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_MARKER = -2,
83 DB_SEARCH = -1,
84 DB_UNCACHED,
85 DB_FILL,
86 DB_NOFILL,
87 DB_READ,
88 DB_CACHED,
89 DB_EVICTING
90 } dbuf_states_t;
91
92 typedef enum dbuf_cached_state {
93 DB_NO_CACHE = -1,
94 DB_DBUF_CACHE,
95 DB_DBUF_METADATA_CACHE,
96 DB_CACHE_MAX
97 } dbuf_cached_state_t;
98
99 struct dnode;
100 struct dmu_tx;
101
102 /*
103 * level = 0 means the user data
104 * level = 1 means the single indirect block
105 * etc.
106 */
107
108 struct dmu_buf_impl;
109
110 typedef enum override_states {
111 DR_NOT_OVERRIDDEN,
112 DR_IN_DMU_SYNC,
113 DR_OVERRIDDEN
114 } override_states_t;
115
116 typedef enum db_lock_type {
117 DLT_NONE,
118 DLT_PARENT,
119 DLT_OBJSET
120 } db_lock_type_t;
121
122 typedef struct dbuf_dirty_record {
123 /* link on our parents dirty list */
124 list_node_t dr_dirty_node;
125
126 /* transaction group this data will sync in */
127 uint64_t dr_txg;
128
129 /* zio of outstanding write IO */
130 zio_t *dr_zio;
131
132 /* pointer back to our dbuf */
133 struct dmu_buf_impl *dr_dbuf;
134
135 /* list link for dbuf dirty records */
136 list_node_t dr_dbuf_node;
137
138 /*
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
144 * on a dbuf.
145 */
146 dnode_t *dr_dnode;
147
148 /* pointer to parent dirty record */
149 struct dbuf_dirty_record *dr_parent;
150
151 /* How much space was changed to dsl_pool_dirty_space() for this? */
152 unsigned int dr_accounted;
153
154 /* A copy of the bp that points to us */
155 blkptr_t dr_bp_copy;
156
157 union dirty_types {
158 struct dirty_indirect {
159
160 /* protect access to list */
161 kmutex_t dr_mtx;
162
163 /* Our list of dirty children */
164 list_t dr_children;
165 } di;
166 struct dirty_leaf {
167
168 /*
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.
172 */
173 arc_buf_t *dr_data;
174 blkptr_t dr_overridden_by;
175 override_states_t dr_override_state;
176 uint8_t dr_copies;
177 boolean_t dr_nopwrite;
178 boolean_t dr_brtwrite;
179 boolean_t dr_has_raw_params;
180
181 /*
182 * If dr_has_raw_params is set, the following crypt
183 * params will be set on the BP that's written.
184 */
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];
189 } dl;
190 struct dirty_lightweight_leaf {
191 /*
192 * This dirty record refers to a leaf (level=0)
193 * block, whose dbuf has not been instantiated for
194 * performance reasons.
195 */
196 uint64_t dr_blkid;
197 abd_t *dr_abd;
198 zio_prop_t dr_props;
199 zio_flag_t dr_flags;
200 } dll;
201 } dt;
202 } dbuf_dirty_record_t;
203
204 typedef struct dmu_buf_impl {
205 /*
206 * The following members are immutable, with the exception of
207 * db.db_data, which is protected by db_mtx.
208 */
209
210 /* the publicly visible structure */
211 dmu_buf_t db;
212
213 /* the objset we belong to */
214 struct objset *db_objset;
215
216 /*
217 * handle to safely access the dnode we belong to (NULL when evicted)
218 */
219 struct dnode_handle *db_dnode_handle;
220
221 /*
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())
228 */
229 struct dmu_buf_impl *db_parent;
230
231 /*
232 * link for hash table of all dmu_buf_impl_t's
233 */
234 struct dmu_buf_impl *db_hash_next;
235
236 /*
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.
240 */
241 avl_node_t db_link;
242
243 /* our block number */
244 uint64_t db_blkid;
245
246 /*
247 * Pointer to the blkptr_t which points to us. May be NULL if we
248 * don't have one yet. (NULL when evicted)
249 */
250 blkptr_t *db_blkptr;
251
252 /*
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.
257 */
258 uint8_t db_level;
259
260 /*
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
266 * for this lock is:
267 * 1) dn_struct_rwlock
268 * 2) db_rwlock
269 * We don't currently grab multiple dbufs' db_rwlocks at once.
270 */
271 krwlock_t db_rwlock;
272
273 /* buffer holding our data */
274 arc_buf_t *db_buf;
275
276 /* db_mtx protects the members below */
277 kmutex_t db_mtx;
278
279 /*
280 * Current state of the buffer
281 */
282 dbuf_states_t db_state;
283
284 /*
285 * Refcount accessed by dmu_buf_{hold,rele}.
286 * If nonzero, the buffer can't be destroyed.
287 * Protected by db_mtx.
288 */
289 zfs_refcount_t db_holds;
290
291 kcondvar_t db_changed;
292 dbuf_dirty_record_t *db_data_pending;
293
294 /* List of dirty records for the buffer sorted newest to oldest. */
295 list_t db_dirty_records;
296
297 /* Link in dbuf_cache or dbuf_metadata_cache */
298 multilist_node_t db_cache_link;
299
300 /* Tells us which dbuf cache this dbuf is in, if any */
301 dbuf_cached_state_t db_caching_status;
302
303 uint64_t db_hash;
304
305 /* Data which is unique to data (leaf) blocks: */
306
307 /* User callback information. */
308 dmu_buf_user_t *db_user;
309
310 /*
311 * Evict user data as soon as the dirty and reference
312 * counts are equal.
313 */
314 uint8_t db_user_immediate_evict;
315
316 /*
317 * This block was freed while a read or write was
318 * active.
319 */
320 uint8_t db_freed_in_flight;
321
322 /*
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.
326 */
327 uint8_t db_pending_evict;
328
329 uint8_t db_dirtycnt;
330
331 /* The buffer was partially read. More reads may follow. */
332 uint8_t db_partial_read;
333 } dmu_buf_impl_t;
334
335 #define DBUF_HASH_MUTEX(h, idx) \
336 (&(h)->hash_mutexes[(idx) & ((h)->hash_mutex_mask)])
337
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;
343 } dbuf_hash_table_t;
344
345 typedef void (*dbuf_prefetch_fn)(void *, uint64_t, uint64_t, boolean_t);
346
347 uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level,
348 const uint64_t offset);
349
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);
352
353 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
354
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,
357 const void *tag);
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);
361
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,
364 void *arg);
365 int dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
366 zio_priority_t prio, arc_flags_t aflags);
367
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);
372
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,
375 boolean_t evicting);
376
377 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
378 uint64_t blkid, uint64_t *hash_out);
379
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,
388 dmu_tx_t *tx);
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);
394
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);
397
398 void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx);
399 void dbuf_destroy(dmu_buf_impl_t *db);
400
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,
405 const void *tag);
406 void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type,
407 const void *tag);
408
409 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
410 struct dmu_tx *);
411
412 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
413
414 void dbuf_stats_init(dbuf_hash_table_t *hash);
415 void dbuf_stats_destroy(void);
416
417 int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid,
418 blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift);
419
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)))
425
426 void dbuf_init(void);
427 void dbuf_fini(void);
428
429 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
430
431 static inline dbuf_dirty_record_t *
dbuf_find_dirty_lte(dmu_buf_impl_t * db,uint64_t txg)432 dbuf_find_dirty_lte(dmu_buf_impl_t *db, uint64_t txg)
433 {
434 dbuf_dirty_record_t *dr;
435
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))
439 continue;
440 return (dr);
441 }
442
443 static inline dbuf_dirty_record_t *
dbuf_find_dirty_eq(dmu_buf_impl_t * db,uint64_t txg)444 dbuf_find_dirty_eq(dmu_buf_impl_t *db, uint64_t txg)
445 {
446 dbuf_dirty_record_t *dr;
447
448 dr = dbuf_find_dirty_lte(db, txg);
449 if (dr && dr->dr_txg == txg)
450 return (dr);
451 return (NULL);
452 }
453
454 #define DBUF_GET_BUFC_TYPE(_db) \
455 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
456
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)))
461
462 boolean_t dbuf_is_l2cacheable(dmu_buf_impl_t *db);
463
464 #ifdef ZFS_DEBUG
465
466 /*
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.
470 */
471 #define dprintf_dbuf(dbuf, fmt, ...) do { \
472 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
473 char __db_buf[32]; \
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)); \
477 else \
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__); \
484 } \
485 } while (0)
486
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); \
493 } \
494 } while (0)
495
496 #define DBUF_VERIFY(db) dbuf_verify(db)
497
498 #else
499
500 #define dprintf_dbuf(db, fmt, ...)
501 #define dprintf_dbuf_bp(db, bp, fmt, ...)
502 #define DBUF_VERIFY(db)
503
504 #endif
505
506
507 #ifdef __cplusplus
508 }
509 #endif
510
511 #endif /* _SYS_DBUF_H */
512