xref: /titanic_41/usr/src/uts/common/fs/zfs/sys/dmu.h (revision e95181dad4cc30742055089ecd61e6ccb97986e0)
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) 2013 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  */
29 
30 /* Portions Copyright 2010 Robert Milkowski */
31 
32 #ifndef	_SYS_DMU_H
33 #define	_SYS_DMU_H
34 
35 /*
36  * This file describes the interface that the DMU provides for its
37  * consumers.
38  *
39  * The DMU also interacts with the SPA.  That interface is described in
40  * dmu_spa.h.
41  */
42 
43 #include <sys/inttypes.h>
44 #include <sys/types.h>
45 #include <sys/param.h>
46 #include <sys/cred.h>
47 #include <sys/time.h>
48 #include <sys/fs/zfs.h>
49 
50 #ifdef	__cplusplus
51 extern "C" {
52 #endif
53 
54 struct uio;
55 struct xuio;
56 struct page;
57 struct vnode;
58 struct spa;
59 struct zilog;
60 struct zio;
61 struct blkptr;
62 struct zap_cursor;
63 struct dsl_dataset;
64 struct dsl_pool;
65 struct dnode;
66 struct drr_begin;
67 struct drr_end;
68 struct zbookmark;
69 struct spa;
70 struct nvlist;
71 struct arc_buf;
72 struct zio_prop;
73 struct sa_handle;
74 
75 typedef struct objset objset_t;
76 typedef struct dmu_tx dmu_tx_t;
77 typedef struct dsl_dir dsl_dir_t;
78 
79 typedef enum dmu_object_byteswap {
80 	DMU_BSWAP_UINT8,
81 	DMU_BSWAP_UINT16,
82 	DMU_BSWAP_UINT32,
83 	DMU_BSWAP_UINT64,
84 	DMU_BSWAP_ZAP,
85 	DMU_BSWAP_DNODE,
86 	DMU_BSWAP_OBJSET,
87 	DMU_BSWAP_ZNODE,
88 	DMU_BSWAP_OLDACL,
89 	DMU_BSWAP_ACL,
90 	/*
91 	 * Allocating a new byteswap type number makes the on-disk format
92 	 * incompatible with any other format that uses the same number.
93 	 *
94 	 * Data can usually be structured to work with one of the
95 	 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
96 	 */
97 	DMU_BSWAP_NUMFUNCS
98 } dmu_object_byteswap_t;
99 
100 #define	DMU_OT_NEWTYPE 0x80
101 #define	DMU_OT_METADATA 0x40
102 #define	DMU_OT_BYTESWAP_MASK 0x3f
103 
104 /*
105  * Defines a uint8_t object type. Object types specify if the data
106  * in the object is metadata (boolean) and how to byteswap the data
107  * (dmu_object_byteswap_t).
108  */
109 #define	DMU_OT(byteswap, metadata) \
110 	(DMU_OT_NEWTYPE | \
111 	((metadata) ? DMU_OT_METADATA : 0) | \
112 	((byteswap) & DMU_OT_BYTESWAP_MASK))
113 
114 #define	DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
115 	((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
116 	(ot) < DMU_OT_NUMTYPES)
117 
118 #define	DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
119 	((ot) & DMU_OT_METADATA) : \
120 	dmu_ot[(ot)].ot_metadata)
121 
122 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
123 	((ot) & DMU_OT_BYTESWAP_MASK) : \
124 	dmu_ot[(ot)].ot_byteswap)
125 
126 typedef enum dmu_object_type {
127 	DMU_OT_NONE,
128 	/* general: */
129 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
130 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
131 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
132 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
133 	DMU_OT_BPOBJ,			/* UINT64 */
134 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
135 	/* spa: */
136 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
137 	DMU_OT_SPACE_MAP,		/* UINT64 */
138 	/* zil: */
139 	DMU_OT_INTENT_LOG,		/* UINT64 */
140 	/* dmu: */
141 	DMU_OT_DNODE,			/* DNODE */
142 	DMU_OT_OBJSET,			/* OBJSET */
143 	/* dsl: */
144 	DMU_OT_DSL_DIR,			/* UINT64 */
145 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
146 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
147 	DMU_OT_DSL_PROPS,		/* ZAP */
148 	DMU_OT_DSL_DATASET,		/* UINT64 */
149 	/* zpl: */
150 	DMU_OT_ZNODE,			/* ZNODE */
151 	DMU_OT_OLDACL,			/* Old ACL */
152 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
153 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
154 	DMU_OT_MASTER_NODE,		/* ZAP */
155 	DMU_OT_UNLINKED_SET,		/* ZAP */
156 	/* zvol: */
157 	DMU_OT_ZVOL,			/* UINT8 */
158 	DMU_OT_ZVOL_PROP,		/* ZAP */
159 	/* other; for testing only! */
160 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
161 	DMU_OT_UINT64_OTHER,		/* UINT64 */
162 	DMU_OT_ZAP_OTHER,		/* ZAP */
163 	/* new object types: */
164 	DMU_OT_ERROR_LOG,		/* ZAP */
165 	DMU_OT_SPA_HISTORY,		/* UINT8 */
166 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
167 	DMU_OT_POOL_PROPS,		/* ZAP */
168 	DMU_OT_DSL_PERMS,		/* ZAP */
169 	DMU_OT_ACL,			/* ACL */
170 	DMU_OT_SYSACL,			/* SYSACL */
171 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
172 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
173 	DMU_OT_NEXT_CLONES,		/* ZAP */
174 	DMU_OT_SCAN_QUEUE,		/* ZAP */
175 	DMU_OT_USERGROUP_USED,		/* ZAP */
176 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
177 	DMU_OT_USERREFS,		/* ZAP */
178 	DMU_OT_DDT_ZAP,			/* ZAP */
179 	DMU_OT_DDT_STATS,		/* ZAP */
180 	DMU_OT_SA,			/* System attr */
181 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
182 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
183 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
184 	DMU_OT_SCAN_XLATE,		/* ZAP */
185 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
186 	DMU_OT_DEADLIST,		/* ZAP */
187 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
188 	DMU_OT_DSL_CLONES,		/* ZAP */
189 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
190 	/*
191 	 * Do not allocate new object types here. Doing so makes the on-disk
192 	 * format incompatible with any other format that uses the same object
193 	 * type number.
194 	 *
195 	 * When creating an object which does not have one of the above types
196 	 * use the DMU_OTN_* type with the correct byteswap and metadata
197 	 * values.
198 	 *
199 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
200 	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
201 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
202 	 * and DMU_OTN_* types).
203 	 */
204 	DMU_OT_NUMTYPES,
205 
206 	/*
207 	 * Names for valid types declared with DMU_OT().
208 	 */
209 	DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
210 	DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
211 	DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
212 	DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
213 	DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
214 	DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
215 	DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
216 	DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
217 	DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
218 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
219 } dmu_object_type_t;
220 
221 typedef enum txg_how {
222 	TXG_WAIT = 1,
223 	TXG_NOWAIT,
224 	TXG_WAITED,
225 } txg_how_t;
226 
227 void byteswap_uint64_array(void *buf, size_t size);
228 void byteswap_uint32_array(void *buf, size_t size);
229 void byteswap_uint16_array(void *buf, size_t size);
230 void byteswap_uint8_array(void *buf, size_t size);
231 void zap_byteswap(void *buf, size_t size);
232 void zfs_oldacl_byteswap(void *buf, size_t size);
233 void zfs_acl_byteswap(void *buf, size_t size);
234 void zfs_znode_byteswap(void *buf, size_t size);
235 
236 #define	DS_FIND_SNAPSHOTS	(1<<0)
237 #define	DS_FIND_CHILDREN	(1<<1)
238 
239 /*
240  * The maximum number of bytes that can be accessed as part of one
241  * operation, including metadata.
242  */
243 #define	DMU_MAX_ACCESS (10<<20) /* 10MB */
244 #define	DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
245 
246 #define	DMU_USERUSED_OBJECT	(-1ULL)
247 #define	DMU_GROUPUSED_OBJECT	(-2ULL)
248 #define	DMU_DEADLIST_OBJECT	(-3ULL)
249 
250 /*
251  * artificial blkids for bonus buffer and spill blocks
252  */
253 #define	DMU_BONUS_BLKID		(-1ULL)
254 #define	DMU_SPILL_BLKID		(-2ULL)
255 /*
256  * Public routines to create, destroy, open, and close objsets.
257  */
258 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
259 int dmu_objset_own(const char *name, dmu_objset_type_t type,
260     boolean_t readonly, void *tag, objset_t **osp);
261 void dmu_objset_rele(objset_t *os, void *tag);
262 void dmu_objset_disown(objset_t *os, void *tag);
263 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
264 
265 void dmu_objset_evict_dbufs(objset_t *os);
266 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
267     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
268 int dmu_objset_clone(const char *name, const char *origin);
269 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
270     struct nvlist *errlist);
271 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
272 int dmu_objset_snapshot_tmp(const char *, const char *, int);
273 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
274     int flags);
275 void dmu_objset_byteswap(void *buf, size_t size);
276 int dsl_dataset_rename_snapshot(const char *fsname,
277     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
278 
279 typedef struct dmu_buf {
280 	uint64_t db_object;		/* object that this buffer is part of */
281 	uint64_t db_offset;		/* byte offset in this object */
282 	uint64_t db_size;		/* size of buffer in bytes */
283 	void *db_data;			/* data in buffer */
284 } dmu_buf_t;
285 
286 typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr);
287 
288 /*
289  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
290  */
291 #define	DMU_POOL_DIRECTORY_OBJECT	1
292 #define	DMU_POOL_CONFIG			"config"
293 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
294 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
295 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
296 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
297 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
298 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
299 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
300 #define	DMU_POOL_SPARES			"spares"
301 #define	DMU_POOL_DEFLATE		"deflate"
302 #define	DMU_POOL_HISTORY		"history"
303 #define	DMU_POOL_PROPS			"pool_props"
304 #define	DMU_POOL_L2CACHE		"l2cache"
305 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
306 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
307 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
308 #define	DMU_POOL_CREATION_VERSION	"creation_version"
309 #define	DMU_POOL_SCAN			"scan"
310 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
311 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
312 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
313 
314 /*
315  * Allocate an object from this objset.  The range of object numbers
316  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
317  *
318  * The transaction must be assigned to a txg.  The newly allocated
319  * object will be "held" in the transaction (ie. you can modify the
320  * newly allocated object in this transaction).
321  *
322  * dmu_object_alloc() chooses an object and returns it in *objectp.
323  *
324  * dmu_object_claim() allocates a specific object number.  If that
325  * number is already allocated, it fails and returns EEXIST.
326  *
327  * Return 0 on success, or ENOSPC or EEXIST as specified above.
328  */
329 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
330     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
331 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
332     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
333 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
334     int blocksize, dmu_object_type_t bonustype, int bonuslen);
335 
336 /*
337  * Free an object from this objset.
338  *
339  * The object's data will be freed as well (ie. you don't need to call
340  * dmu_free(object, 0, -1, tx)).
341  *
342  * The object need not be held in the transaction.
343  *
344  * If there are any holds on this object's buffers (via dmu_buf_hold()),
345  * or tx holds on the object (via dmu_tx_hold_object()), you can not
346  * free it; it fails and returns EBUSY.
347  *
348  * If the object is not allocated, it fails and returns ENOENT.
349  *
350  * Return 0 on success, or EBUSY or ENOENT as specified above.
351  */
352 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
353 
354 /*
355  * Find the next allocated or free object.
356  *
357  * The objectp parameter is in-out.  It will be updated to be the next
358  * object which is allocated.  Ignore objects which have not been
359  * modified since txg.
360  *
361  * XXX Can only be called on a objset with no dirty data.
362  *
363  * Returns 0 on success, or ENOENT if there are no more objects.
364  */
365 int dmu_object_next(objset_t *os, uint64_t *objectp,
366     boolean_t hole, uint64_t txg);
367 
368 /*
369  * Set the data blocksize for an object.
370  *
371  * The object cannot have any blocks allcated beyond the first.  If
372  * the first block is allocated already, the new size must be greater
373  * than the current block size.  If these conditions are not met,
374  * ENOTSUP will be returned.
375  *
376  * Returns 0 on success, or EBUSY if there are any holds on the object
377  * contents, or ENOTSUP as described above.
378  */
379 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
380     int ibs, dmu_tx_t *tx);
381 
382 /*
383  * Set the checksum property on a dnode.  The new checksum algorithm will
384  * apply to all newly written blocks; existing blocks will not be affected.
385  */
386 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
387     dmu_tx_t *tx);
388 
389 /*
390  * Set the compress property on a dnode.  The new compression algorithm will
391  * apply to all newly written blocks; existing blocks will not be affected.
392  */
393 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
394     dmu_tx_t *tx);
395 
396 /*
397  * Decide how to write a block: checksum, compression, number of copies, etc.
398  */
399 #define	WP_NOFILL	0x1
400 #define	WP_DMU_SYNC	0x2
401 #define	WP_SPILL	0x4
402 
403 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
404     struct zio_prop *zp);
405 /*
406  * The bonus data is accessed more or less like a regular buffer.
407  * You must dmu_bonus_hold() to get the buffer, which will give you a
408  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
409  * data.  As with any normal buffer, you must call dmu_buf_read() to
410  * read db_data, dmu_buf_will_dirty() before modifying it, and the
411  * object must be held in an assigned transaction before calling
412  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
413  * buffer as well.  You must release your hold with dmu_buf_rele().
414  *
415  * Returns ENOENT, EIO, or 0.
416  */
417 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
418 int dmu_bonus_max(void);
419 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
420 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
421 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
422 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
423 
424 /*
425  * Special spill buffer support used by "SA" framework
426  */
427 
428 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
429 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
430     void *tag, dmu_buf_t **dbp);
431 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
432 
433 /*
434  * Obtain the DMU buffer from the specified object which contains the
435  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
436  * that it will remain in memory.  You must release the hold with
437  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
438  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
439  *
440  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
441  * on the returned buffer before reading or writing the buffer's
442  * db_data.  The comments for those routines describe what particular
443  * operations are valid after calling them.
444  *
445  * The object number must be a valid, allocated object number.
446  */
447 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
448     void *tag, dmu_buf_t **, int flags);
449 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
450 void dmu_buf_rele(dmu_buf_t *db, void *tag);
451 uint64_t dmu_buf_refcount(dmu_buf_t *db);
452 
453 /*
454  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
455  * range of an object.  A pointer to an array of dmu_buf_t*'s is
456  * returned (in *dbpp).
457  *
458  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
459  * frees the array.  The hold on the array of buffers MUST be released
460  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
461  * individually with dmu_buf_rele.
462  */
463 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
464     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
465 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
466 
467 /*
468  * Returns NULL on success, or the existing user ptr if it's already
469  * been set.
470  *
471  * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
472  *
473  * user_data_ptr_ptr should be NULL, or a pointer to a pointer which
474  * will be set to db->db_data when you are allowed to access it.  Note
475  * that db->db_data (the pointer) can change when you do dmu_buf_read(),
476  * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill().
477  * *user_data_ptr_ptr will be set to the new value when it changes.
478  *
479  * If non-NULL, pageout func will be called when this buffer is being
480  * excised from the cache, so that you can clean up the data structure
481  * pointed to by user_ptr.
482  *
483  * dmu_evict_user() will call the pageout func for all buffers in a
484  * objset with a given pageout func.
485  */
486 void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr,
487     dmu_buf_evict_func_t *pageout_func);
488 /*
489  * set_user_ie is the same as set_user, but request immediate eviction
490  * when hold count goes to zero.
491  */
492 void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
493     void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func);
494 void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
495     void *user_ptr, void *user_data_ptr_ptr,
496     dmu_buf_evict_func_t *pageout_func);
497 void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
498 
499 /*
500  * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
501  */
502 void *dmu_buf_get_user(dmu_buf_t *db);
503 
504 /*
505  * Returns the blkptr associated with this dbuf, or NULL if not set.
506  */
507 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
508 
509 /*
510  * Indicate that you are going to modify the buffer's data (db_data).
511  *
512  * The transaction (tx) must be assigned to a txg (ie. you've called
513  * dmu_tx_assign()).  The buffer's object must be held in the tx
514  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
515  */
516 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
517 
518 /*
519  * Tells if the given dbuf is freeable.
520  */
521 boolean_t dmu_buf_freeable(dmu_buf_t *);
522 
523 /*
524  * You must create a transaction, then hold the objects which you will
525  * (or might) modify as part of this transaction.  Then you must assign
526  * the transaction to a transaction group.  Once the transaction has
527  * been assigned, you can modify buffers which belong to held objects as
528  * part of this transaction.  You can't modify buffers before the
529  * transaction has been assigned; you can't modify buffers which don't
530  * belong to objects which this transaction holds; you can't hold
531  * objects once the transaction has been assigned.  You may hold an
532  * object which you are going to free (with dmu_object_free()), but you
533  * don't have to.
534  *
535  * You can abort the transaction before it has been assigned.
536  *
537  * Note that you may hold buffers (with dmu_buf_hold) at any time,
538  * regardless of transaction state.
539  */
540 
541 #define	DMU_NEW_OBJECT	(-1ULL)
542 #define	DMU_OBJECT_END	(-1ULL)
543 
544 dmu_tx_t *dmu_tx_create(objset_t *os);
545 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
546 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
547     uint64_t len);
548 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
549 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
550 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
551 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
552 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
553 void dmu_tx_abort(dmu_tx_t *tx);
554 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
555 void dmu_tx_wait(dmu_tx_t *tx);
556 void dmu_tx_commit(dmu_tx_t *tx);
557 
558 /*
559  * To register a commit callback, dmu_tx_callback_register() must be called.
560  *
561  * dcb_data is a pointer to caller private data that is passed on as a
562  * callback parameter. The caller is responsible for properly allocating and
563  * freeing it.
564  *
565  * When registering a callback, the transaction must be already created, but
566  * it cannot be committed or aborted. It can be assigned to a txg or not.
567  *
568  * The callback will be called after the transaction has been safely written
569  * to stable storage and will also be called if the dmu_tx is aborted.
570  * If there is any error which prevents the transaction from being committed to
571  * disk, the callback will be called with a value of error != 0.
572  */
573 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
574 
575 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
576     void *dcb_data);
577 
578 /*
579  * Free up the data blocks for a defined range of a file.  If size is
580  * -1, the range from offset to end-of-file is freed.
581  */
582 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
583 	uint64_t size, dmu_tx_t *tx);
584 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
585 	uint64_t size);
586 int dmu_free_long_object(objset_t *os, uint64_t object);
587 
588 /*
589  * Convenience functions.
590  *
591  * Canfail routines will return 0 on success, or an errno if there is a
592  * nonrecoverable I/O error.
593  */
594 #define	DMU_READ_PREFETCH	0 /* prefetch */
595 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
596 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
597 	void *buf, uint32_t flags);
598 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
599 	const void *buf, dmu_tx_t *tx);
600 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
601 	dmu_tx_t *tx);
602 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
603 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
604     dmu_tx_t *tx);
605 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
606     dmu_tx_t *tx);
607 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
608     uint64_t size, struct page *pp, dmu_tx_t *tx);
609 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
610 void dmu_return_arcbuf(struct arc_buf *buf);
611 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
612     dmu_tx_t *tx);
613 int dmu_xuio_init(struct xuio *uio, int niov);
614 void dmu_xuio_fini(struct xuio *uio);
615 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
616     size_t n);
617 int dmu_xuio_cnt(struct xuio *uio);
618 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
619 void dmu_xuio_clear(struct xuio *uio, int i);
620 void xuio_stat_wbuf_copied();
621 void xuio_stat_wbuf_nocopy();
622 
623 extern int zfs_prefetch_disable;
624 
625 /*
626  * Asynchronously try to read in the data.
627  */
628 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
629     uint64_t len);
630 
631 typedef struct dmu_object_info {
632 	/* All sizes are in bytes unless otherwise indicated. */
633 	uint32_t doi_data_block_size;
634 	uint32_t doi_metadata_block_size;
635 	dmu_object_type_t doi_type;
636 	dmu_object_type_t doi_bonus_type;
637 	uint64_t doi_bonus_size;
638 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
639 	uint8_t doi_checksum;
640 	uint8_t doi_compress;
641 	uint8_t doi_pad[5];
642 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
643 	uint64_t doi_max_offset;
644 	uint64_t doi_fill_count;		/* number of non-empty blocks */
645 } dmu_object_info_t;
646 
647 typedef void arc_byteswap_func_t(void *buf, size_t size);
648 
649 typedef struct dmu_object_type_info {
650 	dmu_object_byteswap_t	ot_byteswap;
651 	boolean_t		ot_metadata;
652 	char			*ot_name;
653 } dmu_object_type_info_t;
654 
655 typedef struct dmu_object_byteswap_info {
656 	arc_byteswap_func_t	*ob_func;
657 	char			*ob_name;
658 } dmu_object_byteswap_info_t;
659 
660 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
661 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
662 
663 /*
664  * Get information on a DMU object.
665  *
666  * Return 0 on success or ENOENT if object is not allocated.
667  *
668  * If doi is NULL, just indicates whether the object exists.
669  */
670 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
671 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
672 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
673 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
674 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
675 /*
676  * Like dmu_object_info_from_db, but faster still when you only care about
677  * the size.  This is specifically optimized for zfs_getattr().
678  */
679 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
680     u_longlong_t *nblk512);
681 
682 typedef struct dmu_objset_stats {
683 	uint64_t dds_num_clones; /* number of clones of this */
684 	uint64_t dds_creation_txg;
685 	uint64_t dds_guid;
686 	dmu_objset_type_t dds_type;
687 	uint8_t dds_is_snapshot;
688 	uint8_t dds_inconsistent;
689 	char dds_origin[MAXNAMELEN];
690 } dmu_objset_stats_t;
691 
692 /*
693  * Get stats on a dataset.
694  */
695 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
696 
697 /*
698  * Add entries to the nvlist for all the objset's properties.  See
699  * zfs_prop_table[] and zfs(1m) for details on the properties.
700  */
701 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
702 
703 /*
704  * Get the space usage statistics for statvfs().
705  *
706  * refdbytes is the amount of space "referenced" by this objset.
707  * availbytes is the amount of space available to this objset, taking
708  * into account quotas & reservations, assuming that no other objsets
709  * use the space first.  These values correspond to the 'referenced' and
710  * 'available' properties, described in the zfs(1m) manpage.
711  *
712  * usedobjs and availobjs are the number of objects currently allocated,
713  * and available.
714  */
715 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
716     uint64_t *usedobjsp, uint64_t *availobjsp);
717 
718 /*
719  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
720  * (Contrast with the ds_guid which is a 64-bit ID that will never
721  * change, so there is a small probability that it will collide.)
722  */
723 uint64_t dmu_objset_fsid_guid(objset_t *os);
724 
725 /*
726  * Get the [cm]time for an objset's snapshot dir
727  */
728 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
729 
730 int dmu_objset_is_snapshot(objset_t *os);
731 
732 extern struct spa *dmu_objset_spa(objset_t *os);
733 extern struct zilog *dmu_objset_zil(objset_t *os);
734 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
735 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
736 extern void dmu_objset_name(objset_t *os, char *buf);
737 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
738 extern uint64_t dmu_objset_id(objset_t *os);
739 extern uint64_t dmu_objset_syncprop(objset_t *os);
740 extern uint64_t dmu_objset_logbias(objset_t *os);
741 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
742     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
743 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
744     int maxlen, boolean_t *conflict);
745 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
746     uint64_t *idp, uint64_t *offp);
747 
748 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
749     void *bonus, uint64_t *userp, uint64_t *groupp);
750 extern void dmu_objset_register_type(dmu_objset_type_t ost,
751     objset_used_cb_t *cb);
752 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
753 extern void *dmu_objset_get_user(objset_t *os);
754 
755 /*
756  * Return the txg number for the given assigned transaction.
757  */
758 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
759 
760 /*
761  * Synchronous write.
762  * If a parent zio is provided this function initiates a write on the
763  * provided buffer as a child of the parent zio.
764  * In the absence of a parent zio, the write is completed synchronously.
765  * At write completion, blk is filled with the bp of the written block.
766  * Note that while the data covered by this function will be on stable
767  * storage when the write completes this new data does not become a
768  * permanent part of the file until the associated transaction commits.
769  */
770 
771 /*
772  * {zfs,zvol,ztest}_get_done() args
773  */
774 typedef struct zgd {
775 	struct zilog	*zgd_zilog;
776 	struct blkptr	*zgd_bp;
777 	dmu_buf_t	*zgd_db;
778 	struct rl	*zgd_rl;
779 	void		*zgd_private;
780 } zgd_t;
781 
782 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
783 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
784 
785 /*
786  * Find the next hole or data block in file starting at *off
787  * Return found offset in *off. Return ESRCH for end of file.
788  */
789 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
790     uint64_t *off);
791 
792 /*
793  * Initial setup and final teardown.
794  */
795 extern void dmu_init(void);
796 extern void dmu_fini(void);
797 
798 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
799     uint64_t object, uint64_t offset, int len);
800 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
801     dmu_traverse_cb_t cb, void *arg);
802 
803 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
804     struct vnode *vp, offset_t *offp);
805 
806 /* CRC64 table */
807 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
808 extern uint64_t zfs_crc64_table[256];
809 
810 extern int zfs_mdcomp_disable;
811 
812 #ifdef	__cplusplus
813 }
814 #endif
815 
816 #endif	/* _SYS_DMU_H */
817