dmu.h revision 288569
1167974Sdelphij/* 2167974Sdelphij * CDDL HEADER START 3167974Sdelphij * 478556Sobrien * The contents of this file are subject to the terms of the 5215041Sobrien * Common Development and Distribution License (the "License"). 6215041Sobrien * You may not use this file except in compliance with the License. 778556Sobrien * 8167974Sdelphij * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9167974Sdelphij * or http://www.opensolaris.org/os/licensing. 10167974Sdelphij * See the License for the specific language governing permissions 11167974Sdelphij * and limitations under the License. 12167974Sdelphij * 13167974Sdelphij * When distributing Covered Code, include this CDDL HEADER in each 14167974Sdelphij * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15167974Sdelphij * If applicable, add the following below this CDDL HEADER, with the 1678556Sobrien * fields enclosed by brackets "[]" replaced with your own identifying 1778556Sobrien * information: Portions Copyright [yyyy] [name of copyright owner] 1878556Sobrien * 1978556Sobrien * CDDL HEADER END 2078556Sobrien */ 2178556Sobrien 2278556Sobrien/* 2378556Sobrien * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 2478556Sobrien * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 2578556Sobrien * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 2678556Sobrien * Copyright (c) 2012, Joyent, Inc. All rights reserved. 2778556Sobrien * Copyright 2013 DEY Storage Systems, Inc. 2878556Sobrien * Copyright 2014 HybridCluster. All rights reserved. 2978556Sobrien * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 3078556Sobrien */ 3178556Sobrien 3278556Sobrien/* Portions Copyright 2010 Robert Milkowski */ 3378556Sobrien 3478556Sobrien#ifndef _SYS_DMU_H 3578556Sobrien#define _SYS_DMU_H 3678556Sobrien 3778556Sobrien/* 3878556Sobrien * This file describes the interface that the DMU provides for its 3978556Sobrien * consumers. 4078556Sobrien * 4178556Sobrien * The DMU also interacts with the SPA. That interface is described in 4278556Sobrien * dmu_spa.h. 4378556Sobrien */ 4478556Sobrien 4578556Sobrien#include <sys/zfs_context.h> 4678556Sobrien#include <sys/cred.h> 4778556Sobrien#include <sys/fs/zfs.h> 4878556Sobrien 4978556Sobrien#ifdef __cplusplus 5078556Sobrienextern "C" { 5178556Sobrien#endif 5278556Sobrien 5378556Sobrienstruct uio; 5478556Sobrienstruct xuio; 5578556Sobrienstruct page; 5678556Sobrienstruct vnode; 5778556Sobrienstruct spa; 5878556Sobrienstruct zilog; 5978556Sobrienstruct zio; 6078556Sobrienstruct blkptr; 6178556Sobrienstruct zap_cursor; 6278556Sobrienstruct dsl_dataset; 6378556Sobrienstruct dsl_pool; 6478556Sobrienstruct dnode; 6578556Sobrienstruct drr_begin; 6678556Sobrienstruct drr_end; 6778556Sobrienstruct zbookmark_phys; 6878556Sobrienstruct spa; 6978556Sobrienstruct nvlist; 7078556Sobrienstruct arc_buf; 7178556Sobrienstruct zio_prop; 7278556Sobrienstruct sa_handle; 7378556Sobrienstruct file; 7478556Sobrien 7578556Sobrientypedef struct objset objset_t; 7678556Sobrientypedef struct dmu_tx dmu_tx_t; 7778556Sobrientypedef struct dsl_dir dsl_dir_t; 7878556Sobrien 7978556Sobrientypedef enum dmu_object_byteswap { 8078556Sobrien DMU_BSWAP_UINT8, 8178556Sobrien DMU_BSWAP_UINT16, 8278556Sobrien DMU_BSWAP_UINT32, 8378556Sobrien DMU_BSWAP_UINT64, 8478556Sobrien DMU_BSWAP_ZAP, 8578556Sobrien DMU_BSWAP_DNODE, 8678556Sobrien DMU_BSWAP_OBJSET, 8778556Sobrien DMU_BSWAP_ZNODE, 8878556Sobrien DMU_BSWAP_OLDACL, 8978556Sobrien DMU_BSWAP_ACL, 9078556Sobrien /* 9178556Sobrien * Allocating a new byteswap type number makes the on-disk format 9278556Sobrien * incompatible with any other format that uses the same number. 9378556Sobrien * 9478556Sobrien * Data can usually be structured to work with one of the 9578556Sobrien * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types. 9678556Sobrien */ 9778556Sobrien DMU_BSWAP_NUMFUNCS 9878556Sobrien} dmu_object_byteswap_t; 9978556Sobrien 10078556Sobrien#define DMU_OT_NEWTYPE 0x80 10178556Sobrien#define DMU_OT_METADATA 0x40 10278556Sobrien#define DMU_OT_BYTESWAP_MASK 0x3f 10378556Sobrien 10478556Sobrien/* 10578556Sobrien * Defines a uint8_t object type. Object types specify if the data 10678556Sobrien * in the object is metadata (boolean) and how to byteswap the data 10778556Sobrien * (dmu_object_byteswap_t). 10878556Sobrien */ 10978556Sobrien#define DMU_OT(byteswap, metadata) \ 11078556Sobrien (DMU_OT_NEWTYPE | \ 11178556Sobrien ((metadata) ? DMU_OT_METADATA : 0) | \ 11278556Sobrien ((byteswap) & DMU_OT_BYTESWAP_MASK)) 11378556Sobrien 11478556Sobrien#define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 11578556Sobrien ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \ 11678556Sobrien (ot) < DMU_OT_NUMTYPES) 11778556Sobrien 11878556Sobrien#define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 11978556Sobrien ((ot) & DMU_OT_METADATA) : \ 12078556Sobrien dmu_ot[(ot)].ot_metadata) 12178556Sobrien 12278556Sobrien/* 12378556Sobrien * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't 12478556Sobrien * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill 12578556Sobrien * is repurposed for embedded BPs. 12678556Sobrien */ 12778556Sobrien#define DMU_OT_HAS_FILL(ot) \ 12878556Sobrien ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET) 12978556Sobrien 13078556Sobrien#define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 13178556Sobrien ((ot) & DMU_OT_BYTESWAP_MASK) : \ 13278556Sobrien dmu_ot[(ot)].ot_byteswap) 13378556Sobrien 13478556Sobrientypedef enum dmu_object_type { 13578556Sobrien DMU_OT_NONE, 13678556Sobrien /* general: */ 13778556Sobrien DMU_OT_OBJECT_DIRECTORY, /* ZAP */ 13878556Sobrien DMU_OT_OBJECT_ARRAY, /* UINT64 */ 13978556Sobrien DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */ 14078556Sobrien DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */ 14178556Sobrien DMU_OT_BPOBJ, /* UINT64 */ 14278556Sobrien DMU_OT_BPOBJ_HDR, /* UINT64 */ 14378556Sobrien /* spa: */ 14478556Sobrien DMU_OT_SPACE_MAP_HEADER, /* UINT64 */ 14578556Sobrien DMU_OT_SPACE_MAP, /* UINT64 */ 14678556Sobrien /* zil: */ 14778556Sobrien DMU_OT_INTENT_LOG, /* UINT64 */ 14878556Sobrien /* dmu: */ 14978556Sobrien DMU_OT_DNODE, /* DNODE */ 15090067Ssobomax DMU_OT_OBJSET, /* OBJSET */ 15178556Sobrien /* dsl: */ 15278556Sobrien DMU_OT_DSL_DIR, /* UINT64 */ 15378556Sobrien DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */ 15478556Sobrien DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */ 15578556Sobrien DMU_OT_DSL_PROPS, /* ZAP */ 15678556Sobrien DMU_OT_DSL_DATASET, /* UINT64 */ 15778556Sobrien /* zpl: */ 15878556Sobrien DMU_OT_ZNODE, /* ZNODE */ 15978556Sobrien DMU_OT_OLDACL, /* Old ACL */ 16078556Sobrien DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */ 16178556Sobrien DMU_OT_DIRECTORY_CONTENTS, /* ZAP */ 16278556Sobrien DMU_OT_MASTER_NODE, /* ZAP */ 16378556Sobrien DMU_OT_UNLINKED_SET, /* ZAP */ 16478556Sobrien /* zvol: */ 16578556Sobrien DMU_OT_ZVOL, /* UINT8 */ 16678556Sobrien DMU_OT_ZVOL_PROP, /* ZAP */ 16778556Sobrien /* other; for testing only! */ 16878556Sobrien DMU_OT_PLAIN_OTHER, /* UINT8 */ 16978556Sobrien DMU_OT_UINT64_OTHER, /* UINT64 */ 17078556Sobrien DMU_OT_ZAP_OTHER, /* ZAP */ 17178556Sobrien /* new object types: */ 17278556Sobrien DMU_OT_ERROR_LOG, /* ZAP */ 17378556Sobrien DMU_OT_SPA_HISTORY, /* UINT8 */ 17478556Sobrien DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */ 17578556Sobrien DMU_OT_POOL_PROPS, /* ZAP */ 17678556Sobrien DMU_OT_DSL_PERMS, /* ZAP */ 17778556Sobrien DMU_OT_ACL, /* ACL */ 17878556Sobrien DMU_OT_SYSACL, /* SYSACL */ 17978556Sobrien DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */ 18078556Sobrien DMU_OT_FUID_SIZE, /* FUID table size UINT64 */ 18190067Ssobomax DMU_OT_NEXT_CLONES, /* ZAP */ 18290067Ssobomax DMU_OT_SCAN_QUEUE, /* ZAP */ 18390067Ssobomax DMU_OT_USERGROUP_USED, /* ZAP */ 18490067Ssobomax DMU_OT_USERGROUP_QUOTA, /* ZAP */ 18590067Ssobomax DMU_OT_USERREFS, /* ZAP */ 18690067Ssobomax DMU_OT_DDT_ZAP, /* ZAP */ 18790067Ssobomax DMU_OT_DDT_STATS, /* ZAP */ 18890067Ssobomax DMU_OT_SA, /* System attr */ 18990067Ssobomax DMU_OT_SA_MASTER_NODE, /* ZAP */ 19090067Ssobomax DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */ 19190067Ssobomax DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */ 19290067Ssobomax DMU_OT_SCAN_XLATE, /* ZAP */ 19390067Ssobomax DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */ 19490067Ssobomax DMU_OT_DEADLIST, /* ZAP */ 19590067Ssobomax DMU_OT_DEADLIST_HDR, /* UINT64 */ 19690067Ssobomax DMU_OT_DSL_CLONES, /* ZAP */ 19790067Ssobomax DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */ 19890067Ssobomax /* 19990067Ssobomax * Do not allocate new object types here. Doing so makes the on-disk 20090067Ssobomax * format incompatible with any other format that uses the same object 20190067Ssobomax * type number. 20290067Ssobomax * 20390067Ssobomax * When creating an object which does not have one of the above types 20490067Ssobomax * use the DMU_OTN_* type with the correct byteswap and metadata 20590067Ssobomax * values. 20690067Ssobomax * 20790067Ssobomax * The DMU_OTN_* types do not have entries in the dmu_ot table, 20890067Ssobomax * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead 20990067Ssobomax * of indexing into dmu_ot directly (this works for both DMU_OT_* types 21090067Ssobomax * and DMU_OTN_* types). 21190067Ssobomax */ 21290067Ssobomax DMU_OT_NUMTYPES, 21390067Ssobomax 21490067Ssobomax /* 21590067Ssobomax * Names for valid types declared with DMU_OT(). 21690067Ssobomax */ 21790067Ssobomax DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE), 21890067Ssobomax DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE), 21990067Ssobomax DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE), 22090067Ssobomax DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE), 22190067Ssobomax DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE), 22290067Ssobomax DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE), 22390067Ssobomax DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE), 22490067Ssobomax DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE), 22590067Ssobomax DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE), 22690067Ssobomax DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE), 22790067Ssobomax} dmu_object_type_t; 22890067Ssobomax 22990067Ssobomaxtypedef enum txg_how { 23090067Ssobomax TXG_WAIT = 1, 23190067Ssobomax TXG_NOWAIT, 23290067Ssobomax TXG_WAITED, 23390067Ssobomax} txg_how_t; 23490067Ssobomax 23590067Ssobomaxvoid byteswap_uint64_array(void *buf, size_t size); 23690067Ssobomaxvoid byteswap_uint32_array(void *buf, size_t size); 23790067Ssobomaxvoid byteswap_uint16_array(void *buf, size_t size); 23890067Ssobomaxvoid byteswap_uint8_array(void *buf, size_t size); 23990067Ssobomaxvoid zap_byteswap(void *buf, size_t size); 24090067Ssobomaxvoid zfs_oldacl_byteswap(void *buf, size_t size); 24190067Ssobomaxvoid zfs_acl_byteswap(void *buf, size_t size); 24290067Ssobomaxvoid zfs_znode_byteswap(void *buf, size_t size); 24390067Ssobomax 24490067Ssobomax#define DS_FIND_SNAPSHOTS (1<<0) 24590067Ssobomax#define DS_FIND_CHILDREN (1<<1) 24690067Ssobomax#define DS_FIND_SERIALIZE (1<<2) 24790067Ssobomax 24890067Ssobomax/* 24990067Ssobomax * The maximum number of bytes that can be accessed as part of one 25090067Ssobomax * operation, including metadata. 25190067Ssobomax */ 25290067Ssobomax#define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */ 25390067Ssobomax#define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */ 25490067Ssobomax 25590067Ssobomax#define DMU_USERUSED_OBJECT (-1ULL) 25690067Ssobomax#define DMU_GROUPUSED_OBJECT (-2ULL) 25790067Ssobomax 25890067Ssobomax/* 25990067Ssobomax * artificial blkids for bonus buffer and spill blocks 26090067Ssobomax */ 26190067Ssobomax#define DMU_BONUS_BLKID (-1ULL) 26290067Ssobomax#define DMU_SPILL_BLKID (-2ULL) 26390067Ssobomax/* 26490067Ssobomax * Public routines to create, destroy, open, and close objsets. 26590067Ssobomax */ 26690067Ssobomaxint dmu_objset_hold(const char *name, void *tag, objset_t **osp); 267146293Sobrienint dmu_objset_own(const char *name, dmu_objset_type_t type, 268146293Sobrien boolean_t readonly, void *tag, objset_t **osp); 269146293Sobrienvoid dmu_objset_rele(objset_t *os, void *tag); 270146293Sobrienvoid dmu_objset_disown(objset_t *os, void *tag); 271146293Sobrienint dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp); 272146293Sobrien 273146293Sobrienvoid dmu_objset_evict_dbufs(objset_t *os); 274146293Sobrienint dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 275146293Sobrien void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg); 276146293Sobrienint dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 277167974Sdelphij struct nvlist *snaps); 278146293Sobrienint dmu_objset_clone(const char *name, const char *origin); 279146293Sobrienint dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer, 280146293Sobrien struct nvlist *errlist); 281146293Sobrienint dmu_objset_snapshot_one(const char *fsname, const char *snapname); 282146293Sobrienint dmu_objset_snapshot_tmp(const char *, const char *, int); 283146293Sobrienint dmu_objset_find(char *name, int func(const char *, void *), void *arg, 284146293Sobrien int flags); 285146293Sobrienvoid dmu_objset_byteswap(void *buf, size_t size); 286146293Sobrienint dsl_dataset_rename_snapshot(const char *fsname, 287146293Sobrien const char *oldsnapname, const char *newsnapname, boolean_t recursive); 288146293Sobrien 289167974Sdelphijtypedef struct dmu_buf { 290167974Sdelphij uint64_t db_object; /* object that this buffer is part of */ 291167974Sdelphij uint64_t db_offset; /* byte offset in this object */ 292167974Sdelphij uint64_t db_size; /* size of buffer in bytes */ 293167974Sdelphij void *db_data; /* data in buffer */ 294167974Sdelphij} dmu_buf_t; 295167974Sdelphij 296167974Sdelphij/* 297167974Sdelphij * The names of zap entries in the DIRECTORY_OBJECT of the MOS. 298167974Sdelphij */ 299167974Sdelphij#define DMU_POOL_DIRECTORY_OBJECT 1 300167974Sdelphij#define DMU_POOL_CONFIG "config" 301167974Sdelphij#define DMU_POOL_FEATURES_FOR_WRITE "features_for_write" 302167974Sdelphij#define DMU_POOL_FEATURES_FOR_READ "features_for_read" 303167974Sdelphij#define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions" 304167974Sdelphij#define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg" 305167974Sdelphij#define DMU_POOL_ROOT_DATASET "root_dataset" 306167974Sdelphij#define DMU_POOL_SYNC_BPOBJ "sync_bplist" 307167974Sdelphij#define DMU_POOL_ERRLOG_SCRUB "errlog_scrub" 308167974Sdelphij#define DMU_POOL_ERRLOG_LAST "errlog_last" 309167974Sdelphij#define DMU_POOL_SPARES "spares" 310167974Sdelphij#define DMU_POOL_DEFLATE "deflate" 311167974Sdelphij#define DMU_POOL_HISTORY "history" 312167974Sdelphij#define DMU_POOL_PROPS "pool_props" 313167974Sdelphij#define DMU_POOL_L2CACHE "l2cache" 314167974Sdelphij#define DMU_POOL_TMP_USERREFS "tmp_userrefs" 315177420Sdelphij#define DMU_POOL_DDT "DDT-%s-%s-%s" 316177420Sdelphij#define DMU_POOL_DDT_STATS "DDT-statistics" 317177420Sdelphij#define DMU_POOL_CREATION_VERSION "creation_version" 318177420Sdelphij#define DMU_POOL_SCAN "scan" 319177420Sdelphij#define DMU_POOL_FREE_BPOBJ "free_bpobj" 320215041Sobrien#define DMU_POOL_BPTREE_OBJ "bptree_obj" 321215041Sobrien#define DMU_POOL_EMPTY_BPOBJ "empty_bpobj" 322215041Sobrien 323215041Sobrien/* 324215041Sobrien * Allocate an object from this objset. The range of object numbers 325215041Sobrien * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode. 326215041Sobrien * 327215041Sobrien * The transaction must be assigned to a txg. The newly allocated 328 * object will be "held" in the transaction (ie. you can modify the 329 * newly allocated object in this transaction). 330 * 331 * dmu_object_alloc() chooses an object and returns it in *objectp. 332 * 333 * dmu_object_claim() allocates a specific object number. If that 334 * number is already allocated, it fails and returns EEXIST. 335 * 336 * Return 0 on success, or ENOSPC or EEXIST as specified above. 337 */ 338uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot, 339 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 340int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot, 341 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 342int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot, 343 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp); 344 345/* 346 * Free an object from this objset. 347 * 348 * The object's data will be freed as well (ie. you don't need to call 349 * dmu_free(object, 0, -1, tx)). 350 * 351 * The object need not be held in the transaction. 352 * 353 * If there are any holds on this object's buffers (via dmu_buf_hold()), 354 * or tx holds on the object (via dmu_tx_hold_object()), you can not 355 * free it; it fails and returns EBUSY. 356 * 357 * If the object is not allocated, it fails and returns ENOENT. 358 * 359 * Return 0 on success, or EBUSY or ENOENT as specified above. 360 */ 361int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx); 362 363/* 364 * Find the next allocated or free object. 365 * 366 * The objectp parameter is in-out. It will be updated to be the next 367 * object which is allocated. Ignore objects which have not been 368 * modified since txg. 369 * 370 * XXX Can only be called on a objset with no dirty data. 371 * 372 * Returns 0 on success, or ENOENT if there are no more objects. 373 */ 374int dmu_object_next(objset_t *os, uint64_t *objectp, 375 boolean_t hole, uint64_t txg); 376 377/* 378 * Set the data blocksize for an object. 379 * 380 * The object cannot have any blocks allcated beyond the first. If 381 * the first block is allocated already, the new size must be greater 382 * than the current block size. If these conditions are not met, 383 * ENOTSUP will be returned. 384 * 385 * Returns 0 on success, or EBUSY if there are any holds on the object 386 * contents, or ENOTSUP as described above. 387 */ 388int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, 389 int ibs, dmu_tx_t *tx); 390 391/* 392 * Set the checksum property on a dnode. The new checksum algorithm will 393 * apply to all newly written blocks; existing blocks will not be affected. 394 */ 395void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 396 dmu_tx_t *tx); 397 398/* 399 * Set the compress property on a dnode. The new compression algorithm will 400 * apply to all newly written blocks; existing blocks will not be affected. 401 */ 402void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 403 dmu_tx_t *tx); 404 405void 406dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, 407 void *data, uint8_t etype, uint8_t comp, int uncompressed_size, 408 int compressed_size, int byteorder, dmu_tx_t *tx); 409 410/* 411 * Decide how to write a block: checksum, compression, number of copies, etc. 412 */ 413#define WP_NOFILL 0x1 414#define WP_DMU_SYNC 0x2 415#define WP_SPILL 0x4 416 417void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp, 418 struct zio_prop *zp); 419/* 420 * The bonus data is accessed more or less like a regular buffer. 421 * You must dmu_bonus_hold() to get the buffer, which will give you a 422 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus 423 * data. As with any normal buffer, you must call dmu_buf_read() to 424 * read db_data, dmu_buf_will_dirty() before modifying it, and the 425 * object must be held in an assigned transaction before calling 426 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus 427 * buffer as well. You must release your hold with dmu_buf_rele(). 428 * 429 * Returns ENOENT, EIO, or 0. 430 */ 431int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **); 432int dmu_bonus_max(void); 433int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *); 434int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *); 435dmu_object_type_t dmu_get_bonustype(dmu_buf_t *); 436int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *); 437 438/* 439 * Special spill buffer support used by "SA" framework 440 */ 441 442int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 443int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags, 444 void *tag, dmu_buf_t **dbp); 445int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 446 447/* 448 * Obtain the DMU buffer from the specified object which contains the 449 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so 450 * that it will remain in memory. You must release the hold with 451 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your 452 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU. 453 * 454 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill 455 * on the returned buffer before reading or writing the buffer's 456 * db_data. The comments for those routines describe what particular 457 * operations are valid after calling them. 458 * 459 * The object number must be a valid, allocated object number. 460 */ 461int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 462 void *tag, dmu_buf_t **, int flags); 463 464/* 465 * Add a reference to a dmu buffer that has already been held via 466 * dmu_buf_hold() in the current context. 467 */ 468void dmu_buf_add_ref(dmu_buf_t *db, void* tag); 469 470/* 471 * Attempt to add a reference to a dmu buffer that is in an unknown state, 472 * using a pointer that may have been invalidated by eviction processing. 473 * The request will succeed if the passed in dbuf still represents the 474 * same os/object/blkid, is ineligible for eviction, and has at least 475 * one hold by a user other than the syncer. 476 */ 477boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object, 478 uint64_t blkid, void *tag); 479 480void dmu_buf_rele(dmu_buf_t *db, void *tag); 481uint64_t dmu_buf_refcount(dmu_buf_t *db); 482 483/* 484 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a 485 * range of an object. A pointer to an array of dmu_buf_t*'s is 486 * returned (in *dbpp). 487 * 488 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and 489 * frees the array. The hold on the array of buffers MUST be released 490 * with dmu_buf_rele_array. You can NOT release the hold on each buffer 491 * individually with dmu_buf_rele. 492 */ 493int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, 494 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp); 495void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag); 496 497typedef void dmu_buf_evict_func_t(void *user_ptr); 498 499/* 500 * A DMU buffer user object may be associated with a dbuf for the 501 * duration of its lifetime. This allows the user of a dbuf (client) 502 * to attach private data to a dbuf (e.g. in-core only data such as a 503 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified 504 * when that dbuf has been evicted. Clients typically respond to the 505 * eviction notification by freeing their private data, thus ensuring 506 * the same lifetime for both dbuf and private data. 507 * 508 * The mapping from a dmu_buf_user_t to any client private data is the 509 * client's responsibility. All current consumers of the API with private 510 * data embed a dmu_buf_user_t as the first member of the structure for 511 * their private data. This allows conversions between the two types 512 * with a simple cast. Since the DMU buf user API never needs access 513 * to the private data, other strategies can be employed if necessary 514 * or convenient for the client (e.g. using container_of() to do the 515 * conversion for private data that cannot have the dmu_buf_user_t as 516 * its first member). 517 * 518 * Eviction callbacks are executed without the dbuf mutex held or any 519 * other type of mechanism to guarantee that the dbuf is still available. 520 * For this reason, users must assume the dbuf has already been freed 521 * and not reference the dbuf from the callback context. 522 * 523 * Users requesting "immediate eviction" are notified as soon as the dbuf 524 * is only referenced by dirty records (dirties == holds). Otherwise the 525 * notification occurs after eviction processing for the dbuf begins. 526 */ 527typedef struct dmu_buf_user { 528 /* 529 * Asynchronous user eviction callback state. 530 */ 531 taskq_ent_t dbu_tqent; 532 533 /* This instance's eviction function pointer. */ 534 dmu_buf_evict_func_t *dbu_evict_func; 535#ifdef ZFS_DEBUG 536 /* 537 * Pointer to user's dbuf pointer. NULL for clients that do 538 * not associate a dbuf with their user data. 539 * 540 * The dbuf pointer is cleared upon eviction so as to catch 541 * use-after-evict bugs in clients. 542 */ 543 dmu_buf_t **dbu_clear_on_evict_dbufp; 544#endif 545} dmu_buf_user_t; 546 547/* 548 * Initialize the given dmu_buf_user_t instance with the eviction function 549 * evict_func, to be called when the user is evicted. 550 * 551 * NOTE: This function should only be called once on a given dmu_buf_user_t. 552 * To allow enforcement of this, dbu must already be zeroed on entry. 553 */ 554#ifdef __lint 555/* Very ugly, but it beats issuing suppression directives in many Makefiles. */ 556extern void 557dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func, 558 dmu_buf_t **clear_on_evict_dbufp); 559#else /* __lint */ 560inline void 561dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func, 562 dmu_buf_t **clear_on_evict_dbufp) 563{ 564 ASSERT(dbu->dbu_evict_func == NULL); 565 ASSERT(evict_func != NULL); 566 dbu->dbu_evict_func = evict_func; 567#ifdef ZFS_DEBUG 568 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp; 569#endif 570} 571#endif /* __lint */ 572 573/* 574 * Attach user data to a dbuf and mark it for normal (when the dbuf's 575 * data is cleared or its reference count goes to zero) eviction processing. 576 * 577 * Returns NULL on success, or the existing user if another user currently 578 * owns the buffer. 579 */ 580void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user); 581 582/* 583 * Attach user data to a dbuf and mark it for immediate (its dirty and 584 * reference counts are equal) eviction processing. 585 * 586 * Returns NULL on success, or the existing user if another user currently 587 * owns the buffer. 588 */ 589void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user); 590 591/* 592 * Replace the current user of a dbuf. 593 * 594 * If given the current user of a dbuf, replaces the dbuf's user with 595 * "new_user" and returns the user data pointer that was replaced. 596 * Otherwise returns the current, and unmodified, dbuf user pointer. 597 */ 598void *dmu_buf_replace_user(dmu_buf_t *db, 599 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user); 600 601/* 602 * Remove the specified user data for a DMU buffer. 603 * 604 * Returns the user that was removed on success, or the current user if 605 * another user currently owns the buffer. 606 */ 607void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user); 608 609/* 610 * Returns the user data (dmu_buf_user_t *) associated with this dbuf. 611 */ 612void *dmu_buf_get_user(dmu_buf_t *db); 613 614/* Block until any in-progress dmu buf user evictions complete. */ 615void dmu_buf_user_evict_wait(void); 616 617/* 618 * Returns the blkptr associated with this dbuf, or NULL if not set. 619 */ 620struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db); 621 622/* 623 * Indicate that you are going to modify the buffer's data (db_data). 624 * 625 * The transaction (tx) must be assigned to a txg (ie. you've called 626 * dmu_tx_assign()). The buffer's object must be held in the tx 627 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)). 628 */ 629void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx); 630 631/* 632 * Tells if the given dbuf is freeable. 633 */ 634boolean_t dmu_buf_freeable(dmu_buf_t *); 635 636/* 637 * You must create a transaction, then hold the objects which you will 638 * (or might) modify as part of this transaction. Then you must assign 639 * the transaction to a transaction group. Once the transaction has 640 * been assigned, you can modify buffers which belong to held objects as 641 * part of this transaction. You can't modify buffers before the 642 * transaction has been assigned; you can't modify buffers which don't 643 * belong to objects which this transaction holds; you can't hold 644 * objects once the transaction has been assigned. You may hold an 645 * object which you are going to free (with dmu_object_free()), but you 646 * don't have to. 647 * 648 * You can abort the transaction before it has been assigned. 649 * 650 * Note that you may hold buffers (with dmu_buf_hold) at any time, 651 * regardless of transaction state. 652 */ 653 654#define DMU_NEW_OBJECT (-1ULL) 655#define DMU_OBJECT_END (-1ULL) 656 657dmu_tx_t *dmu_tx_create(objset_t *os); 658void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len); 659void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, 660 uint64_t len); 661void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name); 662void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object); 663void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object); 664void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow); 665void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size); 666void dmu_tx_abort(dmu_tx_t *tx); 667int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how); 668void dmu_tx_wait(dmu_tx_t *tx); 669void dmu_tx_commit(dmu_tx_t *tx); 670void dmu_tx_mark_netfree(dmu_tx_t *tx); 671 672/* 673 * To register a commit callback, dmu_tx_callback_register() must be called. 674 * 675 * dcb_data is a pointer to caller private data that is passed on as a 676 * callback parameter. The caller is responsible for properly allocating and 677 * freeing it. 678 * 679 * When registering a callback, the transaction must be already created, but 680 * it cannot be committed or aborted. It can be assigned to a txg or not. 681 * 682 * The callback will be called after the transaction has been safely written 683 * to stable storage and will also be called if the dmu_tx is aborted. 684 * If there is any error which prevents the transaction from being committed to 685 * disk, the callback will be called with a value of error != 0. 686 */ 687typedef void dmu_tx_callback_func_t(void *dcb_data, int error); 688 689void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func, 690 void *dcb_data); 691 692/* 693 * Free up the data blocks for a defined range of a file. If size is 694 * -1, the range from offset to end-of-file is freed. 695 */ 696int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 697 uint64_t size, dmu_tx_t *tx); 698int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset, 699 uint64_t size); 700int dmu_free_long_object(objset_t *os, uint64_t object); 701 702/* 703 * Convenience functions. 704 * 705 * Canfail routines will return 0 on success, or an errno if there is a 706 * nonrecoverable I/O error. 707 */ 708#define DMU_READ_PREFETCH 0 /* prefetch */ 709#define DMU_READ_NO_PREFETCH 1 /* don't prefetch */ 710int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 711 void *buf, uint32_t flags); 712void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 713 const void *buf, dmu_tx_t *tx); 714void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 715 dmu_tx_t *tx); 716int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size); 717int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size); 718int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size, 719 dmu_tx_t *tx); 720int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size, 721 dmu_tx_t *tx); 722#ifdef _KERNEL 723#ifdef sun 724int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 725 uint64_t size, struct page *pp, dmu_tx_t *tx); 726#else 727int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 728 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx); 729#endif 730#endif 731struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size); 732void dmu_return_arcbuf(struct arc_buf *buf); 733void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf, 734 dmu_tx_t *tx); 735int dmu_xuio_init(struct xuio *uio, int niov); 736void dmu_xuio_fini(struct xuio *uio); 737int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off, 738 size_t n); 739int dmu_xuio_cnt(struct xuio *uio); 740struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i); 741void dmu_xuio_clear(struct xuio *uio, int i); 742void xuio_stat_wbuf_copied(); 743void xuio_stat_wbuf_nocopy(); 744 745extern int zfs_prefetch_disable; 746extern int zfs_max_recordsize; 747 748/* 749 * Asynchronously try to read in the data. 750 */ 751void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, 752 uint64_t len); 753 754typedef struct dmu_object_info { 755 /* All sizes are in bytes unless otherwise indicated. */ 756 uint32_t doi_data_block_size; 757 uint32_t doi_metadata_block_size; 758 dmu_object_type_t doi_type; 759 dmu_object_type_t doi_bonus_type; 760 uint64_t doi_bonus_size; 761 uint8_t doi_indirection; /* 2 = dnode->indirect->data */ 762 uint8_t doi_checksum; 763 uint8_t doi_compress; 764 uint8_t doi_nblkptr; 765 uint8_t doi_pad[4]; 766 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */ 767 uint64_t doi_max_offset; 768 uint64_t doi_fill_count; /* number of non-empty blocks */ 769} dmu_object_info_t; 770 771typedef void arc_byteswap_func_t(void *buf, size_t size); 772 773typedef struct dmu_object_type_info { 774 dmu_object_byteswap_t ot_byteswap; 775 boolean_t ot_metadata; 776 char *ot_name; 777} dmu_object_type_info_t; 778 779typedef struct dmu_object_byteswap_info { 780 arc_byteswap_func_t *ob_func; 781 char *ob_name; 782} dmu_object_byteswap_info_t; 783 784extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES]; 785extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS]; 786 787/* 788 * Get information on a DMU object. 789 * 790 * Return 0 on success or ENOENT if object is not allocated. 791 * 792 * If doi is NULL, just indicates whether the object exists. 793 */ 794int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi); 795/* Like dmu_object_info, but faster if you have a held dnode in hand. */ 796void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi); 797/* Like dmu_object_info, but faster if you have a held dbuf in hand. */ 798void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi); 799/* 800 * Like dmu_object_info_from_db, but faster still when you only care about 801 * the size. This is specifically optimized for zfs_getattr(). 802 */ 803void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, 804 u_longlong_t *nblk512); 805 806typedef struct dmu_objset_stats { 807 uint64_t dds_num_clones; /* number of clones of this */ 808 uint64_t dds_creation_txg; 809 uint64_t dds_guid; 810 dmu_objset_type_t dds_type; 811 uint8_t dds_is_snapshot; 812 uint8_t dds_inconsistent; 813 char dds_origin[MAXNAMELEN]; 814} dmu_objset_stats_t; 815 816/* 817 * Get stats on a dataset. 818 */ 819void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat); 820 821/* 822 * Add entries to the nvlist for all the objset's properties. See 823 * zfs_prop_table[] and zfs(1m) for details on the properties. 824 */ 825void dmu_objset_stats(objset_t *os, struct nvlist *nv); 826 827/* 828 * Get the space usage statistics for statvfs(). 829 * 830 * refdbytes is the amount of space "referenced" by this objset. 831 * availbytes is the amount of space available to this objset, taking 832 * into account quotas & reservations, assuming that no other objsets 833 * use the space first. These values correspond to the 'referenced' and 834 * 'available' properties, described in the zfs(1m) manpage. 835 * 836 * usedobjs and availobjs are the number of objects currently allocated, 837 * and available. 838 */ 839void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 840 uint64_t *usedobjsp, uint64_t *availobjsp); 841 842/* 843 * The fsid_guid is a 56-bit ID that can change to avoid collisions. 844 * (Contrast with the ds_guid which is a 64-bit ID that will never 845 * change, so there is a small probability that it will collide.) 846 */ 847uint64_t dmu_objset_fsid_guid(objset_t *os); 848 849/* 850 * Get the [cm]time for an objset's snapshot dir 851 */ 852timestruc_t dmu_objset_snap_cmtime(objset_t *os); 853 854int dmu_objset_is_snapshot(objset_t *os); 855 856extern struct spa *dmu_objset_spa(objset_t *os); 857extern struct zilog *dmu_objset_zil(objset_t *os); 858extern struct dsl_pool *dmu_objset_pool(objset_t *os); 859extern struct dsl_dataset *dmu_objset_ds(objset_t *os); 860extern void dmu_objset_name(objset_t *os, char *buf); 861extern dmu_objset_type_t dmu_objset_type(objset_t *os); 862extern uint64_t dmu_objset_id(objset_t *os); 863extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os); 864extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os); 865extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 866 uint64_t *id, uint64_t *offp, boolean_t *case_conflict); 867extern int dmu_snapshot_realname(objset_t *os, char *name, char *real, 868 int maxlen, boolean_t *conflict); 869extern int dmu_dir_list_next(objset_t *os, int namelen, char *name, 870 uint64_t *idp, uint64_t *offp); 871 872typedef int objset_used_cb_t(dmu_object_type_t bonustype, 873 void *bonus, uint64_t *userp, uint64_t *groupp); 874extern void dmu_objset_register_type(dmu_objset_type_t ost, 875 objset_used_cb_t *cb); 876extern void dmu_objset_set_user(objset_t *os, void *user_ptr); 877extern void *dmu_objset_get_user(objset_t *os); 878 879/* 880 * Return the txg number for the given assigned transaction. 881 */ 882uint64_t dmu_tx_get_txg(dmu_tx_t *tx); 883 884/* 885 * Synchronous write. 886 * If a parent zio is provided this function initiates a write on the 887 * provided buffer as a child of the parent zio. 888 * In the absence of a parent zio, the write is completed synchronously. 889 * At write completion, blk is filled with the bp of the written block. 890 * Note that while the data covered by this function will be on stable 891 * storage when the write completes this new data does not become a 892 * permanent part of the file until the associated transaction commits. 893 */ 894 895/* 896 * {zfs,zvol,ztest}_get_done() args 897 */ 898typedef struct zgd { 899 struct zilog *zgd_zilog; 900 struct blkptr *zgd_bp; 901 dmu_buf_t *zgd_db; 902 struct rl *zgd_rl; 903 void *zgd_private; 904} zgd_t; 905 906typedef void dmu_sync_cb_t(zgd_t *arg, int error); 907int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd); 908 909/* 910 * Find the next hole or data block in file starting at *off 911 * Return found offset in *off. Return ESRCH for end of file. 912 */ 913int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, 914 uint64_t *off); 915 916/* 917 * Initial setup and final teardown. 918 */ 919extern void dmu_init(void); 920extern void dmu_fini(void); 921 922typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp, 923 uint64_t object, uint64_t offset, int len); 924void dmu_traverse_objset(objset_t *os, uint64_t txg_start, 925 dmu_traverse_cb_t cb, void *arg); 926int dmu_diff(const char *tosnap_name, const char *fromsnap_name, 927 struct file *fp, offset_t *offp); 928 929/* CRC64 table */ 930#define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */ 931extern uint64_t zfs_crc64_table[256]; 932 933extern int zfs_mdcomp_disable; 934 935#ifdef __cplusplus 936} 937#endif 938 939#endif /* _SYS_DMU_H */ 940