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.
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.
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]
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]
34 /* Portions Copyright 2010 Robert Milkowski */
40 * This file describes the interface that the DMU provides for its
43 * The DMU also interacts with the SPA. That interface is described in
47 #include <sys/zfs_context.h>
49 #include <sys/fs/zfs.h>
50 #include <sys/zio_compress.h>
51 #include <sys/zio_priority.h>
71 struct zbookmark_phys;
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;
84 typedef enum dmu_object_byteswap {
96 * Allocating a new byteswap type number makes the on-disk format
97 * incompatible with any other format that uses the same number.
99 * Data can usually be structured to work with one of the
100 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
103 } dmu_object_byteswap_t;
105 #define DMU_OT_NEWTYPE 0x80
106 #define DMU_OT_METADATA 0x40
107 #define DMU_OT_BYTESWAP_MASK 0x3f
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).
114 #define DMU_OT(byteswap, metadata) \
116 ((metadata) ? DMU_OT_METADATA : 0) | \
117 ((byteswap) & DMU_OT_BYTESWAP_MASK))
119 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
120 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
121 (ot) < DMU_OT_NUMTYPES)
123 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
124 ((ot) & DMU_OT_METADATA) : \
125 dmu_ot[(ot)].ot_metadata)
128 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
129 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
130 * is repurposed for embedded BPs.
132 #define DMU_OT_HAS_FILL(ot) \
133 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
135 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
136 ((ot) & DMU_OT_BYTESWAP_MASK) : \
137 dmu_ot[(ot)].ot_byteswap)
139 typedef enum dmu_object_type {
142 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
143 DMU_OT_OBJECT_ARRAY, /* UINT64 */
144 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
145 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
146 DMU_OT_BPOBJ, /* UINT64 */
147 DMU_OT_BPOBJ_HDR, /* UINT64 */
149 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
150 DMU_OT_SPACE_MAP, /* UINT64 */
152 DMU_OT_INTENT_LOG, /* UINT64 */
154 DMU_OT_DNODE, /* DNODE */
155 DMU_OT_OBJSET, /* OBJSET */
157 DMU_OT_DSL_DIR, /* UINT64 */
158 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
159 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
160 DMU_OT_DSL_PROPS, /* ZAP */
161 DMU_OT_DSL_DATASET, /* UINT64 */
163 DMU_OT_ZNODE, /* ZNODE */
164 DMU_OT_OLDACL, /* Old ACL */
165 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
166 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
167 DMU_OT_MASTER_NODE, /* ZAP */
168 DMU_OT_UNLINKED_SET, /* ZAP */
170 DMU_OT_ZVOL, /* UINT8 */
171 DMU_OT_ZVOL_PROP, /* ZAP */
172 /* other; for testing only! */
173 DMU_OT_PLAIN_OTHER, /* UINT8 */
174 DMU_OT_UINT64_OTHER, /* UINT64 */
175 DMU_OT_ZAP_OTHER, /* ZAP */
176 /* new object types: */
177 DMU_OT_ERROR_LOG, /* ZAP */
178 DMU_OT_SPA_HISTORY, /* UINT8 */
179 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
180 DMU_OT_POOL_PROPS, /* ZAP */
181 DMU_OT_DSL_PERMS, /* ZAP */
182 DMU_OT_ACL, /* ACL */
183 DMU_OT_SYSACL, /* SYSACL */
184 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
185 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
186 DMU_OT_NEXT_CLONES, /* ZAP */
187 DMU_OT_SCAN_QUEUE, /* ZAP */
188 DMU_OT_USERGROUP_USED, /* ZAP */
189 DMU_OT_USERGROUP_QUOTA, /* ZAP */
190 DMU_OT_USERREFS, /* ZAP */
191 DMU_OT_DDT_ZAP, /* ZAP */
192 DMU_OT_DDT_STATS, /* ZAP */
193 DMU_OT_SA, /* System attr */
194 DMU_OT_SA_MASTER_NODE, /* ZAP */
195 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
196 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
197 DMU_OT_SCAN_XLATE, /* ZAP */
198 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
199 DMU_OT_DEADLIST, /* ZAP */
200 DMU_OT_DEADLIST_HDR, /* UINT64 */
201 DMU_OT_DSL_CLONES, /* ZAP */
202 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
204 * Do not allocate new object types here. Doing so makes the on-disk
205 * format incompatible with any other format that uses the same object
208 * When creating an object which does not have one of the above types
209 * use the DMU_OTN_* type with the correct byteswap and metadata
212 * The DMU_OTN_* types do not have entries in the dmu_ot table,
213 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
214 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
215 * and DMU_OTN_* types).
220 * Names for valid types declared with DMU_OT().
222 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
223 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
224 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
225 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
226 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
227 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
228 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
229 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
230 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
231 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
234 typedef enum txg_how {
240 void byteswap_uint64_array(void *buf, size_t size);
241 void byteswap_uint32_array(void *buf, size_t size);
242 void byteswap_uint16_array(void *buf, size_t size);
243 void byteswap_uint8_array(void *buf, size_t size);
244 void zap_byteswap(void *buf, size_t size);
245 void zfs_oldacl_byteswap(void *buf, size_t size);
246 void zfs_acl_byteswap(void *buf, size_t size);
247 void zfs_znode_byteswap(void *buf, size_t size);
249 #define DS_FIND_SNAPSHOTS (1<<0)
250 #define DS_FIND_CHILDREN (1<<1)
251 #define DS_FIND_SERIALIZE (1<<2)
254 * The maximum number of bytes that can be accessed as part of one
255 * operation, including metadata.
257 #define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
258 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
260 #define DMU_USERUSED_OBJECT (-1ULL)
261 #define DMU_GROUPUSED_OBJECT (-2ULL)
264 * artificial blkids for bonus buffer and spill blocks
266 #define DMU_BONUS_BLKID (-1ULL)
267 #define DMU_SPILL_BLKID (-2ULL)
269 * Public routines to create, destroy, open, and close objsets.
271 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
272 int dmu_objset_own(const char *name, dmu_objset_type_t type,
273 boolean_t readonly, void *tag, objset_t **osp);
274 void dmu_objset_rele(objset_t *os, void *tag);
275 void dmu_objset_disown(objset_t *os, void *tag);
276 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
278 void dmu_objset_evict_dbufs(objset_t *os);
279 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
280 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
281 int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
282 struct nvlist *snaps);
283 int dmu_objset_clone(const char *name, const char *origin);
284 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
285 struct nvlist *errlist);
286 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
287 int dmu_objset_snapshot_tmp(const char *, const char *, int);
288 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
290 void dmu_objset_byteswap(void *buf, size_t size);
291 int dsl_dataset_rename_snapshot(const char *fsname,
292 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
294 typedef struct dmu_buf {
295 uint64_t db_object; /* object that this buffer is part of */
296 uint64_t db_offset; /* byte offset in this object */
297 uint64_t db_size; /* size of buffer in bytes */
298 void *db_data; /* data in buffer */
302 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
304 #define DMU_POOL_DIRECTORY_OBJECT 1
305 #define DMU_POOL_CONFIG "config"
306 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
307 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
308 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
309 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
310 #define DMU_POOL_ROOT_DATASET "root_dataset"
311 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
312 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
313 #define DMU_POOL_ERRLOG_LAST "errlog_last"
314 #define DMU_POOL_SPARES "spares"
315 #define DMU_POOL_DEFLATE "deflate"
316 #define DMU_POOL_HISTORY "history"
317 #define DMU_POOL_PROPS "pool_props"
318 #define DMU_POOL_L2CACHE "l2cache"
319 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
320 #define DMU_POOL_DDT "DDT-%s-%s-%s"
321 #define DMU_POOL_DDT_STATS "DDT-statistics"
322 #define DMU_POOL_CREATION_VERSION "creation_version"
323 #define DMU_POOL_SCAN "scan"
324 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
325 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
326 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
327 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
328 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
331 * Allocate an object from this objset. The range of object numbers
332 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
334 * The transaction must be assigned to a txg. The newly allocated
335 * object will be "held" in the transaction (ie. you can modify the
336 * newly allocated object in this transaction).
338 * dmu_object_alloc() chooses an object and returns it in *objectp.
340 * dmu_object_claim() allocates a specific object number. If that
341 * number is already allocated, it fails and returns EEXIST.
343 * Return 0 on success, or ENOSPC or EEXIST as specified above.
345 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
346 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
347 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
348 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
349 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
350 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
353 * Free an object from this objset.
355 * The object's data will be freed as well (ie. you don't need to call
356 * dmu_free(object, 0, -1, tx)).
358 * The object need not be held in the transaction.
360 * If there are any holds on this object's buffers (via dmu_buf_hold()),
361 * or tx holds on the object (via dmu_tx_hold_object()), you can not
362 * free it; it fails and returns EBUSY.
364 * If the object is not allocated, it fails and returns ENOENT.
366 * Return 0 on success, or EBUSY or ENOENT as specified above.
368 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
371 * Find the next allocated or free object.
373 * The objectp parameter is in-out. It will be updated to be the next
374 * object which is allocated. Ignore objects which have not been
375 * modified since txg.
377 * XXX Can only be called on a objset with no dirty data.
379 * Returns 0 on success, or ENOENT if there are no more objects.
381 int dmu_object_next(objset_t *os, uint64_t *objectp,
382 boolean_t hole, uint64_t txg);
385 * Set the data blocksize for an object.
387 * The object cannot have any blocks allcated beyond the first. If
388 * the first block is allocated already, the new size must be greater
389 * than the current block size. If these conditions are not met,
390 * ENOTSUP will be returned.
392 * Returns 0 on success, or EBUSY if there are any holds on the object
393 * contents, or ENOTSUP as described above.
395 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
396 int ibs, dmu_tx_t *tx);
399 * Set the checksum property on a dnode. The new checksum algorithm will
400 * apply to all newly written blocks; existing blocks will not be affected.
402 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
406 * Set the compress property on a dnode. The new compression algorithm will
407 * apply to all newly written blocks; existing blocks will not be affected.
409 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
413 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
414 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
415 int compressed_size, int byteorder, dmu_tx_t *tx);
418 * Decide how to write a block: checksum, compression, number of copies, etc.
420 #define WP_NOFILL 0x1
421 #define WP_DMU_SYNC 0x2
424 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
425 struct zio_prop *zp);
427 * The bonus data is accessed more or less like a regular buffer.
428 * You must dmu_bonus_hold() to get the buffer, which will give you a
429 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
430 * data. As with any normal buffer, you must call dmu_buf_read() to
431 * read db_data, dmu_buf_will_dirty() before modifying it, and the
432 * object must be held in an assigned transaction before calling
433 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
434 * buffer as well. You must release your hold with dmu_buf_rele().
436 * Returns ENOENT, EIO, or 0.
438 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
439 int dmu_bonus_max(void);
440 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
441 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
442 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
443 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
446 * Special spill buffer support used by "SA" framework
449 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
450 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
451 void *tag, dmu_buf_t **dbp);
452 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
455 * Obtain the DMU buffer from the specified object which contains the
456 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
457 * that it will remain in memory. You must release the hold with
458 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
459 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
461 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
462 * on the returned buffer before reading or writing the buffer's
463 * db_data. The comments for those routines describe what particular
464 * operations are valid after calling them.
466 * The object number must be a valid, allocated object number.
468 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
469 void *tag, dmu_buf_t **, int flags);
470 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
471 void *tag, dmu_buf_t **dbp, int flags);
474 * Add a reference to a dmu buffer that has already been held via
475 * dmu_buf_hold() in the current context.
477 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
480 * Attempt to add a reference to a dmu buffer that is in an unknown state,
481 * using a pointer that may have been invalidated by eviction processing.
482 * The request will succeed if the passed in dbuf still represents the
483 * same os/object/blkid, is ineligible for eviction, and has at least
484 * one hold by a user other than the syncer.
486 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
487 uint64_t blkid, void *tag);
489 void dmu_buf_rele(dmu_buf_t *db, void *tag);
490 uint64_t dmu_buf_refcount(dmu_buf_t *db);
493 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
494 * range of an object. A pointer to an array of dmu_buf_t*'s is
495 * returned (in *dbpp).
497 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
498 * frees the array. The hold on the array of buffers MUST be released
499 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
500 * individually with dmu_buf_rele.
502 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
503 uint64_t length, boolean_t read, void *tag,
504 int *numbufsp, dmu_buf_t ***dbpp);
505 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
507 typedef void dmu_buf_evict_func_t(void *user_ptr);
510 * A DMU buffer user object may be associated with a dbuf for the
511 * duration of its lifetime. This allows the user of a dbuf (client)
512 * to attach private data to a dbuf (e.g. in-core only data such as a
513 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
514 * when that dbuf has been evicted. Clients typically respond to the
515 * eviction notification by freeing their private data, thus ensuring
516 * the same lifetime for both dbuf and private data.
518 * The mapping from a dmu_buf_user_t to any client private data is the
519 * client's responsibility. All current consumers of the API with private
520 * data embed a dmu_buf_user_t as the first member of the structure for
521 * their private data. This allows conversions between the two types
522 * with a simple cast. Since the DMU buf user API never needs access
523 * to the private data, other strategies can be employed if necessary
524 * or convenient for the client (e.g. using container_of() to do the
525 * conversion for private data that cannot have the dmu_buf_user_t as
528 * Eviction callbacks are executed without the dbuf mutex held or any
529 * other type of mechanism to guarantee that the dbuf is still available.
530 * For this reason, users must assume the dbuf has already been freed
531 * and not reference the dbuf from the callback context.
533 * Users requesting "immediate eviction" are notified as soon as the dbuf
534 * is only referenced by dirty records (dirties == holds). Otherwise the
535 * notification occurs after eviction processing for the dbuf begins.
537 typedef struct dmu_buf_user {
539 * Asynchronous user eviction callback state.
541 taskq_ent_t dbu_tqent;
544 * This instance's eviction function pointers.
546 * dbu_evict_func_sync is called synchronously and then
547 * dbu_evict_func_async is executed asynchronously on a taskq.
549 dmu_buf_evict_func_t *dbu_evict_func_sync;
550 dmu_buf_evict_func_t *dbu_evict_func_async;
553 * Pointer to user's dbuf pointer. NULL for clients that do
554 * not associate a dbuf with their user data.
556 * The dbuf pointer is cleared upon eviction so as to catch
557 * use-after-evict bugs in clients.
559 dmu_buf_t **dbu_clear_on_evict_dbufp;
564 * Initialize the given dmu_buf_user_t instance with the eviction function
565 * evict_func, to be called when the user is evicted.
567 * NOTE: This function should only be called once on a given dmu_buf_user_t.
568 * To allow enforcement of this, dbu must already be zeroed on entry.
572 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
573 dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
575 ASSERT(dbu->dbu_evict_func_sync == NULL);
576 ASSERT(dbu->dbu_evict_func_async == NULL);
578 /* must have at least one evict func */
579 IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
580 dbu->dbu_evict_func_sync = evict_func_sync;
581 dbu->dbu_evict_func_async = evict_func_async;
583 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
588 * Attach user data to a dbuf and mark it for normal (when the dbuf's
589 * data is cleared or its reference count goes to zero) eviction processing.
591 * Returns NULL on success, or the existing user if another user currently
594 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
597 * Attach user data to a dbuf and mark it for immediate (its dirty and
598 * reference counts are equal) eviction processing.
600 * Returns NULL on success, or the existing user if another user currently
603 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
606 * Replace the current user of a dbuf.
608 * If given the current user of a dbuf, replaces the dbuf's user with
609 * "new_user" and returns the user data pointer that was replaced.
610 * Otherwise returns the current, and unmodified, dbuf user pointer.
612 void *dmu_buf_replace_user(dmu_buf_t *db,
613 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
616 * Remove the specified user data for a DMU buffer.
618 * Returns the user that was removed on success, or the current user if
619 * another user currently owns the buffer.
621 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
624 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
626 void *dmu_buf_get_user(dmu_buf_t *db);
628 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
629 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
630 void dmu_buf_dnode_exit(dmu_buf_t *db);
632 /* Block until any in-progress dmu buf user evictions complete. */
633 void dmu_buf_user_evict_wait(void);
636 * Returns the blkptr associated with this dbuf, or NULL if not set.
638 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
641 * Indicate that you are going to modify the buffer's data (db_data).
643 * The transaction (tx) must be assigned to a txg (ie. you've called
644 * dmu_tx_assign()). The buffer's object must be held in the tx
645 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
647 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
650 * You must create a transaction, then hold the objects which you will
651 * (or might) modify as part of this transaction. Then you must assign
652 * the transaction to a transaction group. Once the transaction has
653 * been assigned, you can modify buffers which belong to held objects as
654 * part of this transaction. You can't modify buffers before the
655 * transaction has been assigned; you can't modify buffers which don't
656 * belong to objects which this transaction holds; you can't hold
657 * objects once the transaction has been assigned. You may hold an
658 * object which you are going to free (with dmu_object_free()), but you
661 * You can abort the transaction before it has been assigned.
663 * Note that you may hold buffers (with dmu_buf_hold) at any time,
664 * regardless of transaction state.
667 #define DMU_NEW_OBJECT (-1ULL)
668 #define DMU_OBJECT_END (-1ULL)
670 dmu_tx_t *dmu_tx_create(objset_t *os);
671 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
672 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
674 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
676 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
678 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
679 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
681 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
682 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
683 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
684 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
685 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
686 void dmu_tx_abort(dmu_tx_t *tx);
687 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
688 void dmu_tx_wait(dmu_tx_t *tx);
689 void dmu_tx_commit(dmu_tx_t *tx);
690 void dmu_tx_mark_netfree(dmu_tx_t *tx);
693 * To register a commit callback, dmu_tx_callback_register() must be called.
695 * dcb_data is a pointer to caller private data that is passed on as a
696 * callback parameter. The caller is responsible for properly allocating and
699 * When registering a callback, the transaction must be already created, but
700 * it cannot be committed or aborted. It can be assigned to a txg or not.
702 * The callback will be called after the transaction has been safely written
703 * to stable storage and will also be called if the dmu_tx is aborted.
704 * If there is any error which prevents the transaction from being committed to
705 * disk, the callback will be called with a value of error != 0.
707 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
709 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
713 * Free up the data blocks for a defined range of a file. If size is
714 * -1, the range from offset to end-of-file is freed.
716 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
717 uint64_t size, dmu_tx_t *tx);
718 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
720 int dmu_free_long_object(objset_t *os, uint64_t object);
723 * Convenience functions.
725 * Canfail routines will return 0 on success, or an errno if there is a
726 * nonrecoverable I/O error.
728 #define DMU_READ_PREFETCH 0 /* prefetch */
729 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
730 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
731 void *buf, uint32_t flags);
732 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
734 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
735 const void *buf, dmu_tx_t *tx);
736 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
737 const void *buf, dmu_tx_t *tx);
738 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
740 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
741 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
742 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
744 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
748 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
749 uint64_t size, struct page *pp, dmu_tx_t *tx);
751 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
752 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx);
755 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
756 void dmu_return_arcbuf(struct arc_buf *buf);
757 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
759 int dmu_xuio_init(struct xuio *uio, int niov);
760 void dmu_xuio_fini(struct xuio *uio);
761 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
763 int dmu_xuio_cnt(struct xuio *uio);
764 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
765 void dmu_xuio_clear(struct xuio *uio, int i);
766 void xuio_stat_wbuf_copied(void);
767 void xuio_stat_wbuf_nocopy(void);
769 extern boolean_t zfs_prefetch_disable;
770 extern int zfs_max_recordsize;
773 * Asynchronously try to read in the data.
775 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
776 uint64_t len, enum zio_priority pri);
778 typedef struct dmu_object_info {
779 /* All sizes are in bytes unless otherwise indicated. */
780 uint32_t doi_data_block_size;
781 uint32_t doi_metadata_block_size;
782 dmu_object_type_t doi_type;
783 dmu_object_type_t doi_bonus_type;
784 uint64_t doi_bonus_size;
785 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
786 uint8_t doi_checksum;
787 uint8_t doi_compress;
790 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
791 uint64_t doi_max_offset;
792 uint64_t doi_fill_count; /* number of non-empty blocks */
795 typedef void arc_byteswap_func_t(void *buf, size_t size);
797 typedef struct dmu_object_type_info {
798 dmu_object_byteswap_t ot_byteswap;
799 boolean_t ot_metadata;
801 } dmu_object_type_info_t;
803 typedef struct dmu_object_byteswap_info {
804 arc_byteswap_func_t *ob_func;
806 } dmu_object_byteswap_info_t;
808 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
809 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
812 * Get information on a DMU object.
814 * Return 0 on success or ENOENT if object is not allocated.
816 * If doi is NULL, just indicates whether the object exists.
818 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
819 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
820 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
821 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
822 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
824 * Like dmu_object_info_from_db, but faster still when you only care about
825 * the size. This is specifically optimized for zfs_getattr().
827 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
828 u_longlong_t *nblk512);
830 typedef struct dmu_objset_stats {
831 uint64_t dds_num_clones; /* number of clones of this */
832 uint64_t dds_creation_txg;
834 dmu_objset_type_t dds_type;
835 uint8_t dds_is_snapshot;
836 uint8_t dds_inconsistent;
837 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
838 } dmu_objset_stats_t;
841 * Get stats on a dataset.
843 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
846 * Add entries to the nvlist for all the objset's properties. See
847 * zfs_prop_table[] and zfs(1m) for details on the properties.
849 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
852 * Get the space usage statistics for statvfs().
854 * refdbytes is the amount of space "referenced" by this objset.
855 * availbytes is the amount of space available to this objset, taking
856 * into account quotas & reservations, assuming that no other objsets
857 * use the space first. These values correspond to the 'referenced' and
858 * 'available' properties, described in the zfs(1m) manpage.
860 * usedobjs and availobjs are the number of objects currently allocated,
863 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
864 uint64_t *usedobjsp, uint64_t *availobjsp);
867 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
868 * (Contrast with the ds_guid which is a 64-bit ID that will never
869 * change, so there is a small probability that it will collide.)
871 uint64_t dmu_objset_fsid_guid(objset_t *os);
874 * Get the [cm]time for an objset's snapshot dir
876 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
878 int dmu_objset_is_snapshot(objset_t *os);
880 extern struct spa *dmu_objset_spa(objset_t *os);
881 extern struct zilog *dmu_objset_zil(objset_t *os);
882 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
883 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
884 extern void dmu_objset_name(objset_t *os, char *buf);
885 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
886 extern uint64_t dmu_objset_id(objset_t *os);
887 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
888 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
889 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
890 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
891 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
892 int maxlen, boolean_t *conflict);
893 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
894 uint64_t *idp, uint64_t *offp);
896 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
897 void *bonus, uint64_t *userp, uint64_t *groupp);
898 extern void dmu_objset_register_type(dmu_objset_type_t ost,
899 objset_used_cb_t *cb);
900 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
901 extern void *dmu_objset_get_user(objset_t *os);
904 * Return the txg number for the given assigned transaction.
906 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
910 * If a parent zio is provided this function initiates a write on the
911 * provided buffer as a child of the parent zio.
912 * In the absence of a parent zio, the write is completed synchronously.
913 * At write completion, blk is filled with the bp of the written block.
914 * Note that while the data covered by this function will be on stable
915 * storage when the write completes this new data does not become a
916 * permanent part of the file until the associated transaction commits.
920 * {zfs,zvol,ztest}_get_done() args
923 struct zilog *zgd_zilog;
924 struct blkptr *zgd_bp;
930 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
931 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
934 * Find the next hole or data block in file starting at *off
935 * Return found offset in *off. Return ESRCH for end of file.
937 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
941 * Check if a DMU object has any dirty blocks. If so, sync out
942 * all pending transaction groups. Otherwise, this function
943 * does not alter DMU state. This could be improved to only sync
944 * out the necessary transaction groups for this particular
947 int dmu_object_wait_synced(objset_t *os, uint64_t object);
950 * Initial setup and final teardown.
952 extern void dmu_init(void);
953 extern void dmu_fini(void);
955 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
956 uint64_t object, uint64_t offset, int len);
957 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
958 dmu_traverse_cb_t cb, void *arg);
959 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
960 struct file *fp, offset_t *offp);
963 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
964 extern uint64_t zfs_crc64_table[256];
966 extern int zfs_mdcomp_disable;
972 #endif /* _SYS_DMU_H */