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, 2017 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.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
31 /* Portions Copyright 2010 Robert Milkowski */
37 * This file describes the interface that the DMU provides for its
40 * The DMU also interacts with the SPA. That interface is described in
44 #include <sys/zfs_context.h>
45 #include <sys/inttypes.h>
47 #include <sys/fs/zfs.h>
48 #include <sys/zio_compress.h>
49 #include <sys/zio_priority.h>
68 struct zbookmark_phys
;
75 typedef struct objset objset_t
;
76 typedef struct dmu_tx dmu_tx_t
;
77 typedef struct dsl_dir dsl_dir_t
;
78 typedef struct dnode dnode_t
;
80 typedef enum dmu_object_byteswap
{
92 * Allocating a new byteswap type number makes the on-disk format
93 * incompatible with any other format that uses the same number.
95 * Data can usually be structured to work with one of the
96 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
99 } dmu_object_byteswap_t
;
101 #define DMU_OT_NEWTYPE 0x80
102 #define DMU_OT_METADATA 0x40
103 #define DMU_OT_BYTESWAP_MASK 0x3f
106 * Defines a uint8_t object type. Object types specify if the data
107 * in the object is metadata (boolean) and how to byteswap the data
108 * (dmu_object_byteswap_t).
110 #define DMU_OT(byteswap, metadata) \
112 ((metadata) ? DMU_OT_METADATA : 0) | \
113 ((byteswap) & DMU_OT_BYTESWAP_MASK))
115 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
116 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
117 (ot) < DMU_OT_NUMTYPES)
119 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
120 ((ot) & DMU_OT_METADATA) : \
121 dmu_ot[(int)(ot)].ot_metadata)
124 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
125 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
126 * is repurposed for embedded BPs.
128 #define DMU_OT_HAS_FILL(ot) \
129 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
131 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
132 ((ot) & DMU_OT_BYTESWAP_MASK) : \
133 dmu_ot[(int)(ot)].ot_byteswap)
135 typedef enum dmu_object_type
{
138 DMU_OT_OBJECT_DIRECTORY
, /* ZAP */
139 DMU_OT_OBJECT_ARRAY
, /* UINT64 */
140 DMU_OT_PACKED_NVLIST
, /* UINT8 (XDR by nvlist_pack/unpack) */
141 DMU_OT_PACKED_NVLIST_SIZE
, /* UINT64 */
142 DMU_OT_BPOBJ
, /* UINT64 */
143 DMU_OT_BPOBJ_HDR
, /* UINT64 */
145 DMU_OT_SPACE_MAP_HEADER
, /* UINT64 */
146 DMU_OT_SPACE_MAP
, /* UINT64 */
148 DMU_OT_INTENT_LOG
, /* UINT64 */
150 DMU_OT_DNODE
, /* DNODE */
151 DMU_OT_OBJSET
, /* OBJSET */
153 DMU_OT_DSL_DIR
, /* UINT64 */
154 DMU_OT_DSL_DIR_CHILD_MAP
, /* ZAP */
155 DMU_OT_DSL_DS_SNAP_MAP
, /* ZAP */
156 DMU_OT_DSL_PROPS
, /* ZAP */
157 DMU_OT_DSL_DATASET
, /* UINT64 */
159 DMU_OT_ZNODE
, /* ZNODE */
160 DMU_OT_OLDACL
, /* Old ACL */
161 DMU_OT_PLAIN_FILE_CONTENTS
, /* UINT8 */
162 DMU_OT_DIRECTORY_CONTENTS
, /* ZAP */
163 DMU_OT_MASTER_NODE
, /* ZAP */
164 DMU_OT_UNLINKED_SET
, /* ZAP */
166 DMU_OT_ZVOL
, /* UINT8 */
167 DMU_OT_ZVOL_PROP
, /* ZAP */
168 /* other; for testing only! */
169 DMU_OT_PLAIN_OTHER
, /* UINT8 */
170 DMU_OT_UINT64_OTHER
, /* UINT64 */
171 DMU_OT_ZAP_OTHER
, /* ZAP */
172 /* new object types: */
173 DMU_OT_ERROR_LOG
, /* ZAP */
174 DMU_OT_SPA_HISTORY
, /* UINT8 */
175 DMU_OT_SPA_HISTORY_OFFSETS
, /* spa_his_phys_t */
176 DMU_OT_POOL_PROPS
, /* ZAP */
177 DMU_OT_DSL_PERMS
, /* ZAP */
178 DMU_OT_ACL
, /* ACL */
179 DMU_OT_SYSACL
, /* SYSACL */
180 DMU_OT_FUID
, /* FUID table (Packed NVLIST UINT8) */
181 DMU_OT_FUID_SIZE
, /* FUID table size UINT64 */
182 DMU_OT_NEXT_CLONES
, /* ZAP */
183 DMU_OT_SCAN_QUEUE
, /* ZAP */
184 DMU_OT_USERGROUP_USED
, /* ZAP */
185 DMU_OT_USERGROUP_QUOTA
, /* ZAP */
186 DMU_OT_USERREFS
, /* ZAP */
187 DMU_OT_DDT_ZAP
, /* ZAP */
188 DMU_OT_DDT_STATS
, /* ZAP */
189 DMU_OT_SA
, /* System attr */
190 DMU_OT_SA_MASTER_NODE
, /* ZAP */
191 DMU_OT_SA_ATTR_REGISTRATION
, /* ZAP */
192 DMU_OT_SA_ATTR_LAYOUTS
, /* ZAP */
193 DMU_OT_SCAN_XLATE
, /* ZAP */
194 DMU_OT_DEDUP
, /* fake dedup BP from ddt_bp_create() */
195 DMU_OT_DEADLIST
, /* ZAP */
196 DMU_OT_DEADLIST_HDR
, /* UINT64 */
197 DMU_OT_DSL_CLONES
, /* ZAP */
198 DMU_OT_BPOBJ_SUBOBJ
, /* UINT64 */
200 * Do not allocate new object types here. Doing so makes the on-disk
201 * format incompatible with any other format that uses the same object
204 * When creating an object which does not have one of the above types
205 * use the DMU_OTN_* type with the correct byteswap and metadata
208 * The DMU_OTN_* types do not have entries in the dmu_ot table,
209 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
210 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
211 * and DMU_OTN_* types).
216 * Names for valid types declared with DMU_OT().
218 DMU_OTN_UINT8_DATA
= DMU_OT(DMU_BSWAP_UINT8
, B_FALSE
),
219 DMU_OTN_UINT8_METADATA
= DMU_OT(DMU_BSWAP_UINT8
, B_TRUE
),
220 DMU_OTN_UINT16_DATA
= DMU_OT(DMU_BSWAP_UINT16
, B_FALSE
),
221 DMU_OTN_UINT16_METADATA
= DMU_OT(DMU_BSWAP_UINT16
, B_TRUE
),
222 DMU_OTN_UINT32_DATA
= DMU_OT(DMU_BSWAP_UINT32
, B_FALSE
),
223 DMU_OTN_UINT32_METADATA
= DMU_OT(DMU_BSWAP_UINT32
, B_TRUE
),
224 DMU_OTN_UINT64_DATA
= DMU_OT(DMU_BSWAP_UINT64
, B_FALSE
),
225 DMU_OTN_UINT64_METADATA
= DMU_OT(DMU_BSWAP_UINT64
, B_TRUE
),
226 DMU_OTN_ZAP_DATA
= DMU_OT(DMU_BSWAP_ZAP
, B_FALSE
),
227 DMU_OTN_ZAP_METADATA
= DMU_OT(DMU_BSWAP_ZAP
, B_TRUE
),
231 * These flags are intended to be used to specify the "txg_how"
232 * parameter when calling the dmu_tx_assign() function. See the comment
233 * above dmu_tx_assign() for more details on the meaning of these flags.
235 #define TXG_NOWAIT (0ULL)
236 #define TXG_WAIT (1ULL<<0)
237 #define TXG_NOTHROTTLE (1ULL<<1)
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 (64 * 1024 * 1024) /* 64MB */
257 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
259 #define DMU_USERUSED_OBJECT (-1ULL)
260 #define DMU_GROUPUSED_OBJECT (-2ULL)
263 * Zap prefix for object accounting in DMU_{USER,GROUP}USED_OBJECT.
265 #define DMU_OBJACCT_PREFIX "obj-"
266 #define DMU_OBJACCT_PREFIX_LEN 4
269 * artificial blkids for bonus buffer and spill blocks
271 #define DMU_BONUS_BLKID (-1ULL)
272 #define DMU_SPILL_BLKID (-2ULL)
274 * Public routines to create, destroy, open, and close objsets.
276 int dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
);
277 int dmu_objset_own(const char *name
, dmu_objset_type_t type
,
278 boolean_t readonly
, void *tag
, objset_t
**osp
);
279 void dmu_objset_rele(objset_t
*os
, void *tag
);
280 void dmu_objset_disown(objset_t
*os
, void *tag
);
281 int dmu_objset_open_ds(struct dsl_dataset
*ds
, objset_t
**osp
);
283 void dmu_objset_evict_dbufs(objset_t
*os
);
284 int dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
285 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
);
286 int dmu_objset_clone(const char *name
, const char *origin
);
287 int dsl_destroy_snapshots_nvl(struct nvlist
*snaps
, boolean_t defer
,
288 struct nvlist
*errlist
);
289 int dmu_objset_snapshot_one(const char *fsname
, const char *snapname
);
290 int dmu_objset_snapshot_tmp(const char *, const char *, int);
291 int dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
293 void dmu_objset_byteswap(void *buf
, size_t size
);
294 int dsl_dataset_rename_snapshot(const char *fsname
,
295 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
);
297 typedef struct dmu_buf
{
298 uint64_t db_object
; /* object that this buffer is part of */
299 uint64_t db_offset
; /* byte offset in this object */
300 uint64_t db_size
; /* size of buffer in bytes */
301 void *db_data
; /* data in buffer */
305 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
307 #define DMU_POOL_DIRECTORY_OBJECT 1
308 #define DMU_POOL_CONFIG "config"
309 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
310 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
311 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
312 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
313 #define DMU_POOL_ROOT_DATASET "root_dataset"
314 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
315 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
316 #define DMU_POOL_ERRLOG_LAST "errlog_last"
317 #define DMU_POOL_SPARES "spares"
318 #define DMU_POOL_DEFLATE "deflate"
319 #define DMU_POOL_HISTORY "history"
320 #define DMU_POOL_PROPS "pool_props"
321 #define DMU_POOL_L2CACHE "l2cache"
322 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
323 #define DMU_POOL_DDT "DDT-%s-%s-%s"
324 #define DMU_POOL_DDT_STATS "DDT-statistics"
325 #define DMU_POOL_CREATION_VERSION "creation_version"
326 #define DMU_POOL_SCAN "scan"
327 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
328 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
329 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
330 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
331 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
334 * Allocate an object from this objset. The range of object numbers
335 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
337 * The transaction must be assigned to a txg. The newly allocated
338 * object will be "held" in the transaction (ie. you can modify the
339 * newly allocated object in this transaction).
341 * dmu_object_alloc() chooses an object and returns it in *objectp.
343 * dmu_object_claim() allocates a specific object number. If that
344 * number is already allocated, it fails and returns EEXIST.
346 * Return 0 on success, or ENOSPC or EEXIST as specified above.
348 uint64_t dmu_object_alloc(objset_t
*os
, dmu_object_type_t ot
,
349 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
350 uint64_t dmu_object_alloc_dnsize(objset_t
*os
, dmu_object_type_t ot
,
351 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
,
352 int dnodesize
, dmu_tx_t
*tx
);
353 int dmu_object_claim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
354 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
355 int dmu_object_claim_dnsize(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
356 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
,
357 int dnodesize
, dmu_tx_t
*tx
);
358 int dmu_object_reclaim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
359 int blocksize
, dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*txp
);
360 int dmu_object_reclaim_dnsize(objset_t
*os
, uint64_t object
,
361 dmu_object_type_t ot
, int blocksize
, dmu_object_type_t bonustype
,
362 int bonuslen
, int dnodesize
, dmu_tx_t
*txp
);
365 * Free an object from this objset.
367 * The object's data will be freed as well (ie. you don't need to call
368 * dmu_free(object, 0, -1, tx)).
370 * The object need not be held in the transaction.
372 * If there are any holds on this object's buffers (via dmu_buf_hold()),
373 * or tx holds on the object (via dmu_tx_hold_object()), you can not
374 * free it; it fails and returns EBUSY.
376 * If the object is not allocated, it fails and returns ENOENT.
378 * Return 0 on success, or EBUSY or ENOENT as specified above.
380 int dmu_object_free(objset_t
*os
, uint64_t object
, dmu_tx_t
*tx
);
383 * Find the next allocated or free object.
385 * The objectp parameter is in-out. It will be updated to be the next
386 * object which is allocated. Ignore objects which have not been
387 * modified since txg.
389 * XXX Can only be called on a objset with no dirty data.
391 * Returns 0 on success, or ENOENT if there are no more objects.
393 int dmu_object_next(objset_t
*os
, uint64_t *objectp
,
394 boolean_t hole
, uint64_t txg
);
397 * Set the data blocksize for an object.
399 * The object cannot have any blocks allcated beyond the first. If
400 * the first block is allocated already, the new size must be greater
401 * than the current block size. If these conditions are not met,
402 * ENOTSUP will be returned.
404 * Returns 0 on success, or EBUSY if there are any holds on the object
405 * contents, or ENOTSUP as described above.
407 int dmu_object_set_blocksize(objset_t
*os
, uint64_t object
, uint64_t size
,
408 int ibs
, dmu_tx_t
*tx
);
411 * Set the checksum property on a dnode. The new checksum algorithm will
412 * apply to all newly written blocks; existing blocks will not be affected.
414 void dmu_object_set_checksum(objset_t
*os
, uint64_t object
, uint8_t checksum
,
418 * Set the compress property on a dnode. The new compression algorithm will
419 * apply to all newly written blocks; existing blocks will not be affected.
421 void dmu_object_set_compress(objset_t
*os
, uint64_t object
, uint8_t compress
,
425 dmu_write_embedded(objset_t
*os
, uint64_t object
, uint64_t offset
,
426 void *data
, uint8_t etype
, uint8_t comp
, int uncompressed_size
,
427 int compressed_size
, int byteorder
, dmu_tx_t
*tx
);
430 * Decide how to write a block: checksum, compression, number of copies, etc.
432 #define WP_NOFILL 0x1
433 #define WP_DMU_SYNC 0x2
436 void dmu_write_policy(objset_t
*os
, dnode_t
*dn
, int level
, int wp
,
437 struct zio_prop
*zp
);
439 * The bonus data is accessed more or less like a regular buffer.
440 * You must dmu_bonus_hold() to get the buffer, which will give you a
441 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
442 * data. As with any normal buffer, you must call dmu_buf_read() to
443 * read db_data, dmu_buf_will_dirty() before modifying it, and the
444 * object must be held in an assigned transaction before calling
445 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
446 * buffer as well. You must release what you hold with dmu_buf_rele().
448 * Returns ENOENT, EIO, or 0.
450 int dmu_bonus_hold(objset_t
*os
, uint64_t object
, void *tag
, dmu_buf_t
**);
451 int dmu_bonus_max(void);
452 int dmu_set_bonus(dmu_buf_t
*, int, dmu_tx_t
*);
453 int dmu_set_bonustype(dmu_buf_t
*, dmu_object_type_t
, dmu_tx_t
*);
454 dmu_object_type_t
dmu_get_bonustype(dmu_buf_t
*);
455 int dmu_rm_spill(objset_t
*, uint64_t, dmu_tx_t
*);
458 * Special spill buffer support used by "SA" framework
461 int dmu_spill_hold_by_bonus(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
);
462 int dmu_spill_hold_by_dnode(dnode_t
*dn
, uint32_t flags
,
463 void *tag
, dmu_buf_t
**dbp
);
464 int dmu_spill_hold_existing(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
);
467 * Obtain the DMU buffer from the specified object which contains the
468 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
469 * that it will remain in memory. You must release the hold with
470 * dmu_buf_rele(). You must not access the dmu_buf_t after releasing
471 * what you hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
473 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
474 * on the returned buffer before reading or writing the buffer's
475 * db_data. The comments for those routines describe what particular
476 * operations are valid after calling them.
478 * The object number must be a valid, allocated object number.
480 int dmu_buf_hold(objset_t
*os
, uint64_t object
, uint64_t offset
,
481 void *tag
, dmu_buf_t
**, int flags
);
482 int dmu_buf_hold_by_dnode(dnode_t
*dn
, uint64_t offset
,
483 void *tag
, dmu_buf_t
**dbp
, int flags
);
486 * Add a reference to a dmu buffer that has already been held via
487 * dmu_buf_hold() in the current context.
489 void dmu_buf_add_ref(dmu_buf_t
*db
, void* tag
);
492 * Attempt to add a reference to a dmu buffer that is in an unknown state,
493 * using a pointer that may have been invalidated by eviction processing.
494 * The request will succeed if the passed in dbuf still represents the
495 * same os/object/blkid, is ineligible for eviction, and has at least
496 * one hold by a user other than the syncer.
498 boolean_t
dmu_buf_try_add_ref(dmu_buf_t
*, objset_t
*os
, uint64_t object
,
499 uint64_t blkid
, void *tag
);
501 void dmu_buf_rele(dmu_buf_t
*db
, void *tag
);
502 uint64_t dmu_buf_refcount(dmu_buf_t
*db
);
505 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
506 * range of an object. A pointer to an array of dmu_buf_t*'s is
507 * returned (in *dbpp).
509 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
510 * frees the array. The hold on the array of buffers MUST be released
511 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
512 * individually with dmu_buf_rele.
514 int dmu_buf_hold_array_by_bonus(dmu_buf_t
*db
, uint64_t offset
,
515 uint64_t length
, boolean_t read
, void *tag
,
516 int *numbufsp
, dmu_buf_t
***dbpp
);
517 void dmu_buf_rele_array(dmu_buf_t
**, int numbufs
, void *tag
);
519 typedef void dmu_buf_evict_func_t(void *user_ptr
);
522 * A DMU buffer user object may be associated with a dbuf for the
523 * duration of its lifetime. This allows the user of a dbuf (client)
524 * to attach private data to a dbuf (e.g. in-core only data such as a
525 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
526 * when that dbuf has been evicted. Clients typically respond to the
527 * eviction notification by freeing their private data, thus ensuring
528 * the same lifetime for both dbuf and private data.
530 * The mapping from a dmu_buf_user_t to any client private data is the
531 * client's responsibility. All current consumers of the API with private
532 * data embed a dmu_buf_user_t as the first member of the structure for
533 * their private data. This allows conversions between the two types
534 * with a simple cast. Since the DMU buf user API never needs access
535 * to the private data, other strategies can be employed if necessary
536 * or convenient for the client (e.g. using container_of() to do the
537 * conversion for private data that cannot have the dmu_buf_user_t as
540 * Eviction callbacks are executed without the dbuf mutex held or any
541 * other type of mechanism to guarantee that the dbuf is still available.
542 * For this reason, users must assume the dbuf has already been freed
543 * and not reference the dbuf from the callback context.
545 * Users requesting "immediate eviction" are notified as soon as the dbuf
546 * is only referenced by dirty records (dirties == holds). Otherwise the
547 * notification occurs after eviction processing for the dbuf begins.
549 typedef struct dmu_buf_user
{
551 * Asynchronous user eviction callback state.
553 taskq_ent_t dbu_tqent
;
556 * This instance's eviction function pointers.
558 * dbu_evict_func_sync is called synchronously and then
559 * dbu_evict_func_async is executed asynchronously on a taskq.
561 dmu_buf_evict_func_t
*dbu_evict_func_sync
;
562 dmu_buf_evict_func_t
*dbu_evict_func_async
;
565 * Pointer to user's dbuf pointer. NULL for clients that do
566 * not associate a dbuf with their user data.
568 * The dbuf pointer is cleared upon eviction so as to catch
569 * use-after-evict bugs in clients.
571 dmu_buf_t
**dbu_clear_on_evict_dbufp
;
576 * Initialize the given dmu_buf_user_t instance with the eviction function
577 * evict_func, to be called when the user is evicted.
579 * NOTE: This function should only be called once on a given dmu_buf_user_t.
580 * To allow enforcement of this, dbu must already be zeroed on entry.
584 dmu_buf_init_user(dmu_buf_user_t
*dbu
, dmu_buf_evict_func_t
*evict_func_sync
,
585 dmu_buf_evict_func_t
*evict_func_async
, dmu_buf_t
**clear_on_evict_dbufp
)
587 ASSERT(dbu
->dbu_evict_func_sync
== NULL
);
588 ASSERT(dbu
->dbu_evict_func_async
== NULL
);
590 /* must have at least one evict func */
591 IMPLY(evict_func_sync
== NULL
, evict_func_async
!= NULL
);
592 dbu
->dbu_evict_func_sync
= evict_func_sync
;
593 dbu
->dbu_evict_func_async
= evict_func_async
;
594 taskq_init_ent(&dbu
->dbu_tqent
);
596 dbu
->dbu_clear_on_evict_dbufp
= clear_on_evict_dbufp
;
601 * Attach user data to a dbuf and mark it for normal (when the dbuf's
602 * data is cleared or its reference count goes to zero) eviction processing.
604 * Returns NULL on success, or the existing user if another user currently
607 void *dmu_buf_set_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
610 * Attach user data to a dbuf and mark it for immediate (its dirty and
611 * reference counts are equal) eviction processing.
613 * Returns NULL on success, or the existing user if another user currently
616 void *dmu_buf_set_user_ie(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
619 * Replace the current user of a dbuf.
621 * If given the current user of a dbuf, replaces the dbuf's user with
622 * "new_user" and returns the user data pointer that was replaced.
623 * Otherwise returns the current, and unmodified, dbuf user pointer.
625 void *dmu_buf_replace_user(dmu_buf_t
*db
,
626 dmu_buf_user_t
*old_user
, dmu_buf_user_t
*new_user
);
629 * Remove the specified user data for a DMU buffer.
631 * Returns the user that was removed on success, or the current user if
632 * another user currently owns the buffer.
634 void *dmu_buf_remove_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
637 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
639 void *dmu_buf_get_user(dmu_buf_t
*db
);
641 objset_t
*dmu_buf_get_objset(dmu_buf_t
*db
);
642 dnode_t
*dmu_buf_dnode_enter(dmu_buf_t
*db
);
643 void dmu_buf_dnode_exit(dmu_buf_t
*db
);
645 /* Block until any in-progress dmu buf user evictions complete. */
646 void dmu_buf_user_evict_wait(void);
649 * Returns the blkptr associated with this dbuf, or NULL if not set.
651 struct blkptr
*dmu_buf_get_blkptr(dmu_buf_t
*db
);
654 * Indicate that you are going to modify the buffer's data (db_data).
656 * The transaction (tx) must be assigned to a txg (ie. you've called
657 * dmu_tx_assign()). The buffer's object must be held in the tx
658 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
660 void dmu_buf_will_dirty(dmu_buf_t
*db
, dmu_tx_t
*tx
);
663 * You must create a transaction, then hold the objects which you will
664 * (or might) modify as part of this transaction. Then you must assign
665 * the transaction to a transaction group. Once the transaction has
666 * been assigned, you can modify buffers which belong to held objects as
667 * part of this transaction. You can't modify buffers before the
668 * transaction has been assigned; you can't modify buffers which don't
669 * belong to objects which this transaction holds; you can't hold
670 * objects once the transaction has been assigned. You may hold an
671 * object which you are going to free (with dmu_object_free()), but you
674 * You can abort the transaction before it has been assigned.
676 * Note that you may hold buffers (with dmu_buf_hold) at any time,
677 * regardless of transaction state.
680 #define DMU_NEW_OBJECT (-1ULL)
681 #define DMU_OBJECT_END (-1ULL)
683 dmu_tx_t
*dmu_tx_create(objset_t
*os
);
684 void dmu_tx_hold_write(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
, int len
);
685 void dmu_tx_hold_write_by_dnode(dmu_tx_t
*tx
, dnode_t
*dn
, uint64_t off
,
687 void dmu_tx_hold_free(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
,
689 void dmu_tx_hold_free_by_dnode(dmu_tx_t
*tx
, dnode_t
*dn
, uint64_t off
,
691 void dmu_tx_hold_zap(dmu_tx_t
*tx
, uint64_t object
, int add
, const char *name
);
692 void dmu_tx_hold_zap_by_dnode(dmu_tx_t
*tx
, dnode_t
*dn
, int add
,
694 void dmu_tx_hold_bonus(dmu_tx_t
*tx
, uint64_t object
);
695 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t
*tx
, dnode_t
*dn
);
696 void dmu_tx_hold_spill(dmu_tx_t
*tx
, uint64_t object
);
697 void dmu_tx_hold_sa(dmu_tx_t
*tx
, struct sa_handle
*hdl
, boolean_t may_grow
);
698 void dmu_tx_hold_sa_create(dmu_tx_t
*tx
, int total_size
);
699 void dmu_tx_abort(dmu_tx_t
*tx
);
700 int dmu_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
);
701 void dmu_tx_wait(dmu_tx_t
*tx
);
702 void dmu_tx_commit(dmu_tx_t
*tx
);
703 void dmu_tx_mark_netfree(dmu_tx_t
*tx
);
706 * To register a commit callback, dmu_tx_callback_register() must be called.
708 * dcb_data is a pointer to caller private data that is passed on as a
709 * callback parameter. The caller is responsible for properly allocating and
712 * When registering a callback, the transaction must be already created, but
713 * it cannot be committed or aborted. It can be assigned to a txg or not.
715 * The callback will be called after the transaction has been safely written
716 * to stable storage and will also be called if the dmu_tx is aborted.
717 * If there is any error which prevents the transaction from being committed to
718 * disk, the callback will be called with a value of error != 0.
720 * When multiple callbacks are registered to the transaction, the callbacks
721 * will be called in reverse order to let Lustre, the only user of commit
722 * callback currently, take the fast path of its commit callback handling.
724 typedef void dmu_tx_callback_func_t(void *dcb_data
, int error
);
726 void dmu_tx_callback_register(dmu_tx_t
*tx
, dmu_tx_callback_func_t
*dcb_func
,
728 void dmu_tx_do_callbacks(list_t
*cb_list
, int error
);
731 * Free up the data blocks for a defined range of a file. If size is
732 * -1, the range from offset to end-of-file is freed.
734 int dmu_free_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
735 uint64_t size
, dmu_tx_t
*tx
);
736 int dmu_free_long_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
738 int dmu_free_long_object(objset_t
*os
, uint64_t object
);
741 * Convenience functions.
743 * Canfail routines will return 0 on success, or an errno if there is a
744 * nonrecoverable I/O error.
746 #define DMU_READ_PREFETCH 0 /* prefetch */
747 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
748 int dmu_read(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
749 void *buf
, uint32_t flags
);
750 int dmu_read_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
, void *buf
,
752 void dmu_write(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
753 const void *buf
, dmu_tx_t
*tx
);
754 void dmu_write_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
,
755 const void *buf
, dmu_tx_t
*tx
);
756 void dmu_prealloc(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
759 #include <linux/blkdev_compat.h>
760 int dmu_read_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
);
761 int dmu_read_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
);
762 int dmu_read_uio_dnode(dnode_t
*dn
, struct uio
*uio
, uint64_t size
);
763 int dmu_write_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
,
765 int dmu_write_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
,
767 int dmu_write_uio_dnode(dnode_t
*dn
, struct uio
*uio
, uint64_t size
,
770 struct arc_buf
*dmu_request_arcbuf(dmu_buf_t
*handle
, int size
);
771 void dmu_return_arcbuf(struct arc_buf
*buf
);
772 void dmu_assign_arcbuf(dmu_buf_t
*handle
, uint64_t offset
, struct arc_buf
*buf
,
774 #ifdef HAVE_UIO_ZEROCOPY
775 int dmu_xuio_init(struct xuio
*uio
, int niov
);
776 void dmu_xuio_fini(struct xuio
*uio
);
777 int dmu_xuio_add(struct xuio
*uio
, struct arc_buf
*abuf
, offset_t off
,
779 int dmu_xuio_cnt(struct xuio
*uio
);
780 struct arc_buf
*dmu_xuio_arcbuf(struct xuio
*uio
, int i
);
781 void dmu_xuio_clear(struct xuio
*uio
, int i
);
782 #endif /* HAVE_UIO_ZEROCOPY */
783 void xuio_stat_wbuf_copied(void);
784 void xuio_stat_wbuf_nocopy(void);
786 extern int zfs_prefetch_disable
;
787 extern int zfs_max_recordsize
;
790 * Asynchronously try to read in the data.
792 void dmu_prefetch(objset_t
*os
, uint64_t object
, int64_t level
, uint64_t offset
,
793 uint64_t len
, enum zio_priority pri
);
795 typedef struct dmu_object_info
{
796 /* All sizes are in bytes unless otherwise indicated. */
797 uint32_t doi_data_block_size
;
798 uint32_t doi_metadata_block_size
;
799 dmu_object_type_t doi_type
;
800 dmu_object_type_t doi_bonus_type
;
801 uint64_t doi_bonus_size
;
802 uint8_t doi_indirection
; /* 2 = dnode->indirect->data */
803 uint8_t doi_checksum
;
804 uint8_t doi_compress
;
807 uint64_t doi_dnodesize
;
808 uint64_t doi_physical_blocks_512
; /* data + metadata, 512b blks */
809 uint64_t doi_max_offset
;
810 uint64_t doi_fill_count
; /* number of non-empty blocks */
813 typedef void (*const arc_byteswap_func_t
)(void *buf
, size_t size
);
815 typedef struct dmu_object_type_info
{
816 dmu_object_byteswap_t ot_byteswap
;
817 boolean_t ot_metadata
;
819 } dmu_object_type_info_t
;
821 typedef const struct dmu_object_byteswap_info
{
822 arc_byteswap_func_t ob_func
;
824 } dmu_object_byteswap_info_t
;
826 extern const dmu_object_type_info_t dmu_ot
[DMU_OT_NUMTYPES
];
827 extern const dmu_object_byteswap_info_t dmu_ot_byteswap
[DMU_BSWAP_NUMFUNCS
];
830 * Get information on a DMU object.
832 * Return 0 on success or ENOENT if object is not allocated.
834 * If doi is NULL, just indicates whether the object exists.
836 int dmu_object_info(objset_t
*os
, uint64_t object
, dmu_object_info_t
*doi
);
837 void __dmu_object_info_from_dnode(struct dnode
*dn
, dmu_object_info_t
*doi
);
838 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
839 void dmu_object_info_from_dnode(dnode_t
*dn
, dmu_object_info_t
*doi
);
840 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
841 void dmu_object_info_from_db(dmu_buf_t
*db
, dmu_object_info_t
*doi
);
843 * Like dmu_object_info_from_db, but faster still when you only care about
844 * the size. This is specifically optimized for zfs_getattr().
846 void dmu_object_size_from_db(dmu_buf_t
*db
, uint32_t *blksize
,
847 u_longlong_t
*nblk512
);
849 void dmu_object_dnsize_from_db(dmu_buf_t
*db
, int *dnsize
);
851 typedef struct dmu_objset_stats
{
852 uint64_t dds_num_clones
; /* number of clones of this */
853 uint64_t dds_creation_txg
;
855 dmu_objset_type_t dds_type
;
856 uint8_t dds_is_snapshot
;
857 uint8_t dds_inconsistent
;
858 char dds_origin
[ZFS_MAX_DATASET_NAME_LEN
];
859 } dmu_objset_stats_t
;
862 * Get stats on a dataset.
864 void dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
);
867 * Add entries to the nvlist for all the objset's properties. See
868 * zfs_prop_table[] and zfs(1m) for details on the properties.
870 void dmu_objset_stats(objset_t
*os
, struct nvlist
*nv
);
873 * Get the space usage statistics for statvfs().
875 * refdbytes is the amount of space "referenced" by this objset.
876 * availbytes is the amount of space available to this objset, taking
877 * into account quotas & reservations, assuming that no other objsets
878 * use the space first. These values correspond to the 'referenced' and
879 * 'available' properties, described in the zfs(1m) manpage.
881 * usedobjs and availobjs are the number of objects currently allocated,
884 void dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
885 uint64_t *usedobjsp
, uint64_t *availobjsp
);
888 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
889 * (Contrast with the ds_guid which is a 64-bit ID that will never
890 * change, so there is a small probability that it will collide.)
892 uint64_t dmu_objset_fsid_guid(objset_t
*os
);
895 * Get the [cm]time for an objset's snapshot dir
897 inode_timespec_t
dmu_objset_snap_cmtime(objset_t
*os
);
899 int dmu_objset_is_snapshot(objset_t
*os
);
901 extern struct spa
*dmu_objset_spa(objset_t
*os
);
902 extern struct zilog
*dmu_objset_zil(objset_t
*os
);
903 extern struct dsl_pool
*dmu_objset_pool(objset_t
*os
);
904 extern struct dsl_dataset
*dmu_objset_ds(objset_t
*os
);
905 extern void dmu_objset_name(objset_t
*os
, char *buf
);
906 extern dmu_objset_type_t
dmu_objset_type(objset_t
*os
);
907 extern uint64_t dmu_objset_id(objset_t
*os
);
908 extern uint64_t dmu_objset_dnodesize(objset_t
*os
);
909 extern zfs_sync_type_t
dmu_objset_syncprop(objset_t
*os
);
910 extern zfs_logbias_op_t
dmu_objset_logbias(objset_t
*os
);
911 extern int dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
912 uint64_t *id
, uint64_t *offp
, boolean_t
*case_conflict
);
913 extern int dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *val
);
914 extern int dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
,
915 int maxlen
, boolean_t
*conflict
);
916 extern int dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
917 uint64_t *idp
, uint64_t *offp
);
919 typedef int objset_used_cb_t(dmu_object_type_t bonustype
,
920 void *bonus
, uint64_t *userp
, uint64_t *groupp
);
921 extern void dmu_objset_register_type(dmu_objset_type_t ost
,
922 objset_used_cb_t
*cb
);
923 extern void dmu_objset_set_user(objset_t
*os
, void *user_ptr
);
924 extern void *dmu_objset_get_user(objset_t
*os
);
927 * Return the txg number for the given assigned transaction.
929 uint64_t dmu_tx_get_txg(dmu_tx_t
*tx
);
933 * If a parent zio is provided this function initiates a write on the
934 * provided buffer as a child of the parent zio.
935 * In the absence of a parent zio, the write is completed synchronously.
936 * At write completion, blk is filled with the bp of the written block.
937 * Note that while the data covered by this function will be on stable
938 * storage when the write completes this new data does not become a
939 * permanent part of the file until the associated transaction commits.
943 * {zfs,zvol,ztest}_get_done() args
946 struct zilog
*zgd_zilog
;
947 struct blkptr
*zgd_bp
;
953 typedef void dmu_sync_cb_t(zgd_t
*arg
, int error
);
954 int dmu_sync(struct zio
*zio
, uint64_t txg
, dmu_sync_cb_t
*done
, zgd_t
*zgd
);
957 * Find the next hole or data block in file starting at *off
958 * Return found offset in *off. Return ESRCH for end of file.
960 int dmu_offset_next(objset_t
*os
, uint64_t object
, boolean_t hole
,
964 * Initial setup and final teardown.
966 extern void dmu_init(void);
967 extern void dmu_fini(void);
969 typedef void (*dmu_traverse_cb_t
)(objset_t
*os
, void *arg
, struct blkptr
*bp
,
970 uint64_t object
, uint64_t offset
, int len
);
971 void dmu_traverse_objset(objset_t
*os
, uint64_t txg_start
,
972 dmu_traverse_cb_t cb
, void *arg
);
974 int dmu_diff(const char *tosnap_name
, const char *fromsnap_name
,
975 struct vnode
*vp
, offset_t
*offp
);
978 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
979 extern uint64_t zfs_crc64_table
[256];
981 extern int zfs_mdcomp_disable
;
987 #endif /* _SYS_DMU_H */