xref: /titanic_52/usr/src/uts/common/fs/zfs/sys/dmu.h (revision e6546372a86065934a1b311176ecd1b1c543a222)
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  * Copyright (c) 2014 Integros [integros.com]
32  */
33 
34 /* Portions Copyright 2010 Robert Milkowski */
35 
36 #ifndef	_SYS_DMU_H
37 #define	_SYS_DMU_H
38 
39 /*
40  * This file describes the interface that the DMU provides for its
41  * consumers.
42  *
43  * The DMU also interacts with the SPA.  That interface is described in
44  * dmu_spa.h.
45  */
46 
47 #include <sys/zfs_context.h>
48 #include <sys/inttypes.h>
49 #include <sys/cred.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/zio_priority.h>
52 
53 #ifdef	__cplusplus
54 extern "C" {
55 #endif
56 
57 struct uio;
58 struct xuio;
59 struct page;
60 struct vnode;
61 struct spa;
62 struct zilog;
63 struct zio;
64 struct blkptr;
65 struct zap_cursor;
66 struct dsl_dataset;
67 struct dsl_pool;
68 struct dnode;
69 struct drr_begin;
70 struct drr_end;
71 struct zbookmark_phys;
72 struct spa;
73 struct nvlist;
74 struct arc_buf;
75 struct zio_prop;
76 struct sa_handle;
77 
78 typedef struct objset objset_t;
79 typedef struct dmu_tx dmu_tx_t;
80 typedef struct dsl_dir dsl_dir_t;
81 
82 typedef enum dmu_object_byteswap {
83 	DMU_BSWAP_UINT8,
84 	DMU_BSWAP_UINT16,
85 	DMU_BSWAP_UINT32,
86 	DMU_BSWAP_UINT64,
87 	DMU_BSWAP_ZAP,
88 	DMU_BSWAP_DNODE,
89 	DMU_BSWAP_OBJSET,
90 	DMU_BSWAP_ZNODE,
91 	DMU_BSWAP_OLDACL,
92 	DMU_BSWAP_ACL,
93 	/*
94 	 * Allocating a new byteswap type number makes the on-disk format
95 	 * incompatible with any other format that uses the same number.
96 	 *
97 	 * Data can usually be structured to work with one of the
98 	 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
99 	 */
100 	DMU_BSWAP_NUMFUNCS
101 } dmu_object_byteswap_t;
102 
103 #define	DMU_OT_NEWTYPE 0x80
104 #define	DMU_OT_METADATA 0x40
105 #define	DMU_OT_BYTESWAP_MASK 0x3f
106 
107 /*
108  * Defines a uint8_t object type. Object types specify if the data
109  * in the object is metadata (boolean) and how to byteswap the data
110  * (dmu_object_byteswap_t).
111  */
112 #define	DMU_OT(byteswap, metadata) \
113 	(DMU_OT_NEWTYPE | \
114 	((metadata) ? DMU_OT_METADATA : 0) | \
115 	((byteswap) & DMU_OT_BYTESWAP_MASK))
116 
117 #define	DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
118 	((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
119 	(ot) < DMU_OT_NUMTYPES)
120 
121 #define	DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
122 	((ot) & DMU_OT_METADATA) : \
123 	dmu_ot[(ot)].ot_metadata)
124 
125 /*
126  * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
127  * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
128  * is repurposed for embedded BPs.
129  */
130 #define	DMU_OT_HAS_FILL(ot) \
131 	((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
132 
133 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
134 	((ot) & DMU_OT_BYTESWAP_MASK) : \
135 	dmu_ot[(ot)].ot_byteswap)
136 
137 typedef enum dmu_object_type {
138 	DMU_OT_NONE,
139 	/* general: */
140 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
141 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
142 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
143 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
144 	DMU_OT_BPOBJ,			/* UINT64 */
145 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
146 	/* spa: */
147 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
148 	DMU_OT_SPACE_MAP,		/* UINT64 */
149 	/* zil: */
150 	DMU_OT_INTENT_LOG,		/* UINT64 */
151 	/* dmu: */
152 	DMU_OT_DNODE,			/* DNODE */
153 	DMU_OT_OBJSET,			/* OBJSET */
154 	/* dsl: */
155 	DMU_OT_DSL_DIR,			/* UINT64 */
156 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
157 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
158 	DMU_OT_DSL_PROPS,		/* ZAP */
159 	DMU_OT_DSL_DATASET,		/* UINT64 */
160 	/* zpl: */
161 	DMU_OT_ZNODE,			/* ZNODE */
162 	DMU_OT_OLDACL,			/* Old ACL */
163 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
164 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
165 	DMU_OT_MASTER_NODE,		/* ZAP */
166 	DMU_OT_UNLINKED_SET,		/* ZAP */
167 	/* zvol: */
168 	DMU_OT_ZVOL,			/* UINT8 */
169 	DMU_OT_ZVOL_PROP,		/* ZAP */
170 	/* other; for testing only! */
171 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
172 	DMU_OT_UINT64_OTHER,		/* UINT64 */
173 	DMU_OT_ZAP_OTHER,		/* ZAP */
174 	/* new object types: */
175 	DMU_OT_ERROR_LOG,		/* ZAP */
176 	DMU_OT_SPA_HISTORY,		/* UINT8 */
177 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
178 	DMU_OT_POOL_PROPS,		/* ZAP */
179 	DMU_OT_DSL_PERMS,		/* ZAP */
180 	DMU_OT_ACL,			/* ACL */
181 	DMU_OT_SYSACL,			/* SYSACL */
182 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
183 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
184 	DMU_OT_NEXT_CLONES,		/* ZAP */
185 	DMU_OT_SCAN_QUEUE,		/* ZAP */
186 	DMU_OT_USERGROUP_USED,		/* ZAP */
187 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
188 	DMU_OT_USERREFS,		/* ZAP */
189 	DMU_OT_DDT_ZAP,			/* ZAP */
190 	DMU_OT_DDT_STATS,		/* ZAP */
191 	DMU_OT_SA,			/* System attr */
192 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
193 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
194 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
195 	DMU_OT_SCAN_XLATE,		/* ZAP */
196 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
197 	DMU_OT_DEADLIST,		/* ZAP */
198 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
199 	DMU_OT_DSL_CLONES,		/* ZAP */
200 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
201 	/*
202 	 * Do not allocate new object types here. Doing so makes the on-disk
203 	 * format incompatible with any other format that uses the same object
204 	 * type number.
205 	 *
206 	 * When creating an object which does not have one of the above types
207 	 * use the DMU_OTN_* type with the correct byteswap and metadata
208 	 * values.
209 	 *
210 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
211 	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
212 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
213 	 * and DMU_OTN_* types).
214 	 */
215 	DMU_OT_NUMTYPES,
216 
217 	/*
218 	 * Names for valid types declared with DMU_OT().
219 	 */
220 	DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
221 	DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
222 	DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
223 	DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
224 	DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
225 	DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
226 	DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
227 	DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
228 	DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
229 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
230 } dmu_object_type_t;
231 
232 typedef enum txg_how {
233 	TXG_WAIT = 1,
234 	TXG_NOWAIT,
235 	TXG_WAITED,
236 } txg_how_t;
237 
238 void byteswap_uint64_array(void *buf, size_t size);
239 void byteswap_uint32_array(void *buf, size_t size);
240 void byteswap_uint16_array(void *buf, size_t size);
241 void byteswap_uint8_array(void *buf, size_t size);
242 void zap_byteswap(void *buf, size_t size);
243 void zfs_oldacl_byteswap(void *buf, size_t size);
244 void zfs_acl_byteswap(void *buf, size_t size);
245 void zfs_znode_byteswap(void *buf, size_t size);
246 
247 #define	DS_FIND_SNAPSHOTS	(1<<0)
248 #define	DS_FIND_CHILDREN	(1<<1)
249 #define	DS_FIND_SERIALIZE	(1<<2)
250 
251 /*
252  * The maximum number of bytes that can be accessed as part of one
253  * operation, including metadata.
254  */
255 #define	DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
256 #define	DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
257 
258 #define	DMU_USERUSED_OBJECT	(-1ULL)
259 #define	DMU_GROUPUSED_OBJECT	(-2ULL)
260 
261 /*
262  * artificial blkids for bonus buffer and spill blocks
263  */
264 #define	DMU_BONUS_BLKID		(-1ULL)
265 #define	DMU_SPILL_BLKID		(-2ULL)
266 /*
267  * Public routines to create, destroy, open, and close objsets.
268  */
269 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
270 int dmu_objset_own(const char *name, dmu_objset_type_t type,
271     boolean_t readonly, void *tag, objset_t **osp);
272 void dmu_objset_rele(objset_t *os, void *tag);
273 void dmu_objset_disown(objset_t *os, void *tag);
274 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
275 
276 void dmu_objset_evict_dbufs(objset_t *os);
277 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
278     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
279 int dmu_objset_clone(const char *name, const char *origin);
280 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
281     struct nvlist *errlist);
282 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
283 int dmu_objset_snapshot_tmp(const char *, const char *, int);
284 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
285     int flags);
286 void dmu_objset_byteswap(void *buf, size_t size);
287 int dsl_dataset_rename_snapshot(const char *fsname,
288     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
289 
290 typedef struct dmu_buf {
291 	uint64_t db_object;		/* object that this buffer is part of */
292 	uint64_t db_offset;		/* byte offset in this object */
293 	uint64_t db_size;		/* size of buffer in bytes */
294 	void *db_data;			/* data in buffer */
295 } dmu_buf_t;
296 
297 /*
298  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
299  */
300 #define	DMU_POOL_DIRECTORY_OBJECT	1
301 #define	DMU_POOL_CONFIG			"config"
302 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
303 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
304 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
305 #define	DMU_POOL_FEATURE_ENABLED_TXG	"feature_enabled_txg"
306 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
307 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
308 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
309 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
310 #define	DMU_POOL_SPARES			"spares"
311 #define	DMU_POOL_DEFLATE		"deflate"
312 #define	DMU_POOL_HISTORY		"history"
313 #define	DMU_POOL_PROPS			"pool_props"
314 #define	DMU_POOL_L2CACHE		"l2cache"
315 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
316 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
317 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
318 #define	DMU_POOL_CREATION_VERSION	"creation_version"
319 #define	DMU_POOL_SCAN			"scan"
320 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
321 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
322 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
323 #define	DMU_POOL_CHECKSUM_SALT		"org.illumos:checksum_salt"
324 
325 /*
326  * Allocate an object from this objset.  The range of object numbers
327  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
328  *
329  * The transaction must be assigned to a txg.  The newly allocated
330  * object will be "held" in the transaction (ie. you can modify the
331  * newly allocated object in this transaction).
332  *
333  * dmu_object_alloc() chooses an object and returns it in *objectp.
334  *
335  * dmu_object_claim() allocates a specific object number.  If that
336  * number is already allocated, it fails and returns EEXIST.
337  *
338  * Return 0 on success, or ENOSPC or EEXIST as specified above.
339  */
340 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
341     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
342 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
343     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
344 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
345     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
346 
347 /*
348  * Free an object from this objset.
349  *
350  * The object's data will be freed as well (ie. you don't need to call
351  * dmu_free(object, 0, -1, tx)).
352  *
353  * The object need not be held in the transaction.
354  *
355  * If there are any holds on this object's buffers (via dmu_buf_hold()),
356  * or tx holds on the object (via dmu_tx_hold_object()), you can not
357  * free it; it fails and returns EBUSY.
358  *
359  * If the object is not allocated, it fails and returns ENOENT.
360  *
361  * Return 0 on success, or EBUSY or ENOENT as specified above.
362  */
363 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
364 
365 /*
366  * Find the next allocated or free object.
367  *
368  * The objectp parameter is in-out.  It will be updated to be the next
369  * object which is allocated.  Ignore objects which have not been
370  * modified since txg.
371  *
372  * XXX Can only be called on a objset with no dirty data.
373  *
374  * Returns 0 on success, or ENOENT if there are no more objects.
375  */
376 int dmu_object_next(objset_t *os, uint64_t *objectp,
377     boolean_t hole, uint64_t txg);
378 
379 /*
380  * Set the data blocksize for an object.
381  *
382  * The object cannot have any blocks allcated beyond the first.  If
383  * the first block is allocated already, the new size must be greater
384  * than the current block size.  If these conditions are not met,
385  * ENOTSUP will be returned.
386  *
387  * Returns 0 on success, or EBUSY if there are any holds on the object
388  * contents, or ENOTSUP as described above.
389  */
390 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
391     int ibs, dmu_tx_t *tx);
392 
393 /*
394  * Set the checksum property on a dnode.  The new checksum algorithm will
395  * apply to all newly written blocks; existing blocks will not be affected.
396  */
397 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
398     dmu_tx_t *tx);
399 
400 /*
401  * Set the compress property on a dnode.  The new compression algorithm will
402  * apply to all newly written blocks; existing blocks will not be affected.
403  */
404 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
405     dmu_tx_t *tx);
406 
407 void
408 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
409     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
410     int compressed_size, int byteorder, dmu_tx_t *tx);
411 
412 /*
413  * Decide how to write a block: checksum, compression, number of copies, etc.
414  */
415 #define	WP_NOFILL	0x1
416 #define	WP_DMU_SYNC	0x2
417 #define	WP_SPILL	0x4
418 
419 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
420     struct zio_prop *zp);
421 /*
422  * The bonus data is accessed more or less like a regular buffer.
423  * You must dmu_bonus_hold() to get the buffer, which will give you a
424  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
425  * data.  As with any normal buffer, you must call dmu_buf_read() to
426  * read db_data, dmu_buf_will_dirty() before modifying it, and the
427  * object must be held in an assigned transaction before calling
428  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
429  * buffer as well.  You must release your hold with dmu_buf_rele().
430  *
431  * Returns ENOENT, EIO, or 0.
432  */
433 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
434 int dmu_bonus_max(void);
435 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
436 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
437 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
438 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
439 
440 /*
441  * Special spill buffer support used by "SA" framework
442  */
443 
444 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
445 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
446     void *tag, dmu_buf_t **dbp);
447 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
448 
449 /*
450  * Obtain the DMU buffer from the specified object which contains the
451  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
452  * that it will remain in memory.  You must release the hold with
453  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
454  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
455  *
456  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
457  * on the returned buffer before reading or writing the buffer's
458  * db_data.  The comments for those routines describe what particular
459  * operations are valid after calling them.
460  *
461  * The object number must be a valid, allocated object number.
462  */
463 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
464     void *tag, dmu_buf_t **, int flags);
465 
466 /*
467  * Add a reference to a dmu buffer that has already been held via
468  * dmu_buf_hold() in the current context.
469  */
470 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
471 
472 /*
473  * Attempt to add a reference to a dmu buffer that is in an unknown state,
474  * using a pointer that may have been invalidated by eviction processing.
475  * The request will succeed if the passed in dbuf still represents the
476  * same os/object/blkid, is ineligible for eviction, and has at least
477  * one hold by a user other than the syncer.
478  */
479 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
480     uint64_t blkid, void *tag);
481 
482 void dmu_buf_rele(dmu_buf_t *db, void *tag);
483 uint64_t dmu_buf_refcount(dmu_buf_t *db);
484 
485 /*
486  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
487  * range of an object.  A pointer to an array of dmu_buf_t*'s is
488  * returned (in *dbpp).
489  *
490  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
491  * frees the array.  The hold on the array of buffers MUST be released
492  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
493  * individually with dmu_buf_rele.
494  */
495 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
496     uint64_t length, boolean_t read, void *tag,
497     int *numbufsp, dmu_buf_t ***dbpp);
498 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
499 
500 typedef void dmu_buf_evict_func_t(void *user_ptr);
501 
502 /*
503  * A DMU buffer user object may be associated with a dbuf for the
504  * duration of its lifetime.  This allows the user of a dbuf (client)
505  * to attach private data to a dbuf (e.g. in-core only data such as a
506  * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
507  * when that dbuf has been evicted.  Clients typically respond to the
508  * eviction notification by freeing their private data, thus ensuring
509  * the same lifetime for both dbuf and private data.
510  *
511  * The mapping from a dmu_buf_user_t to any client private data is the
512  * client's responsibility.  All current consumers of the API with private
513  * data embed a dmu_buf_user_t as the first member of the structure for
514  * their private data.  This allows conversions between the two types
515  * with a simple cast.  Since the DMU buf user API never needs access
516  * to the private data, other strategies can be employed if necessary
517  * or convenient for the client (e.g. using container_of() to do the
518  * conversion for private data that cannot have the dmu_buf_user_t as
519  * its first member).
520  *
521  * Eviction callbacks are executed without the dbuf mutex held or any
522  * other type of mechanism to guarantee that the dbuf is still available.
523  * For this reason, users must assume the dbuf has already been freed
524  * and not reference the dbuf from the callback context.
525  *
526  * Users requesting "immediate eviction" are notified as soon as the dbuf
527  * is only referenced by dirty records (dirties == holds).  Otherwise the
528  * notification occurs after eviction processing for the dbuf begins.
529  */
530 typedef struct dmu_buf_user {
531 	/*
532 	 * Asynchronous user eviction callback state.
533 	 */
534 	taskq_ent_t	dbu_tqent;
535 
536 	/*
537 	 * This instance's eviction function pointers.
538 	 *
539 	 * dbu_evict_func_sync is called synchronously and then
540 	 * dbu_evict_func_async is executed asynchronously on a taskq.
541 	 */
542 	dmu_buf_evict_func_t *dbu_evict_func_sync;
543 	dmu_buf_evict_func_t *dbu_evict_func_async;
544 #ifdef ZFS_DEBUG
545 	/*
546 	 * Pointer to user's dbuf pointer.  NULL for clients that do
547 	 * not associate a dbuf with their user data.
548 	 *
549 	 * The dbuf pointer is cleared upon eviction so as to catch
550 	 * use-after-evict bugs in clients.
551 	 */
552 	dmu_buf_t **dbu_clear_on_evict_dbufp;
553 #endif
554 } dmu_buf_user_t;
555 
556 /*
557  * Initialize the given dmu_buf_user_t instance with the eviction function
558  * evict_func, to be called when the user is evicted.
559  *
560  * NOTE: This function should only be called once on a given dmu_buf_user_t.
561  *       To allow enforcement of this, dbu must already be zeroed on entry.
562  */
563 #ifdef __lint
564 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
565 extern void
566 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
567     dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp);
568 #else /* __lint */
569 inline void
570 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
571     dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
572 {
573 	ASSERT(dbu->dbu_evict_func_sync == NULL);
574 	ASSERT(dbu->dbu_evict_func_async == NULL);
575 
576 	/* must have at least one evict func */
577 	IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
578 	dbu->dbu_evict_func_sync = evict_func_sync;
579 	dbu->dbu_evict_func_async = evict_func_async;
580 #ifdef ZFS_DEBUG
581 	dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
582 #endif
583 }
584 #endif /* __lint */
585 
586 /*
587  * Attach user data to a dbuf and mark it for normal (when the dbuf's
588  * data is cleared or its reference count goes to zero) eviction processing.
589  *
590  * Returns NULL on success, or the existing user if another user currently
591  * owns the buffer.
592  */
593 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
594 
595 /*
596  * Attach user data to a dbuf and mark it for immediate (its dirty and
597  * reference counts are equal) eviction processing.
598  *
599  * Returns NULL on success, or the existing user if another user currently
600  * owns the buffer.
601  */
602 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
603 
604 /*
605  * Replace the current user of a dbuf.
606  *
607  * If given the current user of a dbuf, replaces the dbuf's user with
608  * "new_user" and returns the user data pointer that was replaced.
609  * Otherwise returns the current, and unmodified, dbuf user pointer.
610  */
611 void *dmu_buf_replace_user(dmu_buf_t *db,
612     dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
613 
614 /*
615  * Remove the specified user data for a DMU buffer.
616  *
617  * Returns the user that was removed on success, or the current user if
618  * another user currently owns the buffer.
619  */
620 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
621 
622 /*
623  * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
624  */
625 void *dmu_buf_get_user(dmu_buf_t *db);
626 
627 /* Block until any in-progress dmu buf user evictions complete. */
628 void dmu_buf_user_evict_wait(void);
629 
630 /*
631  * Returns the blkptr associated with this dbuf, or NULL if not set.
632  */
633 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
634 
635 /*
636  * Indicate that you are going to modify the buffer's data (db_data).
637  *
638  * The transaction (tx) must be assigned to a txg (ie. you've called
639  * dmu_tx_assign()).  The buffer's object must be held in the tx
640  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
641  */
642 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
643 
644 /*
645  * Tells if the given dbuf is freeable.
646  */
647 boolean_t dmu_buf_freeable(dmu_buf_t *);
648 
649 /*
650  * You must create a transaction, then hold the objects which you will
651  * (or might) modify as part of this transaction.  Then you must assign
652  * the transaction to a transaction group.  Once the transaction has
653  * been assigned, you can modify buffers which belong to held objects as
654  * part of this transaction.  You can't modify buffers before the
655  * transaction has been assigned; you can't modify buffers which don't
656  * belong to objects which this transaction holds; you can't hold
657  * objects once the transaction has been assigned.  You may hold an
658  * object which you are going to free (with dmu_object_free()), but you
659  * don't have to.
660  *
661  * You can abort the transaction before it has been assigned.
662  *
663  * Note that you may hold buffers (with dmu_buf_hold) at any time,
664  * regardless of transaction state.
665  */
666 
667 #define	DMU_NEW_OBJECT	(-1ULL)
668 #define	DMU_OBJECT_END	(-1ULL)
669 
670 dmu_tx_t *dmu_tx_create(objset_t *os);
671 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
672 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
673     uint64_t len);
674 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
675 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
676 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
677 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
678 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
679 void dmu_tx_abort(dmu_tx_t *tx);
680 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
681 void dmu_tx_wait(dmu_tx_t *tx);
682 void dmu_tx_commit(dmu_tx_t *tx);
683 void dmu_tx_mark_netfree(dmu_tx_t *tx);
684 
685 /*
686  * To register a commit callback, dmu_tx_callback_register() must be called.
687  *
688  * dcb_data is a pointer to caller private data that is passed on as a
689  * callback parameter. The caller is responsible for properly allocating and
690  * freeing it.
691  *
692  * When registering a callback, the transaction must be already created, but
693  * it cannot be committed or aborted. It can be assigned to a txg or not.
694  *
695  * The callback will be called after the transaction has been safely written
696  * to stable storage and will also be called if the dmu_tx is aborted.
697  * If there is any error which prevents the transaction from being committed to
698  * disk, the callback will be called with a value of error != 0.
699  */
700 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
701 
702 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
703     void *dcb_data);
704 
705 /*
706  * Free up the data blocks for a defined range of a file.  If size is
707  * -1, the range from offset to end-of-file is freed.
708  */
709 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
710 	uint64_t size, dmu_tx_t *tx);
711 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
712 	uint64_t size);
713 int dmu_free_long_object(objset_t *os, uint64_t object);
714 
715 /*
716  * Convenience functions.
717  *
718  * Canfail routines will return 0 on success, or an errno if there is a
719  * nonrecoverable I/O error.
720  */
721 #define	DMU_READ_PREFETCH	0 /* prefetch */
722 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
723 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
724 	void *buf, uint32_t flags);
725 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
726 	const void *buf, dmu_tx_t *tx);
727 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
728 	dmu_tx_t *tx);
729 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
730 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
731 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
732     dmu_tx_t *tx);
733 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
734     dmu_tx_t *tx);
735 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
736     uint64_t size, struct page *pp, dmu_tx_t *tx);
737 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
738 void dmu_return_arcbuf(struct arc_buf *buf);
739 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
740     dmu_tx_t *tx);
741 int dmu_xuio_init(struct xuio *uio, int niov);
742 void dmu_xuio_fini(struct xuio *uio);
743 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
744     size_t n);
745 int dmu_xuio_cnt(struct xuio *uio);
746 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
747 void dmu_xuio_clear(struct xuio *uio, int i);
748 void xuio_stat_wbuf_copied();
749 void xuio_stat_wbuf_nocopy();
750 
751 extern boolean_t zfs_prefetch_disable;
752 extern int zfs_max_recordsize;
753 
754 /*
755  * Asynchronously try to read in the data.
756  */
757 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
758     uint64_t len, enum zio_priority pri);
759 
760 typedef struct dmu_object_info {
761 	/* All sizes are in bytes unless otherwise indicated. */
762 	uint32_t doi_data_block_size;
763 	uint32_t doi_metadata_block_size;
764 	dmu_object_type_t doi_type;
765 	dmu_object_type_t doi_bonus_type;
766 	uint64_t doi_bonus_size;
767 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
768 	uint8_t doi_checksum;
769 	uint8_t doi_compress;
770 	uint8_t doi_nblkptr;
771 	uint8_t doi_pad[4];
772 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
773 	uint64_t doi_max_offset;
774 	uint64_t doi_fill_count;		/* number of non-empty blocks */
775 } dmu_object_info_t;
776 
777 typedef void arc_byteswap_func_t(void *buf, size_t size);
778 
779 typedef struct dmu_object_type_info {
780 	dmu_object_byteswap_t	ot_byteswap;
781 	boolean_t		ot_metadata;
782 	char			*ot_name;
783 } dmu_object_type_info_t;
784 
785 typedef struct dmu_object_byteswap_info {
786 	arc_byteswap_func_t	*ob_func;
787 	char			*ob_name;
788 } dmu_object_byteswap_info_t;
789 
790 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
791 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
792 
793 /*
794  * Get information on a DMU object.
795  *
796  * Return 0 on success or ENOENT if object is not allocated.
797  *
798  * If doi is NULL, just indicates whether the object exists.
799  */
800 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
801 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
802 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
803 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
804 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
805 /*
806  * Like dmu_object_info_from_db, but faster still when you only care about
807  * the size.  This is specifically optimized for zfs_getattr().
808  */
809 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
810     u_longlong_t *nblk512);
811 
812 typedef struct dmu_objset_stats {
813 	uint64_t dds_num_clones; /* number of clones of this */
814 	uint64_t dds_creation_txg;
815 	uint64_t dds_guid;
816 	dmu_objset_type_t dds_type;
817 	uint8_t dds_is_snapshot;
818 	uint8_t dds_inconsistent;
819 	char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
820 } dmu_objset_stats_t;
821 
822 /*
823  * Get stats on a dataset.
824  */
825 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
826 
827 /*
828  * Add entries to the nvlist for all the objset's properties.  See
829  * zfs_prop_table[] and zfs(1m) for details on the properties.
830  */
831 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
832 
833 /*
834  * Get the space usage statistics for statvfs().
835  *
836  * refdbytes is the amount of space "referenced" by this objset.
837  * availbytes is the amount of space available to this objset, taking
838  * into account quotas & reservations, assuming that no other objsets
839  * use the space first.  These values correspond to the 'referenced' and
840  * 'available' properties, described in the zfs(1m) manpage.
841  *
842  * usedobjs and availobjs are the number of objects currently allocated,
843  * and available.
844  */
845 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
846     uint64_t *usedobjsp, uint64_t *availobjsp);
847 
848 /*
849  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
850  * (Contrast with the ds_guid which is a 64-bit ID that will never
851  * change, so there is a small probability that it will collide.)
852  */
853 uint64_t dmu_objset_fsid_guid(objset_t *os);
854 
855 /*
856  * Get the [cm]time for an objset's snapshot dir
857  */
858 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
859 
860 int dmu_objset_is_snapshot(objset_t *os);
861 
862 extern struct spa *dmu_objset_spa(objset_t *os);
863 extern struct zilog *dmu_objset_zil(objset_t *os);
864 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
865 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
866 extern void dmu_objset_name(objset_t *os, char *buf);
867 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
868 extern uint64_t dmu_objset_id(objset_t *os);
869 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
870 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
871 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
872     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
873 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
874     int maxlen, boolean_t *conflict);
875 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
876     uint64_t *idp, uint64_t *offp);
877 
878 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
879     void *bonus, uint64_t *userp, uint64_t *groupp);
880 extern void dmu_objset_register_type(dmu_objset_type_t ost,
881     objset_used_cb_t *cb);
882 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
883 extern void *dmu_objset_get_user(objset_t *os);
884 
885 /*
886  * Return the txg number for the given assigned transaction.
887  */
888 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
889 
890 /*
891  * Synchronous write.
892  * If a parent zio is provided this function initiates a write on the
893  * provided buffer as a child of the parent zio.
894  * In the absence of a parent zio, the write is completed synchronously.
895  * At write completion, blk is filled with the bp of the written block.
896  * Note that while the data covered by this function will be on stable
897  * storage when the write completes this new data does not become a
898  * permanent part of the file until the associated transaction commits.
899  */
900 
901 /*
902  * {zfs,zvol,ztest}_get_done() args
903  */
904 typedef struct zgd {
905 	struct zilog	*zgd_zilog;
906 	struct blkptr	*zgd_bp;
907 	dmu_buf_t	*zgd_db;
908 	struct rl	*zgd_rl;
909 	void		*zgd_private;
910 } zgd_t;
911 
912 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
913 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
914 
915 /*
916  * Find the next hole or data block in file starting at *off
917  * Return found offset in *off. Return ESRCH for end of file.
918  */
919 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
920     uint64_t *off);
921 
922 /*
923  * Check if a DMU object has any dirty blocks. If so, sync out
924  * all pending transaction groups. Otherwise, this function
925  * does not alter DMU state. This could be improved to only sync
926  * out the necessary transaction groups for this particular
927  * object.
928  */
929 int dmu_object_wait_synced(objset_t *os, uint64_t object);
930 
931 /*
932  * Initial setup and final teardown.
933  */
934 extern void dmu_init(void);
935 extern void dmu_fini(void);
936 
937 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
938     uint64_t object, uint64_t offset, int len);
939 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
940     dmu_traverse_cb_t cb, void *arg);
941 
942 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
943     struct vnode *vp, offset_t *offp);
944 
945 /* CRC64 table */
946 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
947 extern uint64_t zfs_crc64_table[256];
948 
949 extern int zfs_mdcomp_disable;
950 
951 #ifdef	__cplusplus
952 }
953 #endif
954 
955 #endif	/* _SYS_DMU_H */
956