Print this page
4045 zfs write throttle & i/o scheduler performance work
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.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) 2013 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
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 42 #include <sys/inttypes.h>
43 43 #include <sys/types.h>
44 44 #include <sys/param.h>
45 45 #include <sys/cred.h>
46 46 #include <sys/time.h>
47 47 #include <sys/fs/zfs.h>
48 48
49 49 #ifdef __cplusplus
50 50 extern "C" {
51 51 #endif
52 52
53 53 struct uio;
54 54 struct xuio;
55 55 struct page;
56 56 struct vnode;
57 57 struct spa;
58 58 struct zilog;
59 59 struct zio;
60 60 struct blkptr;
61 61 struct zap_cursor;
62 62 struct dsl_dataset;
63 63 struct dsl_pool;
64 64 struct dnode;
65 65 struct drr_begin;
66 66 struct drr_end;
67 67 struct zbookmark;
68 68 struct spa;
69 69 struct nvlist;
70 70 struct arc_buf;
71 71 struct zio_prop;
72 72 struct sa_handle;
73 73
74 74 typedef struct objset objset_t;
75 75 typedef struct dmu_tx dmu_tx_t;
76 76 typedef struct dsl_dir dsl_dir_t;
77 77
78 78 typedef enum dmu_object_byteswap {
79 79 DMU_BSWAP_UINT8,
80 80 DMU_BSWAP_UINT16,
81 81 DMU_BSWAP_UINT32,
82 82 DMU_BSWAP_UINT64,
83 83 DMU_BSWAP_ZAP,
84 84 DMU_BSWAP_DNODE,
85 85 DMU_BSWAP_OBJSET,
86 86 DMU_BSWAP_ZNODE,
87 87 DMU_BSWAP_OLDACL,
88 88 DMU_BSWAP_ACL,
89 89 /*
90 90 * Allocating a new byteswap type number makes the on-disk format
91 91 * incompatible with any other format that uses the same number.
92 92 *
93 93 * Data can usually be structured to work with one of the
94 94 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
95 95 */
96 96 DMU_BSWAP_NUMFUNCS
97 97 } dmu_object_byteswap_t;
98 98
99 99 #define DMU_OT_NEWTYPE 0x80
100 100 #define DMU_OT_METADATA 0x40
101 101 #define DMU_OT_BYTESWAP_MASK 0x3f
102 102
103 103 /*
104 104 * Defines a uint8_t object type. Object types specify if the data
105 105 * in the object is metadata (boolean) and how to byteswap the data
106 106 * (dmu_object_byteswap_t).
107 107 */
108 108 #define DMU_OT(byteswap, metadata) \
109 109 (DMU_OT_NEWTYPE | \
110 110 ((metadata) ? DMU_OT_METADATA : 0) | \
111 111 ((byteswap) & DMU_OT_BYTESWAP_MASK))
112 112
113 113 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
114 114 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
115 115 (ot) < DMU_OT_NUMTYPES)
116 116
117 117 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
118 118 ((ot) & DMU_OT_METADATA) : \
119 119 dmu_ot[(ot)].ot_metadata)
120 120
121 121 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
122 122 ((ot) & DMU_OT_BYTESWAP_MASK) : \
123 123 dmu_ot[(ot)].ot_byteswap)
124 124
125 125 typedef enum dmu_object_type {
126 126 DMU_OT_NONE,
127 127 /* general: */
128 128 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
129 129 DMU_OT_OBJECT_ARRAY, /* UINT64 */
130 130 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
131 131 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
132 132 DMU_OT_BPOBJ, /* UINT64 */
133 133 DMU_OT_BPOBJ_HDR, /* UINT64 */
134 134 /* spa: */
135 135 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
136 136 DMU_OT_SPACE_MAP, /* UINT64 */
137 137 /* zil: */
138 138 DMU_OT_INTENT_LOG, /* UINT64 */
139 139 /* dmu: */
140 140 DMU_OT_DNODE, /* DNODE */
141 141 DMU_OT_OBJSET, /* OBJSET */
142 142 /* dsl: */
143 143 DMU_OT_DSL_DIR, /* UINT64 */
144 144 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
145 145 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
146 146 DMU_OT_DSL_PROPS, /* ZAP */
147 147 DMU_OT_DSL_DATASET, /* UINT64 */
148 148 /* zpl: */
149 149 DMU_OT_ZNODE, /* ZNODE */
150 150 DMU_OT_OLDACL, /* Old ACL */
151 151 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
152 152 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
153 153 DMU_OT_MASTER_NODE, /* ZAP */
154 154 DMU_OT_UNLINKED_SET, /* ZAP */
155 155 /* zvol: */
156 156 DMU_OT_ZVOL, /* UINT8 */
157 157 DMU_OT_ZVOL_PROP, /* ZAP */
158 158 /* other; for testing only! */
159 159 DMU_OT_PLAIN_OTHER, /* UINT8 */
160 160 DMU_OT_UINT64_OTHER, /* UINT64 */
161 161 DMU_OT_ZAP_OTHER, /* ZAP */
162 162 /* new object types: */
163 163 DMU_OT_ERROR_LOG, /* ZAP */
164 164 DMU_OT_SPA_HISTORY, /* UINT8 */
165 165 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
166 166 DMU_OT_POOL_PROPS, /* ZAP */
167 167 DMU_OT_DSL_PERMS, /* ZAP */
168 168 DMU_OT_ACL, /* ACL */
169 169 DMU_OT_SYSACL, /* SYSACL */
170 170 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
171 171 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
172 172 DMU_OT_NEXT_CLONES, /* ZAP */
173 173 DMU_OT_SCAN_QUEUE, /* ZAP */
174 174 DMU_OT_USERGROUP_USED, /* ZAP */
175 175 DMU_OT_USERGROUP_QUOTA, /* ZAP */
176 176 DMU_OT_USERREFS, /* ZAP */
177 177 DMU_OT_DDT_ZAP, /* ZAP */
178 178 DMU_OT_DDT_STATS, /* ZAP */
179 179 DMU_OT_SA, /* System attr */
180 180 DMU_OT_SA_MASTER_NODE, /* ZAP */
181 181 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
182 182 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
183 183 DMU_OT_SCAN_XLATE, /* ZAP */
184 184 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
185 185 DMU_OT_DEADLIST, /* ZAP */
186 186 DMU_OT_DEADLIST_HDR, /* UINT64 */
187 187 DMU_OT_DSL_CLONES, /* ZAP */
188 188 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
189 189 /*
190 190 * Do not allocate new object types here. Doing so makes the on-disk
191 191 * format incompatible with any other format that uses the same object
192 192 * type number.
193 193 *
194 194 * When creating an object which does not have one of the above types
195 195 * use the DMU_OTN_* type with the correct byteswap and metadata
196 196 * values.
197 197 *
198 198 * The DMU_OTN_* types do not have entries in the dmu_ot table,
199 199 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
200 200 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
201 201 * and DMU_OTN_* types).
202 202 */
203 203 DMU_OT_NUMTYPES,
204 204
205 205 /*
206 206 * Names for valid types declared with DMU_OT().
207 207 */
208 208 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
209 209 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
210 210 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
211 211 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
212 212 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
↓ open down ↓ |
212 lines elided |
↑ open up ↑ |
213 213 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
214 214 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
215 215 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
216 216 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
217 217 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
218 218 } dmu_object_type_t;
219 219
220 220 typedef enum txg_how {
221 221 TXG_WAIT = 1,
222 222 TXG_NOWAIT,
223 + TXG_WAITED,
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);
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
284 285 typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr);
285 286
286 287 /*
287 288 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
288 289 */
289 290 #define DMU_POOL_DIRECTORY_OBJECT 1
290 291 #define DMU_POOL_CONFIG "config"
291 292 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
292 293 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
293 294 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
294 295 #define DMU_POOL_ROOT_DATASET "root_dataset"
295 296 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
296 297 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
297 298 #define DMU_POOL_ERRLOG_LAST "errlog_last"
298 299 #define DMU_POOL_SPARES "spares"
299 300 #define DMU_POOL_DEFLATE "deflate"
300 301 #define DMU_POOL_HISTORY "history"
301 302 #define DMU_POOL_PROPS "pool_props"
302 303 #define DMU_POOL_L2CACHE "l2cache"
303 304 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
304 305 #define DMU_POOL_DDT "DDT-%s-%s-%s"
305 306 #define DMU_POOL_DDT_STATS "DDT-statistics"
306 307 #define DMU_POOL_CREATION_VERSION "creation_version"
307 308 #define DMU_POOL_SCAN "scan"
308 309 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
309 310 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
310 311 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
311 312
312 313 /*
313 314 * Allocate an object from this objset. The range of object numbers
314 315 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
315 316 *
316 317 * The transaction must be assigned to a txg. The newly allocated
317 318 * object will be "held" in the transaction (ie. you can modify the
318 319 * newly allocated object in this transaction).
319 320 *
320 321 * dmu_object_alloc() chooses an object and returns it in *objectp.
321 322 *
322 323 * dmu_object_claim() allocates a specific object number. If that
323 324 * number is already allocated, it fails and returns EEXIST.
324 325 *
325 326 * Return 0 on success, or ENOSPC or EEXIST as specified above.
326 327 */
327 328 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
328 329 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
329 330 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
330 331 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
331 332 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
332 333 int blocksize, dmu_object_type_t bonustype, int bonuslen);
333 334
334 335 /*
335 336 * Free an object from this objset.
336 337 *
337 338 * The object's data will be freed as well (ie. you don't need to call
338 339 * dmu_free(object, 0, -1, tx)).
339 340 *
340 341 * The object need not be held in the transaction.
341 342 *
342 343 * If there are any holds on this object's buffers (via dmu_buf_hold()),
343 344 * or tx holds on the object (via dmu_tx_hold_object()), you can not
344 345 * free it; it fails and returns EBUSY.
345 346 *
346 347 * If the object is not allocated, it fails and returns ENOENT.
347 348 *
348 349 * Return 0 on success, or EBUSY or ENOENT as specified above.
349 350 */
350 351 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
351 352
352 353 /*
353 354 * Find the next allocated or free object.
354 355 *
355 356 * The objectp parameter is in-out. It will be updated to be the next
356 357 * object which is allocated. Ignore objects which have not been
357 358 * modified since txg.
358 359 *
359 360 * XXX Can only be called on a objset with no dirty data.
360 361 *
361 362 * Returns 0 on success, or ENOENT if there are no more objects.
362 363 */
363 364 int dmu_object_next(objset_t *os, uint64_t *objectp,
364 365 boolean_t hole, uint64_t txg);
365 366
366 367 /*
367 368 * Set the data blocksize for an object.
368 369 *
369 370 * The object cannot have any blocks allcated beyond the first. If
370 371 * the first block is allocated already, the new size must be greater
371 372 * than the current block size. If these conditions are not met,
372 373 * ENOTSUP will be returned.
373 374 *
374 375 * Returns 0 on success, or EBUSY if there are any holds on the object
375 376 * contents, or ENOTSUP as described above.
376 377 */
377 378 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
378 379 int ibs, dmu_tx_t *tx);
379 380
380 381 /*
381 382 * Set the checksum property on a dnode. The new checksum algorithm will
382 383 * apply to all newly written blocks; existing blocks will not be affected.
383 384 */
384 385 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
385 386 dmu_tx_t *tx);
386 387
387 388 /*
388 389 * Set the compress property on a dnode. The new compression algorithm will
389 390 * apply to all newly written blocks; existing blocks will not be affected.
390 391 */
391 392 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
392 393 dmu_tx_t *tx);
393 394
394 395 /*
395 396 * Decide how to write a block: checksum, compression, number of copies, etc.
396 397 */
397 398 #define WP_NOFILL 0x1
398 399 #define WP_DMU_SYNC 0x2
399 400 #define WP_SPILL 0x4
400 401
401 402 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
402 403 struct zio_prop *zp);
403 404 /*
404 405 * The bonus data is accessed more or less like a regular buffer.
405 406 * You must dmu_bonus_hold() to get the buffer, which will give you a
406 407 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
407 408 * data. As with any normal buffer, you must call dmu_buf_read() to
408 409 * read db_data, dmu_buf_will_dirty() before modifying it, and the
409 410 * object must be held in an assigned transaction before calling
410 411 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
411 412 * buffer as well. You must release your hold with dmu_buf_rele().
412 413 *
413 414 * Returns ENOENT, EIO, or 0.
414 415 */
415 416 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
416 417 int dmu_bonus_max(void);
417 418 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
418 419 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
419 420 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
420 421 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
421 422
422 423 /*
423 424 * Special spill buffer support used by "SA" framework
424 425 */
425 426
426 427 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
427 428 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
428 429 void *tag, dmu_buf_t **dbp);
429 430 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
430 431
431 432 /*
432 433 * Obtain the DMU buffer from the specified object which contains the
433 434 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
434 435 * that it will remain in memory. You must release the hold with
435 436 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
436 437 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
437 438 *
438 439 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
439 440 * on the returned buffer before reading or writing the buffer's
440 441 * db_data. The comments for those routines describe what particular
441 442 * operations are valid after calling them.
442 443 *
443 444 * The object number must be a valid, allocated object number.
444 445 */
445 446 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
446 447 void *tag, dmu_buf_t **, int flags);
447 448 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
448 449 void dmu_buf_rele(dmu_buf_t *db, void *tag);
449 450 uint64_t dmu_buf_refcount(dmu_buf_t *db);
450 451
451 452 /*
452 453 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
453 454 * range of an object. A pointer to an array of dmu_buf_t*'s is
454 455 * returned (in *dbpp).
455 456 *
456 457 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
457 458 * frees the array. The hold on the array of buffers MUST be released
458 459 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
459 460 * individually with dmu_buf_rele.
460 461 */
461 462 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
462 463 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
463 464 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
464 465
465 466 /*
466 467 * Returns NULL on success, or the existing user ptr if it's already
467 468 * been set.
468 469 *
469 470 * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
470 471 *
471 472 * user_data_ptr_ptr should be NULL, or a pointer to a pointer which
472 473 * will be set to db->db_data when you are allowed to access it. Note
473 474 * that db->db_data (the pointer) can change when you do dmu_buf_read(),
474 475 * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill().
475 476 * *user_data_ptr_ptr will be set to the new value when it changes.
476 477 *
477 478 * If non-NULL, pageout func will be called when this buffer is being
478 479 * excised from the cache, so that you can clean up the data structure
479 480 * pointed to by user_ptr.
480 481 *
481 482 * dmu_evict_user() will call the pageout func for all buffers in a
482 483 * objset with a given pageout func.
483 484 */
484 485 void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr,
485 486 dmu_buf_evict_func_t *pageout_func);
486 487 /*
487 488 * set_user_ie is the same as set_user, but request immediate eviction
488 489 * when hold count goes to zero.
489 490 */
490 491 void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
491 492 void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func);
492 493 void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
493 494 void *user_ptr, void *user_data_ptr_ptr,
494 495 dmu_buf_evict_func_t *pageout_func);
495 496 void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
496 497
497 498 /*
498 499 * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
499 500 */
500 501 void *dmu_buf_get_user(dmu_buf_t *db);
501 502
502 503 /*
503 504 * Returns the blkptr associated with this dbuf, or NULL if not set.
504 505 */
505 506 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
506 507
507 508 /*
508 509 * Indicate that you are going to modify the buffer's data (db_data).
509 510 *
510 511 * The transaction (tx) must be assigned to a txg (ie. you've called
511 512 * dmu_tx_assign()). The buffer's object must be held in the tx
512 513 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
513 514 */
514 515 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
515 516
516 517 /*
517 518 * Tells if the given dbuf is freeable.
518 519 */
519 520 boolean_t dmu_buf_freeable(dmu_buf_t *);
520 521
521 522 /*
522 523 * You must create a transaction, then hold the objects which you will
523 524 * (or might) modify as part of this transaction. Then you must assign
524 525 * the transaction to a transaction group. Once the transaction has
525 526 * been assigned, you can modify buffers which belong to held objects as
526 527 * part of this transaction. You can't modify buffers before the
527 528 * transaction has been assigned; you can't modify buffers which don't
528 529 * belong to objects which this transaction holds; you can't hold
529 530 * objects once the transaction has been assigned. You may hold an
530 531 * object which you are going to free (with dmu_object_free()), but you
531 532 * don't have to.
532 533 *
533 534 * You can abort the transaction before it has been assigned.
534 535 *
535 536 * Note that you may hold buffers (with dmu_buf_hold) at any time,
536 537 * regardless of transaction state.
537 538 */
538 539
539 540 #define DMU_NEW_OBJECT (-1ULL)
540 541 #define DMU_OBJECT_END (-1ULL)
541 542
542 543 dmu_tx_t *dmu_tx_create(objset_t *os);
543 544 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
544 545 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
545 546 uint64_t len);
546 547 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
547 548 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
548 549 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
549 550 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
550 551 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
551 552 void dmu_tx_abort(dmu_tx_t *tx);
552 553 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
553 554 void dmu_tx_wait(dmu_tx_t *tx);
554 555 void dmu_tx_commit(dmu_tx_t *tx);
555 556
556 557 /*
557 558 * To register a commit callback, dmu_tx_callback_register() must be called.
558 559 *
559 560 * dcb_data is a pointer to caller private data that is passed on as a
560 561 * callback parameter. The caller is responsible for properly allocating and
561 562 * freeing it.
562 563 *
563 564 * When registering a callback, the transaction must be already created, but
564 565 * it cannot be committed or aborted. It can be assigned to a txg or not.
565 566 *
566 567 * The callback will be called after the transaction has been safely written
567 568 * to stable storage and will also be called if the dmu_tx is aborted.
568 569 * If there is any error which prevents the transaction from being committed to
569 570 * disk, the callback will be called with a value of error != 0.
570 571 */
571 572 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
572 573
573 574 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
574 575 void *dcb_data);
575 576
576 577 /*
577 578 * Free up the data blocks for a defined range of a file. If size is
578 579 * -1, the range from offset to end-of-file is freed.
579 580 */
580 581 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
581 582 uint64_t size, dmu_tx_t *tx);
582 583 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
583 584 uint64_t size);
584 585 int dmu_free_long_object(objset_t *os, uint64_t object);
585 586
586 587 /*
587 588 * Convenience functions.
588 589 *
589 590 * Canfail routines will return 0 on success, or an errno if there is a
590 591 * nonrecoverable I/O error.
591 592 */
592 593 #define DMU_READ_PREFETCH 0 /* prefetch */
593 594 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
594 595 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
595 596 void *buf, uint32_t flags);
596 597 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
597 598 const void *buf, dmu_tx_t *tx);
598 599 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
599 600 dmu_tx_t *tx);
600 601 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
601 602 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
602 603 dmu_tx_t *tx);
603 604 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
604 605 dmu_tx_t *tx);
605 606 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
606 607 uint64_t size, struct page *pp, dmu_tx_t *tx);
607 608 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
608 609 void dmu_return_arcbuf(struct arc_buf *buf);
609 610 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
610 611 dmu_tx_t *tx);
611 612 int dmu_xuio_init(struct xuio *uio, int niov);
612 613 void dmu_xuio_fini(struct xuio *uio);
613 614 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
614 615 size_t n);
615 616 int dmu_xuio_cnt(struct xuio *uio);
616 617 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
617 618 void dmu_xuio_clear(struct xuio *uio, int i);
618 619 void xuio_stat_wbuf_copied();
619 620 void xuio_stat_wbuf_nocopy();
620 621
621 622 extern int zfs_prefetch_disable;
622 623
623 624 /*
624 625 * Asynchronously try to read in the data.
625 626 */
626 627 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
627 628 uint64_t len);
628 629
629 630 typedef struct dmu_object_info {
630 631 /* All sizes are in bytes unless otherwise indicated. */
631 632 uint32_t doi_data_block_size;
632 633 uint32_t doi_metadata_block_size;
633 634 dmu_object_type_t doi_type;
634 635 dmu_object_type_t doi_bonus_type;
635 636 uint64_t doi_bonus_size;
636 637 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
637 638 uint8_t doi_checksum;
638 639 uint8_t doi_compress;
639 640 uint8_t doi_pad[5];
640 641 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
641 642 uint64_t doi_max_offset;
642 643 uint64_t doi_fill_count; /* number of non-empty blocks */
643 644 } dmu_object_info_t;
644 645
645 646 typedef void arc_byteswap_func_t(void *buf, size_t size);
646 647
647 648 typedef struct dmu_object_type_info {
648 649 dmu_object_byteswap_t ot_byteswap;
649 650 boolean_t ot_metadata;
650 651 char *ot_name;
651 652 } dmu_object_type_info_t;
652 653
653 654 typedef struct dmu_object_byteswap_info {
654 655 arc_byteswap_func_t *ob_func;
655 656 char *ob_name;
656 657 } dmu_object_byteswap_info_t;
657 658
658 659 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
659 660 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
660 661
661 662 /*
662 663 * Get information on a DMU object.
663 664 *
664 665 * Return 0 on success or ENOENT if object is not allocated.
665 666 *
666 667 * If doi is NULL, just indicates whether the object exists.
667 668 */
668 669 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
669 670 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
670 671 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
671 672 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
672 673 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
673 674 /*
674 675 * Like dmu_object_info_from_db, but faster still when you only care about
675 676 * the size. This is specifically optimized for zfs_getattr().
676 677 */
677 678 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
678 679 u_longlong_t *nblk512);
679 680
680 681 typedef struct dmu_objset_stats {
681 682 uint64_t dds_num_clones; /* number of clones of this */
682 683 uint64_t dds_creation_txg;
683 684 uint64_t dds_guid;
684 685 dmu_objset_type_t dds_type;
685 686 uint8_t dds_is_snapshot;
686 687 uint8_t dds_inconsistent;
687 688 char dds_origin[MAXNAMELEN];
688 689 } dmu_objset_stats_t;
689 690
690 691 /*
691 692 * Get stats on a dataset.
692 693 */
693 694 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
694 695
695 696 /*
696 697 * Add entries to the nvlist for all the objset's properties. See
697 698 * zfs_prop_table[] and zfs(1m) for details on the properties.
698 699 */
699 700 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
700 701
701 702 /*
702 703 * Get the space usage statistics for statvfs().
703 704 *
704 705 * refdbytes is the amount of space "referenced" by this objset.
705 706 * availbytes is the amount of space available to this objset, taking
706 707 * into account quotas & reservations, assuming that no other objsets
707 708 * use the space first. These values correspond to the 'referenced' and
708 709 * 'available' properties, described in the zfs(1m) manpage.
709 710 *
710 711 * usedobjs and availobjs are the number of objects currently allocated,
711 712 * and available.
712 713 */
713 714 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
714 715 uint64_t *usedobjsp, uint64_t *availobjsp);
715 716
716 717 /*
717 718 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
718 719 * (Contrast with the ds_guid which is a 64-bit ID that will never
719 720 * change, so there is a small probability that it will collide.)
720 721 */
721 722 uint64_t dmu_objset_fsid_guid(objset_t *os);
722 723
723 724 /*
724 725 * Get the [cm]time for an objset's snapshot dir
725 726 */
726 727 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
727 728
728 729 int dmu_objset_is_snapshot(objset_t *os);
729 730
730 731 extern struct spa *dmu_objset_spa(objset_t *os);
731 732 extern struct zilog *dmu_objset_zil(objset_t *os);
732 733 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
733 734 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
734 735 extern void dmu_objset_name(objset_t *os, char *buf);
735 736 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
736 737 extern uint64_t dmu_objset_id(objset_t *os);
737 738 extern uint64_t dmu_objset_syncprop(objset_t *os);
738 739 extern uint64_t dmu_objset_logbias(objset_t *os);
739 740 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
740 741 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
741 742 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
742 743 int maxlen, boolean_t *conflict);
743 744 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
744 745 uint64_t *idp, uint64_t *offp);
745 746
746 747 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
747 748 void *bonus, uint64_t *userp, uint64_t *groupp);
748 749 extern void dmu_objset_register_type(dmu_objset_type_t ost,
749 750 objset_used_cb_t *cb);
750 751 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
751 752 extern void *dmu_objset_get_user(objset_t *os);
752 753
753 754 /*
754 755 * Return the txg number for the given assigned transaction.
755 756 */
756 757 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
757 758
758 759 /*
759 760 * Synchronous write.
760 761 * If a parent zio is provided this function initiates a write on the
761 762 * provided buffer as a child of the parent zio.
762 763 * In the absence of a parent zio, the write is completed synchronously.
763 764 * At write completion, blk is filled with the bp of the written block.
764 765 * Note that while the data covered by this function will be on stable
765 766 * storage when the write completes this new data does not become a
766 767 * permanent part of the file until the associated transaction commits.
767 768 */
768 769
769 770 /*
770 771 * {zfs,zvol,ztest}_get_done() args
771 772 */
772 773 typedef struct zgd {
773 774 struct zilog *zgd_zilog;
774 775 struct blkptr *zgd_bp;
775 776 dmu_buf_t *zgd_db;
776 777 struct rl *zgd_rl;
777 778 void *zgd_private;
778 779 } zgd_t;
779 780
780 781 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
781 782 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
782 783
783 784 /*
784 785 * Find the next hole or data block in file starting at *off
785 786 * Return found offset in *off. Return ESRCH for end of file.
786 787 */
787 788 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
788 789 uint64_t *off);
789 790
790 791 /*
791 792 * Initial setup and final teardown.
792 793 */
793 794 extern void dmu_init(void);
794 795 extern void dmu_fini(void);
795 796
796 797 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
797 798 uint64_t object, uint64_t offset, int len);
798 799 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
799 800 dmu_traverse_cb_t cb, void *arg);
800 801
801 802 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
802 803 struct vnode *vp, offset_t *offp);
803 804
804 805 /* CRC64 table */
805 806 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
806 807 extern uint64_t zfs_crc64_table[256];
807 808
808 809 #ifdef __cplusplus
809 810 }
810 811 #endif
811 812
812 813 #endif /* _SYS_DMU_H */
↓ open down ↓ |
580 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX