dmu_send.c revision 277586
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 25 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2012, Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 * Copyright 2014 HybridCluster. All rights reserved. 28 */ 29 30#include <sys/dmu.h> 31#include <sys/dmu_impl.h> 32#include <sys/dmu_tx.h> 33#include <sys/dbuf.h> 34#include <sys/dnode.h> 35#include <sys/zfs_context.h> 36#include <sys/dmu_objset.h> 37#include <sys/dmu_traverse.h> 38#include <sys/dsl_dataset.h> 39#include <sys/dsl_dir.h> 40#include <sys/dsl_prop.h> 41#include <sys/dsl_pool.h> 42#include <sys/dsl_synctask.h> 43#include <sys/zfs_ioctl.h> 44#include <sys/zap.h> 45#include <sys/zio_checksum.h> 46#include <sys/zfs_znode.h> 47#include <zfs_fletcher.h> 48#include <sys/avl.h> 49#include <sys/ddt.h> 50#include <sys/zfs_onexit.h> 51#include <sys/dmu_send.h> 52#include <sys/dsl_destroy.h> 53#include <sys/blkptr.h> 54#include <sys/dsl_bookmark.h> 55#include <sys/zfeature.h> 56 57#ifdef __FreeBSD__ 58#undef dump_write 59#define dump_write dmu_dump_write 60#endif 61 62/* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */ 63int zfs_send_corrupt_data = B_FALSE; 64 65static char *dmu_recv_tag = "dmu_recv_tag"; 66static const char *recv_clone_name = "%recv"; 67 68static int 69dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) 70{ 71 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset; 72 struct uio auio; 73 struct iovec aiov; 74 ASSERT0(len % 8); 75 76 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc); 77 aiov.iov_base = buf; 78 aiov.iov_len = len; 79 auio.uio_iov = &aiov; 80 auio.uio_iovcnt = 1; 81 auio.uio_resid = len; 82 auio.uio_segflg = UIO_SYSSPACE; 83 auio.uio_rw = UIO_WRITE; 84 auio.uio_offset = (off_t)-1; 85 auio.uio_td = dsp->dsa_td; 86#ifdef _KERNEL 87 if (dsp->dsa_fp->f_type == DTYPE_VNODE) 88 bwillwrite(); 89 dsp->dsa_err = fo_write(dsp->dsa_fp, &auio, dsp->dsa_td->td_ucred, 0, 90 dsp->dsa_td); 91#else 92 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__); 93 dsp->dsa_err = EOPNOTSUPP; 94#endif 95 mutex_enter(&ds->ds_sendstream_lock); 96 *dsp->dsa_off += len; 97 mutex_exit(&ds->ds_sendstream_lock); 98 99 return (dsp->dsa_err); 100} 101 102static int 103dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 104 uint64_t length) 105{ 106 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); 107 108 /* 109 * When we receive a free record, dbuf_free_range() assumes 110 * that the receiving system doesn't have any dbufs in the range 111 * being freed. This is always true because there is a one-record 112 * constraint: we only send one WRITE record for any given 113 * object+offset. We know that the one-record constraint is 114 * true because we always send data in increasing order by 115 * object,offset. 116 * 117 * If the increasing-order constraint ever changes, we should find 118 * another way to assert that the one-record constraint is still 119 * satisfied. 120 */ 121 ASSERT(object > dsp->dsa_last_data_object || 122 (object == dsp->dsa_last_data_object && 123 offset > dsp->dsa_last_data_offset)); 124 125 /* 126 * If we are doing a non-incremental send, then there can't 127 * be any data in the dataset we're receiving into. Therefore 128 * a free record would simply be a no-op. Save space by not 129 * sending it to begin with. 130 */ 131 if (!dsp->dsa_incremental) 132 return (0); 133 134 if (length != -1ULL && offset + length < offset) 135 length = -1ULL; 136 137 /* 138 * If there is a pending op, but it's not PENDING_FREE, push it out, 139 * since free block aggregation can only be done for blocks of the 140 * same type (i.e., DRR_FREE records can only be aggregated with 141 * other DRR_FREE records. DRR_FREEOBJECTS records can only be 142 * aggregated with other DRR_FREEOBJECTS records. 143 */ 144 if (dsp->dsa_pending_op != PENDING_NONE && 145 dsp->dsa_pending_op != PENDING_FREE) { 146 if (dump_bytes(dsp, dsp->dsa_drr, 147 sizeof (dmu_replay_record_t)) != 0) 148 return (SET_ERROR(EINTR)); 149 dsp->dsa_pending_op = PENDING_NONE; 150 } 151 152 if (dsp->dsa_pending_op == PENDING_FREE) { 153 /* 154 * There should never be a PENDING_FREE if length is -1 155 * (because dump_dnode is the only place where this 156 * function is called with a -1, and only after flushing 157 * any pending record). 158 */ 159 ASSERT(length != -1ULL); 160 /* 161 * Check to see whether this free block can be aggregated 162 * with pending one. 163 */ 164 if (drrf->drr_object == object && drrf->drr_offset + 165 drrf->drr_length == offset) { 166 drrf->drr_length += length; 167 return (0); 168 } else { 169 /* not a continuation. Push out pending record */ 170 if (dump_bytes(dsp, dsp->dsa_drr, 171 sizeof (dmu_replay_record_t)) != 0) 172 return (SET_ERROR(EINTR)); 173 dsp->dsa_pending_op = PENDING_NONE; 174 } 175 } 176 /* create a FREE record and make it pending */ 177 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 178 dsp->dsa_drr->drr_type = DRR_FREE; 179 drrf->drr_object = object; 180 drrf->drr_offset = offset; 181 drrf->drr_length = length; 182 drrf->drr_toguid = dsp->dsa_toguid; 183 if (length == -1ULL) { 184 if (dump_bytes(dsp, dsp->dsa_drr, 185 sizeof (dmu_replay_record_t)) != 0) 186 return (SET_ERROR(EINTR)); 187 } else { 188 dsp->dsa_pending_op = PENDING_FREE; 189 } 190 191 return (0); 192} 193 194static int 195dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, 196 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data) 197{ 198 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write); 199 200 /* 201 * We send data in increasing object, offset order. 202 * See comment in dump_free() for details. 203 */ 204 ASSERT(object > dsp->dsa_last_data_object || 205 (object == dsp->dsa_last_data_object && 206 offset > dsp->dsa_last_data_offset)); 207 dsp->dsa_last_data_object = object; 208 dsp->dsa_last_data_offset = offset + blksz - 1; 209 210 /* 211 * If there is any kind of pending aggregation (currently either 212 * a grouping of free objects or free blocks), push it out to 213 * the stream, since aggregation can't be done across operations 214 * of different types. 215 */ 216 if (dsp->dsa_pending_op != PENDING_NONE) { 217 if (dump_bytes(dsp, dsp->dsa_drr, 218 sizeof (dmu_replay_record_t)) != 0) 219 return (SET_ERROR(EINTR)); 220 dsp->dsa_pending_op = PENDING_NONE; 221 } 222 /* write a DATA record */ 223 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 224 dsp->dsa_drr->drr_type = DRR_WRITE; 225 drrw->drr_object = object; 226 drrw->drr_type = type; 227 drrw->drr_offset = offset; 228 drrw->drr_length = blksz; 229 drrw->drr_toguid = dsp->dsa_toguid; 230 if (bp == NULL || BP_IS_EMBEDDED(bp)) { 231 /* 232 * There's no pre-computed checksum for partial-block 233 * writes or embedded BP's, so (like 234 * fletcher4-checkummed blocks) userland will have to 235 * compute a dedup-capable checksum itself. 236 */ 237 drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; 238 } else { 239 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); 240 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup) 241 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP; 242 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); 243 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); 244 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); 245 drrw->drr_key.ddk_cksum = bp->blk_cksum; 246 } 247 248 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) 249 return (SET_ERROR(EINTR)); 250 if (dump_bytes(dsp, data, blksz) != 0) 251 return (SET_ERROR(EINTR)); 252 return (0); 253} 254 255static int 256dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 257 int blksz, const blkptr_t *bp) 258{ 259 char buf[BPE_PAYLOAD_SIZE]; 260 struct drr_write_embedded *drrw = 261 &(dsp->dsa_drr->drr_u.drr_write_embedded); 262 263 if (dsp->dsa_pending_op != PENDING_NONE) { 264 if (dump_bytes(dsp, dsp->dsa_drr, 265 sizeof (dmu_replay_record_t)) != 0) 266 return (EINTR); 267 dsp->dsa_pending_op = PENDING_NONE; 268 } 269 270 ASSERT(BP_IS_EMBEDDED(bp)); 271 272 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 273 dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED; 274 drrw->drr_object = object; 275 drrw->drr_offset = offset; 276 drrw->drr_length = blksz; 277 drrw->drr_toguid = dsp->dsa_toguid; 278 drrw->drr_compression = BP_GET_COMPRESS(bp); 279 drrw->drr_etype = BPE_GET_ETYPE(bp); 280 drrw->drr_lsize = BPE_GET_LSIZE(bp); 281 drrw->drr_psize = BPE_GET_PSIZE(bp); 282 283 decode_embedded_bp_compressed(bp, buf); 284 285 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) 286 return (EINTR); 287 if (dump_bytes(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0) 288 return (EINTR); 289 return (0); 290} 291 292static int 293dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data) 294{ 295 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); 296 297 if (dsp->dsa_pending_op != PENDING_NONE) { 298 if (dump_bytes(dsp, dsp->dsa_drr, 299 sizeof (dmu_replay_record_t)) != 0) 300 return (SET_ERROR(EINTR)); 301 dsp->dsa_pending_op = PENDING_NONE; 302 } 303 304 /* write a SPILL record */ 305 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 306 dsp->dsa_drr->drr_type = DRR_SPILL; 307 drrs->drr_object = object; 308 drrs->drr_length = blksz; 309 drrs->drr_toguid = dsp->dsa_toguid; 310 311 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t))) 312 return (SET_ERROR(EINTR)); 313 if (dump_bytes(dsp, data, blksz)) 314 return (SET_ERROR(EINTR)); 315 return (0); 316} 317 318static int 319dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) 320{ 321 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); 322 323 /* See comment in dump_free(). */ 324 if (!dsp->dsa_incremental) 325 return (0); 326 327 /* 328 * If there is a pending op, but it's not PENDING_FREEOBJECTS, 329 * push it out, since free block aggregation can only be done for 330 * blocks of the same type (i.e., DRR_FREE records can only be 331 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records 332 * can only be aggregated with other DRR_FREEOBJECTS records. 333 */ 334 if (dsp->dsa_pending_op != PENDING_NONE && 335 dsp->dsa_pending_op != PENDING_FREEOBJECTS) { 336 if (dump_bytes(dsp, dsp->dsa_drr, 337 sizeof (dmu_replay_record_t)) != 0) 338 return (SET_ERROR(EINTR)); 339 dsp->dsa_pending_op = PENDING_NONE; 340 } 341 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { 342 /* 343 * See whether this free object array can be aggregated 344 * with pending one 345 */ 346 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { 347 drrfo->drr_numobjs += numobjs; 348 return (0); 349 } else { 350 /* can't be aggregated. Push out pending record */ 351 if (dump_bytes(dsp, dsp->dsa_drr, 352 sizeof (dmu_replay_record_t)) != 0) 353 return (SET_ERROR(EINTR)); 354 dsp->dsa_pending_op = PENDING_NONE; 355 } 356 } 357 358 /* write a FREEOBJECTS record */ 359 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 360 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; 361 drrfo->drr_firstobj = firstobj; 362 drrfo->drr_numobjs = numobjs; 363 drrfo->drr_toguid = dsp->dsa_toguid; 364 365 dsp->dsa_pending_op = PENDING_FREEOBJECTS; 366 367 return (0); 368} 369 370static int 371dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) 372{ 373 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); 374 375 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) 376 return (dump_freeobjects(dsp, object, 1)); 377 378 if (dsp->dsa_pending_op != PENDING_NONE) { 379 if (dump_bytes(dsp, dsp->dsa_drr, 380 sizeof (dmu_replay_record_t)) != 0) 381 return (SET_ERROR(EINTR)); 382 dsp->dsa_pending_op = PENDING_NONE; 383 } 384 385 /* write an OBJECT record */ 386 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 387 dsp->dsa_drr->drr_type = DRR_OBJECT; 388 drro->drr_object = object; 389 drro->drr_type = dnp->dn_type; 390 drro->drr_bonustype = dnp->dn_bonustype; 391 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 392 drro->drr_bonuslen = dnp->dn_bonuslen; 393 drro->drr_checksumtype = dnp->dn_checksum; 394 drro->drr_compress = dnp->dn_compress; 395 drro->drr_toguid = dsp->dsa_toguid; 396 397 if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 398 drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) 399 drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; 400 401 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) 402 return (SET_ERROR(EINTR)); 403 404 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0) 405 return (SET_ERROR(EINTR)); 406 407 /* Free anything past the end of the file. */ 408 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * 409 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0) 410 return (SET_ERROR(EINTR)); 411 if (dsp->dsa_err != 0) 412 return (SET_ERROR(EINTR)); 413 return (0); 414} 415 416static boolean_t 417backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp) 418{ 419 if (!BP_IS_EMBEDDED(bp)) 420 return (B_FALSE); 421 422 /* 423 * Compression function must be legacy, or explicitly enabled. 424 */ 425 if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS && 426 !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4))) 427 return (B_FALSE); 428 429 /* 430 * Embed type must be explicitly enabled. 431 */ 432 switch (BPE_GET_ETYPE(bp)) { 433 case BP_EMBEDDED_TYPE_DATA: 434 if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) 435 return (B_TRUE); 436 break; 437 default: 438 return (B_FALSE); 439 } 440 return (B_FALSE); 441} 442 443#define BP_SPAN(dnp, level) \ 444 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \ 445 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) 446 447/* ARGSUSED */ 448static int 449backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 450 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 451{ 452 dmu_sendarg_t *dsp = arg; 453 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; 454 int err = 0; 455 456 if (issig(JUSTLOOKING) && issig(FORREAL)) 457 return (SET_ERROR(EINTR)); 458 459 if (zb->zb_object != DMU_META_DNODE_OBJECT && 460 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { 461 return (0); 462 } else if (zb->zb_level == ZB_ZIL_LEVEL) { 463 /* 464 * If we are sending a non-snapshot (which is allowed on 465 * read-only pools), it may have a ZIL, which must be ignored. 466 */ 467 return (0); 468 } else if (BP_IS_HOLE(bp) && 469 zb->zb_object == DMU_META_DNODE_OBJECT) { 470 uint64_t span = BP_SPAN(dnp, zb->zb_level); 471 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; 472 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT); 473 } else if (BP_IS_HOLE(bp)) { 474 uint64_t span = BP_SPAN(dnp, zb->zb_level); 475 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span); 476 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { 477 return (0); 478 } else if (type == DMU_OT_DNODE) { 479 dnode_phys_t *blk; 480 int i; 481 int blksz = BP_GET_LSIZE(bp); 482 arc_flags_t aflags = ARC_FLAG_WAIT; 483 arc_buf_t *abuf; 484 485 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 486 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 487 &aflags, zb) != 0) 488 return (SET_ERROR(EIO)); 489 490 blk = abuf->b_data; 491 for (i = 0; i < blksz >> DNODE_SHIFT; i++) { 492 uint64_t dnobj = (zb->zb_blkid << 493 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i; 494 err = dump_dnode(dsp, dnobj, blk+i); 495 if (err != 0) 496 break; 497 } 498 (void) arc_buf_remove_ref(abuf, &abuf); 499 } else if (type == DMU_OT_SA) { 500 arc_flags_t aflags = ARC_FLAG_WAIT; 501 arc_buf_t *abuf; 502 int blksz = BP_GET_LSIZE(bp); 503 504 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 505 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 506 &aflags, zb) != 0) 507 return (SET_ERROR(EIO)); 508 509 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data); 510 (void) arc_buf_remove_ref(abuf, &abuf); 511 } else if (backup_do_embed(dsp, bp)) { 512 /* it's an embedded level-0 block of a regular object */ 513 int blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 514 err = dump_write_embedded(dsp, zb->zb_object, 515 zb->zb_blkid * blksz, blksz, bp); 516 } else { /* it's a level-0 block of a regular object */ 517 arc_flags_t aflags = ARC_FLAG_WAIT; 518 arc_buf_t *abuf; 519 int blksz = BP_GET_LSIZE(bp); 520 uint64_t offset; 521 522 ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 523 ASSERT0(zb->zb_level); 524 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 525 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 526 &aflags, zb) != 0) { 527 if (zfs_send_corrupt_data) { 528 /* Send a block filled with 0x"zfs badd bloc" */ 529 abuf = arc_buf_alloc(spa, blksz, &abuf, 530 ARC_BUFC_DATA); 531 uint64_t *ptr; 532 for (ptr = abuf->b_data; 533 (char *)ptr < (char *)abuf->b_data + blksz; 534 ptr++) 535 *ptr = 0x2f5baddb10c; 536 } else { 537 return (SET_ERROR(EIO)); 538 } 539 } 540 541 offset = zb->zb_blkid * blksz; 542 543 if (!(dsp->dsa_featureflags & 544 DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 545 blksz > SPA_OLD_MAXBLOCKSIZE) { 546 char *buf = abuf->b_data; 547 while (blksz > 0 && err == 0) { 548 int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); 549 err = dump_write(dsp, type, zb->zb_object, 550 offset, n, NULL, buf); 551 offset += n; 552 buf += n; 553 blksz -= n; 554 } 555 } else { 556 err = dump_write(dsp, type, zb->zb_object, 557 offset, blksz, bp, abuf->b_data); 558 } 559 (void) arc_buf_remove_ref(abuf, &abuf); 560 } 561 562 ASSERT(err == 0 || err == EINTR); 563 return (err); 564} 565 566/* 567 * Releases dp using the specified tag. 568 */ 569static int 570dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds, 571 zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok, 572#ifdef illumos 573 boolean_t large_block_ok, int outfd, vnode_t *vp, offset_t *off) 574#else 575 boolean_t large_block_ok, int outfd, struct file *fp, offset_t *off) 576#endif 577{ 578 objset_t *os; 579 dmu_replay_record_t *drr; 580 dmu_sendarg_t *dsp; 581 int err; 582 uint64_t fromtxg = 0; 583 uint64_t featureflags = 0; 584 585 err = dmu_objset_from_ds(ds, &os); 586 if (err != 0) { 587 dsl_pool_rele(dp, tag); 588 return (err); 589 } 590 591 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 592 drr->drr_type = DRR_BEGIN; 593 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 594 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, 595 DMU_SUBSTREAM); 596 597#ifdef _KERNEL 598 if (dmu_objset_type(os) == DMU_OST_ZFS) { 599 uint64_t version; 600 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { 601 kmem_free(drr, sizeof (dmu_replay_record_t)); 602 dsl_pool_rele(dp, tag); 603 return (SET_ERROR(EINVAL)); 604 } 605 if (version >= ZPL_VERSION_SA) { 606 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; 607 } 608 } 609#endif 610 611 if (large_block_ok && ds->ds_large_blocks) 612 featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; 613 if (embedok && 614 spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { 615 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; 616 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 617 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4; 618 } else { 619 embedok = B_FALSE; 620 } 621 622 DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo, 623 featureflags); 624 625 drr->drr_u.drr_begin.drr_creation_time = 626 dsl_dataset_phys(ds)->ds_creation_time; 627 drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); 628 if (is_clone) 629 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; 630 drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(ds)->ds_guid; 631 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 632 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; 633 634 if (fromzb != NULL) { 635 drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid; 636 fromtxg = fromzb->zbm_creation_txg; 637 } 638 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname); 639 if (!dsl_dataset_is_snapshot(ds)) { 640 (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--", 641 sizeof (drr->drr_u.drr_begin.drr_toname)); 642 } 643 644 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); 645 646 dsp->dsa_drr = drr; 647 dsp->dsa_outfd = outfd; 648 dsp->dsa_proc = curproc; 649 dsp->dsa_td = curthread; 650 dsp->dsa_fp = fp; 651 dsp->dsa_os = os; 652 dsp->dsa_off = off; 653 dsp->dsa_toguid = dsl_dataset_phys(ds)->ds_guid; 654 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0); 655 dsp->dsa_pending_op = PENDING_NONE; 656 dsp->dsa_incremental = (fromzb != NULL); 657 dsp->dsa_featureflags = featureflags; 658 659 mutex_enter(&ds->ds_sendstream_lock); 660 list_insert_head(&ds->ds_sendstreams, dsp); 661 mutex_exit(&ds->ds_sendstream_lock); 662 663 dsl_dataset_long_hold(ds, FTAG); 664 dsl_pool_rele(dp, tag); 665 666 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) { 667 err = dsp->dsa_err; 668 goto out; 669 } 670 671 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH, 672 backup_cb, dsp); 673 674 if (dsp->dsa_pending_op != PENDING_NONE) 675 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) 676 err = SET_ERROR(EINTR); 677 678 if (err != 0) { 679 if (err == EINTR && dsp->dsa_err != 0) 680 err = dsp->dsa_err; 681 goto out; 682 } 683 684 bzero(drr, sizeof (dmu_replay_record_t)); 685 drr->drr_type = DRR_END; 686 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; 687 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; 688 689 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) { 690 err = dsp->dsa_err; 691 goto out; 692 } 693 694out: 695 mutex_enter(&ds->ds_sendstream_lock); 696 list_remove(&ds->ds_sendstreams, dsp); 697 mutex_exit(&ds->ds_sendstream_lock); 698 699 kmem_free(drr, sizeof (dmu_replay_record_t)); 700 kmem_free(dsp, sizeof (dmu_sendarg_t)); 701 702 dsl_dataset_long_rele(ds, FTAG); 703 704 return (err); 705} 706 707int 708dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, 709 boolean_t embedok, boolean_t large_block_ok, 710#ifdef illumos 711 int outfd, vnode_t *vp, offset_t *off) 712#else 713 int outfd, struct file *fp, offset_t *off) 714#endif 715{ 716 dsl_pool_t *dp; 717 dsl_dataset_t *ds; 718 dsl_dataset_t *fromds = NULL; 719 int err; 720 721 err = dsl_pool_hold(pool, FTAG, &dp); 722 if (err != 0) 723 return (err); 724 725 err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds); 726 if (err != 0) { 727 dsl_pool_rele(dp, FTAG); 728 return (err); 729 } 730 731 if (fromsnap != 0) { 732 zfs_bookmark_phys_t zb; 733 boolean_t is_clone; 734 735 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); 736 if (err != 0) { 737 dsl_dataset_rele(ds, FTAG); 738 dsl_pool_rele(dp, FTAG); 739 return (err); 740 } 741 if (!dsl_dataset_is_before(ds, fromds, 0)) 742 err = SET_ERROR(EXDEV); 743 zb.zbm_creation_time = 744 dsl_dataset_phys(fromds)->ds_creation_time; 745 zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; 746 zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 747 is_clone = (fromds->ds_dir != ds->ds_dir); 748 dsl_dataset_rele(fromds, FTAG); 749 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 750 embedok, large_block_ok, outfd, fp, off); 751 } else { 752 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 753 embedok, large_block_ok, outfd, fp, off); 754 } 755 dsl_dataset_rele(ds, FTAG); 756 return (err); 757} 758 759int 760dmu_send(const char *tosnap, const char *fromsnap, 761 boolean_t embedok, boolean_t large_block_ok, 762#ifdef illumos 763 int outfd, vnode_t *vp, offset_t *off) 764#else 765 int outfd, struct file *fp, offset_t *off) 766#endif 767{ 768 dsl_pool_t *dp; 769 dsl_dataset_t *ds; 770 int err; 771 boolean_t owned = B_FALSE; 772 773 if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL) 774 return (SET_ERROR(EINVAL)); 775 776 err = dsl_pool_hold(tosnap, FTAG, &dp); 777 if (err != 0) 778 return (err); 779 780 if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) { 781 /* 782 * We are sending a filesystem or volume. Ensure 783 * that it doesn't change by owning the dataset. 784 */ 785 err = dsl_dataset_own(dp, tosnap, FTAG, &ds); 786 owned = B_TRUE; 787 } else { 788 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds); 789 } 790 if (err != 0) { 791 dsl_pool_rele(dp, FTAG); 792 return (err); 793 } 794 795 if (fromsnap != NULL) { 796 zfs_bookmark_phys_t zb; 797 boolean_t is_clone = B_FALSE; 798 int fsnamelen = strchr(tosnap, '@') - tosnap; 799 800 /* 801 * If the fromsnap is in a different filesystem, then 802 * mark the send stream as a clone. 803 */ 804 if (strncmp(tosnap, fromsnap, fsnamelen) != 0 || 805 (fromsnap[fsnamelen] != '@' && 806 fromsnap[fsnamelen] != '#')) { 807 is_clone = B_TRUE; 808 } 809 810 if (strchr(fromsnap, '@')) { 811 dsl_dataset_t *fromds; 812 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); 813 if (err == 0) { 814 if (!dsl_dataset_is_before(ds, fromds, 0)) 815 err = SET_ERROR(EXDEV); 816 zb.zbm_creation_time = 817 dsl_dataset_phys(fromds)->ds_creation_time; 818 zb.zbm_creation_txg = 819 dsl_dataset_phys(fromds)->ds_creation_txg; 820 zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 821 is_clone = (ds->ds_dir != fromds->ds_dir); 822 dsl_dataset_rele(fromds, FTAG); 823 } 824 } else { 825 err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb); 826 } 827 if (err != 0) { 828 dsl_dataset_rele(ds, FTAG); 829 dsl_pool_rele(dp, FTAG); 830 return (err); 831 } 832 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 833 embedok, large_block_ok, outfd, fp, off); 834 } else { 835 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 836 embedok, large_block_ok, outfd, fp, off); 837 } 838 if (owned) 839 dsl_dataset_disown(ds, FTAG); 840 else 841 dsl_dataset_rele(ds, FTAG); 842 return (err); 843} 844 845int 846dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep) 847{ 848 dsl_pool_t *dp = ds->ds_dir->dd_pool; 849 int err; 850 uint64_t size; 851 852 ASSERT(dsl_pool_config_held(dp)); 853 854 /* tosnap must be a snapshot */ 855 if (!dsl_dataset_is_snapshot(ds)) 856 return (SET_ERROR(EINVAL)); 857 858 /* 859 * fromsnap must be an earlier snapshot from the same fs as tosnap, 860 * or the origin's fs. 861 */ 862 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0)) 863 return (SET_ERROR(EXDEV)); 864 865 /* Get uncompressed size estimate of changed data. */ 866 if (fromds == NULL) { 867 size = dsl_dataset_phys(ds)->ds_uncompressed_bytes; 868 } else { 869 uint64_t used, comp; 870 err = dsl_dataset_space_written(fromds, ds, 871 &used, &comp, &size); 872 if (err != 0) 873 return (err); 874 } 875 876 /* 877 * Assume that space (both on-disk and in-stream) is dominated by 878 * data. We will adjust for indirect blocks and the copies property, 879 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). 880 */ 881 882 /* 883 * Subtract out approximate space used by indirect blocks. 884 * Assume most space is used by data blocks (non-indirect, non-dnode). 885 * Assume all blocks are recordsize. Assume ditto blocks and 886 * internal fragmentation counter out compression. 887 * 888 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per 889 * block, which we observe in practice. 890 */ 891 uint64_t recordsize; 892 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize); 893 if (err != 0) 894 return (err); 895 size -= size / recordsize * sizeof (blkptr_t); 896 897 /* Add in the space for the record associated with each block. */ 898 size += size / recordsize * sizeof (dmu_replay_record_t); 899 900 *sizep = size; 901 902 return (0); 903} 904 905typedef struct dmu_recv_begin_arg { 906 const char *drba_origin; 907 dmu_recv_cookie_t *drba_cookie; 908 cred_t *drba_cred; 909 uint64_t drba_snapobj; 910} dmu_recv_begin_arg_t; 911 912static int 913recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, 914 uint64_t fromguid) 915{ 916 uint64_t val; 917 int error; 918 dsl_pool_t *dp = ds->ds_dir->dd_pool; 919 920 /* temporary clone name must not exist */ 921 error = zap_lookup(dp->dp_meta_objset, 922 dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, 923 8, 1, &val); 924 if (error != ENOENT) 925 return (error == 0 ? EBUSY : error); 926 927 /* new snapshot name must not exist */ 928 error = zap_lookup(dp->dp_meta_objset, 929 dsl_dataset_phys(ds)->ds_snapnames_zapobj, 930 drba->drba_cookie->drc_tosnap, 8, 1, &val); 931 if (error != ENOENT) 932 return (error == 0 ? EEXIST : error); 933 934 /* 935 * Check snapshot limit before receiving. We'll recheck again at the 936 * end, but might as well abort before receiving if we're already over 937 * the limit. 938 * 939 * Note that we do not check the file system limit with 940 * dsl_dir_fscount_check because the temporary %clones don't count 941 * against that limit. 942 */ 943 error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, 944 NULL, drba->drba_cred); 945 if (error != 0) 946 return (error); 947 948 if (fromguid != 0) { 949 dsl_dataset_t *snap; 950 uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 951 952 /* Find snapshot in this dir that matches fromguid. */ 953 while (obj != 0) { 954 error = dsl_dataset_hold_obj(dp, obj, FTAG, 955 &snap); 956 if (error != 0) 957 return (SET_ERROR(ENODEV)); 958 if (snap->ds_dir != ds->ds_dir) { 959 dsl_dataset_rele(snap, FTAG); 960 return (SET_ERROR(ENODEV)); 961 } 962 if (dsl_dataset_phys(snap)->ds_guid == fromguid) 963 break; 964 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 965 dsl_dataset_rele(snap, FTAG); 966 } 967 if (obj == 0) 968 return (SET_ERROR(ENODEV)); 969 970 if (drba->drba_cookie->drc_force) { 971 drba->drba_snapobj = obj; 972 } else { 973 /* 974 * If we are not forcing, there must be no 975 * changes since fromsnap. 976 */ 977 if (dsl_dataset_modified_since_snap(ds, snap)) { 978 dsl_dataset_rele(snap, FTAG); 979 return (SET_ERROR(ETXTBSY)); 980 } 981 drba->drba_snapobj = ds->ds_prev->ds_object; 982 } 983 984 dsl_dataset_rele(snap, FTAG); 985 } else { 986 /* if full, most recent snapshot must be $ORIGIN */ 987 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= TXG_INITIAL) 988 return (SET_ERROR(ENODEV)); 989 drba->drba_snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 990 } 991 992 return (0); 993 994} 995 996static int 997dmu_recv_begin_check(void *arg, dmu_tx_t *tx) 998{ 999 dmu_recv_begin_arg_t *drba = arg; 1000 dsl_pool_t *dp = dmu_tx_pool(tx); 1001 struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 1002 uint64_t fromguid = drrb->drr_fromguid; 1003 int flags = drrb->drr_flags; 1004 int error; 1005 uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 1006 dsl_dataset_t *ds; 1007 const char *tofs = drba->drba_cookie->drc_tofs; 1008 1009 /* already checked */ 1010 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 1011 1012 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 1013 DMU_COMPOUNDSTREAM || 1014 drrb->drr_type >= DMU_OST_NUMTYPES || 1015 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) 1016 return (SET_ERROR(EINVAL)); 1017 1018 /* Verify pool version supports SA if SA_SPILL feature set */ 1019 if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && 1020 spa_version(dp->dp_spa) < SPA_VERSION_SA) 1021 return (SET_ERROR(ENOTSUP)); 1022 1023 /* 1024 * The receiving code doesn't know how to translate a WRITE_EMBEDDED 1025 * record to a plan WRITE record, so the pool must have the 1026 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED 1027 * records. Same with WRITE_EMBEDDED records that use LZ4 compression. 1028 */ 1029 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && 1030 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) 1031 return (SET_ERROR(ENOTSUP)); 1032 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) && 1033 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 1034 return (SET_ERROR(ENOTSUP)); 1035 1036 /* 1037 * The receiving code doesn't know how to translate large blocks 1038 * to smaller ones, so the pool must have the LARGE_BLOCKS 1039 * feature enabled if the stream has LARGE_BLOCKS. 1040 */ 1041 if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 1042 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 1043 return (SET_ERROR(ENOTSUP)); 1044 1045 error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 1046 if (error == 0) { 1047 /* target fs already exists; recv into temp clone */ 1048 1049 /* Can't recv a clone into an existing fs */ 1050 if (flags & DRR_FLAG_CLONE) { 1051 dsl_dataset_rele(ds, FTAG); 1052 return (SET_ERROR(EINVAL)); 1053 } 1054 1055 error = recv_begin_check_existing_impl(drba, ds, fromguid); 1056 dsl_dataset_rele(ds, FTAG); 1057 } else if (error == ENOENT) { 1058 /* target fs does not exist; must be a full backup or clone */ 1059 char buf[MAXNAMELEN]; 1060 1061 /* 1062 * If it's a non-clone incremental, we are missing the 1063 * target fs, so fail the recv. 1064 */ 1065 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE)) 1066 return (SET_ERROR(ENOENT)); 1067 1068 /* Open the parent of tofs */ 1069 ASSERT3U(strlen(tofs), <, MAXNAMELEN); 1070 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); 1071 error = dsl_dataset_hold(dp, buf, FTAG, &ds); 1072 if (error != 0) 1073 return (error); 1074 1075 /* 1076 * Check filesystem and snapshot limits before receiving. We'll 1077 * recheck snapshot limits again at the end (we create the 1078 * filesystems and increment those counts during begin_sync). 1079 */ 1080 error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 1081 ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); 1082 if (error != 0) { 1083 dsl_dataset_rele(ds, FTAG); 1084 return (error); 1085 } 1086 1087 error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 1088 ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); 1089 if (error != 0) { 1090 dsl_dataset_rele(ds, FTAG); 1091 return (error); 1092 } 1093 1094 if (drba->drba_origin != NULL) { 1095 dsl_dataset_t *origin; 1096 error = dsl_dataset_hold(dp, drba->drba_origin, 1097 FTAG, &origin); 1098 if (error != 0) { 1099 dsl_dataset_rele(ds, FTAG); 1100 return (error); 1101 } 1102 if (!dsl_dataset_is_snapshot(origin)) { 1103 dsl_dataset_rele(origin, FTAG); 1104 dsl_dataset_rele(ds, FTAG); 1105 return (SET_ERROR(EINVAL)); 1106 } 1107 if (dsl_dataset_phys(origin)->ds_guid != fromguid) { 1108 dsl_dataset_rele(origin, FTAG); 1109 dsl_dataset_rele(ds, FTAG); 1110 return (SET_ERROR(ENODEV)); 1111 } 1112 dsl_dataset_rele(origin, FTAG); 1113 } 1114 dsl_dataset_rele(ds, FTAG); 1115 error = 0; 1116 } 1117 return (error); 1118} 1119 1120static void 1121dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) 1122{ 1123 dmu_recv_begin_arg_t *drba = arg; 1124 dsl_pool_t *dp = dmu_tx_pool(tx); 1125 struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 1126 const char *tofs = drba->drba_cookie->drc_tofs; 1127 dsl_dataset_t *ds, *newds; 1128 uint64_t dsobj; 1129 int error; 1130 uint64_t crflags; 1131 1132 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ? 1133 DS_FLAG_CI_DATASET : 0; 1134 1135 error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 1136 if (error == 0) { 1137 /* create temporary clone */ 1138 dsl_dataset_t *snap = NULL; 1139 if (drba->drba_snapobj != 0) { 1140 VERIFY0(dsl_dataset_hold_obj(dp, 1141 drba->drba_snapobj, FTAG, &snap)); 1142 } 1143 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, 1144 snap, crflags, drba->drba_cred, tx); 1145 dsl_dataset_rele(snap, FTAG); 1146 dsl_dataset_rele(ds, FTAG); 1147 } else { 1148 dsl_dir_t *dd; 1149 const char *tail; 1150 dsl_dataset_t *origin = NULL; 1151 1152 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); 1153 1154 if (drba->drba_origin != NULL) { 1155 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, 1156 FTAG, &origin)); 1157 } 1158 1159 /* Create new dataset. */ 1160 dsobj = dsl_dataset_create_sync(dd, 1161 strrchr(tofs, '/') + 1, 1162 origin, crflags, drba->drba_cred, tx); 1163 if (origin != NULL) 1164 dsl_dataset_rele(origin, FTAG); 1165 dsl_dir_rele(dd, FTAG); 1166 drba->drba_cookie->drc_newfs = B_TRUE; 1167 } 1168 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); 1169 1170 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 1171 DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 1172 !newds->ds_large_blocks) { 1173 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 1174 newds->ds_large_blocks = B_TRUE; 1175 } 1176 1177 dmu_buf_will_dirty(newds->ds_dbuf, tx); 1178 dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; 1179 1180 /* 1181 * If we actually created a non-clone, we need to create the 1182 * objset in our new dataset. 1183 */ 1184 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) { 1185 (void) dmu_objset_create_impl(dp->dp_spa, 1186 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); 1187 } 1188 1189 drba->drba_cookie->drc_ds = newds; 1190 1191 spa_history_log_internal_ds(newds, "receive", tx, ""); 1192} 1193 1194/* 1195 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() 1196 * succeeds; otherwise we will leak the holds on the datasets. 1197 */ 1198int 1199dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb, 1200 boolean_t force, char *origin, dmu_recv_cookie_t *drc) 1201{ 1202 dmu_recv_begin_arg_t drba = { 0 }; 1203 dmu_replay_record_t *drr; 1204 1205 bzero(drc, sizeof (dmu_recv_cookie_t)); 1206 drc->drc_drrb = drrb; 1207 drc->drc_tosnap = tosnap; 1208 drc->drc_tofs = tofs; 1209 drc->drc_force = force; 1210 drc->drc_cred = CRED(); 1211 1212 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) 1213 drc->drc_byteswap = B_TRUE; 1214 else if (drrb->drr_magic != DMU_BACKUP_MAGIC) 1215 return (SET_ERROR(EINVAL)); 1216 1217 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 1218 drr->drr_type = DRR_BEGIN; 1219 drr->drr_u.drr_begin = *drc->drc_drrb; 1220 if (drc->drc_byteswap) { 1221 fletcher_4_incremental_byteswap(drr, 1222 sizeof (dmu_replay_record_t), &drc->drc_cksum); 1223 } else { 1224 fletcher_4_incremental_native(drr, 1225 sizeof (dmu_replay_record_t), &drc->drc_cksum); 1226 } 1227 kmem_free(drr, sizeof (dmu_replay_record_t)); 1228 1229 if (drc->drc_byteswap) { 1230 drrb->drr_magic = BSWAP_64(drrb->drr_magic); 1231 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); 1232 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); 1233 drrb->drr_type = BSWAP_32(drrb->drr_type); 1234 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); 1235 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); 1236 } 1237 1238 drba.drba_origin = origin; 1239 drba.drba_cookie = drc; 1240 drba.drba_cred = CRED(); 1241 1242 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync, 1243 &drba, 5, ZFS_SPACE_CHECK_NORMAL)); 1244} 1245 1246struct restorearg { 1247 int err; 1248 boolean_t byteswap; 1249 kthread_t *td; 1250 struct file *fp; 1251 char *buf; 1252 uint64_t voff; 1253 int bufsize; /* amount of memory allocated for buf */ 1254 zio_cksum_t cksum; 1255 avl_tree_t *guid_to_ds_map; 1256}; 1257 1258typedef struct guid_map_entry { 1259 uint64_t guid; 1260 dsl_dataset_t *gme_ds; 1261 avl_node_t avlnode; 1262} guid_map_entry_t; 1263 1264static int 1265guid_compare(const void *arg1, const void *arg2) 1266{ 1267 const guid_map_entry_t *gmep1 = arg1; 1268 const guid_map_entry_t *gmep2 = arg2; 1269 1270 if (gmep1->guid < gmep2->guid) 1271 return (-1); 1272 else if (gmep1->guid > gmep2->guid) 1273 return (1); 1274 return (0); 1275} 1276 1277static void 1278free_guid_map_onexit(void *arg) 1279{ 1280 avl_tree_t *ca = arg; 1281 void *cookie = NULL; 1282 guid_map_entry_t *gmep; 1283 1284 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { 1285 dsl_dataset_long_rele(gmep->gme_ds, gmep); 1286 dsl_dataset_rele(gmep->gme_ds, gmep); 1287 kmem_free(gmep, sizeof (guid_map_entry_t)); 1288 } 1289 avl_destroy(ca); 1290 kmem_free(ca, sizeof (avl_tree_t)); 1291} 1292 1293static int 1294restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid) 1295{ 1296 struct uio auio; 1297 struct iovec aiov; 1298 int error; 1299 1300 aiov.iov_base = buf; 1301 aiov.iov_len = len; 1302 auio.uio_iov = &aiov; 1303 auio.uio_iovcnt = 1; 1304 auio.uio_resid = len; 1305 auio.uio_segflg = UIO_SYSSPACE; 1306 auio.uio_rw = UIO_READ; 1307 auio.uio_offset = off; 1308 auio.uio_td = ra->td; 1309#ifdef _KERNEL 1310 error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td); 1311#else 1312 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__); 1313 error = EOPNOTSUPP; 1314#endif 1315 *resid = auio.uio_resid; 1316 return (error); 1317} 1318 1319static void * 1320restore_read(struct restorearg *ra, int len, char *buf) 1321{ 1322 int done = 0; 1323 1324 if (buf == NULL) 1325 buf = ra->buf; 1326 1327 /* some things will require 8-byte alignment, so everything must */ 1328 ASSERT0(len % 8); 1329 ASSERT3U(len, <=, ra->bufsize); 1330 1331 while (done < len) { 1332 ssize_t resid; 1333 1334 ra->err = restore_bytes(ra, buf + done, 1335 len - done, ra->voff, &resid); 1336 1337 if (resid == len - done) 1338 ra->err = SET_ERROR(EINVAL); 1339 ra->voff += len - done - resid; 1340 done = len - resid; 1341 if (ra->err != 0) 1342 return (NULL); 1343 } 1344 1345 ASSERT3U(done, ==, len); 1346 if (ra->byteswap) 1347 fletcher_4_incremental_byteswap(buf, len, &ra->cksum); 1348 else 1349 fletcher_4_incremental_native(buf, len, &ra->cksum); 1350 return (buf); 1351} 1352 1353static void 1354backup_byteswap(dmu_replay_record_t *drr) 1355{ 1356#define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) 1357#define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) 1358 drr->drr_type = BSWAP_32(drr->drr_type); 1359 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); 1360 switch (drr->drr_type) { 1361 case DRR_BEGIN: 1362 DO64(drr_begin.drr_magic); 1363 DO64(drr_begin.drr_versioninfo); 1364 DO64(drr_begin.drr_creation_time); 1365 DO32(drr_begin.drr_type); 1366 DO32(drr_begin.drr_flags); 1367 DO64(drr_begin.drr_toguid); 1368 DO64(drr_begin.drr_fromguid); 1369 break; 1370 case DRR_OBJECT: 1371 DO64(drr_object.drr_object); 1372 DO32(drr_object.drr_type); 1373 DO32(drr_object.drr_bonustype); 1374 DO32(drr_object.drr_blksz); 1375 DO32(drr_object.drr_bonuslen); 1376 DO64(drr_object.drr_toguid); 1377 break; 1378 case DRR_FREEOBJECTS: 1379 DO64(drr_freeobjects.drr_firstobj); 1380 DO64(drr_freeobjects.drr_numobjs); 1381 DO64(drr_freeobjects.drr_toguid); 1382 break; 1383 case DRR_WRITE: 1384 DO64(drr_write.drr_object); 1385 DO32(drr_write.drr_type); 1386 DO64(drr_write.drr_offset); 1387 DO64(drr_write.drr_length); 1388 DO64(drr_write.drr_toguid); 1389 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]); 1390 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]); 1391 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]); 1392 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]); 1393 DO64(drr_write.drr_key.ddk_prop); 1394 break; 1395 case DRR_WRITE_BYREF: 1396 DO64(drr_write_byref.drr_object); 1397 DO64(drr_write_byref.drr_offset); 1398 DO64(drr_write_byref.drr_length); 1399 DO64(drr_write_byref.drr_toguid); 1400 DO64(drr_write_byref.drr_refguid); 1401 DO64(drr_write_byref.drr_refobject); 1402 DO64(drr_write_byref.drr_refoffset); 1403 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]); 1404 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]); 1405 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]); 1406 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]); 1407 DO64(drr_write_byref.drr_key.ddk_prop); 1408 break; 1409 case DRR_WRITE_EMBEDDED: 1410 DO64(drr_write_embedded.drr_object); 1411 DO64(drr_write_embedded.drr_offset); 1412 DO64(drr_write_embedded.drr_length); 1413 DO64(drr_write_embedded.drr_toguid); 1414 DO32(drr_write_embedded.drr_lsize); 1415 DO32(drr_write_embedded.drr_psize); 1416 break; 1417 case DRR_FREE: 1418 DO64(drr_free.drr_object); 1419 DO64(drr_free.drr_offset); 1420 DO64(drr_free.drr_length); 1421 DO64(drr_free.drr_toguid); 1422 break; 1423 case DRR_SPILL: 1424 DO64(drr_spill.drr_object); 1425 DO64(drr_spill.drr_length); 1426 DO64(drr_spill.drr_toguid); 1427 break; 1428 case DRR_END: 1429 DO64(drr_end.drr_checksum.zc_word[0]); 1430 DO64(drr_end.drr_checksum.zc_word[1]); 1431 DO64(drr_end.drr_checksum.zc_word[2]); 1432 DO64(drr_end.drr_checksum.zc_word[3]); 1433 DO64(drr_end.drr_toguid); 1434 break; 1435 } 1436#undef DO64 1437#undef DO32 1438} 1439 1440static inline uint8_t 1441deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) 1442{ 1443 if (bonus_type == DMU_OT_SA) { 1444 return (1); 1445 } else { 1446 return (1 + 1447 ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT)); 1448 } 1449} 1450 1451static int 1452restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) 1453{ 1454 dmu_object_info_t doi; 1455 dmu_tx_t *tx; 1456 void *data = NULL; 1457 uint64_t object; 1458 int err; 1459 1460 if (drro->drr_type == DMU_OT_NONE || 1461 !DMU_OT_IS_VALID(drro->drr_type) || 1462 !DMU_OT_IS_VALID(drro->drr_bonustype) || 1463 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || 1464 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || 1465 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || 1466 drro->drr_blksz < SPA_MINBLOCKSIZE || 1467 drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(os)) || 1468 drro->drr_bonuslen > DN_MAX_BONUSLEN) { 1469 return (SET_ERROR(EINVAL)); 1470 } 1471 1472 err = dmu_object_info(os, drro->drr_object, &doi); 1473 1474 if (err != 0 && err != ENOENT) 1475 return (SET_ERROR(EINVAL)); 1476 object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT; 1477 1478 if (drro->drr_bonuslen) { 1479 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8), NULL); 1480 if (ra->err != 0) 1481 return (ra->err); 1482 } 1483 1484 /* 1485 * If we are losing blkptrs or changing the block size this must 1486 * be a new file instance. We must clear out the previous file 1487 * contents before we can change this type of metadata in the dnode. 1488 */ 1489 if (err == 0) { 1490 int nblkptr; 1491 1492 nblkptr = deduce_nblkptr(drro->drr_bonustype, 1493 drro->drr_bonuslen); 1494 1495 if (drro->drr_blksz != doi.doi_data_block_size || 1496 nblkptr < doi.doi_nblkptr) { 1497 err = dmu_free_long_range(os, drro->drr_object, 1498 0, DMU_OBJECT_END); 1499 if (err != 0) 1500 return (SET_ERROR(EINVAL)); 1501 } 1502 } 1503 1504 tx = dmu_tx_create(os); 1505 dmu_tx_hold_bonus(tx, object); 1506 err = dmu_tx_assign(tx, TXG_WAIT); 1507 if (err != 0) { 1508 dmu_tx_abort(tx); 1509 return (err); 1510 } 1511 1512 if (object == DMU_NEW_OBJECT) { 1513 /* currently free, want to be allocated */ 1514 err = dmu_object_claim(os, drro->drr_object, 1515 drro->drr_type, drro->drr_blksz, 1516 drro->drr_bonustype, drro->drr_bonuslen, tx); 1517 } else if (drro->drr_type != doi.doi_type || 1518 drro->drr_blksz != doi.doi_data_block_size || 1519 drro->drr_bonustype != doi.doi_bonus_type || 1520 drro->drr_bonuslen != doi.doi_bonus_size) { 1521 /* currently allocated, but with different properties */ 1522 err = dmu_object_reclaim(os, drro->drr_object, 1523 drro->drr_type, drro->drr_blksz, 1524 drro->drr_bonustype, drro->drr_bonuslen, tx); 1525 } 1526 if (err != 0) { 1527 dmu_tx_commit(tx); 1528 return (SET_ERROR(EINVAL)); 1529 } 1530 1531 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype, 1532 tx); 1533 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx); 1534 1535 if (data != NULL) { 1536 dmu_buf_t *db; 1537 1538 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db)); 1539 dmu_buf_will_dirty(db, tx); 1540 1541 ASSERT3U(db->db_size, >=, drro->drr_bonuslen); 1542 bcopy(data, db->db_data, drro->drr_bonuslen); 1543 if (ra->byteswap) { 1544 dmu_object_byteswap_t byteswap = 1545 DMU_OT_BYTESWAP(drro->drr_bonustype); 1546 dmu_ot_byteswap[byteswap].ob_func(db->db_data, 1547 drro->drr_bonuslen); 1548 } 1549 dmu_buf_rele(db, FTAG); 1550 } 1551 dmu_tx_commit(tx); 1552 return (0); 1553} 1554 1555/* ARGSUSED */ 1556static int 1557restore_freeobjects(struct restorearg *ra, objset_t *os, 1558 struct drr_freeobjects *drrfo) 1559{ 1560 uint64_t obj; 1561 1562 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) 1563 return (SET_ERROR(EINVAL)); 1564 1565 for (obj = drrfo->drr_firstobj; 1566 obj < drrfo->drr_firstobj + drrfo->drr_numobjs; 1567 (void) dmu_object_next(os, &obj, FALSE, 0)) { 1568 int err; 1569 1570 if (dmu_object_info(os, obj, NULL) != 0) 1571 continue; 1572 1573 err = dmu_free_long_object(os, obj); 1574 if (err != 0) 1575 return (err); 1576 } 1577 return (0); 1578} 1579 1580static int 1581restore_write(struct restorearg *ra, objset_t *os, 1582 struct drr_write *drrw) 1583{ 1584 dmu_tx_t *tx; 1585 void *data; 1586 int err; 1587 1588 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset || 1589 !DMU_OT_IS_VALID(drrw->drr_type)) 1590 return (SET_ERROR(EINVAL)); 1591 1592 if (dmu_object_info(os, drrw->drr_object, NULL) != 0) 1593 return (SET_ERROR(EINVAL)); 1594 1595 dmu_buf_t *bonus; 1596 if (dmu_bonus_hold(os, drrw->drr_object, FTAG, &bonus) != 0) 1597 return (SET_ERROR(EINVAL)); 1598 1599 arc_buf_t *abuf = dmu_request_arcbuf(bonus, drrw->drr_length); 1600 1601 data = restore_read(ra, drrw->drr_length, abuf->b_data); 1602 if (data == NULL) { 1603 dmu_return_arcbuf(abuf); 1604 dmu_buf_rele(bonus, FTAG); 1605 return (ra->err); 1606 } 1607 1608 tx = dmu_tx_create(os); 1609 1610 dmu_tx_hold_write(tx, drrw->drr_object, 1611 drrw->drr_offset, drrw->drr_length); 1612 err = dmu_tx_assign(tx, TXG_WAIT); 1613 if (err != 0) { 1614 dmu_return_arcbuf(abuf); 1615 dmu_buf_rele(bonus, FTAG); 1616 dmu_tx_abort(tx); 1617 return (err); 1618 } 1619 if (ra->byteswap) { 1620 dmu_object_byteswap_t byteswap = 1621 DMU_OT_BYTESWAP(drrw->drr_type); 1622 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length); 1623 } 1624 dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx); 1625 dmu_tx_commit(tx); 1626 dmu_buf_rele(bonus, FTAG); 1627 return (0); 1628} 1629 1630/* 1631 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed 1632 * streams to refer to a copy of the data that is already on the 1633 * system because it came in earlier in the stream. This function 1634 * finds the earlier copy of the data, and uses that copy instead of 1635 * data from the stream to fulfill this write. 1636 */ 1637static int 1638restore_write_byref(struct restorearg *ra, objset_t *os, 1639 struct drr_write_byref *drrwbr) 1640{ 1641 dmu_tx_t *tx; 1642 int err; 1643 guid_map_entry_t gmesrch; 1644 guid_map_entry_t *gmep; 1645 avl_index_t where; 1646 objset_t *ref_os = NULL; 1647 dmu_buf_t *dbp; 1648 1649 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) 1650 return (SET_ERROR(EINVAL)); 1651 1652 /* 1653 * If the GUID of the referenced dataset is different from the 1654 * GUID of the target dataset, find the referenced dataset. 1655 */ 1656 if (drrwbr->drr_toguid != drrwbr->drr_refguid) { 1657 gmesrch.guid = drrwbr->drr_refguid; 1658 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch, 1659 &where)) == NULL) { 1660 return (SET_ERROR(EINVAL)); 1661 } 1662 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) 1663 return (SET_ERROR(EINVAL)); 1664 } else { 1665 ref_os = os; 1666 } 1667 1668 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, 1669 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH); 1670 if (err != 0) 1671 return (err); 1672 1673 tx = dmu_tx_create(os); 1674 1675 dmu_tx_hold_write(tx, drrwbr->drr_object, 1676 drrwbr->drr_offset, drrwbr->drr_length); 1677 err = dmu_tx_assign(tx, TXG_WAIT); 1678 if (err != 0) { 1679 dmu_tx_abort(tx); 1680 return (err); 1681 } 1682 dmu_write(os, drrwbr->drr_object, 1683 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); 1684 dmu_buf_rele(dbp, FTAG); 1685 dmu_tx_commit(tx); 1686 return (0); 1687} 1688 1689static int 1690restore_write_embedded(struct restorearg *ra, objset_t *os, 1691 struct drr_write_embedded *drrwnp) 1692{ 1693 dmu_tx_t *tx; 1694 int err; 1695 void *data; 1696 1697 if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset) 1698 return (EINVAL); 1699 1700 if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE) 1701 return (EINVAL); 1702 1703 if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES) 1704 return (EINVAL); 1705 if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS) 1706 return (EINVAL); 1707 1708 data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8), NULL); 1709 if (data == NULL) 1710 return (ra->err); 1711 1712 tx = dmu_tx_create(os); 1713 1714 dmu_tx_hold_write(tx, drrwnp->drr_object, 1715 drrwnp->drr_offset, drrwnp->drr_length); 1716 err = dmu_tx_assign(tx, TXG_WAIT); 1717 if (err != 0) { 1718 dmu_tx_abort(tx); 1719 return (err); 1720 } 1721 1722 dmu_write_embedded(os, drrwnp->drr_object, 1723 drrwnp->drr_offset, data, drrwnp->drr_etype, 1724 drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize, 1725 ra->byteswap ^ ZFS_HOST_BYTEORDER, tx); 1726 1727 dmu_tx_commit(tx); 1728 return (0); 1729} 1730 1731static int 1732restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs) 1733{ 1734 dmu_tx_t *tx; 1735 void *data; 1736 dmu_buf_t *db, *db_spill; 1737 int err; 1738 1739 if (drrs->drr_length < SPA_MINBLOCKSIZE || 1740 drrs->drr_length > spa_maxblocksize(dmu_objset_spa(os))) 1741 return (SET_ERROR(EINVAL)); 1742 1743 data = restore_read(ra, drrs->drr_length, NULL); 1744 if (data == NULL) 1745 return (ra->err); 1746 1747 if (dmu_object_info(os, drrs->drr_object, NULL) != 0) 1748 return (SET_ERROR(EINVAL)); 1749 1750 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db)); 1751 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) { 1752 dmu_buf_rele(db, FTAG); 1753 return (err); 1754 } 1755 1756 tx = dmu_tx_create(os); 1757 1758 dmu_tx_hold_spill(tx, db->db_object); 1759 1760 err = dmu_tx_assign(tx, TXG_WAIT); 1761 if (err != 0) { 1762 dmu_buf_rele(db, FTAG); 1763 dmu_buf_rele(db_spill, FTAG); 1764 dmu_tx_abort(tx); 1765 return (err); 1766 } 1767 dmu_buf_will_dirty(db_spill, tx); 1768 1769 if (db_spill->db_size < drrs->drr_length) 1770 VERIFY(0 == dbuf_spill_set_blksz(db_spill, 1771 drrs->drr_length, tx)); 1772 bcopy(data, db_spill->db_data, drrs->drr_length); 1773 1774 dmu_buf_rele(db, FTAG); 1775 dmu_buf_rele(db_spill, FTAG); 1776 1777 dmu_tx_commit(tx); 1778 return (0); 1779} 1780 1781/* ARGSUSED */ 1782static int 1783restore_free(struct restorearg *ra, objset_t *os, 1784 struct drr_free *drrf) 1785{ 1786 int err; 1787 1788 if (drrf->drr_length != -1ULL && 1789 drrf->drr_offset + drrf->drr_length < drrf->drr_offset) 1790 return (SET_ERROR(EINVAL)); 1791 1792 if (dmu_object_info(os, drrf->drr_object, NULL) != 0) 1793 return (SET_ERROR(EINVAL)); 1794 1795 err = dmu_free_long_range(os, drrf->drr_object, 1796 drrf->drr_offset, drrf->drr_length); 1797 return (err); 1798} 1799 1800/* used to destroy the drc_ds on error */ 1801static void 1802dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) 1803{ 1804 char name[MAXNAMELEN]; 1805 dsl_dataset_name(drc->drc_ds, name); 1806 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 1807 (void) dsl_destroy_head(name); 1808} 1809 1810/* 1811 * NB: callers *must* call dmu_recv_end() if this succeeds. 1812 */ 1813int 1814dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp, 1815 int cleanup_fd, uint64_t *action_handlep) 1816{ 1817 struct restorearg ra = { 0 }; 1818 dmu_replay_record_t *drr; 1819 objset_t *os; 1820 zio_cksum_t pcksum; 1821 int featureflags; 1822 1823 ra.byteswap = drc->drc_byteswap; 1824 ra.cksum = drc->drc_cksum; 1825 ra.td = curthread; 1826 ra.fp = fp; 1827 ra.voff = *voffp; 1828 ra.bufsize = SPA_MAXBLOCKSIZE; 1829 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP); 1830 1831 /* these were verified in dmu_recv_begin */ 1832 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, 1833 DMU_SUBSTREAM); 1834 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); 1835 1836 /* 1837 * Open the objset we are modifying. 1838 */ 1839 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os)); 1840 1841 ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); 1842 1843 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); 1844 1845 /* if this stream is dedup'ed, set up the avl tree for guid mapping */ 1846 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { 1847 minor_t minor; 1848 1849 if (cleanup_fd == -1) { 1850 ra.err = SET_ERROR(EBADF); 1851 goto out; 1852 } 1853 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor); 1854 if (ra.err != 0) { 1855 cleanup_fd = -1; 1856 goto out; 1857 } 1858 1859 if (*action_handlep == 0) { 1860 ra.guid_to_ds_map = 1861 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); 1862 avl_create(ra.guid_to_ds_map, guid_compare, 1863 sizeof (guid_map_entry_t), 1864 offsetof(guid_map_entry_t, avlnode)); 1865 ra.err = zfs_onexit_add_cb(minor, 1866 free_guid_map_onexit, ra.guid_to_ds_map, 1867 action_handlep); 1868 if (ra.err != 0) 1869 goto out; 1870 } else { 1871 ra.err = zfs_onexit_cb_data(minor, *action_handlep, 1872 (void **)&ra.guid_to_ds_map); 1873 if (ra.err != 0) 1874 goto out; 1875 } 1876 1877 drc->drc_guid_to_ds_map = ra.guid_to_ds_map; 1878 } 1879 1880 /* 1881 * Read records and process them. 1882 */ 1883 pcksum = ra.cksum; 1884 while (ra.err == 0 && 1885 NULL != (drr = restore_read(&ra, sizeof (*drr), NULL))) { 1886 if (issig(JUSTLOOKING) && issig(FORREAL)) { 1887 ra.err = SET_ERROR(EINTR); 1888 goto out; 1889 } 1890 1891 if (ra.byteswap) 1892 backup_byteswap(drr); 1893 1894 switch (drr->drr_type) { 1895 case DRR_OBJECT: 1896 { 1897 /* 1898 * We need to make a copy of the record header, 1899 * because restore_{object,write} may need to 1900 * restore_read(), which will invalidate drr. 1901 */ 1902 struct drr_object drro = drr->drr_u.drr_object; 1903 ra.err = restore_object(&ra, os, &drro); 1904 break; 1905 } 1906 case DRR_FREEOBJECTS: 1907 { 1908 struct drr_freeobjects drrfo = 1909 drr->drr_u.drr_freeobjects; 1910 ra.err = restore_freeobjects(&ra, os, &drrfo); 1911 break; 1912 } 1913 case DRR_WRITE: 1914 { 1915 struct drr_write drrw = drr->drr_u.drr_write; 1916 ra.err = restore_write(&ra, os, &drrw); 1917 break; 1918 } 1919 case DRR_WRITE_BYREF: 1920 { 1921 struct drr_write_byref drrwbr = 1922 drr->drr_u.drr_write_byref; 1923 ra.err = restore_write_byref(&ra, os, &drrwbr); 1924 break; 1925 } 1926 case DRR_WRITE_EMBEDDED: 1927 { 1928 struct drr_write_embedded drrwe = 1929 drr->drr_u.drr_write_embedded; 1930 ra.err = restore_write_embedded(&ra, os, &drrwe); 1931 break; 1932 } 1933 case DRR_FREE: 1934 { 1935 struct drr_free drrf = drr->drr_u.drr_free; 1936 ra.err = restore_free(&ra, os, &drrf); 1937 break; 1938 } 1939 case DRR_END: 1940 { 1941 struct drr_end drre = drr->drr_u.drr_end; 1942 /* 1943 * We compare against the *previous* checksum 1944 * value, because the stored checksum is of 1945 * everything before the DRR_END record. 1946 */ 1947 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum)) 1948 ra.err = SET_ERROR(ECKSUM); 1949 goto out; 1950 } 1951 case DRR_SPILL: 1952 { 1953 struct drr_spill drrs = drr->drr_u.drr_spill; 1954 ra.err = restore_spill(&ra, os, &drrs); 1955 break; 1956 } 1957 default: 1958 ra.err = SET_ERROR(EINVAL); 1959 goto out; 1960 } 1961 pcksum = ra.cksum; 1962 } 1963 ASSERT(ra.err != 0); 1964 1965out: 1966 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) 1967 zfs_onexit_fd_rele(cleanup_fd); 1968 1969 if (ra.err != 0) { 1970 /* 1971 * destroy what we created, so we don't leave it in the 1972 * inconsistent restoring state. 1973 */ 1974 dmu_recv_cleanup_ds(drc); 1975 } 1976 1977 kmem_free(ra.buf, ra.bufsize); 1978 *voffp = ra.voff; 1979 return (ra.err); 1980} 1981 1982static int 1983dmu_recv_end_check(void *arg, dmu_tx_t *tx) 1984{ 1985 dmu_recv_cookie_t *drc = arg; 1986 dsl_pool_t *dp = dmu_tx_pool(tx); 1987 int error; 1988 1989 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); 1990 1991 if (!drc->drc_newfs) { 1992 dsl_dataset_t *origin_head; 1993 1994 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); 1995 if (error != 0) 1996 return (error); 1997 if (drc->drc_force) { 1998 /* 1999 * We will destroy any snapshots in tofs (i.e. before 2000 * origin_head) that are after the origin (which is 2001 * the snap before drc_ds, because drc_ds can not 2002 * have any snaps of its own). 2003 */ 2004 uint64_t obj; 2005 2006 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 2007 while (obj != 2008 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 2009 dsl_dataset_t *snap; 2010 error = dsl_dataset_hold_obj(dp, obj, FTAG, 2011 &snap); 2012 if (error != 0) 2013 return (error); 2014 if (snap->ds_dir != origin_head->ds_dir) 2015 error = SET_ERROR(EINVAL); 2016 if (error == 0) { 2017 error = dsl_destroy_snapshot_check_impl( 2018 snap, B_FALSE); 2019 } 2020 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 2021 dsl_dataset_rele(snap, FTAG); 2022 if (error != 0) 2023 return (error); 2024 } 2025 } 2026 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, 2027 origin_head, drc->drc_force, drc->drc_owner, tx); 2028 if (error != 0) { 2029 dsl_dataset_rele(origin_head, FTAG); 2030 return (error); 2031 } 2032 error = dsl_dataset_snapshot_check_impl(origin_head, 2033 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 2034 dsl_dataset_rele(origin_head, FTAG); 2035 if (error != 0) 2036 return (error); 2037 2038 error = dsl_destroy_head_check_impl(drc->drc_ds, 1); 2039 } else { 2040 error = dsl_dataset_snapshot_check_impl(drc->drc_ds, 2041 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 2042 } 2043 return (error); 2044} 2045 2046static void 2047dmu_recv_end_sync(void *arg, dmu_tx_t *tx) 2048{ 2049 dmu_recv_cookie_t *drc = arg; 2050 dsl_pool_t *dp = dmu_tx_pool(tx); 2051 2052 spa_history_log_internal_ds(drc->drc_ds, "finish receiving", 2053 tx, "snap=%s", drc->drc_tosnap); 2054 2055 if (!drc->drc_newfs) { 2056 dsl_dataset_t *origin_head; 2057 2058 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, 2059 &origin_head)); 2060 2061 if (drc->drc_force) { 2062 /* 2063 * Destroy any snapshots of drc_tofs (origin_head) 2064 * after the origin (the snap before drc_ds). 2065 */ 2066 uint64_t obj; 2067 2068 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 2069 while (obj != 2070 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 2071 dsl_dataset_t *snap; 2072 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, 2073 &snap)); 2074 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); 2075 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 2076 dsl_destroy_snapshot_sync_impl(snap, 2077 B_FALSE, tx); 2078 dsl_dataset_rele(snap, FTAG); 2079 } 2080 } 2081 VERIFY3P(drc->drc_ds->ds_prev, ==, 2082 origin_head->ds_prev); 2083 2084 dsl_dataset_clone_swap_sync_impl(drc->drc_ds, 2085 origin_head, tx); 2086 dsl_dataset_snapshot_sync_impl(origin_head, 2087 drc->drc_tosnap, tx); 2088 2089 /* set snapshot's creation time and guid */ 2090 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); 2091 dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = 2092 drc->drc_drrb->drr_creation_time; 2093 dsl_dataset_phys(origin_head->ds_prev)->ds_guid = 2094 drc->drc_drrb->drr_toguid; 2095 dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= 2096 ~DS_FLAG_INCONSISTENT; 2097 2098 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2099 dsl_dataset_phys(origin_head)->ds_flags &= 2100 ~DS_FLAG_INCONSISTENT; 2101 2102 dsl_dataset_rele(origin_head, FTAG); 2103 dsl_destroy_head_sync_impl(drc->drc_ds, tx); 2104 2105 if (drc->drc_owner != NULL) 2106 VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); 2107 } else { 2108 dsl_dataset_t *ds = drc->drc_ds; 2109 2110 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); 2111 2112 /* set snapshot's creation time and guid */ 2113 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 2114 dsl_dataset_phys(ds->ds_prev)->ds_creation_time = 2115 drc->drc_drrb->drr_creation_time; 2116 dsl_dataset_phys(ds->ds_prev)->ds_guid = 2117 drc->drc_drrb->drr_toguid; 2118 dsl_dataset_phys(ds->ds_prev)->ds_flags &= 2119 ~DS_FLAG_INCONSISTENT; 2120 2121 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2122 dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; 2123 } 2124 drc->drc_newsnapobj = dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; 2125 /* 2126 * Release the hold from dmu_recv_begin. This must be done before 2127 * we return to open context, so that when we free the dataset's dnode, 2128 * we can evict its bonus buffer. 2129 */ 2130 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 2131 drc->drc_ds = NULL; 2132} 2133 2134static int 2135add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj) 2136{ 2137 dsl_pool_t *dp; 2138 dsl_dataset_t *snapds; 2139 guid_map_entry_t *gmep; 2140 int err; 2141 2142 ASSERT(guid_map != NULL); 2143 2144 err = dsl_pool_hold(name, FTAG, &dp); 2145 if (err != 0) 2146 return (err); 2147 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); 2148 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds); 2149 if (err == 0) { 2150 gmep->guid = dsl_dataset_phys(snapds)->ds_guid; 2151 gmep->gme_ds = snapds; 2152 avl_add(guid_map, gmep); 2153 dsl_dataset_long_hold(snapds, gmep); 2154 } else 2155 kmem_free(gmep, sizeof (*gmep)); 2156 2157 dsl_pool_rele(dp, FTAG); 2158 return (err); 2159} 2160 2161static int dmu_recv_end_modified_blocks = 3; 2162 2163static int 2164dmu_recv_existing_end(dmu_recv_cookie_t *drc) 2165{ 2166 int error; 2167 char name[MAXNAMELEN]; 2168 2169#ifdef _KERNEL 2170 /* 2171 * We will be destroying the ds; make sure its origin is unmounted if 2172 * necessary. 2173 */ 2174 dsl_dataset_name(drc->drc_ds, name); 2175 zfs_destroy_unmount_origin(name); 2176#endif 2177 2178 error = dsl_sync_task(drc->drc_tofs, 2179 dmu_recv_end_check, dmu_recv_end_sync, drc, 2180 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL); 2181 2182 if (error != 0) 2183 dmu_recv_cleanup_ds(drc); 2184 return (error); 2185} 2186 2187static int 2188dmu_recv_new_end(dmu_recv_cookie_t *drc) 2189{ 2190 int error; 2191 2192 error = dsl_sync_task(drc->drc_tofs, 2193 dmu_recv_end_check, dmu_recv_end_sync, drc, 2194 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL); 2195 2196 if (error != 0) { 2197 dmu_recv_cleanup_ds(drc); 2198 } else if (drc->drc_guid_to_ds_map != NULL) { 2199 (void) add_ds_to_guidmap(drc->drc_tofs, 2200 drc->drc_guid_to_ds_map, 2201 drc->drc_newsnapobj); 2202 } 2203 return (error); 2204} 2205 2206int 2207dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) 2208{ 2209 drc->drc_owner = owner; 2210 2211 if (drc->drc_newfs) 2212 return (dmu_recv_new_end(drc)); 2213 else 2214 return (dmu_recv_existing_end(drc)); 2215} 2216 2217/* 2218 * Return TRUE if this objset is currently being received into. 2219 */ 2220boolean_t 2221dmu_objset_is_receiving(objset_t *os) 2222{ 2223 return (os->os_dsl_dataset != NULL && 2224 os->os_dsl_dataset->ds_owner == dmu_recv_tag); 2225} 2226