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 2008 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26
  27 #include <assert.h>
  28 #include <ctype.h>
  29 #include <errno.h>
  30 #include <libdevinfo.h>
  31 #include <libintl.h>
  32 #include <stdio.h>
  33 #include <stdlib.h>
  34 #include <strings.h>
  35 #include <unistd.h>
  36 #include <stddef.h>
  37 #include <fcntl.h>
  38 #include <sys/mount.h>
  39 #include <sys/mntent.h>
  40 #include <sys/mnttab.h>
  41 #include <sys/avl.h>
  42 #include <stddef.h>
  43
  44 #include <libzfs.h>
  45
  46 #include "zfs_namecheck.h"
  47 #include "zfs_prop.h"
  48 #include "libzfs_impl.h"
  49
  50 #include <fletcher.c> /* XXX */
  51
  52 static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t,
  53     int, avl_tree_t *, char **);
  54
  55 /*
  56  * Routines for dealing with the AVL tree of fs-nvlists
  57  */
  58 typedef struct fsavl_node {
  59         avl_node_t fn_node;
  60         nvlist_t *fn_nvfs;
  61         char *fn_snapname;
  62         uint64_t fn_guid;
  63 } fsavl_node_t;
  64
  65 static int
  66 fsavl_compare(const void *arg1, const void *arg2)
  67 {
  68         const fsavl_node_t *fn1 = arg1;
  69         const fsavl_node_t *fn2 = arg2;
  70
  71         if (fn1->fn_guid > fn2->fn_guid)
  72                 return (+1);
  73         else if (fn1->fn_guid < fn2->fn_guid)
  74                 return (-1);
  75         else
  76                 return (0);
  77 }
  78
  79 /*
  80  * Given the GUID of a snapshot, find its containing filesystem and
  81  * (optionally) name.
  82  */
  83 static nvlist_t *
  84 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
  85 {
  86         fsavl_node_t fn_find;
  87         fsavl_node_t *fn;
  88
  89         fn_find.fn_guid = snapguid;
  90
  91         fn = avl_find(avl, &fn_find, NULL);
  92         if (fn) {
  93                 if (snapname)
  94                         *snapname = fn->fn_snapname;
  95                 return (fn->fn_nvfs);
  96         }
  97         return (NULL);
  98 }
  99
 100 static void
 101 fsavl_destroy(avl_tree_t *avl)
 102 {
 103         fsavl_node_t *fn;
 104         void *cookie;
 105
 106         if (avl == NULL)
 107                 return;
 108
 109         cookie = NULL;
 110         while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
 111                 free(fn);
 112         avl_destroy(avl);
 113         free(avl);
 114 }
 115
 116 static avl_tree_t *
 117 fsavl_create(nvlist_t *fss)
 118 {
 119         avl_tree_t *fsavl;
 120         nvpair_t *fselem = NULL;
 121
 122         if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
 123                 return (NULL);
 124
 125         avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
 126             offsetof(fsavl_node_t, fn_node));
 127
 128         while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
 129                 nvlist_t *nvfs, *snaps;
 130                 nvpair_t *snapelem = NULL;
 131
 132                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
 133                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
 134
 135                 while ((snapelem =
 136                     nvlist_next_nvpair(snaps, snapelem)) != NULL) {
 137                         fsavl_node_t *fn;
 138                         uint64_t guid;
 139
 140                         VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
 141                         if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
 142                                 fsavl_destroy(fsavl);
 143                                 return (NULL);
 144                         }
 145                         fn->fn_nvfs = nvfs;
 146                         fn->fn_snapname = nvpair_name(snapelem);
 147                         fn->fn_guid = guid;
 148
 149                         /*
 150                          * Note: if there are multiple snaps with the
 151                          * same GUID, we ignore all but one.
 152                          */
 153                         if (avl_find(fsavl, fn, NULL) == NULL)
 154                                 avl_add(fsavl, fn);
 155                         else
 156                                 free(fn);
 157                 }
 158         }
 159
 160         return (fsavl);
 161 }
 162
 163 /*
 164  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
 165  */
 166 typedef struct send_data {
 167         uint64_t parent_fromsnap_guid;
 168         nvlist_t *parent_snaps;
 169         nvlist_t *fss;
 170         nvlist_t *snapprops;
 171         const char *fromsnap;
 172         const char *tosnap;
 173
 174         /*
 175          * The header nvlist is of the following format:
 176          * {
 177          *   "tosnap" -> string
 178          *   "fromsnap" -> string (if incremental)
 179          *   "fss" -> {
 180          *      id -> {
 181          *
 182          *       "name" -> string (full name; for debugging)
 183          *       "parentfromsnap" -> number (guid of fromsnap in parent)
 184          *
 185          *       "props" -> { name -> value (only if set here) }
 186          *       "snaps" -> { name (lastname) -> number (guid) }
 187          *       "snapprops" -> { name (lastname) -> { name -> value } }
 188          *
 189          *       "origin" -> number (guid) (if clone)
 190          *       "sent" -> boolean (not on-disk)
 191          *      }
 192          *   }
 193          * }
 194          *
 195          */
 196 } send_data_t;
 197
 198 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
 199
 200 static int
 201 send_iterate_snap(zfs_handle_t *zhp, void *arg)
 202 {
 203         send_data_t *sd = arg;
 204         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 205         char *snapname;
 206         nvlist_t *nv;
 207
 208         snapname = strrchr(zhp->zfs_name, '@')+1;
 209
 210         VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 211         /*
 212          * NB: if there is no fromsnap here (it's a newly created fs in
 213          * an incremental replication), we will substitute the tosnap.
 214          */
 215         if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
 216             (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
 217             strcmp(snapname, sd->tosnap) == 0)) {
 218                 sd->parent_fromsnap_guid = guid;
 219         }
 220
 221         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 222         send_iterate_prop(zhp, nv);
 223         VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
 224         nvlist_free(nv);
 225
 226         zfs_close(zhp);
 227         return (0);
 228 }
 229
 230 static void
 231 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 232 {
 233         nvpair_t *elem = NULL;
 234
 235         while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
 236                 char *propname = nvpair_name(elem);
 237                 zfs_prop_t prop = zfs_name_to_prop(propname);
 238                 nvlist_t *propnv;
 239
 240                 if (!zfs_prop_user(propname) && zfs_prop_readonly(prop))
 241                         continue;
 242
 243                 verify(nvpair_value_nvlist(elem, &propnv) == 0);
 244                 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION) {
 245                         /* these guys are modifyable, but have no source */
 246                         uint64_t value;
 247                         verify(nvlist_lookup_uint64(propnv,
 248                             ZPROP_VALUE, &value) == 0);
 249                         if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
 250                                 continue;
 251                 } else {
 252                         char *source;
 253                         if (nvlist_lookup_string(propnv,
 254                             ZPROP_SOURCE, &source) != 0)
 255                                 continue;
 256                         if (strcmp(source, zhp->zfs_name) != 0)
 257                                 continue;
 258                 }
 259
 260                 if (zfs_prop_user(propname) ||
 261                     zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
 262                         char *value;
 263                         verify(nvlist_lookup_string(propnv,
 264                             ZPROP_VALUE, &value) == 0);
 265                         VERIFY(0 == nvlist_add_string(nv, propname, value));
 266                 } else {
 267                         uint64_t value;
 268                         verify(nvlist_lookup_uint64(propnv,
 269                             ZPROP_VALUE, &value) == 0);
 270                         VERIFY(0 == nvlist_add_uint64(nv, propname, value));
 271                 }
 272         }
 273 }
 274
 275 static int
 276 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 277 {
 278         send_data_t *sd = arg;
 279         nvlist_t *nvfs, *nv;
 280         int rv;
 281         uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
 282         uint64_t guid = zhp->zfs_dmustats.dds_guid;
 283         char guidstring[64];
 284
 285         VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 286         VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 287         VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
 288             sd->parent_fromsnap_guid));
 289
 290         if (zhp->zfs_dmustats.dds_origin[0]) {
 291                 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 292                     zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
 293                 if (origin == NULL)
 294                         return (-1);
 295                 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 296                     origin->zfs_dmustats.dds_guid));
 297         }
 298
 299         /* iterate over props */
 300         VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
 301         send_iterate_prop(zhp, nv);
 302         VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
 303         nvlist_free(nv);
 304
 305         /* iterate over snaps, and set sd->parent_fromsnap_guid */
 306         sd->parent_fromsnap_guid = 0;
 307         VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
 308         VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
 309         (void) zfs_iter_snapshots(zhp, send_iterate_snap, sd);
 310         VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
 311         VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
 312         nvlist_free(sd->parent_snaps);
 313         nvlist_free(sd->snapprops);
 314
 315         /* add this fs to nvlist */
 316         (void) snprintf(guidstring, sizeof (guidstring),
 317             "0x%llx", (longlong_t)guid);
 318         VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
 319         nvlist_free(nvfs);
 320
 321         /* iterate over children */
 322         rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 323
 324         sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
 325
 326         zfs_close(zhp);
 327         return (rv);
 328 }
 329
 330 static int
 331 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 332     const char *tosnap, nvlist_t **nvlp, avl_tree_t **avlp)
 333 {
 334         zfs_handle_t *zhp;
 335         send_data_t sd = { 0 };
 336         int error;
 337
 338         zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
 339         if (zhp == NULL)
 340                 return (EZFS_BADTYPE);
 341
 342         VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
 343         sd.fromsnap = fromsnap;
 344         sd.tosnap = tosnap;
 345
 346         if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 347                 nvlist_free(sd.fss);
 348                 if (avlp != NULL)
 349                         *avlp = NULL;
 350                 *nvlp = NULL;
 351                 return (error);
 352         }
 353
 354         if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
 355                 nvlist_free(sd.fss);
 356                 *nvlp = NULL;
 357                 return (EZFS_NOMEM);
 358         }
 359
 360         *nvlp = sd.fss;
 361         return (0);
 362 }
 363
 364 /*
 365  * Routines for dealing with the sorted snapshot functionality
 366  */
 367 typedef struct zfs_node {
 368         zfs_handle_t    *zn_handle;
 369         avl_node_t      zn_avlnode;
 370 } zfs_node_t;
 371
 372 static int
 373 zfs_sort_snaps(zfs_handle_t *zhp, void *data)
 374 {
 375         avl_tree_t *avl = data;
 376         zfs_node_t *node = zfs_alloc(zhp->zfs_hdl, sizeof (zfs_node_t));
 377
 378         node->zn_handle = zhp;
 379         avl_add(avl, node);
 380         return (0);
 381 }
 382
 383 /* ARGSUSED */
 384 static int
 385 zfs_snapshot_compare(const void *larg, const void *rarg)
 386 {
 387         zfs_handle_t *l = ((zfs_node_t *)larg)->zn_handle;
 388         zfs_handle_t *r = ((zfs_node_t *)rarg)->zn_handle;
 389         uint64_t lcreate, rcreate;
 390
 391         /*
 392          * Sort them according to creation time.  We use the hidden
 393          * CREATETXG property to get an absolute ordering of snapshots.
 394          */
 395         lcreate = zfs_prop_get_int(l, ZFS_PROP_CREATETXG);
 396         rcreate = zfs_prop_get_int(r, ZFS_PROP_CREATETXG);
 397
 398         if (lcreate < rcreate)
 399                 return (-1);
 400         else if (lcreate > rcreate)
 401                 return (+1);
 402         else
 403                 return (0);
 404 }
 405
 406 static int
 407 zfs_iter_snapshots_sorted(zfs_handle_t *zhp, zfs_iter_f callback, void *data)
 408 {
 409         int ret = 0;
 410         zfs_node_t *node;
 411         avl_tree_t avl;
 412         void *cookie = NULL;
 413
 414         avl_create(&avl, zfs_snapshot_compare,
 415             sizeof (zfs_node_t), offsetof(zfs_node_t, zn_avlnode));
 416
 417         ret = zfs_iter_snapshots(zhp, zfs_sort_snaps, &avl);
 418
 419         for (node = avl_first(&avl); node != NULL; node = AVL_NEXT(&avl, node))
 420                 ret |= callback(node->zn_handle, data);
 421
 422         while ((node = avl_destroy_nodes(&avl, &cookie)) != NULL)
 423                 free(node);
 424
 425         avl_destroy(&avl);
 426
 427         return (ret);
 428 }
 429
 430 /*
 431  * Routines specific to "zfs send"
 432  */
 433 typedef struct send_dump_data {
 434         /* these are all just the short snapname (the part after the @) */
 435         const char *fromsnap;
 436         const char *tosnap;
 437         char lastsnap[ZFS_MAXNAMELEN];
 438         boolean_t seenfrom, seento, replicate, doall, fromorigin;
 439         boolean_t verbose;
 440         int outfd;
 441         boolean_t err;
 442         nvlist_t *fss;
 443         avl_tree_t *fsavl;
 444 } send_dump_data_t;
 445
 446 /*
 447  * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
 448  * NULL) to the file descriptor specified by outfd.
 449  */
 450 static int
 451 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, boolean_t fromorigin,
 452     int outfd)
 453 {
 454         zfs_cmd_t zc = { 0 };
 455         libzfs_handle_t *hdl = zhp->zfs_hdl;
 456
 457         assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
 458         assert(fromsnap == NULL || fromsnap[0] == '\0' || !fromorigin);
 459
 460         (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 461         if (fromsnap)
 462                 (void) strlcpy(zc.zc_value, fromsnap, sizeof (zc.zc_value));
 463         zc.zc_cookie = outfd;
 464         zc.zc_obj = fromorigin;
 465
 466         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_SEND, &zc) != 0) {
 467                 char errbuf[1024];
 468                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 469                     "warning: cannot send '%s'"), zhp->zfs_name);
 470
 471                 switch (errno) {
 472
 473                 case EXDEV:
 474                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 475                             "not an earlier snapshot from the same fs"));
 476                         return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
 477
 478                 case ENOENT:
 479                         if (zfs_dataset_exists(hdl, zc.zc_name,
 480                             ZFS_TYPE_SNAPSHOT)) {
 481                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 482                                     "incremental source (@%s) does not exist"),
 483                                     zc.zc_value);
 484                         }
 485                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
 486
 487                 case EDQUOT:
 488                 case EFBIG:
 489                 case EIO:
 490                 case ENOLINK:
 491                 case ENOSPC:
 492                 case ENOSTR:
 493                 case ENXIO:
 494                 case EPIPE:
 495                 case ERANGE:
 496                 case EFAULT:
 497                 case EROFS:
 498                         zfs_error_aux(hdl, strerror(errno));
 499                         return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
 500
 501                 default:
 502                         return (zfs_standard_error(hdl, errno, errbuf));
 503                 }
 504         }
 505
 506         return (0);
 507 }
 508
 509 static int
 510 dump_snapshot(zfs_handle_t *zhp, void *arg)
 511 {
 512         send_dump_data_t *sdd = arg;
 513         const char *thissnap;
 514         int err;
 515
 516         thissnap = strchr(zhp->zfs_name, '@') + 1;
 517
 518         if (sdd->fromsnap && !sdd->seenfrom &&
 519             strcmp(sdd->fromsnap, thissnap) == 0) {
 520                 sdd->seenfrom = B_TRUE;
 521                 (void) strcpy(sdd->lastsnap, thissnap);
 522                 zfs_close(zhp);
 523                 return (0);
 524         }
 525
 526         if (sdd->seento || !sdd->seenfrom) {
 527                 zfs_close(zhp);
 528                 return (0);
 529         }
 530
 531         /* send it */
 532         if (sdd->verbose) {
 533                 (void) fprintf(stderr, "sending from @%s to %s\n",
 534                     sdd->lastsnap, zhp->zfs_name);
 535         }
 536
 537         err = dump_ioctl(zhp, sdd->lastsnap,
 538             sdd->lastsnap[0] == '\0' && (sdd->fromorigin || sdd->replicate),
 539             sdd->outfd);
 540
 541         if (!sdd->seento && strcmp(sdd->tosnap, thissnap) == 0)
 542                 sdd->seento = B_TRUE;
 543
 544         (void) strcpy(sdd->lastsnap, thissnap);
 545         zfs_close(zhp);
 546         return (err);
 547 }
 548
 549 static int
 550 dump_filesystem(zfs_handle_t *zhp, void *arg)
 551 {
 552         int rv = 0;
 553         send_dump_data_t *sdd = arg;
 554         boolean_t missingfrom = B_FALSE;
 555         zfs_cmd_t zc = { 0 };
 556
 557         (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
 558             zhp->zfs_name, sdd->tosnap);
 559         if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
 560                 (void) fprintf(stderr, "WARNING: "
 561                     "could not send %s@%s: does not exist\n",
 562                     zhp->zfs_name, sdd->tosnap);
 563                 sdd->err = B_TRUE;
 564                 return (0);
 565         }
 566
 567         if (sdd->replicate && sdd->fromsnap) {
 568                 /*
 569                  * If this fs does not have fromsnap, and we're doing
 570                  * recursive, we need to send a full stream from the
 571                  * beginning (or an incremental from the origin if this
 572                  * is a clone).  If we're doing non-recursive, then let
 573                  * them get the error.
 574                  */
 575                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
 576                     zhp->zfs_name, sdd->fromsnap);
 577                 if (ioctl(zhp->zfs_hdl->libzfs_fd,
 578                     ZFS_IOC_OBJSET_STATS, &zc) != 0) {
 579                         missingfrom = B_TRUE;
 580                 }
 581         }
 582
 583         if (sdd->doall) {
 584                 sdd->seenfrom = sdd->seento = sdd->lastsnap[0] = 0;
 585                 if (sdd->fromsnap == NULL || missingfrom)
 586                         sdd->seenfrom = B_TRUE;
 587
 588                 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
 589                 if (!sdd->seenfrom) {
 590                         (void) fprintf(stderr,
 591                             "WARNING: could not send %s@%s:\n"
 592                             "incremental source (%s@%s) does not exist\n",
 593                             zhp->zfs_name, sdd->tosnap,
 594                             zhp->zfs_name, sdd->fromsnap);
 595                         sdd->err = B_TRUE;
 596                 } else if (!sdd->seento) {
 597                         (void) fprintf(stderr,
 598                             "WARNING: could not send %s@%s:\n"
 599                             "incremental source (%s@%s) "
 600                             "is not earlier than it\n",
 601                             zhp->zfs_name, sdd->tosnap,
 602                             zhp->zfs_name, sdd->fromsnap);
 603                         sdd->err = B_TRUE;
 604                 }
 605         } else {
 606                 zfs_handle_t *snapzhp;
 607                 char snapname[ZFS_MAXNAMELEN];
 608
 609                 (void) snprintf(snapname, sizeof (snapname), "%s@%s",
 610                     zfs_get_name(zhp), sdd->tosnap);
 611                 snapzhp = zfs_open(zhp->zfs_hdl, snapname, ZFS_TYPE_SNAPSHOT);
 612                 if (snapzhp == NULL) {
 613                         rv = -1;
 614                 } else {
 615                         rv = dump_ioctl(snapzhp,
 616                             missingfrom ? NULL : sdd->fromsnap,
 617                             sdd->fromorigin || missingfrom,
 618                             sdd->outfd);
 619                         sdd->seento = B_TRUE;
 620                         zfs_close(snapzhp);
 621                 }
 622         }
 623
 624         return (rv);
 625 }
 626
 627 static int
 628 dump_filesystems(zfs_handle_t *rzhp, void *arg)
 629 {
 630         send_dump_data_t *sdd = arg;
 631         nvpair_t *fspair;
 632         boolean_t needagain, progress;
 633
 634         if (!sdd->replicate)
 635                 return (dump_filesystem(rzhp, sdd));
 636
 637 again:
 638         needagain = progress = B_FALSE;
 639         for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
 640             fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
 641                 nvlist_t *fslist;
 642                 char *fsname;
 643                 zfs_handle_t *zhp;
 644                 int err;
 645                 uint64_t origin_guid = 0;
 646                 nvlist_t *origin_nv;
 647
 648                 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
 649                 if (nvlist_lookup_boolean(fslist, "sent") == 0)
 650                         continue;
 651
 652                 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
 653                 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
 654
 655                 origin_nv = fsavl_find(sdd->fsavl, origin_guid, NULL);
 656                 if (origin_nv &&
 657                     nvlist_lookup_boolean(origin_nv, "sent") == ENOENT) {
 658                         /*
 659                          * origin has not been sent yet;
 660                          * skip this clone.
 661                          */
 662                         needagain = B_TRUE;
 663                         continue;
 664                 }
 665
 666                 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
 667                 if (zhp == NULL)
 668                         return (-1);
 669                 err = dump_filesystem(zhp, sdd);
 670                 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
 671                 progress = B_TRUE;
 672                 zfs_close(zhp);
 673                 if (err)
 674                         return (err);
 675         }
 676         if (needagain) {
 677                 assert(progress);
 678                 goto again;
 679         }
 680         return (0);
 681 }
 682
 683 /*
 684  * Dumps a backup of tosnap, incremental from fromsnap if it isn't NULL.
 685  * If 'doall', dump all intermediate snaps.
 686  * If 'replicate', dump special header and do recursively.
 687  */
 688 int
 689 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
 690     boolean_t replicate, boolean_t doall, boolean_t fromorigin,
 691     boolean_t verbose, int outfd)
 692 {
 693         char errbuf[1024];
 694         send_dump_data_t sdd = { 0 };
 695         int err;
 696         nvlist_t *fss = NULL;
 697         avl_tree_t *fsavl = NULL;
 698
 699         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 700             "cannot send '%s'"), zhp->zfs_name);
 701
 702         if (fromsnap && fromsnap[0] == '\0') {
 703                 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
 704                     "zero-length incremental source"));
 705                 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
 706         }
 707
 708         if (replicate || doall) {
 709                 dmu_replay_record_t drr = { 0 };
 710                 char *packbuf = NULL;
 711                 size_t buflen = 0;
 712                 zio_cksum_t zc = { 0 };
 713
 714                 assert(fromsnap || doall);
 715
 716                 if (replicate) {
 717                         nvlist_t *hdrnv;
 718
 719                         VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
 720                         if (fromsnap) {
 721                                 VERIFY(0 == nvlist_add_string(hdrnv,
 722                                     "fromsnap", fromsnap));
 723                         }
 724                         VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
 725
 726                         err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
 727                             fromsnap, tosnap, &fss, &fsavl);
 728                         if (err)
 729                                 return (err);
 730                         VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
 731                         err = nvlist_pack(hdrnv, &packbuf, &buflen,
 732                             NV_ENCODE_XDR, 0);
 733                         nvlist_free(hdrnv);
 734                         if (err) {
 735                                 fsavl_destroy(fsavl);
 736                                 nvlist_free(fss);
 737                                 return (zfs_standard_error(zhp->zfs_hdl,
 738                                     err, errbuf));
 739                         }
 740                 }
 741
 742                 /* write first begin record */
 743                 drr.drr_type = DRR_BEGIN;
 744                 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
 745                 drr.drr_u.drr_begin.drr_version = DMU_BACKUP_HEADER_VERSION;
 746                 (void) snprintf(drr.drr_u.drr_begin.drr_toname,
 747                     sizeof (drr.drr_u.drr_begin.drr_toname),
 748                     "%s@%s", zhp->zfs_name, tosnap);
 749                 drr.drr_payloadlen = buflen;
 750                 fletcher_4_incremental_native(&drr, sizeof (drr), &zc);
 751                 err = write(outfd, &drr, sizeof (drr));
 752
 753                 /* write header nvlist */
 754                 if (err != -1) {
 755                         fletcher_4_incremental_native(packbuf, buflen, &zc);
 756                         err = write(outfd, packbuf, buflen);
 757                 }
 758                 free(packbuf);
 759                 if (err == -1) {
 760                         fsavl_destroy(fsavl);
 761                         nvlist_free(fss);
 762                         return (zfs_standard_error(zhp->zfs_hdl,
 763                             errno, errbuf));
 764                 }
 765
 766                 /* write end record */
 767                 if (err != -1) {
 768                         bzero(&drr, sizeof (drr));
 769                         drr.drr_type = DRR_END;
 770                         drr.drr_u.drr_end.drr_checksum = zc;
 771                         err = write(outfd, &drr, sizeof (drr));
 772                         if (err == -1) {
 773                                 fsavl_destroy(fsavl);
 774                                 nvlist_free(fss);
 775                                 return (zfs_standard_error(zhp->zfs_hdl,
 776                                     errno, errbuf));
 777                         }
 778                 }
 779         }
 780
 781         /* dump each stream */
 782         sdd.fromsnap = fromsnap;
 783         sdd.tosnap = tosnap;
 784         sdd.outfd = outfd;
 785         sdd.replicate = replicate;
 786         sdd.doall = doall;
 787         sdd.fromorigin = fromorigin;
 788         sdd.fss = fss;
 789         sdd.fsavl = fsavl;
 790         sdd.verbose = verbose;
 791         err = dump_filesystems(zhp, &sdd);
 792         fsavl_destroy(fsavl);
 793         nvlist_free(fss);
 794
 795         if (replicate || doall) {
 796                 /*
 797                  * write final end record.  NB: want to do this even if
 798                  * there was some error, because it might not be totally
 799                  * failed.
 800                  */
 801                 dmu_replay_record_t drr = { 0 };
 802                 drr.drr_type = DRR_END;
 803                 if (write(outfd, &drr, sizeof (drr)) == -1) {
 804                         return (zfs_standard_error(zhp->zfs_hdl,
 805                             errno, errbuf));
 806                 }
 807         }
 808
 809         return (err || sdd.err);
 810 }
 811
 812 /*
 813  * Routines specific to "zfs recv"
 814  */
 815
 816 static int
 817 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
 818     boolean_t byteswap, zio_cksum_t *zc)
 819 {
 820         char *cp = buf;
 821         int rv;
 822         int len = ilen;
 823
 824         do {
 825                 rv = read(fd, cp, len);
 826                 cp += rv;
 827                 len -= rv;
 828         } while (rv > 0);
 829
 830         if (rv < 0 || len != 0) {
 831                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 832                     "failed to read from stream"));
 833                 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
 834                     "cannot receive")));
 835         }
 836
 837         if (zc) {
 838                 if (byteswap)
 839                         fletcher_4_incremental_byteswap(buf, ilen, zc);
 840                 else
 841                         fletcher_4_incremental_native(buf, ilen, zc);
 842         }
 843         return (0);
 844 }
 845
 846 static int
 847 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
 848     boolean_t byteswap, zio_cksum_t *zc)
 849 {
 850         char *buf;
 851         int err;
 852
 853         buf = zfs_alloc(hdl, len);
 854         if (buf == NULL)
 855                 return (ENOMEM);
 856
 857         err = recv_read(hdl, fd, buf, len, byteswap, zc);
 858         if (err != 0) {
 859                 free(buf);
 860                 return (err);
 861         }
 862
 863         err = nvlist_unpack(buf, len, nvp, 0);
 864         free(buf);
 865         if (err != 0) {
 866                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
 867                     "stream (malformed nvlist)"));
 868                 return (EINVAL);
 869         }
 870         return (0);
 871 }
 872
 873 static int
 874 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
 875     int baselen, char *newname, recvflags_t flags)
 876 {
 877         static int seq;
 878         zfs_cmd_t zc = { 0 };
 879         int err;
 880         prop_changelist_t *clp;
 881         zfs_handle_t *zhp;
 882
 883         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
 884         if (zhp == NULL)
 885                 return (-1);
 886         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
 887             flags.force ? MS_FORCE : 0);
 888         zfs_close(zhp);
 889         if (clp == NULL)
 890                 return (-1);
 891         err = changelist_prefix(clp);
 892         if (err)
 893                 return (err);
 894
 895         if (tryname) {
 896                 (void) strcpy(newname, tryname);
 897
 898                 zc.zc_objset_type = DMU_OST_ZFS;
 899                 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
 900                 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
 901
 902                 if (flags.verbose) {
 903                         (void) printf("attempting rename %s to %s\n",
 904                             zc.zc_name, zc.zc_value);
 905                 }
 906                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
 907                 if (err == 0)
 908                         changelist_rename(clp, name, tryname);
 909         } else {
 910                 err = ENOENT;
 911         }
 912
 913         if (err != 0 && strncmp(name+baselen, "recv-", 5) != 0) {
 914                 seq++;
 915
 916                 (void) strncpy(newname, name, baselen);
 917                 (void) snprintf(newname+baselen, ZFS_MAXNAMELEN-baselen,
 918                     "recv-%u-%u", getpid(), seq);
 919                 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
 920
 921                 if (flags.verbose) {
 922                         (void) printf("failed - trying rename %s to %s\n",
 923                             zc.zc_name, zc.zc_value);
 924                 }
 925                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
 926                 if (err == 0)
 927                         changelist_rename(clp, name, newname);
 928                 if (err && flags.verbose) {
 929                         (void) printf("failed (%u) - "
 930                             "will try again on next pass\n", errno);
 931                 }
 932                 err = EAGAIN;
 933         } else if (flags.verbose) {
 934                 if (err == 0)
 935                         (void) printf("success\n");
 936                 else
 937                         (void) printf("failed (%u)\n", errno);
 938         }
 939
 940         (void) changelist_postfix(clp);
 941         changelist_free(clp);
 942
 943         return (err);
 944 }
 945
 946 static int
 947 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
 948     char *newname, recvflags_t flags)
 949 {
 950         zfs_cmd_t zc = { 0 };
 951         int err = 0;
 952         prop_changelist_t *clp;
 953         zfs_handle_t *zhp;
 954
 955         zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
 956         if (zhp == NULL)
 957                 return (-1);
 958         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
 959             flags.force ? MS_FORCE : 0);
 960         zfs_close(zhp);
 961         if (clp == NULL)
 962                 return (-1);
 963         err = changelist_prefix(clp);
 964         if (err)
 965                 return (err);
 966
 967         zc.zc_objset_type = DMU_OST_ZFS;
 968         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
 969
 970         if (flags.verbose)
 971                 (void) printf("attempting destroy %s\n", zc.zc_name);
 972         err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
 973
 974         if (err == 0) {
 975                 if (flags.verbose)
 976                         (void) printf("success\n");
 977                 changelist_remove(clp, zc.zc_name);
 978         }
 979
 980         (void) changelist_postfix(clp);
 981         changelist_free(clp);
 982
 983         if (err != 0)
 984                 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
 985
 986         return (err);
 987 }
 988
 989 typedef struct guid_to_name_data {
 990         uint64_t guid;
 991         char *name;
 992 } guid_to_name_data_t;
 993
 994 static int
 995 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
 996 {
 997         guid_to_name_data_t *gtnd = arg;
 998         int err;
 999
1000         if (zhp->zfs_dmustats.dds_guid == gtnd->guid) {
1001                 (void) strcpy(gtnd->name, zhp->zfs_name);
1002                 return (EEXIST);
1003         }
1004         err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
1005         zfs_close(zhp);
1006         return (err);
1007 }
1008
1009 static int
1010 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
1011     char *name)
1012 {
1013         /* exhaustive search all local snapshots */
1014         guid_to_name_data_t gtnd;
1015         int err = 0;
1016         zfs_handle_t *zhp;
1017         char *cp;
1018
1019         gtnd.guid = guid;
1020         gtnd.name = name;
1021
1022         if (strchr(parent, '@') == NULL) {
1023                 zhp = make_dataset_handle(hdl, parent);
1024                 if (zhp != NULL) {
1025                         err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1026                         zfs_close(zhp);
1027                         if (err == EEXIST)
1028                                 return (0);
1029                 }
1030         }
1031
1032         cp = strchr(parent, '/');
1033         if (cp)
1034                 *cp = '\0';
1035         zhp = make_dataset_handle(hdl, parent);
1036         if (cp)
1037                 *cp = '/';
1038
1039         if (zhp) {
1040                 err = zfs_iter_children(zhp, guid_to_name_cb, &gtnd);
1041                 zfs_close(zhp);
1042         }
1043
1044         return (err == EEXIST ? 0 : ENOENT);
1045
1046 }
1047
1048 /*
1049  * Return true if dataset guid1 is created before guid2.
1050  */
1051 static int
1052 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
1053     uint64_t guid1, uint64_t guid2)
1054 {
1055         nvlist_t *nvfs;
1056         char *fsname, *snapname;
1057         char buf[ZFS_MAXNAMELEN];
1058         int rv;
1059         zfs_node_t zn1, zn2;
1060
1061         if (guid2 == 0)
1062                 return (0);
1063         if (guid1 == 0)
1064                 return (1);
1065
1066         nvfs = fsavl_find(avl, guid1, &snapname);
1067         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1068         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1069         zn1.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1070         if (zn1.zn_handle == NULL)
1071                 return (-1);
1072
1073         nvfs = fsavl_find(avl, guid2, &snapname);
1074         VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1075         (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
1076         zn2.zn_handle = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
1077         if (zn2.zn_handle == NULL) {
1078                 zfs_close(zn2.zn_handle);
1079                 return (-1);
1080         }
1081
1082         rv = (zfs_snapshot_compare(&zn1, &zn2) == -1);
1083
1084         zfs_close(zn1.zn_handle);
1085         zfs_close(zn2.zn_handle);
1086
1087         return (rv);
1088 }
1089
1090 static int
1091 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
1092     recvflags_t flags, nvlist_t *stream_nv, avl_tree_t *stream_avl)
1093 {
1094         nvlist_t *local_nv;
1095         avl_tree_t *local_avl;
1096         nvpair_t *fselem, *nextfselem;
1097         char *tosnap, *fromsnap;
1098         char newname[ZFS_MAXNAMELEN];
1099         int error;
1100         boolean_t needagain, progress;
1101
1102         VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
1103         VERIFY(0 == nvlist_lookup_string(stream_nv, "tosnap", &tosnap));
1104
1105         if (flags.dryrun)
1106                 return (0);
1107
1108 again:
1109         needagain = progress = B_FALSE;
1110
1111         if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
1112             &local_nv, &local_avl)) != 0)
1113                 return (error);
1114
1115         /*
1116          * Process deletes and renames
1117          */
1118         for (fselem = nvlist_next_nvpair(local_nv, NULL);
1119             fselem; fselem = nextfselem) {
1120                 nvlist_t *nvfs, *snaps;
1121                 nvlist_t *stream_nvfs = NULL;
1122                 nvpair_t *snapelem, *nextsnapelem;
1123                 uint64_t fromguid = 0;
1124                 uint64_t originguid = 0;
1125                 uint64_t stream_originguid = 0;
1126                 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
1127                 char *fsname, *stream_fsname;
1128
1129                 nextfselem = nvlist_next_nvpair(local_nv, fselem);
1130
1131                 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
1132                 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
1133                 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
1134                 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
1135                     &parent_fromsnap_guid));
1136                 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
1137
1138                 /*
1139                  * First find the stream's fs, so we can check for
1140                  * a different origin (due to "zfs promote")
1141                  */
1142                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1143                     snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
1144                         uint64_t thisguid;
1145
1146                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1147                         stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
1148
1149                         if (stream_nvfs != NULL)
1150                                 break;
1151                 }
1152
1153                 /* check for promote */
1154                 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
1155                     &stream_originguid);
1156                 if (stream_nvfs && originguid != stream_originguid) {
1157                         switch (created_before(hdl, local_avl,
1158                             stream_originguid, originguid)) {
1159                         case 1: {
1160                                 /* promote it! */
1161                                 zfs_cmd_t zc = { 0 };
1162                                 nvlist_t *origin_nvfs;
1163                                 char *origin_fsname;
1164
1165                                 if (flags.verbose)
1166                                         (void) printf("promoting %s\n", fsname);
1167
1168                                 origin_nvfs = fsavl_find(local_avl, originguid,
1169                                     NULL);
1170                                 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
1171                                     "name", &origin_fsname));
1172                                 (void) strlcpy(zc.zc_value, origin_fsname,
1173                                     sizeof (zc.zc_value));
1174                                 (void) strlcpy(zc.zc_name, fsname,
1175                                     sizeof (zc.zc_name));
1176                                 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
1177                                 if (error == 0)
1178                                         progress = B_TRUE;
1179                                 break;
1180                         }
1181                         default:
1182                                 break;
1183                         case -1:
1184                                 fsavl_destroy(local_avl);
1185                                 nvlist_free(local_nv);
1186                                 return (-1);
1187                         }
1188                         /*
1189                          * We had/have the wrong origin, therefore our
1190                          * list of snapshots is wrong.  Need to handle
1191                          * them on the next pass.
1192                          */
1193                         needagain = B_TRUE;
1194                         continue;
1195                 }
1196
1197                 for (snapelem = nvlist_next_nvpair(snaps, NULL);
1198                     snapelem; snapelem = nextsnapelem) {
1199                         uint64_t thisguid;
1200                         char *stream_snapname;
1201                         nvlist_t *found, *props;
1202
1203                         nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
1204
1205                         VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
1206                         found = fsavl_find(stream_avl, thisguid,
1207                             &stream_snapname);
1208
1209                         /* check for delete */
1210                         if (found == NULL) {
1211                                 char name[ZFS_MAXNAMELEN];
1212
1213                                 if (!flags.force)
1214                                         continue;
1215
1216                                 (void) snprintf(name, sizeof (name), "%s@%s",
1217                                     fsname, nvpair_name(snapelem));
1218
1219                                 error = recv_destroy(hdl, name,
1220                                     strlen(fsname)+1, newname, flags);
1221                                 if (error)
1222                                         needagain = B_TRUE;
1223                                 else
1224                                         progress = B_TRUE;
1225                                 continue;
1226                         }
1227
1228                         stream_nvfs = found;
1229
1230                         if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
1231                             &props) && 0 == nvlist_lookup_nvlist(props,
1232                             stream_snapname, &props)) {
1233                                 zfs_cmd_t zc = { 0 };
1234
1235                                 zc.zc_cookie = B_TRUE; /* clear current props */
1236                                 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
1237                                     "%s@%s", fsname, nvpair_name(snapelem));
1238                                 if (zcmd_write_src_nvlist(hdl, &zc,
1239                                     props) == 0) {
1240                                         (void) zfs_ioctl(hdl,
1241                                             ZFS_IOC_SET_PROP, &zc);
1242                                         zcmd_free_nvlists(&zc);
1243                                 }
1244                         }
1245
1246                         /* check for different snapname */
1247                         if (strcmp(nvpair_name(snapelem),
1248                             stream_snapname) != 0) {
1249                                 char name[ZFS_MAXNAMELEN];
1250                                 char tryname[ZFS_MAXNAMELEN];
1251
1252                                 (void) snprintf(name, sizeof (name), "%s@%s",
1253                                     fsname, nvpair_name(snapelem));
1254                                 (void) snprintf(tryname, sizeof (name), "%s@%s",
1255                                     fsname, stream_snapname);
1256
1257                                 error = recv_rename(hdl, name, tryname,
1258                                     strlen(fsname)+1, newname, flags);
1259                                 if (error)
1260                                         needagain = B_TRUE;
1261                                 else
1262                                         progress = B_TRUE;
1263                         }
1264
1265                         if (strcmp(stream_snapname, fromsnap) == 0)
1266                                 fromguid = thisguid;
1267                 }
1268
1269                 /* check for delete */
1270                 if (stream_nvfs == NULL) {
1271                         if (!flags.force)
1272                                 continue;
1273
1274                         error = recv_destroy(hdl, fsname, strlen(tofs)+1,
1275                             newname, flags);
1276                         if (error)
1277                                 needagain = B_TRUE;
1278                         else
1279                                 progress = B_TRUE;
1280                         continue;
1281                 }
1282
1283                 if (fromguid == 0 && flags.verbose) {
1284                         (void) printf("local fs %s does not have fromsnap "
1285                             "(%s in stream); must have been deleted locally; "
1286                             "ignoring\n", fsname, fromsnap);
1287                         continue;
1288                 }
1289
1290                 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
1291                     "name", &stream_fsname));
1292                 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
1293                     "parentfromsnap", &stream_parent_fromsnap_guid));
1294
1295                 /* check for rename */
1296                 if ((stream_parent_fromsnap_guid != 0 &&
1297                     stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
1298                     strcmp(strrchr(fsname, '/'),
1299                     strrchr(stream_fsname, '/')) != 0) {
1300                         nvlist_t *parent;
1301                         char tryname[ZFS_MAXNAMELEN];
1302
1303                         parent = fsavl_find(local_avl,
1304                             stream_parent_fromsnap_guid, NULL);
1305                         /*
1306                          * NB: parent might not be found if we used the
1307                          * tosnap for stream_parent_fromsnap_guid,
1308                          * because the parent is a newly-created fs;
1309                          * we'll be able to rename it after we recv the
1310                          * new fs.
1311                          */
1312                         if (parent != NULL) {
1313                                 char *pname;
1314
1315                                 VERIFY(0 == nvlist_lookup_string(parent, "name",
1316                                     &pname));
1317                                 (void) snprintf(tryname, sizeof (tryname),
1318                                     "%s%s", pname, strrchr(stream_fsname, '/'));
1319                         } else {
1320                                 tryname[0] = '\0';
1321                                 if (flags.verbose) {
1322                                         (void) printf("local fs %s new parent "
1323                                             "not found\n", fsname);
1324                                 }
1325                         }
1326
1327                         error = recv_rename(hdl, fsname, tryname,
1328                             strlen(tofs)+1, newname, flags);
1329                         if (error)
1330                                 needagain = B_TRUE;
1331                         else
1332                                 progress = B_TRUE;
1333                 }
1334         }
1335
1336         fsavl_destroy(local_avl);
1337         nvlist_free(local_nv);
1338
1339         if (needagain && progress) {
1340                 /* do another pass to fix up temporary names */
1341                 if (flags.verbose)
1342                         (void) printf("another pass:\n");
1343                 goto again;
1344         }
1345
1346         return (needagain);
1347 }
1348
1349 static int
1350 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
1351     recvflags_t flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
1352     char **top_zfs)
1353 {
1354         nvlist_t *stream_nv = NULL;
1355         avl_tree_t *stream_avl = NULL;
1356         char *fromsnap = NULL;
1357         char tofs[ZFS_MAXNAMELEN];
1358         char errbuf[1024];
1359         dmu_replay_record_t drre;
1360         int error;
1361         boolean_t anyerr = B_FALSE;
1362         boolean_t softerr = B_FALSE;
1363
1364         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1365             "cannot receive"));
1366
1367         if (strchr(destname, '@')) {
1368                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1369                     "can not specify snapshot name for multi-snapshot stream"));
1370                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
1371         }
1372
1373         assert(drr->drr_type == DRR_BEGIN);
1374         assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
1375         assert(drr->drr_u.drr_begin.drr_version == DMU_BACKUP_HEADER_VERSION);
1376
1377         /*
1378          * Read in the nvlist from the stream.
1379          */
1380         if (drr->drr_payloadlen != 0) {
1381                 if (!flags.isprefix) {
1382                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1383                             "must use -d to receive replication "
1384                             "(send -R) stream"));
1385                         return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
1386                 }
1387
1388                 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
1389                     &stream_nv, flags.byteswap, zc);
1390                 if (error) {
1391                         error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
1392                         goto out;
1393                 }
1394         }
1395
1396         /*
1397          * Read in the end record and verify checksum.
1398          */
1399         if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
1400             flags.byteswap, NULL)))
1401                 goto out;
1402         if (flags.byteswap) {
1403                 drre.drr_type = BSWAP_32(drre.drr_type);
1404                 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
1405                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
1406                 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
1407                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
1408                 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
1409                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
1410                 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
1411                     BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
1412         }
1413         if (drre.drr_type != DRR_END) {
1414                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
1415                 goto out;
1416         }
1417         if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
1418                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1419                     "incorrect header checksum"));
1420                 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
1421                 goto out;
1422         }
1423
1424         (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
1425
1426         if (drr->drr_payloadlen != 0) {
1427                 nvlist_t *stream_fss;
1428
1429                 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
1430                     &stream_fss));
1431                 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
1432                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1433                             "couldn't allocate avl tree"));
1434                         error = zfs_error(hdl, EZFS_NOMEM, errbuf);
1435                         goto out;
1436                 }
1437
1438                 if (fromsnap != NULL) {
1439                         (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN);
1440                         if (flags.isprefix) {
1441                                 int i = strcspn(drr->drr_u.drr_begin.drr_toname,
1442                                     "/@");
1443                                 /* zfs_receive_one() will create_parents() */
1444                                 (void) strlcat(tofs,
1445                                     &drr->drr_u.drr_begin.drr_toname[i],
1446                                     ZFS_MAXNAMELEN);
1447                                 *strchr(tofs, '@') = '\0';
1448                         }
1449                         softerr = recv_incremental_replication(hdl, tofs,
1450                             flags, stream_nv, stream_avl);
1451                 }
1452         }
1453
1454
1455         /* Finally, receive each contained stream */
1456         do {
1457                 /*
1458                  * we should figure out if it has a recoverable
1459                  * error, in which case do a recv_skip() and drive on.
1460                  * Note, if we fail due to already having this guid,
1461                  * zfs_receive_one() will take care of it (ie,
1462                  * recv_skip() and return 0).
1463                  */
1464                 error = zfs_receive_impl(hdl, destname, flags, fd,
1465                     stream_avl, top_zfs);
1466                 if (error == ENODATA) {
1467                         error = 0;
1468                         break;
1469                 }
1470                 anyerr |= error;
1471         } while (error == 0);
1472
1473         if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
1474                 /*
1475                  * Now that we have the fs's they sent us, try the
1476                  * renames again.
1477                  */
1478                 softerr = recv_incremental_replication(hdl, tofs, flags,
1479                     stream_nv, stream_avl);
1480         }
1481
1482 out:
1483         fsavl_destroy(stream_avl);
1484         if (stream_nv)
1485                 nvlist_free(stream_nv);
1486         if (softerr)
1487                 error = -2;
1488         if (anyerr)
1489                 error = -1;
1490         return (error);
1491 }
1492
1493 static int
1494 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
1495 {
1496         dmu_replay_record_t *drr;
1497         void *buf = malloc(1<<20);
1498
1499         /* XXX would be great to use lseek if possible... */
1500         drr = buf;
1501
1502         while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
1503             byteswap, NULL) == 0) {
1504                 if (byteswap)
1505                         drr->drr_type = BSWAP_32(drr->drr_type);
1506
1507                 switch (drr->drr_type) {
1508                 case DRR_BEGIN:
1509                         /* NB: not to be used on v2 stream packages */
1510                         assert(drr->drr_payloadlen == 0);
1511                         break;
1512
1513                 case DRR_END:
1514                         free(buf);
1515                         return (0);
1516
1517                 case DRR_OBJECT:
1518                         if (byteswap) {
1519                                 drr->drr_u.drr_object.drr_bonuslen =
1520                                     BSWAP_32(drr->drr_u.drr_object.
1521                                     drr_bonuslen);
1522                         }
1523                         (void) recv_read(hdl, fd, buf,
1524                             P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
1525                             B_FALSE, NULL);
1526                         break;
1527
1528                 case DRR_WRITE:
1529                         if (byteswap) {
1530                                 drr->drr_u.drr_write.drr_length =
1531                                     BSWAP_64(drr->drr_u.drr_write.drr_length);
1532                         }
1533                         (void) recv_read(hdl, fd, buf,
1534                             drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
1535                         break;
1536
1537                 case DRR_FREEOBJECTS:
1538                 case DRR_FREE:
1539                         break;
1540
1541                 default:
1542                         assert(!"invalid record type");
1543                 }
1544         }
1545
1546         free(buf);
1547         return (-1);
1548 }
1549
1550 /*
1551  * Restores a backup of tosnap from the file descriptor specified by infd.
1552  */
1553 static int
1554 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
1555     recvflags_t flags, dmu_replay_record_t *drr,
1556     dmu_replay_record_t *drr_noswap, avl_tree_t *stream_avl,
1557     char **top_zfs)
1558 {
1559         zfs_cmd_t zc = { 0 };
1560         time_t begin_time;
1561         int ioctl_err, ioctl_errno, err, choplen;
1562         char *cp;
1563         struct drr_begin *drrb = &drr->drr_u.drr_begin;
1564         char errbuf[1024];
1565         char chopprefix[ZFS_MAXNAMELEN];
1566         boolean_t newfs = B_FALSE;
1567         boolean_t stream_wantsnewfs;
1568         uint64_t parent_snapguid = 0;
1569         prop_changelist_t *clp = NULL;
1570         nvlist_t *snapprops_nvlist = NULL;
1571
1572         begin_time = time(NULL);
1573
1574         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1575             "cannot receive"));
1576
1577         if (stream_avl != NULL) {
1578                 char *snapname;
1579                 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
1580                     &snapname);
1581                 nvlist_t *props;
1582                 int ret;
1583
1584                 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
1585                     &parent_snapguid);
1586                 err = nvlist_lookup_nvlist(fs, "props", &props);
1587                 if (err)
1588                         VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
1589
1590                 if (flags.canmountoff) {
1591                         VERIFY(0 == nvlist_add_uint64(props,
1592                             zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
1593                 }
1594                 ret = zcmd_write_src_nvlist(hdl, &zc, props);
1595                 if (err)
1596                         nvlist_free(props);
1597
1598                 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
1599                         VERIFY(0 == nvlist_lookup_nvlist(props,
1600                             snapname, &snapprops_nvlist));
1601                 }
1602
1603                 if (ret != 0)
1604                         return (-1);
1605         }
1606
1607         /*
1608          * Determine how much of the snapshot name stored in the stream
1609          * we are going to tack on to the name they specified on the
1610          * command line, and how much we are going to chop off.
1611          *
1612          * If they specified a snapshot, chop the entire name stored in
1613          * the stream.
1614          */
1615         (void) strcpy(chopprefix, drrb->drr_toname);
1616         if (flags.isprefix) {
1617                 /*
1618                  * They specified a fs with -d, we want to tack on
1619                  * everything but the pool name stored in the stream
1620                  */
1621                 if (strchr(tosnap, '@')) {
1622                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
1623                             "argument - snapshot not allowed with -d"));
1624                         return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
1625                 }
1626                 cp = strchr(chopprefix, '/');
1627                 if (cp == NULL)
1628                         cp = strchr(chopprefix, '@');
1629                 *cp = '\0';
1630         } else if (strchr(tosnap, '@') == NULL) {
1631                 /*
1632                  * If they specified a filesystem without -d, we want to
1633                  * tack on everything after the fs specified in the
1634                  * first name from the stream.
1635                  */
1636                 cp = strchr(chopprefix, '@');
1637                 *cp = '\0';
1638         }
1639         choplen = strlen(chopprefix);
1640
1641         /*
1642          * Determine name of destination snapshot, store in zc_value.
1643          */
1644         (void) strcpy(zc.zc_value, tosnap);
1645         (void) strncat(zc.zc_value, drrb->drr_toname+choplen,
1646             sizeof (zc.zc_value));
1647         if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
1648                 zcmd_free_nvlists(&zc);
1649                 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
1650         }
1651
1652         /*
1653          * Determine the name of the origin snapshot, store in zc_string.
1654          */
1655         if (drrb->drr_flags & DRR_FLAG_CLONE) {
1656                 if (guid_to_name(hdl, tosnap,
1657                     drrb->drr_fromguid, zc.zc_string) != 0) {
1658                         zcmd_free_nvlists(&zc);
1659                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1660                             "local origin for clone %s does not exist"),
1661                             zc.zc_value);
1662                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
1663                 }
1664                 if (flags.verbose)
1665                         (void) printf("found clone origin %s\n", zc.zc_string);
1666         }
1667
1668         stream_wantsnewfs = (drrb->drr_fromguid == NULL ||
1669             (drrb->drr_flags & DRR_FLAG_CLONE));
1670
1671         if (stream_wantsnewfs) {
1672                 /*
1673                  * if the parent fs does not exist, look for it based on
1674                  * the parent snap GUID
1675                  */
1676                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1677                     "cannot receive new filesystem stream"));
1678
1679                 (void) strcpy(zc.zc_name, zc.zc_value);
1680                 cp = strrchr(zc.zc_name, '/');
1681                 if (cp)
1682                         *cp = '\0';
1683                 if (cp &&
1684                     !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
1685                         char suffix[ZFS_MAXNAMELEN];
1686                         (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
1687                         if (guid_to_name(hdl, tosnap, parent_snapguid,
1688                             zc.zc_value) == 0) {
1689                                 *strchr(zc.zc_value, '@') = '\0';
1690                                 (void) strcat(zc.zc_value, suffix);
1691                         }
1692                 }
1693         } else {
1694                 /*
1695                  * if the fs does not exist, look for it based on the
1696                  * fromsnap GUID
1697                  */
1698                 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1699                     "cannot receive incremental stream"));
1700
1701                 (void) strcpy(zc.zc_name, zc.zc_value);
1702                 *strchr(zc.zc_name, '@') = '\0';
1703
1704                 if (!zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
1705                         char snap[ZFS_MAXNAMELEN];
1706                         (void) strcpy(snap, strchr(zc.zc_value, '@'));
1707                         if (guid_to_name(hdl, tosnap, drrb->drr_fromguid,
1708                             zc.zc_value) == 0) {
1709                                 *strchr(zc.zc_value, '@') = '\0';
1710                                 (void) strcat(zc.zc_value, snap);
1711                         }
1712                 }
1713         }
1714
1715         (void) strcpy(zc.zc_name, zc.zc_value);
1716         *strchr(zc.zc_name, '@') = '\0';
1717
1718         if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
1719                 zfs_handle_t *zhp;
1720                 /*
1721                  * Destination fs exists.  Therefore this should either
1722                  * be an incremental, or the stream specifies a new fs
1723                  * (full stream or clone) and they want us to blow it
1724                  * away (and have therefore specified -F and removed any
1725                  * snapshots).
1726                  */
1727
1728                 if (stream_wantsnewfs) {
1729                         if (!flags.force) {
1730                                 zcmd_free_nvlists(&zc);
1731                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1732                                     "destination '%s' exists\n"
1733                                     "must specify -F to overwrite it"),
1734                                     zc.zc_name);
1735                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
1736                         }
1737                         if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
1738                             &zc) == 0) {
1739                                 zcmd_free_nvlists(&zc);
1740                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1741                                     "destination has snapshots (eg. %s)\n"
1742                                     "must destroy them to overwrite it"),
1743                                     zc.zc_name);
1744                                 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
1745                         }
1746                 }
1747
1748                 if ((zhp = zfs_open(hdl, zc.zc_name,
1749                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
1750                         zcmd_free_nvlists(&zc);
1751                         return (-1);
1752                 }
1753
1754                 if (stream_wantsnewfs &&
1755                     zhp->zfs_dmustats.dds_origin[0]) {
1756                         zcmd_free_nvlists(&zc);
1757                         zfs_close(zhp);
1758                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1759                             "destination '%s' is a clone\n"
1760                             "must destroy it to overwrite it"),
1761                             zc.zc_name);
1762                         return (zfs_error(hdl, EZFS_EXISTS, errbuf));
1763                 }
1764
1765                 if (!flags.dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
1766                     stream_wantsnewfs) {
1767                         /* We can't do online recv in this case */
1768                         clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
1769                         if (clp == NULL) {
1770                                 zcmd_free_nvlists(&zc);
1771                                 return (-1);
1772                         }
1773                         if (changelist_prefix(clp) != 0) {
1774                                 changelist_free(clp);
1775                                 zcmd_free_nvlists(&zc);
1776                                 return (-1);
1777                         }
1778                 }
1779                 if (!flags.dryrun && zhp->zfs_type == ZFS_TYPE_VOLUME &&
1780                     zvol_remove_link(hdl, zhp->zfs_name) != 0) {
1781                         zfs_close(zhp);
1782                         zcmd_free_nvlists(&zc);
1783                         return (-1);
1784                 }
1785                 zfs_close(zhp);
1786         } else {
1787                 /*
1788                  * Destination filesystem does not exist.  Therefore we better
1789                  * be creating a new filesystem (either from a full backup, or
1790                  * a clone).  It would therefore be invalid if the user
1791                  * specified only the pool name (i.e. if the destination name
1792                  * contained no slash character).
1793                  */
1794                 if (!stream_wantsnewfs ||
1795                     (cp = strrchr(zc.zc_name, '/')) == NULL) {
1796                         zcmd_free_nvlists(&zc);
1797                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1798                             "destination '%s' does not exist"), zc.zc_name);
1799                         return (zfs_error(hdl, EZFS_NOENT, errbuf));
1800                 }
1801
1802                 /*
1803                  * Trim off the final dataset component so we perform the
1804                  * recvbackup ioctl to the filesystems's parent.
1805                  */
1806                 *cp = '\0';
1807
1808                 if (flags.isprefix && !flags.dryrun &&
1809                     create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
1810                         zcmd_free_nvlists(&zc);
1811                         return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
1812                 }
1813
1814                 newfs = B_TRUE;
1815         }
1816
1817         zc.zc_begin_record = drr_noswap->drr_u.drr_begin;
1818         zc.zc_cookie = infd;
1819         zc.zc_guid = flags.force;
1820         if (flags.verbose) {
1821                 (void) printf("%s %s stream of %s into %s\n",
1822                     flags.dryrun ? "would receive" : "receiving",
1823                     drrb->drr_fromguid ? "incremental" : "full",
1824                     drrb->drr_toname, zc.zc_value);
1825                 (void) fflush(stdout);
1826         }
1827
1828         if (flags.dryrun) {
1829                 zcmd_free_nvlists(&zc);
1830                 return (recv_skip(hdl, infd, flags.byteswap));
1831         }
1832
1833         err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
1834         ioctl_errno = errno;
1835         zcmd_free_nvlists(&zc);
1836
1837         if (err == 0 && snapprops_nvlist) {
1838                 zfs_cmd_t zc2 = { 0 };
1839
1840                 (void) strcpy(zc2.zc_name, zc.zc_value);
1841                 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
1842                         (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
1843                         zcmd_free_nvlists(&zc2);
1844                 }
1845         }
1846
1847         if (err && (ioctl_errno == ENOENT || ioctl_errno == ENODEV)) {
1848                 /*
1849                  * It may be that this snapshot already exists,
1850                  * in which case we want to consume & ignore it
1851                  * rather than failing.
1852                  */
1853                 avl_tree_t *local_avl;
1854                 nvlist_t *local_nv, *fs;
1855                 char *cp = strchr(zc.zc_value, '@');
1856
1857                 /*
1858                  * XXX Do this faster by just iterating over snaps in
1859                  * this fs.  Also if zc_value does not exist, we will
1860                  * get a strange "does not exist" error message.
1861                  */
1862                 *cp = '\0';
1863                 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL,
1864                     &local_nv, &local_avl) == 0) {
1865                         *cp = '@';
1866                         fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
1867                         fsavl_destroy(local_avl);
1868                         nvlist_free(local_nv);
1869
1870                         if (fs != NULL) {
1871                                 if (flags.verbose) {
1872                                         (void) printf("snap %s already exists; "
1873                                             "ignoring\n", zc.zc_value);
1874                                 }
1875                                 ioctl_err = recv_skip(hdl, infd,
1876                                     flags.byteswap);
1877                         }
1878                 }
1879                 *cp = '@';
1880         }
1881
1882
1883         if (ioctl_err != 0) {
1884                 switch (ioctl_errno) {
1885                 case ENODEV:
1886                         cp = strchr(zc.zc_value, '@');
1887                         *cp = '\0';
1888                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1889                             "most recent snapshot of %s does not\n"
1890                             "match incremental source"), zc.zc_value);
1891                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
1892                         *cp = '@';
1893                         break;
1894                 case ETXTBSY:
1895                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1896                             "destination %s has been modified\n"
1897                             "since most recent snapshot"), zc.zc_name);
1898                         (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
1899                         break;
1900                 case EEXIST:
1901                         cp = strchr(zc.zc_value, '@');
1902                         if (newfs) {
1903                                 /* it's the containing fs that exists */
1904                                 *cp = '\0';
1905                         }
1906                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1907                             "destination already exists"));
1908                         (void) zfs_error_fmt(hdl, EZFS_EXISTS,
1909                             dgettext(TEXT_DOMAIN, "cannot restore to %s"),
1910                             zc.zc_value);
1911                         *cp = '@';
1912                         break;
1913                 case EINVAL:
1914                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
1915                         break;
1916                 case ECKSUM:
1917                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1918                             "invalid stream (checksum mismatch)"));
1919                         (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
1920                         break;
1921                 default:
1922                         (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
1923                 }
1924         }
1925
1926         /*
1927          * Mount or recreate the /dev links for the target filesystem
1928          * (if created, or if we tore them down to do an incremental
1929          * restore), and the /dev links for the new snapshot (if
1930          * created). Also mount any children of the target filesystem
1931          * if we did an incremental receive.
1932          */
1933         cp = strchr(zc.zc_value, '@');
1934         if (cp && (ioctl_err == 0 || !newfs)) {
1935                 zfs_handle_t *h;
1936
1937                 *cp = '\0';
1938                 h = zfs_open(hdl, zc.zc_value,
1939                     ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
1940                 if (h != NULL) {
1941                         if (h->zfs_type == ZFS_TYPE_VOLUME) {
1942                                 *cp = '@';
1943                                 err = zvol_create_link(hdl, h->zfs_name);
1944                                 if (err == 0 && ioctl_err == 0)
1945                                         err = zvol_create_link(hdl,
1946                                             zc.zc_value);
1947                         } else if (newfs) {
1948                                 /*
1949                                  * Track the first/top of hierarchy fs,
1950                                  * for mounting and sharing later.
1951                                  */
1952                                 if (top_zfs && *top_zfs == NULL)
1953                                         *top_zfs = zfs_strdup(hdl, zc.zc_value);
1954                         }
1955                         zfs_close(h);
1956                 }
1957                 *cp = '@';
1958         }
1959
1960         if (clp) {
1961                 err |= changelist_postfix(clp);
1962                 changelist_free(clp);
1963         }
1964
1965         if (err || ioctl_err)
1966                 return (-1);
1967
1968         if (flags.verbose) {
1969                 char buf1[64];
1970                 char buf2[64];
1971                 uint64_t bytes = zc.zc_cookie;
1972                 time_t delta = time(NULL) - begin_time;
1973                 if (delta == 0)
1974                         delta = 1;
1975                 zfs_nicenum(bytes, buf1, sizeof (buf1));
1976                 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
1977
1978                 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
1979                     buf1, delta, buf2);
1980         }
1981
1982         return (0);
1983 }
1984
1985 static int
1986 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
1987     int infd, avl_tree_t *stream_avl, char **top_zfs)
1988 {
1989         int err;
1990         dmu_replay_record_t drr, drr_noswap;
1991         struct drr_begin *drrb = &drr.drr_u.drr_begin;
1992         char errbuf[1024];
1993         zio_cksum_t zcksum = { 0 };
1994
1995         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1996             "cannot receive"));
1997
1998         if (flags.isprefix &&
1999             !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
2000                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
2001                     "(%s) does not exist"), tosnap);
2002                 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2003         }
2004
2005         /* read in the BEGIN record */
2006         if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
2007             &zcksum)))
2008                 return (err);
2009
2010         if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
2011                 /* It's the double end record at the end of a package */
2012                 return (ENODATA);
2013         }
2014
2015         /* the kernel needs the non-byteswapped begin record */
2016         drr_noswap = drr;
2017
2018         flags.byteswap = B_FALSE;
2019         if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
2020                 /*
2021                  * We computed the checksum in the wrong byteorder in
2022                  * recv_read() above; do it again correctly.
2023                  */
2024                 bzero(&zcksum, sizeof (zio_cksum_t));
2025                 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
2026                 flags.byteswap = B_TRUE;
2027
2028                 drr.drr_type = BSWAP_32(drr.drr_type);
2029                 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
2030                 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
2031                 drrb->drr_version = BSWAP_64(drrb->drr_version);
2032                 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
2033                 drrb->drr_type = BSWAP_32(drrb->drr_type);
2034                 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
2035                 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
2036                 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
2037         }
2038
2039         if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
2040                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2041                     "stream (bad magic number)"));
2042                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2043         }
2044
2045         if (strchr(drrb->drr_toname, '@') == NULL) {
2046                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2047                     "stream (bad snapshot name)"));
2048                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2049         }
2050
2051         if (drrb->drr_version == DMU_BACKUP_STREAM_VERSION) {
2052                 return (zfs_receive_one(hdl, infd, tosnap, flags,
2053                     &drr, &drr_noswap, stream_avl, top_zfs));
2054         } else if (drrb->drr_version == DMU_BACKUP_HEADER_VERSION) {
2055                 return (zfs_receive_package(hdl, infd, tosnap, flags,
2056                     &drr, &zcksum, top_zfs));
2057         } else {
2058                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2059                     "stream is unsupported version %llu"),
2060                     drrb->drr_version);
2061                 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
2062         }
2063 }
2064
2065 /*
2066  * Restores a backup of tosnap from the file descriptor specified by infd.
2067  * Return 0 on total success, -2 if some things couldn't be
2068  * destroyed/renamed/promoted, -1 if some things couldn't be received.
2069  * (-1 will override -2).
2070  */
2071 int
2072 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t flags,
2073     int infd, avl_tree_t *stream_avl)
2074 {
2075         char *top_zfs = NULL;
2076         int err;
2077
2078         err = zfs_receive_impl(hdl, tosnap, flags, infd, stream_avl, &top_zfs);
2079
2080         if (err == 0 && top_zfs) {
2081                 zfs_handle_t *zhp;
2082                 prop_changelist_t *clp;
2083
2084                 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
2085                 if (zhp != NULL) {
2086                         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
2087                             CL_GATHER_MOUNT_ALWAYS, 0);
2088                         zfs_close(zhp);
2089                         if (clp != NULL) {
2090                                 /* mount and share received datasets */
2091                                 err = changelist_postfix(clp);
2092                                 changelist_free(clp);
2093                         }
2094                 }
2095                 if (zhp == NULL || clp == NULL || err)
2096                         err = -1;
2097         }
2098         if (top_zfs)
2099                 free(top_zfs);
2100
2101         return (err);
2102 }