xref: /titanic_51/usr/src/uts/common/fs/zfs/sys/dmu.h (revision 3e30c24aeefdee1631958ecf17f18da671781956)
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) 2012 by Delphix. All rights reserved.
25  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
26  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
27  */
28 
29 /* Portions Copyright 2010 Robert Milkowski */
30 
31 #ifndef	_SYS_DMU_H
32 #define	_SYS_DMU_H
33 
34 /*
35  * This file describes the interface that the DMU provides for its
36  * consumers.
37  *
38  * The DMU also interacts with the SPA.  That interface is described in
39  * dmu_spa.h.
40  */
41 
42 #include <sys/inttypes.h>
43 #include <sys/types.h>
44 #include <sys/param.h>
45 #include <sys/cred.h>
46 #include <sys/time.h>
47 #include <sys/fs/zfs.h>
48 
49 #ifdef	__cplusplus
50 extern "C" {
51 #endif
52 
53 struct uio;
54 struct xuio;
55 struct page;
56 struct vnode;
57 struct spa;
58 struct zilog;
59 struct zio;
60 struct blkptr;
61 struct zap_cursor;
62 struct dsl_dataset;
63 struct dsl_pool;
64 struct dnode;
65 struct drr_begin;
66 struct drr_end;
67 struct zbookmark;
68 struct spa;
69 struct nvlist;
70 struct arc_buf;
71 struct zio_prop;
72 struct sa_handle;
73 
74 typedef struct objset objset_t;
75 typedef struct dmu_tx dmu_tx_t;
76 typedef struct dsl_dir dsl_dir_t;
77 
78 typedef enum dmu_object_byteswap {
79 	DMU_BSWAP_UINT8,
80 	DMU_BSWAP_UINT16,
81 	DMU_BSWAP_UINT32,
82 	DMU_BSWAP_UINT64,
83 	DMU_BSWAP_ZAP,
84 	DMU_BSWAP_DNODE,
85 	DMU_BSWAP_OBJSET,
86 	DMU_BSWAP_ZNODE,
87 	DMU_BSWAP_OLDACL,
88 	DMU_BSWAP_ACL,
89 	/*
90 	 * Allocating a new byteswap type number makes the on-disk format
91 	 * incompatible with any other format that uses the same number.
92 	 *
93 	 * Data can usually be structured to work with one of the
94 	 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
95 	 */
96 	DMU_BSWAP_NUMFUNCS
97 } dmu_object_byteswap_t;
98 
99 #define	DMU_OT_NEWTYPE 0x80
100 #define	DMU_OT_METADATA 0x40
101 #define	DMU_OT_BYTESWAP_MASK 0x3f
102 
103 /*
104  * Defines a uint8_t object type. Object types specify if the data
105  * in the object is metadata (boolean) and how to byteswap the data
106  * (dmu_object_byteswap_t).
107  */
108 #define	DMU_OT(byteswap, metadata) \
109 	(DMU_OT_NEWTYPE | \
110 	((metadata) ? DMU_OT_METADATA : 0) | \
111 	((byteswap) & DMU_OT_BYTESWAP_MASK))
112 
113 #define	DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
114 	((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
115 	(ot) < DMU_OT_NUMTYPES)
116 
117 #define	DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
118 	((ot) & DMU_OT_METADATA) : \
119 	dmu_ot[(ot)].ot_metadata)
120 
121 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
122 	((ot) & DMU_OT_BYTESWAP_MASK) : \
123 	dmu_ot[(ot)].ot_byteswap)
124 
125 typedef enum dmu_object_type {
126 	DMU_OT_NONE,
127 	/* general: */
128 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
129 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
130 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
131 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
132 	DMU_OT_BPOBJ,			/* UINT64 */
133 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
134 	/* spa: */
135 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
136 	DMU_OT_SPACE_MAP,		/* UINT64 */
137 	/* zil: */
138 	DMU_OT_INTENT_LOG,		/* UINT64 */
139 	/* dmu: */
140 	DMU_OT_DNODE,			/* DNODE */
141 	DMU_OT_OBJSET,			/* OBJSET */
142 	/* dsl: */
143 	DMU_OT_DSL_DIR,			/* UINT64 */
144 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
145 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
146 	DMU_OT_DSL_PROPS,		/* ZAP */
147 	DMU_OT_DSL_DATASET,		/* UINT64 */
148 	/* zpl: */
149 	DMU_OT_ZNODE,			/* ZNODE */
150 	DMU_OT_OLDACL,			/* Old ACL */
151 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
152 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
153 	DMU_OT_MASTER_NODE,		/* ZAP */
154 	DMU_OT_UNLINKED_SET,		/* ZAP */
155 	/* zvol: */
156 	DMU_OT_ZVOL,			/* UINT8 */
157 	DMU_OT_ZVOL_PROP,		/* ZAP */
158 	/* other; for testing only! */
159 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
160 	DMU_OT_UINT64_OTHER,		/* UINT64 */
161 	DMU_OT_ZAP_OTHER,		/* ZAP */
162 	/* new object types: */
163 	DMU_OT_ERROR_LOG,		/* ZAP */
164 	DMU_OT_SPA_HISTORY,		/* UINT8 */
165 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
166 	DMU_OT_POOL_PROPS,		/* ZAP */
167 	DMU_OT_DSL_PERMS,		/* ZAP */
168 	DMU_OT_ACL,			/* ACL */
169 	DMU_OT_SYSACL,			/* SYSACL */
170 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
171 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
172 	DMU_OT_NEXT_CLONES,		/* ZAP */
173 	DMU_OT_SCAN_QUEUE,		/* ZAP */
174 	DMU_OT_USERGROUP_USED,		/* ZAP */
175 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
176 	DMU_OT_USERREFS,		/* ZAP */
177 	DMU_OT_DDT_ZAP,			/* ZAP */
178 	DMU_OT_DDT_STATS,		/* ZAP */
179 	DMU_OT_SA,			/* System attr */
180 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
181 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
182 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
183 	DMU_OT_SCAN_XLATE,		/* ZAP */
184 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
185 	DMU_OT_DEADLIST,		/* ZAP */
186 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
187 	DMU_OT_DSL_CLONES,		/* ZAP */
188 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
189 	/*
190 	 * Do not allocate new object types here. Doing so makes the on-disk
191 	 * format incompatible with any other format that uses the same object
192 	 * type number.
193 	 *
194 	 * When creating an object which does not have one of the above types
195 	 * use the DMU_OTN_* type with the correct byteswap and metadata
196 	 * values.
197 	 *
198 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
199 	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
200 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
201 	 * and DMU_OTN_* types).
202 	 */
203 	DMU_OT_NUMTYPES,
204 
205 	/*
206 	 * Names for valid types declared with DMU_OT().
207 	 */
208 	DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
209 	DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
210 	DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
211 	DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
212 	DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
213 	DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
214 	DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
215 	DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
216 	DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
217 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
218 } dmu_object_type_t;
219 
220 typedef enum txg_how {
221 	TXG_WAIT = 1,
222 	TXG_NOWAIT,
223 } txg_how_t;
224 
225 void byteswap_uint64_array(void *buf, size_t size);
226 void byteswap_uint32_array(void *buf, size_t size);
227 void byteswap_uint16_array(void *buf, size_t size);
228 void byteswap_uint8_array(void *buf, size_t size);
229 void zap_byteswap(void *buf, size_t size);
230 void zfs_oldacl_byteswap(void *buf, size_t size);
231 void zfs_acl_byteswap(void *buf, size_t size);
232 void zfs_znode_byteswap(void *buf, size_t size);
233 
234 #define	DS_FIND_SNAPSHOTS	(1<<0)
235 #define	DS_FIND_CHILDREN	(1<<1)
236 
237 /*
238  * The maximum number of bytes that can be accessed as part of one
239  * operation, including metadata.
240  */
241 #define	DMU_MAX_ACCESS (10<<20) /* 10MB */
242 #define	DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
243 
244 #define	DMU_USERUSED_OBJECT	(-1ULL)
245 #define	DMU_GROUPUSED_OBJECT	(-2ULL)
246 #define	DMU_DEADLIST_OBJECT	(-3ULL)
247 
248 /*
249  * artificial blkids for bonus buffer and spill blocks
250  */
251 #define	DMU_BONUS_BLKID		(-1ULL)
252 #define	DMU_SPILL_BLKID		(-2ULL)
253 /*
254  * Public routines to create, destroy, open, and close objsets.
255  */
256 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
257 int dmu_objset_own(const char *name, dmu_objset_type_t type,
258     boolean_t readonly, void *tag, objset_t **osp);
259 void dmu_objset_rele(objset_t *os, void *tag);
260 void dmu_objset_disown(objset_t *os, void *tag);
261 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
262 
263 void dmu_objset_evict_dbufs(objset_t *os);
264 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
265     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
266 int dmu_objset_clone(const char *name, const char *origin);
267 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
268     struct nvlist *errlist);
269 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
270 int dmu_objset_snapshot_tmp(const char *, const char *, int);
271 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
272     int flags);
273 void dmu_objset_byteswap(void *buf, size_t size);
274 int dsl_dataset_rename_snapshot(const char *fsname,
275     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
276 
277 typedef struct dmu_buf {
278 	uint64_t db_object;		/* object that this buffer is part of */
279 	uint64_t db_offset;		/* byte offset in this object */
280 	uint64_t db_size;		/* size of buffer in bytes */
281 	void *db_data;			/* data in buffer */
282 } dmu_buf_t;
283 
284 typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr);
285 
286 /*
287  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
288  */
289 #define	DMU_POOL_DIRECTORY_OBJECT	1
290 #define	DMU_POOL_CONFIG			"config"
291 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
292 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
293 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
294 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
295 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
296 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
297 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
298 #define	DMU_POOL_SPARES			"spares"
299 #define	DMU_POOL_DEFLATE		"deflate"
300 #define	DMU_POOL_HISTORY		"history"
301 #define	DMU_POOL_PROPS			"pool_props"
302 #define	DMU_POOL_L2CACHE		"l2cache"
303 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
304 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
305 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
306 #define	DMU_POOL_CREATION_VERSION	"creation_version"
307 #define	DMU_POOL_SCAN			"scan"
308 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
309 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
310 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
311 
312 /*
313  * Allocate an object from this objset.  The range of object numbers
314  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
315  *
316  * The transaction must be assigned to a txg.  The newly allocated
317  * object will be "held" in the transaction (ie. you can modify the
318  * newly allocated object in this transaction).
319  *
320  * dmu_object_alloc() chooses an object and returns it in *objectp.
321  *
322  * dmu_object_claim() allocates a specific object number.  If that
323  * number is already allocated, it fails and returns EEXIST.
324  *
325  * Return 0 on success, or ENOSPC or EEXIST as specified above.
326  */
327 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
328     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
329 int dmu_object_claim(objset_t *os, uint64_t object, 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_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
332     int blocksize, dmu_object_type_t bonustype, int bonuslen);
333 
334 /*
335  * Free an object from this objset.
336  *
337  * The object's data will be freed as well (ie. you don't need to call
338  * dmu_free(object, 0, -1, tx)).
339  *
340  * The object need not be held in the transaction.
341  *
342  * If there are any holds on this object's buffers (via dmu_buf_hold()),
343  * or tx holds on the object (via dmu_tx_hold_object()), you can not
344  * free it; it fails and returns EBUSY.
345  *
346  * If the object is not allocated, it fails and returns ENOENT.
347  *
348  * Return 0 on success, or EBUSY or ENOENT as specified above.
349  */
350 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
351 
352 /*
353  * Find the next allocated or free object.
354  *
355  * The objectp parameter is in-out.  It will be updated to be the next
356  * object which is allocated.  Ignore objects which have not been
357  * modified since txg.
358  *
359  * XXX Can only be called on a objset with no dirty data.
360  *
361  * Returns 0 on success, or ENOENT if there are no more objects.
362  */
363 int dmu_object_next(objset_t *os, uint64_t *objectp,
364     boolean_t hole, uint64_t txg);
365 
366 /*
367  * Set the data blocksize for an object.
368  *
369  * The object cannot have any blocks allcated beyond the first.  If
370  * the first block is allocated already, the new size must be greater
371  * than the current block size.  If these conditions are not met,
372  * ENOTSUP will be returned.
373  *
374  * Returns 0 on success, or EBUSY if there are any holds on the object
375  * contents, or ENOTSUP as described above.
376  */
377 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
378     int ibs, dmu_tx_t *tx);
379 
380 /*
381  * Set the checksum property on a dnode.  The new checksum algorithm will
382  * apply to all newly written blocks; existing blocks will not be affected.
383  */
384 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
385     dmu_tx_t *tx);
386 
387 /*
388  * Set the compress property on a dnode.  The new compression algorithm will
389  * apply to all newly written blocks; existing blocks will not be affected.
390  */
391 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
392     dmu_tx_t *tx);
393 
394 /*
395  * Decide how to write a block: checksum, compression, number of copies, etc.
396  */
397 #define	WP_NOFILL	0x1
398 #define	WP_DMU_SYNC	0x2
399 #define	WP_SPILL	0x4
400 
401 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
402     struct zio_prop *zp);
403 /*
404  * The bonus data is accessed more or less like a regular buffer.
405  * You must dmu_bonus_hold() to get the buffer, which will give you a
406  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
407  * data.  As with any normal buffer, you must call dmu_buf_read() to
408  * read db_data, dmu_buf_will_dirty() before modifying it, and the
409  * object must be held in an assigned transaction before calling
410  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
411  * buffer as well.  You must release your hold with dmu_buf_rele().
412  *
413  * Returns ENOENT, EIO, or 0.
414  */
415 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
416 int dmu_bonus_max(void);
417 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
418 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
419 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
420 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
421 
422 /*
423  * Special spill buffer support used by "SA" framework
424  */
425 
426 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
427 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
428     void *tag, dmu_buf_t **dbp);
429 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
430 
431 /*
432  * Obtain the DMU buffer from the specified object which contains the
433  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
434  * that it will remain in memory.  You must release the hold with
435  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
436  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
437  *
438  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
439  * on the returned buffer before reading or writing the buffer's
440  * db_data.  The comments for those routines describe what particular
441  * operations are valid after calling them.
442  *
443  * The object number must be a valid, allocated object number.
444  */
445 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
446     void *tag, dmu_buf_t **, int flags);
447 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
448 void dmu_buf_rele(dmu_buf_t *db, void *tag);
449 uint64_t dmu_buf_refcount(dmu_buf_t *db);
450 
451 /*
452  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
453  * range of an object.  A pointer to an array of dmu_buf_t*'s is
454  * returned (in *dbpp).
455  *
456  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
457  * frees the array.  The hold on the array of buffers MUST be released
458  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
459  * individually with dmu_buf_rele.
460  */
461 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
462     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
463 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
464 
465 /*
466  * Returns NULL on success, or the existing user ptr if it's already
467  * been set.
468  *
469  * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
470  *
471  * user_data_ptr_ptr should be NULL, or a pointer to a pointer which
472  * will be set to db->db_data when you are allowed to access it.  Note
473  * that db->db_data (the pointer) can change when you do dmu_buf_read(),
474  * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill().
475  * *user_data_ptr_ptr will be set to the new value when it changes.
476  *
477  * If non-NULL, pageout func will be called when this buffer is being
478  * excised from the cache, so that you can clean up the data structure
479  * pointed to by user_ptr.
480  *
481  * dmu_evict_user() will call the pageout func for all buffers in a
482  * objset with a given pageout func.
483  */
484 void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr,
485     dmu_buf_evict_func_t *pageout_func);
486 /*
487  * set_user_ie is the same as set_user, but request immediate eviction
488  * when hold count goes to zero.
489  */
490 void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
491     void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func);
492 void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
493     void *user_ptr, void *user_data_ptr_ptr,
494     dmu_buf_evict_func_t *pageout_func);
495 void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
496 
497 /*
498  * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
499  */
500 void *dmu_buf_get_user(dmu_buf_t *db);
501 
502 /*
503  * Returns the blkptr associated with this dbuf, or NULL if not set.
504  */
505 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
506 
507 /*
508  * Indicate that you are going to modify the buffer's data (db_data).
509  *
510  * The transaction (tx) must be assigned to a txg (ie. you've called
511  * dmu_tx_assign()).  The buffer's object must be held in the tx
512  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
513  */
514 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
515 
516 /*
517  * Tells if the given dbuf is freeable.
518  */
519 boolean_t dmu_buf_freeable(dmu_buf_t *);
520 
521 /*
522  * You must create a transaction, then hold the objects which you will
523  * (or might) modify as part of this transaction.  Then you must assign
524  * the transaction to a transaction group.  Once the transaction has
525  * been assigned, you can modify buffers which belong to held objects as
526  * part of this transaction.  You can't modify buffers before the
527  * transaction has been assigned; you can't modify buffers which don't
528  * belong to objects which this transaction holds; you can't hold
529  * objects once the transaction has been assigned.  You may hold an
530  * object which you are going to free (with dmu_object_free()), but you
531  * don't have to.
532  *
533  * You can abort the transaction before it has been assigned.
534  *
535  * Note that you may hold buffers (with dmu_buf_hold) at any time,
536  * regardless of transaction state.
537  */
538 
539 #define	DMU_NEW_OBJECT	(-1ULL)
540 #define	DMU_OBJECT_END	(-1ULL)
541 
542 dmu_tx_t *dmu_tx_create(objset_t *os);
543 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
544 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
545     uint64_t len);
546 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
547 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
548 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
549 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
550 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
551 void dmu_tx_abort(dmu_tx_t *tx);
552 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
553 void dmu_tx_wait(dmu_tx_t *tx);
554 void dmu_tx_commit(dmu_tx_t *tx);
555 
556 /*
557  * To register a commit callback, dmu_tx_callback_register() must be called.
558  *
559  * dcb_data is a pointer to caller private data that is passed on as a
560  * callback parameter. The caller is responsible for properly allocating and
561  * freeing it.
562  *
563  * When registering a callback, the transaction must be already created, but
564  * it cannot be committed or aborted. It can be assigned to a txg or not.
565  *
566  * The callback will be called after the transaction has been safely written
567  * to stable storage and will also be called if the dmu_tx is aborted.
568  * If there is any error which prevents the transaction from being committed to
569  * disk, the callback will be called with a value of error != 0.
570  */
571 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
572 
573 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
574     void *dcb_data);
575 
576 /*
577  * Free up the data blocks for a defined range of a file.  If size is
578  * -1, the range from offset to end-of-file is freed.
579  */
580 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
581 	uint64_t size, dmu_tx_t *tx);
582 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
583 	uint64_t size);
584 int dmu_free_object(objset_t *os, uint64_t object);
585 
586 /*
587  * Convenience functions.
588  *
589  * Canfail routines will return 0 on success, or an errno if there is a
590  * nonrecoverable I/O error.
591  */
592 #define	DMU_READ_PREFETCH	0 /* prefetch */
593 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
594 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
595 	void *buf, uint32_t flags);
596 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
597 	const void *buf, dmu_tx_t *tx);
598 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
599 	dmu_tx_t *tx);
600 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
601 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
602     dmu_tx_t *tx);
603 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
604     dmu_tx_t *tx);
605 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
606     uint64_t size, struct page *pp, dmu_tx_t *tx);
607 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
608 void dmu_return_arcbuf(struct arc_buf *buf);
609 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
610     dmu_tx_t *tx);
611 int dmu_xuio_init(struct xuio *uio, int niov);
612 void dmu_xuio_fini(struct xuio *uio);
613 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
614     size_t n);
615 int dmu_xuio_cnt(struct xuio *uio);
616 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
617 void dmu_xuio_clear(struct xuio *uio, int i);
618 void xuio_stat_wbuf_copied();
619 void xuio_stat_wbuf_nocopy();
620 
621 extern int zfs_prefetch_disable;
622 
623 /*
624  * Asynchronously try to read in the data.
625  */
626 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
627     uint64_t len);
628 
629 typedef struct dmu_object_info {
630 	/* All sizes are in bytes unless otherwise indicated. */
631 	uint32_t doi_data_block_size;
632 	uint32_t doi_metadata_block_size;
633 	dmu_object_type_t doi_type;
634 	dmu_object_type_t doi_bonus_type;
635 	uint64_t doi_bonus_size;
636 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
637 	uint8_t doi_checksum;
638 	uint8_t doi_compress;
639 	uint8_t doi_pad[5];
640 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
641 	uint64_t doi_max_offset;
642 	uint64_t doi_fill_count;		/* number of non-empty blocks */
643 } dmu_object_info_t;
644 
645 typedef void arc_byteswap_func_t(void *buf, size_t size);
646 
647 typedef struct dmu_object_type_info {
648 	dmu_object_byteswap_t	ot_byteswap;
649 	boolean_t		ot_metadata;
650 	char			*ot_name;
651 } dmu_object_type_info_t;
652 
653 typedef struct dmu_object_byteswap_info {
654 	arc_byteswap_func_t	*ob_func;
655 	char			*ob_name;
656 } dmu_object_byteswap_info_t;
657 
658 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
659 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
660 
661 /*
662  * Get information on a DMU object.
663  *
664  * Return 0 on success or ENOENT if object is not allocated.
665  *
666  * If doi is NULL, just indicates whether the object exists.
667  */
668 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
669 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
670 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
671 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
672 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
673 /*
674  * Like dmu_object_info_from_db, but faster still when you only care about
675  * the size.  This is specifically optimized for zfs_getattr().
676  */
677 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
678     u_longlong_t *nblk512);
679 
680 typedef struct dmu_objset_stats {
681 	uint64_t dds_num_clones; /* number of clones of this */
682 	uint64_t dds_creation_txg;
683 	uint64_t dds_guid;
684 	dmu_objset_type_t dds_type;
685 	uint8_t dds_is_snapshot;
686 	uint8_t dds_inconsistent;
687 	char dds_origin[MAXNAMELEN];
688 } dmu_objset_stats_t;
689 
690 /*
691  * Get stats on a dataset.
692  */
693 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
694 
695 /*
696  * Add entries to the nvlist for all the objset's properties.  See
697  * zfs_prop_table[] and zfs(1m) for details on the properties.
698  */
699 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
700 
701 /*
702  * Get the space usage statistics for statvfs().
703  *
704  * refdbytes is the amount of space "referenced" by this objset.
705  * availbytes is the amount of space available to this objset, taking
706  * into account quotas & reservations, assuming that no other objsets
707  * use the space first.  These values correspond to the 'referenced' and
708  * 'available' properties, described in the zfs(1m) manpage.
709  *
710  * usedobjs and availobjs are the number of objects currently allocated,
711  * and available.
712  */
713 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
714     uint64_t *usedobjsp, uint64_t *availobjsp);
715 
716 /*
717  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
718  * (Contrast with the ds_guid which is a 64-bit ID that will never
719  * change, so there is a small probability that it will collide.)
720  */
721 uint64_t dmu_objset_fsid_guid(objset_t *os);
722 
723 /*
724  * Get the [cm]time for an objset's snapshot dir
725  */
726 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
727 
728 int dmu_objset_is_snapshot(objset_t *os);
729 
730 extern struct spa *dmu_objset_spa(objset_t *os);
731 extern struct zilog *dmu_objset_zil(objset_t *os);
732 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
733 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
734 extern void dmu_objset_name(objset_t *os, char *buf);
735 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
736 extern uint64_t dmu_objset_id(objset_t *os);
737 extern uint64_t dmu_objset_syncprop(objset_t *os);
738 extern uint64_t dmu_objset_logbias(objset_t *os);
739 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
740     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
741 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
742     int maxlen, boolean_t *conflict);
743 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
744     uint64_t *idp, uint64_t *offp);
745 
746 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
747     void *bonus, uint64_t *userp, uint64_t *groupp);
748 extern void dmu_objset_register_type(dmu_objset_type_t ost,
749     objset_used_cb_t *cb);
750 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
751 extern void *dmu_objset_get_user(objset_t *os);
752 
753 /*
754  * Return the txg number for the given assigned transaction.
755  */
756 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
757 
758 /*
759  * Synchronous write.
760  * If a parent zio is provided this function initiates a write on the
761  * provided buffer as a child of the parent zio.
762  * In the absence of a parent zio, the write is completed synchronously.
763  * At write completion, blk is filled with the bp of the written block.
764  * Note that while the data covered by this function will be on stable
765  * storage when the write completes this new data does not become a
766  * permanent part of the file until the associated transaction commits.
767  */
768 
769 /*
770  * {zfs,zvol,ztest}_get_done() args
771  */
772 typedef struct zgd {
773 	struct zilog	*zgd_zilog;
774 	struct blkptr	*zgd_bp;
775 	dmu_buf_t	*zgd_db;
776 	struct rl	*zgd_rl;
777 	void		*zgd_private;
778 } zgd_t;
779 
780 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
781 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
782 
783 /*
784  * Find the next hole or data block in file starting at *off
785  * Return found offset in *off. Return ESRCH for end of file.
786  */
787 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
788     uint64_t *off);
789 
790 /*
791  * Initial setup and final teardown.
792  */
793 extern void dmu_init(void);
794 extern void dmu_fini(void);
795 
796 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
797     uint64_t object, uint64_t offset, int len);
798 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
799     dmu_traverse_cb_t cb, void *arg);
800 
801 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
802     struct vnode *vp, offset_t *offp);
803 
804 /* CRC64 table */
805 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
806 extern uint64_t zfs_crc64_table[256];
807 
808 #ifdef	__cplusplus
809 }
810 #endif
811 
812 #endif	/* _SYS_DMU_H */
813