1 /*-------------------------------------------------------------------------
4 * PostgreSQL transaction log manager
7 * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
8 * Portions Copyright (c) 1994, Regents of the University of California
10 * $PostgreSQL: pgsql/src/backend/access/transam/xlog.c,v 1.301 2008/05/09 14:27:47 heikki Exp $
12 *-------------------------------------------------------------------------
26 #include "access/clog.h"
27 #include "access/heapam.h"
28 #include "access/multixact.h"
29 #include "access/subtrans.h"
30 #include "access/transam.h"
31 #include "access/tuptoaster.h"
32 #include "access/twophase.h"
33 #include "access/xact.h"
34 #include "access/xlog_internal.h"
35 #include "access/xlogdefs.h"
36 #include "access/xlogutils.h"
37 #include "catalog/catversion.h"
38 #include "catalog/pg_control.h"
39 #include "catalog/pg_type.h"
41 #include "miscadmin.h"
43 #include "postmaster/bgwriter.h"
44 #include "storage/bufpage.h"
45 #include "storage/fd.h"
46 #include "storage/ipc.h"
47 #include "storage/pmsignal.h"
48 #include "storage/procarray.h"
49 #include "storage/smgr.h"
50 #include "storage/spin.h"
51 #include "utils/builtins.h"
52 #include "utils/pg_locale.h"
53 #include "utils/ps_status.h"
56 /* File path names (all relative to $PGDATA) */
57 #define BACKUP_LABEL_FILE "backup_label"
58 #define BACKUP_LABEL_OLD "backup_label.old"
59 #define RECOVERY_COMMAND_FILE "recovery.conf"
60 #define RECOVERY_COMMAND_DONE "recovery.done"
63 /* User-settable parameters */
64 int CheckPointSegments = 3;
66 int XLogArchiveTimeout = 0;
67 bool XLogArchiveMode = false;
68 char *XLogArchiveCommand = NULL;
69 char *XLOG_sync_method = NULL;
70 const char XLOG_sync_method_default[] = DEFAULT_SYNC_METHOD_STR;
71 bool fullPageWrites = true;
72 bool log_checkpoints = false;
75 bool XLOG_DEBUG = false;
79 * XLOGfileslop is the maximum number of preallocated future XLOG segments.
80 * When we are done with an old XLOG segment file, we will recycle it as a
81 * future XLOG segment as long as there aren't already XLOGfileslop future
82 * segments; else we'll delete it. This could be made a separate GUC
83 * variable, but at present I think it's sufficient to hardwire it as
84 * 2*CheckPointSegments+1. Under normal conditions, a checkpoint will free
85 * no more than 2*CheckPointSegments log segments, and we want to recycle all
86 * of them; the +1 allows boundary cases to happen without wasting a
87 * delete/create-segment cycle.
89 #define XLOGfileslop (2*CheckPointSegments + 1)
92 /* these are derived from XLOG_sync_method by assign_xlog_sync_method */
93 int sync_method = DEFAULT_SYNC_METHOD;
94 static int open_sync_bit = DEFAULT_SYNC_FLAGBIT;
96 #define XLOG_SYNC_BIT (enableFsync ? open_sync_bit : 0)
100 * Statistics for current checkpoint are collected in this global struct.
101 * Because only the background writer or a stand-alone backend can perform
102 * checkpoints, this will be unused in normal backends.
104 CheckpointStatsData CheckpointStats;
107 * ThisTimeLineID will be same in all backends --- it identifies current
108 * WAL timeline for the database system.
110 TimeLineID ThisTimeLineID = 0;
112 /* Are we doing recovery from XLOG? */
113 bool InRecovery = false;
115 /* Are we recovering using offline XLOG archives? */
116 static bool InArchiveRecovery = false;
118 /* Was the last xlog file restored from archive, or local? */
119 static bool restoredFromArchive = false;
121 /* options taken from recovery.conf */
122 static char *recoveryRestoreCommand = NULL;
123 static bool recoveryTarget = false;
124 static bool recoveryTargetExact = false;
125 static bool recoveryTargetInclusive = true;
126 static bool recoveryLogRestartpoints = false;
127 static TransactionId recoveryTargetXid;
128 static TimestampTz recoveryTargetTime;
129 static TimestampTz recoveryLastXTime = 0;
131 /* if recoveryStopsHere returns true, it saves actual stop xid/time here */
132 static TransactionId recoveryStopXid;
133 static TimestampTz recoveryStopTime;
134 static bool recoveryStopAfter;
137 * During normal operation, the only timeline we care about is ThisTimeLineID.
138 * During recovery, however, things are more complicated. To simplify life
139 * for rmgr code, we keep ThisTimeLineID set to the "current" timeline as we
140 * scan through the WAL history (that is, it is the line that was active when
141 * the currently-scanned WAL record was generated). We also need these
144 * recoveryTargetTLI: the desired timeline that we want to end in.
146 * expectedTLIs: an integer list of recoveryTargetTLI and the TLIs of
147 * its known parents, newest first (so recoveryTargetTLI is always the
148 * first list member). Only these TLIs are expected to be seen in the WAL
149 * segments we read, and indeed only these TLIs will be considered as
150 * candidate WAL files to open at all.
152 * curFileTLI: the TLI appearing in the name of the current input WAL file.
153 * (This is not necessarily the same as ThisTimeLineID, because we could
154 * be scanning data that was copied from an ancestor timeline when the current
155 * file was created.) During a sequential scan we do not allow this value
158 static TimeLineID recoveryTargetTLI;
159 static List *expectedTLIs;
160 static TimeLineID curFileTLI;
163 * ProcLastRecPtr points to the start of the last XLOG record inserted by the
164 * current backend. It is updated for all inserts. XactLastRecEnd points to
165 * end+1 of the last record, and is reset when we end a top-level transaction,
166 * or start a new one; so it can be used to tell if the current transaction has
167 * created any XLOG records.
169 static XLogRecPtr ProcLastRecPtr = {0, 0};
171 XLogRecPtr XactLastRecEnd = {0, 0};
174 * RedoRecPtr is this backend's local copy of the REDO record pointer
175 * (which is almost but not quite the same as a pointer to the most recent
176 * CHECKPOINT record). We update this from the shared-memory copy,
177 * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
178 * hold the Insert lock). See XLogInsert for details. We are also allowed
179 * to update from XLogCtl->Insert.RedoRecPtr if we hold the info_lck;
180 * see GetRedoRecPtr. A freshly spawned backend obtains the value during
183 static XLogRecPtr RedoRecPtr;
186 * Shared-memory data structures for XLOG control
188 * LogwrtRqst indicates a byte position that we need to write and/or fsync
189 * the log up to (all records before that point must be written or fsynced).
190 * LogwrtResult indicates the byte positions we have already written/fsynced.
191 * These structs are identical but are declared separately to indicate their
192 * slightly different functions.
194 * We do a lot of pushups to minimize the amount of access to lockable
195 * shared memory values. There are actually three shared-memory copies of
196 * LogwrtResult, plus one unshared copy in each backend. Here's how it works:
197 * XLogCtl->LogwrtResult is protected by info_lck
198 * XLogCtl->Write.LogwrtResult is protected by WALWriteLock
199 * XLogCtl->Insert.LogwrtResult is protected by WALInsertLock
200 * One must hold the associated lock to read or write any of these, but
201 * of course no lock is needed to read/write the unshared LogwrtResult.
203 * XLogCtl->LogwrtResult and XLogCtl->Write.LogwrtResult are both "always
204 * right", since both are updated by a write or flush operation before
205 * it releases WALWriteLock. The point of keeping XLogCtl->Write.LogwrtResult
206 * is that it can be examined/modified by code that already holds WALWriteLock
207 * without needing to grab info_lck as well.
209 * XLogCtl->Insert.LogwrtResult may lag behind the reality of the other two,
210 * but is updated when convenient. Again, it exists for the convenience of
211 * code that is already holding WALInsertLock but not the other locks.
213 * The unshared LogwrtResult may lag behind any or all of these, and again
214 * is updated when convenient.
216 * The request bookkeeping is simpler: there is a shared XLogCtl->LogwrtRqst
217 * (protected by info_lck), but we don't need to cache any copies of it.
219 * Note that this all works because the request and result positions can only
220 * advance forward, never back up, and so we can easily determine which of two
221 * values is "more up to date".
223 * info_lck is only held long enough to read/update the protected variables,
224 * so it's a plain spinlock. The other locks are held longer (potentially
225 * over I/O operations), so we use LWLocks for them. These locks are:
227 * WALInsertLock: must be held to insert a record into the WAL buffers.
229 * WALWriteLock: must be held to write WAL buffers to disk (XLogWrite or
232 * ControlFileLock: must be held to read/update control file or create
235 * CheckpointLock: must be held to do a checkpoint (ensures only one
236 * checkpointer at a time; currently, with all checkpoints done by the
237 * bgwriter, this is just pro forma).
242 typedef struct XLogwrtRqst
244 XLogRecPtr Write; /* last byte + 1 to write out */
245 XLogRecPtr Flush; /* last byte + 1 to flush */
248 typedef struct XLogwrtResult
250 XLogRecPtr Write; /* last byte + 1 written out */
251 XLogRecPtr Flush; /* last byte + 1 flushed */
255 * Shared state data for XLogInsert.
257 typedef struct XLogCtlInsert
259 XLogwrtResult LogwrtResult; /* a recent value of LogwrtResult */
260 XLogRecPtr PrevRecord; /* start of previously-inserted record */
261 int curridx; /* current block index in cache */
262 XLogPageHeader currpage; /* points to header of block in cache */
263 char *currpos; /* current insertion point in cache */
264 XLogRecPtr RedoRecPtr; /* current redo point for insertions */
265 bool forcePageWrites; /* forcing full-page writes for PITR? */
269 * Shared state data for XLogWrite/XLogFlush.
271 typedef struct XLogCtlWrite
273 XLogwrtResult LogwrtResult; /* current value of LogwrtResult */
274 int curridx; /* cache index of next block to write */
275 pg_time_t lastSegSwitchTime; /* time of last xlog segment switch */
279 * Total shared-memory state for XLOG.
281 typedef struct XLogCtlData
283 /* Protected by WALInsertLock: */
284 XLogCtlInsert Insert;
286 /* Protected by info_lck: */
287 XLogwrtRqst LogwrtRqst;
288 XLogwrtResult LogwrtResult;
289 uint32 ckptXidEpoch; /* nextXID & epoch of latest checkpoint */
290 TransactionId ckptXid;
291 XLogRecPtr asyncCommitLSN; /* LSN of newest async commit */
293 /* Protected by WALWriteLock: */
297 * These values do not change after startup, although the pointed-to pages
298 * and xlblocks values certainly do. Permission to read/write the pages
299 * and xlblocks values depends on WALInsertLock and WALWriteLock.
301 char *pages; /* buffers for unwritten XLOG pages */
302 XLogRecPtr *xlblocks; /* 1st byte ptr-s + XLOG_BLCKSZ */
303 int XLogCacheBlck; /* highest allocated xlog buffer index */
304 TimeLineID ThisTimeLineID;
306 slock_t info_lck; /* locks shared variables shown above */
309 static XLogCtlData *XLogCtl = NULL;
312 * We maintain an image of pg_control in shared memory.
314 static ControlFileData *ControlFile = NULL;
317 * Macros for managing XLogInsert state. In most cases, the calling routine
318 * has local copies of XLogCtl->Insert and/or XLogCtl->Insert->curridx,
319 * so these are passed as parameters instead of being fetched via XLogCtl.
322 /* Free space remaining in the current xlog page buffer */
323 #define INSERT_FREESPACE(Insert) \
324 (XLOG_BLCKSZ - ((Insert)->currpos - (char *) (Insert)->currpage))
326 /* Construct XLogRecPtr value for current insertion point */
327 #define INSERT_RECPTR(recptr,Insert,curridx) \
329 (recptr).xlogid = XLogCtl->xlblocks[curridx].xlogid, \
331 XLogCtl->xlblocks[curridx].xrecoff - INSERT_FREESPACE(Insert) \
334 #define PrevBufIdx(idx) \
335 (((idx) == 0) ? XLogCtl->XLogCacheBlck : ((idx) - 1))
337 #define NextBufIdx(idx) \
338 (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))
341 * Private, possibly out-of-date copy of shared LogwrtResult.
342 * See discussion above.
344 static XLogwrtResult LogwrtResult = {{0, 0}, {0, 0}};
347 * openLogFile is -1 or a kernel FD for an open log file segment.
348 * When it's open, openLogOff is the current seek offset in the file.
349 * openLogId/openLogSeg identify the segment. These variables are only
350 * used to write the XLOG, and so will normally refer to the active segment.
352 static int openLogFile = -1;
353 static uint32 openLogId = 0;
354 static uint32 openLogSeg = 0;
355 static uint32 openLogOff = 0;
358 * These variables are used similarly to the ones above, but for reading
359 * the XLOG. Note, however, that readOff generally represents the offset
360 * of the page just read, not the seek position of the FD itself, which
361 * will be just past that page.
363 static int readFile = -1;
364 static uint32 readId = 0;
365 static uint32 readSeg = 0;
366 static uint32 readOff = 0;
368 /* Buffer for currently read page (XLOG_BLCKSZ bytes) */
369 static char *readBuf = NULL;
371 /* Buffer for current ReadRecord result (expandable) */
372 static char *readRecordBuf = NULL;
373 static uint32 readRecordBufSize = 0;
375 /* State information for XLOG reading */
376 static XLogRecPtr ReadRecPtr; /* start of last record read */
377 static XLogRecPtr EndRecPtr; /* end+1 of last record read */
378 static XLogRecord *nextRecord = NULL;
379 static TimeLineID lastPageTLI = 0;
381 static bool InRedo = false;
384 static void XLogArchiveNotify(const char *xlog);
385 static void XLogArchiveNotifySeg(uint32 log, uint32 seg);
386 static bool XLogArchiveCheckDone(const char *xlog, bool create_if_missing);
387 static void XLogArchiveCleanup(const char *xlog);
388 static void readRecoveryCommandFile(void);
389 static void exitArchiveRecovery(TimeLineID endTLI,
390 uint32 endLogId, uint32 endLogSeg);
391 static bool recoveryStopsHere(XLogRecord *record, bool *includeThis);
392 static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
394 static bool XLogCheckBuffer(XLogRecData *rdata, bool doPageWrites,
395 XLogRecPtr *lsn, BkpBlock *bkpb);
396 static bool AdvanceXLInsertBuffer(bool new_segment);
397 static void XLogWrite(XLogwrtRqst WriteRqst, bool flexible, bool xlog_switch);
398 static int XLogFileInit(uint32 log, uint32 seg,
399 bool *use_existent, bool use_lock);
400 static bool InstallXLogFileSegment(uint32 *log, uint32 *seg, char *tmppath,
401 bool find_free, int *max_advance,
403 static int XLogFileOpen(uint32 log, uint32 seg);
404 static int XLogFileRead(uint32 log, uint32 seg, int emode);
405 static void XLogFileClose(void);
406 static bool RestoreArchivedFile(char *path, const char *xlogfname,
407 const char *recovername, off_t expectedSize);
408 static void PreallocXlogFiles(XLogRecPtr endptr);
409 static void RemoveOldXlogFiles(uint32 log, uint32 seg, XLogRecPtr endptr);
410 static void CleanupBackupHistory(void);
411 static XLogRecord *ReadRecord(XLogRecPtr *RecPtr, int emode);
412 static bool ValidXLOGHeader(XLogPageHeader hdr, int emode);
413 static XLogRecord *ReadCheckpointRecord(XLogRecPtr RecPtr, int whichChkpt);
414 static List *readTimeLineHistory(TimeLineID targetTLI);
415 static bool existsTimeLineHistory(TimeLineID probeTLI);
416 static TimeLineID findNewestTimeLine(TimeLineID startTLI);
417 static void writeTimeLineHistory(TimeLineID newTLI, TimeLineID parentTLI,
419 uint32 endLogId, uint32 endLogSeg);
420 static void WriteControlFile(void);
421 static void ReadControlFile(void);
422 static char *str_time(pg_time_t tnow);
424 static void xlog_outrec(StringInfo buf, XLogRecord *record);
426 static void issue_xlog_fsync(void);
427 static void pg_start_backup_callback(int code, Datum arg);
428 static bool read_backup_label(XLogRecPtr *checkPointLoc,
429 XLogRecPtr *minRecoveryLoc);
430 static void rm_redo_error_callback(void *arg);
434 * Insert an XLOG record having the specified RMID and info bytes,
435 * with the body of the record being the data chunk(s) described by
436 * the rdata chain (see xlog.h for notes about rdata).
438 * Returns XLOG pointer to end of record (beginning of next record).
439 * This can be used as LSN for data pages affected by the logged action.
440 * (LSN is the XLOG point up to which the XLOG must be flushed to disk
441 * before the data page can be written out. This implements the basic
442 * WAL rule "write the log before the data".)
444 * NB: this routine feels free to scribble on the XLogRecData structs,
445 * though not on the data they reference. This is OK since the XLogRecData
446 * structs are always just temporaries in the calling code.
449 XLogInsert(RmgrId rmid, uint8 info, XLogRecData *rdata)
451 XLogCtlInsert *Insert = &XLogCtl->Insert;
453 XLogContRecord *contrecord;
455 XLogRecPtr WriteRqst;
459 Buffer dtbuf[XLR_MAX_BKP_BLOCKS];
460 bool dtbuf_bkp[XLR_MAX_BKP_BLOCKS];
461 BkpBlock dtbuf_xlg[XLR_MAX_BKP_BLOCKS];
462 XLogRecPtr dtbuf_lsn[XLR_MAX_BKP_BLOCKS];
463 XLogRecData dtbuf_rdt1[XLR_MAX_BKP_BLOCKS];
464 XLogRecData dtbuf_rdt2[XLR_MAX_BKP_BLOCKS];
465 XLogRecData dtbuf_rdt3[XLR_MAX_BKP_BLOCKS];
472 bool isLogSwitch = (rmid == RM_XLOG_ID && info == XLOG_SWITCH);
474 /* info's high bits are reserved for use by me */
475 if (info & XLR_INFO_MASK)
476 elog(PANIC, "invalid xlog info mask %02X", info);
479 * In bootstrap mode, we don't actually log anything but XLOG resources;
480 * return a phony record pointer.
482 if (IsBootstrapProcessingMode() && rmid != RM_XLOG_ID)
485 RecPtr.xrecoff = SizeOfXLogLongPHD; /* start of 1st chkpt record */
490 * Here we scan the rdata chain, determine which buffers must be backed
491 * up, and compute the CRC values for the data. Note that the record
492 * header isn't added into the CRC initially since we don't know the final
493 * length or info bits quite yet. Thus, the CRC will represent the CRC of
494 * the whole record in the order "rdata, then backup blocks, then record
497 * We may have to loop back to here if a race condition is detected below.
498 * We could prevent the race by doing all this work while holding the
499 * insert lock, but it seems better to avoid doing CRC calculations while
500 * holding the lock. This means we have to be careful about modifying the
501 * rdata chain until we know we aren't going to loop back again. The only
502 * change we allow ourselves to make earlier is to set rdt->data = NULL in
503 * chain items we have decided we will have to back up the whole buffer
504 * for. This is OK because we will certainly decide the same thing again
505 * for those items if we do it over; doing it here saves an extra pass
506 * over the chain later.
509 for (i = 0; i < XLR_MAX_BKP_BLOCKS; i++)
511 dtbuf[i] = InvalidBuffer;
512 dtbuf_bkp[i] = false;
516 * Decide if we need to do full-page writes in this XLOG record: true if
517 * full_page_writes is on or we have a PITR request for it. Since we
518 * don't yet have the insert lock, forcePageWrites could change under us,
519 * but we'll recheck it once we have the lock.
521 doPageWrites = fullPageWrites || Insert->forcePageWrites;
523 INIT_CRC32(rdata_crc);
527 if (rdt->buffer == InvalidBuffer)
529 /* Simple data, just include it */
531 COMP_CRC32(rdata_crc, rdt->data, rdt->len);
535 /* Find info for buffer */
536 for (i = 0; i < XLR_MAX_BKP_BLOCKS; i++)
538 if (rdt->buffer == dtbuf[i])
540 /* Buffer already referenced by earlier chain item */
546 COMP_CRC32(rdata_crc, rdt->data, rdt->len);
550 if (dtbuf[i] == InvalidBuffer)
552 /* OK, put it in this slot */
553 dtbuf[i] = rdt->buffer;
554 if (XLogCheckBuffer(rdt, doPageWrites,
555 &(dtbuf_lsn[i]), &(dtbuf_xlg[i])))
563 COMP_CRC32(rdata_crc, rdt->data, rdt->len);
568 if (i >= XLR_MAX_BKP_BLOCKS)
569 elog(PANIC, "can backup at most %d blocks per xlog record",
572 /* Break out of loop when rdt points to last chain item */
573 if (rdt->next == NULL)
579 * Now add the backup block headers and data into the CRC
581 for (i = 0; i < XLR_MAX_BKP_BLOCKS; i++)
585 BkpBlock *bkpb = &(dtbuf_xlg[i]);
588 COMP_CRC32(rdata_crc,
591 page = (char *) BufferGetBlock(dtbuf[i]);
592 if (bkpb->hole_length == 0)
594 COMP_CRC32(rdata_crc,
600 /* must skip the hole */
601 COMP_CRC32(rdata_crc,
604 COMP_CRC32(rdata_crc,
605 page + (bkpb->hole_offset + bkpb->hole_length),
606 BLCKSZ - (bkpb->hole_offset + bkpb->hole_length));
612 * NOTE: We disallow len == 0 because it provides a useful bit of extra
613 * error checking in ReadRecord. This means that all callers of
614 * XLogInsert must supply at least some not-in-a-buffer data. However, we
615 * make an exception for XLOG SWITCH records because we don't want them to
616 * ever cross a segment boundary.
618 if (len == 0 && !isLogSwitch)
619 elog(PANIC, "invalid xlog record length %u", len);
621 START_CRIT_SECTION();
623 /* Now wait to get insert lock */
624 LWLockAcquire(WALInsertLock, LW_EXCLUSIVE);
627 * Check to see if my RedoRecPtr is out of date. If so, may have to go
628 * back and recompute everything. This can only happen just after a
629 * checkpoint, so it's better to be slow in this case and fast otherwise.
631 * If we aren't doing full-page writes then RedoRecPtr doesn't actually
632 * affect the contents of the XLOG record, so we'll update our local copy
633 * but not force a recomputation.
635 if (!XLByteEQ(RedoRecPtr, Insert->RedoRecPtr))
637 Assert(XLByteLT(RedoRecPtr, Insert->RedoRecPtr));
638 RedoRecPtr = Insert->RedoRecPtr;
642 for (i = 0; i < XLR_MAX_BKP_BLOCKS; i++)
644 if (dtbuf[i] == InvalidBuffer)
646 if (dtbuf_bkp[i] == false &&
647 XLByteLE(dtbuf_lsn[i], RedoRecPtr))
650 * Oops, this buffer now needs to be backed up, but we
651 * didn't think so above. Start over.
653 LWLockRelease(WALInsertLock);
662 * Also check to see if forcePageWrites was just turned on; if we weren't
663 * already doing full-page writes then go back and recompute. (If it was
664 * just turned off, we could recompute the record without full pages, but
665 * we choose not to bother.)
667 if (Insert->forcePageWrites && !doPageWrites)
669 /* Oops, must redo it with full-page data */
670 LWLockRelease(WALInsertLock);
676 * Make additional rdata chain entries for the backup blocks, so that we
677 * don't need to special-case them in the write loop. Note that we have
678 * now irrevocably changed the input rdata chain. At the exit of this
679 * loop, write_len includes the backup block data.
681 * Also set the appropriate info bits to show which buffers were backed
682 * up. The i'th XLR_SET_BKP_BLOCK bit corresponds to the i'th distinct
683 * buffer value (ignoring InvalidBuffer) appearing in the rdata chain.
686 for (i = 0; i < XLR_MAX_BKP_BLOCKS; i++)
694 info |= XLR_SET_BKP_BLOCK(i);
696 bkpb = &(dtbuf_xlg[i]);
697 page = (char *) BufferGetBlock(dtbuf[i]);
699 rdt->next = &(dtbuf_rdt1[i]);
702 rdt->data = (char *) bkpb;
703 rdt->len = sizeof(BkpBlock);
704 write_len += sizeof(BkpBlock);
706 rdt->next = &(dtbuf_rdt2[i]);
709 if (bkpb->hole_length == 0)
718 /* must skip the hole */
720 rdt->len = bkpb->hole_offset;
721 write_len += bkpb->hole_offset;
723 rdt->next = &(dtbuf_rdt3[i]);
726 rdt->data = page + (bkpb->hole_offset + bkpb->hole_length);
727 rdt->len = BLCKSZ - (bkpb->hole_offset + bkpb->hole_length);
728 write_len += rdt->len;
734 * If we backed up any full blocks and online backup is not in progress,
735 * mark the backup blocks as removable. This allows the WAL archiver to
736 * know whether it is safe to compress archived WAL data by transforming
737 * full-block records into the non-full-block format.
739 * Note: we could just set the flag whenever !forcePageWrites, but
740 * defining it like this leaves the info bit free for some potential other
741 * use in records without any backup blocks.
743 if ((info & XLR_BKP_BLOCK_MASK) && !Insert->forcePageWrites)
744 info |= XLR_BKP_REMOVABLE;
747 * If there isn't enough space on the current XLOG page for a record
748 * header, advance to the next page (leaving the unused space as zeroes).
751 freespace = INSERT_FREESPACE(Insert);
752 if (freespace < SizeOfXLogRecord)
754 updrqst = AdvanceXLInsertBuffer(false);
755 freespace = INSERT_FREESPACE(Insert);
758 /* Compute record's XLOG location */
759 curridx = Insert->curridx;
760 INSERT_RECPTR(RecPtr, Insert, curridx);
763 * If the record is an XLOG_SWITCH, and we are exactly at the start of a
764 * segment, we need not insert it (and don't want to because we'd like
765 * consecutive switch requests to be no-ops). Instead, make sure
766 * everything is written and flushed through the end of the prior segment,
767 * and return the prior segment's end address.
770 (RecPtr.xrecoff % XLogSegSize) == SizeOfXLogLongPHD)
772 /* We can release insert lock immediately */
773 LWLockRelease(WALInsertLock);
775 RecPtr.xrecoff -= SizeOfXLogLongPHD;
776 if (RecPtr.xrecoff == 0)
778 /* crossing a logid boundary */
780 RecPtr.xrecoff = XLogFileSize;
783 LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
784 LogwrtResult = XLogCtl->Write.LogwrtResult;
785 if (!XLByteLE(RecPtr, LogwrtResult.Flush))
787 XLogwrtRqst FlushRqst;
789 FlushRqst.Write = RecPtr;
790 FlushRqst.Flush = RecPtr;
791 XLogWrite(FlushRqst, false, false);
793 LWLockRelease(WALWriteLock);
800 /* Insert record header */
802 record = (XLogRecord *) Insert->currpos;
803 record->xl_prev = Insert->PrevRecord;
804 record->xl_xid = GetCurrentTransactionIdIfAny();
805 record->xl_tot_len = SizeOfXLogRecord + write_len;
806 record->xl_len = len; /* doesn't include backup blocks */
807 record->xl_info = info;
808 record->xl_rmid = rmid;
810 /* Now we can finish computing the record's CRC */
811 COMP_CRC32(rdata_crc, (char *) record + sizeof(pg_crc32),
812 SizeOfXLogRecord - sizeof(pg_crc32));
813 FIN_CRC32(rdata_crc);
814 record->xl_crc = rdata_crc;
821 initStringInfo(&buf);
822 appendStringInfo(&buf, "INSERT @ %X/%X: ",
823 RecPtr.xlogid, RecPtr.xrecoff);
824 xlog_outrec(&buf, record);
825 if (rdata->data != NULL)
827 appendStringInfo(&buf, " - ");
828 RmgrTable[record->xl_rmid].rm_desc(&buf, record->xl_info, rdata->data);
830 elog(LOG, "%s", buf.data);
835 /* Record begin of record in appropriate places */
836 ProcLastRecPtr = RecPtr;
837 Insert->PrevRecord = RecPtr;
839 Insert->currpos += SizeOfXLogRecord;
840 freespace -= SizeOfXLogRecord;
843 * Append the data, including backup blocks if any
847 while (rdata->data == NULL)
852 if (rdata->len > freespace)
854 memcpy(Insert->currpos, rdata->data, freespace);
855 rdata->data += freespace;
856 rdata->len -= freespace;
857 write_len -= freespace;
861 memcpy(Insert->currpos, rdata->data, rdata->len);
862 freespace -= rdata->len;
863 write_len -= rdata->len;
864 Insert->currpos += rdata->len;
870 /* Use next buffer */
871 updrqst = AdvanceXLInsertBuffer(false);
872 curridx = Insert->curridx;
873 /* Insert cont-record header */
874 Insert->currpage->xlp_info |= XLP_FIRST_IS_CONTRECORD;
875 contrecord = (XLogContRecord *) Insert->currpos;
876 contrecord->xl_rem_len = write_len;
877 Insert->currpos += SizeOfXLogContRecord;
878 freespace = INSERT_FREESPACE(Insert);
881 /* Ensure next record will be properly aligned */
882 Insert->currpos = (char *) Insert->currpage +
883 MAXALIGN(Insert->currpos - (char *) Insert->currpage);
884 freespace = INSERT_FREESPACE(Insert);
887 * The recptr I return is the beginning of the *next* record. This will be
888 * stored as LSN for changed data pages...
890 INSERT_RECPTR(RecPtr, Insert, curridx);
893 * If the record is an XLOG_SWITCH, we must now write and flush all the
894 * existing data, and then forcibly advance to the start of the next
895 * segment. It's not good to do this I/O while holding the insert lock,
896 * but there seems too much risk of confusion if we try to release the
897 * lock sooner. Fortunately xlog switch needn't be a high-performance
898 * operation anyway...
902 XLogCtlWrite *Write = &XLogCtl->Write;
903 XLogwrtRqst FlushRqst;
904 XLogRecPtr OldSegEnd;
906 LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
909 * Flush through the end of the page containing XLOG_SWITCH, and
910 * perform end-of-segment actions (eg, notifying archiver).
912 WriteRqst = XLogCtl->xlblocks[curridx];
913 FlushRqst.Write = WriteRqst;
914 FlushRqst.Flush = WriteRqst;
915 XLogWrite(FlushRqst, false, true);
917 /* Set up the next buffer as first page of next segment */
918 /* Note: AdvanceXLInsertBuffer cannot need to do I/O here */
919 (void) AdvanceXLInsertBuffer(true);
921 /* There should be no unwritten data */
922 curridx = Insert->curridx;
923 Assert(curridx == Write->curridx);
925 /* Compute end address of old segment */
926 OldSegEnd = XLogCtl->xlblocks[curridx];
927 OldSegEnd.xrecoff -= XLOG_BLCKSZ;
928 if (OldSegEnd.xrecoff == 0)
930 /* crossing a logid boundary */
931 OldSegEnd.xlogid -= 1;
932 OldSegEnd.xrecoff = XLogFileSize;
935 /* Make it look like we've written and synced all of old segment */
936 LogwrtResult.Write = OldSegEnd;
937 LogwrtResult.Flush = OldSegEnd;
940 * Update shared-memory status --- this code should match XLogWrite
943 /* use volatile pointer to prevent code rearrangement */
944 volatile XLogCtlData *xlogctl = XLogCtl;
946 SpinLockAcquire(&xlogctl->info_lck);
947 xlogctl->LogwrtResult = LogwrtResult;
948 if (XLByteLT(xlogctl->LogwrtRqst.Write, LogwrtResult.Write))
949 xlogctl->LogwrtRqst.Write = LogwrtResult.Write;
950 if (XLByteLT(xlogctl->LogwrtRqst.Flush, LogwrtResult.Flush))
951 xlogctl->LogwrtRqst.Flush = LogwrtResult.Flush;
952 SpinLockRelease(&xlogctl->info_lck);
955 Write->LogwrtResult = LogwrtResult;
957 LWLockRelease(WALWriteLock);
959 updrqst = false; /* done already */
963 /* normal case, ie not xlog switch */
965 /* Need to update shared LogwrtRqst if some block was filled up */
966 if (freespace < SizeOfXLogRecord)
968 /* curridx is filled and available for writing out */
973 /* if updrqst already set, write through end of previous buf */
974 curridx = PrevBufIdx(curridx);
976 WriteRqst = XLogCtl->xlblocks[curridx];
979 LWLockRelease(WALInsertLock);
983 /* use volatile pointer to prevent code rearrangement */
984 volatile XLogCtlData *xlogctl = XLogCtl;
986 SpinLockAcquire(&xlogctl->info_lck);
987 /* advance global request to include new block(s) */
988 if (XLByteLT(xlogctl->LogwrtRqst.Write, WriteRqst))
989 xlogctl->LogwrtRqst.Write = WriteRqst;
990 /* update local result copy while I have the chance */
991 LogwrtResult = xlogctl->LogwrtResult;
992 SpinLockRelease(&xlogctl->info_lck);
995 XactLastRecEnd = RecPtr;
1003 * Determine whether the buffer referenced by an XLogRecData item has to
1004 * be backed up, and if so fill a BkpBlock struct for it. In any case
1005 * save the buffer's LSN at *lsn.
1008 XLogCheckBuffer(XLogRecData *rdata, bool doPageWrites,
1009 XLogRecPtr *lsn, BkpBlock *bkpb)
1013 page = (PageHeader) BufferGetBlock(rdata->buffer);
1016 * XXX We assume page LSN is first data on *every* page that can be passed
1017 * to XLogInsert, whether it otherwise has the standard page layout or
1020 *lsn = page->pd_lsn;
1023 XLByteLE(page->pd_lsn, RedoRecPtr))
1026 * The page needs to be backed up, so set up *bkpb
1028 bkpb->node = BufferGetFileNode(rdata->buffer);
1029 bkpb->block = BufferGetBlockNumber(rdata->buffer);
1031 if (rdata->buffer_std)
1033 /* Assume we can omit data between pd_lower and pd_upper */
1034 uint16 lower = page->pd_lower;
1035 uint16 upper = page->pd_upper;
1037 if (lower >= SizeOfPageHeaderData &&
1041 bkpb->hole_offset = lower;
1042 bkpb->hole_length = upper - lower;
1046 /* No "hole" to compress out */
1047 bkpb->hole_offset = 0;
1048 bkpb->hole_length = 0;
1053 /* Not a standard page header, don't try to eliminate "hole" */
1054 bkpb->hole_offset = 0;
1055 bkpb->hole_length = 0;
1058 return true; /* buffer requires backup */
1061 return false; /* buffer does not need to be backed up */
1067 * Create an archive notification file
1069 * The name of the notification file is the message that will be picked up
1070 * by the archiver, e.g. we write 0000000100000001000000C6.ready
1071 * and the archiver then knows to archive XLOGDIR/0000000100000001000000C6,
1072 * then when complete, rename it to 0000000100000001000000C6.done
1075 XLogArchiveNotify(const char *xlog)
1077 char archiveStatusPath[MAXPGPATH];
1080 /* insert an otherwise empty file called <XLOG>.ready */
1081 StatusFilePath(archiveStatusPath, xlog, ".ready");
1082 fd = AllocateFile(archiveStatusPath, "w");
1086 (errcode_for_file_access(),
1087 errmsg("could not create archive status file \"%s\": %m",
1088 archiveStatusPath)));
1094 (errcode_for_file_access(),
1095 errmsg("could not write archive status file \"%s\": %m",
1096 archiveStatusPath)));
1100 /* Notify archiver that it's got something to do */
1101 if (IsUnderPostmaster)
1102 SendPostmasterSignal(PMSIGNAL_WAKEN_ARCHIVER);
1106 * Convenience routine to notify using log/seg representation of filename
1109 XLogArchiveNotifySeg(uint32 log, uint32 seg)
1111 char xlog[MAXFNAMELEN];
1113 XLogFileName(xlog, ThisTimeLineID, log, seg);
1114 XLogArchiveNotify(xlog);
1118 * XLogArchiveCheckDone
1120 * This is called when we are ready to delete or recycle an old XLOG segment
1121 * file or backup history file. If it is okay to delete it then return true.
1122 * If it is not time to delete it, make sure a .ready file exists, and return
1125 * If <XLOG>.done exists, then return true; else if <XLOG>.ready exists,
1126 * then return false; else create <XLOG>.ready and return false.
1128 * The reason we do things this way is so that if the original attempt to
1129 * create <XLOG>.ready fails, we'll retry during subsequent checkpoints.
1132 XLogArchiveCheckDone(const char *xlog, bool create_if_missing)
1134 char archiveStatusPath[MAXPGPATH];
1135 struct stat stat_buf;
1137 /* Always deletable if archiving is off */
1138 if (!XLogArchivingActive())
1141 /* First check for .done --- this means archiver is done with it */
1142 StatusFilePath(archiveStatusPath, xlog, ".done");
1143 if (stat(archiveStatusPath, &stat_buf) == 0)
1146 /* check for .ready --- this means archiver is still busy with it */
1147 StatusFilePath(archiveStatusPath, xlog, ".ready");
1148 if (stat(archiveStatusPath, &stat_buf) == 0)
1151 /* Race condition --- maybe archiver just finished, so recheck */
1152 StatusFilePath(archiveStatusPath, xlog, ".done");
1153 if (stat(archiveStatusPath, &stat_buf) == 0)
1156 /* Retry creation of the .ready file */
1157 if (create_if_missing)
1158 XLogArchiveNotify(xlog);
1164 * XLogArchiveCleanup
1166 * Cleanup archive notification file(s) for a particular xlog segment
1169 XLogArchiveCleanup(const char *xlog)
1171 char archiveStatusPath[MAXPGPATH];
1173 /* Remove the .done file */
1174 StatusFilePath(archiveStatusPath, xlog, ".done");
1175 unlink(archiveStatusPath);
1176 /* should we complain about failure? */
1178 /* Remove the .ready file if present --- normally it shouldn't be */
1179 StatusFilePath(archiveStatusPath, xlog, ".ready");
1180 unlink(archiveStatusPath);
1181 /* should we complain about failure? */
1185 * Advance the Insert state to the next buffer page, writing out the next
1186 * buffer if it still contains unwritten data.
1188 * If new_segment is TRUE then we set up the next buffer page as the first
1189 * page of the next xlog segment file, possibly but not usually the next
1190 * consecutive file page.
1192 * The global LogwrtRqst.Write pointer needs to be advanced to include the
1193 * just-filled page. If we can do this for free (without an extra lock),
1194 * we do so here. Otherwise the caller must do it. We return TRUE if the
1195 * request update still needs to be done, FALSE if we did it internally.
1197 * Must be called with WALInsertLock held.
1200 AdvanceXLInsertBuffer(bool new_segment)
1202 XLogCtlInsert *Insert = &XLogCtl->Insert;
1203 XLogCtlWrite *Write = &XLogCtl->Write;
1204 int nextidx = NextBufIdx(Insert->curridx);
1205 bool update_needed = true;
1206 XLogRecPtr OldPageRqstPtr;
1207 XLogwrtRqst WriteRqst;
1208 XLogRecPtr NewPageEndPtr;
1209 XLogPageHeader NewPage;
1211 /* Use Insert->LogwrtResult copy if it's more fresh */
1212 if (XLByteLT(LogwrtResult.Write, Insert->LogwrtResult.Write))
1213 LogwrtResult = Insert->LogwrtResult;
1216 * Get ending-offset of the buffer page we need to replace (this may be
1217 * zero if the buffer hasn't been used yet). Fall through if it's already
1220 OldPageRqstPtr = XLogCtl->xlblocks[nextidx];
1221 if (!XLByteLE(OldPageRqstPtr, LogwrtResult.Write))
1223 /* nope, got work to do... */
1224 XLogRecPtr FinishedPageRqstPtr;
1226 FinishedPageRqstPtr = XLogCtl->xlblocks[Insert->curridx];
1228 /* Before waiting, get info_lck and update LogwrtResult */
1230 /* use volatile pointer to prevent code rearrangement */
1231 volatile XLogCtlData *xlogctl = XLogCtl;
1233 SpinLockAcquire(&xlogctl->info_lck);
1234 if (XLByteLT(xlogctl->LogwrtRqst.Write, FinishedPageRqstPtr))
1235 xlogctl->LogwrtRqst.Write = FinishedPageRqstPtr;
1236 LogwrtResult = xlogctl->LogwrtResult;
1237 SpinLockRelease(&xlogctl->info_lck);
1240 update_needed = false; /* Did the shared-request update */
1242 if (XLByteLE(OldPageRqstPtr, LogwrtResult.Write))
1244 /* OK, someone wrote it already */
1245 Insert->LogwrtResult = LogwrtResult;
1249 /* Must acquire write lock */
1250 LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
1251 LogwrtResult = Write->LogwrtResult;
1252 if (XLByteLE(OldPageRqstPtr, LogwrtResult.Write))
1254 /* OK, someone wrote it already */
1255 LWLockRelease(WALWriteLock);
1256 Insert->LogwrtResult = LogwrtResult;
1261 * Have to write buffers while holding insert lock. This is
1262 * not good, so only write as much as we absolutely must.
1264 WriteRqst.Write = OldPageRqstPtr;
1265 WriteRqst.Flush.xlogid = 0;
1266 WriteRqst.Flush.xrecoff = 0;
1267 XLogWrite(WriteRqst, false, false);
1268 LWLockRelease(WALWriteLock);
1269 Insert->LogwrtResult = LogwrtResult;
1275 * Now the next buffer slot is free and we can set it up to be the next
1278 NewPageEndPtr = XLogCtl->xlblocks[Insert->curridx];
1282 /* force it to a segment start point */
1283 NewPageEndPtr.xrecoff += XLogSegSize - 1;
1284 NewPageEndPtr.xrecoff -= NewPageEndPtr.xrecoff % XLogSegSize;
1287 if (NewPageEndPtr.xrecoff >= XLogFileSize)
1289 /* crossing a logid boundary */
1290 NewPageEndPtr.xlogid += 1;
1291 NewPageEndPtr.xrecoff = XLOG_BLCKSZ;
1294 NewPageEndPtr.xrecoff += XLOG_BLCKSZ;
1295 XLogCtl->xlblocks[nextidx] = NewPageEndPtr;
1296 NewPage = (XLogPageHeader) (XLogCtl->pages + nextidx * (Size) XLOG_BLCKSZ);
1298 Insert->curridx = nextidx;
1299 Insert->currpage = NewPage;
1301 Insert->currpos = ((char *) NewPage) +SizeOfXLogShortPHD;
1304 * Be sure to re-zero the buffer so that bytes beyond what we've written
1305 * will look like zeroes and not valid XLOG records...
1307 MemSet((char *) NewPage, 0, XLOG_BLCKSZ);
1310 * Fill the new page's header
1312 NewPage ->xlp_magic = XLOG_PAGE_MAGIC;
1314 /* NewPage->xlp_info = 0; */ /* done by memset */
1315 NewPage ->xlp_tli = ThisTimeLineID;
1316 NewPage ->xlp_pageaddr.xlogid = NewPageEndPtr.xlogid;
1317 NewPage ->xlp_pageaddr.xrecoff = NewPageEndPtr.xrecoff - XLOG_BLCKSZ;
1320 * If first page of an XLOG segment file, make it a long header.
1322 if ((NewPage->xlp_pageaddr.xrecoff % XLogSegSize) == 0)
1324 XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
1326 NewLongPage->xlp_sysid = ControlFile->system_identifier;
1327 NewLongPage->xlp_seg_size = XLogSegSize;
1328 NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
1329 NewPage ->xlp_info |= XLP_LONG_HEADER;
1331 Insert->currpos = ((char *) NewPage) +SizeOfXLogLongPHD;
1334 return update_needed;
1338 * Check whether we've consumed enough xlog space that a checkpoint is needed.
1340 * Caller must have just finished filling the open log file (so that
1341 * openLogId/openLogSeg are valid). We measure the distance from RedoRecPtr
1342 * to the open log file and see if that exceeds CheckPointSegments.
1344 * Note: it is caller's responsibility that RedoRecPtr is up-to-date.
1347 XLogCheckpointNeeded(void)
1350 * A straight computation of segment number could overflow 32 bits. Rather
1351 * than assuming we have working 64-bit arithmetic, we compare the
1352 * highest-order bits separately, and force a checkpoint immediately when
1357 uint32 old_highbits,
1360 old_segno = (RedoRecPtr.xlogid % XLogSegSize) * XLogSegsPerFile +
1361 (RedoRecPtr.xrecoff / XLogSegSize);
1362 old_highbits = RedoRecPtr.xlogid / XLogSegSize;
1363 new_segno = (openLogId % XLogSegSize) * XLogSegsPerFile + openLogSeg;
1364 new_highbits = openLogId / XLogSegSize;
1365 if (new_highbits != old_highbits ||
1366 new_segno >= old_segno + (uint32) (CheckPointSegments - 1))
1372 * Write and/or fsync the log at least as far as WriteRqst indicates.
1374 * If flexible == TRUE, we don't have to write as far as WriteRqst, but
1375 * may stop at any convenient boundary (such as a cache or logfile boundary).
1376 * This option allows us to avoid uselessly issuing multiple writes when a
1377 * single one would do.
1379 * If xlog_switch == TRUE, we are intending an xlog segment switch, so
1380 * perform end-of-segment actions after writing the last page, even if
1381 * it's not physically the end of its segment. (NB: this will work properly
1382 * only if caller specifies WriteRqst == page-end and flexible == false,
1383 * and there is some data to write.)
1385 * Must be called with WALWriteLock held.
1388 XLogWrite(XLogwrtRqst WriteRqst, bool flexible, bool xlog_switch)
1390 XLogCtlWrite *Write = &XLogCtl->Write;
1392 bool last_iteration;
1400 /* We should always be inside a critical section here */
1401 Assert(CritSectionCount > 0);
1404 * Update local LogwrtResult (caller probably did this already, but...)
1406 LogwrtResult = Write->LogwrtResult;
1409 * Since successive pages in the xlog cache are consecutively allocated,
1410 * we can usually gather multiple pages together and issue just one
1411 * write() call. npages is the number of pages we have determined can be
1412 * written together; startidx is the cache block index of the first one,
1413 * and startoffset is the file offset at which it should go. The latter
1414 * two variables are only valid when npages > 0, but we must initialize
1415 * all of them to keep the compiler quiet.
1422 * Within the loop, curridx is the cache block index of the page to
1423 * consider writing. We advance Write->curridx only after successfully
1424 * writing pages. (Right now, this refinement is useless since we are
1425 * going to PANIC if any error occurs anyway; but someday it may come in
1428 curridx = Write->curridx;
1430 while (XLByteLT(LogwrtResult.Write, WriteRqst.Write))
1433 * Make sure we're not ahead of the insert process. This could happen
1434 * if we're passed a bogus WriteRqst.Write that is past the end of the
1435 * last page that's been initialized by AdvanceXLInsertBuffer.
1437 if (!XLByteLT(LogwrtResult.Write, XLogCtl->xlblocks[curridx]))
1438 elog(PANIC, "xlog write request %X/%X is past end of log %X/%X",
1439 LogwrtResult.Write.xlogid, LogwrtResult.Write.xrecoff,
1440 XLogCtl->xlblocks[curridx].xlogid,
1441 XLogCtl->xlblocks[curridx].xrecoff);
1443 /* Advance LogwrtResult.Write to end of current buffer page */
1444 LogwrtResult.Write = XLogCtl->xlblocks[curridx];
1445 ispartialpage = XLByteLT(WriteRqst.Write, LogwrtResult.Write);
1447 if (!XLByteInPrevSeg(LogwrtResult.Write, openLogId, openLogSeg))
1450 * Switch to new logfile segment. We cannot have any pending
1451 * pages here (since we dump what we have at segment end).
1453 Assert(npages == 0);
1454 if (openLogFile >= 0)
1456 XLByteToPrevSeg(LogwrtResult.Write, openLogId, openLogSeg);
1458 /* create/use new log file */
1459 use_existent = true;
1460 openLogFile = XLogFileInit(openLogId, openLogSeg,
1461 &use_existent, true);
1465 /* Make sure we have the current logfile open */
1466 if (openLogFile < 0)
1468 XLByteToPrevSeg(LogwrtResult.Write, openLogId, openLogSeg);
1469 openLogFile = XLogFileOpen(openLogId, openLogSeg);
1473 /* Add current page to the set of pending pages-to-dump */
1476 /* first of group */
1478 startoffset = (LogwrtResult.Write.xrecoff - XLOG_BLCKSZ) % XLogSegSize;
1483 * Dump the set if this will be the last loop iteration, or if we are
1484 * at the last page of the cache area (since the next page won't be
1485 * contiguous in memory), or if we are at the end of the logfile
1488 last_iteration = !XLByteLT(LogwrtResult.Write, WriteRqst.Write);
1490 finishing_seg = !ispartialpage &&
1491 (startoffset + npages * XLOG_BLCKSZ) >= XLogSegSize;
1493 if (last_iteration ||
1494 curridx == XLogCtl->XLogCacheBlck ||
1500 /* Need to seek in the file? */
1501 if (openLogOff != startoffset)
1503 if (lseek(openLogFile, (off_t) startoffset, SEEK_SET) < 0)
1505 (errcode_for_file_access(),
1506 errmsg("could not seek in log file %u, "
1507 "segment %u to offset %u: %m",
1508 openLogId, openLogSeg, startoffset)));
1509 openLogOff = startoffset;
1512 /* OK to write the page(s) */
1513 from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
1514 nbytes = npages * (Size) XLOG_BLCKSZ;
1516 if (write(openLogFile, from, nbytes) != nbytes)
1518 /* if write didn't set errno, assume no disk space */
1522 (errcode_for_file_access(),
1523 errmsg("could not write to log file %u, segment %u "
1524 "at offset %u, length %lu: %m",
1525 openLogId, openLogSeg,
1526 openLogOff, (unsigned long) nbytes)));
1529 /* Update state for write */
1530 openLogOff += nbytes;
1531 Write->curridx = ispartialpage ? curridx : NextBufIdx(curridx);
1535 * If we just wrote the whole last page of a logfile segment,
1536 * fsync the segment immediately. This avoids having to go back
1537 * and re-open prior segments when an fsync request comes along
1538 * later. Doing it here ensures that one and only one backend will
1539 * perform this fsync.
1541 * We also do this if this is the last page written for an xlog
1544 * This is also the right place to notify the Archiver that the
1545 * segment is ready to copy to archival storage, and to update the
1546 * timer for archive_timeout, and to signal for a checkpoint if
1547 * too many logfile segments have been used since the last
1550 if (finishing_seg || (xlog_switch && last_iteration))
1553 LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
1555 if (XLogArchivingActive())
1556 XLogArchiveNotifySeg(openLogId, openLogSeg);
1558 Write->lastSegSwitchTime = (pg_time_t) time(NULL);
1561 * Signal bgwriter to start a checkpoint if we've consumed too
1562 * much xlog since the last one. For speed, we first check
1563 * using the local copy of RedoRecPtr, which might be out of
1564 * date; if it looks like a checkpoint is needed, forcibly
1565 * update RedoRecPtr and recheck.
1567 if (IsUnderPostmaster &&
1568 XLogCheckpointNeeded())
1570 (void) GetRedoRecPtr();
1571 if (XLogCheckpointNeeded())
1572 RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
1579 /* Only asked to write a partial page */
1580 LogwrtResult.Write = WriteRqst.Write;
1583 curridx = NextBufIdx(curridx);
1585 /* If flexible, break out of loop as soon as we wrote something */
1586 if (flexible && npages == 0)
1590 Assert(npages == 0);
1591 Assert(curridx == Write->curridx);
1594 * If asked to flush, do so
1596 if (XLByteLT(LogwrtResult.Flush, WriteRqst.Flush) &&
1597 XLByteLT(LogwrtResult.Flush, LogwrtResult.Write))
1600 * Could get here without iterating above loop, in which case we might
1601 * have no open file or the wrong one. However, we do not need to
1602 * fsync more than one file.
1604 if (sync_method != SYNC_METHOD_OPEN)
1606 if (openLogFile >= 0 &&
1607 !XLByteInPrevSeg(LogwrtResult.Write, openLogId, openLogSeg))
1609 if (openLogFile < 0)
1611 XLByteToPrevSeg(LogwrtResult.Write, openLogId, openLogSeg);
1612 openLogFile = XLogFileOpen(openLogId, openLogSeg);
1617 LogwrtResult.Flush = LogwrtResult.Write;
1621 * Update shared-memory status
1623 * We make sure that the shared 'request' values do not fall behind the
1624 * 'result' values. This is not absolutely essential, but it saves some
1625 * code in a couple of places.
1628 /* use volatile pointer to prevent code rearrangement */
1629 volatile XLogCtlData *xlogctl = XLogCtl;
1631 SpinLockAcquire(&xlogctl->info_lck);
1632 xlogctl->LogwrtResult = LogwrtResult;
1633 if (XLByteLT(xlogctl->LogwrtRqst.Write, LogwrtResult.Write))
1634 xlogctl->LogwrtRqst.Write = LogwrtResult.Write;
1635 if (XLByteLT(xlogctl->LogwrtRqst.Flush, LogwrtResult.Flush))
1636 xlogctl->LogwrtRqst.Flush = LogwrtResult.Flush;
1637 SpinLockRelease(&xlogctl->info_lck);
1640 Write->LogwrtResult = LogwrtResult;
1644 * Record the LSN for an asynchronous transaction commit.
1645 * (This should not be called for aborts, nor for synchronous commits.)
1648 XLogSetAsyncCommitLSN(XLogRecPtr asyncCommitLSN)
1650 /* use volatile pointer to prevent code rearrangement */
1651 volatile XLogCtlData *xlogctl = XLogCtl;
1653 SpinLockAcquire(&xlogctl->info_lck);
1654 if (XLByteLT(xlogctl->asyncCommitLSN, asyncCommitLSN))
1655 xlogctl->asyncCommitLSN = asyncCommitLSN;
1656 SpinLockRelease(&xlogctl->info_lck);
1660 * Ensure that all XLOG data through the given position is flushed to disk.
1662 * NOTE: this differs from XLogWrite mainly in that the WALWriteLock is not
1663 * already held, and we try to avoid acquiring it if possible.
1666 XLogFlush(XLogRecPtr record)
1668 XLogRecPtr WriteRqstPtr;
1669 XLogwrtRqst WriteRqst;
1671 /* Disabled during REDO */
1675 /* Quick exit if already known flushed */
1676 if (XLByteLE(record, LogwrtResult.Flush))
1681 elog(LOG, "xlog flush request %X/%X; write %X/%X; flush %X/%X",
1682 record.xlogid, record.xrecoff,
1683 LogwrtResult.Write.xlogid, LogwrtResult.Write.xrecoff,
1684 LogwrtResult.Flush.xlogid, LogwrtResult.Flush.xrecoff);
1687 START_CRIT_SECTION();
1690 * Since fsync is usually a horribly expensive operation, we try to
1691 * piggyback as much data as we can on each fsync: if we see any more data
1692 * entered into the xlog buffer, we'll write and fsync that too, so that
1693 * the final value of LogwrtResult.Flush is as large as possible. This
1694 * gives us some chance of avoiding another fsync immediately after.
1697 /* initialize to given target; may increase below */
1698 WriteRqstPtr = record;
1700 /* read LogwrtResult and update local state */
1702 /* use volatile pointer to prevent code rearrangement */
1703 volatile XLogCtlData *xlogctl = XLogCtl;
1705 SpinLockAcquire(&xlogctl->info_lck);
1706 if (XLByteLT(WriteRqstPtr, xlogctl->LogwrtRqst.Write))
1707 WriteRqstPtr = xlogctl->LogwrtRqst.Write;
1708 LogwrtResult = xlogctl->LogwrtResult;
1709 SpinLockRelease(&xlogctl->info_lck);
1713 if (!XLByteLE(record, LogwrtResult.Flush))
1715 /* now wait for the write lock */
1716 LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
1717 LogwrtResult = XLogCtl->Write.LogwrtResult;
1718 if (!XLByteLE(record, LogwrtResult.Flush))
1720 /* try to write/flush later additions to XLOG as well */
1721 if (LWLockConditionalAcquire(WALInsertLock, LW_EXCLUSIVE))
1723 XLogCtlInsert *Insert = &XLogCtl->Insert;
1724 uint32 freespace = INSERT_FREESPACE(Insert);
1726 if (freespace < SizeOfXLogRecord) /* buffer is full */
1727 WriteRqstPtr = XLogCtl->xlblocks[Insert->curridx];
1730 WriteRqstPtr = XLogCtl->xlblocks[Insert->curridx];
1731 WriteRqstPtr.xrecoff -= freespace;
1733 LWLockRelease(WALInsertLock);
1734 WriteRqst.Write = WriteRqstPtr;
1735 WriteRqst.Flush = WriteRqstPtr;
1739 WriteRqst.Write = WriteRqstPtr;
1740 WriteRqst.Flush = record;
1742 XLogWrite(WriteRqst, false, false);
1744 LWLockRelease(WALWriteLock);
1750 * If we still haven't flushed to the request point then we have a
1751 * problem; most likely, the requested flush point is past end of XLOG.
1752 * This has been seen to occur when a disk page has a corrupted LSN.
1754 * Formerly we treated this as a PANIC condition, but that hurts the
1755 * system's robustness rather than helping it: we do not want to take down
1756 * the whole system due to corruption on one data page. In particular, if
1757 * the bad page is encountered again during recovery then we would be
1758 * unable to restart the database at all! (This scenario has actually
1759 * happened in the field several times with 7.1 releases. Note that we
1760 * cannot get here while InRedo is true, but if the bad page is brought in
1761 * and marked dirty during recovery then CreateCheckPoint will try to
1762 * flush it at the end of recovery.)
1764 * The current approach is to ERROR under normal conditions, but only
1765 * WARNING during recovery, so that the system can be brought up even if
1766 * there's a corrupt LSN. Note that for calls from xact.c, the ERROR will
1767 * be promoted to PANIC since xact.c calls this routine inside a critical
1768 * section. However, calls from bufmgr.c are not within critical sections
1769 * and so we will not force a restart for a bad LSN on a data page.
1771 if (XLByteLT(LogwrtResult.Flush, record))
1772 elog(InRecovery ? WARNING : ERROR,
1773 "xlog flush request %X/%X is not satisfied --- flushed only to %X/%X",
1774 record.xlogid, record.xrecoff,
1775 LogwrtResult.Flush.xlogid, LogwrtResult.Flush.xrecoff);
1779 * Flush xlog, but without specifying exactly where to flush to.
1781 * We normally flush only completed blocks; but if there is nothing to do on
1782 * that basis, we check for unflushed async commits in the current incomplete
1783 * block, and flush through the latest one of those. Thus, if async commits
1784 * are not being used, we will flush complete blocks only. We can guarantee
1785 * that async commits reach disk after at most three cycles; normally only
1786 * one or two. (We allow XLogWrite to write "flexibly", meaning it can stop
1787 * at the end of the buffer ring; this makes a difference only with very high
1788 * load or long wal_writer_delay, but imposes one extra cycle for the worst
1789 * case for async commits.)
1791 * This routine is invoked periodically by the background walwriter process.
1794 XLogBackgroundFlush(void)
1796 XLogRecPtr WriteRqstPtr;
1797 bool flexible = true;
1799 /* read LogwrtResult and update local state */
1801 /* use volatile pointer to prevent code rearrangement */
1802 volatile XLogCtlData *xlogctl = XLogCtl;
1804 SpinLockAcquire(&xlogctl->info_lck);
1805 LogwrtResult = xlogctl->LogwrtResult;
1806 WriteRqstPtr = xlogctl->LogwrtRqst.Write;
1807 SpinLockRelease(&xlogctl->info_lck);
1810 /* back off to last completed page boundary */
1811 WriteRqstPtr.xrecoff -= WriteRqstPtr.xrecoff % XLOG_BLCKSZ;
1813 /* if we have already flushed that far, consider async commit records */
1814 if (XLByteLE(WriteRqstPtr, LogwrtResult.Flush))
1816 /* use volatile pointer to prevent code rearrangement */
1817 volatile XLogCtlData *xlogctl = XLogCtl;
1819 SpinLockAcquire(&xlogctl->info_lck);
1820 WriteRqstPtr = xlogctl->asyncCommitLSN;
1821 SpinLockRelease(&xlogctl->info_lck);
1822 flexible = false; /* ensure it all gets written */
1825 /* Done if already known flushed */
1826 if (XLByteLE(WriteRqstPtr, LogwrtResult.Flush))
1831 elog(LOG, "xlog bg flush request %X/%X; write %X/%X; flush %X/%X",
1832 WriteRqstPtr.xlogid, WriteRqstPtr.xrecoff,
1833 LogwrtResult.Write.xlogid, LogwrtResult.Write.xrecoff,
1834 LogwrtResult.Flush.xlogid, LogwrtResult.Flush.xrecoff);
1837 START_CRIT_SECTION();
1839 /* now wait for the write lock */
1840 LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
1841 LogwrtResult = XLogCtl->Write.LogwrtResult;
1842 if (!XLByteLE(WriteRqstPtr, LogwrtResult.Flush))
1844 XLogwrtRqst WriteRqst;
1846 WriteRqst.Write = WriteRqstPtr;
1847 WriteRqst.Flush = WriteRqstPtr;
1848 XLogWrite(WriteRqst, flexible, false);
1850 LWLockRelease(WALWriteLock);
1856 * Flush any previous asynchronously-committed transactions' commit records.
1858 * NOTE: it is unwise to assume that this provides any strong guarantees.
1859 * In particular, because of the inexact LSN bookkeeping used by clog.c,
1860 * we cannot assume that hint bits will be settable for these transactions.
1863 XLogAsyncCommitFlush(void)
1865 XLogRecPtr WriteRqstPtr;
1867 /* use volatile pointer to prevent code rearrangement */
1868 volatile XLogCtlData *xlogctl = XLogCtl;
1870 SpinLockAcquire(&xlogctl->info_lck);
1871 WriteRqstPtr = xlogctl->asyncCommitLSN;
1872 SpinLockRelease(&xlogctl->info_lck);
1874 XLogFlush(WriteRqstPtr);
1878 * Test whether XLOG data has been flushed up to (at least) the given position.
1880 * Returns true if a flush is still needed. (It may be that someone else
1881 * is already in process of flushing that far, however.)
1884 XLogNeedsFlush(XLogRecPtr record)
1886 /* Quick exit if already known flushed */
1887 if (XLByteLE(record, LogwrtResult.Flush))
1890 /* read LogwrtResult and update local state */
1892 /* use volatile pointer to prevent code rearrangement */
1893 volatile XLogCtlData *xlogctl = XLogCtl;
1895 SpinLockAcquire(&xlogctl->info_lck);
1896 LogwrtResult = xlogctl->LogwrtResult;
1897 SpinLockRelease(&xlogctl->info_lck);
1901 if (XLByteLE(record, LogwrtResult.Flush))
1908 * Create a new XLOG file segment, or open a pre-existing one.
1910 * log, seg: identify segment to be created/opened.
1912 * *use_existent: if TRUE, OK to use a pre-existing file (else, any
1913 * pre-existing file will be deleted). On return, TRUE if a pre-existing
1916 * use_lock: if TRUE, acquire ControlFileLock while moving file into
1917 * place. This should be TRUE except during bootstrap log creation. The
1918 * caller must *not* hold the lock at call.
1920 * Returns FD of opened file.
1922 * Note: errors here are ERROR not PANIC because we might or might not be
1923 * inside a critical section (eg, during checkpoint there is no reason to
1924 * take down the system on failure). They will promote to PANIC if we are
1925 * in a critical section.
1928 XLogFileInit(uint32 log, uint32 seg,
1929 bool *use_existent, bool use_lock)
1931 char path[MAXPGPATH];
1932 char tmppath[MAXPGPATH];
1934 uint32 installed_log;
1935 uint32 installed_seg;
1940 XLogFilePath(path, ThisTimeLineID, log, seg);
1943 * Try to use existent file (checkpoint maker may have created it already)
1947 fd = BasicOpenFile(path, O_RDWR | PG_BINARY | XLOG_SYNC_BIT,
1951 if (errno != ENOENT)
1953 (errcode_for_file_access(),
1954 errmsg("could not open file \"%s\" (log file %u, segment %u): %m",
1962 * Initialize an empty (all zeroes) segment. NOTE: it is possible that
1963 * another process is doing the same thing. If so, we will end up
1964 * pre-creating an extra log segment. That seems OK, and better than
1965 * holding the lock throughout this lengthy process.
1967 elog(DEBUG2, "creating and filling new WAL file");
1969 snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
1973 /* do not use XLOG_SYNC_BIT here --- want to fsync only at end of fill */
1974 fd = BasicOpenFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
1978 (errcode_for_file_access(),
1979 errmsg("could not create file \"%s\": %m", tmppath)));
1982 * Zero-fill the file. We have to do this the hard way to ensure that all
1983 * the file space has really been allocated --- on platforms that allow
1984 * "holes" in files, just seeking to the end doesn't allocate intermediate
1985 * space. This way, we know that we have all the space and (after the
1986 * fsync below) that all the indirect blocks are down on disk. Therefore,
1987 * fdatasync(2) or O_DSYNC will be sufficient to sync future writes to the
1990 * Note: palloc zbuffer, instead of just using a local char array, to
1991 * ensure it is reasonably well-aligned; this may save a few cycles
1992 * transferring data to the kernel.
1994 zbuffer = (char *) palloc0(XLOG_BLCKSZ);
1995 for (nbytes = 0; nbytes < XLogSegSize; nbytes += XLOG_BLCKSZ)
1998 if ((int) write(fd, zbuffer, XLOG_BLCKSZ) != (int) XLOG_BLCKSZ)
2000 int save_errno = errno;
2003 * If we fail to make the file, delete it to release disk space
2006 /* if write didn't set errno, assume problem is no disk space */
2007 errno = save_errno ? save_errno : ENOSPC;
2010 (errcode_for_file_access(),
2011 errmsg("could not write to file \"%s\": %m", tmppath)));
2016 if (pg_fsync(fd) != 0)
2018 (errcode_for_file_access(),
2019 errmsg("could not fsync file \"%s\": %m", tmppath)));
2023 (errcode_for_file_access(),
2024 errmsg("could not close file \"%s\": %m", tmppath)));
2027 * Now move the segment into place with its final name.
2029 * If caller didn't want to use a pre-existing file, get rid of any
2030 * pre-existing file. Otherwise, cope with possibility that someone else
2031 * has created the file while we were filling ours: if so, use ours to
2032 * pre-create a future log segment.
2034 installed_log = log;
2035 installed_seg = seg;
2036 max_advance = XLOGfileslop;
2037 if (!InstallXLogFileSegment(&installed_log, &installed_seg, tmppath,
2038 *use_existent, &max_advance,
2041 /* No need for any more future segments... */
2045 elog(DEBUG2, "done creating and filling new WAL file");
2047 /* Set flag to tell caller there was no existent file */
2048 *use_existent = false;
2050 /* Now open original target segment (might not be file I just made) */
2051 fd = BasicOpenFile(path, O_RDWR | PG_BINARY | XLOG_SYNC_BIT,
2055 (errcode_for_file_access(),
2056 errmsg("could not open file \"%s\" (log file %u, segment %u): %m",
2063 * Create a new XLOG file segment by copying a pre-existing one.
2065 * log, seg: identify segment to be created.
2067 * srcTLI, srclog, srcseg: identify segment to be copied (could be from
2068 * a different timeline)
2070 * Currently this is only used during recovery, and so there are no locking
2071 * considerations. But we should be just as tense as XLogFileInit to avoid
2072 * emplacing a bogus file.
2075 XLogFileCopy(uint32 log, uint32 seg,
2076 TimeLineID srcTLI, uint32 srclog, uint32 srcseg)
2078 char path[MAXPGPATH];
2079 char tmppath[MAXPGPATH];
2080 char buffer[XLOG_BLCKSZ];
2086 * Open the source file
2088 XLogFilePath(path, srcTLI, srclog, srcseg);
2089 srcfd = BasicOpenFile(path, O_RDONLY | PG_BINARY, 0);
2092 (errcode_for_file_access(),
2093 errmsg("could not open file \"%s\": %m", path)));
2096 * Copy into a temp file name.
2098 snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
2102 /* do not use XLOG_SYNC_BIT here --- want to fsync only at end of fill */
2103 fd = BasicOpenFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
2107 (errcode_for_file_access(),
2108 errmsg("could not create file \"%s\": %m", tmppath)));
2111 * Do the data copying.
2113 for (nbytes = 0; nbytes < XLogSegSize; nbytes += sizeof(buffer))
2116 if ((int) read(srcfd, buffer, sizeof(buffer)) != (int) sizeof(buffer))
2120 (errcode_for_file_access(),
2121 errmsg("could not read file \"%s\": %m", path)));
2124 (errmsg("not enough data in file \"%s\"", path)));
2127 if ((int) write(fd, buffer, sizeof(buffer)) != (int) sizeof(buffer))
2129 int save_errno = errno;
2132 * If we fail to make the file, delete it to release disk space
2135 /* if write didn't set errno, assume problem is no disk space */
2136 errno = save_errno ? save_errno : ENOSPC;
2139 (errcode_for_file_access(),
2140 errmsg("could not write to file \"%s\": %m", tmppath)));
2144 if (pg_fsync(fd) != 0)
2146 (errcode_for_file_access(),
2147 errmsg("could not fsync file \"%s\": %m", tmppath)));
2151 (errcode_for_file_access(),
2152 errmsg("could not close file \"%s\": %m", tmppath)));
2157 * Now move the segment into place with its final name.
2159 if (!InstallXLogFileSegment(&log, &seg, tmppath, false, NULL, false))
2160 elog(ERROR, "InstallXLogFileSegment should not have failed");
2164 * Install a new XLOG segment file as a current or future log segment.
2166 * This is used both to install a newly-created segment (which has a temp
2167 * filename while it's being created) and to recycle an old segment.
2169 * *log, *seg: identify segment to install as (or first possible target).
2170 * When find_free is TRUE, these are modified on return to indicate the
2171 * actual installation location or last segment searched.
2173 * tmppath: initial name of file to install. It will be renamed into place.
2175 * find_free: if TRUE, install the new segment at the first empty log/seg
2176 * number at or after the passed numbers. If FALSE, install the new segment
2177 * exactly where specified, deleting any existing segment file there.
2179 * *max_advance: maximum number of log/seg slots to advance past the starting
2180 * point. Fail if no free slot is found in this range. On return, reduced
2181 * by the number of slots skipped over. (Irrelevant, and may be NULL,
2182 * when find_free is FALSE.)
2184 * use_lock: if TRUE, acquire ControlFileLock while moving file into
2185 * place. This should be TRUE except during bootstrap log creation. The
2186 * caller must *not* hold the lock at call.
2188 * Returns TRUE if file installed, FALSE if not installed because of
2189 * exceeding max_advance limit. On Windows, we also return FALSE if we
2190 * can't rename the file into place because someone's got it open.
2191 * (Any other kind of failure causes ereport().)
2194 InstallXLogFileSegment(uint32 *log, uint32 *seg, char *tmppath,
2195 bool find_free, int *max_advance,
2198 char path[MAXPGPATH];
2199 struct stat stat_buf;
2201 XLogFilePath(path, ThisTimeLineID, *log, *seg);
2204 * We want to be sure that only one process does this at a time.
2207 LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2211 /* Force installation: get rid of any pre-existing segment file */
2216 /* Find a free slot to put it in */
2217 while (stat(path, &stat_buf) == 0)
2219 if (*max_advance <= 0)
2221 /* Failed to find a free slot within specified range */
2223 LWLockRelease(ControlFileLock);
2226 NextLogSeg(*log, *seg);
2228 XLogFilePath(path, ThisTimeLineID, *log, *seg);
2233 * Prefer link() to rename() here just to be really sure that we don't
2234 * overwrite an existing logfile. However, there shouldn't be one, so
2235 * rename() is an acceptable substitute except for the truly paranoid.
2237 #if HAVE_WORKING_LINK
2238 if (link(tmppath, path) < 0)
2240 (errcode_for_file_access(),
2241 errmsg("could not link file \"%s\" to \"%s\" (initialization of log file %u, segment %u): %m",
2242 tmppath, path, *log, *seg)));
2245 if (rename(tmppath, path) < 0)
2248 #if !defined(__CYGWIN__)
2249 if (GetLastError() == ERROR_ACCESS_DENIED)
2251 if (errno == EACCES)
2255 LWLockRelease(ControlFileLock);
2261 (errcode_for_file_access(),
2262 errmsg("could not rename file \"%s\" to \"%s\" (initialization of log file %u, segment %u): %m",
2263 tmppath, path, *log, *seg)));
2268 LWLockRelease(ControlFileLock);
2274 * Open a pre-existing logfile segment for writing.
2277 XLogFileOpen(uint32 log, uint32 seg)
2279 char path[MAXPGPATH];
2282 XLogFilePath(path, ThisTimeLineID, log, seg);
2284 fd = BasicOpenFile(path, O_RDWR | PG_BINARY | XLOG_SYNC_BIT,
2288 (errcode_for_file_access(),
2289 errmsg("could not open file \"%s\" (log file %u, segment %u): %m",
2296 * Open a logfile segment for reading (during recovery).
2299 XLogFileRead(uint32 log, uint32 seg, int emode)
2301 char path[MAXPGPATH];
2302 char xlogfname[MAXFNAMELEN];
2303 char activitymsg[MAXFNAMELEN + 16];
2308 * Loop looking for a suitable timeline ID: we might need to read any of
2309 * the timelines listed in expectedTLIs.
2311 * We expect curFileTLI on entry to be the TLI of the preceding file in
2312 * sequence, or 0 if there was no predecessor. We do not allow curFileTLI
2313 * to go backwards; this prevents us from picking up the wrong file when a
2314 * parent timeline extends to higher segment numbers than the child we
2317 foreach(cell, expectedTLIs)
2319 TimeLineID tli = (TimeLineID) lfirst_int(cell);
2321 if (tli < curFileTLI)
2322 break; /* don't bother looking at too-old TLIs */
2324 XLogFileName(xlogfname, tli, log, seg);
2326 if (InArchiveRecovery)
2328 /* Report recovery progress in PS display */
2329 snprintf(activitymsg, sizeof(activitymsg), "waiting for %s",
2331 set_ps_display(activitymsg, false);
2333 restoredFromArchive = RestoreArchivedFile(path, xlogfname,
2338 XLogFilePath(path, tli, log, seg);
2340 fd = BasicOpenFile(path, O_RDONLY | PG_BINARY, 0);
2346 /* Report recovery progress in PS display */
2347 snprintf(activitymsg, sizeof(activitymsg), "recovering %s",
2349 set_ps_display(activitymsg, false);
2353 if (errno != ENOENT) /* unexpected failure? */
2355 (errcode_for_file_access(),
2356 errmsg("could not open file \"%s\" (log file %u, segment %u): %m",
2360 /* Couldn't find it. For simplicity, complain about front timeline */
2361 XLogFilePath(path, recoveryTargetTLI, log, seg);
2364 (errcode_for_file_access(),
2365 errmsg("could not open file \"%s\" (log file %u, segment %u): %m",
2371 * Close the current logfile segment for writing.
2376 Assert(openLogFile >= 0);
2379 * posix_fadvise is problematic on many platforms: on older x86 Linux it
2380 * just dumps core, and there are reports of problems on PPC platforms as
2381 * well. The following is therefore disabled for the time being. We could
2382 * consider some kind of configure test to see if it's safe to use, but
2383 * since we lack hard evidence that there's any useful performance gain to
2384 * be had, spending time on that seems unprofitable for now.
2389 * WAL segment files will not be re-read in normal operation, so we advise
2390 * OS to release any cached pages. But do not do so if WAL archiving is
2391 * active, because archiver process could use the cache to read the WAL
2394 * While O_DIRECT works for O_SYNC, posix_fadvise() works for fsync() and
2395 * O_SYNC, and some platforms only have posix_fadvise().
2397 #if defined(HAVE_DECL_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
2398 if (!XLogArchivingActive())
2399 posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
2401 #endif /* NOT_USED */
2403 if (close(openLogFile))
2405 (errcode_for_file_access(),
2406 errmsg("could not close log file %u, segment %u: %m",
2407 openLogId, openLogSeg)));
2412 * Attempt to retrieve the specified file from off-line archival storage.
2413 * If successful, fill "path" with its complete path (note that this will be
2414 * a temp file name that doesn't follow the normal naming convention), and
2417 * If not successful, fill "path" with the name of the normal on-line file
2418 * (which may or may not actually exist, but we'll try to use it), and return
2421 * For fixed-size files, the caller may pass the expected size as an
2422 * additional crosscheck on successful recovery. If the file size is not
2423 * known, set expectedSize = 0.
2426 RestoreArchivedFile(char *path, const char *xlogfname,
2427 const char *recovername, off_t expectedSize)
2429 char xlogpath[MAXPGPATH];
2430 char xlogRestoreCmd[MAXPGPATH];
2431 char lastRestartPointFname[MAXPGPATH];
2437 struct stat stat_buf;
2442 * When doing archive recovery, we always prefer an archived log file even
2443 * if a file of the same name exists in XLOGDIR. The reason is that the
2444 * file in XLOGDIR could be an old, un-filled or partly-filled version
2445 * that was copied and restored as part of backing up $PGDATA.
2447 * We could try to optimize this slightly by checking the local copy
2448 * lastchange timestamp against the archived copy, but we have no API to
2449 * do this, nor can we guarantee that the lastchange timestamp was
2450 * preserved correctly when we copied to archive. Our aim is robustness,
2451 * so we elect not to do this.
2453 * If we cannot obtain the log file from the archive, however, we will try
2454 * to use the XLOGDIR file if it exists. This is so that we can make use
2455 * of log segments that weren't yet transferred to the archive.
2457 * Notice that we don't actually overwrite any files when we copy back
2458 * from archive because the recoveryRestoreCommand may inadvertently
2459 * restore inappropriate xlogs, or they may be corrupt, so we may wish to
2460 * fallback to the segments remaining in current XLOGDIR later. The
2461 * copy-from-archive filename is always the same, ensuring that we don't
2462 * run out of disk space on long recoveries.
2464 snprintf(xlogpath, MAXPGPATH, XLOGDIR "/%s", recovername);
2467 * Make sure there is no existing file named recovername.
2469 if (stat(xlogpath, &stat_buf) != 0)
2471 if (errno != ENOENT)
2473 (errcode_for_file_access(),
2474 errmsg("could not stat file \"%s\": %m",
2479 if (unlink(xlogpath) != 0)
2481 (errcode_for_file_access(),
2482 errmsg("could not remove file \"%s\": %m",
2487 * Calculate the archive file cutoff point for use during log shipping
2488 * replication. All files earlier than this point can be deleted
2489 * from the archive, though there is no requirement to do so.
2491 * We initialise this with the filename of an InvalidXLogRecPtr, which
2492 * will prevent the deletion of any WAL files from the archive
2493 * because of the alphabetic sorting property of WAL filenames.
2495 * Once we have successfully located the redo pointer of the checkpoint
2496 * from which we start recovery we never request a file prior to the redo
2497 * pointer of the last restartpoint. When redo begins we know that we
2498 * have successfully located it, so there is no need for additional
2499 * status flags to signify the point when we can begin deleting WAL files
2504 XLByteToSeg(ControlFile->checkPointCopy.redo,
2505 restartLog, restartSeg);
2506 XLogFileName(lastRestartPointFname,
2507 ControlFile->checkPointCopy.ThisTimeLineID,
2508 restartLog, restartSeg);
2509 /* we shouldn't need anything earlier than last restart point */
2510 Assert(strcmp(lastRestartPointFname, xlogfname) < 0);
2513 XLogFileName(lastRestartPointFname, 0, 0, 0);
2516 * construct the command to be executed
2518 dp = xlogRestoreCmd;
2519 endp = xlogRestoreCmd + MAXPGPATH - 1;
2522 for (sp = recoveryRestoreCommand; *sp; sp++)
2529 /* %p: relative path of target file */
2531 StrNCpy(dp, xlogpath, endp - dp);
2532 make_native_path(dp);
2536 /* %f: filename of desired file */
2538 StrNCpy(dp, xlogfname, endp - dp);
2542 /* %r: filename of last restartpoint */
2544 StrNCpy(dp, lastRestartPointFname, endp - dp);
2548 /* convert %% to a single % */
2554 /* otherwise treat the % as not special */
2569 (errmsg_internal("executing restore command \"%s\"",
2573 * Copy xlog from archival storage to XLOGDIR
2575 rc = system(xlogRestoreCmd);
2579 * command apparently succeeded, but let's make sure the file is
2580 * really there now and has the correct size.
2582 * XXX I made wrong-size a fatal error to ensure the DBA would notice
2583 * it, but is that too strong? We could try to plow ahead with a
2584 * local copy of the file ... but the problem is that there probably
2585 * isn't one, and we'd incorrectly conclude we've reached the end of
2586 * WAL and we're done recovering ...
2588 if (stat(xlogpath, &stat_buf) == 0)
2590 if (expectedSize > 0 && stat_buf.st_size != expectedSize)
2592 (errmsg("archive file \"%s\" has wrong size: %lu instead of %lu",
2594 (unsigned long) stat_buf.st_size,
2595 (unsigned long) expectedSize)));
2599 (errmsg("restored log file \"%s\" from archive",
2601 strcpy(path, xlogpath);
2608 if (errno != ENOENT)
2610 (errcode_for_file_access(),
2611 errmsg("could not stat file \"%s\": %m",
2617 * Remember, we rollforward UNTIL the restore fails so failure here is
2618 * just part of the process... that makes it difficult to determine
2619 * whether the restore failed because there isn't an archive to restore,
2620 * or because the administrator has specified the restore program
2621 * incorrectly. We have to assume the former.
2623 * However, if the failure was due to any sort of signal, it's best to
2624 * punt and abort recovery. (If we "return false" here, upper levels will
2625 * assume that recovery is complete and start up the database!) It's
2626 * essential to abort on child SIGINT and SIGQUIT, because per spec
2627 * system() ignores SIGINT and SIGQUIT while waiting; if we see one of
2628 * those it's a good bet we should have gotten it too. Aborting on other
2629 * signals such as SIGTERM seems a good idea as well.
2631 * Per the Single Unix Spec, shells report exit status > 128 when a called
2632 * command died on a signal. Also, 126 and 127 are used to report
2633 * problems such as an unfindable command; treat those as fatal errors
2636 signaled = WIFSIGNALED(rc) || WEXITSTATUS(rc) > 125;
2638 ereport(signaled ? FATAL : DEBUG2,
2639 (errmsg("could not restore file \"%s\" from archive: return code %d",
2643 * if an archived file is not available, there might still be a version of
2644 * this file in XLOGDIR, so return that as the filename to open.
2646 * In many recovery scenarios we expect this to fail also, but if so that
2647 * just means we've reached the end of WAL.
2649 snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlogfname);
2654 * Preallocate log files beyond the specified log endpoint.
2656 * XXX this is currently extremely conservative, since it forces only one
2657 * future log segment to exist, and even that only if we are 75% done with
2658 * the current one. This is only appropriate for very low-WAL-volume systems.
2659 * High-volume systems will be OK once they've built up a sufficient set of
2660 * recycled log segments, but the startup transient is likely to include
2661 * a lot of segment creations by foreground processes, which is not so good.
2664 PreallocXlogFiles(XLogRecPtr endptr)
2671 XLByteToPrevSeg(endptr, _logId, _logSeg);
2672 if ((endptr.xrecoff - 1) % XLogSegSize >=
2673 (uint32) (0.75 * XLogSegSize))
2675 NextLogSeg(_logId, _logSeg);
2676 use_existent = true;
2677 lf = XLogFileInit(_logId, _logSeg, &use_existent, true);
2680 CheckpointStats.ckpt_segs_added++;
2685 * Recycle or remove all log files older or equal to passed log/seg#
2687 * endptr is current (or recent) end of xlog; this is used to determine
2688 * whether we want to recycle rather than delete no-longer-wanted log files.
2691 RemoveOldXlogFiles(uint32 log, uint32 seg, XLogRecPtr endptr)
2697 struct dirent *xlde;
2698 char lastoff[MAXFNAMELEN];
2699 char path[MAXPGPATH];
2702 * Initialize info about where to try to recycle to. We allow recycling
2703 * segments up to XLOGfileslop segments beyond the current XLOG location.
2705 XLByteToPrevSeg(endptr, endlogId, endlogSeg);
2706 max_advance = XLOGfileslop;
2708 xldir = AllocateDir(XLOGDIR);
2711 (errcode_for_file_access(),
2712 errmsg("could not open transaction log directory \"%s\": %m",
2715 XLogFileName(lastoff, ThisTimeLineID, log, seg);
2717 while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
2720 * We ignore the timeline part of the XLOG segment identifiers in
2721 * deciding whether a segment is still needed. This ensures that we
2722 * won't prematurely remove a segment from a parent timeline. We could
2723 * probably be a little more proactive about removing segments of
2724 * non-parent timelines, but that would be a whole lot more
2727 * We use the alphanumeric sorting property of the filenames to decide
2728 * which ones are earlier than the lastoff segment.
2730 if (strlen(xlde->d_name) == 24 &&
2731 strspn(xlde->d_name, "0123456789ABCDEF") == 24 &&
2732 strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
2734 if (XLogArchiveCheckDone(xlde->d_name, true))
2736 snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
2739 * Before deleting the file, see if it can be recycled as a
2740 * future log segment.
2742 if (InstallXLogFileSegment(&endlogId, &endlogSeg, path,
2747 (errmsg("recycled transaction log file \"%s\"",
2749 CheckpointStats.ckpt_segs_recycled++;
2750 /* Needn't recheck that slot on future iterations */
2751 if (max_advance > 0)
2753 NextLogSeg(endlogId, endlogSeg);
2759 /* No need for any more future segments... */
2761 (errmsg("removing transaction log file \"%s\"",
2764 CheckpointStats.ckpt_segs_removed++;
2767 XLogArchiveCleanup(xlde->d_name);
2776 * Remove previous backup history files. This also retries creation of
2777 * .ready files for any backup history files for which XLogArchiveNotify
2781 CleanupBackupHistory(void)
2784 struct dirent *xlde;
2785 char path[MAXPGPATH];
2787 xldir = AllocateDir(XLOGDIR);
2790 (errcode_for_file_access(),
2791 errmsg("could not open transaction log directory \"%s\": %m",
2794 while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
2796 if (strlen(xlde->d_name) > 24 &&
2797 strspn(xlde->d_name, "0123456789ABCDEF") == 24 &&
2798 strcmp(xlde->d_name + strlen(xlde->d_name) - strlen(".backup"),
2801 if (XLogArchiveCheckDone(xlde->d_name, true))
2804 (errmsg("removing transaction log backup history file \"%s\"",
2806 snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
2808 XLogArchiveCleanup(xlde->d_name);
2817 * Restore the backup blocks present in an XLOG record, if any.
2819 * We assume all of the record has been read into memory at *record.
2821 * Note: when a backup block is available in XLOG, we restore it
2822 * unconditionally, even if the page in the database appears newer.
2823 * This is to protect ourselves against database pages that were partially
2824 * or incorrectly written during a crash. We assume that the XLOG data
2825 * must be good because it has passed a CRC check, while the database
2826 * page might not be. This will force us to replay all subsequent
2827 * modifications of the page that appear in XLOG, rather than possibly
2828 * ignoring them as already applied, but that's not a huge drawback.
2831 RestoreBkpBlocks(XLogRecord *record, XLogRecPtr lsn)
2840 blk = (char *) XLogRecGetData(record) + record->xl_len;
2841 for (i = 0; i < XLR_MAX_BKP_BLOCKS; i++)
2843 if (!(record->xl_info & XLR_SET_BKP_BLOCK(i)))
2846 memcpy(&bkpb, blk, sizeof(BkpBlock));
2847 blk += sizeof(BkpBlock);
2849 reln = XLogOpenRelation(bkpb.node);
2850 buffer = XLogReadBuffer(reln, bkpb.block, true);
2851 Assert(BufferIsValid(buffer));
2852 page = (Page) BufferGetPage(buffer);
2854 if (bkpb.hole_length == 0)
2856 memcpy((char *) page, blk, BLCKSZ);
2860 /* must zero-fill the hole */
2861 MemSet((char *) page, 0, BLCKSZ);
2862 memcpy((char *) page, blk, bkpb.hole_offset);
2863 memcpy((char *) page + (bkpb.hole_offset + bkpb.hole_length),
2864 blk + bkpb.hole_offset,
2865 BLCKSZ - (bkpb.hole_offset + bkpb.hole_length));
2868 PageSetLSN(page, lsn);
2869 PageSetTLI(page, ThisTimeLineID);
2870 MarkBufferDirty(buffer);
2871 UnlockReleaseBuffer(buffer);
2873 blk += BLCKSZ - bkpb.hole_length;
2878 * CRC-check an XLOG record. We do not believe the contents of an XLOG
2879 * record (other than to the minimal extent of computing the amount of
2880 * data to read in) until we've checked the CRCs.
2882 * We assume all of the record has been read into memory at *record.
2885 RecordIsValid(XLogRecord *record, XLogRecPtr recptr, int emode)
2889 uint32 len = record->xl_len;
2893 /* First the rmgr data */
2895 COMP_CRC32(crc, XLogRecGetData(record), len);
2897 /* Add in the backup blocks, if any */
2898 blk = (char *) XLogRecGetData(record) + len;
2899 for (i = 0; i < XLR_MAX_BKP_BLOCKS; i++)
2903 if (!(record->xl_info & XLR_SET_BKP_BLOCK(i)))
2906 memcpy(&bkpb, blk, sizeof(BkpBlock));
2907 if (bkpb.hole_offset + bkpb.hole_length > BLCKSZ)
2910 (errmsg("incorrect hole size in record at %X/%X",
2911 recptr.xlogid, recptr.xrecoff)));
2914 blen = sizeof(BkpBlock) + BLCKSZ - bkpb.hole_length;
2915 COMP_CRC32(crc, blk, blen);
2919 /* Check that xl_tot_len agrees with our calculation */
2920 if (blk != (char *) record + record->xl_tot_len)
2923 (errmsg("incorrect total length in record at %X/%X",
2924 recptr.xlogid, recptr.xrecoff)));
2928 /* Finally include the record header */
2929 COMP_CRC32(crc, (char *) record + sizeof(pg_crc32),
2930 SizeOfXLogRecord - sizeof(pg_crc32));
2933 if (!EQ_CRC32(record->xl_crc, crc))
2936 (errmsg("incorrect resource manager data checksum in record at %X/%X",
2937 recptr.xlogid, recptr.xrecoff)));
2945 * Attempt to read an XLOG record.
2947 * If RecPtr is not NULL, try to read a record at that position. Otherwise
2948 * try to read a record just after the last one previously read.
2950 * If no valid record is available, returns NULL, or fails if emode is PANIC.
2951 * (emode must be either PANIC or LOG.)
2953 * The record is copied into readRecordBuf, so that on successful return,
2954 * the returned record pointer always points there.
2957 ReadRecord(XLogRecPtr *RecPtr, int emode)
2961 XLogRecPtr tmpRecPtr = EndRecPtr;
2962 bool randAccess = false;
2965 uint32 targetPageOff;
2966 uint32 targetRecOff;
2967 uint32 pageHeaderSize;
2969 if (readBuf == NULL)
2972 * First time through, permanently allocate readBuf. We do it this
2973 * way, rather than just making a static array, for two reasons: (1)
2974 * no need to waste the storage in most instantiations of the backend;
2975 * (2) a static char array isn't guaranteed to have any particular
2976 * alignment, whereas malloc() will provide MAXALIGN'd storage.
2978 readBuf = (char *) malloc(XLOG_BLCKSZ);
2979 Assert(readBuf != NULL);
2984 RecPtr = &tmpRecPtr;
2985 /* fast case if next record is on same page */
2986 if (nextRecord != NULL)
2988 record = nextRecord;
2991 /* align old recptr to next page */
2992 if (tmpRecPtr.xrecoff % XLOG_BLCKSZ != 0)
2993 tmpRecPtr.xrecoff += (XLOG_BLCKSZ - tmpRecPtr.xrecoff % XLOG_BLCKSZ);
2994 if (tmpRecPtr.xrecoff >= XLogFileSize)
2996 (tmpRecPtr.xlogid)++;
2997 tmpRecPtr.xrecoff = 0;
2999 /* We will account for page header size below */
3003 if (!XRecOffIsValid(RecPtr->xrecoff))
3005 (errmsg("invalid record offset at %X/%X",
3006 RecPtr->xlogid, RecPtr->xrecoff)));
3009 * Since we are going to a random position in WAL, forget any prior
3010 * state about what timeline we were in, and allow it to be any
3011 * timeline in expectedTLIs. We also set a flag to allow curFileTLI
3012 * to go backwards (but we can't reset that variable right here, since
3013 * we might not change files at all).
3015 lastPageTLI = 0; /* see comment in ValidXLOGHeader */
3016 randAccess = true; /* allow curFileTLI to go backwards too */
3019 if (readFile >= 0 && !XLByteInSeg(*RecPtr, readId, readSeg))
3024 XLByteToSeg(*RecPtr, readId, readSeg);
3027 /* Now it's okay to reset curFileTLI if random fetch */
3031 readFile = XLogFileRead(readId, readSeg, emode);
3033 goto next_record_is_invalid;
3036 * Whenever switching to a new WAL segment, we read the first page of
3037 * the file and validate its header, even if that's not where the
3038 * target record is. This is so that we can check the additional
3039 * identification info that is present in the first page's "long"
3043 if (read(readFile, readBuf, XLOG_BLCKSZ) != XLOG_BLCKSZ)
3046 (errcode_for_file_access(),
3047 errmsg("could not read from log file %u, segment %u, offset %u: %m",
3048 readId, readSeg, readOff)));
3049 goto next_record_is_invalid;
3051 if (!ValidXLOGHeader((XLogPageHeader) readBuf, emode))
3052 goto next_record_is_invalid;
3055 targetPageOff = ((RecPtr->xrecoff % XLogSegSize) / XLOG_BLCKSZ) * XLOG_BLCKSZ;
3056 if (readOff != targetPageOff)
3058 readOff = targetPageOff;
3059 if (lseek(readFile, (off_t) readOff, SEEK_SET) < 0)
3062 (errcode_for_file_access(),
3063 errmsg("could not seek in log file %u, segment %u to offset %u: %m",
3064 readId, readSeg, readOff)));
3065 goto next_record_is_invalid;
3067 if (read(readFile, readBuf, XLOG_BLCKSZ) != XLOG_BLCKSZ)
3070 (errcode_for_file_access(),
3071 errmsg("could not read from log file %u, segment %u, offset %u: %m",
3072 readId, readSeg, readOff)));
3073 goto next_record_is_invalid;
3075 if (!ValidXLOGHeader((XLogPageHeader) readBuf, emode))
3076 goto next_record_is_invalid;
3078 pageHeaderSize = XLogPageHeaderSize((XLogPageHeader) readBuf);
3079 targetRecOff = RecPtr->xrecoff % XLOG_BLCKSZ;
3080 if (targetRecOff == 0)
3083 * Can only get here in the continuing-from-prev-page case, because
3084 * XRecOffIsValid eliminated the zero-page-offset case otherwise. Need
3085 * to skip over the new page's header.
3087 tmpRecPtr.xrecoff += pageHeaderSize;
3088 targetRecOff = pageHeaderSize;
3090 else if (targetRecOff < pageHeaderSize)
3093 (errmsg("invalid record offset at %X/%X",
3094 RecPtr->xlogid, RecPtr->xrecoff)));
3095 goto next_record_is_invalid;
3097 if ((((XLogPageHeader) readBuf)->xlp_info & XLP_FIRST_IS_CONTRECORD) &&
3098 targetRecOff == pageHeaderSize)
3101 (errmsg("contrecord is requested by %X/%X",
3102 RecPtr->xlogid, RecPtr->xrecoff)));
3103 goto next_record_is_invalid;
3105 record = (XLogRecord *) ((char *) readBuf + RecPtr->xrecoff % XLOG_BLCKSZ);
3110 * xl_len == 0 is bad data for everything except XLOG SWITCH, where it is
3113 if (record->xl_rmid == RM_XLOG_ID && record->xl_info == XLOG_SWITCH)
3115 if (record->xl_len != 0)
3118 (errmsg("invalid xlog switch record at %X/%X",
3119 RecPtr->xlogid, RecPtr->xrecoff)));
3120 goto next_record_is_invalid;
3123 else if (record->xl_len == 0)
3126 (errmsg("record with zero length at %X/%X",
3127 RecPtr->xlogid, RecPtr->xrecoff)));
3128 goto next_record_is_invalid;
3130 if (record->xl_tot_len < SizeOfXLogRecord + record->xl_len ||
3131 record->xl_tot_len > SizeOfXLogRecord + record->xl_len +
3132 XLR_MAX_BKP_BLOCKS * (sizeof(BkpBlock) + BLCKSZ))
3135 (errmsg("invalid record length at %X/%X",
3136 RecPtr->xlogid, RecPtr->xrecoff)));
3137 goto next_record_is_invalid;
3139 if (record->xl_rmid > RM_MAX_ID)
3142 (errmsg("invalid resource manager ID %u at %X/%X",
3143 record->xl_rmid, RecPtr->xlogid, RecPtr->xrecoff)));
3144 goto next_record_is_invalid;
3149 * We can't exactly verify the prev-link, but surely it should be less
3150 * than the record's own address.
3152 if (!XLByteLT(record->xl_prev, *RecPtr))
3155 (errmsg("record with incorrect prev-link %X/%X at %X/%X",
3156 record->xl_prev.xlogid, record->xl_prev.xrecoff,
3157 RecPtr->xlogid, RecPtr->xrecoff)));
3158 goto next_record_is_invalid;
3164 * Record's prev-link should exactly match our previous location. This
3165 * check guards against torn WAL pages where a stale but valid-looking
3166 * WAL record starts on a sector boundary.
3168 if (!XLByteEQ(record->xl_prev, ReadRecPtr))
3171 (errmsg("record with incorrect prev-link %X/%X at %X/%X",
3172 record->xl_prev.xlogid, record->xl_prev.xrecoff,
3173 RecPtr->xlogid, RecPtr->xrecoff)));
3174 goto next_record_is_invalid;
3179 * Allocate or enlarge readRecordBuf as needed. To avoid useless small
3180 * increases, round its size to a multiple of XLOG_BLCKSZ, and make sure
3181 * it's at least 4*Max(BLCKSZ, XLOG_BLCKSZ) to start with. (That is
3182 * enough for all "normal" records, but very large commit or abort records
3183 * might need more space.)
3185 total_len = record->xl_tot_len;
3186 if (total_len > readRecordBufSize)
3188 uint32 newSize = total_len;
3190 newSize += XLOG_BLCKSZ - (newSize % XLOG_BLCKSZ);
3191 newSize = Max(newSize, 4 * Max(BLCKSZ, XLOG_BLCKSZ));
3193 free(readRecordBuf);
3194 readRecordBuf = (char *) malloc(newSize);
3197 readRecordBufSize = 0;
3198 /* We treat this as a "bogus data" condition */
3200 (errmsg("record length %u at %X/%X too long",
3201 total_len, RecPtr->xlogid, RecPtr->xrecoff)));
3202 goto next_record_is_invalid;
3204 readRecordBufSize = newSize;
3207 buffer = readRecordBuf;
3209 len = XLOG_BLCKSZ - RecPtr->xrecoff % XLOG_BLCKSZ;
3210 if (total_len > len)
3212 /* Need to reassemble record */
3213 XLogContRecord *contrecord;
3214 uint32 gotlen = len;
3216 memcpy(buffer, record, len);
3217 record = (XLogRecord *) buffer;
3221 readOff += XLOG_BLCKSZ;
3222 if (readOff >= XLogSegSize)
3226 NextLogSeg(readId, readSeg);
3227 readFile = XLogFileRead(readId, readSeg, emode);
3229 goto next_record_is_invalid;
3232 if (read(readFile, readBuf, XLOG_BLCKSZ) != XLOG_BLCKSZ)
3235 (errcode_for_file_access(),
3236 errmsg("could not read from log file %u, segment %u, offset %u: %m",
3237 readId, readSeg, readOff)));
3238 goto next_record_is_invalid;
3240 if (!ValidXLOGHeader((XLogPageHeader) readBuf, emode))
3241 goto next_record_is_invalid;
3242 if (!(((XLogPageHeader) readBuf)->xlp_info & XLP_FIRST_IS_CONTRECORD))
3245 (errmsg("there is no contrecord flag in log file %u, segment %u, offset %u",
3246 readId, readSeg, readOff)));
3247 goto next_record_is_invalid;
3249 pageHeaderSize = XLogPageHeaderSize((XLogPageHeader) readBuf);
3250 contrecord = (XLogContRecord *) ((char *) readBuf + pageHeaderSize);
3251 if (contrecord->xl_rem_len == 0 ||
3252 total_len != (contrecord->xl_rem_len + gotlen))
3255 (errmsg("invalid contrecord length %u in log file %u, segment %u, offset %u",
3256 contrecord->xl_rem_len,
3257 readId, readSeg, readOff)));
3258 goto next_record_is_invalid;
3260 len = XLOG_BLCKSZ - pageHeaderSize - SizeOfXLogContRecord;
3261 if (contrecord->xl_rem_len > len)
3263 memcpy(buffer, (char *) contrecord + SizeOfXLogContRecord, len);
3268 memcpy(buffer, (char *) contrecord + SizeOfXLogContRecord,
3269 contrecord->xl_rem_len);
3272 if (!RecordIsValid(record, *RecPtr, emode))
3273 goto next_record_is_invalid;
3274 pageHeaderSize = XLogPageHeaderSize((XLogPageHeader) readBuf);
3275 if (XLOG_BLCKSZ - SizeOfXLogRecord >= pageHeaderSize +
3276 MAXALIGN(SizeOfXLogContRecord + contrecord->xl_rem_len))
3278 nextRecord = (XLogRecord *) ((char *) contrecord +
3279 MAXALIGN(SizeOfXLogContRecord + contrecord->xl_rem_len));
3281 EndRecPtr.xlogid = readId;
3282 EndRecPtr.xrecoff = readSeg * XLogSegSize + readOff +
3284 MAXALIGN(SizeOfXLogContRecord + contrecord->xl_rem_len);
3285 ReadRecPtr = *RecPtr;
3286 /* needn't worry about XLOG SWITCH, it can't cross page boundaries */
3290 /* Record does not cross a page boundary */
3291 if (!RecordIsValid(record, *RecPtr, emode))
3292 goto next_record_is_invalid;
3293 if (XLOG_BLCKSZ - SizeOfXLogRecord >= RecPtr->xrecoff % XLOG_BLCKSZ +
3294 MAXALIGN(total_len))
3295 nextRecord = (XLogRecord *) ((char *) record + MAXALIGN(total_len));
3296 EndRecPtr.xlogid = RecPtr->xlogid;
3297 EndRecPtr.xrecoff = RecPtr->xrecoff + MAXALIGN(total_len);
3298 ReadRecPtr = *RecPtr;
3299 memcpy(buffer, record, total_len);
3302 * Special processing if it's an XLOG SWITCH record
3304 if (record->xl_rmid == RM_XLOG_ID && record->xl_info == XLOG_SWITCH)
3306 /* Pretend it extends to end of segment */
3307 EndRecPtr.xrecoff += XLogSegSize - 1;
3308 EndRecPtr.xrecoff -= EndRecPtr.xrecoff % XLogSegSize;
3309 nextRecord = NULL; /* definitely not on same page */
3312 * Pretend that readBuf contains the last page of the segment. This is
3313 * just to avoid Assert failure in StartupXLOG if XLOG ends with this
3316 readOff = XLogSegSize - XLOG_BLCKSZ;
3318 return (XLogRecord *) buffer;
3320 next_record_is_invalid:;
3328 * Check whether the xlog header of a page just read in looks valid.
3330 * This is just a convenience subroutine to avoid duplicated code in
3331 * ReadRecord. It's not intended for use from anywhere else.
3334 ValidXLOGHeader(XLogPageHeader hdr, int emode)
3338 if (hdr->xlp_magic != XLOG_PAGE_MAGIC)
3341 (errmsg("invalid magic number %04X in log file %u, segment %u, offset %u",
3342 hdr->xlp_magic, readId, readSeg, readOff)));
3345 if ((hdr->xlp_info & ~XLP_ALL_FLAGS) != 0)
3348 (errmsg("invalid info bits %04X in log file %u, segment %u, offset %u",
3349 hdr->xlp_info, readId, readSeg, readOff)));
3352 if (hdr->xlp_info & XLP_LONG_HEADER)
3354 XLogLongPageHeader longhdr = (XLogLongPageHeader) hdr;
3356 if (longhdr->xlp_sysid != ControlFile->system_identifier)
3358 char fhdrident_str[32];
3359 char sysident_str[32];
3362 * Format sysids separately to keep platform-dependent format code
3363 * out of the translatable message string.
3365 snprintf(fhdrident_str, sizeof(fhdrident_str), UINT64_FORMAT,
3366 longhdr->xlp_sysid);
3367 snprintf(sysident_str, sizeof(sysident_str), UINT64_FORMAT,
3368 ControlFile->system_identifier);
3370 (errmsg("WAL file is from different system"),
3371 errdetail("WAL file SYSID is %s, pg_control SYSID is %s",
3372 fhdrident_str, sysident_str)));
3375 if (longhdr->xlp_seg_size != XLogSegSize)
3378 (errmsg("WAL file is from different system"),
3379 errdetail("Incorrect XLOG_SEG_SIZE in page header.")));
3382 if (longhdr->xlp_xlog_blcksz != XLOG_BLCKSZ)
3385 (errmsg("WAL file is from different system"),
3386 errdetail("Incorrect XLOG_BLCKSZ in page header.")));
3390 else if (readOff == 0)
3392 /* hmm, first page of file doesn't have a long header? */
3394 (errmsg("invalid info bits %04X in log file %u, segment %u, offset %u",
3395 hdr->xlp_info, readId, readSeg, readOff)));
3399 recaddr.xlogid = readId;
3400 recaddr.xrecoff = readSeg * XLogSegSize + readOff;
3401 if (!XLByteEQ(hdr->xlp_pageaddr, recaddr))
3404 (errmsg("unexpected pageaddr %X/%X in log file %u, segment %u, offset %u",
3405 hdr->xlp_pageaddr.xlogid, hdr->xlp_pageaddr.xrecoff,
3406 readId, readSeg, readOff)));
3411 * Check page TLI is one of the expected values.
3413 if (!list_member_int(expectedTLIs, (int) hdr->xlp_tli))
3416 (errmsg("unexpected timeline ID %u in log file %u, segment %u, offset %u",
3418 readId, readSeg, readOff)));
3423 * Since child timelines are always assigned a TLI greater than their
3424 * immediate parent's TLI, we should never see TLI go backwards across
3425 * successive pages of a consistent WAL sequence.
3427 * Of course this check should only be applied when advancing sequentially
3428 * across pages; therefore ReadRecord resets lastPageTLI to zero when
3429 * going to a random page.
3431 if (hdr->xlp_tli < lastPageTLI)
3434 (errmsg("out-of-sequence timeline ID %u (after %u) in log file %u, segment %u, offset %u",
3435 hdr->xlp_tli, lastPageTLI,
3436 readId, readSeg, readOff)));
3439 lastPageTLI = hdr->xlp_tli;
3444 * Try to read a timeline's history file.
3446 * If successful, return the list of component TLIs (the given TLI followed by
3447 * its ancestor TLIs). If we can't find the history file, assume that the
3448 * timeline has no parents, and return a list of just the specified timeline
3452 readTimeLineHistory(TimeLineID targetTLI)
3455 char path[MAXPGPATH];
3456 char histfname[MAXFNAMELEN];
3457 char fline[MAXPGPATH];
3460 if (InArchiveRecovery)
3462 TLHistoryFileName(histfname, targetTLI);
3463 RestoreArchivedFile(path, histfname, "RECOVERYHISTORY", 0);
3466 TLHistoryFilePath(path, targetTLI);
3468 fd = AllocateFile(path, "r");
3471 if (errno != ENOENT)
3473 (errcode_for_file_access(),
3474 errmsg("could not open file \"%s\": %m", path)));
3475 /* Not there, so assume no parents */
3476 return list_make1_int((int) targetTLI);
3484 while (fgets(fline, sizeof(fline), fd) != NULL)
3486 /* skip leading whitespace and check for # comment */
3491 for (ptr = fline; *ptr; ptr++)
3493 if (!isspace((unsigned char) *ptr))
3496 if (*ptr == '\0' || *ptr == '#')
3499 /* expect a numeric timeline ID as first field of line */
3500 tli = (TimeLineID) strtoul(ptr, &endptr, 0);
3503 (errmsg("syntax error in history file: %s", fline),
3504 errhint("Expected a numeric timeline ID.")));
3507 tli <= (TimeLineID) linitial_int(result))
3509 (errmsg("invalid data in history file: %s", fline),
3510 errhint("Timeline IDs must be in increasing sequence.")));
3512 /* Build list with newest item first */
3513 result = lcons_int((int) tli, result);
3515 /* we ignore the remainder of each line */
3521 targetTLI <= (TimeLineID) linitial_int(result))
3523 (errmsg("invalid data in history file \"%s\"", path),
3524 errhint("Timeline IDs must be less than child timeline's ID.")));
3526 result = lcons_int((int) targetTLI, result);
3529 (errmsg_internal("history of timeline %u is %s",
3530 targetTLI, nodeToString(result))));
3536 * Probe whether a timeline history file exists for the given timeline ID
3539 existsTimeLineHistory(TimeLineID probeTLI)
3541 char path[MAXPGPATH];
3542 char histfname[MAXFNAMELEN];
3545 if (InArchiveRecovery)
3547 TLHistoryFileName(histfname, probeTLI);
3548 RestoreArchivedFile(path, histfname, "RECOVERYHISTORY", 0);
3551 TLHistoryFilePath(path, probeTLI);
3553 fd = AllocateFile(path, "r");
3561 if (errno != ENOENT)
3563 (errcode_for_file_access(),
3564 errmsg("could not open file \"%s\": %m", path)));
3570 * Find the newest existing timeline, assuming that startTLI exists.
3572 * Note: while this is somewhat heuristic, it does positively guarantee
3573 * that (result + 1) is not a known timeline, and therefore it should
3574 * be safe to assign that ID to a new timeline.
3577 findNewestTimeLine(TimeLineID startTLI)
3579 TimeLineID newestTLI;
3580 TimeLineID probeTLI;
3583 * The algorithm is just to probe for the existence of timeline history
3584 * files. XXX is it useful to allow gaps in the sequence?
3586 newestTLI = startTLI;
3588 for (probeTLI = startTLI + 1;; probeTLI++)
3590 if (existsTimeLineHistory(probeTLI))
3592 newestTLI = probeTLI; /* probeTLI exists */
3596 /* doesn't exist, assume we're done */
3605 * Create a new timeline history file.
3607 * newTLI: ID of the new timeline
3608 * parentTLI: ID of its immediate parent
3609 * endTLI et al: ID of the last used WAL file, for annotation purposes
3611 * Currently this is only used during recovery, and so there are no locking
3612 * considerations. But we should be just as tense as XLogFileInit to avoid
3613 * emplacing a bogus file.
3616 writeTimeLineHistory(TimeLineID newTLI, TimeLineID parentTLI,
3617 TimeLineID endTLI, uint32 endLogId, uint32 endLogSeg)
3619 char path[MAXPGPATH];
3620 char tmppath[MAXPGPATH];
3621 char histfname[MAXFNAMELEN];
3622 char xlogfname[MAXFNAMELEN];
3623 char buffer[BLCKSZ];
3628 Assert(newTLI > parentTLI); /* else bad selection of newTLI */
3631 * Write into a temp file name.
3633 snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3637 /* do not use XLOG_SYNC_BIT here --- want to fsync only at end of fill */
3638 fd = BasicOpenFile(tmppath, O_RDWR | O_CREAT | O_EXCL,
3642 (errcode_for_file_access(),
3643 errmsg("could not create file \"%s\": %m", tmppath)));
3646 * If a history file exists for the parent, copy it verbatim
3648 if (InArchiveRecovery)
3650 TLHistoryFileName(histfname, parentTLI);
3651 RestoreArchivedFile(path, histfname, "RECOVERYHISTORY", 0);
3654 TLHistoryFilePath(path, parentTLI);
3656 srcfd = BasicOpenFile(path, O_RDONLY, 0);
3659 if (errno != ENOENT)
3661 (errcode_for_file_access(),
3662 errmsg("could not open file \"%s\": %m", path)));
3663 /* Not there, so assume parent has no parents */
3670 nbytes = (int) read(srcfd, buffer, sizeof(buffer));
3671 if (nbytes < 0 || errno != 0)
3673 (errcode_for_file_access(),
3674 errmsg("could not read file \"%s\": %m", path)));
3678 if ((int) write(fd, buffer, nbytes) != nbytes)
3680 int save_errno = errno;
3683 * If we fail to make the file, delete it to release disk
3689 * if write didn't set errno, assume problem is no disk space
3691 errno = save_errno ? save_errno : ENOSPC;
3694 (errcode_for_file_access(),
3695 errmsg("could not write to file \"%s\": %m", tmppath)));
3702 * Append one line with the details of this timeline split.
3704 * If we did have a parent file, insert an extra newline just in case the
3705 * parent file failed to end with one.
3707 XLogFileName(xlogfname, endTLI, endLogId, endLogSeg);
3709 snprintf(buffer, sizeof(buffer),
3710 "%s%u\t%s\t%s transaction %u at %s\n",
3711 (srcfd < 0) ? "" : "\n",
3714 recoveryStopAfter ? "after" : "before",
3716 timestamptz_to_str(recoveryStopTime));
3718 nbytes = strlen(buffer);
3720 if ((int) write(fd, buffer, nbytes) != nbytes)
3722 int save_errno = errno;
3725 * If we fail to make the file, delete it to release disk space
3728 /* if write didn't set errno, assume problem is no disk space */
3729 errno = save_errno ? save_errno : ENOSPC;
3732 (errcode_for_file_access(),
3733 errmsg("could not write to file \"%s\": %m", tmppath)));
3736 if (pg_fsync(fd) != 0)
3738 (errcode_for_file_access(),
3739 errmsg("could not fsync file \"%s\": %m", tmppath)));
3743 (errcode_for_file_access(),
3744 errmsg("could not close file \"%s\": %m", tmppath)));
3748 * Now move the completed history file into place with its final name.
3750 TLHistoryFilePath(path, newTLI);
3753 * Prefer link() to rename() here just to be really sure that we don't
3754 * overwrite an existing logfile. However, there shouldn't be one, so
3755 * rename() is an acceptable substitute except for the truly paranoid.
3757 #if HAVE_WORKING_LINK
3758 if (link(tmppath, path) < 0)
3760 (errcode_for_file_access(),
3761 errmsg("could not link file \"%s\" to \"%s\": %m",
3765 if (rename(tmppath, path) < 0)
3767 (errcode_for_file_access(),
3768 errmsg("could not rename file \"%s\" to \"%s\": %m",
3772 /* The history file can be archived immediately. */
3773 TLHistoryFileName(histfname, newTLI);
3774 XLogArchiveNotify(histfname);
3778 * I/O routines for pg_control
3780 * *ControlFile is a buffer in shared memory that holds an image of the
3781 * contents of pg_control. WriteControlFile() initializes pg_control
3782 * given a preloaded buffer, ReadControlFile() loads the buffer from
3783 * the pg_control file (during postmaster or standalone-backend startup),
3784 * and UpdateControlFile() rewrites pg_control after we modify xlog state.
3786 * For simplicity, WriteControlFile() initializes the fields of pg_control
3787 * that are related to checking backend/database compatibility, and
3788 * ReadControlFile() verifies they are correct. We could split out the
3789 * I/O and compatibility-check functions, but there seems no need currently.
3792 WriteControlFile(void)
3795 char buffer[PG_CONTROL_SIZE]; /* need not be aligned */
3799 * Initialize version and compatibility-check fields
3801 ControlFile->pg_control_version = PG_CONTROL_VERSION;
3802 ControlFile->catalog_version_no = CATALOG_VERSION_NO;
3804 ControlFile->maxAlign = MAXIMUM_ALIGNOF;
3805 ControlFile->floatFormat = FLOATFORMAT_VALUE;
3807 ControlFile->blcksz = BLCKSZ;
3808 ControlFile->relseg_size = RELSEG_SIZE;
3809 ControlFile->xlog_blcksz = XLOG_BLCKSZ;
3810 ControlFile->xlog_seg_size = XLOG_SEG_SIZE;
3812 ControlFile->nameDataLen = NAMEDATALEN;
3813 ControlFile->indexMaxKeys = INDEX_MAX_KEYS;
3815 ControlFile->toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
3817 #ifdef HAVE_INT64_TIMESTAMP
3818 ControlFile->enableIntTimes = true;
3820 ControlFile->enableIntTimes = false;
3822 ControlFile->float4ByVal = FLOAT4PASSBYVAL;
3823 ControlFile->float8ByVal = FLOAT8PASSBYVAL;
3825 ControlFile->localeBuflen = LOCALE_NAME_BUFLEN;
3826 localeptr = setlocale(LC_COLLATE, NULL);
3829 (errmsg("invalid LC_COLLATE setting")));
3830 StrNCpy(ControlFile->lc_collate, localeptr, LOCALE_NAME_BUFLEN);
3831 localeptr = setlocale(LC_CTYPE, NULL);
3834 (errmsg("invalid LC_CTYPE setting")));
3835 StrNCpy(ControlFile->lc_ctype, localeptr, LOCALE_NAME_BUFLEN);
3837 /* Contents are protected with a CRC */
3838 INIT_CRC32(ControlFile->crc);
3839 COMP_CRC32(ControlFile->crc,
3840 (char *) ControlFile,
3841 offsetof(ControlFileData, crc));
3842 FIN_CRC32(ControlFile->crc);
3845 * We write out PG_CONTROL_SIZE bytes into pg_control, zero-padding the
3846 * excess over sizeof(ControlFileData). This reduces the odds of
3847 * premature-EOF errors when reading pg_control. We'll still fail when we
3848 * check the contents of the file, but hopefully with a more specific
3849 * error than "couldn't read pg_control".
3851 if (sizeof(ControlFileData) > PG_CONTROL_SIZE)
3852 elog(PANIC, "sizeof(ControlFileData) is larger than PG_CONTROL_SIZE; fix either one");
3854 memset(buffer, 0, PG_CONTROL_SIZE);
3855 memcpy(buffer, ControlFile, sizeof(ControlFileData));
3857 fd = BasicOpenFile(XLOG_CONTROL_FILE,
3858 O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
3862 (errcode_for_file_access(),
3863 errmsg("could not create control file \"%s\": %m",
3864 XLOG_CONTROL_FILE)));
3867 if (write(fd, buffer, PG_CONTROL_SIZE) != PG_CONTROL_SIZE)
3869 /* if write didn't set errno, assume problem is no disk space */
3873 (errcode_for_file_access(),
3874 errmsg("could not write to control file: %m")));
3877 if (pg_fsync(fd) != 0)
3879 (errcode_for_file_access(),
3880 errmsg("could not fsync control file: %m")));
3884 (errcode_for_file_access(),
3885 errmsg("could not close control file: %m")));
3889 ReadControlFile(void)
3897 fd = BasicOpenFile(XLOG_CONTROL_FILE,
3902 (errcode_for_file_access(),
3903 errmsg("could not open control file \"%s\": %m",
3904 XLOG_CONTROL_FILE)));
3906 if (read(fd, ControlFile, sizeof(ControlFileData)) != sizeof(ControlFileData))
3908 (errcode_for_file_access(),
3909 errmsg("could not read from control file: %m")));
3914 * Check for expected pg_control format version. If this is wrong, the
3915 * CRC check will likely fail because we'll be checking the wrong number
3916 * of bytes. Complaining about wrong version will probably be more
3917 * enlightening than complaining about wrong CRC.
3920 if (ControlFile->pg_control_version != PG_CONTROL_VERSION && ControlFile->pg_control_version % 65536 == 0 && ControlFile->pg_control_version / 65536 != 0)
3922 (errmsg("database files are incompatible with server"),
3923 errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
3924 " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
3925 ControlFile->pg_control_version, ControlFile->pg_control_version,
3926 PG_CONTROL_VERSION, PG_CONTROL_VERSION),
3927 errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
3929 if (ControlFile->pg_control_version != PG_CONTROL_VERSION)
3931 (errmsg("database files are incompatible with server"),
3932 errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
3933 " but the server was compiled with PG_CONTROL_VERSION %d.",
3934 ControlFile->pg_control_version, PG_CONTROL_VERSION),
3935 errhint("It looks like you need to initdb.")));
3937 /* Now check the CRC. */
3940 (char *) ControlFile,
3941 offsetof(ControlFileData, crc));
3944 if (!EQ_CRC32(crc, ControlFile->crc))
3946 (errmsg("incorrect checksum in control file")));
3949 * Do compatibility checking immediately. We do this here for 2 reasons:
3951 * (1) if the database isn't compatible with the backend executable, we
3952 * want to abort before we can possibly do any damage;
3954 * (2) this code is executed in the postmaster, so the setlocale() will
3955 * propagate to forked backends, which aren't going to read this file for
3956 * themselves. (These locale settings are considered critical
3957 * compatibility items because they can affect sort order of indexes.)
3959 if (ControlFile->catalog_version_no != CATALOG_VERSION_NO)
3961 (errmsg("database files are incompatible with server"),
3962 errdetail("The database cluster was initialized with CATALOG_VERSION_NO %d,"
3963 " but the server was compiled with CATALOG_VERSION_NO %d.",
3964 ControlFile->catalog_version_no, CATALOG_VERSION_NO),
3965 errhint("It looks like you need to initdb.")));
3966 if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
3968 (errmsg("database files are incompatible with server"),
3969 errdetail("The database cluster was initialized with MAXALIGN %d,"
3970 " but the server was compiled with MAXALIGN %d.",
3971 ControlFile->maxAlign, MAXIMUM_ALIGNOF),
3972 errhint("It looks like you need to initdb.")));
3973 if (ControlFile->floatFormat != FLOATFORMAT_VALUE)
3975 (errmsg("database files are incompatible with server"),
3976 errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
3977 errhint("It looks like you need to initdb.")));
3978 if (ControlFile->blcksz != BLCKSZ)
3980 (errmsg("database files are incompatible with server"),
3981 errdetail("The database cluster was initialized with BLCKSZ %d,"
3982 " but the server was compiled with BLCKSZ %d.",
3983 ControlFile->blcksz, BLCKSZ),
3984 errhint("It looks like you need to recompile or initdb.")));
3985 if (ControlFile->relseg_size != RELSEG_SIZE)
3987 (errmsg("database files are incompatible with server"),
3988 errdetail("The database cluster was initialized with RELSEG_SIZE %d,"
3989 " but the server was compiled with RELSEG_SIZE %d.",
3990 ControlFile->relseg_size, RELSEG_SIZE),
3991 errhint("It looks like you need to recompile or initdb.")));
3992 if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
3994 (errmsg("database files are incompatible with server"),
3995 errdetail("The database cluster was initialized with XLOG_BLCKSZ %d,"
3996 " but the server was compiled with XLOG_BLCKSZ %d.",
3997 ControlFile->xlog_blcksz, XLOG_BLCKSZ),
3998 errhint("It looks like you need to recompile or initdb.")));
3999 if (ControlFile->xlog_seg_size != XLOG_SEG_SIZE)
4001 (errmsg("database files are incompatible with server"),
4002 errdetail("The database cluster was initialized with XLOG_SEG_SIZE %d,"
4003 " but the server was compiled with XLOG_SEG_SIZE %d.",
4004 ControlFile->xlog_seg_size, XLOG_SEG_SIZE),
4005 errhint("It looks like you need to recompile or initdb.")));
4006 if (ControlFile->nameDataLen != NAMEDATALEN)
4008 (errmsg("database files are incompatible with server"),
4009 errdetail("The database cluster was initialized with NAMEDATALEN %d,"
4010 " but the server was compiled with NAMEDATALEN %d.",
4011 ControlFile->nameDataLen, NAMEDATALEN),
4012 errhint("It looks like you need to recompile or initdb.")));
4013 if (ControlFile->indexMaxKeys != INDEX_MAX_KEYS)
4015 (errmsg("database files are incompatible with server"),
4016 errdetail("The database cluster was initialized with INDEX_MAX_KEYS %d,"
4017 " but the server was compiled with INDEX_MAX_KEYS %d.",
4018 ControlFile->indexMaxKeys, INDEX_MAX_KEYS),
4019 errhint("It looks like you need to recompile or initdb.")));
4020 if (ControlFile->toast_max_chunk_size != TOAST_MAX_CHUNK_SIZE)
4022 (errmsg("database files are incompatible with server"),
4023 errdetail("The database cluster was initialized with TOAST_MAX_CHUNK_SIZE %d,"
4024 " but the server was compiled with TOAST_MAX_CHUNK_SIZE %d.",
4025 ControlFile->toast_max_chunk_size, (int) TOAST_MAX_CHUNK_SIZE),
4026 errhint("It looks like you need to recompile or initdb.")));
4028 #ifdef HAVE_INT64_TIMESTAMP
4029 if (ControlFile->enableIntTimes != true)
4031 (errmsg("database files are incompatible with server"),
4032 errdetail("The database cluster was initialized without HAVE_INT64_TIMESTAMP"
4033 " but the server was compiled with HAVE_INT64_TIMESTAMP."),
4034 errhint("It looks like you need to recompile or initdb.")));
4036 if (ControlFile->enableIntTimes != false)
4038 (errmsg("database files are incompatible with server"),
4039 errdetail("The database cluster was initialized with HAVE_INT64_TIMESTAMP"
4040 " but the server was compiled without HAVE_INT64_TIMESTAMP."),
4041 errhint("It looks like you need to recompile or initdb.")));
4044 #ifdef USE_FLOAT4_BYVAL
4045 if (ControlFile->float4ByVal != true)
4047 (errmsg("database files are incompatible with server"),
4048 errdetail("The database cluster was initialized without USE_FLOAT4_BYVAL"
4049 " but the server was compiled with USE_FLOAT4_BYVAL."),
4050 errhint("It looks like you need to recompile or initdb.")));
4052 if (ControlFile->float4ByVal != false)
4054 (errmsg("database files are incompatible with server"),
4055 errdetail("The database cluster was initialized with USE_FLOAT4_BYVAL"
4056 " but the server was compiled without USE_FLOAT4_BYVAL."),
4057 errhint("It looks like you need to recompile or initdb.")));
4060 #ifdef USE_FLOAT8_BYVAL
4061 if (ControlFile->float8ByVal != true)
4063 (errmsg("database files are incompatible with server"),
4064 errdetail("The database cluster was initialized without USE_FLOAT8_BYVAL"
4065 " but the server was compiled with USE_FLOAT8_BYVAL."),
4066 errhint("It looks like you need to recompile or initdb.")));
4068 if (ControlFile->float8ByVal != false)
4070 (errmsg("database files are incompatible with server"),
4071 errdetail("The database cluster was initialized with USE_FLOAT8_BYVAL"
4072 " but the server was compiled without USE_FLOAT8_BYVAL."),
4073 errhint("It looks like you need to recompile or initdb.")));
4076 if (ControlFile->localeBuflen != LOCALE_NAME_BUFLEN)
4078 (errmsg("database files are incompatible with server"),
4079 errdetail("The database cluster was initialized with LOCALE_NAME_BUFLEN %d,"
4080 " but the server was compiled with LOCALE_NAME_BUFLEN %d.",
4081 ControlFile->localeBuflen, LOCALE_NAME_BUFLEN),
4082 errhint("It looks like you need to recompile or initdb.")));
4083 if (pg_perm_setlocale(LC_COLLATE, ControlFile->lc_collate) == NULL)
4085 (errmsg("database files are incompatible with operating system"),
4086 errdetail("The database cluster was initialized with LC_COLLATE \"%s\","
4087 " which is not recognized by setlocale().",
4088 ControlFile->lc_collate),
4089 errhint("It looks like you need to initdb or install locale support.")));
4090 if (pg_perm_setlocale(LC_CTYPE, ControlFile->lc_ctype) == NULL)
4092 (errmsg("database files are incompatible with operating system"),
4093 errdetail("The database cluster was initialized with LC_CTYPE \"%s\","
4094 " which is not recognized by setlocale().",
4095 ControlFile->lc_ctype),
4096 errhint("It looks like you need to initdb or install locale support.")));
4098 /* Make the fixed locale settings visible as GUC variables, too */
4099 SetConfigOption("lc_collate", ControlFile->lc_collate,
4100 PGC_INTERNAL, PGC_S_OVERRIDE);
4101 SetConfigOption("lc_ctype", ControlFile->lc_ctype,
4102 PGC_INTERNAL, PGC_S_OVERRIDE);
4106 UpdateControlFile(void)
4110 INIT_CRC32(ControlFile->crc);
4111 COMP_CRC32(ControlFile->crc,
4112 (char *) ControlFile,
4113 offsetof(ControlFileData, crc));
4114 FIN_CRC32(ControlFile->crc);
4116 fd = BasicOpenFile(XLOG_CONTROL_FILE,
4121 (errcode_for_file_access(),
4122 errmsg("could not open control file \"%s\": %m",
4123 XLOG_CONTROL_FILE)));
4126 if (write(fd, ControlFile, sizeof(ControlFileData)) != sizeof(ControlFileData))
4128 /* if write didn't set errno, assume problem is no disk space */
4132 (errcode_for_file_access(),
4133 errmsg("could not write to control file: %m")));
4136 if (pg_fsync(fd) != 0)
4138 (errcode_for_file_access(),
4139 errmsg("could not fsync control file: %m")));
4143 (errcode_for_file_access(),
4144 errmsg("could not close control file: %m")));
4148 * Initialization of shared memory for XLOG
4156 size = sizeof(XLogCtlData);
4157 /* xlblocks array */
4158 size = add_size(size, mul_size(sizeof(XLogRecPtr), XLOGbuffers));
4159 /* extra alignment padding for XLOG I/O buffers */
4160 size = add_size(size, ALIGNOF_XLOG_BUFFER);
4161 /* and the buffers themselves */
4162 size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4165 * Note: we don't count ControlFileData, it comes out of the "slop factor"
4166 * added by CreateSharedMemoryAndSemaphores. This lets us use this
4167 * routine again below to compute the actual allocation size.
4180 ControlFile = (ControlFileData *)
4181 ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
4182 XLogCtl = (XLogCtlData *)
4183 ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
4185 if (foundCFile || foundXLog)
4187 /* both should be present or neither */
4188 Assert(foundCFile && foundXLog);
4192 memset(XLogCtl, 0, sizeof(XLogCtlData));
4195 * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
4196 * multiple of the alignment for same, so no extra alignment padding is
4199 allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
4200 XLogCtl->xlblocks = (XLogRecPtr *) allocptr;
4201 memset(XLogCtl->xlblocks, 0, sizeof(XLogRecPtr) * XLOGbuffers);
4202 allocptr += sizeof(XLogRecPtr) * XLOGbuffers;
4205 * Align the start of the page buffers to an ALIGNOF_XLOG_BUFFER boundary.
4207 allocptr = (char *) TYPEALIGN(ALIGNOF_XLOG_BUFFER, allocptr);
4208 XLogCtl->pages = allocptr;
4209 memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
4212 * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
4213 * in additional info.)
4215 XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
4216 XLogCtl->Insert.currpage = (XLogPageHeader) (XLogCtl->pages);
4217 SpinLockInit(&XLogCtl->info_lck);
4220 * If we are not in bootstrap mode, pg_control should already exist. Read
4221 * and validate it immediately (see comments in ReadControlFile() for the
4224 if (!IsBootstrapProcessingMode())
4229 * This func must be called ONCE on system install. It creates pg_control
4230 * and the initial XLOG segment.
4235 CheckPoint checkPoint;
4237 XLogPageHeader page;
4238 XLogLongPageHeader longpage;
4241 uint64 sysidentifier;
4246 * Select a hopefully-unique system identifier code for this installation.
4247 * We use the result of gettimeofday(), including the fractional seconds
4248 * field, as being about as unique as we can easily get. (Think not to
4249 * use random(), since it hasn't been seeded and there's no portable way
4250 * to seed it other than the system clock value...) The upper half of the
4251 * uint64 value is just the tv_sec part, while the lower half is the XOR
4252 * of tv_sec and tv_usec. This is to ensure that we don't lose uniqueness
4253 * unnecessarily if "uint64" is really only 32 bits wide. A person
4254 * knowing this encoding can determine the initialization time of the
4255 * installation, which could perhaps be useful sometimes.
4257 gettimeofday(&tv, NULL);
4258 sysidentifier = ((uint64) tv.tv_sec) << 32;
4259 sysidentifier |= (uint32) (tv.tv_sec | tv.tv_usec);
4261 /* First timeline ID is always 1 */
4264 /* page buffer must be aligned suitably for O_DIRECT */
4265 buffer = (char *) palloc(XLOG_BLCKSZ + ALIGNOF_XLOG_BUFFER);
4266 page = (XLogPageHeader) TYPEALIGN(ALIGNOF_XLOG_BUFFER, buffer);
4267 memset(page, 0, XLOG_BLCKSZ);
4269 /* Set up information for the initial checkpoint record */
4270 checkPoint.redo.xlogid = 0;
4271 checkPoint.redo.xrecoff = SizeOfXLogLongPHD;
4272 checkPoint.ThisTimeLineID = ThisTimeLineID;
4273 checkPoint.nextXidEpoch = 0;
4274 checkPoint.nextXid = FirstNormalTransactionId;
4275 checkPoint.nextOid = FirstBootstrapObjectId;
4276 checkPoint.nextMulti = FirstMultiXactId;
4277 checkPoint.nextMultiOffset = 0;
4278 checkPoint.time = (pg_time_t) time(NULL);
4280 ShmemVariableCache->nextXid = checkPoint.nextXid;
4281 ShmemVariableCache->nextOid = checkPoint.nextOid;
4282 ShmemVariableCache->oidCount = 0;
4283 MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
4285 /* Set up the XLOG page header */
4286 page->xlp_magic = XLOG_PAGE_MAGIC;
4287 page->xlp_info = XLP_LONG_HEADER;
4288 page->xlp_tli = ThisTimeLineID;
4289 page->xlp_pageaddr.xlogid = 0;
4290 page->xlp_pageaddr.xrecoff = 0;
4291 longpage = (XLogLongPageHeader) page;
4292 longpage->xlp_sysid = sysidentifier;
4293 longpage->xlp_seg_size = XLogSegSize;
4294 longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
4296 /* Insert the initial checkpoint record */
4297 record = (XLogRecord *) ((char *) page + SizeOfXLogLongPHD);
4298 record->xl_prev.xlogid = 0;
4299 record->xl_prev.xrecoff = 0;
4300 record->xl_xid = InvalidTransactionId;
4301 record->xl_tot_len = SizeOfXLogRecord + sizeof(checkPoint);
4302 record->xl_len = sizeof(checkPoint);
4303 record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
4304 record->xl_rmid = RM_XLOG_ID;
4305 memcpy(XLogRecGetData(record), &checkPoint, sizeof(checkPoint));
4308 COMP_CRC32(crc, &checkPoint, sizeof(checkPoint));
4309 COMP_CRC32(crc, (char *) record + sizeof(pg_crc32),
4310 SizeOfXLogRecord - sizeof(pg_crc32));
4312 record->xl_crc = crc;
4314 /* Create first XLOG segment file */
4315 use_existent = false;
4316 openLogFile = XLogFileInit(0, 0, &use_existent, false);
4318 /* Write the first page with the initial record */
4320 if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
4322 /* if write didn't set errno, assume problem is no disk space */
4326 (errcode_for_file_access(),
4327 errmsg("could not write bootstrap transaction log file: %m")));
4330 if (pg_fsync(openLogFile) != 0)
4332 (errcode_for_file_access(),
4333 errmsg("could not fsync bootstrap transaction log file: %m")));
4335 if (close(openLogFile))
4337 (errcode_for_file_access(),
4338 errmsg("could not close bootstrap transaction log file: %m")));
4342 /* Now create pg_control */
4344 memset(ControlFile, 0, sizeof(ControlFileData));
4345 /* Initialize pg_control status fields */
4346 ControlFile->system_identifier = sysidentifier;
4347 ControlFile->state = DB_SHUTDOWNED;
4348 ControlFile->time = checkPoint.time;
4349 ControlFile->checkPoint = checkPoint.redo;
4350 ControlFile->checkPointCopy = checkPoint;
4351 /* some additional ControlFile fields are set in WriteControlFile() */
4355 /* Bootstrap the commit log, too */
4357 BootStrapSUBTRANS();
4358 BootStrapMultiXact();
4364 str_time(pg_time_t tnow)
4366 static char buf[128];
4368 pg_strftime(buf, sizeof(buf),
4369 "%Y-%m-%d %H:%M:%S %Z",
4370 pg_localtime(&tnow, log_timezone));
4376 * See if there is a recovery command file (recovery.conf), and if so
4377 * read in parameters for archive recovery.
4379 * XXX longer term intention is to expand this to
4380 * cater for additional parameters and controls
4381 * possibly use a flex lexer similar to the GUC one
4384 readRecoveryCommandFile(void)
4387 char cmdline[MAXPGPATH];
4388 TimeLineID rtli = 0;
4389 bool rtliGiven = false;
4390 bool syntaxError = false;
4392 fd = AllocateFile(RECOVERY_COMMAND_FILE, "r");
4395 if (errno == ENOENT)
4396 return; /* not there, so no archive recovery */
4398 (errcode_for_file_access(),
4399 errmsg("could not open recovery command file \"%s\": %m",
4400 RECOVERY_COMMAND_FILE)));
4404 (errmsg("starting archive recovery")));
4409 while (fgets(cmdline, sizeof(cmdline), fd) != NULL)
4411 /* skip leading whitespace and check for # comment */
4416 for (ptr = cmdline; *ptr; ptr++)
4418 if (!isspace((unsigned char) *ptr))
4421 if (*ptr == '\0' || *ptr == '#')
4424 /* identify the quoted parameter value */
4425 tok1 = strtok(ptr, "'");
4431 tok2 = strtok(NULL, "'");
4437 /* reparse to get just the parameter name */
4438 tok1 = strtok(ptr, " \t=");
4445 if (strcmp(tok1, "restore_command") == 0)
4447 recoveryRestoreCommand = pstrdup(tok2);
4449 (errmsg("restore_command = '%s'",
4450 recoveryRestoreCommand)));
4452 else if (strcmp(tok1, "recovery_target_timeline") == 0)
4455 if (strcmp(tok2, "latest") == 0)
4460 rtli = (TimeLineID) strtoul(tok2, NULL, 0);
4461 if (errno == EINVAL || errno == ERANGE)
4463 (errmsg("recovery_target_timeline is not a valid number: \"%s\"",
4468 (errmsg("recovery_target_timeline = %u", rtli)));
4471 (errmsg("recovery_target_timeline = latest")));
4473 else if (strcmp(tok1, "recovery_target_xid") == 0)
4476 recoveryTargetXid = (TransactionId) strtoul(tok2, NULL, 0);
4477 if (errno == EINVAL || errno == ERANGE)
4479 (errmsg("recovery_target_xid is not a valid number: \"%s\"",
4482 (errmsg("recovery_target_xid = %u",
4483 recoveryTargetXid)));
4484 recoveryTarget = true;
4485 recoveryTargetExact = true;
4487 else if (strcmp(tok1, "recovery_target_time") == 0)
4490 * if recovery_target_xid specified, then this overrides
4491 * recovery_target_time
4493 if (recoveryTargetExact)
4495 recoveryTarget = true;
4496 recoveryTargetExact = false;
4499 * Convert the time string given by the user to TimestampTz form.
4501 recoveryTargetTime =
4502 DatumGetTimestampTz(DirectFunctionCall3(timestamptz_in,
4503 CStringGetDatum(tok2),
4504 ObjectIdGetDatum(InvalidOid),
4505 Int32GetDatum(-1)));
4507 (errmsg("recovery_target_time = '%s'",
4508 timestamptz_to_str(recoveryTargetTime))));
4510 else if (strcmp(tok1, "recovery_target_inclusive") == 0)
4513 * does nothing if a recovery_target is not also set
4515 if (strcmp(tok2, "true") == 0)
4516 recoveryTargetInclusive = true;
4519 recoveryTargetInclusive = false;
4523 (errmsg("recovery_target_inclusive = %s", tok2)));
4525 else if (strcmp(tok1, "log_restartpoints") == 0)
4528 * does nothing if a recovery_target is not also set
4530 if (strcmp(tok2, "true") == 0)
4531 recoveryLogRestartpoints = true;
4534 recoveryLogRestartpoints = false;
4538 (errmsg("log_restartpoints = %s", tok2)));
4542 (errmsg("unrecognized recovery parameter \"%s\"",
4550 (errmsg("syntax error in recovery command file: %s",
4552 errhint("Lines should have the format parameter = 'value'.")));
4554 /* Check that required parameters were supplied */
4555 if (recoveryRestoreCommand == NULL)
4557 (errmsg("recovery command file \"%s\" did not specify restore_command",
4558 RECOVERY_COMMAND_FILE)));
4560 /* Enable fetching from archive recovery area */
4561 InArchiveRecovery = true;
4564 * If user specified recovery_target_timeline, validate it or compute the
4565 * "latest" value. We can't do this until after we've gotten the restore
4566 * command and set InArchiveRecovery, because we need to fetch timeline
4567 * history files from the archive.
4573 /* Timeline 1 does not have a history file, all else should */
4574 if (rtli != 1 && !existsTimeLineHistory(rtli))
4576 (errmsg("recovery target timeline %u does not exist",
4578 recoveryTargetTLI = rtli;
4582 /* We start the "latest" search from pg_control's timeline */
4583 recoveryTargetTLI = findNewestTimeLine(recoveryTargetTLI);
4589 * Exit archive-recovery state
4592 exitArchiveRecovery(TimeLineID endTLI, uint32 endLogId, uint32 endLogSeg)
4594 char recoveryPath[MAXPGPATH];
4595 char xlogpath[MAXPGPATH];
4598 * We are no longer in archive recovery state.
4600 InArchiveRecovery = false;
4603 * We should have the ending log segment currently open. Verify, and then
4604 * close it (to avoid problems on Windows with trying to rename or delete
4607 Assert(readFile >= 0);
4608 Assert(readId == endLogId);
4609 Assert(readSeg == endLogSeg);
4615 * If the segment was fetched from archival storage, we want to replace
4616 * the existing xlog segment (if any) with the archival version. This is
4617 * because whatever is in XLOGDIR is very possibly older than what we have
4618 * from the archives, since it could have come from restoring a PGDATA
4619 * backup. In any case, the archival version certainly is more
4620 * descriptive of what our current database state is, because that is what
4623 * Note that if we are establishing a new timeline, ThisTimeLineID is
4624 * already set to the new value, and so we will create a new file instead
4625 * of overwriting any existing file. (This is, in fact, always the case
4628 snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYXLOG");
4629 XLogFilePath(xlogpath, ThisTimeLineID, endLogId, endLogSeg);
4631 if (restoredFromArchive)
4634 (errmsg_internal("moving last restored xlog to \"%s\"",
4636 unlink(xlogpath); /* might or might not exist */
4637 if (rename(recoveryPath, xlogpath) != 0)
4639 (errcode_for_file_access(),
4640 errmsg("could not rename file \"%s\" to \"%s\": %m",
4641 recoveryPath, xlogpath)));
4642 /* XXX might we need to fix permissions on the file? */
4647 * If the latest segment is not archival, but there's still a
4648 * RECOVERYXLOG laying about, get rid of it.
4650 unlink(recoveryPath); /* ignore any error */
4653 * If we are establishing a new timeline, we have to copy data from
4654 * the last WAL segment of the old timeline to create a starting WAL
4655 * segment for the new timeline.
4657 if (endTLI != ThisTimeLineID)
4658 XLogFileCopy(endLogId, endLogSeg,
4659 endTLI, endLogId, endLogSeg);
4663 * Let's just make real sure there are not .ready or .done flags posted
4664 * for the new segment.
4666 XLogFileName(xlogpath, ThisTimeLineID, endLogId, endLogSeg);
4667 XLogArchiveCleanup(xlogpath);
4669 /* Get rid of any remaining recovered timeline-history file, too */
4670 snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYHISTORY");
4671 unlink(recoveryPath); /* ignore any error */
4674 * Rename the config file out of the way, so that we don't accidentally
4675 * re-enter archive recovery mode in a subsequent crash.
4677 unlink(RECOVERY_COMMAND_DONE);
4678 if (rename(RECOVERY_COMMAND_FILE, RECOVERY_COMMAND_DONE) != 0)
4680 (errcode_for_file_access(),
4681 errmsg("could not rename file \"%s\" to \"%s\": %m",
4682 RECOVERY_COMMAND_FILE, RECOVERY_COMMAND_DONE)));
4685 (errmsg("archive recovery complete")));
4689 * For point-in-time recovery, this function decides whether we want to
4690 * stop applying the XLOG at or after the current record.
4692 * Returns TRUE if we are stopping, FALSE otherwise. On TRUE return,
4693 * *includeThis is set TRUE if we should apply this record before stopping.
4694 * Also, some information is saved in recoveryStopXid et al for use in
4695 * annotating the new timeline's history file.
4698 recoveryStopsHere(XLogRecord *record, bool *includeThis)
4702 TimestampTz recordXtime;
4704 /* We only consider stopping at COMMIT or ABORT records */
4705 if (record->xl_rmid != RM_XACT_ID)
4707 record_info = record->xl_info & ~XLR_INFO_MASK;
4708 if (record_info == XLOG_XACT_COMMIT)
4710 xl_xact_commit *recordXactCommitData;
4712 recordXactCommitData = (xl_xact_commit *) XLogRecGetData(record);
4713 recordXtime = recordXactCommitData->xact_time;
4715 else if (record_info == XLOG_XACT_ABORT)
4717 xl_xact_abort *recordXactAbortData;
4719 recordXactAbortData = (xl_xact_abort *) XLogRecGetData(record);
4720 recordXtime = recordXactAbortData->xact_time;
4725 /* Remember the most recent COMMIT/ABORT time for logging purposes */
4726 recoveryLastXTime = recordXtime;
4728 /* Do we have a PITR target at all? */
4729 if (!recoveryTarget)
4732 if (recoveryTargetExact)
4735 * there can be only one transaction end record with this exact
4738 * when testing for an xid, we MUST test for equality only, since
4739 * transactions are numbered in the order they start, not the order
4740 * they complete. A higher numbered xid will complete before you about
4741 * 50% of the time...
4743 stopsHere = (record->xl_xid == recoveryTargetXid);
4745 *includeThis = recoveryTargetInclusive;
4750 * there can be many transactions that share the same commit time, so
4751 * we stop after the last one, if we are inclusive, or stop at the
4752 * first one if we are exclusive
4754 if (recoveryTargetInclusive)
4755 stopsHere = (recordXtime > recoveryTargetTime);
4757 stopsHere = (recordXtime >= recoveryTargetTime);
4759 *includeThis = false;
4764 recoveryStopXid = record->xl_xid;
4765 recoveryStopTime = recordXtime;
4766 recoveryStopAfter = *includeThis;
4768 if (record_info == XLOG_XACT_COMMIT)
4770 if (recoveryStopAfter)
4772 (errmsg("recovery stopping after commit of transaction %u, time %s",
4774 timestamptz_to_str(recoveryStopTime))));
4777 (errmsg("recovery stopping before commit of transaction %u, time %s",
4779 timestamptz_to_str(recoveryStopTime))));
4783 if (recoveryStopAfter)
4785 (errmsg("recovery stopping after abort of transaction %u, time %s",
4787 timestamptz_to_str(recoveryStopTime))));
4790 (errmsg("recovery stopping before abort of transaction %u, time %s",
4792 timestamptz_to_str(recoveryStopTime))));
4800 * This must be called ONCE during postmaster or standalone-backend startup
4805 XLogCtlInsert *Insert;
4806 CheckPoint checkPoint;
4808 bool reachedStopPoint = false;
4809 bool haveBackupLabel = false;
4819 TransactionId oldestActiveXID;
4822 * Read control file and check XLOG status looks valid.
4824 * Note: in most control paths, *ControlFile is already valid and we need
4825 * not do ReadControlFile() here, but might as well do it to be sure.
4829 if (ControlFile->state < DB_SHUTDOWNED ||
4830 ControlFile->state > DB_IN_PRODUCTION ||
4831 !XRecOffIsValid(ControlFile->checkPoint.xrecoff))
4833 (errmsg("control file contains invalid data")));
4835 if (ControlFile->state == DB_SHUTDOWNED)
4837 (errmsg("database system was shut down at %s",
4838 str_time(ControlFile->time))));
4839 else if (ControlFile->state == DB_SHUTDOWNING)
4841 (errmsg("database system shutdown was interrupted; last known up at %s",
4842 str_time(ControlFile->time))));
4843 else if (ControlFile->state == DB_IN_CRASH_RECOVERY)
4845 (errmsg("database system was interrupted while in recovery at %s",
4846 str_time(ControlFile->time)),
4847 errhint("This probably means that some data is corrupted and"
4848 " you will have to use the last backup for recovery.")));
4849 else if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY)
4851 (errmsg("database system was interrupted while in recovery at log time %s",
4852 str_time(ControlFile->checkPointCopy.time)),
4853 errhint("If this has occurred more than once some data might be corrupted"
4854 " and you might need to choose an earlier recovery target.")));
4855 else if (ControlFile->state == DB_IN_PRODUCTION)
4857 (errmsg("database system was interrupted; last known up at %s",
4858 str_time(ControlFile->time))));
4860 /* This is just to allow attaching to startup process with a debugger */
4861 #ifdef XLOG_REPLAY_DELAY
4862 if (ControlFile->state != DB_SHUTDOWNED)
4863 pg_usleep(60000000L);
4867 * Initialize on the assumption we want to recover to the same timeline
4868 * that's active according to pg_control.
4870 recoveryTargetTLI = ControlFile->checkPointCopy.ThisTimeLineID;
4873 * Check for recovery control file, and if so set up state for offline
4876 readRecoveryCommandFile();
4878 /* Now we can determine the list of expected TLIs */
4879 expectedTLIs = readTimeLineHistory(recoveryTargetTLI);
4882 * If pg_control's timeline is not in expectedTLIs, then we cannot
4883 * proceed: the backup is not part of the history of the requested
4886 if (!list_member_int(expectedTLIs,
4887 (int) ControlFile->checkPointCopy.ThisTimeLineID))
4889 (errmsg("requested timeline %u is not a child of database system timeline %u",
4891 ControlFile->checkPointCopy.ThisTimeLineID)));
4893 if (read_backup_label(&checkPointLoc, &minRecoveryLoc))
4896 * When a backup_label file is present, we want to roll forward from
4897 * the checkpoint it identifies, rather than using pg_control.
4899 record = ReadCheckpointRecord(checkPointLoc, 0);
4903 (errmsg("checkpoint record is at %X/%X",
4904 checkPointLoc.xlogid, checkPointLoc.xrecoff)));
4905 InRecovery = true; /* force recovery even if SHUTDOWNED */
4910 (errmsg("could not locate required checkpoint record"),
4911 errhint("If you are not restoring from a backup, try removing the file \"%s/backup_label\".", DataDir)));
4913 /* set flag to delete it later */
4914 haveBackupLabel = true;
4919 * Get the last valid checkpoint record. If the latest one according
4920 * to pg_control is broken, try the next-to-last one.
4922 checkPointLoc = ControlFile->checkPoint;
4923 record = ReadCheckpointRecord(checkPointLoc, 1);
4927 (errmsg("checkpoint record is at %X/%X",
4928 checkPointLoc.xlogid, checkPointLoc.xrecoff)));
4932 checkPointLoc = ControlFile->prevCheckPoint;
4933 record = ReadCheckpointRecord(checkPointLoc, 2);
4937 (errmsg("using previous checkpoint record at %X/%X",
4938 checkPointLoc.xlogid, checkPointLoc.xrecoff)));
4939 InRecovery = true; /* force recovery even if SHUTDOWNED */
4943 (errmsg("could not locate a valid checkpoint record")));
4947 LastRec = RecPtr = checkPointLoc;
4948 memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
4949 wasShutdown = (record->xl_info == XLOG_CHECKPOINT_SHUTDOWN);
4952 (errmsg("redo record is at %X/%X; shutdown %s",
4953 checkPoint.redo.xlogid, checkPoint.redo.xrecoff,
4954 wasShutdown ? "TRUE" : "FALSE")));
4956 (errmsg("next transaction ID: %u/%u; next OID: %u",
4957 checkPoint.nextXidEpoch, checkPoint.nextXid,
4958 checkPoint.nextOid)));
4960 (errmsg("next MultiXactId: %u; next MultiXactOffset: %u",
4961 checkPoint.nextMulti, checkPoint.nextMultiOffset)));
4962 if (!TransactionIdIsNormal(checkPoint.nextXid))
4964 (errmsg("invalid next transaction ID")));
4966 ShmemVariableCache->nextXid = checkPoint.nextXid;
4967 ShmemVariableCache->nextOid = checkPoint.nextOid;
4968 ShmemVariableCache->oidCount = 0;
4969 MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
4972 * We must replay WAL entries using the same TimeLineID they were created
4973 * under, so temporarily adopt the TLI indicated by the checkpoint (see
4974 * also xlog_redo()).
4976 ThisTimeLineID = checkPoint.ThisTimeLineID;
4978 RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
4980 if (XLByteLT(RecPtr, checkPoint.redo))
4982 (errmsg("invalid redo in checkpoint record")));
4985 * Check whether we need to force recovery from WAL. If it appears to
4986 * have been a clean shutdown and we did not have a recovery.conf file,
4987 * then assume no recovery needed.
4989 if (XLByteLT(checkPoint.redo, RecPtr))
4993 (errmsg("invalid redo record in shutdown checkpoint")));
4996 else if (ControlFile->state != DB_SHUTDOWNED)
4998 else if (InArchiveRecovery)
5000 /* force recovery due to presence of recovery.conf */
5010 * Update pg_control to show that we are recovering and to show the
5011 * selected checkpoint as the place we are starting from. We also mark
5012 * pg_control with any minimum recovery stop point obtained from a
5013 * backup history file.
5015 if (InArchiveRecovery)
5018 (errmsg("automatic recovery in progress")));
5019 ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
5024 (errmsg("database system was not properly shut down; "
5025 "automatic recovery in progress")));
5026 ControlFile->state = DB_IN_CRASH_RECOVERY;
5028 ControlFile->prevCheckPoint = ControlFile->checkPoint;
5029 ControlFile->checkPoint = checkPointLoc;
5030 ControlFile->checkPointCopy = checkPoint;
5031 if (minRecoveryLoc.xlogid != 0 || minRecoveryLoc.xrecoff != 0)
5032 ControlFile->minRecoveryPoint = minRecoveryLoc;
5033 ControlFile->time = (pg_time_t) time(NULL);
5034 UpdateControlFile();
5037 * If there was a backup label file, it's done its job and the info
5038 * has now been propagated into pg_control. We must get rid of the
5039 * label file so that if we crash during recovery, we'll pick up at
5040 * the latest recovery restartpoint instead of going all the way back
5041 * to the backup start point. It seems prudent though to just rename
5042 * the file out of the way rather than delete it completely.
5044 if (haveBackupLabel)
5046 unlink(BACKUP_LABEL_OLD);
5047 if (rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD) != 0)
5049 (errcode_for_file_access(),
5050 errmsg("could not rename file \"%s\" to \"%s\": %m",
5051 BACKUP_LABEL_FILE, BACKUP_LABEL_OLD)));
5054 /* Start up the recovery environment */
5055 XLogInitRelationCache();
5057 for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
5059 if (RmgrTable[rmid].rm_startup != NULL)
5060 RmgrTable[rmid].rm_startup();
5064 * Find the first record that logically follows the checkpoint --- it
5065 * might physically precede it, though.
5067 if (XLByteLT(checkPoint.redo, RecPtr))
5069 /* back up to find the record */
5070 record = ReadRecord(&(checkPoint.redo), PANIC);
5074 /* just have to read next record after CheckPoint */
5075 record = ReadRecord(NULL, LOG);
5080 bool recoveryContinue = true;
5081 bool recoveryApply = true;
5082 ErrorContextCallback errcontext;
5086 (errmsg("redo starts at %X/%X",
5087 ReadRecPtr.xlogid, ReadRecPtr.xrecoff)));
5090 * main redo apply loop
5099 initStringInfo(&buf);
5100 appendStringInfo(&buf, "REDO @ %X/%X; LSN %X/%X: ",
5101 ReadRecPtr.xlogid, ReadRecPtr.xrecoff,
5102 EndRecPtr.xlogid, EndRecPtr.xrecoff);
5103 xlog_outrec(&buf, record);
5104 appendStringInfo(&buf, " - ");
5105 RmgrTable[record->xl_rmid].rm_desc(&buf,
5107 XLogRecGetData(record));
5108 elog(LOG, "%s", buf.data);
5114 * Have we reached our recovery target?
5116 if (recoveryStopsHere(record, &recoveryApply))
5118 reachedStopPoint = true; /* see below */
5119 recoveryContinue = false;
5124 /* Setup error traceback support for ereport() */
5125 errcontext.callback = rm_redo_error_callback;
5126 errcontext.arg = (void *) record;
5127 errcontext.previous = error_context_stack;
5128 error_context_stack = &errcontext;
5130 /* nextXid must be beyond record's xid */
5131 if (TransactionIdFollowsOrEquals(record->xl_xid,
5132 ShmemVariableCache->nextXid))
5134 ShmemVariableCache->nextXid = record->xl_xid;
5135 TransactionIdAdvance(ShmemVariableCache->nextXid);
5138 if (record->xl_info & XLR_BKP_BLOCK_MASK)
5139 RestoreBkpBlocks(record, EndRecPtr);
5141 RmgrTable[record->xl_rmid].rm_redo(EndRecPtr, record);
5143 /* Pop the error context stack */
5144 error_context_stack = errcontext.previous;
5146 LastRec = ReadRecPtr;
5148 record = ReadRecord(NULL, LOG);
5149 } while (record != NULL && recoveryContinue);
5152 * end of main redo apply loop
5156 (errmsg("redo done at %X/%X",
5157 ReadRecPtr.xlogid, ReadRecPtr.xrecoff)));
5158 if (recoveryLastXTime)
5160 (errmsg("last completed transaction was at log time %s",
5161 timestamptz_to_str(recoveryLastXTime))));
5166 /* there are no WAL records following the checkpoint */
5168 (errmsg("redo is not required")));
5173 * Re-fetch the last valid or last applied record, so we can identify the
5174 * exact endpoint of what we consider the valid portion of WAL.
5176 record = ReadRecord(&LastRec, PANIC);
5177 EndOfLog = EndRecPtr;
5178 XLByteToPrevSeg(EndOfLog, endLogId, endLogSeg);
5181 * Complain if we did not roll forward far enough to render the backup
5184 if (XLByteLT(EndOfLog, ControlFile->minRecoveryPoint))
5186 if (reachedStopPoint) /* stopped because of stop request */
5188 (errmsg("requested recovery stop point is before end time of backup dump")));
5189 else /* ran off end of WAL */
5191 (errmsg("WAL ends before end time of backup dump")));
5195 * Consider whether we need to assign a new timeline ID.
5197 * If we are doing an archive recovery, we always assign a new ID. This
5198 * handles a couple of issues. If we stopped short of the end of WAL
5199 * during recovery, then we are clearly generating a new timeline and must
5200 * assign it a unique new ID. Even if we ran to the end, modifying the
5201 * current last segment is problematic because it may result in trying to
5202 * overwrite an already-archived copy of that segment, and we encourage
5203 * DBAs to make their archive_commands reject that. We can dodge the
5204 * problem by making the new active segment have a new timeline ID.
5206 * In a normal crash recovery, we can just extend the timeline we were in.
5208 if (InArchiveRecovery)
5210 ThisTimeLineID = findNewestTimeLine(recoveryTargetTLI) + 1;
5212 (errmsg("selected new timeline ID: %u", ThisTimeLineID)));
5213 writeTimeLineHistory(ThisTimeLineID, recoveryTargetTLI,
5214 curFileTLI, endLogId, endLogSeg);
5217 /* Save the selected TimeLineID in shared memory, too */
5218 XLogCtl->ThisTimeLineID = ThisTimeLineID;
5221 * We are now done reading the old WAL. Turn off archive fetching if it
5222 * was active, and make a writable copy of the last WAL segment. (Note
5223 * that we also have a copy of the last block of the old WAL in readBuf;
5224 * we will use that below.)
5226 if (InArchiveRecovery)
5227 exitArchiveRecovery(curFileTLI, endLogId, endLogSeg);
5230 * Prepare to write WAL starting at EndOfLog position, and init xlog
5231 * buffer cache using the block containing the last record from the
5232 * previous incarnation.
5234 openLogId = endLogId;
5235 openLogSeg = endLogSeg;
5236 openLogFile = XLogFileOpen(openLogId, openLogSeg);
5238 Insert = &XLogCtl->Insert;
5239 Insert->PrevRecord = LastRec;
5240 XLogCtl->xlblocks[0].xlogid = openLogId;
5241 XLogCtl->xlblocks[0].xrecoff =
5242 ((EndOfLog.xrecoff - 1) / XLOG_BLCKSZ + 1) * XLOG_BLCKSZ;
5245 * Tricky point here: readBuf contains the *last* block that the LastRec
5246 * record spans, not the one it starts in. The last block is indeed the
5247 * one we want to use.
5249 Assert(readOff == (XLogCtl->xlblocks[0].xrecoff - XLOG_BLCKSZ) % XLogSegSize);
5250 memcpy((char *) Insert->currpage, readBuf, XLOG_BLCKSZ);
5251 Insert->currpos = (char *) Insert->currpage +
5252 (EndOfLog.xrecoff + XLOG_BLCKSZ - XLogCtl->xlblocks[0].xrecoff);
5254 LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
5256 XLogCtl->Write.LogwrtResult = LogwrtResult;
5257 Insert->LogwrtResult = LogwrtResult;
5258 XLogCtl->LogwrtResult = LogwrtResult;
5260 XLogCtl->LogwrtRqst.Write = EndOfLog;
5261 XLogCtl->LogwrtRqst.Flush = EndOfLog;
5263 freespace = INSERT_FREESPACE(Insert);
5266 /* Make sure rest of page is zero */
5267 MemSet(Insert->currpos, 0, freespace);
5268 XLogCtl->Write.curridx = 0;
5273 * Whenever Write.LogwrtResult points to exactly the end of a page,
5274 * Write.curridx must point to the *next* page (see XLogWrite()).
5276 * Note: it might seem we should do AdvanceXLInsertBuffer() here, but
5277 * this is sufficient. The first actual attempt to insert a log
5278 * record will advance the insert state.
5280 XLogCtl->Write.curridx = NextBufIdx(0);
5283 /* Pre-scan prepared transactions to find out the range of XIDs present */
5284 oldestActiveXID = PrescanPreparedTransactions();
5291 * Allow resource managers to do any required cleanup.
5293 for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
5295 if (RmgrTable[rmid].rm_cleanup != NULL)
5296 RmgrTable[rmid].rm_cleanup();
5300 * Check to see if the XLOG sequence contained any unresolved
5301 * references to uninitialized pages.
5303 XLogCheckInvalidPages();
5306 * Reset pgstat data, because it may be invalid after recovery.
5311 * Perform a checkpoint to update all our recovery activity to disk.
5313 * Note that we write a shutdown checkpoint rather than an on-line
5314 * one. This is not particularly critical, but since we may be
5315 * assigning a new TLI, using a shutdown checkpoint allows us to have
5316 * the rule that TLI only changes in shutdown checkpoints, which
5317 * allows some extra error checking in xlog_redo.
5319 CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
5322 * Close down recovery environment
5324 XLogCloseRelationCache();
5328 * Preallocate additional log files, if wanted.
5330 PreallocXlogFiles(EndOfLog);
5333 * Okay, we're officially UP.
5337 ControlFile->state = DB_IN_PRODUCTION;
5338 ControlFile->time = (pg_time_t) time(NULL);
5339 UpdateControlFile();
5341 /* start the archive_timeout timer running */
5342 XLogCtl->Write.lastSegSwitchTime = ControlFile->time;
5344 /* initialize shared-memory copy of latest checkpoint XID/epoch */
5345 XLogCtl->ckptXidEpoch = ControlFile->checkPointCopy.nextXidEpoch;
5346 XLogCtl->ckptXid = ControlFile->checkPointCopy.nextXid;
5348 /* also initialize latestCompletedXid, to nextXid - 1 */
5349 ShmemVariableCache->latestCompletedXid = ShmemVariableCache->nextXid;
5350 TransactionIdRetreat(ShmemVariableCache->latestCompletedXid);
5352 /* Start up the commit log and related stuff, too */
5354 StartupSUBTRANS(oldestActiveXID);
5357 /* Reload shared-memory state for prepared transactions */
5358 RecoverPreparedTransactions();
5360 /* Shut down readFile facility, free space */
5373 free(readRecordBuf);
5374 readRecordBuf = NULL;
5375 readRecordBufSize = 0;
5380 * Subroutine to try to fetch and validate a prior checkpoint record.
5382 * whichChkpt identifies the checkpoint (merely for reporting purposes).
5383 * 1 for "primary", 2 for "secondary", 0 for "other" (backup_label)
5386 ReadCheckpointRecord(XLogRecPtr RecPtr, int whichChkpt)
5390 if (!XRecOffIsValid(RecPtr.xrecoff))
5396 (errmsg("invalid primary checkpoint link in control file")));
5400 (errmsg("invalid secondary checkpoint link in control file")));
5404 (errmsg("invalid checkpoint link in backup_label file")));
5410 record = ReadRecord(&RecPtr, LOG);
5418 (errmsg("invalid primary checkpoint record")));
5422 (errmsg("invalid secondary checkpoint record")));
5426 (errmsg("invalid checkpoint record")));
5431 if (record->xl_rmid != RM_XLOG_ID)
5437 (errmsg("invalid resource manager ID in primary checkpoint record")));
5441 (errmsg("invalid resource manager ID in secondary checkpoint record")));
5445 (errmsg("invalid resource manager ID in checkpoint record")));
5450 if (record->xl_info != XLOG_CHECKPOINT_SHUTDOWN &&
5451 record->xl_info != XLOG_CHECKPOINT_ONLINE)
5457 (errmsg("invalid xl_info in primary checkpoint record")));
5461 (errmsg("invalid xl_info in secondary checkpoint record")));
5465 (errmsg("invalid xl_info in checkpoint record")));
5470 if (record->xl_len != sizeof(CheckPoint) ||
5471 record->xl_tot_len != SizeOfXLogRecord + sizeof(CheckPoint))
5477 (errmsg("invalid length of primary checkpoint record")));
5481 (errmsg("invalid length of secondary checkpoint record")));
5485 (errmsg("invalid length of checkpoint record")));
5494 * This must be called during startup of a backend process, except that
5495 * it need not be called in a standalone backend (which does StartupXLOG
5496 * instead). We need to initialize the local copies of ThisTimeLineID and
5499 * Note: before Postgres 8.0, we went to some effort to keep the postmaster
5500 * process's copies of ThisTimeLineID and RedoRecPtr valid too. This was
5501 * unnecessary however, since the postmaster itself never touches XLOG anyway.
5504 InitXLOGAccess(void)
5506 /* ThisTimeLineID doesn't change so we need no lock to copy it */
5507 ThisTimeLineID = XLogCtl->ThisTimeLineID;
5508 /* Use GetRedoRecPtr to copy the RedoRecPtr safely */
5509 (void) GetRedoRecPtr();
5513 * Once spawned, a backend may update its local RedoRecPtr from
5514 * XLogCtl->Insert.RedoRecPtr; it must hold the insert lock or info_lck
5515 * to do so. This is done in XLogInsert() or GetRedoRecPtr().
5520 /* use volatile pointer to prevent code rearrangement */
5521 volatile XLogCtlData *xlogctl = XLogCtl;
5523 SpinLockAcquire(&xlogctl->info_lck);
5524 Assert(XLByteLE(RedoRecPtr, xlogctl->Insert.RedoRecPtr));
5525 RedoRecPtr = xlogctl->Insert.RedoRecPtr;
5526 SpinLockRelease(&xlogctl->info_lck);
5532 * GetInsertRecPtr -- Returns the current insert position.
5534 * NOTE: The value *actually* returned is the position of the last full
5535 * xlog page. It lags behind the real insert position by at most 1 page.
5536 * For that, we don't need to acquire WALInsertLock which can be quite
5537 * heavily contended, and an approximation is enough for the current
5538 * usage of this function.
5541 GetInsertRecPtr(void)
5543 /* use volatile pointer to prevent code rearrangement */
5544 volatile XLogCtlData *xlogctl = XLogCtl;
5547 SpinLockAcquire(&xlogctl->info_lck);
5548 recptr = xlogctl->LogwrtRqst.Write;
5549 SpinLockRelease(&xlogctl->info_lck);
5555 * Get the time of the last xlog segment switch
5558 GetLastSegSwitchTime(void)
5562 /* Need WALWriteLock, but shared lock is sufficient */
5563 LWLockAcquire(WALWriteLock, LW_SHARED);
5564 result = XLogCtl->Write.lastSegSwitchTime;
5565 LWLockRelease(WALWriteLock);
5571 * GetNextXidAndEpoch - get the current nextXid value and associated epoch
5573 * This is exported for use by code that would like to have 64-bit XIDs.
5574 * We don't really support such things, but all XIDs within the system
5575 * can be presumed "close to" the result, and thus the epoch associated
5576 * with them can be determined.
5579 GetNextXidAndEpoch(TransactionId *xid, uint32 *epoch)
5581 uint32 ckptXidEpoch;
5582 TransactionId ckptXid;
5583 TransactionId nextXid;
5585 /* Must read checkpoint info first, else have race condition */
5587 /* use volatile pointer to prevent code rearrangement */
5588 volatile XLogCtlData *xlogctl = XLogCtl;
5590 SpinLockAcquire(&xlogctl->info_lck);
5591 ckptXidEpoch = xlogctl->ckptXidEpoch;
5592 ckptXid = xlogctl->ckptXid;
5593 SpinLockRelease(&xlogctl->info_lck);
5596 /* Now fetch current nextXid */
5597 nextXid = ReadNewTransactionId();
5600 * nextXid is certainly logically later than ckptXid. So if it's
5601 * numerically less, it must have wrapped into the next epoch.
5603 if (nextXid < ckptXid)
5607 *epoch = ckptXidEpoch;
5611 * This must be called ONCE during postmaster or standalone-backend shutdown
5614 ShutdownXLOG(int code, Datum arg)
5617 (errmsg("shutting down")));
5619 CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
5622 ShutdownMultiXact();
5625 (errmsg("database system is shut down")));
5629 * Log start of a checkpoint.
5632 LogCheckpointStart(int flags)
5634 elog(LOG, "checkpoint starting:%s%s%s%s%s%s",
5635 (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
5636 (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
5637 (flags & CHECKPOINT_FORCE) ? " force" : "",
5638 (flags & CHECKPOINT_WAIT) ? " wait" : "",
5639 (flags & CHECKPOINT_CAUSE_XLOG) ? " xlog" : "",
5640 (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "");
5644 * Log end of a checkpoint.
5647 LogCheckpointEnd(void)
5656 CheckpointStats.ckpt_end_t = GetCurrentTimestamp();
5658 TimestampDifference(CheckpointStats.ckpt_start_t,
5659 CheckpointStats.ckpt_end_t,
5660 &total_secs, &total_usecs);
5662 TimestampDifference(CheckpointStats.ckpt_write_t,
5663 CheckpointStats.ckpt_sync_t,
5664 &write_secs, &write_usecs);
5666 TimestampDifference(CheckpointStats.ckpt_sync_t,
5667 CheckpointStats.ckpt_sync_end_t,
5668 &sync_secs, &sync_usecs);
5670 elog(LOG, "checkpoint complete: wrote %d buffers (%.1f%%); "
5671 "%d transaction log file(s) added, %d removed, %d recycled; "
5672 "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
5673 CheckpointStats.ckpt_bufs_written,
5674 (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
5675 CheckpointStats.ckpt_segs_added,
5676 CheckpointStats.ckpt_segs_removed,
5677 CheckpointStats.ckpt_segs_recycled,
5678 write_secs, write_usecs / 1000,
5679 sync_secs, sync_usecs / 1000,
5680 total_secs, total_usecs / 1000);
5684 * Perform a checkpoint --- either during shutdown, or on-the-fly
5686 * flags is a bitwise OR of the following:
5687 * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
5688 * CHECKPOINT_IMMEDIATE: finish the checkpoint ASAP,
5689 * ignoring checkpoint_completion_target parameter.
5690 * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occured
5691 * since the last one (implied by CHECKPOINT_IS_SHUTDOWN).
5693 * Note: flags contains other bits, of interest here only for logging purposes.
5694 * In particular note that this routine is synchronous and does not pay
5695 * attention to CHECKPOINT_WAIT.
5698 CreateCheckPoint(int flags)
5700 bool shutdown = (flags & CHECKPOINT_IS_SHUTDOWN) != 0;
5701 CheckPoint checkPoint;
5703 XLogCtlInsert *Insert = &XLogCtl->Insert;
5708 TransactionId *inCommitXids;
5712 * Acquire CheckpointLock to ensure only one checkpoint happens at a time.
5713 * (This is just pro forma, since in the present system structure there is
5714 * only one process that is allowed to issue checkpoints at any given
5717 LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
5720 * Prepare to accumulate statistics.
5722 * Note: because it is possible for log_checkpoints to change while a
5723 * checkpoint proceeds, we always accumulate stats, even if
5724 * log_checkpoints is currently off.
5726 MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
5727 CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
5730 * Use a critical section to force system panic if we have trouble.
5732 START_CRIT_SECTION();
5736 ControlFile->state = DB_SHUTDOWNING;
5737 ControlFile->time = (pg_time_t) time(NULL);
5738 UpdateControlFile();
5742 * Let smgr prepare for checkpoint; this has to happen before we determine
5743 * the REDO pointer. Note that smgr must not do anything that'd have to
5744 * be undone if we decide no checkpoint is needed.
5748 /* Begin filling in the checkpoint WAL record */
5749 MemSet(&checkPoint, 0, sizeof(checkPoint));
5750 checkPoint.ThisTimeLineID = ThisTimeLineID;
5751 checkPoint.time = (pg_time_t) time(NULL);
5754 * We must hold WALInsertLock while examining insert state to determine
5755 * the checkpoint REDO pointer.
5757 LWLockAcquire(WALInsertLock, LW_EXCLUSIVE);
5760 * If this isn't a shutdown or forced checkpoint, and we have not inserted
5761 * any XLOG records since the start of the last checkpoint, skip the
5762 * checkpoint. The idea here is to avoid inserting duplicate checkpoints
5763 * when the system is idle. That wastes log space, and more importantly it
5764 * exposes us to possible loss of both current and previous checkpoint
5765 * records if the machine crashes just as we're writing the update.
5766 * (Perhaps it'd make even more sense to checkpoint only when the previous
5767 * checkpoint record is in a different xlog page?)
5769 * We have to make two tests to determine that nothing has happened since
5770 * the start of the last checkpoint: current insertion point must match
5771 * the end of the last checkpoint record, and its redo pointer must point
5774 if ((flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_FORCE)) == 0)
5776 XLogRecPtr curInsert;
5778 INSERT_RECPTR(curInsert, Insert, Insert->curridx);
5779 if (curInsert.xlogid == ControlFile->checkPoint.xlogid &&
5780 curInsert.xrecoff == ControlFile->checkPoint.xrecoff +
5781 MAXALIGN(SizeOfXLogRecord + sizeof(CheckPoint)) &&
5782 ControlFile->checkPoint.xlogid ==
5783 ControlFile->checkPointCopy.redo.xlogid &&
5784 ControlFile->checkPoint.xrecoff ==
5785 ControlFile->checkPointCopy.redo.xrecoff)
5787 LWLockRelease(WALInsertLock);
5788 LWLockRelease(CheckpointLock);
5795 * Compute new REDO record ptr = location of next XLOG record.
5797 * NB: this is NOT necessarily where the checkpoint record itself will be,
5798 * since other backends may insert more XLOG records while we're off doing
5799 * the buffer flush work. Those XLOG records are logically after the
5800 * checkpoint, even though physically before it. Got that?
5802 freespace = INSERT_FREESPACE(Insert);
5803 if (freespace < SizeOfXLogRecord)
5805 (void) AdvanceXLInsertBuffer(false);
5806 /* OK to ignore update return flag, since we will do flush anyway */
5807 freespace = INSERT_FREESPACE(Insert);
5809 INSERT_RECPTR(checkPoint.redo, Insert, Insert->curridx);
5812 * Here we update the shared RedoRecPtr for future XLogInsert calls; this
5813 * must be done while holding the insert lock AND the info_lck.
5815 * Note: if we fail to complete the checkpoint, RedoRecPtr will be left
5816 * pointing past where it really needs to point. This is okay; the only
5817 * consequence is that XLogInsert might back up whole buffers that it
5818 * didn't really need to. We can't postpone advancing RedoRecPtr because
5819 * XLogInserts that happen while we are dumping buffers must assume that
5820 * their buffer changes are not included in the checkpoint.
5823 /* use volatile pointer to prevent code rearrangement */
5824 volatile XLogCtlData *xlogctl = XLogCtl;
5826 SpinLockAcquire(&xlogctl->info_lck);
5827 RedoRecPtr = xlogctl->Insert.RedoRecPtr = checkPoint.redo;
5828 SpinLockRelease(&xlogctl->info_lck);
5832 * Now we can release WAL insert lock, allowing other xacts to proceed
5833 * while we are flushing disk buffers.
5835 LWLockRelease(WALInsertLock);
5838 * If enabled, log checkpoint start. We postpone this until now so as not
5839 * to log anything if we decided to skip the checkpoint.
5841 if (log_checkpoints)
5842 LogCheckpointStart(flags);
5845 * Before flushing data, we must wait for any transactions that are
5846 * currently in their commit critical sections. If an xact inserted its
5847 * commit record into XLOG just before the REDO point, then a crash
5848 * restart from the REDO point would not replay that record, which means
5849 * that our flushing had better include the xact's update of pg_clog. So
5850 * we wait till he's out of his commit critical section before proceeding.
5851 * See notes in RecordTransactionCommit().
5853 * Because we've already released WALInsertLock, this test is a bit fuzzy:
5854 * it is possible that we will wait for xacts we didn't really need to
5855 * wait for. But the delay should be short and it seems better to make
5856 * checkpoint take a bit longer than to hold locks longer than necessary.
5857 * (In fact, the whole reason we have this issue is that xact.c does
5858 * commit record XLOG insertion and clog update as two separate steps
5859 * protected by different locks, but again that seems best on grounds of
5860 * minimizing lock contention.)
5862 * A transaction that has not yet set inCommit when we look cannot be at
5863 * risk, since he's not inserted his commit record yet; and one that's
5864 * already cleared it is not at risk either, since he's done fixing clog
5865 * and we will correctly flush the update below. So we cannot miss any
5866 * xacts we need to wait for.
5868 nInCommit = GetTransactionsInCommit(&inCommitXids);
5873 pg_usleep(10000L); /* wait for 10 msec */
5874 } while (HaveTransactionsInCommit(inCommitXids, nInCommit));
5876 pfree(inCommitXids);
5879 * Get the other info we need for the checkpoint record.
5881 LWLockAcquire(XidGenLock, LW_SHARED);
5882 checkPoint.nextXid = ShmemVariableCache->nextXid;
5883 LWLockRelease(XidGenLock);
5885 /* Increase XID epoch if we've wrapped around since last checkpoint */
5886 checkPoint.nextXidEpoch = ControlFile->checkPointCopy.nextXidEpoch;
5887 if (checkPoint.nextXid < ControlFile->checkPointCopy.nextXid)
5888 checkPoint.nextXidEpoch++;
5890 LWLockAcquire(OidGenLock, LW_SHARED);
5891 checkPoint.nextOid = ShmemVariableCache->nextOid;
5893 checkPoint.nextOid += ShmemVariableCache->oidCount;
5894 LWLockRelease(OidGenLock);
5896 MultiXactGetCheckptMulti(shutdown,
5897 &checkPoint.nextMulti,
5898 &checkPoint.nextMultiOffset);
5901 * Having constructed the checkpoint record, ensure all shmem disk buffers
5902 * and commit-log buffers are flushed to disk.
5904 * This I/O could fail for various reasons. If so, we will fail to
5905 * complete the checkpoint, but there is no reason to force a system
5906 * panic. Accordingly, exit critical section while doing it.
5910 CheckPointGuts(checkPoint.redo, flags);
5912 START_CRIT_SECTION();
5915 * Now insert the checkpoint record into XLOG.
5917 rdata.data = (char *) (&checkPoint);
5918 rdata.len = sizeof(checkPoint);
5919 rdata.buffer = InvalidBuffer;
5922 recptr = XLogInsert(RM_XLOG_ID,
5923 shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
5924 XLOG_CHECKPOINT_ONLINE,
5930 * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
5931 * = end of actual checkpoint record.
5933 if (shutdown && !XLByteEQ(checkPoint.redo, ProcLastRecPtr))
5935 (errmsg("concurrent transaction log activity while database system is shutting down")));
5938 * Select point at which we can truncate the log, which we base on the
5939 * prior checkpoint's earliest info.
5941 XLByteToSeg(ControlFile->checkPointCopy.redo, _logId, _logSeg);
5944 * Update the control file.
5946 LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
5948 ControlFile->state = DB_SHUTDOWNED;
5949 ControlFile->prevCheckPoint = ControlFile->checkPoint;
5950 ControlFile->checkPoint = ProcLastRecPtr;
5951 ControlFile->checkPointCopy = checkPoint;
5952 ControlFile->time = (pg_time_t) time(NULL);
5953 UpdateControlFile();
5954 LWLockRelease(ControlFileLock);
5956 /* Update shared-memory copy of checkpoint XID/epoch */
5958 /* use volatile pointer to prevent code rearrangement */
5959 volatile XLogCtlData *xlogctl = XLogCtl;
5961 SpinLockAcquire(&xlogctl->info_lck);
5962 xlogctl->ckptXidEpoch = checkPoint.nextXidEpoch;
5963 xlogctl->ckptXid = checkPoint.nextXid;
5964 SpinLockRelease(&xlogctl->info_lck);
5968 * We are now done with critical updates; no need for system panic if we
5969 * have trouble while fooling with old log segments.
5974 * Let smgr do post-checkpoint cleanup (eg, deleting old files).
5979 * Delete old log files (those no longer needed even for previous
5982 if (_logId || _logSeg)
5984 PrevLogSeg(_logId, _logSeg);
5985 RemoveOldXlogFiles(_logId, _logSeg, recptr);
5989 * Make more log segments if needed. (Do this after recycling old log
5990 * segments, since that may supply some of the needed files.)
5993 PreallocXlogFiles(recptr);
5996 * Truncate pg_subtrans if possible. We can throw away all data before
5997 * the oldest XMIN of any running transaction. No future transaction will
5998 * attempt to reference any pg_subtrans entry older than that (see Asserts
5999 * in subtrans.c). During recovery, though, we mustn't do this because
6000 * StartupSUBTRANS hasn't been called yet.
6003 TruncateSUBTRANS(GetOldestXmin(true, false));
6005 /* All real work is done, but log before releasing lock. */
6006 if (log_checkpoints)
6009 LWLockRelease(CheckpointLock);
6013 * Flush all data in shared memory to disk, and fsync
6015 * This is the common code shared between regular checkpoints and
6016 * recovery restartpoints.
6019 CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
6022 CheckPointSUBTRANS();
6023 CheckPointMultiXact();
6024 CheckPointBuffers(flags); /* performs all required fsyncs */
6025 /* We deliberately delay 2PC checkpointing as long as possible */
6026 CheckPointTwoPhase(checkPointRedo);
6030 * Set a recovery restart point if appropriate
6032 * This is similar to CreateCheckPoint, but is used during WAL recovery
6033 * to establish a point from which recovery can roll forward without
6034 * replaying the entire recovery log. This function is called each time
6035 * a checkpoint record is read from XLOG; it must determine whether a
6036 * restartpoint is needed or not.
6039 RecoveryRestartPoint(const CheckPoint *checkPoint)
6045 * Do nothing if the elapsed time since the last restartpoint is less than
6046 * half of checkpoint_timeout. (We use a value less than
6047 * checkpoint_timeout so that variations in the timing of checkpoints on
6048 * the master, or speed of transmission of WAL segments to a slave, won't
6049 * make the slave skip a restartpoint once it's synced with the master.)
6050 * Checking true elapsed time keeps us from doing restartpoints too often
6051 * while rapidly scanning large amounts of WAL.
6053 elapsed_secs = (pg_time_t) time(NULL) - ControlFile->time;
6054 if (elapsed_secs < CheckPointTimeout / 2)
6058 * Is it safe to checkpoint? We must ask each of the resource managers
6059 * whether they have any partial state information that might prevent a
6060 * correct restart from this point. If so, we skip this opportunity, but
6061 * return at the next checkpoint record for another try.
6063 for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
6065 if (RmgrTable[rmid].rm_safe_restartpoint != NULL)
6066 if (!(RmgrTable[rmid].rm_safe_restartpoint()))
6068 elog(DEBUG2, "RM %d not safe to record restart point at %X/%X",
6070 checkPoint->redo.xlogid,
6071 checkPoint->redo.xrecoff);
6077 * OK, force data out to disk
6079 CheckPointGuts(checkPoint->redo, CHECKPOINT_IMMEDIATE);
6082 * Update pg_control so that any subsequent crash will restart from this
6083 * checkpoint. Note: ReadRecPtr gives the XLOG address of the checkpoint
6086 ControlFile->prevCheckPoint = ControlFile->checkPoint;
6087 ControlFile->checkPoint = ReadRecPtr;
6088 ControlFile->checkPointCopy = *checkPoint;
6089 ControlFile->time = (pg_time_t) time(NULL);
6090 UpdateControlFile();
6092 ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
6093 (errmsg("recovery restart point at %X/%X",
6094 checkPoint->redo.xlogid, checkPoint->redo.xrecoff)));
6095 if (recoveryLastXTime)
6096 ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
6097 (errmsg("last completed transaction was at log time %s",
6098 timestamptz_to_str(recoveryLastXTime))));
6102 * Write a NEXTOID log record
6105 XLogPutNextOid(Oid nextOid)
6109 rdata.data = (char *) (&nextOid);
6110 rdata.len = sizeof(Oid);
6111 rdata.buffer = InvalidBuffer;
6113 (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID, &rdata);
6116 * We need not flush the NEXTOID record immediately, because any of the
6117 * just-allocated OIDs could only reach disk as part of a tuple insert or
6118 * update that would have its own XLOG record that must follow the NEXTOID
6119 * record. Therefore, the standard buffer LSN interlock applied to those
6120 * records will ensure no such OID reaches disk before the NEXTOID record
6123 * Note, however, that the above statement only covers state "within" the
6124 * database. When we use a generated OID as a file or directory name, we
6125 * are in a sense violating the basic WAL rule, because that filesystem
6126 * change may reach disk before the NEXTOID WAL record does. The impact
6127 * of this is that if a database crash occurs immediately afterward, we
6128 * might after restart re-generate the same OID and find that it conflicts
6129 * with the leftover file or directory. But since for safety's sake we
6130 * always loop until finding a nonconflicting filename, this poses no real
6131 * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
6136 * Write an XLOG SWITCH record.
6138 * Here we just blindly issue an XLogInsert request for the record.
6139 * All the magic happens inside XLogInsert.
6141 * The return value is either the end+1 address of the switch record,
6142 * or the end+1 address of the prior segment if we did not need to
6143 * write a switch record because we are already at segment start.
6146 RequestXLogSwitch(void)
6151 /* XLOG SWITCH, alone among xlog record types, has no data */
6152 rdata.buffer = InvalidBuffer;
6157 RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH, &rdata);
6163 * XLOG resource manager's routines
6166 xlog_redo(XLogRecPtr lsn, XLogRecord *record)
6168 uint8 info = record->xl_info & ~XLR_INFO_MASK;
6170 if (info == XLOG_NEXTOID)
6174 memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
6175 if (ShmemVariableCache->nextOid < nextOid)
6177 ShmemVariableCache->nextOid = nextOid;
6178 ShmemVariableCache->oidCount = 0;
6181 else if (info == XLOG_CHECKPOINT_SHUTDOWN)
6183 CheckPoint checkPoint;
6185 memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
6186 /* In a SHUTDOWN checkpoint, believe the counters exactly */
6187 ShmemVariableCache->nextXid = checkPoint.nextXid;
6188 ShmemVariableCache->nextOid = checkPoint.nextOid;
6189 ShmemVariableCache->oidCount = 0;
6190 MultiXactSetNextMXact(checkPoint.nextMulti,
6191 checkPoint.nextMultiOffset);
6193 /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
6194 ControlFile->checkPointCopy.nextXidEpoch = checkPoint.nextXidEpoch;
6195 ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
6198 * TLI may change in a shutdown checkpoint, but it shouldn't decrease
6200 if (checkPoint.ThisTimeLineID != ThisTimeLineID)
6202 if (checkPoint.ThisTimeLineID < ThisTimeLineID ||
6203 !list_member_int(expectedTLIs,
6204 (int) checkPoint.ThisTimeLineID))
6206 (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
6207 checkPoint.ThisTimeLineID, ThisTimeLineID)));
6208 /* Following WAL records should be run with new TLI */
6209 ThisTimeLineID = checkPoint.ThisTimeLineID;
6212 RecoveryRestartPoint(&checkPoint);
6214 else if (info == XLOG_CHECKPOINT_ONLINE)
6216 CheckPoint checkPoint;
6218 memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
6219 /* In an ONLINE checkpoint, treat the counters like NEXTOID */
6220 if (TransactionIdPrecedes(ShmemVariableCache->nextXid,
6221 checkPoint.nextXid))
6222 ShmemVariableCache->nextXid = checkPoint.nextXid;
6223 if (ShmemVariableCache->nextOid < checkPoint.nextOid)
6225 ShmemVariableCache->nextOid = checkPoint.nextOid;
6226 ShmemVariableCache->oidCount = 0;
6228 MultiXactAdvanceNextMXact(checkPoint.nextMulti,
6229 checkPoint.nextMultiOffset);
6231 /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
6232 ControlFile->checkPointCopy.nextXidEpoch = checkPoint.nextXidEpoch;
6233 ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
6235 /* TLI should not change in an on-line checkpoint */
6236 if (checkPoint.ThisTimeLineID != ThisTimeLineID)
6238 (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
6239 checkPoint.ThisTimeLineID, ThisTimeLineID)));
6241 RecoveryRestartPoint(&checkPoint);
6243 else if (info == XLOG_NOOP)
6245 /* nothing to do here */
6247 else if (info == XLOG_SWITCH)
6249 /* nothing to do here */
6254 xlog_desc(StringInfo buf, uint8 xl_info, char *rec)
6256 uint8 info = xl_info & ~XLR_INFO_MASK;
6258 if (info == XLOG_CHECKPOINT_SHUTDOWN ||
6259 info == XLOG_CHECKPOINT_ONLINE)
6261 CheckPoint *checkpoint = (CheckPoint *) rec;
6263 appendStringInfo(buf, "checkpoint: redo %X/%X; "
6264 "tli %u; xid %u/%u; oid %u; multi %u; offset %u; %s",
6265 checkpoint->redo.xlogid, checkpoint->redo.xrecoff,
6266 checkpoint->ThisTimeLineID,
6267 checkpoint->nextXidEpoch, checkpoint->nextXid,
6268 checkpoint->nextOid,
6269 checkpoint->nextMulti,
6270 checkpoint->nextMultiOffset,
6271 (info == XLOG_CHECKPOINT_SHUTDOWN) ? "shutdown" : "online");
6273 else if (info == XLOG_NOOP)
6275 appendStringInfo(buf, "xlog no-op");
6277 else if (info == XLOG_NEXTOID)
6281 memcpy(&nextOid, rec, sizeof(Oid));
6282 appendStringInfo(buf, "nextOid: %u", nextOid);
6284 else if (info == XLOG_SWITCH)
6286 appendStringInfo(buf, "xlog switch");
6289 appendStringInfo(buf, "UNKNOWN");
6295 xlog_outrec(StringInfo buf, XLogRecord *record)
6299 appendStringInfo(buf, "prev %X/%X; xid %u",
6300 record->xl_prev.xlogid, record->xl_prev.xrecoff,
6303 for (i = 0; i < XLR_MAX_BKP_BLOCKS; i++)
6305 if (record->xl_info & XLR_SET_BKP_BLOCK(i))
6306 appendStringInfo(buf, "; bkpb%d", i + 1);
6309 appendStringInfo(buf, ": %s", RmgrTable[record->xl_rmid].rm_name);
6311 #endif /* WAL_DEBUG */
6318 assign_xlog_sync_method(const char *method, bool doit, GucSource source)
6320 int new_sync_method;
6323 if (pg_strcasecmp(method, "fsync") == 0)
6325 new_sync_method = SYNC_METHOD_FSYNC;
6328 #ifdef HAVE_FSYNC_WRITETHROUGH
6329 else if (pg_strcasecmp(method, "fsync_writethrough") == 0)
6331 new_sync_method = SYNC_METHOD_FSYNC_WRITETHROUGH;
6335 #ifdef HAVE_FDATASYNC
6336 else if (pg_strcasecmp(method, "fdatasync") == 0)
6338 new_sync_method = SYNC_METHOD_FDATASYNC;
6342 #ifdef OPEN_SYNC_FLAG
6343 else if (pg_strcasecmp(method, "open_sync") == 0)
6345 new_sync_method = SYNC_METHOD_OPEN;
6346 new_sync_bit = OPEN_SYNC_FLAG;
6349 #ifdef OPEN_DATASYNC_FLAG
6350 else if (pg_strcasecmp(method, "open_datasync") == 0)
6352 new_sync_method = SYNC_METHOD_OPEN;
6353 new_sync_bit = OPEN_DATASYNC_FLAG;
6362 if (sync_method != new_sync_method || open_sync_bit != new_sync_bit)
6365 * To ensure that no blocks escape unsynced, force an fsync on the
6366 * currently open log segment (if any). Also, if the open flag is
6367 * changing, close the log file so it will be reopened (with new flag
6370 if (openLogFile >= 0)
6372 if (pg_fsync(openLogFile) != 0)
6374 (errcode_for_file_access(),
6375 errmsg("could not fsync log file %u, segment %u: %m",
6376 openLogId, openLogSeg)));
6377 if (open_sync_bit != new_sync_bit)
6380 sync_method = new_sync_method;
6381 open_sync_bit = new_sync_bit;
6389 * Issue appropriate kind of fsync (if any) on the current XLOG output file
6392 issue_xlog_fsync(void)
6394 switch (sync_method)
6396 case SYNC_METHOD_FSYNC:
6397 if (pg_fsync_no_writethrough(openLogFile) != 0)
6399 (errcode_for_file_access(),
6400 errmsg("could not fsync log file %u, segment %u: %m",
6401 openLogId, openLogSeg)));
6403 #ifdef HAVE_FSYNC_WRITETHROUGH
6404 case SYNC_METHOD_FSYNC_WRITETHROUGH:
6405 if (pg_fsync_writethrough(openLogFile) != 0)
6407 (errcode_for_file_access(),
6408 errmsg("could not fsync write-through log file %u, segment %u: %m",
6409 openLogId, openLogSeg)));
6412 #ifdef HAVE_FDATASYNC
6413 case SYNC_METHOD_FDATASYNC:
6414 if (pg_fdatasync(openLogFile) != 0)
6416 (errcode_for_file_access(),
6417 errmsg("could not fdatasync log file %u, segment %u: %m",
6418 openLogId, openLogSeg)));
6421 case SYNC_METHOD_OPEN:
6422 /* write synced it already */
6425 elog(PANIC, "unrecognized wal_sync_method: %d", sync_method);
6432 * pg_start_backup: set up for taking an on-line backup dump
6434 * Essentially what this does is to create a backup label file in $PGDATA,
6435 * where it will be archived as part of the backup dump. The label file
6436 * contains the user-supplied label string (typically this would be used
6437 * to tell where the backup dump will be stored) and the starting time and
6438 * starting WAL location for the dump.
6441 pg_start_backup(PG_FUNCTION_ARGS)
6443 text *backupid = PG_GETARG_TEXT_P(0);
6445 XLogRecPtr checkpointloc;
6446 XLogRecPtr startpoint;
6447 pg_time_t stamp_time;
6449 char xlogfilename[MAXFNAMELEN];
6452 struct stat stat_buf;
6457 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
6458 errmsg("must be superuser to run a backup")));
6460 if (!XLogArchivingActive())
6462 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6463 errmsg("WAL archiving is not active"),
6464 errhint("archive_mode must be enabled at server start.")));
6466 if (!XLogArchiveCommandSet())
6468 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6469 errmsg("WAL archiving is not active"),
6470 errhint("archive_command must be defined before "
6471 "online backups can be made safely.")));
6473 backupidstr = text_to_cstring(backupid);
6476 * Mark backup active in shared memory. We must do full-page WAL writes
6477 * during an on-line backup even if not doing so at other times, because
6478 * it's quite possible for the backup dump to obtain a "torn" (partially
6479 * written) copy of a database page if it reads the page concurrently with
6480 * our write to the same page. This can be fixed as long as the first
6481 * write to the page in the WAL sequence is a full-page write. Hence, we
6482 * turn on forcePageWrites and then force a CHECKPOINT, to ensure there
6483 * are no dirty pages in shared memory that might get dumped while the
6484 * backup is in progress without having a corresponding WAL record. (Once
6485 * the backup is complete, we need not force full-page writes anymore,
6486 * since we expect that any pages not modified during the backup interval
6487 * must have been correctly captured by the backup.)
6489 * We must hold WALInsertLock to change the value of forcePageWrites, to
6490 * ensure adequate interlocking against XLogInsert().
6492 LWLockAcquire(WALInsertLock, LW_EXCLUSIVE);
6493 if (XLogCtl->Insert.forcePageWrites)
6495 LWLockRelease(WALInsertLock);
6497 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6498 errmsg("a backup is already in progress"),
6499 errhint("Run pg_stop_backup() and try again.")));
6501 XLogCtl->Insert.forcePageWrites = true;
6502 LWLockRelease(WALInsertLock);
6504 /* Ensure we release forcePageWrites if fail below */
6505 PG_ENSURE_ERROR_CLEANUP(pg_start_backup_callback, (Datum) 0);
6508 * Force a CHECKPOINT. Aside from being necessary to prevent torn
6509 * page problems, this guarantees that two successive backup runs will
6510 * have different checkpoint positions and hence different history
6511 * file names, even if nothing happened in between.
6513 * We don't use CHECKPOINT_IMMEDIATE, hence this can take awhile.
6515 RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_WAIT);
6518 * Now we need to fetch the checkpoint record location, and also its
6519 * REDO pointer. The oldest point in WAL that would be needed to
6520 * restore starting from the checkpoint is precisely the REDO pointer.
6522 LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6523 checkpointloc = ControlFile->checkPoint;
6524 startpoint = ControlFile->checkPointCopy.redo;
6525 LWLockRelease(ControlFileLock);
6527 XLByteToSeg(startpoint, _logId, _logSeg);
6528 XLogFileName(xlogfilename, ThisTimeLineID, _logId, _logSeg);
6530 /* Use the log timezone here, not the session timezone */
6531 stamp_time = (pg_time_t) time(NULL);
6532 pg_strftime(strfbuf, sizeof(strfbuf),
6533 "%Y-%m-%d %H:%M:%S %Z",
6534 pg_localtime(&stamp_time, log_timezone));
6537 * Check for existing backup label --- implies a backup is already
6538 * running. (XXX given that we checked forcePageWrites above, maybe
6539 * it would be OK to just unlink any such label file?)
6541 if (stat(BACKUP_LABEL_FILE, &stat_buf) != 0)
6543 if (errno != ENOENT)
6545 (errcode_for_file_access(),
6546 errmsg("could not stat file \"%s\": %m",
6547 BACKUP_LABEL_FILE)));
6551 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6552 errmsg("a backup is already in progress"),
6553 errhint("If you're sure there is no backup in progress, remove file \"%s\" and try again.",
6554 BACKUP_LABEL_FILE)));
6557 * Okay, write the file
6559 fp = AllocateFile(BACKUP_LABEL_FILE, "w");
6562 (errcode_for_file_access(),
6563 errmsg("could not create file \"%s\": %m",
6564 BACKUP_LABEL_FILE)));
6565 fprintf(fp, "START WAL LOCATION: %X/%X (file %s)\n",
6566 startpoint.xlogid, startpoint.xrecoff, xlogfilename);
6567 fprintf(fp, "CHECKPOINT LOCATION: %X/%X\n",
6568 checkpointloc.xlogid, checkpointloc.xrecoff);
6569 fprintf(fp, "START TIME: %s\n", strfbuf);
6570 fprintf(fp, "LABEL: %s\n", backupidstr);
6571 if (fflush(fp) || ferror(fp) || FreeFile(fp))
6573 (errcode_for_file_access(),
6574 errmsg("could not write file \"%s\": %m",
6575 BACKUP_LABEL_FILE)));
6577 PG_END_ENSURE_ERROR_CLEANUP(pg_start_backup_callback, (Datum) 0);
6580 * We're done. As a convenience, return the starting WAL location.
6582 snprintf(xlogfilename, sizeof(xlogfilename), "%X/%X",
6583 startpoint.xlogid, startpoint.xrecoff);
6584 PG_RETURN_TEXT_P(cstring_to_text(xlogfilename));
6587 /* Error cleanup callback for pg_start_backup */
6589 pg_start_backup_callback(int code, Datum arg)
6591 /* Turn off forcePageWrites on failure */
6592 LWLockAcquire(WALInsertLock, LW_EXCLUSIVE);
6593 XLogCtl->Insert.forcePageWrites = false;
6594 LWLockRelease(WALInsertLock);
6598 * pg_stop_backup: finish taking an on-line backup dump
6600 * We remove the backup label file created by pg_start_backup, and instead
6601 * create a backup history file in pg_xlog (whence it will immediately be
6602 * archived). The backup history file contains the same info found in
6603 * the label file, plus the backup-end time and WAL location.
6604 * Note: different from CancelBackup which just cancels online backup mode.
6607 pg_stop_backup(PG_FUNCTION_ARGS)
6609 XLogRecPtr startpoint;
6610 XLogRecPtr stoppoint;
6611 pg_time_t stamp_time;
6613 char histfilepath[MAXPGPATH];
6614 char startxlogfilename[MAXFNAMELEN];
6615 char stopxlogfilename[MAXFNAMELEN];
6622 int seconds_before_warning;
6627 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
6628 (errmsg("must be superuser to run a backup"))));
6631 * OK to clear forcePageWrites
6633 LWLockAcquire(WALInsertLock, LW_EXCLUSIVE);
6634 XLogCtl->Insert.forcePageWrites = false;
6635 LWLockRelease(WALInsertLock);
6638 * Force a switch to a new xlog segment file, so that the backup is valid
6639 * as soon as archiver moves out the current segment file. We'll report
6640 * the end address of the XLOG SWITCH record as the backup stopping point.
6642 stoppoint = RequestXLogSwitch();
6644 XLByteToSeg(stoppoint, _logId, _logSeg);
6645 XLogFileName(stopxlogfilename, ThisTimeLineID, _logId, _logSeg);
6647 /* Use the log timezone here, not the session timezone */
6648 stamp_time = (pg_time_t) time(NULL);
6649 pg_strftime(strfbuf, sizeof(strfbuf),
6650 "%Y-%m-%d %H:%M:%S %Z",
6651 pg_localtime(&stamp_time, log_timezone));
6654 * Open the existing label file
6656 lfp = AllocateFile(BACKUP_LABEL_FILE, "r");
6659 if (errno != ENOENT)
6661 (errcode_for_file_access(),
6662 errmsg("could not read file \"%s\": %m",
6663 BACKUP_LABEL_FILE)));
6665 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6666 errmsg("a backup is not in progress")));
6670 * Read and parse the START WAL LOCATION line (this code is pretty crude,
6671 * but we are not expecting any variability in the file format).
6673 if (fscanf(lfp, "START WAL LOCATION: %X/%X (file %24s)%c",
6674 &startpoint.xlogid, &startpoint.xrecoff, startxlogfilename,
6675 &ch) != 4 || ch != '\n')
6677 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6678 errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
6681 * Write the backup history file
6683 XLByteToSeg(startpoint, _logId, _logSeg);
6684 BackupHistoryFilePath(histfilepath, ThisTimeLineID, _logId, _logSeg,
6685 startpoint.xrecoff % XLogSegSize);
6686 fp = AllocateFile(histfilepath, "w");
6689 (errcode_for_file_access(),
6690 errmsg("could not create file \"%s\": %m",
6692 fprintf(fp, "START WAL LOCATION: %X/%X (file %s)\n",
6693 startpoint.xlogid, startpoint.xrecoff, startxlogfilename);
6694 fprintf(fp, "STOP WAL LOCATION: %X/%X (file %s)\n",
6695 stoppoint.xlogid, stoppoint.xrecoff, stopxlogfilename);
6696 /* transfer remaining lines from label to history file */
6697 while ((ich = fgetc(lfp)) != EOF)
6699 fprintf(fp, "STOP TIME: %s\n", strfbuf);
6700 if (fflush(fp) || ferror(fp) || FreeFile(fp))
6702 (errcode_for_file_access(),
6703 errmsg("could not write file \"%s\": %m",
6707 * Close and remove the backup label file
6709 if (ferror(lfp) || FreeFile(lfp))
6711 (errcode_for_file_access(),
6712 errmsg("could not read file \"%s\": %m",
6713 BACKUP_LABEL_FILE)));
6714 if (unlink(BACKUP_LABEL_FILE) != 0)
6716 (errcode_for_file_access(),
6717 errmsg("could not remove file \"%s\": %m",
6718 BACKUP_LABEL_FILE)));
6721 * Clean out any no-longer-needed history files. As a side effect, this
6722 * will post a .ready file for the newly created history file, notifying
6723 * the archiver that history file may be archived immediately.
6725 CleanupBackupHistory();
6728 * Wait until the history file has been archived. We assume that the
6729 * alphabetic sorting property of the WAL files ensures the last WAL
6730 * file is guaranteed archived by the time the history file is archived.
6732 * We wait forever, since archive_command is supposed to work and
6733 * we assume the admin wanted his backup to work completely. If you
6734 * don't wish to wait, you can SET statement_timeout = xx;
6736 * If the status file is missing, we assume that is because it was
6737 * set to .ready before we slept, then while asleep it has been set
6738 * to .done and then removed by a concurrent checkpoint.
6740 BackupHistoryFileName(histfilepath, ThisTimeLineID, _logId, _logSeg,
6741 startpoint.xrecoff % XLogSegSize);
6743 seconds_before_warning = 60;
6746 while (!XLogArchiveCheckDone(histfilepath, false))
6748 CHECK_FOR_INTERRUPTS();
6750 pg_usleep(1000000L);
6752 if (++waits >= seconds_before_warning)
6754 seconds_before_warning *= 2; /* This wraps in >10 years... */
6755 elog(WARNING, "pg_stop_backup() waiting for archive to complete "
6756 "(%d seconds delay)", waits);
6761 * We're done. As a convenience, return the ending WAL location.
6763 snprintf(stopxlogfilename, sizeof(stopxlogfilename), "%X/%X",
6764 stoppoint.xlogid, stoppoint.xrecoff);
6765 PG_RETURN_TEXT_P(cstring_to_text(stopxlogfilename));
6769 * pg_switch_xlog: switch to next xlog file
6772 pg_switch_xlog(PG_FUNCTION_ARGS)
6774 XLogRecPtr switchpoint;
6775 char location[MAXFNAMELEN];
6779 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
6780 (errmsg("must be superuser to switch transaction log files"))));
6782 switchpoint = RequestXLogSwitch();
6785 * As a convenience, return the WAL location of the switch record
6787 snprintf(location, sizeof(location), "%X/%X",
6788 switchpoint.xlogid, switchpoint.xrecoff);
6789 PG_RETURN_TEXT_P(cstring_to_text(location));
6793 * Report the current WAL write location (same format as pg_start_backup etc)
6795 * This is useful for determining how much of WAL is visible to an external
6796 * archiving process. Note that the data before this point is written out
6797 * to the kernel, but is not necessarily synced to disk.
6800 pg_current_xlog_location(PG_FUNCTION_ARGS)
6802 char location[MAXFNAMELEN];
6804 /* Make sure we have an up-to-date local LogwrtResult */
6806 /* use volatile pointer to prevent code rearrangement */
6807 volatile XLogCtlData *xlogctl = XLogCtl;
6809 SpinLockAcquire(&xlogctl->info_lck);
6810 LogwrtResult = xlogctl->LogwrtResult;
6811 SpinLockRelease(&xlogctl->info_lck);
6814 snprintf(location, sizeof(location), "%X/%X",
6815 LogwrtResult.Write.xlogid, LogwrtResult.Write.xrecoff);
6816 PG_RETURN_TEXT_P(cstring_to_text(location));
6820 * Report the current WAL insert location (same format as pg_start_backup etc)
6822 * This function is mostly for debugging purposes.
6825 pg_current_xlog_insert_location(PG_FUNCTION_ARGS)
6827 XLogCtlInsert *Insert = &XLogCtl->Insert;
6828 XLogRecPtr current_recptr;
6829 char location[MAXFNAMELEN];
6832 * Get the current end-of-WAL position ... shared lock is sufficient
6834 LWLockAcquire(WALInsertLock, LW_SHARED);
6835 INSERT_RECPTR(current_recptr, Insert, Insert->curridx);
6836 LWLockRelease(WALInsertLock);
6838 snprintf(location, sizeof(location), "%X/%X",
6839 current_recptr.xlogid, current_recptr.xrecoff);
6840 PG_RETURN_TEXT_P(cstring_to_text(location));
6844 * Compute an xlog file name and decimal byte offset given a WAL location,
6845 * such as is returned by pg_stop_backup() or pg_xlog_switch().
6847 * Note that a location exactly at a segment boundary is taken to be in
6848 * the previous segment. This is usually the right thing, since the
6849 * expected usage is to determine which xlog file(s) are ready to archive.
6852 pg_xlogfile_name_offset(PG_FUNCTION_ARGS)
6854 text *location = PG_GETARG_TEXT_P(0);
6856 unsigned int uxlogid;
6857 unsigned int uxrecoff;
6861 XLogRecPtr locationpoint;
6862 char xlogfilename[MAXFNAMELEN];
6865 TupleDesc resultTupleDesc;
6866 HeapTuple resultHeapTuple;
6870 * Read input and parse
6872 locationstr = text_to_cstring(location);
6874 if (sscanf(locationstr, "%X/%X", &uxlogid, &uxrecoff) != 2)
6876 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
6877 errmsg("could not parse transaction log location \"%s\"",
6880 locationpoint.xlogid = uxlogid;
6881 locationpoint.xrecoff = uxrecoff;
6884 * Construct a tuple descriptor for the result row. This must match this
6885 * function's pg_proc entry!
6887 resultTupleDesc = CreateTemplateTupleDesc(2, false);
6888 TupleDescInitEntry(resultTupleDesc, (AttrNumber) 1, "file_name",
6890 TupleDescInitEntry(resultTupleDesc, (AttrNumber) 2, "file_offset",
6893 resultTupleDesc = BlessTupleDesc(resultTupleDesc);
6898 XLByteToPrevSeg(locationpoint, xlogid, xlogseg);
6899 XLogFileName(xlogfilename, ThisTimeLineID, xlogid, xlogseg);
6901 values[0] = CStringGetTextDatum(xlogfilename);
6907 xrecoff = locationpoint.xrecoff - xlogseg * XLogSegSize;
6909 values[1] = UInt32GetDatum(xrecoff);
6913 * Tuple jam: Having first prepared your Datums, then squash together
6915 resultHeapTuple = heap_form_tuple(resultTupleDesc, values, isnull);
6917 result = HeapTupleGetDatum(resultHeapTuple);
6919 PG_RETURN_DATUM(result);
6923 * Compute an xlog file name given a WAL location,
6924 * such as is returned by pg_stop_backup() or pg_xlog_switch().
6927 pg_xlogfile_name(PG_FUNCTION_ARGS)
6929 text *location = PG_GETARG_TEXT_P(0);
6931 unsigned int uxlogid;
6932 unsigned int uxrecoff;
6935 XLogRecPtr locationpoint;
6936 char xlogfilename[MAXFNAMELEN];
6938 locationstr = text_to_cstring(location);
6940 if (sscanf(locationstr, "%X/%X", &uxlogid, &uxrecoff) != 2)
6942 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
6943 errmsg("could not parse transaction log location \"%s\"",
6946 locationpoint.xlogid = uxlogid;
6947 locationpoint.xrecoff = uxrecoff;
6949 XLByteToPrevSeg(locationpoint, xlogid, xlogseg);
6950 XLogFileName(xlogfilename, ThisTimeLineID, xlogid, xlogseg);
6952 PG_RETURN_TEXT_P(cstring_to_text(xlogfilename));
6956 * read_backup_label: check to see if a backup_label file is present
6958 * If we see a backup_label during recovery, we assume that we are recovering
6959 * from a backup dump file, and we therefore roll forward from the checkpoint
6960 * identified by the label file, NOT what pg_control says. This avoids the
6961 * problem that pg_control might have been archived one or more checkpoints
6962 * later than the start of the dump, and so if we rely on it as the start
6963 * point, we will fail to restore a consistent database state.
6965 * We also attempt to retrieve the corresponding backup history file.
6966 * If successful, set *minRecoveryLoc to constrain valid PITR stopping
6969 * Returns TRUE if a backup_label was found (and fills the checkpoint
6970 * location into *checkPointLoc); returns FALSE if not.
6973 read_backup_label(XLogRecPtr *checkPointLoc, XLogRecPtr *minRecoveryLoc)
6975 XLogRecPtr startpoint;
6976 XLogRecPtr stoppoint;
6977 char histfilename[MAXFNAMELEN];
6978 char histfilepath[MAXPGPATH];
6979 char startxlogfilename[MAXFNAMELEN];
6980 char stopxlogfilename[MAXFNAMELEN];
6988 /* Default is to not constrain recovery stop point */
6989 minRecoveryLoc->xlogid = 0;
6990 minRecoveryLoc->xrecoff = 0;
6993 * See if label file is present
6995 lfp = AllocateFile(BACKUP_LABEL_FILE, "r");
6998 if (errno != ENOENT)
7000 (errcode_for_file_access(),
7001 errmsg("could not read file \"%s\": %m",
7002 BACKUP_LABEL_FILE)));
7003 return false; /* it's not there, all is fine */
7007 * Read and parse the START WAL LOCATION and CHECKPOINT lines (this code
7008 * is pretty crude, but we are not expecting any variability in the file
7011 if (fscanf(lfp, "START WAL LOCATION: %X/%X (file %08X%16s)%c",
7012 &startpoint.xlogid, &startpoint.xrecoff, &tli,
7013 startxlogfilename, &ch) != 5 || ch != '\n')
7015 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
7016 errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
7017 if (fscanf(lfp, "CHECKPOINT LOCATION: %X/%X%c",
7018 &checkPointLoc->xlogid, &checkPointLoc->xrecoff,
7019 &ch) != 3 || ch != '\n')
7021 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
7022 errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
7023 if (ferror(lfp) || FreeFile(lfp))
7025 (errcode_for_file_access(),
7026 errmsg("could not read file \"%s\": %m",
7027 BACKUP_LABEL_FILE)));
7030 * Try to retrieve the backup history file (no error if we can't)
7032 XLByteToSeg(startpoint, _logId, _logSeg);
7033 BackupHistoryFileName(histfilename, tli, _logId, _logSeg,
7034 startpoint.xrecoff % XLogSegSize);
7036 if (InArchiveRecovery)
7037 RestoreArchivedFile(histfilepath, histfilename, "RECOVERYHISTORY", 0);
7039 BackupHistoryFilePath(histfilepath, tli, _logId, _logSeg,
7040 startpoint.xrecoff % XLogSegSize);
7042 fp = AllocateFile(histfilepath, "r");
7046 * Parse history file to identify stop point.
7048 if (fscanf(fp, "START WAL LOCATION: %X/%X (file %24s)%c",
7049 &startpoint.xlogid, &startpoint.xrecoff, startxlogfilename,
7050 &ch) != 4 || ch != '\n')
7052 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
7053 errmsg("invalid data in file \"%s\"", histfilename)));
7054 if (fscanf(fp, "STOP WAL LOCATION: %X/%X (file %24s)%c",
7055 &stoppoint.xlogid, &stoppoint.xrecoff, stopxlogfilename,
7056 &ch) != 4 || ch != '\n')
7058 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
7059 errmsg("invalid data in file \"%s\"", histfilename)));
7060 *minRecoveryLoc = stoppoint;
7061 if (ferror(fp) || FreeFile(fp))
7063 (errcode_for_file_access(),
7064 errmsg("could not read file \"%s\": %m",
7072 * Error context callback for errors occurring during rm_redo().
7075 rm_redo_error_callback(void *arg)
7077 XLogRecord *record = (XLogRecord *) arg;
7080 initStringInfo(&buf);
7081 RmgrTable[record->xl_rmid].rm_desc(&buf,
7083 XLogRecGetData(record));
7085 /* don't bother emitting empty description */
7087 errcontext("xlog redo %s", buf.data);
7093 * BackupInProgress: check if online backup mode is active
7095 * This is done by checking for existence of the "backup_label" file.
7098 BackupInProgress(void)
7100 struct stat stat_buf;
7102 return (stat(BACKUP_LABEL_FILE, &stat_buf) == 0);
7106 * CancelBackup: rename the "backup_label" file to cancel backup mode
7108 * If the "backup_label" file exists, it will be renamed to "backup_label.old".
7109 * Note that this will render an online backup in progress useless.
7110 * To correctly finish an online backup, pg_stop_backup must be called.
7115 struct stat stat_buf;
7117 /* if the file is not there, return */
7118 if (stat(BACKUP_LABEL_FILE, &stat_buf) < 0)
7121 /* remove leftover file from previously cancelled backup if it exists */
7122 unlink(BACKUP_LABEL_OLD);
7124 if (rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD) == 0)
7127 (errmsg("online backup mode cancelled"),
7128 errdetail("\"%s\" was renamed to \"%s\".",
7129 BACKUP_LABEL_FILE, BACKUP_LABEL_OLD)));
7134 (errcode_for_file_access(),
7135 errmsg("online backup mode was not cancelled"),
7136 errdetail("Could not rename \"%s\" to \"%s\": %m.",
7137 BACKUP_LABEL_FILE, BACKUP_LABEL_OLD)));