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