avl_find(fsavl, fn, NULL) == NULL)
 563                                 avl_add(fsavl, fn);
 564                         else
 565                                 free(fn);
 566                 }
 567         }
 568
 569         return (fsavl);
 570 }
 571
 572 /*
 573  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 574  */
 575 typedef struct send_data {
 576         uint64_t parent_fromsnap_guid;
 577         nvlist_t *parent_snaps;
 578         nvlist_t *fss;
 579         nvlist_t *snapprops;
 580         const char *fromsnap;
 581         const char *tosnap;
 582         boolean_t recursive;
 583
 584         /*
 585          * The header nvlist is of the following format:
 586          * {
 587          *   "tosnap" -> string
 588          *   "fromsnap" -> string (if incremental)
 589          *   "fss" -> {
 590          *      id -> {
 591          *
 592          *       "name" -> string (full name; for debugging)
 593          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 594          *
 595          *       "props" -> { name -> value (only if set here) }
 596          *       "snaps" -> { name (lastname) -> number (guid) }
 597          *       "snapprops" -> { name (lastname) -> { name -> value } }
 598          *
 599          *       "origin" -> number (guid) (if clone)
 600          *       "sent" -> boolean (not on-disk)
 601          *      }
 602          *   }
 603          * }
 604          *
 605          */
 606 } send_data_t;
 607
 608 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 609
 610 static int
 611 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 612 {
 613         send_data_t *sd = arg;
 614         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 615         char *snapname;
 616         nvlist_t *nv;
 617
 618         snapname = strrchr(zhp->zfs_name, '@')+1;
 619
 620         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 621         /*
 622          * NB: if there is no fromsnap here (it's a newly created fs in
 623          * an incremental replication), we will substitute the tosnap.
 624          */
 625         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 626             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 627             strcmp(snapname, sd->tosnap) == 0)) {
 628                 sd->parent_fromsnap_guid = guid;
 629         }
 630
 631         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 632         send_iterate_prop(zhp, nv);
 633         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 634         nvlist_free(nv);
 635
 636         zfs_close(zhp);
 637         return (0);
 638 }
 639
 640 static void
 641 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 642 {
 643         nvpair_t *elem = NULL;
 644
 645         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 646                 char *propname = nvpair_name(elem);
 647                 zfs_prop_t prop = zfs_name_to_prop(propname);
 648                 nvlist_t *propnv;
 649
 650                 if (!zfs_prop_user(propname)) {
 651                         /*
 652                          * Realistically, this should never happen.  However,
 653                          * we want the ability to add DSL properties without
 654                          * needing to make incompatible version changes.  We
 655                          * need to ignore unknown properties to allow older
 656                          * software to still send datasets containing these
 657                          * properties, with the unknown properties elided.
 658                          */
 659                         if (prop == ZPROP_INVAL)
 660                                 continue;
 661
 662                         if (zfs_prop_readonly(prop))
 663                                 continue;
 664                 }
 665
 666                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 667                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 668                     prop == ZFS_PROP_REFQUOTA ||
 669                     prop == ZFS_PROP_REFRESERVATION) {
 670                         char *source;
 671                         uint64_t value;
 672                         verify(nvlist_lookup_uint64(propnv,
 673                             ZPROP_VALUE, &value) == 0);
 674                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 675                                 continue;
 676                         /*
 677                          * May have no source before SPA_VERSION_RECVD_PROPS,
 678                          * but is still modifiable.
 679                          */
 680                         if (nvlist_lookup_string(propnv,
 681                             ZPROP_SOURCE, &source) == 0) {
 682                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 683                                     (strcmp(source,
 684                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 685                                         continue;
 686                         }
 687                 } else {
 688                         char *source;
 689                         if (nvlist_lookup_string(propnv,
 690                             ZPROP_SOURCE, &source) != 0)
 691                                 continue;
 692                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 693                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 694                                 continue;
 695                 }
 696
 697                 if (zfs_prop_user(propname) ||
 698                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 699                         char *value;
 700                         verify(nvlist_lookup_string(propnv,
 701                             ZPROP_VALUE, &value) == 0);
 702                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 703                 } else {
 704                         uint64_t value;
 705                         verify(nvlist_lookup_uint64(propnv,
 706                             ZPROP_VALUE, &value) == 0);
 707                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 708                 }
 709         }
 710 }
 711
 712 /*
 713  * recursively generate nvlists describing datasets.  See comment
 714  * for the data structure send_data_t above for description of contents
 715  * of the nvlist.
 716  */
 717 static int
 718 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 719 {
 720         send_data_t *sd = arg;
 721         nvlist_t *nvfs, *nv;
 722         int rv = 0;
 723         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 724         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 725         char guidstring[64];
 726
 727         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 728         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 729         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 730             sd->parent_fromsnap_guid));
 731
 732         if (zhp->zfs_dmustats.dds_origin[0]) {
 733                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 734                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 735                 if (origin == NULL)
 736                         return (-1);
 737                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 738                     origin->zfs_dmustats.dds_guid));
 739         }
 740
 741         /* iterate over props */
 742         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 743         send_iterate_prop(zhp, nv);
 744         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 745         nvlist_free(nv);
 746
 747         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 748         sd->parent_fromsnap_guid = 0;
 749         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 750         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 751         (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd);
 752         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 753         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 754         nvlist_free(sd->parent_snaps);
 755         nvlist_free(sd->snapprops);
 756
 757         /* add this fs to nvlist */
 758         (void) snprintf(guidstring, sizeof (guidstring),
 759             "0x%llx", (longlong_t)guid);
 760         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 761         nvlist_free(nvfs);
 762
 763         /* iterate over children */
 764         if (sd->recursive)
 765                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 766
 767         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 768
 769         zfs_close(zhp);
 770         return (rv);
 771 }
 772
 773 static int
 774 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 775     const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
 776 {
 777         zfs_handle_t *zhp;
 778         send_data_t sd = { 0 };
 779         int error;
 780
 781         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 782         if (zhp == NULL)
 783                 return (EZFS_BADTYPE);
 784
 785         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 786         sd.fromsnap = fromsnap;
 787         sd.tosnap = tosnap;
 788         sd.recursive = recursive;
 789
 790         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 791                 nvlist_free(sd.fss);
 792                 if (avlp != NULL)
 793                         *avlp = NULL;
 794                 *nvlp = NULL;
 795                 return (error);
 796         }
 797
 798         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 799                 nvlist_free(sd.fss);
 800                 *nvlp = NULL;
 801                 return (EZFS_NOMEM);
 802         }
 803
 804         *nvlp = sd.fss;
 805         return (0);
 806 }
 807
 808 /*
 809  * Routines specific to "zfs send"
 810  */
 811 typedef struct send_dump_data {
 812         /* these are all just the short snapname (the part after the @) */
 813         const char *fromsnap;
 814         const char *tosnap;
 815         char prevsnap[ZFS_MAXNAMELEN];
 816         uint64_t prevsnap_obj;
 817         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 818         boolean_t verbose, dryrun, parsable, progress, embed_data;
 819         int outfd;
 820         boolean_t err;
 821         nvlist_t *fss;
 822         nvlist_t *snapholds;
 823         avl_tree_t *fsavl;
 824         snapfilter_cb_t *filter_cb;
 825         void *filter_cb_arg;
 826         nvlist_t *debugnv;
 827         char holdtag[ZFS_MAXNAMELEN];
 828         int cleanup_fd;
 829         uint64_t size;
 830 } send_dump_data_t;
 831
 832 static int
 833 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
 834     boolean_t fromorigin, uint64_t *sizep)
 835 {
 836         zfs_cmd_t zc = { 0 };
 837         libzfs_handle_t *hdl = zhp->zfs_hdl;
 838
 839         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 840         assert(fromsnap_obj == 0 || !fromorigin);
 841
 842         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 843         zc.zc_obj = fromorigin;
 844         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 845         zc.zc_fromobj = fromsnap_obj;
 846         zc.zc_guid = 1;  /* estimate flag */
 847
 848         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 849                 char errbuf[1024];
 850                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 851                     "warning: cannot estimate space for '%s'"), zhp->zfs_name);
 852
 853                 switch (errno) {
 854                 case EXDEV:
 855                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 856                             "not an earlier snapshot from the same fs"));
 857                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 858
 859                 case ENOENT:
 860                         if (zfs_dataset_exists(hdl, zc.zc_name,
 861                             ZFS_TYPE_SNAPSHOT)) {
 862                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 863                                     "incremental source (@%s) does not exist"),
 864                                     zc.zc_value);
 865                         }
 866                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 867
 868                 case EDQUOT:
 869                 case EFBIG:
 870                 case EIO:
 871                 case ENOLINK:
 872                 case ENOSPC:
 873                 case ENXIO:
 874                 case EPIPE:
 875                 case ERANGE:
 876                 case EFAULT:
 877                 case EROFS:
 878                         zfs_error_aux(hdl, strerror(errno));
 879                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 880
 881                 default:
 882                         return (zfs_standard_error(hdl, errno, errbuf));
 883                 }
 884         }
 885
 886         *sizep = zc.zc_objset_type;
 887
 888         return (0);
 889 }
 890
 891 /*
 892  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 893  * NULL) to the file descriptor specified by outfd.
 894  */
 895 static int
 896 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
 897     boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
 898     nvlist_t *debugnv)
 899 {
 900         zfs_cmd_t zc = { 0 };
 901         libzfs_handle_t *hdl = zhp->zfs_hdl;
 902         nvlist_t *thisdbg;
 903
 904         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 905         assert(fromsnap_obj == 0 || !fromorigin);
 906
 907         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 908         zc.zc_cookie = outfd;
 909         zc.zc_obj = fromorigin;
 910         zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 911         zc.zc_fromobj = fromsnap_obj;
 912         zc.zc_flags = flags;
 913
 914         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 915         if (fromsnap && fromsnap[0] != '\0') {
 916                 VERIFY(0 == nvlist_add_string(thisdbg,
 917                     "fromsnap", fromsnap));
 918         }
 919
 920         if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
 921                 char errbuf[1024];
 922                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 923                     "warning: cannot send '%s'"), zhp->zfs_name);
 924
 925                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
 926                 if (debugnv) {
 927                         VERIFY(0 == nvlist_add_nvlist(debugnv,
 928                             zhp->zfs_name, thisdbg));
 929                 }
 930                 nvlist_free(thisdbg);
 931
 932                 switch (errno) {
 933                 case EXDEV:
 934                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 935                             "not an earlier snapshot from the same fs"));
 936                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 937
 938                 case ENOENT:
 939                         if (zfs_dataset_exists(hdl, zc.zc_name,
 940                             ZFS_TYPE_SNAPSHOT)) {
 941                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 942                                     "incremental source (@%s) does not exist"),
 943                                     zc.zc_value);
 944                         }
 945                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 946
 947                 case EDQUOT:
 948                 case EFBIG:
 949                 case EIO:
 950                 case ENOLINK:
 951                 case ENOSPC:
 952 #ifdef sun
 953                 case ENOSTR:
 954 #endif
 955                 case ENXIO:
 956                 case EPIPE:
 957                 case ERANGE:
 958                 case EFAULT:
 959                 case EROFS:
 960                         zfs_error_aux(hdl, strerror(errno));
 961                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 962
 963                 default:
 964                         return (zfs_standard_error(hdl, errno, errbuf));
 965                 }
 966         }
 967
 968         if (debugnv)
 969                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
 970         nvlist_free(thisdbg);
 971
 972         return (0);
 973 }
 974
 975 static void
 976 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
 977 {
 978         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 979
 980         /*
 981          * zfs_send() only sets snapholds for sends that need them,
 982          * e.g. replication and doall.
 983          */
 984         if (sdd->snapholds == NULL)
 985                 return;
 986
 987         fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
 988 }
 989
 990 static void *
 991 send_progress_thread(void *arg)
 992 {
 993         progress_arg_t *pa = arg;
 994
 995         zfs_cmd_t zc = { 0 };
 996         zfs_handle_t *zhp = pa->pa_zhp;
 997         libzfs_handle_t *hdl = zhp->zfs_hdl;
 998         unsigned long long bytes;
 999         char buf[16];
1000
1001         time_t t;
1002         struct tm *tm;
1003
1004         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
1005         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1006
1007         if (!pa->pa_parsable)
1008                 (void) fprintf(stderr, "TIME        SENT   SNAPSHOT\n");
1009
1010         /*
1011          * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1012          */
1013         for (;;) {
1014                 (void) sleep(1);
1015
1016                 zc.zc_cookie = pa->pa_fd;
1017                 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1018                         return ((void *)-1);
1019
1020                 (void) time(&t);
1021                 tm = localtime(&t);
1022                 bytes = zc.zc_cookie;
1023
1024                 if (pa->pa_parsable) {
1025                         (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1026                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1027                             bytes, zhp->zfs_name);
1028                 } else {
1029                         zfs_nicenum(bytes, buf, sizeof (buf));
1030                         (void) fprintf(stderr, "%02d:%02d:%02d   %5s   %s\n",
1031                             tm->tm_hour, tm->tm_min, tm->tm_sec,
1032                             buf, zhp->zfs_name);
1033                 }
1034         }
1035 }
1036
1037 static int
1038 dump_snapshot(zfs_handle_t *zhp, void *arg)
1039 {
1040         send_dump_data_t *sdd = arg;
1041         progress_arg_t pa = { 0 };
1042         pthread_t tid;
1043         char *thissnap;
1044         int err;
1045         boolean_t isfromsnap, istosnap, fromorigin;
1046         boolean_t exclude = B_FALSE;
1047
1048         err = 0;
1049         thissnap = strchr(zhp->zfs_name, '@') + 1;
1050         isfromsnap = (sdd->fromsnap != NULL &&
1051             strcmp(sdd->fromsnap, thissnap) == 0);
1052
1053         if (!sdd->seenfrom && isfromsnap) {
1054                 gather_holds(zhp, sdd);
1055                 sdd->seenfrom = B_TRUE;
1056                 (void) strcpy(sdd->prevsnap, thissnap);
1057                 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1058                 zfs_close(zhp);
1059                 return (0);
1060         }
1061
1062         if (sdd->seento || !sdd->seenfrom) {
1063                 zfs_close(zhp);
1064                 return (0);
1065         }
1066
1067         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1068         if (istosnap)
1069                 sdd->seento = B_TRUE;
1070
1071         if (!sdd->doall && !isfromsnap && !istosnap) {
1072                 if (sdd->replicate) {
1073                         char *snapname;
1074                         nvlist_t *snapprops;
1075                         /*
1076                          * Filter out all intermediate snapshots except origin
1077                          * snapshots needed to replicate clones.
1078                          */
1079                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1080                             zhp->zfs_dmustats.dds_guid, &snapname);
1081
1082                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1083                             "snapprops", &snapprops));
1084                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1085                             thissnap, &snapprops));
1086                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
1087                 } else {
1088                         exclude = B_TRUE;
1089                 }
1090         }
1091
1092         /*
1093          * If a filter function exists, call it to determine whether
1094          * this snapshot will be sent.
1095          */
1096         if (exclude || (sdd->filter_cb != NULL &&
1097             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1098                 /*
1099                  * This snapshot is filtered out.  Don't send it, and don't
1100                  * set prevsnap_obj, so it will be as if this snapshot didn't
1101                  * exist, and the next accepted snapshot will be sent as
1102                  * an incremental from the last accepted one, or as the
1103                  * first (and full) snapshot in the case of a replication,
1104                  * non-incremental send.
1105                  */
1106                 zfs_close(zhp);
1107                 return (0);
1108         }
1109
1110         gather_holds(zhp, sdd);
1111         fromorigin = sdd->prevsnap[0] == '\0' &&
1112             (sdd->fromorigin || sdd->replicate);
1113
1114         if (sdd->verbose) {
1115                 uint64_t size;
1116                 err = estimate_ioctl(zhp, sdd->prevsnap_obj,
1117                     fromorigin, &size);
1118
1119                 if (sdd->parsable) {
1120                         if (sdd->prevsnap[0] != '\0') {
1121                                 (void) fprintf(stderr, "incremental\t%s\t%s",
1122                                     sdd->prevsnap, zhp->zfs_name);
1123                         } else {
1124                                 (void) fprintf(stderr, "full\t%s",
1125                                     zhp->zfs_name);
1126                         }
1127                 } else {
1128                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1129                             "send from @%s to %s"),
1130                             sdd->prevsnap, zhp->zfs_name);
1131                 }
1132                 if (err == 0) {
1133                         if (sdd->parsable) {
1134                                 (void) fprintf(stderr, "\t%llu\n",
1135                                     (longlong_t)size);
1136                         } else {
1137                                 char buf[16];
1138                                 zfs_nicenum(size, buf, sizeof (buf));
1139                                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1140                                     " estimated size is %s\n"), buf);
1141                         }
1142                         sdd->size += size;
1143                 } else {
1144                         (void) fprintf(stderr, "\n");
1145                 }
1146         }
1147
1148         if (!sdd->dryrun) {
1149                 /*
1150                  * If progress reporting is requested, spawn a new thread to
1151                  * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1152                  */
1153                 if (sdd->progress) {
1154                         pa.pa_zhp = zhp;
1155                         pa.pa_fd = sdd->outfd;
1156                         pa.pa_parsable = sdd->parsable;
1157
1158                         if (err = pthread_create(&tid, NULL,
1159                             send_progress_thread, &pa)) {
1160                                 zfs_close(zhp);
1161                                 return (err);
1162                         }
1163                 }
1164
1165                 enum lzc_send_flags flags = 0;
1166                 if (sdd->embed_data)
1167                         flags |= LZC_SEND_FLAG_EMBED_DATA;
1168
1169                 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1170                     fromorigin, sdd->outfd, flags, sdd->debugnv);
1171
1172                 if (sdd->progress) {
1173                         (void) pthread_cancel(tid);
1174                         (void) pthread_join(tid, NULL);
1175                 }
1176         }
1177
1178         (void) strcpy(sdd->prevsnap, thissnap);
1179         sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1180         zfs_close(zhp);
1181         return (err);
1182 }
1183
1184 static int
1185 dump_filesystem(zfs_handle_t *zhp, void *arg)
1186 {
1187         int rv = 0;
1188         send_dump_data_t *sdd = arg;
1189         boolean_t missingfrom = B_FALSE;
1190         zfs_cmd_t zc = { 0 };
1191
1192         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1193             zhp->zfs_name, sdd->tosnap);
1194         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1195                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1196                     "WARNING: could not send %s@%s: does not exist\n"),
1197                     zhp->zfs_name, sdd->tosnap);
1198                 sdd->err = B_TRUE;
1199                 return (0);
1200         }
1201
1202         if (sdd->replicate && sdd->fromsnap) {
1203                 /*
1204                  * If this fs does not have fromsnap, and we're doing
1205                  * recursive, we need to send a full stream from the
1206                  * beginning (or an incremental from the origin if this
1207                  * is a clone).  If we're doing non-recursive, then let
1208                  * them get the error.
1209                  */
1210                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1211                     zhp->zfs_name, sdd->fromsnap);
1212                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1213                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1214                         missingfrom = B_TRUE;
1215                 }
1216         }
1217
1218         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1219         sdd->prevsnap_obj = 0;
1220         if (sdd->fromsnap == NULL || missingfrom)
1221                 sdd->seenfrom = B_TRUE;
1222
1223         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1224         if (!sdd->seenfrom) {
1225                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1226                     "WARNING: could not send %s@%s:\n"
1227                     "incremental source (%s@%s) does not exist\n"),
1228                     zhp->zfs_name, sdd->tosnap,
1229                     zhp->zfs_name, sdd->fromsnap);
1230                 sdd->err = B_TRUE;
1231         } else if (!sdd->seento) {
1232                 if (sdd->fromsnap) {
1233                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1234                             "WARNING: could not send %s@%s:\n"
1235                             "incremental source (%s@%s) "
1236                             "is not earlier than it\n"),
1237                             zhp->zfs_name, sdd->tosnap,
1238                             zhp->zfs_name, sdd->fromsnap);
1239                 } else {
1240                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1241                             "WARNING: "
1242                             "could not send %s@%s: does not exist\n"),
1243                             zhp->zfs_name, sdd->tosnap);
1244                 }
1245                 sdd->err = B_TRUE;
1246         }
1247
1248         return (rv);
1249 }
1250
1251 static int
1252 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1253 {
1254         send_dump_data_t *sdd = arg;
1255         nvpair_t *fspair;
1256         boolean_t needagain, progress;
1257
1258         if (!sdd->replicate)
1259                 return (dump_filesystem(rzhp, sdd));
1260
1261         /* Mark the clone origin snapshots. */
1262         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1263             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1264                 nvlist_t *nvfs;
1265                 uint64_t origin_guid = 0;
1266
1267                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1268                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1269                 if (origin_guid != 0) {
1270                         char *snapname;
1271                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1272                             origin_guid, &snapname);
1273                         if (origin_nv != NULL) {
1274                                 nvlist_t *snapprops;
1275                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1276                                     "snapprops", &snapprops));
1277                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1278                                     snapname, &snapprops));
1279                                 VERIFY(0 == nvlist_add_boolean(
1280                                     snapprops, "is_clone_origin"));
1281                         }
1282                 }
1283         }
1284 again:
1285         needagain = progress = B_FALSE;
1286         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1287             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1288                 nvlist_t *fslist, *parent_nv;
1289                 char *fsname;
1290                 zfs_handle_t *zhp;
1291                 int err;
1292                 uint64_t origin_guid = 0;
1293                 uint64_t parent_guid = 0;
1294
1295                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1296                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1297                         continue;
1298
1299                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1300                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1301                 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1302                     &parent_guid);
1303
1304                 if (parent_guid != 0) {
1305                         parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1306                         if (!nvlist_exists(parent_nv, "sent")) {
1307                                 /* parent has not been sent; skip this one */
1308                                 needagain = B_TRUE;
1309                                 continue;
1310                         }
1311                 }
1312
1313                 if (origin_guid != 0) {
1314                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1315                             origin_guid, NULL);
1316                         if (origin_nv != NULL &&
1317                             !nvlist_exists(origin_nv, "sent")) {
1318                                 /*
1319                                  * origin has not been sent yet;
1320                                  * skip this clone.
1321                                  */
1322                                 needagain = B_TRUE;
1323                                 continue;
1324                         }
1325                 }
1326
1327                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1328                 if (zhp == NULL)
1329                         return (-1);
1330                 err = dump_filesystem(zhp, sdd);
1331                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1332                 progress = B_TRUE;
1333                 zfs_close(zhp);
1334                 if (err)
1335                         return (err);
1336         }
1337         if (needagain) {
1338                 assert(progress);
1339                 goto again;
1340         }
1341
1342         /* clean out the sent flags in case we reuse this fss */
1343         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1344             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1345                 nvlist_t *fslist;
1346
1347                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1348                 (void) nvlist_remove_all(fslist, "sent");
1349         }
1350
1351         return (0);
1352 }
1353
1354 /*
1355  * Generate a send stream for the dataset identified by the argument zhp.
1356  *
1357  * The content of the send stream is the snapshot identified by
1358  * 'tosnap'.  Incremental streams are requested in two ways:
1359  *     - from the snapshot identified by "fromsnap" (if non-null) or
1360  *     - from the origin of the dataset identified by zhp, which must
1361  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1362  *       is TRUE.
1363  *
1364  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1365  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1366  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1367  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1368  * case too. If "props" is set, send properties.
1369  */
1370 int
1371 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1372     sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1373     void *cb_arg, nvlist_t **debugnvp)
1374 {
1375         char errbuf[1024];
1376         send_dump_data_t sdd = { 0 };
1377         int err = 0;
1378         nvlist_t *fss = NULL;
1379         avl_tree_t *fsavl = NULL;
1380         static uint64_t holdseq;
1381         int spa_version;
1382         pthread_t tid = 0;
1383         int pipefd[2];
1384         dedup_arg_t dda = { 0 };
1385         int featureflags = 0;
1386
1387         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1388             "cannot send '%s'"), zhp->zfs_name);
1389
1390         if (fromsnap && fromsnap[0] == '\0') {
1391                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1392                     "zero-length incremental source"));
1393                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1394         }
1395
1396         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1397                 uint64_t version;
1398                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1399                 if (version >= ZPL_VERSION_SA) {
1400                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1401                 }
1402         }
1403
1404         if (flags->dedup && !flags->dryrun) {
1405                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1406                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1407                 if (err = pipe(pipefd)) {
1408                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1409                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1410                             errbuf));
1411                 }
1412                 dda.outputfd = outfd;
1413                 dda.inputfd = pipefd[1];
1414                 dda.dedup_hdl = zhp->zfs_hdl;
1415                 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1416                         (void) close(pipefd[0]);
1417                         (void) close(pipefd[1]);
1418                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1419                         return (zfs_error(zhp->zfs_hdl,
1420                             EZFS_THREADCREATEFAILED, errbuf));
1421                 }
1422         }
1423
1424         if (flags->replicate || flags->doall || flags->props) {
1425                 dmu_replay_record_t drr = { 0 };
1426                 char *packbuf = NULL;
1427                 size_t buflen = 0;
1428                 zio_cksum_t zc = { 0 };
1429
1430                 if (flags->replicate || flags->props) {
1431                         nvlist_t *hdrnv;
1432
1433                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1434                         if (fromsnap) {
1435                                 VERIFY(0 == nvlist_add_string(hdrnv,
1436                                     "fromsnap", fromsnap));
1437                         }
1438                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1439                         if (!flags->replicate) {
1440                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1441                                     "not_recursive"));
1442                         }
1443
1444                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1445                             fromsnap, tosnap, flags->replicate, &fss, &fsavl);
1446                         if (err)
1447                                 goto err_out;
1448                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1449                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1450                             NV_ENCODE_XDR, 0);
1451                         if (debugnvp)
1452                                 *debugnvp = hdrnv;
1453                         else
1454                                 nvlist_free(hdrnv);
1455                         if (err)
1456                                 goto stderr_out;
1457                 }
1458
1459                 if (!flags->dryrun) {
1460                         /* write first begin record */
1461                         drr.drr_type = DRR_BEGIN;
1462                         drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1463                         DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1464                             drr_versioninfo, DMU_COMPOUNDSTREAM);
1465                         DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1466                             drr_versioninfo, featureflags);
1467                         (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1468                             sizeof (drr.drr_u.drr_begin.drr_toname),
1469                             "%s@%s", zhp->zfs_name, tosnap);
1470                         drr.drr_payloadlen = buflen;
1471                         err = cksum_and_write(&drr, sizeof (drr), &zc, outfd);
1472
1473                         /* write header nvlist */
1474                         if (err != -1 && packbuf != NULL) {
1475                                 err = cksum_and_write(packbuf, buflen, &zc,
1476                                     outfd);
1477                         }
1478                         free(packbuf);
1479                         if (err == -1) {
1480                                 err = errno;
1481                                 goto stderr_out;
1482                         }
1483
1484                         /* write end record */
1485                         bzero(&drr, sizeof (drr));
1486                         drr.drr_type = DRR_END;
1487                         drr.drr_u.drr_end.drr_checksum = zc;
1488                         err = write(outfd, &drr, sizeof (drr));
1489                         if (err == -1) {
1490                                 err = errno;
1491                                 goto stderr_out;
1492                         }
1493
1494                         err = 0;
1495                 }
1496         }
1497
1498         /* dump each stream */
1499         sdd.fromsnap = fromsnap;
1500         sdd.tosnap = tosnap;
1501         if (tid != 0)
1502                 sdd.outfd = pipefd[0];
1503         else
1504                 sdd.outfd = outfd;
1505         sdd.replicate = flags->replicate;
1506         sdd.doall = flags->doall;
1507         sdd.fromorigin = flags->fromorigin;
1508         sdd.fss = fss;
1509         sdd.fsavl = fsavl;
1510         sdd.verbose = flags->verbose;
1511         sdd.parsable = flags->parsable;
1512         sdd.progress = flags->progress;
1513         sdd.dryrun = flags->dryrun;
1514         sdd.embed_data = flags->embed_data;
1515         sdd.filter_cb = filter_func;
1516         sdd.filter_cb_arg = cb_arg;
1517         if (debugnvp)
1518                 sdd.debugnv = *debugnvp;
1519
1520         /*
1521          * Some flags require that we place user holds on the datasets that are
1522          * being sent so they don't get destroyed during the send. We can skip
1523          * this step if the pool is imported read-only since the datasets cannot
1524          * be destroyed.
1525          */
1526         if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1527             ZPOOL_PROP_READONLY, NULL) &&
1528             zfs_spa_version(zhp, &spa_version) == 0 &&
1529             spa_version >= SPA_VERSION_USERREFS &&
1530             (flags->doall || flags->replicate)) {
1531                 ++holdseq;
1532                 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1533                     ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1534                 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1535                 if (sdd.cleanup_fd < 0) {
1536                         err = errno;
1537                         goto stderr_out;
1538                 }
1539                 sdd.snapholds = fnvlist_alloc();
1540         } else {
1541                 sdd.cleanup_fd = -1;
1542                 sdd.snapholds = NULL;
1543         }
1544         if (flags->verbose || sdd.snapholds != NULL) {
1545                 /*
1546                  * Do a verbose no-op dry run to get all the verbose output
1547                  * or to gather snapshot hold's before generating any data,
1548                  * then do a non-verbose real run to generate the streams.
1549                  */
1550                 sdd.dryrun = B_TRUE;
1551                 err = dump_filesystems(zhp, &sdd);
1552
1553                 if (err != 0)
1554                         goto stderr_out;
1555
1556                 if (flags->verbose) {
1557                         if (flags->parsable) {
1558                                 (void) fprintf(stderr, "size\t%llu\n",
1559                                     (longlong_t)sdd.size);
1560                         } else {
1561                                 char buf[16];
1562                                 zfs_nicenum(sdd.size, buf, sizeof (buf));
1563                                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1564                                     "total estimated size is %s\n"), buf);
1565                         }
1566                 }
1567
1568                 /* Ensure no snaps found is treated as an error. */
1569                 if (!sdd.seento) {
1570                         err = ENOENT;
1571                         goto err_out;
1572                 }
1573
1574                 /* Skip the second run if dryrun was requested. */
1575                 if (flags->dryrun)
1576                         goto err_out;
1577
1578                 if (sdd.snapholds != NULL) {
1579                         err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
1580                         if (err != 0)
1581                                 goto stderr_out;
1582
1583                         fnvlist_free(sdd.snapholds);
1584                         sdd.snapholds = NULL;
1585                 }
1586
1587                 sdd.dryrun = B_FALSE;
1588                 sdd.verbose = B_FALSE;
1589         }
1590
1591         err = dump_filesystems(zhp, &sdd);
1592         fsavl_destroy(fsavl);
1593         nvlist_free(fss);
1594
1595         /* Ensure no snaps found is treated as an error. */
1596         if (err == 0 && !sdd.seento)
1597                 err = ENOENT;
1598
1599         if (tid != 0) {
1600                 if (err != 0)
1601                         (void) pthread_cancel(tid);
1602                 (void) close(pipefd[0]);
1603                 (void) pthread_join(tid, NULL);
1604         }
1605
1606         if (sdd.cleanup_fd != -1) {
1607                 VERIFY(0 == close(sdd.cleanup_fd));
1608                 sdd.cleanup_fd = -1;
1609         }
1610
1611         if (!flags->dryrun && (flags->replicate || flags->doall ||
1612             flags->props)) {
1613                 /*
1614                  * write final end record.  NB: want to do this even if
1615                  * there was some error, because it might not be totally
1616                  * failed.
1617                  */
1618                 dmu_replay_record_t drr = { 0 };
1619                 drr.drr_type = DRR_END;
1620                 if (write(outfd, &drr, sizeof (drr)) == -1) {
1621                         return (zfs_standard_error(zhp->zfs_hdl,
1622                             errno, errbuf));
1623                 }
1624         }
1625
1626         return (err || sdd.err);
1627
1628 stderr_out:
1629         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1630 err_out:
1631         fsavl_destroy(fsavl);
1632         nvlist_free(fss);
1633         fnvlist_free(sdd.snapholds);
1634
1635         if (sdd.cleanup_fd != -1)
1636                 VERIFY(0 == close(sdd.cleanup_fd));
1637         if (tid != 0) {
1638                 (void) pthread_cancel(tid);
1639                 (void) close(pipefd[0]);
1640                 (void) pthread_join(tid, NULL);
1641         }
1642         return (err);
1643 }
1644
1645 int
1646 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd,
1647     enum lzc_send_flags flags)
1648 {
1649         int err;
1650         libzfs_handle_t *hdl = zhp->zfs_hdl;
1651
1652         char errbuf[1024];
1653         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1654             "warning: cannot send '%s'"), zhp->zfs_name);
1655
1656         err = lzc_send(zhp->zfs_name, from, fd, flags);
1657         if (err != 0) {
1658                 switch (errno) {
1659                 case EXDEV:
1660                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1661                             "not an earlier snapshot from the same fs"));
1662                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
1663
1664                 case ENOENT:
1665                 case ESRCH:
1666                         if (lzc_exists(zhp->zfs_name)) {
1667                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1668                                     "incremental source (%s) does not exist"),
1669                                     from);
1670                         }
1671                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
1672
1673                 case EBUSY:
1674                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1675                             "target is busy; if a filesystem, "
1676                             "it must not be mounted"));
1677                         return (zfs_error(hdl, EZFS_BUSY, errbuf));
1678
1679                 case EDQUOT:
1680                 case EFBIG:
1681                 case EIO:
1682                 case ENOLINK:
1683                 case ENOSPC:
1684 #ifdef illumos
1685                 case ENOSTR:
1686 #endif
1687                 case ENXIO:
1688                 case EPIPE:
1689                 case ERANGE:
1690                 case EFAULT:
1691                 case EROFS:
1692                         zfs_error_aux(hdl, strerror(errno));
1693                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1694
1695                 default:
1696                         return (zfs_standard_error(hdl, errno, errbuf));
1697                 }
1698         }
1699         return (err != 0);
1700 }
1701
1702 /*
1703  * Routines specific to "zfs recv"
1704  */
1705
1706 static int
1707 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1708     boolean_t byteswap, zio_cksum_t *zc)
1709 {
1710         char *cp = buf;
1711         int rv;
1712         int len = ilen;
1713
1714         do {
1715                 rv = read(fd, cp, len);
1716                 cp += rv;
1717                 len -= rv;
1718         } while (rv > 0);
1719
1720         if (rv < 0 || len != 0) {
1721                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1722                     "failed to read from stream"));
1723                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1724                     "cannot receive")));
1725         }
1726
1727         if (zc) {
1728                 if (byteswap)
1729                         fletcher_4_incremental_byteswap(buf, ilen, zc);
1730                 else
1731                         fletcher_4_incremental_native(buf, ilen, zc);
1732         }
1733         return (0);
1734 }
1735
1736 static int
1737 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1738     boolean_t byteswap, zio_cksum_t *zc)
1739 {
1740         char *buf;
1741         int err;
1742
1743         buf = zfs_alloc(hdl, len);
1744         if (buf == NULL)
1745                 return (ENOMEM);
1746
1747         err = recv_read(hdl, fd, buf, len, byteswap, zc);
1748         if (err != 0) {
1749                 free(buf);
1750                 return (err);
1751         }
1752
1753         err = nvlist_unpack(buf, len, nvp, 0);
1754         free(buf);
1755         if (err != 0) {
1756                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1757                     "stream (malformed nvlist)"));
1758                 return (EINVAL);
1759         }
1760         return (0);
1761 }
1762
1763 static int
1764 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1765     int baselen, char *newname, recvflags_t *flags)
1766 {
1767         static int seq;
1768         zfs_cmd_t zc = { 0 };
1769         int err;
1770         prop_changelist_t *clp;
1771         zfs_handle_t *zhp;
1772
1773         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1774         if (zhp == NULL)
1775                 return (-1);
1776         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1777             flags->force ? MS_FORCE : 0);
1778         zfs_close(zhp);
1779         if (clp == NULL)
1780                 return (-1);
1781         err = changelist_prefix(clp);
1782         if (err)
1783                 return (err);
1784
1785         zc.zc_objset_type = DMU_OST_ZFS;
1786         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1787
1788         if (tryname) {
1789                 (void) strcpy(newname, tryname);
1790
1791                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1792
1793                 if (flags->verbose) {
1794                         (void) printf("attempting rename %s to %s\n",
1795                             zc.zc_name, zc.zc_value);
1796                 }
1797                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1798                 if (err == 0)
1799                         changelist_rename(clp, name, tryname);
1800         } else {
1801                 err = ENOENT;
1802         }
1803
1804         if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
1805                 seq++;
1806
1807                 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u",
1808                     baselen, name, getpid(), seq);
1809                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1810
1811                 if (flags->verbose) {
1812                         (void) printf("failed - trying rename %s to %s\n",
1813                             zc.zc_name, zc.zc_value);
1814                 }
1815                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1816                 if (err == 0)
1817                         changelist_rename(clp, name, newname);
1818                 if (err && flags->verbose) {
1819                         (void) printf("failed (%u) - "
1820                             "will try again on next pass\n", errno);
1821                 }
1822                 err = EAGAIN;
1823         } else if (flags->verbose) {
1824                 if (err == 0)
1825                         (void) printf("success\n");
1826                 else
1827                         (void) printf("failed (%u)\n", errno);
1828         }
1829
1830         (void) changelist_postfix(clp);
1831         changelist_free(clp);
1832
1833         return (err);
1834 }
1835
1836 static int
1837 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1838     char *newname, recvflags_t *flags)
1839 {
1840         zfs_cmd_t zc = { 0 };
1841         int err = 0;
1842         prop_changelist_t *clp;
1843         zfs_handle_t *zhp;
1844         boolean_t defer = B_FALSE;
1845         int spa_version;
1846
1847         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1848         if (zhp == NULL)
1849                 return (-1);
1850         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1851             flags->force ? MS_FORCE : 0);
1852         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1853             zfs_spa_version(zhp, &spa_version) == 0 &&
1854             spa_version >= SPA_VERSION_USERREFS)
1855                 defer = B_TRUE;
1856         zfs_close(zhp);
1857         if (clp == NULL)
1858                 return (-1);
1859         err = changelist_prefix(clp);
1860         if (err)
1861                 return (err);
1862
1863         zc.zc_objset_type = DMU_OST_ZFS;
1864         zc.zc_defer_destroy = defer;
1865         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1866
1867         if (flags->verbose)
1868                 (void) printf("attempting destroy %s\n", zc.zc_name);
1869         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1870         if (err == 0) {
1871                 if (flags->verbose)
1872                         (void) printf("success\n");
1873                 changelist_remove(clp, zc.zc_name);
1874         }
1875
1876         (void) changelist_postfix(clp);
1877         changelist_free(clp);
1878
1879         /*
1880          * Deferred destroy might destroy the snapshot or only mark it to be
1881          * destroyed later, and it returns success in either case.
1882          */
1883         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1884             ZFS_TYPE_SNAPSHOT))) {
1885                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1886         }
1887
1888         return (err);
1889 }
1890
1891 typedef struct guid_to_name_data {
1892         uint64_t guid;
1893         char *name;
1894         char *skip;
1895 } guid_to_name_data_t;
1896
1897 static int
1898 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1899 {
1900         guid_to_name_data_t *gtnd = arg;
1901         int err;
1902
1903         if (gtnd->skip != NULL &&
1904             strcmp(zhp->zfs_name, gtnd->skip) == 0) {
1905                 return (0);
1906         }
1907
1908         if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1909                 (void) strcpy(gtnd->name, zhp->zfs_name);
1910                 zfs_close(zhp);
1911                 return (EEXIST);
1912         }
1913
1914         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1915         zfs_close(zhp);
1916         return (err);
1917 }
1918
1919 /*
1920  * Attempt to find the local dataset associated with this guid.  In the case of
1921  * multiple matches, we attempt to find the "best" match by searching
1922  * progressively larger portions of the hierarchy.  This allows one to send a
1923  * tree of datasets individually and guarantee that we will find the source
1924  * guid within that hierarchy, even if there are multiple matches elsewhere.
1925  */
1926 static int
1927 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1928     char *name)
1929 {
1930         /* exhaustive search all local snapshots */
1931         char pname[ZFS_MAXNAMELEN];
1932         guid_to_name_data_t gtnd;
1933         int err = 0;
1934         zfs_handle_t *zhp;
1935         char *cp;
1936
1937         gtnd.guid = guid;
1938         gtnd.name = name;
1939         gtnd.skip = NULL;
1940
1941         (void) strlcpy(pname, parent, sizeof (pname));
1942
1943         /*
1944          * Search progressively larger portions of the hierarchy.  This will
1945          * select the "most local" version of the origin snapshot in the case
1946          * that there are multiple matching snapshots in the system.
1947          */
1948         while ((cp = strrchr(pname, '/')) != NULL) {
1949
1950                 /* Chop off the last component and open the parent */
1951                 *cp = '\0';
1952                 zhp = make_dataset_handle(hdl, pname);
1953
1954                 if (zhp == NULL)
1955                         continue;
1956
1957                 err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1958                 zfs_close(zhp);
1959                 if (err == EEXIST)
1960                         return (0);
1961
1962                 /*
1963                  * Remember the dataset that we already searched, so we
1964                  * skip it next time through.
1965                  */
1966                 gtnd.skip = pname;
1967         }
1968
1969         return (ENOENT);
1970 }
1971
1972 /*
1973  * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
1974  * guid1 is after guid2.
1975  */
1976 static int
1977 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1978     uint64_t guid1, uint64_t guid2)
1979 {
1980         nvlist_t *nvfs;
1981         char *fsname, *snapname;
1982         char buf[ZFS_MAXNAMELEN];
1983         int rv;
1984         zfs_handle_t *guid1hdl, *guid2hdl;
1985         uint64_t create1, create2;
1986
1987         if (guid2 == 0)
1988                 return (0);
1989         if (guid1 == 0)
1990                 return (1);
1991
1992         nvfs = fsavl_find(avl, guid1, &snapname);
1993         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1994         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1995         guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1996         if (guid1hdl == NULL)
1997                 return (-1);
1998
1999         nvfs = fsavl_find(avl, guid2, &snapname);
2000         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2001         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2002         guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2003         if (guid2hdl == NULL) {
2004                 zfs_close(guid1hdl);
2005                 return (-1);
2006         }
2007
2008         create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
2009         create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
2010
2011         if (create1 < create2)
2012                 rv = -1;
2013         else if (create1 > create2)
2014                 rv = +1;
2015         else
2016                 rv = 0;
2017
2018         zfs_close(guid1hdl);
2019         zfs_close(guid2hdl);
2020
2021         return (rv);
2022 }
2023
2024 static int
2025 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2026     recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2027     nvlist_t *renamed)
2028 {
2029         nvlist_t *local_nv, *deleted = NULL;
2030         avl_tree_t *local_avl;
2031         nvpair_t *fselem, *nextfselem;
2032         char *fromsnap;
2033         char newname[ZFS_MAXNAMELEN];
2034         char guidname[32];
2035         int error;
2036         boolean_t needagain, progress, recursive;
2037         char *s1, *s2;
2038
2039         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2040
2041         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2042             ENOENT);
2043
2044         if (flags->dryrun)
2045                 return (0);
2046
2047 again:
2048         needagain = progress = B_FALSE;
2049
2050         VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0));
2051
2052         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2053             recursive, &local_nv, &local_avl)) != 0)
2054                 return (error);
2055
2056         /*
2057          * Process deletes and renames
2058          */
2059         for (fselem = nvlist_next_nvpair(local_nv, NULL);
2060             fselem; fselem = nextfselem) {
2061                 nvlist_t *nvfs, *snaps;
2062                 nvlist_t *stream_nvfs = NULL;
2063                 nvpair_t *snapelem, *nextsnapelem;
2064                 uint64_t fromguid = 0;
2065                 uint64_t originguid = 0;
2066                 uint64_t stream_originguid = 0;
2067                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2068                 char *fsname, *stream_fsname;
2069
2070                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2071
2072                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2073                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2074                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2075                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2076                     &parent_fromsnap_guid));
2077                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2078
2079                 /*
2080                  * First find the stream's fs, so we can check for
2081                  * a different origin (due to "zfs promote")
2082                  */
2083                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2084                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2085                         uint64_t thisguid;
2086
2087                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2088                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2089
2090                         if (stream_nvfs != NULL)
2091                                 break;
2092                 }
2093
2094                 /* check for promote */
2095                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2096                     &stream_originguid);
2097                 if (stream_nvfs && originguid != stream_originguid) {
2098                         switch (created_before(hdl, local_avl,
2099                             stream_originguid, originguid)) {
2100                         case 1: {
2101                                 /* promote it! */
2102                                 zfs_cmd_t zc = { 0 };
2103                                 nvlist_t *origin_nvfs;
2104                                 char *origin_fsname;
2105
2106                                 if (flags->verbose)
2107                                         (void) printf("promoting %s\n", fsname);
2108
2109                                 origin_nvfs = fsavl_find(local_avl, originguid,
2110                                     NULL);
2111                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2112                                     "name", &origin_fsname));
2113                                 (void) strlcpy(zc.zc_value, origin_fsname,
2114                                     sizeof (zc.zc_value));
2115                                 (void) strlcpy(zc.zc_name, fsname,
2116                                     sizeof (zc.zc_name));
2117                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2118                                 if (error == 0)
2119                                         progress = B_TRUE;
2120                                 break;
2121                         }
2122                         default:
2123                                 break;
2124                         case -1:
2125                                 fsavl_destroy(local_avl);
2126                                 nvlist_free(local_nv);
2127                                 return (-1);
2128                         }
2129                         /*
2130                          * We had/have the wrong origin, therefore our
2131                          * list of snapshots is wrong.  Need to handle
2132                          * them on the next pass.
2133                          */
2134                         needagain = B_TRUE;
2135                         continue;
2136                 }
2137
2138                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2139                     snapelem; snapelem = nextsnapelem) {
2140                         uint64_t thisguid;
2141                         char *stream_snapname;
2142                         nvlist_t *found, *props;
2143
2144                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2145
2146                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2147                         found = fsavl_find(stream_avl, thisguid,
2148                             &stream_snapname);
2149
2150                         /* check for delete */
2151                         if (found == NULL) {
2152                                 char name[ZFS_MAXNAMELEN];
2153
2154                                 if (!flags->force)
2155                                         continue;
2156
2157                                 (void) snprintf(name, sizeof (name), "%s@%s",
2158                                     fsname, nvpair_name(snapelem));
2159
2160                                 error = recv_destroy(hdl, name,
2161                                     strlen(fsname)+1, newname, flags);
2162                                 if (error)
2163                                         needagain = B_TRUE;
2164                                 else
2165                                         progress = B_TRUE;
2166                                 sprintf(guidname, "%lu", thisguid);
2167                                 nvlist_add_boolean(deleted, guidname);
2168                                 continue;
2169                         }
2170
2171                         stream_nvfs = found;
2172
2173                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2174                             &props) && 0 == nvlist_lookup_nvlist(props,
2175                             stream_snapname, &props)) {
2176                                 zfs_cmd_t zc = { 0 };
2177
2178                                 zc.zc_cookie = B_TRUE; /* received */
2179                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2180                                     "%s@%s", fsname, nvpair_name(snapelem));
2181                                 if (zcmd_write_src_nvlist(hdl, &zc,
2182                                     props) == 0) {
2183                                         (void) zfs_ioctl(hdl,
2184                                             ZFS_IOC_SET_PROP, &zc);
2185                                         zcmd_free_nvlists(&zc);
2186                                 }
2187                         }
2188
2189                         /* check for different snapname */
2190                         if (strcmp(nvpair_name(snapelem),
2191                             stream_snapname) != 0) {
2192                                 char name[ZFS_MAXNAMELEN];
2193                                 char tryname[ZFS_MAXNAMELEN];
2194
2195                                 (void) snprintf(name, sizeof (name), "%s@%s",
2196                                     fsname, nvpair_name(snapelem));
2197                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
2198                                     fsname, stream_snapname);
2199
2200                                 error = recv_rename(hdl, name, tryname,
2201                                     strlen(fsname)+1, newname, flags);
2202                                 if (error)
2203                                         needagain = B_TRUE;
2204                                 else
2205                                         progress = B_TRUE;
2206                         }
2207
2208                         if (strcmp(stream_snapname, fromsnap) == 0)
2209                                 fromguid = thisguid;
2210                 }
2211
2212                 /* check for delete */
2213                 if (stream_nvfs == NULL) {
2214                         if (!flags->force)
2215                                 continue;
2216
2217                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2218                             newname, flags);
2219                         if (error)
2220                                 needagain = B_TRUE;
2221                         else
2222                                 progress = B_TRUE;
2223                         sprintf(guidname, "%lu", parent_fromsnap_guid);
2224                         nvlist_add_boolean(deleted, guidname);
2225                         continue;
2226                 }
2227
2228                 if (fromguid == 0) {
2229                         if (flags->verbose) {
2230                                 (void) printf("local fs %s does not have "
2231                                     "fromsnap (%s in stream); must have "
2232                                     "been deleted locally; ignoring\n",
2233                                     fsname, fromsnap);
2234                         }
2235                         continue;
2236                 }
2237
2238                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2239                     "name", &stream_fsname));
2240                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2241                     "parentfromsnap", &stream_parent_fromsnap_guid));
2242
2243                 s1 = strrchr(fsname, '/');
2244                 s2 = strrchr(stream_fsname, '/');
2245
2246                 /*
2247                  * Check if we're going to rename based on parent guid change
2248                  * and the current parent guid was also deleted. If it was then
2249                  * rename will fail and is likely unneeded, so avoid this and
2250                  * force an early retry to determine the new
2251                  * parent_fromsnap_guid.
2252                  */
2253                 if (stream_parent_fromsnap_guid != 0 &&
2254                     parent_fromsnap_guid != 0 &&
2255                     stream_parent_fromsnap_guid != parent_fromsnap_guid) {
2256                         sprintf(guidname, "%lu", parent_fromsnap_guid);
2257                         if (nvlist_exists(deleted, guidname)) {
2258                                 progress = B_TRUE;
2259                                 needagain = B_TRUE;
2260                                 goto doagain;
2261                         }
2262                 }
2263
2264                 /*
2265                  * Check for rename. If the exact receive path is specified, it
2266                  * does not count as a rename, but we still need to check the
2267                  * datasets beneath it.
2268                  */
2269                 if ((stream_parent_fromsnap_guid != 0 &&
2270                     parent_fromsnap_guid != 0 &&
2271                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2272                     ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2273                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2274                         nvlist_t *parent;
2275                         char tryname[ZFS_MAXNAMELEN];
2276
2277                         parent = fsavl_find(local_avl,
2278                             stream_parent_fromsnap_guid, NULL);
2279                         /*
2280                          * NB: parent might not be found if we used the
2281                          * tosnap for stream_parent_fromsnap_guid,
2282                          * because the parent is a newly-created fs;
2283                          * we'll be able to rename it after we recv the
2284                          * new fs.
2285                          */
2286                         if (parent != NULL) {
2287                                 char *pname;
2288
2289                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
2290                                     &pname));
2291                                 (void) snprintf(tryname, sizeof (tryname),
2292                                     "%s%s", pname, strrchr(stream_fsname, '/'));
2293                         } else {
2294                                 tryname[0] = '\0';
2295                                 if (flags->verbose) {
2296                                         (void) printf("local fs %s new parent "
2297                                             "not found\n", fsname);
2298                                 }
2299                         }
2300
2301                         newname[0] = '\0';
2302
2303                         error = recv_rename(hdl, fsname, tryname,
2304                             strlen(tofs)+1, newname, flags);
2305
2306                         if (renamed != NULL && newname[0] != '\0') {
2307                                 VERIFY(0 == nvlist_add_boolean(renamed,
2308                                     newname));
2309                         }
2310
2311                         if (error)
2312                                 needagain = B_TRUE;
2313                         else
2314                                 progress = B_TRUE;
2315                 }
2316         }
2317
2318 doagain:
2319         fsavl_destroy(local_avl);
2320         nvlist_free(local_nv);
2321         nvlist_free(deleted);
2322
2323         if (needagain && progress) {
2324                 /* do another pass to fix up temporary names */
2325                 if (flags->verbose)
2326                         (void) printf("another pass:\n");
2327                 goto again;
2328         }
2329
2330         return (needagain);
2331 }
2332
2333 static int
2334 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2335     recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2336     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2337 {
2338         nvlist_t *stream_nv = NULL;
2339         avl_tree_t *stream_avl = NULL;
2340         char *fromsnap = NULL;
2341         char *cp;
2342         char tofs[ZFS_MAXNAMELEN];
2343         char sendfs[ZFS_MAXNAMELEN];
2344         char errbuf[1024];
2345         dmu_replay_record_t drre;
2346         int error;
2347         boolean_t anyerr = B_FALSE;
2348         boolean_t softerr = B_FALSE;
2349         boolean_t recursive;
2350
2351         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2352             "cannot receive"));
2353
2354         assert(drr->drr_type == DRR_BEGIN);
2355         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2356         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2357             DMU_COMPOUNDSTREAM);
2358
2359         /*
2360          * Read in the nvlist from the stream.
2361          */
2362         if (drr->drr_payloadlen != 0) {
2363                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2364                     &stream_nv, flags->byteswap, zc);
2365                 if (error) {
2366                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2367                         goto out;
2368                 }
2369         }
2370
2371         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2372             ENOENT);
2373
2374         if (recursive && strchr(destname, '@')) {
2375                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2376                     "cannot specify snapshot name for multi-snapshot stream"));
2377                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2378                 goto out;
2379         }
2380
2381         /*
2382          * Read in the end record and verify checksum.
2383          */
2384         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2385             flags->byteswap, NULL)))
2386                 goto out;
2387         if (flags->byteswap) {
2388                 drre.drr_type = BSWAP_32(drre.drr_type);
2389                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2390                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2391                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2392                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2393                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2394                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2395                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2396                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2397         }
2398         if (drre.drr_type != DRR_END) {
2399                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2400                 goto out;
2401         }
2402         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2403                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2404                     "incorrect header checksum"));
2405                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2406                 goto out;
2407         }
2408
2409         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2410
2411         if (drr->drr_payloadlen != 0) {
2412                 nvlist_t *stream_fss;
2413
2414                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2415                     &stream_fss));
2416                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2417                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2418                             "couldn't allocate avl tree"));
2419                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2420                         goto out;
2421                 }
2422
2423                 if (fromsnap != NULL) {
2424                         nvlist_t *renamed = NULL;
2425                         nvpair_t *pair = NULL;
2426
2427                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2428                         if (flags->isprefix) {
2429                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2430                                 int i;
2431
2432                                 if (flags->istail) {
2433                                         cp = strrchr(drrb->drr_toname, '/');
2434                                         if (cp == NULL) {
2435                                                 (void) strlcat(tofs, "/",
2436                                                     ZFS_MAXNAMELEN);
2437                                                 i = 0;
2438                                         } else {
2439                                                 i = (cp - drrb->drr_toname);
2440                                         }
2441                                 } else {
2442                                         i = strcspn(drrb->drr_toname, "/@");
2443                                 }
2444                                 /* zfs_receive_one() will create_parents() */
2445                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2446                                     ZFS_MAXNAMELEN);
2447                                 *strchr(tofs, '@') = '\0';
2448                         }
2449
2450                         if (recursive && !flags->dryrun && !flags->nomount) {
2451                                 VERIFY(0 == nvlist_alloc(&renamed,
2452                                     NV_UNIQUE_NAME, 0));
2453                         }
2454
2455                         softerr = recv_incremental_replication(hdl, tofs, flags,
2456                             stream_nv, stream_avl, renamed);
2457
2458                         /* Unmount renamed filesystems before receiving. */
2459                         while ((pair = nvlist_next_nvpair(renamed,
2460                             pair)) != NULL) {
2461                                 zfs_handle_t *zhp;
2462                                 prop_changelist_t *clp = NULL;
2463
2464                                 zhp = zfs_open(hdl, nvpair_name(pair),
2465                                     ZFS_TYPE_FILESYSTEM);
2466                                 if (zhp != NULL) {
2467                                         clp = changelist_gather(zhp,
2468                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2469                                         zfs_close(zhp);
2470                                         if (clp != NULL) {
2471                                                 softerr |=
2472                                                     changelist_prefix(clp);
2473                                                 changelist_free(clp);
2474                                         }
2475                                 }
2476                         }
2477
2478                         nvlist_free(renamed);
2479                 }
2480         }
2481
2482         /*
2483          * Get the fs specified by the first path in the stream (the top level
2484          * specified by 'zfs send') and pass it to each invocation of
2485          * zfs_receive_one().
2486          */
2487         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2488             ZFS_MAXNAMELEN);
2489         if ((cp = strchr(sendfs, '@')) != NULL)
2490                 *cp = '\0';
2491
2492         /* Finally, receive each contained stream */
2493         do {
2494                 /*
2495                  * we should figure out if it has a recoverable
2496                  * error, in which case do a recv_skip() and drive on.
2497                  * Note, if we fail due to already having this guid,
2498                  * zfs_receive_one() will take care of it (ie,
2499                  * recv_skip() and return 0).
2500                  */
2501                 error = zfs_receive_impl(hdl, destname, flags, fd,
2502                     sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2503                     action_handlep);
2504                 if (error == ENODATA) {
2505                         error = 0;
2506                         break;
2507                 }
2508                 anyerr |= error;
2509         } while (error == 0);
2510
2511         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2512                 /*
2513                  * Now that we have the fs's they sent us, try the
2514                  * renames again.
2515                  */
2516                 softerr = recv_incremental_replication(hdl, tofs, flags,
2517                     stream_nv, stream_avl, NULL);
2518         }
2519
2520 out:
2521         fsavl_destroy(stream_avl);
2522         if (stream_nv)
2523                 nvlist_free(stream_nv);
2524         if (softerr)
2525                 error = -2;
2526         if (anyerr)
2527                 error = -1;
2528         return (error);
2529 }
2530
2531 static void
2532 trunc_prop_errs(int truncated)
2533 {
2534         ASSERT(truncated != 0);
2535
2536         if (truncated == 1)
2537                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2538                     "1 more property could not be set\n"));
2539         else
2540                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2541                     "%d more properties could not be set\n"), truncated);
2542 }
2543
2544 static int
2545 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2546 {
2547         dmu_replay_record_t *drr;
2548         void *buf = malloc(1<<20);
2549         char errbuf[1024];
2550
2551         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2552             "cannot receive:"));
2553
2554         /* XXX would be great to use lseek if possible... */
2555         drr = buf;
2556
2557         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2558             byteswap, NULL) == 0) {
2559                 if (byteswap)
2560                         drr->drr_type = BSWAP_32(drr->drr_type);
2561
2562                 switch (drr->drr_type) {
2563                 case DRR_BEGIN:
2564                         /* NB: not to be used on v2 stream packages */
2565                         if (drr->drr_payloadlen != 0) {
2566                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2567                                     "invalid substream header"));
2568                                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2569                         }
2570                         break;
2571
2572                 case DRR_END:
2573                         free(buf);
2574                         return (0);
2575
2576                 case DRR_OBJECT:
2577                         if (byteswap) {
2578                                 drr->drr_u.drr_object.drr_bonuslen =
2579                                     BSWAP_32(drr->drr_u.drr_object.
2580                                     drr_bonuslen);
2581                         }
2582                         (void) recv_read(hdl, fd, buf,
2583                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2584                             B_FALSE, NULL);
2585                         break;
2586
2587                 case DRR_WRITE:
2588                         if (byteswap) {
2589                                 drr->drr_u.drr_write.drr_length =
2590                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
2591                         }
2592                         (void) recv_read(hdl, fd, buf,
2593                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2594                         break;
2595                 case DRR_SPILL:
2596                         if (byteswap) {
2597                                 drr->drr_u.drr_write.drr_length =
2598                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
2599                         }
2600                         (void) recv_read(hdl, fd, buf,
2601                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2602                         break;
2603                 case DRR_WRITE_EMBEDDED:
2604                         if (byteswap) {
2605                                 drr->drr_u.drr_write_embedded.drr_psize =
2606                                     BSWAP_32(drr->drr_u.drr_write_embedded.
2607                                     drr_psize);
2608                         }
2609                         (void) recv_read(hdl, fd, buf,
2610                             P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
2611                             8), B_FALSE, NULL);
2612                         break;
2613                 case DRR_WRITE_BYREF:
2614                 case DRR_FREEOBJECTS:
2615                 case DRR_FREE:
2616                         break;
2617
2618                 default:
2619                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2620                             "invalid record type"));
2621                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2622                 }
2623         }
2624
2625         free(buf);
2626         return (-1);
2627 }
2628
2629 /*
2630  * Restores a backup of tosnap from the file descriptor specified by infd.
2631  */
2632 static int
2633 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2634     recvflags_t *flags, dmu_replay_record_t *drr,
2635     dmu_replay_record_t *drr_noswap, const char *sendfs,
2636     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
2637     uint64_t *action_handlep)
2638 {
2639         zfs_cmd_t zc = { 0 };
2640         time_t begin_time;
2641         int ioctl_err, ioctl_errno, err;
2642         char *cp;
2643         struct drr_begin *drrb = &drr->drr_u.drr_begin;
2644         char errbuf[1024];
2645         char prop_errbuf[1024];
2646         const char *chopprefix;
2647         boolean_t newfs = B_FALSE;
2648         boolean_t stream_wantsnewfs;
2649         uint64_t parent_snapguid = 0;
2650         prop_changelist_t *clp = NULL;
2651         nvlist_t *snapprops_nvlist = NULL;
2652         zprop_errflags_t prop_errflags;
2653         boolean_t recursive;
2654
2655         begin_time = time(NULL);
2656
2657         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2658             "cannot receive"));
2659
2660         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2661             ENOENT);
2662
2663         if (stream_avl != NULL) {
2664                 char *snapname;
2665                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2666                     &snapname);
2667                 nvlist_t *props;
2668                 int ret;
2669
2670                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2671                     &parent_snapguid);
2672                 err = nvlist_lookup_nvlist(fs, "props", &props);
2673                 if (err)
2674                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2675
2676                 if (flags->canmountoff) {
2677                         VERIFY(0 == nvlist_add_uint64(props,
2678                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2679                 }
2680                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2681                 if (err)
2682                         nvlist_free(props);
2683
2684                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2685                         VERIFY(0 == nvlist_lookup_nvlist(props,
2686                             snapname, &snapprops_nvlist));
2687                 }
2688
2689                 if (ret != 0)
2690                         return (-1);
2691         }
2692
2693         cp = NULL;
2694
2695         /*
2696          * Determine how much of the snapshot name stored in the stream
2697          * we are going to tack on to the name they specified on the
2698          * command line, and how much we are going to chop off.
2699          *
2700          * If they specified a snapshot, chop the entire name stored in
2701          * the stream.
2702          */
2703         if (flags->istail) {
2704                 /*
2705                  * A filesystem was specified with -e. We want to tack on only
2706                  * the tail of the sent snapshot path.
2707                  */
2708                 if (strchr(tosnap, '@')) {
2709                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2710                             "argument - snapshot not allowed with -e"));
2711                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2712                 }
2713
2714                 chopprefix = strrchr(sendfs, '/');
2715
2716                 if (chopprefix == NULL) {
2717                         /*
2718                          * The tail is the poolname, so we need to
2719                          * prepend a path separator.
2720                          */
2721                         int len = strlen(drrb->drr_toname);
2722                         cp = malloc(len + 2);
2723                         cp[0] = '/';
2724                         (void) strcpy(&cp[1], drrb->drr_toname);
2725                         chopprefix = cp;
2726                 } else {
2727                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2728                 }
2729         } else if (flags->isprefix) {
2730                 /*
2731                  * A filesystem was specified with -d. We want to tack on
2732                  * everything but the first element of the sent snapshot path
2733                  * (all but the pool name).
2734                  */
2735                 if (strchr(tosnap, '@')) {
2736                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2737                             "argument - snapshot not allowed with -d"));
2738                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2739                 }
2740
2741                 chopprefix = strchr(drrb->drr_toname, '/');
2742                 if (chopprefix == NULL)
2743                         chopprefix = strchr(drrb->drr_toname, '@');
2744         } else if (strchr(tosnap, '@') == NULL) {
2745                 /*
2746                  * If a filesystem was specified without -d or -e, we want to
2747                  * tack on everything after the fs specified by 'zfs send'.
2748                  */
2749                 chopprefix = drrb->drr_toname + strlen(sendfs);
2750         } else {
2751                 /* A snapshot was specified as an exact path (no -d or -e). */
2752                 if (recursive) {
2753                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2754                             "cannot specify snapshot name for multi-snapshot "
2755                             "stream"));
2756                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2757                 }
2758                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2759         }
2760
2761         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2762         ASSERT(chopprefix > drrb->drr_toname);
2763         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2764         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2765             chopprefix[0] == '\0');
2766
2767         /*
2768          * Determine name of destination snapshot, store in zc_value.
2769          */
2770         (void) strcpy(zc.zc_value, tosnap);
2771         (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2772 #ifdef __FreeBSD__
2773         if (zfs_ioctl_version == ZFS_IOCVER_UNDEF)
2774                 zfs_ioctl_version = get_zfs_ioctl_version();
2775         /*
2776          * For forward compatibility hide tosnap in zc_value
2777          */
2778         if (zfs_ioctl_version < ZFS_IOCVER_LZC)
2779                 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap);
2780 #endif
2781         free(cp);
2782         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2783                 zcmd_free_nvlists(&zc);
2784                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2785         }
2786
2787         /*
2788          * Determine the name of the origin snapshot, store in zc_string.
2789          */
2790         if (drrb->drr_flags & DRR_FLAG_CLONE) {
2791                 if (guid_to_name(hdl, zc.zc_value,
2792                     drrb->drr_fromguid, zc.zc_string) != 0) {
2793                         zcmd_free_nvlists(&zc);
2794                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2795                             "local origin for clone %s does not exist"),
2796                             zc.zc_value);
2797                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2798                 }
2799                 if (flags->verbose)
2800                         (void) printf("found clone origin %s\n", zc.zc_string);
2801         }
2802
2803         stream_wantsnewfs = (drrb->drr_fromguid == 0 ||
2804             (drrb->drr_flags & DRR_FLAG_CLONE));
2805
2806         if (stream_wantsnewfs) {
2807                 /*
2808                  * if the parent fs does not exist, look for it based on
2809                  * the parent snap GUID
2810                  */
2811                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2812                     "cannot receive new filesystem stream"));
2813
2814                 (void) strcpy(zc.zc_name, zc.zc_value);
2815                 cp = strrchr(zc.zc_name, '/');
2816                 if (cp)
2817                         *cp = '\0';
2818                 if (cp &&
2819                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2820                         char suffix[ZFS_MAXNAMELEN];
2821                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2822                         if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
2823                             zc.zc_value) == 0) {
2824                                 *strchr(zc.zc_value, '@') = '\0';
2825                                 (void) strcat(zc.zc_value, suffix);
2826                         }
2827                 }
2828         } else {
2829                 /*
2830                  * if the fs does not exist, look for it based on the
2831                  * fromsnap GUID
2832                  */
2833                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2834                     "cannot receive incremental stream"));
2835
2836                 (void) strcpy(zc.zc_name, zc.zc_value);
2837                 *strchr(zc.zc_name, '@') = '\0';
2838
2839                 /*
2840                  * If the exact receive path was specified and this is the
2841                  * topmost path in the stream, then if the fs does not exist we
2842                  * should look no further.
2843                  */
2844                 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
2845                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2846                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2847                         char snap[ZFS_MAXNAMELEN];
2848                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
2849                         if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
2850                             zc.zc_value) == 0) {
2851                                 *strchr(zc.zc_value, '@') = '\0';
2852                                 (void) strcat(zc.zc_value, snap);
2853                         }
2854                 }
2855         }
2856
2857         (void) strcpy(zc.zc_name, zc.zc_value);
2858         *strchr(zc.zc_name, '@') = '\0';
2859
2860         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2861                 zfs_handle_t *zhp;
2862
2863                 /*
2864                  * Destination fs exists.  Therefore this should either
2865                  * be an incremental, or the stream specifies a new fs
2866                  * (full stream or clone) and they want us to blow it
2867                  * away (and have therefore specified -F and removed any
2868                  * snapshots).
2869                  */
2870                 if (stream_wantsnewfs) {
2871                         if (!flags->force) {
2872                                 zcmd_free_nvlists(&zc);
2873                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2874                                     "destination '%s' exists\n"
2875                                     "must specify -F to overwrite it"),
2876                                     zc.zc_name);
2877                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2878                         }
2879                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2880                             &zc) == 0) {
2881                                 zcmd_free_nvlists(&zc);
2882                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2883                                     "destination has snapshots (eg. %s)\n"
2884                                     "must destroy them to overwrite it"),
2885                                     zc.zc_name);
2886                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2887                         }
2888                 }
2889
2890                 if ((zhp = zfs_open(hdl, zc.zc_name,
2891                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2892                         zcmd_free_nvlists(&zc);
2893                         return (-1);
2894                 }
2895
2896                 if (stream_wantsnewfs &&
2897                     zhp->zfs_dmustats.dds_origin[0]) {
2898                         zcmd_free_nvlists(&zc);
2899                         zfs_close(zhp);
2900                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2901                             "destination '%s' is a clone\n"
2902                             "must destroy it to overwrite it"),
2903                             zc.zc_name);
2904                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2905                 }
2906
2907                 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2908                     stream_wantsnewfs) {
2909                         /* We can't do online recv in this case */
2910                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2911                         if (clp == NULL) {
2912                                 zfs_close(zhp);
2913                                 zcmd_free_nvlists(&zc);
2914                                 return (-1);
2915                         }
2916                         if (changelist_prefix(clp) != 0) {
2917                                 changelist_free(clp);
2918                                 zfs_close(zhp);
2919                                 zcmd_free_nvlists(&zc);
2920                                 return (-1);
2921                         }
2922                 }
2923                 zfs_close(zhp);
2924         } else {
2925                 /*
2926                  * Destination filesystem does not exist.  Therefore we better
2927                  * be creating a new filesystem (either from a full backup, or
2928                  * a clone).  It would therefore be invalid if the user
2929                  * specified only the pool name (i.e. if the destination name
2930                  * contained no slash character).
2931                  */
2932                 if (!stream_wantsnewfs ||
2933                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
2934                         zcmd_free_nvlists(&zc);
2935                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2936                             "destination '%s' does not exist"), zc.zc_name);
2937                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2938                 }
2939
2940                 /*
2941                  * Trim off the final dataset component so we perform the
2942                  * recvbackup ioctl to the filesystems's parent.
2943                  */
2944                 *cp = '\0';
2945
2946                 if (flags->isprefix && !flags->istail && !flags->dryrun &&
2947                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2948                         zcmd_free_nvlists(&zc);
2949                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2950                 }
2951
2952                 newfs = B_TRUE;
2953         }
2954
2955         zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2956         zc.zc_cookie = infd;
2957         zc.zc_guid = flags->force;
2958         if (flags->verbose) {
2959                 (void) printf("%s %s stream of %s into %s\n",
2960                     flags->dryrun ? "would receive" : "receiving",
2961                     drrb->drr_fromguid ? "incremental" : "full",
2962                     drrb->drr_toname, zc.zc_value);
2963                 (void) fflush(stdout);
2964         }
2965
2966         if (flags->dryrun) {
2967                 zcmd_free_nvlists(&zc);
2968                 return (recv_skip(hdl, infd, flags->byteswap));
2969         }
2970
2971         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2972         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2973         zc.zc_cleanup_fd = cleanup_fd;
2974         zc.zc_action_handle = *action_handlep;
2975
2976         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2977         ioctl_errno = errno;
2978         prop_errflags = (zprop_errflags_t)zc.zc_obj;
2979
2980         if (err == 0) {
2981                 nvlist_t *prop_errors;
2982                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2983                     zc.zc_nvlist_dst_size, &prop_errors, 0));
2984
2985                 nvpair_t *prop_err = NULL;
2986
2987                 while ((prop_err = nvlist_next_nvpair(prop_errors,
2988                     prop_err)) != NULL) {
2989                         char tbuf[1024];
2990                         zfs_prop_t prop;
2991                         int intval;
2992
2993                         prop = zfs_name_to_prop(nvpair_name(prop_err));
2994                         (void) nvpair_value_int32(prop_err, &intval);
2995                         if (strcmp(nvpair_name(prop_err),
2996                             ZPROP_N_MORE_ERRORS) == 0) {
2997                                 trunc_prop_errs(intval);
2998                                 break;
2999                         } else {
3000                                 (void) snprintf(tbuf, sizeof (tbuf),
3001                                     dgettext(TEXT_DOMAIN,
3002                                     "cannot receive %s property on %s"),
3003                                     nvpair_name(prop_err), zc.zc_name);
3004                                 zfs_setprop_error(hdl, prop, intval, tbuf);
3005                         }
3006                 }
3007                 nvlist_free(prop_errors);
3008         }
3009
3010         zc.zc_nvlist_dst = 0;
3011         zc.zc_nvlist_dst_size = 0;
3012         zcmd_free_nvlists(&zc);
3013
3014         if (err == 0 && snapprops_nvlist) {
3015                 zfs_cmd_t zc2 = { 0 };
3016
3017                 (void) strcpy(zc2.zc_name, zc.zc_value);
3018                 zc2.zc_cookie = B_TRUE; /* received */
3019                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
3020                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
3021                         zcmd_free_nvlists(&zc2);
3022                 }
3023         }
3024
3025         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
3026                 /*
3027                  * It may be that this snapshot already exists,
3028                  * in which case we want to consume & ignore it
3029                  * rather than failing.
3030                  */
3031                 avl_tree_t *local_avl;
3032                 nvlist_t *local_nv, *fs;
3033                 cp = strchr(zc.zc_value, '@');
3034
3035                 /*
3036                  * XXX Do this faster by just iterating over snaps in
3037                  * this fs.  Also if zc_value does not exist, we will
3038                  * get a strange "does not exist" error message.
3039                  */
3040                 *cp = '\0';
3041                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3042                     &local_nv, &local_avl) == 0) {
3043                         *cp = '@';
3044                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3045                         fsavl_destroy(local_avl);
3046                         nvlist_free(local_nv);
3047
3048                         if (fs != NULL) {
3049                                 if (flags->verbose) {
3050                                         (void) printf("snap %s already exists; "
3051                                             "ignoring\n", zc.zc_value);
3052                                 }
3053                                 err = ioctl_err = recv_skip(hdl, infd,
3054                                     flags->byteswap);
3055                         }
3056                 }
3057                 *cp = '@';
3058         }
3059
3060         if (ioctl_err != 0) {
3061                 switch (ioctl_errno) {
3062                 case ENODEV:
3063                         cp = strchr(zc.zc_value, '@');
3064                         *cp = '\0';
3065                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3066                             "most recent snapshot of %s does not\n"
3067                             "match incremental source"), zc.zc_value);
3068                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3069                         *cp = '@';
3070                         break;
3071                 case ETXTBSY:
3072                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3073                             "destination %s has been modified\n"
3074                             "since most recent snapshot"), zc.zc_name);
3075                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3076                         break;
3077                 case EEXIST:
3078                         cp = strchr(zc.zc_value, '@');
3079                         if (newfs) {
3080                                 /* it's the containing fs that exists */
3081                                 *cp = '\0';
3082                         }
3083                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3084                             "destination already exists"));
3085                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3086                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3087                             zc.zc_value);
3088                         *cp = '@';
3089                         break;
3090                 case EINVAL:
3091                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3092                         break;
3093                 case ECKSUM:
3094                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3095                             "invalid stream (checksum mismatch)"));
3096                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3097                         break;
3098                 case ENOTSUP:
3099                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3100                             "pool must be upgraded to receive this stream."));
3101                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3102                         break;
3103                 case EDQUOT:
3104                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3105                             "destination %s space quota exceeded"), zc.zc_name);
3106                         (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3107                         break;
3108                 default:
3109                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3110                 }
3111         }
3112
3113         /*
3114          * Mount the target filesystem (if created).  Also mount any
3115          * children of the target filesystem if we did a replication
3116          * receive (indicated by stream_avl being non-NULL).
3117          */
3118         cp = strchr(zc.zc_value, '@');
3119         if (cp && (ioctl_err == 0 || !newfs)) {
3120                 zfs_handle_t *h;
3121
3122                 *cp = '\0';
3123                 h = zfs_open(hdl, zc.zc_value,
3124                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3125                 if (h != NULL) {
3126                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
3127                                 *cp = '@';
3128                         } else if (newfs || stream_avl) {
3129                                 /*
3130                                  * Track the first/top of hierarchy fs,
3131                                  * for mounting and sharing later.
3132                                  */
3133                                 if (top_zfs && *top_zfs == NULL)
3134                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
3135                         }
3136                         zfs_close(h);
3137                 }
3138                 *cp = '@';
3139         }
3140
3141         if (clp) {
3142                 err |= changelist_postfix(clp);
3143                 changelist_free(clp);
3144         }
3145
3146         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3147                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3148                     "failed to clear unreceived properties on %s"),
3149                     zc.zc_name);
3150                 (void) fprintf(stderr, "\n");
3151         }
3152         if (prop_errflags & ZPROP_ERR_NORESTORE) {
3153                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3154                     "failed to restore original properties on %s"),
3155                     zc.zc_name);
3156                 (void) fprintf(stderr, "\n");
3157         }
3158
3159         if (err || ioctl_err)
3160                 return (-1);
3161
3162         *action_handlep = zc.zc_action_handle;
3163
3164         if (flags->verbose) {
3165                 char buf1[64];
3166                 char buf2[64];
3167                 uint64_t bytes = zc.zc_cookie;
3168                 time_t delta = time(NULL) - begin_time;
3169                 if (delta == 0)
3170                         delta = 1;
3171                 zfs_nicenum(bytes, buf1, sizeof (buf1));
3172                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3173
3174                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3175                     buf1, delta, buf2);
3176         }
3177
3178         return (0);
3179 }
3180
3181 static int
3182 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3183     int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl,
3184     char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
3185 {
3186         int err;
3187         dmu_replay_record_t drr, drr_noswap;
3188         struct drr_begin *drrb = &drr.drr_u.drr_begin;
3189         char errbuf[1024];
3190         zio_cksum_t zcksum = { 0 };
3191         uint64_t featureflags;
3192         int hdrtype;
3193
3194         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3195             "cannot receive"));
3196
3197         if (flags->isprefix &&
3198             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3199                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3200                     "(%s) does not exist"), tosnap);
3201                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3202         }
3203
3204         /* read in the BEGIN record */
3205         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3206             &zcksum)))
3207                 return (err);
3208
3209         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3210                 /* It's the double end record at the end of a package */
3211                 return (ENODATA);
3212         }
3213
3214         /* the kernel needs the non-byteswapped begin record */
3215         drr_noswap = drr;
3216
3217         flags->byteswap = B_FALSE;
3218         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3219                 /*
3220                  * We computed the checksum in the wrong byteorder in
3221                  * recv_read() above; do it again correctly.
3222                  */
3223                 bzero(&zcksum, sizeof (zio_cksum_t));
3224                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3225                 flags->byteswap = B_TRUE;
3226
3227                 drr.drr_type = BSWAP_32(drr.drr_type);
3228                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3229                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3230                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3231                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3232                 drrb->drr_type = BSWAP_32(drrb->drr_type);
3233                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3234                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3235                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3236         }
3237
3238         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3239                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3240                     "stream (bad magic number)"));
3241                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3242         }
3243
3244         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3245         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3246
3247         if (!DMU_STREAM_SUPPORTED(featureflags) ||
3248             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3249                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3250                     "stream has unsupported feature, feature flags = %lx"),
3251                     featureflags);
3252                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3253         }
3254
3255         if (strchr(drrb->drr_toname, '@') == NULL) {
3256                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3257                     "stream (bad snapshot name)"));
3258                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3259         }
3260
3261         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3262                 char nonpackage_sendfs[ZFS_MAXNAMELEN];
3263                 if (sendfs == NULL) {
3264                         /*
3265                          * We were not called from zfs_receive_package(). Get
3266                          * the fs specified by 'zfs send'.
3267                          */
3268                         char *cp;
3269                         (void) strlcpy(nonpackage_sendfs,
3270                             drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
3271                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3272                                 *cp = '\0';
3273                         sendfs = nonpackage_sendfs;
3274                 }
3275                 return (zfs_receive_one(hdl, infd, tosnap, flags,
3276                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl,
3277                     top_zfs, cleanup_fd, action_handlep));
3278         } else {
3279                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3280                     DMU_COMPOUNDSTREAM);
3281                 return (zfs_receive_package(hdl, infd, tosnap, flags,
3282                     &drr, &zcksum, top_zfs, cleanup_fd, action_handlep));
3283         }
3284 }
3285
3286 /*
3287  * Restores a backup of tosnap from the file descriptor specified by infd.
3288  * Return 0 on total success, -2 if some things couldn't be
3289  * destroyed/renamed/promoted, -1 if some things couldn't be received.
3290  * (-1 will override -2).
3291  */
3292 int
3293 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
3294     int infd, avl_tree_t *stream_avl)
3295 {
3296         char *top_zfs = NULL;
3297         int err;
3298         int cleanup_fd;
3299         uint64_t action_handle = 0;
3300
3301         cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3302         VERIFY(cleanup_fd >= 0);
3303
3304         err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL,
3305             stream_avl, &top_zfs, cleanup_fd, &action_handle);
3306
3307         VERIFY(0 == close(cleanup_fd));
3308
3309         if (err == 0 && !flags->nomount && top_zfs) {
3310                 zfs_handle_t *zhp;
3311                 prop_changelist_t *clp;
3312
3313                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3314                 if (zhp != NULL) {
3315                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3316                             CL_GATHER_MOUNT_ALWAYS, 0);
3317                         zfs_close(zhp);
3318                         if (clp != NULL) {
3319                                 /* mount and share received datasets */
3320                                 err = changelist_postfix(clp);
3321                                 changelist_free(clp);
3322                         }
3323                 }
3324                 if (zhp == NULL || clp == NULL || err)
3325                         err = -1;
3326         }
3327         if (top_zfs)
3328                 free(top_zfs);
3329
3330         return (err);
3331 }