Print this page
3752 want more verifiable dbuf user eviction
Submitted by: Justin Gibbs <justing@spectralogic.com>
Submitted by: Will Andrews <willa@spectralogic.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/sys/dmu.h
+++ new/usr/src/uts/common/fs/zfs/sys/dmu.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright (c) 2012 by Delphix. All rights reserved.
25 25 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
26 26 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
27 27 */
28 28
29 29 /* Portions Copyright 2010 Robert Milkowski */
30 30
31 31 #ifndef _SYS_DMU_H
|
↓ open down ↓ |
31 lines elided |
↑ open up ↑ |
32 32 #define _SYS_DMU_H
33 33
34 34 /*
35 35 * This file describes the interface that the DMU provides for its
36 36 * consumers.
37 37 *
38 38 * The DMU also interacts with the SPA. That interface is described in
39 39 * dmu_spa.h.
40 40 */
41 41
42 +#include <sys/zfs_context.h>
42 43 #include <sys/inttypes.h>
43 44 #include <sys/types.h>
44 45 #include <sys/param.h>
45 46 #include <sys/cred.h>
46 47 #include <sys/time.h>
47 48 #include <sys/fs/zfs.h>
48 49
49 50 #ifdef __cplusplus
50 51 extern "C" {
51 52 #endif
52 53
53 54 struct uio;
54 55 struct xuio;
55 56 struct page;
56 57 struct vnode;
57 58 struct spa;
58 59 struct zilog;
59 60 struct zio;
60 61 struct blkptr;
61 62 struct zap_cursor;
62 63 struct dsl_dataset;
63 64 struct dsl_pool;
64 65 struct dnode;
65 66 struct drr_begin;
66 67 struct drr_end;
67 68 struct zbookmark;
68 69 struct spa;
69 70 struct nvlist;
70 71 struct arc_buf;
71 72 struct zio_prop;
72 73 struct sa_handle;
73 74
74 75 typedef struct objset objset_t;
75 76 typedef struct dmu_tx dmu_tx_t;
76 77 typedef struct dsl_dir dsl_dir_t;
77 78
78 79 typedef enum dmu_object_byteswap {
79 80 DMU_BSWAP_UINT8,
80 81 DMU_BSWAP_UINT16,
81 82 DMU_BSWAP_UINT32,
82 83 DMU_BSWAP_UINT64,
83 84 DMU_BSWAP_ZAP,
84 85 DMU_BSWAP_DNODE,
85 86 DMU_BSWAP_OBJSET,
86 87 DMU_BSWAP_ZNODE,
87 88 DMU_BSWAP_OLDACL,
88 89 DMU_BSWAP_ACL,
89 90 /*
90 91 * Allocating a new byteswap type number makes the on-disk format
91 92 * incompatible with any other format that uses the same number.
92 93 *
93 94 * Data can usually be structured to work with one of the
94 95 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
95 96 */
96 97 DMU_BSWAP_NUMFUNCS
97 98 } dmu_object_byteswap_t;
98 99
99 100 #define DMU_OT_NEWTYPE 0x80
100 101 #define DMU_OT_METADATA 0x40
101 102 #define DMU_OT_BYTESWAP_MASK 0x3f
102 103
103 104 /*
104 105 * Defines a uint8_t object type. Object types specify if the data
105 106 * in the object is metadata (boolean) and how to byteswap the data
106 107 * (dmu_object_byteswap_t).
107 108 */
108 109 #define DMU_OT(byteswap, metadata) \
109 110 (DMU_OT_NEWTYPE | \
110 111 ((metadata) ? DMU_OT_METADATA : 0) | \
111 112 ((byteswap) & DMU_OT_BYTESWAP_MASK))
112 113
113 114 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
114 115 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
115 116 (ot) < DMU_OT_NUMTYPES)
116 117
117 118 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
118 119 ((ot) & DMU_OT_METADATA) : \
119 120 dmu_ot[(ot)].ot_metadata)
120 121
121 122 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
122 123 ((ot) & DMU_OT_BYTESWAP_MASK) : \
123 124 dmu_ot[(ot)].ot_byteswap)
124 125
125 126 typedef enum dmu_object_type {
126 127 DMU_OT_NONE,
127 128 /* general: */
128 129 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
129 130 DMU_OT_OBJECT_ARRAY, /* UINT64 */
130 131 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
131 132 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
132 133 DMU_OT_BPOBJ, /* UINT64 */
133 134 DMU_OT_BPOBJ_HDR, /* UINT64 */
134 135 /* spa: */
135 136 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
136 137 DMU_OT_SPACE_MAP, /* UINT64 */
137 138 /* zil: */
138 139 DMU_OT_INTENT_LOG, /* UINT64 */
139 140 /* dmu: */
140 141 DMU_OT_DNODE, /* DNODE */
141 142 DMU_OT_OBJSET, /* OBJSET */
142 143 /* dsl: */
143 144 DMU_OT_DSL_DIR, /* UINT64 */
144 145 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
145 146 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
146 147 DMU_OT_DSL_PROPS, /* ZAP */
147 148 DMU_OT_DSL_DATASET, /* UINT64 */
148 149 /* zpl: */
149 150 DMU_OT_ZNODE, /* ZNODE */
150 151 DMU_OT_OLDACL, /* Old ACL */
151 152 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
152 153 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
153 154 DMU_OT_MASTER_NODE, /* ZAP */
154 155 DMU_OT_UNLINKED_SET, /* ZAP */
155 156 /* zvol: */
156 157 DMU_OT_ZVOL, /* UINT8 */
157 158 DMU_OT_ZVOL_PROP, /* ZAP */
158 159 /* other; for testing only! */
159 160 DMU_OT_PLAIN_OTHER, /* UINT8 */
160 161 DMU_OT_UINT64_OTHER, /* UINT64 */
161 162 DMU_OT_ZAP_OTHER, /* ZAP */
162 163 /* new object types: */
163 164 DMU_OT_ERROR_LOG, /* ZAP */
164 165 DMU_OT_SPA_HISTORY, /* UINT8 */
165 166 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
166 167 DMU_OT_POOL_PROPS, /* ZAP */
167 168 DMU_OT_DSL_PERMS, /* ZAP */
168 169 DMU_OT_ACL, /* ACL */
169 170 DMU_OT_SYSACL, /* SYSACL */
170 171 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
171 172 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
172 173 DMU_OT_NEXT_CLONES, /* ZAP */
173 174 DMU_OT_SCAN_QUEUE, /* ZAP */
174 175 DMU_OT_USERGROUP_USED, /* ZAP */
175 176 DMU_OT_USERGROUP_QUOTA, /* ZAP */
176 177 DMU_OT_USERREFS, /* ZAP */
177 178 DMU_OT_DDT_ZAP, /* ZAP */
178 179 DMU_OT_DDT_STATS, /* ZAP */
179 180 DMU_OT_SA, /* System attr */
180 181 DMU_OT_SA_MASTER_NODE, /* ZAP */
181 182 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
182 183 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
183 184 DMU_OT_SCAN_XLATE, /* ZAP */
184 185 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
185 186 DMU_OT_DEADLIST, /* ZAP */
186 187 DMU_OT_DEADLIST_HDR, /* UINT64 */
187 188 DMU_OT_DSL_CLONES, /* ZAP */
188 189 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
189 190 /*
190 191 * Do not allocate new object types here. Doing so makes the on-disk
191 192 * format incompatible with any other format that uses the same object
192 193 * type number.
193 194 *
194 195 * When creating an object which does not have one of the above types
195 196 * use the DMU_OTN_* type with the correct byteswap and metadata
196 197 * values.
197 198 *
198 199 * The DMU_OTN_* types do not have entries in the dmu_ot table,
199 200 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
200 201 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
201 202 * and DMU_OTN_* types).
202 203 */
203 204 DMU_OT_NUMTYPES,
204 205
205 206 /*
206 207 * Names for valid types declared with DMU_OT().
207 208 */
208 209 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
209 210 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
210 211 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
211 212 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
212 213 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
213 214 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
214 215 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
215 216 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
216 217 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
217 218 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
218 219 } dmu_object_type_t;
219 220
220 221 typedef enum txg_how {
221 222 TXG_WAIT = 1,
222 223 TXG_NOWAIT,
223 224 } txg_how_t;
224 225
225 226 void byteswap_uint64_array(void *buf, size_t size);
226 227 void byteswap_uint32_array(void *buf, size_t size);
227 228 void byteswap_uint16_array(void *buf, size_t size);
228 229 void byteswap_uint8_array(void *buf, size_t size);
229 230 void zap_byteswap(void *buf, size_t size);
230 231 void zfs_oldacl_byteswap(void *buf, size_t size);
231 232 void zfs_acl_byteswap(void *buf, size_t size);
232 233 void zfs_znode_byteswap(void *buf, size_t size);
233 234
234 235 #define DS_FIND_SNAPSHOTS (1<<0)
235 236 #define DS_FIND_CHILDREN (1<<1)
236 237
237 238 /*
238 239 * The maximum number of bytes that can be accessed as part of one
239 240 * operation, including metadata.
240 241 */
241 242 #define DMU_MAX_ACCESS (10<<20) /* 10MB */
242 243 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
243 244
244 245 #define DMU_USERUSED_OBJECT (-1ULL)
245 246 #define DMU_GROUPUSED_OBJECT (-2ULL)
246 247 #define DMU_DEADLIST_OBJECT (-3ULL)
247 248
248 249 /*
249 250 * artificial blkids for bonus buffer and spill blocks
250 251 */
251 252 #define DMU_BONUS_BLKID (-1ULL)
252 253 #define DMU_SPILL_BLKID (-2ULL)
253 254 /*
254 255 * Public routines to create, destroy, open, and close objsets.
255 256 */
256 257 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
257 258 int dmu_objset_own(const char *name, dmu_objset_type_t type,
258 259 boolean_t readonly, void *tag, objset_t **osp);
259 260 void dmu_objset_rele(objset_t *os, void *tag);
260 261 void dmu_objset_disown(objset_t *os, void *tag);
261 262 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
262 263
263 264 void dmu_objset_evict_dbufs(objset_t *os);
264 265 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
265 266 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
266 267 int dmu_objset_clone(const char *name, const char *origin);
267 268 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
268 269 struct nvlist *errlist);
269 270 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
270 271 int dmu_objset_snapshot_tmp(const char *, const char *, int);
271 272 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
272 273 int flags);
273 274 void dmu_objset_byteswap(void *buf, size_t size);
|
↓ open down ↓ |
222 lines elided |
↑ open up ↑ |
274 275 int dsl_dataset_rename_snapshot(const char *fsname,
275 276 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
276 277
277 278 typedef struct dmu_buf {
278 279 uint64_t db_object; /* object that this buffer is part of */
279 280 uint64_t db_offset; /* byte offset in this object */
280 281 uint64_t db_size; /* size of buffer in bytes */
281 282 void *db_data; /* data in buffer */
282 283 } dmu_buf_t;
283 284
284 -typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr);
285 -
286 285 /*
287 286 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
288 287 */
289 288 #define DMU_POOL_DIRECTORY_OBJECT 1
290 289 #define DMU_POOL_CONFIG "config"
291 290 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
292 291 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
293 292 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
294 293 #define DMU_POOL_ROOT_DATASET "root_dataset"
295 294 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
296 295 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
297 296 #define DMU_POOL_ERRLOG_LAST "errlog_last"
298 297 #define DMU_POOL_SPARES "spares"
299 298 #define DMU_POOL_DEFLATE "deflate"
300 299 #define DMU_POOL_HISTORY "history"
301 300 #define DMU_POOL_PROPS "pool_props"
302 301 #define DMU_POOL_L2CACHE "l2cache"
303 302 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
304 303 #define DMU_POOL_DDT "DDT-%s-%s-%s"
305 304 #define DMU_POOL_DDT_STATS "DDT-statistics"
306 305 #define DMU_POOL_CREATION_VERSION "creation_version"
307 306 #define DMU_POOL_SCAN "scan"
308 307 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
309 308 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
310 309 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
311 310
312 311 /*
313 312 * Allocate an object from this objset. The range of object numbers
314 313 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
315 314 *
316 315 * The transaction must be assigned to a txg. The newly allocated
317 316 * object will be "held" in the transaction (ie. you can modify the
318 317 * newly allocated object in this transaction).
319 318 *
320 319 * dmu_object_alloc() chooses an object and returns it in *objectp.
321 320 *
322 321 * dmu_object_claim() allocates a specific object number. If that
323 322 * number is already allocated, it fails and returns EEXIST.
324 323 *
325 324 * Return 0 on success, or ENOSPC or EEXIST as specified above.
326 325 */
327 326 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
328 327 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
329 328 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
330 329 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
331 330 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
332 331 int blocksize, dmu_object_type_t bonustype, int bonuslen);
333 332
334 333 /*
335 334 * Free an object from this objset.
336 335 *
337 336 * The object's data will be freed as well (ie. you don't need to call
338 337 * dmu_free(object, 0, -1, tx)).
339 338 *
340 339 * The object need not be held in the transaction.
341 340 *
342 341 * If there are any holds on this object's buffers (via dmu_buf_hold()),
343 342 * or tx holds on the object (via dmu_tx_hold_object()), you can not
344 343 * free it; it fails and returns EBUSY.
345 344 *
346 345 * If the object is not allocated, it fails and returns ENOENT.
347 346 *
348 347 * Return 0 on success, or EBUSY or ENOENT as specified above.
349 348 */
350 349 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
351 350
352 351 /*
353 352 * Find the next allocated or free object.
354 353 *
355 354 * The objectp parameter is in-out. It will be updated to be the next
356 355 * object which is allocated. Ignore objects which have not been
357 356 * modified since txg.
358 357 *
359 358 * XXX Can only be called on a objset with no dirty data.
360 359 *
361 360 * Returns 0 on success, or ENOENT if there are no more objects.
362 361 */
363 362 int dmu_object_next(objset_t *os, uint64_t *objectp,
364 363 boolean_t hole, uint64_t txg);
365 364
366 365 /*
367 366 * Set the data blocksize for an object.
368 367 *
369 368 * The object cannot have any blocks allcated beyond the first. If
370 369 * the first block is allocated already, the new size must be greater
371 370 * than the current block size. If these conditions are not met,
372 371 * ENOTSUP will be returned.
373 372 *
374 373 * Returns 0 on success, or EBUSY if there are any holds on the object
375 374 * contents, or ENOTSUP as described above.
376 375 */
377 376 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
378 377 int ibs, dmu_tx_t *tx);
379 378
380 379 /*
381 380 * Set the checksum property on a dnode. The new checksum algorithm will
382 381 * apply to all newly written blocks; existing blocks will not be affected.
383 382 */
384 383 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
385 384 dmu_tx_t *tx);
386 385
387 386 /*
388 387 * Set the compress property on a dnode. The new compression algorithm will
389 388 * apply to all newly written blocks; existing blocks will not be affected.
390 389 */
391 390 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
392 391 dmu_tx_t *tx);
393 392
394 393 /*
395 394 * Decide how to write a block: checksum, compression, number of copies, etc.
396 395 */
397 396 #define WP_NOFILL 0x1
398 397 #define WP_DMU_SYNC 0x2
399 398 #define WP_SPILL 0x4
400 399
401 400 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
402 401 struct zio_prop *zp);
403 402 /*
404 403 * The bonus data is accessed more or less like a regular buffer.
405 404 * You must dmu_bonus_hold() to get the buffer, which will give you a
406 405 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
407 406 * data. As with any normal buffer, you must call dmu_buf_read() to
408 407 * read db_data, dmu_buf_will_dirty() before modifying it, and the
409 408 * object must be held in an assigned transaction before calling
410 409 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
411 410 * buffer as well. You must release your hold with dmu_buf_rele().
412 411 */
413 412 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
414 413 int dmu_bonus_max(void);
415 414 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
416 415 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
417 416 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
418 417 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
419 418
420 419 /*
421 420 * Special spill buffer support used by "SA" framework
422 421 */
423 422
424 423 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
425 424 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
426 425 void *tag, dmu_buf_t **dbp);
427 426 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
428 427
429 428 /*
430 429 * Obtain the DMU buffer from the specified object which contains the
431 430 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
432 431 * that it will remain in memory. You must release the hold with
433 432 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
434 433 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
435 434 *
436 435 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
437 436 * on the returned buffer before reading or writing the buffer's
438 437 * db_data. The comments for those routines describe what particular
439 438 * operations are valid after calling them.
440 439 *
441 440 * The object number must be a valid, allocated object number.
442 441 */
443 442 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
444 443 void *tag, dmu_buf_t **, int flags);
445 444 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
446 445 void dmu_buf_rele(dmu_buf_t *db, void *tag);
447 446 uint64_t dmu_buf_refcount(dmu_buf_t *db);
448 447
449 448 /*
450 449 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
451 450 * range of an object. A pointer to an array of dmu_buf_t*'s is
452 451 * returned (in *dbpp).
|
↓ open down ↓ |
157 lines elided |
↑ open up ↑ |
453 452 *
454 453 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
455 454 * frees the array. The hold on the array of buffers MUST be released
456 455 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
457 456 * individually with dmu_buf_rele.
458 457 */
459 458 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
460 459 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
461 460 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
462 461
463 -/*
464 - * Returns NULL on success, or the existing user ptr if it's already
465 - * been set.
466 - *
467 - * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
468 - *
469 - * user_data_ptr_ptr should be NULL, or a pointer to a pointer which
470 - * will be set to db->db_data when you are allowed to access it. Note
471 - * that db->db_data (the pointer) can change when you do dmu_buf_read(),
472 - * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill().
473 - * *user_data_ptr_ptr will be set to the new value when it changes.
474 - *
475 - * If non-NULL, pageout func will be called when this buffer is being
476 - * excised from the cache, so that you can clean up the data structure
477 - * pointed to by user_ptr.
478 - *
479 - * dmu_evict_user() will call the pageout func for all buffers in a
480 - * objset with a given pageout func.
481 - */
482 -void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr,
483 - dmu_buf_evict_func_t *pageout_func);
484 -/*
485 - * set_user_ie is the same as set_user, but request immediate eviction
486 - * when hold count goes to zero.
487 - */
488 -void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
489 - void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func);
490 -void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
491 - void *user_ptr, void *user_data_ptr_ptr,
492 - dmu_buf_evict_func_t *pageout_func);
493 -void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
462 +struct dmu_buf_user;
463 +
464 +typedef void dmu_buf_evict_func_t(struct dmu_buf_user *);
494 465
495 466 /*
496 - * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
467 + * The DMU buffer user object is used to allow private data to be
468 + * associated with a dbuf for the duration of its lifetime. This private
469 + * data must include a dmu_buf_user_t as its first object, which is passed
470 + * into the DMU user data API and can be attached to a dbuf. Clients can
471 + * regain access to their private data structure with a cast.
472 + *
473 + * DMU buffer users can be notified via a callback when their associated
474 + * dbuf has been evicted. This is typically used to free the user's
475 + * private data. The eviction callback is executed without the dbuf
476 + * mutex held or any other type of mechanism to guarantee that the
477 + * dbuf is still available. For this reason, users must assume the dbuf
478 + * has already been freed and not reference the dbuf from the callback
479 + * context.
480 + *
481 + * Users requestion "immediate eviction" are notified as soon as the dbuf
482 + * is only referenced by dirty records (dirties == holds). Otherwise the
483 + * eviction callback occurs after the last reference to the dbuf is dropped.
484 + *
485 + * Eviction Callback Processing
486 + * ============================
487 + * In any context where a dbuf reference drop may trigger an eviction, an
488 + * eviction queue object must be provided. This queue must then be
489 + * processed while not holding any dbuf locks. In this way, the user can
490 + * perform any work needed in their eviction function without fear of
491 + * lock order reversals.
492 + *
493 + * Implementation Note
494 + * ============================
495 + * Some users will occasionally want to map a structure directly onto the
496 + * backing dbuf. Using an union with an name alias macro to access these
497 + * overlays reduces the ugliness of code that accesses them. Initial work on
498 + * user objects involved using a macro that took the user object as an
499 + * argument to access the fields, which resulted in hundreds of lines of
500 + * needless diffs and wasn't any easier to read.
497 501 */
498 -void *dmu_buf_get_user(dmu_buf_t *db);
502 +typedef struct dmu_buf_user {
503 + /*
504 + * This instance's link in the eviction queue. Set when the buffer
505 + * has evicted and the callback needs to be called.
506 + */
507 + list_node_t evict_queue_link;
508 + /** This instance's eviction function pointer. */
509 + dmu_buf_evict_func_t *evict_func;
510 +} dmu_buf_user_t;
511 +
512 +/*
513 + * Initialize the given dmu_buf_user_t instance with the eviction function
514 + * evict_func, to be called when the user is evicted.
515 + *
516 + * NOTE: This function should only be called once on a given object. To
517 + * help enforce this, dbu should already be zeroed on entry.
518 + */
519 +static inline void
520 +dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func)
521 +{
522 + ASSERT(dbu->evict_func == NULL);
523 + ASSERT(!list_link_active(&dbu->evict_queue_link));
524 + dbu->evict_func = evict_func;
525 +}
526 +
527 +static inline void
528 +dmu_buf_create_user_evict_list(list_t *evict_list_p)
529 +{
530 + list_create(evict_list_p, sizeof(dmu_buf_user_t),
531 + offsetof(dmu_buf_user_t, evict_queue_link));
532 +}
533 +
534 +static inline void
535 +dmu_buf_process_user_evicts(list_t *evict_list_p)
536 +{
537 + dmu_buf_user_t *dbu, *next;
538 +
539 + for (dbu = (dmu_buf_user_t *)list_head(evict_list_p); dbu != NULL;
540 + dbu = next) {
541 + next = (dmu_buf_user_t *)list_next(evict_list_p, dbu);
542 + list_remove(evict_list_p, dbu);
543 + dbu->evict_func(dbu);
544 + }
545 +}
546 +
547 +static inline void
548 +dmu_buf_destroy_user_evict_list(list_t *evict_list_p)
549 +{
550 + dmu_buf_process_user_evicts(evict_list_p);
551 + list_destroy(evict_list_p);
552 +}
553 +
554 +dmu_buf_user_t *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
555 +dmu_buf_user_t *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
556 +dmu_buf_user_t *dmu_buf_replace_user(dmu_buf_t *db,
557 + dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
558 +dmu_buf_user_t *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
559 +dmu_buf_user_t *dmu_buf_get_user(dmu_buf_t *db);
499 560
500 561 /*
501 562 * Returns the blkptr associated with this dbuf, or NULL if not set.
502 563 */
503 564 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
504 565
505 566 /*
506 567 * Indicate that you are going to modify the buffer's data (db_data).
507 568 *
508 569 * The transaction (tx) must be assigned to a txg (ie. you've called
509 570 * dmu_tx_assign()). The buffer's object must be held in the tx
510 571 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
511 572 */
512 573 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
513 574
514 575 /*
515 576 * Tells if the given dbuf is freeable.
516 577 */
517 578 boolean_t dmu_buf_freeable(dmu_buf_t *);
518 579
519 580 /*
520 581 * You must create a transaction, then hold the objects which you will
521 582 * (or might) modify as part of this transaction. Then you must assign
522 583 * the transaction to a transaction group. Once the transaction has
523 584 * been assigned, you can modify buffers which belong to held objects as
524 585 * part of this transaction. You can't modify buffers before the
525 586 * transaction has been assigned; you can't modify buffers which don't
526 587 * belong to objects which this transaction holds; you can't hold
527 588 * objects once the transaction has been assigned. You may hold an
528 589 * object which you are going to free (with dmu_object_free()), but you
529 590 * don't have to.
530 591 *
531 592 * You can abort the transaction before it has been assigned.
532 593 *
533 594 * Note that you may hold buffers (with dmu_buf_hold) at any time,
534 595 * regardless of transaction state.
535 596 */
536 597
537 598 #define DMU_NEW_OBJECT (-1ULL)
538 599 #define DMU_OBJECT_END (-1ULL)
539 600
540 601 dmu_tx_t *dmu_tx_create(objset_t *os);
541 602 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
542 603 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
543 604 uint64_t len);
544 605 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
545 606 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
546 607 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
547 608 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
548 609 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
549 610 void dmu_tx_abort(dmu_tx_t *tx);
550 611 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
551 612 void dmu_tx_wait(dmu_tx_t *tx);
552 613 void dmu_tx_commit(dmu_tx_t *tx);
553 614
554 615 /*
555 616 * To register a commit callback, dmu_tx_callback_register() must be called.
556 617 *
557 618 * dcb_data is a pointer to caller private data that is passed on as a
558 619 * callback parameter. The caller is responsible for properly allocating and
559 620 * freeing it.
560 621 *
561 622 * When registering a callback, the transaction must be already created, but
562 623 * it cannot be committed or aborted. It can be assigned to a txg or not.
563 624 *
564 625 * The callback will be called after the transaction has been safely written
565 626 * to stable storage and will also be called if the dmu_tx is aborted.
566 627 * If there is any error which prevents the transaction from being committed to
567 628 * disk, the callback will be called with a value of error != 0.
568 629 */
569 630 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
570 631
571 632 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
572 633 void *dcb_data);
573 634
574 635 /*
575 636 * Free up the data blocks for a defined range of a file. If size is
576 637 * -1, the range from offset to end-of-file is freed.
577 638 */
578 639 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
579 640 uint64_t size, dmu_tx_t *tx);
580 641 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
581 642 uint64_t size);
582 643 int dmu_free_object(objset_t *os, uint64_t object);
583 644
584 645 /*
585 646 * Convenience functions.
586 647 *
587 648 * Canfail routines will return 0 on success, or an errno if there is a
588 649 * nonrecoverable I/O error.
589 650 */
590 651 #define DMU_READ_PREFETCH 0 /* prefetch */
591 652 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
592 653 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
593 654 void *buf, uint32_t flags);
594 655 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
595 656 const void *buf, dmu_tx_t *tx);
596 657 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
597 658 dmu_tx_t *tx);
598 659 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
599 660 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
600 661 dmu_tx_t *tx);
601 662 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
602 663 dmu_tx_t *tx);
603 664 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
604 665 uint64_t size, struct page *pp, dmu_tx_t *tx);
605 666 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
606 667 void dmu_return_arcbuf(struct arc_buf *buf);
607 668 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
608 669 dmu_tx_t *tx);
609 670 int dmu_xuio_init(struct xuio *uio, int niov);
610 671 void dmu_xuio_fini(struct xuio *uio);
611 672 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
612 673 size_t n);
613 674 int dmu_xuio_cnt(struct xuio *uio);
614 675 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
615 676 void dmu_xuio_clear(struct xuio *uio, int i);
616 677 void xuio_stat_wbuf_copied();
617 678 void xuio_stat_wbuf_nocopy();
618 679
619 680 extern int zfs_prefetch_disable;
620 681
621 682 /*
622 683 * Asynchronously try to read in the data.
623 684 */
624 685 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
625 686 uint64_t len);
626 687
627 688 typedef struct dmu_object_info {
628 689 /* All sizes are in bytes unless otherwise indicated. */
629 690 uint32_t doi_data_block_size;
630 691 uint32_t doi_metadata_block_size;
631 692 dmu_object_type_t doi_type;
632 693 dmu_object_type_t doi_bonus_type;
633 694 uint64_t doi_bonus_size;
634 695 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
635 696 uint8_t doi_checksum;
636 697 uint8_t doi_compress;
637 698 uint8_t doi_pad[5];
638 699 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
639 700 uint64_t doi_max_offset;
640 701 uint64_t doi_fill_count; /* number of non-empty blocks */
641 702 } dmu_object_info_t;
642 703
643 704 typedef void arc_byteswap_func_t(void *buf, size_t size);
644 705
645 706 typedef struct dmu_object_type_info {
646 707 dmu_object_byteswap_t ot_byteswap;
647 708 boolean_t ot_metadata;
648 709 char *ot_name;
649 710 } dmu_object_type_info_t;
650 711
651 712 typedef struct dmu_object_byteswap_info {
652 713 arc_byteswap_func_t *ob_func;
653 714 char *ob_name;
654 715 } dmu_object_byteswap_info_t;
655 716
656 717 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
657 718 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
658 719
659 720 /*
660 721 * Get information on a DMU object.
661 722 *
662 723 * Return 0 on success or ENOENT if object is not allocated.
663 724 *
664 725 * If doi is NULL, just indicates whether the object exists.
665 726 */
666 727 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
667 728 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
668 729 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
669 730 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
670 731 u_longlong_t *nblk512);
671 732
672 733 typedef struct dmu_objset_stats {
673 734 uint64_t dds_num_clones; /* number of clones of this */
674 735 uint64_t dds_creation_txg;
675 736 uint64_t dds_guid;
676 737 dmu_objset_type_t dds_type;
677 738 uint8_t dds_is_snapshot;
678 739 uint8_t dds_inconsistent;
679 740 char dds_origin[MAXNAMELEN];
680 741 } dmu_objset_stats_t;
681 742
682 743 /*
683 744 * Get stats on a dataset.
684 745 */
685 746 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
686 747
687 748 /*
688 749 * Add entries to the nvlist for all the objset's properties. See
689 750 * zfs_prop_table[] and zfs(1m) for details on the properties.
690 751 */
691 752 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
692 753
693 754 /*
694 755 * Get the space usage statistics for statvfs().
695 756 *
696 757 * refdbytes is the amount of space "referenced" by this objset.
697 758 * availbytes is the amount of space available to this objset, taking
698 759 * into account quotas & reservations, assuming that no other objsets
699 760 * use the space first. These values correspond to the 'referenced' and
700 761 * 'available' properties, described in the zfs(1m) manpage.
701 762 *
702 763 * usedobjs and availobjs are the number of objects currently allocated,
703 764 * and available.
704 765 */
705 766 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
706 767 uint64_t *usedobjsp, uint64_t *availobjsp);
707 768
708 769 /*
709 770 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
710 771 * (Contrast with the ds_guid which is a 64-bit ID that will never
711 772 * change, so there is a small probability that it will collide.)
712 773 */
713 774 uint64_t dmu_objset_fsid_guid(objset_t *os);
714 775
715 776 /*
716 777 * Get the [cm]time for an objset's snapshot dir
717 778 */
718 779 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
719 780
720 781 int dmu_objset_is_snapshot(objset_t *os);
721 782
722 783 extern struct spa *dmu_objset_spa(objset_t *os);
723 784 extern struct zilog *dmu_objset_zil(objset_t *os);
724 785 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
725 786 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
726 787 extern void dmu_objset_name(objset_t *os, char *buf);
727 788 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
728 789 extern uint64_t dmu_objset_id(objset_t *os);
729 790 extern uint64_t dmu_objset_syncprop(objset_t *os);
730 791 extern uint64_t dmu_objset_logbias(objset_t *os);
731 792 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
732 793 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
733 794 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
734 795 int maxlen, boolean_t *conflict);
735 796 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
736 797 uint64_t *idp, uint64_t *offp);
737 798
738 799 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
739 800 void *bonus, uint64_t *userp, uint64_t *groupp);
740 801 extern void dmu_objset_register_type(dmu_objset_type_t ost,
741 802 objset_used_cb_t *cb);
742 803 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
743 804 extern void *dmu_objset_get_user(objset_t *os);
744 805
745 806 /*
746 807 * Return the txg number for the given assigned transaction.
747 808 */
748 809 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
749 810
750 811 /*
751 812 * Synchronous write.
752 813 * If a parent zio is provided this function initiates a write on the
753 814 * provided buffer as a child of the parent zio.
754 815 * In the absence of a parent zio, the write is completed synchronously.
755 816 * At write completion, blk is filled with the bp of the written block.
756 817 * Note that while the data covered by this function will be on stable
757 818 * storage when the write completes this new data does not become a
758 819 * permanent part of the file until the associated transaction commits.
759 820 */
760 821
761 822 /*
762 823 * {zfs,zvol,ztest}_get_done() args
763 824 */
764 825 typedef struct zgd {
765 826 struct zilog *zgd_zilog;
766 827 struct blkptr *zgd_bp;
767 828 dmu_buf_t *zgd_db;
768 829 struct rl *zgd_rl;
769 830 void *zgd_private;
770 831 } zgd_t;
771 832
772 833 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
773 834 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
774 835
775 836 /*
776 837 * Find the next hole or data block in file starting at *off
777 838 * Return found offset in *off. Return ESRCH for end of file.
778 839 */
779 840 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
780 841 uint64_t *off);
781 842
782 843 /*
783 844 * Initial setup and final teardown.
784 845 */
785 846 extern void dmu_init(void);
786 847 extern void dmu_fini(void);
787 848
788 849 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
789 850 uint64_t object, uint64_t offset, int len);
790 851 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
791 852 dmu_traverse_cb_t cb, void *arg);
792 853
793 854 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
794 855 struct vnode *vp, offset_t *offp);
795 856
796 857 /* CRC64 table */
797 858 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
798 859 extern uint64_t zfs_crc64_table[256];
799 860
800 861 #ifdef __cplusplus
801 862 }
802 863 #endif
803 864
804 865 #endif /* _SYS_DMU_H */
|
↓ open down ↓ |
296 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX