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) 2012, 2018 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)
60 * The simplified state transition diagram for dbufs looks like:
65 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
71 * +--------> NOFILL -------+
73 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
74 * to find all dbufs in a range of a dnode and must be less than any other
75 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
77 typedef enum dbuf_states {
87 typedef enum dbuf_cached_state {
90 DB_DBUF_METADATA_CACHE,
92 } dbuf_cached_state_t;
98 * level = 0 means the user data
99 * level = 1 means the single indirect block
105 typedef enum override_states {
111 typedef enum db_lock_type {
117 typedef struct dbuf_dirty_record {
118 /* link on our parents dirty list */
119 list_node_t dr_dirty_node;
121 /* transaction group this data will sync in */
124 /* zio of outstanding write IO */
127 /* pointer back to our dbuf */
128 struct dmu_buf_impl *dr_dbuf;
130 /* list link for dbuf dirty records */
131 list_node_t dr_dbuf_node;
133 /* pointer to parent dirty record */
134 struct dbuf_dirty_record *dr_parent;
136 /* How much space was changed to dsl_pool_dirty_space() for this? */
137 unsigned int dr_accounted;
139 /* A copy of the bp that points to us */
143 struct dirty_indirect {
145 /* protect access to list */
148 /* Our list of dirty children */
154 * dr_data is set when we dirty the buffer
155 * so that we can retain the pointer even if it
156 * gets COW'd in a subsequent transaction group.
159 blkptr_t dr_overridden_by;
160 override_states_t dr_override_state;
162 boolean_t dr_nopwrite;
163 boolean_t dr_has_raw_params;
166 * If dr_has_raw_params is set, the following crypt
167 * params will be set on the BP that's written.
169 boolean_t dr_byteorder;
170 uint8_t dr_salt[ZIO_DATA_SALT_LEN];
171 uint8_t dr_iv[ZIO_DATA_IV_LEN];
172 uint8_t dr_mac[ZIO_DATA_MAC_LEN];
175 } dbuf_dirty_record_t;
177 typedef struct dmu_buf_impl {
179 * The following members are immutable, with the exception of
180 * db.db_data, which is protected by db_mtx.
183 /* the publicly visible structure */
186 /* the objset we belong to */
187 struct objset *db_objset;
190 * handle to safely access the dnode we belong to (NULL when evicted)
192 struct dnode_handle *db_dnode_handle;
195 * our parent buffer; if the dnode points to us directly,
196 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
197 * only accessed by sync thread ???
198 * (NULL when evicted)
199 * May change from NULL to non-NULL under the protection of db_mtx
200 * (see dbuf_check_blkptr())
202 struct dmu_buf_impl *db_parent;
205 * link for hash table of all dmu_buf_impl_t's
207 struct dmu_buf_impl *db_hash_next;
210 * Our link on the owner dnodes's dn_dbufs list.
211 * Protected by its dn_dbufs_mtx. Should be on the same cache line
212 * as db_level and db_blkid for the best avl_add() performance.
216 /* our block number */
220 * Pointer to the blkptr_t which points to us. May be NULL if we
221 * don't have one yet. (NULL when evicted)
226 * Our indirection level. Data buffers have db_level==0.
227 * Indirect buffers which point to data buffers have
228 * db_level==1. etc. Buffers which contain dnodes have
229 * db_level==0, since the dnodes are stored in a file.
234 * Protects db_buf's contents if they contain an indirect block or data
235 * block of the meta-dnode. We use this lock to protect the structure of
236 * the block tree. This means that when modifying this dbuf's data, we
237 * grab its rwlock. When modifying its parent's data (including the
238 * blkptr to this dbuf), we grab the parent's rwlock. The lock ordering
240 * 1) dn_struct_rwlock
242 * We don't currently grab multiple dbufs' db_rwlocks at once.
246 /* buffer holding our data */
249 /* db_mtx protects the members below */
253 * Current state of the buffer
255 dbuf_states_t db_state;
258 * Refcount accessed by dmu_buf_{hold,rele}.
259 * If nonzero, the buffer can't be destroyed.
260 * Protected by db_mtx.
262 zfs_refcount_t db_holds;
264 kcondvar_t db_changed;
265 dbuf_dirty_record_t *db_data_pending;
267 /* List of dirty records for the buffer sorted newest to oldest. */
268 list_t db_dirty_records;
270 /* Link in dbuf_cache or dbuf_metadata_cache */
271 multilist_node_t db_cache_link;
273 /* Tells us which dbuf cache this dbuf is in, if any */
274 dbuf_cached_state_t db_caching_status;
276 /* Data which is unique to data (leaf) blocks: */
278 /* User callback information. */
279 dmu_buf_user_t *db_user;
282 * Evict user data as soon as the dirty and reference
285 uint8_t db_user_immediate_evict;
288 * This block was freed while a read or write was
291 uint8_t db_freed_in_flight;
294 * dnode_evict_dbufs() or dnode_evict_bonus() tried to
295 * evict this dbuf, but couldn't due to outstanding
296 * references. Evict once the refcount drops to 0.
298 uint8_t db_pending_evict;
303 /* Note: the dbuf hash table is exposed only for the mdb module */
304 #define DBUF_MUTEXES 8192
305 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
306 typedef struct dbuf_hash_table {
307 uint64_t hash_table_mask;
308 dmu_buf_impl_t **hash_table;
309 kmutex_t hash_mutexes[DBUF_MUTEXES];
312 uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level,
313 const uint64_t offset);
315 void dbuf_create_bonus(struct dnode *dn);
316 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
318 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
320 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
321 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
323 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid,
324 boolean_t fail_sparse, boolean_t fail_uncached,
325 void *tag, dmu_buf_impl_t **dbp);
327 void dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
328 zio_priority_t prio, arc_flags_t aflags);
330 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
331 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj,
332 uint64_t blkid, void *tag);
333 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
335 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
336 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag, boolean_t evicting);
338 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
341 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
342 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
343 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
344 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
345 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
346 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
347 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
348 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
349 bp_embedded_type_t etype, enum zio_compress comp,
350 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
352 void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx);
353 void dbuf_destroy(dmu_buf_impl_t *db);
355 void dbuf_unoverride(dbuf_dirty_record_t *dr);
356 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx);
357 void dbuf_release_bp(dmu_buf_impl_t *db);
358 db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw, void *tag);
359 void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type, void *tag);
361 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
364 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
366 void dbuf_stats_init(dbuf_hash_table_t *hash);
367 void dbuf_stats_destroy(void);
369 int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid,
370 blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift);
372 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
373 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
374 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
375 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
376 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
378 void dbuf_init(void);
379 void dbuf_fini(void);
381 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
383 static inline dbuf_dirty_record_t *
384 dbuf_find_dirty_lte(dmu_buf_impl_t *db, uint64_t txg)
386 dbuf_dirty_record_t *dr;
388 for (dr = list_head(&db->db_dirty_records);
389 dr != NULL && dr->dr_txg > txg;
390 dr = list_next(&db->db_dirty_records, dr))
395 static inline dbuf_dirty_record_t *
396 dbuf_find_dirty_eq(dmu_buf_impl_t *db, uint64_t txg)
398 dbuf_dirty_record_t *dr;
400 dr = dbuf_find_dirty_lte(db, txg);
401 if (dr && dr->dr_txg == txg)
406 #define DBUF_GET_BUFC_TYPE(_db) \
407 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
409 #define DBUF_IS_CACHEABLE(_db) \
410 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
411 (dbuf_is_metadata(_db) && \
412 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
414 #define DBUF_IS_L2CACHEABLE(_db) \
415 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \
416 (dbuf_is_metadata(_db) && \
417 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
419 #define DNODE_LEVEL_IS_L2CACHEABLE(_dn, _level) \
420 ((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_ALL || \
422 DMU_OT_IS_METADATA((_dn)->dn_handle->dnh_dnode->dn_type)) && \
423 ((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
428 * There should be a ## between the string literal and fmt, to make it
429 * clear that we're joining two strings together, but gcc does not
430 * support that preprocessor token.
432 #define dprintf_dbuf(dbuf, fmt, ...) do { \
433 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
435 uint64_t __db_obj = (dbuf)->db.db_object; \
436 if (__db_obj == DMU_META_DNODE_OBJECT) \
437 (void) strcpy(__db_buf, "mdn"); \
439 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
440 (u_longlong_t)__db_obj); \
441 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
442 "obj=%s lvl=%u blkid=%lld " fmt, \
443 __db_buf, (dbuf)->db_level, \
444 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
446 _NOTE(CONSTCOND) } while (0)
448 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
449 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
450 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
451 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
452 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
453 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
455 _NOTE(CONSTCOND) } while (0)
457 #define DBUF_VERIFY(db) dbuf_verify(db)
461 #define dprintf_dbuf(db, fmt, ...)
462 #define dprintf_dbuf_bp(db, bp, fmt, ...)
463 #define DBUF_VERIFY(db)
472 #endif /* _SYS_DBUF_H */