dmu.h revision 290757
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 25 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2012, Joyent, Inc. All rights reserved. 27 * Copyright 2013 DEY Storage Systems, Inc. 28 * Copyright 2014 HybridCluster. All rights reserved. 29 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 30 * Copyright 2013 Saso Kiselkov. All rights reserved. 31 */ 32 33/* Portions Copyright 2010 Robert Milkowski */ 34 35#ifndef _SYS_DMU_H 36#define _SYS_DMU_H 37 38/* 39 * This file describes the interface that the DMU provides for its 40 * consumers. 41 * 42 * The DMU also interacts with the SPA. That interface is described in 43 * dmu_spa.h. 44 */ 45 46#include <sys/zfs_context.h> 47#include <sys/cred.h> 48#include <sys/fs/zfs.h> 49#include <sys/zio_priority.h> 50 51#ifdef __cplusplus 52extern "C" { 53#endif 54 55struct uio; 56struct xuio; 57struct page; 58struct vnode; 59struct spa; 60struct zilog; 61struct zio; 62struct blkptr; 63struct zap_cursor; 64struct dsl_dataset; 65struct dsl_pool; 66struct dnode; 67struct drr_begin; 68struct drr_end; 69struct zbookmark_phys; 70struct spa; 71struct nvlist; 72struct arc_buf; 73struct zio_prop; 74struct sa_handle; 75struct file; 76 77typedef struct objset objset_t; 78typedef struct dmu_tx dmu_tx_t; 79typedef struct dsl_dir dsl_dir_t; 80 81typedef enum dmu_object_byteswap { 82 DMU_BSWAP_UINT8, 83 DMU_BSWAP_UINT16, 84 DMU_BSWAP_UINT32, 85 DMU_BSWAP_UINT64, 86 DMU_BSWAP_ZAP, 87 DMU_BSWAP_DNODE, 88 DMU_BSWAP_OBJSET, 89 DMU_BSWAP_ZNODE, 90 DMU_BSWAP_OLDACL, 91 DMU_BSWAP_ACL, 92 /* 93 * Allocating a new byteswap type number makes the on-disk format 94 * incompatible with any other format that uses the same number. 95 * 96 * Data can usually be structured to work with one of the 97 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types. 98 */ 99 DMU_BSWAP_NUMFUNCS 100} dmu_object_byteswap_t; 101 102#define DMU_OT_NEWTYPE 0x80 103#define DMU_OT_METADATA 0x40 104#define DMU_OT_BYTESWAP_MASK 0x3f 105 106/* 107 * Defines a uint8_t object type. Object types specify if the data 108 * in the object is metadata (boolean) and how to byteswap the data 109 * (dmu_object_byteswap_t). 110 */ 111#define DMU_OT(byteswap, metadata) \ 112 (DMU_OT_NEWTYPE | \ 113 ((metadata) ? DMU_OT_METADATA : 0) | \ 114 ((byteswap) & DMU_OT_BYTESWAP_MASK)) 115 116#define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 117 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \ 118 (ot) < DMU_OT_NUMTYPES) 119 120#define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 121 ((ot) & DMU_OT_METADATA) : \ 122 dmu_ot[(ot)].ot_metadata) 123 124/* 125 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't 126 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill 127 * is repurposed for embedded BPs. 128 */ 129#define DMU_OT_HAS_FILL(ot) \ 130 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET) 131 132#define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 133 ((ot) & DMU_OT_BYTESWAP_MASK) : \ 134 dmu_ot[(ot)].ot_byteswap) 135 136typedef enum dmu_object_type { 137 DMU_OT_NONE, 138 /* general: */ 139 DMU_OT_OBJECT_DIRECTORY, /* ZAP */ 140 DMU_OT_OBJECT_ARRAY, /* UINT64 */ 141 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */ 142 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */ 143 DMU_OT_BPOBJ, /* UINT64 */ 144 DMU_OT_BPOBJ_HDR, /* UINT64 */ 145 /* spa: */ 146 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */ 147 DMU_OT_SPACE_MAP, /* UINT64 */ 148 /* zil: */ 149 DMU_OT_INTENT_LOG, /* UINT64 */ 150 /* dmu: */ 151 DMU_OT_DNODE, /* DNODE */ 152 DMU_OT_OBJSET, /* OBJSET */ 153 /* dsl: */ 154 DMU_OT_DSL_DIR, /* UINT64 */ 155 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */ 156 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */ 157 DMU_OT_DSL_PROPS, /* ZAP */ 158 DMU_OT_DSL_DATASET, /* UINT64 */ 159 /* zpl: */ 160 DMU_OT_ZNODE, /* ZNODE */ 161 DMU_OT_OLDACL, /* Old ACL */ 162 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */ 163 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */ 164 DMU_OT_MASTER_NODE, /* ZAP */ 165 DMU_OT_UNLINKED_SET, /* ZAP */ 166 /* zvol: */ 167 DMU_OT_ZVOL, /* UINT8 */ 168 DMU_OT_ZVOL_PROP, /* ZAP */ 169 /* other; for testing only! */ 170 DMU_OT_PLAIN_OTHER, /* UINT8 */ 171 DMU_OT_UINT64_OTHER, /* UINT64 */ 172 DMU_OT_ZAP_OTHER, /* ZAP */ 173 /* new object types: */ 174 DMU_OT_ERROR_LOG, /* ZAP */ 175 DMU_OT_SPA_HISTORY, /* UINT8 */ 176 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */ 177 DMU_OT_POOL_PROPS, /* ZAP */ 178 DMU_OT_DSL_PERMS, /* ZAP */ 179 DMU_OT_ACL, /* ACL */ 180 DMU_OT_SYSACL, /* SYSACL */ 181 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */ 182 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */ 183 DMU_OT_NEXT_CLONES, /* ZAP */ 184 DMU_OT_SCAN_QUEUE, /* ZAP */ 185 DMU_OT_USERGROUP_USED, /* ZAP */ 186 DMU_OT_USERGROUP_QUOTA, /* ZAP */ 187 DMU_OT_USERREFS, /* ZAP */ 188 DMU_OT_DDT_ZAP, /* ZAP */ 189 DMU_OT_DDT_STATS, /* ZAP */ 190 DMU_OT_SA, /* System attr */ 191 DMU_OT_SA_MASTER_NODE, /* ZAP */ 192 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */ 193 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */ 194 DMU_OT_SCAN_XLATE, /* ZAP */ 195 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */ 196 DMU_OT_DEADLIST, /* ZAP */ 197 DMU_OT_DEADLIST_HDR, /* UINT64 */ 198 DMU_OT_DSL_CLONES, /* ZAP */ 199 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */ 200 /* 201 * Do not allocate new object types here. Doing so makes the on-disk 202 * format incompatible with any other format that uses the same object 203 * type number. 204 * 205 * When creating an object which does not have one of the above types 206 * use the DMU_OTN_* type with the correct byteswap and metadata 207 * values. 208 * 209 * The DMU_OTN_* types do not have entries in the dmu_ot table, 210 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead 211 * of indexing into dmu_ot directly (this works for both DMU_OT_* types 212 * and DMU_OTN_* types). 213 */ 214 DMU_OT_NUMTYPES, 215 216 /* 217 * Names for valid types declared with DMU_OT(). 218 */ 219 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE), 220 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE), 221 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE), 222 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE), 223 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE), 224 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE), 225 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE), 226 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE), 227 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE), 228 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE), 229} dmu_object_type_t; 230 231typedef enum txg_how { 232 TXG_WAIT = 1, 233 TXG_NOWAIT, 234 TXG_WAITED, 235} txg_how_t; 236 237void byteswap_uint64_array(void *buf, size_t size); 238void byteswap_uint32_array(void *buf, size_t size); 239void byteswap_uint16_array(void *buf, size_t size); 240void byteswap_uint8_array(void *buf, size_t size); 241void zap_byteswap(void *buf, size_t size); 242void zfs_oldacl_byteswap(void *buf, size_t size); 243void zfs_acl_byteswap(void *buf, size_t size); 244void zfs_znode_byteswap(void *buf, size_t size); 245 246#define DS_FIND_SNAPSHOTS (1<<0) 247#define DS_FIND_CHILDREN (1<<1) 248#define DS_FIND_SERIALIZE (1<<2) 249 250/* 251 * The maximum number of bytes that can be accessed as part of one 252 * operation, including metadata. 253 */ 254#define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */ 255#define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */ 256 257#define DMU_USERUSED_OBJECT (-1ULL) 258#define DMU_GROUPUSED_OBJECT (-2ULL) 259 260/* 261 * artificial blkids for bonus buffer and spill blocks 262 */ 263#define DMU_BONUS_BLKID (-1ULL) 264#define DMU_SPILL_BLKID (-2ULL) 265/* 266 * Public routines to create, destroy, open, and close objsets. 267 */ 268int dmu_objset_hold(const char *name, void *tag, objset_t **osp); 269int dmu_objset_own(const char *name, dmu_objset_type_t type, 270 boolean_t readonly, void *tag, objset_t **osp); 271void dmu_objset_rele(objset_t *os, void *tag); 272void dmu_objset_disown(objset_t *os, void *tag); 273int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp); 274 275void dmu_objset_evict_dbufs(objset_t *os); 276int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 277 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg); 278int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 279 struct nvlist *snaps); 280int dmu_objset_clone(const char *name, const char *origin); 281int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer, 282 struct nvlist *errlist); 283int dmu_objset_snapshot_one(const char *fsname, const char *snapname); 284int dmu_objset_snapshot_tmp(const char *, const char *, int); 285int dmu_objset_find(char *name, int func(const char *, void *), void *arg, 286 int flags); 287void dmu_objset_byteswap(void *buf, size_t size); 288int dsl_dataset_rename_snapshot(const char *fsname, 289 const char *oldsnapname, const char *newsnapname, boolean_t recursive); 290 291typedef struct dmu_buf { 292 uint64_t db_object; /* object that this buffer is part of */ 293 uint64_t db_offset; /* byte offset in this object */ 294 uint64_t db_size; /* size of buffer in bytes */ 295 void *db_data; /* data in buffer */ 296} dmu_buf_t; 297 298/* 299 * The names of zap entries in the DIRECTORY_OBJECT of the MOS. 300 */ 301#define DMU_POOL_DIRECTORY_OBJECT 1 302#define DMU_POOL_CONFIG "config" 303#define DMU_POOL_FEATURES_FOR_WRITE "features_for_write" 304#define DMU_POOL_FEATURES_FOR_READ "features_for_read" 305#define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions" 306#define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg" 307#define DMU_POOL_ROOT_DATASET "root_dataset" 308#define DMU_POOL_SYNC_BPOBJ "sync_bplist" 309#define DMU_POOL_ERRLOG_SCRUB "errlog_scrub" 310#define DMU_POOL_ERRLOG_LAST "errlog_last" 311#define DMU_POOL_SPARES "spares" 312#define DMU_POOL_DEFLATE "deflate" 313#define DMU_POOL_HISTORY "history" 314#define DMU_POOL_PROPS "pool_props" 315#define DMU_POOL_L2CACHE "l2cache" 316#define DMU_POOL_TMP_USERREFS "tmp_userrefs" 317#define DMU_POOL_DDT "DDT-%s-%s-%s" 318#define DMU_POOL_DDT_STATS "DDT-statistics" 319#define DMU_POOL_CREATION_VERSION "creation_version" 320#define DMU_POOL_SCAN "scan" 321#define DMU_POOL_FREE_BPOBJ "free_bpobj" 322#define DMU_POOL_BPTREE_OBJ "bptree_obj" 323#define DMU_POOL_EMPTY_BPOBJ "empty_bpobj" 324#define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt" 325 326/* 327 * Allocate an object from this objset. The range of object numbers 328 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode. 329 * 330 * The transaction must be assigned to a txg. The newly allocated 331 * object will be "held" in the transaction (ie. you can modify the 332 * newly allocated object in this transaction). 333 * 334 * dmu_object_alloc() chooses an object and returns it in *objectp. 335 * 336 * dmu_object_claim() allocates a specific object number. If that 337 * number is already allocated, it fails and returns EEXIST. 338 * 339 * Return 0 on success, or ENOSPC or EEXIST as specified above. 340 */ 341uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot, 342 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 343int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot, 344 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 345int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot, 346 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp); 347 348/* 349 * Free an object from this objset. 350 * 351 * The object's data will be freed as well (ie. you don't need to call 352 * dmu_free(object, 0, -1, tx)). 353 * 354 * The object need not be held in the transaction. 355 * 356 * If there are any holds on this object's buffers (via dmu_buf_hold()), 357 * or tx holds on the object (via dmu_tx_hold_object()), you can not 358 * free it; it fails and returns EBUSY. 359 * 360 * If the object is not allocated, it fails and returns ENOENT. 361 * 362 * Return 0 on success, or EBUSY or ENOENT as specified above. 363 */ 364int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx); 365 366/* 367 * Find the next allocated or free object. 368 * 369 * The objectp parameter is in-out. It will be updated to be the next 370 * object which is allocated. Ignore objects which have not been 371 * modified since txg. 372 * 373 * XXX Can only be called on a objset with no dirty data. 374 * 375 * Returns 0 on success, or ENOENT if there are no more objects. 376 */ 377int dmu_object_next(objset_t *os, uint64_t *objectp, 378 boolean_t hole, uint64_t txg); 379 380/* 381 * Set the data blocksize for an object. 382 * 383 * The object cannot have any blocks allcated beyond the first. If 384 * the first block is allocated already, the new size must be greater 385 * than the current block size. If these conditions are not met, 386 * ENOTSUP will be returned. 387 * 388 * Returns 0 on success, or EBUSY if there are any holds on the object 389 * contents, or ENOTSUP as described above. 390 */ 391int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, 392 int ibs, dmu_tx_t *tx); 393 394/* 395 * Set the checksum property on a dnode. The new checksum algorithm will 396 * apply to all newly written blocks; existing blocks will not be affected. 397 */ 398void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 399 dmu_tx_t *tx); 400 401/* 402 * Set the compress property on a dnode. The new compression algorithm will 403 * apply to all newly written blocks; existing blocks will not be affected. 404 */ 405void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 406 dmu_tx_t *tx); 407 408void 409dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, 410 void *data, uint8_t etype, uint8_t comp, int uncompressed_size, 411 int compressed_size, int byteorder, dmu_tx_t *tx); 412 413/* 414 * Decide how to write a block: checksum, compression, number of copies, etc. 415 */ 416#define WP_NOFILL 0x1 417#define WP_DMU_SYNC 0x2 418#define WP_SPILL 0x4 419 420void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp, 421 struct zio_prop *zp); 422/* 423 * The bonus data is accessed more or less like a regular buffer. 424 * You must dmu_bonus_hold() to get the buffer, which will give you a 425 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus 426 * data. As with any normal buffer, you must call dmu_buf_read() to 427 * read db_data, dmu_buf_will_dirty() before modifying it, and the 428 * object must be held in an assigned transaction before calling 429 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus 430 * buffer as well. You must release your hold with dmu_buf_rele(). 431 * 432 * Returns ENOENT, EIO, or 0. 433 */ 434int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **); 435int dmu_bonus_max(void); 436int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *); 437int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *); 438dmu_object_type_t dmu_get_bonustype(dmu_buf_t *); 439int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *); 440 441/* 442 * Special spill buffer support used by "SA" framework 443 */ 444 445int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 446int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags, 447 void *tag, dmu_buf_t **dbp); 448int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 449 450/* 451 * Obtain the DMU buffer from the specified object which contains the 452 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so 453 * that it will remain in memory. You must release the hold with 454 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your 455 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU. 456 * 457 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill 458 * on the returned buffer before reading or writing the buffer's 459 * db_data. The comments for those routines describe what particular 460 * operations are valid after calling them. 461 * 462 * The object number must be a valid, allocated object number. 463 */ 464int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 465 void *tag, dmu_buf_t **, int flags); 466 467/* 468 * Add a reference to a dmu buffer that has already been held via 469 * dmu_buf_hold() in the current context. 470 */ 471void dmu_buf_add_ref(dmu_buf_t *db, void* tag); 472 473/* 474 * Attempt to add a reference to a dmu buffer that is in an unknown state, 475 * using a pointer that may have been invalidated by eviction processing. 476 * The request will succeed if the passed in dbuf still represents the 477 * same os/object/blkid, is ineligible for eviction, and has at least 478 * one hold by a user other than the syncer. 479 */ 480boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object, 481 uint64_t blkid, void *tag); 482 483void dmu_buf_rele(dmu_buf_t *db, void *tag); 484uint64_t dmu_buf_refcount(dmu_buf_t *db); 485 486/* 487 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a 488 * range of an object. A pointer to an array of dmu_buf_t*'s is 489 * returned (in *dbpp). 490 * 491 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and 492 * frees the array. The hold on the array of buffers MUST be released 493 * with dmu_buf_rele_array. You can NOT release the hold on each buffer 494 * individually with dmu_buf_rele. 495 */ 496int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, 497 uint64_t length, boolean_t read, void *tag, 498 int *numbufsp, dmu_buf_t ***dbpp); 499void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag); 500 501typedef void dmu_buf_evict_func_t(void *user_ptr); 502 503/* 504 * A DMU buffer user object may be associated with a dbuf for the 505 * duration of its lifetime. This allows the user of a dbuf (client) 506 * to attach private data to a dbuf (e.g. in-core only data such as a 507 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified 508 * when that dbuf has been evicted. Clients typically respond to the 509 * eviction notification by freeing their private data, thus ensuring 510 * the same lifetime for both dbuf and private data. 511 * 512 * The mapping from a dmu_buf_user_t to any client private data is the 513 * client's responsibility. All current consumers of the API with private 514 * data embed a dmu_buf_user_t as the first member of the structure for 515 * their private data. This allows conversions between the two types 516 * with a simple cast. Since the DMU buf user API never needs access 517 * to the private data, other strategies can be employed if necessary 518 * or convenient for the client (e.g. using container_of() to do the 519 * conversion for private data that cannot have the dmu_buf_user_t as 520 * its first member). 521 * 522 * Eviction callbacks are executed without the dbuf mutex held or any 523 * other type of mechanism to guarantee that the dbuf is still available. 524 * For this reason, users must assume the dbuf has already been freed 525 * and not reference the dbuf from the callback context. 526 * 527 * Users requesting "immediate eviction" are notified as soon as the dbuf 528 * is only referenced by dirty records (dirties == holds). Otherwise the 529 * notification occurs after eviction processing for the dbuf begins. 530 */ 531typedef struct dmu_buf_user { 532 /* 533 * Asynchronous user eviction callback state. 534 */ 535 taskq_ent_t dbu_tqent; 536 537 /* This instance's eviction function pointer. */ 538 dmu_buf_evict_func_t *dbu_evict_func; 539#ifdef ZFS_DEBUG 540 /* 541 * Pointer to user's dbuf pointer. NULL for clients that do 542 * not associate a dbuf with their user data. 543 * 544 * The dbuf pointer is cleared upon eviction so as to catch 545 * use-after-evict bugs in clients. 546 */ 547 dmu_buf_t **dbu_clear_on_evict_dbufp; 548#endif 549} dmu_buf_user_t; 550 551/* 552 * Initialize the given dmu_buf_user_t instance with the eviction function 553 * evict_func, to be called when the user is evicted. 554 * 555 * NOTE: This function should only be called once on a given dmu_buf_user_t. 556 * To allow enforcement of this, dbu must already be zeroed on entry. 557 */ 558#ifdef __lint 559/* Very ugly, but it beats issuing suppression directives in many Makefiles. */ 560extern 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#else /* __lint */ 564static inline void 565dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func, 566 dmu_buf_t **clear_on_evict_dbufp) 567{ 568 ASSERT(dbu->dbu_evict_func == NULL); 569 ASSERT(evict_func != NULL); 570 dbu->dbu_evict_func = evict_func; 571#ifdef ZFS_DEBUG 572 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp; 573#endif 574} 575#endif /* __lint */ 576 577/* 578 * Attach user data to a dbuf and mark it for normal (when the dbuf's 579 * data is cleared or its reference count goes to zero) eviction processing. 580 * 581 * Returns NULL on success, or the existing user if another user currently 582 * owns the buffer. 583 */ 584void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user); 585 586/* 587 * Attach user data to a dbuf and mark it for immediate (its dirty and 588 * reference counts are equal) eviction processing. 589 * 590 * Returns NULL on success, or the existing user if another user currently 591 * owns the buffer. 592 */ 593void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user); 594 595/* 596 * Replace the current user of a dbuf. 597 * 598 * If given the current user of a dbuf, replaces the dbuf's user with 599 * "new_user" and returns the user data pointer that was replaced. 600 * Otherwise returns the current, and unmodified, dbuf user pointer. 601 */ 602void *dmu_buf_replace_user(dmu_buf_t *db, 603 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user); 604 605/* 606 * Remove the specified user data for a DMU buffer. 607 * 608 * Returns the user that was removed on success, or the current user if 609 * another user currently owns the buffer. 610 */ 611void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user); 612 613/* 614 * Returns the user data (dmu_buf_user_t *) associated with this dbuf. 615 */ 616void *dmu_buf_get_user(dmu_buf_t *db); 617 618/* Block until any in-progress dmu buf user evictions complete. */ 619void dmu_buf_user_evict_wait(void); 620 621/* 622 * Returns the blkptr associated with this dbuf, or NULL if not set. 623 */ 624struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db); 625 626/* 627 * Indicate that you are going to modify the buffer's data (db_data). 628 * 629 * The transaction (tx) must be assigned to a txg (ie. you've called 630 * dmu_tx_assign()). The buffer's object must be held in the tx 631 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)). 632 */ 633void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx); 634 635/* 636 * Tells if the given dbuf is freeable. 637 */ 638boolean_t dmu_buf_freeable(dmu_buf_t *); 639 640/* 641 * You must create a transaction, then hold the objects which you will 642 * (or might) modify as part of this transaction. Then you must assign 643 * the transaction to a transaction group. Once the transaction has 644 * been assigned, you can modify buffers which belong to held objects as 645 * part of this transaction. You can't modify buffers before the 646 * transaction has been assigned; you can't modify buffers which don't 647 * belong to objects which this transaction holds; you can't hold 648 * objects once the transaction has been assigned. You may hold an 649 * object which you are going to free (with dmu_object_free()), but you 650 * don't have to. 651 * 652 * You can abort the transaction before it has been assigned. 653 * 654 * Note that you may hold buffers (with dmu_buf_hold) at any time, 655 * regardless of transaction state. 656 */ 657 658#define DMU_NEW_OBJECT (-1ULL) 659#define DMU_OBJECT_END (-1ULL) 660 661dmu_tx_t *dmu_tx_create(objset_t *os); 662void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len); 663void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, 664 uint64_t len); 665void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name); 666void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object); 667void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object); 668void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow); 669void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size); 670void dmu_tx_abort(dmu_tx_t *tx); 671int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how); 672void dmu_tx_wait(dmu_tx_t *tx); 673void dmu_tx_commit(dmu_tx_t *tx); 674void dmu_tx_mark_netfree(dmu_tx_t *tx); 675 676/* 677 * To register a commit callback, dmu_tx_callback_register() must be called. 678 * 679 * dcb_data is a pointer to caller private data that is passed on as a 680 * callback parameter. The caller is responsible for properly allocating and 681 * freeing it. 682 * 683 * When registering a callback, the transaction must be already created, but 684 * it cannot be committed or aborted. It can be assigned to a txg or not. 685 * 686 * The callback will be called after the transaction has been safely written 687 * to stable storage and will also be called if the dmu_tx is aborted. 688 * If there is any error which prevents the transaction from being committed to 689 * disk, the callback will be called with a value of error != 0. 690 */ 691typedef void dmu_tx_callback_func_t(void *dcb_data, int error); 692 693void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func, 694 void *dcb_data); 695 696/* 697 * Free up the data blocks for a defined range of a file. If size is 698 * -1, the range from offset to end-of-file is freed. 699 */ 700int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 701 uint64_t size, dmu_tx_t *tx); 702int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset, 703 uint64_t size); 704int dmu_free_long_object(objset_t *os, uint64_t object); 705 706/* 707 * Convenience functions. 708 * 709 * Canfail routines will return 0 on success, or an errno if there is a 710 * nonrecoverable I/O error. 711 */ 712#define DMU_READ_PREFETCH 0 /* prefetch */ 713#define DMU_READ_NO_PREFETCH 1 /* don't prefetch */ 714int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 715 void *buf, uint32_t flags); 716void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 717 const void *buf, dmu_tx_t *tx); 718void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 719 dmu_tx_t *tx); 720int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size); 721int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size); 722int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size, 723 dmu_tx_t *tx); 724int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size, 725 dmu_tx_t *tx); 726#ifdef _KERNEL 727#ifdef sun 728int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 729 uint64_t size, struct page *pp, dmu_tx_t *tx); 730#else 731int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 732 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx); 733#endif 734#endif 735struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size); 736void dmu_return_arcbuf(struct arc_buf *buf); 737void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf, 738 dmu_tx_t *tx); 739int dmu_xuio_init(struct xuio *uio, int niov); 740void dmu_xuio_fini(struct xuio *uio); 741int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off, 742 size_t n); 743int dmu_xuio_cnt(struct xuio *uio); 744struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i); 745void dmu_xuio_clear(struct xuio *uio, int i); 746void xuio_stat_wbuf_copied(); 747void xuio_stat_wbuf_nocopy(); 748 749extern boolean_t zfs_prefetch_disable; 750extern int zfs_max_recordsize; 751 752/* 753 * Asynchronously try to read in the data. 754 */ 755void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, 756 uint64_t len, enum zio_priority pri); 757 758typedef struct dmu_object_info { 759 /* All sizes are in bytes unless otherwise indicated. */ 760 uint32_t doi_data_block_size; 761 uint32_t doi_metadata_block_size; 762 dmu_object_type_t doi_type; 763 dmu_object_type_t doi_bonus_type; 764 uint64_t doi_bonus_size; 765 uint8_t doi_indirection; /* 2 = dnode->indirect->data */ 766 uint8_t doi_checksum; 767 uint8_t doi_compress; 768 uint8_t doi_nblkptr; 769 uint8_t doi_pad[4]; 770 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */ 771 uint64_t doi_max_offset; 772 uint64_t doi_fill_count; /* number of non-empty blocks */ 773} dmu_object_info_t; 774 775typedef void arc_byteswap_func_t(void *buf, size_t size); 776 777typedef struct dmu_object_type_info { 778 dmu_object_byteswap_t ot_byteswap; 779 boolean_t ot_metadata; 780 char *ot_name; 781} dmu_object_type_info_t; 782 783typedef struct dmu_object_byteswap_info { 784 arc_byteswap_func_t *ob_func; 785 char *ob_name; 786} dmu_object_byteswap_info_t; 787 788extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES]; 789extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS]; 790 791/* 792 * Get information on a DMU object. 793 * 794 * Return 0 on success or ENOENT if object is not allocated. 795 * 796 * If doi is NULL, just indicates whether the object exists. 797 */ 798int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi); 799/* Like dmu_object_info, but faster if you have a held dnode in hand. */ 800void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi); 801/* Like dmu_object_info, but faster if you have a held dbuf in hand. */ 802void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi); 803/* 804 * Like dmu_object_info_from_db, but faster still when you only care about 805 * the size. This is specifically optimized for zfs_getattr(). 806 */ 807void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, 808 u_longlong_t *nblk512); 809 810typedef struct dmu_objset_stats { 811 uint64_t dds_num_clones; /* number of clones of this */ 812 uint64_t dds_creation_txg; 813 uint64_t dds_guid; 814 dmu_objset_type_t dds_type; 815 uint8_t dds_is_snapshot; 816 uint8_t dds_inconsistent; 817 char dds_origin[MAXNAMELEN]; 818} dmu_objset_stats_t; 819 820/* 821 * Get stats on a dataset. 822 */ 823void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat); 824 825/* 826 * Add entries to the nvlist for all the objset's properties. See 827 * zfs_prop_table[] and zfs(1m) for details on the properties. 828 */ 829void dmu_objset_stats(objset_t *os, struct nvlist *nv); 830 831/* 832 * Get the space usage statistics for statvfs(). 833 * 834 * refdbytes is the amount of space "referenced" by this objset. 835 * availbytes is the amount of space available to this objset, taking 836 * into account quotas & reservations, assuming that no other objsets 837 * use the space first. These values correspond to the 'referenced' and 838 * 'available' properties, described in the zfs(1m) manpage. 839 * 840 * usedobjs and availobjs are the number of objects currently allocated, 841 * and available. 842 */ 843void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 844 uint64_t *usedobjsp, uint64_t *availobjsp); 845 846/* 847 * The fsid_guid is a 56-bit ID that can change to avoid collisions. 848 * (Contrast with the ds_guid which is a 64-bit ID that will never 849 * change, so there is a small probability that it will collide.) 850 */ 851uint64_t dmu_objset_fsid_guid(objset_t *os); 852 853/* 854 * Get the [cm]time for an objset's snapshot dir 855 */ 856timestruc_t dmu_objset_snap_cmtime(objset_t *os); 857 858int dmu_objset_is_snapshot(objset_t *os); 859 860extern struct spa *dmu_objset_spa(objset_t *os); 861extern struct zilog *dmu_objset_zil(objset_t *os); 862extern struct dsl_pool *dmu_objset_pool(objset_t *os); 863extern struct dsl_dataset *dmu_objset_ds(objset_t *os); 864extern void dmu_objset_name(objset_t *os, char *buf); 865extern dmu_objset_type_t dmu_objset_type(objset_t *os); 866extern uint64_t dmu_objset_id(objset_t *os); 867extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os); 868extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os); 869extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 870 uint64_t *id, uint64_t *offp, boolean_t *case_conflict); 871extern int dmu_snapshot_realname(objset_t *os, char *name, char *real, 872 int maxlen, boolean_t *conflict); 873extern int dmu_dir_list_next(objset_t *os, int namelen, char *name, 874 uint64_t *idp, uint64_t *offp); 875 876typedef int objset_used_cb_t(dmu_object_type_t bonustype, 877 void *bonus, uint64_t *userp, uint64_t *groupp); 878extern void dmu_objset_register_type(dmu_objset_type_t ost, 879 objset_used_cb_t *cb); 880extern void dmu_objset_set_user(objset_t *os, void *user_ptr); 881extern void *dmu_objset_get_user(objset_t *os); 882 883/* 884 * Return the txg number for the given assigned transaction. 885 */ 886uint64_t dmu_tx_get_txg(dmu_tx_t *tx); 887 888/* 889 * Synchronous write. 890 * If a parent zio is provided this function initiates a write on the 891 * provided buffer as a child of the parent zio. 892 * In the absence of a parent zio, the write is completed synchronously. 893 * At write completion, blk is filled with the bp of the written block. 894 * Note that while the data covered by this function will be on stable 895 * storage when the write completes this new data does not become a 896 * permanent part of the file until the associated transaction commits. 897 */ 898 899/* 900 * {zfs,zvol,ztest}_get_done() args 901 */ 902typedef struct zgd { 903 struct zilog *zgd_zilog; 904 struct blkptr *zgd_bp; 905 dmu_buf_t *zgd_db; 906 struct rl *zgd_rl; 907 void *zgd_private; 908} zgd_t; 909 910typedef void dmu_sync_cb_t(zgd_t *arg, int error); 911int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd); 912 913/* 914 * Find the next hole or data block in file starting at *off 915 * Return found offset in *off. Return ESRCH for end of file. 916 */ 917int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, 918 uint64_t *off); 919 920/* 921 * Check if a DMU object has any dirty blocks. If so, sync out 922 * all pending transaction groups. Otherwise, this function 923 * does not alter DMU state. This could be improved to only sync 924 * out the necessary transaction groups for this particular 925 * object. 926 */ 927int dmu_object_wait_synced(objset_t *os, uint64_t object); 928 929/* 930 * Initial setup and final teardown. 931 */ 932extern void dmu_init(void); 933extern void dmu_fini(void); 934 935typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp, 936 uint64_t object, uint64_t offset, int len); 937void dmu_traverse_objset(objset_t *os, uint64_t txg_start, 938 dmu_traverse_cb_t cb, void *arg); 939int dmu_diff(const char *tosnap_name, const char *fromsnap_name, 940 struct file *fp, offset_t *offp); 941 942/* CRC64 table */ 943#define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */ 944extern uint64_t zfs_crc64_table[256]; 945 946extern int zfs_mdcomp_disable; 947 948#ifdef __cplusplus 949} 950#endif 951 952#endif /* _SYS_DMU_H */ 953