xref: /freebsd/sys/contrib/openzfs/include/sys/dmu.h (revision df58e8b1506f241670be86a560fb6e8432043aee)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
25  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
26  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
27  * Copyright 2014 HybridCluster. All rights reserved.
28  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
29  * Copyright 2013 Saso Kiselkov. All rights reserved.
30  * Copyright (c) 2017, Intel Corporation.
31  * Copyright (c) 2022 Hewlett Packard Enterprise Development LP.
32  * Copyright (c) 2025, Klara, Inc.
33  */
34 
35 /* Portions Copyright 2010 Robert Milkowski */
36 
37 #ifndef	_SYS_DMU_H
38 #define	_SYS_DMU_H
39 
40 /*
41  * This file describes the interface that the DMU provides for its
42  * consumers.
43  *
44  * The DMU also interacts with the SPA.  That interface is described in
45  * dmu_spa.h.
46  */
47 
48 #include <sys/zfs_context.h>
49 #include <sys/inttypes.h>
50 #include <sys/cred.h>
51 #include <sys/fs/zfs.h>
52 #include <sys/zio_compress.h>
53 #include <sys/uio.h>
54 #include <sys/zfs_file.h>
55 
56 #ifdef	__cplusplus
57 extern "C" {
58 #endif
59 
60 struct page;
61 struct vnode;
62 struct spa;
63 struct zilog;
64 struct zio;
65 struct blkptr;
66 struct zap_cursor;
67 struct dsl_dataset;
68 struct dsl_pool;
69 struct dnode;
70 struct drr_begin;
71 struct drr_end;
72 struct zbookmark_phys;
73 struct spa;
74 struct nvlist;
75 struct arc_buf;
76 struct zio_prop;
77 struct sa_handle;
78 struct dsl_crypto_params;
79 struct locked_range;
80 
81 typedef struct objset objset_t;
82 typedef struct dmu_tx dmu_tx_t;
83 typedef struct dsl_dir dsl_dir_t;
84 typedef struct dnode dnode_t;
85 
86 typedef enum dmu_object_byteswap {
87 	DMU_BSWAP_UINT8,
88 	DMU_BSWAP_UINT16,
89 	DMU_BSWAP_UINT32,
90 	DMU_BSWAP_UINT64,
91 	DMU_BSWAP_ZAP,
92 	DMU_BSWAP_DNODE,
93 	DMU_BSWAP_OBJSET,
94 	DMU_BSWAP_ZNODE,
95 	DMU_BSWAP_OLDACL,
96 	DMU_BSWAP_ACL,
97 	/*
98 	 * Allocating a new byteswap type number makes the on-disk format
99 	 * incompatible with any other format that uses the same number.
100 	 *
101 	 * Data can usually be structured to work with one of the
102 	 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
103 	 */
104 	DMU_BSWAP_NUMFUNCS
105 } dmu_object_byteswap_t;
106 
107 #define	DMU_OT_NEWTYPE 0x80
108 #define	DMU_OT_METADATA 0x40
109 #define	DMU_OT_ENCRYPTED 0x20
110 #define	DMU_OT_BYTESWAP_MASK 0x1f
111 
112 /*
113  * Defines a uint8_t object type. Object types specify if the data
114  * in the object is metadata (boolean) and how to byteswap the data
115  * (dmu_object_byteswap_t). All of the types created by this method
116  * are cached in the dbuf metadata cache.
117  */
118 #define	DMU_OT(byteswap, metadata, encrypted) \
119 	(DMU_OT_NEWTYPE | \
120 	((metadata) ? DMU_OT_METADATA : 0) | \
121 	((encrypted) ? DMU_OT_ENCRYPTED : 0) | \
122 	((byteswap) & DMU_OT_BYTESWAP_MASK))
123 
124 #define	DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
125 	((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
126 	(ot) < DMU_OT_NUMTYPES)
127 
128 #define	DMU_OT_IS_METADATA_CACHED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
129 	B_TRUE : dmu_ot[(ot)].ot_dbuf_metadata_cache)
130 
131 /*
132  * MDB doesn't have dmu_ot; it defines these macros itself.
133  */
134 #ifndef ZFS_MDB
135 #define	DMU_OT_IS_METADATA_IMPL(ot) (dmu_ot[ot].ot_metadata)
136 #define	DMU_OT_IS_ENCRYPTED_IMPL(ot) (dmu_ot[ot].ot_encrypt)
137 #define	DMU_OT_BYTESWAP_IMPL(ot) (dmu_ot[ot].ot_byteswap)
138 #endif
139 
140 #define	DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
141 	(((ot) & DMU_OT_METADATA) != 0) : \
142 	DMU_OT_IS_METADATA_IMPL(ot))
143 
144 #define	DMU_OT_IS_DDT(ot) \
145 	((ot) == DMU_OT_DDT_ZAP)
146 
147 #define	DMU_OT_IS_CRITICAL(ot, level) \
148 	(DMU_OT_IS_METADATA(ot) && \
149 	((ot) != DMU_OT_DNODE || (level) > 0) && \
150 	(ot) != DMU_OT_DIRECTORY_CONTENTS && \
151 	(ot) != DMU_OT_SA)
152 
153 /* Note: ztest uses DMU_OT_UINT64_OTHER as a proxy for file blocks */
154 #define	DMU_OT_IS_FILE(ot) \
155 	((ot) == DMU_OT_PLAIN_FILE_CONTENTS || (ot) == DMU_OT_UINT64_OTHER)
156 
157 #define	DMU_OT_IS_ENCRYPTED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
158 	(((ot) & DMU_OT_ENCRYPTED) != 0) : \
159 	DMU_OT_IS_ENCRYPTED_IMPL(ot))
160 
161 /*
162  * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
163  * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
164  * is repurposed for embedded BPs.
165  */
166 #define	DMU_OT_HAS_FILL(ot) \
167 	((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
168 
169 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
170 	((ot) & DMU_OT_BYTESWAP_MASK) : \
171 	DMU_OT_BYTESWAP_IMPL(ot))
172 
173 typedef enum dmu_object_type {
174 	DMU_OT_NONE,
175 	/* general: */
176 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
177 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
178 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
179 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
180 	DMU_OT_BPOBJ,			/* UINT64 */
181 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
182 	/* spa: */
183 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
184 	DMU_OT_SPACE_MAP,		/* UINT64 */
185 	/* zil: */
186 	DMU_OT_INTENT_LOG,		/* UINT64 */
187 	/* dmu: */
188 	DMU_OT_DNODE,			/* DNODE */
189 	DMU_OT_OBJSET,			/* OBJSET */
190 	/* dsl: */
191 	DMU_OT_DSL_DIR,			/* UINT64 */
192 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
193 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
194 	DMU_OT_DSL_PROPS,		/* ZAP */
195 	DMU_OT_DSL_DATASET,		/* UINT64 */
196 	/* zpl: */
197 	DMU_OT_ZNODE,			/* ZNODE */
198 	DMU_OT_OLDACL,			/* Old ACL */
199 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
200 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
201 	DMU_OT_MASTER_NODE,		/* ZAP */
202 	DMU_OT_UNLINKED_SET,		/* ZAP */
203 	/* zvol: */
204 	DMU_OT_ZVOL,			/* UINT8 */
205 	DMU_OT_ZVOL_PROP,		/* ZAP */
206 	/* other; for testing only! */
207 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
208 	DMU_OT_UINT64_OTHER,		/* UINT64 */
209 	DMU_OT_ZAP_OTHER,		/* ZAP */
210 	/* new object types: */
211 	DMU_OT_ERROR_LOG,		/* ZAP */
212 	DMU_OT_SPA_HISTORY,		/* UINT8 */
213 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
214 	DMU_OT_POOL_PROPS,		/* ZAP */
215 	DMU_OT_DSL_PERMS,		/* ZAP */
216 	DMU_OT_ACL,			/* ACL */
217 	DMU_OT_SYSACL,			/* SYSACL */
218 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
219 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
220 	DMU_OT_NEXT_CLONES,		/* ZAP */
221 	DMU_OT_SCAN_QUEUE,		/* ZAP */
222 	DMU_OT_USERGROUP_USED,		/* ZAP */
223 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
224 	DMU_OT_USERREFS,		/* ZAP */
225 	DMU_OT_DDT_ZAP,			/* ZAP */
226 	DMU_OT_DDT_STATS,		/* ZAP */
227 	DMU_OT_SA,			/* System attr */
228 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
229 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
230 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
231 	DMU_OT_SCAN_XLATE,		/* ZAP */
232 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
233 	DMU_OT_DEADLIST,		/* ZAP */
234 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
235 	DMU_OT_DSL_CLONES,		/* ZAP */
236 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
237 	/*
238 	 * Do not allocate new object types here. Doing so makes the on-disk
239 	 * format incompatible with any other format that uses the same object
240 	 * type number.
241 	 *
242 	 * When creating an object which does not have one of the above types
243 	 * use the DMU_OTN_* type with the correct byteswap and metadata
244 	 * values.
245 	 *
246 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
247 	 * use the DMU_OT_IS_METADATA() and DMU_OT_BYTESWAP() macros instead
248 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
249 	 * and DMU_OTN_* types).
250 	 */
251 	DMU_OT_NUMTYPES,
252 
253 	/*
254 	 * Names for valid types declared with DMU_OT().
255 	 */
256 	DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_FALSE),
257 	DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_FALSE),
258 	DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_FALSE),
259 	DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_FALSE),
260 	DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_FALSE),
261 	DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_FALSE),
262 	DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_FALSE),
263 	DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_FALSE),
264 	DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_FALSE),
265 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_FALSE),
266 
267 	DMU_OTN_UINT8_ENC_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_TRUE),
268 	DMU_OTN_UINT8_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_TRUE),
269 	DMU_OTN_UINT16_ENC_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_TRUE),
270 	DMU_OTN_UINT16_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_TRUE),
271 	DMU_OTN_UINT32_ENC_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_TRUE),
272 	DMU_OTN_UINT32_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_TRUE),
273 	DMU_OTN_UINT64_ENC_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_TRUE),
274 	DMU_OTN_UINT64_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_TRUE),
275 	DMU_OTN_ZAP_ENC_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_TRUE),
276 	DMU_OTN_ZAP_ENC_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_TRUE),
277 } dmu_object_type_t;
278 
279 /*
280  * These flags are for the dmu_tx_assign() function and describe what to do if
281  * the transaction is full. See the comment above dmu_tx_assign() for more
282  * details on the meaning of these flags.
283  */
284 typedef enum {
285 	/*
286 	 * If the tx cannot be assigned to a transaction for any reason, do
287 	 * not block but return immediately.
288 	 */
289 	DMU_TX_NOWAIT		= 0,
290 
291 	/*
292 	 * Assign the tx to the open transaction. If the open transaction is
293 	 * full, or the write throttle is active, block until the next
294 	 * transaction and try again. If the pool suspends while waiting
295 	 * and failmode=continue, return an error.
296 	 */
297 	DMU_TX_WAIT		= (1 << 0),
298 
299 	/* If the write throttle would prevent the assignment, ignore it. */
300 	DMU_TX_NOTHROTTLE	= (1 << 1),
301 
302 	/*
303 	 * With DMU_TX_WAIT, always block if the pool suspends during
304 	 * assignment, regardless of the value of the failmode= property.
305 	 */
306 	DMU_TX_SUSPEND		= (1 << 2),
307 } dmu_tx_flag_t;
308 
309 void byteswap_uint64_array(void *buf, size_t size);
310 void byteswap_uint32_array(void *buf, size_t size);
311 void byteswap_uint16_array(void *buf, size_t size);
312 void byteswap_uint8_array(void *buf, size_t size);
313 void zap_byteswap(void *buf, size_t size);
314 void zfs_oldacl_byteswap(void *buf, size_t size);
315 void zfs_acl_byteswap(void *buf, size_t size);
316 void zfs_znode_byteswap(void *buf, size_t size);
317 
318 #define	DS_FIND_SNAPSHOTS	(1<<0)
319 #define	DS_FIND_CHILDREN	(1<<1)
320 #define	DS_FIND_SERIALIZE	(1<<2)
321 
322 /*
323  * The maximum number of bytes that can be accessed as part of one
324  * operation, including metadata.
325  */
326 #define	DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
327 #define	DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
328 
329 #define	DMU_USERUSED_OBJECT	(-1ULL)
330 #define	DMU_GROUPUSED_OBJECT	(-2ULL)
331 #define	DMU_PROJECTUSED_OBJECT	(-3ULL)
332 
333 /*
334  * Zap prefix for object accounting in DMU_{USER,GROUP,PROJECT}USED_OBJECT.
335  */
336 #define	DMU_OBJACCT_PREFIX	"obj-"
337 #define	DMU_OBJACCT_PREFIX_LEN	4
338 
339 /*
340  * artificial blkids for bonus buffer and spill blocks
341  */
342 #define	DMU_BONUS_BLKID		(-1ULL)
343 #define	DMU_SPILL_BLKID		(-2ULL)
344 
345 /*
346  * Public routines to create, destroy, open, and close objsets.
347  */
348 typedef void dmu_objset_create_sync_func_t(objset_t *os, void *arg,
349     cred_t *cr, dmu_tx_t *tx);
350 
351 int dmu_objset_hold(const char *name, const void *tag, objset_t **osp);
352 int dmu_objset_own(const char *name, dmu_objset_type_t type,
353     boolean_t readonly, boolean_t key_required, const void *tag,
354     objset_t **osp);
355 void dmu_objset_rele(objset_t *os, const void *tag);
356 void dmu_objset_disown(objset_t *os, boolean_t key_required, const void *tag);
357 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
358 
359 void dmu_objset_evict_dbufs(objset_t *os);
360 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
361     struct dsl_crypto_params *dcp, dmu_objset_create_sync_func_t func,
362     void *arg);
363 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
364     struct nvlist *errlist);
365 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
366 int dmu_objset_find(const char *name, int func(const char *, void *), void *arg,
367     int flags);
368 void dmu_objset_byteswap(void *buf, size_t size);
369 int dsl_dataset_rename_snapshot(const char *fsname,
370     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
371 
372 typedef struct dmu_buf {
373 	uint64_t db_object;		/* object that this buffer is part of */
374 	uint64_t db_offset;		/* byte offset in this object */
375 	uint64_t db_size;		/* size of buffer in bytes */
376 	void *db_data;			/* data in buffer */
377 } dmu_buf_t;
378 
379 /*
380  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
381  */
382 #define	DMU_POOL_DIRECTORY_OBJECT	1
383 #define	DMU_POOL_CONFIG			"config"
384 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
385 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
386 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
387 #define	DMU_POOL_FEATURE_ENABLED_TXG	"feature_enabled_txg"
388 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
389 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
390 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
391 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
392 #define	DMU_POOL_SPARES			"spares"
393 #define	DMU_POOL_DEFLATE		"deflate"
394 #define	DMU_POOL_HISTORY		"history"
395 #define	DMU_POOL_PROPS			"pool_props"
396 #define	DMU_POOL_L2CACHE		"l2cache"
397 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
398 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
399 #define	DMU_POOL_DDT_LOG		"DDT-log-%s-%u"
400 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
401 #define	DMU_POOL_DDT_DIR		"DDT-%s"
402 #define	DMU_POOL_CREATION_VERSION	"creation_version"
403 #define	DMU_POOL_SCAN			"scan"
404 #define	DMU_POOL_ERRORSCRUB		"error_scrub"
405 #define	DMU_POOL_LAST_SCRUBBED_TXG	"last_scrubbed_txg"
406 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
407 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
408 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
409 #define	DMU_POOL_CHECKSUM_SALT		"org.illumos:checksum_salt"
410 #define	DMU_POOL_VDEV_ZAP_MAP		"com.delphix:vdev_zap_map"
411 #define	DMU_POOL_REMOVING		"com.delphix:removing"
412 #define	DMU_POOL_OBSOLETE_BPOBJ		"com.delphix:obsolete_bpobj"
413 #define	DMU_POOL_CONDENSING_INDIRECT	"com.delphix:condensing_indirect"
414 #define	DMU_POOL_ZPOOL_CHECKPOINT	"com.delphix:zpool_checkpoint"
415 #define	DMU_POOL_LOG_SPACEMAP_ZAP	"com.delphix:log_spacemap_zap"
416 #define	DMU_POOL_DELETED_CLONES		"com.delphix:deleted_clones"
417 #define	DMU_POOL_TXG_LOG_TIME_MINUTES	"com.klaraystems:txg_log_time:minutes"
418 #define	DMU_POOL_TXG_LOG_TIME_DAYS	"com.klaraystems:txg_log_time:days"
419 #define	DMU_POOL_TXG_LOG_TIME_MONTHS	"com.klaraystems:txg_log_time:months"
420 
421 /*
422  * Allocate an object from this objset.  The range of object numbers
423  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
424  *
425  * The transaction must be assigned to a txg.  The newly allocated
426  * object will be "held" in the transaction (ie. you can modify the
427  * newly allocated object in this transaction).
428  *
429  * dmu_object_alloc() chooses an object and returns it in *objectp.
430  *
431  * dmu_object_claim() allocates a specific object number.  If that
432  * number is already allocated, it fails and returns EEXIST.
433  *
434  * Return 0 on success, or ENOSPC or EEXIST as specified above.
435  */
436 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
437     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
438 uint64_t dmu_object_alloc_ibs(objset_t *os, dmu_object_type_t ot, int blocksize,
439     int indirect_blockshift,
440     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
441 uint64_t dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot,
442     int blocksize, dmu_object_type_t bonus_type, int bonus_len,
443     int dnodesize, dmu_tx_t *tx);
444 uint64_t dmu_object_alloc_hold(objset_t *os, dmu_object_type_t ot,
445     int blocksize, int indirect_blockshift, dmu_object_type_t bonustype,
446     int bonuslen, int dnodesize, dnode_t **allocated_dnode, const void *tag,
447     dmu_tx_t *tx);
448 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
449     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
450 int dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
451     int blocksize, dmu_object_type_t bonus_type, int bonus_len,
452     int dnodesize, dmu_tx_t *tx);
453 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
454     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
455 int dmu_object_reclaim_dnsize(objset_t *os, uint64_t object,
456     dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype,
457     int bonuslen, int dnodesize, boolean_t keep_spill, dmu_tx_t *tx);
458 int dmu_object_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx);
459 
460 /*
461  * Free an object from this objset.
462  *
463  * The object's data will be freed as well (ie. you don't need to call
464  * dmu_free(object, 0, -1, tx)).
465  *
466  * The object need not be held in the transaction.
467  *
468  * If there are any holds on this object's buffers (via dmu_buf_hold()),
469  * or tx holds on the object (via dmu_tx_hold_object()), you can not
470  * free it; it fails and returns EBUSY.
471  *
472  * If the object is not allocated, it fails and returns ENOENT.
473  *
474  * Return 0 on success, or EBUSY or ENOENT as specified above.
475  */
476 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
477 
478 /*
479  * Find the next allocated or free object.
480  *
481  * The objectp parameter is in-out.  It will be updated to be the next
482  * object which is allocated.  Ignore objects which have not been
483  * modified since txg.
484  *
485  * XXX Can only be called on a objset with no dirty data.
486  *
487  * Returns 0 on success, or ENOENT if there are no more objects.
488  */
489 int dmu_object_next(objset_t *os, uint64_t *objectp,
490     boolean_t hole, uint64_t txg);
491 
492 /*
493  * Set the number of levels on a dnode. nlevels must be greater than the
494  * current number of levels or an EINVAL will be returned.
495  */
496 int dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels,
497     dmu_tx_t *tx);
498 
499 /*
500  * Set the data blocksize for an object.
501  *
502  * The object cannot have any blocks allocated beyond the first.  If
503  * the first block is allocated already, the new size must be greater
504  * than the current block size.  If these conditions are not met,
505  * ENOTSUP will be returned.
506  *
507  * Returns 0 on success, or EBUSY if there are any holds on the object
508  * contents, or ENOTSUP as described above.
509  */
510 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
511     int ibs, dmu_tx_t *tx);
512 
513 /*
514  * Manually set the maxblkid on a dnode. This will adjust nlevels accordingly
515  * to accommodate the change. When calling this function, the caller must
516  * ensure that the object's nlevels can sufficiently support the new maxblkid.
517  */
518 int dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid,
519     dmu_tx_t *tx);
520 
521 /*
522  * Set the checksum property on a dnode.  The new checksum algorithm will
523  * apply to all newly written blocks; existing blocks will not be affected.
524  */
525 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
526     dmu_tx_t *tx);
527 
528 /*
529  * Set the compress property on a dnode.  The new compression algorithm will
530  * apply to all newly written blocks; existing blocks will not be affected.
531  */
532 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
533     dmu_tx_t *tx);
534 
535 /*
536  * Get an estimated cache size for an object. Caller must expect races.
537  */
538 int dmu_object_cached_size(objset_t *os, uint64_t object,
539     uint64_t *l1sz, uint64_t *l2sz);
540 
541 void dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
542     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
543     int compressed_size, int byteorder, dmu_tx_t *tx);
544 void dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
545     dmu_tx_t *tx);
546 
547 /*
548  * Decide how to write a block: checksum, compression, number of copies, etc.
549  */
550 #define	WP_NOFILL	0x1
551 #define	WP_DMU_SYNC	0x2
552 #define	WP_SPILL	0x4
553 #define	WP_DIRECT_WR	0x8
554 
555 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
556     struct zio_prop *zp);
557 
558 /*
559  * DB_RF_* are to be used for dbuf_read() or in limited other cases.
560  */
561 typedef enum dmu_flags {
562 	DB_RF_MUST_SUCCEED	= 0,	  /* Suspend on I/O errors. */
563 	DB_RF_CANFAIL		= 1 << 0, /* Return on I/O errors. */
564 	DB_RF_HAVESTRUCT	= 1 << 1, /* dn_struct_rwlock is locked. */
565 	DB_RF_NEVERWAIT		= 1 << 2,
566 	DMU_READ_PREFETCH	= 0,	  /* Try speculative prefetch. */
567 	DMU_READ_NO_PREFETCH	= 1 << 3, /* Don't prefetch speculatively. */
568 	DB_RF_NOPREFETCH	= DMU_READ_NO_PREFETCH,
569 	DMU_READ_NO_DECRYPT	= 1 << 4, /* Don't decrypt. */
570 	DB_RF_NO_DECRYPT	= DMU_READ_NO_DECRYPT,
571 	DMU_DIRECTIO		= 1 << 5, /* Bypass ARC. */
572 	DMU_UNCACHEDIO		= 1 << 6, /* Reduce caching. */
573 	DMU_PARTIAL_FIRST	= 1 << 7, /* First partial access. */
574 	DMU_PARTIAL_MORE	= 1 << 8, /* Following partial access. */
575 	DMU_KEEP_CACHING	= 1 << 9, /* Don't affect caching. */
576 } dmu_flags_t;
577 
578 /*
579  * The bonus data is accessed more or less like a regular buffer.
580  * You must dmu_bonus_hold() to get the buffer, which will give you a
581  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
582  * data.  As with any normal buffer, you must call dmu_buf_will_dirty()
583  * before modifying it, and the
584  * object must be held in an assigned transaction before calling
585  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
586  * buffer as well.  You must release what you hold with dmu_buf_rele().
587  *
588  * Returns ENOENT, EIO, or 0.
589  */
590 int dmu_bonus_hold(objset_t *os, uint64_t object, const void *tag,
591     dmu_buf_t **dbp);
592 int dmu_bonus_hold_by_dnode(dnode_t *dn, const void *tag, dmu_buf_t **dbp,
593     dmu_flags_t flags);
594 int dmu_bonus_max(void);
595 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
596 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
597 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
598 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
599 
600 /*
601  * Special spill buffer support used by "SA" framework
602  */
603 
604 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, dmu_flags_t flags,
605     const void *tag, dmu_buf_t **dbp);
606 int dmu_spill_hold_by_dnode(dnode_t *dn, dmu_flags_t flags,
607     const void *tag, dmu_buf_t **dbp);
608 int dmu_spill_hold_existing(dmu_buf_t *bonus, const void *tag, dmu_buf_t **dbp);
609 
610 /*
611  * Obtain the DMU buffer from the specified object which contains the
612  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
613  * that it will remain in memory.  You must release the hold with
614  * dmu_buf_rele().  You must not access the dmu_buf_t after releasing
615  * what you hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
616  *
617  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
618  * on the returned buffer before reading or writing the buffer's
619  * db_data.  The comments for those routines describe what particular
620  * operations are valid after calling them.
621  *
622  * The object number must be a valid, allocated object number.
623  */
624 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
625     const void *tag, dmu_buf_t **, dmu_flags_t flags);
626 int dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
627     uint64_t length, int read, const void *tag, int *numbufsp,
628     dmu_buf_t ***dbpp);
629 int dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
630     const void *tag, dmu_buf_t **dbp);
631 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
632     const void *tag, dmu_buf_t **dbp, dmu_flags_t flags);
633 int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset,
634     uint64_t length, boolean_t read, const void *tag, int *numbufsp,
635     dmu_buf_t ***dbpp, dmu_flags_t flags);
636 int dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, const void *tag,
637     dmu_buf_t **dbp);
638 
639 /*
640  * Add a reference to a dmu buffer that has already been held via
641  * dmu_buf_hold() in the current context.
642  */
643 void dmu_buf_add_ref(dmu_buf_t *db, const void *tag);
644 
645 /*
646  * Attempt to add a reference to a dmu buffer that is in an unknown state,
647  * using a pointer that may have been invalidated by eviction processing.
648  * The request will succeed if the passed in dbuf still represents the
649  * same os/object/blkid, is ineligible for eviction, and has at least
650  * one hold by a user other than the syncer.
651  */
652 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
653     uint64_t blkid, const void *tag);
654 
655 void dmu_buf_rele(dmu_buf_t *db, const void *tag);
656 uint64_t dmu_buf_refcount(dmu_buf_t *db);
657 uint64_t dmu_buf_user_refcount(dmu_buf_t *db);
658 
659 /*
660  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
661  * range of an object.  A pointer to an array of dmu_buf_t*'s is
662  * returned (in *dbpp).
663  *
664  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
665  * frees the array.  The hold on the array of buffers MUST be released
666  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
667  * individually with dmu_buf_rele.
668  */
669 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
670     uint64_t length, boolean_t read, const void *tag,
671     int *numbufsp, dmu_buf_t ***dbpp);
672 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, const void *tag);
673 
674 typedef void dmu_buf_evict_func_t(void *user_ptr);
675 
676 /*
677  * A DMU buffer user object may be associated with a dbuf for the
678  * duration of its lifetime.  This allows the user of a dbuf (client)
679  * to attach private data to a dbuf (e.g. in-core only data such as a
680  * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
681  * when that dbuf has been evicted.  Clients typically respond to the
682  * eviction notification by freeing their private data, thus ensuring
683  * the same lifetime for both dbuf and private data.
684  *
685  * The mapping from a dmu_buf_user_t to any client private data is the
686  * client's responsibility.  All current consumers of the API with private
687  * data embed a dmu_buf_user_t as the first member of the structure for
688  * their private data.  This allows conversions between the two types
689  * with a simple cast.  Since the DMU buf user API never needs access
690  * to the private data, other strategies can be employed if necessary
691  * or convenient for the client (e.g. using container_of() to do the
692  * conversion for private data that cannot have the dmu_buf_user_t as
693  * its first member).
694  *
695  * Eviction callbacks are executed without the dbuf mutex held or any
696  * other type of mechanism to guarantee that the dbuf is still available.
697  * For this reason, users must assume the dbuf has already been freed
698  * and not reference the dbuf from the callback context.
699  *
700  * Users requesting "immediate eviction" are notified as soon as the dbuf
701  * is only referenced by dirty records (dirties == holds).  Otherwise the
702  * notification occurs after eviction processing for the dbuf begins.
703  */
704 typedef struct dmu_buf_user {
705 	/*
706 	 * Asynchronous user eviction callback state.
707 	 */
708 	taskq_ent_t	dbu_tqent;
709 
710 	/* Size of user data, for inclusion in dbuf_cache accounting. */
711 	uint64_t	dbu_size;
712 
713 	/*
714 	 * This instance's eviction function pointers.
715 	 *
716 	 * dbu_evict_func_sync is called synchronously and then
717 	 * dbu_evict_func_async is executed asynchronously on a taskq.
718 	 */
719 	dmu_buf_evict_func_t *dbu_evict_func_sync;
720 	dmu_buf_evict_func_t *dbu_evict_func_async;
721 #ifdef ZFS_DEBUG
722 	/*
723 	 * Pointer to user's dbuf pointer.  NULL for clients that do
724 	 * not associate a dbuf with their user data.
725 	 *
726 	 * The dbuf pointer is cleared upon eviction so as to catch
727 	 * use-after-evict bugs in clients.
728 	 */
729 	dmu_buf_t **dbu_clear_on_evict_dbufp;
730 #endif
731 } dmu_buf_user_t;
732 
733 /*
734  * Initialize the given dmu_buf_user_t instance with the eviction function
735  * evict_func, to be called when the user is evicted.
736  *
737  * NOTE: This function should only be called once on a given dmu_buf_user_t.
738  *       To allow enforcement of this, dbu must already be zeroed on entry.
739  */
740 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)741 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
742     dmu_buf_evict_func_t *evict_func_async,
743     dmu_buf_t **clear_on_evict_dbufp __maybe_unused)
744 {
745 	ASSERT(dbu->dbu_evict_func_sync == NULL);
746 	ASSERT(dbu->dbu_evict_func_async == NULL);
747 
748 	/* must have at least one evict func */
749 	IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
750 	dbu->dbu_evict_func_sync = evict_func_sync;
751 	dbu->dbu_evict_func_async = evict_func_async;
752 	taskq_init_ent(&dbu->dbu_tqent);
753 #ifdef ZFS_DEBUG
754 	dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
755 #endif
756 }
757 
758 /*
759  * Attach user data to a dbuf and mark it for normal (when the dbuf's
760  * data is cleared or its reference count goes to zero) 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(dmu_buf_t *db, dmu_buf_user_t *user);
766 
767 /*
768  * Attach user data to a dbuf and mark it for immediate (its dirty and
769  * reference counts are equal) eviction processing.
770  *
771  * Returns NULL on success, or the existing user if another user currently
772  * owns the buffer.
773  */
774 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
775 
776 /*
777  * Replace the current user of a dbuf.
778  *
779  * If given the current user of a dbuf, replaces the dbuf's user with
780  * "new_user" and returns the user data pointer that was replaced.
781  * Otherwise returns the current, and unmodified, dbuf user pointer.
782  */
783 void *dmu_buf_replace_user(dmu_buf_t *db,
784     dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
785 
786 /*
787  * Remove the specified user data for a DMU buffer.
788  *
789  * Returns the user that was removed on success, or the current user if
790  * another user currently owns the buffer.
791  */
792 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
793 
794 /*
795  * User data size accounting. This can be used to artifically inflate the size
796  * of the dbuf during cache accounting, so that dbuf_evict_thread evicts enough
797  * to satisfy memory reclaim requests. It's not used for anything else, and
798  * defaults to 0.
799  */
800 uint64_t dmu_buf_user_size(dmu_buf_t *db);
801 void dmu_buf_add_user_size(dmu_buf_t *db, uint64_t nadd);
802 void dmu_buf_sub_user_size(dmu_buf_t *db, uint64_t nsub);
803 
804 /*
805  * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
806  */
807 void *dmu_buf_get_user(dmu_buf_t *db);
808 
809 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
810 
811 /* Block until any in-progress dmu buf user evictions complete. */
812 void dmu_buf_user_evict_wait(void);
813 
814 /*
815  * Returns the blkptr associated with this dbuf, or NULL if not set.
816  */
817 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
818 
819 /*
820  * Indicate that you are going to modify the buffer's data (db_data).
821  *
822  * The transaction (tx) must be assigned to a txg (ie. you've called
823  * dmu_tx_assign()).  The buffer's object must be held in the tx
824  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
825  */
826 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
827 void dmu_buf_will_dirty_flags(dmu_buf_t *db, dmu_tx_t *tx, dmu_flags_t flags);
828 void dmu_buf_will_rewrite(dmu_buf_t *db, dmu_tx_t *tx);
829 boolean_t dmu_buf_is_dirty(dmu_buf_t *db, dmu_tx_t *tx);
830 void dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder,
831     const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx);
832 
833 /*
834  * You must create a transaction, then hold the objects which you will
835  * (or might) modify as part of this transaction.  Then you must assign
836  * the transaction to a transaction group.  Once the transaction has
837  * been assigned, you can modify buffers which belong to held objects as
838  * part of this transaction.  You can't modify buffers before the
839  * transaction has been assigned; you can't modify buffers which don't
840  * belong to objects which this transaction holds; you can't hold
841  * objects once the transaction has been assigned.  You may hold an
842  * object which you are going to free (with dmu_object_free()), but you
843  * don't have to.
844  *
845  * You can abort the transaction before it has been assigned.
846  *
847  * Note that you may hold buffers (with dmu_buf_hold) at any time,
848  * regardless of transaction state.
849  */
850 
851 #define	DMU_NEW_OBJECT	(-1ULL)
852 #define	DMU_OBJECT_END	(-1ULL)
853 
854 dmu_tx_t *dmu_tx_create(objset_t *os);
855 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
856 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
857     int len);
858 void dmu_tx_hold_append(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
859 void dmu_tx_hold_append_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
860     int len);
861 void dmu_tx_hold_clone_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
862     uint64_t len, uint_t blksz);
863 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
864     uint64_t len);
865 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
866     uint64_t len);
867 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
868 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
869     const char *name);
870 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
871 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
872 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
873 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
874 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
875 void dmu_tx_abort(dmu_tx_t *tx);
876 int dmu_tx_assign(dmu_tx_t *tx, dmu_tx_flag_t flags);
877 void dmu_tx_wait(dmu_tx_t *tx);
878 void dmu_tx_commit(dmu_tx_t *tx);
879 void dmu_tx_mark_netfree(dmu_tx_t *tx);
880 
881 /*
882  * To register a commit callback, dmu_tx_callback_register() must be called.
883  *
884  * dcb_data is a pointer to caller private data that is passed on as a
885  * callback parameter. The caller is responsible for properly allocating and
886  * freeing it.
887  *
888  * When registering a callback, the transaction must be already created, but
889  * it cannot be committed or aborted. It can be assigned to a txg or not.
890  *
891  * The callback will be called after the transaction has been safely written
892  * to stable storage and will also be called if the dmu_tx is aborted.
893  * If there is any error which prevents the transaction from being committed to
894  * disk, the callback will be called with a value of error != 0.
895  *
896  * When multiple callbacks are registered to the transaction, the callbacks
897  * will be called in reverse order to let Lustre, the only user of commit
898  * callback currently, take the fast path of its commit callback handling.
899  */
900 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
901 
902 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
903     void *dcb_data);
904 void dmu_tx_do_callbacks(list_t *cb_list, int error);
905 
906 /*
907  * Free up the data blocks for a defined range of a file.  If size is
908  * -1, the range from offset to end-of-file is freed.
909  */
910 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
911     uint64_t size, dmu_tx_t *tx);
912 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
913     uint64_t size);
914 int dmu_free_long_object(objset_t *os, uint64_t object);
915 
916 /*
917  * Convenience functions.
918  *
919  * Canfail routines will return 0 on success, or an errno if there is a
920  * nonrecoverable I/O error.
921  */
922 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
923     void *buf, dmu_flags_t flags);
924 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
925     dmu_flags_t flags);
926 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
927     const void *buf, dmu_tx_t *tx);
928 int dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
929     const void *buf, dmu_tx_t *tx, dmu_flags_t flags);
930 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
931     dmu_tx_t *tx);
932 #ifdef _KERNEL
933 int dmu_read_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size,
934     dmu_flags_t flags);
935 int dmu_read_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size,
936     dmu_flags_t flags);
937 int dmu_read_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size,
938     dmu_flags_t flags);
939 int dmu_write_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size,
940 	dmu_tx_t *tx, dmu_flags_t flags);
941 int dmu_write_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size,
942 	dmu_tx_t *tx, dmu_flags_t flags);
943 int dmu_write_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size,
944 	dmu_tx_t *tx, dmu_flags_t flags);
945 #endif
946 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
947 void dmu_return_arcbuf(struct arc_buf *buf);
948 int dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset,
949     struct arc_buf *buf, dmu_tx_t *tx, dmu_flags_t flags);
950 int dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset,
951     struct arc_buf *buf, dmu_tx_t *tx, dmu_flags_t flags);
952 #define	dmu_assign_arcbuf	dmu_assign_arcbuf_by_dbuf
953 extern uint_t zfs_max_recordsize;
954 
955 /*
956  * Asynchronously try to read in the data.
957  */
958 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
959 	uint64_t len, enum zio_priority pri);
960 void dmu_prefetch_by_dnode(dnode_t *dn, int64_t level, uint64_t offset,
961 	uint64_t len, enum zio_priority pri);
962 void dmu_prefetch_dnode(objset_t *os, uint64_t object, enum zio_priority pri);
963 int dmu_prefetch_wait(objset_t *os, uint64_t object, uint64_t offset,
964     uint64_t size);
965 
966 typedef struct dmu_object_info {
967 	/* All sizes are in bytes unless otherwise indicated. */
968 	uint32_t doi_data_block_size;
969 	uint32_t doi_metadata_block_size;
970 	dmu_object_type_t doi_type;
971 	dmu_object_type_t doi_bonus_type;
972 	uint64_t doi_bonus_size;
973 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
974 	uint8_t doi_checksum;
975 	uint8_t doi_compress;
976 	uint8_t doi_nblkptr;
977 	uint8_t doi_pad[4];
978 	uint64_t doi_dnodesize;
979 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
980 	uint64_t doi_max_offset;
981 	uint64_t doi_fill_count;		/* number of non-empty blocks */
982 } dmu_object_info_t;
983 
984 typedef void (*const arc_byteswap_func_t)(void *buf, size_t size);
985 
986 typedef struct dmu_object_type_info {
987 	dmu_object_byteswap_t	ot_byteswap;
988 	boolean_t		ot_metadata;
989 	boolean_t		ot_dbuf_metadata_cache;
990 	boolean_t		ot_encrypt;
991 	const char		*ot_name;
992 } dmu_object_type_info_t;
993 
994 typedef const struct dmu_object_byteswap_info {
995 	arc_byteswap_func_t	 ob_func;
996 	const char		*ob_name;
997 } dmu_object_byteswap_info_t;
998 
999 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
1000 extern dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
1001 
1002 /*
1003  * Get information on a DMU object.
1004  *
1005  * Return 0 on success or ENOENT if object is not allocated.
1006  *
1007  * If doi is NULL, just indicates whether the object exists.
1008  */
1009 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
1010 void __dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
1011 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
1012 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
1013 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
1014 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
1015 /*
1016  * Like dmu_object_info_from_db, but faster still when you only care about
1017  * the size.
1018  */
1019 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
1020     u_longlong_t *nblk512);
1021 
1022 void dmu_object_dnsize_from_db(dmu_buf_t *db, int *dnsize);
1023 
1024 typedef enum {
1025 	DDS_FLAG_ENCRYPTED = (1<<0),
1026 	DDS_FLAG_HAS_ENCRYPTED = (1<<7),
1027 } dmu_objset_flag_t;
1028 
1029 typedef struct dmu_objset_stats {
1030 	uint64_t dds_num_clones; /* number of clones of this */
1031 	uint64_t dds_creation_txg;
1032 	uint64_t dds_guid;
1033 	dmu_objset_type_t dds_type;
1034 	uint8_t dds_is_snapshot;
1035 	uint8_t dds_inconsistent;
1036 	uint8_t dds_redacted;
1037 	char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
1038 	uint8_t dds_flags; /* dmu_objset_flag_t */
1039 } dmu_objset_stats_t;
1040 
1041 /*
1042  * Get stats on a dataset.
1043  */
1044 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
1045 
1046 /*
1047  * Add entries to the nvlist for all the objset's properties.  See
1048  * zfs_prop_table[] and zfs(1m) for details on the properties.
1049  */
1050 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
1051 
1052 /*
1053  * Get the space usage statistics for statvfs().
1054  *
1055  * refdbytes is the amount of space "referenced" by this objset.
1056  * availbytes is the amount of space available to this objset, taking
1057  * into account quotas & reservations, assuming that no other objsets
1058  * use the space first.  These values correspond to the 'referenced' and
1059  * 'available' properties, described in the zfs(1m) manpage.
1060  *
1061  * usedobjs and availobjs are the number of objects currently allocated,
1062  * and available.
1063  */
1064 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1065     uint64_t *usedobjsp, uint64_t *availobjsp);
1066 
1067 /*
1068  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
1069  * (Contrast with the ds_guid which is a 64-bit ID that will never
1070  * change, so there is a small probability that it will collide.)
1071  */
1072 uint64_t dmu_objset_fsid_guid(objset_t *os);
1073 
1074 /*
1075  * Get the [cm]time for an objset's snapshot dir
1076  */
1077 inode_timespec_t dmu_objset_snap_cmtime(objset_t *os);
1078 
1079 int dmu_objset_is_snapshot(objset_t *os);
1080 
1081 extern struct spa *dmu_objset_spa(objset_t *os);
1082 extern struct zilog *dmu_objset_zil(objset_t *os);
1083 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
1084 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
1085 extern void dmu_objset_name(objset_t *os, char *buf);
1086 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
1087 extern uint64_t dmu_objset_id(objset_t *os);
1088 extern uint64_t dmu_objset_dnodesize(objset_t *os);
1089 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
1090 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
1091 extern int dmu_objset_blksize(objset_t *os);
1092 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1093     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
1094 extern int dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *val);
1095 extern int dmu_snapshot_realname(objset_t *os, const char *name, char *real,
1096     int maxlen, boolean_t *conflict);
1097 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
1098     uint64_t *idp, uint64_t *offp);
1099 
1100 typedef struct zfs_file_info {
1101 	uint64_t zfi_user;
1102 	uint64_t zfi_group;
1103 	uint64_t zfi_project;
1104 	uint64_t zfi_generation;
1105 } zfs_file_info_t;
1106 
1107 typedef int file_info_cb_t(dmu_object_type_t bonustype, const void *data,
1108     struct zfs_file_info *zoi);
1109 extern void dmu_objset_register_type(dmu_objset_type_t ost,
1110     file_info_cb_t *cb);
1111 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
1112 extern void *dmu_objset_get_user(objset_t *os);
1113 
1114 /*
1115  * Return the txg number for the given assigned transaction.
1116  */
1117 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
1118 
1119 /*
1120  * Synchronous write.
1121  * If a parent zio is provided this function initiates a write on the
1122  * provided buffer as a child of the parent zio.
1123  * In the absence of a parent zio, the write is completed synchronously.
1124  * At write completion, blk is filled with the bp of the written block.
1125  * Note that while the data covered by this function will be on stable
1126  * storage when the write completes this new data does not become a
1127  * permanent part of the file until the associated transaction commits.
1128  */
1129 
1130 /*
1131  * {zfs,zvol,ztest}_get_done() args
1132  */
1133 typedef struct zgd {
1134 	struct lwb	*zgd_lwb;
1135 	struct blkptr	*zgd_bp;
1136 	dmu_buf_t	*zgd_db;
1137 	struct zfs_locked_range *zgd_lr;
1138 	void		*zgd_private;
1139 } zgd_t;
1140 
1141 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
1142 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
1143 
1144 /*
1145  * Find the next hole or data block in file starting at *off
1146  * Return found offset in *off. Return ESRCH for end of file.
1147  */
1148 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
1149     uint64_t *off);
1150 
1151 int dmu_read_l0_bps(objset_t *os, uint64_t object, uint64_t offset,
1152     uint64_t length, struct blkptr *bps, size_t *nbpsp);
1153 int dmu_brt_clone(objset_t *os, uint64_t object, uint64_t offset,
1154     uint64_t length, dmu_tx_t *tx, const struct blkptr *bps, size_t nbps);
1155 
1156 /*
1157  * Initial setup and final teardown.
1158  */
1159 extern void dmu_init(void);
1160 extern void dmu_fini(void);
1161 
1162 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
1163     uint64_t object, uint64_t offset, int len);
1164 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
1165     dmu_traverse_cb_t cb, void *arg);
1166 
1167 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
1168     zfs_file_t *fp, offset_t *offp);
1169 
1170 /* CRC64 table */
1171 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
1172 extern uint64_t zfs_crc64_table[256];
1173 
1174 extern uint_t dmu_prefetch_max;
1175 
1176 #ifdef	__cplusplus
1177 }
1178 #endif
1179 
1180 #endif	/* _SYS_DMU_H */
1181