1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
27 * Copyright 2013 DEY Storage Systems, Inc.
28 * Copyright 2014 HybridCluster. All rights reserved.
29 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30 * Copyright 2013 Saso Kiselkov. All rights reserved.
31 * Copyright (c) 2014 Integros [integros.com]
32 */
33
34 /* Portions Copyright 2010 Robert Milkowski */
35
36 #ifndef _SYS_DMU_H
37 #define _SYS_DMU_H
38
39 /*
40 * This file describes the interface that the DMU provides for its
41 * consumers.
42 *
43 * The DMU also interacts with the SPA. That interface is described in
44 * dmu_spa.h.
45 */
46
47 #include <sys/zfs_context.h>
48 #include <sys/cred.h>
49 #include <sys/fs/zfs.h>
50 #include <sys/zio_compress.h>
51 #include <sys/zio_priority.h>
52
53 #ifdef __cplusplus
54 extern "C" {
55 #endif
56
57 struct uio;
58 struct xuio;
59 struct page;
60 struct vnode;
61 struct spa;
62 struct zilog;
63 struct zio;
64 struct blkptr;
65 struct zap_cursor;
66 struct dsl_dataset;
67 struct dsl_pool;
68 struct dnode;
69 struct drr_begin;
70 struct drr_end;
71 struct zbookmark_phys;
72 struct spa;
73 struct nvlist;
74 struct arc_buf;
75 struct zio_prop;
76 struct sa_handle;
77 struct file;
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 /*
239 * These flags are intended to be used to specify the "txg_how"
240 * parameter when calling the dmu_tx_assign() function. See the comment
241 * above dmu_tx_assign() for more details on the meaning of these flags.
242 */
243 #define TXG_NOWAIT (0ULL)
244 #define TXG_WAIT (1ULL<<0)
245 #define TXG_NOTHROTTLE (1ULL<<1)
246
247 void byteswap_uint64_array(void *buf, size_t size);
248 void byteswap_uint32_array(void *buf, size_t size);
249 void byteswap_uint16_array(void *buf, size_t size);
250 void byteswap_uint8_array(void *buf, size_t size);
251 void zap_byteswap(void *buf, size_t size);
252 void zfs_oldacl_byteswap(void *buf, size_t size);
253 void zfs_acl_byteswap(void *buf, size_t size);
254 void zfs_znode_byteswap(void *buf, size_t size);
255
256 #define DS_FIND_SNAPSHOTS (1<<0)
257 #define DS_FIND_CHILDREN (1<<1)
258 #define DS_FIND_SERIALIZE (1<<2)
259
260 /*
261 * The maximum number of bytes that can be accessed as part of one
262 * operation, including metadata.
263 */
264 #define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
265 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
266
267 #define DMU_USERUSED_OBJECT (-1ULL)
268 #define DMU_GROUPUSED_OBJECT (-2ULL)
269
270 /*
271 * artificial blkids for bonus buffer and spill blocks
272 */
273 #define DMU_BONUS_BLKID (-1ULL)
274 #define DMU_SPILL_BLKID (-2ULL)
275 /*
276 * Public routines to create, destroy, open, and close objsets.
277 */
278 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
279 int dmu_objset_own(const char *name, dmu_objset_type_t type,
280 boolean_t readonly, void *tag, objset_t **osp);
281 void dmu_objset_rele(objset_t *os, void *tag);
282 void dmu_objset_disown(objset_t *os, void *tag);
283 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
284
285 void dmu_objset_evict_dbufs(objset_t *os);
286 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
287 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
288 int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
289 struct nvlist *snaps);
290 int dmu_objset_clone(const char *name, const char *origin);
291 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
292 struct nvlist *errlist);
293 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
294 int dmu_objset_snapshot_tmp(const char *, const char *, int);
295 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
296 int flags);
297 void dmu_objset_byteswap(void *buf, size_t size);
298 int dsl_dataset_rename_snapshot(const char *fsname,
299 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
300 int dmu_objset_remap_indirects(const char *fsname);
301
302 typedef struct dmu_buf {
303 uint64_t db_object; /* object that this buffer is part of */
304 uint64_t db_offset; /* byte offset in this object */
305 uint64_t db_size; /* size of buffer in bytes */
306 void *db_data; /* data in buffer */
307 } dmu_buf_t;
308
309 /*
310 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
311 */
312 #define DMU_POOL_DIRECTORY_OBJECT 1
313 #define DMU_POOL_CONFIG "config"
314 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
315 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
316 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
317 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
318 #define DMU_POOL_ROOT_DATASET "root_dataset"
319 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
320 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
321 #define DMU_POOL_ERRLOG_LAST "errlog_last"
322 #define DMU_POOL_SPARES "spares"
323 #define DMU_POOL_DEFLATE "deflate"
324 #define DMU_POOL_HISTORY "history"
325 #define DMU_POOL_PROPS "pool_props"
326 #define DMU_POOL_L2CACHE "l2cache"
327 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
328 #define DMU_POOL_DDT "DDT-%s-%s-%s"
329 #define DMU_POOL_DDT_STATS "DDT-statistics"
330 #define DMU_POOL_CREATION_VERSION "creation_version"
331 #define DMU_POOL_SCAN "scan"
332 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
333 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
334 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
335 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
336 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
337 #define DMU_POOL_REMOVING "com.delphix:removing"
338 #define DMU_POOL_OBSOLETE_BPOBJ "com.delphix:obsolete_bpobj"
339 #define DMU_POOL_CONDENSING_INDIRECT "com.delphix:condensing_indirect"
340 #define DMU_POOL_ZPOOL_CHECKPOINT "com.delphix:zpool_checkpoint"
341
342 /*
343 * Allocate an object from this objset. The range of object numbers
344 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
345 *
346 * The transaction must be assigned to a txg. The newly allocated
347 * object will be "held" in the transaction (ie. you can modify the
348 * newly allocated object in this transaction).
349 *
350 * dmu_object_alloc() chooses an object and returns it in *objectp.
351 *
352 * dmu_object_claim() allocates a specific object number. If that
353 * number is already allocated, it fails and returns EEXIST.
354 *
355 * Return 0 on success, or ENOSPC or EEXIST as specified above.
356 */
357 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
358 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
359 uint64_t dmu_object_alloc_ibs(objset_t *os, dmu_object_type_t ot, int blocksize,
360 int indirect_blockshift,
361 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
362 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
363 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
364 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
365 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
366
367 /*
368 * Free an object from this objset.
369 *
370 * The object's data will be freed as well (ie. you don't need to call
371 * dmu_free(object, 0, -1, tx)).
372 *
373 * The object need not be held in the transaction.
374 *
375 * If there are any holds on this object's buffers (via dmu_buf_hold()),
376 * or tx holds on the object (via dmu_tx_hold_object()), you can not
377 * free it; it fails and returns EBUSY.
378 *
379 * If the object is not allocated, it fails and returns ENOENT.
380 *
381 * Return 0 on success, or EBUSY or ENOENT as specified above.
382 */
383 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
384
385 /*
386 * Find the next allocated or free object.
387 *
388 * The objectp parameter is in-out. It will be updated to be the next
389 * object which is allocated. Ignore objects which have not been
390 * modified since txg.
391 *
392 * XXX Can only be called on a objset with no dirty data.
393 *
394 * Returns 0 on success, or ENOENT if there are no more objects.
395 */
396 int dmu_object_next(objset_t *os, uint64_t *objectp,
397 boolean_t hole, uint64_t txg);
398
399 /*
400 * Set the data blocksize for an object.
401 *
402 * The object cannot have any blocks allcated beyond the first. If
403 * the first block is allocated already, the new size must be greater
404 * than the current block size. If these conditions are not met,
405 * ENOTSUP will be returned.
406 *
407 * Returns 0 on success, or EBUSY if there are any holds on the object
408 * contents, or ENOTSUP as described above.
409 */
410 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
411 int ibs, dmu_tx_t *tx);
412
413 /*
414 * Set the checksum property on a dnode. The new checksum algorithm will
415 * apply to all newly written blocks; existing blocks will not be affected.
416 */
417 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
418 dmu_tx_t *tx);
419
420 /*
421 * Set the compress property on a dnode. The new compression algorithm will
422 * apply to all newly written blocks; existing blocks will not be affected.
423 */
424 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
425 dmu_tx_t *tx);
426
427 int dmu_object_remap_indirects(objset_t *os, uint64_t object, uint64_t txg);
428
429 void
430 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
431 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
432 int compressed_size, int byteorder, dmu_tx_t *tx);
433
434 /*
435 * Decide how to write a block: checksum, compression, number of copies, etc.
436 */
437 #define WP_NOFILL 0x1
438 #define WP_DMU_SYNC 0x2
439 #define WP_SPILL 0x4
440
441 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
442 struct zio_prop *zp);
443 /*
444 * The bonus data is accessed more or less like a regular buffer.
445 * You must dmu_bonus_hold() to get the buffer, which will give you a
446 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
447 * data. As with any normal buffer, you must call dmu_buf_will_dirty()
448 * before modifying it, and the
449 * object must be held in an assigned transaction before calling
450 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
451 * buffer as well. You must release your hold with dmu_buf_rele().
452 *
453 * Returns ENOENT, EIO, or 0.
454 */
455 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
456 int dmu_bonus_max(void);
457 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
458 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
459 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
460 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
461
462 /*
463 * Special spill buffer support used by "SA" framework
464 */
465
466 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
467 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
468 void *tag, dmu_buf_t **dbp);
469 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
470
471 /*
472 * Obtain the DMU buffer from the specified object which contains the
473 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
474 * that it will remain in memory. You must release the hold with
475 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
476 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
477 *
478 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
479 * on the returned buffer before reading or writing the buffer's
480 * db_data. The comments for those routines describe what particular
481 * operations are valid after calling them.
482 *
483 * The object number must be a valid, allocated object number.
484 */
485 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
486 void *tag, dmu_buf_t **, int flags);
487 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
488 void *tag, dmu_buf_t **dbp, int flags);
489
490 /*
491 * Add a reference to a dmu buffer that has already been held via
492 * dmu_buf_hold() in the current context.
493 */
494 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
495
496 /*
497 * Attempt to add a reference to a dmu buffer that is in an unknown state,
498 * using a pointer that may have been invalidated by eviction processing.
499 * The request will succeed if the passed in dbuf still represents the
500 * same os/object/blkid, is ineligible for eviction, and has at least
501 * one hold by a user other than the syncer.
502 */
503 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
504 uint64_t blkid, void *tag);
505
506 void dmu_buf_rele(dmu_buf_t *db, void *tag);
507 uint64_t dmu_buf_refcount(dmu_buf_t *db);
508
509 /*
510 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
511 * range of an object. A pointer to an array of dmu_buf_t*'s is
512 * returned (in *dbpp).
513 *
514 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
515 * frees the array. The hold on the array of buffers MUST be released
516 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
517 * individually with dmu_buf_rele.
518 */
519 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
520 uint64_t length, boolean_t read, void *tag,
521 int *numbufsp, dmu_buf_t ***dbpp);
522 int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
523 boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp,
524 uint32_t flags);
525 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
526
527 typedef void dmu_buf_evict_func_t(void *user_ptr);
528
529 /*
530 * A DMU buffer user object may be associated with a dbuf for the
531 * duration of its lifetime. This allows the user of a dbuf (client)
532 * to attach private data to a dbuf (e.g. in-core only data such as a
533 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
534 * when that dbuf has been evicted. Clients typically respond to the
535 * eviction notification by freeing their private data, thus ensuring
536 * the same lifetime for both dbuf and private data.
537 *
538 * The mapping from a dmu_buf_user_t to any client private data is the
539 * client's responsibility. All current consumers of the API with private
540 * data embed a dmu_buf_user_t as the first member of the structure for
541 * their private data. This allows conversions between the two types
542 * with a simple cast. Since the DMU buf user API never needs access
543 * to the private data, other strategies can be employed if necessary
544 * or convenient for the client (e.g. using container_of() to do the
545 * conversion for private data that cannot have the dmu_buf_user_t as
546 * its first member).
547 *
548 * Eviction callbacks are executed without the dbuf mutex held or any
549 * other type of mechanism to guarantee that the dbuf is still available.
550 * For this reason, users must assume the dbuf has already been freed
551 * and not reference the dbuf from the callback context.
552 *
553 * Users requesting "immediate eviction" are notified as soon as the dbuf
554 * is only referenced by dirty records (dirties == holds). Otherwise the
555 * notification occurs after eviction processing for the dbuf begins.
556 */
557 typedef struct dmu_buf_user {
558 /*
559 * Asynchronous user eviction callback state.
560 */
561 taskq_ent_t dbu_tqent;
562
563 /*
564 * This instance's eviction function pointers.
565 *
566 * dbu_evict_func_sync is called synchronously and then
567 * dbu_evict_func_async is executed asynchronously on a taskq.
568 */
569 dmu_buf_evict_func_t *dbu_evict_func_sync;
570 dmu_buf_evict_func_t *dbu_evict_func_async;
571 #ifdef ZFS_DEBUG
572 /*
573 * Pointer to user's dbuf pointer. NULL for clients that do
574 * not associate a dbuf with their user data.
575 *
576 * The dbuf pointer is cleared upon eviction so as to catch
577 * use-after-evict bugs in clients.
578 */
579 dmu_buf_t **dbu_clear_on_evict_dbufp;
580 #endif
581 } dmu_buf_user_t;
582
583 /*
584 * Initialize the given dmu_buf_user_t instance with the eviction function
585 * evict_func, to be called when the user is evicted.
586 *
587 * NOTE: This function should only be called once on a given dmu_buf_user_t.
588 * To allow enforcement of this, dbu must already be zeroed on entry.
589 */
590 /*ARGSUSED*/
591 inline void
dmu_buf_init_user(dmu_buf_user_t * dbu,dmu_buf_evict_func_t * evict_func_sync,dmu_buf_evict_func_t * evict_func_async,dmu_buf_t ** clear_on_evict_dbufp)592 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
593 dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
594 {
595 ASSERT(dbu->dbu_evict_func_sync == NULL);
596 ASSERT(dbu->dbu_evict_func_async == NULL);
597
598 /* must have at least one evict func */
599 IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
600 dbu->dbu_evict_func_sync = evict_func_sync;
601 dbu->dbu_evict_func_async = evict_func_async;
602 #ifdef ZFS_DEBUG
603 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
604 #endif
605 }
606
607 /*
608 * Attach user data to a dbuf and mark it for normal (when the dbuf's
609 * data is cleared or its reference count goes to zero) eviction processing.
610 *
611 * Returns NULL on success, or the existing user if another user currently
612 * owns the buffer.
613 */
614 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
615
616 /*
617 * Attach user data to a dbuf and mark it for immediate (its dirty and
618 * reference counts are equal) eviction processing.
619 *
620 * Returns NULL on success, or the existing user if another user currently
621 * owns the buffer.
622 */
623 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
624
625 /*
626 * Replace the current user of a dbuf.
627 *
628 * If given the current user of a dbuf, replaces the dbuf's user with
629 * "new_user" and returns the user data pointer that was replaced.
630 * Otherwise returns the current, and unmodified, dbuf user pointer.
631 */
632 void *dmu_buf_replace_user(dmu_buf_t *db,
633 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
634
635 /*
636 * Remove the specified user data for a DMU buffer.
637 *
638 * Returns the user that was removed on success, or the current user if
639 * another user currently owns the buffer.
640 */
641 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
642
643 /*
644 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
645 */
646 void *dmu_buf_get_user(dmu_buf_t *db);
647
648 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
649 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
650 void dmu_buf_dnode_exit(dmu_buf_t *db);
651
652 /* Block until any in-progress dmu buf user evictions complete. */
653 void dmu_buf_user_evict_wait(void);
654
655 /*
656 * Returns the blkptr associated with this dbuf, or NULL if not set.
657 */
658 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
659
660 /*
661 * Indicate that you are going to modify the buffer's data (db_data).
662 *
663 * The transaction (tx) must be assigned to a txg (ie. you've called
664 * dmu_tx_assign()). The buffer's object must be held in the tx
665 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
666 */
667 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
668
669 /*
670 * You must create a transaction, then hold the objects which you will
671 * (or might) modify as part of this transaction. Then you must assign
672 * the transaction to a transaction group. Once the transaction has
673 * been assigned, you can modify buffers which belong to held objects as
674 * part of this transaction. You can't modify buffers before the
675 * transaction has been assigned; you can't modify buffers which don't
676 * belong to objects which this transaction holds; you can't hold
677 * objects once the transaction has been assigned. You may hold an
678 * object which you are going to free (with dmu_object_free()), but you
679 * don't have to.
680 *
681 * You can abort the transaction before it has been assigned.
682 *
683 * Note that you may hold buffers (with dmu_buf_hold) at any time,
684 * regardless of transaction state.
685 */
686
687 #define DMU_NEW_OBJECT (-1ULL)
688 #define DMU_OBJECT_END (-1ULL)
689
690 dmu_tx_t *dmu_tx_create(objset_t *os);
691 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
692 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
693 int len);
694 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
695 uint64_t len);
696 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
697 uint64_t len);
698 void dmu_tx_hold_remap_l1indirect(dmu_tx_t *tx, uint64_t object);
699 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
700 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
701 const char *name);
702 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
703 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
704 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
705 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
706 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
707 void dmu_tx_abort(dmu_tx_t *tx);
708 int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
709 void dmu_tx_wait(dmu_tx_t *tx);
710 void dmu_tx_commit(dmu_tx_t *tx);
711 void dmu_tx_mark_netfree(dmu_tx_t *tx);
712
713 /*
714 * To register a commit callback, dmu_tx_callback_register() must be called.
715 *
716 * dcb_data is a pointer to caller private data that is passed on as a
717 * callback parameter. The caller is responsible for properly allocating and
718 * freeing it.
719 *
720 * When registering a callback, the transaction must be already created, but
721 * it cannot be committed or aborted. It can be assigned to a txg or not.
722 *
723 * The callback will be called after the transaction has been safely written
724 * to stable storage and will also be called if the dmu_tx is aborted.
725 * If there is any error which prevents the transaction from being committed to
726 * disk, the callback will be called with a value of error != 0.
727 */
728 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
729
730 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
731 void *dcb_data);
732
733 /*
734 * Free up the data blocks for a defined range of a file. If size is
735 * -1, the range from offset to end-of-file is freed.
736 */
737 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
738 uint64_t size, dmu_tx_t *tx);
739 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
740 uint64_t size);
741 int dmu_free_long_object(objset_t *os, uint64_t object);
742
743 /*
744 * Convenience functions.
745 *
746 * Canfail routines will return 0 on success, or an errno if there is a
747 * nonrecoverable I/O error.
748 */
749 #define DMU_READ_PREFETCH 0 /* prefetch */
750 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
751 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
752 void *buf, uint32_t flags);
753 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
754 uint32_t flags);
755 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
756 const void *buf, dmu_tx_t *tx);
757 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
758 const void *buf, dmu_tx_t *tx);
759 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
760 dmu_tx_t *tx);
761 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
762 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
763 int dmu_read_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size);
764 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
765 dmu_tx_t *tx);
766 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
767 dmu_tx_t *tx);
768 int dmu_write_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size,
769 dmu_tx_t *tx);
770 #ifdef _KERNEL
771 #ifdef illumos
772 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
773 uint64_t size, struct page *pp, dmu_tx_t *tx);
774 #else
775 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
776 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx);
777 int dmu_read_pages(objset_t *os, uint64_t object, vm_page_t *ma, int count,
778 int *rbehind, int *rahead, int last_size);
779 #endif
780 #endif
781 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
782 void dmu_return_arcbuf(struct arc_buf *buf);
783 void dmu_assign_arcbuf_dnode(dnode_t *handle, uint64_t offset,
784 struct arc_buf *buf, dmu_tx_t *tx);
785 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
786 dmu_tx_t *tx);
787 int dmu_xuio_init(struct xuio *uio, int niov);
788 void dmu_xuio_fini(struct xuio *uio);
789 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
790 size_t n);
791 int dmu_xuio_cnt(struct xuio *uio);
792 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
793 void dmu_xuio_clear(struct xuio *uio, int i);
794 void xuio_stat_wbuf_copied(void);
795 void xuio_stat_wbuf_nocopy(void);
796
797 extern boolean_t zfs_prefetch_disable;
798 extern int zfs_max_recordsize;
799
800 /*
801 * Asynchronously try to read in the data.
802 */
803 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
804 uint64_t len, enum zio_priority pri);
805
806 typedef struct dmu_object_info {
807 /* All sizes are in bytes unless otherwise indicated. */
808 uint32_t doi_data_block_size;
809 uint32_t doi_metadata_block_size;
810 dmu_object_type_t doi_type;
811 dmu_object_type_t doi_bonus_type;
812 uint64_t doi_bonus_size;
813 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
814 uint8_t doi_checksum;
815 uint8_t doi_compress;
816 uint8_t doi_nblkptr;
817 uint8_t doi_pad[4];
818 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
819 uint64_t doi_max_offset;
820 uint64_t doi_fill_count; /* number of non-empty blocks */
821 } dmu_object_info_t;
822
823 typedef void arc_byteswap_func_t(void *buf, size_t size);
824
825 typedef struct dmu_object_type_info {
826 dmu_object_byteswap_t ot_byteswap;
827 boolean_t ot_metadata;
828 boolean_t ot_dbuf_metadata_cache;
829 char *ot_name;
830 } dmu_object_type_info_t;
831
832 typedef struct dmu_object_byteswap_info {
833 arc_byteswap_func_t *ob_func;
834 char *ob_name;
835 } dmu_object_byteswap_info_t;
836
837 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
838 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
839
840 /*
841 * Get information on a DMU object.
842 *
843 * Return 0 on success or ENOENT if object is not allocated.
844 *
845 * If doi is NULL, just indicates whether the object exists.
846 */
847 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
848 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
849 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
850 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
851 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
852 /*
853 * Like dmu_object_info_from_db, but faster still when you only care about
854 * the size. This is specifically optimized for zfs_getattr().
855 */
856 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
857 u_longlong_t *nblk512);
858
859 typedef struct dmu_objset_stats {
860 uint64_t dds_num_clones; /* number of clones of this */
861 uint64_t dds_creation_txg;
862 uint64_t dds_guid;
863 dmu_objset_type_t dds_type;
864 uint8_t dds_is_snapshot;
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 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
919 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
920 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
921 int maxlen, boolean_t *conflict);
922 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
923 uint64_t *idp, uint64_t *offp);
924
925 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
926 void *bonus, uint64_t *userp, uint64_t *groupp);
927 extern void dmu_objset_register_type(dmu_objset_type_t ost,
928 objset_used_cb_t *cb);
929 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
930 extern void *dmu_objset_get_user(objset_t *os);
931
932 /*
933 * Return the txg number for the given assigned transaction.
934 */
935 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
936
937 /*
938 * Synchronous write.
939 * If a parent zio is provided this function initiates a write on the
940 * provided buffer as a child of the parent zio.
941 * In the absence of a parent zio, the write is completed synchronously.
942 * At write completion, blk is filled with the bp of the written block.
943 * Note that while the data covered by this function will be on stable
944 * storage when the write completes this new data does not become a
945 * permanent part of the file until the associated transaction commits.
946 */
947
948 /*
949 * {zfs,zvol,ztest}_get_done() args
950 */
951 typedef struct zgd {
952 struct lwb *zgd_lwb;
953 struct blkptr *zgd_bp;
954 dmu_buf_t *zgd_db;
955 struct rl *zgd_rl;
956 void *zgd_private;
957 } zgd_t;
958
959 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
960 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
961
962 /*
963 * Find the next hole or data block in file starting at *off
964 * Return found offset in *off. Return ESRCH for end of file.
965 */
966 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
967 uint64_t *off);
968
969 /*
970 * Check if a DMU object has any dirty blocks. If so, sync out
971 * all pending transaction groups. Otherwise, this function
972 * does not alter DMU state. This could be improved to only sync
973 * out the necessary transaction groups for this particular
974 * object.
975 */
976 int dmu_object_wait_synced(objset_t *os, uint64_t object);
977
978 /*
979 * Initial setup and final teardown.
980 */
981 extern void dmu_init(void);
982 extern void dmu_fini(void);
983
984 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
985 uint64_t object, uint64_t offset, int len);
986 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
987 dmu_traverse_cb_t cb, void *arg);
988 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
989 struct file *fp, offset_t *offp);
990
991 /* CRC64 table */
992 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
993 extern uint64_t zfs_crc64_table[256];
994
995 extern int zfs_mdcomp_disable;
996
997 #ifdef __cplusplus
998 }
999 #endif
1000
1001 #endif /* _SYS_DMU_H */
1002