PostgreSQL Source Code  git master
twophase.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * twophase.c
4  * Two-phase commit support functions.
5  *
6  * Portions Copyright (c) 1996-2020, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  * IDENTIFICATION
10  * src/backend/access/transam/twophase.c
11  *
12  * NOTES
13  * Each global transaction is associated with a global transaction
14  * identifier (GID). The client assigns a GID to a postgres
15  * transaction with the PREPARE TRANSACTION command.
16  *
17  * We keep all active global transactions in a shared memory array.
18  * When the PREPARE TRANSACTION command is issued, the GID is
19  * reserved for the transaction in the array. This is done before
20  * a WAL entry is made, because the reservation checks for duplicate
21  * GIDs and aborts the transaction if there already is a global
22  * transaction in prepared state with the same GID.
23  *
24  * A global transaction (gxact) also has dummy PGXACT and PGPROC; this is
25  * what keeps the XID considered running by TransactionIdIsInProgress.
26  * It is also convenient as a PGPROC to hook the gxact's locks to.
27  *
28  * Information to recover prepared transactions in case of crash is
29  * now stored in WAL for the common case. In some cases there will be
30  * an extended period between preparing a GXACT and commit/abort, in
31  * which case we need to separately record prepared transaction data
32  * in permanent storage. This includes locking information, pending
33  * notifications etc. All that state information is written to the
34  * per-transaction state file in the pg_twophase directory.
35  * All prepared transactions will be written prior to shutdown.
36  *
37  * Life track of state data is following:
38  *
39  * * On PREPARE TRANSACTION backend writes state data only to the WAL and
40  * stores pointer to the start of the WAL record in
41  * gxact->prepare_start_lsn.
42  * * If COMMIT occurs before checkpoint then backend reads data from WAL
43  * using prepare_start_lsn.
44  * * On checkpoint state data copied to files in pg_twophase directory and
45  * fsynced
46  * * If COMMIT happens after checkpoint then backend reads state data from
47  * files
48  *
49  * During replay and replication, TwoPhaseState also holds information
50  * about active prepared transactions that haven't been moved to disk yet.
51  *
52  * Replay of twophase records happens by the following rules:
53  *
54  * * At the beginning of recovery, pg_twophase is scanned once, filling
55  * TwoPhaseState with entries marked with gxact->inredo and
56  * gxact->ondisk. Two-phase file data older than the XID horizon of
57  * the redo position are discarded.
58  * * On PREPARE redo, the transaction is added to TwoPhaseState->prepXacts.
59  * gxact->inredo is set to true for such entries.
60  * * On Checkpoint we iterate through TwoPhaseState->prepXacts entries
61  * that have gxact->inredo set and are behind the redo_horizon. We
62  * save them to disk and then switch gxact->ondisk to true.
63  * * On COMMIT/ABORT we delete the entry from TwoPhaseState->prepXacts.
64  * If gxact->ondisk is true, the corresponding entry from the disk
65  * is additionally deleted.
66  * * RecoverPreparedTransactions(), StandbyRecoverPreparedTransactions()
67  * and PrescanPreparedTransactions() have been modified to go through
68  * gxact->inredo entries that have not made it to disk.
69  *
70  *-------------------------------------------------------------------------
71  */
72 #include "postgres.h"
73 
74 #include <fcntl.h>
75 #include <sys/stat.h>
76 #include <time.h>
77 #include <unistd.h>
78 
79 #include "access/commit_ts.h"
80 #include "access/htup_details.h"
81 #include "access/subtrans.h"
82 #include "access/transam.h"
83 #include "access/twophase.h"
84 #include "access/twophase_rmgr.h"
85 #include "access/xact.h"
86 #include "access/xlog.h"
87 #include "access/xloginsert.h"
88 #include "access/xlogreader.h"
89 #include "access/xlogutils.h"
90 #include "catalog/pg_type.h"
91 #include "catalog/storage.h"
92 #include "funcapi.h"
93 #include "miscadmin.h"
94 #include "pg_trace.h"
95 #include "pgstat.h"
96 #include "replication/origin.h"
97 #include "replication/syncrep.h"
98 #include "replication/walsender.h"
99 #include "storage/fd.h"
100 #include "storage/ipc.h"
101 #include "storage/md.h"
102 #include "storage/predicate.h"
103 #include "storage/proc.h"
104 #include "storage/procarray.h"
105 #include "storage/sinvaladt.h"
106 #include "storage/smgr.h"
107 #include "utils/builtins.h"
108 #include "utils/memutils.h"
109 #include "utils/timestamp.h"
110 
111 /*
112  * Directory where Two-phase commit files reside within PGDATA
113  */
114 #define TWOPHASE_DIR "pg_twophase"
115 
116 /* GUC variable, can't be changed after startup */
118 
119 /*
120  * This struct describes one global transaction that is in prepared state
121  * or attempting to become prepared.
122  *
123  * The lifecycle of a global transaction is:
124  *
125  * 1. After checking that the requested GID is not in use, set up an entry in
126  * the TwoPhaseState->prepXacts array with the correct GID and valid = false,
127  * and mark it as locked by my backend.
128  *
129  * 2. After successfully completing prepare, set valid = true and enter the
130  * referenced PGPROC into the global ProcArray.
131  *
132  * 3. To begin COMMIT PREPARED or ROLLBACK PREPARED, check that the entry is
133  * valid and not locked, then mark the entry as locked by storing my current
134  * backend ID into locking_backend. This prevents concurrent attempts to
135  * commit or rollback the same prepared xact.
136  *
137  * 4. On completion of COMMIT PREPARED or ROLLBACK PREPARED, remove the entry
138  * from the ProcArray and the TwoPhaseState->prepXacts array and return it to
139  * the freelist.
140  *
141  * Note that if the preparing transaction fails between steps 1 and 2, the
142  * entry must be removed so that the GID and the GlobalTransaction struct
143  * can be reused. See AtAbort_Twophase().
144  *
145  * typedef struct GlobalTransactionData *GlobalTransaction appears in
146  * twophase.h
147  */
148 
149 typedef struct GlobalTransactionData
150 {
151  GlobalTransaction next; /* list link for free list */
152  int pgprocno; /* ID of associated dummy PGPROC */
153  BackendId dummyBackendId; /* similar to backend id for backends */
154  TimestampTz prepared_at; /* time of preparation */
155 
156  /*
157  * Note that we need to keep track of two LSNs for each GXACT. We keep
158  * track of the start LSN because this is the address we must use to read
159  * state data back from WAL when committing a prepared GXACT. We keep
160  * track of the end LSN because that is the LSN we need to wait for prior
161  * to commit.
162  */
163  XLogRecPtr prepare_start_lsn; /* XLOG offset of prepare record start */
164  XLogRecPtr prepare_end_lsn; /* XLOG offset of prepare record end */
165  TransactionId xid; /* The GXACT id */
166 
167  Oid owner; /* ID of user that executed the xact */
168  BackendId locking_backend; /* backend currently working on the xact */
169  bool valid; /* true if PGPROC entry is in proc array */
170  bool ondisk; /* true if prepare state file is on disk */
171  bool inredo; /* true if entry was added via xlog_redo */
172  char gid[GIDSIZE]; /* The GID assigned to the prepared xact */
174 
175 /*
176  * Two Phase Commit shared state. Access to this struct is protected
177  * by TwoPhaseStateLock.
178  */
179 typedef struct TwoPhaseStateData
180 {
181  /* Head of linked list of free GlobalTransactionData structs */
183 
184  /* Number of valid prepXacts entries. */
186 
187  /* There are max_prepared_xacts items in this array */
190 
192 
193 /*
194  * Global transaction entry currently locked by us, if any. Note that any
195  * access to the entry pointed to by this variable must be protected by
196  * TwoPhaseStateLock, though obviously the pointer itself doesn't need to be
197  * (since it's just local memory).
198  */
200 
201 static bool twophaseExitRegistered = false;
202 
204  int nchildren,
205  TransactionId *children,
206  int nrels,
207  RelFileNode *rels,
208  int ninvalmsgs,
209  SharedInvalidationMessage *invalmsgs,
210  bool initfileinval,
211  const char *gid);
213  int nchildren,
214  TransactionId *children,
215  int nrels,
216  RelFileNode *rels,
217  const char *gid);
218 static void ProcessRecords(char *bufptr, TransactionId xid,
219  const TwoPhaseCallback callbacks[]);
220 static void RemoveGXact(GlobalTransaction gxact);
221 
222 static void XlogReadTwoPhaseData(XLogRecPtr lsn, char **buf, int *len);
225  bool fromdisk, bool setParent, bool setNextXid);
227  const char *gid, TimestampTz prepared_at, Oid owner,
228  Oid databaseid);
229 static void RemoveTwoPhaseFile(TransactionId xid, bool giveWarning);
230 static void RecreateTwoPhaseFile(TransactionId xid, void *content, int len);
231 
232 /*
233  * Initialization of shared memory
234  */
235 Size
237 {
238  Size size;
239 
240  /* Need the fixed struct, the array of pointers, and the GTD structs */
241  size = offsetof(TwoPhaseStateData, prepXacts);
242  size = add_size(size, mul_size(max_prepared_xacts,
243  sizeof(GlobalTransaction)));
244  size = MAXALIGN(size);
245  size = add_size(size, mul_size(max_prepared_xacts,
246  sizeof(GlobalTransactionData)));
247 
248  return size;
249 }
250 
251 void
253 {
254  bool found;
255 
256  TwoPhaseState = ShmemInitStruct("Prepared Transaction Table",
258  &found);
259  if (!IsUnderPostmaster)
260  {
261  GlobalTransaction gxacts;
262  int i;
263 
264  Assert(!found);
265  TwoPhaseState->freeGXacts = NULL;
266  TwoPhaseState->numPrepXacts = 0;
267 
268  /*
269  * Initialize the linked list of free GlobalTransactionData structs
270  */
271  gxacts = (GlobalTransaction)
272  ((char *) TwoPhaseState +
273  MAXALIGN(offsetof(TwoPhaseStateData, prepXacts) +
275  for (i = 0; i < max_prepared_xacts; i++)
276  {
277  /* insert into linked list */
278  gxacts[i].next = TwoPhaseState->freeGXacts;
279  TwoPhaseState->freeGXacts = &gxacts[i];
280 
281  /* associate it with a PGPROC assigned by InitProcGlobal */
283 
284  /*
285  * Assign a unique ID for each dummy proc, so that the range of
286  * dummy backend IDs immediately follows the range of normal
287  * backend IDs. We don't dare to assign a real backend ID to dummy
288  * procs, because prepared transactions don't take part in cache
289  * invalidation like a real backend ID would imply, but having a
290  * unique ID for them is nevertheless handy. This arrangement
291  * allows you to allocate an array of size (MaxBackends +
292  * max_prepared_xacts + 1), and have a slot for every backend and
293  * prepared transaction. Currently multixact.c uses that
294  * technique.
295  */
296  gxacts[i].dummyBackendId = MaxBackends + 1 + i;
297  }
298  }
299  else
300  Assert(found);
301 }
302 
303 /*
304  * Exit hook to unlock the global transaction entry we're working on.
305  */
306 static void
308 {
309  /* same logic as abort */
311 }
312 
313 /*
314  * Abort hook to unlock the global transaction entry we're working on.
315  */
316 void
318 {
319  if (MyLockedGxact == NULL)
320  return;
321 
322  /*
323  * What to do with the locked global transaction entry? If we were in the
324  * process of preparing the transaction, but haven't written the WAL
325  * record and state file yet, the transaction must not be considered as
326  * prepared. Likewise, if we are in the process of finishing an
327  * already-prepared transaction, and fail after having already written the
328  * 2nd phase commit or rollback record to the WAL, the transaction should
329  * not be considered as prepared anymore. In those cases, just remove the
330  * entry from shared memory.
331  *
332  * Otherwise, the entry must be left in place so that the transaction can
333  * be finished later, so just unlock it.
334  *
335  * If we abort during prepare, after having written the WAL record, we
336  * might not have transferred all locks and other state to the prepared
337  * transaction yet. Likewise, if we abort during commit or rollback,
338  * after having written the WAL record, we might not have released all the
339  * resources held by the transaction yet. In those cases, the in-memory
340  * state can be wrong, but it's too late to back out.
341  */
342  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
343  if (!MyLockedGxact->valid)
344  RemoveGXact(MyLockedGxact);
345  else
346  MyLockedGxact->locking_backend = InvalidBackendId;
347  LWLockRelease(TwoPhaseStateLock);
348 
349  MyLockedGxact = NULL;
350 }
351 
352 /*
353  * This is called after we have finished transferring state to the prepared
354  * PGXACT entry.
355  */
356 void
358 {
359  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
360  MyLockedGxact->locking_backend = InvalidBackendId;
361  LWLockRelease(TwoPhaseStateLock);
362 
363  MyLockedGxact = NULL;
364 }
365 
366 
367 /*
368  * MarkAsPreparing
369  * Reserve the GID for the given transaction.
370  */
373  TimestampTz prepared_at, Oid owner, Oid databaseid)
374 {
375  GlobalTransaction gxact;
376  int i;
377 
378  if (strlen(gid) >= GIDSIZE)
379  ereport(ERROR,
380  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
381  errmsg("transaction identifier \"%s\" is too long",
382  gid)));
383 
384  /* fail immediately if feature is disabled */
385  if (max_prepared_xacts == 0)
386  ereport(ERROR,
387  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
388  errmsg("prepared transactions are disabled"),
389  errhint("Set max_prepared_transactions to a nonzero value.")));
390 
391  /* on first call, register the exit hook */
393  {
395  twophaseExitRegistered = true;
396  }
397 
398  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
399 
400  /* Check for conflicting GID */
401  for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
402  {
403  gxact = TwoPhaseState->prepXacts[i];
404  if (strcmp(gxact->gid, gid) == 0)
405  {
406  ereport(ERROR,
408  errmsg("transaction identifier \"%s\" is already in use",
409  gid)));
410  }
411  }
412 
413  /* Get a free gxact from the freelist */
414  if (TwoPhaseState->freeGXacts == NULL)
415  ereport(ERROR,
416  (errcode(ERRCODE_OUT_OF_MEMORY),
417  errmsg("maximum number of prepared transactions reached"),
418  errhint("Increase max_prepared_transactions (currently %d).",
420  gxact = TwoPhaseState->freeGXacts;
421  TwoPhaseState->freeGXacts = gxact->next;
422 
423  MarkAsPreparingGuts(gxact, xid, gid, prepared_at, owner, databaseid);
424 
425  gxact->ondisk = false;
426 
427  /* And insert it into the active array */
428  Assert(TwoPhaseState->numPrepXacts < max_prepared_xacts);
429  TwoPhaseState->prepXacts[TwoPhaseState->numPrepXacts++] = gxact;
430 
431  LWLockRelease(TwoPhaseStateLock);
432 
433  return gxact;
434 }
435 
436 /*
437  * MarkAsPreparingGuts
438  *
439  * This uses a gxact struct and puts it into the active array.
440  * NOTE: this is also used when reloading a gxact after a crash; so avoid
441  * assuming that we can use very much backend context.
442  *
443  * Note: This function should be called with appropriate locks held.
444  */
445 static void
447  TimestampTz prepared_at, Oid owner, Oid databaseid)
448 {
449  PGPROC *proc;
450  PGXACT *pgxact;
451  int i;
452 
453  Assert(LWLockHeldByMeInMode(TwoPhaseStateLock, LW_EXCLUSIVE));
454 
455  Assert(gxact != NULL);
456  proc = &ProcGlobal->allProcs[gxact->pgprocno];
457  pgxact = &ProcGlobal->allPgXact[gxact->pgprocno];
458 
459  /* Initialize the PGPROC entry */
460  MemSet(proc, 0, sizeof(PGPROC));
461  proc->pgprocno = gxact->pgprocno;
462  SHMQueueElemInit(&(proc->links));
463  proc->waitStatus = STATUS_OK;
464  /* We set up the gxact's VXID as InvalidBackendId/XID */
465  proc->lxid = (LocalTransactionId) xid;
466  pgxact->xid = xid;
467  pgxact->xmin = InvalidTransactionId;
468  pgxact->delayChkpt = false;
469  pgxact->vacuumFlags = 0;
470  proc->pid = 0;
471  proc->backendId = InvalidBackendId;
472  proc->databaseId = databaseid;
473  proc->roleId = owner;
474  proc->tempNamespaceId = InvalidOid;
475  proc->isBackgroundWorker = false;
476  proc->lwWaiting = false;
477  proc->lwWaitMode = 0;
478  proc->waitLock = NULL;
479  proc->waitProcLock = NULL;
480  for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
481  SHMQueueInit(&(proc->myProcLocks[i]));
482  /* subxid data must be filled later by GXactLoadSubxactData */
483  pgxact->overflowed = false;
484  pgxact->nxids = 0;
485 
486  gxact->prepared_at = prepared_at;
487  gxact->xid = xid;
488  gxact->owner = owner;
489  gxact->locking_backend = MyBackendId;
490  gxact->valid = false;
491  gxact->inredo = false;
492  strcpy(gxact->gid, gid);
493 
494  /*
495  * Remember that we have this GlobalTransaction entry locked for us. If we
496  * abort after this, we must release it.
497  */
498  MyLockedGxact = gxact;
499 }
500 
501 /*
502  * GXactLoadSubxactData
503  *
504  * If the transaction being persisted had any subtransactions, this must
505  * be called before MarkAsPrepared() to load information into the dummy
506  * PGPROC.
507  */
508 static void
510  TransactionId *children)
511 {
512  PGPROC *proc = &ProcGlobal->allProcs[gxact->pgprocno];
513  PGXACT *pgxact = &ProcGlobal->allPgXact[gxact->pgprocno];
514 
515  /* We need no extra lock since the GXACT isn't valid yet */
516  if (nsubxacts > PGPROC_MAX_CACHED_SUBXIDS)
517  {
518  pgxact->overflowed = true;
519  nsubxacts = PGPROC_MAX_CACHED_SUBXIDS;
520  }
521  if (nsubxacts > 0)
522  {
523  memcpy(proc->subxids.xids, children,
524  nsubxacts * sizeof(TransactionId));
525  pgxact->nxids = nsubxacts;
526  }
527 }
528 
529 /*
530  * MarkAsPrepared
531  * Mark the GXACT as fully valid, and enter it into the global ProcArray.
532  *
533  * lock_held indicates whether caller already holds TwoPhaseStateLock.
534  */
535 static void
536 MarkAsPrepared(GlobalTransaction gxact, bool lock_held)
537 {
538  /* Lock here may be overkill, but I'm not convinced of that ... */
539  if (!lock_held)
540  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
541  Assert(!gxact->valid);
542  gxact->valid = true;
543  if (!lock_held)
544  LWLockRelease(TwoPhaseStateLock);
545 
546  /*
547  * Put it into the global ProcArray so TransactionIdIsInProgress considers
548  * the XID as still running.
549  */
551 }
552 
553 /*
554  * LockGXact
555  * Locate the prepared transaction and mark it busy for COMMIT or PREPARE.
556  */
557 static GlobalTransaction
558 LockGXact(const char *gid, Oid user)
559 {
560  int i;
561 
562  /* on first call, register the exit hook */
564  {
566  twophaseExitRegistered = true;
567  }
568 
569  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
570 
571  for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
572  {
573  GlobalTransaction gxact = TwoPhaseState->prepXacts[i];
574  PGPROC *proc = &ProcGlobal->allProcs[gxact->pgprocno];
575 
576  /* Ignore not-yet-valid GIDs */
577  if (!gxact->valid)
578  continue;
579  if (strcmp(gxact->gid, gid) != 0)
580  continue;
581 
582  /* Found it, but has someone else got it locked? */
583  if (gxact->locking_backend != InvalidBackendId)
584  ereport(ERROR,
585  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
586  errmsg("prepared transaction with identifier \"%s\" is busy",
587  gid)));
588 
589  if (user != gxact->owner && !superuser_arg(user))
590  ereport(ERROR,
591  (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
592  errmsg("permission denied to finish prepared transaction"),
593  errhint("Must be superuser or the user that prepared the transaction.")));
594 
595  /*
596  * Note: it probably would be possible to allow committing from
597  * another database; but at the moment NOTIFY is known not to work and
598  * there may be some other issues as well. Hence disallow until
599  * someone gets motivated to make it work.
600  */
601  if (MyDatabaseId != proc->databaseId)
602  ereport(ERROR,
603  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
604  errmsg("prepared transaction belongs to another database"),
605  errhint("Connect to the database where the transaction was prepared to finish it.")));
606 
607  /* OK for me to lock it */
608  gxact->locking_backend = MyBackendId;
609  MyLockedGxact = gxact;
610 
611  LWLockRelease(TwoPhaseStateLock);
612 
613  return gxact;
614  }
615 
616  LWLockRelease(TwoPhaseStateLock);
617 
618  ereport(ERROR,
619  (errcode(ERRCODE_UNDEFINED_OBJECT),
620  errmsg("prepared transaction with identifier \"%s\" does not exist",
621  gid)));
622 
623  /* NOTREACHED */
624  return NULL;
625 }
626 
627 /*
628  * RemoveGXact
629  * Remove the prepared transaction from the shared memory array.
630  *
631  * NB: caller should have already removed it from ProcArray
632  */
633 static void
635 {
636  int i;
637 
638  Assert(LWLockHeldByMeInMode(TwoPhaseStateLock, LW_EXCLUSIVE));
639 
640  for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
641  {
642  if (gxact == TwoPhaseState->prepXacts[i])
643  {
644  /* remove from the active array */
645  TwoPhaseState->numPrepXacts--;
646  TwoPhaseState->prepXacts[i] = TwoPhaseState->prepXacts[TwoPhaseState->numPrepXacts];
647 
648  /* and put it back in the freelist */
649  gxact->next = TwoPhaseState->freeGXacts;
650  TwoPhaseState->freeGXacts = gxact;
651 
652  return;
653  }
654  }
655 
656  elog(ERROR, "failed to find %p in GlobalTransaction array", gxact);
657 }
658 
659 /*
660  * Returns an array of all prepared transactions for the user-level
661  * function pg_prepared_xact.
662  *
663  * The returned array and all its elements are copies of internal data
664  * structures, to minimize the time we need to hold the TwoPhaseStateLock.
665  *
666  * WARNING -- we return even those transactions that are not fully prepared
667  * yet. The caller should filter them out if he doesn't want them.
668  *
669  * The returned array is palloc'd.
670  */
671 static int
673 {
674  GlobalTransaction array;
675  int num;
676  int i;
677 
678  LWLockAcquire(TwoPhaseStateLock, LW_SHARED);
679 
680  if (TwoPhaseState->numPrepXacts == 0)
681  {
682  LWLockRelease(TwoPhaseStateLock);
683 
684  *gxacts = NULL;
685  return 0;
686  }
687 
688  num = TwoPhaseState->numPrepXacts;
689  array = (GlobalTransaction) palloc(sizeof(GlobalTransactionData) * num);
690  *gxacts = array;
691  for (i = 0; i < num; i++)
692  memcpy(array + i, TwoPhaseState->prepXacts[i],
693  sizeof(GlobalTransactionData));
694 
695  LWLockRelease(TwoPhaseStateLock);
696 
697  return num;
698 }
699 
700 
701 /* Working status for pg_prepared_xact */
702 typedef struct
703 {
705  int ngxacts;
706  int currIdx;
707 } Working_State;
708 
709 /*
710  * pg_prepared_xact
711  * Produce a view with one row per prepared transaction.
712  *
713  * This function is here so we don't have to export the
714  * GlobalTransactionData struct definition.
715  */
716 Datum
718 {
719  FuncCallContext *funcctx;
721 
722  if (SRF_IS_FIRSTCALL())
723  {
724  TupleDesc tupdesc;
725  MemoryContext oldcontext;
726 
727  /* create a function context for cross-call persistence */
728  funcctx = SRF_FIRSTCALL_INIT();
729 
730  /*
731  * Switch to memory context appropriate for multiple function calls
732  */
733  oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
734 
735  /* build tupdesc for result tuples */
736  /* this had better match pg_prepared_xacts view in system_views.sql */
737  tupdesc = CreateTemplateTupleDesc(5);
738  TupleDescInitEntry(tupdesc, (AttrNumber) 1, "transaction",
739  XIDOID, -1, 0);
740  TupleDescInitEntry(tupdesc, (AttrNumber) 2, "gid",
741  TEXTOID, -1, 0);
742  TupleDescInitEntry(tupdesc, (AttrNumber) 3, "prepared",
743  TIMESTAMPTZOID, -1, 0);
744  TupleDescInitEntry(tupdesc, (AttrNumber) 4, "ownerid",
745  OIDOID, -1, 0);
746  TupleDescInitEntry(tupdesc, (AttrNumber) 5, "dbid",
747  OIDOID, -1, 0);
748 
749  funcctx->tuple_desc = BlessTupleDesc(tupdesc);
750 
751  /*
752  * Collect all the 2PC status information that we will format and send
753  * out as a result set.
754  */
755  status = (Working_State *) palloc(sizeof(Working_State));
756  funcctx->user_fctx = (void *) status;
757 
758  status->ngxacts = GetPreparedTransactionList(&status->array);
759  status->currIdx = 0;
760 
761  MemoryContextSwitchTo(oldcontext);
762  }
763 
764  funcctx = SRF_PERCALL_SETUP();
765  status = (Working_State *) funcctx->user_fctx;
766 
767  while (status->array != NULL && status->currIdx < status->ngxacts)
768  {
769  GlobalTransaction gxact = &status->array[status->currIdx++];
770  PGPROC *proc = &ProcGlobal->allProcs[gxact->pgprocno];
771  PGXACT *pgxact = &ProcGlobal->allPgXact[gxact->pgprocno];
772  Datum values[5];
773  bool nulls[5];
774  HeapTuple tuple;
775  Datum result;
776 
777  if (!gxact->valid)
778  continue;
779 
780  /*
781  * Form tuple with appropriate data.
782  */
783  MemSet(values, 0, sizeof(values));
784  MemSet(nulls, 0, sizeof(nulls));
785 
786  values[0] = TransactionIdGetDatum(pgxact->xid);
787  values[1] = CStringGetTextDatum(gxact->gid);
788  values[2] = TimestampTzGetDatum(gxact->prepared_at);
789  values[3] = ObjectIdGetDatum(gxact->owner);
790  values[4] = ObjectIdGetDatum(proc->databaseId);
791 
792  tuple = heap_form_tuple(funcctx->tuple_desc, values, nulls);
793  result = HeapTupleGetDatum(tuple);
794  SRF_RETURN_NEXT(funcctx, result);
795  }
796 
797  SRF_RETURN_DONE(funcctx);
798 }
799 
800 /*
801  * TwoPhaseGetGXact
802  * Get the GlobalTransaction struct for a prepared transaction
803  * specified by XID
804  *
805  * If lock_held is set to true, TwoPhaseStateLock will not be taken, so the
806  * caller had better hold it.
807  */
808 static GlobalTransaction
810 {
811  GlobalTransaction result = NULL;
812  int i;
813 
814  static TransactionId cached_xid = InvalidTransactionId;
815  static GlobalTransaction cached_gxact = NULL;
816 
817  Assert(!lock_held || LWLockHeldByMe(TwoPhaseStateLock));
818 
819  /*
820  * During a recovery, COMMIT PREPARED, or ABORT PREPARED, we'll be called
821  * repeatedly for the same XID. We can save work with a simple cache.
822  */
823  if (xid == cached_xid)
824  return cached_gxact;
825 
826  if (!lock_held)
827  LWLockAcquire(TwoPhaseStateLock, LW_SHARED);
828 
829  for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
830  {
831  GlobalTransaction gxact = TwoPhaseState->prepXacts[i];
832  PGXACT *pgxact = &ProcGlobal->allPgXact[gxact->pgprocno];
833 
834  if (pgxact->xid == xid)
835  {
836  result = gxact;
837  break;
838  }
839  }
840 
841  if (!lock_held)
842  LWLockRelease(TwoPhaseStateLock);
843 
844  if (result == NULL) /* should not happen */
845  elog(ERROR, "failed to find GlobalTransaction for xid %u", xid);
846 
847  cached_xid = xid;
848  cached_gxact = result;
849 
850  return result;
851 }
852 
853 /*
854  * TwoPhaseGetDummyBackendId
855  * Get the dummy backend ID for prepared transaction specified by XID
856  *
857  * Dummy backend IDs are similar to real backend IDs of real backends.
858  * They start at MaxBackends + 1, and are unique across all currently active
859  * real backends and prepared transactions. If lock_held is set to true,
860  * TwoPhaseStateLock will not be taken, so the caller had better hold it.
861  */
862 BackendId
864 {
865  GlobalTransaction gxact = TwoPhaseGetGXact(xid, lock_held);
866 
867  return gxact->dummyBackendId;
868 }
869 
870 /*
871  * TwoPhaseGetDummyProc
872  * Get the PGPROC that represents a prepared transaction specified by XID
873  *
874  * If lock_held is set to true, TwoPhaseStateLock will not be taken, so the
875  * caller had better hold it.
876  */
877 PGPROC *
879 {
880  GlobalTransaction gxact = TwoPhaseGetGXact(xid, lock_held);
881 
882  return &ProcGlobal->allProcs[gxact->pgprocno];
883 }
884 
885 /************************************************************************/
886 /* State file support */
887 /************************************************************************/
888 
889 #define TwoPhaseFilePath(path, xid) \
890  snprintf(path, MAXPGPATH, TWOPHASE_DIR "/%08X", xid)
891 
892 /*
893  * 2PC state file format:
894  *
895  * 1. TwoPhaseFileHeader
896  * 2. TransactionId[] (subtransactions)
897  * 3. RelFileNode[] (files to be deleted at commit)
898  * 4. RelFileNode[] (files to be deleted at abort)
899  * 5. SharedInvalidationMessage[] (inval messages to be sent at commit)
900  * 6. TwoPhaseRecordOnDisk
901  * 7. ...
902  * 8. TwoPhaseRecordOnDisk (end sentinel, rmid == TWOPHASE_RM_END_ID)
903  * 9. checksum (CRC-32C)
904  *
905  * Each segment except the final checksum is MAXALIGN'd.
906  */
907 
908 /*
909  * Header for a 2PC state file
910  */
911 #define TWOPHASE_MAGIC 0x57F94534 /* format identifier */
912 
914 
915 /*
916  * Header for each record in a state file
917  *
918  * NOTE: len counts only the rmgr data, not the TwoPhaseRecordOnDisk header.
919  * The rmgr data will be stored starting on a MAXALIGN boundary.
920  */
921 typedef struct TwoPhaseRecordOnDisk
922 {
923  uint32 len; /* length of rmgr data */
924  TwoPhaseRmgrId rmid; /* resource manager for this record */
925  uint16 info; /* flag bits for use by rmgr */
927 
928 /*
929  * During prepare, the state file is assembled in memory before writing it
930  * to WAL and the actual state file. We use a chain of StateFileChunk blocks
931  * for that.
932  */
933 typedef struct StateFileChunk
934 {
935  char *data;
939 
940 static struct xllist
941 {
942  StateFileChunk *head; /* first data block in the chain */
943  StateFileChunk *tail; /* last block in chain */
945  uint32 bytes_free; /* free bytes left in tail block */
946  uint32 total_len; /* total data bytes in chain */
947 } records;
948 
949 
950 /*
951  * Append a block of data to records data structure.
952  *
953  * NB: each block is padded to a MAXALIGN multiple. This must be
954  * accounted for when the file is later read!
955  *
956  * The data is copied, so the caller is free to modify it afterwards.
957  */
958 static void
959 save_state_data(const void *data, uint32 len)
960 {
961  uint32 padlen = MAXALIGN(len);
962 
963  if (padlen > records.bytes_free)
964  {
965  records.tail->next = palloc0(sizeof(StateFileChunk));
967  records.tail->len = 0;
968  records.tail->next = NULL;
970 
971  records.bytes_free = Max(padlen, 512);
973  }
974 
975  memcpy(((char *) records.tail->data) + records.tail->len, data, len);
976  records.tail->len += padlen;
977  records.bytes_free -= padlen;
978  records.total_len += padlen;
979 }
980 
981 /*
982  * Start preparing a state file.
983  *
984  * Initializes data structure and inserts the 2PC file header record.
985  */
986 void
988 {
989  PGPROC *proc = &ProcGlobal->allProcs[gxact->pgprocno];
990  PGXACT *pgxact = &ProcGlobal->allPgXact[gxact->pgprocno];
991  TransactionId xid = pgxact->xid;
992  TwoPhaseFileHeader hdr;
993  TransactionId *children;
994  RelFileNode *commitrels;
995  RelFileNode *abortrels;
996  SharedInvalidationMessage *invalmsgs;
997 
998  /* Initialize linked list */
999  records.head = palloc0(sizeof(StateFileChunk));
1000  records.head->len = 0;
1001  records.head->next = NULL;
1002 
1003  records.bytes_free = Max(sizeof(TwoPhaseFileHeader), 512);
1005 
1007  records.num_chunks = 1;
1008 
1009  records.total_len = 0;
1010 
1011  /* Create header */
1012  hdr.magic = TWOPHASE_MAGIC;
1013  hdr.total_len = 0; /* EndPrepare will fill this in */
1014  hdr.xid = xid;
1015  hdr.database = proc->databaseId;
1016  hdr.prepared_at = gxact->prepared_at;
1017  hdr.owner = gxact->owner;
1018  hdr.nsubxacts = xactGetCommittedChildren(&children);
1019  hdr.ncommitrels = smgrGetPendingDeletes(true, &commitrels);
1020  hdr.nabortrels = smgrGetPendingDeletes(false, &abortrels);
1022  &hdr.initfileinval);
1023  hdr.gidlen = strlen(gxact->gid) + 1; /* Include '\0' */
1024 
1025  save_state_data(&hdr, sizeof(TwoPhaseFileHeader));
1026  save_state_data(gxact->gid, hdr.gidlen);
1027 
1028  /*
1029  * Add the additional info about subxacts, deletable files and cache
1030  * invalidation messages.
1031  */
1032  if (hdr.nsubxacts > 0)
1033  {
1034  save_state_data(children, hdr.nsubxacts * sizeof(TransactionId));
1035  /* While we have the child-xact data, stuff it in the gxact too */
1036  GXactLoadSubxactData(gxact, hdr.nsubxacts, children);
1037  }
1038  if (hdr.ncommitrels > 0)
1039  {
1040  save_state_data(commitrels, hdr.ncommitrels * sizeof(RelFileNode));
1041  pfree(commitrels);
1042  }
1043  if (hdr.nabortrels > 0)
1044  {
1045  save_state_data(abortrels, hdr.nabortrels * sizeof(RelFileNode));
1046  pfree(abortrels);
1047  }
1048  if (hdr.ninvalmsgs > 0)
1049  {
1050  save_state_data(invalmsgs,
1051  hdr.ninvalmsgs * sizeof(SharedInvalidationMessage));
1052  pfree(invalmsgs);
1053  }
1054 }
1055 
1056 /*
1057  * Finish preparing state data and writing it to WAL.
1058  */
1059 void
1061 {
1062  TwoPhaseFileHeader *hdr;
1063  StateFileChunk *record;
1064  bool replorigin;
1065 
1066  /* Add the end sentinel to the list of 2PC records */
1068  NULL, 0);
1069 
1070  /* Go back and fill in total_len in the file header record */
1071  hdr = (TwoPhaseFileHeader *) records.head->data;
1072  Assert(hdr->magic == TWOPHASE_MAGIC);
1073  hdr->total_len = records.total_len + sizeof(pg_crc32c);
1074 
1075  replorigin = (replorigin_session_origin != InvalidRepOriginId &&
1077 
1078  if (replorigin)
1079  {
1083  }
1084  else
1085  {
1087  hdr->origin_timestamp = 0;
1088  }
1089 
1090  /*
1091  * If the data size exceeds MaxAllocSize, we won't be able to read it in
1092  * ReadTwoPhaseFile. Check for that now, rather than fail in the case
1093  * where we write data to file and then re-read at commit time.
1094  */
1095  if (hdr->total_len > MaxAllocSize)
1096  ereport(ERROR,
1097  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1098  errmsg("two-phase state file maximum length exceeded")));
1099 
1100  /*
1101  * Now writing 2PC state data to WAL. We let the WAL's CRC protection
1102  * cover us, so no need to calculate a separate CRC.
1103  *
1104  * We have to set delayChkpt here, too; otherwise a checkpoint starting
1105  * immediately after the WAL record is inserted could complete without
1106  * fsync'ing our state file. (This is essentially the same kind of race
1107  * condition as the COMMIT-to-clog-write case that RecordTransactionCommit
1108  * uses delayChkpt for; see notes there.)
1109  *
1110  * We save the PREPARE record's location in the gxact for later use by
1111  * CheckPointTwoPhase.
1112  */
1114 
1116 
1117  MyPgXact->delayChkpt = true;
1118 
1119  XLogBeginInsert();
1120  for (record = records.head; record != NULL; record = record->next)
1121  XLogRegisterData(record->data, record->len);
1122 
1124 
1125  gxact->prepare_end_lsn = XLogInsert(RM_XACT_ID, XLOG_XACT_PREPARE);
1126 
1127  if (replorigin)
1128  {
1129  /* Move LSNs forward for this replication origin */
1131  gxact->prepare_end_lsn);
1132  }
1133 
1134  XLogFlush(gxact->prepare_end_lsn);
1135 
1136  /* If we crash now, we have prepared: WAL replay will fix things */
1137 
1138  /* Store record's start location to read that later on Commit */
1140 
1141  /*
1142  * Mark the prepared transaction as valid. As soon as xact.c marks
1143  * MyPgXact as not running our XID (which it will do immediately after
1144  * this function returns), others can commit/rollback the xact.
1145  *
1146  * NB: a side effect of this is to make a dummy ProcArray entry for the
1147  * prepared XID. This must happen before we clear the XID from MyPgXact,
1148  * else there is a window where the XID is not running according to
1149  * TransactionIdIsInProgress, and onlookers would be entitled to assume
1150  * the xact crashed. Instead we have a window where the same XID appears
1151  * twice in ProcArray, which is OK.
1152  */
1153  MarkAsPrepared(gxact, false);
1154 
1155  /*
1156  * Now we can mark ourselves as out of the commit critical section: a
1157  * checkpoint starting after this will certainly see the gxact as a
1158  * candidate for fsyncing.
1159  */
1160  MyPgXact->delayChkpt = false;
1161 
1162  /*
1163  * Remember that we have this GlobalTransaction entry locked for us. If
1164  * we crash after this point, it's too late to abort, but we must unlock
1165  * it so that the prepared transaction can be committed or rolled back.
1166  */
1167  MyLockedGxact = gxact;
1168 
1169  END_CRIT_SECTION();
1170 
1171  /*
1172  * Wait for synchronous replication, if required.
1173  *
1174  * Note that at this stage we have marked the prepare, but still show as
1175  * running in the procarray (twice!) and continue to hold locks.
1176  */
1177  SyncRepWaitForLSN(gxact->prepare_end_lsn, false);
1178 
1179  records.tail = records.head = NULL;
1180  records.num_chunks = 0;
1181 }
1182 
1183 /*
1184  * Register a 2PC record to be written to state file.
1185  */
1186 void
1188  const void *data, uint32 len)
1189 {
1190  TwoPhaseRecordOnDisk record;
1191 
1192  record.rmid = rmid;
1193  record.info = info;
1194  record.len = len;
1195  save_state_data(&record, sizeof(TwoPhaseRecordOnDisk));
1196  if (len > 0)
1197  save_state_data(data, len);
1198 }
1199 
1200 
1201 /*
1202  * Read and validate the state file for xid.
1203  *
1204  * If it looks OK (has a valid magic number and CRC), return the palloc'd
1205  * contents of the file, issuing an error when finding corrupted data. If
1206  * missing_ok is true, which indicates that missing files can be safely
1207  * ignored, then return NULL. This state can be reached when doing recovery.
1208  */
1209 static char *
1211 {
1212  char path[MAXPGPATH];
1213  char *buf;
1214  TwoPhaseFileHeader *hdr;
1215  int fd;
1216  struct stat stat;
1217  uint32 crc_offset;
1218  pg_crc32c calc_crc,
1219  file_crc;
1220  int r;
1221 
1222  TwoPhaseFilePath(path, xid);
1223 
1224  fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
1225  if (fd < 0)
1226  {
1227  if (missing_ok && errno == ENOENT)
1228  return NULL;
1229 
1230  ereport(ERROR,
1232  errmsg("could not open file \"%s\": %m", path)));
1233  }
1234 
1235  /*
1236  * Check file length. We can determine a lower bound pretty easily. We
1237  * set an upper bound to avoid palloc() failure on a corrupt file, though
1238  * we can't guarantee that we won't get an out of memory error anyway,
1239  * even on a valid file.
1240  */
1241  if (fstat(fd, &stat))
1242  ereport(ERROR,
1244  errmsg("could not stat file \"%s\": %m", path)));
1245 
1246  if (stat.st_size < (MAXALIGN(sizeof(TwoPhaseFileHeader)) +
1247  MAXALIGN(sizeof(TwoPhaseRecordOnDisk)) +
1248  sizeof(pg_crc32c)) ||
1249  stat.st_size > MaxAllocSize)
1250  ereport(ERROR,
1252  errmsg_plural("incorrect size of file \"%s\": %zu byte",
1253  "incorrect size of file \"%s\": %zu bytes",
1254  (Size) stat.st_size, path,
1255  (Size) stat.st_size)));
1256 
1257  crc_offset = stat.st_size - sizeof(pg_crc32c);
1258  if (crc_offset != MAXALIGN(crc_offset))
1259  ereport(ERROR,
1261  errmsg("incorrect alignment of CRC offset for file \"%s\"",
1262  path)));
1263 
1264  /*
1265  * OK, slurp in the file.
1266  */
1267  buf = (char *) palloc(stat.st_size);
1268 
1270  r = read(fd, buf, stat.st_size);
1271  if (r != stat.st_size)
1272  {
1273  if (r < 0)
1274  ereport(ERROR,
1276  errmsg("could not read file \"%s\": %m", path)));
1277  else
1278  ereport(ERROR,
1279  (errmsg("could not read file \"%s\": read %d of %zu",
1280  path, r, (Size) stat.st_size)));
1281  }
1282 
1284 
1285  if (CloseTransientFile(fd) != 0)
1286  ereport(ERROR,
1288  errmsg("could not close file \"%s\": %m", path)));
1289 
1290  hdr = (TwoPhaseFileHeader *) buf;
1291  if (hdr->magic != TWOPHASE_MAGIC)
1292  ereport(ERROR,
1294  errmsg("invalid magic number stored in file \"%s\"",
1295  path)));
1296 
1297  if (hdr->total_len != stat.st_size)
1298  ereport(ERROR,
1300  errmsg("invalid size stored in file \"%s\"",
1301  path)));
1302 
1303  INIT_CRC32C(calc_crc);
1304  COMP_CRC32C(calc_crc, buf, crc_offset);
1305  FIN_CRC32C(calc_crc);
1306 
1307  file_crc = *((pg_crc32c *) (buf + crc_offset));
1308 
1309  if (!EQ_CRC32C(calc_crc, file_crc))
1310  ereport(ERROR,
1312  errmsg("calculated CRC checksum does not match value stored in file \"%s\"",
1313  path)));
1314 
1315  return buf;
1316 }
1317 
1318 
1319 /*
1320  * Reads 2PC data from xlog. During checkpoint this data will be moved to
1321  * twophase files and ReadTwoPhaseFile should be used instead.
1322  *
1323  * Note clearly that this function can access WAL during normal operation,
1324  * similarly to the way WALSender or Logical Decoding would do.
1325  */
1326 static void
1327 XlogReadTwoPhaseData(XLogRecPtr lsn, char **buf, int *len)
1328 {
1329  XLogRecord *record;
1330  XLogReaderState *xlogreader;
1331  char *errormsg;
1332 
1333  xlogreader = XLogReaderAllocate(wal_segment_size, NULL,
1334  &read_local_xlog_page, NULL);
1335  if (!xlogreader)
1336  ereport(ERROR,
1337  (errcode(ERRCODE_OUT_OF_MEMORY),
1338  errmsg("out of memory"),
1339  errdetail("Failed while allocating a WAL reading processor.")));
1340 
1341  XLogBeginRead(xlogreader, lsn);
1342  record = XLogReadRecord(xlogreader, &errormsg);
1343  if (record == NULL)
1344  ereport(ERROR,
1346  errmsg("could not read two-phase state from WAL at %X/%X",
1347  (uint32) (lsn >> 32),
1348  (uint32) lsn)));
1349 
1350  if (XLogRecGetRmid(xlogreader) != RM_XACT_ID ||
1352  ereport(ERROR,
1354  errmsg("expected two-phase state data is not present in WAL at %X/%X",
1355  (uint32) (lsn >> 32),
1356  (uint32) lsn)));
1357 
1358  if (len != NULL)
1359  *len = XLogRecGetDataLen(xlogreader);
1360 
1361  *buf = palloc(sizeof(char) * XLogRecGetDataLen(xlogreader));
1362  memcpy(*buf, XLogRecGetData(xlogreader), sizeof(char) * XLogRecGetDataLen(xlogreader));
1363 
1364  XLogReaderFree(xlogreader);
1365 }
1366 
1367 
1368 /*
1369  * Confirms an xid is prepared, during recovery
1370  */
1371 bool
1373 {
1374  char *buf;
1375  TwoPhaseFileHeader *hdr;
1376  bool result;
1377 
1379 
1380  if (max_prepared_xacts <= 0)
1381  return false; /* nothing to do */
1382 
1383  /* Read and validate file */
1384  buf = ReadTwoPhaseFile(xid, true);
1385  if (buf == NULL)
1386  return false;
1387 
1388  /* Check header also */
1389  hdr = (TwoPhaseFileHeader *) buf;
1390  result = TransactionIdEquals(hdr->xid, xid);
1391  pfree(buf);
1392 
1393  return result;
1394 }
1395 
1396 /*
1397  * FinishPreparedTransaction: execute COMMIT PREPARED or ROLLBACK PREPARED
1398  */
1399 void
1400 FinishPreparedTransaction(const char *gid, bool isCommit)
1401 {
1402  GlobalTransaction gxact;
1403  PGPROC *proc;
1404  PGXACT *pgxact;
1406  char *buf;
1407  char *bufptr;
1408  TwoPhaseFileHeader *hdr;
1409  TransactionId latestXid;
1410  TransactionId *children;
1411  RelFileNode *commitrels;
1412  RelFileNode *abortrels;
1413  RelFileNode *delrels;
1414  int ndelrels;
1415  SharedInvalidationMessage *invalmsgs;
1416 
1417  /*
1418  * Validate the GID, and lock the GXACT to ensure that two backends do not
1419  * try to commit the same GID at once.
1420  */
1421  gxact = LockGXact(gid, GetUserId());
1422  proc = &ProcGlobal->allProcs[gxact->pgprocno];
1423  pgxact = &ProcGlobal->allPgXact[gxact->pgprocno];
1424  xid = pgxact->xid;
1425 
1426  /*
1427  * Read and validate 2PC state data. State data will typically be stored
1428  * in WAL files if the LSN is after the last checkpoint record, or moved
1429  * to disk if for some reason they have lived for a long time.
1430  */
1431  if (gxact->ondisk)
1432  buf = ReadTwoPhaseFile(xid, false);
1433  else
1434  XlogReadTwoPhaseData(gxact->prepare_start_lsn, &buf, NULL);
1435 
1436 
1437  /*
1438  * Disassemble the header area
1439  */
1440  hdr = (TwoPhaseFileHeader *) buf;
1441  Assert(TransactionIdEquals(hdr->xid, xid));
1442  bufptr = buf + MAXALIGN(sizeof(TwoPhaseFileHeader));
1443  bufptr += MAXALIGN(hdr->gidlen);
1444  children = (TransactionId *) bufptr;
1445  bufptr += MAXALIGN(hdr->nsubxacts * sizeof(TransactionId));
1446  commitrels = (RelFileNode *) bufptr;
1447  bufptr += MAXALIGN(hdr->ncommitrels * sizeof(RelFileNode));
1448  abortrels = (RelFileNode *) bufptr;
1449  bufptr += MAXALIGN(hdr->nabortrels * sizeof(RelFileNode));
1450  invalmsgs = (SharedInvalidationMessage *) bufptr;
1451  bufptr += MAXALIGN(hdr->ninvalmsgs * sizeof(SharedInvalidationMessage));
1452 
1453  /* compute latestXid among all children */
1454  latestXid = TransactionIdLatest(xid, hdr->nsubxacts, children);
1455 
1456  /* Prevent cancel/die interrupt while cleaning up */
1457  HOLD_INTERRUPTS();
1458 
1459  /*
1460  * The order of operations here is critical: make the XLOG entry for
1461  * commit or abort, then mark the transaction committed or aborted in
1462  * pg_xact, then remove its PGPROC from the global ProcArray (which means
1463  * TransactionIdIsInProgress will stop saying the prepared xact is in
1464  * progress), then run the post-commit or post-abort callbacks. The
1465  * callbacks will release the locks the transaction held.
1466  */
1467  if (isCommit)
1469  hdr->nsubxacts, children,
1470  hdr->ncommitrels, commitrels,
1471  hdr->ninvalmsgs, invalmsgs,
1472  hdr->initfileinval, gid);
1473  else
1475  hdr->nsubxacts, children,
1476  hdr->nabortrels, abortrels,
1477  gid);
1478 
1479  ProcArrayRemove(proc, latestXid);
1480 
1481  /*
1482  * In case we fail while running the callbacks, mark the gxact invalid so
1483  * no one else will try to commit/rollback, and so it will be recycled if
1484  * we fail after this point. It is still locked by our backend so it
1485  * won't go away yet.
1486  *
1487  * (We assume it's safe to do this without taking TwoPhaseStateLock.)
1488  */
1489  gxact->valid = false;
1490 
1491  /*
1492  * We have to remove any files that were supposed to be dropped. For
1493  * consistency with the regular xact.c code paths, must do this before
1494  * releasing locks, so do it before running the callbacks.
1495  *
1496  * NB: this code knows that we couldn't be dropping any temp rels ...
1497  */
1498  if (isCommit)
1499  {
1500  delrels = commitrels;
1501  ndelrels = hdr->ncommitrels;
1502  }
1503  else
1504  {
1505  delrels = abortrels;
1506  ndelrels = hdr->nabortrels;
1507  }
1508 
1509  /* Make sure files supposed to be dropped are dropped */
1510  DropRelationFiles(delrels, ndelrels, false);
1511 
1512  /*
1513  * Handle cache invalidation messages.
1514  *
1515  * Relcache init file invalidation requires processing both before and
1516  * after we send the SI messages. See AtEOXact_Inval()
1517  */
1518  if (hdr->initfileinval)
1520  SendSharedInvalidMessages(invalmsgs, hdr->ninvalmsgs);
1521  if (hdr->initfileinval)
1523 
1524  /*
1525  * Acquire the two-phase lock. We want to work on the two-phase callbacks
1526  * while holding it to avoid potential conflicts with other transactions
1527  * attempting to use the same GID, so the lock is released once the shared
1528  * memory state is cleared.
1529  */
1530  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
1531 
1532  /* And now do the callbacks */
1533  if (isCommit)
1535  else
1537 
1538  PredicateLockTwoPhaseFinish(xid, isCommit);
1539 
1540  /* Clear shared memory state */
1541  RemoveGXact(gxact);
1542 
1543  /*
1544  * Release the lock as all callbacks are called and shared memory cleanup
1545  * is done.
1546  */
1547  LWLockRelease(TwoPhaseStateLock);
1548 
1549  /* Count the prepared xact as committed or aborted */
1550  AtEOXact_PgStat(isCommit, false);
1551 
1552  /*
1553  * And now we can clean up any files we may have left.
1554  */
1555  if (gxact->ondisk)
1556  RemoveTwoPhaseFile(xid, true);
1557 
1558  MyLockedGxact = NULL;
1559 
1561 
1562  pfree(buf);
1563 }
1564 
1565 /*
1566  * Scan 2PC state data in memory and call the indicated callbacks for each 2PC record.
1567  */
1568 static void
1570  const TwoPhaseCallback callbacks[])
1571 {
1572  for (;;)
1573  {
1574  TwoPhaseRecordOnDisk *record = (TwoPhaseRecordOnDisk *) bufptr;
1575 
1576  Assert(record->rmid <= TWOPHASE_RM_MAX_ID);
1577  if (record->rmid == TWOPHASE_RM_END_ID)
1578  break;
1579 
1580  bufptr += MAXALIGN(sizeof(TwoPhaseRecordOnDisk));
1581 
1582  if (callbacks[record->rmid] != NULL)
1583  callbacks[record->rmid] (xid, record->info,
1584  (void *) bufptr, record->len);
1585 
1586  bufptr += MAXALIGN(record->len);
1587  }
1588 }
1589 
1590 /*
1591  * Remove the 2PC file for the specified XID.
1592  *
1593  * If giveWarning is false, do not complain about file-not-present;
1594  * this is an expected case during WAL replay.
1595  */
1596 static void
1598 {
1599  char path[MAXPGPATH];
1600 
1601  TwoPhaseFilePath(path, xid);
1602  if (unlink(path))
1603  if (errno != ENOENT || giveWarning)
1604  ereport(WARNING,
1606  errmsg("could not remove file \"%s\": %m", path)));
1607 }
1608 
1609 /*
1610  * Recreates a state file. This is used in WAL replay and during
1611  * checkpoint creation.
1612  *
1613  * Note: content and len don't include CRC.
1614  */
1615 static void
1616 RecreateTwoPhaseFile(TransactionId xid, void *content, int len)
1617 {
1618  char path[MAXPGPATH];
1619  pg_crc32c statefile_crc;
1620  int fd;
1621 
1622  /* Recompute CRC */
1623  INIT_CRC32C(statefile_crc);
1624  COMP_CRC32C(statefile_crc, content, len);
1625  FIN_CRC32C(statefile_crc);
1626 
1627  TwoPhaseFilePath(path, xid);
1628 
1629  fd = OpenTransientFile(path,
1630  O_CREAT | O_TRUNC | O_WRONLY | PG_BINARY);
1631  if (fd < 0)
1632  ereport(ERROR,
1634  errmsg("could not recreate file \"%s\": %m", path)));
1635 
1636  /* Write content and CRC */
1637  errno = 0;
1639  if (write(fd, content, len) != len)
1640  {
1641  /* if write didn't set errno, assume problem is no disk space */
1642  if (errno == 0)
1643  errno = ENOSPC;
1644  ereport(ERROR,
1646  errmsg("could not write file \"%s\": %m", path)));
1647  }
1648  if (write(fd, &statefile_crc, sizeof(pg_crc32c)) != sizeof(pg_crc32c))
1649  {
1650  /* if write didn't set errno, assume problem is no disk space */
1651  if (errno == 0)
1652  errno = ENOSPC;
1653  ereport(ERROR,
1655  errmsg("could not write file \"%s\": %m", path)));
1656  }
1658 
1659  /*
1660  * We must fsync the file because the end-of-replay checkpoint will not do
1661  * so, there being no GXACT in shared memory yet to tell it to.
1662  */
1664  if (pg_fsync(fd) != 0)
1665  ereport(ERROR,
1667  errmsg("could not fsync file \"%s\": %m", path)));
1669 
1670  if (CloseTransientFile(fd) != 0)
1671  ereport(ERROR,
1673  errmsg("could not close file \"%s\": %m", path)));
1674 }
1675 
1676 /*
1677  * CheckPointTwoPhase -- handle 2PC component of checkpointing.
1678  *
1679  * We must fsync the state file of any GXACT that is valid or has been
1680  * generated during redo and has a PREPARE LSN <= the checkpoint's redo
1681  * horizon. (If the gxact isn't valid yet, has not been generated in
1682  * redo, or has a later LSN, this checkpoint is not responsible for
1683  * fsyncing it.)
1684  *
1685  * This is deliberately run as late as possible in the checkpoint sequence,
1686  * because GXACTs ordinarily have short lifespans, and so it is quite
1687  * possible that GXACTs that were valid at checkpoint start will no longer
1688  * exist if we wait a little bit. With typical checkpoint settings this
1689  * will be about 3 minutes for an online checkpoint, so as a result we
1690  * expect that there will be no GXACTs that need to be copied to disk.
1691  *
1692  * If a GXACT remains valid across multiple checkpoints, it will already
1693  * be on disk so we don't bother to repeat that write.
1694  */
1695 void
1697 {
1698  int i;
1699  int serialized_xacts = 0;
1700 
1701  if (max_prepared_xacts <= 0)
1702  return; /* nothing to do */
1703 
1704  TRACE_POSTGRESQL_TWOPHASE_CHECKPOINT_START();
1705 
1706  /*
1707  * We are expecting there to be zero GXACTs that need to be copied to
1708  * disk, so we perform all I/O while holding TwoPhaseStateLock for
1709  * simplicity. This prevents any new xacts from preparing while this
1710  * occurs, which shouldn't be a problem since the presence of long-lived
1711  * prepared xacts indicates the transaction manager isn't active.
1712  *
1713  * It's also possible to move I/O out of the lock, but on every error we
1714  * should check whether somebody committed our transaction in different
1715  * backend. Let's leave this optimization for future, if somebody will
1716  * spot that this place cause bottleneck.
1717  *
1718  * Note that it isn't possible for there to be a GXACT with a
1719  * prepare_end_lsn set prior to the last checkpoint yet is marked invalid,
1720  * because of the efforts with delayChkpt.
1721  */
1722  LWLockAcquire(TwoPhaseStateLock, LW_SHARED);
1723  for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
1724  {
1725  /*
1726  * Note that we are using gxact not pgxact so this works in recovery
1727  * also
1728  */
1729  GlobalTransaction gxact = TwoPhaseState->prepXacts[i];
1730 
1731  if ((gxact->valid || gxact->inredo) &&
1732  !gxact->ondisk &&
1733  gxact->prepare_end_lsn <= redo_horizon)
1734  {
1735  char *buf;
1736  int len;
1737 
1738  XlogReadTwoPhaseData(gxact->prepare_start_lsn, &buf, &len);
1739  RecreateTwoPhaseFile(gxact->xid, buf, len);
1740  gxact->ondisk = true;
1743  pfree(buf);
1744  serialized_xacts++;
1745  }
1746  }
1747  LWLockRelease(TwoPhaseStateLock);
1748 
1749  /*
1750  * Flush unconditionally the parent directory to make any information
1751  * durable on disk. Two-phase files could have been removed and those
1752  * removals need to be made persistent as well as any files newly created
1753  * previously since the last checkpoint.
1754  */
1755  fsync_fname(TWOPHASE_DIR, true);
1756 
1757  TRACE_POSTGRESQL_TWOPHASE_CHECKPOINT_DONE();
1758 
1759  if (log_checkpoints && serialized_xacts > 0)
1760  ereport(LOG,
1761  (errmsg_plural("%u two-phase state file was written "
1762  "for a long-running prepared transaction",
1763  "%u two-phase state files were written "
1764  "for long-running prepared transactions",
1765  serialized_xacts,
1766  serialized_xacts)));
1767 }
1768 
1769 /*
1770  * restoreTwoPhaseData
1771  *
1772  * Scan pg_twophase and fill TwoPhaseState depending on the on-disk data.
1773  * This is called once at the beginning of recovery, saving any extra
1774  * lookups in the future. Two-phase files that are newer than the
1775  * minimum XID horizon are discarded on the way.
1776  */
1777 void
1779 {
1780  DIR *cldir;
1781  struct dirent *clde;
1782 
1783  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
1784  cldir = AllocateDir(TWOPHASE_DIR);
1785  while ((clde = ReadDir(cldir, TWOPHASE_DIR)) != NULL)
1786  {
1787  if (strlen(clde->d_name) == 8 &&
1788  strspn(clde->d_name, "0123456789ABCDEF") == 8)
1789  {
1791  char *buf;
1792 
1793  xid = (TransactionId) strtoul(clde->d_name, NULL, 16);
1794 
1796  true, false, false);
1797  if (buf == NULL)
1798  continue;
1799 
1802  }
1803  }
1804  LWLockRelease(TwoPhaseStateLock);
1805  FreeDir(cldir);
1806 }
1807 
1808 /*
1809  * PrescanPreparedTransactions
1810  *
1811  * Scan the shared memory entries of TwoPhaseState and determine the range
1812  * of valid XIDs present. This is run during database startup, after we
1813  * have completed reading WAL. ShmemVariableCache->nextFullXid has been set to
1814  * one more than the highest XID for which evidence exists in WAL.
1815  *
1816  * We throw away any prepared xacts with main XID beyond nextFullXid --- if any
1817  * are present, it suggests that the DBA has done a PITR recovery to an
1818  * earlier point in time without cleaning out pg_twophase. We dare not
1819  * try to recover such prepared xacts since they likely depend on database
1820  * state that doesn't exist now.
1821  *
1822  * However, we will advance nextFullXid beyond any subxact XIDs belonging to
1823  * valid prepared xacts. We need to do this since subxact commit doesn't
1824  * write a WAL entry, and so there might be no evidence in WAL of those
1825  * subxact XIDs.
1826  *
1827  * On corrupted two-phase files, fail immediately. Keeping around broken
1828  * entries and let replay continue causes harm on the system, and a new
1829  * backup should be rolled in.
1830  *
1831  * Our other responsibility is to determine and return the oldest valid XID
1832  * among the prepared xacts (if none, return ShmemVariableCache->nextFullXid).
1833  * This is needed to synchronize pg_subtrans startup properly.
1834  *
1835  * If xids_p and nxids_p are not NULL, pointer to a palloc'd array of all
1836  * top-level xids is stored in *xids_p. The number of entries in the array
1837  * is returned in *nxids_p.
1838  */
1841 {
1843  TransactionId origNextXid = XidFromFullTransactionId(nextFullXid);
1844  TransactionId result = origNextXid;
1845  TransactionId *xids = NULL;
1846  int nxids = 0;
1847  int allocsize = 0;
1848  int i;
1849 
1850  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
1851  for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
1852  {
1854  char *buf;
1855  GlobalTransaction gxact = TwoPhaseState->prepXacts[i];
1856 
1857  Assert(gxact->inredo);
1858 
1859  xid = gxact->xid;
1860 
1861  buf = ProcessTwoPhaseBuffer(xid,
1862  gxact->prepare_start_lsn,
1863  gxact->ondisk, false, true);
1864 
1865  if (buf == NULL)
1866  continue;
1867 
1868  /*
1869  * OK, we think this file is valid. Incorporate xid into the
1870  * running-minimum result.
1871  */
1872  if (TransactionIdPrecedes(xid, result))
1873  result = xid;
1874 
1875  if (xids_p)
1876  {
1877  if (nxids == allocsize)
1878  {
1879  if (nxids == 0)
1880  {
1881  allocsize = 10;
1882  xids = palloc(allocsize * sizeof(TransactionId));
1883  }
1884  else
1885  {
1886  allocsize = allocsize * 2;
1887  xids = repalloc(xids, allocsize * sizeof(TransactionId));
1888  }
1889  }
1890  xids[nxids++] = xid;
1891  }
1892 
1893  pfree(buf);
1894  }
1895  LWLockRelease(TwoPhaseStateLock);
1896 
1897  if (xids_p)
1898  {
1899  *xids_p = xids;
1900  *nxids_p = nxids;
1901  }
1902 
1903  return result;
1904 }
1905 
1906 /*
1907  * StandbyRecoverPreparedTransactions
1908  *
1909  * Scan the shared memory entries of TwoPhaseState and setup all the required
1910  * information to allow standby queries to treat prepared transactions as still
1911  * active.
1912  *
1913  * This is never called at the end of recovery - we use
1914  * RecoverPreparedTransactions() at that point.
1915  *
1916  * The lack of calls to SubTransSetParent() calls here is by design;
1917  * those calls are made by RecoverPreparedTransactions() at the end of recovery
1918  * for those xacts that need this.
1919  */
1920 void
1922 {
1923  int i;
1924 
1925  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
1926  for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
1927  {
1929  char *buf;
1930  GlobalTransaction gxact = TwoPhaseState->prepXacts[i];
1931 
1932  Assert(gxact->inredo);
1933 
1934  xid = gxact->xid;
1935 
1936  buf = ProcessTwoPhaseBuffer(xid,
1937  gxact->prepare_start_lsn,
1938  gxact->ondisk, false, false);
1939  if (buf != NULL)
1940  pfree(buf);
1941  }
1942  LWLockRelease(TwoPhaseStateLock);
1943 }
1944 
1945 /*
1946  * RecoverPreparedTransactions
1947  *
1948  * Scan the shared memory entries of TwoPhaseState and reload the state for
1949  * each prepared transaction (reacquire locks, etc).
1950  *
1951  * This is run at the end of recovery, but before we allow backends to write
1952  * WAL.
1953  *
1954  * At the end of recovery the way we take snapshots will change. We now need
1955  * to mark all running transactions with their full SubTransSetParent() info
1956  * to allow normal snapshots to work correctly if snapshots overflow.
1957  * We do this here because by definition prepared transactions are the only
1958  * type of write transaction still running, so this is necessary and
1959  * complete.
1960  */
1961 void
1963 {
1964  int i;
1965 
1966  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
1967  for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
1968  {
1970  char *buf;
1971  GlobalTransaction gxact = TwoPhaseState->prepXacts[i];
1972  char *bufptr;
1973  TwoPhaseFileHeader *hdr;
1974  TransactionId *subxids;
1975  const char *gid;
1976 
1977  xid = gxact->xid;
1978 
1979  /*
1980  * Reconstruct subtrans state for the transaction --- needed because
1981  * pg_subtrans is not preserved over a restart. Note that we are
1982  * linking all the subtransactions directly to the top-level XID;
1983  * there may originally have been a more complex hierarchy, but
1984  * there's no need to restore that exactly. It's possible that
1985  * SubTransSetParent has been set before, if the prepared transaction
1986  * generated xid assignment records.
1987  */
1988  buf = ProcessTwoPhaseBuffer(xid,
1989  gxact->prepare_start_lsn,
1990  gxact->ondisk, true, false);
1991  if (buf == NULL)
1992  continue;
1993 
1994  ereport(LOG,
1995  (errmsg("recovering prepared transaction %u from shared memory", xid)));
1996 
1997  hdr = (TwoPhaseFileHeader *) buf;
1998  Assert(TransactionIdEquals(hdr->xid, xid));
1999  bufptr = buf + MAXALIGN(sizeof(TwoPhaseFileHeader));
2000  gid = (const char *) bufptr;
2001  bufptr += MAXALIGN(hdr->gidlen);
2002  subxids = (TransactionId *) bufptr;
2003  bufptr += MAXALIGN(hdr->nsubxacts * sizeof(TransactionId));
2004  bufptr += MAXALIGN(hdr->ncommitrels * sizeof(RelFileNode));
2005  bufptr += MAXALIGN(hdr->nabortrels * sizeof(RelFileNode));
2006  bufptr += MAXALIGN(hdr->ninvalmsgs * sizeof(SharedInvalidationMessage));
2007 
2008  /*
2009  * Recreate its GXACT and dummy PGPROC. But, check whether it was
2010  * added in redo and already has a shmem entry for it.
2011  */
2012  MarkAsPreparingGuts(gxact, xid, gid,
2013  hdr->prepared_at,
2014  hdr->owner, hdr->database);
2015 
2016  /* recovered, so reset the flag for entries generated by redo */
2017  gxact->inredo = false;
2018 
2019  GXactLoadSubxactData(gxact, hdr->nsubxacts, subxids);
2020  MarkAsPrepared(gxact, true);
2021 
2022  LWLockRelease(TwoPhaseStateLock);
2023 
2024  /*
2025  * Recover other state (notably locks) using resource managers.
2026  */
2028 
2029  /*
2030  * Release locks held by the standby process after we process each
2031  * prepared transaction. As a result, we don't need too many
2032  * additional locks at any one time.
2033  */
2034  if (InHotStandby)
2035  StandbyReleaseLockTree(xid, hdr->nsubxacts, subxids);
2036 
2037  /*
2038  * We're done with recovering this transaction. Clear MyLockedGxact,
2039  * like we do in PrepareTransaction() during normal operation.
2040  */
2042 
2043  pfree(buf);
2044 
2045  LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
2046  }
2047 
2048  LWLockRelease(TwoPhaseStateLock);
2049 }
2050 
2051 /*
2052  * ProcessTwoPhaseBuffer
2053  *
2054  * Given a transaction id, read it either from disk or read it directly
2055  * via shmem xlog record pointer using the provided "prepare_start_lsn".
2056  *
2057  * If setParent is true, set up subtransaction parent linkages.
2058  *
2059  * If setNextXid is true, set ShmemVariableCache->nextFullXid to the newest
2060  * value scanned.
2061  */
2062 static char *
2065  bool fromdisk,
2066  bool setParent, bool setNextXid)
2067 {
2069  TransactionId origNextXid = XidFromFullTransactionId(nextFullXid);
2070  TransactionId *subxids;
2071  char *buf;
2072  TwoPhaseFileHeader *hdr;
2073  int i;
2074 
2075  Assert(LWLockHeldByMeInMode(TwoPhaseStateLock, LW_EXCLUSIVE));
2076 
2077  if (!fromdisk)
2078  Assert(prepare_start_lsn != InvalidXLogRecPtr);
2079 
2080  /* Already processed? */
2082  {
2083  if (fromdisk)
2084  {
2085  ereport(WARNING,
2086  (errmsg("removing stale two-phase state file for transaction %u",
2087  xid)));
2088  RemoveTwoPhaseFile(xid, true);
2089  }
2090  else
2091  {
2092  ereport(WARNING,
2093  (errmsg("removing stale two-phase state from memory for transaction %u",
2094  xid)));
2095  PrepareRedoRemove(xid, true);
2096  }
2097  return NULL;
2098  }
2099 
2100  /* Reject XID if too new */
2101  if (TransactionIdFollowsOrEquals(xid, origNextXid))
2102  {
2103  if (fromdisk)
2104  {
2105  ereport(WARNING,
2106  (errmsg("removing future two-phase state file for transaction %u",
2107  xid)));
2108  RemoveTwoPhaseFile(xid, true);
2109  }
2110  else
2111  {
2112  ereport(WARNING,
2113  (errmsg("removing future two-phase state from memory for transaction %u",
2114  xid)));
2115  PrepareRedoRemove(xid, true);
2116  }
2117  return NULL;
2118  }
2119 
2120  if (fromdisk)
2121  {
2122  /* Read and validate file */
2123  buf = ReadTwoPhaseFile(xid, false);
2124  }
2125  else
2126  {
2127  /* Read xlog data */
2128  XlogReadTwoPhaseData(prepare_start_lsn, &buf, NULL);
2129  }
2130 
2131  /* Deconstruct header */
2132  hdr = (TwoPhaseFileHeader *) buf;
2133  if (!TransactionIdEquals(hdr->xid, xid))
2134  {
2135  if (fromdisk)
2136  ereport(ERROR,
2138  errmsg("corrupted two-phase state file for transaction %u",
2139  xid)));
2140  else
2141  ereport(ERROR,
2143  errmsg("corrupted two-phase state in memory for transaction %u",
2144  xid)));
2145  }
2146 
2147  /*
2148  * Examine subtransaction XIDs ... they should all follow main XID, and
2149  * they may force us to advance nextFullXid.
2150  */
2151  subxids = (TransactionId *) (buf +
2152  MAXALIGN(sizeof(TwoPhaseFileHeader)) +
2153  MAXALIGN(hdr->gidlen));
2154  for (i = 0; i < hdr->nsubxacts; i++)
2155  {
2156  TransactionId subxid = subxids[i];
2157 
2158  Assert(TransactionIdFollows(subxid, xid));
2159 
2160  /* update nextFullXid if needed */
2161  if (setNextXid)
2163 
2164  if (setParent)
2165  SubTransSetParent(subxid, xid);
2166  }
2167 
2168  return buf;
2169 }
2170 
2171 
2172 /*
2173  * RecordTransactionCommitPrepared
2174  *
2175  * This is basically the same as RecordTransactionCommit (q.v. if you change
2176  * this function): in particular, we must set the delayChkpt flag to avoid a
2177  * race condition.
2178  *
2179  * We know the transaction made at least one XLOG entry (its PREPARE),
2180  * so it is never possible to optimize out the commit record.
2181  */
2182 static void
2184  int nchildren,
2185  TransactionId *children,
2186  int nrels,
2187  RelFileNode *rels,
2188  int ninvalmsgs,
2189  SharedInvalidationMessage *invalmsgs,
2190  bool initfileinval,
2191  const char *gid)
2192 {
2193  XLogRecPtr recptr;
2194  TimestampTz committs = GetCurrentTimestamp();
2195  bool replorigin;
2196 
2197  /*
2198  * Are we using the replication origins feature? Or, in other words, are
2199  * we replaying remote actions?
2200  */
2201  replorigin = (replorigin_session_origin != InvalidRepOriginId &&
2203 
2205 
2206  /* See notes in RecordTransactionCommit */
2207  MyPgXact->delayChkpt = true;
2208 
2209  /*
2210  * Emit the XLOG commit record. Note that we mark 2PC commits as
2211  * potentially having AccessExclusiveLocks since we don't know whether or
2212  * not they do.
2213  */
2214  recptr = XactLogCommitRecord(committs,
2215  nchildren, children, nrels, rels,
2216  ninvalmsgs, invalmsgs,
2217  initfileinval, false,
2219  xid, gid);
2220 
2221 
2222  if (replorigin)
2223  /* Move LSNs forward for this replication origin */
2225  XactLastRecEnd);
2226 
2227  /*
2228  * Record commit timestamp. The value comes from plain commit timestamp
2229  * if replorigin is not enabled, or replorigin already set a value for us
2230  * in replorigin_session_origin_timestamp otherwise.
2231  *
2232  * We don't need to WAL-log anything here, as the commit record written
2233  * above already contains the data.
2234  */
2235  if (!replorigin || replorigin_session_origin_timestamp == 0)
2237 
2238  TransactionTreeSetCommitTsData(xid, nchildren, children,
2240  replorigin_session_origin, false);
2241 
2242  /*
2243  * We don't currently try to sleep before flush here ... nor is there any
2244  * support for async commit of a prepared xact (the very idea is probably
2245  * a contradiction)
2246  */
2247 
2248  /* Flush XLOG to disk */
2249  XLogFlush(recptr);
2250 
2251  /* Mark the transaction committed in pg_xact */
2252  TransactionIdCommitTree(xid, nchildren, children);
2253 
2254  /* Checkpoint can proceed now */
2255  MyPgXact->delayChkpt = false;
2256 
2257  END_CRIT_SECTION();
2258 
2259  /*
2260  * Wait for synchronous replication, if required.
2261  *
2262  * Note that at this stage we have marked clog, but still show as running
2263  * in the procarray and continue to hold locks.
2264  */
2265  SyncRepWaitForLSN(recptr, true);
2266 }
2267 
2268 /*
2269  * RecordTransactionAbortPrepared
2270  *
2271  * This is basically the same as RecordTransactionAbort.
2272  *
2273  * We know the transaction made at least one XLOG entry (its PREPARE),
2274  * so it is never possible to optimize out the abort record.
2275  */
2276 static void
2278  int nchildren,
2279  TransactionId *children,
2280  int nrels,
2281  RelFileNode *rels,
2282  const char *gid)
2283 {
2284  XLogRecPtr recptr;
2285 
2286  /*
2287  * Catch the scenario where we aborted partway through
2288  * RecordTransactionCommitPrepared ...
2289  */
2290  if (TransactionIdDidCommit(xid))
2291  elog(PANIC, "cannot abort transaction %u, it was already committed",
2292  xid);
2293 
2295 
2296  /*
2297  * Emit the XLOG commit record. Note that we mark 2PC aborts as
2298  * potentially having AccessExclusiveLocks since we don't know whether or
2299  * not they do.
2300  */
2302  nchildren, children,
2303  nrels, rels,
2305  xid, gid);
2306 
2307  /* Always flush, since we're about to remove the 2PC state file */
2308  XLogFlush(recptr);
2309 
2310  /*
2311  * Mark the transaction aborted in clog. This is not absolutely necessary
2312  * but we may as well do it while we are here.
2313  */
2314  TransactionIdAbortTree(xid, nchildren, children);
2315 
2316  END_CRIT_SECTION();
2317 
2318  /*
2319  * Wait for synchronous replication, if required.
2320  *
2321  * Note that at this stage we have marked clog, but still show as running
2322  * in the procarray and continue to hold locks.
2323  */
2324  SyncRepWaitForLSN(recptr, false);
2325 }
2326 
2327 /*
2328  * PrepareRedoAdd
2329  *
2330  * Store pointers to the start/end of the WAL record along with the xid in
2331  * a gxact entry in shared memory TwoPhaseState structure. If caller
2332  * specifies InvalidXLogRecPtr as WAL location to fetch the two-phase
2333  * data, the entry is marked as located on disk.
2334  */
2335 void
2336 PrepareRedoAdd(char *buf, XLogRecPtr start_lsn,
2337  XLogRecPtr end_lsn, RepOriginId origin_id)
2338 {
2339  TwoPhaseFileHeader *hdr = (TwoPhaseFileHeader *) buf;
2340  char *bufptr;
2341  const char *gid;
2342  GlobalTransaction gxact;
2343 
2344  Assert(LWLockHeldByMeInMode(TwoPhaseStateLock, LW_EXCLUSIVE));
2346 
2347  bufptr = buf + MAXALIGN(sizeof(TwoPhaseFileHeader));
2348  gid = (const char *) bufptr;
2349 
2350  /*
2351  * Reserve the GID for the given transaction in the redo code path.
2352  *
2353  * This creates a gxact struct and puts it into the active array.
2354  *
2355  * In redo, this struct is mainly used to track PREPARE/COMMIT entries in
2356  * shared memory. Hence, we only fill up the bare minimum contents here.
2357  * The gxact also gets marked with gxact->inredo set to true to indicate
2358  * that it got added in the redo phase
2359  */
2360 
2361  /* Get a free gxact from the freelist */
2362  if (TwoPhaseState->freeGXacts == NULL)
2363  ereport(ERROR,
2364  (errcode(ERRCODE_OUT_OF_MEMORY),
2365  errmsg("maximum number of prepared transactions reached"),
2366  errhint("Increase max_prepared_transactions (currently %d).",
2367  max_prepared_xacts)));
2368  gxact = TwoPhaseState->freeGXacts;
2369  TwoPhaseState->freeGXacts = gxact->next;
2370 
2371  gxact->prepared_at = hdr->prepared_at;
2372  gxact->prepare_start_lsn = start_lsn;
2373  gxact->prepare_end_lsn = end_lsn;
2374  gxact->xid = hdr->xid;
2375  gxact->owner = hdr->owner;
2377  gxact->valid = false;
2378  gxact->ondisk = XLogRecPtrIsInvalid(start_lsn);
2379  gxact->inredo = true; /* yes, added in redo */
2380  strcpy(gxact->gid, gid);
2381 
2382  /* And insert it into the active array */
2383  Assert(TwoPhaseState->numPrepXacts < max_prepared_xacts);
2384  TwoPhaseState->prepXacts[TwoPhaseState->numPrepXacts++] = gxact;
2385 
2386  if (origin_id != InvalidRepOriginId)
2387  {
2388  /* recover apply progress */
2389  replorigin_advance(origin_id, hdr->origin_lsn, end_lsn,
2390  false /* backward */ , false /* WAL */ );
2391  }
2392 
2393  elog(DEBUG2, "added 2PC data in shared memory for transaction %u", gxact->xid);
2394 }
2395 
2396 /*
2397  * PrepareRedoRemove
2398  *
2399  * Remove the corresponding gxact entry from TwoPhaseState. Also remove
2400  * the 2PC file if a prepared transaction was saved via an earlier checkpoint.
2401  *
2402  * Caller must hold TwoPhaseStateLock in exclusive mode, because TwoPhaseState
2403  * is updated.
2404  */
2405 void
2407 {
2408  GlobalTransaction gxact = NULL;
2409  int i;
2410  bool found = false;
2411 
2412  Assert(LWLockHeldByMeInMode(TwoPhaseStateLock, LW_EXCLUSIVE));
2414 
2415  for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
2416  {
2417  gxact = TwoPhaseState->prepXacts[i];
2418 
2419  if (gxact->xid == xid)
2420  {
2421  Assert(gxact->inredo);
2422  found = true;
2423  break;
2424  }
2425  }
2426 
2427  /*
2428  * Just leave if there is nothing, this is expected during WAL replay.
2429  */
2430  if (!found)
2431  return;
2432 
2433  /*
2434  * And now we can clean up any files we may have left.
2435  */
2436  elog(DEBUG2, "removing 2PC data for transaction %u", xid);
2437  if (gxact->ondisk)
2438  RemoveTwoPhaseFile(xid, giveWarning);
2439  RemoveGXact(gxact);
2440 }
int32 ninvalmsgs
Definition: xact.h:306
void TransactionIdAbortTree(TransactionId xid, int nxids, TransactionId *xids)
Definition: transam.c:290
#define INIT_CRC32C(crc)
Definition: pg_crc32c.h:41
int xactGetCommittedChildren(TransactionId **ptr)
Definition: xact.c:5436
Datum pg_prepared_xact(PG_FUNCTION_ARGS)
Definition: twophase.c:717
void StartPrepare(GlobalTransaction gxact)
Definition: twophase.c:987
bool log_checkpoints
Definition: xlog.c:104
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
#define TWOPHASE_RM_MAX_ID
Definition: twophase_rmgr.h:29
static void RecordTransactionAbortPrepared(TransactionId xid, int nchildren, TransactionId *children, int nrels, RelFileNode *rels, const char *gid)
Definition: twophase.c:2277
bool LWLockHeldByMeInMode(LWLock *l, LWLockMode mode)
Definition: lwlock.c:1861
void PrepareRedoRemove(TransactionId xid, bool giveWarning)
Definition: twophase.c:2406
int32 nsubxacts
Definition: xact.h:303
int errhint(const char *fmt,...)
Definition: elog.c:1071
BackendId MyBackendId
Definition: globals.c:81
static TwoPhaseStateData * TwoPhaseState
Definition: twophase.c:191
#define XLOG_XACT_PREPARE
Definition: xact.h:144
#define TransactionIdEquals(id1, id2)
Definition: transam.h:43
bool TransactionIdFollows(TransactionId id1, TransactionId id2)
Definition: transam.c:334
BackendId backendId
Definition: proc.h:113
int wal_segment_size
Definition: xlog.c:115
uint32 TransactionId
Definition: c.h:513
void AdvanceNextFullTransactionIdPastXid(TransactionId xid)
Definition: varsup.c:262
TupleDesc CreateTemplateTupleDesc(int natts)
Definition: tupdesc.c:44
Definition: proc.h:222
int32 ncommitrels
Definition: xact.h:304
XLogRecPtr XactLogAbortRecord(TimestampTz abort_time, int nsubxacts, TransactionId *subxacts, int nrels, RelFileNode *rels, int xactflags, TransactionId twophase_xid, const char *twophase_gid)
Definition: xact.c:5615
Oid GetUserId(void)
Definition: miscinit.c:439
#define DoNotReplicateId
Definition: origin.h:34
bool LWLockHeldByMe(LWLock *l)
Definition: lwlock.c:1843
int errmsg_plural(const char *fmt_singular, const char *fmt_plural, unsigned long n,...)
Definition: elog.c:934
#define write(a, b, c)
Definition: win32.h:14
TimestampTz GetCurrentTimestamp(void)
Definition: timestamp.c:1583
Oid tempNamespaceId
Definition: proc.h:117
TransactionId xmin
Definition: proc.h:228
XLogRecPtr XactLastRecEnd
Definition: xlog.c:361
PGXACT * allPgXact
Definition: proc.h:249
int64 TimestampTz
Definition: timestamp.h:39
#define SRF_IS_FIRSTCALL()
Definition: funcapi.h:293
TransactionId xid
Definition: proc.h:224
SHM_QUEUE links
Definition: proc.h:98
StateFileChunk * tail
Definition: twophase.c:943
TransactionId TransactionIdLatest(TransactionId mainxid, int nxids, const TransactionId *xids)
Definition: transam.c:365
static int GetPreparedTransactionList(GlobalTransaction *gxacts)
Definition: twophase.c:672
uint32 pg_crc32c
Definition: pg_crc32c.h:38
XLogRecPtr ProcLastRecPtr
Definition: xlog.c:360
static void ProcessRecords(char *bufptr, TransactionId xid, const TwoPhaseCallback callbacks[])
Definition: twophase.c:1569
void SyncRepWaitForLSN(XLogRecPtr lsn, bool commit)
Definition: syncrep.c:146
bool lwWaiting
Definition: proc.h:130
const TwoPhaseCallback twophase_recover_callbacks[TWOPHASE_RM_MAX_ID+1]
Definition: twophase_rmgr.c:24
#define END_CRIT_SECTION()
Definition: miscadmin.h:134
void fsync_fname(const char *fname, bool isdir)
Definition: fd.c:630
GlobalTransaction freeGXacts
Definition: twophase.c:182
bool TransactionIdFollowsOrEquals(TransactionId id1, TransactionId id2)
Definition: transam.c:349
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
PGPROC * PreparedXactProcs
Definition: proc.c:82
uint16 RepOriginId
Definition: xlogdefs.h:58
#define FLEXIBLE_ARRAY_MEMBER
Definition: c.h:276
Size TwoPhaseShmemSize(void)
Definition: twophase.c:236
#define InHotStandby
Definition: xlog.h:74
#define START_CRIT_SECTION()
Definition: miscadmin.h:132
Oid roleId
Definition: proc.h:115
int errcode(int sqlerrcode)
Definition: elog.c:610
const TwoPhaseCallback twophase_postcommit_callbacks[TWOPHASE_RM_MAX_ID+1]
Definition: twophase_rmgr.c:33
PROC_HDR * ProcGlobal
Definition: proc.c:80
#define TWOPHASE_RM_END_ID
Definition: twophase_rmgr.h:24
#define MemSet(start, val, len)
Definition: c.h:971
static char * ReadTwoPhaseFile(TransactionId xid, bool missing_ok)
Definition: twophase.c:1210
#define XLOG_INCLUDE_ORIGIN
Definition: xlog.h:229
static bool twophaseExitRegistered
Definition: twophase.c:201
uint16 gidlen
Definition: xact.h:308
void replorigin_advance(RepOriginId node, XLogRecPtr remote_commit, XLogRecPtr local_commit, bool go_backward, bool wal_log)
Definition: origin.c:857
uint8 lwWaitMode
Definition: proc.h:131
FullTransactionId nextFullXid
Definition: transam.h:164
uint32 num_chunks
Definition: twophase.c:944
HeapTuple heap_form_tuple(TupleDesc tupleDescriptor, Datum *values, bool *isnull)
Definition: heaptuple.c:1020
GlobalTransaction next
Definition: twophase.c:151
struct GlobalTransactionData * GlobalTransaction
Definition: twophase.h:26
bool TransactionIdDidCommit(TransactionId transactionId)
Definition: transam.c:125
#define LOG
Definition: elog.h:26
unsigned int Oid
Definition: postgres_ext.h:31
bool RecoveryInProgress(void)
Definition: xlog.c:8016
Definition: dirent.h:9
#define PANIC
Definition: elog.h:53
#define SRF_PERCALL_SETUP()
Definition: funcapi.h:297
void XLogFlush(XLogRecPtr record)
Definition: xlog.c:2842
static int fd(const char *x, int i)
Definition: preproc-init.c:105
struct TwoPhaseRecordOnDisk TwoPhaseRecordOnDisk
#define PG_BINARY
Definition: c.h:1234
Oid database
Definition: xact.h:300
char gid[GIDSIZE]
Definition: twophase.c:172
TupleDesc tuple_desc
Definition: funcapi.h:112
void RegisterTwoPhaseRecord(TwoPhaseRmgrId rmid, uint16 info, const void *data, uint32 len)
Definition: twophase.c:1187
PGXACT * MyPgXact
Definition: proc.c:68
uint8 vacuumFlags
Definition: proc.h:233
BackendId locking_backend
Definition: twophase.c:168
XLogRecPtr prepare_start_lsn
Definition: twophase.c:163
char * data
Definition: twophase.c:935
#define XidFromFullTransactionId(x)
Definition: transam.h:48
int32 nabortrels
Definition: xact.h:305
#define XACT_FLAGS_ACQUIREDACCESSEXCLUSIVELOCK
Definition: xact.h:103
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1727
#define RESUME_INTERRUPTS()
Definition: miscadmin.h:118
bool isBackgroundWorker
Definition: proc.h:120
#define SRF_RETURN_NEXT(_funcctx, _result)
Definition: funcapi.h:299
void TransactionIdCommitTree(TransactionId xid, int nxids, TransactionId *xids)
Definition: transam.c:260
GlobalTransaction MarkAsPreparing(TransactionId xid, const char *gid, TimestampTz prepared_at, Oid owner, Oid databaseid)
Definition: twophase.c:372
struct TwoPhaseStateData TwoPhaseStateData
uint8 TwoPhaseRmgrId
Definition: twophase_rmgr.h:19
unsigned short uint16
Definition: c.h:366
void pfree(void *pointer)
Definition: mcxt.c:1056
XLogRecord * XLogReadRecord(XLogReaderState *state, char **errormsg)
Definition: xlogreader.c:261
#define XLogRecGetData(decoder)
Definition: xlogreader.h:288
static void MarkAsPreparingGuts(GlobalTransaction gxact, TransactionId xid, const char *gid, TimestampTz prepared_at, Oid owner, Oid databaseid)
Definition: twophase.c:446
static GlobalTransaction TwoPhaseGetGXact(TransactionId xid, bool lock_held)
Definition: twophase.c:809
static GlobalTransaction LockGXact(const char *gid, Oid user)
Definition: twophase.c:558
Definition: dirent.c:25
#define ObjectIdGetDatum(X)
Definition: postgres.h:507
#define TwoPhaseFilePath(path, xid)
Definition: twophase.c:889
#define ERROR
Definition: elog.h:43
#define XLogRecGetDataLen(decoder)
Definition: xlogreader.h:289
static void GXactLoadSubxactData(GlobalTransaction gxact, int nsubxacts, TransactionId *children)
Definition: twophase.c:509
int max_prepared_xacts
Definition: twophase.c:117
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2370
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:392
#define TimestampTzGetDatum(X)
Definition: timestamp.h:32
void ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
Definition: procarray.c:336
#define MAXPGPATH
static void RemoveGXact(GlobalTransaction gxact)
Definition: twophase.c:634
uint32 bytes_free
Definition: twophase.c:945
void DropRelationFiles(RelFileNode *delrels, int ndelrels, bool isRedo)
Definition: md.c:1015
#define GIDSIZE
Definition: xact.h:31
int MaxBackends
Definition: globals.c:135
void AtAbort_Twophase(void)
Definition: twophase.c:317
void AtEOXact_PgStat(bool isCommit, bool parallel)
Definition: pgstat.c:2141
TupleDesc BlessTupleDesc(TupleDesc tupdesc)
Definition: execTuples.c:2052
#define DEBUG2
Definition: elog.h:24
int smgrGetPendingDeletes(bool forCommit, RelFileNode **ptr)
Definition: storage.c:816
void PredicateLockTwoPhaseFinish(TransactionId xid, bool isCommit)
Definition: predicate.c:4951
PROCLOCK * waitProcLock
Definition: proc.h:140
XLogRecPtr replorigin_session_origin_lsn
Definition: origin.c:153
static char * buf
Definition: pg_test_fsync.c:67
TimestampTz replorigin_session_origin_timestamp
Definition: origin.c:154
bool IsUnderPostmaster
Definition: globals.c:109
uint8 nxids
Definition: proc.h:238
int errdetail(const char *fmt,...)
Definition: elog.c:957
XLogRecPtr origin_lsn
Definition: xact.h:309
int errcode_for_file_access(void)
Definition: elog.c:633
VariableCache ShmemVariableCache
Definition: varsup.c:34
void before_shmem_exit(pg_on_exit_callback function, Datum arg)
Definition: ipc.c:333
#define InvalidTransactionId
Definition: transam.h:31
uint32 total_len
Definition: xact.h:298
void XLogBeginRead(XLogReaderState *state, XLogRecPtr RecPtr)
Definition: xlogreader.c:233
bool StandbyTransactionIdIsPrepared(TransactionId xid)
Definition: twophase.c:1372
uint32 len
Definition: twophase.c:936
Oid databaseId
Definition: proc.h:114
void StandbyReleaseLockTree(TransactionId xid, int nsubxids, TransactionId *subxids)
Definition: standby.c:738
unsigned int uint32
Definition: c.h:367
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2581
GlobalTransaction array
Definition: twophase.c:704
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1386
uint32 LocalTransactionId
Definition: c.h:515
PGPROC * TwoPhaseGetDummyProc(TransactionId xid, bool lock_held)
Definition: twophase.c:878
LOCK * waitLock
Definition: proc.h:139
#define EQ_CRC32C(c1, c2)
Definition: pg_crc32c.h:42
void XLogSetRecordFlags(uint8 flags)
Definition: xloginsert.c:398
void TupleDescInitEntry(TupleDesc desc, AttrNumber attributeNumber, const char *attributeName, Oid oidtypeid, int32 typmod, int attdim)
Definition: tupdesc.c:603
struct XidCache subxids
Definition: proc.h:162
bool TransactionIdDidAbort(TransactionId transactionId)
Definition: transam.c:181
GlobalTransaction prepXacts[FLEXIBLE_ARRAY_MEMBER]
Definition: twophase.c:188
#define STATUS_OK
Definition: c.h:1133
TransactionId PrescanPreparedTransactions(TransactionId **xids_p, int *nxids_p)
Definition: twophase.c:1840
bool delayChkpt
Definition: proc.h:235
bool superuser_arg(Oid roleid)
Definition: superuser.c:56
#define XLogRecGetInfo(decoder)
Definition: xlogreader.h:284
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
static struct xllist records
static void RecreateTwoPhaseFile(TransactionId xid, void *content, int len)
Definition: twophase.c:1616
XLogRecPtr XactLogCommitRecord(TimestampTz commit_time, int nsubxacts, TransactionId *subxacts, int nrels, RelFileNode *rels, int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInval, bool forceSync, int xactflags, TransactionId twophase_xid, const char *twophase_gid)
Definition: xact.c:5460
#define ERRCODE_DATA_CORRUPTED
Definition: pg_basebackup.c:45
int MyXactFlags
Definition: xact.c:118
uint32 total_len
Definition: twophase.c:946
#define MaxAllocSize
Definition: memutils.h:40
int CloseTransientFile(int fd)
Definition: fd.c:2547
#define XLogRecPtrIsInvalid(r)
Definition: xlogdefs.h:29
#define WARNING
Definition: elog.h:40
struct StateFileChunk * next
Definition: twophase.c:937
void XLogReaderFree(XLogReaderState *state)
Definition: xlogreader.c:135
#define stat(a, b)
Definition: win32_port.h:255
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:324
void TwoPhaseShmemInit(void)
Definition: twophase.c:252
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:416
Size mul_size(Size s1, Size s2)
Definition: shmem.c:515
#define TransactionIdGetDatum(X)
Definition: postgres.h:521
BackendId TwoPhaseGetDummyBackendId(TransactionId xid, bool lock_held)
Definition: twophase.c:863
#define InvalidBackendId
Definition: backendid.h:23
const TwoPhaseCallback twophase_postabort_callbacks[TWOPHASE_RM_MAX_ID+1]
Definition: twophase_rmgr.c:42
void * palloc0(Size size)
Definition: mcxt.c:980
uintptr_t Datum
Definition: postgres.h:367
struct StateFileChunk StateFileChunk
Size add_size(Size s1, Size s2)
Definition: shmem.c:498
int BackendId
Definition: backendid.h:21
Oid MyDatabaseId
Definition: globals.c:85
static char * ProcessTwoPhaseBuffer(TransactionId xid, XLogRecPtr prepare_start_lsn, bool fromdisk, bool setParent, bool setNextXid)
Definition: twophase.c:2063
void RelationCacheInitFilePostInvalidate(void)
Definition: relcache.c:6299
bool overflowed
Definition: proc.h:234
void PrepareRedoAdd(char *buf, XLogRecPtr start_lsn, XLogRecPtr end_lsn, RepOriginId origin_id)
Definition: twophase.c:2336
int waitStatus
Definition: proc.h:102
static void RemoveTwoPhaseFile(TransactionId xid, bool giveWarning)
Definition: twophase.c:1597
#define PGPROC_MAX_CACHED_SUBXIDS
Definition: proc.h:36
#define InvalidOid
Definition: postgres_ext.h:36
#define TWOPHASE_MAGIC
Definition: twophase.c:911
#define ereport(elevel,...)
Definition: elog.h:144
void FinishPreparedTransaction(const char *gid, bool isCommit)
Definition: twophase.c:1400
struct GlobalTransactionData GlobalTransactionData
bool initfileinval
Definition: xact.h:307
#define Max(x, y)
Definition: c.h:914
TransactionId xids[PGPROC_MAX_CACHED_SUBXIDS]
Definition: proc.h:40
StateFileChunk * head
Definition: twophase.c:942
void RelationCacheInitFilePreInvalidate(void)
Definition: relcache.c:6274
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:738
RepOriginId replorigin_session_origin
Definition: origin.c:152
struct dirent * ReadDir(DIR *dir, const char *dirname)
Definition: fd.c:2647
void EndPrepare(GlobalTransaction gxact)
Definition: twophase.c:1060
TimestampTz origin_timestamp
Definition: xact.h:310
MemoryContext multi_call_memory_ctx
Definition: funcapi.h:101
size_t Size
Definition: c.h:466
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: pgstat.h:1362
void XLogEnsureRecordSpace(int max_block_id, int ndatas)
Definition: xloginsert.c:147
void replorigin_session_advance(XLogRecPtr remote_commit, XLogRecPtr local_commit)
Definition: origin.c:1170
#define XLOG_XACT_OPMASK
Definition: xact.h:153
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1123
#define MAXALIGN(LEN)
Definition: c.h:691
#define HeapTupleGetDatum(tuple)
Definition: funcapi.h:220
void SendSharedInvalidMessages(const SharedInvalidationMessage *msgs, int n)
Definition: sinval.c:49
static void save_state_data(const void *data, uint32 len)
Definition: twophase.c:959
void * repalloc(void *pointer, Size size)
Definition: mcxt.c:1069
static void MarkAsPrepared(GlobalTransaction gxact, bool lock_held)
Definition: twophase.c:536
TransactionId xid
Definition: xact.h:299
#define InvalidRepOriginId
Definition: origin.h:33
#define TWOPHASE_DIR
Definition: twophase.c:114
TransactionId xid
Definition: twophase.c:165
XLogReaderState * XLogReaderAllocate(int wal_segment_size, const char *waldir, XLogPageReadCB pagereadfunc, void *private_data)
Definition: xlogreader.c:73
static Datum values[MAXATTR]
Definition: bootstrap.c:167
TimestampTz prepared_at
Definition: twophase.c:154
void SHMQueueElemInit(SHM_QUEUE *queue)
Definition: shmqueue.c:57
static void XlogReadTwoPhaseData(XLogRecPtr lsn, char **buf, int *len)
Definition: twophase.c:1327
int pgprocno
Definition: proc.h:110
static char * user
Definition: pg_regress.c:93
void * user_fctx
Definition: funcapi.h:82
void restoreTwoPhaseData(void)
Definition: twophase.c:1778
uint32 magic
Definition: xact.h:297
void * palloc(Size size)
Definition: mcxt.c:949
int errmsg(const char *fmt,...)
Definition: elog.c:824
void SHMQueueInit(SHM_QUEUE *queue)
Definition: shmqueue.c:36
int xactGetCommittedInvalidationMessages(SharedInvalidationMessage **msgs, bool *RelcacheInitFileInval)
Definition: inval.c:828
static GlobalTransaction MyLockedGxact
Definition: twophase.c:199
#define HOLD_INTERRUPTS()
Definition: miscadmin.h:116
#define elog(elevel,...)
Definition: elog.h:214
void PostPrepare_Twophase(void)
Definition: twophase.c:357
int i
#define CStringGetTextDatum(s)
Definition: builtins.h:87
void * arg
#define PG_FUNCTION_ARGS
Definition: fmgr.h:188
TimestampTz prepared_at
Definition: xact.h:301
PGPROC * allProcs
Definition: proc.h:247
int pg_fsync(int fd)
Definition: fd.c:343
void CheckPointTwoPhase(XLogRecPtr redo_horizon)
Definition: twophase.c:1696
SHM_QUEUE myProcLocks[NUM_LOCK_PARTITIONS]
Definition: proc.h:160
char d_name[MAX_PATH]
Definition: dirent.h:14
xl_xact_prepare TwoPhaseFileHeader
Definition: twophase.c:913
#define TransactionIdIsValid(xid)
Definition: transam.h:41
void StandbyRecoverPreparedTransactions(void)
Definition: twophase.c:1921
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:225
static void AtProcExit_Twophase(int code, Datum arg)
Definition: twophase.c:307
void SubTransSetParent(TransactionId xid, TransactionId parent)
Definition: subtrans.c:74
#define COMP_CRC32C(crc, data, len)
Definition: pg_crc32c.h:89
void RecoverPreparedTransactions(void)
Definition: twophase.c:1962
void XLogBeginInsert(void)
Definition: xloginsert.c:121
void TransactionTreeSetCommitTsData(TransactionId xid, int nsubxids, TransactionId *subxids, TimestampTz timestamp, RepOriginId nodeid, bool write_xlog)
Definition: commit_ts.c:145
int read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen, XLogRecPtr targetRecPtr, char *cur_page)
Definition: xlogutils.c:826
#define ERRCODE_DUPLICATE_OBJECT
Definition: streamutil.c:31
Definition: proc.h:95
void(* TwoPhaseCallback)(TransactionId xid, uint16 info, void *recdata, uint32 len)
Definition: twophase_rmgr.h:17
#define FIN_CRC32C(crc)
Definition: pg_crc32c.h:94
int pid
Definition: proc.h:109
int16 AttrNumber
Definition: attnum.h:21
XLogRecPtr prepare_end_lsn
Definition: twophase.c:164
void ProcArrayAdd(PGPROC *proc)
Definition: procarray.c:278
#define read(a, b, c)
Definition: win32.h:13
int FreeDir(DIR *dir)
Definition: fd.c:2699
#define offsetof(type, field)
Definition: c.h:661
BackendId dummyBackendId
Definition: twophase.c:153
#define NUM_LOCK_PARTITIONS
Definition: lwlock.h:117
#define XLogRecGetRmid(decoder)
Definition: xlogreader.h:285
static void RecordTransactionCommitPrepared(TransactionId xid, int nchildren, TransactionId *children, int nrels, RelFileNode *rels, int ninvalmsgs, SharedInvalidationMessage *invalmsgs, bool initfileinval, const char *gid)
Definition: twophase.c:2183
#define SRF_RETURN_DONE(_funcctx)
Definition: funcapi.h:317
LocalTransactionId lxid
Definition: proc.h:106
#define SRF_FIRSTCALL_INIT()
Definition: funcapi.h:295
TwoPhaseRmgrId rmid
Definition: twophase.c:924