xref: /titanic_41/usr/src/uts/common/fs/zfs/sys/dmu.h (revision 7bed0dedcabfbec8beb9b54e3e19eee9692f2722)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
26  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
27  * Copyright 2013 DEY Storage Systems, Inc.
28  * Copyright 2014 HybridCluster. All rights reserved.
29  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30  */
31 
32 /* Portions Copyright 2010 Robert Milkowski */
33 
34 #ifndef	_SYS_DMU_H
35 #define	_SYS_DMU_H
36 
37 /*
38  * This file describes the interface that the DMU provides for its
39  * consumers.
40  *
41  * The DMU also interacts with the SPA.  That interface is described in
42  * dmu_spa.h.
43  */
44 
45 #include <sys/zfs_context.h>
46 #include <sys/inttypes.h>
47 #include <sys/cred.h>
48 #include <sys/fs/zfs.h>
49 
50 #ifdef	__cplusplus
51 extern "C" {
52 #endif
53 
54 struct uio;
55 struct xuio;
56 struct page;
57 struct vnode;
58 struct spa;
59 struct zilog;
60 struct zio;
61 struct blkptr;
62 struct zap_cursor;
63 struct dsl_dataset;
64 struct dsl_pool;
65 struct dnode;
66 struct drr_begin;
67 struct drr_end;
68 struct zbookmark_phys;
69 struct spa;
70 struct nvlist;
71 struct arc_buf;
72 struct zio_prop;
73 struct sa_handle;
74 
75 typedef struct objset objset_t;
76 typedef struct dmu_tx dmu_tx_t;
77 typedef struct dsl_dir dsl_dir_t;
78 
79 typedef enum dmu_object_byteswap {
80 	DMU_BSWAP_UINT8,
81 	DMU_BSWAP_UINT16,
82 	DMU_BSWAP_UINT32,
83 	DMU_BSWAP_UINT64,
84 	DMU_BSWAP_ZAP,
85 	DMU_BSWAP_DNODE,
86 	DMU_BSWAP_OBJSET,
87 	DMU_BSWAP_ZNODE,
88 	DMU_BSWAP_OLDACL,
89 	DMU_BSWAP_ACL,
90 	/*
91 	 * Allocating a new byteswap type number makes the on-disk format
92 	 * incompatible with any other format that uses the same number.
93 	 *
94 	 * Data can usually be structured to work with one of the
95 	 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
96 	 */
97 	DMU_BSWAP_NUMFUNCS
98 } dmu_object_byteswap_t;
99 
100 #define	DMU_OT_NEWTYPE 0x80
101 #define	DMU_OT_METADATA 0x40
102 #define	DMU_OT_BYTESWAP_MASK 0x3f
103 
104 /*
105  * Defines a uint8_t object type. Object types specify if the data
106  * in the object is metadata (boolean) and how to byteswap the data
107  * (dmu_object_byteswap_t).
108  */
109 #define	DMU_OT(byteswap, metadata) \
110 	(DMU_OT_NEWTYPE | \
111 	((metadata) ? DMU_OT_METADATA : 0) | \
112 	((byteswap) & DMU_OT_BYTESWAP_MASK))
113 
114 #define	DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
115 	((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
116 	(ot) < DMU_OT_NUMTYPES)
117 
118 #define	DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
119 	((ot) & DMU_OT_METADATA) : \
120 	dmu_ot[(ot)].ot_metadata)
121 
122 /*
123  * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
124  * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
125  * is repurposed for embedded BPs.
126  */
127 #define	DMU_OT_HAS_FILL(ot) \
128 	((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
129 
130 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
131 	((ot) & DMU_OT_BYTESWAP_MASK) : \
132 	dmu_ot[(ot)].ot_byteswap)
133 
134 typedef enum dmu_object_type {
135 	DMU_OT_NONE,
136 	/* general: */
137 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
138 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
139 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
140 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
141 	DMU_OT_BPOBJ,			/* UINT64 */
142 	DMU_OT_BPOBJ_HDR,		/* UINT64 */
143 	/* spa: */
144 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
145 	DMU_OT_SPACE_MAP,		/* UINT64 */
146 	/* zil: */
147 	DMU_OT_INTENT_LOG,		/* UINT64 */
148 	/* dmu: */
149 	DMU_OT_DNODE,			/* DNODE */
150 	DMU_OT_OBJSET,			/* OBJSET */
151 	/* dsl: */
152 	DMU_OT_DSL_DIR,			/* UINT64 */
153 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
154 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
155 	DMU_OT_DSL_PROPS,		/* ZAP */
156 	DMU_OT_DSL_DATASET,		/* UINT64 */
157 	/* zpl: */
158 	DMU_OT_ZNODE,			/* ZNODE */
159 	DMU_OT_OLDACL,			/* Old ACL */
160 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
161 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
162 	DMU_OT_MASTER_NODE,		/* ZAP */
163 	DMU_OT_UNLINKED_SET,		/* ZAP */
164 	/* zvol: */
165 	DMU_OT_ZVOL,			/* UINT8 */
166 	DMU_OT_ZVOL_PROP,		/* ZAP */
167 	/* other; for testing only! */
168 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
169 	DMU_OT_UINT64_OTHER,		/* UINT64 */
170 	DMU_OT_ZAP_OTHER,		/* ZAP */
171 	/* new object types: */
172 	DMU_OT_ERROR_LOG,		/* ZAP */
173 	DMU_OT_SPA_HISTORY,		/* UINT8 */
174 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
175 	DMU_OT_POOL_PROPS,		/* ZAP */
176 	DMU_OT_DSL_PERMS,		/* ZAP */
177 	DMU_OT_ACL,			/* ACL */
178 	DMU_OT_SYSACL,			/* SYSACL */
179 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
180 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
181 	DMU_OT_NEXT_CLONES,		/* ZAP */
182 	DMU_OT_SCAN_QUEUE,		/* ZAP */
183 	DMU_OT_USERGROUP_USED,		/* ZAP */
184 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
185 	DMU_OT_USERREFS,		/* ZAP */
186 	DMU_OT_DDT_ZAP,			/* ZAP */
187 	DMU_OT_DDT_STATS,		/* ZAP */
188 	DMU_OT_SA,			/* System attr */
189 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
190 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
191 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
192 	DMU_OT_SCAN_XLATE,		/* ZAP */
193 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
194 	DMU_OT_DEADLIST,		/* ZAP */
195 	DMU_OT_DEADLIST_HDR,		/* UINT64 */
196 	DMU_OT_DSL_CLONES,		/* ZAP */
197 	DMU_OT_BPOBJ_SUBOBJ,		/* UINT64 */
198 	/*
199 	 * Do not allocate new object types here. Doing so makes the on-disk
200 	 * format incompatible with any other format that uses the same object
201 	 * type number.
202 	 *
203 	 * When creating an object which does not have one of the above types
204 	 * use the DMU_OTN_* type with the correct byteswap and metadata
205 	 * values.
206 	 *
207 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
208 	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
209 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
210 	 * and DMU_OTN_* types).
211 	 */
212 	DMU_OT_NUMTYPES,
213 
214 	/*
215 	 * Names for valid types declared with DMU_OT().
216 	 */
217 	DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
218 	DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
219 	DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
220 	DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
221 	DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
222 	DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
223 	DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
224 	DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
225 	DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
226 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
227 } dmu_object_type_t;
228 
229 typedef enum txg_how {
230 	TXG_WAIT = 1,
231 	TXG_NOWAIT,
232 	TXG_WAITED,
233 } txg_how_t;
234 
235 void byteswap_uint64_array(void *buf, size_t size);
236 void byteswap_uint32_array(void *buf, size_t size);
237 void byteswap_uint16_array(void *buf, size_t size);
238 void byteswap_uint8_array(void *buf, size_t size);
239 void zap_byteswap(void *buf, size_t size);
240 void zfs_oldacl_byteswap(void *buf, size_t size);
241 void zfs_acl_byteswap(void *buf, size_t size);
242 void zfs_znode_byteswap(void *buf, size_t size);
243 
244 #define	DS_FIND_SNAPSHOTS	(1<<0)
245 #define	DS_FIND_CHILDREN	(1<<1)
246 
247 /*
248  * The maximum number of bytes that can be accessed as part of one
249  * operation, including metadata.
250  */
251 #define	DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
252 #define	DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
253 
254 #define	DMU_USERUSED_OBJECT	(-1ULL)
255 #define	DMU_GROUPUSED_OBJECT	(-2ULL)
256 
257 /*
258  * artificial blkids for bonus buffer and spill blocks
259  */
260 #define	DMU_BONUS_BLKID		(-1ULL)
261 #define	DMU_SPILL_BLKID		(-2ULL)
262 /*
263  * Public routines to create, destroy, open, and close objsets.
264  */
265 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
266 int dmu_objset_own(const char *name, dmu_objset_type_t type,
267     boolean_t readonly, void *tag, objset_t **osp);
268 void dmu_objset_rele(objset_t *os, void *tag);
269 void dmu_objset_disown(objset_t *os, void *tag);
270 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
271 
272 void dmu_objset_evict_dbufs(objset_t *os);
273 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
274     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
275 int dmu_objset_clone(const char *name, const char *origin);
276 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
277     struct nvlist *errlist);
278 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
279 int dmu_objset_snapshot_tmp(const char *, const char *, int);
280 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
281     int flags);
282 void dmu_objset_byteswap(void *buf, size_t size);
283 int dsl_dataset_rename_snapshot(const char *fsname,
284     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
285 
286 typedef struct dmu_buf {
287 	uint64_t db_object;		/* object that this buffer is part of */
288 	uint64_t db_offset;		/* byte offset in this object */
289 	uint64_t db_size;		/* size of buffer in bytes */
290 	void *db_data;			/* data in buffer */
291 } dmu_buf_t;
292 
293 /*
294  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
295  */
296 #define	DMU_POOL_DIRECTORY_OBJECT	1
297 #define	DMU_POOL_CONFIG			"config"
298 #define	DMU_POOL_FEATURES_FOR_WRITE	"features_for_write"
299 #define	DMU_POOL_FEATURES_FOR_READ	"features_for_read"
300 #define	DMU_POOL_FEATURE_DESCRIPTIONS	"feature_descriptions"
301 #define	DMU_POOL_FEATURE_ENABLED_TXG	"feature_enabled_txg"
302 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
303 #define	DMU_POOL_SYNC_BPOBJ		"sync_bplist"
304 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
305 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
306 #define	DMU_POOL_SPARES			"spares"
307 #define	DMU_POOL_DEFLATE		"deflate"
308 #define	DMU_POOL_HISTORY		"history"
309 #define	DMU_POOL_PROPS			"pool_props"
310 #define	DMU_POOL_L2CACHE		"l2cache"
311 #define	DMU_POOL_TMP_USERREFS		"tmp_userrefs"
312 #define	DMU_POOL_DDT			"DDT-%s-%s-%s"
313 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
314 #define	DMU_POOL_CREATION_VERSION	"creation_version"
315 #define	DMU_POOL_SCAN			"scan"
316 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
317 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
318 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
319 
320 /*
321  * Allocate an object from this objset.  The range of object numbers
322  * available is (0, DN_MAX_OBJECT).  Object 0 is the meta-dnode.
323  *
324  * The transaction must be assigned to a txg.  The newly allocated
325  * object will be "held" in the transaction (ie. you can modify the
326  * newly allocated object in this transaction).
327  *
328  * dmu_object_alloc() chooses an object and returns it in *objectp.
329  *
330  * dmu_object_claim() allocates a specific object number.  If that
331  * number is already allocated, it fails and returns EEXIST.
332  *
333  * Return 0 on success, or ENOSPC or EEXIST as specified above.
334  */
335 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
336     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
337 int dmu_object_claim(objset_t *os, uint64_t object, 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_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
340     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
341 
342 /*
343  * Free an object from this objset.
344  *
345  * The object's data will be freed as well (ie. you don't need to call
346  * dmu_free(object, 0, -1, tx)).
347  *
348  * The object need not be held in the transaction.
349  *
350  * If there are any holds on this object's buffers (via dmu_buf_hold()),
351  * or tx holds on the object (via dmu_tx_hold_object()), you can not
352  * free it; it fails and returns EBUSY.
353  *
354  * If the object is not allocated, it fails and returns ENOENT.
355  *
356  * Return 0 on success, or EBUSY or ENOENT as specified above.
357  */
358 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
359 
360 /*
361  * Find the next allocated or free object.
362  *
363  * The objectp parameter is in-out.  It will be updated to be the next
364  * object which is allocated.  Ignore objects which have not been
365  * modified since txg.
366  *
367  * XXX Can only be called on a objset with no dirty data.
368  *
369  * Returns 0 on success, or ENOENT if there are no more objects.
370  */
371 int dmu_object_next(objset_t *os, uint64_t *objectp,
372     boolean_t hole, uint64_t txg);
373 
374 /*
375  * Set the data blocksize for an object.
376  *
377  * The object cannot have any blocks allcated beyond the first.  If
378  * the first block is allocated already, the new size must be greater
379  * than the current block size.  If these conditions are not met,
380  * ENOTSUP will be returned.
381  *
382  * Returns 0 on success, or EBUSY if there are any holds on the object
383  * contents, or ENOTSUP as described above.
384  */
385 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
386     int ibs, dmu_tx_t *tx);
387 
388 /*
389  * Set the checksum property on a dnode.  The new checksum algorithm will
390  * apply to all newly written blocks; existing blocks will not be affected.
391  */
392 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
393     dmu_tx_t *tx);
394 
395 /*
396  * Set the compress property on a dnode.  The new compression algorithm will
397  * apply to all newly written blocks; existing blocks will not be affected.
398  */
399 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
400     dmu_tx_t *tx);
401 
402 void
403 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
404     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
405     int compressed_size, int byteorder, dmu_tx_t *tx);
406 
407 /*
408  * Decide how to write a block: checksum, compression, number of copies, etc.
409  */
410 #define	WP_NOFILL	0x1
411 #define	WP_DMU_SYNC	0x2
412 #define	WP_SPILL	0x4
413 
414 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
415     struct zio_prop *zp);
416 /*
417  * The bonus data is accessed more or less like a regular buffer.
418  * You must dmu_bonus_hold() to get the buffer, which will give you a
419  * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
420  * data.  As with any normal buffer, you must call dmu_buf_read() to
421  * read db_data, dmu_buf_will_dirty() before modifying it, and the
422  * object must be held in an assigned transaction before calling
423  * dmu_buf_will_dirty.  You may use dmu_buf_set_user() on the bonus
424  * buffer as well.  You must release your hold with dmu_buf_rele().
425  *
426  * Returns ENOENT, EIO, or 0.
427  */
428 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
429 int dmu_bonus_max(void);
430 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
431 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
432 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
433 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
434 
435 /*
436  * Special spill buffer support used by "SA" framework
437  */
438 
439 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
440 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
441     void *tag, dmu_buf_t **dbp);
442 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
443 
444 /*
445  * Obtain the DMU buffer from the specified object which contains the
446  * specified offset.  dmu_buf_hold() puts a "hold" on the buffer, so
447  * that it will remain in memory.  You must release the hold with
448  * dmu_buf_rele().  You musn't access the dmu_buf_t after releasing your
449  * hold.  You must have a hold on any dmu_buf_t* you pass to the DMU.
450  *
451  * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
452  * on the returned buffer before reading or writing the buffer's
453  * db_data.  The comments for those routines describe what particular
454  * operations are valid after calling them.
455  *
456  * The object number must be a valid, allocated object number.
457  */
458 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
459     void *tag, dmu_buf_t **, int flags);
460 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
461 void dmu_buf_rele(dmu_buf_t *db, void *tag);
462 uint64_t dmu_buf_refcount(dmu_buf_t *db);
463 
464 /*
465  * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
466  * range of an object.  A pointer to an array of dmu_buf_t*'s is
467  * returned (in *dbpp).
468  *
469  * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
470  * frees the array.  The hold on the array of buffers MUST be released
471  * with dmu_buf_rele_array.  You can NOT release the hold on each buffer
472  * individually with dmu_buf_rele.
473  */
474 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
475     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
476 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
477 
478 typedef void dmu_buf_evict_func_t(void *user_ptr);
479 
480 /*
481  * A DMU buffer user object may be associated with a dbuf for the
482  * duration of its lifetime.  This allows the user of a dbuf (client)
483  * to attach private data to a dbuf (e.g. in-core only data such as a
484  * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
485  * when that dbuf has been evicted.  Clients typically respond to the
486  * eviction notification by freeing their private data, thus ensuring
487  * the same lifetime for both dbuf and private data.
488  *
489  * The mapping from a dmu_buf_user_t to any client private data is the
490  * client's responsibility.  All current consumers of the API with private
491  * data embed a dmu_buf_user_t as the first member of the structure for
492  * their private data.  This allows conversions between the two types
493  * with a simple cast.  Since the DMU buf user API never needs access
494  * to the private data, other strategies can be employed if necessary
495  * or convenient for the client (e.g. using container_of() to do the
496  * conversion for private data that cannot have the dmu_buf_user_t as
497  * its first member).
498  *
499  * Eviction callbacks are executed without the dbuf mutex held or any
500  * other type of mechanism to guarantee that the dbuf is still available.
501  * For this reason, users must assume the dbuf has already been freed
502  * and not reference the dbuf from the callback context.
503  *
504  * Users requesting "immediate eviction" are notified as soon as the dbuf
505  * is only referenced by dirty records (dirties == holds).  Otherwise the
506  * notification occurs after eviction processing for the dbuf begins.
507  */
508 typedef struct dmu_buf_user {
509 	/*
510 	 * Asynchronous user eviction callback state.
511 	 */
512 	taskq_ent_t	dbu_tqent;
513 
514 	/* This instance's eviction function pointer. */
515 	dmu_buf_evict_func_t *dbu_evict_func;
516 #ifdef ZFS_DEBUG
517 	/*
518 	 * Pointer to user's dbuf pointer.  NULL for clients that do
519 	 * not associate a dbuf with their user data.
520 	 *
521 	 * The dbuf pointer is cleared upon eviction so as to catch
522 	 * use-after-evict bugs in clients.
523 	 */
524 	dmu_buf_t **dbu_clear_on_evict_dbufp;
525 #endif
526 } dmu_buf_user_t;
527 
528 /*
529  * Initialize the given dmu_buf_user_t instance with the eviction function
530  * evict_func, to be called when the user is evicted.
531  *
532  * NOTE: This function should only be called once on a given dmu_buf_user_t.
533  *       To allow enforcement of this, dbu must already be zeroed on entry.
534  */
535 #ifdef __lint
536 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
537 extern void
538 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
539     dmu_buf_t **clear_on_evict_dbufp);
540 #else /* __lint */
541 inline void
542 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
543     dmu_buf_t **clear_on_evict_dbufp)
544 {
545 	ASSERT(dbu->dbu_evict_func == NULL);
546 	ASSERT(evict_func != NULL);
547 	dbu->dbu_evict_func = evict_func;
548 #ifdef ZFS_DEBUG
549 	dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
550 #endif
551 }
552 #endif /* __lint */
553 
554 /*
555  * Attach user data to a dbuf and mark it for normal (when the dbuf's
556  * data is cleared or its reference count goes to zero) eviction processing.
557  *
558  * Returns NULL on success, or the existing user if another user currently
559  * owns the buffer.
560  */
561 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
562 
563 /*
564  * Attach user data to a dbuf and mark it for immediate (its dirty and
565  * reference counts are equal) eviction processing.
566  *
567  * Returns NULL on success, or the existing user if another user currently
568  * owns the buffer.
569  */
570 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
571 
572 /*
573  * Replace the current user of a dbuf.
574  *
575  * If given the current user of a dbuf, replaces the dbuf's user with
576  * "new_user" and returns the user data pointer that was replaced.
577  * Otherwise returns the current, and unmodified, dbuf user pointer.
578  */
579 void *dmu_buf_replace_user(dmu_buf_t *db,
580     dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
581 
582 /*
583  * Remove the specified user data for a DMU buffer.
584  *
585  * Returns the user that was removed on success, or the current user if
586  * another user currently owns the buffer.
587  */
588 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
589 
590 /*
591  * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
592  */
593 void *dmu_buf_get_user(dmu_buf_t *db);
594 
595 /* Block until any in-progress dmu buf user evictions complete. */
596 void dmu_buf_user_evict_wait(void);
597 
598 /*
599  * Returns the blkptr associated with this dbuf, or NULL if not set.
600  */
601 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
602 
603 /*
604  * Indicate that you are going to modify the buffer's data (db_data).
605  *
606  * The transaction (tx) must be assigned to a txg (ie. you've called
607  * dmu_tx_assign()).  The buffer's object must be held in the tx
608  * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
609  */
610 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
611 
612 /*
613  * Tells if the given dbuf is freeable.
614  */
615 boolean_t dmu_buf_freeable(dmu_buf_t *);
616 
617 /*
618  * You must create a transaction, then hold the objects which you will
619  * (or might) modify as part of this transaction.  Then you must assign
620  * the transaction to a transaction group.  Once the transaction has
621  * been assigned, you can modify buffers which belong to held objects as
622  * part of this transaction.  You can't modify buffers before the
623  * transaction has been assigned; you can't modify buffers which don't
624  * belong to objects which this transaction holds; you can't hold
625  * objects once the transaction has been assigned.  You may hold an
626  * object which you are going to free (with dmu_object_free()), but you
627  * don't have to.
628  *
629  * You can abort the transaction before it has been assigned.
630  *
631  * Note that you may hold buffers (with dmu_buf_hold) at any time,
632  * regardless of transaction state.
633  */
634 
635 #define	DMU_NEW_OBJECT	(-1ULL)
636 #define	DMU_OBJECT_END	(-1ULL)
637 
638 dmu_tx_t *dmu_tx_create(objset_t *os);
639 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
640 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
641     uint64_t len);
642 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
643 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
644 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
645 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
646 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
647 void dmu_tx_abort(dmu_tx_t *tx);
648 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
649 void dmu_tx_wait(dmu_tx_t *tx);
650 void dmu_tx_commit(dmu_tx_t *tx);
651 void dmu_tx_mark_netfree(dmu_tx_t *tx);
652 
653 /*
654  * To register a commit callback, dmu_tx_callback_register() must be called.
655  *
656  * dcb_data is a pointer to caller private data that is passed on as a
657  * callback parameter. The caller is responsible for properly allocating and
658  * freeing it.
659  *
660  * When registering a callback, the transaction must be already created, but
661  * it cannot be committed or aborted. It can be assigned to a txg or not.
662  *
663  * The callback will be called after the transaction has been safely written
664  * to stable storage and will also be called if the dmu_tx is aborted.
665  * If there is any error which prevents the transaction from being committed to
666  * disk, the callback will be called with a value of error != 0.
667  */
668 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
669 
670 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
671     void *dcb_data);
672 
673 /*
674  * Free up the data blocks for a defined range of a file.  If size is
675  * -1, the range from offset to end-of-file is freed.
676  */
677 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
678 	uint64_t size, dmu_tx_t *tx);
679 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
680 	uint64_t size);
681 int dmu_free_long_object(objset_t *os, uint64_t object);
682 
683 /*
684  * Convenience functions.
685  *
686  * Canfail routines will return 0 on success, or an errno if there is a
687  * nonrecoverable I/O error.
688  */
689 #define	DMU_READ_PREFETCH	0 /* prefetch */
690 #define	DMU_READ_NO_PREFETCH	1 /* don't prefetch */
691 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
692 	void *buf, uint32_t flags);
693 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
694 	const void *buf, dmu_tx_t *tx);
695 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
696 	dmu_tx_t *tx);
697 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
698 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
699 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
700     dmu_tx_t *tx);
701 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
702     dmu_tx_t *tx);
703 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
704     uint64_t size, struct page *pp, dmu_tx_t *tx);
705 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
706 void dmu_return_arcbuf(struct arc_buf *buf);
707 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
708     dmu_tx_t *tx);
709 int dmu_xuio_init(struct xuio *uio, int niov);
710 void dmu_xuio_fini(struct xuio *uio);
711 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
712     size_t n);
713 int dmu_xuio_cnt(struct xuio *uio);
714 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
715 void dmu_xuio_clear(struct xuio *uio, int i);
716 void xuio_stat_wbuf_copied();
717 void xuio_stat_wbuf_nocopy();
718 
719 extern int zfs_prefetch_disable;
720 extern int zfs_max_recordsize;
721 
722 /*
723  * Asynchronously try to read in the data.
724  */
725 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
726     uint64_t len);
727 
728 typedef struct dmu_object_info {
729 	/* All sizes are in bytes unless otherwise indicated. */
730 	uint32_t doi_data_block_size;
731 	uint32_t doi_metadata_block_size;
732 	dmu_object_type_t doi_type;
733 	dmu_object_type_t doi_bonus_type;
734 	uint64_t doi_bonus_size;
735 	uint8_t doi_indirection;		/* 2 = dnode->indirect->data */
736 	uint8_t doi_checksum;
737 	uint8_t doi_compress;
738 	uint8_t doi_nblkptr;
739 	uint8_t doi_pad[4];
740 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
741 	uint64_t doi_max_offset;
742 	uint64_t doi_fill_count;		/* number of non-empty blocks */
743 } dmu_object_info_t;
744 
745 typedef void arc_byteswap_func_t(void *buf, size_t size);
746 
747 typedef struct dmu_object_type_info {
748 	dmu_object_byteswap_t	ot_byteswap;
749 	boolean_t		ot_metadata;
750 	char			*ot_name;
751 } dmu_object_type_info_t;
752 
753 typedef struct dmu_object_byteswap_info {
754 	arc_byteswap_func_t	*ob_func;
755 	char			*ob_name;
756 } dmu_object_byteswap_info_t;
757 
758 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
759 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
760 
761 /*
762  * Get information on a DMU object.
763  *
764  * Return 0 on success or ENOENT if object is not allocated.
765  *
766  * If doi is NULL, just indicates whether the object exists.
767  */
768 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
769 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
770 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
771 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
772 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
773 /*
774  * Like dmu_object_info_from_db, but faster still when you only care about
775  * the size.  This is specifically optimized for zfs_getattr().
776  */
777 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
778     u_longlong_t *nblk512);
779 
780 typedef struct dmu_objset_stats {
781 	uint64_t dds_num_clones; /* number of clones of this */
782 	uint64_t dds_creation_txg;
783 	uint64_t dds_guid;
784 	dmu_objset_type_t dds_type;
785 	uint8_t dds_is_snapshot;
786 	uint8_t dds_inconsistent;
787 	char dds_origin[MAXNAMELEN];
788 } dmu_objset_stats_t;
789 
790 /*
791  * Get stats on a dataset.
792  */
793 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
794 
795 /*
796  * Add entries to the nvlist for all the objset's properties.  See
797  * zfs_prop_table[] and zfs(1m) for details on the properties.
798  */
799 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
800 
801 /*
802  * Get the space usage statistics for statvfs().
803  *
804  * refdbytes is the amount of space "referenced" by this objset.
805  * availbytes is the amount of space available to this objset, taking
806  * into account quotas & reservations, assuming that no other objsets
807  * use the space first.  These values correspond to the 'referenced' and
808  * 'available' properties, described in the zfs(1m) manpage.
809  *
810  * usedobjs and availobjs are the number of objects currently allocated,
811  * and available.
812  */
813 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
814     uint64_t *usedobjsp, uint64_t *availobjsp);
815 
816 /*
817  * The fsid_guid is a 56-bit ID that can change to avoid collisions.
818  * (Contrast with the ds_guid which is a 64-bit ID that will never
819  * change, so there is a small probability that it will collide.)
820  */
821 uint64_t dmu_objset_fsid_guid(objset_t *os);
822 
823 /*
824  * Get the [cm]time for an objset's snapshot dir
825  */
826 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
827 
828 int dmu_objset_is_snapshot(objset_t *os);
829 
830 extern struct spa *dmu_objset_spa(objset_t *os);
831 extern struct zilog *dmu_objset_zil(objset_t *os);
832 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
833 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
834 extern void dmu_objset_name(objset_t *os, char *buf);
835 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
836 extern uint64_t dmu_objset_id(objset_t *os);
837 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
838 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
839 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
840     uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
841 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
842     int maxlen, boolean_t *conflict);
843 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
844     uint64_t *idp, uint64_t *offp);
845 
846 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
847     void *bonus, uint64_t *userp, uint64_t *groupp);
848 extern void dmu_objset_register_type(dmu_objset_type_t ost,
849     objset_used_cb_t *cb);
850 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
851 extern void *dmu_objset_get_user(objset_t *os);
852 
853 /*
854  * Return the txg number for the given assigned transaction.
855  */
856 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
857 
858 /*
859  * Synchronous write.
860  * If a parent zio is provided this function initiates a write on the
861  * provided buffer as a child of the parent zio.
862  * In the absence of a parent zio, the write is completed synchronously.
863  * At write completion, blk is filled with the bp of the written block.
864  * Note that while the data covered by this function will be on stable
865  * storage when the write completes this new data does not become a
866  * permanent part of the file until the associated transaction commits.
867  */
868 
869 /*
870  * {zfs,zvol,ztest}_get_done() args
871  */
872 typedef struct zgd {
873 	struct zilog	*zgd_zilog;
874 	struct blkptr	*zgd_bp;
875 	dmu_buf_t	*zgd_db;
876 	struct rl	*zgd_rl;
877 	void		*zgd_private;
878 } zgd_t;
879 
880 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
881 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
882 
883 /*
884  * Find the next hole or data block in file starting at *off
885  * Return found offset in *off. Return ESRCH for end of file.
886  */
887 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
888     uint64_t *off);
889 
890 /*
891  * Initial setup and final teardown.
892  */
893 extern void dmu_init(void);
894 extern void dmu_fini(void);
895 
896 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
897     uint64_t object, uint64_t offset, int len);
898 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
899     dmu_traverse_cb_t cb, void *arg);
900 
901 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
902     struct vnode *vp, offset_t *offp);
903 
904 /* CRC64 table */
905 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
906 extern uint64_t zfs_crc64_table[256];
907 
908 extern int zfs_mdcomp_disable;
909 
910 #ifdef	__cplusplus
911 }
912 #endif
913 
914 #endif	/* _SYS_DMU_H */
915