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, 2020 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>
52 #include <sys/zfs_file.h>
70 struct zbookmark_phys;
76 struct dsl_crypto_params;
79 typedef struct objset objset_t;
80 typedef struct dmu_tx dmu_tx_t;
81 typedef struct dsl_dir dsl_dir_t;
82 typedef struct dnode dnode_t;
84 typedef enum dmu_object_byteswap {
96 * Allocating a new byteswap type number makes the on-disk format
97 * incompatible with any other format that uses the same number.
99 * Data can usually be structured to work with one of the
100 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
103 } dmu_object_byteswap_t;
105 #define DMU_OT_NEWTYPE 0x80
106 #define DMU_OT_METADATA 0x40
107 #define DMU_OT_ENCRYPTED 0x20
108 #define DMU_OT_BYTESWAP_MASK 0x1f
111 * Defines a uint8_t object type. Object types specify if the data
112 * in the object is metadata (boolean) and how to byteswap the data
113 * (dmu_object_byteswap_t). All of the types created by this method
114 * are cached in the dbuf metadata cache.
116 #define DMU_OT(byteswap, metadata, encrypted) \
118 ((metadata) ? DMU_OT_METADATA : 0) | \
119 ((encrypted) ? DMU_OT_ENCRYPTED : 0) | \
120 ((byteswap) & DMU_OT_BYTESWAP_MASK))
122 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
123 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
124 (ot) < DMU_OT_NUMTYPES)
126 #define DMU_OT_IS_METADATA_CACHED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
127 B_TRUE : dmu_ot[(ot)].ot_dbuf_metadata_cache)
130 * MDB doesn't have dmu_ot; it defines these macros itself.
133 #define DMU_OT_IS_METADATA_IMPL(ot) (dmu_ot[ot].ot_metadata)
134 #define DMU_OT_IS_ENCRYPTED_IMPL(ot) (dmu_ot[ot].ot_encrypt)
135 #define DMU_OT_BYTESWAP_IMPL(ot) (dmu_ot[ot].ot_byteswap)
138 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
139 ((ot) & DMU_OT_METADATA) : \
140 DMU_OT_IS_METADATA_IMPL(ot))
142 #define DMU_OT_IS_DDT(ot) \
143 ((ot) == DMU_OT_DDT_ZAP)
145 #define DMU_OT_IS_ZIL(ot) \
146 ((ot) == DMU_OT_INTENT_LOG)
148 /* Note: ztest uses DMU_OT_UINT64_OTHER as a proxy for file blocks */
149 #define DMU_OT_IS_FILE(ot) \
150 ((ot) == DMU_OT_PLAIN_FILE_CONTENTS || (ot) == DMU_OT_UINT64_OTHER)
152 #define DMU_OT_IS_ENCRYPTED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
153 ((ot) & DMU_OT_ENCRYPTED) : \
154 DMU_OT_IS_ENCRYPTED_IMPL(ot))
157 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
158 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
159 * is repurposed for embedded BPs.
161 #define DMU_OT_HAS_FILL(ot) \
162 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
164 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
165 ((ot) & DMU_OT_BYTESWAP_MASK) : \
166 DMU_OT_BYTESWAP_IMPL(ot))
168 typedef enum dmu_object_type {
171 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
172 DMU_OT_OBJECT_ARRAY, /* UINT64 */
173 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
174 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
175 DMU_OT_BPOBJ, /* UINT64 */
176 DMU_OT_BPOBJ_HDR, /* UINT64 */
178 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
179 DMU_OT_SPACE_MAP, /* UINT64 */
181 DMU_OT_INTENT_LOG, /* UINT64 */
183 DMU_OT_DNODE, /* DNODE */
184 DMU_OT_OBJSET, /* OBJSET */
186 DMU_OT_DSL_DIR, /* UINT64 */
187 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
188 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
189 DMU_OT_DSL_PROPS, /* ZAP */
190 DMU_OT_DSL_DATASET, /* UINT64 */
192 DMU_OT_ZNODE, /* ZNODE */
193 DMU_OT_OLDACL, /* Old ACL */
194 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
195 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
196 DMU_OT_MASTER_NODE, /* ZAP */
197 DMU_OT_UNLINKED_SET, /* ZAP */
199 DMU_OT_ZVOL, /* UINT8 */
200 DMU_OT_ZVOL_PROP, /* ZAP */
201 /* other; for testing only! */
202 DMU_OT_PLAIN_OTHER, /* UINT8 */
203 DMU_OT_UINT64_OTHER, /* UINT64 */
204 DMU_OT_ZAP_OTHER, /* ZAP */
205 /* new object types: */
206 DMU_OT_ERROR_LOG, /* ZAP */
207 DMU_OT_SPA_HISTORY, /* UINT8 */
208 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
209 DMU_OT_POOL_PROPS, /* ZAP */
210 DMU_OT_DSL_PERMS, /* ZAP */
211 DMU_OT_ACL, /* ACL */
212 DMU_OT_SYSACL, /* SYSACL */
213 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
214 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
215 DMU_OT_NEXT_CLONES, /* ZAP */
216 DMU_OT_SCAN_QUEUE, /* ZAP */
217 DMU_OT_USERGROUP_USED, /* ZAP */
218 DMU_OT_USERGROUP_QUOTA, /* ZAP */
219 DMU_OT_USERREFS, /* ZAP */
220 DMU_OT_DDT_ZAP, /* ZAP */
221 DMU_OT_DDT_STATS, /* ZAP */
222 DMU_OT_SA, /* System attr */
223 DMU_OT_SA_MASTER_NODE, /* ZAP */
224 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
225 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
226 DMU_OT_SCAN_XLATE, /* ZAP */
227 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
228 DMU_OT_DEADLIST, /* ZAP */
229 DMU_OT_DEADLIST_HDR, /* UINT64 */
230 DMU_OT_DSL_CLONES, /* ZAP */
231 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
233 * Do not allocate new object types here. Doing so makes the on-disk
234 * format incompatible with any other format that uses the same object
237 * When creating an object which does not have one of the above types
238 * use the DMU_OTN_* type with the correct byteswap and metadata
241 * The DMU_OTN_* types do not have entries in the dmu_ot table,
242 * use the DMU_OT_IS_METADATA() and DMU_OT_BYTESWAP() macros instead
243 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
244 * and DMU_OTN_* types).
249 * Names for valid types declared with DMU_OT().
251 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_FALSE),
252 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_FALSE),
253 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_FALSE),
254 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_FALSE),
255 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_FALSE),
256 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_FALSE),
257 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_FALSE),
258 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_FALSE),
259 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_FALSE),
260 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_FALSE),
262 DMU_OTN_UINT8_ENC_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_TRUE),
263 DMU_OTN_UINT8_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_TRUE),
264 DMU_OTN_UINT16_ENC_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_TRUE),
265 DMU_OTN_UINT16_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_TRUE),
266 DMU_OTN_UINT32_ENC_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_TRUE),
267 DMU_OTN_UINT32_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_TRUE),
268 DMU_OTN_UINT64_ENC_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_TRUE),
269 DMU_OTN_UINT64_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_TRUE),
270 DMU_OTN_ZAP_ENC_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_TRUE),
271 DMU_OTN_ZAP_ENC_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_TRUE),
275 * These flags are intended to be used to specify the "txg_how"
276 * parameter when calling the dmu_tx_assign() function. See the comment
277 * above dmu_tx_assign() for more details on the meaning of these flags.
279 #define TXG_NOWAIT (0ULL)
280 #define TXG_WAIT (1ULL<<0)
281 #define TXG_NOTHROTTLE (1ULL<<1)
283 void byteswap_uint64_array(void *buf, size_t size);
284 void byteswap_uint32_array(void *buf, size_t size);
285 void byteswap_uint16_array(void *buf, size_t size);
286 void byteswap_uint8_array(void *buf, size_t size);
287 void zap_byteswap(void *buf, size_t size);
288 void zfs_oldacl_byteswap(void *buf, size_t size);
289 void zfs_acl_byteswap(void *buf, size_t size);
290 void zfs_znode_byteswap(void *buf, size_t size);
292 #define DS_FIND_SNAPSHOTS (1<<0)
293 #define DS_FIND_CHILDREN (1<<1)
294 #define DS_FIND_SERIALIZE (1<<2)
297 * The maximum number of bytes that can be accessed as part of one
298 * operation, including metadata.
300 #define DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
301 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
303 #define DMU_USERUSED_OBJECT (-1ULL)
304 #define DMU_GROUPUSED_OBJECT (-2ULL)
305 #define DMU_PROJECTUSED_OBJECT (-3ULL)
308 * Zap prefix for object accounting in DMU_{USER,GROUP,PROJECT}USED_OBJECT.
310 #define DMU_OBJACCT_PREFIX "obj-"
311 #define DMU_OBJACCT_PREFIX_LEN 4
314 * artificial blkids for bonus buffer and spill blocks
316 #define DMU_BONUS_BLKID (-1ULL)
317 #define DMU_SPILL_BLKID (-2ULL)
320 * Public routines to create, destroy, open, and close objsets.
322 typedef void dmu_objset_create_sync_func_t(objset_t *os, void *arg,
323 cred_t *cr, dmu_tx_t *tx);
325 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
326 int dmu_objset_own(const char *name, dmu_objset_type_t type,
327 boolean_t readonly, boolean_t key_required, void *tag, objset_t **osp);
328 void dmu_objset_rele(objset_t *os, void *tag);
329 void dmu_objset_disown(objset_t *os, boolean_t key_required, void *tag);
330 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
332 void dmu_objset_evict_dbufs(objset_t *os);
333 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
334 struct dsl_crypto_params *dcp, dmu_objset_create_sync_func_t func,
336 int dmu_objset_clone(const char *name, const char *origin);
337 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
338 struct nvlist *errlist);
339 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
340 int dmu_objset_snapshot_tmp(const char *, const char *, int);
341 int dmu_objset_find(const char *name, int func(const char *, void *), void *arg,
343 void dmu_objset_byteswap(void *buf, size_t size);
344 int dsl_dataset_rename_snapshot(const char *fsname,
345 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
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"
386 #define DMU_POOL_LOG_SPACEMAP_ZAP "com.delphix:log_spacemap_zap"
387 #define DMU_POOL_DELETED_CLONES "com.delphix:deleted_clones"
390 * Allocate an object from this objset. The range of object numbers
391 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
393 * The transaction must be assigned to a txg. The newly allocated
394 * object will be "held" in the transaction (ie. you can modify the
395 * newly allocated object in this transaction).
397 * dmu_object_alloc() chooses an object and returns it in *objectp.
399 * dmu_object_claim() allocates a specific object number. If that
400 * number is already allocated, it fails and returns EEXIST.
402 * Return 0 on success, or ENOSPC or EEXIST as specified above.
404 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
405 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
406 uint64_t dmu_object_alloc_ibs(objset_t *os, dmu_object_type_t ot, int blocksize,
407 int indirect_blockshift,
408 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
409 uint64_t dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot,
410 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
411 int dnodesize, dmu_tx_t *tx);
412 uint64_t dmu_object_alloc_hold(objset_t *os, dmu_object_type_t ot,
413 int blocksize, int indirect_blockshift, dmu_object_type_t bonustype,
414 int bonuslen, int dnodesize, dnode_t **allocated_dnode, void *tag,
416 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
417 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
418 int dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
419 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
420 int dnodesize, dmu_tx_t *tx);
421 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
422 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
423 int dmu_object_reclaim_dnsize(objset_t *os, uint64_t object,
424 dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype,
425 int bonuslen, int dnodesize, boolean_t keep_spill, dmu_tx_t *tx);
426 int dmu_object_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx);
429 * Free an object from this objset.
431 * The object's data will be freed as well (ie. you don't need to call
432 * dmu_free(object, 0, -1, tx)).
434 * The object need not be held in the transaction.
436 * If there are any holds on this object's buffers (via dmu_buf_hold()),
437 * or tx holds on the object (via dmu_tx_hold_object()), you can not
438 * free it; it fails and returns EBUSY.
440 * If the object is not allocated, it fails and returns ENOENT.
442 * Return 0 on success, or EBUSY or ENOENT as specified above.
444 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
447 * Find the next allocated or free object.
449 * The objectp parameter is in-out. It will be updated to be the next
450 * object which is allocated. Ignore objects which have not been
451 * modified since txg.
453 * XXX Can only be called on a objset with no dirty data.
455 * Returns 0 on success, or ENOENT if there are no more objects.
457 int dmu_object_next(objset_t *os, uint64_t *objectp,
458 boolean_t hole, uint64_t txg);
461 * Set the number of levels on a dnode. nlevels must be greater than the
462 * current number of levels or an EINVAL will be returned.
464 int dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels,
468 * Set the data blocksize for an object.
470 * The object cannot have any blocks allocated beyond the first. If
471 * the first block is allocated already, the new size must be greater
472 * than the current block size. If these conditions are not met,
473 * ENOTSUP will be returned.
475 * Returns 0 on success, or EBUSY if there are any holds on the object
476 * contents, or ENOTSUP as described above.
478 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
479 int ibs, dmu_tx_t *tx);
482 * Manually set the maxblkid on a dnode. This will adjust nlevels accordingly
483 * to accommodate the change. When calling this function, the caller must
484 * ensure that the object's nlevels can sufficiently support the new maxblkid.
486 int dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid,
490 * Set the checksum property on a dnode. The new checksum algorithm will
491 * apply to all newly written blocks; existing blocks will not be affected.
493 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
497 * Set the compress property on a dnode. The new compression algorithm will
498 * apply to all newly written blocks; existing blocks will not be affected.
500 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
503 void dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
504 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
505 int compressed_size, int byteorder, dmu_tx_t *tx);
506 void dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
510 * Decide how to write a block: checksum, compression, number of copies, etc.
512 #define WP_NOFILL 0x1
513 #define WP_DMU_SYNC 0x2
516 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
517 struct zio_prop *zp);
520 * The bonus data is accessed more or less like a regular buffer.
521 * You must dmu_bonus_hold() to get the buffer, which will give you a
522 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
523 * data. As with any normal buffer, you must call dmu_buf_will_dirty()
524 * before modifying it, and the
525 * object must be held in an assigned transaction before calling
526 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
527 * buffer as well. You must release what you hold with dmu_buf_rele().
529 * Returns ENOENT, EIO, or 0.
531 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp);
532 int dmu_bonus_hold_by_dnode(dnode_t *dn, void *tag, dmu_buf_t **dbp,
534 int dmu_bonus_max(void);
535 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
536 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
537 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
538 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
541 * Special spill buffer support used by "SA" framework
544 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag,
546 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
547 void *tag, dmu_buf_t **dbp);
548 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
551 * Obtain the DMU buffer from the specified object which contains the
552 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
553 * that it will remain in memory. You must release the hold with
554 * dmu_buf_rele(). You must not access the dmu_buf_t after releasing
555 * what you hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
557 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
558 * on the returned buffer before reading or writing the buffer's
559 * db_data. The comments for those routines describe what particular
560 * operations are valid after calling them.
562 * The object number must be a valid, allocated object number.
564 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
565 void *tag, dmu_buf_t **, int flags);
566 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
567 void *tag, dmu_buf_t **dbp, int flags);
568 int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset,
569 uint64_t length, boolean_t read, void *tag, int *numbufsp,
570 dmu_buf_t ***dbpp, uint32_t flags);
572 * Add a reference to a dmu buffer that has already been held via
573 * dmu_buf_hold() in the current context.
575 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
578 * Attempt to add a reference to a dmu buffer that is in an unknown state,
579 * using a pointer that may have been invalidated by eviction processing.
580 * The request will succeed if the passed in dbuf still represents the
581 * same os/object/blkid, is ineligible for eviction, and has at least
582 * one hold by a user other than the syncer.
584 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
585 uint64_t blkid, void *tag);
587 void dmu_buf_rele(dmu_buf_t *db, void *tag);
588 uint64_t dmu_buf_refcount(dmu_buf_t *db);
589 uint64_t dmu_buf_user_refcount(dmu_buf_t *db);
592 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
593 * range of an object. A pointer to an array of dmu_buf_t*'s is
594 * returned (in *dbpp).
596 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
597 * frees the array. The hold on the array of buffers MUST be released
598 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
599 * individually with dmu_buf_rele.
601 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
602 uint64_t length, boolean_t read, void *tag,
603 int *numbufsp, dmu_buf_t ***dbpp);
604 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
606 typedef void dmu_buf_evict_func_t(void *user_ptr);
609 * A DMU buffer user object may be associated with a dbuf for the
610 * duration of its lifetime. This allows the user of a dbuf (client)
611 * to attach private data to a dbuf (e.g. in-core only data such as a
612 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
613 * when that dbuf has been evicted. Clients typically respond to the
614 * eviction notification by freeing their private data, thus ensuring
615 * the same lifetime for both dbuf and private data.
617 * The mapping from a dmu_buf_user_t to any client private data is the
618 * client's responsibility. All current consumers of the API with private
619 * data embed a dmu_buf_user_t as the first member of the structure for
620 * their private data. This allows conversions between the two types
621 * with a simple cast. Since the DMU buf user API never needs access
622 * to the private data, other strategies can be employed if necessary
623 * or convenient for the client (e.g. using container_of() to do the
624 * conversion for private data that cannot have the dmu_buf_user_t as
627 * Eviction callbacks are executed without the dbuf mutex held or any
628 * other type of mechanism to guarantee that the dbuf is still available.
629 * For this reason, users must assume the dbuf has already been freed
630 * and not reference the dbuf from the callback context.
632 * Users requesting "immediate eviction" are notified as soon as the dbuf
633 * is only referenced by dirty records (dirties == holds). Otherwise the
634 * notification occurs after eviction processing for the dbuf begins.
636 typedef struct dmu_buf_user {
638 * Asynchronous user eviction callback state.
640 taskq_ent_t dbu_tqent;
643 * This instance's eviction function pointers.
645 * dbu_evict_func_sync is called synchronously and then
646 * dbu_evict_func_async is executed asynchronously on a taskq.
648 dmu_buf_evict_func_t *dbu_evict_func_sync;
649 dmu_buf_evict_func_t *dbu_evict_func_async;
652 * Pointer to user's dbuf pointer. NULL for clients that do
653 * not associate a dbuf with their user data.
655 * The dbuf pointer is cleared upon eviction so as to catch
656 * use-after-evict bugs in clients.
658 dmu_buf_t **dbu_clear_on_evict_dbufp;
663 * Initialize the given dmu_buf_user_t instance with the eviction function
664 * evict_func, to be called when the user is evicted.
666 * NOTE: This function should only be called once on a given dmu_buf_user_t.
667 * To allow enforcement of this, dbu must already be zeroed on entry.
671 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
672 dmu_buf_evict_func_t *evict_func_async,
673 dmu_buf_t **clear_on_evict_dbufp __maybe_unused)
675 ASSERT(dbu->dbu_evict_func_sync == NULL);
676 ASSERT(dbu->dbu_evict_func_async == NULL);
678 /* must have at least one evict func */
679 IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
680 dbu->dbu_evict_func_sync = evict_func_sync;
681 dbu->dbu_evict_func_async = evict_func_async;
682 taskq_init_ent(&dbu->dbu_tqent);
684 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
689 * Attach user data to a dbuf and mark it for normal (when the dbuf's
690 * data is cleared or its reference count goes to zero) eviction processing.
692 * Returns NULL on success, or the existing user if another user currently
695 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
698 * Attach user data to a dbuf and mark it for immediate (its dirty and
699 * reference counts are equal) eviction processing.
701 * Returns NULL on success, or the existing user if another user currently
704 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
707 * Replace the current user of a dbuf.
709 * If given the current user of a dbuf, replaces the dbuf's user with
710 * "new_user" and returns the user data pointer that was replaced.
711 * Otherwise returns the current, and unmodified, dbuf user pointer.
713 void *dmu_buf_replace_user(dmu_buf_t *db,
714 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
717 * Remove the specified user data for a DMU buffer.
719 * Returns the user that was removed on success, or the current user if
720 * another user currently owns the buffer.
722 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
725 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
727 void *dmu_buf_get_user(dmu_buf_t *db);
729 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
730 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
731 void dmu_buf_dnode_exit(dmu_buf_t *db);
733 /* Block until any in-progress dmu buf user evictions complete. */
734 void dmu_buf_user_evict_wait(void);
737 * Returns the blkptr associated with this dbuf, or NULL if not set.
739 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
742 * Indicate that you are going to modify the buffer's data (db_data).
744 * The transaction (tx) must be assigned to a txg (ie. you've called
745 * dmu_tx_assign()). The buffer's object must be held in the tx
746 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
748 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
749 boolean_t dmu_buf_is_dirty(dmu_buf_t *db, dmu_tx_t *tx);
750 void dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder,
751 const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx);
754 * You must create a transaction, then hold the objects which you will
755 * (or might) modify as part of this transaction. Then you must assign
756 * the transaction to a transaction group. Once the transaction has
757 * been assigned, you can modify buffers which belong to held objects as
758 * part of this transaction. You can't modify buffers before the
759 * transaction has been assigned; you can't modify buffers which don't
760 * belong to objects which this transaction holds; you can't hold
761 * objects once the transaction has been assigned. You may hold an
762 * object which you are going to free (with dmu_object_free()), but you
765 * You can abort the transaction before it has been assigned.
767 * Note that you may hold buffers (with dmu_buf_hold) at any time,
768 * regardless of transaction state.
771 #define DMU_NEW_OBJECT (-1ULL)
772 #define DMU_OBJECT_END (-1ULL)
774 dmu_tx_t *dmu_tx_create(objset_t *os);
775 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
776 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
778 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
780 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
782 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
783 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
785 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
786 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
787 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
788 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
789 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
790 void dmu_tx_abort(dmu_tx_t *tx);
791 int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
792 void dmu_tx_wait(dmu_tx_t *tx);
793 void dmu_tx_commit(dmu_tx_t *tx);
794 void dmu_tx_mark_netfree(dmu_tx_t *tx);
797 * To register a commit callback, dmu_tx_callback_register() must be called.
799 * dcb_data is a pointer to caller private data that is passed on as a
800 * callback parameter. The caller is responsible for properly allocating and
803 * When registering a callback, the transaction must be already created, but
804 * it cannot be committed or aborted. It can be assigned to a txg or not.
806 * The callback will be called after the transaction has been safely written
807 * to stable storage and will also be called if the dmu_tx is aborted.
808 * If there is any error which prevents the transaction from being committed to
809 * disk, the callback will be called with a value of error != 0.
811 * When multiple callbacks are registered to the transaction, the callbacks
812 * will be called in reverse order to let Lustre, the only user of commit
813 * callback currently, take the fast path of its commit callback handling.
815 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
817 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
819 void dmu_tx_do_callbacks(list_t *cb_list, int error);
822 * Free up the data blocks for a defined range of a file. If size is
823 * -1, the range from offset to end-of-file is freed.
825 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
826 uint64_t size, dmu_tx_t *tx);
827 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
829 int dmu_free_long_object(objset_t *os, uint64_t object);
832 * Convenience functions.
834 * Canfail routines will return 0 on success, or an errno if there is a
835 * nonrecoverable I/O error.
837 #define DMU_READ_PREFETCH 0 /* prefetch */
838 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
839 #define DMU_READ_NO_DECRYPT 2 /* don't decrypt */
840 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
841 void *buf, uint32_t flags);
842 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
844 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
845 const void *buf, dmu_tx_t *tx);
846 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
847 const void *buf, dmu_tx_t *tx);
848 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
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 #ifdef HAVE_UIO_ZEROCOPY
869 int dmu_xuio_init(struct xuio *uio, int niov);
870 void dmu_xuio_fini(struct xuio *uio);
871 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
873 int dmu_xuio_cnt(struct xuio *uio);
874 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
875 void dmu_xuio_clear(struct xuio *uio, int i);
876 #endif /* HAVE_UIO_ZEROCOPY */
877 void xuio_stat_wbuf_copied(void);
878 void xuio_stat_wbuf_nocopy(void);
880 extern int zfs_prefetch_disable;
881 extern int zfs_max_recordsize;
884 * Asynchronously try to read in the data.
886 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
887 uint64_t len, enum zio_priority pri);
889 typedef struct dmu_object_info {
890 /* All sizes are in bytes unless otherwise indicated. */
891 uint32_t doi_data_block_size;
892 uint32_t doi_metadata_block_size;
893 dmu_object_type_t doi_type;
894 dmu_object_type_t doi_bonus_type;
895 uint64_t doi_bonus_size;
896 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
897 uint8_t doi_checksum;
898 uint8_t doi_compress;
901 uint64_t doi_dnodesize;
902 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
903 uint64_t doi_max_offset;
904 uint64_t doi_fill_count; /* number of non-empty blocks */
907 typedef void (*const arc_byteswap_func_t)(void *buf, size_t size);
909 typedef struct dmu_object_type_info {
910 dmu_object_byteswap_t ot_byteswap;
911 boolean_t ot_metadata;
912 boolean_t ot_dbuf_metadata_cache;
913 boolean_t ot_encrypt;
915 } dmu_object_type_info_t;
917 typedef const struct dmu_object_byteswap_info {
918 arc_byteswap_func_t ob_func;
920 } dmu_object_byteswap_info_t;
922 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
923 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
926 * Get information on a DMU object.
928 * Return 0 on success or ENOENT if object is not allocated.
930 * If doi is NULL, just indicates whether the object exists.
932 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
933 void __dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
934 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
935 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
936 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
937 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
939 * Like dmu_object_info_from_db, but faster still when you only care about
942 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
943 u_longlong_t *nblk512);
945 void dmu_object_dnsize_from_db(dmu_buf_t *db, int *dnsize);
947 typedef struct dmu_objset_stats {
948 uint64_t dds_num_clones; /* number of clones of this */
949 uint64_t dds_creation_txg;
951 dmu_objset_type_t dds_type;
952 uint8_t dds_is_snapshot;
953 uint8_t dds_inconsistent;
954 uint8_t dds_redacted;
955 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
956 } dmu_objset_stats_t;
959 * Get stats on a dataset.
961 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
964 * Add entries to the nvlist for all the objset's properties. See
965 * zfs_prop_table[] and zfs(1m) for details on the properties.
967 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
970 * Get the space usage statistics for statvfs().
972 * refdbytes is the amount of space "referenced" by this objset.
973 * availbytes is the amount of space available to this objset, taking
974 * into account quotas & reservations, assuming that no other objsets
975 * use the space first. These values correspond to the 'referenced' and
976 * 'available' properties, described in the zfs(1m) manpage.
978 * usedobjs and availobjs are the number of objects currently allocated,
981 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
982 uint64_t *usedobjsp, uint64_t *availobjsp);
985 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
986 * (Contrast with the ds_guid which is a 64-bit ID that will never
987 * change, so there is a small probability that it will collide.)
989 uint64_t dmu_objset_fsid_guid(objset_t *os);
992 * Get the [cm]time for an objset's snapshot dir
994 inode_timespec_t dmu_objset_snap_cmtime(objset_t *os);
996 int dmu_objset_is_snapshot(objset_t *os);
998 extern struct spa *dmu_objset_spa(objset_t *os);
999 extern struct zilog *dmu_objset_zil(objset_t *os);
1000 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
1001 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
1002 extern void dmu_objset_name(objset_t *os, char *buf);
1003 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
1004 extern uint64_t dmu_objset_id(objset_t *os);
1005 extern uint64_t dmu_objset_dnodesize(objset_t *os);
1006 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
1007 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
1008 extern int dmu_objset_blksize(objset_t *os);
1009 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1010 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
1011 extern int dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *val);
1012 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
1013 int maxlen, boolean_t *conflict);
1014 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
1015 uint64_t *idp, uint64_t *offp);
1017 typedef struct zfs_file_info {
1020 uint64_t zfi_project;
1021 uint64_t zfi_generation;
1024 typedef int file_info_cb_t(dmu_object_type_t bonustype, const void *data,
1025 struct zfs_file_info *zoi);
1026 extern void dmu_objset_register_type(dmu_objset_type_t ost,
1027 file_info_cb_t *cb);
1028 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
1029 extern void *dmu_objset_get_user(objset_t *os);
1032 * Return the txg number for the given assigned transaction.
1034 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
1037 * Synchronous write.
1038 * If a parent zio is provided this function initiates a write on the
1039 * provided buffer as a child of the parent zio.
1040 * In the absence of a parent zio, the write is completed synchronously.
1041 * At write completion, blk is filled with the bp of the written block.
1042 * Note that while the data covered by this function will be on stable
1043 * storage when the write completes this new data does not become a
1044 * permanent part of the file until the associated transaction commits.
1048 * {zfs,zvol,ztest}_get_done() args
1050 typedef struct zgd {
1051 struct lwb *zgd_lwb;
1052 struct blkptr *zgd_bp;
1054 struct zfs_locked_range *zgd_lr;
1058 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
1059 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
1062 * Find the next hole or data block in file starting at *off
1063 * Return found offset in *off. Return ESRCH for end of file.
1065 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
1069 * Initial setup and final teardown.
1071 extern void dmu_init(void);
1072 extern void dmu_fini(void);
1074 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
1075 uint64_t object, uint64_t offset, int len);
1076 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
1077 dmu_traverse_cb_t cb, void *arg);
1079 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
1080 zfs_file_t *fp, offset_t *offp);
1083 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
1084 extern uint64_t zfs_crc64_table[256];
1090 #endif /* _SYS_DMU_H */