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, 2017 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.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
31 /* Portions Copyright 2010 Robert Milkowski */
37 * This file describes the interface that the DMU provides for its
40 * The DMU also interacts with the SPA. That interface is described in
44 #include <sys/zfs_context.h>
45 #include <sys/inttypes.h>
47 #include <sys/fs/zfs.h>
48 #include <sys/zio_compress.h>
49 #include <sys/zio_priority.h>
68 struct zbookmark_phys;
74 struct dsl_crypto_params;
76 typedef struct objset objset_t;
77 typedef struct dmu_tx dmu_tx_t;
78 typedef struct dsl_dir dsl_dir_t;
79 typedef struct dnode dnode_t;
81 typedef enum dmu_object_byteswap {
93 * Allocating a new byteswap type number makes the on-disk format
94 * incompatible with any other format that uses the same number.
96 * Data can usually be structured to work with one of the
97 * DMU_BSWAP_UINT* or DMU_BSWAP_ZAP types.
100 } dmu_object_byteswap_t;
102 #define DMU_OT_NEWTYPE 0x80
103 #define DMU_OT_METADATA 0x40
104 #define DMU_OT_ENCRYPTED 0x20
105 #define DMU_OT_BYTESWAP_MASK 0x1f
108 * Defines a uint8_t object type. Object types specify if the data
109 * in the object is metadata (boolean) and how to byteswap the data
110 * (dmu_object_byteswap_t).
112 #define DMU_OT(byteswap, metadata, encrypted) \
114 ((metadata) ? DMU_OT_METADATA : 0) | \
115 ((encrypted) ? DMU_OT_ENCRYPTED : 0) | \
116 ((byteswap) & DMU_OT_BYTESWAP_MASK))
118 #define DMU_OT_IS_VALID(ot) (((ot) & DMU_OT_NEWTYPE) ? \
119 ((ot) & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS : \
120 (ot) < DMU_OT_NUMTYPES)
122 #define DMU_OT_IS_METADATA(ot) (((ot) & DMU_OT_NEWTYPE) ? \
123 ((ot) & DMU_OT_METADATA) : \
124 dmu_ot[(int)(ot)].ot_metadata)
126 #define DMU_OT_IS_ENCRYPTED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
127 ((ot) & DMU_OT_ENCRYPTED) : \
128 dmu_ot[(int)(ot)].ot_encrypt)
131 * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
132 * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
133 * is repurposed for embedded BPs.
135 #define DMU_OT_HAS_FILL(ot) \
136 ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
138 #define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
139 ((ot) & DMU_OT_BYTESWAP_MASK) : \
140 dmu_ot[(int)(ot)].ot_byteswap)
142 typedef enum dmu_object_type {
145 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
146 DMU_OT_OBJECT_ARRAY, /* UINT64 */
147 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
148 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
149 DMU_OT_BPOBJ, /* UINT64 */
150 DMU_OT_BPOBJ_HDR, /* UINT64 */
152 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
153 DMU_OT_SPACE_MAP, /* UINT64 */
155 DMU_OT_INTENT_LOG, /* UINT64 */
157 DMU_OT_DNODE, /* DNODE */
158 DMU_OT_OBJSET, /* OBJSET */
160 DMU_OT_DSL_DIR, /* UINT64 */
161 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
162 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
163 DMU_OT_DSL_PROPS, /* ZAP */
164 DMU_OT_DSL_DATASET, /* UINT64 */
166 DMU_OT_ZNODE, /* ZNODE */
167 DMU_OT_OLDACL, /* Old ACL */
168 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
169 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
170 DMU_OT_MASTER_NODE, /* ZAP */
171 DMU_OT_UNLINKED_SET, /* ZAP */
173 DMU_OT_ZVOL, /* UINT8 */
174 DMU_OT_ZVOL_PROP, /* ZAP */
175 /* other; for testing only! */
176 DMU_OT_PLAIN_OTHER, /* UINT8 */
177 DMU_OT_UINT64_OTHER, /* UINT64 */
178 DMU_OT_ZAP_OTHER, /* ZAP */
179 /* new object types: */
180 DMU_OT_ERROR_LOG, /* ZAP */
181 DMU_OT_SPA_HISTORY, /* UINT8 */
182 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
183 DMU_OT_POOL_PROPS, /* ZAP */
184 DMU_OT_DSL_PERMS, /* ZAP */
185 DMU_OT_ACL, /* ACL */
186 DMU_OT_SYSACL, /* SYSACL */
187 DMU_OT_FUID, /* FUID table (Packed NVLIST UINT8) */
188 DMU_OT_FUID_SIZE, /* FUID table size UINT64 */
189 DMU_OT_NEXT_CLONES, /* ZAP */
190 DMU_OT_SCAN_QUEUE, /* ZAP */
191 DMU_OT_USERGROUP_USED, /* ZAP */
192 DMU_OT_USERGROUP_QUOTA, /* ZAP */
193 DMU_OT_USERREFS, /* ZAP */
194 DMU_OT_DDT_ZAP, /* ZAP */
195 DMU_OT_DDT_STATS, /* ZAP */
196 DMU_OT_SA, /* System attr */
197 DMU_OT_SA_MASTER_NODE, /* ZAP */
198 DMU_OT_SA_ATTR_REGISTRATION, /* ZAP */
199 DMU_OT_SA_ATTR_LAYOUTS, /* ZAP */
200 DMU_OT_SCAN_XLATE, /* ZAP */
201 DMU_OT_DEDUP, /* fake dedup BP from ddt_bp_create() */
202 DMU_OT_DEADLIST, /* ZAP */
203 DMU_OT_DEADLIST_HDR, /* UINT64 */
204 DMU_OT_DSL_CLONES, /* ZAP */
205 DMU_OT_BPOBJ_SUBOBJ, /* UINT64 */
207 * Do not allocate new object types here. Doing so makes the on-disk
208 * format incompatible with any other format that uses the same object
211 * When creating an object which does not have one of the above types
212 * use the DMU_OTN_* type with the correct byteswap and metadata
215 * The DMU_OTN_* types do not have entries in the dmu_ot table,
216 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
217 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
218 * and DMU_OTN_* types).
223 * Names for valid types declared with DMU_OT().
225 DMU_OTN_UINT8_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_FALSE),
226 DMU_OTN_UINT8_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_FALSE),
227 DMU_OTN_UINT16_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_FALSE),
228 DMU_OTN_UINT16_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_FALSE),
229 DMU_OTN_UINT32_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_FALSE),
230 DMU_OTN_UINT32_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_FALSE),
231 DMU_OTN_UINT64_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_FALSE),
232 DMU_OTN_UINT64_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_FALSE),
233 DMU_OTN_ZAP_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_FALSE),
234 DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_FALSE),
236 DMU_OTN_UINT8_ENC_DATA = DMU_OT(DMU_BSWAP_UINT8, B_FALSE, B_TRUE),
237 DMU_OTN_UINT8_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT8, B_TRUE, B_TRUE),
238 DMU_OTN_UINT16_ENC_DATA = DMU_OT(DMU_BSWAP_UINT16, B_FALSE, B_TRUE),
239 DMU_OTN_UINT16_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT16, B_TRUE, B_TRUE),
240 DMU_OTN_UINT32_ENC_DATA = DMU_OT(DMU_BSWAP_UINT32, B_FALSE, B_TRUE),
241 DMU_OTN_UINT32_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT32, B_TRUE, B_TRUE),
242 DMU_OTN_UINT64_ENC_DATA = DMU_OT(DMU_BSWAP_UINT64, B_FALSE, B_TRUE),
243 DMU_OTN_UINT64_ENC_METADATA = DMU_OT(DMU_BSWAP_UINT64, B_TRUE, B_TRUE),
244 DMU_OTN_ZAP_ENC_DATA = DMU_OT(DMU_BSWAP_ZAP, B_FALSE, B_TRUE),
245 DMU_OTN_ZAP_ENC_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE, B_TRUE),
248 typedef enum txg_how {
254 void byteswap_uint64_array(void *buf, size_t size);
255 void byteswap_uint32_array(void *buf, size_t size);
256 void byteswap_uint16_array(void *buf, size_t size);
257 void byteswap_uint8_array(void *buf, size_t size);
258 void zap_byteswap(void *buf, size_t size);
259 void zfs_oldacl_byteswap(void *buf, size_t size);
260 void zfs_acl_byteswap(void *buf, size_t size);
261 void zfs_znode_byteswap(void *buf, size_t size);
263 #define DS_FIND_SNAPSHOTS (1<<0)
264 #define DS_FIND_CHILDREN (1<<1)
265 #define DS_FIND_SERIALIZE (1<<2)
268 * The maximum number of bytes that can be accessed as part of one
269 * operation, including metadata.
271 #define DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
272 #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
274 #define DMU_USERUSED_OBJECT (-1ULL)
275 #define DMU_GROUPUSED_OBJECT (-2ULL)
278 * Zap prefix for object accounting in DMU_{USER,GROUP}USED_OBJECT.
280 #define DMU_OBJACCT_PREFIX "obj-"
281 #define DMU_OBJACCT_PREFIX_LEN 4
284 * artificial blkids for bonus buffer and spill blocks
286 #define DMU_BONUS_BLKID (-1ULL)
287 #define DMU_SPILL_BLKID (-2ULL)
290 * Public routines to create, destroy, open, and close objsets.
292 typedef void dmu_objset_create_sync_func_t(objset_t *os, void *arg,
293 cred_t *cr, dmu_tx_t *tx);
295 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
296 int dmu_objset_own(const char *name, dmu_objset_type_t type,
297 boolean_t readonly, boolean_t key_required, void *tag, objset_t **osp);
298 void dmu_objset_rele(objset_t *os, void *tag);
299 void dmu_objset_disown(objset_t *os, boolean_t key_required, void *tag);
300 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
302 void dmu_objset_evict_dbufs(objset_t *os);
303 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
304 struct dsl_crypto_params *dcp, dmu_objset_create_sync_func_t func,
306 int dmu_objset_clone(const char *name, const char *origin);
307 int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
308 struct nvlist *errlist);
309 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
310 int dmu_objset_snapshot_tmp(const char *, const char *, int);
311 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
313 void dmu_objset_byteswap(void *buf, size_t size);
314 int dsl_dataset_rename_snapshot(const char *fsname,
315 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
317 typedef struct dmu_buf {
318 uint64_t db_object; /* object that this buffer is part of */
319 uint64_t db_offset; /* byte offset in this object */
320 uint64_t db_size; /* size of buffer in bytes */
321 void *db_data; /* data in buffer */
325 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
327 #define DMU_POOL_DIRECTORY_OBJECT 1
328 #define DMU_POOL_CONFIG "config"
329 #define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
330 #define DMU_POOL_FEATURES_FOR_READ "features_for_read"
331 #define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
332 #define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
333 #define DMU_POOL_ROOT_DATASET "root_dataset"
334 #define DMU_POOL_SYNC_BPOBJ "sync_bplist"
335 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
336 #define DMU_POOL_ERRLOG_LAST "errlog_last"
337 #define DMU_POOL_SPARES "spares"
338 #define DMU_POOL_DEFLATE "deflate"
339 #define DMU_POOL_HISTORY "history"
340 #define DMU_POOL_PROPS "pool_props"
341 #define DMU_POOL_L2CACHE "l2cache"
342 #define DMU_POOL_TMP_USERREFS "tmp_userrefs"
343 #define DMU_POOL_DDT "DDT-%s-%s-%s"
344 #define DMU_POOL_DDT_STATS "DDT-statistics"
345 #define DMU_POOL_CREATION_VERSION "creation_version"
346 #define DMU_POOL_SCAN "scan"
347 #define DMU_POOL_FREE_BPOBJ "free_bpobj"
348 #define DMU_POOL_BPTREE_OBJ "bptree_obj"
349 #define DMU_POOL_EMPTY_BPOBJ "empty_bpobj"
350 #define DMU_POOL_CHECKSUM_SALT "org.illumos:checksum_salt"
351 #define DMU_POOL_VDEV_ZAP_MAP "com.delphix:vdev_zap_map"
354 * Allocate an object from this objset. The range of object numbers
355 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
357 * The transaction must be assigned to a txg. The newly allocated
358 * object will be "held" in the transaction (ie. you can modify the
359 * newly allocated object in this transaction).
361 * dmu_object_alloc() chooses an object and returns it in *objectp.
363 * dmu_object_claim() allocates a specific object number. If that
364 * number is already allocated, it fails and returns EEXIST.
366 * Return 0 on success, or ENOSPC or EEXIST as specified above.
368 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
369 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
370 uint64_t dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot,
371 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
372 int dnodesize, dmu_tx_t *tx);
373 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
374 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
375 int dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
376 int blocksize, dmu_object_type_t bonus_type, int bonus_len,
377 int dnodesize, dmu_tx_t *tx);
378 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
379 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
380 int dmu_object_reclaim_dnsize(objset_t *os, uint64_t object,
381 dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype,
382 int bonuslen, int dnodesize, dmu_tx_t *txp);
385 * Free an object from this objset.
387 * The object's data will be freed as well (ie. you don't need to call
388 * dmu_free(object, 0, -1, tx)).
390 * The object need not be held in the transaction.
392 * If there are any holds on this object's buffers (via dmu_buf_hold()),
393 * or tx holds on the object (via dmu_tx_hold_object()), you can not
394 * free it; it fails and returns EBUSY.
396 * If the object is not allocated, it fails and returns ENOENT.
398 * Return 0 on success, or EBUSY or ENOENT as specified above.
400 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
403 * Find the next allocated or free object.
405 * The objectp parameter is in-out. It will be updated to be the next
406 * object which is allocated. Ignore objects which have not been
407 * modified since txg.
409 * XXX Can only be called on a objset with no dirty data.
411 * Returns 0 on success, or ENOENT if there are no more objects.
413 int dmu_object_next(objset_t *os, uint64_t *objectp,
414 boolean_t hole, uint64_t txg);
417 * Set the number of levels on a dnode. nlevels must be greater than the
418 * current number of levels or an EINVAL will be returned.
420 int dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels,
424 * Set the data blocksize for an object.
426 * The object cannot have any blocks allcated beyond the first. If
427 * the first block is allocated already, the new size must be greater
428 * than the current block size. If these conditions are not met,
429 * ENOTSUP will be returned.
431 * Returns 0 on success, or EBUSY if there are any holds on the object
432 * contents, or ENOTSUP as described above.
434 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
435 int ibs, dmu_tx_t *tx);
438 * Set the checksum property on a dnode. The new checksum algorithm will
439 * apply to all newly written blocks; existing blocks will not be affected.
441 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
445 * Set the compress property on a dnode. The new compression algorithm will
446 * apply to all newly written blocks; existing blocks will not be affected.
448 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
451 int dmu_object_dirty_raw(objset_t *os, uint64_t object, dmu_tx_t *tx);
453 void dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
454 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
455 int compressed_size, int byteorder, dmu_tx_t *tx);
458 * Decide how to write a block: checksum, compression, number of copies, etc.
460 #define WP_NOFILL 0x1
461 #define WP_DMU_SYNC 0x2
464 void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp,
465 struct zio_prop *zp);
468 * The bonus data is accessed more or less like a regular buffer.
469 * You must dmu_bonus_hold() to get the buffer, which will give you a
470 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
471 * data. As with any normal buffer, you must call dmu_buf_read() to
472 * read db_data, dmu_buf_will_dirty() before modifying it, and the
473 * object must be held in an assigned transaction before calling
474 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
475 * buffer as well. You must release what you hold with dmu_buf_rele().
477 * Returns ENOENT, EIO, or 0.
479 int dmu_bonus_hold_impl(objset_t *os, uint64_t object, void *tag,
480 uint32_t flags, dmu_buf_t **dbp);
481 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
482 int dmu_bonus_max(void);
483 int dmu_set_bonus(dmu_buf_t *, int, dmu_tx_t *);
484 int dmu_set_bonustype(dmu_buf_t *, dmu_object_type_t, dmu_tx_t *);
485 dmu_object_type_t dmu_get_bonustype(dmu_buf_t *);
486 int dmu_rm_spill(objset_t *, uint64_t, dmu_tx_t *);
489 * Special spill buffer support used by "SA" framework
492 int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
493 int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags,
494 void *tag, dmu_buf_t **dbp);
495 int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
498 * Obtain the DMU buffer from the specified object which contains the
499 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
500 * that it will remain in memory. You must release the hold with
501 * dmu_buf_rele(). You must not access the dmu_buf_t after releasing
502 * what you hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
504 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
505 * on the returned buffer before reading or writing the buffer's
506 * db_data. The comments for those routines describe what particular
507 * operations are valid after calling them.
509 * The object number must be a valid, allocated object number.
511 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
512 void *tag, dmu_buf_t **, int flags);
513 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
514 void *tag, dmu_buf_t **dbp, int flags);
517 * Add a reference to a dmu buffer that has already been held via
518 * dmu_buf_hold() in the current context.
520 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
523 * Attempt to add a reference to a dmu buffer that is in an unknown state,
524 * using a pointer that may have been invalidated by eviction processing.
525 * The request will succeed if the passed in dbuf still represents the
526 * same os/object/blkid, is ineligible for eviction, and has at least
527 * one hold by a user other than the syncer.
529 boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
530 uint64_t blkid, void *tag);
532 void dmu_buf_rele(dmu_buf_t *db, void *tag);
533 uint64_t dmu_buf_refcount(dmu_buf_t *db);
536 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
537 * range of an object. A pointer to an array of dmu_buf_t*'s is
538 * returned (in *dbpp).
540 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
541 * frees the array. The hold on the array of buffers MUST be released
542 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
543 * individually with dmu_buf_rele.
545 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
546 uint64_t length, boolean_t read, void *tag,
547 int *numbufsp, dmu_buf_t ***dbpp);
548 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
550 typedef void dmu_buf_evict_func_t(void *user_ptr);
553 * A DMU buffer user object may be associated with a dbuf for the
554 * duration of its lifetime. This allows the user of a dbuf (client)
555 * to attach private data to a dbuf (e.g. in-core only data such as a
556 * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
557 * when that dbuf has been evicted. Clients typically respond to the
558 * eviction notification by freeing their private data, thus ensuring
559 * the same lifetime for both dbuf and private data.
561 * The mapping from a dmu_buf_user_t to any client private data is the
562 * client's responsibility. All current consumers of the API with private
563 * data embed a dmu_buf_user_t as the first member of the structure for
564 * their private data. This allows conversions between the two types
565 * with a simple cast. Since the DMU buf user API never needs access
566 * to the private data, other strategies can be employed if necessary
567 * or convenient for the client (e.g. using container_of() to do the
568 * conversion for private data that cannot have the dmu_buf_user_t as
571 * Eviction callbacks are executed without the dbuf mutex held or any
572 * other type of mechanism to guarantee that the dbuf is still available.
573 * For this reason, users must assume the dbuf has already been freed
574 * and not reference the dbuf from the callback context.
576 * Users requesting "immediate eviction" are notified as soon as the dbuf
577 * is only referenced by dirty records (dirties == holds). Otherwise the
578 * notification occurs after eviction processing for the dbuf begins.
580 typedef struct dmu_buf_user {
582 * Asynchronous user eviction callback state.
584 taskq_ent_t dbu_tqent;
587 * This instance's eviction function pointers.
589 * dbu_evict_func_sync is called synchronously and then
590 * dbu_evict_func_async is executed asynchronously on a taskq.
592 dmu_buf_evict_func_t *dbu_evict_func_sync;
593 dmu_buf_evict_func_t *dbu_evict_func_async;
596 * Pointer to user's dbuf pointer. NULL for clients that do
597 * not associate a dbuf with their user data.
599 * The dbuf pointer is cleared upon eviction so as to catch
600 * use-after-evict bugs in clients.
602 dmu_buf_t **dbu_clear_on_evict_dbufp;
607 * Initialize the given dmu_buf_user_t instance with the eviction function
608 * evict_func, to be called when the user is evicted.
610 * NOTE: This function should only be called once on a given dmu_buf_user_t.
611 * To allow enforcement of this, dbu must already be zeroed on entry.
615 dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync,
616 dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp)
618 ASSERT(dbu->dbu_evict_func_sync == NULL);
619 ASSERT(dbu->dbu_evict_func_async == NULL);
621 /* must have at least one evict func */
622 IMPLY(evict_func_sync == NULL, evict_func_async != NULL);
623 dbu->dbu_evict_func_sync = evict_func_sync;
624 dbu->dbu_evict_func_async = evict_func_async;
625 taskq_init_ent(&dbu->dbu_tqent);
627 dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
632 * Attach user data to a dbuf and mark it for normal (when the dbuf's
633 * data is cleared or its reference count goes to zero) eviction processing.
635 * Returns NULL on success, or the existing user if another user currently
638 void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
641 * Attach user data to a dbuf and mark it for immediate (its dirty and
642 * reference counts are equal) eviction processing.
644 * Returns NULL on success, or the existing user if another user currently
647 void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
650 * Replace the current user of a dbuf.
652 * If given the current user of a dbuf, replaces the dbuf's user with
653 * "new_user" and returns the user data pointer that was replaced.
654 * Otherwise returns the current, and unmodified, dbuf user pointer.
656 void *dmu_buf_replace_user(dmu_buf_t *db,
657 dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
660 * Remove the specified user data for a DMU buffer.
662 * Returns the user that was removed on success, or the current user if
663 * another user currently owns the buffer.
665 void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
668 * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
670 void *dmu_buf_get_user(dmu_buf_t *db);
672 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
673 dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
674 void dmu_buf_dnode_exit(dmu_buf_t *db);
676 /* Block until any in-progress dmu buf user evictions complete. */
677 void dmu_buf_user_evict_wait(void);
680 * Returns the blkptr associated with this dbuf, or NULL if not set.
682 struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
685 * Indicate that you are going to modify the buffer's data (db_data).
687 * The transaction (tx) must be assigned to a txg (ie. you've called
688 * dmu_tx_assign()). The buffer's object must be held in the tx
689 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
691 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
692 void dmu_buf_will_change_crypt_params(dmu_buf_t *db, dmu_tx_t *tx);
695 * You must create a transaction, then hold the objects which you will
696 * (or might) modify as part of this transaction. Then you must assign
697 * the transaction to a transaction group. Once the transaction has
698 * been assigned, you can modify buffers which belong to held objects as
699 * part of this transaction. You can't modify buffers before the
700 * transaction has been assigned; you can't modify buffers which don't
701 * belong to objects which this transaction holds; you can't hold
702 * objects once the transaction has been assigned. You may hold an
703 * object which you are going to free (with dmu_object_free()), but you
706 * You can abort the transaction before it has been assigned.
708 * Note that you may hold buffers (with dmu_buf_hold) at any time,
709 * regardless of transaction state.
712 #define DMU_NEW_OBJECT (-1ULL)
713 #define DMU_OBJECT_END (-1ULL)
715 dmu_tx_t *dmu_tx_create(objset_t *os);
716 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
717 void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
719 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
721 void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off,
723 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name);
724 void dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add,
726 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
727 void dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn);
728 void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
729 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
730 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
731 void dmu_tx_abort(dmu_tx_t *tx);
732 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
733 void dmu_tx_wait(dmu_tx_t *tx);
734 void dmu_tx_commit(dmu_tx_t *tx);
735 void dmu_tx_mark_netfree(dmu_tx_t *tx);
738 * To register a commit callback, dmu_tx_callback_register() must be called.
740 * dcb_data is a pointer to caller private data that is passed on as a
741 * callback parameter. The caller is responsible for properly allocating and
744 * When registering a callback, the transaction must be already created, but
745 * it cannot be committed or aborted. It can be assigned to a txg or not.
747 * The callback will be called after the transaction has been safely written
748 * to stable storage and will also be called if the dmu_tx is aborted.
749 * If there is any error which prevents the transaction from being committed to
750 * disk, the callback will be called with a value of error != 0.
752 * When multiple callbacks are registered to the transaction, the callbacks
753 * will be called in reverse order to let Lustre, the only user of commit
754 * callback currently, take the fast path of its commit callback handling.
756 typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
758 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
760 void dmu_tx_do_callbacks(list_t *cb_list, int error);
763 * Free up the data blocks for a defined range of a file. If size is
764 * -1, the range from offset to end-of-file is freed.
766 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
767 uint64_t size, dmu_tx_t *tx);
768 int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
770 int dmu_free_long_range_raw(objset_t *os, uint64_t object, uint64_t offset,
772 int dmu_free_long_object(objset_t *os, uint64_t object);
773 int dmu_free_long_object_raw(objset_t *os, uint64_t object);
776 * Convenience functions.
778 * Canfail routines will return 0 on success, or an errno if there is a
779 * nonrecoverable I/O error.
781 #define DMU_READ_PREFETCH 0 /* prefetch */
782 #define DMU_READ_NO_PREFETCH 1 /* don't prefetch */
783 #define DMU_READ_NO_DECRYPT 2 /* don't decrypt */
784 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
785 void *buf, uint32_t flags);
786 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
788 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
789 const void *buf, dmu_tx_t *tx);
790 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
791 const void *buf, dmu_tx_t *tx);
792 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
795 #include <linux/blkdev_compat.h>
796 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
797 int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
798 int dmu_read_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size);
799 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
801 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
803 int dmu_write_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size,
806 struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
807 void dmu_return_arcbuf(struct arc_buf *buf);
808 void dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset,
809 struct arc_buf *buf, dmu_tx_t *tx);
810 void dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset,
811 struct arc_buf *buf, dmu_tx_t *tx);
812 #define dmu_assign_arcbuf dmu_assign_arcbuf_by_dbuf
813 void dmu_convert_to_raw(dmu_buf_t *handle, boolean_t byteorder,
814 const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx);
815 void dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset,
816 dmu_buf_t *handle, dmu_tx_t *tx);
817 #ifdef HAVE_UIO_ZEROCOPY
818 int dmu_xuio_init(struct xuio *uio, int niov);
819 void dmu_xuio_fini(struct xuio *uio);
820 int dmu_xuio_add(struct xuio *uio, struct arc_buf *abuf, offset_t off,
822 int dmu_xuio_cnt(struct xuio *uio);
823 struct arc_buf *dmu_xuio_arcbuf(struct xuio *uio, int i);
824 void dmu_xuio_clear(struct xuio *uio, int i);
825 #endif /* HAVE_UIO_ZEROCOPY */
826 void xuio_stat_wbuf_copied(void);
827 void xuio_stat_wbuf_nocopy(void);
829 extern int zfs_prefetch_disable;
830 extern int zfs_max_recordsize;
833 * Asynchronously try to read in the data.
835 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
836 uint64_t len, enum zio_priority pri);
838 typedef struct dmu_object_info {
839 /* All sizes are in bytes unless otherwise indicated. */
840 uint32_t doi_data_block_size;
841 uint32_t doi_metadata_block_size;
842 dmu_object_type_t doi_type;
843 dmu_object_type_t doi_bonus_type;
844 uint64_t doi_bonus_size;
845 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
846 uint8_t doi_checksum;
847 uint8_t doi_compress;
850 uint64_t doi_dnodesize;
851 uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
852 uint64_t doi_max_offset;
853 uint64_t doi_fill_count; /* number of non-empty blocks */
856 typedef void (*const arc_byteswap_func_t)(void *buf, size_t size);
858 typedef struct dmu_object_type_info {
859 dmu_object_byteswap_t ot_byteswap;
860 boolean_t ot_metadata;
861 boolean_t ot_encrypt;
863 } dmu_object_type_info_t;
865 typedef const struct dmu_object_byteswap_info {
866 arc_byteswap_func_t ob_func;
868 } dmu_object_byteswap_info_t;
870 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
871 extern const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS];
874 * Get information on a DMU object.
876 * Return 0 on success or ENOENT if object is not allocated.
878 * If doi is NULL, just indicates whether the object exists.
880 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
881 void __dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
882 /* Like dmu_object_info, but faster if you have a held dnode in hand. */
883 void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi);
884 /* Like dmu_object_info, but faster if you have a held dbuf in hand. */
885 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
887 * Like dmu_object_info_from_db, but faster still when you only care about
888 * the size. This is specifically optimized for zfs_getattr().
890 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
891 u_longlong_t *nblk512);
893 void dmu_object_dnsize_from_db(dmu_buf_t *db, int *dnsize);
895 typedef struct dmu_objset_stats {
896 uint64_t dds_num_clones; /* number of clones of this */
897 uint64_t dds_creation_txg;
899 dmu_objset_type_t dds_type;
900 uint8_t dds_is_snapshot;
901 uint8_t dds_inconsistent;
902 char dds_origin[ZFS_MAX_DATASET_NAME_LEN];
903 } dmu_objset_stats_t;
906 * Get stats on a dataset.
908 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
911 * Add entries to the nvlist for all the objset's properties. See
912 * zfs_prop_table[] and zfs(1m) for details on the properties.
914 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
917 * Get the space usage statistics for statvfs().
919 * refdbytes is the amount of space "referenced" by this objset.
920 * availbytes is the amount of space available to this objset, taking
921 * into account quotas & reservations, assuming that no other objsets
922 * use the space first. These values correspond to the 'referenced' and
923 * 'available' properties, described in the zfs(1m) manpage.
925 * usedobjs and availobjs are the number of objects currently allocated,
928 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
929 uint64_t *usedobjsp, uint64_t *availobjsp);
932 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
933 * (Contrast with the ds_guid which is a 64-bit ID that will never
934 * change, so there is a small probability that it will collide.)
936 uint64_t dmu_objset_fsid_guid(objset_t *os);
939 * Get the [cm]time for an objset's snapshot dir
941 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
943 int dmu_objset_is_snapshot(objset_t *os);
945 extern struct spa *dmu_objset_spa(objset_t *os);
946 extern struct zilog *dmu_objset_zil(objset_t *os);
947 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
948 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
949 extern void dmu_objset_name(objset_t *os, char *buf);
950 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
951 extern uint64_t dmu_objset_id(objset_t *os);
952 extern uint64_t dmu_objset_dnodesize(objset_t *os);
953 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
954 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
955 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
956 uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
957 extern int dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *val);
958 extern int dmu_snapshot_realname(objset_t *os, char *name, char *real,
959 int maxlen, boolean_t *conflict);
960 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
961 uint64_t *idp, uint64_t *offp);
963 typedef int objset_used_cb_t(dmu_object_type_t bonustype,
964 void *bonus, uint64_t *userp, uint64_t *groupp);
965 extern void dmu_objset_register_type(dmu_objset_type_t ost,
966 objset_used_cb_t *cb);
967 extern void dmu_objset_set_user(objset_t *os, void *user_ptr);
968 extern void *dmu_objset_get_user(objset_t *os);
971 * Return the txg number for the given assigned transaction.
973 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
977 * If a parent zio is provided this function initiates a write on the
978 * provided buffer as a child of the parent zio.
979 * In the absence of a parent zio, the write is completed synchronously.
980 * At write completion, blk is filled with the bp of the written block.
981 * Note that while the data covered by this function will be on stable
982 * storage when the write completes this new data does not become a
983 * permanent part of the file until the associated transaction commits.
987 * {zfs,zvol,ztest}_get_done() args
991 struct blkptr *zgd_bp;
997 typedef void dmu_sync_cb_t(zgd_t *arg, int error);
998 int dmu_sync(struct zio *zio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd);
1001 * Find the next hole or data block in file starting at *off
1002 * Return found offset in *off. Return ESRCH for end of file.
1004 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
1008 * Initial setup and final teardown.
1010 extern void dmu_init(void);
1011 extern void dmu_fini(void);
1013 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
1014 uint64_t object, uint64_t offset, int len);
1015 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
1016 dmu_traverse_cb_t cb, void *arg);
1018 int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
1019 struct vnode *vp, offset_t *offp);
1022 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
1023 extern uint64_t zfs_crc64_table[256];
1025 extern int zfs_mdcomp_disable;
1031 #endif /* _SYS_DMU_H */