1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  25  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  26  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  27  * Copyright 2013 DEY Storage Systems, Inc.
  28  */
  29 
  30 /* Portions Copyright 2010 Robert Milkowski */
  31 
  32 #ifndef _SYS_DMU_H
  33 #define _SYS_DMU_H
  34 
  35 /*
  36  * This file describes the interface that the DMU provides for its
  37  * consumers.
  38  *
  39  * The DMU also interacts with the SPA.  That interface is described in
  40  * dmu_spa.h.
  41  */
  42 
  43 #include <sys/inttypes.h>
  44 #include <sys/types.h>
  45 #include <sys/param.h>
  46 #include <sys/cred.h>
  47 #include <sys/time.h>
  48 #include <sys/fs/zfs.h>
  49 
  50 #ifdef  __cplusplus
  51 extern "C" {
  52 #endif
  53 
  54 struct uio;
  55 struct xuio;
  56 struct page;
  57 struct vnode;
  58 struct spa;
  59 struct zilog;
  60 struct zio;
  61 struct blkptr;
  62 struct zap_cursor;
  63 struct dsl_dataset;
  64 struct dsl_pool;
  65 struct dnode;
  66 struct drr_begin;
  67 struct drr_end;
  68 struct zbookmark;
  69 struct spa;
  70 struct nvlist;
  71 struct arc_buf;
  72 struct zio_prop;
  73 struct sa_handle;
  74 
  75 typedef struct objset objset_t;
  76 typedef struct dmu_tx dmu_tx_t;
  77 typedef struct dsl_dir dsl_dir_t;
  78 
  79 typedef 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 /*
 123  * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
 124  * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
 125  * is repurposed for embedded BPs.
 126  */
 127 #define DMU_OT_HAS_FILL(ot) \
 128         ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
 129 
 130 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
 131         ((ot) & DMU_OT_BYTESWAP_MASK) : \
 132         dmu_ot[(ot)].ot_byteswap)
 133 
 134 typedef enum dmu_object_type {
 135         DMU_OT_NONE,
 136         /* general: */
 137         DMU_OT_OBJECT_DIRECTORY,        /* ZAP */
 138         DMU_OT_OBJECT_ARRAY,            /* UINT64 */
 139         DMU_OT_PACKED_NVLIST,           /* UINT8 (XDR by nvlist_pack/unpack) */
 140         DMU_OT_PACKED_NVLIST_SIZE,      /* UINT64 */
 141         DMU_OT_BPOBJ,                   /* UINT64 */
 142         DMU_OT_BPOBJ_HDR,               /* UINT64 */
 143         /* spa: */
 144         DMU_OT_SPACE_MAP_HEADER,        /* UINT64 */
 145         DMU_OT_SPACE_MAP,               /* UINT64 */
 146         /* zil: */
 147         DMU_OT_INTENT_LOG,              /* UINT64 */
 148         /* dmu: */
 149         DMU_OT_DNODE,                   /* DNODE */
 150         DMU_OT_OBJSET,                  /* OBJSET */
 151         /* dsl: */
 152         DMU_OT_DSL_DIR,                 /* UINT64 */
 153         DMU_OT_DSL_DIR_CHILD_MAP,       /* ZAP */
 154         DMU_OT_DSL_DS_SNAP_MAP,         /* ZAP */
 155         DMU_OT_DSL_PROPS,               /* ZAP */
 156         DMU_OT_DSL_DATASET,             /* UINT64 */
 157         /* zpl: */
 158         DMU_OT_ZNODE,                   /* ZNODE */
 159         DMU_OT_OLDACL,                  /* Old ACL */
 160         DMU_OT_PLAIN_FILE_CONTENTS,     /* UINT8 */
 161         DMU_OT_DIRECTORY_CONTENTS,      /* ZAP */
 162         DMU_OT_MASTER_NODE,             /* ZAP */
 163         DMU_OT_UNLINKED_SET,            /* ZAP */
 164         /* zvol: */
 165         DMU_OT_ZVOL,                    /* UINT8 */
 166         DMU_OT_ZVOL_PROP,               /* ZAP */
 167         /* other; for testing only! */
 168         DMU_OT_PLAIN_OTHER,             /* UINT8 */
 169         DMU_OT_UINT64_OTHER,            /* UINT64 */
 170         DMU_OT_ZAP_OTHER,               /* ZAP */
 171         /* new object types: */
 172         DMU_OT_ERROR_LOG,               /* ZAP */
 173         DMU_OT_SPA_HISTORY,             /* UINT8 */
 174         DMU_OT_SPA_HISTORY_OFFSETS,     /* spa_his_phys_t */
 175         DMU_OT_POOL_PROPS,              /* ZAP */
 176         DMU_OT_DSL_PERMS,               /* ZAP */
 177         DMU_OT_ACL,                     /* ACL */
 178         DMU_OT_SYSACL,                  /* SYSACL */
 179         DMU_OT_FUID,                    /* FUID table (Packed NVLIST UINT8) */
 180         DMU_OT_FUID_SIZE,               /* FUID table size UINT64 */
 181         DMU_OT_NEXT_CLONES,             /* ZAP */
 182         DMU_OT_SCAN_QUEUE,              /* ZAP */
 183         DMU_OT_USERGROUP_USED,          /* ZAP */
 184         DMU_OT_USERGROUP_QUOTA,         /* ZAP */
 185         DMU_OT_USERREFS,                /* ZAP */
 186         DMU_OT_DDT_ZAP,                 /* ZAP */
 187         DMU_OT_DDT_STATS,               /* ZAP */
 188         DMU_OT_SA,                      /* System attr */
 189         DMU_OT_SA_MASTER_NODE,          /* ZAP */
 190         DMU_OT_SA_ATTR_REGISTRATION,    /* ZAP */
 191         DMU_OT_SA_ATTR_LAYOUTS,         /* ZAP */
 192         DMU_OT_SCAN_XLATE,              /* ZAP */
 193         DMU_OT_DEDUP,                   /* fake dedup BP from ddt_bp_create() */
 194         DMU_OT_DEADLIST,                /* ZAP */
 195         DMU_OT_DEADLIST_HDR,            /* UINT64 */
 196         DMU_OT_DSL_CLONES,              /* ZAP */
 197         DMU_OT_BPOBJ_SUBOBJ,            /* UINT64 */
 198         /*
 199          * Do not allocate new object types here. Doing so makes the on-disk
 200          * format incompatible with any other format that uses the same object
 201          * type number.
 202          *
 203          * When creating an object which does not have one of the above types
 204          * use the DMU_OTN_* type with the correct byteswap and metadata
 205          * values.
 206          *
 207          * The DMU_OTN_* types do not have entries in the dmu_ot table,
 208          * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
 209          * of indexing into dmu_ot directly (this works for both DMU_OT_* types
 210          * and DMU_OTN_* types).
 211          */
 212         DMU_OT_NUMTYPES,
 213 
 214         /*
 215          * Names for valid types declared with DMU_OT().
 216          */
 217         DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
 218         DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
 219         DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
 220         DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
 221         DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
 222         DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
 223         DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
 224         DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
 225         DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
 226         DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
 227 } dmu_object_type_t;
 228 
 229 typedef enum txg_how {
 230         TXG_WAIT = 1,
 231         TXG_NOWAIT,
 232         TXG_WAITED,
 233 } txg_how_t;
 234 
 235 void byteswap_uint64_array(void *buf, size_t size);
 236 void byteswap_uint32_array(void *buf, size_t size);
 237 void byteswap_uint16_array(void *buf, size_t size);
 238 void byteswap_uint8_array(void *buf, size_t size);
 239 void zap_byteswap(void *buf, size_t size);
 240 void zfs_oldacl_byteswap(void *buf, size_t size);
 241 void zfs_acl_byteswap(void *buf, size_t size);
 242 void zfs_znode_byteswap(void *buf, size_t size);
 243 
 244 #define DS_FIND_SNAPSHOTS       (1<<0)
 245 #define DS_FIND_CHILDREN        (1<<1)
 246 
 247 /*
 248  * The maximum number of bytes that can be accessed as part of one
 249  * operation, including metadata.
 250  */
 251 #define DMU_MAX_ACCESS (10<<20) /* 10MB */
 252 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
 253 
 254 #define DMU_USERUSED_OBJECT     (-1ULL)
 255 #define DMU_GROUPUSED_OBJECT    (-2ULL)
 256 
 257 /*
 258  * artificial blkids for bonus buffer and spill blocks
 259  */
 260 #define DMU_BONUS_BLKID         (-1ULL)
 261 #define DMU_SPILL_BLKID         (-2ULL)
 262 /*
 263  * Public routines to create, destroy, open, and close objsets.
 264  */
 265 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
 266 int dmu_objset_own(const char *name, dmu_objset_type_t type,
 267     boolean_t readonly, void *tag, objset_t **osp);
 268 void dmu_objset_rele(objset_t *os, void *tag);
 269 void dmu_objset_rele_obj(objset_t *os, void *tag);
 270 void dmu_objset_disown(objset_t *os, void *tag);
 271 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
 272 
 273 void dmu_objset_evict_dbufs(objset_t *os);
 274 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
 275     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
 276 int dmu_objset_clone(const char *name, const char *origin);
 277 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
 278     struct nvlist *errlist);
 279 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
 280 int dmu_objset_snapshot_tmp(const char *, const char *, int);
 281 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
 282     int flags);
 283 void dmu_objset_byteswap(void *buf, size_t size);
 284 int dsl_dataset_rename_snapshot(const char *fsname,
 285     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
 286 
 287 typedef struct dmu_buf {
 288         uint64_t db_object;             /* object that this buffer is part of */
 289         uint64_t db_offset;             /* byte offset in this object */
 290         uint64_t db_size;               /* size of buffer in bytes */
 291         void *db_data;                  /* data in buffer */
 292 } dmu_buf_t;
 293 
 294 typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr);
 295 
 296 /*
 297  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
 298  */
 299 #define DMU_POOL_DIRECTORY_OBJECT       1
 300 #define DMU_POOL_CONFIG                 "config"
 301 #define DMU_POOL_FEATURES_FOR_WRITE     "features_for_write"
 302 #define DMU_POOL_FEATURES_FOR_READ      "features_for_read"
 303 #define DMU_POOL_FEATURE_DESCRIPTIONS   "feature_descriptions"
 304 #define DMU_POOL_FEATURE_ENABLED_TXG    "feature_enabled_txg"
 305 #define DMU_POOL_ROOT_DATASET           "root_dataset"
 306 #define DMU_POOL_SYNC_BPOBJ             "sync_bplist"
 307 #define DMU_POOL_ERRLOG_SCRUB           "errlog_scrub"
 308 #define DMU_POOL_ERRLOG_LAST            "errlog_last"
 309 #define DMU_POOL_SPARES                 "spares"
 310 #define DMU_POOL_DEFLATE                "deflate"
 311 #define DMU_POOL_HISTORY                "history"
 312 #define DMU_POOL_PROPS                  "pool_props"
 313 #define DMU_POOL_L2CACHE                "l2cache"
 314 #define DMU_POOL_TMP_USERREFS           "tmp_userrefs"
 315 #define DMU_POOL_DDT                    "DDT-%s-%s-%s"
 316 #define DMU_POOL_DDT_STATS              "DDT-statistics"
 317 #define DMU_POOL_CREATION_VERSION       "creation_version"
 318 #define DMU_POOL_SCAN                   "scan"
 319 #define DMU_POOL_FREE_BPOBJ             "free_bpobj"
 320 #define DMU_POOL_BPTREE_OBJ             "bptree_obj"
 321 #define DMU_POOL_EMPTY_BPOBJ            "empty_bpobj"
 322 
 323 /*
 324  * Allocate an object from this objset.  The range of object numbers
 325  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
 326  *
 327  * The transaction must be assigned to a txg.  The newly allocated
 328  * object will be "held" in the transaction (ie. you can modify the
 329  * newly allocated object in this transaction).
 330  *
 331  * dmu_object_alloc() chooses an object and returns it in *objectp.
 332  *
 333  * dmu_object_claim() allocates a specific object number.  If that
 334  * number is already allocated, it fails and returns EEXIST.
 335  *
 336  * Return 0 on success, or ENOSPC or EEXIST as specified above.
 337  */
 338 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
 339     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
 340 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
 341     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
 342 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
 343     int blocksize, dmu_object_type_t bonustype, int bonuslen);
 344 
 345 /*
 346  * Free an object from this objset.
 347  *
 348  * The object's data will be freed as well (ie. you don't need to call
 349  * dmu_free(object, 0, -1, tx)).
 350  *
 351  * The object need not be held in the transaction.
 352  *
 353  * If there are any holds on this object's buffers (via dmu_buf_hold()),
 354  * or tx holds on the object (via dmu_tx_hold_object()), you can not
 355  * free it; it fails and returns EBUSY.
 356  *
 357  * If the object is not allocated, it fails and returns ENOENT.
 358  *
 359  * Return 0 on success, or EBUSY or ENOENT as specified above.
 360  */
 361 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
 362 
 363 /*
 364  * Find the next allocated or free object.
 365  *
 366  * The objectp parameter is in-out.  It will be updated to be the next
 367  * object which is allocated.  Ignore objects which have not been
 368  * modified since txg.
 369  *
 370  * XXX Can only be called on a objset with no dirty data.
 371  *
 372  * Returns 0 on success, or ENOENT if there are no more objects.
 373  */
 374 int dmu_object_next(objset_t *os, uint64_t *objectp,
 375     boolean_t hole, uint64_t txg);
 376 
 377 /*
 378  * Set the data blocksize for an object.
 379  *
 380  * The object cannot have any blocks allcated beyond the first.  If
 381  * the first block is allocated already, the new size must be greater
 382  * than the current block size.  If these conditions are not met,
 383  * ENOTSUP will be returned.
 384  *
 385  * Returns 0 on success, or EBUSY if there are any holds on the object
 386  * contents, or ENOTSUP as described above.
 387  */
 388 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
 389     int ibs, dmu_tx_t *tx);
 390 
 391 /*
 392  * Set the checksum property on a dnode.  The new checksum algorithm will
 393  * apply to all newly written blocks; existing blocks will not be affected.
 394  */
 395 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
 396     dmu_tx_t *tx);
 397 
 398 /*
 399  * Set the compress property on a dnode.  The new compression algorithm will
 400  * apply to all newly written blocks; existing blocks will not be affected.
 401  */
 402 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
 403     dmu_tx_t *tx);
 404 
 405 void
 406 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
 407     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
 408     int compressed_size, int byteorder, dmu_tx_t *tx);
 409 
 410 /*
 411  * Decide how to write a block: checksum, compression, number of copies, etc.
 412  */
 413 #define WP_NOFILL       0x1
 414 #define WP_DMU_SYNC     0x2
 415 #define WP_SPILL        0x4
 416 
 417 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
 418     struct zio_prop *zp);
 419 /*
 420  * The bonus data is accessed more or less like a regular buffer.
 421  * You must dmu_bonus_hold() to get the buffer, which will give you a
 422  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
 423  * data.  As with any normal buffer, you must call dmu_buf_read() to
 424  * read db_data, dmu_buf_will_dirty() before modifying it, and the
 425  * object must be held in an assigned transaction before calling
 426  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
 427  * buffer as well.  You must release your hold with dmu_buf_rele().
 428  *
 429  * Returns ENOENT, EIO, or 0.
 430  */
 431 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
 432 int dmu_bonus_max(void);
 433 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
 434 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
 435 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
 436 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
 437 
 438 /*
 439  * Special spill buffer support used by "SA" framework
 440  */
 441 
 442 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
 443 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
 444     void *tag, dmu_buf_t **dbp);
 445 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
 446 
 447 /*
 448  * Obtain the DMU buffer from the specified object which contains the
 449  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
 450  * that it will remain in memory.  You must release the hold with
 451  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
 452  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
 453  *
 454  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
 455  * on the returned buffer before reading or writing the buffer's
 456  * db_data.  The comments for those routines describe what particular
 457  * operations are valid after calling them.
 458  *
 459  * The object number must be a valid, allocated object number.
 460  */
 461 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
 462     void *tag, dmu_buf_t **, int flags);
 463 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
 464 void dmu_buf_rele(dmu_buf_t *db, void *tag);
 465 uint64_t dmu_buf_refcount(dmu_buf_t *db);
 466 
 467 /*
 468  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
 469  * range of an object.  A pointer to an array of dmu_buf_t*'s is
 470  * returned (in *dbpp).
 471  *
 472  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
 473  * frees the array.  The hold on the array of buffers MUST be released
 474  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
 475  * individually with dmu_buf_rele.
 476  */
 477 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
 478     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
 479 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
 480 
 481 /*
 482  * Returns NULL on success, or the existing user ptr if it's already
 483  * been set.
 484  *
 485  * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
 486  *
 487  * user_data_ptr_ptr should be NULL, or a pointer to a pointer which
 488  * will be set to db->db_data when you are allowed to access it.  Note
 489  * that db->db_data (the pointer) can change when you do dmu_buf_read(),
 490  * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill().
 491  * *user_data_ptr_ptr will be set to the new value when it changes.
 492  *
 493  * If non-NULL, pageout func will be called when this buffer is being
 494  * excised from the cache, so that you can clean up the data structure
 495  * pointed to by user_ptr.
 496  *
 497  * dmu_evict_user() will call the pageout func for all buffers in a
 498  * objset with a given pageout func.
 499  */
 500 void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr,
 501     dmu_buf_evict_func_t *pageout_func);
 502 /*
 503  * set_user_ie is the same as set_user, but request immediate eviction
 504  * when hold count goes to zero.
 505  */
 506 void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
 507     void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func);
 508 void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
 509     void *user_ptr, void *user_data_ptr_ptr,
 510     dmu_buf_evict_func_t *pageout_func);
 511 void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
 512 
 513 /*
 514  * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
 515  */
 516 void *dmu_buf_get_user(dmu_buf_t *db);
 517 
 518 /*
 519  * Returns the blkptr associated with this dbuf, or NULL if not set.
 520  */
 521 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
 522 
 523 /*
 524  * Indicate that you are going to modify the buffer's data (db_data).
 525  *
 526  * The transaction (tx) must be assigned to a txg (ie. you've called
 527  * dmu_tx_assign()).  The buffer's object must be held in the tx
 528  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
 529  */
 530 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
 531 
 532 /*
 533  * Tells if the given dbuf is freeable.
 534  */
 535 boolean_t dmu_buf_freeable(dmu_buf_t *);
 536 
 537 /*
 538  * You must create a transaction, then hold the objects which you will
 539  * (or might) modify as part of this transaction.  Then you must assign
 540  * the transaction to a transaction group.  Once the transaction has
 541  * been assigned, you can modify buffers which belong to held objects as
 542  * part of this transaction.  You can't modify buffers before the
 543  * transaction has been assigned; you can't modify buffers which don't
 544  * belong to objects which this transaction holds; you can't hold
 545  * objects once the transaction has been assigned.  You may hold an
 546  * object which you are going to free (with dmu_object_free()), but you
 547  * don't have to.
 548  *
 549  * You can abort the transaction before it has been assigned.
 550  *
 551  * Note that you may hold buffers (with dmu_buf_hold) at any time,
 552  * regardless of transaction state.
 553  */
 554 
 555 #define DMU_NEW_OBJECT  (-1ULL)
 556 #define DMU_OBJECT_END  (-1ULL)
 557 
 558 dmu_tx_t *dmu_tx_create(objset_t *os);
 559 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
 560 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
 561     uint64_t len);
 562 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
 563 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
 564 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
 565 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
 566 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
 567 void dmu_tx_abort(dmu_tx_t *tx);
 568 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
 569 void dmu_tx_wait(dmu_tx_t *tx);
 570 void dmu_tx_commit(dmu_tx_t *tx);
 571 
 572 /*
 573  * To register a commit callback, dmu_tx_callback_register() must be called.
 574  *
 575  * dcb_data is a pointer to caller private data that is passed on as a
 576  * callback parameter. The caller is responsible for properly allocating and
 577  * freeing it.
 578  *
 579  * When registering a callback, the transaction must be already created, but
 580  * it cannot be committed or aborted. It can be assigned to a txg or not.
 581  *
 582  * The callback will be called after the transaction has been safely written
 583  * to stable storage and will also be called if the dmu_tx is aborted.
 584  * If there is any error which prevents the transaction from being committed to
 585  * disk, the callback will be called with a value of error != 0.
 586  */
 587 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
 588 
 589 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
 590     void *dcb_data);
 591 
 592 /*
 593  * Free up the data blocks for a defined range of a file.  If size is
 594  * -1, the range from offset to end-of-file is freed.
 595  */
 596 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
 597         uint64_t size, dmu_tx_t *tx);
 598 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
 599         uint64_t size);
 600 int dmu_free_long_object(objset_t *os, uint64_t object);
 601 
 602 /*
 603  * Convenience functions.
 604  *
 605  * Canfail routines will return 0 on success, or an errno if there is a
 606  * nonrecoverable I/O error.
 607  */
 608 #define DMU_READ_PREFETCH       0 /* prefetch */
 609 #define DMU_READ_NO_PREFETCH    1 /* don't prefetch */
 610 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 611         void *buf, uint32_t flags);
 612 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 613         const void *buf, dmu_tx_t *tx);
 614 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 615         dmu_tx_t *tx);
 616 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
 617 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
 618     dmu_tx_t *tx);
 619 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
 620     dmu_tx_t *tx);
 621 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
 622     uint64_t size, struct page *pp, dmu_tx_t *tx);
 623 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
 624 void dmu_return_arcbuf(struct arc_buf *buf);
 625 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
 626     dmu_tx_t *tx);
 627 int dmu_xuio_init(struct xuio *uio, int niov);
 628 void dmu_xuio_fini(struct xuio *uio);
 629 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
 630     size_t n);
 631 int dmu_xuio_cnt(struct xuio *uio);
 632 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
 633 void dmu_xuio_clear(struct xuio *uio, int i);
 634 void xuio_stat_wbuf_copied();
 635 void xuio_stat_wbuf_nocopy();
 636 
 637 extern int zfs_prefetch_disable;
 638 
 639 /*
 640  * Asynchronously try to read in the data.
 641  */
 642 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
 643     uint64_t len);
 644 
 645 typedef struct dmu_object_info {
 646         /* All sizes are in bytes unless otherwise indicated. */
 647         uint32_t doi_data_block_size;
 648         uint32_t doi_metadata_block_size;
 649         dmu_object_type_t doi_type;
 650         dmu_object_type_t doi_bonus_type;
 651         uint64_t doi_bonus_size;
 652         uint8_t doi_indirection;                /* 2 = dnode->indirect->data */
 653         uint8_t doi_checksum;
 654         uint8_t doi_compress;
 655         uint8_t doi_pad[5];
 656         uint64_t doi_physical_blocks_512;       /* data + metadata, 512b blks */
 657         uint64_t doi_max_offset;
 658         uint64_t doi_fill_count;                /* number of non-empty blocks */
 659 } dmu_object_info_t;
 660 
 661 typedef void arc_byteswap_func_t(void *buf, size_t size);
 662 
 663 typedef struct dmu_object_type_info {
 664         dmu_object_byteswap_t   ot_byteswap;
 665         boolean_t               ot_metadata;
 666         char                    *ot_name;
 667 } dmu_object_type_info_t;
 668 
 669 typedef struct dmu_object_byteswap_info {
 670         arc_byteswap_func_t     *ob_func;
 671         char                    *ob_name;
 672 } dmu_object_byteswap_info_t;
 673 
 674 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
 675 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
 676 
 677 /*
 678  * Get information on a DMU object.
 679  *
 680  * Return 0 on success or ENOENT if object is not allocated.
 681  *
 682  * If doi is NULL, just indicates whether the object exists.
 683  */
 684 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
 685 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
 686 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
 687 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
 688 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
 689 /*
 690  * Like dmu_object_info_from_db, but faster still when you only care about
 691  * the size.  This is specifically optimized for zfs_getattr().
 692  */
 693 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
 694     u_longlong_t *nblk512);
 695 
 696 typedef struct dmu_objset_stats {
 697         uint64_t dds_num_clones; /* number of clones of this */
 698         uint64_t dds_creation_txg;
 699         uint64_t dds_guid;
 700         dmu_objset_type_t dds_type;
 701         uint8_t dds_is_snapshot;
 702         uint8_t dds_inconsistent;
 703         char dds_origin[MAXNAMELEN];
 704 } dmu_objset_stats_t;
 705 
 706 /*
 707  * Get stats on a dataset.
 708  */
 709 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
 710 
 711 /*
 712  * Add entries to the nvlist for all the objset's properties.  See
 713  * zfs_prop_table[] and zfs(1m) for details on the properties.
 714  */
 715 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
 716 
 717 /*
 718  * Get the space usage statistics for statvfs().
 719  *
 720  * refdbytes is the amount of space "referenced" by this objset.
 721  * availbytes is the amount of space available to this objset, taking
 722  * into account quotas & reservations, assuming that no other objsets
 723  * use the space first.  These values correspond to the 'referenced' and
 724  * 'available' properties, described in the zfs(1m) manpage.
 725  *
 726  * usedobjs and availobjs are the number of objects currently allocated,
 727  * and available.
 728  */
 729 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
 730     uint64_t *usedobjsp, uint64_t *availobjsp);
 731 
 732 /*
 733  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
 734  * (Contrast with the ds_guid which is a 64-bit ID that will never
 735  * change, so there is a small probability that it will collide.)
 736  */
 737 uint64_t dmu_objset_fsid_guid(objset_t *os);
 738 
 739 /*
 740  * Get the [cm]time for an objset's snapshot dir
 741  */
 742 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
 743 
 744 int dmu_objset_is_snapshot(objset_t *os);
 745 
 746 extern struct spa *dmu_objset_spa(objset_t *os);
 747 extern struct zilog *dmu_objset_zil(objset_t *os);
 748 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
 749 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
 750 extern void dmu_objset_name(objset_t *os, char *buf);
 751 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
 752 extern uint64_t dmu_objset_id(objset_t *os);
 753 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
 754 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
 755 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
 756     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
 757 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
 758     int maxlen, boolean_t *conflict);
 759 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
 760     uint64_t *idp, uint64_t *offp);
 761 
 762 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
 763     void *bonus, uint64_t *userp, uint64_t *groupp);
 764 extern void dmu_objset_register_type(dmu_objset_type_t ost,
 765     objset_used_cb_t *cb);
 766 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
 767 extern void *dmu_objset_get_user(objset_t *os);
 768 
 769 /*
 770  * Return the txg number for the given assigned transaction.
 771  */
 772 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
 773 
 774 /*
 775  * Synchronous write.
 776  * If a parent zio is provided this function initiates a write on the
 777  * provided buffer as a child of the parent zio.
 778  * In the absence of a parent zio, the write is completed synchronously.
 779  * At write completion, blk is filled with the bp of the written block.
 780  * Note that while the data covered by this function will be on stable
 781  * storage when the write completes this new data does not become a
 782  * permanent part of the file until the associated transaction commits.
 783  */
 784 
 785 /*
 786  * {zfs,zvol,ztest}_get_done() args
 787  */
 788 typedef struct zgd {
 789         struct zilog    *zgd_zilog;
 790         struct blkptr   *zgd_bp;
 791         dmu_buf_t       *zgd_db;
 792         struct rl       *zgd_rl;
 793         void            *zgd_private;
 794 } zgd_t;
 795 
 796 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
 797 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
 798 
 799 /*
 800  * Find the next hole or data block in file starting at *off
 801  * Return found offset in *off. Return ESRCH for end of file.
 802  */
 803 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
 804     uint64_t *off);
 805 
 806 /*
 807  * Initial setup and final teardown.
 808  */
 809 extern void dmu_init(void);
 810 extern void dmu_fini(void);
 811 
 812 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
 813     uint64_t object, uint64_t offset, int len);
 814 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
 815     dmu_traverse_cb_t cb, void *arg);
 816 
 817 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
 818     struct vnode *vp, offset_t *offp);
 819 
 820 /* CRC64 table */
 821 #define ZFS_CRC64_POLY  0xC96C5795D7870F42ULL   /* ECMA-182, reflected form */
 822 extern uint64_t zfs_crc64_table[256];
 823 
 824 extern int zfs_mdcomp_disable;
 825 
 826 #ifdef  __cplusplus
 827 }
 828 #endif
 829 
 830 #endif  /* _SYS_DMU_H */