PostgreSQL Source Code  git master
heapam.c File Reference
#include "postgres.h"
#include "access/bufmask.h"
#include "access/genam.h"
#include "access/heapam.h"
#include "access/heapam_xlog.h"
#include "access/heaptoast.h"
#include "access/hio.h"
#include "access/multixact.h"
#include "access/parallel.h"
#include "access/relscan.h"
#include "access/subtrans.h"
#include "access/syncscan.h"
#include "access/sysattr.h"
#include "access/tableam.h"
#include "access/transam.h"
#include "access/valid.h"
#include "access/visibilitymap.h"
#include "access/xact.h"
#include "access/xlog.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
#include "catalog/catalog.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/atomics.h"
#include "port/pg_bitutils.h"
#include "storage/bufmgr.h"
#include "storage/freespace.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
#include "storage/procarray.h"
#include "storage/smgr.h"
#include "storage/spin.h"
#include "storage/standby.h"
#include "utils/datum.h"
#include "utils/inval.h"
#include "utils/lsyscache.h"
#include "utils/relcache.h"
#include "utils/snapmgr.h"
#include "utils/spccache.h"
Include dependency graph for heapam.c:

Go to the source code of this file.

Data Structures

struct  IndexDeleteCounts
 

Macros

#define LOCKMODE_from_mxstatus(status)    (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
 
#define LockTupleTuplock(rel, tup, mode)    LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
 
#define UnlockTupleTuplock(rel, tup, mode)    UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
 
#define ConditionalLockTupleTuplock(rel, tup, mode)    ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
 
#define BOTTOMUP_MAX_NBLOCKS   6
 
#define BOTTOMUP_TOLERANCE_NBLOCKS   3
 
#define TUPLOCK_from_mxstatus(status)    (MultiXactStatusLock[(status)])
 
#define FRM_NOOP   0x0001
 
#define FRM_INVALIDATE_XMAX   0x0002
 
#define FRM_RETURN_IS_XID   0x0004
 
#define FRM_RETURN_IS_MULTI   0x0008
 
#define FRM_MARK_COMMITTED   0x0010
 

Typedefs

typedef struct IndexDeleteCounts IndexDeleteCounts
 

Functions

static HeapTuple heap_prepare_insert (Relation relation, HeapTuple tup, TransactionId xid, CommandId cid, int options)
 
static XLogRecPtr log_heap_update (Relation reln, Buffer oldbuf, Buffer newbuf, HeapTuple oldtup, HeapTuple newtup, HeapTuple old_key_tuple, bool all_visible_cleared, bool new_all_visible_cleared)
 
static BitmapsetHeapDetermineModifiedColumns (Relation relation, Bitmapset *interesting_cols, HeapTuple oldtup, HeapTuple newtup)
 
static bool heap_acquire_tuplock (Relation relation, ItemPointer tid, LockTupleMode mode, LockWaitPolicy wait_policy, bool *have_tuple_lock)
 
static void compute_new_xmax_infomask (TransactionId xmax, uint16 old_infomask, uint16 old_infomask2, TransactionId add_to_xmax, LockTupleMode mode, bool is_update, TransactionId *result_xmax, uint16 *result_infomask, uint16 *result_infomask2)
 
static TM_Result heap_lock_updated_tuple (Relation rel, HeapTuple tuple, ItemPointer ctid, TransactionId xid, LockTupleMode mode)
 
static void GetMultiXactIdHintBits (MultiXactId multi, uint16 *new_infomask, uint16 *new_infomask2)
 
static TransactionId MultiXactIdGetUpdateXid (TransactionId xmax, uint16 t_infomask)
 
static bool DoesMultiXactIdConflict (MultiXactId multi, uint16 infomask, LockTupleMode lockmode, bool *current_is_member)
 
static void MultiXactIdWait (MultiXactId multi, MultiXactStatus status, uint16 infomask, Relation rel, ItemPointer ctid, XLTW_Oper oper, int *remaining)
 
static bool ConditionalMultiXactIdWait (MultiXactId multi, MultiXactStatus status, uint16 infomask, Relation rel, int *remaining)
 
static void index_delete_sort (TM_IndexDeleteOp *delstate)
 
static int bottomup_sort_and_shrink (TM_IndexDeleteOp *delstate)
 
static XLogRecPtr log_heap_new_cid (Relation relation, HeapTuple tup)
 
static HeapTuple ExtractReplicaIdentity (Relation rel, HeapTuple tup, bool key_changed, bool *copy)
 
static void initscan (HeapScanDesc scan, ScanKey key, bool keep_startblock)
 
void heap_setscanlimits (TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
 
void heapgetpage (TableScanDesc sscan, BlockNumber page)
 
static void heapgettup (HeapScanDesc scan, ScanDirection dir, int nkeys, ScanKey key)
 
static void heapgettup_pagemode (HeapScanDesc scan, ScanDirection dir, int nkeys, ScanKey key)
 
TableScanDesc heap_beginscan (Relation relation, Snapshot snapshot, int nkeys, ScanKey key, ParallelTableScanDesc parallel_scan, uint32 flags)
 
void heap_rescan (TableScanDesc sscan, ScanKey key, bool set_params, bool allow_strat, bool allow_sync, bool allow_pagemode)
 
void heap_endscan (TableScanDesc sscan)
 
HeapTuple heap_getnext (TableScanDesc sscan, ScanDirection direction)
 
bool heap_getnextslot (TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
 
void heap_set_tidrange (TableScanDesc sscan, ItemPointer mintid, ItemPointer maxtid)
 
bool heap_getnextslot_tidrange (TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
 
bool heap_fetch (Relation relation, Snapshot snapshot, HeapTuple tuple, Buffer *userbuf)
 
bool heap_hot_search_buffer (ItemPointer tid, Relation relation, Buffer buffer, Snapshot snapshot, HeapTuple heapTuple, bool *all_dead, bool first_call)
 
void heap_get_latest_tid (TableScanDesc sscan, ItemPointer tid)
 
static void UpdateXmaxHintBits (HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
 
BulkInsertState GetBulkInsertState (void)
 
void FreeBulkInsertState (BulkInsertState bistate)
 
void ReleaseBulkInsertStatePin (BulkInsertState bistate)
 
void heap_insert (Relation relation, HeapTuple tup, CommandId cid, int options, BulkInsertState bistate)
 
void heap_multi_insert (Relation relation, TupleTableSlot **slots, int ntuples, CommandId cid, int options, BulkInsertState bistate)
 
void simple_heap_insert (Relation relation, HeapTuple tup)
 
static uint8 compute_infobits (uint16 infomask, uint16 infomask2)
 
static bool xmax_infomask_changed (uint16 new_infomask, uint16 old_infomask)
 
TM_Result heap_delete (Relation relation, ItemPointer tid, CommandId cid, Snapshot crosscheck, bool wait, TM_FailureData *tmfd, bool changingPart)
 
void simple_heap_delete (Relation relation, ItemPointer tid)
 
TM_Result heap_update (Relation relation, ItemPointer otid, HeapTuple newtup, CommandId cid, Snapshot crosscheck, bool wait, TM_FailureData *tmfd, LockTupleMode *lockmode)
 
static bool heap_tuple_attr_equals (TupleDesc tupdesc, int attrnum, HeapTuple tup1, HeapTuple tup2)
 
void simple_heap_update (Relation relation, ItemPointer otid, HeapTuple tup)
 
static MultiXactStatus get_mxact_status_for_lock (LockTupleMode mode, bool is_update)
 
TM_Result heap_lock_tuple (Relation relation, HeapTuple tuple, CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy, bool follow_updates, Buffer *buffer, TM_FailureData *tmfd)
 
static TM_Result test_lockmode_for_conflict (MultiXactStatus status, TransactionId xid, LockTupleMode mode, HeapTuple tup, bool *needwait)
 
static TM_Result heap_lock_updated_tuple_rec (Relation rel, ItemPointer tid, TransactionId xid, LockTupleMode mode)
 
void heap_finish_speculative (Relation relation, ItemPointer tid)
 
void heap_abort_speculative (Relation relation, ItemPointer tid)
 
void heap_inplace_update (Relation relation, HeapTuple tuple)
 
static TransactionId FreezeMultiXactId (MultiXactId multi, uint16 t_infomask, TransactionId relfrozenxid, TransactionId relminmxid, TransactionId cutoff_xid, MultiXactId cutoff_multi, uint16 *flags)
 
bool heap_prepare_freeze_tuple (HeapTupleHeader tuple, TransactionId relfrozenxid, TransactionId relminmxid, TransactionId cutoff_xid, TransactionId cutoff_multi, xl_heap_freeze_tuple *frz, bool *totally_frozen_p)
 
void heap_execute_freeze_tuple (HeapTupleHeader tuple, xl_heap_freeze_tuple *frz)
 
bool heap_freeze_tuple (HeapTupleHeader tuple, TransactionId relfrozenxid, TransactionId relminmxid, TransactionId cutoff_xid, TransactionId cutoff_multi)
 
TransactionId HeapTupleGetUpdateXid (HeapTupleHeader tuple)
 
static bool Do_MultiXactIdWait (MultiXactId multi, MultiXactStatus status, uint16 infomask, bool nowait, Relation rel, ItemPointer ctid, XLTW_Oper oper, int *remaining)
 
bool heap_tuple_needs_eventual_freeze (HeapTupleHeader tuple)
 
bool heap_tuple_needs_freeze (HeapTupleHeader tuple, TransactionId cutoff_xid, MultiXactId cutoff_multi, Buffer buf)
 
void HeapTupleHeaderAdvanceLatestRemovedXid (HeapTupleHeader tuple, TransactionId *latestRemovedXid)
 
static void index_delete_check_htid (TM_IndexDeleteOp *delstate, Page page, OffsetNumber maxoff, ItemPointer htid, TM_IndexStatus *istatus)
 
TransactionId heap_index_delete_tuples (Relation rel, TM_IndexDeleteOp *delstate)
 
static int index_delete_sort_cmp (TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
 
static int bottomup_nblocksfavorable (IndexDeleteCounts *blockgroups, int nblockgroups, TM_IndexDelete *deltids)
 
static int bottomup_sort_and_shrink_cmp (const void *arg1, const void *arg2)
 
XLogRecPtr log_heap_freeze (Relation reln, Buffer buffer, TransactionId cutoff_xid, xl_heap_freeze_tuple *tuples, int ntuples)
 
XLogRecPtr log_heap_visible (RelFileNode rnode, Buffer heap_buffer, Buffer vm_buffer, TransactionId cutoff_xid, uint8 vmflags)
 
static void heap_xlog_prune (XLogReaderState *record)
 
static void heap_xlog_vacuum (XLogReaderState *record)
 
static void heap_xlog_visible (XLogReaderState *record)
 
static void heap_xlog_freeze_page (XLogReaderState *record)
 
static void fix_infomask_from_infobits (uint8 infobits, uint16 *infomask, uint16 *infomask2)
 
static void heap_xlog_delete (XLogReaderState *record)
 
static void heap_xlog_insert (XLogReaderState *record)
 
static void heap_xlog_multi_insert (XLogReaderState *record)
 
static void heap_xlog_update (XLogReaderState *record, bool hot_update)
 
static void heap_xlog_confirm (XLogReaderState *record)
 
static void heap_xlog_lock (XLogReaderState *record)
 
static void heap_xlog_lock_updated (XLogReaderState *record)
 
static void heap_xlog_inplace (XLogReaderState *record)
 
void heap_redo (XLogReaderState *record)
 
void heap2_redo (XLogReaderState *record)
 
void heap_mask (char *pagedata, BlockNumber blkno)
 
void HeapCheckForSerializableConflictOut (bool visible, Relation relation, HeapTuple tuple, Buffer buffer, Snapshot snapshot)
 

Variables

struct {
   LOCKMODE   hwlock
 
   int   lockstatus
 
   int   updstatus
 
tupleLockExtraInfo [MaxLockTupleMode+1]
 
static const int MultiXactStatusLock [MaxMultiXactStatus+1]
 

Macro Definition Documentation

◆ BOTTOMUP_MAX_NBLOCKS

#define BOTTOMUP_MAX_NBLOCKS   6

Definition at line 185 of file heapam.c.

◆ BOTTOMUP_TOLERANCE_NBLOCKS

#define BOTTOMUP_TOLERANCE_NBLOCKS   3

Definition at line 186 of file heapam.c.

◆ ConditionalLockTupleTuplock

#define ConditionalLockTupleTuplock (   rel,
  tup,
  mode 
)     ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)

Definition at line 167 of file heapam.c.

◆ FRM_INVALIDATE_XMAX

#define FRM_INVALIDATE_XMAX   0x0002

Definition at line 6065 of file heapam.c.

◆ FRM_MARK_COMMITTED

#define FRM_MARK_COMMITTED   0x0010

Definition at line 6068 of file heapam.c.

◆ FRM_NOOP

#define FRM_NOOP   0x0001

Definition at line 6064 of file heapam.c.

◆ FRM_RETURN_IS_MULTI

#define FRM_RETURN_IS_MULTI   0x0008

Definition at line 6067 of file heapam.c.

◆ FRM_RETURN_IS_XID

#define FRM_RETURN_IS_XID   0x0004

Definition at line 6066 of file heapam.c.

◆ LOCKMODE_from_mxstatus

#define LOCKMODE_from_mxstatus (   status)     (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)

Definition at line 155 of file heapam.c.

◆ LockTupleTuplock

#define LockTupleTuplock (   rel,
  tup,
  mode 
)     LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)

Definition at line 163 of file heapam.c.

◆ TUPLOCK_from_mxstatus

#define TUPLOCK_from_mxstatus (   status)     (MultiXactStatusLock[(status)])

Definition at line 214 of file heapam.c.

◆ UnlockTupleTuplock

#define UnlockTupleTuplock (   rel,
  tup,
  mode 
)     UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)

Definition at line 165 of file heapam.c.

Typedef Documentation

◆ IndexDeleteCounts

Function Documentation

◆ bottomup_nblocksfavorable()

static int bottomup_nblocksfavorable ( IndexDeleteCounts blockgroups,
int  nblockgroups,
TM_IndexDelete deltids 
)
static

Definition at line 7757 of file heapam.c.

7759 {
7760  int64 lastblock = -1;
7761  int nblocksfavorable = 0;
7762 
7763  Assert(nblockgroups >= 1);
7764  Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
7765 
7766  /*
7767  * We tolerate heap blocks that will be accessed only slightly out of
7768  * physical order. Small blips occur when a pair of almost-contiguous
7769  * blocks happen to fall into different buckets (perhaps due only to a
7770  * small difference in npromisingtids that the bucketing scheme didn't
7771  * quite manage to ignore). We effectively ignore these blips by applying
7772  * a small tolerance. The precise tolerance we use is a little arbitrary,
7773  * but it works well enough in practice.
7774  */
7775  for (int b = 0; b < nblockgroups; b++)
7776  {
7777  IndexDeleteCounts *group = blockgroups + b;
7778  TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
7779  BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
7780 
7781  if (lastblock != -1 &&
7782  ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
7783  (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
7784  break;
7785 
7786  nblocksfavorable++;
7787  lastblock = block;
7788  }
7789 
7790  /* Always indicate that there is at least 1 favorable block */
7791  Assert(nblocksfavorable >= 1);
7792 
7793  return nblocksfavorable;
7794 }
uint32 BlockNumber
Definition: block.h:31
#define BOTTOMUP_TOLERANCE_NBLOCKS
Definition: heapam.c:186
#define BOTTOMUP_MAX_NBLOCKS
Definition: heapam.c:185
int b
Definition: isn.c:70
#define ItemPointerGetBlockNumber(pointer)
Definition: itemptr.h:98
Assert(fmt[strlen(fmt) - 1] !='\n')
int16 ifirsttid
Definition: heapam.c:196
ItemPointerData tid
Definition: tableam.h:189

References Assert(), b, BOTTOMUP_MAX_NBLOCKS, BOTTOMUP_TOLERANCE_NBLOCKS, IndexDeleteCounts::ifirsttid, ItemPointerGetBlockNumber, and TM_IndexDelete::tid.

Referenced by bottomup_sort_and_shrink().

◆ bottomup_sort_and_shrink()

static int bottomup_sort_and_shrink ( TM_IndexDeleteOp delstate)
static

Definition at line 7873 of file heapam.c.

7874 {
7875  IndexDeleteCounts *blockgroups;
7876  TM_IndexDelete *reordereddeltids;
7877  BlockNumber curblock = InvalidBlockNumber;
7878  int nblockgroups = 0;
7879  int ncopied = 0;
7880  int nblocksfavorable = 0;
7881 
7882  Assert(delstate->bottomup);
7883  Assert(delstate->ndeltids > 0);
7884 
7885  /* Calculate per-heap-block count of TIDs */
7886  blockgroups = palloc(sizeof(IndexDeleteCounts) * delstate->ndeltids);
7887  for (int i = 0; i < delstate->ndeltids; i++)
7888  {
7889  TM_IndexDelete *ideltid = &delstate->deltids[i];
7890  TM_IndexStatus *istatus = delstate->status + ideltid->id;
7891  ItemPointer htid = &ideltid->tid;
7892  bool promising = istatus->promising;
7893 
7894  if (curblock != ItemPointerGetBlockNumber(htid))
7895  {
7896  /* New block group */
7897  nblockgroups++;
7898 
7899  Assert(curblock < ItemPointerGetBlockNumber(htid) ||
7900  !BlockNumberIsValid(curblock));
7901 
7902  curblock = ItemPointerGetBlockNumber(htid);
7903  blockgroups[nblockgroups - 1].ifirsttid = i;
7904  blockgroups[nblockgroups - 1].ntids = 1;
7905  blockgroups[nblockgroups - 1].npromisingtids = 0;
7906  }
7907  else
7908  {
7909  blockgroups[nblockgroups - 1].ntids++;
7910  }
7911 
7912  if (promising)
7913  blockgroups[nblockgroups - 1].npromisingtids++;
7914  }
7915 
7916  /*
7917  * We're about ready to sort block groups to determine the optimal order
7918  * for visiting heap blocks. But before we do, round the number of
7919  * promising tuples for each block group up to the next power-of-two,
7920  * unless it is very low (less than 4), in which case we round up to 4.
7921  * npromisingtids is far too noisy to trust when choosing between a pair
7922  * of block groups that both have very low values.
7923  *
7924  * This scheme divides heap blocks/block groups into buckets. Each bucket
7925  * contains blocks that have _approximately_ the same number of promising
7926  * TIDs as each other. The goal is to ignore relatively small differences
7927  * in the total number of promising entries, so that the whole process can
7928  * give a little weight to heapam factors (like heap block locality)
7929  * instead. This isn't a trade-off, really -- we have nothing to lose. It
7930  * would be foolish to interpret small differences in npromisingtids
7931  * values as anything more than noise.
7932  *
7933  * We tiebreak on nhtids when sorting block group subsets that have the
7934  * same npromisingtids, but this has the same issues as npromisingtids,
7935  * and so nhtids is subject to the same power-of-two bucketing scheme. The
7936  * only reason that we don't fix nhtids in the same way here too is that
7937  * we'll need accurate nhtids values after the sort. We handle nhtids
7938  * bucketization dynamically instead (in the sort comparator).
7939  *
7940  * See bottomup_nblocksfavorable() for a full explanation of when and how
7941  * heap locality/favorable blocks can significantly influence when and how
7942  * heap blocks are accessed.
7943  */
7944  for (int b = 0; b < nblockgroups; b++)
7945  {
7946  IndexDeleteCounts *group = blockgroups + b;
7947 
7948  /* Better off falling back on nhtids with low npromisingtids */
7949  if (group->npromisingtids <= 4)
7950  group->npromisingtids = 4;
7951  else
7952  group->npromisingtids =
7954  }
7955 
7956  /* Sort groups and rearrange caller's deltids array */
7957  qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
7959  reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
7960 
7961  nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
7962  /* Determine number of favorable blocks at the start of final deltids */
7963  nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
7964  delstate->deltids);
7965 
7966  for (int b = 0; b < nblockgroups; b++)
7967  {
7968  IndexDeleteCounts *group = blockgroups + b;
7969  TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
7970 
7971  memcpy(reordereddeltids + ncopied, firstdtid,
7972  sizeof(TM_IndexDelete) * group->ntids);
7973  ncopied += group->ntids;
7974  }
7975 
7976  /* Copy final grouped and sorted TIDs back into start of caller's array */
7977  memcpy(delstate->deltids, reordereddeltids,
7978  sizeof(TM_IndexDelete) * ncopied);
7979  delstate->ndeltids = ncopied;
7980 
7981  pfree(reordereddeltids);
7982  pfree(blockgroups);
7983 
7984  return nblocksfavorable;
7985 }
#define InvalidBlockNumber
Definition: block.h:33
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
unsigned int uint32
Definition: c.h:441
#define Min(x, y)
Definition: c.h:986
static int bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups, TM_IndexDelete *deltids)
Definition: heapam.c:7757
static int bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
Definition: heapam.c:7800
int i
Definition: isn.c:73
void pfree(void *pointer)
Definition: mcxt.c:1169
void * palloc(Size size)
Definition: mcxt.c:1062
static uint32 pg_nextpower2_32(uint32 num)
Definition: pg_bitutils.h:140
#define qsort(a, b, c, d)
Definition: port.h:495
int16 npromisingtids
Definition: heapam.c:194
TM_IndexStatus * status
Definition: tableam.h:231
TM_IndexDelete * deltids
Definition: tableam.h:230
bool promising
Definition: tableam.h:199

References Assert(), b, BlockNumberIsValid, TM_IndexDeleteOp::bottomup, BOTTOMUP_MAX_NBLOCKS, bottomup_nblocksfavorable(), bottomup_sort_and_shrink_cmp(), TM_IndexDeleteOp::deltids, i, TM_IndexDelete::id, IndexDeleteCounts::ifirsttid, InvalidBlockNumber, ItemPointerGetBlockNumber, Min, TM_IndexDeleteOp::ndeltids, IndexDeleteCounts::npromisingtids, IndexDeleteCounts::ntids, palloc(), pfree(), pg_nextpower2_32(), TM_IndexStatus::promising, qsort, TM_IndexDeleteOp::status, and TM_IndexDelete::tid.

Referenced by heap_index_delete_tuples().

◆ bottomup_sort_and_shrink_cmp()

static int bottomup_sort_and_shrink_cmp ( const void *  arg1,
const void *  arg2 
)
static

Definition at line 7800 of file heapam.c.

7801 {
7802  const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
7803  const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
7804 
7805  /*
7806  * Most significant field is npromisingtids (which we invert the order of
7807  * so as to sort in desc order).
7808  *
7809  * Caller should have already normalized npromisingtids fields into
7810  * power-of-two values (buckets).
7811  */
7812  if (group1->npromisingtids > group2->npromisingtids)
7813  return -1;
7814  if (group1->npromisingtids < group2->npromisingtids)
7815  return 1;
7816 
7817  /*
7818  * Tiebreak: desc ntids sort order.
7819  *
7820  * We cannot expect power-of-two values for ntids fields. We should
7821  * behave as if they were already rounded up for us instead.
7822  */
7823  if (group1->ntids != group2->ntids)
7824  {
7825  uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
7826  uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
7827 
7828  if (ntids1 > ntids2)
7829  return -1;
7830  if (ntids1 < ntids2)
7831  return 1;
7832  }
7833 
7834  /*
7835  * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
7836  * block in deltids array) order.
7837  *
7838  * This is equivalent to sorting in ascending heap block number order
7839  * (among otherwise equal subsets of the array). This approach allows us
7840  * to avoid accessing the out-of-line TID. (We rely on the assumption
7841  * that the deltids array was sorted in ascending heap TID order when
7842  * these offsets to the first TID from each heap block group were formed.)
7843  */
7844  if (group1->ifirsttid > group2->ifirsttid)
7845  return 1;
7846  if (group1->ifirsttid < group2->ifirsttid)
7847  return -1;
7848 
7849  pg_unreachable();
7850 
7851  return 0;
7852 }
#define pg_unreachable()
Definition: c.h:258

References IndexDeleteCounts::ifirsttid, IndexDeleteCounts::npromisingtids, IndexDeleteCounts::ntids, pg_nextpower2_32(), and pg_unreachable.

Referenced by bottomup_sort_and_shrink().

◆ compute_infobits()

static uint8 compute_infobits ( uint16  infomask,
uint16  infomask2 
)
static

Definition at line 2655 of file heapam.c.

2656 {
2657  return
2658  ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2659  ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2660  ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2661  /* note we ignore HEAP_XMAX_SHR_LOCK here */
2662  ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2663  ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2664  XLHL_KEYS_UPDATED : 0);
2665 }
#define XLHL_XMAX_KEYSHR_LOCK
Definition: heapam_xlog.h:271
#define XLHL_XMAX_IS_MULTI
Definition: heapam_xlog.h:268
#define XLHL_XMAX_LOCK_ONLY
Definition: heapam_xlog.h:269
#define XLHL_XMAX_EXCL_LOCK
Definition: heapam_xlog.h:270
#define XLHL_KEYS_UPDATED
Definition: heapam_xlog.h:272
#define HEAP_KEYS_UPDATED
Definition: htup_details.h:274
#define HEAP_XMAX_LOCK_ONLY
Definition: htup_details.h:196
#define HEAP_XMAX_IS_MULTI
Definition: htup_details.h:208
#define HEAP_XMAX_EXCL_LOCK
Definition: htup_details.h:195
#define HEAP_XMAX_KEYSHR_LOCK
Definition: htup_details.h:193

References HEAP_KEYS_UPDATED, HEAP_XMAX_EXCL_LOCK, HEAP_XMAX_IS_MULTI, HEAP_XMAX_KEYSHR_LOCK, HEAP_XMAX_LOCK_ONLY, XLHL_KEYS_UPDATED, XLHL_XMAX_EXCL_LOCK, XLHL_XMAX_IS_MULTI, XLHL_XMAX_KEYSHR_LOCK, and XLHL_XMAX_LOCK_ONLY.

Referenced by heap_abort_speculative(), heap_delete(), heap_lock_tuple(), heap_lock_updated_tuple_rec(), heap_update(), and log_heap_update().

◆ compute_new_xmax_infomask()

static void compute_new_xmax_infomask ( TransactionId  xmax,
uint16  old_infomask,
uint16  old_infomask2,
TransactionId  add_to_xmax,
LockTupleMode  mode,
bool  is_update,
TransactionId result_xmax,
uint16 result_infomask,
uint16 result_infomask2 
)
static

Definition at line 4975 of file heapam.c.

4980 {
4981  TransactionId new_xmax;
4982  uint16 new_infomask,
4983  new_infomask2;
4984 
4986 
4987 l5:
4988  new_infomask = 0;
4989  new_infomask2 = 0;
4990  if (old_infomask & HEAP_XMAX_INVALID)
4991  {
4992  /*
4993  * No previous locker; we just insert our own TransactionId.
4994  *
4995  * Note that it's critical that this case be the first one checked,
4996  * because there are several blocks below that come back to this one
4997  * to implement certain optimizations; old_infomask might contain
4998  * other dirty bits in those cases, but we don't really care.
4999  */
5000  if (is_update)
5001  {
5002  new_xmax = add_to_xmax;
5003  if (mode == LockTupleExclusive)
5004  new_infomask2 |= HEAP_KEYS_UPDATED;
5005  }
5006  else
5007  {
5008  new_infomask |= HEAP_XMAX_LOCK_ONLY;
5009  switch (mode)
5010  {
5011  case LockTupleKeyShare:
5012  new_xmax = add_to_xmax;
5013  new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5014  break;
5015  case LockTupleShare:
5016  new_xmax = add_to_xmax;
5017  new_infomask |= HEAP_XMAX_SHR_LOCK;
5018  break;
5020  new_xmax = add_to_xmax;
5021  new_infomask |= HEAP_XMAX_EXCL_LOCK;
5022  break;
5023  case LockTupleExclusive:
5024  new_xmax = add_to_xmax;
5025  new_infomask |= HEAP_XMAX_EXCL_LOCK;
5026  new_infomask2 |= HEAP_KEYS_UPDATED;
5027  break;
5028  default:
5029  new_xmax = InvalidTransactionId; /* silence compiler */
5030  elog(ERROR, "invalid lock mode");
5031  }
5032  }
5033  }
5034  else if (old_infomask & HEAP_XMAX_IS_MULTI)
5035  {
5036  MultiXactStatus new_status;
5037 
5038  /*
5039  * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5040  * cross-check.
5041  */
5042  Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5043 
5044  /*
5045  * A multixact together with LOCK_ONLY set but neither lock bit set
5046  * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5047  * anymore. This check is critical for databases upgraded by
5048  * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5049  * that such multis are never passed.
5050  */
5051  if (HEAP_LOCKED_UPGRADED(old_infomask))
5052  {
5053  old_infomask &= ~HEAP_XMAX_IS_MULTI;
5054  old_infomask |= HEAP_XMAX_INVALID;
5055  goto l5;
5056  }
5057 
5058  /*
5059  * If the XMAX is already a MultiXactId, then we need to expand it to
5060  * include add_to_xmax; but if all the members were lockers and are
5061  * all gone, we can do away with the IS_MULTI bit and just set
5062  * add_to_xmax as the only locker/updater. If all lockers are gone
5063  * and we have an updater that aborted, we can also do without a
5064  * multi.
5065  *
5066  * The cost of doing GetMultiXactIdMembers would be paid by
5067  * MultiXactIdExpand if we weren't to do this, so this check is not
5068  * incurring extra work anyhow.
5069  */
5070  if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5071  {
5072  if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5074  old_infomask)))
5075  {
5076  /*
5077  * Reset these bits and restart; otherwise fall through to
5078  * create a new multi below.
5079  */
5080  old_infomask &= ~HEAP_XMAX_IS_MULTI;
5081  old_infomask |= HEAP_XMAX_INVALID;
5082  goto l5;
5083  }
5084  }
5085 
5086  new_status = get_mxact_status_for_lock(mode, is_update);
5087 
5088  new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5089  new_status);
5090  GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5091  }
5092  else if (old_infomask & HEAP_XMAX_COMMITTED)
5093  {
5094  /*
5095  * It's a committed update, so we need to preserve him as updater of
5096  * the tuple.
5097  */
5099  MultiXactStatus new_status;
5100 
5101  if (old_infomask2 & HEAP_KEYS_UPDATED)
5103  else
5105 
5106  new_status = get_mxact_status_for_lock(mode, is_update);
5107 
5108  /*
5109  * since it's not running, it's obviously impossible for the old
5110  * updater to be identical to the current one, so we need not check
5111  * for that case as we do in the block above.
5112  */
5113  new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5114  GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5115  }
5116  else if (TransactionIdIsInProgress(xmax))
5117  {
5118  /*
5119  * If the XMAX is a valid, in-progress TransactionId, then we need to
5120  * create a new MultiXactId that includes both the old locker or
5121  * updater and our own TransactionId.
5122  */
5123  MultiXactStatus new_status;
5124  MultiXactStatus old_status;
5125  LockTupleMode old_mode;
5126 
5127  if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5128  {
5129  if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5130  old_status = MultiXactStatusForKeyShare;
5131  else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5132  old_status = MultiXactStatusForShare;
5133  else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5134  {
5135  if (old_infomask2 & HEAP_KEYS_UPDATED)
5136  old_status = MultiXactStatusForUpdate;
5137  else
5138  old_status = MultiXactStatusForNoKeyUpdate;
5139  }
5140  else
5141  {
5142  /*
5143  * LOCK_ONLY can be present alone only when a page has been
5144  * upgraded by pg_upgrade. But in that case,
5145  * TransactionIdIsInProgress() should have returned false. We
5146  * assume it's no longer locked in this case.
5147  */
5148  elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5149  old_infomask |= HEAP_XMAX_INVALID;
5150  old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5151  goto l5;
5152  }
5153  }
5154  else
5155  {
5156  /* it's an update, but which kind? */
5157  if (old_infomask2 & HEAP_KEYS_UPDATED)
5158  old_status = MultiXactStatusUpdate;
5159  else
5160  old_status = MultiXactStatusNoKeyUpdate;
5161  }
5162 
5163  old_mode = TUPLOCK_from_mxstatus(old_status);
5164 
5165  /*
5166  * If the lock to be acquired is for the same TransactionId as the
5167  * existing lock, there's an optimization possible: consider only the
5168  * strongest of both locks as the only one present, and restart.
5169  */
5170  if (xmax == add_to_xmax)
5171  {
5172  /*
5173  * Note that it's not possible for the original tuple to be
5174  * updated: we wouldn't be here because the tuple would have been
5175  * invisible and we wouldn't try to update it. As a subtlety,
5176  * this code can also run when traversing an update chain to lock
5177  * future versions of a tuple. But we wouldn't be here either,
5178  * because the add_to_xmax would be different from the original
5179  * updater.
5180  */
5181  Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5182 
5183  /* acquire the strongest of both */
5184  if (mode < old_mode)
5185  mode = old_mode;
5186  /* mustn't touch is_update */
5187 
5188  old_infomask |= HEAP_XMAX_INVALID;
5189  goto l5;
5190  }
5191 
5192  /* otherwise, just fall back to creating a new multixact */
5193  new_status = get_mxact_status_for_lock(mode, is_update);
5194  new_xmax = MultiXactIdCreate(xmax, old_status,
5195  add_to_xmax, new_status);
5196  GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5197  }
5198  else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5199  TransactionIdDidCommit(xmax))
5200  {
5201  /*
5202  * It's a committed update, so we gotta preserve him as updater of the
5203  * tuple.
5204  */
5206  MultiXactStatus new_status;
5207 
5208  if (old_infomask2 & HEAP_KEYS_UPDATED)
5210  else
5212 
5213  new_status = get_mxact_status_for_lock(mode, is_update);
5214 
5215  /*
5216  * since it's not running, it's obviously impossible for the old
5217  * updater to be identical to the current one, so we need not check
5218  * for that case as we do in the block above.
5219  */
5220  new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5221  GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5222  }
5223  else
5224  {
5225  /*
5226  * Can get here iff the locking/updating transaction was running when
5227  * the infomask was extracted from the tuple, but finished before
5228  * TransactionIdIsInProgress got to run. Deal with it as if there was
5229  * no locker at all in the first place.
5230  */
5231  old_infomask |= HEAP_XMAX_INVALID;
5232  goto l5;
5233  }
5234 
5235  *result_infomask = new_infomask;
5236  *result_infomask2 = new_infomask2;
5237  *result_xmax = new_xmax;
5238 }
unsigned short uint16
Definition: c.h:440
TransactionId MultiXactId
Definition: c.h:597
uint32 TransactionId
Definition: c.h:587
#define WARNING
Definition: elog.h:30
#define ERROR
Definition: elog.h:33
#define elog(elevel,...)
Definition: elog.h:218
static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask, uint16 *new_infomask2)
Definition: heapam.c:6673
#define TUPLOCK_from_mxstatus(status)
Definition: heapam.c:214
static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
Definition: heapam.c:6754
static MultiXactStatus get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
Definition: heapam.c:4180
#define HEAP_XMAX_IS_EXCL_LOCKED(infomask)
Definition: htup_details.h:260
#define HEAP_XMAX_SHR_LOCK
Definition: htup_details.h:199
#define HEAP_XMAX_IS_LOCKED_ONLY(infomask)
Definition: htup_details.h:226
#define HEAP_XMAX_IS_KEYSHR_LOCKED(infomask)
Definition: htup_details.h:262
#define HEAP_XMAX_COMMITTED
Definition: htup_details.h:206
#define HEAP_XMAX_INVALID
Definition: htup_details.h:207
#define HEAP_XMAX_IS_SHR_LOCKED(infomask)
Definition: htup_details.h:258
#define HEAP_LOCKED_UPGRADED(infomask)
Definition: htup_details.h:248
LockTupleMode
Definition: lockoptions.h:50
@ LockTupleExclusive
Definition: lockoptions.h:58
@ LockTupleNoKeyExclusive
Definition: lockoptions.h:56
@ LockTupleShare
Definition: lockoptions.h:54
@ LockTupleKeyShare
Definition: lockoptions.h:52
MultiXactId MultiXactIdExpand(MultiXactId multi, TransactionId xid, MultiXactStatus status)
Definition: multixact.c:439
bool MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
Definition: multixact.c:551
MultiXactId MultiXactIdCreate(TransactionId xid1, MultiXactStatus status1, TransactionId xid2, MultiXactStatus status2)
Definition: multixact.c:386
MultiXactStatus
Definition: multixact.h:42
@ MultiXactStatusForShare
Definition: multixact.h:44
@ MultiXactStatusForNoKeyUpdate
Definition: multixact.h:45
@ MultiXactStatusNoKeyUpdate
Definition: multixact.h:48
@ MultiXactStatusUpdate
Definition: multixact.h:50
@ MultiXactStatusForUpdate
Definition: multixact.h:46
@ MultiXactStatusForKeyShare
Definition: multixact.h:43
static PgChecksumMode mode
Definition: pg_checksums.c:65
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:229
bool TransactionIdIsInProgress(TransactionId xid)
Definition: procarray.c:1359
bool TransactionIdDidCommit(TransactionId transactionId)
Definition: transam.c:125
#define InvalidTransactionId
Definition: transam.h:31
bool TransactionIdIsCurrentTransactionId(TransactionId xid)
Definition: xact.c:921

References Assert(), elog, ERROR, get_mxact_status_for_lock(), GetMultiXactIdHintBits(), HEAP_KEYS_UPDATED, HEAP_LOCKED_UPGRADED, HEAP_XMAX_COMMITTED, HEAP_XMAX_EXCL_LOCK, HEAP_XMAX_INVALID, HEAP_XMAX_IS_EXCL_LOCKED, HEAP_XMAX_IS_KEYSHR_LOCKED, HEAP_XMAX_IS_LOCKED_ONLY, HEAP_XMAX_IS_MULTI, HEAP_XMAX_IS_SHR_LOCKED, HEAP_XMAX_KEYSHR_LOCK, HEAP_XMAX_LOCK_ONLY, HEAP_XMAX_SHR_LOCK, InvalidTransactionId, LockTupleExclusive, LockTupleKeyShare, LockTupleNoKeyExclusive, LockTupleShare, mode, MultiXactIdCreate(), MultiXactIdExpand(), MultiXactIdGetUpdateXid(), MultiXactIdIsRunning(), MultiXactStatusForKeyShare, MultiXactStatusForNoKeyUpdate, MultiXactStatusForShare, MultiXactStatusForUpdate, MultiXactStatusNoKeyUpdate, MultiXactStatusUpdate, status(), TransactionIdDidCommit(), TransactionIdIsCurrentTransactionId(), TransactionIdIsInProgress(), TUPLOCK_from_mxstatus, and WARNING.

Referenced by heap_delete(), heap_lock_tuple(), heap_lock_updated_tuple_rec(), and heap_update().

◆ ConditionalMultiXactIdWait()

static bool ConditionalMultiXactIdWait ( MultiXactId  multi,
MultiXactStatus  status,
uint16  infomask,
Relation  rel,
int *  remaining 
)
static

Definition at line 7021 of file heapam.c.

7023 {
7024  return Do_MultiXactIdWait(multi, status, infomask, true,
7025  rel, NULL, XLTW_None, remaining);
7026 }
static bool Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask, bool nowait, Relation rel, ItemPointer ctid, XLTW_Oper oper, int *remaining)
Definition: heapam.c:6921
int remaining
Definition: informix.c:667
@ XLTW_None
Definition: lmgr.h:26

References Do_MultiXactIdWait(), remaining, status(), and XLTW_None.

Referenced by heap_lock_tuple().

◆ Do_MultiXactIdWait()

static bool Do_MultiXactIdWait ( MultiXactId  multi,
MultiXactStatus  status,
uint16  infomask,
bool  nowait,
Relation  rel,
ItemPointer  ctid,
XLTW_Oper  oper,
int *  remaining 
)
static

Definition at line 6921 of file heapam.c.

6925 {
6926  bool result = true;
6927  MultiXactMember *members;
6928  int nmembers;
6929  int remain = 0;
6930 
6931  /* for pre-pg_upgrade tuples, no need to sleep at all */
6932  nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
6933  GetMultiXactIdMembers(multi, &members, false,
6934  HEAP_XMAX_IS_LOCKED_ONLY(infomask));
6935 
6936  if (nmembers >= 0)
6937  {
6938  int i;
6939 
6940  for (i = 0; i < nmembers; i++)
6941  {
6942  TransactionId memxid = members[i].xid;
6943  MultiXactStatus memstatus = members[i].status;
6944 
6946  {
6947  remain++;
6948  continue;
6949  }
6950 
6953  {
6954  if (remaining && TransactionIdIsInProgress(memxid))
6955  remain++;
6956  continue;
6957  }
6958 
6959  /*
6960  * This member conflicts with our multi, so we have to sleep (or
6961  * return failure, if asked to avoid waiting.)
6962  *
6963  * Note that we don't set up an error context callback ourselves,
6964  * but instead we pass the info down to XactLockTableWait. This
6965  * might seem a bit wasteful because the context is set up and
6966  * tore down for each member of the multixact, but in reality it
6967  * should be barely noticeable, and it avoids duplicate code.
6968  */
6969  if (nowait)
6970  {
6971  result = ConditionalXactLockTableWait(memxid);
6972  if (!result)
6973  break;
6974  }
6975  else
6976  XactLockTableWait(memxid, rel, ctid, oper);
6977  }
6978 
6979  pfree(members);
6980  }
6981 
6982  if (remaining)
6983  *remaining = remain;
6984 
6985  return result;
6986 }
#define LOCKMODE_from_mxstatus(status)
Definition: heapam.c:155
void XactLockTableWait(TransactionId xid, Relation rel, ItemPointer ctid, XLTW_Oper oper)
Definition: lmgr.c:640
bool ConditionalXactLockTableWait(TransactionId xid)
Definition: lmgr.c:713
bool DoLockModesConflict(LOCKMODE mode1, LOCKMODE mode2)
Definition: lock.c:583
int GetMultiXactIdMembers(MultiXactId multi, MultiXactMember **members, bool from_pgupgrade, bool onlyLock)
Definition: multixact.c:1223
Operator oper(ParseState *pstate, List *opname, Oid ltypeId, Oid rtypeId, bool noError, int location)
Definition: parse_oper.c:382
TransactionId xid
Definition: multixact.h:62
MultiXactStatus status
Definition: multixact.h:63

References ConditionalXactLockTableWait(), DoLockModesConflict(), GetMultiXactIdMembers(), HEAP_LOCKED_UPGRADED, HEAP_XMAX_IS_LOCKED_ONLY, i, LOCKMODE_from_mxstatus, oper(), pfree(), remaining, MultiXactMember::status, status(), TransactionIdIsCurrentTransactionId(), TransactionIdIsInProgress(), XactLockTableWait(), and MultiXactMember::xid.

Referenced by ConditionalMultiXactIdWait(), and MultiXactIdWait().

◆ DoesMultiXactIdConflict()

static bool DoesMultiXactIdConflict ( MultiXactId  multi,
uint16  infomask,
LockTupleMode  lockmode,
bool current_is_member 
)
static

Definition at line 6822 of file heapam.c.

6824 {
6825  int nmembers;
6826  MultiXactMember *members;
6827  bool result = false;
6828  LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
6829 
6830  if (HEAP_LOCKED_UPGRADED(infomask))
6831  return false;
6832 
6833  nmembers = GetMultiXactIdMembers(multi, &members, false,
6834  HEAP_XMAX_IS_LOCKED_ONLY(infomask));
6835  if (nmembers >= 0)
6836  {
6837  int i;
6838 
6839  for (i = 0; i < nmembers; i++)
6840  {
6841  TransactionId memxid;
6842  LOCKMODE memlockmode;
6843 
6844  if (result && (current_is_member == NULL || *current_is_member))
6845  break;
6846 
6847  memlockmode = LOCKMODE_from_mxstatus(members[i].status);
6848 
6849  /* ignore members from current xact (but track their presence) */
6850  memxid = members[i].xid;
6852  {
6853  if (current_is_member != NULL)
6854  *current_is_member = true;
6855  continue;
6856  }
6857  else if (result)
6858  continue;
6859 
6860  /* ignore members that don't conflict with the lock we want */
6861  if (!DoLockModesConflict(memlockmode, wanted))
6862  continue;
6863 
6864  if (ISUPDATE_from_mxstatus(members[i].status))
6865  {
6866  /* ignore aborted updaters */
6867  if (TransactionIdDidAbort(memxid))
6868  continue;
6869  }
6870  else
6871  {
6872  /* ignore lockers-only that are no longer in progress */
6873  if (!TransactionIdIsInProgress(memxid))
6874  continue;
6875  }
6876 
6877  /*
6878  * Whatever remains are either live lockers that conflict with our
6879  * wanted lock, and updaters that are not aborted. Those conflict
6880  * with what we want. Set up to return true, but keep going to
6881  * look for the current transaction among the multixact members,
6882  * if needed.
6883  */
6884  result = true;
6885  }
6886  pfree(members);
6887  }
6888 
6889  return result;
6890 }
static const struct @12 tupleLockExtraInfo[MaxLockTupleMode+1]
int LOCKMODE
Definition: lockdefs.h:26
#define ISUPDATE_from_mxstatus(status)
Definition: multixact.h:56
bool TransactionIdDidAbort(TransactionId transactionId)
Definition: transam.c:181

References DoLockModesConflict(), GetMultiXactIdMembers(), HEAP_LOCKED_UPGRADED, HEAP_XMAX_IS_LOCKED_ONLY, i, ISUPDATE_from_mxstatus, LOCKMODE_from_mxstatus, pfree(), status(), TransactionIdDidAbort(), TransactionIdIsCurrentTransactionId(), TransactionIdIsInProgress(), tupleLockExtraInfo, and MultiXactMember::xid.

Referenced by heap_delete(), heap_lock_tuple(), and heap_update().

◆ ExtractReplicaIdentity()

static HeapTuple ExtractReplicaIdentity ( Relation  rel,
HeapTuple  tup,
bool  key_changed,
bool copy 
)
static

Definition at line 8369 of file heapam.c.

8371 {
8372  TupleDesc desc = RelationGetDescr(relation);
8373  char replident = relation->rd_rel->relreplident;
8374  Bitmapset *idattrs;
8375  HeapTuple key_tuple;
8376  bool nulls[MaxHeapAttributeNumber];
8378 
8379  *copy = false;
8380 
8381  if (!RelationIsLogicallyLogged(relation))
8382  return NULL;
8383 
8384  if (replident == REPLICA_IDENTITY_NOTHING)
8385  return NULL;
8386 
8387  if (replident == REPLICA_IDENTITY_FULL)
8388  {
8389  /*
8390  * When logging the entire old tuple, it very well could contain
8391  * toasted columns. If so, force them to be inlined.
8392  */
8393  if (HeapTupleHasExternal(tp))
8394  {
8395  *copy = true;
8396  tp = toast_flatten_tuple(tp, desc);
8397  }
8398  return tp;
8399  }
8400 
8401  /* if the key hasn't changed and we're only logging the key, we're done */
8402  if (!key_changed)
8403  return NULL;
8404 
8405  /* find out the replica identity columns */
8406  idattrs = RelationGetIndexAttrBitmap(relation,
8408 
8409  /*
8410  * If there's no defined replica identity columns, treat as !key_changed.
8411  * (This case should not be reachable from heap_update, since that should
8412  * calculate key_changed accurately. But heap_delete just passes constant
8413  * true for key_changed, so we can hit this case in deletes.)
8414  */
8415  if (bms_is_empty(idattrs))
8416  return NULL;
8417 
8418  /*
8419  * Construct a new tuple containing only the replica identity columns,
8420  * with nulls elsewhere. While we're at it, assert that the replica
8421  * identity columns aren't null.
8422  */
8423  heap_deform_tuple(tp, desc, values, nulls);
8424 
8425  for (int i = 0; i < desc->natts; i++)
8426  {
8428  idattrs))
8429  Assert(!nulls[i]);
8430  else
8431  nulls[i] = true;
8432  }
8433 
8434  key_tuple = heap_form_tuple(desc, values, nulls);
8435  *copy = true;
8436 
8437  bms_free(idattrs);
8438 
8439  /*
8440  * If the tuple, which by here only contains indexed columns, still has
8441  * toasted columns, force them to be inlined. This is somewhat unlikely
8442  * since there's limits on the size of indexed columns, so we don't
8443  * duplicate toast_flatten_tuple()s functionality in the above loop over
8444  * the indexed columns, even if it would be more efficient.
8445  */
8446  if (HeapTupleHasExternal(key_tuple))
8447  {
8448  HeapTuple oldtup = key_tuple;
8449 
8450  key_tuple = toast_flatten_tuple(oldtup, desc);
8451  heap_freetuple(oldtup);
8452  }
8453 
8454  return key_tuple;
8455 }
void bms_free(Bitmapset *a)
Definition: bitmapset.c:208
bool bms_is_member(int x, const Bitmapset *a)
Definition: bitmapset.c:427
bool bms_is_empty(const Bitmapset *a)
Definition: bitmapset.c:703
static Datum values[MAXATTR]
Definition: bootstrap.c:156
HeapTuple toast_flatten_tuple(HeapTuple tup, TupleDesc tupleDesc)
Definition: heaptoast.c:350
HeapTuple heap_form_tuple(TupleDesc tupleDescriptor, Datum *values, bool *isnull)
Definition: heaptuple.c:1020
void heap_deform_tuple(HeapTuple tuple, TupleDesc tupleDesc, Datum *values, bool *isnull)
Definition: heaptuple.c:1249
void heap_freetuple(HeapTuple htup)
Definition: heaptuple.c:1338
#define HeapTupleHasExternal(tuple)
Definition: htup_details.h:667
#define MaxHeapAttributeNumber
Definition: htup_details.h:47
uintptr_t Datum
Definition: postgres.h:411
#define RelationIsLogicallyLogged(relation)
Definition: rel.h:675
#define RelationGetDescr(relation)
Definition: rel.h:504
Bitmapset * RelationGetIndexAttrBitmap(Relation relation, IndexAttrBitmapKind attrKind)
Definition: relcache.c:5079
@ INDEX_ATTR_BITMAP_IDENTITY_KEY
Definition: relcache.h:60
#define FirstLowInvalidHeapAttributeNumber
Definition: sysattr.h:27

References Assert(), bms_free(), bms_is_empty(), bms_is_member(), FirstLowInvalidHeapAttributeNumber, heap_deform_tuple(), heap_form_tuple(), heap_freetuple(), HeapTupleHasExternal, i, INDEX_ATTR_BITMAP_IDENTITY_KEY, MaxHeapAttributeNumber, TupleDescData::natts, RelationData::rd_rel, RelationGetDescr, RelationGetIndexAttrBitmap(), RelationIsLogicallyLogged, toast_flatten_tuple(), and values.

Referenced by heap_delete(), and heap_update().

◆ fix_infomask_from_infobits()

static void fix_infomask_from_infobits ( uint8  infobits,
uint16 infomask,
uint16 infomask2 
)
static

Definition at line 8822 of file heapam.c.

8823 {
8824  *infomask &= ~(HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY |
8826  *infomask2 &= ~HEAP_KEYS_UPDATED;
8827 
8828  if (infobits & XLHL_XMAX_IS_MULTI)
8829  *infomask |= HEAP_XMAX_IS_MULTI;
8830  if (infobits & XLHL_XMAX_LOCK_ONLY)
8831  *infomask |= HEAP_XMAX_LOCK_ONLY;
8832  if (infobits & XLHL_XMAX_EXCL_LOCK)
8833  *infomask |= HEAP_XMAX_EXCL_LOCK;
8834  /* note HEAP_XMAX_SHR_LOCK isn't considered here */
8835  if (infobits & XLHL_XMAX_KEYSHR_LOCK)
8836  *infomask |= HEAP_XMAX_KEYSHR_LOCK;
8837 
8838  if (infobits & XLHL_KEYS_UPDATED)
8839  *infomask2 |= HEAP_KEYS_UPDATED;
8840 }

References HEAP_KEYS_UPDATED, HEAP_XMAX_EXCL_LOCK, HEAP_XMAX_IS_MULTI, HEAP_XMAX_KEYSHR_LOCK, HEAP_XMAX_LOCK_ONLY, XLHL_KEYS_UPDATED, XLHL_XMAX_EXCL_LOCK, XLHL_XMAX_IS_MULTI, XLHL_XMAX_KEYSHR_LOCK, and XLHL_XMAX_LOCK_ONLY.

Referenced by heap_xlog_delete(), heap_xlog_lock(), heap_xlog_lock_updated(), and heap_xlog_update().

◆ FreeBulkInsertState()

void FreeBulkInsertState ( BulkInsertState  bistate)

Definition at line 2021 of file heapam.c.

2022 {
2023  if (bistate->current_buf != InvalidBuffer)
2024  ReleaseBuffer(bistate->current_buf);
2025  FreeAccessStrategy(bistate->strategy);
2026  pfree(bistate);
2027 }
#define InvalidBuffer
Definition: buf.h:25
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3757
void FreeAccessStrategy(BufferAccessStrategy strategy)
Definition: freelist.c:597
BufferAccessStrategy strategy
Definition: hio.h:31
Buffer current_buf
Definition: hio.h:32

References BulkInsertStateData::current_buf, FreeAccessStrategy(), InvalidBuffer, pfree(), ReleaseBuffer(), and BulkInsertStateData::strategy.

Referenced by ATRewriteTable(), CopyFrom(), CopyMultiInsertBufferCleanup(), intorel_shutdown(), and transientrel_shutdown().

◆ FreezeMultiXactId()

static TransactionId FreezeMultiXactId ( MultiXactId  multi,
uint16  t_infomask,
TransactionId  relfrozenxid,
TransactionId  relminmxid,
TransactionId  cutoff_xid,
MultiXactId  cutoff_multi,
uint16 flags 
)
static

Definition at line 6092 of file heapam.c.

6096 {
6098  int i;
6099  MultiXactMember *members;
6100  int nmembers;
6101  bool need_replace;
6102  int nnewmembers;
6103  MultiXactMember *newmembers;
6104  bool has_lockers;
6105  TransactionId update_xid;
6106  bool update_committed;
6107 
6108  *flags = 0;
6109 
6110  /* We should only be called in Multis */
6111  Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6112 
6113  if (!MultiXactIdIsValid(multi) ||
6114  HEAP_LOCKED_UPGRADED(t_infomask))
6115  {
6116  /* Ensure infomask bits are appropriately set/reset */
6117  *flags |= FRM_INVALIDATE_XMAX;
6118  return InvalidTransactionId;
6119  }
6120  else if (MultiXactIdPrecedes(multi, relminmxid))
6121  ereport(ERROR,
6123  errmsg_internal("found multixact %u from before relminmxid %u",
6124  multi, relminmxid)));
6125  else if (MultiXactIdPrecedes(multi, cutoff_multi))
6126  {
6127  /*
6128  * This old multi cannot possibly have members still running, but
6129  * verify just in case. If it was a locker only, it can be removed
6130  * without any further consideration; but if it contained an update,
6131  * we might need to preserve it.
6132  */
6133  if (MultiXactIdIsRunning(multi,
6134  HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6135  ereport(ERROR,
6137  errmsg_internal("multixact %u from before cutoff %u found to be still running",
6138  multi, cutoff_multi)));
6139 
6140  if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6141  {
6142  *flags |= FRM_INVALIDATE_XMAX;
6143  xid = InvalidTransactionId; /* not strictly necessary */
6144  }
6145  else
6146  {
6147  /* replace multi by update xid */
6148  xid = MultiXactIdGetUpdateXid(multi, t_infomask);
6149 
6150  /* wasn't only a lock, xid needs to be valid */
6152 
6153  if (TransactionIdPrecedes(xid, relfrozenxid))
6154  ereport(ERROR,
6156  errmsg_internal("found update xid %u from before relfrozenxid %u",
6157  xid, relfrozenxid)));
6158 
6159  /*
6160  * If the xid is older than the cutoff, it has to have aborted,
6161  * otherwise the tuple would have gotten pruned away.
6162  */
6163  if (TransactionIdPrecedes(xid, cutoff_xid))
6164  {
6165  if (TransactionIdDidCommit(xid))
6166  ereport(ERROR,
6168  errmsg_internal("cannot freeze committed update xid %u", xid)));
6169  *flags |= FRM_INVALIDATE_XMAX;
6170  xid = InvalidTransactionId; /* not strictly necessary */
6171  }
6172  else
6173  {
6174  *flags |= FRM_RETURN_IS_XID;
6175  }
6176  }
6177 
6178  return xid;
6179  }
6180 
6181  /*
6182  * This multixact might have or might not have members still running, but
6183  * we know it's valid and is newer than the cutoff point for multis.
6184  * However, some member(s) of it may be below the cutoff for Xids, so we
6185  * need to walk the whole members array to figure out what to do, if
6186  * anything.
6187  */
6188 
6189  nmembers =
6190  GetMultiXactIdMembers(multi, &members, false,
6191  HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6192  if (nmembers <= 0)
6193  {
6194  /* Nothing worth keeping */
6195  *flags |= FRM_INVALIDATE_XMAX;
6196  return InvalidTransactionId;
6197  }
6198 
6199  /* is there anything older than the cutoff? */
6200  need_replace = false;
6201  for (i = 0; i < nmembers; i++)
6202  {
6203  if (TransactionIdPrecedes(members[i].xid, cutoff_xid))
6204  {
6205  need_replace = true;
6206  break;
6207  }
6208  }
6209 
6210  /*
6211  * In the simplest case, there is no member older than the cutoff; we can
6212  * keep the existing MultiXactId as is.
6213  */
6214  if (!need_replace)
6215  {
6216  *flags |= FRM_NOOP;
6217  pfree(members);
6218  return InvalidTransactionId;
6219  }
6220 
6221  /*
6222  * If the multi needs to be updated, figure out which members do we need
6223  * to keep.
6224  */
6225  nnewmembers = 0;
6226  newmembers = palloc(sizeof(MultiXactMember) * nmembers);
6227  has_lockers = false;
6228  update_xid = InvalidTransactionId;
6229  update_committed = false;
6230 
6231  for (i = 0; i < nmembers; i++)
6232  {
6233  /*
6234  * Determine whether to keep this member or ignore it.
6235  */
6236  if (ISUPDATE_from_mxstatus(members[i].status))
6237  {
6238  TransactionId xid = members[i].xid;
6239 
6241  if (TransactionIdPrecedes(xid, relfrozenxid))
6242  ereport(ERROR,
6244  errmsg_internal("found update xid %u from before relfrozenxid %u",
6245  xid, relfrozenxid)));
6246 
6247  /*
6248  * It's an update; should we keep it? If the transaction is known
6249  * aborted or crashed then it's okay to ignore it, otherwise not.
6250  * Note that an updater older than cutoff_xid cannot possibly be
6251  * committed, because HeapTupleSatisfiesVacuum would have returned
6252  * HEAPTUPLE_DEAD and we would not be trying to freeze the tuple.
6253  *
6254  * As with all tuple visibility routines, it's critical to test
6255  * TransactionIdIsInProgress before TransactionIdDidCommit,
6256  * because of race conditions explained in detail in
6257  * heapam_visibility.c.
6258  */
6261  {
6262  Assert(!TransactionIdIsValid(update_xid));
6263  update_xid = xid;
6264  }
6265  else if (TransactionIdDidCommit(xid))
6266  {
6267  /*
6268  * The transaction committed, so we can tell caller to set
6269  * HEAP_XMAX_COMMITTED. (We can only do this because we know
6270  * the transaction is not running.)
6271  */
6272  Assert(!TransactionIdIsValid(update_xid));
6273  update_committed = true;
6274  update_xid = xid;
6275  }
6276  else
6277  {
6278  /*
6279  * Not in progress, not committed -- must be aborted or
6280  * crashed; we can ignore it.
6281  */
6282  }
6283 
6284  /*
6285  * Since the tuple wasn't marked HEAPTUPLE_DEAD by vacuum, the
6286  * update Xid cannot possibly be older than the xid cutoff. The
6287  * presence of such a tuple would cause corruption, so be paranoid
6288  * and check.
6289  */
6290  if (TransactionIdIsValid(update_xid) &&
6291  TransactionIdPrecedes(update_xid, cutoff_xid))
6292  ereport(ERROR,
6294  errmsg_internal("found update xid %u from before xid cutoff %u",
6295  update_xid, cutoff_xid)));
6296 
6297  /*
6298  * If we determined that it's an Xid corresponding to an update
6299  * that must be retained, additionally add it to the list of
6300  * members of the new Multi, in case we end up using that. (We
6301  * might still decide to use only an update Xid and not a multi,
6302  * but it's easier to maintain the list as we walk the old members
6303  * list.)
6304  */
6305  if (TransactionIdIsValid(update_xid))
6306  newmembers[nnewmembers++] = members[i];
6307  }
6308  else
6309  {
6310  /* We only keep lockers if they are still running */
6311  if (TransactionIdIsCurrentTransactionId(members[i].xid) ||
6312  TransactionIdIsInProgress(members[i].xid))
6313  {
6314  /* running locker cannot possibly be older than the cutoff */
6315  Assert(!TransactionIdPrecedes(members[i].xid, cutoff_xid));
6316  newmembers[nnewmembers++] = members[i];
6317  has_lockers = true;
6318  }
6319  }
6320  }
6321 
6322  pfree(members);
6323 
6324  if (nnewmembers == 0)
6325  {
6326  /* nothing worth keeping!? Tell caller to remove the whole thing */
6327  *flags |= FRM_INVALIDATE_XMAX;
6328  xid = InvalidTransactionId;
6329  }
6330  else if (TransactionIdIsValid(update_xid) && !has_lockers)
6331  {
6332  /*
6333  * If there's a single member and it's an update, pass it back alone
6334  * without creating a new Multi. (XXX we could do this when there's a
6335  * single remaining locker, too, but that would complicate the API too
6336  * much; moreover, the case with the single updater is more
6337  * interesting, because those are longer-lived.)
6338  */
6339  Assert(nnewmembers == 1);
6340  *flags |= FRM_RETURN_IS_XID;
6341  if (update_committed)
6342  *flags |= FRM_MARK_COMMITTED;
6343  xid = update_xid;
6344  }
6345  else
6346  {
6347  /*
6348  * Create a new multixact with the surviving members of the previous
6349  * one, to set as new Xmax in the tuple.
6350  */
6351  xid = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
6352  *flags |= FRM_RETURN_IS_MULTI;
6353  }
6354 
6355  pfree(newmembers);
6356 
6357  return xid;
6358 }
int errmsg_internal(const char *fmt,...)
Definition: elog.c:991
int errcode(int sqlerrcode)
Definition: elog.c:693
#define ereport(elevel,...)
Definition: elog.h:143
#define FRM_RETURN_IS_XID
Definition: heapam.c:6066
#define FRM_MARK_COMMITTED
Definition: heapam.c:6068
#define FRM_NOOP
Definition: heapam.c:6064
#define FRM_RETURN_IS_MULTI
Definition: heapam.c:6067
#define FRM_INVALIDATE_XMAX
Definition: heapam.c:6065
bool MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
Definition: multixact.c:3159
MultiXactId MultiXactIdCreateFromMembers(int nmembers, MultiXactMember *members)
Definition: multixact.c:767
#define MultiXactIdIsValid(multi)
Definition: multixact.h:28
#define ERRCODE_DATA_CORRUPTED
Definition: pg_basebackup.c:42
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
#define TransactionIdIsValid(xid)
Definition: transam.h:41

References Assert(), ereport, errcode(), ERRCODE_DATA_CORRUPTED, errmsg_internal(), ERROR, FRM_INVALIDATE_XMAX, FRM_MARK_COMMITTED, FRM_NOOP, FRM_RETURN_IS_MULTI, FRM_RETURN_IS_XID, GetMultiXactIdMembers(), HEAP_LOCKED_UPGRADED, HEAP_XMAX_IS_LOCKED_ONLY, HEAP_XMAX_IS_MULTI, i, InvalidTransactionId, ISUPDATE_from_mxstatus, MultiXactIdCreateFromMembers(), MultiXactIdGetUpdateXid(), MultiXactIdIsRunning(), MultiXactIdIsValid, MultiXactIdPrecedes(), palloc(), pfree(), status(), TransactionIdDidCommit(), TransactionIdIsCurrentTransactionId(), TransactionIdIsInProgress(), TransactionIdIsValid, TransactionIdPrecedes(), and MultiXactMember::xid.

Referenced by heap_prepare_freeze_tuple().

◆ get_mxact_status_for_lock()

static MultiXactStatus get_mxact_status_for_lock ( LockTupleMode  mode,
bool  is_update 
)
static

Definition at line 4180 of file heapam.c.

4181 {
4182  int retval;
4183 
4184  if (is_update)
4185  retval = tupleLockExtraInfo[mode].updstatus;
4186  else
4187  retval = tupleLockExtraInfo[mode].lockstatus;
4188 
4189  if (retval == -1)
4190  elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4191  is_update ? "true" : "false");
4192 
4193  return (MultiXactStatus) retval;
4194 }

References elog, ERROR, mode, and tupleLockExtraInfo.

Referenced by compute_new_xmax_infomask(), heap_lock_tuple(), and test_lockmode_for_conflict().

◆ GetBulkInsertState()

BulkInsertState GetBulkInsertState ( void  )

Definition at line 2007 of file heapam.c.

2008 {
2009  BulkInsertState bistate;
2010 
2011  bistate = (BulkInsertState) palloc(sizeof(BulkInsertStateData));
2013  bistate->current_buf = InvalidBuffer;
2014  return bistate;
2015 }
@ BAS_BULKWRITE
Definition: bufmgr.h:32
BufferAccessStrategy GetAccessStrategy(BufferAccessStrategyType btype)
Definition: freelist.c:542
struct BulkInsertStateData * BulkInsertState
Definition: heapam.h:39

References BAS_BULKWRITE, BulkInsertStateData::current_buf, GetAccessStrategy(), InvalidBuffer, palloc(), and BulkInsertStateData::strategy.

Referenced by ATRewriteTable(), CopyFrom(), CopyMultiInsertBufferInit(), intorel_startup(), and transientrel_startup().

◆ GetMultiXactIdHintBits()

static void GetMultiXactIdHintBits ( MultiXactId  multi,
uint16 new_infomask,
uint16 new_infomask2 
)
static

Definition at line 6673 of file heapam.c.

6675 {
6676  int nmembers;
6677  MultiXactMember *members;
6678  int i;
6679  uint16 bits = HEAP_XMAX_IS_MULTI;
6680  uint16 bits2 = 0;
6681  bool has_update = false;
6682  LockTupleMode strongest = LockTupleKeyShare;
6683 
6684  /*
6685  * We only use this in multis we just created, so they cannot be values
6686  * pre-pg_upgrade.
6687  */
6688  nmembers = GetMultiXactIdMembers(multi, &members, false, false);
6689 
6690  for (i = 0; i < nmembers; i++)
6691  {
6693 
6694  /*
6695  * Remember the strongest lock mode held by any member of the
6696  * multixact.
6697  */
6698  mode = TUPLOCK_from_mxstatus(members[i].status);
6699  if (mode > strongest)
6700  strongest = mode;
6701 
6702  /* See what other bits we need */
6703  switch (members[i].status)
6704  {
6708  break;
6709 
6711  bits2 |= HEAP_KEYS_UPDATED;
6712  break;
6713 
6715  has_update = true;
6716  break;
6717 
6718  case MultiXactStatusUpdate:
6719  bits2 |= HEAP_KEYS_UPDATED;
6720  has_update = true;
6721  break;
6722  }
6723  }
6724 
6725  if (strongest == LockTupleExclusive ||
6726  strongest == LockTupleNoKeyExclusive)
6727  bits |= HEAP_XMAX_EXCL_LOCK;
6728  else if (strongest == LockTupleShare)
6729  bits |= HEAP_XMAX_SHR_LOCK;
6730  else if (strongest == LockTupleKeyShare)
6731  bits |= HEAP_XMAX_KEYSHR_LOCK;
6732 
6733  if (!has_update)
6734  bits |= HEAP_XMAX_LOCK_ONLY;
6735 
6736  if (nmembers > 0)
6737  pfree(members);
6738 
6739  *new_infomask = bits;
6740  *new_infomask2 = bits2;
6741 }

References GetMultiXactIdMembers(), HEAP_KEYS_UPDATED, HEAP_XMAX_EXCL_LOCK, HEAP_XMAX_IS_MULTI, HEAP_XMAX_KEYSHR_LOCK, HEAP_XMAX_LOCK_ONLY, HEAP_XMAX_SHR_LOCK, i, LockTupleExclusive, LockTupleKeyShare, LockTupleNoKeyExclusive, LockTupleShare, mode, MultiXactStatusForKeyShare, MultiXactStatusForNoKeyUpdate, MultiXactStatusForShare, MultiXactStatusForUpdate, MultiXactStatusNoKeyUpdate, MultiXactStatusUpdate, pfree(), status(), and TUPLOCK_from_mxstatus.

Referenced by compute_new_xmax_infomask(), heap_prepare_freeze_tuple(), and heap_update().

◆ heap2_redo()

void heap2_redo ( XLogReaderState record)

Definition at line 9709 of file heapam.c.

9710 {
9711  uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
9712 
9713  switch (info & XLOG_HEAP_OPMASK)
9714  {
9715  case XLOG_HEAP2_PRUNE:
9716  heap_xlog_prune(record);
9717  break;
9718  case XLOG_HEAP2_VACUUM:
9719  heap_xlog_vacuum(record);
9720  break;
9722  heap_xlog_freeze_page(record);
9723  break;
9724  case XLOG_HEAP2_VISIBLE:
9725  heap_xlog_visible(record);
9726  break;
9728  heap_xlog_multi_insert(record);
9729  break;
9731  heap_xlog_lock_updated(record);
9732  break;
9733  case XLOG_HEAP2_NEW_CID:
9734 
9735  /*
9736  * Nothing to do on a real replay, only used during logical
9737  * decoding.
9738  */
9739  break;
9740  case XLOG_HEAP2_REWRITE:
9741  heap_xlog_logical_rewrite(record);
9742  break;
9743  default:
9744  elog(PANIC, "heap2_redo: unknown op code %u", info);
9745  }
9746 }
unsigned char uint8
Definition: c.h:439
#define PANIC
Definition: elog.h:36
static void heap_xlog_prune(XLogReaderState *record)
Definition: heapam.c:8463
static void heap_xlog_vacuum(XLogReaderState *record)
Definition: heapam.c:8549
static void heap_xlog_lock_updated(XLogReaderState *record)
Definition: heapam.c:9562
static void heap_xlog_multi_insert(XLogReaderState *record)
Definition: heapam.c:9036
static void heap_xlog_visible(XLogReaderState *record)
Definition: heapam.c:8624
static void heap_xlog_freeze_page(XLogReaderState *record)
Definition: heapam.c:8764
#define XLOG_HEAP2_PRUNE
Definition: heapam_xlog.h:54
#define XLOG_HEAP2_MULTI_INSERT
Definition: heapam_xlog.h:58
#define XLOG_HEAP2_VACUUM
Definition: heapam_xlog.h:55
#define XLOG_HEAP2_REWRITE
Definition: heapam_xlog.h:53
#define XLOG_HEAP_OPMASK
Definition: heapam_xlog.h:41
#define XLOG_HEAP2_LOCK_UPDATED
Definition: heapam_xlog.h:59
#define XLOG_HEAP2_FREEZE_PAGE
Definition: heapam_xlog.h:56
#define XLOG_HEAP2_NEW_CID
Definition: heapam_xlog.h:60
#define XLOG_HEAP2_VISIBLE
Definition: heapam_xlog.h:57
void heap_xlog_logical_rewrite(XLogReaderState *r)
Definition: rewriteheap.c:1110
#define XLogRecGetInfo(decoder)
Definition: xlogreader.h:315
#define XLR_INFO_MASK
Definition: xlogrecord.h:62

References elog, heap_xlog_freeze_page(), heap_xlog_lock_updated(), heap_xlog_logical_rewrite(), heap_xlog_multi_insert(), heap_xlog_prune(), heap_xlog_vacuum(), heap_xlog_visible(), PANIC, XLOG_HEAP2_FREEZE_PAGE, XLOG_HEAP2_LOCK_UPDATED, XLOG_HEAP2_MULTI_INSERT, XLOG_HEAP2_NEW_CID, XLOG_HEAP2_PRUNE, XLOG_HEAP2_REWRITE, XLOG_HEAP2_VACUUM, XLOG_HEAP2_VISIBLE, XLOG_HEAP_OPMASK, XLogRecGetInfo, and XLR_INFO_MASK.

◆ heap_abort_speculative()

void heap_abort_speculative ( Relation  relation,
ItemPointer  tid 
)

Definition at line 5828 of file heapam.c.

5829 {
5831  ItemId lp;
5832  HeapTupleData tp;
5833  Page page;
5834  BlockNumber block;
5835  Buffer buffer;
5836  TransactionId prune_xid;
5837 
5838  Assert(ItemPointerIsValid(tid));
5839 
5840  block = ItemPointerGetBlockNumber(tid);
5841  buffer = ReadBuffer(relation, block);
5842  page = BufferGetPage(buffer);
5843 
5845 
5846  /*
5847  * Page can't be all visible, we just inserted into it, and are still
5848  * running.
5849  */
5850  Assert(!PageIsAllVisible(page));
5851 
5852  lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
5853  Assert(ItemIdIsNormal(lp));
5854 
5855  tp.t_tableOid = RelationGetRelid(relation);
5856  tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
5857  tp.t_len = ItemIdGetLength(lp);
5858  tp.t_self = *tid;
5859 
5860  /*
5861  * Sanity check that the tuple really is a speculatively inserted tuple,
5862  * inserted by us.
5863  */
5864  if (tp.t_data->t_choice.t_heap.t_xmin != xid)
5865  elog(ERROR, "attempted to kill a tuple inserted by another transaction");
5866  if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
5867  elog(ERROR, "attempted to kill a non-speculative tuple");
5869 
5870  /*
5871  * No need to check for serializable conflicts here. There is never a
5872  * need for a combo CID, either. No need to extract replica identity, or
5873  * do anything special with infomask bits.
5874  */
5875 
5877 
5878  /*
5879  * The tuple will become DEAD immediately. Flag that this page is a
5880  * candidate for pruning by setting xmin to TransactionXmin. While not
5881  * immediately prunable, it is the oldest xid we can cheaply determine
5882  * that's safe against wraparound / being older than the table's
5883  * relfrozenxid. To defend against the unlikely case of a new relation
5884  * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
5885  * if so (vacuum can't subsequently move relfrozenxid to beyond
5886  * TransactionXmin, so there's no race here).
5887  */
5889  if (TransactionIdPrecedes(TransactionXmin, relation->rd_rel->relfrozenxid))
5890  prune_xid = relation->rd_rel->relfrozenxid;
5891  else
5892  prune_xid = TransactionXmin;
5893  PageSetPrunable(page, prune_xid);
5894 
5895  /* store transaction information of xact deleting the tuple */
5898 
5899  /*
5900  * Set the tuple header xmin to InvalidTransactionId. This makes the
5901  * tuple immediately invisible everyone. (In particular, to any
5902  * transactions waiting on the speculative token, woken up later.)
5903  */
5905 
5906  /* Clear the speculative insertion token too */
5907  tp.t_data->t_ctid = tp.t_self;
5908 
5909  MarkBufferDirty(buffer);
5910 
5911  /*
5912  * XLOG stuff
5913  *
5914  * The WAL records generated here match heap_delete(). The same recovery
5915  * routines are used.
5916  */
5917  if (RelationNeedsWAL(relation))
5918  {
5919  xl_heap_delete xlrec;
5920  XLogRecPtr recptr;
5921 
5922  xlrec.flags = XLH_DELETE_IS_SUPER;
5924  tp.t_data->t_infomask2);
5926  xlrec.xmax = xid;
5927 
5928  XLogBeginInsert();
5929  XLogRegisterData((char *) &xlrec, SizeOfHeapDelete);
5930  XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
5931 
5932  /* No replica identity & replication origin logged */
5933 
5934  recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
5935 
5936  PageSetLSN(page, recptr);
5937  }
5938 
5939  END_CRIT_SECTION();
5940 
5941  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
5942 
5943  if (HeapTupleHasExternal(&tp))
5944  {
5945  Assert(!IsToastRelation(relation));
5946  heap_toast_delete(relation, &tp, true);
5947  }
5948 
5949  /*
5950  * Never need to mark tuple for invalidation, since catalogs don't support
5951  * speculative insertion
5952  */
5953 
5954  /* Now we can release the buffer */
5955  ReleaseBuffer(buffer);
5956 
5957  /* count deletion, as we counted the insertion too */
5958  pgstat_count_heap_delete(relation);
5959 }
int Buffer
Definition: buf.h:23
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:1565
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3996
Buffer ReadBuffer(Relation reln, BlockNumber blockNum)
Definition: bufmgr.c:694
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:96
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:98
#define BufferGetPage(buffer)
Definition: bufmgr.h:169
Pointer Page
Definition: bufpage.h:78
#define PageIsAllVisible(page)
Definition: bufpage.h:384
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:234
#define PageGetItem(page, itemId)
Definition: bufpage.h:339
#define PageSetLSN(page, lsn)
Definition: bufpage.h:367
#define PageSetPrunable(page, xid)
Definition: bufpage.h:391
bool IsToastRelation(Relation relation)
Definition: catalog.c:146
static uint8 compute_infobits(uint16 infomask, uint16 infomask2)
Definition: heapam.c:2655
#define XLOG_HEAP_DELETE
Definition: heapam_xlog.h:33
#define SizeOfHeapDelete
Definition: heapam_xlog.h:115
#define XLH_DELETE_IS_SUPER
Definition: heapam_xlog.h:99
void heap_toast_delete(Relation rel, HeapTuple oldtup, bool is_speculative)
Definition: heaptoast.c:43
HeapTupleHeaderData * HeapTupleHeader
Definition: htup.h:23
#define HeapTupleHeaderIsHeapOnly(tup)
Definition: htup_details.h:495
#define HeapTupleHeaderSetXmin(tup, xid)
Definition: htup_details.h:314
#define HEAP_XMAX_BITS
Definition: htup_details.h:266
#define HEAP_MOVED
Definition: htup_details.h:212
#define HeapTupleHeaderIsSpeculative(tup)
Definition: htup_details.h:424
#define ItemIdGetLength(itemId)
Definition: itemid.h:59
#define ItemIdIsNormal(itemId)
Definition: itemid.h:99
#define ItemPointerIsValid(pointer)
Definition: itemptr.h:82
#define ItemPointerGetOffsetNumber(pointer)
Definition: itemptr.h:117
#define START_CRIT_SECTION()
Definition: miscadmin.h:147
#define END_CRIT_SECTION()
Definition: miscadmin.h:149
void pgstat_count_heap_delete(Relation rel)
Definition: pgstat.c:2404
#define RelationGetRelid(relation)
Definition: rel.h:478
#define RelationNeedsWAL(relation)
Definition: rel.h:602
TransactionId TransactionXmin
Definition: snapmgr.c:112
ItemPointerData t_self
Definition: htup.h:65
uint32 t_len
Definition: htup.h:64
HeapTupleHeader t_data
Definition: htup.h:68
Oid t_tableOid
Definition: htup.h:66
TransactionId t_xmin
Definition: htup_details.h:123
union HeapTupleHeaderData::@42 t_choice
ItemPointerData t_ctid
Definition: htup_details.h:160
HeapTupleFields t_heap
Definition: htup_details.h:156
Form_pg_class rd_rel
Definition: rel.h:109
TransactionId xmax
Definition: heapam_xlog.h:109
OffsetNumber offnum
Definition: heapam_xlog.h:110
uint8 infobits_set
Definition: heapam_xlog.h:111
TransactionId GetCurrentTransactionId(void)
Definition: xact.c:440
uint64 XLogRecPtr
Definition: xlogdefs.h:21
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:429
void XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
Definition: xloginsert.c:229
void XLogBeginInsert(void)
Definition: xloginsert.c:136
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:337
#define REGBUF_STANDARD
Definition: xloginsert.h:34

References Assert(), BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetPage, compute_infobits(), elog, END_CRIT_SECTION, ERROR, xl_heap_delete::flags, GetCurrentTransactionId(), HEAP_KEYS_UPDATED, HEAP_MOVED, heap_toast_delete(), HEAP_XMAX_BITS, HeapTupleHasExternal, HeapTupleHeaderIsHeapOnly, HeapTupleHeaderIsSpeculative, HeapTupleHeaderSetXmin, xl_heap_delete::infobits_set, InvalidTransactionId, IsToastRelation(), ItemIdGetLength, ItemIdIsNormal, ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, ItemPointerIsValid, LockBuffer(), MarkBufferDirty(), xl_heap_delete::offnum, PageGetItem, PageGetItemId, PageIsAllVisible, PageSetLSN, PageSetPrunable, pgstat_count_heap_delete(), RelationData::rd_rel, ReadBuffer(), REGBUF_STANDARD, RelationGetRelid, RelationNeedsWAL, ReleaseBuffer(), SizeOfHeapDelete, START_CRIT_SECTION, HeapTupleHeaderData::t_choice, HeapTupleHeaderData::t_ctid, HeapTupleData::t_data, HeapTupleHeaderData::t_heap, HeapTupleHeaderData::t_infomask, HeapTupleHeaderData::t_infomask2, HeapTupleData::t_len, HeapTupleData::t_self, HeapTupleData::t_tableOid, HeapTupleFields::t_xmin, TransactionIdIsValid, TransactionIdPrecedes(), TransactionXmin, XLH_DELETE_IS_SUPER, XLOG_HEAP_DELETE, XLogBeginInsert(), XLogInsert(), XLogRegisterBuffer(), XLogRegisterData(), and xl_heap_delete::xmax.

Referenced by heapam_tuple_complete_speculative(), and toast_delete_datum().

◆ heap_acquire_tuplock()

static bool heap_acquire_tuplock ( Relation  relation,
ItemPointer  tid,
LockTupleMode  mode,
LockWaitPolicy  wait_policy,
bool have_tuple_lock 
)
static

Definition at line 4926 of file heapam.c.

4928 {
4929  if (*have_tuple_lock)
4930  return true;
4931 
4932  switch (wait_policy)
4933  {
4934  case LockWaitBlock:
4935  LockTupleTuplock(relation, tid, mode);
4936  break;
4937 
4938  case LockWaitSkip:
4939  if (!ConditionalLockTupleTuplock(relation, tid, mode))
4940  return false;
4941  break;
4942 
4943  case LockWaitError:
4944  if (!ConditionalLockTupleTuplock(relation, tid, mode))
4945  ereport(ERROR,
4946  (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4947  errmsg("could not obtain lock on row in relation \"%s\"",
4948  RelationGetRelationName(relation))));
4949  break;
4950  }
4951  *have_tuple_lock = true;
4952 
4953  return true;
4954 }
int errmsg(const char *fmt,...)
Definition: elog.c:904
#define ConditionalLockTupleTuplock(rel, tup, mode)
Definition: heapam.c:167
#define LockTupleTuplock(rel, tup, mode)
Definition: heapam.c:163
@ LockWaitSkip
Definition: lockoptions.h:41
@ LockWaitBlock
Definition: lockoptions.h:39
@ LockWaitError
Definition: lockoptions.h:43
#define RelationGetRelationName(relation)
Definition: rel.h:512

References ConditionalLockTupleTuplock, ereport, errcode(), errmsg(), ERROR, LockTupleTuplock, LockWaitBlock, LockWaitError, LockWaitSkip, mode, and RelationGetRelationName.

Referenced by heap_delete(), heap_lock_tuple(), and heap_update().

◆ heap_beginscan()

TableScanDesc heap_beginscan ( Relation  relation,
Snapshot  snapshot,
int  nkeys,
ScanKey  key,
ParallelTableScanDesc  parallel_scan,
uint32  flags 
)

Definition at line 1185 of file heapam.c.

1189 {
1190  HeapScanDesc scan;
1191 
1192  /*
1193  * increment relation ref count while scanning relation
1194  *
1195  * This is just to make really sure the relcache entry won't go away while
1196  * the scan has a pointer to it. Caller should be holding the rel open
1197  * anyway, so this is redundant in all normal scenarios...
1198  */
1200 
1201  /*
1202  * allocate and initialize scan descriptor
1203  */
1204  scan = (HeapScanDesc) palloc(sizeof(HeapScanDescData));
1205 
1206  scan->rs_base.rs_rd = relation;
1207  scan->rs_base.rs_snapshot = snapshot;
1208  scan->rs_base.rs_nkeys = nkeys;
1209  scan->rs_base.rs_flags = flags;
1210  scan->rs_base.rs_parallel = parallel_scan;
1211  scan->rs_strategy = NULL; /* set in initscan */
1212 
1213  /*
1214  * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1215  */
1216  if (!(snapshot && IsMVCCSnapshot(snapshot)))
1218 
1219  /*
1220  * For seqscan and sample scans in a serializable transaction, acquire a
1221  * predicate lock on the entire relation. This is required not only to
1222  * lock all the matching tuples, but also to conflict with new insertions
1223  * into the table. In an indexscan, we take page locks on the index pages
1224  * covering the range specified in the scan qual, but in a heap scan there
1225  * is nothing more fine-grained to lock. A bitmap scan is a different
1226  * story, there we have already scanned the index and locked the index
1227  * pages covering the predicate. But in that case we still have to lock
1228  * any matching heap tuples. For sample scan we could optimize the locking
1229  * to be at least page-level granularity, but we'd need to add per-tuple
1230  * locking for that.
1231  */
1233  {
1234  /*
1235  * Ensure a missing snapshot is noticed reliably, even if the
1236  * isolation mode means predicate locking isn't performed (and
1237  * therefore the snapshot isn't used here).
1238  */
1239  Assert(snapshot);
1240  PredicateLockRelation(relation, snapshot);
1241  }
1242 
1243  /* we only need to set this up once */
1244  scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1245 
1246  /*
1247  * Allocate memory to keep track of page allocation for parallel workers
1248  * when doing a parallel scan.
1249  */
1250  if (parallel_scan != NULL)
1252  else
1253  scan->rs_parallelworkerdata = NULL;
1254 
1255  /*
1256  * we do this here instead of in initscan() because heap_rescan also calls
1257  * initscan() and we don't want to allocate memory again
1258  */
1259  if (nkeys > 0)
1260  scan->rs_base.rs_key = (ScanKey) palloc(sizeof(ScanKeyData) * nkeys);
1261  else
1262  scan->rs_base.rs_key = NULL;
1263 
1264  initscan(scan, key, false);
1265 
1266  return (TableScanDesc) scan;
1267 }
static void initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
Definition: heapam.c:227
struct HeapScanDescData * HeapScanDesc
Definition: heapam.h:79
void PredicateLockRelation(Relation relation, Snapshot snapshot)
Definition: predicate.c:2571
void RelationIncrementReferenceCount(Relation rel)
Definition: relcache.c:2105
ScanKeyData * ScanKey
Definition: skey.h:75
#define IsMVCCSnapshot(snapshot)
Definition: snapmgr.h:96
BufferAccessStrategy rs_strategy
Definition: heapam.h:64
ParallelBlockTableScanWorkerData * rs_parallelworkerdata
Definition: heapam.h:72
HeapTupleData rs_ctup
Definition: heapam.h:66
TableScanDescData rs_base
Definition: heapam.h:49
Relation rs_rd
Definition: relscan.h:34
uint32 rs_flags
Definition: relscan.h:47
struct ScanKeyData * rs_key
Definition: relscan.h:37
struct SnapshotData * rs_snapshot
Definition: relscan.h:35
struct ParallelTableScanDescData * rs_parallel
Definition: relscan.h:49
@ SO_ALLOW_PAGEMODE
Definition: tableam.h:61
@ SO_TYPE_SAMPLESCAN
Definition: tableam.h:50
@ SO_TYPE_SEQSCAN
Definition: tableam.h:48

References Assert(), if(), initscan(), IsMVCCSnapshot, sort-test::key, palloc(), PredicateLockRelation(), RelationGetRelid, RelationIncrementReferenceCount(), HeapScanDescData::rs_base, HeapScanDescData::rs_ctup, TableScanDescData::rs_flags, TableScanDescData::rs_key, TableScanDescData::rs_nkeys, TableScanDescData::rs_parallel, HeapScanDescData::rs_parallelworkerdata, TableScanDescData::rs_rd, TableScanDescData::rs_snapshot, HeapScanDescData::rs_strategy, SO_ALLOW_PAGEMODE, SO_TYPE_SAMPLESCAN, SO_TYPE_SEQSCAN, and HeapTupleData::t_tableOid.

◆ heap_delete()

TM_Result heap_delete ( Relation  relation,
ItemPointer  tid,
CommandId  cid,
Snapshot  crosscheck,
bool  wait,
TM_FailureData tmfd,
bool  changingPart 
)

Definition at line 2700 of file heapam.c.

2703 {
2704  TM_Result result;
2706  ItemId lp;
2707  HeapTupleData tp;
2708  Page page;
2709  BlockNumber block;
2710  Buffer buffer;
2711  Buffer vmbuffer = InvalidBuffer;
2712  TransactionId new_xmax;
2713  uint16 new_infomask,
2714  new_infomask2;
2715  bool have_tuple_lock = false;
2716  bool iscombo;
2717  bool all_visible_cleared = false;
2718  HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2719  bool old_key_copied = false;
2720 
2721  Assert(ItemPointerIsValid(tid));
2722 
2723  /*
2724  * Forbid this during a parallel operation, lest it allocate a combo CID.
2725  * Other workers might need that combo CID for visibility checks, and we
2726  * have no provision for broadcasting it to them.
2727  */
2728  if (IsInParallelMode())
2729  ereport(ERROR,
2730  (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2731  errmsg("cannot delete tuples during a parallel operation")));
2732 
2733  block = ItemPointerGetBlockNumber(tid);
2734  buffer = ReadBuffer(relation, block);
2735  page = BufferGetPage(buffer);
2736 
2737  /*
2738  * Before locking the buffer, pin the visibility map page if it appears to
2739  * be necessary. Since we haven't got the lock yet, someone else might be
2740  * in the middle of changing this, so we'll need to recheck after we have
2741  * the lock.
2742  */
2743  if (PageIsAllVisible(page))
2744  visibilitymap_pin(relation, block, &vmbuffer);
2745 
2747 
2748  /*
2749  * If we didn't pin the visibility map page and the page has become all
2750  * visible while we were busy locking the buffer, we'll have to unlock and
2751  * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2752  * unfortunate, but hopefully shouldn't happen often.
2753  */
2754  if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2755  {
2756  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2757  visibilitymap_pin(relation, block, &vmbuffer);
2759  }
2760 
2761  lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2762  Assert(ItemIdIsNormal(lp));
2763 
2764  tp.t_tableOid = RelationGetRelid(relation);
2765  tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2766  tp.t_len = ItemIdGetLength(lp);
2767  tp.t_self = *tid;
2768 
2769 l1:
2770  result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2771 
2772  if (result == TM_Invisible)
2773  {
2774  UnlockReleaseBuffer(buffer);
2775  ereport(ERROR,
2776  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2777  errmsg("attempted to delete invisible tuple")));
2778  }
2779  else if (result == TM_BeingModified && wait)
2780  {
2781  TransactionId xwait;
2782  uint16 infomask;
2783 
2784  /* must copy state data before unlocking buffer */
2785  xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2786  infomask = tp.t_data->t_infomask;
2787 
2788  /*
2789  * Sleep until concurrent transaction ends -- except when there's a
2790  * single locker and it's our own transaction. Note we don't care
2791  * which lock mode the locker has, because we need the strongest one.
2792  *
2793  * Before sleeping, we need to acquire tuple lock to establish our
2794  * priority for the tuple (see heap_lock_tuple). LockTuple will
2795  * release us when we are next-in-line for the tuple.
2796  *
2797  * If we are forced to "start over" below, we keep the tuple lock;
2798  * this arranges that we stay at the head of the line while rechecking
2799  * tuple state.
2800  */
2801  if (infomask & HEAP_XMAX_IS_MULTI)
2802  {
2803  bool current_is_member = false;
2804 
2805  if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2806  LockTupleExclusive, &current_is_member))
2807  {
2808  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2809 
2810  /*
2811  * Acquire the lock, if necessary (but skip it when we're
2812  * requesting a lock and already have one; avoids deadlock).
2813  */
2814  if (!current_is_member)
2816  LockWaitBlock, &have_tuple_lock);
2817 
2818  /* wait for multixact */
2820  relation, &(tp.t_self), XLTW_Delete,
2821  NULL);
2823 
2824  /*
2825  * If xwait had just locked the tuple then some other xact
2826  * could update this tuple before we get to this point. Check
2827  * for xmax change, and start over if so.
2828  */
2829  if (xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2831  xwait))
2832  goto l1;
2833  }
2834 
2835  /*
2836  * You might think the multixact is necessarily done here, but not
2837  * so: it could have surviving members, namely our own xact or
2838  * other subxacts of this backend. It is legal for us to delete
2839  * the tuple in either case, however (the latter case is
2840  * essentially a situation of upgrading our former shared lock to
2841  * exclusive). We don't bother changing the on-disk hint bits
2842  * since we are about to overwrite the xmax altogether.
2843  */
2844  }
2845  else if (!TransactionIdIsCurrentTransactionId(xwait))
2846  {
2847  /*
2848  * Wait for regular transaction to end; but first, acquire tuple
2849  * lock.
2850  */
2851  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2853  LockWaitBlock, &have_tuple_lock);
2854  XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
2856 
2857  /*
2858  * xwait is done, but if xwait had just locked the tuple then some
2859  * other xact could update this tuple before we get to this point.
2860  * Check for xmax change, and start over if so.
2861  */
2862  if (xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2864  xwait))
2865  goto l1;
2866 
2867  /* Otherwise check if it committed or aborted */
2868  UpdateXmaxHintBits(tp.t_data, buffer, xwait);
2869  }
2870 
2871  /*
2872  * We may overwrite if previous xmax aborted, or if it committed but
2873  * only locked the tuple without updating it.
2874  */
2875  if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2878  result = TM_Ok;
2879  else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2880  result = TM_Updated;
2881  else
2882  result = TM_Deleted;
2883  }
2884 
2885  if (crosscheck != InvalidSnapshot && result == TM_Ok)
2886  {
2887  /* Perform additional check for transaction-snapshot mode RI updates */
2888  if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
2889  result = TM_Updated;
2890  }
2891 
2892  if (result != TM_Ok)
2893  {
2894  Assert(result == TM_SelfModified ||
2895  result == TM_Updated ||
2896  result == TM_Deleted ||
2897  result == TM_BeingModified);
2899  Assert(result != TM_Updated ||
2900  !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
2901  tmfd->ctid = tp.t_data->t_ctid;
2903  if (result == TM_SelfModified)
2904  tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
2905  else
2906  tmfd->cmax = InvalidCommandId;
2907  UnlockReleaseBuffer(buffer);
2908  if (have_tuple_lock)
2909  UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
2910  if (vmbuffer != InvalidBuffer)
2911  ReleaseBuffer(vmbuffer);
2912  return result;
2913  }
2914 
2915  /*
2916  * We're about to do the actual delete -- check for conflict first, to
2917  * avoid possibly having to roll back work we've just done.
2918  *
2919  * This is safe without a recheck as long as there is no possibility of
2920  * another process scanning the page between this check and the delete
2921  * being visible to the scan (i.e., an exclusive buffer content lock is
2922  * continuously held from this point until the tuple delete is visible).
2923  */
2924  CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
2925 
2926  /* replace cid with a combo CID if necessary */
2927  HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
2928 
2929  /*
2930  * Compute replica identity tuple before entering the critical section so
2931  * we don't PANIC upon a memory allocation failure.
2932  */
2933  old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
2934 
2935  /*
2936  * If this is the first possibly-multixact-able operation in the current
2937  * transaction, set my per-backend OldestMemberMXactId setting. We can be
2938  * certain that the transaction will never become a member of any older
2939  * MultiXactIds than that. (We have to do this even if we end up just
2940  * using our own TransactionId below, since some other backend could
2941  * incorporate our XID into a MultiXact immediately afterwards.)
2942  */
2944 
2947  xid, LockTupleExclusive, true,
2948  &new_xmax, &new_infomask, &new_infomask2);
2949 
2951 
2952  /*
2953  * If this transaction commits, the tuple will become DEAD sooner or
2954  * later. Set flag that this page is a candidate for pruning once our xid
2955  * falls below the OldestXmin horizon. If the transaction finally aborts,
2956  * the subsequent page pruning will be a no-op and the hint will be
2957  * cleared.
2958  */
2959  PageSetPrunable(page, xid);
2960 
2961  if (PageIsAllVisible(page))
2962  {
2963  all_visible_cleared = true;
2964  PageClearAllVisible(page);
2965  visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
2966  vmbuffer, VISIBILITYMAP_VALID_BITS);
2967  }
2968 
2969  /* store transaction information of xact deleting the tuple */
2972  tp.t_data->t_infomask |= new_infomask;
2973  tp.t_data->t_infomask2 |= new_infomask2;
2975  HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
2976  HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
2977  /* Make sure there is no forward chain link in t_ctid */
2978  tp.t_data->t_ctid = tp.t_self;
2979 
2980  /* Signal that this is actually a move into another partition */
2981  if (changingPart)
2983 
2984  MarkBufferDirty(buffer);
2985 
2986  /*
2987  * XLOG stuff
2988  *
2989  * NB: heap_abort_speculative() uses the same xlog record and replay
2990  * routines.
2991  */
2992  if (RelationNeedsWAL(relation))
2993  {
2994  xl_heap_delete xlrec;
2995  xl_heap_header xlhdr;
2996  XLogRecPtr recptr;
2997 
2998  /*
2999  * For logical decode we need combo CIDs to properly decode the
3000  * catalog
3001  */
3003  log_heap_new_cid(relation, &tp);
3004 
3005  xlrec.flags = 0;
3006  if (all_visible_cleared)
3008  if (changingPart)
3011  tp.t_data->t_infomask2);
3013  xlrec.xmax = new_xmax;
3014 
3015  if (old_key_tuple != NULL)
3016  {
3017  if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3019  else
3021  }
3022 
3023  XLogBeginInsert();
3024  XLogRegisterData((char *) &xlrec, SizeOfHeapDelete);
3025 
3026  XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3027 
3028  /*
3029  * Log replica identity of the deleted tuple if there is one
3030  */
3031  if (old_key_tuple != NULL)
3032  {
3033  xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3034  xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3035  xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3036 
3037  XLogRegisterData((char *) &xlhdr, SizeOfHeapHeader);
3038  XLogRegisterData((char *) old_key_tuple->t_data
3040  old_key_tuple->t_len
3042  }
3043 
3044  /* filtering by origin on a row level is much more efficient */
3046 
3047  recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3048 
3049  PageSetLSN(page, recptr);
3050  }
3051 
3052  END_CRIT_SECTION();
3053 
3054  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3055 
3056  if (vmbuffer != InvalidBuffer)
3057  ReleaseBuffer(vmbuffer);
3058 
3059  /*
3060  * If the tuple has toasted out-of-line attributes, we need to delete
3061  * those items too. We have to do this before releasing the buffer
3062  * because we need to look at the contents of the tuple, but it's OK to
3063  * release the content lock on the buffer first.
3064  */
3065  if (relation->rd_rel->relkind != RELKIND_RELATION &&
3066  relation->rd_rel->relkind != RELKIND_MATVIEW)
3067  {
3068  /* toast table entries should never be recursively toasted */
3070  }
3071  else if (HeapTupleHasExternal(&tp))
3072  heap_toast_delete(relation, &tp, false);
3073 
3074  /*
3075  * Mark tuple for invalidation from system caches at next command
3076  * boundary. We have to do this before releasing the buffer because we
3077  * need to look at the contents of the tuple.
3078  */
3079  CacheInvalidateHeapTuple(relation, &tp, NULL);
3080 
3081  /* Now we can release the buffer */
3082  ReleaseBuffer(buffer);
3083 
3084  /*
3085  * Release the lmgr tuple lock, if we had it.
3086  */
3087  if (have_tuple_lock)
3088  UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3089 
3090  pgstat_count_heap_delete(relation);
3091 
3092  if (old_key_tuple != NULL && old_key_copied)
3093  heap_freetuple(old_key_tuple);
3094 
3095  return TM_Ok;
3096 }
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2748
void UnlockReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3780
#define PageClearAllVisible(page)
Definition: bufpage.h:388
#define InvalidCommandId
Definition: c.h:604
void HeapTupleHeaderAdjustCmax(HeapTupleHeader tup, CommandId *cmax, bool *iscombo)
Definition: combocid.c:153
CommandId HeapTupleHeaderGetCmax(HeapTupleHeader tup)
Definition: combocid.c:118
static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask, LockTupleMode lockmode, bool *current_is_member)
Definition: heapam.c:6822
static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup)
Definition: heapam.c:8287
static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask, uint16 old_infomask2, TransactionId add_to_xmax, LockTupleMode mode, bool is_update, TransactionId *result_xmax, uint16 *result_infomask, uint16 *result_infomask2)
Definition: heapam.c:4975
static HeapTuple ExtractReplicaIdentity(Relation rel, HeapTuple tup, bool key_changed, bool *copy)
Definition: heapam.c:8369
static bool heap_acquire_tuplock(Relation relation, ItemPointer tid, LockTupleMode mode, LockWaitPolicy wait_policy, bool *have_tuple_lock)
Definition: heapam.c:4926
static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask, Relation rel, ItemPointer ctid, XLTW_Oper oper, int *remaining)
Definition: heapam.c:6999
static bool xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
Definition: heapam.c:2677
#define UnlockTupleTuplock(rel, tup, mode)
Definition: heapam.c:165
static void UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
Definition: heapam.c:1985
bool HeapTupleSatisfiesVisibility(HeapTuple tup, Snapshot snapshot, Buffer buffer)
bool HeapTupleHeaderIsOnlyLocked(HeapTupleHeader tuple)
TM_Result HeapTupleSatisfiesUpdate(HeapTuple htup, CommandId curcid, Buffer buffer)
#define XLH_DELETE_CONTAINS_OLD_KEY
Definition: heapam_xlog.h:98
#define XLH_DELETE_ALL_VISIBLE_CLEARED
Definition: heapam_xlog.h:96
#define SizeOfHeapHeader
Definition: heapam_xlog.h:151
#define XLH_DELETE_IS_PARTITION_MOVE
Definition: heapam_xlog.h:100
#define XLH_DELETE_CONTAINS_OLD_TUPLE
Definition: heapam_xlog.h:97
#define SizeofHeapTupleHeader
Definition: htup_details.h:184
#define HeapTupleHeaderSetXmax(tup, xid)
Definition: htup_details.h:375
#define HeapTupleHeaderClearHotUpdated(tup)
Definition: htup_details.h:490
#define HeapTupleHeaderSetMovedPartitions(tup)
Definition: htup_details.h:443
#define HeapTupleHeaderGetRawXmax(tup)
Definition: htup_details.h:370
#define HeapTupleHeaderGetUpdateXid(tup)
Definition: htup_details.h:360
#define HeapTupleHeaderSetCmax(tup, cid, iscombo)
Definition: htup_details.h:400
void CacheInvalidateHeapTuple(Relation relation, HeapTuple tuple, HeapTuple newtuple)
Definition: inval.c:1206
bool ItemPointerEquals(ItemPointer pointer1, ItemPointer pointer2)
Definition: itemptr.c:29
@ XLTW_Delete
Definition: lmgr.h:28
void MultiXactIdSetOldestMember(void)
Definition: multixact.c:625
void CheckForSerializableConflictIn(Relation relation, ItemPointer tid, BlockNumber blkno)
Definition: predicate.c:4448
#define RelationIsAccessibleInLogicalDecoding(relation)
Definition: rel.h:658
#define InvalidSnapshot
Definition: snapshot.h:123
TransactionId xmax
Definition: tableam.h:127
CommandId cmax
Definition: tableam.h:128
ItemPointerData ctid
Definition: tableam.h:126
uint16 t_infomask
Definition: heapam_xlog.h:147
uint16 t_infomask2
Definition: heapam_xlog.h:146
TM_Result
Definition: tableam.h:72
@ TM_Ok
Definition: tableam.h:77
@ TM_BeingModified
Definition: tableam.h:99
@ TM_Deleted
Definition: tableam.h:92
@ TM_Updated
Definition: tableam.h:89
@ TM_SelfModified
Definition: tableam.h:83
@ TM_Invisible
Definition: tableam.h:80
#define TransactionIdEquals(id1, id2)
Definition: transam.h:43
bool visibilitymap_clear(Relation rel, BlockNumber heapBlk, Buffer buf, uint8 flags)
void visibilitymap_pin(Relation rel, BlockNumber heapBlk, Buffer *buf)
#define VISIBILITYMAP_VALID_BITS
bool IsInParallelMode(void)
Definition: xact.c:1064
#define XLOG_INCLUDE_ORIGIN
Definition: xlog.h:209
void XLogSetRecordFlags(uint8 flags)
Definition: xloginsert.c:411

References Assert(), BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetBlockNumber(), BufferGetPage, CacheInvalidateHeapTuple(), CheckForSerializableConflictIn(), TM_FailureData::cmax, compute_infobits(), compute_new_xmax_infomask(), TM_FailureData::ctid, DoesMultiXactIdConflict(), END_CRIT_SECTION, ereport, errcode(), errmsg(), ERROR, ExtractReplicaIdentity(), xl_heap_delete::flags, GetCurrentTransactionId(), heap_acquire_tuplock(), heap_freetuple(), HEAP_KEYS_UPDATED, HEAP_MOVED, heap_toast_delete(), HEAP_XMAX_BITS, HEAP_XMAX_INVALID, HEAP_XMAX_IS_LOCKED_ONLY, HEAP_XMAX_IS_MULTI, HeapTupleHasExternal, HeapTupleHeaderAdjustCmax(), HeapTupleHeaderClearHotUpdated, HeapTupleHeaderGetCmax(), HeapTupleHeaderGetRawXmax, HeapTupleHeaderGetUpdateXid, HeapTupleHeaderIsOnlyLocked(), HeapTupleHeaderSetCmax, HeapTupleHeaderSetMovedPartitions, HeapTupleHeaderSetXmax, HeapTupleSatisfiesUpdate(), HeapTupleSatisfiesVisibility(), xl_heap_delete::infobits_set, InvalidBuffer, InvalidCommandId, InvalidSnapshot, IsInParallelMode(), ItemIdGetLength, ItemIdIsNormal, ItemPointerEquals(), ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, ItemPointerIsValid, LockBuffer(), LockTupleExclusive, LockWaitBlock, log_heap_new_cid(), MarkBufferDirty(), MultiXactIdSetOldestMember(), MultiXactIdWait(), MultiXactStatusUpdate, xl_heap_delete::offnum, PageClearAllVisible, PageGetItem, PageGetItemId, PageIsAllVisible, PageSetLSN, PageSetPrunable, pgstat_count_heap_delete(), RelationData::rd_rel, ReadBuffer(), REGBUF_STANDARD, RelationGetRelid, RelationIsAccessibleInLogicalDecoding, RelationNeedsWAL, ReleaseBuffer(), SizeOfHeapDelete, SizeOfHeapHeader, SizeofHeapTupleHeader, START_CRIT_SECTION, HeapTupleHeaderData::t_ctid, HeapTupleData::t_data, xl_heap_header::t_hoff, HeapTupleHeaderData::t_hoff, xl_heap_header::t_infomask, HeapTupleHeaderData::t_infomask, xl_heap_header::t_infomask2, HeapTupleHeaderData::t_infomask2, HeapTupleData::t_len, HeapTupleData::t_self, HeapTupleData::t_tableOid, TM_BeingModified, TM_Deleted, TM_Invisible, TM_Ok, TM_SelfModified, TM_Updated, TransactionIdEquals, TransactionIdIsCurrentTransactionId(), UnlockReleaseBuffer(), UnlockTupleTuplock, UpdateXmaxHintBits(), visibilitymap_clear(), visibilitymap_pin(), VISIBILITYMAP_VALID_BITS, XactLockTableWait(), XLH_DELETE_ALL_VISIBLE_CLEARED, XLH_DELETE_CONTAINS_OLD_KEY, XLH_DELETE_CONTAINS_OLD_TUPLE, XLH_DELETE_IS_PARTITION_MOVE, XLOG_HEAP_DELETE, XLOG_INCLUDE_ORIGIN, XLogBeginInsert(), XLogInsert(), XLogRegisterBuffer(), XLogRegisterData(), XLogSetRecordFlags(), XLTW_Delete, xl_heap_delete::xmax, TM_FailureData::xmax, and xmax_infomask_changed().

Referenced by heapam_tuple_delete(), and simple_heap_delete().

◆ heap_endscan()

void heap_endscan ( TableScanDesc  sscan)

Definition at line 1307 of file heapam.c.

1308 {
1309  HeapScanDesc scan = (HeapScanDesc) sscan;
1310 
1311  /* Note: no locking manipulations needed */
1312 
1313  /*
1314  * unpin scan buffers
1315  */
1316  if (BufferIsValid(scan->rs_cbuf))
1317  ReleaseBuffer(scan->rs_cbuf);
1318 
1319  /*
1320  * decrement relation reference count and free scan descriptor storage
1321  */
1323 
1324  if (scan->rs_base.rs_key)
1325  pfree(scan->rs_base.rs_key);
1326 
1327  if (scan->rs_strategy != NULL)
1329 
1330  if (scan->rs_parallelworkerdata != NULL)
1332 
1333  if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1335 
1336  pfree(scan);
1337 }
#define BufferIsValid(bufnum)
Definition: bufmgr.h:123
void RelationDecrementReferenceCount(Relation rel)
Definition: relcache.c:2118
void UnregisterSnapshot(Snapshot snapshot)
Definition: snapmgr.c:867
Buffer rs_cbuf
Definition: heapam.h:60
@ SO_TEMP_SNAPSHOT
Definition: tableam.h:64

References BufferIsValid, FreeAccessStrategy(), pfree(), RelationDecrementReferenceCount(), ReleaseBuffer(), HeapScanDescData::rs_base, HeapScanDescData::rs_cbuf, TableScanDescData::rs_flags, TableScanDescData::rs_key, HeapScanDescData::rs_parallelworkerdata, TableScanDescData::rs_rd, TableScanDescData::rs_snapshot, HeapScanDescData::rs_strategy, SO_TEMP_SNAPSHOT, and UnregisterSnapshot().

◆ heap_execute_freeze_tuple()

void heap_execute_freeze_tuple ( HeapTupleHeader  tuple,
xl_heap_freeze_tuple frz 
)

◆ heap_fetch()

bool heap_fetch ( Relation  relation,
Snapshot  snapshot,
HeapTuple  tuple,
Buffer userbuf 
)

Definition at line 1595 of file heapam.c.

1599 {
1600  ItemPointer tid = &(tuple->t_self);
1601  ItemId lp;
1602  Buffer buffer;
1603  Page page;
1604  OffsetNumber offnum;
1605  bool valid;
1606 
1607  /*
1608  * Fetch and pin the appropriate page of the relation.
1609  */
1610  buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1611 
1612  /*
1613  * Need share lock on buffer to examine tuple commit status.
1614  */
1615  LockBuffer(buffer, BUFFER_LOCK_SHARE);
1616  page = BufferGetPage(buffer);
1617  TestForOldSnapshot(snapshot, relation, page);
1618 
1619  /*
1620  * We'd better check for out-of-range offnum in case of VACUUM since the
1621  * TID was obtained.
1622  */
1623  offnum = ItemPointerGetOffsetNumber(tid);
1624  if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1625  {
1626  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1627  ReleaseBuffer(buffer);
1628  *userbuf = InvalidBuffer;
1629  tuple->t_data = NULL;
1630  return false;
1631  }
1632 
1633  /*
1634  * get the item line pointer corresponding to the requested tid
1635  */
1636  lp = PageGetItemId(page, offnum);
1637 
1638  /*
1639  * Must check for deleted tuple.
1640  */
1641  if (!ItemIdIsNormal(lp))
1642  {
1643  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1644  ReleaseBuffer(buffer);
1645  *userbuf = InvalidBuffer;
1646  tuple->t_data = NULL;
1647  return false;
1648  }
1649 
1650  /*
1651  * fill in *tuple fields
1652  */
1653  tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1654  tuple->t_len = ItemIdGetLength(lp);
1655  tuple->t_tableOid = RelationGetRelid(relation);
1656 
1657  /*
1658  * check tuple visibility, then release lock
1659  */
1660  valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1661 
1662  if (valid)
1663  PredicateLockTID(relation, &(tuple->t_self), snapshot,
1664  HeapTupleHeaderGetXmin(tuple->t_data));
1665 
1666  HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1667 
1668  LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1669 
1670  if (valid)
1671  {
1672  /*
1673  * All checks passed, so return the tuple as valid. Caller is now
1674  * responsible for releasing the buffer.
1675  */
1676  *userbuf = buffer;
1677 
1678  return true;
1679  }
1680 
1681  /* Tuple failed time qual */
1682  ReleaseBuffer(buffer);
1683  *userbuf = InvalidBuffer;
1684 
1685  return false;
1686 }
#define BUFFER_LOCK_SHARE
Definition: bufmgr.h:97
static void TestForOldSnapshot(Snapshot snapshot, Relation relation, Page page)
Definition: bufmgr.h:278
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:356
void HeapCheckForSerializableConflictOut(bool visible, Relation relation, HeapTuple tuple, Buffer buffer, Snapshot snapshot)
Definition: heapam.c:9848
#define HeapTupleHeaderGetXmin(tup)
Definition: htup_details.h:308
uint16 OffsetNumber
Definition: off.h:24
void PredicateLockTID(Relation relation, ItemPointer tid, Snapshot snapshot, TransactionId tuple_xid)
Definition: predicate.c:2616

References BUFFER_LOCK_SHARE, BUFFER_LOCK_UNLOCK, BufferGetPage, HeapCheckForSerializableConflictOut(), HeapTupleHeaderGetXmin, HeapTupleSatisfiesVisibility(), InvalidBuffer, ItemIdGetLength, ItemIdIsNormal, ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, LockBuffer(), PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, PredicateLockTID(), ReadBuffer(), RelationGetRelid, ReleaseBuffer(), HeapTupleData::t_data, HeapTupleData::t_len, HeapTupleData::t_self, HeapTupleData::t_tableOid, and TestForOldSnapshot().

Referenced by heap_lock_updated_tuple_rec(), heapam_fetch_row_version(), and heapam_tuple_lock().

◆ heap_finish_speculative()

void heap_finish_speculative ( Relation  relation,
ItemPointer  tid 
)

Definition at line 5737 of file heapam.c.

5738 {
5739  Buffer buffer;
5740  Page page;
5741  OffsetNumber offnum;
5742  ItemId lp = NULL;
5743  HeapTupleHeader htup;
5744 
5745  buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
5747  page = (Page) BufferGetPage(buffer);
5748 
5749  offnum = ItemPointerGetOffsetNumber(tid);
5750  if (PageGetMaxOffsetNumber(page) >= offnum)
5751  lp = PageGetItemId(page, offnum);
5752 
5753  if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
5754  elog(ERROR, "invalid lp");
5755 
5756  htup = (HeapTupleHeader) PageGetItem(page, lp);
5757 
5758  /* SpecTokenOffsetNumber should be distinguishable from any real offset */
5760  "invalid speculative token constant");
5761 
5762  /* NO EREPORT(ERROR) from here till changes are logged */
5764 
5766 
5767  MarkBufferDirty(buffer);
5768 
5769  /*
5770  * Replace the speculative insertion token with a real t_ctid, pointing to
5771  * itself like it does on regular tuples.
5772  */
5773  htup->t_ctid = *tid;
5774 
5775  /* XLOG stuff */
5776  if (RelationNeedsWAL(relation))
5777  {
5778  xl_heap_confirm xlrec;
5779  XLogRecPtr recptr;
5780 
5781  xlrec.offnum = ItemPointerGetOffsetNumber(tid);
5782 
5783  XLogBeginInsert();
5784 
5785  /* We want the same filtering on this as on a plain insert */
5787 
5788  XLogRegisterData((char *) &xlrec, SizeOfHeapConfirm);
5789  XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
5790 
5791  recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
5792 
5793  PageSetLSN(page, recptr);
5794  }
5795 
5796  END_CRIT_SECTION();
5797 
5798  UnlockReleaseBuffer(buffer);
5799 }
#define StaticAssertStmt(condition, errmessage)
Definition: c.h:918
#define SizeOfHeapConfirm
Definition: heapam_xlog.h:305
#define XLOG_HEAP_CONFIRM
Definition: heapam_xlog.h:37
#define SpecTokenOffsetNumber
Definition: itemptr.h:63
#define MaxOffsetNumber
Definition: off.h:28
OffsetNumber offnum
Definition: heapam_xlog.h:302

References Assert(), BUFFER_LOCK_EXCLUSIVE, BufferGetPage, elog, END_CRIT_SECTION, ERROR, HeapTupleHeaderIsSpeculative, ItemIdIsNormal, ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, LockBuffer(), MarkBufferDirty(), MaxOffsetNumber, xl_heap_confirm::offnum, PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, PageSetLSN, ReadBuffer(), REGBUF_STANDARD, RelationNeedsWAL, SizeOfHeapConfirm, SpecTokenOffsetNumber, START_CRIT_SECTION, StaticAssertStmt, HeapTupleHeaderData::t_ctid, UnlockReleaseBuffer(), XLOG_HEAP_CONFIRM, XLOG_INCLUDE_ORIGIN, XLogBeginInsert(), XLogInsert(), XLogRegisterBuffer(), XLogRegisterData(), and XLogSetRecordFlags().

Referenced by heapam_tuple_complete_speculative().

◆ heap_freeze_tuple()

bool heap_freeze_tuple ( HeapTupleHeader  tuple,
TransactionId  relfrozenxid,
TransactionId  relminmxid,
TransactionId  cutoff_xid,
TransactionId  cutoff_multi 
)

Definition at line 6642 of file heapam.c.

6645 {
6647  bool do_freeze;
6648  bool tuple_totally_frozen;
6649 
6650  do_freeze = heap_prepare_freeze_tuple(tuple,
6651  relfrozenxid, relminmxid,
6652  cutoff_xid, cutoff_multi,
6653  &frz, &tuple_totally_frozen);
6654 
6655  /*
6656  * Note that because this is not a WAL-logged operation, we don't need to
6657  * fill in the offset in the freeze record.
6658  */
6659 
6660  if (do_freeze)
6661  heap_execute_freeze_tuple(tuple, &frz);
6662  return do_freeze;
6663 }
void heap_execute_freeze_tuple(HeapTupleHeader tuple, xl_heap_freeze_tuple *frz)
Definition: heapam.c:6621
bool heap_prepare_freeze_tuple(HeapTupleHeader tuple, TransactionId relfrozenxid, TransactionId relminmxid, TransactionId cutoff_xid, TransactionId cutoff_multi, xl_heap_freeze_tuple *frz, bool *totally_frozen_p)
Definition: heapam.c:6392

References heap_execute_freeze_tuple(), and heap_prepare_freeze_tuple().

Referenced by rewrite_heap_tuple().

◆ heap_get_latest_tid()

void heap_get_latest_tid ( TableScanDesc  sscan,
ItemPointer  tid 
)

Definition at line 1862 of file heapam.c.

1864 {
1865  Relation relation = sscan->rs_rd;
1866  Snapshot snapshot = sscan->rs_snapshot;
1867  ItemPointerData ctid;
1868  TransactionId priorXmax;
1869 
1870  /*
1871  * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1872  * Assume that t_ctid links are valid however - there shouldn't be invalid
1873  * ones in the table.
1874  */
1875  Assert(ItemPointerIsValid(tid));
1876 
1877  /*
1878  * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1879  * need to examine, and *tid is the TID we will return if ctid turns out
1880  * to be bogus.
1881  *
1882  * Note that we will loop until we reach the end of the t_ctid chain.
1883  * Depending on the snapshot passed, there might be at most one visible
1884  * version of the row, but we don't try to optimize for that.
1885  */
1886  ctid = *tid;
1887  priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1888  for (;;)
1889  {
1890  Buffer buffer;
1891  Page page;
1892  OffsetNumber offnum;
1893  ItemId lp;
1894  HeapTupleData tp;
1895  bool valid;
1896 
1897  /*
1898  * Read, pin, and lock the page.
1899  */
1900  buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1901  LockBuffer(buffer, BUFFER_LOCK_SHARE);
1902  page = BufferGetPage(buffer);
1903  TestForOldSnapshot(snapshot, relation, page);
1904 
1905  /*
1906  * Check for bogus item number. This is not treated as an error
1907  * condition because it can happen while following a t_ctid link. We
1908  * just assume that the prior tid is OK and return it unchanged.
1909  */
1910  offnum = ItemPointerGetOffsetNumber(&ctid);
1911  if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1912  {
1913  UnlockReleaseBuffer(buffer);
1914  break;
1915  }
1916  lp = PageGetItemId(page, offnum);
1917  if (!ItemIdIsNormal(lp))
1918  {
1919  UnlockReleaseBuffer(buffer);
1920  break;
1921  }
1922 
1923  /* OK to access the tuple */
1924  tp.t_self = ctid;
1925  tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1926  tp.t_len = ItemIdGetLength(lp);
1927  tp.t_tableOid = RelationGetRelid(relation);
1928 
1929  /*
1930  * After following a t_ctid link, we might arrive at an unrelated
1931  * tuple. Check for XMIN match.
1932  */
1933  if (TransactionIdIsValid(priorXmax) &&
1935  {
1936  UnlockReleaseBuffer(buffer);
1937  break;
1938  }
1939 
1940  /*
1941  * Check tuple visibility; if visible, set it as the new result
1942  * candidate.
1943  */
1944  valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
1945  HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
1946  if (valid)
1947  *tid = ctid;
1948 
1949  /*
1950  * If there's a valid t_ctid link, follow it, else we're done.
1951  */
1952  if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
1956  {
1957  UnlockReleaseBuffer(buffer);
1958  break;
1959  }
1960 
1961  ctid = tp.t_data->t_ctid;
1962  priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
1963  UnlockReleaseBuffer(buffer);
1964  } /* end of loop */
1965 }
#define HeapTupleHeaderIndicatesMovedPartitions(tup)
Definition: htup_details.h:440

References Assert(), BUFFER_LOCK_SHARE, BufferGetPage, HEAP_XMAX_INVALID, HeapCheckForSerializableConflictOut(), HeapTupleHeaderGetUpdateXid, HeapTupleHeaderGetXmin, HeapTupleHeaderIndicatesMovedPartitions, HeapTupleHeaderIsOnlyLocked(), HeapTupleSatisfiesVisibility(), InvalidTransactionId, ItemIdGetLength, ItemIdIsNormal, ItemPointerEquals(), ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, ItemPointerIsValid, LockBuffer(), PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, ReadBuffer(), RelationGetRelid, TableScanDescData::rs_rd, TableScanDescData::rs_snapshot, HeapTupleHeaderData::t_ctid, HeapTupleData::t_data, HeapTupleHeaderData::t_infomask, HeapTupleData::t_len, HeapTupleData::t_self, HeapTupleData::t_tableOid, TestForOldSnapshot(), TransactionIdEquals, TransactionIdIsValid, and UnlockReleaseBuffer().

◆ heap_getnext()

HeapTuple heap_getnext ( TableScanDesc  sscan,
ScanDirection  direction 
)

Definition at line 1340 of file heapam.c.

1341 {
1342  HeapScanDesc scan = (HeapScanDesc) sscan;
1343 
1344  /*
1345  * This is still widely used directly, without going through table AM, so
1346  * add a safety check. It's possible we should, at a later point,
1347  * downgrade this to an assert. The reason for checking the AM routine,
1348  * rather than the AM oid, is that this allows to write regression tests
1349  * that create another AM reusing the heap handler.
1350  */
1352  ereport(ERROR,
1353  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1354  errmsg_internal("only heap AM is supported")));
1355 
1356  /*
1357  * We don't expect direct calls to heap_getnext with valid CheckXidAlive
1358  * for catalog or regular tables. See detailed comments in xact.c where
1359  * these variables are declared. Normally we have such a check at tableam
1360  * level API but this is called from many places so we need to ensure it
1361  * here.
1362  */
1364  elog(ERROR, "unexpected heap_getnext call during logical decoding");
1365 
1366  /* Note: no locking manipulations needed */
1367 
1368  if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1369  heapgettup_pagemode(scan, direction,
1370  scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1371  else
1372  heapgettup(scan, direction,
1373  scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1374 
1375  if (scan->rs_ctup.t_data == NULL)
1376  return NULL;
1377 
1378  /*
1379  * if we get here it means we have a new current scan tuple, so point to
1380  * the proper return buffer and return the tuple.
1381  */
1382 
1384 
1385  return &scan->rs_ctup;
1386 }
#define unlikely(x)
Definition: c.h:273
static void heapgettup(HeapScanDesc scan, ScanDirection dir, int nkeys, ScanKey key)
Definition: heapam.c:506
static void heapgettup_pagemode(HeapScanDesc scan, ScanDirection dir, int nkeys, ScanKey key)
Definition: heapam.c:844
const TableAmRoutine * GetHeapamTableAmRoutine(void)
#define pgstat_count_heap_getnext(rel)
Definition: pgstat.h:1154
const struct TableAmRoutine * rd_tableam
Definition: rel.h:185
bool bsysscan
Definition: xact.c:98
TransactionId CheckXidAlive
Definition: xact.c:97

References bsysscan, CheckXidAlive, elog, ereport, errcode(), errmsg_internal(), ERROR, GetHeapamTableAmRoutine(), heapgettup(), heapgettup_pagemode(), pgstat_count_heap_getnext, RelationData::rd_tableam, HeapScanDescData::rs_base, HeapScanDescData::rs_ctup, TableScanDescData::rs_flags, TableScanDescData::rs_key, TableScanDescData::rs_nkeys, TableScanDescData::rs_rd, SO_ALLOW_PAGEMODE, HeapTupleData::t_data, TransactionIdIsValid, and unlikely.

Referenced by AlterTableMoveAll(), AlterTableSpaceOptions(), check_db_file_conflict(), createdb(), do_autovacuum(), DropSetting(), DropTableSpace(), find_typed_table_dependencies(), get_all_vacuum_rels(), get_database_list(), get_subscription_list(), get_tables_to_cluster(), get_tablespace_name(), get_tablespace_oid(), GetAllTablesPublicationRelations(), getRelationsInNamespace(), GetSchemaPublicationRelations(), heapam_index_build_range_scan(), heapam_index_validate_scan(), index_update_stats(), objectsInSchemaToOids(), pgrowlocks(), pgstat_collect_oids(), pgstat_heap(), populate_typ_list(), ReindexMultipleTables(), remove_dbtablespaces(), RemoveSubscriptionRel(), RenameTableSpace(), ThereIsAtLeastOneRole(), and vac_truncate_clog().

◆ heap_getnextslot()

bool heap_getnextslot ( TableScanDesc  sscan,
ScanDirection  direction,
TupleTableSlot slot 
)

Definition at line 1389 of file heapam.c.

1390 {
1391  HeapScanDesc scan = (HeapScanDesc) sscan;
1392 
1393  /* Note: no locking manipulations needed */
1394 
1395  if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1396  heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1397  else
1398  heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1399 
1400  if (scan->rs_ctup.t_data == NULL)
1401  {
1402  ExecClearTuple(slot);
1403  return false;
1404  }
1405 
1406  /*
1407  * if we get here it means we have a new current scan tuple, so point to
1408  * the proper return buffer and return the tuple.
1409  */
1410 
1412 
1413  ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1414  scan->rs_cbuf);
1415  return true;
1416 }
TupleTableSlot * ExecStoreBufferHeapTuple(HeapTuple tuple, TupleTableSlot *slot, Buffer buffer)
Definition: execTuples.c:1392
static TupleTableSlot * ExecClearTuple(TupleTableSlot *slot)
Definition: tuptable.h:425

References ExecClearTuple(), ExecStoreBufferHeapTuple(), heapgettup(), heapgettup_pagemode(), pgstat_count_heap_getnext, HeapScanDescData::rs_base, HeapScanDescData::rs_cbuf, HeapScanDescData::rs_ctup, TableScanDescData::rs_flags, TableScanDescData::rs_key, TableScanDescData::rs_nkeys, TableScanDescData::rs_rd, SO_ALLOW_PAGEMODE, and HeapTupleData::t_data.

◆ heap_getnextslot_tidrange()

bool heap_getnextslot_tidrange ( TableScanDesc  sscan,
ScanDirection  direction,
TupleTableSlot slot 
)

Definition at line 1492 of file heapam.c.

1494 {
1495  HeapScanDesc scan = (HeapScanDesc) sscan;
1496  ItemPointer mintid = &sscan->rs_mintid;
1497  ItemPointer maxtid = &sscan->rs_maxtid;
1498 
1499  /* Note: no locking manipulations needed */
1500  for (;;)
1501  {
1502  if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1503  heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1504  else
1505  heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1506 
1507  if (scan->rs_ctup.t_data == NULL)
1508  {
1509  ExecClearTuple(slot);
1510  return false;
1511  }
1512 
1513  /*
1514  * heap_set_tidrange will have used heap_setscanlimits to limit the
1515  * range of pages we scan to only ones that can contain the TID range
1516  * we're scanning for. Here we must filter out any tuples from these
1517  * pages that are outwith that range.
1518  */
1519  if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1520  {
1521  ExecClearTuple(slot);
1522 
1523  /*
1524  * When scanning backwards, the TIDs will be in descending order.
1525  * Future tuples in this direction will be lower still, so we can
1526  * just return false to indicate there will be no more tuples.
1527  */
1528  if (ScanDirectionIsBackward(direction))
1529  return false;
1530 
1531  continue;
1532  }
1533 
1534  /*
1535  * Likewise for the final page, we must filter out TIDs greater than
1536  * maxtid.
1537  */
1538  if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1539  {
1540  ExecClearTuple(slot);
1541 
1542  /*
1543  * When scanning forward, the TIDs will be in ascending order.
1544  * Future tuples in this direction will be higher still, so we can
1545  * just return false to indicate there will be no more tuples.
1546  */
1547  if (ScanDirectionIsForward(direction))
1548  return false;
1549  continue;
1550  }
1551 
1552  break;
1553  }
1554 
1555  /*
1556  * if we get here it means we have a new current scan tuple, so point to
1557  * the proper return buffer and return the tuple.
1558  */
1560 
1561  ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1562  return true;
1563 }
int32 ItemPointerCompare(ItemPointer arg1, ItemPointer arg2)
Definition: itemptr.c:52
#define ScanDirectionIsForward(direction)
Definition: sdir.h:55
#define ScanDirectionIsBackward(direction)
Definition: sdir.h:41
ItemPointerData rs_mintid
Definition: relscan.h:40
ItemPointerData rs_maxtid
Definition: relscan.h:41

References ExecClearTuple(), ExecStoreBufferHeapTuple(), heapgettup(), heapgettup_pagemode(), ItemPointerCompare(), pgstat_count_heap_getnext, HeapScanDescData::rs_base, HeapScanDescData::rs_cbuf, HeapScanDescData::rs_ctup, TableScanDescData::rs_flags, TableScanDescData::rs_key, TableScanDescData::rs_maxtid, TableScanDescData::rs_mintid, TableScanDescData::rs_nkeys, TableScanDescData::rs_rd, ScanDirectionIsBackward, ScanDirectionIsForward, SO_ALLOW_PAGEMODE, HeapTupleData::t_data, and HeapTupleData::t_self.

◆ heap_hot_search_buffer()

bool heap_hot_search_buffer ( ItemPointer  tid,
Relation  relation,
Buffer  buffer,
Snapshot  snapshot,
HeapTuple  heapTuple,
bool all_dead,
bool  first_call 
)

Definition at line 1710 of file heapam.c.

1713 {
1714  Page dp = (Page) BufferGetPage(buffer);
1715  TransactionId prev_xmax = InvalidTransactionId;
1716  BlockNumber blkno;
1717  OffsetNumber offnum;
1718  bool at_chain_start;
1719  bool valid;
1720  bool skip;
1721  GlobalVisState *vistest = NULL;
1722 
1723  /* If this is not the first call, previous call returned a (live!) tuple */
1724  if (all_dead)
1725  *all_dead = first_call;
1726 
1727  blkno = ItemPointerGetBlockNumber(tid);
1728  offnum = ItemPointerGetOffsetNumber(tid);
1729  at_chain_start = first_call;
1730  skip = !first_call;
1731 
1732  /* XXX: we should assert that a snapshot is pushed or registered */
1734  Assert(BufferGetBlockNumber(buffer) == blkno);
1735 
1736  /* Scan through possible multiple members of HOT-chain */
1737  for (;;)
1738  {
1739  ItemId lp;
1740 
1741  /* check for bogus TID */
1742  if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(dp))
1743  break;
1744 
1745  lp = PageGetItemId(dp, offnum);
1746 
1747  /* check for unused, dead, or redirected items */
1748  if (!ItemIdIsNormal(lp))
1749  {
1750  /* We should only see a redirect at start of chain */
1751  if (ItemIdIsRedirected(lp) && at_chain_start)
1752  {
1753  /* Follow the redirect */
1754  offnum = ItemIdGetRedirect(lp);
1755  at_chain_start = false;
1756  continue;
1757  }
1758  /* else must be end of chain */
1759  break;
1760  }
1761 
1762  /*
1763  * Update heapTuple to point to the element of the HOT chain we're
1764  * currently investigating. Having t_self set correctly is important
1765  * because the SSI checks and the *Satisfies routine for historical
1766  * MVCC snapshots need the correct tid to decide about the visibility.
1767  */
1768  heapTuple->t_data = (HeapTupleHeader) PageGetItem(dp, lp);
1769  heapTuple->t_len = ItemIdGetLength(lp);
1770  heapTuple->t_tableOid = RelationGetRelid(relation);
1771  ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1772 
1773  /*
1774  * Shouldn't see a HEAP_ONLY tuple at chain start.
1775  */
1776  if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
1777  break;
1778 
1779  /*
1780  * The xmin should match the previous xmax value, else chain is
1781  * broken.
1782  */
1783  if (TransactionIdIsValid(prev_xmax) &&
1784  !TransactionIdEquals(prev_xmax,
1785  HeapTupleHeaderGetXmin(heapTuple->t_data)))
1786  break;
1787 
1788  /*
1789  * When first_call is true (and thus, skip is initially false) we'll
1790  * return the first tuple we find. But on later passes, heapTuple
1791  * will initially be pointing to the tuple we returned last time.
1792  * Returning it again would be incorrect (and would loop forever), so
1793  * we skip it and return the next match we find.
1794  */
1795  if (!skip)
1796  {
1797  /* If it's visible per the snapshot, we must return it */
1798  valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1799  HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1800  buffer, snapshot);
1801 
1802  if (valid)
1803  {
1804  ItemPointerSetOffsetNumber(tid, offnum);
1805  PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1806  HeapTupleHeaderGetXmin(heapTuple->t_data));
1807  if (all_dead)
1808  *all_dead = false;
1809  return true;
1810  }
1811  }
1812  skip = false;
1813 
1814  /*
1815  * If we can't see it, maybe no one else can either. At caller
1816  * request, check whether all chain members are dead to all
1817  * transactions.
1818  *
1819  * Note: if you change the criterion here for what is "dead", fix the
1820  * planner's get_actual_variable_range() function to match.
1821  */
1822  if (all_dead && *all_dead)
1823  {
1824  if (!vistest)
1825  vistest = GlobalVisTestFor(relation);
1826 
1827  if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1828  *all_dead = false;
1829  }
1830 
1831  /*
1832  * Check to see if HOT chain continues past this tuple; if so fetch
1833  * the next offnum and loop around.
1834  */
1835  if (HeapTupleIsHotUpdated(heapTuple))
1836  {
1838  blkno);
1839  offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
1840  at_chain_start = false;
1841  prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1842  }
1843  else
1844  break; /* end of chain */
1845  }
1846 
1847  return false;
1848 }
bool HeapTupleIsSurelyDead(HeapTuple htup, GlobalVisState *vistest)
#define HeapTupleIsHeapOnly(tuple)
Definition: htup_details.h:679
#define HeapTupleIsHotUpdated(tuple)
Definition: htup_details.h:670
#define ItemIdGetRedirect(itemId)
Definition: itemid.h:78
#define ItemIdIsRedirected(itemId)
Definition: itemid.h:106
#define ItemPointerSet(pointer, blockNumber, offNum)
Definition: itemptr.h:127
#define ItemPointerSetOffsetNumber(pointer, offsetNumber)
Definition: itemptr.h:148
static const struct exclude_list_item skip[]
Definition: pg_checksums.c:116
GlobalVisState * GlobalVisTestFor(Relation rel)
Definition: procarray.c:4042
TransactionId RecentXmin
Definition: snapmgr.c:113

References Assert(), BufferGetBlockNumber(), BufferGetPage, GlobalVisTestFor(), HeapCheckForSerializableConflictOut(), HeapTupleHeaderGetUpdateXid, HeapTupleHeaderGetXmin, HeapTupleIsHeapOnly, HeapTupleIsHotUpdated, HeapTupleIsSurelyDead(), HeapTupleSatisfiesVisibility(), InvalidTransactionId, ItemIdGetLength, ItemIdGetRedirect, ItemIdIsNormal, ItemIdIsRedirected, ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, ItemPointerSet, ItemPointerSetOffsetNumber, PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, PredicateLockTID(), RecentXmin, RelationGetRelid, skip, HeapTupleHeaderData::t_ctid, HeapTupleData::t_data, HeapTupleData::t_len, HeapTupleData::t_self, HeapTupleData::t_tableOid, TransactionIdEquals, and TransactionIdIsValid.

Referenced by heap_index_delete_tuples(), heapam_index_fetch_tuple(), and heapam_scan_bitmap_next_block().

◆ heap_index_delete_tuples()

TransactionId heap_index_delete_tuples ( Relation  rel,
TM_IndexDeleteOp delstate 
)

Definition at line 7316 of file heapam.c.

7317 {
7318  /* Initial assumption is that earlier pruning took care of conflict */
7319  TransactionId latestRemovedXid = InvalidTransactionId;
7322  Page page = NULL;
7324  TransactionId priorXmax;
7325 #ifdef USE_PREFETCH
7326  IndexDeletePrefetchState prefetch_state;
7327  int prefetch_distance;
7328 #endif
7329  SnapshotData SnapshotNonVacuumable;
7330  int finalndeltids = 0,
7331  nblocksaccessed = 0;
7332 
7333  /* State that's only used in bottom-up index deletion case */
7334  int nblocksfavorable = 0;
7335  int curtargetfreespace = delstate->bottomupfreespace,
7336  lastfreespace = 0,
7337  actualfreespace = 0;
7338  bool bottomup_final_block = false;
7339 
7340  InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
7341 
7342  /* Sort caller's deltids array by TID for further processing */
7343  index_delete_sort(delstate);
7344 
7345  /*
7346  * Bottom-up case: resort deltids array in an order attuned to where the
7347  * greatest number of promising TIDs are to be found, and determine how
7348  * many blocks from the start of sorted array should be considered
7349  * favorable. This will also shrink the deltids array in order to
7350  * eliminate completely unfavorable blocks up front.
7351  */
7352  if (delstate->bottomup)
7353  nblocksfavorable = bottomup_sort_and_shrink(delstate);
7354 
7355 #ifdef USE_PREFETCH
7356  /* Initialize prefetch state. */
7357  prefetch_state.cur_hblkno = InvalidBlockNumber;
7358  prefetch_state.next_item = 0;
7359  prefetch_state.ndeltids = delstate->ndeltids;
7360  prefetch_state.deltids = delstate->deltids;
7361 
7362  /*
7363  * Determine the prefetch distance that we will attempt to maintain.
7364  *
7365  * Since the caller holds a buffer lock somewhere in rel, we'd better make
7366  * sure that isn't a catalog relation before we call code that does
7367  * syscache lookups, to avoid risk of deadlock.
7368  */
7369  if (IsCatalogRelation(rel))
7370  prefetch_distance = maintenance_io_concurrency;
7371  else
7372  prefetch_distance =
7374 
7375  /* Cap initial prefetch distance for bottom-up deletion caller */
7376  if (delstate->bottomup)
7377  {
7378  Assert(nblocksfavorable >= 1);
7379  Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
7380  prefetch_distance = Min(prefetch_distance, nblocksfavorable);
7381  }
7382 
7383  /* Start prefetching. */
7384  index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
7385 #endif
7386 
7387  /* Iterate over deltids, determine which to delete, check their horizon */
7388  Assert(delstate->ndeltids > 0);
7389  for (int i = 0; i < delstate->ndeltids; i++)
7390  {
7391  TM_IndexDelete *ideltid = &delstate->deltids[i];
7392  TM_IndexStatus *istatus = delstate->status + ideltid->id;
7393  ItemPointer htid = &ideltid->tid;
7394  OffsetNumber offnum;
7395 
7396  /*
7397  * Read buffer, and perform required extra steps each time a new block
7398  * is encountered. Avoid refetching if it's the same block as the one
7399  * from the last htid.
7400  */
7401  if (blkno == InvalidBlockNumber ||
7402  ItemPointerGetBlockNumber(htid) != blkno)
7403  {
7404  /*
7405  * Consider giving up early for bottom-up index deletion caller
7406  * first. (Only prefetch next-next block afterwards, when it
7407  * becomes clear that we're at least going to access the next
7408  * block in line.)
7409  *
7410  * Sometimes the first block frees so much space for bottom-up
7411  * caller that the deletion process can end without accessing any
7412  * more blocks. It is usually necessary to access 2 or 3 blocks
7413  * per bottom-up deletion operation, though.
7414  */
7415  if (delstate->bottomup)
7416  {
7417  /*
7418  * We often allow caller to delete a few additional items
7419  * whose entries we reached after the point that space target
7420  * from caller was satisfied. The cost of accessing the page
7421  * was already paid at that point, so it made sense to finish
7422  * it off. When that happened, we finalize everything here
7423  * (by finishing off the whole bottom-up deletion operation
7424  * without needlessly paying the cost of accessing any more
7425  * blocks).
7426  */
7427  if (bottomup_final_block)
7428  break;
7429 
7430  /*
7431  * Give up when we didn't enable our caller to free any
7432  * additional space as a result of processing the page that we
7433  * just finished up with. This rule is the main way in which
7434  * we keep the cost of bottom-up deletion under control.
7435  */
7436  if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
7437  break;
7438  lastfreespace = actualfreespace; /* for next time */
7439 
7440  /*
7441  * Deletion operation (which is bottom-up) will definitely
7442  * access the next block in line. Prepare for that now.
7443  *
7444  * Decay target free space so that we don't hang on for too
7445  * long with a marginal case. (Space target is only truly
7446  * helpful when it allows us to recognize that we don't need
7447  * to access more than 1 or 2 blocks to satisfy caller due to
7448  * agreeable workload characteristics.)
7449  *
7450  * We are a bit more patient when we encounter contiguous
7451  * blocks, though: these are treated as favorable blocks. The
7452  * decay process is only applied when the next block in line
7453  * is not a favorable/contiguous block. This is not an
7454  * exception to the general rule; we still insist on finding
7455  * at least one deletable item per block accessed. See
7456  * bottomup_nblocksfavorable() for full details of the theory
7457  * behind favorable blocks and heap block locality in general.
7458  *
7459  * Note: The first block in line is always treated as a
7460  * favorable block, so the earliest possible point that the
7461  * decay can be applied is just before we access the second
7462  * block in line. The Assert() verifies this for us.
7463  */
7464  Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
7465  if (nblocksfavorable > 0)
7466  nblocksfavorable--;
7467  else
7468  curtargetfreespace /= 2;
7469  }
7470 
7471  /* release old buffer */
7472  if (BufferIsValid(buf))
7474 
7475  blkno = ItemPointerGetBlockNumber(htid);
7476  buf = ReadBuffer(rel, blkno);
7477  nblocksaccessed++;
7478  Assert(!delstate->bottomup ||
7479  nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
7480 
7481 #ifdef USE_PREFETCH
7482 
7483  /*
7484  * To maintain the prefetch distance, prefetch one more page for
7485  * each page we read.
7486  */
7487  index_delete_prefetch_buffer(rel, &prefetch_state, 1);
7488 #endif
7489 
7491 
7492  page = BufferGetPage(buf);
7493  maxoff = PageGetMaxOffsetNumber(page);
7494  }
7495 
7496  /*
7497  * In passing, detect index corruption involving an index page with a
7498  * TID that points to a location in the heap that couldn't possibly be
7499  * correct. We only do this with actual TIDs from caller's index page
7500  * (not items reached by traversing through a HOT chain).
7501  */
7502  index_delete_check_htid(delstate, page, maxoff, htid, istatus);
7503 
7504  if (istatus->knowndeletable)
7505  Assert(!delstate->bottomup && !istatus->promising);
7506  else
7507  {
7508  ItemPointerData tmp = *htid;
7509  HeapTupleData heapTuple;
7510 
7511  /* Are any tuples from this HOT chain non-vacuumable? */
7512  if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
7513  &heapTuple, NULL, true))
7514  continue; /* can't delete entry */
7515 
7516  /* Caller will delete, since whole HOT chain is vacuumable */
7517  istatus->knowndeletable = true;
7518 
7519  /* Maintain index free space info for bottom-up deletion case */
7520  if (delstate->bottomup)
7521  {
7522  Assert(istatus->freespace > 0);
7523  actualfreespace += istatus->freespace;
7524  if (actualfreespace >= curtargetfreespace)
7525  bottomup_final_block = true;
7526  }
7527  }
7528 
7529  /*
7530  * Maintain latestRemovedXid value for deletion operation as a whole
7531  * by advancing current value using heap tuple headers. This is
7532  * loosely based on the logic for pruning a HOT chain.
7533  */
7534  offnum = ItemPointerGetOffsetNumber(htid);
7535  priorXmax = InvalidTransactionId; /* cannot check first XMIN */
7536  for (;;)
7537  {
7538  ItemId lp;
7539  HeapTupleHeader htup;
7540 
7541  /* Sanity check (pure paranoia) */
7542  if (offnum < FirstOffsetNumber)
7543  break;
7544 
7545  /*
7546  * An offset past the end of page's line pointer array is possible
7547  * when the array was truncated
7548  */
7549  if (offnum > maxoff)
7550  break;
7551 
7552  lp = PageGetItemId(page, offnum);
7553  if (ItemIdIsRedirected(lp))
7554  {
7555  offnum = ItemIdGetRedirect(lp);
7556  continue;
7557  }
7558 
7559  /*
7560  * We'll often encounter LP_DEAD line pointers (especially with an
7561  * entry marked knowndeletable by our caller up front). No heap
7562  * tuple headers get examined for an htid that leads us to an
7563  * LP_DEAD item. This is okay because the earlier pruning
7564  * operation that made the line pointer LP_DEAD in the first place
7565  * must have considered the original tuple header as part of
7566  * generating its own latestRemovedXid value.
7567  *
7568  * Relying on XLOG_HEAP2_PRUNE records like this is the same
7569  * strategy that index vacuuming uses in all cases. Index VACUUM
7570  * WAL records don't even have a latestRemovedXid field of their
7571  * own for this reason.
7572  */
7573  if (!ItemIdIsNormal(lp))
7574  break;
7575 
7576  htup = (HeapTupleHeader) PageGetItem(page, lp);
7577 
7578  /*
7579  * Check the tuple XMIN against prior XMAX, if any
7580  */
7581  if (TransactionIdIsValid(priorXmax) &&
7582  !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
7583  break;
7584 
7585  HeapTupleHeaderAdvanceLatestRemovedXid(htup, &latestRemovedXid);
7586 
7587  /*
7588  * If the tuple is not HOT-updated, then we are at the end of this
7589  * HOT-chain. No need to visit later tuples from the same update
7590  * chain (they get their own index entries) -- just move on to
7591  * next htid from index AM caller.
7592  */
7593  if (!HeapTupleHeaderIsHotUpdated(htup))
7594  break;
7595 
7596  /* Advance to next HOT chain member */
7597  Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
7598  offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
7599  priorXmax = HeapTupleHeaderGetUpdateXid(htup);
7600  }
7601 
7602  /* Enable further/final shrinking of deltids for caller */
7603  finalndeltids = i + 1;
7604  }
7605 
7607 
7608  /*
7609  * Shrink deltids array to exclude non-deletable entries at the end. This
7610  * is not just a minor optimization. Final deltids array size might be
7611  * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
7612  * ndeltids being zero in all cases with zero total deletable entries.
7613  */
7614  Assert(finalndeltids > 0 || delstate->bottomup);
7615  delstate->ndeltids = finalndeltids;
7616 
7617  return latestRemovedXid;
7618 }
int maintenance_io_concurrency
Definition: bufmgr.c:150
bool IsCatalogRelation(Relation relation)
Definition: catalog.c:104
static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
Definition: heapam.c:7873
void HeapTupleHeaderAdvanceLatestRemovedXid(HeapTupleHeader tuple, TransactionId *latestRemovedXid)
Definition: heapam.c:7169
static void index_delete_check_htid(TM_IndexDeleteOp *delstate, Page page, OffsetNumber maxoff, ItemPointer htid, TM_IndexStatus *istatus)
Definition: heapam.c:7256
bool heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer, Snapshot snapshot, HeapTuple heapTuple, bool *all_dead, bool first_call)
Definition: heapam.c:1710
static void index_delete_sort(TM_IndexDeleteOp *delstate)
Definition: heapam.c:7660
#define HeapTupleHeaderIsHotUpdated(tup)
Definition: htup_details.h:478
#define InvalidOffsetNumber
Definition: off.h:26
#define FirstOffsetNumber
Definition: off.h:27
static char * buf
Definition: pg_test_fsync.c:70
#define InitNonVacuumableSnapshot(snapshotdata, vistestp)
Definition: snapmgr.h:82
int get_tablespace_maintenance_io_concurrency(Oid spcid)
Definition: spccache.c:228
int bottomupfreespace
Definition: tableam.h:226
bool knowndeletable
Definition: tableam.h:196
int16 freespace
Definition: tableam.h:200

References Assert(), TM_IndexDeleteOp::bottomup, BOTTOMUP_MAX_NBLOCKS, bottomup_sort_and_shrink(), TM_IndexDeleteOp::bottomupfreespace, buf, BUFFER_LOCK_SHARE, BufferGetPage, BufferIsValid, TM_IndexDeleteOp::deltids, FirstOffsetNumber, TM_IndexStatus::freespace, get_tablespace_maintenance_io_concurrency(), GlobalVisTestFor(), heap_hot_search_buffer(), HeapTupleHeaderAdvanceLatestRemovedXid(), HeapTupleHeaderGetUpdateXid, HeapTupleHeaderGetXmin, HeapTupleHeaderIsHotUpdated, i, TM_IndexDelete::id, index_delete_check_htid(), index_delete_sort(), InitNonVacuumableSnapshot, InvalidBlockNumber, InvalidBuffer, InvalidOffsetNumber, InvalidTransactionId, IsCatalogRelation(), ItemIdGetRedirect, ItemIdIsNormal, ItemIdIsRedirected, ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, TM_IndexStatus::knowndeletable, LockBuffer(), maintenance_io_concurrency, Min, TM_IndexDeleteOp::ndeltids, PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, TM_IndexStatus::promising, RelationData::rd_rel, ReadBuffer(), TM_IndexDeleteOp::status, HeapTupleHeaderData::t_ctid, TM_IndexDelete::tid, TransactionIdEquals, TransactionIdIsValid, and UnlockReleaseBuffer().

◆ heap_inplace_update()

void heap_inplace_update ( Relation  relation,
HeapTuple  tuple 
)

Definition at line 5981 of file heapam.c.

5982 {
5983  Buffer buffer;
5984  Page page;
5985  OffsetNumber offnum;
5986  ItemId lp = NULL;
5987  HeapTupleHeader htup;
5988  uint32 oldlen;
5989  uint32 newlen;
5990 
5991  /*
5992  * For now, we don't allow parallel updates. Unlike a regular update,
5993  * this should never create a combo CID, so it might be possible to relax
5994  * this restriction, but not without more thought and testing. It's not
5995  * clear that it would be useful, anyway.
5996  */
5997  if (IsInParallelMode())
5998  ereport(ERROR,
5999  (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
6000  errmsg("cannot update tuples during a parallel operation")));
6001 
6002  buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&(tuple->t_self)));
6004  page = (Page) BufferGetPage(buffer);
6005 
6006  offnum = ItemPointerGetOffsetNumber(&(tuple->t_self));
6007  if (PageGetMaxOffsetNumber(page) >= offnum)
6008  lp = PageGetItemId(page, offnum);
6009 
6010  if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
6011  elog(ERROR, "invalid lp");
6012 
6013  htup = (HeapTupleHeader) PageGetItem(page, lp);
6014 
6015  oldlen = ItemIdGetLength(lp) - htup->t_hoff;
6016  newlen = tuple->t_len - tuple->t_data->t_hoff;
6017  if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6018  elog(ERROR, "wrong tuple length");
6019 
6020  /* NO EREPORT(ERROR) from here till changes are logged */
6022 
6023  memcpy((char *) htup + htup->t_hoff,
6024  (char *) tuple->t_data + tuple->t_data->t_hoff,
6025  newlen);
6026 
6027  MarkBufferDirty(buffer);
6028 
6029  /* XLOG stuff */
6030  if (RelationNeedsWAL(relation))
6031  {
6032  xl_heap_inplace xlrec;
6033  XLogRecPtr recptr;
6034 
6035  xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6036 
6037  XLogBeginInsert();
6038  XLogRegisterData((char *) &xlrec, SizeOfHeapInplace);
6039 
6040  XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6041  XLogRegisterBufData(0, (char *) htup + htup->t_hoff, newlen);
6042 
6043  /* inplace updates aren't decoded atm, don't log the origin */
6044 
6045  recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6046 
6047  PageSetLSN(page, recptr);
6048  }
6049 
6050  END_CRIT_SECTION();
6051 
6052  UnlockReleaseBuffer(buffer);
6053 
6054  /*
6055  * Send out shared cache inval if necessary. Note that because we only
6056  * pass the new version of the tuple, this mustn't be used for any
6057  * operations that could change catcache lookup keys. But we aren't
6058  * bothering with index updates either, so that's true a fortiori.
6059  */
6061  CacheInvalidateHeapTuple(relation, tuple, NULL);
6062 }
#define SizeOfHeapInplace
Definition: heapam_xlog.h:314
#define XLOG_HEAP_INPLACE
Definition: heapam_xlog.h:39
#define IsBootstrapProcessingMode()
Definition: miscadmin.h:406
OffsetNumber offnum
Definition: heapam_xlog.h:310
void XLogRegisterBufData(uint8 block_id, char *data, int len)
Definition: xloginsert.c:375

References BUFFER_LOCK_EXCLUSIVE, BufferGetPage, CacheInvalidateHeapTuple(), elog, END_CRIT_SECTION, ereport, errcode(), errmsg(), ERROR, IsBootstrapProcessingMode, IsInParallelMode(), ItemIdGetLength, ItemIdIsNormal, ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, LockBuffer(), MarkBufferDirty(), xl_heap_inplace::offnum, PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, PageSetLSN, ReadBuffer(), REGBUF_STANDARD, RelationNeedsWAL, SizeOfHeapInplace, START_CRIT_SECTION, HeapTupleData::t_data, HeapTupleHeaderData::t_hoff, HeapTupleData::t_len, HeapTupleData::t_self, UnlockReleaseBuffer(), XLOG_HEAP_INPLACE, XLogBeginInsert(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by create_toast_table(), index_update_stats(), vac_update_datfrozenxid(), and vac_update_relstats().

◆ heap_insert()

void heap_insert ( Relation  relation,
HeapTuple  tup,
CommandId  cid,
int  options,
BulkInsertState  bistate 
)

Definition at line 2060 of file heapam.c.

2062 {
2064  HeapTuple heaptup;
2065  Buffer buffer;
2066  Buffer vmbuffer = InvalidBuffer;
2067  bool all_visible_cleared = false;
2068 
2069  /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2071  RelationGetNumberOfAttributes(relation));
2072 
2073  /*
2074  * Fill in tuple header fields and toast the tuple if necessary.
2075  *
2076  * Note: below this point, heaptup is the data we actually intend to store
2077  * into the relation; tup is the caller's original untoasted data.
2078  */
2079  heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2080 
2081  /*
2082  * Find buffer to insert this tuple into. If the page is all visible,
2083  * this will also pin the requisite visibility map page.
2084  */
2085  buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2086  InvalidBuffer, options, bistate,
2087  &vmbuffer, NULL);
2088 
2089  /*
2090  * We're about to do the actual insert -- but check for conflict first, to
2091  * avoid possibly having to roll back work we've just done.
2092  *
2093  * This is safe without a recheck as long as there is no possibility of
2094  * another process scanning the page between this check and the insert
2095  * being visible to the scan (i.e., an exclusive buffer content lock is
2096  * continuously held from this point until the tuple insert is visible).
2097  *
2098  * For a heap insert, we only need to check for table-level SSI locks. Our
2099  * new tuple can't possibly conflict with existing tuple locks, and heap
2100  * page locks are only consolidated versions of tuple locks; they do not
2101  * lock "gaps" as index page locks do. So we don't need to specify a
2102  * buffer when making the call, which makes for a faster check.
2103  */
2105 
2106  /* NO EREPORT(ERROR) from here till changes are logged */
2108 
2109  RelationPutHeapTuple(relation, buffer, heaptup,
2110  (options & HEAP_INSERT_SPECULATIVE) != 0);
2111 
2112  if (PageIsAllVisible(BufferGetPage(buffer)))
2113  {
2114  all_visible_cleared = true;
2116  visibilitymap_clear(relation,
2117  ItemPointerGetBlockNumber(&(heaptup->t_self)),
2118  vmbuffer, VISIBILITYMAP_VALID_BITS);
2119  }
2120 
2121  /*
2122  * XXX Should we set PageSetPrunable on this page ?
2123  *
2124  * The inserting transaction may eventually abort thus making this tuple
2125  * DEAD and hence available for pruning. Though we don't want to optimize
2126  * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2127  * aborted tuple will never be pruned until next vacuum is triggered.
2128  *
2129  * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2130  */
2131 
2132  MarkBufferDirty(buffer);
2133 
2134  /* XLOG stuff */
2135  if (RelationNeedsWAL(relation))
2136  {
2137  xl_heap_insert xlrec;
2138  xl_heap_header xlhdr;
2139  XLogRecPtr recptr;
2140  Page page = BufferGetPage(buffer);
2141  uint8 info = XLOG_HEAP_INSERT;
2142  int bufflags = 0;
2143 
2144  /*
2145  * If this is a catalog, we need to transmit combo CIDs to properly
2146  * decode, so log that as well.
2147  */
2149  log_heap_new_cid(relation, heaptup);
2150 
2151  /*
2152  * If this is the single and first tuple on page, we can reinit the
2153  * page instead of restoring the whole thing. Set flag, and hide
2154  * buffer references from XLogInsert.
2155  */
2156  if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2158  {
2159  info |= XLOG_HEAP_INIT_PAGE;
2160  bufflags |= REGBUF_WILL_INIT;
2161  }
2162 
2163  xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2164  xlrec.flags = 0;
2165  if (all_visible_cleared)
2170 
2171  /*
2172  * For logical decoding, we need the tuple even if we're doing a full
2173  * page write, so make sure it's included even if we take a full-page
2174  * image. (XXX We could alternatively store a pointer into the FPW).
2175  */
2176  if (RelationIsLogicallyLogged(relation) &&
2178  {
2180  bufflags |= REGBUF_KEEP_DATA;
2181 
2182  if (IsToastRelation(relation))
2184  }
2185 
2186  XLogBeginInsert();
2187  XLogRegisterData((char *) &xlrec, SizeOfHeapInsert);
2188 
2189  xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2190  xlhdr.t_infomask = heaptup->t_data->t_infomask;
2191  xlhdr.t_hoff = heaptup->t_data->t_hoff;
2192 
2193  /*
2194  * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2195  * write the whole page to the xlog, we don't need to store
2196  * xl_heap_header in the xlog.
2197  */
2198  XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2199  XLogRegisterBufData(0, (char *) &xlhdr, SizeOfHeapHeader);
2200  /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2202  (char *) heaptup->t_data + SizeofHeapTupleHeader,
2203  heaptup->t_len - SizeofHeapTupleHeader);
2204 
2205  /* filtering by origin on a row level is much more efficient */
2207 
2208  recptr = XLogInsert(RM_HEAP_ID, info);
2209 
2210  PageSetLSN(page, recptr);
2211  }
2212 
2213  END_CRIT_SECTION();
2214 
2215  UnlockReleaseBuffer(buffer);
2216  if (vmbuffer != InvalidBuffer)
2217  ReleaseBuffer(vmbuffer);
2218 
2219  /*
2220  * If tuple is cachable, mark it for invalidation from the caches in case
2221  * we abort. Note it is OK to do this after releasing the buffer, because
2222  * the heaptup data structure is all in local memory, not in the shared
2223  * buffer.
2224  */
2225  CacheInvalidateHeapTuple(relation, heaptup, NULL);
2226 
2227  /* Note: speculative insertions are counted too, even if aborted later */
2228  pgstat_count_heap_insert(relation, 1);
2229 
2230  /*
2231  * If heaptup is a private copy, release it. Don't forget to copy t_self
2232  * back to the caller's image, too.
2233  */
2234  if (heaptup != tup)
2235  {
2236  tup->t_self = heaptup->t_self;
2237  heap_freetuple(heaptup);
2238  }
2239 }
static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid, CommandId cid, int options)
Definition: heapam.c:2248
#define HEAP_INSERT_SPECULATIVE
Definition: heapam.h:37
#define HEAP_INSERT_NO_LOGICAL
Definition: heapam.h:36
#define XLH_INSERT_ON_TOAST_RELATION
Definition: heapam_xlog.h:70
#define XLH_INSERT_IS_SPECULATIVE
Definition: heapam_xlog.h:68
#define XLH_INSERT_ALL_VISIBLE_CLEARED
Definition: heapam_xlog.h:66
#define XLOG_HEAP_INSERT
Definition: heapam_xlog.h:32
#define SizeOfHeapInsert
Definition: heapam_xlog.h:162
#define XLH_INSERT_CONTAINS_NEW_TUPLE
Definition: heapam_xlog.h:69
#define XLOG_HEAP_INIT_PAGE
Definition: heapam_xlog.h:46
void RelationPutHeapTuple(Relation relation, Buffer buffer, HeapTuple tuple, bool token)
Definition: hio.c:36
Buffer RelationGetBufferForTuple(Relation relation, Size len, Buffer otherBuffer, int options, BulkInsertState bistate, Buffer *vmbuffer, Buffer *vmbuffer_other)
Definition: hio.c:333
#define HeapTupleHeaderGetNatts(tup)
Definition: htup_details.h:525
void pgstat_count_heap_insert(Relation rel, PgStat_Counter n)
Definition: pgstat.c:2358
#define RelationGetNumberOfAttributes(relation)
Definition: rel.h:484
OffsetNumber offnum
Definition: heapam_xlog.h:156
#define REGBUF_KEEP_DATA
Definition: xloginsert.h:35
#define REGBUF_WILL_INIT
Definition: xloginsert.h:33

References Assert(), BufferGetBlockNumber(), BufferGetPage, CacheInvalidateHeapTuple(), CheckForSerializableConflictIn(), END_CRIT_SECTION, FirstOffsetNumber, xl_heap_insert::flags, GetCurrentTransactionId(), heap_freetuple(), HEAP_INSERT_NO_LOGICAL, HEAP_INSERT_SPECULATIVE, heap_prepare_insert(), HeapTupleHeaderGetNatts, InvalidBlockNumber, InvalidBuffer, IsToastRelation(), ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, log_heap_new_cid(), MarkBufferDirty(), xl_heap_insert::offnum, PageClearAllVisible, PageGetMaxOffsetNumber, PageIsAllVisible, PageSetLSN, pgstat_count_heap_insert(), REGBUF_KEEP_DATA, REGBUF_STANDARD, REGBUF_WILL_INIT, RelationGetBufferForTuple(), RelationGetNumberOfAttributes, RelationIsAccessibleInLogicalDecoding, RelationIsLogicallyLogged, RelationNeedsWAL, RelationPutHeapTuple(), ReleaseBuffer(), SizeOfHeapHeader, SizeOfHeapInsert, SizeofHeapTupleHeader, START_CRIT_SECTION, HeapTupleData::t_data, xl_heap_header::t_hoff, HeapTupleHeaderData::t_hoff, xl_heap_header::t_infomask, HeapTupleHeaderData::t_infomask, xl_heap_header::t_infomask2, HeapTupleHeaderData::t_infomask2, HeapTupleData::t_len, HeapTupleData::t_self, UnlockReleaseBuffer(), visibilitymap_clear(), VISIBILITYMAP_VALID_BITS, XLH_INSERT_ALL_VISIBLE_CLEARED, XLH_INSERT_CONTAINS_NEW_TUPLE, XLH_INSERT_IS_SPECULATIVE, XLH_INSERT_ON_TOAST_RELATION, XLOG_HEAP_INIT_PAGE, XLOG_HEAP_INSERT, XLOG_INCLUDE_ORIGIN, XLogBeginInsert(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), XLogRegisterData(), and XLogSetRecordFlags().

Referenced by heapam_tuple_insert(), heapam_tuple_insert_speculative(), simple_heap_insert(), and toast_save_datum().

◆ heap_lock_tuple()

TM_Result heap_lock_tuple ( Relation  relation,
HeapTuple  tuple,
CommandId  cid,
LockTupleMode  mode,
LockWaitPolicy  wait_policy,
bool  follow_updates,
Buffer buffer,
TM_FailureData tmfd 
)

Definition at line 4228 of file heapam.c.

4232 {
4233  TM_Result result;
4234  ItemPointer tid = &(tuple->t_self);
4235  ItemId lp;
4236  Page page;
4237  Buffer vmbuffer = InvalidBuffer;
4238  BlockNumber block;
4239  TransactionId xid,
4240  xmax;
4241  uint16 old_infomask,
4242  new_infomask,
4243  new_infomask2;
4244  bool first_time = true;
4245  bool skip_tuple_lock = false;
4246  bool have_tuple_lock = false;
4247  bool cleared_all_frozen = false;
4248 
4249  *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4250  block = ItemPointerGetBlockNumber(tid);
4251 
4252  /*
4253  * Before locking the buffer, pin the visibility map page if it appears to
4254  * be necessary. Since we haven't got the lock yet, someone else might be
4255  * in the middle of changing this, so we'll need to recheck after we have
4256  * the lock.
4257  */
4258  if (PageIsAllVisible(BufferGetPage(*buffer)))
4259  visibilitymap_pin(relation, block, &vmbuffer);
4260 
4262 
4263  page = BufferGetPage(*buffer);
4264  lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4265  Assert(ItemIdIsNormal(lp));
4266 
4267  tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4268  tuple->t_len = ItemIdGetLength(lp);
4269  tuple->t_tableOid = RelationGetRelid(relation);
4270 
4271 l3:
4272  result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4273 
4274  if (result == TM_Invisible)
4275  {
4276  /*
4277  * This is possible, but only when locking a tuple for ON CONFLICT
4278  * UPDATE. We return this value here rather than throwing an error in
4279  * order to give that case the opportunity to throw a more specific
4280  * error.
4281  */
4282  result = TM_Invisible;
4283  goto out_locked;
4284  }
4285  else if (result == TM_BeingModified ||
4286  result == TM_Updated ||
4287  result == TM_Deleted)
4288  {
4289  TransactionId xwait;
4290  uint16 infomask;
4291  uint16 infomask2;
4292  bool require_sleep;
4293  ItemPointerData t_ctid;
4294 
4295  /* must copy state data before unlocking buffer */
4296  xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4297  infomask = tuple->t_data->t_infomask;
4298  infomask2 = tuple->t_data->t_infomask2;
4299  ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4300 
4301  LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4302 
4303  /*
4304  * If any subtransaction of the current top transaction already holds
4305  * a lock as strong as or stronger than what we're requesting, we
4306  * effectively hold the desired lock already. We *must* succeed
4307  * without trying to take the tuple lock, else we will deadlock
4308  * against anyone wanting to acquire a stronger lock.
4309  *
4310  * Note we only do this the first time we loop on the HTSU result;
4311  * there is no point in testing in subsequent passes, because
4312  * evidently our own transaction cannot have acquired a new lock after
4313  * the first time we checked.
4314  */
4315  if (first_time)
4316  {
4317  first_time = false;
4318 
4319  if (infomask & HEAP_XMAX_IS_MULTI)
4320  {
4321  int i;
4322  int nmembers;
4323  MultiXactMember *members;
4324 
4325  /*
4326  * We don't need to allow old multixacts here; if that had
4327  * been the case, HeapTupleSatisfiesUpdate would have returned
4328  * MayBeUpdated and we wouldn't be here.
4329  */
4330  nmembers =
4331  GetMultiXactIdMembers(xwait, &members, false,
4332  HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4333 
4334  for (i = 0; i < nmembers; i++)
4335  {
4336  /* only consider members of our own transaction */
4337  if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4338  continue;
4339 
4340  if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4341  {
4342  pfree(members);
4343  result = TM_Ok;
4344  goto out_unlocked;
4345  }
4346  else
4347  {
4348  /*
4349  * Disable acquisition of the heavyweight tuple lock.
4350  * Otherwise, when promoting a weaker lock, we might
4351  * deadlock with another locker that has acquired the
4352  * heavyweight tuple lock and is waiting for our
4353  * transaction to finish.
4354  *
4355  * Note that in this case we still need to wait for
4356  * the multixact if required, to avoid acquiring
4357  * conflicting locks.
4358  */
4359  skip_tuple_lock = true;
4360  }
4361  }
4362 
4363  if (members)
4364  pfree(members);
4365  }
4366  else if (TransactionIdIsCurrentTransactionId(xwait))
4367  {
4368  switch (mode)
4369  {
4370  case LockTupleKeyShare:
4371  Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4372  HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4373  HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4374  result = TM_Ok;
4375  goto out_unlocked;
4376  case LockTupleShare:
4377  if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4378  HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4379  {
4380  result = TM_Ok;
4381  goto out_unlocked;
4382  }
4383  break;
4385  if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4386  {
4387  result = TM_Ok;
4388  goto out_unlocked;
4389  }
4390  break;
4391  case LockTupleExclusive:
4392  if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4393  infomask2 & HEAP_KEYS_UPDATED)
4394  {
4395  result = TM_Ok;
4396  goto out_unlocked;
4397  }
4398  break;
4399  }
4400  }
4401  }
4402 
4403  /*
4404  * Initially assume that we will have to wait for the locking
4405  * transaction(s) to finish. We check various cases below in which
4406  * this can be turned off.
4407  */
4408  require_sleep = true;
4409  if (mode == LockTupleKeyShare)
4410  {
4411  /*
4412  * If we're requesting KeyShare, and there's no update present, we
4413  * don't need to wait. Even if there is an update, we can still
4414  * continue if the key hasn't been modified.
4415  *
4416  * However, if there are updates, we need to walk the update chain
4417  * to mark future versions of the row as locked, too. That way,
4418  * if somebody deletes that future version, we're protected
4419  * against the key going away. This locking of future versions
4420  * could block momentarily, if a concurrent transaction is
4421  * deleting a key; or it could return a value to the effect that
4422  * the transaction deleting the key has already committed. So we
4423  * do this before re-locking the buffer; otherwise this would be
4424  * prone to deadlocks.
4425  *
4426  * Note that the TID we're locking was grabbed before we unlocked
4427  * the buffer. For it to change while we're not looking, the
4428  * other properties we're testing for below after re-locking the
4429  * buffer would also change, in which case we would restart this
4430  * loop above.
4431  */
4432  if (!(infomask2 & HEAP_KEYS_UPDATED))
4433  {
4434  bool updated;
4435 
4436  updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4437 
4438  /*
4439  * If there are updates, follow the update chain; bail out if
4440  * that cannot be done.
4441  */
4442  if (follow_updates && updated)
4443  {
4444  TM_Result res;
4445 
4446  res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
4448  mode);
4449  if (res != TM_Ok)
4450  {
4451  result = res;
4452  /* recovery code expects to have buffer lock held */
4454  goto failed;
4455  }
4456  }
4457 
4459 
4460  /*
4461  * Make sure it's still an appropriate lock, else start over.
4462  * Also, if it wasn't updated before we released the lock, but
4463  * is updated now, we start over too; the reason is that we
4464  * now need to follow the update chain to lock the new
4465  * versions.
4466  */
4467  if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4468  ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4469  !updated))
4470  goto l3;
4471 
4472  /* Things look okay, so we can skip sleeping */
4473  require_sleep = false;
4474 
4475  /*
4476  * Note we allow Xmax to change here; other updaters/lockers
4477  * could have modified it before we grabbed the buffer lock.
4478  * However, this is not a problem, because with the recheck we
4479  * just did we ensure that they still don't conflict with the
4480  * lock we want.
4481  */
4482  }
4483  }
4484  else if (mode == LockTupleShare)
4485  {
4486  /*
4487  * If we're requesting Share, we can similarly avoid sleeping if
4488  * there's no update and no exclusive lock present.
4489  */
4490  if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4491  !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4492  {
4494 
4495  /*
4496  * Make sure it's still an appropriate lock, else start over.
4497  * See above about allowing xmax to change.
4498  */
4499  if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4501  goto l3;
4502  require_sleep = false;
4503  }
4504  }
4505  else if (mode == LockTupleNoKeyExclusive)
4506  {
4507  /*
4508  * If we're requesting NoKeyExclusive, we might also be able to
4509  * avoid sleeping; just ensure that there no conflicting lock
4510  * already acquired.
4511  */
4512  if (infomask & HEAP_XMAX_IS_MULTI)
4513  {
4514  if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4515  mode, NULL))
4516  {
4517  /*
4518  * No conflict, but if the xmax changed under us in the
4519  * meantime, start over.
4520  */
4522  if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4524  xwait))
4525  goto l3;
4526 
4527  /* otherwise, we're good */
4528  require_sleep = false;
4529  }
4530  }
4531  else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4532  {
4534 
4535  /* if the xmax changed in the meantime, start over */
4536  if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4538  xwait))
4539  goto l3;
4540  /* otherwise, we're good */
4541  require_sleep = false;
4542  }
4543  }
4544 
4545  /*
4546  * As a check independent from those above, we can also avoid sleeping
4547  * if the current transaction is the sole locker of the tuple. Note
4548  * that the strength of the lock already held is irrelevant; this is
4549  * not about recording the lock in Xmax (which will be done regardless
4550  * of this optimization, below). Also, note that the cases where we
4551  * hold a lock stronger than we are requesting are already handled
4552  * above by not doing anything.
4553  *
4554  * Note we only deal with the non-multixact case here; MultiXactIdWait
4555  * is well equipped to deal with this situation on its own.
4556  */
4557  if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4559  {
4560  /* ... but if the xmax changed in the meantime, start over */
4562  if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4564  xwait))
4565  goto l3;
4567  require_sleep = false;
4568  }
4569 
4570  /*
4571  * Time to sleep on the other transaction/multixact, if necessary.
4572  *
4573  * If the other transaction is an update/delete that's already
4574  * committed, then sleeping cannot possibly do any good: if we're
4575  * required to sleep, get out to raise an error instead.
4576  *
4577  * By here, we either have already acquired the buffer exclusive lock,
4578  * or we must wait for the locking transaction or multixact; so below
4579  * we ensure that we grab buffer lock after the sleep.
4580  */
4581  if (require_sleep && (result == TM_Updated || result == TM_Deleted))
4582  {
4584  goto failed;
4585  }
4586  else if (require_sleep)
4587  {
4588  /*
4589  * Acquire tuple lock to establish our priority for the tuple, or
4590  * die trying. LockTuple will release us when we are next-in-line
4591  * for the tuple. We must do this even if we are share-locking,
4592  * but not if we already have a weaker lock on the tuple.
4593  *
4594  * If we are forced to "start over" below, we keep the tuple lock;
4595  * this arranges that we stay at the head of the line while
4596  * rechecking tuple state.
4597  */
4598  if (!skip_tuple_lock &&
4599  !heap_acquire_tuplock(relation, tid, mode, wait_policy,
4600  &have_tuple_lock))
4601  {
4602  /*
4603  * This can only happen if wait_policy is Skip and the lock
4604  * couldn't be obtained.
4605  */
4606  result = TM_WouldBlock;
4607  /* recovery code expects to have buffer lock held */
4609  goto failed;
4610  }
4611 
4612  if (infomask & HEAP_XMAX_IS_MULTI)
4613  {
4615 
4616  /* We only ever lock tuples, never update them */
4618  elog(ERROR, "invalid lock mode in heap_lock_tuple");
4619 
4620  /* wait for multixact to end, or die trying */
4621  switch (wait_policy)
4622  {
4623  case LockWaitBlock:
4624  MultiXactIdWait((MultiXactId) xwait, status, infomask,
4625  relation, &tuple->t_self, XLTW_Lock, NULL);
4626  break;
4627  case LockWaitSkip:
4629  status, infomask, relation,
4630  NULL))
4631  {
4632  result = TM_WouldBlock;
4633  /* recovery code expects to have buffer lock held */
4635  goto failed;
4636  }
4637  break;
4638  case LockWaitError:
4640  status, infomask, relation,
4641  NULL))
4642  ereport(ERROR,
4643  (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4644  errmsg("could not obtain lock on row in relation \"%s\"",
4645  RelationGetRelationName(relation))));
4646 
4647  break;
4648  }
4649 
4650  /*
4651  * Of course, the multixact might not be done here: if we're
4652  * requesting a light lock mode, other transactions with light
4653  * locks could still be alive, as well as locks owned by our
4654  * own xact or other subxacts of this backend. We need to
4655  * preserve the surviving MultiXact members. Note that it
4656  * isn't absolutely necessary in the latter case, but doing so
4657  * is simpler.
4658  */
4659  }
4660  else
4661  {
4662  /* wait for regular transaction to end, or die trying */
4663  switch (wait_policy)
4664  {
4665  case LockWaitBlock:
4666  XactLockTableWait(xwait, relation, &tuple->t_self,
4667  XLTW_Lock);
4668  break;
4669  case LockWaitSkip:
4670  if (!ConditionalXactLockTableWait(xwait))
4671  {
4672  result = TM_WouldBlock;
4673  /* recovery code expects to have buffer lock held */
4675  goto failed;
4676  }
4677  break;
4678  case LockWaitError:
4679  if (!ConditionalXactLockTableWait(xwait))
4680  ereport(ERROR,
4681  (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4682  errmsg("could not obtain lock on row in relation \"%s\"",
4683  RelationGetRelationName(relation))));
4684  break;
4685  }
4686  }
4687 
4688  /* if there are updates, follow the update chain */
4689  if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask))
4690  {
4691  TM_Result res;
4692 
4693  res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
4695  mode);
4696  if (res != TM_Ok)
4697  {
4698  result = res;
4699  /* recovery code expects to have buffer lock held */
4701  goto failed;
4702  }
4703  }
4704 
4706 
4707  /*
4708  * xwait is done, but if xwait had just locked the tuple then some
4709  * other xact could update this tuple before we get to this point.
4710  * Check for xmax change, and start over if so.
4711  */
4712  if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4714  xwait))
4715  goto l3;
4716 
4717  if (!(infomask & HEAP_XMAX_IS_MULTI))
4718  {
4719  /*
4720  * Otherwise check if it committed or aborted. Note we cannot
4721  * be here if the tuple was only locked by somebody who didn't
4722  * conflict with us; that would have been handled above. So
4723  * that transaction must necessarily be gone by now. But
4724  * don't check for this in the multixact case, because some
4725  * locker transactions might still be running.
4726  */
4727  UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
4728  }
4729  }
4730 
4731  /* By here, we're certain that we hold buffer exclusive lock again */
4732 
4733  /*
4734  * We may lock if previous xmax aborted, or if it committed but only
4735  * locked the tuple without updating it; or if we didn't have to wait
4736  * at all for whatever reason.
4737  */
4738  if (!require_sleep ||
4739  (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
4742  result = TM_Ok;
4743  else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
4744  result = TM_Updated;
4745  else
4746  result = TM_Deleted;
4747  }
4748 
4749 failed:
4750  if (result != TM_Ok)
4751  {
4752  Assert(result == TM_SelfModified || result == TM_Updated ||
4753  result == TM_Deleted || result == TM_WouldBlock);
4754 
4755  /*
4756  * When locking a tuple under LockWaitSkip semantics and we fail with
4757  * TM_WouldBlock above, it's possible for concurrent transactions to
4758  * release the lock and set HEAP_XMAX_INVALID in the meantime. So
4759  * this assert is slightly different from the equivalent one in
4760  * heap_delete and heap_update.
4761  */
4762  Assert((result == TM_WouldBlock) ||
4763  !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
4764  Assert(result != TM_Updated ||
4765  !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
4766  tmfd->ctid = tuple->t_data->t_ctid;
4767  tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
4768  if (result == TM_SelfModified)
4769  tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
4770  else
4771  tmfd->cmax = InvalidCommandId;
4772  goto out_locked;
4773  }
4774 
4775  /*
4776  * If we didn't pin the visibility map page and the page has become all
4777  * visible while we were busy locking the buffer, or during some
4778  * subsequent window during which we had it unlocked, we'll have to unlock
4779  * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
4780  * unfortunate, especially since we'll now have to recheck whether the
4781  * tuple has been locked or updated under us, but hopefully it won't
4782  * happen very often.
4783  */
4784  if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
4785  {
4786  LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4787  visibilitymap_pin(relation, block, &vmbuffer);
4789  goto l3;
4790  }
4791 
4792  xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
4793  old_infomask = tuple->t_data->t_infomask;
4794 
4795  /*
4796  * If this is the first possibly-multixact-able operation in the current
4797  * transaction, set my per-backend OldestMemberMXactId setting. We can be
4798  * certain that the transaction will never become a member of any older
4799  * MultiXactIds than that. (We have to do this even if we end up just
4800  * using our own TransactionId below, since some other backend could
4801  * incorporate our XID into a MultiXact immediately afterwards.)
4802  */
4804 
4805  /*
4806  * Compute the new xmax and infomask to store into the tuple. Note we do
4807  * not modify the tuple just yet, because that would leave it in the wrong
4808  * state if multixact.c elogs.
4809  */
4810  compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
4811  GetCurrentTransactionId(), mode, false,
4812  &xid, &new_infomask, &new_infomask2);
4813 
4815 
4816  /*
4817  * Store transaction information of xact locking the tuple.
4818  *
4819  * Note: Cmax is meaningless in this context, so don't set it; this avoids
4820  * possibly generating a useless combo CID. Moreover, if we're locking a
4821  * previously updated tuple, it's important to preserve the Cmax.
4822  *
4823  * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
4824  * we would break the HOT chain.
4825  */
4826  tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
4827  tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4828  tuple->t_data->t_infomask |= new_infomask;
4829  tuple->t_data->t_infomask2 |= new_infomask2;
4830  if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
4832  HeapTupleHeaderSetXmax(tuple->t_data, xid);
4833 
4834  /*
4835  * Make sure there is no forward chain link in t_ctid. Note that in the
4836  * cases where the tuple has been updated, we must not overwrite t_ctid,
4837  * because it was set by the updater. Moreover, if the tuple has been
4838  * updated, we need to follow the update chain to lock the new versions of
4839  * the tuple as well.
4840  */
4841  if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
4842  tuple->t_data->t_ctid = *tid;
4843 
4844  /* Clear only the all-frozen bit on visibility map if needed */
4845  if (PageIsAllVisible(page) &&
4846  visibilitymap_clear(relation, block, vmbuffer,
4848  cleared_all_frozen = true;
4849 
4850 
4851  MarkBufferDirty(*buffer);
4852 
4853  /*
4854  * XLOG stuff. You might think that we don't need an XLOG record because
4855  * there is no state change worth restoring after a crash. You would be
4856  * wrong however: we have just written either a TransactionId or a
4857  * MultiXactId that may never have been seen on disk before, and we need
4858  * to make sure that there are XLOG entries covering those ID numbers.
4859  * Else the same IDs might be re-used after a crash, which would be
4860  * disastrous if this page made it to disk before the crash. Essentially
4861  * we have to enforce the WAL log-before-data rule even in this case.
4862  * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
4863  * entries for everything anyway.)
4864  */
4865  if (RelationNeedsWAL(relation))
4866  {
4867  xl_heap_lock xlrec;
4868  XLogRecPtr recptr;
4869 
4870  XLogBeginInsert();
4871  XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
4872 
4873  xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
4874  xlrec.locking_xid = xid;
4875  xlrec.infobits_set = compute_infobits(new_infomask,
4876  tuple->t_data->t_infomask2);
4877  xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
4878  XLogRegisterData((char *) &xlrec, SizeOfHeapLock);
4879 
4880  /* we don't decode row locks atm, so no need to log the origin */
4881 
4882  recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
4883 
4884  PageSetLSN(page, recptr);
4885  }
4886 
4887  END_CRIT_SECTION();
4888 
4889  result = TM_Ok;
4890 
4891 out_locked:
4892  LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4893 
4894 out_unlocked:
4895  if (BufferIsValid(vmbuffer))
4896  ReleaseBuffer(vmbuffer);
4897 
4898  /*
4899  * Don't update the visibility map here. Locking a tuple doesn't change
4900  * visibility info.
4901  */
4902 
4903  /*
4904  * Now that we have successfully marked the tuple as locked, we can
4905  * release the lmgr tuple lock, if we had it.
4906  */
4907  if (have_tuple_lock)
4908  UnlockTupleTuplock(relation, tid, mode);
4909 
4910  return result;
4911 }
static TM_Result heap_lock_updated_tuple(Relation rel, HeapTuple tuple, ItemPointer ctid, TransactionId xid, LockTupleMode mode)
Definition: heapam.c:5692
static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask, Relation rel, int *remaining)
Definition: heapam.c:7021
#define XLH_LOCK_ALL_FROZEN_CLEARED
Definition: heapam_xlog.h:275
#define XLOG_HEAP_LOCK
Definition: heapam_xlog.h:38
#define SizeOfHeapLock
Definition: heapam_xlog.h:286
#define ItemPointerCopy(fromPointer, toPointer)
Definition: itemptr.h:161
@ XLTW_Lock
Definition: lmgr.h:29
OffsetNumber offnum
Definition: heapam_xlog.h:281
int8 infobits_set
Definition: heapam_xlog.h:282
TransactionId locking_xid
Definition: heapam_xlog.h:280
@ TM_WouldBlock
Definition: tableam.h:102
#define VISIBILITYMAP_ALL_FROZEN

References Assert(), BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetPage, BufferIsValid, TM_FailureData::cmax, compute_infobits(), compute_new_xmax_infomask(), ConditionalMultiXactIdWait(), ConditionalXactLockTableWait(), TM_FailureData::ctid, DoesMultiXactIdConflict(), elog, END_CRIT_SECTION, ereport, errcode(), errmsg(), ERROR, xl_heap_lock::flags, get_mxact_status_for_lock(), GetCurrentTransactionId(), GetMultiXactIdMembers(), heap_acquire_tuplock(), HEAP_KEYS_UPDATED, heap_lock_updated_tuple(), HEAP_XMAX_BITS, HEAP_XMAX_INVALID, HEAP_XMAX_IS_EXCL_LOCKED, HEAP_XMAX_IS_KEYSHR_LOCKED, HEAP_XMAX_IS_LOCKED_ONLY, HEAP_XMAX_IS_MULTI, HEAP_XMAX_IS_SHR_LOCKED, HeapTupleHeaderClearHotUpdated, HeapTupleHeaderGetCmax(), HeapTupleHeaderGetRawXmax, HeapTupleHeaderGetUpdateXid, HeapTupleHeaderIsOnlyLocked(), HeapTupleHeaderSetXmax, HeapTupleSatisfiesUpdate(), i, xl_heap_lock::infobits_set, InvalidBuffer, InvalidCommandId, ItemIdGetLength, ItemIdIsNormal, ItemPointerCopy, ItemPointerEquals(), ItemPointerGetBlockNumber, ItemPointerGetOffsetNumber, LockBuffer(), xl_heap_lock::locking_xid, LockTupleExclusive, LockTupleKeyShare, LockTupleNoKeyExclusive, LockTupleShare, LockWaitBlock, LockWaitError, LockWaitSkip, MarkBufferDirty(), mode, MultiXactIdSetOldestMember(), MultiXactIdWait(), MultiXactStatusNoKeyUpdate, xl_heap_lock::offnum, PageGetItem, PageGetItemId, PageIsAllVisible, PageSetLSN, pfree(), ReadBuffer(), REGBUF_STANDARD, RelationGetRelationName, RelationGetRelid, RelationNeedsWAL, ReleaseBuffer(), res, SizeOfHeapLock, START_CRIT_SECTION, status(), HeapTupleHeaderData::t_ctid, HeapTupleData::t_data, HeapTupleHeaderData::t_infomask, HeapTupleHeaderData::t_infomask2, HeapTupleData::t_len, HeapTupleData::t_self, HeapTupleData::t_tableOid, TM_BeingModified, TM_Deleted, TM_Invisible, TM_Ok, TM_SelfModified, TM_Updated, TM_WouldBlock, TransactionIdEquals, TransactionIdIsCurrentTransactionId(), TUPLOCK_from_mxstatus, UnlockTupleTuplock, UpdateXmaxHintBits(), VISIBILITYMAP_ALL_FROZEN, visibilitymap_clear(), visibilitymap_pin(), XactLockTableWait(), XLH_LOCK_ALL_FROZEN_CLEARED, XLOG_HEAP_LOCK, XLogBeginInsert(), XLogInsert(), XLogRegisterBuffer(), XLogRegisterData(), XLTW_Lock, TM_FailureData::xmax, and xmax_infomask_changed().

Referenced by heapam_tuple_lock().

◆ heap_lock_updated_tuple()

static TM_Result heap_lock_updated_tuple ( Relation  rel,
HeapTuple  tuple,
ItemPointer  ctid,
TransactionId  xid,
LockTupleMode  mode 
)
static

Definition at line 5692 of file heapam.c.

5694 {
5695  /*
5696  * If the tuple has not been updated, or has moved into another partition
5697  * (effectively a delete) stop here.
5698  */
5700  !ItemPointerEquals(&tuple->t_self, ctid))
5701  {
5702  /*
5703  * If this is the first possibly-multixact-able operation in the
5704  * current transaction, set my per-backend OldestMemberMXactId
5705  * setting. We can be certain that the transaction will never become a
5706  * member of any older MultiXactIds than that. (We have to do this
5707  * even if we end up just using our own TransactionId below, since
5708  * some other backend could incorporate our XID into a MultiXact
5709  * immediately afterwards.)
5710  */
5712 
5713  return heap_lock_updated_tuple_rec(rel, ctid, xid, mode);
5714  }
5715 
5716  /* nothing to lock */
5717  return TM_Ok;
5718 }
static TM_Result heap_lock_updated_tuple_rec(Relation rel, ItemPointer tid, TransactionId xid, LockTupleMode mode)
Definition: heapam.c:5347

References heap_lock_updated_tuple_rec(), HeapTupleHeaderIndicatesMovedPartitions, ItemPointerEquals(), mode, MultiXactIdSetOldestMember(), HeapTupleData::t_data, HeapTupleData::t_self, and TM_Ok.

Referenced by heap_lock_tuple().

◆ heap_lock_updated_tuple_rec()