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; 232 233/* 234 * These flags are intended to be used to specify the "txg_how" 235 * parameter when calling the dmu_tx_assign() function. See the comment 236 * above dmu_tx_assign() for more details on the meaning of these flags. 237 */ 238#define TXG_NOWAIT (0ULL) 239#define TXG_WAIT (1ULL<<0) 240#define TXG_NOTHROTTLE (1ULL<<1) 241 242void byteswap_uint64_array(void *buf, size_t size); 243void byteswap_uint32_array(void *buf, size_t size); 244void byteswap_uint16_array(void *buf, size_t size); 245void byteswap_uint8_array(void *buf, size_t size); 246void zap_byteswap(void *buf, size_t size); 247void zfs_oldacl_byteswap(void *buf, size_t size); 248void zfs_acl_byteswap(void *buf, size_t size); 249void zfs_znode_byteswap(void *buf, size_t size); 250 251#define DS_FIND_SNAPSHOTS (1<<0) 252#define DS_FIND_CHILDREN (1<<1) 253#define DS_FIND_SERIALIZE (1<<2) 254 255/* 256 * The maximum number of bytes that can be accessed as part of one 257 * operation, including metadata. 258 */ 259#define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */ 260#define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */ 261 262#define DMU_USERUSED_OBJECT (-1ULL) 263#define DMU_GROUPUSED_OBJECT (-2ULL) 264 265/* 266 * artificial blkids for bonus buffer and spill blocks 267 */ 268#define DMU_BONUS_BLKID (-1ULL) 269#define DMU_SPILL_BLKID (-2ULL) 270/* 271 * Public routines to create, destroy, open, and close objsets. 272 */ 273int dmu_objset_hold(const char *name, void *tag, objset_t **osp); 274int dmu_objset_own(const char *name, dmu_objset_type_t type, 275 boolean_t readonly, void *tag, objset_t **osp); 276void dmu_objset_rele(objset_t *os, void *tag); 277void dmu_objset_disown(objset_t *os, void *tag); 278int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp); 279 280void dmu_objset_evict_dbufs(objset_t *os); 281int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 282 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg); 283int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 284 struct nvlist *snaps); 285int dmu_objset_clone(const char *name, const char *origin); 286int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer, 287 struct nvlist *errlist); 288int dmu_objset_snapshot_one(const char *fsname, const char *snapname); 289int dmu_objset_snapshot_tmp(const char *, const char *, int); 290int dmu_objset_find(char *name, int func(const char *, void *), void *arg, 291 int flags); 292void dmu_objset_byteswap(void *buf, size_t size); 293int dsl_dataset_rename_snapshot(const char *fsname, 294 const char *oldsnapname, const char *newsnapname, boolean_t recursive); 295 296typedef struct dmu_buf { 297 uint64_t db_object; /* object that this buffer is part of */ 298 uint64_t db_offset; /* byte offset in this object */ 299 uint64_t db_size; /* size of buffer in bytes */ 300 void *db_data; /* data in buffer */ 301} dmu_buf_t; 302 303/* 304 * The names of zap entries in the DIRECTORY_OBJECT of the MOS. 305 */ 306#define DMU_POOL_DIRECTORY_OBJECT 1 307#define DMU_POOL_CONFIG "config" 308#define DMU_POOL_FEATURES_FOR_WRITE "features_for_write" 309#define DMU_POOL_FEATURES_FOR_READ "features_for_read" 310#define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions" 311#define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg" 312#define DMU_POOL_ROOT_DATASET "root_dataset" 313#define DMU_POOL_SYNC_BPOBJ "sync_bplist" 314#define DMU_POOL_ERRLOG_SCRUB "errlog_scrub" 315#define DMU_POOL_ERRLOG_LAST "errlog_last" 316#define DMU_POOL_SPARES "spares" 317#define DMU_POOL_DEFLATE "deflate" 318#define DMU_POOL_HISTORY "history" 319#define DMU_POOL_PROPS "pool_props" 320#define DMU_POOL_L2CACHE "l2cache" 321#define DMU_POOL_TMP_USERREFS "tmp_userrefs" 322#define DMU_POOL_DDT "DDT-%s-%s-%s" 323#define DMU_POOL_DDT_STATS "DDT-statistics" 324#define DMU_POOL_CREATION_VERSION "creation_version" 325#define DMU_POOL_SCAN "scan" 326#define DMU_POOL_FREE_BPOBJ "free_bpobj" 327#define DMU_POOL_BPTREE_OBJ "bptree_obj" 328#define DMU_POOL_EMPTY_BPOBJ "empty_bpobj" 329#define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt" 330 331/* 332 * Allocate an object from this objset. The range of object numbers 333 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode. 334 * 335 * The transaction must be assigned to a txg. The newly allocated 336 * object will be "held" in the transaction (ie. you can modify the 337 * newly allocated object in this transaction). 338 * 339 * dmu_object_alloc() chooses an object and returns it in *objectp. 340 * 341 * dmu_object_claim() allocates a specific object number. If that 342 * number is already allocated, it fails and returns EEXIST. 343 * 344 * Return 0 on success, or ENOSPC or EEXIST as specified above. 345 */ 346uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot, 347 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 348int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot, 349 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx); 350int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot, 351 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp); 352 353/* 354 * Free an object from this objset. 355 * 356 * The object's data will be freed as well (ie. you don't need to call 357 * dmu_free(object, 0, -1, tx)). 358 * 359 * The object need not be held in the transaction. 360 * 361 * If there are any holds on this object's buffers (via dmu_buf_hold()), 362 * or tx holds on the object (via dmu_tx_hold_object()), you can not 363 * free it; it fails and returns EBUSY. 364 * 365 * If the object is not allocated, it fails and returns ENOENT. 366 * 367 * Return 0 on success, or EBUSY or ENOENT as specified above. 368 */ 369int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx); 370 371/* 372 * Find the next allocated or free object. 373 * 374 * The objectp parameter is in-out. It will be updated to be the next 375 * object which is allocated. Ignore objects which have not been 376 * modified since txg. 377 * 378 * XXX Can only be called on a objset with no dirty data. 379 * 380 * Returns 0 on success, or ENOENT if there are no more objects. 381 */ 382int dmu_object_next(objset_t *os, uint64_t *objectp, 383 boolean_t hole, uint64_t txg); 384 385/* 386 * Set the data blocksize for an object. 387 * 388 * The object cannot have any blocks allcated beyond the first. If 389 * the first block is allocated already, the new size must be greater 390 * than the current block size. If these conditions are not met, 391 * ENOTSUP will be returned. 392 * 393 * Returns 0 on success, or EBUSY if there are any holds on the object 394 * contents, or ENOTSUP as described above. 395 */ 396int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, 397 int ibs, dmu_tx_t *tx); 398 399/* 400 * Set the checksum property on a dnode. The new checksum algorithm will 401 * apply to all newly written blocks; existing blocks will not be affected. 402 */ 403void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 404 dmu_tx_t *tx); 405 406/* 407 * Set the compress property on a dnode. The new compression algorithm will 408 * apply to all newly written blocks; existing blocks will not be affected. 409 */ 410void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 411 dmu_tx_t *tx); 412 413void 414dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, 415 void *data, uint8_t etype, uint8_t comp, int uncompressed_size, 416 int compressed_size, int byteorder, dmu_tx_t *tx); 417 418/* 419 * Decide how to write a block: checksum, compression, number of copies, etc. 420 */ 421#define WP_NOFILL 0x1 422#define WP_DMU_SYNC 0x2 423#define WP_SPILL 0x4 424 425void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, 426 struct zio_prop *zp); 427/* 428 * The bonus data is accessed more or less like a regular buffer. 429 * You must dmu_bonus_hold() to get the buffer, which will give you a 430 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus 431 * data. As with any normal buffer, you must call dmu_buf_read() to 432 * read db_data, dmu_buf_will_dirty() before modifying it, and the 433 * object must be held in an assigned transaction before calling 434 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus 435 * buffer as well. You must release your hold with dmu_buf_rele(). 436 * 437 * Returns ENOENT, EIO, or 0. 438 */ 439int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **); 440int dmu_bonus_max(void); 441int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *); 442int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *); 443dmu_object_type_t dmu_get_bonustype(dmu_buf_t *); 444int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *); 445 446/* 447 * Special spill buffer support used by "SA" framework 448 */ 449 450int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 451int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, 452 void *tag, dmu_buf_t **dbp); 453int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp); 454 455/* 456 * Obtain the DMU buffer from the specified object which contains the 457 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so 458 * that it will remain in memory. You must release the hold with 459 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your 460 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU. 461 * 462 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill 463 * on the returned buffer before reading or writing the buffer's 464 * db_data. The comments for those routines describe what particular 465 * operations are valid after calling them. 466 * 467 * The object number must be a valid, allocated object number. 468 */ 469int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 470 void *tag, dmu_buf_t **, int flags); 471int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, 472 void *tag, dmu_buf_t **dbp, int flags); 473 474/* 475 * Add a reference to a dmu buffer that has already been held via 476 * dmu_buf_hold() in the current context. 477 */ 478void dmu_buf_add_ref(dmu_buf_t *db, void* tag); 479 480/* 481 * Attempt to add a reference to a dmu buffer that is in an unknown state, 482 * using a pointer that may have been invalidated by eviction processing. 483 * The request will succeed if the passed in dbuf still represents the 484 * same os/object/blkid, is ineligible for eviction, and has at least 485 * one hold by a user other than the syncer. 486 */ 487boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object, 488 uint64_t blkid, void *tag); 489 490void dmu_buf_rele(dmu_buf_t *db, void *tag); 491uint64_t dmu_buf_refcount(dmu_buf_t *db); 492 493/* 494 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a 495 * range of an object. A pointer to an array of dmu_buf_t*'s is 496 * returned (in *dbpp). 497 * 498 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and 499 * frees the array. The hold on the array of buffers MUST be released 500 * with dmu_buf_rele_array. You can NOT release the hold on each buffer 501 * individually with dmu_buf_rele. 502 */ 503int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, 504 uint64_t length, boolean_t read, void *tag, 505 int *numbufsp, dmu_buf_t ***dbpp); 506void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag); 507 508typedef void dmu_buf_evict_func_t(void *user_ptr); 509 510/* 511 * A DMU buffer user object may be associated with a dbuf for the 512 * duration of its lifetime. This allows the user of a dbuf (client) 513 * to attach private data to a dbuf (e.g. in-core only data such as a 514 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified 515 * when that dbuf has been evicted. Clients typically respond to the 516 * eviction notification by freeing their private data, thus ensuring 517 * the same lifetime for both dbuf and private data. 518 * 519 * The mapping from a dmu_buf_user_t to any client private data is the 520 * client's responsibility. All current consumers of the API with private 521 * data embed a dmu_buf_user_t as the first member of the structure for 522 * their private data. This allows conversions between the two types 523 * with a simple cast. Since the DMU buf user API never needs access 524 * to the private data, other strategies can be employed if necessary 525 * or convenient for the client (e.g. using container_of() to do the 526 * conversion for private data that cannot have the dmu_buf_user_t as 527 * its first member). 528 * 529 * Eviction callbacks are executed without the dbuf mutex held or any 530 * other type of mechanism to guarantee that the dbuf is still available. 531 * For this reason, users must assume the dbuf has already been freed 532 * and not reference the dbuf from the callback context. 533 * 534 * Users requesting "immediate eviction" are notified as soon as the dbuf 535 * is only referenced by dirty records (dirties == holds). Otherwise the 536 * notification occurs after eviction processing for the dbuf begins. 537 */ 538typedef struct dmu_buf_user { 539 /* 540 * Asynchronous user eviction callback state. 541 */ 542 taskq_ent_t dbu_tqent; 543 544 /* This instance's eviction function pointer. */ 545 dmu_buf_evict_func_t *dbu_evict_func; 546#ifdef ZFS_DEBUG 547 /* 548 * Pointer to user's dbuf pointer. NULL for clients that do 549 * not associate a dbuf with their user data. 550 * 551 * The dbuf pointer is cleared upon eviction so as to catch 552 * use-after-evict bugs in clients. 553 */ 554 dmu_buf_t **dbu_clear_on_evict_dbufp; 555#endif 556} dmu_buf_user_t; 557 558/* 559 * Initialize the given dmu_buf_user_t instance with the eviction function 560 * evict_func, to be called when the user is evicted. 561 * 562 * NOTE: This function should only be called once on a given dmu_buf_user_t. 563 * To allow enforcement of this, dbu must already be zeroed on entry. 564 */ 565#ifdef __lint 566/* Very ugly, but it beats issuing suppression directives in many Makefiles. */ 567extern void 568dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func, 569 dmu_buf_t **clear_on_evict_dbufp); 570#else /* __lint */ 571static inline void 572dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func, 573 dmu_buf_t **clear_on_evict_dbufp) 574{ 575 ASSERT(dbu->dbu_evict_func == NULL); 576 ASSERT(evict_func != NULL); 577 dbu->dbu_evict_func = evict_func; 578#ifdef ZFS_DEBUG 579 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp; 580#endif 581} 582#endif /* __lint */ 583 584/* 585 * Attach user data to a dbuf and mark it for normal (when the dbuf's 586 * data is cleared or its reference count goes to zero) eviction processing. 587 * 588 * Returns NULL on success, or the existing user if another user currently 589 * owns the buffer. 590 */ 591void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user); 592 593/* 594 * Attach user data to a dbuf and mark it for immediate (its dirty and 595 * reference counts are equal) eviction processing. 596 * 597 * Returns NULL on success, or the existing user if another user currently 598 * owns the buffer. 599 */ 600void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user); 601 602/* 603 * Replace the current user of a dbuf. 604 * 605 * If given the current user of a dbuf, replaces the dbuf's user with 606 * "new_user" and returns the user data pointer that was replaced. 607 * Otherwise returns the current, and unmodified, dbuf user pointer. 608 */ 609void *dmu_buf_replace_user(dmu_buf_t *db, 610 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user); 611 612/* 613 * Remove the specified user data for a DMU buffer. 614 * 615 * Returns the user that was removed on success, or the current user if 616 * another user currently owns the buffer. 617 */ 618void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user); 619 620/* 621 * Returns the user data (dmu_buf_user_t *) associated with this dbuf. 622 */ 623void *dmu_buf_get_user(dmu_buf_t *db); 624 625objset_t *dmu_buf_get_objset(dmu_buf_t *db); 626dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db); 627void dmu_buf_dnode_exit(dmu_buf_t *db); 628 629/* Block until any in-progress dmu buf user evictions complete. */ 630void dmu_buf_user_evict_wait(void); 631 632/* 633 * Returns the blkptr associated with this dbuf, or NULL if not set. 634 */ 635struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db); 636 637/* 638 * Indicate that you are going to modify the buffer's data (db_data). 639 * 640 * The transaction (tx) must be assigned to a txg (ie. you've called 641 * dmu_tx_assign()). The buffer's object must be held in the tx 642 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)). 643 */ 644void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx); 645 646/* 647 * Tells if the given dbuf is freeable. 648 */ 649boolean_t dmu_buf_freeable(dmu_buf_t *); 650 651/* 652 * You must create a transaction, then hold the objects which you will 653 * (or might) modify as part of this transaction. Then you must assign 654 * the transaction to a transaction group. Once the transaction has 655 * been assigned, you can modify buffers which belong to held objects as 656 * part of this transaction. You can't modify buffers before the 657 * transaction has been assigned; you can't modify buffers which don't 658 * belong to objects which this transaction holds; you can't hold 659 * objects once the transaction has been assigned. You may hold an 660 * object which you are going to free (with dmu_object_free()), but you 661 * don't have to. 662 * 663 * You can abort the transaction before it has been assigned. 664 * 665 * Note that you may hold buffers (with dmu_buf_hold) at any time, 666 * regardless of transaction state. 667 */ 668 669#define DMU_NEW_OBJECT (-1ULL) 670#define DMU_OBJECT_END (-1ULL) 671 672dmu_tx_t *dmu_tx_create(objset_t *os); 673void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len); 674void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, 675 uint64_t len); 676void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name); 677void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object); 678void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object); 679void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow); 680void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size); 681void dmu_tx_abort(dmu_tx_t *tx); 682int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how); 683void dmu_tx_wait(dmu_tx_t *tx); 684void dmu_tx_commit(dmu_tx_t *tx); 685void dmu_tx_mark_netfree(dmu_tx_t *tx); 686 687/* 688 * To register a commit callback, dmu_tx_callback_register() must be called. 689 * 690 * dcb_data is a pointer to caller private data that is passed on as a 691 * callback parameter. The caller is responsible for properly allocating and 692 * freeing it. 693 * 694 * When registering a callback, the transaction must be already created, but 695 * it cannot be committed or aborted. It can be assigned to a txg or not. 696 * 697 * The callback will be called after the transaction has been safely written 698 * to stable storage and will also be called if the dmu_tx is aborted. 699 * If there is any error which prevents the transaction from being committed to 700 * disk, the callback will be called with a value of error != 0. 701 */ 702typedef void dmu_tx_callback_func_t(void *dcb_data, int error); 703 704void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func, 705 void *dcb_data); 706 707/* 708 * Free up the data blocks for a defined range of a file. If size is 709 * -1, the range from offset to end-of-file is freed. 710 */ 711int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 712 uint64_t size, dmu_tx_t *tx); 713int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset, 714 uint64_t size); 715int dmu_free_long_object(objset_t *os, uint64_t object); 716 717/* 718 * Convenience functions. 719 * 720 * Canfail routines will return 0 on success, or an errno if there is a 721 * nonrecoverable I/O error. 722 */ 723#define DMU_READ_PREFETCH 0 /* prefetch */ 724#define DMU_READ_NO_PREFETCH 1 /* don't prefetch */ 725int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 726 void *buf, uint32_t flags); 727void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 728 const void *buf, dmu_tx_t *tx); 729void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 730 dmu_tx_t *tx); 731int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size); 732int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size); 733int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size, 734 dmu_tx_t *tx); 735int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size, 736 dmu_tx_t *tx); 737#ifdef _KERNEL 738#ifdef illumos 739int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 740 uint64_t size, struct page *pp, dmu_tx_t *tx); 741#else 742int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, 743 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx); 744#endif 745#endif 746struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size); 747void dmu_return_arcbuf(struct arc_buf *buf); 748void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf, 749 dmu_tx_t *tx); 750int dmu_xuio_init(struct xuio *uio, int niov); 751void dmu_xuio_fini(struct xuio *uio); 752int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off, 753 size_t n); 754int dmu_xuio_cnt(struct xuio *uio); 755struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i); 756void dmu_xuio_clear(struct xuio *uio, int i); 757void xuio_stat_wbuf_copied(); 758void xuio_stat_wbuf_nocopy(); 759 760extern boolean_t zfs_prefetch_disable; 761extern int zfs_max_recordsize; 762 763/* 764 * Asynchronously try to read in the data. 765 */ 766void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, 767 uint64_t len, enum zio_priority pri); 768 769typedef struct dmu_object_info { 770 /* All sizes are in bytes unless otherwise indicated. */ 771 uint32_t doi_data_block_size; 772 uint32_t doi_metadata_block_size; 773 dmu_object_type_t doi_type; 774 dmu_object_type_t doi_bonus_type; 775 uint64_t doi_bonus_size; 776 uint8_t doi_indirection; /* 2 = dnode->indirect->data */ 777 uint8_t doi_checksum; 778 uint8_t doi_compress; 779 uint8_t doi_nblkptr; 780 uint8_t doi_pad[4]; 781 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */ 782 uint64_t doi_max_offset; 783 uint64_t doi_fill_count; /* number of non-empty blocks */ 784} dmu_object_info_t; 785 786typedef void arc_byteswap_func_t(void *buf, size_t size); 787 788typedef struct dmu_object_type_info { 789 dmu_object_byteswap_t ot_byteswap; 790 boolean_t ot_metadata; 791 char *ot_name; 792} dmu_object_type_info_t; 793 794typedef struct dmu_object_byteswap_info { 795 arc_byteswap_func_t *ob_func; 796 char *ob_name; 797} dmu_object_byteswap_info_t; 798 799extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES]; 800extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS]; 801 802/* 803 * Get information on a DMU object. 804 * 805 * Return 0 on success or ENOENT if object is not allocated. 806 * 807 * If doi is NULL, just indicates whether the object exists. 808 */ 809int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi); 810/* Like dmu_object_info, but faster if you have a held dnode in hand. */ 811void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi); 812/* Like dmu_object_info, but faster if you have a held dbuf in hand. */ 813void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi); 814/* 815 * Like dmu_object_info_from_db, but faster still when you only care about 816 * the size. This is specifically optimized for zfs_getattr(). 817 */ 818void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, 819 u_longlong_t *nblk512); 820 821typedef struct dmu_objset_stats { 822 uint64_t dds_num_clones; /* number of clones of this */ 823 uint64_t dds_creation_txg; 824 uint64_t dds_guid; 825 dmu_objset_type_t dds_type; 826 uint8_t dds_is_snapshot; 827 uint8_t dds_inconsistent; 828 char dds_origin[ZFS_MAX_DATASET_NAME_LEN]; 829} dmu_objset_stats_t; 830 831/* 832 * Get stats on a dataset. 833 */ 834void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat); 835 836/* 837 * Add entries to the nvlist for all the objset's properties. See 838 * zfs_prop_table[] and zfs(1m) for details on the properties. 839 */ 840void dmu_objset_stats(objset_t *os, struct nvlist *nv); 841 842/* 843 * Get the space usage statistics for statvfs(). 844 * 845 * refdbytes is the amount of space "referenced" by this objset. 846 * availbytes is the amount of space available to this objset, taking 847 * into account quotas & reservations, assuming that no other objsets 848 * use the space first. These values correspond to the 'referenced' and 849 * 'available' properties, described in the zfs(1m) manpage. 850 * 851 * usedobjs and availobjs are the number of objects currently allocated, 852 * and available. 853 */ 854void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 855 uint64_t *usedobjsp, uint64_t *availobjsp); 856 857/* 858 * The fsid_guid is a 56-bit ID that can change to avoid collisions. 859 * (Contrast with the ds_guid which is a 64-bit ID that will never 860 * change, so there is a small probability that it will collide.) 861 */ 862uint64_t dmu_objset_fsid_guid(objset_t *os); 863 864/* 865 * Get the [cm]time for an objset's snapshot dir 866 */ 867timestruc_t dmu_objset_snap_cmtime(objset_t *os); 868 869int dmu_objset_is_snapshot(objset_t *os); 870 871extern struct spa *dmu_objset_spa(objset_t *os); 872extern struct zilog *dmu_objset_zil(objset_t *os); 873extern struct dsl_pool *dmu_objset_pool(objset_t *os); 874extern struct dsl_dataset *dmu_objset_ds(objset_t *os); 875extern void dmu_objset_name(objset_t *os, char *buf); 876extern dmu_objset_type_t dmu_objset_type(objset_t *os); 877extern uint64_t dmu_objset_id(objset_t *os); 878extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os); 879extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os); 880extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 881 uint64_t *id, uint64_t *offp, boolean_t *case_conflict); 882extern int dmu_snapshot_realname(objset_t *os, char *name, char *real, 883 int maxlen, boolean_t *conflict); 884extern int dmu_dir_list_next(objset_t *os, int namelen, char *name, 885 uint64_t *idp, uint64_t *offp); 886 887typedef int objset_used_cb_t(dmu_object_type_t bonustype, 888 void *bonus, uint64_t *userp, uint64_t *groupp); 889extern void dmu_objset_register_type(dmu_objset_type_t ost, 890 objset_used_cb_t *cb); 891extern void dmu_objset_set_user(objset_t *os, void *user_ptr); 892extern void *dmu_objset_get_user(objset_t *os); 893 894/* 895 * Return the txg number for the given assigned transaction. 896 */ 897uint64_t dmu_tx_get_txg(dmu_tx_t *tx); 898 899/* 900 * Synchronous write. 901 * If a parent zio is provided this function initiates a write on the 902 * provided buffer as a child of the parent zio. 903 * In the absence of a parent zio, the write is completed synchronously. 904 * At write completion, blk is filled with the bp of the written block. 905 * Note that while the data covered by this function will be on stable 906 * storage when the write completes this new data does not become a 907 * permanent part of the file until the associated transaction commits. 908 */ 909 910/* 911 * {zfs,zvol,ztest}_get_done() args 912 */ 913typedef struct zgd { 914 struct zilog *zgd_zilog; 915 struct blkptr *zgd_bp; 916 dmu_buf_t *zgd_db; 917 struct rl *zgd_rl; 918 void *zgd_private; 919} zgd_t; 920 921typedef void dmu_sync_cb_t(zgd_t *arg, int error); 922int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd); 923 924/* 925 * Find the next hole or data block in file starting at *off 926 * Return found offset in *off. Return ESRCH for end of file. 927 */ 928int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, 929 uint64_t *off); 930 931/* 932 * Check if a DMU object has any dirty blocks. If so, sync out 933 * all pending transaction groups. Otherwise, this function 934 * does not alter DMU state. This could be improved to only sync 935 * out the necessary transaction groups for this particular 936 * object. 937 */ 938int dmu_object_wait_synced(objset_t *os, uint64_t object); 939 940/* 941 * Initial setup and final teardown. 942 */ 943extern void dmu_init(void); 944extern void dmu_fini(void); 945 946typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp, 947 uint64_t object, uint64_t offset, int len); 948void dmu_traverse_objset(objset_t *os, uint64_t txg_start, 949 dmu_traverse_cb_t cb, void *arg); 950int dmu_diff(const char *tosnap_name, const char *fromsnap_name, 951 struct file *fp, offset_t *offp); 952 953/* CRC64 table */ 954#define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */ 955extern uint64_t zfs_crc64_table[256]; 956 957extern int zfs_mdcomp_disable; 958 959#ifdef __cplusplus 960} 961#endif 962 963#endif /* _SYS_DMU_H */ 964