1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * This file and its contents are supplied under the terms of the
4 * Common Development and Distribution License ("CDDL"), version 1.0.
5 * You may only use this file in accordance with the terms of version
6 * 1.0 of the CDDL.
7 *
8 * A full copy of the text of the CDDL should have accompanied this
9 * source. A copy of the CDDL is also available via the Internet at
10 * https://opensource.org/license/CDDL-1.0.
11 */
12 /*
13 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
14 * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
15 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
16 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
17 * Copyright 2014 HybridCluster. All rights reserved.
18 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
19 * Copyright 2013 Saso Kiselkov. All rights reserved.
20 * Copyright (c) 2017, Intel Corporation.
21 * Copyright (c) 2022 Hewlett Packard Enterprise Development LP.
22 * Copyright (c) 2025, Klara, Inc.
23 */
24
25 /* Portions Copyright 2010 Robert Milkowski */
26
27 #ifndef _SYS_DMU_H
28 #define _SYS_DMU_H
29
30 /*
31 * This file describes the interface that the DMU provides for its
32 * consumers.
33 *
34 * The DMU also interacts with the SPA. That interface is described in
35 * dmu_spa.h.
36 */
37
38 #include <sys/zfs_context.h>
39 #include <sys/inttypes.h>
40 #include <sys/cred.h>
41 #include <sys/fs/zfs.h>
42 #include <sys/zio_compress.h>
43 #include <sys/uio.h>
44 #include <sys/zfs_file.h>
45
46 #ifdef __cplusplus
47 extern "C" {
48 #endif
49
50 struct page;
51 struct vnode;
52 struct spa;
53 struct zilog;
54 struct zio;
55 struct blkptr;
56 struct zap_cursor;
57 struct dsl_dataset;
58 struct dsl_pool;
59 struct dnode;
60 struct drr_begin;
61 struct drr_end;
62 struct zbookmark_phys;
63 struct spa;
64 struct nvlist;
65 struct arc_buf;
66 struct zio_prop;
67 struct sa_handle;
68 struct dsl_crypto_params;
69 struct locked_range;
70
71 typedef struct objset objset_t;
72 typedef struct dmu_tx dmu_tx_t;
73 typedef struct dsl_dir dsl_dir_t;
74 typedef struct dnode dnode_t;
75
76 typedef enum dmu_object_byteswap {
77 DMU_BSWAP_UINT8,
78 DMU_BSWAP_UINT16,
79 DMU_BSWAP_UINT32,
80 DMU_BSWAP_UINT64,
81 DMU_BSWAP_ZAP,
82 DMU_BSWAP_DNODE,
83 DMU_BSWAP_OBJSET,
84 DMU_BSWAP_ZNODE,
85 DMU_BSWAP_OLDACL,
86 DMU_BSWAP_ACL,
87 /*
88 * Allocating a new byteswap type number makes the on-disk format
89 * incompatible with any other format that uses the same number.
90 *
91 * Data can usually be structured to work with one of the
92 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
93 */
94 DMU_BSWAP_NUMFUNCS
95 } dmu_object_byteswap_t;
96
97 #define DMU_OT_NEWTYPE 0x80
98 #define DMU_OT_METADATA 0x40
99 #define DMU_OT_ENCRYPTED 0x20
100 #define DMU_OT_BYTESWAP_MASK 0x1f
101
102 /*
103 * Defines a uint8_t object type. Object types specify if the data
104 * in the object is metadata (boolean) and how to byteswap the data
105 * (dmu_object_byteswap_t). All of the types created by this method
106 * are cached in the dbuf metadata cache.
107 */
108 #define DMU_OT(byteswap, metadata, encrypted) \
109 (DMU_OT_NEWTYPE | \
110 ((metadata) ? DMU_OT_METADATA : 0) | \
111 ((encrypted) ? DMU_OT_ENCRYPTED : 0) | \
112 ((byteswap) & DMU_OT_BYTESWAP_MASK))
113
114 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
115 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
116 (ot) < DMU_OT_NUMTYPES)
117
118 #define DMU_OT_IS_METADATA_CACHED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
119 ((ot) & DMU_OT_METADATA) != 0 : dmu_ot[(ot)].ot_dbuf_metadata_cache)
120
121 /*
122 * MDB doesn't have dmu_ot; it defines these macros itself.
123 */
124 #ifndef ZFS_MDB
125 #define DMU_OT_IS_METADATA_IMPL(ot) (dmu_ot[ot].ot_metadata)
126 #define DMU_OT_IS_ENCRYPTED_IMPL(ot) (dmu_ot[ot].ot_encrypt)
127 #define DMU_OT_BYTESWAP_IMPL(ot) (dmu_ot[ot].ot_byteswap)
128 #endif
129
130 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
131 (((ot) & DMU_OT_METADATA) != 0) : \
132 DMU_OT_IS_METADATA_IMPL(ot))
133
134 #define DMU_OT_IS_DDT(ot) \
135 ((ot) == DMU_OT_DDT_ZAP)
136
137 #define DMU_OT_IS_CRITICAL(ot, level) \
138 (DMU_OT_IS_METADATA(ot) && \
139 ((ot) != DMU_OT_DNODE || (level) > 0) && \
140 (ot) != DMU_OT_DIRECTORY_CONTENTS && \
141 (ot) != DMU_OT_SA)
142
143 /* Note: ztest uses DMU_OT_UINT64_OTHER as a proxy for file blocks */
144 #define DMU_OT_IS_FILE(ot) \
145 ((ot) == DMU_OT_PLAIN_FILE_CONTENTS || (ot) == DMU_OT_UINT64_OTHER)
146
147 #define DMU_OT_IS_ENCRYPTED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
148 (((ot) & DMU_OT_ENCRYPTED) != 0) : \
149 DMU_OT_IS_ENCRYPTED_IMPL(ot))
150
151 /*
152 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
153 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
154 * is repurposed for embedded BPs.
155 */
156 #define DMU_OT_HAS_FILL(ot) \
157 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
158
159 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
160 ((ot) & DMU_OT_BYTESWAP_MASK) : \
161 DMU_OT_BYTESWAP_IMPL(ot))
162
163 typedef enum dmu_object_type {
164 DMU_OT_NONE,
165 /* general: */
166 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
167 DMU_OT_OBJECT_ARRAY, /* UINT64 */
168 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
169 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
170 DMU_OT_BPOBJ, /* UINT64 */
171 DMU_OT_BPOBJ_HDR, /* UINT64 */
172 /* spa: */
173 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
174 DMU_OT_SPACE_MAP, /* UINT64 */
175 /* zil: */
176 DMU_OT_INTENT_LOG, /* UINT64 */
177 /* dmu: */
178 DMU_OT_DNODE, /* DNODE */
179 DMU_OT_OBJSET, /* OBJSET */
180 /* dsl: */
181 DMU_OT_DSL_DIR, /* UINT64 */
182 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
183 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
184 DMU_OT_DSL_PROPS, /* ZAP */
185 DMU_OT_DSL_DATASET, /* UINT64 */
186 /* zpl: */
187 DMU_OT_ZNODE, /* ZNODE */
188 DMU_OT_OLDACL, /* Old ACL */
189 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
190 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
191 DMU_OT_MASTER_NODE, /* ZAP */
192 DMU_OT_UNLINKED_SET, /* ZAP */
193 /* zvol: */
194 DMU_OT_ZVOL, /* UINT8 */
195 DMU_OT_ZVOL_PROP, /* ZAP */
196 /* other; for testing only! */
197 DMU_OT_PLAIN_OTHER, /* UINT8 */
198 DMU_OT_UINT64_OTHER, /* UINT64 */
199 DMU_OT_ZAP_OTHER, /* ZAP */
200 /* new object types: */
201 DMU_OT_ERROR_LOG, /* ZAP */
202 DMU_OT_SPA_HISTORY, /* UINT8 */
203 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
204 DMU_OT_POOL_PROPS, /* ZAP */
205 DMU_OT_DSL_PERMS, /* ZAP */
206 DMU_OT_ACL, /* ACL */
207 DMU_OT_SYSACL, /* SYSACL */
208 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
209 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
210 DMU_OT_NEXT_CLONES, /* ZAP */
211 DMU_OT_SCAN_QUEUE, /* ZAP */
212 DMU_OT_USERGROUP_USED, /* ZAP */
213 DMU_OT_USERGROUP_QUOTA, /* ZAP */
214 DMU_OT_USERREFS, /* ZAP */
215 DMU_OT_DDT_ZAP, /* ZAP */
216 DMU_OT_DDT_STATS, /* ZAP */
217 DMU_OT_SA, /* System attr */
218 DMU_OT_SA_MASTER_NODE, /* ZAP */
219 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
220 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
221 DMU_OT_SCAN_XLATE, /* ZAP */
222 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
223 DMU_OT_DEADLIST, /* ZAP */
224 DMU_OT_DEADLIST_HDR, /* UINT64 */
225 DMU_OT_DSL_CLONES, /* ZAP */
226 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
227 /*
228 * Do not allocate new object types here. Doing so makes the on-disk
229 * format incompatible with any other format that uses the same object
230 * type number.
231 *
232 * When creating an object which does not have one of the above types
233 * use the DMU_OTN_* type with the correct byteswap and metadata
234 * values.
235 *
236 * The DMU_OTN_* types do not have entries in the dmu_ot table,
237 * use the DMU_OT_IS_METADATA() and DMU_OT_BYTESWAP() macros instead
238 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
239 * and DMU_OTN_* types).
240 */
241 DMU_OT_NUMTYPES,
242
243 /*
244 * Names for valid types declared with DMU_OT().
245 */
246 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_FALSE),
247 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_FALSE),
248 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_FALSE),
249 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_FALSE),
250 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_FALSE),
251 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_FALSE),
252 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_FALSE),
253 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_FALSE),
254 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_FALSE),
255 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_FALSE),
256
257 DMU_OTN_UINT8_ENC_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_TRUE),
258 DMU_OTN_UINT8_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_TRUE),
259 DMU_OTN_UINT16_ENC_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_TRUE),
260 DMU_OTN_UINT16_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_TRUE),
261 DMU_OTN_UINT32_ENC_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_TRUE),
262 DMU_OTN_UINT32_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_TRUE),
263 DMU_OTN_UINT64_ENC_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_TRUE),
264 DMU_OTN_UINT64_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_TRUE),
265 DMU_OTN_ZAP_ENC_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_TRUE),
266 DMU_OTN_ZAP_ENC_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_TRUE),
267 } dmu_object_type_t;
268
269 /*
270 * These flags are for the dmu_tx_assign() function and describe what to do if
271 * the transaction is full. See the comment above dmu_tx_assign() for more
272 * details on the meaning of these flags.
273 */
274 typedef enum {
275 /*
276 * If the tx cannot be assigned to a transaction for any reason, do
277 * not block but return immediately.
278 */
279 DMU_TX_NOWAIT = 0,
280
281 /*
282 * Assign the tx to the open transaction. If the open transaction is
283 * full, or the write throttle is active, block until the next
284 * transaction and try again. If the pool suspends while waiting
285 * and failmode=continue, return an error.
286 */
287 DMU_TX_WAIT = (1 << 0),
288
289 /* If the write throttle would prevent the assignment, ignore it. */
290 DMU_TX_NOTHROTTLE = (1 << 1),
291
292 /*
293 * With DMU_TX_WAIT, always block if the pool suspends during
294 * assignment, regardless of the value of the failmode= property.
295 */
296 DMU_TX_SUSPEND = (1 << 2),
297 } dmu_tx_flag_t;
298
299 void byteswap_uint64_array(void *buf, size_t size);
300 void byteswap_uint32_array(void *buf, size_t size);
301 void byteswap_uint16_array(void *buf, size_t size);
302 void byteswap_uint8_array(void *buf, size_t size);
303 void zap_byteswap(void *buf, size_t size);
304 void zfs_oldacl_byteswap(void *buf, size_t size);
305 void zfs_acl_byteswap(void *buf, size_t size);
306 void zfs_znode_byteswap(void *buf, size_t size);
307
308 #define DS_FIND_SNAPSHOTS (1<<0)
309 #define DS_FIND_CHILDREN (1<<1)
310 #define DS_FIND_SERIALIZE (1<<2)
311
312 /*
313 * The maximum number of bytes that can be accessed as part of one
314 * operation, including metadata.
315 */
316 #define DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
317 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
318
319 #define DMU_USERUSED_OBJECT (-1ULL)
320 #define DMU_GROUPUSED_OBJECT (-2ULL)
321 #define DMU_PROJECTUSED_OBJECT (-3ULL)
322
323 /*
324 * Zap prefix for object accounting in DMU_{USER,GROUP,PROJECT}USED_OBJECT.
325 */
326 #define DMU_OBJACCT_PREFIX "obj-"
327 #define DMU_OBJACCT_PREFIX_LEN 4
328
329 /*
330 * artificial blkids for bonus buffer and spill blocks
331 */
332 #define DMU_BONUS_BLKID (-1ULL)
333 #define DMU_SPILL_BLKID (-2ULL)
334
335 /*
336 * Public routines to create, destroy, open, and close objsets.
337 */
338 typedef void dmu_objset_create_sync_func_t(objset_t *os, void *arg,
339 cred_t *cr, dmu_tx_t *tx);
340
341 int dmu_objset_hold(const char *name, const void *tag, objset_t **osp);
342 int dmu_objset_own(const char *name, dmu_objset_type_t type,
343 boolean_t readonly, boolean_t key_required, const void *tag,
344 objset_t **osp);
345 void dmu_objset_rele(objset_t *os, const void *tag);
346 void dmu_objset_disown(objset_t *os, boolean_t key_required, const void *tag);
347 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
348
349 void dmu_objset_evict_dbufs(objset_t *os);
350 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
351 struct dsl_crypto_params *dcp, dmu_objset_create_sync_func_t func,
352 void *arg);
353 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
354 struct nvlist *errlist);
355 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
356 int dmu_objset_find(const char *name, int func(const char *, void *), void *arg,
357 int flags);
358 void dmu_objset_byteswap(void *buf, size_t size);
359 int dsl_dataset_rename_snapshot(const char *fsname,
360 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
361
362 typedef struct dmu_buf {
363 uint64_t db_object; /* object that this buffer is part of */
364 uint64_t db_offset; /* byte offset in this object */
365 uint64_t db_size; /* size of buffer in bytes */
366 void *db_data; /* data in buffer */
367 } dmu_buf_t;
368
369 /*
370 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
371 */
372 #define DMU_POOL_DIRECTORY_OBJECT 1
373 #define DMU_POOL_CONFIG "config"
374 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
375 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
376 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
377 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
378 #define DMU_POOL_ROOT_DATASET "root_dataset"
379 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
380 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
381 #define DMU_POOL_ERRLOG_LAST "errlog_last"
382 #define DMU_POOL_SPARES "spares"
383 #define DMU_POOL_DEFLATE "deflate"
384 #define DMU_POOL_HISTORY "history"
385 #define DMU_POOL_PROPS "pool_props"
386 #define DMU_POOL_L2CACHE "l2cache"
387 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
388 #define DMU_POOL_DDT "DDT-%s-%s-%s"
389 #define DMU_POOL_DDT_LOG "DDT-log-%s-%u"
390 #define DMU_POOL_DDT_STATS "DDT-statistics"
391 #define DMU_POOL_DDT_DIR "DDT-%s"
392 #define DMU_POOL_CREATION_VERSION "creation_version"
393 #define DMU_POOL_SCAN "scan"
394 #define DMU_POOL_ERRORSCRUB "error_scrub"
395 #define DMU_POOL_LAST_SCRUBBED_TXG "last_scrubbed_txg"
396 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
397 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
398 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
399 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
400 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
401 #define DMU_POOL_REMOVING "com.delphix:removing"
402 #define DMU_POOL_OBSOLETE_BPOBJ "com.delphix:obsolete_bpobj"
403 #define DMU_POOL_CONDENSING_INDIRECT "com.delphix:condensing_indirect"
404 #define DMU_POOL_ZPOOL_CHECKPOINT "com.delphix:zpool_checkpoint"
405 #define DMU_POOL_LOG_SPACEMAP_ZAP "com.delphix:log_spacemap_zap"
406 #define DMU_POOL_DELETED_CLONES "com.delphix:deleted_clones"
407 #define DMU_POOL_TXG_LOG_TIME_MINUTES "com.klarasystems:txg_log_time:minutes"
408 #define DMU_POOL_TXG_LOG_TIME_DAYS "com.klarasystems:txg_log_time:days"
409 #define DMU_POOL_TXG_LOG_TIME_MONTHS "com.klarasystems:txg_log_time:months"
410
411 /*
412 * Allocate an object from this objset. The range of object numbers
413 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
414 *
415 * The transaction must be assigned to a txg. The newly allocated
416 * object will be "held" in the transaction (ie. you can modify the
417 * newly allocated object in this transaction).
418 *
419 * dmu_object_alloc() chooses an object and returns it in *objectp.
420 *
421 * dmu_object_claim() allocates a specific object number. If that
422 * number is already allocated, it fails and returns EEXIST.
423 *
424 * Return 0 on success, or ENOSPC or EEXIST as specified above.
425 */
426 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
427 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
428 uint64_t dmu_object_alloc_ibs(objset_t *os, dmu_object_type_t ot, int blocksize,
429 int indirect_blockshift,
430 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
431 uint64_t dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot,
432 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
433 int dnodesize, dmu_tx_t *tx);
434 uint64_t dmu_object_alloc_hold(objset_t *os, dmu_object_type_t ot,
435 int blocksize, int indirect_blockshift, dmu_object_type_t bonustype,
436 int bonuslen, int dnodesize, dnode_t **allocated_dnode, const void *tag,
437 dmu_tx_t *tx);
438 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
439 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
440 int dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
441 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
442 int dnodesize, dmu_tx_t *tx);
443 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
444 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
445 int dmu_object_reclaim_dnsize(objset_t *os, uint64_t object,
446 dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype,
447 int bonuslen, int dnodesize, boolean_t keep_spill, dmu_tx_t *tx);
448 int dmu_object_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx);
449
450 /*
451 * Free an object from this objset.
452 *
453 * The object's data will be freed as well (ie. you don't need to call
454 * dmu_free(object, 0, -1, tx)).
455 *
456 * The object need not be held in the transaction.
457 *
458 * If there are any holds on this object's buffers (via dmu_buf_hold()),
459 * or tx holds on the object (via dmu_tx_hold_object()), you can not
460 * free it; it fails and returns EBUSY.
461 *
462 * If the object is not allocated, it fails and returns ENOENT.
463 *
464 * Return 0 on success, or EBUSY or ENOENT as specified above.
465 */
466 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
467
468 /*
469 * Find the next allocated or free object.
470 *
471 * The objectp parameter is in-out. It will be updated to be the next
472 * object which is allocated. Ignore objects which have not been
473 * modified since txg.
474 *
475 * XXX Can only be called on a objset with no dirty data.
476 *
477 * Returns 0 on success, or ENOENT if there are no more objects.
478 */
479 int dmu_object_next(objset_t *os, uint64_t *objectp,
480 boolean_t hole, uint64_t txg);
481
482 /*
483 * Set the number of levels on a dnode. nlevels must be greater than the
484 * current number of levels or an EINVAL will be returned.
485 */
486 int dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels,
487 dmu_tx_t *tx);
488
489 /*
490 * Set the data blocksize for an object.
491 *
492 * The object cannot have any blocks allocated beyond the first. If
493 * the first block is allocated already, the new size must be greater
494 * than the current block size. If these conditions are not met,
495 * ENOTSUP will be returned.
496 *
497 * Returns 0 on success, or EBUSY if there are any holds on the object
498 * contents, or ENOTSUP as described above.
499 */
500 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
501 int ibs, dmu_tx_t *tx);
502
503 /*
504 * Manually set the maxblkid on a dnode. This will adjust nlevels accordingly
505 * to accommodate the change. When calling this function, the caller must
506 * ensure that the object's nlevels can sufficiently support the new maxblkid.
507 */
508 int dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid,
509 dmu_tx_t *tx);
510
511 /*
512 * Set the checksum property on a dnode. The new checksum algorithm will
513 * apply to all newly written blocks; existing blocks will not be affected.
514 */
515 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
516 dmu_tx_t *tx);
517
518 /*
519 * Set the compress property on a dnode. The new compression algorithm will
520 * apply to all newly written blocks; existing blocks will not be affected.
521 */
522 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
523 dmu_tx_t *tx);
524
525 /*
526 * Get an estimated cache size for an object. Caller must expect races.
527 */
528 int dmu_object_cached_size(objset_t *os, uint64_t object,
529 uint64_t *l1sz, uint64_t *l2sz);
530
531 void dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
532 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
533 int compressed_size, int byteorder, dmu_tx_t *tx);
534 void dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
535 dmu_tx_t *tx);
536
537 /*
538 * Decide how to write a block: checksum, compression, number of copies, etc.
539 */
540 #define WP_NOFILL 0x1
541 #define WP_DMU_SYNC 0x2
542 #define WP_SPILL 0x4
543 #define WP_DIRECT_WR 0x8
544
545 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
546 struct zio_prop *zp);
547
548 /*
549 * DB_RF_* are to be used for dbuf_read() or in limited other cases.
550 */
551 typedef enum dmu_flags {
552 DB_RF_MUST_SUCCEED = 0, /* Suspend on I/O errors. */
553 DB_RF_CANFAIL = 1 << 0, /* Return on I/O errors. */
554 DB_RF_HAVESTRUCT = 1 << 1, /* dn_struct_rwlock is locked. */
555 DB_RF_NEVERWAIT = 1 << 2,
556 DMU_READ_PREFETCH = 0, /* Try speculative prefetch. */
557 DMU_READ_NO_PREFETCH = 1 << 3, /* Don't prefetch speculatively. */
558 DB_RF_NOPREFETCH = DMU_READ_NO_PREFETCH,
559 DMU_READ_NO_DECRYPT = 1 << 4, /* Don't decrypt. */
560 DB_RF_NO_DECRYPT = DMU_READ_NO_DECRYPT,
561 DMU_DIRECTIO = 1 << 5, /* Bypass ARC. */
562 DMU_UNCACHEDIO = 1 << 6, /* Reduce caching. */
563 DMU_PARTIAL_FIRST = 1 << 7, /* First partial access. */
564 DMU_PARTIAL_MORE = 1 << 8, /* Following partial access. */
565 DMU_KEEP_CACHING = 1 << 9, /* Don't affect caching. */
566 DMU_IS_PREFETCH = 1 << 10, /* This read is a prefetch. */
567 } dmu_flags_t;
568
569 /*
570 * The bonus data is accessed more or less like a regular buffer.
571 * You must dmu_bonus_hold() to get the buffer, which will give you a
572 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
573 * data. As with any normal buffer, you must call dmu_buf_will_dirty()
574 * before modifying it, and the
575 * object must be held in an assigned transaction before calling
576 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
577 * buffer as well. You must release what you hold with dmu_buf_rele().
578 *
579 * Returns ENOENT, EIO, or 0.
580 */
581 int dmu_bonus_hold(objset_t *os, uint64_t object, const void *tag,
582 dmu_buf_t **dbp);
583 int dmu_bonus_hold_by_dnode(dnode_t *dn, const void *tag, dmu_buf_t **dbp,
584 dmu_flags_t flags);
585 int dmu_bonus_max(void);
586 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
587 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
588 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
589 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
590
591 /*
592 * Special spill buffer support used by "SA" framework
593 */
594
595 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, dmu_flags_t flags,
596 const void *tag, dmu_buf_t **dbp);
597 int dmu_spill_hold_by_dnode(dnode_t *dn, dmu_flags_t flags,
598 const void *tag, dmu_buf_t **dbp);
599 int dmu_spill_hold_existing(dmu_buf_t *bonus, const void *tag, dmu_buf_t **dbp);
600
601 /*
602 * Obtain the DMU buffer from the specified object which contains the
603 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
604 * that it will remain in memory. You must release the hold with
605 * dmu_buf_rele(). You must not access the dmu_buf_t after releasing
606 * what you hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
607 *
608 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
609 * on the returned buffer before reading or writing the buffer's
610 * db_data. The comments for those routines describe what particular
611 * operations are valid after calling them.
612 *
613 * The object number must be a valid, allocated object number.
614 */
615 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
616 const void *tag, dmu_buf_t **, dmu_flags_t flags);
617 int dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
618 uint64_t length, int read, const void *tag, int *numbufsp,
619 dmu_buf_t ***dbpp, dmu_flags_t flags);
620 int dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
621 const void *tag, dmu_buf_t **dbp);
622 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
623 const void *tag, dmu_buf_t **dbp, dmu_flags_t flags);
624 int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset,
625 uint64_t length, boolean_t read, const void *tag, int *numbufsp,
626 dmu_buf_t ***dbpp, dmu_flags_t flags);
627 int dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, const void *tag,
628 dmu_buf_t **dbp);
629
630 /*
631 * Add a reference to a dmu buffer that has already been held via
632 * dmu_buf_hold() in the current context.
633 */
634 void dmu_buf_add_ref(dmu_buf_t *db, const void *tag);
635
636 /*
637 * Attempt to add a reference to a dmu buffer that is in an unknown state,
638 * using a pointer that may have been invalidated by eviction processing.
639 * The request will succeed if the passed in dbuf still represents the
640 * same os/object/blkid, is ineligible for eviction, and has at least
641 * one hold by a user other than the syncer.
642 */
643 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
644 uint64_t blkid, const void *tag);
645
646 void dmu_buf_rele(dmu_buf_t *db, const void *tag);
647 uint64_t dmu_buf_refcount(dmu_buf_t *db);
648 uint64_t dmu_buf_user_refcount(dmu_buf_t *db);
649
650 /*
651 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
652 * range of an object. A pointer to an array of dmu_buf_t*'s is
653 * returned (in *dbpp).
654 *
655 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
656 * frees the array. The hold on the array of buffers MUST be released
657 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
658 * individually with dmu_buf_rele.
659 */
660 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
661 uint64_t length, boolean_t read, const void *tag,
662 int *numbufsp, dmu_buf_t ***dbpp, dmu_flags_t flags);
663 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, const void *tag);
664
665 typedef void dmu_buf_evict_func_t(void *user_ptr);
666
667 /*
668 * A DMU buffer user object may be associated with a dbuf for the
669 * duration of its lifetime. This allows the user of a dbuf (client)
670 * to attach private data to a dbuf (e.g. in-core only data such as a
671 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
672 * when that dbuf has been evicted. Clients typically respond to the
673 * eviction notification by freeing their private data, thus ensuring
674 * the same lifetime for both dbuf and private data.
675 *
676 * The mapping from a dmu_buf_user_t to any client private data is the
677 * client's responsibility. All current consumers of the API with private
678 * data embed a dmu_buf_user_t as the first member of the structure for
679 * their private data. This allows conversions between the two types
680 * with a simple cast. Since the DMU buf user API never needs access
681 * to the private data, other strategies can be employed if necessary
682 * or convenient for the client (e.g. using container_of() to do the
683 * conversion for private data that cannot have the dmu_buf_user_t as
684 * its first member).
685 *
686 * Eviction callbacks are executed without the dbuf mutex held or any
687 * other type of mechanism to guarantee that the dbuf is still available.
688 * For this reason, users must assume the dbuf has already been freed
689 * and not reference the dbuf from the callback context.
690 *
691 * Users requesting "immediate eviction" are notified as soon as the dbuf
692 * is only referenced by dirty records (dirties == holds). Otherwise the
693 * notification occurs after eviction processing for the dbuf begins.
694 */
695 typedef struct dmu_buf_user {
696 /*
697 * Asynchronous user eviction callback state.
698 */
699 taskq_ent_t dbu_tqent;
700
701 /* Size of user data, for inclusion in dbuf_cache accounting. */
702 uint64_t dbu_size;
703
704 /*
705 * This instance's eviction function pointers.
706 *
707 * dbu_evict_func_sync is called synchronously and then
708 * dbu_evict_func_async is executed asynchronously on a taskq.
709 */
710 dmu_buf_evict_func_t *dbu_evict_func_sync;
711 dmu_buf_evict_func_t *dbu_evict_func_async;
712 #ifdef ZFS_DEBUG
713 /*
714 * Pointer to user's dbuf pointer. NULL for clients that do
715 * not associate a dbuf with their user data.
716 *
717 * The dbuf pointer is cleared upon eviction so as to catch
718 * use-after-evict bugs in clients.
719 */
720 dmu_buf_t **dbu_clear_on_evict_dbufp;
721 #endif
722 } dmu_buf_user_t;
723
724 /*
725 * Initialize the given dmu_buf_user_t instance with the eviction function
726 * evict_func, to be called when the user is evicted.
727 *
728 * NOTE: This function should only be called once on a given dmu_buf_user_t.
729 * To allow enforcement of this, dbu must already be zeroed on entry.
730 */
731 static inline void
dmu_buf_init_user(dmu_buf_user_t * dbu,dmu_buf_evict_func_t * evict_func_sync,dmu_buf_evict_func_t * evict_func_async,dmu_buf_t ** clear_on_evict_dbufp __maybe_unused)732 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
733 dmu_buf_evict_func_t *evict_func_async,
734 dmu_buf_t **clear_on_evict_dbufp __maybe_unused)
735 {
736 ASSERT0P(dbu->dbu_evict_func_sync);
737 ASSERT0P(dbu->dbu_evict_func_async);
738
739 /* must have at least one evict func */
740 IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
741 dbu->dbu_evict_func_sync = evict_func_sync;
742 dbu->dbu_evict_func_async = evict_func_async;
743 taskq_init_ent(&dbu->dbu_tqent);
744 #ifdef ZFS_DEBUG
745 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
746 #endif
747 }
748
749 /*
750 * Attach user data to a dbuf and mark it for normal (when the dbuf's
751 * data is cleared or its reference count goes to zero) eviction processing.
752 *
753 * Returns NULL on success, or the existing user if another user currently
754 * owns the buffer.
755 */
756 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
757
758 /*
759 * Attach user data to a dbuf and mark it for immediate (its dirty and
760 * reference counts are equal) eviction processing.
761 *
762 * Returns NULL on success, or the existing user if another user currently
763 * owns the buffer.
764 */
765 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
766
767 /*
768 * Replace the current user of a dbuf.
769 *
770 * If given the current user of a dbuf, replaces the dbuf's user with
771 * "new_user" and returns the user data pointer that was replaced.
772 * Otherwise returns the current, and unmodified, dbuf user pointer.
773 */
774 void *dmu_buf_replace_user(dmu_buf_t *db,
775 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
776
777 /*
778 * Remove the specified user data for a DMU buffer.
779 *
780 * Returns the user that was removed on success, or the current user if
781 * another user currently owns the buffer.
782 */
783 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
784
785 /*
786 * User data size accounting. This can be used to artifically inflate the size
787 * of the dbuf during cache accounting, so that dbuf_evict_thread evicts enough
788 * to satisfy memory reclaim requests. It's not used for anything else, and
789 * defaults to 0.
790 */
791 uint64_t dmu_buf_user_size(dmu_buf_t *db);
792 void dmu_buf_add_user_size(dmu_buf_t *db, uint64_t nadd);
793 void dmu_buf_sub_user_size(dmu_buf_t *db, uint64_t nsub);
794
795 /*
796 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
797 */
798 void *dmu_buf_get_user(dmu_buf_t *db);
799
800 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
801
802 /* Block until any in-progress dmu buf user evictions complete. */
803 void dmu_buf_user_evict_wait(void);
804
805 /*
806 * Returns the blkptr associated with this dbuf, or NULL if not set.
807 */
808 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
809
810 /*
811 * Indicate that you are going to modify the buffer's data (db_data).
812 *
813 * The transaction (tx) must be assigned to a txg (ie. you've called
814 * dmu_tx_assign()). The buffer's object must be held in the tx
815 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
816 */
817 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
818 void dmu_buf_will_dirty_flags(dmu_buf_t *db, dmu_tx_t *tx, dmu_flags_t flags);
819 void dmu_buf_will_rewrite(dmu_buf_t *db, dmu_tx_t *tx);
820 boolean_t dmu_buf_is_dirty(dmu_buf_t *db, dmu_tx_t *tx);
821 void dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder,
822 const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx);
823
824 /*
825 * You must create a transaction, then hold the objects which you will
826 * (or might) modify as part of this transaction. Then you must assign
827 * the transaction to a transaction group. Once the transaction has
828 * been assigned, you can modify buffers which belong to held objects as
829 * part of this transaction. You can't modify buffers before the
830 * transaction has been assigned; you can't modify buffers which don't
831 * belong to objects which this transaction holds; you can't hold
832 * objects once the transaction has been assigned. You may hold an
833 * object which you are going to free (with dmu_object_free()), but you
834 * don't have to.
835 *
836 * You can abort the transaction before it has been assigned.
837 *
838 * Note that you may hold buffers (with dmu_buf_hold) at any time,
839 * regardless of transaction state.
840 */
841
842 #define DMU_NEW_OBJECT (-1ULL)
843 #define DMU_OBJECT_END (-1ULL)
844
845 dmu_tx_t *dmu_tx_create(objset_t *os);
846 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
847 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
848 int len);
849 void dmu_tx_hold_append(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
850 void dmu_tx_hold_append_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
851 int len);
852 void dmu_tx_hold_clone_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
853 uint64_t len, uint_t blksz);
854 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
855 uint64_t len);
856 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
857 uint64_t len);
858 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
859 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
860 const char *name);
861 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
862 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
863 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
864 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
865 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
866 void dmu_tx_abort(dmu_tx_t *tx);
867 int dmu_tx_assign(dmu_tx_t *tx, dmu_tx_flag_t flags);
868 void dmu_tx_wait(dmu_tx_t *tx);
869 void dmu_tx_commit(dmu_tx_t *tx);
870 void dmu_tx_mark_netfree(dmu_tx_t *tx);
871
872 /*
873 * To register a commit callback, dmu_tx_callback_register() must be called.
874 *
875 * dcb_data is a pointer to caller private data that is passed on as a
876 * callback parameter. The caller is responsible for properly allocating and
877 * freeing it.
878 *
879 * When registering a callback, the transaction must be already created, but
880 * it cannot be committed or aborted. It can be assigned to a txg or not.
881 *
882 * The callback will be called after the transaction has been safely written
883 * to stable storage and will also be called if the dmu_tx is aborted.
884 * If there is any error which prevents the transaction from being committed to
885 * disk, the callback will be called with a value of error != 0.
886 *
887 * When multiple callbacks are registered to the transaction, the callbacks
888 * will be called in reverse order to let Lustre, the only user of commit
889 * callback currently, take the fast path of its commit callback handling.
890 */
891 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
892
893 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
894 void *dcb_data);
895 void dmu_tx_do_callbacks(list_t *cb_list, int error);
896
897 /*
898 * Free up the data blocks for a defined range of a file. If size is
899 * -1, the range from offset to end-of-file is freed.
900 */
901 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
902 uint64_t size, dmu_tx_t *tx);
903 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
904 uint64_t size);
905 int dmu_free_long_object(objset_t *os, uint64_t object);
906
907 /*
908 * Convenience functions.
909 *
910 * Canfail routines will return 0 on success, or an errno if there is a
911 * nonrecoverable I/O error.
912 */
913 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
914 void *buf, dmu_flags_t flags);
915 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
916 dmu_flags_t flags);
917 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
918 const void *buf, dmu_tx_t *tx, dmu_flags_t flags);
919 int dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
920 const void *buf, dmu_tx_t *tx, dmu_flags_t flags);
921 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
922 dmu_tx_t *tx);
923 #ifdef _KERNEL
924 int dmu_read_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size,
925 dmu_flags_t flags);
926 int dmu_read_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size,
927 dmu_flags_t flags);
928 int dmu_read_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size,
929 dmu_flags_t flags);
930 int dmu_write_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size,
931 dmu_tx_t *tx, dmu_flags_t flags);
932 int dmu_write_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size,
933 dmu_tx_t *tx, dmu_flags_t flags);
934 int dmu_write_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size,
935 dmu_tx_t *tx, dmu_flags_t flags);
936 #endif
937 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
938 void dmu_return_arcbuf(struct arc_buf *buf);
939 int dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset,
940 struct arc_buf *buf, dmu_tx_t *tx, dmu_flags_t flags);
941 int dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset,
942 struct arc_buf *buf, dmu_tx_t *tx, dmu_flags_t flags);
943 #define dmu_assign_arcbuf dmu_assign_arcbuf_by_dbuf
944 extern uint_t zfs_max_recordsize;
945
946 /*
947 * Asynchronously try to read in the data.
948 */
949 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
950 uint64_t len, enum zio_priority pri);
951 void dmu_prefetch_user(objset_t *os, uint64_t object, int64_t level,
952 uint64_t offset, uint64_t len, enum zio_priority pri);
953 void dmu_prefetch_by_dnode(dnode_t *dn, int64_t level, uint64_t offset,
954 uint64_t len, enum zio_priority pri);
955 void dmu_prefetch_dnode(objset_t *os, uint64_t object, enum zio_priority pri);
956 int dmu_prefetch_wait(objset_t *os, uint64_t object, uint64_t offset,
957 uint64_t size);
958 void dmu_prefetch_stream(objset_t *os, uint64_t object, uint64_t offset,
959 uint64_t len, boolean_t start_now);
960 void dmu_prefetch_stream_by_dnode(dnode_t *dn, uint64_t offset,
961 uint64_t len, boolean_t start_now);
962 void dmu_evict_range(objset_t *os, uint64_t object, uint64_t offset,
963 uint64_t len);
964
965 typedef struct dmu_object_info {
966 /* All sizes are in bytes unless otherwise indicated. */
967 uint32_t doi_data_block_size;
968 uint32_t doi_metadata_block_size;
969 dmu_object_type_t doi_type;
970 dmu_object_type_t doi_bonus_type;
971 uint64_t doi_bonus_size;
972 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
973 uint8_t doi_checksum;
974 uint8_t doi_compress;
975 uint8_t doi_nblkptr;
976 uint8_t doi_pad[4];
977 uint64_t doi_dnodesize;
978 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
979 uint64_t doi_max_offset;
980 uint64_t doi_fill_count; /* number of non-empty blocks */
981 } dmu_object_info_t;
982
983 typedef void (*const arc_byteswap_func_t)(void *buf, size_t size);
984
985 typedef struct dmu_object_type_info {
986 dmu_object_byteswap_t ot_byteswap;
987 boolean_t ot_metadata;
988 boolean_t ot_dbuf_metadata_cache;
989 boolean_t ot_encrypt;
990 const char *ot_name;
991 } dmu_object_type_info_t;
992
993 typedef const struct dmu_object_byteswap_info {
994 arc_byteswap_func_t ob_func;
995 const char *ob_name;
996 } dmu_object_byteswap_info_t;
997
998 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
999 extern dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
1000
1001 /*
1002 * Get information on a DMU object.
1003 *
1004 * Return 0 on success or ENOENT if object is not allocated.
1005 *
1006 * If doi is NULL, just indicates whether the object exists.
1007 */
1008 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
1009 void __dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
1010 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
1011 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
1012 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
1013 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
1014 /*
1015 * Like dmu_object_info_from_db, but faster still when you only care about
1016 * the size.
1017 */
1018 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
1019 u_longlong_t *nblk512);
1020
1021 void dmu_object_dnsize_from_db(dmu_buf_t *db, int *dnsize);
1022
1023 typedef enum {
1024 DDS_FLAG_ENCRYPTED = (1<<0),
1025 DDS_FLAG_HAS_ENCRYPTED = (1<<7),
1026 } dmu_objset_flag_t;
1027
1028 typedef struct dmu_objset_stats {
1029 uint64_t dds_num_clones; /* number of clones of this */
1030 uint64_t dds_creation_txg;
1031 uint64_t dds_guid;
1032 dmu_objset_type_t dds_type;
1033 uint8_t dds_is_snapshot;
1034 uint8_t dds_inconsistent;
1035 uint8_t dds_redacted;
1036 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
1037 uint8_t dds_flags; /* dmu_objset_flag_t */
1038 } dmu_objset_stats_t;
1039
1040 /*
1041 * Get stats on a dataset.
1042 */
1043 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
1044
1045 /*
1046 * Add entries to the nvlist for all the objset's properties. See
1047 * zfs_prop_table[] and zfs(1m) for details on the properties.
1048 */
1049 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
1050
1051 /*
1052 * Get the space usage statistics for statvfs().
1053 *
1054 * refdbytes is the amount of space "referenced" by this objset.
1055 * availbytes is the amount of space available to this objset, taking
1056 * into account quotas & reservations, assuming that no other objsets
1057 * use the space first. These values correspond to the 'referenced' and
1058 * 'available' properties, described in the zfs(1m) manpage.
1059 *
1060 * usedobjs and availobjs are the number of objects currently allocated,
1061 * and available.
1062 */
1063 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1064 uint64_t *usedobjsp, uint64_t *availobjsp);
1065
1066 /*
1067 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
1068 * (Contrast with the ds_guid which is a 64-bit ID that will never
1069 * change, so there is a small probability that it will collide.)
1070 */
1071 uint64_t dmu_objset_fsid_guid(objset_t *os);
1072
1073 /*
1074 * Get the [cm]time for an objset's snapshot dir
1075 */
1076 inode_timespec_t dmu_objset_snap_cmtime(objset_t *os);
1077
1078 int dmu_objset_is_snapshot(objset_t *os);
1079
1080 extern struct spa *dmu_objset_spa(objset_t *os);
1081 extern struct zilog *dmu_objset_zil(objset_t *os);
1082 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
1083 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
1084 extern void dmu_objset_name(objset_t *os, char *buf);
1085 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
1086 extern uint64_t dmu_objset_id(objset_t *os);
1087 extern uint64_t dmu_objset_dnodesize(objset_t *os);
1088 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
1089 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
1090 extern int dmu_objset_blksize(objset_t *os);
1091 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1092 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
1093 extern int dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *val);
1094 extern int dmu_snapshot_realname(objset_t *os, const char *name, char *real,
1095 int maxlen, boolean_t *conflict);
1096 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
1097 uint64_t *idp, uint64_t *offp);
1098
1099 typedef struct zfs_file_info {
1100 uint64_t zfi_user;
1101 uint64_t zfi_group;
1102 uint64_t zfi_project;
1103 uint64_t zfi_generation;
1104 } zfs_file_info_t;
1105
1106 typedef int file_info_cb_t(dmu_object_type_t bonustype, const void *data,
1107 struct zfs_file_info *zoi);
1108 extern void dmu_objset_register_type(dmu_objset_type_t ost,
1109 file_info_cb_t *cb);
1110 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
1111 extern void *dmu_objset_get_user(objset_t *os);
1112
1113 /*
1114 * Return the txg number for the given assigned transaction.
1115 */
1116 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
1117
1118 /*
1119 * Synchronous write.
1120 * If a parent zio is provided this function initiates a write on the
1121 * provided buffer as a child of the parent zio.
1122 * In the absence of a parent zio, the write is completed synchronously.
1123 * At write completion, blk is filled with the bp of the written block.
1124 * Note that while the data covered by this function will be on stable
1125 * storage when the write completes this new data does not become a
1126 * permanent part of the file until the associated transaction commits.
1127 */
1128
1129 /*
1130 * {zfs,zvol,ztest}_get_done() args
1131 */
1132 typedef struct zgd {
1133 struct lwb *zgd_lwb;
1134 struct blkptr *zgd_bp;
1135 dmu_buf_t *zgd_db;
1136 struct zfs_locked_range *zgd_lr;
1137 void *zgd_private;
1138 } zgd_t;
1139
1140 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
1141 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
1142
1143 /*
1144 * Find the next hole or data block in file starting at *off
1145 * Return found offset in *off. Return ESRCH for end of file.
1146 */
1147 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
1148 uint64_t *off);
1149
1150 int dmu_read_l0_bps(objset_t *os, uint64_t object, uint64_t offset,
1151 uint64_t length, struct blkptr *bps, size_t *nbpsp);
1152 int dmu_brt_clone(objset_t *os, uint64_t object, uint64_t offset,
1153 uint64_t length, dmu_tx_t *tx, const struct blkptr *bps, size_t nbps);
1154
1155 /*
1156 * Initial setup and final teardown.
1157 */
1158 extern void dmu_init(void);
1159 extern void dmu_fini(void);
1160
1161 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
1162 uint64_t object, uint64_t offset, int len);
1163 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
1164 dmu_traverse_cb_t cb, void *arg);
1165
1166 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
1167 zfs_file_t *fp, offset_t *offp);
1168
1169 /* CRC64 table */
1170 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
1171 extern uint64_t zfs_crc64_table[256];
1172
1173 extern uint_t dmu_prefetch_max;
1174
1175 #ifdef __cplusplus
1176 }
1177 #endif
1178
1179 #endif /* _SYS_DMU_H */
1180