xref: /titanic_50/usr/src/uts/common/fs/zfs/sys/dmu.h (revision dc32d872cbeb56532bcea030255db9cd79bac7da)
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  * Copyright 2013 Saso Kiselkov. All rights reserved.
31  */
32 
33 /* Portions Copyright 2010 Robert Milkowski */
34 
35 #ifndef	_SYS_DMU_H
36 #define	_SYS_DMU_H
37 
38 /*
39  * This file describes the interface that the DMU provides for its
40  * consumers.
41  *
42  * The DMU also interacts with the SPA.  That interface is described in
43  * dmu_spa.h.
44  */
45 
46 #include <sys/zfs_context.h>
47 #include <sys/inttypes.h>
48 #include <sys/cred.h>
49 #include <sys/fs/zfs.h>
50 #include <sys/zio_priority.h>
51 
52 #ifdef	__cplusplus
53 extern "C" {
54 #endif
55 
56 struct uio;
57 struct xuio;
58 struct page;
59 struct vnode;
60 struct spa;
61 struct zilog;
62 struct zio;
63 struct blkptr;
64 struct zap_cursor;
65 struct dsl_dataset;
66 struct dsl_pool;
67 struct dnode;
68 struct drr_begin;
69 struct drr_end;
70 struct zbookmark_phys;
71 struct spa;
72 struct nvlist;
73 struct arc_buf;
74 struct zio_prop;
75 struct sa_handle;
76 
77 typedef struct objset objset_t;
78 typedef struct dmu_tx dmu_tx_t;
79 typedef struct dsl_dir dsl_dir_t;
80 
81 typedef enum dmu_object_byteswap {
82 	DMU_BSWAP_UINT8,
83 	DMU_BSWAP_UINT16,
84 	DMU_BSWAP_UINT32,
85 	DMU_BSWAP_UINT64,
86 	DMU_BSWAP_ZAP,
87 	DMU_BSWAP_DNODE,
88 	DMU_BSWAP_OBJSET,
89 	DMU_BSWAP_ZNODE,
90 	DMU_BSWAP_OLDACL,
91 	DMU_BSWAP_ACL,
92 	/*
93 	 * Allocating a new byteswap type number makes the on-disk format
94 	 * incompatible with any other format that uses the same number.
95 	 *
96 	 * Data can usually be structured to work with one of the
97 	 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
98 	 */
99 	DMU_BSWAP_NUMFUNCS
100 } dmu_object_byteswap_t;
101 
102 #define	DMU_OT_NEWTYPE 0x80
103 #define	DMU_OT_METADATA 0x40
104 #define	DMU_OT_BYTESWAP_MASK 0x3f
105 
106 /*
107  * Defines a uint8_t object type. Object types specify if the data
108  * in the object is metadata (boolean) and how to byteswap the data
109  * (dmu_object_byteswap_t).
110  */
111 #define	DMU_OT(byteswap, metadata) \
112 	(DMU_OT_NEWTYPE | \
113 	((metadata) ? DMU_OT_METADATA : 0) | \
114 	((byteswap) & DMU_OT_BYTESWAP_MASK))
115 
116 #define	DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
117 	((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
118 	(ot) < DMU_OT_NUMTYPES)
119 
120 #define	DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
121 	((ot) & DMU_OT_METADATA) : \
122 	dmu_ot[(ot)].ot_metadata)
123 
124 /*
125  * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
126  * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
127  * is repurposed for embedded BPs.
128  */
129 #define	DMU_OT_HAS_FILL(ot) \
130 	((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
131 
132 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
133 	((ot) & DMU_OT_BYTESWAP_MASK) : \
134 	dmu_ot[(ot)].ot_byteswap)
135 
136 typedef enum dmu_object_type {
137 	DMU_OT_NONE,
138 	/* general: */
139 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
140 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
141 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
142 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
143 	DMU_OT_BPOBJ,			/* UINT64 */
144 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
145 	/* spa: */
146 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
147 	DMU_OT_SPACE_MAP,		/* UINT64 */
148 	/* zil: */
149 	DMU_OT_INTENT_LOG,		/* UINT64 */
150 	/* dmu: */
151 	DMU_OT_DNODE,			/* DNODE */
152 	DMU_OT_OBJSET,			/* OBJSET */
153 	/* dsl: */
154 	DMU_OT_DSL_DIR,			/* UINT64 */
155 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
156 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
157 	DMU_OT_DSL_PROPS,		/* ZAP */
158 	DMU_OT_DSL_DATASET,		/* UINT64 */
159 	/* zpl: */
160 	DMU_OT_ZNODE,			/* ZNODE */
161 	DMU_OT_OLDACL,			/* Old ACL */
162 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
163 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
164 	DMU_OT_MASTER_NODE,		/* ZAP */
165 	DMU_OT_UNLINKED_SET,		/* ZAP */
166 	/* zvol: */
167 	DMU_OT_ZVOL,			/* UINT8 */
168 	DMU_OT_ZVOL_PROP,		/* ZAP */
169 	/* other; for testing only! */
170 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
171 	DMU_OT_UINT64_OTHER,		/* UINT64 */
172 	DMU_OT_ZAP_OTHER,		/* ZAP */
173 	/* new object types: */
174 	DMU_OT_ERROR_LOG,		/* ZAP */
175 	DMU_OT_SPA_HISTORY,		/* UINT8 */
176 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
177 	DMU_OT_POOL_PROPS,		/* ZAP */
178 	DMU_OT_DSL_PERMS,		/* ZAP */
179 	DMU_OT_ACL,			/* ACL */
180 	DMU_OT_SYSACL,			/* SYSACL */
181 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
182 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
183 	DMU_OT_NEXT_CLONES,		/* ZAP */
184 	DMU_OT_SCAN_QUEUE,		/* ZAP */
185 	DMU_OT_USERGROUP_USED,		/* ZAP */
186 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
187 	DMU_OT_USERREFS,		/* ZAP */
188 	DMU_OT_DDT_ZAP,			/* ZAP */
189 	DMU_OT_DDT_STATS,		/* ZAP */
190 	DMU_OT_SA,			/* System attr */
191 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
192 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
193 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
194 	DMU_OT_SCAN_XLATE,		/* ZAP */
195 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
196 	DMU_OT_DEADLIST,		/* ZAP */
197 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
198 	DMU_OT_DSL_CLONES,		/* ZAP */
199 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
200 	/*
201 	 * Do not allocate new object types here. Doing so makes the on-disk
202 	 * format incompatible with any other format that uses the same object
203 	 * type number.
204 	 *
205 	 * When creating an object which does not have one of the above types
206 	 * use the DMU_OTN_* type with the correct byteswap and metadata
207 	 * values.
208 	 *
209 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
210 	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
211 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
212 	 * and DMU_OTN_* types).
213 	 */
214 	DMU_OT_NUMTYPES,
215 
216 	/*
217 	 * Names for valid types declared with DMU_OT().
218 	 */
219 	DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
220 	DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
221 	DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
222 	DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
223 	DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
224 	DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
225 	DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
226 	DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
227 	DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
228 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
229 } dmu_object_type_t;
230 
231 typedef enum txg_how {
232 	TXG_WAIT = 1,
233 	TXG_NOWAIT,
234 	TXG_WAITED,
235 } txg_how_t;
236 
237 void byteswap_uint64_array(void *buf, size_t size);
238 void byteswap_uint32_array(void *buf, size_t size);
239 void byteswap_uint16_array(void *buf, size_t size);
240 void byteswap_uint8_array(void *buf, size_t size);
241 void zap_byteswap(void *buf, size_t size);
242 void zfs_oldacl_byteswap(void *buf, size_t size);
243 void zfs_acl_byteswap(void *buf, size_t size);
244 void zfs_znode_byteswap(void *buf, size_t size);
245 
246 #define	DS_FIND_SNAPSHOTS	(1<<0)
247 #define	DS_FIND_CHILDREN	(1<<1)
248 #define	DS_FIND_SERIALIZE	(1<<2)
249 
250 /*
251  * The maximum number of bytes that can be accessed as part of one
252  * operation, including metadata.
253  */
254 #define	DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
255 #define	DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
256 
257 #define	DMU_USERUSED_OBJECT	(-1ULL)
258 #define	DMU_GROUPUSED_OBJECT	(-2ULL)
259 
260 /*
261  * artificial blkids for bonus buffer and spill blocks
262  */
263 #define	DMU_BONUS_BLKID		(-1ULL)
264 #define	DMU_SPILL_BLKID		(-2ULL)
265 /*
266  * Public routines to create, destroy, open, and close objsets.
267  */
268 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
269 int dmu_objset_own(const char *name, dmu_objset_type_t type,
270     boolean_t readonly, void *tag, objset_t **osp);
271 void dmu_objset_rele(objset_t *os, void *tag);
272 void dmu_objset_disown(objset_t *os, void *tag);
273 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
274 
275 void dmu_objset_evict_dbufs(objset_t *os);
276 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
277     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
278 int dmu_objset_clone(const char *name, const char *origin);
279 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
280     struct nvlist *errlist);
281 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
282 int dmu_objset_snapshot_tmp(const char *, const char *, int);
283 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
284     int flags);
285 void dmu_objset_byteswap(void *buf, size_t size);
286 int dsl_dataset_rename_snapshot(const char *fsname,
287     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
288 
289 typedef struct dmu_buf {
290 	uint64_t db_object;		/* object that this buffer is part of */
291 	uint64_t db_offset;		/* byte offset in this object */
292 	uint64_t db_size;		/* size of buffer in bytes */
293 	void *db_data;			/* data in buffer */
294 } dmu_buf_t;
295 
296 /*
297  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
298  */
299 #define	DMU_POOL_DIRECTORY_OBJECT	1
300 #define	DMU_POOL_CONFIG			"config"
301 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
302 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
303 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
304 #define	DMU_POOL_FEATURE_ENABLED_TXG	"feature_enabled_txg"
305 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
306 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
307 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
308 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
309 #define	DMU_POOL_SPARES			"spares"
310 #define	DMU_POOL_DEFLATE		"deflate"
311 #define	DMU_POOL_HISTORY		"history"
312 #define	DMU_POOL_PROPS			"pool_props"
313 #define	DMU_POOL_L2CACHE		"l2cache"
314 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
315 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
316 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
317 #define	DMU_POOL_CREATION_VERSION	"creation_version"
318 #define	DMU_POOL_SCAN			"scan"
319 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
320 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
321 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
322 #define	DMU_POOL_CHECKSUM_SALT		"org.illumos:checksum_salt"
323 
324 /*
325  * Allocate an object from this objset.  The range of object numbers
326  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
327  *
328  * The transaction must be assigned to a txg.  The newly allocated
329  * object will be "held" in the transaction (ie. you can modify the
330  * newly allocated object in this transaction).
331  *
332  * dmu_object_alloc() chooses an object and returns it in *objectp.
333  *
334  * dmu_object_claim() allocates a specific object number.  If that
335  * number is already allocated, it fails and returns EEXIST.
336  *
337  * Return 0 on success, or ENOSPC or EEXIST as specified above.
338  */
339 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
340     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
341 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
342     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
343 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
344     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
345 
346 /*
347  * Free an object from this objset.
348  *
349  * The object's data will be freed as well (ie. you don't need to call
350  * dmu_free(object, 0, -1, tx)).
351  *
352  * The object need not be held in the transaction.
353  *
354  * If there are any holds on this object's buffers (via dmu_buf_hold()),
355  * or tx holds on the object (via dmu_tx_hold_object()), you can not
356  * free it; it fails and returns EBUSY.
357  *
358  * If the object is not allocated, it fails and returns ENOENT.
359  *
360  * Return 0 on success, or EBUSY or ENOENT as specified above.
361  */
362 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
363 
364 /*
365  * Find the next allocated or free object.
366  *
367  * The objectp parameter is in-out.  It will be updated to be the next
368  * object which is allocated.  Ignore objects which have not been
369  * modified since txg.
370  *
371  * XXX Can only be called on a objset with no dirty data.
372  *
373  * Returns 0 on success, or ENOENT if there are no more objects.
374  */
375 int dmu_object_next(objset_t *os, uint64_t *objectp,
376     boolean_t hole, uint64_t txg);
377 
378 /*
379  * Set the data blocksize for an object.
380  *
381  * The object cannot have any blocks allcated beyond the first.  If
382  * the first block is allocated already, the new size must be greater
383  * than the current block size.  If these conditions are not met,
384  * ENOTSUP will be returned.
385  *
386  * Returns 0 on success, or EBUSY if there are any holds on the object
387  * contents, or ENOTSUP as described above.
388  */
389 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
390     int ibs, dmu_tx_t *tx);
391 
392 /*
393  * Set the checksum property on a dnode.  The new checksum algorithm will
394  * apply to all newly written blocks; existing blocks will not be affected.
395  */
396 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
397     dmu_tx_t *tx);
398 
399 /*
400  * Set the compress property on a dnode.  The new compression algorithm will
401  * apply to all newly written blocks; existing blocks will not be affected.
402  */
403 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
404     dmu_tx_t *tx);
405 
406 void
407 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
408     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
409     int compressed_size, int byteorder, dmu_tx_t *tx);
410 
411 /*
412  * Decide how to write a block: checksum, compression, number of copies, etc.
413  */
414 #define	WP_NOFILL	0x1
415 #define	WP_DMU_SYNC	0x2
416 #define	WP_SPILL	0x4
417 
418 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
419     struct zio_prop *zp);
420 /*
421  * The bonus data is accessed more or less like a regular buffer.
422  * You must dmu_bonus_hold() to get the buffer, which will give you a
423  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
424  * data.  As with any normal buffer, you must call dmu_buf_read() to
425  * read db_data, dmu_buf_will_dirty() before modifying it, and the
426  * object must be held in an assigned transaction before calling
427  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
428  * buffer as well.  You must release your hold with dmu_buf_rele().
429  *
430  * Returns ENOENT, EIO, or 0.
431  */
432 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
433 int dmu_bonus_max(void);
434 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
435 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
436 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
437 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
438 
439 /*
440  * Special spill buffer support used by "SA" framework
441  */
442 
443 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
444 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
445     void *tag, dmu_buf_t **dbp);
446 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
447 
448 /*
449  * Obtain the DMU buffer from the specified object which contains the
450  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
451  * that it will remain in memory.  You must release the hold with
452  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
453  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
454  *
455  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
456  * on the returned buffer before reading or writing the buffer's
457  * db_data.  The comments for those routines describe what particular
458  * operations are valid after calling them.
459  *
460  * The object number must be a valid, allocated object number.
461  */
462 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
463     void *tag, dmu_buf_t **, int flags);
464 
465 /*
466  * Add a reference to a dmu buffer that has already been held via
467  * dmu_buf_hold() in the current context.
468  */
469 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
470 
471 /*
472  * Attempt to add a reference to a dmu buffer that is in an unknown state,
473  * using a pointer that may have been invalidated by eviction processing.
474  * The request will succeed if the passed in dbuf still represents the
475  * same os/object/blkid, is ineligible for eviction, and has at least
476  * one hold by a user other than the syncer.
477  */
478 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
479     uint64_t blkid, void *tag);
480 
481 void dmu_buf_rele(dmu_buf_t *db, void *tag);
482 uint64_t dmu_buf_refcount(dmu_buf_t *db);
483 
484 /*
485  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
486  * range of an object.  A pointer to an array of dmu_buf_t*'s is
487  * returned (in *dbpp).
488  *
489  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
490  * frees the array.  The hold on the array of buffers MUST be released
491  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
492  * individually with dmu_buf_rele.
493  */
494 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
495     uint64_t length, boolean_t read, void *tag,
496     int *numbufsp, dmu_buf_t ***dbpp);
497 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
498 
499 typedef void dmu_buf_evict_func_t(void *user_ptr);
500 
501 /*
502  * A DMU buffer user object may be associated with a dbuf for the
503  * duration of its lifetime.  This allows the user of a dbuf (client)
504  * to attach private data to a dbuf (e.g. in-core only data such as a
505  * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
506  * when that dbuf has been evicted.  Clients typically respond to the
507  * eviction notification by freeing their private data, thus ensuring
508  * the same lifetime for both dbuf and private data.
509  *
510  * The mapping from a dmu_buf_user_t to any client private data is the
511  * client's responsibility.  All current consumers of the API with private
512  * data embed a dmu_buf_user_t as the first member of the structure for
513  * their private data.  This allows conversions between the two types
514  * with a simple cast.  Since the DMU buf user API never needs access
515  * to the private data, other strategies can be employed if necessary
516  * or convenient for the client (e.g. using container_of() to do the
517  * conversion for private data that cannot have the dmu_buf_user_t as
518  * its first member).
519  *
520  * Eviction callbacks are executed without the dbuf mutex held or any
521  * other type of mechanism to guarantee that the dbuf is still available.
522  * For this reason, users must assume the dbuf has already been freed
523  * and not reference the dbuf from the callback context.
524  *
525  * Users requesting "immediate eviction" are notified as soon as the dbuf
526  * is only referenced by dirty records (dirties == holds).  Otherwise the
527  * notification occurs after eviction processing for the dbuf begins.
528  */
529 typedef struct dmu_buf_user {
530 	/*
531 	 * Asynchronous user eviction callback state.
532 	 */
533 	taskq_ent_t	dbu_tqent;
534 
535 	/* This instance's eviction function pointer. */
536 	dmu_buf_evict_func_t *dbu_evict_func;
537 #ifdef ZFS_DEBUG
538 	/*
539 	 * Pointer to user's dbuf pointer.  NULL for clients that do
540 	 * not associate a dbuf with their user data.
541 	 *
542 	 * The dbuf pointer is cleared upon eviction so as to catch
543 	 * use-after-evict bugs in clients.
544 	 */
545 	dmu_buf_t **dbu_clear_on_evict_dbufp;
546 #endif
547 } dmu_buf_user_t;
548 
549 /*
550  * Initialize the given dmu_buf_user_t instance with the eviction function
551  * evict_func, to be called when the user is evicted.
552  *
553  * NOTE: This function should only be called once on a given dmu_buf_user_t.
554  *       To allow enforcement of this, dbu must already be zeroed on entry.
555  */
556 #ifdef __lint
557 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
558 extern 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 #else /* __lint */
562 inline void
563 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
564     dmu_buf_t **clear_on_evict_dbufp)
565 {
566 	ASSERT(dbu->dbu_evict_func == NULL);
567 	ASSERT(evict_func != NULL);
568 	dbu->dbu_evict_func = evict_func;
569 #ifdef ZFS_DEBUG
570 	dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
571 #endif
572 }
573 #endif /* __lint */
574 
575 /*
576  * Attach user data to a dbuf and mark it for normal (when the dbuf's
577  * data is cleared or its reference count goes to zero) eviction processing.
578  *
579  * Returns NULL on success, or the existing user if another user currently
580  * owns the buffer.
581  */
582 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
583 
584 /*
585  * Attach user data to a dbuf and mark it for immediate (its dirty and
586  * reference counts are equal) eviction processing.
587  *
588  * Returns NULL on success, or the existing user if another user currently
589  * owns the buffer.
590  */
591 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
592 
593 /*
594  * Replace the current user of a dbuf.
595  *
596  * If given the current user of a dbuf, replaces the dbuf's user with
597  * "new_user" and returns the user data pointer that was replaced.
598  * Otherwise returns the current, and unmodified, dbuf user pointer.
599  */
600 void *dmu_buf_replace_user(dmu_buf_t *db,
601     dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
602 
603 /*
604  * Remove the specified user data for a DMU buffer.
605  *
606  * Returns the user that was removed on success, or the current user if
607  * another user currently owns the buffer.
608  */
609 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
610 
611 /*
612  * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
613  */
614 void *dmu_buf_get_user(dmu_buf_t *db);
615 
616 /* Block until any in-progress dmu buf user evictions complete. */
617 void dmu_buf_user_evict_wait(void);
618 
619 /*
620  * Returns the blkptr associated with this dbuf, or NULL if not set.
621  */
622 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
623 
624 /*
625  * Indicate that you are going to modify the buffer's data (db_data).
626  *
627  * The transaction (tx) must be assigned to a txg (ie. you've called
628  * dmu_tx_assign()).  The buffer's object must be held in the tx
629  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
630  */
631 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
632 
633 /*
634  * Tells if the given dbuf is freeable.
635  */
636 boolean_t dmu_buf_freeable(dmu_buf_t *);
637 
638 /*
639  * You must create a transaction, then hold the objects which you will
640  * (or might) modify as part of this transaction.  Then you must assign
641  * the transaction to a transaction group.  Once the transaction has
642  * been assigned, you can modify buffers which belong to held objects as
643  * part of this transaction.  You can't modify buffers before the
644  * transaction has been assigned; you can't modify buffers which don't
645  * belong to objects which this transaction holds; you can't hold
646  * objects once the transaction has been assigned.  You may hold an
647  * object which you are going to free (with dmu_object_free()), but you
648  * don't have to.
649  *
650  * You can abort the transaction before it has been assigned.
651  *
652  * Note that you may hold buffers (with dmu_buf_hold) at any time,
653  * regardless of transaction state.
654  */
655 
656 #define	DMU_NEW_OBJECT	(-1ULL)
657 #define	DMU_OBJECT_END	(-1ULL)
658 
659 dmu_tx_t *dmu_tx_create(objset_t *os);
660 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
661 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
662     uint64_t len);
663 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
664 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
665 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
666 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
667 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
668 void dmu_tx_abort(dmu_tx_t *tx);
669 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
670 void dmu_tx_wait(dmu_tx_t *tx);
671 void dmu_tx_commit(dmu_tx_t *tx);
672 void dmu_tx_mark_netfree(dmu_tx_t *tx);
673 
674 /*
675  * To register a commit callback, dmu_tx_callback_register() must be called.
676  *
677  * dcb_data is a pointer to caller private data that is passed on as a
678  * callback parameter. The caller is responsible for properly allocating and
679  * freeing it.
680  *
681  * When registering a callback, the transaction must be already created, but
682  * it cannot be committed or aborted. It can be assigned to a txg or not.
683  *
684  * The callback will be called after the transaction has been safely written
685  * to stable storage and will also be called if the dmu_tx is aborted.
686  * If there is any error which prevents the transaction from being committed to
687  * disk, the callback will be called with a value of error != 0.
688  */
689 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
690 
691 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
692     void *dcb_data);
693 
694 /*
695  * Free up the data blocks for a defined range of a file.  If size is
696  * -1, the range from offset to end-of-file is freed.
697  */
698 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
699 	uint64_t size, dmu_tx_t *tx);
700 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
701 	uint64_t size);
702 int dmu_free_long_object(objset_t *os, uint64_t object);
703 
704 /*
705  * Convenience functions.
706  *
707  * Canfail routines will return 0 on success, or an errno if there is a
708  * nonrecoverable I/O error.
709  */
710 #define	DMU_READ_PREFETCH	0 /* prefetch */
711 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
712 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
713 	void *buf, uint32_t flags);
714 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
715 	const void *buf, dmu_tx_t *tx);
716 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
717 	dmu_tx_t *tx);
718 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
719 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
720 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
721     dmu_tx_t *tx);
722 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
723     dmu_tx_t *tx);
724 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
725     uint64_t size, struct page *pp, dmu_tx_t *tx);
726 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
727 void dmu_return_arcbuf(struct arc_buf *buf);
728 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
729     dmu_tx_t *tx);
730 int dmu_xuio_init(struct xuio *uio, int niov);
731 void dmu_xuio_fini(struct xuio *uio);
732 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
733     size_t n);
734 int dmu_xuio_cnt(struct xuio *uio);
735 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
736 void dmu_xuio_clear(struct xuio *uio, int i);
737 void xuio_stat_wbuf_copied();
738 void xuio_stat_wbuf_nocopy();
739 
740 extern boolean_t zfs_prefetch_disable;
741 extern int zfs_max_recordsize;
742 
743 /*
744  * Asynchronously try to read in the data.
745  */
746 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
747     uint64_t len, enum zio_priority pri);
748 
749 typedef struct dmu_object_info {
750 	/* All sizes are in bytes unless otherwise indicated. */
751 	uint32_t doi_data_block_size;
752 	uint32_t doi_metadata_block_size;
753 	dmu_object_type_t doi_type;
754 	dmu_object_type_t doi_bonus_type;
755 	uint64_t doi_bonus_size;
756 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
757 	uint8_t doi_checksum;
758 	uint8_t doi_compress;
759 	uint8_t doi_nblkptr;
760 	uint8_t doi_pad[4];
761 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
762 	uint64_t doi_max_offset;
763 	uint64_t doi_fill_count;		/* number of non-empty blocks */
764 } dmu_object_info_t;
765 
766 typedef void arc_byteswap_func_t(void *buf, size_t size);
767 
768 typedef struct dmu_object_type_info {
769 	dmu_object_byteswap_t	ot_byteswap;
770 	boolean_t		ot_metadata;
771 	char			*ot_name;
772 } dmu_object_type_info_t;
773 
774 typedef struct dmu_object_byteswap_info {
775 	arc_byteswap_func_t	*ob_func;
776 	char			*ob_name;
777 } dmu_object_byteswap_info_t;
778 
779 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
780 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
781 
782 /*
783  * Get information on a DMU object.
784  *
785  * Return 0 on success or ENOENT if object is not allocated.
786  *
787  * If doi is NULL, just indicates whether the object exists.
788  */
789 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
790 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
791 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
792 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
793 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
794 /*
795  * Like dmu_object_info_from_db, but faster still when you only care about
796  * the size.  This is specifically optimized for zfs_getattr().
797  */
798 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
799     u_longlong_t *nblk512);
800 
801 typedef struct dmu_objset_stats {
802 	uint64_t dds_num_clones; /* number of clones of this */
803 	uint64_t dds_creation_txg;
804 	uint64_t dds_guid;
805 	dmu_objset_type_t dds_type;
806 	uint8_t dds_is_snapshot;
807 	uint8_t dds_inconsistent;
808 	char dds_origin[MAXNAMELEN];
809 } dmu_objset_stats_t;
810 
811 /*
812  * Get stats on a dataset.
813  */
814 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
815 
816 /*
817  * Add entries to the nvlist for all the objset's properties.  See
818  * zfs_prop_table[] and zfs(1m) for details on the properties.
819  */
820 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
821 
822 /*
823  * Get the space usage statistics for statvfs().
824  *
825  * refdbytes is the amount of space "referenced" by this objset.
826  * availbytes is the amount of space available to this objset, taking
827  * into account quotas & reservations, assuming that no other objsets
828  * use the space first.  These values correspond to the 'referenced' and
829  * 'available' properties, described in the zfs(1m) manpage.
830  *
831  * usedobjs and availobjs are the number of objects currently allocated,
832  * and available.
833  */
834 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
835     uint64_t *usedobjsp, uint64_t *availobjsp);
836 
837 /*
838  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
839  * (Contrast with the ds_guid which is a 64-bit ID that will never
840  * change, so there is a small probability that it will collide.)
841  */
842 uint64_t dmu_objset_fsid_guid(objset_t *os);
843 
844 /*
845  * Get the [cm]time for an objset's snapshot dir
846  */
847 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
848 
849 int dmu_objset_is_snapshot(objset_t *os);
850 
851 extern struct spa *dmu_objset_spa(objset_t *os);
852 extern struct zilog *dmu_objset_zil(objset_t *os);
853 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
854 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
855 extern void dmu_objset_name(objset_t *os, char *buf);
856 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
857 extern uint64_t dmu_objset_id(objset_t *os);
858 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
859 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
860 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
861     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
862 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
863     int maxlen, boolean_t *conflict);
864 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
865     uint64_t *idp, uint64_t *offp);
866 
867 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
868     void *bonus, uint64_t *userp, uint64_t *groupp);
869 extern void dmu_objset_register_type(dmu_objset_type_t ost,
870     objset_used_cb_t *cb);
871 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
872 extern void *dmu_objset_get_user(objset_t *os);
873 
874 /*
875  * Return the txg number for the given assigned transaction.
876  */
877 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
878 
879 /*
880  * Synchronous write.
881  * If a parent zio is provided this function initiates a write on the
882  * provided buffer as a child of the parent zio.
883  * In the absence of a parent zio, the write is completed synchronously.
884  * At write completion, blk is filled with the bp of the written block.
885  * Note that while the data covered by this function will be on stable
886  * storage when the write completes this new data does not become a
887  * permanent part of the file until the associated transaction commits.
888  */
889 
890 /*
891  * {zfs,zvol,ztest}_get_done() args
892  */
893 typedef struct zgd {
894 	struct zilog	*zgd_zilog;
895 	struct blkptr	*zgd_bp;
896 	dmu_buf_t	*zgd_db;
897 	struct rl	*zgd_rl;
898 	void		*zgd_private;
899 } zgd_t;
900 
901 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
902 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
903 
904 /*
905  * Find the next hole or data block in file starting at *off
906  * Return found offset in *off. Return ESRCH for end of file.
907  */
908 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
909     uint64_t *off);
910 
911 /*
912  * Check if a DMU object has any dirty blocks. If so, sync out
913  * all pending transaction groups. Otherwise, this function
914  * does not alter DMU state. This could be improved to only sync
915  * out the necessary transaction groups for this particular
916  * object.
917  */
918 int dmu_object_wait_synced(objset_t *os, uint64_t object);
919 
920 /*
921  * Initial setup and final teardown.
922  */
923 extern void dmu_init(void);
924 extern void dmu_fini(void);
925 
926 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
927     uint64_t object, uint64_t offset, int len);
928 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
929     dmu_traverse_cb_t cb, void *arg);
930 
931 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
932     struct vnode *vp, offset_t *offp);
933 
934 /* CRC64 table */
935 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
936 extern uint64_t zfs_crc64_table[256];
937 
938 extern int zfs_mdcomp_disable;
939 
940 #ifdef	__cplusplus
941 }
942 #endif
943 
944 #endif	/* _SYS_DMU_H */
945