Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * bufmgr.h
4 : * POSTGRES buffer manager definitions.
5 : *
6 : *
7 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
8 : * Portions Copyright (c) 1994, Regents of the University of California
9 : *
10 : * src/include/storage/bufmgr.h
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 : #ifndef BUFMGR_H
15 : #define BUFMGR_H
16 :
17 : #include "port/pg_iovec.h"
18 : #include "storage/aio_types.h"
19 : #include "storage/block.h"
20 : #include "storage/buf.h"
21 : #include "storage/bufpage.h"
22 : #include "storage/relfilelocator.h"
23 : #include "utils/relcache.h"
24 : #include "utils/snapmgr.h"
25 :
26 : typedef void *Block;
27 :
28 : /*
29 : * Possible arguments for GetAccessStrategy().
30 : *
31 : * If adding a new BufferAccessStrategyType, also add a new IOContext so
32 : * IO statistics using this strategy are tracked.
33 : */
34 : typedef enum BufferAccessStrategyType
35 : {
36 : BAS_NORMAL, /* Normal random access */
37 : BAS_BULKREAD, /* Large read-only scan (hint bit updates are
38 : * ok) */
39 : BAS_BULKWRITE, /* Large multi-block write (e.g. COPY IN) */
40 : BAS_VACUUM, /* VACUUM */
41 : } BufferAccessStrategyType;
42 :
43 : /* Possible modes for ReadBufferExtended() */
44 : typedef enum
45 : {
46 : RBM_NORMAL, /* Normal read */
47 : RBM_ZERO_AND_LOCK, /* Don't read from disk, caller will
48 : * initialize. Also locks the page. */
49 : RBM_ZERO_AND_CLEANUP_LOCK, /* Like RBM_ZERO_AND_LOCK, but locks the page
50 : * in "cleanup" mode */
51 : RBM_ZERO_ON_ERROR, /* Read, but return an all-zeros page on error */
52 : RBM_NORMAL_NO_LOG, /* Don't log page as invalid during WAL
53 : * replay; otherwise same as RBM_NORMAL */
54 : } ReadBufferMode;
55 :
56 : /*
57 : * Type returned by PrefetchBuffer().
58 : */
59 : typedef struct PrefetchBufferResult
60 : {
61 : Buffer recent_buffer; /* If valid, a hit (recheck needed!) */
62 : bool initiated_io; /* If true, a miss resulting in async I/O */
63 : } PrefetchBufferResult;
64 :
65 : /*
66 : * Flags influencing the behaviour of ExtendBufferedRel*
67 : */
68 : typedef enum ExtendBufferedFlags
69 : {
70 : /*
71 : * Don't acquire extension lock. This is safe only if the relation isn't
72 : * shared, an access exclusive lock is held or if this is the startup
73 : * process.
74 : */
75 : EB_SKIP_EXTENSION_LOCK = (1 << 0),
76 :
77 : /* Is this extension part of recovery? */
78 : EB_PERFORMING_RECOVERY = (1 << 1),
79 :
80 : /*
81 : * Should the fork be created if it does not currently exist? This likely
82 : * only ever makes sense for relation forks.
83 : */
84 : EB_CREATE_FORK_IF_NEEDED = (1 << 2),
85 :
86 : /* Should the first (possibly only) return buffer be returned locked? */
87 : EB_LOCK_FIRST = (1 << 3),
88 :
89 : /* Should the smgr size cache be cleared? */
90 : EB_CLEAR_SIZE_CACHE = (1 << 4),
91 :
92 : /* internal flags follow */
93 : EB_LOCK_TARGET = (1 << 5),
94 : } ExtendBufferedFlags;
95 :
96 : /* forward declared, to avoid including smgr.h here */
97 : typedef struct SMgrRelationData *SMgrRelation;
98 :
99 : /*
100 : * Some functions identify relations either by relation or smgr +
101 : * relpersistence, initialized via the BMR_REL()/BMR_SMGR() macros below.
102 : * This allows us to use the same function for both recovery and normal
103 : * operation. When BMR_REL is used, it's not valid to cache its rd_smgr here,
104 : * because our pointer would be obsolete in case of relcache invalidation.
105 : * For simplicity, use BMR_GET_SMGR to read the smgr.
106 : */
107 : typedef struct BufferManagerRelation
108 : {
109 : Relation rel;
110 : SMgrRelation smgr;
111 : char relpersistence;
112 : } BufferManagerRelation;
113 :
114 : #define BMR_REL(p_rel) \
115 : ((BufferManagerRelation){.rel = p_rel})
116 : #define BMR_SMGR(p_smgr, p_relpersistence) \
117 : ((BufferManagerRelation){.smgr = p_smgr, .relpersistence = p_relpersistence})
118 : #define BMR_GET_SMGR(bmr) \
119 : (RelationIsValid((bmr).rel) ? RelationGetSmgr((bmr).rel) : (bmr).smgr)
120 :
121 : /* Zero out page if reading fails. */
122 : #define READ_BUFFERS_ZERO_ON_ERROR (1 << 0)
123 : /* Call smgrprefetch() if I/O necessary. */
124 : #define READ_BUFFERS_ISSUE_ADVICE (1 << 1)
125 : /* Don't treat page as invalid due to checksum failures. */
126 : #define READ_BUFFERS_IGNORE_CHECKSUM_FAILURES (1 << 2)
127 : /* IO will immediately be waited for */
128 : #define READ_BUFFERS_SYNCHRONOUSLY (1 << 3)
129 :
130 :
131 : struct ReadBuffersOperation
132 : {
133 : /* The following members should be set by the caller. */
134 : Relation rel; /* optional */
135 : SMgrRelation smgr;
136 : char persistence;
137 : ForkNumber forknum;
138 : BufferAccessStrategy strategy;
139 :
140 : /*
141 : * The following private members are private state for communication
142 : * between StartReadBuffers() and WaitReadBuffers(), initialized only if
143 : * an actual read is required, and should not be modified.
144 : */
145 : Buffer *buffers;
146 : BlockNumber blocknum;
147 : int flags;
148 : int16 nblocks;
149 : int16 nblocks_done;
150 : PgAioWaitRef io_wref;
151 : PgAioReturn io_return;
152 : };
153 :
154 : typedef struct ReadBuffersOperation ReadBuffersOperation;
155 :
156 : /* to avoid having to expose buf_internals.h here */
157 : typedef struct WritebackContext WritebackContext;
158 :
159 : /* in globals.c ... this duplicates miscadmin.h */
160 : extern PGDLLIMPORT int NBuffers;
161 :
162 : /* in bufmgr.c */
163 : extern PGDLLIMPORT bool zero_damaged_pages;
164 : extern PGDLLIMPORT int bgwriter_lru_maxpages;
165 : extern PGDLLIMPORT double bgwriter_lru_multiplier;
166 : extern PGDLLIMPORT bool track_io_timing;
167 :
168 : #define DEFAULT_EFFECTIVE_IO_CONCURRENCY 16
169 : #define DEFAULT_MAINTENANCE_IO_CONCURRENCY 16
170 : extern PGDLLIMPORT int effective_io_concurrency;
171 : extern PGDLLIMPORT int maintenance_io_concurrency;
172 :
173 : #define MAX_IO_COMBINE_LIMIT PG_IOV_MAX
174 : #define DEFAULT_IO_COMBINE_LIMIT Min(MAX_IO_COMBINE_LIMIT, (128 * 1024) / BLCKSZ)
175 : extern PGDLLIMPORT int io_combine_limit; /* min of the two GUCs below */
176 : extern PGDLLIMPORT int io_combine_limit_guc;
177 : extern PGDLLIMPORT int io_max_combine_limit;
178 :
179 : extern PGDLLIMPORT int checkpoint_flush_after;
180 : extern PGDLLIMPORT int backend_flush_after;
181 : extern PGDLLIMPORT int bgwriter_flush_after;
182 :
183 : extern PGDLLIMPORT const PgAioHandleCallbacks aio_shared_buffer_readv_cb;
184 : extern PGDLLIMPORT const PgAioHandleCallbacks aio_local_buffer_readv_cb;
185 :
186 : /* in buf_init.c */
187 : extern PGDLLIMPORT char *BufferBlocks;
188 :
189 : /* in localbuf.c */
190 : extern PGDLLIMPORT int NLocBuffer;
191 : extern PGDLLIMPORT Block *LocalBufferBlockPointers;
192 : extern PGDLLIMPORT int32 *LocalRefCount;
193 :
194 : /* upper limit for effective_io_concurrency */
195 : #define MAX_IO_CONCURRENCY 1000
196 :
197 : /* special block number for ReadBuffer() */
198 : #define P_NEW InvalidBlockNumber /* grow the file to get a new page */
199 :
200 : /*
201 : * Buffer content lock modes (mode argument for LockBuffer())
202 : */
203 : typedef enum BufferLockMode
204 : {
205 : BUFFER_LOCK_UNLOCK,
206 : BUFFER_LOCK_SHARE,
207 : BUFFER_LOCK_EXCLUSIVE,
208 : } BufferLockMode;
209 :
210 :
211 : /*
212 : * prototypes for functions in bufmgr.c
213 : */
214 : extern PrefetchBufferResult PrefetchSharedBuffer(SMgrRelation smgr_reln,
215 : ForkNumber forkNum,
216 : BlockNumber blockNum);
217 : extern PrefetchBufferResult PrefetchBuffer(Relation reln, ForkNumber forkNum,
218 : BlockNumber blockNum);
219 : extern bool ReadRecentBuffer(RelFileLocator rlocator, ForkNumber forkNum,
220 : BlockNumber blockNum, Buffer recent_buffer);
221 : extern Buffer ReadBuffer(Relation reln, BlockNumber blockNum);
222 : extern Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum,
223 : BlockNumber blockNum, ReadBufferMode mode,
224 : BufferAccessStrategy strategy);
225 : extern Buffer ReadBufferWithoutRelcache(RelFileLocator rlocator,
226 : ForkNumber forkNum, BlockNumber blockNum,
227 : ReadBufferMode mode, BufferAccessStrategy strategy,
228 : bool permanent);
229 :
230 : extern bool StartReadBuffer(ReadBuffersOperation *operation,
231 : Buffer *buffer,
232 : BlockNumber blocknum,
233 : int flags);
234 : extern bool StartReadBuffers(ReadBuffersOperation *operation,
235 : Buffer *buffers,
236 : BlockNumber blockNum,
237 : int *nblocks,
238 : int flags);
239 : extern void WaitReadBuffers(ReadBuffersOperation *operation);
240 :
241 : extern void ReleaseBuffer(Buffer buffer);
242 : extern void UnlockReleaseBuffer(Buffer buffer);
243 : extern bool BufferIsLockedByMe(Buffer buffer);
244 : extern bool BufferIsLockedByMeInMode(Buffer buffer, BufferLockMode mode);
245 : extern bool BufferIsDirty(Buffer buffer);
246 : extern void MarkBufferDirty(Buffer buffer);
247 : extern void IncrBufferRefCount(Buffer buffer);
248 : extern void CheckBufferIsPinnedOnce(Buffer buffer);
249 : extern Buffer ReleaseAndReadBuffer(Buffer buffer, Relation relation,
250 : BlockNumber blockNum);
251 :
252 : extern Buffer ExtendBufferedRel(BufferManagerRelation bmr,
253 : ForkNumber forkNum,
254 : BufferAccessStrategy strategy,
255 : uint32 flags);
256 : extern BlockNumber ExtendBufferedRelBy(BufferManagerRelation bmr,
257 : ForkNumber fork,
258 : BufferAccessStrategy strategy,
259 : uint32 flags,
260 : uint32 extend_by,
261 : Buffer *buffers,
262 : uint32 *extended_by);
263 : extern Buffer ExtendBufferedRelTo(BufferManagerRelation bmr,
264 : ForkNumber fork,
265 : BufferAccessStrategy strategy,
266 : uint32 flags,
267 : BlockNumber extend_to,
268 : ReadBufferMode mode);
269 :
270 : extern void InitBufferManagerAccess(void);
271 : extern void AtEOXact_Buffers(bool isCommit);
272 : #ifdef USE_ASSERT_CHECKING
273 : extern void AssertBufferLocksPermitCatalogRead(void);
274 : #endif
275 : extern char *DebugPrintBufferRefcount(Buffer buffer);
276 : extern void CheckPointBuffers(int flags);
277 : extern BlockNumber BufferGetBlockNumber(Buffer buffer);
278 : extern BlockNumber RelationGetNumberOfBlocksInFork(Relation relation,
279 : ForkNumber forkNum);
280 : extern void FlushOneBuffer(Buffer buffer);
281 : extern void FlushRelationBuffers(Relation rel);
282 : extern void FlushRelationsAllBuffers(SMgrRelation *smgrs, int nrels);
283 : extern void CreateAndCopyRelationData(RelFileLocator src_rlocator,
284 : RelFileLocator dst_rlocator,
285 : bool permanent);
286 : extern void FlushDatabaseBuffers(Oid dbid);
287 : extern void DropRelationBuffers(SMgrRelation smgr_reln,
288 : ForkNumber *forkNum,
289 : int nforks, BlockNumber *firstDelBlock);
290 : extern void DropRelationsAllBuffers(SMgrRelation *smgr_reln,
291 : int nlocators);
292 : extern void DropDatabaseBuffers(Oid dbid);
293 :
294 : #define RelationGetNumberOfBlocks(reln) \
295 : RelationGetNumberOfBlocksInFork(reln, MAIN_FORKNUM)
296 :
297 : extern bool BufferIsPermanent(Buffer buffer);
298 : extern XLogRecPtr BufferGetLSNAtomic(Buffer buffer);
299 : extern void BufferGetTag(Buffer buffer, RelFileLocator *rlocator,
300 : ForkNumber *forknum, BlockNumber *blknum);
301 :
302 : extern void MarkBufferDirtyHint(Buffer buffer, bool buffer_std);
303 :
304 : extern void UnlockBuffers(void);
305 : extern void LockBuffer(Buffer buffer, BufferLockMode mode);
306 : extern bool ConditionalLockBuffer(Buffer buffer);
307 : extern void LockBufferForCleanup(Buffer buffer);
308 : extern bool ConditionalLockBufferForCleanup(Buffer buffer);
309 : extern bool IsBufferCleanupOK(Buffer buffer);
310 : extern bool HoldingBufferPinThatDelaysRecovery(void);
311 :
312 : extern bool BgBufferSync(WritebackContext *wb_context);
313 :
314 : extern uint32 GetPinLimit(void);
315 : extern uint32 GetLocalPinLimit(void);
316 : extern uint32 GetAdditionalPinLimit(void);
317 : extern uint32 GetAdditionalLocalPinLimit(void);
318 : extern void LimitAdditionalPins(uint32 *additional_pins);
319 : extern void LimitAdditionalLocalPins(uint32 *additional_pins);
320 :
321 : extern bool EvictUnpinnedBuffer(Buffer buf, bool *buffer_flushed);
322 : extern void EvictAllUnpinnedBuffers(int32 *buffers_evicted,
323 : int32 *buffers_flushed,
324 : int32 *buffers_skipped);
325 : extern void EvictRelUnpinnedBuffers(Relation rel,
326 : int32 *buffers_evicted,
327 : int32 *buffers_flushed,
328 : int32 *buffers_skipped);
329 : extern bool MarkDirtyUnpinnedBuffer(Buffer buf, bool *buffer_already_dirty);
330 : extern void MarkDirtyRelUnpinnedBuffers(Relation rel,
331 : int32 *buffers_dirtied,
332 : int32 *buffers_already_dirty,
333 : int32 *buffers_skipped);
334 : extern void MarkDirtyAllUnpinnedBuffers(int32 *buffers_dirtied,
335 : int32 *buffers_already_dirty,
336 : int32 *buffers_skipped);
337 :
338 : /* in buf_init.c */
339 : extern void BufferManagerShmemInit(void);
340 : extern Size BufferManagerShmemSize(void);
341 :
342 : /* in localbuf.c */
343 : extern void AtProcExit_LocalBuffers(void);
344 :
345 : /* in freelist.c */
346 :
347 : extern BufferAccessStrategy GetAccessStrategy(BufferAccessStrategyType btype);
348 : extern BufferAccessStrategy GetAccessStrategyWithSize(BufferAccessStrategyType btype,
349 : int ring_size_kb);
350 : extern int GetAccessStrategyBufferCount(BufferAccessStrategy strategy);
351 : extern int GetAccessStrategyPinLimit(BufferAccessStrategy strategy);
352 :
353 : extern void FreeAccessStrategy(BufferAccessStrategy strategy);
354 :
355 :
356 : /* inline functions */
357 :
358 : /*
359 : * Although this header file is nominally backend-only, certain frontend
360 : * programs like pg_waldump include it. For compilers that emit static
361 : * inline functions even when they're unused, that leads to unsatisfied
362 : * external references; hence hide these with #ifndef FRONTEND.
363 : */
364 :
365 : #ifndef FRONTEND
366 :
367 : /*
368 : * BufferIsValid
369 : * True iff the given buffer number is valid (either as a shared
370 : * or local buffer).
371 : *
372 : * Note: For a long time this was defined the same as BufferIsPinned,
373 : * that is it would say False if you didn't hold a pin on the buffer.
374 : * I believe this was bogus and served only to mask logic errors.
375 : * Code should always know whether it has a buffer reference,
376 : * independently of the pin state.
377 : *
378 : * Note: For a further long time this was not quite the inverse of the
379 : * BufferIsInvalid() macro, in that it also did sanity checks to verify
380 : * that the buffer number was in range. Most likely, this macro was
381 : * originally intended only to be used in assertions, but its use has
382 : * since expanded quite a bit, and the overhead of making those checks
383 : * even in non-assert-enabled builds can be significant. Thus, we've
384 : * now demoted the range checks to assertions within the macro itself.
385 : */
386 : static inline bool
387 639974644 : BufferIsValid(Buffer bufnum)
388 : {
389 : Assert(bufnum <= NBuffers);
390 : Assert(bufnum >= -NLocBuffer);
391 :
392 639974644 : return bufnum != InvalidBuffer;
393 : }
394 :
395 : /*
396 : * BufferGetBlock
397 : * Returns a reference to a disk page image associated with a buffer.
398 : *
399 : * Note:
400 : * Assumes buffer is valid.
401 : */
402 : static inline Block
403 736749282 : BufferGetBlock(Buffer buffer)
404 : {
405 : Assert(BufferIsValid(buffer));
406 :
407 736749282 : if (BufferIsLocal(buffer))
408 23450254 : return LocalBufferBlockPointers[-buffer - 1];
409 : else
410 713299028 : return (Block) (BufferBlocks + ((Size) (buffer - 1)) * BLCKSZ);
411 : }
412 :
413 : /*
414 : * BufferGetPageSize
415 : * Returns the page size within a buffer.
416 : *
417 : * Notes:
418 : * Assumes buffer is valid.
419 : *
420 : * The buffer can be a raw disk block and need not contain a valid
421 : * (formatted) disk page.
422 : */
423 : /* XXX should dig out of buffer descriptor */
424 : static inline Size
425 459618 : BufferGetPageSize(Buffer buffer)
426 : {
427 : Assert(BufferIsValid(buffer));
428 459618 : return (Size) BLCKSZ;
429 : }
430 :
431 : /*
432 : * BufferGetPage
433 : * Returns the page associated with a buffer.
434 : */
435 : static inline Page
436 730810432 : BufferGetPage(Buffer buffer)
437 : {
438 730810432 : return (Page) BufferGetBlock(buffer);
439 : }
440 :
441 : #endif /* FRONTEND */
442 :
443 : #endif /* BUFMGR_H */
|
