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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
25 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
27 * Copyright 2013 DEY Storage Systems, Inc.
28 * Copyright 2014 HybridCluster. All rights reserved.
29 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30 * Copyright 2013 Saso Kiselkov. All rights reserved.
31 * Copyright (c) 2014 Integros [integros.com]
34 /* Portions Copyright 2010 Robert Milkowski */
40 * This file describes the interface that the DMU provides for its
43 * The DMU also interacts with the SPA. That interface is described in
47 #include <sys/zfs_context.h>
48 #include <sys/inttypes.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/zio_priority.h>
71 struct zbookmark_phys
;
78 typedef struct objset objset_t
;
79 typedef struct dmu_tx dmu_tx_t
;
80 typedef struct dsl_dir dsl_dir_t
;
81 typedef struct dnode dnode_t
;
83 typedef enum dmu_object_byteswap
{
95 * Allocating a new byteswap type number makes the on-disk format
96 * incompatible with any other format that uses the same number.
98 * Data can usually be structured to work with one of the
99 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
102 } dmu_object_byteswap_t
;
104 #define DMU_OT_NEWTYPE 0x80
105 #define DMU_OT_METADATA 0x40
106 #define DMU_OT_BYTESWAP_MASK 0x3f
109 * Defines a uint8_t object type. Object types specify if the data
110 * in the object is metadata (boolean) and how to byteswap the data
111 * (dmu_object_byteswap_t).
113 #define DMU_OT(byteswap, metadata) \
115 ((metadata) ? DMU_OT_METADATA : 0) | \
116 ((byteswap) & DMU_OT_BYTESWAP_MASK))
118 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
119 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
120 (ot) < DMU_OT_NUMTYPES)
122 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
123 ((ot) & DMU_OT_METADATA) : \
124 dmu_ot[(ot)].ot_metadata)
127 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
128 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
129 * is repurposed for embedded BPs.
131 #define DMU_OT_HAS_FILL(ot) \
132 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
134 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
135 ((ot) & DMU_OT_BYTESWAP_MASK) : \
136 dmu_ot[(ot)].ot_byteswap)
138 typedef enum dmu_object_type
{
141 DMU_OT_OBJECT_DIRECTORY
, /* ZAP */
142 DMU_OT_OBJECT_ARRAY
, /* UINT64 */
143 DMU_OT_PACKED_NVLIST
, /* UINT8 (XDR by nvlist_pack/unpack) */
144 DMU_OT_PACKED_NVLIST_SIZE
, /* UINT64 */
145 DMU_OT_BPOBJ
, /* UINT64 */
146 DMU_OT_BPOBJ_HDR
, /* UINT64 */
148 DMU_OT_SPACE_MAP_HEADER
, /* UINT64 */
149 DMU_OT_SPACE_MAP
, /* UINT64 */
151 DMU_OT_INTENT_LOG
, /* UINT64 */
153 DMU_OT_DNODE
, /* DNODE */
154 DMU_OT_OBJSET
, /* OBJSET */
156 DMU_OT_DSL_DIR
, /* UINT64 */
157 DMU_OT_DSL_DIR_CHILD_MAP
, /* ZAP */
158 DMU_OT_DSL_DS_SNAP_MAP
, /* ZAP */
159 DMU_OT_DSL_PROPS
, /* ZAP */
160 DMU_OT_DSL_DATASET
, /* UINT64 */
162 DMU_OT_ZNODE
, /* ZNODE */
163 DMU_OT_OLDACL
, /* Old ACL */
164 DMU_OT_PLAIN_FILE_CONTENTS
, /* UINT8 */
165 DMU_OT_DIRECTORY_CONTENTS
, /* ZAP */
166 DMU_OT_MASTER_NODE
, /* ZAP */
167 DMU_OT_UNLINKED_SET
, /* ZAP */
169 DMU_OT_ZVOL
, /* UINT8 */
170 DMU_OT_ZVOL_PROP
, /* ZAP */
171 /* other; for testing only! */
172 DMU_OT_PLAIN_OTHER
, /* UINT8 */
173 DMU_OT_UINT64_OTHER
, /* UINT64 */
174 DMU_OT_ZAP_OTHER
, /* ZAP */
175 /* new object types: */
176 DMU_OT_ERROR_LOG
, /* ZAP */
177 DMU_OT_SPA_HISTORY
, /* UINT8 */
178 DMU_OT_SPA_HISTORY_OFFSETS
, /* spa_his_phys_t */
179 DMU_OT_POOL_PROPS
, /* ZAP */
180 DMU_OT_DSL_PERMS
, /* ZAP */
181 DMU_OT_ACL
, /* ACL */
182 DMU_OT_SYSACL
, /* SYSACL */
183 DMU_OT_FUID
, /* FUID table (Packed NVLIST UINT8) */
184 DMU_OT_FUID_SIZE
, /* FUID table size UINT64 */
185 DMU_OT_NEXT_CLONES
, /* ZAP */
186 DMU_OT_SCAN_QUEUE
, /* ZAP */
187 DMU_OT_USERGROUP_USED
, /* ZAP */
188 DMU_OT_USERGROUP_QUOTA
, /* ZAP */
189 DMU_OT_USERREFS
, /* ZAP */
190 DMU_OT_DDT_ZAP
, /* ZAP */
191 DMU_OT_DDT_STATS
, /* ZAP */
192 DMU_OT_SA
, /* System attr */
193 DMU_OT_SA_MASTER_NODE
, /* ZAP */
194 DMU_OT_SA_ATTR_REGISTRATION
, /* ZAP */
195 DMU_OT_SA_ATTR_LAYOUTS
, /* ZAP */
196 DMU_OT_SCAN_XLATE
, /* ZAP */
197 DMU_OT_DEDUP
, /* fake dedup BP from ddt_bp_create() */
198 DMU_OT_DEADLIST
, /* ZAP */
199 DMU_OT_DEADLIST_HDR
, /* UINT64 */
200 DMU_OT_DSL_CLONES
, /* ZAP */
201 DMU_OT_BPOBJ_SUBOBJ
, /* UINT64 */
203 * Do not allocate new object types here. Doing so makes the on-disk
204 * format incompatible with any other format that uses the same object
207 * When creating an object which does not have one of the above types
208 * use the DMU_OTN_* type with the correct byteswap and metadata
211 * The DMU_OTN_* types do not have entries in the dmu_ot table,
212 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
213 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
214 * and DMU_OTN_* types).
219 * Names for valid types declared with DMU_OT().
221 DMU_OTN_UINT8_DATA
= DMU_OT(DMU_BSWAP_UINT8
, B_FALSE
),
222 DMU_OTN_UINT8_METADATA
= DMU_OT(DMU_BSWAP_UINT8
, B_TRUE
),
223 DMU_OTN_UINT16_DATA
= DMU_OT(DMU_BSWAP_UINT16
, B_FALSE
),
224 DMU_OTN_UINT16_METADATA
= DMU_OT(DMU_BSWAP_UINT16
, B_TRUE
),
225 DMU_OTN_UINT32_DATA
= DMU_OT(DMU_BSWAP_UINT32
, B_FALSE
),
226 DMU_OTN_UINT32_METADATA
= DMU_OT(DMU_BSWAP_UINT32
, B_TRUE
),
227 DMU_OTN_UINT64_DATA
= DMU_OT(DMU_BSWAP_UINT64
, B_FALSE
),
228 DMU_OTN_UINT64_METADATA
= DMU_OT(DMU_BSWAP_UINT64
, B_TRUE
),
229 DMU_OTN_ZAP_DATA
= DMU_OT(DMU_BSWAP_ZAP
, B_FALSE
),
230 DMU_OTN_ZAP_METADATA
= DMU_OT(DMU_BSWAP_ZAP
, B_TRUE
),
233 typedef enum txg_how
{
239 void byteswap_uint64_array(void *buf
, size_t size
);
240 void byteswap_uint32_array(void *buf
, size_t size
);
241 void byteswap_uint16_array(void *buf
, size_t size
);
242 void byteswap_uint8_array(void *buf
, size_t size
);
243 void zap_byteswap(void *buf
, size_t size
);
244 void zfs_oldacl_byteswap(void *buf
, size_t size
);
245 void zfs_acl_byteswap(void *buf
, size_t size
);
246 void zfs_znode_byteswap(void *buf
, size_t size
);
248 #define DS_FIND_SNAPSHOTS (1<<0)
249 #define DS_FIND_CHILDREN (1<<1)
250 #define DS_FIND_SERIALIZE (1<<2)
253 * The maximum number of bytes that can be accessed as part of one
254 * operation, including metadata.
256 #define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
257 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
259 #define DMU_USERUSED_OBJECT (-1ULL)
260 #define DMU_GROUPUSED_OBJECT (-2ULL)
263 * artificial blkids for bonus buffer and spill blocks
265 #define DMU_BONUS_BLKID (-1ULL)
266 #define DMU_SPILL_BLKID (-2ULL)
268 * Public routines to create, destroy, open, and close objsets.
270 int dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
);
271 int dmu_objset_own(const char *name
, dmu_objset_type_t type
,
272 boolean_t readonly
, void *tag
, objset_t
**osp
);
273 void dmu_objset_rele(objset_t
*os
, void *tag
);
274 void dmu_objset_disown(objset_t
*os
, void *tag
);
275 int dmu_objset_open_ds(struct dsl_dataset
*ds
, objset_t
**osp
);
277 void dmu_objset_evict_dbufs(objset_t
*os
);
278 int dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
279 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
);
280 int dmu_objset_clone(const char *name
, const char *origin
);
281 int dsl_destroy_snapshots_nvl(struct nvlist
*snaps
, boolean_t defer
,
282 struct nvlist
*errlist
);
283 int dmu_objset_snapshot_one(const char *fsname
, const char *snapname
);
284 int dmu_objset_snapshot_tmp(const char *, const char *, int);
285 int dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
287 void dmu_objset_byteswap(void *buf
, size_t size
);
288 int dsl_dataset_rename_snapshot(const char *fsname
,
289 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
);
291 typedef struct dmu_buf
{
292 uint64_t db_object
; /* object that this buffer is part of */
293 uint64_t db_offset
; /* byte offset in this object */
294 uint64_t db_size
; /* size of buffer in bytes */
295 void *db_data
; /* data in buffer */
299 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
301 #define DMU_POOL_DIRECTORY_OBJECT 1
302 #define DMU_POOL_CONFIG "config"
303 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
304 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
305 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
306 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
307 #define DMU_POOL_ROOT_DATASET "root_dataset"
308 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
309 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
310 #define DMU_POOL_ERRLOG_LAST "errlog_last"
311 #define DMU_POOL_SPARES "spares"
312 #define DMU_POOL_DEFLATE "deflate"
313 #define DMU_POOL_HISTORY "history"
314 #define DMU_POOL_PROPS "pool_props"
315 #define DMU_POOL_L2CACHE "l2cache"
316 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
317 #define DMU_POOL_DDT "DDT-%s-%s-%s"
318 #define DMU_POOL_DDT_STATS "DDT-statistics"
319 #define DMU_POOL_CREATION_VERSION "creation_version"
320 #define DMU_POOL_SCAN "scan"
321 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
322 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
323 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
324 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
325 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
328 * Allocate an object from this objset. The range of object numbers
329 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
331 * The transaction must be assigned to a txg. The newly allocated
332 * object will be "held" in the transaction (ie. you can modify the
333 * newly allocated object in this transaction).
335 * dmu_object_alloc() chooses an object and returns it in *objectp.
337 * dmu_object_claim() allocates a specific object number. If that
338 * number is already allocated, it fails and returns EEXIST.
340 * Return 0 on success, or ENOSPC or EEXIST as specified above.
342 uint64_t dmu_object_alloc(objset_t
*os
, dmu_object_type_t ot
,
343 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
344 int dmu_object_claim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
345 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
346 int dmu_object_reclaim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
347 int blocksize
, dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*txp
);
350 * Free an object from this objset.
352 * The object's data will be freed as well (ie. you don't need to call
353 * dmu_free(object, 0, -1, tx)).
355 * The object need not be held in the transaction.
357 * If there are any holds on this object's buffers (via dmu_buf_hold()),
358 * or tx holds on the object (via dmu_tx_hold_object()), you can not
359 * free it; it fails and returns EBUSY.
361 * If the object is not allocated, it fails and returns ENOENT.
363 * Return 0 on success, or EBUSY or ENOENT as specified above.
365 int dmu_object_free(objset_t
*os
, uint64_t object
, dmu_tx_t
*tx
);
368 * Find the next allocated or free object.
370 * The objectp parameter is in-out. It will be updated to be the next
371 * object which is allocated. Ignore objects which have not been
372 * modified since txg.
374 * XXX Can only be called on a objset with no dirty data.
376 * Returns 0 on success, or ENOENT if there are no more objects.
378 int dmu_object_next(objset_t
*os
, uint64_t *objectp
,
379 boolean_t hole
, uint64_t txg
);
382 * Set the data blocksize for an object.
384 * The object cannot have any blocks allcated beyond the first. If
385 * the first block is allocated already, the new size must be greater
386 * than the current block size. If these conditions are not met,
387 * ENOTSUP will be returned.
389 * Returns 0 on success, or EBUSY if there are any holds on the object
390 * contents, or ENOTSUP as described above.
392 int dmu_object_set_blocksize(objset_t
*os
, uint64_t object
, uint64_t size
,
393 int ibs
, dmu_tx_t
*tx
);
396 * Set the checksum property on a dnode. The new checksum algorithm will
397 * apply to all newly written blocks; existing blocks will not be affected.
399 void dmu_object_set_checksum(objset_t
*os
, uint64_t object
, uint8_t checksum
,
403 * Set the compress property on a dnode. The new compression algorithm will
404 * apply to all newly written blocks; existing blocks will not be affected.
406 void dmu_object_set_compress(objset_t
*os
, uint64_t object
, uint8_t compress
,
410 dmu_write_embedded(objset_t
*os
, uint64_t object
, uint64_t offset
,
411 void *data
, uint8_t etype
, uint8_t comp
, int uncompressed_size
,
412 int compressed_size
, int byteorder
, dmu_tx_t
*tx
);
415 * Decide how to write a block: checksum, compression, number of copies, etc.
417 #define WP_NOFILL 0x1
418 #define WP_DMU_SYNC 0x2
421 void dmu_write_policy(objset_t
*os
, dnode_t
*dn
, int level
, int wp
,
422 struct zio_prop
*zp
);
424 * The bonus data is accessed more or less like a regular buffer.
425 * You must dmu_bonus_hold() to get the buffer, which will give you a
426 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
427 * data. As with any normal buffer, you must call dmu_buf_read() to
428 * read db_data, dmu_buf_will_dirty() before modifying it, and the
429 * object must be held in an assigned transaction before calling
430 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
431 * buffer as well. You must release your hold with dmu_buf_rele().
433 * Returns ENOENT, EIO, or 0.
435 int dmu_bonus_hold(objset_t
*os
, uint64_t object
, void *tag
, dmu_buf_t
**);
436 int dmu_bonus_max(void);
437 int dmu_set_bonus(dmu_buf_t
*, int, dmu_tx_t
*);
438 int dmu_set_bonustype(dmu_buf_t
*, dmu_object_type_t
, dmu_tx_t
*);
439 dmu_object_type_t
dmu_get_bonustype(dmu_buf_t
*);
440 int dmu_rm_spill(objset_t
*, uint64_t, dmu_tx_t
*);
443 * Special spill buffer support used by "SA" framework
446 int dmu_spill_hold_by_bonus(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
);
447 int dmu_spill_hold_by_dnode(dnode_t
*dn
, uint32_t flags
,
448 void *tag
, dmu_buf_t
**dbp
);
449 int dmu_spill_hold_existing(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
);
452 * Obtain the DMU buffer from the specified object which contains the
453 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
454 * that it will remain in memory. You must release the hold with
455 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
456 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
458 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
459 * on the returned buffer before reading or writing the buffer's
460 * db_data. The comments for those routines describe what particular
461 * operations are valid after calling them.
463 * The object number must be a valid, allocated object number.
465 int dmu_buf_hold(objset_t
*os
, uint64_t object
, uint64_t offset
,
466 void *tag
, dmu_buf_t
**, int flags
);
467 int dmu_buf_hold_by_dnode(dnode_t
*dn
, uint64_t offset
,
468 void *tag
, dmu_buf_t
**dbp
, int flags
);
471 * Add a reference to a dmu buffer that has already been held via
472 * dmu_buf_hold() in the current context.
474 void dmu_buf_add_ref(dmu_buf_t
*db
, void* tag
);
477 * Attempt to add a reference to a dmu buffer that is in an unknown state,
478 * using a pointer that may have been invalidated by eviction processing.
479 * The request will succeed if the passed in dbuf still represents the
480 * same os/object/blkid, is ineligible for eviction, and has at least
481 * one hold by a user other than the syncer.
483 boolean_t
dmu_buf_try_add_ref(dmu_buf_t
*, objset_t
*os
, uint64_t object
,
484 uint64_t blkid
, void *tag
);
486 void dmu_buf_rele(dmu_buf_t
*db
, void *tag
);
487 uint64_t dmu_buf_refcount(dmu_buf_t
*db
);
490 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
491 * range of an object. A pointer to an array of dmu_buf_t*'s is
492 * returned (in *dbpp).
494 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
495 * frees the array. The hold on the array of buffers MUST be released
496 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
497 * individually with dmu_buf_rele.
499 int dmu_buf_hold_array_by_bonus(dmu_buf_t
*db
, uint64_t offset
,
500 uint64_t length
, boolean_t read
, void *tag
,
501 int *numbufsp
, dmu_buf_t
***dbpp
);
502 void dmu_buf_rele_array(dmu_buf_t
**, int numbufs
, void *tag
);
504 typedef void dmu_buf_evict_func_t(void *user_ptr
);
507 * A DMU buffer user object may be associated with a dbuf for the
508 * duration of its lifetime. This allows the user of a dbuf (client)
509 * to attach private data to a dbuf (e.g. in-core only data such as a
510 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
511 * when that dbuf has been evicted. Clients typically respond to the
512 * eviction notification by freeing their private data, thus ensuring
513 * the same lifetime for both dbuf and private data.
515 * The mapping from a dmu_buf_user_t to any client private data is the
516 * client's responsibility. All current consumers of the API with private
517 * data embed a dmu_buf_user_t as the first member of the structure for
518 * their private data. This allows conversions between the two types
519 * with a simple cast. Since the DMU buf user API never needs access
520 * to the private data, other strategies can be employed if necessary
521 * or convenient for the client (e.g. using container_of() to do the
522 * conversion for private data that cannot have the dmu_buf_user_t as
525 * Eviction callbacks are executed without the dbuf mutex held or any
526 * other type of mechanism to guarantee that the dbuf is still available.
527 * For this reason, users must assume the dbuf has already been freed
528 * and not reference the dbuf from the callback context.
530 * Users requesting "immediate eviction" are notified as soon as the dbuf
531 * is only referenced by dirty records (dirties == holds). Otherwise the
532 * notification occurs after eviction processing for the dbuf begins.
534 typedef struct dmu_buf_user
{
536 * Asynchronous user eviction callback state.
538 taskq_ent_t dbu_tqent
;
541 * This instance's eviction function pointers.
543 * dbu_evict_func_sync is called synchronously and then
544 * dbu_evict_func_async is executed asynchronously on a taskq.
546 dmu_buf_evict_func_t
*dbu_evict_func_sync
;
547 dmu_buf_evict_func_t
*dbu_evict_func_async
;
550 * Pointer to user's dbuf pointer. NULL for clients that do
551 * not associate a dbuf with their user data.
553 * The dbuf pointer is cleared upon eviction so as to catch
554 * use-after-evict bugs in clients.
556 dmu_buf_t
**dbu_clear_on_evict_dbufp
;
561 * Initialize the given dmu_buf_user_t instance with the eviction function
562 * evict_func, to be called when the user is evicted.
564 * NOTE: This function should only be called once on a given dmu_buf_user_t.
565 * To allow enforcement of this, dbu must already be zeroed on entry.
568 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
570 dmu_buf_init_user(dmu_buf_user_t
*dbu
, dmu_buf_evict_func_t
*evict_func
,
571 dmu_buf_evict_func_t
*evict_func_async
, dmu_buf_t
**clear_on_evict_dbufp
);
574 dmu_buf_init_user(dmu_buf_user_t
*dbu
, dmu_buf_evict_func_t
*evict_func_sync
,
575 dmu_buf_evict_func_t
*evict_func_async
, dmu_buf_t
**clear_on_evict_dbufp
)
577 ASSERT(dbu
->dbu_evict_func_sync
== NULL
);
578 ASSERT(dbu
->dbu_evict_func_async
== NULL
);
580 /* must have at least one evict func */
581 IMPLY(evict_func_sync
== NULL
, evict_func_async
!= NULL
);
582 dbu
->dbu_evict_func_sync
= evict_func_sync
;
583 dbu
->dbu_evict_func_async
= evict_func_async
;
585 dbu
->dbu_clear_on_evict_dbufp
= clear_on_evict_dbufp
;
591 * Attach user data to a dbuf and mark it for normal (when the dbuf's
592 * data is cleared or its reference count goes to zero) eviction processing.
594 * Returns NULL on success, or the existing user if another user currently
597 void *dmu_buf_set_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
600 * Attach user data to a dbuf and mark it for immediate (its dirty and
601 * reference counts are equal) eviction processing.
603 * Returns NULL on success, or the existing user if another user currently
606 void *dmu_buf_set_user_ie(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
609 * Replace the current user of a dbuf.
611 * If given the current user of a dbuf, replaces the dbuf's user with
612 * "new_user" and returns the user data pointer that was replaced.
613 * Otherwise returns the current, and unmodified, dbuf user pointer.
615 void *dmu_buf_replace_user(dmu_buf_t
*db
,
616 dmu_buf_user_t
*old_user
, dmu_buf_user_t
*new_user
);
619 * Remove the specified user data for a DMU buffer.
621 * Returns the user that was removed on success, or the current user if
622 * another user currently owns the buffer.
624 void *dmu_buf_remove_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
627 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
629 void *dmu_buf_get_user(dmu_buf_t
*db
);
631 objset_t
*dmu_buf_get_objset(dmu_buf_t
*db
);
632 dnode_t
*dmu_buf_dnode_enter(dmu_buf_t
*db
);
633 void dmu_buf_dnode_exit(dmu_buf_t
*db
);
635 /* Block until any in-progress dmu buf user evictions complete. */
636 void dmu_buf_user_evict_wait(void);
639 * Returns the blkptr associated with this dbuf, or NULL if not set.
641 struct blkptr
*dmu_buf_get_blkptr(dmu_buf_t
*db
);
644 * Indicate that you are going to modify the buffer's data (db_data).
646 * The transaction (tx) must be assigned to a txg (ie. you've called
647 * dmu_tx_assign()). The buffer's object must be held in the tx
648 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
650 void dmu_buf_will_dirty(dmu_buf_t
*db
, dmu_tx_t
*tx
);
653 * Tells if the given dbuf is freeable.
655 boolean_t
dmu_buf_freeable(dmu_buf_t
*);
658 * You must create a transaction, then hold the objects which you will
659 * (or might) modify as part of this transaction. Then you must assign
660 * the transaction to a transaction group. Once the transaction has
661 * been assigned, you can modify buffers which belong to held objects as
662 * part of this transaction. You can't modify buffers before the
663 * transaction has been assigned; you can't modify buffers which don't
664 * belong to objects which this transaction holds; you can't hold
665 * objects once the transaction has been assigned. You may hold an
666 * object which you are going to free (with dmu_object_free()), but you
669 * You can abort the transaction before it has been assigned.
671 * Note that you may hold buffers (with dmu_buf_hold) at any time,
672 * regardless of transaction state.
675 #define DMU_NEW_OBJECT (-1ULL)
676 #define DMU_OBJECT_END (-1ULL)
678 dmu_tx_t
*dmu_tx_create(objset_t
*os
);
679 void dmu_tx_hold_write(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
, int len
);
680 void dmu_tx_hold_free(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
,
682 void dmu_tx_hold_zap(dmu_tx_t
*tx
, uint64_t object
, int add
, const char *name
);
683 void dmu_tx_hold_bonus(dmu_tx_t
*tx
, uint64_t object
);
684 void dmu_tx_hold_spill(dmu_tx_t
*tx
, uint64_t object
);
685 void dmu_tx_hold_sa(dmu_tx_t
*tx
, struct sa_handle
*hdl
, boolean_t may_grow
);
686 void dmu_tx_hold_sa_create(dmu_tx_t
*tx
, int total_size
);
687 void dmu_tx_abort(dmu_tx_t
*tx
);
688 int dmu_tx_assign(dmu_tx_t
*tx
, enum txg_how txg_how
);
689 void dmu_tx_wait(dmu_tx_t
*tx
);
690 void dmu_tx_commit(dmu_tx_t
*tx
);
691 void dmu_tx_mark_netfree(dmu_tx_t
*tx
);
694 * To register a commit callback, dmu_tx_callback_register() must be called.
696 * dcb_data is a pointer to caller private data that is passed on as a
697 * callback parameter. The caller is responsible for properly allocating and
700 * When registering a callback, the transaction must be already created, but
701 * it cannot be committed or aborted. It can be assigned to a txg or not.
703 * The callback will be called after the transaction has been safely written
704 * to stable storage and will also be called if the dmu_tx is aborted.
705 * If there is any error which prevents the transaction from being committed to
706 * disk, the callback will be called with a value of error != 0.
708 typedef void dmu_tx_callback_func_t(void *dcb_data
, int error
);
710 void dmu_tx_callback_register(dmu_tx_t
*tx
, dmu_tx_callback_func_t
*dcb_func
,
714 * Free up the data blocks for a defined range of a file. If size is
715 * -1, the range from offset to end-of-file is freed.
717 int dmu_free_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
718 uint64_t size
, dmu_tx_t
*tx
);
719 int dmu_free_long_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
721 int dmu_free_long_object(objset_t
*os
, uint64_t object
);
724 * Convenience functions.
726 * Canfail routines will return 0 on success, or an errno if there is a
727 * nonrecoverable I/O error.
729 #define DMU_READ_PREFETCH 0 /* prefetch */
730 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
731 int dmu_read(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
732 void *buf
, uint32_t flags
);
733 void dmu_write(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
734 const void *buf
, dmu_tx_t
*tx
);
735 void dmu_prealloc(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
737 int dmu_read_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
);
738 int dmu_read_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
);
739 int dmu_write_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
,
741 int dmu_write_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
,
743 int dmu_write_pages(objset_t
*os
, uint64_t object
, uint64_t offset
,
744 uint64_t size
, struct page
*pp
, dmu_tx_t
*tx
);
745 struct arc_buf
*dmu_request_arcbuf(dmu_buf_t
*handle
, int size
);
746 void dmu_return_arcbuf(struct arc_buf
*buf
);
747 void dmu_assign_arcbuf(dmu_buf_t
*handle
, uint64_t offset
, struct arc_buf
*buf
,
749 int dmu_xuio_init(struct xuio
*uio
, int niov
);
750 void dmu_xuio_fini(struct xuio
*uio
);
751 int dmu_xuio_add(struct xuio
*uio
, struct arc_buf
*abuf
, offset_t off
,
753 int dmu_xuio_cnt(struct xuio
*uio
);
754 struct arc_buf
*dmu_xuio_arcbuf(struct xuio
*uio
, int i
);
755 void dmu_xuio_clear(struct xuio
*uio
, int i
);
756 void xuio_stat_wbuf_copied();
757 void xuio_stat_wbuf_nocopy();
759 extern boolean_t zfs_prefetch_disable
;
760 extern int zfs_max_recordsize
;
763 * Asynchronously try to read in the data.
765 void dmu_prefetch(objset_t
*os
, uint64_t object
, int64_t level
, uint64_t offset
,
766 uint64_t len
, enum zio_priority pri
);
768 typedef struct dmu_object_info
{
769 /* All sizes are in bytes unless otherwise indicated. */
770 uint32_t doi_data_block_size
;
771 uint32_t doi_metadata_block_size
;
772 dmu_object_type_t doi_type
;
773 dmu_object_type_t doi_bonus_type
;
774 uint64_t doi_bonus_size
;
775 uint8_t doi_indirection
; /* 2 = dnode->indirect->data */
776 uint8_t doi_checksum
;
777 uint8_t doi_compress
;
780 uint64_t doi_physical_blocks_512
; /* data + metadata, 512b blks */
781 uint64_t doi_max_offset
;
782 uint64_t doi_fill_count
; /* number of non-empty blocks */
785 typedef void arc_byteswap_func_t(void *buf
, size_t size
);
787 typedef struct dmu_object_type_info
{
788 dmu_object_byteswap_t ot_byteswap
;
789 boolean_t ot_metadata
;
791 } dmu_object_type_info_t
;
793 typedef struct dmu_object_byteswap_info
{
794 arc_byteswap_func_t
*ob_func
;
796 } dmu_object_byteswap_info_t
;
798 extern const dmu_object_type_info_t dmu_ot
[DMU_OT_NUMTYPES
];
799 extern const dmu_object_byteswap_info_t dmu_ot_byteswap
[DMU_BSWAP_NUMFUNCS
];
802 * Get information on a DMU object.
804 * Return 0 on success or ENOENT if object is not allocated.
806 * If doi is NULL, just indicates whether the object exists.
808 int dmu_object_info(objset_t
*os
, uint64_t object
, dmu_object_info_t
*doi
);
809 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
810 void dmu_object_info_from_dnode(dnode_t
*dn
, dmu_object_info_t
*doi
);
811 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
812 void dmu_object_info_from_db(dmu_buf_t
*db
, dmu_object_info_t
*doi
);
814 * Like dmu_object_info_from_db, but faster still when you only care about
815 * the size. This is specifically optimized for zfs_getattr().
817 void dmu_object_size_from_db(dmu_buf_t
*db
, uint32_t *blksize
,
818 u_longlong_t
*nblk512
);
820 typedef struct dmu_objset_stats
{
821 uint64_t dds_num_clones
; /* number of clones of this */
822 uint64_t dds_creation_txg
;
824 dmu_objset_type_t dds_type
;
825 uint8_t dds_is_snapshot
;
826 uint8_t dds_inconsistent
;
827 char dds_origin
[ZFS_MAX_DATASET_NAME_LEN
];
828 } dmu_objset_stats_t
;
831 * Get stats on a dataset.
833 void dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
);
836 * Add entries to the nvlist for all the objset's properties. See
837 * zfs_prop_table[] and zfs(1m) for details on the properties.
839 void dmu_objset_stats(objset_t
*os
, struct nvlist
*nv
);
842 * Get the space usage statistics for statvfs().
844 * refdbytes is the amount of space "referenced" by this objset.
845 * availbytes is the amount of space available to this objset, taking
846 * into account quotas & reservations, assuming that no other objsets
847 * use the space first. These values correspond to the 'referenced' and
848 * 'available' properties, described in the zfs(1m) manpage.
850 * usedobjs and availobjs are the number of objects currently allocated,
853 void dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
854 uint64_t *usedobjsp
, uint64_t *availobjsp
);
857 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
858 * (Contrast with the ds_guid which is a 64-bit ID that will never
859 * change, so there is a small probability that it will collide.)
861 uint64_t dmu_objset_fsid_guid(objset_t
*os
);
864 * Get the [cm]time for an objset's snapshot dir
866 timestruc_t
dmu_objset_snap_cmtime(objset_t
*os
);
868 int dmu_objset_is_snapshot(objset_t
*os
);
870 extern struct spa
*dmu_objset_spa(objset_t
*os
);
871 extern struct zilog
*dmu_objset_zil(objset_t
*os
);
872 extern struct dsl_pool
*dmu_objset_pool(objset_t
*os
);
873 extern struct dsl_dataset
*dmu_objset_ds(objset_t
*os
);
874 extern void dmu_objset_name(objset_t
*os
, char *buf
);
875 extern dmu_objset_type_t
dmu_objset_type(objset_t
*os
);
876 extern uint64_t dmu_objset_id(objset_t
*os
);
877 extern zfs_sync_type_t
dmu_objset_syncprop(objset_t
*os
);
878 extern zfs_logbias_op_t
dmu_objset_logbias(objset_t
*os
);
879 extern int dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
880 uint64_t *id
, uint64_t *offp
, boolean_t
*case_conflict
);
881 extern int dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
,
882 int maxlen
, boolean_t
*conflict
);
883 extern int dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
884 uint64_t *idp
, uint64_t *offp
);
886 typedef int objset_used_cb_t(dmu_object_type_t bonustype
,
887 void *bonus
, uint64_t *userp
, uint64_t *groupp
);
888 extern void dmu_objset_register_type(dmu_objset_type_t ost
,
889 objset_used_cb_t
*cb
);
890 extern void dmu_objset_set_user(objset_t
*os
, void *user_ptr
);
891 extern void *dmu_objset_get_user(objset_t
*os
);
894 * Return the txg number for the given assigned transaction.
896 uint64_t dmu_tx_get_txg(dmu_tx_t
*tx
);
900 * If a parent zio is provided this function initiates a write on the
901 * provided buffer as a child of the parent zio.
902 * In the absence of a parent zio, the write is completed synchronously.
903 * At write completion, blk is filled with the bp of the written block.
904 * Note that while the data covered by this function will be on stable
905 * storage when the write completes this new data does not become a
906 * permanent part of the file until the associated transaction commits.
910 * {zfs,zvol,ztest}_get_done() args
913 struct zilog
*zgd_zilog
;
914 struct blkptr
*zgd_bp
;
920 typedef void dmu_sync_cb_t(zgd_t
*arg
, int error
);
921 int dmu_sync(struct zio
*zio
, uint64_t txg
, dmu_sync_cb_t
*done
, zgd_t
*zgd
);
924 * Find the next hole or data block in file starting at *off
925 * Return found offset in *off. Return ESRCH for end of file.
927 int dmu_offset_next(objset_t
*os
, uint64_t object
, boolean_t hole
,
931 * Check if a DMU object has any dirty blocks. If so, sync out
932 * all pending transaction groups. Otherwise, this function
933 * does not alter DMU state. This could be improved to only sync
934 * out the necessary transaction groups for this particular
937 int dmu_object_wait_synced(objset_t
*os
, uint64_t object
);
940 * Initial setup and final teardown.
942 extern void dmu_init(void);
943 extern void dmu_fini(void);
945 typedef void (*dmu_traverse_cb_t
)(objset_t
*os
, void *arg
, struct blkptr
*bp
,
946 uint64_t object
, uint64_t offset
, int len
);
947 void dmu_traverse_objset(objset_t
*os
, uint64_t txg_start
,
948 dmu_traverse_cb_t cb
, void *arg
);
950 int dmu_diff(const char *tosnap_name
, const char *fromsnap_name
,
951 struct vnode
*vp
, offset_t
*offp
);
954 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
955 extern uint64_t zfs_crc64_table
[256];
957 extern int zfs_mdcomp_disable
;
963 #endif /* _SYS_DMU_H */