dmu.h revision 307292
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, 2016 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 * Copyright (c) 2014 Integros [integros.com] 32 */ 33 34/* Portions Copyright 2010 Robert Milkowski */ 35 36#ifndef _SYS_DMU_H 37#define _SYS_DMU_H 38 39/* 40 * This file describes the interface that the DMU provides for its 41 * consumers. 42 * 43 * The DMU also interacts with the SPA. That interface is described in 44 * dmu_spa.h. 45 */ 46 47#include <sys/zfs_context.h> 48#include <sys/cred.h> 49#include <sys/fs/zfs.h> 50#include <sys/zio_priority.h> 51 52#ifdef __cplusplus 53extern "C" { 54#endif 55 56struct uio; 57struct xuio; 58struct page; 59struct vnode; 60struct spa; 61struct zilog; 62struct zio; 63struct blkptr; 64struct zap_cursor; 65struct dsl_dataset; 66struct dsl_pool; 67struct dnode; 68struct drr_begin; 69struct drr_end; 70struct zbookmark_phys; 71struct spa; 72struct nvlist; 73struct arc_buf; 74struct zio_prop; 75struct sa_handle; 76struct file; 77 78typedef struct objset objset_t; 79typedef struct dmu_tx dmu_tx_t; 80typedef struct dsl_dir dsl_dir_t; 81typedef struct dnode dnode_t; 82 83typedef enum dmu_object_byteswap { 84 DMU_BSWAP_UINT8, 85 DMU_BSWAP_UINT16, 86 DMU_BSWAP_UINT32, 87 DMU_BSWAP_UINT64, 88 DMU_BSWAP_ZAP, 89 DMU_BSWAP_DNODE, 90 DMU_BSWAP_OBJSET, 91 DMU_BSWAP_ZNODE, 92 DMU_BSWAP_OLDACL, 93 DMU_BSWAP_ACL, 94 /* 95 * Allocating a new byteswap type number makes the on-disk format 96 * incompatible with any other format that uses the same number. 97 * 98 * Data can usually be structured to work with one of the 99 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types. 100 */ 101 DMU_BSWAP_NUMFUNCS 102} dmu_object_byteswap_t; 103 104#define DMU_OT_NEWTYPE 0x80 105#define DMU_OT_METADATA 0x40 106#define DMU_OT_BYTESWAP_MASK 0x3f 107 108/* 109 * Defines a uint8_t object type. Object types specify if the data 110 * in the object is metadata (boolean) and how to byteswap the data 111 * (dmu_object_byteswap_t). 112 */ 113#define DMU_OT(byteswap, metadata) \ 114 (DMU_OT_NEWTYPE | \ 115 ((metadata) ? DMU_OT_METADATA : 0) | \ 116 ((byteswap) & DMU_OT_BYTESWAP_MASK)) 117 118#define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 119 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \ 120 (ot) < DMU_OT_NUMTYPES) 121 122#define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 123 ((ot) & DMU_OT_METADATA) : \ 124 dmu_ot[(ot)].ot_metadata) 125 126/* 127 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't 128 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill 129 * is repurposed for embedded BPs. 130 */ 131#define DMU_OT_HAS_FILL(ot) \ 132 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET) 133 134#define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 135 ((ot) & DMU_OT_BYTESWAP_MASK) : \ 136 dmu_ot[(ot)].ot_byteswap) 137 138typedef enum dmu_object_type { 139 DMU_OT_NONE, 140 /* general: */ 141 DMU_OT_OBJECT_DIRECTORY, /* ZAP */ 142 DMU_OT_OBJECT_ARRAY, /* UINT64 */ 143 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */ 144 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */ 145 DMU_OT_BPOBJ, /* UINT64 */ 146 DMU_OT_BPOBJ_HDR, /* UINT64 */ 147 /* spa: */ 148 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */ 149 DMU_OT_SPACE_MAP, /* UINT64 */ 150 /* zil: */ 151 DMU_OT_INTENT_LOG, /* UINT64 */ 152 /* dmu: */ 153 DMU_OT_DNODE, /* DNODE */ 154 DMU_OT_OBJSET, /* OBJSET */ 155 /* dsl: */ 156 DMU_OT_DSL_DIR, /* UINT64 */ 157 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */ 158 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */ 159 DMU_OT_DSL_PROPS, /* ZAP */ 160 DMU_OT_DSL_DATASET, /* UINT64 */ 161 /* zpl: */ 162 DMU_OT_ZNODE, /* ZNODE */ 163 DMU_OT_OLDACL, /* Old ACL */ 164 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */ 165 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */ 166 DMU_OT_MASTER_NODE, /* ZAP */ 167 DMU_OT_UNLINKED_SET, /* ZAP */ 168 /* zvol: */ 169 DMU_OT_ZVOL, /* UINT8 */ 170 DMU_OT_ZVOL_PROP, /* ZAP */ 171 /* other; for testing only! */ 172 DMU_OT_PLAIN_OTHER, /* UINT8 */ 173 DMU_OT_UINT64_OTHER, /* UINT64 */ 174 DMU_OT_ZAP_OTHER, /* ZAP */ 175 /* new object types: */ 176 DMU_OT_ERROR_LOG, /* ZAP */ 177 DMU_OT_SPA_HISTORY, /* UINT8 */ 178 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */ 179 DMU_OT_POOL_PROPS, /* ZAP */ 180 DMU_OT_DSL_PERMS, /* ZAP */ 181 DMU_OT_ACL, /* ACL */ 182 DMU_OT_SYSACL, /* SYSACL */ 183 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */ 184 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */ 185 DMU_OT_NEXT_CLONES, /* ZAP */ 186 DMU_OT_SCAN_QUEUE, /* ZAP */ 187 DMU_OT_USERGROUP_USED, /* ZAP */ 188 DMU_OT_USERGROUP_QUOTA, /* ZAP */ 189 DMU_OT_USERREFS, /* ZAP */ 190 DMU_OT_DDT_ZAP, /* ZAP */ 191 DMU_OT_DDT_STATS, /* ZAP */ 192 DMU_OT_SA, /* System attr */ 193 DMU_OT_SA_MASTER_NODE, /* ZAP */ 194 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */ 195 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */ 196 DMU_OT_SCAN_XLATE, /* ZAP */ 197 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */ 198 DMU_OT_DEADLIST, /* ZAP */ 199 DMU_OT_DEADLIST_HDR, /* UINT64 */ 200 DMU_OT_DSL_CLONES, /* ZAP */ 201 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */ 202 /* 203 * Do not allocate new object types here. Doing so makes the on-disk 204 * format incompatible with any other format that uses the same object 205 * type number. 206 * 207 * When creating an object which does not have one of the above types 208 * use the DMU_OTN_* type with the correct byteswap and metadata 209 * values. 210 * 211 * The DMU_OTN_* types do not have entries in the dmu_ot table, 212 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead 213 * of indexing into dmu_ot directly (this works for both DMU_OT_* types 214 * and DMU_OTN_* types). 215 */ 216 DMU_OT_NUMTYPES, 217 218 /* 219 * Names for valid types declared with DMU_OT(). 220 */ 221 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE), 222 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE), 223 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE), 224 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE), 225 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE), 226 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE), 227 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE), 228 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE), 229 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE), 230 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE), 231} dmu_object_type_t; 232 233typedef enum txg_how { 234 TXG_WAIT = 1, 235 TXG_NOWAIT, 236 TXG_WAITED, 237} txg_how_t; 238 239void byteswap_uint64_array(void *buf, size_t size); 240void byteswap_uint32_array(void *buf, size_t size); 241void byteswap_uint16_array(void *buf, size_t size); 242void byteswap_uint8_array(void *buf, size_t size); 243void zap_byteswap(void *buf, size_t size); 244void zfs_oldacl_byteswap(void *buf, size_t size); 245void zfs_acl_byteswap(void *buf, size_t size); 246void zfs_znode_byteswap(void *buf, size_t size); 247 248#define DS_FIND_SNAPSHOTS (1<<0) 249#define DS_FIND_CHILDREN (1<<1) 250#define DS_FIND_SERIALIZE (1<<2) 251 252/* 253 * The maximum number of bytes that can be accessed as part of one 254 * operation, including metadata. 255 */ 256#define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */ 257#define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */ 258 259#define DMU_USERUSED_OBJECT (-1ULL) 260#define DMU_GROUPUSED_OBJECT (-2ULL) 261 262/* 263 * artificial blkids for bonus buffer and spill blocks 264 */ 265#define DMU_BONUS_BLKID (-1ULL) 266#define DMU_SPILL_BLKID (-2ULL) 267/* 268 * Public routines to create, destroy, open, and close objsets. 269 */ 270int dmu_objset_hold(const char *name, void *tag, objset_t **osp); 271int dmu_objset_own(const char *name, dmu_objset_type_t type, 272 boolean_t readonly, void *tag, objset_t **osp); 273void dmu_objset_rele(objset_t *os, void *tag); 274void dmu_objset_disown(objset_t *os, void *tag); 275int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp); 276 277void dmu_objset_evict_dbufs(objset_t *os); 278int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 279 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg); 280int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 281 struct nvlist *snaps); 282int dmu_objset_clone(const char *name, const char *origin); 283int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer, 284 struct nvlist *errlist); 285int dmu_objset_snapshot_one(const char *fsname, const char *snapname); 286int dmu_objset_snapshot_tmp(const char *, const char *, int); 287int dmu_objset_find(char *name, int func(const char *, void *), void *arg, 288 int flags); 289void dmu_objset_byteswap(void *buf, size_t size); 290int dsl_dataset_rename_snapshot(const char *fsname, 291 const char *oldsnapname, const char *newsnapname, boolean_t recursive); 292 293typedef struct dmu_buf { 294 uint64_t db_object; /* object that this buffer is part of */ 295 uint64_t db_offset; /* byte offset in this object */ 296 uint64_t db_size; /* size of buffer in bytes */ 297 void *db_data; /* data in buffer */ 298} dmu_buf_t; 299 300/* 301 * The names of zap entries in the DIRECTORY_OBJECT of the MOS. 302 */ 303#define DMU_POOL_DIRECTORY_OBJECT 1 304#define DMU_POOL_CONFIG "config" 305#define DMU_POOL_FEATURES_FOR_WRITE "features_for_write" 306#define DMU_POOL_FEATURES_FOR_READ "features_for_read" 307#define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions" 308#define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg" 309#define DMU_POOL_ROOT_DATASET "root_dataset" 310#define DMU_POOL_SYNC_BPOBJ "sync_bplist" 311#define DMU_POOL_ERRLOG_SCRUB "errlog_scrub" 312#define DMU_POOL_ERRLOG_LAST "errlog_last" 313#define DMU_POOL_SPARES "spares" 314#define DMU_POOL_DEFLATE "deflate" 315#define DMU_POOL_HISTORY "history" 316#define DMU_POOL_PROPS "pool_props" 317#define DMU_POOL_L2CACHE "l2cache" 318#define DMU_POOL_TMP_USERREFS "tmp_userrefs" 319#define DMU_POOL_DDT "DDT-%s-%s-%s" 320#define DMU_POOL_DDT_STATS "DDT-statistics" 321#define DMU_POOL_CREATION_VERSION "creation_version" 322#define DMU_POOL_SCAN "scan" 323#define DMU_POOL_FREE_BPOBJ "free_bpobj" 324#define DMU_POOL_BPTREE_OBJ "bptree_obj" 325#define DMU_POOL_EMPTY_BPOBJ "empty_bpobj" 326#define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt" 327 328/* 329 * Allocate an object from this objset. The range of object numbers 330 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode. 331 * 332 * The transaction must be assigned to a txg. The newly allocated 333 * object will be "held" in the transaction (ie. you can modify the 334 * newly allocated object in this transaction). 335 * 336 * dmu_object_alloc() chooses an object and returns it in *objectp. 337 * 338 * dmu_object_claim() allocates a specific object number. If that 339 * number is already allocated, it fails and returns EEXIST. 340 * 341 * Return 0 on success, or ENOSPC or EEXIST as specified above. 342 */ 343uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot, 344 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 345int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot, 346 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 347int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot, 348 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp); 349 350/* 351 * Free an object from this objset. 352 * 353 * The object's data will be freed as well (ie. you don't need to call 354 * dmu_free(object, 0, -1, tx)). 355 * 356 * The object need not be held in the transaction. 357 * 358 * If there are any holds on this object's buffers (via dmu_buf_hold()), 359 * or tx holds on the object (via dmu_tx_hold_object()), you can not 360 * free it; it fails and returns EBUSY. 361 * 362 * If the object is not allocated, it fails and returns ENOENT. 363 * 364 * Return 0 on success, or EBUSY or ENOENT as specified above. 365 */ 366int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx); 367 368/* 369 * Find the next allocated or free object. 370 * 371 * The objectp parameter is in-out. It will be updated to be the next 372 * object which is allocated. Ignore objects which have not been 373 * modified since txg. 374 * 375 * XXX Can only be called on a objset with no dirty data. 376 * 377 * Returns 0 on success, or ENOENT if there are no more objects. 378 */ 379int dmu_object_next(objset_t *os, uint64_t *objectp, 380 boolean_t hole, uint64_t txg); 381 382/* 383 * Set the data blocksize for an object. 384 * 385 * The object cannot have any blocks allcated beyond the first. If 386 * the first block is allocated already, the new size must be greater 387 * than the current block size. If these conditions are not met, 388 * ENOTSUP will be returned. 389 * 390 * Returns 0 on success, or EBUSY if there are any holds on the object 391 * contents, or ENOTSUP as described above. 392 */ 393int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, 394 int ibs, dmu_tx_t *tx); 395 396/* 397 * Set the checksum property on a dnode. The new checksum algorithm will 398 * apply to all newly written blocks; existing blocks will not be affected. 399 */ 400void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 401 dmu_tx_t *tx); 402 403/* 404 * Set the compress property on a dnode. The new compression algorithm will 405 * apply to all newly written blocks; existing blocks will not be affected. 406 */ 407void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 408 dmu_tx_t *tx); 409 410void 411dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, 412 void *data, uint8_t etype, uint8_t comp, int uncompressed_size, 413 int compressed_size, int byteorder, dmu_tx_t *tx); 414 415/* 416 * Decide how to write a block: checksum, compression, number of copies, etc. 417 */ 418#define WP_NOFILL 0x1 419#define WP_DMU_SYNC 0x2 420#define WP_SPILL 0x4 421 422void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, 423 struct zio_prop *zp); 424/* 425 * The bonus data is accessed more or less like a regular buffer. 426 * You must dmu_bonus_hold() to get the buffer, which will give you a 427 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus 428 * data. As with any normal buffer, you must call dmu_buf_read() to 429 * read db_data, dmu_buf_will_dirty() before modifying it, and the 430 * object must be held in an assigned transaction before calling 431 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus 432 * buffer as well. You must release your hold with dmu_buf_rele(). 433 * 434 * Returns ENOENT, EIO, or 0. 435 */ 436int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **); 437int dmu_bonus_max(void); 438int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *); 439int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *); 440dmu_object_type_t dmu_get_bonustype(dmu_buf_t *); 441int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *); 442 443/* 444 * Special spill buffer support used by "SA" framework 445 */ 446 447int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 448int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, 449 void *tag, dmu_buf_t **dbp); 450int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 451 452/* 453 * Obtain the DMU buffer from the specified object which contains the 454 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so 455 * that it will remain in memory. You must release the hold with 456 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your 457 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU. 458 * 459 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill 460 * on the returned buffer before reading or writing the buffer's 461 * db_data. The comments for those routines describe what particular 462 * operations are valid after calling them. 463 * 464 * The object number must be a valid, allocated object number. 465 */ 466int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 467 void *tag, dmu_buf_t **, int flags); 468int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, 469 void *tag, dmu_buf_t **dbp, int flags); 470 471/* 472 * Add a reference to a dmu buffer that has already been held via 473 * dmu_buf_hold() in the current context. 474 */ 475void dmu_buf_add_ref(dmu_buf_t *db, void* tag); 476 477/* 478 * Attempt to add a reference to a dmu buffer that is in an unknown state, 479 * using a pointer that may have been invalidated by eviction processing. 480 * The request will succeed if the passed in dbuf still represents the 481 * same os/object/blkid, is ineligible for eviction, and has at least 482 * one hold by a user other than the syncer. 483 */ 484boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object, 485 uint64_t blkid, void *tag); 486 487void dmu_buf_rele(dmu_buf_t *db, void *tag); 488uint64_t dmu_buf_refcount(dmu_buf_t *db); 489 490/* 491 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a 492 * range of an object. A pointer to an array of dmu_buf_t*'s is 493 * returned (in *dbpp). 494 * 495 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and 496 * frees the array. The hold on the array of buffers MUST be released 497 * with dmu_buf_rele_array. You can NOT release the hold on each buffer 498 * individually with dmu_buf_rele. 499 */ 500int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, 501 uint64_t length, boolean_t read, void *tag, 502 int *numbufsp, dmu_buf_t ***dbpp); 503void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag); 504 505typedef void dmu_buf_evict_func_t(void *user_ptr); 506 507/* 508 * A DMU buffer user object may be associated with a dbuf for the 509 * duration of its lifetime. This allows the user of a dbuf (client) 510 * to attach private data to a dbuf (e.g. in-core only data such as a 511 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified 512 * when that dbuf has been evicted. Clients typically respond to the 513 * eviction notification by freeing their private data, thus ensuring 514 * the same lifetime for both dbuf and private data. 515 * 516 * The mapping from a dmu_buf_user_t to any client private data is the 517 * client's responsibility. All current consumers of the API with private 518 * data embed a dmu_buf_user_t as the first member of the structure for 519 * their private data. This allows conversions between the two types 520 * with a simple cast. Since the DMU buf user API never needs access 521 * to the private data, other strategies can be employed if necessary 522 * or convenient for the client (e.g. using container_of() to do the 523 * conversion for private data that cannot have the dmu_buf_user_t as 524 * its first member). 525 * 526 * Eviction callbacks are executed without the dbuf mutex held or any 527 * other type of mechanism to guarantee that the dbuf is still available. 528 * For this reason, users must assume the dbuf has already been freed 529 * and not reference the dbuf from the callback context. 530 * 531 * Users requesting "immediate eviction" are notified as soon as the dbuf 532 * is only referenced by dirty records (dirties == holds). Otherwise the 533 * notification occurs after eviction processing for the dbuf begins. 534 */ 535typedef struct dmu_buf_user { 536 /* 537 * Asynchronous user eviction callback state. 538 */ 539 taskq_ent_t dbu_tqent; 540 541 /* This instance's eviction function pointer. */ 542 dmu_buf_evict_func_t *dbu_evict_func; 543#ifdef ZFS_DEBUG 544 /* 545 * Pointer to user's dbuf pointer. NULL for clients that do 546 * not associate a dbuf with their user data. 547 * 548 * The dbuf pointer is cleared upon eviction so as to catch 549 * use-after-evict bugs in clients. 550 */ 551 dmu_buf_t **dbu_clear_on_evict_dbufp; 552#endif 553} dmu_buf_user_t; 554 555/* 556 * Initialize the given dmu_buf_user_t instance with the eviction function 557 * evict_func, to be called when the user is evicted. 558 * 559 * NOTE: This function should only be called once on a given dmu_buf_user_t. 560 * To allow enforcement of this, dbu must already be zeroed on entry. 561 */ 562#ifdef __lint 563/* Very ugly, but it beats issuing suppression directives in many Makefiles. */ 564extern 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#else /* __lint */ 568static inline void 569dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func, 570 dmu_buf_t **clear_on_evict_dbufp) 571{ 572 ASSERT(dbu->dbu_evict_func == NULL); 573 ASSERT(evict_func != NULL); 574 dbu->dbu_evict_func = evict_func; 575#ifdef ZFS_DEBUG 576 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp; 577#endif 578} 579#endif /* __lint */ 580 581/* 582 * Attach user data to a dbuf and mark it for normal (when the dbuf's 583 * data is cleared or its reference count goes to zero) eviction processing. 584 * 585 * Returns NULL on success, or the existing user if another user currently 586 * owns the buffer. 587 */ 588void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user); 589 590/* 591 * Attach user data to a dbuf and mark it for immediate (its dirty and 592 * reference counts are equal) eviction processing. 593 * 594 * Returns NULL on success, or the existing user if another user currently 595 * owns the buffer. 596 */ 597void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user); 598 599/* 600 * Replace the current user of a dbuf. 601 * 602 * If given the current user of a dbuf, replaces the dbuf's user with 603 * "new_user" and returns the user data pointer that was replaced. 604 * Otherwise returns the current, and unmodified, dbuf user pointer. 605 */ 606void *dmu_buf_replace_user(dmu_buf_t *db, 607 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user); 608 609/* 610 * Remove the specified user data for a DMU buffer. 611 * 612 * Returns the user that was removed on success, or the current user if 613 * another user currently owns the buffer. 614 */ 615void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user); 616 617/* 618 * Returns the user data (dmu_buf_user_t *) associated with this dbuf. 619 */ 620void *dmu_buf_get_user(dmu_buf_t *db); 621 622objset_t *dmu_buf_get_objset(dmu_buf_t *db); 623dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db); 624void dmu_buf_dnode_exit(dmu_buf_t *db); 625 626/* Block until any in-progress dmu buf user evictions complete. */ 627void dmu_buf_user_evict_wait(void); 628 629/* 630 * Returns the blkptr associated with this dbuf, or NULL if not set. 631 */ 632struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db); 633 634/* 635 * Indicate that you are going to modify the buffer's data (db_data). 636 * 637 * The transaction (tx) must be assigned to a txg (ie. you've called 638 * dmu_tx_assign()). The buffer's object must be held in the tx 639 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)). 640 */ 641void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx); 642 643/* 644 * Tells if the given dbuf is freeable. 645 */ 646boolean_t dmu_buf_freeable(dmu_buf_t *); 647 648/* 649 * You must create a transaction, then hold the objects which you will 650 * (or might) modify as part of this transaction. Then you must assign 651 * the transaction to a transaction group. Once the transaction has 652 * been assigned, you can modify buffers which belong to held objects as 653 * part of this transaction. You can't modify buffers before the 654 * transaction has been assigned; you can't modify buffers which don't 655 * belong to objects which this transaction holds; you can't hold 656 * objects once the transaction has been assigned. You may hold an 657 * object which you are going to free (with dmu_object_free()), but you 658 * don't have to. 659 * 660 * You can abort the transaction before it has been assigned. 661 * 662 * Note that you may hold buffers (with dmu_buf_hold) at any time, 663 * regardless of transaction state. 664 */ 665 666#define DMU_NEW_OBJECT (-1ULL) 667#define DMU_OBJECT_END (-1ULL) 668 669dmu_tx_t *dmu_tx_create(objset_t *os); 670void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len); 671void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, 672 uint64_t len); 673void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name); 674void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object); 675void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object); 676void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow); 677void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size); 678void dmu_tx_abort(dmu_tx_t *tx); 679int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how); 680void dmu_tx_wait(dmu_tx_t *tx); 681void dmu_tx_commit(dmu_tx_t *tx); 682void dmu_tx_mark_netfree(dmu_tx_t *tx); 683 684/* 685 * To register a commit callback, dmu_tx_callback_register() must be called. 686 * 687 * dcb_data is a pointer to caller private data that is passed on as a 688 * callback parameter. The caller is responsible for properly allocating and 689 * freeing it. 690 * 691 * When registering a callback, the transaction must be already created, but 692 * it cannot be committed or aborted. It can be assigned to a txg or not. 693 * 694 * The callback will be called after the transaction has been safely written 695 * to stable storage and will also be called if the dmu_tx is aborted. 696 * If there is any error which prevents the transaction from being committed to 697 * disk, the callback will be called with a value of error != 0. 698 */ 699typedef void dmu_tx_callback_func_t(void *dcb_data, int error); 700 701void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func, 702 void *dcb_data); 703 704/* 705 * Free up the data blocks for a defined range of a file. If size is 706 * -1, the range from offset to end-of-file is freed. 707 */ 708int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 709 uint64_t size, dmu_tx_t *tx); 710int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset, 711 uint64_t size); 712int dmu_free_long_object(objset_t *os, uint64_t object); 713 714/* 715 * Convenience functions. 716 * 717 * Canfail routines will return 0 on success, or an errno if there is a 718 * nonrecoverable I/O error. 719 */ 720#define DMU_READ_PREFETCH 0 /* prefetch */ 721#define DMU_READ_NO_PREFETCH 1 /* don't prefetch */ 722int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 723 void *buf, uint32_t flags); 724void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 725 const void *buf, dmu_tx_t *tx); 726void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 727 dmu_tx_t *tx); 728int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size); 729int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size); 730int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size, 731 dmu_tx_t *tx); 732int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size, 733 dmu_tx_t *tx); 734#ifdef _KERNEL 735#ifdef illumos 736int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 737 uint64_t size, struct page *pp, dmu_tx_t *tx); 738#else 739int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 740 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx); 741#endif 742#endif 743struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size); 744void dmu_return_arcbuf(struct arc_buf *buf); 745void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf, 746 dmu_tx_t *tx); 747int dmu_xuio_init(struct xuio *uio, int niov); 748void dmu_xuio_fini(struct xuio *uio); 749int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off, 750 size_t n); 751int dmu_xuio_cnt(struct xuio *uio); 752struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i); 753void dmu_xuio_clear(struct xuio *uio, int i); 754void xuio_stat_wbuf_copied(); 755void xuio_stat_wbuf_nocopy(); 756 757extern boolean_t zfs_prefetch_disable; 758extern int zfs_max_recordsize; 759 760/* 761 * Asynchronously try to read in the data. 762 */ 763void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, 764 uint64_t len, enum zio_priority pri); 765 766typedef struct dmu_object_info { 767 /* All sizes are in bytes unless otherwise indicated. */ 768 uint32_t doi_data_block_size; 769 uint32_t doi_metadata_block_size; 770 dmu_object_type_t doi_type; 771 dmu_object_type_t doi_bonus_type; 772 uint64_t doi_bonus_size; 773 uint8_t doi_indirection; /* 2 = dnode->indirect->data */ 774 uint8_t doi_checksum; 775 uint8_t doi_compress; 776 uint8_t doi_nblkptr; 777 uint8_t doi_pad[4]; 778 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */ 779 uint64_t doi_max_offset; 780 uint64_t doi_fill_count; /* number of non-empty blocks */ 781} dmu_object_info_t; 782 783typedef void arc_byteswap_func_t(void *buf, size_t size); 784 785typedef struct dmu_object_type_info { 786 dmu_object_byteswap_t ot_byteswap; 787 boolean_t ot_metadata; 788 char *ot_name; 789} dmu_object_type_info_t; 790 791typedef struct dmu_object_byteswap_info { 792 arc_byteswap_func_t *ob_func; 793 char *ob_name; 794} dmu_object_byteswap_info_t; 795 796extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES]; 797extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS]; 798 799/* 800 * Get information on a DMU object. 801 * 802 * Return 0 on success or ENOENT if object is not allocated. 803 * 804 * If doi is NULL, just indicates whether the object exists. 805 */ 806int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi); 807/* Like dmu_object_info, but faster if you have a held dnode in hand. */ 808void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi); 809/* Like dmu_object_info, but faster if you have a held dbuf in hand. */ 810void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi); 811/* 812 * Like dmu_object_info_from_db, but faster still when you only care about 813 * the size. This is specifically optimized for zfs_getattr(). 814 */ 815void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, 816 u_longlong_t *nblk512); 817 818typedef struct dmu_objset_stats { 819 uint64_t dds_num_clones; /* number of clones of this */ 820 uint64_t dds_creation_txg; 821 uint64_t dds_guid; 822 dmu_objset_type_t dds_type; 823 uint8_t dds_is_snapshot; 824 uint8_t dds_inconsistent; 825 char dds_origin[ZFS_MAX_DATASET_NAME_LEN]; 826} dmu_objset_stats_t; 827 828/* 829 * Get stats on a dataset. 830 */ 831void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat); 832 833/* 834 * Add entries to the nvlist for all the objset's properties. See 835 * zfs_prop_table[] and zfs(1m) for details on the properties. 836 */ 837void dmu_objset_stats(objset_t *os, struct nvlist *nv); 838 839/* 840 * Get the space usage statistics for statvfs(). 841 * 842 * refdbytes is the amount of space "referenced" by this objset. 843 * availbytes is the amount of space available to this objset, taking 844 * into account quotas & reservations, assuming that no other objsets 845 * use the space first. These values correspond to the 'referenced' and 846 * 'available' properties, described in the zfs(1m) manpage. 847 * 848 * usedobjs and availobjs are the number of objects currently allocated, 849 * and available. 850 */ 851void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 852 uint64_t *usedobjsp, uint64_t *availobjsp); 853 854/* 855 * The fsid_guid is a 56-bit ID that can change to avoid collisions. 856 * (Contrast with the ds_guid which is a 64-bit ID that will never 857 * change, so there is a small probability that it will collide.) 858 */ 859uint64_t dmu_objset_fsid_guid(objset_t *os); 860 861/* 862 * Get the [cm]time for an objset's snapshot dir 863 */ 864timestruc_t dmu_objset_snap_cmtime(objset_t *os); 865 866int dmu_objset_is_snapshot(objset_t *os); 867 868extern struct spa *dmu_objset_spa(objset_t *os); 869extern struct zilog *dmu_objset_zil(objset_t *os); 870extern struct dsl_pool *dmu_objset_pool(objset_t *os); 871extern struct dsl_dataset *dmu_objset_ds(objset_t *os); 872extern void dmu_objset_name(objset_t *os, char *buf); 873extern dmu_objset_type_t dmu_objset_type(objset_t *os); 874extern uint64_t dmu_objset_id(objset_t *os); 875extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os); 876extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os); 877extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 878 uint64_t *id, uint64_t *offp, boolean_t *case_conflict); 879extern int dmu_snapshot_realname(objset_t *os, char *name, char *real, 880 int maxlen, boolean_t *conflict); 881extern int dmu_dir_list_next(objset_t *os, int namelen, char *name, 882 uint64_t *idp, uint64_t *offp); 883 884typedef int objset_used_cb_t(dmu_object_type_t bonustype, 885 void *bonus, uint64_t *userp, uint64_t *groupp); 886extern void dmu_objset_register_type(dmu_objset_type_t ost, 887 objset_used_cb_t *cb); 888extern void dmu_objset_set_user(objset_t *os, void *user_ptr); 889extern void *dmu_objset_get_user(objset_t *os); 890 891/* 892 * Return the txg number for the given assigned transaction. 893 */ 894uint64_t dmu_tx_get_txg(dmu_tx_t *tx); 895 896/* 897 * Synchronous write. 898 * If a parent zio is provided this function initiates a write on the 899 * provided buffer as a child of the parent zio. 900 * In the absence of a parent zio, the write is completed synchronously. 901 * At write completion, blk is filled with the bp of the written block. 902 * Note that while the data covered by this function will be on stable 903 * storage when the write completes this new data does not become a 904 * permanent part of the file until the associated transaction commits. 905 */ 906 907/* 908 * {zfs,zvol,ztest}_get_done() args 909 */ 910typedef struct zgd { 911 struct zilog *zgd_zilog; 912 struct blkptr *zgd_bp; 913 dmu_buf_t *zgd_db; 914 struct rl *zgd_rl; 915 void *zgd_private; 916} zgd_t; 917 918typedef void dmu_sync_cb_t(zgd_t *arg, int error); 919int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd); 920 921/* 922 * Find the next hole or data block in file starting at *off 923 * Return found offset in *off. Return ESRCH for end of file. 924 */ 925int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, 926 uint64_t *off); 927 928/* 929 * Check if a DMU object has any dirty blocks. If so, sync out 930 * all pending transaction groups. Otherwise, this function 931 * does not alter DMU state. This could be improved to only sync 932 * out the necessary transaction groups for this particular 933 * object. 934 */ 935int dmu_object_wait_synced(objset_t *os, uint64_t object); 936 937/* 938 * Initial setup and final teardown. 939 */ 940extern void dmu_init(void); 941extern void dmu_fini(void); 942 943typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp, 944 uint64_t object, uint64_t offset, int len); 945void dmu_traverse_objset(objset_t *os, uint64_t txg_start, 946 dmu_traverse_cb_t cb, void *arg); 947int dmu_diff(const char *tosnap_name, const char *fromsnap_name, 948 struct file *fp, offset_t *offp); 949 950/* CRC64 table */ 951#define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */ 952extern uint64_t zfs_crc64_table[256]; 953 954extern int zfs_mdcomp_disable; 955 956#ifdef __cplusplus 957} 958#endif 959 960#endif /* _SYS_DMU_H */ 961