lib/libzfs/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  */
  25
  26 #include <assert.h>
  27 #include <ctype.h>
  28 #include <errno.h>
  29 #include <libintl.h>
  30 #include <stdio.h>
  31 #include <stdlib.h>
  32 #include <strings.h>
  33 #include <unistd.h>
  34 #include <stddef.h>
  35 #include <fcntl.h>
  36 #include <sys/mount.h>
  37 #include <pthread.h>
  38 #include <umem.h>
  39
  40 #include <libzfs.h>
  41
  42 #include "zfs_namecheck.h"
  43 #include "zfs_prop.h"
  44 #include "zfs_fletcher.h"
  45 #include "libzfs_impl.h"
  46 #include <sha2.h>
  47 #include <sys/zio_checksum.h>
  48 #include <sys/ddt.h>
  49
  50 /* in libzfs_dataset.c */
  51 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
  52
  53 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t,
  54     int, const char *, nvlist_t *, avl_tree_t *, char **);
  55
  56 static const zio_cksum_t zero_cksum = { 0 };
  57
  58 typedef struct dedup_arg {
  59         int     inputfd;
  60         int     outputfd;
  61         libzfs_handle_t  *dedup_hdl;
  62 } dedup_arg_t;
  63
  64 typedef struct dataref {
  65         uint64_t ref_guid;
  66         uint64_t ref_object;
  67         uint64_t ref_offset;
  68 } dataref_t;
  69
  70 typedef struct dedup_entry {
  71         struct dedup_entry      *dde_next;
  72         zio_cksum_t dde_chksum;
  73         uint64_t dde_prop;
  74         dataref_t dde_ref;
  75 } dedup_entry_t;
  76
  77 #define MAX_DDT_PHYSMEM_PERCENT         20
  78 #define SMALLEST_POSSIBLE_MAX_DDT_MB            128
  79
  80 typedef struct dedup_table {
  81         dedup_entry_t   **dedup_hash_array;
  82         umem_cache_t    *ddecache;
  83         uint64_t        max_ddt_size;  /* max dedup table size in bytes */
  84         uint64_t        cur_ddt_size;  /* current dedup table size in bytes */
  85         uint64_t        ddt_count;
  86         int             numhashbits;
  87         boolean_t       ddt_full;
  88 } dedup_table_t;
  89
  90 static int
  91 high_order_bit(uint64_t n)
  92 {
  93         int count;
  94
  95         for (count = 0; n != 0; count++)
  96                 n >>= 1;
  97         return (count);
  98 }
  99
 100 static size_t
 101 ssread(void *buf, size_t len, FILE *stream)
 102 {
 103         size_t outlen;
 104
 105         if ((outlen = fread(buf, len, 1, stream)) == 0)
 106                 return (0);
 107
 108         return (outlen);
 109 }
 110
 111 static void
 112 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
 113     zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
 114 {
 115         dedup_entry_t   *dde;
 116
 117         if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
 118                 if (ddt->ddt_full == B_FALSE) {
 119                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 120                             "Dedup table full.  Deduplication will continue "
 121                             "with existing table entries"));
 122                         ddt->ddt_full = B_TRUE;
 123                 }
 124                 return;
 125         }
 126
 127         if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
 128             != NULL) {
 129                 assert(*ddepp == NULL);
 130                 dde->dde_next = NULL;
 131                 dde->dde_chksum = *cs;
 132                 dde->dde_prop = prop;
 133                 dde->dde_ref = *dr;
 134                 *ddepp = dde;
 135                 ddt->cur_ddt_size += sizeof (dedup_entry_t);
 136                 ddt->ddt_count++;
 137         }
 138 }
 139
 140 /*
 141  * Using the specified dedup table, do a lookup for an entry with
 142  * the checksum cs.  If found, return the block's reference info
 143  * in *dr. Otherwise, insert a new entry in the dedup table, using
 144  * the reference information specified by *dr.
 145  *
 146  * return value:  true - entry was found
 147  *                false - entry was not found
 148  */
 149 static boolean_t
 150 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
 151     uint64_t prop, dataref_t *dr)
 152 {
 153         uint32_t hashcode;
 154         dedup_entry_t **ddepp;
 155
 156         hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
 157
 158         for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
 159             ddepp = &((*ddepp)->dde_next)) {
 160                 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
 161                     (*ddepp)->dde_prop == prop) {
 162                         *dr = (*ddepp)->dde_ref;
 163                         return (B_TRUE);
 164                 }
 165         }
 166         ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
 167         return (B_FALSE);
 168 }
 169
 170 static int
 171 cksum_and_write(const void *buf, uint64_t len, zio_cksum_t *zc, int outfd)
 172 {
 173         fletcher_4_incremental_native(buf, len, zc);
 174         return (write(outfd, buf, len));
 175 }
 176
 177 /*
 178  * This function is started in a separate thread when the dedup option
 179  * has been requested.  The main send thread determines the list of
 180  * snapshots to be included in the send stream and makes the ioctl calls
 181  * for each one.  But instead of having the ioctl send the output to the
 182  * the output fd specified by the caller of zfs_send()), the
 183  * ioctl is told to direct the output to a pipe, which is read by the
 184  * alternate thread running THIS function.  This function does the
 185  * dedup'ing by:
 186  *  1. building a dedup table (the DDT)
 187  *  2. doing checksums on each data block and inserting a record in the DDT
 188  *  3. looking for matching checksums, and
 189  *  4.  sending a DRR_WRITE_BYREF record instead of a write record whenever
 190  *      a duplicate block is found.
 191  * The output of this function then goes to the output fd requested
 192  * by the caller of zfs_send().
 193  */
 194 static void *
 195 cksummer(void *arg)
 196 {
 197         dedup_arg_t *dda = arg;
 198         char *buf = malloc(1<<20);
 199         dmu_replay_record_t thedrr;
 200         dmu_replay_record_t *drr = &thedrr;
 201         struct drr_begin *drrb = &thedrr.drr_u.drr_begin;
 202         struct drr_end *drre = &thedrr.drr_u.drr_end;
 203         struct drr_object *drro = &thedrr.drr_u.drr_object;
 204         struct drr_write *drrw = &thedrr.drr_u.drr_write;
 205         struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
 206         FILE *ofp;
 207         int outfd;
 208         dmu_replay_record_t wbr_drr = {0};
 209         struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref;
 210         dedup_table_t ddt;
 211         zio_cksum_t stream_cksum;
 212         uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
 213         uint64_t numbuckets;
 214
 215         ddt.max_ddt_size =
 216             MAX((physmem * MAX_DDT_PHYSMEM_PERCENT)/100,
 217             SMALLEST_POSSIBLE_MAX_DDT_MB<<20);
 218
 219         numbuckets = ddt.max_ddt_size/(sizeof (dedup_entry_t));
 220
 221         /*
 222          * numbuckets must be a power of 2.  Increase number to
 223          * a power of 2 if necessary.
 224          */
 225         if (!ISP2(numbuckets))
 226                 numbuckets = 1 << high_order_bit(numbuckets);
 227
 228         ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
 229         ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
 230             NULL, NULL, NULL, NULL, NULL, 0);
 231         ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
 232         ddt.numhashbits = high_order_bit(numbuckets) - 1;
 233         ddt.ddt_full = B_FALSE;
 234
 235         /* Initialize the write-by-reference block. */
 236         wbr_drr.drr_type = DRR_WRITE_BYREF;
 237         wbr_drr.drr_payloadlen = 0;
 238
 239         outfd = dda->outputfd;
 240         ofp = fdopen(dda->inputfd, "r");
 241         while (ssread(drr, sizeof (dmu_replay_record_t), ofp) != 0) {
 242
 243                 switch (drr->drr_type) {
 244                 case DRR_BEGIN:
 245                 {
 246                         int     fflags;
 247                         ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
 248
 249                         /* set the DEDUP feature flag for this stream */
 250                         fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 251                         fflags |= (DMU_BACKUP_FEATURE_DEDUP |
 252                             DMU_BACKUP_FEATURE_DEDUPPROPS);
 253                         DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
 254
 255                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 256                             &stream_cksum, outfd) == -1)
 257                                 goto out;
 258                         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
 259                             DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) {
 260                                 int sz = drr->drr_payloadlen;
 261
 262                                 if (sz > 1<<20) {
 263                                         free(buf);
 264                                         buf = malloc(sz);
 265                                 }
 266                                 (void) ssread(buf, sz, ofp);
 267                                 if (ferror(stdin))
 268                                         perror("fread");
 269                                 if (cksum_and_write(buf, sz, &stream_cksum,
 270                                     outfd) == -1)
 271                                         goto out;
 272                         }
 273                         break;
 274                 }
 275
 276                 case DRR_END:
 277                 {
 278                         /* use the recalculated checksum */
 279                         ZIO_SET_CHECKSUM(&drre->drr_checksum,
 280                             stream_cksum.zc_word[0], stream_cksum.zc_word[1],
 281                             stream_cksum.zc_word[2], stream_cksum.zc_word[3]);
 282                         if ((write(outfd, drr,
 283                             sizeof (dmu_replay_record_t))) == -1)
 284                                 goto out;
 285                         break;
 286                 }
 287
 288                 case DRR_OBJECT:
 289                 {
 290                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 291                             &stream_cksum, outfd) == -1)
 292                                 goto out;
 293                         if (drro->drr_bonuslen > 0) {
 294                                 (void) ssread(buf,
 295                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 296                                     ofp);
 297                                 if (cksum_and_write(buf,
 298                                     P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
 299                                     &stream_cksum, outfd) == -1)
 300                                         goto out;
 301                         }
 302                         break;
 303                 }
 304
 305                 case DRR_SPILL:
 306                 {
 307                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 308                             &stream_cksum, outfd) == -1)
 309                                 goto out;
 310                         (void) ssread(buf, drrs->drr_length, ofp);
 311                         if (cksum_and_write(buf, drrs->drr_length,
 312                             &stream_cksum, outfd) == -1)
 313                                 goto out;
 314                         break;
 315                 }
 316
 317                 case DRR_FREEOBJECTS:
 318                 {
 319                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 320                             &stream_cksum, outfd) == -1)
 321                                 goto out;
 322                         break;
 323                 }
 324
 325                 case DRR_WRITE:
 326                 {
 327                         dataref_t       dataref;
 328
 329                         (void) ssread(buf, drrw->drr_length, ofp);
 330
 331                         /*
 332                          * Use the existing checksum if it's dedup-capable,
 333                          * else calculate a SHA256 checksum for it.
 334                          */
 335
 336                         if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
 337                             zero_cksum) ||
 338                             !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
 339                                 SHA256_CTX      ctx;
 340                                 zio_cksum_t     tmpsha256;
 341
 342                                 SHA256Init(&ctx);
 343                                 SHA256Update(&ctx, buf, drrw->drr_length);
 344                                 SHA256Final(&tmpsha256, &ctx);
 345                                 drrw->drr_key.ddk_cksum.zc_word[0] =
 346                                     BE_64(tmpsha256.zc_word[0]);
 347                                 drrw->drr_key.ddk_cksum.zc_word[1] =
 348                                     BE_64(tmpsha256.zc_word[1]);
 349                                 drrw->drr_key.ddk_cksum.zc_word[2] =
 350                                     BE_64(tmpsha256.zc_word[2]);
 351                                 drrw->drr_key.ddk_cksum.zc_word[3] =
 352                                     BE_64(tmpsha256.zc_word[3]);
 353                                 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
 354                                 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
 355                         }
 356
 357                         dataref.ref_guid = drrw->drr_toguid;
 358                         dataref.ref_object = drrw->drr_object;
 359                         dataref.ref_offset = drrw->drr_offset;
 360
 361                         if (ddt_update(dda->dedup_hdl, &ddt,
 362                             &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
 363                             &dataref)) {
 364                                 /* block already present in stream */
 365                                 wbr_drrr->drr_object = drrw->drr_object;
 366                                 wbr_drrr->drr_offset = drrw->drr_offset;
 367                                 wbr_drrr->drr_length = drrw->drr_length;
 368                                 wbr_drrr->drr_toguid = drrw->drr_toguid;
 369                                 wbr_drrr->drr_refguid = dataref.ref_guid;
 370                                 wbr_drrr->drr_refobject =
 371                                     dataref.ref_object;
 372                                 wbr_drrr->drr_refoffset =
 373                                     dataref.ref_offset;
 374
 375                                 wbr_drrr->drr_checksumtype =
 376                                     drrw->drr_checksumtype;
 377                                 wbr_drrr->drr_checksumflags =
 378                                     drrw->drr_checksumtype;
 379                                 wbr_drrr->drr_key.ddk_cksum =
 380                                     drrw->drr_key.ddk_cksum;
 381                                 wbr_drrr->drr_key.ddk_prop =
 382                                     drrw->drr_key.ddk_prop;
 383
 384                                 if (cksum_and_write(&wbr_drr,
 385                                     sizeof (dmu_replay_record_t), &stream_cksum,
 386                                     outfd) == -1)
 387                                         goto out;
 388                         } else {
 389                                 /* block not previously seen */
 390                                 if (cksum_and_write(drr,
 391                                     sizeof (dmu_replay_record_t), &stream_cksum,
 392                                     outfd) == -1)
 393                                         goto out;
 394                                 if (cksum_and_write(buf,
 395                                     drrw->drr_length,
 396                                     &stream_cksum, outfd) == -1)
 397                                         goto out;
 398                         }
 399                         break;
 400                 }
 401
 402                 case DRR_FREE:
 403                 {
 404                         if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
 405                             &stream_cksum, outfd) == -1)
 406                                 goto out;
 407                         break;
 408                 }
 409
 410                 default:
 411                         (void) printf("INVALID record type 0x%x\n",
 412                             drr->drr_type);
 413                         /* should never happen, so assert */
 414                         assert(B_FALSE);
 415                 }
 416         }
 417 out:
 418         umem_cache_destroy(ddt.ddecache);
 419         free(ddt.dedup_hash_array);
 420         free(buf);
 421         (void) fclose(ofp);
 422
 423         return (NULL);
 424 }
 425
 426 /*
 427  * Routines for dealing with the AVL tree of fs-nvlists
 428  */
 429 typedef struct fsavl_node {
 430         avl_node_t fn_node;
 431         nvlist_t *fn_nvfs;
 432         char *fn_snapname;
 433         uint64_t fn_guid;
 434 } fsavl_node_t;
 435
 436 static int
 437 fsavl_compare(const void *arg1, const void *arg2)
 438 {
 439         const fsavl_node_t *fn1 = arg1;
 440         const fsavl_node_t *fn2 = arg2;
 441
 442         if (fn1->fn_guid > fn2->fn_guid)
 443                 return (+1);
 444         else if (fn1->fn_guid < fn2->fn_guid)
 445                 return (-1);
 446         else
 447                 return (0);
 448 }
 449
 450 /*
 451  * Given the GUID of a snapshot, find its containing filesystem and
 452  * (optionally) name.
 453  */
 454 static nvlist_t *
 455 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
 456 {
 457         fsavl_node_t fn_find;
 458         fsavl_node_t *fn;
 459
 460         fn_find.fn_guid = snapguid;
 461
 462         fn = avl_find(avl, &fn_find, NULL);
 463         if (fn) {
 464                 if (snapname)
 465                         *snapname = fn->fn_snapname;
 466                 return (fn->fn_nvfs);
 467         }
 468         return (NULL);
 469 }
 470
 471 static void
 472 fsavl_destroy(avl_tree_t *avl)
 473 {
 474         fsavl_node_t *fn;
 475         void *cookie;
 476
 477         if (avl == NULL)
 478                 return;
 479
 480         cookie = NULL;
 481         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 482                 free(fn);
 483         avl_destroy(avl);
 484         free(avl);
 485 }
 486
 487 /*
 488  * Given an nvlist, produce an avl tree of snapshots, ordered by guid
 489  */
 490 static avl_tree_t *
 491 fsavl_create(nvlist_t *fss)
 492 {
 493         avl_tree_t *fsavl;
 494         nvpair_t *fselem = NULL;
 495
 496         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 497                 return (NULL);
 498
 499         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 500             offsetof(fsavl_node_t, fn_node));
 501
 502         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 503                 nvlist_t *nvfs, *snaps;
 504                 nvpair_t *snapelem = NULL;
 505
 506                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 507                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 508
 509                 while ((snapelem =
 510                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 511                         fsavl_node_t *fn;
 512                         uint64_t guid;
 513
 514                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 515                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 516                                 fsavl_destroy(fsavl);
 517                                 return (NULL);
 518                         }
 519                         fn->fn_nvfs = nvfs;
 520                         fn->fn_snapname = nvpair_name(snapelem);
 521                         fn->fn_guid = guid;
 522
 523                         /*
 524                          * Note: if there are multiple snaps with the
 525                          * same GUID, we ignore all but one.
 526                          */
 527                         if (avl_find(fsavl, fn, NULL) == NULL)
 528                                 avl_add(fsavl, fn);
 529                         else
 530                                 free(fn);
 531                 }
 532         }
 533
 534         return (fsavl);
 535 }
 536
 537 /*
 538  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 539  */
 540 typedef struct send_data {
 541         uint64_t parent_fromsnap_guid;
 542         nvlist_t *parent_snaps;
 543         nvlist_t *fss;
 544         nvlist_t *snapprops;
 545         const char *fromsnap;
 546         const char *tosnap;
 547         boolean_t recursive;
 548
 549         /*
 550          * The header nvlist is of the following format:
 551          * {
 552          *   "tosnap" -> string
 553          *   "fromsnap" -> string (if incremental)
 554          *   "fss" -> {
 555          *      id -> {
 556          *
 557          *       "name" -> string (full name; for debugging)
 558          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 559          *
 560          *       "props" -> { name -> value (only if set here) }
 561          *       "snaps" -> { name (lastname) -> number (guid) }
 562          *       "snapprops" -> { name (lastname) -> { name -> value } }
 563          *
 564          *       "origin" -> number (guid) (if clone)
 565          *       "sent" -> boolean (not on-disk)
 566          *      }
 567          *   }
 568          * }
 569          *
 570          */
 571 } send_data_t;
 572
 573 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 574
 575 static int
 576 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 577 {
 578         send_data_t *sd = arg;
 579         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 580         char *snapname;
 581         nvlist_t *nv;
 582
 583         snapname = strrchr(zhp->zfs_name, '@')+1;
 584
 585         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 586         /*
 587          * NB: if there is no fromsnap here (it's a newly created fs in
 588          * an incremental replication), we will substitute the tosnap.
 589          */
 590         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 591             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 592             strcmp(snapname, sd->tosnap) == 0)) {
 593                 sd->parent_fromsnap_guid = guid;
 594         }
 595
 596         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 597         send_iterate_prop(zhp, nv);
 598         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 599         nvlist_free(nv);
 600
 601         zfs_close(zhp);
 602         return (0);
 603 }
 604
 605 static void
 606 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 607 {
 608         nvpair_t *elem = NULL;
 609
 610         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 611                 char *propname = nvpair_name(elem);
 612                 zfs_prop_t prop = zfs_name_to_prop(propname);
 613                 nvlist_t *propnv;
 614
 615                 if (!zfs_prop_user(propname)) {
 616                         /*
 617                          * Realistically, this should never happen.  However,
 618                          * we want the ability to add DSL properties without
 619                          * needing to make incompatible version changes.  We
 620                          * need to ignore unknown properties to allow older
 621                          * software to still send datasets containing these
 622                          * properties, with the unknown properties elided.
 623                          */
 624                         if (prop == ZPROP_INVAL)
 625                                 continue;
 626
 627                         if (zfs_prop_readonly(prop))
 628                                 continue;
 629                 }
 630
 631                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 632                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
 633                     prop == ZFS_PROP_REFQUOTA ||
 634                     prop == ZFS_PROP_REFRESERVATION) {
 635                         char *source;
 636                         uint64_t value;
 637                         verify(nvlist_lookup_uint64(propnv,
 638                             ZPROP_VALUE, &value) == 0);
 639                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 640                                 continue;
 641                         /*
 642                          * May have no source before SPA_VERSION_RECVD_PROPS,
 643                          * but is still modifiable.
 644                          */
 645                         if (nvlist_lookup_string(propnv,
 646                             ZPROP_SOURCE, &source) == 0) {
 647                                 if ((strcmp(source, zhp->zfs_name) != 0) &&
 648                                     (strcmp(source,
 649                                     ZPROP_SOURCE_VAL_RECVD) != 0))
 650                                         continue;
 651                         }
 652                 } else {
 653                         char *source;
 654                         if (nvlist_lookup_string(propnv,
 655                             ZPROP_SOURCE, &source) != 0)
 656                                 continue;
 657                         if ((strcmp(source, zhp->zfs_name) != 0) &&
 658                             (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
 659                                 continue;
 660                 }
 661
 662                 if (zfs_prop_user(propname) ||
 663                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 664                         char *value;
 665                         verify(nvlist_lookup_string(propnv,
 666                             ZPROP_VALUE, &value) == 0);
 667                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 668                 } else {
 669                         uint64_t value;
 670                         verify(nvlist_lookup_uint64(propnv,
 671                             ZPROP_VALUE, &value) == 0);
 672                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 673                 }
 674         }
 675 }
 676
 677 /*
 678  * recursively generate nvlists describing datasets.  See comment
 679  * for the data structure send_data_t above for description of contents
 680  * of the nvlist.
 681  */
 682 static int
 683 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 684 {
 685         send_data_t *sd = arg;
 686         nvlist_t *nvfs, *nv;
 687         int rv = 0;
 688         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 689         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 690         char guidstring[64];
 691
 692         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 693         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 694         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 695             sd->parent_fromsnap_guid));
 696
 697         if (zhp->zfs_dmustats.dds_origin[0]) {
 698                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 699                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 700                 if (origin == NULL)
 701                         return (-1);
 702                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 703                     origin->zfs_dmustats.dds_guid));
 704         }
 705
 706         /* iterate over props */
 707         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 708         send_iterate_prop(zhp, nv);
 709         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 710         nvlist_free(nv);
 711
 712         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 713         sd->parent_fromsnap_guid = 0;
 714         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 715         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 716         (void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
 717         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 718         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 719         nvlist_free(sd->parent_snaps);
 720         nvlist_free(sd->snapprops);
 721
 722         /* add this fs to nvlist */
 723         (void) snprintf(guidstring, sizeof (guidstring),
 724             "0x%llx", (longlong_t)guid);
 725         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 726         nvlist_free(nvfs);
 727
 728         /* iterate over children */
 729         if (sd->recursive)
 730                 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 731
 732         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 733
 734         zfs_close(zhp);
 735         return (rv);
 736 }
 737
 738 static int
 739 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 740     const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
 741 {
 742         zfs_handle_t *zhp;
 743         send_data_t sd = { 0 };
 744         int error;
 745
 746         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 747         if (zhp == NULL)
 748                 return (EZFS_BADTYPE);
 749
 750         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 751         sd.fromsnap = fromsnap;
 752         sd.tosnap = tosnap;
 753         sd.recursive = recursive;
 754
 755         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 756                 nvlist_free(sd.fss);
 757                 if (avlp != NULL)
 758                         *avlp = NULL;
 759                 *nvlp = NULL;
 760                 return (error);
 761         }
 762
 763         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 764                 nvlist_free(sd.fss);
 765                 *nvlp = NULL;
 766                 return (EZFS_NOMEM);
 767         }
 768
 769         *nvlp = sd.fss;
 770         return (0);
 771 }
 772
 773 /*
 774  * Routines for dealing with the sorted snapshot functionality
 775  */
 776 typedef struct zfs_node {
 777         zfs_handle_t    *zn_handle;
 778         avl_node_t      zn_avlnode;
 779 } zfs_node_t;
 780
 781 static int
 782 zfs_sort_snaps(zfs_handle_t *zhp, void *data)
 783 {
 784         avl_tree_t *avl = data;
 785         zfs_node_t *node = zfs_alloc(zhp->zfs_hdl, sizeof (zfs_node_t));
 786
 787         node->zn_handle = zhp;
 788         avl_add(avl, node);
 789         return (0);
 790 }
 791
 792 /* ARGSUSED */
 793 static int
 794 zfs_snapshot_compare(const void *larg, const void *rarg)
 795 {
 796         zfs_handle_t *l = ((zfs_node_t *)larg)->zn_handle;
 797         zfs_handle_t *r = ((zfs_node_t *)rarg)->zn_handle;
 798         uint64_t lcreate, rcreate;
 799
 800         /*
 801          * Sort them according to creation time.  We use the hidden
 802          * CREATETXG property to get an absolute ordering of snapshots.
 803          */
 804         lcreate = zfs_prop_get_int(l, ZFS_PROP_CREATETXG);
 805         rcreate = zfs_prop_get_int(r, ZFS_PROP_CREATETXG);
 806
 807         if (lcreate < rcreate)
 808                 return (-1);
 809         else if (lcreate > rcreate)
 810                 return (+1);
 811         else
 812                 return (0);
 813 }
 814
 815 int
 816 zfs_iter_snapshots_sorted(zfs_handle_t *zhp, zfs_iter_f callback, void *data)
 817 {
 818         int ret = 0;
 819         zfs_node_t *node;
 820         avl_tree_t avl;
 821         void *cookie = NULL;
 822
 823         avl_create(&avl, zfs_snapshot_compare,
 824             sizeof (zfs_node_t), offsetof(zfs_node_t, zn_avlnode));
 825
 826         ret = zfs_iter_snapshots(zhp, zfs_sort_snaps, &avl);
 827
 828         for (node = avl_first(&avl); node != NULL; node = AVL_NEXT(&avl, node))
 829                 ret |= callback(node->zn_handle, data);
 830
 831         while ((node = avl_destroy_nodes(&avl, &cookie)) != NULL)
 832                 free(node);
 833
 834         avl_destroy(&avl);
 835
 836         return (ret);
 837 }
 838
 839 /*
 840  * Routines specific to "zfs send"
 841  */
 842 typedef struct send_dump_data {
 843         /* these are all just the short snapname (the part after the @) */
 844         const char *fromsnap;
 845         const char *tosnap;
 846         char prevsnap[ZFS_MAXNAMELEN];
 847         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 848         boolean_t verbose;
 849         int outfd;
 850         boolean_t err;
 851         nvlist_t *fss;
 852         avl_tree_t *fsavl;
 853         snapfilter_cb_t *filter_cb;
 854         void *filter_cb_arg;
 855         nvlist_t *debugnv;
 856 } send_dump_data_t;
 857
 858 /*
 859  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 860  * NULL) to the file descriptor specified by outfd.
 861  */
 862 static int
 863 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, boolean_t fromorigin,
 864     int outfd, boolean_t enoent_ok, boolean_t *got_enoent, nvlist_t *debugnv)
 865 {
 866         zfs_cmd_t zc = { 0 };
 867         libzfs_handle_t *hdl = zhp->zfs_hdl;
 868         nvlist_t *thisdbg;
 869
 870         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 871         assert(fromsnap == NULL || fromsnap[0] == '\0' || !fromorigin);
 872
 873         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 874         if (fromsnap)
 875                 (void) strlcpy(zc.zc_value, fromsnap, sizeof (zc.zc_value));
 876         zc.zc_cookie = outfd;
 877         zc.zc_obj = fromorigin;
 878
 879         *got_enoent = B_FALSE;
 880
 881         VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 882         if (fromsnap && fromsnap[0] != '\0') {
 883                 VERIFY(0 == nvlist_add_string(thisdbg,
 884                     "fromsnap", fromsnap));
 885         }
 886
 887         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_SEND, &zc) != 0) {
 888                 char errbuf[1024];
 889                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 890                     "warning: cannot send '%s'"), zhp->zfs_name);
 891
 892                 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
 893                 if (debugnv) {
 894                         VERIFY(0 == nvlist_add_nvlist(debugnv,
 895                             zhp->zfs_name, thisdbg));
 896                 }
 897                 nvlist_free(thisdbg);
 898
 899                 switch (errno) {
 900
 901                 case EXDEV:
 902                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 903                             "not an earlier snapshot from the same fs"));
 904                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 905
 906                 case ENOENT:
 907                         if (enoent_ok) {
 908                                 *got_enoent = B_TRUE;
 909                                 return (0);
 910                         }
 911                         if (zfs_dataset_exists(hdl, zc.zc_name,
 912                             ZFS_TYPE_SNAPSHOT)) {
 913                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 914                                     "incremental source (@%s) does not exist"),
 915                                     zc.zc_value);
 916                         }
 917                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 918
 919                 case EDQUOT:
 920                 case EFBIG:
 921                 case EIO:
 922                 case ENOLINK:
 923                 case ENOSPC:
 924                 case ENOSTR:
 925                 case ENXIO:
 926                 case EPIPE:
 927                 case ERANGE:
 928                 case EFAULT:
 929                 case EROFS:
 930                         zfs_error_aux(hdl, strerror(errno));
 931                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 932
 933                 default:
 934                         return (zfs_standard_error(hdl, errno, errbuf));
 935                 }
 936         }
 937
 938         if (debugnv)
 939                 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
 940         nvlist_free(thisdbg);
 941
 942         return (0);
 943 }
 944
 945 static int
 946 dump_snapshot(zfs_handle_t *zhp, void *arg)
 947 {
 948         send_dump_data_t *sdd = arg;
 949         const char *thissnap;
 950         int err;
 951         boolean_t got_enoent;
 952         boolean_t isfromsnap, istosnap;
 953         boolean_t exclude = B_FALSE;
 954
 955         thissnap = strchr(zhp->zfs_name, '@') + 1;
 956         isfromsnap = (sdd->fromsnap != NULL &&
 957             strcmp(sdd->fromsnap, thissnap) == 0);
 958
 959         if (!sdd->seenfrom && isfromsnap) {
 960                 sdd->seenfrom = B_TRUE;
 961                 (void) strcpy(sdd->prevsnap, thissnap);
 962                 zfs_close(zhp);
 963                 return (0);
 964         }
 965
 966         if (sdd->seento || !sdd->seenfrom) {
 967                 zfs_close(zhp);
 968                 return (0);
 969         }
 970
 971         istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
 972         if (istosnap)
 973                 sdd->seento = B_TRUE;
 974
 975         if (!sdd->doall && !isfromsnap && !istosnap) {
 976                 if (sdd->replicate) {
 977                         char *snapname;
 978                         nvlist_t *snapprops;
 979                         /*
 980                          * Filter out all intermediate snapshots except origin
 981                          * snapshots needed to replicate clones.
 982                          */
 983                         nvlist_t *nvfs = fsavl_find(sdd->fsavl,
 984                             zhp->zfs_dmustats.dds_guid, &snapname);
 985
 986                         VERIFY(0 == nvlist_lookup_nvlist(nvfs,
 987                             "snapprops", &snapprops));
 988                         VERIFY(0 == nvlist_lookup_nvlist(snapprops,
 989                             thissnap, &snapprops));
 990                         exclude = !nvlist_exists(snapprops, "is_clone_origin");
 991                 } else {
 992                         exclude = B_TRUE;
 993                 }
 994         }
 995
 996         /*
 997          * If a filter function exists, call it to determine whether
 998          * this snapshot will be sent.
 999          */
1000         if (exclude || (sdd->filter_cb != NULL &&
1001             sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1002                 /*
1003                  * This snapshot is filtered out.  Don't send it, and don't
1004                  * set prevsnap, so it will be as if this snapshot didn't
1005                  * exist, and the next accepted snapshot will be sent as
1006                  * an incremental from the last accepted one, or as the
1007                  * first (and full) snapshot in the case of a replication,
1008                  * non-incremental send.
1009                  */
1010                 zfs_close(zhp);
1011                 return (0);
1012         }
1013
1014         /* send it */
1015         if (sdd->verbose) {
1016                 (void) fprintf(stderr, "sending from @%s to %s\n",
1017                     sdd->prevsnap, zhp->zfs_name);
1018         }
1019
1020         err = dump_ioctl(zhp, sdd->prevsnap,
1021             sdd->prevsnap[0] == '\0' && (sdd->fromorigin || sdd->replicate),
1022             sdd->outfd, B_TRUE, &got_enoent, sdd->debugnv);
1023
1024         if (got_enoent)
1025                 err = 0;
1026         else
1027                 (void) strcpy(sdd->prevsnap, thissnap);
1028         zfs_close(zhp);
1029         return (err);
1030 }
1031
1032 static int
1033 dump_filesystem(zfs_handle_t *zhp, void *arg)
1034 {
1035         int rv = 0;
1036         send_dump_data_t *sdd = arg;
1037         boolean_t missingfrom = B_FALSE;
1038         zfs_cmd_t zc = { 0 };
1039
1040         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1041             zhp->zfs_name, sdd->tosnap);
1042         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1043                 (void) fprintf(stderr, "WARNING: "
1044                     "could not send %s@%s: does not exist\n",
1045                     zhp->zfs_name, sdd->tosnap);
1046                 sdd->err = B_TRUE;
1047                 return (0);
1048         }
1049
1050         if (sdd->replicate && sdd->fromsnap) {
1051                 /*
1052                  * If this fs does not have fromsnap, and we're doing
1053                  * recursive, we need to send a full stream from the
1054                  * beginning (or an incremental from the origin if this
1055                  * is a clone).  If we're doing non-recursive, then let
1056                  * them get the error.
1057                  */
1058                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1059                     zhp->zfs_name, sdd->fromsnap);
1060                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1061                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1062                         missingfrom = B_TRUE;
1063                 }
1064         }
1065
1066         sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1067         if (sdd->fromsnap == NULL || missingfrom)
1068                 sdd->seenfrom = B_TRUE;
1069
1070         rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1071         if (!sdd->seenfrom) {
1072                 (void) fprintf(stderr,
1073                     "WARNING: could not send %s@%s:\n"
1074                     "incremental source (%s@%s) does not exist\n",
1075                     zhp->zfs_name, sdd->tosnap,
1076                     zhp->zfs_name, sdd->fromsnap);
1077                 sdd->err = B_TRUE;
1078         } else if (!sdd->seento) {
1079                 if (sdd->fromsnap) {
1080                         (void) fprintf(stderr,
1081                             "WARNING: could not send %s@%s:\n"
1082                             "incremental source (%s@%s) "
1083                             "is not earlier than it\n",
1084                             zhp->zfs_name, sdd->tosnap,
1085                             zhp->zfs_name, sdd->fromsnap);
1086                 } else {
1087                         (void) fprintf(stderr, "WARNING: "
1088                             "could not send %s@%s: does not exist\n",
1089                             zhp->zfs_name, sdd->tosnap);
1090                 }
1091                 sdd->err = B_TRUE;
1092         }
1093
1094         return (rv);
1095 }
1096
1097 static int
1098 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1099 {
1100         send_dump_data_t *sdd = arg;
1101         nvpair_t *fspair;
1102         boolean_t needagain, progress;
1103
1104         if (!sdd->replicate)
1105                 return (dump_filesystem(rzhp, sdd));
1106
1107         /* Mark the clone origin snapshots. */
1108         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1109             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1110                 nvlist_t *nvfs;
1111                 uint64_t origin_guid = 0;
1112
1113                 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1114                 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1115                 if (origin_guid != 0) {
1116                         char *snapname;
1117                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1118                             origin_guid, &snapname);
1119                         if (origin_nv != NULL) {
1120                                 nvlist_t *snapprops;
1121                                 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1122                                     "snapprops", &snapprops));
1123                                 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1124                                     snapname, &snapprops));
1125                                 VERIFY(0 == nvlist_add_boolean(
1126                                     snapprops, "is_clone_origin"));
1127                         }
1128                 }
1129         }
1130 again:
1131         needagain = progress = B_FALSE;
1132         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1133             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1134                 nvlist_t *fslist;
1135                 char *fsname;
1136                 zfs_handle_t *zhp;
1137                 int err;
1138                 uint64_t origin_guid = 0;
1139
1140                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1141                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1142                         continue;
1143
1144                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1145                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1146
1147                 if (origin_guid != 0) {
1148                         nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1149                             origin_guid, NULL);
1150                         if (origin_nv != NULL &&
1151                             nvlist_lookup_boolean(origin_nv,
1152                             "sent") == ENOENT) {
1153                                 /*
1154                                  * origin has not been sent yet;
1155                                  * skip this clone.
1156                                  */
1157                                 needagain = B_TRUE;
1158                                 continue;
1159                         }
1160                 }
1161
1162                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1163                 if (zhp == NULL)
1164                         return (-1);
1165                 err = dump_filesystem(zhp, sdd);
1166                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1167                 progress = B_TRUE;
1168                 zfs_close(zhp);
1169                 if (err)
1170                         return (err);
1171         }
1172         if (needagain) {
1173                 assert(progress);
1174                 goto again;
1175         }
1176         return (0);
1177 }
1178
1179 /*
1180  * Generate a send stream for the dataset identified by the argument zhp.
1181  *
1182  * The content of the send stream is the snapshot identified by
1183  * 'tosnap'.  Incremental streams are requested in two ways:
1184  *     - from the snapshot identified by "fromsnap" (if non-null) or
1185  *     - from the origin of the dataset identified by zhp, which must
1186  *       be a clone.  In this case, "fromsnap" is null and "fromorigin"
1187  *       is TRUE.
1188  *
1189  * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1190  * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1191  * if "replicate" is set.  If "doall" is set, dump all the intermediate
1192  * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1193  * case too. If "props" is set, send properties.
1194  */
1195 int
1196 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1197     sendflags_t flags, int outfd, snapfilter_cb_t filter_func,
1198     void *cb_arg, nvlist_t **debugnvp)
1199 {
1200         char errbuf[1024];
1201         send_dump_data_t sdd = { 0 };
1202         int err;
1203         nvlist_t *fss = NULL;
1204         avl_tree_t *fsavl = NULL;
1205         char holdtag[128];
1206         static uint64_t holdseq;
1207         int spa_version;
1208         boolean_t holdsnaps = B_FALSE;
1209         pthread_t tid;
1210         int pipefd[2];
1211         dedup_arg_t dda = { 0 };
1212         int featureflags = 0;
1213
1214         if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1215                 uint64_t version;
1216                 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1217                 if (version >= ZPL_VERSION_SA) {
1218                         featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1219                 }
1220         }
1221
1222         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1223             "cannot send '%s'"), zhp->zfs_name);
1224
1225         if (fromsnap && fromsnap[0] == '\0') {
1226                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1227                     "zero-length incremental source"));
1228                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1229         }
1230
1231         if (zfs_spa_version(zhp, &spa_version) == 0 &&
1232             spa_version >= SPA_VERSION_USERREFS)
1233                 holdsnaps = B_TRUE;
1234
1235         if (flags.dedup) {
1236                 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1237                     DMU_BACKUP_FEATURE_DEDUPPROPS);
1238                 if (err = pipe(pipefd)) {
1239                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1240                         return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1241                             errbuf));
1242                 }
1243                 dda.outputfd = outfd;
1244                 dda.inputfd = pipefd[1];
1245                 dda.dedup_hdl = zhp->zfs_hdl;
1246                 if (err = pthread_create(&tid, NULL, cksummer, &dda)) {
1247                         (void) close(pipefd[0]);
1248                         (void) close(pipefd[1]);
1249                         zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1250                         return (zfs_error(zhp->zfs_hdl,
1251                             EZFS_THREADCREATEFAILED, errbuf));
1252                 }
1253         }
1254
1255         if (flags.replicate || flags.doall || flags.props) {
1256                 dmu_replay_record_t drr = { 0 };
1257                 char *packbuf = NULL;
1258                 size_t buflen = 0;
1259                 zio_cksum_t zc = { 0 };
1260
1261                 if (holdsnaps) {
1262                         (void) snprintf(holdtag, sizeof (holdtag),
1263                             ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1264                         ++holdseq;
1265                         err = zfs_hold_range(zhp, fromsnap, tosnap,
1266                             holdtag, flags.replicate, B_TRUE, filter_func,
1267                             cb_arg);
1268                         if (err)
1269                                 goto err_out;
1270                 }
1271
1272                 if (flags.replicate || flags.props) {
1273                         nvlist_t *hdrnv;
1274
1275                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1276                         if (fromsnap) {
1277                                 VERIFY(0 == nvlist_add_string(hdrnv,
1278                                     "fromsnap", fromsnap));
1279                         }
1280                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1281                         if (!flags.replicate) {
1282                                 VERIFY(0 == nvlist_add_boolean(hdrnv,
1283                                     "not_recursive"));
1284                         }
1285
1286                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1287                             fromsnap, tosnap, flags.replicate, &fss, &fsavl);
1288                         if (err) {
1289                                 if (holdsnaps) {
1290                                         (void) zfs_release_range(zhp, fromsnap,
1291                                             tosnap, holdtag, flags.replicate);
1292                                 }
1293                                 goto err_out;
1294                         }
1295                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1296                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
1297                             NV_ENCODE_XDR, 0);
1298                         if (debugnvp)
1299                                 *debugnvp = hdrnv;
1300                         else
1301                                 nvlist_free(hdrnv);
1302                         if (err) {
1303                                 fsavl_destroy(fsavl);
1304                                 nvlist_free(fss);
1305                                 if (holdsnaps) {
1306                                         (void) zfs_release_range(zhp, fromsnap,
1307                                             tosnap, holdtag, flags.replicate);
1308                                 }
1309                                 goto stderr_out;
1310                         }
1311                 }
1312
1313                 /* write first begin record */
1314                 drr.drr_type = DRR_BEGIN;
1315                 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1316                 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.drr_versioninfo,
1317                     DMU_COMPOUNDSTREAM);
1318                 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.drr_versioninfo,
1319                     featureflags);
1320                 (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1321                     sizeof (drr.drr_u.drr_begin.drr_toname),
1322                     "%s@%s", zhp->zfs_name, tosnap);
1323                 drr.drr_payloadlen = buflen;
1324                 err = cksum_and_write(&drr, sizeof (drr), &zc, outfd);
1325
1326                 /* write header nvlist */
1327                 if (err != -1 && packbuf != NULL) {
1328                         err = cksum_and_write(packbuf, buflen, &zc, outfd);
1329                 }
1330                 free(packbuf);
1331                 if (err == -1) {
1332                         fsavl_destroy(fsavl);
1333                         nvlist_free(fss);
1334                         if (holdsnaps) {
1335                                 (void) zfs_release_range(zhp, fromsnap, tosnap,
1336                                     holdtag, flags.replicate);
1337                         }
1338                         err = errno;
1339                         goto stderr_out;
1340                 }
1341
1342                 /* write end record */
1343                 if (err != -1) {
1344                         bzero(&drr, sizeof (drr));
1345                         drr.drr_type = DRR_END;
1346                         drr.drr_u.drr_end.drr_checksum = zc;
1347                         err = write(outfd, &drr, sizeof (drr));
1348                         if (err == -1) {
1349                                 fsavl_destroy(fsavl);
1350                                 nvlist_free(fss);
1351                                 err = errno;
1352                                 if (holdsnaps) {
1353                                         (void) zfs_release_range(zhp, fromsnap,
1354                                             tosnap, holdtag, flags.replicate);
1355                                 }
1356                                 goto stderr_out;
1357                         }
1358                 }
1359         }
1360
1361         /* dump each stream */
1362         sdd.fromsnap = fromsnap;
1363         sdd.tosnap = tosnap;
1364         if (flags.dedup)
1365                 sdd.outfd = pipefd[0];
1366         else
1367                 sdd.outfd = outfd;
1368         sdd.replicate = flags.replicate;
1369         sdd.doall = flags.doall;
1370         sdd.fromorigin = flags.fromorigin;
1371         sdd.fss = fss;
1372         sdd.fsavl = fsavl;
1373         sdd.verbose = flags.verbose;
1374         sdd.filter_cb = filter_func;
1375         sdd.filter_cb_arg = cb_arg;
1376         if (debugnvp)
1377                 sdd.debugnv = *debugnvp;
1378         err = dump_filesystems(zhp, &sdd);
1379         fsavl_destroy(fsavl);
1380         nvlist_free(fss);
1381
1382         if (flags.dedup) {
1383                 (void) close(pipefd[0]);
1384                 (void) pthread_join(tid, NULL);
1385         }
1386
1387         if (flags.replicate || flags.doall || flags.props) {
1388                 /*
1389                  * write final end record.  NB: want to do this even if
1390                  * there was some error, because it might not be totally
1391                  * failed.
1392                  */
1393                 dmu_replay_record_t drr = { 0 };
1394                 drr.drr_type = DRR_END;
1395                 if (holdsnaps) {
1396                         (void) zfs_release_range(zhp, fromsnap, tosnap,
1397                             holdtag, flags.replicate);
1398                 }
1399                 if (write(outfd, &drr, sizeof (drr)) == -1) {
1400                         return (zfs_standard_error(zhp->zfs_hdl,
1401                             errno, errbuf));
1402                 }
1403         }
1404
1405         return (err || sdd.err);
1406
1407 stderr_out:
1408         err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1409 err_out:
1410         if (flags.dedup) {
1411                 (void) pthread_cancel(tid);
1412                 (void) pthread_join(tid, NULL);
1413                 (void) close(pipefd[0]);
1414         }
1415         return (err);
1416 }
1417
1418 /*
1419  * Routines specific to "zfs recv"
1420  */
1421
1422 static int
1423 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
1424     boolean_t byteswap, zio_cksum_t *zc)
1425 {
1426         char *cp = buf;
1427         int rv;
1428         int len = ilen;
1429
1430         do {
1431                 rv = read(fd, cp, len);
1432                 cp += rv;
1433                 len -= rv;
1434         } while (rv > 0);
1435
1436         if (rv < 0 || len != 0) {
1437                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1438                     "failed to read from stream"));
1439                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
1440                     "cannot receive")));
1441         }
1442
1443         if (zc) {
1444                 if (byteswap)
1445                         fletcher_4_incremental_byteswap(buf, ilen, zc);
1446                 else
1447                         fletcher_4_incremental_native(buf, ilen, zc);
1448         }
1449         return (0);
1450 }
1451
1452 static int
1453 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
1454     boolean_t byteswap, zio_cksum_t *zc)
1455 {
1456         char *buf;
1457         int err;
1458
1459         buf = zfs_alloc(hdl, len);
1460         if (buf == NULL)
1461                 return (ENOMEM);
1462
1463         err = recv_read(hdl, fd, buf, len, byteswap, zc);
1464         if (err != 0) {
1465                 free(buf);
1466                 return (err);
1467         }
1468
1469         err = nvlist_unpack(buf, len, nvp, 0);
1470         free(buf);
1471         if (err != 0) {
1472                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1473                     "stream (malformed nvlist)"));
1474                 return (EINVAL);
1475         }
1476         return (0);
1477 }
1478
1479 static int
1480 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
1481     int baselen, char *newname, recvflags_t flags)
1482 {
1483         static int seq;
1484         zfs_cmd_t zc = { 0 };
1485         int err;
1486         prop_changelist_t *clp;
1487         zfs_handle_t *zhp;
1488
1489         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1490         if (zhp == NULL)
1491                 return (-1);
1492         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1493             flags.force ? MS_FORCE : 0);
1494         zfs_close(zhp);
1495         if (clp == NULL)
1496                 return (-1);
1497         err = changelist_prefix(clp);
1498         if (err)
1499                 return (err);
1500
1501         zc.zc_objset_type = DMU_OST_ZFS;
1502         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1503
1504         if (tryname) {
1505                 (void) strcpy(newname, tryname);
1506
1507                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
1508
1509                 if (flags.verbose) {
1510                         (void) printf("attempting rename %s to %s\n",
1511                             zc.zc_name, zc.zc_value);
1512                 }
1513                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1514                 if (err == 0)
1515                         changelist_rename(clp, name, tryname);
1516         } else {
1517                 err = ENOENT;
1518         }
1519
1520         if (err != 0 && strncmp(name+baselen, "recv-", 5) != 0) {
1521                 seq++;
1522
1523                 (void) strncpy(newname, name, baselen);
1524                 (void) snprintf(newname+baselen, ZFS_MAXNAMELEN-baselen,
1525                     "recv-%u-%u", getpid(), seq);
1526                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
1527
1528                 if (flags.verbose) {
1529                         (void) printf("failed - trying rename %s to %s\n",
1530                             zc.zc_name, zc.zc_value);
1531                 }
1532                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
1533                 if (err == 0)
1534                         changelist_rename(clp, name, newname);
1535                 if (err && flags.verbose) {
1536                         (void) printf("failed (%u) - "
1537                             "will try again on next pass\n", errno);
1538                 }
1539                 err = EAGAIN;
1540         } else if (flags.verbose) {
1541                 if (err == 0)
1542                         (void) printf("success\n");
1543                 else
1544                         (void) printf("failed (%u)\n", errno);
1545         }
1546
1547         (void) changelist_postfix(clp);
1548         changelist_free(clp);
1549
1550         return (err);
1551 }
1552
1553 static int
1554 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
1555     char *newname, recvflags_t flags)
1556 {
1557         zfs_cmd_t zc = { 0 };
1558         int err = 0;
1559         prop_changelist_t *clp;
1560         zfs_handle_t *zhp;
1561         boolean_t defer = B_FALSE;
1562         int spa_version;
1563
1564         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1565         if (zhp == NULL)
1566                 return (-1);
1567         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
1568             flags.force ? MS_FORCE : 0);
1569         if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
1570             zfs_spa_version(zhp, &spa_version) == 0 &&
1571             spa_version >= SPA_VERSION_USERREFS)
1572                 defer = B_TRUE;
1573         zfs_close(zhp);
1574         if (clp == NULL)
1575                 return (-1);
1576         err = changelist_prefix(clp);
1577         if (err)
1578                 return (err);
1579
1580         zc.zc_objset_type = DMU_OST_ZFS;
1581         zc.zc_defer_destroy = defer;
1582         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
1583
1584         if (flags.verbose)
1585                 (void) printf("attempting destroy %s\n", zc.zc_name);
1586         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
1587         if (err == 0) {
1588                 if (flags.verbose)
1589                         (void) printf("success\n");
1590                 changelist_remove(clp, zc.zc_name);
1591         }
1592
1593         (void) changelist_postfix(clp);
1594         changelist_free(clp);
1595
1596         /*
1597          * Deferred destroy might destroy the snapshot or only mark it to be
1598          * destroyed later, and it returns success in either case.
1599          */
1600         if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
1601             ZFS_TYPE_SNAPSHOT))) {
1602                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
1603         }
1604
1605         return (err);
1606 }
1607
1608 typedef struct guid_to_name_data {
1609         uint64_t guid;
1610         char *name;
1611 } guid_to_name_data_t;
1612
1613 static int
1614 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
1615 {
1616         guid_to_name_data_t *gtnd = arg;
1617         int err;
1618
1619         if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1620                 (void) strcpy(gtnd->name, zhp->zfs_name);
1621                 zfs_close(zhp);
1622                 return (EEXIST);
1623         }
1624         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1625         zfs_close(zhp);
1626         return (err);
1627 }
1628
1629 static int
1630 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1631     char *name)
1632 {
1633         /* exhaustive search all local snapshots */
1634         guid_to_name_data_t gtnd;
1635         int err = 0;
1636         zfs_handle_t *zhp;
1637         char *cp;
1638
1639         gtnd.guid = guid;
1640         gtnd.name = name;
1641
1642         if (strchr(parent, '@') == NULL) {
1643                 zhp = make_dataset_handle(hdl, parent);
1644                 if (zhp != NULL) {
1645                         err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1646                         zfs_close(zhp);
1647                         if (err == EEXIST)
1648                                 return (0);
1649                 }
1650         }
1651
1652         cp = strchr(parent, '/');
1653         if (cp)
1654                 *cp = '\0';
1655         zhp = make_dataset_handle(hdl, parent);
1656         if (cp)
1657                 *cp = '/';
1658
1659         if (zhp) {
1660                 err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1661                 zfs_close(zhp);
1662         }
1663
1664         return (err == EEXIST ? 0 : ENOENT);
1665
1666 }
1667
1668 /*
1669  * Return true if dataset guid1 is created before guid2.
1670  */
1671 static int
1672 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1673     uint64_t guid1, uint64_t guid2)
1674 {
1675         nvlist_t *nvfs;
1676         char *fsname, *snapname;
1677         char buf[ZFS_MAXNAMELEN];
1678         int rv;
1679         zfs_node_t zn1, zn2;
1680
1681         if (guid2 == 0)
1682                 return (0);
1683         if (guid1 == 0)
1684                 return (1);
1685
1686         nvfs = fsavl_find(avl, guid1, &snapname);
1687         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1688         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1689         zn1.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1690         if (zn1.zn_handle == NULL)
1691                 return (-1);
1692
1693         nvfs = fsavl_find(avl, guid2, &snapname);
1694         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1695         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1696         zn2.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1697         if (zn2.zn_handle == NULL) {
1698                 zfs_close(zn2.zn_handle);
1699                 return (-1);
1700         }
1701
1702         rv = (zfs_snapshot_compare(&zn1, &zn2) == -1);
1703
1704         zfs_close(zn1.zn_handle);
1705         zfs_close(zn2.zn_handle);
1706
1707         return (rv);
1708 }
1709
1710 static int
1711 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
1712     recvflags_t flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
1713     nvlist_t *renamed)
1714 {
1715         nvlist_t *local_nv;
1716         avl_tree_t *local_avl;
1717         nvpair_t *fselem, *nextfselem;
1718         char *fromsnap;
1719         char newname[ZFS_MAXNAMELEN];
1720         int error;
1721         boolean_t needagain, progress, recursive;
1722         char *s1, *s2;
1723
1724         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
1725
1726         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
1727             ENOENT);
1728
1729         if (flags.dryrun)
1730                 return (0);
1731
1732 again:
1733         needagain = progress = B_FALSE;
1734
1735         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
1736             recursive, &local_nv, &local_avl)) != 0)
1737                 return (error);
1738
1739         /*
1740          * Process deletes and renames
1741          */
1742         for (fselem = nvlist_next_nvpair(local_nv, NULL);
1743             fselem; fselem = nextfselem) {
1744                 nvlist_t *nvfs, *snaps;
1745                 nvlist_t *stream_nvfs = NULL;
1746                 nvpair_t *snapelem, *nextsnapelem;
1747                 uint64_t fromguid = 0;
1748                 uint64_t originguid = 0;
1749                 uint64_t stream_originguid = 0;
1750                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
1751                 char *fsname, *stream_fsname;
1752
1753                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
1754
1755                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
1756                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
1757                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1758                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
1759                     &parent_fromsnap_guid));
1760                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
1761
1762                 /*
1763                  * First find the stream's fs, so we can check for
1764                  * a different origin (due to "zfs promote")
1765                  */
1766                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1767                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
1768                         uint64_t thisguid;
1769
1770                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1771                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
1772
1773                         if (stream_nvfs != NULL)
1774                                 break;
1775                 }
1776
1777                 /* check for promote */
1778                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
1779                     &stream_originguid);
1780                 if (stream_nvfs && originguid != stream_originguid) {
1781                         switch (created_before(hdl, local_avl,
1782                             stream_originguid, originguid)) {
1783                         case 1: {
1784                                 /* promote it! */
1785                                 zfs_cmd_t zc = { 0 };
1786                                 nvlist_t *origin_nvfs;
1787                                 char *origin_fsname;
1788
1789                                 if (flags.verbose)
1790                                         (void) printf("promoting %s\n", fsname);
1791
1792                                 origin_nvfs = fsavl_find(local_avl, originguid,
1793                                     NULL);
1794                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
1795                                     "name", &origin_fsname));
1796                                 (void) strlcpy(zc.zc_value, origin_fsname,
1797                                     sizeof (zc.zc_value));
1798                                 (void) strlcpy(zc.zc_name, fsname,
1799                                     sizeof (zc.zc_name));
1800                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
1801                                 if (error == 0)
1802                                         progress = B_TRUE;
1803                                 break;
1804                         }
1805                         default:
1806                                 break;
1807                         case -1:
1808                                 fsavl_destroy(local_avl);
1809                                 nvlist_free(local_nv);
1810                                 return (-1);
1811                         }
1812                         /*
1813                          * We had/have the wrong origin, therefore our
1814                          * list of snapshots is wrong.  Need to handle
1815                          * them on the next pass.
1816                          */
1817                         needagain = B_TRUE;
1818                         continue;
1819                 }
1820
1821                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1822                     snapelem; snapelem = nextsnapelem) {
1823                         uint64_t thisguid;
1824                         char *stream_snapname;
1825                         nvlist_t *found, *props;
1826
1827                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
1828
1829                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1830                         found = fsavl_find(stream_avl, thisguid,
1831                             &stream_snapname);
1832
1833                         /* check for delete */
1834                         if (found == NULL) {
1835                                 char name[ZFS_MAXNAMELEN];
1836
1837                                 if (!flags.force)
1838                                         continue;
1839
1840                                 (void) snprintf(name, sizeof (name), "%s@%s",
1841                                     fsname, nvpair_name(snapelem));
1842
1843                                 error = recv_destroy(hdl, name,
1844                                     strlen(fsname)+1, newname, flags);
1845                                 if (error)
1846                                         needagain = B_TRUE;
1847                                 else
1848                                         progress = B_TRUE;
1849                                 continue;
1850                         }
1851
1852                         stream_nvfs = found;
1853
1854                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
1855                             &props) && 0 == nvlist_lookup_nvlist(props,
1856                             stream_snapname, &props)) {
1857                                 zfs_cmd_t zc = { 0 };
1858
1859                                 zc.zc_cookie = B_TRUE; /* received */
1860                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
1861                                     "%s@%s", fsname, nvpair_name(snapelem));
1862                                 if (zcmd_write_src_nvlist(hdl, &zc,
1863                                     props) == 0) {
1864                                         (void) zfs_ioctl(hdl,
1865                                             ZFS_IOC_SET_PROP, &zc);
1866                                         zcmd_free_nvlists(&zc);
1867                                 }
1868                         }
1869
1870                         /* check for different snapname */
1871                         if (strcmp(nvpair_name(snapelem),
1872                             stream_snapname) != 0) {
1873                                 char name[ZFS_MAXNAMELEN];
1874                                 char tryname[ZFS_MAXNAMELEN];
1875
1876                                 (void) snprintf(name, sizeof (name), "%s@%s",
1877                                     fsname, nvpair_name(snapelem));
1878                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
1879                                     fsname, stream_snapname);
1880
1881                                 error = recv_rename(hdl, name, tryname,
1882                                     strlen(fsname)+1, newname, flags);
1883                                 if (error)
1884                                         needagain = B_TRUE;
1885                                 else
1886                                         progress = B_TRUE;
1887                         }
1888
1889                         if (strcmp(stream_snapname, fromsnap) == 0)
1890                                 fromguid = thisguid;
1891                 }
1892
1893                 /* check for delete */
1894                 if (stream_nvfs == NULL) {
1895                         if (!flags.force)
1896                                 continue;
1897
1898                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
1899                             newname, flags);
1900                         if (error)
1901                                 needagain = B_TRUE;
1902                         else
1903                                 progress = B_TRUE;
1904                         continue;
1905                 }
1906
1907                 if (fromguid == 0) {
1908                         if (flags.verbose) {
1909                                 (void) printf("local fs %s does not have "
1910                                     "fromsnap (%s in stream); must have "
1911                                     "been deleted locally; ignoring\n",
1912                                     fsname, fromsnap);
1913                         }
1914                         continue;
1915                 }
1916
1917                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
1918                     "name", &stream_fsname));
1919                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
1920                     "parentfromsnap", &stream_parent_fromsnap_guid));
1921
1922                 s1 = strrchr(fsname, '/');
1923                 s2 = strrchr(stream_fsname, '/');
1924
1925                 /*
1926                  * Check for rename. If the exact receive path is specified, it
1927                  * does not count as a rename, but we still need to check the
1928                  * datasets beneath it.
1929                  */
1930                 if ((stream_parent_fromsnap_guid != 0 &&
1931                     parent_fromsnap_guid != 0 &&
1932                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
1933                     ((flags.isprefix || strcmp(tofs, fsname) != 0) &&
1934                     (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
1935                         nvlist_t *parent;
1936                         char tryname[ZFS_MAXNAMELEN];
1937
1938                         parent = fsavl_find(local_avl,
1939                             stream_parent_fromsnap_guid, NULL);
1940                         /*
1941                          * NB: parent might not be found if we used the
1942                          * tosnap for stream_parent_fromsnap_guid,
1943                          * because the parent is a newly-created fs;
1944                          * we'll be able to rename it after we recv the
1945                          * new fs.
1946                          */
1947                         if (parent != NULL) {
1948                                 char *pname;
1949
1950                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
1951                                     &pname));
1952                                 (void) snprintf(tryname, sizeof (tryname),
1953                                     "%s%s", pname, strrchr(stream_fsname, '/'));
1954                         } else {
1955                                 tryname[0] = '\0';
1956                                 if (flags.verbose) {
1957                                         (void) printf("local fs %s new parent "
1958                                             "not found\n", fsname);
1959                                 }
1960                         }
1961
1962                         newname[0] = '\0';
1963
1964                         error = recv_rename(hdl, fsname, tryname,
1965                             strlen(tofs)+1, newname, flags);
1966
1967                         if (renamed != NULL && newname[0] != '\0') {
1968                                 VERIFY(0 == nvlist_add_boolean(renamed,
1969                                     newname));
1970                         }
1971
1972                         if (error)
1973                                 needagain = B_TRUE;
1974                         else
1975                                 progress = B_TRUE;
1976                 }
1977         }
1978
1979         fsavl_destroy(local_avl);
1980         nvlist_free(local_nv);
1981
1982         if (needagain && progress) {
1983                 /* do another pass to fix up temporary names */
1984                 if (flags.verbose)
1985                         (void) printf("another pass:\n");
1986                 goto again;
1987         }
1988
1989         return (needagain);
1990 }
1991
1992 static int
1993 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
1994     recvflags_t flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
1995     char **top_zfs)
1996 {
1997         nvlist_t *stream_nv = NULL;
1998         avl_tree_t *stream_avl = NULL;
1999         char *fromsnap = NULL;
2000         char *cp;
2001         char tofs[ZFS_MAXNAMELEN];
2002         char sendfs[ZFS_MAXNAMELEN];
2003         char errbuf[1024];
2004         dmu_replay_record_t drre;
2005         int error;
2006         boolean_t anyerr = B_FALSE;
2007         boolean_t softerr = B_FALSE;
2008         boolean_t recursive;
2009
2010         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2011             "cannot receive"));
2012
2013         assert(drr->drr_type == DRR_BEGIN);
2014         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2015         assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2016             DMU_COMPOUNDSTREAM);
2017
2018         /*
2019          * Read in the nvlist from the stream.
2020          */
2021         if (drr->drr_payloadlen != 0) {
2022                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2023                     &stream_nv, flags.byteswap, zc);
2024                 if (error) {
2025                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2026                         goto out;
2027                 }
2028         }
2029
2030         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2031             ENOENT);
2032
2033         if (recursive && strchr(destname, '@')) {
2034                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2035                     "cannot specify snapshot name for multi-snapshot stream"));
2036                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2037                 goto out;
2038         }
2039
2040         /*
2041          * Read in the end record and verify checksum.
2042          */
2043         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2044             flags.byteswap, NULL)))
2045                 goto out;
2046         if (flags.byteswap) {
2047                 drre.drr_type = BSWAP_32(drre.drr_type);
2048                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2049                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2050                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2051                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2052                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2053                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2054                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2055                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2056         }
2057         if (drre.drr_type != DRR_END) {
2058                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2059                 goto out;
2060         }
2061         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2062                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2063                     "incorrect header checksum"));
2064                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2065                 goto out;
2066         }
2067
2068         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2069
2070         if (drr->drr_payloadlen != 0) {
2071                 nvlist_t *stream_fss;
2072
2073                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2074                     &stream_fss));
2075                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2076                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2077                             "couldn't allocate avl tree"));
2078                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2079                         goto out;
2080                 }
2081
2082                 if (fromsnap != NULL) {
2083                         nvlist_t *renamed = NULL;
2084                         nvpair_t *pair = NULL;
2085
2086                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
2087                         if (flags.isprefix) {
2088                                 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2089                                 int i;
2090
2091                                 if (flags.istail) {
2092                                         cp = strrchr(drrb->drr_toname, '/');
2093                                         if (cp == NULL) {
2094                                                 (void) strlcat(tofs, "/",
2095                                                     ZFS_MAXNAMELEN);
2096                                                 i = 0;
2097                                         } else {
2098                                                 i = (cp - drrb->drr_toname);
2099                                         }
2100                                 } else {
2101                                         i = strcspn(drrb->drr_toname, "/@");
2102                                 }
2103                                 /* zfs_receive_one() will create_parents() */
2104                                 (void) strlcat(tofs, &drrb->drr_toname[i],
2105                                     ZFS_MAXNAMELEN);
2106                                 *strchr(tofs, '@') = '\0';
2107                         }
2108
2109                         if (recursive && !flags.dryrun && !flags.nomount) {
2110                                 VERIFY(0 == nvlist_alloc(&renamed,
2111                                     NV_UNIQUE_NAME, 0));
2112                         }
2113
2114                         softerr = recv_incremental_replication(hdl, tofs, flags,
2115                             stream_nv, stream_avl, renamed);
2116
2117                         /* Unmount renamed filesystems before receiving. */
2118                         while ((pair = nvlist_next_nvpair(renamed,
2119                             pair)) != NULL) {
2120                                 zfs_handle_t *zhp;
2121                                 prop_changelist_t *clp = NULL;
2122
2123                                 zhp = zfs_open(hdl, nvpair_name(pair),
2124                                     ZFS_TYPE_FILESYSTEM);
2125                                 if (zhp != NULL) {
2126                                         clp = changelist_gather(zhp,
2127                                             ZFS_PROP_MOUNTPOINT, 0, 0);
2128                                         zfs_close(zhp);
2129                                         if (clp != NULL) {
2130                                                 softerr |=
2131                                                     changelist_prefix(clp);
2132                                                 changelist_free(clp);
2133                                         }
2134                                 }
2135                         }
2136
2137                         nvlist_free(renamed);
2138                 }
2139         }
2140
2141         /*
2142          * Get the fs specified by the first path in the stream (the top level
2143          * specified by 'zfs send') and pass it to each invocation of
2144          * zfs_receive_one().
2145          */
2146         (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2147             ZFS_MAXNAMELEN);
2148         if ((cp = strchr(sendfs, '@')) != NULL)
2149                 *cp = '\0';
2150
2151         /* Finally, receive each contained stream */
2152         do {
2153                 /*
2154                  * we should figure out if it has a recoverable
2155                  * error, in which case do a recv_skip() and drive on.
2156                  * Note, if we fail due to already having this guid,
2157                  * zfs_receive_one() will take care of it (ie,
2158                  * recv_skip() and return 0).
2159                  */
2160                 error = zfs_receive_impl(hdl, destname, flags, fd,
2161                     sendfs, stream_nv, stream_avl, top_zfs);
2162                 if (error == ENODATA) {
2163                         error = 0;
2164                         break;
2165                 }
2166                 anyerr |= error;
2167         } while (error == 0);
2168
2169         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2170                 /*
2171                  * Now that we have the fs's they sent us, try the
2172                  * renames again.
2173                  */
2174                 softerr = recv_incremental_replication(hdl, tofs, flags,
2175                     stream_nv, stream_avl, NULL);
2176         }
2177
2178 out:
2179         fsavl_destroy(stream_avl);
2180         if (stream_nv)
2181                 nvlist_free(stream_nv);
2182         if (softerr)
2183                 error = -2;
2184         if (anyerr)
2185                 error = -1;
2186         return (error);
2187 }
2188
2189 static void
2190 trunc_prop_errs(int truncated)
2191 {
2192         ASSERT(truncated != 0);
2193
2194         if (truncated == 1)
2195                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2196                     "1 more property could not be set\n"));
2197         else
2198                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2199                     "%d more properties could not be set\n"), truncated);
2200 }
2201
2202 static int
2203 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2204 {
2205         dmu_replay_record_t *drr;
2206         void *buf = malloc(1<<20);
2207         char errbuf[1024];
2208
2209         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2210             "cannot receive:"));
2211
2212         /* XXX would be great to use lseek if possible... */
2213         drr = buf;
2214
2215         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2216             byteswap, NULL) == 0) {
2217                 if (byteswap)
2218                         drr->drr_type = BSWAP_32(drr->drr_type);
2219
2220                 switch (drr->drr_type) {
2221                 case DRR_BEGIN:
2222                         /* NB: not to be used on v2 stream packages */
2223                         if (drr->drr_payloadlen != 0) {
2224                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2225                                     "invalid substream header"));
2226                                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2227                         }
2228                         break;
2229
2230                 case DRR_END:
2231                         free(buf);
2232                         return (0);
2233
2234                 case DRR_OBJECT:
2235                         if (byteswap) {
2236                                 drr->drr_u.drr_object.drr_bonuslen =
2237                                     BSWAP_32(drr->drr_u.drr_object.
2238                                     drr_bonuslen);
2239                         }
2240                         (void) recv_read(hdl, fd, buf,
2241                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2242                             B_FALSE, NULL);
2243                         break;
2244
2245                 case DRR_WRITE:
2246                         if (byteswap) {
2247                                 drr->drr_u.drr_write.drr_length =
2248                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
2249                         }
2250                         (void) recv_read(hdl, fd, buf,
2251                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2252                         break;
2253                 case DRR_SPILL:
2254                         if (byteswap) {
2255                                 drr->drr_u.drr_write.drr_length =
2256                                     BSWAP_64(drr->drr_u.drr_spill.drr_length);
2257                         }
2258                         (void) recv_read(hdl, fd, buf,
2259                             drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2260                         break;
2261                 case DRR_WRITE_BYREF:
2262                 case DRR_FREEOBJECTS:
2263                 case DRR_FREE:
2264                         break;
2265
2266                 default:
2267                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2268                             "invalid record type"));
2269                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2270                 }
2271         }
2272
2273         free(buf);
2274         return (-1);
2275 }
2276
2277 /*
2278  * Restores a backup of tosnap from the file descriptor specified by infd.
2279  */
2280 static int
2281 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
2282     recvflags_t flags, dmu_replay_record_t *drr,
2283     dmu_replay_record_t *drr_noswap, const char *sendfs,
2284     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs)
2285 {
2286         zfs_cmd_t zc = { 0 };
2287         time_t begin_time;
2288         int ioctl_err, ioctl_errno, err;
2289         char *cp;
2290         struct drr_begin *drrb = &drr->drr_u.drr_begin;
2291         char errbuf[1024];
2292         char prop_errbuf[1024];
2293         const char *chopprefix;
2294         boolean_t newfs = B_FALSE;
2295         boolean_t stream_wantsnewfs;
2296         uint64_t parent_snapguid = 0;
2297         prop_changelist_t *clp = NULL;
2298         nvlist_t *snapprops_nvlist = NULL;
2299         zprop_errflags_t prop_errflags;
2300         boolean_t recursive;
2301
2302         begin_time = time(NULL);
2303
2304         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2305             "cannot receive"));
2306
2307         recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2308             ENOENT);
2309
2310         if (stream_avl != NULL) {
2311                 char *snapname;
2312                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
2313                     &snapname);
2314                 nvlist_t *props;
2315                 int ret;
2316
2317                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
2318                     &parent_snapguid);
2319                 err = nvlist_lookup_nvlist(fs, "props", &props);
2320                 if (err)
2321                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
2322
2323                 if (flags.canmountoff) {
2324                         VERIFY(0 == nvlist_add_uint64(props,
2325                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
2326                 }
2327                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
2328                 if (err)
2329                         nvlist_free(props);
2330
2331                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
2332                         VERIFY(0 == nvlist_lookup_nvlist(props,
2333                             snapname, &snapprops_nvlist));
2334                 }
2335
2336                 if (ret != 0)
2337                         return (-1);
2338         }
2339
2340         cp = NULL;
2341
2342         /*
2343          * Determine how much of the snapshot name stored in the stream
2344          * we are going to tack on to the name they specified on the
2345          * command line, and how much we are going to chop off.
2346          *
2347          * If they specified a snapshot, chop the entire name stored in
2348          * the stream.
2349          */
2350         if (flags.istail) {
2351                 /*
2352                  * A filesystem was specified with -e. We want to tack on only
2353                  * the tail of the sent snapshot path.
2354                  */
2355                 if (strchr(tosnap, '@')) {
2356                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2357                             "argument - snapshot not allowed with -e"));
2358                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2359                 }
2360
2361                 chopprefix = strrchr(sendfs, '/');
2362
2363                 if (chopprefix == NULL) {
2364                         /*
2365                          * The tail is the poolname, so we need to
2366                          * prepend a path separator.
2367                          */
2368                         int len = strlen(drrb->drr_toname);
2369                         cp = malloc(len + 2);
2370                         cp[0] = '/';
2371                         (void) strcpy(&cp[1], drrb->drr_toname);
2372                         chopprefix = cp;
2373                 } else {
2374                         chopprefix = drrb->drr_toname + (chopprefix - sendfs);
2375                 }
2376         } else if (flags.isprefix) {
2377                 /*
2378                  * A filesystem was specified with -d. We want to tack on
2379                  * everything but the first element of the sent snapshot path
2380                  * (all but the pool name).
2381                  */
2382                 if (strchr(tosnap, '@')) {
2383                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2384                             "argument - snapshot not allowed with -d"));
2385                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2386                 }
2387
2388                 chopprefix = strchr(drrb->drr_toname, '/');
2389                 if (chopprefix == NULL)
2390                         chopprefix = strchr(drrb->drr_toname, '@');
2391         } else if (strchr(tosnap, '@') == NULL) {
2392                 /*
2393                  * If a filesystem was specified without -d or -e, we want to
2394                  * tack on everything after the fs specified by 'zfs send'.
2395                  */
2396                 chopprefix = drrb->drr_toname + strlen(sendfs);
2397         } else {
2398                 /* A snapshot was specified as an exact path (no -d or -e). */
2399                 if (recursive) {
2400                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2401                             "cannot specify snapshot name for multi-snapshot "
2402                             "stream"));
2403                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2404                 }
2405                 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
2406         }
2407
2408         ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
2409         ASSERT(chopprefix > drrb->drr_toname);
2410         ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
2411         ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
2412             chopprefix[0] == '\0');
2413
2414         /*
2415          * Determine name of destination snapshot, store in zc_value.
2416          */
2417         (void) strcpy(zc.zc_top_ds, tosnap);
2418         (void) strcpy(zc.zc_value, tosnap);
2419         (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
2420         free(cp);
2421         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
2422                 zcmd_free_nvlists(&zc);
2423                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2424         }
2425
2426         /*
2427          * Determine the name of the origin snapshot, store in zc_string.
2428          */
2429         if (drrb->drr_flags & DRR_FLAG_CLONE) {
2430                 if (guid_to_name(hdl, tosnap,
2431                     drrb->drr_fromguid, zc.zc_string) != 0) {
2432                         zcmd_free_nvlists(&zc);
2433                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2434                             "local origin for clone %s does not exist"),
2435                             zc.zc_value);
2436                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2437                 }
2438                 if (flags.verbose)
2439                         (void) printf("found clone origin %s\n", zc.zc_string);
2440         }
2441
2442         stream_wantsnewfs = (drrb->drr_fromguid == NULL ||
2443             (drrb->drr_flags & DRR_FLAG_CLONE));
2444
2445         if (stream_wantsnewfs) {
2446                 /*
2447                  * if the parent fs does not exist, look for it based on
2448                  * the parent snap GUID
2449                  */
2450                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2451                     "cannot receive new filesystem stream"));
2452
2453                 (void) strcpy(zc.zc_name, zc.zc_value);
2454                 cp = strrchr(zc.zc_name, '/');
2455                 if (cp)
2456                         *cp = '\0';
2457                 if (cp &&
2458                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2459                         char suffix[ZFS_MAXNAMELEN];
2460                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
2461                         if (guid_to_name(hdl, tosnap, parent_snapguid,
2462                             zc.zc_value) == 0) {
2463                                 *strchr(zc.zc_value, '@') = '\0';
2464                                 (void) strcat(zc.zc_value, suffix);
2465                         }
2466                 }
2467         } else {
2468                 /*
2469                  * if the fs does not exist, look for it based on the
2470                  * fromsnap GUID
2471                  */
2472                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2473                     "cannot receive incremental stream"));
2474
2475                 (void) strcpy(zc.zc_name, zc.zc_value);
2476                 *strchr(zc.zc_name, '@') = '\0';
2477
2478                 /*
2479                  * If the exact receive path was specified and this is the
2480                  * topmost path in the stream, then if the fs does not exist we
2481                  * should look no further.
2482                  */
2483                 if ((flags.isprefix || (*(chopprefix = drrb->drr_toname +
2484                     strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
2485                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2486                         char snap[ZFS_MAXNAMELEN];
2487                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
2488                         if (guid_to_name(hdl, tosnap, drrb->drr_fromguid,
2489                             zc.zc_value) == 0) {
2490                                 *strchr(zc.zc_value, '@') = '\0';
2491                                 (void) strcat(zc.zc_value, snap);
2492                         }
2493                 }
2494         }
2495
2496         (void) strcpy(zc.zc_name, zc.zc_value);
2497         *strchr(zc.zc_name, '@') = '\0';
2498
2499         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2500                 zfs_handle_t *zhp;
2501
2502                 /*
2503                  * Destination fs exists.  Therefore this should either
2504                  * be an incremental, or the stream specifies a new fs
2505                  * (full stream or clone) and they want us to blow it
2506                  * away (and have therefore specified -F and removed any
2507                  * snapshots).
2508                  */
2509                 if (stream_wantsnewfs) {
2510                         if (!flags.force) {
2511                                 zcmd_free_nvlists(&zc);
2512                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2513                                     "destination '%s' exists\n"
2514                                     "must specify -F to overwrite it"),
2515                                     zc.zc_name);
2516                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2517                         }
2518                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2519                             &zc) == 0) {
2520                                 zcmd_free_nvlists(&zc);
2521                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2522                                     "destination has snapshots (eg. %s)\n"
2523                                     "must destroy them to overwrite it"),
2524                                     zc.zc_name);
2525                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2526                         }
2527                 }
2528
2529                 if ((zhp = zfs_open(hdl, zc.zc_name,
2530                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
2531                         zcmd_free_nvlists(&zc);
2532                         return (-1);
2533                 }
2534
2535                 if (stream_wantsnewfs &&
2536                     zhp->zfs_dmustats.dds_origin[0]) {
2537                         zcmd_free_nvlists(&zc);
2538                         zfs_close(zhp);
2539                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2540                             "destination '%s' is a clone\n"
2541                             "must destroy it to overwrite it"),
2542                             zc.zc_name);
2543                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2544                 }
2545
2546                 if (!flags.dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
2547                     stream_wantsnewfs) {
2548                         /* We can't do online recv in this case */
2549                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
2550                         if (clp == NULL) {
2551                                 zfs_close(zhp);
2552                                 zcmd_free_nvlists(&zc);
2553                                 return (-1);
2554                         }
2555                         if (changelist_prefix(clp) != 0) {
2556                                 changelist_free(clp);
2557                                 zfs_close(zhp);
2558                                 zcmd_free_nvlists(&zc);
2559                                 return (-1);
2560                         }
2561                 }
2562                 zfs_close(zhp);
2563         } else {
2564                 /*
2565                  * Destination filesystem does not exist.  Therefore we better
2566                  * be creating a new filesystem (either from a full backup, or
2567                  * a clone).  It would therefore be invalid if the user
2568                  * specified only the pool name (i.e. if the destination name
2569                  * contained no slash character).
2570                  */
2571                 if (!stream_wantsnewfs ||
2572                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
2573                         zcmd_free_nvlists(&zc);
2574                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2575                             "destination '%s' does not exist"), zc.zc_name);
2576                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
2577                 }
2578
2579                 /*
2580                  * Trim off the final dataset component so we perform the
2581                  * recvbackup ioctl to the filesystems's parent.
2582                  */
2583                 *cp = '\0';
2584
2585                 if (flags.isprefix && !flags.istail && !flags.dryrun &&
2586                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
2587                         zcmd_free_nvlists(&zc);
2588                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
2589                 }
2590
2591                 newfs = B_TRUE;
2592         }
2593
2594         zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
2595         zc.zc_cookie = infd;
2596         zc.zc_guid = flags.force;
2597         if (flags.verbose) {
2598                 (void) printf("%s %s stream of %s into %s\n",
2599                     flags.dryrun ? "would receive" : "receiving",
2600                     drrb->drr_fromguid ? "incremental" : "full",
2601                     drrb->drr_toname, zc.zc_value);
2602                 (void) fflush(stdout);
2603         }
2604
2605         if (flags.dryrun) {
2606                 zcmd_free_nvlists(&zc);
2607                 return (recv_skip(hdl, infd, flags.byteswap));
2608         }
2609
2610         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
2611         zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
2612
2613         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
2614         ioctl_errno = errno;
2615         prop_errflags = (zprop_errflags_t)zc.zc_obj;
2616
2617         if (err == 0) {
2618                 nvlist_t *prop_errors;
2619                 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
2620                     zc.zc_nvlist_dst_size, &prop_errors, 0));
2621
2622                 nvpair_t *prop_err = NULL;
2623
2624                 while ((prop_err = nvlist_next_nvpair(prop_errors,
2625                     prop_err)) != NULL) {
2626                         char tbuf[1024];
2627                         zfs_prop_t prop;
2628                         int intval;
2629
2630                         prop = zfs_name_to_prop(nvpair_name(prop_err));
2631                         (void) nvpair_value_int32(prop_err, &intval);
2632                         if (strcmp(nvpair_name(prop_err),
2633                             ZPROP_N_MORE_ERRORS) == 0) {
2634                                 trunc_prop_errs(intval);
2635                                 break;
2636                         } else {
2637                                 (void) snprintf(tbuf, sizeof (tbuf),
2638                                     dgettext(TEXT_DOMAIN,
2639                                     "cannot receive %s property on %s"),
2640                                     nvpair_name(prop_err), zc.zc_name);
2641                                 zfs_setprop_error(hdl, prop, intval, tbuf);
2642                         }
2643                 }
2644                 nvlist_free(prop_errors);
2645         }
2646
2647         zc.zc_nvlist_dst = 0;
2648         zc.zc_nvlist_dst_size = 0;
2649         zcmd_free_nvlists(&zc);
2650
2651         if (err == 0 && snapprops_nvlist) {
2652                 zfs_cmd_t zc2 = { 0 };
2653
2654                 (void) strcpy(zc2.zc_name, zc.zc_value);
2655                 zc2.zc_cookie = B_TRUE; /* received */
2656                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
2657                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
2658                         zcmd_free_nvlists(&zc2);
2659                 }
2660         }
2661
2662         if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
2663                 /*
2664                  * It may be that this snapshot already exists,
2665                  * in which case we want to consume & ignore it
2666                  * rather than failing.
2667                  */
2668                 avl_tree_t *local_avl;
2669                 nvlist_t *local_nv, *fs;
2670                 cp = strchr(zc.zc_value, '@');
2671
2672                 /*
2673                  * XXX Do this faster by just iterating over snaps in
2674                  * this fs.  Also if zc_value does not exist, we will
2675                  * get a strange "does not exist" error message.
2676                  */
2677                 *cp = '\0';
2678                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
2679                     &local_nv, &local_avl) == 0) {
2680                         *cp = '@';
2681                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
2682                         fsavl_destroy(local_avl);
2683                         nvlist_free(local_nv);
2684
2685                         if (fs != NULL) {
2686                                 if (flags.verbose) {
2687                                         (void) printf("snap %s already exists; "
2688                                             "ignoring\n", zc.zc_value);
2689                                 }
2690                                 err = ioctl_err = recv_skip(hdl, infd,
2691                                     flags.byteswap);
2692                         }
2693                 }
2694                 *cp = '@';
2695         }
2696
2697         if (ioctl_err != 0) {
2698                 switch (ioctl_errno) {
2699                 case ENODEV:
2700                         cp = strchr(zc.zc_value, '@');
2701                         *cp = '\0';
2702                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2703                             "most recent snapshot of %s does not\n"
2704                             "match incremental source"), zc.zc_value);
2705                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2706                         *cp = '@';
2707                         break;
2708                 case ETXTBSY:
2709                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2710                             "destination %s has been modified\n"
2711                             "since most recent snapshot"), zc.zc_name);
2712                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2713                         break;
2714                 case EEXIST:
2715                         cp = strchr(zc.zc_value, '@');
2716                         if (newfs) {
2717                                 /* it's the containing fs that exists */
2718                                 *cp = '\0';
2719                         }
2720                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2721                             "destination already exists"));
2722                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
2723                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
2724                             zc.zc_value);
2725                         *cp = '@';
2726                         break;
2727                 case EINVAL:
2728                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2729                         break;
2730                 case ECKSUM:
2731                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2732                             "invalid stream (checksum mismatch)"));
2733                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2734                         break;
2735                 case ENOTSUP:
2736                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2737                             "pool must be upgraded to receive this stream."));
2738                         (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
2739                         break;
2740                 case EDQUOT:
2741                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2742                             "destination %s space quota exceeded"), zc.zc_name);
2743                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
2744                         break;
2745                 default:
2746                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
2747                 }
2748         }
2749
2750         /*
2751          * Mount the target filesystem (if created).  Also mount any
2752          * children of the target filesystem if we did a replication
2753          * receive (indicated by stream_avl being non-NULL).
2754          */
2755         cp = strchr(zc.zc_value, '@');
2756         if (cp && (ioctl_err == 0 || !newfs)) {
2757                 zfs_handle_t *h;
2758
2759                 *cp = '\0';
2760                 h = zfs_open(hdl, zc.zc_value,
2761                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
2762                 if (h != NULL) {
2763                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
2764                                 *cp = '@';
2765                         } else if (newfs || stream_avl) {
2766                                 /*
2767                                  * Track the first/top of hierarchy fs,
2768                                  * for mounting and sharing later.
2769                                  */
2770                                 if (top_zfs && *top_zfs == NULL)
2771                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
2772                         }
2773                         zfs_close(h);
2774                 }
2775                 *cp = '@';
2776         }
2777
2778         if (clp) {
2779                 err |= changelist_postfix(clp);
2780                 changelist_free(clp);
2781         }
2782
2783         if (prop_errflags & ZPROP_ERR_NOCLEAR) {
2784                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
2785                     "failed to clear unreceived properties on %s"),
2786                     zc.zc_name);
2787                 (void) fprintf(stderr, "\n");
2788         }
2789         if (prop_errflags & ZPROP_ERR_NORESTORE) {
2790                 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
2791                     "failed to restore original properties on %s"),
2792                     zc.zc_name);
2793                 (void) fprintf(stderr, "\n");
2794         }
2795
2796         if (err || ioctl_err)
2797                 return (-1);
2798
2799         if (flags.verbose) {
2800                 char buf1[64];
2801                 char buf2[64];
2802                 uint64_t bytes = zc.zc_cookie;
2803                 time_t delta = time(NULL) - begin_time;
2804                 if (delta == 0)
2805                         delta = 1;
2806                 zfs_nicenum(bytes, buf1, sizeof (buf1));
2807                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
2808
2809                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
2810                     buf1, delta, buf2);
2811         }
2812
2813         return (0);
2814 }
2815
2816 static int
2817 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
2818     int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2819     char **top_zfs)
2820 {
2821         int err;
2822         dmu_replay_record_t drr, drr_noswap;
2823         struct drr_begin *drrb = &drr.drr_u.drr_begin;
2824         char errbuf[1024];
2825         zio_cksum_t zcksum = { 0 };
2826         uint64_t featureflags;
2827         int hdrtype;
2828
2829         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2830             "cannot receive"));
2831
2832         if (flags.isprefix &&
2833             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
2834                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
2835                     "(%s) does not exist"), tosnap);
2836                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2837         }
2838
2839         /* read in the BEGIN record */
2840         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
2841             &zcksum)))
2842                 return (err);
2843
2844         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
2845                 /* It's the double end record at the end of a package */
2846                 return (ENODATA);
2847         }
2848
2849         /* the kernel needs the non-byteswapped begin record */
2850         drr_noswap = drr;
2851
2852         flags.byteswap = B_FALSE;
2853         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
2854                 /*
2855                  * We computed the checksum in the wrong byteorder in
2856                  * recv_read() above; do it again correctly.
2857                  */
2858                 bzero(&zcksum, sizeof (zio_cksum_t));
2859                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
2860                 flags.byteswap = B_TRUE;
2861
2862                 drr.drr_type = BSWAP_32(drr.drr_type);
2863                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
2864                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
2865                 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
2866                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
2867                 drrb->drr_type = BSWAP_32(drrb->drr_type);
2868                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
2869                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
2870                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
2871         }
2872
2873         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
2874                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2875                     "stream (bad magic number)"));
2876                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2877         }
2878
2879         featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
2880         hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
2881
2882         if (!DMU_STREAM_SUPPORTED(featureflags) ||
2883             (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
2884                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2885                     "stream has unsupported feature, feature flags = %lx"),
2886                     featureflags);
2887                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2888         }
2889
2890         if (strchr(drrb->drr_toname, '@') == NULL) {
2891                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2892                     "stream (bad snapshot name)"));
2893                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2894         }
2895
2896         if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
2897                 char nonpackage_sendfs[ZFS_MAXNAMELEN];
2898                 if (sendfs == NULL) {
2899                         /*
2900                          * We were not called from zfs_receive_package(). Get
2901                          * the fs specified by 'zfs send'.
2902                          */
2903                         char *cp;
2904                         (void) strlcpy(nonpackage_sendfs,
2905                             drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN);
2906                         if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
2907                                 *cp = '\0';
2908                         sendfs = nonpackage_sendfs;
2909                 }
2910                 return (zfs_receive_one(hdl, infd, tosnap, flags,
2911                     &drr, &drr_noswap, sendfs, stream_nv, stream_avl,
2912                     top_zfs));
2913         } else {
2914                 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
2915                     DMU_COMPOUNDSTREAM);
2916                 return (zfs_receive_package(hdl, infd, tosnap, flags,
2917                     &drr, &zcksum, top_zfs));
2918         }
2919 }
2920
2921 /*
2922  * Restores a backup of tosnap from the file descriptor specified by infd.
2923  * Return 0 on total success, -2 if some things couldn't be
2924  * destroyed/renamed/promoted, -1 if some things couldn't be received.
2925  * (-1 will override -2).
2926  */
2927 int
2928 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
2929     int infd, avl_tree_t *stream_avl)
2930 {
2931         char *top_zfs = NULL;
2932         int err;
2933
2934         err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL,
2935             stream_avl, &top_zfs);
2936
2937         if (err == 0 && !flags.nomount && top_zfs) {
2938                 zfs_handle_t *zhp;
2939                 prop_changelist_t *clp;
2940
2941                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
2942                 if (zhp != NULL) {
2943                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
2944                             CL_GATHER_MOUNT_ALWAYS, 0);
2945                         zfs_close(zhp);
2946                         if (clp != NULL) {
2947                                 /* mount and share received datasets */
2948                                 err = changelist_postfix(clp);
2949                                 changelist_free(clp);
2950                         }
2951                 }
2952                 if (zhp == NULL || clp == NULL || err)
2953                         err = -1;
2954         }
2955         if (top_zfs)
2956                 free(top_zfs);
2957
2958         return (err);
2959 }