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. 237802d7bfSMatthew Ahrens * Copyright (c) 2011, 2014 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. */ 27b8289d24SDaniil Lunev /* Copyright (c) 2014, Nexenta Systems, Inc. All rights reserved. */ 28aad02571SSaso Kiselkov 29fa9e4066Sahrens #include <sys/dmu.h> 30fa9e4066Sahrens #include <sys/dmu_impl.h> 31fa9e4066Sahrens #include <sys/dmu_tx.h> 32fa9e4066Sahrens #include <sys/dbuf.h> 33fa9e4066Sahrens #include <sys/dnode.h> 34fa9e4066Sahrens #include <sys/zfs_context.h> 35fa9e4066Sahrens #include <sys/dmu_objset.h> 36fa9e4066Sahrens #include <sys/dmu_traverse.h> 37fa9e4066Sahrens #include <sys/dsl_dataset.h> 38fa9e4066Sahrens #include <sys/dsl_dir.h> 39fa9e4066Sahrens #include <sys/dsl_pool.h> 401d452cf5Sahrens #include <sys/dsl_synctask.h> 41a2eea2e1Sahrens #include <sys/dsl_prop.h> 42fa9e4066Sahrens #include <sys/dmu_zfetch.h> 43fa9e4066Sahrens #include <sys/zfs_ioctl.h> 44fa9e4066Sahrens #include <sys/zap.h> 45ea8dc4b6Seschrock #include <sys/zio_checksum.h> 4680901aeaSGeorge Wilson #include <sys/zio_compress.h> 470a586ceaSMark Shellenbaum #include <sys/sa.h> 48b8289d24SDaniil Lunev #include <sys/zfeature.h> 4944eda4d7Smaybee #ifdef _KERNEL 5044eda4d7Smaybee #include <sys/vmsystm.h> 510fab61baSJonathan W Adams #include <sys/zfs_znode.h> 5244eda4d7Smaybee #endif 53fa9e4066Sahrens 5480901aeaSGeorge Wilson /* 5580901aeaSGeorge Wilson * Enable/disable nopwrite feature. 5680901aeaSGeorge Wilson */ 5780901aeaSGeorge Wilson int zfs_nopwrite_enabled = 1; 5880901aeaSGeorge Wilson 59fa9e4066Sahrens const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { 60ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "unallocated" }, 61ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "object directory" }, 62ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "object array" }, 63ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "packed nvlist" }, 64ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" }, 65ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "bpobj" }, 66ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "bpobj header" }, 67ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "SPA space map header" }, 68ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "SPA space map" }, 69ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" }, 70ad135b5dSChristopher Siden { DMU_BSWAP_DNODE, TRUE, "DMU dnode" }, 71ad135b5dSChristopher Siden { DMU_BSWAP_OBJSET, TRUE, "DMU objset" }, 72ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "DSL directory" }, 73ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"}, 74ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" }, 75ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL props" }, 76ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "DSL dataset" }, 77ad135b5dSChristopher Siden { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" }, 78ad135b5dSChristopher Siden { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" }, 79ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" }, 80ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS directory" }, 81ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS master node" }, 82ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" }, 83ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "zvol object" }, 84ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "zvol prop" }, 85ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "other uint8[]" }, 86ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, FALSE, "other uint64[]" }, 87ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "other ZAP" }, 88ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "persistent error log" }, 89ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "SPA history" }, 90ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" }, 91ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "Pool properties" }, 92ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL permissions" }, 93ad135b5dSChristopher Siden { DMU_BSWAP_ACL, TRUE, "ZFS ACL" }, 94ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" }, 95ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "FUID table" }, 96ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "FUID table size" }, 97ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"}, 98ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "scan work queue" }, 99ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" }, 100ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" }, 101ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"}, 102ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" }, 103ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DDT statistics" }, 104ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, TRUE, "System attributes" }, 105ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "SA master node" }, 106ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "SA attr registration" }, 107ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" }, 108ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "scan translations" }, 109ad135b5dSChristopher Siden { DMU_BSWAP_UINT8, FALSE, "deduplicated block" }, 110ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" }, 111ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" }, 112ad135b5dSChristopher Siden { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" }, 113ad135b5dSChristopher Siden { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" } 114ad135b5dSChristopher Siden }; 115ad135b5dSChristopher Siden 116ad135b5dSChristopher Siden const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { 117ad135b5dSChristopher Siden { byteswap_uint8_array, "uint8" }, 118ad135b5dSChristopher Siden { byteswap_uint16_array, "uint16" }, 119ad135b5dSChristopher Siden { byteswap_uint32_array, "uint32" }, 120ad135b5dSChristopher Siden { byteswap_uint64_array, "uint64" }, 121ad135b5dSChristopher Siden { zap_byteswap, "zap" }, 122ad135b5dSChristopher Siden { dnode_buf_byteswap, "dnode" }, 123ad135b5dSChristopher Siden { dmu_objset_byteswap, "objset" }, 124ad135b5dSChristopher Siden { zfs_znode_byteswap, "znode" }, 125ad135b5dSChristopher Siden { zfs_oldacl_byteswap, "oldacl" }, 126ad135b5dSChristopher Siden { zfs_acl_byteswap, "acl" } 1273f9d6ad7SLin Ling }; 128fa9e4066Sahrens 129fa9e4066Sahrens int 1305d7b4d43SMatthew Ahrens dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, 1315d7b4d43SMatthew Ahrens void *tag, dmu_buf_t **dbp) 132fa9e4066Sahrens { 133fa9e4066Sahrens dnode_t *dn; 134fa9e4066Sahrens uint64_t blkid; 135fa9e4066Sahrens dmu_buf_impl_t *db; 136ea8dc4b6Seschrock int err; 137fa9e4066Sahrens 138503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 139ea8dc4b6Seschrock if (err) 140ea8dc4b6Seschrock return (err); 141*a2cdcdd2SPaul Dagnelie blkid = dbuf_whichblock(dn, 0, offset); 142fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 143ea8dc4b6Seschrock db = dbuf_hold(dn, blkid, tag); 144fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 1455d7b4d43SMatthew Ahrens dnode_rele(dn, FTAG); 1465d7b4d43SMatthew Ahrens 147ea8dc4b6Seschrock if (db == NULL) { 1485d7b4d43SMatthew Ahrens *dbp = NULL; 1495d7b4d43SMatthew Ahrens return (SET_ERROR(EIO)); 1505d7b4d43SMatthew Ahrens } 1515d7b4d43SMatthew Ahrens 1525d7b4d43SMatthew Ahrens *dbp = &db->db; 1535d7b4d43SMatthew Ahrens return (err); 1545d7b4d43SMatthew Ahrens } 1555d7b4d43SMatthew Ahrens 1565d7b4d43SMatthew Ahrens int 1575d7b4d43SMatthew Ahrens dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 1585d7b4d43SMatthew Ahrens void *tag, dmu_buf_t **dbp, int flags) 1595d7b4d43SMatthew Ahrens { 1605d7b4d43SMatthew Ahrens int err; 1615d7b4d43SMatthew Ahrens int db_flags = DB_RF_CANFAIL; 1625d7b4d43SMatthew Ahrens 1635d7b4d43SMatthew Ahrens if (flags & DMU_READ_NO_PREFETCH) 1645d7b4d43SMatthew Ahrens db_flags |= DB_RF_NOPREFETCH; 1655d7b4d43SMatthew Ahrens 1665d7b4d43SMatthew Ahrens err = dmu_buf_hold_noread(os, object, offset, tag, dbp); 1675d7b4d43SMatthew Ahrens if (err == 0) { 1685d7b4d43SMatthew Ahrens dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); 16947cb52daSJeff Bonwick err = dbuf_read(db, NULL, db_flags); 1705d7b4d43SMatthew Ahrens if (err != 0) { 171ea8dc4b6Seschrock dbuf_rele(db, tag); 1725d7b4d43SMatthew Ahrens *dbp = NULL; 173ea8dc4b6Seschrock } 174ea8dc4b6Seschrock } 175fa9e4066Sahrens 176ea8dc4b6Seschrock return (err); 177fa9e4066Sahrens } 178fa9e4066Sahrens 179fa9e4066Sahrens int 180fa9e4066Sahrens dmu_bonus_max(void) 181fa9e4066Sahrens { 182fa9e4066Sahrens return (DN_MAX_BONUSLEN); 183fa9e4066Sahrens } 184fa9e4066Sahrens 1851934e92fSmaybee int 186744947dcSTom Erickson dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) 1871934e92fSmaybee { 188744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 189744947dcSTom Erickson dnode_t *dn; 190744947dcSTom Erickson int error; 1911934e92fSmaybee 192744947dcSTom Erickson DB_DNODE_ENTER(db); 193744947dcSTom Erickson dn = DB_DNODE(db); 194744947dcSTom Erickson 195744947dcSTom Erickson if (dn->dn_bonus != db) { 196be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 197744947dcSTom Erickson } else if (newsize < 0 || newsize > db_fake->db_size) { 198be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 199744947dcSTom Erickson } else { 200744947dcSTom Erickson dnode_setbonuslen(dn, newsize, tx); 201744947dcSTom Erickson error = 0; 202744947dcSTom Erickson } 203744947dcSTom Erickson 204744947dcSTom Erickson DB_DNODE_EXIT(db); 205744947dcSTom Erickson return (error); 2061934e92fSmaybee } 2071934e92fSmaybee 2080a586ceaSMark Shellenbaum int 209744947dcSTom Erickson dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) 2100a586ceaSMark Shellenbaum { 211744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 212744947dcSTom Erickson dnode_t *dn; 213744947dcSTom Erickson int error; 2140a586ceaSMark Shellenbaum 215744947dcSTom Erickson DB_DNODE_ENTER(db); 216744947dcSTom Erickson dn = DB_DNODE(db); 217744947dcSTom Erickson 218ad135b5dSChristopher Siden if (!DMU_OT_IS_VALID(type)) { 219be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 220744947dcSTom Erickson } else if (dn->dn_bonus != db) { 221be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 222744947dcSTom Erickson } else { 223744947dcSTom Erickson dnode_setbonus_type(dn, type, tx); 224744947dcSTom Erickson error = 0; 225744947dcSTom Erickson } 2260a586ceaSMark Shellenbaum 227744947dcSTom Erickson DB_DNODE_EXIT(db); 228744947dcSTom Erickson return (error); 229744947dcSTom Erickson } 2300a586ceaSMark Shellenbaum 231744947dcSTom Erickson dmu_object_type_t 232744947dcSTom Erickson dmu_get_bonustype(dmu_buf_t *db_fake) 233744947dcSTom Erickson { 234744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 235744947dcSTom Erickson dnode_t *dn; 236744947dcSTom Erickson dmu_object_type_t type; 237744947dcSTom Erickson 238744947dcSTom Erickson DB_DNODE_ENTER(db); 239744947dcSTom Erickson dn = DB_DNODE(db); 240744947dcSTom Erickson type = dn->dn_bonustype; 241744947dcSTom Erickson DB_DNODE_EXIT(db); 242744947dcSTom Erickson 243744947dcSTom Erickson return (type); 2440a586ceaSMark Shellenbaum } 2450a586ceaSMark Shellenbaum 2460a586ceaSMark Shellenbaum int 2470a586ceaSMark Shellenbaum dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) 2480a586ceaSMark Shellenbaum { 2490a586ceaSMark Shellenbaum dnode_t *dn; 2500a586ceaSMark Shellenbaum int error; 2510a586ceaSMark Shellenbaum 2520a586ceaSMark Shellenbaum error = dnode_hold(os, object, FTAG, &dn); 2530a586ceaSMark Shellenbaum dbuf_rm_spill(dn, tx); 25406e0070dSMark Shellenbaum rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 25506e0070dSMark Shellenbaum dnode_rm_spill(dn, tx); 25606e0070dSMark Shellenbaum rw_exit(&dn->dn_struct_rwlock); 2570a586ceaSMark Shellenbaum dnode_rele(dn, FTAG); 2580a586ceaSMark Shellenbaum return (error); 2590a586ceaSMark Shellenbaum } 2600a586ceaSMark Shellenbaum 261fa9e4066Sahrens /* 262ea8dc4b6Seschrock * returns ENOENT, EIO, or 0. 263fa9e4066Sahrens */ 264ea8dc4b6Seschrock int 265ea8dc4b6Seschrock dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) 266fa9e4066Sahrens { 267ea8dc4b6Seschrock dnode_t *dn; 268fa9e4066Sahrens dmu_buf_impl_t *db; 2691934e92fSmaybee int error; 270fa9e4066Sahrens 271503ad85cSMatthew Ahrens error = dnode_hold(os, object, FTAG, &dn); 2721934e92fSmaybee if (error) 2731934e92fSmaybee return (error); 274fa9e4066Sahrens 275ea8dc4b6Seschrock rw_enter(&dn->dn_struct_rwlock, RW_READER); 276ea8dc4b6Seschrock if (dn->dn_bonus == NULL) { 277fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 278ea8dc4b6Seschrock rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 279ea8dc4b6Seschrock if (dn->dn_bonus == NULL) 2801934e92fSmaybee dbuf_create_bonus(dn); 281fa9e4066Sahrens } 282ea8dc4b6Seschrock db = dn->dn_bonus; 2831934e92fSmaybee 2841934e92fSmaybee /* as long as the bonus buf is held, the dnode will be held */ 285744947dcSTom Erickson if (refcount_add(&db->db_holds, tag) == 1) { 2861934e92fSmaybee VERIFY(dnode_add_ref(dn, db)); 287640c1670SJosef 'Jeff' Sipek atomic_inc_32(&dn->dn_dbufs_count); 288744947dcSTom Erickson } 289744947dcSTom Erickson 290744947dcSTom Erickson /* 291744947dcSTom Erickson * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's 292744947dcSTom Erickson * hold and incrementing the dbuf count to ensure that dnode_move() sees 293744947dcSTom Erickson * a dnode hold for every dbuf. 294744947dcSTom Erickson */ 295744947dcSTom Erickson rw_exit(&dn->dn_struct_rwlock); 2961934e92fSmaybee 297fa9e4066Sahrens dnode_rele(dn, FTAG); 298ea8dc4b6Seschrock 29947cb52daSJeff Bonwick VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); 300ea8dc4b6Seschrock 301ea8dc4b6Seschrock *dbp = &db->db; 302ea8dc4b6Seschrock return (0); 303fa9e4066Sahrens } 304fa9e4066Sahrens 3050a586ceaSMark Shellenbaum /* 3060a586ceaSMark Shellenbaum * returns ENOENT, EIO, or 0. 3070a586ceaSMark Shellenbaum * 3080a586ceaSMark Shellenbaum * This interface will allocate a blank spill dbuf when a spill blk 3090a586ceaSMark Shellenbaum * doesn't already exist on the dnode. 3100a586ceaSMark Shellenbaum * 3110a586ceaSMark Shellenbaum * if you only want to find an already existing spill db, then 3120a586ceaSMark Shellenbaum * dmu_spill_hold_existing() should be used. 3130a586ceaSMark Shellenbaum */ 3140a586ceaSMark Shellenbaum int 3150a586ceaSMark Shellenbaum dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) 3160a586ceaSMark Shellenbaum { 3170a586ceaSMark Shellenbaum dmu_buf_impl_t *db = NULL; 3180a586ceaSMark Shellenbaum int err; 3190a586ceaSMark Shellenbaum 3200a586ceaSMark Shellenbaum if ((flags & DB_RF_HAVESTRUCT) == 0) 3210a586ceaSMark Shellenbaum rw_enter(&dn->dn_struct_rwlock, RW_READER); 3220a586ceaSMark Shellenbaum 3230a586ceaSMark Shellenbaum db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); 3240a586ceaSMark Shellenbaum 3250a586ceaSMark Shellenbaum if ((flags & DB_RF_HAVESTRUCT) == 0) 3260a586ceaSMark Shellenbaum rw_exit(&dn->dn_struct_rwlock); 3270a586ceaSMark Shellenbaum 3280a586ceaSMark Shellenbaum ASSERT(db != NULL); 3291d8ccc7bSMark Shellenbaum err = dbuf_read(db, NULL, flags); 3301d8ccc7bSMark Shellenbaum if (err == 0) 3311d8ccc7bSMark Shellenbaum *dbp = &db->db; 3321d8ccc7bSMark Shellenbaum else 3331d8ccc7bSMark Shellenbaum dbuf_rele(db, tag); 3340a586ceaSMark Shellenbaum return (err); 3350a586ceaSMark Shellenbaum } 3360a586ceaSMark Shellenbaum 3370a586ceaSMark Shellenbaum int 3380a586ceaSMark Shellenbaum dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 3390a586ceaSMark Shellenbaum { 340744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; 341744947dcSTom Erickson dnode_t *dn; 3420a586ceaSMark Shellenbaum int err; 3430a586ceaSMark Shellenbaum 344744947dcSTom Erickson DB_DNODE_ENTER(db); 345744947dcSTom Erickson dn = DB_DNODE(db); 346744947dcSTom Erickson 347744947dcSTom Erickson if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { 348be6fd75aSMatthew Ahrens err = SET_ERROR(EINVAL); 349744947dcSTom Erickson } else { 350744947dcSTom Erickson rw_enter(&dn->dn_struct_rwlock, RW_READER); 351744947dcSTom Erickson 352744947dcSTom Erickson if (!dn->dn_have_spill) { 353be6fd75aSMatthew Ahrens err = SET_ERROR(ENOENT); 354744947dcSTom Erickson } else { 355744947dcSTom Erickson err = dmu_spill_hold_by_dnode(dn, 356744947dcSTom Erickson DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); 357744947dcSTom Erickson } 3580a586ceaSMark Shellenbaum 3590a586ceaSMark Shellenbaum rw_exit(&dn->dn_struct_rwlock); 3600a586ceaSMark Shellenbaum } 361744947dcSTom Erickson 362744947dcSTom Erickson DB_DNODE_EXIT(db); 3630a586ceaSMark Shellenbaum return (err); 3640a586ceaSMark Shellenbaum } 3650a586ceaSMark Shellenbaum 3660a586ceaSMark Shellenbaum int 3670a586ceaSMark Shellenbaum dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 3680a586ceaSMark Shellenbaum { 369744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; 370744947dcSTom Erickson dnode_t *dn; 371744947dcSTom Erickson int err; 372744947dcSTom Erickson 373744947dcSTom Erickson DB_DNODE_ENTER(db); 374744947dcSTom Erickson dn = DB_DNODE(db); 375744947dcSTom Erickson err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); 376744947dcSTom Erickson DB_DNODE_EXIT(db); 377744947dcSTom Erickson 378744947dcSTom Erickson return (err); 3790a586ceaSMark Shellenbaum } 3800a586ceaSMark Shellenbaum 38113506d1eSmaybee /* 38213506d1eSmaybee * Note: longer-term, we should modify all of the dmu_buf_*() interfaces 38313506d1eSmaybee * to take a held dnode rather than <os, object> -- the lookup is wasteful, 38413506d1eSmaybee * and can induce severe lock contention when writing to several files 38513506d1eSmaybee * whose dnodes are in the same block. 38613506d1eSmaybee */ 38713506d1eSmaybee static int 3887bfdf011SNeil Perrin dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, 3897bfdf011SNeil Perrin int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) 390fa9e4066Sahrens { 391fa9e4066Sahrens dmu_buf_t **dbp; 392fa9e4066Sahrens uint64_t blkid, nblks, i; 3937bfdf011SNeil Perrin uint32_t dbuf_flags; 394ea8dc4b6Seschrock int err; 395ea8dc4b6Seschrock zio_t *zio; 396ea8dc4b6Seschrock 397ea8dc4b6Seschrock ASSERT(length <= DMU_MAX_ACCESS); 398fa9e4066Sahrens 399c87b8fc5SMark J Musante dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT; 4007bfdf011SNeil Perrin if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz) 4017bfdf011SNeil Perrin dbuf_flags |= DB_RF_NOPREFETCH; 402ea8dc4b6Seschrock 403fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 404fa9e4066Sahrens if (dn->dn_datablkshift) { 405fa9e4066Sahrens int blkshift = dn->dn_datablkshift; 406fa9e4066Sahrens nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - 407b1b8ab34Slling P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; 408fa9e4066Sahrens } else { 4090125049cSahrens if (offset + length > dn->dn_datablksz) { 4100125049cSahrens zfs_panic_recover("zfs: accessing past end of object " 4110125049cSahrens "%llx/%llx (size=%u access=%llu+%llu)", 4120125049cSahrens (longlong_t)dn->dn_objset-> 4130125049cSahrens os_dsl_dataset->ds_object, 4140125049cSahrens (longlong_t)dn->dn_object, dn->dn_datablksz, 4150125049cSahrens (longlong_t)offset, (longlong_t)length); 416c87b8fc5SMark J Musante rw_exit(&dn->dn_struct_rwlock); 417be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 4180125049cSahrens } 419fa9e4066Sahrens nblks = 1; 420fa9e4066Sahrens } 421ea8dc4b6Seschrock dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); 422fa9e4066Sahrens 423e14bb325SJeff Bonwick zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); 424*a2cdcdd2SPaul Dagnelie blkid = dbuf_whichblock(dn, 0, offset); 425fa9e4066Sahrens for (i = 0; i < nblks; i++) { 426ea8dc4b6Seschrock dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); 427ea8dc4b6Seschrock if (db == NULL) { 428ea8dc4b6Seschrock rw_exit(&dn->dn_struct_rwlock); 429ea8dc4b6Seschrock dmu_buf_rele_array(dbp, nblks, tag); 430ea8dc4b6Seschrock zio_nowait(zio); 431be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 432ea8dc4b6Seschrock } 433ea8dc4b6Seschrock /* initiate async i/o */ 43413506d1eSmaybee if (read) { 4357bfdf011SNeil Perrin (void) dbuf_read(db, zio, dbuf_flags); 436ea8dc4b6Seschrock } 437ea8dc4b6Seschrock dbp[i] = &db->db; 438fa9e4066Sahrens } 439fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 440fa9e4066Sahrens 441ea8dc4b6Seschrock /* wait for async i/o */ 442ea8dc4b6Seschrock err = zio_wait(zio); 443ea8dc4b6Seschrock if (err) { 444ea8dc4b6Seschrock dmu_buf_rele_array(dbp, nblks, tag); 445ea8dc4b6Seschrock return (err); 446ea8dc4b6Seschrock } 447fa9e4066Sahrens 448ea8dc4b6Seschrock /* wait for other io to complete */ 449ea8dc4b6Seschrock if (read) { 450ea8dc4b6Seschrock for (i = 0; i < nblks; i++) { 451ea8dc4b6Seschrock dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; 452ea8dc4b6Seschrock mutex_enter(&db->db_mtx); 453ea8dc4b6Seschrock while (db->db_state == DB_READ || 454ea8dc4b6Seschrock db->db_state == DB_FILL) 455ea8dc4b6Seschrock cv_wait(&db->db_changed, &db->db_mtx); 456ea8dc4b6Seschrock if (db->db_state == DB_UNCACHED) 457be6fd75aSMatthew Ahrens err = SET_ERROR(EIO); 458ea8dc4b6Seschrock mutex_exit(&db->db_mtx); 459ea8dc4b6Seschrock if (err) { 460ea8dc4b6Seschrock dmu_buf_rele_array(dbp, nblks, tag); 461ea8dc4b6Seschrock return (err); 462ea8dc4b6Seschrock } 463ea8dc4b6Seschrock } 464ea8dc4b6Seschrock } 465fa9e4066Sahrens 466ea8dc4b6Seschrock *numbufsp = nblks; 467ea8dc4b6Seschrock *dbpp = dbp; 468ea8dc4b6Seschrock return (0); 469fa9e4066Sahrens } 470fa9e4066Sahrens 471a2eea2e1Sahrens static int 47213506d1eSmaybee dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, 47313506d1eSmaybee uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 47413506d1eSmaybee { 47513506d1eSmaybee dnode_t *dn; 47613506d1eSmaybee int err; 47713506d1eSmaybee 478503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 47913506d1eSmaybee if (err) 48013506d1eSmaybee return (err); 48113506d1eSmaybee 48213506d1eSmaybee err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 4837bfdf011SNeil Perrin numbufsp, dbpp, DMU_READ_PREFETCH); 48413506d1eSmaybee 48513506d1eSmaybee dnode_rele(dn, FTAG); 48613506d1eSmaybee 48713506d1eSmaybee return (err); 48813506d1eSmaybee } 48913506d1eSmaybee 49013506d1eSmaybee int 491744947dcSTom Erickson dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, 49213506d1eSmaybee uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 49313506d1eSmaybee { 494744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 495744947dcSTom Erickson dnode_t *dn; 49613506d1eSmaybee int err; 49713506d1eSmaybee 498744947dcSTom Erickson DB_DNODE_ENTER(db); 499744947dcSTom Erickson dn = DB_DNODE(db); 50013506d1eSmaybee err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 5017bfdf011SNeil Perrin numbufsp, dbpp, DMU_READ_PREFETCH); 502744947dcSTom Erickson DB_DNODE_EXIT(db); 50313506d1eSmaybee 50413506d1eSmaybee return (err); 50513506d1eSmaybee } 50613506d1eSmaybee 507fa9e4066Sahrens void 508ea8dc4b6Seschrock dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) 509fa9e4066Sahrens { 510fa9e4066Sahrens int i; 511fa9e4066Sahrens dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; 512fa9e4066Sahrens 513fa9e4066Sahrens if (numbufs == 0) 514fa9e4066Sahrens return; 515fa9e4066Sahrens 516ea8dc4b6Seschrock for (i = 0; i < numbufs; i++) { 517ea8dc4b6Seschrock if (dbp[i]) 518ea8dc4b6Seschrock dbuf_rele(dbp[i], tag); 519ea8dc4b6Seschrock } 520fa9e4066Sahrens 521fa9e4066Sahrens kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); 522fa9e4066Sahrens } 523fa9e4066Sahrens 52469962b56SMatthew Ahrens /* 525*a2cdcdd2SPaul Dagnelie * Issue prefetch i/os for the given blocks. If level is greater than 0, the 526*a2cdcdd2SPaul Dagnelie * indirect blocks prefeteched will be those that point to the blocks containing 527*a2cdcdd2SPaul Dagnelie * the data starting at offset, and continuing to offset + len. 52869962b56SMatthew Ahrens * 529*a2cdcdd2SPaul Dagnelie * Note that if the indirect blocks above the blocks being prefetched are not in 530*a2cdcdd2SPaul Dagnelie * cache, they will be asychronously read in. 53169962b56SMatthew Ahrens */ 532fa9e4066Sahrens void 533*a2cdcdd2SPaul Dagnelie dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, 534*a2cdcdd2SPaul Dagnelie uint64_t len, zio_priority_t pri) 535fa9e4066Sahrens { 536fa9e4066Sahrens dnode_t *dn; 537fa9e4066Sahrens uint64_t blkid; 53869962b56SMatthew Ahrens int nblks, err; 539fa9e4066Sahrens 540416e0cd8Sek if (zfs_prefetch_disable) 541416e0cd8Sek return; 542416e0cd8Sek 543fa9e4066Sahrens if (len == 0) { /* they're interested in the bonus buffer */ 544744947dcSTom Erickson dn = DMU_META_DNODE(os); 545fa9e4066Sahrens 546fa9e4066Sahrens if (object == 0 || object >= DN_MAX_OBJECT) 547fa9e4066Sahrens return; 548fa9e4066Sahrens 549fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 550*a2cdcdd2SPaul Dagnelie blkid = dbuf_whichblock(dn, level, 551*a2cdcdd2SPaul Dagnelie object * sizeof (dnode_phys_t)); 552*a2cdcdd2SPaul Dagnelie dbuf_prefetch(dn, level, blkid, pri, 0); 553fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 554fa9e4066Sahrens return; 555fa9e4066Sahrens } 556fa9e4066Sahrens 557fa9e4066Sahrens /* 558fa9e4066Sahrens * XXX - Note, if the dnode for the requested object is not 559fa9e4066Sahrens * already cached, we will do a *synchronous* read in the 560fa9e4066Sahrens * dnode_hold() call. The same is true for any indirects. 561fa9e4066Sahrens */ 562503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 563ea8dc4b6Seschrock if (err != 0) 564fa9e4066Sahrens return; 565fa9e4066Sahrens 566fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 567*a2cdcdd2SPaul Dagnelie /* 568*a2cdcdd2SPaul Dagnelie * offset + len - 1 is the last byte we want to prefetch for, and offset 569*a2cdcdd2SPaul Dagnelie * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the 570*a2cdcdd2SPaul Dagnelie * last block we want to prefetch, and dbuf_whichblock(dn, level, 571*a2cdcdd2SPaul Dagnelie * offset) is the first. Then the number we need to prefetch is the 572*a2cdcdd2SPaul Dagnelie * last - first + 1. 573*a2cdcdd2SPaul Dagnelie */ 574*a2cdcdd2SPaul Dagnelie if (level > 0 || dn->dn_datablkshift != 0) { 575*a2cdcdd2SPaul Dagnelie nblks = dbuf_whichblock(dn, level, offset + len - 1) - 576*a2cdcdd2SPaul Dagnelie dbuf_whichblock(dn, level, offset) + 1; 577fa9e4066Sahrens } else { 578fa9e4066Sahrens nblks = (offset < dn->dn_datablksz); 579fa9e4066Sahrens } 580fa9e4066Sahrens 581fa9e4066Sahrens if (nblks != 0) { 582*a2cdcdd2SPaul Dagnelie blkid = dbuf_whichblock(dn, level, offset); 58369962b56SMatthew Ahrens for (int i = 0; i < nblks; i++) 584*a2cdcdd2SPaul Dagnelie dbuf_prefetch(dn, level, blkid + i, pri, 0); 585fa9e4066Sahrens } 586fa9e4066Sahrens 587fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 588fa9e4066Sahrens 589fa9e4066Sahrens dnode_rele(dn, FTAG); 590fa9e4066Sahrens } 591fa9e4066Sahrens 59276256205SMark Maybee /* 59376256205SMark Maybee * Get the next "chunk" of file data to free. We traverse the file from 59476256205SMark Maybee * the end so that the file gets shorter over time (if we crashes in the 59576256205SMark Maybee * middle, this will leave us in a better state). We find allocated file 59676256205SMark Maybee * data by simply searching the allocated level 1 indirects. 597713d6c20SMatthew Ahrens * 598713d6c20SMatthew Ahrens * On input, *start should be the first offset that does not need to be 599713d6c20SMatthew Ahrens * freed (e.g. "offset + length"). On return, *start will be the first 600713d6c20SMatthew Ahrens * offset that should be freed. 60176256205SMark Maybee */ 602cdb0ab79Smaybee static int 603713d6c20SMatthew Ahrens get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum) 604cdb0ab79Smaybee { 605713d6c20SMatthew Ahrens uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); 606713d6c20SMatthew Ahrens /* bytes of data covered by a level-1 indirect block */ 60776256205SMark Maybee uint64_t iblkrange = 6081c8564a7SMark Maybee dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); 609cdb0ab79Smaybee 610713d6c20SMatthew Ahrens ASSERT3U(minimum, <=, *start); 611cdb0ab79Smaybee 612713d6c20SMatthew Ahrens if (*start - minimum <= iblkrange * maxblks) { 613713d6c20SMatthew Ahrens *start = minimum; 614cdb0ab79Smaybee return (0); 615cdb0ab79Smaybee } 61676256205SMark Maybee ASSERT(ISP2(iblkrange)); 617cdb0ab79Smaybee 618713d6c20SMatthew Ahrens for (uint64_t blks = 0; *start > minimum && blks < maxblks; blks++) { 6191c8564a7SMark Maybee int err; 620cdb0ab79Smaybee 621713d6c20SMatthew Ahrens /* 622713d6c20SMatthew Ahrens * dnode_next_offset(BACKWARDS) will find an allocated L1 623713d6c20SMatthew Ahrens * indirect block at or before the input offset. We must 624713d6c20SMatthew Ahrens * decrement *start so that it is at the end of the region 625713d6c20SMatthew Ahrens * to search. 626713d6c20SMatthew Ahrens */ 627713d6c20SMatthew Ahrens (*start)--; 628cdb0ab79Smaybee err = dnode_next_offset(dn, 62976256205SMark Maybee DNODE_FIND_BACKWARDS, start, 2, 1, 0); 630cdb0ab79Smaybee 631713d6c20SMatthew Ahrens /* if there are no indirect blocks before start, we are done */ 63276256205SMark Maybee if (err == ESRCH) { 633713d6c20SMatthew Ahrens *start = minimum; 634713d6c20SMatthew Ahrens break; 635713d6c20SMatthew Ahrens } else if (err != 0) { 636cdb0ab79Smaybee return (err); 63776256205SMark Maybee } 63876256205SMark Maybee 639713d6c20SMatthew Ahrens /* set start to the beginning of this L1 indirect */ 64076256205SMark Maybee *start = P2ALIGN(*start, iblkrange); 641cdb0ab79Smaybee } 642713d6c20SMatthew Ahrens if (*start < minimum) 643713d6c20SMatthew Ahrens *start = minimum; 644cdb0ab79Smaybee return (0); 645cdb0ab79Smaybee } 646cdb0ab79Smaybee 647cdb0ab79Smaybee static int 648cdb0ab79Smaybee dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, 649713d6c20SMatthew Ahrens uint64_t length) 650cdb0ab79Smaybee { 651713d6c20SMatthew Ahrens uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; 652713d6c20SMatthew Ahrens int err; 653713d6c20SMatthew Ahrens 654713d6c20SMatthew Ahrens if (offset >= object_size) 655cdb0ab79Smaybee return (0); 656cdb0ab79Smaybee 657713d6c20SMatthew Ahrens if (length == DMU_OBJECT_END || offset + length > object_size) 658713d6c20SMatthew Ahrens length = object_size - offset; 659713d6c20SMatthew Ahrens 660713d6c20SMatthew Ahrens while (length != 0) { 661713d6c20SMatthew Ahrens uint64_t chunk_end, chunk_begin; 662713d6c20SMatthew Ahrens 663713d6c20SMatthew Ahrens chunk_end = chunk_begin = offset + length; 664713d6c20SMatthew Ahrens 665713d6c20SMatthew Ahrens /* move chunk_begin backwards to the beginning of this chunk */ 666713d6c20SMatthew Ahrens err = get_next_chunk(dn, &chunk_begin, offset); 667cdb0ab79Smaybee if (err) 668cdb0ab79Smaybee return (err); 669713d6c20SMatthew Ahrens ASSERT3U(chunk_begin, >=, offset); 670713d6c20SMatthew Ahrens ASSERT3U(chunk_begin, <=, chunk_end); 671cdb0ab79Smaybee 672713d6c20SMatthew Ahrens dmu_tx_t *tx = dmu_tx_create(os); 673713d6c20SMatthew Ahrens dmu_tx_hold_free(tx, dn->dn_object, 674713d6c20SMatthew Ahrens chunk_begin, chunk_end - chunk_begin); 6754bb73804SMatthew Ahrens 6764bb73804SMatthew Ahrens /* 6774bb73804SMatthew Ahrens * Mark this transaction as typically resulting in a net 6784bb73804SMatthew Ahrens * reduction in space used. 6794bb73804SMatthew Ahrens */ 6804bb73804SMatthew Ahrens dmu_tx_mark_netfree(tx); 681cdb0ab79Smaybee err = dmu_tx_assign(tx, TXG_WAIT); 682cdb0ab79Smaybee if (err) { 683cdb0ab79Smaybee dmu_tx_abort(tx); 684cdb0ab79Smaybee return (err); 685cdb0ab79Smaybee } 686713d6c20SMatthew Ahrens dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx); 687cdb0ab79Smaybee dmu_tx_commit(tx); 688713d6c20SMatthew Ahrens 689713d6c20SMatthew Ahrens length -= chunk_end - chunk_begin; 690cdb0ab79Smaybee } 691cdb0ab79Smaybee return (0); 692cdb0ab79Smaybee } 693cdb0ab79Smaybee 694cdb0ab79Smaybee int 695cdb0ab79Smaybee dmu_free_long_range(objset_t *os, uint64_t object, 696cdb0ab79Smaybee uint64_t offset, uint64_t length) 697cdb0ab79Smaybee { 698cdb0ab79Smaybee dnode_t *dn; 699cdb0ab79Smaybee int err; 700cdb0ab79Smaybee 701503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 702cdb0ab79Smaybee if (err != 0) 703cdb0ab79Smaybee return (err); 704713d6c20SMatthew Ahrens err = dmu_free_long_range_impl(os, dn, offset, length); 7055253393bSMatthew Ahrens 7065253393bSMatthew Ahrens /* 7075253393bSMatthew Ahrens * It is important to zero out the maxblkid when freeing the entire 7085253393bSMatthew Ahrens * file, so that (a) subsequent calls to dmu_free_long_range_impl() 7095253393bSMatthew Ahrens * will take the fast path, and (b) dnode_reallocate() can verify 7105253393bSMatthew Ahrens * that the entire file has been freed. 7115253393bSMatthew Ahrens */ 71243466aaeSMax Grossman if (err == 0 && offset == 0 && length == DMU_OBJECT_END) 7135253393bSMatthew Ahrens dn->dn_maxblkid = 0; 7145253393bSMatthew Ahrens 715cdb0ab79Smaybee dnode_rele(dn, FTAG); 716cdb0ab79Smaybee return (err); 717cdb0ab79Smaybee } 718cdb0ab79Smaybee 719cdb0ab79Smaybee int 720713d6c20SMatthew Ahrens dmu_free_long_object(objset_t *os, uint64_t object) 721cdb0ab79Smaybee { 722cdb0ab79Smaybee dmu_tx_t *tx; 723cdb0ab79Smaybee int err; 724cdb0ab79Smaybee 725713d6c20SMatthew Ahrens err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); 726cdb0ab79Smaybee if (err != 0) 727cdb0ab79Smaybee return (err); 728713d6c20SMatthew Ahrens 729713d6c20SMatthew Ahrens tx = dmu_tx_create(os); 730713d6c20SMatthew Ahrens dmu_tx_hold_bonus(tx, object); 731713d6c20SMatthew Ahrens dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 7324bb73804SMatthew Ahrens dmu_tx_mark_netfree(tx); 733713d6c20SMatthew Ahrens err = dmu_tx_assign(tx, TXG_WAIT); 734713d6c20SMatthew Ahrens if (err == 0) { 735713d6c20SMatthew Ahrens err = dmu_object_free(os, object, tx); 736713d6c20SMatthew Ahrens dmu_tx_commit(tx); 737cdb0ab79Smaybee } else { 738713d6c20SMatthew Ahrens dmu_tx_abort(tx); 739cdb0ab79Smaybee } 740713d6c20SMatthew Ahrens 741cdb0ab79Smaybee return (err); 742cdb0ab79Smaybee } 743cdb0ab79Smaybee 744ea8dc4b6Seschrock int 745fa9e4066Sahrens dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 746fa9e4066Sahrens uint64_t size, dmu_tx_t *tx) 747fa9e4066Sahrens { 748ea8dc4b6Seschrock dnode_t *dn; 749503ad85cSMatthew Ahrens int err = dnode_hold(os, object, FTAG, &dn); 750ea8dc4b6Seschrock if (err) 751ea8dc4b6Seschrock return (err); 752fa9e4066Sahrens ASSERT(offset < UINT64_MAX); 753fa9e4066Sahrens ASSERT(size == -1ULL || size <= UINT64_MAX - offset); 754fa9e4066Sahrens dnode_free_range(dn, offset, size, tx); 755fa9e4066Sahrens dnode_rele(dn, FTAG); 756ea8dc4b6Seschrock return (0); 757fa9e4066Sahrens } 758fa9e4066Sahrens 759ea8dc4b6Seschrock int 760ea8dc4b6Seschrock dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 7617bfdf011SNeil Perrin void *buf, uint32_t flags) 762fa9e4066Sahrens { 763fa9e4066Sahrens dnode_t *dn; 764fa9e4066Sahrens dmu_buf_t **dbp; 765c87b8fc5SMark J Musante int numbufs, err; 766fa9e4066Sahrens 767503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 768ea8dc4b6Seschrock if (err) 769ea8dc4b6Seschrock return (err); 770feb08c6bSbillm 771feb08c6bSbillm /* 772feb08c6bSbillm * Deal with odd block sizes, where there can't be data past the first 773feb08c6bSbillm * block. If we ever do the tail block optimization, we will need to 774feb08c6bSbillm * handle that here as well. 775feb08c6bSbillm */ 776c87b8fc5SMark J Musante if (dn->dn_maxblkid == 0) { 777fa9e4066Sahrens int newsz = offset > dn->dn_datablksz ? 0 : 778fa9e4066Sahrens MIN(size, dn->dn_datablksz - offset); 779fa9e4066Sahrens bzero((char *)buf + newsz, size - newsz); 780fa9e4066Sahrens size = newsz; 781fa9e4066Sahrens } 782fa9e4066Sahrens 783fa9e4066Sahrens while (size > 0) { 784fa9e4066Sahrens uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); 785c87b8fc5SMark J Musante int i; 786fa9e4066Sahrens 787fa9e4066Sahrens /* 788fa9e4066Sahrens * NB: we could do this block-at-a-time, but it's nice 789fa9e4066Sahrens * to be reading in parallel. 790fa9e4066Sahrens */ 791a2eea2e1Sahrens err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, 7927bfdf011SNeil Perrin TRUE, FTAG, &numbufs, &dbp, flags); 793ea8dc4b6Seschrock if (err) 7941934e92fSmaybee break; 795fa9e4066Sahrens 796fa9e4066Sahrens for (i = 0; i < numbufs; i++) { 797fa9e4066Sahrens int tocpy; 798fa9e4066Sahrens int bufoff; 799fa9e4066Sahrens dmu_buf_t *db = dbp[i]; 800fa9e4066Sahrens 801fa9e4066Sahrens ASSERT(size > 0); 802fa9e4066Sahrens 803fa9e4066Sahrens bufoff = offset - db->db_offset; 804fa9e4066Sahrens tocpy = (int)MIN(db->db_size - bufoff, size); 805fa9e4066Sahrens 806fa9e4066Sahrens bcopy((char *)db->db_data + bufoff, buf, tocpy); 807fa9e4066Sahrens 808fa9e4066Sahrens offset += tocpy; 809fa9e4066Sahrens size -= tocpy; 810fa9e4066Sahrens buf = (char *)buf + tocpy; 811fa9e4066Sahrens } 812ea8dc4b6Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG); 813fa9e4066Sahrens } 814a2eea2e1Sahrens dnode_rele(dn, FTAG); 8151934e92fSmaybee return (err); 816fa9e4066Sahrens } 817fa9e4066Sahrens 818fa9e4066Sahrens void 819fa9e4066Sahrens dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 820fa9e4066Sahrens const void *buf, dmu_tx_t *tx) 821fa9e4066Sahrens { 822fa9e4066Sahrens dmu_buf_t **dbp; 823fa9e4066Sahrens int numbufs, i; 824fa9e4066Sahrens 82513506d1eSmaybee if (size == 0) 82613506d1eSmaybee return; 82713506d1eSmaybee 828ea8dc4b6Seschrock VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 829ea8dc4b6Seschrock FALSE, FTAG, &numbufs, &dbp)); 830fa9e4066Sahrens 831fa9e4066Sahrens for (i = 0; i < numbufs; i++) { 832fa9e4066Sahrens int tocpy; 833fa9e4066Sahrens int bufoff; 834fa9e4066Sahrens dmu_buf_t *db = dbp[i]; 835fa9e4066Sahrens 836fa9e4066Sahrens ASSERT(size > 0); 837fa9e4066Sahrens 838fa9e4066Sahrens bufoff = offset - db->db_offset; 839fa9e4066Sahrens tocpy = (int)MIN(db->db_size - bufoff, size); 840fa9e4066Sahrens 841fa9e4066Sahrens ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 842fa9e4066Sahrens 843fa9e4066Sahrens if (tocpy == db->db_size) 844fa9e4066Sahrens dmu_buf_will_fill(db, tx); 845fa9e4066Sahrens else 846fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 847fa9e4066Sahrens 848fa9e4066Sahrens bcopy(buf, (char *)db->db_data + bufoff, tocpy); 849fa9e4066Sahrens 850fa9e4066Sahrens if (tocpy == db->db_size) 851fa9e4066Sahrens dmu_buf_fill_done(db, tx); 852fa9e4066Sahrens 853fa9e4066Sahrens offset += tocpy; 854fa9e4066Sahrens size -= tocpy; 855fa9e4066Sahrens buf = (char *)buf + tocpy; 856fa9e4066Sahrens } 857ea8dc4b6Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG); 858fa9e4066Sahrens } 859fa9e4066Sahrens 86082c9918fSTim Haley void 86182c9918fSTim Haley dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 86282c9918fSTim Haley dmu_tx_t *tx) 86382c9918fSTim Haley { 86482c9918fSTim Haley dmu_buf_t **dbp; 86582c9918fSTim Haley int numbufs, i; 86682c9918fSTim Haley 86782c9918fSTim Haley if (size == 0) 86882c9918fSTim Haley return; 86982c9918fSTim Haley 87082c9918fSTim Haley VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 87182c9918fSTim Haley FALSE, FTAG, &numbufs, &dbp)); 87282c9918fSTim Haley 87382c9918fSTim Haley for (i = 0; i < numbufs; i++) { 87482c9918fSTim Haley dmu_buf_t *db = dbp[i]; 87582c9918fSTim Haley 87682c9918fSTim Haley dmu_buf_will_not_fill(db, tx); 87782c9918fSTim Haley } 87882c9918fSTim Haley dmu_buf_rele_array(dbp, numbufs, FTAG); 87982c9918fSTim Haley } 88082c9918fSTim Haley 8815d7b4d43SMatthew Ahrens void 8825d7b4d43SMatthew Ahrens dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, 8835d7b4d43SMatthew Ahrens void *data, uint8_t etype, uint8_t comp, int uncompressed_size, 8845d7b4d43SMatthew Ahrens int compressed_size, int byteorder, dmu_tx_t *tx) 8855d7b4d43SMatthew Ahrens { 8865d7b4d43SMatthew Ahrens dmu_buf_t *db; 8875d7b4d43SMatthew Ahrens 8885d7b4d43SMatthew Ahrens ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); 8895d7b4d43SMatthew Ahrens ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); 8905d7b4d43SMatthew Ahrens VERIFY0(dmu_buf_hold_noread(os, object, offset, 8915d7b4d43SMatthew Ahrens FTAG, &db)); 8925d7b4d43SMatthew Ahrens 8935d7b4d43SMatthew Ahrens dmu_buf_write_embedded(db, 8945d7b4d43SMatthew Ahrens data, (bp_embedded_type_t)etype, (enum zio_compress)comp, 8955d7b4d43SMatthew Ahrens uncompressed_size, compressed_size, byteorder, tx); 8965d7b4d43SMatthew Ahrens 8975d7b4d43SMatthew Ahrens dmu_buf_rele(db, FTAG); 8985d7b4d43SMatthew Ahrens } 8995d7b4d43SMatthew Ahrens 900c242f9a0Schunli zhang - Sun Microsystems - Irvine United States /* 901c242f9a0Schunli zhang - Sun Microsystems - Irvine United States * DMU support for xuio 902c242f9a0Schunli zhang - Sun Microsystems - Irvine United States */ 903c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_t *xuio_ksp = NULL; 904c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 905c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int 906c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_init(xuio_t *xuio, int nblk) 907c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 908c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv; 909c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio_t *uio = &xuio->xu_uio; 910c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 911c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_iovcnt = nblk; 912c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); 913c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 914c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); 915c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->cnt = nblk; 916c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); 917c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->iovp = uio->uio_iov; 918c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIO_XUZC_PRIV(xuio) = priv; 919c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 920c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (XUIO_XUZC_RW(xuio) == UIO_READ) 921c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); 922c242f9a0Schunli zhang - Sun Microsystems - Irvine United States else 923c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); 924c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 925c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (0); 926c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 927c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 928c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 929c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_fini(xuio_t *xuio) 930c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 931c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 932c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int nblk = priv->cnt; 933c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 934c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kmem_free(priv->iovp, nblk * sizeof (iovec_t)); 935c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); 936c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kmem_free(priv, sizeof (dmu_xuio_t)); 937c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 938c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (XUIO_XUZC_RW(xuio) == UIO_READ) 939c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); 940c242f9a0Schunli zhang - Sun Microsystems - Irvine United States else 941c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); 942c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 943c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 944c242f9a0Schunli zhang - Sun Microsystems - Irvine United States /* 945c242f9a0Schunli zhang - Sun Microsystems - Irvine United States * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } 946c242f9a0Schunli zhang - Sun Microsystems - Irvine United States * and increase priv->next by 1. 947c242f9a0Schunli zhang - Sun Microsystems - Irvine United States */ 948c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int 949c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) 950c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 951c242f9a0Schunli zhang - Sun Microsystems - Irvine United States struct iovec *iov; 952c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio_t *uio = &xuio->xu_uio; 953c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 954c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int i = priv->next++; 955c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 956c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(i < priv->cnt); 957c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(off + n <= arc_buf_size(abuf)); 958c242f9a0Schunli zhang - Sun Microsystems - Irvine United States iov = uio->uio_iov + i; 959c242f9a0Schunli zhang - Sun Microsystems - Irvine United States iov->iov_base = (char *)abuf->b_data + off; 960c242f9a0Schunli zhang - Sun Microsystems - Irvine United States iov->iov_len = n; 961c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->bufs[i] = abuf; 962c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (0); 963c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 964c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 965c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int 966c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_cnt(xuio_t *xuio) 967c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 968c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 969c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (priv->cnt); 970c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 971c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 972c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t * 973c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_arcbuf(xuio_t *xuio, int i) 974c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 975c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 976c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 977c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(i < priv->cnt); 978c242f9a0Schunli zhang - Sun Microsystems - Irvine United States return (priv->bufs[i]); 979c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 980c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 981c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 982c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_clear(xuio_t *xuio, int i) 983c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 984c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 985c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 986c242f9a0Schunli zhang - Sun Microsystems - Irvine United States ASSERT(i < priv->cnt); 987c242f9a0Schunli zhang - Sun Microsystems - Irvine United States priv->bufs[i] = NULL; 988c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 989c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 990c242f9a0Schunli zhang - Sun Microsystems - Irvine United States static void 991c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_init(void) 992c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 993c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", 994c242f9a0Schunli zhang - Sun Microsystems - Irvine United States KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), 995c242f9a0Schunli zhang - Sun Microsystems - Irvine United States KSTAT_FLAG_VIRTUAL); 996c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (xuio_ksp != NULL) { 997c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_ksp->ks_data = &xuio_stats; 998c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_install(xuio_ksp); 999c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1000c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1001c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 1002c242f9a0Schunli zhang - Sun Microsystems - Irvine United States static void 1003c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_fini(void) 1004c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 1005c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (xuio_ksp != NULL) { 1006c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_delete(xuio_ksp); 1007c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_ksp = NULL; 1008c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1009c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1010c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 1011c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 1012c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_wbuf_copied() 1013c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 1014c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_wbuf_copied); 1015c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1016c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 1017c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void 1018c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_wbuf_nocopy() 1019c242f9a0Schunli zhang - Sun Microsystems - Irvine United States { 1020c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); 1021c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1022c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 1023fa9e4066Sahrens #ifdef _KERNEL 1024f8554bb9SMatthew Ahrens static int 1025f8554bb9SMatthew Ahrens dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) 1026feb08c6bSbillm { 1027feb08c6bSbillm dmu_buf_t **dbp; 1028feb08c6bSbillm int numbufs, i, err; 1029c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_t *xuio = NULL; 1030feb08c6bSbillm 1031feb08c6bSbillm /* 1032feb08c6bSbillm * NB: we could do this block-at-a-time, but it's nice 1033feb08c6bSbillm * to be reading in parallel. 1034feb08c6bSbillm */ 1035f8554bb9SMatthew Ahrens err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, 1036f8554bb9SMatthew Ahrens TRUE, FTAG, &numbufs, &dbp, 0); 1037feb08c6bSbillm if (err) 1038feb08c6bSbillm return (err); 1039feb08c6bSbillm 1040c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (uio->uio_extflg == UIO_XUIO) 1041c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio = (xuio_t *)uio; 1042c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 1043feb08c6bSbillm for (i = 0; i < numbufs; i++) { 1044feb08c6bSbillm int tocpy; 1045feb08c6bSbillm int bufoff; 1046feb08c6bSbillm dmu_buf_t *db = dbp[i]; 1047feb08c6bSbillm 1048feb08c6bSbillm ASSERT(size > 0); 1049feb08c6bSbillm 1050feb08c6bSbillm bufoff = uio->uio_loffset - db->db_offset; 1051feb08c6bSbillm tocpy = (int)MIN(db->db_size - bufoff, size); 1052feb08c6bSbillm 1053c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (xuio) { 1054c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; 1055c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t *dbuf_abuf = dbi->db_buf; 1056c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); 1057c242f9a0Schunli zhang - Sun Microsystems - Irvine United States err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); 1058c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (!err) { 1059c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_resid -= tocpy; 1060c242f9a0Schunli zhang - Sun Microsystems - Irvine United States uio->uio_loffset += tocpy; 1061c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1062c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 1063c242f9a0Schunli zhang - Sun Microsystems - Irvine United States if (abuf == dbuf_abuf) 1064c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); 1065c242f9a0Schunli zhang - Sun Microsystems - Irvine United States else 1066c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_rbuf_copied); 1067c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } else { 1068c242f9a0Schunli zhang - Sun Microsystems - Irvine United States err = uiomove((char *)db->db_data + bufoff, tocpy, 1069c242f9a0Schunli zhang - Sun Microsystems - Irvine United States UIO_READ, uio); 1070c242f9a0Schunli zhang - Sun Microsystems - Irvine United States } 1071feb08c6bSbillm if (err) 1072feb08c6bSbillm break; 1073feb08c6bSbillm 1074feb08c6bSbillm size -= tocpy; 1075feb08c6bSbillm } 1076feb08c6bSbillm dmu_buf_rele_array(dbp, numbufs, FTAG); 1077feb08c6bSbillm 1078feb08c6bSbillm return (err); 1079feb08c6bSbillm } 1080feb08c6bSbillm 1081f8554bb9SMatthew Ahrens /* 1082f8554bb9SMatthew Ahrens * Read 'size' bytes into the uio buffer. 1083f8554bb9SMatthew Ahrens * From object zdb->db_object. 1084f8554bb9SMatthew Ahrens * Starting at offset uio->uio_loffset. 1085f8554bb9SMatthew Ahrens * 1086f8554bb9SMatthew Ahrens * If the caller already has a dbuf in the target object 1087f8554bb9SMatthew Ahrens * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), 1088f8554bb9SMatthew Ahrens * because we don't have to find the dnode_t for the object. 1089f8554bb9SMatthew Ahrens */ 1090f8554bb9SMatthew Ahrens int 1091f8554bb9SMatthew Ahrens dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) 1092f8554bb9SMatthew Ahrens { 1093f8554bb9SMatthew Ahrens dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; 1094f8554bb9SMatthew Ahrens dnode_t *dn; 1095f8554bb9SMatthew Ahrens int err; 1096f8554bb9SMatthew Ahrens 1097f8554bb9SMatthew Ahrens if (size == 0) 1098f8554bb9SMatthew Ahrens return (0); 1099f8554bb9SMatthew Ahrens 1100f8554bb9SMatthew Ahrens DB_DNODE_ENTER(db); 1101f8554bb9SMatthew Ahrens dn = DB_DNODE(db); 1102f8554bb9SMatthew Ahrens err = dmu_read_uio_dnode(dn, uio, size); 1103f8554bb9SMatthew Ahrens DB_DNODE_EXIT(db); 1104f8554bb9SMatthew Ahrens 1105f8554bb9SMatthew Ahrens return (err); 1106f8554bb9SMatthew Ahrens } 1107f8554bb9SMatthew Ahrens 1108f8554bb9SMatthew Ahrens /* 1109f8554bb9SMatthew Ahrens * Read 'size' bytes into the uio buffer. 1110f8554bb9SMatthew Ahrens * From the specified object 1111f8554bb9SMatthew Ahrens * Starting at offset uio->uio_loffset. 1112f8554bb9SMatthew Ahrens */ 1113f8554bb9SMatthew Ahrens int 1114f8554bb9SMatthew Ahrens dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) 1115f8554bb9SMatthew Ahrens { 1116f8554bb9SMatthew Ahrens dnode_t *dn; 1117f8554bb9SMatthew Ahrens int err; 1118f8554bb9SMatthew Ahrens 1119f8554bb9SMatthew Ahrens if (size == 0) 1120f8554bb9SMatthew Ahrens return (0); 1121f8554bb9SMatthew Ahrens 1122f8554bb9SMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 1123f8554bb9SMatthew Ahrens if (err) 1124f8554bb9SMatthew Ahrens return (err); 1125f8554bb9SMatthew Ahrens 1126f8554bb9SMatthew Ahrens err = dmu_read_uio_dnode(dn, uio, size); 1127f8554bb9SMatthew Ahrens 1128f8554bb9SMatthew Ahrens dnode_rele(dn, FTAG); 1129f8554bb9SMatthew Ahrens 1130f8554bb9SMatthew Ahrens return (err); 1131f8554bb9SMatthew Ahrens } 1132f8554bb9SMatthew Ahrens 113394d1a210STim Haley static int 113494d1a210STim Haley dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) 1135fa9e4066Sahrens { 1136fa9e4066Sahrens dmu_buf_t **dbp; 113794d1a210STim Haley int numbufs; 1138fa9e4066Sahrens int err = 0; 113994d1a210STim Haley int i; 1140fa9e4066Sahrens 114194d1a210STim Haley err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, 114294d1a210STim Haley FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); 1143ea8dc4b6Seschrock if (err) 1144ea8dc4b6Seschrock return (err); 1145fa9e4066Sahrens 1146fa9e4066Sahrens for (i = 0; i < numbufs; i++) { 1147fa9e4066Sahrens int tocpy; 1148fa9e4066Sahrens int bufoff; 1149fa9e4066Sahrens dmu_buf_t *db = dbp[i]; 1150fa9e4066Sahrens 1151fa9e4066Sahrens ASSERT(size > 0); 1152fa9e4066Sahrens 1153feb08c6bSbillm bufoff = uio->uio_loffset - db->db_offset; 1154fa9e4066Sahrens tocpy = (int)MIN(db->db_size - bufoff, size); 1155fa9e4066Sahrens 1156fa9e4066Sahrens ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 1157fa9e4066Sahrens 1158fa9e4066Sahrens if (tocpy == db->db_size) 1159fa9e4066Sahrens dmu_buf_will_fill(db, tx); 1160fa9e4066Sahrens else 1161fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 1162fa9e4066Sahrens 1163fa9e4066Sahrens /* 1164fa9e4066Sahrens * XXX uiomove could block forever (eg. nfs-backed 1165fa9e4066Sahrens * pages). There needs to be a uiolockdown() function 1166fa9e4066Sahrens * to lock the pages in memory, so that uiomove won't 1167fa9e4066Sahrens * block. 1168fa9e4066Sahrens */ 1169fa9e4066Sahrens err = uiomove((char *)db->db_data + bufoff, tocpy, 1170fa9e4066Sahrens UIO_WRITE, uio); 1171fa9e4066Sahrens 1172fa9e4066Sahrens if (tocpy == db->db_size) 1173fa9e4066Sahrens dmu_buf_fill_done(db, tx); 1174fa9e4066Sahrens 1175fa9e4066Sahrens if (err) 1176fa9e4066Sahrens break; 1177fa9e4066Sahrens 1178fa9e4066Sahrens size -= tocpy; 1179fa9e4066Sahrens } 118094d1a210STim Haley 1181ea8dc4b6Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG); 1182fa9e4066Sahrens return (err); 1183fa9e4066Sahrens } 118444eda4d7Smaybee 1185f8554bb9SMatthew Ahrens /* 1186f8554bb9SMatthew Ahrens * Write 'size' bytes from the uio buffer. 1187f8554bb9SMatthew Ahrens * To object zdb->db_object. 1188f8554bb9SMatthew Ahrens * Starting at offset uio->uio_loffset. 1189f8554bb9SMatthew Ahrens * 1190f8554bb9SMatthew Ahrens * If the caller already has a dbuf in the target object 1191f8554bb9SMatthew Ahrens * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), 1192f8554bb9SMatthew Ahrens * because we don't have to find the dnode_t for the object. 1193f8554bb9SMatthew Ahrens */ 119494d1a210STim Haley int 119594d1a210STim Haley dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, 119694d1a210STim Haley dmu_tx_t *tx) 119794d1a210STim Haley { 1198744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; 1199744947dcSTom Erickson dnode_t *dn; 1200744947dcSTom Erickson int err; 1201744947dcSTom Erickson 120294d1a210STim Haley if (size == 0) 120394d1a210STim Haley return (0); 120494d1a210STim Haley 1205744947dcSTom Erickson DB_DNODE_ENTER(db); 1206744947dcSTom Erickson dn = DB_DNODE(db); 1207744947dcSTom Erickson err = dmu_write_uio_dnode(dn, uio, size, tx); 1208744947dcSTom Erickson DB_DNODE_EXIT(db); 1209744947dcSTom Erickson 1210744947dcSTom Erickson return (err); 121194d1a210STim Haley } 121294d1a210STim Haley 1213f8554bb9SMatthew Ahrens /* 1214f8554bb9SMatthew Ahrens * Write 'size' bytes from the uio buffer. 1215f8554bb9SMatthew Ahrens * To the specified object. 1216f8554bb9SMatthew Ahrens * Starting at offset uio->uio_loffset. 1217f8554bb9SMatthew Ahrens */ 121894d1a210STim Haley int 121994d1a210STim Haley dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, 122094d1a210STim Haley dmu_tx_t *tx) 122194d1a210STim Haley { 122294d1a210STim Haley dnode_t *dn; 122394d1a210STim Haley int err; 122494d1a210STim Haley 122594d1a210STim Haley if (size == 0) 122694d1a210STim Haley return (0); 122794d1a210STim Haley 122894d1a210STim Haley err = dnode_hold(os, object, FTAG, &dn); 122994d1a210STim Haley if (err) 123094d1a210STim Haley return (err); 123194d1a210STim Haley 123294d1a210STim Haley err = dmu_write_uio_dnode(dn, uio, size, tx); 123394d1a210STim Haley 123494d1a210STim Haley dnode_rele(dn, FTAG); 123594d1a210STim Haley 123694d1a210STim Haley return (err); 123794d1a210STim Haley } 123894d1a210STim Haley 123944eda4d7Smaybee int 124044eda4d7Smaybee dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 124144eda4d7Smaybee page_t *pp, dmu_tx_t *tx) 124244eda4d7Smaybee { 124344eda4d7Smaybee dmu_buf_t **dbp; 124444eda4d7Smaybee int numbufs, i; 124544eda4d7Smaybee int err; 124644eda4d7Smaybee 124744eda4d7Smaybee if (size == 0) 124844eda4d7Smaybee return (0); 124944eda4d7Smaybee 125044eda4d7Smaybee err = dmu_buf_hold_array(os, object, offset, size, 125144eda4d7Smaybee FALSE, FTAG, &numbufs, &dbp); 125244eda4d7Smaybee if (err) 125344eda4d7Smaybee return (err); 125444eda4d7Smaybee 125544eda4d7Smaybee for (i = 0; i < numbufs; i++) { 125644eda4d7Smaybee int tocpy, copied, thiscpy; 125744eda4d7Smaybee int bufoff; 125844eda4d7Smaybee dmu_buf_t *db = dbp[i]; 125944eda4d7Smaybee caddr_t va; 126044eda4d7Smaybee 126144eda4d7Smaybee ASSERT(size > 0); 126244eda4d7Smaybee ASSERT3U(db->db_size, >=, PAGESIZE); 126344eda4d7Smaybee 126444eda4d7Smaybee bufoff = offset - db->db_offset; 126544eda4d7Smaybee tocpy = (int)MIN(db->db_size - bufoff, size); 126644eda4d7Smaybee 126744eda4d7Smaybee ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 126844eda4d7Smaybee 126944eda4d7Smaybee if (tocpy == db->db_size) 127044eda4d7Smaybee dmu_buf_will_fill(db, tx); 127144eda4d7Smaybee else 127244eda4d7Smaybee dmu_buf_will_dirty(db, tx); 127344eda4d7Smaybee 127444eda4d7Smaybee for (copied = 0; copied < tocpy; copied += PAGESIZE) { 127544eda4d7Smaybee ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); 127644eda4d7Smaybee thiscpy = MIN(PAGESIZE, tocpy - copied); 12770fab61baSJonathan W Adams va = zfs_map_page(pp, S_READ); 127844eda4d7Smaybee bcopy(va, (char *)db->db_data + bufoff, thiscpy); 12790fab61baSJonathan W Adams zfs_unmap_page(pp, va); 128044eda4d7Smaybee pp = pp->p_next; 128144eda4d7Smaybee bufoff += PAGESIZE; 128244eda4d7Smaybee } 128344eda4d7Smaybee 128444eda4d7Smaybee if (tocpy == db->db_size) 128544eda4d7Smaybee dmu_buf_fill_done(db, tx); 128644eda4d7Smaybee 128744eda4d7Smaybee offset += tocpy; 128844eda4d7Smaybee size -= tocpy; 128944eda4d7Smaybee } 129044eda4d7Smaybee dmu_buf_rele_array(dbp, numbufs, FTAG); 129144eda4d7Smaybee return (err); 129244eda4d7Smaybee } 1293fa9e4066Sahrens #endif 1294fa9e4066Sahrens 12952fdbea25SAleksandr Guzovskiy /* 12962fdbea25SAleksandr Guzovskiy * Allocate a loaned anonymous arc buffer. 12972fdbea25SAleksandr Guzovskiy */ 12982fdbea25SAleksandr Guzovskiy arc_buf_t * 12992fdbea25SAleksandr Guzovskiy dmu_request_arcbuf(dmu_buf_t *handle, int size) 13002fdbea25SAleksandr Guzovskiy { 1301744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; 13022fdbea25SAleksandr Guzovskiy 130343466aaeSMax Grossman return (arc_loan_buf(db->db_objset->os_spa, size)); 13042fdbea25SAleksandr Guzovskiy } 13052fdbea25SAleksandr Guzovskiy 13062fdbea25SAleksandr Guzovskiy /* 13072fdbea25SAleksandr Guzovskiy * Free a loaned arc buffer. 13082fdbea25SAleksandr Guzovskiy */ 13092fdbea25SAleksandr Guzovskiy void 13102fdbea25SAleksandr Guzovskiy dmu_return_arcbuf(arc_buf_t *buf) 13112fdbea25SAleksandr Guzovskiy { 13122fdbea25SAleksandr Guzovskiy arc_return_buf(buf, FTAG); 13133b2aab18SMatthew Ahrens VERIFY(arc_buf_remove_ref(buf, FTAG)); 13142fdbea25SAleksandr Guzovskiy } 13152fdbea25SAleksandr Guzovskiy 13162fdbea25SAleksandr Guzovskiy /* 13172fdbea25SAleksandr Guzovskiy * When possible directly assign passed loaned arc buffer to a dbuf. 13182fdbea25SAleksandr Guzovskiy * If this is not possible copy the contents of passed arc buf via 13192fdbea25SAleksandr Guzovskiy * dmu_write(). 13202fdbea25SAleksandr Guzovskiy */ 13212fdbea25SAleksandr Guzovskiy void 13222fdbea25SAleksandr Guzovskiy dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, 13232fdbea25SAleksandr Guzovskiy dmu_tx_t *tx) 13242fdbea25SAleksandr Guzovskiy { 1325744947dcSTom Erickson dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; 1326744947dcSTom Erickson dnode_t *dn; 13272fdbea25SAleksandr Guzovskiy dmu_buf_impl_t *db; 13282fdbea25SAleksandr Guzovskiy uint32_t blksz = (uint32_t)arc_buf_size(buf); 13292fdbea25SAleksandr Guzovskiy uint64_t blkid; 13302fdbea25SAleksandr Guzovskiy 1331744947dcSTom Erickson DB_DNODE_ENTER(dbuf); 1332744947dcSTom Erickson dn = DB_DNODE(dbuf); 13332fdbea25SAleksandr Guzovskiy rw_enter(&dn->dn_struct_rwlock, RW_READER); 1334*a2cdcdd2SPaul Dagnelie blkid = dbuf_whichblock(dn, 0, offset); 13352fdbea25SAleksandr Guzovskiy VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); 13362fdbea25SAleksandr Guzovskiy rw_exit(&dn->dn_struct_rwlock); 1337744947dcSTom Erickson DB_DNODE_EXIT(dbuf); 13382fdbea25SAleksandr Guzovskiy 13398a904709SMatthew Ahrens /* 13408a904709SMatthew Ahrens * We can only assign if the offset is aligned, the arc buf is the 13418a904709SMatthew Ahrens * same size as the dbuf, and the dbuf is not metadata. It 13428a904709SMatthew Ahrens * can't be metadata because the loaned arc buf comes from the 13438a904709SMatthew Ahrens * user-data kmem arena. 13448a904709SMatthew Ahrens */ 13458a904709SMatthew Ahrens if (offset == db->db.db_offset && blksz == db->db.db_size && 13468a904709SMatthew Ahrens DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA) { 13472fdbea25SAleksandr Guzovskiy dbuf_assign_arcbuf(db, buf, tx); 13482fdbea25SAleksandr Guzovskiy dbuf_rele(db, FTAG); 13492fdbea25SAleksandr Guzovskiy } else { 1350744947dcSTom Erickson objset_t *os; 1351744947dcSTom Erickson uint64_t object; 1352744947dcSTom Erickson 1353744947dcSTom Erickson DB_DNODE_ENTER(dbuf); 1354744947dcSTom Erickson dn = DB_DNODE(dbuf); 1355744947dcSTom Erickson os = dn->dn_objset; 1356744947dcSTom Erickson object = dn->dn_object; 1357744947dcSTom Erickson DB_DNODE_EXIT(dbuf); 1358744947dcSTom Erickson 13592fdbea25SAleksandr Guzovskiy dbuf_rele(db, FTAG); 1360744947dcSTom Erickson dmu_write(os, object, offset, blksz, buf->b_data, tx); 13612fdbea25SAleksandr Guzovskiy dmu_return_arcbuf(buf); 1362c242f9a0Schunli zhang - Sun Microsystems - Irvine United States XUIOSTAT_BUMP(xuiostat_wbuf_copied); 13632fdbea25SAleksandr Guzovskiy } 13642fdbea25SAleksandr Guzovskiy } 13652fdbea25SAleksandr Guzovskiy 1366c5c6ffa0Smaybee typedef struct { 1367b24ab676SJeff Bonwick dbuf_dirty_record_t *dsa_dr; 1368b24ab676SJeff Bonwick dmu_sync_cb_t *dsa_done; 1369b24ab676SJeff Bonwick zgd_t *dsa_zgd; 1370b24ab676SJeff Bonwick dmu_tx_t *dsa_tx; 1371c717a561Smaybee } dmu_sync_arg_t; 1372c5c6ffa0Smaybee 1373e14bb325SJeff Bonwick /* ARGSUSED */ 1374e14bb325SJeff Bonwick static void 1375e14bb325SJeff Bonwick dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) 1376e14bb325SJeff Bonwick { 1377b24ab676SJeff Bonwick dmu_sync_arg_t *dsa = varg; 1378b24ab676SJeff Bonwick dmu_buf_t *db = dsa->dsa_zgd->zgd_db; 1379e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1380975c32a0SNeil Perrin 1381b24ab676SJeff Bonwick if (zio->io_error == 0) { 1382b24ab676SJeff Bonwick if (BP_IS_HOLE(bp)) { 1383b24ab676SJeff Bonwick /* 1384b24ab676SJeff Bonwick * A block of zeros may compress to a hole, but the 1385b24ab676SJeff Bonwick * block size still needs to be known for replay. 1386b24ab676SJeff Bonwick */ 1387b24ab676SJeff Bonwick BP_SET_LSIZE(bp, db->db_size); 13885d7b4d43SMatthew Ahrens } else if (!BP_IS_EMBEDDED(bp)) { 1389b24ab676SJeff Bonwick ASSERT(BP_GET_LEVEL(bp) == 0); 1390b24ab676SJeff Bonwick bp->blk_fill = 1; 1391b24ab676SJeff Bonwick } 1392e14bb325SJeff Bonwick } 1393e14bb325SJeff Bonwick } 1394e14bb325SJeff Bonwick 1395b24ab676SJeff Bonwick static void 1396b24ab676SJeff Bonwick dmu_sync_late_arrival_ready(zio_t *zio) 1397b24ab676SJeff Bonwick { 1398b24ab676SJeff Bonwick dmu_sync_ready(zio, NULL, zio->io_private); 1399b24ab676SJeff Bonwick } 1400b24ab676SJeff Bonwick 1401c5c6ffa0Smaybee /* ARGSUSED */ 1402c5c6ffa0Smaybee static void 1403c5c6ffa0Smaybee dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) 1404c5c6ffa0Smaybee { 1405b24ab676SJeff Bonwick dmu_sync_arg_t *dsa = varg; 1406b24ab676SJeff Bonwick dbuf_dirty_record_t *dr = dsa->dsa_dr; 1407c717a561Smaybee dmu_buf_impl_t *db = dr->dr_dbuf; 1408c5c6ffa0Smaybee 1409b50a0fe0SNeil Perrin mutex_enter(&db->db_mtx); 1410b50a0fe0SNeil Perrin ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); 1411b24ab676SJeff Bonwick if (zio->io_error == 0) { 141280901aeaSGeorge Wilson dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); 141380901aeaSGeorge Wilson if (dr->dt.dl.dr_nopwrite) { 141480901aeaSGeorge Wilson blkptr_t *bp = zio->io_bp; 141580901aeaSGeorge Wilson blkptr_t *bp_orig = &zio->io_bp_orig; 141680901aeaSGeorge Wilson uint8_t chksum = BP_GET_CHECKSUM(bp_orig); 141780901aeaSGeorge Wilson 141880901aeaSGeorge Wilson ASSERT(BP_EQUAL(bp, bp_orig)); 141980901aeaSGeorge Wilson ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); 142080901aeaSGeorge Wilson ASSERT(zio_checksum_table[chksum].ci_dedup); 142180901aeaSGeorge Wilson } 1422b24ab676SJeff Bonwick dr->dt.dl.dr_overridden_by = *zio->io_bp; 1423b24ab676SJeff Bonwick dr->dt.dl.dr_override_state = DR_OVERRIDDEN; 1424b24ab676SJeff Bonwick dr->dt.dl.dr_copies = zio->io_prop.zp_copies; 142570163ac5SPrakash Surya 142670163ac5SPrakash Surya /* 142770163ac5SPrakash Surya * Old style holes are filled with all zeros, whereas 142870163ac5SPrakash Surya * new-style holes maintain their lsize, type, level, 142970163ac5SPrakash Surya * and birth time (see zio_write_compress). While we 143070163ac5SPrakash Surya * need to reset the BP_SET_LSIZE() call that happened 143170163ac5SPrakash Surya * in dmu_sync_ready for old style holes, we do *not* 143270163ac5SPrakash Surya * want to wipe out the information contained in new 143370163ac5SPrakash Surya * style holes. Thus, only zero out the block pointer if 143470163ac5SPrakash Surya * it's an old style hole. 143570163ac5SPrakash Surya */ 143670163ac5SPrakash Surya if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && 143770163ac5SPrakash Surya dr->dt.dl.dr_overridden_by.blk_birth == 0) 1438b24ab676SJeff Bonwick BP_ZERO(&dr->dt.dl.dr_overridden_by); 1439b24ab676SJeff Bonwick } else { 1440b24ab676SJeff Bonwick dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; 1441b24ab676SJeff Bonwick } 1442c5c6ffa0Smaybee cv_broadcast(&db->db_changed); 1443b50a0fe0SNeil Perrin mutex_exit(&db->db_mtx); 1444b50a0fe0SNeil Perrin 1445b24ab676SJeff Bonwick dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1446b24ab676SJeff Bonwick 1447b24ab676SJeff Bonwick kmem_free(dsa, sizeof (*dsa)); 1448b24ab676SJeff Bonwick } 1449b24ab676SJeff Bonwick 1450b24ab676SJeff Bonwick static void 1451b24ab676SJeff Bonwick dmu_sync_late_arrival_done(zio_t *zio) 1452b24ab676SJeff Bonwick { 1453b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1454b24ab676SJeff Bonwick dmu_sync_arg_t *dsa = zio->io_private; 145580901aeaSGeorge Wilson blkptr_t *bp_orig = &zio->io_bp_orig; 1456b24ab676SJeff Bonwick 1457b24ab676SJeff Bonwick if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { 145880901aeaSGeorge Wilson /* 145980901aeaSGeorge Wilson * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE) 146080901aeaSGeorge Wilson * then there is nothing to do here. Otherwise, free the 146180901aeaSGeorge Wilson * newly allocated block in this txg. 146280901aeaSGeorge Wilson */ 146380901aeaSGeorge Wilson if (zio->io_flags & ZIO_FLAG_NOPWRITE) { 146480901aeaSGeorge Wilson ASSERT(BP_EQUAL(bp, bp_orig)); 146580901aeaSGeorge Wilson } else { 146680901aeaSGeorge Wilson ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); 146780901aeaSGeorge Wilson ASSERT(zio->io_bp->blk_birth == zio->io_txg); 146880901aeaSGeorge Wilson ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); 146980901aeaSGeorge Wilson zio_free(zio->io_spa, zio->io_txg, zio->io_bp); 147080901aeaSGeorge Wilson } 1471b24ab676SJeff Bonwick } 1472b24ab676SJeff Bonwick 1473b24ab676SJeff Bonwick dmu_tx_commit(dsa->dsa_tx); 1474b24ab676SJeff Bonwick 1475b24ab676SJeff Bonwick dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1476b24ab676SJeff Bonwick 1477b24ab676SJeff Bonwick kmem_free(dsa, sizeof (*dsa)); 1478b24ab676SJeff Bonwick } 1479b24ab676SJeff Bonwick 1480b24ab676SJeff Bonwick static int 1481b24ab676SJeff Bonwick dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, 14827802d7bfSMatthew Ahrens zio_prop_t *zp, zbookmark_phys_t *zb) 1483b24ab676SJeff Bonwick { 1484b24ab676SJeff Bonwick dmu_sync_arg_t *dsa; 1485b24ab676SJeff Bonwick dmu_tx_t *tx; 1486b24ab676SJeff Bonwick 1487b24ab676SJeff Bonwick tx = dmu_tx_create(os); 1488b24ab676SJeff Bonwick dmu_tx_hold_space(tx, zgd->zgd_db->db_size); 14896e1f5caaSNeil Perrin if (dmu_tx_assign(tx, TXG_WAIT) != 0) { 1490b24ab676SJeff Bonwick dmu_tx_abort(tx); 1491be6fd75aSMatthew Ahrens /* Make zl_get_data do txg_waited_synced() */ 1492be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 1493b24ab676SJeff Bonwick } 1494b24ab676SJeff Bonwick 1495b24ab676SJeff Bonwick dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1496b24ab676SJeff Bonwick dsa->dsa_dr = NULL; 1497b24ab676SJeff Bonwick dsa->dsa_done = done; 1498b24ab676SJeff Bonwick dsa->dsa_zgd = zgd; 1499b24ab676SJeff Bonwick dsa->dsa_tx = tx; 1500c717a561Smaybee 1501b24ab676SJeff Bonwick zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, 1502b24ab676SJeff Bonwick zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, 150369962b56SMatthew Ahrens dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa, 1504b24ab676SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); 1505b24ab676SJeff Bonwick 1506b24ab676SJeff Bonwick return (0); 1507c5c6ffa0Smaybee } 1508c5c6ffa0Smaybee 1509fa9e4066Sahrens /* 1510c5c6ffa0Smaybee * Intent log support: sync the block associated with db to disk. 1511c5c6ffa0Smaybee * N.B. and XXX: the caller is responsible for making sure that the 1512c5c6ffa0Smaybee * data isn't changing while dmu_sync() is writing it. 1513fa9e4066Sahrens * 1514fa9e4066Sahrens * Return values: 1515fa9e4066Sahrens * 151680901aeaSGeorge Wilson * EEXIST: this txg has already been synced, so there's nothing to do. 1517fa9e4066Sahrens * The caller should not log the write. 1518fa9e4066Sahrens * 1519fa9e4066Sahrens * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. 1520fa9e4066Sahrens * The caller should not log the write. 1521fa9e4066Sahrens * 1522c5c6ffa0Smaybee * EALREADY: this block is already in the process of being synced. 1523c5c6ffa0Smaybee * The caller should track its progress (somehow). 1524fa9e4066Sahrens * 1525b24ab676SJeff Bonwick * EIO: could not do the I/O. 1526b24ab676SJeff Bonwick * The caller should do a txg_wait_synced(). 1527fa9e4066Sahrens * 1528b24ab676SJeff Bonwick * 0: the I/O has been initiated. 1529b24ab676SJeff Bonwick * The caller should log this blkptr in the done callback. 1530b24ab676SJeff Bonwick * It is possible that the I/O will fail, in which case 1531b24ab676SJeff Bonwick * the error will be reported to the done callback and 1532b24ab676SJeff Bonwick * propagated to pio from zio_done(). 1533fa9e4066Sahrens */ 1534fa9e4066Sahrens int 1535b24ab676SJeff Bonwick dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) 1536fa9e4066Sahrens { 1537b24ab676SJeff Bonwick blkptr_t *bp = zgd->zgd_bp; 1538b24ab676SJeff Bonwick dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; 1539503ad85cSMatthew Ahrens objset_t *os = db->db_objset; 1540b24ab676SJeff Bonwick dsl_dataset_t *ds = os->os_dsl_dataset; 1541c717a561Smaybee dbuf_dirty_record_t *dr; 1542b24ab676SJeff Bonwick dmu_sync_arg_t *dsa; 15437802d7bfSMatthew Ahrens zbookmark_phys_t zb; 1544b24ab676SJeff Bonwick zio_prop_t zp; 1545744947dcSTom Erickson dnode_t *dn; 1546fa9e4066Sahrens 1547b24ab676SJeff Bonwick ASSERT(pio != NULL); 1548fa9e4066Sahrens ASSERT(txg != 0); 1549fa9e4066Sahrens 1550b24ab676SJeff Bonwick SET_BOOKMARK(&zb, ds->ds_object, 1551b24ab676SJeff Bonwick db->db.db_object, db->db_level, db->db_blkid); 1552b24ab676SJeff Bonwick 1553744947dcSTom Erickson DB_DNODE_ENTER(db); 1554744947dcSTom Erickson dn = DB_DNODE(db); 1555744947dcSTom Erickson dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); 1556744947dcSTom Erickson DB_DNODE_EXIT(db); 1557fa9e4066Sahrens 1558ea8dc4b6Seschrock /* 1559b24ab676SJeff Bonwick * If we're frozen (running ziltest), we always need to generate a bp. 1560ea8dc4b6Seschrock */ 1561b24ab676SJeff Bonwick if (txg > spa_freeze_txg(os->os_spa)) 1562b24ab676SJeff Bonwick return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1563ea8dc4b6Seschrock 1564fa9e4066Sahrens /* 1565b24ab676SJeff Bonwick * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() 1566b24ab676SJeff Bonwick * and us. If we determine that this txg is not yet syncing, 1567b24ab676SJeff Bonwick * but it begins to sync a moment later, that's OK because the 1568b24ab676SJeff Bonwick * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. 1569fa9e4066Sahrens */ 1570b24ab676SJeff Bonwick mutex_enter(&db->db_mtx); 1571b24ab676SJeff Bonwick 1572b24ab676SJeff Bonwick if (txg <= spa_last_synced_txg(os->os_spa)) { 1573fa9e4066Sahrens /* 1574b24ab676SJeff Bonwick * This txg has already synced. There's nothing to do. 1575fa9e4066Sahrens */ 1576b24ab676SJeff Bonwick mutex_exit(&db->db_mtx); 1577be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 1578fa9e4066Sahrens } 1579fa9e4066Sahrens 1580b24ab676SJeff Bonwick if (txg <= spa_syncing_txg(os->os_spa)) { 1581b24ab676SJeff Bonwick /* 1582b24ab676SJeff Bonwick * This txg is currently syncing, so we can't mess with 1583b24ab676SJeff Bonwick * the dirty record anymore; just write a new log block. 1584b24ab676SJeff Bonwick */ 1585b24ab676SJeff Bonwick mutex_exit(&db->db_mtx); 1586b24ab676SJeff Bonwick return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1587fa9e4066Sahrens } 1588fa9e4066Sahrens 1589c717a561Smaybee dr = db->db_last_dirty; 1590b24ab676SJeff Bonwick while (dr && dr->dr_txg != txg) 1591c717a561Smaybee dr = dr->dr_next; 1592b24ab676SJeff Bonwick 1593b24ab676SJeff Bonwick if (dr == NULL) { 1594c5c6ffa0Smaybee /* 1595b24ab676SJeff Bonwick * There's no dr for this dbuf, so it must have been freed. 1596c5c6ffa0Smaybee * There's no need to log writes to freed blocks, so we're done. 1597c5c6ffa0Smaybee */ 1598c5c6ffa0Smaybee mutex_exit(&db->db_mtx); 1599be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 1600c5c6ffa0Smaybee } 1601c5c6ffa0Smaybee 160280901aeaSGeorge Wilson ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); 160380901aeaSGeorge Wilson 160480901aeaSGeorge Wilson /* 160534e8acefSMatthew Ahrens * Assume the on-disk data is X, the current syncing data (in 160634e8acefSMatthew Ahrens * txg - 1) is Y, and the current in-memory data is Z (currently 160734e8acefSMatthew Ahrens * in dmu_sync). 160834e8acefSMatthew Ahrens * 160934e8acefSMatthew Ahrens * We usually want to perform a nopwrite if X and Z are the 161034e8acefSMatthew Ahrens * same. However, if Y is different (i.e. the BP is going to 161134e8acefSMatthew Ahrens * change before this write takes effect), then a nopwrite will 161234e8acefSMatthew Ahrens * be incorrect - we would override with X, which could have 161334e8acefSMatthew Ahrens * been freed when Y was written. 161434e8acefSMatthew Ahrens * 161534e8acefSMatthew Ahrens * (Note that this is not a concern when we are nop-writing from 161634e8acefSMatthew Ahrens * syncing context, because X and Y must be identical, because 161734e8acefSMatthew Ahrens * all previous txgs have been synced.) 161834e8acefSMatthew Ahrens * 161934e8acefSMatthew Ahrens * Therefore, we disable nopwrite if the current BP could change 162034e8acefSMatthew Ahrens * before this TXG. There are two ways it could change: by 162134e8acefSMatthew Ahrens * being dirty (dr_next is non-NULL), or by being freed 162234e8acefSMatthew Ahrens * (dnode_block_freed()). This behavior is verified by 162334e8acefSMatthew Ahrens * zio_done(), which VERIFYs that the override BP is identical 162434e8acefSMatthew Ahrens * to the on-disk BP. 162580901aeaSGeorge Wilson */ 162634e8acefSMatthew Ahrens DB_DNODE_ENTER(db); 162734e8acefSMatthew Ahrens dn = DB_DNODE(db); 162834e8acefSMatthew Ahrens if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) 162980901aeaSGeorge Wilson zp.zp_nopwrite = B_FALSE; 163034e8acefSMatthew Ahrens DB_DNODE_EXIT(db); 163180901aeaSGeorge Wilson 1632c717a561Smaybee ASSERT(dr->dr_txg == txg); 1633b24ab676SJeff Bonwick if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || 1634b24ab676SJeff Bonwick dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { 1635c717a561Smaybee /* 1636b24ab676SJeff Bonwick * We have already issued a sync write for this buffer, 1637b24ab676SJeff Bonwick * or this buffer has already been synced. It could not 1638c717a561Smaybee * have been dirtied since, or we would have cleared the state. 1639c717a561Smaybee */ 1640c717a561Smaybee mutex_exit(&db->db_mtx); 1641be6fd75aSMatthew Ahrens return (SET_ERROR(EALREADY)); 1642c717a561Smaybee } 1643c717a561Smaybee 1644b24ab676SJeff Bonwick ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); 1645c717a561Smaybee dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; 1646fa9e4066Sahrens mutex_exit(&db->db_mtx); 1647e14bb325SJeff Bonwick 1648b24ab676SJeff Bonwick dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1649b24ab676SJeff Bonwick dsa->dsa_dr = dr; 1650b24ab676SJeff Bonwick dsa->dsa_done = done; 1651b24ab676SJeff Bonwick dsa->dsa_zgd = zgd; 1652b24ab676SJeff Bonwick dsa->dsa_tx = NULL; 1653e14bb325SJeff Bonwick 1654b24ab676SJeff Bonwick zio_nowait(arc_write(pio, os->os_spa, txg, 1655aad02571SSaso Kiselkov bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), 165669962b56SMatthew Ahrens DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready, 165769962b56SMatthew Ahrens NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE, 165869962b56SMatthew Ahrens ZIO_FLAG_CANFAIL, &zb)); 1659e14bb325SJeff Bonwick 1660b24ab676SJeff Bonwick return (0); 1661fa9e4066Sahrens } 1662fa9e4066Sahrens 1663fa9e4066Sahrens int 1664fa9e4066Sahrens dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, 1665fa9e4066Sahrens dmu_tx_t *tx) 1666fa9e4066Sahrens { 1667ea8dc4b6Seschrock dnode_t *dn; 1668ea8dc4b6Seschrock int err; 1669ea8dc4b6Seschrock 1670503ad85cSMatthew Ahrens err = dnode_hold(os, object, FTAG, &dn); 1671ea8dc4b6Seschrock if (err) 1672ea8dc4b6Seschrock return (err); 1673ea8dc4b6Seschrock err = dnode_set_blksz(dn, size, ibs, tx); 1674fa9e4066Sahrens dnode_rele(dn, FTAG); 1675fa9e4066Sahrens return (err); 1676fa9e4066Sahrens } 1677fa9e4066Sahrens 1678fa9e4066Sahrens void 1679fa9e4066Sahrens dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 1680fa9e4066Sahrens dmu_tx_t *tx) 1681fa9e4066Sahrens { 1682ea8dc4b6Seschrock dnode_t *dn; 1683ea8dc4b6Seschrock 16845d7b4d43SMatthew Ahrens /* 16855d7b4d43SMatthew Ahrens * Send streams include each object's checksum function. This 16865d7b4d43SMatthew Ahrens * check ensures that the receiving system can understand the 16875d7b4d43SMatthew Ahrens * checksum function transmitted. 16885d7b4d43SMatthew Ahrens */ 16895d7b4d43SMatthew Ahrens ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); 16905d7b4d43SMatthew Ahrens 16915d7b4d43SMatthew Ahrens VERIFY0(dnode_hold(os, object, FTAG, &dn)); 16925d7b4d43SMatthew Ahrens ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); 1693fa9e4066Sahrens dn->dn_checksum = checksum; 1694fa9e4066Sahrens dnode_setdirty(dn, tx); 1695fa9e4066Sahrens dnode_rele(dn, FTAG); 1696fa9e4066Sahrens } 1697fa9e4066Sahrens 1698fa9e4066Sahrens void 1699fa9e4066Sahrens dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 1700fa9e4066Sahrens dmu_tx_t *tx) 1701fa9e4066Sahrens { 1702ea8dc4b6Seschrock dnode_t *dn; 1703ea8dc4b6Seschrock 17045d7b4d43SMatthew Ahrens /* 17055d7b4d43SMatthew Ahrens * Send streams include each object's compression function. This 17065d7b4d43SMatthew Ahrens * check ensures that the receiving system can understand the 17075d7b4d43SMatthew Ahrens * compression function transmitted. 17085d7b4d43SMatthew Ahrens */ 17095d7b4d43SMatthew Ahrens ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); 17105d7b4d43SMatthew Ahrens 17115d7b4d43SMatthew Ahrens VERIFY0(dnode_hold(os, object, FTAG, &dn)); 1712fa9e4066Sahrens dn->dn_compress = compress; 1713fa9e4066Sahrens dnode_setdirty(dn, tx); 1714fa9e4066Sahrens dnode_rele(dn, FTAG); 1715fa9e4066Sahrens } 1716fa9e4066Sahrens 1717b24ab676SJeff Bonwick int zfs_mdcomp_disable = 0; 1718b24ab676SJeff Bonwick 1719edf345e6SMatthew Ahrens /* 1720edf345e6SMatthew Ahrens * When the "redundant_metadata" property is set to "most", only indirect 1721edf345e6SMatthew Ahrens * blocks of this level and higher will have an additional ditto block. 1722edf345e6SMatthew Ahrens */ 1723edf345e6SMatthew Ahrens int zfs_redundant_metadata_most_ditto_level = 2; 1724edf345e6SMatthew Ahrens 1725b24ab676SJeff Bonwick void 1726b24ab676SJeff Bonwick dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) 1727b24ab676SJeff Bonwick { 1728b24ab676SJeff Bonwick dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; 1729ad135b5dSChristopher Siden boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || 17301d8ccc7bSMark Shellenbaum (wp & WP_SPILL)); 1731b24ab676SJeff Bonwick enum zio_checksum checksum = os->os_checksum; 1732b24ab676SJeff Bonwick enum zio_compress compress = os->os_compress; 1733b24ab676SJeff Bonwick enum zio_checksum dedup_checksum = os->os_dedup_checksum; 17347540df39SGeorge Wilson boolean_t dedup = B_FALSE; 17357540df39SGeorge Wilson boolean_t nopwrite = B_FALSE; 1736b24ab676SJeff Bonwick boolean_t dedup_verify = os->os_dedup_verify; 1737b24ab676SJeff Bonwick int copies = os->os_copies; 1738b24ab676SJeff Bonwick 1739b24ab676SJeff Bonwick /* 174080901aeaSGeorge Wilson * We maintain different write policies for each of the following 174180901aeaSGeorge Wilson * types of data: 174280901aeaSGeorge Wilson * 1. metadata 174380901aeaSGeorge Wilson * 2. preallocated blocks (i.e. level-0 blocks of a dump device) 174480901aeaSGeorge Wilson * 3. all other level 0 blocks 1745b24ab676SJeff Bonwick */ 1746b24ab676SJeff Bonwick if (ismd) { 1747b8289d24SDaniil Lunev if (zfs_mdcomp_disable) { 1748b8289d24SDaniil Lunev compress = ZIO_COMPRESS_EMPTY; 1749b8289d24SDaniil Lunev } else { 1750db1741f5SJustin T. Gibbs /* 1751db1741f5SJustin T. Gibbs * XXX -- we should design a compression algorithm 1752db1741f5SJustin T. Gibbs * that specializes in arrays of bps. 1753db1741f5SJustin T. Gibbs */ 1754db1741f5SJustin T. Gibbs compress = zio_compress_select(os->os_spa, 1755db1741f5SJustin T. Gibbs ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); 1756b8289d24SDaniil Lunev } 175780901aeaSGeorge Wilson 1758b24ab676SJeff Bonwick /* 1759b24ab676SJeff Bonwick * Metadata always gets checksummed. If the data 1760b24ab676SJeff Bonwick * checksum is multi-bit correctable, and it's not a 1761b24ab676SJeff Bonwick * ZBT-style checksum, then it's suitable for metadata 1762b24ab676SJeff Bonwick * as well. Otherwise, the metadata checksum defaults 1763b24ab676SJeff Bonwick * to fletcher4. 1764b24ab676SJeff Bonwick */ 1765b24ab676SJeff Bonwick if (zio_checksum_table[checksum].ci_correctable < 1 || 17666e1f5caaSNeil Perrin zio_checksum_table[checksum].ci_eck) 1767b24ab676SJeff Bonwick checksum = ZIO_CHECKSUM_FLETCHER_4; 1768edf345e6SMatthew Ahrens 1769edf345e6SMatthew Ahrens if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || 1770edf345e6SMatthew Ahrens (os->os_redundant_metadata == 1771edf345e6SMatthew Ahrens ZFS_REDUNDANT_METADATA_MOST && 1772edf345e6SMatthew Ahrens (level >= zfs_redundant_metadata_most_ditto_level || 1773edf345e6SMatthew Ahrens DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) 1774edf345e6SMatthew Ahrens copies++; 177580901aeaSGeorge Wilson } else if (wp & WP_NOFILL) { 177680901aeaSGeorge Wilson ASSERT(level == 0); 177780901aeaSGeorge Wilson 1778b24ab676SJeff Bonwick /* 177980901aeaSGeorge Wilson * If we're writing preallocated blocks, we aren't actually 178080901aeaSGeorge Wilson * writing them so don't set any policy properties. These 178180901aeaSGeorge Wilson * blocks are currently only used by an external subsystem 178280901aeaSGeorge Wilson * outside of zfs (i.e. dump) and not written by the zio 178380901aeaSGeorge Wilson * pipeline. 1784b24ab676SJeff Bonwick */ 178580901aeaSGeorge Wilson compress = ZIO_COMPRESS_OFF; 1786810e43b2SBill Pijewski checksum = ZIO_CHECKSUM_NOPARITY; 1787b24ab676SJeff Bonwick } else { 1788db1741f5SJustin T. Gibbs compress = zio_compress_select(os->os_spa, dn->dn_compress, 1789db1741f5SJustin T. Gibbs compress); 1790b24ab676SJeff Bonwick 179180901aeaSGeorge Wilson checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? 179280901aeaSGeorge Wilson zio_checksum_select(dn->dn_checksum, checksum) : 179380901aeaSGeorge Wilson dedup_checksum; 1794b24ab676SJeff Bonwick 179580901aeaSGeorge Wilson /* 179680901aeaSGeorge Wilson * Determine dedup setting. If we are in dmu_sync(), 179780901aeaSGeorge Wilson * we won't actually dedup now because that's all 179880901aeaSGeorge Wilson * done in syncing context; but we do want to use the 179980901aeaSGeorge Wilson * dedup checkum. If the checksum is not strong 180080901aeaSGeorge Wilson * enough to ensure unique signatures, force 180180901aeaSGeorge Wilson * dedup_verify. 180280901aeaSGeorge Wilson */ 180380901aeaSGeorge Wilson if (dedup_checksum != ZIO_CHECKSUM_OFF) { 180480901aeaSGeorge Wilson dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; 180580901aeaSGeorge Wilson if (!zio_checksum_table[checksum].ci_dedup) 180680901aeaSGeorge Wilson dedup_verify = B_TRUE; 180780901aeaSGeorge Wilson } 1808b24ab676SJeff Bonwick 180980901aeaSGeorge Wilson /* 181080901aeaSGeorge Wilson * Enable nopwrite if we have a cryptographically secure 181180901aeaSGeorge Wilson * checksum that has no known collisions (i.e. SHA-256) 181280901aeaSGeorge Wilson * and compression is enabled. We don't enable nopwrite if 181380901aeaSGeorge Wilson * dedup is enabled as the two features are mutually exclusive. 181480901aeaSGeorge Wilson */ 181580901aeaSGeorge Wilson nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup && 181680901aeaSGeorge Wilson compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); 1817b24ab676SJeff Bonwick } 1818b24ab676SJeff Bonwick 1819b24ab676SJeff Bonwick zp->zp_checksum = checksum; 1820b24ab676SJeff Bonwick zp->zp_compress = compress; 18210a586ceaSMark Shellenbaum zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; 1822b24ab676SJeff Bonwick zp->zp_level = level; 1823edf345e6SMatthew Ahrens zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); 1824b24ab676SJeff Bonwick zp->zp_dedup = dedup; 1825b24ab676SJeff Bonwick zp->zp_dedup_verify = dedup && dedup_verify; 182680901aeaSGeorge Wilson zp->zp_nopwrite = nopwrite; 1827b24ab676SJeff Bonwick } 1828b24ab676SJeff Bonwick 1829fa9e4066Sahrens int 1830fa9e4066Sahrens dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) 1831fa9e4066Sahrens { 1832fa9e4066Sahrens dnode_t *dn; 18332bcf0248SMax Grossman int err; 1834fa9e4066Sahrens 1835fa9e4066Sahrens /* 1836fa9e4066Sahrens * Sync any current changes before 1837fa9e4066Sahrens * we go trundling through the block pointers. 1838fa9e4066Sahrens */ 18392bcf0248SMax Grossman err = dmu_object_wait_synced(os, object); 18402bcf0248SMax Grossman if (err) { 18412bcf0248SMax Grossman return (err); 1842fa9e4066Sahrens } 18432bcf0248SMax Grossman 18442bcf0248SMax Grossman err = dnode_hold(os, object, FTAG, &dn); 18452bcf0248SMax Grossman if (err) { 18462bcf0248SMax Grossman return (err); 1847fa9e4066Sahrens } 1848fa9e4066Sahrens 1849cdb0ab79Smaybee err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); 1850fa9e4066Sahrens dnode_rele(dn, FTAG); 1851fa9e4066Sahrens 1852fa9e4066Sahrens return (err); 1853fa9e4066Sahrens } 1854fa9e4066Sahrens 18552bcf0248SMax Grossman /* 18562bcf0248SMax Grossman * Given the ZFS object, if it contains any dirty nodes 18572bcf0248SMax Grossman * this function flushes all dirty blocks to disk. This 18582bcf0248SMax Grossman * ensures the DMU object info is updated. A more efficient 18592bcf0248SMax Grossman * future version might just find the TXG with the maximum 18602bcf0248SMax Grossman * ID and wait for that to be synced. 18612bcf0248SMax Grossman */ 18622bcf0248SMax Grossman int 18632bcf0248SMax Grossman dmu_object_wait_synced(objset_t *os, uint64_t object) { 18642bcf0248SMax Grossman dnode_t *dn; 18652bcf0248SMax Grossman int error, i; 18662bcf0248SMax Grossman 18672bcf0248SMax Grossman error = dnode_hold(os, object, FTAG, &dn); 18682bcf0248SMax Grossman if (error) { 18692bcf0248SMax Grossman return (error); 18702bcf0248SMax Grossman } 18712bcf0248SMax Grossman 18722bcf0248SMax Grossman for (i = 0; i < TXG_SIZE; i++) { 18732bcf0248SMax Grossman if (list_link_active(&dn->dn_dirty_link[i])) { 18742bcf0248SMax Grossman break; 18752bcf0248SMax Grossman } 18762bcf0248SMax Grossman } 18772bcf0248SMax Grossman dnode_rele(dn, FTAG); 18782bcf0248SMax Grossman if (i != TXG_SIZE) { 18792bcf0248SMax Grossman txg_wait_synced(dmu_objset_pool(os), 0); 18802bcf0248SMax Grossman } 18812bcf0248SMax Grossman 18822bcf0248SMax Grossman return (0); 18832bcf0248SMax Grossman } 18842bcf0248SMax Grossman 1885fa9e4066Sahrens void 1886fa9e4066Sahrens dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) 1887fa9e4066Sahrens { 1888b24ab676SJeff Bonwick dnode_phys_t *dnp; 1889b24ab676SJeff Bonwick 1890fa9e4066Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER); 1891fa9e4066Sahrens mutex_enter(&dn->dn_mtx); 1892fa9e4066Sahrens 1893b24ab676SJeff Bonwick dnp = dn->dn_phys; 1894b24ab676SJeff Bonwick 1895fa9e4066Sahrens doi->doi_data_block_size = dn->dn_datablksz; 1896fa9e4066Sahrens doi->doi_metadata_block_size = dn->dn_indblkshift ? 1897fa9e4066Sahrens 1ULL << dn->dn_indblkshift : 0; 1898b24ab676SJeff Bonwick doi->doi_type = dn->dn_type; 1899b24ab676SJeff Bonwick doi->doi_bonus_type = dn->dn_bonustype; 1900b24ab676SJeff Bonwick doi->doi_bonus_size = dn->dn_bonuslen; 1901fa9e4066Sahrens doi->doi_indirection = dn->dn_nlevels; 1902fa9e4066Sahrens doi->doi_checksum = dn->dn_checksum; 1903fa9e4066Sahrens doi->doi_compress = dn->dn_compress; 1904e77d42eaSMatthew Ahrens doi->doi_nblkptr = dn->dn_nblkptr; 1905b24ab676SJeff Bonwick doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; 1906d0475637SMatthew Ahrens doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; 1907b24ab676SJeff Bonwick doi->doi_fill_count = 0; 1908b24ab676SJeff Bonwick for (int i = 0; i < dnp->dn_nblkptr; i++) 19095d7b4d43SMatthew Ahrens doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); 1910fa9e4066Sahrens 1911fa9e4066Sahrens mutex_exit(&dn->dn_mtx); 1912fa9e4066Sahrens rw_exit(&dn->dn_struct_rwlock); 1913fa9e4066Sahrens } 1914fa9e4066Sahrens 1915fa9e4066Sahrens /* 1916fa9e4066Sahrens * Get information on a DMU object. 1917fa9e4066Sahrens * If doi is NULL, just indicates whether the object exists. 1918fa9e4066Sahrens */ 1919fa9e4066Sahrens int 1920fa9e4066Sahrens dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) 1921fa9e4066Sahrens { 1922ea8dc4b6Seschrock dnode_t *dn; 1923503ad85cSMatthew Ahrens int err = dnode_hold(os, object, FTAG, &dn); 1924fa9e4066Sahrens 1925ea8dc4b6Seschrock if (err) 1926ea8dc4b6Seschrock return (err); 1927fa9e4066Sahrens 1928fa9e4066Sahrens if (doi != NULL) 1929fa9e4066Sahrens dmu_object_info_from_dnode(dn, doi); 1930fa9e4066Sahrens 1931fa9e4066Sahrens dnode_rele(dn, FTAG); 1932fa9e4066Sahrens return (0); 1933fa9e4066Sahrens } 1934fa9e4066Sahrens 1935fa9e4066Sahrens /* 1936fa9e4066Sahrens * As above, but faster; can be used when you have a held dbuf in hand. 1937fa9e4066Sahrens */ 1938fa9e4066Sahrens void 1939744947dcSTom Erickson dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) 1940fa9e4066Sahrens { 1941744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 1942744947dcSTom Erickson 1943744947dcSTom Erickson DB_DNODE_ENTER(db); 1944744947dcSTom Erickson dmu_object_info_from_dnode(DB_DNODE(db), doi); 1945744947dcSTom Erickson DB_DNODE_EXIT(db); 1946fa9e4066Sahrens } 1947fa9e4066Sahrens 1948fa9e4066Sahrens /* 1949fa9e4066Sahrens * Faster still when you only care about the size. 1950fa9e4066Sahrens * This is specifically optimized for zfs_getattr(). 1951fa9e4066Sahrens */ 1952fa9e4066Sahrens void 1953744947dcSTom Erickson dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, 1954744947dcSTom Erickson u_longlong_t *nblk512) 1955fa9e4066Sahrens { 1956744947dcSTom Erickson dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 1957744947dcSTom Erickson dnode_t *dn; 1958744947dcSTom Erickson 1959744947dcSTom Erickson DB_DNODE_ENTER(db); 1960744947dcSTom Erickson dn = DB_DNODE(db); 1961fa9e4066Sahrens 1962fa9e4066Sahrens *blksize = dn->dn_datablksz; 196399653d4eSeschrock /* add 1 for dnode space */ 196499653d4eSeschrock *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> 196599653d4eSeschrock SPA_MINBLOCKSHIFT) + 1; 1966744947dcSTom Erickson DB_DNODE_EXIT(db); 1967fa9e4066Sahrens } 1968fa9e4066Sahrens 1969fa9e4066Sahrens void 1970fa9e4066Sahrens byteswap_uint64_array(void *vbuf, size_t size) 1971fa9e4066Sahrens { 1972fa9e4066Sahrens uint64_t *buf = vbuf; 1973fa9e4066Sahrens size_t count = size >> 3; 1974fa9e4066Sahrens int i; 1975fa9e4066Sahrens 1976fa9e4066Sahrens ASSERT((size & 7) == 0); 1977fa9e4066Sahrens 1978fa9e4066Sahrens for (i = 0; i < count; i++) 1979fa9e4066Sahrens buf[i] = BSWAP_64(buf[i]); 1980fa9e4066Sahrens } 1981fa9e4066Sahrens 1982fa9e4066Sahrens void 1983fa9e4066Sahrens byteswap_uint32_array(void *vbuf, size_t size) 1984fa9e4066Sahrens { 1985fa9e4066Sahrens uint32_t *buf = vbuf; 1986fa9e4066Sahrens size_t count = size >> 2; 1987fa9e4066Sahrens int i; 1988fa9e4066Sahrens 1989fa9e4066Sahrens ASSERT((size & 3) == 0); 1990fa9e4066Sahrens 1991fa9e4066Sahrens for (i = 0; i < count; i++) 1992fa9e4066Sahrens buf[i] = BSWAP_32(buf[i]); 1993fa9e4066Sahrens } 1994fa9e4066Sahrens 1995fa9e4066Sahrens void 1996fa9e4066Sahrens byteswap_uint16_array(void *vbuf, size_t size) 1997fa9e4066Sahrens { 1998fa9e4066Sahrens uint16_t *buf = vbuf; 1999fa9e4066Sahrens size_t count = size >> 1; 2000fa9e4066Sahrens int i; 2001fa9e4066Sahrens 2002fa9e4066Sahrens ASSERT((size & 1) == 0); 2003fa9e4066Sahrens 2004fa9e4066Sahrens for (i = 0; i < count; i++) 2005fa9e4066Sahrens buf[i] = BSWAP_16(buf[i]); 2006fa9e4066Sahrens } 2007fa9e4066Sahrens 2008fa9e4066Sahrens /* ARGSUSED */ 2009fa9e4066Sahrens void 2010fa9e4066Sahrens byteswap_uint8_array(void *vbuf, size_t size) 2011fa9e4066Sahrens { 2012fa9e4066Sahrens } 2013fa9e4066Sahrens 2014fa9e4066Sahrens void 2015fa9e4066Sahrens dmu_init(void) 2016fa9e4066Sahrens { 20173f9d6ad7SLin Ling zfs_dbgmsg_init(); 2018744947dcSTom Erickson sa_cache_init(); 2019744947dcSTom Erickson xuio_stat_init(); 2020744947dcSTom Erickson dmu_objset_init(); 2021fa9e4066Sahrens dnode_init(); 2022744947dcSTom Erickson dbuf_init(); 20237cbf8b43SRich Morris zfetch_init(); 2024fa94a07fSbrendan l2arc_init(); 2025ce636f8bSMatthew Ahrens arc_init(); 2026fa9e4066Sahrens } 2027fa9e4066Sahrens 2028fa9e4066Sahrens void 2029fa9e4066Sahrens dmu_fini(void) 2030fa9e4066Sahrens { 20313e30c24aSWill Andrews arc_fini(); /* arc depends on l2arc, so arc must go first */ 2032ce636f8bSMatthew Ahrens l2arc_fini(); 20337cbf8b43SRich Morris zfetch_fini(); 2034fa9e4066Sahrens dbuf_fini(); 2035744947dcSTom Erickson dnode_fini(); 2036744947dcSTom Erickson dmu_objset_fini(); 2037c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_fini(); 20380a586ceaSMark Shellenbaum sa_cache_fini(); 20393f9d6ad7SLin Ling zfs_dbgmsg_fini(); 2040fa9e4066Sahrens } 2041