libzfs_sendrecv.c revision 290755
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 *, const char *, 68 recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **, int, 69 uint64_t *); 70 71static const zio_cksum_t zero_cksum = { 0 }; 72 73typedef struct dedup_arg { 74 int inputfd; 75 int outputfd; 76 libzfs_handle_t *dedup_hdl; 77} dedup_arg_t; 78 79typedef struct progress_arg { 80 zfs_handle_t *pa_zhp; 81 int pa_fd; 82 boolean_t pa_parsable; 83} progress_arg_t; 84 85typedef struct dataref { 86 uint64_t ref_guid; 87 uint64_t ref_object; 88 uint64_t ref_offset; 89} dataref_t; 90 91typedef struct dedup_entry { 92 struct dedup_entry *dde_next; 93 zio_cksum_t dde_chksum; 94 uint64_t dde_prop; 95 dataref_t dde_ref; 96} dedup_entry_t; 97 98#define MAX_DDT_PHYSMEM_PERCENT 20 99#define SMALLEST_POSSIBLE_MAX_DDT_MB 128 100 101typedef struct dedup_table { 102 dedup_entry_t **dedup_hash_array; 103 umem_cache_t *ddecache; 104 uint64_t max_ddt_size; /* max dedup table size in bytes */ 105 uint64_t cur_ddt_size; /* current dedup table size in bytes */ 106 uint64_t ddt_count; 107 int numhashbits; 108 boolean_t ddt_full; 109} dedup_table_t; 110 111static int 112high_order_bit(uint64_t n) 113{ 114 int count; 115 116 for (count = 0; n != 0; count++) 117 n >>= 1; 118 return (count); 119} 120 121static size_t 122ssread(void *buf, size_t len, FILE *stream) 123{ 124 size_t outlen; 125 126 if ((outlen = fread(buf, len, 1, stream)) == 0) 127 return (0); 128 129 return (outlen); 130} 131 132static void 133ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp, 134 zio_cksum_t *cs, uint64_t prop, dataref_t *dr) 135{ 136 dedup_entry_t *dde; 137 138 if (ddt->cur_ddt_size >= ddt->max_ddt_size) { 139 if (ddt->ddt_full == B_FALSE) { 140 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 141 "Dedup table full. Deduplication will continue " 142 "with existing table entries")); 143 ddt->ddt_full = B_TRUE; 144 } 145 return; 146 } 147 148 if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT)) 149 != NULL) { 150 assert(*ddepp == NULL); 151 dde->dde_next = NULL; 152 dde->dde_chksum = *cs; 153 dde->dde_prop = prop; 154 dde->dde_ref = *dr; 155 *ddepp = dde; 156 ddt->cur_ddt_size += sizeof (dedup_entry_t); 157 ddt->ddt_count++; 158 } 159} 160 161/* 162 * Using the specified dedup table, do a lookup for an entry with 163 * the checksum cs. If found, return the block's reference info 164 * in *dr. Otherwise, insert a new entry in the dedup table, using 165 * the reference information specified by *dr. 166 * 167 * return value: true - entry was found 168 * false - entry was not found 169 */ 170static boolean_t 171ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs, 172 uint64_t prop, dataref_t *dr) 173{ 174 uint32_t hashcode; 175 dedup_entry_t **ddepp; 176 177 hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits); 178 179 for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL; 180 ddepp = &((*ddepp)->dde_next)) { 181 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) && 182 (*ddepp)->dde_prop == prop) { 183 *dr = (*ddepp)->dde_ref; 184 return (B_TRUE); 185 } 186 } 187 ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr); 188 return (B_FALSE); 189} 190 191static int 192dump_record(dmu_replay_record_t *drr, void *payload, int payload_len, 193 zio_cksum_t *zc, int outfd) 194{ 195 ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 196 ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 197 fletcher_4_incremental_native(drr, 198 offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc); 199 if (drr->drr_type != DRR_BEGIN) { 200 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u. 201 drr_checksum.drr_checksum)); 202 drr->drr_u.drr_checksum.drr_checksum = *zc; 203 } 204 fletcher_4_incremental_native(&drr->drr_u.drr_checksum.drr_checksum, 205 sizeof (zio_cksum_t), zc); 206 if (write(outfd, drr, sizeof (*drr)) == -1) 207 return (errno); 208 if (payload_len != 0) { 209 fletcher_4_incremental_native(payload, payload_len, zc); 210 if (write(outfd, payload, payload_len) == -1) 211 return (errno); 212 } 213 return (0); 214} 215 216/* 217 * This function is started in a separate thread when the dedup option 218 * has been requested. The main send thread determines the list of 219 * snapshots to be included in the send stream and makes the ioctl calls 220 * for each one. But instead of having the ioctl send the output to the 221 * the output fd specified by the caller of zfs_send()), the 222 * ioctl is told to direct the output to a pipe, which is read by the 223 * alternate thread running THIS function. This function does the 224 * dedup'ing by: 225 * 1. building a dedup table (the DDT) 226 * 2. doing checksums on each data block and inserting a record in the DDT 227 * 3. looking for matching checksums, and 228 * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever 229 * a duplicate block is found. 230 * The output of this function then goes to the output fd requested 231 * by the caller of zfs_send(). 232 */ 233static void * 234cksummer(void *arg) 235{ 236 dedup_arg_t *dda = arg; 237 char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE); 238 dmu_replay_record_t thedrr; 239 dmu_replay_record_t *drr = &thedrr; 240 FILE *ofp; 241 int outfd; 242 dedup_table_t ddt; 243 zio_cksum_t stream_cksum; 244 uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE); 245 uint64_t numbuckets; 246 247 ddt.max_ddt_size = 248 MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100, 249 SMALLEST_POSSIBLE_MAX_DDT_MB << 20); 250 251 numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t)); 252 253 /* 254 * numbuckets must be a power of 2. Increase number to 255 * a power of 2 if necessary. 256 */ 257 if (!ISP2(numbuckets)) 258 numbuckets = 1 << high_order_bit(numbuckets); 259 260 ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *)); 261 ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0, 262 NULL, NULL, NULL, NULL, NULL, 0); 263 ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *); 264 ddt.numhashbits = high_order_bit(numbuckets) - 1; 265 ddt.ddt_full = B_FALSE; 266 267 outfd = dda->outputfd; 268 ofp = fdopen(dda->inputfd, "r"); 269 while (ssread(drr, sizeof (*drr), ofp) != 0) { 270 271 switch (drr->drr_type) { 272 case DRR_BEGIN: 273 { 274 struct drr_begin *drrb = &drr->drr_u.drr_begin; 275 int fflags; 276 int sz = 0; 277 ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0); 278 279 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 280 281 /* set the DEDUP feature flag for this stream */ 282 fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 283 fflags |= (DMU_BACKUP_FEATURE_DEDUP | 284 DMU_BACKUP_FEATURE_DEDUPPROPS); 285 DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags); 286 287 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 288 DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) { 289 sz = drr->drr_payloadlen; 290 291 if (sz > SPA_MAXBLOCKSIZE) { 292 buf = zfs_realloc(dda->dedup_hdl, buf, 293 SPA_MAXBLOCKSIZE, sz); 294 } 295 (void) ssread(buf, sz, ofp); 296 if (ferror(stdin)) 297 perror("fread"); 298 } 299 if (dump_record(drr, buf, sz, &stream_cksum, 300 outfd) != 0) 301 goto out; 302 break; 303 } 304 305 case DRR_END: 306 { 307 struct drr_end *drre = &drr->drr_u.drr_end; 308 /* use the recalculated checksum */ 309 drre->drr_checksum = stream_cksum; 310 if (dump_record(drr, NULL, 0, &stream_cksum, 311 outfd) != 0) 312 goto out; 313 break; 314 } 315 316 case DRR_OBJECT: 317 { 318 struct drr_object *drro = &drr->drr_u.drr_object; 319 if (drro->drr_bonuslen > 0) { 320 (void) ssread(buf, 321 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), 322 ofp); 323 } 324 if (dump_record(drr, buf, 325 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), 326 &stream_cksum, outfd) != 0) 327 goto out; 328 break; 329 } 330 331 case DRR_SPILL: 332 { 333 struct drr_spill *drrs = &drr->drr_u.drr_spill; 334 (void) ssread(buf, drrs->drr_length, ofp); 335 if (dump_record(drr, buf, drrs->drr_length, 336 &stream_cksum, outfd) != 0) 337 goto out; 338 break; 339 } 340 341 case DRR_FREEOBJECTS: 342 { 343 if (dump_record(drr, NULL, 0, &stream_cksum, 344 outfd) != 0) 345 goto out; 346 break; 347 } 348 349 case DRR_WRITE: 350 { 351 struct drr_write *drrw = &drr->drr_u.drr_write; 352 dataref_t dataref; 353 354 (void) ssread(buf, drrw->drr_length, ofp); 355 356 /* 357 * Use the existing checksum if it's dedup-capable, 358 * else calculate a SHA256 checksum for it. 359 */ 360 361 if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum, 362 zero_cksum) || 363 !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) { 364 SHA256_CTX ctx; 365 zio_cksum_t tmpsha256; 366 367 SHA256Init(&ctx); 368 SHA256Update(&ctx, buf, drrw->drr_length); 369 SHA256Final(&tmpsha256, &ctx); 370 drrw->drr_key.ddk_cksum.zc_word[0] = 371 BE_64(tmpsha256.zc_word[0]); 372 drrw->drr_key.ddk_cksum.zc_word[1] = 373 BE_64(tmpsha256.zc_word[1]); 374 drrw->drr_key.ddk_cksum.zc_word[2] = 375 BE_64(tmpsha256.zc_word[2]); 376 drrw->drr_key.ddk_cksum.zc_word[3] = 377 BE_64(tmpsha256.zc_word[3]); 378 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256; 379 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP; 380 } 381 382 dataref.ref_guid = drrw->drr_toguid; 383 dataref.ref_object = drrw->drr_object; 384 dataref.ref_offset = drrw->drr_offset; 385 386 if (ddt_update(dda->dedup_hdl, &ddt, 387 &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop, 388 &dataref)) { 389 dmu_replay_record_t wbr_drr = {0}; 390 struct drr_write_byref *wbr_drrr = 391 &wbr_drr.drr_u.drr_write_byref; 392 393 /* block already present in stream */ 394 wbr_drr.drr_type = DRR_WRITE_BYREF; 395 396 wbr_drrr->drr_object = drrw->drr_object; 397 wbr_drrr->drr_offset = drrw->drr_offset; 398 wbr_drrr->drr_length = drrw->drr_length; 399 wbr_drrr->drr_toguid = drrw->drr_toguid; 400 wbr_drrr->drr_refguid = dataref.ref_guid; 401 wbr_drrr->drr_refobject = 402 dataref.ref_object; 403 wbr_drrr->drr_refoffset = 404 dataref.ref_offset; 405 406 wbr_drrr->drr_checksumtype = 407 drrw->drr_checksumtype; 408 wbr_drrr->drr_checksumflags = 409 drrw->drr_checksumtype; 410 wbr_drrr->drr_key.ddk_cksum = 411 drrw->drr_key.ddk_cksum; 412 wbr_drrr->drr_key.ddk_prop = 413 drrw->drr_key.ddk_prop; 414 415 if (dump_record(&wbr_drr, NULL, 0, 416 &stream_cksum, outfd) != 0) 417 goto out; 418 } else { 419 /* block not previously seen */ 420 if (dump_record(drr, buf, drrw->drr_length, 421 &stream_cksum, outfd) != 0) 422 goto out; 423 } 424 break; 425 } 426 427 case DRR_WRITE_EMBEDDED: 428 { 429 struct drr_write_embedded *drrwe = 430 &drr->drr_u.drr_write_embedded; 431 (void) ssread(buf, 432 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp); 433 if (dump_record(drr, buf, 434 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), 435 &stream_cksum, outfd) != 0) 436 goto out; 437 break; 438 } 439 440 case DRR_FREE: 441 { 442 if (dump_record(drr, NULL, 0, &stream_cksum, 443 outfd) != 0) 444 goto out; 445 break; 446 } 447 448 default: 449 (void) fprintf(stderr, "INVALID record type 0x%x\n", 450 drr->drr_type); 451 /* should never happen, so assert */ 452 assert(B_FALSE); 453 } 454 } 455out: 456 umem_cache_destroy(ddt.ddecache); 457 free(ddt.dedup_hash_array); 458 free(buf); 459 (void) fclose(ofp); 460 461 return (NULL); 462} 463 464/* 465 * Routines for dealing with the AVL tree of fs-nvlists 466 */ 467typedef struct fsavl_node { 468 avl_node_t fn_node; 469 nvlist_t *fn_nvfs; 470 char *fn_snapname; 471 uint64_t fn_guid; 472} fsavl_node_t; 473 474static int 475fsavl_compare(const void *arg1, const void *arg2) 476{ 477 const fsavl_node_t *fn1 = arg1; 478 const fsavl_node_t *fn2 = arg2; 479 480 if (fn1->fn_guid > fn2->fn_guid) 481 return (+1); 482 else if (fn1->fn_guid < fn2->fn_guid) 483 return (-1); 484 else 485 return (0); 486} 487 488/* 489 * Given the GUID of a snapshot, find its containing filesystem and 490 * (optionally) name. 491 */ 492static nvlist_t * 493fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname) 494{ 495 fsavl_node_t fn_find; 496 fsavl_node_t *fn; 497 498 fn_find.fn_guid = snapguid; 499 500 fn = avl_find(avl, &fn_find, NULL); 501 if (fn) { 502 if (snapname) 503 *snapname = fn->fn_snapname; 504 return (fn->fn_nvfs); 505 } 506 return (NULL); 507} 508 509static void 510fsavl_destroy(avl_tree_t *avl) 511{ 512 fsavl_node_t *fn; 513 void *cookie; 514 515 if (avl == NULL) 516 return; 517 518 cookie = NULL; 519 while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL) 520 free(fn); 521 avl_destroy(avl); 522 free(avl); 523} 524 525/* 526 * Given an nvlist, produce an avl tree of snapshots, ordered by guid 527 */ 528static avl_tree_t * 529fsavl_create(nvlist_t *fss) 530{ 531 avl_tree_t *fsavl; 532 nvpair_t *fselem = NULL; 533 534 if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL) 535 return (NULL); 536 537 avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t), 538 offsetof(fsavl_node_t, fn_node)); 539 540 while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) { 541 nvlist_t *nvfs, *snaps; 542 nvpair_t *snapelem = NULL; 543 544 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); 545 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); 546 547 while ((snapelem = 548 nvlist_next_nvpair(snaps, snapelem)) != NULL) { 549 fsavl_node_t *fn; 550 uint64_t guid; 551 552 VERIFY(0 == nvpair_value_uint64(snapelem, &guid)); 553 if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) { 554 fsavl_destroy(fsavl); 555 return (NULL); 556 } 557 fn->fn_nvfs = nvfs; 558 fn->fn_snapname = nvpair_name(snapelem); 559 fn->fn_guid = guid; 560 561 /* 562 * Note: if there are multiple snaps with the 563 * same GUID, we ignore all but one. 564 */ 565 if (avl_find(fsavl, fn, NULL) == NULL) 566 avl_add(fsavl, fn); 567 else 568 free(fn); 569 } 570 } 571 572 return (fsavl); 573} 574 575/* 576 * Routines for dealing with the giant nvlist of fs-nvlists, etc. 577 */ 578typedef struct send_data { 579 uint64_t parent_fromsnap_guid; 580 nvlist_t *parent_snaps; 581 nvlist_t *fss; 582 nvlist_t *snapprops; 583 const char *fromsnap; 584 const char *tosnap; 585 boolean_t recursive; 586 587 /* 588 * The header nvlist is of the following format: 589 * { 590 * "tosnap" -> string 591 * "fromsnap" -> string (if incremental) 592 * "fss" -> { 593 * id -> { 594 * 595 * "name" -> string (full name; for debugging) 596 * "parentfromsnap" -> number (guid of fromsnap in parent) 597 * 598 * "props" -> { name -> value (only if set here) } 599 * "snaps" -> { name (lastname) -> number (guid) } 600 * "snapprops" -> { name (lastname) -> { name -> value } } 601 * 602 * "origin" -> number (guid) (if clone) 603 * "sent" -> boolean (not on-disk) 604 * } 605 * } 606 * } 607 * 608 */ 609} send_data_t; 610 611static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv); 612 613static int 614send_iterate_snap(zfs_handle_t *zhp, void *arg) 615{ 616 send_data_t *sd = arg; 617 uint64_t guid = zhp->zfs_dmustats.dds_guid; 618 char *snapname; 619 nvlist_t *nv; 620 621 snapname = strrchr(zhp->zfs_name, '@')+1; 622 623 VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid)); 624 /* 625 * NB: if there is no fromsnap here (it's a newly created fs in 626 * an incremental replication), we will substitute the tosnap. 627 */ 628 if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) || 629 (sd->parent_fromsnap_guid == 0 && sd->tosnap && 630 strcmp(snapname, sd->tosnap) == 0)) { 631 sd->parent_fromsnap_guid = guid; 632 } 633 634 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); 635 send_iterate_prop(zhp, nv); 636 VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv)); 637 nvlist_free(nv); 638 639 zfs_close(zhp); 640 return (0); 641} 642 643static void 644send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv) 645{ 646 nvpair_t *elem = NULL; 647 648 while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) { 649 char *propname = nvpair_name(elem); 650 zfs_prop_t prop = zfs_name_to_prop(propname); 651 nvlist_t *propnv; 652 653 if (!zfs_prop_user(propname)) { 654 /* 655 * Realistically, this should never happen. However, 656 * we want the ability to add DSL properties without 657 * needing to make incompatible version changes. We 658 * need to ignore unknown properties to allow older 659 * software to still send datasets containing these 660 * properties, with the unknown properties elided. 661 */ 662 if (prop == ZPROP_INVAL) 663 continue; 664 665 if (zfs_prop_readonly(prop)) 666 continue; 667 } 668 669 verify(nvpair_value_nvlist(elem, &propnv) == 0); 670 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION || 671 prop == ZFS_PROP_REFQUOTA || 672 prop == ZFS_PROP_REFRESERVATION) { 673 char *source; 674 uint64_t value; 675 verify(nvlist_lookup_uint64(propnv, 676 ZPROP_VALUE, &value) == 0); 677 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) 678 continue; 679 /* 680 * May have no source before SPA_VERSION_RECVD_PROPS, 681 * but is still modifiable. 682 */ 683 if (nvlist_lookup_string(propnv, 684 ZPROP_SOURCE, &source) == 0) { 685 if ((strcmp(source, zhp->zfs_name) != 0) && 686 (strcmp(source, 687 ZPROP_SOURCE_VAL_RECVD) != 0)) 688 continue; 689 } 690 } else { 691 char *source; 692 if (nvlist_lookup_string(propnv, 693 ZPROP_SOURCE, &source) != 0) 694 continue; 695 if ((strcmp(source, zhp->zfs_name) != 0) && 696 (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0)) 697 continue; 698 } 699 700 if (zfs_prop_user(propname) || 701 zfs_prop_get_type(prop) == PROP_TYPE_STRING) { 702 char *value; 703 verify(nvlist_lookup_string(propnv, 704 ZPROP_VALUE, &value) == 0); 705 VERIFY(0 == nvlist_add_string(nv, propname, value)); 706 } else { 707 uint64_t value; 708 verify(nvlist_lookup_uint64(propnv, 709 ZPROP_VALUE, &value) == 0); 710 VERIFY(0 == nvlist_add_uint64(nv, propname, value)); 711 } 712 } 713} 714 715/* 716 * recursively generate nvlists describing datasets. See comment 717 * for the data structure send_data_t above for description of contents 718 * of the nvlist. 719 */ 720static int 721send_iterate_fs(zfs_handle_t *zhp, void *arg) 722{ 723 send_data_t *sd = arg; 724 nvlist_t *nvfs, *nv; 725 int rv = 0; 726 uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid; 727 uint64_t guid = zhp->zfs_dmustats.dds_guid; 728 char guidstring[64]; 729 730 VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0)); 731 VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name)); 732 VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap", 733 sd->parent_fromsnap_guid)); 734 735 if (zhp->zfs_dmustats.dds_origin[0]) { 736 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl, 737 zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT); 738 if (origin == NULL) 739 return (-1); 740 VERIFY(0 == nvlist_add_uint64(nvfs, "origin", 741 origin->zfs_dmustats.dds_guid)); 742 } 743 744 /* iterate over props */ 745 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); 746 send_iterate_prop(zhp, nv); 747 VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv)); 748 nvlist_free(nv); 749 750 /* iterate over snaps, and set sd->parent_fromsnap_guid */ 751 sd->parent_fromsnap_guid = 0; 752 VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0)); 753 VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0)); 754 (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd); 755 VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps)); 756 VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops)); 757 nvlist_free(sd->parent_snaps); 758 nvlist_free(sd->snapprops); 759 760 /* add this fs to nvlist */ 761 (void) snprintf(guidstring, sizeof (guidstring), 762 "0x%llx", (longlong_t)guid); 763 VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs)); 764 nvlist_free(nvfs); 765 766 /* iterate over children */ 767 if (sd->recursive) 768 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd); 769 770 sd->parent_fromsnap_guid = parent_fromsnap_guid_save; 771 772 zfs_close(zhp); 773 return (rv); 774} 775 776static int 777gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap, 778 const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp) 779{ 780 zfs_handle_t *zhp; 781 send_data_t sd = { 0 }; 782 int error; 783 784 zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 785 if (zhp == NULL) 786 return (EZFS_BADTYPE); 787 788 VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0)); 789 sd.fromsnap = fromsnap; 790 sd.tosnap = tosnap; 791 sd.recursive = recursive; 792 793 if ((error = send_iterate_fs(zhp, &sd)) != 0) { 794 nvlist_free(sd.fss); 795 if (avlp != NULL) 796 *avlp = NULL; 797 *nvlp = NULL; 798 return (error); 799 } 800 801 if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) { 802 nvlist_free(sd.fss); 803 *nvlp = NULL; 804 return (EZFS_NOMEM); 805 } 806 807 *nvlp = sd.fss; 808 return (0); 809} 810 811/* 812 * Routines specific to "zfs send" 813 */ 814typedef struct send_dump_data { 815 /* these are all just the short snapname (the part after the @) */ 816 const char *fromsnap; 817 const char *tosnap; 818 char prevsnap[ZFS_MAXNAMELEN]; 819 uint64_t prevsnap_obj; 820 boolean_t seenfrom, seento, replicate, doall, fromorigin; 821 boolean_t verbose, dryrun, parsable, progress, embed_data, std_out; 822 boolean_t large_block; 823 int outfd; 824 boolean_t err; 825 nvlist_t *fss; 826 nvlist_t *snapholds; 827 avl_tree_t *fsavl; 828 snapfilter_cb_t *filter_cb; 829 void *filter_cb_arg; 830 nvlist_t *debugnv; 831 char holdtag[ZFS_MAXNAMELEN]; 832 int cleanup_fd; 833 uint64_t size; 834} send_dump_data_t; 835 836static int 837estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj, 838 boolean_t fromorigin, uint64_t *sizep) 839{ 840 zfs_cmd_t zc = { 0 }; 841 libzfs_handle_t *hdl = zhp->zfs_hdl; 842 843 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 844 assert(fromsnap_obj == 0 || !fromorigin); 845 846 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 847 zc.zc_obj = fromorigin; 848 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 849 zc.zc_fromobj = fromsnap_obj; 850 zc.zc_guid = 1; /* estimate flag */ 851 852 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { 853 char errbuf[1024]; 854 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 855 "warning: cannot estimate space for '%s'"), zhp->zfs_name); 856 857 switch (errno) { 858 case EXDEV: 859 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 860 "not an earlier snapshot from the same fs")); 861 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 862 863 case ENOENT: 864 if (zfs_dataset_exists(hdl, zc.zc_name, 865 ZFS_TYPE_SNAPSHOT)) { 866 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 867 "incremental source (@%s) does not exist"), 868 zc.zc_value); 869 } 870 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 871 872 case EDQUOT: 873 case EFBIG: 874 case EIO: 875 case ENOLINK: 876 case ENOSPC: 877 case ENXIO: 878 case EPIPE: 879 case ERANGE: 880 case EFAULT: 881 case EROFS: 882 zfs_error_aux(hdl, strerror(errno)); 883 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 884 885 default: 886 return (zfs_standard_error(hdl, errno, errbuf)); 887 } 888 } 889 890 *sizep = zc.zc_objset_type; 891 892 return (0); 893} 894 895/* 896 * Dumps a backup of the given snapshot (incremental from fromsnap if it's not 897 * NULL) to the file descriptor specified by outfd. 898 */ 899static int 900dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj, 901 boolean_t fromorigin, int outfd, enum lzc_send_flags flags, 902 nvlist_t *debugnv) 903{ 904 zfs_cmd_t zc = { 0 }; 905 libzfs_handle_t *hdl = zhp->zfs_hdl; 906 nvlist_t *thisdbg; 907 908 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 909 assert(fromsnap_obj == 0 || !fromorigin); 910 911 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 912 zc.zc_cookie = outfd; 913 zc.zc_obj = fromorigin; 914 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 915 zc.zc_fromobj = fromsnap_obj; 916 zc.zc_flags = flags; 917 918 VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0)); 919 if (fromsnap && fromsnap[0] != '\0') { 920 VERIFY(0 == nvlist_add_string(thisdbg, 921 "fromsnap", fromsnap)); 922 } 923 924 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { 925 char errbuf[1024]; 926 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 927 "warning: cannot send '%s'"), zhp->zfs_name); 928 929 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno)); 930 if (debugnv) { 931 VERIFY(0 == nvlist_add_nvlist(debugnv, 932 zhp->zfs_name, thisdbg)); 933 } 934 nvlist_free(thisdbg); 935 936 switch (errno) { 937 case EXDEV: 938 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 939 "not an earlier snapshot from the same fs")); 940 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 941 942 case ENOENT: 943 if (zfs_dataset_exists(hdl, zc.zc_name, 944 ZFS_TYPE_SNAPSHOT)) { 945 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 946 "incremental source (@%s) does not exist"), 947 zc.zc_value); 948 } 949 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 950 951 case EDQUOT: 952 case EFBIG: 953 case EIO: 954 case ENOLINK: 955 case ENOSPC: 956#ifdef sun 957 case ENOSTR: 958#endif 959 case ENXIO: 960 case EPIPE: 961 case ERANGE: 962 case EFAULT: 963 case EROFS: 964 zfs_error_aux(hdl, strerror(errno)); 965 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 966 967 default: 968 return (zfs_standard_error(hdl, errno, errbuf)); 969 } 970 } 971 972 if (debugnv) 973 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg)); 974 nvlist_free(thisdbg); 975 976 return (0); 977} 978 979static void 980gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd) 981{ 982 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 983 984 /* 985 * zfs_send() only sets snapholds for sends that need them, 986 * e.g. replication and doall. 987 */ 988 if (sdd->snapholds == NULL) 989 return; 990 991 fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag); 992} 993 994static void * 995send_progress_thread(void *arg) 996{ 997 progress_arg_t *pa = arg; 998 999 zfs_cmd_t zc = { 0 }; 1000 zfs_handle_t *zhp = pa->pa_zhp; 1001 libzfs_handle_t *hdl = zhp->zfs_hdl; 1002 unsigned long long bytes; 1003 char buf[16]; 1004 1005 time_t t; 1006 struct tm *tm; 1007 1008 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); 1009 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); 1010 1011 if (!pa->pa_parsable) 1012 (void) fprintf(stderr, "TIME SENT SNAPSHOT\n"); 1013 1014 /* 1015 * Print the progress from ZFS_IOC_SEND_PROGRESS every second. 1016 */ 1017 for (;;) { 1018 (void) sleep(1); 1019 1020 zc.zc_cookie = pa->pa_fd; 1021 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0) 1022 return ((void *)-1); 1023 1024 (void) time(&t); 1025 tm = localtime(&t); 1026 bytes = zc.zc_cookie; 1027 1028 if (pa->pa_parsable) { 1029 (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n", 1030 tm->tm_hour, tm->tm_min, tm->tm_sec, 1031 bytes, zhp->zfs_name); 1032 } else { 1033 zfs_nicenum(bytes, buf, sizeof (buf)); 1034 (void) fprintf(stderr, "%02d:%02d:%02d %5s %s\n", 1035 tm->tm_hour, tm->tm_min, tm->tm_sec, 1036 buf, zhp->zfs_name); 1037 } 1038 } 1039} 1040 1041static int 1042dump_snapshot(zfs_handle_t *zhp, void *arg) 1043{ 1044 send_dump_data_t *sdd = arg; 1045 progress_arg_t pa = { 0 }; 1046 pthread_t tid; 1047 char *thissnap; 1048 int err; 1049 boolean_t isfromsnap, istosnap, fromorigin; 1050 boolean_t exclude = B_FALSE; 1051 FILE *fout = sdd->std_out ? stdout : stderr; 1052 1053 err = 0; 1054 thissnap = strchr(zhp->zfs_name, '@') + 1; 1055 isfromsnap = (sdd->fromsnap != NULL && 1056 strcmp(sdd->fromsnap, thissnap) == 0); 1057 1058 if (!sdd->seenfrom && isfromsnap) { 1059 gather_holds(zhp, sdd); 1060 sdd->seenfrom = B_TRUE; 1061 (void) strcpy(sdd->prevsnap, thissnap); 1062 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 1063 zfs_close(zhp); 1064 return (0); 1065 } 1066 1067 if (sdd->seento || !sdd->seenfrom) { 1068 zfs_close(zhp); 1069 return (0); 1070 } 1071 1072 istosnap = (strcmp(sdd->tosnap, thissnap) == 0); 1073 if (istosnap) 1074 sdd->seento = B_TRUE; 1075 1076 if (!sdd->doall && !isfromsnap && !istosnap) { 1077 if (sdd->replicate) { 1078 char *snapname; 1079 nvlist_t *snapprops; 1080 /* 1081 * Filter out all intermediate snapshots except origin 1082 * snapshots needed to replicate clones. 1083 */ 1084 nvlist_t *nvfs = fsavl_find(sdd->fsavl, 1085 zhp->zfs_dmustats.dds_guid, &snapname); 1086 1087 VERIFY(0 == nvlist_lookup_nvlist(nvfs, 1088 "snapprops", &snapprops)); 1089 VERIFY(0 == nvlist_lookup_nvlist(snapprops, 1090 thissnap, &snapprops)); 1091 exclude = !nvlist_exists(snapprops, "is_clone_origin"); 1092 } else { 1093 exclude = B_TRUE; 1094 } 1095 } 1096 1097 /* 1098 * If a filter function exists, call it to determine whether 1099 * this snapshot will be sent. 1100 */ 1101 if (exclude || (sdd->filter_cb != NULL && 1102 sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) { 1103 /* 1104 * This snapshot is filtered out. Don't send it, and don't 1105 * set prevsnap_obj, so it will be as if this snapshot didn't 1106 * exist, and the next accepted snapshot will be sent as 1107 * an incremental from the last accepted one, or as the 1108 * first (and full) snapshot in the case of a replication, 1109 * non-incremental send. 1110 */ 1111 zfs_close(zhp); 1112 return (0); 1113 } 1114 1115 gather_holds(zhp, sdd); 1116 fromorigin = sdd->prevsnap[0] == '\0' && 1117 (sdd->fromorigin || sdd->replicate); 1118 1119 if (sdd->verbose) { 1120 uint64_t size; 1121 err = estimate_ioctl(zhp, sdd->prevsnap_obj, 1122 fromorigin, &size); 1123 1124 if (sdd->parsable) { 1125 if (sdd->prevsnap[0] != '\0') { 1126 (void) fprintf(fout, "incremental\t%s\t%s", 1127 sdd->prevsnap, zhp->zfs_name); 1128 } else { 1129 (void) fprintf(fout, "full\t%s", 1130 zhp->zfs_name); 1131 } 1132 } else { 1133 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1134 "send from @%s to %s"), 1135 sdd->prevsnap, zhp->zfs_name); 1136 } 1137 if (err == 0) { 1138 if (sdd->parsable) { 1139 (void) fprintf(fout, "\t%llu\n", 1140 (longlong_t)size); 1141 } else { 1142 char buf[16]; 1143 zfs_nicenum(size, buf, sizeof (buf)); 1144 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1145 " estimated size is %s\n"), buf); 1146 } 1147 sdd->size += size; 1148 } else { 1149 (void) fprintf(fout, "\n"); 1150 } 1151 } 1152 1153 if (!sdd->dryrun) { 1154 /* 1155 * If progress reporting is requested, spawn a new thread to 1156 * poll ZFS_IOC_SEND_PROGRESS at a regular interval. 1157 */ 1158 if (sdd->progress) { 1159 pa.pa_zhp = zhp; 1160 pa.pa_fd = sdd->outfd; 1161 pa.pa_parsable = sdd->parsable; 1162 1163 if (err = pthread_create(&tid, NULL, 1164 send_progress_thread, &pa)) { 1165 zfs_close(zhp); 1166 return (err); 1167 } 1168 } 1169 1170 enum lzc_send_flags flags = 0; 1171 if (sdd->large_block) 1172 flags |= LZC_SEND_FLAG_LARGE_BLOCK; 1173 if (sdd->embed_data) 1174 flags |= LZC_SEND_FLAG_EMBED_DATA; 1175 1176 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj, 1177 fromorigin, sdd->outfd, flags, sdd->debugnv); 1178 1179 if (sdd->progress) { 1180 (void) pthread_cancel(tid); 1181 (void) pthread_join(tid, NULL); 1182 } 1183 } 1184 1185 (void) strcpy(sdd->prevsnap, thissnap); 1186 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 1187 zfs_close(zhp); 1188 return (err); 1189} 1190 1191static int 1192dump_filesystem(zfs_handle_t *zhp, void *arg) 1193{ 1194 int rv = 0; 1195 send_dump_data_t *sdd = arg; 1196 boolean_t missingfrom = B_FALSE; 1197 zfs_cmd_t zc = { 0 }; 1198 1199 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", 1200 zhp->zfs_name, sdd->tosnap); 1201 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) { 1202 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1203 "WARNING: could not send %s@%s: does not exist\n"), 1204 zhp->zfs_name, sdd->tosnap); 1205 sdd->err = B_TRUE; 1206 return (0); 1207 } 1208 1209 if (sdd->replicate && sdd->fromsnap) { 1210 /* 1211 * If this fs does not have fromsnap, and we're doing 1212 * recursive, we need to send a full stream from the 1213 * beginning (or an incremental from the origin if this 1214 * is a clone). If we're doing non-recursive, then let 1215 * them get the error. 1216 */ 1217 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", 1218 zhp->zfs_name, sdd->fromsnap); 1219 if (ioctl(zhp->zfs_hdl->libzfs_fd, 1220 ZFS_IOC_OBJSET_STATS, &zc) != 0) { 1221 missingfrom = B_TRUE; 1222 } 1223 } 1224 1225 sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0; 1226 sdd->prevsnap_obj = 0; 1227 if (sdd->fromsnap == NULL || missingfrom) 1228 sdd->seenfrom = B_TRUE; 1229 1230 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg); 1231 if (!sdd->seenfrom) { 1232 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1233 "WARNING: could not send %s@%s:\n" 1234 "incremental source (%s@%s) does not exist\n"), 1235 zhp->zfs_name, sdd->tosnap, 1236 zhp->zfs_name, sdd->fromsnap); 1237 sdd->err = B_TRUE; 1238 } else if (!sdd->seento) { 1239 if (sdd->fromsnap) { 1240 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1241 "WARNING: could not send %s@%s:\n" 1242 "incremental source (%s@%s) " 1243 "is not earlier than it\n"), 1244 zhp->zfs_name, sdd->tosnap, 1245 zhp->zfs_name, sdd->fromsnap); 1246 } else { 1247 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 1248 "WARNING: " 1249 "could not send %s@%s: does not exist\n"), 1250 zhp->zfs_name, sdd->tosnap); 1251 } 1252 sdd->err = B_TRUE; 1253 } 1254 1255 return (rv); 1256} 1257 1258static int 1259dump_filesystems(zfs_handle_t *rzhp, void *arg) 1260{ 1261 send_dump_data_t *sdd = arg; 1262 nvpair_t *fspair; 1263 boolean_t needagain, progress; 1264 1265 if (!sdd->replicate) 1266 return (dump_filesystem(rzhp, sdd)); 1267 1268 /* Mark the clone origin snapshots. */ 1269 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1270 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1271 nvlist_t *nvfs; 1272 uint64_t origin_guid = 0; 1273 1274 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs)); 1275 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid); 1276 if (origin_guid != 0) { 1277 char *snapname; 1278 nvlist_t *origin_nv = fsavl_find(sdd->fsavl, 1279 origin_guid, &snapname); 1280 if (origin_nv != NULL) { 1281 nvlist_t *snapprops; 1282 VERIFY(0 == nvlist_lookup_nvlist(origin_nv, 1283 "snapprops", &snapprops)); 1284 VERIFY(0 == nvlist_lookup_nvlist(snapprops, 1285 snapname, &snapprops)); 1286 VERIFY(0 == nvlist_add_boolean( 1287 snapprops, "is_clone_origin")); 1288 } 1289 } 1290 } 1291again: 1292 needagain = progress = B_FALSE; 1293 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1294 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1295 nvlist_t *fslist, *parent_nv; 1296 char *fsname; 1297 zfs_handle_t *zhp; 1298 int err; 1299 uint64_t origin_guid = 0; 1300 uint64_t parent_guid = 0; 1301 1302 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); 1303 if (nvlist_lookup_boolean(fslist, "sent") == 0) 1304 continue; 1305 1306 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0); 1307 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid); 1308 (void) nvlist_lookup_uint64(fslist, "parentfromsnap", 1309 &parent_guid); 1310 1311 if (parent_guid != 0) { 1312 parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL); 1313 if (!nvlist_exists(parent_nv, "sent")) { 1314 /* parent has not been sent; skip this one */ 1315 needagain = B_TRUE; 1316 continue; 1317 } 1318 } 1319 1320 if (origin_guid != 0) { 1321 nvlist_t *origin_nv = fsavl_find(sdd->fsavl, 1322 origin_guid, NULL); 1323 if (origin_nv != NULL && 1324 !nvlist_exists(origin_nv, "sent")) { 1325 /* 1326 * origin has not been sent yet; 1327 * skip this clone. 1328 */ 1329 needagain = B_TRUE; 1330 continue; 1331 } 1332 } 1333 1334 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET); 1335 if (zhp == NULL) 1336 return (-1); 1337 err = dump_filesystem(zhp, sdd); 1338 VERIFY(nvlist_add_boolean(fslist, "sent") == 0); 1339 progress = B_TRUE; 1340 zfs_close(zhp); 1341 if (err) 1342 return (err); 1343 } 1344 if (needagain) { 1345 assert(progress); 1346 goto again; 1347 } 1348 1349 /* clean out the sent flags in case we reuse this fss */ 1350 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; 1351 fspair = nvlist_next_nvpair(sdd->fss, fspair)) { 1352 nvlist_t *fslist; 1353 1354 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); 1355 (void) nvlist_remove_all(fslist, "sent"); 1356 } 1357 1358 return (0); 1359} 1360 1361/* 1362 * Generate a send stream for the dataset identified by the argument zhp. 1363 * 1364 * The content of the send stream is the snapshot identified by 1365 * 'tosnap'. Incremental streams are requested in two ways: 1366 * - from the snapshot identified by "fromsnap" (if non-null) or 1367 * - from the origin of the dataset identified by zhp, which must 1368 * be a clone. In this case, "fromsnap" is null and "fromorigin" 1369 * is TRUE. 1370 * 1371 * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and 1372 * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM) 1373 * if "replicate" is set. If "doall" is set, dump all the intermediate 1374 * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall" 1375 * case too. If "props" is set, send properties. 1376 */ 1377int 1378zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, 1379 sendflags_t *flags, int outfd, snapfilter_cb_t filter_func, 1380 void *cb_arg, nvlist_t **debugnvp) 1381{ 1382 char errbuf[1024]; 1383 send_dump_data_t sdd = { 0 }; 1384 int err = 0; 1385 nvlist_t *fss = NULL; 1386 avl_tree_t *fsavl = NULL; 1387 static uint64_t holdseq; 1388 int spa_version; 1389 pthread_t tid = 0; 1390 int pipefd[2]; 1391 dedup_arg_t dda = { 0 }; 1392 int featureflags = 0; 1393 FILE *fout; 1394 1395 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1396 "cannot send '%s'"), zhp->zfs_name); 1397 1398 if (fromsnap && fromsnap[0] == '\0') { 1399 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN, 1400 "zero-length incremental source")); 1401 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf)); 1402 } 1403 1404 if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) { 1405 uint64_t version; 1406 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION); 1407 if (version >= ZPL_VERSION_SA) { 1408 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; 1409 } 1410 } 1411 1412 if (flags->dedup && !flags->dryrun) { 1413 featureflags |= (DMU_BACKUP_FEATURE_DEDUP | 1414 DMU_BACKUP_FEATURE_DEDUPPROPS); 1415 if (err = pipe(pipefd)) { 1416 zfs_error_aux(zhp->zfs_hdl, strerror(errno)); 1417 return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED, 1418 errbuf)); 1419 } 1420 dda.outputfd = outfd; 1421 dda.inputfd = pipefd[1]; 1422 dda.dedup_hdl = zhp->zfs_hdl; 1423 if (err = pthread_create(&tid, NULL, cksummer, &dda)) { 1424 (void) close(pipefd[0]); 1425 (void) close(pipefd[1]); 1426 zfs_error_aux(zhp->zfs_hdl, strerror(errno)); 1427 return (zfs_error(zhp->zfs_hdl, 1428 EZFS_THREADCREATEFAILED, errbuf)); 1429 } 1430 } 1431 1432 if (flags->replicate || flags->doall || flags->props) { 1433 dmu_replay_record_t drr = { 0 }; 1434 char *packbuf = NULL; 1435 size_t buflen = 0; 1436 zio_cksum_t zc = { 0 }; 1437 1438 if (flags->replicate || flags->props) { 1439 nvlist_t *hdrnv; 1440 1441 VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0)); 1442 if (fromsnap) { 1443 VERIFY(0 == nvlist_add_string(hdrnv, 1444 "fromsnap", fromsnap)); 1445 } 1446 VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap)); 1447 if (!flags->replicate) { 1448 VERIFY(0 == nvlist_add_boolean(hdrnv, 1449 "not_recursive")); 1450 } 1451 1452 err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name, 1453 fromsnap, tosnap, flags->replicate, &fss, &fsavl); 1454 if (err) 1455 goto err_out; 1456 VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss)); 1457 err = nvlist_pack(hdrnv, &packbuf, &buflen, 1458 NV_ENCODE_XDR, 0); 1459 if (debugnvp) 1460 *debugnvp = hdrnv; 1461 else 1462 nvlist_free(hdrnv); 1463 if (err) 1464 goto stderr_out; 1465 } 1466 1467 if (!flags->dryrun) { 1468 /* write first begin record */ 1469 drr.drr_type = DRR_BEGIN; 1470 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 1471 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin. 1472 drr_versioninfo, DMU_COMPOUNDSTREAM); 1473 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin. 1474 drr_versioninfo, featureflags); 1475 (void) snprintf(drr.drr_u.drr_begin.drr_toname, 1476 sizeof (drr.drr_u.drr_begin.drr_toname), 1477 "%s@%s", zhp->zfs_name, tosnap); 1478 drr.drr_payloadlen = buflen; 1479 1480 err = dump_record(&drr, packbuf, buflen, &zc, outfd); 1481 free(packbuf); 1482 if (err != 0) 1483 goto stderr_out; 1484 1485 /* write end record */ 1486 bzero(&drr, sizeof (drr)); 1487 drr.drr_type = DRR_END; 1488 drr.drr_u.drr_end.drr_checksum = zc; 1489 err = write(outfd, &drr, sizeof (drr)); 1490 if (err == -1) { 1491 err = errno; 1492 goto stderr_out; 1493 } 1494 1495 err = 0; 1496 } 1497 } 1498 1499 /* dump each stream */ 1500 sdd.fromsnap = fromsnap; 1501 sdd.tosnap = tosnap; 1502 if (tid != 0) 1503 sdd.outfd = pipefd[0]; 1504 else 1505 sdd.outfd = outfd; 1506 sdd.replicate = flags->replicate; 1507 sdd.doall = flags->doall; 1508 sdd.fromorigin = flags->fromorigin; 1509 sdd.fss = fss; 1510 sdd.fsavl = fsavl; 1511 sdd.verbose = flags->verbose; 1512 sdd.parsable = flags->parsable; 1513 sdd.progress = flags->progress; 1514 sdd.dryrun = flags->dryrun; 1515 sdd.large_block = flags->largeblock; 1516 sdd.embed_data = flags->embed_data; 1517 sdd.filter_cb = filter_func; 1518 sdd.filter_cb_arg = cb_arg; 1519 if (debugnvp) 1520 sdd.debugnv = *debugnvp; 1521 if (sdd.verbose && sdd.dryrun) 1522 sdd.std_out = B_TRUE; 1523 fout = sdd.std_out ? stdout : stderr; 1524 1525 /* 1526 * Some flags require that we place user holds on the datasets that are 1527 * being sent so they don't get destroyed during the send. We can skip 1528 * this step if the pool is imported read-only since the datasets cannot 1529 * be destroyed. 1530 */ 1531 if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp), 1532 ZPOOL_PROP_READONLY, NULL) && 1533 zfs_spa_version(zhp, &spa_version) == 0 && 1534 spa_version >= SPA_VERSION_USERREFS && 1535 (flags->doall || flags->replicate)) { 1536 ++holdseq; 1537 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag), 1538 ".send-%d-%llu", getpid(), (u_longlong_t)holdseq); 1539 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); 1540 if (sdd.cleanup_fd < 0) { 1541 err = errno; 1542 goto stderr_out; 1543 } 1544 sdd.snapholds = fnvlist_alloc(); 1545 } else { 1546 sdd.cleanup_fd = -1; 1547 sdd.snapholds = NULL; 1548 } 1549 if (flags->verbose || sdd.snapholds != NULL) { 1550 /* 1551 * Do a verbose no-op dry run to get all the verbose output 1552 * or to gather snapshot hold's before generating any data, 1553 * then do a non-verbose real run to generate the streams. 1554 */ 1555 sdd.dryrun = B_TRUE; 1556 err = dump_filesystems(zhp, &sdd); 1557 1558 if (err != 0) 1559 goto stderr_out; 1560 1561 if (flags->verbose) { 1562 if (flags->parsable) { 1563 (void) fprintf(fout, "size\t%llu\n", 1564 (longlong_t)sdd.size); 1565 } else { 1566 char buf[16]; 1567 zfs_nicenum(sdd.size, buf, sizeof (buf)); 1568 (void) fprintf(fout, dgettext(TEXT_DOMAIN, 1569 "total estimated size is %s\n"), buf); 1570 } 1571 } 1572 1573 /* Ensure no snaps found is treated as an error. */ 1574 if (!sdd.seento) { 1575 err = ENOENT; 1576 goto err_out; 1577 } 1578 1579 /* Skip the second run if dryrun was requested. */ 1580 if (flags->dryrun) 1581 goto err_out; 1582 1583 if (sdd.snapholds != NULL) { 1584 err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds); 1585 if (err != 0) 1586 goto stderr_out; 1587 1588 fnvlist_free(sdd.snapholds); 1589 sdd.snapholds = NULL; 1590 } 1591 1592 sdd.dryrun = B_FALSE; 1593 sdd.verbose = B_FALSE; 1594 } 1595 1596 err = dump_filesystems(zhp, &sdd); 1597 fsavl_destroy(fsavl); 1598 nvlist_free(fss); 1599 1600 /* Ensure no snaps found is treated as an error. */ 1601 if (err == 0 && !sdd.seento) 1602 err = ENOENT; 1603 1604 if (tid != 0) { 1605 if (err != 0) 1606 (void) pthread_cancel(tid); 1607 (void) close(pipefd[0]); 1608 (void) pthread_join(tid, NULL); 1609 } 1610 1611 if (sdd.cleanup_fd != -1) { 1612 VERIFY(0 == close(sdd.cleanup_fd)); 1613 sdd.cleanup_fd = -1; 1614 } 1615 1616 if (!flags->dryrun && (flags->replicate || flags->doall || 1617 flags->props)) { 1618 /* 1619 * write final end record. NB: want to do this even if 1620 * there was some error, because it might not be totally 1621 * failed. 1622 */ 1623 dmu_replay_record_t drr = { 0 }; 1624 drr.drr_type = DRR_END; 1625 if (write(outfd, &drr, sizeof (drr)) == -1) { 1626 return (zfs_standard_error(zhp->zfs_hdl, 1627 errno, errbuf)); 1628 } 1629 } 1630 1631 return (err || sdd.err); 1632 1633stderr_out: 1634 err = zfs_standard_error(zhp->zfs_hdl, err, errbuf); 1635err_out: 1636 fsavl_destroy(fsavl); 1637 nvlist_free(fss); 1638 fnvlist_free(sdd.snapholds); 1639 1640 if (sdd.cleanup_fd != -1) 1641 VERIFY(0 == close(sdd.cleanup_fd)); 1642 if (tid != 0) { 1643 (void) pthread_cancel(tid); 1644 (void) close(pipefd[0]); 1645 (void) pthread_join(tid, NULL); 1646 } 1647 return (err); 1648} 1649 1650int 1651zfs_send_one(zfs_handle_t *zhp, const char *from, int fd, 1652 enum lzc_send_flags flags) 1653{ 1654 int err; 1655 libzfs_handle_t *hdl = zhp->zfs_hdl; 1656 1657 char errbuf[1024]; 1658 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 1659 "warning: cannot send '%s'"), zhp->zfs_name); 1660 1661 err = lzc_send(zhp->zfs_name, from, fd, flags); 1662 if (err != 0) { 1663 switch (errno) { 1664 case EXDEV: 1665 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1666 "not an earlier snapshot from the same fs")); 1667 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); 1668 1669 case ENOENT: 1670 case ESRCH: 1671 if (lzc_exists(zhp->zfs_name)) { 1672 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1673 "incremental source (%s) does not exist"), 1674 from); 1675 } 1676 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 1677 1678 case EBUSY: 1679 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1680 "target is busy; if a filesystem, " 1681 "it must not be mounted")); 1682 return (zfs_error(hdl, EZFS_BUSY, errbuf)); 1683 1684 case EDQUOT: 1685 case EFBIG: 1686 case EIO: 1687 case ENOLINK: 1688 case ENOSPC: 1689#ifdef illumos 1690 case ENOSTR: 1691#endif 1692 case ENXIO: 1693 case EPIPE: 1694 case ERANGE: 1695 case EFAULT: 1696 case EROFS: 1697 zfs_error_aux(hdl, strerror(errno)); 1698 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); 1699 1700 default: 1701 return (zfs_standard_error(hdl, errno, errbuf)); 1702 } 1703 } 1704 return (err != 0); 1705} 1706 1707/* 1708 * Routines specific to "zfs recv" 1709 */ 1710 1711static int 1712recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen, 1713 boolean_t byteswap, zio_cksum_t *zc) 1714{ 1715 char *cp = buf; 1716 int rv; 1717 int len = ilen; 1718 1719 assert(ilen <= SPA_MAXBLOCKSIZE); 1720 1721 do { 1722 rv = read(fd, cp, len); 1723 cp += rv; 1724 len -= rv; 1725 } while (rv > 0); 1726 1727 if (rv < 0 || len != 0) { 1728 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 1729 "failed to read from stream")); 1730 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN, 1731 "cannot receive"))); 1732 } 1733 1734 if (zc) { 1735 if (byteswap) 1736 fletcher_4_incremental_byteswap(buf, ilen, zc); 1737 else 1738 fletcher_4_incremental_native(buf, ilen, zc); 1739 } 1740 return (0); 1741} 1742 1743static int 1744recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp, 1745 boolean_t byteswap, zio_cksum_t *zc) 1746{ 1747 char *buf; 1748 int err; 1749 1750 buf = zfs_alloc(hdl, len); 1751 if (buf == NULL) 1752 return (ENOMEM); 1753 1754 err = recv_read(hdl, fd, buf, len, byteswap, zc); 1755 if (err != 0) { 1756 free(buf); 1757 return (err); 1758 } 1759 1760 err = nvlist_unpack(buf, len, nvp, 0); 1761 free(buf); 1762 if (err != 0) { 1763 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 1764 "stream (malformed nvlist)")); 1765 return (EINVAL); 1766 } 1767 return (0); 1768} 1769 1770static int 1771recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname, 1772 int baselen, char *newname, recvflags_t *flags) 1773{ 1774 static int seq; 1775 zfs_cmd_t zc = { 0 }; 1776 int err; 1777 prop_changelist_t *clp; 1778 zfs_handle_t *zhp; 1779 1780 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 1781 if (zhp == NULL) 1782 return (-1); 1783 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 1784 flags->force ? MS_FORCE : 0); 1785 zfs_close(zhp); 1786 if (clp == NULL) 1787 return (-1); 1788 err = changelist_prefix(clp); 1789 if (err) 1790 return (err); 1791 1792 zc.zc_objset_type = DMU_OST_ZFS; 1793 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 1794 1795 if (tryname) { 1796 (void) strcpy(newname, tryname); 1797 1798 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value)); 1799 1800 if (flags->verbose) { 1801 (void) printf("attempting rename %s to %s\n", 1802 zc.zc_name, zc.zc_value); 1803 } 1804 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc); 1805 if (err == 0) 1806 changelist_rename(clp, name, tryname); 1807 } else { 1808 err = ENOENT; 1809 } 1810 1811 if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) { 1812 seq++; 1813 1814 (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u", 1815 baselen, name, getpid(), seq); 1816 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value)); 1817 1818 if (flags->verbose) { 1819 (void) printf("failed - trying rename %s to %s\n", 1820 zc.zc_name, zc.zc_value); 1821 } 1822 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc); 1823 if (err == 0) 1824 changelist_rename(clp, name, newname); 1825 if (err && flags->verbose) { 1826 (void) printf("failed (%u) - " 1827 "will try again on next pass\n", errno); 1828 } 1829 err = EAGAIN; 1830 } else if (flags->verbose) { 1831 if (err == 0) 1832 (void) printf("success\n"); 1833 else 1834 (void) printf("failed (%u)\n", errno); 1835 } 1836 1837 (void) changelist_postfix(clp); 1838 changelist_free(clp); 1839 1840 return (err); 1841} 1842 1843static int 1844recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen, 1845 char *newname, recvflags_t *flags) 1846{ 1847 zfs_cmd_t zc = { 0 }; 1848 int err = 0; 1849 prop_changelist_t *clp; 1850 zfs_handle_t *zhp; 1851 boolean_t defer = B_FALSE; 1852 int spa_version; 1853 1854 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); 1855 if (zhp == NULL) 1856 return (-1); 1857 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 1858 flags->force ? MS_FORCE : 0); 1859 if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT && 1860 zfs_spa_version(zhp, &spa_version) == 0 && 1861 spa_version >= SPA_VERSION_USERREFS) 1862 defer = B_TRUE; 1863 zfs_close(zhp); 1864 if (clp == NULL) 1865 return (-1); 1866 err = changelist_prefix(clp); 1867 if (err) 1868 return (err); 1869 1870 zc.zc_objset_type = DMU_OST_ZFS; 1871 zc.zc_defer_destroy = defer; 1872 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 1873 1874 if (flags->verbose) 1875 (void) printf("attempting destroy %s\n", zc.zc_name); 1876 err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc); 1877 if (err == 0) { 1878 if (flags->verbose) 1879 (void) printf("success\n"); 1880 changelist_remove(clp, zc.zc_name); 1881 } 1882 1883 (void) changelist_postfix(clp); 1884 changelist_free(clp); 1885 1886 /* 1887 * Deferred destroy might destroy the snapshot or only mark it to be 1888 * destroyed later, and it returns success in either case. 1889 */ 1890 if (err != 0 || (defer && zfs_dataset_exists(hdl, name, 1891 ZFS_TYPE_SNAPSHOT))) { 1892 err = recv_rename(hdl, name, NULL, baselen, newname, flags); 1893 } 1894 1895 return (err); 1896} 1897 1898typedef struct guid_to_name_data { 1899 uint64_t guid; 1900 char *name; 1901 char *skip; 1902} guid_to_name_data_t; 1903 1904static int 1905guid_to_name_cb(zfs_handle_t *zhp, void *arg) 1906{ 1907 guid_to_name_data_t *gtnd = arg; 1908 int err; 1909 1910 if (gtnd->skip != NULL && 1911 strcmp(zhp->zfs_name, gtnd->skip) == 0) { 1912 return (0); 1913 } 1914 1915 if (zhp->zfs_dmustats.dds_guid == gtnd->guid) { 1916 (void) strcpy(gtnd->name, zhp->zfs_name); 1917 zfs_close(zhp); 1918 return (EEXIST); 1919 } 1920 1921 err = zfs_iter_children(zhp, guid_to_name_cb, gtnd); 1922 zfs_close(zhp); 1923 return (err); 1924} 1925 1926/* 1927 * Attempt to find the local dataset associated with this guid. In the case of 1928 * multiple matches, we attempt to find the "best" match by searching 1929 * progressively larger portions of the hierarchy. This allows one to send a 1930 * tree of datasets individually and guarantee that we will find the source 1931 * guid within that hierarchy, even if there are multiple matches elsewhere. 1932 */ 1933static int 1934guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid, 1935 char *name) 1936{ 1937 /* exhaustive search all local snapshots */ 1938 char pname[ZFS_MAXNAMELEN]; 1939 guid_to_name_data_t gtnd; 1940 int err = 0; 1941 zfs_handle_t *zhp; 1942 char *cp; 1943 1944 gtnd.guid = guid; 1945 gtnd.name = name; 1946 gtnd.skip = NULL; 1947 1948 (void) strlcpy(pname, parent, sizeof (pname)); 1949 1950 /* 1951 * Search progressively larger portions of the hierarchy. This will 1952 * select the "most local" version of the origin snapshot in the case 1953 * that there are multiple matching snapshots in the system. 1954 */ 1955 while ((cp = strrchr(pname, '/')) != NULL) { 1956 1957 /* Chop off the last component and open the parent */ 1958 *cp = '\0'; 1959 zhp = make_dataset_handle(hdl, pname); 1960 1961 if (zhp == NULL) 1962 continue; 1963 1964 err = zfs_iter_children(zhp, guid_to_name_cb, >nd); 1965 zfs_close(zhp); 1966 if (err == EEXIST) 1967 return (0); 1968 1969 /* 1970 * Remember the dataset that we already searched, so we 1971 * skip it next time through. 1972 */ 1973 gtnd.skip = pname; 1974 } 1975 1976 return (ENOENT); 1977} 1978 1979/* 1980 * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if 1981 * guid1 is after guid2. 1982 */ 1983static int 1984created_before(libzfs_handle_t *hdl, avl_tree_t *avl, 1985 uint64_t guid1, uint64_t guid2) 1986{ 1987 nvlist_t *nvfs; 1988 char *fsname, *snapname; 1989 char buf[ZFS_MAXNAMELEN]; 1990 int rv; 1991 zfs_handle_t *guid1hdl, *guid2hdl; 1992 uint64_t create1, create2; 1993 1994 if (guid2 == 0) 1995 return (0); 1996 if (guid1 == 0) 1997 return (1); 1998 1999 nvfs = fsavl_find(avl, guid1, &snapname); 2000 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2001 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); 2002 guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); 2003 if (guid1hdl == NULL) 2004 return (-1); 2005 2006 nvfs = fsavl_find(avl, guid2, &snapname); 2007 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2008 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); 2009 guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); 2010 if (guid2hdl == NULL) { 2011 zfs_close(guid1hdl); 2012 return (-1); 2013 } 2014 2015 create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG); 2016 create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG); 2017 2018 if (create1 < create2) 2019 rv = -1; 2020 else if (create1 > create2) 2021 rv = +1; 2022 else 2023 rv = 0; 2024 2025 zfs_close(guid1hdl); 2026 zfs_close(guid2hdl); 2027 2028 return (rv); 2029} 2030 2031static int 2032recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs, 2033 recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl, 2034 nvlist_t *renamed) 2035{ 2036 nvlist_t *local_nv, *deleted = NULL; 2037 avl_tree_t *local_avl; 2038 nvpair_t *fselem, *nextfselem; 2039 char *fromsnap; 2040 char newname[ZFS_MAXNAMELEN]; 2041 char guidname[32]; 2042 int error; 2043 boolean_t needagain, progress, recursive; 2044 char *s1, *s2; 2045 2046 VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap)); 2047 2048 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2049 ENOENT); 2050 2051 if (flags->dryrun) 2052 return (0); 2053 2054again: 2055 needagain = progress = B_FALSE; 2056 2057 VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0)); 2058 2059 if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL, 2060 recursive, &local_nv, &local_avl)) != 0) 2061 return (error); 2062 2063 /* 2064 * Process deletes and renames 2065 */ 2066 for (fselem = nvlist_next_nvpair(local_nv, NULL); 2067 fselem; fselem = nextfselem) { 2068 nvlist_t *nvfs, *snaps; 2069 nvlist_t *stream_nvfs = NULL; 2070 nvpair_t *snapelem, *nextsnapelem; 2071 uint64_t fromguid = 0; 2072 uint64_t originguid = 0; 2073 uint64_t stream_originguid = 0; 2074 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid; 2075 char *fsname, *stream_fsname; 2076 2077 nextfselem = nvlist_next_nvpair(local_nv, fselem); 2078 2079 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); 2080 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); 2081 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); 2082 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap", 2083 &parent_fromsnap_guid)); 2084 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid); 2085 2086 /* 2087 * First find the stream's fs, so we can check for 2088 * a different origin (due to "zfs promote") 2089 */ 2090 for (snapelem = nvlist_next_nvpair(snaps, NULL); 2091 snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) { 2092 uint64_t thisguid; 2093 2094 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); 2095 stream_nvfs = fsavl_find(stream_avl, thisguid, NULL); 2096 2097 if (stream_nvfs != NULL) 2098 break; 2099 } 2100 2101 /* check for promote */ 2102 (void) nvlist_lookup_uint64(stream_nvfs, "origin", 2103 &stream_originguid); 2104 if (stream_nvfs && originguid != stream_originguid) { 2105 switch (created_before(hdl, local_avl, 2106 stream_originguid, originguid)) { 2107 case 1: { 2108 /* promote it! */ 2109 zfs_cmd_t zc = { 0 }; 2110 nvlist_t *origin_nvfs; 2111 char *origin_fsname; 2112 2113 if (flags->verbose) 2114 (void) printf("promoting %s\n", fsname); 2115 2116 origin_nvfs = fsavl_find(local_avl, originguid, 2117 NULL); 2118 VERIFY(0 == nvlist_lookup_string(origin_nvfs, 2119 "name", &origin_fsname)); 2120 (void) strlcpy(zc.zc_value, origin_fsname, 2121 sizeof (zc.zc_value)); 2122 (void) strlcpy(zc.zc_name, fsname, 2123 sizeof (zc.zc_name)); 2124 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc); 2125 if (error == 0) 2126 progress = B_TRUE; 2127 break; 2128 } 2129 default: 2130 break; 2131 case -1: 2132 fsavl_destroy(local_avl); 2133 nvlist_free(local_nv); 2134 return (-1); 2135 } 2136 /* 2137 * We had/have the wrong origin, therefore our 2138 * list of snapshots is wrong. Need to handle 2139 * them on the next pass. 2140 */ 2141 needagain = B_TRUE; 2142 continue; 2143 } 2144 2145 for (snapelem = nvlist_next_nvpair(snaps, NULL); 2146 snapelem; snapelem = nextsnapelem) { 2147 uint64_t thisguid; 2148 char *stream_snapname; 2149 nvlist_t *found, *props; 2150 2151 nextsnapelem = nvlist_next_nvpair(snaps, snapelem); 2152 2153 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); 2154 found = fsavl_find(stream_avl, thisguid, 2155 &stream_snapname); 2156 2157 /* check for delete */ 2158 if (found == NULL) { 2159 char name[ZFS_MAXNAMELEN]; 2160 2161 if (!flags->force) 2162 continue; 2163 2164 (void) snprintf(name, sizeof (name), "%s@%s", 2165 fsname, nvpair_name(snapelem)); 2166 2167 error = recv_destroy(hdl, name, 2168 strlen(fsname)+1, newname, flags); 2169 if (error) 2170 needagain = B_TRUE; 2171 else 2172 progress = B_TRUE; 2173 sprintf(guidname, "%lu", thisguid); 2174 nvlist_add_boolean(deleted, guidname); 2175 continue; 2176 } 2177 2178 stream_nvfs = found; 2179 2180 if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops", 2181 &props) && 0 == nvlist_lookup_nvlist(props, 2182 stream_snapname, &props)) { 2183 zfs_cmd_t zc = { 0 }; 2184 2185 zc.zc_cookie = B_TRUE; /* received */ 2186 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), 2187 "%s@%s", fsname, nvpair_name(snapelem)); 2188 if (zcmd_write_src_nvlist(hdl, &zc, 2189 props) == 0) { 2190 (void) zfs_ioctl(hdl, 2191 ZFS_IOC_SET_PROP, &zc); 2192 zcmd_free_nvlists(&zc); 2193 } 2194 } 2195 2196 /* check for different snapname */ 2197 if (strcmp(nvpair_name(snapelem), 2198 stream_snapname) != 0) { 2199 char name[ZFS_MAXNAMELEN]; 2200 char tryname[ZFS_MAXNAMELEN]; 2201 2202 (void) snprintf(name, sizeof (name), "%s@%s", 2203 fsname, nvpair_name(snapelem)); 2204 (void) snprintf(tryname, sizeof (name), "%s@%s", 2205 fsname, stream_snapname); 2206 2207 error = recv_rename(hdl, name, tryname, 2208 strlen(fsname)+1, newname, flags); 2209 if (error) 2210 needagain = B_TRUE; 2211 else 2212 progress = B_TRUE; 2213 } 2214 2215 if (strcmp(stream_snapname, fromsnap) == 0) 2216 fromguid = thisguid; 2217 } 2218 2219 /* check for delete */ 2220 if (stream_nvfs == NULL) { 2221 if (!flags->force) 2222 continue; 2223 2224 error = recv_destroy(hdl, fsname, strlen(tofs)+1, 2225 newname, flags); 2226 if (error) 2227 needagain = B_TRUE; 2228 else 2229 progress = B_TRUE; 2230 sprintf(guidname, "%lu", parent_fromsnap_guid); 2231 nvlist_add_boolean(deleted, guidname); 2232 continue; 2233 } 2234 2235 if (fromguid == 0) { 2236 if (flags->verbose) { 2237 (void) printf("local fs %s does not have " 2238 "fromsnap (%s in stream); must have " 2239 "been deleted locally; ignoring\n", 2240 fsname, fromsnap); 2241 } 2242 continue; 2243 } 2244 2245 VERIFY(0 == nvlist_lookup_string(stream_nvfs, 2246 "name", &stream_fsname)); 2247 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs, 2248 "parentfromsnap", &stream_parent_fromsnap_guid)); 2249 2250 s1 = strrchr(fsname, '/'); 2251 s2 = strrchr(stream_fsname, '/'); 2252 2253 /* 2254 * Check if we're going to rename based on parent guid change 2255 * and the current parent guid was also deleted. If it was then 2256 * rename will fail and is likely unneeded, so avoid this and 2257 * force an early retry to determine the new 2258 * parent_fromsnap_guid. 2259 */ 2260 if (stream_parent_fromsnap_guid != 0 && 2261 parent_fromsnap_guid != 0 && 2262 stream_parent_fromsnap_guid != parent_fromsnap_guid) { 2263 sprintf(guidname, "%lu", parent_fromsnap_guid); 2264 if (nvlist_exists(deleted, guidname)) { 2265 progress = B_TRUE; 2266 needagain = B_TRUE; 2267 goto doagain; 2268 } 2269 } 2270 2271 /* 2272 * Check for rename. If the exact receive path is specified, it 2273 * does not count as a rename, but we still need to check the 2274 * datasets beneath it. 2275 */ 2276 if ((stream_parent_fromsnap_guid != 0 && 2277 parent_fromsnap_guid != 0 && 2278 stream_parent_fromsnap_guid != parent_fromsnap_guid) || 2279 ((flags->isprefix || strcmp(tofs, fsname) != 0) && 2280 (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) { 2281 nvlist_t *parent; 2282 char tryname[ZFS_MAXNAMELEN]; 2283 2284 parent = fsavl_find(local_avl, 2285 stream_parent_fromsnap_guid, NULL); 2286 /* 2287 * NB: parent might not be found if we used the 2288 * tosnap for stream_parent_fromsnap_guid, 2289 * because the parent is a newly-created fs; 2290 * we'll be able to rename it after we recv the 2291 * new fs. 2292 */ 2293 if (parent != NULL) { 2294 char *pname; 2295 2296 VERIFY(0 == nvlist_lookup_string(parent, "name", 2297 &pname)); 2298 (void) snprintf(tryname, sizeof (tryname), 2299 "%s%s", pname, strrchr(stream_fsname, '/')); 2300 } else { 2301 tryname[0] = '\0'; 2302 if (flags->verbose) { 2303 (void) printf("local fs %s new parent " 2304 "not found\n", fsname); 2305 } 2306 } 2307 2308 newname[0] = '\0'; 2309 2310 error = recv_rename(hdl, fsname, tryname, 2311 strlen(tofs)+1, newname, flags); 2312 2313 if (renamed != NULL && newname[0] != '\0') { 2314 VERIFY(0 == nvlist_add_boolean(renamed, 2315 newname)); 2316 } 2317 2318 if (error) 2319 needagain = B_TRUE; 2320 else 2321 progress = B_TRUE; 2322 } 2323 } 2324 2325doagain: 2326 fsavl_destroy(local_avl); 2327 nvlist_free(local_nv); 2328 nvlist_free(deleted); 2329 2330 if (needagain && progress) { 2331 /* do another pass to fix up temporary names */ 2332 if (flags->verbose) 2333 (void) printf("another pass:\n"); 2334 goto again; 2335 } 2336 2337 return (needagain); 2338} 2339 2340static int 2341zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname, 2342 recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc, 2343 char **top_zfs, int cleanup_fd, uint64_t *action_handlep) 2344{ 2345 nvlist_t *stream_nv = NULL; 2346 avl_tree_t *stream_avl = NULL; 2347 char *fromsnap = NULL; 2348 char *cp; 2349 char tofs[ZFS_MAXNAMELEN]; 2350 char sendfs[ZFS_MAXNAMELEN]; 2351 char errbuf[1024]; 2352 dmu_replay_record_t drre; 2353 int error; 2354 boolean_t anyerr = B_FALSE; 2355 boolean_t softerr = B_FALSE; 2356 boolean_t recursive; 2357 2358 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2359 "cannot receive")); 2360 2361 assert(drr->drr_type == DRR_BEGIN); 2362 assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC); 2363 assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) == 2364 DMU_COMPOUNDSTREAM); 2365 2366 /* 2367 * Read in the nvlist from the stream. 2368 */ 2369 if (drr->drr_payloadlen != 0) { 2370 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen, 2371 &stream_nv, flags->byteswap, zc); 2372 if (error) { 2373 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2374 goto out; 2375 } 2376 } 2377 2378 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2379 ENOENT); 2380 2381 if (recursive && strchr(destname, '@')) { 2382 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2383 "cannot specify snapshot name for multi-snapshot stream")); 2384 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2385 goto out; 2386 } 2387 2388 /* 2389 * Read in the end record and verify checksum. 2390 */ 2391 if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre), 2392 flags->byteswap, NULL))) 2393 goto out; 2394 if (flags->byteswap) { 2395 drre.drr_type = BSWAP_32(drre.drr_type); 2396 drre.drr_u.drr_end.drr_checksum.zc_word[0] = 2397 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]); 2398 drre.drr_u.drr_end.drr_checksum.zc_word[1] = 2399 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]); 2400 drre.drr_u.drr_end.drr_checksum.zc_word[2] = 2401 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]); 2402 drre.drr_u.drr_end.drr_checksum.zc_word[3] = 2403 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]); 2404 } 2405 if (drre.drr_type != DRR_END) { 2406 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2407 goto out; 2408 } 2409 if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) { 2410 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2411 "incorrect header checksum")); 2412 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); 2413 goto out; 2414 } 2415 2416 (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap); 2417 2418 if (drr->drr_payloadlen != 0) { 2419 nvlist_t *stream_fss; 2420 2421 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss", 2422 &stream_fss)); 2423 if ((stream_avl = fsavl_create(stream_fss)) == NULL) { 2424 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2425 "couldn't allocate avl tree")); 2426 error = zfs_error(hdl, EZFS_NOMEM, errbuf); 2427 goto out; 2428 } 2429 2430 if (fromsnap != NULL) { 2431 nvlist_t *renamed = NULL; 2432 nvpair_t *pair = NULL; 2433 2434 (void) strlcpy(tofs, destname, ZFS_MAXNAMELEN); 2435 if (flags->isprefix) { 2436 struct drr_begin *drrb = &drr->drr_u.drr_begin; 2437 int i; 2438 2439 if (flags->istail) { 2440 cp = strrchr(drrb->drr_toname, '/'); 2441 if (cp == NULL) { 2442 (void) strlcat(tofs, "/", 2443 ZFS_MAXNAMELEN); 2444 i = 0; 2445 } else { 2446 i = (cp - drrb->drr_toname); 2447 } 2448 } else { 2449 i = strcspn(drrb->drr_toname, "/@"); 2450 } 2451 /* zfs_receive_one() will create_parents() */ 2452 (void) strlcat(tofs, &drrb->drr_toname[i], 2453 ZFS_MAXNAMELEN); 2454 *strchr(tofs, '@') = '\0'; 2455 } 2456 2457 if (recursive && !flags->dryrun && !flags->nomount) { 2458 VERIFY(0 == nvlist_alloc(&renamed, 2459 NV_UNIQUE_NAME, 0)); 2460 } 2461 2462 softerr = recv_incremental_replication(hdl, tofs, flags, 2463 stream_nv, stream_avl, renamed); 2464 2465 /* Unmount renamed filesystems before receiving. */ 2466 while ((pair = nvlist_next_nvpair(renamed, 2467 pair)) != NULL) { 2468 zfs_handle_t *zhp; 2469 prop_changelist_t *clp = NULL; 2470 2471 zhp = zfs_open(hdl, nvpair_name(pair), 2472 ZFS_TYPE_FILESYSTEM); 2473 if (zhp != NULL) { 2474 clp = changelist_gather(zhp, 2475 ZFS_PROP_MOUNTPOINT, 0, 0); 2476 zfs_close(zhp); 2477 if (clp != NULL) { 2478 softerr |= 2479 changelist_prefix(clp); 2480 changelist_free(clp); 2481 } 2482 } 2483 } 2484 2485 nvlist_free(renamed); 2486 } 2487 } 2488 2489 /* 2490 * Get the fs specified by the first path in the stream (the top level 2491 * specified by 'zfs send') and pass it to each invocation of 2492 * zfs_receive_one(). 2493 */ 2494 (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname, 2495 ZFS_MAXNAMELEN); 2496 if ((cp = strchr(sendfs, '@')) != NULL) 2497 *cp = '\0'; 2498 2499 /* Finally, receive each contained stream */ 2500 do { 2501 /* 2502 * we should figure out if it has a recoverable 2503 * error, in which case do a recv_skip() and drive on. 2504 * Note, if we fail due to already having this guid, 2505 * zfs_receive_one() will take care of it (ie, 2506 * recv_skip() and return 0). 2507 */ 2508 error = zfs_receive_impl(hdl, destname, NULL, flags, fd, 2509 sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd, 2510 action_handlep); 2511 if (error == ENODATA) { 2512 error = 0; 2513 break; 2514 } 2515 anyerr |= error; 2516 } while (error == 0); 2517 2518 if (drr->drr_payloadlen != 0 && fromsnap != NULL) { 2519 /* 2520 * Now that we have the fs's they sent us, try the 2521 * renames again. 2522 */ 2523 softerr = recv_incremental_replication(hdl, tofs, flags, 2524 stream_nv, stream_avl, NULL); 2525 } 2526 2527out: 2528 fsavl_destroy(stream_avl); 2529 if (stream_nv) 2530 nvlist_free(stream_nv); 2531 if (softerr) 2532 error = -2; 2533 if (anyerr) 2534 error = -1; 2535 return (error); 2536} 2537 2538static void 2539trunc_prop_errs(int truncated) 2540{ 2541 ASSERT(truncated != 0); 2542 2543 if (truncated == 1) 2544 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 2545 "1 more property could not be set\n")); 2546 else 2547 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, 2548 "%d more properties could not be set\n"), truncated); 2549} 2550 2551static int 2552recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap) 2553{ 2554 dmu_replay_record_t *drr; 2555 void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE); 2556 char errbuf[1024]; 2557 2558 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2559 "cannot receive:")); 2560 2561 /* XXX would be great to use lseek if possible... */ 2562 drr = buf; 2563 2564 while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t), 2565 byteswap, NULL) == 0) { 2566 if (byteswap) 2567 drr->drr_type = BSWAP_32(drr->drr_type); 2568 2569 switch (drr->drr_type) { 2570 case DRR_BEGIN: 2571 /* NB: not to be used on v2 stream packages */ 2572 if (drr->drr_payloadlen != 0) { 2573 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2574 "invalid substream header")); 2575 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 2576 } 2577 break; 2578 2579 case DRR_END: 2580 free(buf); 2581 return (0); 2582 2583 case DRR_OBJECT: 2584 if (byteswap) { 2585 drr->drr_u.drr_object.drr_bonuslen = 2586 BSWAP_32(drr->drr_u.drr_object. 2587 drr_bonuslen); 2588 } 2589 (void) recv_read(hdl, fd, buf, 2590 P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8), 2591 B_FALSE, NULL); 2592 break; 2593 2594 case DRR_WRITE: 2595 if (byteswap) { 2596 drr->drr_u.drr_write.drr_length = 2597 BSWAP_64(drr->drr_u.drr_write.drr_length); 2598 } 2599 (void) recv_read(hdl, fd, buf, 2600 drr->drr_u.drr_write.drr_length, B_FALSE, NULL); 2601 break; 2602 case DRR_SPILL: 2603 if (byteswap) { 2604 drr->drr_u.drr_write.drr_length = 2605 BSWAP_64(drr->drr_u.drr_spill.drr_length); 2606 } 2607 (void) recv_read(hdl, fd, buf, 2608 drr->drr_u.drr_spill.drr_length, B_FALSE, NULL); 2609 break; 2610 case DRR_WRITE_EMBEDDED: 2611 if (byteswap) { 2612 drr->drr_u.drr_write_embedded.drr_psize = 2613 BSWAP_32(drr->drr_u.drr_write_embedded. 2614 drr_psize); 2615 } 2616 (void) recv_read(hdl, fd, buf, 2617 P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize, 2618 8), B_FALSE, NULL); 2619 break; 2620 case DRR_WRITE_BYREF: 2621 case DRR_FREEOBJECTS: 2622 case DRR_FREE: 2623 break; 2624 2625 default: 2626 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2627 "invalid record type")); 2628 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 2629 } 2630 } 2631 2632 free(buf); 2633 return (-1); 2634} 2635 2636/* 2637 * Restores a backup of tosnap from the file descriptor specified by infd. 2638 */ 2639static int 2640zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, 2641 const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr, 2642 dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv, 2643 avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd, 2644 uint64_t *action_handlep) 2645{ 2646 zfs_cmd_t zc = { 0 }; 2647 time_t begin_time; 2648 int ioctl_err, ioctl_errno, err; 2649 char *cp; 2650 struct drr_begin *drrb = &drr->drr_u.drr_begin; 2651 char errbuf[1024]; 2652 char prop_errbuf[1024]; 2653 const char *chopprefix; 2654 boolean_t newfs = B_FALSE; 2655 boolean_t stream_wantsnewfs; 2656 uint64_t parent_snapguid = 0; 2657 prop_changelist_t *clp = NULL; 2658 nvlist_t *snapprops_nvlist = NULL; 2659 zprop_errflags_t prop_errflags; 2660 boolean_t recursive; 2661 2662 begin_time = time(NULL); 2663 2664 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2665 "cannot receive")); 2666 2667 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == 2668 ENOENT); 2669 2670 if (stream_avl != NULL) { 2671 char *snapname; 2672 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid, 2673 &snapname); 2674 nvlist_t *props; 2675 int ret; 2676 2677 (void) nvlist_lookup_uint64(fs, "parentfromsnap", 2678 &parent_snapguid); 2679 err = nvlist_lookup_nvlist(fs, "props", &props); 2680 if (err) 2681 VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0)); 2682 2683 if (flags->canmountoff) { 2684 VERIFY(0 == nvlist_add_uint64(props, 2685 zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0)); 2686 } 2687 ret = zcmd_write_src_nvlist(hdl, &zc, props); 2688 if (err) 2689 nvlist_free(props); 2690 2691 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) { 2692 VERIFY(0 == nvlist_lookup_nvlist(props, 2693 snapname, &snapprops_nvlist)); 2694 } 2695 2696 if (ret != 0) 2697 return (-1); 2698 } 2699 2700 cp = NULL; 2701 2702 /* 2703 * Determine how much of the snapshot name stored in the stream 2704 * we are going to tack on to the name they specified on the 2705 * command line, and how much we are going to chop off. 2706 * 2707 * If they specified a snapshot, chop the entire name stored in 2708 * the stream. 2709 */ 2710 if (flags->istail) { 2711 /* 2712 * A filesystem was specified with -e. We want to tack on only 2713 * the tail of the sent snapshot path. 2714 */ 2715 if (strchr(tosnap, '@')) { 2716 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 2717 "argument - snapshot not allowed with -e")); 2718 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 2719 } 2720 2721 chopprefix = strrchr(sendfs, '/'); 2722 2723 if (chopprefix == NULL) { 2724 /* 2725 * The tail is the poolname, so we need to 2726 * prepend a path separator. 2727 */ 2728 int len = strlen(drrb->drr_toname); 2729 cp = malloc(len + 2); 2730 cp[0] = '/'; 2731 (void) strcpy(&cp[1], drrb->drr_toname); 2732 chopprefix = cp; 2733 } else { 2734 chopprefix = drrb->drr_toname + (chopprefix - sendfs); 2735 } 2736 } else if (flags->isprefix) { 2737 /* 2738 * A filesystem was specified with -d. We want to tack on 2739 * everything but the first element of the sent snapshot path 2740 * (all but the pool name). 2741 */ 2742 if (strchr(tosnap, '@')) { 2743 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 2744 "argument - snapshot not allowed with -d")); 2745 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 2746 } 2747 2748 chopprefix = strchr(drrb->drr_toname, '/'); 2749 if (chopprefix == NULL) 2750 chopprefix = strchr(drrb->drr_toname, '@'); 2751 } else if (strchr(tosnap, '@') == NULL) { 2752 /* 2753 * If a filesystem was specified without -d or -e, we want to 2754 * tack on everything after the fs specified by 'zfs send'. 2755 */ 2756 chopprefix = drrb->drr_toname + strlen(sendfs); 2757 } else { 2758 /* A snapshot was specified as an exact path (no -d or -e). */ 2759 if (recursive) { 2760 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2761 "cannot specify snapshot name for multi-snapshot " 2762 "stream")); 2763 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 2764 } 2765 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname); 2766 } 2767 2768 ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname); 2769 ASSERT(chopprefix > drrb->drr_toname); 2770 ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname)); 2771 ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' || 2772 chopprefix[0] == '\0'); 2773 2774 /* 2775 * Determine name of destination snapshot, store in zc_value. 2776 */ 2777 (void) strcpy(zc.zc_value, tosnap); 2778 (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value)); 2779#ifdef __FreeBSD__ 2780 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF) 2781 zfs_ioctl_version = get_zfs_ioctl_version(); 2782 /* 2783 * For forward compatibility hide tosnap in zc_value 2784 */ 2785 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 2786 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap); 2787#endif 2788 free(cp); 2789 if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) { 2790 zcmd_free_nvlists(&zc); 2791 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); 2792 } 2793 2794 /* 2795 * Determine the name of the origin snapshot, store in zc_string. 2796 */ 2797 if (drrb->drr_flags & DRR_FLAG_CLONE) { 2798 if (guid_to_name(hdl, zc.zc_value, 2799 drrb->drr_fromguid, zc.zc_string) != 0) { 2800 zcmd_free_nvlists(&zc); 2801 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2802 "local origin for clone %s does not exist"), 2803 zc.zc_value); 2804 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 2805 } 2806 if (flags->verbose) 2807 (void) printf("found clone origin %s\n", zc.zc_string); 2808 } else if (originsnap) { 2809 (void) strncpy(zc.zc_string, originsnap, ZFS_MAXNAMELEN); 2810 if (flags->verbose) 2811 (void) printf("using provided clone origin %s\n", 2812 zc.zc_string); 2813 } 2814 2815 stream_wantsnewfs = (drrb->drr_fromguid == 0 || 2816 (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap); 2817 2818 if (stream_wantsnewfs) { 2819 /* 2820 * if the parent fs does not exist, look for it based on 2821 * the parent snap GUID 2822 */ 2823 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2824 "cannot receive new filesystem stream")); 2825 2826 (void) strcpy(zc.zc_name, zc.zc_value); 2827 cp = strrchr(zc.zc_name, '/'); 2828 if (cp) 2829 *cp = '\0'; 2830 if (cp && 2831 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 2832 char suffix[ZFS_MAXNAMELEN]; 2833 (void) strcpy(suffix, strrchr(zc.zc_value, '/')); 2834 if (guid_to_name(hdl, zc.zc_name, parent_snapguid, 2835 zc.zc_value) == 0) { 2836 *strchr(zc.zc_value, '@') = '\0'; 2837 (void) strcat(zc.zc_value, suffix); 2838 } 2839 } 2840 } else { 2841 /* 2842 * if the fs does not exist, look for it based on the 2843 * fromsnap GUID 2844 */ 2845 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 2846 "cannot receive incremental stream")); 2847 2848 (void) strcpy(zc.zc_name, zc.zc_value); 2849 *strchr(zc.zc_name, '@') = '\0'; 2850 2851 /* 2852 * If the exact receive path was specified and this is the 2853 * topmost path in the stream, then if the fs does not exist we 2854 * should look no further. 2855 */ 2856 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname + 2857 strlen(sendfs)) != '\0' && *chopprefix != '@')) && 2858 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 2859 char snap[ZFS_MAXNAMELEN]; 2860 (void) strcpy(snap, strchr(zc.zc_value, '@')); 2861 if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid, 2862 zc.zc_value) == 0) { 2863 *strchr(zc.zc_value, '@') = '\0'; 2864 (void) strcat(zc.zc_value, snap); 2865 } 2866 } 2867 } 2868 2869 (void) strcpy(zc.zc_name, zc.zc_value); 2870 *strchr(zc.zc_name, '@') = '\0'; 2871 2872 if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { 2873 zfs_handle_t *zhp; 2874 2875 /* 2876 * Destination fs exists. Therefore this should either 2877 * be an incremental, or the stream specifies a new fs 2878 * (full stream or clone) and they want us to blow it 2879 * away (and have therefore specified -F and removed any 2880 * snapshots). 2881 */ 2882 if (stream_wantsnewfs) { 2883 if (!flags->force) { 2884 zcmd_free_nvlists(&zc); 2885 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2886 "destination '%s' exists\n" 2887 "must specify -F to overwrite it"), 2888 zc.zc_name); 2889 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 2890 } 2891 if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT, 2892 &zc) == 0) { 2893 zcmd_free_nvlists(&zc); 2894 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2895 "destination has snapshots (eg. %s)\n" 2896 "must destroy them to overwrite it"), 2897 zc.zc_name); 2898 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 2899 } 2900 } 2901 2902 if ((zhp = zfs_open(hdl, zc.zc_name, 2903 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) { 2904 zcmd_free_nvlists(&zc); 2905 return (-1); 2906 } 2907 2908 if (stream_wantsnewfs && 2909 zhp->zfs_dmustats.dds_origin[0]) { 2910 zcmd_free_nvlists(&zc); 2911 zfs_close(zhp); 2912 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2913 "destination '%s' is a clone\n" 2914 "must destroy it to overwrite it"), 2915 zc.zc_name); 2916 return (zfs_error(hdl, EZFS_EXISTS, errbuf)); 2917 } 2918 2919 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM && 2920 stream_wantsnewfs) { 2921 /* We can't do online recv in this case */ 2922 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0); 2923 if (clp == NULL) { 2924 zfs_close(zhp); 2925 zcmd_free_nvlists(&zc); 2926 return (-1); 2927 } 2928 if (changelist_prefix(clp) != 0) { 2929 changelist_free(clp); 2930 zfs_close(zhp); 2931 zcmd_free_nvlists(&zc); 2932 return (-1); 2933 } 2934 } 2935 zfs_close(zhp); 2936 } else { 2937 /* 2938 * Destination filesystem does not exist. Therefore we better 2939 * be creating a new filesystem (either from a full backup, or 2940 * a clone). It would therefore be invalid if the user 2941 * specified only the pool name (i.e. if the destination name 2942 * contained no slash character). 2943 */ 2944 if (!stream_wantsnewfs || 2945 (cp = strrchr(zc.zc_name, '/')) == NULL) { 2946 zcmd_free_nvlists(&zc); 2947 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 2948 "destination '%s' does not exist"), zc.zc_name); 2949 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 2950 } 2951 2952 /* 2953 * Trim off the final dataset component so we perform the 2954 * recvbackup ioctl to the filesystems's parent. 2955 */ 2956 *cp = '\0'; 2957 2958 if (flags->isprefix && !flags->istail && !flags->dryrun && 2959 create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) { 2960 zcmd_free_nvlists(&zc); 2961 return (zfs_error(hdl, EZFS_BADRESTORE, errbuf)); 2962 } 2963 2964 newfs = B_TRUE; 2965 } 2966 2967 zc.zc_begin_record = drr_noswap->drr_u.drr_begin; 2968 zc.zc_cookie = infd; 2969 zc.zc_guid = flags->force; 2970 if (flags->verbose) { 2971 (void) printf("%s %s stream of %s into %s\n", 2972 flags->dryrun ? "would receive" : "receiving", 2973 drrb->drr_fromguid ? "incremental" : "full", 2974 drrb->drr_toname, zc.zc_value); 2975 (void) fflush(stdout); 2976 } 2977 2978 if (flags->dryrun) { 2979 zcmd_free_nvlists(&zc); 2980 return (recv_skip(hdl, infd, flags->byteswap)); 2981 } 2982 2983 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf; 2984 zc.zc_nvlist_dst_size = sizeof (prop_errbuf); 2985 zc.zc_cleanup_fd = cleanup_fd; 2986 zc.zc_action_handle = *action_handlep; 2987 2988 err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc); 2989 ioctl_errno = errno; 2990 prop_errflags = (zprop_errflags_t)zc.zc_obj; 2991 2992 if (err == 0) { 2993 nvlist_t *prop_errors; 2994 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, 2995 zc.zc_nvlist_dst_size, &prop_errors, 0)); 2996 2997 nvpair_t *prop_err = NULL; 2998 2999 while ((prop_err = nvlist_next_nvpair(prop_errors, 3000 prop_err)) != NULL) { 3001 char tbuf[1024]; 3002 zfs_prop_t prop; 3003 int intval; 3004 3005 prop = zfs_name_to_prop(nvpair_name(prop_err)); 3006 (void) nvpair_value_int32(prop_err, &intval); 3007 if (strcmp(nvpair_name(prop_err), 3008 ZPROP_N_MORE_ERRORS) == 0) { 3009 trunc_prop_errs(intval); 3010 break; 3011 } else { 3012 (void) snprintf(tbuf, sizeof (tbuf), 3013 dgettext(TEXT_DOMAIN, 3014 "cannot receive %s property on %s"), 3015 nvpair_name(prop_err), zc.zc_name); 3016 zfs_setprop_error(hdl, prop, intval, tbuf); 3017 } 3018 } 3019 nvlist_free(prop_errors); 3020 } 3021 3022 zc.zc_nvlist_dst = 0; 3023 zc.zc_nvlist_dst_size = 0; 3024 zcmd_free_nvlists(&zc); 3025 3026 if (err == 0 && snapprops_nvlist) { 3027 zfs_cmd_t zc2 = { 0 }; 3028 3029 (void) strcpy(zc2.zc_name, zc.zc_value); 3030 zc2.zc_cookie = B_TRUE; /* received */ 3031 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) { 3032 (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2); 3033 zcmd_free_nvlists(&zc2); 3034 } 3035 } 3036 3037 if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) { 3038 /* 3039 * It may be that this snapshot already exists, 3040 * in which case we want to consume & ignore it 3041 * rather than failing. 3042 */ 3043 avl_tree_t *local_avl; 3044 nvlist_t *local_nv, *fs; 3045 cp = strchr(zc.zc_value, '@'); 3046 3047 /* 3048 * XXX Do this faster by just iterating over snaps in 3049 * this fs. Also if zc_value does not exist, we will 3050 * get a strange "does not exist" error message. 3051 */ 3052 *cp = '\0'; 3053 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE, 3054 &local_nv, &local_avl) == 0) { 3055 *cp = '@'; 3056 fs = fsavl_find(local_avl, drrb->drr_toguid, NULL); 3057 fsavl_destroy(local_avl); 3058 nvlist_free(local_nv); 3059 3060 if (fs != NULL) { 3061 if (flags->verbose) { 3062 (void) printf("snap %s already exists; " 3063 "ignoring\n", zc.zc_value); 3064 } 3065 err = ioctl_err = recv_skip(hdl, infd, 3066 flags->byteswap); 3067 } 3068 } 3069 *cp = '@'; 3070 } 3071 3072 if (ioctl_err != 0) { 3073 switch (ioctl_errno) { 3074 case ENODEV: 3075 cp = strchr(zc.zc_value, '@'); 3076 *cp = '\0'; 3077 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3078 "most recent snapshot of %s does not\n" 3079 "match incremental source"), zc.zc_value); 3080 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); 3081 *cp = '@'; 3082 break; 3083 case ETXTBSY: 3084 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3085 "destination %s has been modified\n" 3086 "since most recent snapshot"), zc.zc_name); 3087 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); 3088 break; 3089 case EEXIST: 3090 cp = strchr(zc.zc_value, '@'); 3091 if (newfs) { 3092 /* it's the containing fs that exists */ 3093 *cp = '\0'; 3094 } 3095 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3096 "destination already exists")); 3097 (void) zfs_error_fmt(hdl, EZFS_EXISTS, 3098 dgettext(TEXT_DOMAIN, "cannot restore to %s"), 3099 zc.zc_value); 3100 *cp = '@'; 3101 break; 3102 case EINVAL: 3103 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); 3104 break; 3105 case ECKSUM: 3106 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3107 "invalid stream (checksum mismatch)")); 3108 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); 3109 break; 3110 case ENOTSUP: 3111 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3112 "pool must be upgraded to receive this stream.")); 3113 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf); 3114 break; 3115 case EDQUOT: 3116 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3117 "destination %s space quota exceeded"), zc.zc_name); 3118 (void) zfs_error(hdl, EZFS_NOSPC, errbuf); 3119 break; 3120 default: 3121 (void) zfs_standard_error(hdl, ioctl_errno, errbuf); 3122 } 3123 } 3124 3125 /* 3126 * Mount the target filesystem (if created). Also mount any 3127 * children of the target filesystem if we did a replication 3128 * receive (indicated by stream_avl being non-NULL). 3129 */ 3130 cp = strchr(zc.zc_value, '@'); 3131 if (cp && (ioctl_err == 0 || !newfs)) { 3132 zfs_handle_t *h; 3133 3134 *cp = '\0'; 3135 h = zfs_open(hdl, zc.zc_value, 3136 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); 3137 if (h != NULL) { 3138 if (h->zfs_type == ZFS_TYPE_VOLUME) { 3139 *cp = '@'; 3140 } else if (newfs || stream_avl) { 3141 /* 3142 * Track the first/top of hierarchy fs, 3143 * for mounting and sharing later. 3144 */ 3145 if (top_zfs && *top_zfs == NULL) 3146 *top_zfs = zfs_strdup(hdl, zc.zc_value); 3147 } 3148 zfs_close(h); 3149 } 3150 *cp = '@'; 3151 } 3152 3153 if (clp) { 3154 err |= changelist_postfix(clp); 3155 changelist_free(clp); 3156 } 3157 3158 if (prop_errflags & ZPROP_ERR_NOCLEAR) { 3159 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " 3160 "failed to clear unreceived properties on %s"), 3161 zc.zc_name); 3162 (void) fprintf(stderr, "\n"); 3163 } 3164 if (prop_errflags & ZPROP_ERR_NORESTORE) { 3165 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " 3166 "failed to restore original properties on %s"), 3167 zc.zc_name); 3168 (void) fprintf(stderr, "\n"); 3169 } 3170 3171 if (err || ioctl_err) 3172 return (-1); 3173 3174 *action_handlep = zc.zc_action_handle; 3175 3176 if (flags->verbose) { 3177 char buf1[64]; 3178 char buf2[64]; 3179 uint64_t bytes = zc.zc_cookie; 3180 time_t delta = time(NULL) - begin_time; 3181 if (delta == 0) 3182 delta = 1; 3183 zfs_nicenum(bytes, buf1, sizeof (buf1)); 3184 zfs_nicenum(bytes/delta, buf2, sizeof (buf1)); 3185 3186 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n", 3187 buf1, delta, buf2); 3188 } 3189 3190 return (0); 3191} 3192 3193static int 3194zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, 3195 const char *originsnap, recvflags_t *flags, int infd, const char *sendfs, 3196 nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd, 3197 uint64_t *action_handlep) 3198{ 3199 int err; 3200 dmu_replay_record_t drr, drr_noswap; 3201 struct drr_begin *drrb = &drr.drr_u.drr_begin; 3202 char errbuf[1024]; 3203 zio_cksum_t zcksum = { 0 }; 3204 uint64_t featureflags; 3205 int hdrtype; 3206 3207 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, 3208 "cannot receive")); 3209 3210 if (flags->isprefix && 3211 !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) { 3212 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs " 3213 "(%s) does not exist"), tosnap); 3214 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3215 } 3216 if (originsnap && 3217 !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) { 3218 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs " 3219 "(%s) does not exist"), originsnap); 3220 return (zfs_error(hdl, EZFS_NOENT, errbuf)); 3221 } 3222 3223 /* read in the BEGIN record */ 3224 if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE, 3225 &zcksum))) 3226 return (err); 3227 3228 if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) { 3229 /* It's the double end record at the end of a package */ 3230 return (ENODATA); 3231 } 3232 3233 /* the kernel needs the non-byteswapped begin record */ 3234 drr_noswap = drr; 3235 3236 flags->byteswap = B_FALSE; 3237 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { 3238 /* 3239 * We computed the checksum in the wrong byteorder in 3240 * recv_read() above; do it again correctly. 3241 */ 3242 bzero(&zcksum, sizeof (zio_cksum_t)); 3243 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum); 3244 flags->byteswap = B_TRUE; 3245 3246 drr.drr_type = BSWAP_32(drr.drr_type); 3247 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen); 3248 drrb->drr_magic = BSWAP_64(drrb->drr_magic); 3249 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); 3250 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); 3251 drrb->drr_type = BSWAP_32(drrb->drr_type); 3252 drrb->drr_flags = BSWAP_32(drrb->drr_flags); 3253 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); 3254 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); 3255 } 3256 3257 if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) { 3258 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3259 "stream (bad magic number)")); 3260 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3261 } 3262 3263 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 3264 hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo); 3265 3266 if (!DMU_STREAM_SUPPORTED(featureflags) || 3267 (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) { 3268 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 3269 "stream has unsupported feature, feature flags = %lx"), 3270 featureflags); 3271 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3272 } 3273 3274 if (strchr(drrb->drr_toname, '@') == NULL) { 3275 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " 3276 "stream (bad snapshot name)")); 3277 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); 3278 } 3279 3280 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) { 3281 char nonpackage_sendfs[ZFS_MAXNAMELEN]; 3282 if (sendfs == NULL) { 3283 /* 3284 * We were not called from zfs_receive_package(). Get 3285 * the fs specified by 'zfs send'. 3286 */ 3287 char *cp; 3288 (void) strlcpy(nonpackage_sendfs, 3289 drr.drr_u.drr_begin.drr_toname, ZFS_MAXNAMELEN); 3290 if ((cp = strchr(nonpackage_sendfs, '@')) != NULL) 3291 *cp = '\0'; 3292 sendfs = nonpackage_sendfs; 3293 } 3294 return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags, 3295 &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs, 3296 cleanup_fd, action_handlep)); 3297 } else { 3298 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 3299 DMU_COMPOUNDSTREAM); 3300 return (zfs_receive_package(hdl, infd, tosnap, flags, &drr, 3301 &zcksum, top_zfs, cleanup_fd, action_handlep)); 3302 } 3303} 3304 3305/* 3306 * Restores a backup of tosnap from the file descriptor specified by infd. 3307 * Return 0 on total success, -2 if some things couldn't be 3308 * destroyed/renamed/promoted, -1 if some things couldn't be received. 3309 * (-1 will override -2). 3310 */ 3311int 3312zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props, 3313 recvflags_t *flags, int infd, avl_tree_t *stream_avl) 3314{ 3315 char *top_zfs = NULL; 3316 int err; 3317 int cleanup_fd; 3318 uint64_t action_handle = 0; 3319 char *originsnap = NULL; 3320 if (props) { 3321 err = nvlist_lookup_string(props, "origin", &originsnap); 3322 if (err && err != ENOENT) 3323 return (err); 3324 } 3325 3326 cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); 3327 VERIFY(cleanup_fd >= 0); 3328 3329 err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL, 3330 stream_avl, &top_zfs, cleanup_fd, &action_handle); 3331 3332 VERIFY(0 == close(cleanup_fd)); 3333 3334 if (err == 0 && !flags->nomount && top_zfs) { 3335 zfs_handle_t *zhp; 3336 prop_changelist_t *clp; 3337 3338 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM); 3339 if (zhp != NULL) { 3340 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT, 3341 CL_GATHER_MOUNT_ALWAYS, 0); 3342 zfs_close(zhp); 3343 if (clp != NULL) { 3344 /* mount and share received datasets */ 3345 err = changelist_postfix(clp); 3346 changelist_free(clp); 3347 } 3348 } 3349 if (zhp == NULL || clp == NULL || err) 3350 err = -1; 3351 } 3352 if (top_zfs) 3353 free(top_zfs); 3354 3355 return (err); 3356} 3357