1 /*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21 /*
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.
26 */
27
28 #ifndef _SYS_DBUF_H
29 #define _SYS_DBUF_H
30
31 #include <sys/dmu.h>
32 #include <sys/spa.h>
33 #include <sys/txg.h>
34 #include <sys/zio.h>
35 #include <sys/arc.h>
36 #include <sys/zfs_context.h>
37 #include <sys/zfs_refcount.h>
38 #include <sys/zrlock.h>
39 #include <sys/multilist.h>
40
41 #ifdef __cplusplus
42 extern "C" {
43 #endif
44
45 #define IN_DMU_SYNC 2
46
47 /*
48 * define flags for dbuf_read
49 */
50
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)
60
61 /*
62 * The simplified state transition diagram for dbufs looks like:
63 *
64 * +--> READ --+
65 * | |
66 * | V
67 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
68 * ^ | ^ ^
69 * | | | |
70 * | +--> FILL --+ |
71 * | | |
72 * | | |
73 * | +------> NOFILL -----+
74 * | |
75 * +---------------+
76 *
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).
80 */
81 typedef enum dbuf_states {
82 DB_SEARCH = -1,
83 DB_UNCACHED,
84 DB_FILL,
85 DB_NOFILL,
86 DB_READ,
87 DB_CACHED,
88 DB_EVICTING
89 } dbuf_states_t;
90
91 typedef enum dbuf_cached_state {
92 DB_NO_CACHE = -1,
93 DB_DBUF_CACHE,
94 DB_DBUF_METADATA_CACHE,
95 DB_CACHE_MAX
96 } dbuf_cached_state_t;
97
98 struct dnode;
99 struct dmu_tx;
100
101 /*
102 * level = 0 means the user data
103 * level = 1 means the single indirect block
104 * etc.
105 */
106
107 struct dmu_buf_impl;
108
109 typedef enum override_states {
110 DR_NOT_OVERRIDDEN,
111 DR_IN_DMU_SYNC,
112 DR_OVERRIDDEN
113 } override_states_t;
114
115 typedef enum db_lock_type {
116 DLT_NONE,
117 DLT_PARENT,
118 DLT_OBJSET
119 } db_lock_type_t;
120
121 typedef struct dbuf_dirty_record {
122 /* link on our parents dirty list */
123 list_node_t dr_dirty_node;
124
125 /* transaction group this data will sync in */
126 uint64_t dr_txg;
127
128 /* zio of outstanding write IO */
129 zio_t *dr_zio;
130
131 /* pointer back to our dbuf */
132 struct dmu_buf_impl *dr_dbuf;
133
134 /* list link for dbuf dirty records */
135 list_node_t dr_dbuf_node;
136
137 /*
138 * The dnode we are part of. Note that the dnode can not be moved or
139 * evicted due to the hold that's added by dnode_setdirty() or
140 * dmu_objset_sync_dnodes(), and released by dnode_rele_task() or
141 * userquota_updates_task(). This hold is necessary for
142 * dirty_lightweight_leaf-type dirty records, which don't have a hold
143 * on a dbuf.
144 */
145 dnode_t *dr_dnode;
146
147 /* pointer to parent dirty record */
148 struct dbuf_dirty_record *dr_parent;
149
150 /* How much space was changed to dsl_pool_dirty_space() for this? */
151 unsigned int dr_accounted;
152
153 /* A copy of the bp that points to us */
154 blkptr_t dr_bp_copy;
155
156 union dirty_types {
157 struct dirty_indirect {
158
159 /* protect access to list */
160 kmutex_t dr_mtx;
161
162 /* Our list of dirty children */
163 list_t dr_children;
164 } di;
165 struct dirty_leaf {
166
167 /*
168 * dr_data is set when we dirty the buffer
169 * so that we can retain the pointer even if it
170 * gets COW'd in a subsequent transaction group.
171 */
172 arc_buf_t *dr_data;
173 blkptr_t dr_overridden_by;
174 override_states_t dr_override_state;
175 uint8_t dr_copies;
176 boolean_t dr_nopwrite;
177 boolean_t dr_brtwrite;
178 boolean_t dr_has_raw_params;
179
180 /*
181 * If dr_has_raw_params is set, the following crypt
182 * params will be set on the BP that's written.
183 */
184 boolean_t dr_byteorder;
185 uint8_t dr_salt[ZIO_DATA_SALT_LEN];
186 uint8_t dr_iv[ZIO_DATA_IV_LEN];
187 uint8_t dr_mac[ZIO_DATA_MAC_LEN];
188 } dl;
189 struct dirty_lightweight_leaf {
190 /*
191 * This dirty record refers to a leaf (level=0)
192 * block, whose dbuf has not been instantiated for
193 * performance reasons.
194 */
195 uint64_t dr_blkid;
196 abd_t *dr_abd;
197 zio_prop_t dr_props;
198 zio_flag_t dr_flags;
199 } dll;
200 } dt;
201 } dbuf_dirty_record_t;
202
203 typedef struct dmu_buf_impl {
204 /*
205 * The following members are immutable, with the exception of
206 * db.db_data, which is protected by db_mtx.
207 */
208
209 /* the publicly visible structure */
210 dmu_buf_t db;
211
212 /* the objset we belong to */
213 struct objset *db_objset;
214
215 /*
216 * handle to safely access the dnode we belong to (NULL when evicted)
217 */
218 struct dnode_handle *db_dnode_handle;
219
220 /*
221 * our parent buffer; if the dnode points to us directly,
222 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
223 * only accessed by sync thread ???
224 * (NULL when evicted)
225 * May change from NULL to non-NULL under the protection of db_mtx
226 * (see dbuf_check_blkptr())
227 */
228 struct dmu_buf_impl *db_parent;
229
230 /*
231 * link for hash table of all dmu_buf_impl_t's
232 */
233 struct dmu_buf_impl *db_hash_next;
234
235 /*
236 * Our link on the owner dnodes's dn_dbufs list.
237 * Protected by its dn_dbufs_mtx. Should be on the same cache line
238 * as db_level and db_blkid for the best avl_add() performance.
239 */
240 avl_node_t db_link;
241
242 /* our block number */
243 uint64_t db_blkid;
244
245 /*
246 * Pointer to the blkptr_t which points to us. May be NULL if we
247 * don't have one yet. (NULL when evicted)
248 */
249 blkptr_t *db_blkptr;
250
251 /*
252 * Our indirection level. Data buffers have db_level==0.
253 * Indirect buffers which point to data buffers have
254 * db_level==1. etc. Buffers which contain dnodes have
255 * db_level==0, since the dnodes are stored in a file.
256 */
257 uint8_t db_level;
258
259 /* This block was freed while a read or write was active. */
260 uint8_t db_freed_in_flight;
261
262 /*
263 * Evict user data as soon as the dirty and reference counts are equal.
264 */
265 uint8_t db_user_immediate_evict;
266
267 /*
268 * dnode_evict_dbufs() or dnode_evict_bonus() tried to evict this dbuf,
269 * but couldn't due to outstanding references. Evict once the refcount
270 * drops to 0.
271 */
272 uint8_t db_pending_evict;
273
274 /* Number of TXGs in which this buffer is dirty. */
275 uint8_t db_dirtycnt;
276
277 /* The buffer was partially read. More reads may follow. */
278 uint8_t db_partial_read;
279
280 /*
281 * Protects db_buf's contents if they contain an indirect block or data
282 * block of the meta-dnode. We use this lock to protect the structure of
283 * the block tree. This means that when modifying this dbuf's data, we
284 * grab its rwlock. When modifying its parent's data (including the
285 * blkptr to this dbuf), we grab the parent's rwlock. The lock ordering
286 * for this lock is:
287 * 1) dn_struct_rwlock
288 * 2) db_rwlock
289 * We don't currently grab multiple dbufs' db_rwlocks at once.
290 */
291 krwlock_t db_rwlock;
292
293 /* buffer holding our data */
294 arc_buf_t *db_buf;
295
296 /* db_mtx protects the members below */
297 kmutex_t db_mtx;
298
299 /*
300 * Current state of the buffer
301 */
302 dbuf_states_t db_state;
303
304 /* In which dbuf cache this dbuf is, if any. */
305 dbuf_cached_state_t db_caching_status;
306
307 /*
308 * Refcount accessed by dmu_buf_{hold,rele}.
309 * If nonzero, the buffer can't be destroyed.
310 * Protected by db_mtx.
311 */
312 zfs_refcount_t db_holds;
313
314 kcondvar_t db_changed;
315 dbuf_dirty_record_t *db_data_pending;
316
317 /* List of dirty records for the buffer sorted newest to oldest. */
318 list_t db_dirty_records;
319
320 /* Link in dbuf_cache or dbuf_metadata_cache */
321 multilist_node_t db_cache_link;
322
323 uint64_t db_hash;
324
325 /* User callback information. */
326 dmu_buf_user_t *db_user;
327 } dmu_buf_impl_t;
328
329 #define DBUF_HASH_MUTEX(h, idx) \
330 (&(h)->hash_mutexes[(idx) & ((h)->hash_mutex_mask)])
331
332 typedef struct dbuf_hash_table {
333 uint64_t hash_table_mask;
334 uint64_t hash_mutex_mask;
335 dmu_buf_impl_t **hash_table;
336 kmutex_t *hash_mutexes;
337 } dbuf_hash_table_t;
338
339 typedef void (*dbuf_prefetch_fn)(void *, uint64_t, uint64_t, boolean_t);
340
341 uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level,
342 const uint64_t offset);
343
344 void dbuf_create_bonus(struct dnode *dn);
345 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
346
347 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
348
349 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, const void *tag);
350 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
351 const void *tag);
352 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid,
353 boolean_t fail_sparse, boolean_t fail_uncached,
354 const void *tag, dmu_buf_impl_t **dbp);
355
356 int dbuf_prefetch_impl(struct dnode *dn, int64_t level, uint64_t blkid,
357 zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb,
358 void *arg);
359 int dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
360 zio_priority_t prio, arc_flags_t aflags);
361
362 void dbuf_add_ref(dmu_buf_impl_t *db, const void *tag);
363 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj,
364 uint64_t blkid, const void *tag);
365 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
366
367 void dbuf_rele(dmu_buf_impl_t *db, const void *tag);
368 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, const void *tag,
369 boolean_t evicting);
370
371 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
372 uint64_t blkid, uint64_t *hash_out);
373
374 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
375 void dmu_buf_will_clone(dmu_buf_t *db, dmu_tx_t *tx);
376 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
377 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx, boolean_t canfail);
378 boolean_t dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx, boolean_t failed);
379 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
380 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
381 dbuf_dirty_record_t *dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid,
382 dmu_tx_t *tx);
383 boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
384 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
385 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
386 bp_embedded_type_t etype, enum zio_compress comp,
387 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
388
389 int dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd,
390 const struct zio_prop *zp, zio_flag_t flags, dmu_tx_t *tx);
391
392 void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx);
393 void dbuf_destroy(dmu_buf_impl_t *db);
394
395 void dbuf_unoverride(dbuf_dirty_record_t *dr);
396 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx);
397 void dbuf_release_bp(dmu_buf_impl_t *db);
398 db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw,
399 const void *tag);
400 void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type,
401 const void *tag);
402
403 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
404 struct dmu_tx *);
405
406 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
407
408 void dbuf_stats_init(dbuf_hash_table_t *hash);
409 void dbuf_stats_destroy(void);
410
411 int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid,
412 blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift);
413
414 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
415 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
416 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
417 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
418 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
419
420 void dbuf_init(void);
421 void dbuf_fini(void);
422
423 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
424
425 static inline dbuf_dirty_record_t *
dbuf_find_dirty_lte(dmu_buf_impl_t * db,uint64_t txg)426 dbuf_find_dirty_lte(dmu_buf_impl_t *db, uint64_t txg)
427 {
428 dbuf_dirty_record_t *dr;
429
430 for (dr = list_head(&db->db_dirty_records);
431 dr != NULL && dr->dr_txg > txg;
432 dr = list_next(&db->db_dirty_records, dr))
433 continue;
434 return (dr);
435 }
436
437 static inline dbuf_dirty_record_t *
dbuf_find_dirty_eq(dmu_buf_impl_t * db,uint64_t txg)438 dbuf_find_dirty_eq(dmu_buf_impl_t *db, uint64_t txg)
439 {
440 dbuf_dirty_record_t *dr;
441
442 dr = dbuf_find_dirty_lte(db, txg);
443 if (dr && dr->dr_txg == txg)
444 return (dr);
445 return (NULL);
446 }
447
448 #define DBUF_GET_BUFC_TYPE(_db) \
449 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
450
451 #define DBUF_IS_CACHEABLE(_db) \
452 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
453 (dbuf_is_metadata(_db) && \
454 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
455
456 boolean_t dbuf_is_l2cacheable(dmu_buf_impl_t *db);
457
458 #ifdef ZFS_DEBUG
459
460 /*
461 * There should be a ## between the string literal and fmt, to make it
462 * clear that we're joining two strings together, but gcc does not
463 * support that preprocessor token.
464 */
465 #define dprintf_dbuf(dbuf, fmt, ...) do { \
466 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
467 char __db_buf[32]; \
468 uint64_t __db_obj = (dbuf)->db.db_object; \
469 if (__db_obj == DMU_META_DNODE_OBJECT) \
470 (void) strlcpy(__db_buf, "mdn", sizeof (__db_buf)); \
471 else \
472 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
473 (u_longlong_t)__db_obj); \
474 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
475 "obj=%s lvl=%u blkid=%lld " fmt, \
476 __db_buf, (dbuf)->db_level, \
477 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
478 } \
479 } while (0)
480
481 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
482 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
483 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
484 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
485 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
486 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
487 } \
488 } while (0)
489
490 #define DBUF_VERIFY(db) dbuf_verify(db)
491
492 #else
493
494 #define dprintf_dbuf(db, fmt, ...)
495 #define dprintf_dbuf_bp(db, bp, fmt, ...)
496 #define DBUF_VERIFY(db)
497
498 #endif
499
500
501 #ifdef __cplusplus
502 }
503 #endif
504
505 #endif /* _SYS_DBUF_H */
506