10fa1b3ccSPaul Dagnelie /* 20fa1b3ccSPaul Dagnelie * CDDL HEADER START 30fa1b3ccSPaul Dagnelie * 40fa1b3ccSPaul Dagnelie * The contents of this file are subject to the terms of the 50fa1b3ccSPaul Dagnelie * Common Development and Distribution License (the "License"). 60fa1b3ccSPaul Dagnelie * You may not use this file except in compliance with the License. 70fa1b3ccSPaul Dagnelie * 80fa1b3ccSPaul Dagnelie * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 90fa1b3ccSPaul Dagnelie * or http://www.opensolaris.org/os/licensing. 100fa1b3ccSPaul Dagnelie * See the License for the specific language governing permissions 110fa1b3ccSPaul Dagnelie * and limitations under the License. 120fa1b3ccSPaul Dagnelie * 130fa1b3ccSPaul Dagnelie * When distributing Covered Code, include this CDDL HEADER in each 140fa1b3ccSPaul Dagnelie * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 150fa1b3ccSPaul Dagnelie * If applicable, add the following below this CDDL HEADER, with the 160fa1b3ccSPaul Dagnelie * fields enclosed by brackets "[]" replaced with your own identifying 170fa1b3ccSPaul Dagnelie * information: Portions Copyright [yyyy] [name of copyright owner] 180fa1b3ccSPaul Dagnelie * 190fa1b3ccSPaul Dagnelie * CDDL HEADER END 200fa1b3ccSPaul Dagnelie */ 210fa1b3ccSPaul Dagnelie /* 220fa1b3ccSPaul Dagnelie * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 230fa1b3ccSPaul Dagnelie * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 240fa1b3ccSPaul Dagnelie * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 250fa1b3ccSPaul Dagnelie * Copyright (c) 2014, Joyent, Inc. All rights reserved. 260fa1b3ccSPaul Dagnelie * Copyright 2014 HybridCluster. All rights reserved. 270fa1b3ccSPaul Dagnelie * Copyright 2016 RackTop Systems. 280fa1b3ccSPaul Dagnelie * Copyright (c) 2014 Integros [integros.com] 290fa1b3ccSPaul Dagnelie */ 300fa1b3ccSPaul Dagnelie 310fa1b3ccSPaul Dagnelie #include <sys/dmu.h> 320fa1b3ccSPaul Dagnelie #include <sys/dmu_impl.h> 330fa1b3ccSPaul Dagnelie #include <sys/dmu_tx.h> 340fa1b3ccSPaul Dagnelie #include <sys/dbuf.h> 350fa1b3ccSPaul Dagnelie #include <sys/dnode.h> 360fa1b3ccSPaul Dagnelie #include <sys/zfs_context.h> 370fa1b3ccSPaul Dagnelie #include <sys/dmu_objset.h> 380fa1b3ccSPaul Dagnelie #include <sys/dmu_traverse.h> 390fa1b3ccSPaul Dagnelie #include <sys/dsl_dataset.h> 400fa1b3ccSPaul Dagnelie #include <sys/dsl_dir.h> 410fa1b3ccSPaul Dagnelie #include <sys/dsl_prop.h> 420fa1b3ccSPaul Dagnelie #include <sys/dsl_pool.h> 430fa1b3ccSPaul Dagnelie #include <sys/dsl_synctask.h> 440fa1b3ccSPaul Dagnelie #include <sys/zfs_ioctl.h> 450fa1b3ccSPaul Dagnelie #include <sys/zap.h> 460fa1b3ccSPaul Dagnelie #include <sys/zio_checksum.h> 470fa1b3ccSPaul Dagnelie #include <sys/zfs_znode.h> 480fa1b3ccSPaul Dagnelie #include <zfs_fletcher.h> 490fa1b3ccSPaul Dagnelie #include <sys/avl.h> 500fa1b3ccSPaul Dagnelie #include <sys/ddt.h> 510fa1b3ccSPaul Dagnelie #include <sys/zfs_onexit.h> 520fa1b3ccSPaul Dagnelie #include <sys/dmu_recv.h> 530fa1b3ccSPaul Dagnelie #include <sys/dsl_destroy.h> 540fa1b3ccSPaul Dagnelie #include <sys/blkptr.h> 550fa1b3ccSPaul Dagnelie #include <sys/dsl_bookmark.h> 560fa1b3ccSPaul Dagnelie #include <sys/zfeature.h> 570fa1b3ccSPaul Dagnelie #include <sys/bqueue.h> 580fa1b3ccSPaul Dagnelie 590fa1b3ccSPaul Dagnelie int zfs_recv_queue_length = SPA_MAXBLOCKSIZE; 600fa1b3ccSPaul Dagnelie 610fa1b3ccSPaul Dagnelie static char *dmu_recv_tag = "dmu_recv_tag"; 620fa1b3ccSPaul Dagnelie const char *recv_clone_name = "%recv"; 630fa1b3ccSPaul Dagnelie 640fa1b3ccSPaul Dagnelie static void byteswap_record(dmu_replay_record_t *drr); 650fa1b3ccSPaul Dagnelie 660fa1b3ccSPaul Dagnelie typedef struct dmu_recv_begin_arg { 670fa1b3ccSPaul Dagnelie const char *drba_origin; 680fa1b3ccSPaul Dagnelie dmu_recv_cookie_t *drba_cookie; 690fa1b3ccSPaul Dagnelie cred_t *drba_cred; 70eb633035STom Caputi dsl_crypto_params_t *drba_dcp; 710fa1b3ccSPaul Dagnelie } dmu_recv_begin_arg_t; 720fa1b3ccSPaul Dagnelie 730fa1b3ccSPaul Dagnelie static int 740fa1b3ccSPaul Dagnelie recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, 75*6ccda740Sloli uint64_t fromguid, uint64_t featureflags) 760fa1b3ccSPaul Dagnelie { 770fa1b3ccSPaul Dagnelie uint64_t val; 780fa1b3ccSPaul Dagnelie int error; 790fa1b3ccSPaul Dagnelie dsl_pool_t *dp = ds->ds_dir->dd_pool; 80eb633035STom Caputi boolean_t encrypted = ds->ds_dir->dd_crypto_obj != 0; 81eb633035STom Caputi boolean_t raw = (featureflags & DMU_BACKUP_FEATURE_RAW) != 0; 82eb633035STom Caputi boolean_t embed = (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) != 0; 830fa1b3ccSPaul Dagnelie 840fa1b3ccSPaul Dagnelie /* temporary clone name must not exist */ 850fa1b3ccSPaul Dagnelie error = zap_lookup(dp->dp_meta_objset, 860fa1b3ccSPaul Dagnelie dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, 870fa1b3ccSPaul Dagnelie 8, 1, &val); 880fa1b3ccSPaul Dagnelie if (error != ENOENT) 890fa1b3ccSPaul Dagnelie return (error == 0 ? EBUSY : error); 900fa1b3ccSPaul Dagnelie 910fa1b3ccSPaul Dagnelie /* new snapshot name must not exist */ 920fa1b3ccSPaul Dagnelie error = zap_lookup(dp->dp_meta_objset, 930fa1b3ccSPaul Dagnelie dsl_dataset_phys(ds)->ds_snapnames_zapobj, 940fa1b3ccSPaul Dagnelie drba->drba_cookie->drc_tosnap, 8, 1, &val); 950fa1b3ccSPaul Dagnelie if (error != ENOENT) 960fa1b3ccSPaul Dagnelie return (error == 0 ? EEXIST : error); 970fa1b3ccSPaul Dagnelie 980fa1b3ccSPaul Dagnelie /* 990fa1b3ccSPaul Dagnelie * Check snapshot limit before receiving. We'll recheck again at the 1000fa1b3ccSPaul Dagnelie * end, but might as well abort before receiving if we're already over 1010fa1b3ccSPaul Dagnelie * the limit. 1020fa1b3ccSPaul Dagnelie * 1030fa1b3ccSPaul Dagnelie * Note that we do not check the file system limit with 1040fa1b3ccSPaul Dagnelie * dsl_dir_fscount_check because the temporary %clones don't count 1050fa1b3ccSPaul Dagnelie * against that limit. 1060fa1b3ccSPaul Dagnelie */ 1070fa1b3ccSPaul Dagnelie error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, 1080fa1b3ccSPaul Dagnelie NULL, drba->drba_cred); 1090fa1b3ccSPaul Dagnelie if (error != 0) 1100fa1b3ccSPaul Dagnelie return (error); 1110fa1b3ccSPaul Dagnelie 1120fa1b3ccSPaul Dagnelie if (fromguid != 0) { 1130fa1b3ccSPaul Dagnelie dsl_dataset_t *snap; 1140fa1b3ccSPaul Dagnelie uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1150fa1b3ccSPaul Dagnelie 116eb633035STom Caputi /* Can't raw receive on top of an unencrypted dataset */ 117eb633035STom Caputi if (!encrypted && raw) 118eb633035STom Caputi return (SET_ERROR(EINVAL)); 119eb633035STom Caputi 120eb633035STom Caputi /* Encryption is incompatible with embedded data */ 121eb633035STom Caputi if (encrypted && embed) 122eb633035STom Caputi return (SET_ERROR(EINVAL)); 123eb633035STom Caputi 1240fa1b3ccSPaul Dagnelie /* Find snapshot in this dir that matches fromguid. */ 1250fa1b3ccSPaul Dagnelie while (obj != 0) { 1260fa1b3ccSPaul Dagnelie error = dsl_dataset_hold_obj(dp, obj, FTAG, 1270fa1b3ccSPaul Dagnelie &snap); 1280fa1b3ccSPaul Dagnelie if (error != 0) 1290fa1b3ccSPaul Dagnelie return (SET_ERROR(ENODEV)); 1300fa1b3ccSPaul Dagnelie if (snap->ds_dir != ds->ds_dir) { 1310fa1b3ccSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 1320fa1b3ccSPaul Dagnelie return (SET_ERROR(ENODEV)); 1330fa1b3ccSPaul Dagnelie } 1340fa1b3ccSPaul Dagnelie if (dsl_dataset_phys(snap)->ds_guid == fromguid) 1350fa1b3ccSPaul Dagnelie break; 1360fa1b3ccSPaul Dagnelie obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 1370fa1b3ccSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 1380fa1b3ccSPaul Dagnelie } 1390fa1b3ccSPaul Dagnelie if (obj == 0) 1400fa1b3ccSPaul Dagnelie return (SET_ERROR(ENODEV)); 1410fa1b3ccSPaul Dagnelie 1420fa1b3ccSPaul Dagnelie if (drba->drba_cookie->drc_force) { 143eb633035STom Caputi drba->drba_cookie->drc_fromsnapobj = obj; 1440fa1b3ccSPaul Dagnelie } else { 1450fa1b3ccSPaul Dagnelie /* 1460fa1b3ccSPaul Dagnelie * If we are not forcing, there must be no 1470fa1b3ccSPaul Dagnelie * changes since fromsnap. 1480fa1b3ccSPaul Dagnelie */ 1490fa1b3ccSPaul Dagnelie if (dsl_dataset_modified_since_snap(ds, snap)) { 1500fa1b3ccSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 1510fa1b3ccSPaul Dagnelie return (SET_ERROR(ETXTBSY)); 1520fa1b3ccSPaul Dagnelie } 153eb633035STom Caputi drba->drba_cookie->drc_fromsnapobj = 154eb633035STom Caputi ds->ds_prev->ds_object; 1550fa1b3ccSPaul Dagnelie } 1560fa1b3ccSPaul Dagnelie 1570fa1b3ccSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 1580fa1b3ccSPaul Dagnelie } else { 1590fa1b3ccSPaul Dagnelie /* if full, then must be forced */ 1600fa1b3ccSPaul Dagnelie if (!drba->drba_cookie->drc_force) 1610fa1b3ccSPaul Dagnelie return (SET_ERROR(EEXIST)); 162eb633035STom Caputi 163eb633035STom Caputi /* 164eb633035STom Caputi * We don't support using zfs recv -F to blow away 165eb633035STom Caputi * encrypted filesystems. This would require the 166eb633035STom Caputi * dsl dir to point to the old encryption key and 167eb633035STom Caputi * the new one at the same time during the receive. 168eb633035STom Caputi */ 169eb633035STom Caputi if ((!encrypted && raw) || encrypted) 170eb633035STom Caputi return (SET_ERROR(EINVAL)); 171eb633035STom Caputi 172eb633035STom Caputi /* 173eb633035STom Caputi * Perform the same encryption checks we would if 174eb633035STom Caputi * we were creating a new dataset from scratch. 175eb633035STom Caputi */ 176eb633035STom Caputi if (!raw) { 177eb633035STom Caputi boolean_t will_encrypt; 178eb633035STom Caputi 179eb633035STom Caputi error = dmu_objset_create_crypt_check( 180eb633035STom Caputi ds->ds_dir->dd_parent, drba->drba_dcp, 181eb633035STom Caputi &will_encrypt); 182eb633035STom Caputi if (error != 0) 183eb633035STom Caputi return (error); 184eb633035STom Caputi 185eb633035STom Caputi if (will_encrypt && embed) 186eb633035STom Caputi return (SET_ERROR(EINVAL)); 187eb633035STom Caputi } 188eb633035STom Caputi 189eb633035STom Caputi drba->drba_cookie->drc_fromsnapobj = 0; 1900fa1b3ccSPaul Dagnelie } 1910fa1b3ccSPaul Dagnelie 1920fa1b3ccSPaul Dagnelie return (0); 1930fa1b3ccSPaul Dagnelie 1940fa1b3ccSPaul Dagnelie } 1950fa1b3ccSPaul Dagnelie 1960fa1b3ccSPaul Dagnelie static int 1970fa1b3ccSPaul Dagnelie dmu_recv_begin_check(void *arg, dmu_tx_t *tx) 1980fa1b3ccSPaul Dagnelie { 1990fa1b3ccSPaul Dagnelie dmu_recv_begin_arg_t *drba = arg; 2000fa1b3ccSPaul Dagnelie dsl_pool_t *dp = dmu_tx_pool(tx); 2010fa1b3ccSPaul Dagnelie struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 2020fa1b3ccSPaul Dagnelie uint64_t fromguid = drrb->drr_fromguid; 2030fa1b3ccSPaul Dagnelie int flags = drrb->drr_flags; 204eb633035STom Caputi ds_hold_flags_t dsflags = 0; 2050fa1b3ccSPaul Dagnelie int error; 2060fa1b3ccSPaul Dagnelie uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 2070fa1b3ccSPaul Dagnelie dsl_dataset_t *ds; 2080fa1b3ccSPaul Dagnelie const char *tofs = drba->drba_cookie->drc_tofs; 2090fa1b3ccSPaul Dagnelie 2100fa1b3ccSPaul Dagnelie /* already checked */ 2110fa1b3ccSPaul Dagnelie ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 2120fa1b3ccSPaul Dagnelie ASSERT(!(featureflags & DMU_BACKUP_FEATURE_RESUMING)); 2130fa1b3ccSPaul Dagnelie 2140fa1b3ccSPaul Dagnelie if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 2150fa1b3ccSPaul Dagnelie DMU_COMPOUNDSTREAM || 2160fa1b3ccSPaul Dagnelie drrb->drr_type >= DMU_OST_NUMTYPES || 2170fa1b3ccSPaul Dagnelie ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) 2180fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 2190fa1b3ccSPaul Dagnelie 2200fa1b3ccSPaul Dagnelie /* Verify pool version supports SA if SA_SPILL feature set */ 2210fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && 2220fa1b3ccSPaul Dagnelie spa_version(dp->dp_spa) < SPA_VERSION_SA) 2230fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 2240fa1b3ccSPaul Dagnelie 2250fa1b3ccSPaul Dagnelie if (drba->drba_cookie->drc_resumable && 2260fa1b3ccSPaul Dagnelie !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EXTENSIBLE_DATASET)) 2270fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 2280fa1b3ccSPaul Dagnelie 2290fa1b3ccSPaul Dagnelie /* 2300fa1b3ccSPaul Dagnelie * The receiving code doesn't know how to translate a WRITE_EMBEDDED 2310fa1b3ccSPaul Dagnelie * record to a plain WRITE record, so the pool must have the 2320fa1b3ccSPaul Dagnelie * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED 2330fa1b3ccSPaul Dagnelie * records. Same with WRITE_EMBEDDED records that use LZ4 compression. 2340fa1b3ccSPaul Dagnelie */ 2350fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && 2360fa1b3ccSPaul Dagnelie !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) 2370fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 2380fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && 2390fa1b3ccSPaul Dagnelie !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 2400fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 2410fa1b3ccSPaul Dagnelie 2420fa1b3ccSPaul Dagnelie /* 2430fa1b3ccSPaul Dagnelie * The receiving code doesn't know how to translate large blocks 2440fa1b3ccSPaul Dagnelie * to smaller ones, so the pool must have the LARGE_BLOCKS 2450fa1b3ccSPaul Dagnelie * feature enabled if the stream has LARGE_BLOCKS. Same with 2460fa1b3ccSPaul Dagnelie * large dnodes. 2470fa1b3ccSPaul Dagnelie */ 2480fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 2490fa1b3ccSPaul Dagnelie !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 2500fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 2510fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) && 2520fa1b3ccSPaul Dagnelie !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_DNODE)) 2530fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 2540fa1b3ccSPaul Dagnelie 255*6ccda740Sloli if (featureflags & DMU_BACKUP_FEATURE_RAW) { 256eb633035STom Caputi /* raw receives require the encryption feature */ 257eb633035STom Caputi if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ENCRYPTION)) 258eb633035STom Caputi return (SET_ERROR(ENOTSUP)); 259eb633035STom Caputi 260eb633035STom Caputi /* embedded data is incompatible with encryption and raw recv */ 261eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) 262eb633035STom Caputi return (SET_ERROR(EINVAL)); 263eb633035STom Caputi 264eb633035STom Caputi /* raw receives require spill block allocation flag */ 265eb633035STom Caputi if (!(flags & DRR_FLAG_SPILL_BLOCK)) 266eb633035STom Caputi return (SET_ERROR(ZFS_ERR_SPILL_BLOCK_FLAG_MISSING)); 267eb633035STom Caputi } else { 268eb633035STom Caputi dsflags |= DS_HOLD_FLAG_DECRYPT; 269eb633035STom Caputi } 270eb633035STom Caputi 271eb633035STom Caputi error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); 2720fa1b3ccSPaul Dagnelie if (error == 0) { 2730fa1b3ccSPaul Dagnelie /* target fs already exists; recv into temp clone */ 2740fa1b3ccSPaul Dagnelie 2750fa1b3ccSPaul Dagnelie /* Can't recv a clone into an existing fs */ 2760fa1b3ccSPaul Dagnelie if (flags & DRR_FLAG_CLONE || drba->drba_origin) { 277eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 2780fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 2790fa1b3ccSPaul Dagnelie } 2800fa1b3ccSPaul Dagnelie 281*6ccda740Sloli error = recv_begin_check_existing_impl(drba, ds, fromguid, 282*6ccda740Sloli featureflags); 283eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 2840fa1b3ccSPaul Dagnelie } else if (error == ENOENT) { 2850fa1b3ccSPaul Dagnelie /* target fs does not exist; must be a full backup or clone */ 2860fa1b3ccSPaul Dagnelie char buf[ZFS_MAX_DATASET_NAME_LEN]; 2870fa1b3ccSPaul Dagnelie 2880fa1b3ccSPaul Dagnelie /* 2890fa1b3ccSPaul Dagnelie * If it's a non-clone incremental, we are missing the 2900fa1b3ccSPaul Dagnelie * target fs, so fail the recv. 2910fa1b3ccSPaul Dagnelie */ 2920fa1b3ccSPaul Dagnelie if (fromguid != 0 && !(flags & DRR_FLAG_CLONE || 2930fa1b3ccSPaul Dagnelie drba->drba_origin)) 2940fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOENT)); 2950fa1b3ccSPaul Dagnelie 2960fa1b3ccSPaul Dagnelie /* 2970fa1b3ccSPaul Dagnelie * If we're receiving a full send as a clone, and it doesn't 2980fa1b3ccSPaul Dagnelie * contain all the necessary free records and freeobject 2990fa1b3ccSPaul Dagnelie * records, reject it. 3000fa1b3ccSPaul Dagnelie */ 3010fa1b3ccSPaul Dagnelie if (fromguid == 0 && drba->drba_origin && 3020fa1b3ccSPaul Dagnelie !(flags & DRR_FLAG_FREERECORDS)) 3030fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 3040fa1b3ccSPaul Dagnelie 3050fa1b3ccSPaul Dagnelie /* Open the parent of tofs */ 3060fa1b3ccSPaul Dagnelie ASSERT3U(strlen(tofs), <, sizeof (buf)); 3070fa1b3ccSPaul Dagnelie (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); 308a60ca23dSTom Caputi error = dsl_dataset_hold(dp, buf, FTAG, &ds); 3090fa1b3ccSPaul Dagnelie if (error != 0) 3100fa1b3ccSPaul Dagnelie return (error); 3110fa1b3ccSPaul Dagnelie 312eb633035STom Caputi if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0 && 313eb633035STom Caputi drba->drba_origin == NULL) { 314eb633035STom Caputi boolean_t will_encrypt; 315eb633035STom Caputi 316eb633035STom Caputi /* 317eb633035STom Caputi * Check that we aren't breaking any encryption rules 318eb633035STom Caputi * and that we have all the parameters we need to 319eb633035STom Caputi * create an encrypted dataset if necessary. If we are 320eb633035STom Caputi * making an encrypted dataset the stream can't have 321eb633035STom Caputi * embedded data. 322eb633035STom Caputi */ 323eb633035STom Caputi error = dmu_objset_create_crypt_check(ds->ds_dir, 324eb633035STom Caputi drba->drba_dcp, &will_encrypt); 325eb633035STom Caputi if (error != 0) { 326a60ca23dSTom Caputi dsl_dataset_rele(ds, FTAG); 327eb633035STom Caputi return (error); 328eb633035STom Caputi } 329eb633035STom Caputi 330eb633035STom Caputi if (will_encrypt && 331eb633035STom Caputi (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)) { 332a60ca23dSTom Caputi dsl_dataset_rele(ds, FTAG); 333eb633035STom Caputi return (SET_ERROR(EINVAL)); 334eb633035STom Caputi } 335eb633035STom Caputi } 336eb633035STom Caputi 3370fa1b3ccSPaul Dagnelie /* 3380fa1b3ccSPaul Dagnelie * Check filesystem and snapshot limits before receiving. We'll 3390fa1b3ccSPaul Dagnelie * recheck snapshot limits again at the end (we create the 3400fa1b3ccSPaul Dagnelie * filesystems and increment those counts during begin_sync). 3410fa1b3ccSPaul Dagnelie */ 3420fa1b3ccSPaul Dagnelie error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 3430fa1b3ccSPaul Dagnelie ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); 3440fa1b3ccSPaul Dagnelie if (error != 0) { 345a60ca23dSTom Caputi dsl_dataset_rele(ds, FTAG); 3460fa1b3ccSPaul Dagnelie return (error); 3470fa1b3ccSPaul Dagnelie } 3480fa1b3ccSPaul Dagnelie 3490fa1b3ccSPaul Dagnelie error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 3500fa1b3ccSPaul Dagnelie ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); 3510fa1b3ccSPaul Dagnelie if (error != 0) { 352a60ca23dSTom Caputi dsl_dataset_rele(ds, FTAG); 3530fa1b3ccSPaul Dagnelie return (error); 3540fa1b3ccSPaul Dagnelie } 3550fa1b3ccSPaul Dagnelie 3560fa1b3ccSPaul Dagnelie if (drba->drba_origin != NULL) { 3570fa1b3ccSPaul Dagnelie dsl_dataset_t *origin; 358eb633035STom Caputi 359a60ca23dSTom Caputi error = dsl_dataset_hold(dp, drba->drba_origin, 360a60ca23dSTom Caputi FTAG, &origin); 3610fa1b3ccSPaul Dagnelie if (error != 0) { 362a60ca23dSTom Caputi dsl_dataset_rele(ds, FTAG); 3630fa1b3ccSPaul Dagnelie return (error); 3640fa1b3ccSPaul Dagnelie } 3650fa1b3ccSPaul Dagnelie if (!origin->ds_is_snapshot) { 366a60ca23dSTom Caputi dsl_dataset_rele(origin, FTAG); 367a60ca23dSTom Caputi dsl_dataset_rele(ds, FTAG); 3680fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 3690fa1b3ccSPaul Dagnelie } 3700fa1b3ccSPaul Dagnelie if (dsl_dataset_phys(origin)->ds_guid != fromguid && 3710fa1b3ccSPaul Dagnelie fromguid != 0) { 372a60ca23dSTom Caputi dsl_dataset_rele(origin, FTAG); 373a60ca23dSTom Caputi dsl_dataset_rele(ds, FTAG); 3740fa1b3ccSPaul Dagnelie return (SET_ERROR(ENODEV)); 3750fa1b3ccSPaul Dagnelie } 376eb633035STom Caputi if (origin->ds_dir->dd_crypto_obj != 0 && 377eb633035STom Caputi (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)) { 378a60ca23dSTom Caputi dsl_dataset_rele(origin, FTAG); 379a60ca23dSTom Caputi dsl_dataset_rele(ds, FTAG); 380eb633035STom Caputi return (SET_ERROR(EINVAL)); 381eb633035STom Caputi } 382a60ca23dSTom Caputi dsl_dataset_rele(origin, FTAG); 3830fa1b3ccSPaul Dagnelie } 384a60ca23dSTom Caputi dsl_dataset_rele(ds, FTAG); 3850fa1b3ccSPaul Dagnelie error = 0; 3860fa1b3ccSPaul Dagnelie } 3870fa1b3ccSPaul Dagnelie return (error); 3880fa1b3ccSPaul Dagnelie } 3890fa1b3ccSPaul Dagnelie 3900fa1b3ccSPaul Dagnelie static void 3910fa1b3ccSPaul Dagnelie dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) 3920fa1b3ccSPaul Dagnelie { 3930fa1b3ccSPaul Dagnelie dmu_recv_begin_arg_t *drba = arg; 3940fa1b3ccSPaul Dagnelie dsl_pool_t *dp = dmu_tx_pool(tx); 3950fa1b3ccSPaul Dagnelie objset_t *mos = dp->dp_meta_objset; 3960fa1b3ccSPaul Dagnelie struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 3970fa1b3ccSPaul Dagnelie const char *tofs = drba->drba_cookie->drc_tofs; 398eb633035STom Caputi uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 3990fa1b3ccSPaul Dagnelie dsl_dataset_t *ds, *newds; 400eb633035STom Caputi objset_t *os; 4010fa1b3ccSPaul Dagnelie uint64_t dsobj; 402eb633035STom Caputi ds_hold_flags_t dsflags = 0; 4030fa1b3ccSPaul Dagnelie int error; 4040fa1b3ccSPaul Dagnelie uint64_t crflags = 0; 405eb633035STom Caputi dsl_crypto_params_t dummy_dcp = { 0 }; 406eb633035STom Caputi dsl_crypto_params_t *dcp = drba->drba_dcp; 4070fa1b3ccSPaul Dagnelie 4080fa1b3ccSPaul Dagnelie if (drrb->drr_flags & DRR_FLAG_CI_DATA) 4090fa1b3ccSPaul Dagnelie crflags |= DS_FLAG_CI_DATASET; 410*6ccda740Sloli 411eb633035STom Caputi if ((featureflags & DMU_BACKUP_FEATURE_RAW) == 0) 412eb633035STom Caputi dsflags |= DS_HOLD_FLAG_DECRYPT; 413eb633035STom Caputi 414eb633035STom Caputi /* 415eb633035STom Caputi * Raw, non-incremental recvs always use a dummy dcp with 416eb633035STom Caputi * the raw cmd set. Raw incremental recvs do not use a dcp 417eb633035STom Caputi * since the encryption parameters are already set in stone. 418eb633035STom Caputi */ 419eb633035STom Caputi if (dcp == NULL && drba->drba_cookie->drc_fromsnapobj == 0 && 420eb633035STom Caputi drba->drba_origin == NULL) { 421eb633035STom Caputi ASSERT3P(dcp, ==, NULL); 422eb633035STom Caputi dcp = &dummy_dcp; 4230fa1b3ccSPaul Dagnelie 424eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_RAW) 425eb633035STom Caputi dcp->cp_cmd = DCP_CMD_RAW_RECV; 426eb633035STom Caputi } 427eb633035STom Caputi 428eb633035STom Caputi error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); 4290fa1b3ccSPaul Dagnelie if (error == 0) { 4300fa1b3ccSPaul Dagnelie /* create temporary clone */ 4310fa1b3ccSPaul Dagnelie dsl_dataset_t *snap = NULL; 432eb633035STom Caputi 433eb633035STom Caputi if (drba->drba_cookie->drc_fromsnapobj != 0) { 4340fa1b3ccSPaul Dagnelie VERIFY0(dsl_dataset_hold_obj(dp, 435eb633035STom Caputi drba->drba_cookie->drc_fromsnapobj, FTAG, &snap)); 436eb633035STom Caputi ASSERT3P(dcp, ==, NULL); 4370fa1b3ccSPaul Dagnelie } 438eb633035STom Caputi 4390fa1b3ccSPaul Dagnelie dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, 440eb633035STom Caputi snap, crflags, drba->drba_cred, dcp, tx); 441eb633035STom Caputi if (drba->drba_cookie->drc_fromsnapobj != 0) 4420fa1b3ccSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 443eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 4440fa1b3ccSPaul Dagnelie } else { 4450fa1b3ccSPaul Dagnelie dsl_dir_t *dd; 4460fa1b3ccSPaul Dagnelie const char *tail; 4470fa1b3ccSPaul Dagnelie dsl_dataset_t *origin = NULL; 4480fa1b3ccSPaul Dagnelie 4490fa1b3ccSPaul Dagnelie VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); 4500fa1b3ccSPaul Dagnelie 4510fa1b3ccSPaul Dagnelie if (drba->drba_origin != NULL) { 4520fa1b3ccSPaul Dagnelie VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, 4530fa1b3ccSPaul Dagnelie FTAG, &origin)); 454eb633035STom Caputi ASSERT3P(dcp, ==, NULL); 4550fa1b3ccSPaul Dagnelie } 4560fa1b3ccSPaul Dagnelie 4570fa1b3ccSPaul Dagnelie /* Create new dataset. */ 458eb633035STom Caputi dsobj = dsl_dataset_create_sync(dd, strrchr(tofs, '/') + 1, 459eb633035STom Caputi origin, crflags, drba->drba_cred, dcp, tx); 4600fa1b3ccSPaul Dagnelie if (origin != NULL) 4610fa1b3ccSPaul Dagnelie dsl_dataset_rele(origin, FTAG); 4620fa1b3ccSPaul Dagnelie dsl_dir_rele(dd, FTAG); 4630fa1b3ccSPaul Dagnelie drba->drba_cookie->drc_newfs = B_TRUE; 4640fa1b3ccSPaul Dagnelie } 465eb633035STom Caputi 466eb633035STom Caputi VERIFY0(dsl_dataset_own_obj(dp, dsobj, dsflags, dmu_recv_tag, &newds)); 467eb633035STom Caputi VERIFY0(dmu_objset_from_ds(newds, &os)); 4680fa1b3ccSPaul Dagnelie 4690fa1b3ccSPaul Dagnelie if (drba->drba_cookie->drc_resumable) { 4700fa1b3ccSPaul Dagnelie dsl_dataset_zapify(newds, tx); 4710fa1b3ccSPaul Dagnelie if (drrb->drr_fromguid != 0) { 4720fa1b3ccSPaul Dagnelie VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_FROMGUID, 4730fa1b3ccSPaul Dagnelie 8, 1, &drrb->drr_fromguid, tx)); 4740fa1b3ccSPaul Dagnelie } 4750fa1b3ccSPaul Dagnelie VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TOGUID, 4760fa1b3ccSPaul Dagnelie 8, 1, &drrb->drr_toguid, tx)); 4770fa1b3ccSPaul Dagnelie VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TONAME, 4780fa1b3ccSPaul Dagnelie 1, strlen(drrb->drr_toname) + 1, drrb->drr_toname, tx)); 4790fa1b3ccSPaul Dagnelie uint64_t one = 1; 4800fa1b3ccSPaul Dagnelie uint64_t zero = 0; 4810fa1b3ccSPaul Dagnelie VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OBJECT, 4820fa1b3ccSPaul Dagnelie 8, 1, &one, tx)); 4830fa1b3ccSPaul Dagnelie VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OFFSET, 4840fa1b3ccSPaul Dagnelie 8, 1, &zero, tx)); 4850fa1b3ccSPaul Dagnelie VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_BYTES, 4860fa1b3ccSPaul Dagnelie 8, 1, &zero, tx)); 487eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) { 4880fa1b3ccSPaul Dagnelie VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_LARGEBLOCK, 4890fa1b3ccSPaul Dagnelie 8, 1, &one, tx)); 4900fa1b3ccSPaul Dagnelie } 491eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) { 4920fa1b3ccSPaul Dagnelie VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_EMBEDOK, 4930fa1b3ccSPaul Dagnelie 8, 1, &one, tx)); 4940fa1b3ccSPaul Dagnelie } 495eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_COMPRESSED) { 4960fa1b3ccSPaul Dagnelie VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_COMPRESSOK, 4970fa1b3ccSPaul Dagnelie 8, 1, &one, tx)); 4980fa1b3ccSPaul Dagnelie } 499eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_RAW) { 500eb633035STom Caputi VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_RAWOK, 501eb633035STom Caputi 8, 1, &one, tx)); 502eb633035STom Caputi } 503eb633035STom Caputi } 504eb633035STom Caputi 505eb633035STom Caputi /* 506eb633035STom Caputi * Usually the os->os_encrypted value is tied to the presence of a 507eb633035STom Caputi * DSL Crypto Key object in the dd. However, that will not be received 508eb633035STom Caputi * until dmu_recv_stream(), so we set the value manually for now. 509eb633035STom Caputi */ 510eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_RAW) { 511eb633035STom Caputi os->os_encrypted = B_TRUE; 512eb633035STom Caputi drba->drba_cookie->drc_raw = B_TRUE; 5130fa1b3ccSPaul Dagnelie } 5140fa1b3ccSPaul Dagnelie 5150fa1b3ccSPaul Dagnelie dmu_buf_will_dirty(newds->ds_dbuf, tx); 5160fa1b3ccSPaul Dagnelie dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; 5170fa1b3ccSPaul Dagnelie 5180fa1b3ccSPaul Dagnelie /* 519eb633035STom Caputi * If we actually created a non-clone, we need to create the objset 520eb633035STom Caputi * in our new dataset. If this is a raw send we postpone this until 521eb633035STom Caputi * dmu_recv_stream() so that we can allocate the metadnode with the 522eb633035STom Caputi * properties from the DRR_BEGIN payload. 5230fa1b3ccSPaul Dagnelie */ 5240fa1b3ccSPaul Dagnelie rrw_enter(&newds->ds_bp_rwlock, RW_READER, FTAG); 525eb633035STom Caputi if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds)) && 526eb633035STom Caputi (featureflags & DMU_BACKUP_FEATURE_RAW) == 0) { 5270fa1b3ccSPaul Dagnelie (void) dmu_objset_create_impl(dp->dp_spa, 5280fa1b3ccSPaul Dagnelie newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); 5290fa1b3ccSPaul Dagnelie } 5300fa1b3ccSPaul Dagnelie rrw_exit(&newds->ds_bp_rwlock, FTAG); 5310fa1b3ccSPaul Dagnelie 5320fa1b3ccSPaul Dagnelie drba->drba_cookie->drc_ds = newds; 5330fa1b3ccSPaul Dagnelie 5340fa1b3ccSPaul Dagnelie spa_history_log_internal_ds(newds, "receive", tx, ""); 5350fa1b3ccSPaul Dagnelie } 5360fa1b3ccSPaul Dagnelie 5370fa1b3ccSPaul Dagnelie static int 5380fa1b3ccSPaul Dagnelie dmu_recv_resume_begin_check(void *arg, dmu_tx_t *tx) 5390fa1b3ccSPaul Dagnelie { 5400fa1b3ccSPaul Dagnelie dmu_recv_begin_arg_t *drba = arg; 5410fa1b3ccSPaul Dagnelie dsl_pool_t *dp = dmu_tx_pool(tx); 5420fa1b3ccSPaul Dagnelie struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 5430fa1b3ccSPaul Dagnelie int error; 544eb633035STom Caputi ds_hold_flags_t dsflags = 0; 5450fa1b3ccSPaul Dagnelie uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 5460fa1b3ccSPaul Dagnelie dsl_dataset_t *ds; 5470fa1b3ccSPaul Dagnelie const char *tofs = drba->drba_cookie->drc_tofs; 5480fa1b3ccSPaul Dagnelie 5490fa1b3ccSPaul Dagnelie /* 6 extra bytes for /%recv */ 5500fa1b3ccSPaul Dagnelie char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 5510fa1b3ccSPaul Dagnelie 5520fa1b3ccSPaul Dagnelie /* already checked */ 5530fa1b3ccSPaul Dagnelie ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 5540fa1b3ccSPaul Dagnelie ASSERT(featureflags & DMU_BACKUP_FEATURE_RESUMING); 5550fa1b3ccSPaul Dagnelie 5560fa1b3ccSPaul Dagnelie if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 5570fa1b3ccSPaul Dagnelie DMU_COMPOUNDSTREAM || 5580fa1b3ccSPaul Dagnelie drrb->drr_type >= DMU_OST_NUMTYPES) 5590fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 5600fa1b3ccSPaul Dagnelie 5610fa1b3ccSPaul Dagnelie /* Verify pool version supports SA if SA_SPILL feature set */ 5620fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && 5630fa1b3ccSPaul Dagnelie spa_version(dp->dp_spa) < SPA_VERSION_SA) 5640fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 5650fa1b3ccSPaul Dagnelie 5660fa1b3ccSPaul Dagnelie /* 5670fa1b3ccSPaul Dagnelie * The receiving code doesn't know how to translate a WRITE_EMBEDDED 5680fa1b3ccSPaul Dagnelie * record to a plain WRITE record, so the pool must have the 5690fa1b3ccSPaul Dagnelie * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED 5700fa1b3ccSPaul Dagnelie * records. Same with WRITE_EMBEDDED records that use LZ4 compression. 5710fa1b3ccSPaul Dagnelie */ 5720fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && 5730fa1b3ccSPaul Dagnelie !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) 5740fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 5750fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && 5760fa1b3ccSPaul Dagnelie !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 5770fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 5780fa1b3ccSPaul Dagnelie 5790fa1b3ccSPaul Dagnelie /* 5800fa1b3ccSPaul Dagnelie * The receiving code doesn't know how to translate large blocks 5810fa1b3ccSPaul Dagnelie * to smaller ones, so the pool must have the LARGE_BLOCKS 5820fa1b3ccSPaul Dagnelie * feature enabled if the stream has LARGE_BLOCKS. Same with 5830fa1b3ccSPaul Dagnelie * large dnodes. 5840fa1b3ccSPaul Dagnelie */ 5850fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 5860fa1b3ccSPaul Dagnelie !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 5870fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 5880fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) && 5890fa1b3ccSPaul Dagnelie !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_DNODE)) 5900fa1b3ccSPaul Dagnelie return (SET_ERROR(ENOTSUP)); 5910fa1b3ccSPaul Dagnelie 5920fa1b3ccSPaul Dagnelie (void) snprintf(recvname, sizeof (recvname), "%s/%s", 5930fa1b3ccSPaul Dagnelie tofs, recv_clone_name); 5940fa1b3ccSPaul Dagnelie 595eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_RAW) { 596eb633035STom Caputi /* raw receives require spill block allocation flag */ 597eb633035STom Caputi if (!(drrb->drr_flags & DRR_FLAG_SPILL_BLOCK)) 598eb633035STom Caputi return (SET_ERROR(ZFS_ERR_SPILL_BLOCK_FLAG_MISSING)); 599eb633035STom Caputi } else { 600eb633035STom Caputi dsflags |= DS_HOLD_FLAG_DECRYPT; 601eb633035STom Caputi } 602eb633035STom Caputi 603eb633035STom Caputi if (dsl_dataset_hold_flags(dp, recvname, dsflags, FTAG, &ds) != 0) { 6040fa1b3ccSPaul Dagnelie /* %recv does not exist; continue in tofs */ 605eb633035STom Caputi error = dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds); 6060fa1b3ccSPaul Dagnelie if (error != 0) 6070fa1b3ccSPaul Dagnelie return (error); 6080fa1b3ccSPaul Dagnelie } 6090fa1b3ccSPaul Dagnelie 6100fa1b3ccSPaul Dagnelie /* check that ds is marked inconsistent */ 6110fa1b3ccSPaul Dagnelie if (!DS_IS_INCONSISTENT(ds)) { 612eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 6130fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 6140fa1b3ccSPaul Dagnelie } 6150fa1b3ccSPaul Dagnelie 6160fa1b3ccSPaul Dagnelie /* check that there is resuming data, and that the toguid matches */ 6170fa1b3ccSPaul Dagnelie if (!dsl_dataset_is_zapified(ds)) { 618eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 6190fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 6200fa1b3ccSPaul Dagnelie } 6210fa1b3ccSPaul Dagnelie uint64_t val; 6220fa1b3ccSPaul Dagnelie error = zap_lookup(dp->dp_meta_objset, ds->ds_object, 6230fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val); 6240fa1b3ccSPaul Dagnelie if (error != 0 || drrb->drr_toguid != val) { 625eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 6260fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 6270fa1b3ccSPaul Dagnelie } 6280fa1b3ccSPaul Dagnelie 6290fa1b3ccSPaul Dagnelie /* 6300fa1b3ccSPaul Dagnelie * Check if the receive is still running. If so, it will be owned. 6310fa1b3ccSPaul Dagnelie * Note that nothing else can own the dataset (e.g. after the receive 6320fa1b3ccSPaul Dagnelie * fails) because it will be marked inconsistent. 6330fa1b3ccSPaul Dagnelie */ 6340fa1b3ccSPaul Dagnelie if (dsl_dataset_has_owner(ds)) { 635eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 6360fa1b3ccSPaul Dagnelie return (SET_ERROR(EBUSY)); 6370fa1b3ccSPaul Dagnelie } 6380fa1b3ccSPaul Dagnelie 6390fa1b3ccSPaul Dagnelie /* There should not be any snapshots of this fs yet. */ 6400fa1b3ccSPaul Dagnelie if (ds->ds_prev != NULL && ds->ds_prev->ds_dir == ds->ds_dir) { 641eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 6420fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 6430fa1b3ccSPaul Dagnelie } 6440fa1b3ccSPaul Dagnelie 6450fa1b3ccSPaul Dagnelie /* 6460fa1b3ccSPaul Dagnelie * Note: resume point will be checked when we process the first WRITE 6470fa1b3ccSPaul Dagnelie * record. 6480fa1b3ccSPaul Dagnelie */ 6490fa1b3ccSPaul Dagnelie 6500fa1b3ccSPaul Dagnelie /* check that the origin matches */ 6510fa1b3ccSPaul Dagnelie val = 0; 6520fa1b3ccSPaul Dagnelie (void) zap_lookup(dp->dp_meta_objset, ds->ds_object, 6530fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val); 6540fa1b3ccSPaul Dagnelie if (drrb->drr_fromguid != val) { 655eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 6560fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 6570fa1b3ccSPaul Dagnelie } 6580fa1b3ccSPaul Dagnelie 659eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 6600fa1b3ccSPaul Dagnelie return (0); 6610fa1b3ccSPaul Dagnelie } 6620fa1b3ccSPaul Dagnelie 6630fa1b3ccSPaul Dagnelie static void 6640fa1b3ccSPaul Dagnelie dmu_recv_resume_begin_sync(void *arg, dmu_tx_t *tx) 6650fa1b3ccSPaul Dagnelie { 6660fa1b3ccSPaul Dagnelie dmu_recv_begin_arg_t *drba = arg; 6670fa1b3ccSPaul Dagnelie dsl_pool_t *dp = dmu_tx_pool(tx); 6680fa1b3ccSPaul Dagnelie const char *tofs = drba->drba_cookie->drc_tofs; 669eb633035STom Caputi struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 670eb633035STom Caputi uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 6710fa1b3ccSPaul Dagnelie dsl_dataset_t *ds; 672eb633035STom Caputi objset_t *os; 673eb633035STom Caputi ds_hold_flags_t dsflags = 0; 6740fa1b3ccSPaul Dagnelie uint64_t dsobj; 6750fa1b3ccSPaul Dagnelie /* 6 extra bytes for /%recv */ 6760fa1b3ccSPaul Dagnelie char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 6770fa1b3ccSPaul Dagnelie 6780fa1b3ccSPaul Dagnelie (void) snprintf(recvname, sizeof (recvname), "%s/%s", 6790fa1b3ccSPaul Dagnelie tofs, recv_clone_name); 6800fa1b3ccSPaul Dagnelie 681eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_RAW) { 682eb633035STom Caputi drba->drba_cookie->drc_raw = B_TRUE; 683eb633035STom Caputi } else { 684eb633035STom Caputi dsflags |= DS_HOLD_FLAG_DECRYPT; 685eb633035STom Caputi } 686eb633035STom Caputi 687eb633035STom Caputi if (dsl_dataset_hold_flags(dp, recvname, dsflags, FTAG, &ds) != 0) { 6880fa1b3ccSPaul Dagnelie /* %recv does not exist; continue in tofs */ 689eb633035STom Caputi VERIFY0(dsl_dataset_hold_flags(dp, tofs, dsflags, FTAG, &ds)); 6900fa1b3ccSPaul Dagnelie drba->drba_cookie->drc_newfs = B_TRUE; 6910fa1b3ccSPaul Dagnelie } 6920fa1b3ccSPaul Dagnelie 6930fa1b3ccSPaul Dagnelie /* clear the inconsistent flag so that we can own it */ 6940fa1b3ccSPaul Dagnelie ASSERT(DS_IS_INCONSISTENT(ds)); 6950fa1b3ccSPaul Dagnelie dmu_buf_will_dirty(ds->ds_dbuf, tx); 6960fa1b3ccSPaul Dagnelie dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; 6970fa1b3ccSPaul Dagnelie dsobj = ds->ds_object; 698eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 6990fa1b3ccSPaul Dagnelie 700eb633035STom Caputi VERIFY0(dsl_dataset_own_obj(dp, dsobj, dsflags, dmu_recv_tag, &ds)); 701eb633035STom Caputi VERIFY0(dmu_objset_from_ds(ds, &os)); 7020fa1b3ccSPaul Dagnelie 7030fa1b3ccSPaul Dagnelie dmu_buf_will_dirty(ds->ds_dbuf, tx); 7040fa1b3ccSPaul Dagnelie dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT; 7050fa1b3ccSPaul Dagnelie 7060fa1b3ccSPaul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 707eb633035STom Caputi ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds)) || 708eb633035STom Caputi drba->drba_cookie->drc_raw); 7090fa1b3ccSPaul Dagnelie rrw_exit(&ds->ds_bp_rwlock, FTAG); 7100fa1b3ccSPaul Dagnelie 7110fa1b3ccSPaul Dagnelie drba->drba_cookie->drc_ds = ds; 7120fa1b3ccSPaul Dagnelie 7130fa1b3ccSPaul Dagnelie spa_history_log_internal_ds(ds, "resume receive", tx, ""); 7140fa1b3ccSPaul Dagnelie } 7150fa1b3ccSPaul Dagnelie 7160fa1b3ccSPaul Dagnelie /* 7170fa1b3ccSPaul Dagnelie * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() 7180fa1b3ccSPaul Dagnelie * succeeds; otherwise we will leak the holds on the datasets. 7190fa1b3ccSPaul Dagnelie */ 7200fa1b3ccSPaul Dagnelie int 7210fa1b3ccSPaul Dagnelie dmu_recv_begin(char *tofs, char *tosnap, dmu_replay_record_t *drr_begin, 722*6ccda740Sloli boolean_t force, boolean_t resumable, nvlist_t *localprops, 723*6ccda740Sloli nvlist_t *hidden_args, char *origin, dmu_recv_cookie_t *drc) 7240fa1b3ccSPaul Dagnelie { 7250fa1b3ccSPaul Dagnelie dmu_recv_begin_arg_t drba = { 0 }; 7260fa1b3ccSPaul Dagnelie 7270fa1b3ccSPaul Dagnelie bzero(drc, sizeof (dmu_recv_cookie_t)); 7280fa1b3ccSPaul Dagnelie drc->drc_drr_begin = drr_begin; 7290fa1b3ccSPaul Dagnelie drc->drc_drrb = &drr_begin->drr_u.drr_begin; 7300fa1b3ccSPaul Dagnelie drc->drc_tosnap = tosnap; 7310fa1b3ccSPaul Dagnelie drc->drc_tofs = tofs; 7320fa1b3ccSPaul Dagnelie drc->drc_force = force; 7330fa1b3ccSPaul Dagnelie drc->drc_resumable = resumable; 7340fa1b3ccSPaul Dagnelie drc->drc_cred = CRED(); 7350fa1b3ccSPaul Dagnelie drc->drc_clone = (origin != NULL); 7360fa1b3ccSPaul Dagnelie 7370fa1b3ccSPaul Dagnelie if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { 7380fa1b3ccSPaul Dagnelie drc->drc_byteswap = B_TRUE; 7390fa1b3ccSPaul Dagnelie (void) fletcher_4_incremental_byteswap(drr_begin, 7400fa1b3ccSPaul Dagnelie sizeof (dmu_replay_record_t), &drc->drc_cksum); 7410fa1b3ccSPaul Dagnelie byteswap_record(drr_begin); 7420fa1b3ccSPaul Dagnelie } else if (drc->drc_drrb->drr_magic == DMU_BACKUP_MAGIC) { 7430fa1b3ccSPaul Dagnelie (void) fletcher_4_incremental_native(drr_begin, 7440fa1b3ccSPaul Dagnelie sizeof (dmu_replay_record_t), &drc->drc_cksum); 7450fa1b3ccSPaul Dagnelie } else { 7460fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 7470fa1b3ccSPaul Dagnelie } 7480fa1b3ccSPaul Dagnelie 749eb633035STom Caputi if (drc->drc_drrb->drr_flags & DRR_FLAG_SPILL_BLOCK) 750eb633035STom Caputi drc->drc_spill = B_TRUE; 751eb633035STom Caputi 7520fa1b3ccSPaul Dagnelie drba.drba_origin = origin; 7530fa1b3ccSPaul Dagnelie drba.drba_cookie = drc; 7540fa1b3ccSPaul Dagnelie drba.drba_cred = CRED(); 7550fa1b3ccSPaul Dagnelie 7560fa1b3ccSPaul Dagnelie if (DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & 7570fa1b3ccSPaul Dagnelie DMU_BACKUP_FEATURE_RESUMING) { 7580fa1b3ccSPaul Dagnelie return (dsl_sync_task(tofs, 7590fa1b3ccSPaul Dagnelie dmu_recv_resume_begin_check, dmu_recv_resume_begin_sync, 7600fa1b3ccSPaul Dagnelie &drba, 5, ZFS_SPACE_CHECK_NORMAL)); 7610fa1b3ccSPaul Dagnelie } else { 762*6ccda740Sloli int err; 763*6ccda740Sloli 764*6ccda740Sloli /* 765*6ccda740Sloli * For non-raw, non-incremental, non-resuming receives the 766*6ccda740Sloli * user can specify encryption parameters on the command line 767*6ccda740Sloli * with "zfs recv -o". For these receives we create a dcp and 768*6ccda740Sloli * pass it to the sync task. Creating the dcp will implicitly 769*6ccda740Sloli * remove the encryption params from the localprops nvlist, 770*6ccda740Sloli * which avoids errors when trying to set these normally 771*6ccda740Sloli * read-only properties. Any other kind of receive that 772*6ccda740Sloli * attempts to set these properties will fail as a result. 773*6ccda740Sloli */ 774*6ccda740Sloli if ((DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & 775*6ccda740Sloli DMU_BACKUP_FEATURE_RAW) == 0 && 776*6ccda740Sloli origin == NULL && drc->drc_drrb->drr_fromguid == 0) { 777*6ccda740Sloli err = dsl_crypto_params_create_nvlist(DCP_CMD_NONE, 778*6ccda740Sloli localprops, hidden_args, &drba.drba_dcp); 779*6ccda740Sloli if (err != 0) 780*6ccda740Sloli return (err); 781*6ccda740Sloli } 782*6ccda740Sloli 783*6ccda740Sloli err = dsl_sync_task(tofs, 7840fa1b3ccSPaul Dagnelie dmu_recv_begin_check, dmu_recv_begin_sync, 785*6ccda740Sloli &drba, 5, ZFS_SPACE_CHECK_NORMAL); 786*6ccda740Sloli dsl_crypto_params_free(drba.drba_dcp, !!err); 787*6ccda740Sloli 788*6ccda740Sloli return (err); 7890fa1b3ccSPaul Dagnelie } 7900fa1b3ccSPaul Dagnelie } 7910fa1b3ccSPaul Dagnelie 7920fa1b3ccSPaul Dagnelie struct receive_record_arg { 7930fa1b3ccSPaul Dagnelie dmu_replay_record_t header; 7940fa1b3ccSPaul Dagnelie void *payload; /* Pointer to a buffer containing the payload */ 7950fa1b3ccSPaul Dagnelie /* 7960fa1b3ccSPaul Dagnelie * If the record is a write, pointer to the arc_buf_t containing the 7970fa1b3ccSPaul Dagnelie * payload. 7980fa1b3ccSPaul Dagnelie */ 799eb633035STom Caputi arc_buf_t *arc_buf; 8000fa1b3ccSPaul Dagnelie int payload_size; 8010fa1b3ccSPaul Dagnelie uint64_t bytes_read; /* bytes read from stream when record created */ 8020fa1b3ccSPaul Dagnelie boolean_t eos_marker; /* Marks the end of the stream */ 8030fa1b3ccSPaul Dagnelie bqueue_node_t node; 8040fa1b3ccSPaul Dagnelie }; 8050fa1b3ccSPaul Dagnelie 8060fa1b3ccSPaul Dagnelie struct receive_writer_arg { 8070fa1b3ccSPaul Dagnelie objset_t *os; 8080fa1b3ccSPaul Dagnelie boolean_t byteswap; 8090fa1b3ccSPaul Dagnelie bqueue_t q; 8100fa1b3ccSPaul Dagnelie 8110fa1b3ccSPaul Dagnelie /* 8120fa1b3ccSPaul Dagnelie * These three args are used to signal to the main thread that we're 8130fa1b3ccSPaul Dagnelie * done. 8140fa1b3ccSPaul Dagnelie */ 8150fa1b3ccSPaul Dagnelie kmutex_t mutex; 8160fa1b3ccSPaul Dagnelie kcondvar_t cv; 8170fa1b3ccSPaul Dagnelie boolean_t done; 8180fa1b3ccSPaul Dagnelie 8190fa1b3ccSPaul Dagnelie int err; 8200fa1b3ccSPaul Dagnelie /* A map from guid to dataset to help handle dedup'd streams. */ 8210fa1b3ccSPaul Dagnelie avl_tree_t *guid_to_ds_map; 8220fa1b3ccSPaul Dagnelie boolean_t resumable; 823eb633035STom Caputi boolean_t raw; /* DMU_BACKUP_FEATURE_RAW set */ 824eb633035STom Caputi boolean_t spill; /* DRR_FLAG_SPILL_BLOCK set */ 8250fa1b3ccSPaul Dagnelie uint64_t last_object; 8260fa1b3ccSPaul Dagnelie uint64_t last_offset; 8270fa1b3ccSPaul Dagnelie uint64_t max_object; /* highest object ID referenced in stream */ 8280fa1b3ccSPaul Dagnelie uint64_t bytes_read; /* bytes read when current record created */ 829eb633035STom Caputi 830eb633035STom Caputi /* Encryption parameters for the last received DRR_OBJECT_RANGE */ 831eb633035STom Caputi boolean_t or_crypt_params_present; 832eb633035STom Caputi uint64_t or_firstobj; 833eb633035STom Caputi uint64_t or_numslots; 834eb633035STom Caputi uint8_t or_salt[ZIO_DATA_SALT_LEN]; 835eb633035STom Caputi uint8_t or_iv[ZIO_DATA_IV_LEN]; 836eb633035STom Caputi uint8_t or_mac[ZIO_DATA_MAC_LEN]; 837eb633035STom Caputi boolean_t or_byteorder; 8380fa1b3ccSPaul Dagnelie }; 8390fa1b3ccSPaul Dagnelie 8400fa1b3ccSPaul Dagnelie struct objlist { 8410fa1b3ccSPaul Dagnelie list_t list; /* List of struct receive_objnode. */ 8420fa1b3ccSPaul Dagnelie /* 8430fa1b3ccSPaul Dagnelie * Last object looked up. Used to assert that objects are being looked 8440fa1b3ccSPaul Dagnelie * up in ascending order. 8450fa1b3ccSPaul Dagnelie */ 8460fa1b3ccSPaul Dagnelie uint64_t last_lookup; 8470fa1b3ccSPaul Dagnelie }; 8480fa1b3ccSPaul Dagnelie 8490fa1b3ccSPaul Dagnelie struct receive_objnode { 8500fa1b3ccSPaul Dagnelie list_node_t node; 8510fa1b3ccSPaul Dagnelie uint64_t object; 8520fa1b3ccSPaul Dagnelie }; 8530fa1b3ccSPaul Dagnelie 8540fa1b3ccSPaul Dagnelie struct receive_arg { 8550fa1b3ccSPaul Dagnelie objset_t *os; 8560fa1b3ccSPaul Dagnelie vnode_t *vp; /* The vnode to read the stream from */ 8570fa1b3ccSPaul Dagnelie uint64_t voff; /* The current offset in the stream */ 8580fa1b3ccSPaul Dagnelie uint64_t bytes_read; 8590fa1b3ccSPaul Dagnelie /* 8600fa1b3ccSPaul Dagnelie * A record that has had its payload read in, but hasn't yet been handed 8610fa1b3ccSPaul Dagnelie * off to the worker thread. 8620fa1b3ccSPaul Dagnelie */ 8630fa1b3ccSPaul Dagnelie struct receive_record_arg *rrd; 8640fa1b3ccSPaul Dagnelie /* A record that has had its header read in, but not its payload. */ 8650fa1b3ccSPaul Dagnelie struct receive_record_arg *next_rrd; 8660fa1b3ccSPaul Dagnelie zio_cksum_t cksum; 8670fa1b3ccSPaul Dagnelie zio_cksum_t prev_cksum; 8680fa1b3ccSPaul Dagnelie int err; 8690fa1b3ccSPaul Dagnelie boolean_t byteswap; 870eb633035STom Caputi boolean_t raw; 871eb633035STom Caputi uint64_t featureflags; 8720fa1b3ccSPaul Dagnelie /* Sorted list of objects not to issue prefetches for. */ 8730fa1b3ccSPaul Dagnelie struct objlist ignore_objlist; 8740fa1b3ccSPaul Dagnelie }; 8750fa1b3ccSPaul Dagnelie 8760fa1b3ccSPaul Dagnelie typedef struct guid_map_entry { 8770fa1b3ccSPaul Dagnelie uint64_t guid; 878eb633035STom Caputi boolean_t raw; 8790fa1b3ccSPaul Dagnelie dsl_dataset_t *gme_ds; 8800fa1b3ccSPaul Dagnelie avl_node_t avlnode; 8810fa1b3ccSPaul Dagnelie } guid_map_entry_t; 8820fa1b3ccSPaul Dagnelie 8830fa1b3ccSPaul Dagnelie static int 8840fa1b3ccSPaul Dagnelie guid_compare(const void *arg1, const void *arg2) 8850fa1b3ccSPaul Dagnelie { 8860fa1b3ccSPaul Dagnelie const guid_map_entry_t *gmep1 = arg1; 8870fa1b3ccSPaul Dagnelie const guid_map_entry_t *gmep2 = arg2; 8880fa1b3ccSPaul Dagnelie 8890fa1b3ccSPaul Dagnelie if (gmep1->guid < gmep2->guid) 8900fa1b3ccSPaul Dagnelie return (-1); 8910fa1b3ccSPaul Dagnelie else if (gmep1->guid > gmep2->guid) 8920fa1b3ccSPaul Dagnelie return (1); 8930fa1b3ccSPaul Dagnelie return (0); 8940fa1b3ccSPaul Dagnelie } 8950fa1b3ccSPaul Dagnelie 8960fa1b3ccSPaul Dagnelie static void 8970fa1b3ccSPaul Dagnelie free_guid_map_onexit(void *arg) 8980fa1b3ccSPaul Dagnelie { 8990fa1b3ccSPaul Dagnelie avl_tree_t *ca = arg; 9000fa1b3ccSPaul Dagnelie void *cookie = NULL; 9010fa1b3ccSPaul Dagnelie guid_map_entry_t *gmep; 9020fa1b3ccSPaul Dagnelie 9030fa1b3ccSPaul Dagnelie while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { 904eb633035STom Caputi ds_hold_flags_t dsflags = DS_HOLD_FLAG_DECRYPT; 905eb633035STom Caputi 906eb633035STom Caputi if (gmep->raw) { 907eb633035STom Caputi gmep->gme_ds->ds_objset->os_raw_receive = B_FALSE; 908eb633035STom Caputi dsflags &= ~DS_HOLD_FLAG_DECRYPT; 909eb633035STom Caputi } 910eb633035STom Caputi 911eb633035STom Caputi dsl_dataset_disown(gmep->gme_ds, dsflags, gmep); 9120fa1b3ccSPaul Dagnelie kmem_free(gmep, sizeof (guid_map_entry_t)); 9130fa1b3ccSPaul Dagnelie } 9140fa1b3ccSPaul Dagnelie avl_destroy(ca); 9150fa1b3ccSPaul Dagnelie kmem_free(ca, sizeof (avl_tree_t)); 9160fa1b3ccSPaul Dagnelie } 9170fa1b3ccSPaul Dagnelie 9180fa1b3ccSPaul Dagnelie static int 9190fa1b3ccSPaul Dagnelie receive_read(struct receive_arg *ra, int len, void *buf) 9200fa1b3ccSPaul Dagnelie { 9210fa1b3ccSPaul Dagnelie int done = 0; 9220fa1b3ccSPaul Dagnelie 9230fa1b3ccSPaul Dagnelie /* 9240fa1b3ccSPaul Dagnelie * The code doesn't rely on this (lengths being multiples of 8). See 9250fa1b3ccSPaul Dagnelie * comment in dump_bytes. 9260fa1b3ccSPaul Dagnelie */ 927eb633035STom Caputi ASSERT(len % 8 == 0 || 928eb633035STom Caputi (ra->featureflags & DMU_BACKUP_FEATURE_RAW) != 0); 9290fa1b3ccSPaul Dagnelie 9300fa1b3ccSPaul Dagnelie while (done < len) { 9310fa1b3ccSPaul Dagnelie ssize_t resid; 9320fa1b3ccSPaul Dagnelie 9330fa1b3ccSPaul Dagnelie ra->err = vn_rdwr(UIO_READ, ra->vp, 9340fa1b3ccSPaul Dagnelie (char *)buf + done, len - done, 9350fa1b3ccSPaul Dagnelie ra->voff, UIO_SYSSPACE, FAPPEND, 9360fa1b3ccSPaul Dagnelie RLIM64_INFINITY, CRED(), &resid); 9370fa1b3ccSPaul Dagnelie 9380fa1b3ccSPaul Dagnelie if (resid == len - done) { 9390fa1b3ccSPaul Dagnelie /* 9400fa1b3ccSPaul Dagnelie * Note: ECKSUM indicates that the receive 9410fa1b3ccSPaul Dagnelie * was interrupted and can potentially be resumed. 9420fa1b3ccSPaul Dagnelie */ 9430fa1b3ccSPaul Dagnelie ra->err = SET_ERROR(ECKSUM); 9440fa1b3ccSPaul Dagnelie } 9450fa1b3ccSPaul Dagnelie ra->voff += len - done - resid; 9460fa1b3ccSPaul Dagnelie done = len - resid; 9470fa1b3ccSPaul Dagnelie if (ra->err != 0) 9480fa1b3ccSPaul Dagnelie return (ra->err); 9490fa1b3ccSPaul Dagnelie } 9500fa1b3ccSPaul Dagnelie 9510fa1b3ccSPaul Dagnelie ra->bytes_read += len; 9520fa1b3ccSPaul Dagnelie 9530fa1b3ccSPaul Dagnelie ASSERT3U(done, ==, len); 9540fa1b3ccSPaul Dagnelie return (0); 9550fa1b3ccSPaul Dagnelie } 9560fa1b3ccSPaul Dagnelie 9570fa1b3ccSPaul Dagnelie static void 9580fa1b3ccSPaul Dagnelie byteswap_record(dmu_replay_record_t *drr) 9590fa1b3ccSPaul Dagnelie { 9600fa1b3ccSPaul Dagnelie #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) 9610fa1b3ccSPaul Dagnelie #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) 9620fa1b3ccSPaul Dagnelie drr->drr_type = BSWAP_32(drr->drr_type); 9630fa1b3ccSPaul Dagnelie drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); 9640fa1b3ccSPaul Dagnelie 9650fa1b3ccSPaul Dagnelie switch (drr->drr_type) { 9660fa1b3ccSPaul Dagnelie case DRR_BEGIN: 9670fa1b3ccSPaul Dagnelie DO64(drr_begin.drr_magic); 9680fa1b3ccSPaul Dagnelie DO64(drr_begin.drr_versioninfo); 9690fa1b3ccSPaul Dagnelie DO64(drr_begin.drr_creation_time); 9700fa1b3ccSPaul Dagnelie DO32(drr_begin.drr_type); 9710fa1b3ccSPaul Dagnelie DO32(drr_begin.drr_flags); 9720fa1b3ccSPaul Dagnelie DO64(drr_begin.drr_toguid); 9730fa1b3ccSPaul Dagnelie DO64(drr_begin.drr_fromguid); 9740fa1b3ccSPaul Dagnelie break; 9750fa1b3ccSPaul Dagnelie case DRR_OBJECT: 9760fa1b3ccSPaul Dagnelie DO64(drr_object.drr_object); 9770fa1b3ccSPaul Dagnelie DO32(drr_object.drr_type); 9780fa1b3ccSPaul Dagnelie DO32(drr_object.drr_bonustype); 9790fa1b3ccSPaul Dagnelie DO32(drr_object.drr_blksz); 9800fa1b3ccSPaul Dagnelie DO32(drr_object.drr_bonuslen); 981eb633035STom Caputi DO32(drr_object.drr_raw_bonuslen); 9820fa1b3ccSPaul Dagnelie DO64(drr_object.drr_toguid); 983eb633035STom Caputi DO64(drr_object.drr_maxblkid); 9840fa1b3ccSPaul Dagnelie break; 9850fa1b3ccSPaul Dagnelie case DRR_FREEOBJECTS: 9860fa1b3ccSPaul Dagnelie DO64(drr_freeobjects.drr_firstobj); 9870fa1b3ccSPaul Dagnelie DO64(drr_freeobjects.drr_numobjs); 9880fa1b3ccSPaul Dagnelie DO64(drr_freeobjects.drr_toguid); 9890fa1b3ccSPaul Dagnelie break; 9900fa1b3ccSPaul Dagnelie case DRR_WRITE: 9910fa1b3ccSPaul Dagnelie DO64(drr_write.drr_object); 9920fa1b3ccSPaul Dagnelie DO32(drr_write.drr_type); 9930fa1b3ccSPaul Dagnelie DO64(drr_write.drr_offset); 9940fa1b3ccSPaul Dagnelie DO64(drr_write.drr_logical_size); 9950fa1b3ccSPaul Dagnelie DO64(drr_write.drr_toguid); 9960fa1b3ccSPaul Dagnelie ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write.drr_key.ddk_cksum); 9970fa1b3ccSPaul Dagnelie DO64(drr_write.drr_key.ddk_prop); 9980fa1b3ccSPaul Dagnelie DO64(drr_write.drr_compressed_size); 9990fa1b3ccSPaul Dagnelie break; 10000fa1b3ccSPaul Dagnelie case DRR_WRITE_BYREF: 10010fa1b3ccSPaul Dagnelie DO64(drr_write_byref.drr_object); 10020fa1b3ccSPaul Dagnelie DO64(drr_write_byref.drr_offset); 10030fa1b3ccSPaul Dagnelie DO64(drr_write_byref.drr_length); 10040fa1b3ccSPaul Dagnelie DO64(drr_write_byref.drr_toguid); 10050fa1b3ccSPaul Dagnelie DO64(drr_write_byref.drr_refguid); 10060fa1b3ccSPaul Dagnelie DO64(drr_write_byref.drr_refobject); 10070fa1b3ccSPaul Dagnelie DO64(drr_write_byref.drr_refoffset); 10080fa1b3ccSPaul Dagnelie ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write_byref. 10090fa1b3ccSPaul Dagnelie drr_key.ddk_cksum); 10100fa1b3ccSPaul Dagnelie DO64(drr_write_byref.drr_key.ddk_prop); 10110fa1b3ccSPaul Dagnelie break; 10120fa1b3ccSPaul Dagnelie case DRR_WRITE_EMBEDDED: 10130fa1b3ccSPaul Dagnelie DO64(drr_write_embedded.drr_object); 10140fa1b3ccSPaul Dagnelie DO64(drr_write_embedded.drr_offset); 10150fa1b3ccSPaul Dagnelie DO64(drr_write_embedded.drr_length); 10160fa1b3ccSPaul Dagnelie DO64(drr_write_embedded.drr_toguid); 10170fa1b3ccSPaul Dagnelie DO32(drr_write_embedded.drr_lsize); 10180fa1b3ccSPaul Dagnelie DO32(drr_write_embedded.drr_psize); 10190fa1b3ccSPaul Dagnelie break; 10200fa1b3ccSPaul Dagnelie case DRR_FREE: 10210fa1b3ccSPaul Dagnelie DO64(drr_free.drr_object); 10220fa1b3ccSPaul Dagnelie DO64(drr_free.drr_offset); 10230fa1b3ccSPaul Dagnelie DO64(drr_free.drr_length); 10240fa1b3ccSPaul Dagnelie DO64(drr_free.drr_toguid); 10250fa1b3ccSPaul Dagnelie break; 10260fa1b3ccSPaul Dagnelie case DRR_SPILL: 10270fa1b3ccSPaul Dagnelie DO64(drr_spill.drr_object); 10280fa1b3ccSPaul Dagnelie DO64(drr_spill.drr_length); 10290fa1b3ccSPaul Dagnelie DO64(drr_spill.drr_toguid); 1030eb633035STom Caputi DO64(drr_spill.drr_compressed_size); 1031eb633035STom Caputi DO32(drr_spill.drr_type); 1032eb633035STom Caputi break; 1033eb633035STom Caputi case DRR_OBJECT_RANGE: 1034eb633035STom Caputi DO64(drr_object_range.drr_firstobj); 1035eb633035STom Caputi DO64(drr_object_range.drr_numslots); 1036eb633035STom Caputi DO64(drr_object_range.drr_toguid); 10370fa1b3ccSPaul Dagnelie break; 10380fa1b3ccSPaul Dagnelie case DRR_END: 10390fa1b3ccSPaul Dagnelie DO64(drr_end.drr_toguid); 10400fa1b3ccSPaul Dagnelie ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_end.drr_checksum); 10410fa1b3ccSPaul Dagnelie break; 10420fa1b3ccSPaul Dagnelie } 10430fa1b3ccSPaul Dagnelie 10440fa1b3ccSPaul Dagnelie if (drr->drr_type != DRR_BEGIN) { 10450fa1b3ccSPaul Dagnelie ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_checksum.drr_checksum); 10460fa1b3ccSPaul Dagnelie } 10470fa1b3ccSPaul Dagnelie 10480fa1b3ccSPaul Dagnelie #undef DO64 10490fa1b3ccSPaul Dagnelie #undef DO32 10500fa1b3ccSPaul Dagnelie } 10510fa1b3ccSPaul Dagnelie 10520fa1b3ccSPaul Dagnelie static inline uint8_t 10530fa1b3ccSPaul Dagnelie deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) 10540fa1b3ccSPaul Dagnelie { 10550fa1b3ccSPaul Dagnelie if (bonus_type == DMU_OT_SA) { 10560fa1b3ccSPaul Dagnelie return (1); 10570fa1b3ccSPaul Dagnelie } else { 10580fa1b3ccSPaul Dagnelie return (1 + 10590fa1b3ccSPaul Dagnelie ((DN_OLD_MAX_BONUSLEN - 10600fa1b3ccSPaul Dagnelie MIN(DN_OLD_MAX_BONUSLEN, bonus_size)) >> SPA_BLKPTRSHIFT)); 10610fa1b3ccSPaul Dagnelie } 10620fa1b3ccSPaul Dagnelie } 10630fa1b3ccSPaul Dagnelie 10640fa1b3ccSPaul Dagnelie static void 10650fa1b3ccSPaul Dagnelie save_resume_state(struct receive_writer_arg *rwa, 10660fa1b3ccSPaul Dagnelie uint64_t object, uint64_t offset, dmu_tx_t *tx) 10670fa1b3ccSPaul Dagnelie { 10680fa1b3ccSPaul Dagnelie int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; 10690fa1b3ccSPaul Dagnelie 10700fa1b3ccSPaul Dagnelie if (!rwa->resumable) 10710fa1b3ccSPaul Dagnelie return; 10720fa1b3ccSPaul Dagnelie 10730fa1b3ccSPaul Dagnelie /* 10740fa1b3ccSPaul Dagnelie * We use ds_resume_bytes[] != 0 to indicate that we need to 10750fa1b3ccSPaul Dagnelie * update this on disk, so it must not be 0. 10760fa1b3ccSPaul Dagnelie */ 10770fa1b3ccSPaul Dagnelie ASSERT(rwa->bytes_read != 0); 10780fa1b3ccSPaul Dagnelie 10790fa1b3ccSPaul Dagnelie /* 10800fa1b3ccSPaul Dagnelie * We only resume from write records, which have a valid 10810fa1b3ccSPaul Dagnelie * (non-meta-dnode) object number. 10820fa1b3ccSPaul Dagnelie */ 10830fa1b3ccSPaul Dagnelie ASSERT(object != 0); 10840fa1b3ccSPaul Dagnelie 10850fa1b3ccSPaul Dagnelie /* 10860fa1b3ccSPaul Dagnelie * For resuming to work correctly, we must receive records in order, 10870fa1b3ccSPaul Dagnelie * sorted by object,offset. This is checked by the callers, but 10880fa1b3ccSPaul Dagnelie * assert it here for good measure. 10890fa1b3ccSPaul Dagnelie */ 10900fa1b3ccSPaul Dagnelie ASSERT3U(object, >=, rwa->os->os_dsl_dataset->ds_resume_object[txgoff]); 10910fa1b3ccSPaul Dagnelie ASSERT(object != rwa->os->os_dsl_dataset->ds_resume_object[txgoff] || 10920fa1b3ccSPaul Dagnelie offset >= rwa->os->os_dsl_dataset->ds_resume_offset[txgoff]); 10930fa1b3ccSPaul Dagnelie ASSERT3U(rwa->bytes_read, >=, 10940fa1b3ccSPaul Dagnelie rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff]); 10950fa1b3ccSPaul Dagnelie 10960fa1b3ccSPaul Dagnelie rwa->os->os_dsl_dataset->ds_resume_object[txgoff] = object; 10970fa1b3ccSPaul Dagnelie rwa->os->os_dsl_dataset->ds_resume_offset[txgoff] = offset; 10980fa1b3ccSPaul Dagnelie rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff] = rwa->bytes_read; 10990fa1b3ccSPaul Dagnelie } 11000fa1b3ccSPaul Dagnelie 1101eb633035STom Caputi int receive_object_delay_frac = 0; 1102eb633035STom Caputi 11030fa1b3ccSPaul Dagnelie static int 11040fa1b3ccSPaul Dagnelie receive_object(struct receive_writer_arg *rwa, struct drr_object *drro, 11050fa1b3ccSPaul Dagnelie void *data) 11060fa1b3ccSPaul Dagnelie { 11070fa1b3ccSPaul Dagnelie dmu_object_info_t doi; 11080fa1b3ccSPaul Dagnelie dmu_tx_t *tx; 11090fa1b3ccSPaul Dagnelie uint64_t object; 11100fa1b3ccSPaul Dagnelie int err; 11110fa1b3ccSPaul Dagnelie uint8_t dn_slots = drro->drr_dn_slots != 0 ? 11120fa1b3ccSPaul Dagnelie drro->drr_dn_slots : DNODE_MIN_SLOTS; 11130fa1b3ccSPaul Dagnelie 1114eb633035STom Caputi if (receive_object_delay_frac != 0 && 1115eb633035STom Caputi spa_get_random(receive_object_delay_frac) == 0) 1116eb633035STom Caputi delay(1); 1117eb633035STom Caputi 11180fa1b3ccSPaul Dagnelie if (drro->drr_type == DMU_OT_NONE || 11190fa1b3ccSPaul Dagnelie !DMU_OT_IS_VALID(drro->drr_type) || 11200fa1b3ccSPaul Dagnelie !DMU_OT_IS_VALID(drro->drr_bonustype) || 11210fa1b3ccSPaul Dagnelie drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || 11220fa1b3ccSPaul Dagnelie drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || 11230fa1b3ccSPaul Dagnelie P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || 11240fa1b3ccSPaul Dagnelie drro->drr_blksz < SPA_MINBLOCKSIZE || 11250fa1b3ccSPaul Dagnelie drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) || 11260fa1b3ccSPaul Dagnelie drro->drr_bonuslen > 11270fa1b3ccSPaul Dagnelie DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa->os))) || 11280fa1b3ccSPaul Dagnelie dn_slots > 11290fa1b3ccSPaul Dagnelie (spa_maxdnodesize(dmu_objset_spa(rwa->os)) >> DNODE_SHIFT)) { 11300fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 11310fa1b3ccSPaul Dagnelie } 11320fa1b3ccSPaul Dagnelie 1133eb633035STom Caputi if (rwa->raw) { 1134eb633035STom Caputi /* 1135eb633035STom Caputi * We should have received a DRR_OBJECT_RANGE record 1136eb633035STom Caputi * containing this block and stored it in rwa. 1137eb633035STom Caputi */ 1138eb633035STom Caputi if (drro->drr_object < rwa->or_firstobj || 1139eb633035STom Caputi drro->drr_object >= rwa->or_firstobj + rwa->or_numslots || 1140eb633035STom Caputi drro->drr_raw_bonuslen < drro->drr_bonuslen || 1141eb633035STom Caputi drro->drr_indblkshift > SPA_MAXBLOCKSHIFT || 1142eb633035STom Caputi drro->drr_nlevels > DN_MAX_LEVELS || 1143eb633035STom Caputi drro->drr_nblkptr > DN_MAX_NBLKPTR || 1144eb633035STom Caputi DN_SLOTS_TO_BONUSLEN(drro->drr_dn_slots) < 1145eb633035STom Caputi drro->drr_raw_bonuslen) 1146eb633035STom Caputi return (SET_ERROR(EINVAL)); 1147eb633035STom Caputi } else { 1148eb633035STom Caputi 1149eb633035STom Caputi /* 1150eb633035STom Caputi * The DRR_OBJECT_SPILL flag is valid when the DRR_BEGIN 1151eb633035STom Caputi * record indicates this by setting DRR_FLAG_SPILL_BLOCK. 1152eb633035STom Caputi */ 1153eb633035STom Caputi if (((drro->drr_flags & ~(DRR_OBJECT_SPILL))) || 1154eb633035STom Caputi (!rwa->spill && DRR_OBJECT_HAS_SPILL(drro->drr_flags))) { 1155eb633035STom Caputi return (SET_ERROR(EINVAL)); 1156eb633035STom Caputi } 1157eb633035STom Caputi 1158eb633035STom Caputi if (drro->drr_raw_bonuslen != 0 || drro->drr_nblkptr != 0 || 1159eb633035STom Caputi drro->drr_indblkshift != 0 || drro->drr_nlevels != 0) { 1160eb633035STom Caputi return (SET_ERROR(EINVAL)); 1161eb633035STom Caputi } 1162eb633035STom Caputi } 1163eb633035STom Caputi 11640fa1b3ccSPaul Dagnelie err = dmu_object_info(rwa->os, drro->drr_object, &doi); 11650fa1b3ccSPaul Dagnelie 11660fa1b3ccSPaul Dagnelie if (err != 0 && err != ENOENT && err != EEXIST) 11670fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 11680fa1b3ccSPaul Dagnelie 11690fa1b3ccSPaul Dagnelie if (drro->drr_object > rwa->max_object) 11700fa1b3ccSPaul Dagnelie rwa->max_object = drro->drr_object; 11710fa1b3ccSPaul Dagnelie 11720fa1b3ccSPaul Dagnelie /* 11730fa1b3ccSPaul Dagnelie * If we are losing blkptrs or changing the block size this must 11740fa1b3ccSPaul Dagnelie * be a new file instance. We must clear out the previous file 11750fa1b3ccSPaul Dagnelie * contents before we can change this type of metadata in the dnode. 1176eb633035STom Caputi * Raw receives will also check that the indirect structure of the 1177eb633035STom Caputi * dnode hasn't changed. 11780fa1b3ccSPaul Dagnelie */ 11790fa1b3ccSPaul Dagnelie if (err == 0) { 1180eb633035STom Caputi uint32_t indblksz = drro->drr_indblkshift ? 1181eb633035STom Caputi 1ULL << drro->drr_indblkshift : 0; 1182eb633035STom Caputi int nblkptr = deduce_nblkptr(drro->drr_bonustype, 1183eb633035STom Caputi drro->drr_bonuslen); 1184eb633035STom Caputi boolean_t did_free = B_FALSE; 11850fa1b3ccSPaul Dagnelie 11860fa1b3ccSPaul Dagnelie object = drro->drr_object; 11870fa1b3ccSPaul Dagnelie 1188eb633035STom Caputi /* nblkptr should be bounded by the bonus size and type */ 1189eb633035STom Caputi if (rwa->raw && nblkptr != drro->drr_nblkptr) 1190eb633035STom Caputi return (SET_ERROR(EINVAL)); 11910fa1b3ccSPaul Dagnelie 1192eb633035STom Caputi /* 1193eb633035STom Caputi * Check for indicators that the object was freed and 1194eb633035STom Caputi * reallocated. For all sends, these indicators are: 1195eb633035STom Caputi * - A changed block size 1196eb633035STom Caputi * - A smaller nblkptr 1197eb633035STom Caputi * - A changed dnode size 1198eb633035STom Caputi * For raw sends we also check a few other fields to 1199eb633035STom Caputi * ensure we are preserving the objset structure exactly 1200eb633035STom Caputi * as it was on the receive side: 1201eb633035STom Caputi * - A changed indirect block size 1202eb633035STom Caputi * - A smaller nlevels 1203eb633035STom Caputi */ 12040fa1b3ccSPaul Dagnelie if (drro->drr_blksz != doi.doi_data_block_size || 12050fa1b3ccSPaul Dagnelie nblkptr < doi.doi_nblkptr || 1206eb633035STom Caputi dn_slots != doi.doi_dnodesize >> DNODE_SHIFT || 1207eb633035STom Caputi (rwa->raw && 1208eb633035STom Caputi (indblksz != doi.doi_metadata_block_size || 1209eb633035STom Caputi drro->drr_nlevels < doi.doi_indirection))) { 1210eb633035STom Caputi err = dmu_free_long_range(rwa->os, 1211eb633035STom Caputi drro->drr_object, 0, DMU_OBJECT_END); 1212eb633035STom Caputi if (err != 0) 1213eb633035STom Caputi return (SET_ERROR(EINVAL)); 1214eb633035STom Caputi else 1215eb633035STom Caputi did_free = B_TRUE; 1216eb633035STom Caputi } 1217eb633035STom Caputi 1218eb633035STom Caputi /* 1219eb633035STom Caputi * The dmu does not currently support decreasing nlevels 1220eb633035STom Caputi * or changing the number of dnode slots on an object. For 1221eb633035STom Caputi * non-raw sends, this does not matter and the new object 1222eb633035STom Caputi * can just use the previous one's nlevels. For raw sends, 1223eb633035STom Caputi * however, the structure of the received dnode (including 1224eb633035STom Caputi * nlevels and dnode slots) must match that of the send 1225eb633035STom Caputi * side. Therefore, instead of using dmu_object_reclaim(), 1226eb633035STom Caputi * we must free the object completely and call 1227eb633035STom Caputi * dmu_object_claim_dnsize() instead. 1228eb633035STom Caputi */ 1229eb633035STom Caputi if ((rwa->raw && drro->drr_nlevels < doi.doi_indirection) || 12300fa1b3ccSPaul Dagnelie dn_slots != doi.doi_dnodesize >> DNODE_SHIFT) { 1231eb633035STom Caputi err = dmu_free_long_object(rwa->os, drro->drr_object); 1232eb633035STom Caputi if (err != 0) 1233eb633035STom Caputi return (SET_ERROR(EINVAL)); 1234eb633035STom Caputi 1235eb633035STom Caputi txg_wait_synced(dmu_objset_pool(rwa->os), 0); 1236eb633035STom Caputi object = DMU_NEW_OBJECT; 1237eb633035STom Caputi } 1238eb633035STom Caputi 1239eb633035STom Caputi /* 1240eb633035STom Caputi * For raw receives, free everything beyond the new incoming 1241eb633035STom Caputi * maxblkid. Normally this would be done with a DRR_FREE 1242eb633035STom Caputi * record that would come after this DRR_OBJECT record is 1243eb633035STom Caputi * processed. However, for raw receives we manually set the 1244eb633035STom Caputi * maxblkid from the drr_maxblkid and so we must first free 1245eb633035STom Caputi * everything above that blkid to ensure the DMU is always 1246eb633035STom Caputi * consistent with itself. We will never free the first block 1247eb633035STom Caputi * of the object here because a maxblkid of 0 could indicate 1248eb633035STom Caputi * an object with a single block or one with no blocks. This 1249eb633035STom Caputi * free may be skipped when dmu_free_long_range() was called 1250eb633035STom Caputi * above since it covers the entire object's contents. 1251eb633035STom Caputi */ 1252eb633035STom Caputi if (rwa->raw && object != DMU_NEW_OBJECT && !did_free) { 12530fa1b3ccSPaul Dagnelie err = dmu_free_long_range(rwa->os, drro->drr_object, 1254eb633035STom Caputi (drro->drr_maxblkid + 1) * doi.doi_data_block_size, 1255eb633035STom Caputi DMU_OBJECT_END); 12560fa1b3ccSPaul Dagnelie if (err != 0) 12570fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 12580fa1b3ccSPaul Dagnelie } 12590fa1b3ccSPaul Dagnelie } else if (err == EEXIST) { 12600fa1b3ccSPaul Dagnelie /* 12610fa1b3ccSPaul Dagnelie * The object requested is currently an interior slot of a 12620fa1b3ccSPaul Dagnelie * multi-slot dnode. This will be resolved when the next txg 12630fa1b3ccSPaul Dagnelie * is synced out, since the send stream will have told us 12640fa1b3ccSPaul Dagnelie * to free this slot when we freed the associated dnode 12650fa1b3ccSPaul Dagnelie * earlier in the stream. 12660fa1b3ccSPaul Dagnelie */ 12670fa1b3ccSPaul Dagnelie txg_wait_synced(dmu_objset_pool(rwa->os), 0); 1268*6ccda740Sloli 1269eb633035STom Caputi if (dmu_object_info(rwa->os, drro->drr_object, NULL) != ENOENT) 1270eb633035STom Caputi return (SET_ERROR(EINVAL)); 1271eb633035STom Caputi 1272eb633035STom Caputi /* object was freed and we are about to allocate a new one */ 1273eb633035STom Caputi object = DMU_NEW_OBJECT; 12740fa1b3ccSPaul Dagnelie } else { 12750fa1b3ccSPaul Dagnelie /* object is free and we are about to allocate a new one */ 12760fa1b3ccSPaul Dagnelie object = DMU_NEW_OBJECT; 12770fa1b3ccSPaul Dagnelie } 12780fa1b3ccSPaul Dagnelie 12790fa1b3ccSPaul Dagnelie /* 12800fa1b3ccSPaul Dagnelie * If this is a multi-slot dnode there is a chance that this 12810fa1b3ccSPaul Dagnelie * object will expand into a slot that is already used by 12820fa1b3ccSPaul Dagnelie * another object from the previous snapshot. We must free 12830fa1b3ccSPaul Dagnelie * these objects before we attempt to allocate the new dnode. 12840fa1b3ccSPaul Dagnelie */ 12850fa1b3ccSPaul Dagnelie if (dn_slots > 1) { 12860fa1b3ccSPaul Dagnelie boolean_t need_sync = B_FALSE; 12870fa1b3ccSPaul Dagnelie 12880fa1b3ccSPaul Dagnelie for (uint64_t slot = drro->drr_object + 1; 12890fa1b3ccSPaul Dagnelie slot < drro->drr_object + dn_slots; 12900fa1b3ccSPaul Dagnelie slot++) { 12910fa1b3ccSPaul Dagnelie dmu_object_info_t slot_doi; 12920fa1b3ccSPaul Dagnelie 12930fa1b3ccSPaul Dagnelie err = dmu_object_info(rwa->os, slot, &slot_doi); 12940fa1b3ccSPaul Dagnelie if (err == ENOENT || err == EEXIST) 12950fa1b3ccSPaul Dagnelie continue; 12960fa1b3ccSPaul Dagnelie else if (err != 0) 12970fa1b3ccSPaul Dagnelie return (err); 12980fa1b3ccSPaul Dagnelie 12990fa1b3ccSPaul Dagnelie err = dmu_free_long_object(rwa->os, slot); 13000fa1b3ccSPaul Dagnelie 13010fa1b3ccSPaul Dagnelie if (err != 0) 13020fa1b3ccSPaul Dagnelie return (err); 13030fa1b3ccSPaul Dagnelie 13040fa1b3ccSPaul Dagnelie need_sync = B_TRUE; 13050fa1b3ccSPaul Dagnelie } 13060fa1b3ccSPaul Dagnelie 13070fa1b3ccSPaul Dagnelie if (need_sync) 13080fa1b3ccSPaul Dagnelie txg_wait_synced(dmu_objset_pool(rwa->os), 0); 13090fa1b3ccSPaul Dagnelie } 13100fa1b3ccSPaul Dagnelie 13110fa1b3ccSPaul Dagnelie tx = dmu_tx_create(rwa->os); 13120fa1b3ccSPaul Dagnelie dmu_tx_hold_bonus(tx, object); 1313eb633035STom Caputi dmu_tx_hold_write(tx, object, 0, 0); 13140fa1b3ccSPaul Dagnelie err = dmu_tx_assign(tx, TXG_WAIT); 13150fa1b3ccSPaul Dagnelie if (err != 0) { 13160fa1b3ccSPaul Dagnelie dmu_tx_abort(tx); 13170fa1b3ccSPaul Dagnelie return (err); 13180fa1b3ccSPaul Dagnelie } 13190fa1b3ccSPaul Dagnelie 13200fa1b3ccSPaul Dagnelie if (object == DMU_NEW_OBJECT) { 1321eb633035STom Caputi /* Currently free, wants to be allocated */ 13220fa1b3ccSPaul Dagnelie err = dmu_object_claim_dnsize(rwa->os, drro->drr_object, 13230fa1b3ccSPaul Dagnelie drro->drr_type, drro->drr_blksz, 13240fa1b3ccSPaul Dagnelie drro->drr_bonustype, drro->drr_bonuslen, 13250fa1b3ccSPaul Dagnelie dn_slots << DNODE_SHIFT, tx); 13260fa1b3ccSPaul Dagnelie } else if (drro->drr_type != doi.doi_type || 13270fa1b3ccSPaul Dagnelie drro->drr_blksz != doi.doi_data_block_size || 13280fa1b3ccSPaul Dagnelie drro->drr_bonustype != doi.doi_bonus_type || 1329eb633035STom Caputi drro->drr_bonuslen != doi.doi_bonus_size) { 1330eb633035STom Caputi /* Currently allocated, but with different properties */ 13310fa1b3ccSPaul Dagnelie err = dmu_object_reclaim_dnsize(rwa->os, drro->drr_object, 13320fa1b3ccSPaul Dagnelie drro->drr_type, drro->drr_blksz, 13330fa1b3ccSPaul Dagnelie drro->drr_bonustype, drro->drr_bonuslen, 1334eb633035STom Caputi dn_slots << DNODE_SHIFT, rwa->spill ? 1335eb633035STom Caputi DRR_OBJECT_HAS_SPILL(drro->drr_flags) : B_FALSE, tx); 1336eb633035STom Caputi } else if (rwa->spill && !DRR_OBJECT_HAS_SPILL(drro->drr_flags)) { 1337eb633035STom Caputi /* 1338eb633035STom Caputi * Currently allocated, the existing version of this object 1339eb633035STom Caputi * may reference a spill block that is no longer allocated 1340eb633035STom Caputi * at the source and needs to be freed. 1341eb633035STom Caputi */ 1342eb633035STom Caputi err = dmu_object_rm_spill(rwa->os, drro->drr_object, tx); 13430fa1b3ccSPaul Dagnelie } 1344eb633035STom Caputi 13450fa1b3ccSPaul Dagnelie if (err != 0) { 13460fa1b3ccSPaul Dagnelie dmu_tx_commit(tx); 13470fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 13480fa1b3ccSPaul Dagnelie } 13490fa1b3ccSPaul Dagnelie 1350eb633035STom Caputi if (rwa->or_crypt_params_present) { 1351eb633035STom Caputi /* 1352eb633035STom Caputi * Set the crypt params for the buffer associated with this 1353eb633035STom Caputi * range of dnodes. This causes the blkptr_t to have the 1354eb633035STom Caputi * same crypt params (byteorder, salt, iv, mac) as on the 1355eb633035STom Caputi * sending side. 1356eb633035STom Caputi * 1357eb633035STom Caputi * Since we are committing this tx now, it is possible for 1358eb633035STom Caputi * the dnode block to end up on-disk with the incorrect MAC, 1359eb633035STom Caputi * if subsequent objects in this block are received in a 1360eb633035STom Caputi * different txg. However, since the dataset is marked as 1361eb633035STom Caputi * inconsistent, no code paths will do a non-raw read (or 1362eb633035STom Caputi * decrypt the block / verify the MAC). The receive code and 1363eb633035STom Caputi * scrub code can safely do raw reads and verify the 1364eb633035STom Caputi * checksum. They don't need to verify the MAC. 1365eb633035STom Caputi */ 1366eb633035STom Caputi dmu_buf_t *db = NULL; 1367eb633035STom Caputi uint64_t offset = rwa->or_firstobj * DNODE_MIN_SIZE; 1368eb633035STom Caputi 1369eb633035STom Caputi err = dmu_buf_hold_by_dnode(DMU_META_DNODE(rwa->os), 1370eb633035STom Caputi offset, FTAG, &db, DMU_READ_PREFETCH | DMU_READ_NO_DECRYPT); 1371eb633035STom Caputi if (err != 0) { 1372eb633035STom Caputi dmu_tx_commit(tx); 1373eb633035STom Caputi return (SET_ERROR(EINVAL)); 1374eb633035STom Caputi } 1375eb633035STom Caputi 1376eb633035STom Caputi dmu_buf_set_crypt_params(db, rwa->or_byteorder, 1377eb633035STom Caputi rwa->or_salt, rwa->or_iv, rwa->or_mac, tx); 1378eb633035STom Caputi 1379eb633035STom Caputi dmu_buf_rele(db, FTAG); 1380eb633035STom Caputi 1381eb633035STom Caputi rwa->or_crypt_params_present = B_FALSE; 1382eb633035STom Caputi } 1383eb633035STom Caputi 13840fa1b3ccSPaul Dagnelie dmu_object_set_checksum(rwa->os, drro->drr_object, 13850fa1b3ccSPaul Dagnelie drro->drr_checksumtype, tx); 13860fa1b3ccSPaul Dagnelie dmu_object_set_compress(rwa->os, drro->drr_object, 13870fa1b3ccSPaul Dagnelie drro->drr_compress, tx); 13880fa1b3ccSPaul Dagnelie 1389eb633035STom Caputi /* handle more restrictive dnode structuring for raw recvs */ 1390eb633035STom Caputi if (rwa->raw) { 1391eb633035STom Caputi /* 1392eb633035STom Caputi * Set the indirect block size, block shift, nlevels. 1393eb633035STom Caputi * This will not fail because we ensured all of the 1394eb633035STom Caputi * blocks were freed earlier if this is a new object. 1395eb633035STom Caputi * For non-new objects block size and indirect block 1396eb633035STom Caputi * shift cannot change and nlevels can only increase. 1397eb633035STom Caputi */ 1398eb633035STom Caputi VERIFY0(dmu_object_set_blocksize(rwa->os, drro->drr_object, 1399eb633035STom Caputi drro->drr_blksz, drro->drr_indblkshift, tx)); 1400eb633035STom Caputi VERIFY0(dmu_object_set_nlevels(rwa->os, drro->drr_object, 1401eb633035STom Caputi drro->drr_nlevels, tx)); 1402eb633035STom Caputi 1403eb633035STom Caputi /* 1404eb633035STom Caputi * Set the maxblkid. This will always succeed because 1405eb633035STom Caputi * we freed all blocks beyond the new maxblkid above. 1406eb633035STom Caputi */ 1407eb633035STom Caputi VERIFY0(dmu_object_set_maxblkid(rwa->os, drro->drr_object, 1408eb633035STom Caputi drro->drr_maxblkid, tx)); 1409eb633035STom Caputi } 1410eb633035STom Caputi 14110fa1b3ccSPaul Dagnelie if (data != NULL) { 14120fa1b3ccSPaul Dagnelie dmu_buf_t *db; 1413eb633035STom Caputi dnode_t *dn; 1414eb633035STom Caputi uint32_t flags = DMU_READ_NO_PREFETCH; 1415eb633035STom Caputi 1416eb633035STom Caputi if (rwa->raw) 1417eb633035STom Caputi flags |= DMU_READ_NO_DECRYPT; 1418eb633035STom Caputi 1419eb633035STom Caputi VERIFY0(dnode_hold(rwa->os, drro->drr_object, FTAG, &dn)); 1420eb633035STom Caputi VERIFY0(dmu_bonus_hold_by_dnode(dn, FTAG, &db, flags)); 14210fa1b3ccSPaul Dagnelie 14220fa1b3ccSPaul Dagnelie dmu_buf_will_dirty(db, tx); 14230fa1b3ccSPaul Dagnelie 14240fa1b3ccSPaul Dagnelie ASSERT3U(db->db_size, >=, drro->drr_bonuslen); 1425eb633035STom Caputi bcopy(data, db->db_data, DRR_OBJECT_PAYLOAD_SIZE(drro)); 1426eb633035STom Caputi 1427eb633035STom Caputi /* 1428eb633035STom Caputi * Raw bonus buffers have their byteorder determined by the 1429eb633035STom Caputi * DRR_OBJECT_RANGE record. 1430eb633035STom Caputi */ 1431eb633035STom Caputi if (rwa->byteswap && !rwa->raw) { 14320fa1b3ccSPaul Dagnelie dmu_object_byteswap_t byteswap = 14330fa1b3ccSPaul Dagnelie DMU_OT_BYTESWAP(drro->drr_bonustype); 14340fa1b3ccSPaul Dagnelie dmu_ot_byteswap[byteswap].ob_func(db->db_data, 1435eb633035STom Caputi DRR_OBJECT_PAYLOAD_SIZE(drro)); 14360fa1b3ccSPaul Dagnelie } 14370fa1b3ccSPaul Dagnelie dmu_buf_rele(db, FTAG); 1438eb633035STom Caputi dnode_rele(dn, FTAG); 14390fa1b3ccSPaul Dagnelie } 14400fa1b3ccSPaul Dagnelie dmu_tx_commit(tx); 14410fa1b3ccSPaul Dagnelie 14420fa1b3ccSPaul Dagnelie return (0); 14430fa1b3ccSPaul Dagnelie } 14440fa1b3ccSPaul Dagnelie 14450fa1b3ccSPaul Dagnelie /* ARGSUSED */ 14460fa1b3ccSPaul Dagnelie static int 14470fa1b3ccSPaul Dagnelie receive_freeobjects(struct receive_writer_arg *rwa, 14480fa1b3ccSPaul Dagnelie struct drr_freeobjects *drrfo) 14490fa1b3ccSPaul Dagnelie { 14500fa1b3ccSPaul Dagnelie uint64_t obj; 14510fa1b3ccSPaul Dagnelie int next_err = 0; 14520fa1b3ccSPaul Dagnelie 14530fa1b3ccSPaul Dagnelie if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) 14540fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 14550fa1b3ccSPaul Dagnelie 14560fa1b3ccSPaul Dagnelie for (obj = drrfo->drr_firstobj == 0 ? 1 : drrfo->drr_firstobj; 14570fa1b3ccSPaul Dagnelie obj < drrfo->drr_firstobj + drrfo->drr_numobjs && next_err == 0; 14580fa1b3ccSPaul Dagnelie next_err = dmu_object_next(rwa->os, &obj, FALSE, 0)) { 1459eb633035STom Caputi dmu_object_info_t doi; 14600fa1b3ccSPaul Dagnelie int err; 14610fa1b3ccSPaul Dagnelie 1462eb633035STom Caputi err = dmu_object_info(rwa->os, obj, &doi); 14630fa1b3ccSPaul Dagnelie if (err == ENOENT) 14640fa1b3ccSPaul Dagnelie continue; 14650fa1b3ccSPaul Dagnelie else if (err != 0) 14660fa1b3ccSPaul Dagnelie return (err); 14670fa1b3ccSPaul Dagnelie 14680fa1b3ccSPaul Dagnelie err = dmu_free_long_object(rwa->os, obj); 1469eb633035STom Caputi 14700fa1b3ccSPaul Dagnelie if (err != 0) 14710fa1b3ccSPaul Dagnelie return (err); 14720fa1b3ccSPaul Dagnelie 14730fa1b3ccSPaul Dagnelie if (obj > rwa->max_object) 14740fa1b3ccSPaul Dagnelie rwa->max_object = obj; 14750fa1b3ccSPaul Dagnelie } 14760fa1b3ccSPaul Dagnelie if (next_err != ESRCH) 14770fa1b3ccSPaul Dagnelie return (next_err); 14780fa1b3ccSPaul Dagnelie return (0); 14790fa1b3ccSPaul Dagnelie } 14800fa1b3ccSPaul Dagnelie 14810fa1b3ccSPaul Dagnelie static int 14820fa1b3ccSPaul Dagnelie receive_write(struct receive_writer_arg *rwa, struct drr_write *drrw, 14830fa1b3ccSPaul Dagnelie arc_buf_t *abuf) 14840fa1b3ccSPaul Dagnelie { 14850fa1b3ccSPaul Dagnelie int err; 1486eb633035STom Caputi dmu_tx_t *tx; 1487eb633035STom Caputi dnode_t *dn; 14880fa1b3ccSPaul Dagnelie 14890fa1b3ccSPaul Dagnelie if (drrw->drr_offset + drrw->drr_logical_size < drrw->drr_offset || 14900fa1b3ccSPaul Dagnelie !DMU_OT_IS_VALID(drrw->drr_type)) 14910fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 14920fa1b3ccSPaul Dagnelie 14930fa1b3ccSPaul Dagnelie /* 14940fa1b3ccSPaul Dagnelie * For resuming to work, records must be in increasing order 14950fa1b3ccSPaul Dagnelie * by (object, offset). 14960fa1b3ccSPaul Dagnelie */ 14970fa1b3ccSPaul Dagnelie if (drrw->drr_object < rwa->last_object || 14980fa1b3ccSPaul Dagnelie (drrw->drr_object == rwa->last_object && 14990fa1b3ccSPaul Dagnelie drrw->drr_offset < rwa->last_offset)) { 15000fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 15010fa1b3ccSPaul Dagnelie } 15020fa1b3ccSPaul Dagnelie rwa->last_object = drrw->drr_object; 15030fa1b3ccSPaul Dagnelie rwa->last_offset = drrw->drr_offset; 15040fa1b3ccSPaul Dagnelie 15050fa1b3ccSPaul Dagnelie if (rwa->last_object > rwa->max_object) 15060fa1b3ccSPaul Dagnelie rwa->max_object = rwa->last_object; 15070fa1b3ccSPaul Dagnelie 15080fa1b3ccSPaul Dagnelie if (dmu_object_info(rwa->os, drrw->drr_object, NULL) != 0) 15090fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 15100fa1b3ccSPaul Dagnelie 15110fa1b3ccSPaul Dagnelie tx = dmu_tx_create(rwa->os); 15120fa1b3ccSPaul Dagnelie dmu_tx_hold_write(tx, drrw->drr_object, 15130fa1b3ccSPaul Dagnelie drrw->drr_offset, drrw->drr_logical_size); 15140fa1b3ccSPaul Dagnelie err = dmu_tx_assign(tx, TXG_WAIT); 15150fa1b3ccSPaul Dagnelie if (err != 0) { 15160fa1b3ccSPaul Dagnelie dmu_tx_abort(tx); 15170fa1b3ccSPaul Dagnelie return (err); 15180fa1b3ccSPaul Dagnelie } 1519eb633035STom Caputi 1520eb633035STom Caputi if (rwa->byteswap && !arc_is_encrypted(abuf) && 1521eb633035STom Caputi arc_get_compression(abuf) == ZIO_COMPRESS_OFF) { 15220fa1b3ccSPaul Dagnelie dmu_object_byteswap_t byteswap = 15230fa1b3ccSPaul Dagnelie DMU_OT_BYTESWAP(drrw->drr_type); 15240fa1b3ccSPaul Dagnelie dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, 15250fa1b3ccSPaul Dagnelie DRR_WRITE_PAYLOAD_SIZE(drrw)); 15260fa1b3ccSPaul Dagnelie } 15270fa1b3ccSPaul Dagnelie 1528eb633035STom Caputi VERIFY0(dnode_hold(rwa->os, drrw->drr_object, FTAG, &dn)); 1529eb633035STom Caputi err = dmu_assign_arcbuf_by_dnode(dn, drrw->drr_offset, abuf, tx); 1530eb633035STom Caputi if (err != 0) { 1531eb633035STom Caputi dnode_rele(dn, FTAG); 1532eb633035STom Caputi dmu_tx_commit(tx); 1533eb633035STom Caputi return (err); 1534eb633035STom Caputi } 1535eb633035STom Caputi dnode_rele(dn, FTAG); 15360fa1b3ccSPaul Dagnelie 15370fa1b3ccSPaul Dagnelie /* 15380fa1b3ccSPaul Dagnelie * Note: If the receive fails, we want the resume stream to start 15390fa1b3ccSPaul Dagnelie * with the same record that we last successfully received (as opposed 15400fa1b3ccSPaul Dagnelie * to the next record), so that we can verify that we are 15410fa1b3ccSPaul Dagnelie * resuming from the correct location. 15420fa1b3ccSPaul Dagnelie */ 15430fa1b3ccSPaul Dagnelie save_resume_state(rwa, drrw->drr_object, drrw->drr_offset, tx); 15440fa1b3ccSPaul Dagnelie dmu_tx_commit(tx); 15450fa1b3ccSPaul Dagnelie 15460fa1b3ccSPaul Dagnelie return (0); 15470fa1b3ccSPaul Dagnelie } 15480fa1b3ccSPaul Dagnelie 15490fa1b3ccSPaul Dagnelie /* 15500fa1b3ccSPaul Dagnelie * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed 15510fa1b3ccSPaul Dagnelie * streams to refer to a copy of the data that is already on the 15520fa1b3ccSPaul Dagnelie * system because it came in earlier in the stream. This function 15530fa1b3ccSPaul Dagnelie * finds the earlier copy of the data, and uses that copy instead of 15540fa1b3ccSPaul Dagnelie * data from the stream to fulfill this write. 15550fa1b3ccSPaul Dagnelie */ 15560fa1b3ccSPaul Dagnelie static int 15570fa1b3ccSPaul Dagnelie receive_write_byref(struct receive_writer_arg *rwa, 15580fa1b3ccSPaul Dagnelie struct drr_write_byref *drrwbr) 15590fa1b3ccSPaul Dagnelie { 15600fa1b3ccSPaul Dagnelie dmu_tx_t *tx; 15610fa1b3ccSPaul Dagnelie int err; 15620fa1b3ccSPaul Dagnelie guid_map_entry_t gmesrch; 15630fa1b3ccSPaul Dagnelie guid_map_entry_t *gmep; 15640fa1b3ccSPaul Dagnelie avl_index_t where; 15650fa1b3ccSPaul Dagnelie objset_t *ref_os = NULL; 1566eb633035STom Caputi int flags = DMU_READ_PREFETCH; 15670fa1b3ccSPaul Dagnelie dmu_buf_t *dbp; 15680fa1b3ccSPaul Dagnelie 15690fa1b3ccSPaul Dagnelie if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) 15700fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 15710fa1b3ccSPaul Dagnelie 15720fa1b3ccSPaul Dagnelie /* 15730fa1b3ccSPaul Dagnelie * If the GUID of the referenced dataset is different from the 15740fa1b3ccSPaul Dagnelie * GUID of the target dataset, find the referenced dataset. 15750fa1b3ccSPaul Dagnelie */ 15760fa1b3ccSPaul Dagnelie if (drrwbr->drr_toguid != drrwbr->drr_refguid) { 15770fa1b3ccSPaul Dagnelie gmesrch.guid = drrwbr->drr_refguid; 15780fa1b3ccSPaul Dagnelie if ((gmep = avl_find(rwa->guid_to_ds_map, &gmesrch, 15790fa1b3ccSPaul Dagnelie &where)) == NULL) { 15800fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 15810fa1b3ccSPaul Dagnelie } 15820fa1b3ccSPaul Dagnelie if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) 15830fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 15840fa1b3ccSPaul Dagnelie } else { 15850fa1b3ccSPaul Dagnelie ref_os = rwa->os; 15860fa1b3ccSPaul Dagnelie } 15870fa1b3ccSPaul Dagnelie 15880fa1b3ccSPaul Dagnelie if (drrwbr->drr_object > rwa->max_object) 15890fa1b3ccSPaul Dagnelie rwa->max_object = drrwbr->drr_object; 15900fa1b3ccSPaul Dagnelie 1591eb633035STom Caputi if (rwa->raw) 1592eb633035STom Caputi flags |= DMU_READ_NO_DECRYPT; 1593eb633035STom Caputi 1594eb633035STom Caputi /* may return either a regular db or an encrypted one */ 15950fa1b3ccSPaul Dagnelie err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, 1596eb633035STom Caputi drrwbr->drr_refoffset, FTAG, &dbp, flags); 15970fa1b3ccSPaul Dagnelie if (err != 0) 15980fa1b3ccSPaul Dagnelie return (err); 15990fa1b3ccSPaul Dagnelie 16000fa1b3ccSPaul Dagnelie tx = dmu_tx_create(rwa->os); 16010fa1b3ccSPaul Dagnelie 16020fa1b3ccSPaul Dagnelie dmu_tx_hold_write(tx, drrwbr->drr_object, 16030fa1b3ccSPaul Dagnelie drrwbr->drr_offset, drrwbr->drr_length); 16040fa1b3ccSPaul Dagnelie err = dmu_tx_assign(tx, TXG_WAIT); 16050fa1b3ccSPaul Dagnelie if (err != 0) { 16060fa1b3ccSPaul Dagnelie dmu_tx_abort(tx); 16070fa1b3ccSPaul Dagnelie return (err); 16080fa1b3ccSPaul Dagnelie } 1609eb633035STom Caputi 1610eb633035STom Caputi if (rwa->raw) { 1611eb633035STom Caputi dmu_copy_from_buf(rwa->os, drrwbr->drr_object, 1612eb633035STom Caputi drrwbr->drr_offset, dbp, tx); 1613eb633035STom Caputi } else { 1614eb633035STom Caputi dmu_write(rwa->os, drrwbr->drr_object, 1615eb633035STom Caputi drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); 1616eb633035STom Caputi } 16170fa1b3ccSPaul Dagnelie dmu_buf_rele(dbp, FTAG); 16180fa1b3ccSPaul Dagnelie 16190fa1b3ccSPaul Dagnelie /* See comment in restore_write. */ 16200fa1b3ccSPaul Dagnelie save_resume_state(rwa, drrwbr->drr_object, drrwbr->drr_offset, tx); 16210fa1b3ccSPaul Dagnelie dmu_tx_commit(tx); 16220fa1b3ccSPaul Dagnelie return (0); 16230fa1b3ccSPaul Dagnelie } 16240fa1b3ccSPaul Dagnelie 16250fa1b3ccSPaul Dagnelie static int 16260fa1b3ccSPaul Dagnelie receive_write_embedded(struct receive_writer_arg *rwa, 16270fa1b3ccSPaul Dagnelie struct drr_write_embedded *drrwe, void *data) 16280fa1b3ccSPaul Dagnelie { 16290fa1b3ccSPaul Dagnelie dmu_tx_t *tx; 16300fa1b3ccSPaul Dagnelie int err; 16310fa1b3ccSPaul Dagnelie 16320fa1b3ccSPaul Dagnelie if (drrwe->drr_offset + drrwe->drr_length < drrwe->drr_offset) 16330fa1b3ccSPaul Dagnelie return (EINVAL); 16340fa1b3ccSPaul Dagnelie 16350fa1b3ccSPaul Dagnelie if (drrwe->drr_psize > BPE_PAYLOAD_SIZE) 16360fa1b3ccSPaul Dagnelie return (EINVAL); 16370fa1b3ccSPaul Dagnelie 16380fa1b3ccSPaul Dagnelie if (drrwe->drr_etype >= NUM_BP_EMBEDDED_TYPES) 16390fa1b3ccSPaul Dagnelie return (EINVAL); 16400fa1b3ccSPaul Dagnelie if (drrwe->drr_compression >= ZIO_COMPRESS_FUNCTIONS) 16410fa1b3ccSPaul Dagnelie return (EINVAL); 1642eb633035STom Caputi if (rwa->raw) 1643eb633035STom Caputi return (SET_ERROR(EINVAL)); 16440fa1b3ccSPaul Dagnelie 16450fa1b3ccSPaul Dagnelie if (drrwe->drr_object > rwa->max_object) 16460fa1b3ccSPaul Dagnelie rwa->max_object = drrwe->drr_object; 16470fa1b3ccSPaul Dagnelie 16480fa1b3ccSPaul Dagnelie tx = dmu_tx_create(rwa->os); 16490fa1b3ccSPaul Dagnelie 16500fa1b3ccSPaul Dagnelie dmu_tx_hold_write(tx, drrwe->drr_object, 16510fa1b3ccSPaul Dagnelie drrwe->drr_offset, drrwe->drr_length); 16520fa1b3ccSPaul Dagnelie err = dmu_tx_assign(tx, TXG_WAIT); 16530fa1b3ccSPaul Dagnelie if (err != 0) { 16540fa1b3ccSPaul Dagnelie dmu_tx_abort(tx); 16550fa1b3ccSPaul Dagnelie return (err); 16560fa1b3ccSPaul Dagnelie } 16570fa1b3ccSPaul Dagnelie 16580fa1b3ccSPaul Dagnelie dmu_write_embedded(rwa->os, drrwe->drr_object, 16590fa1b3ccSPaul Dagnelie drrwe->drr_offset, data, drrwe->drr_etype, 16600fa1b3ccSPaul Dagnelie drrwe->drr_compression, drrwe->drr_lsize, drrwe->drr_psize, 16610fa1b3ccSPaul Dagnelie rwa->byteswap ^ ZFS_HOST_BYTEORDER, tx); 16620fa1b3ccSPaul Dagnelie 16630fa1b3ccSPaul Dagnelie /* See comment in restore_write. */ 16640fa1b3ccSPaul Dagnelie save_resume_state(rwa, drrwe->drr_object, drrwe->drr_offset, tx); 16650fa1b3ccSPaul Dagnelie dmu_tx_commit(tx); 16660fa1b3ccSPaul Dagnelie return (0); 16670fa1b3ccSPaul Dagnelie } 16680fa1b3ccSPaul Dagnelie 16690fa1b3ccSPaul Dagnelie static int 16700fa1b3ccSPaul Dagnelie receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs, 1671eb633035STom Caputi arc_buf_t *abuf) 16720fa1b3ccSPaul Dagnelie { 16730fa1b3ccSPaul Dagnelie dmu_tx_t *tx; 16740fa1b3ccSPaul Dagnelie dmu_buf_t *db, *db_spill; 16750fa1b3ccSPaul Dagnelie int err; 1676eb633035STom Caputi uint32_t flags = 0; 16770fa1b3ccSPaul Dagnelie 16780fa1b3ccSPaul Dagnelie if (drrs->drr_length < SPA_MINBLOCKSIZE || 16790fa1b3ccSPaul Dagnelie drrs->drr_length > spa_maxblocksize(dmu_objset_spa(rwa->os))) 16800fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 16810fa1b3ccSPaul Dagnelie 1682eb633035STom Caputi /* 1683eb633035STom Caputi * This is an unmodified spill block which was added to the stream 1684eb633035STom Caputi * to resolve an issue with incorrectly removing spill blocks. It 1685eb633035STom Caputi * should be ignored by current versions of the code which support 1686eb633035STom Caputi * the DRR_FLAG_SPILL_BLOCK flag. 1687eb633035STom Caputi */ 1688eb633035STom Caputi if (rwa->spill && DRR_SPILL_IS_UNMODIFIED(drrs->drr_flags)) { 1689eb633035STom Caputi dmu_return_arcbuf(abuf); 1690eb633035STom Caputi return (0); 1691eb633035STom Caputi } 1692eb633035STom Caputi 1693eb633035STom Caputi if (rwa->raw) { 1694eb633035STom Caputi if (!DMU_OT_IS_VALID(drrs->drr_type) || 1695eb633035STom Caputi drrs->drr_compressiontype >= ZIO_COMPRESS_FUNCTIONS || 1696eb633035STom Caputi drrs->drr_compressed_size == 0) 1697eb633035STom Caputi return (SET_ERROR(EINVAL)); 1698eb633035STom Caputi 1699eb633035STom Caputi flags |= DMU_READ_NO_DECRYPT; 1700eb633035STom Caputi } 1701eb633035STom Caputi 17020fa1b3ccSPaul Dagnelie if (dmu_object_info(rwa->os, drrs->drr_object, NULL) != 0) 17030fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 17040fa1b3ccSPaul Dagnelie 17050fa1b3ccSPaul Dagnelie if (drrs->drr_object > rwa->max_object) 17060fa1b3ccSPaul Dagnelie rwa->max_object = drrs->drr_object; 17070fa1b3ccSPaul Dagnelie 17080fa1b3ccSPaul Dagnelie VERIFY0(dmu_bonus_hold(rwa->os, drrs->drr_object, FTAG, &db)); 1709eb633035STom Caputi if ((err = dmu_spill_hold_by_bonus(db, DMU_READ_NO_DECRYPT, FTAG, 1710eb633035STom Caputi &db_spill)) != 0) { 17110fa1b3ccSPaul Dagnelie dmu_buf_rele(db, FTAG); 17120fa1b3ccSPaul Dagnelie return (err); 17130fa1b3ccSPaul Dagnelie } 17140fa1b3ccSPaul Dagnelie 17150fa1b3ccSPaul Dagnelie tx = dmu_tx_create(rwa->os); 17160fa1b3ccSPaul Dagnelie 17170fa1b3ccSPaul Dagnelie dmu_tx_hold_spill(tx, db->db_object); 17180fa1b3ccSPaul Dagnelie 17190fa1b3ccSPaul Dagnelie err = dmu_tx_assign(tx, TXG_WAIT); 17200fa1b3ccSPaul Dagnelie if (err != 0) { 17210fa1b3ccSPaul Dagnelie dmu_buf_rele(db, FTAG); 17220fa1b3ccSPaul Dagnelie dmu_buf_rele(db_spill, FTAG); 17230fa1b3ccSPaul Dagnelie dmu_tx_abort(tx); 17240fa1b3ccSPaul Dagnelie return (err); 17250fa1b3ccSPaul Dagnelie } 17260fa1b3ccSPaul Dagnelie 1727eb633035STom Caputi /* 1728eb633035STom Caputi * Spill blocks may both grow and shrink. When a change in size 1729eb633035STom Caputi * occurs any existing dbuf must be updated to match the logical 1730eb633035STom Caputi * size of the provided arc_buf_t. 1731eb633035STom Caputi */ 1732eb633035STom Caputi if (db_spill->db_size != drrs->drr_length) { 1733eb633035STom Caputi dmu_buf_will_fill(db_spill, tx); 17340fa1b3ccSPaul Dagnelie VERIFY(0 == dbuf_spill_set_blksz(db_spill, 17350fa1b3ccSPaul Dagnelie drrs->drr_length, tx)); 1736eb633035STom Caputi } 1737eb633035STom Caputi 1738eb633035STom Caputi if (rwa->byteswap && !arc_is_encrypted(abuf) && 1739eb633035STom Caputi arc_get_compression(abuf) == ZIO_COMPRESS_OFF) { 1740eb633035STom Caputi dmu_object_byteswap_t byteswap = 1741eb633035STom Caputi DMU_OT_BYTESWAP(drrs->drr_type); 1742eb633035STom Caputi dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, 1743eb633035STom Caputi DRR_SPILL_PAYLOAD_SIZE(drrs)); 1744eb633035STom Caputi } 1745eb633035STom Caputi 1746eb633035STom Caputi dbuf_assign_arcbuf((dmu_buf_impl_t *)db_spill, abuf, tx); 17470fa1b3ccSPaul Dagnelie 17480fa1b3ccSPaul Dagnelie dmu_buf_rele(db, FTAG); 17490fa1b3ccSPaul Dagnelie dmu_buf_rele(db_spill, FTAG); 17500fa1b3ccSPaul Dagnelie 17510fa1b3ccSPaul Dagnelie dmu_tx_commit(tx); 17520fa1b3ccSPaul Dagnelie return (0); 17530fa1b3ccSPaul Dagnelie } 17540fa1b3ccSPaul Dagnelie 17550fa1b3ccSPaul Dagnelie /* ARGSUSED */ 17560fa1b3ccSPaul Dagnelie static int 17570fa1b3ccSPaul Dagnelie receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf) 17580fa1b3ccSPaul Dagnelie { 17590fa1b3ccSPaul Dagnelie int err; 17600fa1b3ccSPaul Dagnelie 1761eb633035STom Caputi if (drrf->drr_length != DMU_OBJECT_END && 17620fa1b3ccSPaul Dagnelie drrf->drr_offset + drrf->drr_length < drrf->drr_offset) 17630fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 17640fa1b3ccSPaul Dagnelie 17650fa1b3ccSPaul Dagnelie if (dmu_object_info(rwa->os, drrf->drr_object, NULL) != 0) 17660fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 17670fa1b3ccSPaul Dagnelie 17680fa1b3ccSPaul Dagnelie if (drrf->drr_object > rwa->max_object) 17690fa1b3ccSPaul Dagnelie rwa->max_object = drrf->drr_object; 17700fa1b3ccSPaul Dagnelie 17710fa1b3ccSPaul Dagnelie err = dmu_free_long_range(rwa->os, drrf->drr_object, 17720fa1b3ccSPaul Dagnelie drrf->drr_offset, drrf->drr_length); 17730fa1b3ccSPaul Dagnelie 17740fa1b3ccSPaul Dagnelie return (err); 17750fa1b3ccSPaul Dagnelie } 17760fa1b3ccSPaul Dagnelie 1777eb633035STom Caputi static int 1778eb633035STom Caputi receive_object_range(struct receive_writer_arg *rwa, 1779eb633035STom Caputi struct drr_object_range *drror) 1780eb633035STom Caputi { 1781eb633035STom Caputi /* 1782eb633035STom Caputi * By default, we assume this block is in our native format 1783eb633035STom Caputi * (ZFS_HOST_BYTEORDER). We then take into account whether 1784eb633035STom Caputi * the send stream is byteswapped (rwa->byteswap). Finally, 1785eb633035STom Caputi * we need to byteswap again if this particular block was 1786eb633035STom Caputi * in non-native format on the send side. 1787eb633035STom Caputi */ 1788eb633035STom Caputi boolean_t byteorder = ZFS_HOST_BYTEORDER ^ rwa->byteswap ^ 1789eb633035STom Caputi !!DRR_IS_RAW_BYTESWAPPED(drror->drr_flags); 1790eb633035STom Caputi 1791eb633035STom Caputi /* 1792eb633035STom Caputi * Since dnode block sizes are constant, we should not need to worry 1793eb633035STom Caputi * about making sure that the dnode block size is the same on the 1794eb633035STom Caputi * sending and receiving sides for the time being. For non-raw sends, 1795eb633035STom Caputi * this does not matter (and in fact we do not send a DRR_OBJECT_RANGE 1796eb633035STom Caputi * record at all). Raw sends require this record type because the 1797eb633035STom Caputi * encryption parameters are used to protect an entire block of bonus 1798eb633035STom Caputi * buffers. If the size of dnode blocks ever becomes variable, 1799eb633035STom Caputi * handling will need to be added to ensure that dnode block sizes 1800eb633035STom Caputi * match on the sending and receiving side. 1801eb633035STom Caputi */ 1802eb633035STom Caputi if (drror->drr_numslots != DNODES_PER_BLOCK || 1803eb633035STom Caputi P2PHASE(drror->drr_firstobj, DNODES_PER_BLOCK) != 0 || 1804eb633035STom Caputi !rwa->raw) 1805eb633035STom Caputi return (SET_ERROR(EINVAL)); 1806eb633035STom Caputi 1807eb633035STom Caputi if (drror->drr_firstobj > rwa->max_object) 1808eb633035STom Caputi rwa->max_object = drror->drr_firstobj; 1809eb633035STom Caputi 1810eb633035STom Caputi /* 1811eb633035STom Caputi * The DRR_OBJECT_RANGE handling must be deferred to receive_object() 1812eb633035STom Caputi * so that the block of dnodes is not written out when it's empty, 1813eb633035STom Caputi * and converted to a HOLE BP. 1814eb633035STom Caputi */ 1815eb633035STom Caputi rwa->or_crypt_params_present = B_TRUE; 1816eb633035STom Caputi rwa->or_firstobj = drror->drr_firstobj; 1817eb633035STom Caputi rwa->or_numslots = drror->drr_numslots; 1818eb633035STom Caputi bcopy(drror->drr_salt, rwa->or_salt, ZIO_DATA_SALT_LEN); 1819eb633035STom Caputi bcopy(drror->drr_iv, rwa->or_iv, ZIO_DATA_IV_LEN); 1820eb633035STom Caputi bcopy(drror->drr_mac, rwa->or_mac, ZIO_DATA_MAC_LEN); 1821eb633035STom Caputi rwa->or_byteorder = byteorder; 1822eb633035STom Caputi 1823eb633035STom Caputi return (0); 1824eb633035STom Caputi } 1825eb633035STom Caputi 18260fa1b3ccSPaul Dagnelie /* used to destroy the drc_ds on error */ 18270fa1b3ccSPaul Dagnelie static void 18280fa1b3ccSPaul Dagnelie dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) 18290fa1b3ccSPaul Dagnelie { 1830eb633035STom Caputi dsl_dataset_t *ds = drc->drc_ds; 1831eb633035STom Caputi ds_hold_flags_t dsflags = (drc->drc_raw) ? 0 : DS_HOLD_FLAG_DECRYPT; 1832eb633035STom Caputi 1833eb633035STom Caputi /* 1834eb633035STom Caputi * Wait for the txg sync before cleaning up the receive. For 1835eb633035STom Caputi * resumable receives, this ensures that our resume state has 1836eb633035STom Caputi * been written out to disk. For raw receives, this ensures 1837eb633035STom Caputi * that the user accounting code will not attempt to do anything 1838eb633035STom Caputi * after we stopped receiving the dataset. 1839eb633035STom Caputi */ 1840eb633035STom Caputi txg_wait_synced(ds->ds_dir->dd_pool, 0); 1841eb633035STom Caputi ds->ds_objset->os_raw_receive = B_FALSE; 1842eb633035STom Caputi 1843eb633035STom Caputi rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1844eb633035STom Caputi if (drc->drc_resumable && !BP_IS_HOLE(dsl_dataset_get_blkptr(ds))) { 1845eb633035STom Caputi rrw_exit(&ds->ds_bp_rwlock, FTAG); 1846eb633035STom Caputi dsl_dataset_disown(ds, dsflags, dmu_recv_tag); 18470fa1b3ccSPaul Dagnelie } else { 18480fa1b3ccSPaul Dagnelie char name[ZFS_MAX_DATASET_NAME_LEN]; 1849eb633035STom Caputi rrw_exit(&ds->ds_bp_rwlock, FTAG); 1850eb633035STom Caputi dsl_dataset_name(ds, name); 1851eb633035STom Caputi dsl_dataset_disown(ds, dsflags, dmu_recv_tag); 18520fa1b3ccSPaul Dagnelie (void) dsl_destroy_head(name); 18530fa1b3ccSPaul Dagnelie } 18540fa1b3ccSPaul Dagnelie } 18550fa1b3ccSPaul Dagnelie 18560fa1b3ccSPaul Dagnelie static void 18570fa1b3ccSPaul Dagnelie receive_cksum(struct receive_arg *ra, int len, void *buf) 18580fa1b3ccSPaul Dagnelie { 18590fa1b3ccSPaul Dagnelie if (ra->byteswap) { 18600fa1b3ccSPaul Dagnelie (void) fletcher_4_incremental_byteswap(buf, len, &ra->cksum); 18610fa1b3ccSPaul Dagnelie } else { 18620fa1b3ccSPaul Dagnelie (void) fletcher_4_incremental_native(buf, len, &ra->cksum); 18630fa1b3ccSPaul Dagnelie } 18640fa1b3ccSPaul Dagnelie } 18650fa1b3ccSPaul Dagnelie 18660fa1b3ccSPaul Dagnelie /* 18670fa1b3ccSPaul Dagnelie * Read the payload into a buffer of size len, and update the current record's 18680fa1b3ccSPaul Dagnelie * payload field. 18690fa1b3ccSPaul Dagnelie * Allocate ra->next_rrd and read the next record's header into 18700fa1b3ccSPaul Dagnelie * ra->next_rrd->header. 18710fa1b3ccSPaul Dagnelie * Verify checksum of payload and next record. 18720fa1b3ccSPaul Dagnelie */ 18730fa1b3ccSPaul Dagnelie static int 18740fa1b3ccSPaul Dagnelie receive_read_payload_and_next_header(struct receive_arg *ra, int len, void *buf) 18750fa1b3ccSPaul Dagnelie { 18760fa1b3ccSPaul Dagnelie int err; 18770fa1b3ccSPaul Dagnelie 18780fa1b3ccSPaul Dagnelie if (len != 0) { 18790fa1b3ccSPaul Dagnelie ASSERT3U(len, <=, SPA_MAXBLOCKSIZE); 18800fa1b3ccSPaul Dagnelie err = receive_read(ra, len, buf); 18810fa1b3ccSPaul Dagnelie if (err != 0) 18820fa1b3ccSPaul Dagnelie return (err); 18830fa1b3ccSPaul Dagnelie receive_cksum(ra, len, buf); 18840fa1b3ccSPaul Dagnelie 18850fa1b3ccSPaul Dagnelie /* note: rrd is NULL when reading the begin record's payload */ 18860fa1b3ccSPaul Dagnelie if (ra->rrd != NULL) { 18870fa1b3ccSPaul Dagnelie ra->rrd->payload = buf; 18880fa1b3ccSPaul Dagnelie ra->rrd->payload_size = len; 18890fa1b3ccSPaul Dagnelie ra->rrd->bytes_read = ra->bytes_read; 18900fa1b3ccSPaul Dagnelie } 18910fa1b3ccSPaul Dagnelie } 18920fa1b3ccSPaul Dagnelie 18930fa1b3ccSPaul Dagnelie ra->prev_cksum = ra->cksum; 18940fa1b3ccSPaul Dagnelie 18950fa1b3ccSPaul Dagnelie ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); 18960fa1b3ccSPaul Dagnelie err = receive_read(ra, sizeof (ra->next_rrd->header), 18970fa1b3ccSPaul Dagnelie &ra->next_rrd->header); 18980fa1b3ccSPaul Dagnelie ra->next_rrd->bytes_read = ra->bytes_read; 1899eb633035STom Caputi 19000fa1b3ccSPaul Dagnelie if (err != 0) { 19010fa1b3ccSPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 19020fa1b3ccSPaul Dagnelie ra->next_rrd = NULL; 19030fa1b3ccSPaul Dagnelie return (err); 19040fa1b3ccSPaul Dagnelie } 19050fa1b3ccSPaul Dagnelie if (ra->next_rrd->header.drr_type == DRR_BEGIN) { 19060fa1b3ccSPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 19070fa1b3ccSPaul Dagnelie ra->next_rrd = NULL; 19080fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 19090fa1b3ccSPaul Dagnelie } 19100fa1b3ccSPaul Dagnelie 19110fa1b3ccSPaul Dagnelie /* 19120fa1b3ccSPaul Dagnelie * Note: checksum is of everything up to but not including the 19130fa1b3ccSPaul Dagnelie * checksum itself. 19140fa1b3ccSPaul Dagnelie */ 19150fa1b3ccSPaul Dagnelie ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 19160fa1b3ccSPaul Dagnelie ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 19170fa1b3ccSPaul Dagnelie receive_cksum(ra, 19180fa1b3ccSPaul Dagnelie offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 19190fa1b3ccSPaul Dagnelie &ra->next_rrd->header); 19200fa1b3ccSPaul Dagnelie 19210fa1b3ccSPaul Dagnelie zio_cksum_t cksum_orig = 19220fa1b3ccSPaul Dagnelie ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; 19230fa1b3ccSPaul Dagnelie zio_cksum_t *cksump = 19240fa1b3ccSPaul Dagnelie &ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; 19250fa1b3ccSPaul Dagnelie 19260fa1b3ccSPaul Dagnelie if (ra->byteswap) 19270fa1b3ccSPaul Dagnelie byteswap_record(&ra->next_rrd->header); 19280fa1b3ccSPaul Dagnelie 19290fa1b3ccSPaul Dagnelie if ((!ZIO_CHECKSUM_IS_ZERO(cksump)) && 19300fa1b3ccSPaul Dagnelie !ZIO_CHECKSUM_EQUAL(ra->cksum, *cksump)) { 19310fa1b3ccSPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 19320fa1b3ccSPaul Dagnelie ra->next_rrd = NULL; 19330fa1b3ccSPaul Dagnelie return (SET_ERROR(ECKSUM)); 19340fa1b3ccSPaul Dagnelie } 19350fa1b3ccSPaul Dagnelie 19360fa1b3ccSPaul Dagnelie receive_cksum(ra, sizeof (cksum_orig), &cksum_orig); 19370fa1b3ccSPaul Dagnelie 19380fa1b3ccSPaul Dagnelie return (0); 19390fa1b3ccSPaul Dagnelie } 19400fa1b3ccSPaul Dagnelie 19410fa1b3ccSPaul Dagnelie static void 19420fa1b3ccSPaul Dagnelie objlist_create(struct objlist *list) 19430fa1b3ccSPaul Dagnelie { 19440fa1b3ccSPaul Dagnelie list_create(&list->list, sizeof (struct receive_objnode), 19450fa1b3ccSPaul Dagnelie offsetof(struct receive_objnode, node)); 19460fa1b3ccSPaul Dagnelie list->last_lookup = 0; 19470fa1b3ccSPaul Dagnelie } 19480fa1b3ccSPaul Dagnelie 19490fa1b3ccSPaul Dagnelie static void 19500fa1b3ccSPaul Dagnelie objlist_destroy(struct objlist *list) 19510fa1b3ccSPaul Dagnelie { 19520fa1b3ccSPaul Dagnelie for (struct receive_objnode *n = list_remove_head(&list->list); 19530fa1b3ccSPaul Dagnelie n != NULL; n = list_remove_head(&list->list)) { 19540fa1b3ccSPaul Dagnelie kmem_free(n, sizeof (*n)); 19550fa1b3ccSPaul Dagnelie } 19560fa1b3ccSPaul Dagnelie list_destroy(&list->list); 19570fa1b3ccSPaul Dagnelie } 19580fa1b3ccSPaul Dagnelie 19590fa1b3ccSPaul Dagnelie /* 19600fa1b3ccSPaul Dagnelie * This function looks through the objlist to see if the specified object number 19610fa1b3ccSPaul Dagnelie * is contained in the objlist. In the process, it will remove all object 19620fa1b3ccSPaul Dagnelie * numbers in the list that are smaller than the specified object number. Thus, 19630fa1b3ccSPaul Dagnelie * any lookup of an object number smaller than a previously looked up object 19640fa1b3ccSPaul Dagnelie * number will always return false; therefore, all lookups should be done in 19650fa1b3ccSPaul Dagnelie * ascending order. 19660fa1b3ccSPaul Dagnelie */ 19670fa1b3ccSPaul Dagnelie static boolean_t 19680fa1b3ccSPaul Dagnelie objlist_exists(struct objlist *list, uint64_t object) 19690fa1b3ccSPaul Dagnelie { 19700fa1b3ccSPaul Dagnelie struct receive_objnode *node = list_head(&list->list); 19710fa1b3ccSPaul Dagnelie ASSERT3U(object, >=, list->last_lookup); 19720fa1b3ccSPaul Dagnelie list->last_lookup = object; 19730fa1b3ccSPaul Dagnelie while (node != NULL && node->object < object) { 19740fa1b3ccSPaul Dagnelie VERIFY3P(node, ==, list_remove_head(&list->list)); 19750fa1b3ccSPaul Dagnelie kmem_free(node, sizeof (*node)); 19760fa1b3ccSPaul Dagnelie node = list_head(&list->list); 19770fa1b3ccSPaul Dagnelie } 19780fa1b3ccSPaul Dagnelie return (node != NULL && node->object == object); 19790fa1b3ccSPaul Dagnelie } 19800fa1b3ccSPaul Dagnelie 19810fa1b3ccSPaul Dagnelie /* 19820fa1b3ccSPaul Dagnelie * The objlist is a list of object numbers stored in ascending order. However, 19830fa1b3ccSPaul Dagnelie * the insertion of new object numbers does not seek out the correct location to 19840fa1b3ccSPaul Dagnelie * store a new object number; instead, it appends it to the list for simplicity. 19850fa1b3ccSPaul Dagnelie * Thus, any users must take care to only insert new object numbers in ascending 19860fa1b3ccSPaul Dagnelie * order. 19870fa1b3ccSPaul Dagnelie */ 19880fa1b3ccSPaul Dagnelie static void 19890fa1b3ccSPaul Dagnelie objlist_insert(struct objlist *list, uint64_t object) 19900fa1b3ccSPaul Dagnelie { 19910fa1b3ccSPaul Dagnelie struct receive_objnode *node = kmem_zalloc(sizeof (*node), KM_SLEEP); 19920fa1b3ccSPaul Dagnelie node->object = object; 19930fa1b3ccSPaul Dagnelie #ifdef ZFS_DEBUG 19940fa1b3ccSPaul Dagnelie struct receive_objnode *last_object = list_tail(&list->list); 19950fa1b3ccSPaul Dagnelie uint64_t last_objnum = (last_object != NULL ? last_object->object : 0); 19960fa1b3ccSPaul Dagnelie ASSERT3U(node->object, >, last_objnum); 19970fa1b3ccSPaul Dagnelie #endif 19980fa1b3ccSPaul Dagnelie list_insert_tail(&list->list, node); 19990fa1b3ccSPaul Dagnelie } 20000fa1b3ccSPaul Dagnelie 20010fa1b3ccSPaul Dagnelie /* 20020fa1b3ccSPaul Dagnelie * Issue the prefetch reads for any necessary indirect blocks. 20030fa1b3ccSPaul Dagnelie * 20040fa1b3ccSPaul Dagnelie * We use the object ignore list to tell us whether or not to issue prefetches 20050fa1b3ccSPaul Dagnelie * for a given object. We do this for both correctness (in case the blocksize 20060fa1b3ccSPaul Dagnelie * of an object has changed) and performance (if the object doesn't exist, don't 20070fa1b3ccSPaul Dagnelie * needlessly try to issue prefetches). We also trim the list as we go through 20080fa1b3ccSPaul Dagnelie * the stream to prevent it from growing to an unbounded size. 20090fa1b3ccSPaul Dagnelie * 20100fa1b3ccSPaul Dagnelie * The object numbers within will always be in sorted order, and any write 20110fa1b3ccSPaul Dagnelie * records we see will also be in sorted order, but they're not sorted with 20120fa1b3ccSPaul Dagnelie * respect to each other (i.e. we can get several object records before 20130fa1b3ccSPaul Dagnelie * receiving each object's write records). As a result, once we've reached a 20140fa1b3ccSPaul Dagnelie * given object number, we can safely remove any reference to lower object 20150fa1b3ccSPaul Dagnelie * numbers in the ignore list. In practice, we receive up to 32 object records 20160fa1b3ccSPaul Dagnelie * before receiving write records, so the list can have up to 32 nodes in it. 20170fa1b3ccSPaul Dagnelie */ 20180fa1b3ccSPaul Dagnelie /* ARGSUSED */ 20190fa1b3ccSPaul Dagnelie static void 20200fa1b3ccSPaul Dagnelie receive_read_prefetch(struct receive_arg *ra, 20210fa1b3ccSPaul Dagnelie uint64_t object, uint64_t offset, uint64_t length) 20220fa1b3ccSPaul Dagnelie { 20230fa1b3ccSPaul Dagnelie if (!objlist_exists(&ra->ignore_objlist, object)) { 20240fa1b3ccSPaul Dagnelie dmu_prefetch(ra->os, object, 1, offset, length, 20250fa1b3ccSPaul Dagnelie ZIO_PRIORITY_SYNC_READ); 20260fa1b3ccSPaul Dagnelie } 20270fa1b3ccSPaul Dagnelie } 20280fa1b3ccSPaul Dagnelie 20290fa1b3ccSPaul Dagnelie /* 20300fa1b3ccSPaul Dagnelie * Read records off the stream, issuing any necessary prefetches. 20310fa1b3ccSPaul Dagnelie */ 20320fa1b3ccSPaul Dagnelie static int 20330fa1b3ccSPaul Dagnelie receive_read_record(struct receive_arg *ra) 20340fa1b3ccSPaul Dagnelie { 20350fa1b3ccSPaul Dagnelie int err; 20360fa1b3ccSPaul Dagnelie 20370fa1b3ccSPaul Dagnelie switch (ra->rrd->header.drr_type) { 20380fa1b3ccSPaul Dagnelie case DRR_OBJECT: 20390fa1b3ccSPaul Dagnelie { 20400fa1b3ccSPaul Dagnelie struct drr_object *drro = &ra->rrd->header.drr_u.drr_object; 2041eb633035STom Caputi uint32_t size = DRR_OBJECT_PAYLOAD_SIZE(drro); 2042eb633035STom Caputi void *buf = NULL; 20430fa1b3ccSPaul Dagnelie dmu_object_info_t doi; 2044eb633035STom Caputi 2045eb633035STom Caputi if (size != 0) 2046eb633035STom Caputi buf = kmem_zalloc(size, KM_SLEEP); 2047eb633035STom Caputi 20480fa1b3ccSPaul Dagnelie err = receive_read_payload_and_next_header(ra, size, buf); 20490fa1b3ccSPaul Dagnelie if (err != 0) { 20500fa1b3ccSPaul Dagnelie kmem_free(buf, size); 20510fa1b3ccSPaul Dagnelie return (err); 20520fa1b3ccSPaul Dagnelie } 20530fa1b3ccSPaul Dagnelie err = dmu_object_info(ra->os, drro->drr_object, &doi); 20540fa1b3ccSPaul Dagnelie /* 20550fa1b3ccSPaul Dagnelie * See receive_read_prefetch for an explanation why we're 20560fa1b3ccSPaul Dagnelie * storing this object in the ignore_obj_list. 20570fa1b3ccSPaul Dagnelie */ 2058eb633035STom Caputi if (err == ENOENT || err == EEXIST || 20590fa1b3ccSPaul Dagnelie (err == 0 && doi.doi_data_block_size != drro->drr_blksz)) { 20600fa1b3ccSPaul Dagnelie objlist_insert(&ra->ignore_objlist, drro->drr_object); 20610fa1b3ccSPaul Dagnelie err = 0; 20620fa1b3ccSPaul Dagnelie } 20630fa1b3ccSPaul Dagnelie return (err); 20640fa1b3ccSPaul Dagnelie } 20650fa1b3ccSPaul Dagnelie case DRR_FREEOBJECTS: 20660fa1b3ccSPaul Dagnelie { 20670fa1b3ccSPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 20680fa1b3ccSPaul Dagnelie return (err); 20690fa1b3ccSPaul Dagnelie } 20700fa1b3ccSPaul Dagnelie case DRR_WRITE: 20710fa1b3ccSPaul Dagnelie { 20720fa1b3ccSPaul Dagnelie struct drr_write *drrw = &ra->rrd->header.drr_u.drr_write; 20730fa1b3ccSPaul Dagnelie arc_buf_t *abuf; 20740fa1b3ccSPaul Dagnelie boolean_t is_meta = DMU_OT_IS_METADATA(drrw->drr_type); 2075eb633035STom Caputi 2076eb633035STom Caputi if (ra->raw) { 2077eb633035STom Caputi boolean_t byteorder = ZFS_HOST_BYTEORDER ^ 2078eb633035STom Caputi !!DRR_IS_RAW_BYTESWAPPED(drrw->drr_flags) ^ 2079eb633035STom Caputi ra->byteswap; 2080eb633035STom Caputi 2081eb633035STom Caputi abuf = arc_loan_raw_buf(dmu_objset_spa(ra->os), 2082eb633035STom Caputi drrw->drr_object, byteorder, drrw->drr_salt, 2083eb633035STom Caputi drrw->drr_iv, drrw->drr_mac, drrw->drr_type, 2084eb633035STom Caputi drrw->drr_compressed_size, drrw->drr_logical_size, 2085eb633035STom Caputi drrw->drr_compressiontype); 2086eb633035STom Caputi } else if (DRR_WRITE_COMPRESSED(drrw)) { 20870fa1b3ccSPaul Dagnelie ASSERT3U(drrw->drr_compressed_size, >, 0); 20880fa1b3ccSPaul Dagnelie ASSERT3U(drrw->drr_logical_size, >=, 20890fa1b3ccSPaul Dagnelie drrw->drr_compressed_size); 20900fa1b3ccSPaul Dagnelie ASSERT(!is_meta); 20910fa1b3ccSPaul Dagnelie abuf = arc_loan_compressed_buf( 20920fa1b3ccSPaul Dagnelie dmu_objset_spa(ra->os), 20930fa1b3ccSPaul Dagnelie drrw->drr_compressed_size, drrw->drr_logical_size, 20940fa1b3ccSPaul Dagnelie drrw->drr_compressiontype); 20950fa1b3ccSPaul Dagnelie } else { 20960fa1b3ccSPaul Dagnelie abuf = arc_loan_buf(dmu_objset_spa(ra->os), 20970fa1b3ccSPaul Dagnelie is_meta, drrw->drr_logical_size); 20980fa1b3ccSPaul Dagnelie } 20990fa1b3ccSPaul Dagnelie 21000fa1b3ccSPaul Dagnelie err = receive_read_payload_and_next_header(ra, 21010fa1b3ccSPaul Dagnelie DRR_WRITE_PAYLOAD_SIZE(drrw), abuf->b_data); 21020fa1b3ccSPaul Dagnelie if (err != 0) { 21030fa1b3ccSPaul Dagnelie dmu_return_arcbuf(abuf); 21040fa1b3ccSPaul Dagnelie return (err); 21050fa1b3ccSPaul Dagnelie } 2106eb633035STom Caputi ra->rrd->arc_buf = abuf; 21070fa1b3ccSPaul Dagnelie receive_read_prefetch(ra, drrw->drr_object, drrw->drr_offset, 21080fa1b3ccSPaul Dagnelie drrw->drr_logical_size); 21090fa1b3ccSPaul Dagnelie return (err); 21100fa1b3ccSPaul Dagnelie } 21110fa1b3ccSPaul Dagnelie case DRR_WRITE_BYREF: 21120fa1b3ccSPaul Dagnelie { 21130fa1b3ccSPaul Dagnelie struct drr_write_byref *drrwb = 21140fa1b3ccSPaul Dagnelie &ra->rrd->header.drr_u.drr_write_byref; 21150fa1b3ccSPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 21160fa1b3ccSPaul Dagnelie receive_read_prefetch(ra, drrwb->drr_object, drrwb->drr_offset, 21170fa1b3ccSPaul Dagnelie drrwb->drr_length); 21180fa1b3ccSPaul Dagnelie return (err); 21190fa1b3ccSPaul Dagnelie } 21200fa1b3ccSPaul Dagnelie case DRR_WRITE_EMBEDDED: 21210fa1b3ccSPaul Dagnelie { 21220fa1b3ccSPaul Dagnelie struct drr_write_embedded *drrwe = 21230fa1b3ccSPaul Dagnelie &ra->rrd->header.drr_u.drr_write_embedded; 21240fa1b3ccSPaul Dagnelie uint32_t size = P2ROUNDUP(drrwe->drr_psize, 8); 21250fa1b3ccSPaul Dagnelie void *buf = kmem_zalloc(size, KM_SLEEP); 21260fa1b3ccSPaul Dagnelie 21270fa1b3ccSPaul Dagnelie err = receive_read_payload_and_next_header(ra, size, buf); 21280fa1b3ccSPaul Dagnelie if (err != 0) { 21290fa1b3ccSPaul Dagnelie kmem_free(buf, size); 21300fa1b3ccSPaul Dagnelie return (err); 21310fa1b3ccSPaul Dagnelie } 21320fa1b3ccSPaul Dagnelie 21330fa1b3ccSPaul Dagnelie receive_read_prefetch(ra, drrwe->drr_object, drrwe->drr_offset, 21340fa1b3ccSPaul Dagnelie drrwe->drr_length); 21350fa1b3ccSPaul Dagnelie return (err); 21360fa1b3ccSPaul Dagnelie } 21370fa1b3ccSPaul Dagnelie case DRR_FREE: 21380fa1b3ccSPaul Dagnelie { 21390fa1b3ccSPaul Dagnelie /* 21400fa1b3ccSPaul Dagnelie * It might be beneficial to prefetch indirect blocks here, but 21410fa1b3ccSPaul Dagnelie * we don't really have the data to decide for sure. 21420fa1b3ccSPaul Dagnelie */ 21430fa1b3ccSPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 21440fa1b3ccSPaul Dagnelie return (err); 21450fa1b3ccSPaul Dagnelie } 21460fa1b3ccSPaul Dagnelie case DRR_END: 21470fa1b3ccSPaul Dagnelie { 21480fa1b3ccSPaul Dagnelie struct drr_end *drre = &ra->rrd->header.drr_u.drr_end; 21490fa1b3ccSPaul Dagnelie if (!ZIO_CHECKSUM_EQUAL(ra->prev_cksum, drre->drr_checksum)) 21500fa1b3ccSPaul Dagnelie return (SET_ERROR(ECKSUM)); 21510fa1b3ccSPaul Dagnelie return (0); 21520fa1b3ccSPaul Dagnelie } 21530fa1b3ccSPaul Dagnelie case DRR_SPILL: 21540fa1b3ccSPaul Dagnelie { 21550fa1b3ccSPaul Dagnelie struct drr_spill *drrs = &ra->rrd->header.drr_u.drr_spill; 2156eb633035STom Caputi arc_buf_t *abuf; 2157eb633035STom Caputi int len = DRR_SPILL_PAYLOAD_SIZE(drrs); 2158eb633035STom Caputi 2159eb633035STom Caputi /* DRR_SPILL records are either raw or uncompressed */ 2160eb633035STom Caputi if (ra->raw) { 2161eb633035STom Caputi boolean_t byteorder = ZFS_HOST_BYTEORDER ^ 2162eb633035STom Caputi !!DRR_IS_RAW_BYTESWAPPED(drrs->drr_flags) ^ 2163eb633035STom Caputi ra->byteswap; 2164eb633035STom Caputi 2165eb633035STom Caputi abuf = arc_loan_raw_buf(dmu_objset_spa(ra->os), 2166eb633035STom Caputi dmu_objset_id(ra->os), byteorder, drrs->drr_salt, 2167eb633035STom Caputi drrs->drr_iv, drrs->drr_mac, drrs->drr_type, 2168eb633035STom Caputi drrs->drr_compressed_size, drrs->drr_length, 2169eb633035STom Caputi drrs->drr_compressiontype); 2170eb633035STom Caputi } else { 2171eb633035STom Caputi abuf = arc_loan_buf(dmu_objset_spa(ra->os), 2172eb633035STom Caputi DMU_OT_IS_METADATA(drrs->drr_type), 2173eb633035STom Caputi drrs->drr_length); 2174eb633035STom Caputi } 2175eb633035STom Caputi 2176eb633035STom Caputi err = receive_read_payload_and_next_header(ra, len, 2177eb633035STom Caputi abuf->b_data); 2178eb633035STom Caputi if (err != 0) { 2179eb633035STom Caputi dmu_return_arcbuf(abuf); 2180eb633035STom Caputi return (err); 2181eb633035STom Caputi } 2182eb633035STom Caputi ra->rrd->arc_buf = abuf; 2183eb633035STom Caputi return (err); 2184eb633035STom Caputi } 2185eb633035STom Caputi case DRR_OBJECT_RANGE: 2186eb633035STom Caputi { 2187eb633035STom Caputi err = receive_read_payload_and_next_header(ra, 0, NULL); 21880fa1b3ccSPaul Dagnelie return (err); 21890fa1b3ccSPaul Dagnelie } 21900fa1b3ccSPaul Dagnelie default: 21910fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 21920fa1b3ccSPaul Dagnelie } 21930fa1b3ccSPaul Dagnelie } 21940fa1b3ccSPaul Dagnelie 21950fa1b3ccSPaul Dagnelie /* 21960fa1b3ccSPaul Dagnelie * Commit the records to the pool. 21970fa1b3ccSPaul Dagnelie */ 21980fa1b3ccSPaul Dagnelie static int 21990fa1b3ccSPaul Dagnelie receive_process_record(struct receive_writer_arg *rwa, 22000fa1b3ccSPaul Dagnelie struct receive_record_arg *rrd) 22010fa1b3ccSPaul Dagnelie { 22020fa1b3ccSPaul Dagnelie int err; 22030fa1b3ccSPaul Dagnelie 22040fa1b3ccSPaul Dagnelie /* Processing in order, therefore bytes_read should be increasing. */ 22050fa1b3ccSPaul Dagnelie ASSERT3U(rrd->bytes_read, >=, rwa->bytes_read); 22060fa1b3ccSPaul Dagnelie rwa->bytes_read = rrd->bytes_read; 22070fa1b3ccSPaul Dagnelie 22080fa1b3ccSPaul Dagnelie switch (rrd->header.drr_type) { 22090fa1b3ccSPaul Dagnelie case DRR_OBJECT: 22100fa1b3ccSPaul Dagnelie { 22110fa1b3ccSPaul Dagnelie struct drr_object *drro = &rrd->header.drr_u.drr_object; 22120fa1b3ccSPaul Dagnelie err = receive_object(rwa, drro, rrd->payload); 22130fa1b3ccSPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 22140fa1b3ccSPaul Dagnelie rrd->payload = NULL; 22150fa1b3ccSPaul Dagnelie return (err); 22160fa1b3ccSPaul Dagnelie } 22170fa1b3ccSPaul Dagnelie case DRR_FREEOBJECTS: 22180fa1b3ccSPaul Dagnelie { 22190fa1b3ccSPaul Dagnelie struct drr_freeobjects *drrfo = 22200fa1b3ccSPaul Dagnelie &rrd->header.drr_u.drr_freeobjects; 22210fa1b3ccSPaul Dagnelie return (receive_freeobjects(rwa, drrfo)); 22220fa1b3ccSPaul Dagnelie } 22230fa1b3ccSPaul Dagnelie case DRR_WRITE: 22240fa1b3ccSPaul Dagnelie { 22250fa1b3ccSPaul Dagnelie struct drr_write *drrw = &rrd->header.drr_u.drr_write; 2226eb633035STom Caputi err = receive_write(rwa, drrw, rrd->arc_buf); 22270fa1b3ccSPaul Dagnelie /* if receive_write() is successful, it consumes the arc_buf */ 22280fa1b3ccSPaul Dagnelie if (err != 0) 2229eb633035STom Caputi dmu_return_arcbuf(rrd->arc_buf); 2230eb633035STom Caputi rrd->arc_buf = NULL; 22310fa1b3ccSPaul Dagnelie rrd->payload = NULL; 22320fa1b3ccSPaul Dagnelie return (err); 22330fa1b3ccSPaul Dagnelie } 22340fa1b3ccSPaul Dagnelie case DRR_WRITE_BYREF: 22350fa1b3ccSPaul Dagnelie { 22360fa1b3ccSPaul Dagnelie struct drr_write_byref *drrwbr = 22370fa1b3ccSPaul Dagnelie &rrd->header.drr_u.drr_write_byref; 22380fa1b3ccSPaul Dagnelie return (receive_write_byref(rwa, drrwbr)); 22390fa1b3ccSPaul Dagnelie } 22400fa1b3ccSPaul Dagnelie case DRR_WRITE_EMBEDDED: 22410fa1b3ccSPaul Dagnelie { 22420fa1b3ccSPaul Dagnelie struct drr_write_embedded *drrwe = 22430fa1b3ccSPaul Dagnelie &rrd->header.drr_u.drr_write_embedded; 22440fa1b3ccSPaul Dagnelie err = receive_write_embedded(rwa, drrwe, rrd->payload); 22450fa1b3ccSPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 22460fa1b3ccSPaul Dagnelie rrd->payload = NULL; 22470fa1b3ccSPaul Dagnelie return (err); 22480fa1b3ccSPaul Dagnelie } 22490fa1b3ccSPaul Dagnelie case DRR_FREE: 22500fa1b3ccSPaul Dagnelie { 22510fa1b3ccSPaul Dagnelie struct drr_free *drrf = &rrd->header.drr_u.drr_free; 22520fa1b3ccSPaul Dagnelie return (receive_free(rwa, drrf)); 22530fa1b3ccSPaul Dagnelie } 22540fa1b3ccSPaul Dagnelie case DRR_SPILL: 22550fa1b3ccSPaul Dagnelie { 22560fa1b3ccSPaul Dagnelie struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; 2257eb633035STom Caputi err = receive_spill(rwa, drrs, rrd->arc_buf); 2258eb633035STom Caputi /* if receive_spill() is successful, it consumes the arc_buf */ 2259eb633035STom Caputi if (err != 0) 2260eb633035STom Caputi dmu_return_arcbuf(rrd->arc_buf); 2261eb633035STom Caputi rrd->arc_buf = NULL; 22620fa1b3ccSPaul Dagnelie rrd->payload = NULL; 22630fa1b3ccSPaul Dagnelie return (err); 22640fa1b3ccSPaul Dagnelie } 2265eb633035STom Caputi case DRR_OBJECT_RANGE: 2266eb633035STom Caputi { 2267eb633035STom Caputi struct drr_object_range *drror = 2268eb633035STom Caputi &rrd->header.drr_u.drr_object_range; 2269eb633035STom Caputi return (receive_object_range(rwa, drror)); 2270eb633035STom Caputi } 22710fa1b3ccSPaul Dagnelie default: 22720fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 22730fa1b3ccSPaul Dagnelie } 22740fa1b3ccSPaul Dagnelie } 22750fa1b3ccSPaul Dagnelie 22760fa1b3ccSPaul Dagnelie /* 22770fa1b3ccSPaul Dagnelie * dmu_recv_stream's worker thread; pull records off the queue, and then call 22780fa1b3ccSPaul Dagnelie * receive_process_record When we're done, signal the main thread and exit. 22790fa1b3ccSPaul Dagnelie */ 22800fa1b3ccSPaul Dagnelie static void 22810fa1b3ccSPaul Dagnelie receive_writer_thread(void *arg) 22820fa1b3ccSPaul Dagnelie { 22830fa1b3ccSPaul Dagnelie struct receive_writer_arg *rwa = arg; 22840fa1b3ccSPaul Dagnelie struct receive_record_arg *rrd; 22850fa1b3ccSPaul Dagnelie for (rrd = bqueue_dequeue(&rwa->q); !rrd->eos_marker; 22860fa1b3ccSPaul Dagnelie rrd = bqueue_dequeue(&rwa->q)) { 22870fa1b3ccSPaul Dagnelie /* 22880fa1b3ccSPaul Dagnelie * If there's an error, the main thread will stop putting things 22890fa1b3ccSPaul Dagnelie * on the queue, but we need to clear everything in it before we 22900fa1b3ccSPaul Dagnelie * can exit. 22910fa1b3ccSPaul Dagnelie */ 22920fa1b3ccSPaul Dagnelie if (rwa->err == 0) { 22930fa1b3ccSPaul Dagnelie rwa->err = receive_process_record(rwa, rrd); 2294eb633035STom Caputi } else if (rrd->arc_buf != NULL) { 2295eb633035STom Caputi dmu_return_arcbuf(rrd->arc_buf); 2296eb633035STom Caputi rrd->arc_buf = NULL; 22970fa1b3ccSPaul Dagnelie rrd->payload = NULL; 22980fa1b3ccSPaul Dagnelie } else if (rrd->payload != NULL) { 22990fa1b3ccSPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 23000fa1b3ccSPaul Dagnelie rrd->payload = NULL; 23010fa1b3ccSPaul Dagnelie } 23020fa1b3ccSPaul Dagnelie kmem_free(rrd, sizeof (*rrd)); 23030fa1b3ccSPaul Dagnelie } 23040fa1b3ccSPaul Dagnelie kmem_free(rrd, sizeof (*rrd)); 23050fa1b3ccSPaul Dagnelie mutex_enter(&rwa->mutex); 23060fa1b3ccSPaul Dagnelie rwa->done = B_TRUE; 23070fa1b3ccSPaul Dagnelie cv_signal(&rwa->cv); 23080fa1b3ccSPaul Dagnelie mutex_exit(&rwa->mutex); 23090fa1b3ccSPaul Dagnelie thread_exit(); 23100fa1b3ccSPaul Dagnelie } 23110fa1b3ccSPaul Dagnelie 23120fa1b3ccSPaul Dagnelie static int 23130fa1b3ccSPaul Dagnelie resume_check(struct receive_arg *ra, nvlist_t *begin_nvl) 23140fa1b3ccSPaul Dagnelie { 23150fa1b3ccSPaul Dagnelie uint64_t val; 23160fa1b3ccSPaul Dagnelie objset_t *mos = dmu_objset_pool(ra->os)->dp_meta_objset; 23170fa1b3ccSPaul Dagnelie uint64_t dsobj = dmu_objset_id(ra->os); 23180fa1b3ccSPaul Dagnelie uint64_t resume_obj, resume_off; 23190fa1b3ccSPaul Dagnelie 23200fa1b3ccSPaul Dagnelie if (nvlist_lookup_uint64(begin_nvl, 23210fa1b3ccSPaul Dagnelie "resume_object", &resume_obj) != 0 || 23220fa1b3ccSPaul Dagnelie nvlist_lookup_uint64(begin_nvl, 23230fa1b3ccSPaul Dagnelie "resume_offset", &resume_off) != 0) { 23240fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 23250fa1b3ccSPaul Dagnelie } 23260fa1b3ccSPaul Dagnelie VERIFY0(zap_lookup(mos, dsobj, 23270fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val)); 23280fa1b3ccSPaul Dagnelie if (resume_obj != val) 23290fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 23300fa1b3ccSPaul Dagnelie VERIFY0(zap_lookup(mos, dsobj, 23310fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val)); 23320fa1b3ccSPaul Dagnelie if (resume_off != val) 23330fa1b3ccSPaul Dagnelie return (SET_ERROR(EINVAL)); 23340fa1b3ccSPaul Dagnelie 23350fa1b3ccSPaul Dagnelie return (0); 23360fa1b3ccSPaul Dagnelie } 23370fa1b3ccSPaul Dagnelie 23380fa1b3ccSPaul Dagnelie /* 23390fa1b3ccSPaul Dagnelie * Read in the stream's records, one by one, and apply them to the pool. There 23400fa1b3ccSPaul Dagnelie * are two threads involved; the thread that calls this function will spin up a 23410fa1b3ccSPaul Dagnelie * worker thread, read the records off the stream one by one, and issue 23420fa1b3ccSPaul Dagnelie * prefetches for any necessary indirect blocks. It will then push the records 23430fa1b3ccSPaul Dagnelie * onto an internal blocking queue. The worker thread will pull the records off 23440fa1b3ccSPaul Dagnelie * the queue, and actually write the data into the DMU. This way, the worker 23450fa1b3ccSPaul Dagnelie * thread doesn't have to wait for reads to complete, since everything it needs 23460fa1b3ccSPaul Dagnelie * (the indirect blocks) will be prefetched. 23470fa1b3ccSPaul Dagnelie * 23480fa1b3ccSPaul Dagnelie * NB: callers *must* call dmu_recv_end() if this succeeds. 23490fa1b3ccSPaul Dagnelie */ 23500fa1b3ccSPaul Dagnelie int 23510fa1b3ccSPaul Dagnelie dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, 23520fa1b3ccSPaul Dagnelie int cleanup_fd, uint64_t *action_handlep) 23530fa1b3ccSPaul Dagnelie { 23540fa1b3ccSPaul Dagnelie int err = 0; 23550fa1b3ccSPaul Dagnelie struct receive_arg ra = { 0 }; 23560fa1b3ccSPaul Dagnelie struct receive_writer_arg rwa = { 0 }; 23570fa1b3ccSPaul Dagnelie int featureflags; 23580fa1b3ccSPaul Dagnelie nvlist_t *begin_nvl = NULL; 23590fa1b3ccSPaul Dagnelie 23600fa1b3ccSPaul Dagnelie ra.byteswap = drc->drc_byteswap; 2361eb633035STom Caputi ra.raw = drc->drc_raw; 23620fa1b3ccSPaul Dagnelie ra.cksum = drc->drc_cksum; 23630fa1b3ccSPaul Dagnelie ra.vp = vp; 23640fa1b3ccSPaul Dagnelie ra.voff = *voffp; 23650fa1b3ccSPaul Dagnelie 23660fa1b3ccSPaul Dagnelie if (dsl_dataset_is_zapified(drc->drc_ds)) { 23670fa1b3ccSPaul Dagnelie (void) zap_lookup(drc->drc_ds->ds_dir->dd_pool->dp_meta_objset, 23680fa1b3ccSPaul Dagnelie drc->drc_ds->ds_object, DS_FIELD_RESUME_BYTES, 23690fa1b3ccSPaul Dagnelie sizeof (ra.bytes_read), 1, &ra.bytes_read); 23700fa1b3ccSPaul Dagnelie } 23710fa1b3ccSPaul Dagnelie 23720fa1b3ccSPaul Dagnelie objlist_create(&ra.ignore_objlist); 23730fa1b3ccSPaul Dagnelie 23740fa1b3ccSPaul Dagnelie /* these were verified in dmu_recv_begin */ 23750fa1b3ccSPaul Dagnelie ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, 23760fa1b3ccSPaul Dagnelie DMU_SUBSTREAM); 23770fa1b3ccSPaul Dagnelie ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); 23780fa1b3ccSPaul Dagnelie 23790fa1b3ccSPaul Dagnelie /* 23800fa1b3ccSPaul Dagnelie * Open the objset we are modifying. 23810fa1b3ccSPaul Dagnelie */ 23820fa1b3ccSPaul Dagnelie VERIFY0(dmu_objset_from_ds(drc->drc_ds, &ra.os)); 23830fa1b3ccSPaul Dagnelie 23840fa1b3ccSPaul Dagnelie ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); 23850fa1b3ccSPaul Dagnelie 23860fa1b3ccSPaul Dagnelie featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); 2387eb633035STom Caputi ra.featureflags = featureflags; 2388eb633035STom Caputi 2389eb633035STom Caputi ASSERT0(ra.os->os_encrypted && 2390eb633035STom Caputi (featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)); 23910fa1b3ccSPaul Dagnelie 23920fa1b3ccSPaul Dagnelie /* if this stream is dedup'ed, set up the avl tree for guid mapping */ 23930fa1b3ccSPaul Dagnelie if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { 23940fa1b3ccSPaul Dagnelie minor_t minor; 23950fa1b3ccSPaul Dagnelie 23960fa1b3ccSPaul Dagnelie if (cleanup_fd == -1) { 2397eb633035STom Caputi err = SET_ERROR(EBADF); 23980fa1b3ccSPaul Dagnelie goto out; 23990fa1b3ccSPaul Dagnelie } 2400eb633035STom Caputi err = zfs_onexit_fd_hold(cleanup_fd, &minor); 2401eb633035STom Caputi if (err != 0) { 24020fa1b3ccSPaul Dagnelie cleanup_fd = -1; 24030fa1b3ccSPaul Dagnelie goto out; 24040fa1b3ccSPaul Dagnelie } 24050fa1b3ccSPaul Dagnelie 24060fa1b3ccSPaul Dagnelie if (*action_handlep == 0) { 24070fa1b3ccSPaul Dagnelie rwa.guid_to_ds_map = 24080fa1b3ccSPaul Dagnelie kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); 24090fa1b3ccSPaul Dagnelie avl_create(rwa.guid_to_ds_map, guid_compare, 24100fa1b3ccSPaul Dagnelie sizeof (guid_map_entry_t), 24110fa1b3ccSPaul Dagnelie offsetof(guid_map_entry_t, avlnode)); 24120fa1b3ccSPaul Dagnelie err = zfs_onexit_add_cb(minor, 24130fa1b3ccSPaul Dagnelie free_guid_map_onexit, rwa.guid_to_ds_map, 24140fa1b3ccSPaul Dagnelie action_handlep); 2415eb633035STom Caputi if (err != 0) 24160fa1b3ccSPaul Dagnelie goto out; 24170fa1b3ccSPaul Dagnelie } else { 24180fa1b3ccSPaul Dagnelie err = zfs_onexit_cb_data(minor, *action_handlep, 24190fa1b3ccSPaul Dagnelie (void **)&rwa.guid_to_ds_map); 2420eb633035STom Caputi if (err != 0) 24210fa1b3ccSPaul Dagnelie goto out; 24220fa1b3ccSPaul Dagnelie } 24230fa1b3ccSPaul Dagnelie 24240fa1b3ccSPaul Dagnelie drc->drc_guid_to_ds_map = rwa.guid_to_ds_map; 24250fa1b3ccSPaul Dagnelie } 24260fa1b3ccSPaul Dagnelie 24270fa1b3ccSPaul Dagnelie uint32_t payloadlen = drc->drc_drr_begin->drr_payloadlen; 24280fa1b3ccSPaul Dagnelie void *payload = NULL; 24290fa1b3ccSPaul Dagnelie if (payloadlen != 0) 24300fa1b3ccSPaul Dagnelie payload = kmem_alloc(payloadlen, KM_SLEEP); 24310fa1b3ccSPaul Dagnelie 24320fa1b3ccSPaul Dagnelie err = receive_read_payload_and_next_header(&ra, payloadlen, payload); 24330fa1b3ccSPaul Dagnelie if (err != 0) { 24340fa1b3ccSPaul Dagnelie if (payloadlen != 0) 24350fa1b3ccSPaul Dagnelie kmem_free(payload, payloadlen); 24360fa1b3ccSPaul Dagnelie goto out; 24370fa1b3ccSPaul Dagnelie } 24380fa1b3ccSPaul Dagnelie if (payloadlen != 0) { 24390fa1b3ccSPaul Dagnelie err = nvlist_unpack(payload, payloadlen, &begin_nvl, KM_SLEEP); 24400fa1b3ccSPaul Dagnelie kmem_free(payload, payloadlen); 24410fa1b3ccSPaul Dagnelie if (err != 0) 24420fa1b3ccSPaul Dagnelie goto out; 24430fa1b3ccSPaul Dagnelie } 24440fa1b3ccSPaul Dagnelie 2445eb633035STom Caputi /* handle DSL encryption key payload */ 2446eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_RAW) { 2447eb633035STom Caputi nvlist_t *keynvl = NULL; 2448eb633035STom Caputi 2449eb633035STom Caputi ASSERT(ra.os->os_encrypted); 2450eb633035STom Caputi ASSERT(drc->drc_raw); 2451eb633035STom Caputi 2452eb633035STom Caputi err = nvlist_lookup_nvlist(begin_nvl, "crypt_keydata", &keynvl); 2453eb633035STom Caputi if (err != 0) 2454eb633035STom Caputi goto out; 2455eb633035STom Caputi 2456eb633035STom Caputi /* 2457eb633035STom Caputi * If this is a new dataset we set the key immediately. 2458eb633035STom Caputi * Otherwise we don't want to change the key until we 2459eb633035STom Caputi * are sure the rest of the receive succeeded so we stash 2460eb633035STom Caputi * the keynvl away until then. 2461eb633035STom Caputi */ 2462eb633035STom Caputi err = dsl_crypto_recv_raw(spa_name(ra.os->os_spa), 2463eb633035STom Caputi drc->drc_ds->ds_object, drc->drc_fromsnapobj, 2464eb633035STom Caputi drc->drc_drrb->drr_type, keynvl, drc->drc_newfs); 2465eb633035STom Caputi if (err != 0) 2466eb633035STom Caputi goto out; 2467eb633035STom Caputi 2468eb633035STom Caputi /* see comment in dmu_recv_end_sync() */ 2469eb633035STom Caputi drc->drc_ivset_guid = 0; 2470eb633035STom Caputi (void) nvlist_lookup_uint64(keynvl, "to_ivset_guid", 2471eb633035STom Caputi &drc->drc_ivset_guid); 2472eb633035STom Caputi 2473eb633035STom Caputi if (!drc->drc_newfs) 2474eb633035STom Caputi drc->drc_keynvl = fnvlist_dup(keynvl); 2475eb633035STom Caputi } 2476eb633035STom Caputi 24770fa1b3ccSPaul Dagnelie if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { 24780fa1b3ccSPaul Dagnelie err = resume_check(&ra, begin_nvl); 24790fa1b3ccSPaul Dagnelie if (err != 0) 24800fa1b3ccSPaul Dagnelie goto out; 24810fa1b3ccSPaul Dagnelie } 24820fa1b3ccSPaul Dagnelie 2483*6ccda740Sloli (void) bqueue_init(&rwa.q, 2484*6ccda740Sloli MAX(zfs_recv_queue_length, 2 * zfs_max_recordsize), 24850fa1b3ccSPaul Dagnelie offsetof(struct receive_record_arg, node)); 24860fa1b3ccSPaul Dagnelie cv_init(&rwa.cv, NULL, CV_DEFAULT, NULL); 24870fa1b3ccSPaul Dagnelie mutex_init(&rwa.mutex, NULL, MUTEX_DEFAULT, NULL); 24880fa1b3ccSPaul Dagnelie rwa.os = ra.os; 24890fa1b3ccSPaul Dagnelie rwa.byteswap = drc->drc_byteswap; 24900fa1b3ccSPaul Dagnelie rwa.resumable = drc->drc_resumable; 2491eb633035STom Caputi rwa.raw = drc->drc_raw; 2492eb633035STom Caputi rwa.spill = drc->drc_spill; 2493eb633035STom Caputi rwa.os->os_raw_receive = drc->drc_raw; 24940fa1b3ccSPaul Dagnelie 24950fa1b3ccSPaul Dagnelie (void) thread_create(NULL, 0, receive_writer_thread, &rwa, 0, curproc, 24960fa1b3ccSPaul Dagnelie TS_RUN, minclsyspri); 24970fa1b3ccSPaul Dagnelie /* 24980fa1b3ccSPaul Dagnelie * We're reading rwa.err without locks, which is safe since we are the 24990fa1b3ccSPaul Dagnelie * only reader, and the worker thread is the only writer. It's ok if we 25000fa1b3ccSPaul Dagnelie * miss a write for an iteration or two of the loop, since the writer 25010fa1b3ccSPaul Dagnelie * thread will keep freeing records we send it until we send it an eos 25020fa1b3ccSPaul Dagnelie * marker. 25030fa1b3ccSPaul Dagnelie * 25040fa1b3ccSPaul Dagnelie * We can leave this loop in 3 ways: First, if rwa.err is 25050fa1b3ccSPaul Dagnelie * non-zero. In that case, the writer thread will free the rrd we just 25060fa1b3ccSPaul Dagnelie * pushed. Second, if we're interrupted; in that case, either it's the 25070fa1b3ccSPaul Dagnelie * first loop and ra.rrd was never allocated, or it's later, and ra.rrd 25080fa1b3ccSPaul Dagnelie * has been handed off to the writer thread who will free it. Finally, 25090fa1b3ccSPaul Dagnelie * if receive_read_record fails or we're at the end of the stream, then 25100fa1b3ccSPaul Dagnelie * we free ra.rrd and exit. 25110fa1b3ccSPaul Dagnelie */ 25120fa1b3ccSPaul Dagnelie while (rwa.err == 0) { 25130fa1b3ccSPaul Dagnelie if (issig(JUSTLOOKING) && issig(FORREAL)) { 25140fa1b3ccSPaul Dagnelie err = SET_ERROR(EINTR); 25150fa1b3ccSPaul Dagnelie break; 25160fa1b3ccSPaul Dagnelie } 25170fa1b3ccSPaul Dagnelie 25180fa1b3ccSPaul Dagnelie ASSERT3P(ra.rrd, ==, NULL); 25190fa1b3ccSPaul Dagnelie ra.rrd = ra.next_rrd; 25200fa1b3ccSPaul Dagnelie ra.next_rrd = NULL; 25210fa1b3ccSPaul Dagnelie /* Allocates and loads header into ra.next_rrd */ 25220fa1b3ccSPaul Dagnelie err = receive_read_record(&ra); 25230fa1b3ccSPaul Dagnelie 25240fa1b3ccSPaul Dagnelie if (ra.rrd->header.drr_type == DRR_END || err != 0) { 25250fa1b3ccSPaul Dagnelie kmem_free(ra.rrd, sizeof (*ra.rrd)); 25260fa1b3ccSPaul Dagnelie ra.rrd = NULL; 25270fa1b3ccSPaul Dagnelie break; 25280fa1b3ccSPaul Dagnelie } 25290fa1b3ccSPaul Dagnelie 25300fa1b3ccSPaul Dagnelie bqueue_enqueue(&rwa.q, ra.rrd, 25310fa1b3ccSPaul Dagnelie sizeof (struct receive_record_arg) + ra.rrd->payload_size); 25320fa1b3ccSPaul Dagnelie ra.rrd = NULL; 25330fa1b3ccSPaul Dagnelie } 2534eb633035STom Caputi ASSERT3P(ra.rrd, ==, NULL); 2535eb633035STom Caputi ra.rrd = kmem_zalloc(sizeof (*ra.rrd), KM_SLEEP); 2536eb633035STom Caputi ra.rrd->eos_marker = B_TRUE; 2537eb633035STom Caputi bqueue_enqueue(&rwa.q, ra.rrd, 1); 25380fa1b3ccSPaul Dagnelie 25390fa1b3ccSPaul Dagnelie mutex_enter(&rwa.mutex); 25400fa1b3ccSPaul Dagnelie while (!rwa.done) { 25410fa1b3ccSPaul Dagnelie cv_wait(&rwa.cv, &rwa.mutex); 25420fa1b3ccSPaul Dagnelie } 25430fa1b3ccSPaul Dagnelie mutex_exit(&rwa.mutex); 25440fa1b3ccSPaul Dagnelie 25450fa1b3ccSPaul Dagnelie /* 25460fa1b3ccSPaul Dagnelie * If we are receiving a full stream as a clone, all object IDs which 25470fa1b3ccSPaul Dagnelie * are greater than the maximum ID referenced in the stream are 25480fa1b3ccSPaul Dagnelie * by definition unused and must be freed. Note that it's possible that 25490fa1b3ccSPaul Dagnelie * we've resumed this send and the first record we received was the END 25500fa1b3ccSPaul Dagnelie * record. In that case, max_object would be 0, but we shouldn't start 25510fa1b3ccSPaul Dagnelie * freeing all objects from there; instead we should start from the 25520fa1b3ccSPaul Dagnelie * resumeobj. 25530fa1b3ccSPaul Dagnelie */ 25540fa1b3ccSPaul Dagnelie if (drc->drc_clone && drc->drc_drrb->drr_fromguid == 0) { 25550fa1b3ccSPaul Dagnelie uint64_t obj; 25560fa1b3ccSPaul Dagnelie if (nvlist_lookup_uint64(begin_nvl, "resume_object", &obj) != 0) 25570fa1b3ccSPaul Dagnelie obj = 0; 25580fa1b3ccSPaul Dagnelie if (rwa.max_object > obj) 25590fa1b3ccSPaul Dagnelie obj = rwa.max_object; 25600fa1b3ccSPaul Dagnelie obj++; 25610fa1b3ccSPaul Dagnelie int free_err = 0; 25620fa1b3ccSPaul Dagnelie int next_err = 0; 25630fa1b3ccSPaul Dagnelie 25640fa1b3ccSPaul Dagnelie while (next_err == 0) { 25650fa1b3ccSPaul Dagnelie free_err = dmu_free_long_object(rwa.os, obj); 25660fa1b3ccSPaul Dagnelie if (free_err != 0 && free_err != ENOENT) 25670fa1b3ccSPaul Dagnelie break; 25680fa1b3ccSPaul Dagnelie 25690fa1b3ccSPaul Dagnelie next_err = dmu_object_next(rwa.os, &obj, FALSE, 0); 25700fa1b3ccSPaul Dagnelie } 25710fa1b3ccSPaul Dagnelie 25720fa1b3ccSPaul Dagnelie if (err == 0) { 25730fa1b3ccSPaul Dagnelie if (free_err != 0 && free_err != ENOENT) 25740fa1b3ccSPaul Dagnelie err = free_err; 25750fa1b3ccSPaul Dagnelie else if (next_err != ESRCH) 25760fa1b3ccSPaul Dagnelie err = next_err; 25770fa1b3ccSPaul Dagnelie } 25780fa1b3ccSPaul Dagnelie } 25790fa1b3ccSPaul Dagnelie 25800fa1b3ccSPaul Dagnelie cv_destroy(&rwa.cv); 25810fa1b3ccSPaul Dagnelie mutex_destroy(&rwa.mutex); 25820fa1b3ccSPaul Dagnelie bqueue_destroy(&rwa.q); 25830fa1b3ccSPaul Dagnelie if (err == 0) 25840fa1b3ccSPaul Dagnelie err = rwa.err; 25850fa1b3ccSPaul Dagnelie 25860fa1b3ccSPaul Dagnelie out: 2587eb633035STom Caputi /* 2588eb633035STom Caputi * If we hit an error before we started the receive_writer_thread 2589eb633035STom Caputi * we need to clean up the next_rrd we create by processing the 2590eb633035STom Caputi * DRR_BEGIN record. 2591eb633035STom Caputi */ 2592eb633035STom Caputi if (ra.next_rrd != NULL) 2593eb633035STom Caputi kmem_free(ra.next_rrd, sizeof (*ra.next_rrd)); 2594eb633035STom Caputi 25950fa1b3ccSPaul Dagnelie nvlist_free(begin_nvl); 25960fa1b3ccSPaul Dagnelie if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) 25970fa1b3ccSPaul Dagnelie zfs_onexit_fd_rele(cleanup_fd); 25980fa1b3ccSPaul Dagnelie 25990fa1b3ccSPaul Dagnelie if (err != 0) { 26000fa1b3ccSPaul Dagnelie /* 26010fa1b3ccSPaul Dagnelie * Clean up references. If receive is not resumable, 26020fa1b3ccSPaul Dagnelie * destroy what we created, so we don't leave it in 26030fa1b3ccSPaul Dagnelie * the inconsistent state. 26040fa1b3ccSPaul Dagnelie */ 26050fa1b3ccSPaul Dagnelie dmu_recv_cleanup_ds(drc); 2606eb633035STom Caputi nvlist_free(drc->drc_keynvl); 26070fa1b3ccSPaul Dagnelie } 26080fa1b3ccSPaul Dagnelie 26090fa1b3ccSPaul Dagnelie *voffp = ra.voff; 26100fa1b3ccSPaul Dagnelie objlist_destroy(&ra.ignore_objlist); 26110fa1b3ccSPaul Dagnelie return (err); 26120fa1b3ccSPaul Dagnelie } 26130fa1b3ccSPaul Dagnelie 26140fa1b3ccSPaul Dagnelie static int 26150fa1b3ccSPaul Dagnelie dmu_recv_end_check(void *arg, dmu_tx_t *tx) 26160fa1b3ccSPaul Dagnelie { 26170fa1b3ccSPaul Dagnelie dmu_recv_cookie_t *drc = arg; 26180fa1b3ccSPaul Dagnelie dsl_pool_t *dp = dmu_tx_pool(tx); 26190fa1b3ccSPaul Dagnelie int error; 26200fa1b3ccSPaul Dagnelie 26210fa1b3ccSPaul Dagnelie ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); 26220fa1b3ccSPaul Dagnelie 26230fa1b3ccSPaul Dagnelie if (!drc->drc_newfs) { 26240fa1b3ccSPaul Dagnelie dsl_dataset_t *origin_head; 26250fa1b3ccSPaul Dagnelie 26260fa1b3ccSPaul Dagnelie error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); 26270fa1b3ccSPaul Dagnelie if (error != 0) 26280fa1b3ccSPaul Dagnelie return (error); 26290fa1b3ccSPaul Dagnelie if (drc->drc_force) { 26300fa1b3ccSPaul Dagnelie /* 26310fa1b3ccSPaul Dagnelie * We will destroy any snapshots in tofs (i.e. before 26320fa1b3ccSPaul Dagnelie * origin_head) that are after the origin (which is 26330fa1b3ccSPaul Dagnelie * the snap before drc_ds, because drc_ds can not 26340fa1b3ccSPaul Dagnelie * have any snaps of its own). 26350fa1b3ccSPaul Dagnelie */ 26360fa1b3ccSPaul Dagnelie uint64_t obj; 26370fa1b3ccSPaul Dagnelie 26380fa1b3ccSPaul Dagnelie obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 26390fa1b3ccSPaul Dagnelie while (obj != 26400fa1b3ccSPaul Dagnelie dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 26410fa1b3ccSPaul Dagnelie dsl_dataset_t *snap; 26420fa1b3ccSPaul Dagnelie error = dsl_dataset_hold_obj(dp, obj, FTAG, 26430fa1b3ccSPaul Dagnelie &snap); 26440fa1b3ccSPaul Dagnelie if (error != 0) 26450fa1b3ccSPaul Dagnelie break; 26460fa1b3ccSPaul Dagnelie if (snap->ds_dir != origin_head->ds_dir) 26470fa1b3ccSPaul Dagnelie error = SET_ERROR(EINVAL); 26480fa1b3ccSPaul Dagnelie if (error == 0) { 26490fa1b3ccSPaul Dagnelie error = dsl_destroy_snapshot_check_impl( 26500fa1b3ccSPaul Dagnelie snap, B_FALSE); 26510fa1b3ccSPaul Dagnelie } 26520fa1b3ccSPaul Dagnelie obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 26530fa1b3ccSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 26540fa1b3ccSPaul Dagnelie if (error != 0) 26550fa1b3ccSPaul Dagnelie break; 26560fa1b3ccSPaul Dagnelie } 26570fa1b3ccSPaul Dagnelie if (error != 0) { 26580fa1b3ccSPaul Dagnelie dsl_dataset_rele(origin_head, FTAG); 26590fa1b3ccSPaul Dagnelie return (error); 26600fa1b3ccSPaul Dagnelie } 26610fa1b3ccSPaul Dagnelie } 2662eb633035STom Caputi if (drc->drc_keynvl != NULL) { 2663eb633035STom Caputi error = dsl_crypto_recv_raw_key_check(drc->drc_ds, 2664eb633035STom Caputi drc->drc_keynvl, tx); 2665eb633035STom Caputi if (error != 0) { 2666eb633035STom Caputi dsl_dataset_rele(origin_head, FTAG); 2667eb633035STom Caputi return (error); 2668eb633035STom Caputi } 2669eb633035STom Caputi } 2670eb633035STom Caputi 26710fa1b3ccSPaul Dagnelie error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, 26720fa1b3ccSPaul Dagnelie origin_head, drc->drc_force, drc->drc_owner, tx); 26730fa1b3ccSPaul Dagnelie if (error != 0) { 26740fa1b3ccSPaul Dagnelie dsl_dataset_rele(origin_head, FTAG); 26750fa1b3ccSPaul Dagnelie return (error); 26760fa1b3ccSPaul Dagnelie } 26770fa1b3ccSPaul Dagnelie error = dsl_dataset_snapshot_check_impl(origin_head, 26780fa1b3ccSPaul Dagnelie drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 26790fa1b3ccSPaul Dagnelie dsl_dataset_rele(origin_head, FTAG); 26800fa1b3ccSPaul Dagnelie if (error != 0) 26810fa1b3ccSPaul Dagnelie return (error); 26820fa1b3ccSPaul Dagnelie 26830fa1b3ccSPaul Dagnelie error = dsl_destroy_head_check_impl(drc->drc_ds, 1); 26840fa1b3ccSPaul Dagnelie } else { 26850fa1b3ccSPaul Dagnelie error = dsl_dataset_snapshot_check_impl(drc->drc_ds, 26860fa1b3ccSPaul Dagnelie drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 26870fa1b3ccSPaul Dagnelie } 26880fa1b3ccSPaul Dagnelie return (error); 26890fa1b3ccSPaul Dagnelie } 26900fa1b3ccSPaul Dagnelie 26910fa1b3ccSPaul Dagnelie static void 26920fa1b3ccSPaul Dagnelie dmu_recv_end_sync(void *arg, dmu_tx_t *tx) 26930fa1b3ccSPaul Dagnelie { 26940fa1b3ccSPaul Dagnelie dmu_recv_cookie_t *drc = arg; 26950fa1b3ccSPaul Dagnelie dsl_pool_t *dp = dmu_tx_pool(tx); 2696eb633035STom Caputi boolean_t encrypted = drc->drc_ds->ds_dir->dd_crypto_obj != 0; 26970fa1b3ccSPaul Dagnelie 26980fa1b3ccSPaul Dagnelie spa_history_log_internal_ds(drc->drc_ds, "finish receiving", 26990fa1b3ccSPaul Dagnelie tx, "snap=%s", drc->drc_tosnap); 2700eb633035STom Caputi drc->drc_ds->ds_objset->os_raw_receive = B_FALSE; 27010fa1b3ccSPaul Dagnelie 27020fa1b3ccSPaul Dagnelie if (!drc->drc_newfs) { 27030fa1b3ccSPaul Dagnelie dsl_dataset_t *origin_head; 27040fa1b3ccSPaul Dagnelie 27050fa1b3ccSPaul Dagnelie VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, 27060fa1b3ccSPaul Dagnelie &origin_head)); 27070fa1b3ccSPaul Dagnelie 27080fa1b3ccSPaul Dagnelie if (drc->drc_force) { 27090fa1b3ccSPaul Dagnelie /* 27100fa1b3ccSPaul Dagnelie * Destroy any snapshots of drc_tofs (origin_head) 27110fa1b3ccSPaul Dagnelie * after the origin (the snap before drc_ds). 27120fa1b3ccSPaul Dagnelie */ 27130fa1b3ccSPaul Dagnelie uint64_t obj; 27140fa1b3ccSPaul Dagnelie 27150fa1b3ccSPaul Dagnelie obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 27160fa1b3ccSPaul Dagnelie while (obj != 27170fa1b3ccSPaul Dagnelie dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 27180fa1b3ccSPaul Dagnelie dsl_dataset_t *snap; 27190fa1b3ccSPaul Dagnelie VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, 27200fa1b3ccSPaul Dagnelie &snap)); 27210fa1b3ccSPaul Dagnelie ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); 27220fa1b3ccSPaul Dagnelie obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 27230fa1b3ccSPaul Dagnelie dsl_destroy_snapshot_sync_impl(snap, 27240fa1b3ccSPaul Dagnelie B_FALSE, tx); 27250fa1b3ccSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 27260fa1b3ccSPaul Dagnelie } 27270fa1b3ccSPaul Dagnelie } 2728eb633035STom Caputi if (drc->drc_keynvl != NULL) { 2729eb633035STom Caputi dsl_crypto_recv_raw_key_sync(drc->drc_ds, 2730eb633035STom Caputi drc->drc_keynvl, tx); 2731eb633035STom Caputi nvlist_free(drc->drc_keynvl); 2732eb633035STom Caputi drc->drc_keynvl = NULL; 2733eb633035STom Caputi } 2734eb633035STom Caputi 2735eb633035STom Caputi VERIFY3P(drc->drc_ds->ds_prev, ==, origin_head->ds_prev); 27360fa1b3ccSPaul Dagnelie 27370fa1b3ccSPaul Dagnelie dsl_dataset_clone_swap_sync_impl(drc->drc_ds, 27380fa1b3ccSPaul Dagnelie origin_head, tx); 27390fa1b3ccSPaul Dagnelie dsl_dataset_snapshot_sync_impl(origin_head, 27400fa1b3ccSPaul Dagnelie drc->drc_tosnap, tx); 27410fa1b3ccSPaul Dagnelie 27420fa1b3ccSPaul Dagnelie /* set snapshot's creation time and guid */ 27430fa1b3ccSPaul Dagnelie dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); 27440fa1b3ccSPaul Dagnelie dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = 27450fa1b3ccSPaul Dagnelie drc->drc_drrb->drr_creation_time; 27460fa1b3ccSPaul Dagnelie dsl_dataset_phys(origin_head->ds_prev)->ds_guid = 27470fa1b3ccSPaul Dagnelie drc->drc_drrb->drr_toguid; 27480fa1b3ccSPaul Dagnelie dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= 27490fa1b3ccSPaul Dagnelie ~DS_FLAG_INCONSISTENT; 27500fa1b3ccSPaul Dagnelie 27510fa1b3ccSPaul Dagnelie dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 27520fa1b3ccSPaul Dagnelie dsl_dataset_phys(origin_head)->ds_flags &= 27530fa1b3ccSPaul Dagnelie ~DS_FLAG_INCONSISTENT; 27540fa1b3ccSPaul Dagnelie 27550fa1b3ccSPaul Dagnelie drc->drc_newsnapobj = 27560fa1b3ccSPaul Dagnelie dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 27570fa1b3ccSPaul Dagnelie 27580fa1b3ccSPaul Dagnelie dsl_dataset_rele(origin_head, FTAG); 27590fa1b3ccSPaul Dagnelie dsl_destroy_head_sync_impl(drc->drc_ds, tx); 27600fa1b3ccSPaul Dagnelie 27610fa1b3ccSPaul Dagnelie if (drc->drc_owner != NULL) 27620fa1b3ccSPaul Dagnelie VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); 27630fa1b3ccSPaul Dagnelie } else { 27640fa1b3ccSPaul Dagnelie dsl_dataset_t *ds = drc->drc_ds; 27650fa1b3ccSPaul Dagnelie 27660fa1b3ccSPaul Dagnelie dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); 27670fa1b3ccSPaul Dagnelie 27680fa1b3ccSPaul Dagnelie /* set snapshot's creation time and guid */ 27690fa1b3ccSPaul Dagnelie dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 27700fa1b3ccSPaul Dagnelie dsl_dataset_phys(ds->ds_prev)->ds_creation_time = 27710fa1b3ccSPaul Dagnelie drc->drc_drrb->drr_creation_time; 27720fa1b3ccSPaul Dagnelie dsl_dataset_phys(ds->ds_prev)->ds_guid = 27730fa1b3ccSPaul Dagnelie drc->drc_drrb->drr_toguid; 27740fa1b3ccSPaul Dagnelie dsl_dataset_phys(ds->ds_prev)->ds_flags &= 27750fa1b3ccSPaul Dagnelie ~DS_FLAG_INCONSISTENT; 27760fa1b3ccSPaul Dagnelie 27770fa1b3ccSPaul Dagnelie dmu_buf_will_dirty(ds->ds_dbuf, tx); 27780fa1b3ccSPaul Dagnelie dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; 27790fa1b3ccSPaul Dagnelie if (dsl_dataset_has_resume_receive_state(ds)) { 27800fa1b3ccSPaul Dagnelie (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 27810fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_FROMGUID, tx); 27820fa1b3ccSPaul Dagnelie (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 27830fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_OBJECT, tx); 27840fa1b3ccSPaul Dagnelie (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 27850fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_OFFSET, tx); 27860fa1b3ccSPaul Dagnelie (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 27870fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_BYTES, tx); 27880fa1b3ccSPaul Dagnelie (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 27890fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_TOGUID, tx); 27900fa1b3ccSPaul Dagnelie (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 27910fa1b3ccSPaul Dagnelie DS_FIELD_RESUME_TONAME, tx); 27920fa1b3ccSPaul Dagnelie } 27930fa1b3ccSPaul Dagnelie drc->drc_newsnapobj = 27940fa1b3ccSPaul Dagnelie dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; 27950fa1b3ccSPaul Dagnelie } 2796eb633035STom Caputi 2797eb633035STom Caputi /* 2798eb633035STom Caputi * If this is a raw receive, the crypt_keydata nvlist will include 2799eb633035STom Caputi * a to_ivset_guid for us to set on the new snapshot. This value 2800eb633035STom Caputi * will override the value generated by the snapshot code. However, 2801eb633035STom Caputi * this value may not be present, because older implementations of 2802eb633035STom Caputi * the raw send code did not include this value, and we are still 2803eb633035STom Caputi * allowed to receive them if the zfs_disable_ivset_guid_check 2804eb633035STom Caputi * tunable is set, in which case we will leave the newly-generated 2805eb633035STom Caputi * value. 2806eb633035STom Caputi */ 2807eb633035STom Caputi if (drc->drc_raw && drc->drc_ivset_guid != 0) { 2808eb633035STom Caputi dmu_object_zapify(dp->dp_meta_objset, drc->drc_newsnapobj, 2809eb633035STom Caputi DMU_OT_DSL_DATASET, tx); 2810eb633035STom Caputi VERIFY0(zap_update(dp->dp_meta_objset, drc->drc_newsnapobj, 2811eb633035STom Caputi DS_FIELD_IVSET_GUID, sizeof (uint64_t), 1, 2812eb633035STom Caputi &drc->drc_ivset_guid, tx)); 2813eb633035STom Caputi } 2814eb633035STom Caputi 28150fa1b3ccSPaul Dagnelie /* 28160fa1b3ccSPaul Dagnelie * Release the hold from dmu_recv_begin. This must be done before 2817eb633035STom Caputi * we return to open context, so that when we free the dataset's dnode 2818eb633035STom Caputi * we can evict its bonus buffer. Since the dataset may be destroyed 2819eb633035STom Caputi * at this point (and therefore won't have a valid pointer to the spa) 2820eb633035STom Caputi * we release the key mapping manually here while we do have a valid 2821eb633035STom Caputi * pointer, if it exists. 28220fa1b3ccSPaul Dagnelie */ 2823eb633035STom Caputi if (!drc->drc_raw && encrypted) { 2824eb633035STom Caputi (void) spa_keystore_remove_mapping(dmu_tx_pool(tx)->dp_spa, 2825eb633035STom Caputi drc->drc_ds->ds_object, drc->drc_ds); 2826eb633035STom Caputi } 2827eb633035STom Caputi dsl_dataset_disown(drc->drc_ds, 0, dmu_recv_tag); 28280fa1b3ccSPaul Dagnelie drc->drc_ds = NULL; 28290fa1b3ccSPaul Dagnelie } 28300fa1b3ccSPaul Dagnelie 28310fa1b3ccSPaul Dagnelie static int 2832eb633035STom Caputi add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj, 2833eb633035STom Caputi boolean_t raw) 28340fa1b3ccSPaul Dagnelie { 28350fa1b3ccSPaul Dagnelie dsl_pool_t *dp; 28360fa1b3ccSPaul Dagnelie dsl_dataset_t *snapds; 28370fa1b3ccSPaul Dagnelie guid_map_entry_t *gmep; 2838eb633035STom Caputi objset_t *os; 2839eb633035STom Caputi ds_hold_flags_t dsflags = (raw) ? 0 : DS_HOLD_FLAG_DECRYPT; 28400fa1b3ccSPaul Dagnelie int err; 28410fa1b3ccSPaul Dagnelie 28420fa1b3ccSPaul Dagnelie ASSERT(guid_map != NULL); 28430fa1b3ccSPaul Dagnelie 28440fa1b3ccSPaul Dagnelie err = dsl_pool_hold(name, FTAG, &dp); 28450fa1b3ccSPaul Dagnelie if (err != 0) 28460fa1b3ccSPaul Dagnelie return (err); 28470fa1b3ccSPaul Dagnelie gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); 2848eb633035STom Caputi err = dsl_dataset_own_obj(dp, snapobj, dsflags, gmep, &snapds); 28490fa1b3ccSPaul Dagnelie if (err == 0) { 2850eb633035STom Caputi /* 2851eb633035STom Caputi * If this is a deduplicated raw send stream, we need 2852eb633035STom Caputi * to make sure that we can still read raw blocks from 2853eb633035STom Caputi * earlier datasets in the stream, so we set the 2854eb633035STom Caputi * os_raw_receive flag now. 2855eb633035STom Caputi */ 2856eb633035STom Caputi if (raw) { 2857eb633035STom Caputi err = dmu_objset_from_ds(snapds, &os); 2858eb633035STom Caputi if (err != 0) { 2859eb633035STom Caputi dsl_dataset_disown(snapds, dsflags, FTAG); 2860eb633035STom Caputi dsl_pool_rele(dp, FTAG); 2861eb633035STom Caputi kmem_free(gmep, sizeof (*gmep)); 2862eb633035STom Caputi return (err); 2863eb633035STom Caputi } 2864eb633035STom Caputi os->os_raw_receive = B_TRUE; 2865eb633035STom Caputi } 2866eb633035STom Caputi 2867eb633035STom Caputi gmep->raw = raw; 28680fa1b3ccSPaul Dagnelie gmep->guid = dsl_dataset_phys(snapds)->ds_guid; 28690fa1b3ccSPaul Dagnelie gmep->gme_ds = snapds; 28700fa1b3ccSPaul Dagnelie avl_add(guid_map, gmep); 28710fa1b3ccSPaul Dagnelie } else { 28720fa1b3ccSPaul Dagnelie kmem_free(gmep, sizeof (*gmep)); 28730fa1b3ccSPaul Dagnelie } 28740fa1b3ccSPaul Dagnelie 28750fa1b3ccSPaul Dagnelie dsl_pool_rele(dp, FTAG); 28760fa1b3ccSPaul Dagnelie return (err); 28770fa1b3ccSPaul Dagnelie } 28780fa1b3ccSPaul Dagnelie 28790fa1b3ccSPaul Dagnelie static int dmu_recv_end_modified_blocks = 3; 28800fa1b3ccSPaul Dagnelie 28810fa1b3ccSPaul Dagnelie static int 28820fa1b3ccSPaul Dagnelie dmu_recv_existing_end(dmu_recv_cookie_t *drc) 28830fa1b3ccSPaul Dagnelie { 28840fa1b3ccSPaul Dagnelie #ifdef _KERNEL 28850fa1b3ccSPaul Dagnelie /* 28860fa1b3ccSPaul Dagnelie * We will be destroying the ds; make sure its origin is unmounted if 28870fa1b3ccSPaul Dagnelie * necessary. 28880fa1b3ccSPaul Dagnelie */ 28890fa1b3ccSPaul Dagnelie char name[ZFS_MAX_DATASET_NAME_LEN]; 28900fa1b3ccSPaul Dagnelie dsl_dataset_name(drc->drc_ds, name); 28910fa1b3ccSPaul Dagnelie zfs_destroy_unmount_origin(name); 28920fa1b3ccSPaul Dagnelie #endif 28930fa1b3ccSPaul Dagnelie 28940fa1b3ccSPaul Dagnelie return (dsl_sync_task(drc->drc_tofs, 28950fa1b3ccSPaul Dagnelie dmu_recv_end_check, dmu_recv_end_sync, drc, 28960fa1b3ccSPaul Dagnelie dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); 28970fa1b3ccSPaul Dagnelie } 28980fa1b3ccSPaul Dagnelie 28990fa1b3ccSPaul Dagnelie static int 29000fa1b3ccSPaul Dagnelie dmu_recv_new_end(dmu_recv_cookie_t *drc) 29010fa1b3ccSPaul Dagnelie { 29020fa1b3ccSPaul Dagnelie return (dsl_sync_task(drc->drc_tofs, 29030fa1b3ccSPaul Dagnelie dmu_recv_end_check, dmu_recv_end_sync, drc, 29040fa1b3ccSPaul Dagnelie dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); 29050fa1b3ccSPaul Dagnelie } 29060fa1b3ccSPaul Dagnelie 29070fa1b3ccSPaul Dagnelie int 29080fa1b3ccSPaul Dagnelie dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) 29090fa1b3ccSPaul Dagnelie { 29100fa1b3ccSPaul Dagnelie int error; 29110fa1b3ccSPaul Dagnelie 29120fa1b3ccSPaul Dagnelie drc->drc_owner = owner; 29130fa1b3ccSPaul Dagnelie 29140fa1b3ccSPaul Dagnelie if (drc->drc_newfs) 29150fa1b3ccSPaul Dagnelie error = dmu_recv_new_end(drc); 29160fa1b3ccSPaul Dagnelie else 29170fa1b3ccSPaul Dagnelie error = dmu_recv_existing_end(drc); 29180fa1b3ccSPaul Dagnelie 29190fa1b3ccSPaul Dagnelie if (error != 0) { 29200fa1b3ccSPaul Dagnelie dmu_recv_cleanup_ds(drc); 2921eb633035STom Caputi nvlist_free(drc->drc_keynvl); 29220fa1b3ccSPaul Dagnelie } else if (drc->drc_guid_to_ds_map != NULL) { 2923eb633035STom Caputi (void) add_ds_to_guidmap(drc->drc_tofs, drc->drc_guid_to_ds_map, 2924eb633035STom Caputi drc->drc_newsnapobj, drc->drc_raw); 29250fa1b3ccSPaul Dagnelie } 29260fa1b3ccSPaul Dagnelie return (error); 29270fa1b3ccSPaul Dagnelie } 29280fa1b3ccSPaul Dagnelie 29290fa1b3ccSPaul Dagnelie /* 29300fa1b3ccSPaul Dagnelie * Return TRUE if this objset is currently being received into. 29310fa1b3ccSPaul Dagnelie */ 29320fa1b3ccSPaul Dagnelie boolean_t 29330fa1b3ccSPaul Dagnelie dmu_objset_is_receiving(objset_t *os) 29340fa1b3ccSPaul Dagnelie { 29350fa1b3ccSPaul Dagnelie return (os->os_dsl_dataset != NULL && 29360fa1b3ccSPaul Dagnelie os->os_dsl_dataset->ds_owner == dmu_recv_tag); 29370fa1b3ccSPaul Dagnelie } 2938