1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or https://opensource.org/licenses/CDDL-1.0.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 */
28
29 #ifndef _SYS_DBUF_H
30 #define _SYS_DBUF_H
31
32 #include <sys/dmu.h>
33 #include <sys/spa.h>
34 #include <sys/txg.h>
35 #include <sys/zio.h>
36 #include <sys/arc.h>
37 #include <sys/zfs_context.h>
38 #include <sys/zfs_refcount.h>
39 #include <sys/zrlock.h>
40 #include <sys/multilist.h>
41
42 #ifdef __cplusplus
43 extern "C" {
44 #endif
45
46 #define IN_DMU_SYNC 2
47
48 /*
49 * define flags for dbuf_read
50 */
51
52 #define DB_RF_MUST_SUCCEED (1 << 0)
53 #define DB_RF_CANFAIL (1 << 1)
54 #define DB_RF_HAVESTRUCT (1 << 2)
55 #define DB_RF_NOPREFETCH (1 << 3)
56 #define DB_RF_NEVERWAIT (1 << 4)
57 #define DB_RF_CACHED (1 << 5)
58 #define DB_RF_NO_DECRYPT (1 << 6)
59 #define DB_RF_PARTIAL_FIRST (1 << 7)
60 #define DB_RF_PARTIAL_MORE (1 << 8)
61
62 /*
63 * The simplified state transition diagram for dbufs looks like:
64 *
65 * +-------> READ ------+
66 * | |
67 * | V
68 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
69 * ^ | ^ ^
70 * | | | |
71 * | +-------> FILL ------+ |
72 * | | | |
73 * | | | |
74 * | +------> NOFILL -----+-----> UNCACHED
75 * | | (Direct I/O)
76 * +---------------+
77 *
78 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
79 * to find all dbufs in a range of a dnode and must be less than any other
80 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
81 */
82 typedef enum dbuf_states {
83 DB_MARKER = -2,
84 DB_SEARCH = -1,
85 DB_UNCACHED,
86 DB_FILL,
87 DB_NOFILL,
88 DB_READ,
89 DB_CACHED,
90 DB_EVICTING
91 } dbuf_states_t;
92
93 typedef enum dbuf_cached_state {
94 DB_NO_CACHE = -1,
95 DB_DBUF_CACHE,
96 DB_DBUF_METADATA_CACHE,
97 DB_CACHE_MAX
98 } dbuf_cached_state_t;
99
100 struct dnode;
101 struct dmu_tx;
102
103 /*
104 * level = 0 means the user data
105 * level = 1 means the single indirect block
106 * etc.
107 */
108
109 struct dmu_buf_impl;
110
111 typedef enum override_states {
112 DR_NOT_OVERRIDDEN,
113 DR_IN_DMU_SYNC,
114 DR_OVERRIDDEN
115 } override_states_t;
116
117 typedef enum db_lock_type {
118 DLT_NONE,
119 DLT_PARENT,
120 DLT_OBJSET
121 } db_lock_type_t;
122
123 typedef struct dbuf_dirty_record {
124 /* link on our parents dirty list */
125 list_node_t dr_dirty_node;
126
127 /* transaction group this data will sync in */
128 uint64_t dr_txg;
129
130 /* zio of outstanding write IO */
131 zio_t *dr_zio;
132
133 /* pointer back to our dbuf */
134 struct dmu_buf_impl *dr_dbuf;
135
136 /* list link for dbuf dirty records */
137 list_node_t dr_dbuf_node;
138
139 /*
140 * The dnode we are part of. Note that the dnode can not be moved or
141 * evicted due to the hold that's added by dnode_setdirty() or
142 * dmu_objset_sync_dnodes(), and released by dnode_rele_task() or
143 * userquota_updates_task(). This hold is necessary for
144 * dirty_lightweight_leaf-type dirty records, which don't have a hold
145 * on a dbuf.
146 */
147 dnode_t *dr_dnode;
148
149 /* pointer to parent dirty record */
150 struct dbuf_dirty_record *dr_parent;
151
152 /* How much space was changed to dsl_pool_dirty_space() for this? */
153 unsigned int dr_accounted;
154
155 /* A copy of the bp that points to us */
156 blkptr_t dr_bp_copy;
157
158 union dirty_types {
159 struct dirty_indirect {
160
161 /* protect access to list */
162 kmutex_t dr_mtx;
163
164 /* Our list of dirty children */
165 list_t dr_children;
166 } di;
167 struct dirty_leaf {
168
169 /*
170 * dr_data is set when we dirty the buffer
171 * so that we can retain the pointer even if it
172 * gets COW'd in a subsequent transaction group.
173 */
174 arc_buf_t *dr_data;
175 override_states_t dr_override_state;
176 uint8_t dr_copies;
177 uint8_t dr_gang_copies;
178 boolean_t dr_nopwrite;
179 boolean_t dr_brtwrite;
180 boolean_t dr_diowrite;
181 boolean_t dr_has_raw_params;
182
183 /* Override and raw params are mutually exclusive. */
184 union {
185 blkptr_t dr_overridden_by;
186 struct {
187 /*
188 * If dr_has_raw_params is set, the
189 * following crypt params will be set
190 * on the BP that's written.
191 */
192 boolean_t dr_byteorder;
193 uint8_t dr_salt[ZIO_DATA_SALT_LEN];
194 uint8_t dr_iv[ZIO_DATA_IV_LEN];
195 uint8_t dr_mac[ZIO_DATA_MAC_LEN];
196 };
197 };
198 } dl;
199 struct dirty_lightweight_leaf {
200 /*
201 * This dirty record refers to a leaf (level=0)
202 * block, whose dbuf has not been instantiated for
203 * performance reasons.
204 */
205 uint64_t dr_blkid;
206 abd_t *dr_abd;
207 zio_prop_t dr_props;
208 zio_flag_t dr_flags;
209 } dll;
210 } dt;
211 } dbuf_dirty_record_t;
212
213 typedef struct dmu_buf_impl {
214 /*
215 * The following members are immutable, with the exception of
216 * db.db_data, which is protected by db_mtx.
217 */
218
219 /* the publicly visible structure */
220 dmu_buf_t db;
221
222 /* the objset we belong to */
223 struct objset *db_objset;
224
225 /*
226 * Handle to safely access the dnode we belong to (NULL when evicted)
227 * if dnode_move() is used on the platform, or just dnode otherwise.
228 */
229 #if !defined(__linux__) && !defined(__FreeBSD__)
230 #define USE_DNODE_HANDLE 1
231 struct dnode_handle *db_dnode_handle;
232 #else
233 struct dnode *db_dnode;
234 #endif
235
236 /*
237 * our parent buffer; if the dnode points to us directly,
238 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
239 * only accessed by sync thread ???
240 * (NULL when evicted)
241 * May change from NULL to non-NULL under the protection of db_mtx
242 * (see dbuf_check_blkptr())
243 */
244 struct dmu_buf_impl *db_parent;
245
246 /*
247 * link for hash table of all dmu_buf_impl_t's
248 */
249 struct dmu_buf_impl *db_hash_next;
250
251 /*
252 * Our link on the owner dnodes's dn_dbufs list.
253 * Protected by its dn_dbufs_mtx. Should be on the same cache line
254 * as db_level and db_blkid for the best avl_add() performance.
255 */
256 avl_node_t db_link;
257
258 /* our block number */
259 uint64_t db_blkid;
260
261 /*
262 * Pointer to the blkptr_t which points to us. May be NULL if we
263 * don't have one yet. (NULL when evicted)
264 */
265 blkptr_t *db_blkptr;
266
267 /*
268 * Our indirection level. Data buffers have db_level==0.
269 * Indirect buffers which point to data buffers have
270 * db_level==1. etc. Buffers which contain dnodes have
271 * db_level==0, since the dnodes are stored in a file.
272 */
273 uint8_t db_level;
274
275 /* This block was freed while a read or write was active. */
276 uint8_t db_freed_in_flight;
277
278 /*
279 * Evict user data as soon as the dirty and reference counts are equal.
280 */
281 uint8_t db_user_immediate_evict;
282
283 /*
284 * dnode_evict_dbufs() or dnode_evict_bonus() tried to evict this dbuf,
285 * but couldn't due to outstanding references. Evict once the refcount
286 * drops to 0.
287 */
288 uint8_t db_pending_evict;
289
290 /* Number of TXGs in which this buffer is dirty. */
291 uint8_t db_dirtycnt;
292
293 /* The buffer was partially read. More reads may follow. */
294 uint8_t db_partial_read;
295
296 /*
297 * Protects db_buf's contents if they contain an indirect block or data
298 * block of the meta-dnode. We use this lock to protect the structure of
299 * the block tree. This means that when modifying this dbuf's data, we
300 * grab its rwlock. When modifying its parent's data (including the
301 * blkptr to this dbuf), we grab the parent's rwlock. The lock ordering
302 * for this lock is:
303 * 1) dn_struct_rwlock
304 * 2) db_rwlock
305 * We don't currently grab multiple dbufs' db_rwlocks at once.
306 */
307 krwlock_t db_rwlock;
308
309 /* buffer holding our data */
310 arc_buf_t *db_buf;
311
312 /* db_mtx protects the members below */
313 kmutex_t db_mtx;
314
315 /*
316 * Current state of the buffer
317 */
318 dbuf_states_t db_state;
319
320 /* In which dbuf cache this dbuf is, if any. */
321 dbuf_cached_state_t db_caching_status;
322
323 /*
324 * Refcount accessed by dmu_buf_{hold,rele}.
325 * If nonzero, the buffer can't be destroyed.
326 * Protected by db_mtx.
327 */
328 zfs_refcount_t db_holds;
329
330 kcondvar_t db_changed;
331 dbuf_dirty_record_t *db_data_pending;
332
333 /* List of dirty records for the buffer sorted newest to oldest. */
334 list_t db_dirty_records;
335
336 /* Link in dbuf_cache or dbuf_metadata_cache */
337 multilist_node_t db_cache_link;
338
339 uint64_t db_hash;
340
341 /* User callback information. */
342 dmu_buf_user_t *db_user;
343 } dmu_buf_impl_t;
344
345 #define DBUF_HASH_MUTEX(h, idx) \
346 (&(h)->hash_mutexes[(idx) & ((h)->hash_mutex_mask)])
347
348 typedef struct dbuf_hash_table {
349 uint64_t hash_table_mask;
350 uint64_t hash_mutex_mask;
351 dmu_buf_impl_t **hash_table;
352 kmutex_t *hash_mutexes;
353 } dbuf_hash_table_t;
354
355 typedef void (*dbuf_prefetch_fn)(void *, uint64_t, uint64_t, boolean_t);
356
357 extern kmem_cache_t *dbuf_dirty_kmem_cache;
358
359 uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level,
360 const uint64_t offset);
361
362 void dbuf_create_bonus(struct dnode *dn);
363 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
364
365 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
366
367 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, const void *tag);
368 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
369 const void *tag);
370 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid,
371 boolean_t fail_sparse, boolean_t fail_uncached,
372 const void *tag, dmu_buf_impl_t **dbp);
373
374 int dbuf_prefetch_impl(struct dnode *dn, int64_t level, uint64_t blkid,
375 zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb,
376 void *arg);
377 int dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
378 zio_priority_t prio, arc_flags_t aflags);
379
380 void dbuf_add_ref(dmu_buf_impl_t *db, const void *tag);
381 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj,
382 uint64_t blkid, const void *tag);
383 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
384
385 void dbuf_rele(dmu_buf_impl_t *db, const void *tag);
386 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, const void *tag,
387 boolean_t evicting);
388
389 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
390 uint64_t blkid, uint64_t *hash_out);
391
392 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
393 void dmu_buf_will_clone_or_dio(dmu_buf_t *db, dmu_tx_t *tx);
394 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
395 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx, boolean_t canfail);
396 boolean_t dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx, boolean_t failed);
397 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
398 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
399 dbuf_dirty_record_t *dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid,
400 dmu_tx_t *tx);
401 boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
402 int dmu_buf_get_bp_from_dbuf(dmu_buf_impl_t *db, blkptr_t **bp);
403 int dmu_buf_untransform_direct(dmu_buf_impl_t *db, spa_t *spa);
404 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
405 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
406 bp_embedded_type_t etype, enum zio_compress comp,
407 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
408
409 int dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd,
410 const struct zio_prop *zp, zio_flag_t flags, dmu_tx_t *tx);
411
412 void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx);
413 void dbuf_destroy(dmu_buf_impl_t *db);
414
415 void dbuf_unoverride(dbuf_dirty_record_t *dr);
416 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx);
417 void dbuf_release_bp(dmu_buf_impl_t *db);
418 db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw,
419 const void *tag);
420 void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type,
421 const void *tag);
422
423 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
424 struct dmu_tx *);
425
426 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
427
428 void dbuf_stats_init(dbuf_hash_table_t *hash);
429 void dbuf_stats_destroy(void);
430
431 int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid,
432 blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift);
433
434 #ifdef USE_DNODE_HANDLE
435 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
436 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
437 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
438 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
439 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
440 #else
441 #define DB_DNODE(_db) ((_db)->db_dnode)
442 #define DB_DNODE_LOCK(_db)
443 #define DB_DNODE_ENTER(_db)
444 #define DB_DNODE_EXIT(_db)
445 #define DB_DNODE_HELD(_db) (B_TRUE)
446 #endif
447
448 void dbuf_init(void);
449 void dbuf_fini(void);
450
451 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
452
453 static inline dbuf_dirty_record_t *
dbuf_find_dirty_lte(dmu_buf_impl_t * db,uint64_t txg)454 dbuf_find_dirty_lte(dmu_buf_impl_t *db, uint64_t txg)
455 {
456 dbuf_dirty_record_t *dr;
457
458 for (dr = list_head(&db->db_dirty_records);
459 dr != NULL && dr->dr_txg > txg;
460 dr = list_next(&db->db_dirty_records, dr))
461 continue;
462 return (dr);
463 }
464
465 static inline dbuf_dirty_record_t *
dbuf_find_dirty_eq(dmu_buf_impl_t * db,uint64_t txg)466 dbuf_find_dirty_eq(dmu_buf_impl_t *db, uint64_t txg)
467 {
468 dbuf_dirty_record_t *dr;
469
470 dr = dbuf_find_dirty_lte(db, txg);
471 if (dr && dr->dr_txg == txg)
472 return (dr);
473 return (NULL);
474 }
475
476 #define DBUF_GET_BUFC_TYPE(_db) \
477 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
478
479 #define DBUF_IS_CACHEABLE(_db) \
480 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
481 (dbuf_is_metadata(_db) && \
482 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
483
484 boolean_t dbuf_is_l2cacheable(dmu_buf_impl_t *db, blkptr_t *db_bp);
485
486 #ifdef ZFS_DEBUG
487
488 /*
489 * There should be a ## between the string literal and fmt, to make it
490 * clear that we're joining two strings together, but gcc does not
491 * support that preprocessor token.
492 */
493 #define dprintf_dbuf(dbuf, fmt, ...) do { \
494 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
495 char __db_buf[32]; \
496 uint64_t __db_obj = (dbuf)->db.db_object; \
497 if (__db_obj == DMU_META_DNODE_OBJECT) \
498 (void) strlcpy(__db_buf, "mdn", sizeof (__db_buf)); \
499 else \
500 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
501 (u_longlong_t)__db_obj); \
502 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
503 "obj=%s lvl=%u blkid=%lld " fmt, \
504 __db_buf, (dbuf)->db_level, \
505 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
506 } \
507 } while (0)
508
509 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
510 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
511 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
512 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
513 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
514 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
515 } \
516 } while (0)
517
518 #define DBUF_VERIFY(db) dbuf_verify(db)
519
520 #else
521
522 #define dprintf_dbuf(db, fmt, ...)
523 #define dprintf_dbuf_bp(db, bp, fmt, ...)
524 #define DBUF_VERIFY(db)
525
526 #endif
527
528
529 #ifdef __cplusplus
530 }
531 #endif
532
533 #endif /* _SYS_DBUF_H */
534