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 2007 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
29 #pragma ident "%Z%%M% %I% %E% SMI"
32 * This file describes the interface that the DMU provides for its
35 * The DMU also interacts with the SPA. That interface is described in
39 #include <sys/types.h>
40 #include <sys/param.h>
65 typedef struct objset objset_t;
66 typedef struct dmu_tx dmu_tx_t;
67 typedef struct dsl_dir dsl_dir_t;
69 typedef enum dmu_object_type {
72 DMU_OT_OBJECT_DIRECTORY, /* ZAP */
73 DMU_OT_OBJECT_ARRAY, /* UINT64 */
74 DMU_OT_PACKED_NVLIST, /* UINT8 (XDR by nvlist_pack/unpack) */
75 DMU_OT_PACKED_NVLIST_SIZE, /* UINT64 */
76 DMU_OT_BPLIST, /* UINT64 */
77 DMU_OT_BPLIST_HDR, /* UINT64 */
79 DMU_OT_SPACE_MAP_HEADER, /* UINT64 */
80 DMU_OT_SPACE_MAP, /* UINT64 */
82 DMU_OT_INTENT_LOG, /* UINT64 */
84 DMU_OT_DNODE, /* DNODE */
85 DMU_OT_OBJSET, /* OBJSET */
87 DMU_OT_DSL_DIR, /* UINT64 */
88 DMU_OT_DSL_DIR_CHILD_MAP, /* ZAP */
89 DMU_OT_DSL_DS_SNAP_MAP, /* ZAP */
90 DMU_OT_DSL_PROPS, /* ZAP */
91 DMU_OT_DSL_DATASET, /* UINT64 */
93 DMU_OT_ZNODE, /* ZNODE */
95 DMU_OT_PLAIN_FILE_CONTENTS, /* UINT8 */
96 DMU_OT_DIRECTORY_CONTENTS, /* ZAP */
97 DMU_OT_MASTER_NODE, /* ZAP */
98 DMU_OT_UNLINKED_SET, /* ZAP */
100 DMU_OT_ZVOL, /* UINT8 */
101 DMU_OT_ZVOL_PROP, /* ZAP */
102 /* other; for testing only! */
103 DMU_OT_PLAIN_OTHER, /* UINT8 */
104 DMU_OT_UINT64_OTHER, /* UINT64 */
105 DMU_OT_ZAP_OTHER, /* ZAP */
106 /* new object types: */
107 DMU_OT_ERROR_LOG, /* ZAP */
108 DMU_OT_SPA_HISTORY, /* UINT8 */
109 DMU_OT_SPA_HISTORY_OFFSETS, /* spa_his_phys_t */
110 DMU_OT_POOL_PROPS, /* ZAP */
115 typedef enum dmu_objset_type {
120 DMU_OST_OTHER, /* For testing only! */
121 DMU_OST_ANY, /* Be careful! */
125 void byteswap_uint64_array(void *buf, size_t size);
126 void byteswap_uint32_array(void *buf, size_t size);
127 void byteswap_uint16_array(void *buf, size_t size);
128 void byteswap_uint8_array(void *buf, size_t size);
129 void zap_byteswap(void *buf, size_t size);
130 void zfs_acl_byteswap(void *buf, size_t size);
131 void zfs_znode_byteswap(void *buf, size_t size);
133 #define DS_MODE_NONE 0 /* invalid, to aid debugging */
134 #define DS_MODE_STANDARD 1 /* normal access, no special needs */
135 #define DS_MODE_PRIMARY 2 /* the "main" access, e.g. a mount */
136 #define DS_MODE_EXCLUSIVE 3 /* exclusive access, e.g. to destroy */
137 #define DS_MODE_LEVELS 4
138 #define DS_MODE_LEVEL(x) ((x) & (DS_MODE_LEVELS - 1))
139 #define DS_MODE_READONLY 0x8
140 #define DS_MODE_IS_READONLY(x) ((x) & DS_MODE_READONLY)
141 #define DS_MODE_INCONSISTENT 0x10
142 #define DS_MODE_IS_INCONSISTENT(x) ((x) & DS_MODE_INCONSISTENT)
144 #define DS_FIND_SNAPSHOTS (1<<0)
145 #define DS_FIND_CHILDREN (1<<1)
148 * The maximum number of bytes that can be accessed as part of one
149 * operation, including metadata.
151 #define DMU_MAX_ACCESS (10<<20) /* 10MB */
154 * Public routines to create, destroy, open, and close objsets.
156 int dmu_objset_open(const char *name, dmu_objset_type_t type, int mode,
158 void dmu_objset_close(objset_t *os);
159 int dmu_objset_evict_dbufs(objset_t *os, int try);
160 int dmu_objset_create(const char *name, dmu_objset_type_t type,
161 objset_t *clone_parent,
162 void (*func)(objset_t *os, void *arg, dmu_tx_t *tx), void *arg);
163 int dmu_objset_destroy(const char *name);
164 int dmu_snapshots_destroy(char *fsname, char *snapname);
165 int dmu_objset_rollback(const char *name);
166 int dmu_objset_snapshot(char *fsname, char *snapname, boolean_t recursive);
167 int dmu_objset_rename(const char *name, const char *newname,
168 boolean_t recursive);
169 int dmu_objset_find(char *name, int func(char *, void *), void *arg,
171 void dmu_objset_byteswap(void *buf, size_t size);
173 typedef struct dmu_buf {
174 uint64_t db_object; /* object that this buffer is part of */
175 uint64_t db_offset; /* byte offset in this object */
176 uint64_t db_size; /* size of buffer in bytes */
177 void *db_data; /* data in buffer */
180 typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr);
183 * Callback function to perform byte swapping on a block.
185 typedef void dmu_byteswap_func_t(void *buf, size_t size);
188 * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
190 #define DMU_POOL_DIRECTORY_OBJECT 1
191 #define DMU_POOL_CONFIG "config"
192 #define DMU_POOL_ROOT_DATASET "root_dataset"
193 #define DMU_POOL_SYNC_BPLIST "sync_bplist"
194 #define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
195 #define DMU_POOL_ERRLOG_LAST "errlog_last"
196 #define DMU_POOL_SPARES "spares"
197 #define DMU_POOL_DEFLATE "deflate"
198 #define DMU_POOL_HISTORY "history"
199 #define DMU_POOL_PROPS "pool_props"
202 * Allocate an object from this objset. The range of object numbers
203 * available is (0, DN_MAX_OBJECT). Object 0 is the meta-dnode.
205 * The transaction must be assigned to a txg. The newly allocated
206 * object will be "held" in the transaction (ie. you can modify the
207 * newly allocated object in this transaction).
209 * dmu_object_alloc() chooses an object and returns it in *objectp.
211 * dmu_object_claim() allocates a specific object number. If that
212 * number is already allocated, it fails and returns EEXIST.
214 * Return 0 on success, or ENOSPC or EEXIST as specified above.
216 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
217 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
218 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
219 int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
220 int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
221 int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
224 * Free an object from this objset.
226 * The object's data will be freed as well (ie. you don't need to call
227 * dmu_free(object, 0, -1, tx)).
229 * The object need not be held in the transaction.
231 * If there are any holds on this object's buffers (via dmu_buf_hold()),
232 * or tx holds on the object (via dmu_tx_hold_object()), you can not
233 * free it; it fails and returns EBUSY.
235 * If the object is not allocated, it fails and returns ENOENT.
237 * Return 0 on success, or EBUSY or ENOENT as specified above.
239 int dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx);
242 * Find the next allocated or free object.
244 * The objectp parameter is in-out. It will be updated to be the next
245 * object which is allocated. Ignore objects which have not been
246 * modified since txg.
248 * XXX Can only be called on a objset with no dirty data.
250 * Returns 0 on success, or ENOENT if there are no more objects.
252 int dmu_object_next(objset_t *os, uint64_t *objectp,
253 boolean_t hole, uint64_t txg);
256 * Set the data blocksize for an object.
258 * The object cannot have any blocks allcated beyond the first. If
259 * the first block is allocated already, the new size must be greater
260 * than the current block size. If these conditions are not met,
261 * ENOTSUP will be returned.
263 * Returns 0 on success, or EBUSY if there are any holds on the object
264 * contents, or ENOTSUP as described above.
266 int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size,
267 int ibs, dmu_tx_t *tx);
270 * Set the checksum property on a dnode. The new checksum algorithm will
271 * apply to all newly written blocks; existing blocks will not be affected.
273 void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
277 * Set the compress property on a dnode. The new compression algorithm will
278 * apply to all newly written blocks; existing blocks will not be affected.
280 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
284 * Decide how many copies of a given block we should make. Can be from
285 * 1 to SPA_DVAS_PER_BP.
287 int dmu_get_replication_level(struct objset_impl *, struct zbookmark *zb,
288 dmu_object_type_t ot);
290 * The bonus data is accessed more or less like a regular buffer.
291 * You must dmu_bonus_hold() to get the buffer, which will give you a
292 * dmu_buf_t with db_offset==-1ULL, and db_size = the size of the bonus
293 * data. As with any normal buffer, you must call dmu_buf_read() to
294 * read db_data, dmu_buf_will_dirty() before modifying it, and the
295 * object must be held in an assigned transaction before calling
296 * dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
297 * buffer as well. You must release your hold with dmu_buf_rele().
299 int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
300 int dmu_bonus_max(void);
303 * Obtain the DMU buffer from the specified object which contains the
304 * specified offset. dmu_buf_hold() puts a "hold" on the buffer, so
305 * that it will remain in memory. You must release the hold with
306 * dmu_buf_rele(). You musn't access the dmu_buf_t after releasing your
307 * hold. You must have a hold on any dmu_buf_t* you pass to the DMU.
309 * You must call dmu_buf_read, dmu_buf_will_dirty, or dmu_buf_will_fill
310 * on the returned buffer before reading or writing the buffer's
311 * db_data. The comments for those routines describe what particular
312 * operations are valid after calling them.
314 * The object number must be a valid, allocated object number.
316 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
317 void *tag, dmu_buf_t **);
318 void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
319 void dmu_buf_rele(dmu_buf_t *db, void *tag);
320 uint64_t dmu_buf_refcount(dmu_buf_t *db);
323 * dmu_buf_hold_array holds the DMU buffers which contain all bytes in a
324 * range of an object. A pointer to an array of dmu_buf_t*'s is
325 * returned (in *dbpp).
327 * dmu_buf_rele_array releases the hold on an array of dmu_buf_t*'s, and
328 * frees the array. The hold on the array of buffers MUST be released
329 * with dmu_buf_rele_array. You can NOT release the hold on each buffer
330 * individually with dmu_buf_rele.
332 int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
333 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
334 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
337 * Returns NULL on success, or the existing user ptr if it's already
340 * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
342 * user_data_ptr_ptr should be NULL, or a pointer to a pointer which
343 * will be set to db->db_data when you are allowed to access it. Note
344 * that db->db_data (the pointer) can change when you do dmu_buf_read(),
345 * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill().
346 * *user_data_ptr_ptr will be set to the new value when it changes.
348 * If non-NULL, pageout func will be called when this buffer is being
349 * excised from the cache, so that you can clean up the data structure
350 * pointed to by user_ptr.
352 * dmu_evict_user() will call the pageout func for all buffers in a
353 * objset with a given pageout func.
355 void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr,
356 dmu_buf_evict_func_t *pageout_func);
358 * set_user_ie is the same as set_user, but request immediate eviction
359 * when hold count goes to zero.
361 void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
362 void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func);
363 void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
364 void *user_ptr, void *user_data_ptr_ptr,
365 dmu_buf_evict_func_t *pageout_func);
366 void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
369 * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
371 void *dmu_buf_get_user(dmu_buf_t *db);
374 * Indicate that you are going to modify the buffer's data (db_data).
376 * The transaction (tx) must be assigned to a txg (ie. you've called
377 * dmu_tx_assign()). The buffer's object must be held in the tx
378 * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
380 void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
383 * You must create a transaction, then hold the objects which you will
384 * (or might) modify as part of this transaction. Then you must assign
385 * the transaction to a transaction group. Once the transaction has
386 * been assigned, you can modify buffers which belong to held objects as
387 * part of this transaction. You can't modify buffers before the
388 * transaction has been assigned; you can't modify buffers which don't
389 * belong to objects which this transaction holds; you can't hold
390 * objects once the transaction has been assigned. You may hold an
391 * object which you are going to free (with dmu_object_free()), but you
394 * You can abort the transaction before it has been assigned.
396 * Note that you may hold buffers (with dmu_buf_hold) at any time,
397 * regardless of transaction state.
400 #define DMU_NEW_OBJECT (-1ULL)
401 #define DMU_OBJECT_END (-1ULL)
403 dmu_tx_t *dmu_tx_create(objset_t *os);
404 void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len);
405 void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off,
407 void dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, char *name);
408 void dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object);
409 void dmu_tx_abort(dmu_tx_t *tx);
410 int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
411 void dmu_tx_wait(dmu_tx_t *tx);
412 void dmu_tx_commit(dmu_tx_t *tx);
415 * Free up the data blocks for a defined range of a file. If size is
416 * zero, the range from offset to end-of-file is freed.
418 int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
419 uint64_t size, dmu_tx_t *tx);
422 * Convenience functions.
424 * Canfail routines will return 0 on success, or an errno if there is a
425 * nonrecoverable I/O error.
427 int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
429 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
430 const void *buf, dmu_tx_t *tx);
431 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
432 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
434 int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
435 uint64_t size, struct page *pp, dmu_tx_t *tx);
437 extern int zfs_prefetch_disable;
440 * Asynchronously try to read in the data.
442 void dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset,
445 typedef struct dmu_object_info {
446 /* All sizes are in bytes. */
447 uint32_t doi_data_block_size;
448 uint32_t doi_metadata_block_size;
449 uint64_t doi_bonus_size;
450 dmu_object_type_t doi_type;
451 dmu_object_type_t doi_bonus_type;
452 uint8_t doi_indirection; /* 2 = dnode->indirect->data */
453 uint8_t doi_checksum;
454 uint8_t doi_compress;
456 /* Values below are number of 512-byte blocks. */
457 uint64_t doi_physical_blks; /* data + metadata */
458 uint64_t doi_max_block_offset;
461 typedef struct dmu_object_type_info {
462 dmu_byteswap_func_t *ot_byteswap;
463 boolean_t ot_metadata;
465 } dmu_object_type_info_t;
467 extern const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES];
470 * Get information on a DMU object.
472 * Return 0 on success or ENOENT if object is not allocated.
474 * If doi is NULL, just indicates whether the object exists.
476 int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
477 void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
478 void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
479 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
480 u_longlong_t *nblk512);
482 typedef struct dmu_objset_stats {
483 uint64_t dds_num_clones; /* number of clones of this */
484 uint64_t dds_creation_txg;
485 dmu_objset_type_t dds_type;
486 uint8_t dds_is_snapshot;
487 uint8_t dds_inconsistent;
488 char dds_clone_of[MAXNAMELEN];
489 } dmu_objset_stats_t;
492 * Get stats on a dataset.
494 void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
497 * Add entries to the nvlist for all the objset's properties. See
498 * zfs_prop_table[] and zfs(1m) for details on the properties.
500 void dmu_objset_stats(objset_t *os, struct nvlist *nv);
503 * Get the space usage statistics for statvfs().
505 * refdbytes is the amount of space "referenced" by this objset.
506 * availbytes is the amount of space available to this objset, taking
507 * into account quotas & reservations, assuming that no other objsets
508 * use the space first. These values correspond to the 'referenced' and
509 * 'available' properties, described in the zfs(1m) manpage.
511 * usedobjs and availobjs are the number of objects currently allocated,
514 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
515 uint64_t *usedobjsp, uint64_t *availobjsp);
518 * The fsid_guid is a 56-bit ID that can change to avoid collisions.
519 * (Contrast with the ds_guid which is a 64-bit ID that will never
520 * change, so there is a small probability that it will collide.)
522 uint64_t dmu_objset_fsid_guid(objset_t *os);
524 int dmu_objset_is_snapshot(objset_t *os);
526 extern struct spa *dmu_objset_spa(objset_t *os);
527 extern struct zilog *dmu_objset_zil(objset_t *os);
528 extern struct dsl_pool *dmu_objset_pool(objset_t *os);
529 extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
530 extern void dmu_objset_name(objset_t *os, char *buf);
531 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
532 extern uint64_t dmu_objset_id(objset_t *os);
533 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
534 uint64_t *id, uint64_t *offp);
535 extern int dmu_dir_list_next(objset_t *os, int namelen, char *name,
536 uint64_t *idp, uint64_t *offp);
539 * Return the txg number for the given assigned transaction.
541 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
545 * If a parent zio is provided this function initiates a write on the
546 * provided buffer as a child of the parent zio.
547 * In the absense of a parent zio, the write is completed synchronously.
548 * At write completion, blk is filled with the bp of the written block.
549 * Note that while the data covered by this function will be on stable
550 * storage when the write completes this new data does not become a
551 * permanent part of the file until the associated transaction commits.
553 typedef void dmu_sync_cb_t(dmu_buf_t *db, void *arg);
554 int dmu_sync(struct zio *zio, dmu_buf_t *db,
555 struct blkptr *bp, uint64_t txg, dmu_sync_cb_t *done, void *arg);
558 * Find the next hole or data block in file starting at *off
559 * Return found offset in *off. Return ESRCH for end of file.
561 int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole,
565 * Initial setup and final teardown.
567 extern void dmu_init(void);
568 extern void dmu_fini(void);
570 typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
571 uint64_t object, uint64_t offset, int len);
572 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
573 dmu_traverse_cb_t cb, void *arg);
575 int dmu_sendbackup(objset_t *tosnap, objset_t *fromsnap, struct file *fp);
576 int dmu_recvbackup(char *tosnap, struct drr_begin *drrb, uint64_t *sizep,
577 boolean_t force, struct file *fp, uint64_t voffset);
580 #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
581 extern uint64_t zfs_crc64_table[256];
587 #endif /* _SYS_DMU_H */