libzfs_sendrecv.c revision 352722
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 25 * Copyright (c) 2012, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2012 Pawel Jakub Dawidek. All rights reserved. 27 * Copyright (c) 2013 Steven Hartland. All rights reserved. 28 * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com> 31 */ 32 33#include <assert.h> 34#include <ctype.h> 35#include <errno.h> 36#include <libintl.h> 37#include <stdio.h> 38#include <stdlib.h> 39#include <strings.h> 40#include <unistd.h> 41#include <stddef.h> 42#include <fcntl.h> 43#include <sys/param.h> 44#include <sys/mount.h> 45#include <pthread.h> 46#include <umem.h> 47#include <time.h> 48 49#include <libzfs.h> 50#include <libzfs_core.h> 51 52#include "zfs_namecheck.h" 53#include "zfs_prop.h" 54#include "zfs_fletcher.h" 55#include "libzfs_impl.h" 56#include <zlib.h> 57#include <sha2.h> 58#include <sys/zio_checksum.h> 59#include <sys/ddt.h> 60 61#ifdef __FreeBSD__ 62extern int zfs_ioctl_version; 63#endif 64 65/* in libzfs_dataset.c */ 66extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *); 67/* We need to use something for ENODATA. */ 68#define ENODATA EIDRM 69 70static int zfs_receive_impl(libzfs_handle_t *, const char *, const char *, 71 recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **, int, 72 uint64_t *, const char *); 73static int guid_to_name(libzfs_handle_t *, const char *, 74 uint64_t, boolean_t, char *); 75 76static const zio_cksum_t zero_cksum = { 0 }; 77 78typedef struct dedup_arg { 79 int inputfd; 80 int outputfd; 81 libzfs_handle_t *dedup_hdl; 82} dedup_arg_t; 83 84typedef struct progress_arg { 85 zfs_handle_t *pa_zhp; 86 int pa_fd; 87 boolean_t pa_parsable; 88 boolean_t pa_astitle; 89 uint64_t pa_size; 90} progress_arg_t; 91 92typedef struct dataref { 93 uint64_t ref_guid; 94 uint64_t ref_object; 95 uint64_t ref_offset; 96} dataref_t; 97 98typedef struct dedup_entry { 99 struct dedup_entry *dde_next; 100 zio_cksum_t dde_chksum; 101 uint64_t dde_prop; 102 dataref_t dde_ref; 103} dedup_entry_t; 104 105#define MAX_DDT_PHYSMEM_PERCENT 20 106#define SMALLEST_POSSIBLE_MAX_DDT_MB 128 107 108typedef struct dedup_table { 109 dedup_entry_t **dedup_hash_array; 110 umem_cache_t *ddecache; 111 uint64_t max_ddt_size; /* max dedup table size in bytes */ 112 uint64_t cur_ddt_size; /* current dedup table size in bytes */ 113 uint64_t ddt_count; 114 int numhashbits; 115 boolean_t ddt_full; 116} dedup_table_t; 117 118static int 119high_order_bit(uint64_t n) 120{ 121 int count; 122 123 for (count = 0; n != 0; count++) 124 n >>= 1; 125 return (count); 126} 127 128static size_t 129ssread(void *buf, size_t len, FILE *stream) 130{ 131 size_t outlen; 132 133 if ((outlen = fread(buf, len, 1, stream)) == 0) 134 return (0); 135 136 return (outlen); 137} 138 139static void 140ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp, 141 zio_cksum_t *cs, uint64_t prop, dataref_t *dr) 142{ 143 dedup_entry_t *dde; 144 145 if (ddt->cur_ddt_size >= ddt->max_ddt_size) { 146 if (ddt->ddt_full == B_FALSE) { 147 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 148 "Dedup table full. Deduplication will continue " 149 "with existing table entries")); 150 ddt->ddt_full = B_TRUE; 151 } 152 return; 153 } 154 155 if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT)) 156 != NULL) { 157 assert(*ddepp == NULL); 158 dde->dde_next = NULL; 159 dde->dde_chksum = *cs; 160 dde->dde_prop = prop; 161 dde->dde_ref = *dr; 162 *ddepp = dde; 163 ddt->cur_ddt_size += sizeof (dedup_entry_t); 164 ddt->ddt_count++; 165 } 166} 167 168/* 169 * Using the specified dedup table, do a lookup for an entry with 170 * the checksum cs. If found, return the block's reference info 171 * in *dr. Otherwise, insert a new entry in the dedup table, using 172 * the reference information specified by *dr. 173 * 174 * return value: true - entry was found 175 * false - entry was not found 176 */ 177static boolean_t 178ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs, 179 uint64_t prop, dataref_t *dr) 180{ 181 uint32_t hashcode; 182 dedup_entry_t **ddepp; 183 184 hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits); 185 186 for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL; 187 ddepp = &((*ddepp)->dde_next)) { 188 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) && 189 (*ddepp)->dde_prop == prop) { 190 *dr = (*ddepp)->dde_ref; 191 return (B_TRUE); 192 } 193 } 194 ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr); 195 return (B_FALSE); 196} 197 198static int 199dump_record(dmu_replay_record_t *drr, void *payload, int payload_len, 200 zio_cksum_t *zc, int outfd) 201{ 202 ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 203 ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 204 (void) fletcher_4_incremental_native(drr, 205 offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc); 206 if (drr->drr_type != DRR_BEGIN) { 207 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u. 208 drr_checksum.drr_checksum)); 209 drr->drr_u.drr_checksum.drr_checksum = *zc; 210 } 211 (void) fletcher_4_incremental_native( 212 &drr->drr_u.drr_checksum.drr_checksum, sizeof (zio_cksum_t), zc); 213 if (write(outfd, drr, sizeof (*drr)) == -1) 214 return (errno); 215 if (payload_len != 0) { 216 (void) fletcher_4_incremental_native(payload, payload_len, zc); 217 if (write(outfd, payload, payload_len) == -1) 218 return (errno); 219 } 220 return (0); 221} 222 223/* 224 * This function is started in a separate thread when the dedup option 225 * has been requested. The main send thread determines the list of 226 * snapshots to be included in the send stream and makes the ioctl calls 227 * for each one. But instead of having the ioctl send the output to the 228 * the output fd specified by the caller of zfs_send()), the 229 * ioctl is told to direct the output to a pipe, which is read by the 230 * alternate thread running THIS function. This function does the 231 * dedup'ing by: 232 * 1. building a dedup table (the DDT) 233 * 2. doing checksums on each data block and inserting a record in the DDT 234 * 3. looking for matching checksums, and 235 * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever 236 * a duplicate block is found. 237 * The output of this function then goes to the output fd requested 238 * by the caller of zfs_send(). 239 */ 240static void * 241cksummer(void *arg) 242{ 243 dedup_arg_t *dda = arg; 244 char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE); 245 dmu_replay_record_t thedrr; 246 dmu_replay_record_t *drr = &thedrr; 247 FILE *ofp; 248 int outfd; 249 dedup_table_t ddt; 250 zio_cksum_t stream_cksum; 251 uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE); 252 uint64_t numbuckets; 253 254 ddt.max_ddt_size = 255 MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100, 256 SMALLEST_POSSIBLE_MAX_DDT_MB << 20); 257 258 numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t)); 259 260 /* 261 * numbuckets must be a power of 2. Increase number to 262 * a power of 2 if necessary. 263 */ 264 if (!ISP2(numbuckets)) 265 numbuckets = 1 << high_order_bit(numbuckets); 266 267 ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *)); 268 ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0, 269 NULL, NULL, NULL, NULL, NULL, 0); 270 ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *); 271 ddt.numhashbits = high_order_bit(numbuckets) - 1; 272 ddt.ddt_full = B_FALSE; 273 274 outfd = dda->outputfd; 275 ofp = fdopen(dda->inputfd, "r"); 276 while (ssread(drr, sizeof (*drr), ofp) != 0) { 277 278 /* 279 * kernel filled in checksum, we are going to write same 280 * record, but need to regenerate checksum. 281 */ 282 if (drr->drr_type != DRR_BEGIN) { 283 bzero(&drr->drr_u.drr_checksum.drr_checksum, 284 sizeof (drr->drr_u.drr_checksum.drr_checksum)); 285 } 286 287 switch (drr->drr_type) { 288 case DRR_BEGIN: 289 { 290 struct drr_begin *drrb = &drr->drr_u.drr_begin; 291 int fflags; 292 int sz = 0; 293 ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0); 294 295 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 296 297 /* set the DEDUP feature flag for this stream */ 298 fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 299 fflags |= (DMU_BACKUP_FEATURE_DEDUP | 300 DMU_BACKUP_FEATURE_DEDUPPROPS); 301 DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags); 302 303 if (drr->drr_payloadlen != 0) { 304 sz = drr->drr_payloadlen; 305 306 if (sz > SPA_MAXBLOCKSIZE) { 307 buf = zfs_realloc(dda->dedup_hdl, buf, 308 SPA_MAXBLOCKSIZE, sz); 309 } 310 (void) ssread(buf, sz, ofp); 311 if (ferror(stdin)) 312 perror("fread"); 313 } 314 if (dump_record(drr, buf, sz, &stream_cksum, 315 outfd) != 0) 316 goto out; 317 break; 318 } 319 320 case DRR_END: 321 { 322 struct drr_end *drre = &drr->drr_u.drr_end; 323 /* use the recalculated checksum */ 324 drre->drr_checksum = stream_cksum; 325 if (dump_record(drr, NULL, 0, &stream_cksum, 326 outfd) != 0) 327 goto out; 328 break; 329 } 330 331 case DRR_OBJECT: 332 { 333 struct drr_object *drro = &drr->drr_u.drr_object; 334 if (drro->drr_bonuslen > 0) { 335 (void) ssread(buf, 336 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), 337 ofp); 338 } 339 if (dump_record(drr, buf, 340 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), 341 &stream_cksum, outfd) != 0) 342 goto out; 343 break; 344 } 345 346 case DRR_SPILL: 347 { 348 struct drr_spill *drrs = &drr->drr_u.drr_spill; 349 (void) ssread(buf, drrs->drr_length, ofp); 350 if (dump_record(drr, buf, drrs->drr_length, 351 &stream_cksum, outfd) != 0) 352 goto out; 353 break; 354 } 355 356 case DRR_FREEOBJECTS: 357 { 358 if (dump_record(drr, NULL, 0, &stream_cksum, 359 outfd) != 0) 360 goto out; 361 break; 362 } 363 364 case DRR_WRITE: 365 { 366 struct drr_write *drrw = &drr->drr_u.drr_write; 367 dataref_t dataref; 368 uint64_t payload_size; 369 370 payload_size = DRR_WRITE_PAYLOAD_SIZE(drrw); 371 (void) ssread(buf, payload_size, ofp); 372 373 /* 374 * Use the existing checksum if it's dedup-capable, 375 * else calculate a SHA256 checksum for it. 376 */ 377 378 if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum, 379 zero_cksum) || 380 !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) { 381 SHA256_CTX ctx; 382 zio_cksum_t tmpsha256; 383 384 SHA256Init(&ctx); 385 SHA256Update(&ctx, buf, payload_size); 386 SHA256Final(&tmpsha256, &ctx); 387 drrw->drr_key.ddk_cksum.zc_word[0] = 388 BE_64(tmpsha256.zc_word[0]); 389 drrw->drr_key.ddk_cksum.zc_word[1] = 390 BE_64(tmpsha256.zc_word[1]); 391 drrw->drr_key.ddk_cksum.zc_word[2] = 392 BE_64(tmpsha256.zc_word[2]); 393 drrw->drr_key.ddk_cksum.zc_word[3] = 394 BE_64(tmpsha256.zc_word[3]); 395 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256; 396 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP; 397 } 398 399 dataref.ref_guid = drrw->drr_toguid; 400 dataref.ref_object = drrw->drr_object; 401 dataref.ref_offset = drrw->drr_offset; 402 403 if (ddt_update(dda->dedup_hdl, &ddt, 404 &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop, 405 &dataref)) { 406 dmu_replay_record_t wbr_drr = {0}; 407 struct drr_write_byref *wbr_drrr = 408 &wbr_drr.drr_u.drr_write_byref; 409 410 /* block already present in stream */ 411 wbr_drr.drr_type = DRR_WRITE_BYREF; 412 413 wbr_drrr->drr_object = drrw->drr_object; 414 wbr_drrr->drr_offset = drrw->drr_offset; 415 wbr_drrr->drr_length = drrw->drr_logical_size; 416 wbr_drrr->drr_toguid = drrw->drr_toguid; 417 wbr_drrr->drr_refguid = dataref.ref_guid; 418 wbr_drrr->drr_refobject = 419 dataref.ref_object; 420 wbr_drrr->drr_refoffset = 421 dataref.ref_offset; 422 423 wbr_drrr->drr_checksumtype = 424 drrw->drr_checksumtype; 425 wbr_drrr->drr_checksumflags = 426 drrw->drr_checksumtype; 427 wbr_drrr->drr_key.ddk_cksum = 428 drrw->drr_key.ddk_cksum; 429 wbr_drrr->drr_key.ddk_prop = 430 drrw->drr_key.ddk_prop; 431 432 if (dump_record(&wbr_drr, NULL, 0, 433 &stream_cksum, outfd) != 0) 434 goto out; 435 } else { 436 /* block not previously seen */ 437 if (dump_record(drr, buf, payload_size, 438 &stream_cksum, outfd) != 0) 439 goto out; 440 } 441 break; 442 } 443 444 case DRR_WRITE_EMBEDDED: 445 { 446 struct drr_write_embedded *drrwe = 447 &drr->drr_u.drr_write_embedded; 448 (void) ssread(buf, 449 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp); 450 if (dump_record(drr, buf, 451 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), 452 &stream_cksum, outfd) != 0) 453 goto out; 454 break; 455 } 456 457 case DRR_FREE: 458 { 459 if (dump_record(drr, NULL, 0, &stream_cksum, 460 outfd) != 0) 461 goto out; 462 break; 463 } 464 465 default: 466 (void) fprintf(stderr, "INVALID record type 0x%x\n", 467 drr->drr_type); 468 /* should never happen, so assert */ 469 assert(B_FALSE); 470 } 471 } 472out: 473 umem_cache_destroy(ddt.ddecache); 474 free(ddt.dedup_hash_array); 475 free(buf); 476 (void) fclose(ofp); 477 478 return (NULL); 479} 480 481/* 482 * Routines for dealing with the AVL tree of fs-nvlists 483 */ 484typedef struct fsavl_node { 485 avl_node_t fn_node; 486 nvlist_t *fn_nvfs; 487 char *fn_snapname; 488 uint64_t fn_guid; 489} fsavl_node_t; 490 491static int 492fsavl_compare(const void *arg1, const void *arg2) 493{ 494 const fsavl_node_t *fn1 = (const fsavl_node_t *)arg1; 495 const fsavl_node_t *fn2 = (const fsavl_node_t *)arg2; 496 497 return (AVL_CMP(fn1->fn_guid, fn2->fn_guid)); 498} 499 500/* 501 * Given the GUID of a snapshot, find its containing filesystem and 502 * (optionally) name. 503 */ 504static nvlist_t * 505fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname) 506{ 507 fsavl_node_t fn_find; 508 fsavl_node_t *fn; 509 510 fn_find.fn_guid = snapguid; 511 512 fn = avl_find(avl, &fn_find, NULL); 513 if (fn) { 514 if (snapname) 515 *snapname = fn->fn_snapname; 516 return (fn->fn_nvfs); 517 } 518 return (NULL); 519} 520 521static void 522fsavl_destroy(avl_tree_t *avl) 523{ 524 fsavl_node_t *fn; 525 void *cookie; 526 527 if (avl == NULL) 528 return; 529 530 cookie = NULL; 531 while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL) 532 free(fn); 533 avl_destroy(avl); 534 free(avl); 535} 536 537/* 538 * Given an nvlist, produce an avl tree of snapshots, ordered by guid 539 */ 540static avl_tree_t * 541fsavl_create(nvlist_t *fss) 542{ 543 avl_tree_t *fsavl; 544 nvpair_t *fselem = NULL; 545 546 if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL) 547 return (NULL); 548 549 avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t), 550 offsetof(fsavl_node_t, fn_node)); 551 552 while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) { 553 nvlist_t *nvfs, *snaps; 554 nvpair_t *snapelem = NULL; 555 556 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); 557 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); 558 559 while ((snapelem = 560 nvlist_next_nvpair(snaps, snapelem)) != NULL) { 561 fsavl_node_t *fn; 562 uint64_t guid; 563 564 VERIFY(0 == nvpair_value_uint64(snapelem, &guid)); 565 if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) { 566 fsavl_destroy(fsavl); 567 return (NULL); 568 } 569 fn->fn_nvfs = nvfs; 570 fn->fn_snapname = nvpair_name(snapelem); 571 fn->fn_guid = guid; 572 573 /* 574 * Note: if there are multiple snaps with the 575 * same GUID, we ignore all but one. 576 */ 577 if (avl_find(fsavl, fn, NULL) == NULL) 578 avl_add(fsavl, fn); 579 else 580 free(fn); 581 } 582 } 583 584 return (fsavl); 585} 586 587/* 588 * Routines for dealing with the giant nvlist of fs-nvlists, etc. 589 */ 590typedef struct send_data { 591 /* 592 * assigned inside every recursive call, 593 * restored from *_save on return: 594 * 595 * guid of fromsnap snapshot in parent dataset 596 * txg of fromsnap snapshot in current dataset 597 * txg of tosnap snapshot in current dataset 598 */ 599 600 uint64_t parent_fromsnap_guid; 601 uint64_t fromsnap_txg; 602 uint64_t tosnap_txg; 603 604 /* the nvlists get accumulated during depth-first traversal */ 605 nvlist_t *parent_snaps; 606 nvlist_t *fss; 607 nvlist_t *snapprops; 608 609 /* send-receive configuration, does not change during traversal */ 610 const char *fsname; 611 const char *fromsnap; 612 const char *tosnap; 613 boolean_t recursive; 614 boolean_t verbose; 615 616 /* 617 * The header nvlist is of the following format: 618 * { 619 * "tosnap" -> string 620 * "fromsnap" -> string (if incremental) 621 * "fss" -> { 622 * id -> { 623 * 624 * "name" -> string (full name; for debugging) 625 * "parentfromsnap" -> number (guid of fromsnap in parent) 626 * 627 * "props" -> { name -> value (only if set here) } 628 * "snaps" -> { name (lastname) -> number (guid) } 629 * "snapprops" -> { name (lastname) -> { name -> value } } 630 * 631 * "origin" -> number (guid) (if clone) 632 * "sent" -> boolean (not on-disk) 633 * } 634 * } 635 * } 636 * 637 */ 638} send_data_t; 639 640static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv); 641 642static int 643send_iterate_snap(zfs_handle_t *zhp, void *arg) 644{ 645 send_data_t *sd = arg; 646 uint64_t guid = zhp->zfs_dmustats.dds_guid; 647 uint64_t txg = zhp->zfs_dmustats.dds_creation_txg; 648 char *snapname; 649 nvlist_t *nv; 650 651 snapname = strrchr(zhp->zfs_name, '@')+1; 652 653 if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) { 654 if (sd->verbose) { 655 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 656 "skipping snapshot %s because it was created " 657 "after the destination snapshot (%s)\n"), 658 zhp->zfs_name, sd->tosnap); 659 } 660 zfs_close(zhp); 661 return (0); 662 } 663 664 VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid)); 665 /* 666 * NB: if there is no fromsnap here (it's a newly created fs in 667 * an incremental replication), we will substitute the tosnap. 668 */ 669 if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) || 670 (sd->parent_fromsnap_guid == 0 && sd->tosnap && 671 strcmp(snapname, sd->tosnap) == 0)) { 672 sd->parent_fromsnap_guid = guid; 673 } 674 675 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); 676 send_iterate_prop(zhp, nv); 677 VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv)); 678 nvlist_free(nv); 679 680 zfs_close(zhp); 681 return (0); 682} 683 684static void 685send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv) 686{ 687 nvpair_t *elem = NULL; 688 689 while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) { 690 char *propname = nvpair_name(elem); 691 zfs_prop_t prop = zfs_name_to_prop(propname); 692 nvlist_t *propnv; 693 694 if (!zfs_prop_user(propname)) { 695 /* 696 * Realistically, this should never happen. However, 697 * we want the ability to add DSL properties without 698 * needing to make incompatible version changes. We 699 * need to ignore unknown properties to allow older 700 * software to still send datasets containing these 701 * properties, with the unknown properties elided. 702 */ 703 if (prop == ZPROP_INVAL) 704 continue; 705 706 if (zfs_prop_readonly(prop)) 707 continue; 708 } 709 710 verify(nvpair_value_nvlist(elem, &propnv) == 0); 711 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION || 712 prop == ZFS_PROP_REFQUOTA || 713 prop == ZFS_PROP_REFRESERVATION) { 714 char *source; 715 uint64_t value; 716 verify(nvlist_lookup_uint64(propnv, 717 ZPROP_VALUE, &value) == 0); 718 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) 719 continue; 720 /* 721 * May have no source before SPA_VERSION_RECVD_PROPS, 722 * but is still modifiable. 723 */ 724 if (nvlist_lookup_string(propnv, 725 ZPROP_SOURCE, &source) == 0) { 726 if ((strcmp(source, zhp->zfs_name) != 0) && 727 (strcmp(source, 728 ZPROP_SOURCE_VAL_RECVD) != 0)) 729 continue; 730 } 731 } else { 732 char *source; 733 if (nvlist_lookup_string(propnv, 734 ZPROP_SOURCE, &source) != 0) 735 continue; 736 if ((strcmp(source, zhp->zfs_name) != 0) && 737 (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0)) 738 continue; 739 } 740 741 if (zfs_prop_user(propname) || 742 zfs_prop_get_type(prop) == PROP_TYPE_STRING) { 743 char *value; 744 verify(nvlist_lookup_string(propnv, 745 ZPROP_VALUE, &value) == 0); 746 VERIFY(0 == nvlist_add_string(nv, propname, value)); 747 } else { 748 uint64_t value; 749 verify(nvlist_lookup_uint64(propnv, 750 ZPROP_VALUE, &value) == 0); 751 VERIFY(0 == nvlist_add_uint64(nv, propname, value)); 752 } 753 } 754} 755 756/* 757 * returns snapshot creation txg 758 * and returns 0 if the snapshot does not exist 759 */ 760static uint64_t 761get_snap_txg(libzfs_handle_t *hdl, const char *fs, const char *snap) 762{ 763 char name[ZFS_MAX_DATASET_NAME_LEN]; 764 uint64_t txg = 0; 765 766 if (fs == NULL || fs[0] == '\0' || snap == NULL || snap[0] == '\0') 767 return (txg); 768 769 (void) snprintf(name, sizeof (name), "%s@%s", fs, snap); 770 if (zfs_dataset_exists(hdl, name, ZFS_TYPE_SNAPSHOT)) { 771 zfs_handle_t *zhp = zfs_open(hdl, name, ZFS_TYPE_SNAPSHOT); 772 if (zhp != NULL) { 773 txg = zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG); 774 zfs_close(zhp); 775 } 776 } 777 778 return (txg); 779} 780 781/* 782 * recursively generate nvlists describing datasets. See comment 783 * for the data structure send_data_t above for description of contents 784 * of the nvlist. 785 */ 786static int 787send_iterate_fs(zfs_handle_t *zhp, void *arg) 788{ 789 send_data_t *sd = arg; 790 nvlist_t *nvfs, *nv; 791 int rv = 0; 792 uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid; 793 uint64_t fromsnap_txg_save = sd->fromsnap_txg; 794 uint64_t tosnap_txg_save = sd->tosnap_txg; 795 uint64_t txg = zhp->zfs_dmustats.dds_creation_txg; 796 uint64_t guid = zhp->zfs_dmustats.dds_guid; 797 uint64_t fromsnap_txg, tosnap_txg; 798 char guidstring[64]; 799 800 fromsnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->fromsnap); 801 if (fromsnap_txg != 0) 802 sd->fromsnap_txg = fromsnap_txg; 803 804 tosnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->tosnap); 805 if (tosnap_txg != 0) 806 sd->tosnap_txg = tosnap_txg; 807 808 /* 809 * on the send side, if the current dataset does not have tosnap, 810 * perform two additional checks: 811 * 812 * - skip sending the current dataset if it was created later than 813 * the parent tosnap 814 * - return error if the current dataset was created earlier than 815 * the parent tosnap 816 */ 817 if (sd->tosnap != NULL && tosnap_txg == 0) { 818 if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) { 819 if (sd->verbose) { 820 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 821 "skipping dataset %s: snapshot %s does " 822 "not exist\n"), zhp->zfs_name, sd->tosnap); 823 } 824 } else { 825 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 826 "cannot send %s@%s%s: snapshot %s@%s does not " 827 "exist\n"), sd->fsname, sd->tosnap, sd->recursive ? 828 dgettext(TEXT_DOMAIN, " recursively") : "", 829 zhp->zfs_name, sd->tosnap); 830 rv = -1; 831 } 832 goto out; 833 } 834 835 VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0)); 836 VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name)); 837 VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap", 838 sd->parent_fromsnap_guid)); 839 840 if (zhp->zfs_dmustats.dds_origin[0]) { 841 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl, 842 zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT); 843 if (origin == NULL) { 844 rv = -1; 845 goto out; 846 } 847 VERIFY(0 == nvlist_add_uint64(nvfs, "origin", 848 origin->zfs_dmustats.dds_guid)); 849 } 850 851 /* iterate over props */ 852 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); 853 send_iterate_prop(zhp, nv); 854 VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv)); 855 nvlist_free(nv); 856 857 /* iterate over snaps, and set sd->parent_fromsnap_guid */ 858 sd->parent_fromsnap_guid = 0; 859 VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0)); 860 VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0)); 861 (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd); 862 VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps)); 863 VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops)); 864 nvlist_free(sd->parent_snaps); 865 nvlist_free(sd->snapprops); 866 867 /* add this fs to nvlist */ 868 (void) snprintf(guidstring, sizeof (guidstring), 869 "0x%llx", (longlong_t)guid); 870 VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs)); 871 nvlist_free(nvfs); 872 873 /* iterate over children */ 874 if (sd->recursive) 875 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd); 876 877out: 878 sd->parent_fromsnap_guid = parent_fromsnap_guid_save; 879 sd->fromsnap_txg = fromsnap_txg_save; 880 sd->tosnap_txg = tosnap_txg_save; 881 882 zfs_close(zhp); 883 return (rv); 884} 885 886static int 887gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap, 888 const char *tosnap, boolean_t recursive, boolean_t verbose, 889 nvlist_t **nvlp, avl_tree_t **avlp) 890{ 891 zfs_handle_t *zhp; 892 send_data_t sd = { 0 }; 893 int error; 894 895 zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 896 if (zhp == NULL) 897 return (EZFS_BADTYPE); 898 899 VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0)); 900 sd.fsname = fsname; 901 sd.fromsnap = fromsnap; 902 sd.tosnap = tosnap; 903 sd.recursive = recursive; 904 sd.verbose = verbose; 905 906 if ((error = send_iterate_fs(zhp, &sd)) != 0) { 907 nvlist_free(sd.fss); 908 if (avlp != NULL) 909 *avlp = NULL; 910 *nvlp = NULL; 911 return (error); 912 } 913 914 if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) { 915 nvlist_free(sd.fss); 916 *nvlp = NULL; 917 return (EZFS_NOMEM); 918 } 919 920 *nvlp = sd.fss; 921 return (0); 922} 923 924/* 925 * Routines specific to "zfs send" 926 */ 927typedef struct send_dump_data { 928 /* these are all just the short snapname (the part after the @) */ 929 const char *fromsnap; 930 const char *tosnap; 931 char prevsnap[ZFS_MAX_DATASET_NAME_LEN]; 932 uint64_t prevsnap_obj; 933 boolean_t seenfrom, seento, replicate, doall, fromorigin; 934 boolean_t verbose, dryrun, parsable, progress, embed_data, std_out; 935 boolean_t progressastitle; 936 boolean_t large_block, compress; 937 int outfd; 938 boolean_t err; 939 nvlist_t *fss; 940 nvlist_t *snapholds; 941 avl_tree_t *fsavl; 942 snapfilter_cb_t *filter_cb; 943 void *filter_cb_arg; 944 nvlist_t *debugnv; 945 char holdtag[ZFS_MAX_DATASET_NAME_LEN]; 946 int cleanup_fd; 947 uint64_t size; 948} send_dump_data_t; 949 950static int 951estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj, 952 boolean_t fromorigin, enum lzc_send_flags flags, uint64_t *sizep) 953{ 954 zfs_cmd_t zc = { 0 }; 955 libzfs_handle_t *hdl = zhp->zfs_hdl; 956 957 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 958 assert(fromsnap_obj == 0 || !fromorigin); 959 960 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 961 zc.zc_obj = fromorigin; 962 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 963 zc.zc_fromobj = fromsnap_obj; 964 zc.zc_guid = 1; /* estimate flag */ 965 zc.zc_flags = flags; 966 967 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { 968 char errbuf[1024]; 969 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 970 "warning: cannot estimate space for '%s'"), zhp->zfs_name); 971 972 switch (errno) { 973 case EXDEV: 974 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 975 "not an earlier snapshot from the same fs")); 976 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 977 978 case ENOENT: 979 if (zfs_dataset_exists(hdl, zc.zc_name, 980 ZFS_TYPE_SNAPSHOT)) { 981 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 982 "incremental source (@%s) does not exist"), 983 zc.zc_value); 984 } 985 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 986 987 case EDQUOT: 988 case EFBIG: 989 case EIO: 990 case ENOLINK: 991 case ENOSPC: 992 case ENXIO: 993 case EPIPE: 994 case ERANGE: 995 case EFAULT: 996 case EROFS: 997 zfs_error_aux(hdl, strerror(errno)); 998 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 999 1000 default: 1001 return (zfs_standard_error(hdl, errno, errbuf)); 1002 } 1003 } 1004 1005 *sizep = zc.zc_objset_type; 1006 1007 return (0); 1008} 1009 1010/* 1011 * Dumps a backup of the given snapshot (incremental from fromsnap if it's not 1012 * NULL) to the file descriptor specified by outfd. 1013 */ 1014static int 1015dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj, 1016 boolean_t fromorigin, int outfd, enum lzc_send_flags flags, 1017 nvlist_t *debugnv) 1018{ 1019 zfs_cmd_t zc = { 0 }; 1020 libzfs_handle_t *hdl = zhp->zfs_hdl; 1021 nvlist_t *thisdbg; 1022 1023 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 1024 assert(fromsnap_obj == 0 || !fromorigin); 1025 1026 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 1027 zc.zc_cookie = outfd; 1028 zc.zc_obj = fromorigin; 1029 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 1030 zc.zc_fromobj = fromsnap_obj; 1031 zc.zc_flags = flags; 1032 1033 VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0)); 1034 if (fromsnap && fromsnap[0] != '\0') { 1035 VERIFY(0 == nvlist_add_string(thisdbg, 1036 "fromsnap", fromsnap)); 1037 } 1038 1039 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { 1040 char errbuf[1024]; 1041 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1042 "warning: cannot send '%s'"), zhp->zfs_name); 1043 1044 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno)); 1045 if (debugnv) { 1046 VERIFY(0 == nvlist_add_nvlist(debugnv, 1047 zhp->zfs_name, thisdbg)); 1048 } 1049 nvlist_free(thisdbg); 1050 1051 switch (errno) { 1052 case EXDEV: 1053 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1054 "not an earlier snapshot from the same fs")); 1055 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 1056 1057 case ENOENT: 1058 if (zfs_dataset_exists(hdl, zc.zc_name, 1059 ZFS_TYPE_SNAPSHOT)) { 1060 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1061 "incremental source (@%s) does not exist"), 1062 zc.zc_value); 1063 } 1064 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 1065 1066 case EDQUOT: 1067 case EFBIG: 1068 case EIO: 1069 case ENOLINK: 1070 case ENOSPC: 1071#ifdef illumos 1072 case ENOSTR: 1073#endif 1074 case ENXIO: 1075 case EPIPE: 1076 case ERANGE: 1077 case EFAULT: 1078 case EROFS: 1079 zfs_error_aux(hdl, strerror(errno)); 1080 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 1081 1082 default: 1083 return (zfs_standard_error(hdl, errno, errbuf)); 1084 } 1085 } 1086 1087 if (debugnv) 1088 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg)); 1089 nvlist_free(thisdbg); 1090 1091 return (0); 1092} 1093 1094static void 1095gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd) 1096{ 1097 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 1098 1099 /* 1100 * zfs_send() only sets snapholds for sends that need them, 1101 * e.g. replication and doall. 1102 */ 1103 if (sdd->snapholds == NULL) 1104 return; 1105 1106 fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag); 1107} 1108 1109static void * 1110send_progress_thread(void *arg) 1111{ 1112 progress_arg_t *pa = arg; 1113 zfs_cmd_t zc = { 0 }; 1114 zfs_handle_t *zhp = pa->pa_zhp; 1115 libzfs_handle_t *hdl = zhp->zfs_hdl; 1116 unsigned long long bytes, total; 1117 char buf[16]; 1118 time_t t; 1119 struct tm *tm; 1120 1121 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 1122 1123 if (!pa->pa_parsable && !pa->pa_astitle) 1124 (void) fprintf(stderr, "TIME SENT SNAPSHOT\n"); 1125 1126 /* 1127 * Print the progress from ZFS_IOC_SEND_PROGRESS every second. 1128 */ 1129 for (;;) { 1130 (void) sleep(1); 1131 1132 zc.zc_cookie = pa->pa_fd; 1133 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0) 1134 return ((void *)-1); 1135 1136 (void) time(&t); 1137 tm = localtime(&t); 1138 bytes = zc.zc_cookie; 1139 1140 if (pa->pa_astitle) { 1141 int pct; 1142 if (pa->pa_size > bytes) 1143 pct = 100 * bytes / pa->pa_size; 1144 else 1145 pct = 100; 1146 1147 setproctitle("sending %s (%d%%: %llu/%llu)", 1148 zhp->zfs_name, pct, bytes, pa->pa_size); 1149 } else if (pa->pa_parsable) { 1150 (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n", 1151 tm->tm_hour, tm->tm_min, tm->tm_sec, 1152 bytes, zhp->zfs_name); 1153 } else { 1154 zfs_nicenum(bytes, buf, sizeof (buf)); 1155 (void) fprintf(stderr, "%02d:%02d:%02d %5s %s\n", 1156 tm->tm_hour, tm->tm_min, tm->tm_sec, 1157 buf, zhp->zfs_name); 1158 } 1159 } 1160} 1161 1162static void 1163send_print_verbose(FILE *fout, const char *tosnap, const char *fromsnap, 1164 uint64_t size, boolean_t parsable) 1165{ 1166 if (parsable) { 1167 if (fromsnap != NULL) { 1168 (void) fprintf(fout, "incremental\t%s\t%s", 1169 fromsnap, tosnap); 1170 } else { 1171 (void) fprintf(fout, "full\t%s", 1172 tosnap); 1173 } 1174 } else { 1175 if (fromsnap != NULL) { 1176 if (strchr(fromsnap, '@') == NULL && 1177 strchr(fromsnap, '#') == NULL) { 1178 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1179 "send from @%s to %s"), 1180 fromsnap, tosnap); 1181 } else { 1182 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1183 "send from %s to %s"), 1184 fromsnap, tosnap); 1185 } 1186 } else { 1187 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1188 "full send of %s"), 1189 tosnap); 1190 } 1191 } 1192 1193 if (parsable) { 1194 (void) fprintf(fout, "\t%llu", 1195 (longlong_t)size); 1196 } else if (size != 0) { 1197 char buf[16]; 1198 zfs_nicenum(size, buf, sizeof (buf)); 1199 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1200 " estimated size is %s"), buf); 1201 } 1202 (void) fprintf(fout, "\n"); 1203} 1204 1205static int 1206dump_snapshot(zfs_handle_t *zhp, void *arg) 1207{ 1208 send_dump_data_t *sdd = arg; 1209 progress_arg_t pa = { 0 }; 1210 pthread_t tid; 1211 char *thissnap; 1212 enum lzc_send_flags flags = 0; 1213 int err; 1214 boolean_t isfromsnap, istosnap, fromorigin; 1215 boolean_t exclude = B_FALSE; 1216 FILE *fout = sdd->std_out ? stdout : stderr; 1217 uint64_t size = 0; 1218 1219 err = 0; 1220 thissnap = strchr(zhp->zfs_name, '@') + 1; 1221 isfromsnap = (sdd->fromsnap != NULL && 1222 strcmp(sdd->fromsnap, thissnap) == 0); 1223 1224 if (!sdd->seenfrom && isfromsnap) { 1225 gather_holds(zhp, sdd); 1226 sdd->seenfrom = B_TRUE; 1227 (void) strcpy(sdd->prevsnap, thissnap); 1228 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 1229 zfs_close(zhp); 1230 return (0); 1231 } 1232 1233 if (sdd->seento || !sdd->seenfrom) { 1234 zfs_close(zhp); 1235 return (0); 1236 } 1237 1238 istosnap = (strcmp(sdd->tosnap, thissnap) == 0); 1239 if (istosnap) 1240 sdd->seento = B_TRUE; 1241 1242 if (sdd->large_block) 1243 flags |= LZC_SEND_FLAG_LARGE_BLOCK; 1244 if (sdd->embed_data) 1245 flags |= LZC_SEND_FLAG_EMBED_DATA; 1246 if (sdd->compress) 1247 flags |= LZC_SEND_FLAG_COMPRESS; 1248 1249 if (!sdd->doall && !isfromsnap && !istosnap) { 1250 if (sdd->replicate) { 1251 char *snapname; 1252 nvlist_t *snapprops; 1253 /* 1254 * Filter out all intermediate snapshots except origin 1255 * snapshots needed to replicate clones. 1256 */ 1257 nvlist_t *nvfs = fsavl_find(sdd->fsavl, 1258 zhp->zfs_dmustats.dds_guid, &snapname); 1259 1260 VERIFY(0 == nvlist_lookup_nvlist(nvfs, 1261 "snapprops", &snapprops)); 1262 VERIFY(0 == nvlist_lookup_nvlist(snapprops, 1263 thissnap, &snapprops)); 1264 exclude = !nvlist_exists(snapprops, "is_clone_origin"); 1265 } else { 1266 exclude = B_TRUE; 1267 } 1268 } 1269 1270 /* 1271 * If a filter function exists, call it to determine whether 1272 * this snapshot will be sent. 1273 */ 1274 if (exclude || (sdd->filter_cb != NULL && 1275 sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) { 1276 /* 1277 * This snapshot is filtered out. Don't send it, and don't 1278 * set prevsnap_obj, so it will be as if this snapshot didn't 1279 * exist, and the next accepted snapshot will be sent as 1280 * an incremental from the last accepted one, or as the 1281 * first (and full) snapshot in the case of a replication, 1282 * non-incremental send. 1283 */ 1284 zfs_close(zhp); 1285 return (0); 1286 } 1287 1288 gather_holds(zhp, sdd); 1289 fromorigin = sdd->prevsnap[0] == '\0' && 1290 (sdd->fromorigin || sdd->replicate); 1291 1292 if (sdd->progress && sdd->dryrun) { 1293 (void) estimate_ioctl(zhp, sdd->prevsnap_obj, 1294 fromorigin, flags, &size); 1295 sdd->size += size; 1296 } 1297 1298 if (sdd->verbose) { 1299 send_print_verbose(fout, zhp->zfs_name, 1300 sdd->prevsnap[0] ? sdd->prevsnap : NULL, 1301 size, sdd->parsable); 1302 } 1303 1304 if (!sdd->dryrun) { 1305 /* 1306 * If progress reporting is requested, spawn a new thread to 1307 * poll ZFS_IOC_SEND_PROGRESS at a regular interval. 1308 */ 1309 if (sdd->progress) { 1310 pa.pa_zhp = zhp; 1311 pa.pa_fd = sdd->outfd; 1312 pa.pa_parsable = sdd->parsable; 1313 pa.pa_size = sdd->size; 1314 pa.pa_astitle = sdd->progressastitle; 1315 1316 if ((err = pthread_create(&tid, NULL, 1317 send_progress_thread, &pa)) != 0) { 1318 zfs_close(zhp); 1319 return (err); 1320 } 1321 } 1322 1323 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj, 1324 fromorigin, sdd->outfd, flags, sdd->debugnv); 1325 1326 if (sdd->progress) { 1327 (void) pthread_cancel(tid); 1328 (void) pthread_join(tid, NULL); 1329 } 1330 } 1331 1332 (void) strcpy(sdd->prevsnap, thissnap); 1333 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 1334 zfs_close(zhp); 1335 return (err); 1336} 1337 1338static int 1339dump_filesystem(zfs_handle_t *zhp, void *arg) 1340{ 1341 int rv = 0; 1342 send_dump_data_t *sdd = arg; 1343 boolean_t missingfrom = B_FALSE; 1344 zfs_cmd_t zc = { 0 }; 1345 1346 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", 1347 zhp->zfs_name, sdd->tosnap); 1348 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) { 1349 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1350 "WARNING: could not send %s@%s: does not exist\n"), 1351 zhp->zfs_name, sdd->tosnap); 1352 sdd->err = B_TRUE; 1353 return (0); 1354 } 1355 1356 if (sdd->replicate && sdd->fromsnap) { 1357 /* 1358 * If this fs does not have fromsnap, and we're doing 1359 * recursive, we need to send a full stream from the 1360 * beginning (or an incremental from the origin if this 1361 * is a clone). If we're doing non-recursive, then let 1362 * them get the error. 1363 */ 1364 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", 1365 zhp->zfs_name, sdd->fromsnap); 1366 if (ioctl(zhp->zfs_hdl->libzfs_fd, 1367 ZFS_IOC_OBJSET_STATS, &zc) != 0) { 1368 missingfrom = B_TRUE; 1369 } 1370 } 1371 1372 sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0; 1373 sdd->prevsnap_obj = 0; 1374 if (sdd->fromsnap == NULL || missingfrom) 1375 sdd->seenfrom = B_TRUE; 1376 1377 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg); 1378 if (!sdd->seenfrom) { 1379 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1380 "WARNING: could not send %s@%s:\n" 1381 "incremental source (%s@%s) does not exist\n"), 1382 zhp->zfs_name, sdd->tosnap, 1383 zhp->zfs_name, sdd->fromsnap); 1384 sdd->err = B_TRUE; 1385 } else if (!sdd->seento) { 1386 if (sdd->fromsnap) { 1387 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1388 "WARNING: could not send %s@%s:\n" 1389 "incremental source (%s@%s) " 1390 "is not earlier than it\n"), 1391 zhp->zfs_name, sdd->tosnap, 1392 zhp->zfs_name, sdd->fromsnap); 1393 } else { 1394 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1395 "WARNING: " 1396 "could not send %s@%s: does not exist\n"), 1397 zhp->zfs_name, sdd->tosnap); 1398 } 1399 sdd->err = B_TRUE; 1400 } 1401 1402 return (rv); 1403} 1404 1405static int 1406dump_filesystems(zfs_handle_t *rzhp, void *arg) 1407{ 1408 send_dump_data_t *sdd = arg; 1409 nvpair_t *fspair; 1410 boolean_t needagain, progress; 1411 1412 if (!sdd->replicate) 1413 return (dump_filesystem(rzhp, sdd)); 1414 1415 /* Mark the clone origin snapshots. */ 1416 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1417 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1418 nvlist_t *nvfs; 1419 uint64_t origin_guid = 0; 1420 1421 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs)); 1422 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid); 1423 if (origin_guid != 0) { 1424 char *snapname; 1425 nvlist_t *origin_nv = fsavl_find(sdd->fsavl, 1426 origin_guid, &snapname); 1427 if (origin_nv != NULL) { 1428 nvlist_t *snapprops; 1429 VERIFY(0 == nvlist_lookup_nvlist(origin_nv, 1430 "snapprops", &snapprops)); 1431 VERIFY(0 == nvlist_lookup_nvlist(snapprops, 1432 snapname, &snapprops)); 1433 VERIFY(0 == nvlist_add_boolean( 1434 snapprops, "is_clone_origin")); 1435 } 1436 } 1437 } 1438again: 1439 needagain = progress = B_FALSE; 1440 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1441 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1442 nvlist_t *fslist, *parent_nv; 1443 char *fsname; 1444 zfs_handle_t *zhp; 1445 int err; 1446 uint64_t origin_guid = 0; 1447 uint64_t parent_guid = 0; 1448 1449 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); 1450 if (nvlist_lookup_boolean(fslist, "sent") == 0) 1451 continue; 1452 1453 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0); 1454 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid); 1455 (void) nvlist_lookup_uint64(fslist, "parentfromsnap", 1456 &parent_guid); 1457 1458 if (parent_guid != 0) { 1459 parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL); 1460 if (!nvlist_exists(parent_nv, "sent")) { 1461 /* parent has not been sent; skip this one */ 1462 needagain = B_TRUE; 1463 continue; 1464 } 1465 } 1466 1467 if (origin_guid != 0) { 1468 nvlist_t *origin_nv = fsavl_find(sdd->fsavl, 1469 origin_guid, NULL); 1470 if (origin_nv != NULL && 1471 !nvlist_exists(origin_nv, "sent")) { 1472 /* 1473 * origin has not been sent yet; 1474 * skip this clone. 1475 */ 1476 needagain = B_TRUE; 1477 continue; 1478 } 1479 } 1480 1481 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET); 1482 if (zhp == NULL) 1483 return (-1); 1484 err = dump_filesystem(zhp, sdd); 1485 VERIFY(nvlist_add_boolean(fslist, "sent") == 0); 1486 progress = B_TRUE; 1487 zfs_close(zhp); 1488 if (err) 1489 return (err); 1490 } 1491 if (needagain) { 1492 assert(progress); 1493 goto again; 1494 } 1495 1496 /* clean out the sent flags in case we reuse this fss */ 1497 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1498 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1499 nvlist_t *fslist; 1500 1501 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); 1502 (void) nvlist_remove_all(fslist, "sent"); 1503 } 1504 1505 return (0); 1506} 1507 1508nvlist_t * 1509zfs_send_resume_token_to_nvlist(libzfs_handle_t *hdl, const char *token) 1510{ 1511 unsigned int version; 1512 int nread; 1513 unsigned long long checksum, packed_len; 1514 1515 /* 1516 * Decode token header, which is: 1517 * <token version>-<checksum of payload>-<uncompressed payload length> 1518 * Note that the only supported token version is 1. 1519 */ 1520 nread = sscanf(token, "%u-%llx-%llx-", 1521 &version, &checksum, &packed_len); 1522 if (nread != 3) { 1523 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1524 "resume token is corrupt (invalid format)")); 1525 return (NULL); 1526 } 1527 1528 if (version != ZFS_SEND_RESUME_TOKEN_VERSION) { 1529 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1530 "resume token is corrupt (invalid version %u)"), 1531 version); 1532 return (NULL); 1533 } 1534 1535 /* convert hexadecimal representation to binary */ 1536 token = strrchr(token, '-') + 1; 1537 int len = strlen(token) / 2; 1538 unsigned char *compressed = zfs_alloc(hdl, len); 1539 for (int i = 0; i < len; i++) { 1540 nread = sscanf(token + i * 2, "%2hhx", compressed + i); 1541 if (nread != 1) { 1542 free(compressed); 1543 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1544 "resume token is corrupt " 1545 "(payload is not hex-encoded)")); 1546 return (NULL); 1547 } 1548 } 1549 1550 /* verify checksum */ 1551 zio_cksum_t cksum; 1552 fletcher_4_native(compressed, len, NULL, &cksum); 1553 if (cksum.zc_word[0] != checksum) { 1554 free(compressed); 1555 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1556 "resume token is corrupt (incorrect checksum)")); 1557 return (NULL); 1558 } 1559 1560 /* uncompress */ 1561 void *packed = zfs_alloc(hdl, packed_len); 1562 uLongf packed_len_long = packed_len; 1563 if (uncompress(packed, &packed_len_long, compressed, len) != Z_OK || 1564 packed_len_long != packed_len) { 1565 free(packed); 1566 free(compressed); 1567 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1568 "resume token is corrupt (decompression failed)")); 1569 return (NULL); 1570 } 1571 1572 /* unpack nvlist */ 1573 nvlist_t *nv; 1574 int error = nvlist_unpack(packed, packed_len, &nv, KM_SLEEP); 1575 free(packed); 1576 free(compressed); 1577 if (error != 0) { 1578 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1579 "resume token is corrupt (nvlist_unpack failed)")); 1580 return (NULL); 1581 } 1582 return (nv); 1583} 1584 1585int 1586zfs_send_resume(libzfs_handle_t *hdl, sendflags_t *flags, int outfd, 1587 const char *resume_token) 1588{ 1589 char errbuf[1024]; 1590 char *toname; 1591 char *fromname = NULL; 1592 uint64_t resumeobj, resumeoff, toguid, fromguid, bytes; 1593 zfs_handle_t *zhp; 1594 int error = 0; 1595 char name[ZFS_MAX_DATASET_NAME_LEN]; 1596 enum lzc_send_flags lzc_flags = 0; 1597 uint64_t size = 0; 1598 FILE *fout = (flags->verbose && flags->dryrun) ? stdout : stderr; 1599 1600 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1601 "cannot resume send")); 1602 1603 nvlist_t *resume_nvl = 1604 zfs_send_resume_token_to_nvlist(hdl, resume_token); 1605 if (resume_nvl == NULL) { 1606 /* 1607 * zfs_error_aux has already been set by 1608 * zfs_send_resume_token_to_nvlist 1609 */ 1610 return (zfs_error(hdl, EZFS_FAULT, errbuf)); 1611 } 1612 if (flags->verbose) { 1613 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1614 "resume token contents:\n")); 1615 nvlist_print(fout, resume_nvl); 1616 } 1617 1618 if (nvlist_lookup_string(resume_nvl, "toname", &toname) != 0 || 1619 nvlist_lookup_uint64(resume_nvl, "object", &resumeobj) != 0 || 1620 nvlist_lookup_uint64(resume_nvl, "offset", &resumeoff) != 0 || 1621 nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 || 1622 nvlist_lookup_uint64(resume_nvl, "toguid", &toguid) != 0) { 1623 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1624 "resume token is corrupt")); 1625 return (zfs_error(hdl, EZFS_FAULT, errbuf)); 1626 } 1627 fromguid = 0; 1628 (void) nvlist_lookup_uint64(resume_nvl, "fromguid", &fromguid); 1629 1630 if (flags->largeblock || nvlist_exists(resume_nvl, "largeblockok")) 1631 lzc_flags |= LZC_SEND_FLAG_LARGE_BLOCK; 1632 if (flags->embed_data || nvlist_exists(resume_nvl, "embedok")) 1633 lzc_flags |= LZC_SEND_FLAG_EMBED_DATA; 1634 if (flags->compress || nvlist_exists(resume_nvl, "compressok")) 1635 lzc_flags |= LZC_SEND_FLAG_COMPRESS; 1636 1637 if (guid_to_name(hdl, toname, toguid, B_FALSE, name) != 0) { 1638 if (zfs_dataset_exists(hdl, toname, ZFS_TYPE_DATASET)) { 1639 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1640 "'%s' is no longer the same snapshot used in " 1641 "the initial send"), toname); 1642 } else { 1643 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1644 "'%s' used in the initial send no longer exists"), 1645 toname); 1646 } 1647 return (zfs_error(hdl, EZFS_BADPATH, errbuf)); 1648 } 1649 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 1650 if (zhp == NULL) { 1651 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1652 "unable to access '%s'"), name); 1653 return (zfs_error(hdl, EZFS_BADPATH, errbuf)); 1654 } 1655 1656 if (fromguid != 0) { 1657 if (guid_to_name(hdl, toname, fromguid, B_TRUE, name) != 0) { 1658 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1659 "incremental source %#llx no longer exists"), 1660 (longlong_t)fromguid); 1661 return (zfs_error(hdl, EZFS_BADPATH, errbuf)); 1662 } 1663 fromname = name; 1664 } 1665 1666 if (flags->progress) { 1667 error = lzc_send_space(zhp->zfs_name, fromname, 1668 lzc_flags, &size); 1669 if (error == 0) 1670 size = MAX(0, (int64_t)(size - bytes)); 1671 } 1672 if (flags->verbose) { 1673 send_print_verbose(fout, zhp->zfs_name, fromname, 1674 size, flags->parsable); 1675 } 1676 1677 if (!flags->dryrun) { 1678 progress_arg_t pa = { 0 }; 1679 pthread_t tid; 1680 /* 1681 * If progress reporting is requested, spawn a new thread to 1682 * poll ZFS_IOC_SEND_PROGRESS at a regular interval. 1683 */ 1684 if (flags->progress) { 1685 pa.pa_zhp = zhp; 1686 pa.pa_fd = outfd; 1687 pa.pa_parsable = flags->parsable; 1688 pa.pa_size = size; 1689 pa.pa_astitle = flags->progressastitle; 1690 1691 error = pthread_create(&tid, NULL, 1692 send_progress_thread, &pa); 1693 if (error != 0) { 1694 zfs_close(zhp); 1695 return (error); 1696 } 1697 } 1698 1699 error = lzc_send_resume(zhp->zfs_name, fromname, outfd, 1700 lzc_flags, resumeobj, resumeoff); 1701 1702 if (flags->progress) { 1703 (void) pthread_cancel(tid); 1704 (void) pthread_join(tid, NULL); 1705 } 1706 1707 char errbuf[1024]; 1708 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1709 "warning: cannot send '%s'"), zhp->zfs_name); 1710 1711 zfs_close(zhp); 1712 1713 switch (error) { 1714 case 0: 1715 return (0); 1716 case EXDEV: 1717 case ENOENT: 1718 case EDQUOT: 1719 case EFBIG: 1720 case EIO: 1721 case ENOLINK: 1722 case ENOSPC: 1723#ifdef illumos 1724 case ENOSTR: 1725#endif 1726 case ENXIO: 1727 case EPIPE: 1728 case ERANGE: 1729 case EFAULT: 1730 case EROFS: 1731 zfs_error_aux(hdl, strerror(errno)); 1732 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 1733 1734 default: 1735 return (zfs_standard_error(hdl, errno, errbuf)); 1736 } 1737 } 1738 1739 1740 zfs_close(zhp); 1741 1742 return (error); 1743} 1744 1745/* 1746 * Generate a send stream for the dataset identified by the argument zhp. 1747 * 1748 * The content of the send stream is the snapshot identified by 1749 * 'tosnap'. Incremental streams are requested in two ways: 1750 * - from the snapshot identified by "fromsnap" (if non-null) or 1751 * - from the origin of the dataset identified by zhp, which must 1752 * be a clone. In this case, "fromsnap" is null and "fromorigin" 1753 * is TRUE. 1754 * 1755 * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and 1756 * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM) 1757 * if "replicate" is set. If "doall" is set, dump all the intermediate 1758 * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall" 1759 * case too. If "props" is set, send properties. 1760 */ 1761int 1762zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, 1763 sendflags_t *flags, int outfd, snapfilter_cb_t filter_func, 1764 void *cb_arg, nvlist_t **debugnvp) 1765{ 1766 char errbuf[1024]; 1767 send_dump_data_t sdd = { 0 }; 1768 int err = 0; 1769 nvlist_t *fss = NULL; 1770 avl_tree_t *fsavl = NULL; 1771 static uint64_t holdseq; 1772 int spa_version; 1773 pthread_t tid = 0; 1774 int pipefd[2]; 1775 dedup_arg_t dda = { 0 }; 1776 int featureflags = 0; 1777 FILE *fout; 1778 1779 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1780 "cannot send '%s'"), zhp->zfs_name); 1781 1782 if (fromsnap && fromsnap[0] == '\0') { 1783 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN, 1784 "zero-length incremental source")); 1785 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf)); 1786 } 1787 1788 if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) { 1789 uint64_t version; 1790 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION); 1791 if (version >= ZPL_VERSION_SA) { 1792 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; 1793 } 1794 } 1795 1796 if (flags->dedup && !flags->dryrun) { 1797 featureflags |= (DMU_BACKUP_FEATURE_DEDUP | 1798 DMU_BACKUP_FEATURE_DEDUPPROPS); 1799 if ((err = pipe(pipefd)) != 0) { 1800 zfs_error_aux(zhp->zfs_hdl, strerror(errno)); 1801 return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED, 1802 errbuf)); 1803 } 1804 dda.outputfd = outfd; 1805 dda.inputfd = pipefd[1]; 1806 dda.dedup_hdl = zhp->zfs_hdl; 1807 if ((err = pthread_create(&tid, NULL, cksummer, &dda)) != 0) { 1808 (void) close(pipefd[0]); 1809 (void) close(pipefd[1]); 1810 zfs_error_aux(zhp->zfs_hdl, strerror(errno)); 1811 return (zfs_error(zhp->zfs_hdl, 1812 EZFS_THREADCREATEFAILED, errbuf)); 1813 } 1814 } 1815 1816 if (flags->replicate || flags->doall || flags->props) { 1817 dmu_replay_record_t drr = { 0 }; 1818 char *packbuf = NULL; 1819 size_t buflen = 0; 1820 zio_cksum_t zc = { 0 }; 1821 1822 if (flags->replicate || flags->props) { 1823 nvlist_t *hdrnv; 1824 1825 VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0)); 1826 if (fromsnap) { 1827 VERIFY(0 == nvlist_add_string(hdrnv, 1828 "fromsnap", fromsnap)); 1829 } 1830 VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap)); 1831 if (!flags->replicate) { 1832 VERIFY(0 == nvlist_add_boolean(hdrnv, 1833 "not_recursive")); 1834 } 1835 1836 err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name, 1837 fromsnap, tosnap, flags->replicate, flags->verbose, 1838 &fss, &fsavl); 1839 if (err) 1840 goto err_out; 1841 VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss)); 1842 err = nvlist_pack(hdrnv, &packbuf, &buflen, 1843 NV_ENCODE_XDR, 0); 1844 if (debugnvp) 1845 *debugnvp = hdrnv; 1846 else 1847 nvlist_free(hdrnv); 1848 if (err) 1849 goto stderr_out; 1850 } 1851 1852 if (!flags->dryrun) { 1853 /* write first begin record */ 1854 drr.drr_type = DRR_BEGIN; 1855 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 1856 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin. 1857 drr_versioninfo, DMU_COMPOUNDSTREAM); 1858 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin. 1859 drr_versioninfo, featureflags); 1860 (void) snprintf(drr.drr_u.drr_begin.drr_toname, 1861 sizeof (drr.drr_u.drr_begin.drr_toname), 1862 "%s@%s", zhp->zfs_name, tosnap); 1863 drr.drr_payloadlen = buflen; 1864 1865 err = dump_record(&drr, packbuf, buflen, &zc, outfd); 1866 free(packbuf); 1867 if (err != 0) 1868 goto stderr_out; 1869 1870 /* write end record */ 1871 bzero(&drr, sizeof (drr)); 1872 drr.drr_type = DRR_END; 1873 drr.drr_u.drr_end.drr_checksum = zc; 1874 err = write(outfd, &drr, sizeof (drr)); 1875 if (err == -1) { 1876 err = errno; 1877 goto stderr_out; 1878 } 1879 1880 err = 0; 1881 } 1882 } 1883 1884 /* dump each stream */ 1885 sdd.fromsnap = fromsnap; 1886 sdd.tosnap = tosnap; 1887 if (tid != 0) 1888 sdd.outfd = pipefd[0]; 1889 else 1890 sdd.outfd = outfd; 1891 sdd.replicate = flags->replicate; 1892 sdd.doall = flags->doall; 1893 sdd.fromorigin = flags->fromorigin; 1894 sdd.fss = fss; 1895 sdd.fsavl = fsavl; 1896 sdd.verbose = flags->verbose; 1897 sdd.parsable = flags->parsable; 1898 sdd.progress = flags->progress; 1899 sdd.progressastitle = flags->progressastitle; 1900 sdd.dryrun = flags->dryrun; 1901 sdd.large_block = flags->largeblock; 1902 sdd.embed_data = flags->embed_data; 1903 sdd.compress = flags->compress; 1904 sdd.filter_cb = filter_func; 1905 sdd.filter_cb_arg = cb_arg; 1906 if (debugnvp) 1907 sdd.debugnv = *debugnvp; 1908 if (sdd.verbose && sdd.dryrun) 1909 sdd.std_out = B_TRUE; 1910 fout = sdd.std_out ? stdout : stderr; 1911 1912 /* 1913 * Some flags require that we place user holds on the datasets that are 1914 * being sent so they don't get destroyed during the send. We can skip 1915 * this step if the pool is imported read-only since the datasets cannot 1916 * be destroyed. 1917 */ 1918 if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp), 1919 ZPOOL_PROP_READONLY, NULL) && 1920 zfs_spa_version(zhp, &spa_version) == 0 && 1921 spa_version >= SPA_VERSION_USERREFS && 1922 (flags->doall || flags->replicate)) { 1923 ++holdseq; 1924 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag), 1925 ".send-%d-%llu", getpid(), (u_longlong_t)holdseq); 1926 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); 1927 if (sdd.cleanup_fd < 0) { 1928 err = errno; 1929 goto stderr_out; 1930 } 1931 sdd.snapholds = fnvlist_alloc(); 1932 } else { 1933 sdd.cleanup_fd = -1; 1934 sdd.snapholds = NULL; 1935 } 1936 if (flags->progress || sdd.snapholds != NULL) { 1937 /* 1938 * Do a verbose no-op dry run to get all the verbose output 1939 * or to gather snapshot hold's before generating any data, 1940 * then do a non-verbose real run to generate the streams. 1941 */ 1942 sdd.dryrun = B_TRUE; 1943 err = dump_filesystems(zhp, &sdd); 1944 1945 if (err != 0) 1946 goto stderr_out; 1947 1948 if (flags->verbose) { 1949 if (flags->parsable) { 1950 (void) fprintf(fout, "size\t%llu\n", 1951 (longlong_t)sdd.size); 1952 } else { 1953 char buf[16]; 1954 zfs_nicenum(sdd.size, buf, sizeof (buf)); 1955 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1956 "total estimated size is %s\n"), buf); 1957 } 1958 } 1959 1960 /* Ensure no snaps found is treated as an error. */ 1961 if (!sdd.seento) { 1962 err = ENOENT; 1963 goto err_out; 1964 } 1965 1966 /* Skip the second run if dryrun was requested. */ 1967 if (flags->dryrun) 1968 goto err_out; 1969 1970 if (sdd.snapholds != NULL) { 1971 err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds); 1972 if (err != 0) 1973 goto stderr_out; 1974 1975 fnvlist_free(sdd.snapholds); 1976 sdd.snapholds = NULL; 1977 } 1978 1979 sdd.dryrun = B_FALSE; 1980 sdd.verbose = B_FALSE; 1981 } 1982 1983 err = dump_filesystems(zhp, &sdd); 1984 fsavl_destroy(fsavl); 1985 nvlist_free(fss); 1986 1987 /* Ensure no snaps found is treated as an error. */ 1988 if (err == 0 && !sdd.seento) 1989 err = ENOENT; 1990 1991 if (tid != 0) { 1992 if (err != 0) 1993 (void) pthread_cancel(tid); 1994 (void) close(pipefd[0]); 1995 (void) pthread_join(tid, NULL); 1996 } 1997 1998 if (sdd.cleanup_fd != -1) { 1999 VERIFY(0 == close(sdd.cleanup_fd)); 2000 sdd.cleanup_fd = -1; 2001 } 2002 2003 if (!flags->dryrun && (flags->replicate || flags->doall || 2004 flags->props)) { 2005 /* 2006 * write final end record. NB: want to do this even if 2007 * there was some error, because it might not be totally 2008 * failed. 2009 */ 2010 dmu_replay_record_t drr = { 0 }; 2011 drr.drr_type = DRR_END; 2012 if (write(outfd, &drr, sizeof (drr)) == -1) { 2013 return (zfs_standard_error(zhp->zfs_hdl, 2014 errno, errbuf)); 2015 } 2016 } 2017 2018 return (err || sdd.err); 2019 2020stderr_out: 2021 err = zfs_standard_error(zhp->zfs_hdl, err, errbuf); 2022err_out: 2023 fsavl_destroy(fsavl); 2024 nvlist_free(fss); 2025 fnvlist_free(sdd.snapholds); 2026 2027 if (sdd.cleanup_fd != -1) 2028 VERIFY(0 == close(sdd.cleanup_fd)); 2029 if (tid != 0) { 2030 (void) pthread_cancel(tid); 2031 (void) close(pipefd[0]); 2032 (void) pthread_join(tid, NULL); 2033 } 2034 return (err); 2035} 2036 2037int 2038zfs_send_one(zfs_handle_t *zhp, const char *from, int fd, sendflags_t flags) 2039{ 2040 int err = 0; 2041 libzfs_handle_t *hdl = zhp->zfs_hdl; 2042 enum lzc_send_flags lzc_flags = 0; 2043 FILE *fout = (flags.verbose && flags.dryrun) ? stdout : stderr; 2044 char errbuf[1024]; 2045 2046 if (flags.largeblock) 2047 lzc_flags |= LZC_SEND_FLAG_LARGE_BLOCK; 2048 if (flags.embed_data) 2049 lzc_flags |= LZC_SEND_FLAG_EMBED_DATA; 2050 if (flags.compress) 2051 lzc_flags |= LZC_SEND_FLAG_COMPRESS; 2052 2053 if (flags.verbose) { 2054 uint64_t size = 0; 2055 err = lzc_send_space(zhp->zfs_name, from, lzc_flags, &size); 2056 if (err == 0) { 2057 send_print_verbose(fout, zhp->zfs_name, from, size, 2058 flags.parsable); 2059 if (flags.parsable) { 2060 (void) fprintf(fout, "size\t%llu\n", 2061 (longlong_t)size); 2062 } else { 2063 char buf[16]; 2064 zfs_nicenum(size, buf, sizeof (buf)); 2065 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 2066 "total estimated size is %s\n"), buf); 2067 } 2068 } else { 2069 (void) fprintf(stderr, "Cannot estimate send size: " 2070 "%s\n", strerror(errno)); 2071 } 2072 } 2073 2074 if (flags.dryrun) 2075 return (err); 2076 2077 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2078 "warning: cannot send '%s'"), zhp->zfs_name); 2079 2080 err = lzc_send(zhp->zfs_name, from, fd, lzc_flags); 2081 if (err != 0) { 2082 switch (errno) { 2083 case EXDEV: 2084 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2085 "not an earlier snapshot from the same fs")); 2086 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 2087 2088 case ENOENT: 2089 case ESRCH: 2090 if (lzc_exists(zhp->zfs_name)) { 2091 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2092 "incremental source (%s) does not exist"), 2093 from); 2094 } 2095 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 2096 2097 case EBUSY: 2098 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2099 "target is busy; if a filesystem, " 2100 "it must not be mounted")); 2101 return (zfs_error(hdl, EZFS_BUSY, errbuf)); 2102 2103 case EDQUOT: 2104 case EFBIG: 2105 case EIO: 2106 case ENOLINK: 2107 case ENOSPC: 2108#ifdef illumos 2109 case ENOSTR: 2110#endif 2111 case ENXIO: 2112 case EPIPE: 2113 case ERANGE: 2114 case EFAULT: 2115 case EROFS: 2116 zfs_error_aux(hdl, strerror(errno)); 2117 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 2118 2119 default: 2120 return (zfs_standard_error(hdl, errno, errbuf)); 2121 } 2122 } 2123 return (err != 0); 2124} 2125 2126/* 2127 * Routines specific to "zfs recv" 2128 */ 2129 2130static int 2131recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen, 2132 boolean_t byteswap, zio_cksum_t *zc) 2133{ 2134 char *cp = buf; 2135 int rv; 2136 int len = ilen; 2137 2138 assert(ilen <= SPA_MAXBLOCKSIZE); 2139 2140 do { 2141 rv = read(fd, cp, len); 2142 cp += rv; 2143 len -= rv; 2144 } while (rv > 0); 2145 2146 if (rv < 0 || len != 0) { 2147 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2148 "failed to read from stream")); 2149 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN, 2150 "cannot receive"))); 2151 } 2152 2153 if (zc) { 2154 if (byteswap) 2155 (void) fletcher_4_incremental_byteswap(buf, ilen, zc); 2156 else 2157 (void) fletcher_4_incremental_native(buf, ilen, zc); 2158 } 2159 return (0); 2160} 2161 2162static int 2163recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp, 2164 boolean_t byteswap, zio_cksum_t *zc) 2165{ 2166 char *buf; 2167 int err; 2168 2169 buf = zfs_alloc(hdl, len); 2170 if (buf == NULL) 2171 return (ENOMEM); 2172 2173 err = recv_read(hdl, fd, buf, len, byteswap, zc); 2174 if (err != 0) { 2175 free(buf); 2176 return (err); 2177 } 2178 2179 err = nvlist_unpack(buf, len, nvp, 0); 2180 free(buf); 2181 if (err != 0) { 2182 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 2183 "stream (malformed nvlist)")); 2184 return (EINVAL); 2185 } 2186 return (0); 2187} 2188 2189static int 2190recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname, 2191 int baselen, char *newname, recvflags_t *flags) 2192{ 2193 static int seq; 2194 int err; 2195 prop_changelist_t *clp; 2196 zfs_handle_t *zhp; 2197 2198 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 2199 if (zhp == NULL) 2200 return (-1); 2201 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 2202 flags->force ? MS_FORCE : 0); 2203 zfs_close(zhp); 2204 if (clp == NULL) 2205 return (-1); 2206 err = changelist_prefix(clp); 2207 if (err) 2208 return (err); 2209 2210 if (tryname) { 2211 (void) strcpy(newname, tryname); 2212 if (flags->verbose) { 2213 (void) printf("attempting rename %s to %s\n", 2214 name, newname); 2215 } 2216 err = lzc_rename(name, newname); 2217 if (err == 0) 2218 changelist_rename(clp, name, tryname); 2219 } else { 2220 err = ENOENT; 2221 } 2222 2223 if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) { 2224 seq++; 2225 2226 (void) snprintf(newname, ZFS_MAX_DATASET_NAME_LEN, 2227 "%.*srecv-%u-%u", baselen, name, getpid(), seq); 2228 if (flags->verbose) { 2229 (void) printf("failed - trying rename %s to %s\n", 2230 name, newname); 2231 } 2232 err = lzc_rename(name, newname); 2233 if (err == 0) 2234 changelist_rename(clp, name, newname); 2235 if (err && flags->verbose) { 2236 (void) printf("failed (%u) - " 2237 "will try again on next pass\n", errno); 2238 } 2239 err = EAGAIN; 2240 } else if (flags->verbose) { 2241 if (err == 0) 2242 (void) printf("success\n"); 2243 else 2244 (void) printf("failed (%u)\n", errno); 2245 } 2246 2247 (void) changelist_postfix(clp); 2248 changelist_free(clp); 2249 2250 return (err); 2251} 2252 2253static int 2254recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen, 2255 char *newname, recvflags_t *flags) 2256{ 2257 int err = 0; 2258 prop_changelist_t *clp; 2259 zfs_handle_t *zhp; 2260 boolean_t defer = B_FALSE; 2261 int spa_version; 2262 2263 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 2264 if (zhp == NULL) 2265 return (-1); 2266 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 2267 flags->force ? MS_FORCE : 0); 2268 if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT && 2269 zfs_spa_version(zhp, &spa_version) == 0 && 2270 spa_version >= SPA_VERSION_USERREFS) 2271 defer = B_TRUE; 2272 zfs_close(zhp); 2273 if (clp == NULL) 2274 return (-1); 2275 err = changelist_prefix(clp); 2276 if (err) 2277 return (err); 2278 2279 if (flags->verbose) 2280 (void) printf("attempting destroy %s\n", name); 2281 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) { 2282 nvlist_t *nv = fnvlist_alloc(); 2283 fnvlist_add_boolean(nv, name); 2284 err = lzc_destroy_snaps(nv, defer, NULL); 2285 fnvlist_free(nv); 2286 } else { 2287 err = lzc_destroy(name); 2288 } 2289 if (err == 0) { 2290 if (flags->verbose) 2291 (void) printf("success\n"); 2292 changelist_remove(clp, name); 2293 } 2294 2295 (void) changelist_postfix(clp); 2296 changelist_free(clp); 2297 2298 /* 2299 * Deferred destroy might destroy the snapshot or only mark it to be 2300 * destroyed later, and it returns success in either case. 2301 */ 2302 if (err != 0 || (defer && zfs_dataset_exists(hdl, name, 2303 ZFS_TYPE_SNAPSHOT))) { 2304 err = recv_rename(hdl, name, NULL, baselen, newname, flags); 2305 } 2306 2307 return (err); 2308} 2309 2310typedef struct guid_to_name_data { 2311 uint64_t guid; 2312 boolean_t bookmark_ok; 2313 char *name; 2314 char *skip; 2315} guid_to_name_data_t; 2316 2317static int 2318guid_to_name_cb(zfs_handle_t *zhp, void *arg) 2319{ 2320 guid_to_name_data_t *gtnd = arg; 2321 const char *slash; 2322 int err; 2323 2324 if (gtnd->skip != NULL && 2325 (slash = strrchr(zhp->zfs_name, '/')) != NULL && 2326 strcmp(slash + 1, gtnd->skip) == 0) { 2327 zfs_close(zhp); 2328 return (0); 2329 } 2330 2331 if (zfs_prop_get_int(zhp, ZFS_PROP_GUID) == gtnd->guid) { 2332 (void) strcpy(gtnd->name, zhp->zfs_name); 2333 zfs_close(zhp); 2334 return (EEXIST); 2335 } 2336 2337 err = zfs_iter_children(zhp, guid_to_name_cb, gtnd); 2338 if (err != EEXIST && gtnd->bookmark_ok) 2339 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, gtnd); 2340 zfs_close(zhp); 2341 return (err); 2342} 2343 2344/* 2345 * Attempt to find the local dataset associated with this guid. In the case of 2346 * multiple matches, we attempt to find the "best" match by searching 2347 * progressively larger portions of the hierarchy. This allows one to send a 2348 * tree of datasets individually and guarantee that we will find the source 2349 * guid within that hierarchy, even if there are multiple matches elsewhere. 2350 */ 2351static int 2352guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid, 2353 boolean_t bookmark_ok, char *name) 2354{ 2355 char pname[ZFS_MAX_DATASET_NAME_LEN]; 2356 guid_to_name_data_t gtnd; 2357 2358 gtnd.guid = guid; 2359 gtnd.bookmark_ok = bookmark_ok; 2360 gtnd.name = name; 2361 gtnd.skip = NULL; 2362 2363 /* 2364 * Search progressively larger portions of the hierarchy, starting 2365 * with the filesystem specified by 'parent'. This will 2366 * select the "most local" version of the origin snapshot in the case 2367 * that there are multiple matching snapshots in the system. 2368 */ 2369 (void) strlcpy(pname, parent, sizeof (pname)); 2370 char *cp = strrchr(pname, '@'); 2371 if (cp == NULL) 2372 cp = strchr(pname, '\0'); 2373 for (; cp != NULL; cp = strrchr(pname, '/')) { 2374 /* Chop off the last component and open the parent */ 2375 *cp = '\0'; 2376 zfs_handle_t *zhp = make_dataset_handle(hdl, pname); 2377 2378 if (zhp == NULL) 2379 continue; 2380 int err = guid_to_name_cb(zfs_handle_dup(zhp), >nd); 2381 if (err != EEXIST) 2382 err = zfs_iter_children(zhp, guid_to_name_cb, >nd); 2383 if (err != EEXIST && bookmark_ok) 2384 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, >nd); 2385 zfs_close(zhp); 2386 if (err == EEXIST) 2387 return (0); 2388 2389 /* 2390 * Remember the last portion of the dataset so we skip it next 2391 * time through (as we've already searched that portion of the 2392 * hierarchy). 2393 */ 2394 gtnd.skip = strrchr(pname, '/') + 1; 2395 } 2396 2397 return (ENOENT); 2398} 2399 2400/* 2401 * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if 2402 * guid1 is after guid2. 2403 */ 2404static int 2405created_before(libzfs_handle_t *hdl, avl_tree_t *avl, 2406 uint64_t guid1, uint64_t guid2) 2407{ 2408 nvlist_t *nvfs; 2409 char *fsname, *snapname; 2410 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2411 int rv; 2412 zfs_handle_t *guid1hdl, *guid2hdl; 2413 uint64_t create1, create2; 2414 2415 if (guid2 == 0) 2416 return (0); 2417 if (guid1 == 0) 2418 return (1); 2419 2420 nvfs = fsavl_find(avl, guid1, &snapname); 2421 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2422 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); 2423 guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); 2424 if (guid1hdl == NULL) 2425 return (-1); 2426 2427 nvfs = fsavl_find(avl, guid2, &snapname); 2428 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2429 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); 2430 guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); 2431 if (guid2hdl == NULL) { 2432 zfs_close(guid1hdl); 2433 return (-1); 2434 } 2435 2436 create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG); 2437 create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG); 2438 2439 if (create1 < create2) 2440 rv = -1; 2441 else if (create1 > create2) 2442 rv = +1; 2443 else 2444 rv = 0; 2445 2446 zfs_close(guid1hdl); 2447 zfs_close(guid2hdl); 2448 2449 return (rv); 2450} 2451 2452static int 2453recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs, 2454 recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl, 2455 nvlist_t *renamed) 2456{ 2457 nvlist_t *local_nv, *deleted = NULL; 2458 avl_tree_t *local_avl; 2459 nvpair_t *fselem, *nextfselem; 2460 char *fromsnap; 2461 char newname[ZFS_MAX_DATASET_NAME_LEN]; 2462 char guidname[32]; 2463 int error; 2464 boolean_t needagain, progress, recursive; 2465 char *s1, *s2; 2466 2467 VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap)); 2468 2469 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2470 ENOENT); 2471 2472 if (flags->dryrun) 2473 return (0); 2474 2475again: 2476 needagain = progress = B_FALSE; 2477 2478 VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0)); 2479 2480 if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL, 2481 recursive, B_FALSE, &local_nv, &local_avl)) != 0) 2482 return (error); 2483 2484 /* 2485 * Process deletes and renames 2486 */ 2487 for (fselem = nvlist_next_nvpair(local_nv, NULL); 2488 fselem; fselem = nextfselem) { 2489 nvlist_t *nvfs, *snaps; 2490 nvlist_t *stream_nvfs = NULL; 2491 nvpair_t *snapelem, *nextsnapelem; 2492 uint64_t fromguid = 0; 2493 uint64_t originguid = 0; 2494 uint64_t stream_originguid = 0; 2495 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid; 2496 char *fsname, *stream_fsname; 2497 2498 nextfselem = nvlist_next_nvpair(local_nv, fselem); 2499 2500 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); 2501 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); 2502 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2503 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap", 2504 &parent_fromsnap_guid)); 2505 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid); 2506 2507 /* 2508 * First find the stream's fs, so we can check for 2509 * a different origin (due to "zfs promote") 2510 */ 2511 for (snapelem = nvlist_next_nvpair(snaps, NULL); 2512 snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) { 2513 uint64_t thisguid; 2514 2515 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); 2516 stream_nvfs = fsavl_find(stream_avl, thisguid, NULL); 2517 2518 if (stream_nvfs != NULL) 2519 break; 2520 } 2521 2522 /* check for promote */ 2523 (void) nvlist_lookup_uint64(stream_nvfs, "origin", 2524 &stream_originguid); 2525 if (stream_nvfs && originguid != stream_originguid) { 2526 switch (created_before(hdl, local_avl, 2527 stream_originguid, originguid)) { 2528 case 1: { 2529 /* promote it! */ 2530 zfs_cmd_t zc = { 0 }; 2531 nvlist_t *origin_nvfs; 2532 char *origin_fsname; 2533 2534 if (flags->verbose) 2535 (void) printf("promoting %s\n", fsname); 2536 2537 origin_nvfs = fsavl_find(local_avl, originguid, 2538 NULL); 2539 VERIFY(0 == nvlist_lookup_string(origin_nvfs, 2540 "name", &origin_fsname)); 2541 (void) strlcpy(zc.zc_value, origin_fsname, 2542 sizeof (zc.zc_value)); 2543 (void) strlcpy(zc.zc_name, fsname, 2544 sizeof (zc.zc_name)); 2545 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc); 2546 if (error == 0) 2547 progress = B_TRUE; 2548 break; 2549 } 2550 default: 2551 break; 2552 case -1: 2553 fsavl_destroy(local_avl); 2554 nvlist_free(local_nv); 2555 return (-1); 2556 } 2557 /* 2558 * We had/have the wrong origin, therefore our 2559 * list of snapshots is wrong. Need to handle 2560 * them on the next pass. 2561 */ 2562 needagain = B_TRUE; 2563 continue; 2564 } 2565 2566 for (snapelem = nvlist_next_nvpair(snaps, NULL); 2567 snapelem; snapelem = nextsnapelem) { 2568 uint64_t thisguid; 2569 char *stream_snapname; 2570 nvlist_t *found, *props; 2571 2572 nextsnapelem = nvlist_next_nvpair(snaps, snapelem); 2573 2574 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); 2575 found = fsavl_find(stream_avl, thisguid, 2576 &stream_snapname); 2577 2578 /* check for delete */ 2579 if (found == NULL) { 2580 char name[ZFS_MAX_DATASET_NAME_LEN]; 2581 2582 if (!flags->force) 2583 continue; 2584 2585 (void) snprintf(name, sizeof (name), "%s@%s", 2586 fsname, nvpair_name(snapelem)); 2587 2588 error = recv_destroy(hdl, name, 2589 strlen(fsname)+1, newname, flags); 2590 if (error) 2591 needagain = B_TRUE; 2592 else 2593 progress = B_TRUE; 2594 sprintf(guidname, "%" PRIu64, thisguid); 2595 nvlist_add_boolean(deleted, guidname); 2596 continue; 2597 } 2598 2599 stream_nvfs = found; 2600 2601 if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops", 2602 &props) && 0 == nvlist_lookup_nvlist(props, 2603 stream_snapname, &props)) { 2604 zfs_cmd_t zc = { 0 }; 2605 2606 zc.zc_cookie = B_TRUE; /* received */ 2607 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), 2608 "%s@%s", fsname, nvpair_name(snapelem)); 2609 if (zcmd_write_src_nvlist(hdl, &zc, 2610 props) == 0) { 2611 (void) zfs_ioctl(hdl, 2612 ZFS_IOC_SET_PROP, &zc); 2613 zcmd_free_nvlists(&zc); 2614 } 2615 } 2616 2617 /* check for different snapname */ 2618 if (strcmp(nvpair_name(snapelem), 2619 stream_snapname) != 0) { 2620 char name[ZFS_MAX_DATASET_NAME_LEN]; 2621 char tryname[ZFS_MAX_DATASET_NAME_LEN]; 2622 2623 (void) snprintf(name, sizeof (name), "%s@%s", 2624 fsname, nvpair_name(snapelem)); 2625 (void) snprintf(tryname, sizeof (name), "%s@%s", 2626 fsname, stream_snapname); 2627 2628 error = recv_rename(hdl, name, tryname, 2629 strlen(fsname)+1, newname, flags); 2630 if (error) 2631 needagain = B_TRUE; 2632 else 2633 progress = B_TRUE; 2634 } 2635 2636 if (strcmp(stream_snapname, fromsnap) == 0) 2637 fromguid = thisguid; 2638 } 2639 2640 /* check for delete */ 2641 if (stream_nvfs == NULL) { 2642 if (!flags->force) 2643 continue; 2644 2645 error = recv_destroy(hdl, fsname, strlen(tofs)+1, 2646 newname, flags); 2647 if (error) 2648 needagain = B_TRUE; 2649 else 2650 progress = B_TRUE; 2651 sprintf(guidname, "%" PRIu64, parent_fromsnap_guid); 2652 nvlist_add_boolean(deleted, guidname); 2653 continue; 2654 } 2655 2656 if (fromguid == 0) { 2657 if (flags->verbose) { 2658 (void) printf("local fs %s does not have " 2659 "fromsnap (%s in stream); must have " 2660 "been deleted locally; ignoring\n", 2661 fsname, fromsnap); 2662 } 2663 continue; 2664 } 2665 2666 VERIFY(0 == nvlist_lookup_string(stream_nvfs, 2667 "name", &stream_fsname)); 2668 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs, 2669 "parentfromsnap", &stream_parent_fromsnap_guid)); 2670 2671 s1 = strrchr(fsname, '/'); 2672 s2 = strrchr(stream_fsname, '/'); 2673 2674 /* 2675 * Check if we're going to rename based on parent guid change 2676 * and the current parent guid was also deleted. If it was then 2677 * rename will fail and is likely unneeded, so avoid this and 2678 * force an early retry to determine the new 2679 * parent_fromsnap_guid. 2680 */ 2681 if (stream_parent_fromsnap_guid != 0 && 2682 parent_fromsnap_guid != 0 && 2683 stream_parent_fromsnap_guid != parent_fromsnap_guid) { 2684 sprintf(guidname, "%" PRIu64, parent_fromsnap_guid); 2685 if (nvlist_exists(deleted, guidname)) { 2686 progress = B_TRUE; 2687 needagain = B_TRUE; 2688 goto doagain; 2689 } 2690 } 2691 2692 /* 2693 * Check for rename. If the exact receive path is specified, it 2694 * does not count as a rename, but we still need to check the 2695 * datasets beneath it. 2696 */ 2697 if ((stream_parent_fromsnap_guid != 0 && 2698 parent_fromsnap_guid != 0 && 2699 stream_parent_fromsnap_guid != parent_fromsnap_guid) || 2700 ((flags->isprefix || strcmp(tofs, fsname) != 0) && 2701 (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) { 2702 nvlist_t *parent; 2703 char tryname[ZFS_MAX_DATASET_NAME_LEN]; 2704 2705 parent = fsavl_find(local_avl, 2706 stream_parent_fromsnap_guid, NULL); 2707 /* 2708 * NB: parent might not be found if we used the 2709 * tosnap for stream_parent_fromsnap_guid, 2710 * because the parent is a newly-created fs; 2711 * we'll be able to rename it after we recv the 2712 * new fs. 2713 */ 2714 if (parent != NULL) { 2715 char *pname; 2716 2717 VERIFY(0 == nvlist_lookup_string(parent, "name", 2718 &pname)); 2719 (void) snprintf(tryname, sizeof (tryname), 2720 "%s%s", pname, strrchr(stream_fsname, '/')); 2721 } else { 2722 tryname[0] = '\0'; 2723 if (flags->verbose) { 2724 (void) printf("local fs %s new parent " 2725 "not found\n", fsname); 2726 } 2727 } 2728 2729 newname[0] = '\0'; 2730 2731 error = recv_rename(hdl, fsname, tryname, 2732 strlen(tofs)+1, newname, flags); 2733 2734 if (renamed != NULL && newname[0] != '\0') { 2735 VERIFY(0 == nvlist_add_boolean(renamed, 2736 newname)); 2737 } 2738 2739 if (error) 2740 needagain = B_TRUE; 2741 else 2742 progress = B_TRUE; 2743 } 2744 } 2745 2746doagain: 2747 fsavl_destroy(local_avl); 2748 nvlist_free(local_nv); 2749 nvlist_free(deleted); 2750 2751 if (needagain && progress) { 2752 /* do another pass to fix up temporary names */ 2753 if (flags->verbose) 2754 (void) printf("another pass:\n"); 2755 goto again; 2756 } 2757 2758 return (needagain); 2759} 2760 2761static int 2762zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname, 2763 recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc, 2764 char **top_zfs, int cleanup_fd, uint64_t *action_handlep) 2765{ 2766 nvlist_t *stream_nv = NULL; 2767 avl_tree_t *stream_avl = NULL; 2768 char *fromsnap = NULL; 2769 char *sendsnap = NULL; 2770 char *cp; 2771 char tofs[ZFS_MAX_DATASET_NAME_LEN]; 2772 char sendfs[ZFS_MAX_DATASET_NAME_LEN]; 2773 char errbuf[1024]; 2774 dmu_replay_record_t drre; 2775 int error; 2776 boolean_t anyerr = B_FALSE; 2777 boolean_t softerr = B_FALSE; 2778 boolean_t recursive; 2779 2780 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2781 "cannot receive")); 2782 2783 assert(drr->drr_type == DRR_BEGIN); 2784 assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC); 2785 assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) == 2786 DMU_COMPOUNDSTREAM); 2787 2788 /* 2789 * Read in the nvlist from the stream. 2790 */ 2791 if (drr->drr_payloadlen != 0) { 2792 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen, 2793 &stream_nv, flags->byteswap, zc); 2794 if (error) { 2795 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2796 goto out; 2797 } 2798 } 2799 2800 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2801 ENOENT); 2802 2803 if (recursive && strchr(destname, '@')) { 2804 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2805 "cannot specify snapshot name for multi-snapshot stream")); 2806 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2807 goto out; 2808 } 2809 2810 /* 2811 * Read in the end record and verify checksum. 2812 */ 2813 if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre), 2814 flags->byteswap, NULL))) 2815 goto out; 2816 if (flags->byteswap) { 2817 drre.drr_type = BSWAP_32(drre.drr_type); 2818 drre.drr_u.drr_end.drr_checksum.zc_word[0] = 2819 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]); 2820 drre.drr_u.drr_end.drr_checksum.zc_word[1] = 2821 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]); 2822 drre.drr_u.drr_end.drr_checksum.zc_word[2] = 2823 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]); 2824 drre.drr_u.drr_end.drr_checksum.zc_word[3] = 2825 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]); 2826 } 2827 if (drre.drr_type != DRR_END) { 2828 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2829 goto out; 2830 } 2831 if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) { 2832 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2833 "incorrect header checksum")); 2834 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2835 goto out; 2836 } 2837 2838 (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap); 2839 2840 if (drr->drr_payloadlen != 0) { 2841 nvlist_t *stream_fss; 2842 2843 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss", 2844 &stream_fss)); 2845 if ((stream_avl = fsavl_create(stream_fss)) == NULL) { 2846 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2847 "couldn't allocate avl tree")); 2848 error = zfs_error(hdl, EZFS_NOMEM, errbuf); 2849 goto out; 2850 } 2851 2852 if (fromsnap != NULL && recursive) { 2853 nvlist_t *renamed = NULL; 2854 nvpair_t *pair = NULL; 2855 2856 (void) strlcpy(tofs, destname, sizeof (tofs)); 2857 if (flags->isprefix) { 2858 struct drr_begin *drrb = &drr->drr_u.drr_begin; 2859 int i; 2860 2861 if (flags->istail) { 2862 cp = strrchr(drrb->drr_toname, '/'); 2863 if (cp == NULL) { 2864 (void) strlcat(tofs, "/", 2865 sizeof (tofs)); 2866 i = 0; 2867 } else { 2868 i = (cp - drrb->drr_toname); 2869 } 2870 } else { 2871 i = strcspn(drrb->drr_toname, "/@"); 2872 } 2873 /* zfs_receive_one() will create_parents() */ 2874 (void) strlcat(tofs, &drrb->drr_toname[i], 2875 sizeof (tofs)); 2876 *strchr(tofs, '@') = '\0'; 2877 } 2878 2879 if (!flags->dryrun && !flags->nomount) { 2880 VERIFY(0 == nvlist_alloc(&renamed, 2881 NV_UNIQUE_NAME, 0)); 2882 } 2883 2884 softerr = recv_incremental_replication(hdl, tofs, flags, 2885 stream_nv, stream_avl, renamed); 2886 2887 /* Unmount renamed filesystems before receiving. */ 2888 while ((pair = nvlist_next_nvpair(renamed, 2889 pair)) != NULL) { 2890 zfs_handle_t *zhp; 2891 prop_changelist_t *clp = NULL; 2892 2893 zhp = zfs_open(hdl, nvpair_name(pair), 2894 ZFS_TYPE_FILESYSTEM); 2895 if (zhp != NULL) { 2896 clp = changelist_gather(zhp, 2897 ZFS_PROP_MOUNTPOINT, 0, 0); 2898 zfs_close(zhp); 2899 if (clp != NULL) { 2900 softerr |= 2901 changelist_prefix(clp); 2902 changelist_free(clp); 2903 } 2904 } 2905 } 2906 2907 nvlist_free(renamed); 2908 } 2909 } 2910 2911 /* 2912 * Get the fs specified by the first path in the stream (the top level 2913 * specified by 'zfs send') and pass it to each invocation of 2914 * zfs_receive_one(). 2915 */ 2916 (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname, 2917 sizeof (sendfs)); 2918 if ((cp = strchr(sendfs, '@')) != NULL) { 2919 *cp = '\0'; 2920 /* 2921 * Find the "sendsnap", the final snapshot in a replication 2922 * stream. zfs_receive_one() handles certain errors 2923 * differently, depending on if the contained stream is the 2924 * last one or not. 2925 */ 2926 sendsnap = (cp + 1); 2927 } 2928 2929 /* Finally, receive each contained stream */ 2930 do { 2931 /* 2932 * we should figure out if it has a recoverable 2933 * error, in which case do a recv_skip() and drive on. 2934 * Note, if we fail due to already having this guid, 2935 * zfs_receive_one() will take care of it (ie, 2936 * recv_skip() and return 0). 2937 */ 2938 error = zfs_receive_impl(hdl, destname, NULL, flags, fd, 2939 sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd, 2940 action_handlep, sendsnap); 2941 if (error == ENODATA) { 2942 error = 0; 2943 break; 2944 } 2945 anyerr |= error; 2946 } while (error == 0); 2947 2948 if (drr->drr_payloadlen != 0 && recursive && fromsnap != NULL) { 2949 /* 2950 * Now that we have the fs's they sent us, try the 2951 * renames again. 2952 */ 2953 softerr = recv_incremental_replication(hdl, tofs, flags, 2954 stream_nv, stream_avl, NULL); 2955 } 2956 2957out: 2958 fsavl_destroy(stream_avl); 2959 nvlist_free(stream_nv); 2960 if (softerr) 2961 error = -2; 2962 if (anyerr) 2963 error = -1; 2964 return (error); 2965} 2966 2967static void 2968trunc_prop_errs(int truncated) 2969{ 2970 ASSERT(truncated != 0); 2971 2972 if (truncated == 1) 2973 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 2974 "1 more property could not be set\n")); 2975 else 2976 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 2977 "%d more properties could not be set\n"), truncated); 2978} 2979 2980static int 2981recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap) 2982{ 2983 dmu_replay_record_t *drr; 2984 void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE); 2985 char errbuf[1024]; 2986 2987 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2988 "cannot receive:")); 2989 2990 /* XXX would be great to use lseek if possible... */ 2991 drr = buf; 2992 2993 while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t), 2994 byteswap, NULL) == 0) { 2995 if (byteswap) 2996 drr->drr_type = BSWAP_32(drr->drr_type); 2997 2998 switch (drr->drr_type) { 2999 case DRR_BEGIN: 3000 if (drr->drr_payloadlen != 0) { 3001 (void) recv_read(hdl, fd, buf, 3002 drr->drr_payloadlen, B_FALSE, NULL); 3003 } 3004 break; 3005 3006 case DRR_END: 3007 free(buf); 3008 return (0); 3009 3010 case DRR_OBJECT: 3011 if (byteswap) { 3012 drr->drr_u.drr_object.drr_bonuslen = 3013 BSWAP_32(drr->drr_u.drr_object. 3014 drr_bonuslen); 3015 } 3016 (void) recv_read(hdl, fd, buf, 3017 P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8), 3018 B_FALSE, NULL); 3019 break; 3020 3021 case DRR_WRITE: 3022 if (byteswap) { 3023 drr->drr_u.drr_write.drr_logical_size = 3024 BSWAP_64( 3025 drr->drr_u.drr_write.drr_logical_size); 3026 drr->drr_u.drr_write.drr_compressed_size = 3027 BSWAP_64( 3028 drr->drr_u.drr_write.drr_compressed_size); 3029 } 3030 uint64_t payload_size = 3031 DRR_WRITE_PAYLOAD_SIZE(&drr->drr_u.drr_write); 3032 (void) recv_read(hdl, fd, buf, 3033 payload_size, B_FALSE, NULL); 3034 break; 3035 case DRR_SPILL: 3036 if (byteswap) { 3037 drr->drr_u.drr_spill.drr_length = 3038 BSWAP_64(drr->drr_u.drr_spill.drr_length); 3039 } 3040 (void) recv_read(hdl, fd, buf, 3041 drr->drr_u.drr_spill.drr_length, B_FALSE, NULL); 3042 break; 3043 case DRR_WRITE_EMBEDDED: 3044 if (byteswap) { 3045 drr->drr_u.drr_write_embedded.drr_psize = 3046 BSWAP_32(drr->drr_u.drr_write_embedded. 3047 drr_psize); 3048 } 3049 (void) recv_read(hdl, fd, buf, 3050 P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize, 3051 8), B_FALSE, NULL); 3052 break; 3053 case DRR_WRITE_BYREF: 3054 case DRR_FREEOBJECTS: 3055 case DRR_FREE: 3056 break; 3057 3058 default: 3059 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3060 "invalid record type")); 3061 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3062 } 3063 } 3064 3065 free(buf); 3066 return (-1); 3067} 3068 3069static void 3070recv_ecksum_set_aux(libzfs_handle_t *hdl, const char *target_snap, 3071 boolean_t resumable) 3072{ 3073 char target_fs[ZFS_MAX_DATASET_NAME_LEN]; 3074 3075 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3076 "checksum mismatch or incomplete stream")); 3077 3078 if (!resumable) 3079 return; 3080 (void) strlcpy(target_fs, target_snap, sizeof (target_fs)); 3081 *strchr(target_fs, '@') = '\0'; 3082 zfs_handle_t *zhp = zfs_open(hdl, target_fs, 3083 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 3084 if (zhp == NULL) 3085 return; 3086 3087 char token_buf[ZFS_MAXPROPLEN]; 3088 int error = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN, 3089 token_buf, sizeof (token_buf), 3090 NULL, NULL, 0, B_TRUE); 3091 if (error == 0) { 3092 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3093 "checksum mismatch or incomplete stream.\n" 3094 "Partially received snapshot is saved.\n" 3095 "A resuming stream can be generated on the sending " 3096 "system by running:\n" 3097 " zfs send -t %s"), 3098 token_buf); 3099 } 3100 zfs_close(zhp); 3101} 3102 3103/* 3104 * Restores a backup of tosnap from the file descriptor specified by infd. 3105 */ 3106static int 3107zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, 3108 const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr, 3109 dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv, 3110 avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd, 3111 uint64_t *action_handlep, const char *finalsnap) 3112{ 3113 zfs_cmd_t zc = { 0 }; 3114 time_t begin_time; 3115 int ioctl_err, ioctl_errno, err; 3116 char *cp; 3117 struct drr_begin *drrb = &drr->drr_u.drr_begin; 3118 char errbuf[1024]; 3119 char prop_errbuf[1024]; 3120 const char *chopprefix; 3121 boolean_t newfs = B_FALSE; 3122 boolean_t stream_wantsnewfs; 3123 uint64_t parent_snapguid = 0; 3124 prop_changelist_t *clp = NULL; 3125 nvlist_t *snapprops_nvlist = NULL; 3126 zprop_errflags_t prop_errflags; 3127 boolean_t recursive; 3128 char *snapname = NULL; 3129 3130 begin_time = time(NULL); 3131 3132 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 3133 "cannot receive")); 3134 3135 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 3136 ENOENT); 3137 3138 if (stream_avl != NULL) { 3139 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid, 3140 &snapname); 3141 nvlist_t *props; 3142 int ret; 3143 3144 (void) nvlist_lookup_uint64(fs, "parentfromsnap", 3145 &parent_snapguid); 3146 err = nvlist_lookup_nvlist(fs, "props", &props); 3147 if (err) 3148 VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0)); 3149 3150 if (flags->canmountoff) { 3151 VERIFY(0 == nvlist_add_uint64(props, 3152 zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0)); 3153 } 3154 ret = zcmd_write_src_nvlist(hdl, &zc, props); 3155 if (err) 3156 nvlist_free(props); 3157 3158 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) { 3159 VERIFY(0 == nvlist_lookup_nvlist(props, 3160 snapname, &snapprops_nvlist)); 3161 } 3162 3163 if (ret != 0) 3164 return (-1); 3165 } 3166 3167 cp = NULL; 3168 3169 /* 3170 * Determine how much of the snapshot name stored in the stream 3171 * we are going to tack on to the name they specified on the 3172 * command line, and how much we are going to chop off. 3173 * 3174 * If they specified a snapshot, chop the entire name stored in 3175 * the stream. 3176 */ 3177 if (flags->istail) { 3178 /* 3179 * A filesystem was specified with -e. We want to tack on only 3180 * the tail of the sent snapshot path. 3181 */ 3182 if (strchr(tosnap, '@')) { 3183 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3184 "argument - snapshot not allowed with -e")); 3185 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 3186 } 3187 3188 chopprefix = strrchr(sendfs, '/'); 3189 3190 if (chopprefix == NULL) { 3191 /* 3192 * The tail is the poolname, so we need to 3193 * prepend a path separator. 3194 */ 3195 int len = strlen(drrb->drr_toname); 3196 cp = malloc(len + 2); 3197 cp[0] = '/'; 3198 (void) strcpy(&cp[1], drrb->drr_toname); 3199 chopprefix = cp; 3200 } else { 3201 chopprefix = drrb->drr_toname + (chopprefix - sendfs); 3202 } 3203 } else if (flags->isprefix) { 3204 /* 3205 * A filesystem was specified with -d. We want to tack on 3206 * everything but the first element of the sent snapshot path 3207 * (all but the pool name). 3208 */ 3209 if (strchr(tosnap, '@')) { 3210 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3211 "argument - snapshot not allowed with -d")); 3212 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 3213 } 3214 3215 chopprefix = strchr(drrb->drr_toname, '/'); 3216 if (chopprefix == NULL) 3217 chopprefix = strchr(drrb->drr_toname, '@'); 3218 } else if (strchr(tosnap, '@') == NULL) { 3219 /* 3220 * If a filesystem was specified without -d or -e, we want to 3221 * tack on everything after the fs specified by 'zfs send'. 3222 */ 3223 chopprefix = drrb->drr_toname + strlen(sendfs); 3224 } else { 3225 /* A snapshot was specified as an exact path (no -d or -e). */ 3226 if (recursive) { 3227 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3228 "cannot specify snapshot name for multi-snapshot " 3229 "stream")); 3230 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3231 } 3232 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname); 3233 } 3234 3235 ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname); 3236 ASSERT(chopprefix > drrb->drr_toname); 3237 ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname)); 3238 ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' || 3239 chopprefix[0] == '\0'); 3240 3241 /* 3242 * Determine name of destination snapshot, store in zc_value. 3243 */ 3244 (void) strcpy(zc.zc_value, tosnap); 3245 (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value)); 3246#ifdef __FreeBSD__ 3247 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF) 3248 zfs_ioctl_version = get_zfs_ioctl_version(); 3249 /* 3250 * For forward compatibility hide tosnap in zc_value 3251 */ 3252 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 3253 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap); 3254#endif 3255 free(cp); 3256 if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) { 3257 zcmd_free_nvlists(&zc); 3258 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 3259 } 3260 3261 /* 3262 * Determine the name of the origin snapshot, store in zc_string. 3263 */ 3264 if (originsnap) { 3265 (void) strncpy(zc.zc_string, originsnap, sizeof (zc.zc_string)); 3266 if (flags->verbose) 3267 (void) printf("using provided clone origin %s\n", 3268 zc.zc_string); 3269 } else if (drrb->drr_flags & DRR_FLAG_CLONE) { 3270 if (guid_to_name(hdl, zc.zc_value, 3271 drrb->drr_fromguid, B_FALSE, zc.zc_string) != 0) { 3272 zcmd_free_nvlists(&zc); 3273 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3274 "local origin for clone %s does not exist"), 3275 zc.zc_value); 3276 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3277 } 3278 if (flags->verbose) 3279 (void) printf("found clone origin %s\n", zc.zc_string); 3280 } 3281 3282 boolean_t resuming = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 3283 DMU_BACKUP_FEATURE_RESUMING; 3284 stream_wantsnewfs = (drrb->drr_fromguid == 0 || 3285 (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && !resuming; 3286 3287 if (stream_wantsnewfs) { 3288 /* 3289 * if the parent fs does not exist, look for it based on 3290 * the parent snap GUID 3291 */ 3292 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 3293 "cannot receive new filesystem stream")); 3294 3295 (void) strcpy(zc.zc_name, zc.zc_value); 3296 cp = strrchr(zc.zc_name, '/'); 3297 if (cp) 3298 *cp = '\0'; 3299 if (cp && 3300 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 3301 char suffix[ZFS_MAX_DATASET_NAME_LEN]; 3302 (void) strcpy(suffix, strrchr(zc.zc_value, '/')); 3303 if (guid_to_name(hdl, zc.zc_name, parent_snapguid, 3304 B_FALSE, zc.zc_value) == 0) { 3305 *strchr(zc.zc_value, '@') = '\0'; 3306 (void) strcat(zc.zc_value, suffix); 3307 } 3308 } 3309 } else { 3310 /* 3311 * If the fs does not exist, look for it based on the 3312 * fromsnap GUID. 3313 */ 3314 if (resuming) { 3315 (void) snprintf(errbuf, sizeof (errbuf), 3316 dgettext(TEXT_DOMAIN, 3317 "cannot receive resume stream")); 3318 } else { 3319 (void) snprintf(errbuf, sizeof (errbuf), 3320 dgettext(TEXT_DOMAIN, 3321 "cannot receive incremental stream")); 3322 } 3323 3324 (void) strcpy(zc.zc_name, zc.zc_value); 3325 *strchr(zc.zc_name, '@') = '\0'; 3326 3327 /* 3328 * If the exact receive path was specified and this is the 3329 * topmost path in the stream, then if the fs does not exist we 3330 * should look no further. 3331 */ 3332 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname + 3333 strlen(sendfs)) != '\0' && *chopprefix != '@')) && 3334 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 3335 char snap[ZFS_MAX_DATASET_NAME_LEN]; 3336 (void) strcpy(snap, strchr(zc.zc_value, '@')); 3337 if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid, 3338 B_FALSE, zc.zc_value) == 0) { 3339 *strchr(zc.zc_value, '@') = '\0'; 3340 (void) strcat(zc.zc_value, snap); 3341 } 3342 } 3343 } 3344 3345 (void) strcpy(zc.zc_name, zc.zc_value); 3346 *strchr(zc.zc_name, '@') = '\0'; 3347 3348 if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 3349 zfs_handle_t *zhp; 3350 3351 /* 3352 * Destination fs exists. It must be one of these cases: 3353 * - an incremental send stream 3354 * - the stream specifies a new fs (full stream or clone) 3355 * and they want us to blow away the existing fs (and 3356 * have therefore specified -F and removed any snapshots) 3357 * - we are resuming a failed receive. 3358 */ 3359 if (stream_wantsnewfs) { 3360 if (!flags->force) { 3361 zcmd_free_nvlists(&zc); 3362 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3363 "destination '%s' exists\n" 3364 "must specify -F to overwrite it"), 3365 zc.zc_name); 3366 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 3367 } 3368 if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT, 3369 &zc) == 0) { 3370 zcmd_free_nvlists(&zc); 3371 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3372 "destination has snapshots (eg. %s)\n" 3373 "must destroy them to overwrite it"), 3374 zc.zc_name); 3375 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 3376 } 3377 } 3378 3379 if ((zhp = zfs_open(hdl, zc.zc_name, 3380 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) { 3381 zcmd_free_nvlists(&zc); 3382 return (-1); 3383 } 3384 3385 if (stream_wantsnewfs && 3386 zhp->zfs_dmustats.dds_origin[0]) { 3387 zcmd_free_nvlists(&zc); 3388 zfs_close(zhp); 3389 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3390 "destination '%s' is a clone\n" 3391 "must destroy it to overwrite it"), 3392 zc.zc_name); 3393 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 3394 } 3395 3396 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM && 3397 stream_wantsnewfs) { 3398 /* We can't do online recv in this case */ 3399 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0); 3400 if (clp == NULL) { 3401 zfs_close(zhp); 3402 zcmd_free_nvlists(&zc); 3403 return (-1); 3404 } 3405 if (changelist_prefix(clp) != 0) { 3406 changelist_free(clp); 3407 zfs_close(zhp); 3408 zcmd_free_nvlists(&zc); 3409 return (-1); 3410 } 3411 } 3412 3413 /* 3414 * If we are resuming a newfs, set newfs here so that we will 3415 * mount it if the recv succeeds this time. We can tell 3416 * that it was a newfs on the first recv because the fs 3417 * itself will be inconsistent (if the fs existed when we 3418 * did the first recv, we would have received it into 3419 * .../%recv). 3420 */ 3421 if (resuming && zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT)) 3422 newfs = B_TRUE; 3423 3424 zfs_close(zhp); 3425 } else { 3426 /* 3427 * Destination filesystem does not exist. Therefore we better 3428 * be creating a new filesystem (either from a full backup, or 3429 * a clone). It would therefore be invalid if the user 3430 * specified only the pool name (i.e. if the destination name 3431 * contained no slash character). 3432 */ 3433 if (!stream_wantsnewfs || 3434 (cp = strrchr(zc.zc_name, '/')) == NULL) { 3435 zcmd_free_nvlists(&zc); 3436 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3437 "destination '%s' does not exist"), zc.zc_name); 3438 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3439 } 3440 3441 /* 3442 * Trim off the final dataset component so we perform the 3443 * recvbackup ioctl to the filesystems's parent. 3444 */ 3445 *cp = '\0'; 3446 3447 if (flags->isprefix && !flags->istail && !flags->dryrun && 3448 create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) { 3449 zcmd_free_nvlists(&zc); 3450 return (zfs_error(hdl, EZFS_BADRESTORE, errbuf)); 3451 } 3452 3453 newfs = B_TRUE; 3454 } 3455 3456 zc.zc_begin_record = *drr_noswap; 3457 zc.zc_cookie = infd; 3458 zc.zc_guid = flags->force; 3459 zc.zc_resumable = flags->resumable; 3460 if (flags->verbose) { 3461 (void) printf("%s %s stream of %s into %s\n", 3462 flags->dryrun ? "would receive" : "receiving", 3463 drrb->drr_fromguid ? "incremental" : "full", 3464 drrb->drr_toname, zc.zc_value); 3465 (void) fflush(stdout); 3466 } 3467 3468 if (flags->dryrun) { 3469 zcmd_free_nvlists(&zc); 3470 return (recv_skip(hdl, infd, flags->byteswap)); 3471 } 3472 3473 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf; 3474 zc.zc_nvlist_dst_size = sizeof (prop_errbuf); 3475 zc.zc_cleanup_fd = cleanup_fd; 3476 zc.zc_action_handle = *action_handlep; 3477 3478 err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc); 3479 ioctl_errno = errno; 3480 prop_errflags = (zprop_errflags_t)zc.zc_obj; 3481 3482 if (err == 0) { 3483 nvlist_t *prop_errors; 3484 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, 3485 zc.zc_nvlist_dst_size, &prop_errors, 0)); 3486 3487 nvpair_t *prop_err = NULL; 3488 3489 while ((prop_err = nvlist_next_nvpair(prop_errors, 3490 prop_err)) != NULL) { 3491 char tbuf[1024]; 3492 zfs_prop_t prop; 3493 int intval; 3494 3495 prop = zfs_name_to_prop(nvpair_name(prop_err)); 3496 (void) nvpair_value_int32(prop_err, &intval); 3497 if (strcmp(nvpair_name(prop_err), 3498 ZPROP_N_MORE_ERRORS) == 0) { 3499 trunc_prop_errs(intval); 3500 break; 3501 } else if (snapname == NULL || finalsnap == NULL || 3502 strcmp(finalsnap, snapname) == 0 || 3503 strcmp(nvpair_name(prop_err), 3504 zfs_prop_to_name(ZFS_PROP_REFQUOTA)) != 0) { 3505 /* 3506 * Skip the special case of, for example, 3507 * "refquota", errors on intermediate 3508 * snapshots leading up to a final one. 3509 * That's why we have all of the checks above. 3510 * 3511 * See zfs_ioctl.c's extract_delay_props() for 3512 * a list of props which can fail on 3513 * intermediate snapshots, but shouldn't 3514 * affect the overall receive. 3515 */ 3516 (void) snprintf(tbuf, sizeof (tbuf), 3517 dgettext(TEXT_DOMAIN, 3518 "cannot receive %s property on %s"), 3519 nvpair_name(prop_err), zc.zc_name); 3520 zfs_setprop_error(hdl, prop, intval, tbuf); 3521 } 3522 } 3523 nvlist_free(prop_errors); 3524 } 3525 3526 zc.zc_nvlist_dst = 0; 3527 zc.zc_nvlist_dst_size = 0; 3528 zcmd_free_nvlists(&zc); 3529 3530 if (err == 0 && snapprops_nvlist) { 3531 zfs_cmd_t zc2 = { 0 }; 3532 3533 (void) strcpy(zc2.zc_name, zc.zc_value); 3534 zc2.zc_cookie = B_TRUE; /* received */ 3535 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) { 3536 (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2); 3537 zcmd_free_nvlists(&zc2); 3538 } 3539 } 3540 3541 if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) { 3542 /* 3543 * It may be that this snapshot already exists, 3544 * in which case we want to consume & ignore it 3545 * rather than failing. 3546 */ 3547 avl_tree_t *local_avl; 3548 nvlist_t *local_nv, *fs; 3549 cp = strchr(zc.zc_value, '@'); 3550 3551 /* 3552 * XXX Do this faster by just iterating over snaps in 3553 * this fs. Also if zc_value does not exist, we will 3554 * get a strange "does not exist" error message. 3555 */ 3556 *cp = '\0'; 3557 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE, 3558 B_FALSE, &local_nv, &local_avl) == 0) { 3559 *cp = '@'; 3560 fs = fsavl_find(local_avl, drrb->drr_toguid, NULL); 3561 fsavl_destroy(local_avl); 3562 nvlist_free(local_nv); 3563 3564 if (fs != NULL) { 3565 if (flags->verbose) { 3566 (void) printf("snap %s already exists; " 3567 "ignoring\n", zc.zc_value); 3568 } 3569 err = ioctl_err = recv_skip(hdl, infd, 3570 flags->byteswap); 3571 } 3572 } 3573 *cp = '@'; 3574 } 3575 3576 if (ioctl_err != 0) { 3577 switch (ioctl_errno) { 3578 case ENODEV: 3579 cp = strchr(zc.zc_value, '@'); 3580 *cp = '\0'; 3581 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3582 "most recent snapshot of %s does not\n" 3583 "match incremental source"), zc.zc_value); 3584 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); 3585 *cp = '@'; 3586 break; 3587 case ETXTBSY: 3588 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3589 "destination %s has been modified\n" 3590 "since most recent snapshot"), zc.zc_name); 3591 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); 3592 break; 3593 case EEXIST: 3594 cp = strchr(zc.zc_value, '@'); 3595 if (newfs) { 3596 /* it's the containing fs that exists */ 3597 *cp = '\0'; 3598 } 3599 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3600 "destination already exists")); 3601 (void) zfs_error_fmt(hdl, EZFS_EXISTS, 3602 dgettext(TEXT_DOMAIN, "cannot restore to %s"), 3603 zc.zc_value); 3604 *cp = '@'; 3605 break; 3606 case EINVAL: 3607 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); 3608 break; 3609 case ECKSUM: 3610 recv_ecksum_set_aux(hdl, zc.zc_value, flags->resumable); 3611 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); 3612 break; 3613 case ENOTSUP: 3614 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3615 "pool must be upgraded to receive this stream.")); 3616 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf); 3617 break; 3618 case EDQUOT: 3619 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3620 "destination %s space quota exceeded"), zc.zc_name); 3621 (void) zfs_error(hdl, EZFS_NOSPC, errbuf); 3622 break; 3623 default: 3624 (void) zfs_standard_error(hdl, ioctl_errno, errbuf); 3625 } 3626 } 3627 3628 /* 3629 * Mount the target filesystem (if created). Also mount any 3630 * children of the target filesystem if we did a replication 3631 * receive (indicated by stream_avl being non-NULL). 3632 */ 3633 cp = strchr(zc.zc_value, '@'); 3634 if (cp && (ioctl_err == 0 || !newfs)) { 3635 zfs_handle_t *h; 3636 3637 *cp = '\0'; 3638 h = zfs_open(hdl, zc.zc_value, 3639 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 3640 if (h != NULL) { 3641 if (h->zfs_type == ZFS_TYPE_VOLUME) { 3642 *cp = '@'; 3643 } else if (newfs || stream_avl) { 3644 /* 3645 * Track the first/top of hierarchy fs, 3646 * for mounting and sharing later. 3647 */ 3648 if (top_zfs && *top_zfs == NULL) 3649 *top_zfs = zfs_strdup(hdl, zc.zc_value); 3650 } 3651 zfs_close(h); 3652 } 3653 *cp = '@'; 3654 } 3655 3656 if (clp) { 3657 if (!flags->nomount) 3658 err |= changelist_postfix(clp); 3659 changelist_free(clp); 3660 } 3661 3662 if (prop_errflags & ZPROP_ERR_NOCLEAR) { 3663 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " 3664 "failed to clear unreceived properties on %s"), 3665 zc.zc_name); 3666 (void) fprintf(stderr, "\n"); 3667 } 3668 if (prop_errflags & ZPROP_ERR_NORESTORE) { 3669 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " 3670 "failed to restore original properties on %s"), 3671 zc.zc_name); 3672 (void) fprintf(stderr, "\n"); 3673 } 3674 3675 if (err || ioctl_err) 3676 return (-1); 3677 3678 *action_handlep = zc.zc_action_handle; 3679 3680 if (flags->verbose) { 3681 char buf1[64]; 3682 char buf2[64]; 3683 uint64_t bytes = zc.zc_cookie; 3684 time_t delta = time(NULL) - begin_time; 3685 if (delta == 0) 3686 delta = 1; 3687 zfs_nicenum(bytes, buf1, sizeof (buf1)); 3688 zfs_nicenum(bytes/delta, buf2, sizeof (buf1)); 3689 3690 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n", 3691 buf1, delta, buf2); 3692 } 3693 3694 return (0); 3695} 3696 3697static int 3698zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, 3699 const char *originsnap, recvflags_t *flags, int infd, const char *sendfs, 3700 nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd, 3701 uint64_t *action_handlep, const char *finalsnap) 3702{ 3703 int err; 3704 dmu_replay_record_t drr, drr_noswap; 3705 struct drr_begin *drrb = &drr.drr_u.drr_begin; 3706 char errbuf[1024]; 3707 zio_cksum_t zcksum = { 0 }; 3708 uint64_t featureflags; 3709 int hdrtype; 3710 3711 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 3712 "cannot receive")); 3713 3714 if (flags->isprefix && 3715 !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) { 3716 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs " 3717 "(%s) does not exist"), tosnap); 3718 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3719 } 3720 if (originsnap && 3721 !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) { 3722 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs " 3723 "(%s) does not exist"), originsnap); 3724 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3725 } 3726 3727 /* read in the BEGIN record */ 3728 if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE, 3729 &zcksum))) 3730 return (err); 3731 3732 if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) { 3733 /* It's the double end record at the end of a package */ 3734 return (ENODATA); 3735 } 3736 3737 /* the kernel needs the non-byteswapped begin record */ 3738 drr_noswap = drr; 3739 3740 flags->byteswap = B_FALSE; 3741 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { 3742 /* 3743 * We computed the checksum in the wrong byteorder in 3744 * recv_read() above; do it again correctly. 3745 */ 3746 bzero(&zcksum, sizeof (zio_cksum_t)); 3747 (void) fletcher_4_incremental_byteswap(&drr, 3748 sizeof (drr), &zcksum); 3749 flags->byteswap = B_TRUE; 3750 3751 drr.drr_type = BSWAP_32(drr.drr_type); 3752 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen); 3753 drrb->drr_magic = BSWAP_64(drrb->drr_magic); 3754 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); 3755 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); 3756 drrb->drr_type = BSWAP_32(drrb->drr_type); 3757 drrb->drr_flags = BSWAP_32(drrb->drr_flags); 3758 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); 3759 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); 3760 } 3761 3762 if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) { 3763 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3764 "stream (bad magic number)")); 3765 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3766 } 3767 3768 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 3769 hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo); 3770 3771 if (!DMU_STREAM_SUPPORTED(featureflags) || 3772 (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) { 3773 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3774 "stream has unsupported feature, feature flags = %lx"), 3775 featureflags); 3776 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3777 } 3778 3779 if (strchr(drrb->drr_toname, '@') == NULL) { 3780 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3781 "stream (bad snapshot name)")); 3782 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3783 } 3784 3785 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) { 3786 char nonpackage_sendfs[ZFS_MAX_DATASET_NAME_LEN]; 3787 if (sendfs == NULL) { 3788 /* 3789 * We were not called from zfs_receive_package(). Get 3790 * the fs specified by 'zfs send'. 3791 */ 3792 char *cp; 3793 (void) strlcpy(nonpackage_sendfs, 3794 drr.drr_u.drr_begin.drr_toname, 3795 sizeof (nonpackage_sendfs)); 3796 if ((cp = strchr(nonpackage_sendfs, '@')) != NULL) 3797 *cp = '\0'; 3798 sendfs = nonpackage_sendfs; 3799 VERIFY(finalsnap == NULL); 3800 } 3801 return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags, 3802 &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs, 3803 cleanup_fd, action_handlep, finalsnap)); 3804 } else { 3805 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 3806 DMU_COMPOUNDSTREAM); 3807 return (zfs_receive_package(hdl, infd, tosnap, flags, &drr, 3808 &zcksum, top_zfs, cleanup_fd, action_handlep)); 3809 } 3810} 3811 3812/* 3813 * Restores a backup of tosnap from the file descriptor specified by infd. 3814 * Return 0 on total success, -2 if some things couldn't be 3815 * destroyed/renamed/promoted, -1 if some things couldn't be received. 3816 * (-1 will override -2, if -1 and the resumable flag was specified the 3817 * transfer can be resumed if the sending side supports it). 3818 */ 3819int 3820zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props, 3821 recvflags_t *flags, int infd, avl_tree_t *stream_avl) 3822{ 3823 char *top_zfs = NULL; 3824 int err; 3825 int cleanup_fd; 3826 uint64_t action_handle = 0; 3827 char *originsnap = NULL; 3828 if (props) { 3829 err = nvlist_lookup_string(props, "origin", &originsnap); 3830 if (err && err != ENOENT) 3831 return (err); 3832 } 3833 3834 cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); 3835 VERIFY(cleanup_fd >= 0); 3836 3837 err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL, 3838 stream_avl, &top_zfs, cleanup_fd, &action_handle, NULL); 3839 3840 VERIFY(0 == close(cleanup_fd)); 3841 3842 if (err == 0 && !flags->nomount && top_zfs) { 3843 zfs_handle_t *zhp; 3844 prop_changelist_t *clp; 3845 3846 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM); 3847 if (zhp != NULL) { 3848 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT, 3849 CL_GATHER_MOUNT_ALWAYS, 0); 3850 zfs_close(zhp); 3851 if (clp != NULL) { 3852 /* mount and share received datasets */ 3853 err = changelist_postfix(clp); 3854 changelist_free(clp); 3855 } 3856 } 3857 if (zhp == NULL || clp == NULL || err) 3858 err = -1; 3859 } 3860 if (top_zfs) 3861 free(top_zfs); 3862 3863 return (err); 3864} 3865