dmu_send.c revision 249042
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) 2012 by Delphix. All rights reserved. 25 * Copyright (c) 2012, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2012, Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 */ 28 29#include <sys/dmu.h> 30#include <sys/dmu_impl.h> 31#include <sys/dmu_tx.h> 32#include <sys/dbuf.h> 33#include <sys/dnode.h> 34#include <sys/zfs_context.h> 35#include <sys/dmu_objset.h> 36#include <sys/dmu_traverse.h> 37#include <sys/dsl_dataset.h> 38#include <sys/dsl_dir.h> 39#include <sys/dsl_prop.h> 40#include <sys/dsl_pool.h> 41#include <sys/dsl_synctask.h> 42#include <sys/zfs_ioctl.h> 43#include <sys/zap.h> 44#include <sys/zio_checksum.h> 45#include <sys/zfs_znode.h> 46#include <zfs_fletcher.h> 47#include <sys/avl.h> 48#include <sys/ddt.h> 49#include <sys/zfs_onexit.h> 50#include <sys/dmu_send.h> 51#include <sys/dsl_destroy.h> 52 53/* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */ 54int zfs_send_corrupt_data = B_FALSE; 55 56static char *dmu_recv_tag = "dmu_recv_tag"; 57static const char *recv_clone_name = "%recv"; 58 59static int 60dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) 61{ 62 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset; 63 struct uio auio; 64 struct iovec aiov; 65 ASSERT0(len % 8); 66 67 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc); 68 aiov.iov_base = buf; 69 aiov.iov_len = len; 70 auio.uio_iov = &aiov; 71 auio.uio_iovcnt = 1; 72 auio.uio_resid = len; 73 auio.uio_segflg = UIO_SYSSPACE; 74 auio.uio_rw = UIO_WRITE; 75 auio.uio_offset = (off_t)-1; 76 auio.uio_td = dsp->dsa_td; 77#ifdef _KERNEL 78 if (dsp->dsa_fp->f_type == DTYPE_VNODE) 79 bwillwrite(); 80 dsp->dsa_err = fo_write(dsp->dsa_fp, &auio, dsp->dsa_td->td_ucred, 0, 81 dsp->dsa_td); 82#else 83 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__); 84 dsp->dsa_err = EOPNOTSUPP; 85#endif 86 mutex_enter(&ds->ds_sendstream_lock); 87 *dsp->dsa_off += len; 88 mutex_exit(&ds->ds_sendstream_lock); 89 90 return (dsp->dsa_err); 91} 92 93static int 94dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 95 uint64_t length) 96{ 97 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); 98 99 if (length != -1ULL && offset + length < offset) 100 length = -1ULL; 101 102 /* 103 * If there is a pending op, but it's not PENDING_FREE, push it out, 104 * since free block aggregation can only be done for blocks of the 105 * same type (i.e., DRR_FREE records can only be aggregated with 106 * other DRR_FREE records. DRR_FREEOBJECTS records can only be 107 * aggregated with other DRR_FREEOBJECTS records. 108 */ 109 if (dsp->dsa_pending_op != PENDING_NONE && 110 dsp->dsa_pending_op != PENDING_FREE) { 111 if (dump_bytes(dsp, dsp->dsa_drr, 112 sizeof (dmu_replay_record_t)) != 0) 113 return (EINTR); 114 dsp->dsa_pending_op = PENDING_NONE; 115 } 116 117 if (dsp->dsa_pending_op == PENDING_FREE) { 118 /* 119 * There should never be a PENDING_FREE if length is -1 120 * (because dump_dnode is the only place where this 121 * function is called with a -1, and only after flushing 122 * any pending record). 123 */ 124 ASSERT(length != -1ULL); 125 /* 126 * Check to see whether this free block can be aggregated 127 * with pending one. 128 */ 129 if (drrf->drr_object == object && drrf->drr_offset + 130 drrf->drr_length == offset) { 131 drrf->drr_length += length; 132 return (0); 133 } else { 134 /* not a continuation. Push out pending record */ 135 if (dump_bytes(dsp, dsp->dsa_drr, 136 sizeof (dmu_replay_record_t)) != 0) 137 return (EINTR); 138 dsp->dsa_pending_op = PENDING_NONE; 139 } 140 } 141 /* create a FREE record and make it pending */ 142 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 143 dsp->dsa_drr->drr_type = DRR_FREE; 144 drrf->drr_object = object; 145 drrf->drr_offset = offset; 146 drrf->drr_length = length; 147 drrf->drr_toguid = dsp->dsa_toguid; 148 if (length == -1ULL) { 149 if (dump_bytes(dsp, dsp->dsa_drr, 150 sizeof (dmu_replay_record_t)) != 0) 151 return (EINTR); 152 } else { 153 dsp->dsa_pending_op = PENDING_FREE; 154 } 155 156 return (0); 157} 158 159static int 160dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type, 161 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data) 162{ 163 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write); 164 165 166 /* 167 * If there is any kind of pending aggregation (currently either 168 * a grouping of free objects or free blocks), push it out to 169 * the stream, since aggregation can't be done across operations 170 * of different types. 171 */ 172 if (dsp->dsa_pending_op != PENDING_NONE) { 173 if (dump_bytes(dsp, dsp->dsa_drr, 174 sizeof (dmu_replay_record_t)) != 0) 175 return (EINTR); 176 dsp->dsa_pending_op = PENDING_NONE; 177 } 178 /* write a DATA record */ 179 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 180 dsp->dsa_drr->drr_type = DRR_WRITE; 181 drrw->drr_object = object; 182 drrw->drr_type = type; 183 drrw->drr_offset = offset; 184 drrw->drr_length = blksz; 185 drrw->drr_toguid = dsp->dsa_toguid; 186 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); 187 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup) 188 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP; 189 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); 190 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); 191 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); 192 drrw->drr_key.ddk_cksum = bp->blk_cksum; 193 194 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) 195 return (EINTR); 196 if (dump_bytes(dsp, data, blksz) != 0) 197 return (EINTR); 198 return (0); 199} 200 201static int 202dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data) 203{ 204 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); 205 206 if (dsp->dsa_pending_op != PENDING_NONE) { 207 if (dump_bytes(dsp, dsp->dsa_drr, 208 sizeof (dmu_replay_record_t)) != 0) 209 return (EINTR); 210 dsp->dsa_pending_op = PENDING_NONE; 211 } 212 213 /* write a SPILL record */ 214 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 215 dsp->dsa_drr->drr_type = DRR_SPILL; 216 drrs->drr_object = object; 217 drrs->drr_length = blksz; 218 drrs->drr_toguid = dsp->dsa_toguid; 219 220 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t))) 221 return (EINTR); 222 if (dump_bytes(dsp, data, blksz)) 223 return (EINTR); 224 return (0); 225} 226 227static int 228dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) 229{ 230 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); 231 232 /* 233 * If there is a pending op, but it's not PENDING_FREEOBJECTS, 234 * push it out, since free block aggregation can only be done for 235 * blocks of the same type (i.e., DRR_FREE records can only be 236 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records 237 * can only be aggregated with other DRR_FREEOBJECTS records. 238 */ 239 if (dsp->dsa_pending_op != PENDING_NONE && 240 dsp->dsa_pending_op != PENDING_FREEOBJECTS) { 241 if (dump_bytes(dsp, dsp->dsa_drr, 242 sizeof (dmu_replay_record_t)) != 0) 243 return (EINTR); 244 dsp->dsa_pending_op = PENDING_NONE; 245 } 246 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { 247 /* 248 * See whether this free object array can be aggregated 249 * with pending one 250 */ 251 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { 252 drrfo->drr_numobjs += numobjs; 253 return (0); 254 } else { 255 /* can't be aggregated. Push out pending record */ 256 if (dump_bytes(dsp, dsp->dsa_drr, 257 sizeof (dmu_replay_record_t)) != 0) 258 return (EINTR); 259 dsp->dsa_pending_op = PENDING_NONE; 260 } 261 } 262 263 /* write a FREEOBJECTS record */ 264 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 265 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; 266 drrfo->drr_firstobj = firstobj; 267 drrfo->drr_numobjs = numobjs; 268 drrfo->drr_toguid = dsp->dsa_toguid; 269 270 dsp->dsa_pending_op = PENDING_FREEOBJECTS; 271 272 return (0); 273} 274 275static int 276dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) 277{ 278 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); 279 280 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) 281 return (dump_freeobjects(dsp, object, 1)); 282 283 if (dsp->dsa_pending_op != PENDING_NONE) { 284 if (dump_bytes(dsp, dsp->dsa_drr, 285 sizeof (dmu_replay_record_t)) != 0) 286 return (EINTR); 287 dsp->dsa_pending_op = PENDING_NONE; 288 } 289 290 /* write an OBJECT record */ 291 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 292 dsp->dsa_drr->drr_type = DRR_OBJECT; 293 drro->drr_object = object; 294 drro->drr_type = dnp->dn_type; 295 drro->drr_bonustype = dnp->dn_bonustype; 296 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 297 drro->drr_bonuslen = dnp->dn_bonuslen; 298 drro->drr_checksumtype = dnp->dn_checksum; 299 drro->drr_compress = dnp->dn_compress; 300 drro->drr_toguid = dsp->dsa_toguid; 301 302 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) 303 return (EINTR); 304 305 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0) 306 return (EINTR); 307 308 /* free anything past the end of the file */ 309 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * 310 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL)) 311 return (EINTR); 312 if (dsp->dsa_err != 0) 313 return (EINTR); 314 return (0); 315} 316 317#define BP_SPAN(dnp, level) \ 318 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \ 319 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) 320 321/* ARGSUSED */ 322static int 323backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 324 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 325{ 326 dmu_sendarg_t *dsp = arg; 327 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; 328 int err = 0; 329 330 if (issig(JUSTLOOKING) && issig(FORREAL)) 331 return (EINTR); 332 333 if (zb->zb_object != DMU_META_DNODE_OBJECT && 334 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { 335 return (0); 336 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) { 337 uint64_t span = BP_SPAN(dnp, zb->zb_level); 338 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; 339 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT); 340 } else if (bp == NULL) { 341 uint64_t span = BP_SPAN(dnp, zb->zb_level); 342 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span); 343 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { 344 return (0); 345 } else if (type == DMU_OT_DNODE) { 346 dnode_phys_t *blk; 347 int i; 348 int blksz = BP_GET_LSIZE(bp); 349 uint32_t aflags = ARC_WAIT; 350 arc_buf_t *abuf; 351 352 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 353 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 354 &aflags, zb) != 0) 355 return (EIO); 356 357 blk = abuf->b_data; 358 for (i = 0; i < blksz >> DNODE_SHIFT; i++) { 359 uint64_t dnobj = (zb->zb_blkid << 360 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i; 361 err = dump_dnode(dsp, dnobj, blk+i); 362 if (err != 0) 363 break; 364 } 365 (void) arc_buf_remove_ref(abuf, &abuf); 366 } else if (type == DMU_OT_SA) { 367 uint32_t aflags = ARC_WAIT; 368 arc_buf_t *abuf; 369 int blksz = BP_GET_LSIZE(bp); 370 371 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 372 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 373 &aflags, zb) != 0) 374 return (EIO); 375 376 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data); 377 (void) arc_buf_remove_ref(abuf, &abuf); 378 } else { /* it's a level-0 block of a regular object */ 379 uint32_t aflags = ARC_WAIT; 380 arc_buf_t *abuf; 381 int blksz = BP_GET_LSIZE(bp); 382 383 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 384 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 385 &aflags, zb) != 0) { 386 if (zfs_send_corrupt_data) { 387 /* Send a block filled with 0x"zfs badd bloc" */ 388 abuf = arc_buf_alloc(spa, blksz, &abuf, 389 ARC_BUFC_DATA); 390 uint64_t *ptr; 391 for (ptr = abuf->b_data; 392 (char *)ptr < (char *)abuf->b_data + blksz; 393 ptr++) 394 *ptr = 0x2f5baddb10c; 395 } else { 396 return (EIO); 397 } 398 } 399 400 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz, 401 blksz, bp, abuf->b_data); 402 (void) arc_buf_remove_ref(abuf, &abuf); 403 } 404 405 ASSERT(err == 0 || err == EINTR); 406 return (err); 407} 408 409/* 410 * Releases dp, ds, and fromds, using the specified tag. 411 */ 412static int 413dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds, 414#ifdef illumos 415 dsl_dataset_t *fromds, int outfd, vnode_t *vp, offset_t *off) 416#else 417 dsl_dataset_t *fromds, int outfd, struct file *fp, offset_t *off) 418#endif 419{ 420 objset_t *os; 421 dmu_replay_record_t *drr; 422 dmu_sendarg_t *dsp; 423 int err; 424 uint64_t fromtxg = 0; 425 426 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds)) { 427 dsl_dataset_rele(fromds, tag); 428 dsl_dataset_rele(ds, tag); 429 dsl_pool_rele(dp, tag); 430 return (EXDEV); 431 } 432 433 err = dmu_objset_from_ds(ds, &os); 434 if (err != 0) { 435 if (fromds != NULL) 436 dsl_dataset_rele(fromds, tag); 437 dsl_dataset_rele(ds, tag); 438 dsl_pool_rele(dp, tag); 439 return (err); 440 } 441 442 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 443 drr->drr_type = DRR_BEGIN; 444 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 445 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, 446 DMU_SUBSTREAM); 447 448#ifdef _KERNEL 449 if (dmu_objset_type(os) == DMU_OST_ZFS) { 450 uint64_t version; 451 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { 452 kmem_free(drr, sizeof (dmu_replay_record_t)); 453 if (fromds != NULL) 454 dsl_dataset_rele(fromds, tag); 455 dsl_dataset_rele(ds, tag); 456 dsl_pool_rele(dp, tag); 457 return (EINVAL); 458 } 459 if (version >= ZPL_VERSION_SA) { 460 DMU_SET_FEATUREFLAGS( 461 drr->drr_u.drr_begin.drr_versioninfo, 462 DMU_BACKUP_FEATURE_SA_SPILL); 463 } 464 } 465#endif 466 467 drr->drr_u.drr_begin.drr_creation_time = 468 ds->ds_phys->ds_creation_time; 469 drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); 470 if (fromds != NULL && ds->ds_dir != fromds->ds_dir) 471 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; 472 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid; 473 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 474 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; 475 476 if (fromds != NULL) 477 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid; 478 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname); 479 480 if (fromds != NULL) { 481 fromtxg = fromds->ds_phys->ds_creation_txg; 482 dsl_dataset_rele(fromds, tag); 483 fromds = NULL; 484 } 485 486 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); 487 488 dsp->dsa_drr = drr; 489 dsp->dsa_outfd = outfd; 490 dsp->dsa_proc = curproc; 491 dsp->dsa_td = curthread; 492 dsp->dsa_fp = fp; 493 dsp->dsa_os = os; 494 dsp->dsa_off = off; 495 dsp->dsa_toguid = ds->ds_phys->ds_guid; 496 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0); 497 dsp->dsa_pending_op = PENDING_NONE; 498 499 mutex_enter(&ds->ds_sendstream_lock); 500 list_insert_head(&ds->ds_sendstreams, dsp); 501 mutex_exit(&ds->ds_sendstream_lock); 502 503 dsl_dataset_long_hold(ds, FTAG); 504 dsl_pool_rele(dp, tag); 505 506 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) { 507 err = dsp->dsa_err; 508 goto out; 509 } 510 511 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH, 512 backup_cb, dsp); 513 514 if (dsp->dsa_pending_op != PENDING_NONE) 515 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) 516 err = EINTR; 517 518 if (err != 0) { 519 if (err == EINTR && dsp->dsa_err != 0) 520 err = dsp->dsa_err; 521 goto out; 522 } 523 524 bzero(drr, sizeof (dmu_replay_record_t)); 525 drr->drr_type = DRR_END; 526 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; 527 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; 528 529 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) { 530 err = dsp->dsa_err; 531 goto out; 532 } 533 534out: 535 mutex_enter(&ds->ds_sendstream_lock); 536 list_remove(&ds->ds_sendstreams, dsp); 537 mutex_exit(&ds->ds_sendstream_lock); 538 539 kmem_free(drr, sizeof (dmu_replay_record_t)); 540 kmem_free(dsp, sizeof (dmu_sendarg_t)); 541 542 dsl_dataset_long_rele(ds, FTAG); 543 dsl_dataset_rele(ds, tag); 544 545 return (err); 546} 547 548int 549dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, 550#ifdef illumos 551 int outfd, vnode_t *vp, offset_t *off) 552#else 553 int outfd, struct file *fp, offset_t *off) 554#endif 555{ 556 dsl_pool_t *dp; 557 dsl_dataset_t *ds; 558 dsl_dataset_t *fromds = NULL; 559 int err; 560 561 err = dsl_pool_hold(pool, FTAG, &dp); 562 if (err != 0) 563 return (err); 564 565 err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds); 566 if (err != 0) { 567 dsl_pool_rele(dp, FTAG); 568 return (err); 569 } 570 571 if (fromsnap != 0) { 572 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); 573 if (err != 0) { 574 dsl_dataset_rele(ds, FTAG); 575 dsl_pool_rele(dp, FTAG); 576 return (err); 577 } 578 } 579 580 return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, fp, off)); 581} 582 583int 584dmu_send(const char *tosnap, const char *fromsnap, 585#ifdef illumos 586 int outfd, vnode_t *vp, offset_t *off) 587#else 588 int outfd, struct file *fp, offset_t *off) 589#endif 590{ 591 dsl_pool_t *dp; 592 dsl_dataset_t *ds; 593 dsl_dataset_t *fromds = NULL; 594 int err; 595 596 if (strchr(tosnap, '@') == NULL) 597 return (EINVAL); 598 if (fromsnap != NULL && strchr(fromsnap, '@') == NULL) 599 return (EINVAL); 600 601 err = dsl_pool_hold(tosnap, FTAG, &dp); 602 if (err != 0) 603 return (err); 604 605 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds); 606 if (err != 0) { 607 dsl_pool_rele(dp, FTAG); 608 return (err); 609 } 610 611 if (fromsnap != NULL) { 612 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); 613 if (err != 0) { 614 dsl_dataset_rele(ds, FTAG); 615 dsl_pool_rele(dp, FTAG); 616 return (err); 617 } 618 } 619 return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, fp, off)); 620} 621 622int 623dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep) 624{ 625 dsl_pool_t *dp = ds->ds_dir->dd_pool; 626 int err; 627 uint64_t size; 628 629 ASSERT(dsl_pool_config_held(dp)); 630 631 /* tosnap must be a snapshot */ 632 if (!dsl_dataset_is_snapshot(ds)) 633 return (EINVAL); 634 635 /* 636 * fromsnap must be an earlier snapshot from the same fs as tosnap, 637 * or the origin's fs. 638 */ 639 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds)) 640 return (EXDEV); 641 642 /* Get uncompressed size estimate of changed data. */ 643 if (fromds == NULL) { 644 size = ds->ds_phys->ds_uncompressed_bytes; 645 } else { 646 uint64_t used, comp; 647 err = dsl_dataset_space_written(fromds, ds, 648 &used, &comp, &size); 649 if (err != 0) 650 return (err); 651 } 652 653 /* 654 * Assume that space (both on-disk and in-stream) is dominated by 655 * data. We will adjust for indirect blocks and the copies property, 656 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). 657 */ 658 659 /* 660 * Subtract out approximate space used by indirect blocks. 661 * Assume most space is used by data blocks (non-indirect, non-dnode). 662 * Assume all blocks are recordsize. Assume ditto blocks and 663 * internal fragmentation counter out compression. 664 * 665 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per 666 * block, which we observe in practice. 667 */ 668 uint64_t recordsize; 669 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize); 670 if (err != 0) 671 return (err); 672 size -= size / recordsize * sizeof (blkptr_t); 673 674 /* Add in the space for the record associated with each block. */ 675 size += size / recordsize * sizeof (dmu_replay_record_t); 676 677 *sizep = size; 678 679 return (0); 680} 681 682typedef struct dmu_recv_begin_arg { 683 const char *drba_origin; 684 dmu_recv_cookie_t *drba_cookie; 685 cred_t *drba_cred; 686} dmu_recv_begin_arg_t; 687 688static int 689recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, 690 uint64_t fromguid) 691{ 692 uint64_t val; 693 int error; 694 dsl_pool_t *dp = ds->ds_dir->dd_pool; 695 696 /* must not have any changes since most recent snapshot */ 697 if (!drba->drba_cookie->drc_force && 698 dsl_dataset_modified_since_lastsnap(ds)) 699 return (ETXTBSY); 700 701 /* temporary clone name must not exist */ 702 error = zap_lookup(dp->dp_meta_objset, 703 ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name, 704 8, 1, &val); 705 if (error != ENOENT) 706 return (error == 0 ? EBUSY : error); 707 708 /* new snapshot name must not exist */ 709 error = zap_lookup(dp->dp_meta_objset, 710 ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap, 711 8, 1, &val); 712 if (error != ENOENT) 713 return (error == 0 ? EEXIST : error); 714 715 if (fromguid != 0) { 716 /* if incremental, most recent snapshot must match fromguid */ 717 if (ds->ds_prev == NULL) 718 return (ENODEV); 719 720 /* 721 * most recent snapshot must match fromguid, or there are no 722 * changes since the fromguid one 723 */ 724 if (ds->ds_prev->ds_phys->ds_guid != fromguid) { 725 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth; 726 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj; 727 while (obj != 0) { 728 dsl_dataset_t *snap; 729 error = dsl_dataset_hold_obj(dp, obj, FTAG, 730 &snap); 731 if (error != 0) 732 return (ENODEV); 733 if (snap->ds_phys->ds_creation_txg < birth) { 734 dsl_dataset_rele(snap, FTAG); 735 return (ENODEV); 736 } 737 if (snap->ds_phys->ds_guid == fromguid) { 738 dsl_dataset_rele(snap, FTAG); 739 break; /* it's ok */ 740 } 741 obj = snap->ds_phys->ds_prev_snap_obj; 742 dsl_dataset_rele(snap, FTAG); 743 } 744 if (obj == 0) 745 return (ENODEV); 746 } 747 } else { 748 /* if full, most recent snapshot must be $ORIGIN */ 749 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL) 750 return (ENODEV); 751 } 752 753 return (0); 754 755} 756 757static int 758dmu_recv_begin_check(void *arg, dmu_tx_t *tx) 759{ 760 dmu_recv_begin_arg_t *drba = arg; 761 dsl_pool_t *dp = dmu_tx_pool(tx); 762 struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 763 uint64_t fromguid = drrb->drr_fromguid; 764 int flags = drrb->drr_flags; 765 int error; 766 dsl_dataset_t *ds; 767 const char *tofs = drba->drba_cookie->drc_tofs; 768 769 /* already checked */ 770 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 771 772 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 773 DMU_COMPOUNDSTREAM || 774 drrb->drr_type >= DMU_OST_NUMTYPES || 775 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) 776 return (EINVAL); 777 778 /* Verify pool version supports SA if SA_SPILL feature set */ 779 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 780 DMU_BACKUP_FEATURE_SA_SPILL) && 781 spa_version(dp->dp_spa) < SPA_VERSION_SA) { 782 return (ENOTSUP); 783 } 784 785 error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 786 if (error == 0) { 787 /* target fs already exists; recv into temp clone */ 788 789 /* Can't recv a clone into an existing fs */ 790 if (flags & DRR_FLAG_CLONE) { 791 dsl_dataset_rele(ds, FTAG); 792 return (EINVAL); 793 } 794 795 error = recv_begin_check_existing_impl(drba, ds, fromguid); 796 dsl_dataset_rele(ds, FTAG); 797 } else if (error == ENOENT) { 798 /* target fs does not exist; must be a full backup or clone */ 799 char buf[MAXNAMELEN]; 800 801 /* 802 * If it's a non-clone incremental, we are missing the 803 * target fs, so fail the recv. 804 */ 805 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE)) 806 return (ENOENT); 807 808 /* Open the parent of tofs */ 809 ASSERT3U(strlen(tofs), <, MAXNAMELEN); 810 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); 811 error = dsl_dataset_hold(dp, buf, FTAG, &ds); 812 if (error != 0) 813 return (error); 814 815 if (drba->drba_origin != NULL) { 816 dsl_dataset_t *origin; 817 error = dsl_dataset_hold(dp, drba->drba_origin, 818 FTAG, &origin); 819 if (error != 0) { 820 dsl_dataset_rele(ds, FTAG); 821 return (error); 822 } 823 if (!dsl_dataset_is_snapshot(origin)) { 824 dsl_dataset_rele(origin, FTAG); 825 dsl_dataset_rele(ds, FTAG); 826 return (EINVAL); 827 } 828 if (origin->ds_phys->ds_guid != fromguid) { 829 dsl_dataset_rele(origin, FTAG); 830 dsl_dataset_rele(ds, FTAG); 831 return (ENODEV); 832 } 833 dsl_dataset_rele(origin, FTAG); 834 } 835 dsl_dataset_rele(ds, FTAG); 836 error = 0; 837 } 838 return (error); 839} 840 841static void 842dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) 843{ 844 dmu_recv_begin_arg_t *drba = arg; 845 dsl_pool_t *dp = dmu_tx_pool(tx); 846 struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 847 const char *tofs = drba->drba_cookie->drc_tofs; 848 dsl_dataset_t *ds, *newds; 849 uint64_t dsobj; 850 int error; 851 uint64_t crflags; 852 853 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ? 854 DS_FLAG_CI_DATASET : 0; 855 856 error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 857 if (error == 0) { 858 /* create temporary clone */ 859 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, 860 ds->ds_prev, crflags, drba->drba_cred, tx); 861 dsl_dataset_rele(ds, FTAG); 862 } else { 863 dsl_dir_t *dd; 864 const char *tail; 865 dsl_dataset_t *origin = NULL; 866 867 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); 868 869 if (drba->drba_origin != NULL) { 870 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, 871 FTAG, &origin)); 872 } 873 874 /* Create new dataset. */ 875 dsobj = dsl_dataset_create_sync(dd, 876 strrchr(tofs, '/') + 1, 877 origin, crflags, drba->drba_cred, tx); 878 if (origin != NULL) 879 dsl_dataset_rele(origin, FTAG); 880 dsl_dir_rele(dd, FTAG); 881 drba->drba_cookie->drc_newfs = B_TRUE; 882 } 883 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); 884 885 dmu_buf_will_dirty(newds->ds_dbuf, tx); 886 newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; 887 888 /* 889 * If we actually created a non-clone, we need to create the 890 * objset in our new dataset. 891 */ 892 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) { 893 (void) dmu_objset_create_impl(dp->dp_spa, 894 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); 895 } 896 897 drba->drba_cookie->drc_ds = newds; 898 899 spa_history_log_internal_ds(newds, "receive", tx, ""); 900} 901 902/* 903 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() 904 * succeeds; otherwise we will leak the holds on the datasets. 905 */ 906int 907dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb, 908 boolean_t force, char *origin, dmu_recv_cookie_t *drc) 909{ 910 dmu_recv_begin_arg_t drba = { 0 }; 911 dmu_replay_record_t *drr; 912 913 bzero(drc, sizeof (dmu_recv_cookie_t)); 914 drc->drc_drrb = drrb; 915 drc->drc_tosnap = tosnap; 916 drc->drc_tofs = tofs; 917 drc->drc_force = force; 918 919 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) 920 drc->drc_byteswap = B_TRUE; 921 else if (drrb->drr_magic != DMU_BACKUP_MAGIC) 922 return (EINVAL); 923 924 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 925 drr->drr_type = DRR_BEGIN; 926 drr->drr_u.drr_begin = *drc->drc_drrb; 927 if (drc->drc_byteswap) { 928 fletcher_4_incremental_byteswap(drr, 929 sizeof (dmu_replay_record_t), &drc->drc_cksum); 930 } else { 931 fletcher_4_incremental_native(drr, 932 sizeof (dmu_replay_record_t), &drc->drc_cksum); 933 } 934 kmem_free(drr, sizeof (dmu_replay_record_t)); 935 936 if (drc->drc_byteswap) { 937 drrb->drr_magic = BSWAP_64(drrb->drr_magic); 938 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); 939 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); 940 drrb->drr_type = BSWAP_32(drrb->drr_type); 941 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); 942 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); 943 } 944 945 drba.drba_origin = origin; 946 drba.drba_cookie = drc; 947 drba.drba_cred = CRED(); 948 949 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync, 950 &drba, 5)); 951} 952 953struct restorearg { 954 int err; 955 boolean_t byteswap; 956 kthread_t *td; 957 struct file *fp; 958 char *buf; 959 uint64_t voff; 960 int bufsize; /* amount of memory allocated for buf */ 961 zio_cksum_t cksum; 962 avl_tree_t *guid_to_ds_map; 963}; 964 965typedef struct guid_map_entry { 966 uint64_t guid; 967 dsl_dataset_t *gme_ds; 968 avl_node_t avlnode; 969} guid_map_entry_t; 970 971static int 972guid_compare(const void *arg1, const void *arg2) 973{ 974 const guid_map_entry_t *gmep1 = arg1; 975 const guid_map_entry_t *gmep2 = arg2; 976 977 if (gmep1->guid < gmep2->guid) 978 return (-1); 979 else if (gmep1->guid > gmep2->guid) 980 return (1); 981 return (0); 982} 983 984static void 985free_guid_map_onexit(void *arg) 986{ 987 avl_tree_t *ca = arg; 988 void *cookie = NULL; 989 guid_map_entry_t *gmep; 990 991 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { 992 dsl_dataset_long_rele(gmep->gme_ds, gmep); 993 kmem_free(gmep, sizeof (guid_map_entry_t)); 994 } 995 avl_destroy(ca); 996 kmem_free(ca, sizeof (avl_tree_t)); 997} 998 999static int 1000restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid) 1001{ 1002 struct uio auio; 1003 struct iovec aiov; 1004 int error; 1005 1006 aiov.iov_base = buf; 1007 aiov.iov_len = len; 1008 auio.uio_iov = &aiov; 1009 auio.uio_iovcnt = 1; 1010 auio.uio_resid = len; 1011 auio.uio_segflg = UIO_SYSSPACE; 1012 auio.uio_rw = UIO_READ; 1013 auio.uio_offset = off; 1014 auio.uio_td = ra->td; 1015#ifdef _KERNEL 1016 error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td); 1017#else 1018 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__); 1019 error = EOPNOTSUPP; 1020#endif 1021 *resid = auio.uio_resid; 1022 return (error); 1023} 1024 1025static void * 1026restore_read(struct restorearg *ra, int len) 1027{ 1028 void *rv; 1029 int done = 0; 1030 1031 /* some things will require 8-byte alignment, so everything must */ 1032 ASSERT0(len % 8); 1033 1034 while (done < len) { 1035 ssize_t resid; 1036 1037 ra->err = restore_bytes(ra, (caddr_t)ra->buf + done, 1038 len - done, ra->voff, &resid); 1039 1040 if (resid == len - done) 1041 ra->err = EINVAL; 1042 ra->voff += len - done - resid; 1043 done = len - resid; 1044 if (ra->err != 0) 1045 return (NULL); 1046 } 1047 1048 ASSERT3U(done, ==, len); 1049 rv = ra->buf; 1050 if (ra->byteswap) 1051 fletcher_4_incremental_byteswap(rv, len, &ra->cksum); 1052 else 1053 fletcher_4_incremental_native(rv, len, &ra->cksum); 1054 return (rv); 1055} 1056 1057static void 1058backup_byteswap(dmu_replay_record_t *drr) 1059{ 1060#define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) 1061#define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) 1062 drr->drr_type = BSWAP_32(drr->drr_type); 1063 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); 1064 switch (drr->drr_type) { 1065 case DRR_BEGIN: 1066 DO64(drr_begin.drr_magic); 1067 DO64(drr_begin.drr_versioninfo); 1068 DO64(drr_begin.drr_creation_time); 1069 DO32(drr_begin.drr_type); 1070 DO32(drr_begin.drr_flags); 1071 DO64(drr_begin.drr_toguid); 1072 DO64(drr_begin.drr_fromguid); 1073 break; 1074 case DRR_OBJECT: 1075 DO64(drr_object.drr_object); 1076 /* DO64(drr_object.drr_allocation_txg); */ 1077 DO32(drr_object.drr_type); 1078 DO32(drr_object.drr_bonustype); 1079 DO32(drr_object.drr_blksz); 1080 DO32(drr_object.drr_bonuslen); 1081 DO64(drr_object.drr_toguid); 1082 break; 1083 case DRR_FREEOBJECTS: 1084 DO64(drr_freeobjects.drr_firstobj); 1085 DO64(drr_freeobjects.drr_numobjs); 1086 DO64(drr_freeobjects.drr_toguid); 1087 break; 1088 case DRR_WRITE: 1089 DO64(drr_write.drr_object); 1090 DO32(drr_write.drr_type); 1091 DO64(drr_write.drr_offset); 1092 DO64(drr_write.drr_length); 1093 DO64(drr_write.drr_toguid); 1094 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]); 1095 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]); 1096 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]); 1097 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]); 1098 DO64(drr_write.drr_key.ddk_prop); 1099 break; 1100 case DRR_WRITE_BYREF: 1101 DO64(drr_write_byref.drr_object); 1102 DO64(drr_write_byref.drr_offset); 1103 DO64(drr_write_byref.drr_length); 1104 DO64(drr_write_byref.drr_toguid); 1105 DO64(drr_write_byref.drr_refguid); 1106 DO64(drr_write_byref.drr_refobject); 1107 DO64(drr_write_byref.drr_refoffset); 1108 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]); 1109 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]); 1110 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]); 1111 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]); 1112 DO64(drr_write_byref.drr_key.ddk_prop); 1113 break; 1114 case DRR_FREE: 1115 DO64(drr_free.drr_object); 1116 DO64(drr_free.drr_offset); 1117 DO64(drr_free.drr_length); 1118 DO64(drr_free.drr_toguid); 1119 break; 1120 case DRR_SPILL: 1121 DO64(drr_spill.drr_object); 1122 DO64(drr_spill.drr_length); 1123 DO64(drr_spill.drr_toguid); 1124 break; 1125 case DRR_END: 1126 DO64(drr_end.drr_checksum.zc_word[0]); 1127 DO64(drr_end.drr_checksum.zc_word[1]); 1128 DO64(drr_end.drr_checksum.zc_word[2]); 1129 DO64(drr_end.drr_checksum.zc_word[3]); 1130 DO64(drr_end.drr_toguid); 1131 break; 1132 } 1133#undef DO64 1134#undef DO32 1135} 1136 1137static int 1138restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) 1139{ 1140 int err; 1141 dmu_tx_t *tx; 1142 void *data = NULL; 1143 1144 if (drro->drr_type == DMU_OT_NONE || 1145 !DMU_OT_IS_VALID(drro->drr_type) || 1146 !DMU_OT_IS_VALID(drro->drr_bonustype) || 1147 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || 1148 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || 1149 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || 1150 drro->drr_blksz < SPA_MINBLOCKSIZE || 1151 drro->drr_blksz > SPA_MAXBLOCKSIZE || 1152 drro->drr_bonuslen > DN_MAX_BONUSLEN) { 1153 return (EINVAL); 1154 } 1155 1156 err = dmu_object_info(os, drro->drr_object, NULL); 1157 1158 if (err != 0 && err != ENOENT) 1159 return (EINVAL); 1160 1161 if (drro->drr_bonuslen) { 1162 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8)); 1163 if (ra->err != 0) 1164 return (ra->err); 1165 } 1166 1167 if (err == ENOENT) { 1168 /* currently free, want to be allocated */ 1169 tx = dmu_tx_create(os); 1170 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1171 err = dmu_tx_assign(tx, TXG_WAIT); 1172 if (err != 0) { 1173 dmu_tx_abort(tx); 1174 return (err); 1175 } 1176 err = dmu_object_claim(os, drro->drr_object, 1177 drro->drr_type, drro->drr_blksz, 1178 drro->drr_bonustype, drro->drr_bonuslen, tx); 1179 dmu_tx_commit(tx); 1180 } else { 1181 /* currently allocated, want to be allocated */ 1182 err = dmu_object_reclaim(os, drro->drr_object, 1183 drro->drr_type, drro->drr_blksz, 1184 drro->drr_bonustype, drro->drr_bonuslen); 1185 } 1186 if (err != 0) { 1187 return (EINVAL); 1188 } 1189 1190 tx = dmu_tx_create(os); 1191 dmu_tx_hold_bonus(tx, drro->drr_object); 1192 err = dmu_tx_assign(tx, TXG_WAIT); 1193 if (err != 0) { 1194 dmu_tx_abort(tx); 1195 return (err); 1196 } 1197 1198 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype, 1199 tx); 1200 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx); 1201 1202 if (data != NULL) { 1203 dmu_buf_t *db; 1204 1205 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db)); 1206 dmu_buf_will_dirty(db, tx); 1207 1208 ASSERT3U(db->db_size, >=, drro->drr_bonuslen); 1209 bcopy(data, db->db_data, drro->drr_bonuslen); 1210 if (ra->byteswap) { 1211 dmu_object_byteswap_t byteswap = 1212 DMU_OT_BYTESWAP(drro->drr_bonustype); 1213 dmu_ot_byteswap[byteswap].ob_func(db->db_data, 1214 drro->drr_bonuslen); 1215 } 1216 dmu_buf_rele(db, FTAG); 1217 } 1218 dmu_tx_commit(tx); 1219 return (0); 1220} 1221 1222/* ARGSUSED */ 1223static int 1224restore_freeobjects(struct restorearg *ra, objset_t *os, 1225 struct drr_freeobjects *drrfo) 1226{ 1227 uint64_t obj; 1228 1229 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) 1230 return (EINVAL); 1231 1232 for (obj = drrfo->drr_firstobj; 1233 obj < drrfo->drr_firstobj + drrfo->drr_numobjs; 1234 (void) dmu_object_next(os, &obj, FALSE, 0)) { 1235 int err; 1236 1237 if (dmu_object_info(os, obj, NULL) != 0) 1238 continue; 1239 1240 err = dmu_free_object(os, obj); 1241 if (err != 0) 1242 return (err); 1243 } 1244 return (0); 1245} 1246 1247static int 1248restore_write(struct restorearg *ra, objset_t *os, 1249 struct drr_write *drrw) 1250{ 1251 dmu_tx_t *tx; 1252 void *data; 1253 int err; 1254 1255 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset || 1256 !DMU_OT_IS_VALID(drrw->drr_type)) 1257 return (EINVAL); 1258 1259 data = restore_read(ra, drrw->drr_length); 1260 if (data == NULL) 1261 return (ra->err); 1262 1263 if (dmu_object_info(os, drrw->drr_object, NULL) != 0) 1264 return (EINVAL); 1265 1266 tx = dmu_tx_create(os); 1267 1268 dmu_tx_hold_write(tx, drrw->drr_object, 1269 drrw->drr_offset, drrw->drr_length); 1270 err = dmu_tx_assign(tx, TXG_WAIT); 1271 if (err != 0) { 1272 dmu_tx_abort(tx); 1273 return (err); 1274 } 1275 if (ra->byteswap) { 1276 dmu_object_byteswap_t byteswap = 1277 DMU_OT_BYTESWAP(drrw->drr_type); 1278 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length); 1279 } 1280 dmu_write(os, drrw->drr_object, 1281 drrw->drr_offset, drrw->drr_length, data, tx); 1282 dmu_tx_commit(tx); 1283 return (0); 1284} 1285 1286/* 1287 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed 1288 * streams to refer to a copy of the data that is already on the 1289 * system because it came in earlier in the stream. This function 1290 * finds the earlier copy of the data, and uses that copy instead of 1291 * data from the stream to fulfill this write. 1292 */ 1293static int 1294restore_write_byref(struct restorearg *ra, objset_t *os, 1295 struct drr_write_byref *drrwbr) 1296{ 1297 dmu_tx_t *tx; 1298 int err; 1299 guid_map_entry_t gmesrch; 1300 guid_map_entry_t *gmep; 1301 avl_index_t where; 1302 objset_t *ref_os = NULL; 1303 dmu_buf_t *dbp; 1304 1305 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) 1306 return (EINVAL); 1307 1308 /* 1309 * If the GUID of the referenced dataset is different from the 1310 * GUID of the target dataset, find the referenced dataset. 1311 */ 1312 if (drrwbr->drr_toguid != drrwbr->drr_refguid) { 1313 gmesrch.guid = drrwbr->drr_refguid; 1314 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch, 1315 &where)) == NULL) { 1316 return (EINVAL); 1317 } 1318 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) 1319 return (EINVAL); 1320 } else { 1321 ref_os = os; 1322 } 1323 1324 if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, 1325 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH)) 1326 return (err); 1327 1328 tx = dmu_tx_create(os); 1329 1330 dmu_tx_hold_write(tx, drrwbr->drr_object, 1331 drrwbr->drr_offset, drrwbr->drr_length); 1332 err = dmu_tx_assign(tx, TXG_WAIT); 1333 if (err != 0) { 1334 dmu_tx_abort(tx); 1335 return (err); 1336 } 1337 dmu_write(os, drrwbr->drr_object, 1338 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); 1339 dmu_buf_rele(dbp, FTAG); 1340 dmu_tx_commit(tx); 1341 return (0); 1342} 1343 1344static int 1345restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs) 1346{ 1347 dmu_tx_t *tx; 1348 void *data; 1349 dmu_buf_t *db, *db_spill; 1350 int err; 1351 1352 if (drrs->drr_length < SPA_MINBLOCKSIZE || 1353 drrs->drr_length > SPA_MAXBLOCKSIZE) 1354 return (EINVAL); 1355 1356 data = restore_read(ra, drrs->drr_length); 1357 if (data == NULL) 1358 return (ra->err); 1359 1360 if (dmu_object_info(os, drrs->drr_object, NULL) != 0) 1361 return (EINVAL); 1362 1363 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db)); 1364 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) { 1365 dmu_buf_rele(db, FTAG); 1366 return (err); 1367 } 1368 1369 tx = dmu_tx_create(os); 1370 1371 dmu_tx_hold_spill(tx, db->db_object); 1372 1373 err = dmu_tx_assign(tx, TXG_WAIT); 1374 if (err != 0) { 1375 dmu_buf_rele(db, FTAG); 1376 dmu_buf_rele(db_spill, FTAG); 1377 dmu_tx_abort(tx); 1378 return (err); 1379 } 1380 dmu_buf_will_dirty(db_spill, tx); 1381 1382 if (db_spill->db_size < drrs->drr_length) 1383 VERIFY(0 == dbuf_spill_set_blksz(db_spill, 1384 drrs->drr_length, tx)); 1385 bcopy(data, db_spill->db_data, drrs->drr_length); 1386 1387 dmu_buf_rele(db, FTAG); 1388 dmu_buf_rele(db_spill, FTAG); 1389 1390 dmu_tx_commit(tx); 1391 return (0); 1392} 1393 1394/* ARGSUSED */ 1395static int 1396restore_free(struct restorearg *ra, objset_t *os, 1397 struct drr_free *drrf) 1398{ 1399 int err; 1400 1401 if (drrf->drr_length != -1ULL && 1402 drrf->drr_offset + drrf->drr_length < drrf->drr_offset) 1403 return (EINVAL); 1404 1405 if (dmu_object_info(os, drrf->drr_object, NULL) != 0) 1406 return (EINVAL); 1407 1408 err = dmu_free_long_range(os, drrf->drr_object, 1409 drrf->drr_offset, drrf->drr_length); 1410 return (err); 1411} 1412 1413/* used to destroy the drc_ds on error */ 1414static void 1415dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) 1416{ 1417 char name[MAXNAMELEN]; 1418 dsl_dataset_name(drc->drc_ds, name); 1419 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 1420 (void) dsl_destroy_head(name); 1421} 1422 1423/* 1424 * NB: callers *must* call dmu_recv_end() if this succeeds. 1425 */ 1426int 1427dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp, 1428 int cleanup_fd, uint64_t *action_handlep) 1429{ 1430 struct restorearg ra = { 0 }; 1431 dmu_replay_record_t *drr; 1432 objset_t *os; 1433 zio_cksum_t pcksum; 1434 int featureflags; 1435 1436 ra.byteswap = drc->drc_byteswap; 1437 ra.cksum = drc->drc_cksum; 1438 ra.td = curthread; 1439 ra.fp = fp; 1440 ra.voff = *voffp; 1441 ra.bufsize = 1<<20; 1442 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP); 1443 1444 /* these were verified in dmu_recv_begin */ 1445 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, 1446 DMU_SUBSTREAM); 1447 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); 1448 1449 /* 1450 * Open the objset we are modifying. 1451 */ 1452 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os)); 1453 1454 ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT); 1455 1456 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); 1457 1458 /* if this stream is dedup'ed, set up the avl tree for guid mapping */ 1459 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { 1460 minor_t minor; 1461 1462 if (cleanup_fd == -1) { 1463 ra.err = EBADF; 1464 goto out; 1465 } 1466 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor); 1467 if (ra.err != 0) { 1468 cleanup_fd = -1; 1469 goto out; 1470 } 1471 1472 if (*action_handlep == 0) { 1473 ra.guid_to_ds_map = 1474 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); 1475 avl_create(ra.guid_to_ds_map, guid_compare, 1476 sizeof (guid_map_entry_t), 1477 offsetof(guid_map_entry_t, avlnode)); 1478 ra.err = zfs_onexit_add_cb(minor, 1479 free_guid_map_onexit, ra.guid_to_ds_map, 1480 action_handlep); 1481 if (ra.err != 0) 1482 goto out; 1483 } else { 1484 ra.err = zfs_onexit_cb_data(minor, *action_handlep, 1485 (void **)&ra.guid_to_ds_map); 1486 if (ra.err != 0) 1487 goto out; 1488 } 1489 1490 drc->drc_guid_to_ds_map = ra.guid_to_ds_map; 1491 } 1492 1493 /* 1494 * Read records and process them. 1495 */ 1496 pcksum = ra.cksum; 1497 while (ra.err == 0 && 1498 NULL != (drr = restore_read(&ra, sizeof (*drr)))) { 1499 if (issig(JUSTLOOKING) && issig(FORREAL)) { 1500 ra.err = EINTR; 1501 goto out; 1502 } 1503 1504 if (ra.byteswap) 1505 backup_byteswap(drr); 1506 1507 switch (drr->drr_type) { 1508 case DRR_OBJECT: 1509 { 1510 /* 1511 * We need to make a copy of the record header, 1512 * because restore_{object,write} may need to 1513 * restore_read(), which will invalidate drr. 1514 */ 1515 struct drr_object drro = drr->drr_u.drr_object; 1516 ra.err = restore_object(&ra, os, &drro); 1517 break; 1518 } 1519 case DRR_FREEOBJECTS: 1520 { 1521 struct drr_freeobjects drrfo = 1522 drr->drr_u.drr_freeobjects; 1523 ra.err = restore_freeobjects(&ra, os, &drrfo); 1524 break; 1525 } 1526 case DRR_WRITE: 1527 { 1528 struct drr_write drrw = drr->drr_u.drr_write; 1529 ra.err = restore_write(&ra, os, &drrw); 1530 break; 1531 } 1532 case DRR_WRITE_BYREF: 1533 { 1534 struct drr_write_byref drrwbr = 1535 drr->drr_u.drr_write_byref; 1536 ra.err = restore_write_byref(&ra, os, &drrwbr); 1537 break; 1538 } 1539 case DRR_FREE: 1540 { 1541 struct drr_free drrf = drr->drr_u.drr_free; 1542 ra.err = restore_free(&ra, os, &drrf); 1543 break; 1544 } 1545 case DRR_END: 1546 { 1547 struct drr_end drre = drr->drr_u.drr_end; 1548 /* 1549 * We compare against the *previous* checksum 1550 * value, because the stored checksum is of 1551 * everything before the DRR_END record. 1552 */ 1553 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum)) 1554 ra.err = ECKSUM; 1555 goto out; 1556 } 1557 case DRR_SPILL: 1558 { 1559 struct drr_spill drrs = drr->drr_u.drr_spill; 1560 ra.err = restore_spill(&ra, os, &drrs); 1561 break; 1562 } 1563 default: 1564 ra.err = EINVAL; 1565 goto out; 1566 } 1567 pcksum = ra.cksum; 1568 } 1569 ASSERT(ra.err != 0); 1570 1571out: 1572 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) 1573 zfs_onexit_fd_rele(cleanup_fd); 1574 1575 if (ra.err != 0) { 1576 /* 1577 * destroy what we created, so we don't leave it in the 1578 * inconsistent restoring state. 1579 */ 1580 dmu_recv_cleanup_ds(drc); 1581 } 1582 1583 kmem_free(ra.buf, ra.bufsize); 1584 *voffp = ra.voff; 1585 return (ra.err); 1586} 1587 1588static int 1589dmu_recv_end_check(void *arg, dmu_tx_t *tx) 1590{ 1591 dmu_recv_cookie_t *drc = arg; 1592 dsl_pool_t *dp = dmu_tx_pool(tx); 1593 int error; 1594 1595 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); 1596 1597 if (!drc->drc_newfs) { 1598 dsl_dataset_t *origin_head; 1599 1600 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); 1601 if (error != 0) 1602 return (error); 1603 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, 1604 origin_head, drc->drc_force); 1605 if (error != 0) { 1606 dsl_dataset_rele(origin_head, FTAG); 1607 return (error); 1608 } 1609 error = dsl_dataset_snapshot_check_impl(origin_head, 1610 drc->drc_tosnap, tx); 1611 dsl_dataset_rele(origin_head, FTAG); 1612 if (error != 0) 1613 return (error); 1614 1615 error = dsl_destroy_head_check_impl(drc->drc_ds, 1); 1616 } else { 1617 error = dsl_dataset_snapshot_check_impl(drc->drc_ds, 1618 drc->drc_tosnap, tx); 1619 } 1620 return (error); 1621} 1622 1623static void 1624dmu_recv_end_sync(void *arg, dmu_tx_t *tx) 1625{ 1626 dmu_recv_cookie_t *drc = arg; 1627 dsl_pool_t *dp = dmu_tx_pool(tx); 1628 1629 spa_history_log_internal_ds(drc->drc_ds, "finish receiving", 1630 tx, "snap=%s", drc->drc_tosnap); 1631 1632 if (!drc->drc_newfs) { 1633 dsl_dataset_t *origin_head; 1634 1635 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, 1636 &origin_head)); 1637 dsl_dataset_clone_swap_sync_impl(drc->drc_ds, 1638 origin_head, tx); 1639 dsl_dataset_snapshot_sync_impl(origin_head, 1640 drc->drc_tosnap, tx); 1641 1642 /* set snapshot's creation time and guid */ 1643 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); 1644 origin_head->ds_prev->ds_phys->ds_creation_time = 1645 drc->drc_drrb->drr_creation_time; 1646 origin_head->ds_prev->ds_phys->ds_guid = 1647 drc->drc_drrb->drr_toguid; 1648 origin_head->ds_prev->ds_phys->ds_flags &= 1649 ~DS_FLAG_INCONSISTENT; 1650 1651 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 1652 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; 1653 1654 dsl_dataset_rele(origin_head, FTAG); 1655 dsl_destroy_head_sync_impl(drc->drc_ds, tx); 1656 } else { 1657 dsl_dataset_t *ds = drc->drc_ds; 1658 1659 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); 1660 1661 /* set snapshot's creation time and guid */ 1662 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1663 ds->ds_prev->ds_phys->ds_creation_time = 1664 drc->drc_drrb->drr_creation_time; 1665 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid; 1666 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; 1667 1668 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1669 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; 1670 } 1671 drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj; 1672 /* 1673 * Release the hold from dmu_recv_begin. This must be done before 1674 * we return to open context, so that when we free the dataset's dnode, 1675 * we can evict its bonus buffer. 1676 */ 1677 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 1678 drc->drc_ds = NULL; 1679} 1680 1681static int 1682add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj) 1683{ 1684 dsl_pool_t *dp; 1685 dsl_dataset_t *snapds; 1686 guid_map_entry_t *gmep; 1687 int err; 1688 1689 ASSERT(guid_map != NULL); 1690 1691 err = dsl_pool_hold(name, FTAG, &dp); 1692 if (err != 0) 1693 return (err); 1694 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snapds); 1695 if (err == 0) { 1696 gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP); 1697 gmep->guid = snapds->ds_phys->ds_guid; 1698 gmep->gme_ds = snapds; 1699 avl_add(guid_map, gmep); 1700 dsl_dataset_long_hold(snapds, gmep); 1701 dsl_dataset_rele(snapds, FTAG); 1702 } 1703 1704 dsl_pool_rele(dp, FTAG); 1705 return (err); 1706} 1707 1708static int dmu_recv_end_modified_blocks = 3; 1709 1710static int 1711dmu_recv_existing_end(dmu_recv_cookie_t *drc) 1712{ 1713 int error; 1714 char name[MAXNAMELEN]; 1715 1716#ifdef _KERNEL 1717 /* 1718 * We will be destroying the ds; make sure its origin is unmounted if 1719 * necessary. 1720 */ 1721 dsl_dataset_name(drc->drc_ds, name); 1722 zfs_destroy_unmount_origin(name); 1723#endif 1724 1725 error = dsl_sync_task(drc->drc_tofs, 1726 dmu_recv_end_check, dmu_recv_end_sync, drc, 1727 dmu_recv_end_modified_blocks); 1728 1729 if (error != 0) 1730 dmu_recv_cleanup_ds(drc); 1731 return (error); 1732} 1733 1734static int 1735dmu_recv_new_end(dmu_recv_cookie_t *drc) 1736{ 1737 int error; 1738 1739 error = dsl_sync_task(drc->drc_tofs, 1740 dmu_recv_end_check, dmu_recv_end_sync, drc, 1741 dmu_recv_end_modified_blocks); 1742 1743 if (error != 0) { 1744 dmu_recv_cleanup_ds(drc); 1745 } else if (drc->drc_guid_to_ds_map != NULL) { 1746 (void) add_ds_to_guidmap(drc->drc_tofs, 1747 drc->drc_guid_to_ds_map, 1748 drc->drc_newsnapobj); 1749 } 1750 return (error); 1751} 1752 1753int 1754dmu_recv_end(dmu_recv_cookie_t *drc) 1755{ 1756 if (drc->drc_newfs) 1757 return (dmu_recv_new_end(drc)); 1758 else 1759 return (dmu_recv_existing_end(drc)); 1760} 1761