PostgreSQL Source Code  git master
bufmgr.c File Reference
#include "postgres.h"
#include <sys/file.h>
#include <unistd.h>
#include "access/tableam.h"
#include "access/xlog.h"
#include "catalog/catalog.h"
#include "catalog/storage.h"
#include "executor/instrument.h"
#include "lib/binaryheap.h"
#include "miscadmin.h"
#include "pg_trace.h"
#include "pgstat.h"
#include "postmaster/bgwriter.h"
#include "storage/buf_internals.h"
#include "storage/bufmgr.h"
#include "storage/ipc.h"
#include "storage/proc.h"
#include "storage/smgr.h"
#include "storage/standby.h"
#include "utils/rel.h"
#include "utils/resowner_private.h"
#include "utils/timestamp.h"
Include dependency graph for bufmgr.c:

Go to the source code of this file.

Data Structures

struct  PrivateRefCountEntry
 
struct  CkptTsStatus
 

Macros

#define BufHdrGetBlock(bufHdr)   ((Block) (BufferBlocks + ((Size) (bufHdr)->buf_id) * BLCKSZ))
 
#define BufferGetLSN(bufHdr)   (PageGetLSN(BufHdrGetBlock(bufHdr)))
 
#define LocalBufHdrGetBlock(bufHdr)   LocalBufferBlockPointers[-((bufHdr)->buf_id + 2)]
 
#define BUF_WRITTEN   0x01
 
#define BUF_REUSABLE   0x02
 
#define DROP_RELS_BSEARCH_THRESHOLD   20
 
#define REFCOUNT_ARRAY_ENTRIES   8
 
#define BufferIsPinned(bufnum)
 

Typedefs

typedef struct PrivateRefCountEntry PrivateRefCountEntry
 
typedef struct CkptTsStatus CkptTsStatus
 

Functions

static void ReservePrivateRefCountEntry (void)
 
static PrivateRefCountEntryNewPrivateRefCountEntry (Buffer buffer)
 
static PrivateRefCountEntryGetPrivateRefCountEntry (Buffer buffer, bool do_move)
 
static int32 GetPrivateRefCount (Buffer buffer)
 
static void ForgetPrivateRefCountEntry (PrivateRefCountEntry *ref)
 
static Buffer ReadBuffer_common (SMgrRelation reln, char relpersistence, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy, bool *hit)
 
static bool PinBuffer (BufferDesc *buf, BufferAccessStrategy strategy)
 
static void PinBuffer_Locked (BufferDesc *buf)
 
static void UnpinBuffer (BufferDesc *buf, bool fixOwner)
 
static void BufferSync (int flags)
 
static uint32 WaitBufHdrUnlocked (BufferDesc *buf)
 
static int SyncOneBuffer (int buf_id, bool skip_recently_used, WritebackContext *wb_context)
 
static void WaitIO (BufferDesc *buf)
 
static bool StartBufferIO (BufferDesc *buf, bool forInput)
 
static void TerminateBufferIO (BufferDesc *buf, bool clear_dirty, uint32 set_flag_bits)
 
static void shared_buffer_write_error_callback (void *arg)
 
static void local_buffer_write_error_callback (void *arg)
 
static BufferDescBufferAlloc (SMgrRelation smgr, char relpersistence, ForkNumber forkNum, BlockNumber blockNum, BufferAccessStrategy strategy, bool *foundPtr)
 
static void FlushBuffer (BufferDesc *buf, SMgrRelation reln)
 
static void AtProcExit_Buffers (int code, Datum arg)
 
static void CheckForBufferLeaks (void)
 
static int rnode_comparator (const void *p1, const void *p2)
 
static int buffertag_comparator (const void *p1, const void *p2)
 
static int ckpt_buforder_comparator (const void *pa, const void *pb)
 
static int ts_ckpt_progress_comparator (Datum a, Datum b, void *arg)
 
bool ComputeIoConcurrency (int io_concurrency, double *target)
 
void PrefetchBuffer (Relation reln, ForkNumber forkNum, BlockNumber blockNum)
 
Buffer ReadBuffer (Relation reln, BlockNumber blockNum)
 
Buffer ReadBufferExtended (Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
 
Buffer ReadBufferWithoutRelcache (RelFileNode rnode, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
 
static void InvalidateBuffer (BufferDesc *buf)
 
void MarkBufferDirty (Buffer buffer)
 
Buffer ReleaseAndReadBuffer (Buffer buffer, Relation relation, BlockNumber blockNum)
 
bool BgBufferSync (WritebackContext *wb_context)
 
void AtEOXact_Buffers (bool isCommit)
 
void InitBufferPoolAccess (void)
 
void InitBufferPoolBackend (void)
 
void PrintBufferLeakWarning (Buffer buffer)
 
void CheckPointBuffers (int flags)
 
void BufmgrCommit (void)
 
BlockNumber BufferGetBlockNumber (Buffer buffer)
 
void BufferGetTag (Buffer buffer, RelFileNode *rnode, ForkNumber *forknum, BlockNumber *blknum)
 
BlockNumber RelationGetNumberOfBlocksInFork (Relation relation, ForkNumber forkNum)
 
bool BufferIsPermanent (Buffer buffer)
 
XLogRecPtr BufferGetLSNAtomic (Buffer buffer)
 
void DropRelFileNodeBuffers (RelFileNodeBackend rnode, ForkNumber *forkNum, int nforks, BlockNumber *firstDelBlock)
 
void DropRelFileNodesAllBuffers (RelFileNodeBackend *rnodes, int nnodes)
 
void DropDatabaseBuffers (Oid dbid)
 
void FlushRelationBuffers (Relation rel)
 
void FlushDatabaseBuffers (Oid dbid)
 
void FlushOneBuffer (Buffer buffer)
 
void ReleaseBuffer (Buffer buffer)
 
void UnlockReleaseBuffer (Buffer buffer)
 
void IncrBufferRefCount (Buffer buffer)
 
void MarkBufferDirtyHint (Buffer buffer, bool buffer_std)
 
void UnlockBuffers (void)
 
void LockBuffer (Buffer buffer, int mode)
 
bool ConditionalLockBuffer (Buffer buffer)
 
void LockBufferForCleanup (Buffer buffer)
 
bool HoldingBufferPinThatDelaysRecovery (void)
 
bool ConditionalLockBufferForCleanup (Buffer buffer)
 
bool IsBufferCleanupOK (Buffer buffer)
 
void AbortBufferIO (void)
 
uint32 LockBufHdr (BufferDesc *desc)
 
void WritebackContextInit (WritebackContext *context, int *max_pending)
 
void ScheduleBufferTagForWriteback (WritebackContext *context, BufferTag *tag)
 
void IssuePendingWritebacks (WritebackContext *context)
 
void TestForOldSnapshot_impl (Snapshot snapshot, Relation relation)
 

Variables

bool zero_damaged_pages = false
 
int bgwriter_lru_maxpages = 100
 
double bgwriter_lru_multiplier = 2.0
 
bool track_io_timing = false
 
int effective_io_concurrency = 0
 
int checkpoint_flush_after = 0
 
int bgwriter_flush_after = 0
 
int backend_flush_after = 0
 
int target_prefetch_pages = 0
 
static BufferDescInProgressBuf = NULL
 
static bool IsForInput
 
static BufferDescPinCountWaitBuf = NULL
 
static struct PrivateRefCountEntry PrivateRefCountArray [REFCOUNT_ARRAY_ENTRIES]
 
static HTABPrivateRefCountHash = NULL
 
static int32 PrivateRefCountOverflowed = 0
 
static uint32 PrivateRefCountClock = 0
 
static PrivateRefCountEntryReservedRefCountEntry = NULL
 

Macro Definition Documentation

◆ BUF_REUSABLE

#define BUF_REUSABLE   0x02

Definition at line 67 of file bufmgr.c.

Referenced by BgBufferSync(), and SyncOneBuffer().

◆ BUF_WRITTEN

#define BUF_WRITTEN   0x01

Definition at line 66 of file bufmgr.c.

Referenced by BgBufferSync(), BufferSync(), and SyncOneBuffer().

◆ BufferGetLSN

#define BufferGetLSN (   bufHdr)    (PageGetLSN(BufHdrGetBlock(bufHdr)))

Definition at line 59 of file bufmgr.c.

Referenced by BufferAlloc(), and FlushBuffer().

◆ BufferIsPinned

#define BufferIsPinned (   bufnum)
Value:
( \
!BufferIsValid(bufnum) ? \
false \
: \
BufferIsLocal(bufnum) ? \
(LocalRefCount[-(bufnum) - 1] > 0) \
: \
(GetPrivateRefCount(bufnum) > 0) \
)
static int32 GetPrivateRefCount(Buffer buffer)
Definition: bufmgr.c:359
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
int32 * LocalRefCount
Definition: localbuf.c:45

Definition at line 420 of file bufmgr.c.

Referenced by BufferGetBlockNumber(), BufferGetLSNAtomic(), BufferGetTag(), BufferIsPermanent(), FlushOneBuffer(), IncrBufferRefCount(), MarkBufferDirty(), and ReleaseAndReadBuffer().

◆ BufHdrGetBlock

#define BufHdrGetBlock (   bufHdr)    ((Block) (BufferBlocks + ((Size) (bufHdr)->buf_id) * BLCKSZ))

Definition at line 58 of file bufmgr.c.

Referenced by FlushBuffer(), and ReadBuffer_common().

◆ DROP_RELS_BSEARCH_THRESHOLD

#define DROP_RELS_BSEARCH_THRESHOLD   20

Definition at line 69 of file bufmgr.c.

Referenced by DropRelFileNodesAllBuffers().

◆ LocalBufHdrGetBlock

#define LocalBufHdrGetBlock (   bufHdr)    LocalBufferBlockPointers[-((bufHdr)->buf_id + 2)]

Definition at line 62 of file bufmgr.c.

Referenced by FlushRelationBuffers(), and ReadBuffer_common().

◆ REFCOUNT_ARRAY_ENTRIES

#define REFCOUNT_ARRAY_ENTRIES   8

Typedef Documentation

◆ CkptTsStatus

typedef struct CkptTsStatus CkptTsStatus

◆ PrivateRefCountEntry

Function Documentation

◆ AbortBufferIO()

void AbortBufferIO ( void  )

Definition at line 4040 of file bufmgr.c.

References Assert, buftag::blockNum, BM_DIRTY, BM_IO_ERROR, BM_IO_IN_PROGRESS, BM_VALID, buf, BufferDescriptorGetIOLock, ereport, errcode(), errdetail(), errmsg(), buftag::forkNum, InProgressBuf, IsForInput, LockBufHdr(), LW_EXCLUSIVE, LWLockAcquire(), pfree(), relpathperm, buftag::rnode, BufferDesc::tag, TerminateBufferIO(), UnlockBufHdr, and WARNING.

Referenced by AbortSubTransaction(), AbortTransaction(), AtProcExit_Buffers(), AutoVacLauncherMain(), BackgroundWriterMain(), CheckpointerMain(), and WalWriterMain().

4041 {
4043 
4044  if (buf)
4045  {
4046  uint32 buf_state;
4047 
4048  /*
4049  * Since LWLockReleaseAll has already been called, we're not holding
4050  * the buffer's io_in_progress_lock. We have to re-acquire it so that
4051  * we can use TerminateBufferIO. Anyone who's executing WaitIO on the
4052  * buffer will be in a busy spin until we succeed in doing this.
4053  */
4055 
4056  buf_state = LockBufHdr(buf);
4057  Assert(buf_state & BM_IO_IN_PROGRESS);
4058  if (IsForInput)
4059  {
4060  Assert(!(buf_state & BM_DIRTY));
4061 
4062  /* We'd better not think buffer is valid yet */
4063  Assert(!(buf_state & BM_VALID));
4064  UnlockBufHdr(buf, buf_state);
4065  }
4066  else
4067  {
4068  Assert(buf_state & BM_DIRTY);
4069  UnlockBufHdr(buf, buf_state);
4070  /* Issue notice if this is not the first failure... */
4071  if (buf_state & BM_IO_ERROR)
4072  {
4073  /* Buffer is pinned, so we can read tag without spinlock */
4074  char *path;
4075 
4076  path = relpathperm(buf->tag.rnode, buf->tag.forkNum);
4077  ereport(WARNING,
4078  (errcode(ERRCODE_IO_ERROR),
4079  errmsg("could not write block %u of %s",
4080  buf->tag.blockNum, path),
4081  errdetail("Multiple failures --- write error might be permanent.")));
4082  pfree(path);
4083  }
4084  }
4085  TerminateBufferIO(buf, false, BM_IO_ERROR);
4086  }
4087 }
#define relpathperm(rnode, forknum)
Definition: relpath.h:83
ForkNumber forkNum
Definition: buf_internals.h:93
int errcode(int sqlerrcode)
Definition: elog.c:608
#define BM_DIRTY
Definition: buf_internals.h:58
#define BufferDescriptorGetIOLock(bdesc)
static BufferDesc * InProgressBuf
Definition: bufmgr.c:133
void pfree(void *pointer)
Definition: mcxt.c:1056
static char * buf
Definition: pg_test_fsync.c:67
int errdetail(const char *fmt,...)
Definition: elog.c:955
unsigned int uint32
Definition: c.h:359
static bool IsForInput
Definition: bufmgr.c:134
#define ereport(elevel, rest)
Definition: elog.h:141
#define WARNING
Definition: elog.h:40
static void TerminateBufferIO(BufferDesc *buf, bool clear_dirty, uint32 set_flag_bits)
Definition: bufmgr.c:4008
#define BM_VALID
Definition: buf_internals.h:59
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
#define Assert(condition)
Definition: c.h:739
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
BlockNumber blockNum
Definition: buf_internals.h:94
RelFileNode rnode
Definition: buf_internals.h:92
#define BM_IO_ERROR
Definition: buf_internals.h:62
BufferTag tag
int errmsg(const char *fmt,...)
Definition: elog.c:822
#define UnlockBufHdr(desc, s)
#define BM_IO_IN_PROGRESS
Definition: buf_internals.h:61

◆ AtEOXact_Buffers()

void AtEOXact_Buffers ( bool  isCommit)

Definition at line 2433 of file bufmgr.c.

References Assert, AtEOXact_LocalBuffers(), CheckForBufferLeaks(), and PrivateRefCountOverflowed.

Referenced by AbortTransaction(), AutoVacLauncherMain(), BackgroundWriterMain(), CheckpointerMain(), CommitTransaction(), PrepareTransaction(), and WalWriterMain().

2434 {
2436 
2437  AtEOXact_LocalBuffers(isCommit);
2438 
2440 }
static int32 PrivateRefCountOverflowed
Definition: bufmgr.c:171
#define Assert(condition)
Definition: c.h:739
static void CheckForBufferLeaks(void)
Definition: bufmgr.c:2508
void AtEOXact_LocalBuffers(bool isCommit)
Definition: localbuf.c:572

◆ AtProcExit_Buffers()

static void AtProcExit_Buffers ( int  code,
Datum  arg 
)
static

Definition at line 2489 of file bufmgr.c.

References AbortBufferIO(), AtProcExit_LocalBuffers(), CheckForBufferLeaks(), and UnlockBuffers().

Referenced by InitBufferPoolBackend().

2490 {
2491  AbortBufferIO();
2492  UnlockBuffers();
2493 
2495 
2496  /* localbuf.c needs a chance too */
2498 }
void AtProcExit_LocalBuffers(void)
Definition: localbuf.c:583
void UnlockBuffers(void)
Definition: bufmgr.c:3584
static void CheckForBufferLeaks(void)
Definition: bufmgr.c:2508
void AbortBufferIO(void)
Definition: bufmgr.c:4040

◆ BgBufferSync()

bool BgBufferSync ( WritebackContext wb_context)

Definition at line 2063 of file bufmgr.c.

References Assert, bgwriter_lru_maxpages, bgwriter_lru_multiplier, BgWriterDelay, BgWriterStats, BUF_REUSABLE, BUF_WRITTEN, CurrentResourceOwner, DEBUG1, DEBUG2, elog, PgStat_MsgBgWriter::m_buf_alloc, PgStat_MsgBgWriter::m_buf_written_clean, PgStat_MsgBgWriter::m_maxwritten_clean, NBuffers, ResourceOwnerEnlargeBuffers(), StrategySyncStart(), and SyncOneBuffer().

Referenced by BackgroundWriterMain().

2064 {
2065  /* info obtained from freelist.c */
2066  int strategy_buf_id;
2067  uint32 strategy_passes;
2068  uint32 recent_alloc;
2069 
2070  /*
2071  * Information saved between calls so we can determine the strategy
2072  * point's advance rate and avoid scanning already-cleaned buffers.
2073  */
2074  static bool saved_info_valid = false;
2075  static int prev_strategy_buf_id;
2076  static uint32 prev_strategy_passes;
2077  static int next_to_clean;
2078  static uint32 next_passes;
2079 
2080  /* Moving averages of allocation rate and clean-buffer density */
2081  static float smoothed_alloc = 0;
2082  static float smoothed_density = 10.0;
2083 
2084  /* Potentially these could be tunables, but for now, not */
2085  float smoothing_samples = 16;
2086  float scan_whole_pool_milliseconds = 120000.0;
2087 
2088  /* Used to compute how far we scan ahead */
2089  long strategy_delta;
2090  int bufs_to_lap;
2091  int bufs_ahead;
2092  float scans_per_alloc;
2093  int reusable_buffers_est;
2094  int upcoming_alloc_est;
2095  int min_scan_buffers;
2096 
2097  /* Variables for the scanning loop proper */
2098  int num_to_scan;
2099  int num_written;
2100  int reusable_buffers;
2101 
2102  /* Variables for final smoothed_density update */
2103  long new_strategy_delta;
2104  uint32 new_recent_alloc;
2105 
2106  /*
2107  * Find out where the freelist clock sweep currently is, and how many
2108  * buffer allocations have happened since our last call.
2109  */
2110  strategy_buf_id = StrategySyncStart(&strategy_passes, &recent_alloc);
2111 
2112  /* Report buffer alloc counts to pgstat */
2113  BgWriterStats.m_buf_alloc += recent_alloc;
2114 
2115  /*
2116  * If we're not running the LRU scan, just stop after doing the stats
2117  * stuff. We mark the saved state invalid so that we can recover sanely
2118  * if LRU scan is turned back on later.
2119  */
2120  if (bgwriter_lru_maxpages <= 0)
2121  {
2122  saved_info_valid = false;
2123  return true;
2124  }
2125 
2126  /*
2127  * Compute strategy_delta = how many buffers have been scanned by the
2128  * clock sweep since last time. If first time through, assume none. Then
2129  * see if we are still ahead of the clock sweep, and if so, how many
2130  * buffers we could scan before we'd catch up with it and "lap" it. Note:
2131  * weird-looking coding of xxx_passes comparisons are to avoid bogus
2132  * behavior when the passes counts wrap around.
2133  */
2134  if (saved_info_valid)
2135  {
2136  int32 passes_delta = strategy_passes - prev_strategy_passes;
2137 
2138  strategy_delta = strategy_buf_id - prev_strategy_buf_id;
2139  strategy_delta += (long) passes_delta * NBuffers;
2140 
2141  Assert(strategy_delta >= 0);
2142 
2143  if ((int32) (next_passes - strategy_passes) > 0)
2144  {
2145  /* we're one pass ahead of the strategy point */
2146  bufs_to_lap = strategy_buf_id - next_to_clean;
2147 #ifdef BGW_DEBUG
2148  elog(DEBUG2, "bgwriter ahead: bgw %u-%u strategy %u-%u delta=%ld lap=%d",
2149  next_passes, next_to_clean,
2150  strategy_passes, strategy_buf_id,
2151  strategy_delta, bufs_to_lap);
2152 #endif
2153  }
2154  else if (next_passes == strategy_passes &&
2155  next_to_clean >= strategy_buf_id)
2156  {
2157  /* on same pass, but ahead or at least not behind */
2158  bufs_to_lap = NBuffers - (next_to_clean - strategy_buf_id);
2159 #ifdef BGW_DEBUG
2160  elog(DEBUG2, "bgwriter ahead: bgw %u-%u strategy %u-%u delta=%ld lap=%d",
2161  next_passes, next_to_clean,
2162  strategy_passes, strategy_buf_id,
2163  strategy_delta, bufs_to_lap);
2164 #endif
2165  }
2166  else
2167  {
2168  /*
2169  * We're behind, so skip forward to the strategy point and start
2170  * cleaning from there.
2171  */
2172 #ifdef BGW_DEBUG
2173  elog(DEBUG2, "bgwriter behind: bgw %u-%u strategy %u-%u delta=%ld",
2174  next_passes, next_to_clean,
2175  strategy_passes, strategy_buf_id,
2176  strategy_delta);
2177 #endif
2178  next_to_clean = strategy_buf_id;
2179  next_passes = strategy_passes;
2180  bufs_to_lap = NBuffers;
2181  }
2182  }
2183  else
2184  {
2185  /*
2186  * Initializing at startup or after LRU scanning had been off. Always
2187  * start at the strategy point.
2188  */
2189 #ifdef BGW_DEBUG
2190  elog(DEBUG2, "bgwriter initializing: strategy %u-%u",
2191  strategy_passes, strategy_buf_id);
2192 #endif
2193  strategy_delta = 0;
2194  next_to_clean = strategy_buf_id;
2195  next_passes = strategy_passes;
2196  bufs_to_lap = NBuffers;
2197  }
2198 
2199  /* Update saved info for next time */
2200  prev_strategy_buf_id = strategy_buf_id;
2201  prev_strategy_passes = strategy_passes;
2202  saved_info_valid = true;
2203 
2204  /*
2205  * Compute how many buffers had to be scanned for each new allocation, ie,
2206  * 1/density of reusable buffers, and track a moving average of that.
2207  *
2208  * If the strategy point didn't move, we don't update the density estimate
2209  */
2210  if (strategy_delta > 0 && recent_alloc > 0)
2211  {
2212  scans_per_alloc = (float) strategy_delta / (float) recent_alloc;
2213  smoothed_density += (scans_per_alloc - smoothed_density) /
2214  smoothing_samples;
2215  }
2216 
2217  /*
2218  * Estimate how many reusable buffers there are between the current
2219  * strategy point and where we've scanned ahead to, based on the smoothed
2220  * density estimate.
2221  */
2222  bufs_ahead = NBuffers - bufs_to_lap;
2223  reusable_buffers_est = (float) bufs_ahead / smoothed_density;
2224 
2225  /*
2226  * Track a moving average of recent buffer allocations. Here, rather than
2227  * a true average we want a fast-attack, slow-decline behavior: we
2228  * immediately follow any increase.
2229  */
2230  if (smoothed_alloc <= (float) recent_alloc)
2231  smoothed_alloc = recent_alloc;
2232  else
2233  smoothed_alloc += ((float) recent_alloc - smoothed_alloc) /
2234  smoothing_samples;
2235 
2236  /* Scale the estimate by a GUC to allow more aggressive tuning. */
2237  upcoming_alloc_est = (int) (smoothed_alloc * bgwriter_lru_multiplier);
2238 
2239  /*
2240  * If recent_alloc remains at zero for many cycles, smoothed_alloc will
2241  * eventually underflow to zero, and the underflows produce annoying
2242  * kernel warnings on some platforms. Once upcoming_alloc_est has gone to
2243  * zero, there's no point in tracking smaller and smaller values of
2244  * smoothed_alloc, so just reset it to exactly zero to avoid this
2245  * syndrome. It will pop back up as soon as recent_alloc increases.
2246  */
2247  if (upcoming_alloc_est == 0)
2248  smoothed_alloc = 0;
2249 
2250  /*
2251  * Even in cases where there's been little or no buffer allocation
2252  * activity, we want to make a small amount of progress through the buffer
2253  * cache so that as many reusable buffers as possible are clean after an
2254  * idle period.
2255  *
2256  * (scan_whole_pool_milliseconds / BgWriterDelay) computes how many times
2257  * the BGW will be called during the scan_whole_pool time; slice the
2258  * buffer pool into that many sections.
2259  */
2260  min_scan_buffers = (int) (NBuffers / (scan_whole_pool_milliseconds / BgWriterDelay));
2261 
2262  if (upcoming_alloc_est < (min_scan_buffers + reusable_buffers_est))
2263  {
2264 #ifdef BGW_DEBUG
2265  elog(DEBUG2, "bgwriter: alloc_est=%d too small, using min=%d + reusable_est=%d",
2266  upcoming_alloc_est, min_scan_buffers, reusable_buffers_est);
2267 #endif
2268  upcoming_alloc_est = min_scan_buffers + reusable_buffers_est;
2269  }
2270 
2271  /*
2272  * Now write out dirty reusable buffers, working forward from the
2273  * next_to_clean point, until we have lapped the strategy scan, or cleaned
2274  * enough buffers to match our estimate of the next cycle's allocation
2275  * requirements, or hit the bgwriter_lru_maxpages limit.
2276  */
2277 
2278  /* Make sure we can handle the pin inside SyncOneBuffer */
2280 
2281  num_to_scan = bufs_to_lap;
2282  num_written = 0;
2283  reusable_buffers = reusable_buffers_est;
2284 
2285  /* Execute the LRU scan */
2286  while (num_to_scan > 0 && reusable_buffers < upcoming_alloc_est)
2287  {
2288  int sync_state = SyncOneBuffer(next_to_clean, true,
2289  wb_context);
2290 
2291  if (++next_to_clean >= NBuffers)
2292  {
2293  next_to_clean = 0;
2294  next_passes++;
2295  }
2296  num_to_scan--;
2297 
2298  if (sync_state & BUF_WRITTEN)
2299  {
2300  reusable_buffers++;
2301  if (++num_written >= bgwriter_lru_maxpages)
2302  {
2304  break;
2305  }
2306  }
2307  else if (sync_state & BUF_REUSABLE)
2308  reusable_buffers++;
2309  }
2310 
2311  BgWriterStats.m_buf_written_clean += num_written;
2312 
2313 #ifdef BGW_DEBUG
2314  elog(DEBUG1, "bgwriter: recent_alloc=%u smoothed=%.2f delta=%ld ahead=%d density=%.2f reusable_est=%d upcoming_est=%d scanned=%d wrote=%d reusable=%d",
2315  recent_alloc, smoothed_alloc, strategy_delta, bufs_ahead,
2316  smoothed_density, reusable_buffers_est, upcoming_alloc_est,
2317  bufs_to_lap - num_to_scan,
2318  num_written,
2319  reusable_buffers - reusable_buffers_est);
2320 #endif
2321 
2322  /*
2323  * Consider the above scan as being like a new allocation scan.
2324  * Characterize its density and update the smoothed one based on it. This
2325  * effectively halves the moving average period in cases where both the
2326  * strategy and the background writer are doing some useful scanning,
2327  * which is helpful because a long memory isn't as desirable on the
2328  * density estimates.
2329  */
2330  new_strategy_delta = bufs_to_lap - num_to_scan;
2331  new_recent_alloc = reusable_buffers - reusable_buffers_est;
2332  if (new_strategy_delta > 0 && new_recent_alloc > 0)
2333  {
2334  scans_per_alloc = (float) new_strategy_delta / (float) new_recent_alloc;
2335  smoothed_density += (scans_per_alloc - smoothed_density) /
2336  smoothing_samples;
2337 
2338 #ifdef BGW_DEBUG
2339  elog(DEBUG2, "bgwriter: cleaner density alloc=%u scan=%ld density=%.2f new smoothed=%.2f",
2340  new_recent_alloc, new_strategy_delta,
2341  scans_per_alloc, smoothed_density);
2342 #endif
2343  }
2344 
2345  /* Return true if OK to hibernate */
2346  return (bufs_to_lap == 0 && recent_alloc == 0);
2347 }
PgStat_Counter m_buf_alloc
Definition: pgstat.h:420
int StrategySyncStart(uint32 *complete_passes, uint32 *num_buf_alloc)
Definition: freelist.c:395
#define DEBUG1
Definition: elog.h:25
int BgWriterDelay
Definition: bgwriter.c:64
ResourceOwner CurrentResourceOwner
Definition: resowner.c:142
PgStat_Counter m_maxwritten_clean
Definition: pgstat.h:417
PgStat_Counter m_buf_written_clean
Definition: pgstat.h:416
PgStat_MsgBgWriter BgWriterStats
Definition: pgstat.c:142
double bgwriter_lru_multiplier
Definition: bufmgr.c:111
static int SyncOneBuffer(int buf_id, bool skip_recently_used, WritebackContext *wb_context)
Definition: bufmgr.c:2366
signed int int32
Definition: c.h:347
#define BUF_REUSABLE
Definition: bufmgr.c:67
int bgwriter_lru_maxpages
Definition: bufmgr.c:110
#define DEBUG2
Definition: elog.h:24
unsigned int uint32
Definition: c.h:359
#define BUF_WRITTEN
Definition: bufmgr.c:66
void ResourceOwnerEnlargeBuffers(ResourceOwner owner)
Definition: resowner.c:893
#define Assert(condition)
Definition: c.h:739
#define elog(elevel,...)
Definition: elog.h:228
int NBuffers
Definition: globals.c:131

◆ BufferAlloc()

static BufferDesc * BufferAlloc ( SMgrRelation  smgr,
char  relpersistence,
ForkNumber  forkNum,
BlockNumber  blockNum,
BufferAccessStrategy  strategy,
bool foundPtr 
)
static

Definition at line 995 of file bufmgr.c.

References Assert, BackendWritebackContext, BM_CHECKPOINT_NEEDED, BM_DIRTY, BM_IO_ERROR, BM_JUST_DIRTIED, BM_PERMANENT, BM_TAG_VALID, BM_VALID, buf, BUF_FLAG_MASK, BufferDesc::buf_id, BUF_STATE_GET_REFCOUNT, BUF_USAGECOUNT_MASK, BUF_USAGECOUNT_ONE, BufferDescriptorGetContentLock, BufferGetLSN, BufMappingPartitionLock, BufTableDelete(), BufTableHashCode(), BufTableInsert(), BufTableLookup(), RelFileNode::dbNode, FlushBuffer(), GetBufferDescriptor, INIT_BUFFERTAG, INIT_FORKNUM, LockBufHdr(), LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockConditionalAcquire(), LWLockRelease(), RelFileNodeBackend::node, PinBuffer(), PinBuffer_Locked(), RelFileNode::relNode, ReservePrivateRefCountEntry(), ScheduleBufferTagForWriteback(), SMgrRelationData::smgr_rnode, RelFileNode::spcNode, StartBufferIO(), StrategyGetBuffer(), StrategyRejectBuffer(), BufferDesc::tag, UnlockBufHdr, UnpinBuffer(), and XLogNeedsFlush().

Referenced by ReadBuffer_common().

999 {
1000  BufferTag newTag; /* identity of requested block */
1001  uint32 newHash; /* hash value for newTag */
1002  LWLock *newPartitionLock; /* buffer partition lock for it */
1003  BufferTag oldTag; /* previous identity of selected buffer */
1004  uint32 oldHash; /* hash value for oldTag */
1005  LWLock *oldPartitionLock; /* buffer partition lock for it */
1006  uint32 oldFlags;
1007  int buf_id;
1008  BufferDesc *buf;
1009  bool valid;
1010  uint32 buf_state;
1011 
1012  /* create a tag so we can lookup the buffer */
1013  INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
1014 
1015  /* determine its hash code and partition lock ID */
1016  newHash = BufTableHashCode(&newTag);
1017  newPartitionLock = BufMappingPartitionLock(newHash);
1018 
1019  /* see if the block is in the buffer pool already */
1020  LWLockAcquire(newPartitionLock, LW_SHARED);
1021  buf_id = BufTableLookup(&newTag, newHash);
1022  if (buf_id >= 0)
1023  {
1024  /*
1025  * Found it. Now, pin the buffer so no one can steal it from the
1026  * buffer pool, and check to see if the correct data has been loaded
1027  * into the buffer.
1028  */
1029  buf = GetBufferDescriptor(buf_id);
1030 
1031  valid = PinBuffer(buf, strategy);
1032 
1033  /* Can release the mapping lock as soon as we've pinned it */
1034  LWLockRelease(newPartitionLock);
1035 
1036  *foundPtr = true;
1037 
1038  if (!valid)
1039  {
1040  /*
1041  * We can only get here if (a) someone else is still reading in
1042  * the page, or (b) a previous read attempt failed. We have to
1043  * wait for any active read attempt to finish, and then set up our
1044  * own read attempt if the page is still not BM_VALID.
1045  * StartBufferIO does it all.
1046  */
1047  if (StartBufferIO(buf, true))
1048  {
1049  /*
1050  * If we get here, previous attempts to read the buffer must
1051  * have failed ... but we shall bravely try again.
1052  */
1053  *foundPtr = false;
1054  }
1055  }
1056 
1057  return buf;
1058  }
1059 
1060  /*
1061  * Didn't find it in the buffer pool. We'll have to initialize a new
1062  * buffer. Remember to unlock the mapping lock while doing the work.
1063  */
1064  LWLockRelease(newPartitionLock);
1065 
1066  /* Loop here in case we have to try another victim buffer */
1067  for (;;)
1068  {
1069  /*
1070  * Ensure, while the spinlock's not yet held, that there's a free
1071  * refcount entry.
1072  */
1074 
1075  /*
1076  * Select a victim buffer. The buffer is returned with its header
1077  * spinlock still held!
1078  */
1079  buf = StrategyGetBuffer(strategy, &buf_state);
1080 
1081  Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
1082 
1083  /* Must copy buffer flags while we still hold the spinlock */
1084  oldFlags = buf_state & BUF_FLAG_MASK;
1085 
1086  /* Pin the buffer and then release the buffer spinlock */
1087  PinBuffer_Locked(buf);
1088 
1089  /*
1090  * If the buffer was dirty, try to write it out. There is a race
1091  * condition here, in that someone might dirty it after we released it
1092  * above, or even while we are writing it out (since our share-lock
1093  * won't prevent hint-bit updates). We will recheck the dirty bit
1094  * after re-locking the buffer header.
1095  */
1096  if (oldFlags & BM_DIRTY)
1097  {
1098  /*
1099  * We need a share-lock on the buffer contents to write it out
1100  * (else we might write invalid data, eg because someone else is
1101  * compacting the page contents while we write). We must use a
1102  * conditional lock acquisition here to avoid deadlock. Even
1103  * though the buffer was not pinned (and therefore surely not
1104  * locked) when StrategyGetBuffer returned it, someone else could
1105  * have pinned and exclusive-locked it by the time we get here. If
1106  * we try to get the lock unconditionally, we'd block waiting for
1107  * them; if they later block waiting for us, deadlock ensues.
1108  * (This has been observed to happen when two backends are both
1109  * trying to split btree index pages, and the second one just
1110  * happens to be trying to split the page the first one got from
1111  * StrategyGetBuffer.)
1112  */
1114  LW_SHARED))
1115  {
1116  /*
1117  * If using a nondefault strategy, and writing the buffer
1118  * would require a WAL flush, let the strategy decide whether
1119  * to go ahead and write/reuse the buffer or to choose another
1120  * victim. We need lock to inspect the page LSN, so this
1121  * can't be done inside StrategyGetBuffer.
1122  */
1123  if (strategy != NULL)
1124  {
1125  XLogRecPtr lsn;
1126 
1127  /* Read the LSN while holding buffer header lock */
1128  buf_state = LockBufHdr(buf);
1129  lsn = BufferGetLSN(buf);
1130  UnlockBufHdr(buf, buf_state);
1131 
1132  if (XLogNeedsFlush(lsn) &&
1133  StrategyRejectBuffer(strategy, buf))
1134  {
1135  /* Drop lock/pin and loop around for another buffer */
1137  UnpinBuffer(buf, true);
1138  continue;
1139  }
1140  }
1141 
1142  /* OK, do the I/O */
1143  TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
1144  smgr->smgr_rnode.node.spcNode,
1145  smgr->smgr_rnode.node.dbNode,
1146  smgr->smgr_rnode.node.relNode);
1147 
1148  FlushBuffer(buf, NULL);
1150 
1152  &buf->tag);
1153 
1154  TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
1155  smgr->smgr_rnode.node.spcNode,
1156  smgr->smgr_rnode.node.dbNode,
1157  smgr->smgr_rnode.node.relNode);
1158  }
1159  else
1160  {
1161  /*
1162  * Someone else has locked the buffer, so give it up and loop
1163  * back to get another one.
1164  */
1165  UnpinBuffer(buf, true);
1166  continue;
1167  }
1168  }
1169 
1170  /*
1171  * To change the association of a valid buffer, we'll need to have
1172  * exclusive lock on both the old and new mapping partitions.
1173  */
1174  if (oldFlags & BM_TAG_VALID)
1175  {
1176  /*
1177  * Need to compute the old tag's hashcode and partition lock ID.
1178  * XXX is it worth storing the hashcode in BufferDesc so we need
1179  * not recompute it here? Probably not.
1180  */
1181  oldTag = buf->tag;
1182  oldHash = BufTableHashCode(&oldTag);
1183  oldPartitionLock = BufMappingPartitionLock(oldHash);
1184 
1185  /*
1186  * Must lock the lower-numbered partition first to avoid
1187  * deadlocks.
1188  */
1189  if (oldPartitionLock < newPartitionLock)
1190  {
1191  LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
1192  LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
1193  }
1194  else if (oldPartitionLock > newPartitionLock)
1195  {
1196  LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
1197  LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
1198  }
1199  else
1200  {
1201  /* only one partition, only one lock */
1202  LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
1203  }
1204  }
1205  else
1206  {
1207  /* if it wasn't valid, we need only the new partition */
1208  LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
1209  /* remember we have no old-partition lock or tag */
1210  oldPartitionLock = NULL;
1211  /* keep the compiler quiet about uninitialized variables */
1212  oldHash = 0;
1213  }
1214 
1215  /*
1216  * Try to make a hashtable entry for the buffer under its new tag.
1217  * This could fail because while we were writing someone else
1218  * allocated another buffer for the same block we want to read in.
1219  * Note that we have not yet removed the hashtable entry for the old
1220  * tag.
1221  */
1222  buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
1223 
1224  if (buf_id >= 0)
1225  {
1226  /*
1227  * Got a collision. Someone has already done what we were about to
1228  * do. We'll just handle this as if it were found in the buffer
1229  * pool in the first place. First, give up the buffer we were
1230  * planning to use.
1231  */
1232  UnpinBuffer(buf, true);
1233 
1234  /* Can give up that buffer's mapping partition lock now */
1235  if (oldPartitionLock != NULL &&
1236  oldPartitionLock != newPartitionLock)
1237  LWLockRelease(oldPartitionLock);
1238 
1239  /* remaining code should match code at top of routine */
1240 
1241  buf = GetBufferDescriptor(buf_id);
1242 
1243  valid = PinBuffer(buf, strategy);
1244 
1245  /* Can release the mapping lock as soon as we've pinned it */
1246  LWLockRelease(newPartitionLock);
1247 
1248  *foundPtr = true;
1249 
1250  if (!valid)
1251  {
1252  /*
1253  * We can only get here if (a) someone else is still reading
1254  * in the page, or (b) a previous read attempt failed. We
1255  * have to wait for any active read attempt to finish, and
1256  * then set up our own read attempt if the page is still not
1257  * BM_VALID. StartBufferIO does it all.
1258  */
1259  if (StartBufferIO(buf, true))
1260  {
1261  /*
1262  * If we get here, previous attempts to read the buffer
1263  * must have failed ... but we shall bravely try again.
1264  */
1265  *foundPtr = false;
1266  }
1267  }
1268 
1269  return buf;
1270  }
1271 
1272  /*
1273  * Need to lock the buffer header too in order to change its tag.
1274  */
1275  buf_state = LockBufHdr(buf);
1276 
1277  /*
1278  * Somebody could have pinned or re-dirtied the buffer while we were
1279  * doing the I/O and making the new hashtable entry. If so, we can't
1280  * recycle this buffer; we must undo everything we've done and start
1281  * over with a new victim buffer.
1282  */
1283  oldFlags = buf_state & BUF_FLAG_MASK;
1284  if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
1285  break;
1286 
1287  UnlockBufHdr(buf, buf_state);
1288  BufTableDelete(&newTag, newHash);
1289  if (oldPartitionLock != NULL &&
1290  oldPartitionLock != newPartitionLock)
1291  LWLockRelease(oldPartitionLock);
1292  LWLockRelease(newPartitionLock);
1293  UnpinBuffer(buf, true);
1294  }
1295 
1296  /*
1297  * Okay, it's finally safe to rename the buffer.
1298  *
1299  * Clearing BM_VALID here is necessary, clearing the dirtybits is just
1300  * paranoia. We also reset the usage_count since any recency of use of
1301  * the old content is no longer relevant. (The usage_count starts out at
1302  * 1 so that the buffer can survive one clock-sweep pass.)
1303  *
1304  * Make sure BM_PERMANENT is set for buffers that must be written at every
1305  * checkpoint. Unlogged buffers only need to be written at shutdown
1306  * checkpoints, except for their "init" forks, which need to be treated
1307  * just like permanent relations.
1308  */
1309  buf->tag = newTag;
1310  buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
1313  if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
1314  buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
1315  else
1316  buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
1317 
1318  UnlockBufHdr(buf, buf_state);
1319 
1320  if (oldPartitionLock != NULL)
1321  {
1322  BufTableDelete(&oldTag, oldHash);
1323  if (oldPartitionLock != newPartitionLock)
1324  LWLockRelease(oldPartitionLock);
1325  }
1326 
1327  LWLockRelease(newPartitionLock);
1328 
1329  /*
1330  * Buffer contents are currently invalid. Try to get the io_in_progress
1331  * lock. If StartBufferIO returns false, then someone else managed to
1332  * read it before we did, so there's nothing left for BufferAlloc() to do.
1333  */
1334  if (StartBufferIO(buf, true))
1335  *foundPtr = false;
1336  else
1337  *foundPtr = true;
1338 
1339  return buf;
1340 }
static bool PinBuffer(BufferDesc *buf, BufferAccessStrategy strategy)
Definition: bufmgr.c:1579
BufferDesc * StrategyGetBuffer(BufferAccessStrategy strategy, uint32 *buf_state)
Definition: freelist.c:201
void BufTableDelete(BufferTag *tagPtr, uint32 hashcode)
Definition: buf_table.c:149
Definition: lwlock.h:32
#define BM_PERMANENT
Definition: buf_internals.h:66
#define BufMappingPartitionLock(hashcode)
#define BM_TAG_VALID
Definition: buf_internals.h:60
bool XLogNeedsFlush(XLogRecPtr record)
Definition: xlog.c:3125
#define BM_CHECKPOINT_NEEDED
Definition: buf_internals.h:65
uint32 BufTableHashCode(BufferTag *tagPtr)
Definition: buf_table.c:79
int BufTableLookup(BufferTag *tagPtr, uint32 hashcode)
Definition: buf_table.c:91
#define BM_DIRTY
Definition: buf_internals.h:58
static void FlushBuffer(BufferDesc *buf, SMgrRelation reln)
Definition: bufmgr.c:2683
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1726
static bool StartBufferIO(BufferDesc *buf, bool forInput)
Definition: bufmgr.c:3941
int BufTableInsert(BufferTag *tagPtr, uint32 hashcode, int buf_id)
Definition: buf_table.c:119
char relpersistence
Definition: pg_class.h:78
void ScheduleBufferTagForWriteback(WritebackContext *context, BufferTag *tag)
Definition: bufmgr.c:4310
#define BUF_FLAG_MASK
Definition: buf_internals.h:45
RelFileNodeBackend smgr_rnode
Definition: smgr.h:42
WritebackContext BackendWritebackContext
Definition: buf_init.c:23
bool LWLockConditionalAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1294
static char * buf
Definition: pg_test_fsync.c:67
#define BUF_USAGECOUNT_ONE
Definition: buf_internals.h:43
#define GetBufferDescriptor(id)
#define BM_JUST_DIRTIED
Definition: buf_internals.h:63
unsigned int uint32
Definition: c.h:359
static void UnpinBuffer(BufferDesc *buf, bool fixOwner)
Definition: bufmgr.c:1702
#define BM_VALID
Definition: buf_internals.h:59
bool StrategyRejectBuffer(BufferAccessStrategy strategy, BufferDesc *buf)
Definition: freelist.c:686
RelFileNode node
Definition: relfilenode.h:74
#define BufferDescriptorGetContentLock(bdesc)
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:739
#define INIT_BUFFERTAG(a, xx_rnode, xx_forkNum, xx_blockNum)
static void PinBuffer_Locked(BufferDesc *buf)
Definition: bufmgr.c:1664
#define BUF_USAGECOUNT_MASK
Definition: buf_internals.h:42
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
#define BM_IO_ERROR
Definition: buf_internals.h:62
BufferTag tag
#define UnlockBufHdr(desc, s)
static void ReservePrivateRefCountEntry(void)
Definition: bufmgr.c:187
#define BufferGetLSN(bufHdr)
Definition: bufmgr.c:59
#define BUF_STATE_GET_REFCOUNT(state)
Definition: buf_internals.h:48

◆ BufferGetBlockNumber()

BlockNumber BufferGetBlockNumber ( Buffer  buffer)

Definition at line 2623 of file bufmgr.c.

References Assert, buftag::blockNum, BufferIsLocal, BufferIsPinned, GetBufferDescriptor, GetLocalBufferDescriptor, and BufferDesc::tag.

Referenced by _bt_checkpage(), _bt_endpoint(), _bt_finish_split(), _bt_first(), _bt_getroot(), _bt_insert_parent(), _bt_insertonpg(), _bt_mark_page_halfdead(), _bt_moveright(), _bt_newroot(), _bt_pagedel(), _bt_readnextpage(), _bt_readpage(), _bt_restore_meta(), _bt_search(), _bt_split(), _bt_unlink_halfdead_page(), _bt_walk_left(), _hash_addovflpage(), _hash_checkpage(), _hash_first(), _hash_freeovflpage(), _hash_getnewbuf(), _hash_readnext(), _hash_readpage(), _hash_splitbucket(), allocNewBuffer(), blinsert(), BloomInitMetapage(), brin_doinsert(), brin_doupdate(), brin_getinsertbuffer(), brin_initialize_empty_new_buffer(), brin_page_cleanup(), brin_xlog_insert_update(), brinbuild(), brinGetTupleForHeapBlock(), CheckForSerializableConflictIn(), createPostingTree(), dataBeginPlaceToPageLeaf(), dataPrepareDownlink(), doPickSplit(), entryPrepareDownlink(), fill_seq_with_data(), ginFindParents(), ginFinishSplit(), ginPlaceToPage(), ginRedoDeleteListPages(), ginRedoUpdateMetapage(), ginScanToDelete(), gistbufferinginserttuples(), gistbuild(), gistcheckpage(), gistdeletepage(), gistformdownlink(), gistMemorizeAllDownlinks(), gistplacetopage(), gistRelocateBuildBuffersOnSplit(), gistScanPage(), hash_xlog_add_ovfl_page(), heap_delete(), heap_hot_search_buffer(), heap_insert(), heap_multi_insert(), heap_page_is_all_visible(), heap_prune_chain(), heap_update(), heap_xlog_confirm(), heap_xlog_lock(), makeSublist(), moveLeafs(), moveRightIfItNeeded(), pgstathashindex(), ReadBufferBI(), RelationAddExtraBlocks(), RelationGetBufferForTuple(), RelationPutHeapTuple(), revmap_get_buffer(), revmap_physical_extend(), spgAddNodeAction(), spgbuild(), spgdoinsert(), SpGistSetLastUsedPage(), spgSplitNodeAction(), spgWalk(), startScanEntry(), terminate_brin_buildstate(), vacuumLeafPage(), visibilitymap_clear(), visibilitymap_get_status(), visibilitymap_pin(), visibilitymap_pin_ok(), visibilitymap_set(), and XLogReadBufferExtended().

2624 {
2625  BufferDesc *bufHdr;
2626 
2627  Assert(BufferIsPinned(buffer));
2628 
2629  if (BufferIsLocal(buffer))
2630  bufHdr = GetLocalBufferDescriptor(-buffer - 1);
2631  else
2632  bufHdr = GetBufferDescriptor(buffer - 1);
2633 
2634  /* pinned, so OK to read tag without spinlock */
2635  return bufHdr->tag.blockNum;
2636 }
#define BufferIsPinned(bufnum)
Definition: bufmgr.c:420
#define GetLocalBufferDescriptor(id)
#define GetBufferDescriptor(id)
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
BlockNumber blockNum
Definition: buf_internals.h:94
BufferTag tag

◆ BufferGetLSNAtomic()

XLogRecPtr BufferGetLSNAtomic ( Buffer  buffer)

Definition at line 2886 of file bufmgr.c.

References Assert, BufferGetPage, BufferIsLocal, BufferIsPinned, BufferIsValid, GetBufferDescriptor, LockBufHdr(), PageGetLSN, UnlockBufHdr, and XLogHintBitIsNeeded.

Referenced by _bt_killitems(), _bt_readpage(), gistdoinsert(), gistFindPath(), gistkillitems(), gistScanPage(), SetHintBits(), and XLogSaveBufferForHint().

2887 {
2888  BufferDesc *bufHdr = GetBufferDescriptor(buffer - 1);
2889  char *page = BufferGetPage(buffer);
2890  XLogRecPtr lsn;
2891  uint32 buf_state;
2892 
2893  /*
2894  * If we don't need locking for correctness, fastpath out.
2895  */
2896  if (!XLogHintBitIsNeeded() || BufferIsLocal(buffer))
2897  return PageGetLSN(page);
2898 
2899  /* Make sure we've got a real buffer, and that we hold a pin on it. */
2900  Assert(BufferIsValid(buffer));
2901  Assert(BufferIsPinned(buffer));
2902 
2903  buf_state = LockBufHdr(bufHdr);
2904  lsn = PageGetLSN(page);
2905  UnlockBufHdr(bufHdr, buf_state);
2906 
2907  return lsn;
2908 }
#define BufferIsPinned(bufnum)
Definition: bufmgr.c:420
#define GetBufferDescriptor(id)
unsigned int uint32
Definition: c.h:359
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
#define PageGetLSN(page)
Definition: bufpage.h:366
#define UnlockBufHdr(desc, s)
#define XLogHintBitIsNeeded()
Definition: xlog.h:192

◆ BufferGetTag()

void BufferGetTag ( Buffer  buffer,
RelFileNode rnode,
ForkNumber forknum,
BlockNumber blknum 
)

Definition at line 2644 of file bufmgr.c.

References Assert, buftag::blockNum, BufferIsLocal, BufferIsPinned, buftag::forkNum, GetBufferDescriptor, GetLocalBufferDescriptor, buftag::rnode, and BufferDesc::tag.

Referenced by fsm_search_avail(), ginRedoInsertEntry(), log_newpage_buffer(), ResolveCminCmaxDuringDecoding(), XLogRegisterBuffer(), and XLogSaveBufferForHint().

2646 {
2647  BufferDesc *bufHdr;
2648 
2649  /* Do the same checks as BufferGetBlockNumber. */
2650  Assert(BufferIsPinned(buffer));
2651 
2652  if (BufferIsLocal(buffer))
2653  bufHdr = GetLocalBufferDescriptor(-buffer - 1);
2654  else
2655  bufHdr = GetBufferDescriptor(buffer - 1);
2656 
2657  /* pinned, so OK to read tag without spinlock */
2658  *rnode = bufHdr->tag.rnode;
2659  *forknum = bufHdr->tag.forkNum;
2660  *blknum = bufHdr->tag.blockNum;
2661 }
#define BufferIsPinned(bufnum)
Definition: bufmgr.c:420
ForkNumber forkNum
Definition: buf_internals.h:93
#define GetLocalBufferDescriptor(id)
#define GetBufferDescriptor(id)
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
BlockNumber blockNum
Definition: buf_internals.h:94
RelFileNode rnode
Definition: buf_internals.h:92
BufferTag tag

◆ BufferIsPermanent()

bool BufferIsPermanent ( Buffer  buffer)

Definition at line 2856 of file bufmgr.c.

References Assert, BM_PERMANENT, BufferIsLocal, BufferIsPinned, BufferIsValid, GetBufferDescriptor, pg_atomic_read_u32(), and BufferDesc::state.

Referenced by SetHintBits().

2857 {
2858  BufferDesc *bufHdr;
2859 
2860  /* Local buffers are used only for temp relations. */
2861  if (BufferIsLocal(buffer))
2862  return false;
2863 
2864  /* Make sure we've got a real buffer, and that we hold a pin on it. */
2865  Assert(BufferIsValid(buffer));
2866  Assert(BufferIsPinned(buffer));
2867 
2868  /*
2869  * BM_PERMANENT can't be changed while we hold a pin on the buffer, so we
2870  * need not bother with the buffer header spinlock. Even if someone else
2871  * changes the buffer header state while we're doing this, the state is
2872  * changed atomically, so we'll read the old value or the new value, but
2873  * not random garbage.
2874  */
2875  bufHdr = GetBufferDescriptor(buffer - 1);
2876  return (pg_atomic_read_u32(&bufHdr->state) & BM_PERMANENT) != 0;
2877 }
#define BufferIsPinned(bufnum)
Definition: bufmgr.c:420
#define BM_PERMANENT
Definition: buf_internals.h:66
#define GetBufferDescriptor(id)
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
pg_atomic_uint32 state
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241

◆ BufferSync()

static void BufferSync ( int  flags)
static

Definition at line 1786 of file bufmgr.c.

References Assert, BgWriterStats, binaryheap_add_unordered(), binaryheap_allocate(), binaryheap_build(), binaryheap_empty, binaryheap_first(), binaryheap_free(), binaryheap_remove_first(), binaryheap_replace_first(), buftag::blockNum, CkptSortItem::blockNum, BM_CHECKPOINT_NEEDED, BM_DIRTY, BM_PERMANENT, CkptSortItem::buf_id, BUF_WRITTEN, CHECKPOINT_END_OF_RECOVERY, checkpoint_flush_after, CHECKPOINT_FLUSH_ALL, CHECKPOINT_IS_SHUTDOWN, CheckpointStats, CheckpointWriteDelay(), ckpt_buforder_comparator(), CheckpointStatsData::ckpt_bufs_written, CkptBufferIds, CurrentResourceOwner, DatumGetPointer, buftag::forkNum, CkptSortItem::forkNum, GetBufferDescriptor, i, CkptTsStatus::index, InvalidOid, IssuePendingWritebacks(), LockBufHdr(), PgStat_MsgBgWriter::m_buf_written_checkpoints, NBuffers, CkptTsStatus::num_scanned, CkptTsStatus::num_to_scan, palloc(), pfree(), pg_atomic_read_u32(), PointerGetDatum, ProcessProcSignalBarrier(), ProcSignalBarrierPending, CkptTsStatus::progress, CkptTsStatus::progress_slice, qsort, RelFileNode::relNode, CkptSortItem::relNode, repalloc(), ResourceOwnerEnlargeBuffers(), buftag::rnode, RelFileNode::spcNode, BufferDesc::state, SyncOneBuffer(), BufferDesc::tag, ts_ckpt_progress_comparator(), CkptTsStatus::tsId, CkptSortItem::tsId, UnlockBufHdr, and WritebackContextInit().

Referenced by CheckPointBuffers().

1787 {
1788  uint32 buf_state;
1789  int buf_id;
1790  int num_to_scan;
1791  int num_spaces;
1792  int num_processed;
1793  int num_written;
1794  CkptTsStatus *per_ts_stat = NULL;
1795  Oid last_tsid;
1796  binaryheap *ts_heap;
1797  int i;
1798  int mask = BM_DIRTY;
1799  WritebackContext wb_context;
1800 
1801  /* Make sure we can handle the pin inside SyncOneBuffer */
1803 
1804  /*
1805  * Unless this is a shutdown checkpoint or we have been explicitly told,
1806  * we write only permanent, dirty buffers. But at shutdown or end of
1807  * recovery, we write all dirty buffers.
1808  */
1811  mask |= BM_PERMANENT;
1812 
1813  /*
1814  * Loop over all buffers, and mark the ones that need to be written with
1815  * BM_CHECKPOINT_NEEDED. Count them as we go (num_to_scan), so that we
1816  * can estimate how much work needs to be done.
1817  *
1818  * This allows us to write only those pages that were dirty when the
1819  * checkpoint began, and not those that get dirtied while it proceeds.
1820  * Whenever a page with BM_CHECKPOINT_NEEDED is written out, either by us
1821  * later in this function, or by normal backends or the bgwriter cleaning
1822  * scan, the flag is cleared. Any buffer dirtied after this point won't
1823  * have the flag set.
1824  *
1825  * Note that if we fail to write some buffer, we may leave buffers with
1826  * BM_CHECKPOINT_NEEDED still set. This is OK since any such buffer would
1827  * certainly need to be written for the next checkpoint attempt, too.
1828  */
1829  num_to_scan = 0;
1830  for (buf_id = 0; buf_id < NBuffers; buf_id++)
1831  {
1832  BufferDesc *bufHdr = GetBufferDescriptor(buf_id);
1833 
1834  /*
1835  * Header spinlock is enough to examine BM_DIRTY, see comment in
1836  * SyncOneBuffer.
1837  */
1838  buf_state = LockBufHdr(bufHdr);
1839 
1840  if ((buf_state & mask) == mask)
1841  {
1842  CkptSortItem *item;
1843 
1844  buf_state |= BM_CHECKPOINT_NEEDED;
1845 
1846  item = &CkptBufferIds[num_to_scan++];
1847  item->buf_id = buf_id;
1848  item->tsId = bufHdr->tag.rnode.spcNode;
1849  item->relNode = bufHdr->tag.rnode.relNode;
1850  item->forkNum = bufHdr->tag.forkNum;
1851  item->blockNum = bufHdr->tag.blockNum;
1852  }
1853 
1854  UnlockBufHdr(bufHdr, buf_state);
1855 
1856  /* Check for barrier events in case NBuffers is large. */
1859  }
1860 
1861  if (num_to_scan == 0)
1862  return; /* nothing to do */
1863 
1865 
1866  TRACE_POSTGRESQL_BUFFER_SYNC_START(NBuffers, num_to_scan);
1867 
1868  /*
1869  * Sort buffers that need to be written to reduce the likelihood of random
1870  * IO. The sorting is also important for the implementation of balancing
1871  * writes between tablespaces. Without balancing writes we'd potentially
1872  * end up writing to the tablespaces one-by-one; possibly overloading the
1873  * underlying system.
1874  */
1875  qsort(CkptBufferIds, num_to_scan, sizeof(CkptSortItem),
1877 
1878  num_spaces = 0;
1879 
1880  /*
1881  * Allocate progress status for each tablespace with buffers that need to
1882  * be flushed. This requires the to-be-flushed array to be sorted.
1883  */
1884  last_tsid = InvalidOid;
1885  for (i = 0; i < num_to_scan; i++)
1886  {
1887  CkptTsStatus *s;
1888  Oid cur_tsid;
1889 
1890  cur_tsid = CkptBufferIds[i].tsId;
1891 
1892  /*
1893  * Grow array of per-tablespace status structs, every time a new
1894  * tablespace is found.
1895  */
1896  if (last_tsid == InvalidOid || last_tsid != cur_tsid)
1897  {
1898  Size sz;
1899 
1900  num_spaces++;
1901 
1902  /*
1903  * Not worth adding grow-by-power-of-2 logic here - even with a
1904  * few hundred tablespaces this should be fine.
1905  */
1906  sz = sizeof(CkptTsStatus) * num_spaces;
1907 
1908  if (per_ts_stat == NULL)
1909  per_ts_stat = (CkptTsStatus *) palloc(sz);
1910  else
1911  per_ts_stat = (CkptTsStatus *) repalloc(per_ts_stat, sz);
1912 
1913  s = &per_ts_stat[num_spaces - 1];
1914  memset(s, 0, sizeof(*s));
1915  s->tsId = cur_tsid;
1916 
1917  /*
1918  * The first buffer in this tablespace. As CkptBufferIds is sorted
1919  * by tablespace all (s->num_to_scan) buffers in this tablespace
1920  * will follow afterwards.
1921  */
1922  s->index = i;
1923 
1924  /*
1925  * progress_slice will be determined once we know how many buffers
1926  * are in each tablespace, i.e. after this loop.
1927  */
1928 
1929  last_tsid = cur_tsid;
1930  }
1931  else
1932  {
1933  s = &per_ts_stat[num_spaces - 1];
1934  }
1935 
1936  s->num_to_scan++;
1937 
1938  /* Check for barrier events. */
1941  }
1942 
1943  Assert(num_spaces > 0);
1944 
1945  /*
1946  * Build a min-heap over the write-progress in the individual tablespaces,
1947  * and compute how large a portion of the total progress a single
1948  * processed buffer is.
1949  */
1950  ts_heap = binaryheap_allocate(num_spaces,
1952  NULL);
1953 
1954  for (i = 0; i < num_spaces; i++)
1955  {
1956  CkptTsStatus *ts_stat = &per_ts_stat[i];
1957 
1958  ts_stat->progress_slice = (float8) num_to_scan / ts_stat->num_to_scan;
1959 
1960  binaryheap_add_unordered(ts_heap, PointerGetDatum(ts_stat));
1961  }
1962 
1963  binaryheap_build(ts_heap);
1964 
1965  /*
1966  * Iterate through to-be-checkpointed buffers and write the ones (still)
1967  * marked with BM_CHECKPOINT_NEEDED. The writes are balanced between
1968  * tablespaces; otherwise the sorting would lead to only one tablespace
1969  * receiving writes at a time, making inefficient use of the hardware.
1970  */
1971  num_processed = 0;
1972  num_written = 0;
1973  while (!binaryheap_empty(ts_heap))
1974  {
1975  BufferDesc *bufHdr = NULL;
1976  CkptTsStatus *ts_stat = (CkptTsStatus *)
1978 
1979  buf_id = CkptBufferIds[ts_stat->index].buf_id;
1980  Assert(buf_id != -1);
1981 
1982  bufHdr = GetBufferDescriptor(buf_id);
1983 
1984  num_processed++;
1985 
1986  /*
1987  * We don't need to acquire the lock here, because we're only looking
1988  * at a single bit. It's possible that someone else writes the buffer
1989  * and clears the flag right after we check, but that doesn't matter
1990  * since SyncOneBuffer will then do nothing. However, there is a
1991  * further race condition: it's conceivable that between the time we
1992  * examine the bit here and the time SyncOneBuffer acquires the lock,
1993  * someone else not only wrote the buffer but replaced it with another
1994  * page and dirtied it. In that improbable case, SyncOneBuffer will
1995  * write the buffer though we didn't need to. It doesn't seem worth
1996  * guarding against this, though.
1997  */
1999  {
2000  if (SyncOneBuffer(buf_id, false, &wb_context) & BUF_WRITTEN)
2001  {
2002  TRACE_POSTGRESQL_BUFFER_SYNC_WRITTEN(buf_id);
2004  num_written++;
2005  }
2006  }
2007 
2008  /*
2009  * Measure progress independent of actually having to flush the buffer
2010  * - otherwise writing become unbalanced.
2011  */
2012  ts_stat->progress += ts_stat->progress_slice;
2013  ts_stat->num_scanned++;
2014  ts_stat->index++;
2015 
2016  /* Have all the buffers from the tablespace been processed? */
2017  if (ts_stat->num_scanned == ts_stat->num_to_scan)
2018  {
2019  binaryheap_remove_first(ts_heap);
2020  }
2021  else
2022  {
2023  /* update heap with the new progress */
2024  binaryheap_replace_first(ts_heap, PointerGetDatum(ts_stat));
2025  }
2026 
2027  /*
2028  * Sleep to throttle our I/O rate.
2029  *
2030  * (This will check for barrier events even if it doesn't sleep.)
2031  */
2032  CheckpointWriteDelay(flags, (double) num_processed / num_to_scan);
2033  }
2034 
2035  /* issue all pending flushes */
2036  IssuePendingWritebacks(&wb_context);
2037 
2038  pfree(per_ts_stat);
2039  per_ts_stat = NULL;
2040  binaryheap_free(ts_heap);
2041 
2042  /*
2043  * Update checkpoint statistics. As noted above, this doesn't include
2044  * buffers written by other backends or bgwriter scan.
2045  */
2046  CheckpointStats.ckpt_bufs_written += num_written;
2047 
2048  TRACE_POSTGRESQL_BUFFER_SYNC_DONE(NBuffers, num_written, num_to_scan);
2049 }
void CheckpointWriteDelay(int flags, double progress)
Definition: checkpointer.c:663
PgStat_Counter m_buf_written_checkpoints
Definition: pgstat.h:415
#define BM_PERMANENT
Definition: buf_internals.h:66
#define CHECKPOINT_FLUSH_ALL
Definition: xlog.h:216
Oid tsId
Definition: bufmgr.c:87
#define binaryheap_empty(h)
Definition: binaryheap.h:52
ForkNumber forkNum
Definition: buf_internals.h:93
#define PointerGetDatum(X)
Definition: postgres.h:556
ResourceOwner CurrentResourceOwner
Definition: resowner.c:142
#define BM_CHECKPOINT_NEEDED
Definition: buf_internals.h:65
void ProcessProcSignalBarrier(void)
Definition: procsignal.c:429
static int ts_ckpt_progress_comparator(Datum a, Datum b, void *arg)
Definition: bufmgr.c:4275
PgStat_MsgBgWriter BgWriterStats
Definition: pgstat.c:142
int checkpoint_flush_after
Definition: bufmgr.c:119
void binaryheap_replace_first(binaryheap *heap, Datum d)
Definition: binaryheap.c:204
unsigned int Oid
Definition: postgres_ext.h:31
#define BM_DIRTY
Definition: buf_internals.h:58
void binaryheap_add_unordered(binaryheap *heap, Datum d)
Definition: binaryheap.c:110
static int SyncOneBuffer(int buf_id, bool skip_recently_used, WritebackContext *wb_context)
Definition: bufmgr.c:2366
void IssuePendingWritebacks(WritebackContext *context)
Definition: bufmgr.c:4344
void WritebackContextInit(WritebackContext *context, int *max_pending)
Definition: bufmgr.c:4298
void pfree(void *pointer)
Definition: mcxt.c:1056
double float8
Definition: c.h:492
Datum binaryheap_first(binaryheap *heap)
Definition: binaryheap.c:159
int num_to_scan
Definition: bufmgr.c:100
float8 progress_slice
Definition: bufmgr.c:97
int index
Definition: bufmgr.c:105
float8 progress
Definition: bufmgr.c:96
static int ckpt_buforder_comparator(const void *pa, const void *pb)
Definition: bufmgr.c:4241
#define CHECKPOINT_END_OF_RECOVERY
Definition: xlog.h:212
#define GetBufferDescriptor(id)
unsigned int uint32
Definition: c.h:359
#define BUF_WRITTEN
Definition: bufmgr.c:66
void ResourceOwnerEnlargeBuffers(ResourceOwner owner)
Definition: resowner.c:893
int ckpt_bufs_written
Definition: xlog.h:241
BlockNumber blockNum
#define InvalidOid
Definition: postgres_ext.h:36
void binaryheap_build(binaryheap *heap)
Definition: binaryheap.c:126
const symbol * s
Definition: header.h:17
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
#define Assert(condition)
Definition: c.h:739
volatile sig_atomic_t ProcSignalBarrierPending
Definition: globals.c:35
CheckpointStatsData CheckpointStats
Definition: xlog.c:181
CkptSortItem * CkptBufferIds
Definition: buf_init.c:24
size_t Size
Definition: c.h:467
void binaryheap_free(binaryheap *heap)
Definition: binaryheap.c:69
BlockNumber blockNum
Definition: buf_internals.h:94
RelFileNode rnode
Definition: buf_internals.h:92
void * repalloc(void *pointer, Size size)
Definition: mcxt.c:1069
binaryheap * binaryheap_allocate(int capacity, binaryheap_comparator compare, void *arg)
Definition: binaryheap.c:33
#define DatumGetPointer(X)
Definition: postgres.h:549
BufferTag tag
void * palloc(Size size)
Definition: mcxt.c:949
#define UnlockBufHdr(desc, s)
int i
int NBuffers
Definition: globals.c:131
pg_atomic_uint32 state
Datum binaryheap_remove_first(binaryheap *heap)
Definition: binaryheap.c:174
int num_scanned
Definition: bufmgr.c:102
#define qsort(a, b, c, d)
Definition: port.h:491
ForkNumber forkNum
struct CkptTsStatus CkptTsStatus
#define CHECKPOINT_IS_SHUTDOWN
Definition: xlog.h:211
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241

◆ buffertag_comparator()

static int buffertag_comparator ( const void *  p1,
const void *  p2 
)
static

Definition at line 4210 of file bufmgr.c.

References buftag::blockNum, buftag::forkNum, buftag::rnode, and rnode_comparator().

Referenced by IssuePendingWritebacks().

4211 {
4212  const BufferTag *ba = (const BufferTag *) a;
4213  const BufferTag *bb = (const BufferTag *) b;
4214  int ret;
4215 
4216  ret = rnode_comparator(&ba->rnode, &bb->rnode);
4217 
4218  if (ret != 0)
4219  return ret;
4220 
4221  if (ba->forkNum < bb->forkNum)
4222  return -1;
4223  if (ba->forkNum > bb->forkNum)
4224  return 1;
4225 
4226  if (ba->blockNum < bb->blockNum)
4227  return -1;
4228  if (ba->blockNum > bb->blockNum)
4229  return 1;
4230 
4231  return 0;
4232 }
ForkNumber forkNum
Definition: buf_internals.h:93
static int rnode_comparator(const void *p1, const void *p2)
Definition: bufmgr.c:4131
BlockNumber blockNum
Definition: buf_internals.h:94
RelFileNode rnode
Definition: buf_internals.h:92

◆ BufmgrCommit()

void BufmgrCommit ( void  )

Definition at line 2609 of file bufmgr.c.

Referenced by PrepareTransaction(), and RecordTransactionCommit().

2610 {
2611  /* Nothing to do in bufmgr anymore... */
2612 }

◆ CheckForBufferLeaks()

static void CheckForBufferLeaks ( void  )
static

Definition at line 2508 of file bufmgr.c.

References Assert, PrivateRefCountEntry::buffer, hash_seq_init(), hash_seq_search(), i, InvalidBuffer, PrintBufferLeakWarning(), PrivateRefCountArray, PrivateRefCountOverflowed, and REFCOUNT_ARRAY_ENTRIES.

Referenced by AtEOXact_Buffers(), and AtProcExit_Buffers().

2509 {
2510 #ifdef USE_ASSERT_CHECKING
2511  int RefCountErrors = 0;
2512  PrivateRefCountEntry *res;
2513  int i;
2514 
2515  /* check the array */
2516  for (i = 0; i < REFCOUNT_ARRAY_ENTRIES; i++)
2517  {
2518  res = &PrivateRefCountArray[i];
2519 
2520  if (res->buffer != InvalidBuffer)
2521  {
2523  RefCountErrors++;
2524  }
2525  }
2526 
2527  /* if necessary search the hash */
2529  {
2530  HASH_SEQ_STATUS hstat;
2531 
2533  while ((res = (PrivateRefCountEntry *) hash_seq_search(&hstat)) != NULL)
2534  {
2536  RefCountErrors++;
2537  }
2538 
2539  }
2540 
2541  Assert(RefCountErrors == 0);
2542 #endif
2543 }
void PrintBufferLeakWarning(Buffer buffer)
Definition: bufmgr.c:2549
static int32 PrivateRefCountOverflowed
Definition: bufmgr.c:171
#define InvalidBuffer
Definition: buf.h:25
static struct PrivateRefCountEntry PrivateRefCountArray[REFCOUNT_ARRAY_ENTRIES]
Definition: bufmgr.c:169
#define Assert(condition)
Definition: c.h:739
#define REFCOUNT_ARRAY_ENTRIES
Definition: bufmgr.c:78
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1389
void hash_seq_init(HASH_SEQ_STATUS *status, HTAB *hashp)
Definition: dynahash.c:1379
static HTAB * PrivateRefCountHash
Definition: bufmgr.c:170
int i

◆ CheckPointBuffers()

void CheckPointBuffers ( int  flags)

Definition at line 2592 of file bufmgr.c.

References BufferSync(), CheckpointStats, CheckpointStatsData::ckpt_sync_end_t, CheckpointStatsData::ckpt_sync_t, CheckpointStatsData::ckpt_write_t, GetCurrentTimestamp(), and ProcessSyncRequests().

Referenced by CheckPointGuts().

2593 {
2594  TRACE_POSTGRESQL_BUFFER_CHECKPOINT_START(flags);
2596  BufferSync(flags);
2598  TRACE_POSTGRESQL_BUFFER_CHECKPOINT_SYNC_START();
2601  TRACE_POSTGRESQL_BUFFER_CHECKPOINT_DONE();
2602 }
void ProcessSyncRequests(void)
Definition: sync.c:236
TimestampTz ckpt_sync_end_t
Definition: xlog.h:238
TimestampTz GetCurrentTimestamp(void)
Definition: timestamp.c:1583
static void BufferSync(int flags)
Definition: bufmgr.c:1786
CheckpointStatsData CheckpointStats
Definition: xlog.c:181
TimestampTz ckpt_write_t
Definition: xlog.h:236
TimestampTz ckpt_sync_t
Definition: xlog.h:237

◆ ckpt_buforder_comparator()

static int ckpt_buforder_comparator ( const void *  pa,
const void *  pb 
)
static

Definition at line 4241 of file bufmgr.c.

References CkptSortItem::blockNum, CkptSortItem::forkNum, CkptSortItem::relNode, and CkptSortItem::tsId.

Referenced by BufferSync().

4242 {
4243  const CkptSortItem *a = (const CkptSortItem *) pa;
4244  const CkptSortItem *b = (const CkptSortItem *) pb;
4245 
4246  /* compare tablespace */
4247  if (a->tsId < b->tsId)
4248  return -1;
4249  else if (a->tsId > b->tsId)
4250  return 1;
4251  /* compare relation */
4252  if (a->relNode < b->relNode)
4253  return -1;
4254  else if (a->relNode > b->relNode)
4255  return 1;
4256  /* compare fork */
4257  else if (a->forkNum < b->forkNum)
4258  return -1;
4259  else if (a->forkNum > b->forkNum)
4260  return 1;
4261  /* compare block number */
4262  else if (a->blockNum < b->blockNum)
4263  return -1;
4264  else if (a->blockNum > b->blockNum)
4265  return 1;
4266  /* equal page IDs are unlikely, but not impossible */
4267  return 0;
4268 }
BlockNumber blockNum
ForkNumber forkNum

◆ ComputeIoConcurrency()

bool ComputeIoConcurrency ( int  io_concurrency,
double *  target 
)

Definition at line 469 of file bufmgr.c.

References i, Max, MAX_IO_CONCURRENCY, and Min.

Referenced by check_effective_io_concurrency(), and ExecInitBitmapHeapScan().

470 {
471  double new_prefetch_pages = 0.0;
472  int i;
473 
474  /*
475  * Make sure the io_concurrency value is within valid range; it may have
476  * been forced with a manual pg_tablespace update.
477  */
478  io_concurrency = Min(Max(io_concurrency, 0), MAX_IO_CONCURRENCY);
479 
480  /*----------
481  * The user-visible GUC parameter is the number of drives (spindles),
482  * which we need to translate to a number-of-pages-to-prefetch target.
483  * The target value is stashed in *extra and then assigned to the actual
484  * variable by assign_effective_io_concurrency.
485  *
486  * The expected number of prefetch pages needed to keep N drives busy is:
487  *
488  * drives | I/O requests
489  * -------+----------------
490  * 1 | 1
491  * 2 | 2/1 + 2/2 = 3
492  * 3 | 3/1 + 3/2 + 3/3 = 5 1/2
493  * 4 | 4/1 + 4/2 + 4/3 + 4/4 = 8 1/3
494  * n | n * H(n)
495  *
496  * This is called the "coupon collector problem" and H(n) is called the
497  * harmonic series. This could be approximated by n * ln(n), but for
498  * reasonable numbers of drives we might as well just compute the series.
499  *
500  * Alternatively we could set the target to the number of pages necessary
501  * so that the expected number of active spindles is some arbitrary
502  * percentage of the total. This sounds the same but is actually slightly
503  * different. The result ends up being ln(1-P)/ln((n-1)/n) where P is
504  * that desired fraction.
505  *
506  * Experimental results show that both of these formulas aren't aggressive
507  * enough, but we don't really have any better proposals.
508  *
509  * Note that if io_concurrency = 0 (disabled), we must set target = 0.
510  *----------
511  */
512 
513  for (i = 1; i <= io_concurrency; i++)
514  new_prefetch_pages += (double) io_concurrency / (double) i;
515 
516  *target = new_prefetch_pages;
517 
518  /* This range check shouldn't fail, but let's be paranoid */
519  return (new_prefetch_pages >= 0.0 && new_prefetch_pages < (double) INT_MAX);
520 }
#define MAX_IO_CONCURRENCY
Definition: bufmgr.h:78
#define Min(x, y)
Definition: c.h:911
#define Max(x, y)
Definition: c.h:905
int i

◆ ConditionalLockBuffer()

bool ConditionalLockBuffer ( Buffer  buffer)

Definition at line 3638 of file bufmgr.c.

References Assert, buf, BufferDescriptorGetContentLock, BufferIsLocal, BufferIsValid, GetBufferDescriptor, LW_EXCLUSIVE, and LWLockConditionalAcquire().

Referenced by _bt_doinsert(), _bt_getbuf(), BloomNewBuffer(), ConditionalLockBufferForCleanup(), GinNewBuffer(), gistNewBuffer(), RelationGetBufferForTuple(), spgdoinsert(), SpGistGetBuffer(), SpGistNewBuffer(), and SpGistUpdateMetaPage().

3639 {
3640  BufferDesc *buf;
3641 
3642  Assert(BufferIsValid(buffer));
3643  if (BufferIsLocal(buffer))
3644  return true; /* act as though we got it */
3645 
3646  buf = GetBufferDescriptor(buffer - 1);
3647 
3649  LW_EXCLUSIVE);
3650 }
bool LWLockConditionalAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1294
static char * buf
Definition: pg_test_fsync.c:67
#define GetBufferDescriptor(id)
#define BufferDescriptorGetContentLock(bdesc)
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113

◆ ConditionalLockBufferForCleanup()

bool ConditionalLockBufferForCleanup ( Buffer  buffer)

Definition at line 3784 of file bufmgr.c.

References Assert, BUF_STATE_GET_REFCOUNT, BUFFER_LOCK_UNLOCK, BufferIsLocal, BufferIsValid, ConditionalLockBuffer(), GetBufferDescriptor, GetPrivateRefCount(), LocalRefCount, LockBuffer(), LockBufHdr(), PrivateRefCountEntry::refcount, and UnlockBufHdr.

Referenced by _hash_finish_split(), _hash_getbuf_with_condlock_cleanup(), heap_page_prune_opt(), lazy_scan_heap(), and lazy_vacuum_heap().

3785 {
3786  BufferDesc *bufHdr;
3787  uint32 buf_state,
3788  refcount;
3789 
3790  Assert(BufferIsValid(buffer));
3791 
3792  if (BufferIsLocal(buffer))
3793  {
3794  refcount = LocalRefCount[-buffer - 1];
3795  /* There should be exactly one pin */
3796  Assert(refcount > 0);
3797  if (refcount != 1)
3798  return false;
3799  /* Nobody else to wait for */
3800  return true;
3801  }
3802 
3803  /* There should be exactly one local pin */
3804  refcount = GetPrivateRefCount(buffer);
3805  Assert(refcount);
3806  if (refcount != 1)
3807  return false;
3808 
3809  /* Try to acquire lock */
3810  if (!ConditionalLockBuffer(buffer))
3811  return false;
3812 
3813  bufHdr = GetBufferDescriptor(buffer - 1);
3814  buf_state = LockBufHdr(bufHdr);
3815  refcount = BUF_STATE_GET_REFCOUNT(buf_state);
3816 
3817  Assert(refcount > 0);
3818  if (refcount == 1)
3819  {
3820  /* Successfully acquired exclusive lock with pincount 1 */
3821  UnlockBufHdr(bufHdr, buf_state);
3822  return true;
3823  }
3824 
3825  /* Failed, so release the lock */
3826  UnlockBufHdr(bufHdr, buf_state);
3827  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3828  return false;
3829 }
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:86
#define GetBufferDescriptor(id)
static int32 GetPrivateRefCount(Buffer buffer)
Definition: bufmgr.c:359
unsigned int uint32
Definition: c.h:359
bool ConditionalLockBuffer(Buffer buffer)
Definition: bufmgr.c:3638
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3612
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
#define UnlockBufHdr(desc, s)
int32 * LocalRefCount
Definition: localbuf.c:45
#define BUF_STATE_GET_REFCOUNT(state)
Definition: buf_internals.h:48

◆ DropDatabaseBuffers()

void DropDatabaseBuffers ( Oid  dbid)

Definition at line 3109 of file bufmgr.c.

References buftag::blockNum, buf, BufferDescriptorGetBuffer, RelFileNode::dbNode, elog, buftag::forkNum, BufferDesc::freeNext, GetBufferDescriptor, GetPrivateRefCount(), i, InvalidateBuffer(), InvalidBackendId, LockBufHdr(), LOG, NBuffers, relpathbackend, relpathperm, buftag::rnode, BufferDesc::tag, and UnlockBufHdr.

Referenced by dbase_redo(), dropdb(), and movedb().

3110 {
3111  int i;
3112 
3113  /*
3114  * We needn't consider local buffers, since by assumption the target
3115  * database isn't our own.
3116  */
3117 
3118  for (i = 0; i < NBuffers; i++)
3119  {
3120  BufferDesc *bufHdr = GetBufferDescriptor(i);
3121  uint32 buf_state;
3122 
3123  /*
3124  * As in DropRelFileNodeBuffers, an unlocked precheck should be safe
3125  * and saves some cycles.
3126  */
3127  if (bufHdr->tag.rnode.dbNode != dbid)
3128  continue;
3129 
3130  buf_state = LockBufHdr(bufHdr);
3131  if (bufHdr->tag.rnode.dbNode == dbid)
3132  InvalidateBuffer(bufHdr); /* releases spinlock */
3133  else
3134  UnlockBufHdr(bufHdr, buf_state);
3135  }
3136 }
static void InvalidateBuffer(BufferDesc *buf)
Definition: bufmgr.c:1360
#define GetBufferDescriptor(id)
unsigned int uint32
Definition: c.h:359
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
RelFileNode rnode
Definition: buf_internals.h:92
BufferTag tag
#define UnlockBufHdr(desc, s)
int i
int NBuffers
Definition: globals.c:131

◆ DropRelFileNodeBuffers()

void DropRelFileNodeBuffers ( RelFileNodeBackend  rnode,
ForkNumber forkNum,
int  nforks,
BlockNumber firstDelBlock 
)

Definition at line 2937 of file bufmgr.c.

References RelFileNodeBackend::backend, buftag::blockNum, DropRelFileNodeLocalBuffers(), buftag::forkNum, GetBufferDescriptor, i, InvalidateBuffer(), LockBufHdr(), MyBackendId, NBuffers, RelFileNodeBackend::node, RelFileNodeBackendIsTemp, RelFileNodeEquals, buftag::rnode, BufferDesc::tag, and UnlockBufHdr.

Referenced by smgrtruncate().

2939 {
2940  int i;
2941  int j;
2942 
2943  /* If it's a local relation, it's localbuf.c's problem. */
2944  if (RelFileNodeBackendIsTemp(rnode))
2945  {
2946  if (rnode.backend == MyBackendId)
2947  {
2948  for (j = 0; j < nforks; j++)
2949  DropRelFileNodeLocalBuffers(rnode.node, forkNum[j],
2950  firstDelBlock[j]);
2951  }
2952  return;
2953  }
2954 
2955  for (i = 0; i < NBuffers; i++)
2956  {
2957  BufferDesc *bufHdr = GetBufferDescriptor(i);
2958  uint32 buf_state;
2959 
2960  /*
2961  * We can make this a tad faster by prechecking the buffer tag before
2962  * we attempt to lock the buffer; this saves a lot of lock
2963  * acquisitions in typical cases. It should be safe because the
2964  * caller must have AccessExclusiveLock on the relation, or some other
2965  * reason to be certain that no one is loading new pages of the rel
2966  * into the buffer pool. (Otherwise we might well miss such pages
2967  * entirely.) Therefore, while the tag might be changing while we
2968  * look at it, it can't be changing *to* a value we care about, only
2969  * *away* from such a value. So false negatives are impossible, and
2970  * false positives are safe because we'll recheck after getting the
2971  * buffer lock.
2972  *
2973  * We could check forkNum and blockNum as well as the rnode, but the
2974  * incremental win from doing so seems small.
2975  */
2976  if (!RelFileNodeEquals(bufHdr->tag.rnode, rnode.node))
2977  continue;
2978 
2979  buf_state = LockBufHdr(bufHdr);
2980 
2981  for (j = 0; j < nforks; j++)
2982  {
2983  if (RelFileNodeEquals(bufHdr->tag.rnode, rnode.node) &&
2984  bufHdr->tag.forkNum == forkNum[j] &&
2985  bufHdr->tag.blockNum >= firstDelBlock[j])
2986  {
2987  InvalidateBuffer(bufHdr); /* releases spinlock */
2988  break;
2989  }
2990  }
2991  if (j >= nforks)
2992  UnlockBufHdr(bufHdr, buf_state);
2993  }
2994 }
BackendId MyBackendId
Definition: globals.c:81
#define RelFileNodeBackendIsTemp(rnode)
Definition: relfilenode.h:78
ForkNumber forkNum
Definition: buf_internals.h:93
static void InvalidateBuffer(BufferDesc *buf)
Definition: bufmgr.c:1360
void DropRelFileNodeLocalBuffers(RelFileNode rnode, ForkNumber forkNum, BlockNumber firstDelBlock)
Definition: localbuf.c:320
#define GetBufferDescriptor(id)
unsigned int uint32
Definition: c.h:359
RelFileNode node
Definition: relfilenode.h:74
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
BackendId backend
Definition: relfilenode.h:75
BlockNumber blockNum
Definition: buf_internals.h:94
RelFileNode rnode
Definition: buf_internals.h:92
BufferTag tag
#define UnlockBufHdr(desc, s)
int i
int NBuffers
Definition: globals.c:131
#define RelFileNodeEquals(node1, node2)
Definition: relfilenode.h:88

◆ DropRelFileNodesAllBuffers()

void DropRelFileNodesAllBuffers ( RelFileNodeBackend rnodes,
int  nnodes 
)

Definition at line 3006 of file bufmgr.c.

References DROP_RELS_BSEARCH_THRESHOLD, DropRelFileNodeAllLocalBuffers(), GetBufferDescriptor, i, InvalidateBuffer(), LockBufHdr(), MyBackendId, NBuffers, RelFileNodeBackend::node, palloc(), pfree(), pg_qsort(), RelFileNodeBackendIsTemp, RelFileNodeEquals, buftag::rnode, rnode_comparator(), BufferDesc::tag, and UnlockBufHdr.

Referenced by smgrdounlink(), and smgrdounlinkall().

3007 {
3008  int i,
3009  n = 0;
3010  RelFileNode *nodes;
3011  bool use_bsearch;
3012 
3013  if (nnodes == 0)
3014  return;
3015 
3016  nodes = palloc(sizeof(RelFileNode) * nnodes); /* non-local relations */
3017 
3018  /* If it's a local relation, it's localbuf.c's problem. */
3019  for (i = 0; i < nnodes; i++)
3020  {
3021  if (RelFileNodeBackendIsTemp(rnodes[i]))
3022  {
3023  if (rnodes[i].backend == MyBackendId)
3024  DropRelFileNodeAllLocalBuffers(rnodes[i].node);
3025  }
3026  else
3027  nodes[n++] = rnodes[i].node;
3028  }
3029 
3030  /*
3031  * If there are no non-local relations, then we're done. Release the
3032  * memory and return.
3033  */
3034  if (n == 0)
3035  {
3036  pfree(nodes);
3037  return;
3038  }
3039 
3040  /*
3041  * For low number of relations to drop just use a simple walk through, to
3042  * save the bsearch overhead. The threshold to use is rather a guess than
3043  * an exactly determined value, as it depends on many factors (CPU and RAM
3044  * speeds, amount of shared buffers etc.).
3045  */
3046  use_bsearch = n > DROP_RELS_BSEARCH_THRESHOLD;
3047 
3048  /* sort the list of rnodes if necessary */
3049  if (use_bsearch)
3050  pg_qsort(nodes, n, sizeof(RelFileNode), rnode_comparator);
3051 
3052  for (i = 0; i < NBuffers; i++)
3053  {
3054  RelFileNode *rnode = NULL;
3055  BufferDesc *bufHdr = GetBufferDescriptor(i);
3056  uint32 buf_state;
3057 
3058  /*
3059  * As in DropRelFileNodeBuffers, an unlocked precheck should be safe
3060  * and saves some cycles.
3061  */
3062 
3063  if (!use_bsearch)
3064  {
3065  int j;
3066 
3067  for (j = 0; j < n; j++)
3068  {
3069  if (RelFileNodeEquals(bufHdr->tag.rnode, nodes[j]))
3070  {
3071  rnode = &nodes[j];
3072  break;
3073  }
3074  }
3075  }
3076  else
3077  {
3078  rnode = bsearch((const void *) &(bufHdr->tag.rnode),
3079  nodes, n, sizeof(RelFileNode),
3081  }
3082 
3083  /* buffer doesn't belong to any of the given relfilenodes; skip it */
3084  if (rnode == NULL)
3085  continue;
3086 
3087  buf_state = LockBufHdr(bufHdr);
3088  if (RelFileNodeEquals(bufHdr->tag.rnode, (*rnode)))
3089  InvalidateBuffer(bufHdr); /* releases spinlock */
3090  else
3091  UnlockBufHdr(bufHdr, buf_state);
3092  }
3093 
3094  pfree(nodes);
3095 }
BackendId MyBackendId
Definition: globals.c:81
#define RelFileNodeBackendIsTemp(rnode)
Definition: relfilenode.h:78
static void InvalidateBuffer(BufferDesc *buf)
Definition: bufmgr.c:1360
#define DROP_RELS_BSEARCH_THRESHOLD
Definition: bufmgr.c:69
void DropRelFileNodeAllLocalBuffers(RelFileNode rnode)
Definition: localbuf.c:367
void pfree(void *pointer)
Definition: mcxt.c:1056
#define GetBufferDescriptor(id)
unsigned int uint32
Definition: c.h:359
static int rnode_comparator(const void *p1, const void *p2)
Definition: bufmgr.c:4131
RelFileNode node
Definition: relfilenode.h:74
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
void pg_qsort(void *base, size_t nel, size_t elsize, int(*cmp)(const void *, const void *))
Definition: qsort.c:113
RelFileNode rnode
Definition: buf_internals.h:92
BufferTag tag
void * palloc(Size size)
Definition: mcxt.c:949
#define UnlockBufHdr(desc, s)
int i
int NBuffers
Definition: globals.c:131
#define RelFileNodeEquals(node1, node2)
Definition: relfilenode.h:88

◆ FlushBuffer()

static void FlushBuffer ( BufferDesc buf,
SMgrRelation  reln 
)
static

Definition at line 2683 of file bufmgr.c.

References ErrorContextCallback::arg, BufferUsage::blk_write_time, buftag::blockNum, BM_JUST_DIRTIED, BM_PERMANENT, BufferGetLSN, BufHdrGetBlock, ErrorContextCallback::callback, RelFileNode::dbNode, error_context_stack, buftag::forkNum, INSTR_TIME_ADD, INSTR_TIME_GET_MICROSEC, INSTR_TIME_SET_CURRENT, INSTR_TIME_SUBTRACT, InvalidBackendId, LockBufHdr(), RelFileNodeBackend::node, PageSetChecksumCopy(), pgBufferUsage, pgstat_count_buffer_write_time, ErrorContextCallback::previous, RelFileNode::relNode, buftag::rnode, BufferUsage::shared_blks_written, shared_buffer_write_error_callback(), SMgrRelationData::smgr_rnode, smgropen(), smgrwrite(), RelFileNode::spcNode, StartBufferIO(), BufferDesc::tag, TerminateBufferIO(), track_io_timing, UnlockBufHdr, and XLogFlush().

Referenced by BufferAlloc(), FlushDatabaseBuffers(), FlushOneBuffer(), FlushRelationBuffers(), and SyncOneBuffer().

2684 {
2685  XLogRecPtr recptr;
2686  ErrorContextCallback errcallback;
2687  instr_time io_start,
2688  io_time;
2689  Block bufBlock;
2690  char *bufToWrite;
2691  uint32 buf_state;
2692 
2693  /*
2694  * Acquire the buffer's io_in_progress lock. If StartBufferIO returns
2695  * false, then someone else flushed the buffer before we could, so we need
2696  * not do anything.
2697  */
2698  if (!StartBufferIO(buf, false))
2699  return;
2700 
2701  /* Setup error traceback support for ereport() */
2703  errcallback.arg = (void *) buf;
2704  errcallback.previous = error_context_stack;
2705  error_context_stack = &errcallback;
2706 
2707  /* Find smgr relation for buffer */
2708  if (reln == NULL)
2709  reln = smgropen(buf->tag.rnode, InvalidBackendId);
2710 
2711  TRACE_POSTGRESQL_BUFFER_FLUSH_START(buf->tag.forkNum,
2712  buf->tag.blockNum,
2713  reln->smgr_rnode.node.spcNode,
2714  reln->smgr_rnode.node.dbNode,
2715  reln->smgr_rnode.node.relNode);
2716 
2717  buf_state = LockBufHdr(buf);
2718 
2719  /*
2720  * Run PageGetLSN while holding header lock, since we don't have the
2721  * buffer locked exclusively in all cases.
2722  */
2723  recptr = BufferGetLSN(buf);
2724 
2725  /* To check if block content changes while flushing. - vadim 01/17/97 */
2726  buf_state &= ~BM_JUST_DIRTIED;
2727  UnlockBufHdr(buf, buf_state);
2728 
2729  /*
2730  * Force XLOG flush up to buffer's LSN. This implements the basic WAL
2731  * rule that log updates must hit disk before any of the data-file changes
2732  * they describe do.
2733  *
2734  * However, this rule does not apply to unlogged relations, which will be
2735  * lost after a crash anyway. Most unlogged relation pages do not bear
2736  * LSNs since we never emit WAL records for them, and therefore flushing
2737  * up through the buffer LSN would be useless, but harmless. However,
2738  * GiST indexes use LSNs internally to track page-splits, and therefore
2739  * unlogged GiST pages bear "fake" LSNs generated by
2740  * GetFakeLSNForUnloggedRel. It is unlikely but possible that the fake
2741  * LSN counter could advance past the WAL insertion point; and if it did
2742  * happen, attempting to flush WAL through that location would fail, with
2743  * disastrous system-wide consequences. To make sure that can't happen,
2744  * skip the flush if the buffer isn't permanent.
2745  */
2746  if (buf_state & BM_PERMANENT)
2747  XLogFlush(recptr);
2748 
2749  /*
2750  * Now it's safe to write buffer to disk. Note that no one else should
2751  * have been able to write it while we were busy with log flushing because
2752  * we have the io_in_progress lock.
2753  */
2754  bufBlock = BufHdrGetBlock(buf);
2755 
2756  /*
2757  * Update page checksum if desired. Since we have only shared lock on the
2758  * buffer, other processes might be updating hint bits in it, so we must
2759  * copy the page to private storage if we do checksumming.
2760  */
2761  bufToWrite = PageSetChecksumCopy((Page) bufBlock, buf->tag.blockNum);
2762 
2763  if (track_io_timing)
2764  INSTR_TIME_SET_CURRENT(io_start);
2765 
2766  /*
2767  * bufToWrite is either the shared buffer or a copy, as appropriate.
2768  */
2769  smgrwrite(reln,
2770  buf->tag.forkNum,
2771  buf->tag.blockNum,
2772  bufToWrite,
2773  false);
2774 
2775  if (track_io_timing)
2776  {
2777  INSTR_TIME_SET_CURRENT(io_time);
2778  INSTR_TIME_SUBTRACT(io_time, io_start);
2781  }
2782 
2784 
2785  /*
2786  * Mark the buffer as clean (unless BM_JUST_DIRTIED has become set) and
2787  * end the io_in_progress state.
2788  */
2789  TerminateBufferIO(buf, true, 0);
2790 
2791  TRACE_POSTGRESQL_BUFFER_FLUSH_DONE(buf->tag.forkNum,
2792  buf->tag.blockNum,
2793  reln->smgr_rnode.node.spcNode,
2794  reln->smgr_rnode.node.dbNode,
2795  reln->smgr_rnode.node.relNode);
2796 
2797  /* Pop the error context stack */
2798  error_context_stack = errcallback.previous;
2799 }
#define BM_PERMANENT
Definition: buf_internals.h:66
ForkNumber forkNum
Definition: buf_internals.h:93
struct timeval instr_time
Definition: instr_time.h:150
char * PageSetChecksumCopy(Page page, BlockNumber blkno)
Definition: bufpage.c:1169
void(* callback)(void *arg)
Definition: elog.h:256
struct ErrorContextCallback * previous
Definition: elog.h:255
void XLogFlush(XLogRecPtr record)
Definition: xlog.c:2804
ErrorContextCallback * error_context_stack
Definition: elog.c:91
long shared_blks_written
Definition: instrument.h:24
static bool StartBufferIO(BufferDesc *buf, bool forInput)
Definition: bufmgr.c:3941
#define INSTR_TIME_SUBTRACT(x, y)
Definition: instr_time.h:170
RelFileNodeBackend smgr_rnode
Definition: smgr.h:42
#define INSTR_TIME_ADD(x, y)
Definition: instr_time.h:158
void smgrwrite(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: smgr.c:530
#define BM_JUST_DIRTIED
Definition: buf_internals.h:63
unsigned int uint32
Definition: c.h:359
SMgrRelation smgropen(RelFileNode rnode, BackendId backend)
Definition: smgr.c:145
static void TerminateBufferIO(BufferDesc *buf, bool clear_dirty, uint32 set_flag_bits)
Definition: bufmgr.c:4008
#define InvalidBackendId
Definition: backendid.h:23
#define BufHdrGetBlock(bufHdr)
Definition: bufmgr.c:58
RelFileNode node
Definition: relfilenode.h:74
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define INSTR_TIME_GET_MICROSEC(t)
Definition: instr_time.h:205
instr_time blk_write_time
Definition: instrument.h:32
#define pgstat_count_buffer_write_time(n)
Definition: pgstat.h:1397
BlockNumber blockNum
Definition: buf_internals.h:94
RelFileNode rnode
Definition: buf_internals.h:92
#define INSTR_TIME_SET_CURRENT(t)
Definition: instr_time.h:156
BufferTag tag
#define UnlockBufHdr(desc, s)
#define BufferGetLSN(bufHdr)
Definition: bufmgr.c:59
static void shared_buffer_write_error_callback(void *arg)
Definition: bufmgr.c:4093
bool track_io_timing
Definition: bufmgr.c:112
Pointer Page
Definition: bufpage.h:78
BufferUsage pgBufferUsage
Definition: instrument.c:20
void * Block
Definition: bufmgr.h:24

◆ FlushDatabaseBuffers()

void FlushDatabaseBuffers ( Oid  dbid)

Definition at line 3312 of file bufmgr.c.

References BM_DIRTY, BM_VALID, BufferDescriptorGetContentLock, CurrentResourceOwner, RelFileNode::dbNode, FlushBuffer(), GetBufferDescriptor, i, LockBufHdr(), LW_SHARED, LWLockAcquire(), LWLockRelease(), NBuffers, PinBuffer_Locked(), ReservePrivateRefCountEntry(), ResourceOwnerEnlargeBuffers(), buftag::rnode, BufferDesc::tag, UnlockBufHdr, and UnpinBuffer().

Referenced by dbase_redo().

3313 {
3314  int i;
3315  BufferDesc *bufHdr;
3316 
3317  /* Make sure we can handle the pin inside the loop */
3319 
3320  for (i = 0; i < NBuffers; i++)
3321  {
3322  uint32 buf_state;
3323 
3324  bufHdr = GetBufferDescriptor(i);
3325 
3326  /*
3327  * As in DropRelFileNodeBuffers, an unlocked precheck should be safe
3328  * and saves some cycles.
3329  */
3330  if (bufHdr->tag.rnode.dbNode != dbid)
3331  continue;
3332 
3334 
3335  buf_state = LockBufHdr(bufHdr);
3336  if (bufHdr->tag.rnode.dbNode == dbid &&
3337  (buf_state & (BM_VALID | BM_DIRTY)) == (BM_VALID | BM_DIRTY))
3338  {
3339  PinBuffer_Locked(bufHdr);
3341  FlushBuffer(bufHdr, NULL);
3343  UnpinBuffer(bufHdr, true);
3344  }
3345  else
3346  UnlockBufHdr(bufHdr, buf_state);
3347  }
3348 }
ResourceOwner CurrentResourceOwner
Definition: resowner.c:142
#define BM_DIRTY
Definition: buf_internals.h:58
static void FlushBuffer(BufferDesc *buf, SMgrRelation reln)
Definition: bufmgr.c:2683
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1726
#define GetBufferDescriptor(id)
unsigned int uint32
Definition: c.h:359
static void UnpinBuffer(BufferDesc *buf, bool fixOwner)
Definition: bufmgr.c:1702
void ResourceOwnerEnlargeBuffers(ResourceOwner owner)
Definition: resowner.c:893
#define BM_VALID
Definition: buf_internals.h:59
#define BufferDescriptorGetContentLock(bdesc)
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
static void PinBuffer_Locked(BufferDesc *buf)
Definition: bufmgr.c:1664
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
RelFileNode rnode
Definition: buf_internals.h:92
BufferTag tag
#define UnlockBufHdr(desc, s)
int i
int NBuffers
Definition: globals.c:131
static void ReservePrivateRefCountEntry(void)
Definition: bufmgr.c:187

◆ FlushOneBuffer()

void FlushOneBuffer ( Buffer  buffer)

Definition at line 3355 of file bufmgr.c.

References Assert, BufferDescriptorGetContentLock, BufferIsLocal, BufferIsPinned, FlushBuffer(), GetBufferDescriptor, and LWLockHeldByMe().

Referenced by hash_xlog_init_bitmap_page(), hash_xlog_init_meta_page(), and XLogReadBufferForRedoExtended().

3356 {
3357  BufferDesc *bufHdr;
3358 
3359  /* currently not needed, but no fundamental reason not to support */
3360  Assert(!BufferIsLocal(buffer));
3361 
3362  Assert(BufferIsPinned(buffer));
3363 
3364  bufHdr = GetBufferDescriptor(buffer - 1);
3365 
3367 
3368  FlushBuffer(bufHdr, NULL);
3369 }
#define BufferIsPinned(bufnum)
Definition: bufmgr.c:420
bool LWLockHeldByMe(LWLock *l)
Definition: lwlock.c:1842
static void FlushBuffer(BufferDesc *buf, SMgrRelation reln)
Definition: bufmgr.c:2683
#define GetBufferDescriptor(id)
#define BufferDescriptorGetContentLock(bdesc)
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37

◆ FlushRelationBuffers()

void FlushRelationBuffers ( Relation  rel)

Definition at line 3214 of file bufmgr.c.

References ErrorContextCallback::arg, buftag::blockNum, BM_DIRTY, BM_JUST_DIRTIED, BM_VALID, BufferDescriptorGetContentLock, ErrorContextCallback::callback, CurrentResourceOwner, error_context_stack, FlushBuffer(), buftag::forkNum, GetBufferDescriptor, GetLocalBufferDescriptor, i, local_buffer_write_error_callback(), LocalBufHdrGetBlock, LockBufHdr(), LW_SHARED, LWLockAcquire(), LWLockRelease(), NBuffers, NLocBuffer, PageSetChecksumInplace(), pg_atomic_read_u32(), pg_atomic_unlocked_write_u32(), PinBuffer_Locked(), ErrorContextCallback::previous, RelationData::rd_node, RelationData::rd_smgr, RelationOpenSmgr, RelationUsesLocalBuffers, RelFileNodeEquals, ReservePrivateRefCountEntry(), ResourceOwnerEnlargeBuffers(), buftag::rnode, smgrwrite(), BufferDesc::state, BufferDesc::tag, UnlockBufHdr, and UnpinBuffer().

Referenced by heap_sync(), heapam_relation_copy_data(), and index_copy_data().

3215 {
3216  int i;
3217  BufferDesc *bufHdr;
3218 
3219  /* Open rel at the smgr level if not already done */
3220  RelationOpenSmgr(rel);
3221 
3222  if (RelationUsesLocalBuffers(rel))
3223  {
3224  for (i = 0; i < NLocBuffer; i++)
3225  {
3226  uint32 buf_state;
3227 
3228  bufHdr = GetLocalBufferDescriptor(i);
3229  if (RelFileNodeEquals(bufHdr->tag.rnode, rel->rd_node) &&
3230  ((buf_state = pg_atomic_read_u32(&bufHdr->state)) &
3231  (BM_VALID | BM_DIRTY)) == (BM_VALID | BM_DIRTY))
3232  {
3233  ErrorContextCallback errcallback;
3234  Page localpage;
3235 
3236  localpage = (char *) LocalBufHdrGetBlock(bufHdr);
3237 
3238  /* Setup error traceback support for ereport() */
3240  errcallback.arg = (void *) bufHdr;
3241  errcallback.previous = error_context_stack;
3242  error_context_stack = &errcallback;
3243 
3244  PageSetChecksumInplace(localpage, bufHdr->tag.blockNum);
3245 
3246  smgrwrite(rel->rd_smgr,
3247  bufHdr->tag.forkNum,
3248  bufHdr->tag.blockNum,
3249  localpage,
3250  false);
3251 
3252  buf_state &= ~(BM_DIRTY | BM_JUST_DIRTIED);
3253  pg_atomic_unlocked_write_u32(&bufHdr->state, buf_state);
3254 
3255  /* Pop the error context stack */
3256  error_context_stack = errcallback.previous;
3257  }
3258  }
3259 
3260  return;
3261  }
3262 
3263  /* Make sure we can handle the pin inside the loop */
3265 
3266  for (i = 0; i < NBuffers; i++)
3267  {
3268  uint32 buf_state;
3269 
3270  bufHdr = GetBufferDescriptor(i);
3271 
3272  /*
3273  * As in DropRelFileNodeBuffers, an unlocked precheck should be safe
3274  * and saves some cycles.
3275  */
3276  if (!RelFileNodeEquals(bufHdr->tag.rnode, rel->rd_node))
3277  continue;
3278 
3280 
3281  buf_state = LockBufHdr(bufHdr);
3282  if (RelFileNodeEquals(bufHdr->tag.rnode, rel->rd_node) &&
3283  (buf_state & (BM_VALID | BM_DIRTY)) == (BM_VALID | BM_DIRTY))
3284  {
3285  PinBuffer_Locked(bufHdr);
3287  FlushBuffer(bufHdr, rel->rd_smgr);
3289  UnpinBuffer(bufHdr, true);
3290  }
3291  else
3292  UnlockBufHdr(bufHdr, buf_state);
3293  }
3294 }
#define LocalBufHdrGetBlock(bufHdr)
Definition: bufmgr.c:62
ForkNumber forkNum
Definition: buf_internals.h:93
static void local_buffer_write_error_callback(void *arg)
Definition: bufmgr.c:4112
ResourceOwner CurrentResourceOwner
Definition: resowner.c:142
struct SMgrRelationData * rd_smgr
Definition: rel.h:57
#define GetLocalBufferDescriptor(id)
#define BM_DIRTY
Definition: buf_internals.h:58
void(* callback)(void *arg)
Definition: elog.h:256
struct ErrorContextCallback * previous
Definition: elog.h:255
static void FlushBuffer(BufferDesc *buf, SMgrRelation reln)
Definition: bufmgr.c:2683
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1726
ErrorContextCallback * error_context_stack
Definition: elog.c:91
#define RelationOpenSmgr(relation)
Definition: rel.h:485
int NLocBuffer
Definition: localbuf.c:41
void smgrwrite(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: smgr.c:530
#define GetBufferDescriptor(id)
#define BM_JUST_DIRTIED
Definition: buf_internals.h:63
unsigned int uint32
Definition: c.h:359
static void UnpinBuffer(BufferDesc *buf, bool fixOwner)
Definition: bufmgr.c:1702
void ResourceOwnerEnlargeBuffers(ResourceOwner owner)
Definition: resowner.c:893
#define BM_VALID
Definition: buf_internals.h:59
RelFileNode rd_node
Definition: rel.h:55
#define BufferDescriptorGetContentLock(bdesc)
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
static void PinBuffer_Locked(BufferDesc *buf)
Definition: bufmgr.c:1664
void PageSetChecksumInplace(Page page, BlockNumber blkno)
Definition: bufpage.c:1198
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
BlockNumber blockNum
Definition: buf_internals.h:94
RelFileNode rnode
Definition: buf_internals.h:92
#define RelationUsesLocalBuffers(relation)
Definition: rel.h:537
BufferTag tag
#define UnlockBufHdr(desc, s)
int i
int NBuffers
Definition: globals.c:131
static void pg_atomic_unlocked_write_u32(volatile pg_atomic_uint32 *ptr, uint32 val)
Definition: atomics.h:277
static void ReservePrivateRefCountEntry(void)
Definition: bufmgr.c:187
pg_atomic_uint32 state
Pointer Page
Definition: bufpage.h:78
#define RelFileNodeEquals(node1, node2)
Definition: relfilenode.h:88
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241

◆ ForgetPrivateRefCountEntry()

static void ForgetPrivateRefCountEntry ( PrivateRefCountEntry ref)
static

Definition at line 382 of file bufmgr.c.

References Assert, PrivateRefCountEntry::buffer, HASH_REMOVE, hash_search(), InvalidBuffer, PrivateRefCountArray, PrivateRefCountOverflowed, PrivateRefCountEntry::refcount, and REFCOUNT_ARRAY_ENTRIES.

Referenced by UnpinBuffer().

383 {
384  Assert(ref->refcount == 0);
385 
386  if (ref >= &PrivateRefCountArray[0] &&
388  {
389  ref->buffer = InvalidBuffer;
390 
391  /*
392  * Mark the just used entry as reserved - in many scenarios that
393  * allows us to avoid ever having to search the array/hash for free
394  * entries.
395  */
396  ReservedRefCountEntry = ref;
397  }
398  else
399  {
400  bool found;
401  Buffer buffer = ref->buffer;
402 
404  (void *) &buffer,
405  HASH_REMOVE,
406  &found);
407  Assert(found);
410  }
411 }
static int32 PrivateRefCountOverflowed
Definition: bufmgr.c:171
#define InvalidBuffer
Definition: buf.h:25
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:906
static struct PrivateRefCountEntry PrivateRefCountArray[REFCOUNT_ARRAY_ENTRIES]
Definition: bufmgr.c:169
static PrivateRefCountEntry * ReservedRefCountEntry
Definition: bufmgr.c:173
#define Assert(condition)
Definition: c.h:739
#define REFCOUNT_ARRAY_ENTRIES
Definition: bufmgr.c:78
static HTAB * PrivateRefCountHash
Definition: bufmgr.c:170
int Buffer
Definition: buf.h:23

◆ GetPrivateRefCount()

static int32 GetPrivateRefCount ( Buffer  buffer)
inlinestatic

Definition at line 359 of file bufmgr.c.

References Assert, BufferIsLocal, BufferIsValid, GetPrivateRefCountEntry(), and PrivateRefCountEntry::refcount.

Referenced by ConditionalLockBufferForCleanup(), DropDatabaseBuffers(), HoldingBufferPinThatDelaysRecovery(), InvalidateBuffer(), IsBufferCleanupOK(), LockBufferForCleanup(), MarkBufferDirtyHint(), and PrintBufferLeakWarning().

360 {
362 
363  Assert(BufferIsValid(buffer));
364  Assert(!BufferIsLocal(buffer));
365 
366  /*
367  * Not moving the entry - that's ok for the current users, but we might
368  * want to change this one day.
369  */
370  ref = GetPrivateRefCountEntry(buffer, false);
371 
372  if (ref == NULL)
373  return 0;
374  return ref->refcount;
375 }
static PrivateRefCountEntry * GetPrivateRefCountEntry(Buffer buffer, bool do_move)
Definition: bufmgr.c:279
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113

◆ GetPrivateRefCountEntry()

static PrivateRefCountEntry * GetPrivateRefCountEntry ( Buffer  buffer,
bool  do_move 
)
static

Definition at line 279 of file bufmgr.c.

References Assert, PrivateRefCountEntry::buffer, BufferIsLocal, BufferIsValid, free, HASH_FIND, HASH_REMOVE, hash_search(), i, InvalidBuffer, PrivateRefCountArray, PrivateRefCountOverflowed, PrivateRefCountEntry::refcount, REFCOUNT_ARRAY_ENTRIES, ReservedRefCountEntry, and ReservePrivateRefCountEntry().

Referenced by GetPrivateRefCount(), IncrBufferRefCount(), PinBuffer(), PinBuffer_Locked(), and UnpinBuffer().

280 {
282  int i;
283 
284  Assert(BufferIsValid(buffer));
285  Assert(!BufferIsLocal(buffer));
286 
287  /*
288  * First search for references in the array, that'll be sufficient in the
289  * majority of cases.
290  */
291  for (i = 0; i < REFCOUNT_ARRAY_ENTRIES; i++)
292  {
293  res = &PrivateRefCountArray[i];
294 
295  if (res->buffer == buffer)
296  return res;
297  }
298 
299  /*
300  * By here we know that the buffer, if already pinned, isn't residing in
301  * the array.
302  *
303  * Only look up the buffer in the hashtable if we've previously overflowed
304  * into it.
305  */
306  if (PrivateRefCountOverflowed == 0)
307  return NULL;
308 
310  (void *) &buffer,
311  HASH_FIND,
312  NULL);
313 
314  if (res == NULL)
315  return NULL;
316  else if (!do_move)
317  {
318  /* caller doesn't want us to move the hash entry into the array */
319  return res;
320  }
321  else
322  {
323  /* move buffer from hashtable into the free array slot */
324  bool found;
326 
327  /* Ensure there's a free array slot */
329 
330  /* Use up the reserved slot */
331  Assert(ReservedRefCountEntry != NULL);
332  free = ReservedRefCountEntry;
333  ReservedRefCountEntry = NULL;
334  Assert(free->buffer == InvalidBuffer);
335 
336  /* and fill it */
337  free->buffer = buffer;
338  free->refcount = res->refcount;
339 
340  /* delete from hashtable */
342  (void *) &buffer,
343  HASH_REMOVE,
344  &found);
345  Assert(found);
348 
349  return free;
350  }
351 }
static int32 PrivateRefCountOverflowed
Definition: bufmgr.c:171
#define InvalidBuffer
Definition: buf.h:25
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:906
static struct PrivateRefCountEntry PrivateRefCountArray[REFCOUNT_ARRAY_ENTRIES]
Definition: bufmgr.c:169
static PrivateRefCountEntry * ReservedRefCountEntry
Definition: bufmgr.c:173
#define free(a)
Definition: header.h:65
#define Assert(condition)
Definition: c.h:739
#define REFCOUNT_ARRAY_ENTRIES
Definition: bufmgr.c:78
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
static HTAB * PrivateRefCountHash
Definition: bufmgr.c:170
int i
static void ReservePrivateRefCountEntry(void)
Definition: bufmgr.c:187

◆ HoldingBufferPinThatDelaysRecovery()

bool HoldingBufferPinThatDelaysRecovery ( void  )

Definition at line 3758 of file bufmgr.c.

References GetPrivateRefCount(), and GetStartupBufferPinWaitBufId().

Referenced by CheckRecoveryConflictDeadlock(), and RecoveryConflictInterrupt().

3759 {
3760  int bufid = GetStartupBufferPinWaitBufId();
3761 
3762  /*
3763  * If we get woken slowly then it's possible that the Startup process was
3764  * already woken by other backends before we got here. Also possible that
3765  * we get here by multiple interrupts or interrupts at inappropriate
3766  * times, so make sure we do nothing if the bufid is not set.
3767  */
3768  if (bufid < 0)
3769  return false;
3770 
3771  if (GetPrivateRefCount(bufid + 1) > 0)
3772  return true;
3773 
3774  return false;
3775 }
static int32 GetPrivateRefCount(Buffer buffer)
Definition: bufmgr.c:359
int GetStartupBufferPinWaitBufId(void)
Definition: proc.c:657

◆ IncrBufferRefCount()

void IncrBufferRefCount ( Buffer  buffer)

Definition at line 3413 of file bufmgr.c.

References Assert, BufferIsLocal, BufferIsPinned, CurrentResourceOwner, GetPrivateRefCountEntry(), LocalRefCount, PrivateRefCountEntry::refcount, ResourceOwnerEnlargeBuffers(), and ResourceOwnerRememberBuffer().

Referenced by _bt_steppage(), btrestrpos(), entryLoadMoreItems(), ReadBufferBI(), scanPostingTree(), startScanEntry(), and tts_buffer_heap_store_tuple().

3414 {
3415  Assert(BufferIsPinned(buffer));
3417  if (BufferIsLocal(buffer))
3418  LocalRefCount[-buffer - 1]++;
3419  else
3420  {
3421  PrivateRefCountEntry *ref;
3422 
3423  ref = GetPrivateRefCountEntry(buffer, true);
3424  Assert(ref != NULL);
3425  ref->refcount++;
3426  }
3428 }
static PrivateRefCountEntry * GetPrivateRefCountEntry(Buffer buffer, bool do_move)
Definition: bufmgr.c:279
#define BufferIsPinned(bufnum)
Definition: bufmgr.c:420
ResourceOwner CurrentResourceOwner
Definition: resowner.c:142
void ResourceOwnerRememberBuffer(ResourceOwner owner, Buffer buffer)
Definition: resowner.c:906
void ResourceOwnerEnlargeBuffers(ResourceOwner owner)
Definition: resowner.c:893
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
int32 * LocalRefCount
Definition: localbuf.c:45

◆ InitBufferPoolAccess()

void InitBufferPoolAccess ( void  )

Definition at line 2455 of file bufmgr.c.

References HASHCTL::entrysize, HASH_BLOBS, hash_create(), HASH_ELEM, HASHCTL::keysize, MemSet, and PrivateRefCountArray.

Referenced by BaseInit().

2456 {
2457  HASHCTL hash_ctl;
2458 
2459  memset(&PrivateRefCountArray, 0, sizeof(PrivateRefCountArray));
2460 
2461  MemSet(&hash_ctl, 0, sizeof(hash_ctl));
2462  hash_ctl.keysize = sizeof(int32);
2463  hash_ctl.entrysize = sizeof(PrivateRefCountEntry);
2464 
2465  PrivateRefCountHash = hash_create("PrivateRefCount", 100, &hash_ctl,
2466  HASH_ELEM | HASH_BLOBS);
2467 }
struct PrivateRefCountEntry PrivateRefCountEntry
#define HASH_ELEM
Definition: hsearch.h:87
Size entrysize
Definition: hsearch.h:73
#define MemSet(start, val, len)
Definition: c.h:962
signed int int32
Definition: c.h:347
static struct PrivateRefCountEntry PrivateRefCountArray[REFCOUNT_ARRAY_ENTRIES]
Definition: bufmgr.c:169
#define HASH_BLOBS
Definition: hsearch.h:88
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:316
Size keysize
Definition: hsearch.h:72
static HTAB * PrivateRefCountHash
Definition: bufmgr.c:170

◆ InitBufferPoolBackend()

void InitBufferPoolBackend ( void  )

Definition at line 2479 of file bufmgr.c.

References AtProcExit_Buffers(), and on_shmem_exit().

Referenced by AuxiliaryProcessMain(), and InitPostgres().

2480 {
2482 }
void on_shmem_exit(pg_on_exit_callback function, Datum arg)
Definition: ipc.c:361
static void AtProcExit_Buffers(int code, Datum arg)
Definition: bufmgr.c:2489

◆ InvalidateBuffer()

static void InvalidateBuffer ( BufferDesc buf)
static

Definition at line 1360 of file bufmgr.c.

References Assert, BM_LOCKED, BM_TAG_VALID, BUF_FLAG_MASK, BUF_STATE_GET_REFCOUNT, BUF_USAGECOUNT_MASK, BufferDescriptorGetBuffer, BUFFERTAGS_EQUAL, BufMappingPartitionLock, BufTableDelete(), BufTableHashCode(), CLEAR_BUFFERTAG, elog, ERROR, GetPrivateRefCount(), LockBufHdr(), LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), pg_atomic_read_u32(), BufferDesc::state, StrategyFreeBuffer(), BufferDesc::tag, UnlockBufHdr, and WaitIO().

Referenced by DropDatabaseBuffers(), DropRelFileNodeBuffers(), and DropRelFileNodesAllBuffers().

1361 {
1362  BufferTag oldTag;
1363  uint32 oldHash; /* hash value for oldTag */
1364  LWLock *oldPartitionLock; /* buffer partition lock for it */
1365  uint32 oldFlags;
1366  uint32 buf_state;
1367 
1368  /* Save the original buffer tag before dropping the spinlock */
1369  oldTag = buf->tag;
1370 
1371  buf_state = pg_atomic_read_u32(&buf->state);
1372  Assert(buf_state & BM_LOCKED);
1373  UnlockBufHdr(buf, buf_state);
1374 
1375  /*
1376  * Need to compute the old tag's hashcode and partition lock ID. XXX is it
1377  * worth storing the hashcode in BufferDesc so we need not recompute it
1378  * here? Probably not.
1379  */
1380  oldHash = BufTableHashCode(&oldTag);
1381  oldPartitionLock = BufMappingPartitionLock(oldHash);
1382 
1383 retry:
1384 
1385  /*
1386  * Acquire exclusive mapping lock in preparation for changing the buffer's
1387  * association.
1388  */
1389  LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
1390 
1391  /* Re-lock the buffer header */
1392  buf_state = LockBufHdr(buf);
1393 
1394  /* If it's changed while we were waiting for lock, do nothing */
1395  if (!BUFFERTAGS_EQUAL(buf->tag, oldTag))
1396  {
1397  UnlockBufHdr(buf, buf_state);
1398  LWLockRelease(oldPartitionLock);
1399  return;
1400  }
1401 
1402  /*
1403  * We assume the only reason for it to be pinned is that someone else is
1404  * flushing the page out. Wait for them to finish. (This could be an
1405  * infinite loop if the refcount is messed up... it would be nice to time
1406  * out after awhile, but there seems no way to be sure how many loops may
1407  * be needed. Note that if the other guy has pinned the buffer but not
1408  * yet done StartBufferIO, WaitIO will fall through and we'll effectively
1409  * be busy-looping here.)
1410  */
1411  if (BUF_STATE_GET_REFCOUNT(buf_state) != 0)
1412  {
1413  UnlockBufHdr(buf, buf_state);
1414  LWLockRelease(oldPartitionLock);
1415  /* safety check: should definitely not be our *own* pin */
1417  elog(ERROR, "buffer is pinned in InvalidateBuffer");
1418  WaitIO(buf);
1419  goto retry;
1420  }
1421 
1422  /*
1423  * Clear out the buffer's tag and flags. We must do this to ensure that
1424  * linear scans of the buffer array don't think the buffer is valid.
1425  */
1426  oldFlags = buf_state & BUF_FLAG_MASK;
1427  CLEAR_BUFFERTAG(buf->tag);
1428  buf_state &= ~(BUF_FLAG_MASK | BUF_USAGECOUNT_MASK);
1429  UnlockBufHdr(buf, buf_state);
1430 
1431  /*
1432  * Remove the buffer from the lookup hashtable, if it was in there.
1433  */
1434  if (oldFlags & BM_TAG_VALID)
1435  BufTableDelete(&oldTag, oldHash);
1436 
1437  /*
1438  * Done with mapping lock.
1439  */
1440  LWLockRelease(oldPartitionLock);
1441 
1442  /*
1443  * Insert the buffer at the head of the list of free buffers.
1444  */
1445  StrategyFreeBuffer(buf);
1446 }
void BufTableDelete(BufferTag *tagPtr, uint32 hashcode)
Definition: buf_table.c:149
Definition: lwlock.h:32
#define BufMappingPartitionLock(hashcode)
#define BM_TAG_VALID
Definition: buf_internals.h:60
static void WaitIO(BufferDesc *buf)
Definition: bufmgr.c:3894
uint32 BufTableHashCode(BufferTag *tagPtr)
Definition: buf_table.c:79
void StrategyFreeBuffer(BufferDesc *buf)
Definition: freelist.c:364
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1726
#define ERROR
Definition: elog.h:43
#define BUF_FLAG_MASK
Definition: buf_internals.h:45
static int32 GetPrivateRefCount(Buffer buffer)
Definition: bufmgr.c:359
unsigned int uint32
Definition: c.h:359
#define BUFFERTAGS_EQUAL(a, b)
#define BM_LOCKED
Definition: buf_internals.h:57
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
#define Assert(condition)
Definition: c.h:739
#define CLEAR_BUFFERTAG(a)
Definition: buf_internals.h:97
#define BUF_USAGECOUNT_MASK
Definition: buf_internals.h:42
#define BufferDescriptorGetBuffer(bdesc)
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
BufferTag tag
#define UnlockBufHdr(desc, s)
#define elog(elevel,...)
Definition: elog.h:228
pg_atomic_uint32 state
#define BUF_STATE_GET_REFCOUNT(state)
Definition: buf_internals.h:48
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241

◆ IsBufferCleanupOK()

bool IsBufferCleanupOK ( Buffer  buffer)

Definition at line 3840 of file bufmgr.c.

References Assert, BUF_STATE_GET_REFCOUNT, BufferDescriptorGetContentLock, BufferIsLocal, BufferIsValid, GetBufferDescriptor, GetPrivateRefCount(), LocalRefCount, LockBufHdr(), LW_EXCLUSIVE, LWLockHeldByMeInMode(), and UnlockBufHdr.

Referenced by _hash_doinsert(), _hash_expandtable(), _hash_splitbucket(), hash_xlog_split_allocate_page(), and hashbucketcleanup().

3841 {
3842  BufferDesc *bufHdr;
3843  uint32 buf_state;
3844 
3845  Assert(BufferIsValid(buffer));
3846 
3847  if (BufferIsLocal(buffer))
3848  {
3849  /* There should be exactly one pin */
3850  if (LocalRefCount[-buffer - 1] != 1)
3851  return false;
3852  /* Nobody else to wait for */
3853  return true;
3854  }
3855 
3856  /* There should be exactly one local pin */
3857  if (GetPrivateRefCount(buffer) != 1)
3858  return false;
3859 
3860  bufHdr = GetBufferDescriptor(buffer - 1);
3861 
3862  /* caller must hold exclusive lock on buffer */
3864  LW_EXCLUSIVE));
3865 
3866  buf_state = LockBufHdr(bufHdr);
3867 
3868  Assert(BUF_STATE_GET_REFCOUNT(buf_state) > 0);
3869  if (BUF_STATE_GET_REFCOUNT(buf_state) == 1)
3870  {
3871  /* pincount is OK. */
3872  UnlockBufHdr(bufHdr, buf_state);
3873  return true;
3874  }
3875 
3876  UnlockBufHdr(bufHdr, buf_state);
3877  return false;
3878 }
bool LWLockHeldByMeInMode(LWLock *l, LWLockMode mode)
Definition: lwlock.c:1860
#define GetBufferDescriptor(id)
static int32 GetPrivateRefCount(Buffer buffer)
Definition: bufmgr.c:359
unsigned int uint32
Definition: c.h:359
#define BufferDescriptorGetContentLock(bdesc)
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
#define UnlockBufHdr(desc, s)
int32 * LocalRefCount
Definition: localbuf.c:45
#define BUF_STATE_GET_REFCOUNT(state)
Definition: buf_internals.h:48

◆ IssuePendingWritebacks()

void IssuePendingWritebacks ( WritebackContext context)

Definition at line 4344 of file bufmgr.c.

References buftag::blockNum, buffertag_comparator(), cur, buftag::forkNum, i, InvalidBackendId, next, WritebackContext::nr_pending, WritebackContext::pending_writebacks, qsort, RelFileNodeEquals, buftag::rnode, smgropen(), smgrwriteback(), and PendingWriteback::tag.

Referenced by BufferSync(), and ScheduleBufferTagForWriteback().

4345 {
4346  int i;
4347 
4348  if (context->nr_pending == 0)
4349  return;
4350 
4351  /*
4352  * Executing the writes in-order can make them a lot faster, and allows to
4353  * merge writeback requests to consecutive blocks into larger writebacks.
4354  */
4355  qsort(&context->pending_writebacks, context->nr_pending,
4357 
4358  /*
4359  * Coalesce neighbouring writes, but nothing else. For that we iterate
4360  * through the, now sorted, array of pending flushes, and look forward to
4361  * find all neighbouring (or identical) writes.
4362  */
4363  for (i = 0; i < context->nr_pending; i++)
4364  {
4367  SMgrRelation reln;
4368  int ahead;
4369  BufferTag tag;
4370  Size nblocks = 1;
4371 
4372  cur = &context->pending_writebacks[i];
4373  tag = cur->tag;
4374 
4375  /*
4376  * Peek ahead, into following writeback requests, to see if they can
4377  * be combined with the current one.
4378  */
4379  for (ahead = 0; i + ahead + 1 < context->nr_pending; ahead++)
4380  {
4381  next = &context->pending_writebacks[i + ahead + 1];
4382 
4383  /* different file, stop */
4384  if (!RelFileNodeEquals(cur->tag.rnode, next->tag.rnode) ||
4385  cur->tag.forkNum != next->tag.forkNum)
4386  break;
4387 
4388  /* ok, block queued twice, skip */
4389  if (cur->tag.blockNum == next->tag.blockNum)
4390  continue;
4391 
4392  /* only merge consecutive writes */
4393  if (cur->tag.blockNum + 1 != next->tag.blockNum)
4394  break;
4395 
4396  nblocks++;
4397  cur = next;
4398  }
4399 
4400  i += ahead;
4401 
4402  /* and finally tell the kernel to write the data to storage */
4403  reln = smgropen(tag.rnode, InvalidBackendId);
4404  smgrwriteback(reln, tag.forkNum, tag.blockNum, nblocks);
4405  }
4406 
4407  context->nr_pending = 0;
4408 }
static int32 next
Definition: blutils.c:217
ForkNumber forkNum
Definition: buf_internals.h:93
struct cursor * cur
Definition: ecpg.c:28
PendingWriteback pending_writebacks[WRITEBACK_MAX_PENDING_FLUSHES]
void smgrwriteback(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, BlockNumber nblocks)
Definition: smgr.c:543
static int buffertag_comparator(const void *p1, const void *p2)
Definition: bufmgr.c:4210
SMgrRelation smgropen(RelFileNode rnode, BackendId backend)
Definition: smgr.c:145
#define InvalidBackendId
Definition: backendid.h:23
size_t Size
Definition: c.h:467
BlockNumber blockNum
Definition: buf_internals.h:94
RelFileNode rnode
Definition: buf_internals.h:92
int i
#define qsort(a, b, c, d)
Definition: port.h:491
#define RelFileNodeEquals(node1, node2)
Definition: relfilenode.h:88

◆ local_buffer_write_error_callback()

static void local_buffer_write_error_callback ( void *  arg)
static

Definition at line 4112 of file bufmgr.c.

References buftag::blockNum, errcontext, buftag::forkNum, MyBackendId, pfree(), relpathbackend, buftag::rnode, and BufferDesc::tag.

Referenced by FlushRelationBuffers().

4113 {
4114  BufferDesc *bufHdr = (BufferDesc *) arg;
4115 
4116  if (bufHdr != NULL)
4117  {
4118  char *path = relpathbackend(bufHdr->tag.rnode, MyBackendId,
4119  bufHdr->tag.forkNum);
4120 
4121  errcontext("writing block %u of relation %s",
4122  bufHdr->tag.blockNum, path);
4123  pfree(path);
4124  }
4125 }
BackendId MyBackendId
Definition: globals.c:81
ForkNumber forkNum
Definition: buf_internals.h:93
void pfree(void *pointer)
Definition: mcxt.c:1056
BlockNumber blockNum
Definition: buf_internals.h:94
RelFileNode rnode
Definition: buf_internals.h:92
BufferTag tag
#define errcontext
Definition: elog.h:183
void * arg
#define relpathbackend(rnode, backend, forknum)
Definition: relpath.h:78

◆ LockBuffer()

void LockBuffer ( Buffer  buffer,
int  mode 
)

Definition at line 3612 of file bufmgr.c.

References Assert, buf, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_SHARE, BUFFER_LOCK_UNLOCK, BufferDescriptorGetContentLock, BufferIsLocal, BufferIsValid, elog, ERROR, GetBufferDescriptor, LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), and LWLockRelease().

Referenced by _bt_drop_lock_and_maybe_pin(), _bt_endpoint(), _bt_first(), _bt_getbuf(), _bt_getroot(), _bt_killitems(), _bt_moveright(), _bt_pagedel(), _bt_readnextpage(), _bt_relandgetbuf(), _bt_search(), _bt_unlink_halfdead_page(), _bt_update_meta_cleanup_info(), _hash_addovflpage(), _hash_doinsert(), _hash_expandtable(), _hash_finish_split(), _hash_first(), _hash_freeovflpage(), _hash_getbuf(), _hash_getbuf_with_strategy(), _hash_getcachedmetap(), _hash_getnewbuf(), _hash_init(), _hash_kill_items(), _hash_readnext(), _hash_readpage(), _hash_readprev(), _hash_splitbucket(), _hash_squeezebucket(), _hash_vacuum_one_page(), blbulkdelete(), blgetbitmap(), blinsert(), BloomNewBuffer(), blvacuumcleanup(), brin_doinsert(), brin_doupdate(), brin_evacuate_page(), brin_getinsertbuffer(), brin_page_cleanup(), brinbuild(), brinbuildempty(), bringetbitmap(), brinGetStats(), brinGetTupleForHeapBlock(), brininsert(), brinLockRevmapPageForUpdate(), brinRevmapDesummarizeRange(), brinRevmapInitialize(), brinsummarize(), bt_metap(), bt_page_items(), bt_page_stats(), btvacuumpage(), checkXLogConsistency(), collect_corrupt_items(), collect_visibility_data(), collectMatchBitmap(), ConditionalLockBufferForCleanup(), count_nondeletable_pages(), entryLoadMoreItems(), fill_seq_with_data(), FreeSpaceMapPrepareTruncateRel(), fsm_readbuf(), fsm_search(), fsm_search_avail(), fsm_set_and_search(), fsm_vacuum_page(), get_raw_page_internal(), GetVisibilityMapPins(), ginbuildempty(), ginbulkdelete(), ginEntryInsert(), ginFindLeafPage(), ginFindParents(), ginFinishSplit(), ginGetStats(), ginHeapTupleFastInsert(), ginInsertCleanup(), ginInsertValue(), GinNewBuffer(), ginScanToDelete(), ginStepRight(), ginTraverseLock(), ginUpdateStats(), ginvacuumcleanup(), ginVacuumPostingTreeLeaves(), gistBufferingFindCorrectParent(), gistbufferinginserttuples(), gistbuildempty(), gistdoinsert(), gistFindCorrectParent(), gistFindPath(), gistfinishsplit(), gistfixsplit(), gistformdownlink(), gistGetMaxLevel(), gistinserttuples(), gistkillitems(), gistNewBuffer(), gistProcessItup(), gistScanPage(), gistvacuum_delete_empty_pages(), gistvacuumpage(), hashbucketcleanup(), hashbulkdelete(), heap_abort_speculative(), heap_compute_xid_horizon_for_tuples(), heap_delete(), heap_fetch(), heap_finish_speculative(), heap_get_latest_tid(), heap_inplace_update(), heap_lock_tuple(), heap_lock_updated_tuple_rec(), heap_page_prune_opt(), heap_update(), heap_xlog_visible(), heapam_index_build_range_scan(), heapam_index_fetch_tuple(), heapam_index_validate_scan(), heapam_relation_copy_for_cluster(), heapam_scan_analyze_next_block(), heapam_scan_bitmap_next_block(), heapam_scan_sample_next_tuple(), heapam_tuple_satisfies_snapshot(), heapgetpage(), heapgettup(), initBloomState(), lazy_scan_heap(), LockBufferForCleanup(), log_newpage_range(), palloc_btree_page(), pg_visibility(), pgrowlocks(), pgstat_btree_page(), pgstat_gist_page(), pgstat_heap(), pgstatginindex_internal(), pgstathashindex(), pgstatindex_impl(), read_seq_tuple(), RelationGetBufferForTuple(), revmap_physical_extend(), scanGetCandidate(), scanPendingInsert(), shiftList(), spgdoinsert(), spgGetCache(), SpGistNewBuffer(), spgprocesspending(), spgvacuumpage(), spgWalk(), startScanEntry(), statapprox_heap(), summarize_range(), UnlockReleaseBuffer(), visibilitymap_clear(), visibilitymap_prepare_truncate(), visibilitymap_set(), vm_readbuf(), XLogReadBufferExtended(), XLogReadBufferForRedoExtended(), and XLogRecordPageWithFreeSpace().

3613 {
3614  BufferDesc *buf;
3615 
3616  Assert(BufferIsValid(buffer));
3617  if (BufferIsLocal(buffer))
3618  return; /* local buffers need no lock */
3619 
3620  buf = GetBufferDescriptor(buffer - 1);
3621 
3622  if (mode == BUFFER_LOCK_UNLOCK)
3624  else if (mode == BUFFER_LOCK_SHARE)
3626  else if (mode == BUFFER_LOCK_EXCLUSIVE)
3628  else
3629  elog(ERROR, "unrecognized buffer lock mode: %d", mode);
3630 }
static PgChecksumMode mode
Definition: pg_checksums.c:61
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:86
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:88
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1726
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
#define GetBufferDescriptor(id)
#define BufferDescriptorGetContentLock(bdesc)
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
#define elog(elevel,...)
Definition: elog.h:228
#define BUFFER_LOCK_SHARE
Definition: bufmgr.h:87

◆ LockBufferForCleanup()

void LockBufferForCleanup ( Buffer  buffer)

Definition at line 3669 of file bufmgr.c.

References Assert, BM_PIN_COUNT_WAITER, BUF_STATE_GET_REFCOUNT, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferIsLocal, BufferIsValid, elog, ERROR, GetBufferDescriptor, GetPrivateRefCount(), InHotStandby, LocalRefCount, LockBuffer(), LockBufHdr(), MyProcPid, PG_WAIT_BUFFER_PIN, ProcWaitForSignal(), ResolveRecoveryConflictWithBufferPin(), SetStartupBufferPinWaitBufId(), UnlockBufHdr, and BufferDesc::wait_backend_pid.

Referenced by btvacuumpage(), ginVacuumPostingTree(), hashbulkdelete(), lazy_scan_heap(), ReadBuffer_common(), and XLogReadBufferForRedoExtended().

3670 {
3671  BufferDesc *bufHdr;
3672 
3673  Assert(BufferIsValid(buffer));
3674  Assert(PinCountWaitBuf == NULL);
3675 
3676  if (BufferIsLocal(buffer))
3677  {
3678  /* There should be exactly one pin */
3679  if (LocalRefCount[-buffer - 1] != 1)
3680  elog(ERROR, "incorrect local pin count: %d",
3681  LocalRefCount[-buffer - 1]);
3682  /* Nobody else to wait for */
3683  return;
3684  }
3685 
3686  /* There should be exactly one local pin */
3687  if (GetPrivateRefCount(buffer) != 1)
3688  elog(ERROR, "incorrect local pin count: %d",
3689  GetPrivateRefCount(buffer));
3690 
3691  bufHdr = GetBufferDescriptor(buffer - 1);
3692 
3693  for (;;)
3694  {
3695  uint32 buf_state;
3696 
3697  /* Try to acquire lock */
3699  buf_state = LockBufHdr(bufHdr);
3700 
3701  Assert(BUF_STATE_GET_REFCOUNT(buf_state) > 0);
3702  if (BUF_STATE_GET_REFCOUNT(buf_state) == 1)
3703  {
3704  /* Successfully acquired exclusive lock with pincount 1 */
3705  UnlockBufHdr(bufHdr, buf_state);
3706  return;
3707  }
3708  /* Failed, so mark myself as waiting for pincount 1 */
3709  if (buf_state & BM_PIN_COUNT_WAITER)
3710  {
3711  UnlockBufHdr(bufHdr, buf_state);
3712  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3713  elog(ERROR, "multiple backends attempting to wait for pincount 1");
3714  }
3715  bufHdr->wait_backend_pid = MyProcPid;
3716  PinCountWaitBuf = bufHdr;
3717  buf_state |= BM_PIN_COUNT_WAITER;
3718  UnlockBufHdr(bufHdr, buf_state);
3719  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3720 
3721  /* Wait to be signaled by UnpinBuffer() */
3722  if (InHotStandby)
3723  {
3724  /* Publish the bufid that Startup process waits on */
3725  SetStartupBufferPinWaitBufId(buffer - 1);
3726  /* Set alarm and then wait to be signaled by UnpinBuffer() */
3728  /* Reset the published bufid */
3730  }
3731  else
3733 
3734  /*
3735  * Remove flag marking us as waiter. Normally this will not be set
3736  * anymore, but ProcWaitForSignal() can return for other signals as
3737  * well. We take care to only reset the flag if we're the waiter, as
3738  * theoretically another backend could have started waiting. That's
3739  * impossible with the current usages due to table level locking, but
3740  * better be safe.
3741  */
3742  buf_state = LockBufHdr(bufHdr);
3743  if ((buf_state & BM_PIN_COUNT_WAITER) != 0 &&
3744  bufHdr->wait_backend_pid == MyProcPid)
3745  buf_state &= ~BM_PIN_COUNT_WAITER;
3746  UnlockBufHdr(bufHdr, buf_state);
3747 
3748  PinCountWaitBuf = NULL;
3749  /* Loop back and try again */
3750  }
3751 }
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:86
int MyProcPid
Definition: globals.c:40
int wait_backend_pid
#define InHotStandby
Definition: xlog.h:74
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:88
#define ERROR
Definition: elog.h:43
void ResolveRecoveryConflictWithBufferPin(void)
Definition: standby.c:461
void SetStartupBufferPinWaitBufId(int bufid)
Definition: proc.c:645
#define GetBufferDescriptor(id)
static int32 GetPrivateRefCount(Buffer buffer)
Definition: bufmgr.c:359
unsigned int uint32
Definition: c.h:359
void ProcWaitForSignal(uint32 wait_event_info)
Definition: proc.c:1794
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3612
#define PG_WAIT_BUFFER_PIN
Definition: pgstat.h:756
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
#define UnlockBufHdr(desc, s)
#define elog(elevel,...)
Definition: elog.h:228
static BufferDesc * PinCountWaitBuf
Definition: bufmgr.c:137
int32 * LocalRefCount
Definition: localbuf.c:45
#define BUF_STATE_GET_REFCOUNT(state)
Definition: buf_internals.h:48
#define BM_PIN_COUNT_WAITER
Definition: buf_internals.h:64

◆ LockBufHdr()

uint32 LockBufHdr ( BufferDesc desc)

Definition at line 4158 of file bufmgr.c.

References BM_LOCKED, finish_spin_delay(), init_local_spin_delay, perform_spin_delay(), pg_atomic_fetch_or_u32(), and BufferDesc::state.

Referenced by AbortBufferIO(), apw_dump_now(), BufferAlloc(), BufferGetLSNAtomic(), BufferSync(), ConditionalLockBufferForCleanup(), DropDatabaseBuffers(), DropRelFileNodeBuffers(), DropRelFileNodesAllBuffers(), FlushBuffer(), FlushDatabaseBuffers(), FlushRelationBuffers(), GetBufferFromRing(), InvalidateBuffer(), IsBufferCleanupOK(), LockBufferForCleanup(), MarkBufferDirtyHint(), pg_buffercache_pages(), ReadBuffer_common(), StartBufferIO(), StrategyGetBuffer(), SyncOneBuffer(), TerminateBufferIO(), UnlockBuffers(), UnpinBuffer(), and WaitIO().

4159 {
4160  SpinDelayStatus delayStatus;
4161  uint32 old_buf_state;
4162 
4163  init_local_spin_delay(&delayStatus);
4164 
4165  while (true)
4166  {
4167  /* set BM_LOCKED flag */
4168  old_buf_state = pg_atomic_fetch_or_u32(&desc->state, BM_LOCKED);
4169  /* if it wasn't set before we're OK */
4170  if (!(old_buf_state & BM_LOCKED))
4171  break;
4172  perform_spin_delay(&delayStatus);
4173  }
4174  finish_spin_delay(&delayStatus);
4175  return old_buf_state | BM_LOCKED;
4176 }
#define init_local_spin_delay(status)
Definition: s_lock.h:1043
void finish_spin_delay(SpinDelayStatus *status)
Definition: s_lock.c:174
unsigned int uint32
Definition: c.h:359
#define BM_LOCKED
Definition: buf_internals.h:57
pg_atomic_uint32 state
static uint32 pg_atomic_fetch_or_u32(volatile pg_atomic_uint32 *ptr, uint32 or_)
Definition: atomics.h:372
void perform_spin_delay(SpinDelayStatus *status)
Definition: s_lock.c:124

◆ MarkBufferDirty()

void MarkBufferDirty ( Buffer  buffer)

Definition at line 1458 of file bufmgr.c.

References Assert, BM_DIRTY, BM_JUST_DIRTIED, BM_LOCKED, BUF_STATE_GET_REFCOUNT, BufferDescriptorGetContentLock, BufferIsLocal, BufferIsPinned, BufferIsValid, elog, ERROR, GetBufferDescriptor, LW_EXCLUSIVE, LWLockHeldByMeInMode(), MarkLocalBufferDirty(), pg_atomic_compare_exchange_u32(), pg_atomic_read_u32(), pgBufferUsage, BufferUsage::shared_blks_dirtied, BufferDesc::state, VacuumCostActive, VacuumCostBalance, VacuumCostPageDirty, VacuumPageDirty, and WaitBufHdrUnlocked().

Referenced by _bt_clear_incomplete_split(), _bt_delitems_delete(), _bt_delitems_vacuum(), _bt_getroot(), _bt_insertonpg(), _bt_mark_page_halfdead(), _bt_newroot(), _bt_restore_meta(), _bt_split(), _bt_unlink_halfdead_page(), _bt_update_meta_cleanup_info(), _hash_addovflpage(), _hash_doinsert(), _hash_expandtable(), _hash_freeovflpage(), _hash_init(), _hash_splitbucket(), _hash_squeezebucket(), _hash_vacuum_one_page(), addLeafTuple(), brin_doinsert(), brin_doupdate(), brin_initialize_empty_new_buffer(), brin_xlog_createidx(), brin_xlog_desummarize_page(), brin_xlog_insert_update(), brin_xlog_revmap_extend(), brin_xlog_samepage_update(), brin_xlog_update(), brinbuild(), brinbuildempty(), brinRevmapDesummarizeRange(), btree_xlog_delete(), btree_xlog_insert(), btree_xlog_mark_page_halfdead(), btree_xlog_newroot(), btree_xlog_split(), btree_xlog_unlink_page(), btree_xlog_vacuum(), createPostingTree(), do_setval(), doPickSplit(), fill_seq_with_data(), FreeSpaceMapPrepareTruncateRel(), generic_redo(), GenericXLogFinish(), ginbuild(), ginbuildempty(), ginbulkdelete(), ginDeletePage(), ginHeapTupleFastInsert(), ginPlaceToPage(), ginRedoClearIncompleteSplit(), ginRedoCreatePTree(), ginRedoDeleteListPages(), ginRedoDeletePage(), ginRedoInsert(), ginRedoInsertListPage(), ginRedoUpdateMetapage(), ginRedoVacuumDataLeafPage(), ginUpdateStats(), ginVacuumPostingTreeLeaf(), gistbuild(), gistbuildempty(), gistdeletepage(), gistplacetopage(), gistprunepage(), gistRedoClearFollowRight(), gistRedoDeleteRecord(), gistRedoPageDelete(), gistRedoPageSplitRecord(), gistRedoPageUpdateRecord(), gistvacuumpage(), hash_xlog_add_ovfl_page(), hash_xlog_delete(), hash_xlog_init_bitmap_page(), hash_xlog_init_meta_page(), hash_xlog_insert(), hash_xlog_move_page_contents(), hash_xlog_split_allocate_page(), hash_xlog_split_cleanup(), hash_xlog_split_complete(), hash_xlog_squeeze_page(), hash_xlog_update_meta_page(), hash_xlog_vacuum_one_page(), hashbucketcleanup(), hashbulkdelete(), heap_abort_speculative(), heap_delete(), heap_finish_speculative(), heap_inplace_update(), heap_insert(), heap_lock_tuple(), heap_lock_updated_tuple_rec(), heap_multi_insert(), heap_page_prune(), heap_update(), heap_xlog_clean(), heap_xlog_confirm(), heap_xlog_delete(), heap_xlog_freeze_page(), heap_xlog_inplace(), heap_xlog_insert(), heap_xlog_lock(), heap_xlog_lock_updated(), heap_xlog_multi_insert(), heap_xlog_update(), heap_xlog_visible(), lazy_scan_heap(), lazy_vacuum_page(), log_newpage_range(), moveLeafs(), nextval_internal(), RelationGetBufferForTuple(), revmap_physical_extend(), saveNodeLink(), seq_redo(), shiftList(), spgAddNodeAction(), spgbuild(), SpGistUpdateMetaPage(), spgRedoAddLeaf(), spgRedoAddNode(), spgRedoMoveLeafs(), spgRedoPickSplit(), spgRedoSplitTuple(), spgRedoVacuumLeaf(), spgRedoVacuumRedirect(), spgRedoVacuumRoot(), spgSplitNodeAction(), vacuumLeafPage(), vacuumLeafRoot(), vacuumRedirectAndPlaceholder(), visibilitymap_clear(), visibilitymap_prepare_truncate(), visibilitymap_set(), writeListPage(), and XLogReadBufferForRedoExtended().

1459 {
1460  BufferDesc *bufHdr;
1461  uint32 buf_state;
1462  uint32 old_buf_state;
1463 
1464  if (!BufferIsValid(buffer))
1465  elog(ERROR, "bad buffer ID: %d", buffer);
1466 
1467  if (BufferIsLocal(buffer))
1468  {
1469  MarkLocalBufferDirty(buffer);
1470  return;
1471  }
1472 
1473  bufHdr = GetBufferDescriptor(buffer - 1);
1474 
1475  Assert(BufferIsPinned(buffer));
1477  LW_EXCLUSIVE));
1478 
1479  old_buf_state = pg_atomic_read_u32(&bufHdr->state);
1480  for (;;)
1481  {
1482  if (old_buf_state & BM_LOCKED)
1483  old_buf_state = WaitBufHdrUnlocked(bufHdr);
1484 
1485  buf_state = old_buf_state;
1486 
1487  Assert(BUF_STATE_GET_REFCOUNT(buf_state) > 0);
1488  buf_state |= BM_DIRTY | BM_JUST_DIRTIED;
1489 
1490  if (pg_atomic_compare_exchange_u32(&bufHdr->state, &old_buf_state,
1491  buf_state))
1492  break;
1493  }
1494 
1495  /*
1496  * If the buffer was not dirty already, do vacuum accounting.
1497  */
1498  if (!(old_buf_state & BM_DIRTY))
1499  {
1500  VacuumPageDirty++;
1502  if (VacuumCostActive)
1504  }
1505 }
#define BufferIsPinned(bufnum)
Definition: bufmgr.c:420
bool LWLockHeldByMeInMode(LWLock *l, LWLockMode mode)
Definition: lwlock.c:1860
int VacuumCostBalance
Definition: globals.c:147
static bool pg_atomic_compare_exchange_u32(volatile pg_atomic_uint32 *ptr, uint32 *expected, uint32 newval)
Definition: atomics.h:311
#define BM_DIRTY
Definition: buf_internals.h:58
int VacuumCostPageDirty
Definition: globals.c:139
#define ERROR
Definition: elog.h:43
#define GetBufferDescriptor(id)
#define BM_JUST_DIRTIED
Definition: buf_internals.h:63
long shared_blks_dirtied
Definition: instrument.h:23
unsigned int uint32
Definition: c.h:359
void MarkLocalBufferDirty(Buffer buffer)
Definition: localbuf.c:280
#define BM_LOCKED
Definition: buf_internals.h:57
int VacuumPageDirty
Definition: globals.c:145
#define BufferDescriptorGetContentLock(bdesc)
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
static uint32 WaitBufHdrUnlocked(BufferDesc *buf)
Definition: bufmgr.c:4186
#define elog(elevel,...)
Definition: elog.h:228
pg_atomic_uint32 state
BufferUsage pgBufferUsage
Definition: instrument.c:20
#define BUF_STATE_GET_REFCOUNT(state)
Definition: buf_internals.h:48
bool VacuumCostActive
Definition: globals.c:148
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241

◆ MarkBufferDirtyHint()

void MarkBufferDirtyHint ( Buffer  buffer,
bool  buffer_std 
)

Definition at line 3445 of file bufmgr.c.

References Assert, BM_DIRTY, BM_JUST_DIRTIED, BM_PERMANENT, BUF_STATE_GET_REFCOUNT, BufferDescriptorGetContentLock, BufferGetPage, BufferIsLocal, BufferIsValid, PGXACT::delayChkpt, elog, ERROR, GetBufferDescriptor, GetPrivateRefCount(), InvalidXLogRecPtr, LockBufHdr(), LWLockHeldByMe(), MarkLocalBufferDirty(), MyPgXact, PageSetLSN, pg_atomic_read_u32(), pgBufferUsage, RecoveryInProgress(), BufferUsage::shared_blks_dirtied, BufferDesc::state, UnlockBufHdr, VacuumCostActive, VacuumCostBalance, VacuumCostPageDirty, VacuumPageDirty, XLogHintBitIsNeeded, XLogRecPtrIsInvalid, and XLogSaveBufferForHint().

Referenced by _bt_check_unique(), _bt_killitems(), _hash_kill_items(), brin_start_evacuating_page(), btvacuumpage(), fsm_search_avail(), fsm_set_and_search(), fsm_vacuum_page(), gistkillitems(), heap_page_prune(), read_seq_tuple(), SetHintBits(), and XLogRecordPageWithFreeSpace().

3446 {
3447  BufferDesc *bufHdr;
3448  Page page = BufferGetPage(buffer);
3449 
3450  if (!BufferIsValid(buffer))
3451  elog(ERROR, "bad buffer ID: %d", buffer);
3452 
3453  if (BufferIsLocal(buffer))
3454  {
3455  MarkLocalBufferDirty(buffer);
3456  return;
3457  }
3458 
3459  bufHdr = GetBufferDescriptor(buffer - 1);
3460 
3461  Assert(GetPrivateRefCount(buffer) > 0);
3462  /* here, either share or exclusive lock is OK */
3464 
3465  /*
3466  * This routine might get called many times on the same page, if we are
3467  * making the first scan after commit of an xact that added/deleted many
3468  * tuples. So, be as quick as we can if the buffer is already dirty. We
3469  * do this by not acquiring spinlock if it looks like the status bits are
3470  * already set. Since we make this test unlocked, there's a chance we
3471  * might fail to notice that the flags have just been cleared, and failed
3472  * to reset them, due to memory-ordering issues. But since this function
3473  * is only intended to be used in cases where failing to write out the
3474  * data would be harmless anyway, it doesn't really matter.
3475  */
3476  if ((pg_atomic_read_u32(&bufHdr->state) & (BM_DIRTY | BM_JUST_DIRTIED)) !=
3478  {
3480  bool dirtied = false;
3481  bool delayChkpt = false;
3482  uint32 buf_state;
3483 
3484  /*
3485  * If we need to protect hint bit updates from torn writes, WAL-log a
3486  * full page image of the page. This full page image is only necessary
3487  * if the hint bit update is the first change to the page since the
3488  * last checkpoint.
3489  *
3490  * We don't check full_page_writes here because that logic is included
3491  * when we call XLogInsert() since the value changes dynamically.
3492  */
3493  if (XLogHintBitIsNeeded() &&
3494  (pg_atomic_read_u32(&bufHdr->state) & BM_PERMANENT))
3495  {
3496  /*
3497  * If we're in recovery we cannot dirty a page because of a hint.
3498  * We can set the hint, just not dirty the page as a result so the
3499  * hint is lost when we evict the page or shutdown.
3500  *
3501  * See src/backend/storage/page/README for longer discussion.
3502  */
3503  if (RecoveryInProgress())
3504  return;
3505 
3506  /*
3507  * If the block is already dirty because we either made a change
3508  * or set a hint already, then we don't need to write a full page
3509  * image. Note that aggressive cleaning of blocks dirtied by hint
3510  * bit setting would increase the call rate. Bulk setting of hint
3511  * bits would reduce the call rate...
3512  *
3513  * We must issue the WAL record before we mark the buffer dirty.
3514  * Otherwise we might write the page before we write the WAL. That
3515  * causes a race condition, since a checkpoint might occur between
3516  * writing the WAL record and marking the buffer dirty. We solve
3517  * that with a kluge, but one that is already in use during
3518  * transaction commit to prevent race conditions. Basically, we
3519  * simply prevent the checkpoint WAL record from being written
3520  * until we have marked the buffer dirty. We don't start the
3521  * checkpoint flush until we have marked dirty, so our checkpoint
3522  * must flush the change to disk successfully or the checkpoint
3523  * never gets written, so crash recovery will fix.
3524  *
3525  * It's possible we may enter here without an xid, so it is
3526  * essential that CreateCheckpoint waits for virtual transactions
3527  * rather than full transactionids.
3528  */
3529  MyPgXact->delayChkpt = delayChkpt = true;
3530  lsn = XLogSaveBufferForHint(buffer, buffer_std);
3531  }
3532 
3533  buf_state = LockBufHdr(bufHdr);
3534 
3535  Assert(BUF_STATE_GET_REFCOUNT(buf_state) > 0);
3536 
3537  if (!(buf_state & BM_DIRTY))
3538  {
3539  dirtied = true; /* Means "will be dirtied by this action" */
3540 
3541  /*
3542  * Set the page LSN if we wrote a backup block. We aren't supposed
3543  * to set this when only holding a share lock but as long as we
3544  * serialise it somehow we're OK. We choose to set LSN while
3545  * holding the buffer header lock, which causes any reader of an
3546  * LSN who holds only a share lock to also obtain a buffer header
3547  * lock before using PageGetLSN(), which is enforced in
3548  * BufferGetLSNAtomic().
3549  *
3550  * If checksums are enabled, you might think we should reset the
3551  * checksum here. That will happen when the page is written
3552  * sometime later in this checkpoint cycle.
3553  */
3554  if (!XLogRecPtrIsInvalid(lsn))
3555  PageSetLSN(page, lsn);
3556  }
3557 
3558  buf_state |= BM_DIRTY | BM_JUST_DIRTIED;
3559  UnlockBufHdr(bufHdr, buf_state);
3560 
3561  if (delayChkpt)
3562  MyPgXact->delayChkpt = false;
3563 
3564  if (dirtied)
3565  {
3566  VacuumPageDirty++;
3568  if (VacuumCostActive)
3570  }
3571  }
3572 }
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
#define BM_PERMANENT
Definition: buf_internals.h:66
int VacuumCostBalance
Definition: globals.c:147
bool LWLockHeldByMe(LWLock *l)
Definition: lwlock.c:1842
XLogRecPtr XLogSaveBufferForHint(Buffer buffer, bool buffer_std)
Definition: xloginsert.c:893
bool RecoveryInProgress(void)
Definition: xlog.c:7930
#define BM_DIRTY
Definition: buf_internals.h:58
int VacuumCostPageDirty
Definition: globals.c:139
PGXACT * MyPgXact
Definition: proc.c:68
#define ERROR
Definition: elog.h:43
#define GetBufferDescriptor(id)
#define BM_JUST_DIRTIED
Definition: buf_internals.h:63
long shared_blks_dirtied
Definition: instrument.h:23
static int32 GetPrivateRefCount(Buffer buffer)
Definition: bufmgr.c:359
unsigned int uint32
Definition: c.h:359
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
bool delayChkpt
Definition: proc.h:235
void MarkLocalBufferDirty(Buffer buffer)
Definition: localbuf.c:280
#define XLogRecPtrIsInvalid(r)
Definition: xlogdefs.h:29
int VacuumPageDirty
Definition: globals.c:145
#define BufferDescriptorGetContentLock(bdesc)
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
#define UnlockBufHdr(desc, s)
#define elog(elevel,...)
Definition: elog.h:228
pg_atomic_uint32 state
#define PageSetLSN(page, lsn)
Definition: bufpage.h:368
#define XLogHintBitIsNeeded()
Definition: xlog.h:192
Pointer Page
Definition: bufpage.h:78
BufferUsage pgBufferUsage
Definition: instrument.c:20
#define BUF_STATE_GET_REFCOUNT(state)
Definition: buf_internals.h:48
bool VacuumCostActive
Definition: globals.c:148
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241

◆ NewPrivateRefCountEntry()

static PrivateRefCountEntry * NewPrivateRefCountEntry ( Buffer  buffer)
static

Definition at line 253 of file bufmgr.c.

References Assert, PrivateRefCountEntry::buffer, PrivateRefCountEntry::refcount, and ReservedRefCountEntry.

Referenced by PinBuffer(), and PinBuffer_Locked().

254 {
256 
257  /* only allowed to be called when a reservation has been made */
258  Assert(ReservedRefCountEntry != NULL);
259 
260  /* use up the reserved entry */
261  res = ReservedRefCountEntry;
262  ReservedRefCountEntry = NULL;
263 
264  /* and fill it */
265  res->buffer = buffer;
266  res->refcount = 0;
267 
268  return res;
269 }
static PrivateRefCountEntry * ReservedRefCountEntry
Definition: bufmgr.c:173
#define Assert(condition)
Definition: c.h:739

◆ PinBuffer()

static bool PinBuffer ( BufferDesc buf,
BufferAccessStrategy  strategy 
)
static

Definition at line 1579 of file bufmgr.c.

References Assert, BM_LOCKED, BM_MAX_USAGE_COUNT, BM_VALID, BUF_REFCOUNT_ONE, BUF_STATE_GET_USAGECOUNT, BUF_USAGECOUNT_ONE, BufferDescriptorGetBuffer, CurrentResourceOwner, GetPrivateRefCountEntry(), NewPrivateRefCountEntry(), pg_atomic_compare_exchange_u32(), pg_atomic_read_u32(), PrivateRefCountEntry::refcount, ReservePrivateRefCountEntry(), ResourceOwnerRememberBuffer(), BufferDesc::state, and WaitBufHdrUnlocked().

Referenced by BufferAlloc().

1580 {
1582  bool result;
1583  PrivateRefCountEntry *ref;
1584 
1585  ref = GetPrivateRefCountEntry(b, true);
1586 
1587  if (ref == NULL)
1588  {
1589  uint32 buf_state;
1590  uint32 old_buf_state;
1591 
1593  ref = NewPrivateRefCountEntry(b);
1594 
1595  old_buf_state = pg_atomic_read_u32(&buf->state);
1596  for (;;)
1597  {
1598  if (old_buf_state & BM_LOCKED)
1599  old_buf_state = WaitBufHdrUnlocked(buf);
1600 
1601  buf_state = old_buf_state;
1602 
1603  /* increase refcount */
1604  buf_state += BUF_REFCOUNT_ONE;
1605 
1606  if (strategy == NULL)
1607  {
1608  /* Default case: increase usagecount unless already max. */
1610  buf_state += BUF_USAGECOUNT_ONE;
1611  }
1612  else
1613  {
1614  /*
1615  * Ring buffers shouldn't evict others from pool. Thus we
1616  * don't make usagecount more than 1.
1617  */
1618  if (BUF_STATE_GET_USAGECOUNT(buf_state) == 0)
1619  buf_state += BUF_USAGECOUNT_ONE;
1620  }
1621 
1622  if (pg_atomic_compare_exchange_u32(&buf->state, &old_buf_state,
1623  buf_state))
1624  {
1625  result = (buf_state & BM_VALID) != 0;
1626  break;
1627  }
1628  }
1629  }
1630  else
1631  {
1632  /* If we previously pinned the buffer, it must surely be valid */
1633  result = true;
1634  }
1635 
1636  ref->refcount++;
1637  Assert(ref->refcount > 0);
1639  return result;
1640 }
static PrivateRefCountEntry * GetPrivateRefCountEntry(Buffer buffer, bool do_move)
Definition: bufmgr.c:279
static bool pg_atomic_compare_exchange_u32(volatile pg_atomic_uint32 *ptr, uint32 *expected, uint32 newval)
Definition: atomics.h:311
ResourceOwner CurrentResourceOwner
Definition: resowner.c:142
void ResourceOwnerRememberBuffer(ResourceOwner owner, Buffer buffer)
Definition: resowner.c:906
static PrivateRefCountEntry * NewPrivateRefCountEntry(Buffer buffer)
Definition: bufmgr.c:253
#define BUF_USAGECOUNT_ONE
Definition: buf_internals.h:43
unsigned int uint32
Definition: c.h:359
#define BUF_REFCOUNT_ONE
Definition: buf_internals.h:40
#define BM_LOCKED
Definition: buf_internals.h:57
#define BM_VALID
Definition: buf_internals.h:59
int result
Definition: header.h:19
#define Assert(condition)
Definition: c.h:739
#define BufferDescriptorGetBuffer(bdesc)
static uint32 WaitBufHdrUnlocked(BufferDesc *buf)
Definition: bufmgr.c:4186
#define BM_MAX_USAGE_COUNT
Definition: buf_internals.h:76
static void ReservePrivateRefCountEntry(void)
Definition: bufmgr.c:187
pg_atomic_uint32 state
#define BUF_STATE_GET_USAGECOUNT(state)
Definition: buf_internals.h:49
int Buffer
Definition: buf.h:23
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241

◆ PinBuffer_Locked()

static void PinBuffer_Locked ( BufferDesc buf)
static

Definition at line 1664 of file bufmgr.c.

References Assert, BM_LOCKED, BUF_REFCOUNT_ONE, BufferDescriptorGetBuffer, CurrentResourceOwner, GetPrivateRefCountEntry(), NewPrivateRefCountEntry(), pg_atomic_read_u32(), PrivateRefCountEntry::refcount, ResourceOwnerRememberBuffer(), BufferDesc::state, and UnlockBufHdr.

Referenced by BufferAlloc(), FlushDatabaseBuffers(), FlushRelationBuffers(), and SyncOneBuffer().

1665 {
1666  Buffer b;
1667  PrivateRefCountEntry *ref;
1668  uint32 buf_state;
1669 
1670  /*
1671  * As explained, We don't expect any preexisting pins. That allows us to
1672  * manipulate the PrivateRefCount after releasing the spinlock
1673  */
1675 
1676  /*
1677  * Since we hold the buffer spinlock, we can update the buffer state and
1678  * release the lock in one operation.
1679  */
1680  buf_state = pg_atomic_read_u32(&buf->state);
1681  Assert(buf_state & BM_LOCKED);
1682  buf_state += BUF_REFCOUNT_ONE;
1683  UnlockBufHdr(buf, buf_state);
1684 
1685  b = BufferDescriptorGetBuffer(buf);
1686 
1687  ref = NewPrivateRefCountEntry(b);
1688  ref->refcount++;
1689 
1691 }
static PrivateRefCountEntry * GetPrivateRefCountEntry(Buffer buffer, bool do_move)
Definition: bufmgr.c:279
ResourceOwner CurrentResourceOwner
Definition: resowner.c:142
void ResourceOwnerRememberBuffer(ResourceOwner owner, Buffer buffer)
Definition: resowner.c:906
static PrivateRefCountEntry * NewPrivateRefCountEntry(Buffer buffer)
Definition: bufmgr.c:253
unsigned int uint32
Definition: c.h:359
#define BUF_REFCOUNT_ONE
Definition: buf_internals.h:40
#define BM_LOCKED
Definition: buf_internals.h:57
#define Assert(condition)
Definition: c.h:739
#define BufferDescriptorGetBuffer(bdesc)
#define UnlockBufHdr(desc, s)
pg_atomic_uint32 state
int Buffer
Definition: buf.h:23
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241

◆ PrefetchBuffer()

void PrefetchBuffer ( Relation  reln,
ForkNumber  forkNum,
BlockNumber  blockNum 
)

Definition at line 531 of file bufmgr.c.

References Assert, BlockNumberIsValid, BufMappingPartitionLock, BufTableHashCode(), BufTableLookup(), ereport, errcode(), errmsg(), ERROR, INIT_BUFFERTAG, LocalPrefetchBuffer(), LW_SHARED, LWLockAcquire(), LWLockRelease(), RelFileNodeBackend::node, RelationData::rd_smgr, RELATION_IS_OTHER_TEMP, RelationIsValid, RelationOpenSmgr, RelationUsesLocalBuffers, SMgrRelationData::smgr_rnode, and smgrprefetch().

Referenced by BitmapPrefetch(), count_nondeletable_pages(), HeapTupleHeaderAdvanceLatestRemovedXid(), and pg_prewarm().

532 {
533 #ifdef USE_PREFETCH
534  Assert(RelationIsValid(reln));
535  Assert(BlockNumberIsValid(blockNum));
536 
537  /* Open it at the smgr level if not already done */
538  RelationOpenSmgr(reln);
539 
540  if (RelationUsesLocalBuffers(reln))
541  {
542  /* see comments in ReadBufferExtended */
543  if (RELATION_IS_OTHER_TEMP(reln))
544  ereport(ERROR,
545  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
546  errmsg("cannot access temporary tables of other sessions")));
547 
548  /* pass it off to localbuf.c */
549  LocalPrefetchBuffer(reln->rd_smgr, forkNum, blockNum);
550  }
551  else
552  {
553  BufferTag newTag; /* identity of requested block */
554  uint32 newHash; /* hash value for newTag */
555  LWLock *newPartitionLock; /* buffer partition lock for it */
556  int buf_id;
557 
558  /* create a tag so we can lookup the buffer */
559  INIT_BUFFERTAG(newTag, reln->rd_smgr->smgr_rnode.node,
560  forkNum, blockNum);
561 
562  /* determine its hash code and partition lock ID */
563  newHash = BufTableHashCode(&newTag);
564  newPartitionLock = BufMappingPartitionLock(newHash);
565 
566  /* see if the block is in the buffer pool already */
567  LWLockAcquire(newPartitionLock, LW_SHARED);
568  buf_id = BufTableLookup(&newTag, newHash);
569  LWLockRelease(newPartitionLock);
570 
571  /* If not in buffers, initiate prefetch */
572  if (buf_id < 0)
573  smgrprefetch(reln->rd_smgr, forkNum, blockNum);
574 
575  /*
576  * If the block *is* in buffers, we do nothing. This is not really
577  * ideal: the block might be just about to be evicted, which would be
578  * stupid since we know we are going to need it soon. But the only
579  * easy answer is to bump the usage_count, which does not seem like a
580  * great solution: when the caller does ultimately touch the block,
581  * usage_count would get bumped again, resulting in too much
582  * favoritism for blocks that are involved in a prefetch sequence. A
583  * real fix would involve some additional per-buffer state, and it's
584  * not clear that there's enough of a problem to justify that.
585  */
586  }
587 #endif /* USE_PREFETCH */
588 }
Definition: lwlock.h:32
#define BufMappingPartitionLock(hashcode)
void LocalPrefetchBuffer(SMgrRelation smgr, ForkNumber forkNum, BlockNumber blockNum)
Definition: localbuf.c:64
struct SMgrRelationData * rd_smgr
Definition: rel.h:57
uint32 BufTableHashCode(BufferTag *tagPtr)
Definition: buf_table.c:79
int errcode(int sqlerrcode)
Definition: elog.c:608
int BufTableLookup(BufferTag *tagPtr, uint32 hashcode)
Definition: buf_table.c:91
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1726
#define RelationOpenSmgr(relation)
Definition: rel.h:485
#define ERROR
Definition: elog.h:43
#define RelationIsValid(relation)
Definition: rel.h:401
RelFileNodeBackend smgr_rnode
Definition: smgr.h:42
unsigned int uint32
Definition: c.h:359
#define ereport(elevel, rest)
Definition: elog.h:141
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
RelFileNode node
Definition: relfilenode.h:74
#define Assert(condition)
Definition: c.h:739
#define RELATION_IS_OTHER_TEMP(relation)
Definition: rel.h:558
#define INIT_BUFFERTAG(a, xx_rnode, xx_forkNum, xx_blockNum)
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
#define RelationUsesLocalBuffers(relation)
Definition: rel.h:537
int errmsg(const char *fmt,...)
Definition: elog.c:822
void smgrprefetch(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum)
Definition: smgr.c:494

◆ PrintBufferLeakWarning()

void PrintBufferLeakWarning ( Buffer  buffer)

Definition at line 2549 of file bufmgr.c.

References Assert, buftag::blockNum, buf, BUF_FLAG_MASK, BUF_STATE_GET_REFCOUNT, BufferIsLocal, BufferIsValid, elog, buftag::forkNum, GetBufferDescriptor, GetLocalBufferDescriptor, GetPrivateRefCount(), InvalidBackendId, LocalRefCount, MyBackendId, pfree(), pg_atomic_read_u32(), relpathbackend, buftag::rnode, BufferDesc::state, BufferDesc::tag, and WARNING.

Referenced by CheckForBufferLeaks(), CheckForLocalBufferLeaks(), and ResourceOwnerReleaseInternal().

2550 {
2551  BufferDesc *buf;
2552  int32 loccount;
2553  char *path;
2554  BackendId backend;
2555  uint32 buf_state;
2556 
2557  Assert(BufferIsValid(buffer));
2558  if (BufferIsLocal(buffer))
2559  {
2560  buf = GetLocalBufferDescriptor(-buffer - 1);
2561  loccount = LocalRefCount[-buffer - 1];
2562  backend = MyBackendId;
2563  }
2564  else
2565  {
2566  buf = GetBufferDescriptor(buffer - 1);
2567  loccount = GetPrivateRefCount(buffer);
2568  backend = InvalidBackendId;
2569  }
2570 
2571  /* theoretically we should lock the bufhdr here */
2572  path = relpathbackend(buf->tag.rnode, backend, buf->tag.forkNum);
2573  buf_state = pg_atomic_read_u32(&buf->state);
2574  elog(WARNING,
2575  "buffer refcount leak: [%03d] "
2576  "(rel=%s, blockNum=%u, flags=0x%x, refcount=%u %d)",
2577  buffer, path,
2578  buf->tag.blockNum, buf_state & BUF_FLAG_MASK,
2579  BUF_STATE_GET_REFCOUNT(buf_state), loccount);
2580  pfree(path);
2581 }
BackendId MyBackendId
Definition: globals.c:81
ForkNumber forkNum
Definition: buf_internals.h:93
#define GetLocalBufferDescriptor(id)
signed int int32
Definition: c.h:347
void pfree(void *pointer)
Definition: mcxt.c:1056
#define BUF_FLAG_MASK
Definition: buf_internals.h:45
static char * buf
Definition: pg_test_fsync.c:67
#define GetBufferDescriptor(id)
static int32 GetPrivateRefCount(Buffer buffer)
Definition: bufmgr.c:359
unsigned int uint32
Definition: c.h:359
#define WARNING
Definition: elog.h:40
#define InvalidBackendId
Definition: backendid.h:23
int BackendId
Definition: backendid.h:21
#define Assert(condition)
Definition: c.h:739
#define BufferIsLocal(buffer)
Definition: buf.h:37
BlockNumber blockNum
Definition: buf_internals.h:94
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
RelFileNode rnode
Definition: buf_internals.h:92
BufferTag tag
#define elog(elevel,...)
Definition: elog.h:228
pg_atomic_uint32 state
#define relpathbackend(rnode, backend, forknum)
Definition: relpath.h:78
int32 * LocalRefCount
Definition: localbuf.c:45
#define BUF_STATE_GET_REFCOUNT(state)
Definition: buf_internals.h:48
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241

◆ ReadBuffer()

Buffer ReadBuffer ( Relation  reln,
BlockNumber  blockNum 
)

Definition at line 596 of file bufmgr.c.

References MAIN_FORKNUM, RBM_NORMAL, and ReadBufferExtended().

Referenced by _bt_doinsert(), _bt_getbuf(), _hash_getbuf(), _hash_getbuf_with_condlock_cleanup(), blbulkdelete(), blinsert(), BloomNewBuffer(), brin_getinsertbuffer(), brinbuild(), brinGetStats(), brinGetTupleForHeapBlock(), brinRevmapDesummarizeRange(), brinRevmapInitialize(), bt_metap(), bt_page_items(), bt_page_stats(), fill_seq_with_data(), ginFindLeafPage(), ginFindParents(), ginGetStats(), ginHeapTupleFastInsert(), ginInsertCleanup(), GinNewBuffer(), ginStepRight(), ginUpdateStats(), gistBufferingFindCorrectParent(), gistbufferinginserttuples(), gistdoinsert(), gistFindCorrectParent(), gistFindPath(), gistfixsplit(), gistGetMaxLevel(), gistkillitems(), gistNewBuffer(), gistProcessItup(), gistScanPage(), heap_abort_speculative(), heap_compute_xid_horizon_for_tuples(), heap_delete(), heap_fetch(), heap_finish_speculative(), heap_get_latest_tid(), heap_inplace_update(), heap_lock_tuple(), heap_update(), initBloomState(), log_newpage_range(), pg_visibility(), pgstatginindex_internal(), read_seq_tuple(), RelationGetBufferForTuple(), ReleaseAndReadBuffer(), revmap_get_buffer(), revmap_physical_extend(), scanGetCandidate(), scanPendingInsert(), shiftList(), spgdoinsert(), spgGetCache(), SpGistGetBuffer(), SpGistNewBuffer(), SpGistUpdateMetaPage(), and spgWalk().

597 {
598  return ReadBufferExtended(reln, MAIN_FORKNUM, blockNum, RBM_NORMAL, NULL);
599 }
Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:642

◆ ReadBuffer_common()

static Buffer ReadBuffer_common ( SMgrRelation  reln,
char  relpersistence,
ForkNumber  forkNum,
BlockNumber  blockNum,
ReadBufferMode  mode,
BufferAccessStrategy  strategy,
bool hit 
)
static

Definition at line 705 of file bufmgr.c.

References Assert, RelFileNodeBackend::backend, BufferUsage::blk_read_time, BM_VALID, BufferAlloc(), BufferDescriptorGetBuffer, BufferDescriptorGetContentLock, BufHdrGetBlock, CurrentResourceOwner, RelFileNode::dbNode, ereport, errcode(), ERRCODE_DATA_CORRUPTED, errhint(), errmsg(), ERROR, INSTR_TIME_ADD, INSTR_TIME_GET_MICROSEC, INSTR_TIME_SET_CURRENT, INSTR_TIME_SUBTRACT, BufferUsage::local_blks_hit, BufferUsage::local_blks_read, BufferUsage::local_blks_written, LocalBufferAlloc(), LocalBufHdrGetBlock, LockBufferForCleanup(), LockBufHdr(), LW_EXCLUSIVE, LWLockAcquire(), MemSet, RelFileNodeBackend::node, P_NEW, PageIsNew, PageIsVerified(), pg_atomic_read_u32(), pg_atomic_unlocked_write_u32(), pgBufferUsage, pgstat_count_buffer_read_time, RBM_NORMAL, RBM_NORMAL_NO_LOG, RBM_ZERO_AND_CLEANUP_LOCK, RBM_ZERO_AND_LOCK, RBM_ZERO_ON_ERROR, RelFileNode::relNode, relpath, ResourceOwnerEnlargeBuffers(), BufferUsage::shared_blks_hit, BufferUsage::shared_blks_read, BufferUsage::shared_blks_written, SMgrRelationData::smgr_rnode, smgrextend(), SmgrIsTemp, smgrnblocks(), smgrread(), RelFileNode::spcNode, StartBufferIO(), BufferDesc::state, TerminateBufferIO(), track_io_timing, UnlockBufHdr, VacuumCostActive, VacuumCostBalance, VacuumCostPageHit, VacuumCostPageMiss, VacuumPageHit, VacuumPageMiss, WARNING, and zero_damaged_pages.

Referenced by ReadBufferExtended(), and ReadBufferWithoutRelcache().

708 {
709  BufferDesc *bufHdr;
710  Block bufBlock;
711  bool found;
712  bool isExtend;
713  bool isLocalBuf = SmgrIsTemp(smgr);
714 
715  *hit = false;
716 
717  /* Make sure we will have room to remember the buffer pin */
719 
720  isExtend = (blockNum == P_NEW);
721 
722  TRACE_POSTGRESQL_BUFFER_READ_START(forkNum, blockNum,
723  smgr->smgr_rnode.node.spcNode,
724  smgr->smgr_rnode.node.dbNode,
725  smgr->smgr_rnode.node.relNode,
726  smgr->smgr_rnode.backend,
727  isExtend);
728 
729  /* Substitute proper block number if caller asked for P_NEW */
730  if (isExtend)
731  blockNum = smgrnblocks(smgr, forkNum);
732 
733  if (isLocalBuf)
734  {
735  bufHdr = LocalBufferAlloc(smgr, forkNum, blockNum, &found);
736  if (found)
738  else if (isExtend)
740  else if (mode == RBM_NORMAL || mode == RBM_NORMAL_NO_LOG ||
743  }
744  else
745  {
746  /*
747  * lookup the buffer. IO_IN_PROGRESS is set if the requested block is
748  * not currently in memory.
749  */
750  bufHdr = BufferAlloc(smgr, relpersistence, forkNum, blockNum,
751  strategy, &found);
752  if (found)
754  else if (isExtend)
756  else if (mode == RBM_NORMAL || mode == RBM_NORMAL_NO_LOG ||
759  }
760 
761  /* At this point we do NOT hold any locks. */
762 
763  /* if it was already in the buffer pool, we're done */
764  if (found)
765  {
766  if (!isExtend)
767  {
768  /* Just need to update stats before we exit */
769  *hit = true;
770  VacuumPageHit++;
771 
772  if (VacuumCostActive)
774 
775  TRACE_POSTGRESQL_BUFFER_READ_DONE(forkNum, blockNum,
776  smgr->smgr_rnode.node.spcNode,
777  smgr->smgr_rnode.node.dbNode,
778  smgr->smgr_rnode.node.relNode,
779  smgr->smgr_rnode.backend,
780  isExtend,
781  found);
782 
783  /*
784  * In RBM_ZERO_AND_LOCK mode the caller expects the page to be
785  * locked on return.
786  */
787  if (!isLocalBuf)
788  {
789  if (mode == RBM_ZERO_AND_LOCK)
791  LW_EXCLUSIVE);
792  else if (mode == RBM_ZERO_AND_CLEANUP_LOCK)
794  }
795 
796  return BufferDescriptorGetBuffer(bufHdr);
797  }
798 
799  /*
800  * We get here only in the corner case where we are trying to extend
801  * the relation but we found a pre-existing buffer marked BM_VALID.
802  * This can happen because mdread doesn't complain about reads beyond
803  * EOF (when zero_damaged_pages is ON) and so a previous attempt to
804  * read a block beyond EOF could have left a "valid" zero-filled
805  * buffer. Unfortunately, we have also seen this case occurring
806  * because of buggy Linux kernels that sometimes return an
807  * lseek(SEEK_END) result that doesn't account for a recent write. In
808  * that situation, the pre-existing buffer would contain valid data
809  * that we don't want to overwrite. Since the legitimate case should
810  * always have left a zero-filled buffer, complain if not PageIsNew.
811  */
812  bufBlock = isLocalBuf ? LocalBufHdrGetBlock(bufHdr) : BufHdrGetBlock(bufHdr);
813  if (!PageIsNew((Page) bufBlock))
814  ereport(ERROR,
815  (errmsg("unexpected data beyond EOF in block %u of relation %s",
816  blockNum, relpath(smgr->smgr_rnode, forkNum)),
817  errhint("This has been seen to occur with buggy kernels; consider updating your system.")));
818 
819  /*
820  * We *must* do smgrextend before succeeding, else the page will not
821  * be reserved by the kernel, and the next P_NEW call will decide to
822  * return the same page. Clear the BM_VALID bit, do the StartBufferIO
823  * call that BufferAlloc didn't, and proceed.
824  */
825  if (isLocalBuf)
826  {
827  /* Only need to adjust flags */
828  uint32 buf_state = pg_atomic_read_u32(&bufHdr->state);
829 
830  Assert(buf_state & BM_VALID);
831  buf_state &= ~BM_VALID;
832  pg_atomic_unlocked_write_u32(&bufHdr->state, buf_state);
833  }
834  else
835  {
836  /*
837  * Loop to handle the very small possibility that someone re-sets
838  * BM_VALID between our clearing it and StartBufferIO inspecting
839  * it.
840  */
841  do
842  {
843  uint32 buf_state = LockBufHdr(bufHdr);
844 
845  Assert(buf_state & BM_VALID);
846  buf_state &= ~BM_VALID;
847  UnlockBufHdr(bufHdr, buf_state);
848  } while (!StartBufferIO(bufHdr, true));
849  }
850  }
851 
852  /*
853  * if we have gotten to this point, we have allocated a buffer for the
854  * page but its contents are not yet valid. IO_IN_PROGRESS is set for it,
855  * if it's a shared buffer.
856  *
857  * Note: if smgrextend fails, we will end up with a buffer that is
858  * allocated but not marked BM_VALID. P_NEW will still select the same
859  * block number (because the relation didn't get any longer on disk) and
860  * so future attempts to extend the relation will find the same buffer (if
861  * it's not been recycled) but come right back here to try smgrextend
862  * again.
863  */
864  Assert(!(pg_atomic_read_u32(&bufHdr->state) & BM_VALID)); /* spinlock not needed */
865 
866  bufBlock = isLocalBuf ? LocalBufHdrGetBlock(bufHdr) : BufHdrGetBlock(bufHdr);
867 
868  if (isExtend)
869  {
870  /* new buffers are zero-filled */
871  MemSet((char *) bufBlock, 0, BLCKSZ);
872  /* don't set checksum for all-zero page */
873  smgrextend(smgr, forkNum, blockNum, (char *) bufBlock, false);
874 
875  /*
876  * NB: we're *not* doing a ScheduleBufferTagForWriteback here;
877  * although we're essentially performing a write. At least on linux
878  * doing so defeats the 'delayed allocation' mechanism, leading to
879  * increased file fragmentation.
880  */
881  }
882  else
883  {
884  /*
885  * Read in the page, unless the caller intends to overwrite it and
886  * just wants us to allocate a buffer.
887  */
889  MemSet((char *) bufBlock, 0, BLCKSZ);
890  else
891  {
892  instr_time io_start,
893  io_time;
894 
895  if (track_io_timing)
896  INSTR_TIME_SET_CURRENT(io_start);
897 
898  smgrread(smgr, forkNum, blockNum, (char *) bufBlock);
899 
900  if (track_io_timing)
901  {
902  INSTR_TIME_SET_CURRENT(io_time);
903  INSTR_TIME_SUBTRACT(io_time, io_start);
906  }
907 
908  /* check for garbage data */
909  if (!PageIsVerified((Page) bufBlock, blockNum))
910  {
912  {
915  errmsg("invalid page in block %u of relation %s; zeroing out page",
916  blockNum,
917  relpath(smgr->smgr_rnode, forkNum))));
918  MemSet((char *) bufBlock, 0, BLCKSZ);
919  }
920  else
921  ereport(ERROR,
923  errmsg("invalid page in block %u of relation %s",
924  blockNum,
925  relpath(smgr->smgr_rnode, forkNum))));
926  }
927  }
928  }
929 
930  /*
931  * In RBM_ZERO_AND_LOCK mode, grab the buffer content lock before marking
932  * the page as valid, to make sure that no other backend sees the zeroed
933  * page before the caller has had a chance to initialize it.
934  *
935  * Since no-one else can be looking at the page contents yet, there is no
936  * difference between an exclusive lock and a cleanup-strength lock. (Note
937  * that we cannot use LockBuffer() or LockBufferForCleanup() here, because
938  * they assert that the buffer is already valid.)
939  */
941  !isLocalBuf)
942  {
944  }
945 
946  if (isLocalBuf)
947  {
948  /* Only need to adjust flags */
949  uint32 buf_state = pg_atomic_read_u32(&bufHdr->state);
950 
951  buf_state |= BM_VALID;
952  pg_atomic_unlocked_write_u32(&bufHdr->state, buf_state);
953  }
954  else
955  {
956  /* Set BM_VALID, terminate IO, and wake up any waiters */
957  TerminateBufferIO(bufHdr, false, BM_VALID);
958  }
959 
960  VacuumPageMiss++;
961  if (VacuumCostActive)
963 
964  TRACE_POSTGRESQL_BUFFER_READ_DONE(forkNum, blockNum,
965  smgr->smgr_rnode.node.spcNode,
966  smgr->smgr_rnode.node.dbNode,
967  smgr->smgr_rnode.node.relNode,
968  smgr->smgr_rnode.backend,
969  isExtend,
970  found);
971 
972  return BufferDescriptorGetBuffer(bufHdr);
973 }
#define LocalBufHdrGetBlock(bufHdr)
Definition: bufmgr.c:62
BufferDesc * LocalBufferAlloc(SMgrRelation smgr, ForkNumber forkNum, BlockNumber blockNum, bool *foundPtr)
Definition: localbuf.c:103
static PgChecksumMode mode
Definition: pg_checksums.c:61
long local_blks_hit
Definition: instrument.h:25
void LockBufferForCleanup(Buffer buffer)
Definition: bufmgr.c:3669
int errhint(const char *fmt,...)
Definition: elog.c:1069
long local_blks_read
Definition: instrument.h:26
int VacuumCostBalance
Definition: globals.c:147
bool PageIsVerified(Page page, BlockNumber blkno)
Definition: bufpage.c:82
instr_time blk_read_time
Definition: instrument.h:31
int VacuumPageHit
Definition: globals.c:143
ResourceOwner CurrentResourceOwner
Definition: resowner.c:142
struct timeval instr_time
Definition: instr_time.h:150
long shared_blks_read
Definition: instrument.h:22
int errcode(int sqlerrcode)
Definition: elog.c:608
#define MemSet(start, val, len)
Definition: c.h:962
#define P_NEW
Definition: bufmgr.h:81
void smgrread(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer)
Definition: smgr.c:508
#define SmgrIsTemp(smgr)
Definition: smgr.h:79
long shared_blks_written
Definition: instrument.h:24
static bool StartBufferIO(BufferDesc *buf, bool forInput)
Definition: bufmgr.c:3941
#define ERROR
Definition: elog.h:43
char relpersistence
Definition: pg_class.h:78
#define INSTR_TIME_SUBTRACT(x, y)
Definition: instr_time.h:170
#define INSTR_TIME_ADD(x, y)
Definition: instr_time.h:158
unsigned int uint32
Definition: c.h:359
#define ereport(elevel, rest)
Definition: elog.h:141
int VacuumCostPageHit
Definition: globals.c:137
#define ERRCODE_DATA_CORRUPTED
Definition: pg_basebackup.c:45
void ResourceOwnerEnlargeBuffers(ResourceOwner owner)
Definition: resowner.c:893
#define WARNING
Definition: elog.h:40
static void TerminateBufferIO(BufferDesc *buf, bool clear_dirty, uint32 set_flag_bits)
Definition: bufmgr.c:4008
#define BM_VALID
Definition: buf_internals.h:59
#define BufHdrGetBlock(bufHdr)
Definition: bufmgr.c:58
static BufferDesc * BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum, BlockNumber blockNum, BufferAccessStrategy strategy, bool *foundPtr)
Definition: bufmgr.c:995
#define BufferDescriptorGetContentLock(bdesc)
uint32 LockBufHdr(BufferDesc *desc)
Definition: bufmgr.c:4158
BlockNumber smgrnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: smgr.c:555
#define Assert(condition)
Definition: c.h:739
#define pgstat_count_buffer_read_time(n)
Definition: pgstat.h:1395
#define INSTR_TIME_GET_MICROSEC(t)
Definition: instr_time.h:205
#define BufferDescriptorGetBuffer(bdesc)
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
#define INSTR_TIME_SET_CURRENT(t)
Definition: instr_time.h:156
void smgrextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: smgr.c:483
#define PageIsNew(page)
Definition: bufpage.h:229
int errmsg(const char *fmt,...)
Definition: elog.c:822
long shared_blks_hit
Definition: instrument.h:21
#define UnlockBufHdr(desc, s)
long local_blks_written
Definition: instrument.h:28
#define relpath(rnode, forknum)
Definition: relpath.h:87
static void pg_atomic_unlocked_write_u32(volatile pg_atomic_uint32 *ptr, uint32 val)
Definition: atomics.h:277
pg_atomic_uint32 state
int VacuumPageMiss
Definition: globals.c:144
int VacuumCostPageMiss
Definition: globals.c:138
bool track_io_timing
Definition: bufmgr.c:112
Pointer Page
Definition: bufpage.h:78
BufferUsage pgBufferUsage
Definition: instrument.c:20
void * Block
Definition: bufmgr.h:24
bool VacuumCostActive
Definition: globals.c:148
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241
bool zero_damaged_pages
Definition: bufmgr.c:109

◆ ReadBufferExtended()

Buffer ReadBufferExtended ( Relation  reln,
ForkNumber  forkNum,
BlockNumber  blockNum,
ReadBufferMode  mode,
BufferAccessStrategy  strategy 
)

Definition at line 642 of file bufmgr.c.

References buf, ereport, errcode(), errmsg(), ERROR, pgstat_count_buffer_hit, pgstat_count_buffer_read, RelationData::rd_rel, RelationData::rd_smgr, ReadBuffer_common(), RELATION_IS_OTHER_TEMP, and RelationOpenSmgr.

Referenced by _hash_getbuf_with_strategy(), _hash_getinitbuf(), _hash_getnewbuf(), autoprewarm_database_main(), blbulkdelete(), blgetbitmap(), blvacuumcleanup(), brin_vacuum_scan(), brinbuildempty(), btvacuumpage(), collect_corrupt_items(), collect_visibility_data(), count_nondeletable_pages(), fsm_readbuf(), get_raw_page_internal(), ginbuildempty(), ginbulkdelete(), ginDeletePage(), ginScanToDelete(), ginvacuumcleanup(), ginVacuumPostingTree(), ginVacuumPostingTreeLeaves(), gistbuildempty(), gistvacuum_delete_empty_pages(), gistvacuumpage(), hashbulkdelete(), heapam_scan_analyze_next_block(), heapgetpage(), lazy_scan_heap(), lazy_vacuum_heap(), palloc_btree_page(), pg_prewarm(), pgstat_btree_page(), pgstat_gist_page(), pgstat_heap(), pgstathashindex(), pgstatindex_impl(), ReadBuffer(), ReadBufferBI(), spgprocesspending(), spgvacuumpage(), statapprox_heap(), and vm_readbuf().

644 {
645  bool hit;
646  Buffer buf;
647 
648  /* Open it at the smgr level if not already done */
649  RelationOpenSmgr(reln);
650 
651  /*
652  * Reject attempts to read non-local temporary relations; we would be
653  * likely to get wrong data since we have no visibility into the owning
654  * session's local buffers.
655  */
656  if (RELATION_IS_OTHER_TEMP(reln))
657  ereport(ERROR,
658  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
659  errmsg("cannot access temporary tables of other sessions")));
660 
661  /*
662  * Read the buffer, and update pgstat counters to reflect a cache hit or
663  * miss.
664  */
666  buf = ReadBuffer_common(reln->rd_smgr, reln->rd_rel->relpersistence,
667  forkNum, blockNum, mode, strategy, &hit);
668  if (hit)
670  return buf;
671 }
static PgChecksumMode mode
Definition: pg_checksums.c:61
struct SMgrRelationData * rd_smgr
Definition: rel.h:57
int errcode(int sqlerrcode)
Definition: elog.c:608
Form_pg_class rd_rel
Definition: rel.h:84
#define RelationOpenSmgr(relation)
Definition: rel.h:485
#define ERROR
Definition: elog.h:43
#define pgstat_count_buffer_read(rel)
Definition: pgstat.h:1385
static char * buf
Definition: pg_test_fsync.c:67
#define ereport(elevel, rest)
Definition: elog.h:141
#define pgstat_count_buffer_hit(rel)
Definition: pgstat.h:1390
static Buffer ReadBuffer_common(SMgrRelation reln, char relpersistence, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy, bool *hit)
Definition: bufmgr.c:705
#define RELATION_IS_OTHER_TEMP(relation)
Definition: rel.h:558
int errmsg(const char *fmt,...)
Definition: elog.c:822
int Buffer
Definition: buf.h:23

◆ ReadBufferWithoutRelcache()

Buffer ReadBufferWithoutRelcache ( RelFileNode  rnode,
ForkNumber  forkNum,
BlockNumber  blockNum,
ReadBufferMode  mode,
BufferAccessStrategy  strategy 
)

Definition at line 684 of file bufmgr.c.

References Assert, InRecovery, InvalidBackendId, ReadBuffer_common(), and smgropen().

Referenced by XLogReadBufferExtended().

687 {
688  bool hit;
689 
690  SMgrRelation smgr = smgropen(rnode, InvalidBackendId);
691 
693 
694  return ReadBuffer_common(smgr, RELPERSISTENCE_PERMANENT, forkNum, blockNum,
695  mode, strategy, &hit);
696 }
static PgChecksumMode mode
Definition: pg_checksums.c:61
bool InRecovery
Definition: xlog.c:200
SMgrRelation smgropen(RelFileNode rnode, BackendId backend)
Definition: smgr.c:145
#define InvalidBackendId
Definition: backendid.h:23
static Buffer ReadBuffer_common(SMgrRelation reln, char relpersistence, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy, bool *hit)
Definition: bufmgr.c:705
#define Assert(condition)
Definition: c.h:739

◆ RelationGetNumberOfBlocksInFork()

BlockNumber RelationGetNumberOfBlocksInFork ( Relation  relation,
ForkNumber  forkNum 
)

Definition at line 2810 of file bufmgr.c.

References Assert, RelationData::rd_rel, RelationData::rd_smgr, RelationOpenSmgr, smgrnblocks(), and table_relation_size().

Referenced by _hash_getnewbuf(), _hash_init(), autoprewarm_database_main(), get_raw_page_internal(), and pg_prewarm().

2811 {
2812  switch (relation->rd_rel->relkind)
2813  {
2814  case RELKIND_SEQUENCE:
2815  case RELKIND_INDEX:
2816  case RELKIND_PARTITIONED_INDEX:
2817  /* Open it at the smgr level if not already done */
2818  RelationOpenSmgr(relation);
2819 
2820  return smgrnblocks(relation->rd_smgr, forkNum);
2821 
2822  case RELKIND_RELATION:
2823  case RELKIND_TOASTVALUE:
2824  case RELKIND_MATVIEW:
2825  {
2826  /*
2827  * Not every table AM uses BLCKSZ wide fixed size blocks.
2828  * Therefore tableam returns the size in bytes - but for the
2829  * purpose of this routine, we want the number of blocks.
2830  * Therefore divide, rounding up.
2831  */
2832  uint64 szbytes;
2833 
2834  szbytes = table_relation_size(relation, forkNum);
2835 
2836  return (szbytes + (BLCKSZ - 1)) / BLCKSZ;
2837  }
2838  case RELKIND_VIEW:
2839  case RELKIND_COMPOSITE_TYPE:
2840  case RELKIND_FOREIGN_TABLE:
2841  case RELKIND_PARTITIONED_TABLE:
2842  default:
2843  Assert(false);
2844  break;
2845  }
2846 
2847  return 0; /* keep compiler quiet */
2848 }
struct SMgrRelationData * rd_smgr
Definition: rel.h:57
Form_pg_class rd_rel
Definition: rel.h:84
#define RelationOpenSmgr(relation)
Definition: rel.h:485
static uint64 table_relation_size(Relation rel, ForkNumber forkNumber)
Definition: tableam.h:1610
BlockNumber smgrnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: smgr.c:555
#define Assert(condition)
Definition: c.h:739

◆ ReleaseAndReadBuffer()

Buffer ReleaseAndReadBuffer ( Buffer  buffer,
Relation  relation,
BlockNumber  blockNum 
)

Definition at line 1521 of file bufmgr.c.

References Assert, buftag::blockNum, PrivateRefCountEntry::buffer, BufferIsLocal, BufferIsPinned, BufferIsValid, CurrentResourceOwner, buftag::forkNum, GetBufferDescriptor, GetLocalBufferDescriptor, LocalRefCount, MAIN_FORKNUM, RelationData::rd_node, ReadBuffer(), RelFileNodeEquals, ResourceOwnerForgetBuffer(), buftag::rnode, BufferDesc::tag, and UnpinBuffer().

Referenced by _bt_relandgetbuf(), ginFindLeafPage(), heapam_index_fetch_tuple(), and heapam_scan_bitmap_next_block().

1524 {
1525  ForkNumber forkNum = MAIN_FORKNUM;
1526  BufferDesc *bufHdr;
1527 
1528  if (BufferIsValid(buffer))
1529  {
1530  Assert(BufferIsPinned(buffer));
1531  if (BufferIsLocal(buffer))
1532  {
1533  bufHdr = GetLocalBufferDescriptor(-buffer - 1);
1534  if (bufHdr->tag.blockNum == blockNum &&
1535  RelFileNodeEquals(bufHdr->tag.rnode, relation->rd_node) &&
1536  bufHdr->tag.forkNum == forkNum)
1537  return buffer;
1539  LocalRefCount[-buffer - 1]--;
1540  }
1541  else
1542  {
1543  bufHdr = GetBufferDescriptor(buffer - 1);
1544  /* we have pin, so it's ok to examine tag without spinlock */
1545  if (bufHdr->tag.blockNum == blockNum &&
1546  RelFileNodeEquals(bufHdr->tag.rnode, relation->rd_node) &&
1547  bufHdr->tag.forkNum == forkNum)
1548  return buffer;
1549  UnpinBuffer(bufHdr, true);
1550  }
1551  }
1552 
1553  return ReadBuffer(relation, blockNum);
1554 }
#define BufferIsPinned(bufnum)
Definition: bufmgr.c:420
ForkNumber forkNum
Definition: buf_internals.h:93
ResourceOwner CurrentResourceOwner
Definition: resowner.c:142
#define GetLocalBufferDescriptor(id)
#define GetBufferDescriptor(id)
Definit