1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 2294d1a210STim Haley * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23be6fd75aSMatthew Ahrens * Copyright (c) 2013 by Delphix. All rights reserved. 24fa9e4066Sahrens */ 25aad02571SSaso Kiselkov /* Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */ 26810e43b2SBill Pijewski /* Copyright (c) 2013, Joyent, Inc. All rights reserved. */ 27aad02571SSaso Kiselkov 28fa9e4066Sahrens #include <sys/dmu.h> 29fa9e4066Sahrens #include <sys/dmu_impl.h> 30fa9e4066Sahrens #include <sys/dmu_tx.h> 31fa9e4066Sahrens #include <sys/dbuf.h> 32fa9e4066Sahrens #include <sys/dnode.h> 33fa9e4066Sahrens #include <sys/zfs_context.h> 34fa9e4066Sahrens #include <sys/dmu_objset.h> 35fa9e4066Sahrens #include <sys/dmu_traverse.h> 36fa9e4066Sahrens #include <sys/dsl_dataset.h> 37fa9e4066Sahrens #include <sys/dsl_dir.h> 38fa9e4066Sahrens #include <sys/dsl_pool.h> 391d452cf5Sahrens #include <sys/dsl_synctask.h> 40a2eea2e1Sahrens #include <sys/dsl_prop.h> 41fa9e4066Sahrens #include <sys/dmu_zfetch.h> 42fa9e4066Sahrens #include <sys/zfs_ioctl.h> 43fa9e4066Sahrens #include <sys/zap.h> 44ea8dc4b6Seschrock #include <sys/zio_checksum.h> 4580901aeaSGeorge Wilson #include <sys/zio_compress.h> 460a586ceaSMark Shellenbaum #include <sys/sa.h> 4744eda4d7Smaybee #ifdef _KERNEL 4844eda4d7Smaybee #include <sys/vmsystm.h> 490fab61baSJonathan W Adams #include <sys/zfs_znode.h> 5044eda4d7Smaybee #endif 51fa9e4066Sahrens 5280901aeaSGeorge Wilson /* 5380901aeaSGeorge Wilson * Enable/disable nopwrite feature. 5480901aeaSGeorge Wilson */ 5580901aeaSGeorge Wilson int zfs_nopwrite_enabled = 1; 5680901aeaSGeorge Wilson 57fa9e4066Sahrens const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { 58ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "unallocated" }, 59ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "object directory" }, 60ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "object array" }, 61ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "packed nvlist" }, 62ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" }, 63ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "bpobj" }, 64ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "bpobj header" }, 65ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "SPA space map header" }, 66ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "SPA space map" }, 67ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" }, 68ad135b5dSChristopher Siden { DMU_BSWAP_DNODE, TRUE, "DMU dnode" }, 69ad135b5dSChristopher Siden { DMU_BSWAP_OBJSET, TRUE, "DMU objset" }, 70ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "DSL directory" }, 71ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"}, 72ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" }, 73ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL props" }, 74ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "DSL dataset" }, 75ad135b5dSChristopher Siden { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" }, 76ad135b5dSChristopher Siden { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" }, 77ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" }, 78ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS directory" }, 79ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS master node" }, 80ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" }, 81ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "zvol object" }, 82ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "zvol prop" }, 83ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "other uint8[]" }, 84ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, FALSE, "other uint64[]" }, 85ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "other ZAP" }, 86ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "persistent error log" }, 87ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "SPA history" }, 88ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" }, 89ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "Pool properties" }, 90ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL permissions" }, 91ad135b5dSChristopher Siden { DMU_BSWAP_ACL, TRUE, "ZFS ACL" }, 92ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" }, 93ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "FUID table" }, 94ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "FUID table size" }, 95ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"}, 96ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "scan work queue" }, 97ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" }, 98ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" }, 99ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"}, 100ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" }, 101ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DDT statistics" }, 102ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "System attributes" }, 103ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "SA master node" }, 104ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "SA attr registration" }, 105ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" }, 106ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "scan translations" }, 107ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "deduplicated block" }, 108ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" }, 109ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" }, 110ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" }, 111ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" } 112ad135b5dSChristopher Siden }; 113ad135b5dSChristopher Siden 114ad135b5dSChristopher Siden const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { 115ad135b5dSChristopher Siden { byteswap_uint8_array, "uint8" }, 116ad135b5dSChristopher Siden { byteswap_uint16_array, "uint16" }, 117ad135b5dSChristopher Siden { byteswap_uint32_array, "uint32" }, 118ad135b5dSChristopher Siden { byteswap_uint64_array, "uint64" }, 119ad135b5dSChristopher Siden { zap_byteswap, "zap" }, 120ad135b5dSChristopher Siden { dnode_buf_byteswap, "dnode" }, 121ad135b5dSChristopher Siden { dmu_objset_byteswap, "objset" }, 122ad135b5dSChristopher Siden { zfs_znode_byteswap, "znode" }, 123ad135b5dSChristopher Siden { zfs_oldacl_byteswap, "oldacl" }, 124ad135b5dSChristopher Siden { zfs_acl_byteswap, "acl" } 1253f9d6ad7SLin Ling }; 126fa9e4066Sahrens 127fa9e4066Sahrens int 128ea8dc4b6Seschrock dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 12947cb52daSJeff Bonwick void *tag, dmu_buf_t **dbp, int flags) 130fa9e4066Sahrens { 131fa9e4066Sahrens dnode_t *dn; 132fa9e4066Sahrens uint64_t blkid; 133fa9e4066Sahrens dmu_buf_impl_t *db; 134ea8dc4b6Seschrock int err; 13547cb52daSJeff Bonwick int db_flags = DB_RF_CANFAIL; 13647cb52daSJeff Bonwick 13747cb52daSJeff Bonwick if (flags & DMU_READ_NO_PREFETCH) 13847cb52daSJeff Bonwick db_flags |= DB_RF_NOPREFETCH; 139fa9e4066Sahrens 140503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 141ea8dc4b6Seschrock if (err) 142ea8dc4b6Seschrock return (err); 143fa9e4066Sahrens blkid = dbuf_whichblock(dn, offset); 144fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 145ea8dc4b6Seschrock db = dbuf_hold(dn, blkid, tag); 146fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 147ea8dc4b6Seschrock if (db == NULL) { 148be6fd75aSMatthew Ahrens err = SET_ERROR(EIO); 149ea8dc4b6Seschrock } else { 15047cb52daSJeff Bonwick err = dbuf_read(db, NULL, db_flags); 151ea8dc4b6Seschrock if (err) { 152ea8dc4b6Seschrock dbuf_rele(db, tag); 153ea8dc4b6Seschrock db = NULL; 154ea8dc4b6Seschrock } 155ea8dc4b6Seschrock } 156fa9e4066Sahrens 157ea8dc4b6Seschrock dnode_rele(dn, FTAG); 158744947dcSTom Erickson *dbp = &db->db; /* NULL db plus first field offset is NULL */ 159ea8dc4b6Seschrock return (err); 160fa9e4066Sahrens } 161fa9e4066Sahrens 162fa9e4066Sahrens int 163fa9e4066Sahrens dmu_bonus_max(void) 164fa9e4066Sahrens { 165fa9e4066Sahrens return (DN_MAX_BONUSLEN); 166fa9e4066Sahrens } 167fa9e4066Sahrens 1681934e92fSmaybee int 169744947dcSTom Erickson dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) 1701934e92fSmaybee { 171744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 172744947dcSTom Erickson dnode_t *dn; 173744947dcSTom Erickson int error; 1741934e92fSmaybee 175744947dcSTom Erickson DB_DNODE_ENTER(db); 176744947dcSTom Erickson dn = DB_DNODE(db); 177744947dcSTom Erickson 178744947dcSTom Erickson if (dn->dn_bonus != db) { 179be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 180744947dcSTom Erickson } else if (newsize < 0 || newsize > db_fake->db_size) { 181be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 182744947dcSTom Erickson } else { 183744947dcSTom Erickson dnode_setbonuslen(dn, newsize, tx); 184744947dcSTom Erickson error = 0; 185744947dcSTom Erickson } 186744947dcSTom Erickson 187744947dcSTom Erickson DB_DNODE_EXIT(db); 188744947dcSTom Erickson return (error); 1891934e92fSmaybee } 1901934e92fSmaybee 1910a586ceaSMark Shellenbaum int 192744947dcSTom Erickson dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) 1930a586ceaSMark Shellenbaum { 194744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 195744947dcSTom Erickson dnode_t *dn; 196744947dcSTom Erickson int error; 1970a586ceaSMark Shellenbaum 198744947dcSTom Erickson DB_DNODE_ENTER(db); 199744947dcSTom Erickson dn = DB_DNODE(db); 200744947dcSTom Erickson 201ad135b5dSChristopher Siden if (!DMU_OT_IS_VALID(type)) { 202be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 203744947dcSTom Erickson } else if (dn->dn_bonus != db) { 204be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 205744947dcSTom Erickson } else { 206744947dcSTom Erickson dnode_setbonus_type(dn, type, tx); 207744947dcSTom Erickson error = 0; 208744947dcSTom Erickson } 2090a586ceaSMark Shellenbaum 210744947dcSTom Erickson DB_DNODE_EXIT(db); 211744947dcSTom Erickson return (error); 212744947dcSTom Erickson } 2130a586ceaSMark Shellenbaum 214744947dcSTom Erickson dmu_object_type_t 215744947dcSTom Erickson dmu_get_bonustype(dmu_buf_t *db_fake) 216744947dcSTom Erickson { 217744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 218744947dcSTom Erickson dnode_t *dn; 219744947dcSTom Erickson dmu_object_type_t type; 220744947dcSTom Erickson 221744947dcSTom Erickson DB_DNODE_ENTER(db); 222744947dcSTom Erickson dn = DB_DNODE(db); 223744947dcSTom Erickson type = dn->dn_bonustype; 224744947dcSTom Erickson DB_DNODE_EXIT(db); 225744947dcSTom Erickson 226744947dcSTom Erickson return (type); 2270a586ceaSMark Shellenbaum } 2280a586ceaSMark Shellenbaum 2290a586ceaSMark Shellenbaum int 2300a586ceaSMark Shellenbaum dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) 2310a586ceaSMark Shellenbaum { 2320a586ceaSMark Shellenbaum dnode_t *dn; 2330a586ceaSMark Shellenbaum int error; 2340a586ceaSMark Shellenbaum 2350a586ceaSMark Shellenbaum error = dnode_hold(os, object, FTAG, &dn); 2360a586ceaSMark Shellenbaum dbuf_rm_spill(dn, tx); 23706e0070dSMark Shellenbaum rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 23806e0070dSMark Shellenbaum dnode_rm_spill(dn, tx); 23906e0070dSMark Shellenbaum rw_exit(&dn->dn_struct_rwlock); 2400a586ceaSMark Shellenbaum dnode_rele(dn, FTAG); 2410a586ceaSMark Shellenbaum return (error); 2420a586ceaSMark Shellenbaum } 2430a586ceaSMark Shellenbaum 244fa9e4066Sahrens /* 245ea8dc4b6Seschrock * returns ENOENT, EIO, or 0. 246fa9e4066Sahrens */ 247ea8dc4b6Seschrock int 248ea8dc4b6Seschrock dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) 249fa9e4066Sahrens { 250ea8dc4b6Seschrock dnode_t *dn; 251fa9e4066Sahrens dmu_buf_impl_t *db; 2521934e92fSmaybee int error; 253fa9e4066Sahrens 254503ad85cSMatthew Ahrens error = dnode_hold(os, object, FTAG, &dn); 2551934e92fSmaybee if (error) 2561934e92fSmaybee return (error); 257fa9e4066Sahrens 258ea8dc4b6Seschrock rw_enter(&dn->dn_struct_rwlock, RW_READER); 259ea8dc4b6Seschrock if (dn->dn_bonus == NULL) { 260fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 261ea8dc4b6Seschrock rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 262ea8dc4b6Seschrock if (dn->dn_bonus == NULL) 2631934e92fSmaybee dbuf_create_bonus(dn); 264fa9e4066Sahrens } 265ea8dc4b6Seschrock db = dn->dn_bonus; 2661934e92fSmaybee 2671934e92fSmaybee /* as long as the bonus buf is held, the dnode will be held */ 268744947dcSTom Erickson if (refcount_add(&db->db_holds, tag) == 1) { 2691934e92fSmaybee VERIFY(dnode_add_ref(dn, db)); 270744947dcSTom Erickson (void) atomic_inc_32_nv(&dn->dn_dbufs_count); 271744947dcSTom Erickson } 272744947dcSTom Erickson 273744947dcSTom Erickson /* 274744947dcSTom Erickson * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's 275744947dcSTom Erickson * hold and incrementing the dbuf count to ensure that dnode_move() sees 276744947dcSTom Erickson * a dnode hold for every dbuf. 277744947dcSTom Erickson */ 278744947dcSTom Erickson rw_exit(&dn->dn_struct_rwlock); 2791934e92fSmaybee 280fa9e4066Sahrens dnode_rele(dn, FTAG); 281ea8dc4b6Seschrock 28247cb52daSJeff Bonwick VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); 283ea8dc4b6Seschrock 284ea8dc4b6Seschrock *dbp = &db->db; 285ea8dc4b6Seschrock return (0); 286fa9e4066Sahrens } 287fa9e4066Sahrens 2880a586ceaSMark Shellenbaum /* 2890a586ceaSMark Shellenbaum * returns ENOENT, EIO, or 0. 2900a586ceaSMark Shellenbaum * 2910a586ceaSMark Shellenbaum * This interface will allocate a blank spill dbuf when a spill blk 2920a586ceaSMark Shellenbaum * doesn't already exist on the dnode. 2930a586ceaSMark Shellenbaum * 2940a586ceaSMark Shellenbaum * if you only want to find an already existing spill db, then 2950a586ceaSMark Shellenbaum * dmu_spill_hold_existing() should be used. 2960a586ceaSMark Shellenbaum */ 2970a586ceaSMark Shellenbaum int 2980a586ceaSMark Shellenbaum dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) 2990a586ceaSMark Shellenbaum { 3000a586ceaSMark Shellenbaum dmu_buf_impl_t *db = NULL; 3010a586ceaSMark Shellenbaum int err; 3020a586ceaSMark Shellenbaum 3030a586ceaSMark Shellenbaum if ((flags & DB_RF_HAVESTRUCT) == 0) 3040a586ceaSMark Shellenbaum rw_enter(&dn->dn_struct_rwlock, RW_READER); 3050a586ceaSMark Shellenbaum 3060a586ceaSMark Shellenbaum db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); 3070a586ceaSMark Shellenbaum 3080a586ceaSMark Shellenbaum if ((flags & DB_RF_HAVESTRUCT) == 0) 3090a586ceaSMark Shellenbaum rw_exit(&dn->dn_struct_rwlock); 3100a586ceaSMark Shellenbaum 3110a586ceaSMark Shellenbaum ASSERT(db != NULL); 3121d8ccc7bSMark Shellenbaum err = dbuf_read(db, NULL, flags); 3131d8ccc7bSMark Shellenbaum if (err == 0) 3141d8ccc7bSMark Shellenbaum *dbp = &db->db; 3151d8ccc7bSMark Shellenbaum else 3161d8ccc7bSMark Shellenbaum dbuf_rele(db, tag); 3170a586ceaSMark Shellenbaum return (err); 3180a586ceaSMark Shellenbaum } 3190a586ceaSMark Shellenbaum 3200a586ceaSMark Shellenbaum int 3210a586ceaSMark Shellenbaum dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 3220a586ceaSMark Shellenbaum { 323744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; 324744947dcSTom Erickson dnode_t *dn; 3250a586ceaSMark Shellenbaum int err; 3260a586ceaSMark Shellenbaum 327744947dcSTom Erickson DB_DNODE_ENTER(db); 328744947dcSTom Erickson dn = DB_DNODE(db); 329744947dcSTom Erickson 330744947dcSTom Erickson if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { 331be6fd75aSMatthew Ahrens err = SET_ERROR(EINVAL); 332744947dcSTom Erickson } else { 333744947dcSTom Erickson rw_enter(&dn->dn_struct_rwlock, RW_READER); 334744947dcSTom Erickson 335744947dcSTom Erickson if (!dn->dn_have_spill) { 336be6fd75aSMatthew Ahrens err = SET_ERROR(ENOENT); 337744947dcSTom Erickson } else { 338744947dcSTom Erickson err = dmu_spill_hold_by_dnode(dn, 339744947dcSTom Erickson DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); 340744947dcSTom Erickson } 3410a586ceaSMark Shellenbaum 3420a586ceaSMark Shellenbaum rw_exit(&dn->dn_struct_rwlock); 3430a586ceaSMark Shellenbaum } 344744947dcSTom Erickson 345744947dcSTom Erickson DB_DNODE_EXIT(db); 3460a586ceaSMark Shellenbaum return (err); 3470a586ceaSMark Shellenbaum } 3480a586ceaSMark Shellenbaum 3490a586ceaSMark Shellenbaum int 3500a586ceaSMark Shellenbaum dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 3510a586ceaSMark Shellenbaum { 352744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; 353744947dcSTom Erickson dnode_t *dn; 354744947dcSTom Erickson int err; 355744947dcSTom Erickson 356744947dcSTom Erickson DB_DNODE_ENTER(db); 357744947dcSTom Erickson dn = DB_DNODE(db); 358744947dcSTom Erickson err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); 359744947dcSTom Erickson DB_DNODE_EXIT(db); 360744947dcSTom Erickson 361744947dcSTom Erickson return (err); 3620a586ceaSMark Shellenbaum } 3630a586ceaSMark Shellenbaum 36413506d1eSmaybee /* 36513506d1eSmaybee * Note: longer-term, we should modify all of the dmu_buf_*() interfaces 36613506d1eSmaybee * to take a held dnode rather than <os, object> -- the lookup is wasteful, 36713506d1eSmaybee * and can induce severe lock contention when writing to several files 36813506d1eSmaybee * whose dnodes are in the same block. 36913506d1eSmaybee */ 37013506d1eSmaybee static int 3717bfdf011SNeil Perrin dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, 3727bfdf011SNeil Perrin int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) 373fa9e4066Sahrens { 374fa9e4066Sahrens dmu_buf_t **dbp; 375fa9e4066Sahrens uint64_t blkid, nblks, i; 3767bfdf011SNeil Perrin uint32_t dbuf_flags; 377ea8dc4b6Seschrock int err; 378ea8dc4b6Seschrock zio_t *zio; 379ea8dc4b6Seschrock 380ea8dc4b6Seschrock ASSERT(length <= DMU_MAX_ACCESS); 381fa9e4066Sahrens 382c87b8fc5SMark J Musante dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT; 3837bfdf011SNeil Perrin if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz) 3847bfdf011SNeil Perrin dbuf_flags |= DB_RF_NOPREFETCH; 385ea8dc4b6Seschrock 386fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 387fa9e4066Sahrens if (dn->dn_datablkshift) { 388fa9e4066Sahrens int blkshift = dn->dn_datablkshift; 389fa9e4066Sahrens nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - 390b1b8ab34Slling P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; 391fa9e4066Sahrens } else { 3920125049cSahrens if (offset + length > dn->dn_datablksz) { 3930125049cSahrens zfs_panic_recover("zfs: accessing past end of object " 3940125049cSahrens "%llx/%llx (size=%u access=%llu+%llu)", 3950125049cSahrens (longlong_t)dn->dn_objset-> 3960125049cSahrens os_dsl_dataset->ds_object, 3970125049cSahrens (longlong_t)dn->dn_object, dn->dn_datablksz, 3980125049cSahrens (longlong_t)offset, (longlong_t)length); 399c87b8fc5SMark J Musante rw_exit(&dn->dn_struct_rwlock); 400be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 4010125049cSahrens } 402fa9e4066Sahrens nblks = 1; 403fa9e4066Sahrens } 404ea8dc4b6Seschrock dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); 405fa9e4066Sahrens 406e14bb325SJeff Bonwick zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); 407fa9e4066Sahrens blkid = dbuf_whichblock(dn, offset); 408fa9e4066Sahrens for (i = 0; i < nblks; i++) { 409ea8dc4b6Seschrock dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); 410ea8dc4b6Seschrock if (db == NULL) { 411ea8dc4b6Seschrock rw_exit(&dn->dn_struct_rwlock); 412ea8dc4b6Seschrock dmu_buf_rele_array(dbp, nblks, tag); 413ea8dc4b6Seschrock zio_nowait(zio); 414be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 415ea8dc4b6Seschrock } 416ea8dc4b6Seschrock /* initiate async i/o */ 41713506d1eSmaybee if (read) { 4187bfdf011SNeil Perrin (void) dbuf_read(db, zio, dbuf_flags); 419ea8dc4b6Seschrock } 420ea8dc4b6Seschrock dbp[i] = &db->db; 421fa9e4066Sahrens } 422fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 423fa9e4066Sahrens 424ea8dc4b6Seschrock /* wait for async i/o */ 425ea8dc4b6Seschrock err = zio_wait(zio); 426ea8dc4b6Seschrock if (err) { 427ea8dc4b6Seschrock dmu_buf_rele_array(dbp, nblks, tag); 428ea8dc4b6Seschrock return (err); 429ea8dc4b6Seschrock } 430fa9e4066Sahrens 431ea8dc4b6Seschrock /* wait for other io to complete */ 432ea8dc4b6Seschrock if (read) { 433ea8dc4b6Seschrock for (i = 0; i < nblks; i++) { 434ea8dc4b6Seschrock dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; 435ea8dc4b6Seschrock mutex_enter(&db->db_mtx); 436ea8dc4b6Seschrock while (db->db_state == DB_READ || 437ea8dc4b6Seschrock db->db_state == DB_FILL) 438ea8dc4b6Seschrock cv_wait(&db->db_changed, &db->db_mtx); 439ea8dc4b6Seschrock if (db->db_state == DB_UNCACHED) 440be6fd75aSMatthew Ahrens err = SET_ERROR(EIO); 441ea8dc4b6Seschrock mutex_exit(&db->db_mtx); 442ea8dc4b6Seschrock if (err) { 443ea8dc4b6Seschrock dmu_buf_rele_array(dbp, nblks, tag); 444ea8dc4b6Seschrock return (err); 445ea8dc4b6Seschrock } 446ea8dc4b6Seschrock } 447ea8dc4b6Seschrock } 448fa9e4066Sahrens 449ea8dc4b6Seschrock *numbufsp = nblks; 450ea8dc4b6Seschrock *dbpp = dbp; 451ea8dc4b6Seschrock return (0); 452fa9e4066Sahrens } 453fa9e4066Sahrens 454a2eea2e1Sahrens static int 45513506d1eSmaybee dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, 45613506d1eSmaybee uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 45713506d1eSmaybee { 45813506d1eSmaybee dnode_t *dn; 45913506d1eSmaybee int err; 46013506d1eSmaybee 461503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 46213506d1eSmaybee if (err) 46313506d1eSmaybee return (err); 46413506d1eSmaybee 46513506d1eSmaybee err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 4667bfdf011SNeil Perrin numbufsp, dbpp, DMU_READ_PREFETCH); 46713506d1eSmaybee 46813506d1eSmaybee dnode_rele(dn, FTAG); 46913506d1eSmaybee 47013506d1eSmaybee return (err); 47113506d1eSmaybee } 47213506d1eSmaybee 47313506d1eSmaybee int 474744947dcSTom Erickson dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, 47513506d1eSmaybee uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 47613506d1eSmaybee { 477744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 478744947dcSTom Erickson dnode_t *dn; 47913506d1eSmaybee int err; 48013506d1eSmaybee 481744947dcSTom Erickson DB_DNODE_ENTER(db); 482744947dcSTom Erickson dn = DB_DNODE(db); 48313506d1eSmaybee err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 4847bfdf011SNeil Perrin numbufsp, dbpp, DMU_READ_PREFETCH); 485744947dcSTom Erickson DB_DNODE_EXIT(db); 48613506d1eSmaybee 48713506d1eSmaybee return (err); 48813506d1eSmaybee } 48913506d1eSmaybee 490fa9e4066Sahrens void 491ea8dc4b6Seschrock dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) 492fa9e4066Sahrens { 493fa9e4066Sahrens int i; 494fa9e4066Sahrens dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; 495fa9e4066Sahrens 496fa9e4066Sahrens if (numbufs == 0) 497fa9e4066Sahrens return; 498fa9e4066Sahrens 499ea8dc4b6Seschrock for (i = 0; i < numbufs; i++) { 500ea8dc4b6Seschrock if (dbp[i]) 501ea8dc4b6Seschrock dbuf_rele(dbp[i], tag); 502ea8dc4b6Seschrock } 503fa9e4066Sahrens 504fa9e4066Sahrens kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); 505fa9e4066Sahrens } 506fa9e4066Sahrens 507*69962b56SMatthew Ahrens /* 508*69962b56SMatthew Ahrens * Issue prefetch i/os for the given blocks. 509*69962b56SMatthew Ahrens * 510*69962b56SMatthew Ahrens * Note: The assumption is that we *know* these blocks will be needed 511*69962b56SMatthew Ahrens * almost immediately. Therefore, the prefetch i/os will be issued at 512*69962b56SMatthew Ahrens * ZIO_PRIORITY_SYNC_READ 513*69962b56SMatthew Ahrens * 514*69962b56SMatthew Ahrens * Note: indirect blocks and other metadata will be read synchronously, 515*69962b56SMatthew Ahrens * causing this function to block if they are not already cached. 516*69962b56SMatthew Ahrens */ 517fa9e4066Sahrens void 518fa9e4066Sahrens dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) 519fa9e4066Sahrens { 520fa9e4066Sahrens dnode_t *dn; 521fa9e4066Sahrens uint64_t blkid; 522*69962b56SMatthew Ahrens int nblks, err; 523fa9e4066Sahrens 524416e0cd8Sek if (zfs_prefetch_disable) 525416e0cd8Sek return; 526416e0cd8Sek 527fa9e4066Sahrens if (len == 0) { /* they're interested in the bonus buffer */ 528744947dcSTom Erickson dn = DMU_META_DNODE(os); 529fa9e4066Sahrens 530fa9e4066Sahrens if (object == 0 || object >= DN_MAX_OBJECT) 531fa9e4066Sahrens return; 532fa9e4066Sahrens 533fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 534fa9e4066Sahrens blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); 535*69962b56SMatthew Ahrens dbuf_prefetch(dn, blkid, ZIO_PRIORITY_SYNC_READ); 536fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 537fa9e4066Sahrens return; 538fa9e4066Sahrens } 539fa9e4066Sahrens 540fa9e4066Sahrens /* 541fa9e4066Sahrens * XXX - Note, if the dnode for the requested object is not 542fa9e4066Sahrens * already cached, we will do a *synchronous* read in the 543fa9e4066Sahrens * dnode_hold() call. The same is true for any indirects. 544fa9e4066Sahrens */ 545503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 546ea8dc4b6Seschrock if (err != 0) 547fa9e4066Sahrens return; 548fa9e4066Sahrens 549fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 550fa9e4066Sahrens if (dn->dn_datablkshift) { 551fa9e4066Sahrens int blkshift = dn->dn_datablkshift; 552*69962b56SMatthew Ahrens nblks = (P2ROUNDUP(offset + len, 1 << blkshift) - 553*69962b56SMatthew Ahrens P2ALIGN(offset, 1 << blkshift)) >> blkshift; 554fa9e4066Sahrens } else { 555fa9e4066Sahrens nblks = (offset < dn->dn_datablksz); 556fa9e4066Sahrens } 557fa9e4066Sahrens 558fa9e4066Sahrens if (nblks != 0) { 559fa9e4066Sahrens blkid = dbuf_whichblock(dn, offset); 560*69962b56SMatthew Ahrens for (int i = 0; i < nblks; i++) 561*69962b56SMatthew Ahrens dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_SYNC_READ); 562fa9e4066Sahrens } 563fa9e4066Sahrens 564fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 565fa9e4066Sahrens 566fa9e4066Sahrens dnode_rele(dn, FTAG); 567fa9e4066Sahrens } 568fa9e4066Sahrens 56976256205SMark Maybee /* 57076256205SMark Maybee * Get the next "chunk" of file data to free. We traverse the file from 57176256205SMark Maybee * the end so that the file gets shorter over time (if we crashes in the 57276256205SMark Maybee * middle, this will leave us in a better state). We find allocated file 57376256205SMark Maybee * data by simply searching the allocated level 1 indirects. 574713d6c20SMatthew Ahrens * 575713d6c20SMatthew Ahrens * On input, *start should be the first offset that does not need to be 576713d6c20SMatthew Ahrens * freed (e.g. "offset + length"). On return, *start will be the first 577713d6c20SMatthew Ahrens * offset that should be freed. 57876256205SMark Maybee */ 579cdb0ab79Smaybee static int 580713d6c20SMatthew Ahrens get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum) 581cdb0ab79Smaybee { 582713d6c20SMatthew Ahrens uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); 583713d6c20SMatthew Ahrens /* bytes of data covered by a level-1 indirect block */ 58476256205SMark Maybee uint64_t iblkrange = 5851c8564a7SMark Maybee dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); 586cdb0ab79Smaybee 587713d6c20SMatthew Ahrens ASSERT3U(minimum, <=, *start); 588cdb0ab79Smaybee 589713d6c20SMatthew Ahrens if (*start - minimum <= iblkrange * maxblks) { 590713d6c20SMatthew Ahrens *start = minimum; 591cdb0ab79Smaybee return (0); 592cdb0ab79Smaybee } 59376256205SMark Maybee ASSERT(ISP2(iblkrange)); 594cdb0ab79Smaybee 595713d6c20SMatthew Ahrens for (uint64_t blks = 0; *start > minimum && blks < maxblks; blks++) { 5961c8564a7SMark Maybee int err; 597cdb0ab79Smaybee 598713d6c20SMatthew Ahrens /* 599713d6c20SMatthew Ahrens * dnode_next_offset(BACKWARDS) will find an allocated L1 600713d6c20SMatthew Ahrens * indirect block at or before the input offset. We must 601713d6c20SMatthew Ahrens * decrement *start so that it is at the end of the region 602713d6c20SMatthew Ahrens * to search. 603713d6c20SMatthew Ahrens */ 604713d6c20SMatthew Ahrens (*start)--; 605cdb0ab79Smaybee err = dnode_next_offset(dn, 60676256205SMark Maybee DNODE_FIND_BACKWARDS, start, 2, 1, 0); 607cdb0ab79Smaybee 608713d6c20SMatthew Ahrens /* if there are no indirect blocks before start, we are done */ 60976256205SMark Maybee if (err == ESRCH) { 610713d6c20SMatthew Ahrens *start = minimum; 611713d6c20SMatthew Ahrens break; 612713d6c20SMatthew Ahrens } else if (err != 0) { 613cdb0ab79Smaybee return (err); 61476256205SMark Maybee } 61576256205SMark Maybee 616713d6c20SMatthew Ahrens /* set start to the beginning of this L1 indirect */ 61776256205SMark Maybee *start = P2ALIGN(*start, iblkrange); 618cdb0ab79Smaybee } 619713d6c20SMatthew Ahrens if (*start < minimum) 620713d6c20SMatthew Ahrens *start = minimum; 621cdb0ab79Smaybee return (0); 622cdb0ab79Smaybee } 623cdb0ab79Smaybee 624cdb0ab79Smaybee static int 625cdb0ab79Smaybee dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, 626713d6c20SMatthew Ahrens uint64_t length) 627cdb0ab79Smaybee { 628713d6c20SMatthew Ahrens uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; 629713d6c20SMatthew Ahrens int err; 630713d6c20SMatthew Ahrens 631713d6c20SMatthew Ahrens if (offset >= object_size) 632cdb0ab79Smaybee return (0); 633cdb0ab79Smaybee 634713d6c20SMatthew Ahrens if (length == DMU_OBJECT_END || offset + length > object_size) 635713d6c20SMatthew Ahrens length = object_size - offset; 636713d6c20SMatthew Ahrens 637713d6c20SMatthew Ahrens while (length != 0) { 638713d6c20SMatthew Ahrens uint64_t chunk_end, chunk_begin; 639713d6c20SMatthew Ahrens 640713d6c20SMatthew Ahrens chunk_end = chunk_begin = offset + length; 641713d6c20SMatthew Ahrens 642713d6c20SMatthew Ahrens /* move chunk_begin backwards to the beginning of this chunk */ 643713d6c20SMatthew Ahrens err = get_next_chunk(dn, &chunk_begin, offset); 644cdb0ab79Smaybee if (err) 645cdb0ab79Smaybee return (err); 646713d6c20SMatthew Ahrens ASSERT3U(chunk_begin, >=, offset); 647713d6c20SMatthew Ahrens ASSERT3U(chunk_begin, <=, chunk_end); 648cdb0ab79Smaybee 649713d6c20SMatthew Ahrens dmu_tx_t *tx = dmu_tx_create(os); 650713d6c20SMatthew Ahrens dmu_tx_hold_free(tx, dn->dn_object, 651713d6c20SMatthew Ahrens chunk_begin, chunk_end - chunk_begin); 652cdb0ab79Smaybee err = dmu_tx_assign(tx, TXG_WAIT); 653cdb0ab79Smaybee if (err) { 654cdb0ab79Smaybee dmu_tx_abort(tx); 655cdb0ab79Smaybee return (err); 656cdb0ab79Smaybee } 657713d6c20SMatthew Ahrens dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx); 658cdb0ab79Smaybee dmu_tx_commit(tx); 659713d6c20SMatthew Ahrens 660713d6c20SMatthew Ahrens length -= chunk_end - chunk_begin; 661cdb0ab79Smaybee } 662cdb0ab79Smaybee return (0); 663cdb0ab79Smaybee } 664cdb0ab79Smaybee 665cdb0ab79Smaybee int 666cdb0ab79Smaybee dmu_free_long_range(objset_t *os, uint64_t object, 667cdb0ab79Smaybee uint64_t offset, uint64_t length) 668cdb0ab79Smaybee { 669cdb0ab79Smaybee dnode_t *dn; 670cdb0ab79Smaybee int err; 671cdb0ab79Smaybee 672503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 673cdb0ab79Smaybee if (err != 0) 674cdb0ab79Smaybee return (err); 675713d6c20SMatthew Ahrens err = dmu_free_long_range_impl(os, dn, offset, length); 676cdb0ab79Smaybee dnode_rele(dn, FTAG); 677cdb0ab79Smaybee return (err); 678cdb0ab79Smaybee } 679cdb0ab79Smaybee 680cdb0ab79Smaybee int 681713d6c20SMatthew Ahrens dmu_free_long_object(objset_t *os, uint64_t object) 682cdb0ab79Smaybee { 683cdb0ab79Smaybee dmu_tx_t *tx; 684cdb0ab79Smaybee int err; 685cdb0ab79Smaybee 686713d6c20SMatthew Ahrens err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); 687cdb0ab79Smaybee if (err != 0) 688cdb0ab79Smaybee return (err); 689713d6c20SMatthew Ahrens 690713d6c20SMatthew Ahrens tx = dmu_tx_create(os); 691713d6c20SMatthew Ahrens dmu_tx_hold_bonus(tx, object); 692713d6c20SMatthew Ahrens dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 693713d6c20SMatthew Ahrens err = dmu_tx_assign(tx, TXG_WAIT); 694713d6c20SMatthew Ahrens if (err == 0) { 695713d6c20SMatthew Ahrens err = dmu_object_free(os, object, tx); 696713d6c20SMatthew Ahrens dmu_tx_commit(tx); 697cdb0ab79Smaybee } else { 698713d6c20SMatthew Ahrens dmu_tx_abort(tx); 699cdb0ab79Smaybee } 700713d6c20SMatthew Ahrens 701cdb0ab79Smaybee return (err); 702cdb0ab79Smaybee } 703cdb0ab79Smaybee 704ea8dc4b6Seschrock int 705fa9e4066Sahrens dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 706fa9e4066Sahrens uint64_t size, dmu_tx_t *tx) 707fa9e4066Sahrens { 708ea8dc4b6Seschrock dnode_t *dn; 709503ad85cSMatthew Ahrens int err = dnode_hold(os, object, FTAG, &dn); 710ea8dc4b6Seschrock if (err) 711ea8dc4b6Seschrock return (err); 712fa9e4066Sahrens ASSERT(offset < UINT64_MAX); 713fa9e4066Sahrens ASSERT(size == -1ULL || size <= UINT64_MAX - offset); 714fa9e4066Sahrens dnode_free_range(dn, offset, size, tx); 715fa9e4066Sahrens dnode_rele(dn, FTAG); 716ea8dc4b6Seschrock return (0); 717fa9e4066Sahrens } 718fa9e4066Sahrens 719ea8dc4b6Seschrock int 720ea8dc4b6Seschrock dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 7217bfdf011SNeil Perrin void *buf, uint32_t flags) 722fa9e4066Sahrens { 723fa9e4066Sahrens dnode_t *dn; 724fa9e4066Sahrens dmu_buf_t **dbp; 725c87b8fc5SMark J Musante int numbufs, err; 726fa9e4066Sahrens 727503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 728ea8dc4b6Seschrock if (err) 729ea8dc4b6Seschrock return (err); 730feb08c6bSbillm 731feb08c6bSbillm /* 732feb08c6bSbillm * Deal with odd block sizes, where there can't be data past the first 733feb08c6bSbillm * block. If we ever do the tail block optimization, we will need to 734feb08c6bSbillm * handle that here as well. 735feb08c6bSbillm */ 736c87b8fc5SMark J Musante if (dn->dn_maxblkid == 0) { 737fa9e4066Sahrens int newsz = offset > dn->dn_datablksz ? 0 : 738fa9e4066Sahrens MIN(size, dn->dn_datablksz - offset); 739fa9e4066Sahrens bzero((char *)buf + newsz, size - newsz); 740fa9e4066Sahrens size = newsz; 741fa9e4066Sahrens } 742fa9e4066Sahrens 743fa9e4066Sahrens while (size > 0) { 744fa9e4066Sahrens uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); 745c87b8fc5SMark J Musante int i; 746fa9e4066Sahrens 747fa9e4066Sahrens /* 748fa9e4066Sahrens * NB: we could do this block-at-a-time, but it's nice 749fa9e4066Sahrens * to be reading in parallel. 750fa9e4066Sahrens */ 751a2eea2e1Sahrens err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, 7527bfdf011SNeil Perrin TRUE, FTAG, &numbufs, &dbp, flags); 753ea8dc4b6Seschrock if (err) 7541934e92fSmaybee break; 755fa9e4066Sahrens 756fa9e4066Sahrens for (i = 0; i < numbufs; i++) { 757fa9e4066Sahrens int tocpy; 758fa9e4066Sahrens int bufoff; 759fa9e4066Sahrens dmu_buf_t *db = dbp[i]; 760fa9e4066Sahrens 761fa9e4066Sahrens ASSERT(size > 0); 762fa9e4066Sahrens 763fa9e4066Sahrens bufoff = offset - db->db_offset; 764fa9e4066Sahrens tocpy = (int)MIN(db->db_size - bufoff, size); 765fa9e4066Sahrens 766fa9e4066Sahrens bcopy((char *)db->db_data + bufoff, buf, tocpy); 767fa9e4066Sahrens 768fa9e4066Sahrens offset += tocpy; 769fa9e4066Sahrens size -= tocpy; 770fa9e4066Sahrens buf = (char *)buf + tocpy; 771fa9e4066Sahrens } 772ea8dc4b6Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG); 773fa9e4066Sahrens } 774a2eea2e1Sahrens dnode_rele(dn, FTAG); 7751934e92fSmaybee return (err); 776fa9e4066Sahrens } 777fa9e4066Sahrens 778fa9e4066Sahrens void 779fa9e4066Sahrens dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 780fa9e4066Sahrens const void *buf, dmu_tx_t *tx) 781fa9e4066Sahrens { 782fa9e4066Sahrens dmu_buf_t **dbp; 783fa9e4066Sahrens int numbufs, i; 784fa9e4066Sahrens 78513506d1eSmaybee if (size == 0) 78613506d1eSmaybee return; 78713506d1eSmaybee 788ea8dc4b6Seschrock VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 789ea8dc4b6Seschrock FALSE, FTAG, &numbufs, &dbp)); 790fa9e4066Sahrens 791fa9e4066Sahrens for (i = 0; i < numbufs; i++) { 792fa9e4066Sahrens int tocpy; 793fa9e4066Sahrens int bufoff; 794fa9e4066Sahrens dmu_buf_t *db = dbp[i]; 795fa9e4066Sahrens 796fa9e4066Sahrens ASSERT(size > 0); 797fa9e4066Sahrens 798fa9e4066Sahrens bufoff = offset - db->db_offset; 799fa9e4066Sahrens tocpy = (int)MIN(db->db_size - bufoff, size); 800fa9e4066Sahrens 801fa9e4066Sahrens ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 802fa9e4066Sahrens 803fa9e4066Sahrens if (tocpy == db->db_size) 804fa9e4066Sahrens dmu_buf_will_fill(db, tx); 805fa9e4066Sahrens else 806fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 807fa9e4066Sahrens 808fa9e4066Sahrens bcopy(buf, (char *)db->db_data + bufoff, tocpy); 809fa9e4066Sahrens 810fa9e4066Sahrens if (tocpy == db->db_size) 811fa9e4066Sahrens dmu_buf_fill_done(db, tx); 812fa9e4066Sahrens 813fa9e4066Sahrens offset += tocpy; 814fa9e4066Sahrens size -= tocpy; 815fa9e4066Sahrens buf = (char *)buf + tocpy; 816fa9e4066Sahrens } 817ea8dc4b6Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG); 818fa9e4066Sahrens } 819fa9e4066Sahrens 82082c9918fSTim Haley void 82182c9918fSTim Haley dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 82282c9918fSTim Haley dmu_tx_t *tx) 82382c9918fSTim Haley { 82482c9918fSTim Haley dmu_buf_t **dbp; 82582c9918fSTim Haley int numbufs, i; 82682c9918fSTim Haley 82782c9918fSTim Haley if (size == 0) 82882c9918fSTim Haley return; 82982c9918fSTim Haley 83082c9918fSTim Haley VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 83182c9918fSTim Haley FALSE, FTAG, &numbufs, &dbp)); 83282c9918fSTim Haley 83382c9918fSTim Haley for (i = 0; i < numbufs; i++) { 83482c9918fSTim Haley dmu_buf_t *db = dbp[i]; 83582c9918fSTim Haley 83682c9918fSTim Haley dmu_buf_will_not_fill(db, tx); 83782c9918fSTim Haley } 83882c9918fSTim Haley dmu_buf_rele_array(dbp, numbufs, FTAG); 83982c9918fSTim Haley } 84082c9918fSTim Haley 841c242f9a0Schunli zhang - Sun Microsystems - Irvine United States /* 842c242f9a0Schunli zhang - Sun Microsystems - Irvine United States * DMU support for xuio 843c242f9a0Schunli zhang - Sun Microsystems - Irvine United States */ 844c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_t *xuio_ksp = NULL; 845c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 846c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int 847c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_init(xuio_t *xuio, int nblk) 848c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 849c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv; 850c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio_t *uio = &xuio->xu_uio; 851c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 852c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_iovcnt = nblk; 853c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); 854c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 855c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); 856c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->cnt = nblk; 857c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); 858c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->iovp = uio->uio_iov; 859c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIO_XUZC_PRIV(xuio) = priv; 860c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 861c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (XUIO_XUZC_RW(xuio) == UIO_READ) 862c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); 863c242f9a0Schunli zhang - Sun Microsystems - Irvine United States else 864c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); 865c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 866c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (0); 867c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 868c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 869c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 870c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_fini(xuio_t *xuio) 871c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 872c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 873c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int nblk = priv->cnt; 874c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 875c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kmem_free(priv->iovp, nblk * sizeof (iovec_t)); 876c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); 877c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kmem_free(priv, sizeof (dmu_xuio_t)); 878c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 879c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (XUIO_XUZC_RW(xuio) == UIO_READ) 880c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); 881c242f9a0Schunli zhang - Sun Microsystems - Irvine United States else 882c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); 883c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 884c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 885c242f9a0Schunli zhang - Sun Microsystems - Irvine United States /* 886c242f9a0Schunli zhang - Sun Microsystems - Irvine United States * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } 887c242f9a0Schunli zhang - Sun Microsystems - Irvine United States * and increase priv->next by 1. 888c242f9a0Schunli zhang - Sun Microsystems - Irvine United States */ 889c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int 890c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) 891c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 892c242f9a0Schunli zhang - Sun Microsystems - Irvine United States struct iovec *iov; 893c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio_t *uio = &xuio->xu_uio; 894c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 895c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int i = priv->next++; 896c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 897c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(i < priv->cnt); 898c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(off + n <= arc_buf_size(abuf)); 899c242f9a0Schunli zhang - Sun Microsystems - Irvine United States iov = uio->uio_iov + i; 900c242f9a0Schunli zhang - Sun Microsystems - Irvine United States iov->iov_base = (char *)abuf->b_data + off; 901c242f9a0Schunli zhang - Sun Microsystems - Irvine United States iov->iov_len = n; 902c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->bufs[i] = abuf; 903c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (0); 904c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 905c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 906c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int 907c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_cnt(xuio_t *xuio) 908c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 909c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 910c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (priv->cnt); 911c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 912c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 913c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t * 914c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_arcbuf(xuio_t *xuio, int i) 915c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 916c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 917c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 918c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(i < priv->cnt); 919c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (priv->bufs[i]); 920c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 921c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 922c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 923c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_clear(xuio_t *xuio, int i) 924c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 925c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 926c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 927c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(i < priv->cnt); 928c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->bufs[i] = NULL; 929c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 930c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 931c242f9a0Schunli zhang - Sun Microsystems - Irvine United States static void 932c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_init(void) 933c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 934c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", 935c242f9a0Schunli zhang - Sun Microsystems - Irvine United States KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), 936c242f9a0Schunli zhang - Sun Microsystems - Irvine United States KSTAT_FLAG_VIRTUAL); 937c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (xuio_ksp != NULL) { 938c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_ksp->ks_data = &xuio_stats; 939c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_install(xuio_ksp); 940c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 941c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 942c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 943c242f9a0Schunli zhang - Sun Microsystems - Irvine United States static void 944c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_fini(void) 945c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 946c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (xuio_ksp != NULL) { 947c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_delete(xuio_ksp); 948c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_ksp = NULL; 949c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 950c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 951c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 952c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 953c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_wbuf_copied() 954c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 955c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_wbuf_copied); 956c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 957c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 958c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 959c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_wbuf_nocopy() 960c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 961c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); 962c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 963c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 964fa9e4066Sahrens #ifdef _KERNEL 965fa9e4066Sahrens int 966feb08c6bSbillm dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) 967feb08c6bSbillm { 968feb08c6bSbillm dmu_buf_t **dbp; 969feb08c6bSbillm int numbufs, i, err; 970c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_t *xuio = NULL; 971feb08c6bSbillm 972feb08c6bSbillm /* 973feb08c6bSbillm * NB: we could do this block-at-a-time, but it's nice 974feb08c6bSbillm * to be reading in parallel. 975feb08c6bSbillm */ 976feb08c6bSbillm err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, 977feb08c6bSbillm &numbufs, &dbp); 978feb08c6bSbillm if (err) 979feb08c6bSbillm return (err); 980feb08c6bSbillm 981c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (uio->uio_extflg == UIO_XUIO) 982c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio = (xuio_t *)uio; 983c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 984feb08c6bSbillm for (i = 0; i < numbufs; i++) { 985feb08c6bSbillm int tocpy; 986feb08c6bSbillm int bufoff; 987feb08c6bSbillm dmu_buf_t *db = dbp[i]; 988feb08c6bSbillm 989feb08c6bSbillm ASSERT(size > 0); 990feb08c6bSbillm 991feb08c6bSbillm bufoff = uio->uio_loffset - db->db_offset; 992feb08c6bSbillm tocpy = (int)MIN(db->db_size - bufoff, size); 993feb08c6bSbillm 994c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (xuio) { 995c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; 996c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t *dbuf_abuf = dbi->db_buf; 997c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); 998c242f9a0Schunli zhang - Sun Microsystems - Irvine United States err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); 999c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (!err) { 1000c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_resid -= tocpy; 1001c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_loffset += tocpy; 1002c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1003c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 1004c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (abuf == dbuf_abuf) 1005c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); 1006c242f9a0Schunli zhang - Sun Microsystems - Irvine United States else 1007c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_rbuf_copied); 1008c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } else { 1009c242f9a0Schunli zhang - Sun Microsystems - Irvine United States err = uiomove((char *)db->db_data + bufoff, tocpy, 1010c242f9a0Schunli zhang - Sun Microsystems - Irvine United States UIO_READ, uio); 1011c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1012feb08c6bSbillm if (err) 1013feb08c6bSbillm break; 1014feb08c6bSbillm 1015feb08c6bSbillm size -= tocpy; 1016feb08c6bSbillm } 1017feb08c6bSbillm dmu_buf_rele_array(dbp, numbufs, FTAG); 1018feb08c6bSbillm 1019feb08c6bSbillm return (err); 1020feb08c6bSbillm } 1021feb08c6bSbillm 102294d1a210STim Haley static int 102394d1a210STim Haley dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) 1024fa9e4066Sahrens { 1025fa9e4066Sahrens dmu_buf_t **dbp; 102694d1a210STim Haley int numbufs; 1027fa9e4066Sahrens int err = 0; 102894d1a210STim Haley int i; 1029fa9e4066Sahrens 103094d1a210STim Haley err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, 103194d1a210STim Haley FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); 1032ea8dc4b6Seschrock if (err) 1033ea8dc4b6Seschrock return (err); 1034fa9e4066Sahrens 1035fa9e4066Sahrens for (i = 0; i < numbufs; i++) { 1036fa9e4066Sahrens int tocpy; 1037fa9e4066Sahrens int bufoff; 1038fa9e4066Sahrens dmu_buf_t *db = dbp[i]; 1039fa9e4066Sahrens 1040fa9e4066Sahrens ASSERT(size > 0); 1041fa9e4066Sahrens 1042feb08c6bSbillm bufoff = uio->uio_loffset - db->db_offset; 1043fa9e4066Sahrens tocpy = (int)MIN(db->db_size - bufoff, size); 1044fa9e4066Sahrens 1045fa9e4066Sahrens ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 1046fa9e4066Sahrens 1047fa9e4066Sahrens if (tocpy == db->db_size) 1048fa9e4066Sahrens dmu_buf_will_fill(db, tx); 1049fa9e4066Sahrens else 1050fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 1051fa9e4066Sahrens 1052fa9e4066Sahrens /* 1053fa9e4066Sahrens * XXX uiomove could block forever (eg. nfs-backed 1054fa9e4066Sahrens * pages). There needs to be a uiolockdown() function 1055fa9e4066Sahrens * to lock the pages in memory, so that uiomove won't 1056fa9e4066Sahrens * block. 1057fa9e4066Sahrens */ 1058fa9e4066Sahrens err = uiomove((char *)db->db_data + bufoff, tocpy, 1059fa9e4066Sahrens UIO_WRITE, uio); 1060fa9e4066Sahrens 1061fa9e4066Sahrens if (tocpy == db->db_size) 1062fa9e4066Sahrens dmu_buf_fill_done(db, tx); 1063fa9e4066Sahrens 1064fa9e4066Sahrens if (err) 1065fa9e4066Sahrens break; 1066fa9e4066Sahrens 1067fa9e4066Sahrens size -= tocpy; 1068fa9e4066Sahrens } 106994d1a210STim Haley 1070ea8dc4b6Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG); 1071fa9e4066Sahrens return (err); 1072fa9e4066Sahrens } 107344eda4d7Smaybee 107494d1a210STim Haley int 107594d1a210STim Haley dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, 107694d1a210STim Haley dmu_tx_t *tx) 107794d1a210STim Haley { 1078744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; 1079744947dcSTom Erickson dnode_t *dn; 1080744947dcSTom Erickson int err; 1081744947dcSTom Erickson 108294d1a210STim Haley if (size == 0) 108394d1a210STim Haley return (0); 108494d1a210STim Haley 1085744947dcSTom Erickson DB_DNODE_ENTER(db); 1086744947dcSTom Erickson dn = DB_DNODE(db); 1087744947dcSTom Erickson err = dmu_write_uio_dnode(dn, uio, size, tx); 1088744947dcSTom Erickson DB_DNODE_EXIT(db); 1089744947dcSTom Erickson 1090744947dcSTom Erickson return (err); 109194d1a210STim Haley } 109294d1a210STim Haley 109394d1a210STim Haley int 109494d1a210STim Haley dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, 109594d1a210STim Haley dmu_tx_t *tx) 109694d1a210STim Haley { 109794d1a210STim Haley dnode_t *dn; 109894d1a210STim Haley int err; 109994d1a210STim Haley 110094d1a210STim Haley if (size == 0) 110194d1a210STim Haley return (0); 110294d1a210STim Haley 110394d1a210STim Haley err = dnode_hold(os, object, FTAG, &dn); 110494d1a210STim Haley if (err) 110594d1a210STim Haley return (err); 110694d1a210STim Haley 110794d1a210STim Haley err = dmu_write_uio_dnode(dn, uio, size, tx); 110894d1a210STim Haley 110994d1a210STim Haley dnode_rele(dn, FTAG); 111094d1a210STim Haley 111194d1a210STim Haley return (err); 111294d1a210STim Haley } 111394d1a210STim Haley 111444eda4d7Smaybee int 111544eda4d7Smaybee dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 111644eda4d7Smaybee page_t *pp, dmu_tx_t *tx) 111744eda4d7Smaybee { 111844eda4d7Smaybee dmu_buf_t **dbp; 111944eda4d7Smaybee int numbufs, i; 112044eda4d7Smaybee int err; 112144eda4d7Smaybee 112244eda4d7Smaybee if (size == 0) 112344eda4d7Smaybee return (0); 112444eda4d7Smaybee 112544eda4d7Smaybee err = dmu_buf_hold_array(os, object, offset, size, 112644eda4d7Smaybee FALSE, FTAG, &numbufs, &dbp); 112744eda4d7Smaybee if (err) 112844eda4d7Smaybee return (err); 112944eda4d7Smaybee 113044eda4d7Smaybee for (i = 0; i < numbufs; i++) { 113144eda4d7Smaybee int tocpy, copied, thiscpy; 113244eda4d7Smaybee int bufoff; 113344eda4d7Smaybee dmu_buf_t *db = dbp[i]; 113444eda4d7Smaybee caddr_t va; 113544eda4d7Smaybee 113644eda4d7Smaybee ASSERT(size > 0); 113744eda4d7Smaybee ASSERT3U(db->db_size, >=, PAGESIZE); 113844eda4d7Smaybee 113944eda4d7Smaybee bufoff = offset - db->db_offset; 114044eda4d7Smaybee tocpy = (int)MIN(db->db_size - bufoff, size); 114144eda4d7Smaybee 114244eda4d7Smaybee ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 114344eda4d7Smaybee 114444eda4d7Smaybee if (tocpy == db->db_size) 114544eda4d7Smaybee dmu_buf_will_fill(db, tx); 114644eda4d7Smaybee else 114744eda4d7Smaybee dmu_buf_will_dirty(db, tx); 114844eda4d7Smaybee 114944eda4d7Smaybee for (copied = 0; copied < tocpy; copied += PAGESIZE) { 115044eda4d7Smaybee ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); 115144eda4d7Smaybee thiscpy = MIN(PAGESIZE, tocpy - copied); 11520fab61baSJonathan W Adams va = zfs_map_page(pp, S_READ); 115344eda4d7Smaybee bcopy(va, (char *)db->db_data + bufoff, thiscpy); 11540fab61baSJonathan W Adams zfs_unmap_page(pp, va); 115544eda4d7Smaybee pp = pp->p_next; 115644eda4d7Smaybee bufoff += PAGESIZE; 115744eda4d7Smaybee } 115844eda4d7Smaybee 115944eda4d7Smaybee if (tocpy == db->db_size) 116044eda4d7Smaybee dmu_buf_fill_done(db, tx); 116144eda4d7Smaybee 116244eda4d7Smaybee offset += tocpy; 116344eda4d7Smaybee size -= tocpy; 116444eda4d7Smaybee } 116544eda4d7Smaybee dmu_buf_rele_array(dbp, numbufs, FTAG); 116644eda4d7Smaybee return (err); 116744eda4d7Smaybee } 1168fa9e4066Sahrens #endif 1169fa9e4066Sahrens 11702fdbea25SAleksandr Guzovskiy /* 11712fdbea25SAleksandr Guzovskiy * Allocate a loaned anonymous arc buffer. 11722fdbea25SAleksandr Guzovskiy */ 11732fdbea25SAleksandr Guzovskiy arc_buf_t * 11742fdbea25SAleksandr Guzovskiy dmu_request_arcbuf(dmu_buf_t *handle, int size) 11752fdbea25SAleksandr Guzovskiy { 1176744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; 1177744947dcSTom Erickson spa_t *spa; 11782fdbea25SAleksandr Guzovskiy 1179744947dcSTom Erickson DB_GET_SPA(&spa, db); 1180744947dcSTom Erickson return (arc_loan_buf(spa, size)); 11812fdbea25SAleksandr Guzovskiy } 11822fdbea25SAleksandr Guzovskiy 11832fdbea25SAleksandr Guzovskiy /* 11842fdbea25SAleksandr Guzovskiy * Free a loaned arc buffer. 11852fdbea25SAleksandr Guzovskiy */ 11862fdbea25SAleksandr Guzovskiy void 11872fdbea25SAleksandr Guzovskiy dmu_return_arcbuf(arc_buf_t *buf) 11882fdbea25SAleksandr Guzovskiy { 11892fdbea25SAleksandr Guzovskiy arc_return_buf(buf, FTAG); 11903b2aab18SMatthew Ahrens VERIFY(arc_buf_remove_ref(buf, FTAG)); 11912fdbea25SAleksandr Guzovskiy } 11922fdbea25SAleksandr Guzovskiy 11932fdbea25SAleksandr Guzovskiy /* 11942fdbea25SAleksandr Guzovskiy * When possible directly assign passed loaned arc buffer to a dbuf. 11952fdbea25SAleksandr Guzovskiy * If this is not possible copy the contents of passed arc buf via 11962fdbea25SAleksandr Guzovskiy * dmu_write(). 11972fdbea25SAleksandr Guzovskiy */ 11982fdbea25SAleksandr Guzovskiy void 11992fdbea25SAleksandr Guzovskiy dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, 12002fdbea25SAleksandr Guzovskiy dmu_tx_t *tx) 12012fdbea25SAleksandr Guzovskiy { 1202744947dcSTom Erickson dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; 1203744947dcSTom Erickson dnode_t *dn; 12042fdbea25SAleksandr Guzovskiy dmu_buf_impl_t *db; 12052fdbea25SAleksandr Guzovskiy uint32_t blksz = (uint32_t)arc_buf_size(buf); 12062fdbea25SAleksandr Guzovskiy uint64_t blkid; 12072fdbea25SAleksandr Guzovskiy 1208744947dcSTom Erickson DB_DNODE_ENTER(dbuf); 1209744947dcSTom Erickson dn = DB_DNODE(dbuf); 12102fdbea25SAleksandr Guzovskiy rw_enter(&dn->dn_struct_rwlock, RW_READER); 12112fdbea25SAleksandr Guzovskiy blkid = dbuf_whichblock(dn, offset); 12122fdbea25SAleksandr Guzovskiy VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); 12132fdbea25SAleksandr Guzovskiy rw_exit(&dn->dn_struct_rwlock); 1214744947dcSTom Erickson DB_DNODE_EXIT(dbuf); 12152fdbea25SAleksandr Guzovskiy 12162fdbea25SAleksandr Guzovskiy if (offset == db->db.db_offset && blksz == db->db.db_size) { 12172fdbea25SAleksandr Guzovskiy dbuf_assign_arcbuf(db, buf, tx); 12182fdbea25SAleksandr Guzovskiy dbuf_rele(db, FTAG); 12192fdbea25SAleksandr Guzovskiy } else { 1220744947dcSTom Erickson objset_t *os; 1221744947dcSTom Erickson uint64_t object; 1222744947dcSTom Erickson 1223744947dcSTom Erickson DB_DNODE_ENTER(dbuf); 1224744947dcSTom Erickson dn = DB_DNODE(dbuf); 1225744947dcSTom Erickson os = dn->dn_objset; 1226744947dcSTom Erickson object = dn->dn_object; 1227744947dcSTom Erickson DB_DNODE_EXIT(dbuf); 1228744947dcSTom Erickson 12292fdbea25SAleksandr Guzovskiy dbuf_rele(db, FTAG); 1230744947dcSTom Erickson dmu_write(os, object, offset, blksz, buf->b_data, tx); 12312fdbea25SAleksandr Guzovskiy dmu_return_arcbuf(buf); 1232c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_wbuf_copied); 12332fdbea25SAleksandr Guzovskiy } 12342fdbea25SAleksandr Guzovskiy } 12352fdbea25SAleksandr Guzovskiy 1236c5c6ffa0Smaybee typedef struct { 1237b24ab676SJeff Bonwick dbuf_dirty_record_t *dsa_dr; 1238b24ab676SJeff Bonwick dmu_sync_cb_t *dsa_done; 1239b24ab676SJeff Bonwick zgd_t *dsa_zgd; 1240b24ab676SJeff Bonwick dmu_tx_t *dsa_tx; 1241c717a561Smaybee } dmu_sync_arg_t; 1242c5c6ffa0Smaybee 1243e14bb325SJeff Bonwick /* ARGSUSED */ 1244e14bb325SJeff Bonwick static void 1245e14bb325SJeff Bonwick dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) 1246e14bb325SJeff Bonwick { 1247b24ab676SJeff Bonwick dmu_sync_arg_t *dsa = varg; 1248b24ab676SJeff Bonwick dmu_buf_t *db = dsa->dsa_zgd->zgd_db; 1249e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1250975c32a0SNeil Perrin 1251b24ab676SJeff Bonwick if (zio->io_error == 0) { 1252b24ab676SJeff Bonwick if (BP_IS_HOLE(bp)) { 1253b24ab676SJeff Bonwick /* 1254b24ab676SJeff Bonwick * A block of zeros may compress to a hole, but the 1255b24ab676SJeff Bonwick * block size still needs to be known for replay. 1256b24ab676SJeff Bonwick */ 1257b24ab676SJeff Bonwick BP_SET_LSIZE(bp, db->db_size); 1258b24ab676SJeff Bonwick } else { 1259b24ab676SJeff Bonwick ASSERT(BP_GET_LEVEL(bp) == 0); 1260b24ab676SJeff Bonwick bp->blk_fill = 1; 1261b24ab676SJeff Bonwick } 1262e14bb325SJeff Bonwick } 1263e14bb325SJeff Bonwick } 1264e14bb325SJeff Bonwick 1265b24ab676SJeff Bonwick static void 1266b24ab676SJeff Bonwick dmu_sync_late_arrival_ready(zio_t *zio) 1267b24ab676SJeff Bonwick { 1268b24ab676SJeff Bonwick dmu_sync_ready(zio, NULL, zio->io_private); 1269b24ab676SJeff Bonwick } 1270b24ab676SJeff Bonwick 1271c5c6ffa0Smaybee /* ARGSUSED */ 1272c5c6ffa0Smaybee static void 1273c5c6ffa0Smaybee dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) 1274c5c6ffa0Smaybee { 1275b24ab676SJeff Bonwick dmu_sync_arg_t *dsa = varg; 1276b24ab676SJeff Bonwick dbuf_dirty_record_t *dr = dsa->dsa_dr; 1277c717a561Smaybee dmu_buf_impl_t *db = dr->dr_dbuf; 1278c5c6ffa0Smaybee 1279b50a0fe0SNeil Perrin mutex_enter(&db->db_mtx); 1280b50a0fe0SNeil Perrin ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); 1281b24ab676SJeff Bonwick if (zio->io_error == 0) { 128280901aeaSGeorge Wilson dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); 128380901aeaSGeorge Wilson if (dr->dt.dl.dr_nopwrite) { 128480901aeaSGeorge Wilson blkptr_t *bp = zio->io_bp; 128580901aeaSGeorge Wilson blkptr_t *bp_orig = &zio->io_bp_orig; 128680901aeaSGeorge Wilson uint8_t chksum = BP_GET_CHECKSUM(bp_orig); 128780901aeaSGeorge Wilson 128880901aeaSGeorge Wilson ASSERT(BP_EQUAL(bp, bp_orig)); 128980901aeaSGeorge Wilson ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); 129080901aeaSGeorge Wilson ASSERT(zio_checksum_table[chksum].ci_dedup); 129180901aeaSGeorge Wilson } 1292b24ab676SJeff Bonwick dr->dt.dl.dr_overridden_by = *zio->io_bp; 1293b24ab676SJeff Bonwick dr->dt.dl.dr_override_state = DR_OVERRIDDEN; 1294b24ab676SJeff Bonwick dr->dt.dl.dr_copies = zio->io_prop.zp_copies; 1295b24ab676SJeff Bonwick if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by)) 1296b24ab676SJeff Bonwick BP_ZERO(&dr->dt.dl.dr_overridden_by); 1297b24ab676SJeff Bonwick } else { 1298b24ab676SJeff Bonwick dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; 1299b24ab676SJeff Bonwick } 1300c5c6ffa0Smaybee cv_broadcast(&db->db_changed); 1301b50a0fe0SNeil Perrin mutex_exit(&db->db_mtx); 1302b50a0fe0SNeil Perrin 1303b24ab676SJeff Bonwick dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1304b24ab676SJeff Bonwick 1305b24ab676SJeff Bonwick kmem_free(dsa, sizeof (*dsa)); 1306b24ab676SJeff Bonwick } 1307b24ab676SJeff Bonwick 1308b24ab676SJeff Bonwick static void 1309b24ab676SJeff Bonwick dmu_sync_late_arrival_done(zio_t *zio) 1310b24ab676SJeff Bonwick { 1311b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1312b24ab676SJeff Bonwick dmu_sync_arg_t *dsa = zio->io_private; 131380901aeaSGeorge Wilson blkptr_t *bp_orig = &zio->io_bp_orig; 1314b24ab676SJeff Bonwick 1315b24ab676SJeff Bonwick if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { 131680901aeaSGeorge Wilson /* 131780901aeaSGeorge Wilson * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE) 131880901aeaSGeorge Wilson * then there is nothing to do here. Otherwise, free the 131980901aeaSGeorge Wilson * newly allocated block in this txg. 132080901aeaSGeorge Wilson */ 132180901aeaSGeorge Wilson if (zio->io_flags & ZIO_FLAG_NOPWRITE) { 132280901aeaSGeorge Wilson ASSERT(BP_EQUAL(bp, bp_orig)); 132380901aeaSGeorge Wilson } else { 132480901aeaSGeorge Wilson ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); 132580901aeaSGeorge Wilson ASSERT(zio->io_bp->blk_birth == zio->io_txg); 132680901aeaSGeorge Wilson ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); 132780901aeaSGeorge Wilson zio_free(zio->io_spa, zio->io_txg, zio->io_bp); 132880901aeaSGeorge Wilson } 1329b24ab676SJeff Bonwick } 1330b24ab676SJeff Bonwick 1331b24ab676SJeff Bonwick dmu_tx_commit(dsa->dsa_tx); 1332b24ab676SJeff Bonwick 1333b24ab676SJeff Bonwick dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1334b24ab676SJeff Bonwick 1335b24ab676SJeff Bonwick kmem_free(dsa, sizeof (*dsa)); 1336b24ab676SJeff Bonwick } 1337b24ab676SJeff Bonwick 1338b24ab676SJeff Bonwick static int 1339b24ab676SJeff Bonwick dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, 1340b24ab676SJeff Bonwick zio_prop_t *zp, zbookmark_t *zb) 1341b24ab676SJeff Bonwick { 1342b24ab676SJeff Bonwick dmu_sync_arg_t *dsa; 1343b24ab676SJeff Bonwick dmu_tx_t *tx; 1344b24ab676SJeff Bonwick 1345b24ab676SJeff Bonwick tx = dmu_tx_create(os); 1346b24ab676SJeff Bonwick dmu_tx_hold_space(tx, zgd->zgd_db->db_size); 13476e1f5caaSNeil Perrin if (dmu_tx_assign(tx, TXG_WAIT) != 0) { 1348b24ab676SJeff Bonwick dmu_tx_abort(tx); 1349be6fd75aSMatthew Ahrens /* Make zl_get_data do txg_waited_synced() */ 1350be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 1351b24ab676SJeff Bonwick } 1352b24ab676SJeff Bonwick 1353b24ab676SJeff Bonwick dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1354b24ab676SJeff Bonwick dsa->dsa_dr = NULL; 1355b24ab676SJeff Bonwick dsa->dsa_done = done; 1356b24ab676SJeff Bonwick dsa->dsa_zgd = zgd; 1357b24ab676SJeff Bonwick dsa->dsa_tx = tx; 1358c717a561Smaybee 1359b24ab676SJeff Bonwick zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, 1360b24ab676SJeff Bonwick zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, 1361*69962b56SMatthew Ahrens dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa, 1362b24ab676SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); 1363b24ab676SJeff Bonwick 1364b24ab676SJeff Bonwick return (0); 1365c5c6ffa0Smaybee } 1366c5c6ffa0Smaybee 1367fa9e4066Sahrens /* 1368c5c6ffa0Smaybee * Intent log support: sync the block associated with db to disk. 1369c5c6ffa0Smaybee * N.B. and XXX: the caller is responsible for making sure that the 1370c5c6ffa0Smaybee * data isn't changing while dmu_sync() is writing it. 1371fa9e4066Sahrens * 1372fa9e4066Sahrens * Return values: 1373fa9e4066Sahrens * 137480901aeaSGeorge Wilson * EEXIST: this txg has already been synced, so there's nothing to do. 1375fa9e4066Sahrens * The caller should not log the write. 1376fa9e4066Sahrens * 1377fa9e4066Sahrens * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. 1378fa9e4066Sahrens * The caller should not log the write. 1379fa9e4066Sahrens * 1380c5c6ffa0Smaybee * EALREADY: this block is already in the process of being synced. 1381c5c6ffa0Smaybee * The caller should track its progress (somehow). 1382fa9e4066Sahrens * 1383b24ab676SJeff Bonwick * EIO: could not do the I/O. 1384b24ab676SJeff Bonwick * The caller should do a txg_wait_synced(). 1385fa9e4066Sahrens * 1386b24ab676SJeff Bonwick * 0: the I/O has been initiated. 1387b24ab676SJeff Bonwick * The caller should log this blkptr in the done callback. 1388b24ab676SJeff Bonwick * It is possible that the I/O will fail, in which case 1389b24ab676SJeff Bonwick * the error will be reported to the done callback and 1390b24ab676SJeff Bonwick * propagated to pio from zio_done(). 1391fa9e4066Sahrens */ 1392fa9e4066Sahrens int 1393b24ab676SJeff Bonwick dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) 1394fa9e4066Sahrens { 1395b24ab676SJeff Bonwick blkptr_t *bp = zgd->zgd_bp; 1396b24ab676SJeff Bonwick dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; 1397503ad85cSMatthew Ahrens objset_t *os = db->db_objset; 1398b24ab676SJeff Bonwick dsl_dataset_t *ds = os->os_dsl_dataset; 1399c717a561Smaybee dbuf_dirty_record_t *dr; 1400b24ab676SJeff Bonwick dmu_sync_arg_t *dsa; 1401ea8dc4b6Seschrock zbookmark_t zb; 1402b24ab676SJeff Bonwick zio_prop_t zp; 1403744947dcSTom Erickson dnode_t *dn; 1404fa9e4066Sahrens 1405b24ab676SJeff Bonwick ASSERT(pio != NULL); 1406fa9e4066Sahrens ASSERT(txg != 0); 1407fa9e4066Sahrens 1408b24ab676SJeff Bonwick SET_BOOKMARK(&zb, ds->ds_object, 1409b24ab676SJeff Bonwick db->db.db_object, db->db_level, db->db_blkid); 1410b24ab676SJeff Bonwick 1411744947dcSTom Erickson DB_DNODE_ENTER(db); 1412744947dcSTom Erickson dn = DB_DNODE(db); 1413744947dcSTom Erickson dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); 1414744947dcSTom Erickson DB_DNODE_EXIT(db); 1415fa9e4066Sahrens 1416ea8dc4b6Seschrock /* 1417b24ab676SJeff Bonwick * If we're frozen (running ziltest), we always need to generate a bp. 1418ea8dc4b6Seschrock */ 1419b24ab676SJeff Bonwick if (txg > spa_freeze_txg(os->os_spa)) 1420b24ab676SJeff Bonwick return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1421ea8dc4b6Seschrock 1422fa9e4066Sahrens /* 1423b24ab676SJeff Bonwick * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() 1424b24ab676SJeff Bonwick * and us. If we determine that this txg is not yet syncing, 1425b24ab676SJeff Bonwick * but it begins to sync a moment later, that's OK because the 1426b24ab676SJeff Bonwick * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. 1427fa9e4066Sahrens */ 1428b24ab676SJeff Bonwick mutex_enter(&db->db_mtx); 1429b24ab676SJeff Bonwick 1430b24ab676SJeff Bonwick if (txg <= spa_last_synced_txg(os->os_spa)) { 1431fa9e4066Sahrens /* 1432b24ab676SJeff Bonwick * This txg has already synced. There's nothing to do. 1433fa9e4066Sahrens */ 1434b24ab676SJeff Bonwick mutex_exit(&db->db_mtx); 1435be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 1436fa9e4066Sahrens } 1437fa9e4066Sahrens 1438b24ab676SJeff Bonwick if (txg <= spa_syncing_txg(os->os_spa)) { 1439b24ab676SJeff Bonwick /* 1440b24ab676SJeff Bonwick * This txg is currently syncing, so we can't mess with 1441b24ab676SJeff Bonwick * the dirty record anymore; just write a new log block. 1442b24ab676SJeff Bonwick */ 1443b24ab676SJeff Bonwick mutex_exit(&db->db_mtx); 1444b24ab676SJeff Bonwick return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1445fa9e4066Sahrens } 1446fa9e4066Sahrens 1447c717a561Smaybee dr = db->db_last_dirty; 1448b24ab676SJeff Bonwick while (dr && dr->dr_txg != txg) 1449c717a561Smaybee dr = dr->dr_next; 1450b24ab676SJeff Bonwick 1451b24ab676SJeff Bonwick if (dr == NULL) { 1452c5c6ffa0Smaybee /* 1453b24ab676SJeff Bonwick * There's no dr for this dbuf, so it must have been freed. 1454c5c6ffa0Smaybee * There's no need to log writes to freed blocks, so we're done. 1455c5c6ffa0Smaybee */ 1456c5c6ffa0Smaybee mutex_exit(&db->db_mtx); 1457be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 1458c5c6ffa0Smaybee } 1459c5c6ffa0Smaybee 146080901aeaSGeorge Wilson ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); 146180901aeaSGeorge Wilson 146280901aeaSGeorge Wilson /* 146380901aeaSGeorge Wilson * Assume the on-disk data is X, the current syncing data is Y, 146480901aeaSGeorge Wilson * and the current in-memory data is Z (currently in dmu_sync). 146580901aeaSGeorge Wilson * X and Z are identical but Y is has been modified. Normally, 146680901aeaSGeorge Wilson * when X and Z are the same we will perform a nopwrite but if Y 146780901aeaSGeorge Wilson * is different we must disable nopwrite since the resulting write 146880901aeaSGeorge Wilson * of Y to disk can free the block containing X. If we allowed a 146980901aeaSGeorge Wilson * nopwrite to occur the block pointing to Z would reference a freed 147080901aeaSGeorge Wilson * block. Since this is a rare case we simplify this by disabling 147180901aeaSGeorge Wilson * nopwrite if the current dmu_sync-ing dbuf has been modified in 147280901aeaSGeorge Wilson * a previous transaction. 147380901aeaSGeorge Wilson */ 147480901aeaSGeorge Wilson if (dr->dr_next) 147580901aeaSGeorge Wilson zp.zp_nopwrite = B_FALSE; 147680901aeaSGeorge Wilson 1477c717a561Smaybee ASSERT(dr->dr_txg == txg); 1478b24ab676SJeff Bonwick if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || 1479b24ab676SJeff Bonwick dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { 1480c717a561Smaybee /* 1481b24ab676SJeff Bonwick * We have already issued a sync write for this buffer, 1482b24ab676SJeff Bonwick * or this buffer has already been synced. It could not 1483c717a561Smaybee * have been dirtied since, or we would have cleared the state. 1484c717a561Smaybee */ 1485c717a561Smaybee mutex_exit(&db->db_mtx); 1486be6fd75aSMatthew Ahrens return (SET_ERROR(EALREADY)); 1487c717a561Smaybee } 1488c717a561Smaybee 1489b24ab676SJeff Bonwick ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); 1490c717a561Smaybee dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; 1491fa9e4066Sahrens mutex_exit(&db->db_mtx); 1492e14bb325SJeff Bonwick 1493b24ab676SJeff Bonwick dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1494b24ab676SJeff Bonwick dsa->dsa_dr = dr; 1495b24ab676SJeff Bonwick dsa->dsa_done = done; 1496b24ab676SJeff Bonwick dsa->dsa_zgd = zgd; 1497b24ab676SJeff Bonwick dsa->dsa_tx = NULL; 1498e14bb325SJeff Bonwick 1499b24ab676SJeff Bonwick zio_nowait(arc_write(pio, os->os_spa, txg, 1500aad02571SSaso Kiselkov bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), 1501*69962b56SMatthew Ahrens DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready, 1502*69962b56SMatthew Ahrens NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE, 1503*69962b56SMatthew Ahrens ZIO_FLAG_CANFAIL, &zb)); 1504e14bb325SJeff Bonwick 1505b24ab676SJeff Bonwick return (0); 1506fa9e4066Sahrens } 1507fa9e4066Sahrens 1508fa9e4066Sahrens int 1509fa9e4066Sahrens dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, 1510fa9e4066Sahrens dmu_tx_t *tx) 1511fa9e4066Sahrens { 1512ea8dc4b6Seschrock dnode_t *dn; 1513ea8dc4b6Seschrock int err; 1514ea8dc4b6Seschrock 1515503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 1516ea8dc4b6Seschrock if (err) 1517ea8dc4b6Seschrock return (err); 1518ea8dc4b6Seschrock err = dnode_set_blksz(dn, size, ibs, tx); 1519fa9e4066Sahrens dnode_rele(dn, FTAG); 1520fa9e4066Sahrens return (err); 1521fa9e4066Sahrens } 1522fa9e4066Sahrens 1523fa9e4066Sahrens void 1524fa9e4066Sahrens dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 1525fa9e4066Sahrens dmu_tx_t *tx) 1526fa9e4066Sahrens { 1527ea8dc4b6Seschrock dnode_t *dn; 1528ea8dc4b6Seschrock 1529ea8dc4b6Seschrock /* XXX assumes dnode_hold will not get an i/o error */ 1530503ad85cSMatthew Ahrens (void) dnode_hold(os, object, FTAG, &dn); 1531fa9e4066Sahrens ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS); 1532fa9e4066Sahrens dn->dn_checksum = checksum; 1533fa9e4066Sahrens dnode_setdirty(dn, tx); 1534fa9e4066Sahrens dnode_rele(dn, FTAG); 1535fa9e4066Sahrens } 1536fa9e4066Sahrens 1537fa9e4066Sahrens void 1538fa9e4066Sahrens dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 1539fa9e4066Sahrens dmu_tx_t *tx) 1540fa9e4066Sahrens { 1541ea8dc4b6Seschrock dnode_t *dn; 1542ea8dc4b6Seschrock 1543ea8dc4b6Seschrock /* XXX assumes dnode_hold will not get an i/o error */ 1544503ad85cSMatthew Ahrens (void) dnode_hold(os, object, FTAG, &dn); 1545fa9e4066Sahrens ASSERT(compress < ZIO_COMPRESS_FUNCTIONS); 1546fa9e4066Sahrens dn->dn_compress = compress; 1547fa9e4066Sahrens dnode_setdirty(dn, tx); 1548fa9e4066Sahrens dnode_rele(dn, FTAG); 1549fa9e4066Sahrens } 1550fa9e4066Sahrens 1551b24ab676SJeff Bonwick int zfs_mdcomp_disable = 0; 1552b24ab676SJeff Bonwick 1553b24ab676SJeff Bonwick void 1554b24ab676SJeff Bonwick dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) 1555b24ab676SJeff Bonwick { 1556b24ab676SJeff Bonwick dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; 1557ad135b5dSChristopher Siden boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || 15581d8ccc7bSMark Shellenbaum (wp & WP_SPILL)); 1559b24ab676SJeff Bonwick enum zio_checksum checksum = os->os_checksum; 1560b24ab676SJeff Bonwick enum zio_compress compress = os->os_compress; 1561b24ab676SJeff Bonwick enum zio_checksum dedup_checksum = os->os_dedup_checksum; 15627540df39SGeorge Wilson boolean_t dedup = B_FALSE; 15637540df39SGeorge Wilson boolean_t nopwrite = B_FALSE; 1564b24ab676SJeff Bonwick boolean_t dedup_verify = os->os_dedup_verify; 1565b24ab676SJeff Bonwick int copies = os->os_copies; 1566b24ab676SJeff Bonwick 1567b24ab676SJeff Bonwick /* 156880901aeaSGeorge Wilson * We maintain different write policies for each of the following 156980901aeaSGeorge Wilson * types of data: 157080901aeaSGeorge Wilson * 1. metadata 157180901aeaSGeorge Wilson * 2. preallocated blocks (i.e. level-0 blocks of a dump device) 157280901aeaSGeorge Wilson * 3. all other level 0 blocks 1573b24ab676SJeff Bonwick */ 1574b24ab676SJeff Bonwick if (ismd) { 157580901aeaSGeorge Wilson /* 157680901aeaSGeorge Wilson * XXX -- we should design a compression algorithm 157780901aeaSGeorge Wilson * that specializes in arrays of bps. 157880901aeaSGeorge Wilson */ 157980901aeaSGeorge Wilson compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY : 158080901aeaSGeorge Wilson ZIO_COMPRESS_LZJB; 158180901aeaSGeorge Wilson 1582b24ab676SJeff Bonwick /* 1583b24ab676SJeff Bonwick * Metadata always gets checksummed. If the data 1584b24ab676SJeff Bonwick * checksum is multi-bit correctable, and it's not a 1585b24ab676SJeff Bonwick * ZBT-style checksum, then it's suitable for metadata 1586b24ab676SJeff Bonwick * as well. Otherwise, the metadata checksum defaults 1587b24ab676SJeff Bonwick * to fletcher4. 1588b24ab676SJeff Bonwick */ 1589b24ab676SJeff Bonwick if (zio_checksum_table[checksum].ci_correctable < 1 || 15906e1f5caaSNeil Perrin zio_checksum_table[checksum].ci_eck) 1591b24ab676SJeff Bonwick checksum = ZIO_CHECKSUM_FLETCHER_4; 159280901aeaSGeorge Wilson } else if (wp & WP_NOFILL) { 159380901aeaSGeorge Wilson ASSERT(level == 0); 159480901aeaSGeorge Wilson 1595b24ab676SJeff Bonwick /* 159680901aeaSGeorge Wilson * If we're writing preallocated blocks, we aren't actually 159780901aeaSGeorge Wilson * writing them so don't set any policy properties. These 159880901aeaSGeorge Wilson * blocks are currently only used by an external subsystem 159980901aeaSGeorge Wilson * outside of zfs (i.e. dump) and not written by the zio 160080901aeaSGeorge Wilson * pipeline. 1601b24ab676SJeff Bonwick */ 160280901aeaSGeorge Wilson compress = ZIO_COMPRESS_OFF; 1603810e43b2SBill Pijewski checksum = ZIO_CHECKSUM_NOPARITY; 1604b24ab676SJeff Bonwick } else { 1605b24ab676SJeff Bonwick compress = zio_compress_select(dn->dn_compress, compress); 1606b24ab676SJeff Bonwick 160780901aeaSGeorge Wilson checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? 160880901aeaSGeorge Wilson zio_checksum_select(dn->dn_checksum, checksum) : 160980901aeaSGeorge Wilson dedup_checksum; 1610b24ab676SJeff Bonwick 161180901aeaSGeorge Wilson /* 161280901aeaSGeorge Wilson * Determine dedup setting. If we are in dmu_sync(), 161380901aeaSGeorge Wilson * we won't actually dedup now because that's all 161480901aeaSGeorge Wilson * done in syncing context; but we do want to use the 161580901aeaSGeorge Wilson * dedup checkum. If the checksum is not strong 161680901aeaSGeorge Wilson * enough to ensure unique signatures, force 161780901aeaSGeorge Wilson * dedup_verify. 161880901aeaSGeorge Wilson */ 161980901aeaSGeorge Wilson if (dedup_checksum != ZIO_CHECKSUM_OFF) { 162080901aeaSGeorge Wilson dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; 162180901aeaSGeorge Wilson if (!zio_checksum_table[checksum].ci_dedup) 162280901aeaSGeorge Wilson dedup_verify = B_TRUE; 162380901aeaSGeorge Wilson } 1624b24ab676SJeff Bonwick 162580901aeaSGeorge Wilson /* 162680901aeaSGeorge Wilson * Enable nopwrite if we have a cryptographically secure 162780901aeaSGeorge Wilson * checksum that has no known collisions (i.e. SHA-256) 162880901aeaSGeorge Wilson * and compression is enabled. We don't enable nopwrite if 162980901aeaSGeorge Wilson * dedup is enabled as the two features are mutually exclusive. 163080901aeaSGeorge Wilson */ 163180901aeaSGeorge Wilson nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup && 163280901aeaSGeorge Wilson compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); 1633b24ab676SJeff Bonwick } 1634b24ab676SJeff Bonwick 1635b24ab676SJeff Bonwick zp->zp_checksum = checksum; 1636b24ab676SJeff Bonwick zp->zp_compress = compress; 16370a586ceaSMark Shellenbaum zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; 1638b24ab676SJeff Bonwick zp->zp_level = level; 1639b24ab676SJeff Bonwick zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa)); 1640b24ab676SJeff Bonwick zp->zp_dedup = dedup; 1641b24ab676SJeff Bonwick zp->zp_dedup_verify = dedup && dedup_verify; 164280901aeaSGeorge Wilson zp->zp_nopwrite = nopwrite; 1643b24ab676SJeff Bonwick } 1644b24ab676SJeff Bonwick 1645fa9e4066Sahrens int 1646fa9e4066Sahrens dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) 1647fa9e4066Sahrens { 1648fa9e4066Sahrens dnode_t *dn; 1649fa9e4066Sahrens int i, err; 1650fa9e4066Sahrens 1651503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 1652ea8dc4b6Seschrock if (err) 1653ea8dc4b6Seschrock return (err); 1654fa9e4066Sahrens /* 1655fa9e4066Sahrens * Sync any current changes before 1656fa9e4066Sahrens * we go trundling through the block pointers. 1657fa9e4066Sahrens */ 1658fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 1659c543ec06Sahrens if (list_link_active(&dn->dn_dirty_link[i])) 1660fa9e4066Sahrens break; 1661fa9e4066Sahrens } 1662fa9e4066Sahrens if (i != TXG_SIZE) { 1663fa9e4066Sahrens dnode_rele(dn, FTAG); 1664fa9e4066Sahrens txg_wait_synced(dmu_objset_pool(os), 0); 1665503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 1666ea8dc4b6Seschrock if (err) 1667ea8dc4b6Seschrock return (err); 1668fa9e4066Sahrens } 1669fa9e4066Sahrens 1670cdb0ab79Smaybee err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); 1671fa9e4066Sahrens dnode_rele(dn, FTAG); 1672fa9e4066Sahrens 1673fa9e4066Sahrens return (err); 1674fa9e4066Sahrens } 1675fa9e4066Sahrens 1676fa9e4066Sahrens void 1677fa9e4066Sahrens dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) 1678fa9e4066Sahrens { 1679b24ab676SJeff Bonwick dnode_phys_t *dnp; 1680b24ab676SJeff Bonwick 1681fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 1682fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 1683fa9e4066Sahrens 1684b24ab676SJeff Bonwick dnp = dn->dn_phys; 1685b24ab676SJeff Bonwick 1686fa9e4066Sahrens doi->doi_data_block_size = dn->dn_datablksz; 1687fa9e4066Sahrens doi->doi_metadata_block_size = dn->dn_indblkshift ? 1688fa9e4066Sahrens 1ULL << dn->dn_indblkshift : 0; 1689b24ab676SJeff Bonwick doi->doi_type = dn->dn_type; 1690b24ab676SJeff Bonwick doi->doi_bonus_type = dn->dn_bonustype; 1691b24ab676SJeff Bonwick doi->doi_bonus_size = dn->dn_bonuslen; 1692fa9e4066Sahrens doi->doi_indirection = dn->dn_nlevels; 1693fa9e4066Sahrens doi->doi_checksum = dn->dn_checksum; 1694fa9e4066Sahrens doi->doi_compress = dn->dn_compress; 1695b24ab676SJeff Bonwick doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; 1696d0475637SMatthew Ahrens doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; 1697b24ab676SJeff Bonwick doi->doi_fill_count = 0; 1698b24ab676SJeff Bonwick for (int i = 0; i < dnp->dn_nblkptr; i++) 1699b24ab676SJeff Bonwick doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill; 1700fa9e4066Sahrens 1701fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 1702fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 1703fa9e4066Sahrens } 1704fa9e4066Sahrens 1705fa9e4066Sahrens /* 1706fa9e4066Sahrens * Get information on a DMU object. 1707fa9e4066Sahrens * If doi is NULL, just indicates whether the object exists. 1708fa9e4066Sahrens */ 1709fa9e4066Sahrens int 1710fa9e4066Sahrens dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) 1711fa9e4066Sahrens { 1712ea8dc4b6Seschrock dnode_t *dn; 1713503ad85cSMatthew Ahrens int err = dnode_hold(os, object, FTAG, &dn); 1714fa9e4066Sahrens 1715ea8dc4b6Seschrock if (err) 1716ea8dc4b6Seschrock return (err); 1717fa9e4066Sahrens 1718fa9e4066Sahrens if (doi != NULL) 1719fa9e4066Sahrens dmu_object_info_from_dnode(dn, doi); 1720fa9e4066Sahrens 1721fa9e4066Sahrens dnode_rele(dn, FTAG); 1722fa9e4066Sahrens return (0); 1723fa9e4066Sahrens } 1724fa9e4066Sahrens 1725fa9e4066Sahrens /* 1726fa9e4066Sahrens * As above, but faster; can be used when you have a held dbuf in hand. 1727fa9e4066Sahrens */ 1728fa9e4066Sahrens void 1729744947dcSTom Erickson dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) 1730fa9e4066Sahrens { 1731744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 1732744947dcSTom Erickson 1733744947dcSTom Erickson DB_DNODE_ENTER(db); 1734744947dcSTom Erickson dmu_object_info_from_dnode(DB_DNODE(db), doi); 1735744947dcSTom Erickson DB_DNODE_EXIT(db); 1736fa9e4066Sahrens } 1737fa9e4066Sahrens 1738fa9e4066Sahrens /* 1739fa9e4066Sahrens * Faster still when you only care about the size. 1740fa9e4066Sahrens * This is specifically optimized for zfs_getattr(). 1741fa9e4066Sahrens */ 1742fa9e4066Sahrens void 1743744947dcSTom Erickson dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, 1744744947dcSTom Erickson u_longlong_t *nblk512) 1745fa9e4066Sahrens { 1746744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 1747744947dcSTom Erickson dnode_t *dn; 1748744947dcSTom Erickson 1749744947dcSTom Erickson DB_DNODE_ENTER(db); 1750744947dcSTom Erickson dn = DB_DNODE(db); 1751fa9e4066Sahrens 1752fa9e4066Sahrens *blksize = dn->dn_datablksz; 175399653d4eSeschrock /* add 1 for dnode space */ 175499653d4eSeschrock *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> 175599653d4eSeschrock SPA_MINBLOCKSHIFT) + 1; 1756744947dcSTom Erickson DB_DNODE_EXIT(db); 1757fa9e4066Sahrens } 1758fa9e4066Sahrens 1759fa9e4066Sahrens void 1760fa9e4066Sahrens byteswap_uint64_array(void *vbuf, size_t size) 1761fa9e4066Sahrens { 1762fa9e4066Sahrens uint64_t *buf = vbuf; 1763fa9e4066Sahrens size_t count = size >> 3; 1764fa9e4066Sahrens int i; 1765fa9e4066Sahrens 1766fa9e4066Sahrens ASSERT((size & 7) == 0); 1767fa9e4066Sahrens 1768fa9e4066Sahrens for (i = 0; i < count; i++) 1769fa9e4066Sahrens buf[i] = BSWAP_64(buf[i]); 1770fa9e4066Sahrens } 1771fa9e4066Sahrens 1772fa9e4066Sahrens void 1773fa9e4066Sahrens byteswap_uint32_array(void *vbuf, size_t size) 1774fa9e4066Sahrens { 1775fa9e4066Sahrens uint32_t *buf = vbuf; 1776fa9e4066Sahrens size_t count = size >> 2; 1777fa9e4066Sahrens int i; 1778fa9e4066Sahrens 1779fa9e4066Sahrens ASSERT((size & 3) == 0); 1780fa9e4066Sahrens 1781fa9e4066Sahrens for (i = 0; i < count; i++) 1782fa9e4066Sahrens buf[i] = BSWAP_32(buf[i]); 1783fa9e4066Sahrens } 1784fa9e4066Sahrens 1785fa9e4066Sahrens void 1786fa9e4066Sahrens byteswap_uint16_array(void *vbuf, size_t size) 1787fa9e4066Sahrens { 1788fa9e4066Sahrens uint16_t *buf = vbuf; 1789fa9e4066Sahrens size_t count = size >> 1; 1790fa9e4066Sahrens int i; 1791fa9e4066Sahrens 1792fa9e4066Sahrens ASSERT((size & 1) == 0); 1793fa9e4066Sahrens 1794fa9e4066Sahrens for (i = 0; i < count; i++) 1795fa9e4066Sahrens buf[i] = BSWAP_16(buf[i]); 1796fa9e4066Sahrens } 1797fa9e4066Sahrens 1798fa9e4066Sahrens /* ARGSUSED */ 1799fa9e4066Sahrens void 1800fa9e4066Sahrens byteswap_uint8_array(void *vbuf, size_t size) 1801fa9e4066Sahrens { 1802fa9e4066Sahrens } 1803fa9e4066Sahrens 1804fa9e4066Sahrens void 1805fa9e4066Sahrens dmu_init(void) 1806fa9e4066Sahrens { 18073f9d6ad7SLin Ling zfs_dbgmsg_init(); 1808744947dcSTom Erickson sa_cache_init(); 1809744947dcSTom Erickson xuio_stat_init(); 1810744947dcSTom Erickson dmu_objset_init(); 1811fa9e4066Sahrens dnode_init(); 1812744947dcSTom Erickson dbuf_init(); 18137cbf8b43SRich Morris zfetch_init(); 1814fa94a07fSbrendan l2arc_init(); 1815ce636f8bSMatthew Ahrens arc_init(); 1816fa9e4066Sahrens } 1817fa9e4066Sahrens 1818fa9e4066Sahrens void 1819fa9e4066Sahrens dmu_fini(void) 1820fa9e4066Sahrens { 18213e30c24aSWill Andrews arc_fini(); /* arc depends on l2arc, so arc must go first */ 1822ce636f8bSMatthew Ahrens l2arc_fini(); 18237cbf8b43SRich Morris zfetch_fini(); 1824fa9e4066Sahrens dbuf_fini(); 1825744947dcSTom Erickson dnode_fini(); 1826744947dcSTom Erickson dmu_objset_fini(); 1827c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_fini(); 18280a586ceaSMark Shellenbaum sa_cache_fini(); 18293f9d6ad7SLin Ling zfs_dbgmsg_fini(); 1830fa9e4066Sahrens } 1831