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),
235 * These flags are intended to be used to specify the "txg_how"
236 * parameter when calling the dmu_tx_assign() function. See the comment
237 * above dmu_tx_assign() for more details on the meaning of these flags.
239 #define TXG_NOWAIT (0ULL)
240 #define TXG_WAIT (1ULL<<0)
241 #define TXG_NOTHROTTLE (1ULL<<1)
243 void byteswap_uint64_array(void *buf, size_t size);
244 void byteswap_uint32_array(void *buf, size_t size);
245 void byteswap_uint16_array(void *buf, size_t size);
246 void byteswap_uint8_array(void *buf, size_t size);
247 void zap_byteswap(void *buf, size_t size);
248 void zfs_oldacl_byteswap(void *buf, size_t size);
249 void zfs_acl_byteswap(void *buf, size_t size);
250 void zfs_znode_byteswap(void *buf, size_t size);
252 #define DS_FIND_SNAPSHOTS (1<<0)
253 #define DS_FIND_CHILDREN (1<<1)
254 #define DS_FIND_SERIALIZE (1<<2)
257 * The maximum number of bytes that can be accessed as part of one
258 * operation, including metadata.
260 #define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
261 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
263 #define DMU_USERUSED_OBJECT (-1ULL)
264 #define DMU_GROUPUSED_OBJECT (-2ULL)
267 * artificial blkids for bonus buffer and spill blocks
269 #define DMU_BONUS_BLKID (-1ULL)
270 #define DMU_SPILL_BLKID (-2ULL)
272 * Public routines to create, destroy, open, and close objsets.
274 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
275 int dmu_objset_own(const char *name, dmu_objset_type_t type,
276 boolean_t readonly, void *tag, objset_t **osp);
277 void dmu_objset_rele(objset_t *os, void *tag);
278 void dmu_objset_disown(objset_t *os, void *tag);
279 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
281 void dmu_objset_evict_dbufs(objset_t *os);
282 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
283 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
284 int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
285 struct nvlist *snaps);
286 int dmu_objset_clone(const char *name, const char *origin);
287 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
288 struct nvlist *errlist);
289 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
290 int dmu_objset_snapshot_tmp(const char *, const char *, int);
291 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
293 void dmu_objset_byteswap(void *buf, size_t size);
294 int dsl_dataset_rename_snapshot(const char *fsname,
295 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
296 int dmu_objset_remap_indirects(const char *fsname);
298 typedef struct dmu_buf {
299 uint64_t db_object; /* object that this buffer is part of */
300 uint64_t db_offset; /* byte offset in this object */
301 uint64_t db_size; /* size of buffer in bytes */
302 void *db_data; /* data in buffer */
306 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
308 #define DMU_POOL_DIRECTORY_OBJECT 1
309 #define DMU_POOL_CONFIG "config"
310 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
311 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
312 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
313 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
314 #define DMU_POOL_ROOT_DATASET "root_dataset"
315 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
316 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
317 #define DMU_POOL_ERRLOG_LAST "errlog_last"
318 #define DMU_POOL_SPARES "spares"
319 #define DMU_POOL_DEFLATE "deflate"
320 #define DMU_POOL_HISTORY "history"
321 #define DMU_POOL_PROPS "pool_props"
322 #define DMU_POOL_L2CACHE "l2cache"
323 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
324 #define DMU_POOL_DDT "DDT-%s-%s-%s"
325 #define DMU_POOL_DDT_STATS "DDT-statistics"
326 #define DMU_POOL_CREATION_VERSION "creation_version"
327 #define DMU_POOL_SCAN "scan"
328 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
329 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
330 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
331 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
332 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
333 #define DMU_POOL_REMOVING "com.delphix:removing"
334 #define DMU_POOL_OBSOLETE_BPOBJ "com.delphix:obsolete_bpobj"
335 #define DMU_POOL_CONDENSING_INDIRECT "com.delphix:condensing_indirect"
336 #define DMU_POOL_ZPOOL_CHECKPOINT "com.delphix:zpool_checkpoint"
339 * Allocate an object from this objset. The range of object numbers
340 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
342 * The transaction must be assigned to a txg. The newly allocated
343 * object will be "held" in the transaction (ie. you can modify the
344 * newly allocated object in this transaction).
346 * dmu_object_alloc() chooses an object and returns it in *objectp.
348 * dmu_object_claim() allocates a specific object number. If that
349 * number is already allocated, it fails and returns EEXIST.
351 * Return 0 on success, or ENOSPC or EEXIST as specified above.
353 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
354 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
355 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
356 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
357 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
358 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
361 * Free an object from this objset.
363 * The object's data will be freed as well (ie. you don't need to call
364 * dmu_free(object, 0, -1, tx)).
366 * The object need not be held in the transaction.
368 * If there are any holds on this object's buffers (via dmu_buf_hold()),
369 * or tx holds on the object (via dmu_tx_hold_object()), you can not
370 * free it; it fails and returns EBUSY.
372 * If the object is not allocated, it fails and returns ENOENT.
374 * Return 0 on success, or EBUSY or ENOENT as specified above.
376 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
379 * Find the next allocated or free object.
381 * The objectp parameter is in-out. It will be updated to be the next
382 * object which is allocated. Ignore objects which have not been
383 * modified since txg.
385 * XXX Can only be called on a objset with no dirty data.
387 * Returns 0 on success, or ENOENT if there are no more objects.
389 int dmu_object_next(objset_t *os, uint64_t *objectp,
390 boolean_t hole, uint64_t txg);
393 * Set the data blocksize for an object.
395 * The object cannot have any blocks allcated beyond the first. If
396 * the first block is allocated already, the new size must be greater
397 * than the current block size. If these conditions are not met,
398 * ENOTSUP will be returned.
400 * Returns 0 on success, or EBUSY if there are any holds on the object
401 * contents, or ENOTSUP as described above.
403 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
404 int ibs, dmu_tx_t *tx);
407 * Set the checksum property on a dnode. The new checksum algorithm will
408 * apply to all newly written blocks; existing blocks will not be affected.
410 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
414 * Set the compress property on a dnode. The new compression algorithm will
415 * apply to all newly written blocks; existing blocks will not be affected.
417 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
420 int dmu_object_remap_indirects(objset_t *os, uint64_t object, uint64_t txg);
423 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
424 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
425 int compressed_size, int byteorder, dmu_tx_t *tx);
428 * Decide how to write a block: checksum, compression, number of copies, etc.
430 #define WP_NOFILL 0x1
431 #define WP_DMU_SYNC 0x2
434 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
435 struct zio_prop *zp);
437 * The bonus data is accessed more or less like a regular buffer.
438 * You must dmu_bonus_hold() to get the buffer, which will give you a
439 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
440 * data. As with any normal buffer, you must call dmu_buf_will_dirty()
441 * before modifying it, and the
442 * object must be held in an assigned transaction before calling
443 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
444 * buffer as well. You must release your hold with dmu_buf_rele().
446 * Returns ENOENT, EIO, or 0.
448 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
449 int dmu_bonus_max(void);
450 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
451 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
452 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
453 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
456 * Special spill buffer support used by "SA" framework
459 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
460 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
461 void *tag, dmu_buf_t **dbp);
462 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
465 * Obtain the DMU buffer from the specified object which contains the
466 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
467 * that it will remain in memory. You must release the hold with
468 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
469 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
471 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
472 * on the returned buffer before reading or writing the buffer's
473 * db_data. The comments for those routines describe what particular
474 * operations are valid after calling them.
476 * The object number must be a valid, allocated object number.
478 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
479 void *tag, dmu_buf_t **, int flags);
480 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
481 void *tag, dmu_buf_t **dbp, int flags);
484 * Add a reference to a dmu buffer that has already been held via
485 * dmu_buf_hold() in the current context.
487 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
490 * Attempt to add a reference to a dmu buffer that is in an unknown state,
491 * using a pointer that may have been invalidated by eviction processing.
492 * The request will succeed if the passed in dbuf still represents the
493 * same os/object/blkid, is ineligible for eviction, and has at least
494 * one hold by a user other than the syncer.
496 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
497 uint64_t blkid, void *tag);
499 void dmu_buf_rele(dmu_buf_t *db, void *tag);
500 uint64_t dmu_buf_refcount(dmu_buf_t *db);
503 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
504 * range of an object. A pointer to an array of dmu_buf_t*'s is
505 * returned (in *dbpp).
507 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
508 * frees the array. The hold on the array of buffers MUST be released
509 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
510 * individually with dmu_buf_rele.
512 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
513 uint64_t length, boolean_t read, void *tag,
514 int *numbufsp, dmu_buf_t ***dbpp);
515 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
517 typedef void dmu_buf_evict_func_t(void *user_ptr);
520 * A DMU buffer user object may be associated with a dbuf for the
521 * duration of its lifetime. This allows the user of a dbuf (client)
522 * to attach private data to a dbuf (e.g. in-core only data such as a
523 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
524 * when that dbuf has been evicted. Clients typically respond to the
525 * eviction notification by freeing their private data, thus ensuring
526 * the same lifetime for both dbuf and private data.
528 * The mapping from a dmu_buf_user_t to any client private data is the
529 * client's responsibility. All current consumers of the API with private
530 * data embed a dmu_buf_user_t as the first member of the structure for
531 * their private data. This allows conversions between the two types
532 * with a simple cast. Since the DMU buf user API never needs access
533 * to the private data, other strategies can be employed if necessary
534 * or convenient for the client (e.g. using container_of() to do the
535 * conversion for private data that cannot have the dmu_buf_user_t as
538 * Eviction callbacks are executed without the dbuf mutex held or any
539 * other type of mechanism to guarantee that the dbuf is still available.
540 * For this reason, users must assume the dbuf has already been freed
541 * and not reference the dbuf from the callback context.
543 * Users requesting "immediate eviction" are notified as soon as the dbuf
544 * is only referenced by dirty records (dirties == holds). Otherwise the
545 * notification occurs after eviction processing for the dbuf begins.
547 typedef struct dmu_buf_user {
549 * Asynchronous user eviction callback state.
551 taskq_ent_t dbu_tqent;
554 * This instance's eviction function pointers.
556 * dbu_evict_func_sync is called synchronously and then
557 * dbu_evict_func_async is executed asynchronously on a taskq.
559 dmu_buf_evict_func_t *dbu_evict_func_sync;
560 dmu_buf_evict_func_t *dbu_evict_func_async;
563 * Pointer to user's dbuf pointer. NULL for clients that do
564 * not associate a dbuf with their user data.
566 * The dbuf pointer is cleared upon eviction so as to catch
567 * use-after-evict bugs in clients.
569 dmu_buf_t **dbu_clear_on_evict_dbufp;
574 * Initialize the given dmu_buf_user_t instance with the eviction function
575 * evict_func, to be called when the user is evicted.
577 * NOTE: This function should only be called once on a given dmu_buf_user_t.
578 * To allow enforcement of this, dbu must already be zeroed on entry.
582 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
583 dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
585 ASSERT(dbu->dbu_evict_func_sync == NULL);
586 ASSERT(dbu->dbu_evict_func_async == NULL);
588 /* must have at least one evict func */
589 IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
590 dbu->dbu_evict_func_sync = evict_func_sync;
591 dbu->dbu_evict_func_async = evict_func_async;
593 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
598 * Attach user data to a dbuf and mark it for normal (when the dbuf's
599 * data is cleared or its reference count goes to zero) eviction processing.
601 * Returns NULL on success, or the existing user if another user currently
604 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
607 * Attach user data to a dbuf and mark it for immediate (its dirty and
608 * reference counts are equal) eviction processing.
610 * Returns NULL on success, or the existing user if another user currently
613 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
616 * Replace the current user of a dbuf.
618 * If given the current user of a dbuf, replaces the dbuf's user with
619 * "new_user" and returns the user data pointer that was replaced.
620 * Otherwise returns the current, and unmodified, dbuf user pointer.
622 void *dmu_buf_replace_user(dmu_buf_t *db,
623 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
626 * Remove the specified user data for a DMU buffer.
628 * Returns the user that was removed on success, or the current user if
629 * another user currently owns the buffer.
631 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
634 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
636 void *dmu_buf_get_user(dmu_buf_t *db);
638 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
639 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
640 void dmu_buf_dnode_exit(dmu_buf_t *db);
642 /* Block until any in-progress dmu buf user evictions complete. */
643 void dmu_buf_user_evict_wait(void);
646 * Returns the blkptr associated with this dbuf, or NULL if not set.
648 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
651 * Indicate that you are going to modify the buffer's data (db_data).
653 * The transaction (tx) must be assigned to a txg (ie. you've called
654 * dmu_tx_assign()). The buffer's object must be held in the tx
655 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
657 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
660 * You must create a transaction, then hold the objects which you will
661 * (or might) modify as part of this transaction. Then you must assign
662 * the transaction to a transaction group. Once the transaction has
663 * been assigned, you can modify buffers which belong to held objects as
664 * part of this transaction. You can't modify buffers before the
665 * transaction has been assigned; you can't modify buffers which don't
666 * belong to objects which this transaction holds; you can't hold
667 * objects once the transaction has been assigned. You may hold an
668 * object which you are going to free (with dmu_object_free()), but you
671 * You can abort the transaction before it has been assigned.
673 * Note that you may hold buffers (with dmu_buf_hold) at any time,
674 * regardless of transaction state.
677 #define DMU_NEW_OBJECT (-1ULL)
678 #define DMU_OBJECT_END (-1ULL)
680 dmu_tx_t *dmu_tx_create(objset_t *os);
681 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
682 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
684 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
686 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
688 void dmu_tx_hold_remap_l1indirect(dmu_tx_t *tx, uint64_t object);
689 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
690 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
692 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
693 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
694 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
695 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
696 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
697 void dmu_tx_abort(dmu_tx_t *tx);
698 int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
699 void dmu_tx_wait(dmu_tx_t *tx);
700 void dmu_tx_commit(dmu_tx_t *tx);
701 void dmu_tx_mark_netfree(dmu_tx_t *tx);
704 * To register a commit callback, dmu_tx_callback_register() must be called.
706 * dcb_data is a pointer to caller private data that is passed on as a
707 * callback parameter. The caller is responsible for properly allocating and
710 * When registering a callback, the transaction must be already created, but
711 * it cannot be committed or aborted. It can be assigned to a txg or not.
713 * The callback will be called after the transaction has been safely written
714 * to stable storage and will also be called if the dmu_tx is aborted.
715 * If there is any error which prevents the transaction from being committed to
716 * disk, the callback will be called with a value of error != 0.
718 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
720 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
724 * Free up the data blocks for a defined range of a file. If size is
725 * -1, the range from offset to end-of-file is freed.
727 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
728 uint64_t size, dmu_tx_t *tx);
729 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
731 int dmu_free_long_object(objset_t *os, uint64_t object);
734 * Convenience functions.
736 * Canfail routines will return 0 on success, or an errno if there is a
737 * nonrecoverable I/O error.
739 #define DMU_READ_PREFETCH 0 /* prefetch */
740 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
741 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
742 void *buf, uint32_t flags);
743 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
745 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
746 const void *buf, dmu_tx_t *tx);
747 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
748 const void *buf, dmu_tx_t *tx);
749 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
751 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
752 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
753 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
755 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
759 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
760 uint64_t size, struct page *pp, dmu_tx_t *tx);
762 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
763 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx);
764 int dmu_read_pages(objset_t *os, uint64_t object, vm_page_t *ma, int count,
765 int *rbehind, int *rahead, int last_size);
768 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
769 void dmu_return_arcbuf(struct arc_buf *buf);
770 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
772 int dmu_xuio_init(struct xuio *uio, int niov);
773 void dmu_xuio_fini(struct xuio *uio);
774 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
776 int dmu_xuio_cnt(struct xuio *uio);
777 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
778 void dmu_xuio_clear(struct xuio *uio, int i);
779 void xuio_stat_wbuf_copied(void);
780 void xuio_stat_wbuf_nocopy(void);
782 extern boolean_t zfs_prefetch_disable;
783 extern int zfs_max_recordsize;
786 * Asynchronously try to read in the data.
788 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
789 uint64_t len, enum zio_priority pri);
791 typedef struct dmu_object_info {
792 /* All sizes are in bytes unless otherwise indicated. */
793 uint32_t doi_data_block_size;
794 uint32_t doi_metadata_block_size;
795 dmu_object_type_t doi_type;
796 dmu_object_type_t doi_bonus_type;
797 uint64_t doi_bonus_size;
798 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
799 uint8_t doi_checksum;
800 uint8_t doi_compress;
803 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
804 uint64_t doi_max_offset;
805 uint64_t doi_fill_count; /* number of non-empty blocks */
808 typedef void arc_byteswap_func_t(void *buf, size_t size);
810 typedef struct dmu_object_type_info {
811 dmu_object_byteswap_t ot_byteswap;
812 boolean_t ot_metadata;
814 } dmu_object_type_info_t;
816 typedef struct dmu_object_byteswap_info {
817 arc_byteswap_func_t *ob_func;
819 } dmu_object_byteswap_info_t;
821 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
822 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
825 * Get information on a DMU object.
827 * Return 0 on success or ENOENT if object is not allocated.
829 * If doi is NULL, just indicates whether the object exists.
831 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
832 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
833 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
834 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
835 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
837 * Like dmu_object_info_from_db, but faster still when you only care about
838 * the size. This is specifically optimized for zfs_getattr().
840 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
841 u_longlong_t *nblk512);
843 typedef struct dmu_objset_stats {
844 uint64_t dds_num_clones; /* number of clones of this */
845 uint64_t dds_creation_txg;
847 dmu_objset_type_t dds_type;
848 uint8_t dds_is_snapshot;
849 uint8_t dds_inconsistent;
850 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
851 } dmu_objset_stats_t;
854 * Get stats on a dataset.
856 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
859 * Add entries to the nvlist for all the objset's properties. See
860 * zfs_prop_table[] and zfs(1m) for details on the properties.
862 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
865 * Get the space usage statistics for statvfs().
867 * refdbytes is the amount of space "referenced" by this objset.
868 * availbytes is the amount of space available to this objset, taking
869 * into account quotas & reservations, assuming that no other objsets
870 * use the space first. These values correspond to the 'referenced' and
871 * 'available' properties, described in the zfs(1m) manpage.
873 * usedobjs and availobjs are the number of objects currently allocated,
876 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
877 uint64_t *usedobjsp, uint64_t *availobjsp);
880 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
881 * (Contrast with the ds_guid which is a 64-bit ID that will never
882 * change, so there is a small probability that it will collide.)
884 uint64_t dmu_objset_fsid_guid(objset_t *os);
887 * Get the [cm]time for an objset's snapshot dir
889 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
891 int dmu_objset_is_snapshot(objset_t *os);
893 extern struct spa *dmu_objset_spa(objset_t *os);
894 extern struct zilog *dmu_objset_zil(objset_t *os);
895 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
896 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
897 extern void dmu_objset_name(objset_t *os, char *buf);
898 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
899 extern uint64_t dmu_objset_id(objset_t *os);
900 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
901 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
902 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
903 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
904 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
905 int maxlen, boolean_t *conflict);
906 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
907 uint64_t *idp, uint64_t *offp);
909 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
910 void *bonus, uint64_t *userp, uint64_t *groupp);
911 extern void dmu_objset_register_type(dmu_objset_type_t ost,
912 objset_used_cb_t *cb);
913 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
914 extern void *dmu_objset_get_user(objset_t *os);
917 * Return the txg number for the given assigned transaction.
919 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
923 * If a parent zio is provided this function initiates a write on the
924 * provided buffer as a child of the parent zio.
925 * In the absence of a parent zio, the write is completed synchronously.
926 * At write completion, blk is filled with the bp of the written block.
927 * Note that while the data covered by this function will be on stable
928 * storage when the write completes this new data does not become a
929 * permanent part of the file until the associated transaction commits.
933 * {zfs,zvol,ztest}_get_done() args
937 struct blkptr *zgd_bp;
943 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
944 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
947 * Find the next hole or data block in file starting at *off
948 * Return found offset in *off. Return ESRCH for end of file.
950 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
954 * Check if a DMU object has any dirty blocks. If so, sync out
955 * all pending transaction groups. Otherwise, this function
956 * does not alter DMU state. This could be improved to only sync
957 * out the necessary transaction groups for this particular
960 int dmu_object_wait_synced(objset_t *os, uint64_t object);
963 * Initial setup and final teardown.
965 extern void dmu_init(void);
966 extern void dmu_fini(void);
968 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
969 uint64_t object, uint64_t offset, int len);
970 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
971 dmu_traverse_cb_t cb, void *arg);
972 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
973 struct file *fp, offset_t *offp);
976 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
977 extern uint64_t zfs_crc64_table[256];
979 extern int zfs_mdcomp_disable;
985 #endif /* _SYS_DMU_H */