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. 23ad135b5dSChristopher Siden * Copyright (c) 2012 by Delphix. All rights reserved. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens #include <sys/dmu.h> 27fa9e4066Sahrens #include <sys/dmu_impl.h> 28fa9e4066Sahrens #include <sys/dmu_tx.h> 29fa9e4066Sahrens #include <sys/dbuf.h> 30fa9e4066Sahrens #include <sys/dnode.h> 31fa9e4066Sahrens #include <sys/zfs_context.h> 32fa9e4066Sahrens #include <sys/dmu_objset.h> 33fa9e4066Sahrens #include <sys/dmu_traverse.h> 34fa9e4066Sahrens #include <sys/dsl_dataset.h> 35fa9e4066Sahrens #include <sys/dsl_dir.h> 36fa9e4066Sahrens #include <sys/dsl_pool.h> 371d452cf5Sahrens #include <sys/dsl_synctask.h> 38a2eea2e1Sahrens #include <sys/dsl_prop.h> 39fa9e4066Sahrens #include <sys/dmu_zfetch.h> 40fa9e4066Sahrens #include <sys/zfs_ioctl.h> 41fa9e4066Sahrens #include <sys/zap.h> 42ea8dc4b6Seschrock #include <sys/zio_checksum.h> 4380901aeaSGeorge Wilson #include <sys/zio_compress.h> 440a586ceaSMark Shellenbaum #include <sys/sa.h> 4544eda4d7Smaybee #ifdef _KERNEL 4644eda4d7Smaybee #include <sys/vmsystm.h> 470fab61baSJonathan W Adams #include <sys/zfs_znode.h> 4844eda4d7Smaybee #endif 49fa9e4066Sahrens 5080901aeaSGeorge Wilson /* 5180901aeaSGeorge Wilson * Enable/disable nopwrite feature. 5280901aeaSGeorge Wilson */ 5380901aeaSGeorge Wilson int zfs_nopwrite_enabled = 1; 5480901aeaSGeorge Wilson 55fa9e4066Sahrens const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { 56ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "unallocated" }, 57ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "object directory" }, 58ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "object array" }, 59ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "packed nvlist" }, 60ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" }, 61ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "bpobj" }, 62ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "bpobj header" }, 63ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "SPA space map header" }, 64ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "SPA space map" }, 65ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" }, 66ad135b5dSChristopher Siden { DMU_BSWAP_DNODE, TRUE, "DMU dnode" }, 67ad135b5dSChristopher Siden { DMU_BSWAP_OBJSET, TRUE, "DMU objset" }, 68ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "DSL directory" }, 69ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"}, 70ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" }, 71ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL props" }, 72ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "DSL dataset" }, 73ad135b5dSChristopher Siden { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" }, 74ad135b5dSChristopher Siden { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" }, 75ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" }, 76ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS directory" }, 77ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS master node" }, 78ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" }, 79ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "zvol object" }, 80ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "zvol prop" }, 81ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "other uint8[]" }, 82ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, FALSE, "other uint64[]" }, 83ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "other ZAP" }, 84ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "persistent error log" }, 85ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "SPA history" }, 86ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" }, 87ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "Pool properties" }, 88ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL permissions" }, 89ad135b5dSChristopher Siden { DMU_BSWAP_ACL, TRUE, "ZFS ACL" }, 90ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" }, 91ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "FUID table" }, 92ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "FUID table size" }, 93ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"}, 94ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "scan work queue" }, 95ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" }, 96ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" }, 97ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"}, 98ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" }, 99ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DDT statistics" }, 100ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "System attributes" }, 101ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "SA master node" }, 102ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "SA attr registration" }, 103ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" }, 104ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "scan translations" }, 105ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "deduplicated block" }, 106ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" }, 107ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" }, 108ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" }, 109ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" } 110ad135b5dSChristopher Siden }; 111ad135b5dSChristopher Siden 112ad135b5dSChristopher Siden const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { 113ad135b5dSChristopher Siden { byteswap_uint8_array, "uint8" }, 114ad135b5dSChristopher Siden { byteswap_uint16_array, "uint16" }, 115ad135b5dSChristopher Siden { byteswap_uint32_array, "uint32" }, 116ad135b5dSChristopher Siden { byteswap_uint64_array, "uint64" }, 117ad135b5dSChristopher Siden { zap_byteswap, "zap" }, 118ad135b5dSChristopher Siden { dnode_buf_byteswap, "dnode" }, 119ad135b5dSChristopher Siden { dmu_objset_byteswap, "objset" }, 120ad135b5dSChristopher Siden { zfs_znode_byteswap, "znode" }, 121ad135b5dSChristopher Siden { zfs_oldacl_byteswap, "oldacl" }, 122ad135b5dSChristopher Siden { zfs_acl_byteswap, "acl" } 1233f9d6ad7SLin Ling }; 124fa9e4066Sahrens 125fa9e4066Sahrens int 126ea8dc4b6Seschrock dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 12747cb52daSJeff Bonwick void *tag, dmu_buf_t **dbp, int flags) 128fa9e4066Sahrens { 129fa9e4066Sahrens dnode_t *dn; 130fa9e4066Sahrens uint64_t blkid; 131fa9e4066Sahrens dmu_buf_impl_t *db; 132ea8dc4b6Seschrock int err; 13347cb52daSJeff Bonwick int db_flags = DB_RF_CANFAIL; 13447cb52daSJeff Bonwick 13547cb52daSJeff Bonwick if (flags & DMU_READ_NO_PREFETCH) 13647cb52daSJeff Bonwick db_flags |= DB_RF_NOPREFETCH; 137fa9e4066Sahrens 138503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 139ea8dc4b6Seschrock if (err) 140ea8dc4b6Seschrock return (err); 141fa9e4066Sahrens blkid = dbuf_whichblock(dn, offset); 142fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 143ea8dc4b6Seschrock db = dbuf_hold(dn, blkid, tag); 144fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 145ea8dc4b6Seschrock if (db == NULL) { 146ea8dc4b6Seschrock err = EIO; 147ea8dc4b6Seschrock } else { 14847cb52daSJeff Bonwick err = dbuf_read(db, NULL, db_flags); 149ea8dc4b6Seschrock if (err) { 150ea8dc4b6Seschrock dbuf_rele(db, tag); 151ea8dc4b6Seschrock db = NULL; 152ea8dc4b6Seschrock } 153fa9e4066Sahrens } 154fa9e4066Sahrens 155ea8dc4b6Seschrock dnode_rele(dn, FTAG); 156744947dcSTom Erickson *dbp = &db->db; /* NULL db plus first field offset is NULL */ 157ea8dc4b6Seschrock return (err); 158fa9e4066Sahrens } 159fa9e4066Sahrens 160fa9e4066Sahrens int 161fa9e4066Sahrens dmu_bonus_max(void) 162fa9e4066Sahrens { 163fa9e4066Sahrens return (DN_MAX_BONUSLEN); 164fa9e4066Sahrens } 165fa9e4066Sahrens 1661934e92fSmaybee int 167744947dcSTom Erickson dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) 1681934e92fSmaybee { 169744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 170744947dcSTom Erickson dnode_t *dn; 171744947dcSTom Erickson int error; 1721934e92fSmaybee 173744947dcSTom Erickson DB_DNODE_ENTER(db); 174744947dcSTom Erickson dn = DB_DNODE(db); 175744947dcSTom Erickson 176744947dcSTom Erickson if (dn->dn_bonus != db) { 177744947dcSTom Erickson error = EINVAL; 178744947dcSTom Erickson } else if (newsize < 0 || newsize > db_fake->db_size) { 179744947dcSTom Erickson error = EINVAL; 180744947dcSTom Erickson } else { 1811934e92fSmaybee dnode_setbonuslen(dn, newsize, tx); 182744947dcSTom Erickson error = 0; 183744947dcSTom Erickson } 184744947dcSTom Erickson 185744947dcSTom Erickson DB_DNODE_EXIT(db); 186744947dcSTom Erickson return (error); 1871934e92fSmaybee } 1881934e92fSmaybee 1890a586ceaSMark Shellenbaum int 190744947dcSTom Erickson dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) 1910a586ceaSMark Shellenbaum { 192744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 193744947dcSTom Erickson dnode_t *dn; 194744947dcSTom Erickson int error; 1950a586ceaSMark Shellenbaum 196744947dcSTom Erickson DB_DNODE_ENTER(db); 197744947dcSTom Erickson dn = DB_DNODE(db); 1980a586ceaSMark Shellenbaum 199ad135b5dSChristopher Siden if (!DMU_OT_IS_VALID(type)) { 200744947dcSTom Erickson error = EINVAL; 201744947dcSTom Erickson } else if (dn->dn_bonus != db) { 202744947dcSTom Erickson error = EINVAL; 203744947dcSTom Erickson } else { 2040a586ceaSMark Shellenbaum dnode_setbonus_type(dn, type, tx); 205744947dcSTom Erickson error = 0; 206744947dcSTom Erickson } 207744947dcSTom Erickson 208744947dcSTom Erickson DB_DNODE_EXIT(db); 209744947dcSTom Erickson return (error); 210744947dcSTom Erickson } 211744947dcSTom Erickson 212744947dcSTom Erickson dmu_object_type_t 213744947dcSTom Erickson dmu_get_bonustype(dmu_buf_t *db_fake) 214744947dcSTom Erickson { 215744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 216744947dcSTom Erickson dnode_t *dn; 217744947dcSTom Erickson dmu_object_type_t type; 218744947dcSTom Erickson 219744947dcSTom Erickson DB_DNODE_ENTER(db); 220744947dcSTom Erickson dn = DB_DNODE(db); 221744947dcSTom Erickson type = dn->dn_bonustype; 222744947dcSTom Erickson DB_DNODE_EXIT(db); 223744947dcSTom Erickson 224744947dcSTom Erickson return (type); 2250a586ceaSMark Shellenbaum } 2260a586ceaSMark Shellenbaum 2270a586ceaSMark Shellenbaum int 2280a586ceaSMark Shellenbaum dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) 2290a586ceaSMark Shellenbaum { 2300a586ceaSMark Shellenbaum dnode_t *dn; 2310a586ceaSMark Shellenbaum int error; 2320a586ceaSMark Shellenbaum 2330a586ceaSMark Shellenbaum error = dnode_hold(os, object, FTAG, &dn); 2340a586ceaSMark Shellenbaum dbuf_rm_spill(dn, tx); 23506e0070dSMark Shellenbaum rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 23606e0070dSMark Shellenbaum dnode_rm_spill(dn, tx); 23706e0070dSMark Shellenbaum rw_exit(&dn->dn_struct_rwlock); 2380a586ceaSMark Shellenbaum dnode_rele(dn, FTAG); 2390a586ceaSMark Shellenbaum return (error); 2400a586ceaSMark Shellenbaum } 2410a586ceaSMark Shellenbaum 242fa9e4066Sahrens /* 243ea8dc4b6Seschrock * returns ENOENT, EIO, or 0. 244fa9e4066Sahrens */ 245ea8dc4b6Seschrock int 246ea8dc4b6Seschrock dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) 247fa9e4066Sahrens { 248ea8dc4b6Seschrock dnode_t *dn; 249fa9e4066Sahrens dmu_buf_impl_t *db; 2501934e92fSmaybee int error; 251fa9e4066Sahrens 252503ad85cSMatthew Ahrens error = dnode_hold(os, object, FTAG, &dn); 2531934e92fSmaybee if (error) 2541934e92fSmaybee return (error); 255fa9e4066Sahrens 256fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 257ea8dc4b6Seschrock if (dn->dn_bonus == NULL) { 258fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 259ea8dc4b6Seschrock rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 260ea8dc4b6Seschrock if (dn->dn_bonus == NULL) 2611934e92fSmaybee dbuf_create_bonus(dn); 262fa9e4066Sahrens } 263ea8dc4b6Seschrock db = dn->dn_bonus; 2641934e92fSmaybee 2651934e92fSmaybee /* as long as the bonus buf is held, the dnode will be held */ 266744947dcSTom Erickson if (refcount_add(&db->db_holds, tag) == 1) { 2671934e92fSmaybee VERIFY(dnode_add_ref(dn, db)); 268744947dcSTom Erickson (void) atomic_inc_32_nv(&dn->dn_dbufs_count); 269744947dcSTom Erickson } 270744947dcSTom Erickson 271744947dcSTom Erickson /* 272744947dcSTom Erickson * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's 273744947dcSTom Erickson * hold and incrementing the dbuf count to ensure that dnode_move() sees 274744947dcSTom Erickson * a dnode hold for every dbuf. 275744947dcSTom Erickson */ 276744947dcSTom Erickson rw_exit(&dn->dn_struct_rwlock); 2771934e92fSmaybee 278fa9e4066Sahrens dnode_rele(dn, FTAG); 279ea8dc4b6Seschrock 28047cb52daSJeff Bonwick VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); 281ea8dc4b6Seschrock 282ea8dc4b6Seschrock *dbp = &db->db; 283ea8dc4b6Seschrock return (0); 284fa9e4066Sahrens } 285fa9e4066Sahrens 28613506d1eSmaybee /* 2870a586ceaSMark Shellenbaum * returns ENOENT, EIO, or 0. 2880a586ceaSMark Shellenbaum * 2890a586ceaSMark Shellenbaum * This interface will allocate a blank spill dbuf when a spill blk 2900a586ceaSMark Shellenbaum * doesn't already exist on the dnode. 2910a586ceaSMark Shellenbaum * 2920a586ceaSMark Shellenbaum * if you only want to find an already existing spill db, then 2930a586ceaSMark Shellenbaum * dmu_spill_hold_existing() should be used. 2940a586ceaSMark Shellenbaum */ 2950a586ceaSMark Shellenbaum int 2960a586ceaSMark Shellenbaum dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) 2970a586ceaSMark Shellenbaum { 2980a586ceaSMark Shellenbaum dmu_buf_impl_t *db = NULL; 2990a586ceaSMark Shellenbaum int err; 3000a586ceaSMark Shellenbaum 3010a586ceaSMark Shellenbaum if ((flags & DB_RF_HAVESTRUCT) == 0) 3020a586ceaSMark Shellenbaum rw_enter(&dn->dn_struct_rwlock, RW_READER); 3030a586ceaSMark Shellenbaum 3040a586ceaSMark Shellenbaum db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); 3050a586ceaSMark Shellenbaum 3060a586ceaSMark Shellenbaum if ((flags & DB_RF_HAVESTRUCT) == 0) 3070a586ceaSMark Shellenbaum rw_exit(&dn->dn_struct_rwlock); 3080a586ceaSMark Shellenbaum 3090a586ceaSMark Shellenbaum ASSERT(db != NULL); 3101d8ccc7bSMark Shellenbaum err = dbuf_read(db, NULL, flags); 3111d8ccc7bSMark Shellenbaum if (err == 0) 3120a586ceaSMark Shellenbaum *dbp = &db->db; 3131d8ccc7bSMark Shellenbaum else 3141d8ccc7bSMark Shellenbaum dbuf_rele(db, tag); 3150a586ceaSMark Shellenbaum return (err); 3160a586ceaSMark Shellenbaum } 3170a586ceaSMark Shellenbaum 3180a586ceaSMark Shellenbaum int 3190a586ceaSMark Shellenbaum dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 3200a586ceaSMark Shellenbaum { 321744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; 322744947dcSTom Erickson dnode_t *dn; 3230a586ceaSMark Shellenbaum int err; 3240a586ceaSMark Shellenbaum 325744947dcSTom Erickson DB_DNODE_ENTER(db); 326744947dcSTom Erickson dn = DB_DNODE(db); 327744947dcSTom Erickson 328744947dcSTom Erickson if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { 329744947dcSTom Erickson err = EINVAL; 330744947dcSTom Erickson } else { 3310a586ceaSMark Shellenbaum rw_enter(&dn->dn_struct_rwlock, RW_READER); 3320a586ceaSMark Shellenbaum 3330a586ceaSMark Shellenbaum if (!dn->dn_have_spill) { 334744947dcSTom Erickson err = ENOENT; 335744947dcSTom Erickson } else { 3361d8ccc7bSMark Shellenbaum err = dmu_spill_hold_by_dnode(dn, 3371d8ccc7bSMark Shellenbaum DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); 338744947dcSTom Erickson } 339744947dcSTom Erickson 3400a586ceaSMark Shellenbaum rw_exit(&dn->dn_struct_rwlock); 341744947dcSTom Erickson } 342744947dcSTom Erickson 343744947dcSTom Erickson DB_DNODE_EXIT(db); 3440a586ceaSMark Shellenbaum return (err); 3450a586ceaSMark Shellenbaum } 3460a586ceaSMark Shellenbaum 3470a586ceaSMark Shellenbaum int 3480a586ceaSMark Shellenbaum dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 3490a586ceaSMark Shellenbaum { 350744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; 351744947dcSTom Erickson dnode_t *dn; 352744947dcSTom Erickson int err; 353744947dcSTom Erickson 354744947dcSTom Erickson DB_DNODE_ENTER(db); 355744947dcSTom Erickson dn = DB_DNODE(db); 356744947dcSTom Erickson err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); 357744947dcSTom Erickson DB_DNODE_EXIT(db); 358744947dcSTom Erickson 359744947dcSTom Erickson return (err); 3600a586ceaSMark Shellenbaum } 3610a586ceaSMark Shellenbaum 3620a586ceaSMark Shellenbaum /* 36313506d1eSmaybee * Note: longer-term, we should modify all of the dmu_buf_*() interfaces 36413506d1eSmaybee * to take a held dnode rather than <os, object> -- the lookup is wasteful, 36513506d1eSmaybee * and can induce severe lock contention when writing to several files 36613506d1eSmaybee * whose dnodes are in the same block. 36713506d1eSmaybee */ 36813506d1eSmaybee static int 3697bfdf011SNeil Perrin dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, 3707bfdf011SNeil Perrin int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) 371fa9e4066Sahrens { 37205715f94SMark Maybee dsl_pool_t *dp = NULL; 373fa9e4066Sahrens dmu_buf_t **dbp; 374fa9e4066Sahrens uint64_t blkid, nblks, i; 3757bfdf011SNeil Perrin uint32_t dbuf_flags; 376ea8dc4b6Seschrock int err; 377ea8dc4b6Seschrock zio_t *zio; 37805715f94SMark Maybee hrtime_t start; 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) - 390fa9e4066Sahrens 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); 4000125049cSahrens return (EIO); 4010125049cSahrens } 402fa9e4066Sahrens nblks = 1; 403fa9e4066Sahrens } 404ea8dc4b6Seschrock dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); 405fa9e4066Sahrens 40605715f94SMark Maybee if (dn->dn_objset->os_dsl_dataset) 40705715f94SMark Maybee dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool; 40805715f94SMark Maybee start = gethrtime(); 409e14bb325SJeff Bonwick zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); 410fa9e4066Sahrens blkid = dbuf_whichblock(dn, offset); 411fa9e4066Sahrens for (i = 0; i < nblks; i++) { 412ea8dc4b6Seschrock dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); 413ea8dc4b6Seschrock if (db == NULL) { 414ea8dc4b6Seschrock rw_exit(&dn->dn_struct_rwlock); 415ea8dc4b6Seschrock dmu_buf_rele_array(dbp, nblks, tag); 416ea8dc4b6Seschrock zio_nowait(zio); 417ea8dc4b6Seschrock return (EIO); 418ea8dc4b6Seschrock } 419ea8dc4b6Seschrock /* initiate async i/o */ 42013506d1eSmaybee if (read) { 4217bfdf011SNeil Perrin (void) dbuf_read(db, zio, dbuf_flags); 422ea8dc4b6Seschrock } 423ea8dc4b6Seschrock dbp[i] = &db->db; 424fa9e4066Sahrens } 425fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 426fa9e4066Sahrens 427ea8dc4b6Seschrock /* wait for async i/o */ 428ea8dc4b6Seschrock err = zio_wait(zio); 42905715f94SMark Maybee /* track read overhead when we are in sync context */ 43005715f94SMark Maybee if (dp && dsl_pool_sync_context(dp)) 43105715f94SMark Maybee dp->dp_read_overhead += gethrtime() - start; 432ea8dc4b6Seschrock if (err) { 433ea8dc4b6Seschrock dmu_buf_rele_array(dbp, nblks, tag); 434ea8dc4b6Seschrock return (err); 435ea8dc4b6Seschrock } 436ea8dc4b6Seschrock 437ea8dc4b6Seschrock /* wait for other io to complete */ 438ea8dc4b6Seschrock if (read) { 439ea8dc4b6Seschrock for (i = 0; i < nblks; i++) { 440ea8dc4b6Seschrock dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; 441ea8dc4b6Seschrock mutex_enter(&db->db_mtx); 442ea8dc4b6Seschrock while (db->db_state == DB_READ || 443ea8dc4b6Seschrock db->db_state == DB_FILL) 444ea8dc4b6Seschrock cv_wait(&db->db_changed, &db->db_mtx); 445ea8dc4b6Seschrock if (db->db_state == DB_UNCACHED) 446ea8dc4b6Seschrock err = EIO; 447ea8dc4b6Seschrock mutex_exit(&db->db_mtx); 448ea8dc4b6Seschrock if (err) { 449ea8dc4b6Seschrock dmu_buf_rele_array(dbp, nblks, tag); 450ea8dc4b6Seschrock return (err); 451ea8dc4b6Seschrock } 452ea8dc4b6Seschrock } 453ea8dc4b6Seschrock } 454ea8dc4b6Seschrock 455ea8dc4b6Seschrock *numbufsp = nblks; 456ea8dc4b6Seschrock *dbpp = dbp; 457ea8dc4b6Seschrock return (0); 458fa9e4066Sahrens } 459fa9e4066Sahrens 460a2eea2e1Sahrens static int 46113506d1eSmaybee dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, 46213506d1eSmaybee uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 46313506d1eSmaybee { 46413506d1eSmaybee dnode_t *dn; 46513506d1eSmaybee int err; 46613506d1eSmaybee 467503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 46813506d1eSmaybee if (err) 46913506d1eSmaybee return (err); 47013506d1eSmaybee 47113506d1eSmaybee err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 4727bfdf011SNeil Perrin numbufsp, dbpp, DMU_READ_PREFETCH); 47313506d1eSmaybee 47413506d1eSmaybee dnode_rele(dn, FTAG); 47513506d1eSmaybee 47613506d1eSmaybee return (err); 47713506d1eSmaybee } 47813506d1eSmaybee 47913506d1eSmaybee int 480744947dcSTom Erickson dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, 48113506d1eSmaybee uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 48213506d1eSmaybee { 483744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 484744947dcSTom Erickson dnode_t *dn; 48513506d1eSmaybee int err; 48613506d1eSmaybee 487744947dcSTom Erickson DB_DNODE_ENTER(db); 488744947dcSTom Erickson dn = DB_DNODE(db); 48913506d1eSmaybee err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 4907bfdf011SNeil Perrin numbufsp, dbpp, DMU_READ_PREFETCH); 491744947dcSTom Erickson DB_DNODE_EXIT(db); 49213506d1eSmaybee 49313506d1eSmaybee return (err); 49413506d1eSmaybee } 49513506d1eSmaybee 496fa9e4066Sahrens void 497ea8dc4b6Seschrock dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) 498fa9e4066Sahrens { 499fa9e4066Sahrens int i; 500fa9e4066Sahrens dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; 501fa9e4066Sahrens 502fa9e4066Sahrens if (numbufs == 0) 503fa9e4066Sahrens return; 504fa9e4066Sahrens 505ea8dc4b6Seschrock for (i = 0; i < numbufs; i++) { 506ea8dc4b6Seschrock if (dbp[i]) 507ea8dc4b6Seschrock dbuf_rele(dbp[i], tag); 508ea8dc4b6Seschrock } 509fa9e4066Sahrens 510fa9e4066Sahrens kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); 511fa9e4066Sahrens } 512fa9e4066Sahrens 513fa9e4066Sahrens void 514fa9e4066Sahrens dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) 515fa9e4066Sahrens { 516fa9e4066Sahrens dnode_t *dn; 517fa9e4066Sahrens uint64_t blkid; 518ea8dc4b6Seschrock int nblks, i, err; 519fa9e4066Sahrens 520416e0cd8Sek110237 if (zfs_prefetch_disable) 521416e0cd8Sek110237 return; 522416e0cd8Sek110237 523fa9e4066Sahrens if (len == 0) { /* they're interested in the bonus buffer */ 524744947dcSTom Erickson dn = DMU_META_DNODE(os); 525fa9e4066Sahrens 526fa9e4066Sahrens if (object == 0 || object >= DN_MAX_OBJECT) 527fa9e4066Sahrens return; 528fa9e4066Sahrens 529fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 530fa9e4066Sahrens blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); 531fa9e4066Sahrens dbuf_prefetch(dn, blkid); 532fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 533fa9e4066Sahrens return; 534fa9e4066Sahrens } 535fa9e4066Sahrens 536fa9e4066Sahrens /* 537fa9e4066Sahrens * XXX - Note, if the dnode for the requested object is not 538fa9e4066Sahrens * already cached, we will do a *synchronous* read in the 539fa9e4066Sahrens * dnode_hold() call. The same is true for any indirects. 540fa9e4066Sahrens */ 541503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 542ea8dc4b6Seschrock if (err != 0) 543fa9e4066Sahrens return; 544fa9e4066Sahrens 545fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 546fa9e4066Sahrens if (dn->dn_datablkshift) { 547fa9e4066Sahrens int blkshift = dn->dn_datablkshift; 548fa9e4066Sahrens nblks = (P2ROUNDUP(offset+len, 1<<blkshift) - 549fa9e4066Sahrens P2ALIGN(offset, 1<<blkshift)) >> blkshift; 550fa9e4066Sahrens } else { 551fa9e4066Sahrens nblks = (offset < dn->dn_datablksz); 552fa9e4066Sahrens } 553fa9e4066Sahrens 554fa9e4066Sahrens if (nblks != 0) { 555fa9e4066Sahrens blkid = dbuf_whichblock(dn, offset); 556fa9e4066Sahrens for (i = 0; i < nblks; i++) 557fa9e4066Sahrens dbuf_prefetch(dn, blkid+i); 558fa9e4066Sahrens } 559fa9e4066Sahrens 560fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 561fa9e4066Sahrens 562fa9e4066Sahrens dnode_rele(dn, FTAG); 563fa9e4066Sahrens } 564fa9e4066Sahrens 56576256205SMark Maybee /* 56676256205SMark Maybee * Get the next "chunk" of file data to free. We traverse the file from 56776256205SMark Maybee * the end so that the file gets shorter over time (if we crashes in the 56876256205SMark Maybee * middle, this will leave us in a better state). We find allocated file 56976256205SMark Maybee * data by simply searching the allocated level 1 indirects. 57076256205SMark Maybee */ 571cdb0ab79Smaybee static int 57276256205SMark Maybee get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit) 573cdb0ab79Smaybee { 57476256205SMark Maybee uint64_t len = *start - limit; 57576256205SMark Maybee uint64_t blkcnt = 0; 57676256205SMark Maybee uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1)); 57776256205SMark Maybee uint64_t iblkrange = 5781c8564a7SMark Maybee dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); 579cdb0ab79Smaybee 58076256205SMark Maybee ASSERT(limit <= *start); 581cdb0ab79Smaybee 58276256205SMark Maybee if (len <= iblkrange * maxblks) { 58376256205SMark Maybee *start = limit; 584cdb0ab79Smaybee return (0); 585cdb0ab79Smaybee } 58676256205SMark Maybee ASSERT(ISP2(iblkrange)); 587cdb0ab79Smaybee 58876256205SMark Maybee while (*start > limit && blkcnt < maxblks) { 5891c8564a7SMark Maybee int err; 590cdb0ab79Smaybee 59176256205SMark Maybee /* find next allocated L1 indirect */ 592cdb0ab79Smaybee err = dnode_next_offset(dn, 59376256205SMark Maybee DNODE_FIND_BACKWARDS, start, 2, 1, 0); 594cdb0ab79Smaybee 59576256205SMark Maybee /* if there are no more, then we are done */ 59676256205SMark Maybee if (err == ESRCH) { 59776256205SMark Maybee *start = limit; 598cdb0ab79Smaybee return (0); 59976256205SMark Maybee } else if (err) { 600cdb0ab79Smaybee return (err); 60176256205SMark Maybee } 60276256205SMark Maybee blkcnt += 1; 603cdb0ab79Smaybee 60476256205SMark Maybee /* reset offset to end of "next" block back */ 60576256205SMark Maybee *start = P2ALIGN(*start, iblkrange); 60676256205SMark Maybee if (*start <= limit) 60776256205SMark Maybee *start = limit; 60876256205SMark Maybee else 60976256205SMark Maybee *start -= 1; 610cdb0ab79Smaybee } 611cdb0ab79Smaybee return (0); 612cdb0ab79Smaybee } 613cdb0ab79Smaybee 614cdb0ab79Smaybee static int 615cdb0ab79Smaybee dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, 616cdb0ab79Smaybee uint64_t length, boolean_t free_dnode) 617cdb0ab79Smaybee { 618cdb0ab79Smaybee dmu_tx_t *tx; 619cdb0ab79Smaybee uint64_t object_size, start, end, len; 620cdb0ab79Smaybee boolean_t trunc = (length == DMU_OBJECT_END); 621cdb0ab79Smaybee int align, err; 622cdb0ab79Smaybee 623cdb0ab79Smaybee align = 1 << dn->dn_datablkshift; 624cdb0ab79Smaybee ASSERT(align > 0); 625cdb0ab79Smaybee object_size = align == 1 ? dn->dn_datablksz : 626cdb0ab79Smaybee (dn->dn_maxblkid + 1) << dn->dn_datablkshift; 627cdb0ab79Smaybee 62814843421SMatthew Ahrens end = offset + length; 62914843421SMatthew Ahrens if (trunc || end > object_size) 630cdb0ab79Smaybee end = object_size; 631cdb0ab79Smaybee if (end <= offset) 632cdb0ab79Smaybee return (0); 633cdb0ab79Smaybee length = end - offset; 634cdb0ab79Smaybee 635cdb0ab79Smaybee while (length) { 636cdb0ab79Smaybee start = end; 63714843421SMatthew Ahrens /* assert(offset <= start) */ 638cdb0ab79Smaybee err = get_next_chunk(dn, &start, offset); 639cdb0ab79Smaybee if (err) 640cdb0ab79Smaybee return (err); 641cdb0ab79Smaybee len = trunc ? DMU_OBJECT_END : end - start; 642cdb0ab79Smaybee 643cdb0ab79Smaybee tx = dmu_tx_create(os); 644cdb0ab79Smaybee dmu_tx_hold_free(tx, dn->dn_object, start, len); 645cdb0ab79Smaybee err = dmu_tx_assign(tx, TXG_WAIT); 646cdb0ab79Smaybee if (err) { 647cdb0ab79Smaybee dmu_tx_abort(tx); 648cdb0ab79Smaybee return (err); 649cdb0ab79Smaybee } 650cdb0ab79Smaybee 651cdb0ab79Smaybee dnode_free_range(dn, start, trunc ? -1 : len, tx); 652cdb0ab79Smaybee 6531c8564a7SMark Maybee if (start == 0 && free_dnode) { 6541c8564a7SMark Maybee ASSERT(trunc); 655cdb0ab79Smaybee dnode_free(dn, tx); 6561c8564a7SMark Maybee } 657cdb0ab79Smaybee 658cdb0ab79Smaybee length -= end - start; 659cdb0ab79Smaybee 660cdb0ab79Smaybee dmu_tx_commit(tx); 661cdb0ab79Smaybee end = start; 662cdb0ab79Smaybee } 663cdb0ab79Smaybee return (0); 664cdb0ab79Smaybee } 665cdb0ab79Smaybee 666cdb0ab79Smaybee int 667cdb0ab79Smaybee dmu_free_long_range(objset_t *os, uint64_t object, 668cdb0ab79Smaybee uint64_t offset, uint64_t length) 669cdb0ab79Smaybee { 670cdb0ab79Smaybee dnode_t *dn; 671cdb0ab79Smaybee int err; 672cdb0ab79Smaybee 673503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 674cdb0ab79Smaybee if (err != 0) 675cdb0ab79Smaybee return (err); 676cdb0ab79Smaybee err = dmu_free_long_range_impl(os, dn, offset, length, FALSE); 677cdb0ab79Smaybee dnode_rele(dn, FTAG); 678cdb0ab79Smaybee return (err); 679cdb0ab79Smaybee } 680cdb0ab79Smaybee 681cdb0ab79Smaybee int 682cdb0ab79Smaybee dmu_free_object(objset_t *os, uint64_t object) 683cdb0ab79Smaybee { 684cdb0ab79Smaybee dnode_t *dn; 685cdb0ab79Smaybee dmu_tx_t *tx; 686cdb0ab79Smaybee int err; 687cdb0ab79Smaybee 688503ad85cSMatthew Ahrens err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 689cdb0ab79Smaybee FTAG, &dn); 690cdb0ab79Smaybee if (err != 0) 691cdb0ab79Smaybee return (err); 692cdb0ab79Smaybee if (dn->dn_nlevels == 1) { 693cdb0ab79Smaybee tx = dmu_tx_create(os); 694cdb0ab79Smaybee dmu_tx_hold_bonus(tx, object); 695cdb0ab79Smaybee dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END); 696cdb0ab79Smaybee err = dmu_tx_assign(tx, TXG_WAIT); 697cdb0ab79Smaybee if (err == 0) { 698cdb0ab79Smaybee dnode_free_range(dn, 0, DMU_OBJECT_END, tx); 699cdb0ab79Smaybee dnode_free(dn, tx); 700cdb0ab79Smaybee dmu_tx_commit(tx); 701cdb0ab79Smaybee } else { 702cdb0ab79Smaybee dmu_tx_abort(tx); 703cdb0ab79Smaybee } 704cdb0ab79Smaybee } else { 705cdb0ab79Smaybee err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE); 706cdb0ab79Smaybee } 707cdb0ab79Smaybee dnode_rele(dn, FTAG); 708cdb0ab79Smaybee return (err); 709cdb0ab79Smaybee } 710cdb0ab79Smaybee 711ea8dc4b6Seschrock int 712fa9e4066Sahrens dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 713fa9e4066Sahrens uint64_t size, dmu_tx_t *tx) 714fa9e4066Sahrens { 715ea8dc4b6Seschrock dnode_t *dn; 716503ad85cSMatthew Ahrens int err = dnode_hold(os, object, FTAG, &dn); 717ea8dc4b6Seschrock if (err) 718ea8dc4b6Seschrock return (err); 719fa9e4066Sahrens ASSERT(offset < UINT64_MAX); 720fa9e4066Sahrens ASSERT(size == -1ULL || size <= UINT64_MAX - offset); 721fa9e4066Sahrens dnode_free_range(dn, offset, size, tx); 722fa9e4066Sahrens dnode_rele(dn, FTAG); 723ea8dc4b6Seschrock return (0); 724fa9e4066Sahrens } 725fa9e4066Sahrens 726ea8dc4b6Seschrock int 727ea8dc4b6Seschrock dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 7287bfdf011SNeil Perrin void *buf, uint32_t flags) 729fa9e4066Sahrens { 730fa9e4066Sahrens dnode_t *dn; 731fa9e4066Sahrens dmu_buf_t **dbp; 732c87b8fc5SMark J Musante int numbufs, err; 733fa9e4066Sahrens 734503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 735ea8dc4b6Seschrock if (err) 736ea8dc4b6Seschrock return (err); 737feb08c6bSbillm 738feb08c6bSbillm /* 739feb08c6bSbillm * Deal with odd block sizes, where there can't be data past the first 740feb08c6bSbillm * block. If we ever do the tail block optimization, we will need to 741feb08c6bSbillm * handle that here as well. 742feb08c6bSbillm */ 743c87b8fc5SMark J Musante if (dn->dn_maxblkid == 0) { 744fa9e4066Sahrens int newsz = offset > dn->dn_datablksz ? 0 : 745fa9e4066Sahrens MIN(size, dn->dn_datablksz - offset); 746fa9e4066Sahrens bzero((char *)buf + newsz, size - newsz); 747fa9e4066Sahrens size = newsz; 748fa9e4066Sahrens } 749fa9e4066Sahrens 750fa9e4066Sahrens while (size > 0) { 751fa9e4066Sahrens uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); 752c87b8fc5SMark J Musante int i; 753fa9e4066Sahrens 754fa9e4066Sahrens /* 755fa9e4066Sahrens * NB: we could do this block-at-a-time, but it's nice 756fa9e4066Sahrens * to be reading in parallel. 757fa9e4066Sahrens */ 758a2eea2e1Sahrens err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, 7597bfdf011SNeil Perrin TRUE, FTAG, &numbufs, &dbp, flags); 760ea8dc4b6Seschrock if (err) 7611934e92fSmaybee break; 762fa9e4066Sahrens 763fa9e4066Sahrens for (i = 0; i < numbufs; i++) { 764fa9e4066Sahrens int tocpy; 765fa9e4066Sahrens int bufoff; 766fa9e4066Sahrens dmu_buf_t *db = dbp[i]; 767fa9e4066Sahrens 768fa9e4066Sahrens ASSERT(size > 0); 769fa9e4066Sahrens 770fa9e4066Sahrens bufoff = offset - db->db_offset; 771fa9e4066Sahrens tocpy = (int)MIN(db->db_size - bufoff, size); 772fa9e4066Sahrens 773fa9e4066Sahrens bcopy((char *)db->db_data + bufoff, buf, tocpy); 774fa9e4066Sahrens 775fa9e4066Sahrens offset += tocpy; 776fa9e4066Sahrens size -= tocpy; 777fa9e4066Sahrens buf = (char *)buf + tocpy; 778fa9e4066Sahrens } 779ea8dc4b6Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG); 780fa9e4066Sahrens } 781a2eea2e1Sahrens dnode_rele(dn, FTAG); 7821934e92fSmaybee return (err); 783fa9e4066Sahrens } 784fa9e4066Sahrens 785fa9e4066Sahrens void 786fa9e4066Sahrens dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 787fa9e4066Sahrens const void *buf, dmu_tx_t *tx) 788fa9e4066Sahrens { 789fa9e4066Sahrens dmu_buf_t **dbp; 790fa9e4066Sahrens int numbufs, i; 791fa9e4066Sahrens 79213506d1eSmaybee if (size == 0) 79313506d1eSmaybee return; 79413506d1eSmaybee 795ea8dc4b6Seschrock VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 796ea8dc4b6Seschrock FALSE, FTAG, &numbufs, &dbp)); 797fa9e4066Sahrens 798fa9e4066Sahrens for (i = 0; i < numbufs; i++) { 799fa9e4066Sahrens int tocpy; 800fa9e4066Sahrens int bufoff; 801fa9e4066Sahrens dmu_buf_t *db = dbp[i]; 802fa9e4066Sahrens 803fa9e4066Sahrens ASSERT(size > 0); 804fa9e4066Sahrens 805fa9e4066Sahrens bufoff = offset - db->db_offset; 806fa9e4066Sahrens tocpy = (int)MIN(db->db_size - bufoff, size); 807fa9e4066Sahrens 808fa9e4066Sahrens ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 809fa9e4066Sahrens 810fa9e4066Sahrens if (tocpy == db->db_size) 811fa9e4066Sahrens dmu_buf_will_fill(db, tx); 812fa9e4066Sahrens else 813fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 814fa9e4066Sahrens 815fa9e4066Sahrens bcopy(buf, (char *)db->db_data + bufoff, tocpy); 816fa9e4066Sahrens 817fa9e4066Sahrens if (tocpy == db->db_size) 818fa9e4066Sahrens dmu_buf_fill_done(db, tx); 819fa9e4066Sahrens 820fa9e4066Sahrens offset += tocpy; 821fa9e4066Sahrens size -= tocpy; 822fa9e4066Sahrens buf = (char *)buf + tocpy; 823fa9e4066Sahrens } 824ea8dc4b6Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG); 825fa9e4066Sahrens } 826fa9e4066Sahrens 82782c9918fSTim Haley void 82882c9918fSTim Haley dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 82982c9918fSTim Haley dmu_tx_t *tx) 83082c9918fSTim Haley { 83182c9918fSTim Haley dmu_buf_t **dbp; 83282c9918fSTim Haley int numbufs, i; 83382c9918fSTim Haley 83482c9918fSTim Haley if (size == 0) 83582c9918fSTim Haley return; 83682c9918fSTim Haley 83782c9918fSTim Haley VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 83882c9918fSTim Haley FALSE, FTAG, &numbufs, &dbp)); 83982c9918fSTim Haley 84082c9918fSTim Haley for (i = 0; i < numbufs; i++) { 84182c9918fSTim Haley dmu_buf_t *db = dbp[i]; 84282c9918fSTim Haley 84382c9918fSTim Haley dmu_buf_will_not_fill(db, tx); 84482c9918fSTim Haley } 84582c9918fSTim Haley dmu_buf_rele_array(dbp, numbufs, FTAG); 84682c9918fSTim Haley } 84782c9918fSTim Haley 848c242f9a0Schunli zhang - Sun Microsystems - Irvine United States /* 849c242f9a0Schunli zhang - Sun Microsystems - Irvine United States * DMU support for xuio 850c242f9a0Schunli zhang - Sun Microsystems - Irvine United States */ 851c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_t *xuio_ksp = NULL; 852c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 853c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int 854c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_init(xuio_t *xuio, int nblk) 855c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 856c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv; 857c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio_t *uio = &xuio->xu_uio; 858c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 859c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_iovcnt = nblk; 860c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); 861c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 862c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); 863c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->cnt = nblk; 864c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); 865c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->iovp = uio->uio_iov; 866c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIO_XUZC_PRIV(xuio) = priv; 867c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 868c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (XUIO_XUZC_RW(xuio) == UIO_READ) 869c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); 870c242f9a0Schunli zhang - Sun Microsystems - Irvine United States else 871c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); 872c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 873c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (0); 874c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 875c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 876c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 877c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_fini(xuio_t *xuio) 878c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 879c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 880c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int nblk = priv->cnt; 881c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 882c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kmem_free(priv->iovp, nblk * sizeof (iovec_t)); 883c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); 884c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kmem_free(priv, sizeof (dmu_xuio_t)); 885c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 886c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (XUIO_XUZC_RW(xuio) == UIO_READ) 887c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); 888c242f9a0Schunli zhang - Sun Microsystems - Irvine United States else 889c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); 890c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 891c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 892c242f9a0Schunli zhang - Sun Microsystems - Irvine United States /* 893c242f9a0Schunli zhang - Sun Microsystems - Irvine United States * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } 894c242f9a0Schunli zhang - Sun Microsystems - Irvine United States * and increase priv->next by 1. 895c242f9a0Schunli zhang - Sun Microsystems - Irvine United States */ 896c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int 897c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) 898c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 899c242f9a0Schunli zhang - Sun Microsystems - Irvine United States struct iovec *iov; 900c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio_t *uio = &xuio->xu_uio; 901c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 902c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int i = priv->next++; 903c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 904c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(i < priv->cnt); 905c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(off + n <= arc_buf_size(abuf)); 906c242f9a0Schunli zhang - Sun Microsystems - Irvine United States iov = uio->uio_iov + i; 907c242f9a0Schunli zhang - Sun Microsystems - Irvine United States iov->iov_base = (char *)abuf->b_data + off; 908c242f9a0Schunli zhang - Sun Microsystems - Irvine United States iov->iov_len = n; 909c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->bufs[i] = abuf; 910c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (0); 911c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 912c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 913c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int 914c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_cnt(xuio_t *xuio) 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 return (priv->cnt); 918c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 919c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 920c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t * 921c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_arcbuf(xuio_t *xuio, int i) 922c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 923c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 924c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 925c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(i < priv->cnt); 926c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (priv->bufs[i]); 927c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 928c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 929c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 930c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_clear(xuio_t *xuio, int i) 931c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 932c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 933c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 934c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(i < priv->cnt); 935c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->bufs[i] = NULL; 936c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 937c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 938c242f9a0Schunli zhang - Sun Microsystems - Irvine United States static void 939c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_init(void) 940c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 941c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", 942c242f9a0Schunli zhang - Sun Microsystems - Irvine United States KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), 943c242f9a0Schunli zhang - Sun Microsystems - Irvine United States KSTAT_FLAG_VIRTUAL); 944c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (xuio_ksp != NULL) { 945c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_ksp->ks_data = &xuio_stats; 946c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_install(xuio_ksp); 947c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 948c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 949c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 950c242f9a0Schunli zhang - Sun Microsystems - Irvine United States static void 951c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_fini(void) 952c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 953c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (xuio_ksp != NULL) { 954c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_delete(xuio_ksp); 955c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_ksp = NULL; 956c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 957c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 958c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 959c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 960c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_wbuf_copied() 961c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 962c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_wbuf_copied); 963c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 964c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 965c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 966c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_wbuf_nocopy() 967c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 968c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); 969c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 970c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 971fa9e4066Sahrens #ifdef _KERNEL 972fa9e4066Sahrens int 973feb08c6bSbillm dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) 974feb08c6bSbillm { 975feb08c6bSbillm dmu_buf_t **dbp; 976feb08c6bSbillm int numbufs, i, err; 977c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_t *xuio = NULL; 978feb08c6bSbillm 979feb08c6bSbillm /* 980feb08c6bSbillm * NB: we could do this block-at-a-time, but it's nice 981feb08c6bSbillm * to be reading in parallel. 982feb08c6bSbillm */ 983feb08c6bSbillm err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, 984feb08c6bSbillm &numbufs, &dbp); 985feb08c6bSbillm if (err) 986feb08c6bSbillm return (err); 987feb08c6bSbillm 988c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (uio->uio_extflg == UIO_XUIO) 989c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio = (xuio_t *)uio; 990c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 991feb08c6bSbillm for (i = 0; i < numbufs; i++) { 992feb08c6bSbillm int tocpy; 993feb08c6bSbillm int bufoff; 994feb08c6bSbillm dmu_buf_t *db = dbp[i]; 995feb08c6bSbillm 996feb08c6bSbillm ASSERT(size > 0); 997feb08c6bSbillm 998feb08c6bSbillm bufoff = uio->uio_loffset - db->db_offset; 999feb08c6bSbillm tocpy = (int)MIN(db->db_size - bufoff, size); 1000feb08c6bSbillm 1001c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (xuio) { 1002c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; 1003c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t *dbuf_abuf = dbi->db_buf; 1004c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); 1005c242f9a0Schunli zhang - Sun Microsystems - Irvine United States err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); 1006c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (!err) { 1007c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_resid -= tocpy; 1008c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_loffset += tocpy; 1009c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1010c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 1011c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (abuf == dbuf_abuf) 1012c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); 1013c242f9a0Schunli zhang - Sun Microsystems - Irvine United States else 1014c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_rbuf_copied); 1015c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } else { 1016feb08c6bSbillm err = uiomove((char *)db->db_data + bufoff, tocpy, 1017feb08c6bSbillm UIO_READ, uio); 1018c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1019feb08c6bSbillm if (err) 1020feb08c6bSbillm break; 1021feb08c6bSbillm 1022feb08c6bSbillm size -= tocpy; 1023feb08c6bSbillm } 1024feb08c6bSbillm dmu_buf_rele_array(dbp, numbufs, FTAG); 1025feb08c6bSbillm 1026feb08c6bSbillm return (err); 1027feb08c6bSbillm } 1028feb08c6bSbillm 102994d1a210STim Haley static int 103094d1a210STim Haley dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) 1031fa9e4066Sahrens { 1032fa9e4066Sahrens dmu_buf_t **dbp; 103394d1a210STim Haley int numbufs; 1034fa9e4066Sahrens int err = 0; 103594d1a210STim Haley int i; 1036fa9e4066Sahrens 103794d1a210STim Haley err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, 103894d1a210STim Haley FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); 1039ea8dc4b6Seschrock if (err) 1040ea8dc4b6Seschrock return (err); 1041fa9e4066Sahrens 1042fa9e4066Sahrens for (i = 0; i < numbufs; i++) { 1043fa9e4066Sahrens int tocpy; 1044fa9e4066Sahrens int bufoff; 1045fa9e4066Sahrens dmu_buf_t *db = dbp[i]; 1046fa9e4066Sahrens 1047fa9e4066Sahrens ASSERT(size > 0); 1048fa9e4066Sahrens 1049feb08c6bSbillm bufoff = uio->uio_loffset - db->db_offset; 1050fa9e4066Sahrens tocpy = (int)MIN(db->db_size - bufoff, size); 1051fa9e4066Sahrens 1052fa9e4066Sahrens ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 1053fa9e4066Sahrens 1054fa9e4066Sahrens if (tocpy == db->db_size) 1055fa9e4066Sahrens dmu_buf_will_fill(db, tx); 1056fa9e4066Sahrens else 1057fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 1058fa9e4066Sahrens 1059fa9e4066Sahrens /* 1060fa9e4066Sahrens * XXX uiomove could block forever (eg. nfs-backed 1061fa9e4066Sahrens * pages). There needs to be a uiolockdown() function 1062fa9e4066Sahrens * to lock the pages in memory, so that uiomove won't 1063fa9e4066Sahrens * block. 1064fa9e4066Sahrens */ 1065fa9e4066Sahrens err = uiomove((char *)db->db_data + bufoff, tocpy, 1066fa9e4066Sahrens UIO_WRITE, uio); 1067fa9e4066Sahrens 1068fa9e4066Sahrens if (tocpy == db->db_size) 1069fa9e4066Sahrens dmu_buf_fill_done(db, tx); 1070fa9e4066Sahrens 1071fa9e4066Sahrens if (err) 1072fa9e4066Sahrens break; 1073fa9e4066Sahrens 1074fa9e4066Sahrens size -= tocpy; 1075fa9e4066Sahrens } 107694d1a210STim Haley 1077ea8dc4b6Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG); 1078fa9e4066Sahrens return (err); 1079fa9e4066Sahrens } 108044eda4d7Smaybee 108144eda4d7Smaybee int 108294d1a210STim Haley dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, 108394d1a210STim Haley dmu_tx_t *tx) 108494d1a210STim Haley { 1085744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; 1086744947dcSTom Erickson dnode_t *dn; 1087744947dcSTom Erickson int err; 1088744947dcSTom Erickson 108994d1a210STim Haley if (size == 0) 109094d1a210STim Haley return (0); 109194d1a210STim Haley 1092744947dcSTom Erickson DB_DNODE_ENTER(db); 1093744947dcSTom Erickson dn = DB_DNODE(db); 1094744947dcSTom Erickson err = dmu_write_uio_dnode(dn, uio, size, tx); 1095744947dcSTom Erickson DB_DNODE_EXIT(db); 1096744947dcSTom Erickson 1097744947dcSTom Erickson return (err); 109894d1a210STim Haley } 109994d1a210STim Haley 110094d1a210STim Haley int 110194d1a210STim Haley dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, 110294d1a210STim Haley dmu_tx_t *tx) 110394d1a210STim Haley { 110494d1a210STim Haley dnode_t *dn; 110594d1a210STim Haley int err; 110694d1a210STim Haley 110794d1a210STim Haley if (size == 0) 110894d1a210STim Haley return (0); 110994d1a210STim Haley 111094d1a210STim Haley err = dnode_hold(os, object, FTAG, &dn); 111194d1a210STim Haley if (err) 111294d1a210STim Haley return (err); 111394d1a210STim Haley 111494d1a210STim Haley err = dmu_write_uio_dnode(dn, uio, size, tx); 111594d1a210STim Haley 111694d1a210STim Haley dnode_rele(dn, FTAG); 111794d1a210STim Haley 111894d1a210STim Haley return (err); 111994d1a210STim Haley } 112094d1a210STim Haley 112194d1a210STim Haley int 112244eda4d7Smaybee dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 112344eda4d7Smaybee page_t *pp, dmu_tx_t *tx) 112444eda4d7Smaybee { 112544eda4d7Smaybee dmu_buf_t **dbp; 112644eda4d7Smaybee int numbufs, i; 112744eda4d7Smaybee int err; 112844eda4d7Smaybee 112944eda4d7Smaybee if (size == 0) 113044eda4d7Smaybee return (0); 113144eda4d7Smaybee 113244eda4d7Smaybee err = dmu_buf_hold_array(os, object, offset, size, 113344eda4d7Smaybee FALSE, FTAG, &numbufs, &dbp); 113444eda4d7Smaybee if (err) 113544eda4d7Smaybee return (err); 113644eda4d7Smaybee 113744eda4d7Smaybee for (i = 0; i < numbufs; i++) { 113844eda4d7Smaybee int tocpy, copied, thiscpy; 113944eda4d7Smaybee int bufoff; 114044eda4d7Smaybee dmu_buf_t *db = dbp[i]; 114144eda4d7Smaybee caddr_t va; 114244eda4d7Smaybee 114344eda4d7Smaybee ASSERT(size > 0); 114444eda4d7Smaybee ASSERT3U(db->db_size, >=, PAGESIZE); 114544eda4d7Smaybee 114644eda4d7Smaybee bufoff = offset - db->db_offset; 114744eda4d7Smaybee tocpy = (int)MIN(db->db_size - bufoff, size); 114844eda4d7Smaybee 114944eda4d7Smaybee ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 115044eda4d7Smaybee 115144eda4d7Smaybee if (tocpy == db->db_size) 115244eda4d7Smaybee dmu_buf_will_fill(db, tx); 115344eda4d7Smaybee else 115444eda4d7Smaybee dmu_buf_will_dirty(db, tx); 115544eda4d7Smaybee 115644eda4d7Smaybee for (copied = 0; copied < tocpy; copied += PAGESIZE) { 115744eda4d7Smaybee ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); 115844eda4d7Smaybee thiscpy = MIN(PAGESIZE, tocpy - copied); 11590fab61baSJonathan W Adams va = zfs_map_page(pp, S_READ); 116044eda4d7Smaybee bcopy(va, (char *)db->db_data + bufoff, thiscpy); 11610fab61baSJonathan W Adams zfs_unmap_page(pp, va); 116244eda4d7Smaybee pp = pp->p_next; 116344eda4d7Smaybee bufoff += PAGESIZE; 116444eda4d7Smaybee } 116544eda4d7Smaybee 116644eda4d7Smaybee if (tocpy == db->db_size) 116744eda4d7Smaybee dmu_buf_fill_done(db, tx); 116844eda4d7Smaybee 116944eda4d7Smaybee offset += tocpy; 117044eda4d7Smaybee size -= tocpy; 117144eda4d7Smaybee } 117244eda4d7Smaybee dmu_buf_rele_array(dbp, numbufs, FTAG); 117344eda4d7Smaybee return (err); 117444eda4d7Smaybee } 1175fa9e4066Sahrens #endif 1176fa9e4066Sahrens 11772fdbea25SAleksandr Guzovskiy /* 11782fdbea25SAleksandr Guzovskiy * Allocate a loaned anonymous arc buffer. 11792fdbea25SAleksandr Guzovskiy */ 11802fdbea25SAleksandr Guzovskiy arc_buf_t * 11812fdbea25SAleksandr Guzovskiy dmu_request_arcbuf(dmu_buf_t *handle, int size) 11822fdbea25SAleksandr Guzovskiy { 1183744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; 1184744947dcSTom Erickson spa_t *spa; 11852fdbea25SAleksandr Guzovskiy 1186744947dcSTom Erickson DB_GET_SPA(&spa, db); 1187744947dcSTom Erickson return (arc_loan_buf(spa, size)); 11882fdbea25SAleksandr Guzovskiy } 11892fdbea25SAleksandr Guzovskiy 11902fdbea25SAleksandr Guzovskiy /* 11912fdbea25SAleksandr Guzovskiy * Free a loaned arc buffer. 11922fdbea25SAleksandr Guzovskiy */ 11932fdbea25SAleksandr Guzovskiy void 11942fdbea25SAleksandr Guzovskiy dmu_return_arcbuf(arc_buf_t *buf) 11952fdbea25SAleksandr Guzovskiy { 11962fdbea25SAleksandr Guzovskiy arc_return_buf(buf, FTAG); 1197*3b2aab18SMatthew Ahrens VERIFY(arc_buf_remove_ref(buf, FTAG)); 11982fdbea25SAleksandr Guzovskiy } 11992fdbea25SAleksandr Guzovskiy 12002fdbea25SAleksandr Guzovskiy /* 12012fdbea25SAleksandr Guzovskiy * When possible directly assign passed loaned arc buffer to a dbuf. 12022fdbea25SAleksandr Guzovskiy * If this is not possible copy the contents of passed arc buf via 12032fdbea25SAleksandr Guzovskiy * dmu_write(). 12042fdbea25SAleksandr Guzovskiy */ 12052fdbea25SAleksandr Guzovskiy void 12062fdbea25SAleksandr Guzovskiy dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, 12072fdbea25SAleksandr Guzovskiy dmu_tx_t *tx) 12082fdbea25SAleksandr Guzovskiy { 1209744947dcSTom Erickson dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; 1210744947dcSTom Erickson dnode_t *dn; 12112fdbea25SAleksandr Guzovskiy dmu_buf_impl_t *db; 12122fdbea25SAleksandr Guzovskiy uint32_t blksz = (uint32_t)arc_buf_size(buf); 12132fdbea25SAleksandr Guzovskiy uint64_t blkid; 12142fdbea25SAleksandr Guzovskiy 1215744947dcSTom Erickson DB_DNODE_ENTER(dbuf); 1216744947dcSTom Erickson dn = DB_DNODE(dbuf); 12172fdbea25SAleksandr Guzovskiy rw_enter(&dn->dn_struct_rwlock, RW_READER); 12182fdbea25SAleksandr Guzovskiy blkid = dbuf_whichblock(dn, offset); 12192fdbea25SAleksandr Guzovskiy VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); 12202fdbea25SAleksandr Guzovskiy rw_exit(&dn->dn_struct_rwlock); 1221744947dcSTom Erickson DB_DNODE_EXIT(dbuf); 12222fdbea25SAleksandr Guzovskiy 12232fdbea25SAleksandr Guzovskiy if (offset == db->db.db_offset && blksz == db->db.db_size) { 12242fdbea25SAleksandr Guzovskiy dbuf_assign_arcbuf(db, buf, tx); 12252fdbea25SAleksandr Guzovskiy dbuf_rele(db, FTAG); 12262fdbea25SAleksandr Guzovskiy } else { 1227744947dcSTom Erickson objset_t *os; 1228744947dcSTom Erickson uint64_t object; 1229744947dcSTom Erickson 1230744947dcSTom Erickson DB_DNODE_ENTER(dbuf); 1231744947dcSTom Erickson dn = DB_DNODE(dbuf); 1232744947dcSTom Erickson os = dn->dn_objset; 1233744947dcSTom Erickson object = dn->dn_object; 1234744947dcSTom Erickson DB_DNODE_EXIT(dbuf); 1235744947dcSTom Erickson 12362fdbea25SAleksandr Guzovskiy dbuf_rele(db, FTAG); 1237744947dcSTom Erickson dmu_write(os, object, offset, blksz, buf->b_data, tx); 12382fdbea25SAleksandr Guzovskiy dmu_return_arcbuf(buf); 1239c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_wbuf_copied); 12402fdbea25SAleksandr Guzovskiy } 12412fdbea25SAleksandr Guzovskiy } 12422fdbea25SAleksandr Guzovskiy 1243c5c6ffa0Smaybee typedef struct { 1244b24ab676SJeff Bonwick dbuf_dirty_record_t *dsa_dr; 1245b24ab676SJeff Bonwick dmu_sync_cb_t *dsa_done; 1246b24ab676SJeff Bonwick zgd_t *dsa_zgd; 1247b24ab676SJeff Bonwick dmu_tx_t *dsa_tx; 1248c717a561Smaybee } dmu_sync_arg_t; 1249c5c6ffa0Smaybee 1250c5c6ffa0Smaybee /* ARGSUSED */ 1251c5c6ffa0Smaybee static void 1252e14bb325SJeff Bonwick dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) 1253e14bb325SJeff Bonwick { 1254b24ab676SJeff Bonwick dmu_sync_arg_t *dsa = varg; 1255b24ab676SJeff Bonwick dmu_buf_t *db = dsa->dsa_zgd->zgd_db; 1256e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1257975c32a0SNeil Perrin 1258b24ab676SJeff Bonwick if (zio->io_error == 0) { 1259b24ab676SJeff Bonwick if (BP_IS_HOLE(bp)) { 1260b24ab676SJeff Bonwick /* 1261b24ab676SJeff Bonwick * A block of zeros may compress to a hole, but the 1262b24ab676SJeff Bonwick * block size still needs to be known for replay. 1263b24ab676SJeff Bonwick */ 1264b24ab676SJeff Bonwick BP_SET_LSIZE(bp, db->db_size); 1265b24ab676SJeff Bonwick } else { 1266e14bb325SJeff Bonwick ASSERT(BP_GET_LEVEL(bp) == 0); 1267e14bb325SJeff Bonwick bp->blk_fill = 1; 1268e14bb325SJeff Bonwick } 1269e14bb325SJeff Bonwick } 1270b24ab676SJeff Bonwick } 1271b24ab676SJeff Bonwick 1272b24ab676SJeff Bonwick static void 1273b24ab676SJeff Bonwick dmu_sync_late_arrival_ready(zio_t *zio) 1274b24ab676SJeff Bonwick { 1275b24ab676SJeff Bonwick dmu_sync_ready(zio, NULL, zio->io_private); 1276b24ab676SJeff Bonwick } 1277e14bb325SJeff Bonwick 1278e14bb325SJeff Bonwick /* ARGSUSED */ 1279e14bb325SJeff Bonwick static void 1280c5c6ffa0Smaybee dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) 1281c5c6ffa0Smaybee { 1282b24ab676SJeff Bonwick dmu_sync_arg_t *dsa = varg; 1283b24ab676SJeff Bonwick dbuf_dirty_record_t *dr = dsa->dsa_dr; 1284c717a561Smaybee dmu_buf_impl_t *db = dr->dr_dbuf; 1285c5c6ffa0Smaybee 1286b50a0fe0SNeil Perrin mutex_enter(&db->db_mtx); 1287b50a0fe0SNeil Perrin ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); 1288b24ab676SJeff Bonwick if (zio->io_error == 0) { 128980901aeaSGeorge Wilson dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); 129080901aeaSGeorge Wilson if (dr->dt.dl.dr_nopwrite) { 129180901aeaSGeorge Wilson blkptr_t *bp = zio->io_bp; 129280901aeaSGeorge Wilson blkptr_t *bp_orig = &zio->io_bp_orig; 129380901aeaSGeorge Wilson uint8_t chksum = BP_GET_CHECKSUM(bp_orig); 129480901aeaSGeorge Wilson 129580901aeaSGeorge Wilson ASSERT(BP_EQUAL(bp, bp_orig)); 129680901aeaSGeorge Wilson ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); 129780901aeaSGeorge Wilson ASSERT(zio_checksum_table[chksum].ci_dedup); 129880901aeaSGeorge Wilson } 1299b24ab676SJeff Bonwick dr->dt.dl.dr_overridden_by = *zio->io_bp; 1300b24ab676SJeff Bonwick dr->dt.dl.dr_override_state = DR_OVERRIDDEN; 1301b24ab676SJeff Bonwick dr->dt.dl.dr_copies = zio->io_prop.zp_copies; 1302b50a0fe0SNeil Perrin if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by)) 1303b50a0fe0SNeil Perrin BP_ZERO(&dr->dt.dl.dr_overridden_by); 1304b24ab676SJeff Bonwick } else { 1305b24ab676SJeff Bonwick dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; 1306b24ab676SJeff Bonwick } 1307c5c6ffa0Smaybee cv_broadcast(&db->db_changed); 1308b50a0fe0SNeil Perrin mutex_exit(&db->db_mtx); 1309b50a0fe0SNeil Perrin 1310b24ab676SJeff Bonwick dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1311c717a561Smaybee 1312b24ab676SJeff Bonwick kmem_free(dsa, sizeof (*dsa)); 1313b24ab676SJeff Bonwick } 1314b24ab676SJeff Bonwick 1315b24ab676SJeff Bonwick static void 1316b24ab676SJeff Bonwick dmu_sync_late_arrival_done(zio_t *zio) 1317b24ab676SJeff Bonwick { 1318b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1319b24ab676SJeff Bonwick dmu_sync_arg_t *dsa = zio->io_private; 132080901aeaSGeorge Wilson blkptr_t *bp_orig = &zio->io_bp_orig; 1321b24ab676SJeff Bonwick 1322b24ab676SJeff Bonwick if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { 132380901aeaSGeorge Wilson /* 132480901aeaSGeorge Wilson * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE) 132580901aeaSGeorge Wilson * then there is nothing to do here. Otherwise, free the 132680901aeaSGeorge Wilson * newly allocated block in this txg. 132780901aeaSGeorge Wilson */ 132880901aeaSGeorge Wilson if (zio->io_flags & ZIO_FLAG_NOPWRITE) { 132980901aeaSGeorge Wilson ASSERT(BP_EQUAL(bp, bp_orig)); 133080901aeaSGeorge Wilson } else { 133180901aeaSGeorge Wilson ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); 1332b24ab676SJeff Bonwick ASSERT(zio->io_bp->blk_birth == zio->io_txg); 1333b24ab676SJeff Bonwick ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); 1334b24ab676SJeff Bonwick zio_free(zio->io_spa, zio->io_txg, zio->io_bp); 1335b24ab676SJeff Bonwick } 133680901aeaSGeorge Wilson } 1337b24ab676SJeff Bonwick 1338b24ab676SJeff Bonwick dmu_tx_commit(dsa->dsa_tx); 1339b24ab676SJeff Bonwick 1340b24ab676SJeff Bonwick dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1341b24ab676SJeff Bonwick 1342b24ab676SJeff Bonwick kmem_free(dsa, sizeof (*dsa)); 1343b24ab676SJeff Bonwick } 1344b24ab676SJeff Bonwick 1345b24ab676SJeff Bonwick static int 1346b24ab676SJeff Bonwick dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, 1347b24ab676SJeff Bonwick zio_prop_t *zp, zbookmark_t *zb) 1348b24ab676SJeff Bonwick { 1349b24ab676SJeff Bonwick dmu_sync_arg_t *dsa; 1350b24ab676SJeff Bonwick dmu_tx_t *tx; 1351b24ab676SJeff Bonwick 1352b24ab676SJeff Bonwick tx = dmu_tx_create(os); 1353b24ab676SJeff Bonwick dmu_tx_hold_space(tx, zgd->zgd_db->db_size); 13546e1f5caaSNeil Perrin if (dmu_tx_assign(tx, TXG_WAIT) != 0) { 1355b24ab676SJeff Bonwick dmu_tx_abort(tx); 1356b24ab676SJeff Bonwick return (EIO); /* Make zl_get_data do txg_waited_synced() */ 1357b24ab676SJeff Bonwick } 1358b24ab676SJeff Bonwick 1359b24ab676SJeff Bonwick dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1360b24ab676SJeff Bonwick dsa->dsa_dr = NULL; 1361b24ab676SJeff Bonwick dsa->dsa_done = done; 1362b24ab676SJeff Bonwick dsa->dsa_zgd = zgd; 1363b24ab676SJeff Bonwick dsa->dsa_tx = tx; 1364b24ab676SJeff Bonwick 1365b24ab676SJeff Bonwick zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, 1366b24ab676SJeff Bonwick zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, 1367b24ab676SJeff Bonwick dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa, 1368b24ab676SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); 1369b24ab676SJeff Bonwick 1370b24ab676SJeff Bonwick return (0); 1371c5c6ffa0Smaybee } 1372c5c6ffa0Smaybee 1373fa9e4066Sahrens /* 1374c5c6ffa0Smaybee * Intent log support: sync the block associated with db to disk. 1375c5c6ffa0Smaybee * N.B. and XXX: the caller is responsible for making sure that the 1376c5c6ffa0Smaybee * data isn't changing while dmu_sync() is writing it. 1377fa9e4066Sahrens * 1378fa9e4066Sahrens * Return values: 1379fa9e4066Sahrens * 138080901aeaSGeorge Wilson * EEXIST: this txg has already been synced, so there's nothing to do. 1381fa9e4066Sahrens * The caller should not log the write. 1382fa9e4066Sahrens * 1383fa9e4066Sahrens * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. 1384fa9e4066Sahrens * The caller should not log the write. 1385fa9e4066Sahrens * 1386c5c6ffa0Smaybee * EALREADY: this block is already in the process of being synced. 1387c5c6ffa0Smaybee * The caller should track its progress (somehow). 1388fa9e4066Sahrens * 1389b24ab676SJeff Bonwick * EIO: could not do the I/O. 1390b24ab676SJeff Bonwick * The caller should do a txg_wait_synced(). 1391fa9e4066Sahrens * 1392b24ab676SJeff Bonwick * 0: the I/O has been initiated. 1393b24ab676SJeff Bonwick * The caller should log this blkptr in the done callback. 1394b24ab676SJeff Bonwick * It is possible that the I/O will fail, in which case 1395b24ab676SJeff Bonwick * the error will be reported to the done callback and 1396b24ab676SJeff Bonwick * propagated to pio from zio_done(). 1397fa9e4066Sahrens */ 1398fa9e4066Sahrens int 1399b24ab676SJeff Bonwick dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) 1400fa9e4066Sahrens { 1401b24ab676SJeff Bonwick blkptr_t *bp = zgd->zgd_bp; 1402b24ab676SJeff Bonwick dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; 1403503ad85cSMatthew Ahrens objset_t *os = db->db_objset; 1404b24ab676SJeff Bonwick dsl_dataset_t *ds = os->os_dsl_dataset; 1405c717a561Smaybee dbuf_dirty_record_t *dr; 1406b24ab676SJeff Bonwick dmu_sync_arg_t *dsa; 1407ea8dc4b6Seschrock zbookmark_t zb; 1408b24ab676SJeff Bonwick zio_prop_t zp; 1409744947dcSTom Erickson dnode_t *dn; 1410fa9e4066Sahrens 1411b24ab676SJeff Bonwick ASSERT(pio != NULL); 1412fa9e4066Sahrens ASSERT(txg != 0); 1413fa9e4066Sahrens 1414b24ab676SJeff Bonwick SET_BOOKMARK(&zb, ds->ds_object, 1415b24ab676SJeff Bonwick db->db.db_object, db->db_level, db->db_blkid); 1416b24ab676SJeff Bonwick 1417744947dcSTom Erickson DB_DNODE_ENTER(db); 1418744947dcSTom Erickson dn = DB_DNODE(db); 1419744947dcSTom Erickson dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); 1420744947dcSTom Erickson DB_DNODE_EXIT(db); 1421fa9e4066Sahrens 1422fa9e4066Sahrens /* 1423b24ab676SJeff Bonwick * If we're frozen (running ziltest), we always need to generate a bp. 1424ea8dc4b6Seschrock */ 1425b24ab676SJeff Bonwick if (txg > spa_freeze_txg(os->os_spa)) 1426b24ab676SJeff Bonwick return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1427ea8dc4b6Seschrock 1428ea8dc4b6Seschrock /* 1429b24ab676SJeff Bonwick * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() 1430b24ab676SJeff Bonwick * and us. If we determine that this txg is not yet syncing, 1431b24ab676SJeff Bonwick * but it begins to sync a moment later, that's OK because the 1432b24ab676SJeff Bonwick * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. 1433fa9e4066Sahrens */ 1434b24ab676SJeff Bonwick mutex_enter(&db->db_mtx); 1435b24ab676SJeff Bonwick 1436b24ab676SJeff Bonwick if (txg <= spa_last_synced_txg(os->os_spa)) { 1437fa9e4066Sahrens /* 1438b24ab676SJeff Bonwick * This txg has already synced. There's nothing to do. 1439fa9e4066Sahrens */ 1440b24ab676SJeff Bonwick mutex_exit(&db->db_mtx); 1441c5c6ffa0Smaybee return (EEXIST); 1442fa9e4066Sahrens } 1443fa9e4066Sahrens 1444b24ab676SJeff Bonwick if (txg <= spa_syncing_txg(os->os_spa)) { 1445c5c6ffa0Smaybee /* 1446b24ab676SJeff Bonwick * This txg is currently syncing, so we can't mess with 1447b24ab676SJeff Bonwick * the dirty record anymore; just write a new log block. 1448c5c6ffa0Smaybee */ 144913506d1eSmaybee mutex_exit(&db->db_mtx); 1450b24ab676SJeff Bonwick return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1451c5c6ffa0Smaybee } 1452c5c6ffa0Smaybee 1453c717a561Smaybee dr = db->db_last_dirty; 1454b24ab676SJeff Bonwick while (dr && dr->dr_txg != txg) 1455c717a561Smaybee dr = dr->dr_next; 1456b24ab676SJeff Bonwick 1457b24ab676SJeff Bonwick if (dr == NULL) { 1458c5c6ffa0Smaybee /* 1459b24ab676SJeff Bonwick * There's no dr for this dbuf, so it must have been freed. 1460fa9e4066Sahrens * There's no need to log writes to freed blocks, so we're done. 1461fa9e4066Sahrens */ 1462fa9e4066Sahrens mutex_exit(&db->db_mtx); 1463fa9e4066Sahrens return (ENOENT); 1464fa9e4066Sahrens } 1465fa9e4066Sahrens 146680901aeaSGeorge Wilson ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); 146780901aeaSGeorge Wilson 146880901aeaSGeorge Wilson /* 146980901aeaSGeorge Wilson * Assume the on-disk data is X, the current syncing data is Y, 147080901aeaSGeorge Wilson * and the current in-memory data is Z (currently in dmu_sync). 147180901aeaSGeorge Wilson * X and Z are identical but Y is has been modified. Normally, 147280901aeaSGeorge Wilson * when X and Z are the same we will perform a nopwrite but if Y 147380901aeaSGeorge Wilson * is different we must disable nopwrite since the resulting write 147480901aeaSGeorge Wilson * of Y to disk can free the block containing X. If we allowed a 147580901aeaSGeorge Wilson * nopwrite to occur the block pointing to Z would reference a freed 147680901aeaSGeorge Wilson * block. Since this is a rare case we simplify this by disabling 147780901aeaSGeorge Wilson * nopwrite if the current dmu_sync-ing dbuf has been modified in 147880901aeaSGeorge Wilson * a previous transaction. 147980901aeaSGeorge Wilson */ 148080901aeaSGeorge Wilson if (dr->dr_next) 148180901aeaSGeorge Wilson zp.zp_nopwrite = B_FALSE; 148280901aeaSGeorge Wilson 1483c717a561Smaybee ASSERT(dr->dr_txg == txg); 1484b24ab676SJeff Bonwick if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || 1485b24ab676SJeff Bonwick dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { 1486c5c6ffa0Smaybee /* 1487b24ab676SJeff Bonwick * We have already issued a sync write for this buffer, 1488b24ab676SJeff Bonwick * or this buffer has already been synced. It could not 1489c717a561Smaybee * have been dirtied since, or we would have cleared the state. 1490c717a561Smaybee */ 1491c717a561Smaybee mutex_exit(&db->db_mtx); 1492b24ab676SJeff Bonwick return (EALREADY); 1493c717a561Smaybee } 1494c717a561Smaybee 1495b24ab676SJeff Bonwick ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); 1496c717a561Smaybee dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; 1497fa9e4066Sahrens mutex_exit(&db->db_mtx); 1498fa9e4066Sahrens 1499b24ab676SJeff Bonwick dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1500b24ab676SJeff Bonwick dsa->dsa_dr = dr; 1501b24ab676SJeff Bonwick dsa->dsa_done = done; 1502b24ab676SJeff Bonwick dsa->dsa_zgd = zgd; 1503b24ab676SJeff Bonwick dsa->dsa_tx = NULL; 1504e14bb325SJeff Bonwick 1505b24ab676SJeff Bonwick zio_nowait(arc_write(pio, os->os_spa, txg, 1506b24ab676SJeff Bonwick bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp, 1507b24ab676SJeff Bonwick dmu_sync_ready, dmu_sync_done, dsa, 1508b24ab676SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); 1509e14bb325SJeff Bonwick 1510b24ab676SJeff Bonwick return (0); 1511fa9e4066Sahrens } 1512fa9e4066Sahrens 1513fa9e4066Sahrens int 1514fa9e4066Sahrens dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, 1515fa9e4066Sahrens dmu_tx_t *tx) 1516fa9e4066Sahrens { 1517ea8dc4b6Seschrock dnode_t *dn; 1518ea8dc4b6Seschrock int err; 1519ea8dc4b6Seschrock 1520503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 1521ea8dc4b6Seschrock if (err) 1522ea8dc4b6Seschrock return (err); 1523ea8dc4b6Seschrock err = dnode_set_blksz(dn, size, ibs, tx); 1524fa9e4066Sahrens dnode_rele(dn, FTAG); 1525fa9e4066Sahrens return (err); 1526fa9e4066Sahrens } 1527fa9e4066Sahrens 1528fa9e4066Sahrens void 1529fa9e4066Sahrens dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 1530fa9e4066Sahrens dmu_tx_t *tx) 1531fa9e4066Sahrens { 1532ea8dc4b6Seschrock dnode_t *dn; 1533ea8dc4b6Seschrock 1534ea8dc4b6Seschrock /* XXX assumes dnode_hold will not get an i/o error */ 1535503ad85cSMatthew Ahrens (void) dnode_hold(os, object, FTAG, &dn); 1536fa9e4066Sahrens ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS); 1537fa9e4066Sahrens dn->dn_checksum = checksum; 1538fa9e4066Sahrens dnode_setdirty(dn, tx); 1539fa9e4066Sahrens dnode_rele(dn, FTAG); 1540fa9e4066Sahrens } 1541fa9e4066Sahrens 1542fa9e4066Sahrens void 1543fa9e4066Sahrens dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 1544fa9e4066Sahrens dmu_tx_t *tx) 1545fa9e4066Sahrens { 1546ea8dc4b6Seschrock dnode_t *dn; 1547ea8dc4b6Seschrock 1548ea8dc4b6Seschrock /* XXX assumes dnode_hold will not get an i/o error */ 1549503ad85cSMatthew Ahrens (void) dnode_hold(os, object, FTAG, &dn); 1550fa9e4066Sahrens ASSERT(compress < ZIO_COMPRESS_FUNCTIONS); 1551fa9e4066Sahrens dn->dn_compress = compress; 1552fa9e4066Sahrens dnode_setdirty(dn, tx); 1553fa9e4066Sahrens dnode_rele(dn, FTAG); 1554fa9e4066Sahrens } 1555fa9e4066Sahrens 1556b24ab676SJeff Bonwick int zfs_mdcomp_disable = 0; 1557b24ab676SJeff Bonwick 1558b24ab676SJeff Bonwick void 1559b24ab676SJeff Bonwick dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) 1560b24ab676SJeff Bonwick { 1561b24ab676SJeff Bonwick dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; 1562ad135b5dSChristopher Siden boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || 15631d8ccc7bSMark Shellenbaum (wp & WP_SPILL)); 1564b24ab676SJeff Bonwick enum zio_checksum checksum = os->os_checksum; 1565b24ab676SJeff Bonwick enum zio_compress compress = os->os_compress; 1566b24ab676SJeff Bonwick enum zio_checksum dedup_checksum = os->os_dedup_checksum; 15677540df39SGeorge Wilson boolean_t dedup = B_FALSE; 15687540df39SGeorge Wilson boolean_t nopwrite = B_FALSE; 1569b24ab676SJeff Bonwick boolean_t dedup_verify = os->os_dedup_verify; 1570b24ab676SJeff Bonwick int copies = os->os_copies; 1571b24ab676SJeff Bonwick 1572b24ab676SJeff Bonwick /* 157380901aeaSGeorge Wilson * We maintain different write policies for each of the following 157480901aeaSGeorge Wilson * types of data: 157580901aeaSGeorge Wilson * 1. metadata 157680901aeaSGeorge Wilson * 2. preallocated blocks (i.e. level-0 blocks of a dump device) 157780901aeaSGeorge Wilson * 3. all other level 0 blocks 1578b24ab676SJeff Bonwick */ 1579b24ab676SJeff Bonwick if (ismd) { 1580b24ab676SJeff Bonwick /* 158180901aeaSGeorge Wilson * XXX -- we should design a compression algorithm 158280901aeaSGeorge Wilson * that specializes in arrays of bps. 158380901aeaSGeorge Wilson */ 158480901aeaSGeorge Wilson compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY : 158580901aeaSGeorge Wilson ZIO_COMPRESS_LZJB; 158680901aeaSGeorge Wilson 158780901aeaSGeorge Wilson /* 1588b24ab676SJeff Bonwick * Metadata always gets checksummed. If the data 1589b24ab676SJeff Bonwick * checksum is multi-bit correctable, and it's not a 1590b24ab676SJeff Bonwick * ZBT-style checksum, then it's suitable for metadata 1591b24ab676SJeff Bonwick * as well. Otherwise, the metadata checksum defaults 1592b24ab676SJeff Bonwick * to fletcher4. 1593b24ab676SJeff Bonwick */ 1594b24ab676SJeff Bonwick if (zio_checksum_table[checksum].ci_correctable < 1 || 15956e1f5caaSNeil Perrin zio_checksum_table[checksum].ci_eck) 1596b24ab676SJeff Bonwick checksum = ZIO_CHECKSUM_FLETCHER_4; 159780901aeaSGeorge Wilson } else if (wp & WP_NOFILL) { 159880901aeaSGeorge Wilson ASSERT(level == 0); 1599b24ab676SJeff Bonwick 1600b24ab676SJeff Bonwick /* 160180901aeaSGeorge Wilson * If we're writing preallocated blocks, we aren't actually 160280901aeaSGeorge Wilson * writing them so don't set any policy properties. These 160380901aeaSGeorge Wilson * blocks are currently only used by an external subsystem 160480901aeaSGeorge Wilson * outside of zfs (i.e. dump) and not written by the zio 160580901aeaSGeorge Wilson * pipeline. 1606b24ab676SJeff Bonwick */ 160780901aeaSGeorge Wilson compress = ZIO_COMPRESS_OFF; 160880901aeaSGeorge Wilson checksum = ZIO_CHECKSUM_OFF; 1609b24ab676SJeff Bonwick } else { 1610b24ab676SJeff Bonwick compress = zio_compress_select(dn->dn_compress, compress); 161180901aeaSGeorge Wilson 161280901aeaSGeorge Wilson checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? 161380901aeaSGeorge Wilson zio_checksum_select(dn->dn_checksum, checksum) : 161480901aeaSGeorge Wilson dedup_checksum; 161580901aeaSGeorge Wilson 161680901aeaSGeorge Wilson /* 161780901aeaSGeorge Wilson * Determine dedup setting. If we are in dmu_sync(), 161880901aeaSGeorge Wilson * we won't actually dedup now because that's all 161980901aeaSGeorge Wilson * done in syncing context; but we do want to use the 162080901aeaSGeorge Wilson * dedup checkum. If the checksum is not strong 162180901aeaSGeorge Wilson * enough to ensure unique signatures, force 162280901aeaSGeorge Wilson * dedup_verify. 162380901aeaSGeorge Wilson */ 162480901aeaSGeorge Wilson if (dedup_checksum != ZIO_CHECKSUM_OFF) { 162580901aeaSGeorge Wilson dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; 162680901aeaSGeorge Wilson if (!zio_checksum_table[checksum].ci_dedup) 162780901aeaSGeorge Wilson dedup_verify = B_TRUE; 1628b24ab676SJeff Bonwick } 1629b24ab676SJeff Bonwick 1630b24ab676SJeff Bonwick /* 163180901aeaSGeorge Wilson * Enable nopwrite if we have a cryptographically secure 163280901aeaSGeorge Wilson * checksum that has no known collisions (i.e. SHA-256) 163380901aeaSGeorge Wilson * and compression is enabled. We don't enable nopwrite if 163480901aeaSGeorge Wilson * dedup is enabled as the two features are mutually exclusive. 1635b24ab676SJeff Bonwick */ 163680901aeaSGeorge Wilson nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup && 163780901aeaSGeorge Wilson compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); 1638b24ab676SJeff Bonwick } 1639b24ab676SJeff Bonwick 1640b24ab676SJeff Bonwick zp->zp_checksum = checksum; 1641b24ab676SJeff Bonwick zp->zp_compress = compress; 16420a586ceaSMark Shellenbaum zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; 1643b24ab676SJeff Bonwick zp->zp_level = level; 1644b24ab676SJeff Bonwick zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa)); 1645b24ab676SJeff Bonwick zp->zp_dedup = dedup; 1646b24ab676SJeff Bonwick zp->zp_dedup_verify = dedup && dedup_verify; 164780901aeaSGeorge Wilson zp->zp_nopwrite = nopwrite; 1648b24ab676SJeff Bonwick } 1649b24ab676SJeff Bonwick 165044cd46caSbillm int 1651fa9e4066Sahrens dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) 1652fa9e4066Sahrens { 1653fa9e4066Sahrens dnode_t *dn; 1654fa9e4066Sahrens int i, err; 1655fa9e4066Sahrens 1656503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 1657ea8dc4b6Seschrock if (err) 1658ea8dc4b6Seschrock return (err); 1659fa9e4066Sahrens /* 1660fa9e4066Sahrens * Sync any current changes before 1661fa9e4066Sahrens * we go trundling through the block pointers. 1662fa9e4066Sahrens */ 1663fa9e4066Sahrens for (i = 0; i < TXG_SIZE; i++) { 1664c543ec06Sahrens if (list_link_active(&dn->dn_dirty_link[i])) 1665fa9e4066Sahrens break; 1666fa9e4066Sahrens } 1667fa9e4066Sahrens if (i != TXG_SIZE) { 1668fa9e4066Sahrens dnode_rele(dn, FTAG); 1669fa9e4066Sahrens txg_wait_synced(dmu_objset_pool(os), 0); 1670503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 1671ea8dc4b6Seschrock if (err) 1672ea8dc4b6Seschrock return (err); 1673fa9e4066Sahrens } 1674fa9e4066Sahrens 1675cdb0ab79Smaybee err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); 1676fa9e4066Sahrens dnode_rele(dn, FTAG); 1677fa9e4066Sahrens 1678fa9e4066Sahrens return (err); 1679fa9e4066Sahrens } 1680fa9e4066Sahrens 1681fa9e4066Sahrens void 1682fa9e4066Sahrens dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) 1683fa9e4066Sahrens { 1684b24ab676SJeff Bonwick dnode_phys_t *dnp; 1685b24ab676SJeff Bonwick 1686fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 1687fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 1688fa9e4066Sahrens 1689b24ab676SJeff Bonwick dnp = dn->dn_phys; 1690b24ab676SJeff Bonwick 1691fa9e4066Sahrens doi->doi_data_block_size = dn->dn_datablksz; 1692fa9e4066Sahrens doi->doi_metadata_block_size = dn->dn_indblkshift ? 1693fa9e4066Sahrens 1ULL << dn->dn_indblkshift : 0; 1694b24ab676SJeff Bonwick doi->doi_type = dn->dn_type; 1695b24ab676SJeff Bonwick doi->doi_bonus_type = dn->dn_bonustype; 1696b24ab676SJeff Bonwick doi->doi_bonus_size = dn->dn_bonuslen; 1697fa9e4066Sahrens doi->doi_indirection = dn->dn_nlevels; 1698fa9e4066Sahrens doi->doi_checksum = dn->dn_checksum; 1699fa9e4066Sahrens doi->doi_compress = dn->dn_compress; 1700b24ab676SJeff Bonwick doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; 1701b24ab676SJeff Bonwick doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz; 1702b24ab676SJeff Bonwick doi->doi_fill_count = 0; 1703b24ab676SJeff Bonwick for (int i = 0; i < dnp->dn_nblkptr; i++) 1704b24ab676SJeff Bonwick doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill; 1705fa9e4066Sahrens 1706fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 1707fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 1708fa9e4066Sahrens } 1709fa9e4066Sahrens 1710fa9e4066Sahrens /* 1711fa9e4066Sahrens * Get information on a DMU object. 1712fa9e4066Sahrens * If doi is NULL, just indicates whether the object exists. 1713fa9e4066Sahrens */ 1714fa9e4066Sahrens int 1715fa9e4066Sahrens dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) 1716fa9e4066Sahrens { 1717ea8dc4b6Seschrock dnode_t *dn; 1718503ad85cSMatthew Ahrens int err = dnode_hold(os, object, FTAG, &dn); 1719fa9e4066Sahrens 1720ea8dc4b6Seschrock if (err) 1721ea8dc4b6Seschrock return (err); 1722fa9e4066Sahrens 1723fa9e4066Sahrens if (doi != NULL) 1724fa9e4066Sahrens dmu_object_info_from_dnode(dn, doi); 1725fa9e4066Sahrens 1726fa9e4066Sahrens dnode_rele(dn, FTAG); 1727fa9e4066Sahrens return (0); 1728fa9e4066Sahrens } 1729fa9e4066Sahrens 1730fa9e4066Sahrens /* 1731fa9e4066Sahrens * As above, but faster; can be used when you have a held dbuf in hand. 1732fa9e4066Sahrens */ 1733fa9e4066Sahrens void 1734744947dcSTom Erickson dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) 1735fa9e4066Sahrens { 1736744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 1737744947dcSTom Erickson 1738744947dcSTom Erickson DB_DNODE_ENTER(db); 1739744947dcSTom Erickson dmu_object_info_from_dnode(DB_DNODE(db), doi); 1740744947dcSTom Erickson DB_DNODE_EXIT(db); 1741fa9e4066Sahrens } 1742fa9e4066Sahrens 1743fa9e4066Sahrens /* 1744fa9e4066Sahrens * Faster still when you only care about the size. 1745fa9e4066Sahrens * This is specifically optimized for zfs_getattr(). 1746fa9e4066Sahrens */ 1747fa9e4066Sahrens void 1748744947dcSTom Erickson dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, 1749744947dcSTom Erickson u_longlong_t *nblk512) 1750fa9e4066Sahrens { 1751744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 1752744947dcSTom Erickson dnode_t *dn; 1753744947dcSTom Erickson 1754744947dcSTom Erickson DB_DNODE_ENTER(db); 1755744947dcSTom Erickson dn = DB_DNODE(db); 1756fa9e4066Sahrens 1757fa9e4066Sahrens *blksize = dn->dn_datablksz; 175899653d4eSeschrock /* add 1 for dnode space */ 175999653d4eSeschrock *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> 176099653d4eSeschrock SPA_MINBLOCKSHIFT) + 1; 1761744947dcSTom Erickson DB_DNODE_EXIT(db); 1762fa9e4066Sahrens } 1763fa9e4066Sahrens 1764fa9e4066Sahrens void 1765fa9e4066Sahrens byteswap_uint64_array(void *vbuf, size_t size) 1766fa9e4066Sahrens { 1767fa9e4066Sahrens uint64_t *buf = vbuf; 1768fa9e4066Sahrens size_t count = size >> 3; 1769fa9e4066Sahrens int i; 1770fa9e4066Sahrens 1771fa9e4066Sahrens ASSERT((size & 7) == 0); 1772fa9e4066Sahrens 1773fa9e4066Sahrens for (i = 0; i < count; i++) 1774fa9e4066Sahrens buf[i] = BSWAP_64(buf[i]); 1775fa9e4066Sahrens } 1776fa9e4066Sahrens 1777fa9e4066Sahrens void 1778fa9e4066Sahrens byteswap_uint32_array(void *vbuf, size_t size) 1779fa9e4066Sahrens { 1780fa9e4066Sahrens uint32_t *buf = vbuf; 1781fa9e4066Sahrens size_t count = size >> 2; 1782fa9e4066Sahrens int i; 1783fa9e4066Sahrens 1784fa9e4066Sahrens ASSERT((size & 3) == 0); 1785fa9e4066Sahrens 1786fa9e4066Sahrens for (i = 0; i < count; i++) 1787fa9e4066Sahrens buf[i] = BSWAP_32(buf[i]); 1788fa9e4066Sahrens } 1789fa9e4066Sahrens 1790fa9e4066Sahrens void 1791fa9e4066Sahrens byteswap_uint16_array(void *vbuf, size_t size) 1792fa9e4066Sahrens { 1793fa9e4066Sahrens uint16_t *buf = vbuf; 1794fa9e4066Sahrens size_t count = size >> 1; 1795fa9e4066Sahrens int i; 1796fa9e4066Sahrens 1797fa9e4066Sahrens ASSERT((size & 1) == 0); 1798fa9e4066Sahrens 1799fa9e4066Sahrens for (i = 0; i < count; i++) 1800fa9e4066Sahrens buf[i] = BSWAP_16(buf[i]); 1801fa9e4066Sahrens } 1802fa9e4066Sahrens 1803fa9e4066Sahrens /* ARGSUSED */ 1804fa9e4066Sahrens void 1805fa9e4066Sahrens byteswap_uint8_array(void *vbuf, size_t size) 1806fa9e4066Sahrens { 1807fa9e4066Sahrens } 1808fa9e4066Sahrens 1809fa9e4066Sahrens void 1810fa9e4066Sahrens dmu_init(void) 1811fa9e4066Sahrens { 18123f9d6ad7SLin Ling zfs_dbgmsg_init(); 1813744947dcSTom Erickson sa_cache_init(); 1814744947dcSTom Erickson xuio_stat_init(); 1815744947dcSTom Erickson dmu_objset_init(); 1816fa9e4066Sahrens dnode_init(); 1817744947dcSTom Erickson dbuf_init(); 18187cbf8b43SRich Morris zfetch_init(); 1819fa94a07fSbrendan l2arc_init(); 1820ce636f8bSMatthew Ahrens arc_init(); 1821fa9e4066Sahrens } 1822fa9e4066Sahrens 1823fa9e4066Sahrens void 1824fa9e4066Sahrens dmu_fini(void) 1825fa9e4066Sahrens { 1826fa9e4066Sahrens arc_fini(); 1827ce636f8bSMatthew Ahrens l2arc_fini(); 18287cbf8b43SRich Morris zfetch_fini(); 1829fa9e4066Sahrens dbuf_fini(); 1830744947dcSTom Erickson dnode_fini(); 1831744947dcSTom Erickson dmu_objset_fini(); 1832c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_fini(); 18330a586ceaSMark Shellenbaum sa_cache_fini(); 18343f9d6ad7SLin Ling zfs_dbgmsg_fini(); 1835fa9e4066Sahrens } 1836