xref: /titanic_50/usr/src/uts/common/fs/zfs/sys/dmu.h (revision 59927d313a821b7f3822314ed16fc0a44c128431)
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_FEATURE_ENABLED_TXG	"feature_enabled_txg"
297 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
298 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
299 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
300 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
301 #define	DMU_POOL_SPARES			"spares"
302 #define	DMU_POOL_DEFLATE		"deflate"
303 #define	DMU_POOL_HISTORY		"history"
304 #define	DMU_POOL_PROPS			"pool_props"
305 #define	DMU_POOL_L2CACHE		"l2cache"
306 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
307 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
308 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
309 #define	DMU_POOL_CREATION_VERSION	"creation_version"
310 #define	DMU_POOL_SCAN			"scan"
311 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
312 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
313 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
314 
315 /*
316  * Allocate an object from this objset.  The range of object numbers
317  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
318  *
319  * The transaction must be assigned to a txg.  The newly allocated
320  * object will be "held" in the transaction (ie. you can modify the
321  * newly allocated object in this transaction).
322  *
323  * dmu_object_alloc() chooses an object and returns it in *objectp.
324  *
325  * dmu_object_claim() allocates a specific object number.  If that
326  * number is already allocated, it fails and returns EEXIST.
327  *
328  * Return 0 on success, or ENOSPC or EEXIST as specified above.
329  */
330 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
331     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
332 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
333     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
334 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
335     int blocksize, dmu_object_type_t bonustype, int bonuslen);
336 
337 /*
338  * Free an object from this objset.
339  *
340  * The object's data will be freed as well (ie. you don't need to call
341  * dmu_free(object, 0, -1, tx)).
342  *
343  * The object need not be held in the transaction.
344  *
345  * If there are any holds on this object's buffers (via dmu_buf_hold()),
346  * or tx holds on the object (via dmu_tx_hold_object()), you can not
347  * free it; it fails and returns EBUSY.
348  *
349  * If the object is not allocated, it fails and returns ENOENT.
350  *
351  * Return 0 on success, or EBUSY or ENOENT as specified above.
352  */
353 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
354 
355 /*
356  * Find the next allocated or free object.
357  *
358  * The objectp parameter is in-out.  It will be updated to be the next
359  * object which is allocated.  Ignore objects which have not been
360  * modified since txg.
361  *
362  * XXX Can only be called on a objset with no dirty data.
363  *
364  * Returns 0 on success, or ENOENT if there are no more objects.
365  */
366 int dmu_object_next(objset_t *os, uint64_t *objectp,
367     boolean_t hole, uint64_t txg);
368 
369 /*
370  * Set the data blocksize for an object.
371  *
372  * The object cannot have any blocks allcated beyond the first.  If
373  * the first block is allocated already, the new size must be greater
374  * than the current block size.  If these conditions are not met,
375  * ENOTSUP will be returned.
376  *
377  * Returns 0 on success, or EBUSY if there are any holds on the object
378  * contents, or ENOTSUP as described above.
379  */
380 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
381     int ibs, dmu_tx_t *tx);
382 
383 /*
384  * Set the checksum property on a dnode.  The new checksum algorithm will
385  * apply to all newly written blocks; existing blocks will not be affected.
386  */
387 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
388     dmu_tx_t *tx);
389 
390 /*
391  * Set the compress property on a dnode.  The new compression algorithm will
392  * apply to all newly written blocks; existing blocks will not be affected.
393  */
394 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
395     dmu_tx_t *tx);
396 
397 /*
398  * Decide how to write a block: checksum, compression, number of copies, etc.
399  */
400 #define	WP_NOFILL	0x1
401 #define	WP_DMU_SYNC	0x2
402 #define	WP_SPILL	0x4
403 
404 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
405     struct zio_prop *zp);
406 /*
407  * The bonus data is accessed more or less like a regular buffer.
408  * You must dmu_bonus_hold() to get the buffer, which will give you a
409  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
410  * data.  As with any normal buffer, you must call dmu_buf_read() to
411  * read db_data, dmu_buf_will_dirty() before modifying it, and the
412  * object must be held in an assigned transaction before calling
413  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
414  * buffer as well.  You must release your hold with dmu_buf_rele().
415  *
416  * Returns ENOENT, EIO, or 0.
417  */
418 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
419 int dmu_bonus_max(void);
420 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
421 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
422 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
423 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
424 
425 /*
426  * Special spill buffer support used by "SA" framework
427  */
428 
429 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
430 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
431     void *tag, dmu_buf_t **dbp);
432 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
433 
434 /*
435  * Obtain the DMU buffer from the specified object which contains the
436  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
437  * that it will remain in memory.  You must release the hold with
438  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
439  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
440  *
441  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
442  * on the returned buffer before reading or writing the buffer's
443  * db_data.  The comments for those routines describe what particular
444  * operations are valid after calling them.
445  *
446  * The object number must be a valid, allocated object number.
447  */
448 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
449     void *tag, dmu_buf_t **, int flags);
450 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
451 void dmu_buf_rele(dmu_buf_t *db, void *tag);
452 uint64_t dmu_buf_refcount(dmu_buf_t *db);
453 
454 /*
455  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
456  * range of an object.  A pointer to an array of dmu_buf_t*'s is
457  * returned (in *dbpp).
458  *
459  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
460  * frees the array.  The hold on the array of buffers MUST be released
461  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
462  * individually with dmu_buf_rele.
463  */
464 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
465     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
466 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
467 
468 /*
469  * Returns NULL on success, or the existing user ptr if it's already
470  * been set.
471  *
472  * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
473  *
474  * user_data_ptr_ptr should be NULL, or a pointer to a pointer which
475  * will be set to db->db_data when you are allowed to access it.  Note
476  * that db->db_data (the pointer) can change when you do dmu_buf_read(),
477  * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill().
478  * *user_data_ptr_ptr will be set to the new value when it changes.
479  *
480  * If non-NULL, pageout func will be called when this buffer is being
481  * excised from the cache, so that you can clean up the data structure
482  * pointed to by user_ptr.
483  *
484  * dmu_evict_user() will call the pageout func for all buffers in a
485  * objset with a given pageout func.
486  */
487 void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr,
488     dmu_buf_evict_func_t *pageout_func);
489 /*
490  * set_user_ie is the same as set_user, but request immediate eviction
491  * when hold count goes to zero.
492  */
493 void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
494     void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func);
495 void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
496     void *user_ptr, void *user_data_ptr_ptr,
497     dmu_buf_evict_func_t *pageout_func);
498 void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
499 
500 /*
501  * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
502  */
503 void *dmu_buf_get_user(dmu_buf_t *db);
504 
505 /*
506  * Returns the blkptr associated with this dbuf, or NULL if not set.
507  */
508 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
509 
510 /*
511  * Indicate that you are going to modify the buffer's data (db_data).
512  *
513  * The transaction (tx) must be assigned to a txg (ie. you've called
514  * dmu_tx_assign()).  The buffer's object must be held in the tx
515  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
516  */
517 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
518 
519 /*
520  * Tells if the given dbuf is freeable.
521  */
522 boolean_t dmu_buf_freeable(dmu_buf_t *);
523 
524 /*
525  * You must create a transaction, then hold the objects which you will
526  * (or might) modify as part of this transaction.  Then you must assign
527  * the transaction to a transaction group.  Once the transaction has
528  * been assigned, you can modify buffers which belong to held objects as
529  * part of this transaction.  You can't modify buffers before the
530  * transaction has been assigned; you can't modify buffers which don't
531  * belong to objects which this transaction holds; you can't hold
532  * objects once the transaction has been assigned.  You may hold an
533  * object which you are going to free (with dmu_object_free()), but you
534  * don't have to.
535  *
536  * You can abort the transaction before it has been assigned.
537  *
538  * Note that you may hold buffers (with dmu_buf_hold) at any time,
539  * regardless of transaction state.
540  */
541 
542 #define	DMU_NEW_OBJECT	(-1ULL)
543 #define	DMU_OBJECT_END	(-1ULL)
544 
545 dmu_tx_t *dmu_tx_create(objset_t *os);
546 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
547 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
548     uint64_t len);
549 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
550 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
551 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
552 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
553 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
554 void dmu_tx_abort(dmu_tx_t *tx);
555 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
556 void dmu_tx_wait(dmu_tx_t *tx);
557 void dmu_tx_commit(dmu_tx_t *tx);
558 
559 /*
560  * To register a commit callback, dmu_tx_callback_register() must be called.
561  *
562  * dcb_data is a pointer to caller private data that is passed on as a
563  * callback parameter. The caller is responsible for properly allocating and
564  * freeing it.
565  *
566  * When registering a callback, the transaction must be already created, but
567  * it cannot be committed or aborted. It can be assigned to a txg or not.
568  *
569  * The callback will be called after the transaction has been safely written
570  * to stable storage and will also be called if the dmu_tx is aborted.
571  * If there is any error which prevents the transaction from being committed to
572  * disk, the callback will be called with a value of error != 0.
573  */
574 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
575 
576 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
577     void *dcb_data);
578 
579 /*
580  * Free up the data blocks for a defined range of a file.  If size is
581  * -1, the range from offset to end-of-file is freed.
582  */
583 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
584 	uint64_t size, dmu_tx_t *tx);
585 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
586 	uint64_t size);
587 int dmu_free_long_object(objset_t *os, uint64_t object);
588 
589 /*
590  * Convenience functions.
591  *
592  * Canfail routines will return 0 on success, or an errno if there is a
593  * nonrecoverable I/O error.
594  */
595 #define	DMU_READ_PREFETCH	0 /* prefetch */
596 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
597 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
598 	void *buf, uint32_t flags);
599 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
600 	const void *buf, dmu_tx_t *tx);
601 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
602 	dmu_tx_t *tx);
603 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
604 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
605     dmu_tx_t *tx);
606 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
607     dmu_tx_t *tx);
608 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
609     uint64_t size, struct page *pp, dmu_tx_t *tx);
610 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
611 void dmu_return_arcbuf(struct arc_buf *buf);
612 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
613     dmu_tx_t *tx);
614 int dmu_xuio_init(struct xuio *uio, int niov);
615 void dmu_xuio_fini(struct xuio *uio);
616 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
617     size_t n);
618 int dmu_xuio_cnt(struct xuio *uio);
619 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
620 void dmu_xuio_clear(struct xuio *uio, int i);
621 void xuio_stat_wbuf_copied();
622 void xuio_stat_wbuf_nocopy();
623 
624 extern int zfs_prefetch_disable;
625 
626 /*
627  * Asynchronously try to read in the data.
628  */
629 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
630     uint64_t len);
631 
632 typedef struct dmu_object_info {
633 	/* All sizes are in bytes unless otherwise indicated. */
634 	uint32_t doi_data_block_size;
635 	uint32_t doi_metadata_block_size;
636 	dmu_object_type_t doi_type;
637 	dmu_object_type_t doi_bonus_type;
638 	uint64_t doi_bonus_size;
639 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
640 	uint8_t doi_checksum;
641 	uint8_t doi_compress;
642 	uint8_t doi_pad[5];
643 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
644 	uint64_t doi_max_offset;
645 	uint64_t doi_fill_count;		/* number of non-empty blocks */
646 } dmu_object_info_t;
647 
648 typedef void arc_byteswap_func_t(void *buf, size_t size);
649 
650 typedef struct dmu_object_type_info {
651 	dmu_object_byteswap_t	ot_byteswap;
652 	boolean_t		ot_metadata;
653 	char			*ot_name;
654 } dmu_object_type_info_t;
655 
656 typedef struct dmu_object_byteswap_info {
657 	arc_byteswap_func_t	*ob_func;
658 	char			*ob_name;
659 } dmu_object_byteswap_info_t;
660 
661 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
662 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
663 
664 /*
665  * Get information on a DMU object.
666  *
667  * Return 0 on success or ENOENT if object is not allocated.
668  *
669  * If doi is NULL, just indicates whether the object exists.
670  */
671 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
672 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
673 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
674 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
675 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
676 /*
677  * Like dmu_object_info_from_db, but faster still when you only care about
678  * the size.  This is specifically optimized for zfs_getattr().
679  */
680 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
681     u_longlong_t *nblk512);
682 
683 typedef struct dmu_objset_stats {
684 	uint64_t dds_num_clones; /* number of clones of this */
685 	uint64_t dds_creation_txg;
686 	uint64_t dds_guid;
687 	dmu_objset_type_t dds_type;
688 	uint8_t dds_is_snapshot;
689 	uint8_t dds_inconsistent;
690 	char dds_origin[MAXNAMELEN];
691 } dmu_objset_stats_t;
692 
693 /*
694  * Get stats on a dataset.
695  */
696 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
697 
698 /*
699  * Add entries to the nvlist for all the objset's properties.  See
700  * zfs_prop_table[] and zfs(1m) for details on the properties.
701  */
702 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
703 
704 /*
705  * Get the space usage statistics for statvfs().
706  *
707  * refdbytes is the amount of space "referenced" by this objset.
708  * availbytes is the amount of space available to this objset, taking
709  * into account quotas & reservations, assuming that no other objsets
710  * use the space first.  These values correspond to the 'referenced' and
711  * 'available' properties, described in the zfs(1m) manpage.
712  *
713  * usedobjs and availobjs are the number of objects currently allocated,
714  * and available.
715  */
716 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
717     uint64_t *usedobjsp, uint64_t *availobjsp);
718 
719 /*
720  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
721  * (Contrast with the ds_guid which is a 64-bit ID that will never
722  * change, so there is a small probability that it will collide.)
723  */
724 uint64_t dmu_objset_fsid_guid(objset_t *os);
725 
726 /*
727  * Get the [cm]time for an objset's snapshot dir
728  */
729 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
730 
731 int dmu_objset_is_snapshot(objset_t *os);
732 
733 extern struct spa *dmu_objset_spa(objset_t *os);
734 extern struct zilog *dmu_objset_zil(objset_t *os);
735 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
736 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
737 extern void dmu_objset_name(objset_t *os, char *buf);
738 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
739 extern uint64_t dmu_objset_id(objset_t *os);
740 extern uint64_t dmu_objset_syncprop(objset_t *os);
741 extern uint64_t dmu_objset_logbias(objset_t *os);
742 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
743     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
744 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
745     int maxlen, boolean_t *conflict);
746 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
747     uint64_t *idp, uint64_t *offp);
748 
749 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
750     void *bonus, uint64_t *userp, uint64_t *groupp);
751 extern void dmu_objset_register_type(dmu_objset_type_t ost,
752     objset_used_cb_t *cb);
753 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
754 extern void *dmu_objset_get_user(objset_t *os);
755 
756 /*
757  * Return the txg number for the given assigned transaction.
758  */
759 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
760 
761 /*
762  * Synchronous write.
763  * If a parent zio is provided this function initiates a write on the
764  * provided buffer as a child of the parent zio.
765  * In the absence of a parent zio, the write is completed synchronously.
766  * At write completion, blk is filled with the bp of the written block.
767  * Note that while the data covered by this function will be on stable
768  * storage when the write completes this new data does not become a
769  * permanent part of the file until the associated transaction commits.
770  */
771 
772 /*
773  * {zfs,zvol,ztest}_get_done() args
774  */
775 typedef struct zgd {
776 	struct zilog	*zgd_zilog;
777 	struct blkptr	*zgd_bp;
778 	dmu_buf_t	*zgd_db;
779 	struct rl	*zgd_rl;
780 	void		*zgd_private;
781 } zgd_t;
782 
783 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
784 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
785 
786 /*
787  * Find the next hole or data block in file starting at *off
788  * Return found offset in *off. Return ESRCH for end of file.
789  */
790 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
791     uint64_t *off);
792 
793 /*
794  * Initial setup and final teardown.
795  */
796 extern void dmu_init(void);
797 extern void dmu_fini(void);
798 
799 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
800     uint64_t object, uint64_t offset, int len);
801 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
802     dmu_traverse_cb_t cb, void *arg);
803 
804 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
805     struct vnode *vp, offset_t *offp);
806 
807 /* CRC64 table */
808 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
809 extern uint64_t zfs_crc64_table[256];
810 
811 extern int zfs_mdcomp_disable;
812 
813 #ifdef	__cplusplus
814 }
815 #endif
816 
817 #endif	/* _SYS_DMU_H */
818