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>
65 struct zbookmark_phys
;
72 typedef struct objset objset_t
;
73 typedef struct dmu_tx dmu_tx_t
;
74 typedef struct dsl_dir dsl_dir_t
;
76 typedef enum dmu_object_byteswap
{
88 * Allocating a new byteswap type number makes the on-disk format
89 * incompatible with any other format that uses the same number.
91 * Data can usually be structured to work with one of the
92 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
95 } dmu_object_byteswap_t
;
97 #define DMU_OT_NEWTYPE 0x80
98 #define DMU_OT_METADATA 0x40
99 #define DMU_OT_BYTESWAP_MASK 0x3f
102 * Defines a uint8_t object type. Object types specify if the data
103 * in the object is metadata (boolean) and how to byteswap the data
104 * (dmu_object_byteswap_t).
106 #define DMU_OT(byteswap, metadata) \
108 ((metadata) ? DMU_OT_METADATA : 0) | \
109 ((byteswap) & DMU_OT_BYTESWAP_MASK))
111 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
112 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
113 (ot) < DMU_OT_NUMTYPES)
115 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
116 ((ot) & DMU_OT_METADATA) : \
117 dmu_ot[(int)(ot)].ot_metadata)
120 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
121 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
122 * is repurposed for embedded BPs.
124 #define DMU_OT_HAS_FILL(ot) \
125 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
127 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
128 ((ot) & DMU_OT_BYTESWAP_MASK) : \
129 dmu_ot[(int)(ot)].ot_byteswap)
131 typedef enum dmu_object_type
{
134 DMU_OT_OBJECT_DIRECTORY
, /* ZAP */
135 DMU_OT_OBJECT_ARRAY
, /* UINT64 */
136 DMU_OT_PACKED_NVLIST
, /* UINT8 (XDR by nvlist_pack/unpack) */
137 DMU_OT_PACKED_NVLIST_SIZE
, /* UINT64 */
138 DMU_OT_BPOBJ
, /* UINT64 */
139 DMU_OT_BPOBJ_HDR
, /* UINT64 */
141 DMU_OT_SPACE_MAP_HEADER
, /* UINT64 */
142 DMU_OT_SPACE_MAP
, /* UINT64 */
144 DMU_OT_INTENT_LOG
, /* UINT64 */
146 DMU_OT_DNODE
, /* DNODE */
147 DMU_OT_OBJSET
, /* OBJSET */
149 DMU_OT_DSL_DIR
, /* UINT64 */
150 DMU_OT_DSL_DIR_CHILD_MAP
, /* ZAP */
151 DMU_OT_DSL_DS_SNAP_MAP
, /* ZAP */
152 DMU_OT_DSL_PROPS
, /* ZAP */
153 DMU_OT_DSL_DATASET
, /* UINT64 */
155 DMU_OT_ZNODE
, /* ZNODE */
156 DMU_OT_OLDACL
, /* Old ACL */
157 DMU_OT_PLAIN_FILE_CONTENTS
, /* UINT8 */
158 DMU_OT_DIRECTORY_CONTENTS
, /* ZAP */
159 DMU_OT_MASTER_NODE
, /* ZAP */
160 DMU_OT_UNLINKED_SET
, /* ZAP */
162 DMU_OT_ZVOL
, /* UINT8 */
163 DMU_OT_ZVOL_PROP
, /* ZAP */
164 /* other; for testing only! */
165 DMU_OT_PLAIN_OTHER
, /* UINT8 */
166 DMU_OT_UINT64_OTHER
, /* UINT64 */
167 DMU_OT_ZAP_OTHER
, /* ZAP */
168 /* new object types: */
169 DMU_OT_ERROR_LOG
, /* ZAP */
170 DMU_OT_SPA_HISTORY
, /* UINT8 */
171 DMU_OT_SPA_HISTORY_OFFSETS
, /* spa_his_phys_t */
172 DMU_OT_POOL_PROPS
, /* ZAP */
173 DMU_OT_DSL_PERMS
, /* ZAP */
174 DMU_OT_ACL
, /* ACL */
175 DMU_OT_SYSACL
, /* SYSACL */
176 DMU_OT_FUID
, /* FUID table (Packed NVLIST UINT8) */
177 DMU_OT_FUID_SIZE
, /* FUID table size UINT64 */
178 DMU_OT_NEXT_CLONES
, /* ZAP */
179 DMU_OT_SCAN_QUEUE
, /* ZAP */
180 DMU_OT_USERGROUP_USED
, /* ZAP */
181 DMU_OT_USERGROUP_QUOTA
, /* ZAP */
182 DMU_OT_USERREFS
, /* ZAP */
183 DMU_OT_DDT_ZAP
, /* ZAP */
184 DMU_OT_DDT_STATS
, /* ZAP */
185 DMU_OT_SA
, /* System attr */
186 DMU_OT_SA_MASTER_NODE
, /* ZAP */
187 DMU_OT_SA_ATTR_REGISTRATION
, /* ZAP */
188 DMU_OT_SA_ATTR_LAYOUTS
, /* ZAP */
189 DMU_OT_SCAN_XLATE
, /* ZAP */
190 DMU_OT_DEDUP
, /* fake dedup BP from ddt_bp_create() */
191 DMU_OT_DEADLIST
, /* ZAP */
192 DMU_OT_DEADLIST_HDR
, /* UINT64 */
193 DMU_OT_DSL_CLONES
, /* ZAP */
194 DMU_OT_BPOBJ_SUBOBJ
, /* UINT64 */
196 * Do not allocate new object types here. Doing so makes the on-disk
197 * format incompatible with any other format that uses the same object
200 * When creating an object which does not have one of the above types
201 * use the DMU_OTN_* type with the correct byteswap and metadata
204 * The DMU_OTN_* types do not have entries in the dmu_ot table,
205 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
206 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
207 * and DMU_OTN_* types).
212 * Names for valid types declared with DMU_OT().
214 DMU_OTN_UINT8_DATA
= DMU_OT(DMU_BSWAP_UINT8
, B_FALSE
),
215 DMU_OTN_UINT8_METADATA
= DMU_OT(DMU_BSWAP_UINT8
, B_TRUE
),
216 DMU_OTN_UINT16_DATA
= DMU_OT(DMU_BSWAP_UINT16
, B_FALSE
),
217 DMU_OTN_UINT16_METADATA
= DMU_OT(DMU_BSWAP_UINT16
, B_TRUE
),
218 DMU_OTN_UINT32_DATA
= DMU_OT(DMU_BSWAP_UINT32
, B_FALSE
),
219 DMU_OTN_UINT32_METADATA
= DMU_OT(DMU_BSWAP_UINT32
, B_TRUE
),
220 DMU_OTN_UINT64_DATA
= DMU_OT(DMU_BSWAP_UINT64
, B_FALSE
),
221 DMU_OTN_UINT64_METADATA
= DMU_OT(DMU_BSWAP_UINT64
, B_TRUE
),
222 DMU_OTN_ZAP_DATA
= DMU_OT(DMU_BSWAP_ZAP
, B_FALSE
),
223 DMU_OTN_ZAP_METADATA
= DMU_OT(DMU_BSWAP_ZAP
, B_TRUE
),
226 typedef enum txg_how
{
232 void byteswap_uint64_array(void *buf
, size_t size
);
233 void byteswap_uint32_array(void *buf
, size_t size
);
234 void byteswap_uint16_array(void *buf
, size_t size
);
235 void byteswap_uint8_array(void *buf
, size_t size
);
236 void zap_byteswap(void *buf
, size_t size
);
237 void zfs_oldacl_byteswap(void *buf
, size_t size
);
238 void zfs_acl_byteswap(void *buf
, size_t size
);
239 void zfs_znode_byteswap(void *buf
, size_t size
);
241 #define DS_FIND_SNAPSHOTS (1<<0)
242 #define DS_FIND_CHILDREN (1<<1)
243 #define DS_FIND_SERIALIZE (1<<2)
246 * The maximum number of bytes that can be accessed as part of one
247 * operation, including metadata.
249 #define DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
250 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
252 #define DMU_USERUSED_OBJECT (-1ULL)
253 #define DMU_GROUPUSED_OBJECT (-2ULL)
256 * artificial blkids for bonus buffer and spill blocks
258 #define DMU_BONUS_BLKID (-1ULL)
259 #define DMU_SPILL_BLKID (-2ULL)
261 * Public routines to create, destroy, open, and close objsets.
263 int dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
);
264 int dmu_objset_own(const char *name
, dmu_objset_type_t type
,
265 boolean_t readonly
, void *tag
, objset_t
**osp
);
266 void dmu_objset_rele(objset_t
*os
, void *tag
);
267 void dmu_objset_disown(objset_t
*os
, void *tag
);
268 int dmu_objset_open_ds(struct dsl_dataset
*ds
, objset_t
**osp
);
270 void dmu_objset_evict_dbufs(objset_t
*os
);
271 int dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
272 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
);
273 int dmu_objset_clone(const char *name
, const char *origin
);
274 int dsl_destroy_snapshots_nvl(struct nvlist
*snaps
, boolean_t defer
,
275 struct nvlist
*errlist
);
276 int dmu_objset_snapshot_one(const char *fsname
, const char *snapname
);
277 int dmu_objset_snapshot_tmp(const char *, const char *, int);
278 int dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
280 void dmu_objset_byteswap(void *buf
, size_t size
);
281 int dsl_dataset_rename_snapshot(const char *fsname
,
282 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
);
284 typedef struct dmu_buf
{
285 uint64_t db_object
; /* object that this buffer is part of */
286 uint64_t db_offset
; /* byte offset in this object */
287 uint64_t db_size
; /* size of buffer in bytes */
288 void *db_data
; /* data in buffer */
292 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
294 #define DMU_POOL_DIRECTORY_OBJECT 1
295 #define DMU_POOL_CONFIG "config"
296 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
297 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
298 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
299 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
300 #define DMU_POOL_ROOT_DATASET "root_dataset"
301 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
302 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
303 #define DMU_POOL_ERRLOG_LAST "errlog_last"
304 #define DMU_POOL_SPARES "spares"
305 #define DMU_POOL_DEFLATE "deflate"
306 #define DMU_POOL_HISTORY "history"
307 #define DMU_POOL_PROPS "pool_props"
308 #define DMU_POOL_L2CACHE "l2cache"
309 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
310 #define DMU_POOL_DDT "DDT-%s-%s-%s"
311 #define DMU_POOL_DDT_STATS "DDT-statistics"
312 #define DMU_POOL_CREATION_VERSION "creation_version"
313 #define DMU_POOL_SCAN "scan"
314 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
315 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
316 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
319 * Allocate an object from this objset. The range of object numbers
320 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
322 * The transaction must be assigned to a txg. The newly allocated
323 * object will be "held" in the transaction (ie. you can modify the
324 * newly allocated object in this transaction).
326 * dmu_object_alloc() chooses an object and returns it in *objectp.
328 * dmu_object_claim() allocates a specific object number. If that
329 * number is already allocated, it fails and returns EEXIST.
331 * Return 0 on success, or ENOSPC or EEXIST as specified above.
333 uint64_t dmu_object_alloc(objset_t
*os
, dmu_object_type_t ot
,
334 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
335 int dmu_object_claim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
336 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
337 int dmu_object_reclaim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
338 int blocksize
, dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*txp
);
341 * Free an object from this objset.
343 * The object's data will be freed as well (ie. you don't need to call
344 * dmu_free(object, 0, -1, tx)).
346 * The object need not be held in the transaction.
348 * If there are any holds on this object's buffers (via dmu_buf_hold()),
349 * or tx holds on the object (via dmu_tx_hold_object()), you can not
350 * free it; it fails and returns EBUSY.
352 * If the object is not allocated, it fails and returns ENOENT.
354 * Return 0 on success, or EBUSY or ENOENT as specified above.
356 int dmu_object_free(objset_t
*os
, uint64_t object
, dmu_tx_t
*tx
);
359 * Find the next allocated or free object.
361 * The objectp parameter is in-out. It will be updated to be the next
362 * object which is allocated. Ignore objects which have not been
363 * modified since txg.
365 * XXX Can only be called on a objset with no dirty data.
367 * Returns 0 on success, or ENOENT if there are no more objects.
369 int dmu_object_next(objset_t
*os
, uint64_t *objectp
,
370 boolean_t hole
, uint64_t txg
);
373 * Set the data blocksize for an object.
375 * The object cannot have any blocks allcated beyond the first. If
376 * the first block is allocated already, the new size must be greater
377 * than the current block size. If these conditions are not met,
378 * ENOTSUP will be returned.
380 * Returns 0 on success, or EBUSY if there are any holds on the object
381 * contents, or ENOTSUP as described above.
383 int dmu_object_set_blocksize(objset_t
*os
, uint64_t object
, uint64_t size
,
384 int ibs
, dmu_tx_t
*tx
);
387 * Set the checksum property on a dnode. The new checksum algorithm will
388 * apply to all newly written blocks; existing blocks will not be affected.
390 void dmu_object_set_checksum(objset_t
*os
, uint64_t object
, uint8_t checksum
,
394 * Set the compress property on a dnode. The new compression algorithm will
395 * apply to all newly written blocks; existing blocks will not be affected.
397 void dmu_object_set_compress(objset_t
*os
, uint64_t object
, uint8_t compress
,
401 dmu_write_embedded(objset_t
*os
, uint64_t object
, uint64_t offset
,
402 void *data
, uint8_t etype
, uint8_t comp
, int uncompressed_size
,
403 int compressed_size
, int byteorder
, dmu_tx_t
*tx
);
406 * Decide how to write a block: checksum, compression, number of copies, etc.
408 #define WP_NOFILL 0x1
409 #define WP_DMU_SYNC 0x2
412 void dmu_write_policy(objset_t
*os
, struct dnode
*dn
, int level
, int wp
,
413 struct zio_prop
*zp
);
415 * The bonus data is accessed more or less like a regular buffer.
416 * You must dmu_bonus_hold() to get the buffer, which will give you a
417 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
418 * data. As with any normal buffer, you must call dmu_buf_read() to
419 * read db_data, dmu_buf_will_dirty() before modifying it, and the
420 * object must be held in an assigned transaction before calling
421 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
422 * buffer as well. You must release what you hold with dmu_buf_rele().
424 * Returns ENOENT, EIO, or 0.
426 int dmu_bonus_hold(objset_t
*os
, uint64_t object
, void *tag
, dmu_buf_t
**);
427 int dmu_bonus_max(void);
428 int dmu_set_bonus(dmu_buf_t
*, int, dmu_tx_t
*);
429 int dmu_set_bonustype(dmu_buf_t
*, dmu_object_type_t
, dmu_tx_t
*);
430 dmu_object_type_t
dmu_get_bonustype(dmu_buf_t
*);
431 int dmu_rm_spill(objset_t
*, uint64_t, dmu_tx_t
*);
434 * Special spill buffer support used by "SA" framework
437 int dmu_spill_hold_by_bonus(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
);
438 int dmu_spill_hold_by_dnode(struct dnode
*dn
, uint32_t flags
,
439 void *tag
, dmu_buf_t
**dbp
);
440 int dmu_spill_hold_existing(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
);
443 * Obtain the DMU buffer from the specified object which contains the
444 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
445 * that it will remain in memory. You must release the hold with
446 * dmu_buf_rele(). You must not access the dmu_buf_t after releasing
447 * what you hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
449 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
450 * on the returned buffer before reading or writing the buffer's
451 * db_data. The comments for those routines describe what particular
452 * operations are valid after calling them.
454 * The object number must be a valid, allocated object number.
456 int dmu_buf_hold(objset_t
*os
, uint64_t object
, uint64_t offset
,
457 void *tag
, dmu_buf_t
**, int flags
);
460 * Add a reference to a dmu buffer that has already been held via
461 * dmu_buf_hold() in the current context.
463 void dmu_buf_add_ref(dmu_buf_t
*db
, void* tag
);
466 * Attempt to add a reference to a dmu buffer that is in an unknown state,
467 * using a pointer that may have been invalidated by eviction processing.
468 * The request will succeed if the passed in dbuf still represents the
469 * same os/object/blkid, is ineligible for eviction, and has at least
470 * one hold by a user other than the syncer.
472 boolean_t
dmu_buf_try_add_ref(dmu_buf_t
*, objset_t
*os
, uint64_t object
,
473 uint64_t blkid
, void *tag
);
475 void dmu_buf_rele(dmu_buf_t
*db
, void *tag
);
476 uint64_t dmu_buf_refcount(dmu_buf_t
*db
);
479 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
480 * range of an object. A pointer to an array of dmu_buf_t*'s is
481 * returned (in *dbpp).
483 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
484 * frees the array. The hold on the array of buffers MUST be released
485 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
486 * individually with dmu_buf_rele.
488 int dmu_buf_hold_array_by_bonus(dmu_buf_t
*db
, uint64_t offset
,
489 uint64_t length
, int read
, void *tag
, int *numbufsp
, dmu_buf_t
***dbpp
);
490 void dmu_buf_rele_array(dmu_buf_t
**, int numbufs
, void *tag
);
492 typedef void dmu_buf_evict_func_t(void *user_ptr
);
495 * A DMU buffer user object may be associated with a dbuf for the
496 * duration of its lifetime. This allows the user of a dbuf (client)
497 * to attach private data to a dbuf (e.g. in-core only data such as a
498 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
499 * when that dbuf has been evicted. Clients typically respond to the
500 * eviction notification by freeing their private data, thus ensuring
501 * the same lifetime for both dbuf and private data.
503 * The mapping from a dmu_buf_user_t to any client private data is the
504 * client's responsibility. All current consumers of the API with private
505 * data embed a dmu_buf_user_t as the first member of the structure for
506 * their private data. This allows conversions between the two types
507 * with a simple cast. Since the DMU buf user API never needs access
508 * to the private data, other strategies can be employed if necessary
509 * or convenient for the client (e.g. using container_of() to do the
510 * conversion for private data that cannot have the dmu_buf_user_t as
513 * Eviction callbacks are executed without the dbuf mutex held or any
514 * other type of mechanism to guarantee that the dbuf is still available.
515 * For this reason, users must assume the dbuf has already been freed
516 * and not reference the dbuf from the callback context.
518 * Users requesting "immediate eviction" are notified as soon as the dbuf
519 * is only referenced by dirty records (dirties == holds). Otherwise the
520 * notification occurs after eviction processing for the dbuf begins.
522 typedef struct dmu_buf_user
{
524 * Asynchronous user eviction callback state.
526 taskq_ent_t dbu_tqent
;
528 /* This instance's eviction function pointer. */
529 dmu_buf_evict_func_t
*dbu_evict_func
;
532 * Pointer to user's dbuf pointer. NULL for clients that do
533 * not associate a dbuf with their user data.
535 * The dbuf pointer is cleared upon eviction so as to catch
536 * use-after-evict bugs in clients.
538 dmu_buf_t
**dbu_clear_on_evict_dbufp
;
543 * Initialize the given dmu_buf_user_t instance with the eviction function
544 * evict_func, to be called when the user is evicted.
546 * NOTE: This function should only be called once on a given dmu_buf_user_t.
547 * To allow enforcement of this, dbu must already be zeroed on entry.
550 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
552 dmu_buf_init_user(dmu_buf_user_t
*dbu
, dmu_buf_evict_func_t
*evict_func
,
553 dmu_buf_t
**clear_on_evict_dbufp
);
556 dmu_buf_init_user(dmu_buf_user_t
*dbu
, dmu_buf_evict_func_t
*evict_func
,
557 dmu_buf_t
**clear_on_evict_dbufp
)
559 ASSERT(dbu
->dbu_evict_func
== NULL
);
560 ASSERT(evict_func
!= NULL
);
561 dbu
->dbu_evict_func
= evict_func
;
562 taskq_init_ent(&dbu
->dbu_tqent
);
564 dbu
->dbu_clear_on_evict_dbufp
= clear_on_evict_dbufp
;
570 * Attach user data to a dbuf and mark it for normal (when the dbuf's
571 * data is cleared or its reference count goes to zero) eviction processing.
573 * Returns NULL on success, or the existing user if another user currently
576 void *dmu_buf_set_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
579 * Attach user data to a dbuf and mark it for immediate (its dirty and
580 * reference counts are equal) eviction processing.
582 * Returns NULL on success, or the existing user if another user currently
585 void *dmu_buf_set_user_ie(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
588 * Replace the current user of a dbuf.
590 * If given the current user of a dbuf, replaces the dbuf's user with
591 * "new_user" and returns the user data pointer that was replaced.
592 * Otherwise returns the current, and unmodified, dbuf user pointer.
594 void *dmu_buf_replace_user(dmu_buf_t
*db
,
595 dmu_buf_user_t
*old_user
, dmu_buf_user_t
*new_user
);
598 * Remove the specified user data for a DMU buffer.
600 * Returns the user that was removed on success, or the current user if
601 * another user currently owns the buffer.
603 void *dmu_buf_remove_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
606 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
608 void *dmu_buf_get_user(dmu_buf_t
*db
);
610 /* Block until any in-progress dmu buf user evictions complete. */
611 void dmu_buf_user_evict_wait(void);
614 * Returns the blkptr associated with this dbuf, or NULL if not set.
616 struct blkptr
*dmu_buf_get_blkptr(dmu_buf_t
*db
);
619 * Indicate that you are going to modify the buffer's data (db_data).
621 * The transaction (tx) must be assigned to a txg (ie. you've called
622 * dmu_tx_assign()). The buffer's object must be held in the tx
623 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
625 void dmu_buf_will_dirty(dmu_buf_t
*db
, dmu_tx_t
*tx
);
628 * Tells if the given dbuf is freeable.
630 boolean_t
dmu_buf_freeable(dmu_buf_t
*);
633 * You must create a transaction, then hold the objects which you will
634 * (or might) modify as part of this transaction. Then you must assign
635 * the transaction to a transaction group. Once the transaction has
636 * been assigned, you can modify buffers which belong to held objects as
637 * part of this transaction. You can't modify buffers before the
638 * transaction has been assigned; you can't modify buffers which don't
639 * belong to objects which this transaction holds; you can't hold
640 * objects once the transaction has been assigned. You may hold an
641 * object which you are going to free (with dmu_object_free()), but you
644 * You can abort the transaction before it has been assigned.
646 * Note that you may hold buffers (with dmu_buf_hold) at any time,
647 * regardless of transaction state.
650 #define DMU_NEW_OBJECT (-1ULL)
651 #define DMU_OBJECT_END (-1ULL)
653 dmu_tx_t
*dmu_tx_create(objset_t
*os
);
654 void dmu_tx_hold_write(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
, int len
);
655 void dmu_tx_hold_free(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
,
657 void dmu_tx_hold_zap(dmu_tx_t
*tx
, uint64_t object
, int add
, const char *name
);
658 void dmu_tx_hold_bonus(dmu_tx_t
*tx
, uint64_t object
);
659 void dmu_tx_hold_spill(dmu_tx_t
*tx
, uint64_t object
);
660 void dmu_tx_hold_sa(dmu_tx_t
*tx
, struct sa_handle
*hdl
, boolean_t may_grow
);
661 void dmu_tx_hold_sa_create(dmu_tx_t
*tx
, int total_size
);
662 void dmu_tx_abort(dmu_tx_t
*tx
);
663 int dmu_tx_assign(dmu_tx_t
*tx
, enum txg_how txg_how
);
664 void dmu_tx_wait(dmu_tx_t
*tx
);
665 void dmu_tx_commit(dmu_tx_t
*tx
);
668 * To register a commit callback, dmu_tx_callback_register() must be called.
670 * dcb_data is a pointer to caller private data that is passed on as a
671 * callback parameter. The caller is responsible for properly allocating and
674 * When registering a callback, the transaction must be already created, but
675 * it cannot be committed or aborted. It can be assigned to a txg or not.
677 * The callback will be called after the transaction has been safely written
678 * to stable storage and will also be called if the dmu_tx is aborted.
679 * If there is any error which prevents the transaction from being committed to
680 * disk, the callback will be called with a value of error != 0.
682 typedef void dmu_tx_callback_func_t(void *dcb_data
, int error
);
684 void dmu_tx_callback_register(dmu_tx_t
*tx
, dmu_tx_callback_func_t
*dcb_func
,
688 * Free up the data blocks for a defined range of a file. If size is
689 * -1, the range from offset to end-of-file is freed.
691 int dmu_free_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
692 uint64_t size
, dmu_tx_t
*tx
);
693 int dmu_free_long_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
695 int dmu_free_long_object(objset_t
*os
, uint64_t object
);
698 * Convenience functions.
700 * Canfail routines will return 0 on success, or an errno if there is a
701 * nonrecoverable I/O error.
703 #define DMU_READ_PREFETCH 0 /* prefetch */
704 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
705 int dmu_read(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
706 void *buf
, uint32_t flags
);
707 void dmu_write(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
708 const void *buf
, dmu_tx_t
*tx
);
709 void dmu_prealloc(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
712 #include <linux/blkdev_compat.h>
713 int dmu_read_bio(objset_t
*os
, uint64_t object
, struct bio
*bio
);
714 int dmu_write_bio(objset_t
*os
, uint64_t object
, struct bio
*bio
,
716 int dmu_read_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
);
717 int dmu_read_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
);
718 int dmu_write_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
,
720 int dmu_write_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
,
723 struct arc_buf
*dmu_request_arcbuf(dmu_buf_t
*handle
, int size
);
724 void dmu_return_arcbuf(struct arc_buf
*buf
);
725 void dmu_assign_arcbuf(dmu_buf_t
*handle
, uint64_t offset
, struct arc_buf
*buf
,
727 int dmu_xuio_init(struct xuio
*uio
, int niov
);
728 void dmu_xuio_fini(struct xuio
*uio
);
729 int dmu_xuio_add(struct xuio
*uio
, struct arc_buf
*abuf
, offset_t off
,
731 int dmu_xuio_cnt(struct xuio
*uio
);
732 struct arc_buf
*dmu_xuio_arcbuf(struct xuio
*uio
, int i
);
733 void dmu_xuio_clear(struct xuio
*uio
, int i
);
734 void xuio_stat_wbuf_copied(void);
735 void xuio_stat_wbuf_nocopy(void);
737 extern int zfs_prefetch_disable
;
738 extern int zfs_max_recordsize
;
741 * Asynchronously try to read in the data.
743 void dmu_prefetch(objset_t
*os
, uint64_t object
, uint64_t offset
,
746 typedef struct dmu_object_info
{
747 /* All sizes are in bytes unless otherwise indicated. */
748 uint32_t doi_data_block_size
;
749 uint32_t doi_metadata_block_size
;
750 dmu_object_type_t doi_type
;
751 dmu_object_type_t doi_bonus_type
;
752 uint64_t doi_bonus_size
;
753 uint8_t doi_indirection
; /* 2 = dnode->indirect->data */
754 uint8_t doi_checksum
;
755 uint8_t doi_compress
;
758 uint64_t doi_physical_blocks_512
; /* data + metadata, 512b blks */
759 uint64_t doi_max_offset
;
760 uint64_t doi_fill_count
; /* number of non-empty blocks */
763 typedef void (*const arc_byteswap_func_t
)(void *buf
, size_t size
);
765 typedef struct dmu_object_type_info
{
766 dmu_object_byteswap_t ot_byteswap
;
767 boolean_t ot_metadata
;
769 } dmu_object_type_info_t
;
771 typedef const struct dmu_object_byteswap_info
{
772 arc_byteswap_func_t ob_func
;
774 } dmu_object_byteswap_info_t
;
776 extern const dmu_object_type_info_t dmu_ot
[DMU_OT_NUMTYPES
];
777 extern const dmu_object_byteswap_info_t dmu_ot_byteswap
[DMU_BSWAP_NUMFUNCS
];
780 * Get information on a DMU object.
782 * Return 0 on success or ENOENT if object is not allocated.
784 * If doi is NULL, just indicates whether the object exists.
786 int dmu_object_info(objset_t
*os
, uint64_t object
, dmu_object_info_t
*doi
);
787 void __dmu_object_info_from_dnode(struct dnode
*dn
, dmu_object_info_t
*doi
);
788 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
789 void dmu_object_info_from_dnode(struct dnode
*dn
, dmu_object_info_t
*doi
);
790 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
791 void dmu_object_info_from_db(dmu_buf_t
*db
, dmu_object_info_t
*doi
);
793 * Like dmu_object_info_from_db, but faster still when you only care about
794 * the size. This is specifically optimized for zfs_getattr().
796 void dmu_object_size_from_db(dmu_buf_t
*db
, uint32_t *blksize
,
797 u_longlong_t
*nblk512
);
799 typedef struct dmu_objset_stats
{
800 uint64_t dds_num_clones
; /* number of clones of this */
801 uint64_t dds_creation_txg
;
803 dmu_objset_type_t dds_type
;
804 uint8_t dds_is_snapshot
;
805 uint8_t dds_inconsistent
;
806 char dds_origin
[MAXNAMELEN
];
807 } dmu_objset_stats_t
;
810 * Get stats on a dataset.
812 void dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
);
815 * Add entries to the nvlist for all the objset's properties. See
816 * zfs_prop_table[] and zfs(1m) for details on the properties.
818 void dmu_objset_stats(objset_t
*os
, struct nvlist
*nv
);
821 * Get the space usage statistics for statvfs().
823 * refdbytes is the amount of space "referenced" by this objset.
824 * availbytes is the amount of space available to this objset, taking
825 * into account quotas & reservations, assuming that no other objsets
826 * use the space first. These values correspond to the 'referenced' and
827 * 'available' properties, described in the zfs(1m) manpage.
829 * usedobjs and availobjs are the number of objects currently allocated,
832 void dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
833 uint64_t *usedobjsp
, uint64_t *availobjsp
);
836 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
837 * (Contrast with the ds_guid which is a 64-bit ID that will never
838 * change, so there is a small probability that it will collide.)
840 uint64_t dmu_objset_fsid_guid(objset_t
*os
);
843 * Get the [cm]time for an objset's snapshot dir
845 timestruc_t
dmu_objset_snap_cmtime(objset_t
*os
);
847 int dmu_objset_is_snapshot(objset_t
*os
);
849 extern struct spa
*dmu_objset_spa(objset_t
*os
);
850 extern struct zilog
*dmu_objset_zil(objset_t
*os
);
851 extern struct dsl_pool
*dmu_objset_pool(objset_t
*os
);
852 extern struct dsl_dataset
*dmu_objset_ds(objset_t
*os
);
853 extern void dmu_objset_name(objset_t
*os
, char *buf
);
854 extern dmu_objset_type_t
dmu_objset_type(objset_t
*os
);
855 extern uint64_t dmu_objset_id(objset_t
*os
);
856 extern zfs_sync_type_t
dmu_objset_syncprop(objset_t
*os
);
857 extern zfs_logbias_op_t
dmu_objset_logbias(objset_t
*os
);
858 extern int dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
859 uint64_t *id
, uint64_t *offp
, boolean_t
*case_conflict
);
860 extern int dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *val
);
861 extern int dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
,
862 int maxlen
, boolean_t
*conflict
);
863 extern int dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
864 uint64_t *idp
, uint64_t *offp
);
866 typedef int objset_used_cb_t(dmu_object_type_t bonustype
,
867 void *bonus
, uint64_t *userp
, uint64_t *groupp
);
868 extern void dmu_objset_register_type(dmu_objset_type_t ost
,
869 objset_used_cb_t
*cb
);
870 extern void dmu_objset_set_user(objset_t
*os
, void *user_ptr
);
871 extern void *dmu_objset_get_user(objset_t
*os
);
874 * Return the txg number for the given assigned transaction.
876 uint64_t dmu_tx_get_txg(dmu_tx_t
*tx
);
880 * If a parent zio is provided this function initiates a write on the
881 * provided buffer as a child of the parent zio.
882 * In the absence of a parent zio, the write is completed synchronously.
883 * At write completion, blk is filled with the bp of the written block.
884 * Note that while the data covered by this function will be on stable
885 * storage when the write completes this new data does not become a
886 * permanent part of the file until the associated transaction commits.
890 * {zfs,zvol,ztest}_get_done() args
893 struct zilog
*zgd_zilog
;
894 struct blkptr
*zgd_bp
;
900 typedef void dmu_sync_cb_t(zgd_t
*arg
, int error
);
901 int dmu_sync(struct zio
*zio
, uint64_t txg
, dmu_sync_cb_t
*done
, zgd_t
*zgd
);
904 * Find the next hole or data block in file starting at *off
905 * Return found offset in *off. Return ESRCH for end of file.
907 int dmu_offset_next(objset_t
*os
, uint64_t object
, boolean_t hole
,
911 * Initial setup and final teardown.
913 extern void dmu_init(void);
914 extern void dmu_fini(void);
916 typedef void (*dmu_traverse_cb_t
)(objset_t
*os
, void *arg
, struct blkptr
*bp
,
917 uint64_t object
, uint64_t offset
, int len
);
918 void dmu_traverse_objset(objset_t
*os
, uint64_t txg_start
,
919 dmu_traverse_cb_t cb
, void *arg
);
921 int dmu_diff(const char *tosnap_name
, const char *fromsnap_name
,
922 struct vnode
*vp
, offset_t
*offp
);
925 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
926 extern uint64_t zfs_crc64_table
[256];
928 extern int zfs_mdcomp_disable
;
934 #endif /* _SYS_DMU_H */