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 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/dmu.h> 29 #include <sys/dmu_impl.h> 30 #include <sys/dmu_tx.h> 31 #include <sys/dbuf.h> 32 #include <sys/dnode.h> 33 #include <sys/zfs_context.h> 34 #include <sys/dmu_objset.h> 35 #include <sys/dmu_traverse.h> 36 #include <sys/dsl_dataset.h> 37 #include <sys/dsl_dir.h> 38 #include <sys/dsl_pool.h> 39 #include <sys/dsl_synctask.h> 40 #include <sys/zfs_ioctl.h> 41 #include <sys/zap.h> 42 #include <sys/zio_checksum.h> 43 44 static char *dmu_recv_tag = "dmu_recv_tag"; 45 46 struct backuparg { 47 dmu_replay_record_t *drr; 48 vnode_t *vp; 49 offset_t *off; 50 objset_t *os; 51 zio_cksum_t zc; 52 int err; 53 }; 54 55 static int 56 dump_bytes(struct backuparg *ba, void *buf, int len) 57 { 58 ssize_t resid; /* have to get resid to get detailed errno */ 59 ASSERT3U(len % 8, ==, 0); 60 61 fletcher_4_incremental_native(buf, len, &ba->zc); 62 ba->err = vn_rdwr(UIO_WRITE, ba->vp, 63 (caddr_t)buf, len, 64 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid); 65 *ba->off += len; 66 return (ba->err); 67 } 68 69 static int 70 dump_free(struct backuparg *ba, uint64_t object, uint64_t offset, 71 uint64_t length) 72 { 73 /* write a FREE record */ 74 bzero(ba->drr, sizeof (dmu_replay_record_t)); 75 ba->drr->drr_type = DRR_FREE; 76 ba->drr->drr_u.drr_free.drr_object = object; 77 ba->drr->drr_u.drr_free.drr_offset = offset; 78 ba->drr->drr_u.drr_free.drr_length = length; 79 80 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t))) 81 return (EINTR); 82 return (0); 83 } 84 85 static int 86 dump_data(struct backuparg *ba, dmu_object_type_t type, 87 uint64_t object, uint64_t offset, int blksz, void *data) 88 { 89 /* write a DATA record */ 90 bzero(ba->drr, sizeof (dmu_replay_record_t)); 91 ba->drr->drr_type = DRR_WRITE; 92 ba->drr->drr_u.drr_write.drr_object = object; 93 ba->drr->drr_u.drr_write.drr_type = type; 94 ba->drr->drr_u.drr_write.drr_offset = offset; 95 ba->drr->drr_u.drr_write.drr_length = blksz; 96 97 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t))) 98 return (EINTR); 99 if (dump_bytes(ba, data, blksz)) 100 return (EINTR); 101 return (0); 102 } 103 104 static int 105 dump_freeobjects(struct backuparg *ba, uint64_t firstobj, uint64_t numobjs) 106 { 107 /* write a FREEOBJECTS record */ 108 bzero(ba->drr, sizeof (dmu_replay_record_t)); 109 ba->drr->drr_type = DRR_FREEOBJECTS; 110 ba->drr->drr_u.drr_freeobjects.drr_firstobj = firstobj; 111 ba->drr->drr_u.drr_freeobjects.drr_numobjs = numobjs; 112 113 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t))) 114 return (EINTR); 115 return (0); 116 } 117 118 static int 119 dump_dnode(struct backuparg *ba, uint64_t object, dnode_phys_t *dnp) 120 { 121 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) 122 return (dump_freeobjects(ba, object, 1)); 123 124 /* write an OBJECT record */ 125 bzero(ba->drr, sizeof (dmu_replay_record_t)); 126 ba->drr->drr_type = DRR_OBJECT; 127 ba->drr->drr_u.drr_object.drr_object = object; 128 ba->drr->drr_u.drr_object.drr_type = dnp->dn_type; 129 ba->drr->drr_u.drr_object.drr_bonustype = dnp->dn_bonustype; 130 ba->drr->drr_u.drr_object.drr_blksz = 131 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 132 ba->drr->drr_u.drr_object.drr_bonuslen = dnp->dn_bonuslen; 133 ba->drr->drr_u.drr_object.drr_checksum = dnp->dn_checksum; 134 ba->drr->drr_u.drr_object.drr_compress = dnp->dn_compress; 135 136 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t))) 137 return (EINTR); 138 139 if (dump_bytes(ba, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8))) 140 return (EINTR); 141 142 /* free anything past the end of the file */ 143 if (dump_free(ba, object, (dnp->dn_maxblkid + 1) * 144 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL)) 145 return (EINTR); 146 if (ba->err) 147 return (EINTR); 148 return (0); 149 } 150 151 #define BP_SPAN(dnp, level) \ 152 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \ 153 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) 154 155 static int 156 backup_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg) 157 { 158 struct backuparg *ba = arg; 159 uint64_t object = bc->bc_bookmark.zb_object; 160 int level = bc->bc_bookmark.zb_level; 161 uint64_t blkid = bc->bc_bookmark.zb_blkid; 162 blkptr_t *bp = bc->bc_blkptr.blk_birth ? &bc->bc_blkptr : NULL; 163 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; 164 void *data = bc->bc_data; 165 int err = 0; 166 167 if (issig(JUSTLOOKING) && issig(FORREAL)) 168 return (EINTR); 169 170 ASSERT(data || bp == NULL); 171 172 if (bp == NULL && object == 0) { 173 uint64_t span = BP_SPAN(bc->bc_dnode, level); 174 uint64_t dnobj = (blkid * span) >> DNODE_SHIFT; 175 err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT); 176 } else if (bp == NULL) { 177 uint64_t span = BP_SPAN(bc->bc_dnode, level); 178 err = dump_free(ba, object, blkid * span, span); 179 } else if (data && level == 0 && type == DMU_OT_DNODE) { 180 dnode_phys_t *blk = data; 181 int i; 182 int blksz = BP_GET_LSIZE(bp); 183 184 for (i = 0; i < blksz >> DNODE_SHIFT; i++) { 185 uint64_t dnobj = 186 (blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i; 187 err = dump_dnode(ba, dnobj, blk+i); 188 if (err) 189 break; 190 } 191 } else if (level == 0 && 192 type != DMU_OT_DNODE && type != DMU_OT_OBJSET) { 193 int blksz = BP_GET_LSIZE(bp); 194 if (data == NULL) { 195 uint32_t aflags = ARC_WAIT; 196 arc_buf_t *abuf; 197 zbookmark_t zb; 198 199 zb.zb_objset = ba->os->os->os_dsl_dataset->ds_object; 200 zb.zb_object = object; 201 zb.zb_level = level; 202 zb.zb_blkid = blkid; 203 (void) arc_read(NULL, spa, bp, 204 dmu_ot[type].ot_byteswap, arc_getbuf_func, &abuf, 205 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_MUSTSUCCEED, 206 &aflags, &zb); 207 208 if (abuf) { 209 err = dump_data(ba, type, object, blkid * blksz, 210 blksz, abuf->b_data); 211 (void) arc_buf_remove_ref(abuf, &abuf); 212 } 213 } else { 214 err = dump_data(ba, type, object, blkid * blksz, 215 blksz, data); 216 } 217 } 218 219 ASSERT(err == 0 || err == EINTR); 220 return (err); 221 } 222 223 int 224 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin, 225 vnode_t *vp, offset_t *off) 226 { 227 dsl_dataset_t *ds = tosnap->os->os_dsl_dataset; 228 dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL; 229 dmu_replay_record_t *drr; 230 struct backuparg ba; 231 int err; 232 uint64_t fromtxg = 0; 233 234 /* tosnap must be a snapshot */ 235 if (ds->ds_phys->ds_next_snap_obj == 0) 236 return (EINVAL); 237 238 /* fromsnap must be an earlier snapshot from the same fs as tosnap */ 239 if (fromds && (ds->ds_dir != fromds->ds_dir || 240 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg)) 241 return (EXDEV); 242 243 if (fromorigin) { 244 if (fromsnap) 245 return (EINVAL); 246 247 if (ds->ds_dir->dd_phys->dd_origin_obj != NULL) { 248 dsl_pool_t *dp = ds->ds_dir->dd_pool; 249 rw_enter(&dp->dp_config_rwlock, RW_READER); 250 err = dsl_dataset_hold_obj(dp, 251 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds); 252 rw_exit(&dp->dp_config_rwlock); 253 if (err) 254 return (err); 255 } else { 256 fromorigin = B_FALSE; 257 } 258 } 259 260 261 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 262 drr->drr_type = DRR_BEGIN; 263 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 264 drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION; 265 drr->drr_u.drr_begin.drr_creation_time = 266 ds->ds_phys->ds_creation_time; 267 drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type; 268 if (fromorigin) 269 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; 270 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid; 271 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 272 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; 273 274 if (fromds) 275 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid; 276 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname); 277 278 if (fromds) 279 fromtxg = fromds->ds_phys->ds_creation_txg; 280 if (fromorigin) 281 dsl_dataset_rele(fromds, FTAG); 282 283 ba.drr = drr; 284 ba.vp = vp; 285 ba.os = tosnap; 286 ba.off = off; 287 ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0); 288 289 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) { 290 kmem_free(drr, sizeof (dmu_replay_record_t)); 291 return (ba.err); 292 } 293 294 err = traverse_dsl_dataset(ds, fromtxg, 295 ADVANCE_PRE | ADVANCE_HOLES | ADVANCE_DATA | ADVANCE_NOLOCK, 296 backup_cb, &ba); 297 298 if (err) { 299 if (err == EINTR && ba.err) 300 err = ba.err; 301 kmem_free(drr, sizeof (dmu_replay_record_t)); 302 return (err); 303 } 304 305 bzero(drr, sizeof (dmu_replay_record_t)); 306 drr->drr_type = DRR_END; 307 drr->drr_u.drr_end.drr_checksum = ba.zc; 308 309 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) { 310 kmem_free(drr, sizeof (dmu_replay_record_t)); 311 return (ba.err); 312 } 313 314 kmem_free(drr, sizeof (dmu_replay_record_t)); 315 316 return (0); 317 } 318 319 struct recvbeginsyncarg { 320 const char *tofs; 321 const char *tosnap; 322 dsl_dataset_t *origin; 323 uint64_t fromguid; 324 dmu_objset_type_t type; 325 void *tag; 326 boolean_t force; 327 uint64_t dsflags; 328 char clonelastname[MAXNAMELEN]; 329 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */ 330 }; 331 332 static dsl_dataset_t * 333 recv_full_sync_impl(dsl_pool_t *dp, uint64_t dsobj, dmu_objset_type_t type, 334 cred_t *cr, dmu_tx_t *tx) 335 { 336 dsl_dataset_t *ds; 337 338 /* This should always work, since we just created it */ 339 /* XXX - create should return an owned ds */ 340 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, 341 DS_MODE_INCONSISTENT, dmu_recv_tag, &ds)); 342 343 if (type != DMU_OST_NONE) { 344 (void) dmu_objset_create_impl(dp->dp_spa, 345 ds, &ds->ds_phys->ds_bp, type, tx); 346 } 347 348 spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC, 349 dp->dp_spa, tx, cr, "dataset = %lld", dsobj); 350 351 return (ds); 352 } 353 354 /* ARGSUSED */ 355 static int 356 recv_full_check(void *arg1, void *arg2, dmu_tx_t *tx) 357 { 358 dsl_dir_t *dd = arg1; 359 struct recvbeginsyncarg *rbsa = arg2; 360 objset_t *mos = dd->dd_pool->dp_meta_objset; 361 uint64_t val; 362 int err; 363 364 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj, 365 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val); 366 367 if (err != ENOENT) 368 return (err ? err : EEXIST); 369 370 if (rbsa->origin) { 371 /* make sure it's a snap in the same pool */ 372 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool) 373 return (EXDEV); 374 if (rbsa->origin->ds_phys->ds_num_children == 0) 375 return (EINVAL); 376 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid) 377 return (ENODEV); 378 } 379 380 return (0); 381 } 382 383 static void 384 recv_full_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 385 { 386 dsl_dir_t *dd = arg1; 387 struct recvbeginsyncarg *rbsa = arg2; 388 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags; 389 uint64_t dsobj; 390 391 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1, 392 rbsa->origin, flags, cr, tx); 393 394 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj, 395 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx); 396 } 397 398 static int 399 recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx) 400 { 401 dsl_dataset_t *ds = arg1; 402 struct recvbeginsyncarg *rbsa = arg2; 403 int err; 404 405 /* must be a head ds */ 406 if (ds->ds_phys->ds_next_snap_obj != 0) 407 return (EINVAL); 408 409 /* must not be a clone ds */ 410 if (ds->ds_prev != NULL) 411 return (EINVAL); 412 413 err = dsl_dataset_destroy_check(ds, rbsa->tag, tx); 414 if (err) 415 return (err); 416 417 if (rbsa->origin) { 418 /* make sure it's a snap in the same pool */ 419 if (rbsa->origin->ds_dir->dd_pool != ds->ds_dir->dd_pool) 420 return (EXDEV); 421 if (rbsa->origin->ds_phys->ds_num_children == 0) 422 return (EINVAL); 423 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid) 424 return (ENODEV); 425 } 426 427 return (0); 428 } 429 430 static void 431 recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 432 { 433 dsl_dataset_t *ds = arg1; 434 struct recvbeginsyncarg *rbsa = arg2; 435 dsl_dir_t *dd = ds->ds_dir; 436 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags; 437 uint64_t dsobj; 438 439 /* 440 * NB: caller must provide an extra hold on the dsl_dir_t, so it 441 * won't go away when dsl_dataset_destroy_sync() closes the 442 * dataset. 443 */ 444 dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx); 445 446 dsobj = dsl_dataset_create_sync_impl(dd, rbsa->origin, flags, tx); 447 448 rbsa->ds = recv_full_sync_impl(dd->dd_pool, dsobj, 449 rbsa->origin ? DMU_OST_NONE : rbsa->type, cr, tx); 450 } 451 452 /* ARGSUSED */ 453 static int 454 recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx) 455 { 456 dsl_dataset_t *ds = arg1; 457 struct recvbeginsyncarg *rbsa = arg2; 458 int err; 459 uint64_t val; 460 461 /* must not have any changes since most recent snapshot */ 462 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds)) 463 return (ETXTBSY); 464 465 /* must already be a snapshot of this fs */ 466 if (ds->ds_phys->ds_prev_snap_obj == 0) 467 return (ENODEV); 468 469 /* most recent snapshot must match fromguid */ 470 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) 471 return (ENODEV); 472 473 /* temporary clone name must not exist */ 474 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, 475 ds->ds_dir->dd_phys->dd_child_dir_zapobj, 476 rbsa->clonelastname, 8, 1, &val); 477 if (err == 0) 478 return (EEXIST); 479 if (err != ENOENT) 480 return (err); 481 482 /* new snapshot name must not exist */ 483 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, 484 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val); 485 if (err == 0) 486 return (EEXIST); 487 if (err != ENOENT) 488 return (err); 489 return (0); 490 } 491 492 /* ARGSUSED */ 493 static void 494 recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 495 { 496 dsl_dataset_t *ohds = arg1; 497 struct recvbeginsyncarg *rbsa = arg2; 498 dsl_pool_t *dp = ohds->ds_dir->dd_pool; 499 dsl_dataset_t *ods, *cds; 500 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags; 501 uint64_t dsobj; 502 503 /* create the temporary clone */ 504 VERIFY(0 == dsl_dataset_hold_obj(dp, ohds->ds_phys->ds_prev_snap_obj, 505 FTAG, &ods)); 506 dsobj = dsl_dataset_create_sync(ohds->ds_dir, 507 rbsa->clonelastname, ods, flags, cr, tx); 508 dsl_dataset_rele(ods, FTAG); 509 510 /* open the temporary clone */ 511 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, 512 DS_MODE_INCONSISTENT, dmu_recv_tag, &cds)); 513 514 /* copy the refquota from the target fs to the clone */ 515 if (ohds->ds_quota > 0) 516 dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx); 517 518 rbsa->ds = cds; 519 520 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC, 521 dp->dp_spa, tx, cr, "dataset = %lld", dsobj); 522 } 523 524 /* ARGSUSED */ 525 static void 526 recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 527 { 528 dsl_dataset_t *ds = arg1; 529 530 dmu_buf_will_dirty(ds->ds_dbuf, tx); 531 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; 532 533 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC, 534 ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld", 535 ds->ds_object); 536 } 537 538 /* 539 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() 540 * succeeds; otherwise we will leak the holds on the datasets. 541 */ 542 int 543 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb, 544 boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc) 545 { 546 int err = 0; 547 boolean_t byteswap; 548 struct recvbeginsyncarg rbsa; 549 uint64_t version; 550 int flags; 551 dsl_dataset_t *ds; 552 553 if (drrb->drr_magic == DMU_BACKUP_MAGIC) 554 byteswap = FALSE; 555 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) 556 byteswap = TRUE; 557 else 558 return (EINVAL); 559 560 rbsa.tofs = tofs; 561 rbsa.tosnap = tosnap; 562 rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL; 563 rbsa.fromguid = drrb->drr_fromguid; 564 rbsa.type = drrb->drr_type; 565 rbsa.tag = FTAG; 566 rbsa.dsflags = 0; 567 version = drrb->drr_version; 568 flags = drrb->drr_flags; 569 570 if (byteswap) { 571 rbsa.type = BSWAP_32(rbsa.type); 572 rbsa.fromguid = BSWAP_64(rbsa.fromguid); 573 version = BSWAP_64(version); 574 flags = BSWAP_32(flags); 575 } 576 577 if (version != DMU_BACKUP_STREAM_VERSION || 578 rbsa.type >= DMU_OST_NUMTYPES || 579 ((flags & DRR_FLAG_CLONE) && origin == NULL)) 580 return (EINVAL); 581 582 if (flags & DRR_FLAG_CI_DATA) 583 rbsa.dsflags = DS_FLAG_CI_DATASET; 584 585 bzero(drc, sizeof (dmu_recv_cookie_t)); 586 drc->drc_drrb = drrb; 587 drc->drc_tosnap = tosnap; 588 drc->drc_force = force; 589 590 /* 591 * Process the begin in syncing context. 592 */ 593 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) { 594 /* offline incremental receive */ 595 err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds); 596 if (err) 597 return (err); 598 599 /* 600 * Only do the rollback if the most recent snapshot 601 * matches the incremental source 602 */ 603 if (force) { 604 if (ds->ds_prev == NULL || 605 ds->ds_prev->ds_phys->ds_guid != 606 rbsa.fromguid) { 607 dsl_dataset_disown(ds, dmu_recv_tag); 608 return (ENODEV); 609 } 610 (void) dsl_dataset_rollback(ds, DMU_OST_NONE); 611 } 612 rbsa.force = B_FALSE; 613 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 614 recv_incremental_check, 615 recv_offline_incremental_sync, ds, &rbsa, 1); 616 if (err) { 617 dsl_dataset_disown(ds, dmu_recv_tag); 618 return (err); 619 } 620 drc->drc_logical_ds = drc->drc_real_ds = ds; 621 } else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) { 622 /* online incremental receive */ 623 624 /* tmp clone name is: tofs/%tosnap" */ 625 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname), 626 "%%%s", tosnap); 627 628 /* open the dataset we are logically receiving into */ 629 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds); 630 if (err) 631 return (err); 632 633 rbsa.force = force; 634 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 635 recv_incremental_check, 636 recv_online_incremental_sync, ds, &rbsa, 5); 637 if (err) { 638 dsl_dataset_rele(ds, dmu_recv_tag); 639 return (err); 640 } 641 drc->drc_logical_ds = ds; 642 drc->drc_real_ds = rbsa.ds; 643 } else { 644 /* create new fs -- full backup or clone */ 645 dsl_dir_t *dd = NULL; 646 const char *tail; 647 648 err = dsl_dir_open(tofs, FTAG, &dd, &tail); 649 if (err) 650 return (err); 651 if (tail == NULL) { 652 if (!force) { 653 dsl_dir_close(dd, FTAG); 654 return (EEXIST); 655 } 656 657 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 658 err = dsl_dataset_own_obj(dd->dd_pool, 659 dd->dd_phys->dd_head_dataset_obj, 660 DS_MODE_INCONSISTENT, FTAG, &ds); 661 rw_exit(&dd->dd_pool->dp_config_rwlock); 662 if (err) { 663 dsl_dir_close(dd, FTAG); 664 return (err); 665 } 666 667 dsl_dataset_make_exclusive(ds, FTAG); 668 err = dsl_sync_task_do(dd->dd_pool, 669 recv_full_existing_check, 670 recv_full_existing_sync, ds, &rbsa, 5); 671 dsl_dataset_disown(ds, FTAG); 672 } else { 673 err = dsl_sync_task_do(dd->dd_pool, recv_full_check, 674 recv_full_sync, dd, &rbsa, 5); 675 } 676 dsl_dir_close(dd, FTAG); 677 if (err) 678 return (err); 679 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds; 680 drc->drc_newfs = B_TRUE; 681 } 682 683 return (0); 684 } 685 686 struct restorearg { 687 int err; 688 int byteswap; 689 vnode_t *vp; 690 char *buf; 691 uint64_t voff; 692 int bufsize; /* amount of memory allocated for buf */ 693 zio_cksum_t cksum; 694 }; 695 696 static void * 697 restore_read(struct restorearg *ra, int len) 698 { 699 void *rv; 700 int done = 0; 701 702 /* some things will require 8-byte alignment, so everything must */ 703 ASSERT3U(len % 8, ==, 0); 704 705 while (done < len) { 706 ssize_t resid; 707 708 ra->err = vn_rdwr(UIO_READ, ra->vp, 709 (caddr_t)ra->buf + done, len - done, 710 ra->voff, UIO_SYSSPACE, FAPPEND, 711 RLIM64_INFINITY, CRED(), &resid); 712 713 if (resid == len - done) 714 ra->err = EINVAL; 715 ra->voff += len - done - resid; 716 done = len - resid; 717 if (ra->err) 718 return (NULL); 719 } 720 721 ASSERT3U(done, ==, len); 722 rv = ra->buf; 723 if (ra->byteswap) 724 fletcher_4_incremental_byteswap(rv, len, &ra->cksum); 725 else 726 fletcher_4_incremental_native(rv, len, &ra->cksum); 727 return (rv); 728 } 729 730 static void 731 backup_byteswap(dmu_replay_record_t *drr) 732 { 733 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) 734 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) 735 drr->drr_type = BSWAP_32(drr->drr_type); 736 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); 737 switch (drr->drr_type) { 738 case DRR_BEGIN: 739 DO64(drr_begin.drr_magic); 740 DO64(drr_begin.drr_version); 741 DO64(drr_begin.drr_creation_time); 742 DO32(drr_begin.drr_type); 743 DO32(drr_begin.drr_flags); 744 DO64(drr_begin.drr_toguid); 745 DO64(drr_begin.drr_fromguid); 746 break; 747 case DRR_OBJECT: 748 DO64(drr_object.drr_object); 749 /* DO64(drr_object.drr_allocation_txg); */ 750 DO32(drr_object.drr_type); 751 DO32(drr_object.drr_bonustype); 752 DO32(drr_object.drr_blksz); 753 DO32(drr_object.drr_bonuslen); 754 break; 755 case DRR_FREEOBJECTS: 756 DO64(drr_freeobjects.drr_firstobj); 757 DO64(drr_freeobjects.drr_numobjs); 758 break; 759 case DRR_WRITE: 760 DO64(drr_write.drr_object); 761 DO32(drr_write.drr_type); 762 DO64(drr_write.drr_offset); 763 DO64(drr_write.drr_length); 764 break; 765 case DRR_FREE: 766 DO64(drr_free.drr_object); 767 DO64(drr_free.drr_offset); 768 DO64(drr_free.drr_length); 769 break; 770 case DRR_END: 771 DO64(drr_end.drr_checksum.zc_word[0]); 772 DO64(drr_end.drr_checksum.zc_word[1]); 773 DO64(drr_end.drr_checksum.zc_word[2]); 774 DO64(drr_end.drr_checksum.zc_word[3]); 775 break; 776 } 777 #undef DO64 778 #undef DO32 779 } 780 781 static int 782 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) 783 { 784 int err; 785 dmu_tx_t *tx; 786 787 err = dmu_object_info(os, drro->drr_object, NULL); 788 789 if (err != 0 && err != ENOENT) 790 return (EINVAL); 791 792 if (drro->drr_type == DMU_OT_NONE || 793 drro->drr_type >= DMU_OT_NUMTYPES || 794 drro->drr_bonustype >= DMU_OT_NUMTYPES || 795 drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS || 796 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || 797 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || 798 drro->drr_blksz < SPA_MINBLOCKSIZE || 799 drro->drr_blksz > SPA_MAXBLOCKSIZE || 800 drro->drr_bonuslen > DN_MAX_BONUSLEN) { 801 return (EINVAL); 802 } 803 804 tx = dmu_tx_create(os); 805 806 if (err == ENOENT) { 807 /* currently free, want to be allocated */ 808 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 809 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1); 810 err = dmu_tx_assign(tx, TXG_WAIT); 811 if (err) { 812 dmu_tx_abort(tx); 813 return (err); 814 } 815 err = dmu_object_claim(os, drro->drr_object, 816 drro->drr_type, drro->drr_blksz, 817 drro->drr_bonustype, drro->drr_bonuslen, tx); 818 } else { 819 /* currently allocated, want to be allocated */ 820 dmu_tx_hold_bonus(tx, drro->drr_object); 821 /* 822 * We may change blocksize, so need to 823 * hold_write 824 */ 825 dmu_tx_hold_write(tx, drro->drr_object, 0, 1); 826 err = dmu_tx_assign(tx, TXG_WAIT); 827 if (err) { 828 dmu_tx_abort(tx); 829 return (err); 830 } 831 832 err = dmu_object_reclaim(os, drro->drr_object, 833 drro->drr_type, drro->drr_blksz, 834 drro->drr_bonustype, drro->drr_bonuslen, tx); 835 } 836 if (err) { 837 dmu_tx_commit(tx); 838 return (EINVAL); 839 } 840 841 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx); 842 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx); 843 844 if (drro->drr_bonuslen) { 845 dmu_buf_t *db; 846 void *data; 847 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db)); 848 dmu_buf_will_dirty(db, tx); 849 850 ASSERT3U(db->db_size, >=, drro->drr_bonuslen); 851 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8)); 852 if (data == NULL) { 853 dmu_tx_commit(tx); 854 return (ra->err); 855 } 856 bcopy(data, db->db_data, drro->drr_bonuslen); 857 if (ra->byteswap) { 858 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data, 859 drro->drr_bonuslen); 860 } 861 dmu_buf_rele(db, FTAG); 862 } 863 dmu_tx_commit(tx); 864 return (0); 865 } 866 867 /* ARGSUSED */ 868 static int 869 restore_freeobjects(struct restorearg *ra, objset_t *os, 870 struct drr_freeobjects *drrfo) 871 { 872 uint64_t obj; 873 874 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) 875 return (EINVAL); 876 877 for (obj = drrfo->drr_firstobj; 878 obj < drrfo->drr_firstobj + drrfo->drr_numobjs; 879 (void) dmu_object_next(os, &obj, FALSE, 0)) { 880 int err; 881 882 if (dmu_object_info(os, obj, NULL) != 0) 883 continue; 884 885 err = dmu_free_object(os, obj); 886 if (err) 887 return (err); 888 } 889 return (0); 890 } 891 892 static int 893 restore_write(struct restorearg *ra, objset_t *os, 894 struct drr_write *drrw) 895 { 896 dmu_tx_t *tx; 897 void *data; 898 int err; 899 900 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset || 901 drrw->drr_type >= DMU_OT_NUMTYPES) 902 return (EINVAL); 903 904 data = restore_read(ra, drrw->drr_length); 905 if (data == NULL) 906 return (ra->err); 907 908 if (dmu_object_info(os, drrw->drr_object, NULL) != 0) 909 return (EINVAL); 910 911 tx = dmu_tx_create(os); 912 913 dmu_tx_hold_write(tx, drrw->drr_object, 914 drrw->drr_offset, drrw->drr_length); 915 err = dmu_tx_assign(tx, TXG_WAIT); 916 if (err) { 917 dmu_tx_abort(tx); 918 return (err); 919 } 920 if (ra->byteswap) 921 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length); 922 dmu_write(os, drrw->drr_object, 923 drrw->drr_offset, drrw->drr_length, data, tx); 924 dmu_tx_commit(tx); 925 return (0); 926 } 927 928 /* ARGSUSED */ 929 static int 930 restore_free(struct restorearg *ra, objset_t *os, 931 struct drr_free *drrf) 932 { 933 int err; 934 935 if (drrf->drr_length != -1ULL && 936 drrf->drr_offset + drrf->drr_length < drrf->drr_offset) 937 return (EINVAL); 938 939 if (dmu_object_info(os, drrf->drr_object, NULL) != 0) 940 return (EINVAL); 941 942 err = dmu_free_long_range(os, drrf->drr_object, 943 drrf->drr_offset, drrf->drr_length); 944 return (err); 945 } 946 947 void 948 dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc) 949 { 950 if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) { 951 /* 952 * online incremental or new fs: destroy the fs (which 953 * may be a clone) that we created 954 */ 955 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag); 956 if (drc->drc_real_ds != drc->drc_logical_ds) 957 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag); 958 } else { 959 /* 960 * offline incremental: rollback to most recent snapshot. 961 */ 962 (void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE); 963 dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag); 964 } 965 } 966 967 /* 968 * NB: callers *must* call dmu_recv_end() if this succeeds. 969 */ 970 int 971 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp) 972 { 973 struct restorearg ra = { 0 }; 974 dmu_replay_record_t *drr; 975 objset_t *os; 976 zio_cksum_t pcksum; 977 978 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) 979 ra.byteswap = TRUE; 980 981 { 982 /* compute checksum of drr_begin record */ 983 dmu_replay_record_t *drr; 984 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 985 986 drr->drr_type = DRR_BEGIN; 987 drr->drr_u.drr_begin = *drc->drc_drrb; 988 if (ra.byteswap) { 989 fletcher_4_incremental_byteswap(drr, 990 sizeof (dmu_replay_record_t), &ra.cksum); 991 } else { 992 fletcher_4_incremental_native(drr, 993 sizeof (dmu_replay_record_t), &ra.cksum); 994 } 995 kmem_free(drr, sizeof (dmu_replay_record_t)); 996 } 997 998 if (ra.byteswap) { 999 struct drr_begin *drrb = drc->drc_drrb; 1000 drrb->drr_magic = BSWAP_64(drrb->drr_magic); 1001 drrb->drr_version = BSWAP_64(drrb->drr_version); 1002 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); 1003 drrb->drr_type = BSWAP_32(drrb->drr_type); 1004 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); 1005 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); 1006 } 1007 1008 ra.vp = vp; 1009 ra.voff = *voffp; 1010 ra.bufsize = 1<<20; 1011 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP); 1012 1013 /* these were verified in dmu_recv_begin */ 1014 ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION); 1015 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES); 1016 1017 /* 1018 * Open the objset we are modifying. 1019 */ 1020 VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0); 1021 1022 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT); 1023 1024 /* 1025 * Read records and process them. 1026 */ 1027 pcksum = ra.cksum; 1028 while (ra.err == 0 && 1029 NULL != (drr = restore_read(&ra, sizeof (*drr)))) { 1030 if (issig(JUSTLOOKING) && issig(FORREAL)) { 1031 ra.err = EINTR; 1032 goto out; 1033 } 1034 1035 if (ra.byteswap) 1036 backup_byteswap(drr); 1037 1038 switch (drr->drr_type) { 1039 case DRR_OBJECT: 1040 { 1041 /* 1042 * We need to make a copy of the record header, 1043 * because restore_{object,write} may need to 1044 * restore_read(), which will invalidate drr. 1045 */ 1046 struct drr_object drro = drr->drr_u.drr_object; 1047 ra.err = restore_object(&ra, os, &drro); 1048 break; 1049 } 1050 case DRR_FREEOBJECTS: 1051 { 1052 struct drr_freeobjects drrfo = 1053 drr->drr_u.drr_freeobjects; 1054 ra.err = restore_freeobjects(&ra, os, &drrfo); 1055 break; 1056 } 1057 case DRR_WRITE: 1058 { 1059 struct drr_write drrw = drr->drr_u.drr_write; 1060 ra.err = restore_write(&ra, os, &drrw); 1061 break; 1062 } 1063 case DRR_FREE: 1064 { 1065 struct drr_free drrf = drr->drr_u.drr_free; 1066 ra.err = restore_free(&ra, os, &drrf); 1067 break; 1068 } 1069 case DRR_END: 1070 { 1071 struct drr_end drre = drr->drr_u.drr_end; 1072 /* 1073 * We compare against the *previous* checksum 1074 * value, because the stored checksum is of 1075 * everything before the DRR_END record. 1076 */ 1077 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum)) 1078 ra.err = ECKSUM; 1079 goto out; 1080 } 1081 default: 1082 ra.err = EINVAL; 1083 goto out; 1084 } 1085 pcksum = ra.cksum; 1086 } 1087 ASSERT(ra.err != 0); 1088 1089 out: 1090 dmu_objset_close(os); 1091 1092 if (ra.err != 0) { 1093 /* 1094 * rollback or destroy what we created, so we don't 1095 * leave it in the restoring state. 1096 */ 1097 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0); 1098 dmu_recv_abort_cleanup(drc); 1099 } 1100 1101 kmem_free(ra.buf, ra.bufsize); 1102 *voffp = ra.voff; 1103 return (ra.err); 1104 } 1105 1106 struct recvendsyncarg { 1107 char *tosnap; 1108 uint64_t creation_time; 1109 uint64_t toguid; 1110 }; 1111 1112 static int 1113 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx) 1114 { 1115 dsl_dataset_t *ds = arg1; 1116 struct recvendsyncarg *resa = arg2; 1117 1118 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx)); 1119 } 1120 1121 static void 1122 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 1123 { 1124 dsl_dataset_t *ds = arg1; 1125 struct recvendsyncarg *resa = arg2; 1126 1127 dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx); 1128 1129 /* set snapshot's creation time and guid */ 1130 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1131 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time; 1132 ds->ds_prev->ds_phys->ds_guid = resa->toguid; 1133 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; 1134 1135 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1136 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; 1137 } 1138 1139 int 1140 dmu_recv_end(dmu_recv_cookie_t *drc) 1141 { 1142 struct recvendsyncarg resa; 1143 dsl_dataset_t *ds = drc->drc_logical_ds; 1144 int err; 1145 1146 /* 1147 * XXX hack; seems the ds is still dirty and 1148 * dsl_pool_zil_clean() expects it to have a ds_user_ptr 1149 * (and zil), but clone_swap() can close it. 1150 */ 1151 txg_wait_synced(ds->ds_dir->dd_pool, 0); 1152 1153 if (ds != drc->drc_real_ds) { 1154 /* we are doing an online recv */ 1155 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) { 1156 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds, 1157 drc->drc_force); 1158 if (err) 1159 dsl_dataset_disown(ds, dmu_recv_tag); 1160 } else { 1161 err = EBUSY; 1162 dsl_dataset_rele(ds, dmu_recv_tag); 1163 } 1164 /* dsl_dataset_destroy() will disown the ds */ 1165 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag); 1166 if (err) 1167 return (err); 1168 } 1169 1170 resa.creation_time = drc->drc_drrb->drr_creation_time; 1171 resa.toguid = drc->drc_drrb->drr_toguid; 1172 resa.tosnap = drc->drc_tosnap; 1173 1174 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 1175 recv_end_check, recv_end_sync, ds, &resa, 3); 1176 if (err) { 1177 if (drc->drc_newfs) { 1178 ASSERT(ds == drc->drc_real_ds); 1179 (void) dsl_dataset_destroy(ds, dmu_recv_tag); 1180 return (err); 1181 } else { 1182 (void) dsl_dataset_rollback(ds, DMU_OST_NONE); 1183 } 1184 } 1185 1186 /* release the hold from dmu_recv_begin */ 1187 dsl_dataset_disown(ds, dmu_recv_tag); 1188 return (err); 1189 } 1190