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 https://opensource.org/licenses/CDDL-1.0.
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) 2012, 2020 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
36 #include <sys/zfs_context.h>
37 #include <sys/zfs_refcount.h>
38 #include <sys/zrlock.h>
39 #include <sys/multilist.h>
48 * define flags for dbuf_read
51 #define DB_RF_MUST_SUCCEED (1 << 0)
52 #define DB_RF_CANFAIL (1 << 1)
53 #define DB_RF_HAVESTRUCT (1 << 2)
54 #define DB_RF_NOPREFETCH (1 << 3)
55 #define DB_RF_NEVERWAIT (1 << 4)
56 #define DB_RF_CACHED (1 << 5)
57 #define DB_RF_NO_DECRYPT (1 << 6)
58 #define DB_RF_PARTIAL_FIRST (1 << 7)
59 #define DB_RF_PARTIAL_MORE (1 << 8)
62 * The simplified state transition diagram for dbufs looks like:
67 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
73 * | +------> NOFILL -----+
77 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
78 * to find all dbufs in a range of a dnode and must be less than any other
79 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
81 typedef enum dbuf_states
{
92 typedef enum dbuf_cached_state
{
95 DB_DBUF_METADATA_CACHE
,
97 } dbuf_cached_state_t
;
103 * level = 0 means the user data
104 * level = 1 means the single indirect block
110 typedef enum override_states
{
116 typedef enum db_lock_type
{
122 typedef struct dbuf_dirty_record
{
123 /* link on our parents dirty list */
124 list_node_t dr_dirty_node
;
126 /* transaction group this data will sync in */
129 /* zio of outstanding write IO */
132 /* pointer back to our dbuf */
133 struct dmu_buf_impl
*dr_dbuf
;
135 /* list link for dbuf dirty records */
136 list_node_t dr_dbuf_node
;
139 * The dnode we are part of. Note that the dnode can not be moved or
140 * evicted due to the hold that's added by dnode_setdirty() or
141 * dmu_objset_sync_dnodes(), and released by dnode_rele_task() or
142 * userquota_updates_task(). This hold is necessary for
143 * dirty_lightweight_leaf-type dirty records, which don't have a hold
148 /* pointer to parent dirty record */
149 struct dbuf_dirty_record
*dr_parent
;
151 /* How much space was changed to dsl_pool_dirty_space() for this? */
152 unsigned int dr_accounted
;
154 /* A copy of the bp that points to us */
158 struct dirty_indirect
{
160 /* protect access to list */
163 /* Our list of dirty children */
169 * dr_data is set when we dirty the buffer
170 * so that we can retain the pointer even if it
171 * gets COW'd in a subsequent transaction group.
174 blkptr_t dr_overridden_by
;
175 override_states_t dr_override_state
;
177 boolean_t dr_nopwrite
;
178 boolean_t dr_brtwrite
;
179 boolean_t dr_has_raw_params
;
182 * If dr_has_raw_params is set, the following crypt
183 * params will be set on the BP that's written.
185 boolean_t dr_byteorder
;
186 uint8_t dr_salt
[ZIO_DATA_SALT_LEN
];
187 uint8_t dr_iv
[ZIO_DATA_IV_LEN
];
188 uint8_t dr_mac
[ZIO_DATA_MAC_LEN
];
190 struct dirty_lightweight_leaf
{
192 * This dirty record refers to a leaf (level=0)
193 * block, whose dbuf has not been instantiated for
194 * performance reasons.
202 } dbuf_dirty_record_t
;
204 typedef struct dmu_buf_impl
{
206 * The following members are immutable, with the exception of
207 * db.db_data, which is protected by db_mtx.
210 /* the publicly visible structure */
213 /* the objset we belong to */
214 struct objset
*db_objset
;
217 * handle to safely access the dnode we belong to (NULL when evicted)
219 struct dnode_handle
*db_dnode_handle
;
222 * our parent buffer; if the dnode points to us directly,
223 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
224 * only accessed by sync thread ???
225 * (NULL when evicted)
226 * May change from NULL to non-NULL under the protection of db_mtx
227 * (see dbuf_check_blkptr())
229 struct dmu_buf_impl
*db_parent
;
232 * link for hash table of all dmu_buf_impl_t's
234 struct dmu_buf_impl
*db_hash_next
;
237 * Our link on the owner dnodes's dn_dbufs list.
238 * Protected by its dn_dbufs_mtx. Should be on the same cache line
239 * as db_level and db_blkid for the best avl_add() performance.
243 /* our block number */
247 * Pointer to the blkptr_t which points to us. May be NULL if we
248 * don't have one yet. (NULL when evicted)
253 * Our indirection level. Data buffers have db_level==0.
254 * Indirect buffers which point to data buffers have
255 * db_level==1. etc. Buffers which contain dnodes have
256 * db_level==0, since the dnodes are stored in a file.
261 * Protects db_buf's contents if they contain an indirect block or data
262 * block of the meta-dnode. We use this lock to protect the structure of
263 * the block tree. This means that when modifying this dbuf's data, we
264 * grab its rwlock. When modifying its parent's data (including the
265 * blkptr to this dbuf), we grab the parent's rwlock. The lock ordering
267 * 1) dn_struct_rwlock
269 * We don't currently grab multiple dbufs' db_rwlocks at once.
273 /* buffer holding our data */
276 /* db_mtx protects the members below */
280 * Current state of the buffer
282 dbuf_states_t db_state
;
285 * Refcount accessed by dmu_buf_{hold,rele}.
286 * If nonzero, the buffer can't be destroyed.
287 * Protected by db_mtx.
289 zfs_refcount_t db_holds
;
291 kcondvar_t db_changed
;
292 dbuf_dirty_record_t
*db_data_pending
;
294 /* List of dirty records for the buffer sorted newest to oldest. */
295 list_t db_dirty_records
;
297 /* Link in dbuf_cache or dbuf_metadata_cache */
298 multilist_node_t db_cache_link
;
300 /* Tells us which dbuf cache this dbuf is in, if any */
301 dbuf_cached_state_t db_caching_status
;
305 /* Data which is unique to data (leaf) blocks: */
307 /* User callback information. */
308 dmu_buf_user_t
*db_user
;
311 * Evict user data as soon as the dirty and reference
314 uint8_t db_user_immediate_evict
;
317 * This block was freed while a read or write was
320 uint8_t db_freed_in_flight
;
323 * dnode_evict_dbufs() or dnode_evict_bonus() tried to
324 * evict this dbuf, but couldn't due to outstanding
325 * references. Evict once the refcount drops to 0.
327 uint8_t db_pending_evict
;
331 /* The buffer was partially read. More reads may follow. */
332 uint8_t db_partial_read
;
335 #define DBUF_HASH_MUTEX(h, idx) \
336 (&(h)->hash_mutexes[(idx) & ((h)->hash_mutex_mask)])
338 typedef struct dbuf_hash_table
{
339 uint64_t hash_table_mask
;
340 uint64_t hash_mutex_mask
;
341 dmu_buf_impl_t
**hash_table
;
342 kmutex_t
*hash_mutexes
;
345 typedef void (*dbuf_prefetch_fn
)(void *, uint64_t, uint64_t, boolean_t
);
347 uint64_t dbuf_whichblock(const struct dnode
*di
, const int64_t level
,
348 const uint64_t offset
);
350 void dbuf_create_bonus(struct dnode
*dn
);
351 int dbuf_spill_set_blksz(dmu_buf_t
*db
, uint64_t blksz
, dmu_tx_t
*tx
);
353 void dbuf_rm_spill(struct dnode
*dn
, dmu_tx_t
*tx
);
355 dmu_buf_impl_t
*dbuf_hold(struct dnode
*dn
, uint64_t blkid
, const void *tag
);
356 dmu_buf_impl_t
*dbuf_hold_level(struct dnode
*dn
, int level
, uint64_t blkid
,
358 int dbuf_hold_impl(struct dnode
*dn
, uint8_t level
, uint64_t blkid
,
359 boolean_t fail_sparse
, boolean_t fail_uncached
,
360 const void *tag
, dmu_buf_impl_t
**dbp
);
362 int dbuf_prefetch_impl(struct dnode
*dn
, int64_t level
, uint64_t blkid
,
363 zio_priority_t prio
, arc_flags_t aflags
, dbuf_prefetch_fn cb
,
365 int dbuf_prefetch(struct dnode
*dn
, int64_t level
, uint64_t blkid
,
366 zio_priority_t prio
, arc_flags_t aflags
);
368 void dbuf_add_ref(dmu_buf_impl_t
*db
, const void *tag
);
369 boolean_t
dbuf_try_add_ref(dmu_buf_t
*db
, objset_t
*os
, uint64_t obj
,
370 uint64_t blkid
, const void *tag
);
371 uint64_t dbuf_refcount(dmu_buf_impl_t
*db
);
373 void dbuf_rele(dmu_buf_impl_t
*db
, const void *tag
);
374 void dbuf_rele_and_unlock(dmu_buf_impl_t
*db
, const void *tag
,
377 dmu_buf_impl_t
*dbuf_find(struct objset
*os
, uint64_t object
, uint8_t level
,
378 uint64_t blkid
, uint64_t *hash_out
);
380 int dbuf_read(dmu_buf_impl_t
*db
, zio_t
*zio
, uint32_t flags
);
381 void dmu_buf_will_clone(dmu_buf_t
*db
, dmu_tx_t
*tx
);
382 void dmu_buf_will_not_fill(dmu_buf_t
*db
, dmu_tx_t
*tx
);
383 void dmu_buf_will_fill(dmu_buf_t
*db
, dmu_tx_t
*tx
, boolean_t canfail
);
384 boolean_t
dmu_buf_fill_done(dmu_buf_t
*db
, dmu_tx_t
*tx
, boolean_t failed
);
385 void dbuf_assign_arcbuf(dmu_buf_impl_t
*db
, arc_buf_t
*buf
, dmu_tx_t
*tx
);
386 dbuf_dirty_record_t
*dbuf_dirty(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
);
387 dbuf_dirty_record_t
*dbuf_dirty_lightweight(dnode_t
*dn
, uint64_t blkid
,
389 boolean_t
dbuf_undirty(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
);
390 arc_buf_t
*dbuf_loan_arcbuf(dmu_buf_impl_t
*db
);
391 void dmu_buf_write_embedded(dmu_buf_t
*dbuf
, void *data
,
392 bp_embedded_type_t etype
, enum zio_compress comp
,
393 int uncompressed_size
, int compressed_size
, int byteorder
, dmu_tx_t
*tx
);
395 int dmu_lightweight_write_by_dnode(dnode_t
*dn
, uint64_t offset
, abd_t
*abd
,
396 const struct zio_prop
*zp
, zio_flag_t flags
, dmu_tx_t
*tx
);
398 void dmu_buf_redact(dmu_buf_t
*dbuf
, dmu_tx_t
*tx
);
399 void dbuf_destroy(dmu_buf_impl_t
*db
);
401 void dbuf_unoverride(dbuf_dirty_record_t
*dr
);
402 void dbuf_sync_list(list_t
*list
, int level
, dmu_tx_t
*tx
);
403 void dbuf_release_bp(dmu_buf_impl_t
*db
);
404 db_lock_type_t
dmu_buf_lock_parent(dmu_buf_impl_t
*db
, krw_t rw
,
406 void dmu_buf_unlock_parent(dmu_buf_impl_t
*db
, db_lock_type_t type
,
409 void dbuf_free_range(struct dnode
*dn
, uint64_t start
, uint64_t end
,
412 void dbuf_new_size(dmu_buf_impl_t
*db
, int size
, dmu_tx_t
*tx
);
414 void dbuf_stats_init(dbuf_hash_table_t
*hash
);
415 void dbuf_stats_destroy(void);
417 int dbuf_dnode_findbp(dnode_t
*dn
, uint64_t level
, uint64_t blkid
,
418 blkptr_t
*bp
, uint16_t *datablkszsec
, uint8_t *indblkshift
);
420 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
421 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
422 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
423 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
424 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
426 void dbuf_init(void);
427 void dbuf_fini(void);
429 boolean_t
dbuf_is_metadata(dmu_buf_impl_t
*db
);
431 static inline dbuf_dirty_record_t
*
432 dbuf_find_dirty_lte(dmu_buf_impl_t
*db
, uint64_t txg
)
434 dbuf_dirty_record_t
*dr
;
436 for (dr
= list_head(&db
->db_dirty_records
);
437 dr
!= NULL
&& dr
->dr_txg
> txg
;
438 dr
= list_next(&db
->db_dirty_records
, dr
))
443 static inline dbuf_dirty_record_t
*
444 dbuf_find_dirty_eq(dmu_buf_impl_t
*db
, uint64_t txg
)
446 dbuf_dirty_record_t
*dr
;
448 dr
= dbuf_find_dirty_lte(db
, txg
);
449 if (dr
&& dr
->dr_txg
== txg
)
454 #define DBUF_GET_BUFC_TYPE(_db) \
455 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
457 #define DBUF_IS_CACHEABLE(_db) \
458 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
459 (dbuf_is_metadata(_db) && \
460 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
462 boolean_t
dbuf_is_l2cacheable(dmu_buf_impl_t
*db
);
467 * There should be a ## between the string literal and fmt, to make it
468 * clear that we're joining two strings together, but gcc does not
469 * support that preprocessor token.
471 #define dprintf_dbuf(dbuf, fmt, ...) do { \
472 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
474 uint64_t __db_obj = (dbuf)->db.db_object; \
475 if (__db_obj == DMU_META_DNODE_OBJECT) \
476 (void) strlcpy(__db_buf, "mdn", sizeof (__db_buf)); \
478 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
479 (u_longlong_t)__db_obj); \
480 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
481 "obj=%s lvl=%u blkid=%lld " fmt, \
482 __db_buf, (dbuf)->db_level, \
483 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
487 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
488 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
489 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
490 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
491 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
492 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
496 #define DBUF_VERIFY(db) dbuf_verify(db)
500 #define dprintf_dbuf(db, fmt, ...)
501 #define dprintf_dbuf_bp(db, bp, fmt, ...)
502 #define DBUF_VERIFY(db)
511 #endif /* _SYS_DBUF_H */