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/zfs_context.h>
  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 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
 123         ((ot) & DMU_OT_BYTESWAP_MASK) : \
 124         dmu_ot[(ot)].ot_byteswap)
 125 
 126 typedef enum dmu_object_type {
 127         DMU_OT_NONE,
 128         /* general: */
 129         DMU_OT_OBJECT_DIRECTORY,        /* ZAP */
 130         DMU_OT_OBJECT_ARRAY,            /* UINT64 */
 131         DMU_OT_PACKED_NVLIST,           /* UINT8 (XDR by nvlist_pack/unpack) */
 132         DMU_OT_PACKED_NVLIST_SIZE,      /* UINT64 */
 133         DMU_OT_BPOBJ,                   /* UINT64 */
 134         DMU_OT_BPOBJ_HDR,               /* UINT64 */
 135         /* spa: */
 136         DMU_OT_SPACE_MAP_HEADER,        /* UINT64 */
 137         DMU_OT_SPACE_MAP,               /* UINT64 */
 138         /* zil: */
 139         DMU_OT_INTENT_LOG,              /* UINT64 */
 140         /* dmu: */
 141         DMU_OT_DNODE,                   /* DNODE */
 142         DMU_OT_OBJSET,                  /* OBJSET */
 143         /* dsl: */
 144         DMU_OT_DSL_DIR,                 /* UINT64 */
 145         DMU_OT_DSL_DIR_CHILD_MAP,       /* ZAP */
 146         DMU_OT_DSL_DS_SNAP_MAP,         /* ZAP */
 147         DMU_OT_DSL_PROPS,               /* ZAP */
 148         DMU_OT_DSL_DATASET,             /* UINT64 */
 149         /* zpl: */
 150         DMU_OT_ZNODE,                   /* ZNODE */
 151         DMU_OT_OLDACL,                  /* Old ACL */
 152         DMU_OT_PLAIN_FILE_CONTENTS,     /* UINT8 */
 153         DMU_OT_DIRECTORY_CONTENTS,      /* ZAP */
 154         DMU_OT_MASTER_NODE,             /* ZAP */
 155         DMU_OT_UNLINKED_SET,            /* ZAP */
 156         /* zvol: */
 157         DMU_OT_ZVOL,                    /* UINT8 */
 158         DMU_OT_ZVOL_PROP,               /* ZAP */
 159         /* other; for testing only! */
 160         DMU_OT_PLAIN_OTHER,             /* UINT8 */
 161         DMU_OT_UINT64_OTHER,            /* UINT64 */
 162         DMU_OT_ZAP_OTHER,               /* ZAP */
 163         /* new object types: */
 164         DMU_OT_ERROR_LOG,               /* ZAP */
 165         DMU_OT_SPA_HISTORY,             /* UINT8 */
 166         DMU_OT_SPA_HISTORY_OFFSETS,     /* spa_his_phys_t */
 167         DMU_OT_POOL_PROPS,              /* ZAP */
 168         DMU_OT_DSL_PERMS,               /* ZAP */
 169         DMU_OT_ACL,                     /* ACL */
 170         DMU_OT_SYSACL,                  /* SYSACL */
 171         DMU_OT_FUID,                    /* FUID table (Packed NVLIST UINT8) */
 172         DMU_OT_FUID_SIZE,               /* FUID table size UINT64 */
 173         DMU_OT_NEXT_CLONES,             /* ZAP */
 174         DMU_OT_SCAN_QUEUE,              /* ZAP */
 175         DMU_OT_USERGROUP_USED,          /* ZAP */
 176         DMU_OT_USERGROUP_QUOTA,         /* ZAP */
 177         DMU_OT_USERREFS,                /* ZAP */
 178         DMU_OT_DDT_ZAP,                 /* ZAP */
 179         DMU_OT_DDT_STATS,               /* ZAP */
 180         DMU_OT_SA,                      /* System attr */
 181         DMU_OT_SA_MASTER_NODE,          /* ZAP */
 182         DMU_OT_SA_ATTR_REGISTRATION,    /* ZAP */
 183         DMU_OT_SA_ATTR_LAYOUTS,         /* ZAP */
 184         DMU_OT_SCAN_XLATE,              /* ZAP */
 185         DMU_OT_DEDUP,                   /* fake dedup BP from ddt_bp_create() */
 186         DMU_OT_DEADLIST,                /* ZAP */
 187         DMU_OT_DEADLIST_HDR,            /* UINT64 */
 188         DMU_OT_DSL_CLONES,              /* ZAP */
 189         DMU_OT_BPOBJ_SUBOBJ,            /* UINT64 */
 190         /*
 191          * Do not allocate new object types here. Doing so makes the on-disk
 192          * format incompatible with any other format that uses the same object
 193          * type number.
 194          *
 195          * When creating an object which does not have one of the above types
 196          * use the DMU_OTN_* type with the correct byteswap and metadata
 197          * values.
 198          *
 199          * The DMU_OTN_* types do not have entries in the dmu_ot table,
 200          * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
 201          * of indexing into dmu_ot directly (this works for both DMU_OT_* types
 202          * and DMU_OTN_* types).
 203          */
 204         DMU_OT_NUMTYPES,
 205 
 206         /*
 207          * Names for valid types declared with DMU_OT().
 208          */
 209         DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
 210         DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
 211         DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
 212         DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
 213         DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
 214         DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
 215         DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
 216         DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
 217         DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
 218         DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
 219 } dmu_object_type_t;
 220 
 221 typedef enum txg_how {
 222         TXG_WAIT = 1,
 223         TXG_NOWAIT,
 224 } txg_how_t;
 225 
 226 void byteswap_uint64_array(void *buf, size_t size);
 227 void byteswap_uint32_array(void *buf, size_t size);
 228 void byteswap_uint16_array(void *buf, size_t size);
 229 void byteswap_uint8_array(void *buf, size_t size);
 230 void zap_byteswap(void *buf, size_t size);
 231 void zfs_oldacl_byteswap(void *buf, size_t size);
 232 void zfs_acl_byteswap(void *buf, size_t size);
 233 void zfs_znode_byteswap(void *buf, size_t size);
 234 
 235 #define DS_FIND_SNAPSHOTS       (1<<0)
 236 #define DS_FIND_CHILDREN        (1<<1)
 237 
 238 /*
 239  * The maximum number of bytes that can be accessed as part of one
 240  * operation, including metadata.
 241  */
 242 #define DMU_MAX_ACCESS (10<<20) /* 10MB */
 243 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
 244 
 245 #define DMU_USERUSED_OBJECT     (-1ULL)
 246 #define DMU_GROUPUSED_OBJECT    (-2ULL)
 247 #define DMU_DEADLIST_OBJECT     (-3ULL)
 248 
 249 /*
 250  * artificial blkids for bonus buffer and spill blocks
 251  */
 252 #define DMU_BONUS_BLKID         (-1ULL)
 253 #define DMU_SPILL_BLKID         (-2ULL)
 254 /*
 255  * Public routines to create, destroy, open, and close objsets.
 256  */
 257 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
 258 int dmu_objset_own(const char *name, dmu_objset_type_t type,
 259     boolean_t readonly, void *tag, objset_t **osp);
 260 void dmu_objset_rele(objset_t *os, void *tag);
 261 void dmu_objset_disown(objset_t *os, void *tag);
 262 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
 263 
 264 void dmu_objset_evict_dbufs(objset_t *os);
 265 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
 266     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
 267 int dmu_objset_clone(const char *name, const char *origin);
 268 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
 269     struct nvlist *errlist);
 270 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
 271 int dmu_objset_snapshot_tmp(const char *, const char *, int);
 272 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
 273     int flags);
 274 void dmu_objset_byteswap(void *buf, size_t size);
 275 int dsl_dataset_rename_snapshot(const char *fsname,
 276     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
 277 
 278 typedef struct dmu_buf {
 279         uint64_t db_object;             /* object that this buffer is part of */
 280         uint64_t db_offset;             /* byte offset in this object */
 281         uint64_t db_size;               /* size of buffer in bytes */
 282         void *db_data;                  /* data in buffer */
 283 } dmu_buf_t;
 284 
 285 /*
 286  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
 287  */
 288 #define DMU_POOL_DIRECTORY_OBJECT       1
 289 #define DMU_POOL_CONFIG                 "config"
 290 #define DMU_POOL_FEATURES_FOR_WRITE     "features_for_write"
 291 #define DMU_POOL_FEATURES_FOR_READ      "features_for_read"
 292 #define DMU_POOL_FEATURE_DESCRIPTIONS   "feature_descriptions"
 293 #define DMU_POOL_ROOT_DATASET           "root_dataset"
 294 #define DMU_POOL_SYNC_BPOBJ             "sync_bplist"
 295 #define DMU_POOL_ERRLOG_SCRUB           "errlog_scrub"
 296 #define DMU_POOL_ERRLOG_LAST            "errlog_last"
 297 #define DMU_POOL_SPARES                 "spares"
 298 #define DMU_POOL_DEFLATE                "deflate"
 299 #define DMU_POOL_HISTORY                "history"
 300 #define DMU_POOL_PROPS                  "pool_props"
 301 #define DMU_POOL_L2CACHE                "l2cache"
 302 #define DMU_POOL_TMP_USERREFS           "tmp_userrefs"
 303 #define DMU_POOL_DDT                    "DDT-%s-%s-%s"
 304 #define DMU_POOL_DDT_STATS              "DDT-statistics"
 305 #define DMU_POOL_CREATION_VERSION       "creation_version"
 306 #define DMU_POOL_SCAN                   "scan"
 307 #define DMU_POOL_FREE_BPOBJ             "free_bpobj"
 308 #define DMU_POOL_BPTREE_OBJ             "bptree_obj"
 309 #define DMU_POOL_EMPTY_BPOBJ            "empty_bpobj"
 310 
 311 /*
 312  * Allocate an object from this objset.  The range of object numbers
 313  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
 314  *
 315  * The transaction must be assigned to a txg.  The newly allocated
 316  * object will be "held" in the transaction (ie. you can modify the
 317  * newly allocated object in this transaction).
 318  *
 319  * dmu_object_alloc() chooses an object and returns it in *objectp.
 320  *
 321  * dmu_object_claim() allocates a specific object number.  If that
 322  * number is already allocated, it fails and returns EEXIST.
 323  *
 324  * Return 0 on success, or ENOSPC or EEXIST as specified above.
 325  */
 326 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
 327     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
 328 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
 329     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
 330 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
 331     int blocksize, dmu_object_type_t bonustype, int bonuslen);
 332 
 333 /*
 334  * Free an object from this objset.
 335  *
 336  * The object's data will be freed as well (ie. you don't need to call
 337  * dmu_free(object, 0, -1, tx)).
 338  *
 339  * The object need not be held in the transaction.
 340  *
 341  * If there are any holds on this object's buffers (via dmu_buf_hold()),
 342  * or tx holds on the object (via dmu_tx_hold_object()), you can not
 343  * free it; it fails and returns EBUSY.
 344  *
 345  * If the object is not allocated, it fails and returns ENOENT.
 346  *
 347  * Return 0 on success, or EBUSY or ENOENT as specified above.
 348  */
 349 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
 350 
 351 /*
 352  * Find the next allocated or free object.
 353  *
 354  * The objectp parameter is in-out.  It will be updated to be the next
 355  * object which is allocated.  Ignore objects which have not been
 356  * modified since txg.
 357  *
 358  * XXX Can only be called on a objset with no dirty data.
 359  *
 360  * Returns 0 on success, or ENOENT if there are no more objects.
 361  */
 362 int dmu_object_next(objset_t *os, uint64_t *objectp,
 363     boolean_t hole, uint64_t txg);
 364 
 365 /*
 366  * Set the data blocksize for an object.
 367  *
 368  * The object cannot have any blocks allcated beyond the first.  If
 369  * the first block is allocated already, the new size must be greater
 370  * than the current block size.  If these conditions are not met,
 371  * ENOTSUP will be returned.
 372  *
 373  * Returns 0 on success, or EBUSY if there are any holds on the object
 374  * contents, or ENOTSUP as described above.
 375  */
 376 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
 377     int ibs, dmu_tx_t *tx);
 378 
 379 /*
 380  * Set the checksum property on a dnode.  The new checksum algorithm will
 381  * apply to all newly written blocks; existing blocks will not be affected.
 382  */
 383 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
 384     dmu_tx_t *tx);
 385 
 386 /*
 387  * Set the compress property on a dnode.  The new compression algorithm will
 388  * apply to all newly written blocks; existing blocks will not be affected.
 389  */
 390 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
 391     dmu_tx_t *tx);
 392 
 393 /*
 394  * Decide how to write a block: checksum, compression, number of copies, etc.
 395  */
 396 #define WP_NOFILL       0x1
 397 #define WP_DMU_SYNC     0x2
 398 #define WP_SPILL        0x4
 399 
 400 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
 401     struct zio_prop *zp);
 402 /*
 403  * The bonus data is accessed more or less like a regular buffer.
 404  * You must dmu_bonus_hold() to get the buffer, which will give you a
 405  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
 406  * data.  As with any normal buffer, you must call dmu_buf_read() to
 407  * read db_data, dmu_buf_will_dirty() before modifying it, and the
 408  * object must be held in an assigned transaction before calling
 409  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
 410  * buffer as well.  You must release your hold with dmu_buf_rele().
 411  */
 412 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
 413 int dmu_bonus_max(void);
 414 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
 415 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
 416 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
 417 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
 418 
 419 /*
 420  * Special spill buffer support used by "SA" framework
 421  */
 422 
 423 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
 424 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
 425     void *tag, dmu_buf_t **dbp);
 426 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
 427 
 428 /*
 429  * Obtain the DMU buffer from the specified object which contains the
 430  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
 431  * that it will remain in memory.  You must release the hold with
 432  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
 433  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
 434  *
 435  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
 436  * on the returned buffer before reading or writing the buffer's
 437  * db_data.  The comments for those routines describe what particular
 438  * operations are valid after calling them.
 439  *
 440  * The object number must be a valid, allocated object number.
 441  */
 442 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
 443     void *tag, dmu_buf_t **, int flags);
 444 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
 445 void dmu_buf_rele(dmu_buf_t *db, void *tag);
 446 uint64_t dmu_buf_refcount(dmu_buf_t *db);
 447 
 448 /*
 449  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
 450  * range of an object.  A pointer to an array of dmu_buf_t*'s is
 451  * returned (in *dbpp).
 452  *
 453  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
 454  * frees the array.  The hold on the array of buffers MUST be released
 455  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
 456  * individually with dmu_buf_rele.
 457  */
 458 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
 459     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
 460 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
 461 
 462 struct dmu_buf_user;
 463 
 464 typedef void dmu_buf_evict_func_t(struct dmu_buf_user *);
 465 
 466 /*
 467  * The DMU buffer user object is used to allow private data to be
 468  * associated with a dbuf for the duration of its lifetime.  This private
 469  * data must include a dmu_buf_user_t as its first object, which is passed
 470  * into the DMU user data API and can be attached to a dbuf.  Clients can
 471  * regain access to their private data structure with a cast.
 472  *
 473  * DMU buffer users can be notified via a callback when their associated
 474  * dbuf has been evicted.  This is typically used to free the user's
 475  * private data.  The eviction callback is executed without the dbuf
 476  * mutex held or any other type of mechanism to guarantee that the
 477  * dbuf is still available.  For this reason, users must assume the dbuf
 478  * has already been freed and not reference the dbuf from the callback
 479  * context.
 480  *
 481  * Users requestion "immediate eviction" are notified as soon as the dbuf
 482  * is only referenced by dirty records (dirties == holds).  Otherwise the
 483  * eviction callback occurs after the last reference to the dbuf is dropped.    
 484  *
 485  * Eviction Callback Processing
 486  * ============================
 487  * In any context where a dbuf reference drop may trigger an eviction, an       
 488  * eviction queue object must be provided.  This queue must then be
 489  * processed while not holding any dbuf locks.  In this way, the user can       
 490  * perform any work needed in their eviction function without fear of
 491  * lock order reversals.
 492  *
 493  * Implementation Note
 494  * ============================
 495  * Some users will occasionally want to map a structure directly onto the
 496  * backing dbuf.  Using an union with an name alias macro to access these
 497  * overlays reduces the ugliness of code that accesses them.  Initial work on
 498  * user objects involved using a macro that took the user object as an
 499  * argument to access the fields, which resulted in hundreds of lines of
 500  * needless diffs and wasn't any easier to read.
 501  */
 502 typedef struct dmu_buf_user {
 503         /*
 504          * This instance's link in the eviction queue.  Set when the buffer
 505          * has evicted and the callback needs to be called.
 506          */
 507         list_node_t evict_queue_link;
 508         /** This instance's eviction function pointer. */
 509         dmu_buf_evict_func_t *evict_func;
 510 } dmu_buf_user_t;
 511 
 512 /*
 513  * Initialize the given dmu_buf_user_t instance with the eviction function
 514  * evict_func, to be called when the user is evicted.
 515  *
 516  * NOTE: This function should only be called once on a given object.  To
 517  *       help enforce this, dbu should already be zeroed on entry.
 518  */
 519 static inline void
 520 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func)
 521 {
 522         ASSERT(dbu->evict_func == NULL);
 523         ASSERT(!list_link_active(&dbu->evict_queue_link));
 524         dbu->evict_func = evict_func;
 525 }
 526 
 527 static inline void
 528 dmu_buf_create_user_evict_list(list_t *evict_list_p)
 529 {
 530         list_create(evict_list_p, sizeof(dmu_buf_user_t),
 531             offsetof(dmu_buf_user_t, evict_queue_link));
 532 }
 533 
 534 static inline void
 535 dmu_buf_process_user_evicts(list_t *evict_list_p)
 536 {
 537         dmu_buf_user_t *dbu, *next;
 538 
 539         for (dbu = (dmu_buf_user_t *)list_head(evict_list_p); dbu != NULL;
 540             dbu = next) {
 541                 next = (dmu_buf_user_t *)list_next(evict_list_p, dbu);
 542                 list_remove(evict_list_p, dbu);
 543                 dbu->evict_func(dbu);
 544         }
 545 }
 546 
 547 static inline void
 548 dmu_buf_destroy_user_evict_list(list_t *evict_list_p)
 549 {
 550         dmu_buf_process_user_evicts(evict_list_p);
 551         list_destroy(evict_list_p);
 552 }
 553 
 554 dmu_buf_user_t *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
 555 dmu_buf_user_t *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
 556 dmu_buf_user_t *dmu_buf_replace_user(dmu_buf_t *db,
 557     dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
 558 dmu_buf_user_t *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
 559 dmu_buf_user_t *dmu_buf_get_user(dmu_buf_t *db);
 560 
 561 /*
 562  * Returns the blkptr associated with this dbuf, or NULL if not set.
 563  */
 564 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
 565 
 566 /*
 567  * Indicate that you are going to modify the buffer's data (db_data).
 568  *
 569  * The transaction (tx) must be assigned to a txg (ie. you've called
 570  * dmu_tx_assign()).  The buffer's object must be held in the tx
 571  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
 572  */
 573 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
 574 
 575 /*
 576  * Tells if the given dbuf is freeable.
 577  */
 578 boolean_t dmu_buf_freeable(dmu_buf_t *);
 579 
 580 /*
 581  * You must create a transaction, then hold the objects which you will
 582  * (or might) modify as part of this transaction.  Then you must assign
 583  * the transaction to a transaction group.  Once the transaction has
 584  * been assigned, you can modify buffers which belong to held objects as
 585  * part of this transaction.  You can't modify buffers before the
 586  * transaction has been assigned; you can't modify buffers which don't
 587  * belong to objects which this transaction holds; you can't hold
 588  * objects once the transaction has been assigned.  You may hold an
 589  * object which you are going to free (with dmu_object_free()), but you
 590  * don't have to.
 591  *
 592  * You can abort the transaction before it has been assigned.
 593  *
 594  * Note that you may hold buffers (with dmu_buf_hold) at any time,
 595  * regardless of transaction state.
 596  */
 597 
 598 #define DMU_NEW_OBJECT  (-1ULL)
 599 #define DMU_OBJECT_END  (-1ULL)
 600 
 601 dmu_tx_t *dmu_tx_create(objset_t *os);
 602 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
 603 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
 604     uint64_t len);
 605 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
 606 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
 607 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
 608 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
 609 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
 610 void dmu_tx_abort(dmu_tx_t *tx);
 611 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
 612 void dmu_tx_wait(dmu_tx_t *tx);
 613 void dmu_tx_commit(dmu_tx_t *tx);
 614 
 615 /*
 616  * To register a commit callback, dmu_tx_callback_register() must be called.
 617  *
 618  * dcb_data is a pointer to caller private data that is passed on as a
 619  * callback parameter. The caller is responsible for properly allocating and
 620  * freeing it.
 621  *
 622  * When registering a callback, the transaction must be already created, but
 623  * it cannot be committed or aborted. It can be assigned to a txg or not.
 624  *
 625  * The callback will be called after the transaction has been safely written
 626  * to stable storage and will also be called if the dmu_tx is aborted.
 627  * If there is any error which prevents the transaction from being committed to
 628  * disk, the callback will be called with a value of error != 0.
 629  */
 630 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
 631 
 632 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
 633     void *dcb_data);
 634 
 635 /*
 636  * Free up the data blocks for a defined range of a file.  If size is
 637  * -1, the range from offset to end-of-file is freed.
 638  */
 639 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
 640         uint64_t size, dmu_tx_t *tx);
 641 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
 642         uint64_t size);
 643 int dmu_free_object(objset_t *os, uint64_t object);
 644 
 645 /*
 646  * Convenience functions.
 647  *
 648  * Canfail routines will return 0 on success, or an errno if there is a
 649  * nonrecoverable I/O error.
 650  */
 651 #define DMU_READ_PREFETCH       0 /* prefetch */
 652 #define DMU_READ_NO_PREFETCH    1 /* don't prefetch */
 653 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 654         void *buf, uint32_t flags);
 655 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 656         const void *buf, dmu_tx_t *tx);
 657 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 658         dmu_tx_t *tx);
 659 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
 660 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
 661     dmu_tx_t *tx);
 662 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
 663     dmu_tx_t *tx);
 664 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
 665     uint64_t size, struct page *pp, dmu_tx_t *tx);
 666 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
 667 void dmu_return_arcbuf(struct arc_buf *buf);
 668 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
 669     dmu_tx_t *tx);
 670 int dmu_xuio_init(struct xuio *uio, int niov);
 671 void dmu_xuio_fini(struct xuio *uio);
 672 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
 673     size_t n);
 674 int dmu_xuio_cnt(struct xuio *uio);
 675 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
 676 void dmu_xuio_clear(struct xuio *uio, int i);
 677 void xuio_stat_wbuf_copied();
 678 void xuio_stat_wbuf_nocopy();
 679 
 680 extern int zfs_prefetch_disable;
 681 
 682 /*
 683  * Asynchronously try to read in the data.
 684  */
 685 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
 686     uint64_t len);
 687 
 688 typedef struct dmu_object_info {
 689         /* All sizes are in bytes unless otherwise indicated. */
 690         uint32_t doi_data_block_size;
 691         uint32_t doi_metadata_block_size;
 692         dmu_object_type_t doi_type;
 693         dmu_object_type_t doi_bonus_type;
 694         uint64_t doi_bonus_size;
 695         uint8_t doi_indirection;                /* 2 = dnode->indirect->data */
 696         uint8_t doi_checksum;
 697         uint8_t doi_compress;
 698         uint8_t doi_pad[5];
 699         uint64_t doi_physical_blocks_512;       /* data + metadata, 512b blks */
 700         uint64_t doi_max_offset;
 701         uint64_t doi_fill_count;                /* number of non-empty blocks */
 702 } dmu_object_info_t;
 703 
 704 typedef void arc_byteswap_func_t(void *buf, size_t size);
 705 
 706 typedef struct dmu_object_type_info {
 707         dmu_object_byteswap_t   ot_byteswap;
 708         boolean_t               ot_metadata;
 709         char                    *ot_name;
 710 } dmu_object_type_info_t;
 711 
 712 typedef struct dmu_object_byteswap_info {
 713         arc_byteswap_func_t     *ob_func;
 714         char                    *ob_name;
 715 } dmu_object_byteswap_info_t;
 716 
 717 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
 718 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
 719 
 720 /*
 721  * Get information on a DMU object.
 722  *
 723  * Return 0 on success or ENOENT if object is not allocated.
 724  *
 725  * If doi is NULL, just indicates whether the object exists.
 726  */
 727 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
 728 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
 729 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
 730 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
 731     u_longlong_t *nblk512);
 732 
 733 typedef struct dmu_objset_stats {
 734         uint64_t dds_num_clones; /* number of clones of this */
 735         uint64_t dds_creation_txg;
 736         uint64_t dds_guid;
 737         dmu_objset_type_t dds_type;
 738         uint8_t dds_is_snapshot;
 739         uint8_t dds_inconsistent;
 740         char dds_origin[MAXNAMELEN];
 741 } dmu_objset_stats_t;
 742 
 743 /*
 744  * Get stats on a dataset.
 745  */
 746 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
 747 
 748 /*
 749  * Add entries to the nvlist for all the objset's properties.  See
 750  * zfs_prop_table[] and zfs(1m) for details on the properties.
 751  */
 752 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
 753 
 754 /*
 755  * Get the space usage statistics for statvfs().
 756  *
 757  * refdbytes is the amount of space "referenced" by this objset.
 758  * availbytes is the amount of space available to this objset, taking
 759  * into account quotas & reservations, assuming that no other objsets
 760  * use the space first.  These values correspond to the 'referenced' and
 761  * 'available' properties, described in the zfs(1m) manpage.
 762  *
 763  * usedobjs and availobjs are the number of objects currently allocated,
 764  * and available.
 765  */
 766 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
 767     uint64_t *usedobjsp, uint64_t *availobjsp);
 768 
 769 /*
 770  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
 771  * (Contrast with the ds_guid which is a 64-bit ID that will never
 772  * change, so there is a small probability that it will collide.)
 773  */
 774 uint64_t dmu_objset_fsid_guid(objset_t *os);
 775 
 776 /*
 777  * Get the [cm]time for an objset's snapshot dir
 778  */
 779 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
 780 
 781 int dmu_objset_is_snapshot(objset_t *os);
 782 
 783 extern struct spa *dmu_objset_spa(objset_t *os);
 784 extern struct zilog *dmu_objset_zil(objset_t *os);
 785 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
 786 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
 787 extern void dmu_objset_name(objset_t *os, char *buf);
 788 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
 789 extern uint64_t dmu_objset_id(objset_t *os);
 790 extern uint64_t dmu_objset_syncprop(objset_t *os);
 791 extern uint64_t dmu_objset_logbias(objset_t *os);
 792 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
 793     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
 794 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
 795     int maxlen, boolean_t *conflict);
 796 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
 797     uint64_t *idp, uint64_t *offp);
 798 
 799 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
 800     void *bonus, uint64_t *userp, uint64_t *groupp);
 801 extern void dmu_objset_register_type(dmu_objset_type_t ost,
 802     objset_used_cb_t *cb);
 803 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
 804 extern void *dmu_objset_get_user(objset_t *os);
 805 
 806 /*
 807  * Return the txg number for the given assigned transaction.
 808  */
 809 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
 810 
 811 /*
 812  * Synchronous write.
 813  * If a parent zio is provided this function initiates a write on the
 814  * provided buffer as a child of the parent zio.
 815  * In the absence of a parent zio, the write is completed synchronously.
 816  * At write completion, blk is filled with the bp of the written block.
 817  * Note that while the data covered by this function will be on stable
 818  * storage when the write completes this new data does not become a
 819  * permanent part of the file until the associated transaction commits.
 820  */
 821 
 822 /*
 823  * {zfs,zvol,ztest}_get_done() args
 824  */
 825 typedef struct zgd {
 826         struct zilog    *zgd_zilog;
 827         struct blkptr   *zgd_bp;
 828         dmu_buf_t       *zgd_db;
 829         struct rl       *zgd_rl;
 830         void            *zgd_private;
 831 } zgd_t;
 832 
 833 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
 834 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
 835 
 836 /*
 837  * Find the next hole or data block in file starting at *off
 838  * Return found offset in *off. Return ESRCH for end of file.
 839  */
 840 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
 841     uint64_t *off);
 842 
 843 /*
 844  * Initial setup and final teardown.
 845  */
 846 extern void dmu_init(void);
 847 extern void dmu_fini(void);
 848 
 849 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
 850     uint64_t object, uint64_t offset, int len);
 851 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
 852     dmu_traverse_cb_t cb, void *arg);
 853 
 854 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
 855     struct vnode *vp, offset_t *offp);
 856 
 857 /* CRC64 table */
 858 #define ZFS_CRC64_POLY  0xC96C5795D7870F42ULL   /* ECMA-182, reflected form */
 859 extern uint64_t zfs_crc64_table[256];
 860 
 861 #ifdef  __cplusplus
 862 }
 863 #endif
 864 
 865 #endif  /* _SYS_DMU_H */