xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/dmu.h (revision 0463c8009260e445fd1885cfaf98e04a586d495b)
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 http://www.opensolaris.org/os/licensing.
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 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2018 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 2013 DEY Storage Systems, Inc.
28  * Copyright 2014 HybridCluster. All rights reserved.
29  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30  * Copyright 2013 Saso Kiselkov. All rights reserved.
31  * Copyright (c) 2017, Intel Corporation.
32  * Copyright (c) 2014 Integros [integros.com]
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/zio_priority.h>
54 
55 #ifdef	__cplusplus
56 extern "C" {
57 #endif
58 
59 struct uio;
60 struct xuio;
61 struct page;
62 struct vnode;
63 struct spa;
64 struct zilog;
65 struct zio;
66 struct blkptr;
67 struct zap_cursor;
68 struct dsl_dataset;
69 struct dsl_pool;
70 struct dnode;
71 struct drr_begin;
72 struct drr_end;
73 struct zbookmark_phys;
74 struct spa;
75 struct nvlist;
76 struct arc_buf;
77 struct zio_prop;
78 struct sa_handle;
79 struct locked_range;
80 struct dsl_crypto_params;
81 
82 typedef struct objset objset_t;
83 typedef struct dmu_tx dmu_tx_t;
84 typedef struct dsl_dir dsl_dir_t;
85 typedef struct dnode dnode_t;
86 
87 typedef enum dmu_object_byteswap {
88 	DMU_BSWAP_UINT8,
89 	DMU_BSWAP_UINT16,
90 	DMU_BSWAP_UINT32,
91 	DMU_BSWAP_UINT64,
92 	DMU_BSWAP_ZAP,
93 	DMU_BSWAP_DNODE,
94 	DMU_BSWAP_OBJSET,
95 	DMU_BSWAP_ZNODE,
96 	DMU_BSWAP_OLDACL,
97 	DMU_BSWAP_ACL,
98 	/*
99 	 * Allocating a new byteswap type number makes the on-disk format
100 	 * incompatible with any other format that uses the same number.
101 	 *
102 	 * Data can usually be structured to work with one of the
103 	 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
104 	 */
105 	DMU_BSWAP_NUMFUNCS
106 } dmu_object_byteswap_t;
107 
108 #define	DMU_OT_NEWTYPE 0x80
109 #define	DMU_OT_METADATA 0x40
110 #define	DMU_OT_ENCRYPTED 0x20
111 #define	DMU_OT_BYTESWAP_MASK 0x1f
112 
113 /*
114  * Defines a uint8_t object type. Object types specify if the data
115  * in the object is metadata (boolean) and how to byteswap the data
116  * (dmu_object_byteswap_t). All of the types created by this method
117  * are cached in the dbuf metadata cache.
118  */
119 #define	DMU_OT(byteswap, metadata, encrypted) \
120 	(DMU_OT_NEWTYPE | \
121 	((metadata) ? DMU_OT_METADATA : 0) | \
122 	((encrypted) ? DMU_OT_ENCRYPTED : 0) | \
123 	((byteswap) & DMU_OT_BYTESWAP_MASK))
124 
125 #define	DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
126 	((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
127 	(ot) < DMU_OT_NUMTYPES)
128 
129 /*
130  * MDB doesn't have dmu_ot; it defines these macros itself.
131  */
132 #ifndef ZFS_MDB
133 #define	DMU_OT_IS_METADATA_IMPL(ot) (dmu_ot[ot].ot_metadata)
134 #define	DMU_OT_IS_ENCRYPTED_IMPL(ot) (dmu_ot[ot].ot_encrypt)
135 #define	DMU_OT_BYTESWAP_IMPL(ot) (dmu_ot[ot].ot_byteswap)
136 #endif
137 
138 #define	DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
139 	((ot) & DMU_OT_METADATA) : \
140 	DMU_OT_IS_METADATA_IMPL(ot))
141 
142 #define	DMU_OT_IS_DDT(ot) \
143 	((ot) == DMU_OT_DDT_ZAP)
144 
145 #define	DMU_OT_IS_ZIL(ot) \
146 	((ot) == DMU_OT_INTENT_LOG)
147 
148 /* Note: ztest uses DMU_OT_UINT64_OTHER as a proxy for file blocks */
149 #define	DMU_OT_IS_FILE(ot) \
150 	((ot) == DMU_OT_PLAIN_FILE_CONTENTS || (ot) == DMU_OT_UINT64_OTHER)
151 
152 #define	DMU_OT_IS_METADATA_CACHED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
153 	B_TRUE : dmu_ot[(ot)].ot_dbuf_metadata_cache)
154 
155 #define	DMU_OT_IS_ENCRYPTED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
156 	((ot) & DMU_OT_ENCRYPTED) : \
157 	DMU_OT_IS_ENCRYPTED_IMPL(ot))
158 
159 /*
160  * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
161  * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
162  * is repurposed for embedded BPs.
163  */
164 #define	DMU_OT_HAS_FILL(ot) \
165 	((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
166 
167 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
168 	((ot) & DMU_OT_BYTESWAP_MASK) : \
169 	DMU_OT_BYTESWAP_IMPL(ot))
170 
171 typedef enum dmu_object_type {
172 	DMU_OT_NONE,
173 	/* general: */
174 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
175 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
176 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
177 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
178 	DMU_OT_BPOBJ,			/* UINT64 */
179 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
180 	/* spa: */
181 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
182 	DMU_OT_SPACE_MAP,		/* UINT64 */
183 	/* zil: */
184 	DMU_OT_INTENT_LOG,		/* UINT64 */
185 	/* dmu: */
186 	DMU_OT_DNODE,			/* DNODE */
187 	DMU_OT_OBJSET,			/* OBJSET */
188 	/* dsl: */
189 	DMU_OT_DSL_DIR,			/* UINT64 */
190 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
191 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
192 	DMU_OT_DSL_PROPS,		/* ZAP */
193 	DMU_OT_DSL_DATASET,		/* UINT64 */
194 	/* zpl: */
195 	DMU_OT_ZNODE,			/* ZNODE */
196 	DMU_OT_OLDACL,			/* Old ACL */
197 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
198 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
199 	DMU_OT_MASTER_NODE,		/* ZAP */
200 	DMU_OT_UNLINKED_SET,		/* ZAP */
201 	/* zvol: */
202 	DMU_OT_ZVOL,			/* UINT8 */
203 	DMU_OT_ZVOL_PROP,		/* ZAP */
204 	/* other; for testing only! */
205 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
206 	DMU_OT_UINT64_OTHER,		/* UINT64 */
207 	DMU_OT_ZAP_OTHER,		/* ZAP */
208 	/* new object types: */
209 	DMU_OT_ERROR_LOG,		/* ZAP */
210 	DMU_OT_SPA_HISTORY,		/* UINT8 */
211 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
212 	DMU_OT_POOL_PROPS,		/* ZAP */
213 	DMU_OT_DSL_PERMS,		/* ZAP */
214 	DMU_OT_ACL,			/* ACL */
215 	DMU_OT_SYSACL,			/* SYSACL */
216 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
217 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
218 	DMU_OT_NEXT_CLONES,		/* ZAP */
219 	DMU_OT_SCAN_QUEUE,		/* ZAP */
220 	DMU_OT_USERGROUP_USED,		/* ZAP */
221 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
222 	DMU_OT_USERREFS,		/* ZAP */
223 	DMU_OT_DDT_ZAP,			/* ZAP */
224 	DMU_OT_DDT_STATS,		/* ZAP */
225 	DMU_OT_SA,			/* System attr */
226 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
227 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
228 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
229 	DMU_OT_SCAN_XLATE,		/* ZAP */
230 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
231 	DMU_OT_DEADLIST,		/* ZAP */
232 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
233 	DMU_OT_DSL_CLONES,		/* ZAP */
234 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
235 	/*
236 	 * Do not allocate new object types here. Doing so makes the on-disk
237 	 * format incompatible with any other format that uses the same object
238 	 * type number.
239 	 *
240 	 * When creating an object which does not have one of the above types
241 	 * use the DMU_OTN_* type with the correct byteswap and metadata
242 	 * values.
243 	 *
244 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
245 	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
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 (32 * 1024 * 1024) /* 32MB */
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, 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, void *tag, objset_t **osp);
332 void dmu_objset_rele(objset_t *os, void *tag);
333 void dmu_objset_disown(objset_t *os, boolean_t key_required, void *tag);
334 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
335 
336 void dmu_objset_evict_dbufs(objset_t *os);
337 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
338     struct dsl_crypto_params *dcp, dmu_objset_create_sync_func_t func,
339     void *arg);
340 int dmu_objset_clone(const char *name, const char *origin);
341 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
342     struct nvlist *errlist);
343 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
344 int dmu_objset_snapshot_tmp(const char *, const char *, int);
345 int dmu_objset_find(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 int dmu_objset_remap_indirects(const char *fsname);
351 
352 typedef struct dmu_buf {
353 	uint64_t db_object;		/* object that this buffer is part of */
354 	uint64_t db_offset;		/* byte offset in this object */
355 	uint64_t db_size;		/* size of buffer in bytes */
356 	void *db_data;			/* data in buffer */
357 } dmu_buf_t;
358 
359 /*
360  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
361  */
362 #define	DMU_POOL_DIRECTORY_OBJECT	1
363 #define	DMU_POOL_CONFIG			"config"
364 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
365 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
366 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
367 #define	DMU_POOL_FEATURE_ENABLED_TXG	"feature_enabled_txg"
368 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
369 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
370 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
371 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
372 #define	DMU_POOL_SPARES			"spares"
373 #define	DMU_POOL_DEFLATE		"deflate"
374 #define	DMU_POOL_HISTORY		"history"
375 #define	DMU_POOL_PROPS			"pool_props"
376 #define	DMU_POOL_L2CACHE		"l2cache"
377 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
378 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
379 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
380 #define	DMU_POOL_CREATION_VERSION	"creation_version"
381 #define	DMU_POOL_SCAN			"scan"
382 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
383 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
384 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
385 #define	DMU_POOL_CHECKSUM_SALT		"org.illumos:checksum_salt"
386 #define	DMU_POOL_VDEV_ZAP_MAP		"com.delphix:vdev_zap_map"
387 #define	DMU_POOL_REMOVING		"com.delphix:removing"
388 #define	DMU_POOL_OBSOLETE_BPOBJ		"com.delphix:obsolete_bpobj"
389 #define	DMU_POOL_CONDENSING_INDIRECT	"com.delphix:condensing_indirect"
390 #define	DMU_POOL_ZPOOL_CHECKPOINT	"com.delphix:zpool_checkpoint"
391 
392 /*
393  * Allocate an object from this objset.  The range of object numbers
394  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
395  *
396  * The transaction must be assigned to a txg.  The newly allocated
397  * object will be "held" in the transaction (ie. you can modify the
398  * newly allocated object in this transaction).
399  *
400  * dmu_object_alloc() chooses an object and returns it in *objectp.
401  *
402  * dmu_object_claim() allocates a specific object number.  If that
403  * number is already allocated, it fails and returns EEXIST.
404  *
405  * Return 0 on success, or ENOSPC or EEXIST as specified above.
406  */
407 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
408     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
409 uint64_t dmu_object_alloc_ibs(objset_t *os, dmu_object_type_t ot, int blocksize,
410     int indirect_blockshift,
411     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
412 uint64_t dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot,
413     int blocksize, dmu_object_type_t bonus_type, int bonus_len,
414     int dnodesize, dmu_tx_t *tx);
415 int dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
416     int blocksize, dmu_object_type_t bonus_type, int bonus_len,
417     int dnodesize, dmu_tx_t *tx);
418 int dmu_object_reclaim_dnsize(objset_t *os, uint64_t object,
419     dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype,
420     int bonuslen, int dnodesize, boolean_t keep_spill, dmu_tx_t *txp);
421 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
422     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
423 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
424     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
425 int dmu_object_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx);
426 
427 /*
428  * Free an object from this objset.
429  *
430  * The object's data will be freed as well (ie. you don't need to call
431  * dmu_free(object, 0, -1, tx)).
432  *
433  * The object need not be held in the transaction.
434  *
435  * If there are any holds on this object's buffers (via dmu_buf_hold()),
436  * or tx holds on the object (via dmu_tx_hold_object()), you can not
437  * free it; it fails and returns EBUSY.
438  *
439  * If the object is not allocated, it fails and returns ENOENT.
440  *
441  * Return 0 on success, or EBUSY or ENOENT as specified above.
442  */
443 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
444 
445 /*
446  * Find the next allocated or free object.
447  *
448  * The objectp parameter is in-out.  It will be updated to be the next
449  * object which is allocated.  Ignore objects which have not been
450  * modified since txg.
451  *
452  * XXX Can only be called on a objset with no dirty data.
453  *
454  * Returns 0 on success, or ENOENT if there are no more objects.
455  */
456 int dmu_object_next(objset_t *os, uint64_t *objectp,
457     boolean_t hole, uint64_t txg);
458 
459 /*
460  * Set the number of levels on a dnode. nlevels must be greater than the
461  * current number of levels or an EINVAL will be returned.
462  */
463 int dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels,
464     dmu_tx_t *tx);
465 
466 /*
467  * Set the data blocksize for an object.
468  *
469  * The object cannot have any blocks allcated beyond the first.  If
470  * the first block is allocated already, the new size must be greater
471  * than the current block size.  If these conditions are not met,
472  * ENOTSUP will be returned.
473  *
474  * Returns 0 on success, or EBUSY if there are any holds on the object
475  * contents, or ENOTSUP as described above.
476  */
477 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
478     int ibs, dmu_tx_t *tx);
479 
480 /*
481  * Manually set the maxblkid on a dnode. This will adjust nlevels accordingly
482  * to accommodate the change. When calling this function, the caller must
483  * ensure that the object's nlevels can sufficiently support the new maxblkid.
484  */
485 int dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid,
486     dmu_tx_t *tx);
487 
488 /*
489  * Set the checksum property on a dnode.  The new checksum algorithm will
490  * apply to all newly written blocks; existing blocks will not be affected.
491  */
492 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
493     dmu_tx_t *tx);
494 
495 /*
496  * Set the compress property on a dnode.  The new compression algorithm will
497  * apply to all newly written blocks; existing blocks will not be affected.
498  */
499 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
500     dmu_tx_t *tx);
501 
502 int dmu_object_remap_indirects(objset_t *os, uint64_t object, uint64_t txg);
503 
504 void
505 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
506     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
507     int compressed_size, int byteorder, dmu_tx_t *tx);
508 
509 /*
510  * Decide how to write a block: checksum, compression, number of copies, etc.
511  */
512 #define	WP_NOFILL	0x1
513 #define	WP_DMU_SYNC	0x2
514 #define	WP_SPILL	0x4
515 
516 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
517     struct zio_prop *zp);
518 void dmu_write_policy_override_compress(struct zio_prop *zp,
519     enum zio_compress compress);
520 void dmu_write_policy_override_encrypt(struct zio_prop *zp,
521     dmu_object_type_t ot, boolean_t byteorder, enum zio_compress compress,
522     const uint8_t *salt, const uint8_t *iv, const uint8_t *mac);
523 /*
524  * The bonus data is accessed more or less like a regular buffer.
525  * You must dmu_bonus_hold() to get the buffer, which will give you a
526  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
527  * data.  As with any normal buffer, you must call dmu_buf_will_dirty()
528  * before modifying it, and the
529  * object must be held in an assigned transaction before calling
530  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
531  * buffer as well.  You must release your hold with dmu_buf_rele().
532  *
533  * Returns ENOENT, EIO, or 0.
534  */
535 int dmu_bonus_hold_impl(objset_t *os, uint64_t object, void *tag,
536     uint32_t flags, dmu_buf_t **dbp);
537 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
538 int dmu_bonus_hold_by_dnode(dnode_t *dn, void *tag, dmu_buf_t **dbp,
539     uint32_t flags);
540 int dmu_bonus_max(void);
541 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
542 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
543 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
544 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
545 
546 /*
547  * Special spill buffer support used by "SA" framework
548  */
549 
550 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag,
551     dmu_buf_t **dbp);
552 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
553     void *tag, dmu_buf_t **dbp);
554 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
555 
556 /*
557  * Obtain the DMU buffer from the specified object which contains the
558  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
559  * that it will remain in memory.  You must release the hold with
560  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
561  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
562  *
563  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
564  * on the returned buffer before reading or writing the buffer's
565  * db_data.  The comments for those routines describe what particular
566  * operations are valid after calling them.
567  *
568  * The object number must be a valid, allocated object number.
569  */
570 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
571     void *tag, dmu_buf_t **, int flags);
572 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
573     void *tag, dmu_buf_t **dbp, int flags);
574 
575 /*
576  * Add a reference to a dmu buffer that has already been held via
577  * dmu_buf_hold() in the current context.
578  */
579 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
580 
581 /*
582  * Attempt to add a reference to a dmu buffer that is in an unknown state,
583  * using a pointer that may have been invalidated by eviction processing.
584  * The request will succeed if the passed in dbuf still represents the
585  * same os/object/blkid, is ineligible for eviction, and has at least
586  * one hold by a user other than the syncer.
587  */
588 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
589     uint64_t blkid, void *tag);
590 
591 void dmu_buf_rele(dmu_buf_t *db, void *tag);
592 uint64_t dmu_buf_refcount(dmu_buf_t *db);
593 uint64_t dmu_buf_user_refcount(dmu_buf_t *db);
594 
595 /*
596  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
597  * range of an object.  A pointer to an array of dmu_buf_t*'s is
598  * returned (in *dbpp).
599  *
600  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
601  * frees the array.  The hold on the array of buffers MUST be released
602  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
603  * individually with dmu_buf_rele.
604  */
605 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
606     uint64_t length, boolean_t read, void *tag,
607     int *numbufsp, dmu_buf_t ***dbpp);
608 int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
609     boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp,
610     uint32_t flags);
611 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
612 
613 typedef void dmu_buf_evict_func_t(void *user_ptr);
614 
615 /*
616  * A DMU buffer user object may be associated with a dbuf for the
617  * duration of its lifetime.  This allows the user of a dbuf (client)
618  * to attach private data to a dbuf (e.g. in-core only data such as a
619  * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
620  * when that dbuf has been evicted.  Clients typically respond to the
621  * eviction notification by freeing their private data, thus ensuring
622  * the same lifetime for both dbuf and private data.
623  *
624  * The mapping from a dmu_buf_user_t to any client private data is the
625  * client's responsibility.  All current consumers of the API with private
626  * data embed a dmu_buf_user_t as the first member of the structure for
627  * their private data.  This allows conversions between the two types
628  * with a simple cast.  Since the DMU buf user API never needs access
629  * to the private data, other strategies can be employed if necessary
630  * or convenient for the client (e.g. using __containerof() to do the
631  * conversion for private data that cannot have the dmu_buf_user_t as
632  * its first member).
633  *
634  * Eviction callbacks are executed without the dbuf mutex held or any
635  * other type of mechanism to guarantee that the dbuf is still available.
636  * For this reason, users must assume the dbuf has already been freed
637  * and not reference the dbuf from the callback context.
638  *
639  * Users requesting "immediate eviction" are notified as soon as the dbuf
640  * is only referenced by dirty records (dirties == holds).  Otherwise the
641  * notification occurs after eviction processing for the dbuf begins.
642  */
643 typedef struct dmu_buf_user {
644 	/*
645 	 * Asynchronous user eviction callback state.
646 	 */
647 	taskq_ent_t	dbu_tqent;
648 
649 	/*
650 	 * This instance's eviction function pointers.
651 	 *
652 	 * dbu_evict_func_sync is called synchronously and then
653 	 * dbu_evict_func_async is executed asynchronously on a taskq.
654 	 */
655 	dmu_buf_evict_func_t *dbu_evict_func_sync;
656 	dmu_buf_evict_func_t *dbu_evict_func_async;
657 #ifdef ZFS_DEBUG
658 	/*
659 	 * Pointer to user's dbuf pointer.  NULL for clients that do
660 	 * not associate a dbuf with their user data.
661 	 *
662 	 * The dbuf pointer is cleared upon eviction so as to catch
663 	 * use-after-evict bugs in clients.
664 	 */
665 	dmu_buf_t **dbu_clear_on_evict_dbufp;
666 #endif
667 } dmu_buf_user_t;
668 
669 /*
670  * Initialize the given dmu_buf_user_t instance with the eviction function
671  * evict_func, to be called when the user is evicted.
672  *
673  * NOTE: This function should only be called once on a given dmu_buf_user_t.
674  *       To allow enforcement of this, dbu must already be zeroed on entry.
675  */
676 /*ARGSUSED*/
677 inline void
678 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
679     dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
680 {
681 	ASSERT(dbu->dbu_evict_func_sync == NULL);
682 	ASSERT(dbu->dbu_evict_func_async == NULL);
683 
684 	/* must have at least one evict func */
685 	IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
686 	dbu->dbu_evict_func_sync = evict_func_sync;
687 	dbu->dbu_evict_func_async = evict_func_async;
688 #ifdef ZFS_DEBUG
689 	dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
690 #endif
691 }
692 
693 /*
694  * Attach user data to a dbuf and mark it for normal (when the dbuf's
695  * data is cleared or its reference count goes to zero) eviction processing.
696  *
697  * Returns NULL on success, or the existing user if another user currently
698  * owns the buffer.
699  */
700 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
701 
702 /*
703  * Attach user data to a dbuf and mark it for immediate (its dirty and
704  * reference counts are equal) eviction processing.
705  *
706  * Returns NULL on success, or the existing user if another user currently
707  * owns the buffer.
708  */
709 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
710 
711 /*
712  * Replace the current user of a dbuf.
713  *
714  * If given the current user of a dbuf, replaces the dbuf's user with
715  * "new_user" and returns the user data pointer that was replaced.
716  * Otherwise returns the current, and unmodified, dbuf user pointer.
717  */
718 void *dmu_buf_replace_user(dmu_buf_t *db,
719     dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
720 
721 /*
722  * Remove the specified user data for a DMU buffer.
723  *
724  * Returns the user that was removed on success, or the current user if
725  * another user currently owns the buffer.
726  */
727 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
728 
729 /*
730  * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
731  */
732 void *dmu_buf_get_user(dmu_buf_t *db);
733 
734 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
735 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
736 void dmu_buf_dnode_exit(dmu_buf_t *db);
737 
738 /* Block until any in-progress dmu buf user evictions complete. */
739 void dmu_buf_user_evict_wait(void);
740 
741 /*
742  * Returns the blkptr associated with this dbuf, or NULL if not set.
743  */
744 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
745 
746 /*
747  * Indicate that you are going to modify the buffer's data (db_data).
748  *
749  * The transaction (tx) must be assigned to a txg (ie. you've called
750  * dmu_tx_assign()).  The buffer's object must be held in the tx
751  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
752  */
753 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
754 void dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder,
755     const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx);
756 
757 /*
758  * You must create a transaction, then hold the objects which you will
759  * (or might) modify as part of this transaction.  Then you must assign
760  * the transaction to a transaction group.  Once the transaction has
761  * been assigned, you can modify buffers which belong to held objects as
762  * part of this transaction.  You can't modify buffers before the
763  * transaction has been assigned; you can't modify buffers which don't
764  * belong to objects which this transaction holds; you can't hold
765  * objects once the transaction has been assigned.  You may hold an
766  * object which you are going to free (with dmu_object_free()), but you
767  * don't have to.
768  *
769  * You can abort the transaction before it has been assigned.
770  *
771  * Note that you may hold buffers (with dmu_buf_hold) at any time,
772  * regardless of transaction state.
773  */
774 
775 #define	DMU_NEW_OBJECT	(-1ULL)
776 #define	DMU_OBJECT_END	(-1ULL)
777 
778 dmu_tx_t *dmu_tx_create(objset_t *os);
779 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
780 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
781     int len);
782 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
783     uint64_t len);
784 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
785     uint64_t len);
786 void dmu_tx_hold_remap_l1indirect(dmu_tx_t *tx, uint64_t object);
787 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
788 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
789     const char *name);
790 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
791 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
792 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
793 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
794 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
795 void dmu_tx_abort(dmu_tx_t *tx);
796 int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
797 void dmu_tx_wait(dmu_tx_t *tx);
798 void dmu_tx_commit(dmu_tx_t *tx);
799 void dmu_tx_mark_netfree(dmu_tx_t *tx);
800 
801 /*
802  * To register a commit callback, dmu_tx_callback_register() must be called.
803  *
804  * dcb_data is a pointer to caller private data that is passed on as a
805  * callback parameter. The caller is responsible for properly allocating and
806  * freeing it.
807  *
808  * When registering a callback, the transaction must be already created, but
809  * it cannot be committed or aborted. It can be assigned to a txg or not.
810  *
811  * The callback will be called after the transaction has been safely written
812  * to stable storage and will also be called if the dmu_tx is aborted.
813  * If there is any error which prevents the transaction from being committed to
814  * disk, the callback will be called with a value of error != 0.
815  */
816 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
817 
818 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
819     void *dcb_data);
820 
821 /*
822  * Free up the data blocks for a defined range of a file.  If size is
823  * -1, the range from offset to end-of-file is freed.
824  */
825 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
826     uint64_t size, dmu_tx_t *tx);
827 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
828     uint64_t size);
829 int dmu_free_long_object(objset_t *os, uint64_t object);
830 
831 /*
832  * Convenience functions.
833  *
834  * Canfail routines will return 0 on success, or an errno if there is a
835  * nonrecoverable I/O error.
836  */
837 #define	DMU_READ_PREFETCH	0 /* prefetch */
838 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
839 #define	DMU_READ_NO_DECRYPT	2 /* don't decrypt */
840 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
841 	void *buf, uint32_t flags);
842 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
843     uint32_t flags);
844 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
845 	const void *buf, dmu_tx_t *tx);
846 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
847     const void *buf, dmu_tx_t *tx);
848 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
849 	dmu_tx_t *tx);
850 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
851 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
852 int dmu_read_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size);
853 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
854     dmu_tx_t *tx);
855 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
856     dmu_tx_t *tx);
857 int dmu_write_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size,
858     dmu_tx_t *tx);
859 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
860     uint64_t size, struct page *pp, dmu_tx_t *tx);
861 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
862 void dmu_return_arcbuf(struct arc_buf *buf);
863 int dmu_assign_arcbuf_by_dnode(dnode_t *handle, uint64_t offset,
864     struct arc_buf *buf, dmu_tx_t *tx);
865 int dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset,
866     struct arc_buf *buf, dmu_tx_t *tx);
867 void dmu_convert_to_raw(dmu_buf_t *handle, boolean_t byteorder,
868     const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx);
869 #define	dmu_assign_arcbuf	dmu_assign_arcbuf_by_dbuf
870 void dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset,
871     dmu_buf_t *handle, dmu_tx_t *tx);
872 int dmu_xuio_init(struct xuio *uio, int niov);
873 void dmu_xuio_fini(struct xuio *uio);
874 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
875     size_t n);
876 int dmu_xuio_cnt(struct xuio *uio);
877 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
878 void dmu_xuio_clear(struct xuio *uio, int i);
879 void xuio_stat_wbuf_copied(void);
880 void xuio_stat_wbuf_nocopy(void);
881 
882 extern boolean_t zfs_prefetch_disable;
883 extern int zfs_max_recordsize;
884 
885 /*
886  * Asynchronously try to read in the data.
887  */
888 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
889     uint64_t len, enum zio_priority pri);
890 
891 typedef struct dmu_object_info {
892 	/* All sizes are in bytes unless otherwise indicated. */
893 	uint32_t doi_data_block_size;
894 	uint32_t doi_metadata_block_size;
895 	dmu_object_type_t doi_type;
896 	dmu_object_type_t doi_bonus_type;
897 	uint64_t doi_bonus_size;
898 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
899 	uint8_t doi_checksum;
900 	uint8_t doi_compress;
901 	uint8_t doi_nblkptr;
902 	int8_t doi_pad[4];
903 	uint64_t doi_dnodesize;
904 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
905 	uint64_t doi_max_offset;
906 	uint64_t doi_fill_count;		/* number of non-empty blocks */
907 } dmu_object_info_t;
908 
909 typedef void arc_byteswap_func_t(void *buf, size_t size);
910 
911 typedef struct dmu_object_type_info {
912 	dmu_object_byteswap_t	ot_byteswap;
913 	boolean_t		ot_metadata;
914 	boolean_t		ot_dbuf_metadata_cache;
915 	boolean_t		ot_encrypt;
916 	char			*ot_name;
917 } dmu_object_type_info_t;
918 
919 typedef struct dmu_object_byteswap_info {
920 	arc_byteswap_func_t	*ob_func;
921 	char			*ob_name;
922 } dmu_object_byteswap_info_t;
923 
924 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
925 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
926 
927 /*
928  * Get information on a DMU object.
929  *
930  * Return 0 on success or ENOENT if object is not allocated.
931  *
932  * If doi is NULL, just indicates whether the object exists.
933  */
934 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
935 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
936 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
937 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
938 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
939 /*
940  * Like dmu_object_info_from_db, but faster still when you only care about
941  * the size.  This is specifically optimized for zfs_getattr().
942  */
943 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
944     u_longlong_t *nblk512);
945 
946 void dmu_object_dnsize_from_db(dmu_buf_t *db, int *dnsize);
947 
948 typedef struct dmu_objset_stats {
949 	uint64_t dds_num_clones; /* number of clones of this */
950 	uint64_t dds_creation_txg;
951 	uint64_t dds_guid;
952 	dmu_objset_type_t dds_type;
953 	uint8_t dds_is_snapshot;
954 	uint8_t dds_inconsistent;
955 	char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
956 } dmu_objset_stats_t;
957 
958 /*
959  * Get stats on a dataset.
960  */
961 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
962 
963 /*
964  * Add entries to the nvlist for all the objset's properties.  See
965  * zfs_prop_table[] and zfs(1m) for details on the properties.
966  */
967 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
968 
969 /*
970  * Get the space usage statistics for statvfs().
971  *
972  * refdbytes is the amount of space "referenced" by this objset.
973  * availbytes is the amount of space available to this objset, taking
974  * into account quotas & reservations, assuming that no other objsets
975  * use the space first.  These values correspond to the 'referenced' and
976  * 'available' properties, described in the zfs(1m) manpage.
977  *
978  * usedobjs and availobjs are the number of objects currently allocated,
979  * and available.
980  */
981 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
982     uint64_t *usedobjsp, uint64_t *availobjsp);
983 
984 /*
985  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
986  * (Contrast with the ds_guid which is a 64-bit ID that will never
987  * change, so there is a small probability that it will collide.)
988  */
989 uint64_t dmu_objset_fsid_guid(objset_t *os);
990 
991 /*
992  * Get the [cm]time for an objset's snapshot dir
993  */
994 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
995 
996 int dmu_objset_is_snapshot(objset_t *os);
997 
998 extern struct spa *dmu_objset_spa(objset_t *os);
999 extern struct zilog *dmu_objset_zil(objset_t *os);
1000 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
1001 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
1002 extern void dmu_objset_name(objset_t *os, char *buf);
1003 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
1004 extern uint64_t dmu_objset_id(objset_t *os);
1005 extern uint64_t dmu_objset_dnodesize(objset_t *os);
1006 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
1007 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
1008 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1009     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
1010 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
1011     int maxlen, boolean_t *conflict);
1012 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
1013     uint64_t *idp, uint64_t *offp);
1014 
1015 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
1016     void *bonus, uint64_t *userp, uint64_t *groupp, uint64_t *projectp);
1017 extern void dmu_objset_register_type(dmu_objset_type_t ost,
1018     objset_used_cb_t *cb);
1019 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
1020 extern void *dmu_objset_get_user(objset_t *os);
1021 
1022 /*
1023  * Return the txg number for the given assigned transaction.
1024  */
1025 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
1026 
1027 /*
1028  * Synchronous write.
1029  * If a parent zio is provided this function initiates a write on the
1030  * provided buffer as a child of the parent zio.
1031  * In the absence of a parent zio, the write is completed synchronously.
1032  * At write completion, blk is filled with the bp of the written block.
1033  * Note that while the data covered by this function will be on stable
1034  * storage when the write completes this new data does not become a
1035  * permanent part of the file until the associated transaction commits.
1036  */
1037 
1038 /*
1039  * {zfs,zvol,ztest}_get_done() args
1040  */
1041 typedef struct zgd {
1042 	struct lwb	*zgd_lwb;
1043 	struct blkptr	*zgd_bp;
1044 	dmu_buf_t	*zgd_db;
1045 	struct locked_range *zgd_lr;
1046 	void		*zgd_private;
1047 } zgd_t;
1048 
1049 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
1050 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
1051 
1052 /*
1053  * Find the next hole or data block in file starting at *off
1054  * Return found offset in *off. Return ESRCH for end of file.
1055  */
1056 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
1057     uint64_t *off);
1058 
1059 /*
1060  * Check if a DMU object has any dirty blocks. If so, sync out
1061  * all pending transaction groups. Otherwise, this function
1062  * does not alter DMU state. This could be improved to only sync
1063  * out the necessary transaction groups for this particular
1064  * object.
1065  */
1066 int dmu_object_wait_synced(objset_t *os, uint64_t object);
1067 
1068 /*
1069  * Initial setup and final teardown.
1070  */
1071 extern void dmu_init(void);
1072 extern void dmu_fini(void);
1073 
1074 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
1075     uint64_t object, uint64_t offset, int len);
1076 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
1077     dmu_traverse_cb_t cb, void *arg);
1078 
1079 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
1080     struct vnode *vp, offset_t *offp);
1081 
1082 /* CRC64 table */
1083 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
1084 extern uint64_t zfs_crc64_table[256];
1085 
1086 #ifdef	__cplusplus
1087 }
1088 #endif
1089 
1090 #endif	/* _SYS_DMU_H */
1091