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, 2018 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.
29 * Copyright (c) 2017, Intel Corporation.
32 /* Portions Copyright 2010 Robert Milkowski */
38 * This file describes the interface that the DMU provides for its
41 * The DMU also interacts with the SPA. That interface is described in
45 #include <sys/zfs_context.h>
46 #include <sys/inttypes.h>
48 #include <sys/fs/zfs.h>
49 #include <sys/zio_compress.h>
50 #include <sys/zio_priority.h>
69 struct zbookmark_phys;
75 struct dsl_crypto_params;
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_ENCRYPTED 0x20
107 #define DMU_OT_BYTESWAP_MASK 0x1f
110 * Defines a uint8_t object type. Object types specify if the data
111 * in the object is metadata (boolean) and how to byteswap the data
112 * (dmu_object_byteswap_t). All of the types created by this method
113 * are cached in the dbuf metadata cache.
115 #define DMU_OT(byteswap, metadata, encrypted) \
117 ((metadata) ? DMU_OT_METADATA : 0) | \
118 ((encrypted) ? DMU_OT_ENCRYPTED : 0) | \
119 ((byteswap) & DMU_OT_BYTESWAP_MASK))
121 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
122 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
123 (ot) < DMU_OT_NUMTYPES)
125 #define DMU_OT_IS_METADATA_CACHED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
126 B_TRUE : dmu_ot[(ot)].ot_dbuf_metadata_cache)
129 * MDB doesn't have dmu_ot; it defines these macros itself.
132 #define DMU_OT_IS_METADATA_IMPL(ot) (dmu_ot[ot].ot_metadata)
133 #define DMU_OT_IS_ENCRYPTED_IMPL(ot) (dmu_ot[ot].ot_encrypt)
134 #define DMU_OT_BYTESWAP_IMPL(ot) (dmu_ot[ot].ot_byteswap)
137 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
138 ((ot) & DMU_OT_METADATA) : \
139 DMU_OT_IS_METADATA_IMPL(ot))
141 #define DMU_OT_IS_DDT(ot) \
142 ((ot) == DMU_OT_DDT_ZAP)
144 #define DMU_OT_IS_ZIL(ot) \
145 ((ot) == DMU_OT_INTENT_LOG)
147 /* Note: ztest uses DMU_OT_UINT64_OTHER as a proxy for file blocks */
148 #define DMU_OT_IS_FILE(ot) \
149 ((ot) == DMU_OT_PLAIN_FILE_CONTENTS || (ot) == DMU_OT_UINT64_OTHER)
151 #define DMU_OT_IS_ENCRYPTED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
152 ((ot) & DMU_OT_ENCRYPTED) : \
153 DMU_OT_IS_ENCRYPTED_IMPL(ot))
156 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
157 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
158 * is repurposed for embedded BPs.
160 #define DMU_OT_HAS_FILL(ot) \
161 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
163 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
164 ((ot) & DMU_OT_BYTESWAP_MASK) : \
165 DMU_OT_BYTESWAP_IMPL(ot))
167 typedef enum dmu_object_type {
170 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
171 DMU_OT_OBJECT_ARRAY, /* UINT64 */
172 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
173 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
174 DMU_OT_BPOBJ, /* UINT64 */
175 DMU_OT_BPOBJ_HDR, /* UINT64 */
177 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
178 DMU_OT_SPACE_MAP, /* UINT64 */
180 DMU_OT_INTENT_LOG, /* UINT64 */
182 DMU_OT_DNODE, /* DNODE */
183 DMU_OT_OBJSET, /* OBJSET */
185 DMU_OT_DSL_DIR, /* UINT64 */
186 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
187 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
188 DMU_OT_DSL_PROPS, /* ZAP */
189 DMU_OT_DSL_DATASET, /* UINT64 */
191 DMU_OT_ZNODE, /* ZNODE */
192 DMU_OT_OLDACL, /* Old ACL */
193 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
194 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
195 DMU_OT_MASTER_NODE, /* ZAP */
196 DMU_OT_UNLINKED_SET, /* ZAP */
198 DMU_OT_ZVOL, /* UINT8 */
199 DMU_OT_ZVOL_PROP, /* ZAP */
200 /* other; for testing only! */
201 DMU_OT_PLAIN_OTHER, /* UINT8 */
202 DMU_OT_UINT64_OTHER, /* UINT64 */
203 DMU_OT_ZAP_OTHER, /* ZAP */
204 /* new object types: */
205 DMU_OT_ERROR_LOG, /* ZAP */
206 DMU_OT_SPA_HISTORY, /* UINT8 */
207 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
208 DMU_OT_POOL_PROPS, /* ZAP */
209 DMU_OT_DSL_PERMS, /* ZAP */
210 DMU_OT_ACL, /* ACL */
211 DMU_OT_SYSACL, /* SYSACL */
212 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
213 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
214 DMU_OT_NEXT_CLONES, /* ZAP */
215 DMU_OT_SCAN_QUEUE, /* ZAP */
216 DMU_OT_USERGROUP_USED, /* ZAP */
217 DMU_OT_USERGROUP_QUOTA, /* ZAP */
218 DMU_OT_USERREFS, /* ZAP */
219 DMU_OT_DDT_ZAP, /* ZAP */
220 DMU_OT_DDT_STATS, /* ZAP */
221 DMU_OT_SA, /* System attr */
222 DMU_OT_SA_MASTER_NODE, /* ZAP */
223 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
224 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
225 DMU_OT_SCAN_XLATE, /* ZAP */
226 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
227 DMU_OT_DEADLIST, /* ZAP */
228 DMU_OT_DEADLIST_HDR, /* UINT64 */
229 DMU_OT_DSL_CLONES, /* ZAP */
230 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
232 * Do not allocate new object types here. Doing so makes the on-disk
233 * format incompatible with any other format that uses the same object
236 * When creating an object which does not have one of the above types
237 * use the DMU_OTN_* type with the correct byteswap and metadata
240 * The DMU_OTN_* types do not have entries in the dmu_ot table,
241 * use the DMU_OT_IS_METADATA() and DMU_OT_BYTESWAP() macros instead
242 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
243 * and DMU_OTN_* types).
248 * Names for valid types declared with DMU_OT().
250 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_FALSE),
251 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_FALSE),
252 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_FALSE),
253 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_FALSE),
254 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_FALSE),
255 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_FALSE),
256 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_FALSE),
257 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_FALSE),
258 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_FALSE),
259 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_FALSE),
261 DMU_OTN_UINT8_ENC_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_TRUE),
262 DMU_OTN_UINT8_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_TRUE),
263 DMU_OTN_UINT16_ENC_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_TRUE),
264 DMU_OTN_UINT16_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_TRUE),
265 DMU_OTN_UINT32_ENC_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_TRUE),
266 DMU_OTN_UINT32_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_TRUE),
267 DMU_OTN_UINT64_ENC_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_TRUE),
268 DMU_OTN_UINT64_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_TRUE),
269 DMU_OTN_ZAP_ENC_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_TRUE),
270 DMU_OTN_ZAP_ENC_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_TRUE),
274 * These flags are intended to be used to specify the "txg_how"
275 * parameter when calling the dmu_tx_assign() function. See the comment
276 * above dmu_tx_assign() for more details on the meaning of these flags.
278 #define TXG_NOWAIT (0ULL)
279 #define TXG_WAIT (1ULL<<0)
280 #define TXG_NOTHROTTLE (1ULL<<1)
282 void byteswap_uint64_array(void *buf, size_t size);
283 void byteswap_uint32_array(void *buf, size_t size);
284 void byteswap_uint16_array(void *buf, size_t size);
285 void byteswap_uint8_array(void *buf, size_t size);
286 void zap_byteswap(void *buf, size_t size);
287 void zfs_oldacl_byteswap(void *buf, size_t size);
288 void zfs_acl_byteswap(void *buf, size_t size);
289 void zfs_znode_byteswap(void *buf, size_t size);
291 #define DS_FIND_SNAPSHOTS (1<<0)
292 #define DS_FIND_CHILDREN (1<<1)
293 #define DS_FIND_SERIALIZE (1<<2)
296 * The maximum number of bytes that can be accessed as part of one
297 * operation, including metadata.
299 #define DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
300 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
302 #define DMU_USERUSED_OBJECT (-1ULL)
303 #define DMU_GROUPUSED_OBJECT (-2ULL)
304 #define DMU_PROJECTUSED_OBJECT (-3ULL)
307 * Zap prefix for object accounting in DMU_{USER,GROUP,PROJECT}USED_OBJECT.
309 #define DMU_OBJACCT_PREFIX "obj-"
310 #define DMU_OBJACCT_PREFIX_LEN 4
313 * artificial blkids for bonus buffer and spill blocks
315 #define DMU_BONUS_BLKID (-1ULL)
316 #define DMU_SPILL_BLKID (-2ULL)
319 * Public routines to create, destroy, open, and close objsets.
321 typedef void dmu_objset_create_sync_func_t(objset_t *os, void *arg,
322 cred_t *cr, dmu_tx_t *tx);
324 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
325 int dmu_objset_own(const char *name, dmu_objset_type_t type,
326 boolean_t readonly, boolean_t key_required, void *tag, objset_t **osp);
327 void dmu_objset_rele(objset_t *os, void *tag);
328 void dmu_objset_disown(objset_t *os, boolean_t key_required, void *tag);
329 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
331 void dmu_objset_evict_dbufs(objset_t *os);
332 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
333 struct dsl_crypto_params *dcp, dmu_objset_create_sync_func_t func,
335 int dmu_objset_clone(const char *name, const char *origin);
336 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
337 struct nvlist *errlist);
338 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
339 int dmu_objset_snapshot_tmp(const char *, const char *, int);
340 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
342 void dmu_objset_byteswap(void *buf, size_t size);
343 int dsl_dataset_rename_snapshot(const char *fsname,
344 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
345 int dmu_objset_remap_indirects(const char *fsname);
347 typedef struct dmu_buf {
348 uint64_t db_object; /* object that this buffer is part of */
349 uint64_t db_offset; /* byte offset in this object */
350 uint64_t db_size; /* size of buffer in bytes */
351 void *db_data; /* data in buffer */
355 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
357 #define DMU_POOL_DIRECTORY_OBJECT 1
358 #define DMU_POOL_CONFIG "config"
359 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
360 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
361 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
362 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
363 #define DMU_POOL_ROOT_DATASET "root_dataset"
364 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
365 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
366 #define DMU_POOL_ERRLOG_LAST "errlog_last"
367 #define DMU_POOL_SPARES "spares"
368 #define DMU_POOL_DEFLATE "deflate"
369 #define DMU_POOL_HISTORY "history"
370 #define DMU_POOL_PROPS "pool_props"
371 #define DMU_POOL_L2CACHE "l2cache"
372 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
373 #define DMU_POOL_DDT "DDT-%s-%s-%s"
374 #define DMU_POOL_DDT_STATS "DDT-statistics"
375 #define DMU_POOL_CREATION_VERSION "creation_version"
376 #define DMU_POOL_SCAN "scan"
377 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
378 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
379 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
380 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
381 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
382 #define DMU_POOL_REMOVING "com.delphix:removing"
383 #define DMU_POOL_OBSOLETE_BPOBJ "com.delphix:obsolete_bpobj"
384 #define DMU_POOL_CONDENSING_INDIRECT "com.delphix:condensing_indirect"
385 #define DMU_POOL_ZPOOL_CHECKPOINT "com.delphix:zpool_checkpoint"
388 * Allocate an object from this objset. The range of object numbers
389 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
391 * The transaction must be assigned to a txg. The newly allocated
392 * object will be "held" in the transaction (ie. you can modify the
393 * newly allocated object in this transaction).
395 * dmu_object_alloc() chooses an object and returns it in *objectp.
397 * dmu_object_claim() allocates a specific object number. If that
398 * number is already allocated, it fails and returns EEXIST.
400 * Return 0 on success, or ENOSPC or EEXIST as specified above.
402 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
403 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
404 uint64_t dmu_object_alloc_ibs(objset_t *os, dmu_object_type_t ot, int blocksize,
405 int indirect_blockshift,
406 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
407 uint64_t dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot,
408 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
409 int dnodesize, dmu_tx_t *tx);
410 uint64_t dmu_object_alloc_hold(objset_t *os, dmu_object_type_t ot,
411 int blocksize, int indirect_blockshift, dmu_object_type_t bonustype,
412 int bonuslen, int dnodesize, dnode_t **allocated_dnode, void *tag,
414 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
415 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
416 int dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
417 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
418 int dnodesize, dmu_tx_t *tx);
419 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
420 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
421 int dmu_object_reclaim_dnsize(objset_t *os, uint64_t object,
422 dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype,
423 int bonuslen, int dnodesize, boolean_t keep_spill, dmu_tx_t *tx);
424 int dmu_object_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx);
427 * Free an object from this objset.
429 * The object's data will be freed as well (ie. you don't need to call
430 * dmu_free(object, 0, -1, tx)).
432 * The object need not be held in the transaction.
434 * If there are any holds on this object's buffers (via dmu_buf_hold()),
435 * or tx holds on the object (via dmu_tx_hold_object()), you can not
436 * free it; it fails and returns EBUSY.
438 * If the object is not allocated, it fails and returns ENOENT.
440 * Return 0 on success, or EBUSY or ENOENT as specified above.
442 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
445 * Find the next allocated or free object.
447 * The objectp parameter is in-out. It will be updated to be the next
448 * object which is allocated. Ignore objects which have not been
449 * modified since txg.
451 * XXX Can only be called on a objset with no dirty data.
453 * Returns 0 on success, or ENOENT if there are no more objects.
455 int dmu_object_next(objset_t *os, uint64_t *objectp,
456 boolean_t hole, uint64_t txg);
459 * Set the number of levels on a dnode. nlevels must be greater than the
460 * current number of levels or an EINVAL will be returned.
462 int dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels,
466 * Set the data blocksize for an object.
468 * The object cannot have any blocks allcated beyond the first. If
469 * the first block is allocated already, the new size must be greater
470 * than the current block size. If these conditions are not met,
471 * ENOTSUP will be returned.
473 * Returns 0 on success, or EBUSY if there are any holds on the object
474 * contents, or ENOTSUP as described above.
476 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
477 int ibs, dmu_tx_t *tx);
480 * Manually set the maxblkid on a dnode. This will adjust nlevels accordingly
481 * to accommodate the change. When calling this function, the caller must
482 * ensure that the object's nlevels can sufficiently support the new maxblkid.
484 int dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid,
488 * Set the checksum property on a dnode. The new checksum algorithm will
489 * apply to all newly written blocks; existing blocks will not be affected.
491 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
495 * Set the compress property on a dnode. The new compression algorithm will
496 * apply to all newly written blocks; existing blocks will not be affected.
498 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
502 int dmu_object_remap_indirects(objset_t *os, uint64_t object, uint64_t txg);
504 void dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
505 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
506 int compressed_size, int byteorder, dmu_tx_t *tx);
507 void dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
511 * Decide how to write a block: checksum, compression, number of copies, etc.
513 #define WP_NOFILL 0x1
514 #define WP_DMU_SYNC 0x2
517 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
518 struct zio_prop *zp);
521 * The bonus data is accessed more or less like a regular buffer.
522 * You must dmu_bonus_hold() to get the buffer, which will give you a
523 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
524 * data. As with any normal buffer, you must call dmu_buf_will_dirty()
525 * before modifying it, and the
526 * object must be held in an assigned transaction before calling
527 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
528 * buffer as well. You must release what you hold with dmu_buf_rele().
530 * Returns ENOENT, EIO, or 0.
532 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp);
533 int dmu_bonus_hold_by_dnode(dnode_t *dn, void *tag, dmu_buf_t **dbp,
535 int dmu_bonus_max(void);
536 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
537 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
538 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
539 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
542 * Special spill buffer support used by "SA" framework
545 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag,
547 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
548 void *tag, dmu_buf_t **dbp);
549 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
552 * Obtain the DMU buffer from the specified object which contains the
553 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
554 * that it will remain in memory. You must release the hold with
555 * dmu_buf_rele(). You must not access the dmu_buf_t after releasing
556 * what you hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
558 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
559 * on the returned buffer before reading or writing the buffer's
560 * db_data. The comments for those routines describe what particular
561 * operations are valid after calling them.
563 * The object number must be a valid, allocated object number.
565 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
566 void *tag, dmu_buf_t **, int flags);
567 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
568 void *tag, dmu_buf_t **dbp, int flags);
571 * Add a reference to a dmu buffer that has already been held via
572 * dmu_buf_hold() in the current context.
574 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
577 * Attempt to add a reference to a dmu buffer that is in an unknown state,
578 * using a pointer that may have been invalidated by eviction processing.
579 * The request will succeed if the passed in dbuf still represents the
580 * same os/object/blkid, is ineligible for eviction, and has at least
581 * one hold by a user other than the syncer.
583 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
584 uint64_t blkid, void *tag);
586 void dmu_buf_rele(dmu_buf_t *db, void *tag);
587 uint64_t dmu_buf_refcount(dmu_buf_t *db);
588 uint64_t dmu_buf_user_refcount(dmu_buf_t *db);
591 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
592 * range of an object. A pointer to an array of dmu_buf_t*'s is
593 * returned (in *dbpp).
595 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
596 * frees the array. The hold on the array of buffers MUST be released
597 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
598 * individually with dmu_buf_rele.
600 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
601 uint64_t length, boolean_t read, void *tag,
602 int *numbufsp, dmu_buf_t ***dbpp);
603 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
605 typedef void dmu_buf_evict_func_t(void *user_ptr);
608 * A DMU buffer user object may be associated with a dbuf for the
609 * duration of its lifetime. This allows the user of a dbuf (client)
610 * to attach private data to a dbuf (e.g. in-core only data such as a
611 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
612 * when that dbuf has been evicted. Clients typically respond to the
613 * eviction notification by freeing their private data, thus ensuring
614 * the same lifetime for both dbuf and private data.
616 * The mapping from a dmu_buf_user_t to any client private data is the
617 * client's responsibility. All current consumers of the API with private
618 * data embed a dmu_buf_user_t as the first member of the structure for
619 * their private data. This allows conversions between the two types
620 * with a simple cast. Since the DMU buf user API never needs access
621 * to the private data, other strategies can be employed if necessary
622 * or convenient for the client (e.g. using container_of() to do the
623 * conversion for private data that cannot have the dmu_buf_user_t as
626 * Eviction callbacks are executed without the dbuf mutex held or any
627 * other type of mechanism to guarantee that the dbuf is still available.
628 * For this reason, users must assume the dbuf has already been freed
629 * and not reference the dbuf from the callback context.
631 * Users requesting "immediate eviction" are notified as soon as the dbuf
632 * is only referenced by dirty records (dirties == holds). Otherwise the
633 * notification occurs after eviction processing for the dbuf begins.
635 typedef struct dmu_buf_user {
637 * Asynchronous user eviction callback state.
639 taskq_ent_t dbu_tqent;
642 * This instance's eviction function pointers.
644 * dbu_evict_func_sync is called synchronously and then
645 * dbu_evict_func_async is executed asynchronously on a taskq.
647 dmu_buf_evict_func_t *dbu_evict_func_sync;
648 dmu_buf_evict_func_t *dbu_evict_func_async;
651 * Pointer to user's dbuf pointer. NULL for clients that do
652 * not associate a dbuf with their user data.
654 * The dbuf pointer is cleared upon eviction so as to catch
655 * use-after-evict bugs in clients.
657 dmu_buf_t **dbu_clear_on_evict_dbufp;
662 * Initialize the given dmu_buf_user_t instance with the eviction function
663 * evict_func, to be called when the user is evicted.
665 * NOTE: This function should only be called once on a given dmu_buf_user_t.
666 * To allow enforcement of this, dbu must already be zeroed on entry.
670 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
671 dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
673 ASSERT(dbu->dbu_evict_func_sync == NULL);
674 ASSERT(dbu->dbu_evict_func_async == NULL);
676 /* must have at least one evict func */
677 IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
678 dbu->dbu_evict_func_sync = evict_func_sync;
679 dbu->dbu_evict_func_async = evict_func_async;
680 taskq_init_ent(&dbu->dbu_tqent);
682 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
687 * Attach user data to a dbuf and mark it for normal (when the dbuf's
688 * data is cleared or its reference count goes to zero) eviction processing.
690 * Returns NULL on success, or the existing user if another user currently
693 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
696 * Attach user data to a dbuf and mark it for immediate (its dirty and
697 * reference counts are equal) eviction processing.
699 * Returns NULL on success, or the existing user if another user currently
702 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
705 * Replace the current user of a dbuf.
707 * If given the current user of a dbuf, replaces the dbuf's user with
708 * "new_user" and returns the user data pointer that was replaced.
709 * Otherwise returns the current, and unmodified, dbuf user pointer.
711 void *dmu_buf_replace_user(dmu_buf_t *db,
712 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
715 * Remove the specified user data for a DMU buffer.
717 * Returns the user that was removed on success, or the current user if
718 * another user currently owns the buffer.
720 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
723 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
725 void *dmu_buf_get_user(dmu_buf_t *db);
727 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
728 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
729 void dmu_buf_dnode_exit(dmu_buf_t *db);
731 /* Block until any in-progress dmu buf user evictions complete. */
732 void dmu_buf_user_evict_wait(void);
735 * Returns the blkptr associated with this dbuf, or NULL if not set.
737 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
740 * Indicate that you are going to modify the buffer's data (db_data).
742 * The transaction (tx) must be assigned to a txg (ie. you've called
743 * dmu_tx_assign()). The buffer's object must be held in the tx
744 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
746 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
747 boolean_t dmu_buf_is_dirty(dmu_buf_t *db, dmu_tx_t *tx);
748 void dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder,
749 const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx);
752 * You must create a transaction, then hold the objects which you will
753 * (or might) modify as part of this transaction. Then you must assign
754 * the transaction to a transaction group. Once the transaction has
755 * been assigned, you can modify buffers which belong to held objects as
756 * part of this transaction. You can't modify buffers before the
757 * transaction has been assigned; you can't modify buffers which don't
758 * belong to objects which this transaction holds; you can't hold
759 * objects once the transaction has been assigned. You may hold an
760 * object which you are going to free (with dmu_object_free()), but you
763 * You can abort the transaction before it has been assigned.
765 * Note that you may hold buffers (with dmu_buf_hold) at any time,
766 * regardless of transaction state.
769 #define DMU_NEW_OBJECT (-1ULL)
770 #define DMU_OBJECT_END (-1ULL)
772 dmu_tx_t *dmu_tx_create(objset_t *os);
773 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
774 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
776 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
778 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
780 void dmu_tx_hold_remap_l1indirect(dmu_tx_t *tx, uint64_t object);
781 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
782 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
784 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
785 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
786 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
787 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
788 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
789 void dmu_tx_abort(dmu_tx_t *tx);
790 int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
791 void dmu_tx_wait(dmu_tx_t *tx);
792 void dmu_tx_commit(dmu_tx_t *tx);
793 void dmu_tx_mark_netfree(dmu_tx_t *tx);
796 * To register a commit callback, dmu_tx_callback_register() must be called.
798 * dcb_data is a pointer to caller private data that is passed on as a
799 * callback parameter. The caller is responsible for properly allocating and
802 * When registering a callback, the transaction must be already created, but
803 * it cannot be committed or aborted. It can be assigned to a txg or not.
805 * The callback will be called after the transaction has been safely written
806 * to stable storage and will also be called if the dmu_tx is aborted.
807 * If there is any error which prevents the transaction from being committed to
808 * disk, the callback will be called with a value of error != 0.
810 * When multiple callbacks are registered to the transaction, the callbacks
811 * will be called in reverse order to let Lustre, the only user of commit
812 * callback currently, take the fast path of its commit callback handling.
814 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
816 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
818 void dmu_tx_do_callbacks(list_t *cb_list, int error);
821 * Free up the data blocks for a defined range of a file. If size is
822 * -1, the range from offset to end-of-file is freed.
824 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
825 uint64_t size, dmu_tx_t *tx);
826 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
828 int dmu_free_long_object(objset_t *os, uint64_t object);
831 * Convenience functions.
833 * Canfail routines will return 0 on success, or an errno if there is a
834 * nonrecoverable I/O error.
836 #define DMU_READ_PREFETCH 0 /* prefetch */
837 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
838 #define DMU_READ_NO_DECRYPT 2 /* don't decrypt */
839 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
840 void *buf, uint32_t flags);
841 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
843 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
844 const void *buf, dmu_tx_t *tx);
845 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
846 const void *buf, dmu_tx_t *tx);
847 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
850 #include <linux/blkdev_compat.h>
851 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
852 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
853 int dmu_read_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size);
854 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
856 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
858 int dmu_write_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size,
861 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
862 void dmu_return_arcbuf(struct arc_buf *buf);
863 int dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset,
864 struct arc_buf *buf, dmu_tx_t *tx);
865 int dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset,
866 struct arc_buf *buf, dmu_tx_t *tx);
867 #define dmu_assign_arcbuf dmu_assign_arcbuf_by_dbuf
868 void dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset,
869 dmu_buf_t *handle, dmu_tx_t *tx);
870 #ifdef HAVE_UIO_ZEROCOPY
871 int dmu_xuio_init(struct xuio *uio, int niov);
872 void dmu_xuio_fini(struct xuio *uio);
873 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
875 int dmu_xuio_cnt(struct xuio *uio);
876 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
877 void dmu_xuio_clear(struct xuio *uio, int i);
878 #endif /* HAVE_UIO_ZEROCOPY */
879 void xuio_stat_wbuf_copied(void);
880 void xuio_stat_wbuf_nocopy(void);
882 extern int zfs_prefetch_disable;
883 extern int zfs_max_recordsize;
886 * Asynchronously try to read in the data.
888 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
889 uint64_t len, enum zio_priority pri);
891 typedef struct dmu_object_info {
892 /* All sizes are in bytes unless otherwise indicated. */
893 uint32_t doi_data_block_size;
894 uint32_t doi_metadata_block_size;
895 dmu_object_type_t doi_type;
896 dmu_object_type_t doi_bonus_type;
897 uint64_t doi_bonus_size;
898 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
899 uint8_t doi_checksum;
900 uint8_t doi_compress;
903 uint64_t doi_dnodesize;
904 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
905 uint64_t doi_max_offset;
906 uint64_t doi_fill_count; /* number of non-empty blocks */
909 typedef void (*const arc_byteswap_func_t)(void *buf, size_t size);
911 typedef struct dmu_object_type_info {
912 dmu_object_byteswap_t ot_byteswap;
913 boolean_t ot_metadata;
914 boolean_t ot_dbuf_metadata_cache;
915 boolean_t ot_encrypt;
917 } dmu_object_type_info_t;
919 typedef const struct dmu_object_byteswap_info {
920 arc_byteswap_func_t ob_func;
922 } dmu_object_byteswap_info_t;
924 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
925 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
928 * Get information on a DMU object.
930 * Return 0 on success or ENOENT if object is not allocated.
932 * If doi is NULL, just indicates whether the object exists.
934 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
935 void __dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
936 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
937 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
938 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
939 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
941 * Like dmu_object_info_from_db, but faster still when you only care about
942 * the size. This is specifically optimized for zfs_getattr().
944 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
945 u_longlong_t *nblk512);
947 void dmu_object_dnsize_from_db(dmu_buf_t *db, int *dnsize);
949 typedef struct dmu_objset_stats {
950 uint64_t dds_num_clones; /* number of clones of this */
951 uint64_t dds_creation_txg;
953 dmu_objset_type_t dds_type;
954 uint8_t dds_is_snapshot;
955 uint8_t dds_inconsistent;
956 uint8_t dds_redacted;
957 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
958 } dmu_objset_stats_t;
961 * Get stats on a dataset.
963 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
966 * Add entries to the nvlist for all the objset's properties. See
967 * zfs_prop_table[] and zfs(1m) for details on the properties.
969 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
972 * Get the space usage statistics for statvfs().
974 * refdbytes is the amount of space "referenced" by this objset.
975 * availbytes is the amount of space available to this objset, taking
976 * into account quotas & reservations, assuming that no other objsets
977 * use the space first. These values correspond to the 'referenced' and
978 * 'available' properties, described in the zfs(1m) manpage.
980 * usedobjs and availobjs are the number of objects currently allocated,
983 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
984 uint64_t *usedobjsp, uint64_t *availobjsp);
987 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
988 * (Contrast with the ds_guid which is a 64-bit ID that will never
989 * change, so there is a small probability that it will collide.)
991 uint64_t dmu_objset_fsid_guid(objset_t *os);
994 * Get the [cm]time for an objset's snapshot dir
996 inode_timespec_t dmu_objset_snap_cmtime(objset_t *os);
998 int dmu_objset_is_snapshot(objset_t *os);
1000 extern struct spa *dmu_objset_spa(objset_t *os);
1001 extern struct zilog *dmu_objset_zil(objset_t *os);
1002 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
1003 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
1004 extern void dmu_objset_name(objset_t *os, char *buf);
1005 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
1006 extern uint64_t dmu_objset_id(objset_t *os);
1007 extern uint64_t dmu_objset_dnodesize(objset_t *os);
1008 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
1009 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
1010 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1011 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
1012 extern int dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *val);
1013 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
1014 int maxlen, boolean_t *conflict);
1015 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
1016 uint64_t *idp, uint64_t *offp);
1018 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
1019 void *bonus, uint64_t *userp, uint64_t *groupp, uint64_t *projectp);
1020 extern void dmu_objset_register_type(dmu_objset_type_t ost,
1021 objset_used_cb_t *cb);
1022 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
1023 extern void *dmu_objset_get_user(objset_t *os);
1026 * Return the txg number for the given assigned transaction.
1028 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
1031 * Synchronous write.
1032 * If a parent zio is provided this function initiates a write on the
1033 * provided buffer as a child of the parent zio.
1034 * In the absence of a parent zio, the write is completed synchronously.
1035 * At write completion, blk is filled with the bp of the written block.
1036 * Note that while the data covered by this function will be on stable
1037 * storage when the write completes this new data does not become a
1038 * permanent part of the file until the associated transaction commits.
1042 * {zfs,zvol,ztest}_get_done() args
1044 typedef struct zgd {
1045 struct lwb *zgd_lwb;
1046 struct blkptr *zgd_bp;
1048 struct locked_range *zgd_lr;
1052 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
1053 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
1056 * Find the next hole or data block in file starting at *off
1057 * Return found offset in *off. Return ESRCH for end of file.
1059 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
1063 * Initial setup and final teardown.
1065 extern void dmu_init(void);
1066 extern void dmu_fini(void);
1068 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
1069 uint64_t object, uint64_t offset, int len);
1070 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
1071 dmu_traverse_cb_t cb, void *arg);
1073 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
1074 struct vnode *vp, offset_t *offp);
1077 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
1078 extern uint64_t zfs_crc64_table[256];
1084 #endif /* _SYS_DMU_H */