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). All of the types created by this method
113 * are cached in the dbuf metadata cache.
115 #define DMU_OT(byteswap, metadata) \
117 ((metadata) ? DMU_OT_METADATA : 0) | \
118 ((byteswap) & DMU_OT_BYTESWAP_MASK))
120 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
121 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
122 (ot) < DMU_OT_NUMTYPES)
124 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
125 ((ot) & DMU_OT_METADATA) : \
126 dmu_ot[(ot)].ot_metadata)
128 #define DMU_OT_IS_METADATA_CACHED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
129 B_TRUE : dmu_ot[(ot)].ot_dbuf_metadata_cache)
132 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
133 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
134 * is repurposed for embedded BPs.
136 #define DMU_OT_HAS_FILL(ot) \
137 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
139 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
140 ((ot) & DMU_OT_BYTESWAP_MASK) : \
141 dmu_ot[(ot)].ot_byteswap)
143 typedef enum dmu_object_type {
146 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
147 DMU_OT_OBJECT_ARRAY, /* UINT64 */
148 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
149 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
150 DMU_OT_BPOBJ, /* UINT64 */
151 DMU_OT_BPOBJ_HDR, /* UINT64 */
153 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
154 DMU_OT_SPACE_MAP, /* UINT64 */
156 DMU_OT_INTENT_LOG, /* UINT64 */
158 DMU_OT_DNODE, /* DNODE */
159 DMU_OT_OBJSET, /* OBJSET */
161 DMU_OT_DSL_DIR, /* UINT64 */
162 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
163 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
164 DMU_OT_DSL_PROPS, /* ZAP */
165 DMU_OT_DSL_DATASET, /* UINT64 */
167 DMU_OT_ZNODE, /* ZNODE */
168 DMU_OT_OLDACL, /* Old ACL */
169 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
170 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
171 DMU_OT_MASTER_NODE, /* ZAP */
172 DMU_OT_UNLINKED_SET, /* ZAP */
174 DMU_OT_ZVOL, /* UINT8 */
175 DMU_OT_ZVOL_PROP, /* ZAP */
176 /* other; for testing only! */
177 DMU_OT_PLAIN_OTHER, /* UINT8 */
178 DMU_OT_UINT64_OTHER, /* UINT64 */
179 DMU_OT_ZAP_OTHER, /* ZAP */
180 /* new object types: */
181 DMU_OT_ERROR_LOG, /* ZAP */
182 DMU_OT_SPA_HISTORY, /* UINT8 */
183 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
184 DMU_OT_POOL_PROPS, /* ZAP */
185 DMU_OT_DSL_PERMS, /* ZAP */
186 DMU_OT_ACL, /* ACL */
187 DMU_OT_SYSACL, /* SYSACL */
188 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
189 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
190 DMU_OT_NEXT_CLONES, /* ZAP */
191 DMU_OT_SCAN_QUEUE, /* ZAP */
192 DMU_OT_USERGROUP_USED, /* ZAP */
193 DMU_OT_USERGROUP_QUOTA, /* ZAP */
194 DMU_OT_USERREFS, /* ZAP */
195 DMU_OT_DDT_ZAP, /* ZAP */
196 DMU_OT_DDT_STATS, /* ZAP */
197 DMU_OT_SA, /* System attr */
198 DMU_OT_SA_MASTER_NODE, /* ZAP */
199 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
200 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
201 DMU_OT_SCAN_XLATE, /* ZAP */
202 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
203 DMU_OT_DEADLIST, /* ZAP */
204 DMU_OT_DEADLIST_HDR, /* UINT64 */
205 DMU_OT_DSL_CLONES, /* ZAP */
206 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
208 * Do not allocate new object types here. Doing so makes the on-disk
209 * format incompatible with any other format that uses the same object
212 * When creating an object which does not have one of the above types
213 * use the DMU_OTN_* type with the correct byteswap and metadata
216 * The DMU_OTN_* types do not have entries in the dmu_ot table,
217 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
218 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
219 * and DMU_OTN_* types).
224 * Names for valid types declared with DMU_OT().
226 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE),
227 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE),
228 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE),
229 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE),
230 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE),
231 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE),
232 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE),
233 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE),
234 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE),
235 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
239 * These flags are intended to be used to specify the "txg_how"
240 * parameter when calling the dmu_tx_assign() function. See the comment
241 * above dmu_tx_assign() for more details on the meaning of these flags.
243 #define TXG_NOWAIT (0ULL)
244 #define TXG_WAIT (1ULL<<0)
245 #define TXG_NOTHROTTLE (1ULL<<1)
247 void byteswap_uint64_array(void *buf, size_t size);
248 void byteswap_uint32_array(void *buf, size_t size);
249 void byteswap_uint16_array(void *buf, size_t size);
250 void byteswap_uint8_array(void *buf, size_t size);
251 void zap_byteswap(void *buf, size_t size);
252 void zfs_oldacl_byteswap(void *buf, size_t size);
253 void zfs_acl_byteswap(void *buf, size_t size);
254 void zfs_znode_byteswap(void *buf, size_t size);
256 #define DS_FIND_SNAPSHOTS (1<<0)
257 #define DS_FIND_CHILDREN (1<<1)
258 #define DS_FIND_SERIALIZE (1<<2)
261 * The maximum number of bytes that can be accessed as part of one
262 * operation, including metadata.
264 #define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */
265 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
267 #define DMU_USERUSED_OBJECT (-1ULL)
268 #define DMU_GROUPUSED_OBJECT (-2ULL)
271 * artificial blkids for bonus buffer and spill blocks
273 #define DMU_BONUS_BLKID (-1ULL)
274 #define DMU_SPILL_BLKID (-2ULL)
276 * Public routines to create, destroy, open, and close objsets.
278 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
279 int dmu_objset_own(const char *name, dmu_objset_type_t type,
280 boolean_t readonly, void *tag, objset_t **osp);
281 void dmu_objset_rele(objset_t *os, void *tag);
282 void dmu_objset_disown(objset_t *os, void *tag);
283 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
285 void dmu_objset_evict_dbufs(objset_t *os);
286 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
287 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
288 int dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
289 struct nvlist *snaps);
290 int dmu_objset_clone(const char *name, const char *origin);
291 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
292 struct nvlist *errlist);
293 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
294 int dmu_objset_snapshot_tmp(const char *, const char *, int);
295 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
297 void dmu_objset_byteswap(void *buf, size_t size);
298 int dsl_dataset_rename_snapshot(const char *fsname,
299 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
300 int dmu_objset_remap_indirects(const char *fsname);
302 typedef struct dmu_buf {
303 uint64_t db_object; /* object that this buffer is part of */
304 uint64_t db_offset; /* byte offset in this object */
305 uint64_t db_size; /* size of buffer in bytes */
306 void *db_data; /* data in buffer */
310 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
312 #define DMU_POOL_DIRECTORY_OBJECT 1
313 #define DMU_POOL_CONFIG "config"
314 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
315 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
316 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
317 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
318 #define DMU_POOL_ROOT_DATASET "root_dataset"
319 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
320 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
321 #define DMU_POOL_ERRLOG_LAST "errlog_last"
322 #define DMU_POOL_SPARES "spares"
323 #define DMU_POOL_DEFLATE "deflate"
324 #define DMU_POOL_HISTORY "history"
325 #define DMU_POOL_PROPS "pool_props"
326 #define DMU_POOL_L2CACHE "l2cache"
327 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
328 #define DMU_POOL_DDT "DDT-%s-%s-%s"
329 #define DMU_POOL_DDT_STATS "DDT-statistics"
330 #define DMU_POOL_CREATION_VERSION "creation_version"
331 #define DMU_POOL_SCAN "scan"
332 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
333 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
334 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
335 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
336 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
337 #define DMU_POOL_REMOVING "com.delphix:removing"
338 #define DMU_POOL_OBSOLETE_BPOBJ "com.delphix:obsolete_bpobj"
339 #define DMU_POOL_CONDENSING_INDIRECT "com.delphix:condensing_indirect"
340 #define DMU_POOL_ZPOOL_CHECKPOINT "com.delphix:zpool_checkpoint"
343 * Allocate an object from this objset. The range of object numbers
344 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
346 * The transaction must be assigned to a txg. The newly allocated
347 * object will be "held" in the transaction (ie. you can modify the
348 * newly allocated object in this transaction).
350 * dmu_object_alloc() chooses an object and returns it in *objectp.
352 * dmu_object_claim() allocates a specific object number. If that
353 * number is already allocated, it fails and returns EEXIST.
355 * Return 0 on success, or ENOSPC or EEXIST as specified above.
357 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
358 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
359 uint64_t dmu_object_alloc_ibs(objset_t *os, dmu_object_type_t ot, int blocksize,
360 int indirect_blockshift,
361 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
362 uint64_t dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot,
363 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
364 int dnodesize, dmu_tx_t *tx);
365 int dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
366 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
367 int dnodesize, dmu_tx_t *tx);
368 int dmu_object_reclaim_dnsize(objset_t *os, uint64_t object,
369 dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype,
370 int bonuslen, int dnodesize, dmu_tx_t *txp);
371 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
372 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
373 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
374 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
377 * Free an object from this objset.
379 * The object's data will be freed as well (ie. you don't need to call
380 * dmu_free(object, 0, -1, tx)).
382 * The object need not be held in the transaction.
384 * If there are any holds on this object's buffers (via dmu_buf_hold()),
385 * or tx holds on the object (via dmu_tx_hold_object()), you can not
386 * free it; it fails and returns EBUSY.
388 * If the object is not allocated, it fails and returns ENOENT.
390 * Return 0 on success, or EBUSY or ENOENT as specified above.
392 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
395 * Find the next allocated or free object.
397 * The objectp parameter is in-out. It will be updated to be the next
398 * object which is allocated. Ignore objects which have not been
399 * modified since txg.
401 * XXX Can only be called on a objset with no dirty data.
403 * Returns 0 on success, or ENOENT if there are no more objects.
405 int dmu_object_next(objset_t *os, uint64_t *objectp,
406 boolean_t hole, uint64_t txg);
409 * Set the data blocksize for an object.
411 * The object cannot have any blocks allcated beyond the first. If
412 * the first block is allocated already, the new size must be greater
413 * than the current block size. If these conditions are not met,
414 * ENOTSUP will be returned.
416 * Returns 0 on success, or EBUSY if there are any holds on the object
417 * contents, or ENOTSUP as described above.
419 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
420 int ibs, dmu_tx_t *tx);
423 * Set the checksum property on a dnode. The new checksum algorithm will
424 * apply to all newly written blocks; existing blocks will not be affected.
426 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
430 * Set the compress property on a dnode. The new compression algorithm will
431 * apply to all newly written blocks; existing blocks will not be affected.
433 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
436 int dmu_object_remap_indirects(objset_t *os, uint64_t object, uint64_t txg);
439 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
440 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
441 int compressed_size, int byteorder, dmu_tx_t *tx);
444 * Decide how to write a block: checksum, compression, number of copies, etc.
446 #define WP_NOFILL 0x1
447 #define WP_DMU_SYNC 0x2
450 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
451 struct zio_prop *zp);
453 * The bonus data is accessed more or less like a regular buffer.
454 * You must dmu_bonus_hold() to get the buffer, which will give you a
455 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
456 * data. As with any normal buffer, you must call dmu_buf_will_dirty()
457 * before modifying it, and the
458 * object must be held in an assigned transaction before calling
459 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
460 * buffer as well. You must release your hold with dmu_buf_rele().
462 * Returns ENOENT, EIO, or 0.
464 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
465 int dmu_bonus_max(void);
466 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
467 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
468 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
469 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
472 * Special spill buffer support used by "SA" framework
475 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
476 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
477 void *tag, dmu_buf_t **dbp);
478 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
481 * Obtain the DMU buffer from the specified object which contains the
482 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
483 * that it will remain in memory. You must release the hold with
484 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
485 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
487 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
488 * on the returned buffer before reading or writing the buffer's
489 * db_data. The comments for those routines describe what particular
490 * operations are valid after calling them.
492 * The object number must be a valid, allocated object number.
494 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
495 void *tag, dmu_buf_t **, int flags);
496 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
497 void *tag, dmu_buf_t **dbp, int flags);
500 * Add a reference to a dmu buffer that has already been held via
501 * dmu_buf_hold() in the current context.
503 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
506 * Attempt to add a reference to a dmu buffer that is in an unknown state,
507 * using a pointer that may have been invalidated by eviction processing.
508 * The request will succeed if the passed in dbuf still represents the
509 * same os/object/blkid, is ineligible for eviction, and has at least
510 * one hold by a user other than the syncer.
512 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
513 uint64_t blkid, void *tag);
515 void dmu_buf_rele(dmu_buf_t *db, void *tag);
516 uint64_t dmu_buf_refcount(dmu_buf_t *db);
519 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
520 * range of an object. A pointer to an array of dmu_buf_t*'s is
521 * returned (in *dbpp).
523 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
524 * frees the array. The hold on the array of buffers MUST be released
525 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
526 * individually with dmu_buf_rele.
528 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
529 uint64_t length, boolean_t read, void *tag,
530 int *numbufsp, dmu_buf_t ***dbpp);
531 int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
532 boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp,
534 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
536 typedef void dmu_buf_evict_func_t(void *user_ptr);
539 * A DMU buffer user object may be associated with a dbuf for the
540 * duration of its lifetime. This allows the user of a dbuf (client)
541 * to attach private data to a dbuf (e.g. in-core only data such as a
542 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
543 * when that dbuf has been evicted. Clients typically respond to the
544 * eviction notification by freeing their private data, thus ensuring
545 * the same lifetime for both dbuf and private data.
547 * The mapping from a dmu_buf_user_t to any client private data is the
548 * client's responsibility. All current consumers of the API with private
549 * data embed a dmu_buf_user_t as the first member of the structure for
550 * their private data. This allows conversions between the two types
551 * with a simple cast. Since the DMU buf user API never needs access
552 * to the private data, other strategies can be employed if necessary
553 * or convenient for the client (e.g. using container_of() to do the
554 * conversion for private data that cannot have the dmu_buf_user_t as
557 * Eviction callbacks are executed without the dbuf mutex held or any
558 * other type of mechanism to guarantee that the dbuf is still available.
559 * For this reason, users must assume the dbuf has already been freed
560 * and not reference the dbuf from the callback context.
562 * Users requesting "immediate eviction" are notified as soon as the dbuf
563 * is only referenced by dirty records (dirties == holds). Otherwise the
564 * notification occurs after eviction processing for the dbuf begins.
566 typedef struct dmu_buf_user {
568 * Asynchronous user eviction callback state.
570 taskq_ent_t dbu_tqent;
573 * This instance's eviction function pointers.
575 * dbu_evict_func_sync is called synchronously and then
576 * dbu_evict_func_async is executed asynchronously on a taskq.
578 dmu_buf_evict_func_t *dbu_evict_func_sync;
579 dmu_buf_evict_func_t *dbu_evict_func_async;
582 * Pointer to user's dbuf pointer. NULL for clients that do
583 * not associate a dbuf with their user data.
585 * The dbuf pointer is cleared upon eviction so as to catch
586 * use-after-evict bugs in clients.
588 dmu_buf_t **dbu_clear_on_evict_dbufp;
593 * Initialize the given dmu_buf_user_t instance with the eviction function
594 * evict_func, to be called when the user is evicted.
596 * NOTE: This function should only be called once on a given dmu_buf_user_t.
597 * To allow enforcement of this, dbu must already be zeroed on entry.
601 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
602 dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
604 ASSERT(dbu->dbu_evict_func_sync == NULL);
605 ASSERT(dbu->dbu_evict_func_async == NULL);
607 /* must have at least one evict func */
608 IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
609 dbu->dbu_evict_func_sync = evict_func_sync;
610 dbu->dbu_evict_func_async = evict_func_async;
612 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
617 * Attach user data to a dbuf and mark it for normal (when the dbuf's
618 * data is cleared or its reference count goes to zero) eviction processing.
620 * Returns NULL on success, or the existing user if another user currently
623 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
626 * Attach user data to a dbuf and mark it for immediate (its dirty and
627 * reference counts are equal) eviction processing.
629 * Returns NULL on success, or the existing user if another user currently
632 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
635 * Replace the current user of a dbuf.
637 * If given the current user of a dbuf, replaces the dbuf's user with
638 * "new_user" and returns the user data pointer that was replaced.
639 * Otherwise returns the current, and unmodified, dbuf user pointer.
641 void *dmu_buf_replace_user(dmu_buf_t *db,
642 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
645 * Remove the specified user data for a DMU buffer.
647 * Returns the user that was removed on success, or the current user if
648 * another user currently owns the buffer.
650 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
653 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
655 void *dmu_buf_get_user(dmu_buf_t *db);
657 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
658 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
659 void dmu_buf_dnode_exit(dmu_buf_t *db);
661 /* Block until any in-progress dmu buf user evictions complete. */
662 void dmu_buf_user_evict_wait(void);
665 * Returns the blkptr associated with this dbuf, or NULL if not set.
667 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
670 * Indicate that you are going to modify the buffer's data (db_data).
672 * The transaction (tx) must be assigned to a txg (ie. you've called
673 * dmu_tx_assign()). The buffer's object must be held in the tx
674 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
676 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
679 * You must create a transaction, then hold the objects which you will
680 * (or might) modify as part of this transaction. Then you must assign
681 * the transaction to a transaction group. Once the transaction has
682 * been assigned, you can modify buffers which belong to held objects as
683 * part of this transaction. You can't modify buffers before the
684 * transaction has been assigned; you can't modify buffers which don't
685 * belong to objects which this transaction holds; you can't hold
686 * objects once the transaction has been assigned. You may hold an
687 * object which you are going to free (with dmu_object_free()), but you
690 * You can abort the transaction before it has been assigned.
692 * Note that you may hold buffers (with dmu_buf_hold) at any time,
693 * regardless of transaction state.
696 #define DMU_NEW_OBJECT (-1ULL)
697 #define DMU_OBJECT_END (-1ULL)
699 dmu_tx_t *dmu_tx_create(objset_t *os);
700 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
701 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
703 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
705 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
707 void dmu_tx_hold_remap_l1indirect(dmu_tx_t *tx, uint64_t object);
708 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
709 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
711 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
712 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
713 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
714 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
715 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
716 void dmu_tx_abort(dmu_tx_t *tx);
717 int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
718 void dmu_tx_wait(dmu_tx_t *tx);
719 void dmu_tx_commit(dmu_tx_t *tx);
720 void dmu_tx_mark_netfree(dmu_tx_t *tx);
723 * To register a commit callback, dmu_tx_callback_register() must be called.
725 * dcb_data is a pointer to caller private data that is passed on as a
726 * callback parameter. The caller is responsible for properly allocating and
729 * When registering a callback, the transaction must be already created, but
730 * it cannot be committed or aborted. It can be assigned to a txg or not.
732 * The callback will be called after the transaction has been safely written
733 * to stable storage and will also be called if the dmu_tx is aborted.
734 * If there is any error which prevents the transaction from being committed to
735 * disk, the callback will be called with a value of error != 0.
737 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
739 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
743 * Free up the data blocks for a defined range of a file. If size is
744 * -1, the range from offset to end-of-file is freed.
746 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
747 uint64_t size, dmu_tx_t *tx);
748 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
750 int dmu_free_long_object(objset_t *os, uint64_t object);
753 * Convenience functions.
755 * Canfail routines will return 0 on success, or an errno if there is a
756 * nonrecoverable I/O error.
758 #define DMU_READ_PREFETCH 0 /* prefetch */
759 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
760 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
761 void *buf, uint32_t flags);
762 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
764 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
765 const void *buf, dmu_tx_t *tx);
766 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
767 const void *buf, dmu_tx_t *tx);
768 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
770 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
771 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
772 int dmu_read_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size);
773 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
775 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
777 int dmu_write_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size,
781 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
782 uint64_t size, struct page *pp, dmu_tx_t *tx);
784 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
785 uint64_t size, struct vm_page **ppa, dmu_tx_t *tx);
786 int dmu_read_pages(objset_t *os, uint64_t object, vm_page_t *ma, int count,
787 int *rbehind, int *rahead, int last_size);
790 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
791 void dmu_return_arcbuf(struct arc_buf *buf);
792 void dmu_assign_arcbuf_dnode(dnode_t *handle, uint64_t offset,
793 struct arc_buf *buf, dmu_tx_t *tx);
794 void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
796 int dmu_xuio_init(struct xuio *uio, int niov);
797 void dmu_xuio_fini(struct xuio *uio);
798 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
800 int dmu_xuio_cnt(struct xuio *uio);
801 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
802 void dmu_xuio_clear(struct xuio *uio, int i);
803 void xuio_stat_wbuf_copied(void);
804 void xuio_stat_wbuf_nocopy(void);
806 extern boolean_t zfs_prefetch_disable;
807 extern int zfs_max_recordsize;
810 * Asynchronously try to read in the data.
812 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
813 uint64_t len, enum zio_priority pri);
815 typedef struct dmu_object_info {
816 /* All sizes are in bytes unless otherwise indicated. */
817 uint32_t doi_data_block_size;
818 uint32_t doi_metadata_block_size;
819 dmu_object_type_t doi_type;
820 dmu_object_type_t doi_bonus_type;
821 uint64_t doi_bonus_size;
822 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
823 uint8_t doi_checksum;
824 uint8_t doi_compress;
827 uint64_t doi_dnodesize;
828 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
829 uint64_t doi_max_offset;
830 uint64_t doi_fill_count; /* number of non-empty blocks */
833 typedef void arc_byteswap_func_t(void *buf, size_t size);
835 typedef struct dmu_object_type_info {
836 dmu_object_byteswap_t ot_byteswap;
837 boolean_t ot_metadata;
838 boolean_t ot_dbuf_metadata_cache;
840 } dmu_object_type_info_t;
842 typedef struct dmu_object_byteswap_info {
843 arc_byteswap_func_t *ob_func;
845 } dmu_object_byteswap_info_t;
847 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
848 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
851 * Get information on a DMU object.
853 * Return 0 on success or ENOENT if object is not allocated.
855 * If doi is NULL, just indicates whether the object exists.
857 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
858 void __dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
859 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
860 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
861 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
862 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
864 * Like dmu_object_info_from_db, but faster still when you only care about
865 * the size. This is specifically optimized for zfs_getattr().
867 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
868 u_longlong_t *nblk512);
870 void dmu_object_dnsize_from_db(dmu_buf_t *db, int *dnsize);
872 typedef struct dmu_objset_stats {
873 uint64_t dds_num_clones; /* number of clones of this */
874 uint64_t dds_creation_txg;
876 dmu_objset_type_t dds_type;
877 uint8_t dds_is_snapshot;
878 uint8_t dds_inconsistent;
879 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
880 } dmu_objset_stats_t;
883 * Get stats on a dataset.
885 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
888 * Add entries to the nvlist for all the objset's properties. See
889 * zfs_prop_table[] and zfs(1m) for details on the properties.
891 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
894 * Get the space usage statistics for statvfs().
896 * refdbytes is the amount of space "referenced" by this objset.
897 * availbytes is the amount of space available to this objset, taking
898 * into account quotas & reservations, assuming that no other objsets
899 * use the space first. These values correspond to the 'referenced' and
900 * 'available' properties, described in the zfs(1m) manpage.
902 * usedobjs and availobjs are the number of objects currently allocated,
905 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
906 uint64_t *usedobjsp, uint64_t *availobjsp);
909 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
910 * (Contrast with the ds_guid which is a 64-bit ID that will never
911 * change, so there is a small probability that it will collide.)
913 uint64_t dmu_objset_fsid_guid(objset_t *os);
916 * Get the [cm]time for an objset's snapshot dir
918 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
920 int dmu_objset_is_snapshot(objset_t *os);
922 extern struct spa *dmu_objset_spa(objset_t *os);
923 extern struct zilog *dmu_objset_zil(objset_t *os);
924 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
925 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
926 extern void dmu_objset_name(objset_t *os, char *buf);
927 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
928 extern uint64_t dmu_objset_id(objset_t *os);
929 extern uint64_t dmu_objset_dnodesize(objset_t *os);
930 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
931 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
932 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
933 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
934 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
935 int maxlen, boolean_t *conflict);
936 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
937 uint64_t *idp, uint64_t *offp);
939 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
940 void *bonus, uint64_t *userp, uint64_t *groupp);
941 extern void dmu_objset_register_type(dmu_objset_type_t ost,
942 objset_used_cb_t *cb);
943 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
944 extern void *dmu_objset_get_user(objset_t *os);
947 * Return the txg number for the given assigned transaction.
949 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
953 * If a parent zio is provided this function initiates a write on the
954 * provided buffer as a child of the parent zio.
955 * In the absence of a parent zio, the write is completed synchronously.
956 * At write completion, blk is filled with the bp of the written block.
957 * Note that while the data covered by this function will be on stable
958 * storage when the write completes this new data does not become a
959 * permanent part of the file until the associated transaction commits.
963 * {zfs,zvol,ztest}_get_done() args
967 struct blkptr *zgd_bp;
973 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
974 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
977 * Find the next hole or data block in file starting at *off
978 * Return found offset in *off. Return ESRCH for end of file.
980 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
984 * Check if a DMU object has any dirty blocks. If so, sync out
985 * all pending transaction groups. Otherwise, this function
986 * does not alter DMU state. This could be improved to only sync
987 * out the necessary transaction groups for this particular
990 int dmu_object_wait_synced(objset_t *os, uint64_t object);
993 * Initial setup and final teardown.
995 extern void dmu_init(void);
996 extern void dmu_fini(void);
998 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
999 uint64_t object, uint64_t offset, int len);
1000 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
1001 dmu_traverse_cb_t cb, void *arg);
1002 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
1003 struct file *fp, offset_t *offp);
1006 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
1007 extern uint64_t zfs_crc64_table[256];
1009 extern int zfs_mdcomp_disable;
1015 #endif /* _SYS_DMU_H */