xref: /titanic_41/usr/src/uts/common/fs/zfs/sys/dmu.h (revision c4cb95767ae6931f185dc5c02edb165c2ff75b10)
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 #include <sys/zio_priority.h>
50 
51 #ifdef	__cplusplus
52 extern "C" {
53 #endif
54 
55 struct uio;
56 struct xuio;
57 struct page;
58 struct vnode;
59 struct spa;
60 struct zilog;
61 struct zio;
62 struct blkptr;
63 struct zap_cursor;
64 struct dsl_dataset;
65 struct dsl_pool;
66 struct dnode;
67 struct drr_begin;
68 struct drr_end;
69 struct zbookmark_phys;
70 struct spa;
71 struct nvlist;
72 struct arc_buf;
73 struct zio_prop;
74 struct sa_handle;
75 
76 typedef struct objset objset_t;
77 typedef struct dmu_tx dmu_tx_t;
78 typedef struct dsl_dir dsl_dir_t;
79 
80 typedef enum dmu_object_byteswap {
81 	DMU_BSWAP_UINT8,
82 	DMU_BSWAP_UINT16,
83 	DMU_BSWAP_UINT32,
84 	DMU_BSWAP_UINT64,
85 	DMU_BSWAP_ZAP,
86 	DMU_BSWAP_DNODE,
87 	DMU_BSWAP_OBJSET,
88 	DMU_BSWAP_ZNODE,
89 	DMU_BSWAP_OLDACL,
90 	DMU_BSWAP_ACL,
91 	/*
92 	 * Allocating a new byteswap type number makes the on-disk format
93 	 * incompatible with any other format that uses the same number.
94 	 *
95 	 * Data can usually be structured to work with one of the
96 	 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
97 	 */
98 	DMU_BSWAP_NUMFUNCS
99 } dmu_object_byteswap_t;
100 
101 #define	DMU_OT_NEWTYPE 0x80
102 #define	DMU_OT_METADATA 0x40
103 #define	DMU_OT_BYTESWAP_MASK 0x3f
104 
105 /*
106  * Defines a uint8_t object type. Object types specify if the data
107  * in the object is metadata (boolean) and how to byteswap the data
108  * (dmu_object_byteswap_t).
109  */
110 #define	DMU_OT(byteswap, metadata) \
111 	(DMU_OT_NEWTYPE | \
112 	((metadata) ? DMU_OT_METADATA : 0) | \
113 	((byteswap) & DMU_OT_BYTESWAP_MASK))
114 
115 #define	DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
116 	((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
117 	(ot) < DMU_OT_NUMTYPES)
118 
119 #define	DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
120 	((ot) & DMU_OT_METADATA) : \
121 	dmu_ot[(ot)].ot_metadata)
122 
123 /*
124  * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
125  * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
126  * is repurposed for embedded BPs.
127  */
128 #define	DMU_OT_HAS_FILL(ot) \
129 	((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
130 
131 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
132 	((ot) & DMU_OT_BYTESWAP_MASK) : \
133 	dmu_ot[(ot)].ot_byteswap)
134 
135 typedef enum dmu_object_type {
136 	DMU_OT_NONE,
137 	/* general: */
138 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
139 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
140 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
141 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
142 	DMU_OT_BPOBJ,			/* UINT64 */
143 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
144 	/* spa: */
145 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
146 	DMU_OT_SPACE_MAP,		/* UINT64 */
147 	/* zil: */
148 	DMU_OT_INTENT_LOG,		/* UINT64 */
149 	/* dmu: */
150 	DMU_OT_DNODE,			/* DNODE */
151 	DMU_OT_OBJSET,			/* OBJSET */
152 	/* dsl: */
153 	DMU_OT_DSL_DIR,			/* UINT64 */
154 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
155 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
156 	DMU_OT_DSL_PROPS,		/* ZAP */
157 	DMU_OT_DSL_DATASET,		/* UINT64 */
158 	/* zpl: */
159 	DMU_OT_ZNODE,			/* ZNODE */
160 	DMU_OT_OLDACL,			/* Old ACL */
161 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
162 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
163 	DMU_OT_MASTER_NODE,		/* ZAP */
164 	DMU_OT_UNLINKED_SET,		/* ZAP */
165 	/* zvol: */
166 	DMU_OT_ZVOL,			/* UINT8 */
167 	DMU_OT_ZVOL_PROP,		/* ZAP */
168 	/* other; for testing only! */
169 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
170 	DMU_OT_UINT64_OTHER,		/* UINT64 */
171 	DMU_OT_ZAP_OTHER,		/* ZAP */
172 	/* new object types: */
173 	DMU_OT_ERROR_LOG,		/* ZAP */
174 	DMU_OT_SPA_HISTORY,		/* UINT8 */
175 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
176 	DMU_OT_POOL_PROPS,		/* ZAP */
177 	DMU_OT_DSL_PERMS,		/* ZAP */
178 	DMU_OT_ACL,			/* ACL */
179 	DMU_OT_SYSACL,			/* SYSACL */
180 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
181 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
182 	DMU_OT_NEXT_CLONES,		/* ZAP */
183 	DMU_OT_SCAN_QUEUE,		/* ZAP */
184 	DMU_OT_USERGROUP_USED,		/* ZAP */
185 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
186 	DMU_OT_USERREFS,		/* ZAP */
187 	DMU_OT_DDT_ZAP,			/* ZAP */
188 	DMU_OT_DDT_STATS,		/* ZAP */
189 	DMU_OT_SA,			/* System attr */
190 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
191 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
192 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
193 	DMU_OT_SCAN_XLATE,		/* ZAP */
194 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
195 	DMU_OT_DEADLIST,		/* ZAP */
196 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
197 	DMU_OT_DSL_CLONES,		/* ZAP */
198 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
199 	/*
200 	 * Do not allocate new object types here. Doing so makes the on-disk
201 	 * format incompatible with any other format that uses the same object
202 	 * type number.
203 	 *
204 	 * When creating an object which does not have one of the above types
205 	 * use the DMU_OTN_* type with the correct byteswap and metadata
206 	 * values.
207 	 *
208 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
209 	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
210 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
211 	 * and DMU_OTN_* types).
212 	 */
213 	DMU_OT_NUMTYPES,
214 
215 	/*
216 	 * Names for valid types declared with DMU_OT().
217 	 */
218 	DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
219 	DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
220 	DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
221 	DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
222 	DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
223 	DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
224 	DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
225 	DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
226 	DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
227 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
228 } dmu_object_type_t;
229 
230 typedef enum txg_how {
231 	TXG_WAIT = 1,
232 	TXG_NOWAIT,
233 	TXG_WAITED,
234 } txg_how_t;
235 
236 void byteswap_uint64_array(void *buf, size_t size);
237 void byteswap_uint32_array(void *buf, size_t size);
238 void byteswap_uint16_array(void *buf, size_t size);
239 void byteswap_uint8_array(void *buf, size_t size);
240 void zap_byteswap(void *buf, size_t size);
241 void zfs_oldacl_byteswap(void *buf, size_t size);
242 void zfs_acl_byteswap(void *buf, size_t size);
243 void zfs_znode_byteswap(void *buf, size_t size);
244 
245 #define	DS_FIND_SNAPSHOTS	(1<<0)
246 #define	DS_FIND_CHILDREN	(1<<1)
247 #define	DS_FIND_SERIALIZE	(1<<2)
248 
249 /*
250  * The maximum number of bytes that can be accessed as part of one
251  * operation, including metadata.
252  */
253 #define	DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
254 #define	DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
255 
256 #define	DMU_USERUSED_OBJECT	(-1ULL)
257 #define	DMU_GROUPUSED_OBJECT	(-2ULL)
258 
259 /*
260  * artificial blkids for bonus buffer and spill blocks
261  */
262 #define	DMU_BONUS_BLKID		(-1ULL)
263 #define	DMU_SPILL_BLKID		(-2ULL)
264 /*
265  * Public routines to create, destroy, open, and close objsets.
266  */
267 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
268 int dmu_objset_own(const char *name, dmu_objset_type_t type,
269     boolean_t readonly, void *tag, objset_t **osp);
270 void dmu_objset_rele(objset_t *os, void *tag);
271 void dmu_objset_disown(objset_t *os, void *tag);
272 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
273 
274 void dmu_objset_evict_dbufs(objset_t *os);
275 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
276     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
277 int dmu_objset_clone(const char *name, const char *origin);
278 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
279     struct nvlist *errlist);
280 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
281 int dmu_objset_snapshot_tmp(const char *, const char *, int);
282 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
283     int flags);
284 void dmu_objset_byteswap(void *buf, size_t size);
285 int dsl_dataset_rename_snapshot(const char *fsname,
286     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
287 
288 typedef struct dmu_buf {
289 	uint64_t db_object;		/* object that this buffer is part of */
290 	uint64_t db_offset;		/* byte offset in this object */
291 	uint64_t db_size;		/* size of buffer in bytes */
292 	void *db_data;			/* data in buffer */
293 } dmu_buf_t;
294 
295 /*
296  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
297  */
298 #define	DMU_POOL_DIRECTORY_OBJECT	1
299 #define	DMU_POOL_CONFIG			"config"
300 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
301 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
302 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
303 #define	DMU_POOL_FEATURE_ENABLED_TXG	"feature_enabled_txg"
304 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
305 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
306 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
307 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
308 #define	DMU_POOL_SPARES			"spares"
309 #define	DMU_POOL_DEFLATE		"deflate"
310 #define	DMU_POOL_HISTORY		"history"
311 #define	DMU_POOL_PROPS			"pool_props"
312 #define	DMU_POOL_L2CACHE		"l2cache"
313 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
314 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
315 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
316 #define	DMU_POOL_CREATION_VERSION	"creation_version"
317 #define	DMU_POOL_SCAN			"scan"
318 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
319 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
320 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
321 
322 /*
323  * Allocate an object from this objset.  The range of object numbers
324  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
325  *
326  * The transaction must be assigned to a txg.  The newly allocated
327  * object will be "held" in the transaction (ie. you can modify the
328  * newly allocated object in this transaction).
329  *
330  * dmu_object_alloc() chooses an object and returns it in *objectp.
331  *
332  * dmu_object_claim() allocates a specific object number.  If that
333  * number is already allocated, it fails and returns EEXIST.
334  *
335  * Return 0 on success, or ENOSPC or EEXIST as specified above.
336  */
337 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
338     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
339 int dmu_object_claim(objset_t *os, uint64_t object, 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_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
342     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
343 
344 /*
345  * Free an object from this objset.
346  *
347  * The object's data will be freed as well (ie. you don't need to call
348  * dmu_free(object, 0, -1, tx)).
349  *
350  * The object need not be held in the transaction.
351  *
352  * If there are any holds on this object's buffers (via dmu_buf_hold()),
353  * or tx holds on the object (via dmu_tx_hold_object()), you can not
354  * free it; it fails and returns EBUSY.
355  *
356  * If the object is not allocated, it fails and returns ENOENT.
357  *
358  * Return 0 on success, or EBUSY or ENOENT as specified above.
359  */
360 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
361 
362 /*
363  * Find the next allocated or free object.
364  *
365  * The objectp parameter is in-out.  It will be updated to be the next
366  * object which is allocated.  Ignore objects which have not been
367  * modified since txg.
368  *
369  * XXX Can only be called on a objset with no dirty data.
370  *
371  * Returns 0 on success, or ENOENT if there are no more objects.
372  */
373 int dmu_object_next(objset_t *os, uint64_t *objectp,
374     boolean_t hole, uint64_t txg);
375 
376 /*
377  * Set the data blocksize for an object.
378  *
379  * The object cannot have any blocks allcated beyond the first.  If
380  * the first block is allocated already, the new size must be greater
381  * than the current block size.  If these conditions are not met,
382  * ENOTSUP will be returned.
383  *
384  * Returns 0 on success, or EBUSY if there are any holds on the object
385  * contents, or ENOTSUP as described above.
386  */
387 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
388     int ibs, dmu_tx_t *tx);
389 
390 /*
391  * Set the checksum property on a dnode.  The new checksum algorithm will
392  * apply to all newly written blocks; existing blocks will not be affected.
393  */
394 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
395     dmu_tx_t *tx);
396 
397 /*
398  * Set the compress property on a dnode.  The new compression algorithm will
399  * apply to all newly written blocks; existing blocks will not be affected.
400  */
401 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
402     dmu_tx_t *tx);
403 
404 void
405 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
406     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
407     int compressed_size, int byteorder, dmu_tx_t *tx);
408 
409 /*
410  * Decide how to write a block: checksum, compression, number of copies, etc.
411  */
412 #define	WP_NOFILL	0x1
413 #define	WP_DMU_SYNC	0x2
414 #define	WP_SPILL	0x4
415 
416 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
417     struct zio_prop *zp);
418 /*
419  * The bonus data is accessed more or less like a regular buffer.
420  * You must dmu_bonus_hold() to get the buffer, which will give you a
421  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
422  * data.  As with any normal buffer, you must call dmu_buf_read() to
423  * read db_data, dmu_buf_will_dirty() before modifying it, and the
424  * object must be held in an assigned transaction before calling
425  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
426  * buffer as well.  You must release your hold with dmu_buf_rele().
427  *
428  * Returns ENOENT, EIO, or 0.
429  */
430 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
431 int dmu_bonus_max(void);
432 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
433 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
434 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
435 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
436 
437 /*
438  * Special spill buffer support used by "SA" framework
439  */
440 
441 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
442 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
443     void *tag, dmu_buf_t **dbp);
444 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
445 
446 /*
447  * Obtain the DMU buffer from the specified object which contains the
448  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
449  * that it will remain in memory.  You must release the hold with
450  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
451  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
452  *
453  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
454  * on the returned buffer before reading or writing the buffer's
455  * db_data.  The comments for those routines describe what particular
456  * operations are valid after calling them.
457  *
458  * The object number must be a valid, allocated object number.
459  */
460 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
461     void *tag, dmu_buf_t **, int flags);
462 
463 /*
464  * Add a reference to a dmu buffer that has already been held via
465  * dmu_buf_hold() in the current context.
466  */
467 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
468 
469 /*
470  * Attempt to add a reference to a dmu buffer that is in an unknown state,
471  * using a pointer that may have been invalidated by eviction processing.
472  * The request will succeed if the passed in dbuf still represents the
473  * same os/object/blkid, is ineligible for eviction, and has at least
474  * one hold by a user other than the syncer.
475  */
476 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
477     uint64_t blkid, void *tag);
478 
479 void dmu_buf_rele(dmu_buf_t *db, void *tag);
480 uint64_t dmu_buf_refcount(dmu_buf_t *db);
481 
482 /*
483  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
484  * range of an object.  A pointer to an array of dmu_buf_t*'s is
485  * returned (in *dbpp).
486  *
487  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
488  * frees the array.  The hold on the array of buffers MUST be released
489  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
490  * individually with dmu_buf_rele.
491  */
492 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
493     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
494 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
495 
496 typedef void dmu_buf_evict_func_t(void *user_ptr);
497 
498 /*
499  * A DMU buffer user object may be associated with a dbuf for the
500  * duration of its lifetime.  This allows the user of a dbuf (client)
501  * to attach private data to a dbuf (e.g. in-core only data such as a
502  * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
503  * when that dbuf has been evicted.  Clients typically respond to the
504  * eviction notification by freeing their private data, thus ensuring
505  * the same lifetime for both dbuf and private data.
506  *
507  * The mapping from a dmu_buf_user_t to any client private data is the
508  * client's responsibility.  All current consumers of the API with private
509  * data embed a dmu_buf_user_t as the first member of the structure for
510  * their private data.  This allows conversions between the two types
511  * with a simple cast.  Since the DMU buf user API never needs access
512  * to the private data, other strategies can be employed if necessary
513  * or convenient for the client (e.g. using container_of() to do the
514  * conversion for private data that cannot have the dmu_buf_user_t as
515  * its first member).
516  *
517  * Eviction callbacks are executed without the dbuf mutex held or any
518  * other type of mechanism to guarantee that the dbuf is still available.
519  * For this reason, users must assume the dbuf has already been freed
520  * and not reference the dbuf from the callback context.
521  *
522  * Users requesting "immediate eviction" are notified as soon as the dbuf
523  * is only referenced by dirty records (dirties == holds).  Otherwise the
524  * notification occurs after eviction processing for the dbuf begins.
525  */
526 typedef struct dmu_buf_user {
527 	/*
528 	 * Asynchronous user eviction callback state.
529 	 */
530 	taskq_ent_t	dbu_tqent;
531 
532 	/*
533 	 * This instance's eviction function pointers.
534 	 *
535 	 * dbu_evict_func_sync is called synchronously and then
536 	 * dbu_evict_func_async is executed asynchronously on a taskq.
537 	 */
538 	dmu_buf_evict_func_t *dbu_evict_func_sync;
539 	dmu_buf_evict_func_t *dbu_evict_func_async;
540 #ifdef ZFS_DEBUG
541 	/*
542 	 * Pointer to user's dbuf pointer.  NULL for clients that do
543 	 * not associate a dbuf with their user data.
544 	 *
545 	 * The dbuf pointer is cleared upon eviction so as to catch
546 	 * use-after-evict bugs in clients.
547 	 */
548 	dmu_buf_t **dbu_clear_on_evict_dbufp;
549 #endif
550 } dmu_buf_user_t;
551 
552 /*
553  * Initialize the given dmu_buf_user_t instance with the eviction function
554  * evict_func, to be called when the user is evicted.
555  *
556  * NOTE: This function should only be called once on a given dmu_buf_user_t.
557  *       To allow enforcement of this, dbu must already be zeroed on entry.
558  */
559 #ifdef __lint
560 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
561 extern void
562 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
563     dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp);
564 #else /* __lint */
565 inline void
dmu_buf_init_user(dmu_buf_user_t * dbu,dmu_buf_evict_func_t * evict_func_sync,dmu_buf_evict_func_t * evict_func_async,dmu_buf_t ** clear_on_evict_dbufp)566 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
567     dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
568 {
569 	ASSERT(dbu->dbu_evict_func_sync == NULL);
570 	ASSERT(dbu->dbu_evict_func_async == NULL);
571 
572 	/* must have at least one evict func */
573 	IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
574 	dbu->dbu_evict_func_sync = evict_func_sync;
575 	dbu->dbu_evict_func_async = evict_func_async;
576 #ifdef ZFS_DEBUG
577 	dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
578 #endif
579 }
580 #endif /* __lint */
581 
582 /*
583  * Attach user data to a dbuf and mark it for normal (when the dbuf's
584  * data is cleared or its reference count goes to zero) eviction processing.
585  *
586  * Returns NULL on success, or the existing user if another user currently
587  * owns the buffer.
588  */
589 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
590 
591 /*
592  * Attach user data to a dbuf and mark it for immediate (its dirty and
593  * reference counts are equal) eviction processing.
594  *
595  * Returns NULL on success, or the existing user if another user currently
596  * owns the buffer.
597  */
598 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
599 
600 /*
601  * Replace the current user of a dbuf.
602  *
603  * If given the current user of a dbuf, replaces the dbuf's user with
604  * "new_user" and returns the user data pointer that was replaced.
605  * Otherwise returns the current, and unmodified, dbuf user pointer.
606  */
607 void *dmu_buf_replace_user(dmu_buf_t *db,
608     dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
609 
610 /*
611  * Remove the specified user data for a DMU buffer.
612  *
613  * Returns the user that was removed on success, or the current user if
614  * another user currently owns the buffer.
615  */
616 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
617 
618 /*
619  * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
620  */
621 void *dmu_buf_get_user(dmu_buf_t *db);
622 
623 /* Block until any in-progress dmu buf user evictions complete. */
624 void dmu_buf_user_evict_wait(void);
625 
626 /*
627  * Returns the blkptr associated with this dbuf, or NULL if not set.
628  */
629 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
630 
631 /*
632  * Indicate that you are going to modify the buffer's data (db_data).
633  *
634  * The transaction (tx) must be assigned to a txg (ie. you've called
635  * dmu_tx_assign()).  The buffer's object must be held in the tx
636  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
637  */
638 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
639 
640 /*
641  * Tells if the given dbuf is freeable.
642  */
643 boolean_t dmu_buf_freeable(dmu_buf_t *);
644 
645 /*
646  * You must create a transaction, then hold the objects which you will
647  * (or might) modify as part of this transaction.  Then you must assign
648  * the transaction to a transaction group.  Once the transaction has
649  * been assigned, you can modify buffers which belong to held objects as
650  * part of this transaction.  You can't modify buffers before the
651  * transaction has been assigned; you can't modify buffers which don't
652  * belong to objects which this transaction holds; you can't hold
653  * objects once the transaction has been assigned.  You may hold an
654  * object which you are going to free (with dmu_object_free()), but you
655  * don't have to.
656  *
657  * You can abort the transaction before it has been assigned.
658  *
659  * Note that you may hold buffers (with dmu_buf_hold) at any time,
660  * regardless of transaction state.
661  */
662 
663 #define	DMU_NEW_OBJECT	(-1ULL)
664 #define	DMU_OBJECT_END	(-1ULL)
665 
666 dmu_tx_t *dmu_tx_create(objset_t *os);
667 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
668 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
669     uint64_t len);
670 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
671 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
672 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
673 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
674 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
675 void dmu_tx_abort(dmu_tx_t *tx);
676 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
677 void dmu_tx_wait(dmu_tx_t *tx);
678 void dmu_tx_commit(dmu_tx_t *tx);
679 void dmu_tx_mark_netfree(dmu_tx_t *tx);
680 
681 /*
682  * To register a commit callback, dmu_tx_callback_register() must be called.
683  *
684  * dcb_data is a pointer to caller private data that is passed on as a
685  * callback parameter. The caller is responsible for properly allocating and
686  * freeing it.
687  *
688  * When registering a callback, the transaction must be already created, but
689  * it cannot be committed or aborted. It can be assigned to a txg or not.
690  *
691  * The callback will be called after the transaction has been safely written
692  * to stable storage and will also be called if the dmu_tx is aborted.
693  * If there is any error which prevents the transaction from being committed to
694  * disk, the callback will be called with a value of error != 0.
695  */
696 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
697 
698 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
699     void *dcb_data);
700 
701 /*
702  * Free up the data blocks for a defined range of a file.  If size is
703  * -1, the range from offset to end-of-file is freed.
704  */
705 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
706 	uint64_t size, dmu_tx_t *tx);
707 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
708 	uint64_t size);
709 int dmu_free_long_object(objset_t *os, uint64_t object);
710 
711 /*
712  * Convenience functions.
713  *
714  * Canfail routines will return 0 on success, or an errno if there is a
715  * nonrecoverable I/O error.
716  */
717 #define	DMU_READ_PREFETCH	0 /* prefetch */
718 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
719 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
720 	void *buf, uint32_t flags);
721 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
722 	const void *buf, dmu_tx_t *tx);
723 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
724 	dmu_tx_t *tx);
725 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
726 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
727 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
728     dmu_tx_t *tx);
729 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
730     dmu_tx_t *tx);
731 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
732     uint64_t size, struct page *pp, dmu_tx_t *tx);
733 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
734 void dmu_return_arcbuf(struct arc_buf *buf);
735 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
736     dmu_tx_t *tx);
737 int dmu_xuio_init(struct xuio *uio, int niov);
738 void dmu_xuio_fini(struct xuio *uio);
739 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
740     size_t n);
741 int dmu_xuio_cnt(struct xuio *uio);
742 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
743 void dmu_xuio_clear(struct xuio *uio, int i);
744 void xuio_stat_wbuf_copied();
745 void xuio_stat_wbuf_nocopy();
746 
747 extern int zfs_prefetch_disable;
748 extern int zfs_max_recordsize;
749 
750 /*
751  * Asynchronously try to read in the data.
752  */
753 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
754     uint64_t len, enum zio_priority pri);
755 
756 typedef struct dmu_object_info {
757 	/* All sizes are in bytes unless otherwise indicated. */
758 	uint32_t doi_data_block_size;
759 	uint32_t doi_metadata_block_size;
760 	dmu_object_type_t doi_type;
761 	dmu_object_type_t doi_bonus_type;
762 	uint64_t doi_bonus_size;
763 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
764 	uint8_t doi_checksum;
765 	uint8_t doi_compress;
766 	uint8_t doi_nblkptr;
767 	uint8_t doi_pad[4];
768 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
769 	uint64_t doi_max_offset;
770 	uint64_t doi_fill_count;		/* number of non-empty blocks */
771 } dmu_object_info_t;
772 
773 typedef void arc_byteswap_func_t(void *buf, size_t size);
774 
775 typedef struct dmu_object_type_info {
776 	dmu_object_byteswap_t	ot_byteswap;
777 	boolean_t		ot_metadata;
778 	char			*ot_name;
779 } dmu_object_type_info_t;
780 
781 typedef struct dmu_object_byteswap_info {
782 	arc_byteswap_func_t	*ob_func;
783 	char			*ob_name;
784 } dmu_object_byteswap_info_t;
785 
786 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
787 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
788 
789 /*
790  * Get information on a DMU object.
791  *
792  * Return 0 on success or ENOENT if object is not allocated.
793  *
794  * If doi is NULL, just indicates whether the object exists.
795  */
796 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
797 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
798 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
799 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
800 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
801 /*
802  * Like dmu_object_info_from_db, but faster still when you only care about
803  * the size.  This is specifically optimized for zfs_getattr().
804  */
805 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
806     u_longlong_t *nblk512);
807 
808 typedef struct dmu_objset_stats {
809 	uint64_t dds_num_clones; /* number of clones of this */
810 	uint64_t dds_creation_txg;
811 	uint64_t dds_guid;
812 	dmu_objset_type_t dds_type;
813 	uint8_t dds_is_snapshot;
814 	uint8_t dds_inconsistent;
815 	char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
816 } dmu_objset_stats_t;
817 
818 /*
819  * Get stats on a dataset.
820  */
821 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
822 
823 /*
824  * Add entries to the nvlist for all the objset's properties.  See
825  * zfs_prop_table[] and zfs(1m) for details on the properties.
826  */
827 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
828 
829 /*
830  * Get the space usage statistics for statvfs().
831  *
832  * refdbytes is the amount of space "referenced" by this objset.
833  * availbytes is the amount of space available to this objset, taking
834  * into account quotas & reservations, assuming that no other objsets
835  * use the space first.  These values correspond to the 'referenced' and
836  * 'available' properties, described in the zfs(1m) manpage.
837  *
838  * usedobjs and availobjs are the number of objects currently allocated,
839  * and available.
840  */
841 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
842     uint64_t *usedobjsp, uint64_t *availobjsp);
843 
844 /*
845  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
846  * (Contrast with the ds_guid which is a 64-bit ID that will never
847  * change, so there is a small probability that it will collide.)
848  */
849 uint64_t dmu_objset_fsid_guid(objset_t *os);
850 
851 /*
852  * Get the [cm]time for an objset's snapshot dir
853  */
854 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
855 
856 int dmu_objset_is_snapshot(objset_t *os);
857 
858 extern struct spa *dmu_objset_spa(objset_t *os);
859 extern struct zilog *dmu_objset_zil(objset_t *os);
860 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
861 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
862 extern void dmu_objset_name(objset_t *os, char *buf);
863 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
864 extern uint64_t dmu_objset_id(objset_t *os);
865 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
866 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
867 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
868     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
869 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
870     int maxlen, boolean_t *conflict);
871 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
872     uint64_t *idp, uint64_t *offp);
873 
874 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
875     void *bonus, uint64_t *userp, uint64_t *groupp);
876 extern void dmu_objset_register_type(dmu_objset_type_t ost,
877     objset_used_cb_t *cb);
878 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
879 extern void *dmu_objset_get_user(objset_t *os);
880 
881 /*
882  * Return the txg number for the given assigned transaction.
883  */
884 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
885 
886 /*
887  * Synchronous write.
888  * If a parent zio is provided this function initiates a write on the
889  * provided buffer as a child of the parent zio.
890  * In the absence of a parent zio, the write is completed synchronously.
891  * At write completion, blk is filled with the bp of the written block.
892  * Note that while the data covered by this function will be on stable
893  * storage when the write completes this new data does not become a
894  * permanent part of the file until the associated transaction commits.
895  */
896 
897 /*
898  * {zfs,zvol,ztest}_get_done() args
899  */
900 typedef struct zgd {
901 	struct zilog	*zgd_zilog;
902 	struct blkptr	*zgd_bp;
903 	dmu_buf_t	*zgd_db;
904 	struct rl	*zgd_rl;
905 	void		*zgd_private;
906 } zgd_t;
907 
908 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
909 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
910 
911 /*
912  * Find the next hole or data block in file starting at *off
913  * Return found offset in *off. Return ESRCH for end of file.
914  */
915 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
916     uint64_t *off);
917 
918 /*
919  * Check if a DMU object has any dirty blocks. If so, sync out
920  * all pending transaction groups. Otherwise, this function
921  * does not alter DMU state. This could be improved to only sync
922  * out the necessary transaction groups for this particular
923  * object.
924  */
925 int dmu_object_wait_synced(objset_t *os, uint64_t object);
926 
927 /*
928  * Initial setup and final teardown.
929  */
930 extern void dmu_init(void);
931 extern void dmu_fini(void);
932 
933 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
934     uint64_t object, uint64_t offset, int len);
935 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
936     dmu_traverse_cb_t cb, void *arg);
937 
938 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
939     struct vnode *vp, offset_t *offp);
940 
941 /* CRC64 table */
942 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
943 extern uint64_t zfs_crc64_table[256];
944 
945 extern int zfs_mdcomp_disable;
946 
947 #ifdef	__cplusplus
948 }
949 #endif
950 
951 #endif	/* _SYS_DMU_H */
952