dmu.h revision 268647
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 */ 29 30/* Portions Copyright 2010 Robert Milkowski */ 31 32#ifndef _SYS_DMU_H 33#define _SYS_DMU_H 34 35/* 36 * This file describes the interface that the DMU provides for its 37 * consumers. 38 * 39 * The DMU also interacts with the SPA. That interface is described in 40 * dmu_spa.h. 41 */ 42 43#include <sys/types.h> 44#include <sys/param.h> 45#include <sys/cred.h> 46#include <sys/time.h> 47#include <sys/fs/zfs.h> 48 49#ifdef __cplusplus 50extern "C" { 51#endif 52 53struct uio; 54struct xuio; 55struct page; 56struct vnode; 57struct spa; 58struct zilog; 59struct zio; 60struct blkptr; 61struct zap_cursor; 62struct dsl_dataset; 63struct dsl_pool; 64struct dnode; 65struct drr_begin; 66struct drr_end; 67struct zbookmark; 68struct spa; 69struct nvlist; 70struct arc_buf; 71struct zio_prop; 72struct sa_handle; 73struct file; 74 75typedef struct objset objset_t; 76typedef struct dmu_tx dmu_tx_t; 77typedef struct dsl_dir dsl_dir_t; 78 79typedef enum dmu_object_byteswap { 80 DMU_BSWAP_UINT8, 81 DMU_BSWAP_UINT16, 82 DMU_BSWAP_UINT32, 83 DMU_BSWAP_UINT64, 84 DMU_BSWAP_ZAP, 85 DMU_BSWAP_DNODE, 86 DMU_BSWAP_OBJSET, 87 DMU_BSWAP_ZNODE, 88 DMU_BSWAP_OLDACL, 89 DMU_BSWAP_ACL, 90 /* 91 * Allocating a new byteswap type number makes the on-disk format 92 * incompatible with any other format that uses the same number. 93 * 94 * Data can usually be structured to work with one of the 95 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types. 96 */ 97 DMU_BSWAP_NUMFUNCS 98} dmu_object_byteswap_t; 99 100#define DMU_OT_NEWTYPE 0x80 101#define DMU_OT_METADATA 0x40 102#define DMU_OT_BYTESWAP_MASK 0x3f 103 104/* 105 * Defines a uint8_t object type. Object types specify if the data 106 * in the object is metadata (boolean) and how to byteswap the data 107 * (dmu_object_byteswap_t). 108 */ 109#define DMU_OT(byteswap, metadata) \ 110 (DMU_OT_NEWTYPE | \ 111 ((metadata) ? DMU_OT_METADATA : 0) | \ 112 ((byteswap) & DMU_OT_BYTESWAP_MASK)) 113 114#define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 115 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \ 116 (ot) < DMU_OT_NUMTYPES) 117 118#define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 119 ((ot) & DMU_OT_METADATA) : \ 120 dmu_ot[(ot)].ot_metadata) 121 122#define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \ 123 ((ot) & DMU_OT_BYTESWAP_MASK) : \ 124 dmu_ot[(ot)].ot_byteswap) 125 126typedef enum dmu_object_type { 127 DMU_OT_NONE, 128 /* general: */ 129 DMU_OT_OBJECT_DIRECTORY, /* ZAP */ 130 DMU_OT_OBJECT_ARRAY, /* UINT64 */ 131 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */ 132 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */ 133 DMU_OT_BPOBJ, /* UINT64 */ 134 DMU_OT_BPOBJ_HDR, /* UINT64 */ 135 /* spa: */ 136 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */ 137 DMU_OT_SPACE_MAP, /* UINT64 */ 138 /* zil: */ 139 DMU_OT_INTENT_LOG, /* UINT64 */ 140 /* dmu: */ 141 DMU_OT_DNODE, /* DNODE */ 142 DMU_OT_OBJSET, /* OBJSET */ 143 /* dsl: */ 144 DMU_OT_DSL_DIR, /* UINT64 */ 145 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */ 146 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */ 147 DMU_OT_DSL_PROPS, /* ZAP */ 148 DMU_OT_DSL_DATASET, /* UINT64 */ 149 /* zpl: */ 150 DMU_OT_ZNODE, /* ZNODE */ 151 DMU_OT_OLDACL, /* Old ACL */ 152 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */ 153 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */ 154 DMU_OT_MASTER_NODE, /* ZAP */ 155 DMU_OT_UNLINKED_SET, /* ZAP */ 156 /* zvol: */ 157 DMU_OT_ZVOL, /* UINT8 */ 158 DMU_OT_ZVOL_PROP, /* ZAP */ 159 /* other; for testing only! */ 160 DMU_OT_PLAIN_OTHER, /* UINT8 */ 161 DMU_OT_UINT64_OTHER, /* UINT64 */ 162 DMU_OT_ZAP_OTHER, /* ZAP */ 163 /* new object types: */ 164 DMU_OT_ERROR_LOG, /* ZAP */ 165 DMU_OT_SPA_HISTORY, /* UINT8 */ 166 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */ 167 DMU_OT_POOL_PROPS, /* ZAP */ 168 DMU_OT_DSL_PERMS, /* ZAP */ 169 DMU_OT_ACL, /* ACL */ 170 DMU_OT_SYSACL, /* SYSACL */ 171 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */ 172 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */ 173 DMU_OT_NEXT_CLONES, /* ZAP */ 174 DMU_OT_SCAN_QUEUE, /* ZAP */ 175 DMU_OT_USERGROUP_USED, /* ZAP */ 176 DMU_OT_USERGROUP_QUOTA, /* ZAP */ 177 DMU_OT_USERREFS, /* ZAP */ 178 DMU_OT_DDT_ZAP, /* ZAP */ 179 DMU_OT_DDT_STATS, /* ZAP */ 180 DMU_OT_SA, /* System attr */ 181 DMU_OT_SA_MASTER_NODE, /* ZAP */ 182 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */ 183 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */ 184 DMU_OT_SCAN_XLATE, /* ZAP */ 185 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */ 186 DMU_OT_DEADLIST, /* ZAP */ 187 DMU_OT_DEADLIST_HDR, /* UINT64 */ 188 DMU_OT_DSL_CLONES, /* ZAP */ 189 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */ 190 /* 191 * Do not allocate new object types here. Doing so makes the on-disk 192 * format incompatible with any other format that uses the same object 193 * type number. 194 * 195 * When creating an object which does not have one of the above types 196 * use the DMU_OTN_* type with the correct byteswap and metadata 197 * values. 198 * 199 * The DMU_OTN_* types do not have entries in the dmu_ot table, 200 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead 201 * of indexing into dmu_ot directly (this works for both DMU_OT_* types 202 * and DMU_OTN_* types). 203 */ 204 DMU_OT_NUMTYPES, 205 206 /* 207 * Names for valid types declared with DMU_OT(). 208 */ 209 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE), 210 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE), 211 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE), 212 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE), 213 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE), 214 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE), 215 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE), 216 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE), 217 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE), 218 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE), 219} dmu_object_type_t; 220 221typedef enum txg_how { 222 TXG_WAIT = 1, 223 TXG_NOWAIT, 224 TXG_WAITED, 225} txg_how_t; 226 227void byteswap_uint64_array(void *buf, size_t size); 228void byteswap_uint32_array(void *buf, size_t size); 229void byteswap_uint16_array(void *buf, size_t size); 230void byteswap_uint8_array(void *buf, size_t size); 231void zap_byteswap(void *buf, size_t size); 232void zfs_oldacl_byteswap(void *buf, size_t size); 233void zfs_acl_byteswap(void *buf, size_t size); 234void zfs_znode_byteswap(void *buf, size_t size); 235 236#define DS_FIND_SNAPSHOTS (1<<0) 237#define DS_FIND_CHILDREN (1<<1) 238 239/* 240 * The maximum number of bytes that can be accessed as part of one 241 * operation, including metadata. 242 */ 243#define DMU_MAX_ACCESS (10<<20) /* 10MB */ 244#define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */ 245 246#define DMU_USERUSED_OBJECT (-1ULL) 247#define DMU_GROUPUSED_OBJECT (-2ULL) 248#define DMU_DEADLIST_OBJECT (-3ULL) 249 250/* 251 * artificial blkids for bonus buffer and spill blocks 252 */ 253#define DMU_BONUS_BLKID (-1ULL) 254#define DMU_SPILL_BLKID (-2ULL) 255/* 256 * Public routines to create, destroy, open, and close objsets. 257 */ 258int dmu_objset_hold(const char *name, void *tag, objset_t **osp); 259int dmu_objset_own(const char *name, dmu_objset_type_t type, 260 boolean_t readonly, void *tag, objset_t **osp); 261void dmu_objset_rele(objset_t *os, void *tag); 262void dmu_objset_disown(objset_t *os, void *tag); 263int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp); 264 265void dmu_objset_evict_dbufs(objset_t *os); 266int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 267 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg); 268int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 269 struct nvlist *snaps); 270int dmu_objset_clone(const char *name, const char *origin); 271int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer, 272 struct nvlist *errlist); 273int dmu_objset_snapshot_one(const char *fsname, const char *snapname); 274int dmu_objset_snapshot_tmp(const char *, const char *, int); 275int dmu_objset_find(char *name, int func(const char *, void *), void *arg, 276 int flags); 277void dmu_objset_byteswap(void *buf, size_t size); 278int dsl_dataset_rename_snapshot(const char *fsname, 279 const char *oldsnapname, const char *newsnapname, boolean_t recursive); 280 281typedef struct dmu_buf { 282 uint64_t db_object; /* object that this buffer is part of */ 283 uint64_t db_offset; /* byte offset in this object */ 284 uint64_t db_size; /* size of buffer in bytes */ 285 void *db_data; /* data in buffer */ 286} dmu_buf_t; 287 288typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr); 289 290/* 291 * The names of zap entries in the DIRECTORY_OBJECT of the MOS. 292 */ 293#define DMU_POOL_DIRECTORY_OBJECT 1 294#define DMU_POOL_CONFIG "config" 295#define DMU_POOL_FEATURES_FOR_WRITE "features_for_write" 296#define DMU_POOL_FEATURES_FOR_READ "features_for_read" 297#define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions" 298#define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg" 299#define DMU_POOL_ROOT_DATASET "root_dataset" 300#define DMU_POOL_SYNC_BPOBJ "sync_bplist" 301#define DMU_POOL_ERRLOG_SCRUB "errlog_scrub" 302#define DMU_POOL_ERRLOG_LAST "errlog_last" 303#define DMU_POOL_SPARES "spares" 304#define DMU_POOL_DEFLATE "deflate" 305#define DMU_POOL_HISTORY "history" 306#define DMU_POOL_PROPS "pool_props" 307#define DMU_POOL_L2CACHE "l2cache" 308#define DMU_POOL_TMP_USERREFS "tmp_userrefs" 309#define DMU_POOL_DDT "DDT-%s-%s-%s" 310#define DMU_POOL_DDT_STATS "DDT-statistics" 311#define DMU_POOL_CREATION_VERSION "creation_version" 312#define DMU_POOL_SCAN "scan" 313#define DMU_POOL_FREE_BPOBJ "free_bpobj" 314#define DMU_POOL_BPTREE_OBJ "bptree_obj" 315#define DMU_POOL_EMPTY_BPOBJ "empty_bpobj" 316 317/* 318 * Allocate an object from this objset. The range of object numbers 319 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode. 320 * 321 * The transaction must be assigned to a txg. The newly allocated 322 * object will be "held" in the transaction (ie. you can modify the 323 * newly allocated object in this transaction). 324 * 325 * dmu_object_alloc() chooses an object and returns it in *objectp. 326 * 327 * dmu_object_claim() allocates a specific object number. If that 328 * number is already allocated, it fails and returns EEXIST. 329 * 330 * Return 0 on success, or ENOSPC or EEXIST as specified above. 331 */ 332uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot, 333 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 334int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot, 335 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 336int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot, 337 int blocksize, dmu_object_type_t bonustype, int bonuslen); 338 339/* 340 * Free an object from this objset. 341 * 342 * The object's data will be freed as well (ie. you don't need to call 343 * dmu_free(object, 0, -1, tx)). 344 * 345 * The object need not be held in the transaction. 346 * 347 * If there are any holds on this object's buffers (via dmu_buf_hold()), 348 * or tx holds on the object (via dmu_tx_hold_object()), you can not 349 * free it; it fails and returns EBUSY. 350 * 351 * If the object is not allocated, it fails and returns ENOENT. 352 * 353 * Return 0 on success, or EBUSY or ENOENT as specified above. 354 */ 355int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx); 356 357/* 358 * Find the next allocated or free object. 359 * 360 * The objectp parameter is in-out. It will be updated to be the next 361 * object which is allocated. Ignore objects which have not been 362 * modified since txg. 363 * 364 * XXX Can only be called on a objset with no dirty data. 365 * 366 * Returns 0 on success, or ENOENT if there are no more objects. 367 */ 368int dmu_object_next(objset_t *os, uint64_t *objectp, 369 boolean_t hole, uint64_t txg); 370 371/* 372 * Set the data blocksize for an object. 373 * 374 * The object cannot have any blocks allcated beyond the first. If 375 * the first block is allocated already, the new size must be greater 376 * than the current block size. If these conditions are not met, 377 * ENOTSUP will be returned. 378 * 379 * Returns 0 on success, or EBUSY if there are any holds on the object 380 * contents, or ENOTSUP as described above. 381 */ 382int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, 383 int ibs, dmu_tx_t *tx); 384 385/* 386 * Set the checksum property on a dnode. The new checksum algorithm will 387 * apply to all newly written blocks; existing blocks will not be affected. 388 */ 389void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 390 dmu_tx_t *tx); 391 392/* 393 * Set the compress property on a dnode. The new compression algorithm will 394 * apply to all newly written blocks; existing blocks will not be affected. 395 */ 396void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 397 dmu_tx_t *tx); 398 399/* 400 * Decide how to write a block: checksum, compression, number of copies, etc. 401 */ 402#define WP_NOFILL 0x1 403#define WP_DMU_SYNC 0x2 404#define WP_SPILL 0x4 405 406void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp, 407 struct zio_prop *zp); 408/* 409 * The bonus data is accessed more or less like a regular buffer. 410 * You must dmu_bonus_hold() to get the buffer, which will give you a 411 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus 412 * data. As with any normal buffer, you must call dmu_buf_read() to 413 * read db_data, dmu_buf_will_dirty() before modifying it, and the 414 * object must be held in an assigned transaction before calling 415 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus 416 * buffer as well. You must release your hold with dmu_buf_rele(). 417 * 418 * Returns ENOENT, EIO, or 0. 419 */ 420int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **); 421int dmu_bonus_max(void); 422int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *); 423int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *); 424dmu_object_type_t dmu_get_bonustype(dmu_buf_t *); 425int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *); 426 427/* 428 * Special spill buffer support used by "SA" framework 429 */ 430 431int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 432int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags, 433 void *tag, dmu_buf_t **dbp); 434int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 435 436/* 437 * Obtain the DMU buffer from the specified object which contains the 438 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so 439 * that it will remain in memory. You must release the hold with 440 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your 441 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU. 442 * 443 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill 444 * on the returned buffer before reading or writing the buffer's 445 * db_data. The comments for those routines describe what particular 446 * operations are valid after calling them. 447 * 448 * The object number must be a valid, allocated object number. 449 */ 450int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 451 void *tag, dmu_buf_t **, int flags); 452void dmu_buf_add_ref(dmu_buf_t *db, void* tag); 453void dmu_buf_rele(dmu_buf_t *db, void *tag); 454uint64_t dmu_buf_refcount(dmu_buf_t *db); 455 456/* 457 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a 458 * range of an object. A pointer to an array of dmu_buf_t*'s is 459 * returned (in *dbpp). 460 * 461 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and 462 * frees the array. The hold on the array of buffers MUST be released 463 * with dmu_buf_rele_array. You can NOT release the hold on each buffer 464 * individually with dmu_buf_rele. 465 */ 466int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, 467 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp); 468void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag); 469 470/* 471 * Returns NULL on success, or the existing user ptr if it's already 472 * been set. 473 * 474 * user_ptr is for use by the user and can be obtained via dmu_buf_get_user(). 475 * 476 * user_data_ptr_ptr should be NULL, or a pointer to a pointer which 477 * will be set to db->db_data when you are allowed to access it. Note 478 * that db->db_data (the pointer) can change when you do dmu_buf_read(), 479 * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill(). 480 * *user_data_ptr_ptr will be set to the new value when it changes. 481 * 482 * If non-NULL, pageout func will be called when this buffer is being 483 * excised from the cache, so that you can clean up the data structure 484 * pointed to by user_ptr. 485 * 486 * dmu_evict_user() will call the pageout func for all buffers in a 487 * objset with a given pageout func. 488 */ 489void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr, 490 dmu_buf_evict_func_t *pageout_func); 491/* 492 * set_user_ie is the same as set_user, but request immediate eviction 493 * when hold count goes to zero. 494 */ 495void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr, 496 void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func); 497void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr, 498 void *user_ptr, void *user_data_ptr_ptr, 499 dmu_buf_evict_func_t *pageout_func); 500void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func); 501 502/* 503 * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set. 504 */ 505void *dmu_buf_get_user(dmu_buf_t *db); 506 507/* 508 * Returns the blkptr associated with this dbuf, or NULL if not set. 509 */ 510struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db); 511 512/* 513 * Indicate that you are going to modify the buffer's data (db_data). 514 * 515 * The transaction (tx) must be assigned to a txg (ie. you've called 516 * dmu_tx_assign()). The buffer's object must be held in the tx 517 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)). 518 */ 519void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx); 520 521/* 522 * Tells if the given dbuf is freeable. 523 */ 524boolean_t dmu_buf_freeable(dmu_buf_t *); 525 526/* 527 * You must create a transaction, then hold the objects which you will 528 * (or might) modify as part of this transaction. Then you must assign 529 * the transaction to a transaction group. Once the transaction has 530 * been assigned, you can modify buffers which belong to held objects as 531 * part of this transaction. You can't modify buffers before the 532 * transaction has been assigned; you can't modify buffers which don't 533 * belong to objects which this transaction holds; you can't hold 534 * objects once the transaction has been assigned. You may hold an 535 * object which you are going to free (with dmu_object_free()), but you 536 * don't have to. 537 * 538 * You can abort the transaction before it has been assigned. 539 * 540 * Note that you may hold buffers (with dmu_buf_hold) at any time, 541 * regardless of transaction state. 542 */ 543 544#define DMU_NEW_OBJECT (-1ULL) 545#define DMU_OBJECT_END (-1ULL) 546 547dmu_tx_t *dmu_tx_create(objset_t *os); 548void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len); 549void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, 550 uint64_t len); 551void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name); 552void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object); 553void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object); 554void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow); 555void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size); 556void dmu_tx_abort(dmu_tx_t *tx); 557int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how); 558void dmu_tx_wait(dmu_tx_t *tx); 559void dmu_tx_commit(dmu_tx_t *tx); 560 561/* 562 * To register a commit callback, dmu_tx_callback_register() must be called. 563 * 564 * dcb_data is a pointer to caller private data that is passed on as a 565 * callback parameter. The caller is responsible for properly allocating and 566 * freeing it. 567 * 568 * When registering a callback, the transaction must be already created, but 569 * it cannot be committed or aborted. It can be assigned to a txg or not. 570 * 571 * The callback will be called after the transaction has been safely written 572 * to stable storage and will also be called if the dmu_tx is aborted. 573 * If there is any error which prevents the transaction from being committed to 574 * disk, the callback will be called with a value of error != 0. 575 */ 576typedef void dmu_tx_callback_func_t(void *dcb_data, int error); 577 578void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func, 579 void *dcb_data); 580 581/* 582 * Free up the data blocks for a defined range of a file. If size is 583 * -1, the range from offset to end-of-file is freed. 584 */ 585int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 586 uint64_t size, dmu_tx_t *tx); 587int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset, 588 uint64_t size); 589int dmu_free_long_object(objset_t *os, uint64_t object); 590 591/* 592 * Convenience functions. 593 * 594 * Canfail routines will return 0 on success, or an errno if there is a 595 * nonrecoverable I/O error. 596 */ 597#define DMU_READ_PREFETCH 0 /* prefetch */ 598#define DMU_READ_NO_PREFETCH 1 /* don't prefetch */ 599int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 600 void *buf, uint32_t flags); 601void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 602 const void *buf, dmu_tx_t *tx); 603void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 604 dmu_tx_t *tx); 605int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size); 606int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size, 607 dmu_tx_t *tx); 608int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size, 609 dmu_tx_t *tx); 610#ifdef _KERNEL 611#ifdef sun 612int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 613 uint64_t size, struct page *pp, dmu_tx_t *tx); 614#else 615int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 616 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx); 617#endif 618#endif 619struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size); 620void dmu_return_arcbuf(struct arc_buf *buf); 621void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf, 622 dmu_tx_t *tx); 623int dmu_xuio_init(struct xuio *uio, int niov); 624void dmu_xuio_fini(struct xuio *uio); 625int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off, 626 size_t n); 627int dmu_xuio_cnt(struct xuio *uio); 628struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i); 629void dmu_xuio_clear(struct xuio *uio, int i); 630void xuio_stat_wbuf_copied(); 631void xuio_stat_wbuf_nocopy(); 632 633extern int zfs_prefetch_disable; 634 635/* 636 * Asynchronously try to read in the data. 637 */ 638void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, 639 uint64_t len); 640 641typedef struct dmu_object_info { 642 /* All sizes are in bytes unless otherwise indicated. */ 643 uint32_t doi_data_block_size; 644 uint32_t doi_metadata_block_size; 645 dmu_object_type_t doi_type; 646 dmu_object_type_t doi_bonus_type; 647 uint64_t doi_bonus_size; 648 uint8_t doi_indirection; /* 2 = dnode->indirect->data */ 649 uint8_t doi_checksum; 650 uint8_t doi_compress; 651 uint8_t doi_pad[5]; 652 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */ 653 uint64_t doi_max_offset; 654 uint64_t doi_fill_count; /* number of non-empty blocks */ 655} dmu_object_info_t; 656 657typedef void arc_byteswap_func_t(void *buf, size_t size); 658 659typedef struct dmu_object_type_info { 660 dmu_object_byteswap_t ot_byteswap; 661 boolean_t ot_metadata; 662 char *ot_name; 663} dmu_object_type_info_t; 664 665typedef struct dmu_object_byteswap_info { 666 arc_byteswap_func_t *ob_func; 667 char *ob_name; 668} dmu_object_byteswap_info_t; 669 670extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES]; 671extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS]; 672 673/* 674 * Get information on a DMU object. 675 * 676 * Return 0 on success or ENOENT if object is not allocated. 677 * 678 * If doi is NULL, just indicates whether the object exists. 679 */ 680int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi); 681/* Like dmu_object_info, but faster if you have a held dnode in hand. */ 682void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi); 683/* Like dmu_object_info, but faster if you have a held dbuf in hand. */ 684void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi); 685/* 686 * Like dmu_object_info_from_db, but faster still when you only care about 687 * the size. This is specifically optimized for zfs_getattr(). 688 */ 689void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, 690 u_longlong_t *nblk512); 691 692typedef struct dmu_objset_stats { 693 uint64_t dds_num_clones; /* number of clones of this */ 694 uint64_t dds_creation_txg; 695 uint64_t dds_guid; 696 dmu_objset_type_t dds_type; 697 uint8_t dds_is_snapshot; 698 uint8_t dds_inconsistent; 699 char dds_origin[MAXNAMELEN]; 700} dmu_objset_stats_t; 701 702/* 703 * Get stats on a dataset. 704 */ 705void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat); 706 707/* 708 * Add entries to the nvlist for all the objset's properties. See 709 * zfs_prop_table[] and zfs(1m) for details on the properties. 710 */ 711void dmu_objset_stats(objset_t *os, struct nvlist *nv); 712 713/* 714 * Get the space usage statistics for statvfs(). 715 * 716 * refdbytes is the amount of space "referenced" by this objset. 717 * availbytes is the amount of space available to this objset, taking 718 * into account quotas & reservations, assuming that no other objsets 719 * use the space first. These values correspond to the 'referenced' and 720 * 'available' properties, described in the zfs(1m) manpage. 721 * 722 * usedobjs and availobjs are the number of objects currently allocated, 723 * and available. 724 */ 725void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 726 uint64_t *usedobjsp, uint64_t *availobjsp); 727 728/* 729 * The fsid_guid is a 56-bit ID that can change to avoid collisions. 730 * (Contrast with the ds_guid which is a 64-bit ID that will never 731 * change, so there is a small probability that it will collide.) 732 */ 733uint64_t dmu_objset_fsid_guid(objset_t *os); 734 735/* 736 * Get the [cm]time for an objset's snapshot dir 737 */ 738timestruc_t dmu_objset_snap_cmtime(objset_t *os); 739 740int dmu_objset_is_snapshot(objset_t *os); 741 742extern struct spa *dmu_objset_spa(objset_t *os); 743extern struct zilog *dmu_objset_zil(objset_t *os); 744extern struct dsl_pool *dmu_objset_pool(objset_t *os); 745extern struct dsl_dataset *dmu_objset_ds(objset_t *os); 746extern void dmu_objset_name(objset_t *os, char *buf); 747extern dmu_objset_type_t dmu_objset_type(objset_t *os); 748extern uint64_t dmu_objset_id(objset_t *os); 749extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os); 750extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os); 751extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 752 uint64_t *id, uint64_t *offp, boolean_t *case_conflict); 753extern int dmu_snapshot_realname(objset_t *os, char *name, char *real, 754 int maxlen, boolean_t *conflict); 755extern int dmu_dir_list_next(objset_t *os, int namelen, char *name, 756 uint64_t *idp, uint64_t *offp); 757 758typedef int objset_used_cb_t(dmu_object_type_t bonustype, 759 void *bonus, uint64_t *userp, uint64_t *groupp); 760extern void dmu_objset_register_type(dmu_objset_type_t ost, 761 objset_used_cb_t *cb); 762extern void dmu_objset_set_user(objset_t *os, void *user_ptr); 763extern void *dmu_objset_get_user(objset_t *os); 764 765/* 766 * Return the txg number for the given assigned transaction. 767 */ 768uint64_t dmu_tx_get_txg(dmu_tx_t *tx); 769 770/* 771 * Synchronous write. 772 * If a parent zio is provided this function initiates a write on the 773 * provided buffer as a child of the parent zio. 774 * In the absence of a parent zio, the write is completed synchronously. 775 * At write completion, blk is filled with the bp of the written block. 776 * Note that while the data covered by this function will be on stable 777 * storage when the write completes this new data does not become a 778 * permanent part of the file until the associated transaction commits. 779 */ 780 781/* 782 * {zfs,zvol,ztest}_get_done() args 783 */ 784typedef struct zgd { 785 struct zilog *zgd_zilog; 786 struct blkptr *zgd_bp; 787 dmu_buf_t *zgd_db; 788 struct rl *zgd_rl; 789 void *zgd_private; 790} zgd_t; 791 792typedef void dmu_sync_cb_t(zgd_t *arg, int error); 793int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd); 794 795/* 796 * Find the next hole or data block in file starting at *off 797 * Return found offset in *off. Return ESRCH for end of file. 798 */ 799int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, 800 uint64_t *off); 801 802/* 803 * Initial setup and final teardown. 804 */ 805extern void dmu_init(void); 806extern void dmu_fini(void); 807 808typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp, 809 uint64_t object, uint64_t offset, int len); 810void dmu_traverse_objset(objset_t *os, uint64_t txg_start, 811 dmu_traverse_cb_t cb, void *arg); 812int dmu_diff(const char *tosnap_name, const char *fromsnap_name, 813 struct file *fp, offset_t *offp); 814 815/* CRC64 table */ 816#define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */ 817extern uint64_t zfs_crc64_table[256]; 818 819extern int zfs_mdcomp_disable; 820 821#ifdef __cplusplus 822} 823#endif 824 825#endif /* _SYS_DMU_H */ 826