libzfs_sendrecv.c revision 307050
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} progress_arg_t; 89 90typedef struct dataref { 91 uint64_t ref_guid; 92 uint64_t ref_object; 93 uint64_t ref_offset; 94} dataref_t; 95 96typedef struct dedup_entry { 97 struct dedup_entry *dde_next; 98 zio_cksum_t dde_chksum; 99 uint64_t dde_prop; 100 dataref_t dde_ref; 101} dedup_entry_t; 102 103#define MAX_DDT_PHYSMEM_PERCENT 20 104#define SMALLEST_POSSIBLE_MAX_DDT_MB 128 105 106typedef struct dedup_table { 107 dedup_entry_t **dedup_hash_array; 108 umem_cache_t *ddecache; 109 uint64_t max_ddt_size; /* max dedup table size in bytes */ 110 uint64_t cur_ddt_size; /* current dedup table size in bytes */ 111 uint64_t ddt_count; 112 int numhashbits; 113 boolean_t ddt_full; 114} dedup_table_t; 115 116static int 117high_order_bit(uint64_t n) 118{ 119 int count; 120 121 for (count = 0; n != 0; count++) 122 n >>= 1; 123 return (count); 124} 125 126static size_t 127ssread(void *buf, size_t len, FILE *stream) 128{ 129 size_t outlen; 130 131 if ((outlen = fread(buf, len, 1, stream)) == 0) 132 return (0); 133 134 return (outlen); 135} 136 137static void 138ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp, 139 zio_cksum_t *cs, uint64_t prop, dataref_t *dr) 140{ 141 dedup_entry_t *dde; 142 143 if (ddt->cur_ddt_size >= ddt->max_ddt_size) { 144 if (ddt->ddt_full == B_FALSE) { 145 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 146 "Dedup table full. Deduplication will continue " 147 "with existing table entries")); 148 ddt->ddt_full = B_TRUE; 149 } 150 return; 151 } 152 153 if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT)) 154 != NULL) { 155 assert(*ddepp == NULL); 156 dde->dde_next = NULL; 157 dde->dde_chksum = *cs; 158 dde->dde_prop = prop; 159 dde->dde_ref = *dr; 160 *ddepp = dde; 161 ddt->cur_ddt_size += sizeof (dedup_entry_t); 162 ddt->ddt_count++; 163 } 164} 165 166/* 167 * Using the specified dedup table, do a lookup for an entry with 168 * the checksum cs. If found, return the block's reference info 169 * in *dr. Otherwise, insert a new entry in the dedup table, using 170 * the reference information specified by *dr. 171 * 172 * return value: true - entry was found 173 * false - entry was not found 174 */ 175static boolean_t 176ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs, 177 uint64_t prop, dataref_t *dr) 178{ 179 uint32_t hashcode; 180 dedup_entry_t **ddepp; 181 182 hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits); 183 184 for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL; 185 ddepp = &((*ddepp)->dde_next)) { 186 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) && 187 (*ddepp)->dde_prop == prop) { 188 *dr = (*ddepp)->dde_ref; 189 return (B_TRUE); 190 } 191 } 192 ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr); 193 return (B_FALSE); 194} 195 196static int 197dump_record(dmu_replay_record_t *drr, void *payload, int payload_len, 198 zio_cksum_t *zc, int outfd) 199{ 200 ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 201 ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 202 fletcher_4_incremental_native(drr, 203 offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc); 204 if (drr->drr_type != DRR_BEGIN) { 205 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u. 206 drr_checksum.drr_checksum)); 207 drr->drr_u.drr_checksum.drr_checksum = *zc; 208 } 209 fletcher_4_incremental_native(&drr->drr_u.drr_checksum.drr_checksum, 210 sizeof (zio_cksum_t), zc); 211 if (write(outfd, drr, sizeof (*drr)) == -1) 212 return (errno); 213 if (payload_len != 0) { 214 fletcher_4_incremental_native(payload, payload_len, zc); 215 if (write(outfd, payload, payload_len) == -1) 216 return (errno); 217 } 218 return (0); 219} 220 221/* 222 * This function is started in a separate thread when the dedup option 223 * has been requested. The main send thread determines the list of 224 * snapshots to be included in the send stream and makes the ioctl calls 225 * for each one. But instead of having the ioctl send the output to the 226 * the output fd specified by the caller of zfs_send()), the 227 * ioctl is told to direct the output to a pipe, which is read by the 228 * alternate thread running THIS function. This function does the 229 * dedup'ing by: 230 * 1. building a dedup table (the DDT) 231 * 2. doing checksums on each data block and inserting a record in the DDT 232 * 3. looking for matching checksums, and 233 * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever 234 * a duplicate block is found. 235 * The output of this function then goes to the output fd requested 236 * by the caller of zfs_send(). 237 */ 238static void * 239cksummer(void *arg) 240{ 241 dedup_arg_t *dda = arg; 242 char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE); 243 dmu_replay_record_t thedrr; 244 dmu_replay_record_t *drr = &thedrr; 245 FILE *ofp; 246 int outfd; 247 dedup_table_t ddt; 248 zio_cksum_t stream_cksum; 249 uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE); 250 uint64_t numbuckets; 251 252 ddt.max_ddt_size = 253 MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100, 254 SMALLEST_POSSIBLE_MAX_DDT_MB << 20); 255 256 numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t)); 257 258 /* 259 * numbuckets must be a power of 2. Increase number to 260 * a power of 2 if necessary. 261 */ 262 if (!ISP2(numbuckets)) 263 numbuckets = 1 << high_order_bit(numbuckets); 264 265 ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *)); 266 ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0, 267 NULL, NULL, NULL, NULL, NULL, 0); 268 ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *); 269 ddt.numhashbits = high_order_bit(numbuckets) - 1; 270 ddt.ddt_full = B_FALSE; 271 272 outfd = dda->outputfd; 273 ofp = fdopen(dda->inputfd, "r"); 274 while (ssread(drr, sizeof (*drr), ofp) != 0) { 275 276 switch (drr->drr_type) { 277 case DRR_BEGIN: 278 { 279 struct drr_begin *drrb = &drr->drr_u.drr_begin; 280 int fflags; 281 int sz = 0; 282 ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0); 283 284 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 285 286 /* set the DEDUP feature flag for this stream */ 287 fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 288 fflags |= (DMU_BACKUP_FEATURE_DEDUP | 289 DMU_BACKUP_FEATURE_DEDUPPROPS); 290 DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags); 291 292 if (drr->drr_payloadlen != 0) { 293 sz = drr->drr_payloadlen; 294 295 if (sz > SPA_MAXBLOCKSIZE) { 296 buf = zfs_realloc(dda->dedup_hdl, buf, 297 SPA_MAXBLOCKSIZE, sz); 298 } 299 (void) ssread(buf, sz, ofp); 300 if (ferror(stdin)) 301 perror("fread"); 302 } 303 if (dump_record(drr, buf, sz, &stream_cksum, 304 outfd) != 0) 305 goto out; 306 break; 307 } 308 309 case DRR_END: 310 { 311 struct drr_end *drre = &drr->drr_u.drr_end; 312 /* use the recalculated checksum */ 313 drre->drr_checksum = stream_cksum; 314 if (dump_record(drr, NULL, 0, &stream_cksum, 315 outfd) != 0) 316 goto out; 317 break; 318 } 319 320 case DRR_OBJECT: 321 { 322 struct drr_object *drro = &drr->drr_u.drr_object; 323 if (drro->drr_bonuslen > 0) { 324 (void) ssread(buf, 325 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), 326 ofp); 327 } 328 if (dump_record(drr, buf, 329 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), 330 &stream_cksum, outfd) != 0) 331 goto out; 332 break; 333 } 334 335 case DRR_SPILL: 336 { 337 struct drr_spill *drrs = &drr->drr_u.drr_spill; 338 (void) ssread(buf, drrs->drr_length, ofp); 339 if (dump_record(drr, buf, drrs->drr_length, 340 &stream_cksum, outfd) != 0) 341 goto out; 342 break; 343 } 344 345 case DRR_FREEOBJECTS: 346 { 347 if (dump_record(drr, NULL, 0, &stream_cksum, 348 outfd) != 0) 349 goto out; 350 break; 351 } 352 353 case DRR_WRITE: 354 { 355 struct drr_write *drrw = &drr->drr_u.drr_write; 356 dataref_t dataref; 357 358 (void) ssread(buf, drrw->drr_length, ofp); 359 360 /* 361 * Use the existing checksum if it's dedup-capable, 362 * else calculate a SHA256 checksum for it. 363 */ 364 365 if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum, 366 zero_cksum) || 367 !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) { 368 SHA256_CTX ctx; 369 zio_cksum_t tmpsha256; 370 371 SHA256Init(&ctx); 372 SHA256Update(&ctx, buf, drrw->drr_length); 373 SHA256Final(&tmpsha256, &ctx); 374 drrw->drr_key.ddk_cksum.zc_word[0] = 375 BE_64(tmpsha256.zc_word[0]); 376 drrw->drr_key.ddk_cksum.zc_word[1] = 377 BE_64(tmpsha256.zc_word[1]); 378 drrw->drr_key.ddk_cksum.zc_word[2] = 379 BE_64(tmpsha256.zc_word[2]); 380 drrw->drr_key.ddk_cksum.zc_word[3] = 381 BE_64(tmpsha256.zc_word[3]); 382 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256; 383 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP; 384 } 385 386 dataref.ref_guid = drrw->drr_toguid; 387 dataref.ref_object = drrw->drr_object; 388 dataref.ref_offset = drrw->drr_offset; 389 390 if (ddt_update(dda->dedup_hdl, &ddt, 391 &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop, 392 &dataref)) { 393 dmu_replay_record_t wbr_drr = {0}; 394 struct drr_write_byref *wbr_drrr = 395 &wbr_drr.drr_u.drr_write_byref; 396 397 /* block already present in stream */ 398 wbr_drr.drr_type = DRR_WRITE_BYREF; 399 400 wbr_drrr->drr_object = drrw->drr_object; 401 wbr_drrr->drr_offset = drrw->drr_offset; 402 wbr_drrr->drr_length = drrw->drr_length; 403 wbr_drrr->drr_toguid = drrw->drr_toguid; 404 wbr_drrr->drr_refguid = dataref.ref_guid; 405 wbr_drrr->drr_refobject = 406 dataref.ref_object; 407 wbr_drrr->drr_refoffset = 408 dataref.ref_offset; 409 410 wbr_drrr->drr_checksumtype = 411 drrw->drr_checksumtype; 412 wbr_drrr->drr_checksumflags = 413 drrw->drr_checksumtype; 414 wbr_drrr->drr_key.ddk_cksum = 415 drrw->drr_key.ddk_cksum; 416 wbr_drrr->drr_key.ddk_prop = 417 drrw->drr_key.ddk_prop; 418 419 if (dump_record(&wbr_drr, NULL, 0, 420 &stream_cksum, outfd) != 0) 421 goto out; 422 } else { 423 /* block not previously seen */ 424 if (dump_record(drr, buf, drrw->drr_length, 425 &stream_cksum, outfd) != 0) 426 goto out; 427 } 428 break; 429 } 430 431 case DRR_WRITE_EMBEDDED: 432 { 433 struct drr_write_embedded *drrwe = 434 &drr->drr_u.drr_write_embedded; 435 (void) ssread(buf, 436 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp); 437 if (dump_record(drr, buf, 438 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), 439 &stream_cksum, outfd) != 0) 440 goto out; 441 break; 442 } 443 444 case DRR_FREE: 445 { 446 if (dump_record(drr, NULL, 0, &stream_cksum, 447 outfd) != 0) 448 goto out; 449 break; 450 } 451 452 default: 453 (void) fprintf(stderr, "INVALID record type 0x%x\n", 454 drr->drr_type); 455 /* should never happen, so assert */ 456 assert(B_FALSE); 457 } 458 } 459out: 460 umem_cache_destroy(ddt.ddecache); 461 free(ddt.dedup_hash_array); 462 free(buf); 463 (void) fclose(ofp); 464 465 return (NULL); 466} 467 468/* 469 * Routines for dealing with the AVL tree of fs-nvlists 470 */ 471typedef struct fsavl_node { 472 avl_node_t fn_node; 473 nvlist_t *fn_nvfs; 474 char *fn_snapname; 475 uint64_t fn_guid; 476} fsavl_node_t; 477 478static int 479fsavl_compare(const void *arg1, const void *arg2) 480{ 481 const fsavl_node_t *fn1 = arg1; 482 const fsavl_node_t *fn2 = arg2; 483 484 if (fn1->fn_guid > fn2->fn_guid) 485 return (+1); 486 else if (fn1->fn_guid < fn2->fn_guid) 487 return (-1); 488 else 489 return (0); 490} 491 492/* 493 * Given the GUID of a snapshot, find its containing filesystem and 494 * (optionally) name. 495 */ 496static nvlist_t * 497fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname) 498{ 499 fsavl_node_t fn_find; 500 fsavl_node_t *fn; 501 502 fn_find.fn_guid = snapguid; 503 504 fn = avl_find(avl, &fn_find, NULL); 505 if (fn) { 506 if (snapname) 507 *snapname = fn->fn_snapname; 508 return (fn->fn_nvfs); 509 } 510 return (NULL); 511} 512 513static void 514fsavl_destroy(avl_tree_t *avl) 515{ 516 fsavl_node_t *fn; 517 void *cookie; 518 519 if (avl == NULL) 520 return; 521 522 cookie = NULL; 523 while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL) 524 free(fn); 525 avl_destroy(avl); 526 free(avl); 527} 528 529/* 530 * Given an nvlist, produce an avl tree of snapshots, ordered by guid 531 */ 532static avl_tree_t * 533fsavl_create(nvlist_t *fss) 534{ 535 avl_tree_t *fsavl; 536 nvpair_t *fselem = NULL; 537 538 if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL) 539 return (NULL); 540 541 avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t), 542 offsetof(fsavl_node_t, fn_node)); 543 544 while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) { 545 nvlist_t *nvfs, *snaps; 546 nvpair_t *snapelem = NULL; 547 548 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); 549 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); 550 551 while ((snapelem = 552 nvlist_next_nvpair(snaps, snapelem)) != NULL) { 553 fsavl_node_t *fn; 554 uint64_t guid; 555 556 VERIFY(0 == nvpair_value_uint64(snapelem, &guid)); 557 if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) { 558 fsavl_destroy(fsavl); 559 return (NULL); 560 } 561 fn->fn_nvfs = nvfs; 562 fn->fn_snapname = nvpair_name(snapelem); 563 fn->fn_guid = guid; 564 565 /* 566 * Note: if there are multiple snaps with the 567 * same GUID, we ignore all but one. 568 */ 569 if (avl_find(fsavl, fn, NULL) == NULL) 570 avl_add(fsavl, fn); 571 else 572 free(fn); 573 } 574 } 575 576 return (fsavl); 577} 578 579/* 580 * Routines for dealing with the giant nvlist of fs-nvlists, etc. 581 */ 582typedef struct send_data { 583 uint64_t parent_fromsnap_guid; 584 nvlist_t *parent_snaps; 585 nvlist_t *fss; 586 nvlist_t *snapprops; 587 const char *fromsnap; 588 const char *tosnap; 589 boolean_t recursive; 590 591 /* 592 * The header nvlist is of the following format: 593 * { 594 * "tosnap" -> string 595 * "fromsnap" -> string (if incremental) 596 * "fss" -> { 597 * id -> { 598 * 599 * "name" -> string (full name; for debugging) 600 * "parentfromsnap" -> number (guid of fromsnap in parent) 601 * 602 * "props" -> { name -> value (only if set here) } 603 * "snaps" -> { name (lastname) -> number (guid) } 604 * "snapprops" -> { name (lastname) -> { name -> value } } 605 * 606 * "origin" -> number (guid) (if clone) 607 * "sent" -> boolean (not on-disk) 608 * } 609 * } 610 * } 611 * 612 */ 613} send_data_t; 614 615static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv); 616 617static int 618send_iterate_snap(zfs_handle_t *zhp, void *arg) 619{ 620 send_data_t *sd = arg; 621 uint64_t guid = zhp->zfs_dmustats.dds_guid; 622 char *snapname; 623 nvlist_t *nv; 624 625 snapname = strrchr(zhp->zfs_name, '@')+1; 626 627 VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid)); 628 /* 629 * NB: if there is no fromsnap here (it's a newly created fs in 630 * an incremental replication), we will substitute the tosnap. 631 */ 632 if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) || 633 (sd->parent_fromsnap_guid == 0 && sd->tosnap && 634 strcmp(snapname, sd->tosnap) == 0)) { 635 sd->parent_fromsnap_guid = guid; 636 } 637 638 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); 639 send_iterate_prop(zhp, nv); 640 VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv)); 641 nvlist_free(nv); 642 643 zfs_close(zhp); 644 return (0); 645} 646 647static void 648send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv) 649{ 650 nvpair_t *elem = NULL; 651 652 while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) { 653 char *propname = nvpair_name(elem); 654 zfs_prop_t prop = zfs_name_to_prop(propname); 655 nvlist_t *propnv; 656 657 if (!zfs_prop_user(propname)) { 658 /* 659 * Realistically, this should never happen. However, 660 * we want the ability to add DSL properties without 661 * needing to make incompatible version changes. We 662 * need to ignore unknown properties to allow older 663 * software to still send datasets containing these 664 * properties, with the unknown properties elided. 665 */ 666 if (prop == ZPROP_INVAL) 667 continue; 668 669 if (zfs_prop_readonly(prop)) 670 continue; 671 } 672 673 verify(nvpair_value_nvlist(elem, &propnv) == 0); 674 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION || 675 prop == ZFS_PROP_REFQUOTA || 676 prop == ZFS_PROP_REFRESERVATION) { 677 char *source; 678 uint64_t value; 679 verify(nvlist_lookup_uint64(propnv, 680 ZPROP_VALUE, &value) == 0); 681 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) 682 continue; 683 /* 684 * May have no source before SPA_VERSION_RECVD_PROPS, 685 * but is still modifiable. 686 */ 687 if (nvlist_lookup_string(propnv, 688 ZPROP_SOURCE, &source) == 0) { 689 if ((strcmp(source, zhp->zfs_name) != 0) && 690 (strcmp(source, 691 ZPROP_SOURCE_VAL_RECVD) != 0)) 692 continue; 693 } 694 } else { 695 char *source; 696 if (nvlist_lookup_string(propnv, 697 ZPROP_SOURCE, &source) != 0) 698 continue; 699 if ((strcmp(source, zhp->zfs_name) != 0) && 700 (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0)) 701 continue; 702 } 703 704 if (zfs_prop_user(propname) || 705 zfs_prop_get_type(prop) == PROP_TYPE_STRING) { 706 char *value; 707 verify(nvlist_lookup_string(propnv, 708 ZPROP_VALUE, &value) == 0); 709 VERIFY(0 == nvlist_add_string(nv, propname, value)); 710 } else { 711 uint64_t value; 712 verify(nvlist_lookup_uint64(propnv, 713 ZPROP_VALUE, &value) == 0); 714 VERIFY(0 == nvlist_add_uint64(nv, propname, value)); 715 } 716 } 717} 718 719/* 720 * recursively generate nvlists describing datasets. See comment 721 * for the data structure send_data_t above for description of contents 722 * of the nvlist. 723 */ 724static int 725send_iterate_fs(zfs_handle_t *zhp, void *arg) 726{ 727 send_data_t *sd = arg; 728 nvlist_t *nvfs, *nv; 729 int rv = 0; 730 uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid; 731 uint64_t guid = zhp->zfs_dmustats.dds_guid; 732 char guidstring[64]; 733 734 VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0)); 735 VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name)); 736 VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap", 737 sd->parent_fromsnap_guid)); 738 739 if (zhp->zfs_dmustats.dds_origin[0]) { 740 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl, 741 zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT); 742 if (origin == NULL) 743 return (-1); 744 VERIFY(0 == nvlist_add_uint64(nvfs, "origin", 745 origin->zfs_dmustats.dds_guid)); 746 } 747 748 /* iterate over props */ 749 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); 750 send_iterate_prop(zhp, nv); 751 VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv)); 752 nvlist_free(nv); 753 754 /* iterate over snaps, and set sd->parent_fromsnap_guid */ 755 sd->parent_fromsnap_guid = 0; 756 VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0)); 757 VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0)); 758 (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd); 759 VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps)); 760 VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops)); 761 nvlist_free(sd->parent_snaps); 762 nvlist_free(sd->snapprops); 763 764 /* add this fs to nvlist */ 765 (void) snprintf(guidstring, sizeof (guidstring), 766 "0x%llx", (longlong_t)guid); 767 VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs)); 768 nvlist_free(nvfs); 769 770 /* iterate over children */ 771 if (sd->recursive) 772 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd); 773 774 sd->parent_fromsnap_guid = parent_fromsnap_guid_save; 775 776 zfs_close(zhp); 777 return (rv); 778} 779 780static int 781gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap, 782 const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp) 783{ 784 zfs_handle_t *zhp; 785 send_data_t sd = { 0 }; 786 int error; 787 788 zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 789 if (zhp == NULL) 790 return (EZFS_BADTYPE); 791 792 VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0)); 793 sd.fromsnap = fromsnap; 794 sd.tosnap = tosnap; 795 sd.recursive = recursive; 796 797 if ((error = send_iterate_fs(zhp, &sd)) != 0) { 798 nvlist_free(sd.fss); 799 if (avlp != NULL) 800 *avlp = NULL; 801 *nvlp = NULL; 802 return (error); 803 } 804 805 if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) { 806 nvlist_free(sd.fss); 807 *nvlp = NULL; 808 return (EZFS_NOMEM); 809 } 810 811 *nvlp = sd.fss; 812 return (0); 813} 814 815/* 816 * Routines specific to "zfs send" 817 */ 818typedef struct send_dump_data { 819 /* these are all just the short snapname (the part after the @) */ 820 const char *fromsnap; 821 const char *tosnap; 822 char prevsnap[ZFS_MAXNAMELEN]; 823 uint64_t prevsnap_obj; 824 boolean_t seenfrom, seento, replicate, doall, fromorigin; 825 boolean_t verbose, dryrun, parsable, progress, embed_data, std_out; 826 boolean_t large_block; 827 int outfd; 828 boolean_t err; 829 nvlist_t *fss; 830 nvlist_t *snapholds; 831 avl_tree_t *fsavl; 832 snapfilter_cb_t *filter_cb; 833 void *filter_cb_arg; 834 nvlist_t *debugnv; 835 char holdtag[ZFS_MAXNAMELEN]; 836 int cleanup_fd; 837 uint64_t size; 838} send_dump_data_t; 839 840static int 841estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj, 842 boolean_t fromorigin, uint64_t *sizep) 843{ 844 zfs_cmd_t zc = { 0 }; 845 libzfs_handle_t *hdl = zhp->zfs_hdl; 846 847 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 848 assert(fromsnap_obj == 0 || !fromorigin); 849 850 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 851 zc.zc_obj = fromorigin; 852 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 853 zc.zc_fromobj = fromsnap_obj; 854 zc.zc_guid = 1; /* estimate flag */ 855 856 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { 857 char errbuf[1024]; 858 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 859 "warning: cannot estimate space for '%s'"), zhp->zfs_name); 860 861 switch (errno) { 862 case EXDEV: 863 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 864 "not an earlier snapshot from the same fs")); 865 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 866 867 case ENOENT: 868 if (zfs_dataset_exists(hdl, zc.zc_name, 869 ZFS_TYPE_SNAPSHOT)) { 870 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 871 "incremental source (@%s) does not exist"), 872 zc.zc_value); 873 } 874 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 875 876 case EDQUOT: 877 case EFBIG: 878 case EIO: 879 case ENOLINK: 880 case ENOSPC: 881 case ENXIO: 882 case EPIPE: 883 case ERANGE: 884 case EFAULT: 885 case EROFS: 886 zfs_error_aux(hdl, strerror(errno)); 887 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 888 889 default: 890 return (zfs_standard_error(hdl, errno, errbuf)); 891 } 892 } 893 894 *sizep = zc.zc_objset_type; 895 896 return (0); 897} 898 899/* 900 * Dumps a backup of the given snapshot (incremental from fromsnap if it's not 901 * NULL) to the file descriptor specified by outfd. 902 */ 903static int 904dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj, 905 boolean_t fromorigin, int outfd, enum lzc_send_flags flags, 906 nvlist_t *debugnv) 907{ 908 zfs_cmd_t zc = { 0 }; 909 libzfs_handle_t *hdl = zhp->zfs_hdl; 910 nvlist_t *thisdbg; 911 912 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 913 assert(fromsnap_obj == 0 || !fromorigin); 914 915 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 916 zc.zc_cookie = outfd; 917 zc.zc_obj = fromorigin; 918 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 919 zc.zc_fromobj = fromsnap_obj; 920 zc.zc_flags = flags; 921 922 VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0)); 923 if (fromsnap && fromsnap[0] != '\0') { 924 VERIFY(0 == nvlist_add_string(thisdbg, 925 "fromsnap", fromsnap)); 926 } 927 928 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { 929 char errbuf[1024]; 930 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 931 "warning: cannot send '%s'"), zhp->zfs_name); 932 933 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno)); 934 if (debugnv) { 935 VERIFY(0 == nvlist_add_nvlist(debugnv, 936 zhp->zfs_name, thisdbg)); 937 } 938 nvlist_free(thisdbg); 939 940 switch (errno) { 941 case EXDEV: 942 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 943 "not an earlier snapshot from the same fs")); 944 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 945 946 case ENOENT: 947 if (zfs_dataset_exists(hdl, zc.zc_name, 948 ZFS_TYPE_SNAPSHOT)) { 949 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 950 "incremental source (@%s) does not exist"), 951 zc.zc_value); 952 } 953 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 954 955 case EDQUOT: 956 case EFBIG: 957 case EIO: 958 case ENOLINK: 959 case ENOSPC: 960#ifdef illumos 961 case ENOSTR: 962#endif 963 case ENXIO: 964 case EPIPE: 965 case ERANGE: 966 case EFAULT: 967 case EROFS: 968 zfs_error_aux(hdl, strerror(errno)); 969 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 970 971 default: 972 return (zfs_standard_error(hdl, errno, errbuf)); 973 } 974 } 975 976 if (debugnv) 977 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg)); 978 nvlist_free(thisdbg); 979 980 return (0); 981} 982 983static void 984gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd) 985{ 986 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 987 988 /* 989 * zfs_send() only sets snapholds for sends that need them, 990 * e.g. replication and doall. 991 */ 992 if (sdd->snapholds == NULL) 993 return; 994 995 fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag); 996} 997 998static void * 999send_progress_thread(void *arg) 1000{ 1001 progress_arg_t *pa = arg; 1002 zfs_cmd_t zc = { 0 }; 1003 zfs_handle_t *zhp = pa->pa_zhp; 1004 libzfs_handle_t *hdl = zhp->zfs_hdl; 1005 unsigned long long bytes; 1006 char buf[16]; 1007 time_t t; 1008 struct tm *tm; 1009 1010 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 1011 1012 if (!pa->pa_parsable) 1013 (void) fprintf(stderr, "TIME SENT SNAPSHOT\n"); 1014 1015 /* 1016 * Print the progress from ZFS_IOC_SEND_PROGRESS every second. 1017 */ 1018 for (;;) { 1019 (void) sleep(1); 1020 1021 zc.zc_cookie = pa->pa_fd; 1022 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0) 1023 return ((void *)-1); 1024 1025 (void) time(&t); 1026 tm = localtime(&t); 1027 bytes = zc.zc_cookie; 1028 1029 if (pa->pa_parsable) { 1030 (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n", 1031 tm->tm_hour, tm->tm_min, tm->tm_sec, 1032 bytes, zhp->zfs_name); 1033 } else { 1034 zfs_nicenum(bytes, buf, sizeof (buf)); 1035 (void) fprintf(stderr, "%02d:%02d:%02d %5s %s\n", 1036 tm->tm_hour, tm->tm_min, tm->tm_sec, 1037 buf, zhp->zfs_name); 1038 } 1039 } 1040} 1041 1042static void 1043send_print_verbose(FILE *fout, const char *tosnap, const char *fromsnap, 1044 uint64_t size, boolean_t parsable) 1045{ 1046 if (parsable) { 1047 if (fromsnap != NULL) { 1048 (void) fprintf(fout, "incremental\t%s\t%s", 1049 fromsnap, tosnap); 1050 } else { 1051 (void) fprintf(fout, "full\t%s", 1052 tosnap); 1053 } 1054 } else { 1055 if (fromsnap != NULL) { 1056 if (strchr(fromsnap, '@') == NULL && 1057 strchr(fromsnap, '#') == NULL) { 1058 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1059 "send from @%s to %s"), 1060 fromsnap, tosnap); 1061 } else { 1062 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1063 "send from %s to %s"), 1064 fromsnap, tosnap); 1065 } 1066 } else { 1067 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1068 "full send of %s"), 1069 tosnap); 1070 } 1071 } 1072 1073 if (size != 0) { 1074 if (parsable) { 1075 (void) fprintf(fout, "\t%llu", 1076 (longlong_t)size); 1077 } else { 1078 char buf[16]; 1079 zfs_nicenum(size, buf, sizeof (buf)); 1080 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1081 " estimated size is %s"), buf); 1082 } 1083 } 1084 (void) fprintf(fout, "\n"); 1085} 1086 1087static int 1088dump_snapshot(zfs_handle_t *zhp, void *arg) 1089{ 1090 send_dump_data_t *sdd = arg; 1091 progress_arg_t pa = { 0 }; 1092 pthread_t tid; 1093 char *thissnap; 1094 int err; 1095 boolean_t isfromsnap, istosnap, fromorigin; 1096 boolean_t exclude = B_FALSE; 1097 FILE *fout = sdd->std_out ? stdout : stderr; 1098 1099 err = 0; 1100 thissnap = strchr(zhp->zfs_name, '@') + 1; 1101 isfromsnap = (sdd->fromsnap != NULL && 1102 strcmp(sdd->fromsnap, thissnap) == 0); 1103 1104 if (!sdd->seenfrom && isfromsnap) { 1105 gather_holds(zhp, sdd); 1106 sdd->seenfrom = B_TRUE; 1107 (void) strcpy(sdd->prevsnap, thissnap); 1108 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 1109 zfs_close(zhp); 1110 return (0); 1111 } 1112 1113 if (sdd->seento || !sdd->seenfrom) { 1114 zfs_close(zhp); 1115 return (0); 1116 } 1117 1118 istosnap = (strcmp(sdd->tosnap, thissnap) == 0); 1119 if (istosnap) 1120 sdd->seento = B_TRUE; 1121 1122 if (!sdd->doall && !isfromsnap && !istosnap) { 1123 if (sdd->replicate) { 1124 char *snapname; 1125 nvlist_t *snapprops; 1126 /* 1127 * Filter out all intermediate snapshots except origin 1128 * snapshots needed to replicate clones. 1129 */ 1130 nvlist_t *nvfs = fsavl_find(sdd->fsavl, 1131 zhp->zfs_dmustats.dds_guid, &snapname); 1132 1133 VERIFY(0 == nvlist_lookup_nvlist(nvfs, 1134 "snapprops", &snapprops)); 1135 VERIFY(0 == nvlist_lookup_nvlist(snapprops, 1136 thissnap, &snapprops)); 1137 exclude = !nvlist_exists(snapprops, "is_clone_origin"); 1138 } else { 1139 exclude = B_TRUE; 1140 } 1141 } 1142 1143 /* 1144 * If a filter function exists, call it to determine whether 1145 * this snapshot will be sent. 1146 */ 1147 if (exclude || (sdd->filter_cb != NULL && 1148 sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) { 1149 /* 1150 * This snapshot is filtered out. Don't send it, and don't 1151 * set prevsnap_obj, so it will be as if this snapshot didn't 1152 * exist, and the next accepted snapshot will be sent as 1153 * an incremental from the last accepted one, or as the 1154 * first (and full) snapshot in the case of a replication, 1155 * non-incremental send. 1156 */ 1157 zfs_close(zhp); 1158 return (0); 1159 } 1160 1161 gather_holds(zhp, sdd); 1162 fromorigin = sdd->prevsnap[0] == '\0' && 1163 (sdd->fromorigin || sdd->replicate); 1164 1165 if (sdd->verbose) { 1166 uint64_t size = 0; 1167 (void) estimate_ioctl(zhp, sdd->prevsnap_obj, 1168 fromorigin, &size); 1169 1170 send_print_verbose(fout, zhp->zfs_name, 1171 sdd->prevsnap[0] ? sdd->prevsnap : NULL, 1172 size, sdd->parsable); 1173 sdd->size += size; 1174 } 1175 1176 if (!sdd->dryrun) { 1177 /* 1178 * If progress reporting is requested, spawn a new thread to 1179 * poll ZFS_IOC_SEND_PROGRESS at a regular interval. 1180 */ 1181 if (sdd->progress) { 1182 pa.pa_zhp = zhp; 1183 pa.pa_fd = sdd->outfd; 1184 pa.pa_parsable = sdd->parsable; 1185 1186 if ((err = pthread_create(&tid, NULL, 1187 send_progress_thread, &pa)) != 0) { 1188 zfs_close(zhp); 1189 return (err); 1190 } 1191 } 1192 1193 enum lzc_send_flags flags = 0; 1194 if (sdd->large_block) 1195 flags |= LZC_SEND_FLAG_LARGE_BLOCK; 1196 if (sdd->embed_data) 1197 flags |= LZC_SEND_FLAG_EMBED_DATA; 1198 1199 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj, 1200 fromorigin, sdd->outfd, flags, sdd->debugnv); 1201 1202 if (sdd->progress) { 1203 (void) pthread_cancel(tid); 1204 (void) pthread_join(tid, NULL); 1205 } 1206 } 1207 1208 (void) strcpy(sdd->prevsnap, thissnap); 1209 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 1210 zfs_close(zhp); 1211 return (err); 1212} 1213 1214static int 1215dump_filesystem(zfs_handle_t *zhp, void *arg) 1216{ 1217 int rv = 0; 1218 send_dump_data_t *sdd = arg; 1219 boolean_t missingfrom = B_FALSE; 1220 zfs_cmd_t zc = { 0 }; 1221 1222 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", 1223 zhp->zfs_name, sdd->tosnap); 1224 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) { 1225 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1226 "WARNING: could not send %s@%s: does not exist\n"), 1227 zhp->zfs_name, sdd->tosnap); 1228 sdd->err = B_TRUE; 1229 return (0); 1230 } 1231 1232 if (sdd->replicate && sdd->fromsnap) { 1233 /* 1234 * If this fs does not have fromsnap, and we're doing 1235 * recursive, we need to send a full stream from the 1236 * beginning (or an incremental from the origin if this 1237 * is a clone). If we're doing non-recursive, then let 1238 * them get the error. 1239 */ 1240 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", 1241 zhp->zfs_name, sdd->fromsnap); 1242 if (ioctl(zhp->zfs_hdl->libzfs_fd, 1243 ZFS_IOC_OBJSET_STATS, &zc) != 0) { 1244 missingfrom = B_TRUE; 1245 } 1246 } 1247 1248 sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0; 1249 sdd->prevsnap_obj = 0; 1250 if (sdd->fromsnap == NULL || missingfrom) 1251 sdd->seenfrom = B_TRUE; 1252 1253 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg); 1254 if (!sdd->seenfrom) { 1255 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1256 "WARNING: could not send %s@%s:\n" 1257 "incremental source (%s@%s) does not exist\n"), 1258 zhp->zfs_name, sdd->tosnap, 1259 zhp->zfs_name, sdd->fromsnap); 1260 sdd->err = B_TRUE; 1261 } else if (!sdd->seento) { 1262 if (sdd->fromsnap) { 1263 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1264 "WARNING: could not send %s@%s:\n" 1265 "incremental source (%s@%s) " 1266 "is not earlier than it\n"), 1267 zhp->zfs_name, sdd->tosnap, 1268 zhp->zfs_name, sdd->fromsnap); 1269 } else { 1270 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1271 "WARNING: " 1272 "could not send %s@%s: does not exist\n"), 1273 zhp->zfs_name, sdd->tosnap); 1274 } 1275 sdd->err = B_TRUE; 1276 } 1277 1278 return (rv); 1279} 1280 1281static int 1282dump_filesystems(zfs_handle_t *rzhp, void *arg) 1283{ 1284 send_dump_data_t *sdd = arg; 1285 nvpair_t *fspair; 1286 boolean_t needagain, progress; 1287 1288 if (!sdd->replicate) 1289 return (dump_filesystem(rzhp, sdd)); 1290 1291 /* Mark the clone origin snapshots. */ 1292 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1293 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1294 nvlist_t *nvfs; 1295 uint64_t origin_guid = 0; 1296 1297 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs)); 1298 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid); 1299 if (origin_guid != 0) { 1300 char *snapname; 1301 nvlist_t *origin_nv = fsavl_find(sdd->fsavl, 1302 origin_guid, &snapname); 1303 if (origin_nv != NULL) { 1304 nvlist_t *snapprops; 1305 VERIFY(0 == nvlist_lookup_nvlist(origin_nv, 1306 "snapprops", &snapprops)); 1307 VERIFY(0 == nvlist_lookup_nvlist(snapprops, 1308 snapname, &snapprops)); 1309 VERIFY(0 == nvlist_add_boolean( 1310 snapprops, "is_clone_origin")); 1311 } 1312 } 1313 } 1314again: 1315 needagain = progress = B_FALSE; 1316 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1317 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1318 nvlist_t *fslist, *parent_nv; 1319 char *fsname; 1320 zfs_handle_t *zhp; 1321 int err; 1322 uint64_t origin_guid = 0; 1323 uint64_t parent_guid = 0; 1324 1325 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); 1326 if (nvlist_lookup_boolean(fslist, "sent") == 0) 1327 continue; 1328 1329 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0); 1330 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid); 1331 (void) nvlist_lookup_uint64(fslist, "parentfromsnap", 1332 &parent_guid); 1333 1334 if (parent_guid != 0) { 1335 parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL); 1336 if (!nvlist_exists(parent_nv, "sent")) { 1337 /* parent has not been sent; skip this one */ 1338 needagain = B_TRUE; 1339 continue; 1340 } 1341 } 1342 1343 if (origin_guid != 0) { 1344 nvlist_t *origin_nv = fsavl_find(sdd->fsavl, 1345 origin_guid, NULL); 1346 if (origin_nv != NULL && 1347 !nvlist_exists(origin_nv, "sent")) { 1348 /* 1349 * origin has not been sent yet; 1350 * skip this clone. 1351 */ 1352 needagain = B_TRUE; 1353 continue; 1354 } 1355 } 1356 1357 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET); 1358 if (zhp == NULL) 1359 return (-1); 1360 err = dump_filesystem(zhp, sdd); 1361 VERIFY(nvlist_add_boolean(fslist, "sent") == 0); 1362 progress = B_TRUE; 1363 zfs_close(zhp); 1364 if (err) 1365 return (err); 1366 } 1367 if (needagain) { 1368 assert(progress); 1369 goto again; 1370 } 1371 1372 /* clean out the sent flags in case we reuse this fss */ 1373 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1374 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1375 nvlist_t *fslist; 1376 1377 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); 1378 (void) nvlist_remove_all(fslist, "sent"); 1379 } 1380 1381 return (0); 1382} 1383 1384nvlist_t * 1385zfs_send_resume_token_to_nvlist(libzfs_handle_t *hdl, const char *token) 1386{ 1387 unsigned int version; 1388 int nread; 1389 unsigned long long checksum, packed_len; 1390 1391 /* 1392 * Decode token header, which is: 1393 * <token version>-<checksum of payload>-<uncompressed payload length> 1394 * Note that the only supported token version is 1. 1395 */ 1396 nread = sscanf(token, "%u-%llx-%llx-", 1397 &version, &checksum, &packed_len); 1398 if (nread != 3) { 1399 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1400 "resume token is corrupt (invalid format)")); 1401 return (NULL); 1402 } 1403 1404 if (version != ZFS_SEND_RESUME_TOKEN_VERSION) { 1405 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1406 "resume token is corrupt (invalid version %u)"), 1407 version); 1408 return (NULL); 1409 } 1410 1411 /* convert hexadecimal representation to binary */ 1412 token = strrchr(token, '-') + 1; 1413 int len = strlen(token) / 2; 1414 unsigned char *compressed = zfs_alloc(hdl, len); 1415 for (int i = 0; i < len; i++) { 1416 nread = sscanf(token + i * 2, "%2hhx", compressed + i); 1417 if (nread != 1) { 1418 free(compressed); 1419 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1420 "resume token is corrupt " 1421 "(payload is not hex-encoded)")); 1422 return (NULL); 1423 } 1424 } 1425 1426 /* verify checksum */ 1427 zio_cksum_t cksum; 1428 fletcher_4_native(compressed, len, NULL, &cksum); 1429 if (cksum.zc_word[0] != checksum) { 1430 free(compressed); 1431 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1432 "resume token is corrupt (incorrect checksum)")); 1433 return (NULL); 1434 } 1435 1436 /* uncompress */ 1437 void *packed = zfs_alloc(hdl, packed_len); 1438 uLongf packed_len_long = packed_len; 1439 if (uncompress(packed, &packed_len_long, compressed, len) != Z_OK || 1440 packed_len_long != packed_len) { 1441 free(packed); 1442 free(compressed); 1443 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1444 "resume token is corrupt (decompression failed)")); 1445 return (NULL); 1446 } 1447 1448 /* unpack nvlist */ 1449 nvlist_t *nv; 1450 int error = nvlist_unpack(packed, packed_len, &nv, KM_SLEEP); 1451 free(packed); 1452 free(compressed); 1453 if (error != 0) { 1454 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1455 "resume token is corrupt (nvlist_unpack failed)")); 1456 return (NULL); 1457 } 1458 return (nv); 1459} 1460 1461int 1462zfs_send_resume(libzfs_handle_t *hdl, sendflags_t *flags, int outfd, 1463 const char *resume_token) 1464{ 1465 char errbuf[1024]; 1466 char *toname; 1467 char *fromname = NULL; 1468 uint64_t resumeobj, resumeoff, toguid, fromguid, bytes; 1469 zfs_handle_t *zhp; 1470 int error = 0; 1471 char name[ZFS_MAXNAMELEN]; 1472 enum lzc_send_flags lzc_flags = 0; 1473 1474 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1475 "cannot resume send")); 1476 1477 nvlist_t *resume_nvl = 1478 zfs_send_resume_token_to_nvlist(hdl, resume_token); 1479 if (resume_nvl == NULL) { 1480 /* 1481 * zfs_error_aux has already been set by 1482 * zfs_send_resume_token_to_nvlist 1483 */ 1484 return (zfs_error(hdl, EZFS_FAULT, errbuf)); 1485 } 1486 if (flags->verbose) { 1487 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1488 "resume token contents:\n")); 1489 nvlist_print(stderr, resume_nvl); 1490 } 1491 1492 if (nvlist_lookup_string(resume_nvl, "toname", &toname) != 0 || 1493 nvlist_lookup_uint64(resume_nvl, "object", &resumeobj) != 0 || 1494 nvlist_lookup_uint64(resume_nvl, "offset", &resumeoff) != 0 || 1495 nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 || 1496 nvlist_lookup_uint64(resume_nvl, "toguid", &toguid) != 0) { 1497 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1498 "resume token is corrupt")); 1499 return (zfs_error(hdl, EZFS_FAULT, errbuf)); 1500 } 1501 fromguid = 0; 1502 (void) nvlist_lookup_uint64(resume_nvl, "fromguid", &fromguid); 1503 1504 if (flags->embed_data || nvlist_exists(resume_nvl, "embedok")) 1505 lzc_flags |= LZC_SEND_FLAG_EMBED_DATA; 1506 1507 if (guid_to_name(hdl, toname, toguid, B_FALSE, name) != 0) { 1508 if (zfs_dataset_exists(hdl, toname, ZFS_TYPE_DATASET)) { 1509 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1510 "'%s' is no longer the same snapshot used in " 1511 "the initial send"), toname); 1512 } else { 1513 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1514 "'%s' used in the initial send no longer exists"), 1515 toname); 1516 } 1517 return (zfs_error(hdl, EZFS_BADPATH, errbuf)); 1518 } 1519 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 1520 if (zhp == NULL) { 1521 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1522 "unable to access '%s'"), name); 1523 return (zfs_error(hdl, EZFS_BADPATH, errbuf)); 1524 } 1525 1526 if (fromguid != 0) { 1527 if (guid_to_name(hdl, toname, fromguid, B_TRUE, name) != 0) { 1528 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1529 "incremental source %#llx no longer exists"), 1530 (longlong_t)fromguid); 1531 return (zfs_error(hdl, EZFS_BADPATH, errbuf)); 1532 } 1533 fromname = name; 1534 } 1535 1536 if (flags->verbose) { 1537 uint64_t size = 0; 1538 error = lzc_send_space(zhp->zfs_name, fromname, &size); 1539 if (error == 0) 1540 size = MAX(0, (int64_t)(size - bytes)); 1541 send_print_verbose(stderr, zhp->zfs_name, fromname, 1542 size, flags->parsable); 1543 } 1544 1545 if (!flags->dryrun) { 1546 progress_arg_t pa = { 0 }; 1547 pthread_t tid; 1548 /* 1549 * If progress reporting is requested, spawn a new thread to 1550 * poll ZFS_IOC_SEND_PROGRESS at a regular interval. 1551 */ 1552 if (flags->progress) { 1553 pa.pa_zhp = zhp; 1554 pa.pa_fd = outfd; 1555 pa.pa_parsable = flags->parsable; 1556 1557 error = pthread_create(&tid, NULL, 1558 send_progress_thread, &pa); 1559 if (error != 0) { 1560 zfs_close(zhp); 1561 return (error); 1562 } 1563 } 1564 1565 error = lzc_send_resume(zhp->zfs_name, fromname, outfd, 1566 lzc_flags, resumeobj, resumeoff); 1567 1568 if (flags->progress) { 1569 (void) pthread_cancel(tid); 1570 (void) pthread_join(tid, NULL); 1571 } 1572 1573 char errbuf[1024]; 1574 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1575 "warning: cannot send '%s'"), zhp->zfs_name); 1576 1577 zfs_close(zhp); 1578 1579 switch (error) { 1580 case 0: 1581 return (0); 1582 case EXDEV: 1583 case ENOENT: 1584 case EDQUOT: 1585 case EFBIG: 1586 case EIO: 1587 case ENOLINK: 1588 case ENOSPC: 1589#ifdef illumos 1590 case ENOSTR: 1591#endif 1592 case ENXIO: 1593 case EPIPE: 1594 case ERANGE: 1595 case EFAULT: 1596 case EROFS: 1597 zfs_error_aux(hdl, strerror(errno)); 1598 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 1599 1600 default: 1601 return (zfs_standard_error(hdl, errno, errbuf)); 1602 } 1603 } 1604 1605 1606 zfs_close(zhp); 1607 1608 return (error); 1609} 1610 1611/* 1612 * Generate a send stream for the dataset identified by the argument zhp. 1613 * 1614 * The content of the send stream is the snapshot identified by 1615 * 'tosnap'. Incremental streams are requested in two ways: 1616 * - from the snapshot identified by "fromsnap" (if non-null) or 1617 * - from the origin of the dataset identified by zhp, which must 1618 * be a clone. In this case, "fromsnap" is null and "fromorigin" 1619 * is TRUE. 1620 * 1621 * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and 1622 * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM) 1623 * if "replicate" is set. If "doall" is set, dump all the intermediate 1624 * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall" 1625 * case too. If "props" is set, send properties. 1626 */ 1627int 1628zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, 1629 sendflags_t *flags, int outfd, snapfilter_cb_t filter_func, 1630 void *cb_arg, nvlist_t **debugnvp) 1631{ 1632 char errbuf[1024]; 1633 send_dump_data_t sdd = { 0 }; 1634 int err = 0; 1635 nvlist_t *fss = NULL; 1636 avl_tree_t *fsavl = NULL; 1637 static uint64_t holdseq; 1638 int spa_version; 1639 pthread_t tid = 0; 1640 int pipefd[2]; 1641 dedup_arg_t dda = { 0 }; 1642 int featureflags = 0; 1643 FILE *fout; 1644 1645 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1646 "cannot send '%s'"), zhp->zfs_name); 1647 1648 if (fromsnap && fromsnap[0] == '\0') { 1649 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN, 1650 "zero-length incremental source")); 1651 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf)); 1652 } 1653 1654 if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) { 1655 uint64_t version; 1656 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION); 1657 if (version >= ZPL_VERSION_SA) { 1658 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; 1659 } 1660 } 1661 1662 if (flags->dedup && !flags->dryrun) { 1663 featureflags |= (DMU_BACKUP_FEATURE_DEDUP | 1664 DMU_BACKUP_FEATURE_DEDUPPROPS); 1665 if ((err = pipe(pipefd)) != 0) { 1666 zfs_error_aux(zhp->zfs_hdl, strerror(errno)); 1667 return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED, 1668 errbuf)); 1669 } 1670 dda.outputfd = outfd; 1671 dda.inputfd = pipefd[1]; 1672 dda.dedup_hdl = zhp->zfs_hdl; 1673 if ((err = pthread_create(&tid, NULL, cksummer, &dda)) != 0) { 1674 (void) close(pipefd[0]); 1675 (void) close(pipefd[1]); 1676 zfs_error_aux(zhp->zfs_hdl, strerror(errno)); 1677 return (zfs_error(zhp->zfs_hdl, 1678 EZFS_THREADCREATEFAILED, errbuf)); 1679 } 1680 } 1681 1682 if (flags->replicate || flags->doall || flags->props) { 1683 dmu_replay_record_t drr = { 0 }; 1684 char *packbuf = NULL; 1685 size_t buflen = 0; 1686 zio_cksum_t zc = { 0 }; 1687 1688 if (flags->replicate || flags->props) { 1689 nvlist_t *hdrnv; 1690 1691 VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0)); 1692 if (fromsnap) { 1693 VERIFY(0 == nvlist_add_string(hdrnv, 1694 "fromsnap", fromsnap)); 1695 } 1696 VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap)); 1697 if (!flags->replicate) { 1698 VERIFY(0 == nvlist_add_boolean(hdrnv, 1699 "not_recursive")); 1700 } 1701 1702 err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name, 1703 fromsnap, tosnap, flags->replicate, &fss, &fsavl); 1704 if (err) 1705 goto err_out; 1706 VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss)); 1707 err = nvlist_pack(hdrnv, &packbuf, &buflen, 1708 NV_ENCODE_XDR, 0); 1709 if (debugnvp) 1710 *debugnvp = hdrnv; 1711 else 1712 nvlist_free(hdrnv); 1713 if (err) 1714 goto stderr_out; 1715 } 1716 1717 if (!flags->dryrun) { 1718 /* write first begin record */ 1719 drr.drr_type = DRR_BEGIN; 1720 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 1721 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin. 1722 drr_versioninfo, DMU_COMPOUNDSTREAM); 1723 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin. 1724 drr_versioninfo, featureflags); 1725 (void) snprintf(drr.drr_u.drr_begin.drr_toname, 1726 sizeof (drr.drr_u.drr_begin.drr_toname), 1727 "%s@%s", zhp->zfs_name, tosnap); 1728 drr.drr_payloadlen = buflen; 1729 1730 err = dump_record(&drr, packbuf, buflen, &zc, outfd); 1731 free(packbuf); 1732 if (err != 0) 1733 goto stderr_out; 1734 1735 /* write end record */ 1736 bzero(&drr, sizeof (drr)); 1737 drr.drr_type = DRR_END; 1738 drr.drr_u.drr_end.drr_checksum = zc; 1739 err = write(outfd, &drr, sizeof (drr)); 1740 if (err == -1) { 1741 err = errno; 1742 goto stderr_out; 1743 } 1744 1745 err = 0; 1746 } 1747 } 1748 1749 /* dump each stream */ 1750 sdd.fromsnap = fromsnap; 1751 sdd.tosnap = tosnap; 1752 if (tid != 0) 1753 sdd.outfd = pipefd[0]; 1754 else 1755 sdd.outfd = outfd; 1756 sdd.replicate = flags->replicate; 1757 sdd.doall = flags->doall; 1758 sdd.fromorigin = flags->fromorigin; 1759 sdd.fss = fss; 1760 sdd.fsavl = fsavl; 1761 sdd.verbose = flags->verbose; 1762 sdd.parsable = flags->parsable; 1763 sdd.progress = flags->progress; 1764 sdd.dryrun = flags->dryrun; 1765 sdd.large_block = flags->largeblock; 1766 sdd.embed_data = flags->embed_data; 1767 sdd.filter_cb = filter_func; 1768 sdd.filter_cb_arg = cb_arg; 1769 if (debugnvp) 1770 sdd.debugnv = *debugnvp; 1771 if (sdd.verbose && sdd.dryrun) 1772 sdd.std_out = B_TRUE; 1773 fout = sdd.std_out ? stdout : stderr; 1774 1775 /* 1776 * Some flags require that we place user holds on the datasets that are 1777 * being sent so they don't get destroyed during the send. We can skip 1778 * this step if the pool is imported read-only since the datasets cannot 1779 * be destroyed. 1780 */ 1781 if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp), 1782 ZPOOL_PROP_READONLY, NULL) && 1783 zfs_spa_version(zhp, &spa_version) == 0 && 1784 spa_version >= SPA_VERSION_USERREFS && 1785 (flags->doall || flags->replicate)) { 1786 ++holdseq; 1787 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag), 1788 ".send-%d-%llu", getpid(), (u_longlong_t)holdseq); 1789 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); 1790 if (sdd.cleanup_fd < 0) { 1791 err = errno; 1792 goto stderr_out; 1793 } 1794 sdd.snapholds = fnvlist_alloc(); 1795 } else { 1796 sdd.cleanup_fd = -1; 1797 sdd.snapholds = NULL; 1798 } 1799 if (flags->verbose || sdd.snapholds != NULL) { 1800 /* 1801 * Do a verbose no-op dry run to get all the verbose output 1802 * or to gather snapshot hold's before generating any data, 1803 * then do a non-verbose real run to generate the streams. 1804 */ 1805 sdd.dryrun = B_TRUE; 1806 err = dump_filesystems(zhp, &sdd); 1807 1808 if (err != 0) 1809 goto stderr_out; 1810 1811 if (flags->verbose) { 1812 if (flags->parsable) { 1813 (void) fprintf(fout, "size\t%llu\n", 1814 (longlong_t)sdd.size); 1815 } else { 1816 char buf[16]; 1817 zfs_nicenum(sdd.size, buf, sizeof (buf)); 1818 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1819 "total estimated size is %s\n"), buf); 1820 } 1821 } 1822 1823 /* Ensure no snaps found is treated as an error. */ 1824 if (!sdd.seento) { 1825 err = ENOENT; 1826 goto err_out; 1827 } 1828 1829 /* Skip the second run if dryrun was requested. */ 1830 if (flags->dryrun) 1831 goto err_out; 1832 1833 if (sdd.snapholds != NULL) { 1834 err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds); 1835 if (err != 0) 1836 goto stderr_out; 1837 1838 fnvlist_free(sdd.snapholds); 1839 sdd.snapholds = NULL; 1840 } 1841 1842 sdd.dryrun = B_FALSE; 1843 sdd.verbose = B_FALSE; 1844 } 1845 1846 err = dump_filesystems(zhp, &sdd); 1847 fsavl_destroy(fsavl); 1848 nvlist_free(fss); 1849 1850 /* Ensure no snaps found is treated as an error. */ 1851 if (err == 0 && !sdd.seento) 1852 err = ENOENT; 1853 1854 if (tid != 0) { 1855 if (err != 0) 1856 (void) pthread_cancel(tid); 1857 (void) close(pipefd[0]); 1858 (void) pthread_join(tid, NULL); 1859 } 1860 1861 if (sdd.cleanup_fd != -1) { 1862 VERIFY(0 == close(sdd.cleanup_fd)); 1863 sdd.cleanup_fd = -1; 1864 } 1865 1866 if (!flags->dryrun && (flags->replicate || flags->doall || 1867 flags->props)) { 1868 /* 1869 * write final end record. NB: want to do this even if 1870 * there was some error, because it might not be totally 1871 * failed. 1872 */ 1873 dmu_replay_record_t drr = { 0 }; 1874 drr.drr_type = DRR_END; 1875 if (write(outfd, &drr, sizeof (drr)) == -1) { 1876 return (zfs_standard_error(zhp->zfs_hdl, 1877 errno, errbuf)); 1878 } 1879 } 1880 1881 return (err || sdd.err); 1882 1883stderr_out: 1884 err = zfs_standard_error(zhp->zfs_hdl, err, errbuf); 1885err_out: 1886 fsavl_destroy(fsavl); 1887 nvlist_free(fss); 1888 fnvlist_free(sdd.snapholds); 1889 1890 if (sdd.cleanup_fd != -1) 1891 VERIFY(0 == close(sdd.cleanup_fd)); 1892 if (tid != 0) { 1893 (void) pthread_cancel(tid); 1894 (void) close(pipefd[0]); 1895 (void) pthread_join(tid, NULL); 1896 } 1897 return (err); 1898} 1899 1900int 1901zfs_send_one(zfs_handle_t *zhp, const char *from, int fd, 1902 enum lzc_send_flags flags) 1903{ 1904 int err; 1905 libzfs_handle_t *hdl = zhp->zfs_hdl; 1906 1907 char errbuf[1024]; 1908 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1909 "warning: cannot send '%s'"), zhp->zfs_name); 1910 1911 err = lzc_send(zhp->zfs_name, from, fd, flags); 1912 if (err != 0) { 1913 switch (errno) { 1914 case EXDEV: 1915 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1916 "not an earlier snapshot from the same fs")); 1917 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 1918 1919 case ENOENT: 1920 case ESRCH: 1921 if (lzc_exists(zhp->zfs_name)) { 1922 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1923 "incremental source (%s) does not exist"), 1924 from); 1925 } 1926 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 1927 1928 case EBUSY: 1929 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1930 "target is busy; if a filesystem, " 1931 "it must not be mounted")); 1932 return (zfs_error(hdl, EZFS_BUSY, errbuf)); 1933 1934 case EDQUOT: 1935 case EFBIG: 1936 case EIO: 1937 case ENOLINK: 1938 case ENOSPC: 1939#ifdef illumos 1940 case ENOSTR: 1941#endif 1942 case ENXIO: 1943 case EPIPE: 1944 case ERANGE: 1945 case EFAULT: 1946 case EROFS: 1947 zfs_error_aux(hdl, strerror(errno)); 1948 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 1949 1950 default: 1951 return (zfs_standard_error(hdl, errno, errbuf)); 1952 } 1953 } 1954 return (err != 0); 1955} 1956 1957/* 1958 * Routines specific to "zfs recv" 1959 */ 1960 1961static int 1962recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen, 1963 boolean_t byteswap, zio_cksum_t *zc) 1964{ 1965 char *cp = buf; 1966 int rv; 1967 int len = ilen; 1968 1969 assert(ilen <= SPA_MAXBLOCKSIZE); 1970 1971 do { 1972 rv = read(fd, cp, len); 1973 cp += rv; 1974 len -= rv; 1975 } while (rv > 0); 1976 1977 if (rv < 0 || len != 0) { 1978 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1979 "failed to read from stream")); 1980 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN, 1981 "cannot receive"))); 1982 } 1983 1984 if (zc) { 1985 if (byteswap) 1986 fletcher_4_incremental_byteswap(buf, ilen, zc); 1987 else 1988 fletcher_4_incremental_native(buf, ilen, zc); 1989 } 1990 return (0); 1991} 1992 1993static int 1994recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp, 1995 boolean_t byteswap, zio_cksum_t *zc) 1996{ 1997 char *buf; 1998 int err; 1999 2000 buf = zfs_alloc(hdl, len); 2001 if (buf == NULL) 2002 return (ENOMEM); 2003 2004 err = recv_read(hdl, fd, buf, len, byteswap, zc); 2005 if (err != 0) { 2006 free(buf); 2007 return (err); 2008 } 2009 2010 err = nvlist_unpack(buf, len, nvp, 0); 2011 free(buf); 2012 if (err != 0) { 2013 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 2014 "stream (malformed nvlist)")); 2015 return (EINVAL); 2016 } 2017 return (0); 2018} 2019 2020static int 2021recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname, 2022 int baselen, char *newname, recvflags_t *flags) 2023{ 2024 static int seq; 2025 zfs_cmd_t zc = { 0 }; 2026 int err; 2027 prop_changelist_t *clp; 2028 zfs_handle_t *zhp; 2029 2030 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 2031 if (zhp == NULL) 2032 return (-1); 2033 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 2034 flags->force ? MS_FORCE : 0); 2035 zfs_close(zhp); 2036 if (clp == NULL) 2037 return (-1); 2038 err = changelist_prefix(clp); 2039 if (err) 2040 return (err); 2041 2042 zc.zc_objset_type = DMU_OST_ZFS; 2043 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 2044 2045 if (tryname) { 2046 (void) strcpy(newname, tryname); 2047 2048 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value)); 2049 2050 if (flags->verbose) { 2051 (void) printf("attempting rename %s to %s\n", 2052 zc.zc_name, zc.zc_value); 2053 } 2054 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc); 2055 if (err == 0) 2056 changelist_rename(clp, name, tryname); 2057 } else { 2058 err = ENOENT; 2059 } 2060 2061 if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) { 2062 seq++; 2063 2064 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u", 2065 baselen, name, getpid(), seq); 2066 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value)); 2067 2068 if (flags->verbose) { 2069 (void) printf("failed - trying rename %s to %s\n", 2070 zc.zc_name, zc.zc_value); 2071 } 2072 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc); 2073 if (err == 0) 2074 changelist_rename(clp, name, newname); 2075 if (err && flags->verbose) { 2076 (void) printf("failed (%u) - " 2077 "will try again on next pass\n", errno); 2078 } 2079 err = EAGAIN; 2080 } else if (flags->verbose) { 2081 if (err == 0) 2082 (void) printf("success\n"); 2083 else 2084 (void) printf("failed (%u)\n", errno); 2085 } 2086 2087 (void) changelist_postfix(clp); 2088 changelist_free(clp); 2089 2090 return (err); 2091} 2092 2093static int 2094recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen, 2095 char *newname, recvflags_t *flags) 2096{ 2097 zfs_cmd_t zc = { 0 }; 2098 int err = 0; 2099 prop_changelist_t *clp; 2100 zfs_handle_t *zhp; 2101 boolean_t defer = B_FALSE; 2102 int spa_version; 2103 2104 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 2105 if (zhp == NULL) 2106 return (-1); 2107 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 2108 flags->force ? MS_FORCE : 0); 2109 if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT && 2110 zfs_spa_version(zhp, &spa_version) == 0 && 2111 spa_version >= SPA_VERSION_USERREFS) 2112 defer = B_TRUE; 2113 zfs_close(zhp); 2114 if (clp == NULL) 2115 return (-1); 2116 err = changelist_prefix(clp); 2117 if (err) 2118 return (err); 2119 2120 zc.zc_objset_type = DMU_OST_ZFS; 2121 zc.zc_defer_destroy = defer; 2122 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 2123 2124 if (flags->verbose) 2125 (void) printf("attempting destroy %s\n", zc.zc_name); 2126 err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc); 2127 if (err == 0) { 2128 if (flags->verbose) 2129 (void) printf("success\n"); 2130 changelist_remove(clp, zc.zc_name); 2131 } 2132 2133 (void) changelist_postfix(clp); 2134 changelist_free(clp); 2135 2136 /* 2137 * Deferred destroy might destroy the snapshot or only mark it to be 2138 * destroyed later, and it returns success in either case. 2139 */ 2140 if (err != 0 || (defer && zfs_dataset_exists(hdl, name, 2141 ZFS_TYPE_SNAPSHOT))) { 2142 err = recv_rename(hdl, name, NULL, baselen, newname, flags); 2143 } 2144 2145 return (err); 2146} 2147 2148typedef struct guid_to_name_data { 2149 uint64_t guid; 2150 boolean_t bookmark_ok; 2151 char *name; 2152 char *skip; 2153} guid_to_name_data_t; 2154 2155static int 2156guid_to_name_cb(zfs_handle_t *zhp, void *arg) 2157{ 2158 guid_to_name_data_t *gtnd = arg; 2159 const char *slash; 2160 int err; 2161 2162 if (gtnd->skip != NULL && 2163 (slash = strrchr(zhp->zfs_name, '/')) != NULL && 2164 strcmp(slash + 1, gtnd->skip) == 0) { 2165 zfs_close(zhp); 2166 return (0); 2167 } 2168 2169 if (zfs_prop_get_int(zhp, ZFS_PROP_GUID) == gtnd->guid) { 2170 (void) strcpy(gtnd->name, zhp->zfs_name); 2171 zfs_close(zhp); 2172 return (EEXIST); 2173 } 2174 2175 err = zfs_iter_children(zhp, guid_to_name_cb, gtnd); 2176 if (err != EEXIST && gtnd->bookmark_ok) 2177 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, gtnd); 2178 zfs_close(zhp); 2179 return (err); 2180} 2181 2182/* 2183 * Attempt to find the local dataset associated with this guid. In the case of 2184 * multiple matches, we attempt to find the "best" match by searching 2185 * progressively larger portions of the hierarchy. This allows one to send a 2186 * tree of datasets individually and guarantee that we will find the source 2187 * guid within that hierarchy, even if there are multiple matches elsewhere. 2188 */ 2189static int 2190guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid, 2191 boolean_t bookmark_ok, char *name) 2192{ 2193 char pname[ZFS_MAXNAMELEN]; 2194 guid_to_name_data_t gtnd; 2195 2196 gtnd.guid = guid; 2197 gtnd.bookmark_ok = bookmark_ok; 2198 gtnd.name = name; 2199 gtnd.skip = NULL; 2200 2201 /* 2202 * Search progressively larger portions of the hierarchy, starting 2203 * with the filesystem specified by 'parent'. This will 2204 * select the "most local" version of the origin snapshot in the case 2205 * that there are multiple matching snapshots in the system. 2206 */ 2207 (void) strlcpy(pname, parent, sizeof (pname)); 2208 char *cp = strrchr(pname, '@'); 2209 if (cp == NULL) 2210 cp = strchr(pname, '\0'); 2211 for (; cp != NULL; cp = strrchr(pname, '/')) { 2212 /* Chop off the last component and open the parent */ 2213 *cp = '\0'; 2214 zfs_handle_t *zhp = make_dataset_handle(hdl, pname); 2215 2216 if (zhp == NULL) 2217 continue; 2218 int err = guid_to_name_cb(zfs_handle_dup(zhp), >nd); 2219 if (err != EEXIST) 2220 err = zfs_iter_children(zhp, guid_to_name_cb, >nd); 2221 if (err != EEXIST && bookmark_ok) 2222 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, >nd); 2223 zfs_close(zhp); 2224 if (err == EEXIST) 2225 return (0); 2226 2227 /* 2228 * Remember the last portion of the dataset so we skip it next 2229 * time through (as we've already searched that portion of the 2230 * hierarchy). 2231 */ 2232 gtnd.skip = strrchr(pname, '/') + 1; 2233 } 2234 2235 return (ENOENT); 2236} 2237 2238/* 2239 * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if 2240 * guid1 is after guid2. 2241 */ 2242static int 2243created_before(libzfs_handle_t *hdl, avl_tree_t *avl, 2244 uint64_t guid1, uint64_t guid2) 2245{ 2246 nvlist_t *nvfs; 2247 char *fsname, *snapname; 2248 char buf[ZFS_MAXNAMELEN]; 2249 int rv; 2250 zfs_handle_t *guid1hdl, *guid2hdl; 2251 uint64_t create1, create2; 2252 2253 if (guid2 == 0) 2254 return (0); 2255 if (guid1 == 0) 2256 return (1); 2257 2258 nvfs = fsavl_find(avl, guid1, &snapname); 2259 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2260 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); 2261 guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); 2262 if (guid1hdl == NULL) 2263 return (-1); 2264 2265 nvfs = fsavl_find(avl, guid2, &snapname); 2266 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2267 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); 2268 guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); 2269 if (guid2hdl == NULL) { 2270 zfs_close(guid1hdl); 2271 return (-1); 2272 } 2273 2274 create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG); 2275 create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG); 2276 2277 if (create1 < create2) 2278 rv = -1; 2279 else if (create1 > create2) 2280 rv = +1; 2281 else 2282 rv = 0; 2283 2284 zfs_close(guid1hdl); 2285 zfs_close(guid2hdl); 2286 2287 return (rv); 2288} 2289 2290static int 2291recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs, 2292 recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl, 2293 nvlist_t *renamed) 2294{ 2295 nvlist_t *local_nv, *deleted = NULL; 2296 avl_tree_t *local_avl; 2297 nvpair_t *fselem, *nextfselem; 2298 char *fromsnap; 2299 char newname[ZFS_MAXNAMELEN]; 2300 char guidname[32]; 2301 int error; 2302 boolean_t needagain, progress, recursive; 2303 char *s1, *s2; 2304 2305 VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap)); 2306 2307 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2308 ENOENT); 2309 2310 if (flags->dryrun) 2311 return (0); 2312 2313again: 2314 needagain = progress = B_FALSE; 2315 2316 VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0)); 2317 2318 if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL, 2319 recursive, &local_nv, &local_avl)) != 0) 2320 return (error); 2321 2322 /* 2323 * Process deletes and renames 2324 */ 2325 for (fselem = nvlist_next_nvpair(local_nv, NULL); 2326 fselem; fselem = nextfselem) { 2327 nvlist_t *nvfs, *snaps; 2328 nvlist_t *stream_nvfs = NULL; 2329 nvpair_t *snapelem, *nextsnapelem; 2330 uint64_t fromguid = 0; 2331 uint64_t originguid = 0; 2332 uint64_t stream_originguid = 0; 2333 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid; 2334 char *fsname, *stream_fsname; 2335 2336 nextfselem = nvlist_next_nvpair(local_nv, fselem); 2337 2338 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); 2339 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); 2340 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2341 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap", 2342 &parent_fromsnap_guid)); 2343 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid); 2344 2345 /* 2346 * First find the stream's fs, so we can check for 2347 * a different origin (due to "zfs promote") 2348 */ 2349 for (snapelem = nvlist_next_nvpair(snaps, NULL); 2350 snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) { 2351 uint64_t thisguid; 2352 2353 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); 2354 stream_nvfs = fsavl_find(stream_avl, thisguid, NULL); 2355 2356 if (stream_nvfs != NULL) 2357 break; 2358 } 2359 2360 /* check for promote */ 2361 (void) nvlist_lookup_uint64(stream_nvfs, "origin", 2362 &stream_originguid); 2363 if (stream_nvfs && originguid != stream_originguid) { 2364 switch (created_before(hdl, local_avl, 2365 stream_originguid, originguid)) { 2366 case 1: { 2367 /* promote it! */ 2368 zfs_cmd_t zc = { 0 }; 2369 nvlist_t *origin_nvfs; 2370 char *origin_fsname; 2371 2372 if (flags->verbose) 2373 (void) printf("promoting %s\n", fsname); 2374 2375 origin_nvfs = fsavl_find(local_avl, originguid, 2376 NULL); 2377 VERIFY(0 == nvlist_lookup_string(origin_nvfs, 2378 "name", &origin_fsname)); 2379 (void) strlcpy(zc.zc_value, origin_fsname, 2380 sizeof (zc.zc_value)); 2381 (void) strlcpy(zc.zc_name, fsname, 2382 sizeof (zc.zc_name)); 2383 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc); 2384 if (error == 0) 2385 progress = B_TRUE; 2386 break; 2387 } 2388 default: 2389 break; 2390 case -1: 2391 fsavl_destroy(local_avl); 2392 nvlist_free(local_nv); 2393 return (-1); 2394 } 2395 /* 2396 * We had/have the wrong origin, therefore our 2397 * list of snapshots is wrong. Need to handle 2398 * them on the next pass. 2399 */ 2400 needagain = B_TRUE; 2401 continue; 2402 } 2403 2404 for (snapelem = nvlist_next_nvpair(snaps, NULL); 2405 snapelem; snapelem = nextsnapelem) { 2406 uint64_t thisguid; 2407 char *stream_snapname; 2408 nvlist_t *found, *props; 2409 2410 nextsnapelem = nvlist_next_nvpair(snaps, snapelem); 2411 2412 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); 2413 found = fsavl_find(stream_avl, thisguid, 2414 &stream_snapname); 2415 2416 /* check for delete */ 2417 if (found == NULL) { 2418 char name[ZFS_MAXNAMELEN]; 2419 2420 if (!flags->force) 2421 continue; 2422 2423 (void) snprintf(name, sizeof (name), "%s@%s", 2424 fsname, nvpair_name(snapelem)); 2425 2426 error = recv_destroy(hdl, name, 2427 strlen(fsname)+1, newname, flags); 2428 if (error) 2429 needagain = B_TRUE; 2430 else 2431 progress = B_TRUE; 2432 sprintf(guidname, "%lu", thisguid); 2433 nvlist_add_boolean(deleted, guidname); 2434 continue; 2435 } 2436 2437 stream_nvfs = found; 2438 2439 if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops", 2440 &props) && 0 == nvlist_lookup_nvlist(props, 2441 stream_snapname, &props)) { 2442 zfs_cmd_t zc = { 0 }; 2443 2444 zc.zc_cookie = B_TRUE; /* received */ 2445 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), 2446 "%s@%s", fsname, nvpair_name(snapelem)); 2447 if (zcmd_write_src_nvlist(hdl, &zc, 2448 props) == 0) { 2449 (void) zfs_ioctl(hdl, 2450 ZFS_IOC_SET_PROP, &zc); 2451 zcmd_free_nvlists(&zc); 2452 } 2453 } 2454 2455 /* check for different snapname */ 2456 if (strcmp(nvpair_name(snapelem), 2457 stream_snapname) != 0) { 2458 char name[ZFS_MAXNAMELEN]; 2459 char tryname[ZFS_MAXNAMELEN]; 2460 2461 (void) snprintf(name, sizeof (name), "%s@%s", 2462 fsname, nvpair_name(snapelem)); 2463 (void) snprintf(tryname, sizeof (name), "%s@%s", 2464 fsname, stream_snapname); 2465 2466 error = recv_rename(hdl, name, tryname, 2467 strlen(fsname)+1, newname, flags); 2468 if (error) 2469 needagain = B_TRUE; 2470 else 2471 progress = B_TRUE; 2472 } 2473 2474 if (strcmp(stream_snapname, fromsnap) == 0) 2475 fromguid = thisguid; 2476 } 2477 2478 /* check for delete */ 2479 if (stream_nvfs == NULL) { 2480 if (!flags->force) 2481 continue; 2482 2483 error = recv_destroy(hdl, fsname, strlen(tofs)+1, 2484 newname, flags); 2485 if (error) 2486 needagain = B_TRUE; 2487 else 2488 progress = B_TRUE; 2489 sprintf(guidname, "%lu", parent_fromsnap_guid); 2490 nvlist_add_boolean(deleted, guidname); 2491 continue; 2492 } 2493 2494 if (fromguid == 0) { 2495 if (flags->verbose) { 2496 (void) printf("local fs %s does not have " 2497 "fromsnap (%s in stream); must have " 2498 "been deleted locally; ignoring\n", 2499 fsname, fromsnap); 2500 } 2501 continue; 2502 } 2503 2504 VERIFY(0 == nvlist_lookup_string(stream_nvfs, 2505 "name", &stream_fsname)); 2506 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs, 2507 "parentfromsnap", &stream_parent_fromsnap_guid)); 2508 2509 s1 = strrchr(fsname, '/'); 2510 s2 = strrchr(stream_fsname, '/'); 2511 2512 /* 2513 * Check if we're going to rename based on parent guid change 2514 * and the current parent guid was also deleted. If it was then 2515 * rename will fail and is likely unneeded, so avoid this and 2516 * force an early retry to determine the new 2517 * parent_fromsnap_guid. 2518 */ 2519 if (stream_parent_fromsnap_guid != 0 && 2520 parent_fromsnap_guid != 0 && 2521 stream_parent_fromsnap_guid != parent_fromsnap_guid) { 2522 sprintf(guidname, "%lu", parent_fromsnap_guid); 2523 if (nvlist_exists(deleted, guidname)) { 2524 progress = B_TRUE; 2525 needagain = B_TRUE; 2526 goto doagain; 2527 } 2528 } 2529 2530 /* 2531 * Check for rename. If the exact receive path is specified, it 2532 * does not count as a rename, but we still need to check the 2533 * datasets beneath it. 2534 */ 2535 if ((stream_parent_fromsnap_guid != 0 && 2536 parent_fromsnap_guid != 0 && 2537 stream_parent_fromsnap_guid != parent_fromsnap_guid) || 2538 ((flags->isprefix || strcmp(tofs, fsname) != 0) && 2539 (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) { 2540 nvlist_t *parent; 2541 char tryname[ZFS_MAXNAMELEN]; 2542 2543 parent = fsavl_find(local_avl, 2544 stream_parent_fromsnap_guid, NULL); 2545 /* 2546 * NB: parent might not be found if we used the 2547 * tosnap for stream_parent_fromsnap_guid, 2548 * because the parent is a newly-created fs; 2549 * we'll be able to rename it after we recv the 2550 * new fs. 2551 */ 2552 if (parent != NULL) { 2553 char *pname; 2554 2555 VERIFY(0 == nvlist_lookup_string(parent, "name", 2556 &pname)); 2557 (void) snprintf(tryname, sizeof (tryname), 2558 "%s%s", pname, strrchr(stream_fsname, '/')); 2559 } else { 2560 tryname[0] = '\0'; 2561 if (flags->verbose) { 2562 (void) printf("local fs %s new parent " 2563 "not found\n", fsname); 2564 } 2565 } 2566 2567 newname[0] = '\0'; 2568 2569 error = recv_rename(hdl, fsname, tryname, 2570 strlen(tofs)+1, newname, flags); 2571 2572 if (renamed != NULL && newname[0] != '\0') { 2573 VERIFY(0 == nvlist_add_boolean(renamed, 2574 newname)); 2575 } 2576 2577 if (error) 2578 needagain = B_TRUE; 2579 else 2580 progress = B_TRUE; 2581 } 2582 } 2583 2584doagain: 2585 fsavl_destroy(local_avl); 2586 nvlist_free(local_nv); 2587 nvlist_free(deleted); 2588 2589 if (needagain && progress) { 2590 /* do another pass to fix up temporary names */ 2591 if (flags->verbose) 2592 (void) printf("another pass:\n"); 2593 goto again; 2594 } 2595 2596 return (needagain); 2597} 2598 2599static int 2600zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname, 2601 recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc, 2602 char **top_zfs, int cleanup_fd, uint64_t *action_handlep) 2603{ 2604 nvlist_t *stream_nv = NULL; 2605 avl_tree_t *stream_avl = NULL; 2606 char *fromsnap = NULL; 2607 char *sendsnap = NULL; 2608 char *cp; 2609 char tofs[ZFS_MAXNAMELEN]; 2610 char sendfs[ZFS_MAXNAMELEN]; 2611 char errbuf[1024]; 2612 dmu_replay_record_t drre; 2613 int error; 2614 boolean_t anyerr = B_FALSE; 2615 boolean_t softerr = B_FALSE; 2616 boolean_t recursive; 2617 2618 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2619 "cannot receive")); 2620 2621 assert(drr->drr_type == DRR_BEGIN); 2622 assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC); 2623 assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) == 2624 DMU_COMPOUNDSTREAM); 2625 2626 /* 2627 * Read in the nvlist from the stream. 2628 */ 2629 if (drr->drr_payloadlen != 0) { 2630 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen, 2631 &stream_nv, flags->byteswap, zc); 2632 if (error) { 2633 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2634 goto out; 2635 } 2636 } 2637 2638 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2639 ENOENT); 2640 2641 if (recursive && strchr(destname, '@')) { 2642 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2643 "cannot specify snapshot name for multi-snapshot stream")); 2644 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2645 goto out; 2646 } 2647 2648 /* 2649 * Read in the end record and verify checksum. 2650 */ 2651 if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre), 2652 flags->byteswap, NULL))) 2653 goto out; 2654 if (flags->byteswap) { 2655 drre.drr_type = BSWAP_32(drre.drr_type); 2656 drre.drr_u.drr_end.drr_checksum.zc_word[0] = 2657 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]); 2658 drre.drr_u.drr_end.drr_checksum.zc_word[1] = 2659 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]); 2660 drre.drr_u.drr_end.drr_checksum.zc_word[2] = 2661 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]); 2662 drre.drr_u.drr_end.drr_checksum.zc_word[3] = 2663 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]); 2664 } 2665 if (drre.drr_type != DRR_END) { 2666 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2667 goto out; 2668 } 2669 if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) { 2670 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2671 "incorrect header checksum")); 2672 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2673 goto out; 2674 } 2675 2676 (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap); 2677 2678 if (drr->drr_payloadlen != 0) { 2679 nvlist_t *stream_fss; 2680 2681 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss", 2682 &stream_fss)); 2683 if ((stream_avl = fsavl_create(stream_fss)) == NULL) { 2684 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2685 "couldn't allocate avl tree")); 2686 error = zfs_error(hdl, EZFS_NOMEM, errbuf); 2687 goto out; 2688 } 2689 2690 if (fromsnap != NULL) { 2691 nvlist_t *renamed = NULL; 2692 nvpair_t *pair = NULL; 2693 2694 (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN); 2695 if (flags->isprefix) { 2696 struct drr_begin *drrb = &drr->drr_u.drr_begin; 2697 int i; 2698 2699 if (flags->istail) { 2700 cp = strrchr(drrb->drr_toname, '/'); 2701 if (cp == NULL) { 2702 (void) strlcat(tofs, "/", 2703 ZFS_MAXNAMELEN); 2704 i = 0; 2705 } else { 2706 i = (cp - drrb->drr_toname); 2707 } 2708 } else { 2709 i = strcspn(drrb->drr_toname, "/@"); 2710 } 2711 /* zfs_receive_one() will create_parents() */ 2712 (void) strlcat(tofs, &drrb->drr_toname[i], 2713 ZFS_MAXNAMELEN); 2714 *strchr(tofs, '@') = '\0'; 2715 } 2716 2717 if (recursive && !flags->dryrun && !flags->nomount) { 2718 VERIFY(0 == nvlist_alloc(&renamed, 2719 NV_UNIQUE_NAME, 0)); 2720 } 2721 2722 softerr = recv_incremental_replication(hdl, tofs, flags, 2723 stream_nv, stream_avl, renamed); 2724 2725 /* Unmount renamed filesystems before receiving. */ 2726 while ((pair = nvlist_next_nvpair(renamed, 2727 pair)) != NULL) { 2728 zfs_handle_t *zhp; 2729 prop_changelist_t *clp = NULL; 2730 2731 zhp = zfs_open(hdl, nvpair_name(pair), 2732 ZFS_TYPE_FILESYSTEM); 2733 if (zhp != NULL) { 2734 clp = changelist_gather(zhp, 2735 ZFS_PROP_MOUNTPOINT, 0, 0); 2736 zfs_close(zhp); 2737 if (clp != NULL) { 2738 softerr |= 2739 changelist_prefix(clp); 2740 changelist_free(clp); 2741 } 2742 } 2743 } 2744 2745 nvlist_free(renamed); 2746 } 2747 } 2748 2749 /* 2750 * Get the fs specified by the first path in the stream (the top level 2751 * specified by 'zfs send') and pass it to each invocation of 2752 * zfs_receive_one(). 2753 */ 2754 (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname, 2755 ZFS_MAXNAMELEN); 2756 if ((cp = strchr(sendfs, '@')) != NULL) { 2757 *cp = '\0'; 2758 /* 2759 * Find the "sendsnap", the final snapshot in a replication 2760 * stream. zfs_receive_one() handles certain errors 2761 * differently, depending on if the contained stream is the 2762 * last one or not. 2763 */ 2764 sendsnap = (cp + 1); 2765 } 2766 2767 /* Finally, receive each contained stream */ 2768 do { 2769 /* 2770 * we should figure out if it has a recoverable 2771 * error, in which case do a recv_skip() and drive on. 2772 * Note, if we fail due to already having this guid, 2773 * zfs_receive_one() will take care of it (ie, 2774 * recv_skip() and return 0). 2775 */ 2776 error = zfs_receive_impl(hdl, destname, NULL, flags, fd, 2777 sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd, 2778 action_handlep, sendsnap); 2779 if (error == ENODATA) { 2780 error = 0; 2781 break; 2782 } 2783 anyerr |= error; 2784 } while (error == 0); 2785 2786 if (drr->drr_payloadlen != 0 && fromsnap != NULL) { 2787 /* 2788 * Now that we have the fs's they sent us, try the 2789 * renames again. 2790 */ 2791 softerr = recv_incremental_replication(hdl, tofs, flags, 2792 stream_nv, stream_avl, NULL); 2793 } 2794 2795out: 2796 fsavl_destroy(stream_avl); 2797 nvlist_free(stream_nv); 2798 if (softerr) 2799 error = -2; 2800 if (anyerr) 2801 error = -1; 2802 return (error); 2803} 2804 2805static void 2806trunc_prop_errs(int truncated) 2807{ 2808 ASSERT(truncated != 0); 2809 2810 if (truncated == 1) 2811 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 2812 "1 more property could not be set\n")); 2813 else 2814 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 2815 "%d more properties could not be set\n"), truncated); 2816} 2817 2818static int 2819recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap) 2820{ 2821 dmu_replay_record_t *drr; 2822 void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE); 2823 char errbuf[1024]; 2824 2825 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2826 "cannot receive:")); 2827 2828 /* XXX would be great to use lseek if possible... */ 2829 drr = buf; 2830 2831 while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t), 2832 byteswap, NULL) == 0) { 2833 if (byteswap) 2834 drr->drr_type = BSWAP_32(drr->drr_type); 2835 2836 switch (drr->drr_type) { 2837 case DRR_BEGIN: 2838 if (drr->drr_payloadlen != 0) { 2839 (void) recv_read(hdl, fd, buf, 2840 drr->drr_payloadlen, B_FALSE, NULL); 2841 } 2842 break; 2843 2844 case DRR_END: 2845 free(buf); 2846 return (0); 2847 2848 case DRR_OBJECT: 2849 if (byteswap) { 2850 drr->drr_u.drr_object.drr_bonuslen = 2851 BSWAP_32(drr->drr_u.drr_object. 2852 drr_bonuslen); 2853 } 2854 (void) recv_read(hdl, fd, buf, 2855 P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8), 2856 B_FALSE, NULL); 2857 break; 2858 2859 case DRR_WRITE: 2860 if (byteswap) { 2861 drr->drr_u.drr_write.drr_length = 2862 BSWAP_64(drr->drr_u.drr_write.drr_length); 2863 } 2864 (void) recv_read(hdl, fd, buf, 2865 drr->drr_u.drr_write.drr_length, B_FALSE, NULL); 2866 break; 2867 case DRR_SPILL: 2868 if (byteswap) { 2869 drr->drr_u.drr_write.drr_length = 2870 BSWAP_64(drr->drr_u.drr_spill.drr_length); 2871 } 2872 (void) recv_read(hdl, fd, buf, 2873 drr->drr_u.drr_spill.drr_length, B_FALSE, NULL); 2874 break; 2875 case DRR_WRITE_EMBEDDED: 2876 if (byteswap) { 2877 drr->drr_u.drr_write_embedded.drr_psize = 2878 BSWAP_32(drr->drr_u.drr_write_embedded. 2879 drr_psize); 2880 } 2881 (void) recv_read(hdl, fd, buf, 2882 P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize, 2883 8), B_FALSE, NULL); 2884 break; 2885 case DRR_WRITE_BYREF: 2886 case DRR_FREEOBJECTS: 2887 case DRR_FREE: 2888 break; 2889 2890 default: 2891 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2892 "invalid record type")); 2893 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 2894 } 2895 } 2896 2897 free(buf); 2898 return (-1); 2899} 2900 2901static void 2902recv_ecksum_set_aux(libzfs_handle_t *hdl, const char *target_snap, 2903 boolean_t resumable) 2904{ 2905 char target_fs[ZFS_MAXNAMELEN]; 2906 2907 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2908 "checksum mismatch or incomplete stream")); 2909 2910 if (!resumable) 2911 return; 2912 (void) strlcpy(target_fs, target_snap, sizeof (target_fs)); 2913 *strchr(target_fs, '@') = '\0'; 2914 zfs_handle_t *zhp = zfs_open(hdl, target_fs, 2915 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 2916 if (zhp == NULL) 2917 return; 2918 2919 char token_buf[ZFS_MAXPROPLEN]; 2920 int error = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN, 2921 token_buf, sizeof (token_buf), 2922 NULL, NULL, 0, B_TRUE); 2923 if (error == 0) { 2924 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2925 "checksum mismatch or incomplete stream.\n" 2926 "Partially received snapshot is saved.\n" 2927 "A resuming stream can be generated on the sending " 2928 "system by running:\n" 2929 " zfs send -t %s"), 2930 token_buf); 2931 } 2932 zfs_close(zhp); 2933} 2934 2935/* 2936 * Restores a backup of tosnap from the file descriptor specified by infd. 2937 */ 2938static int 2939zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, 2940 const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr, 2941 dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv, 2942 avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd, 2943 uint64_t *action_handlep, const char *finalsnap) 2944{ 2945 zfs_cmd_t zc = { 0 }; 2946 time_t begin_time; 2947 int ioctl_err, ioctl_errno, err; 2948 char *cp; 2949 struct drr_begin *drrb = &drr->drr_u.drr_begin; 2950 char errbuf[1024]; 2951 char prop_errbuf[1024]; 2952 const char *chopprefix; 2953 boolean_t newfs = B_FALSE; 2954 boolean_t stream_wantsnewfs; 2955 uint64_t parent_snapguid = 0; 2956 prop_changelist_t *clp = NULL; 2957 nvlist_t *snapprops_nvlist = NULL; 2958 zprop_errflags_t prop_errflags; 2959 boolean_t recursive; 2960 char *snapname = NULL; 2961 2962 begin_time = time(NULL); 2963 2964 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2965 "cannot receive")); 2966 2967 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2968 ENOENT); 2969 2970 if (stream_avl != NULL) { 2971 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid, 2972 &snapname); 2973 nvlist_t *props; 2974 int ret; 2975 2976 (void) nvlist_lookup_uint64(fs, "parentfromsnap", 2977 &parent_snapguid); 2978 err = nvlist_lookup_nvlist(fs, "props", &props); 2979 if (err) 2980 VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0)); 2981 2982 if (flags->canmountoff) { 2983 VERIFY(0 == nvlist_add_uint64(props, 2984 zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0)); 2985 } 2986 ret = zcmd_write_src_nvlist(hdl, &zc, props); 2987 if (err) 2988 nvlist_free(props); 2989 2990 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) { 2991 VERIFY(0 == nvlist_lookup_nvlist(props, 2992 snapname, &snapprops_nvlist)); 2993 } 2994 2995 if (ret != 0) 2996 return (-1); 2997 } 2998 2999 cp = NULL; 3000 3001 /* 3002 * Determine how much of the snapshot name stored in the stream 3003 * we are going to tack on to the name they specified on the 3004 * command line, and how much we are going to chop off. 3005 * 3006 * If they specified a snapshot, chop the entire name stored in 3007 * the stream. 3008 */ 3009 if (flags->istail) { 3010 /* 3011 * A filesystem was specified with -e. We want to tack on only 3012 * the tail of the sent snapshot path. 3013 */ 3014 if (strchr(tosnap, '@')) { 3015 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3016 "argument - snapshot not allowed with -e")); 3017 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 3018 } 3019 3020 chopprefix = strrchr(sendfs, '/'); 3021 3022 if (chopprefix == NULL) { 3023 /* 3024 * The tail is the poolname, so we need to 3025 * prepend a path separator. 3026 */ 3027 int len = strlen(drrb->drr_toname); 3028 cp = malloc(len + 2); 3029 cp[0] = '/'; 3030 (void) strcpy(&cp[1], drrb->drr_toname); 3031 chopprefix = cp; 3032 } else { 3033 chopprefix = drrb->drr_toname + (chopprefix - sendfs); 3034 } 3035 } else if (flags->isprefix) { 3036 /* 3037 * A filesystem was specified with -d. We want to tack on 3038 * everything but the first element of the sent snapshot path 3039 * (all but the pool name). 3040 */ 3041 if (strchr(tosnap, '@')) { 3042 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3043 "argument - snapshot not allowed with -d")); 3044 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 3045 } 3046 3047 chopprefix = strchr(drrb->drr_toname, '/'); 3048 if (chopprefix == NULL) 3049 chopprefix = strchr(drrb->drr_toname, '@'); 3050 } else if (strchr(tosnap, '@') == NULL) { 3051 /* 3052 * If a filesystem was specified without -d or -e, we want to 3053 * tack on everything after the fs specified by 'zfs send'. 3054 */ 3055 chopprefix = drrb->drr_toname + strlen(sendfs); 3056 } else { 3057 /* A snapshot was specified as an exact path (no -d or -e). */ 3058 if (recursive) { 3059 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3060 "cannot specify snapshot name for multi-snapshot " 3061 "stream")); 3062 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3063 } 3064 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname); 3065 } 3066 3067 ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname); 3068 ASSERT(chopprefix > drrb->drr_toname); 3069 ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname)); 3070 ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' || 3071 chopprefix[0] == '\0'); 3072 3073 /* 3074 * Determine name of destination snapshot, store in zc_value. 3075 */ 3076 (void) strcpy(zc.zc_value, tosnap); 3077 (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value)); 3078#ifdef __FreeBSD__ 3079 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF) 3080 zfs_ioctl_version = get_zfs_ioctl_version(); 3081 /* 3082 * For forward compatibility hide tosnap in zc_value 3083 */ 3084 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 3085 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap); 3086#endif 3087 free(cp); 3088 if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) { 3089 zcmd_free_nvlists(&zc); 3090 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 3091 } 3092 3093 /* 3094 * Determine the name of the origin snapshot, store in zc_string. 3095 */ 3096 if (drrb->drr_flags & DRR_FLAG_CLONE) { 3097 if (guid_to_name(hdl, zc.zc_value, 3098 drrb->drr_fromguid, B_FALSE, zc.zc_string) != 0) { 3099 zcmd_free_nvlists(&zc); 3100 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3101 "local origin for clone %s does not exist"), 3102 zc.zc_value); 3103 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3104 } 3105 if (flags->verbose) 3106 (void) printf("found clone origin %s\n", zc.zc_string); 3107 } else if (originsnap) { 3108 (void) strncpy(zc.zc_string, originsnap, ZFS_MAXNAMELEN); 3109 if (flags->verbose) 3110 (void) printf("using provided clone origin %s\n", 3111 zc.zc_string); 3112 } 3113 3114 boolean_t resuming = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 3115 DMU_BACKUP_FEATURE_RESUMING; 3116 stream_wantsnewfs = (drrb->drr_fromguid == 0 || 3117 (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && !resuming; 3118 3119 if (stream_wantsnewfs) { 3120 /* 3121 * if the parent fs does not exist, look for it based on 3122 * the parent snap GUID 3123 */ 3124 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 3125 "cannot receive new filesystem stream")); 3126 3127 (void) strcpy(zc.zc_name, zc.zc_value); 3128 cp = strrchr(zc.zc_name, '/'); 3129 if (cp) 3130 *cp = '\0'; 3131 if (cp && 3132 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 3133 char suffix[ZFS_MAXNAMELEN]; 3134 (void) strcpy(suffix, strrchr(zc.zc_value, '/')); 3135 if (guid_to_name(hdl, zc.zc_name, parent_snapguid, 3136 B_FALSE, zc.zc_value) == 0) { 3137 *strchr(zc.zc_value, '@') = '\0'; 3138 (void) strcat(zc.zc_value, suffix); 3139 } 3140 } 3141 } else { 3142 /* 3143 * if the fs does not exist, look for it based on the 3144 * fromsnap GUID 3145 */ 3146 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 3147 "cannot receive incremental stream")); 3148 3149 (void) strcpy(zc.zc_name, zc.zc_value); 3150 *strchr(zc.zc_name, '@') = '\0'; 3151 3152 /* 3153 * If the exact receive path was specified and this is the 3154 * topmost path in the stream, then if the fs does not exist we 3155 * should look no further. 3156 */ 3157 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname + 3158 strlen(sendfs)) != '\0' && *chopprefix != '@')) && 3159 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 3160 char snap[ZFS_MAXNAMELEN]; 3161 (void) strcpy(snap, strchr(zc.zc_value, '@')); 3162 if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid, 3163 B_FALSE, zc.zc_value) == 0) { 3164 *strchr(zc.zc_value, '@') = '\0'; 3165 (void) strcat(zc.zc_value, snap); 3166 } 3167 } 3168 } 3169 3170 (void) strcpy(zc.zc_name, zc.zc_value); 3171 *strchr(zc.zc_name, '@') = '\0'; 3172 3173 if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 3174 zfs_handle_t *zhp; 3175 3176 /* 3177 * Destination fs exists. It must be one of these cases: 3178 * - an incremental send stream 3179 * - the stream specifies a new fs (full stream or clone) 3180 * and they want us to blow away the existing fs (and 3181 * have therefore specified -F and removed any snapshots) 3182 * - we are resuming a failed receive. 3183 */ 3184 if (stream_wantsnewfs) { 3185 if (!flags->force) { 3186 zcmd_free_nvlists(&zc); 3187 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3188 "destination '%s' exists\n" 3189 "must specify -F to overwrite it"), 3190 zc.zc_name); 3191 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 3192 } 3193 if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT, 3194 &zc) == 0) { 3195 zcmd_free_nvlists(&zc); 3196 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3197 "destination has snapshots (eg. %s)\n" 3198 "must destroy them to overwrite it"), 3199 zc.zc_name); 3200 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 3201 } 3202 } 3203 3204 if ((zhp = zfs_open(hdl, zc.zc_name, 3205 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) { 3206 zcmd_free_nvlists(&zc); 3207 return (-1); 3208 } 3209 3210 if (stream_wantsnewfs && 3211 zhp->zfs_dmustats.dds_origin[0]) { 3212 zcmd_free_nvlists(&zc); 3213 zfs_close(zhp); 3214 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3215 "destination '%s' is a clone\n" 3216 "must destroy it to overwrite it"), 3217 zc.zc_name); 3218 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 3219 } 3220 3221 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM && 3222 stream_wantsnewfs) { 3223 /* We can't do online recv in this case */ 3224 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0); 3225 if (clp == NULL) { 3226 zfs_close(zhp); 3227 zcmd_free_nvlists(&zc); 3228 return (-1); 3229 } 3230 if (changelist_prefix(clp) != 0) { 3231 changelist_free(clp); 3232 zfs_close(zhp); 3233 zcmd_free_nvlists(&zc); 3234 return (-1); 3235 } 3236 } 3237 3238 /* 3239 * If we are resuming a newfs, set newfs here so that we will 3240 * mount it if the recv succeeds this time. We can tell 3241 * that it was a newfs on the first recv because the fs 3242 * itself will be inconsistent (if the fs existed when we 3243 * did the first recv, we would have received it into 3244 * .../%recv). 3245 */ 3246 if (resuming && zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT)) 3247 newfs = B_TRUE; 3248 3249 zfs_close(zhp); 3250 } else { 3251 /* 3252 * Destination filesystem does not exist. Therefore we better 3253 * be creating a new filesystem (either from a full backup, or 3254 * a clone). It would therefore be invalid if the user 3255 * specified only the pool name (i.e. if the destination name 3256 * contained no slash character). 3257 */ 3258 if (!stream_wantsnewfs || 3259 (cp = strrchr(zc.zc_name, '/')) == NULL) { 3260 zcmd_free_nvlists(&zc); 3261 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3262 "destination '%s' does not exist"), zc.zc_name); 3263 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3264 } 3265 3266 /* 3267 * Trim off the final dataset component so we perform the 3268 * recvbackup ioctl to the filesystems's parent. 3269 */ 3270 *cp = '\0'; 3271 3272 if (flags->isprefix && !flags->istail && !flags->dryrun && 3273 create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) { 3274 zcmd_free_nvlists(&zc); 3275 return (zfs_error(hdl, EZFS_BADRESTORE, errbuf)); 3276 } 3277 3278 newfs = B_TRUE; 3279 } 3280 3281 zc.zc_begin_record = *drr_noswap; 3282 zc.zc_cookie = infd; 3283 zc.zc_guid = flags->force; 3284 zc.zc_resumable = flags->resumable; 3285 if (flags->verbose) { 3286 (void) printf("%s %s stream of %s into %s\n", 3287 flags->dryrun ? "would receive" : "receiving", 3288 drrb->drr_fromguid ? "incremental" : "full", 3289 drrb->drr_toname, zc.zc_value); 3290 (void) fflush(stdout); 3291 } 3292 3293 if (flags->dryrun) { 3294 zcmd_free_nvlists(&zc); 3295 return (recv_skip(hdl, infd, flags->byteswap)); 3296 } 3297 3298 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf; 3299 zc.zc_nvlist_dst_size = sizeof (prop_errbuf); 3300 zc.zc_cleanup_fd = cleanup_fd; 3301 zc.zc_action_handle = *action_handlep; 3302 3303 err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc); 3304 ioctl_errno = errno; 3305 prop_errflags = (zprop_errflags_t)zc.zc_obj; 3306 3307 if (err == 0) { 3308 nvlist_t *prop_errors; 3309 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, 3310 zc.zc_nvlist_dst_size, &prop_errors, 0)); 3311 3312 nvpair_t *prop_err = NULL; 3313 3314 while ((prop_err = nvlist_next_nvpair(prop_errors, 3315 prop_err)) != NULL) { 3316 char tbuf[1024]; 3317 zfs_prop_t prop; 3318 int intval; 3319 3320 prop = zfs_name_to_prop(nvpair_name(prop_err)); 3321 (void) nvpair_value_int32(prop_err, &intval); 3322 if (strcmp(nvpair_name(prop_err), 3323 ZPROP_N_MORE_ERRORS) == 0) { 3324 trunc_prop_errs(intval); 3325 break; 3326 } else if (snapname == NULL || finalsnap == NULL || 3327 strcmp(finalsnap, snapname) == 0 || 3328 strcmp(nvpair_name(prop_err), 3329 zfs_prop_to_name(ZFS_PROP_REFQUOTA)) != 0) { 3330 /* 3331 * Skip the special case of, for example, 3332 * "refquota", errors on intermediate 3333 * snapshots leading up to a final one. 3334 * That's why we have all of the checks above. 3335 * 3336 * See zfs_ioctl.c's extract_delay_props() for 3337 * a list of props which can fail on 3338 * intermediate snapshots, but shouldn't 3339 * affect the overall receive. 3340 */ 3341 (void) snprintf(tbuf, sizeof (tbuf), 3342 dgettext(TEXT_DOMAIN, 3343 "cannot receive %s property on %s"), 3344 nvpair_name(prop_err), zc.zc_name); 3345 zfs_setprop_error(hdl, prop, intval, tbuf); 3346 } 3347 } 3348 nvlist_free(prop_errors); 3349 } 3350 3351 zc.zc_nvlist_dst = 0; 3352 zc.zc_nvlist_dst_size = 0; 3353 zcmd_free_nvlists(&zc); 3354 3355 if (err == 0 && snapprops_nvlist) { 3356 zfs_cmd_t zc2 = { 0 }; 3357 3358 (void) strcpy(zc2.zc_name, zc.zc_value); 3359 zc2.zc_cookie = B_TRUE; /* received */ 3360 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) { 3361 (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2); 3362 zcmd_free_nvlists(&zc2); 3363 } 3364 } 3365 3366 if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) { 3367 /* 3368 * It may be that this snapshot already exists, 3369 * in which case we want to consume & ignore it 3370 * rather than failing. 3371 */ 3372 avl_tree_t *local_avl; 3373 nvlist_t *local_nv, *fs; 3374 cp = strchr(zc.zc_value, '@'); 3375 3376 /* 3377 * XXX Do this faster by just iterating over snaps in 3378 * this fs. Also if zc_value does not exist, we will 3379 * get a strange "does not exist" error message. 3380 */ 3381 *cp = '\0'; 3382 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE, 3383 &local_nv, &local_avl) == 0) { 3384 *cp = '@'; 3385 fs = fsavl_find(local_avl, drrb->drr_toguid, NULL); 3386 fsavl_destroy(local_avl); 3387 nvlist_free(local_nv); 3388 3389 if (fs != NULL) { 3390 if (flags->verbose) { 3391 (void) printf("snap %s already exists; " 3392 "ignoring\n", zc.zc_value); 3393 } 3394 err = ioctl_err = recv_skip(hdl, infd, 3395 flags->byteswap); 3396 } 3397 } 3398 *cp = '@'; 3399 } 3400 3401 if (ioctl_err != 0) { 3402 switch (ioctl_errno) { 3403 case ENODEV: 3404 cp = strchr(zc.zc_value, '@'); 3405 *cp = '\0'; 3406 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3407 "most recent snapshot of %s does not\n" 3408 "match incremental source"), zc.zc_value); 3409 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); 3410 *cp = '@'; 3411 break; 3412 case ETXTBSY: 3413 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3414 "destination %s has been modified\n" 3415 "since most recent snapshot"), zc.zc_name); 3416 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); 3417 break; 3418 case EEXIST: 3419 cp = strchr(zc.zc_value, '@'); 3420 if (newfs) { 3421 /* it's the containing fs that exists */ 3422 *cp = '\0'; 3423 } 3424 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3425 "destination already exists")); 3426 (void) zfs_error_fmt(hdl, EZFS_EXISTS, 3427 dgettext(TEXT_DOMAIN, "cannot restore to %s"), 3428 zc.zc_value); 3429 *cp = '@'; 3430 break; 3431 case EINVAL: 3432 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); 3433 break; 3434 case ECKSUM: 3435 recv_ecksum_set_aux(hdl, zc.zc_value, flags->resumable); 3436 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); 3437 break; 3438 case ENOTSUP: 3439 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3440 "pool must be upgraded to receive this stream.")); 3441 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf); 3442 break; 3443 case EDQUOT: 3444 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3445 "destination %s space quota exceeded"), zc.zc_name); 3446 (void) zfs_error(hdl, EZFS_NOSPC, errbuf); 3447 break; 3448 default: 3449 (void) zfs_standard_error(hdl, ioctl_errno, errbuf); 3450 } 3451 } 3452 3453 /* 3454 * Mount the target filesystem (if created). Also mount any 3455 * children of the target filesystem if we did a replication 3456 * receive (indicated by stream_avl being non-NULL). 3457 */ 3458 cp = strchr(zc.zc_value, '@'); 3459 if (cp && (ioctl_err == 0 || !newfs)) { 3460 zfs_handle_t *h; 3461 3462 *cp = '\0'; 3463 h = zfs_open(hdl, zc.zc_value, 3464 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 3465 if (h != NULL) { 3466 if (h->zfs_type == ZFS_TYPE_VOLUME) { 3467 *cp = '@'; 3468 } else if (newfs || stream_avl) { 3469 /* 3470 * Track the first/top of hierarchy fs, 3471 * for mounting and sharing later. 3472 */ 3473 if (top_zfs && *top_zfs == NULL) 3474 *top_zfs = zfs_strdup(hdl, zc.zc_value); 3475 } 3476 zfs_close(h); 3477 } 3478 *cp = '@'; 3479 } 3480 3481 if (clp) { 3482 if (!flags->nomount) 3483 err |= changelist_postfix(clp); 3484 changelist_free(clp); 3485 } 3486 3487 if (prop_errflags & ZPROP_ERR_NOCLEAR) { 3488 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " 3489 "failed to clear unreceived properties on %s"), 3490 zc.zc_name); 3491 (void) fprintf(stderr, "\n"); 3492 } 3493 if (prop_errflags & ZPROP_ERR_NORESTORE) { 3494 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " 3495 "failed to restore original properties on %s"), 3496 zc.zc_name); 3497 (void) fprintf(stderr, "\n"); 3498 } 3499 3500 if (err || ioctl_err) 3501 return (-1); 3502 3503 *action_handlep = zc.zc_action_handle; 3504 3505 if (flags->verbose) { 3506 char buf1[64]; 3507 char buf2[64]; 3508 uint64_t bytes = zc.zc_cookie; 3509 time_t delta = time(NULL) - begin_time; 3510 if (delta == 0) 3511 delta = 1; 3512 zfs_nicenum(bytes, buf1, sizeof (buf1)); 3513 zfs_nicenum(bytes/delta, buf2, sizeof (buf1)); 3514 3515 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n", 3516 buf1, delta, buf2); 3517 } 3518 3519 return (0); 3520} 3521 3522static int 3523zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, 3524 const char *originsnap, recvflags_t *flags, int infd, const char *sendfs, 3525 nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd, 3526 uint64_t *action_handlep, const char *finalsnap) 3527{ 3528 int err; 3529 dmu_replay_record_t drr, drr_noswap; 3530 struct drr_begin *drrb = &drr.drr_u.drr_begin; 3531 char errbuf[1024]; 3532 zio_cksum_t zcksum = { 0 }; 3533 uint64_t featureflags; 3534 int hdrtype; 3535 3536 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 3537 "cannot receive")); 3538 3539 if (flags->isprefix && 3540 !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) { 3541 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs " 3542 "(%s) does not exist"), tosnap); 3543 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3544 } 3545 if (originsnap && 3546 !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) { 3547 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs " 3548 "(%s) does not exist"), originsnap); 3549 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3550 } 3551 3552 /* read in the BEGIN record */ 3553 if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE, 3554 &zcksum))) 3555 return (err); 3556 3557 if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) { 3558 /* It's the double end record at the end of a package */ 3559 return (ENODATA); 3560 } 3561 3562 /* the kernel needs the non-byteswapped begin record */ 3563 drr_noswap = drr; 3564 3565 flags->byteswap = B_FALSE; 3566 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { 3567 /* 3568 * We computed the checksum in the wrong byteorder in 3569 * recv_read() above; do it again correctly. 3570 */ 3571 bzero(&zcksum, sizeof (zio_cksum_t)); 3572 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum); 3573 flags->byteswap = B_TRUE; 3574 3575 drr.drr_type = BSWAP_32(drr.drr_type); 3576 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen); 3577 drrb->drr_magic = BSWAP_64(drrb->drr_magic); 3578 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); 3579 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); 3580 drrb->drr_type = BSWAP_32(drrb->drr_type); 3581 drrb->drr_flags = BSWAP_32(drrb->drr_flags); 3582 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); 3583 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); 3584 } 3585 3586 if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) { 3587 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3588 "stream (bad magic number)")); 3589 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3590 } 3591 3592 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 3593 hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo); 3594 3595 if (!DMU_STREAM_SUPPORTED(featureflags) || 3596 (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) { 3597 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3598 "stream has unsupported feature, feature flags = %lx"), 3599 featureflags); 3600 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3601 } 3602 3603 if (strchr(drrb->drr_toname, '@') == NULL) { 3604 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3605 "stream (bad snapshot name)")); 3606 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3607 } 3608 3609 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) { 3610 char nonpackage_sendfs[ZFS_MAXNAMELEN]; 3611 if (sendfs == NULL) { 3612 /* 3613 * We were not called from zfs_receive_package(). Get 3614 * the fs specified by 'zfs send'. 3615 */ 3616 char *cp; 3617 (void) strlcpy(nonpackage_sendfs, 3618 drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN); 3619 if ((cp = strchr(nonpackage_sendfs, '@')) != NULL) 3620 *cp = '\0'; 3621 sendfs = nonpackage_sendfs; 3622 VERIFY(finalsnap == NULL); 3623 } 3624 return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags, 3625 &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs, 3626 cleanup_fd, action_handlep, finalsnap)); 3627 } else { 3628 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 3629 DMU_COMPOUNDSTREAM); 3630 return (zfs_receive_package(hdl, infd, tosnap, flags, &drr, 3631 &zcksum, top_zfs, cleanup_fd, action_handlep)); 3632 } 3633} 3634 3635/* 3636 * Restores a backup of tosnap from the file descriptor specified by infd. 3637 * Return 0 on total success, -2 if some things couldn't be 3638 * destroyed/renamed/promoted, -1 if some things couldn't be received. 3639 * (-1 will override -2, if -1 and the resumable flag was specified the 3640 * transfer can be resumed if the sending side supports it). 3641 */ 3642int 3643zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props, 3644 recvflags_t *flags, int infd, avl_tree_t *stream_avl) 3645{ 3646 char *top_zfs = NULL; 3647 int err; 3648 int cleanup_fd; 3649 uint64_t action_handle = 0; 3650 char *originsnap = NULL; 3651 if (props) { 3652 err = nvlist_lookup_string(props, "origin", &originsnap); 3653 if (err && err != ENOENT) 3654 return (err); 3655 } 3656 3657 cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); 3658 VERIFY(cleanup_fd >= 0); 3659 3660 err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL, 3661 stream_avl, &top_zfs, cleanup_fd, &action_handle, NULL); 3662 3663 VERIFY(0 == close(cleanup_fd)); 3664 3665 if (err == 0 && !flags->nomount && top_zfs) { 3666 zfs_handle_t *zhp; 3667 prop_changelist_t *clp; 3668 3669 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM); 3670 if (zhp != NULL) { 3671 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT, 3672 CL_GATHER_MOUNT_ALWAYS, 0); 3673 zfs_close(zhp); 3674 if (clp != NULL) { 3675 /* mount and share received datasets */ 3676 err = changelist_postfix(clp); 3677 changelist_free(clp); 3678 } 3679 } 3680 if (zhp == NULL || clp == NULL || err) 3681 err = -1; 3682 } 3683 if (top_zfs) 3684 free(top_zfs); 3685 3686 return (err); 3687} 3688