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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 * Copyright 2014 HybridCluster. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30 /* Portions Copyright 2010 Robert Milkowski */
36 * This file describes the interface that the DMU provides for its
39 * The DMU also interacts with the SPA. That interface is described in
43 #include <sys/zfs_context.h>
44 #include <sys/inttypes.h>
46 #include <sys/fs/zfs.h>
47 #include <sys/zio_compress.h>
48 #include <sys/zio_priority.h>
67 struct zbookmark_phys;
74 typedef struct objset objset_t;
75 typedef struct dmu_tx dmu_tx_t;
76 typedef struct dsl_dir dsl_dir_t;
77 typedef struct dnode dnode_t;
79 typedef enum dmu_object_byteswap {
91 * Allocating a new byteswap type number makes the on-disk format
92 * incompatible with any other format that uses the same number.
94 * Data can usually be structured to work with one of the
95 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
98 } dmu_object_byteswap_t;
100 #define DMU_OT_NEWTYPE 0x80
101 #define DMU_OT_METADATA 0x40
102 #define DMU_OT_BYTESWAP_MASK 0x3f
105 * Defines a uint8_t object type. Object types specify if the data
106 * in the object is metadata (boolean) and how to byteswap the data
107 * (dmu_object_byteswap_t).
109 #define DMU_OT(byteswap, metadata) \
111 ((metadata) ? DMU_OT_METADATA : 0) | \
112 ((byteswap) & DMU_OT_BYTESWAP_MASK))
114 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
115 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
116 (ot) < DMU_OT_NUMTYPES)
118 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
119 ((ot) & DMU_OT_METADATA) : \
120 dmu_ot[(int)(ot)].ot_metadata)
123 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
124 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
125 * is repurposed for embedded BPs.
127 #define DMU_OT_HAS_FILL(ot) \
128 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
130 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
131 ((ot) & DMU_OT_BYTESWAP_MASK) : \
132 dmu_ot[(int)(ot)].ot_byteswap)
134 typedef enum dmu_object_type {
137 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
138 DMU_OT_OBJECT_ARRAY, /* UINT64 */
139 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
140 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
141 DMU_OT_BPOBJ, /* UINT64 */
142 DMU_OT_BPOBJ_HDR, /* UINT64 */
144 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
145 DMU_OT_SPACE_MAP, /* UINT64 */
147 DMU_OT_INTENT_LOG, /* UINT64 */
149 DMU_OT_DNODE, /* DNODE */
150 DMU_OT_OBJSET, /* OBJSET */
152 DMU_OT_DSL_DIR, /* UINT64 */
153 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
154 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
155 DMU_OT_DSL_PROPS, /* ZAP */
156 DMU_OT_DSL_DATASET, /* UINT64 */
158 DMU_OT_ZNODE, /* ZNODE */
159 DMU_OT_OLDACL, /* Old ACL */
160 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
161 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
162 DMU_OT_MASTER_NODE, /* ZAP */
163 DMU_OT_UNLINKED_SET, /* ZAP */
165 DMU_OT_ZVOL, /* UINT8 */
166 DMU_OT_ZVOL_PROP, /* ZAP */
167 /* other; for testing only! */
168 DMU_OT_PLAIN_OTHER, /* UINT8 */
169 DMU_OT_UINT64_OTHER, /* UINT64 */
170 DMU_OT_ZAP_OTHER, /* ZAP */
171 /* new object types: */
172 DMU_OT_ERROR_LOG, /* ZAP */
173 DMU_OT_SPA_HISTORY, /* UINT8 */
174 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
175 DMU_OT_POOL_PROPS, /* ZAP */
176 DMU_OT_DSL_PERMS, /* ZAP */
177 DMU_OT_ACL, /* ACL */
178 DMU_OT_SYSACL, /* SYSACL */
179 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
180 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
181 DMU_OT_NEXT_CLONES, /* ZAP */
182 DMU_OT_SCAN_QUEUE, /* ZAP */
183 DMU_OT_USERGROUP_USED, /* ZAP */
184 DMU_OT_USERGROUP_QUOTA, /* ZAP */
185 DMU_OT_USERREFS, /* ZAP */
186 DMU_OT_DDT_ZAP, /* ZAP */
187 DMU_OT_DDT_STATS, /* ZAP */
188 DMU_OT_SA, /* System attr */
189 DMU_OT_SA_MASTER_NODE, /* ZAP */
190 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
191 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
192 DMU_OT_SCAN_XLATE, /* ZAP */
193 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
194 DMU_OT_DEADLIST, /* ZAP */
195 DMU_OT_DEADLIST_HDR, /* UINT64 */
196 DMU_OT_DSL_CLONES, /* ZAP */
197 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
199 * Do not allocate new object types here. Doing so makes the on-disk
200 * format incompatible with any other format that uses the same object
203 * When creating an object which does not have one of the above types
204 * use the DMU_OTN_* type with the correct byteswap and metadata
207 * The DMU_OTN_* types do not have entries in the dmu_ot table,
208 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
209 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
210 * and DMU_OTN_* types).
215 * Names for valid types declared with DMU_OT().
217 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
218 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
219 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
220 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
221 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
222 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
223 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
224 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
225 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
226 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
229 typedef enum txg_how {
235 void byteswap_uint64_array(void *buf, size_t size);
236 void byteswap_uint32_array(void *buf, size_t size);
237 void byteswap_uint16_array(void *buf, size_t size);
238 void byteswap_uint8_array(void *buf, size_t size);
239 void zap_byteswap(void *buf, size_t size);
240 void zfs_oldacl_byteswap(void *buf, size_t size);
241 void zfs_acl_byteswap(void *buf, size_t size);
242 void zfs_znode_byteswap(void *buf, size_t size);
244 #define DS_FIND_SNAPSHOTS (1<<0)
245 #define DS_FIND_CHILDREN (1<<1)
246 #define DS_FIND_SERIALIZE (1<<2)
249 * The maximum number of bytes that can be accessed as part of one
250 * operation, including metadata.
252 #define DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
253 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
255 #define DMU_USERUSED_OBJECT (-1ULL)
256 #define DMU_GROUPUSED_OBJECT (-2ULL)
259 * artificial blkids for bonus buffer and spill blocks
261 #define DMU_BONUS_BLKID (-1ULL)
262 #define DMU_SPILL_BLKID (-2ULL)
264 * Public routines to create, destroy, open, and close objsets.
266 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
267 int dmu_objset_own(const char *name, dmu_objset_type_t type,
268 boolean_t readonly, void *tag, objset_t **osp);
269 void dmu_objset_rele(objset_t *os, void *tag);
270 void dmu_objset_disown(objset_t *os, void *tag);
271 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
273 void dmu_objset_evict_dbufs(objset_t *os);
274 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
275 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
276 int dmu_objset_clone(const char *name, const char *origin);
277 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
278 struct nvlist *errlist);
279 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
280 int dmu_objset_snapshot_tmp(const char *, const char *, int);
281 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
283 void dmu_objset_byteswap(void *buf, size_t size);
284 int dsl_dataset_rename_snapshot(const char *fsname,
285 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
287 typedef struct dmu_buf {
288 uint64_t db_object; /* object that this buffer is part of */
289 uint64_t db_offset; /* byte offset in this object */
290 uint64_t db_size; /* size of buffer in bytes */
291 void *db_data; /* data in buffer */
295 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
297 #define DMU_POOL_DIRECTORY_OBJECT 1
298 #define DMU_POOL_CONFIG "config"
299 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
300 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
301 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
302 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
303 #define DMU_POOL_ROOT_DATASET "root_dataset"
304 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
305 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
306 #define DMU_POOL_ERRLOG_LAST "errlog_last"
307 #define DMU_POOL_SPARES "spares"
308 #define DMU_POOL_DEFLATE "deflate"
309 #define DMU_POOL_HISTORY "history"
310 #define DMU_POOL_PROPS "pool_props"
311 #define DMU_POOL_L2CACHE "l2cache"
312 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
313 #define DMU_POOL_DDT "DDT-%s-%s-%s"
314 #define DMU_POOL_DDT_STATS "DDT-statistics"
315 #define DMU_POOL_CREATION_VERSION "creation_version"
316 #define DMU_POOL_SCAN "scan"
317 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
318 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
319 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
320 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
323 * Allocate an object from this objset. The range of object numbers
324 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
326 * The transaction must be assigned to a txg. The newly allocated
327 * object will be "held" in the transaction (ie. you can modify the
328 * newly allocated object in this transaction).
330 * dmu_object_alloc() chooses an object and returns it in *objectp.
332 * dmu_object_claim() allocates a specific object number. If that
333 * number is already allocated, it fails and returns EEXIST.
335 * Return 0 on success, or ENOSPC or EEXIST as specified above.
337 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
338 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
339 uint64_t dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot,
340 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
341 int dnodesize, dmu_tx_t *tx);
342 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
343 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
344 int dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
345 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
346 int dnodesize, dmu_tx_t *tx);
347 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
348 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
349 int dmu_object_reclaim_dnsize(objset_t *os, uint64_t object,
350 dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype,
351 int bonuslen, int dnodesize, dmu_tx_t *txp);
354 * Free an object from this objset.
356 * The object's data will be freed as well (ie. you don't need to call
357 * dmu_free(object, 0, -1, tx)).
359 * The object need not be held in the transaction.
361 * If there are any holds on this object's buffers (via dmu_buf_hold()),
362 * or tx holds on the object (via dmu_tx_hold_object()), you can not
363 * free it; it fails and returns EBUSY.
365 * If the object is not allocated, it fails and returns ENOENT.
367 * Return 0 on success, or EBUSY or ENOENT as specified above.
369 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
372 * Find the next allocated or free object.
374 * The objectp parameter is in-out. It will be updated to be the next
375 * object which is allocated. Ignore objects which have not been
376 * modified since txg.
378 * XXX Can only be called on a objset with no dirty data.
380 * Returns 0 on success, or ENOENT if there are no more objects.
382 int dmu_object_next(objset_t *os, uint64_t *objectp,
383 boolean_t hole, uint64_t txg);
386 * Set the data blocksize for an object.
388 * The object cannot have any blocks allcated beyond the first. If
389 * the first block is allocated already, the new size must be greater
390 * than the current block size. If these conditions are not met,
391 * ENOTSUP will be returned.
393 * Returns 0 on success, or EBUSY if there are any holds on the object
394 * contents, or ENOTSUP as described above.
396 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
397 int ibs, dmu_tx_t *tx);
400 * Set the checksum property on a dnode. The new checksum algorithm will
401 * apply to all newly written blocks; existing blocks will not be affected.
403 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
407 * Set the compress property on a dnode. The new compression algorithm will
408 * apply to all newly written blocks; existing blocks will not be affected.
410 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
414 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
415 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
416 int compressed_size, int byteorder, dmu_tx_t *tx);
419 * Decide how to write a block: checksum, compression, number of copies, etc.
421 #define WP_NOFILL 0x1
422 #define WP_DMU_SYNC 0x2
425 void dmu_write_policy(objset_t *os, struct dnode *dn, int level, int wp,
426 enum zio_compress compress_override, struct zio_prop *zp);
428 * The bonus data is accessed more or less like a regular buffer.
429 * You must dmu_bonus_hold() to get the buffer, which will give you a
430 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
431 * data. As with any normal buffer, you must call dmu_buf_read() to
432 * read db_data, dmu_buf_will_dirty() before modifying it, and the
433 * object must be held in an assigned transaction before calling
434 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
435 * buffer as well. You must release what you hold with dmu_buf_rele().
437 * Returns ENOENT, EIO, or 0.
439 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
440 int dmu_bonus_max(void);
441 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
442 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
443 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
444 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
447 * Special spill buffer support used by "SA" framework
450 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
451 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
452 void *tag, dmu_buf_t **dbp);
453 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
456 * Obtain the DMU buffer from the specified object which contains the
457 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
458 * that it will remain in memory. You must release the hold with
459 * dmu_buf_rele(). You must not access the dmu_buf_t after releasing
460 * what you hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
462 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
463 * on the returned buffer before reading or writing the buffer's
464 * db_data. The comments for those routines describe what particular
465 * operations are valid after calling them.
467 * The object number must be a valid, allocated object number.
469 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
470 void *tag, dmu_buf_t **, int flags);
471 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
472 void *tag, dmu_buf_t **dbp, int flags);
475 * Add a reference to a dmu buffer that has already been held via
476 * dmu_buf_hold() in the current context.
478 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
481 * Attempt to add a reference to a dmu buffer that is in an unknown state,
482 * using a pointer that may have been invalidated by eviction processing.
483 * The request will succeed if the passed in dbuf still represents the
484 * same os/object/blkid, is ineligible for eviction, and has at least
485 * one hold by a user other than the syncer.
487 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
488 uint64_t blkid, void *tag);
490 void dmu_buf_rele(dmu_buf_t *db, void *tag);
491 uint64_t dmu_buf_refcount(dmu_buf_t *db);
494 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
495 * range of an object. A pointer to an array of dmu_buf_t*'s is
496 * returned (in *dbpp).
498 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
499 * frees the array. The hold on the array of buffers MUST be released
500 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
501 * individually with dmu_buf_rele.
503 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
504 uint64_t length, boolean_t read, void *tag,
505 int *numbufsp, dmu_buf_t ***dbpp);
506 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
508 typedef void dmu_buf_evict_func_t(void *user_ptr);
511 * A DMU buffer user object may be associated with a dbuf for the
512 * duration of its lifetime. This allows the user of a dbuf (client)
513 * to attach private data to a dbuf (e.g. in-core only data such as a
514 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
515 * when that dbuf has been evicted. Clients typically respond to the
516 * eviction notification by freeing their private data, thus ensuring
517 * the same lifetime for both dbuf and private data.
519 * The mapping from a dmu_buf_user_t to any client private data is the
520 * client's responsibility. All current consumers of the API with private
521 * data embed a dmu_buf_user_t as the first member of the structure for
522 * their private data. This allows conversions between the two types
523 * with a simple cast. Since the DMU buf user API never needs access
524 * to the private data, other strategies can be employed if necessary
525 * or convenient for the client (e.g. using container_of() to do the
526 * conversion for private data that cannot have the dmu_buf_user_t as
529 * Eviction callbacks are executed without the dbuf mutex held or any
530 * other type of mechanism to guarantee that the dbuf is still available.
531 * For this reason, users must assume the dbuf has already been freed
532 * and not reference the dbuf from the callback context.
534 * Users requesting "immediate eviction" are notified as soon as the dbuf
535 * is only referenced by dirty records (dirties == holds). Otherwise the
536 * notification occurs after eviction processing for the dbuf begins.
538 typedef struct dmu_buf_user {
540 * Asynchronous user eviction callback state.
542 taskq_ent_t dbu_tqent;
544 /* This instance's eviction function pointer. */
545 dmu_buf_evict_func_t *dbu_evict_func;
548 * Pointer to user's dbuf pointer. NULL for clients that do
549 * not associate a dbuf with their user data.
551 * The dbuf pointer is cleared upon eviction so as to catch
552 * use-after-evict bugs in clients.
554 dmu_buf_t **dbu_clear_on_evict_dbufp;
559 * Initialize the given dmu_buf_user_t instance with the eviction function
560 * evict_func, to be called when the user is evicted.
562 * NOTE: This function should only be called once on a given dmu_buf_user_t.
563 * To allow enforcement of this, dbu must already be zeroed on entry.
567 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
568 dmu_buf_t **clear_on_evict_dbufp)
570 ASSERT(dbu->dbu_evict_func == NULL);
571 ASSERT(evict_func != NULL);
572 dbu->dbu_evict_func = evict_func;
573 taskq_init_ent(&dbu->dbu_tqent);
575 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
580 * Attach user data to a dbuf and mark it for normal (when the dbuf's
581 * data is cleared or its reference count goes to zero) eviction processing.
583 * Returns NULL on success, or the existing user if another user currently
586 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
589 * Attach user data to a dbuf and mark it for immediate (its dirty and
590 * reference counts are equal) eviction processing.
592 * Returns NULL on success, or the existing user if another user currently
595 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
598 * Replace the current user of a dbuf.
600 * If given the current user of a dbuf, replaces the dbuf's user with
601 * "new_user" and returns the user data pointer that was replaced.
602 * Otherwise returns the current, and unmodified, dbuf user pointer.
604 void *dmu_buf_replace_user(dmu_buf_t *db,
605 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
608 * Remove the specified user data for a DMU buffer.
610 * Returns the user that was removed on success, or the current user if
611 * another user currently owns the buffer.
613 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
616 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
618 void *dmu_buf_get_user(dmu_buf_t *db);
620 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
621 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
622 void dmu_buf_dnode_exit(dmu_buf_t *db);
624 /* Block until any in-progress dmu buf user evictions complete. */
625 void dmu_buf_user_evict_wait(void);
628 * Returns the blkptr associated with this dbuf, or NULL if not set.
630 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
633 * Indicate that you are going to modify the buffer's data (db_data).
635 * The transaction (tx) must be assigned to a txg (ie. you've called
636 * dmu_tx_assign()). The buffer's object must be held in the tx
637 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
639 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
642 * Tells if the given dbuf is freeable.
644 boolean_t dmu_buf_freeable(dmu_buf_t *);
647 * You must create a transaction, then hold the objects which you will
648 * (or might) modify as part of this transaction. Then you must assign
649 * the transaction to a transaction group. Once the transaction has
650 * been assigned, you can modify buffers which belong to held objects as
651 * part of this transaction. You can't modify buffers before the
652 * transaction has been assigned; you can't modify buffers which don't
653 * belong to objects which this transaction holds; you can't hold
654 * objects once the transaction has been assigned. You may hold an
655 * object which you are going to free (with dmu_object_free()), but you
658 * You can abort the transaction before it has been assigned.
660 * Note that you may hold buffers (with dmu_buf_hold) at any time,
661 * regardless of transaction state.
664 #define DMU_NEW_OBJECT (-1ULL)
665 #define DMU_OBJECT_END (-1ULL)
667 dmu_tx_t *dmu_tx_create(objset_t *os);
668 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
669 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
671 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
672 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
673 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
674 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
675 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
676 void dmu_tx_abort(dmu_tx_t *tx);
677 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
678 void dmu_tx_wait(dmu_tx_t *tx);
679 void dmu_tx_commit(dmu_tx_t *tx);
680 void dmu_tx_mark_netfree(dmu_tx_t *tx);
683 * To register a commit callback, dmu_tx_callback_register() must be called.
685 * dcb_data is a pointer to caller private data that is passed on as a
686 * callback parameter. The caller is responsible for properly allocating and
689 * When registering a callback, the transaction must be already created, but
690 * it cannot be committed or aborted. It can be assigned to a txg or not.
692 * The callback will be called after the transaction has been safely written
693 * to stable storage and will also be called if the dmu_tx is aborted.
694 * If there is any error which prevents the transaction from being committed to
695 * disk, the callback will be called with a value of error != 0.
697 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
699 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
703 * Free up the data blocks for a defined range of a file. If size is
704 * -1, the range from offset to end-of-file is freed.
706 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
707 uint64_t size, dmu_tx_t *tx);
708 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
710 int dmu_free_long_object(objset_t *os, uint64_t object);
713 * Convenience functions.
715 * Canfail routines will return 0 on success, or an errno if there is a
716 * nonrecoverable I/O error.
718 #define DMU_READ_PREFETCH 0 /* prefetch */
719 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
720 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
721 void *buf, uint32_t flags);
722 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
723 const void *buf, dmu_tx_t *tx);
724 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
727 #include <linux/blkdev_compat.h>
728 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
729 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
730 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
732 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
735 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
736 void dmu_return_arcbuf(struct arc_buf *buf);
737 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
739 int dmu_xuio_init(struct xuio *uio, int niov);
740 void dmu_xuio_fini(struct xuio *uio);
741 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
743 int dmu_xuio_cnt(struct xuio *uio);
744 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
745 void dmu_xuio_clear(struct xuio *uio, int i);
746 void xuio_stat_wbuf_copied(void);
747 void xuio_stat_wbuf_nocopy(void);
749 extern int zfs_prefetch_disable;
750 extern int zfs_max_recordsize;
753 * Asynchronously try to read in the data.
755 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
756 uint64_t len, enum zio_priority pri);
758 typedef struct dmu_object_info {
759 /* All sizes are in bytes unless otherwise indicated. */
760 uint32_t doi_data_block_size;
761 uint32_t doi_metadata_block_size;
762 dmu_object_type_t doi_type;
763 dmu_object_type_t doi_bonus_type;
764 uint64_t doi_bonus_size;
765 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
766 uint8_t doi_checksum;
767 uint8_t doi_compress;
770 uint64_t doi_dnodesize;
771 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
772 uint64_t doi_max_offset;
773 uint64_t doi_fill_count; /* number of non-empty blocks */
776 typedef void (*const arc_byteswap_func_t)(void *buf, size_t size);
778 typedef struct dmu_object_type_info {
779 dmu_object_byteswap_t ot_byteswap;
780 boolean_t ot_metadata;
782 } dmu_object_type_info_t;
784 typedef const struct dmu_object_byteswap_info {
785 arc_byteswap_func_t ob_func;
787 } dmu_object_byteswap_info_t;
789 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
790 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
793 * Get information on a DMU object.
795 * Return 0 on success or ENOENT if object is not allocated.
797 * If doi is NULL, just indicates whether the object exists.
799 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
800 void __dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
801 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
802 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
803 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
804 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
806 * Like dmu_object_info_from_db, but faster still when you only care about
807 * the size. This is specifically optimized for zfs_getattr().
809 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
810 u_longlong_t *nblk512);
812 void dmu_object_dnsize_from_db(dmu_buf_t *db, int *dnsize);
814 typedef struct dmu_objset_stats {
815 uint64_t dds_num_clones; /* number of clones of this */
816 uint64_t dds_creation_txg;
818 dmu_objset_type_t dds_type;
819 uint8_t dds_is_snapshot;
820 uint8_t dds_inconsistent;
821 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
822 } dmu_objset_stats_t;
825 * Get stats on a dataset.
827 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
830 * Add entries to the nvlist for all the objset's properties. See
831 * zfs_prop_table[] and zfs(1m) for details on the properties.
833 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
836 * Get the space usage statistics for statvfs().
838 * refdbytes is the amount of space "referenced" by this objset.
839 * availbytes is the amount of space available to this objset, taking
840 * into account quotas & reservations, assuming that no other objsets
841 * use the space first. These values correspond to the 'referenced' and
842 * 'available' properties, described in the zfs(1m) manpage.
844 * usedobjs and availobjs are the number of objects currently allocated,
847 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
848 uint64_t *usedobjsp, uint64_t *availobjsp);
851 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
852 * (Contrast with the ds_guid which is a 64-bit ID that will never
853 * change, so there is a small probability that it will collide.)
855 uint64_t dmu_objset_fsid_guid(objset_t *os);
858 * Get the [cm]time for an objset's snapshot dir
860 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
862 int dmu_objset_is_snapshot(objset_t *os);
864 extern struct spa *dmu_objset_spa(objset_t *os);
865 extern struct zilog *dmu_objset_zil(objset_t *os);
866 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
867 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
868 extern void dmu_objset_name(objset_t *os, char *buf);
869 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
870 extern uint64_t dmu_objset_id(objset_t *os);
871 extern uint64_t dmu_objset_dnodesize(objset_t *os);
872 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
873 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
874 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
875 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
876 extern int dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *val);
877 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
878 int maxlen, boolean_t *conflict);
879 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
880 uint64_t *idp, uint64_t *offp);
882 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
883 void *bonus, uint64_t *userp, uint64_t *groupp);
884 extern void dmu_objset_register_type(dmu_objset_type_t ost,
885 objset_used_cb_t *cb);
886 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
887 extern void *dmu_objset_get_user(objset_t *os);
890 * Return the txg number for the given assigned transaction.
892 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
896 * If a parent zio is provided this function initiates a write on the
897 * provided buffer as a child of the parent zio.
898 * In the absence of a parent zio, the write is completed synchronously.
899 * At write completion, blk is filled with the bp of the written block.
900 * Note that while the data covered by this function will be on stable
901 * storage when the write completes this new data does not become a
902 * permanent part of the file until the associated transaction commits.
906 * {zfs,zvol,ztest}_get_done() args
909 struct zilog *zgd_zilog;
910 struct blkptr *zgd_bp;
916 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
917 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
920 * Find the next hole or data block in file starting at *off
921 * Return found offset in *off. Return ESRCH for end of file.
923 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
927 * Initial setup and final teardown.
929 extern void dmu_init(void);
930 extern void dmu_fini(void);
932 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
933 uint64_t object, uint64_t offset, int len);
934 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
935 dmu_traverse_cb_t cb, void *arg);
937 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
938 struct vnode *vp, offset_t *offp);
941 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
942 extern uint64_t zfs_crc64_table[256];
944 extern int zfs_mdcomp_disable;
950 #endif /* _SYS_DMU_H */