dmu.h revision 260786
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) 2013 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_ROOT_DATASET "root_dataset" 299#define DMU_POOL_SYNC_BPOBJ "sync_bplist" 300#define DMU_POOL_ERRLOG_SCRUB "errlog_scrub" 301#define DMU_POOL_ERRLOG_LAST "errlog_last" 302#define DMU_POOL_SPARES "spares" 303#define DMU_POOL_DEFLATE "deflate" 304#define DMU_POOL_HISTORY "history" 305#define DMU_POOL_PROPS "pool_props" 306#define DMU_POOL_L2CACHE "l2cache" 307#define DMU_POOL_TMP_USERREFS "tmp_userrefs" 308#define DMU_POOL_DDT "DDT-%s-%s-%s" 309#define DMU_POOL_DDT_STATS "DDT-statistics" 310#define DMU_POOL_CREATION_VERSION "creation_version" 311#define DMU_POOL_SCAN "scan" 312#define DMU_POOL_FREE_BPOBJ "free_bpobj" 313#define DMU_POOL_BPTREE_OBJ "bptree_obj" 314#define DMU_POOL_EMPTY_BPOBJ "empty_bpobj" 315 316/* 317 * Allocate an object from this objset. The range of object numbers 318 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode. 319 * 320 * The transaction must be assigned to a txg. The newly allocated 321 * object will be "held" in the transaction (ie. you can modify the 322 * newly allocated object in this transaction). 323 * 324 * dmu_object_alloc() chooses an object and returns it in *objectp. 325 * 326 * dmu_object_claim() allocates a specific object number. If that 327 * number is already allocated, it fails and returns EEXIST. 328 * 329 * Return 0 on success, or ENOSPC or EEXIST as specified above. 330 */ 331uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot, 332 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 333int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot, 334 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 335int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot, 336 int blocksize, dmu_object_type_t bonustype, int bonuslen); 337 338/* 339 * Free an object from this objset. 340 * 341 * The object's data will be freed as well (ie. you don't need to call 342 * dmu_free(object, 0, -1, tx)). 343 * 344 * The object need not be held in the transaction. 345 * 346 * If there are any holds on this object's buffers (via dmu_buf_hold()), 347 * or tx holds on the object (via dmu_tx_hold_object()), you can not 348 * free it; it fails and returns EBUSY. 349 * 350 * If the object is not allocated, it fails and returns ENOENT. 351 * 352 * Return 0 on success, or EBUSY or ENOENT as specified above. 353 */ 354int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx); 355 356/* 357 * Find the next allocated or free object. 358 * 359 * The objectp parameter is in-out. It will be updated to be the next 360 * object which is allocated. Ignore objects which have not been 361 * modified since txg. 362 * 363 * XXX Can only be called on a objset with no dirty data. 364 * 365 * Returns 0 on success, or ENOENT if there are no more objects. 366 */ 367int dmu_object_next(objset_t *os, uint64_t *objectp, 368 boolean_t hole, uint64_t txg); 369 370/* 371 * Set the data blocksize for an object. 372 * 373 * The object cannot have any blocks allcated beyond the first. If 374 * the first block is allocated already, the new size must be greater 375 * than the current block size. If these conditions are not met, 376 * ENOTSUP will be returned. 377 * 378 * Returns 0 on success, or EBUSY if there are any holds on the object 379 * contents, or ENOTSUP as described above. 380 */ 381int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, 382 int ibs, dmu_tx_t *tx); 383 384/* 385 * Set the checksum property on a dnode. The new checksum algorithm will 386 * apply to all newly written blocks; existing blocks will not be affected. 387 */ 388void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 389 dmu_tx_t *tx); 390 391/* 392 * Set the compress property on a dnode. The new compression algorithm will 393 * apply to all newly written blocks; existing blocks will not be affected. 394 */ 395void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 396 dmu_tx_t *tx); 397 398/* 399 * Decide how to write a block: checksum, compression, number of copies, etc. 400 */ 401#define WP_NOFILL 0x1 402#define WP_DMU_SYNC 0x2 403#define WP_SPILL 0x4 404 405void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp, 406 struct zio_prop *zp); 407/* 408 * The bonus data is accessed more or less like a regular buffer. 409 * You must dmu_bonus_hold() to get the buffer, which will give you a 410 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus 411 * data. As with any normal buffer, you must call dmu_buf_read() to 412 * read db_data, dmu_buf_will_dirty() before modifying it, and the 413 * object must be held in an assigned transaction before calling 414 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus 415 * buffer as well. You must release your hold with dmu_buf_rele(). 416 * 417 * Returns ENOENT, EIO, or 0. 418 */ 419int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **); 420int dmu_bonus_max(void); 421int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *); 422int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *); 423dmu_object_type_t dmu_get_bonustype(dmu_buf_t *); 424int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *); 425 426/* 427 * Special spill buffer support used by "SA" framework 428 */ 429 430int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 431int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags, 432 void *tag, dmu_buf_t **dbp); 433int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 434 435/* 436 * Obtain the DMU buffer from the specified object which contains the 437 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so 438 * that it will remain in memory. You must release the hold with 439 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your 440 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU. 441 * 442 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill 443 * on the returned buffer before reading or writing the buffer's 444 * db_data. The comments for those routines describe what particular 445 * operations are valid after calling them. 446 * 447 * The object number must be a valid, allocated object number. 448 */ 449int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 450 void *tag, dmu_buf_t **, int flags); 451void dmu_buf_add_ref(dmu_buf_t *db, void* tag); 452void dmu_buf_rele(dmu_buf_t *db, void *tag); 453uint64_t dmu_buf_refcount(dmu_buf_t *db); 454 455/* 456 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a 457 * range of an object. A pointer to an array of dmu_buf_t*'s is 458 * returned (in *dbpp). 459 * 460 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and 461 * frees the array. The hold on the array of buffers MUST be released 462 * with dmu_buf_rele_array. You can NOT release the hold on each buffer 463 * individually with dmu_buf_rele. 464 */ 465int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, 466 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp); 467void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag); 468 469/* 470 * Returns NULL on success, or the existing user ptr if it's already 471 * been set. 472 * 473 * user_ptr is for use by the user and can be obtained via dmu_buf_get_user(). 474 * 475 * user_data_ptr_ptr should be NULL, or a pointer to a pointer which 476 * will be set to db->db_data when you are allowed to access it. Note 477 * that db->db_data (the pointer) can change when you do dmu_buf_read(), 478 * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill(). 479 * *user_data_ptr_ptr will be set to the new value when it changes. 480 * 481 * If non-NULL, pageout func will be called when this buffer is being 482 * excised from the cache, so that you can clean up the data structure 483 * pointed to by user_ptr. 484 * 485 * dmu_evict_user() will call the pageout func for all buffers in a 486 * objset with a given pageout func. 487 */ 488void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr, 489 dmu_buf_evict_func_t *pageout_func); 490/* 491 * set_user_ie is the same as set_user, but request immediate eviction 492 * when hold count goes to zero. 493 */ 494void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr, 495 void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func); 496void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr, 497 void *user_ptr, void *user_data_ptr_ptr, 498 dmu_buf_evict_func_t *pageout_func); 499void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func); 500 501/* 502 * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set. 503 */ 504void *dmu_buf_get_user(dmu_buf_t *db); 505 506/* 507 * Returns the blkptr associated with this dbuf, or NULL if not set. 508 */ 509struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db); 510 511/* 512 * Indicate that you are going to modify the buffer's data (db_data). 513 * 514 * The transaction (tx) must be assigned to a txg (ie. you've called 515 * dmu_tx_assign()). The buffer's object must be held in the tx 516 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)). 517 */ 518void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx); 519 520/* 521 * Tells if the given dbuf is freeable. 522 */ 523boolean_t dmu_buf_freeable(dmu_buf_t *); 524 525/* 526 * You must create a transaction, then hold the objects which you will 527 * (or might) modify as part of this transaction. Then you must assign 528 * the transaction to a transaction group. Once the transaction has 529 * been assigned, you can modify buffers which belong to held objects as 530 * part of this transaction. You can't modify buffers before the 531 * transaction has been assigned; you can't modify buffers which don't 532 * belong to objects which this transaction holds; you can't hold 533 * objects once the transaction has been assigned. You may hold an 534 * object which you are going to free (with dmu_object_free()), but you 535 * don't have to. 536 * 537 * You can abort the transaction before it has been assigned. 538 * 539 * Note that you may hold buffers (with dmu_buf_hold) at any time, 540 * regardless of transaction state. 541 */ 542 543#define DMU_NEW_OBJECT (-1ULL) 544#define DMU_OBJECT_END (-1ULL) 545 546dmu_tx_t *dmu_tx_create(objset_t *os); 547void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len); 548void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, 549 uint64_t len); 550void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name); 551void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object); 552void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object); 553void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow); 554void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size); 555void dmu_tx_abort(dmu_tx_t *tx); 556int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how); 557void dmu_tx_wait(dmu_tx_t *tx); 558void dmu_tx_commit(dmu_tx_t *tx); 559 560/* 561 * To register a commit callback, dmu_tx_callback_register() must be called. 562 * 563 * dcb_data is a pointer to caller private data that is passed on as a 564 * callback parameter. The caller is responsible for properly allocating and 565 * freeing it. 566 * 567 * When registering a callback, the transaction must be already created, but 568 * it cannot be committed or aborted. It can be assigned to a txg or not. 569 * 570 * The callback will be called after the transaction has been safely written 571 * to stable storage and will also be called if the dmu_tx is aborted. 572 * If there is any error which prevents the transaction from being committed to 573 * disk, the callback will be called with a value of error != 0. 574 */ 575typedef void dmu_tx_callback_func_t(void *dcb_data, int error); 576 577void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func, 578 void *dcb_data); 579 580/* 581 * Free up the data blocks for a defined range of a file. If size is 582 * -1, the range from offset to end-of-file is freed. 583 */ 584int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 585 uint64_t size, dmu_tx_t *tx); 586int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset, 587 uint64_t size); 588int dmu_free_long_object(objset_t *os, uint64_t object); 589 590/* 591 * Convenience functions. 592 * 593 * Canfail routines will return 0 on success, or an errno if there is a 594 * nonrecoverable I/O error. 595 */ 596#define DMU_READ_PREFETCH 0 /* prefetch */ 597#define DMU_READ_NO_PREFETCH 1 /* don't prefetch */ 598int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 599 void *buf, uint32_t flags); 600void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 601 const void *buf, dmu_tx_t *tx); 602void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 603 dmu_tx_t *tx); 604int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size); 605int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size, 606 dmu_tx_t *tx); 607int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size, 608 dmu_tx_t *tx); 609#ifdef _KERNEL 610#ifdef sun 611int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 612 uint64_t size, struct page *pp, dmu_tx_t *tx); 613#else 614int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 615 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx); 616#endif 617#endif 618struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size); 619void dmu_return_arcbuf(struct arc_buf *buf); 620void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf, 621 dmu_tx_t *tx); 622int dmu_xuio_init(struct xuio *uio, int niov); 623void dmu_xuio_fini(struct xuio *uio); 624int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off, 625 size_t n); 626int dmu_xuio_cnt(struct xuio *uio); 627struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i); 628void dmu_xuio_clear(struct xuio *uio, int i); 629void xuio_stat_wbuf_copied(); 630void xuio_stat_wbuf_nocopy(); 631 632extern int zfs_prefetch_disable; 633 634/* 635 * Asynchronously try to read in the data. 636 */ 637void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, 638 uint64_t len); 639 640typedef struct dmu_object_info { 641 /* All sizes are in bytes unless otherwise indicated. */ 642 uint32_t doi_data_block_size; 643 uint32_t doi_metadata_block_size; 644 dmu_object_type_t doi_type; 645 dmu_object_type_t doi_bonus_type; 646 uint64_t doi_bonus_size; 647 uint8_t doi_indirection; /* 2 = dnode->indirect->data */ 648 uint8_t doi_checksum; 649 uint8_t doi_compress; 650 uint8_t doi_pad[5]; 651 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */ 652 uint64_t doi_max_offset; 653 uint64_t doi_fill_count; /* number of non-empty blocks */ 654} dmu_object_info_t; 655 656typedef void arc_byteswap_func_t(void *buf, size_t size); 657 658typedef struct dmu_object_type_info { 659 dmu_object_byteswap_t ot_byteswap; 660 boolean_t ot_metadata; 661 char *ot_name; 662} dmu_object_type_info_t; 663 664typedef struct dmu_object_byteswap_info { 665 arc_byteswap_func_t *ob_func; 666 char *ob_name; 667} dmu_object_byteswap_info_t; 668 669extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES]; 670extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS]; 671 672/* 673 * Get information on a DMU object. 674 * 675 * Return 0 on success or ENOENT if object is not allocated. 676 * 677 * If doi is NULL, just indicates whether the object exists. 678 */ 679int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi); 680/* Like dmu_object_info, but faster if you have a held dnode in hand. */ 681void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi); 682/* Like dmu_object_info, but faster if you have a held dbuf in hand. */ 683void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi); 684/* 685 * Like dmu_object_info_from_db, but faster still when you only care about 686 * the size. This is specifically optimized for zfs_getattr(). 687 */ 688void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, 689 u_longlong_t *nblk512); 690 691typedef struct dmu_objset_stats { 692 uint64_t dds_num_clones; /* number of clones of this */ 693 uint64_t dds_creation_txg; 694 uint64_t dds_guid; 695 dmu_objset_type_t dds_type; 696 uint8_t dds_is_snapshot; 697 uint8_t dds_inconsistent; 698 char dds_origin[MAXNAMELEN]; 699} dmu_objset_stats_t; 700 701/* 702 * Get stats on a dataset. 703 */ 704void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat); 705 706/* 707 * Add entries to the nvlist for all the objset's properties. See 708 * zfs_prop_table[] and zfs(1m) for details on the properties. 709 */ 710void dmu_objset_stats(objset_t *os, struct nvlist *nv); 711 712/* 713 * Get the space usage statistics for statvfs(). 714 * 715 * refdbytes is the amount of space "referenced" by this objset. 716 * availbytes is the amount of space available to this objset, taking 717 * into account quotas & reservations, assuming that no other objsets 718 * use the space first. These values correspond to the 'referenced' and 719 * 'available' properties, described in the zfs(1m) manpage. 720 * 721 * usedobjs and availobjs are the number of objects currently allocated, 722 * and available. 723 */ 724void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 725 uint64_t *usedobjsp, uint64_t *availobjsp); 726 727/* 728 * The fsid_guid is a 56-bit ID that can change to avoid collisions. 729 * (Contrast with the ds_guid which is a 64-bit ID that will never 730 * change, so there is a small probability that it will collide.) 731 */ 732uint64_t dmu_objset_fsid_guid(objset_t *os); 733 734/* 735 * Get the [cm]time for an objset's snapshot dir 736 */ 737timestruc_t dmu_objset_snap_cmtime(objset_t *os); 738 739int dmu_objset_is_snapshot(objset_t *os); 740 741extern struct spa *dmu_objset_spa(objset_t *os); 742extern struct zilog *dmu_objset_zil(objset_t *os); 743extern struct dsl_pool *dmu_objset_pool(objset_t *os); 744extern struct dsl_dataset *dmu_objset_ds(objset_t *os); 745extern void dmu_objset_name(objset_t *os, char *buf); 746extern dmu_objset_type_t dmu_objset_type(objset_t *os); 747extern uint64_t dmu_objset_id(objset_t *os); 748extern uint64_t dmu_objset_syncprop(objset_t *os); 749extern uint64_t dmu_objset_logbias(objset_t *os); 750extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 751 uint64_t *id, uint64_t *offp, boolean_t *case_conflict); 752extern int dmu_snapshot_realname(objset_t *os, char *name, char *real, 753 int maxlen, boolean_t *conflict); 754extern int dmu_dir_list_next(objset_t *os, int namelen, char *name, 755 uint64_t *idp, uint64_t *offp); 756 757typedef int objset_used_cb_t(dmu_object_type_t bonustype, 758 void *bonus, uint64_t *userp, uint64_t *groupp); 759extern void dmu_objset_register_type(dmu_objset_type_t ost, 760 objset_used_cb_t *cb); 761extern void dmu_objset_set_user(objset_t *os, void *user_ptr); 762extern void *dmu_objset_get_user(objset_t *os); 763 764/* 765 * Return the txg number for the given assigned transaction. 766 */ 767uint64_t dmu_tx_get_txg(dmu_tx_t *tx); 768 769/* 770 * Synchronous write. 771 * If a parent zio is provided this function initiates a write on the 772 * provided buffer as a child of the parent zio. 773 * In the absence of a parent zio, the write is completed synchronously. 774 * At write completion, blk is filled with the bp of the written block. 775 * Note that while the data covered by this function will be on stable 776 * storage when the write completes this new data does not become a 777 * permanent part of the file until the associated transaction commits. 778 */ 779 780/* 781 * {zfs,zvol,ztest}_get_done() args 782 */ 783typedef struct zgd { 784 struct zilog *zgd_zilog; 785 struct blkptr *zgd_bp; 786 dmu_buf_t *zgd_db; 787 struct rl *zgd_rl; 788 void *zgd_private; 789} zgd_t; 790 791typedef void dmu_sync_cb_t(zgd_t *arg, int error); 792int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd); 793 794/* 795 * Find the next hole or data block in file starting at *off 796 * Return found offset in *off. Return ESRCH for end of file. 797 */ 798int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, 799 uint64_t *off); 800 801/* 802 * Initial setup and final teardown. 803 */ 804extern void dmu_init(void); 805extern void dmu_fini(void); 806 807typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp, 808 uint64_t object, uint64_t offset, int len); 809void dmu_traverse_objset(objset_t *os, uint64_t txg_start, 810 dmu_traverse_cb_t cb, void *arg); 811int dmu_diff(const char *tosnap_name, const char *fromsnap_name, 812 struct file *fp, offset_t *offp); 813 814/* CRC64 table */ 815#define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */ 816extern uint64_t zfs_crc64_table[256]; 817 818extern int zfs_mdcomp_disable; 819 820#ifdef __cplusplus 821} 822#endif 823 824#endif /* _SYS_DMU_H */ 825