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) 2012 by Delphix. All rights reserved.
  25  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  26  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  27  */
  28 
  29 /* Portions Copyright 2010 Robert Milkowski */
  30 
  31 #ifndef _SYS_DMU_H
  32 #define _SYS_DMU_H
  33 
  34 /*
  35  * This file describes the interface that the DMU provides for its
  36  * consumers.
  37  *
  38  * The DMU also interacts with the SPA.  That interface is described in
  39  * dmu_spa.h.
  40  */
  41 
  42 #include <sys/inttypes.h>
  43 #include <sys/types.h>
  44 #include <sys/param.h>
  45 #include <sys/cred.h>
  46 #include <sys/time.h>
  47 
  48 #ifdef  __cplusplus
  49 extern "C" {
  50 #endif
  51 
  52 struct uio;
  53 struct xuio;
  54 struct page;
  55 struct vnode;
  56 struct spa;
  57 struct zilog;
  58 struct zio;
  59 struct blkptr;
  60 struct zap_cursor;
  61 struct dsl_dataset;
  62 struct dsl_pool;
  63 struct dnode;
  64 struct drr_begin;
  65 struct drr_end;
  66 struct zbookmark;
  67 struct spa;
  68 struct nvlist;
  69 struct arc_buf;
  70 struct zio_prop;
  71 struct sa_handle;
  72 
  73 typedef struct objset objset_t;
  74 typedef struct dmu_tx dmu_tx_t;
  75 typedef struct dsl_dir dsl_dir_t;
  76 
  77 typedef enum dmu_object_byteswap {
  78         DMU_BSWAP_UINT8,
  79         DMU_BSWAP_UINT16,
  80         DMU_BSWAP_UINT32,
  81         DMU_BSWAP_UINT64,
  82         DMU_BSWAP_ZAP,
  83         DMU_BSWAP_DNODE,
  84         DMU_BSWAP_OBJSET,
  85         DMU_BSWAP_ZNODE,
  86         DMU_BSWAP_OLDACL,
  87         DMU_BSWAP_ACL,
  88         /*
  89          * Allocating a new byteswap type number makes the on-disk format
  90          * incompatible with any other format that uses the same number.
  91          *
  92          * Data can usually be structured to work with one of the
  93          * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
  94          */
  95         DMU_BSWAP_NUMFUNCS
  96 } dmu_object_byteswap_t;
  97 
  98 #define DMU_OT_NEWTYPE 0x80
  99 #define DMU_OT_METADATA 0x40
 100 #define DMU_OT_BYTESWAP_MASK 0x3f
 101 
 102 /*
 103  * Defines a uint8_t object type. Object types specify if the data
 104  * in the object is metadata (boolean) and how to byteswap the data
 105  * (dmu_object_byteswap_t).
 106  */
 107 #define DMU_OT(byteswap, metadata) \
 108         (DMU_OT_NEWTYPE | \
 109         ((metadata) ? DMU_OT_METADATA : 0) | \
 110         ((byteswap) & DMU_OT_BYTESWAP_MASK))
 111 
 112 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
 113         ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
 114         (ot) < DMU_OT_NUMTYPES)
 115 
 116 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
 117         ((ot) & DMU_OT_METADATA) : \
 118         dmu_ot[(ot)].ot_metadata)
 119 
 120 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
 121         ((ot) & DMU_OT_BYTESWAP_MASK) : \
 122         dmu_ot[(ot)].ot_byteswap)
 123 
 124 typedef enum dmu_object_type {
 125         DMU_OT_NONE,
 126         /* general: */
 127         DMU_OT_OBJECT_DIRECTORY,        /* ZAP */
 128         DMU_OT_OBJECT_ARRAY,            /* UINT64 */
 129         DMU_OT_PACKED_NVLIST,           /* UINT8 (XDR by nvlist_pack/unpack) */
 130         DMU_OT_PACKED_NVLIST_SIZE,      /* UINT64 */
 131         DMU_OT_BPOBJ,                   /* UINT64 */
 132         DMU_OT_BPOBJ_HDR,               /* UINT64 */
 133         /* spa: */
 134         DMU_OT_SPACE_MAP_HEADER,        /* UINT64 */
 135         DMU_OT_SPACE_MAP,               /* UINT64 */
 136         /* zil: */
 137         DMU_OT_INTENT_LOG,              /* UINT64 */
 138         /* dmu: */
 139         DMU_OT_DNODE,                   /* DNODE */
 140         DMU_OT_OBJSET,                  /* OBJSET */
 141         /* dsl: */
 142         DMU_OT_DSL_DIR,                 /* UINT64 */
 143         DMU_OT_DSL_DIR_CHILD_MAP,       /* ZAP */
 144         DMU_OT_DSL_DS_SNAP_MAP,         /* ZAP */
 145         DMU_OT_DSL_PROPS,               /* ZAP */
 146         DMU_OT_DSL_DATASET,             /* UINT64 */
 147         /* zpl: */
 148         DMU_OT_ZNODE,                   /* ZNODE */
 149         DMU_OT_OLDACL,                  /* Old ACL */
 150         DMU_OT_PLAIN_FILE_CONTENTS,     /* UINT8 */
 151         DMU_OT_DIRECTORY_CONTENTS,      /* ZAP */
 152         DMU_OT_MASTER_NODE,             /* ZAP */
 153         DMU_OT_UNLINKED_SET,            /* ZAP */
 154         /* zvol: */
 155         DMU_OT_ZVOL,                    /* UINT8 */
 156         DMU_OT_ZVOL_PROP,               /* ZAP */
 157         /* other; for testing only! */
 158         DMU_OT_PLAIN_OTHER,             /* UINT8 */
 159         DMU_OT_UINT64_OTHER,            /* UINT64 */
 160         DMU_OT_ZAP_OTHER,               /* ZAP */
 161         /* new object types: */
 162         DMU_OT_ERROR_LOG,               /* ZAP */
 163         DMU_OT_SPA_HISTORY,             /* UINT8 */
 164         DMU_OT_SPA_HISTORY_OFFSETS,     /* spa_his_phys_t */
 165         DMU_OT_POOL_PROPS,              /* ZAP */
 166         DMU_OT_DSL_PERMS,               /* ZAP */
 167         DMU_OT_ACL,                     /* ACL */
 168         DMU_OT_SYSACL,                  /* SYSACL */
 169         DMU_OT_FUID,                    /* FUID table (Packed NVLIST UINT8) */
 170         DMU_OT_FUID_SIZE,               /* FUID table size UINT64 */
 171         DMU_OT_NEXT_CLONES,             /* ZAP */
 172         DMU_OT_SCAN_QUEUE,              /* ZAP */
 173         DMU_OT_USERGROUP_USED,          /* ZAP */
 174         DMU_OT_USERGROUP_QUOTA,         /* ZAP */
 175         DMU_OT_USERREFS,                /* ZAP */
 176         DMU_OT_DDT_ZAP,                 /* ZAP */
 177         DMU_OT_DDT_STATS,               /* ZAP */
 178         DMU_OT_SA,                      /* System attr */
 179         DMU_OT_SA_MASTER_NODE,          /* ZAP */
 180         DMU_OT_SA_ATTR_REGISTRATION,    /* ZAP */
 181         DMU_OT_SA_ATTR_LAYOUTS,         /* ZAP */
 182         DMU_OT_SCAN_XLATE,              /* ZAP */
 183         DMU_OT_DEDUP,                   /* fake dedup BP from ddt_bp_create() */
 184         DMU_OT_DEADLIST,                /* ZAP */
 185         DMU_OT_DEADLIST_HDR,            /* UINT64 */
 186         DMU_OT_DSL_CLONES,              /* ZAP */
 187         DMU_OT_BPOBJ_SUBOBJ,            /* UINT64 */
 188         /*
 189          * Do not allocate new object types here. Doing so makes the on-disk
 190          * format incompatible with any other format that uses the same object
 191          * type number.
 192          *
 193          * When creating an object which does not have one of the above types
 194          * use the DMU_OTN_* type with the correct byteswap and metadata
 195          * values.
 196          *
 197          * The DMU_OTN_* types do not have entries in the dmu_ot table,
 198          * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
 199          * of indexing into dmu_ot directly (this works for both DMU_OT_* types
 200          * and DMU_OTN_* types).
 201          */
 202         DMU_OT_NUMTYPES,
 203 
 204         /*
 205          * Names for valid types declared with DMU_OT().
 206          */
 207         DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
 208         DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
 209         DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
 210         DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
 211         DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
 212         DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
 213         DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
 214         DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
 215         DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
 216         DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
 217 } dmu_object_type_t;
 218 
 219 typedef enum dmu_objset_type {
 220         DMU_OST_NONE,
 221         DMU_OST_META,
 222         DMU_OST_ZFS,
 223         DMU_OST_ZVOL,
 224         DMU_OST_OTHER,                  /* For testing only! */
 225         DMU_OST_ANY,                    /* Be careful! */
 226         DMU_OST_NUMTYPES
 227 } dmu_objset_type_t;
 228 
 229 void byteswap_uint64_array(void *buf, size_t size);
 230 void byteswap_uint32_array(void *buf, size_t size);
 231 void byteswap_uint16_array(void *buf, size_t size);
 232 void byteswap_uint8_array(void *buf, size_t size);
 233 void zap_byteswap(void *buf, size_t size);
 234 void zfs_oldacl_byteswap(void *buf, size_t size);
 235 void zfs_acl_byteswap(void *buf, size_t size);
 236 void zfs_znode_byteswap(void *buf, size_t size);
 237 
 238 #define DS_FIND_SNAPSHOTS       (1<<0)
 239 #define DS_FIND_CHILDREN        (1<<1)
 240 
 241 /*
 242  * The maximum number of bytes that can be accessed as part of one
 243  * operation, including metadata.
 244  */
 245 #define DMU_MAX_ACCESS (10<<20) /* 10MB */
 246 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
 247 
 248 #define DMU_USERUSED_OBJECT     (-1ULL)
 249 #define DMU_GROUPUSED_OBJECT    (-2ULL)
 250 #define DMU_DEADLIST_OBJECT     (-3ULL)
 251 
 252 /*
 253  * artificial blkids for bonus buffer and spill blocks
 254  */
 255 #define DMU_BONUS_BLKID         (-1ULL)
 256 #define DMU_SPILL_BLKID         (-2ULL)
 257 /*
 258  * Public routines to create, destroy, open, and close objsets.
 259  */
 260 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
 261 int dmu_objset_own(const char *name, dmu_objset_type_t type,
 262     boolean_t readonly, void *tag, objset_t **osp);
 263 void dmu_objset_rele(objset_t *os, void *tag);
 264 void dmu_objset_disown(objset_t *os, void *tag);
 265 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
 266 
 267 int dmu_objset_evict_dbufs(objset_t *os);
 268 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
 269     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
 270 int dmu_objset_clone(const char *name, struct dsl_dataset *clone_origin,
 271     uint64_t flags);
 272 int dmu_objset_destroy(const char *name, boolean_t defer);
 273 int dmu_snapshots_destroy_nvl(struct nvlist *snaps, boolean_t defer, char *);
 274 int dmu_objset_snapshot(char *fsname, char *snapname, char *tag,
 275     struct nvlist *props, boolean_t recursive, boolean_t temporary, int fd);
 276 int dmu_objset_rename(const char *name, const char *newname,
 277     boolean_t recursive);
 278 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
 279     int flags);
 280 void dmu_objset_byteswap(void *buf, size_t size);
 281 
 282 typedef struct dmu_buf {
 283         uint64_t db_object;             /* object that this buffer is part of */
 284         uint64_t db_offset;             /* byte offset in this object */
 285         uint64_t db_size;               /* size of buffer in bytes */
 286         void *db_data;                  /* data in buffer */
 287 } dmu_buf_t;
 288 
 289 typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr);
 290 
 291 /*
 292  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
 293  */
 294 #define DMU_POOL_DIRECTORY_OBJECT       1
 295 #define DMU_POOL_CONFIG                 "config"
 296 #define DMU_POOL_FEATURES_FOR_WRITE     "features_for_write"
 297 #define DMU_POOL_FEATURES_FOR_READ      "features_for_read"
 298 #define DMU_POOL_FEATURE_DESCRIPTIONS   "feature_descriptions"
 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 
 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  */
 331 uint64_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);
 333 int 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);
 335 int 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  */
 354 int 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  */
 367 int 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  */
 381 int 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  */
 388 void 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  */
 395 void 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 
 405 void 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 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
 418 int dmu_bonus_max(void);
 419 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
 420 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
 421 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
 422 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
 423 
 424 /*
 425  * Special spill buffer support used by "SA" framework
 426  */
 427 
 428 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
 429 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
 430     void *tag, dmu_buf_t **dbp);
 431 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
 432 
 433 /*
 434  * Obtain the DMU buffer from the specified object which contains the
 435  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
 436  * that it will remain in memory.  You must release the hold with
 437  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
 438  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
 439  *
 440  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
 441  * on the returned buffer before reading or writing the buffer's
 442  * db_data.  The comments for those routines describe what particular
 443  * operations are valid after calling them.
 444  *
 445  * The object number must be a valid, allocated object number.
 446  */
 447 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
 448     void *tag, dmu_buf_t **, int flags);
 449 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
 450 void dmu_buf_rele(dmu_buf_t *db, void *tag);
 451 uint64_t dmu_buf_refcount(dmu_buf_t *db);
 452 
 453 /*
 454  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
 455  * range of an object.  A pointer to an array of dmu_buf_t*'s is
 456  * returned (in *dbpp).
 457  *
 458  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
 459  * frees the array.  The hold on the array of buffers MUST be released
 460  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
 461  * individually with dmu_buf_rele.
 462  */
 463 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
 464     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
 465 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
 466 
 467 /*
 468  * Returns NULL on success, or the existing user ptr if it's already
 469  * been set.
 470  *
 471  * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
 472  *
 473  * user_data_ptr_ptr should be NULL, or a pointer to a pointer which
 474  * will be set to db->db_data when you are allowed to access it.  Note
 475  * that db->db_data (the pointer) can change when you do dmu_buf_read(),
 476  * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill().
 477  * *user_data_ptr_ptr will be set to the new value when it changes.
 478  *
 479  * If non-NULL, pageout func will be called when this buffer is being
 480  * excised from the cache, so that you can clean up the data structure
 481  * pointed to by user_ptr.
 482  *
 483  * dmu_evict_user() will call the pageout func for all buffers in a
 484  * objset with a given pageout func.
 485  */
 486 void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr,
 487     dmu_buf_evict_func_t *pageout_func);
 488 /*
 489  * set_user_ie is the same as set_user, but request immediate eviction
 490  * when hold count goes to zero.
 491  */
 492 void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
 493     void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func);
 494 void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
 495     void *user_ptr, void *user_data_ptr_ptr,
 496     dmu_buf_evict_func_t *pageout_func);
 497 void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
 498 
 499 /*
 500  * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
 501  */
 502 void *dmu_buf_get_user(dmu_buf_t *db);
 503 
 504 /*
 505  * Indicate that you are going to modify the buffer's data (db_data).
 506  *
 507  * The transaction (tx) must be assigned to a txg (ie. you've called
 508  * dmu_tx_assign()).  The buffer's object must be held in the tx
 509  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
 510  */
 511 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
 512 
 513 /*
 514  * Tells if the given dbuf is freeable.
 515  */
 516 boolean_t dmu_buf_freeable(dmu_buf_t *);
 517 
 518 /*
 519  * You must create a transaction, then hold the objects which you will
 520  * (or might) modify as part of this transaction.  Then you must assign
 521  * the transaction to a transaction group.  Once the transaction has
 522  * been assigned, you can modify buffers which belong to held objects as
 523  * part of this transaction.  You can't modify buffers before the
 524  * transaction has been assigned; you can't modify buffers which don't
 525  * belong to objects which this transaction holds; you can't hold
 526  * objects once the transaction has been assigned.  You may hold an
 527  * object which you are going to free (with dmu_object_free()), but you
 528  * don't have to.
 529  *
 530  * You can abort the transaction before it has been assigned.
 531  *
 532  * Note that you may hold buffers (with dmu_buf_hold) at any time,
 533  * regardless of transaction state.
 534  */
 535 
 536 #define DMU_NEW_OBJECT  (-1ULL)
 537 #define DMU_OBJECT_END  (-1ULL)
 538 
 539 dmu_tx_t *dmu_tx_create(objset_t *os);
 540 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
 541 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
 542     uint64_t len);
 543 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
 544 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
 545 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
 546 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
 547 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
 548 void dmu_tx_abort(dmu_tx_t *tx);
 549 int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
 550 void dmu_tx_wait(dmu_tx_t *tx);
 551 void dmu_tx_commit(dmu_tx_t *tx);
 552 
 553 /*
 554  * To register a commit callback, dmu_tx_callback_register() must be called.
 555  *
 556  * dcb_data is a pointer to caller private data that is passed on as a
 557  * callback parameter. The caller is responsible for properly allocating and
 558  * freeing it.
 559  *
 560  * When registering a callback, the transaction must be already created, but
 561  * it cannot be committed or aborted. It can be assigned to a txg or not.
 562  *
 563  * The callback will be called after the transaction has been safely written
 564  * to stable storage and will also be called if the dmu_tx is aborted.
 565  * If there is any error which prevents the transaction from being committed to
 566  * disk, the callback will be called with a value of error != 0.
 567  */
 568 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
 569 
 570 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
 571     void *dcb_data);
 572 
 573 /*
 574  * Free up the data blocks for a defined range of a file.  If size is
 575  * -1, the range from offset to end-of-file is freed.
 576  */
 577 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
 578         uint64_t size, dmu_tx_t *tx);
 579 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
 580         uint64_t size);
 581 int dmu_free_object(objset_t *os, uint64_t object);
 582 
 583 /*
 584  * Convenience functions.
 585  *
 586  * Canfail routines will return 0 on success, or an errno if there is a
 587  * nonrecoverable I/O error.
 588  */
 589 #define DMU_READ_PREFETCH       0 /* prefetch */
 590 #define DMU_READ_NO_PREFETCH    1 /* don't prefetch */
 591 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 592         void *buf, uint32_t flags);
 593 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 594         const void *buf, dmu_tx_t *tx);
 595 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 596         dmu_tx_t *tx);
 597 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
 598 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
 599     dmu_tx_t *tx);
 600 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
 601     dmu_tx_t *tx);
 602 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
 603     uint64_t size, struct page *pp, dmu_tx_t *tx);
 604 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
 605 void dmu_return_arcbuf(struct arc_buf *buf);
 606 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
 607     dmu_tx_t *tx);
 608 int dmu_xuio_init(struct xuio *uio, int niov);
 609 void dmu_xuio_fini(struct xuio *uio);
 610 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
 611     size_t n);
 612 int dmu_xuio_cnt(struct xuio *uio);
 613 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
 614 void dmu_xuio_clear(struct xuio *uio, int i);
 615 void xuio_stat_wbuf_copied();
 616 void xuio_stat_wbuf_nocopy();
 617 
 618 extern int zfs_prefetch_disable;
 619 
 620 /*
 621  * Asynchronously try to read in the data.
 622  */
 623 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
 624     uint64_t len);
 625 
 626 typedef struct dmu_object_info {
 627         /* All sizes are in bytes unless otherwise indicated. */
 628         uint32_t doi_data_block_size;
 629         uint32_t doi_metadata_block_size;
 630         dmu_object_type_t doi_type;
 631         dmu_object_type_t doi_bonus_type;
 632         uint64_t doi_bonus_size;
 633         uint8_t doi_indirection;                /* 2 = dnode->indirect->data */
 634         uint8_t doi_checksum;
 635         uint8_t doi_compress;
 636         uint8_t doi_pad[5];
 637         uint64_t doi_physical_blocks_512;       /* data + metadata, 512b blks */
 638         uint64_t doi_max_offset;
 639         uint64_t doi_fill_count;                /* number of non-empty blocks */
 640 } dmu_object_info_t;
 641 
 642 typedef void arc_byteswap_func_t(void *buf, size_t size);
 643 
 644 typedef struct dmu_object_type_info {
 645         dmu_object_byteswap_t   ot_byteswap;
 646         boolean_t               ot_metadata;
 647         char                    *ot_name;
 648 } dmu_object_type_info_t;
 649 
 650 typedef struct dmu_object_byteswap_info {
 651         arc_byteswap_func_t     *ob_func;
 652         char                    *ob_name;
 653 } dmu_object_byteswap_info_t;
 654 
 655 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
 656 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
 657 
 658 /*
 659  * Get information on a DMU object.
 660  *
 661  * Return 0 on success or ENOENT if object is not allocated.
 662  *
 663  * If doi is NULL, just indicates whether the object exists.
 664  */
 665 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
 666 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
 667 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
 668 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
 669     u_longlong_t *nblk512);
 670 
 671 typedef struct dmu_objset_stats {
 672         uint64_t dds_num_clones; /* number of clones of this */
 673         uint64_t dds_creation_txg;
 674         uint64_t dds_guid;
 675         dmu_objset_type_t dds_type;
 676         uint8_t dds_is_snapshot;
 677         uint8_t dds_inconsistent;
 678         char dds_origin[MAXNAMELEN];
 679 } dmu_objset_stats_t;
 680 
 681 /*
 682  * Get stats on a dataset.
 683  */
 684 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
 685 
 686 /*
 687  * Add entries to the nvlist for all the objset's properties.  See
 688  * zfs_prop_table[] and zfs(1m) for details on the properties.
 689  */
 690 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
 691 
 692 /*
 693  * Get the space usage statistics for statvfs().
 694  *
 695  * refdbytes is the amount of space "referenced" by this objset.
 696  * availbytes is the amount of space available to this objset, taking
 697  * into account quotas & reservations, assuming that no other objsets
 698  * use the space first.  These values correspond to the 'referenced' and
 699  * 'available' properties, described in the zfs(1m) manpage.
 700  *
 701  * usedobjs and availobjs are the number of objects currently allocated,
 702  * and available.
 703  */
 704 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
 705     uint64_t *usedobjsp, uint64_t *availobjsp);
 706 
 707 /*
 708  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
 709  * (Contrast with the ds_guid which is a 64-bit ID that will never
 710  * change, so there is a small probability that it will collide.)
 711  */
 712 uint64_t dmu_objset_fsid_guid(objset_t *os);
 713 
 714 /*
 715  * Get the [cm]time for an objset's snapshot dir
 716  */
 717 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
 718 
 719 int dmu_objset_is_snapshot(objset_t *os);
 720 
 721 extern struct spa *dmu_objset_spa(objset_t *os);
 722 extern struct zilog *dmu_objset_zil(objset_t *os);
 723 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
 724 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
 725 extern void dmu_objset_name(objset_t *os, char *buf);
 726 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
 727 extern uint64_t dmu_objset_id(objset_t *os);
 728 extern uint64_t dmu_objset_syncprop(objset_t *os);
 729 extern uint64_t dmu_objset_logbias(objset_t *os);
 730 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
 731     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
 732 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
 733     int maxlen, boolean_t *conflict);
 734 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
 735     uint64_t *idp, uint64_t *offp);
 736 
 737 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
 738     void *bonus, uint64_t *userp, uint64_t *groupp);
 739 extern void dmu_objset_register_type(dmu_objset_type_t ost,
 740     objset_used_cb_t *cb);
 741 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
 742 extern void *dmu_objset_get_user(objset_t *os);
 743 
 744 /*
 745  * Return the txg number for the given assigned transaction.
 746  */
 747 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
 748 
 749 /*
 750  * Synchronous write.
 751  * If a parent zio is provided this function initiates a write on the
 752  * provided buffer as a child of the parent zio.
 753  * In the absence of a parent zio, the write is completed synchronously.
 754  * At write completion, blk is filled with the bp of the written block.
 755  * Note that while the data covered by this function will be on stable
 756  * storage when the write completes this new data does not become a
 757  * permanent part of the file until the associated transaction commits.
 758  */
 759 
 760 /*
 761  * {zfs,zvol,ztest}_get_done() args
 762  */
 763 typedef struct zgd {
 764         struct zilog    *zgd_zilog;
 765         struct blkptr   *zgd_bp;
 766         dmu_buf_t       *zgd_db;
 767         struct rl       *zgd_rl;
 768         void            *zgd_private;
 769 } zgd_t;
 770 
 771 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
 772 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
 773 
 774 /*
 775  * Find the next hole or data block in file starting at *off
 776  * Return found offset in *off. Return ESRCH for end of file.
 777  */
 778 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
 779     uint64_t *off);
 780 
 781 /*
 782  * Initial setup and final teardown.
 783  */
 784 extern void dmu_init(void);
 785 extern void dmu_fini(void);
 786 
 787 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
 788     uint64_t object, uint64_t offset, int len);
 789 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
 790     dmu_traverse_cb_t cb, void *arg);
 791 
 792 int dmu_send(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
 793     int outfd, struct vnode *vp, offset_t *off);
 794 int dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorign,
 795     uint64_t *sizep);
 796 
 797 typedef struct dmu_recv_cookie {
 798         /*
 799          * This structure is opaque!
 800          *
 801          * If logical and real are different, we are recving the stream
 802          * into the "real" temporary clone, and then switching it with
 803          * the "logical" target.
 804          */
 805         uint64_t drc_logical_dsobj;
 806         struct dsl_dataset *drc_real_ds;
 807         struct drr_begin *drc_drrb;
 808         char *drc_tosnap;
 809         char *drc_top_ds;
 810         boolean_t drc_newfs;
 811         boolean_t drc_force;
 812         struct avl_tree *drc_guid_to_ds_map;
 813 } dmu_recv_cookie_t;
 814 
 815 int dmu_recv_begin(char *tofs, char *tosnap, char *topds, struct drr_begin *,
 816     boolean_t force, objset_t *origin, dmu_recv_cookie_t *);
 817 int dmu_recv_stream(dmu_recv_cookie_t *drc, struct vnode *vp, offset_t *voffp,
 818     int cleanup_fd, uint64_t *action_handlep);
 819 int dmu_recv_end(dmu_recv_cookie_t *drc);
 820 
 821 int dmu_diff(objset_t *tosnap, objset_t *fromsnap, struct vnode *vp,
 822     offset_t *off);
 823 
 824 /* CRC64 table */
 825 #define ZFS_CRC64_POLY  0xC96C5795D7870F42ULL   /* ECMA-182, reflected form */
 826 extern uint64_t zfs_crc64_table[256];
 827 
 828 #ifdef  __cplusplus
 829 }
 830 #endif
 831 
 832 #endif  /* _SYS_DMU_H */