PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
clog.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * clog.c
4  * PostgreSQL transaction-commit-log manager
5  *
6  * This module replaces the old "pg_log" access code, which treated pg_log
7  * essentially like a relation, in that it went through the regular buffer
8  * manager. The problem with that was that there wasn't any good way to
9  * recycle storage space for transactions so old that they'll never be
10  * looked up again. Now we use specialized access code so that the commit
11  * log can be broken into relatively small, independent segments.
12  *
13  * XLOG interactions: this module generates an XLOG record whenever a new
14  * CLOG page is initialized to zeroes. Other writes of CLOG come from
15  * recording of transaction commit or abort in xact.c, which generates its
16  * own XLOG records for these events and will re-perform the status update
17  * on redo; so we need make no additional XLOG entry here. For synchronous
18  * transaction commits, the XLOG is guaranteed flushed through the XLOG commit
19  * record before we are called to log a commit, so the WAL rule "write xlog
20  * before data" is satisfied automatically. However, for async commits we
21  * must track the latest LSN affecting each CLOG page, so that we can flush
22  * XLOG that far and satisfy the WAL rule. We don't have to worry about this
23  * for aborts (whether sync or async), since the post-crash assumption would
24  * be that such transactions failed anyway.
25  *
26  * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
27  * Portions Copyright (c) 1994, Regents of the University of California
28  *
29  * src/backend/access/transam/clog.c
30  *
31  *-------------------------------------------------------------------------
32  */
33 #include "postgres.h"
34 
35 #include "access/clog.h"
36 #include "access/slru.h"
37 #include "access/transam.h"
38 #include "access/xlog.h"
39 #include "access/xloginsert.h"
40 #include "access/xlogutils.h"
41 #include "miscadmin.h"
42 #include "pg_trace.h"
43 
44 /*
45  * Defines for CLOG page sizes. A page is the same BLCKSZ as is used
46  * everywhere else in Postgres.
47  *
48  * Note: because TransactionIds are 32 bits and wrap around at 0xFFFFFFFF,
49  * CLOG page numbering also wraps around at 0xFFFFFFFF/CLOG_XACTS_PER_PAGE,
50  * and CLOG segment numbering at
51  * 0xFFFFFFFF/CLOG_XACTS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need take no
52  * explicit notice of that fact in this module, except when comparing segment
53  * and page numbers in TruncateCLOG (see CLOGPagePrecedes).
54  */
55 
56 /* We need two bits per xact, so four xacts fit in a byte */
57 #define CLOG_BITS_PER_XACT 2
58 #define CLOG_XACTS_PER_BYTE 4
59 #define CLOG_XACTS_PER_PAGE (BLCKSZ * CLOG_XACTS_PER_BYTE)
60 #define CLOG_XACT_BITMASK ((1 << CLOG_BITS_PER_XACT) - 1)
61 
62 #define TransactionIdToPage(xid) ((xid) / (TransactionId) CLOG_XACTS_PER_PAGE)
63 #define TransactionIdToPgIndex(xid) ((xid) % (TransactionId) CLOG_XACTS_PER_PAGE)
64 #define TransactionIdToByte(xid) (TransactionIdToPgIndex(xid) / CLOG_XACTS_PER_BYTE)
65 #define TransactionIdToBIndex(xid) ((xid) % (TransactionId) CLOG_XACTS_PER_BYTE)
66 
67 /* We store the latest async LSN for each group of transactions */
68 #define CLOG_XACTS_PER_LSN_GROUP 32 /* keep this a power of 2 */
69 #define CLOG_LSNS_PER_PAGE (CLOG_XACTS_PER_PAGE / CLOG_XACTS_PER_LSN_GROUP)
70 
71 #define GetLSNIndex(slotno, xid) ((slotno) * CLOG_LSNS_PER_PAGE + \
72  ((xid) % (TransactionId) CLOG_XACTS_PER_PAGE) / CLOG_XACTS_PER_LSN_GROUP)
73 
74 
75 /*
76  * Link to shared-memory data structures for CLOG control
77  */
79 
80 #define ClogCtl (&ClogCtlData)
81 
82 
83 static int ZeroCLOGPage(int pageno, bool writeXlog);
84 static bool CLOGPagePrecedes(int page1, int page2);
85 static void WriteZeroPageXlogRec(int pageno);
86 static void WriteTruncateXlogRec(int pageno, TransactionId oldestXact,
87  Oid oldestXidDb);
88 static void TransactionIdSetPageStatus(TransactionId xid, int nsubxids,
89  TransactionId *subxids, XidStatus status,
90  XLogRecPtr lsn, int pageno);
92  XLogRecPtr lsn, int slotno);
93 static void set_status_by_pages(int nsubxids, TransactionId *subxids,
95 
96 
97 /*
98  * TransactionIdSetTreeStatus
99  *
100  * Record the final state of transaction entries in the commit log for
101  * a transaction and its subtransaction tree. Take care to ensure this is
102  * efficient, and as atomic as possible.
103  *
104  * xid is a single xid to set status for. This will typically be
105  * the top level transactionid for a top level commit or abort. It can
106  * also be a subtransaction when we record transaction aborts.
107  *
108  * subxids is an array of xids of length nsubxids, representing subtransactions
109  * in the tree of xid. In various cases nsubxids may be zero.
110  *
111  * lsn must be the WAL location of the commit record when recording an async
112  * commit. For a synchronous commit it can be InvalidXLogRecPtr, since the
113  * caller guarantees the commit record is already flushed in that case. It
114  * should be InvalidXLogRecPtr for abort cases, too.
115  *
116  * In the commit case, atomicity is limited by whether all the subxids are in
117  * the same CLOG page as xid. If they all are, then the lock will be grabbed
118  * only once, and the status will be set to committed directly. Otherwise
119  * we must
120  * 1. set sub-committed all subxids that are not on the same page as the
121  * main xid
122  * 2. atomically set committed the main xid and the subxids on the same page
123  * 3. go over the first bunch again and set them committed
124  * Note that as far as concurrent checkers are concerned, main transaction
125  * commit as a whole is still atomic.
126  *
127  * Example:
128  * TransactionId t commits and has subxids t1, t2, t3, t4
129  * t is on page p1, t1 is also on p1, t2 and t3 are on p2, t4 is on p3
130  * 1. update pages2-3:
131  * page2: set t2,t3 as sub-committed
132  * page3: set t4 as sub-committed
133  * 2. update page1:
134  * set t1 as sub-committed,
135  * then set t as committed,
136  then set t1 as committed
137  * 3. update pages2-3:
138  * page2: set t2,t3 as committed
139  * page3: set t4 as committed
140  *
141  * NB: this is a low-level routine and is NOT the preferred entry point
142  * for most uses; functions in transam.c are the intended callers.
143  *
144  * XXX Think about issuing FADVISE_WILLNEED on pages that we will need,
145  * but aren't yet in cache, as well as hinting pages not to fall out of
146  * cache yet.
147  */
148 void
150  TransactionId *subxids, XidStatus status, XLogRecPtr lsn)
151 {
152  int pageno = TransactionIdToPage(xid); /* get page of parent */
153  int i;
154 
156  status == TRANSACTION_STATUS_ABORTED);
157 
158  /*
159  * See how many subxids, if any, are on the same page as the parent, if
160  * any.
161  */
162  for (i = 0; i < nsubxids; i++)
163  {
164  if (TransactionIdToPage(subxids[i]) != pageno)
165  break;
166  }
167 
168  /*
169  * Do all items fit on a single page?
170  */
171  if (i == nsubxids)
172  {
173  /*
174  * Set the parent and all subtransactions in a single call
175  */
176  TransactionIdSetPageStatus(xid, nsubxids, subxids, status, lsn,
177  pageno);
178  }
179  else
180  {
181  int nsubxids_on_first_page = i;
182 
183  /*
184  * If this is a commit then we care about doing this correctly (i.e.
185  * using the subcommitted intermediate status). By here, we know
186  * we're updating more than one page of clog, so we must mark entries
187  * that are *not* on the first page so that they show as subcommitted
188  * before we then return to update the status to fully committed.
189  *
190  * To avoid touching the first page twice, skip marking subcommitted
191  * for the subxids on that first page.
192  */
193  if (status == TRANSACTION_STATUS_COMMITTED)
194  set_status_by_pages(nsubxids - nsubxids_on_first_page,
195  subxids + nsubxids_on_first_page,
197 
198  /*
199  * Now set the parent and subtransactions on same page as the parent,
200  * if any
201  */
202  pageno = TransactionIdToPage(xid);
203  TransactionIdSetPageStatus(xid, nsubxids_on_first_page, subxids, status,
204  lsn, pageno);
205 
206  /*
207  * Now work through the rest of the subxids one clog page at a time,
208  * starting from the second page onwards, like we did above.
209  */
210  set_status_by_pages(nsubxids - nsubxids_on_first_page,
211  subxids + nsubxids_on_first_page,
212  status, lsn);
213  }
214 }
215 
216 /*
217  * Helper for TransactionIdSetTreeStatus: set the status for a bunch of
218  * transactions, chunking in the separate CLOG pages involved. We never
219  * pass the whole transaction tree to this function, only subtransactions
220  * that are on different pages to the top level transaction id.
221  */
222 static void
223 set_status_by_pages(int nsubxids, TransactionId *subxids,
225 {
226  int pageno = TransactionIdToPage(subxids[0]);
227  int offset = 0;
228  int i = 0;
229 
230  while (i < nsubxids)
231  {
232  int num_on_page = 0;
233 
234  while (TransactionIdToPage(subxids[i]) == pageno && i < nsubxids)
235  {
236  num_on_page++;
237  i++;
238  }
239 
241  num_on_page, subxids + offset,
242  status, lsn, pageno);
243  offset = i;
244  pageno = TransactionIdToPage(subxids[offset]);
245  }
246 }
247 
248 /*
249  * Record the final state of transaction entries in the commit log for
250  * all entries on a single page. Atomic only on this page.
251  *
252  * Otherwise API is same as TransactionIdSetTreeStatus()
253  */
254 static void
256  TransactionId *subxids, XidStatus status,
257  XLogRecPtr lsn, int pageno)
258 {
259  int slotno;
260  int i;
261 
263  status == TRANSACTION_STATUS_ABORTED ||
265 
266  LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
267 
268  /*
269  * If we're doing an async commit (ie, lsn is valid), then we must wait
270  * for any active write on the page slot to complete. Otherwise our
271  * update could reach disk in that write, which will not do since we
272  * mustn't let it reach disk until we've done the appropriate WAL flush.
273  * But when lsn is invalid, it's OK to scribble on a page while it is
274  * write-busy, since we don't care if the update reaches disk sooner than
275  * we think.
276  */
277  slotno = SimpleLruReadPage(ClogCtl, pageno, XLogRecPtrIsInvalid(lsn), xid);
278 
279  /*
280  * Set the main transaction id, if any.
281  *
282  * If we update more than one xid on this page while it is being written
283  * out, we might find that some of the bits go to disk and others don't.
284  * If we are updating commits on the page with the top-level xid that
285  * could break atomicity, so we subcommit the subxids first before we mark
286  * the top-level commit.
287  */
288  if (TransactionIdIsValid(xid))
289  {
290  /* Subtransactions first, if needed ... */
291  if (status == TRANSACTION_STATUS_COMMITTED)
292  {
293  for (i = 0; i < nsubxids; i++)
294  {
295  Assert(ClogCtl->shared->page_number[slotno] == TransactionIdToPage(subxids[i]));
296  TransactionIdSetStatusBit(subxids[i],
298  lsn, slotno);
299  }
300  }
301 
302  /* ... then the main transaction */
303  TransactionIdSetStatusBit(xid, status, lsn, slotno);
304  }
305 
306  /* Set the subtransactions */
307  for (i = 0; i < nsubxids; i++)
308  {
309  Assert(ClogCtl->shared->page_number[slotno] == TransactionIdToPage(subxids[i]));
310  TransactionIdSetStatusBit(subxids[i], status, lsn, slotno);
311  }
312 
313  ClogCtl->shared->page_dirty[slotno] = true;
314 
315  LWLockRelease(CLogControlLock);
316 }
317 
318 /*
319  * Sets the commit status of a single transaction.
320  *
321  * Must be called with CLogControlLock held
322  */
323 static void
325 {
326  int byteno = TransactionIdToByte(xid);
327  int bshift = TransactionIdToBIndex(xid) * CLOG_BITS_PER_XACT;
328  char *byteptr;
329  char byteval;
330  char curval;
331 
332  byteptr = ClogCtl->shared->page_buffer[slotno] + byteno;
333  curval = (*byteptr >> bshift) & CLOG_XACT_BITMASK;
334 
335  /*
336  * When replaying transactions during recovery we still need to perform
337  * the two phases of subcommit and then commit. However, some transactions
338  * are already correctly marked, so we just treat those as a no-op which
339  * allows us to keep the following Assert as restrictive as possible.
340  */
341  if (InRecovery && status == TRANSACTION_STATUS_SUB_COMMITTED &&
343  return;
344 
345  /*
346  * Current state change should be from 0 or subcommitted to target state
347  * or we should already be there when replaying changes during recovery.
348  */
349  Assert(curval == 0 ||
351  status != TRANSACTION_STATUS_IN_PROGRESS) ||
352  curval == status);
353 
354  /* note this assumes exclusive access to the clog page */
355  byteval = *byteptr;
356  byteval &= ~(((1 << CLOG_BITS_PER_XACT) - 1) << bshift);
357  byteval |= (status << bshift);
358  *byteptr = byteval;
359 
360  /*
361  * Update the group LSN if the transaction completion LSN is higher.
362  *
363  * Note: lsn will be invalid when supplied during InRecovery processing,
364  * so we don't need to do anything special to avoid LSN updates during
365  * recovery. After recovery completes the next clog change will set the
366  * LSN correctly.
367  */
368  if (!XLogRecPtrIsInvalid(lsn))
369  {
370  int lsnindex = GetLSNIndex(slotno, xid);
371 
372  if (ClogCtl->shared->group_lsn[lsnindex] < lsn)
373  ClogCtl->shared->group_lsn[lsnindex] = lsn;
374  }
375 }
376 
377 /*
378  * Interrogate the state of a transaction in the commit log.
379  *
380  * Aside from the actual commit status, this function returns (into *lsn)
381  * an LSN that is late enough to be able to guarantee that if we flush up to
382  * that LSN then we will have flushed the transaction's commit record to disk.
383  * The result is not necessarily the exact LSN of the transaction's commit
384  * record! For example, for long-past transactions (those whose clog pages
385  * already migrated to disk), we'll return InvalidXLogRecPtr. Also, because
386  * we group transactions on the same clog page to conserve storage, we might
387  * return the LSN of a later transaction that falls into the same group.
388  *
389  * NB: this is a low-level routine and is NOT the preferred entry point
390  * for most uses; TransactionLogFetch() in transam.c is the intended caller.
391  */
392 XidStatus
394 {
395  int pageno = TransactionIdToPage(xid);
396  int byteno = TransactionIdToByte(xid);
397  int bshift = TransactionIdToBIndex(xid) * CLOG_BITS_PER_XACT;
398  int slotno;
399  int lsnindex;
400  char *byteptr;
402 
403  /* lock is acquired by SimpleLruReadPage_ReadOnly */
404 
405  slotno = SimpleLruReadPage_ReadOnly(ClogCtl, pageno, xid);
406  byteptr = ClogCtl->shared->page_buffer[slotno] + byteno;
407 
408  status = (*byteptr >> bshift) & CLOG_XACT_BITMASK;
409 
410  lsnindex = GetLSNIndex(slotno, xid);
411  *lsn = ClogCtl->shared->group_lsn[lsnindex];
412 
413  LWLockRelease(CLogControlLock);
414 
415  return status;
416 }
417 
418 /*
419  * Number of shared CLOG buffers.
420  *
421  * On larger multi-processor systems, it is possible to have many CLOG page
422  * requests in flight at one time which could lead to disk access for CLOG
423  * page if the required page is not found in memory. Testing revealed that we
424  * can get the best performance by having 128 CLOG buffers, more than that it
425  * doesn't improve performance.
426  *
427  * Unconditionally keeping the number of CLOG buffers to 128 did not seem like
428  * a good idea, because it would increase the minimum amount of shared memory
429  * required to start, which could be a problem for people running very small
430  * configurations. The following formula seems to represent a reasonable
431  * compromise: people with very low values for shared_buffers will get fewer
432  * CLOG buffers as well, and everyone else will get 128.
433  */
434 Size
436 {
437  return Min(128, Max(4, NBuffers / 512));
438 }
439 
440 /*
441  * Initialization of shared memory for CLOG
442  */
443 Size
445 {
447 }
448 
449 void
451 {
452  ClogCtl->PagePrecedes = CLOGPagePrecedes;
454  CLogControlLock, "pg_xact", LWTRANCHE_CLOG_BUFFERS);
455 }
456 
457 /*
458  * This func must be called ONCE on system install. It creates
459  * the initial CLOG segment. (The CLOG directory is assumed to
460  * have been created by initdb, and CLOGShmemInit must have been
461  * called already.)
462  */
463 void
465 {
466  int slotno;
467 
468  LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
469 
470  /* Create and zero the first page of the commit log */
471  slotno = ZeroCLOGPage(0, false);
472 
473  /* Make sure it's written out */
474  SimpleLruWritePage(ClogCtl, slotno);
475  Assert(!ClogCtl->shared->page_dirty[slotno]);
476 
477  LWLockRelease(CLogControlLock);
478 }
479 
480 /*
481  * Initialize (or reinitialize) a page of CLOG to zeroes.
482  * If writeXlog is TRUE, also emit an XLOG record saying we did this.
483  *
484  * The page is not actually written, just set up in shared memory.
485  * The slot number of the new page is returned.
486  *
487  * Control lock must be held at entry, and will be held at exit.
488  */
489 static int
490 ZeroCLOGPage(int pageno, bool writeXlog)
491 {
492  int slotno;
493 
494  slotno = SimpleLruZeroPage(ClogCtl, pageno);
495 
496  if (writeXlog)
497  WriteZeroPageXlogRec(pageno);
498 
499  return slotno;
500 }
501 
502 /*
503  * This must be called ONCE during postmaster or standalone-backend startup,
504  * after StartupXLOG has initialized ShmemVariableCache->nextXid.
505  */
506 void
508 {
510  int pageno = TransactionIdToPage(xid);
511 
512  LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
513 
514  /*
515  * Initialize our idea of the latest page number.
516  */
517  ClogCtl->shared->latest_page_number = pageno;
518 
519  LWLockRelease(CLogControlLock);
520 }
521 
522 /*
523  * This must be called ONCE at the end of startup/recovery.
524  */
525 void
526 TrimCLOG(void)
527 {
529  int pageno = TransactionIdToPage(xid);
530 
531  LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
532 
533  /*
534  * Re-Initialize our idea of the latest page number.
535  */
536  ClogCtl->shared->latest_page_number = pageno;
537 
538  /*
539  * Zero out the remainder of the current clog page. Under normal
540  * circumstances it should be zeroes already, but it seems at least
541  * theoretically possible that XLOG replay will have settled on a nextXID
542  * value that is less than the last XID actually used and marked by the
543  * previous database lifecycle (since subtransaction commit writes clog
544  * but makes no WAL entry). Let's just be safe. (We need not worry about
545  * pages beyond the current one, since those will be zeroed when first
546  * used. For the same reason, there is no need to do anything when
547  * nextXid is exactly at a page boundary; and it's likely that the
548  * "current" page doesn't exist yet in that case.)
549  */
550  if (TransactionIdToPgIndex(xid) != 0)
551  {
552  int byteno = TransactionIdToByte(xid);
553  int bshift = TransactionIdToBIndex(xid) * CLOG_BITS_PER_XACT;
554  int slotno;
555  char *byteptr;
556 
557  slotno = SimpleLruReadPage(ClogCtl, pageno, false, xid);
558  byteptr = ClogCtl->shared->page_buffer[slotno] + byteno;
559 
560  /* Zero so-far-unused positions in the current byte */
561  *byteptr &= (1 << bshift) - 1;
562  /* Zero the rest of the page */
563  MemSet(byteptr + 1, 0, BLCKSZ - byteno - 1);
564 
565  ClogCtl->shared->page_dirty[slotno] = true;
566  }
567 
568  LWLockRelease(CLogControlLock);
569 }
570 
571 /*
572  * This must be called ONCE during postmaster or standalone-backend shutdown
573  */
574 void
576 {
577  /* Flush dirty CLOG pages to disk */
578  TRACE_POSTGRESQL_CLOG_CHECKPOINT_START(false);
579  SimpleLruFlush(ClogCtl, false);
580 
581  /*
582  * fsync pg_xact to ensure that any files flushed previously are durably
583  * on disk.
584  */
585  fsync_fname("pg_xact", true);
586 
587  TRACE_POSTGRESQL_CLOG_CHECKPOINT_DONE(false);
588 }
589 
590 /*
591  * Perform a checkpoint --- either during shutdown, or on-the-fly
592  */
593 void
595 {
596  /* Flush dirty CLOG pages to disk */
597  TRACE_POSTGRESQL_CLOG_CHECKPOINT_START(true);
598  SimpleLruFlush(ClogCtl, true);
599 
600  /*
601  * fsync pg_xact to ensure that any files flushed previously are durably
602  * on disk.
603  */
604  fsync_fname("pg_xact", true);
605 
606  TRACE_POSTGRESQL_CLOG_CHECKPOINT_DONE(true);
607 }
608 
609 
610 /*
611  * Make sure that CLOG has room for a newly-allocated XID.
612  *
613  * NB: this is called while holding XidGenLock. We want it to be very fast
614  * most of the time; even when it's not so fast, no actual I/O need happen
615  * unless we're forced to write out a dirty clog or xlog page to make room
616  * in shared memory.
617  */
618 void
620 {
621  int pageno;
622 
623  /*
624  * No work except at first XID of a page. But beware: just after
625  * wraparound, the first XID of page zero is FirstNormalTransactionId.
626  */
627  if (TransactionIdToPgIndex(newestXact) != 0 &&
629  return;
630 
631  pageno = TransactionIdToPage(newestXact);
632 
633  LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
634 
635  /* Zero the page and make an XLOG entry about it */
636  ZeroCLOGPage(pageno, true);
637 
638  LWLockRelease(CLogControlLock);
639 }
640 
641 
642 /*
643  * Remove all CLOG segments before the one holding the passed transaction ID
644  *
645  * Before removing any CLOG data, we must flush XLOG to disk, to ensure
646  * that any recently-emitted HEAP_FREEZE records have reached disk; otherwise
647  * a crash and restart might leave us with some unfrozen tuples referencing
648  * removed CLOG data. We choose to emit a special TRUNCATE XLOG record too.
649  * Replaying the deletion from XLOG is not critical, since the files could
650  * just as well be removed later, but doing so prevents a long-running hot
651  * standby server from acquiring an unreasonably bloated CLOG directory.
652  *
653  * Since CLOG segments hold a large number of transactions, the opportunity to
654  * actually remove a segment is fairly rare, and so it seems best not to do
655  * the XLOG flush unless we have confirmed that there is a removable segment.
656  */
657 void
658 TruncateCLOG(TransactionId oldestXact, Oid oldestxid_datoid)
659 {
660  int cutoffPage;
661 
662  /*
663  * The cutoff point is the start of the segment containing oldestXact. We
664  * pass the *page* containing oldestXact to SimpleLruTruncate.
665  */
666  cutoffPage = TransactionIdToPage(oldestXact);
667 
668  /* Check to see if there's any files that could be removed */
670  return; /* nothing to remove */
671 
672  /*
673  * Advance oldestClogXid before truncating clog, so concurrent xact status
674  * lookups can ensure they don't attempt to access truncated-away clog.
675  *
676  * It's only necessary to do this if we will actually truncate away clog
677  * pages.
678  */
679  AdvanceOldestClogXid(oldestXact);
680 
681  /* vac_truncate_clog already advanced oldestXid */
684 
685  /*
686  * Write XLOG record and flush XLOG to disk. We record the oldest xid
687  * we're keeping information about here so we can ensure that it's always
688  * ahead of clog truncation in case we crash, and so a standby finds out
689  * the new valid xid before the next checkpoint.
690  */
691  WriteTruncateXlogRec(cutoffPage, oldestXact, oldestxid_datoid);
692 
693  /* Now we can remove the old CLOG segment(s) */
694  SimpleLruTruncate(ClogCtl, cutoffPage);
695 }
696 
697 
698 /*
699  * Decide which of two CLOG page numbers is "older" for truncation purposes.
700  *
701  * We need to use comparison of TransactionIds here in order to do the right
702  * thing with wraparound XID arithmetic. However, if we are asked about
703  * page number zero, we don't want to hand InvalidTransactionId to
704  * TransactionIdPrecedes: it'll get weird about permanent xact IDs. So,
705  * offset both xids by FirstNormalTransactionId to avoid that.
706  */
707 static bool
708 CLOGPagePrecedes(int page1, int page2)
709 {
710  TransactionId xid1;
711  TransactionId xid2;
712 
713  xid1 = ((TransactionId) page1) * CLOG_XACTS_PER_PAGE;
714  xid1 += FirstNormalTransactionId;
715  xid2 = ((TransactionId) page2) * CLOG_XACTS_PER_PAGE;
716  xid2 += FirstNormalTransactionId;
717 
718  return TransactionIdPrecedes(xid1, xid2);
719 }
720 
721 
722 /*
723  * Write a ZEROPAGE xlog record
724  */
725 static void
727 {
728  XLogBeginInsert();
729  XLogRegisterData((char *) (&pageno), sizeof(int));
730  (void) XLogInsert(RM_CLOG_ID, CLOG_ZEROPAGE);
731 }
732 
733 /*
734  * Write a TRUNCATE xlog record
735  *
736  * We must flush the xlog record to disk before returning --- see notes
737  * in TruncateCLOG().
738  */
739 static void
740 WriteTruncateXlogRec(int pageno, TransactionId oldestXact, Oid oldestXactDb)
741 {
742  XLogRecPtr recptr;
743  xl_clog_truncate xlrec;
744 
745  xlrec.pageno = pageno;
746  xlrec.oldestXact = oldestXact;
747  xlrec.oldestXactDb = oldestXactDb;
748 
749  XLogBeginInsert();
750  XLogRegisterData((char *) (&xlrec), sizeof(xl_clog_truncate));
751  recptr = XLogInsert(RM_CLOG_ID, CLOG_TRUNCATE);
752  XLogFlush(recptr);
753 }
754 
755 /*
756  * CLOG resource manager's routines
757  */
758 void
760 {
761  uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
762 
763  /* Backup blocks are not used in clog records */
764  Assert(!XLogRecHasAnyBlockRefs(record));
765 
766  if (info == CLOG_ZEROPAGE)
767  {
768  int pageno;
769  int slotno;
770 
771  memcpy(&pageno, XLogRecGetData(record), sizeof(int));
772 
773  LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
774 
775  slotno = ZeroCLOGPage(pageno, false);
776  SimpleLruWritePage(ClogCtl, slotno);
777  Assert(!ClogCtl->shared->page_dirty[slotno]);
778 
779  LWLockRelease(CLogControlLock);
780  }
781  else if (info == CLOG_TRUNCATE)
782  {
783  xl_clog_truncate xlrec;
784 
785  memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_clog_truncate));
786 
787  /*
788  * During XLOG replay, latest_page_number isn't set up yet; insert a
789  * suitable value to bypass the sanity test in SimpleLruTruncate.
790  */
791  ClogCtl->shared->latest_page_number = xlrec.pageno;
792 
794 
796  }
797  else
798  elog(PANIC, "clog_redo: unknown op code %u", info);
799 }
#define TransactionIdToBIndex(xid)
Definition: clog.c:65
#define CLOG_XACT_BITMASK
Definition: clog.c:60
#define CLOG_XACTS_PER_PAGE
Definition: clog.c:59
#define TransactionIdToPage(xid)
Definition: clog.c:62
static void WriteZeroPageXlogRec(int pageno)
Definition: clog.c:726
void TruncateCLOG(TransactionId oldestXact, Oid oldestxid_datoid)
Definition: clog.c:658
#define TransactionIdEquals(id1, id2)
Definition: transam.h:43
#define TRANSACTION_STATUS_COMMITTED
Definition: clog.h:27
uint32 TransactionId
Definition: c.h:397
#define GetLSNIndex(slotno, xid)
Definition: clog.c:71
void clog_redo(XLogReaderState *record)
Definition: clog.c:759
static void TransactionIdSetStatusBit(TransactionId xid, XidStatus status, XLogRecPtr lsn, int slotno)
Definition: clog.c:324
void SimpleLruTruncate(SlruCtl ctl, int cutoffPage)
Definition: slru.c:1169
void ExtendCLOG(TransactionId newestXact)
Definition: clog.c:619
bool InRecovery
Definition: xlog.c:192
#define Min(x, y)
Definition: c.h:806
#define TransactionIdToPgIndex(xid)
Definition: clog.c:63
void ShutdownCLOG(void)
Definition: clog.c:575
void fsync_fname(const char *fname, bool isdir)
Definition: fd.c:567
unsigned char uint8
Definition: c.h:266
int XidStatus
Definition: clog.h:24
TransactionId oldestXid
Definition: transam.h:119
#define MemSet(start, val, len)
Definition: c.h:857
unsigned int Oid
Definition: postgres_ext.h:31
static SlruCtlData ClogCtlData
Definition: clog.c:78
#define PANIC
Definition: elog.h:53
Size SimpleLruShmemSize(int nslots, int nlsns)
Definition: slru.c:145
void SimpleLruFlush(SlruCtl ctl, bool allow_redirtied)
Definition: slru.c:1104
void XLogFlush(XLogRecPtr record)
Definition: xlog.c:2757
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1715
#define ClogCtl
Definition: clog.c:80
#define CLOG_LSNS_PER_PAGE
Definition: clog.c:69
#define CLOG_ZEROPAGE
Definition: clog.h:54
void AdvanceOldestClogXid(TransactionId oldest_datfrozenxid)
Definition: varsup.c:271
void TransactionIdSetTreeStatus(TransactionId xid, int nsubxids, TransactionId *subxids, XidStatus status, XLogRecPtr lsn)
Definition: clog.c:149
#define XLogRecGetData(decoder)
Definition: xlogreader.h:220
bool TransactionIdPrecedesOrEquals(TransactionId id1, TransactionId id2)
Definition: transam.c:319
#define FirstNormalTransactionId
Definition: transam.h:34
TransactionId nextXid
Definition: transam.h:117
int SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok, TransactionId xid)
Definition: slru.c:375
#define TRANSACTION_STATUS_ABORTED
Definition: clog.h:28
static bool CLOGPagePrecedes(int page1, int page2)
Definition: clog.c:708
VariableCache ShmemVariableCache
Definition: varsup.c:34
Size CLOGShmemBuffers(void)
Definition: clog.c:435
#define TRANSACTION_STATUS_SUB_COMMITTED
Definition: clog.h:29
#define InvalidTransactionId
Definition: transam.h:31
#define CLOG_BITS_PER_XACT
Definition: clog.c:57
void SimpleLruWritePage(SlruCtl ctl, int slotno)
Definition: slru.c:578
int pageno
Definition: clog.h:33
void CheckPointCLOG(void)
Definition: clog.c:594
#define CLOG_TRUNCATE
Definition: clog.h:55
#define XLogRecGetInfo(decoder)
Definition: xlogreader.h:216
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
#define XLogRecPtrIsInvalid(r)
Definition: xlogdefs.h:29
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:323
TransactionId oldestXact
Definition: clog.h:34
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
bool SlruScanDirCbReportPresence(SlruCtl ctl, char *filename, int segpage, void *data)
Definition: slru.c:1322
int SimpleLruReadPage_ReadOnly(SlruCtl ctl, int pageno, TransactionId xid)
Definition: slru.c:467
Size CLOGShmemSize(void)
Definition: clog.c:444
static void set_status_by_pages(int nsubxids, TransactionId *subxids, XidStatus status, XLogRecPtr lsn)
Definition: clog.c:223
void BootStrapCLOG(void)
Definition: clog.c:464
#define Max(x, y)
Definition: c.h:800
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:675
#define XLR_INFO_MASK
Definition: xlogrecord.h:62
bool SlruScanDirectory(SlruCtl ctl, SlruScanCallback callback, void *data)
Definition: slru.c:1377
size_t Size
Definition: c.h:356
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1111
void StartupCLOG(void)
Definition: clog.c:507
XidStatus TransactionIdGetStatus(TransactionId xid, XLogRecPtr *lsn)
Definition: clog.c:393
#define TransactionIdToByte(xid)
Definition: clog.c:64
static int ZeroCLOGPage(int pageno, bool writeXlog)
Definition: clog.c:490
int i
void CLOGShmemInit(void)
Definition: clog.c:450
int NBuffers
Definition: globals.c:123
Oid oldestXactDb
Definition: clog.h:35
#define XLogRecHasAnyBlockRefs(decoder)
Definition: xlogreader.h:222
#define elog
Definition: elog.h:219
#define TransactionIdIsValid(xid)
Definition: transam.h:41
static void WriteTruncateXlogRec(int pageno, TransactionId oldestXact, Oid oldestXidDb)
Definition: clog.c:740
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:224
void XLogBeginInsert(void)
Definition: xloginsert.c:120
static void TransactionIdSetPageStatus(TransactionId xid, int nsubxids, TransactionId *subxids, XidStatus status, XLogRecPtr lsn, int pageno)
Definition: clog.c:255
int SimpleLruZeroPage(SlruCtl ctl, int pageno)
Definition: slru.c:263
void TrimCLOG(void)
Definition: clog.c:526
#define TRANSACTION_STATUS_IN_PROGRESS
Definition: clog.h:26
void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns, LWLock *ctllock, const char *subdir, int tranche_id)
Definition: slru.c:165