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 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #include <sys/dmu.h> 27 #include <sys/dmu_impl.h> 28 #include <sys/dmu_tx.h> 29 #include <sys/dbuf.h> 30 #include <sys/dnode.h> 31 #include <sys/zfs_context.h> 32 #include <sys/dmu_objset.h> 33 #include <sys/dmu_traverse.h> 34 #include <sys/dsl_dataset.h> 35 #include <sys/dsl_dir.h> 36 #include <sys/dsl_pool.h> 37 #include <sys/dsl_synctask.h> 38 #include <sys/zfs_ioctl.h> 39 #include <sys/zap.h> 40 #include <sys/zio_checksum.h> 41 42 static char *dmu_recv_tag = "dmu_recv_tag"; 43 44 struct backuparg { 45 dmu_replay_record_t *drr; 46 vnode_t *vp; 47 offset_t *off; 48 objset_t *os; 49 zio_cksum_t zc; 50 int err; 51 }; 52 53 static int 54 dump_bytes(struct backuparg *ba, void *buf, int len) 55 { 56 ssize_t resid; /* have to get resid to get detailed errno */ 57 ASSERT3U(len % 8, ==, 0); 58 59 fletcher_4_incremental_native(buf, len, &ba->zc); 60 ba->err = vn_rdwr(UIO_WRITE, ba->vp, 61 (caddr_t)buf, len, 62 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid); 63 *ba->off += len; 64 return (ba->err); 65 } 66 67 static int 68 dump_free(struct backuparg *ba, uint64_t object, uint64_t offset, 69 uint64_t length) 70 { 71 /* write a FREE record */ 72 bzero(ba->drr, sizeof (dmu_replay_record_t)); 73 ba->drr->drr_type = DRR_FREE; 74 ba->drr->drr_u.drr_free.drr_object = object; 75 ba->drr->drr_u.drr_free.drr_offset = offset; 76 ba->drr->drr_u.drr_free.drr_length = length; 77 78 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t))) 79 return (EINTR); 80 return (0); 81 } 82 83 static int 84 dump_data(struct backuparg *ba, dmu_object_type_t type, 85 uint64_t object, uint64_t offset, int blksz, void *data) 86 { 87 /* write a DATA record */ 88 bzero(ba->drr, sizeof (dmu_replay_record_t)); 89 ba->drr->drr_type = DRR_WRITE; 90 ba->drr->drr_u.drr_write.drr_object = object; 91 ba->drr->drr_u.drr_write.drr_type = type; 92 ba->drr->drr_u.drr_write.drr_offset = offset; 93 ba->drr->drr_u.drr_write.drr_length = blksz; 94 95 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t))) 96 return (EINTR); 97 if (dump_bytes(ba, data, blksz)) 98 return (EINTR); 99 return (0); 100 } 101 102 static int 103 dump_freeobjects(struct backuparg *ba, uint64_t firstobj, uint64_t numobjs) 104 { 105 /* write a FREEOBJECTS record */ 106 bzero(ba->drr, sizeof (dmu_replay_record_t)); 107 ba->drr->drr_type = DRR_FREEOBJECTS; 108 ba->drr->drr_u.drr_freeobjects.drr_firstobj = firstobj; 109 ba->drr->drr_u.drr_freeobjects.drr_numobjs = numobjs; 110 111 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t))) 112 return (EINTR); 113 return (0); 114 } 115 116 static int 117 dump_dnode(struct backuparg *ba, uint64_t object, dnode_phys_t *dnp) 118 { 119 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) 120 return (dump_freeobjects(ba, object, 1)); 121 122 /* write an OBJECT record */ 123 bzero(ba->drr, sizeof (dmu_replay_record_t)); 124 ba->drr->drr_type = DRR_OBJECT; 125 ba->drr->drr_u.drr_object.drr_object = object; 126 ba->drr->drr_u.drr_object.drr_type = dnp->dn_type; 127 ba->drr->drr_u.drr_object.drr_bonustype = dnp->dn_bonustype; 128 ba->drr->drr_u.drr_object.drr_blksz = 129 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 130 ba->drr->drr_u.drr_object.drr_bonuslen = dnp->dn_bonuslen; 131 ba->drr->drr_u.drr_object.drr_checksum = dnp->dn_checksum; 132 ba->drr->drr_u.drr_object.drr_compress = dnp->dn_compress; 133 134 if (dump_bytes(ba, ba->drr, sizeof (dmu_replay_record_t))) 135 return (EINTR); 136 137 if (dump_bytes(ba, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8))) 138 return (EINTR); 139 140 /* free anything past the end of the file */ 141 if (dump_free(ba, object, (dnp->dn_maxblkid + 1) * 142 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL)) 143 return (EINTR); 144 if (ba->err) 145 return (EINTR); 146 return (0); 147 } 148 149 #define BP_SPAN(dnp, level) \ 150 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \ 151 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) 152 153 static int 154 backup_cb(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb, 155 const dnode_phys_t *dnp, void *arg) 156 { 157 struct backuparg *ba = arg; 158 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; 159 int err = 0; 160 161 if (issig(JUSTLOOKING) && issig(FORREAL)) 162 return (EINTR); 163 164 if (zb->zb_object != 0 && DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { 165 return (0); 166 } else if (bp == NULL && zb->zb_object == 0) { 167 uint64_t span = BP_SPAN(dnp, zb->zb_level); 168 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; 169 err = dump_freeobjects(ba, dnobj, span >> DNODE_SHIFT); 170 } else if (bp == NULL) { 171 uint64_t span = BP_SPAN(dnp, zb->zb_level); 172 err = dump_free(ba, zb->zb_object, zb->zb_blkid * span, span); 173 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { 174 return (0); 175 } else if (type == DMU_OT_DNODE) { 176 dnode_phys_t *blk; 177 int i; 178 int blksz = BP_GET_LSIZE(bp); 179 uint32_t aflags = ARC_WAIT; 180 arc_buf_t *abuf; 181 182 if (arc_read_nolock(NULL, spa, bp, 183 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ, 184 ZIO_FLAG_CANFAIL, &aflags, zb) != 0) 185 return (EIO); 186 187 blk = abuf->b_data; 188 for (i = 0; i < blksz >> DNODE_SHIFT; i++) { 189 uint64_t dnobj = (zb->zb_blkid << 190 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i; 191 err = dump_dnode(ba, dnobj, blk+i); 192 if (err) 193 break; 194 } 195 (void) arc_buf_remove_ref(abuf, &abuf); 196 } else { /* it's a level-0 block of a regular object */ 197 uint32_t aflags = ARC_WAIT; 198 arc_buf_t *abuf; 199 int blksz = BP_GET_LSIZE(bp); 200 201 if (arc_read_nolock(NULL, spa, bp, 202 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ, 203 ZIO_FLAG_CANFAIL, &aflags, zb) != 0) 204 return (EIO); 205 206 err = dump_data(ba, type, zb->zb_object, zb->zb_blkid * blksz, 207 blksz, abuf->b_data); 208 (void) arc_buf_remove_ref(abuf, &abuf); 209 } 210 211 ASSERT(err == 0 || err == EINTR); 212 return (err); 213 } 214 215 int 216 dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin, 217 vnode_t *vp, offset_t *off) 218 { 219 dsl_dataset_t *ds = tosnap->os->os_dsl_dataset; 220 dsl_dataset_t *fromds = fromsnap ? fromsnap->os->os_dsl_dataset : NULL; 221 dmu_replay_record_t *drr; 222 struct backuparg ba; 223 int err; 224 uint64_t fromtxg = 0; 225 226 /* tosnap must be a snapshot */ 227 if (ds->ds_phys->ds_next_snap_obj == 0) 228 return (EINVAL); 229 230 /* fromsnap must be an earlier snapshot from the same fs as tosnap */ 231 if (fromds && (ds->ds_dir != fromds->ds_dir || 232 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg)) 233 return (EXDEV); 234 235 if (fromorigin) { 236 dsl_pool_t *dp = ds->ds_dir->dd_pool; 237 238 if (fromsnap) 239 return (EINVAL); 240 241 if (dsl_dir_is_clone(ds->ds_dir)) { 242 rw_enter(&dp->dp_config_rwlock, RW_READER); 243 err = dsl_dataset_hold_obj(dp, 244 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds); 245 rw_exit(&dp->dp_config_rwlock); 246 if (err) 247 return (err); 248 } else { 249 fromorigin = B_FALSE; 250 } 251 } 252 253 254 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 255 drr->drr_type = DRR_BEGIN; 256 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 257 drr->drr_u.drr_begin.drr_version = DMU_BACKUP_STREAM_VERSION; 258 drr->drr_u.drr_begin.drr_creation_time = 259 ds->ds_phys->ds_creation_time; 260 drr->drr_u.drr_begin.drr_type = tosnap->os->os_phys->os_type; 261 if (fromorigin) 262 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; 263 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid; 264 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 265 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; 266 267 if (fromds) 268 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid; 269 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname); 270 271 if (fromds) 272 fromtxg = fromds->ds_phys->ds_creation_txg; 273 if (fromorigin) 274 dsl_dataset_rele(fromds, FTAG); 275 276 ba.drr = drr; 277 ba.vp = vp; 278 ba.os = tosnap; 279 ba.off = off; 280 ZIO_SET_CHECKSUM(&ba.zc, 0, 0, 0, 0); 281 282 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) { 283 kmem_free(drr, sizeof (dmu_replay_record_t)); 284 return (ba.err); 285 } 286 287 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH, 288 backup_cb, &ba); 289 290 if (err) { 291 if (err == EINTR && ba.err) 292 err = ba.err; 293 kmem_free(drr, sizeof (dmu_replay_record_t)); 294 return (err); 295 } 296 297 bzero(drr, sizeof (dmu_replay_record_t)); 298 drr->drr_type = DRR_END; 299 drr->drr_u.drr_end.drr_checksum = ba.zc; 300 301 if (dump_bytes(&ba, drr, sizeof (dmu_replay_record_t))) { 302 kmem_free(drr, sizeof (dmu_replay_record_t)); 303 return (ba.err); 304 } 305 306 kmem_free(drr, sizeof (dmu_replay_record_t)); 307 308 return (0); 309 } 310 311 struct recvbeginsyncarg { 312 const char *tofs; 313 const char *tosnap; 314 dsl_dataset_t *origin; 315 uint64_t fromguid; 316 dmu_objset_type_t type; 317 void *tag; 318 boolean_t force; 319 uint64_t dsflags; 320 char clonelastname[MAXNAMELEN]; 321 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */ 322 }; 323 324 /* ARGSUSED */ 325 static int 326 recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx) 327 { 328 dsl_dir_t *dd = arg1; 329 struct recvbeginsyncarg *rbsa = arg2; 330 objset_t *mos = dd->dd_pool->dp_meta_objset; 331 uint64_t val; 332 int err; 333 334 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj, 335 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val); 336 337 if (err != ENOENT) 338 return (err ? err : EEXIST); 339 340 if (rbsa->origin) { 341 /* make sure it's a snap in the same pool */ 342 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool) 343 return (EXDEV); 344 if (!dsl_dataset_is_snapshot(rbsa->origin)) 345 return (EINVAL); 346 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid) 347 return (ENODEV); 348 } 349 350 return (0); 351 } 352 353 static void 354 recv_new_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 355 { 356 dsl_dir_t *dd = arg1; 357 struct recvbeginsyncarg *rbsa = arg2; 358 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags; 359 uint64_t dsobj; 360 361 /* Create and open new dataset. */ 362 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1, 363 rbsa->origin, flags, cr, tx); 364 VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj, 365 DS_MODE_INCONSISTENT, dmu_recv_tag, &rbsa->ds)); 366 367 if (rbsa->origin == NULL) { 368 (void) dmu_objset_create_impl(dd->dd_pool->dp_spa, 369 rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx); 370 } 371 372 spa_history_internal_log(LOG_DS_REPLAY_FULL_SYNC, 373 dd->dd_pool->dp_spa, tx, cr, "dataset = %lld", dsobj); 374 } 375 376 /* ARGSUSED */ 377 static int 378 recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx) 379 { 380 dsl_dataset_t *ds = arg1; 381 struct recvbeginsyncarg *rbsa = arg2; 382 int err; 383 uint64_t val; 384 385 /* must not have any changes since most recent snapshot */ 386 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds)) 387 return (ETXTBSY); 388 389 if (rbsa->fromguid) { 390 /* if incremental, most recent snapshot must match fromguid */ 391 if (ds->ds_prev == NULL) 392 return (ENODEV); 393 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) 394 return (ENODEV); 395 } else { 396 /* if full, most recent snapshot must be $ORIGIN */ 397 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL) 398 return (ENODEV); 399 } 400 401 /* temporary clone name must not exist */ 402 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, 403 ds->ds_dir->dd_phys->dd_child_dir_zapobj, 404 rbsa->clonelastname, 8, 1, &val); 405 if (err == 0) 406 return (EEXIST); 407 if (err != ENOENT) 408 return (err); 409 410 /* new snapshot name must not exist */ 411 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, 412 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val); 413 if (err == 0) 414 return (EEXIST); 415 if (err != ENOENT) 416 return (err); 417 return (0); 418 } 419 420 /* ARGSUSED */ 421 static void 422 recv_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 423 { 424 dsl_dataset_t *ohds = arg1; 425 struct recvbeginsyncarg *rbsa = arg2; 426 dsl_pool_t *dp = ohds->ds_dir->dd_pool; 427 dsl_dataset_t *cds; 428 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags; 429 uint64_t dsobj; 430 431 /* create and open the temporary clone */ 432 dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname, 433 ohds->ds_prev, flags, cr, tx); 434 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, 435 DS_MODE_INCONSISTENT, dmu_recv_tag, &cds)); 436 437 /* 438 * If we actually created a non-clone, we need to create the 439 * objset in our new dataset. 440 */ 441 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) { 442 (void) dmu_objset_create_impl(dp->dp_spa, 443 cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx); 444 } 445 446 /* copy the refquota from the target fs to the clone */ 447 if (ohds->ds_quota > 0) 448 dsl_dataset_set_quota_sync(cds, &ohds->ds_quota, cr, tx); 449 450 rbsa->ds = cds; 451 452 spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC, 453 dp->dp_spa, tx, cr, "dataset = %lld", dsobj); 454 } 455 456 /* 457 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() 458 * succeeds; otherwise we will leak the holds on the datasets. 459 */ 460 int 461 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb, 462 boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc) 463 { 464 int err = 0; 465 boolean_t byteswap; 466 struct recvbeginsyncarg rbsa = { 0 }; 467 uint64_t version; 468 int flags; 469 dsl_dataset_t *ds; 470 471 if (drrb->drr_magic == DMU_BACKUP_MAGIC) 472 byteswap = FALSE; 473 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) 474 byteswap = TRUE; 475 else 476 return (EINVAL); 477 478 rbsa.tofs = tofs; 479 rbsa.tosnap = tosnap; 480 rbsa.origin = origin ? origin->os->os_dsl_dataset : NULL; 481 rbsa.fromguid = drrb->drr_fromguid; 482 rbsa.type = drrb->drr_type; 483 rbsa.tag = FTAG; 484 rbsa.dsflags = 0; 485 version = drrb->drr_version; 486 flags = drrb->drr_flags; 487 488 if (byteswap) { 489 rbsa.type = BSWAP_32(rbsa.type); 490 rbsa.fromguid = BSWAP_64(rbsa.fromguid); 491 version = BSWAP_64(version); 492 flags = BSWAP_32(flags); 493 } 494 495 if (version != DMU_BACKUP_STREAM_VERSION || 496 rbsa.type >= DMU_OST_NUMTYPES || 497 ((flags & DRR_FLAG_CLONE) && origin == NULL)) 498 return (EINVAL); 499 500 if (flags & DRR_FLAG_CI_DATA) 501 rbsa.dsflags = DS_FLAG_CI_DATASET; 502 503 bzero(drc, sizeof (dmu_recv_cookie_t)); 504 drc->drc_drrb = drrb; 505 drc->drc_tosnap = tosnap; 506 drc->drc_force = force; 507 508 /* 509 * Process the begin in syncing context. 510 */ 511 512 /* open the dataset we are logically receiving into */ 513 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds); 514 if (err == 0) { 515 /* target fs already exists; recv into temp clone */ 516 517 /* Can't recv a clone into an existing fs */ 518 if (flags & DRR_FLAG_CLONE) { 519 dsl_dataset_rele(ds, dmu_recv_tag); 520 return (EINVAL); 521 } 522 523 /* must not have an incremental recv already in progress */ 524 if (!mutex_tryenter(&ds->ds_recvlock)) { 525 dsl_dataset_rele(ds, dmu_recv_tag); 526 return (EBUSY); 527 } 528 529 /* tmp clone name is: tofs/%tosnap" */ 530 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname), 531 "%%%s", tosnap); 532 rbsa.force = force; 533 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 534 recv_existing_check, recv_existing_sync, ds, &rbsa, 5); 535 if (err) { 536 mutex_exit(&ds->ds_recvlock); 537 dsl_dataset_rele(ds, dmu_recv_tag); 538 return (err); 539 } 540 drc->drc_logical_ds = ds; 541 drc->drc_real_ds = rbsa.ds; 542 } else if (err == ENOENT) { 543 /* target fs does not exist; must be a full backup or clone */ 544 dsl_dataset_t *parent; 545 char *cp; 546 547 /* 548 * If it's a non-clone incremental, we are missing the 549 * target fs, so fail the recv. 550 */ 551 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) 552 return (ENOENT); 553 554 /* Open the parent of tofs */ 555 cp = strrchr(tofs, '/'); 556 *cp = '\0'; 557 err = dsl_dataset_hold(tofs, FTAG, &parent); 558 *cp = '/'; 559 if (err) 560 return (err); 561 562 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 563 recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5); 564 dsl_dataset_rele(parent, FTAG); 565 if (err) 566 return (err); 567 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds; 568 drc->drc_newfs = B_TRUE; 569 } 570 571 return (err); 572 } 573 574 struct restorearg { 575 int err; 576 int byteswap; 577 vnode_t *vp; 578 char *buf; 579 uint64_t voff; 580 int bufsize; /* amount of memory allocated for buf */ 581 zio_cksum_t cksum; 582 }; 583 584 static void * 585 restore_read(struct restorearg *ra, int len) 586 { 587 void *rv; 588 int done = 0; 589 590 /* some things will require 8-byte alignment, so everything must */ 591 ASSERT3U(len % 8, ==, 0); 592 593 while (done < len) { 594 ssize_t resid; 595 596 ra->err = vn_rdwr(UIO_READ, ra->vp, 597 (caddr_t)ra->buf + done, len - done, 598 ra->voff, UIO_SYSSPACE, FAPPEND, 599 RLIM64_INFINITY, CRED(), &resid); 600 601 if (resid == len - done) 602 ra->err = EINVAL; 603 ra->voff += len - done - resid; 604 done = len - resid; 605 if (ra->err) 606 return (NULL); 607 } 608 609 ASSERT3U(done, ==, len); 610 rv = ra->buf; 611 if (ra->byteswap) 612 fletcher_4_incremental_byteswap(rv, len, &ra->cksum); 613 else 614 fletcher_4_incremental_native(rv, len, &ra->cksum); 615 return (rv); 616 } 617 618 static void 619 backup_byteswap(dmu_replay_record_t *drr) 620 { 621 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) 622 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) 623 drr->drr_type = BSWAP_32(drr->drr_type); 624 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); 625 switch (drr->drr_type) { 626 case DRR_BEGIN: 627 DO64(drr_begin.drr_magic); 628 DO64(drr_begin.drr_version); 629 DO64(drr_begin.drr_creation_time); 630 DO32(drr_begin.drr_type); 631 DO32(drr_begin.drr_flags); 632 DO64(drr_begin.drr_toguid); 633 DO64(drr_begin.drr_fromguid); 634 break; 635 case DRR_OBJECT: 636 DO64(drr_object.drr_object); 637 /* DO64(drr_object.drr_allocation_txg); */ 638 DO32(drr_object.drr_type); 639 DO32(drr_object.drr_bonustype); 640 DO32(drr_object.drr_blksz); 641 DO32(drr_object.drr_bonuslen); 642 break; 643 case DRR_FREEOBJECTS: 644 DO64(drr_freeobjects.drr_firstobj); 645 DO64(drr_freeobjects.drr_numobjs); 646 break; 647 case DRR_WRITE: 648 DO64(drr_write.drr_object); 649 DO32(drr_write.drr_type); 650 DO64(drr_write.drr_offset); 651 DO64(drr_write.drr_length); 652 break; 653 case DRR_FREE: 654 DO64(drr_free.drr_object); 655 DO64(drr_free.drr_offset); 656 DO64(drr_free.drr_length); 657 break; 658 case DRR_END: 659 DO64(drr_end.drr_checksum.zc_word[0]); 660 DO64(drr_end.drr_checksum.zc_word[1]); 661 DO64(drr_end.drr_checksum.zc_word[2]); 662 DO64(drr_end.drr_checksum.zc_word[3]); 663 break; 664 } 665 #undef DO64 666 #undef DO32 667 } 668 669 static int 670 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) 671 { 672 int err; 673 dmu_tx_t *tx; 674 void *data = NULL; 675 676 if (drro->drr_type == DMU_OT_NONE || 677 drro->drr_type >= DMU_OT_NUMTYPES || 678 drro->drr_bonustype >= DMU_OT_NUMTYPES || 679 drro->drr_checksum >= ZIO_CHECKSUM_FUNCTIONS || 680 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || 681 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || 682 drro->drr_blksz < SPA_MINBLOCKSIZE || 683 drro->drr_blksz > SPA_MAXBLOCKSIZE || 684 drro->drr_bonuslen > DN_MAX_BONUSLEN) { 685 return (EINVAL); 686 } 687 688 err = dmu_object_info(os, drro->drr_object, NULL); 689 690 if (err != 0 && err != ENOENT) 691 return (EINVAL); 692 693 if (drro->drr_bonuslen) { 694 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8)); 695 if (ra->err) 696 return (ra->err); 697 } 698 699 if (err == ENOENT) { 700 /* currently free, want to be allocated */ 701 tx = dmu_tx_create(os); 702 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 703 err = dmu_tx_assign(tx, TXG_WAIT); 704 if (err) { 705 dmu_tx_abort(tx); 706 return (err); 707 } 708 err = dmu_object_claim(os, drro->drr_object, 709 drro->drr_type, drro->drr_blksz, 710 drro->drr_bonustype, drro->drr_bonuslen, tx); 711 dmu_tx_commit(tx); 712 } else { 713 /* currently allocated, want to be allocated */ 714 err = dmu_object_reclaim(os, drro->drr_object, 715 drro->drr_type, drro->drr_blksz, 716 drro->drr_bonustype, drro->drr_bonuslen); 717 } 718 if (err) 719 return (EINVAL); 720 721 tx = dmu_tx_create(os); 722 dmu_tx_hold_bonus(tx, drro->drr_object); 723 err = dmu_tx_assign(tx, TXG_WAIT); 724 if (err) { 725 dmu_tx_abort(tx); 726 return (err); 727 } 728 729 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksum, tx); 730 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx); 731 732 if (data != NULL) { 733 dmu_buf_t *db; 734 735 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db)); 736 dmu_buf_will_dirty(db, tx); 737 738 ASSERT3U(db->db_size, >=, drro->drr_bonuslen); 739 bcopy(data, db->db_data, drro->drr_bonuslen); 740 if (ra->byteswap) { 741 dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data, 742 drro->drr_bonuslen); 743 } 744 dmu_buf_rele(db, FTAG); 745 } 746 dmu_tx_commit(tx); 747 return (0); 748 } 749 750 /* ARGSUSED */ 751 static int 752 restore_freeobjects(struct restorearg *ra, objset_t *os, 753 struct drr_freeobjects *drrfo) 754 { 755 uint64_t obj; 756 757 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) 758 return (EINVAL); 759 760 for (obj = drrfo->drr_firstobj; 761 obj < drrfo->drr_firstobj + drrfo->drr_numobjs; 762 (void) dmu_object_next(os, &obj, FALSE, 0)) { 763 int err; 764 765 if (dmu_object_info(os, obj, NULL) != 0) 766 continue; 767 768 err = dmu_free_object(os, obj); 769 if (err) 770 return (err); 771 } 772 return (0); 773 } 774 775 static int 776 restore_write(struct restorearg *ra, objset_t *os, 777 struct drr_write *drrw) 778 { 779 dmu_tx_t *tx; 780 void *data; 781 int err; 782 783 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset || 784 drrw->drr_type >= DMU_OT_NUMTYPES) 785 return (EINVAL); 786 787 data = restore_read(ra, drrw->drr_length); 788 if (data == NULL) 789 return (ra->err); 790 791 if (dmu_object_info(os, drrw->drr_object, NULL) != 0) 792 return (EINVAL); 793 794 tx = dmu_tx_create(os); 795 796 dmu_tx_hold_write(tx, drrw->drr_object, 797 drrw->drr_offset, drrw->drr_length); 798 err = dmu_tx_assign(tx, TXG_WAIT); 799 if (err) { 800 dmu_tx_abort(tx); 801 return (err); 802 } 803 if (ra->byteswap) 804 dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length); 805 dmu_write(os, drrw->drr_object, 806 drrw->drr_offset, drrw->drr_length, data, tx); 807 dmu_tx_commit(tx); 808 return (0); 809 } 810 811 /* ARGSUSED */ 812 static int 813 restore_free(struct restorearg *ra, objset_t *os, 814 struct drr_free *drrf) 815 { 816 int err; 817 818 if (drrf->drr_length != -1ULL && 819 drrf->drr_offset + drrf->drr_length < drrf->drr_offset) 820 return (EINVAL); 821 822 if (dmu_object_info(os, drrf->drr_object, NULL) != 0) 823 return (EINVAL); 824 825 err = dmu_free_long_range(os, drrf->drr_object, 826 drrf->drr_offset, drrf->drr_length); 827 return (err); 828 } 829 830 /* 831 * NB: callers *must* call dmu_recv_end() if this succeeds. 832 */ 833 int 834 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp) 835 { 836 struct restorearg ra = { 0 }; 837 dmu_replay_record_t *drr; 838 objset_t *os; 839 zio_cksum_t pcksum; 840 841 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) 842 ra.byteswap = TRUE; 843 844 { 845 /* compute checksum of drr_begin record */ 846 dmu_replay_record_t *drr; 847 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 848 849 drr->drr_type = DRR_BEGIN; 850 drr->drr_u.drr_begin = *drc->drc_drrb; 851 if (ra.byteswap) { 852 fletcher_4_incremental_byteswap(drr, 853 sizeof (dmu_replay_record_t), &ra.cksum); 854 } else { 855 fletcher_4_incremental_native(drr, 856 sizeof (dmu_replay_record_t), &ra.cksum); 857 } 858 kmem_free(drr, sizeof (dmu_replay_record_t)); 859 } 860 861 if (ra.byteswap) { 862 struct drr_begin *drrb = drc->drc_drrb; 863 drrb->drr_magic = BSWAP_64(drrb->drr_magic); 864 drrb->drr_version = BSWAP_64(drrb->drr_version); 865 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); 866 drrb->drr_type = BSWAP_32(drrb->drr_type); 867 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); 868 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); 869 } 870 871 ra.vp = vp; 872 ra.voff = *voffp; 873 ra.bufsize = 1<<20; 874 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP); 875 876 /* these were verified in dmu_recv_begin */ 877 ASSERT(drc->drc_drrb->drr_version == DMU_BACKUP_STREAM_VERSION); 878 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES); 879 880 /* 881 * Open the objset we are modifying. 882 */ 883 VERIFY(dmu_objset_open_ds(drc->drc_real_ds, DMU_OST_ANY, &os) == 0); 884 885 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT); 886 887 /* 888 * Read records and process them. 889 */ 890 pcksum = ra.cksum; 891 while (ra.err == 0 && 892 NULL != (drr = restore_read(&ra, sizeof (*drr)))) { 893 if (issig(JUSTLOOKING) && issig(FORREAL)) { 894 ra.err = EINTR; 895 goto out; 896 } 897 898 if (ra.byteswap) 899 backup_byteswap(drr); 900 901 switch (drr->drr_type) { 902 case DRR_OBJECT: 903 { 904 /* 905 * We need to make a copy of the record header, 906 * because restore_{object,write} may need to 907 * restore_read(), which will invalidate drr. 908 */ 909 struct drr_object drro = drr->drr_u.drr_object; 910 ra.err = restore_object(&ra, os, &drro); 911 break; 912 } 913 case DRR_FREEOBJECTS: 914 { 915 struct drr_freeobjects drrfo = 916 drr->drr_u.drr_freeobjects; 917 ra.err = restore_freeobjects(&ra, os, &drrfo); 918 break; 919 } 920 case DRR_WRITE: 921 { 922 struct drr_write drrw = drr->drr_u.drr_write; 923 ra.err = restore_write(&ra, os, &drrw); 924 break; 925 } 926 case DRR_FREE: 927 { 928 struct drr_free drrf = drr->drr_u.drr_free; 929 ra.err = restore_free(&ra, os, &drrf); 930 break; 931 } 932 case DRR_END: 933 { 934 struct drr_end drre = drr->drr_u.drr_end; 935 /* 936 * We compare against the *previous* checksum 937 * value, because the stored checksum is of 938 * everything before the DRR_END record. 939 */ 940 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum)) 941 ra.err = ECKSUM; 942 goto out; 943 } 944 default: 945 ra.err = EINVAL; 946 goto out; 947 } 948 pcksum = ra.cksum; 949 } 950 ASSERT(ra.err != 0); 951 952 out: 953 dmu_objset_close(os); 954 955 if (ra.err != 0) { 956 /* 957 * destroy what we created, so we don't leave it in the 958 * inconsistent restoring state. 959 */ 960 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0); 961 962 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, 963 B_FALSE); 964 if (drc->drc_real_ds != drc->drc_logical_ds) { 965 mutex_exit(&drc->drc_logical_ds->ds_recvlock); 966 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag); 967 } 968 } 969 970 kmem_free(ra.buf, ra.bufsize); 971 *voffp = ra.voff; 972 return (ra.err); 973 } 974 975 struct recvendsyncarg { 976 char *tosnap; 977 uint64_t creation_time; 978 uint64_t toguid; 979 }; 980 981 static int 982 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx) 983 { 984 dsl_dataset_t *ds = arg1; 985 struct recvendsyncarg *resa = arg2; 986 987 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx)); 988 } 989 990 static void 991 recv_end_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 992 { 993 dsl_dataset_t *ds = arg1; 994 struct recvendsyncarg *resa = arg2; 995 996 dsl_dataset_snapshot_sync(ds, resa->tosnap, cr, tx); 997 998 /* set snapshot's creation time and guid */ 999 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1000 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time; 1001 ds->ds_prev->ds_phys->ds_guid = resa->toguid; 1002 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; 1003 1004 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1005 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; 1006 } 1007 1008 static int 1009 dmu_recv_existing_end(dmu_recv_cookie_t *drc) 1010 { 1011 struct recvendsyncarg resa; 1012 dsl_dataset_t *ds = drc->drc_logical_ds; 1013 int err; 1014 1015 /* 1016 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean() 1017 * expects it to have a ds_user_ptr (and zil), but clone_swap() 1018 * can close it. 1019 */ 1020 txg_wait_synced(ds->ds_dir->dd_pool, 0); 1021 1022 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) { 1023 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds, 1024 drc->drc_force); 1025 if (err) 1026 goto out; 1027 } else { 1028 mutex_exit(&ds->ds_recvlock); 1029 dsl_dataset_rele(ds, dmu_recv_tag); 1030 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, 1031 B_FALSE); 1032 return (EBUSY); 1033 } 1034 1035 resa.creation_time = drc->drc_drrb->drr_creation_time; 1036 resa.toguid = drc->drc_drrb->drr_toguid; 1037 resa.tosnap = drc->drc_tosnap; 1038 1039 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 1040 recv_end_check, recv_end_sync, ds, &resa, 3); 1041 if (err) { 1042 /* swap back */ 1043 (void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE); 1044 } 1045 1046 out: 1047 mutex_exit(&ds->ds_recvlock); 1048 dsl_dataset_disown(ds, dmu_recv_tag); 1049 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE); 1050 return (err); 1051 } 1052 1053 static int 1054 dmu_recv_new_end(dmu_recv_cookie_t *drc) 1055 { 1056 struct recvendsyncarg resa; 1057 dsl_dataset_t *ds = drc->drc_logical_ds; 1058 int err; 1059 1060 /* 1061 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean() 1062 * expects it to have a ds_user_ptr (and zil), but clone_swap() 1063 * can close it. 1064 */ 1065 txg_wait_synced(ds->ds_dir->dd_pool, 0); 1066 1067 resa.creation_time = drc->drc_drrb->drr_creation_time; 1068 resa.toguid = drc->drc_drrb->drr_toguid; 1069 resa.tosnap = drc->drc_tosnap; 1070 1071 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 1072 recv_end_check, recv_end_sync, ds, &resa, 3); 1073 if (err) { 1074 /* clean up the fs we just recv'd into */ 1075 (void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE); 1076 } else { 1077 /* release the hold from dmu_recv_begin */ 1078 dsl_dataset_disown(ds, dmu_recv_tag); 1079 } 1080 return (err); 1081 } 1082 1083 int 1084 dmu_recv_end(dmu_recv_cookie_t *drc) 1085 { 1086 if (drc->drc_logical_ds != drc->drc_real_ds) 1087 return (dmu_recv_existing_end(drc)); 1088 else 1089 return (dmu_recv_new_end(drc)); 1090 } 1091