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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 * Copyright 2014 HybridCluster. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30 /* Portions Copyright 2010 Robert Milkowski */
36 * This file describes the interface that the DMU provides for its
39 * The DMU also interacts with the SPA. That interface is described in
43 #include <sys/zfs_context.h>
44 #include <sys/inttypes.h>
46 #include <sys/fs/zfs.h>
47 #include <sys/zio_priority.h>
66 struct zbookmark_phys
;
73 typedef struct objset objset_t
;
74 typedef struct dmu_tx dmu_tx_t
;
75 typedef struct dsl_dir dsl_dir_t
;
77 typedef enum dmu_object_byteswap
{
89 * Allocating a new byteswap type number makes the on-disk format
90 * incompatible with any other format that uses the same number.
92 * Data can usually be structured to work with one of the
93 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
96 } dmu_object_byteswap_t
;
98 #define DMU_OT_NEWTYPE 0x80
99 #define DMU_OT_METADATA 0x40
100 #define DMU_OT_BYTESWAP_MASK 0x3f
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).
107 #define DMU_OT(byteswap, metadata) \
109 ((metadata) ? DMU_OT_METADATA : 0) | \
110 ((byteswap) & DMU_OT_BYTESWAP_MASK))
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)
116 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
117 ((ot) & DMU_OT_METADATA) : \
118 dmu_ot[(int)(ot)].ot_metadata)
121 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
122 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
123 * is repurposed for embedded BPs.
125 #define DMU_OT_HAS_FILL(ot) \
126 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
128 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
129 ((ot) & DMU_OT_BYTESWAP_MASK) : \
130 dmu_ot[(int)(ot)].ot_byteswap)
132 typedef enum dmu_object_type
{
135 DMU_OT_OBJECT_DIRECTORY
, /* ZAP */
136 DMU_OT_OBJECT_ARRAY
, /* UINT64 */
137 DMU_OT_PACKED_NVLIST
, /* UINT8 (XDR by nvlist_pack/unpack) */
138 DMU_OT_PACKED_NVLIST_SIZE
, /* UINT64 */
139 DMU_OT_BPOBJ
, /* UINT64 */
140 DMU_OT_BPOBJ_HDR
, /* UINT64 */
142 DMU_OT_SPACE_MAP_HEADER
, /* UINT64 */
143 DMU_OT_SPACE_MAP
, /* UINT64 */
145 DMU_OT_INTENT_LOG
, /* UINT64 */
147 DMU_OT_DNODE
, /* DNODE */
148 DMU_OT_OBJSET
, /* OBJSET */
150 DMU_OT_DSL_DIR
, /* UINT64 */
151 DMU_OT_DSL_DIR_CHILD_MAP
, /* ZAP */
152 DMU_OT_DSL_DS_SNAP_MAP
, /* ZAP */
153 DMU_OT_DSL_PROPS
, /* ZAP */
154 DMU_OT_DSL_DATASET
, /* UINT64 */
156 DMU_OT_ZNODE
, /* ZNODE */
157 DMU_OT_OLDACL
, /* Old ACL */
158 DMU_OT_PLAIN_FILE_CONTENTS
, /* UINT8 */
159 DMU_OT_DIRECTORY_CONTENTS
, /* ZAP */
160 DMU_OT_MASTER_NODE
, /* ZAP */
161 DMU_OT_UNLINKED_SET
, /* ZAP */
163 DMU_OT_ZVOL
, /* UINT8 */
164 DMU_OT_ZVOL_PROP
, /* ZAP */
165 /* other; for testing only! */
166 DMU_OT_PLAIN_OTHER
, /* UINT8 */
167 DMU_OT_UINT64_OTHER
, /* UINT64 */
168 DMU_OT_ZAP_OTHER
, /* ZAP */
169 /* new object types: */
170 DMU_OT_ERROR_LOG
, /* ZAP */
171 DMU_OT_SPA_HISTORY
, /* UINT8 */
172 DMU_OT_SPA_HISTORY_OFFSETS
, /* spa_his_phys_t */
173 DMU_OT_POOL_PROPS
, /* ZAP */
174 DMU_OT_DSL_PERMS
, /* ZAP */
175 DMU_OT_ACL
, /* ACL */
176 DMU_OT_SYSACL
, /* SYSACL */
177 DMU_OT_FUID
, /* FUID table (Packed NVLIST UINT8) */
178 DMU_OT_FUID_SIZE
, /* FUID table size UINT64 */
179 DMU_OT_NEXT_CLONES
, /* ZAP */
180 DMU_OT_SCAN_QUEUE
, /* ZAP */
181 DMU_OT_USERGROUP_USED
, /* ZAP */
182 DMU_OT_USERGROUP_QUOTA
, /* ZAP */
183 DMU_OT_USERREFS
, /* ZAP */
184 DMU_OT_DDT_ZAP
, /* ZAP */
185 DMU_OT_DDT_STATS
, /* ZAP */
186 DMU_OT_SA
, /* System attr */
187 DMU_OT_SA_MASTER_NODE
, /* ZAP */
188 DMU_OT_SA_ATTR_REGISTRATION
, /* ZAP */
189 DMU_OT_SA_ATTR_LAYOUTS
, /* ZAP */
190 DMU_OT_SCAN_XLATE
, /* ZAP */
191 DMU_OT_DEDUP
, /* fake dedup BP from ddt_bp_create() */
192 DMU_OT_DEADLIST
, /* ZAP */
193 DMU_OT_DEADLIST_HDR
, /* UINT64 */
194 DMU_OT_DSL_CLONES
, /* ZAP */
195 DMU_OT_BPOBJ_SUBOBJ
, /* UINT64 */
197 * Do not allocate new object types here. Doing so makes the on-disk
198 * format incompatible with any other format that uses the same object
201 * When creating an object which does not have one of the above types
202 * use the DMU_OTN_* type with the correct byteswap and metadata
205 * The DMU_OTN_* types do not have entries in the dmu_ot table,
206 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
207 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
208 * and DMU_OTN_* types).
213 * Names for valid types declared with DMU_OT().
215 DMU_OTN_UINT8_DATA
= DMU_OT(DMU_BSWAP_UINT8
, B_FALSE
),
216 DMU_OTN_UINT8_METADATA
= DMU_OT(DMU_BSWAP_UINT8
, B_TRUE
),
217 DMU_OTN_UINT16_DATA
= DMU_OT(DMU_BSWAP_UINT16
, B_FALSE
),
218 DMU_OTN_UINT16_METADATA
= DMU_OT(DMU_BSWAP_UINT16
, B_TRUE
),
219 DMU_OTN_UINT32_DATA
= DMU_OT(DMU_BSWAP_UINT32
, B_FALSE
),
220 DMU_OTN_UINT32_METADATA
= DMU_OT(DMU_BSWAP_UINT32
, B_TRUE
),
221 DMU_OTN_UINT64_DATA
= DMU_OT(DMU_BSWAP_UINT64
, B_FALSE
),
222 DMU_OTN_UINT64_METADATA
= DMU_OT(DMU_BSWAP_UINT64
, B_TRUE
),
223 DMU_OTN_ZAP_DATA
= DMU_OT(DMU_BSWAP_ZAP
, B_FALSE
),
224 DMU_OTN_ZAP_METADATA
= DMU_OT(DMU_BSWAP_ZAP
, B_TRUE
),
227 typedef enum txg_how
{
233 void byteswap_uint64_array(void *buf
, size_t size
);
234 void byteswap_uint32_array(void *buf
, size_t size
);
235 void byteswap_uint16_array(void *buf
, size_t size
);
236 void byteswap_uint8_array(void *buf
, size_t size
);
237 void zap_byteswap(void *buf
, size_t size
);
238 void zfs_oldacl_byteswap(void *buf
, size_t size
);
239 void zfs_acl_byteswap(void *buf
, size_t size
);
240 void zfs_znode_byteswap(void *buf
, size_t size
);
242 #define DS_FIND_SNAPSHOTS (1<<0)
243 #define DS_FIND_CHILDREN (1<<1)
244 #define DS_FIND_SERIALIZE (1<<2)
247 * The maximum number of bytes that can be accessed as part of one
248 * operation, including metadata.
250 #define DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
251 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
253 #define DMU_USERUSED_OBJECT (-1ULL)
254 #define DMU_GROUPUSED_OBJECT (-2ULL)
257 * artificial blkids for bonus buffer and spill blocks
259 #define DMU_BONUS_BLKID (-1ULL)
260 #define DMU_SPILL_BLKID (-2ULL)
262 * Public routines to create, destroy, open, and close objsets.
264 int dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
);
265 int dmu_objset_own(const char *name
, dmu_objset_type_t type
,
266 boolean_t readonly
, void *tag
, objset_t
**osp
);
267 void dmu_objset_rele(objset_t
*os
, void *tag
);
268 void dmu_objset_disown(objset_t
*os
, void *tag
);
269 int dmu_objset_open_ds(struct dsl_dataset
*ds
, objset_t
**osp
);
271 void dmu_objset_evict_dbufs(objset_t
*os
);
272 int dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
273 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
);
274 int dmu_objset_clone(const char *name
, const char *origin
);
275 int dsl_destroy_snapshots_nvl(struct nvlist
*snaps
, boolean_t defer
,
276 struct nvlist
*errlist
);
277 int dmu_objset_snapshot_one(const char *fsname
, const char *snapname
);
278 int dmu_objset_snapshot_tmp(const char *, const char *, int);
279 int dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
281 void dmu_objset_byteswap(void *buf
, size_t size
);
282 int dsl_dataset_rename_snapshot(const char *fsname
,
283 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
);
285 typedef struct dmu_buf
{
286 uint64_t db_object
; /* object that this buffer is part of */
287 uint64_t db_offset
; /* byte offset in this object */
288 uint64_t db_size
; /* size of buffer in bytes */
289 void *db_data
; /* data in buffer */
293 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
295 #define DMU_POOL_DIRECTORY_OBJECT 1
296 #define DMU_POOL_CONFIG "config"
297 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
298 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
299 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
300 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
301 #define DMU_POOL_ROOT_DATASET "root_dataset"
302 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
303 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
304 #define DMU_POOL_ERRLOG_LAST "errlog_last"
305 #define DMU_POOL_SPARES "spares"
306 #define DMU_POOL_DEFLATE "deflate"
307 #define DMU_POOL_HISTORY "history"
308 #define DMU_POOL_PROPS "pool_props"
309 #define DMU_POOL_L2CACHE "l2cache"
310 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
311 #define DMU_POOL_DDT "DDT-%s-%s-%s"
312 #define DMU_POOL_DDT_STATS "DDT-statistics"
313 #define DMU_POOL_CREATION_VERSION "creation_version"
314 #define DMU_POOL_SCAN "scan"
315 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
316 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
317 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
320 * Allocate an object from this objset. The range of object numbers
321 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
323 * The transaction must be assigned to a txg. The newly allocated
324 * object will be "held" in the transaction (ie. you can modify the
325 * newly allocated object in this transaction).
327 * dmu_object_alloc() chooses an object and returns it in *objectp.
329 * dmu_object_claim() allocates a specific object number. If that
330 * number is already allocated, it fails and returns EEXIST.
332 * Return 0 on success, or ENOSPC or EEXIST as specified above.
334 uint64_t dmu_object_alloc(objset_t
*os
, dmu_object_type_t ot
,
335 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
336 int dmu_object_claim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
337 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
338 int dmu_object_reclaim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
339 int blocksize
, dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*txp
);
342 * Free an object from this objset.
344 * The object's data will be freed as well (ie. you don't need to call
345 * dmu_free(object, 0, -1, tx)).
347 * The object need not be held in the transaction.
349 * If there are any holds on this object's buffers (via dmu_buf_hold()),
350 * or tx holds on the object (via dmu_tx_hold_object()), you can not
351 * free it; it fails and returns EBUSY.
353 * If the object is not allocated, it fails and returns ENOENT.
355 * Return 0 on success, or EBUSY or ENOENT as specified above.
357 int dmu_object_free(objset_t
*os
, uint64_t object
, dmu_tx_t
*tx
);
360 * Find the next allocated or free object.
362 * The objectp parameter is in-out. It will be updated to be the next
363 * object which is allocated. Ignore objects which have not been
364 * modified since txg.
366 * XXX Can only be called on a objset with no dirty data.
368 * Returns 0 on success, or ENOENT if there are no more objects.
370 int dmu_object_next(objset_t
*os
, uint64_t *objectp
,
371 boolean_t hole
, uint64_t txg
);
374 * Set the data blocksize for an object.
376 * The object cannot have any blocks allcated beyond the first. If
377 * the first block is allocated already, the new size must be greater
378 * than the current block size. If these conditions are not met,
379 * ENOTSUP will be returned.
381 * Returns 0 on success, or EBUSY if there are any holds on the object
382 * contents, or ENOTSUP as described above.
384 int dmu_object_set_blocksize(objset_t
*os
, uint64_t object
, uint64_t size
,
385 int ibs
, dmu_tx_t
*tx
);
388 * Set the checksum property on a dnode. The new checksum algorithm will
389 * apply to all newly written blocks; existing blocks will not be affected.
391 void dmu_object_set_checksum(objset_t
*os
, uint64_t object
, uint8_t checksum
,
395 * Set the compress property on a dnode. The new compression algorithm will
396 * apply to all newly written blocks; existing blocks will not be affected.
398 void dmu_object_set_compress(objset_t
*os
, uint64_t object
, uint8_t compress
,
402 dmu_write_embedded(objset_t
*os
, uint64_t object
, uint64_t offset
,
403 void *data
, uint8_t etype
, uint8_t comp
, int uncompressed_size
,
404 int compressed_size
, int byteorder
, dmu_tx_t
*tx
);
407 * Decide how to write a block: checksum, compression, number of copies, etc.
409 #define WP_NOFILL 0x1
410 #define WP_DMU_SYNC 0x2
413 void dmu_write_policy(objset_t
*os
, struct dnode
*dn
, int level
, int wp
,
414 struct zio_prop
*zp
);
416 * The bonus data is accessed more or less like a regular buffer.
417 * You must dmu_bonus_hold() to get the buffer, which will give you a
418 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
419 * data. As with any normal buffer, you must call dmu_buf_read() to
420 * read db_data, dmu_buf_will_dirty() before modifying it, and the
421 * object must be held in an assigned transaction before calling
422 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
423 * buffer as well. You must release what you hold with dmu_buf_rele().
425 * Returns ENOENT, EIO, or 0.
427 int dmu_bonus_hold(objset_t
*os
, uint64_t object
, void *tag
, dmu_buf_t
**);
428 int dmu_bonus_max(void);
429 int dmu_set_bonus(dmu_buf_t
*, int, dmu_tx_t
*);
430 int dmu_set_bonustype(dmu_buf_t
*, dmu_object_type_t
, dmu_tx_t
*);
431 dmu_object_type_t
dmu_get_bonustype(dmu_buf_t
*);
432 int dmu_rm_spill(objset_t
*, uint64_t, dmu_tx_t
*);
435 * Special spill buffer support used by "SA" framework
438 int dmu_spill_hold_by_bonus(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
);
439 int dmu_spill_hold_by_dnode(struct dnode
*dn
, uint32_t flags
,
440 void *tag
, dmu_buf_t
**dbp
);
441 int dmu_spill_hold_existing(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
);
444 * Obtain the DMU buffer from the specified object which contains the
445 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
446 * that it will remain in memory. You must release the hold with
447 * dmu_buf_rele(). You must not access the dmu_buf_t after releasing
448 * what you hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
450 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
451 * on the returned buffer before reading or writing the buffer's
452 * db_data. The comments for those routines describe what particular
453 * operations are valid after calling them.
455 * The object number must be a valid, allocated object number.
457 int dmu_buf_hold(objset_t
*os
, uint64_t object
, uint64_t offset
,
458 void *tag
, dmu_buf_t
**, int flags
);
461 * Add a reference to a dmu buffer that has already been held via
462 * dmu_buf_hold() in the current context.
464 void dmu_buf_add_ref(dmu_buf_t
*db
, void* tag
);
467 * Attempt to add a reference to a dmu buffer that is in an unknown state,
468 * using a pointer that may have been invalidated by eviction processing.
469 * The request will succeed if the passed in dbuf still represents the
470 * same os/object/blkid, is ineligible for eviction, and has at least
471 * one hold by a user other than the syncer.
473 boolean_t
dmu_buf_try_add_ref(dmu_buf_t
*, objset_t
*os
, uint64_t object
,
474 uint64_t blkid
, void *tag
);
476 void dmu_buf_rele(dmu_buf_t
*db
, void *tag
);
477 uint64_t dmu_buf_refcount(dmu_buf_t
*db
);
480 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
481 * range of an object. A pointer to an array of dmu_buf_t*'s is
482 * returned (in *dbpp).
484 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
485 * frees the array. The hold on the array of buffers MUST be released
486 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
487 * individually with dmu_buf_rele.
489 int dmu_buf_hold_array_by_bonus(dmu_buf_t
*db
, uint64_t offset
,
490 uint64_t length
, boolean_t read
, void *tag
,
491 int *numbufsp
, dmu_buf_t
***dbpp
);
492 void dmu_buf_rele_array(dmu_buf_t
**, int numbufs
, void *tag
);
494 typedef void dmu_buf_evict_func_t(void *user_ptr
);
497 * A DMU buffer user object may be associated with a dbuf for the
498 * duration of its lifetime. This allows the user of a dbuf (client)
499 * to attach private data to a dbuf (e.g. in-core only data such as a
500 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
501 * when that dbuf has been evicted. Clients typically respond to the
502 * eviction notification by freeing their private data, thus ensuring
503 * the same lifetime for both dbuf and private data.
505 * The mapping from a dmu_buf_user_t to any client private data is the
506 * client's responsibility. All current consumers of the API with private
507 * data embed a dmu_buf_user_t as the first member of the structure for
508 * their private data. This allows conversions between the two types
509 * with a simple cast. Since the DMU buf user API never needs access
510 * to the private data, other strategies can be employed if necessary
511 * or convenient for the client (e.g. using container_of() to do the
512 * conversion for private data that cannot have the dmu_buf_user_t as
515 * Eviction callbacks are executed without the dbuf mutex held or any
516 * other type of mechanism to guarantee that the dbuf is still available.
517 * For this reason, users must assume the dbuf has already been freed
518 * and not reference the dbuf from the callback context.
520 * Users requesting "immediate eviction" are notified as soon as the dbuf
521 * is only referenced by dirty records (dirties == holds). Otherwise the
522 * notification occurs after eviction processing for the dbuf begins.
524 typedef struct dmu_buf_user
{
526 * Asynchronous user eviction callback state.
528 taskq_ent_t dbu_tqent
;
530 /* This instance's eviction function pointer. */
531 dmu_buf_evict_func_t
*dbu_evict_func
;
534 * Pointer to user's dbuf pointer. NULL for clients that do
535 * not associate a dbuf with their user data.
537 * The dbuf pointer is cleared upon eviction so as to catch
538 * use-after-evict bugs in clients.
540 dmu_buf_t
**dbu_clear_on_evict_dbufp
;
545 * Initialize the given dmu_buf_user_t instance with the eviction function
546 * evict_func, to be called when the user is evicted.
548 * NOTE: This function should only be called once on a given dmu_buf_user_t.
549 * To allow enforcement of this, dbu must already be zeroed on entry.
552 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
554 dmu_buf_init_user(dmu_buf_user_t
*dbu
, dmu_buf_evict_func_t
*evict_func
,
555 dmu_buf_t
**clear_on_evict_dbufp
);
558 dmu_buf_init_user(dmu_buf_user_t
*dbu
, dmu_buf_evict_func_t
*evict_func
,
559 dmu_buf_t
**clear_on_evict_dbufp
)
561 ASSERT(dbu
->dbu_evict_func
== NULL
);
562 ASSERT(evict_func
!= NULL
);
563 dbu
->dbu_evict_func
= evict_func
;
564 taskq_init_ent(&dbu
->dbu_tqent
);
566 dbu
->dbu_clear_on_evict_dbufp
= clear_on_evict_dbufp
;
572 * Attach user data to a dbuf and mark it for normal (when the dbuf's
573 * data is cleared or its reference count goes to zero) eviction processing.
575 * Returns NULL on success, or the existing user if another user currently
578 void *dmu_buf_set_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
581 * Attach user data to a dbuf and mark it for immediate (its dirty and
582 * reference counts are equal) eviction processing.
584 * Returns NULL on success, or the existing user if another user currently
587 void *dmu_buf_set_user_ie(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
590 * Replace the current user of a dbuf.
592 * If given the current user of a dbuf, replaces the dbuf's user with
593 * "new_user" and returns the user data pointer that was replaced.
594 * Otherwise returns the current, and unmodified, dbuf user pointer.
596 void *dmu_buf_replace_user(dmu_buf_t
*db
,
597 dmu_buf_user_t
*old_user
, dmu_buf_user_t
*new_user
);
600 * Remove the specified user data for a DMU buffer.
602 * Returns the user that was removed on success, or the current user if
603 * another user currently owns the buffer.
605 void *dmu_buf_remove_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
608 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
610 void *dmu_buf_get_user(dmu_buf_t
*db
);
612 /* Block until any in-progress dmu buf user evictions complete. */
613 void dmu_buf_user_evict_wait(void);
616 * Returns the blkptr associated with this dbuf, or NULL if not set.
618 struct blkptr
*dmu_buf_get_blkptr(dmu_buf_t
*db
);
621 * Indicate that you are going to modify the buffer's data (db_data).
623 * The transaction (tx) must be assigned to a txg (ie. you've called
624 * dmu_tx_assign()). The buffer's object must be held in the tx
625 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
627 void dmu_buf_will_dirty(dmu_buf_t
*db
, dmu_tx_t
*tx
);
630 * Tells if the given dbuf is freeable.
632 boolean_t
dmu_buf_freeable(dmu_buf_t
*);
635 * You must create a transaction, then hold the objects which you will
636 * (or might) modify as part of this transaction. Then you must assign
637 * the transaction to a transaction group. Once the transaction has
638 * been assigned, you can modify buffers which belong to held objects as
639 * part of this transaction. You can't modify buffers before the
640 * transaction has been assigned; you can't modify buffers which don't
641 * belong to objects which this transaction holds; you can't hold
642 * objects once the transaction has been assigned. You may hold an
643 * object which you are going to free (with dmu_object_free()), but you
646 * You can abort the transaction before it has been assigned.
648 * Note that you may hold buffers (with dmu_buf_hold) at any time,
649 * regardless of transaction state.
652 #define DMU_NEW_OBJECT (-1ULL)
653 #define DMU_OBJECT_END (-1ULL)
655 dmu_tx_t
*dmu_tx_create(objset_t
*os
);
656 void dmu_tx_hold_write(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
, int len
);
657 void dmu_tx_hold_free(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
,
659 void dmu_tx_hold_zap(dmu_tx_t
*tx
, uint64_t object
, int add
, const char *name
);
660 void dmu_tx_hold_bonus(dmu_tx_t
*tx
, uint64_t object
);
661 void dmu_tx_hold_spill(dmu_tx_t
*tx
, uint64_t object
);
662 void dmu_tx_hold_sa(dmu_tx_t
*tx
, struct sa_handle
*hdl
, boolean_t may_grow
);
663 void dmu_tx_hold_sa_create(dmu_tx_t
*tx
, int total_size
);
664 void dmu_tx_abort(dmu_tx_t
*tx
);
665 int dmu_tx_assign(dmu_tx_t
*tx
, enum txg_how txg_how
);
666 void dmu_tx_wait(dmu_tx_t
*tx
);
667 void dmu_tx_commit(dmu_tx_t
*tx
);
670 * To register a commit callback, dmu_tx_callback_register() must be called.
672 * dcb_data is a pointer to caller private data that is passed on as a
673 * callback parameter. The caller is responsible for properly allocating and
676 * When registering a callback, the transaction must be already created, but
677 * it cannot be committed or aborted. It can be assigned to a txg or not.
679 * The callback will be called after the transaction has been safely written
680 * to stable storage and will also be called if the dmu_tx is aborted.
681 * If there is any error which prevents the transaction from being committed to
682 * disk, the callback will be called with a value of error != 0.
684 typedef void dmu_tx_callback_func_t(void *dcb_data
, int error
);
686 void dmu_tx_callback_register(dmu_tx_t
*tx
, dmu_tx_callback_func_t
*dcb_func
,
690 * Free up the data blocks for a defined range of a file. If size is
691 * -1, the range from offset to end-of-file is freed.
693 int dmu_free_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
694 uint64_t size
, dmu_tx_t
*tx
);
695 int dmu_free_long_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
697 int dmu_free_long_object(objset_t
*os
, uint64_t object
);
700 * Convenience functions.
702 * Canfail routines will return 0 on success, or an errno if there is a
703 * nonrecoverable I/O error.
705 #define DMU_READ_PREFETCH 0 /* prefetch */
706 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
707 int dmu_read(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
708 void *buf
, uint32_t flags
);
709 void dmu_write(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
710 const void *buf
, dmu_tx_t
*tx
);
711 void dmu_prealloc(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
714 #include <linux/blkdev_compat.h>
715 int dmu_read_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
);
716 int dmu_read_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
);
717 int dmu_write_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
,
719 int dmu_write_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
,
722 struct arc_buf
*dmu_request_arcbuf(dmu_buf_t
*handle
, int size
);
723 void dmu_return_arcbuf(struct arc_buf
*buf
);
724 void dmu_assign_arcbuf(dmu_buf_t
*handle
, uint64_t offset
, struct arc_buf
*buf
,
726 int dmu_xuio_init(struct xuio
*uio
, int niov
);
727 void dmu_xuio_fini(struct xuio
*uio
);
728 int dmu_xuio_add(struct xuio
*uio
, struct arc_buf
*abuf
, offset_t off
,
730 int dmu_xuio_cnt(struct xuio
*uio
);
731 struct arc_buf
*dmu_xuio_arcbuf(struct xuio
*uio
, int i
);
732 void dmu_xuio_clear(struct xuio
*uio
, int i
);
733 void xuio_stat_wbuf_copied(void);
734 void xuio_stat_wbuf_nocopy(void);
736 extern int zfs_prefetch_disable
;
737 extern int zfs_max_recordsize
;
740 * Asynchronously try to read in the data.
742 void dmu_prefetch(objset_t
*os
, uint64_t object
, int64_t level
, uint64_t offset
,
743 uint64_t len
, enum zio_priority pri
);
745 typedef struct dmu_object_info
{
746 /* All sizes are in bytes unless otherwise indicated. */
747 uint32_t doi_data_block_size
;
748 uint32_t doi_metadata_block_size
;
749 dmu_object_type_t doi_type
;
750 dmu_object_type_t doi_bonus_type
;
751 uint64_t doi_bonus_size
;
752 uint8_t doi_indirection
; /* 2 = dnode->indirect->data */
753 uint8_t doi_checksum
;
754 uint8_t doi_compress
;
757 uint64_t doi_physical_blocks_512
; /* data + metadata, 512b blks */
758 uint64_t doi_max_offset
;
759 uint64_t doi_fill_count
; /* number of non-empty blocks */
762 typedef void (*const arc_byteswap_func_t
)(void *buf
, size_t size
);
764 typedef struct dmu_object_type_info
{
765 dmu_object_byteswap_t ot_byteswap
;
766 boolean_t ot_metadata
;
768 } dmu_object_type_info_t
;
770 typedef const struct dmu_object_byteswap_info
{
771 arc_byteswap_func_t ob_func
;
773 } dmu_object_byteswap_info_t
;
775 extern const dmu_object_type_info_t dmu_ot
[DMU_OT_NUMTYPES
];
776 extern const dmu_object_byteswap_info_t dmu_ot_byteswap
[DMU_BSWAP_NUMFUNCS
];
779 * Get information on a DMU object.
781 * Return 0 on success or ENOENT if object is not allocated.
783 * If doi is NULL, just indicates whether the object exists.
785 int dmu_object_info(objset_t
*os
, uint64_t object
, dmu_object_info_t
*doi
);
786 void __dmu_object_info_from_dnode(struct dnode
*dn
, dmu_object_info_t
*doi
);
787 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
788 void dmu_object_info_from_dnode(struct dnode
*dn
, dmu_object_info_t
*doi
);
789 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
790 void dmu_object_info_from_db(dmu_buf_t
*db
, dmu_object_info_t
*doi
);
792 * Like dmu_object_info_from_db, but faster still when you only care about
793 * the size. This is specifically optimized for zfs_getattr().
795 void dmu_object_size_from_db(dmu_buf_t
*db
, uint32_t *blksize
,
796 u_longlong_t
*nblk512
);
798 typedef struct dmu_objset_stats
{
799 uint64_t dds_num_clones
; /* number of clones of this */
800 uint64_t dds_creation_txg
;
802 dmu_objset_type_t dds_type
;
803 uint8_t dds_is_snapshot
;
804 uint8_t dds_inconsistent
;
805 char dds_origin
[MAXNAMELEN
];
806 } dmu_objset_stats_t
;
809 * Get stats on a dataset.
811 void dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
);
814 * Add entries to the nvlist for all the objset's properties. See
815 * zfs_prop_table[] and zfs(1m) for details on the properties.
817 void dmu_objset_stats(objset_t
*os
, struct nvlist
*nv
);
820 * Get the space usage statistics for statvfs().
822 * refdbytes is the amount of space "referenced" by this objset.
823 * availbytes is the amount of space available to this objset, taking
824 * into account quotas & reservations, assuming that no other objsets
825 * use the space first. These values correspond to the 'referenced' and
826 * 'available' properties, described in the zfs(1m) manpage.
828 * usedobjs and availobjs are the number of objects currently allocated,
831 void dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
832 uint64_t *usedobjsp
, uint64_t *availobjsp
);
835 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
836 * (Contrast with the ds_guid which is a 64-bit ID that will never
837 * change, so there is a small probability that it will collide.)
839 uint64_t dmu_objset_fsid_guid(objset_t
*os
);
842 * Get the [cm]time for an objset's snapshot dir
844 timestruc_t
dmu_objset_snap_cmtime(objset_t
*os
);
846 int dmu_objset_is_snapshot(objset_t
*os
);
848 extern struct spa
*dmu_objset_spa(objset_t
*os
);
849 extern struct zilog
*dmu_objset_zil(objset_t
*os
);
850 extern struct dsl_pool
*dmu_objset_pool(objset_t
*os
);
851 extern struct dsl_dataset
*dmu_objset_ds(objset_t
*os
);
852 extern void dmu_objset_name(objset_t
*os
, char *buf
);
853 extern dmu_objset_type_t
dmu_objset_type(objset_t
*os
);
854 extern uint64_t dmu_objset_id(objset_t
*os
);
855 extern zfs_sync_type_t
dmu_objset_syncprop(objset_t
*os
);
856 extern zfs_logbias_op_t
dmu_objset_logbias(objset_t
*os
);
857 extern int dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
858 uint64_t *id
, uint64_t *offp
, boolean_t
*case_conflict
);
859 extern int dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *val
);
860 extern int dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
,
861 int maxlen
, boolean_t
*conflict
);
862 extern int dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
863 uint64_t *idp
, uint64_t *offp
);
865 typedef int objset_used_cb_t(dmu_object_type_t bonustype
,
866 void *bonus
, uint64_t *userp
, uint64_t *groupp
);
867 extern void dmu_objset_register_type(dmu_objset_type_t ost
,
868 objset_used_cb_t
*cb
);
869 extern void dmu_objset_set_user(objset_t
*os
, void *user_ptr
);
870 extern void *dmu_objset_get_user(objset_t
*os
);
873 * Return the txg number for the given assigned transaction.
875 uint64_t dmu_tx_get_txg(dmu_tx_t
*tx
);
879 * If a parent zio is provided this function initiates a write on the
880 * provided buffer as a child of the parent zio.
881 * In the absence of a parent zio, the write is completed synchronously.
882 * At write completion, blk is filled with the bp of the written block.
883 * Note that while the data covered by this function will be on stable
884 * storage when the write completes this new data does not become a
885 * permanent part of the file until the associated transaction commits.
889 * {zfs,zvol,ztest}_get_done() args
892 struct zilog
*zgd_zilog
;
893 struct blkptr
*zgd_bp
;
899 typedef void dmu_sync_cb_t(zgd_t
*arg
, int error
);
900 int dmu_sync(struct zio
*zio
, uint64_t txg
, dmu_sync_cb_t
*done
, zgd_t
*zgd
);
903 * Find the next hole or data block in file starting at *off
904 * Return found offset in *off. Return ESRCH for end of file.
906 int dmu_offset_next(objset_t
*os
, uint64_t object
, boolean_t hole
,
910 * Initial setup and final teardown.
912 extern void dmu_init(void);
913 extern void dmu_fini(void);
915 typedef void (*dmu_traverse_cb_t
)(objset_t
*os
, void *arg
, struct blkptr
*bp
,
916 uint64_t object
, uint64_t offset
, int len
);
917 void dmu_traverse_objset(objset_t
*os
, uint64_t txg_start
,
918 dmu_traverse_cb_t cb
, void *arg
);
920 int dmu_diff(const char *tosnap_name
, const char *fromsnap_name
,
921 struct vnode
*vp
, offset_t
*offp
);
924 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
925 extern uint64_t zfs_crc64_table
[256];
927 extern int zfs_mdcomp_disable
;
933 #endif /* _SYS_DMU_H */