1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
27 * Copyright 2013 DEY Storage Systems, Inc.
28 * Copyright 2014 HybridCluster. All rights reserved.
29 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30 * Copyright 2013 Saso Kiselkov. All rights reserved.
31 * Copyright (c) 2014 Integros [integros.com]
32 */
33
34 /* Portions Copyright 2010 Robert Milkowski */
35
36 #ifndef _SYS_DMU_H
37 #define _SYS_DMU_H
38
39 /*
40 * This file describes the interface that the DMU provides for its
41 * consumers.
42 *
43 * The DMU also interacts with the SPA. That interface is described in
44 * dmu_spa.h.
45 */
46
47 #include <sys/zfs_context.h>
48 #include <sys/inttypes.h>
49 #include <sys/cred.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/zio_compress.h>
52 #include <sys/zio_priority.h>
53
54 #ifdef __cplusplus
55 extern "C" {
56 #endif
57
58 struct uio;
59 struct xuio;
60 struct page;
61 struct vnode;
62 struct spa;
63 struct zilog;
64 struct zio;
65 struct blkptr;
66 struct zap_cursor;
67 struct dsl_dataset;
68 struct dsl_pool;
69 struct dnode;
70 struct drr_begin;
71 struct drr_end;
72 struct zbookmark_phys;
73 struct spa;
74 struct nvlist;
75 struct arc_buf;
76 struct zio_prop;
77 struct sa_handle;
78
79 typedef struct objset objset_t;
80 typedef struct dmu_tx dmu_tx_t;
81 typedef struct dsl_dir dsl_dir_t;
82 typedef struct dnode dnode_t;
83
84 typedef enum dmu_object_byteswap {
85 DMU_BSWAP_UINT8,
86 DMU_BSWAP_UINT16,
87 DMU_BSWAP_UINT32,
88 DMU_BSWAP_UINT64,
89 DMU_BSWAP_ZAP,
90 DMU_BSWAP_DNODE,
91 DMU_BSWAP_OBJSET,
92 DMU_BSWAP_ZNODE,
93 DMU_BSWAP_OLDACL,
94 DMU_BSWAP_ACL,
95 /*
96 * Allocating a new byteswap type number makes the on-disk format
97 * incompatible with any other format that uses the same number.
98 *
99 * Data can usually be structured to work with one of the
100 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
101 */
102 DMU_BSWAP_NUMFUNCS
103 } dmu_object_byteswap_t;
104
105 #define DMU_OT_NEWTYPE 0x80
106 #define DMU_OT_METADATA 0x40
107 #define DMU_OT_BYTESWAP_MASK 0x3f
108
109 /*
110 * Defines a uint8_t object type. Object types specify if the data
111 * in the object is metadata (boolean) and how to byteswap the data
112 * (dmu_object_byteswap_t). All of the types created by this method
113 * are cached in the dbuf metadata cache.
114 */
115 #define DMU_OT(byteswap, metadata) \
116 (DMU_OT_NEWTYPE | \
117 ((metadata) ? DMU_OT_METADATA : 0) | \
118 ((byteswap) & DMU_OT_BYTESWAP_MASK))
119
120 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
121 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
122 (ot) < DMU_OT_NUMTYPES)
123
124 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
125 ((ot) & DMU_OT_METADATA) : \
126 dmu_ot[(ot)].ot_metadata)
127
128 #define DMU_OT_IS_METADATA_CACHED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
129 B_TRUE : dmu_ot[(ot)].ot_dbuf_metadata_cache)
130
131 /*
132 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
133 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
134 * is repurposed for embedded BPs.
135 */
136 #define DMU_OT_HAS_FILL(ot) \
137 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
138
139 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
140 ((ot) & DMU_OT_BYTESWAP_MASK) : \
141 dmu_ot[(ot)].ot_byteswap)
142
143 typedef enum dmu_object_type {
144 DMU_OT_NONE,
145 /* general: */
146 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
147 DMU_OT_OBJECT_ARRAY, /* UINT64 */
148 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
149 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
150 DMU_OT_BPOBJ, /* UINT64 */
151 DMU_OT_BPOBJ_HDR, /* UINT64 */
152 /* spa: */
153 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
154 DMU_OT_SPACE_MAP, /* UINT64 */
155 /* zil: */
156 DMU_OT_INTENT_LOG, /* UINT64 */
157 /* dmu: */
158 DMU_OT_DNODE, /* DNODE */
159 DMU_OT_OBJSET, /* OBJSET */
160 /* dsl: */
161 DMU_OT_DSL_DIR, /* UINT64 */
162 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
163 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
164 DMU_OT_DSL_PROPS, /* ZAP */
165 DMU_OT_DSL_DATASET, /* UINT64 */
166 /* zpl: */
167 DMU_OT_ZNODE, /* ZNODE */
168 DMU_OT_OLDACL, /* Old ACL */
169 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
170 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
171 DMU_OT_MASTER_NODE, /* ZAP */
172 DMU_OT_UNLINKED_SET, /* ZAP */
173 /* zvol: */
174 DMU_OT_ZVOL, /* UINT8 */
175 DMU_OT_ZVOL_PROP, /* ZAP */
176 /* other; for testing only! */
177 DMU_OT_PLAIN_OTHER, /* UINT8 */
178 DMU_OT_UINT64_OTHER, /* UINT64 */
179 DMU_OT_ZAP_OTHER, /* ZAP */
180 /* new object types: */
181 DMU_OT_ERROR_LOG, /* ZAP */
182 DMU_OT_SPA_HISTORY, /* UINT8 */
183 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
184 DMU_OT_POOL_PROPS, /* ZAP */
185 DMU_OT_DSL_PERMS, /* ZAP */
186 DMU_OT_ACL, /* ACL */
187 DMU_OT_SYSACL, /* SYSACL */
188 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
189 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
190 DMU_OT_NEXT_CLONES, /* ZAP */
191 DMU_OT_SCAN_QUEUE, /* ZAP */
192 DMU_OT_USERGROUP_USED, /* ZAP */
193 DMU_OT_USERGROUP_QUOTA, /* ZAP */
194 DMU_OT_USERREFS, /* ZAP */
195 DMU_OT_DDT_ZAP, /* ZAP */
196 DMU_OT_DDT_STATS, /* ZAP */
197 DMU_OT_SA, /* System attr */
198 DMU_OT_SA_MASTER_NODE, /* ZAP */
199 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
200 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
201 DMU_OT_SCAN_XLATE, /* ZAP */
202 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
203 DMU_OT_DEADLIST, /* ZAP */
204 DMU_OT_DEADLIST_HDR, /* UINT64 */
205 DMU_OT_DSL_CLONES, /* ZAP */
206 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
207 /*
208 * Do not allocate new object types here. Doing so makes the on-disk
209 * format incompatible with any other format that uses the same object
210 * type number.
211 *
212 * When creating an object which does not have one of the above types
213 * use the DMU_OTN_* type with the correct byteswap and metadata
214 * values.
215 *
216 * The DMU_OTN_* types do not have entries in the dmu_ot table,
217 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
218 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
219 * and DMU_OTN_* types).
220 */
221 DMU_OT_NUMTYPES,
222
223 /*
224 * Names for valid types declared with DMU_OT().
225 */
226 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
227 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
228 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
229 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
230 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
231 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
232 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
233 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
234 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
235 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
236 } dmu_object_type_t;
237
238 typedef enum txg_how {
239 TXG_WAIT = 1,
240 TXG_NOWAIT,
241 TXG_WAITED,
242 } txg_how_t;
243
244 /*
245 * Selected classes of metadata
246 */
247 #define DMU_OT_IS_DDT_META(type) \
248 ((type == DMU_OT_DDT_ZAP) || \
249 (type == DMU_OT_DDT_STATS))
250
251 #define DMU_OT_IS_ZPL_META(type) \
252 ((type == DMU_OT_ZNODE) || \
253 (type == DMU_OT_OLDACL) || \
254 (type == DMU_OT_DIRECTORY_CONTENTS) || \
255 (type == DMU_OT_MASTER_NODE) || \
256 (type == DMU_OT_UNLINKED_SET))
257
258 void byteswap_uint64_array(void *buf, size_t size);
259 void byteswap_uint32_array(void *buf, size_t size);
260 void byteswap_uint16_array(void *buf, size_t size);
261 void byteswap_uint8_array(void *buf, size_t size);
262 void zap_byteswap(void *buf, size_t size);
263 void zfs_oldacl_byteswap(void *buf, size_t size);
264 void zfs_acl_byteswap(void *buf, size_t size);
265 void zfs_znode_byteswap(void *buf, size_t size);
266
267 #define DS_FIND_SNAPSHOTS (1<<0)
268 #define DS_FIND_CHILDREN (1<<1)
269 #define DS_FIND_SERIALIZE (1<<2)
270
271 /*
272 * The maximum number of bytes that can be accessed as part of one
273 * operation, including metadata.
274 */
275 #define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
276 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
277
278 #define DMU_USERUSED_OBJECT (-1ULL)
279 #define DMU_GROUPUSED_OBJECT (-2ULL)
280
281 /*
282 * artificial blkids for bonus buffer and spill blocks
283 */
284 #define DMU_BONUS_BLKID (-1ULL)
285 #define DMU_SPILL_BLKID (-2ULL)
286 /*
287 * Public routines to create, destroy, open, and close objsets.
288 */
289 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
290 int dmu_objset_own(const char *name, dmu_objset_type_t type,
291 boolean_t readonly, void *tag, objset_t **osp);
292 void dmu_objset_rele(objset_t *os, void *tag);
293 void dmu_objset_disown(objset_t *os, void *tag);
294 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
295
296 void dmu_objset_evict_dbufs(objset_t *os);
297 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
298 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
299 int dmu_objset_clone(const char *name, const char *origin);
300 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
301 struct nvlist *errlist);
302 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
303 int dmu_objset_snapshot_tmp(const char *, const char *, int);
304 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
305 int flags);
306 void dmu_objset_byteswap(void *buf, size_t size);
307 int dsl_dataset_rename_snapshot(const char *fsname,
308 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
309
310 typedef struct dmu_buf {
311 uint64_t db_object; /* object that this buffer is part of */
312 uint64_t db_offset; /* byte offset in this object */
313 uint64_t db_size; /* size of buffer in bytes */
314 void *db_data; /* data in buffer */
315 } dmu_buf_t;
316
317 /*
318 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
319 */
320 #define DMU_POOL_DIRECTORY_OBJECT 1
321 #define DMU_POOL_CONFIG "config"
322 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
323 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
324 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
325 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
326 #define DMU_POOL_ROOT_DATASET "root_dataset"
327 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
328 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
329 #define DMU_POOL_ERRLOG_LAST "errlog_last"
330 #define DMU_POOL_SPARES "spares"
331 #define DMU_POOL_DEFLATE "deflate"
332 #define DMU_POOL_HISTORY "history"
333 #define DMU_POOL_PROPS "pool_props"
334 #define DMU_POOL_L2CACHE "l2cache"
335 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
336 #define DMU_POOL_DDT "DDT-%s-%s-%s"
337 #define DMU_POOL_DDT_STATS "DDT-statistics"
338 #define DMU_POOL_CREATION_VERSION "creation_version"
339 #define DMU_POOL_SCAN "scan"
340 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
341 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
342 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
343 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
344 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
345
346 #define DMU_POOL_COS_PROPS "cos_props"
347 #define DMU_POOL_VDEV_PROPS "vdev_props"
348 #define DMU_POOL_TRIM_START_TIME "trim_start_time"
349 #define DMU_POOL_TRIM_STOP_TIME "trim_stop_time"
350
351 /*
352 * Allocate an object from this objset. The range of object numbers
353 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
354 *
355 * The transaction must be assigned to a txg. The newly allocated
356 * object will be "held" in the transaction (ie. you can modify the
357 * newly allocated object in this transaction).
358 *
359 * dmu_object_alloc() chooses an object and returns it in *objectp.
360 *
361 * dmu_object_claim() allocates a specific object number. If that
362 * number is already allocated, it fails and returns EEXIST.
363 *
364 * Return 0 on success, or ENOSPC or EEXIST as specified above.
365 */
366 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
367 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
368 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
369 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
370 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
371 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
372
373 /*
374 * Free an object from this objset.
375 *
376 * The object's data will be freed as well (ie. you don't need to call
377 * dmu_free(object, 0, -1, tx)).
378 *
379 * The object need not be held in the transaction.
380 *
381 * If there are any holds on this object's buffers (via dmu_buf_hold()),
382 * or tx holds on the object (via dmu_tx_hold_object()), you can not
383 * free it; it fails and returns EBUSY.
384 *
385 * If the object is not allocated, it fails and returns ENOENT.
386 *
387 * Return 0 on success, or EBUSY or ENOENT as specified above.
388 */
389 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
390
391 /*
392 * Find the next allocated or free object.
393 *
394 * The objectp parameter is in-out. It will be updated to be the next
395 * object which is allocated. Ignore objects which have not been
396 * modified since txg.
397 *
398 * XXX Can only be called on a objset with no dirty data.
399 *
400 * Returns 0 on success, or ENOENT if there are no more objects.
401 */
402 int dmu_object_next(objset_t *os, uint64_t *objectp,
403 boolean_t hole, uint64_t txg);
404
405 /*
406 * Set the data blocksize for an object.
407 *
408 * The object cannot have any blocks allcated beyond the first. If
409 * the first block is allocated already, the new size must be greater
410 * than the current block size. If these conditions are not met,
411 * ENOTSUP will be returned.
412 *
413 * Returns 0 on success, or EBUSY if there are any holds on the object
414 * contents, or ENOTSUP as described above.
415 */
416 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
417 int ibs, dmu_tx_t *tx);
418
419 /*
420 * Set the checksum property on a dnode. The new checksum algorithm will
421 * apply to all newly written blocks; existing blocks will not be affected.
422 */
423 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
424 dmu_tx_t *tx);
425
426 /*
427 * Set the compress property on a dnode. The new compression algorithm will
428 * apply to all newly written blocks; existing blocks will not be affected.
429 */
430 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
431 dmu_tx_t *tx);
432
433 void
434 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
435 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
436 int compressed_size, int byteorder, dmu_tx_t *tx);
437
438 /*
439 * Decide how to write a block: checksum, compression, number of copies, etc.
440 */
441 #define WP_NOFILL 0x1
442 #define WP_DMU_SYNC 0x2
443 #define WP_SPILL 0x4
444
445 #define WP_SPECIALCLASS_SHIFT (16)
446 #define WP_SPECIALCLASS_BITS (1) /* 1 bits per storage class */
447 #define WP_SPECIALCLASS_MASK (((1 << WP_SPECIALCLASS_BITS) - 1) \
448 << WP_SPECIALCLASS_SHIFT)
449
450 #define WP_SET_SPECIALCLASS(flags, sclass) { \
451 flags |= ((sclass << WP_SPECIALCLASS_SHIFT) & WP_SPECIALCLASS_MASK); \
452 }
453
454 #define WP_GET_SPECIALCLASS(flags) \
455 ((flags & WP_SPECIALCLASS_MASK) >> WP_SPECIALCLASS_SHIFT)
456
457 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
458 struct zio_prop *zp);
459 /*
460 * The bonus data is accessed more or less like a regular buffer.
461 * You must dmu_bonus_hold() to get the buffer, which will give you a
462 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
463 * data. As with any normal buffer, you must call dmu_buf_read() to
464 * read db_data, dmu_buf_will_dirty() before modifying it, and the
465 * object must be held in an assigned transaction before calling
466 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
467 * buffer as well. You must release your hold with dmu_buf_rele().
468 *
469 * Returns ENOENT, EIO, or 0.
470 */
471 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
472 int dmu_bonus_max(void);
473 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
474 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
475 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
476 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
477
478 /*
479 * Special spill buffer support used by "SA" framework
480 */
481
482 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
483 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
484 void *tag, dmu_buf_t **dbp);
485 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
486
487 /*
488 * Obtain the DMU buffer from the specified object which contains the
489 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
490 * that it will remain in memory. You must release the hold with
491 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
492 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
493 *
494 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
495 * on the returned buffer before reading or writing the buffer's
496 * db_data. The comments for those routines describe what particular
497 * operations are valid after calling them.
498 *
499 * The object number must be a valid, allocated object number.
500 */
501 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
502 void *tag, dmu_buf_t **, int flags);
503 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
504 void *tag, dmu_buf_t **dbp, int flags);
505
506 /*
507 * Add a reference to a dmu buffer that has already been held via
508 * dmu_buf_hold() in the current context.
509 */
510 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
511
512 /*
513 * Attempt to add a reference to a dmu buffer that is in an unknown state,
514 * using a pointer that may have been invalidated by eviction processing.
515 * The request will succeed if the passed in dbuf still represents the
516 * same os/object/blkid, is ineligible for eviction, and has at least
517 * one hold by a user other than the syncer.
518 */
519 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
520 uint64_t blkid, void *tag);
521
522 void dmu_buf_rele(dmu_buf_t *db, void *tag);
523 uint64_t dmu_buf_refcount(dmu_buf_t *db);
524
525 /*
526 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
527 * range of an object. A pointer to an array of dmu_buf_t*'s is
528 * returned (in *dbpp).
529 *
530 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
531 * frees the array. The hold on the array of buffers MUST be released
532 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
533 * individually with dmu_buf_rele.
534 */
535 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
536 uint64_t length, boolean_t read, void *tag,
537 int *numbufsp, dmu_buf_t ***dbpp);
538 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
539
540 typedef void dmu_buf_evict_func_t(void *user_ptr);
541
542 /*
543 * A DMU buffer user object may be associated with a dbuf for the
544 * duration of its lifetime. This allows the user of a dbuf (client)
545 * to attach private data to a dbuf (e.g. in-core only data such as a
546 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
547 * when that dbuf has been evicted. Clients typically respond to the
548 * eviction notification by freeing their private data, thus ensuring
549 * the same lifetime for both dbuf and private data.
550 *
551 * The mapping from a dmu_buf_user_t to any client private data is the
552 * client's responsibility. All current consumers of the API with private
553 * data embed a dmu_buf_user_t as the first member of the structure for
554 * their private data. This allows conversions between the two types
555 * with a simple cast. Since the DMU buf user API never needs access
556 * to the private data, other strategies can be employed if necessary
557 * or convenient for the client (e.g. using container_of() to do the
558 * conversion for private data that cannot have the dmu_buf_user_t as
559 * its first member).
560 *
561 * Eviction callbacks are executed without the dbuf mutex held or any
562 * other type of mechanism to guarantee that the dbuf is still available.
563 * For this reason, users must assume the dbuf has already been freed
564 * and not reference the dbuf from the callback context.
565 *
566 * Users requesting "immediate eviction" are notified as soon as the dbuf
567 * is only referenced by dirty records (dirties == holds). Otherwise the
568 * notification occurs after eviction processing for the dbuf begins.
569 */
570 typedef struct dmu_buf_user {
571 /*
572 * Asynchronous user eviction callback state.
573 */
574 taskq_ent_t dbu_tqent;
575
576 /*
577 * This instance's eviction function pointers.
578 *
579 * dbu_evict_func_sync is called synchronously and then
580 * dbu_evict_func_async is executed asynchronously on a taskq.
581 */
582 dmu_buf_evict_func_t *dbu_evict_func_sync;
583 dmu_buf_evict_func_t *dbu_evict_func_async;
584 #ifdef ZFS_DEBUG
585 /*
586 * Pointer to user's dbuf pointer. NULL for clients that do
587 * not associate a dbuf with their user data.
588 *
589 * The dbuf pointer is cleared upon eviction so as to catch
590 * use-after-evict bugs in clients.
591 */
592 dmu_buf_t **dbu_clear_on_evict_dbufp;
593 #endif
594 } dmu_buf_user_t;
595
596 /*
597 * Initialize the given dmu_buf_user_t instance with the eviction function
598 * evict_func, to be called when the user is evicted.
599 *
600 * NOTE: This function should only be called once on a given dmu_buf_user_t.
601 * To allow enforcement of this, dbu must already be zeroed on entry.
602 */
603 /*ARGSUSED*/
604 inline void
605 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
606 dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
607 {
608 ASSERT(dbu->dbu_evict_func_sync == NULL);
609 ASSERT(dbu->dbu_evict_func_async == NULL);
610
611 /* must have at least one evict func */
612 IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
613 dbu->dbu_evict_func_sync = evict_func_sync;
614 dbu->dbu_evict_func_async = evict_func_async;
615 #ifdef ZFS_DEBUG
616 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
617 #endif
618 }
619
620 /*
621 * Attach user data to a dbuf and mark it for normal (when the dbuf's
622 * data is cleared or its reference count goes to zero) eviction processing.
623 *
624 * Returns NULL on success, or the existing user if another user currently
625 * owns the buffer.
626 */
627 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
628
629 /*
630 * Attach user data to a dbuf and mark it for immediate (its dirty and
631 * reference counts are equal) eviction processing.
632 *
633 * Returns NULL on success, or the existing user if another user currently
634 * owns the buffer.
635 */
636 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
637
638 /*
639 * Replace the current user of a dbuf.
640 *
641 * If given the current user of a dbuf, replaces the dbuf's user with
642 * "new_user" and returns the user data pointer that was replaced.
643 * Otherwise returns the current, and unmodified, dbuf user pointer.
644 */
645 void *dmu_buf_replace_user(dmu_buf_t *db,
646 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
647
648 /*
649 * Remove the specified user data for a DMU buffer.
650 *
651 * Returns the user that was removed on success, or the current user if
652 * another user currently owns the buffer.
653 */
654 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
655
656 /*
657 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
658 */
659 void *dmu_buf_get_user(dmu_buf_t *db);
660
661 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
662 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
663 void dmu_buf_dnode_exit(dmu_buf_t *db);
664
665 /* Block until any in-progress dmu buf user evictions complete. */
666 void dmu_buf_user_evict_wait(void);
667
668 /*
669 * Returns the blkptr associated with this dbuf, or NULL if not set.
670 */
671 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
672
673 /*
674 * Indicate that you are going to modify the buffer's data (db_data).
675 *
676 * The transaction (tx) must be assigned to a txg (ie. you've called
677 * dmu_tx_assign()). The buffer's object must be held in the tx
678 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
679 */
680 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
681 void dmu_buf_will_dirty_sc(dmu_buf_t *db, dmu_tx_t *tx, boolean_t sc);
682
683 /*
684 * You must create a transaction, then hold the objects which you will
685 * (or might) modify as part of this transaction. Then you must assign
686 * the transaction to a transaction group. Once the transaction has
687 * been assigned, you can modify buffers which belong to held objects as
688 * part of this transaction. You can't modify buffers before the
689 * transaction has been assigned; you can't modify buffers which don't
690 * belong to objects which this transaction holds; you can't hold
691 * objects once the transaction has been assigned. You may hold an
692 * object which you are going to free (with dmu_object_free()), but you
693 * don't have to.
694 *
695 * You can abort the transaction before it has been assigned.
696 *
697 * Note that you may hold buffers (with dmu_buf_hold) at any time,
698 * regardless of transaction state.
699 */
700
701 #define DMU_NEW_OBJECT (-1ULL)
702 #define DMU_OBJECT_END (-1ULL)
703
704 dmu_tx_t *dmu_tx_create(objset_t *os);
705 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
706 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
707 int len);
708 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
709 uint64_t len);
710 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
711 uint64_t len);
712 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
713 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
714 const char *name);
715 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
716 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
717 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
718 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
719 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
720 void dmu_tx_abort(dmu_tx_t *tx);
721 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
722 void dmu_tx_wait(dmu_tx_t *tx);
723 void dmu_tx_commit(dmu_tx_t *tx);
724 void dmu_tx_mark_netfree(dmu_tx_t *tx);
725
726 /*
727 * To register a commit callback, dmu_tx_callback_register() must be called.
728 *
729 * dcb_data is a pointer to caller private data that is passed on as a
730 * callback parameter. The caller is responsible for properly allocating and
731 * freeing it.
732 *
733 * When registering a callback, the transaction must be already created, but
734 * it cannot be committed or aborted. It can be assigned to a txg or not.
735 *
736 * The callback will be called after the transaction has been safely written
737 * to stable storage and will also be called if the dmu_tx is aborted.
738 * If there is any error which prevents the transaction from being committed to
739 * disk, the callback will be called with a value of error != 0.
740 */
741 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
742
743 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
744 void *dcb_data);
745
746 /*
747 * Free up the data blocks for a defined range of a file. If size is
748 * -1, the range from offset to end-of-file is freed.
749 */
750 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
751 uint64_t size, dmu_tx_t *tx);
752 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
753 uint64_t size);
754 int dmu_free_long_object(objset_t *os, uint64_t object);
755
756 /*
757 * Convenience functions.
758 *
759 * Canfail routines will return 0 on success, or an errno if there is a
760 * nonrecoverable I/O error.
761 */
762 #define DMU_READ_PREFETCH 0 /* prefetch */
763 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
764 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
765 void *buf, uint32_t flags);
766 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
767 uint32_t flags);
768 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
769 const void *buf, dmu_tx_t *tx);
770 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
771 const void *buf, dmu_tx_t *tx);
772 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
773 dmu_tx_t *tx);
774 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
775 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
776 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
777 dmu_tx_t *tx);
778 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
779 dmu_tx_t *tx);
780 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
781 uint64_t size, struct page *pp, dmu_tx_t *tx);
782 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
783 void dmu_return_arcbuf(struct arc_buf *buf);
784 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
785 dmu_tx_t *tx);
786 int dmu_xuio_init(struct xuio *uio, int niov);
787 void dmu_xuio_fini(struct xuio *uio);
788 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
789 size_t n);
790 int dmu_xuio_cnt(struct xuio *uio);
791 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
792 void dmu_xuio_clear(struct xuio *uio, int i);
793 void xuio_stat_wbuf_copied(void);
794 void xuio_stat_wbuf_nocopy(void);
795
796 extern boolean_t zfs_prefetch_disable;
797 extern int zfs_max_recordsize;
798
799 /*
800 * Asynchronously try to read in the data.
801 */
802 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
803 uint64_t len, enum zio_priority pri);
804
805 typedef struct dmu_object_info {
806 /* All sizes are in bytes unless otherwise indicated. */
807 uint32_t doi_data_block_size;
808 uint32_t doi_metadata_block_size;
809 dmu_object_type_t doi_type;
810 dmu_object_type_t doi_bonus_type;
811 uint64_t doi_bonus_size;
812 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
813 uint8_t doi_checksum;
814 uint8_t doi_compress;
815 uint8_t doi_nblkptr;
816 uint8_t doi_pad[4];
817 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
818 uint64_t doi_max_offset;
819 uint64_t doi_fill_count; /* number of non-empty blocks */
820 } dmu_object_info_t;
821
822 typedef void arc_byteswap_func_t(void *buf, size_t size);
823
824 typedef struct dmu_object_type_info {
825 dmu_object_byteswap_t ot_byteswap;
826 boolean_t ot_metadata;
827 boolean_t ot_dbuf_metadata_cache;
828 char *ot_name;
829 } dmu_object_type_info_t;
830
831 typedef struct dmu_object_byteswap_info {
832 arc_byteswap_func_t *ob_func;
833 char *ob_name;
834 } dmu_object_byteswap_info_t;
835
836 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
837 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
838
839 /*
840 * Get information on a DMU object.
841 *
842 * Return 0 on success or ENOENT if object is not allocated.
843 *
844 * If doi is NULL, just indicates whether the object exists.
845 */
846 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
847 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
848 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
849 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
850 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
851 /*
852 * Like dmu_object_info_from_db, but faster still when you only care about
853 * the size. This is specifically optimized for zfs_getattr().
854 */
855 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
856 u_longlong_t *nblk512);
857
858 typedef struct dmu_objset_stats {
859 uint64_t dds_num_clones; /* number of clones of this */
860 uint64_t dds_creation_txg;
861 uint64_t dds_guid;
862 dmu_objset_type_t dds_type;
863 uint8_t dds_is_snapshot;
864 uint8_t dds_is_autosnapshot;
865 uint8_t dds_inconsistent;
866 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
867 } dmu_objset_stats_t;
868
869 /*
870 * Get stats on a dataset.
871 */
872 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
873
874 /*
875 * Add entries to the nvlist for all the objset's properties. See
876 * zfs_prop_table[] and zfs(1m) for details on the properties.
877 */
878 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
879
880 /*
881 * Get the space usage statistics for statvfs().
882 *
883 * refdbytes is the amount of space "referenced" by this objset.
884 * availbytes is the amount of space available to this objset, taking
885 * into account quotas & reservations, assuming that no other objsets
886 * use the space first. These values correspond to the 'referenced' and
887 * 'available' properties, described in the zfs(1m) manpage.
888 *
889 * usedobjs and availobjs are the number of objects currently allocated,
890 * and available.
891 */
892 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
893 uint64_t *usedobjsp, uint64_t *availobjsp);
894
895 /*
896 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
897 * (Contrast with the ds_guid which is a 64-bit ID that will never
898 * change, so there is a small probability that it will collide.)
899 */
900 uint64_t dmu_objset_fsid_guid(objset_t *os);
901
902 /*
903 * Get the [cm]time for an objset's snapshot dir
904 */
905 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
906
907 int dmu_objset_is_snapshot(objset_t *os);
908
909 extern struct spa *dmu_objset_spa(objset_t *os);
910 extern struct zilog *dmu_objset_zil(objset_t *os);
911 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
912 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
913 extern void dmu_objset_name(objset_t *os, char *buf);
914 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
915 extern uint64_t dmu_objset_id(objset_t *os);
916 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
917 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
918 int dmu_clone_list_next(objset_t *os, int len, char *name,
919 uint64_t *idp, uint64_t *offp);
920 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
921 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
922 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
923 int maxlen, boolean_t *conflict);
924 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
925 uint64_t *idp, uint64_t *offp);
926
927 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
928 void *bonus, uint64_t *userp, uint64_t *groupp);
929 extern void dmu_objset_register_type(dmu_objset_type_t ost,
930 objset_used_cb_t *cb);
931 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
932 extern void *dmu_objset_get_user(objset_t *os);
933
934 /*
935 * Return the txg number for the given assigned transaction.
936 */
937 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
938
939 /*
940 * Synchronous write.
941 * If a parent zio is provided this function initiates a write on the
942 * provided buffer as a child of the parent zio.
943 * In the absence of a parent zio, the write is completed synchronously.
944 * At write completion, blk is filled with the bp of the written block.
945 * Note that while the data covered by this function will be on stable
946 * storage when the write completes this new data does not become a
947 * permanent part of the file until the associated transaction commits.
948 */
949
950 /*
951 * {zfs,zvol,ztest}_get_done() args
952 */
953 typedef struct zgd {
954 struct lwb *zgd_lwb;
955 struct blkptr *zgd_bp;
956 dmu_buf_t *zgd_db;
957 struct rl *zgd_rl;
958 void *zgd_private;
959 } zgd_t;
960
961 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
962 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
963
964 /*
965 * Find the next hole or data block in file starting at *off
966 * Return found offset in *off. Return ESRCH for end of file.
967 */
968 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
969 uint64_t *off);
970
971 /*
972 * Check if a DMU object has any dirty blocks. If so, sync out
973 * all pending transaction groups. Otherwise, this function
974 * does not alter DMU state. This could be improved to only sync
975 * out the necessary transaction groups for this particular
976 * object.
977 */
978 int dmu_object_wait_synced(objset_t *os, uint64_t object);
979
980 /*
981 * Initial setup and final teardown.
982 */
983 extern void dmu_init(void);
984 extern void dmu_fini(void);
985
986 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
987 uint64_t object, uint64_t offset, int len);
988 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
989 dmu_traverse_cb_t cb, void *arg);
990
991 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
992 struct vnode *vp, offset_t *offp);
993
994 /* CRC64 table */
995 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
996 extern uint64_t zfs_crc64_table[256];
997
998 extern int zfs_mdcomp_disable;
999
1000 #ifdef __cplusplus
1001 }
1002 #endif
1003
1004 #endif /* _SYS_DMU_H */