xref: /titanic_41/usr/src/uts/common/fs/zfs/sys/dmu.h (revision 2a62b4a97f3b5b0ef5f3288ebbfc8509fb793b60)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
26  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
27  * Copyright 2013 DEY Storage Systems, Inc.
28  * Copyright 2014 HybridCluster. All rights reserved.
29  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30  */
31 
32 /* Portions Copyright 2010 Robert Milkowski */
33 
34 #ifndef	_SYS_DMU_H
35 #define	_SYS_DMU_H
36 
37 /*
38  * This file describes the interface that the DMU provides for its
39  * consumers.
40  *
41  * The DMU also interacts with the SPA.  That interface is described in
42  * dmu_spa.h.
43  */
44 
45 #include <sys/zfs_context.h>
46 #include <sys/inttypes.h>
47 #include <sys/cred.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_phys;
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 /*
123  * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
124  * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
125  * is repurposed for embedded BPs.
126  */
127 #define	DMU_OT_HAS_FILL(ot) \
128 	((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
129 
130 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
131 	((ot) & DMU_OT_BYTESWAP_MASK) : \
132 	dmu_ot[(ot)].ot_byteswap)
133 
134 typedef enum dmu_object_type {
135 	DMU_OT_NONE,
136 	/* general: */
137 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
138 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
139 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
140 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
141 	DMU_OT_BPOBJ,			/* UINT64 */
142 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
143 	/* spa: */
144 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
145 	DMU_OT_SPACE_MAP,		/* UINT64 */
146 	/* zil: */
147 	DMU_OT_INTENT_LOG,		/* UINT64 */
148 	/* dmu: */
149 	DMU_OT_DNODE,			/* DNODE */
150 	DMU_OT_OBJSET,			/* OBJSET */
151 	/* dsl: */
152 	DMU_OT_DSL_DIR,			/* UINT64 */
153 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
154 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
155 	DMU_OT_DSL_PROPS,		/* ZAP */
156 	DMU_OT_DSL_DATASET,		/* UINT64 */
157 	/* zpl: */
158 	DMU_OT_ZNODE,			/* ZNODE */
159 	DMU_OT_OLDACL,			/* Old ACL */
160 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
161 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
162 	DMU_OT_MASTER_NODE,		/* ZAP */
163 	DMU_OT_UNLINKED_SET,		/* ZAP */
164 	/* zvol: */
165 	DMU_OT_ZVOL,			/* UINT8 */
166 	DMU_OT_ZVOL_PROP,		/* ZAP */
167 	/* other; for testing only! */
168 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
169 	DMU_OT_UINT64_OTHER,		/* UINT64 */
170 	DMU_OT_ZAP_OTHER,		/* ZAP */
171 	/* new object types: */
172 	DMU_OT_ERROR_LOG,		/* ZAP */
173 	DMU_OT_SPA_HISTORY,		/* UINT8 */
174 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
175 	DMU_OT_POOL_PROPS,		/* ZAP */
176 	DMU_OT_DSL_PERMS,		/* ZAP */
177 	DMU_OT_ACL,			/* ACL */
178 	DMU_OT_SYSACL,			/* SYSACL */
179 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
180 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
181 	DMU_OT_NEXT_CLONES,		/* ZAP */
182 	DMU_OT_SCAN_QUEUE,		/* ZAP */
183 	DMU_OT_USERGROUP_USED,		/* ZAP */
184 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
185 	DMU_OT_USERREFS,		/* ZAP */
186 	DMU_OT_DDT_ZAP,			/* ZAP */
187 	DMU_OT_DDT_STATS,		/* ZAP */
188 	DMU_OT_SA,			/* System attr */
189 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
190 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
191 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
192 	DMU_OT_SCAN_XLATE,		/* ZAP */
193 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
194 	DMU_OT_DEADLIST,		/* ZAP */
195 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
196 	DMU_OT_DSL_CLONES,		/* ZAP */
197 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
198 	/*
199 	 * Do not allocate new object types here. Doing so makes the on-disk
200 	 * format incompatible with any other format that uses the same object
201 	 * type number.
202 	 *
203 	 * When creating an object which does not have one of the above types
204 	 * use the DMU_OTN_* type with the correct byteswap and metadata
205 	 * values.
206 	 *
207 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
208 	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
209 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
210 	 * and DMU_OTN_* types).
211 	 */
212 	DMU_OT_NUMTYPES,
213 
214 	/*
215 	 * Names for valid types declared with DMU_OT().
216 	 */
217 	DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
218 	DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
219 	DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
220 	DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
221 	DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
222 	DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
223 	DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
224 	DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
225 	DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
226 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
227 } dmu_object_type_t;
228 
229 typedef enum txg_how {
230 	TXG_WAIT = 1,
231 	TXG_NOWAIT,
232 	TXG_WAITED,
233 } txg_how_t;
234 
235 void byteswap_uint64_array(void *buf, size_t size);
236 void byteswap_uint32_array(void *buf, size_t size);
237 void byteswap_uint16_array(void *buf, size_t size);
238 void byteswap_uint8_array(void *buf, size_t size);
239 void zap_byteswap(void *buf, size_t size);
240 void zfs_oldacl_byteswap(void *buf, size_t size);
241 void zfs_acl_byteswap(void *buf, size_t size);
242 void zfs_znode_byteswap(void *buf, size_t size);
243 
244 #define	DS_FIND_SNAPSHOTS	(1<<0)
245 #define	DS_FIND_CHILDREN	(1<<1)
246 #define	DS_FIND_SERIALIZE	(1<<2)
247 
248 /*
249  * The maximum number of bytes that can be accessed as part of one
250  * operation, including metadata.
251  */
252 #define	DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
253 #define	DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
254 
255 #define	DMU_USERUSED_OBJECT	(-1ULL)
256 #define	DMU_GROUPUSED_OBJECT	(-2ULL)
257 
258 /*
259  * artificial blkids for bonus buffer and spill blocks
260  */
261 #define	DMU_BONUS_BLKID		(-1ULL)
262 #define	DMU_SPILL_BLKID		(-2ULL)
263 /*
264  * Public routines to create, destroy, open, and close objsets.
265  */
266 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
267 int dmu_objset_own(const char *name, dmu_objset_type_t type,
268     boolean_t readonly, void *tag, objset_t **osp);
269 void dmu_objset_rele(objset_t *os, void *tag);
270 void dmu_objset_disown(objset_t *os, void *tag);
271 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
272 
273 void dmu_objset_evict_dbufs(objset_t *os);
274 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
275     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
276 int dmu_objset_clone(const char *name, const char *origin);
277 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
278     struct nvlist *errlist);
279 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
280 int dmu_objset_snapshot_tmp(const char *, const char *, int);
281 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
282     int flags);
283 void dmu_objset_byteswap(void *buf, size_t size);
284 int dsl_dataset_rename_snapshot(const char *fsname,
285     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
286 
287 typedef struct dmu_buf {
288 	uint64_t db_object;		/* object that this buffer is part of */
289 	uint64_t db_offset;		/* byte offset in this object */
290 	uint64_t db_size;		/* size of buffer in bytes */
291 	void *db_data;			/* data in buffer */
292 } dmu_buf_t;
293 
294 /*
295  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
296  */
297 #define	DMU_POOL_DIRECTORY_OBJECT	1
298 #define	DMU_POOL_CONFIG			"config"
299 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
300 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
301 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
302 #define	DMU_POOL_FEATURE_ENABLED_TXG	"feature_enabled_txg"
303 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
304 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
305 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
306 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
307 #define	DMU_POOL_SPARES			"spares"
308 #define	DMU_POOL_DEFLATE		"deflate"
309 #define	DMU_POOL_HISTORY		"history"
310 #define	DMU_POOL_PROPS			"pool_props"
311 #define	DMU_POOL_L2CACHE		"l2cache"
312 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
313 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
314 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
315 #define	DMU_POOL_CREATION_VERSION	"creation_version"
316 #define	DMU_POOL_SCAN			"scan"
317 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
318 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
319 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
320 
321 /*
322  * Allocate an object from this objset.  The range of object numbers
323  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
324  *
325  * The transaction must be assigned to a txg.  The newly allocated
326  * object will be "held" in the transaction (ie. you can modify the
327  * newly allocated object in this transaction).
328  *
329  * dmu_object_alloc() chooses an object and returns it in *objectp.
330  *
331  * dmu_object_claim() allocates a specific object number.  If that
332  * number is already allocated, it fails and returns EEXIST.
333  *
334  * Return 0 on success, or ENOSPC or EEXIST as specified above.
335  */
336 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
337     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
338 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
339     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
340 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
341     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
342 
343 /*
344  * Free an object from this objset.
345  *
346  * The object's data will be freed as well (ie. you don't need to call
347  * dmu_free(object, 0, -1, tx)).
348  *
349  * The object need not be held in the transaction.
350  *
351  * If there are any holds on this object's buffers (via dmu_buf_hold()),
352  * or tx holds on the object (via dmu_tx_hold_object()), you can not
353  * free it; it fails and returns EBUSY.
354  *
355  * If the object is not allocated, it fails and returns ENOENT.
356  *
357  * Return 0 on success, or EBUSY or ENOENT as specified above.
358  */
359 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
360 
361 /*
362  * Find the next allocated or free object.
363  *
364  * The objectp parameter is in-out.  It will be updated to be the next
365  * object which is allocated.  Ignore objects which have not been
366  * modified since txg.
367  *
368  * XXX Can only be called on a objset with no dirty data.
369  *
370  * Returns 0 on success, or ENOENT if there are no more objects.
371  */
372 int dmu_object_next(objset_t *os, uint64_t *objectp,
373     boolean_t hole, uint64_t txg);
374 
375 /*
376  * Set the data blocksize for an object.
377  *
378  * The object cannot have any blocks allcated beyond the first.  If
379  * the first block is allocated already, the new size must be greater
380  * than the current block size.  If these conditions are not met,
381  * ENOTSUP will be returned.
382  *
383  * Returns 0 on success, or EBUSY if there are any holds on the object
384  * contents, or ENOTSUP as described above.
385  */
386 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
387     int ibs, dmu_tx_t *tx);
388 
389 /*
390  * Set the checksum property on a dnode.  The new checksum algorithm will
391  * apply to all newly written blocks; existing blocks will not be affected.
392  */
393 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
394     dmu_tx_t *tx);
395 
396 /*
397  * Set the compress property on a dnode.  The new compression algorithm will
398  * apply to all newly written blocks; existing blocks will not be affected.
399  */
400 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
401     dmu_tx_t *tx);
402 
403 void
404 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
405     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
406     int compressed_size, int byteorder, dmu_tx_t *tx);
407 
408 /*
409  * Decide how to write a block: checksum, compression, number of copies, etc.
410  */
411 #define	WP_NOFILL	0x1
412 #define	WP_DMU_SYNC	0x2
413 #define	WP_SPILL	0x4
414 
415 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
416     struct zio_prop *zp);
417 /*
418  * The bonus data is accessed more or less like a regular buffer.
419  * You must dmu_bonus_hold() to get the buffer, which will give you a
420  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
421  * data.  As with any normal buffer, you must call dmu_buf_read() to
422  * read db_data, dmu_buf_will_dirty() before modifying it, and the
423  * object must be held in an assigned transaction before calling
424  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
425  * buffer as well.  You must release your hold with dmu_buf_rele().
426  *
427  * Returns ENOENT, EIO, or 0.
428  */
429 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
430 int dmu_bonus_max(void);
431 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
432 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
433 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
434 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
435 
436 /*
437  * Special spill buffer support used by "SA" framework
438  */
439 
440 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
441 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
442     void *tag, dmu_buf_t **dbp);
443 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
444 
445 /*
446  * Obtain the DMU buffer from the specified object which contains the
447  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
448  * that it will remain in memory.  You must release the hold with
449  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
450  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
451  *
452  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
453  * on the returned buffer before reading or writing the buffer's
454  * db_data.  The comments for those routines describe what particular
455  * operations are valid after calling them.
456  *
457  * The object number must be a valid, allocated object number.
458  */
459 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
460     void *tag, dmu_buf_t **, int flags);
461 
462 /*
463  * Add a reference to a dmu buffer that has already been held via
464  * dmu_buf_hold() in the current context.
465  */
466 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
467 
468 /*
469  * Attempt to add a reference to a dmu buffer that is in an unknown state,
470  * using a pointer that may have been invalidated by eviction processing.
471  * The request will succeed if the passed in dbuf still represents the
472  * same os/object/blkid, is ineligible for eviction, and has at least
473  * one hold by a user other than the syncer.
474  */
475 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
476     uint64_t blkid, void *tag);
477 
478 void dmu_buf_rele(dmu_buf_t *db, void *tag);
479 uint64_t dmu_buf_refcount(dmu_buf_t *db);
480 
481 /*
482  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
483  * range of an object.  A pointer to an array of dmu_buf_t*'s is
484  * returned (in *dbpp).
485  *
486  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
487  * frees the array.  The hold on the array of buffers MUST be released
488  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
489  * individually with dmu_buf_rele.
490  */
491 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
492     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
493 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
494 
495 typedef void dmu_buf_evict_func_t(void *user_ptr);
496 
497 /*
498  * A DMU buffer user object may be associated with a dbuf for the
499  * duration of its lifetime.  This allows the user of a dbuf (client)
500  * to attach private data to a dbuf (e.g. in-core only data such as a
501  * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
502  * when that dbuf has been evicted.  Clients typically respond to the
503  * eviction notification by freeing their private data, thus ensuring
504  * the same lifetime for both dbuf and private data.
505  *
506  * The mapping from a dmu_buf_user_t to any client private data is the
507  * client's responsibility.  All current consumers of the API with private
508  * data embed a dmu_buf_user_t as the first member of the structure for
509  * their private data.  This allows conversions between the two types
510  * with a simple cast.  Since the DMU buf user API never needs access
511  * to the private data, other strategies can be employed if necessary
512  * or convenient for the client (e.g. using container_of() to do the
513  * conversion for private data that cannot have the dmu_buf_user_t as
514  * its first member).
515  *
516  * Eviction callbacks are executed without the dbuf mutex held or any
517  * other type of mechanism to guarantee that the dbuf is still available.
518  * For this reason, users must assume the dbuf has already been freed
519  * and not reference the dbuf from the callback context.
520  *
521  * Users requesting "immediate eviction" are notified as soon as the dbuf
522  * is only referenced by dirty records (dirties == holds).  Otherwise the
523  * notification occurs after eviction processing for the dbuf begins.
524  */
525 typedef struct dmu_buf_user {
526 	/*
527 	 * Asynchronous user eviction callback state.
528 	 */
529 	taskq_ent_t	dbu_tqent;
530 
531 	/* This instance's eviction function pointer. */
532 	dmu_buf_evict_func_t *dbu_evict_func;
533 #ifdef ZFS_DEBUG
534 	/*
535 	 * Pointer to user's dbuf pointer.  NULL for clients that do
536 	 * not associate a dbuf with their user data.
537 	 *
538 	 * The dbuf pointer is cleared upon eviction so as to catch
539 	 * use-after-evict bugs in clients.
540 	 */
541 	dmu_buf_t **dbu_clear_on_evict_dbufp;
542 #endif
543 } dmu_buf_user_t;
544 
545 /*
546  * Initialize the given dmu_buf_user_t instance with the eviction function
547  * evict_func, to be called when the user is evicted.
548  *
549  * NOTE: This function should only be called once on a given dmu_buf_user_t.
550  *       To allow enforcement of this, dbu must already be zeroed on entry.
551  */
552 #ifdef __lint
553 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
554 extern void
555 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
556     dmu_buf_t **clear_on_evict_dbufp);
557 #else /* __lint */
558 inline void
559 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
560     dmu_buf_t **clear_on_evict_dbufp)
561 {
562 	ASSERT(dbu->dbu_evict_func == NULL);
563 	ASSERT(evict_func != NULL);
564 	dbu->dbu_evict_func = evict_func;
565 #ifdef ZFS_DEBUG
566 	dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
567 #endif
568 }
569 #endif /* __lint */
570 
571 /*
572  * Attach user data to a dbuf and mark it for normal (when the dbuf's
573  * data is cleared or its reference count goes to zero) eviction processing.
574  *
575  * Returns NULL on success, or the existing user if another user currently
576  * owns the buffer.
577  */
578 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
579 
580 /*
581  * Attach user data to a dbuf and mark it for immediate (its dirty and
582  * reference counts are equal) eviction processing.
583  *
584  * Returns NULL on success, or the existing user if another user currently
585  * owns the buffer.
586  */
587 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
588 
589 /*
590  * Replace the current user of a dbuf.
591  *
592  * If given the current user of a dbuf, replaces the dbuf's user with
593  * "new_user" and returns the user data pointer that was replaced.
594  * Otherwise returns the current, and unmodified, dbuf user pointer.
595  */
596 void *dmu_buf_replace_user(dmu_buf_t *db,
597     dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
598 
599 /*
600  * Remove the specified user data for a DMU buffer.
601  *
602  * Returns the user that was removed on success, or the current user if
603  * another user currently owns the buffer.
604  */
605 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
606 
607 /*
608  * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
609  */
610 void *dmu_buf_get_user(dmu_buf_t *db);
611 
612 /* Block until any in-progress dmu buf user evictions complete. */
613 void dmu_buf_user_evict_wait(void);
614 
615 /*
616  * Returns the blkptr associated with this dbuf, or NULL if not set.
617  */
618 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
619 
620 /*
621  * Indicate that you are going to modify the buffer's data (db_data).
622  *
623  * The transaction (tx) must be assigned to a txg (ie. you've called
624  * dmu_tx_assign()).  The buffer's object must be held in the tx
625  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
626  */
627 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
628 
629 /*
630  * Tells if the given dbuf is freeable.
631  */
632 boolean_t dmu_buf_freeable(dmu_buf_t *);
633 
634 /*
635  * You must create a transaction, then hold the objects which you will
636  * (or might) modify as part of this transaction.  Then you must assign
637  * the transaction to a transaction group.  Once the transaction has
638  * been assigned, you can modify buffers which belong to held objects as
639  * part of this transaction.  You can't modify buffers before the
640  * transaction has been assigned; you can't modify buffers which don't
641  * belong to objects which this transaction holds; you can't hold
642  * objects once the transaction has been assigned.  You may hold an
643  * object which you are going to free (with dmu_object_free()), but you
644  * don't have to.
645  *
646  * You can abort the transaction before it has been assigned.
647  *
648  * Note that you may hold buffers (with dmu_buf_hold) at any time,
649  * regardless of transaction state.
650  */
651 
652 #define	DMU_NEW_OBJECT	(-1ULL)
653 #define	DMU_OBJECT_END	(-1ULL)
654 
655 dmu_tx_t *dmu_tx_create(objset_t *os);
656 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
657 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
658     uint64_t len);
659 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
660 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
661 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
662 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
663 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
664 void dmu_tx_abort(dmu_tx_t *tx);
665 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
666 void dmu_tx_wait(dmu_tx_t *tx);
667 void dmu_tx_commit(dmu_tx_t *tx);
668 void dmu_tx_mark_netfree(dmu_tx_t *tx);
669 
670 /*
671  * To register a commit callback, dmu_tx_callback_register() must be called.
672  *
673  * dcb_data is a pointer to caller private data that is passed on as a
674  * callback parameter. The caller is responsible for properly allocating and
675  * freeing it.
676  *
677  * When registering a callback, the transaction must be already created, but
678  * it cannot be committed or aborted. It can be assigned to a txg or not.
679  *
680  * The callback will be called after the transaction has been safely written
681  * to stable storage and will also be called if the dmu_tx is aborted.
682  * If there is any error which prevents the transaction from being committed to
683  * disk, the callback will be called with a value of error != 0.
684  */
685 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
686 
687 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
688     void *dcb_data);
689 
690 /*
691  * Free up the data blocks for a defined range of a file.  If size is
692  * -1, the range from offset to end-of-file is freed.
693  */
694 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
695 	uint64_t size, dmu_tx_t *tx);
696 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
697 	uint64_t size);
698 int dmu_free_long_object(objset_t *os, uint64_t object);
699 
700 /*
701  * Convenience functions.
702  *
703  * Canfail routines will return 0 on success, or an errno if there is a
704  * nonrecoverable I/O error.
705  */
706 #define	DMU_READ_PREFETCH	0 /* prefetch */
707 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
708 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
709 	void *buf, uint32_t flags);
710 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
711 	const void *buf, dmu_tx_t *tx);
712 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
713 	dmu_tx_t *tx);
714 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
715 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
716 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
717     dmu_tx_t *tx);
718 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
719     dmu_tx_t *tx);
720 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
721     uint64_t size, struct page *pp, dmu_tx_t *tx);
722 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
723 void dmu_return_arcbuf(struct arc_buf *buf);
724 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
725     dmu_tx_t *tx);
726 int dmu_xuio_init(struct xuio *uio, int niov);
727 void dmu_xuio_fini(struct xuio *uio);
728 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
729     size_t n);
730 int dmu_xuio_cnt(struct xuio *uio);
731 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
732 void dmu_xuio_clear(struct xuio *uio, int i);
733 void xuio_stat_wbuf_copied();
734 void xuio_stat_wbuf_nocopy();
735 
736 extern int zfs_prefetch_disable;
737 extern int zfs_max_recordsize;
738 
739 /*
740  * Asynchronously try to read in the data.
741  */
742 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
743     uint64_t len);
744 
745 typedef struct dmu_object_info {
746 	/* All sizes are in bytes unless otherwise indicated. */
747 	uint32_t doi_data_block_size;
748 	uint32_t doi_metadata_block_size;
749 	dmu_object_type_t doi_type;
750 	dmu_object_type_t doi_bonus_type;
751 	uint64_t doi_bonus_size;
752 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
753 	uint8_t doi_checksum;
754 	uint8_t doi_compress;
755 	uint8_t doi_nblkptr;
756 	uint8_t doi_pad[4];
757 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
758 	uint64_t doi_max_offset;
759 	uint64_t doi_fill_count;		/* number of non-empty blocks */
760 } dmu_object_info_t;
761 
762 typedef void arc_byteswap_func_t(void *buf, size_t size);
763 
764 typedef struct dmu_object_type_info {
765 	dmu_object_byteswap_t	ot_byteswap;
766 	boolean_t		ot_metadata;
767 	char			*ot_name;
768 } dmu_object_type_info_t;
769 
770 typedef struct dmu_object_byteswap_info {
771 	arc_byteswap_func_t	*ob_func;
772 	char			*ob_name;
773 } dmu_object_byteswap_info_t;
774 
775 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
776 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
777 
778 /*
779  * Get information on a DMU object.
780  *
781  * Return 0 on success or ENOENT if object is not allocated.
782  *
783  * If doi is NULL, just indicates whether the object exists.
784  */
785 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
786 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
787 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
788 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
789 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
790 /*
791  * Like dmu_object_info_from_db, but faster still when you only care about
792  * the size.  This is specifically optimized for zfs_getattr().
793  */
794 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
795     u_longlong_t *nblk512);
796 
797 typedef struct dmu_objset_stats {
798 	uint64_t dds_num_clones; /* number of clones of this */
799 	uint64_t dds_creation_txg;
800 	uint64_t dds_guid;
801 	dmu_objset_type_t dds_type;
802 	uint8_t dds_is_snapshot;
803 	uint8_t dds_inconsistent;
804 	char dds_origin[MAXNAMELEN];
805 } dmu_objset_stats_t;
806 
807 /*
808  * Get stats on a dataset.
809  */
810 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
811 
812 /*
813  * Add entries to the nvlist for all the objset's properties.  See
814  * zfs_prop_table[] and zfs(1m) for details on the properties.
815  */
816 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
817 
818 /*
819  * Get the space usage statistics for statvfs().
820  *
821  * refdbytes is the amount of space "referenced" by this objset.
822  * availbytes is the amount of space available to this objset, taking
823  * into account quotas & reservations, assuming that no other objsets
824  * use the space first.  These values correspond to the 'referenced' and
825  * 'available' properties, described in the zfs(1m) manpage.
826  *
827  * usedobjs and availobjs are the number of objects currently allocated,
828  * and available.
829  */
830 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
831     uint64_t *usedobjsp, uint64_t *availobjsp);
832 
833 /*
834  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
835  * (Contrast with the ds_guid which is a 64-bit ID that will never
836  * change, so there is a small probability that it will collide.)
837  */
838 uint64_t dmu_objset_fsid_guid(objset_t *os);
839 
840 /*
841  * Get the [cm]time for an objset's snapshot dir
842  */
843 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
844 
845 int dmu_objset_is_snapshot(objset_t *os);
846 
847 extern struct spa *dmu_objset_spa(objset_t *os);
848 extern struct zilog *dmu_objset_zil(objset_t *os);
849 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
850 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
851 extern void dmu_objset_name(objset_t *os, char *buf);
852 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
853 extern uint64_t dmu_objset_id(objset_t *os);
854 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
855 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
856 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
857     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
858 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
859     int maxlen, boolean_t *conflict);
860 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
861     uint64_t *idp, uint64_t *offp);
862 
863 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
864     void *bonus, uint64_t *userp, uint64_t *groupp);
865 extern void dmu_objset_register_type(dmu_objset_type_t ost,
866     objset_used_cb_t *cb);
867 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
868 extern void *dmu_objset_get_user(objset_t *os);
869 
870 /*
871  * Return the txg number for the given assigned transaction.
872  */
873 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
874 
875 /*
876  * Synchronous write.
877  * If a parent zio is provided this function initiates a write on the
878  * provided buffer as a child of the parent zio.
879  * In the absence of a parent zio, the write is completed synchronously.
880  * At write completion, blk is filled with the bp of the written block.
881  * Note that while the data covered by this function will be on stable
882  * storage when the write completes this new data does not become a
883  * permanent part of the file until the associated transaction commits.
884  */
885 
886 /*
887  * {zfs,zvol,ztest}_get_done() args
888  */
889 typedef struct zgd {
890 	struct zilog	*zgd_zilog;
891 	struct blkptr	*zgd_bp;
892 	dmu_buf_t	*zgd_db;
893 	struct rl	*zgd_rl;
894 	void		*zgd_private;
895 } zgd_t;
896 
897 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
898 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
899 
900 /*
901  * Find the next hole or data block in file starting at *off
902  * Return found offset in *off. Return ESRCH for end of file.
903  */
904 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
905     uint64_t *off);
906 
907 /*
908  * Check if a DMU object has any dirty blocks. If so, sync out
909  * all pending transaction groups. Otherwise, this function
910  * does not alter DMU state. This could be improved to only sync
911  * out the necessary transaction groups for this particular
912  * object.
913  */
914 int dmu_object_wait_synced(objset_t *os, uint64_t object);
915 
916 /*
917  * Initial setup and final teardown.
918  */
919 extern void dmu_init(void);
920 extern void dmu_fini(void);
921 
922 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
923     uint64_t object, uint64_t offset, int len);
924 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
925     dmu_traverse_cb_t cb, void *arg);
926 
927 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
928     struct vnode *vp, offset_t *offp);
929 
930 /* CRC64 table */
931 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
932 extern uint64_t zfs_crc64_table[256];
933 
934 extern int zfs_mdcomp_disable;
935 
936 #ifdef	__cplusplus
937 }
938 #endif
939 
940 #endif	/* _SYS_DMU_H */
941