xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/dmu.h (revision 8671400134a11c848244896ca51a7db4d0f69da4)
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, 2017 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_compress.h>
52 #include <sys/zio_priority.h>
53 
54 #ifdef	__cplusplus
55 extern "C" {
56 #endif
57 
58 struct uio;
59 struct xuio;
60 struct page;
61 struct vnode;
62 struct spa;
63 struct zilog;
64 struct zio;
65 struct blkptr;
66 struct zap_cursor;
67 struct dsl_dataset;
68 struct dsl_pool;
69 struct dnode;
70 struct drr_begin;
71 struct drr_end;
72 struct zbookmark_phys;
73 struct spa;
74 struct nvlist;
75 struct arc_buf;
76 struct zio_prop;
77 struct sa_handle;
78 
79 typedef struct objset objset_t;
80 typedef struct dmu_tx dmu_tx_t;
81 typedef struct dsl_dir dsl_dir_t;
82 typedef struct dnode dnode_t;
83 
84 typedef enum dmu_object_byteswap {
85 	DMU_BSWAP_UINT8,
86 	DMU_BSWAP_UINT16,
87 	DMU_BSWAP_UINT32,
88 	DMU_BSWAP_UINT64,
89 	DMU_BSWAP_ZAP,
90 	DMU_BSWAP_DNODE,
91 	DMU_BSWAP_OBJSET,
92 	DMU_BSWAP_ZNODE,
93 	DMU_BSWAP_OLDACL,
94 	DMU_BSWAP_ACL,
95 	/*
96 	 * Allocating a new byteswap type number makes the on-disk format
97 	 * incompatible with any other format that uses the same number.
98 	 *
99 	 * Data can usually be structured to work with one of the
100 	 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
101 	 */
102 	DMU_BSWAP_NUMFUNCS
103 } dmu_object_byteswap_t;
104 
105 #define	DMU_OT_NEWTYPE 0x80
106 #define	DMU_OT_METADATA 0x40
107 #define	DMU_OT_BYTESWAP_MASK 0x3f
108 
109 /*
110  * Defines a uint8_t object type. Object types specify if the data
111  * in the object is metadata (boolean) and how to byteswap the data
112  * (dmu_object_byteswap_t).
113  */
114 #define	DMU_OT(byteswap, metadata) \
115 	(DMU_OT_NEWTYPE | \
116 	((metadata) ? DMU_OT_METADATA : 0) | \
117 	((byteswap) & DMU_OT_BYTESWAP_MASK))
118 
119 #define	DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
120 	((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
121 	(ot) < DMU_OT_NUMTYPES)
122 
123 #define	DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
124 	((ot) & DMU_OT_METADATA) : \
125 	dmu_ot[(ot)].ot_metadata)
126 
127 /*
128  * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
129  * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
130  * is repurposed for embedded BPs.
131  */
132 #define	DMU_OT_HAS_FILL(ot) \
133 	((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
134 
135 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
136 	((ot) & DMU_OT_BYTESWAP_MASK) : \
137 	dmu_ot[(ot)].ot_byteswap)
138 
139 typedef enum dmu_object_type {
140 	DMU_OT_NONE,
141 	/* general: */
142 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
143 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
144 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
145 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
146 	DMU_OT_BPOBJ,			/* UINT64 */
147 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
148 	/* spa: */
149 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
150 	DMU_OT_SPACE_MAP,		/* UINT64 */
151 	/* zil: */
152 	DMU_OT_INTENT_LOG,		/* UINT64 */
153 	/* dmu: */
154 	DMU_OT_DNODE,			/* DNODE */
155 	DMU_OT_OBJSET,			/* OBJSET */
156 	/* dsl: */
157 	DMU_OT_DSL_DIR,			/* UINT64 */
158 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
159 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
160 	DMU_OT_DSL_PROPS,		/* ZAP */
161 	DMU_OT_DSL_DATASET,		/* UINT64 */
162 	/* zpl: */
163 	DMU_OT_ZNODE,			/* ZNODE */
164 	DMU_OT_OLDACL,			/* Old ACL */
165 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
166 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
167 	DMU_OT_MASTER_NODE,		/* ZAP */
168 	DMU_OT_UNLINKED_SET,		/* ZAP */
169 	/* zvol: */
170 	DMU_OT_ZVOL,			/* UINT8 */
171 	DMU_OT_ZVOL_PROP,		/* ZAP */
172 	/* other; for testing only! */
173 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
174 	DMU_OT_UINT64_OTHER,		/* UINT64 */
175 	DMU_OT_ZAP_OTHER,		/* ZAP */
176 	/* new object types: */
177 	DMU_OT_ERROR_LOG,		/* ZAP */
178 	DMU_OT_SPA_HISTORY,		/* UINT8 */
179 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
180 	DMU_OT_POOL_PROPS,		/* ZAP */
181 	DMU_OT_DSL_PERMS,		/* ZAP */
182 	DMU_OT_ACL,			/* ACL */
183 	DMU_OT_SYSACL,			/* SYSACL */
184 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
185 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
186 	DMU_OT_NEXT_CLONES,		/* ZAP */
187 	DMU_OT_SCAN_QUEUE,		/* ZAP */
188 	DMU_OT_USERGROUP_USED,		/* ZAP */
189 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
190 	DMU_OT_USERREFS,		/* ZAP */
191 	DMU_OT_DDT_ZAP,			/* ZAP */
192 	DMU_OT_DDT_STATS,		/* ZAP */
193 	DMU_OT_SA,			/* System attr */
194 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
195 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
196 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
197 	DMU_OT_SCAN_XLATE,		/* ZAP */
198 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
199 	DMU_OT_DEADLIST,		/* ZAP */
200 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
201 	DMU_OT_DSL_CLONES,		/* ZAP */
202 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
203 	/*
204 	 * Do not allocate new object types here. Doing so makes the on-disk
205 	 * format incompatible with any other format that uses the same object
206 	 * type number.
207 	 *
208 	 * When creating an object which does not have one of the above types
209 	 * use the DMU_OTN_* type with the correct byteswap and metadata
210 	 * values.
211 	 *
212 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
213 	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
214 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
215 	 * and DMU_OTN_* types).
216 	 */
217 	DMU_OT_NUMTYPES,
218 
219 	/*
220 	 * Names for valid types declared with DMU_OT().
221 	 */
222 	DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
223 	DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
224 	DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
225 	DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
226 	DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
227 	DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
228 	DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
229 	DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
230 	DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
231 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
232 } dmu_object_type_t;
233 
234 /*
235  * These flags are intended to be used to specify the "txg_how"
236  * parameter when calling the dmu_tx_assign() function. See the comment
237  * above dmu_tx_assign() for more details on the meaning of these flags.
238  */
239 #define	TXG_NOWAIT	(0ULL)
240 #define	TXG_WAIT	(1ULL<<0)
241 #define	TXG_NOTHROTTLE	(1ULL<<1)
242 
243 void byteswap_uint64_array(void *buf, size_t size);
244 void byteswap_uint32_array(void *buf, size_t size);
245 void byteswap_uint16_array(void *buf, size_t size);
246 void byteswap_uint8_array(void *buf, size_t size);
247 void zap_byteswap(void *buf, size_t size);
248 void zfs_oldacl_byteswap(void *buf, size_t size);
249 void zfs_acl_byteswap(void *buf, size_t size);
250 void zfs_znode_byteswap(void *buf, size_t size);
251 
252 #define	DS_FIND_SNAPSHOTS	(1<<0)
253 #define	DS_FIND_CHILDREN	(1<<1)
254 #define	DS_FIND_SERIALIZE	(1<<2)
255 
256 /*
257  * The maximum number of bytes that can be accessed as part of one
258  * operation, including metadata.
259  */
260 #define	DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
261 #define	DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
262 
263 #define	DMU_USERUSED_OBJECT	(-1ULL)
264 #define	DMU_GROUPUSED_OBJECT	(-2ULL)
265 
266 /*
267  * artificial blkids for bonus buffer and spill blocks
268  */
269 #define	DMU_BONUS_BLKID		(-1ULL)
270 #define	DMU_SPILL_BLKID		(-2ULL)
271 /*
272  * Public routines to create, destroy, open, and close objsets.
273  */
274 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
275 int dmu_objset_own(const char *name, dmu_objset_type_t type,
276     boolean_t readonly, void *tag, objset_t **osp);
277 void dmu_objset_rele(objset_t *os, void *tag);
278 void dmu_objset_disown(objset_t *os, void *tag);
279 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
280 
281 void dmu_objset_evict_dbufs(objset_t *os);
282 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
283     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
284 int dmu_objset_clone(const char *name, const char *origin);
285 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
286     struct nvlist *errlist);
287 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
288 int dmu_objset_snapshot_tmp(const char *, const char *, int);
289 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
290     int flags);
291 void dmu_objset_byteswap(void *buf, size_t size);
292 int dsl_dataset_rename_snapshot(const char *fsname,
293     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
294 int dmu_objset_remap_indirects(const char *fsname);
295 
296 typedef struct dmu_buf {
297 	uint64_t db_object;		/* object that this buffer is part of */
298 	uint64_t db_offset;		/* byte offset in this object */
299 	uint64_t db_size;		/* size of buffer in bytes */
300 	void *db_data;			/* data in buffer */
301 } dmu_buf_t;
302 
303 /*
304  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
305  */
306 #define	DMU_POOL_DIRECTORY_OBJECT	1
307 #define	DMU_POOL_CONFIG			"config"
308 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
309 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
310 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
311 #define	DMU_POOL_FEATURE_ENABLED_TXG	"feature_enabled_txg"
312 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
313 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
314 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
315 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
316 #define	DMU_POOL_SPARES			"spares"
317 #define	DMU_POOL_DEFLATE		"deflate"
318 #define	DMU_POOL_HISTORY		"history"
319 #define	DMU_POOL_PROPS			"pool_props"
320 #define	DMU_POOL_L2CACHE		"l2cache"
321 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
322 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
323 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
324 #define	DMU_POOL_CREATION_VERSION	"creation_version"
325 #define	DMU_POOL_SCAN			"scan"
326 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
327 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
328 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
329 #define	DMU_POOL_CHECKSUM_SALT		"org.illumos:checksum_salt"
330 #define	DMU_POOL_VDEV_ZAP_MAP		"com.delphix:vdev_zap_map"
331 #define	DMU_POOL_REMOVING		"com.delphix:removing"
332 #define	DMU_POOL_OBSOLETE_BPOBJ		"com.delphix:obsolete_bpobj"
333 #define	DMU_POOL_CONDENSING_INDIRECT	"com.delphix:condensing_indirect"
334 #define	DMU_POOL_ZPOOL_CHECKPOINT	"com.delphix:zpool_checkpoint"
335 
336 /*
337  * Allocate an object from this objset.  The range of object numbers
338  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
339  *
340  * The transaction must be assigned to a txg.  The newly allocated
341  * object will be "held" in the transaction (ie. you can modify the
342  * newly allocated object in this transaction).
343  *
344  * dmu_object_alloc() chooses an object and returns it in *objectp.
345  *
346  * dmu_object_claim() allocates a specific object number.  If that
347  * number is already allocated, it fails and returns EEXIST.
348  *
349  * Return 0 on success, or ENOSPC or EEXIST as specified above.
350  */
351 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
352     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
353 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
354     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
355 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
356     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
357 
358 /*
359  * Free an object from this objset.
360  *
361  * The object's data will be freed as well (ie. you don't need to call
362  * dmu_free(object, 0, -1, tx)).
363  *
364  * The object need not be held in the transaction.
365  *
366  * If there are any holds on this object's buffers (via dmu_buf_hold()),
367  * or tx holds on the object (via dmu_tx_hold_object()), you can not
368  * free it; it fails and returns EBUSY.
369  *
370  * If the object is not allocated, it fails and returns ENOENT.
371  *
372  * Return 0 on success, or EBUSY or ENOENT as specified above.
373  */
374 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
375 
376 /*
377  * Find the next allocated or free object.
378  *
379  * The objectp parameter is in-out.  It will be updated to be the next
380  * object which is allocated.  Ignore objects which have not been
381  * modified since txg.
382  *
383  * XXX Can only be called on a objset with no dirty data.
384  *
385  * Returns 0 on success, or ENOENT if there are no more objects.
386  */
387 int dmu_object_next(objset_t *os, uint64_t *objectp,
388     boolean_t hole, uint64_t txg);
389 
390 /*
391  * Set the data blocksize for an object.
392  *
393  * The object cannot have any blocks allcated beyond the first.  If
394  * the first block is allocated already, the new size must be greater
395  * than the current block size.  If these conditions are not met,
396  * ENOTSUP will be returned.
397  *
398  * Returns 0 on success, or EBUSY if there are any holds on the object
399  * contents, or ENOTSUP as described above.
400  */
401 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
402     int ibs, dmu_tx_t *tx);
403 
404 /*
405  * Set the checksum property on a dnode.  The new checksum algorithm will
406  * apply to all newly written blocks; existing blocks will not be affected.
407  */
408 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
409     dmu_tx_t *tx);
410 
411 /*
412  * Set the compress property on a dnode.  The new compression algorithm will
413  * apply to all newly written blocks; existing blocks will not be affected.
414  */
415 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
416     dmu_tx_t *tx);
417 
418 int dmu_object_remap_indirects(objset_t *os, uint64_t object, uint64_t txg);
419 
420 void
421 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
422     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
423     int compressed_size, int byteorder, dmu_tx_t *tx);
424 
425 /*
426  * Decide how to write a block: checksum, compression, number of copies, etc.
427  */
428 #define	WP_NOFILL	0x1
429 #define	WP_DMU_SYNC	0x2
430 #define	WP_SPILL	0x4
431 
432 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
433     struct zio_prop *zp);
434 /*
435  * The bonus data is accessed more or less like a regular buffer.
436  * You must dmu_bonus_hold() to get the buffer, which will give you a
437  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
438  * data.  As with any normal buffer, you must call dmu_buf_will_dirty()
439  * before modifying it, and the
440  * object must be held in an assigned transaction before calling
441  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
442  * buffer as well.  You must release your hold with dmu_buf_rele().
443  *
444  * Returns ENOENT, EIO, or 0.
445  */
446 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
447 int dmu_bonus_max(void);
448 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
449 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
450 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
451 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
452 
453 /*
454  * Special spill buffer support used by "SA" framework
455  */
456 
457 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
458 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
459     void *tag, dmu_buf_t **dbp);
460 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
461 
462 /*
463  * Obtain the DMU buffer from the specified object which contains the
464  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
465  * that it will remain in memory.  You must release the hold with
466  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
467  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
468  *
469  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
470  * on the returned buffer before reading or writing the buffer's
471  * db_data.  The comments for those routines describe what particular
472  * operations are valid after calling them.
473  *
474  * The object number must be a valid, allocated object number.
475  */
476 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
477     void *tag, dmu_buf_t **, int flags);
478 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
479     void *tag, dmu_buf_t **dbp, int flags);
480 
481 /*
482  * Add a reference to a dmu buffer that has already been held via
483  * dmu_buf_hold() in the current context.
484  */
485 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
486 
487 /*
488  * Attempt to add a reference to a dmu buffer that is in an unknown state,
489  * using a pointer that may have been invalidated by eviction processing.
490  * The request will succeed if the passed in dbuf still represents the
491  * same os/object/blkid, is ineligible for eviction, and has at least
492  * one hold by a user other than the syncer.
493  */
494 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
495     uint64_t blkid, void *tag);
496 
497 void dmu_buf_rele(dmu_buf_t *db, void *tag);
498 uint64_t dmu_buf_refcount(dmu_buf_t *db);
499 
500 /*
501  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
502  * range of an object.  A pointer to an array of dmu_buf_t*'s is
503  * returned (in *dbpp).
504  *
505  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
506  * frees the array.  The hold on the array of buffers MUST be released
507  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
508  * individually with dmu_buf_rele.
509  */
510 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
511     uint64_t length, boolean_t read, void *tag,
512     int *numbufsp, dmu_buf_t ***dbpp);
513 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
514 
515 typedef void dmu_buf_evict_func_t(void *user_ptr);
516 
517 /*
518  * A DMU buffer user object may be associated with a dbuf for the
519  * duration of its lifetime.  This allows the user of a dbuf (client)
520  * to attach private data to a dbuf (e.g. in-core only data such as a
521  * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
522  * when that dbuf has been evicted.  Clients typically respond to the
523  * eviction notification by freeing their private data, thus ensuring
524  * the same lifetime for both dbuf and private data.
525  *
526  * The mapping from a dmu_buf_user_t to any client private data is the
527  * client's responsibility.  All current consumers of the API with private
528  * data embed a dmu_buf_user_t as the first member of the structure for
529  * their private data.  This allows conversions between the two types
530  * with a simple cast.  Since the DMU buf user API never needs access
531  * to the private data, other strategies can be employed if necessary
532  * or convenient for the client (e.g. using container_of() to do the
533  * conversion for private data that cannot have the dmu_buf_user_t as
534  * its first member).
535  *
536  * Eviction callbacks are executed without the dbuf mutex held or any
537  * other type of mechanism to guarantee that the dbuf is still available.
538  * For this reason, users must assume the dbuf has already been freed
539  * and not reference the dbuf from the callback context.
540  *
541  * Users requesting "immediate eviction" are notified as soon as the dbuf
542  * is only referenced by dirty records (dirties == holds).  Otherwise the
543  * notification occurs after eviction processing for the dbuf begins.
544  */
545 typedef struct dmu_buf_user {
546 	/*
547 	 * Asynchronous user eviction callback state.
548 	 */
549 	taskq_ent_t	dbu_tqent;
550 
551 	/*
552 	 * This instance's eviction function pointers.
553 	 *
554 	 * dbu_evict_func_sync is called synchronously and then
555 	 * dbu_evict_func_async is executed asynchronously on a taskq.
556 	 */
557 	dmu_buf_evict_func_t *dbu_evict_func_sync;
558 	dmu_buf_evict_func_t *dbu_evict_func_async;
559 #ifdef ZFS_DEBUG
560 	/*
561 	 * Pointer to user's dbuf pointer.  NULL for clients that do
562 	 * not associate a dbuf with their user data.
563 	 *
564 	 * The dbuf pointer is cleared upon eviction so as to catch
565 	 * use-after-evict bugs in clients.
566 	 */
567 	dmu_buf_t **dbu_clear_on_evict_dbufp;
568 #endif
569 } dmu_buf_user_t;
570 
571 /*
572  * Initialize the given dmu_buf_user_t instance with the eviction function
573  * evict_func, to be called when the user is evicted.
574  *
575  * NOTE: This function should only be called once on a given dmu_buf_user_t.
576  *       To allow enforcement of this, dbu must already be zeroed on entry.
577  */
578 /*ARGSUSED*/
579 inline void
580 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
581     dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
582 {
583 	ASSERT(dbu->dbu_evict_func_sync == NULL);
584 	ASSERT(dbu->dbu_evict_func_async == NULL);
585 
586 	/* must have at least one evict func */
587 	IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
588 	dbu->dbu_evict_func_sync = evict_func_sync;
589 	dbu->dbu_evict_func_async = evict_func_async;
590 #ifdef ZFS_DEBUG
591 	dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
592 #endif
593 }
594 
595 /*
596  * Attach user data to a dbuf and mark it for normal (when the dbuf's
597  * data is cleared or its reference count goes to zero) 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(dmu_buf_t *db, dmu_buf_user_t *user);
603 
604 /*
605  * Attach user data to a dbuf and mark it for immediate (its dirty and
606  * reference counts are equal) eviction processing.
607  *
608  * Returns NULL on success, or the existing user if another user currently
609  * owns the buffer.
610  */
611 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
612 
613 /*
614  * Replace the current user of a dbuf.
615  *
616  * If given the current user of a dbuf, replaces the dbuf's user with
617  * "new_user" and returns the user data pointer that was replaced.
618  * Otherwise returns the current, and unmodified, dbuf user pointer.
619  */
620 void *dmu_buf_replace_user(dmu_buf_t *db,
621     dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
622 
623 /*
624  * Remove the specified user data for a DMU buffer.
625  *
626  * Returns the user that was removed on success, or the current user if
627  * another user currently owns the buffer.
628  */
629 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
630 
631 /*
632  * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
633  */
634 void *dmu_buf_get_user(dmu_buf_t *db);
635 
636 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
637 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
638 void dmu_buf_dnode_exit(dmu_buf_t *db);
639 
640 /* Block until any in-progress dmu buf user evictions complete. */
641 void dmu_buf_user_evict_wait(void);
642 
643 /*
644  * Returns the blkptr associated with this dbuf, or NULL if not set.
645  */
646 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
647 
648 /*
649  * Indicate that you are going to modify the buffer's data (db_data).
650  *
651  * The transaction (tx) must be assigned to a txg (ie. you've called
652  * dmu_tx_assign()).  The buffer's object must be held in the tx
653  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
654  */
655 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
656 
657 /*
658  * You must create a transaction, then hold the objects which you will
659  * (or might) modify as part of this transaction.  Then you must assign
660  * the transaction to a transaction group.  Once the transaction has
661  * been assigned, you can modify buffers which belong to held objects as
662  * part of this transaction.  You can't modify buffers before the
663  * transaction has been assigned; you can't modify buffers which don't
664  * belong to objects which this transaction holds; you can't hold
665  * objects once the transaction has been assigned.  You may hold an
666  * object which you are going to free (with dmu_object_free()), but you
667  * don't have to.
668  *
669  * You can abort the transaction before it has been assigned.
670  *
671  * Note that you may hold buffers (with dmu_buf_hold) at any time,
672  * regardless of transaction state.
673  */
674 
675 #define	DMU_NEW_OBJECT	(-1ULL)
676 #define	DMU_OBJECT_END	(-1ULL)
677 
678 dmu_tx_t *dmu_tx_create(objset_t *os);
679 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
680 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
681     int len);
682 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
683     uint64_t len);
684 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
685     uint64_t len);
686 void dmu_tx_hold_remap_l1indirect(dmu_tx_t *tx, uint64_t object);
687 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
688 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
689     const char *name);
690 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
691 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
692 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
693 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
694 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
695 void dmu_tx_abort(dmu_tx_t *tx);
696 int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
697 void dmu_tx_wait(dmu_tx_t *tx);
698 void dmu_tx_commit(dmu_tx_t *tx);
699 void dmu_tx_mark_netfree(dmu_tx_t *tx);
700 
701 /*
702  * To register a commit callback, dmu_tx_callback_register() must be called.
703  *
704  * dcb_data is a pointer to caller private data that is passed on as a
705  * callback parameter. The caller is responsible for properly allocating and
706  * freeing it.
707  *
708  * When registering a callback, the transaction must be already created, but
709  * it cannot be committed or aborted. It can be assigned to a txg or not.
710  *
711  * The callback will be called after the transaction has been safely written
712  * to stable storage and will also be called if the dmu_tx is aborted.
713  * If there is any error which prevents the transaction from being committed to
714  * disk, the callback will be called with a value of error != 0.
715  */
716 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
717 
718 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
719     void *dcb_data);
720 
721 /*
722  * Free up the data blocks for a defined range of a file.  If size is
723  * -1, the range from offset to end-of-file is freed.
724  */
725 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
726 	uint64_t size, dmu_tx_t *tx);
727 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
728 	uint64_t size);
729 int dmu_free_long_object(objset_t *os, uint64_t object);
730 
731 /*
732  * Convenience functions.
733  *
734  * Canfail routines will return 0 on success, or an errno if there is a
735  * nonrecoverable I/O error.
736  */
737 #define	DMU_READ_PREFETCH	0 /* prefetch */
738 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
739 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
740 	void *buf, uint32_t flags);
741 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
742     uint32_t flags);
743 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
744 	const void *buf, dmu_tx_t *tx);
745 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
746     const void *buf, dmu_tx_t *tx);
747 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
748 	dmu_tx_t *tx);
749 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
750 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
751 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
752     dmu_tx_t *tx);
753 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
754     dmu_tx_t *tx);
755 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
756     uint64_t size, struct page *pp, dmu_tx_t *tx);
757 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
758 void dmu_return_arcbuf(struct arc_buf *buf);
759 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
760     dmu_tx_t *tx);
761 int dmu_xuio_init(struct xuio *uio, int niov);
762 void dmu_xuio_fini(struct xuio *uio);
763 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
764     size_t n);
765 int dmu_xuio_cnt(struct xuio *uio);
766 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
767 void dmu_xuio_clear(struct xuio *uio, int i);
768 void xuio_stat_wbuf_copied(void);
769 void xuio_stat_wbuf_nocopy(void);
770 
771 extern boolean_t zfs_prefetch_disable;
772 extern int zfs_max_recordsize;
773 
774 /*
775  * Asynchronously try to read in the data.
776  */
777 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
778     uint64_t len, enum zio_priority pri);
779 
780 typedef struct dmu_object_info {
781 	/* All sizes are in bytes unless otherwise indicated. */
782 	uint32_t doi_data_block_size;
783 	uint32_t doi_metadata_block_size;
784 	dmu_object_type_t doi_type;
785 	dmu_object_type_t doi_bonus_type;
786 	uint64_t doi_bonus_size;
787 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
788 	uint8_t doi_checksum;
789 	uint8_t doi_compress;
790 	uint8_t doi_nblkptr;
791 	uint8_t doi_pad[4];
792 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
793 	uint64_t doi_max_offset;
794 	uint64_t doi_fill_count;		/* number of non-empty blocks */
795 } dmu_object_info_t;
796 
797 typedef void arc_byteswap_func_t(void *buf, size_t size);
798 
799 typedef struct dmu_object_type_info {
800 	dmu_object_byteswap_t	ot_byteswap;
801 	boolean_t		ot_metadata;
802 	char			*ot_name;
803 } dmu_object_type_info_t;
804 
805 typedef struct dmu_object_byteswap_info {
806 	arc_byteswap_func_t	*ob_func;
807 	char			*ob_name;
808 } dmu_object_byteswap_info_t;
809 
810 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
811 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
812 
813 /*
814  * Get information on a DMU object.
815  *
816  * Return 0 on success or ENOENT if object is not allocated.
817  *
818  * If doi is NULL, just indicates whether the object exists.
819  */
820 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
821 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
822 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
823 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
824 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
825 /*
826  * Like dmu_object_info_from_db, but faster still when you only care about
827  * the size.  This is specifically optimized for zfs_getattr().
828  */
829 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
830     u_longlong_t *nblk512);
831 
832 typedef struct dmu_objset_stats {
833 	uint64_t dds_num_clones; /* number of clones of this */
834 	uint64_t dds_creation_txg;
835 	uint64_t dds_guid;
836 	dmu_objset_type_t dds_type;
837 	uint8_t dds_is_snapshot;
838 	uint8_t dds_inconsistent;
839 	char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
840 } dmu_objset_stats_t;
841 
842 /*
843  * Get stats on a dataset.
844  */
845 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
846 
847 /*
848  * Add entries to the nvlist for all the objset's properties.  See
849  * zfs_prop_table[] and zfs(1m) for details on the properties.
850  */
851 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
852 
853 /*
854  * Get the space usage statistics for statvfs().
855  *
856  * refdbytes is the amount of space "referenced" by this objset.
857  * availbytes is the amount of space available to this objset, taking
858  * into account quotas & reservations, assuming that no other objsets
859  * use the space first.  These values correspond to the 'referenced' and
860  * 'available' properties, described in the zfs(1m) manpage.
861  *
862  * usedobjs and availobjs are the number of objects currently allocated,
863  * and available.
864  */
865 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
866     uint64_t *usedobjsp, uint64_t *availobjsp);
867 
868 /*
869  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
870  * (Contrast with the ds_guid which is a 64-bit ID that will never
871  * change, so there is a small probability that it will collide.)
872  */
873 uint64_t dmu_objset_fsid_guid(objset_t *os);
874 
875 /*
876  * Get the [cm]time for an objset's snapshot dir
877  */
878 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
879 
880 int dmu_objset_is_snapshot(objset_t *os);
881 
882 extern struct spa *dmu_objset_spa(objset_t *os);
883 extern struct zilog *dmu_objset_zil(objset_t *os);
884 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
885 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
886 extern void dmu_objset_name(objset_t *os, char *buf);
887 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
888 extern uint64_t dmu_objset_id(objset_t *os);
889 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
890 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
891 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
892     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
893 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
894     int maxlen, boolean_t *conflict);
895 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
896     uint64_t *idp, uint64_t *offp);
897 
898 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
899     void *bonus, uint64_t *userp, uint64_t *groupp);
900 extern void dmu_objset_register_type(dmu_objset_type_t ost,
901     objset_used_cb_t *cb);
902 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
903 extern void *dmu_objset_get_user(objset_t *os);
904 
905 /*
906  * Return the txg number for the given assigned transaction.
907  */
908 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
909 
910 /*
911  * Synchronous write.
912  * If a parent zio is provided this function initiates a write on the
913  * provided buffer as a child of the parent zio.
914  * In the absence of a parent zio, the write is completed synchronously.
915  * At write completion, blk is filled with the bp of the written block.
916  * Note that while the data covered by this function will be on stable
917  * storage when the write completes this new data does not become a
918  * permanent part of the file until the associated transaction commits.
919  */
920 
921 /*
922  * {zfs,zvol,ztest}_get_done() args
923  */
924 typedef struct zgd {
925 	struct lwb	*zgd_lwb;
926 	struct blkptr	*zgd_bp;
927 	dmu_buf_t	*zgd_db;
928 	struct rl	*zgd_rl;
929 	void		*zgd_private;
930 } zgd_t;
931 
932 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
933 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
934 
935 /*
936  * Find the next hole or data block in file starting at *off
937  * Return found offset in *off. Return ESRCH for end of file.
938  */
939 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
940     uint64_t *off);
941 
942 /*
943  * Check if a DMU object has any dirty blocks. If so, sync out
944  * all pending transaction groups. Otherwise, this function
945  * does not alter DMU state. This could be improved to only sync
946  * out the necessary transaction groups for this particular
947  * object.
948  */
949 int dmu_object_wait_synced(objset_t *os, uint64_t object);
950 
951 /*
952  * Initial setup and final teardown.
953  */
954 extern void dmu_init(void);
955 extern void dmu_fini(void);
956 
957 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
958     uint64_t object, uint64_t offset, int len);
959 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
960     dmu_traverse_cb_t cb, void *arg);
961 
962 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
963     struct vnode *vp, offset_t *offp);
964 
965 /* CRC64 table */
966 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
967 extern uint64_t zfs_crc64_table[256];
968 
969 extern int zfs_mdcomp_disable;
970 
971 #ifdef	__cplusplus
972 }
973 #endif
974 
975 #endif	/* _SYS_DMU_H */
976