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, 2016 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_priority.h>
70 struct zbookmark_phys;
78 typedef struct objset objset_t;
79 typedef struct dmu_tx dmu_tx_t;
80 typedef struct dsl_dir dsl_dir_t;
81 typedef struct dnode dnode_t;
83 typedef enum dmu_object_byteswap {
95 * Allocating a new byteswap type number makes the on-disk format
96 * incompatible with any other format that uses the same number.
98 * Data can usually be structured to work with one of the
99 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
102 } dmu_object_byteswap_t;
104 #define DMU_OT_NEWTYPE 0x80
105 #define DMU_OT_METADATA 0x40
106 #define DMU_OT_BYTESWAP_MASK 0x3f
109 * Defines a uint8_t object type. Object types specify if the data
110 * in the object is metadata (boolean) and how to byteswap the data
111 * (dmu_object_byteswap_t).
113 #define DMU_OT(byteswap, metadata) \
115 ((metadata) ? DMU_OT_METADATA : 0) | \
116 ((byteswap) & DMU_OT_BYTESWAP_MASK))
118 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
119 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
120 (ot) < DMU_OT_NUMTYPES)
122 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
123 ((ot) & DMU_OT_METADATA) : \
124 dmu_ot[(ot)].ot_metadata)
127 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
128 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
129 * is repurposed for embedded BPs.
131 #define DMU_OT_HAS_FILL(ot) \
132 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
134 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
135 ((ot) & DMU_OT_BYTESWAP_MASK) : \
136 dmu_ot[(ot)].ot_byteswap)
138 typedef enum dmu_object_type {
141 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
142 DMU_OT_OBJECT_ARRAY, /* UINT64 */
143 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
144 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
145 DMU_OT_BPOBJ, /* UINT64 */
146 DMU_OT_BPOBJ_HDR, /* UINT64 */
148 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
149 DMU_OT_SPACE_MAP, /* UINT64 */
151 DMU_OT_INTENT_LOG, /* UINT64 */
153 DMU_OT_DNODE, /* DNODE */
154 DMU_OT_OBJSET, /* OBJSET */
156 DMU_OT_DSL_DIR, /* UINT64 */
157 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
158 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
159 DMU_OT_DSL_PROPS, /* ZAP */
160 DMU_OT_DSL_DATASET, /* UINT64 */
162 DMU_OT_ZNODE, /* ZNODE */
163 DMU_OT_OLDACL, /* Old ACL */
164 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
165 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
166 DMU_OT_MASTER_NODE, /* ZAP */
167 DMU_OT_UNLINKED_SET, /* ZAP */
169 DMU_OT_ZVOL, /* UINT8 */
170 DMU_OT_ZVOL_PROP, /* ZAP */
171 /* other; for testing only! */
172 DMU_OT_PLAIN_OTHER, /* UINT8 */
173 DMU_OT_UINT64_OTHER, /* UINT64 */
174 DMU_OT_ZAP_OTHER, /* ZAP */
175 /* new object types: */
176 DMU_OT_ERROR_LOG, /* ZAP */
177 DMU_OT_SPA_HISTORY, /* UINT8 */
178 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
179 DMU_OT_POOL_PROPS, /* ZAP */
180 DMU_OT_DSL_PERMS, /* ZAP */
181 DMU_OT_ACL, /* ACL */
182 DMU_OT_SYSACL, /* SYSACL */
183 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
184 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
185 DMU_OT_NEXT_CLONES, /* ZAP */
186 DMU_OT_SCAN_QUEUE, /* ZAP */
187 DMU_OT_USERGROUP_USED, /* ZAP */
188 DMU_OT_USERGROUP_QUOTA, /* ZAP */
189 DMU_OT_USERREFS, /* ZAP */
190 DMU_OT_DDT_ZAP, /* ZAP */
191 DMU_OT_DDT_STATS, /* ZAP */
192 DMU_OT_SA, /* System attr */
193 DMU_OT_SA_MASTER_NODE, /* ZAP */
194 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
195 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
196 DMU_OT_SCAN_XLATE, /* ZAP */
197 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
198 DMU_OT_DEADLIST, /* ZAP */
199 DMU_OT_DEADLIST_HDR, /* UINT64 */
200 DMU_OT_DSL_CLONES, /* ZAP */
201 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
203 * Do not allocate new object types here. Doing so makes the on-disk
204 * format incompatible with any other format that uses the same object
207 * When creating an object which does not have one of the above types
208 * use the DMU_OTN_* type with the correct byteswap and metadata
211 * The DMU_OTN_* types do not have entries in the dmu_ot table,
212 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
213 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
214 * and DMU_OTN_* types).
219 * Names for valid types declared with DMU_OT().
221 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
222 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
223 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
224 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
225 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
226 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
227 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
228 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
229 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
230 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
233 typedef enum txg_how {
239 void byteswap_uint64_array(void *buf, size_t size);
240 void byteswap_uint32_array(void *buf, size_t size);
241 void byteswap_uint16_array(void *buf, size_t size);
242 void byteswap_uint8_array(void *buf, size_t size);
243 void zap_byteswap(void *buf, size_t size);
244 void zfs_oldacl_byteswap(void *buf, size_t size);
245 void zfs_acl_byteswap(void *buf, size_t size);
246 void zfs_znode_byteswap(void *buf, size_t size);
248 #define DS_FIND_SNAPSHOTS (1<<0)
249 #define DS_FIND_CHILDREN (1<<1)
250 #define DS_FIND_SERIALIZE (1<<2)
253 * The maximum number of bytes that can be accessed as part of one
254 * operation, including metadata.
256 #define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
257 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
259 #define DMU_USERUSED_OBJECT (-1ULL)
260 #define DMU_GROUPUSED_OBJECT (-2ULL)
263 * artificial blkids for bonus buffer and spill blocks
265 #define DMU_BONUS_BLKID (-1ULL)
266 #define DMU_SPILL_BLKID (-2ULL)
268 * Public routines to create, destroy, open, and close objsets.
270 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
271 int dmu_objset_own(const char *name, dmu_objset_type_t type,
272 boolean_t readonly, void *tag, objset_t **osp);
273 void dmu_objset_rele(objset_t *os, void *tag);
274 void dmu_objset_disown(objset_t *os, void *tag);
275 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
277 void dmu_objset_evict_dbufs(objset_t *os);
278 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
279 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
280 int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
281 struct nvlist *snaps);
282 int dmu_objset_clone(const char *name, const char *origin);
283 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
284 struct nvlist *errlist);
285 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
286 int dmu_objset_snapshot_tmp(const char *, const char *, int);
287 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
289 void dmu_objset_byteswap(void *buf, size_t size);
290 int dsl_dataset_rename_snapshot(const char *fsname,
291 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
293 typedef struct dmu_buf {
294 uint64_t db_object; /* object that this buffer is part of */
295 uint64_t db_offset; /* byte offset in this object */
296 uint64_t db_size; /* size of buffer in bytes */
297 void *db_data; /* data in buffer */
301 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
303 #define DMU_POOL_DIRECTORY_OBJECT 1
304 #define DMU_POOL_CONFIG "config"
305 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
306 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
307 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
308 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
309 #define DMU_POOL_ROOT_DATASET "root_dataset"
310 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
311 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
312 #define DMU_POOL_ERRLOG_LAST "errlog_last"
313 #define DMU_POOL_SPARES "spares"
314 #define DMU_POOL_DEFLATE "deflate"
315 #define DMU_POOL_HISTORY "history"
316 #define DMU_POOL_PROPS "pool_props"
317 #define DMU_POOL_L2CACHE "l2cache"
318 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
319 #define DMU_POOL_DDT "DDT-%s-%s-%s"
320 #define DMU_POOL_DDT_STATS "DDT-statistics"
321 #define DMU_POOL_CREATION_VERSION "creation_version"
322 #define DMU_POOL_SCAN "scan"
323 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
324 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
325 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
326 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
327 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
330 * Allocate an object from this objset. The range of object numbers
331 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
333 * The transaction must be assigned to a txg. The newly allocated
334 * object will be "held" in the transaction (ie. you can modify the
335 * newly allocated object in this transaction).
337 * dmu_object_alloc() chooses an object and returns it in *objectp.
339 * dmu_object_claim() allocates a specific object number. If that
340 * number is already allocated, it fails and returns EEXIST.
342 * Return 0 on success, or ENOSPC or EEXIST as specified above.
344 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
345 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
346 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
347 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
348 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
349 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
352 * Free an object from this objset.
354 * The object's data will be freed as well (ie. you don't need to call
355 * dmu_free(object, 0, -1, tx)).
357 * The object need not be held in the transaction.
359 * If there are any holds on this object's buffers (via dmu_buf_hold()),
360 * or tx holds on the object (via dmu_tx_hold_object()), you can not
361 * free it; it fails and returns EBUSY.
363 * If the object is not allocated, it fails and returns ENOENT.
365 * Return 0 on success, or EBUSY or ENOENT as specified above.
367 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
370 * Find the next allocated or free object.
372 * The objectp parameter is in-out. It will be updated to be the next
373 * object which is allocated. Ignore objects which have not been
374 * modified since txg.
376 * XXX Can only be called on a objset with no dirty data.
378 * Returns 0 on success, or ENOENT if there are no more objects.
380 int dmu_object_next(objset_t *os, uint64_t *objectp,
381 boolean_t hole, uint64_t txg);
384 * Set the data blocksize for an object.
386 * The object cannot have any blocks allcated beyond the first. If
387 * the first block is allocated already, the new size must be greater
388 * than the current block size. If these conditions are not met,
389 * ENOTSUP will be returned.
391 * Returns 0 on success, or EBUSY if there are any holds on the object
392 * contents, or ENOTSUP as described above.
394 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
395 int ibs, dmu_tx_t *tx);
398 * Set the checksum property on a dnode. The new checksum algorithm will
399 * apply to all newly written blocks; existing blocks will not be affected.
401 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
405 * Set the compress property on a dnode. The new compression algorithm will
406 * apply to all newly written blocks; existing blocks will not be affected.
408 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
412 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
413 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
414 int compressed_size, int byteorder, dmu_tx_t *tx);
417 * Decide how to write a block: checksum, compression, number of copies, etc.
419 #define WP_NOFILL 0x1
420 #define WP_DMU_SYNC 0x2
423 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
424 struct zio_prop *zp);
426 * The bonus data is accessed more or less like a regular buffer.
427 * You must dmu_bonus_hold() to get the buffer, which will give you a
428 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
429 * data. As with any normal buffer, you must call dmu_buf_read() to
430 * read db_data, dmu_buf_will_dirty() before modifying it, and the
431 * object must be held in an assigned transaction before calling
432 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
433 * buffer as well. You must release your hold with dmu_buf_rele().
435 * Returns ENOENT, EIO, or 0.
437 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
438 int dmu_bonus_max(void);
439 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
440 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
441 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
442 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
445 * Special spill buffer support used by "SA" framework
448 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
449 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
450 void *tag, dmu_buf_t **dbp);
451 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
454 * Obtain the DMU buffer from the specified object which contains the
455 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
456 * that it will remain in memory. You must release the hold with
457 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
458 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
460 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
461 * on the returned buffer before reading or writing the buffer's
462 * db_data. The comments for those routines describe what particular
463 * operations are valid after calling them.
465 * The object number must be a valid, allocated object number.
467 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
468 void *tag, dmu_buf_t **, int flags);
469 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
470 void *tag, dmu_buf_t **dbp, int flags);
473 * Add a reference to a dmu buffer that has already been held via
474 * dmu_buf_hold() in the current context.
476 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
479 * Attempt to add a reference to a dmu buffer that is in an unknown state,
480 * using a pointer that may have been invalidated by eviction processing.
481 * The request will succeed if the passed in dbuf still represents the
482 * same os/object/blkid, is ineligible for eviction, and has at least
483 * one hold by a user other than the syncer.
485 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
486 uint64_t blkid, void *tag);
488 void dmu_buf_rele(dmu_buf_t *db, void *tag);
489 uint64_t dmu_buf_refcount(dmu_buf_t *db);
492 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
493 * range of an object. A pointer to an array of dmu_buf_t*'s is
494 * returned (in *dbpp).
496 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
497 * frees the array. The hold on the array of buffers MUST be released
498 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
499 * individually with dmu_buf_rele.
501 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
502 uint64_t length, boolean_t read, void *tag,
503 int *numbufsp, dmu_buf_t ***dbpp);
504 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
506 typedef void dmu_buf_evict_func_t(void *user_ptr);
509 * A DMU buffer user object may be associated with a dbuf for the
510 * duration of its lifetime. This allows the user of a dbuf (client)
511 * to attach private data to a dbuf (e.g. in-core only data such as a
512 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
513 * when that dbuf has been evicted. Clients typically respond to the
514 * eviction notification by freeing their private data, thus ensuring
515 * the same lifetime for both dbuf and private data.
517 * The mapping from a dmu_buf_user_t to any client private data is the
518 * client's responsibility. All current consumers of the API with private
519 * data embed a dmu_buf_user_t as the first member of the structure for
520 * their private data. This allows conversions between the two types
521 * with a simple cast. Since the DMU buf user API never needs access
522 * to the private data, other strategies can be employed if necessary
523 * or convenient for the client (e.g. using container_of() to do the
524 * conversion for private data that cannot have the dmu_buf_user_t as
527 * Eviction callbacks are executed without the dbuf mutex held or any
528 * other type of mechanism to guarantee that the dbuf is still available.
529 * For this reason, users must assume the dbuf has already been freed
530 * and not reference the dbuf from the callback context.
532 * Users requesting "immediate eviction" are notified as soon as the dbuf
533 * is only referenced by dirty records (dirties == holds). Otherwise the
534 * notification occurs after eviction processing for the dbuf begins.
536 typedef struct dmu_buf_user {
538 * Asynchronous user eviction callback state.
540 taskq_ent_t dbu_tqent;
542 /* This instance's eviction function pointer. */
543 dmu_buf_evict_func_t *dbu_evict_func;
546 * Pointer to user's dbuf pointer. NULL for clients that do
547 * not associate a dbuf with their user data.
549 * The dbuf pointer is cleared upon eviction so as to catch
550 * use-after-evict bugs in clients.
552 dmu_buf_t **dbu_clear_on_evict_dbufp;
557 * Initialize the given dmu_buf_user_t instance with the eviction function
558 * evict_func, to be called when the user is evicted.
560 * NOTE: This function should only be called once on a given dmu_buf_user_t.
561 * To allow enforcement of this, dbu must already be zeroed on entry.
564 /* Very ugly, but it beats issuing suppression directives in many Makefiles. */
566 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
567 dmu_buf_t **clear_on_evict_dbufp);
570 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
571 dmu_buf_t **clear_on_evict_dbufp)
573 ASSERT(dbu->dbu_evict_func == NULL);
574 ASSERT(evict_func != NULL);
575 dbu->dbu_evict_func = evict_func;
577 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
583 * Attach user data to a dbuf and mark it for normal (when the dbuf's
584 * data is cleared or its reference count goes to zero) eviction processing.
586 * Returns NULL on success, or the existing user if another user currently
589 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
592 * Attach user data to a dbuf and mark it for immediate (its dirty and
593 * reference counts are equal) eviction processing.
595 * Returns NULL on success, or the existing user if another user currently
598 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
601 * Replace the current user of a dbuf.
603 * If given the current user of a dbuf, replaces the dbuf's user with
604 * "new_user" and returns the user data pointer that was replaced.
605 * Otherwise returns the current, and unmodified, dbuf user pointer.
607 void *dmu_buf_replace_user(dmu_buf_t *db,
608 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
611 * Remove the specified user data for a DMU buffer.
613 * Returns the user that was removed on success, or the current user if
614 * another user currently owns the buffer.
616 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
619 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
621 void *dmu_buf_get_user(dmu_buf_t *db);
623 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
624 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
625 void dmu_buf_dnode_exit(dmu_buf_t *db);
627 /* Block until any in-progress dmu buf user evictions complete. */
628 void dmu_buf_user_evict_wait(void);
631 * Returns the blkptr associated with this dbuf, or NULL if not set.
633 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
636 * Indicate that you are going to modify the buffer's data (db_data).
638 * The transaction (tx) must be assigned to a txg (ie. you've called
639 * dmu_tx_assign()). The buffer's object must be held in the tx
640 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
642 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
645 * Tells if the given dbuf is freeable.
647 boolean_t dmu_buf_freeable(dmu_buf_t *);
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_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
674 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
675 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
676 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
677 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
678 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
679 void dmu_tx_abort(dmu_tx_t *tx);
680 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
681 void dmu_tx_wait(dmu_tx_t *tx);
682 void dmu_tx_commit(dmu_tx_t *tx);
683 void dmu_tx_mark_netfree(dmu_tx_t *tx);
686 * To register a commit callback, dmu_tx_callback_register() must be called.
688 * dcb_data is a pointer to caller private data that is passed on as a
689 * callback parameter. The caller is responsible for properly allocating and
692 * When registering a callback, the transaction must be already created, but
693 * it cannot be committed or aborted. It can be assigned to a txg or not.
695 * The callback will be called after the transaction has been safely written
696 * to stable storage and will also be called if the dmu_tx is aborted.
697 * If there is any error which prevents the transaction from being committed to
698 * disk, the callback will be called with a value of error != 0.
700 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
702 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
706 * Free up the data blocks for a defined range of a file. If size is
707 * -1, the range from offset to end-of-file is freed.
709 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
710 uint64_t size, dmu_tx_t *tx);
711 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
713 int dmu_free_long_object(objset_t *os, uint64_t object);
716 * Convenience functions.
718 * Canfail routines will return 0 on success, or an errno if there is a
719 * nonrecoverable I/O error.
721 #define DMU_READ_PREFETCH 0 /* prefetch */
722 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
723 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
724 void *buf, uint32_t flags);
725 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
726 const void *buf, dmu_tx_t *tx);
727 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
729 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
730 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
731 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
733 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
737 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
738 uint64_t size, struct page *pp, dmu_tx_t *tx);
740 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
741 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx);
744 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
745 void dmu_return_arcbuf(struct arc_buf *buf);
746 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
748 int dmu_xuio_init(struct xuio *uio, int niov);
749 void dmu_xuio_fini(struct xuio *uio);
750 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
752 int dmu_xuio_cnt(struct xuio *uio);
753 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
754 void dmu_xuio_clear(struct xuio *uio, int i);
755 void xuio_stat_wbuf_copied();
756 void xuio_stat_wbuf_nocopy();
758 extern boolean_t zfs_prefetch_disable;
759 extern int zfs_max_recordsize;
762 * Asynchronously try to read in the data.
764 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
765 uint64_t len, enum zio_priority pri);
767 typedef struct dmu_object_info {
768 /* All sizes are in bytes unless otherwise indicated. */
769 uint32_t doi_data_block_size;
770 uint32_t doi_metadata_block_size;
771 dmu_object_type_t doi_type;
772 dmu_object_type_t doi_bonus_type;
773 uint64_t doi_bonus_size;
774 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
775 uint8_t doi_checksum;
776 uint8_t doi_compress;
779 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
780 uint64_t doi_max_offset;
781 uint64_t doi_fill_count; /* number of non-empty blocks */
784 typedef void arc_byteswap_func_t(void *buf, size_t size);
786 typedef struct dmu_object_type_info {
787 dmu_object_byteswap_t ot_byteswap;
788 boolean_t ot_metadata;
790 } dmu_object_type_info_t;
792 typedef struct dmu_object_byteswap_info {
793 arc_byteswap_func_t *ob_func;
795 } dmu_object_byteswap_info_t;
797 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
798 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
801 * Get information on a DMU object.
803 * Return 0 on success or ENOENT if object is not allocated.
805 * If doi is NULL, just indicates whether the object exists.
807 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
808 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
809 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
810 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
811 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
813 * Like dmu_object_info_from_db, but faster still when you only care about
814 * the size. This is specifically optimized for zfs_getattr().
816 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
817 u_longlong_t *nblk512);
819 typedef struct dmu_objset_stats {
820 uint64_t dds_num_clones; /* number of clones of this */
821 uint64_t dds_creation_txg;
823 dmu_objset_type_t dds_type;
824 uint8_t dds_is_snapshot;
825 uint8_t dds_inconsistent;
826 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
827 } dmu_objset_stats_t;
830 * Get stats on a dataset.
832 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
835 * Add entries to the nvlist for all the objset's properties. See
836 * zfs_prop_table[] and zfs(1m) for details on the properties.
838 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
841 * Get the space usage statistics for statvfs().
843 * refdbytes is the amount of space "referenced" by this objset.
844 * availbytes is the amount of space available to this objset, taking
845 * into account quotas & reservations, assuming that no other objsets
846 * use the space first. These values correspond to the 'referenced' and
847 * 'available' properties, described in the zfs(1m) manpage.
849 * usedobjs and availobjs are the number of objects currently allocated,
852 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
853 uint64_t *usedobjsp, uint64_t *availobjsp);
856 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
857 * (Contrast with the ds_guid which is a 64-bit ID that will never
858 * change, so there is a small probability that it will collide.)
860 uint64_t dmu_objset_fsid_guid(objset_t *os);
863 * Get the [cm]time for an objset's snapshot dir
865 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
867 int dmu_objset_is_snapshot(objset_t *os);
869 extern struct spa *dmu_objset_spa(objset_t *os);
870 extern struct zilog *dmu_objset_zil(objset_t *os);
871 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
872 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
873 extern void dmu_objset_name(objset_t *os, char *buf);
874 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
875 extern uint64_t dmu_objset_id(objset_t *os);
876 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
877 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
878 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
879 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
880 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
881 int maxlen, boolean_t *conflict);
882 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
883 uint64_t *idp, uint64_t *offp);
885 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
886 void *bonus, uint64_t *userp, uint64_t *groupp);
887 extern void dmu_objset_register_type(dmu_objset_type_t ost,
888 objset_used_cb_t *cb);
889 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
890 extern void *dmu_objset_get_user(objset_t *os);
893 * Return the txg number for the given assigned transaction.
895 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
899 * If a parent zio is provided this function initiates a write on the
900 * provided buffer as a child of the parent zio.
901 * In the absence of a parent zio, the write is completed synchronously.
902 * At write completion, blk is filled with the bp of the written block.
903 * Note that while the data covered by this function will be on stable
904 * storage when the write completes this new data does not become a
905 * permanent part of the file until the associated transaction commits.
909 * {zfs,zvol,ztest}_get_done() args
912 struct zilog *zgd_zilog;
913 struct blkptr *zgd_bp;
919 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
920 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
923 * Find the next hole or data block in file starting at *off
924 * Return found offset in *off. Return ESRCH for end of file.
926 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
930 * Check if a DMU object has any dirty blocks. If so, sync out
931 * all pending transaction groups. Otherwise, this function
932 * does not alter DMU state. This could be improved to only sync
933 * out the necessary transaction groups for this particular
936 int dmu_object_wait_synced(objset_t *os, uint64_t object);
939 * Initial setup and final teardown.
941 extern void dmu_init(void);
942 extern void dmu_fini(void);
944 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
945 uint64_t object, uint64_t offset, int len);
946 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
947 dmu_traverse_cb_t cb, void *arg);
948 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
949 struct file *fp, offset_t *offp);
952 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
953 extern uint64_t zfs_crc64_table[256];
955 extern int zfs_mdcomp_disable;
961 #endif /* _SYS_DMU_H */