libzfs_sendrecv.c revision 276081
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 25 * Copyright (c) 2012, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2012 Pawel Jakub Dawidek <pawel@dawidek.net>. 27 * All rights reserved. 28 * Copyright (c) 2013 Steven Hartland. All rights reserved. 29 */ 30 31#include <assert.h> 32#include <ctype.h> 33#include <errno.h> 34#include <libintl.h> 35#include <stdio.h> 36#include <stdlib.h> 37#include <strings.h> 38#include <unistd.h> 39#include <stddef.h> 40#include <fcntl.h> 41#include <sys/param.h> 42#include <sys/mount.h> 43#include <pthread.h> 44#include <umem.h> 45#include <time.h> 46 47#include <libzfs.h> 48#include <libzfs_core.h> 49 50#include "zfs_namecheck.h" 51#include "zfs_prop.h" 52#include "zfs_fletcher.h" 53#include "libzfs_impl.h" 54#include <sha2.h> 55#include <sys/zio_checksum.h> 56#include <sys/ddt.h> 57 58#ifdef __FreeBSD__ 59extern int zfs_ioctl_version; 60#endif 61 62/* in libzfs_dataset.c */ 63extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *); 64/* We need to use something for ENODATA. */ 65#define ENODATA EIDRM 66 67static int zfs_receive_impl(libzfs_handle_t *, const char *, recvflags_t *, 68 int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *); 69 70static const zio_cksum_t zero_cksum = { 0 }; 71 72typedef struct dedup_arg { 73 int inputfd; 74 int outputfd; 75 libzfs_handle_t *dedup_hdl; 76} dedup_arg_t; 77 78typedef struct progress_arg { 79 zfs_handle_t *pa_zhp; 80 int pa_fd; 81 boolean_t pa_parsable; 82} progress_arg_t; 83 84typedef struct dataref { 85 uint64_t ref_guid; 86 uint64_t ref_object; 87 uint64_t ref_offset; 88} dataref_t; 89 90typedef struct dedup_entry { 91 struct dedup_entry *dde_next; 92 zio_cksum_t dde_chksum; 93 uint64_t dde_prop; 94 dataref_t dde_ref; 95} dedup_entry_t; 96 97#define MAX_DDT_PHYSMEM_PERCENT 20 98#define SMALLEST_POSSIBLE_MAX_DDT_MB 128 99 100typedef struct dedup_table { 101 dedup_entry_t **dedup_hash_array; 102 umem_cache_t *ddecache; 103 uint64_t max_ddt_size; /* max dedup table size in bytes */ 104 uint64_t cur_ddt_size; /* current dedup table size in bytes */ 105 uint64_t ddt_count; 106 int numhashbits; 107 boolean_t ddt_full; 108} dedup_table_t; 109 110static int 111high_order_bit(uint64_t n) 112{ 113 int count; 114 115 for (count = 0; n != 0; count++) 116 n >>= 1; 117 return (count); 118} 119 120static size_t 121ssread(void *buf, size_t len, FILE *stream) 122{ 123 size_t outlen; 124 125 if ((outlen = fread(buf, len, 1, stream)) == 0) 126 return (0); 127 128 return (outlen); 129} 130 131static void 132ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp, 133 zio_cksum_t *cs, uint64_t prop, dataref_t *dr) 134{ 135 dedup_entry_t *dde; 136 137 if (ddt->cur_ddt_size >= ddt->max_ddt_size) { 138 if (ddt->ddt_full == B_FALSE) { 139 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 140 "Dedup table full. Deduplication will continue " 141 "with existing table entries")); 142 ddt->ddt_full = B_TRUE; 143 } 144 return; 145 } 146 147 if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT)) 148 != NULL) { 149 assert(*ddepp == NULL); 150 dde->dde_next = NULL; 151 dde->dde_chksum = *cs; 152 dde->dde_prop = prop; 153 dde->dde_ref = *dr; 154 *ddepp = dde; 155 ddt->cur_ddt_size += sizeof (dedup_entry_t); 156 ddt->ddt_count++; 157 } 158} 159 160/* 161 * Using the specified dedup table, do a lookup for an entry with 162 * the checksum cs. If found, return the block's reference info 163 * in *dr. Otherwise, insert a new entry in the dedup table, using 164 * the reference information specified by *dr. 165 * 166 * return value: true - entry was found 167 * false - entry was not found 168 */ 169static boolean_t 170ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs, 171 uint64_t prop, dataref_t *dr) 172{ 173 uint32_t hashcode; 174 dedup_entry_t **ddepp; 175 176 hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits); 177 178 for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL; 179 ddepp = &((*ddepp)->dde_next)) { 180 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) && 181 (*ddepp)->dde_prop == prop) { 182 *dr = (*ddepp)->dde_ref; 183 return (B_TRUE); 184 } 185 } 186 ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr); 187 return (B_FALSE); 188} 189 190static int 191cksum_and_write(const void *buf, uint64_t len, zio_cksum_t *zc, int outfd) 192{ 193 fletcher_4_incremental_native(buf, len, zc); 194 return (write(outfd, buf, len)); 195} 196 197/* 198 * This function is started in a separate thread when the dedup option 199 * has been requested. The main send thread determines the list of 200 * snapshots to be included in the send stream and makes the ioctl calls 201 * for each one. But instead of having the ioctl send the output to the 202 * the output fd specified by the caller of zfs_send()), the 203 * ioctl is told to direct the output to a pipe, which is read by the 204 * alternate thread running THIS function. This function does the 205 * dedup'ing by: 206 * 1. building a dedup table (the DDT) 207 * 2. doing checksums on each data block and inserting a record in the DDT 208 * 3. looking for matching checksums, and 209 * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever 210 * a duplicate block is found. 211 * The output of this function then goes to the output fd requested 212 * by the caller of zfs_send(). 213 */ 214static void * 215cksummer(void *arg) 216{ 217 dedup_arg_t *dda = arg; 218 char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE); 219 dmu_replay_record_t thedrr; 220 dmu_replay_record_t *drr = &thedrr; 221 struct drr_begin *drrb = &thedrr.drr_u.drr_begin; 222 struct drr_end *drre = &thedrr.drr_u.drr_end; 223 struct drr_object *drro = &thedrr.drr_u.drr_object; 224 struct drr_write *drrw = &thedrr.drr_u.drr_write; 225 struct drr_spill *drrs = &thedrr.drr_u.drr_spill; 226 struct drr_write_embedded *drrwe = &thedrr.drr_u.drr_write_embedded; 227 FILE *ofp; 228 int outfd; 229 dmu_replay_record_t wbr_drr = {0}; 230 struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref; 231 dedup_table_t ddt; 232 zio_cksum_t stream_cksum; 233 uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE); 234 uint64_t numbuckets; 235 236 ddt.max_ddt_size = 237 MAX((physmem * MAX_DDT_PHYSMEM_PERCENT)/100, 238 SMALLEST_POSSIBLE_MAX_DDT_MB<<20); 239 240 numbuckets = ddt.max_ddt_size/(sizeof (dedup_entry_t)); 241 242 /* 243 * numbuckets must be a power of 2. Increase number to 244 * a power of 2 if necessary. 245 */ 246 if (!ISP2(numbuckets)) 247 numbuckets = 1 << high_order_bit(numbuckets); 248 249 ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *)); 250 ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0, 251 NULL, NULL, NULL, NULL, NULL, 0); 252 ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *); 253 ddt.numhashbits = high_order_bit(numbuckets) - 1; 254 ddt.ddt_full = B_FALSE; 255 256 /* Initialize the write-by-reference block. */ 257 wbr_drr.drr_type = DRR_WRITE_BYREF; 258 wbr_drr.drr_payloadlen = 0; 259 260 outfd = dda->outputfd; 261 ofp = fdopen(dda->inputfd, "r"); 262 while (ssread(drr, sizeof (dmu_replay_record_t), ofp) != 0) { 263 264 switch (drr->drr_type) { 265 case DRR_BEGIN: 266 { 267 int fflags; 268 ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0); 269 270 /* set the DEDUP feature flag for this stream */ 271 fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 272 fflags |= (DMU_BACKUP_FEATURE_DEDUP | 273 DMU_BACKUP_FEATURE_DEDUPPROPS); 274 DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags); 275 276 if (cksum_and_write(drr, sizeof (dmu_replay_record_t), 277 &stream_cksum, outfd) == -1) 278 goto out; 279 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 280 DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) { 281 int sz = drr->drr_payloadlen; 282 283 if (sz > SPA_MAXBLOCKSIZE) { 284 buf = zfs_realloc(dda->dedup_hdl, buf, 285 SPA_MAXBLOCKSIZE, sz); 286 } 287 (void) ssread(buf, sz, ofp); 288 if (ferror(stdin)) 289 perror("fread"); 290 if (cksum_and_write(buf, sz, &stream_cksum, 291 outfd) == -1) 292 goto out; 293 } 294 break; 295 } 296 297 case DRR_END: 298 { 299 /* use the recalculated checksum */ 300 ZIO_SET_CHECKSUM(&drre->drr_checksum, 301 stream_cksum.zc_word[0], stream_cksum.zc_word[1], 302 stream_cksum.zc_word[2], stream_cksum.zc_word[3]); 303 if ((write(outfd, drr, 304 sizeof (dmu_replay_record_t))) == -1) 305 goto out; 306 break; 307 } 308 309 case DRR_OBJECT: 310 { 311 if (cksum_and_write(drr, sizeof (dmu_replay_record_t), 312 &stream_cksum, outfd) == -1) 313 goto out; 314 if (drro->drr_bonuslen > 0) { 315 (void) ssread(buf, 316 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), 317 ofp); 318 if (cksum_and_write(buf, 319 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), 320 &stream_cksum, outfd) == -1) 321 goto out; 322 } 323 break; 324 } 325 326 case DRR_SPILL: 327 { 328 if (cksum_and_write(drr, sizeof (dmu_replay_record_t), 329 &stream_cksum, outfd) == -1) 330 goto out; 331 (void) ssread(buf, drrs->drr_length, ofp); 332 if (cksum_and_write(buf, drrs->drr_length, 333 &stream_cksum, outfd) == -1) 334 goto out; 335 break; 336 } 337 338 case DRR_FREEOBJECTS: 339 { 340 if (cksum_and_write(drr, sizeof (dmu_replay_record_t), 341 &stream_cksum, outfd) == -1) 342 goto out; 343 break; 344 } 345 346 case DRR_WRITE: 347 { 348 dataref_t dataref; 349 350 (void) ssread(buf, drrw->drr_length, ofp); 351 352 /* 353 * Use the existing checksum if it's dedup-capable, 354 * else calculate a SHA256 checksum for it. 355 */ 356 357 if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum, 358 zero_cksum) || 359 !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) { 360 SHA256_CTX ctx; 361 zio_cksum_t tmpsha256; 362 363 SHA256Init(&ctx); 364 SHA256Update(&ctx, buf, drrw->drr_length); 365 SHA256Final(&tmpsha256, &ctx); 366 drrw->drr_key.ddk_cksum.zc_word[0] = 367 BE_64(tmpsha256.zc_word[0]); 368 drrw->drr_key.ddk_cksum.zc_word[1] = 369 BE_64(tmpsha256.zc_word[1]); 370 drrw->drr_key.ddk_cksum.zc_word[2] = 371 BE_64(tmpsha256.zc_word[2]); 372 drrw->drr_key.ddk_cksum.zc_word[3] = 373 BE_64(tmpsha256.zc_word[3]); 374 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256; 375 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP; 376 } 377 378 dataref.ref_guid = drrw->drr_toguid; 379 dataref.ref_object = drrw->drr_object; 380 dataref.ref_offset = drrw->drr_offset; 381 382 if (ddt_update(dda->dedup_hdl, &ddt, 383 &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop, 384 &dataref)) { 385 /* block already present in stream */ 386 wbr_drrr->drr_object = drrw->drr_object; 387 wbr_drrr->drr_offset = drrw->drr_offset; 388 wbr_drrr->drr_length = drrw->drr_length; 389 wbr_drrr->drr_toguid = drrw->drr_toguid; 390 wbr_drrr->drr_refguid = dataref.ref_guid; 391 wbr_drrr->drr_refobject = 392 dataref.ref_object; 393 wbr_drrr->drr_refoffset = 394 dataref.ref_offset; 395 396 wbr_drrr->drr_checksumtype = 397 drrw->drr_checksumtype; 398 wbr_drrr->drr_checksumflags = 399 drrw->drr_checksumtype; 400 wbr_drrr->drr_key.ddk_cksum = 401 drrw->drr_key.ddk_cksum; 402 wbr_drrr->drr_key.ddk_prop = 403 drrw->drr_key.ddk_prop; 404 405 if (cksum_and_write(&wbr_drr, 406 sizeof (dmu_replay_record_t), &stream_cksum, 407 outfd) == -1) 408 goto out; 409 } else { 410 /* block not previously seen */ 411 if (cksum_and_write(drr, 412 sizeof (dmu_replay_record_t), &stream_cksum, 413 outfd) == -1) 414 goto out; 415 if (cksum_and_write(buf, 416 drrw->drr_length, 417 &stream_cksum, outfd) == -1) 418 goto out; 419 } 420 break; 421 } 422 423 case DRR_WRITE_EMBEDDED: 424 { 425 if (cksum_and_write(drr, sizeof (dmu_replay_record_t), 426 &stream_cksum, outfd) == -1) 427 goto out; 428 (void) ssread(buf, 429 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp); 430 if (cksum_and_write(buf, 431 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), 432 &stream_cksum, outfd) == -1) 433 goto out; 434 break; 435 } 436 437 case DRR_FREE: 438 { 439 if (cksum_and_write(drr, sizeof (dmu_replay_record_t), 440 &stream_cksum, outfd) == -1) 441 goto out; 442 break; 443 } 444 445 default: 446 (void) printf("INVALID record type 0x%x\n", 447 drr->drr_type); 448 /* should never happen, so assert */ 449 assert(B_FALSE); 450 } 451 } 452out: 453 umem_cache_destroy(ddt.ddecache); 454 free(ddt.dedup_hash_array); 455 free(buf); 456 (void) fclose(ofp); 457 458 return (NULL); 459} 460 461/* 462 * Routines for dealing with the AVL tree of fs-nvlists 463 */ 464typedef struct fsavl_node { 465 avl_node_t fn_node; 466 nvlist_t *fn_nvfs; 467 char *fn_snapname; 468 uint64_t fn_guid; 469} fsavl_node_t; 470 471static int 472fsavl_compare(const void *arg1, const void *arg2) 473{ 474 const fsavl_node_t *fn1 = arg1; 475 const fsavl_node_t *fn2 = arg2; 476 477 if (fn1->fn_guid > fn2->fn_guid) 478 return (+1); 479 else if (fn1->fn_guid < fn2->fn_guid) 480 return (-1); 481 else 482 return (0); 483} 484 485/* 486 * Given the GUID of a snapshot, find its containing filesystem and 487 * (optionally) name. 488 */ 489static nvlist_t * 490fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname) 491{ 492 fsavl_node_t fn_find; 493 fsavl_node_t *fn; 494 495 fn_find.fn_guid = snapguid; 496 497 fn = avl_find(avl, &fn_find, NULL); 498 if (fn) { 499 if (snapname) 500 *snapname = fn->fn_snapname; 501 return (fn->fn_nvfs); 502 } 503 return (NULL); 504} 505 506static void 507fsavl_destroy(avl_tree_t *avl) 508{ 509 fsavl_node_t *fn; 510 void *cookie; 511 512 if (avl == NULL) 513 return; 514 515 cookie = NULL; 516 while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL) 517 free(fn); 518 avl_destroy(avl); 519 free(avl); 520} 521 522/* 523 * Given an nvlist, produce an avl tree of snapshots, ordered by guid 524 */ 525static avl_tree_t * 526fsavl_create(nvlist_t *fss) 527{ 528 avl_tree_t *fsavl; 529 nvpair_t *fselem = NULL; 530 531 if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL) 532 return (NULL); 533 534 avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t), 535 offsetof(fsavl_node_t, fn_node)); 536 537 while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) { 538 nvlist_t *nvfs, *snaps; 539 nvpair_t *snapelem = NULL; 540 541 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); 542 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); 543 544 while ((snapelem = 545 nvlist_next_nvpair(snaps, snapelem)) != NULL) { 546 fsavl_node_t *fn; 547 uint64_t guid; 548 549 VERIFY(0 == nvpair_value_uint64(snapelem, &guid)); 550 if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) { 551 fsavl_destroy(fsavl); 552 return (NULL); 553 } 554 fn->fn_nvfs = nvfs; 555 fn->fn_snapname = nvpair_name(snapelem); 556 fn->fn_guid = guid; 557 558 /* 559 * Note: if there are multiple snaps with the 560 * same GUID, we ignore all but one. 561 */ 562 if (avl_find(fsavl, fn, NULL) == NULL) 563 avl_add(fsavl, fn); 564 else 565 free(fn); 566 } 567 } 568 569 return (fsavl); 570} 571 572/* 573 * Routines for dealing with the giant nvlist of fs-nvlists, etc. 574 */ 575typedef struct send_data { 576 uint64_t parent_fromsnap_guid; 577 nvlist_t *parent_snaps; 578 nvlist_t *fss; 579 nvlist_t *snapprops; 580 const char *fromsnap; 581 const char *tosnap; 582 boolean_t recursive; 583 584 /* 585 * The header nvlist is of the following format: 586 * { 587 * "tosnap" -> string 588 * "fromsnap" -> string (if incremental) 589 * "fss" -> { 590 * id -> { 591 * 592 * "name" -> string (full name; for debugging) 593 * "parentfromsnap" -> number (guid of fromsnap in parent) 594 * 595 * "props" -> { name -> value (only if set here) } 596 * "snaps" -> { name (lastname) -> number (guid) } 597 * "snapprops" -> { name (lastname) -> { name -> value } } 598 * 599 * "origin" -> number (guid) (if clone) 600 * "sent" -> boolean (not on-disk) 601 * } 602 * } 603 * } 604 * 605 */ 606} send_data_t; 607 608static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv); 609 610static int 611send_iterate_snap(zfs_handle_t *zhp, void *arg) 612{ 613 send_data_t *sd = arg; 614 uint64_t guid = zhp->zfs_dmustats.dds_guid; 615 char *snapname; 616 nvlist_t *nv; 617 618 snapname = strrchr(zhp->zfs_name, '@')+1; 619 620 VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid)); 621 /* 622 * NB: if there is no fromsnap here (it's a newly created fs in 623 * an incremental replication), we will substitute the tosnap. 624 */ 625 if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) || 626 (sd->parent_fromsnap_guid == 0 && sd->tosnap && 627 strcmp(snapname, sd->tosnap) == 0)) { 628 sd->parent_fromsnap_guid = guid; 629 } 630 631 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); 632 send_iterate_prop(zhp, nv); 633 VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv)); 634 nvlist_free(nv); 635 636 zfs_close(zhp); 637 return (0); 638} 639 640static void 641send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv) 642{ 643 nvpair_t *elem = NULL; 644 645 while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) { 646 char *propname = nvpair_name(elem); 647 zfs_prop_t prop = zfs_name_to_prop(propname); 648 nvlist_t *propnv; 649 650 if (!zfs_prop_user(propname)) { 651 /* 652 * Realistically, this should never happen. However, 653 * we want the ability to add DSL properties without 654 * needing to make incompatible version changes. We 655 * need to ignore unknown properties to allow older 656 * software to still send datasets containing these 657 * properties, with the unknown properties elided. 658 */ 659 if (prop == ZPROP_INVAL) 660 continue; 661 662 if (zfs_prop_readonly(prop)) 663 continue; 664 } 665 666 verify(nvpair_value_nvlist(elem, &propnv) == 0); 667 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION || 668 prop == ZFS_PROP_REFQUOTA || 669 prop == ZFS_PROP_REFRESERVATION) { 670 char *source; 671 uint64_t value; 672 verify(nvlist_lookup_uint64(propnv, 673 ZPROP_VALUE, &value) == 0); 674 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) 675 continue; 676 /* 677 * May have no source before SPA_VERSION_RECVD_PROPS, 678 * but is still modifiable. 679 */ 680 if (nvlist_lookup_string(propnv, 681 ZPROP_SOURCE, &source) == 0) { 682 if ((strcmp(source, zhp->zfs_name) != 0) && 683 (strcmp(source, 684 ZPROP_SOURCE_VAL_RECVD) != 0)) 685 continue; 686 } 687 } else { 688 char *source; 689 if (nvlist_lookup_string(propnv, 690 ZPROP_SOURCE, &source) != 0) 691 continue; 692 if ((strcmp(source, zhp->zfs_name) != 0) && 693 (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0)) 694 continue; 695 } 696 697 if (zfs_prop_user(propname) || 698 zfs_prop_get_type(prop) == PROP_TYPE_STRING) { 699 char *value; 700 verify(nvlist_lookup_string(propnv, 701 ZPROP_VALUE, &value) == 0); 702 VERIFY(0 == nvlist_add_string(nv, propname, value)); 703 } else { 704 uint64_t value; 705 verify(nvlist_lookup_uint64(propnv, 706 ZPROP_VALUE, &value) == 0); 707 VERIFY(0 == nvlist_add_uint64(nv, propname, value)); 708 } 709 } 710} 711 712/* 713 * recursively generate nvlists describing datasets. See comment 714 * for the data structure send_data_t above for description of contents 715 * of the nvlist. 716 */ 717static int 718send_iterate_fs(zfs_handle_t *zhp, void *arg) 719{ 720 send_data_t *sd = arg; 721 nvlist_t *nvfs, *nv; 722 int rv = 0; 723 uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid; 724 uint64_t guid = zhp->zfs_dmustats.dds_guid; 725 char guidstring[64]; 726 727 VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0)); 728 VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name)); 729 VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap", 730 sd->parent_fromsnap_guid)); 731 732 if (zhp->zfs_dmustats.dds_origin[0]) { 733 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl, 734 zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT); 735 if (origin == NULL) 736 return (-1); 737 VERIFY(0 == nvlist_add_uint64(nvfs, "origin", 738 origin->zfs_dmustats.dds_guid)); 739 } 740 741 /* iterate over props */ 742 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); 743 send_iterate_prop(zhp, nv); 744 VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv)); 745 nvlist_free(nv); 746 747 /* iterate over snaps, and set sd->parent_fromsnap_guid */ 748 sd->parent_fromsnap_guid = 0; 749 VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0)); 750 VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0)); 751 (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd); 752 VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps)); 753 VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops)); 754 nvlist_free(sd->parent_snaps); 755 nvlist_free(sd->snapprops); 756 757 /* add this fs to nvlist */ 758 (void) snprintf(guidstring, sizeof (guidstring), 759 "0x%llx", (longlong_t)guid); 760 VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs)); 761 nvlist_free(nvfs); 762 763 /* iterate over children */ 764 if (sd->recursive) 765 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd); 766 767 sd->parent_fromsnap_guid = parent_fromsnap_guid_save; 768 769 zfs_close(zhp); 770 return (rv); 771} 772 773static int 774gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap, 775 const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp) 776{ 777 zfs_handle_t *zhp; 778 send_data_t sd = { 0 }; 779 int error; 780 781 zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 782 if (zhp == NULL) 783 return (EZFS_BADTYPE); 784 785 VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0)); 786 sd.fromsnap = fromsnap; 787 sd.tosnap = tosnap; 788 sd.recursive = recursive; 789 790 if ((error = send_iterate_fs(zhp, &sd)) != 0) { 791 nvlist_free(sd.fss); 792 if (avlp != NULL) 793 *avlp = NULL; 794 *nvlp = NULL; 795 return (error); 796 } 797 798 if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) { 799 nvlist_free(sd.fss); 800 *nvlp = NULL; 801 return (EZFS_NOMEM); 802 } 803 804 *nvlp = sd.fss; 805 return (0); 806} 807 808/* 809 * Routines specific to "zfs send" 810 */ 811typedef struct send_dump_data { 812 /* these are all just the short snapname (the part after the @) */ 813 const char *fromsnap; 814 const char *tosnap; 815 char prevsnap[ZFS_MAXNAMELEN]; 816 uint64_t prevsnap_obj; 817 boolean_t seenfrom, seento, replicate, doall, fromorigin; 818 boolean_t verbose, dryrun, parsable, progress, embed_data, large_block; 819 int outfd; 820 boolean_t err; 821 nvlist_t *fss; 822 nvlist_t *snapholds; 823 avl_tree_t *fsavl; 824 snapfilter_cb_t *filter_cb; 825 void *filter_cb_arg; 826 nvlist_t *debugnv; 827 char holdtag[ZFS_MAXNAMELEN]; 828 int cleanup_fd; 829 uint64_t size; 830} send_dump_data_t; 831 832static int 833estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj, 834 boolean_t fromorigin, uint64_t *sizep) 835{ 836 zfs_cmd_t zc = { 0 }; 837 libzfs_handle_t *hdl = zhp->zfs_hdl; 838 839 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 840 assert(fromsnap_obj == 0 || !fromorigin); 841 842 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 843 zc.zc_obj = fromorigin; 844 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 845 zc.zc_fromobj = fromsnap_obj; 846 zc.zc_guid = 1; /* estimate flag */ 847 848 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { 849 char errbuf[1024]; 850 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 851 "warning: cannot estimate space for '%s'"), zhp->zfs_name); 852 853 switch (errno) { 854 case EXDEV: 855 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 856 "not an earlier snapshot from the same fs")); 857 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 858 859 case ENOENT: 860 if (zfs_dataset_exists(hdl, zc.zc_name, 861 ZFS_TYPE_SNAPSHOT)) { 862 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 863 "incremental source (@%s) does not exist"), 864 zc.zc_value); 865 } 866 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 867 868 case EDQUOT: 869 case EFBIG: 870 case EIO: 871 case ENOLINK: 872 case ENOSPC: 873 case ENXIO: 874 case EPIPE: 875 case ERANGE: 876 case EFAULT: 877 case EROFS: 878 zfs_error_aux(hdl, strerror(errno)); 879 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 880 881 default: 882 return (zfs_standard_error(hdl, errno, errbuf)); 883 } 884 } 885 886 *sizep = zc.zc_objset_type; 887 888 return (0); 889} 890 891/* 892 * Dumps a backup of the given snapshot (incremental from fromsnap if it's not 893 * NULL) to the file descriptor specified by outfd. 894 */ 895static int 896dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj, 897 boolean_t fromorigin, int outfd, enum lzc_send_flags flags, 898 nvlist_t *debugnv) 899{ 900 zfs_cmd_t zc = { 0 }; 901 libzfs_handle_t *hdl = zhp->zfs_hdl; 902 nvlist_t *thisdbg; 903 904 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 905 assert(fromsnap_obj == 0 || !fromorigin); 906 907 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 908 zc.zc_cookie = outfd; 909 zc.zc_obj = fromorigin; 910 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 911 zc.zc_fromobj = fromsnap_obj; 912 zc.zc_flags = flags; 913 914 VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0)); 915 if (fromsnap && fromsnap[0] != '\0') { 916 VERIFY(0 == nvlist_add_string(thisdbg, 917 "fromsnap", fromsnap)); 918 } 919 920 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { 921 char errbuf[1024]; 922 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 923 "warning: cannot send '%s'"), zhp->zfs_name); 924 925 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno)); 926 if (debugnv) { 927 VERIFY(0 == nvlist_add_nvlist(debugnv, 928 zhp->zfs_name, thisdbg)); 929 } 930 nvlist_free(thisdbg); 931 932 switch (errno) { 933 case EXDEV: 934 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 935 "not an earlier snapshot from the same fs")); 936 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 937 938 case ENOENT: 939 if (zfs_dataset_exists(hdl, zc.zc_name, 940 ZFS_TYPE_SNAPSHOT)) { 941 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 942 "incremental source (@%s) does not exist"), 943 zc.zc_value); 944 } 945 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 946 947 case EDQUOT: 948 case EFBIG: 949 case EIO: 950 case ENOLINK: 951 case ENOSPC: 952#ifdef sun 953 case ENOSTR: 954#endif 955 case ENXIO: 956 case EPIPE: 957 case ERANGE: 958 case EFAULT: 959 case EROFS: 960 zfs_error_aux(hdl, strerror(errno)); 961 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 962 963 default: 964 return (zfs_standard_error(hdl, errno, errbuf)); 965 } 966 } 967 968 if (debugnv) 969 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg)); 970 nvlist_free(thisdbg); 971 972 return (0); 973} 974 975static void 976gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd) 977{ 978 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 979 980 /* 981 * zfs_send() only sets snapholds for sends that need them, 982 * e.g. replication and doall. 983 */ 984 if (sdd->snapholds == NULL) 985 return; 986 987 fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag); 988} 989 990static void * 991send_progress_thread(void *arg) 992{ 993 progress_arg_t *pa = arg; 994 995 zfs_cmd_t zc = { 0 }; 996 zfs_handle_t *zhp = pa->pa_zhp; 997 libzfs_handle_t *hdl = zhp->zfs_hdl; 998 unsigned long long bytes; 999 char buf[16]; 1000 1001 time_t t; 1002 struct tm *tm; 1003 1004 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 1005 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 1006 1007 if (!pa->pa_parsable) 1008 (void) fprintf(stderr, "TIME SENT SNAPSHOT\n"); 1009 1010 /* 1011 * Print the progress from ZFS_IOC_SEND_PROGRESS every second. 1012 */ 1013 for (;;) { 1014 (void) sleep(1); 1015 1016 zc.zc_cookie = pa->pa_fd; 1017 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0) 1018 return ((void *)-1); 1019 1020 (void) time(&t); 1021 tm = localtime(&t); 1022 bytes = zc.zc_cookie; 1023 1024 if (pa->pa_parsable) { 1025 (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n", 1026 tm->tm_hour, tm->tm_min, tm->tm_sec, 1027 bytes, zhp->zfs_name); 1028 } else { 1029 zfs_nicenum(bytes, buf, sizeof (buf)); 1030 (void) fprintf(stderr, "%02d:%02d:%02d %5s %s\n", 1031 tm->tm_hour, tm->tm_min, tm->tm_sec, 1032 buf, zhp->zfs_name); 1033 } 1034 } 1035} 1036 1037static int 1038dump_snapshot(zfs_handle_t *zhp, void *arg) 1039{ 1040 send_dump_data_t *sdd = arg; 1041 progress_arg_t pa = { 0 }; 1042 pthread_t tid; 1043 char *thissnap; 1044 int err; 1045 boolean_t isfromsnap, istosnap, fromorigin; 1046 boolean_t exclude = B_FALSE; 1047 1048 err = 0; 1049 thissnap = strchr(zhp->zfs_name, '@') + 1; 1050 isfromsnap = (sdd->fromsnap != NULL && 1051 strcmp(sdd->fromsnap, thissnap) == 0); 1052 1053 if (!sdd->seenfrom && isfromsnap) { 1054 gather_holds(zhp, sdd); 1055 sdd->seenfrom = B_TRUE; 1056 (void) strcpy(sdd->prevsnap, thissnap); 1057 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 1058 zfs_close(zhp); 1059 return (0); 1060 } 1061 1062 if (sdd->seento || !sdd->seenfrom) { 1063 zfs_close(zhp); 1064 return (0); 1065 } 1066 1067 istosnap = (strcmp(sdd->tosnap, thissnap) == 0); 1068 if (istosnap) 1069 sdd->seento = B_TRUE; 1070 1071 if (!sdd->doall && !isfromsnap && !istosnap) { 1072 if (sdd->replicate) { 1073 char *snapname; 1074 nvlist_t *snapprops; 1075 /* 1076 * Filter out all intermediate snapshots except origin 1077 * snapshots needed to replicate clones. 1078 */ 1079 nvlist_t *nvfs = fsavl_find(sdd->fsavl, 1080 zhp->zfs_dmustats.dds_guid, &snapname); 1081 1082 VERIFY(0 == nvlist_lookup_nvlist(nvfs, 1083 "snapprops", &snapprops)); 1084 VERIFY(0 == nvlist_lookup_nvlist(snapprops, 1085 thissnap, &snapprops)); 1086 exclude = !nvlist_exists(snapprops, "is_clone_origin"); 1087 } else { 1088 exclude = B_TRUE; 1089 } 1090 } 1091 1092 /* 1093 * If a filter function exists, call it to determine whether 1094 * this snapshot will be sent. 1095 */ 1096 if (exclude || (sdd->filter_cb != NULL && 1097 sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) { 1098 /* 1099 * This snapshot is filtered out. Don't send it, and don't 1100 * set prevsnap_obj, so it will be as if this snapshot didn't 1101 * exist, and the next accepted snapshot will be sent as 1102 * an incremental from the last accepted one, or as the 1103 * first (and full) snapshot in the case of a replication, 1104 * non-incremental send. 1105 */ 1106 zfs_close(zhp); 1107 return (0); 1108 } 1109 1110 gather_holds(zhp, sdd); 1111 fromorigin = sdd->prevsnap[0] == '\0' && 1112 (sdd->fromorigin || sdd->replicate); 1113 1114 if (sdd->verbose) { 1115 uint64_t size; 1116 err = estimate_ioctl(zhp, sdd->prevsnap_obj, 1117 fromorigin, &size); 1118 1119 if (sdd->parsable) { 1120 if (sdd->prevsnap[0] != '\0') { 1121 (void) fprintf(stderr, "incremental\t%s\t%s", 1122 sdd->prevsnap, zhp->zfs_name); 1123 } else { 1124 (void) fprintf(stderr, "full\t%s", 1125 zhp->zfs_name); 1126 } 1127 } else { 1128 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1129 "send from @%s to %s"), 1130 sdd->prevsnap, zhp->zfs_name); 1131 } 1132 if (err == 0) { 1133 if (sdd->parsable) { 1134 (void) fprintf(stderr, "\t%llu\n", 1135 (longlong_t)size); 1136 } else { 1137 char buf[16]; 1138 zfs_nicenum(size, buf, sizeof (buf)); 1139 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1140 " estimated size is %s\n"), buf); 1141 } 1142 sdd->size += size; 1143 } else { 1144 (void) fprintf(stderr, "\n"); 1145 } 1146 } 1147 1148 if (!sdd->dryrun) { 1149 /* 1150 * If progress reporting is requested, spawn a new thread to 1151 * poll ZFS_IOC_SEND_PROGRESS at a regular interval. 1152 */ 1153 if (sdd->progress) { 1154 pa.pa_zhp = zhp; 1155 pa.pa_fd = sdd->outfd; 1156 pa.pa_parsable = sdd->parsable; 1157 1158 if (err = pthread_create(&tid, NULL, 1159 send_progress_thread, &pa)) { 1160 zfs_close(zhp); 1161 return (err); 1162 } 1163 } 1164 1165 enum lzc_send_flags flags = 0; 1166 if (sdd->large_block) 1167 flags |= LZC_SEND_FLAG_LARGE_BLOCK; 1168 if (sdd->embed_data) 1169 flags |= LZC_SEND_FLAG_EMBED_DATA; 1170 1171 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj, 1172 fromorigin, sdd->outfd, flags, sdd->debugnv); 1173 1174 if (sdd->progress) { 1175 (void) pthread_cancel(tid); 1176 (void) pthread_join(tid, NULL); 1177 } 1178 } 1179 1180 (void) strcpy(sdd->prevsnap, thissnap); 1181 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 1182 zfs_close(zhp); 1183 return (err); 1184} 1185 1186static int 1187dump_filesystem(zfs_handle_t *zhp, void *arg) 1188{ 1189 int rv = 0; 1190 send_dump_data_t *sdd = arg; 1191 boolean_t missingfrom = B_FALSE; 1192 zfs_cmd_t zc = { 0 }; 1193 1194 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", 1195 zhp->zfs_name, sdd->tosnap); 1196 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) { 1197 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1198 "WARNING: could not send %s@%s: does not exist\n"), 1199 zhp->zfs_name, sdd->tosnap); 1200 sdd->err = B_TRUE; 1201 return (0); 1202 } 1203 1204 if (sdd->replicate && sdd->fromsnap) { 1205 /* 1206 * If this fs does not have fromsnap, and we're doing 1207 * recursive, we need to send a full stream from the 1208 * beginning (or an incremental from the origin if this 1209 * is a clone). If we're doing non-recursive, then let 1210 * them get the error. 1211 */ 1212 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", 1213 zhp->zfs_name, sdd->fromsnap); 1214 if (ioctl(zhp->zfs_hdl->libzfs_fd, 1215 ZFS_IOC_OBJSET_STATS, &zc) != 0) { 1216 missingfrom = B_TRUE; 1217 } 1218 } 1219 1220 sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0; 1221 sdd->prevsnap_obj = 0; 1222 if (sdd->fromsnap == NULL || missingfrom) 1223 sdd->seenfrom = B_TRUE; 1224 1225 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg); 1226 if (!sdd->seenfrom) { 1227 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1228 "WARNING: could not send %s@%s:\n" 1229 "incremental source (%s@%s) does not exist\n"), 1230 zhp->zfs_name, sdd->tosnap, 1231 zhp->zfs_name, sdd->fromsnap); 1232 sdd->err = B_TRUE; 1233 } else if (!sdd->seento) { 1234 if (sdd->fromsnap) { 1235 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1236 "WARNING: could not send %s@%s:\n" 1237 "incremental source (%s@%s) " 1238 "is not earlier than it\n"), 1239 zhp->zfs_name, sdd->tosnap, 1240 zhp->zfs_name, sdd->fromsnap); 1241 } else { 1242 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1243 "WARNING: " 1244 "could not send %s@%s: does not exist\n"), 1245 zhp->zfs_name, sdd->tosnap); 1246 } 1247 sdd->err = B_TRUE; 1248 } 1249 1250 return (rv); 1251} 1252 1253static int 1254dump_filesystems(zfs_handle_t *rzhp, void *arg) 1255{ 1256 send_dump_data_t *sdd = arg; 1257 nvpair_t *fspair; 1258 boolean_t needagain, progress; 1259 1260 if (!sdd->replicate) 1261 return (dump_filesystem(rzhp, sdd)); 1262 1263 /* Mark the clone origin snapshots. */ 1264 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1265 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1266 nvlist_t *nvfs; 1267 uint64_t origin_guid = 0; 1268 1269 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs)); 1270 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid); 1271 if (origin_guid != 0) { 1272 char *snapname; 1273 nvlist_t *origin_nv = fsavl_find(sdd->fsavl, 1274 origin_guid, &snapname); 1275 if (origin_nv != NULL) { 1276 nvlist_t *snapprops; 1277 VERIFY(0 == nvlist_lookup_nvlist(origin_nv, 1278 "snapprops", &snapprops)); 1279 VERIFY(0 == nvlist_lookup_nvlist(snapprops, 1280 snapname, &snapprops)); 1281 VERIFY(0 == nvlist_add_boolean( 1282 snapprops, "is_clone_origin")); 1283 } 1284 } 1285 } 1286again: 1287 needagain = progress = B_FALSE; 1288 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1289 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1290 nvlist_t *fslist, *parent_nv; 1291 char *fsname; 1292 zfs_handle_t *zhp; 1293 int err; 1294 uint64_t origin_guid = 0; 1295 uint64_t parent_guid = 0; 1296 1297 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); 1298 if (nvlist_lookup_boolean(fslist, "sent") == 0) 1299 continue; 1300 1301 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0); 1302 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid); 1303 (void) nvlist_lookup_uint64(fslist, "parentfromsnap", 1304 &parent_guid); 1305 1306 if (parent_guid != 0) { 1307 parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL); 1308 if (!nvlist_exists(parent_nv, "sent")) { 1309 /* parent has not been sent; skip this one */ 1310 needagain = B_TRUE; 1311 continue; 1312 } 1313 } 1314 1315 if (origin_guid != 0) { 1316 nvlist_t *origin_nv = fsavl_find(sdd->fsavl, 1317 origin_guid, NULL); 1318 if (origin_nv != NULL && 1319 !nvlist_exists(origin_nv, "sent")) { 1320 /* 1321 * origin has not been sent yet; 1322 * skip this clone. 1323 */ 1324 needagain = B_TRUE; 1325 continue; 1326 } 1327 } 1328 1329 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET); 1330 if (zhp == NULL) 1331 return (-1); 1332 err = dump_filesystem(zhp, sdd); 1333 VERIFY(nvlist_add_boolean(fslist, "sent") == 0); 1334 progress = B_TRUE; 1335 zfs_close(zhp); 1336 if (err) 1337 return (err); 1338 } 1339 if (needagain) { 1340 assert(progress); 1341 goto again; 1342 } 1343 1344 /* clean out the sent flags in case we reuse this fss */ 1345 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1346 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1347 nvlist_t *fslist; 1348 1349 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); 1350 (void) nvlist_remove_all(fslist, "sent"); 1351 } 1352 1353 return (0); 1354} 1355 1356/* 1357 * Generate a send stream for the dataset identified by the argument zhp. 1358 * 1359 * The content of the send stream is the snapshot identified by 1360 * 'tosnap'. Incremental streams are requested in two ways: 1361 * - from the snapshot identified by "fromsnap" (if non-null) or 1362 * - from the origin of the dataset identified by zhp, which must 1363 * be a clone. In this case, "fromsnap" is null and "fromorigin" 1364 * is TRUE. 1365 * 1366 * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and 1367 * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM) 1368 * if "replicate" is set. If "doall" is set, dump all the intermediate 1369 * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall" 1370 * case too. If "props" is set, send properties. 1371 */ 1372int 1373zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, 1374 sendflags_t *flags, int outfd, snapfilter_cb_t filter_func, 1375 void *cb_arg, nvlist_t **debugnvp) 1376{ 1377 char errbuf[1024]; 1378 send_dump_data_t sdd = { 0 }; 1379 int err = 0; 1380 nvlist_t *fss = NULL; 1381 avl_tree_t *fsavl = NULL; 1382 static uint64_t holdseq; 1383 int spa_version; 1384 pthread_t tid = 0; 1385 int pipefd[2]; 1386 dedup_arg_t dda = { 0 }; 1387 int featureflags = 0; 1388 1389 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1390 "cannot send '%s'"), zhp->zfs_name); 1391 1392 if (fromsnap && fromsnap[0] == '\0') { 1393 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN, 1394 "zero-length incremental source")); 1395 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf)); 1396 } 1397 1398 if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) { 1399 uint64_t version; 1400 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION); 1401 if (version >= ZPL_VERSION_SA) { 1402 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; 1403 } 1404 } 1405 1406 if (flags->dedup && !flags->dryrun) { 1407 featureflags |= (DMU_BACKUP_FEATURE_DEDUP | 1408 DMU_BACKUP_FEATURE_DEDUPPROPS); 1409 if (err = pipe(pipefd)) { 1410 zfs_error_aux(zhp->zfs_hdl, strerror(errno)); 1411 return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED, 1412 errbuf)); 1413 } 1414 dda.outputfd = outfd; 1415 dda.inputfd = pipefd[1]; 1416 dda.dedup_hdl = zhp->zfs_hdl; 1417 if (err = pthread_create(&tid, NULL, cksummer, &dda)) { 1418 (void) close(pipefd[0]); 1419 (void) close(pipefd[1]); 1420 zfs_error_aux(zhp->zfs_hdl, strerror(errno)); 1421 return (zfs_error(zhp->zfs_hdl, 1422 EZFS_THREADCREATEFAILED, errbuf)); 1423 } 1424 } 1425 1426 if (flags->replicate || flags->doall || flags->props) { 1427 dmu_replay_record_t drr = { 0 }; 1428 char *packbuf = NULL; 1429 size_t buflen = 0; 1430 zio_cksum_t zc = { 0 }; 1431 1432 if (flags->replicate || flags->props) { 1433 nvlist_t *hdrnv; 1434 1435 VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0)); 1436 if (fromsnap) { 1437 VERIFY(0 == nvlist_add_string(hdrnv, 1438 "fromsnap", fromsnap)); 1439 } 1440 VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap)); 1441 if (!flags->replicate) { 1442 VERIFY(0 == nvlist_add_boolean(hdrnv, 1443 "not_recursive")); 1444 } 1445 1446 err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name, 1447 fromsnap, tosnap, flags->replicate, &fss, &fsavl); 1448 if (err) 1449 goto err_out; 1450 VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss)); 1451 err = nvlist_pack(hdrnv, &packbuf, &buflen, 1452 NV_ENCODE_XDR, 0); 1453 if (debugnvp) 1454 *debugnvp = hdrnv; 1455 else 1456 nvlist_free(hdrnv); 1457 if (err) 1458 goto stderr_out; 1459 } 1460 1461 if (!flags->dryrun) { 1462 /* write first begin record */ 1463 drr.drr_type = DRR_BEGIN; 1464 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 1465 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin. 1466 drr_versioninfo, DMU_COMPOUNDSTREAM); 1467 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin. 1468 drr_versioninfo, featureflags); 1469 (void) snprintf(drr.drr_u.drr_begin.drr_toname, 1470 sizeof (drr.drr_u.drr_begin.drr_toname), 1471 "%s@%s", zhp->zfs_name, tosnap); 1472 drr.drr_payloadlen = buflen; 1473 err = cksum_and_write(&drr, sizeof (drr), &zc, outfd); 1474 1475 /* write header nvlist */ 1476 if (err != -1 && packbuf != NULL) { 1477 err = cksum_and_write(packbuf, buflen, &zc, 1478 outfd); 1479 } 1480 free(packbuf); 1481 if (err == -1) { 1482 err = errno; 1483 goto stderr_out; 1484 } 1485 1486 /* write end record */ 1487 bzero(&drr, sizeof (drr)); 1488 drr.drr_type = DRR_END; 1489 drr.drr_u.drr_end.drr_checksum = zc; 1490 err = write(outfd, &drr, sizeof (drr)); 1491 if (err == -1) { 1492 err = errno; 1493 goto stderr_out; 1494 } 1495 1496 err = 0; 1497 } 1498 } 1499 1500 /* dump each stream */ 1501 sdd.fromsnap = fromsnap; 1502 sdd.tosnap = tosnap; 1503 if (tid != 0) 1504 sdd.outfd = pipefd[0]; 1505 else 1506 sdd.outfd = outfd; 1507 sdd.replicate = flags->replicate; 1508 sdd.doall = flags->doall; 1509 sdd.fromorigin = flags->fromorigin; 1510 sdd.fss = fss; 1511 sdd.fsavl = fsavl; 1512 sdd.verbose = flags->verbose; 1513 sdd.parsable = flags->parsable; 1514 sdd.progress = flags->progress; 1515 sdd.dryrun = flags->dryrun; 1516 sdd.large_block = flags->largeblock; 1517 sdd.embed_data = flags->embed_data; 1518 sdd.filter_cb = filter_func; 1519 sdd.filter_cb_arg = cb_arg; 1520 if (debugnvp) 1521 sdd.debugnv = *debugnvp; 1522 1523 /* 1524 * Some flags require that we place user holds on the datasets that are 1525 * being sent so they don't get destroyed during the send. We can skip 1526 * this step if the pool is imported read-only since the datasets cannot 1527 * be destroyed. 1528 */ 1529 if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp), 1530 ZPOOL_PROP_READONLY, NULL) && 1531 zfs_spa_version(zhp, &spa_version) == 0 && 1532 spa_version >= SPA_VERSION_USERREFS && 1533 (flags->doall || flags->replicate)) { 1534 ++holdseq; 1535 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag), 1536 ".send-%d-%llu", getpid(), (u_longlong_t)holdseq); 1537 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); 1538 if (sdd.cleanup_fd < 0) { 1539 err = errno; 1540 goto stderr_out; 1541 } 1542 sdd.snapholds = fnvlist_alloc(); 1543 } else { 1544 sdd.cleanup_fd = -1; 1545 sdd.snapholds = NULL; 1546 } 1547 if (flags->verbose || sdd.snapholds != NULL) { 1548 /* 1549 * Do a verbose no-op dry run to get all the verbose output 1550 * or to gather snapshot hold's before generating any data, 1551 * then do a non-verbose real run to generate the streams. 1552 */ 1553 sdd.dryrun = B_TRUE; 1554 err = dump_filesystems(zhp, &sdd); 1555 1556 if (err != 0) 1557 goto stderr_out; 1558 1559 if (flags->verbose) { 1560 if (flags->parsable) { 1561 (void) fprintf(stderr, "size\t%llu\n", 1562 (longlong_t)sdd.size); 1563 } else { 1564 char buf[16]; 1565 zfs_nicenum(sdd.size, buf, sizeof (buf)); 1566 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1567 "total estimated size is %s\n"), buf); 1568 } 1569 } 1570 1571 /* Ensure no snaps found is treated as an error. */ 1572 if (!sdd.seento) { 1573 err = ENOENT; 1574 goto err_out; 1575 } 1576 1577 /* Skip the second run if dryrun was requested. */ 1578 if (flags->dryrun) 1579 goto err_out; 1580 1581 if (sdd.snapholds != NULL) { 1582 err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds); 1583 if (err != 0) 1584 goto stderr_out; 1585 1586 fnvlist_free(sdd.snapholds); 1587 sdd.snapholds = NULL; 1588 } 1589 1590 sdd.dryrun = B_FALSE; 1591 sdd.verbose = B_FALSE; 1592 } 1593 1594 err = dump_filesystems(zhp, &sdd); 1595 fsavl_destroy(fsavl); 1596 nvlist_free(fss); 1597 1598 /* Ensure no snaps found is treated as an error. */ 1599 if (err == 0 && !sdd.seento) 1600 err = ENOENT; 1601 1602 if (tid != 0) { 1603 if (err != 0) 1604 (void) pthread_cancel(tid); 1605 (void) close(pipefd[0]); 1606 (void) pthread_join(tid, NULL); 1607 } 1608 1609 if (sdd.cleanup_fd != -1) { 1610 VERIFY(0 == close(sdd.cleanup_fd)); 1611 sdd.cleanup_fd = -1; 1612 } 1613 1614 if (!flags->dryrun && (flags->replicate || flags->doall || 1615 flags->props)) { 1616 /* 1617 * write final end record. NB: want to do this even if 1618 * there was some error, because it might not be totally 1619 * failed. 1620 */ 1621 dmu_replay_record_t drr = { 0 }; 1622 drr.drr_type = DRR_END; 1623 if (write(outfd, &drr, sizeof (drr)) == -1) { 1624 return (zfs_standard_error(zhp->zfs_hdl, 1625 errno, errbuf)); 1626 } 1627 } 1628 1629 return (err || sdd.err); 1630 1631stderr_out: 1632 err = zfs_standard_error(zhp->zfs_hdl, err, errbuf); 1633err_out: 1634 fsavl_destroy(fsavl); 1635 nvlist_free(fss); 1636 fnvlist_free(sdd.snapholds); 1637 1638 if (sdd.cleanup_fd != -1) 1639 VERIFY(0 == close(sdd.cleanup_fd)); 1640 if (tid != 0) { 1641 (void) pthread_cancel(tid); 1642 (void) close(pipefd[0]); 1643 (void) pthread_join(tid, NULL); 1644 } 1645 return (err); 1646} 1647 1648int 1649zfs_send_one(zfs_handle_t *zhp, const char *from, int fd, 1650 enum lzc_send_flags flags) 1651{ 1652 int err; 1653 libzfs_handle_t *hdl = zhp->zfs_hdl; 1654 1655 char errbuf[1024]; 1656 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1657 "warning: cannot send '%s'"), zhp->zfs_name); 1658 1659 err = lzc_send(zhp->zfs_name, from, fd, flags); 1660 if (err != 0) { 1661 switch (errno) { 1662 case EXDEV: 1663 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1664 "not an earlier snapshot from the same fs")); 1665 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 1666 1667 case ENOENT: 1668 case ESRCH: 1669 if (lzc_exists(zhp->zfs_name)) { 1670 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1671 "incremental source (%s) does not exist"), 1672 from); 1673 } 1674 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 1675 1676 case EBUSY: 1677 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1678 "target is busy; if a filesystem, " 1679 "it must not be mounted")); 1680 return (zfs_error(hdl, EZFS_BUSY, errbuf)); 1681 1682 case EDQUOT: 1683 case EFBIG: 1684 case EIO: 1685 case ENOLINK: 1686 case ENOSPC: 1687#ifdef illumos 1688 case ENOSTR: 1689#endif 1690 case ENXIO: 1691 case EPIPE: 1692 case ERANGE: 1693 case EFAULT: 1694 case EROFS: 1695 zfs_error_aux(hdl, strerror(errno)); 1696 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 1697 1698 default: 1699 return (zfs_standard_error(hdl, errno, errbuf)); 1700 } 1701 } 1702 return (err != 0); 1703} 1704 1705/* 1706 * Routines specific to "zfs recv" 1707 */ 1708 1709static int 1710recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen, 1711 boolean_t byteswap, zio_cksum_t *zc) 1712{ 1713 char *cp = buf; 1714 int rv; 1715 int len = ilen; 1716 1717 do { 1718 rv = read(fd, cp, len); 1719 cp += rv; 1720 len -= rv; 1721 } while (rv > 0); 1722 1723 if (rv < 0 || len != 0) { 1724 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1725 "failed to read from stream")); 1726 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN, 1727 "cannot receive"))); 1728 } 1729 1730 if (zc) { 1731 if (byteswap) 1732 fletcher_4_incremental_byteswap(buf, ilen, zc); 1733 else 1734 fletcher_4_incremental_native(buf, ilen, zc); 1735 } 1736 return (0); 1737} 1738 1739static int 1740recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp, 1741 boolean_t byteswap, zio_cksum_t *zc) 1742{ 1743 char *buf; 1744 int err; 1745 1746 buf = zfs_alloc(hdl, len); 1747 if (buf == NULL) 1748 return (ENOMEM); 1749 1750 err = recv_read(hdl, fd, buf, len, byteswap, zc); 1751 if (err != 0) { 1752 free(buf); 1753 return (err); 1754 } 1755 1756 err = nvlist_unpack(buf, len, nvp, 0); 1757 free(buf); 1758 if (err != 0) { 1759 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 1760 "stream (malformed nvlist)")); 1761 return (EINVAL); 1762 } 1763 return (0); 1764} 1765 1766static int 1767recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname, 1768 int baselen, char *newname, recvflags_t *flags) 1769{ 1770 static int seq; 1771 zfs_cmd_t zc = { 0 }; 1772 int err; 1773 prop_changelist_t *clp; 1774 zfs_handle_t *zhp; 1775 1776 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 1777 if (zhp == NULL) 1778 return (-1); 1779 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 1780 flags->force ? MS_FORCE : 0); 1781 zfs_close(zhp); 1782 if (clp == NULL) 1783 return (-1); 1784 err = changelist_prefix(clp); 1785 if (err) 1786 return (err); 1787 1788 zc.zc_objset_type = DMU_OST_ZFS; 1789 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 1790 1791 if (tryname) { 1792 (void) strcpy(newname, tryname); 1793 1794 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value)); 1795 1796 if (flags->verbose) { 1797 (void) printf("attempting rename %s to %s\n", 1798 zc.zc_name, zc.zc_value); 1799 } 1800 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc); 1801 if (err == 0) 1802 changelist_rename(clp, name, tryname); 1803 } else { 1804 err = ENOENT; 1805 } 1806 1807 if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) { 1808 seq++; 1809 1810 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u", 1811 baselen, name, getpid(), seq); 1812 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value)); 1813 1814 if (flags->verbose) { 1815 (void) printf("failed - trying rename %s to %s\n", 1816 zc.zc_name, zc.zc_value); 1817 } 1818 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc); 1819 if (err == 0) 1820 changelist_rename(clp, name, newname); 1821 if (err && flags->verbose) { 1822 (void) printf("failed (%u) - " 1823 "will try again on next pass\n", errno); 1824 } 1825 err = EAGAIN; 1826 } else if (flags->verbose) { 1827 if (err == 0) 1828 (void) printf("success\n"); 1829 else 1830 (void) printf("failed (%u)\n", errno); 1831 } 1832 1833 (void) changelist_postfix(clp); 1834 changelist_free(clp); 1835 1836 return (err); 1837} 1838 1839static int 1840recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen, 1841 char *newname, recvflags_t *flags) 1842{ 1843 zfs_cmd_t zc = { 0 }; 1844 int err = 0; 1845 prop_changelist_t *clp; 1846 zfs_handle_t *zhp; 1847 boolean_t defer = B_FALSE; 1848 int spa_version; 1849 1850 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 1851 if (zhp == NULL) 1852 return (-1); 1853 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 1854 flags->force ? MS_FORCE : 0); 1855 if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT && 1856 zfs_spa_version(zhp, &spa_version) == 0 && 1857 spa_version >= SPA_VERSION_USERREFS) 1858 defer = B_TRUE; 1859 zfs_close(zhp); 1860 if (clp == NULL) 1861 return (-1); 1862 err = changelist_prefix(clp); 1863 if (err) 1864 return (err); 1865 1866 zc.zc_objset_type = DMU_OST_ZFS; 1867 zc.zc_defer_destroy = defer; 1868 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 1869 1870 if (flags->verbose) 1871 (void) printf("attempting destroy %s\n", zc.zc_name); 1872 err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc); 1873 if (err == 0) { 1874 if (flags->verbose) 1875 (void) printf("success\n"); 1876 changelist_remove(clp, zc.zc_name); 1877 } 1878 1879 (void) changelist_postfix(clp); 1880 changelist_free(clp); 1881 1882 /* 1883 * Deferred destroy might destroy the snapshot or only mark it to be 1884 * destroyed later, and it returns success in either case. 1885 */ 1886 if (err != 0 || (defer && zfs_dataset_exists(hdl, name, 1887 ZFS_TYPE_SNAPSHOT))) { 1888 err = recv_rename(hdl, name, NULL, baselen, newname, flags); 1889 } 1890 1891 return (err); 1892} 1893 1894typedef struct guid_to_name_data { 1895 uint64_t guid; 1896 char *name; 1897 char *skip; 1898} guid_to_name_data_t; 1899 1900static int 1901guid_to_name_cb(zfs_handle_t *zhp, void *arg) 1902{ 1903 guid_to_name_data_t *gtnd = arg; 1904 int err; 1905 1906 if (gtnd->skip != NULL && 1907 strcmp(zhp->zfs_name, gtnd->skip) == 0) { 1908 return (0); 1909 } 1910 1911 if (zhp->zfs_dmustats.dds_guid == gtnd->guid) { 1912 (void) strcpy(gtnd->name, zhp->zfs_name); 1913 zfs_close(zhp); 1914 return (EEXIST); 1915 } 1916 1917 err = zfs_iter_children(zhp, guid_to_name_cb, gtnd); 1918 zfs_close(zhp); 1919 return (err); 1920} 1921 1922/* 1923 * Attempt to find the local dataset associated with this guid. In the case of 1924 * multiple matches, we attempt to find the "best" match by searching 1925 * progressively larger portions of the hierarchy. This allows one to send a 1926 * tree of datasets individually and guarantee that we will find the source 1927 * guid within that hierarchy, even if there are multiple matches elsewhere. 1928 */ 1929static int 1930guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid, 1931 char *name) 1932{ 1933 /* exhaustive search all local snapshots */ 1934 char pname[ZFS_MAXNAMELEN]; 1935 guid_to_name_data_t gtnd; 1936 int err = 0; 1937 zfs_handle_t *zhp; 1938 char *cp; 1939 1940 gtnd.guid = guid; 1941 gtnd.name = name; 1942 gtnd.skip = NULL; 1943 1944 (void) strlcpy(pname, parent, sizeof (pname)); 1945 1946 /* 1947 * Search progressively larger portions of the hierarchy. This will 1948 * select the "most local" version of the origin snapshot in the case 1949 * that there are multiple matching snapshots in the system. 1950 */ 1951 while ((cp = strrchr(pname, '/')) != NULL) { 1952 1953 /* Chop off the last component and open the parent */ 1954 *cp = '\0'; 1955 zhp = make_dataset_handle(hdl, pname); 1956 1957 if (zhp == NULL) 1958 continue; 1959 1960 err = zfs_iter_children(zhp, guid_to_name_cb, >nd); 1961 zfs_close(zhp); 1962 if (err == EEXIST) 1963 return (0); 1964 1965 /* 1966 * Remember the dataset that we already searched, so we 1967 * skip it next time through. 1968 */ 1969 gtnd.skip = pname; 1970 } 1971 1972 return (ENOENT); 1973} 1974 1975/* 1976 * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if 1977 * guid1 is after guid2. 1978 */ 1979static int 1980created_before(libzfs_handle_t *hdl, avl_tree_t *avl, 1981 uint64_t guid1, uint64_t guid2) 1982{ 1983 nvlist_t *nvfs; 1984 char *fsname, *snapname; 1985 char buf[ZFS_MAXNAMELEN]; 1986 int rv; 1987 zfs_handle_t *guid1hdl, *guid2hdl; 1988 uint64_t create1, create2; 1989 1990 if (guid2 == 0) 1991 return (0); 1992 if (guid1 == 0) 1993 return (1); 1994 1995 nvfs = fsavl_find(avl, guid1, &snapname); 1996 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 1997 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); 1998 guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); 1999 if (guid1hdl == NULL) 2000 return (-1); 2001 2002 nvfs = fsavl_find(avl, guid2, &snapname); 2003 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2004 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); 2005 guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); 2006 if (guid2hdl == NULL) { 2007 zfs_close(guid1hdl); 2008 return (-1); 2009 } 2010 2011 create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG); 2012 create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG); 2013 2014 if (create1 < create2) 2015 rv = -1; 2016 else if (create1 > create2) 2017 rv = +1; 2018 else 2019 rv = 0; 2020 2021 zfs_close(guid1hdl); 2022 zfs_close(guid2hdl); 2023 2024 return (rv); 2025} 2026 2027static int 2028recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs, 2029 recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl, 2030 nvlist_t *renamed) 2031{ 2032 nvlist_t *local_nv, *deleted = NULL; 2033 avl_tree_t *local_avl; 2034 nvpair_t *fselem, *nextfselem; 2035 char *fromsnap; 2036 char newname[ZFS_MAXNAMELEN]; 2037 char guidname[32]; 2038 int error; 2039 boolean_t needagain, progress, recursive; 2040 char *s1, *s2; 2041 2042 VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap)); 2043 2044 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2045 ENOENT); 2046 2047 if (flags->dryrun) 2048 return (0); 2049 2050again: 2051 needagain = progress = B_FALSE; 2052 2053 VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0)); 2054 2055 if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL, 2056 recursive, &local_nv, &local_avl)) != 0) 2057 return (error); 2058 2059 /* 2060 * Process deletes and renames 2061 */ 2062 for (fselem = nvlist_next_nvpair(local_nv, NULL); 2063 fselem; fselem = nextfselem) { 2064 nvlist_t *nvfs, *snaps; 2065 nvlist_t *stream_nvfs = NULL; 2066 nvpair_t *snapelem, *nextsnapelem; 2067 uint64_t fromguid = 0; 2068 uint64_t originguid = 0; 2069 uint64_t stream_originguid = 0; 2070 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid; 2071 char *fsname, *stream_fsname; 2072 2073 nextfselem = nvlist_next_nvpair(local_nv, fselem); 2074 2075 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); 2076 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); 2077 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2078 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap", 2079 &parent_fromsnap_guid)); 2080 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid); 2081 2082 /* 2083 * First find the stream's fs, so we can check for 2084 * a different origin (due to "zfs promote") 2085 */ 2086 for (snapelem = nvlist_next_nvpair(snaps, NULL); 2087 snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) { 2088 uint64_t thisguid; 2089 2090 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); 2091 stream_nvfs = fsavl_find(stream_avl, thisguid, NULL); 2092 2093 if (stream_nvfs != NULL) 2094 break; 2095 } 2096 2097 /* check for promote */ 2098 (void) nvlist_lookup_uint64(stream_nvfs, "origin", 2099 &stream_originguid); 2100 if (stream_nvfs && originguid != stream_originguid) { 2101 switch (created_before(hdl, local_avl, 2102 stream_originguid, originguid)) { 2103 case 1: { 2104 /* promote it! */ 2105 zfs_cmd_t zc = { 0 }; 2106 nvlist_t *origin_nvfs; 2107 char *origin_fsname; 2108 2109 if (flags->verbose) 2110 (void) printf("promoting %s\n", fsname); 2111 2112 origin_nvfs = fsavl_find(local_avl, originguid, 2113 NULL); 2114 VERIFY(0 == nvlist_lookup_string(origin_nvfs, 2115 "name", &origin_fsname)); 2116 (void) strlcpy(zc.zc_value, origin_fsname, 2117 sizeof (zc.zc_value)); 2118 (void) strlcpy(zc.zc_name, fsname, 2119 sizeof (zc.zc_name)); 2120 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc); 2121 if (error == 0) 2122 progress = B_TRUE; 2123 break; 2124 } 2125 default: 2126 break; 2127 case -1: 2128 fsavl_destroy(local_avl); 2129 nvlist_free(local_nv); 2130 return (-1); 2131 } 2132 /* 2133 * We had/have the wrong origin, therefore our 2134 * list of snapshots is wrong. Need to handle 2135 * them on the next pass. 2136 */ 2137 needagain = B_TRUE; 2138 continue; 2139 } 2140 2141 for (snapelem = nvlist_next_nvpair(snaps, NULL); 2142 snapelem; snapelem = nextsnapelem) { 2143 uint64_t thisguid; 2144 char *stream_snapname; 2145 nvlist_t *found, *props; 2146 2147 nextsnapelem = nvlist_next_nvpair(snaps, snapelem); 2148 2149 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); 2150 found = fsavl_find(stream_avl, thisguid, 2151 &stream_snapname); 2152 2153 /* check for delete */ 2154 if (found == NULL) { 2155 char name[ZFS_MAXNAMELEN]; 2156 2157 if (!flags->force) 2158 continue; 2159 2160 (void) snprintf(name, sizeof (name), "%s@%s", 2161 fsname, nvpair_name(snapelem)); 2162 2163 error = recv_destroy(hdl, name, 2164 strlen(fsname)+1, newname, flags); 2165 if (error) 2166 needagain = B_TRUE; 2167 else 2168 progress = B_TRUE; 2169 sprintf(guidname, "%lu", thisguid); 2170 nvlist_add_boolean(deleted, guidname); 2171 continue; 2172 } 2173 2174 stream_nvfs = found; 2175 2176 if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops", 2177 &props) && 0 == nvlist_lookup_nvlist(props, 2178 stream_snapname, &props)) { 2179 zfs_cmd_t zc = { 0 }; 2180 2181 zc.zc_cookie = B_TRUE; /* received */ 2182 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), 2183 "%s@%s", fsname, nvpair_name(snapelem)); 2184 if (zcmd_write_src_nvlist(hdl, &zc, 2185 props) == 0) { 2186 (void) zfs_ioctl(hdl, 2187 ZFS_IOC_SET_PROP, &zc); 2188 zcmd_free_nvlists(&zc); 2189 } 2190 } 2191 2192 /* check for different snapname */ 2193 if (strcmp(nvpair_name(snapelem), 2194 stream_snapname) != 0) { 2195 char name[ZFS_MAXNAMELEN]; 2196 char tryname[ZFS_MAXNAMELEN]; 2197 2198 (void) snprintf(name, sizeof (name), "%s@%s", 2199 fsname, nvpair_name(snapelem)); 2200 (void) snprintf(tryname, sizeof (name), "%s@%s", 2201 fsname, stream_snapname); 2202 2203 error = recv_rename(hdl, name, tryname, 2204 strlen(fsname)+1, newname, flags); 2205 if (error) 2206 needagain = B_TRUE; 2207 else 2208 progress = B_TRUE; 2209 } 2210 2211 if (strcmp(stream_snapname, fromsnap) == 0) 2212 fromguid = thisguid; 2213 } 2214 2215 /* check for delete */ 2216 if (stream_nvfs == NULL) { 2217 if (!flags->force) 2218 continue; 2219 2220 error = recv_destroy(hdl, fsname, strlen(tofs)+1, 2221 newname, flags); 2222 if (error) 2223 needagain = B_TRUE; 2224 else 2225 progress = B_TRUE; 2226 sprintf(guidname, "%lu", parent_fromsnap_guid); 2227 nvlist_add_boolean(deleted, guidname); 2228 continue; 2229 } 2230 2231 if (fromguid == 0) { 2232 if (flags->verbose) { 2233 (void) printf("local fs %s does not have " 2234 "fromsnap (%s in stream); must have " 2235 "been deleted locally; ignoring\n", 2236 fsname, fromsnap); 2237 } 2238 continue; 2239 } 2240 2241 VERIFY(0 == nvlist_lookup_string(stream_nvfs, 2242 "name", &stream_fsname)); 2243 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs, 2244 "parentfromsnap", &stream_parent_fromsnap_guid)); 2245 2246 s1 = strrchr(fsname, '/'); 2247 s2 = strrchr(stream_fsname, '/'); 2248 2249 /* 2250 * Check if we're going to rename based on parent guid change 2251 * and the current parent guid was also deleted. If it was then 2252 * rename will fail and is likely unneeded, so avoid this and 2253 * force an early retry to determine the new 2254 * parent_fromsnap_guid. 2255 */ 2256 if (stream_parent_fromsnap_guid != 0 && 2257 parent_fromsnap_guid != 0 && 2258 stream_parent_fromsnap_guid != parent_fromsnap_guid) { 2259 sprintf(guidname, "%lu", parent_fromsnap_guid); 2260 if (nvlist_exists(deleted, guidname)) { 2261 progress = B_TRUE; 2262 needagain = B_TRUE; 2263 goto doagain; 2264 } 2265 } 2266 2267 /* 2268 * Check for rename. If the exact receive path is specified, it 2269 * does not count as a rename, but we still need to check the 2270 * datasets beneath it. 2271 */ 2272 if ((stream_parent_fromsnap_guid != 0 && 2273 parent_fromsnap_guid != 0 && 2274 stream_parent_fromsnap_guid != parent_fromsnap_guid) || 2275 ((flags->isprefix || strcmp(tofs, fsname) != 0) && 2276 (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) { 2277 nvlist_t *parent; 2278 char tryname[ZFS_MAXNAMELEN]; 2279 2280 parent = fsavl_find(local_avl, 2281 stream_parent_fromsnap_guid, NULL); 2282 /* 2283 * NB: parent might not be found if we used the 2284 * tosnap for stream_parent_fromsnap_guid, 2285 * because the parent is a newly-created fs; 2286 * we'll be able to rename it after we recv the 2287 * new fs. 2288 */ 2289 if (parent != NULL) { 2290 char *pname; 2291 2292 VERIFY(0 == nvlist_lookup_string(parent, "name", 2293 &pname)); 2294 (void) snprintf(tryname, sizeof (tryname), 2295 "%s%s", pname, strrchr(stream_fsname, '/')); 2296 } else { 2297 tryname[0] = '\0'; 2298 if (flags->verbose) { 2299 (void) printf("local fs %s new parent " 2300 "not found\n", fsname); 2301 } 2302 } 2303 2304 newname[0] = '\0'; 2305 2306 error = recv_rename(hdl, fsname, tryname, 2307 strlen(tofs)+1, newname, flags); 2308 2309 if (renamed != NULL && newname[0] != '\0') { 2310 VERIFY(0 == nvlist_add_boolean(renamed, 2311 newname)); 2312 } 2313 2314 if (error) 2315 needagain = B_TRUE; 2316 else 2317 progress = B_TRUE; 2318 } 2319 } 2320 2321doagain: 2322 fsavl_destroy(local_avl); 2323 nvlist_free(local_nv); 2324 nvlist_free(deleted); 2325 2326 if (needagain && progress) { 2327 /* do another pass to fix up temporary names */ 2328 if (flags->verbose) 2329 (void) printf("another pass:\n"); 2330 goto again; 2331 } 2332 2333 return (needagain); 2334} 2335 2336static int 2337zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname, 2338 recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc, 2339 char **top_zfs, int cleanup_fd, uint64_t *action_handlep) 2340{ 2341 nvlist_t *stream_nv = NULL; 2342 avl_tree_t *stream_avl = NULL; 2343 char *fromsnap = NULL; 2344 char *cp; 2345 char tofs[ZFS_MAXNAMELEN]; 2346 char sendfs[ZFS_MAXNAMELEN]; 2347 char errbuf[1024]; 2348 dmu_replay_record_t drre; 2349 int error; 2350 boolean_t anyerr = B_FALSE; 2351 boolean_t softerr = B_FALSE; 2352 boolean_t recursive; 2353 2354 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2355 "cannot receive")); 2356 2357 assert(drr->drr_type == DRR_BEGIN); 2358 assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC); 2359 assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) == 2360 DMU_COMPOUNDSTREAM); 2361 2362 /* 2363 * Read in the nvlist from the stream. 2364 */ 2365 if (drr->drr_payloadlen != 0) { 2366 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen, 2367 &stream_nv, flags->byteswap, zc); 2368 if (error) { 2369 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2370 goto out; 2371 } 2372 } 2373 2374 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2375 ENOENT); 2376 2377 if (recursive && strchr(destname, '@')) { 2378 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2379 "cannot specify snapshot name for multi-snapshot stream")); 2380 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2381 goto out; 2382 } 2383 2384 /* 2385 * Read in the end record and verify checksum. 2386 */ 2387 if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre), 2388 flags->byteswap, NULL))) 2389 goto out; 2390 if (flags->byteswap) { 2391 drre.drr_type = BSWAP_32(drre.drr_type); 2392 drre.drr_u.drr_end.drr_checksum.zc_word[0] = 2393 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]); 2394 drre.drr_u.drr_end.drr_checksum.zc_word[1] = 2395 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]); 2396 drre.drr_u.drr_end.drr_checksum.zc_word[2] = 2397 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]); 2398 drre.drr_u.drr_end.drr_checksum.zc_word[3] = 2399 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]); 2400 } 2401 if (drre.drr_type != DRR_END) { 2402 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2403 goto out; 2404 } 2405 if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) { 2406 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2407 "incorrect header checksum")); 2408 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2409 goto out; 2410 } 2411 2412 (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap); 2413 2414 if (drr->drr_payloadlen != 0) { 2415 nvlist_t *stream_fss; 2416 2417 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss", 2418 &stream_fss)); 2419 if ((stream_avl = fsavl_create(stream_fss)) == NULL) { 2420 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2421 "couldn't allocate avl tree")); 2422 error = zfs_error(hdl, EZFS_NOMEM, errbuf); 2423 goto out; 2424 } 2425 2426 if (fromsnap != NULL) { 2427 nvlist_t *renamed = NULL; 2428 nvpair_t *pair = NULL; 2429 2430 (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN); 2431 if (flags->isprefix) { 2432 struct drr_begin *drrb = &drr->drr_u.drr_begin; 2433 int i; 2434 2435 if (flags->istail) { 2436 cp = strrchr(drrb->drr_toname, '/'); 2437 if (cp == NULL) { 2438 (void) strlcat(tofs, "/", 2439 ZFS_MAXNAMELEN); 2440 i = 0; 2441 } else { 2442 i = (cp - drrb->drr_toname); 2443 } 2444 } else { 2445 i = strcspn(drrb->drr_toname, "/@"); 2446 } 2447 /* zfs_receive_one() will create_parents() */ 2448 (void) strlcat(tofs, &drrb->drr_toname[i], 2449 ZFS_MAXNAMELEN); 2450 *strchr(tofs, '@') = '\0'; 2451 } 2452 2453 if (recursive && !flags->dryrun && !flags->nomount) { 2454 VERIFY(0 == nvlist_alloc(&renamed, 2455 NV_UNIQUE_NAME, 0)); 2456 } 2457 2458 softerr = recv_incremental_replication(hdl, tofs, flags, 2459 stream_nv, stream_avl, renamed); 2460 2461 /* Unmount renamed filesystems before receiving. */ 2462 while ((pair = nvlist_next_nvpair(renamed, 2463 pair)) != NULL) { 2464 zfs_handle_t *zhp; 2465 prop_changelist_t *clp = NULL; 2466 2467 zhp = zfs_open(hdl, nvpair_name(pair), 2468 ZFS_TYPE_FILESYSTEM); 2469 if (zhp != NULL) { 2470 clp = changelist_gather(zhp, 2471 ZFS_PROP_MOUNTPOINT, 0, 0); 2472 zfs_close(zhp); 2473 if (clp != NULL) { 2474 softerr |= 2475 changelist_prefix(clp); 2476 changelist_free(clp); 2477 } 2478 } 2479 } 2480 2481 nvlist_free(renamed); 2482 } 2483 } 2484 2485 /* 2486 * Get the fs specified by the first path in the stream (the top level 2487 * specified by 'zfs send') and pass it to each invocation of 2488 * zfs_receive_one(). 2489 */ 2490 (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname, 2491 ZFS_MAXNAMELEN); 2492 if ((cp = strchr(sendfs, '@')) != NULL) 2493 *cp = '\0'; 2494 2495 /* Finally, receive each contained stream */ 2496 do { 2497 /* 2498 * we should figure out if it has a recoverable 2499 * error, in which case do a recv_skip() and drive on. 2500 * Note, if we fail due to already having this guid, 2501 * zfs_receive_one() will take care of it (ie, 2502 * recv_skip() and return 0). 2503 */ 2504 error = zfs_receive_impl(hdl, destname, flags, fd, 2505 sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd, 2506 action_handlep); 2507 if (error == ENODATA) { 2508 error = 0; 2509 break; 2510 } 2511 anyerr |= error; 2512 } while (error == 0); 2513 2514 if (drr->drr_payloadlen != 0 && fromsnap != NULL) { 2515 /* 2516 * Now that we have the fs's they sent us, try the 2517 * renames again. 2518 */ 2519 softerr = recv_incremental_replication(hdl, tofs, flags, 2520 stream_nv, stream_avl, NULL); 2521 } 2522 2523out: 2524 fsavl_destroy(stream_avl); 2525 if (stream_nv) 2526 nvlist_free(stream_nv); 2527 if (softerr) 2528 error = -2; 2529 if (anyerr) 2530 error = -1; 2531 return (error); 2532} 2533 2534static void 2535trunc_prop_errs(int truncated) 2536{ 2537 ASSERT(truncated != 0); 2538 2539 if (truncated == 1) 2540 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 2541 "1 more property could not be set\n")); 2542 else 2543 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 2544 "%d more properties could not be set\n"), truncated); 2545} 2546 2547static int 2548recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap) 2549{ 2550 dmu_replay_record_t *drr; 2551 void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE); 2552 char errbuf[1024]; 2553 2554 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2555 "cannot receive:")); 2556 2557 /* XXX would be great to use lseek if possible... */ 2558 drr = buf; 2559 2560 while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t), 2561 byteswap, NULL) == 0) { 2562 if (byteswap) 2563 drr->drr_type = BSWAP_32(drr->drr_type); 2564 2565 switch (drr->drr_type) { 2566 case DRR_BEGIN: 2567 /* NB: not to be used on v2 stream packages */ 2568 if (drr->drr_payloadlen != 0) { 2569 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2570 "invalid substream header")); 2571 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 2572 } 2573 break; 2574 2575 case DRR_END: 2576 free(buf); 2577 return (0); 2578 2579 case DRR_OBJECT: 2580 if (byteswap) { 2581 drr->drr_u.drr_object.drr_bonuslen = 2582 BSWAP_32(drr->drr_u.drr_object. 2583 drr_bonuslen); 2584 } 2585 (void) recv_read(hdl, fd, buf, 2586 P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8), 2587 B_FALSE, NULL); 2588 break; 2589 2590 case DRR_WRITE: 2591 if (byteswap) { 2592 drr->drr_u.drr_write.drr_length = 2593 BSWAP_64(drr->drr_u.drr_write.drr_length); 2594 } 2595 (void) recv_read(hdl, fd, buf, 2596 drr->drr_u.drr_write.drr_length, B_FALSE, NULL); 2597 break; 2598 case DRR_SPILL: 2599 if (byteswap) { 2600 drr->drr_u.drr_write.drr_length = 2601 BSWAP_64(drr->drr_u.drr_spill.drr_length); 2602 } 2603 (void) recv_read(hdl, fd, buf, 2604 drr->drr_u.drr_spill.drr_length, B_FALSE, NULL); 2605 break; 2606 case DRR_WRITE_EMBEDDED: 2607 if (byteswap) { 2608 drr->drr_u.drr_write_embedded.drr_psize = 2609 BSWAP_32(drr->drr_u.drr_write_embedded. 2610 drr_psize); 2611 } 2612 (void) recv_read(hdl, fd, buf, 2613 P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize, 2614 8), B_FALSE, NULL); 2615 break; 2616 case DRR_WRITE_BYREF: 2617 case DRR_FREEOBJECTS: 2618 case DRR_FREE: 2619 break; 2620 2621 default: 2622 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2623 "invalid record type")); 2624 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 2625 } 2626 } 2627 2628 free(buf); 2629 return (-1); 2630} 2631 2632/* 2633 * Restores a backup of tosnap from the file descriptor specified by infd. 2634 */ 2635static int 2636zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, 2637 recvflags_t *flags, dmu_replay_record_t *drr, 2638 dmu_replay_record_t *drr_noswap, const char *sendfs, 2639 nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd, 2640 uint64_t *action_handlep) 2641{ 2642 zfs_cmd_t zc = { 0 }; 2643 time_t begin_time; 2644 int ioctl_err, ioctl_errno, err; 2645 char *cp; 2646 struct drr_begin *drrb = &drr->drr_u.drr_begin; 2647 char errbuf[1024]; 2648 char prop_errbuf[1024]; 2649 const char *chopprefix; 2650 boolean_t newfs = B_FALSE; 2651 boolean_t stream_wantsnewfs; 2652 uint64_t parent_snapguid = 0; 2653 prop_changelist_t *clp = NULL; 2654 nvlist_t *snapprops_nvlist = NULL; 2655 zprop_errflags_t prop_errflags; 2656 boolean_t recursive; 2657 2658 begin_time = time(NULL); 2659 2660 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2661 "cannot receive")); 2662 2663 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2664 ENOENT); 2665 2666 if (stream_avl != NULL) { 2667 char *snapname; 2668 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid, 2669 &snapname); 2670 nvlist_t *props; 2671 int ret; 2672 2673 (void) nvlist_lookup_uint64(fs, "parentfromsnap", 2674 &parent_snapguid); 2675 err = nvlist_lookup_nvlist(fs, "props", &props); 2676 if (err) 2677 VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0)); 2678 2679 if (flags->canmountoff) { 2680 VERIFY(0 == nvlist_add_uint64(props, 2681 zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0)); 2682 } 2683 ret = zcmd_write_src_nvlist(hdl, &zc, props); 2684 if (err) 2685 nvlist_free(props); 2686 2687 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) { 2688 VERIFY(0 == nvlist_lookup_nvlist(props, 2689 snapname, &snapprops_nvlist)); 2690 } 2691 2692 if (ret != 0) 2693 return (-1); 2694 } 2695 2696 cp = NULL; 2697 2698 /* 2699 * Determine how much of the snapshot name stored in the stream 2700 * we are going to tack on to the name they specified on the 2701 * command line, and how much we are going to chop off. 2702 * 2703 * If they specified a snapshot, chop the entire name stored in 2704 * the stream. 2705 */ 2706 if (flags->istail) { 2707 /* 2708 * A filesystem was specified with -e. We want to tack on only 2709 * the tail of the sent snapshot path. 2710 */ 2711 if (strchr(tosnap, '@')) { 2712 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 2713 "argument - snapshot not allowed with -e")); 2714 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 2715 } 2716 2717 chopprefix = strrchr(sendfs, '/'); 2718 2719 if (chopprefix == NULL) { 2720 /* 2721 * The tail is the poolname, so we need to 2722 * prepend a path separator. 2723 */ 2724 int len = strlen(drrb->drr_toname); 2725 cp = malloc(len + 2); 2726 cp[0] = '/'; 2727 (void) strcpy(&cp[1], drrb->drr_toname); 2728 chopprefix = cp; 2729 } else { 2730 chopprefix = drrb->drr_toname + (chopprefix - sendfs); 2731 } 2732 } else if (flags->isprefix) { 2733 /* 2734 * A filesystem was specified with -d. We want to tack on 2735 * everything but the first element of the sent snapshot path 2736 * (all but the pool name). 2737 */ 2738 if (strchr(tosnap, '@')) { 2739 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 2740 "argument - snapshot not allowed with -d")); 2741 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 2742 } 2743 2744 chopprefix = strchr(drrb->drr_toname, '/'); 2745 if (chopprefix == NULL) 2746 chopprefix = strchr(drrb->drr_toname, '@'); 2747 } else if (strchr(tosnap, '@') == NULL) { 2748 /* 2749 * If a filesystem was specified without -d or -e, we want to 2750 * tack on everything after the fs specified by 'zfs send'. 2751 */ 2752 chopprefix = drrb->drr_toname + strlen(sendfs); 2753 } else { 2754 /* A snapshot was specified as an exact path (no -d or -e). */ 2755 if (recursive) { 2756 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2757 "cannot specify snapshot name for multi-snapshot " 2758 "stream")); 2759 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 2760 } 2761 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname); 2762 } 2763 2764 ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname); 2765 ASSERT(chopprefix > drrb->drr_toname); 2766 ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname)); 2767 ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' || 2768 chopprefix[0] == '\0'); 2769 2770 /* 2771 * Determine name of destination snapshot, store in zc_value. 2772 */ 2773 (void) strcpy(zc.zc_value, tosnap); 2774 (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value)); 2775#ifdef __FreeBSD__ 2776 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF) 2777 zfs_ioctl_version = get_zfs_ioctl_version(); 2778 /* 2779 * For forward compatibility hide tosnap in zc_value 2780 */ 2781 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 2782 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap); 2783#endif 2784 free(cp); 2785 if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) { 2786 zcmd_free_nvlists(&zc); 2787 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 2788 } 2789 2790 /* 2791 * Determine the name of the origin snapshot, store in zc_string. 2792 */ 2793 if (drrb->drr_flags & DRR_FLAG_CLONE) { 2794 if (guid_to_name(hdl, zc.zc_value, 2795 drrb->drr_fromguid, zc.zc_string) != 0) { 2796 zcmd_free_nvlists(&zc); 2797 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2798 "local origin for clone %s does not exist"), 2799 zc.zc_value); 2800 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 2801 } 2802 if (flags->verbose) 2803 (void) printf("found clone origin %s\n", zc.zc_string); 2804 } 2805 2806 stream_wantsnewfs = (drrb->drr_fromguid == 0 || 2807 (drrb->drr_flags & DRR_FLAG_CLONE)); 2808 2809 if (stream_wantsnewfs) { 2810 /* 2811 * if the parent fs does not exist, look for it based on 2812 * the parent snap GUID 2813 */ 2814 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2815 "cannot receive new filesystem stream")); 2816 2817 (void) strcpy(zc.zc_name, zc.zc_value); 2818 cp = strrchr(zc.zc_name, '/'); 2819 if (cp) 2820 *cp = '\0'; 2821 if (cp && 2822 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 2823 char suffix[ZFS_MAXNAMELEN]; 2824 (void) strcpy(suffix, strrchr(zc.zc_value, '/')); 2825 if (guid_to_name(hdl, zc.zc_name, parent_snapguid, 2826 zc.zc_value) == 0) { 2827 *strchr(zc.zc_value, '@') = '\0'; 2828 (void) strcat(zc.zc_value, suffix); 2829 } 2830 } 2831 } else { 2832 /* 2833 * if the fs does not exist, look for it based on the 2834 * fromsnap GUID 2835 */ 2836 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2837 "cannot receive incremental stream")); 2838 2839 (void) strcpy(zc.zc_name, zc.zc_value); 2840 *strchr(zc.zc_name, '@') = '\0'; 2841 2842 /* 2843 * If the exact receive path was specified and this is the 2844 * topmost path in the stream, then if the fs does not exist we 2845 * should look no further. 2846 */ 2847 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname + 2848 strlen(sendfs)) != '\0' && *chopprefix != '@')) && 2849 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 2850 char snap[ZFS_MAXNAMELEN]; 2851 (void) strcpy(snap, strchr(zc.zc_value, '@')); 2852 if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid, 2853 zc.zc_value) == 0) { 2854 *strchr(zc.zc_value, '@') = '\0'; 2855 (void) strcat(zc.zc_value, snap); 2856 } 2857 } 2858 } 2859 2860 (void) strcpy(zc.zc_name, zc.zc_value); 2861 *strchr(zc.zc_name, '@') = '\0'; 2862 2863 if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 2864 zfs_handle_t *zhp; 2865 2866 /* 2867 * Destination fs exists. Therefore this should either 2868 * be an incremental, or the stream specifies a new fs 2869 * (full stream or clone) and they want us to blow it 2870 * away (and have therefore specified -F and removed any 2871 * snapshots). 2872 */ 2873 if (stream_wantsnewfs) { 2874 if (!flags->force) { 2875 zcmd_free_nvlists(&zc); 2876 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2877 "destination '%s' exists\n" 2878 "must specify -F to overwrite it"), 2879 zc.zc_name); 2880 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 2881 } 2882 if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT, 2883 &zc) == 0) { 2884 zcmd_free_nvlists(&zc); 2885 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2886 "destination has snapshots (eg. %s)\n" 2887 "must destroy them to overwrite it"), 2888 zc.zc_name); 2889 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 2890 } 2891 } 2892 2893 if ((zhp = zfs_open(hdl, zc.zc_name, 2894 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) { 2895 zcmd_free_nvlists(&zc); 2896 return (-1); 2897 } 2898 2899 if (stream_wantsnewfs && 2900 zhp->zfs_dmustats.dds_origin[0]) { 2901 zcmd_free_nvlists(&zc); 2902 zfs_close(zhp); 2903 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2904 "destination '%s' is a clone\n" 2905 "must destroy it to overwrite it"), 2906 zc.zc_name); 2907 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 2908 } 2909 2910 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM && 2911 stream_wantsnewfs) { 2912 /* We can't do online recv in this case */ 2913 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0); 2914 if (clp == NULL) { 2915 zfs_close(zhp); 2916 zcmd_free_nvlists(&zc); 2917 return (-1); 2918 } 2919 if (changelist_prefix(clp) != 0) { 2920 changelist_free(clp); 2921 zfs_close(zhp); 2922 zcmd_free_nvlists(&zc); 2923 return (-1); 2924 } 2925 } 2926 zfs_close(zhp); 2927 } else { 2928 /* 2929 * Destination filesystem does not exist. Therefore we better 2930 * be creating a new filesystem (either from a full backup, or 2931 * a clone). It would therefore be invalid if the user 2932 * specified only the pool name (i.e. if the destination name 2933 * contained no slash character). 2934 */ 2935 if (!stream_wantsnewfs || 2936 (cp = strrchr(zc.zc_name, '/')) == NULL) { 2937 zcmd_free_nvlists(&zc); 2938 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2939 "destination '%s' does not exist"), zc.zc_name); 2940 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 2941 } 2942 2943 /* 2944 * Trim off the final dataset component so we perform the 2945 * recvbackup ioctl to the filesystems's parent. 2946 */ 2947 *cp = '\0'; 2948 2949 if (flags->isprefix && !flags->istail && !flags->dryrun && 2950 create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) { 2951 zcmd_free_nvlists(&zc); 2952 return (zfs_error(hdl, EZFS_BADRESTORE, errbuf)); 2953 } 2954 2955 newfs = B_TRUE; 2956 } 2957 2958 zc.zc_begin_record = drr_noswap->drr_u.drr_begin; 2959 zc.zc_cookie = infd; 2960 zc.zc_guid = flags->force; 2961 if (flags->verbose) { 2962 (void) printf("%s %s stream of %s into %s\n", 2963 flags->dryrun ? "would receive" : "receiving", 2964 drrb->drr_fromguid ? "incremental" : "full", 2965 drrb->drr_toname, zc.zc_value); 2966 (void) fflush(stdout); 2967 } 2968 2969 if (flags->dryrun) { 2970 zcmd_free_nvlists(&zc); 2971 return (recv_skip(hdl, infd, flags->byteswap)); 2972 } 2973 2974 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf; 2975 zc.zc_nvlist_dst_size = sizeof (prop_errbuf); 2976 zc.zc_cleanup_fd = cleanup_fd; 2977 zc.zc_action_handle = *action_handlep; 2978 2979 err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc); 2980 ioctl_errno = errno; 2981 prop_errflags = (zprop_errflags_t)zc.zc_obj; 2982 2983 if (err == 0) { 2984 nvlist_t *prop_errors; 2985 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, 2986 zc.zc_nvlist_dst_size, &prop_errors, 0)); 2987 2988 nvpair_t *prop_err = NULL; 2989 2990 while ((prop_err = nvlist_next_nvpair(prop_errors, 2991 prop_err)) != NULL) { 2992 char tbuf[1024]; 2993 zfs_prop_t prop; 2994 int intval; 2995 2996 prop = zfs_name_to_prop(nvpair_name(prop_err)); 2997 (void) nvpair_value_int32(prop_err, &intval); 2998 if (strcmp(nvpair_name(prop_err), 2999 ZPROP_N_MORE_ERRORS) == 0) { 3000 trunc_prop_errs(intval); 3001 break; 3002 } else { 3003 (void) snprintf(tbuf, sizeof (tbuf), 3004 dgettext(TEXT_DOMAIN, 3005 "cannot receive %s property on %s"), 3006 nvpair_name(prop_err), zc.zc_name); 3007 zfs_setprop_error(hdl, prop, intval, tbuf); 3008 } 3009 } 3010 nvlist_free(prop_errors); 3011 } 3012 3013 zc.zc_nvlist_dst = 0; 3014 zc.zc_nvlist_dst_size = 0; 3015 zcmd_free_nvlists(&zc); 3016 3017 if (err == 0 && snapprops_nvlist) { 3018 zfs_cmd_t zc2 = { 0 }; 3019 3020 (void) strcpy(zc2.zc_name, zc.zc_value); 3021 zc2.zc_cookie = B_TRUE; /* received */ 3022 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) { 3023 (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2); 3024 zcmd_free_nvlists(&zc2); 3025 } 3026 } 3027 3028 if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) { 3029 /* 3030 * It may be that this snapshot already exists, 3031 * in which case we want to consume & ignore it 3032 * rather than failing. 3033 */ 3034 avl_tree_t *local_avl; 3035 nvlist_t *local_nv, *fs; 3036 cp = strchr(zc.zc_value, '@'); 3037 3038 /* 3039 * XXX Do this faster by just iterating over snaps in 3040 * this fs. Also if zc_value does not exist, we will 3041 * get a strange "does not exist" error message. 3042 */ 3043 *cp = '\0'; 3044 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE, 3045 &local_nv, &local_avl) == 0) { 3046 *cp = '@'; 3047 fs = fsavl_find(local_avl, drrb->drr_toguid, NULL); 3048 fsavl_destroy(local_avl); 3049 nvlist_free(local_nv); 3050 3051 if (fs != NULL) { 3052 if (flags->verbose) { 3053 (void) printf("snap %s already exists; " 3054 "ignoring\n", zc.zc_value); 3055 } 3056 err = ioctl_err = recv_skip(hdl, infd, 3057 flags->byteswap); 3058 } 3059 } 3060 *cp = '@'; 3061 } 3062 3063 if (ioctl_err != 0) { 3064 switch (ioctl_errno) { 3065 case ENODEV: 3066 cp = strchr(zc.zc_value, '@'); 3067 *cp = '\0'; 3068 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3069 "most recent snapshot of %s does not\n" 3070 "match incremental source"), zc.zc_value); 3071 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); 3072 *cp = '@'; 3073 break; 3074 case ETXTBSY: 3075 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3076 "destination %s has been modified\n" 3077 "since most recent snapshot"), zc.zc_name); 3078 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); 3079 break; 3080 case EEXIST: 3081 cp = strchr(zc.zc_value, '@'); 3082 if (newfs) { 3083 /* it's the containing fs that exists */ 3084 *cp = '\0'; 3085 } 3086 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3087 "destination already exists")); 3088 (void) zfs_error_fmt(hdl, EZFS_EXISTS, 3089 dgettext(TEXT_DOMAIN, "cannot restore to %s"), 3090 zc.zc_value); 3091 *cp = '@'; 3092 break; 3093 case EINVAL: 3094 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); 3095 break; 3096 case ECKSUM: 3097 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3098 "invalid stream (checksum mismatch)")); 3099 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); 3100 break; 3101 case ENOTSUP: 3102 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3103 "pool must be upgraded to receive this stream.")); 3104 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf); 3105 break; 3106 case EDQUOT: 3107 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3108 "destination %s space quota exceeded"), zc.zc_name); 3109 (void) zfs_error(hdl, EZFS_NOSPC, errbuf); 3110 break; 3111 default: 3112 (void) zfs_standard_error(hdl, ioctl_errno, errbuf); 3113 } 3114 } 3115 3116 /* 3117 * Mount the target filesystem (if created). Also mount any 3118 * children of the target filesystem if we did a replication 3119 * receive (indicated by stream_avl being non-NULL). 3120 */ 3121 cp = strchr(zc.zc_value, '@'); 3122 if (cp && (ioctl_err == 0 || !newfs)) { 3123 zfs_handle_t *h; 3124 3125 *cp = '\0'; 3126 h = zfs_open(hdl, zc.zc_value, 3127 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 3128 if (h != NULL) { 3129 if (h->zfs_type == ZFS_TYPE_VOLUME) { 3130 *cp = '@'; 3131 } else if (newfs || stream_avl) { 3132 /* 3133 * Track the first/top of hierarchy fs, 3134 * for mounting and sharing later. 3135 */ 3136 if (top_zfs && *top_zfs == NULL) 3137 *top_zfs = zfs_strdup(hdl, zc.zc_value); 3138 } 3139 zfs_close(h); 3140 } 3141 *cp = '@'; 3142 } 3143 3144 if (clp) { 3145 err |= changelist_postfix(clp); 3146 changelist_free(clp); 3147 } 3148 3149 if (prop_errflags & ZPROP_ERR_NOCLEAR) { 3150 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " 3151 "failed to clear unreceived properties on %s"), 3152 zc.zc_name); 3153 (void) fprintf(stderr, "\n"); 3154 } 3155 if (prop_errflags & ZPROP_ERR_NORESTORE) { 3156 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " 3157 "failed to restore original properties on %s"), 3158 zc.zc_name); 3159 (void) fprintf(stderr, "\n"); 3160 } 3161 3162 if (err || ioctl_err) 3163 return (-1); 3164 3165 *action_handlep = zc.zc_action_handle; 3166 3167 if (flags->verbose) { 3168 char buf1[64]; 3169 char buf2[64]; 3170 uint64_t bytes = zc.zc_cookie; 3171 time_t delta = time(NULL) - begin_time; 3172 if (delta == 0) 3173 delta = 1; 3174 zfs_nicenum(bytes, buf1, sizeof (buf1)); 3175 zfs_nicenum(bytes/delta, buf2, sizeof (buf1)); 3176 3177 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n", 3178 buf1, delta, buf2); 3179 } 3180 3181 return (0); 3182} 3183 3184static int 3185zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags, 3186 int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl, 3187 char **top_zfs, int cleanup_fd, uint64_t *action_handlep) 3188{ 3189 int err; 3190 dmu_replay_record_t drr, drr_noswap; 3191 struct drr_begin *drrb = &drr.drr_u.drr_begin; 3192 char errbuf[1024]; 3193 zio_cksum_t zcksum = { 0 }; 3194 uint64_t featureflags; 3195 int hdrtype; 3196 3197 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 3198 "cannot receive")); 3199 3200 if (flags->isprefix && 3201 !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) { 3202 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs " 3203 "(%s) does not exist"), tosnap); 3204 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3205 } 3206 3207 /* read in the BEGIN record */ 3208 if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE, 3209 &zcksum))) 3210 return (err); 3211 3212 if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) { 3213 /* It's the double end record at the end of a package */ 3214 return (ENODATA); 3215 } 3216 3217 /* the kernel needs the non-byteswapped begin record */ 3218 drr_noswap = drr; 3219 3220 flags->byteswap = B_FALSE; 3221 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { 3222 /* 3223 * We computed the checksum in the wrong byteorder in 3224 * recv_read() above; do it again correctly. 3225 */ 3226 bzero(&zcksum, sizeof (zio_cksum_t)); 3227 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum); 3228 flags->byteswap = B_TRUE; 3229 3230 drr.drr_type = BSWAP_32(drr.drr_type); 3231 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen); 3232 drrb->drr_magic = BSWAP_64(drrb->drr_magic); 3233 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); 3234 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); 3235 drrb->drr_type = BSWAP_32(drrb->drr_type); 3236 drrb->drr_flags = BSWAP_32(drrb->drr_flags); 3237 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); 3238 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); 3239 } 3240 3241 if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) { 3242 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3243 "stream (bad magic number)")); 3244 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3245 } 3246 3247 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 3248 hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo); 3249 3250 if (!DMU_STREAM_SUPPORTED(featureflags) || 3251 (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) { 3252 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3253 "stream has unsupported feature, feature flags = %lx"), 3254 featureflags); 3255 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3256 } 3257 3258 if (strchr(drrb->drr_toname, '@') == NULL) { 3259 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3260 "stream (bad snapshot name)")); 3261 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3262 } 3263 3264 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) { 3265 char nonpackage_sendfs[ZFS_MAXNAMELEN]; 3266 if (sendfs == NULL) { 3267 /* 3268 * We were not called from zfs_receive_package(). Get 3269 * the fs specified by 'zfs send'. 3270 */ 3271 char *cp; 3272 (void) strlcpy(nonpackage_sendfs, 3273 drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN); 3274 if ((cp = strchr(nonpackage_sendfs, '@')) != NULL) 3275 *cp = '\0'; 3276 sendfs = nonpackage_sendfs; 3277 } 3278 return (zfs_receive_one(hdl, infd, tosnap, flags, 3279 &drr, &drr_noswap, sendfs, stream_nv, stream_avl, 3280 top_zfs, cleanup_fd, action_handlep)); 3281 } else { 3282 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 3283 DMU_COMPOUNDSTREAM); 3284 return (zfs_receive_package(hdl, infd, tosnap, flags, 3285 &drr, &zcksum, top_zfs, cleanup_fd, action_handlep)); 3286 } 3287} 3288 3289/* 3290 * Restores a backup of tosnap from the file descriptor specified by infd. 3291 * Return 0 on total success, -2 if some things couldn't be 3292 * destroyed/renamed/promoted, -1 if some things couldn't be received. 3293 * (-1 will override -2). 3294 */ 3295int 3296zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags, 3297 int infd, avl_tree_t *stream_avl) 3298{ 3299 char *top_zfs = NULL; 3300 int err; 3301 int cleanup_fd; 3302 uint64_t action_handle = 0; 3303 3304 cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); 3305 VERIFY(cleanup_fd >= 0); 3306 3307 err = zfs_receive_impl(hdl, tosnap, flags, infd, NULL, NULL, 3308 stream_avl, &top_zfs, cleanup_fd, &action_handle); 3309 3310 VERIFY(0 == close(cleanup_fd)); 3311 3312 if (err == 0 && !flags->nomount && top_zfs) { 3313 zfs_handle_t *zhp; 3314 prop_changelist_t *clp; 3315 3316 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM); 3317 if (zhp != NULL) { 3318 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT, 3319 CL_GATHER_MOUNT_ALWAYS, 0); 3320 zfs_close(zhp); 3321 if (clp != NULL) { 3322 /* mount and share received datasets */ 3323 err = changelist_postfix(clp); 3324 changelist_free(clp); 3325 } 3326 } 3327 if (zhp == NULL || clp == NULL || err) 3328 err = -1; 3329 } 3330 if (top_zfs) 3331 free(top_zfs); 3332 3333 return (err); 3334} 3335