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
;
74 struct dsl_crypto_params
;
76 typedef struct objset objset_t
;
77 typedef struct dmu_tx dmu_tx_t
;
78 typedef struct dsl_dir dsl_dir_t
;
79 typedef struct dnode dnode_t
;
81 typedef enum dmu_object_byteswap
{
93 * Allocating a new byteswap type number makes the on-disk format
94 * incompatible with any other format that uses the same number.
96 * Data can usually be structured to work with one of the
97 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
100 } dmu_object_byteswap_t
;
102 #define DMU_OT_NEWTYPE 0x80
103 #define DMU_OT_METADATA 0x40
104 #define DMU_OT_ENCRYPTED 0x20
105 #define DMU_OT_BYTESWAP_MASK 0x1f
108 * Defines a uint8_t object type. Object types specify if the data
109 * in the object is metadata (boolean) and how to byteswap the data
110 * (dmu_object_byteswap_t).
112 #define DMU_OT(byteswap, metadata, encrypted) \
114 ((metadata) ? DMU_OT_METADATA : 0) | \
115 ((encrypted) ? DMU_OT_ENCRYPTED : 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)
123 * MDB doesn't have dmu_ot; it defines these macros itself.
126 #define DMU_OT_IS_METADATA_IMPL(ot) (dmu_ot[ot].ot_metadata)
127 #define DMU_OT_IS_ENCRYPTED_IMPL(ot) (dmu_ot[ot].ot_encrypt)
128 #define DMU_OT_BYTESWAP_IMPL(ot) (dmu_ot[ot].ot_byteswap)
131 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
132 ((ot) & DMU_OT_METADATA) : \
133 DMU_OT_IS_METADATA_IMPL(ot))
135 #define DMU_OT_IS_ENCRYPTED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
136 ((ot) & DMU_OT_ENCRYPTED) : \
137 DMU_OT_IS_ENCRYPTED_IMPL(ot))
140 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
141 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
142 * is repurposed for embedded BPs.
144 #define DMU_OT_HAS_FILL(ot) \
145 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
147 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
148 ((ot) & DMU_OT_BYTESWAP_MASK) : \
149 DMU_OT_BYTESWAP_IMPL(ot))
151 typedef enum dmu_object_type
{
154 DMU_OT_OBJECT_DIRECTORY
, /* ZAP */
155 DMU_OT_OBJECT_ARRAY
, /* UINT64 */
156 DMU_OT_PACKED_NVLIST
, /* UINT8 (XDR by nvlist_pack/unpack) */
157 DMU_OT_PACKED_NVLIST_SIZE
, /* UINT64 */
158 DMU_OT_BPOBJ
, /* UINT64 */
159 DMU_OT_BPOBJ_HDR
, /* UINT64 */
161 DMU_OT_SPACE_MAP_HEADER
, /* UINT64 */
162 DMU_OT_SPACE_MAP
, /* UINT64 */
164 DMU_OT_INTENT_LOG
, /* UINT64 */
166 DMU_OT_DNODE
, /* DNODE */
167 DMU_OT_OBJSET
, /* OBJSET */
169 DMU_OT_DSL_DIR
, /* UINT64 */
170 DMU_OT_DSL_DIR_CHILD_MAP
, /* ZAP */
171 DMU_OT_DSL_DS_SNAP_MAP
, /* ZAP */
172 DMU_OT_DSL_PROPS
, /* ZAP */
173 DMU_OT_DSL_DATASET
, /* UINT64 */
175 DMU_OT_ZNODE
, /* ZNODE */
176 DMU_OT_OLDACL
, /* Old ACL */
177 DMU_OT_PLAIN_FILE_CONTENTS
, /* UINT8 */
178 DMU_OT_DIRECTORY_CONTENTS
, /* ZAP */
179 DMU_OT_MASTER_NODE
, /* ZAP */
180 DMU_OT_UNLINKED_SET
, /* ZAP */
182 DMU_OT_ZVOL
, /* UINT8 */
183 DMU_OT_ZVOL_PROP
, /* ZAP */
184 /* other; for testing only! */
185 DMU_OT_PLAIN_OTHER
, /* UINT8 */
186 DMU_OT_UINT64_OTHER
, /* UINT64 */
187 DMU_OT_ZAP_OTHER
, /* ZAP */
188 /* new object types: */
189 DMU_OT_ERROR_LOG
, /* ZAP */
190 DMU_OT_SPA_HISTORY
, /* UINT8 */
191 DMU_OT_SPA_HISTORY_OFFSETS
, /* spa_his_phys_t */
192 DMU_OT_POOL_PROPS
, /* ZAP */
193 DMU_OT_DSL_PERMS
, /* ZAP */
194 DMU_OT_ACL
, /* ACL */
195 DMU_OT_SYSACL
, /* SYSACL */
196 DMU_OT_FUID
, /* FUID table (Packed NVLIST UINT8) */
197 DMU_OT_FUID_SIZE
, /* FUID table size UINT64 */
198 DMU_OT_NEXT_CLONES
, /* ZAP */
199 DMU_OT_SCAN_QUEUE
, /* ZAP */
200 DMU_OT_USERGROUP_USED
, /* ZAP */
201 DMU_OT_USERGROUP_QUOTA
, /* ZAP */
202 DMU_OT_USERREFS
, /* ZAP */
203 DMU_OT_DDT_ZAP
, /* ZAP */
204 DMU_OT_DDT_STATS
, /* ZAP */
205 DMU_OT_SA
, /* System attr */
206 DMU_OT_SA_MASTER_NODE
, /* ZAP */
207 DMU_OT_SA_ATTR_REGISTRATION
, /* ZAP */
208 DMU_OT_SA_ATTR_LAYOUTS
, /* ZAP */
209 DMU_OT_SCAN_XLATE
, /* ZAP */
210 DMU_OT_DEDUP
, /* fake dedup BP from ddt_bp_create() */
211 DMU_OT_DEADLIST
, /* ZAP */
212 DMU_OT_DEADLIST_HDR
, /* UINT64 */
213 DMU_OT_DSL_CLONES
, /* ZAP */
214 DMU_OT_BPOBJ_SUBOBJ
, /* UINT64 */
216 * Do not allocate new object types here. Doing so makes the on-disk
217 * format incompatible with any other format that uses the same object
220 * When creating an object which does not have one of the above types
221 * use the DMU_OTN_* type with the correct byteswap and metadata
224 * The DMU_OTN_* types do not have entries in the dmu_ot table,
225 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
226 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
227 * and DMU_OTN_* types).
232 * Names for valid types declared with DMU_OT().
234 DMU_OTN_UINT8_DATA
= DMU_OT(DMU_BSWAP_UINT8
, B_FALSE
, B_FALSE
),
235 DMU_OTN_UINT8_METADATA
= DMU_OT(DMU_BSWAP_UINT8
, B_TRUE
, B_FALSE
),
236 DMU_OTN_UINT16_DATA
= DMU_OT(DMU_BSWAP_UINT16
, B_FALSE
, B_FALSE
),
237 DMU_OTN_UINT16_METADATA
= DMU_OT(DMU_BSWAP_UINT16
, B_TRUE
, B_FALSE
),
238 DMU_OTN_UINT32_DATA
= DMU_OT(DMU_BSWAP_UINT32
, B_FALSE
, B_FALSE
),
239 DMU_OTN_UINT32_METADATA
= DMU_OT(DMU_BSWAP_UINT32
, B_TRUE
, B_FALSE
),
240 DMU_OTN_UINT64_DATA
= DMU_OT(DMU_BSWAP_UINT64
, B_FALSE
, B_FALSE
),
241 DMU_OTN_UINT64_METADATA
= DMU_OT(DMU_BSWAP_UINT64
, B_TRUE
, B_FALSE
),
242 DMU_OTN_ZAP_DATA
= DMU_OT(DMU_BSWAP_ZAP
, B_FALSE
, B_FALSE
),
243 DMU_OTN_ZAP_METADATA
= DMU_OT(DMU_BSWAP_ZAP
, B_TRUE
, B_FALSE
),
245 DMU_OTN_UINT8_ENC_DATA
= DMU_OT(DMU_BSWAP_UINT8
, B_FALSE
, B_TRUE
),
246 DMU_OTN_UINT8_ENC_METADATA
= DMU_OT(DMU_BSWAP_UINT8
, B_TRUE
, B_TRUE
),
247 DMU_OTN_UINT16_ENC_DATA
= DMU_OT(DMU_BSWAP_UINT16
, B_FALSE
, B_TRUE
),
248 DMU_OTN_UINT16_ENC_METADATA
= DMU_OT(DMU_BSWAP_UINT16
, B_TRUE
, B_TRUE
),
249 DMU_OTN_UINT32_ENC_DATA
= DMU_OT(DMU_BSWAP_UINT32
, B_FALSE
, B_TRUE
),
250 DMU_OTN_UINT32_ENC_METADATA
= DMU_OT(DMU_BSWAP_UINT32
, B_TRUE
, B_TRUE
),
251 DMU_OTN_UINT64_ENC_DATA
= DMU_OT(DMU_BSWAP_UINT64
, B_FALSE
, B_TRUE
),
252 DMU_OTN_UINT64_ENC_METADATA
= DMU_OT(DMU_BSWAP_UINT64
, B_TRUE
, B_TRUE
),
253 DMU_OTN_ZAP_ENC_DATA
= DMU_OT(DMU_BSWAP_ZAP
, B_FALSE
, B_TRUE
),
254 DMU_OTN_ZAP_ENC_METADATA
= DMU_OT(DMU_BSWAP_ZAP
, B_TRUE
, B_TRUE
),
258 * These flags are intended to be used to specify the "txg_how"
259 * parameter when calling the dmu_tx_assign() function. See the comment
260 * above dmu_tx_assign() for more details on the meaning of these flags.
262 #define TXG_NOWAIT (0ULL)
263 #define TXG_WAIT (1ULL<<0)
264 #define TXG_NOTHROTTLE (1ULL<<1)
266 void byteswap_uint64_array(void *buf
, size_t size
);
267 void byteswap_uint32_array(void *buf
, size_t size
);
268 void byteswap_uint16_array(void *buf
, size_t size
);
269 void byteswap_uint8_array(void *buf
, size_t size
);
270 void zap_byteswap(void *buf
, size_t size
);
271 void zfs_oldacl_byteswap(void *buf
, size_t size
);
272 void zfs_acl_byteswap(void *buf
, size_t size
);
273 void zfs_znode_byteswap(void *buf
, size_t size
);
275 #define DS_FIND_SNAPSHOTS (1<<0)
276 #define DS_FIND_CHILDREN (1<<1)
277 #define DS_FIND_SERIALIZE (1<<2)
280 * The maximum number of bytes that can be accessed as part of one
281 * operation, including metadata.
283 #define DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
284 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
286 #define DMU_USERUSED_OBJECT (-1ULL)
287 #define DMU_GROUPUSED_OBJECT (-2ULL)
288 #define DMU_PROJECTUSED_OBJECT (-3ULL)
291 * Zap prefix for object accounting in DMU_{USER,GROUP,PROJECT}USED_OBJECT.
293 #define DMU_OBJACCT_PREFIX "obj-"
294 #define DMU_OBJACCT_PREFIX_LEN 4
297 * artificial blkids for bonus buffer and spill blocks
299 #define DMU_BONUS_BLKID (-1ULL)
300 #define DMU_SPILL_BLKID (-2ULL)
303 * Public routines to create, destroy, open, and close objsets.
305 typedef void dmu_objset_create_sync_func_t(objset_t
*os
, void *arg
,
306 cred_t
*cr
, dmu_tx_t
*tx
);
308 int dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
);
309 int dmu_objset_own(const char *name
, dmu_objset_type_t type
,
310 boolean_t readonly
, boolean_t key_required
, void *tag
, objset_t
**osp
);
311 void dmu_objset_rele(objset_t
*os
, void *tag
);
312 void dmu_objset_disown(objset_t
*os
, boolean_t key_required
, void *tag
);
313 int dmu_objset_open_ds(struct dsl_dataset
*ds
, objset_t
**osp
);
315 void dmu_objset_evict_dbufs(objset_t
*os
);
316 int dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
317 struct dsl_crypto_params
*dcp
, dmu_objset_create_sync_func_t func
,
319 int dmu_objset_clone(const char *name
, const char *origin
);
320 int dsl_destroy_snapshots_nvl(struct nvlist
*snaps
, boolean_t defer
,
321 struct nvlist
*errlist
);
322 int dmu_objset_snapshot_one(const char *fsname
, const char *snapname
);
323 int dmu_objset_snapshot_tmp(const char *, const char *, int);
324 int dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
326 void dmu_objset_byteswap(void *buf
, size_t size
);
327 int dsl_dataset_rename_snapshot(const char *fsname
,
328 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
);
329 int dmu_objset_remap_indirects(const char *fsname
);
331 typedef struct dmu_buf
{
332 uint64_t db_object
; /* object that this buffer is part of */
333 uint64_t db_offset
; /* byte offset in this object */
334 uint64_t db_size
; /* size of buffer in bytes */
335 void *db_data
; /* data in buffer */
339 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
341 #define DMU_POOL_DIRECTORY_OBJECT 1
342 #define DMU_POOL_CONFIG "config"
343 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
344 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
345 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
346 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
347 #define DMU_POOL_ROOT_DATASET "root_dataset"
348 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
349 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
350 #define DMU_POOL_ERRLOG_LAST "errlog_last"
351 #define DMU_POOL_SPARES "spares"
352 #define DMU_POOL_DEFLATE "deflate"
353 #define DMU_POOL_HISTORY "history"
354 #define DMU_POOL_PROPS "pool_props"
355 #define DMU_POOL_L2CACHE "l2cache"
356 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
357 #define DMU_POOL_DDT "DDT-%s-%s-%s"
358 #define DMU_POOL_DDT_STATS "DDT-statistics"
359 #define DMU_POOL_CREATION_VERSION "creation_version"
360 #define DMU_POOL_SCAN "scan"
361 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
362 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
363 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
364 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
365 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
366 #define DMU_POOL_REMOVING "com.delphix:removing"
367 #define DMU_POOL_OBSOLETE_BPOBJ "com.delphix:obsolete_bpobj"
368 #define DMU_POOL_CONDENSING_INDIRECT "com.delphix:condensing_indirect"
369 #define DMU_POOL_ZPOOL_CHECKPOINT "com.delphix:zpool_checkpoint"
372 * Allocate an object from this objset. The range of object numbers
373 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
375 * The transaction must be assigned to a txg. The newly allocated
376 * object will be "held" in the transaction (ie. you can modify the
377 * newly allocated object in this transaction).
379 * dmu_object_alloc() chooses an object and returns it in *objectp.
381 * dmu_object_claim() allocates a specific object number. If that
382 * number is already allocated, it fails and returns EEXIST.
384 * Return 0 on success, or ENOSPC or EEXIST as specified above.
386 uint64_t dmu_object_alloc(objset_t
*os
, dmu_object_type_t ot
,
387 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
388 uint64_t dmu_object_alloc_dnsize(objset_t
*os
, dmu_object_type_t ot
,
389 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
,
390 int dnodesize
, dmu_tx_t
*tx
);
391 int dmu_object_claim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
392 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
, dmu_tx_t
*tx
);
393 int dmu_object_claim_dnsize(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
394 int blocksize
, dmu_object_type_t bonus_type
, int bonus_len
,
395 int dnodesize
, dmu_tx_t
*tx
);
396 int dmu_object_reclaim(objset_t
*os
, uint64_t object
, dmu_object_type_t ot
,
397 int blocksize
, dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*txp
);
398 int dmu_object_reclaim_dnsize(objset_t
*os
, uint64_t object
,
399 dmu_object_type_t ot
, int blocksize
, dmu_object_type_t bonustype
,
400 int bonuslen
, int dnodesize
, dmu_tx_t
*txp
);
403 * Free an object from this objset.
405 * The object's data will be freed as well (ie. you don't need to call
406 * dmu_free(object, 0, -1, tx)).
408 * The object need not be held in the transaction.
410 * If there are any holds on this object's buffers (via dmu_buf_hold()),
411 * or tx holds on the object (via dmu_tx_hold_object()), you can not
412 * free it; it fails and returns EBUSY.
414 * If the object is not allocated, it fails and returns ENOENT.
416 * Return 0 on success, or EBUSY or ENOENT as specified above.
418 int dmu_object_free(objset_t
*os
, uint64_t object
, dmu_tx_t
*tx
);
421 * Find the next allocated or free object.
423 * The objectp parameter is in-out. It will be updated to be the next
424 * object which is allocated. Ignore objects which have not been
425 * modified since txg.
427 * XXX Can only be called on a objset with no dirty data.
429 * Returns 0 on success, or ENOENT if there are no more objects.
431 int dmu_object_next(objset_t
*os
, uint64_t *objectp
,
432 boolean_t hole
, uint64_t txg
);
435 * Set the number of levels on a dnode. nlevels must be greater than the
436 * current number of levels or an EINVAL will be returned.
438 int dmu_object_set_nlevels(objset_t
*os
, uint64_t object
, int nlevels
,
442 * Set the data blocksize for an object.
444 * The object cannot have any blocks allcated beyond the first. If
445 * the first block is allocated already, the new size must be greater
446 * than the current block size. If these conditions are not met,
447 * ENOTSUP will be returned.
449 * Returns 0 on success, or EBUSY if there are any holds on the object
450 * contents, or ENOTSUP as described above.
452 int dmu_object_set_blocksize(objset_t
*os
, uint64_t object
, uint64_t size
,
453 int ibs
, dmu_tx_t
*tx
);
456 * Manually set the maxblkid on a dnode. This will adjust nlevels accordingly
457 * to accommodate the change.
459 int dmu_object_set_maxblkid(objset_t
*os
, uint64_t object
, uint64_t maxblkid
,
463 * Set the checksum property on a dnode. The new checksum algorithm will
464 * apply to all newly written blocks; existing blocks will not be affected.
466 void dmu_object_set_checksum(objset_t
*os
, uint64_t object
, uint8_t checksum
,
470 * Set the compress property on a dnode. The new compression algorithm will
471 * apply to all newly written blocks; existing blocks will not be affected.
473 void dmu_object_set_compress(objset_t
*os
, uint64_t object
, uint8_t compress
,
477 int dmu_object_remap_indirects(objset_t
*os
, uint64_t object
, uint64_t txg
);
479 void dmu_write_embedded(objset_t
*os
, uint64_t object
, uint64_t offset
,
480 void *data
, uint8_t etype
, uint8_t comp
, int uncompressed_size
,
481 int compressed_size
, int byteorder
, dmu_tx_t
*tx
);
484 * Decide how to write a block: checksum, compression, number of copies, etc.
486 #define WP_NOFILL 0x1
487 #define WP_DMU_SYNC 0x2
490 void dmu_write_policy(objset_t
*os
, dnode_t
*dn
, int level
, int wp
,
491 struct zio_prop
*zp
);
494 * The bonus data is accessed more or less like a regular buffer.
495 * You must dmu_bonus_hold() to get the buffer, which will give you a
496 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
497 * data. As with any normal buffer, you must call dmu_buf_will_dirty()
498 * before modifying it, and the
499 * object must be held in an assigned transaction before calling
500 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
501 * buffer as well. You must release what you hold with dmu_buf_rele().
503 * Returns ENOENT, EIO, or 0.
505 int dmu_bonus_hold_impl(objset_t
*os
, uint64_t object
, void *tag
,
506 uint32_t flags
, dmu_buf_t
**dbp
);
507 int dmu_bonus_hold(objset_t
*os
, uint64_t object
, void *tag
, dmu_buf_t
**);
508 int dmu_bonus_max(void);
509 int dmu_set_bonus(dmu_buf_t
*, int, dmu_tx_t
*);
510 int dmu_set_bonustype(dmu_buf_t
*, dmu_object_type_t
, dmu_tx_t
*);
511 dmu_object_type_t
dmu_get_bonustype(dmu_buf_t
*);
512 int dmu_rm_spill(objset_t
*, uint64_t, dmu_tx_t
*);
515 * Special spill buffer support used by "SA" framework
518 int dmu_spill_hold_by_bonus(dmu_buf_t
*bonus
, uint32_t flags
, void *tag
,
520 int dmu_spill_hold_by_dnode(dnode_t
*dn
, uint32_t flags
,
521 void *tag
, dmu_buf_t
**dbp
);
522 int dmu_spill_hold_existing(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
);
525 * Obtain the DMU buffer from the specified object which contains the
526 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
527 * that it will remain in memory. You must release the hold with
528 * dmu_buf_rele(). You must not access the dmu_buf_t after releasing
529 * what you hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
531 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
532 * on the returned buffer before reading or writing the buffer's
533 * db_data. The comments for those routines describe what particular
534 * operations are valid after calling them.
536 * The object number must be a valid, allocated object number.
538 int dmu_buf_hold(objset_t
*os
, uint64_t object
, uint64_t offset
,
539 void *tag
, dmu_buf_t
**, int flags
);
540 int dmu_buf_hold_by_dnode(dnode_t
*dn
, uint64_t offset
,
541 void *tag
, dmu_buf_t
**dbp
, int flags
);
544 * Add a reference to a dmu buffer that has already been held via
545 * dmu_buf_hold() in the current context.
547 void dmu_buf_add_ref(dmu_buf_t
*db
, void* tag
);
550 * Attempt to add a reference to a dmu buffer that is in an unknown state,
551 * using a pointer that may have been invalidated by eviction processing.
552 * The request will succeed if the passed in dbuf still represents the
553 * same os/object/blkid, is ineligible for eviction, and has at least
554 * one hold by a user other than the syncer.
556 boolean_t
dmu_buf_try_add_ref(dmu_buf_t
*, objset_t
*os
, uint64_t object
,
557 uint64_t blkid
, void *tag
);
559 void dmu_buf_rele(dmu_buf_t
*db
, void *tag
);
560 uint64_t dmu_buf_refcount(dmu_buf_t
*db
);
561 uint64_t dmu_buf_user_refcount(dmu_buf_t
*db
);
564 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
565 * range of an object. A pointer to an array of dmu_buf_t*'s is
566 * returned (in *dbpp).
568 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
569 * frees the array. The hold on the array of buffers MUST be released
570 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
571 * individually with dmu_buf_rele.
573 int dmu_buf_hold_array_by_bonus(dmu_buf_t
*db
, uint64_t offset
,
574 uint64_t length
, boolean_t read
, void *tag
,
575 int *numbufsp
, dmu_buf_t
***dbpp
);
576 void dmu_buf_rele_array(dmu_buf_t
**, int numbufs
, void *tag
);
578 typedef void dmu_buf_evict_func_t(void *user_ptr
);
581 * A DMU buffer user object may be associated with a dbuf for the
582 * duration of its lifetime. This allows the user of a dbuf (client)
583 * to attach private data to a dbuf (e.g. in-core only data such as a
584 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
585 * when that dbuf has been evicted. Clients typically respond to the
586 * eviction notification by freeing their private data, thus ensuring
587 * the same lifetime for both dbuf and private data.
589 * The mapping from a dmu_buf_user_t to any client private data is the
590 * client's responsibility. All current consumers of the API with private
591 * data embed a dmu_buf_user_t as the first member of the structure for
592 * their private data. This allows conversions between the two types
593 * with a simple cast. Since the DMU buf user API never needs access
594 * to the private data, other strategies can be employed if necessary
595 * or convenient for the client (e.g. using container_of() to do the
596 * conversion for private data that cannot have the dmu_buf_user_t as
599 * Eviction callbacks are executed without the dbuf mutex held or any
600 * other type of mechanism to guarantee that the dbuf is still available.
601 * For this reason, users must assume the dbuf has already been freed
602 * and not reference the dbuf from the callback context.
604 * Users requesting "immediate eviction" are notified as soon as the dbuf
605 * is only referenced by dirty records (dirties == holds). Otherwise the
606 * notification occurs after eviction processing for the dbuf begins.
608 typedef struct dmu_buf_user
{
610 * Asynchronous user eviction callback state.
612 taskq_ent_t dbu_tqent
;
615 * This instance's eviction function pointers.
617 * dbu_evict_func_sync is called synchronously and then
618 * dbu_evict_func_async is executed asynchronously on a taskq.
620 dmu_buf_evict_func_t
*dbu_evict_func_sync
;
621 dmu_buf_evict_func_t
*dbu_evict_func_async
;
624 * Pointer to user's dbuf pointer. NULL for clients that do
625 * not associate a dbuf with their user data.
627 * The dbuf pointer is cleared upon eviction so as to catch
628 * use-after-evict bugs in clients.
630 dmu_buf_t
**dbu_clear_on_evict_dbufp
;
635 * Initialize the given dmu_buf_user_t instance with the eviction function
636 * evict_func, to be called when the user is evicted.
638 * NOTE: This function should only be called once on a given dmu_buf_user_t.
639 * To allow enforcement of this, dbu must already be zeroed on entry.
643 dmu_buf_init_user(dmu_buf_user_t
*dbu
, dmu_buf_evict_func_t
*evict_func_sync
,
644 dmu_buf_evict_func_t
*evict_func_async
, dmu_buf_t
**clear_on_evict_dbufp
)
646 ASSERT(dbu
->dbu_evict_func_sync
== NULL
);
647 ASSERT(dbu
->dbu_evict_func_async
== NULL
);
649 /* must have at least one evict func */
650 IMPLY(evict_func_sync
== NULL
, evict_func_async
!= NULL
);
651 dbu
->dbu_evict_func_sync
= evict_func_sync
;
652 dbu
->dbu_evict_func_async
= evict_func_async
;
653 taskq_init_ent(&dbu
->dbu_tqent
);
655 dbu
->dbu_clear_on_evict_dbufp
= clear_on_evict_dbufp
;
660 * Attach user data to a dbuf and mark it for normal (when the dbuf's
661 * data is cleared or its reference count goes to zero) eviction processing.
663 * Returns NULL on success, or the existing user if another user currently
666 void *dmu_buf_set_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
669 * Attach user data to a dbuf and mark it for immediate (its dirty and
670 * reference counts are equal) eviction processing.
672 * Returns NULL on success, or the existing user if another user currently
675 void *dmu_buf_set_user_ie(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
678 * Replace the current user of a dbuf.
680 * If given the current user of a dbuf, replaces the dbuf's user with
681 * "new_user" and returns the user data pointer that was replaced.
682 * Otherwise returns the current, and unmodified, dbuf user pointer.
684 void *dmu_buf_replace_user(dmu_buf_t
*db
,
685 dmu_buf_user_t
*old_user
, dmu_buf_user_t
*new_user
);
688 * Remove the specified user data for a DMU buffer.
690 * Returns the user that was removed on success, or the current user if
691 * another user currently owns the buffer.
693 void *dmu_buf_remove_user(dmu_buf_t
*db
, dmu_buf_user_t
*user
);
696 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
698 void *dmu_buf_get_user(dmu_buf_t
*db
);
700 objset_t
*dmu_buf_get_objset(dmu_buf_t
*db
);
701 dnode_t
*dmu_buf_dnode_enter(dmu_buf_t
*db
);
702 void dmu_buf_dnode_exit(dmu_buf_t
*db
);
704 /* Block until any in-progress dmu buf user evictions complete. */
705 void dmu_buf_user_evict_wait(void);
708 * Returns the blkptr associated with this dbuf, or NULL if not set.
710 struct blkptr
*dmu_buf_get_blkptr(dmu_buf_t
*db
);
713 * Indicate that you are going to modify the buffer's data (db_data).
715 * The transaction (tx) must be assigned to a txg (ie. you've called
716 * dmu_tx_assign()). The buffer's object must be held in the tx
717 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
719 void dmu_buf_will_dirty(dmu_buf_t
*db
, dmu_tx_t
*tx
);
720 void dmu_buf_set_crypt_params(dmu_buf_t
*db_fake
, boolean_t byteorder
,
721 const uint8_t *salt
, const uint8_t *iv
, const uint8_t *mac
, dmu_tx_t
*tx
);
724 * You must create a transaction, then hold the objects which you will
725 * (or might) modify as part of this transaction. Then you must assign
726 * the transaction to a transaction group. Once the transaction has
727 * been assigned, you can modify buffers which belong to held objects as
728 * part of this transaction. You can't modify buffers before the
729 * transaction has been assigned; you can't modify buffers which don't
730 * belong to objects which this transaction holds; you can't hold
731 * objects once the transaction has been assigned. You may hold an
732 * object which you are going to free (with dmu_object_free()), but you
735 * You can abort the transaction before it has been assigned.
737 * Note that you may hold buffers (with dmu_buf_hold) at any time,
738 * regardless of transaction state.
741 #define DMU_NEW_OBJECT (-1ULL)
742 #define DMU_OBJECT_END (-1ULL)
744 dmu_tx_t
*dmu_tx_create(objset_t
*os
);
745 void dmu_tx_hold_write(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
, int len
);
746 void dmu_tx_hold_write_by_dnode(dmu_tx_t
*tx
, dnode_t
*dn
, uint64_t off
,
748 void dmu_tx_hold_free(dmu_tx_t
*tx
, uint64_t object
, uint64_t off
,
750 void dmu_tx_hold_free_by_dnode(dmu_tx_t
*tx
, dnode_t
*dn
, uint64_t off
,
752 void dmu_tx_hold_remap_l1indirect(dmu_tx_t
*tx
, uint64_t object
);
753 void dmu_tx_hold_zap(dmu_tx_t
*tx
, uint64_t object
, int add
, const char *name
);
754 void dmu_tx_hold_zap_by_dnode(dmu_tx_t
*tx
, dnode_t
*dn
, int add
,
756 void dmu_tx_hold_bonus(dmu_tx_t
*tx
, uint64_t object
);
757 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t
*tx
, dnode_t
*dn
);
758 void dmu_tx_hold_spill(dmu_tx_t
*tx
, uint64_t object
);
759 void dmu_tx_hold_sa(dmu_tx_t
*tx
, struct sa_handle
*hdl
, boolean_t may_grow
);
760 void dmu_tx_hold_sa_create(dmu_tx_t
*tx
, int total_size
);
761 void dmu_tx_abort(dmu_tx_t
*tx
);
762 int dmu_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
);
763 void dmu_tx_wait(dmu_tx_t
*tx
);
764 void dmu_tx_commit(dmu_tx_t
*tx
);
765 void dmu_tx_mark_netfree(dmu_tx_t
*tx
);
768 * To register a commit callback, dmu_tx_callback_register() must be called.
770 * dcb_data is a pointer to caller private data that is passed on as a
771 * callback parameter. The caller is responsible for properly allocating and
774 * When registering a callback, the transaction must be already created, but
775 * it cannot be committed or aborted. It can be assigned to a txg or not.
777 * The callback will be called after the transaction has been safely written
778 * to stable storage and will also be called if the dmu_tx is aborted.
779 * If there is any error which prevents the transaction from being committed to
780 * disk, the callback will be called with a value of error != 0.
782 * When multiple callbacks are registered to the transaction, the callbacks
783 * will be called in reverse order to let Lustre, the only user of commit
784 * callback currently, take the fast path of its commit callback handling.
786 typedef void dmu_tx_callback_func_t(void *dcb_data
, int error
);
788 void dmu_tx_callback_register(dmu_tx_t
*tx
, dmu_tx_callback_func_t
*dcb_func
,
790 void dmu_tx_do_callbacks(list_t
*cb_list
, int error
);
793 * Free up the data blocks for a defined range of a file. If size is
794 * -1, the range from offset to end-of-file is freed.
796 int dmu_free_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
797 uint64_t size
, dmu_tx_t
*tx
);
798 int dmu_free_long_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
800 int dmu_free_long_object(objset_t
*os
, uint64_t object
);
803 * Convenience functions.
805 * Canfail routines will return 0 on success, or an errno if there is a
806 * nonrecoverable I/O error.
808 #define DMU_READ_PREFETCH 0 /* prefetch */
809 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
810 #define DMU_READ_NO_DECRYPT 2 /* don't decrypt */
811 int dmu_read(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
812 void *buf
, uint32_t flags
);
813 int dmu_read_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
, void *buf
,
815 void dmu_write(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
816 const void *buf
, dmu_tx_t
*tx
);
817 void dmu_write_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
,
818 const void *buf
, dmu_tx_t
*tx
);
819 void dmu_prealloc(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
822 #include <linux/blkdev_compat.h>
823 int dmu_read_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
);
824 int dmu_read_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
);
825 int dmu_read_uio_dnode(dnode_t
*dn
, struct uio
*uio
, uint64_t size
);
826 int dmu_write_uio(objset_t
*os
, uint64_t object
, struct uio
*uio
, uint64_t size
,
828 int dmu_write_uio_dbuf(dmu_buf_t
*zdb
, struct uio
*uio
, uint64_t size
,
830 int dmu_write_uio_dnode(dnode_t
*dn
, struct uio
*uio
, uint64_t size
,
833 struct arc_buf
*dmu_request_arcbuf(dmu_buf_t
*handle
, int size
);
834 void dmu_return_arcbuf(struct arc_buf
*buf
);
835 void dmu_assign_arcbuf_by_dnode(dnode_t
*dn
, uint64_t offset
,
836 struct arc_buf
*buf
, dmu_tx_t
*tx
);
837 void dmu_assign_arcbuf_by_dbuf(dmu_buf_t
*handle
, uint64_t offset
,
838 struct arc_buf
*buf
, dmu_tx_t
*tx
);
839 #define dmu_assign_arcbuf dmu_assign_arcbuf_by_dbuf
840 void dmu_copy_from_buf(objset_t
*os
, uint64_t object
, uint64_t offset
,
841 dmu_buf_t
*handle
, dmu_tx_t
*tx
);
842 #ifdef HAVE_UIO_ZEROCOPY
843 int dmu_xuio_init(struct xuio
*uio
, int niov
);
844 void dmu_xuio_fini(struct xuio
*uio
);
845 int dmu_xuio_add(struct xuio
*uio
, struct arc_buf
*abuf
, offset_t off
,
847 int dmu_xuio_cnt(struct xuio
*uio
);
848 struct arc_buf
*dmu_xuio_arcbuf(struct xuio
*uio
, int i
);
849 void dmu_xuio_clear(struct xuio
*uio
, int i
);
850 #endif /* HAVE_UIO_ZEROCOPY */
851 void xuio_stat_wbuf_copied(void);
852 void xuio_stat_wbuf_nocopy(void);
854 extern int zfs_prefetch_disable
;
855 extern int zfs_max_recordsize
;
858 * Asynchronously try to read in the data.
860 void dmu_prefetch(objset_t
*os
, uint64_t object
, int64_t level
, uint64_t offset
,
861 uint64_t len
, enum zio_priority pri
);
863 typedef struct dmu_object_info
{
864 /* All sizes are in bytes unless otherwise indicated. */
865 uint32_t doi_data_block_size
;
866 uint32_t doi_metadata_block_size
;
867 dmu_object_type_t doi_type
;
868 dmu_object_type_t doi_bonus_type
;
869 uint64_t doi_bonus_size
;
870 uint8_t doi_indirection
; /* 2 = dnode->indirect->data */
871 uint8_t doi_checksum
;
872 uint8_t doi_compress
;
875 uint64_t doi_dnodesize
;
876 uint64_t doi_physical_blocks_512
; /* data + metadata, 512b blks */
877 uint64_t doi_max_offset
;
878 uint64_t doi_fill_count
; /* number of non-empty blocks */
881 typedef void (*const arc_byteswap_func_t
)(void *buf
, size_t size
);
883 typedef struct dmu_object_type_info
{
884 dmu_object_byteswap_t ot_byteswap
;
885 boolean_t ot_metadata
;
886 boolean_t ot_encrypt
;
888 } dmu_object_type_info_t
;
890 typedef const struct dmu_object_byteswap_info
{
891 arc_byteswap_func_t ob_func
;
893 } dmu_object_byteswap_info_t
;
895 extern const dmu_object_type_info_t dmu_ot
[DMU_OT_NUMTYPES
];
896 extern const dmu_object_byteswap_info_t dmu_ot_byteswap
[DMU_BSWAP_NUMFUNCS
];
899 * Get information on a DMU object.
901 * Return 0 on success or ENOENT if object is not allocated.
903 * If doi is NULL, just indicates whether the object exists.
905 int dmu_object_info(objset_t
*os
, uint64_t object
, dmu_object_info_t
*doi
);
906 void __dmu_object_info_from_dnode(struct dnode
*dn
, dmu_object_info_t
*doi
);
907 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
908 void dmu_object_info_from_dnode(dnode_t
*dn
, dmu_object_info_t
*doi
);
909 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
910 void dmu_object_info_from_db(dmu_buf_t
*db
, dmu_object_info_t
*doi
);
912 * Like dmu_object_info_from_db, but faster still when you only care about
913 * the size. This is specifically optimized for zfs_getattr().
915 void dmu_object_size_from_db(dmu_buf_t
*db
, uint32_t *blksize
,
916 u_longlong_t
*nblk512
);
918 void dmu_object_dnsize_from_db(dmu_buf_t
*db
, int *dnsize
);
920 typedef struct dmu_objset_stats
{
921 uint64_t dds_num_clones
; /* number of clones of this */
922 uint64_t dds_creation_txg
;
924 dmu_objset_type_t dds_type
;
925 uint8_t dds_is_snapshot
;
926 uint8_t dds_inconsistent
;
927 char dds_origin
[ZFS_MAX_DATASET_NAME_LEN
];
928 } dmu_objset_stats_t
;
931 * Get stats on a dataset.
933 void dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
);
936 * Add entries to the nvlist for all the objset's properties. See
937 * zfs_prop_table[] and zfs(1m) for details on the properties.
939 void dmu_objset_stats(objset_t
*os
, struct nvlist
*nv
);
942 * Get the space usage statistics for statvfs().
944 * refdbytes is the amount of space "referenced" by this objset.
945 * availbytes is the amount of space available to this objset, taking
946 * into account quotas & reservations, assuming that no other objsets
947 * use the space first. These values correspond to the 'referenced' and
948 * 'available' properties, described in the zfs(1m) manpage.
950 * usedobjs and availobjs are the number of objects currently allocated,
953 void dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
954 uint64_t *usedobjsp
, uint64_t *availobjsp
);
957 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
958 * (Contrast with the ds_guid which is a 64-bit ID that will never
959 * change, so there is a small probability that it will collide.)
961 uint64_t dmu_objset_fsid_guid(objset_t
*os
);
964 * Get the [cm]time for an objset's snapshot dir
966 inode_timespec_t
dmu_objset_snap_cmtime(objset_t
*os
);
968 int dmu_objset_is_snapshot(objset_t
*os
);
970 extern struct spa
*dmu_objset_spa(objset_t
*os
);
971 extern struct zilog
*dmu_objset_zil(objset_t
*os
);
972 extern struct dsl_pool
*dmu_objset_pool(objset_t
*os
);
973 extern struct dsl_dataset
*dmu_objset_ds(objset_t
*os
);
974 extern void dmu_objset_name(objset_t
*os
, char *buf
);
975 extern dmu_objset_type_t
dmu_objset_type(objset_t
*os
);
976 extern uint64_t dmu_objset_id(objset_t
*os
);
977 extern uint64_t dmu_objset_dnodesize(objset_t
*os
);
978 extern zfs_sync_type_t
dmu_objset_syncprop(objset_t
*os
);
979 extern zfs_logbias_op_t
dmu_objset_logbias(objset_t
*os
);
980 extern int dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
981 uint64_t *id
, uint64_t *offp
, boolean_t
*case_conflict
);
982 extern int dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *val
);
983 extern int dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
,
984 int maxlen
, boolean_t
*conflict
);
985 extern int dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
986 uint64_t *idp
, uint64_t *offp
);
988 typedef int objset_used_cb_t(dmu_object_type_t bonustype
,
989 void *bonus
, uint64_t *userp
, uint64_t *groupp
, uint64_t *projectp
);
990 extern void dmu_objset_register_type(dmu_objset_type_t ost
,
991 objset_used_cb_t
*cb
);
992 extern void dmu_objset_set_user(objset_t
*os
, void *user_ptr
);
993 extern void *dmu_objset_get_user(objset_t
*os
);
996 * Return the txg number for the given assigned transaction.
998 uint64_t dmu_tx_get_txg(dmu_tx_t
*tx
);
1001 * Synchronous write.
1002 * If a parent zio is provided this function initiates a write on the
1003 * provided buffer as a child of the parent zio.
1004 * In the absence of a parent zio, the write is completed synchronously.
1005 * At write completion, blk is filled with the bp of the written block.
1006 * Note that while the data covered by this function will be on stable
1007 * storage when the write completes this new data does not become a
1008 * permanent part of the file until the associated transaction commits.
1012 * {zfs,zvol,ztest}_get_done() args
1014 typedef struct zgd
{
1015 struct lwb
*zgd_lwb
;
1016 struct blkptr
*zgd_bp
;
1022 typedef void dmu_sync_cb_t(zgd_t
*arg
, int error
);
1023 int dmu_sync(struct zio
*zio
, uint64_t txg
, dmu_sync_cb_t
*done
, zgd_t
*zgd
);
1026 * Find the next hole or data block in file starting at *off
1027 * Return found offset in *off. Return ESRCH for end of file.
1029 int dmu_offset_next(objset_t
*os
, uint64_t object
, boolean_t hole
,
1033 * Initial setup and final teardown.
1035 extern void dmu_init(void);
1036 extern void dmu_fini(void);
1038 typedef void (*dmu_traverse_cb_t
)(objset_t
*os
, void *arg
, struct blkptr
*bp
,
1039 uint64_t object
, uint64_t offset
, int len
);
1040 void dmu_traverse_objset(objset_t
*os
, uint64_t txg_start
,
1041 dmu_traverse_cb_t cb
, void *arg
);
1043 int dmu_diff(const char *tosnap_name
, const char *fromsnap_name
,
1044 struct vnode
*vp
, offset_t
*offp
);
1047 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
1048 extern uint64_t zfs_crc64_table
[256];
1054 #endif /* _SYS_DMU_H */