LCOV - code coverage report
Current view: top level - src/include/storage - bufmgr.h (source / functions) Hit Total Coverage
Test: PostgreSQL 19devel Lines: 10 10 100.0 %
Date: 2025-10-23 17:17:24 Functions: 4 4 100.0 %
Legend: Lines: hit not hit

          Line data    Source code
       1             : /*-------------------------------------------------------------------------
       2             :  *
       3             :  * bufmgr.h
       4             :  *    POSTGRES buffer manager definitions.
       5             :  *
       6             :  *
       7             :  * Portions Copyright (c) 1996-2025, 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             : #define BUFFER_LOCK_UNLOCK      0
     204             : #define BUFFER_LOCK_SHARE       1
     205             : #define BUFFER_LOCK_EXCLUSIVE   2
     206             : 
     207             : 
     208             : /*
     209             :  * prototypes for functions in bufmgr.c
     210             :  */
     211             : extern PrefetchBufferResult PrefetchSharedBuffer(SMgrRelation smgr_reln,
     212             :                                                  ForkNumber forkNum,
     213             :                                                  BlockNumber blockNum);
     214             : extern PrefetchBufferResult PrefetchBuffer(Relation reln, ForkNumber forkNum,
     215             :                                            BlockNumber blockNum);
     216             : extern bool ReadRecentBuffer(RelFileLocator rlocator, ForkNumber forkNum,
     217             :                              BlockNumber blockNum, Buffer recent_buffer);
     218             : extern Buffer ReadBuffer(Relation reln, BlockNumber blockNum);
     219             : extern Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum,
     220             :                                  BlockNumber blockNum, ReadBufferMode mode,
     221             :                                  BufferAccessStrategy strategy);
     222             : extern Buffer ReadBufferWithoutRelcache(RelFileLocator rlocator,
     223             :                                         ForkNumber forkNum, BlockNumber blockNum,
     224             :                                         ReadBufferMode mode, BufferAccessStrategy strategy,
     225             :                                         bool permanent);
     226             : 
     227             : extern bool StartReadBuffer(ReadBuffersOperation *operation,
     228             :                             Buffer *buffer,
     229             :                             BlockNumber blocknum,
     230             :                             int flags);
     231             : extern bool StartReadBuffers(ReadBuffersOperation *operation,
     232             :                              Buffer *buffers,
     233             :                              BlockNumber blockNum,
     234             :                              int *nblocks,
     235             :                              int flags);
     236             : extern void WaitReadBuffers(ReadBuffersOperation *operation);
     237             : 
     238             : extern void ReleaseBuffer(Buffer buffer);
     239             : extern void UnlockReleaseBuffer(Buffer buffer);
     240             : extern bool BufferIsLockedByMe(Buffer buffer);
     241             : extern bool BufferIsLockedByMeInMode(Buffer buffer, int mode);
     242             : extern bool BufferIsDirty(Buffer buffer);
     243             : extern void MarkBufferDirty(Buffer buffer);
     244             : extern void IncrBufferRefCount(Buffer buffer);
     245             : extern void CheckBufferIsPinnedOnce(Buffer buffer);
     246             : extern Buffer ReleaseAndReadBuffer(Buffer buffer, Relation relation,
     247             :                                    BlockNumber blockNum);
     248             : 
     249             : extern Buffer ExtendBufferedRel(BufferManagerRelation bmr,
     250             :                                 ForkNumber forkNum,
     251             :                                 BufferAccessStrategy strategy,
     252             :                                 uint32 flags);
     253             : extern BlockNumber ExtendBufferedRelBy(BufferManagerRelation bmr,
     254             :                                        ForkNumber fork,
     255             :                                        BufferAccessStrategy strategy,
     256             :                                        uint32 flags,
     257             :                                        uint32 extend_by,
     258             :                                        Buffer *buffers,
     259             :                                        uint32 *extended_by);
     260             : extern Buffer ExtendBufferedRelTo(BufferManagerRelation bmr,
     261             :                                   ForkNumber fork,
     262             :                                   BufferAccessStrategy strategy,
     263             :                                   uint32 flags,
     264             :                                   BlockNumber extend_to,
     265             :                                   ReadBufferMode mode);
     266             : 
     267             : extern void InitBufferManagerAccess(void);
     268             : extern void AtEOXact_Buffers(bool isCommit);
     269             : #ifdef USE_ASSERT_CHECKING
     270             : extern void AssertBufferLocksPermitCatalogRead(void);
     271             : #endif
     272             : extern char *DebugPrintBufferRefcount(Buffer buffer);
     273             : extern void CheckPointBuffers(int flags);
     274             : extern BlockNumber BufferGetBlockNumber(Buffer buffer);
     275             : extern BlockNumber RelationGetNumberOfBlocksInFork(Relation relation,
     276             :                                                    ForkNumber forkNum);
     277             : extern void FlushOneBuffer(Buffer buffer);
     278             : extern void FlushRelationBuffers(Relation rel);
     279             : extern void FlushRelationsAllBuffers(SMgrRelation *smgrs, int nrels);
     280             : extern void CreateAndCopyRelationData(RelFileLocator src_rlocator,
     281             :                                       RelFileLocator dst_rlocator,
     282             :                                       bool permanent);
     283             : extern void FlushDatabaseBuffers(Oid dbid);
     284             : extern void DropRelationBuffers(SMgrRelation smgr_reln,
     285             :                                 ForkNumber *forkNum,
     286             :                                 int nforks, BlockNumber *firstDelBlock);
     287             : extern void DropRelationsAllBuffers(SMgrRelation *smgr_reln,
     288             :                                     int nlocators);
     289             : extern void DropDatabaseBuffers(Oid dbid);
     290             : 
     291             : #define RelationGetNumberOfBlocks(reln) \
     292             :     RelationGetNumberOfBlocksInFork(reln, MAIN_FORKNUM)
     293             : 
     294             : extern bool BufferIsPermanent(Buffer buffer);
     295             : extern XLogRecPtr BufferGetLSNAtomic(Buffer buffer);
     296             : extern void BufferGetTag(Buffer buffer, RelFileLocator *rlocator,
     297             :                          ForkNumber *forknum, BlockNumber *blknum);
     298             : 
     299             : extern void MarkBufferDirtyHint(Buffer buffer, bool buffer_std);
     300             : 
     301             : extern void UnlockBuffers(void);
     302             : extern void LockBuffer(Buffer buffer, int mode);
     303             : extern bool ConditionalLockBuffer(Buffer buffer);
     304             : extern void LockBufferForCleanup(Buffer buffer);
     305             : extern bool ConditionalLockBufferForCleanup(Buffer buffer);
     306             : extern bool IsBufferCleanupOK(Buffer buffer);
     307             : extern bool HoldingBufferPinThatDelaysRecovery(void);
     308             : 
     309             : extern bool BgBufferSync(WritebackContext *wb_context);
     310             : 
     311             : extern uint32 GetPinLimit(void);
     312             : extern uint32 GetLocalPinLimit(void);
     313             : extern uint32 GetAdditionalPinLimit(void);
     314             : extern uint32 GetAdditionalLocalPinLimit(void);
     315             : extern void LimitAdditionalPins(uint32 *additional_pins);
     316             : extern void LimitAdditionalLocalPins(uint32 *additional_pins);
     317             : 
     318             : extern bool EvictUnpinnedBuffer(Buffer buf, bool *buffer_flushed);
     319             : extern void EvictAllUnpinnedBuffers(int32 *buffers_evicted,
     320             :                                     int32 *buffers_flushed,
     321             :                                     int32 *buffers_skipped);
     322             : extern void EvictRelUnpinnedBuffers(Relation rel,
     323             :                                     int32 *buffers_evicted,
     324             :                                     int32 *buffers_flushed,
     325             :                                     int32 *buffers_skipped);
     326             : 
     327             : /* in buf_init.c */
     328             : extern void BufferManagerShmemInit(void);
     329             : extern Size BufferManagerShmemSize(void);
     330             : 
     331             : /* in localbuf.c */
     332             : extern void AtProcExit_LocalBuffers(void);
     333             : 
     334             : /* in freelist.c */
     335             : 
     336             : extern BufferAccessStrategy GetAccessStrategy(BufferAccessStrategyType btype);
     337             : extern BufferAccessStrategy GetAccessStrategyWithSize(BufferAccessStrategyType btype,
     338             :                                                       int ring_size_kb);
     339             : extern int  GetAccessStrategyBufferCount(BufferAccessStrategy strategy);
     340             : extern int  GetAccessStrategyPinLimit(BufferAccessStrategy strategy);
     341             : 
     342             : extern void FreeAccessStrategy(BufferAccessStrategy strategy);
     343             : 
     344             : 
     345             : /* inline functions */
     346             : 
     347             : /*
     348             :  * Although this header file is nominally backend-only, certain frontend
     349             :  * programs like pg_waldump include it.  For compilers that emit static
     350             :  * inline functions even when they're unused, that leads to unsatisfied
     351             :  * external references; hence hide these with #ifndef FRONTEND.
     352             :  */
     353             : 
     354             : #ifndef FRONTEND
     355             : 
     356             : /*
     357             :  * BufferIsValid
     358             :  *      True iff the given buffer number is valid (either as a shared
     359             :  *      or local buffer).
     360             :  *
     361             :  * Note: For a long time this was defined the same as BufferIsPinned,
     362             :  * that is it would say False if you didn't hold a pin on the buffer.
     363             :  * I believe this was bogus and served only to mask logic errors.
     364             :  * Code should always know whether it has a buffer reference,
     365             :  * independently of the pin state.
     366             :  *
     367             :  * Note: For a further long time this was not quite the inverse of the
     368             :  * BufferIsInvalid() macro, in that it also did sanity checks to verify
     369             :  * that the buffer number was in range.  Most likely, this macro was
     370             :  * originally intended only to be used in assertions, but its use has
     371             :  * since expanded quite a bit, and the overhead of making those checks
     372             :  * even in non-assert-enabled builds can be significant.  Thus, we've
     373             :  * now demoted the range checks to assertions within the macro itself.
     374             :  */
     375             : static inline bool
     376   630170646 : BufferIsValid(Buffer bufnum)
     377             : {
     378             :     Assert(bufnum <= NBuffers);
     379             :     Assert(bufnum >= -NLocBuffer);
     380             : 
     381   630170646 :     return bufnum != InvalidBuffer;
     382             : }
     383             : 
     384             : /*
     385             :  * BufferGetBlock
     386             :  *      Returns a reference to a disk page image associated with a buffer.
     387             :  *
     388             :  * Note:
     389             :  *      Assumes buffer is valid.
     390             :  */
     391             : static inline Block
     392   724180166 : BufferGetBlock(Buffer buffer)
     393             : {
     394             :     Assert(BufferIsValid(buffer));
     395             : 
     396   724180166 :     if (BufferIsLocal(buffer))
     397    23460706 :         return LocalBufferBlockPointers[-buffer - 1];
     398             :     else
     399   700719460 :         return (Block) (BufferBlocks + ((Size) (buffer - 1)) * BLCKSZ);
     400             : }
     401             : 
     402             : /*
     403             :  * BufferGetPageSize
     404             :  *      Returns the page size within a buffer.
     405             :  *
     406             :  * Notes:
     407             :  *      Assumes buffer is valid.
     408             :  *
     409             :  *      The buffer can be a raw disk block and need not contain a valid
     410             :  *      (formatted) disk page.
     411             :  */
     412             : /* XXX should dig out of buffer descriptor */
     413             : static inline Size
     414      452462 : BufferGetPageSize(Buffer buffer)
     415             : {
     416             :     Assert(BufferIsValid(buffer));
     417      452462 :     return (Size) BLCKSZ;
     418             : }
     419             : 
     420             : /*
     421             :  * BufferGetPage
     422             :  *      Returns the page associated with a buffer.
     423             :  */
     424             : static inline Page
     425   718711572 : BufferGetPage(Buffer buffer)
     426             : {
     427   718711572 :     return (Page) BufferGetBlock(buffer);
     428             : }
     429             : 
     430             : #endif                          /* FRONTEND */
     431             : 
     432             : #endif                          /* BUFMGR_H */

Generated by: LCOV version 1.16