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, 2014 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.
32 /* Portions Copyright 2010 Robert Milkowski */
38 * This file describes the interface that the DMU provides for its
41 * The DMU also interacts with the SPA. That interface is described in
45 #include <sys/zfs_context.h>
47 #include <sys/fs/zfs.h>
48 #include <sys/zio_priority.h>
68 struct zbookmark_phys;
76 typedef struct objset objset_t;
77 typedef struct dmu_tx dmu_tx_t;
78 typedef struct dsl_dir dsl_dir_t;
80 typedef enum dmu_object_byteswap {
92 * Allocating a new byteswap type number makes the on-disk format
93 * incompatible with any other format that uses the same number.
95 * Data can usually be structured to work with one of the
96 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
99 } dmu_object_byteswap_t;
101 #define DMU_OT_NEWTYPE 0x80
102 #define DMU_OT_METADATA 0x40
103 #define DMU_OT_BYTESWAP_MASK 0x3f
106 * Defines a uint8_t object type. Object types specify if the data
107 * in the object is metadata (boolean) and how to byteswap the data
108 * (dmu_object_byteswap_t).
110 #define DMU_OT(byteswap, metadata) \
112 ((metadata) ? DMU_OT_METADATA : 0) | \
113 ((byteswap) & DMU_OT_BYTESWAP_MASK))
115 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
116 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
117 (ot) < DMU_OT_NUMTYPES)
119 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
120 ((ot) & DMU_OT_METADATA) : \
121 dmu_ot[(ot)].ot_metadata)
124 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
125 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
126 * is repurposed for embedded BPs.
128 #define DMU_OT_HAS_FILL(ot) \
129 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
131 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
132 ((ot) & DMU_OT_BYTESWAP_MASK) : \
133 dmu_ot[(ot)].ot_byteswap)
135 typedef enum dmu_object_type {
138 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
139 DMU_OT_OBJECT_ARRAY, /* UINT64 */
140 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
141 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
142 DMU_OT_BPOBJ, /* UINT64 */
143 DMU_OT_BPOBJ_HDR, /* UINT64 */
145 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
146 DMU_OT_SPACE_MAP, /* UINT64 */
148 DMU_OT_INTENT_LOG, /* UINT64 */
150 DMU_OT_DNODE, /* DNODE */
151 DMU_OT_OBJSET, /* OBJSET */
153 DMU_OT_DSL_DIR, /* UINT64 */
154 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
155 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
156 DMU_OT_DSL_PROPS, /* ZAP */
157 DMU_OT_DSL_DATASET, /* UINT64 */
159 DMU_OT_ZNODE, /* ZNODE */
160 DMU_OT_OLDACL, /* Old ACL */
161 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
162 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
163 DMU_OT_MASTER_NODE, /* ZAP */
164 DMU_OT_UNLINKED_SET, /* ZAP */
166 DMU_OT_ZVOL, /* UINT8 */
167 DMU_OT_ZVOL_PROP, /* ZAP */
168 /* other; for testing only! */
169 DMU_OT_PLAIN_OTHER, /* UINT8 */
170 DMU_OT_UINT64_OTHER, /* UINT64 */
171 DMU_OT_ZAP_OTHER, /* ZAP */
172 /* new object types: */
173 DMU_OT_ERROR_LOG, /* ZAP */
174 DMU_OT_SPA_HISTORY, /* UINT8 */
175 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
176 DMU_OT_POOL_PROPS, /* ZAP */
177 DMU_OT_DSL_PERMS, /* ZAP */
178 DMU_OT_ACL, /* ACL */
179 DMU_OT_SYSACL, /* SYSACL */
180 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
181 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
182 DMU_OT_NEXT_CLONES, /* ZAP */
183 DMU_OT_SCAN_QUEUE, /* ZAP */
184 DMU_OT_USERGROUP_USED, /* ZAP */
185 DMU_OT_USERGROUP_QUOTA, /* ZAP */
186 DMU_OT_USERREFS, /* ZAP */
187 DMU_OT_DDT_ZAP, /* ZAP */
188 DMU_OT_DDT_STATS, /* ZAP */
189 DMU_OT_SA, /* System attr */
190 DMU_OT_SA_MASTER_NODE, /* ZAP */
191 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
192 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
193 DMU_OT_SCAN_XLATE, /* ZAP */
194 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
195 DMU_OT_DEADLIST, /* ZAP */
196 DMU_OT_DEADLIST_HDR, /* UINT64 */
197 DMU_OT_DSL_CLONES, /* ZAP */
198 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
200 * Do not allocate new object types here. Doing so makes the on-disk
201 * format incompatible with any other format that uses the same object
204 * When creating an object which does not have one of the above types
205 * use the DMU_OTN_* type with the correct byteswap and metadata
208 * The DMU_OTN_* types do not have entries in the dmu_ot table,
209 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
210 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
211 * and DMU_OTN_* types).
216 * Names for valid types declared with DMU_OT().
218 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
219 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
220 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
221 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
222 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
223 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
224 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
225 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
226 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
227 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
230 typedef enum txg_how {
236 void byteswap_uint64_array(void *buf, size_t size);
237 void byteswap_uint32_array(void *buf, size_t size);
238 void byteswap_uint16_array(void *buf, size_t size);
239 void byteswap_uint8_array(void *buf, size_t size);
240 void zap_byteswap(void *buf, size_t size);
241 void zfs_oldacl_byteswap(void *buf, size_t size);
242 void zfs_acl_byteswap(void *buf, size_t size);
243 void zfs_znode_byteswap(void *buf, size_t size);
245 #define DS_FIND_SNAPSHOTS (1<<0)
246 #define DS_FIND_CHILDREN (1<<1)
247 #define DS_FIND_SERIALIZE (1<<2)
250 * The maximum number of bytes that can be accessed as part of one
251 * operation, including metadata.
253 #define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
254 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
256 #define DMU_USERUSED_OBJECT (-1ULL)
257 #define DMU_GROUPUSED_OBJECT (-2ULL)
260 * artificial blkids for bonus buffer and spill blocks
262 #define DMU_BONUS_BLKID (-1ULL)
263 #define DMU_SPILL_BLKID (-2ULL)
265 * Public routines to create, destroy, open, and close objsets.
267 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
268 int dmu_objset_own(const char *name, dmu_objset_type_t type,
269 boolean_t readonly, void *tag, objset_t **osp);
270 void dmu_objset_rele(objset_t *os, void *tag);
271 void dmu_objset_disown(objset_t *os, void *tag);
272 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
274 void dmu_objset_evict_dbufs(objset_t *os);
275 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
276 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
277 int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
278 struct nvlist *snaps);
279 int dmu_objset_clone(const char *name, const char *origin);
280 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
281 struct nvlist *errlist);
282 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
283 int dmu_objset_snapshot_tmp(const char *, const char *, int);
284 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
286 void dmu_objset_byteswap(void *buf, size_t size);
287 int dsl_dataset_rename_snapshot(const char *fsname,
288 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
290 typedef struct dmu_buf {
291 uint64_t db_object; /* object that this buffer is part of */
292 uint64_t db_offset; /* byte offset in this object */
293 uint64_t db_size; /* size of buffer in bytes */
294 void *db_data; /* data in buffer */
298 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
300 #define DMU_POOL_DIRECTORY_OBJECT 1
301 #define DMU_POOL_CONFIG "config"
302 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
303 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
304 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
305 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
306 #define DMU_POOL_ROOT_DATASET "root_dataset"
307 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
308 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
309 #define DMU_POOL_ERRLOG_LAST "errlog_last"
310 #define DMU_POOL_SPARES "spares"
311 #define DMU_POOL_DEFLATE "deflate"
312 #define DMU_POOL_HISTORY "history"
313 #define DMU_POOL_PROPS "pool_props"
314 #define DMU_POOL_L2CACHE "l2cache"
315 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
316 #define DMU_POOL_DDT "DDT-%s-%s-%s"
317 #define DMU_POOL_DDT_STATS "DDT-statistics"
318 #define DMU_POOL_CREATION_VERSION "creation_version"
319 #define DMU_POOL_SCAN "scan"
320 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
321 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
322 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
325 * Allocate an object from this objset. The range of object numbers
326 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
328 * The transaction must be assigned to a txg. The newly allocated
329 * object will be "held" in the transaction (ie. you can modify the
330 * newly allocated object in this transaction).
332 * dmu_object_alloc() chooses an object and returns it in *objectp.
334 * dmu_object_claim() allocates a specific object number. If that
335 * number is already allocated, it fails and returns EEXIST.
337 * Return 0 on success, or ENOSPC or EEXIST as specified above.
339 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
340 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
341 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
342 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
343 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
344 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
347 * Free an object from this objset.
349 * The object's data will be freed as well (ie. you don't need to call
350 * dmu_free(object, 0, -1, tx)).
352 * The object need not be held in the transaction.
354 * If there are any holds on this object's buffers (via dmu_buf_hold()),
355 * or tx holds on the object (via dmu_tx_hold_object()), you can not
356 * free it; it fails and returns EBUSY.
358 * If the object is not allocated, it fails and returns ENOENT.
360 * Return 0 on success, or EBUSY or ENOENT as specified above.
362 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
365 * Find the next allocated or free object.
367 * The objectp parameter is in-out. It will be updated to be the next
368 * object which is allocated. Ignore objects which have not been
369 * modified since txg.
371 * XXX Can only be called on a objset with no dirty data.
373 * Returns 0 on success, or ENOENT if there are no more objects.
375 int dmu_object_next(objset_t *os, uint64_t *objectp,
376 boolean_t hole, uint64_t txg);
379 * Set the data blocksize for an object.
381 * The object cannot have any blocks allcated beyond the first. If
382 * the first block is allocated already, the new size must be greater
383 * than the current block size. If these conditions are not met,
384 * ENOTSUP will be returned.
386 * Returns 0 on success, or EBUSY if there are any holds on the object
387 * contents, or ENOTSUP as described above.
389 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
390 int ibs, dmu_tx_t *tx);
393 * Set the checksum property on a dnode. The new checksum algorithm will
394 * apply to all newly written blocks; existing blocks will not be affected.
396 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
400 * Set the compress property on a dnode. The new compression algorithm will
401 * apply to all newly written blocks; existing blocks will not be affected.
403 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
407 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
408 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
409 int compressed_size, int byteorder, dmu_tx_t *tx);
412 * Decide how to write a block: checksum, compression, number of copies, etc.
414 #define WP_NOFILL 0x1
415 #define WP_DMU_SYNC 0x2
418 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
419 struct zio_prop *zp);
421 * The bonus data is accessed more or less like a regular buffer.
422 * You must dmu_bonus_hold() to get the buffer, which will give you a
423 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
424 * data. As with any normal buffer, you must call dmu_buf_read() to
425 * read db_data, dmu_buf_will_dirty() before modifying it, and the
426 * object must be held in an assigned transaction before calling
427 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
428 * buffer as well. You must release your hold with dmu_buf_rele().
430 * Returns ENOENT, EIO, or 0.
432 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
433 int dmu_bonus_max(void);
434 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
435 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
436 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
437 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
440 * Special spill buffer support used by "SA" framework
443 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
444 int dmu_spill_hold_by_dnode(struct dnode *dn, uint32_t flags,
445 void *tag, dmu_buf_t **dbp);
446 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
449 * Obtain the DMU buffer from the specified object which contains the
450 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
451 * that it will remain in memory. You must release the hold with
452 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
453 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
455 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
456 * on the returned buffer before reading or writing the buffer's
457 * db_data. The comments for those routines describe what particular
458 * operations are valid after calling them.
460 * The object number must be a valid, allocated object number.
462 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
463 void *tag, dmu_buf_t **, int flags);
466 * Add a reference to a dmu buffer that has already been held via
467 * dmu_buf_hold() in the current context.
469 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
472 * Attempt to add a reference to a dmu buffer that is in an unknown state,
473 * using a pointer that may have been invalidated by eviction processing.
474 * The request will succeed if the passed in dbuf still represents the
475 * same os/object/blkid, is ineligible for eviction, and has at least
476 * one hold by a user other than the syncer.
478 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
479 uint64_t blkid, void *tag);
481 void dmu_buf_rele(dmu_buf_t *db, void *tag);
482 uint64_t dmu_buf_refcount(dmu_buf_t *db);
485 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
486 * range of an object. A pointer to an array of dmu_buf_t*'s is
487 * returned (in *dbpp).
489 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
490 * frees the array. The hold on the array of buffers MUST be released
491 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
492 * individually with dmu_buf_rele.
494 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
495 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
496 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
498 typedef void dmu_buf_evict_func_t(void *user_ptr);
501 * A DMU buffer user object may be associated with a dbuf for the
502 * duration of its lifetime. This allows the user of a dbuf (client)
503 * to attach private data to a dbuf (e.g. in-core only data such as a
504 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
505 * when that dbuf has been evicted. Clients typically respond to the
506 * eviction notification by freeing their private data, thus ensuring
507 * the same lifetime for both dbuf and private data.
509 * The mapping from a dmu_buf_user_t to any client private data is the
510 * client's responsibility. All current consumers of the API with private
511 * data embed a dmu_buf_user_t as the first member of the structure for
512 * their private data. This allows conversions between the two types
513 * with a simple cast. Since the DMU buf user API never needs access
514 * to the private data, other strategies can be employed if necessary
515 * or convenient for the client (e.g. using container_of() to do the
516 * conversion for private data that cannot have the dmu_buf_user_t as
519 * Eviction callbacks are executed without the dbuf mutex held or any
520 * other type of mechanism to guarantee that the dbuf is still available.
521 * For this reason, users must assume the dbuf has already been freed
522 * and not reference the dbuf from the callback context.
524 * Users requesting "immediate eviction" are notified as soon as the dbuf
525 * is only referenced by dirty records (dirties == holds). Otherwise the
526 * notification occurs after eviction processing for the dbuf begins.
528 typedef struct dmu_buf_user {
530 * Asynchronous user eviction callback state.
532 taskq_ent_t dbu_tqent;
534 /* This instance's eviction function pointer. */
535 dmu_buf_evict_func_t *dbu_evict_func;
538 * Pointer to user's dbuf pointer. NULL for clients that do
539 * not associate a dbuf with their user data.
541 * The dbuf pointer is cleared upon eviction so as to catch
542 * use-after-evict bugs in clients.
544 dmu_buf_t **dbu_clear_on_evict_dbufp;
549 * Initialize the given dmu_buf_user_t instance with the eviction function
550 * evict_func, to be called when the user is evicted.
552 * NOTE: This function should only be called once on a given dmu_buf_user_t.
553 * To allow enforcement of this, dbu must already be zeroed on entry.
556 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
558 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
559 dmu_buf_t **clear_on_evict_dbufp);
562 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
563 dmu_buf_t **clear_on_evict_dbufp)
565 ASSERT(dbu->dbu_evict_func == NULL);
566 ASSERT(evict_func != NULL);
567 dbu->dbu_evict_func = evict_func;
569 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
575 * Attach user data to a dbuf and mark it for normal (when the dbuf's
576 * data is cleared or its reference count goes to zero) eviction processing.
578 * Returns NULL on success, or the existing user if another user currently
581 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
584 * Attach user data to a dbuf and mark it for immediate (its dirty and
585 * reference counts are equal) eviction processing.
587 * Returns NULL on success, or the existing user if another user currently
590 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
593 * Replace the current user of a dbuf.
595 * If given the current user of a dbuf, replaces the dbuf's user with
596 * "new_user" and returns the user data pointer that was replaced.
597 * Otherwise returns the current, and unmodified, dbuf user pointer.
599 void *dmu_buf_replace_user(dmu_buf_t *db,
600 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
603 * Remove the specified user data for a DMU buffer.
605 * Returns the user that was removed on success, or the current user if
606 * another user currently owns the buffer.
608 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
611 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
613 void *dmu_buf_get_user(dmu_buf_t *db);
615 /* Block until any in-progress dmu buf user evictions complete. */
616 void dmu_buf_user_evict_wait(void);
619 * Returns the blkptr associated with this dbuf, or NULL if not set.
621 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
624 * Indicate that you are going to modify the buffer's data (db_data).
626 * The transaction (tx) must be assigned to a txg (ie. you've called
627 * dmu_tx_assign()). The buffer's object must be held in the tx
628 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
630 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
633 * Tells if the given dbuf is freeable.
635 boolean_t dmu_buf_freeable(dmu_buf_t *);
638 * You must create a transaction, then hold the objects which you will
639 * (or might) modify as part of this transaction. Then you must assign
640 * the transaction to a transaction group. Once the transaction has
641 * been assigned, you can modify buffers which belong to held objects as
642 * part of this transaction. You can't modify buffers before the
643 * transaction has been assigned; you can't modify buffers which don't
644 * belong to objects which this transaction holds; you can't hold
645 * objects once the transaction has been assigned. You may hold an
646 * object which you are going to free (with dmu_object_free()), but you
649 * You can abort the transaction before it has been assigned.
651 * Note that you may hold buffers (with dmu_buf_hold) at any time,
652 * regardless of transaction state.
655 #define DMU_NEW_OBJECT (-1ULL)
656 #define DMU_OBJECT_END (-1ULL)
658 dmu_tx_t *dmu_tx_create(objset_t *os);
659 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
660 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
662 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
663 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
664 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
665 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
666 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
667 void dmu_tx_abort(dmu_tx_t *tx);
668 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
669 void dmu_tx_wait(dmu_tx_t *tx);
670 void dmu_tx_commit(dmu_tx_t *tx);
671 void dmu_tx_mark_netfree(dmu_tx_t *tx);
674 * To register a commit callback, dmu_tx_callback_register() must be called.
676 * dcb_data is a pointer to caller private data that is passed on as a
677 * callback parameter. The caller is responsible for properly allocating and
680 * When registering a callback, the transaction must be already created, but
681 * it cannot be committed or aborted. It can be assigned to a txg or not.
683 * The callback will be called after the transaction has been safely written
684 * to stable storage and will also be called if the dmu_tx is aborted.
685 * If there is any error which prevents the transaction from being committed to
686 * disk, the callback will be called with a value of error != 0.
688 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
690 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
694 * Free up the data blocks for a defined range of a file. If size is
695 * -1, the range from offset to end-of-file is freed.
697 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
698 uint64_t size, dmu_tx_t *tx);
699 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
701 int dmu_free_long_object(objset_t *os, uint64_t object);
704 * Convenience functions.
706 * Canfail routines will return 0 on success, or an errno if there is a
707 * nonrecoverable I/O error.
709 #define DMU_READ_PREFETCH 0 /* prefetch */
710 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
711 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
712 void *buf, uint32_t flags);
713 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
714 const void *buf, dmu_tx_t *tx);
715 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
717 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
718 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
719 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
721 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
725 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
726 uint64_t size, struct page *pp, dmu_tx_t *tx);
728 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
729 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx);
732 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
733 void dmu_return_arcbuf(struct arc_buf *buf);
734 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
736 int dmu_xuio_init(struct xuio *uio, int niov);
737 void dmu_xuio_fini(struct xuio *uio);
738 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
740 int dmu_xuio_cnt(struct xuio *uio);
741 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
742 void dmu_xuio_clear(struct xuio *uio, int i);
743 void xuio_stat_wbuf_copied();
744 void xuio_stat_wbuf_nocopy();
746 extern int zfs_prefetch_disable;
747 extern int zfs_max_recordsize;
750 * Asynchronously try to read in the data.
752 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
753 uint64_t len, enum zio_priority pri);
755 typedef struct dmu_object_info {
756 /* All sizes are in bytes unless otherwise indicated. */
757 uint32_t doi_data_block_size;
758 uint32_t doi_metadata_block_size;
759 dmu_object_type_t doi_type;
760 dmu_object_type_t doi_bonus_type;
761 uint64_t doi_bonus_size;
762 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
763 uint8_t doi_checksum;
764 uint8_t doi_compress;
767 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
768 uint64_t doi_max_offset;
769 uint64_t doi_fill_count; /* number of non-empty blocks */
772 typedef void arc_byteswap_func_t(void *buf, size_t size);
774 typedef struct dmu_object_type_info {
775 dmu_object_byteswap_t ot_byteswap;
776 boolean_t ot_metadata;
778 } dmu_object_type_info_t;
780 typedef struct dmu_object_byteswap_info {
781 arc_byteswap_func_t *ob_func;
783 } dmu_object_byteswap_info_t;
785 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
786 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
789 * Get information on a DMU object.
791 * Return 0 on success or ENOENT if object is not allocated.
793 * If doi is NULL, just indicates whether the object exists.
795 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
796 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
797 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
798 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
799 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
801 * Like dmu_object_info_from_db, but faster still when you only care about
802 * the size. This is specifically optimized for zfs_getattr().
804 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
805 u_longlong_t *nblk512);
807 typedef struct dmu_objset_stats {
808 uint64_t dds_num_clones; /* number of clones of this */
809 uint64_t dds_creation_txg;
811 dmu_objset_type_t dds_type;
812 uint8_t dds_is_snapshot;
813 uint8_t dds_inconsistent;
814 char dds_origin[MAXNAMELEN];
815 } dmu_objset_stats_t;
818 * Get stats on a dataset.
820 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
823 * Add entries to the nvlist for all the objset's properties. See
824 * zfs_prop_table[] and zfs(1m) for details on the properties.
826 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
829 * Get the space usage statistics for statvfs().
831 * refdbytes is the amount of space "referenced" by this objset.
832 * availbytes is the amount of space available to this objset, taking
833 * into account quotas & reservations, assuming that no other objsets
834 * use the space first. These values correspond to the 'referenced' and
835 * 'available' properties, described in the zfs(1m) manpage.
837 * usedobjs and availobjs are the number of objects currently allocated,
840 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
841 uint64_t *usedobjsp, uint64_t *availobjsp);
844 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
845 * (Contrast with the ds_guid which is a 64-bit ID that will never
846 * change, so there is a small probability that it will collide.)
848 uint64_t dmu_objset_fsid_guid(objset_t *os);
851 * Get the [cm]time for an objset's snapshot dir
853 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
855 int dmu_objset_is_snapshot(objset_t *os);
857 extern struct spa *dmu_objset_spa(objset_t *os);
858 extern struct zilog *dmu_objset_zil(objset_t *os);
859 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
860 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
861 extern void dmu_objset_name(objset_t *os, char *buf);
862 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
863 extern uint64_t dmu_objset_id(objset_t *os);
864 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
865 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
866 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
867 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
868 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
869 int maxlen, boolean_t *conflict);
870 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
871 uint64_t *idp, uint64_t *offp);
873 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
874 void *bonus, uint64_t *userp, uint64_t *groupp);
875 extern void dmu_objset_register_type(dmu_objset_type_t ost,
876 objset_used_cb_t *cb);
877 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
878 extern void *dmu_objset_get_user(objset_t *os);
881 * Return the txg number for the given assigned transaction.
883 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
887 * If a parent zio is provided this function initiates a write on the
888 * provided buffer as a child of the parent zio.
889 * In the absence of a parent zio, the write is completed synchronously.
890 * At write completion, blk is filled with the bp of the written block.
891 * Note that while the data covered by this function will be on stable
892 * storage when the write completes this new data does not become a
893 * permanent part of the file until the associated transaction commits.
897 * {zfs,zvol,ztest}_get_done() args
900 struct zilog *zgd_zilog;
901 struct blkptr *zgd_bp;
907 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
908 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
911 * Find the next hole or data block in file starting at *off
912 * Return found offset in *off. Return ESRCH for end of file.
914 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
918 * Initial setup and final teardown.
920 extern void dmu_init(void);
921 extern void dmu_fini(void);
923 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
924 uint64_t object, uint64_t offset, int len);
925 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
926 dmu_traverse_cb_t cb, void *arg);
927 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
928 struct file *fp, offset_t *offp);
931 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
932 extern uint64_t zfs_crc64_table[256];
934 extern int zfs_mdcomp_disable;
940 #endif /* _SYS_DMU_H */