Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * xlog.c
4 : * PostgreSQL write-ahead log manager
5 : *
6 : *
7 : * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
8 : * Portions Copyright (c) 1994, Regents of the University of California
9 : *
10 : * src/backend/access/transam/xlog.c
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 :
15 : #include "postgres.h"
16 :
17 : #include <ctype.h>
18 : #include <math.h>
19 : #include <time.h>
20 : #include <fcntl.h>
21 : #include <sys/stat.h>
22 : #include <sys/time.h>
23 : #include <unistd.h>
24 :
25 : #include "access/clog.h"
26 : #include "access/commit_ts.h"
27 : #include "access/multixact.h"
28 : #include "access/rewriteheap.h"
29 : #include "access/subtrans.h"
30 : #include "access/timeline.h"
31 : #include "access/transam.h"
32 : #include "access/tuptoaster.h"
33 : #include "access/twophase.h"
34 : #include "access/xact.h"
35 : #include "access/xlog_internal.h"
36 : #include "access/xloginsert.h"
37 : #include "access/xlogreader.h"
38 : #include "access/xlogutils.h"
39 : #include "catalog/catversion.h"
40 : #include "catalog/pg_control.h"
41 : #include "catalog/pg_database.h"
42 : #include "commands/tablespace.h"
43 : #include "miscadmin.h"
44 : #include "pgstat.h"
45 : #include "port/atomics.h"
46 : #include "postmaster/bgwriter.h"
47 : #include "postmaster/walwriter.h"
48 : #include "postmaster/startup.h"
49 : #include "replication/basebackup.h"
50 : #include "replication/logical.h"
51 : #include "replication/slot.h"
52 : #include "replication/origin.h"
53 : #include "replication/snapbuild.h"
54 : #include "replication/walreceiver.h"
55 : #include "replication/walsender.h"
56 : #include "storage/bufmgr.h"
57 : #include "storage/fd.h"
58 : #include "storage/ipc.h"
59 : #include "storage/large_object.h"
60 : #include "storage/latch.h"
61 : #include "storage/pmsignal.h"
62 : #include "storage/predicate.h"
63 : #include "storage/proc.h"
64 : #include "storage/procarray.h"
65 : #include "storage/reinit.h"
66 : #include "storage/smgr.h"
67 : #include "storage/spin.h"
68 : #include "utils/backend_random.h"
69 : #include "utils/builtins.h"
70 : #include "utils/guc.h"
71 : #include "utils/memutils.h"
72 : #include "utils/pg_lsn.h"
73 : #include "utils/ps_status.h"
74 : #include "utils/relmapper.h"
75 : #include "utils/snapmgr.h"
76 : #include "utils/timestamp.h"
77 : #include "pg_trace.h"
78 :
79 : extern uint32 bootstrap_data_checksum_version;
80 :
81 : /* File path names (all relative to $PGDATA) */
82 : #define RECOVERY_COMMAND_FILE "recovery.conf"
83 : #define RECOVERY_COMMAND_DONE "recovery.done"
84 : #define PROMOTE_SIGNAL_FILE "promote"
85 : #define FALLBACK_PROMOTE_SIGNAL_FILE "fallback_promote"
86 :
87 :
88 : /* User-settable parameters */
89 : int max_wal_size_mb = 1024; /* 1 GB */
90 : int min_wal_size_mb = 80; /* 80 MB */
91 : int wal_keep_segments = 0;
92 : int XLOGbuffers = -1;
93 : int XLogArchiveTimeout = 0;
94 : int XLogArchiveMode = ARCHIVE_MODE_OFF;
95 : char *XLogArchiveCommand = NULL;
96 : bool EnableHotStandby = false;
97 : bool fullPageWrites = true;
98 : bool wal_log_hints = false;
99 : bool wal_compression = false;
100 : char *wal_consistency_checking_string = NULL;
101 : bool *wal_consistency_checking = NULL;
102 : bool log_checkpoints = false;
103 : int sync_method = DEFAULT_SYNC_METHOD;
104 : int wal_level = WAL_LEVEL_MINIMAL;
105 : int CommitDelay = 0; /* precommit delay in microseconds */
106 : int CommitSiblings = 5; /* # concurrent xacts needed to sleep */
107 : int wal_retrieve_retry_interval = 5000;
108 :
109 : #ifdef WAL_DEBUG
110 : bool XLOG_DEBUG = false;
111 : #endif
112 :
113 : /*
114 : * Number of WAL insertion locks to use. A higher value allows more insertions
115 : * to happen concurrently, but adds some CPU overhead to flushing the WAL,
116 : * which needs to iterate all the locks.
117 : */
118 : #define NUM_XLOGINSERT_LOCKS 8
119 :
120 : /*
121 : * Max distance from last checkpoint, before triggering a new xlog-based
122 : * checkpoint.
123 : */
124 : int CheckPointSegments;
125 :
126 : /* Estimated distance between checkpoints, in bytes */
127 : static double CheckPointDistanceEstimate = 0;
128 : static double PrevCheckPointDistance = 0;
129 :
130 : /*
131 : * GUC support
132 : */
133 : const struct config_enum_entry sync_method_options[] = {
134 : {"fsync", SYNC_METHOD_FSYNC, false},
135 : #ifdef HAVE_FSYNC_WRITETHROUGH
136 : {"fsync_writethrough", SYNC_METHOD_FSYNC_WRITETHROUGH, false},
137 : #endif
138 : #ifdef HAVE_FDATASYNC
139 : {"fdatasync", SYNC_METHOD_FDATASYNC, false},
140 : #endif
141 : #ifdef OPEN_SYNC_FLAG
142 : {"open_sync", SYNC_METHOD_OPEN, false},
143 : #endif
144 : #ifdef OPEN_DATASYNC_FLAG
145 : {"open_datasync", SYNC_METHOD_OPEN_DSYNC, false},
146 : #endif
147 : {NULL, 0, false}
148 : };
149 :
150 :
151 : /*
152 : * Although only "on", "off", and "always" are documented,
153 : * we accept all the likely variants of "on" and "off".
154 : */
155 : const struct config_enum_entry archive_mode_options[] = {
156 : {"always", ARCHIVE_MODE_ALWAYS, false},
157 : {"on", ARCHIVE_MODE_ON, false},
158 : {"off", ARCHIVE_MODE_OFF, false},
159 : {"true", ARCHIVE_MODE_ON, true},
160 : {"false", ARCHIVE_MODE_OFF, true},
161 : {"yes", ARCHIVE_MODE_ON, true},
162 : {"no", ARCHIVE_MODE_OFF, true},
163 : {"1", ARCHIVE_MODE_ON, true},
164 : {"0", ARCHIVE_MODE_OFF, true},
165 : {NULL, 0, false}
166 : };
167 :
168 : /*
169 : * Statistics for current checkpoint are collected in this global struct.
170 : * Because only the checkpointer or a stand-alone backend can perform
171 : * checkpoints, this will be unused in normal backends.
172 : */
173 : CheckpointStatsData CheckpointStats;
174 :
175 : /*
176 : * ThisTimeLineID will be same in all backends --- it identifies current
177 : * WAL timeline for the database system.
178 : */
179 : TimeLineID ThisTimeLineID = 0;
180 :
181 : /*
182 : * Are we doing recovery from XLOG?
183 : *
184 : * This is only ever true in the startup process; it should be read as meaning
185 : * "this process is replaying WAL records", rather than "the system is in
186 : * recovery mode". It should be examined primarily by functions that need
187 : * to act differently when called from a WAL redo function (e.g., to skip WAL
188 : * logging). To check whether the system is in recovery regardless of which
189 : * process you're running in, use RecoveryInProgress() but only after shared
190 : * memory startup and lock initialization.
191 : */
192 : bool InRecovery = false;
193 :
194 : /* Are we in Hot Standby mode? Only valid in startup process, see xlog.h */
195 : HotStandbyState standbyState = STANDBY_DISABLED;
196 :
197 : static XLogRecPtr LastRec;
198 :
199 : /* Local copy of WalRcv->receivedUpto */
200 : static XLogRecPtr receivedUpto = 0;
201 : static TimeLineID receiveTLI = 0;
202 :
203 : /*
204 : * During recovery, lastFullPageWrites keeps track of full_page_writes that
205 : * the replayed WAL records indicate. It's initialized with full_page_writes
206 : * that the recovery starting checkpoint record indicates, and then updated
207 : * each time XLOG_FPW_CHANGE record is replayed.
208 : */
209 : static bool lastFullPageWrites;
210 :
211 : /*
212 : * Local copy of SharedRecoveryInProgress variable. True actually means "not
213 : * known, need to check the shared state".
214 : */
215 : static bool LocalRecoveryInProgress = true;
216 :
217 : /*
218 : * Local copy of SharedHotStandbyActive variable. False actually means "not
219 : * known, need to check the shared state".
220 : */
221 : static bool LocalHotStandbyActive = false;
222 :
223 : /*
224 : * Local state for XLogInsertAllowed():
225 : * 1: unconditionally allowed to insert XLOG
226 : * 0: unconditionally not allowed to insert XLOG
227 : * -1: must check RecoveryInProgress(); disallow until it is false
228 : * Most processes start with -1 and transition to 1 after seeing that recovery
229 : * is not in progress. But we can also force the value for special cases.
230 : * The coding in XLogInsertAllowed() depends on the first two of these states
231 : * being numerically the same as bool true and false.
232 : */
233 : static int LocalXLogInsertAllowed = -1;
234 :
235 : /*
236 : * When ArchiveRecoveryRequested is set, archive recovery was requested,
237 : * ie. recovery.conf file was present. When InArchiveRecovery is set, we are
238 : * currently recovering using offline XLOG archives. These variables are only
239 : * valid in the startup process.
240 : *
241 : * When ArchiveRecoveryRequested is true, but InArchiveRecovery is false, we're
242 : * currently performing crash recovery using only XLOG files in pg_wal, but
243 : * will switch to using offline XLOG archives as soon as we reach the end of
244 : * WAL in pg_wal.
245 : */
246 : bool ArchiveRecoveryRequested = false;
247 : bool InArchiveRecovery = false;
248 :
249 : /* Was the last xlog file restored from archive, or local? */
250 : static bool restoredFromArchive = false;
251 :
252 : /* Buffers dedicated to consistency checks of size BLCKSZ */
253 : static char *replay_image_masked = NULL;
254 : static char *master_image_masked = NULL;
255 :
256 : /* options taken from recovery.conf for archive recovery */
257 : char *recoveryRestoreCommand = NULL;
258 : static char *recoveryEndCommand = NULL;
259 : static char *archiveCleanupCommand = NULL;
260 : static RecoveryTargetType recoveryTarget = RECOVERY_TARGET_UNSET;
261 : static bool recoveryTargetInclusive = true;
262 : static RecoveryTargetAction recoveryTargetAction = RECOVERY_TARGET_ACTION_PAUSE;
263 : static TransactionId recoveryTargetXid;
264 : static TimestampTz recoveryTargetTime;
265 : static char *recoveryTargetName;
266 : static XLogRecPtr recoveryTargetLSN;
267 : static int recovery_min_apply_delay = 0;
268 : static TimestampTz recoveryDelayUntilTime;
269 :
270 : /* options taken from recovery.conf for XLOG streaming */
271 : static bool StandbyModeRequested = false;
272 : static char *PrimaryConnInfo = NULL;
273 : static char *PrimarySlotName = NULL;
274 : static char *TriggerFile = NULL;
275 :
276 : /* are we currently in standby mode? */
277 : bool StandbyMode = false;
278 :
279 : /* whether request for fast promotion has been made yet */
280 : static bool fast_promote = false;
281 :
282 : /*
283 : * if recoveryStopsBefore/After returns true, it saves information of the stop
284 : * point here
285 : */
286 : static TransactionId recoveryStopXid;
287 : static TimestampTz recoveryStopTime;
288 : static XLogRecPtr recoveryStopLSN;
289 : static char recoveryStopName[MAXFNAMELEN];
290 : static bool recoveryStopAfter;
291 :
292 : /*
293 : * During normal operation, the only timeline we care about is ThisTimeLineID.
294 : * During recovery, however, things are more complicated. To simplify life
295 : * for rmgr code, we keep ThisTimeLineID set to the "current" timeline as we
296 : * scan through the WAL history (that is, it is the line that was active when
297 : * the currently-scanned WAL record was generated). We also need these
298 : * timeline values:
299 : *
300 : * recoveryTargetTLI: the desired timeline that we want to end in.
301 : *
302 : * recoveryTargetIsLatest: was the requested target timeline 'latest'?
303 : *
304 : * expectedTLEs: a list of TimeLineHistoryEntries for recoveryTargetTLI and the timelines of
305 : * its known parents, newest first (so recoveryTargetTLI is always the
306 : * first list member). Only these TLIs are expected to be seen in the WAL
307 : * segments we read, and indeed only these TLIs will be considered as
308 : * candidate WAL files to open at all.
309 : *
310 : * curFileTLI: the TLI appearing in the name of the current input WAL file.
311 : * (This is not necessarily the same as ThisTimeLineID, because we could
312 : * be scanning data that was copied from an ancestor timeline when the current
313 : * file was created.) During a sequential scan we do not allow this value
314 : * to decrease.
315 : */
316 : static TimeLineID recoveryTargetTLI;
317 : static bool recoveryTargetIsLatest = false;
318 : static List *expectedTLEs;
319 : static TimeLineID curFileTLI;
320 :
321 : /*
322 : * ProcLastRecPtr points to the start of the last XLOG record inserted by the
323 : * current backend. It is updated for all inserts. XactLastRecEnd points to
324 : * end+1 of the last record, and is reset when we end a top-level transaction,
325 : * or start a new one; so it can be used to tell if the current transaction has
326 : * created any XLOG records.
327 : *
328 : * While in parallel mode, this may not be fully up to date. When committing,
329 : * a transaction can assume this covers all xlog records written either by the
330 : * user backend or by any parallel worker which was present at any point during
331 : * the transaction. But when aborting, or when still in parallel mode, other
332 : * parallel backends may have written WAL records at later LSNs than the value
333 : * stored here. The parallel leader advances its own copy, when necessary,
334 : * in WaitForParallelWorkersToFinish.
335 : */
336 : XLogRecPtr ProcLastRecPtr = InvalidXLogRecPtr;
337 : XLogRecPtr XactLastRecEnd = InvalidXLogRecPtr;
338 : XLogRecPtr XactLastCommitEnd = InvalidXLogRecPtr;
339 :
340 : /*
341 : * RedoRecPtr is this backend's local copy of the REDO record pointer
342 : * (which is almost but not quite the same as a pointer to the most recent
343 : * CHECKPOINT record). We update this from the shared-memory copy,
344 : * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
345 : * hold an insertion lock). See XLogInsertRecord for details. We are also
346 : * allowed to update from XLogCtl->RedoRecPtr if we hold the info_lck;
347 : * see GetRedoRecPtr. A freshly spawned backend obtains the value during
348 : * InitXLOGAccess.
349 : */
350 : static XLogRecPtr RedoRecPtr;
351 :
352 : /*
353 : * doPageWrites is this backend's local copy of (forcePageWrites ||
354 : * fullPageWrites). It is used together with RedoRecPtr to decide whether
355 : * a full-page image of a page need to be taken.
356 : */
357 : static bool doPageWrites;
358 :
359 : /* Has the recovery code requested a walreceiver wakeup? */
360 : static bool doRequestWalReceiverReply;
361 :
362 : /*
363 : * RedoStartLSN points to the checkpoint's REDO location which is specified
364 : * in a backup label file, backup history file or control file. In standby
365 : * mode, XLOG streaming usually starts from the position where an invalid
366 : * record was found. But if we fail to read even the initial checkpoint
367 : * record, we use the REDO location instead of the checkpoint location as
368 : * the start position of XLOG streaming. Otherwise we would have to jump
369 : * backwards to the REDO location after reading the checkpoint record,
370 : * because the REDO record can precede the checkpoint record.
371 : */
372 : static XLogRecPtr RedoStartLSN = InvalidXLogRecPtr;
373 :
374 : /*----------
375 : * Shared-memory data structures for XLOG control
376 : *
377 : * LogwrtRqst indicates a byte position that we need to write and/or fsync
378 : * the log up to (all records before that point must be written or fsynced).
379 : * LogwrtResult indicates the byte positions we have already written/fsynced.
380 : * These structs are identical but are declared separately to indicate their
381 : * slightly different functions.
382 : *
383 : * To read XLogCtl->LogwrtResult, you must hold either info_lck or
384 : * WALWriteLock. To update it, you need to hold both locks. The point of
385 : * this arrangement is that the value can be examined by code that already
386 : * holds WALWriteLock without needing to grab info_lck as well. In addition
387 : * to the shared variable, each backend has a private copy of LogwrtResult,
388 : * which is updated when convenient.
389 : *
390 : * The request bookkeeping is simpler: there is a shared XLogCtl->LogwrtRqst
391 : * (protected by info_lck), but we don't need to cache any copies of it.
392 : *
393 : * info_lck is only held long enough to read/update the protected variables,
394 : * so it's a plain spinlock. The other locks are held longer (potentially
395 : * over I/O operations), so we use LWLocks for them. These locks are:
396 : *
397 : * WALBufMappingLock: must be held to replace a page in the WAL buffer cache.
398 : * It is only held while initializing and changing the mapping. If the
399 : * contents of the buffer being replaced haven't been written yet, the mapping
400 : * lock is released while the write is done, and reacquired afterwards.
401 : *
402 : * WALWriteLock: must be held to write WAL buffers to disk (XLogWrite or
403 : * XLogFlush).
404 : *
405 : * ControlFileLock: must be held to read/update control file or create
406 : * new log file.
407 : *
408 : * CheckpointLock: must be held to do a checkpoint or restartpoint (ensures
409 : * only one checkpointer at a time; currently, with all checkpoints done by
410 : * the checkpointer, this is just pro forma).
411 : *
412 : *----------
413 : */
414 :
415 : typedef struct XLogwrtRqst
416 : {
417 : XLogRecPtr Write; /* last byte + 1 to write out */
418 : XLogRecPtr Flush; /* last byte + 1 to flush */
419 : } XLogwrtRqst;
420 :
421 : typedef struct XLogwrtResult
422 : {
423 : XLogRecPtr Write; /* last byte + 1 written out */
424 : XLogRecPtr Flush; /* last byte + 1 flushed */
425 : } XLogwrtResult;
426 :
427 : /*
428 : * Inserting to WAL is protected by a small fixed number of WAL insertion
429 : * locks. To insert to the WAL, you must hold one of the locks - it doesn't
430 : * matter which one. To lock out other concurrent insertions, you must hold
431 : * of them. Each WAL insertion lock consists of a lightweight lock, plus an
432 : * indicator of how far the insertion has progressed (insertingAt).
433 : *
434 : * The insertingAt values are read when a process wants to flush WAL from
435 : * the in-memory buffers to disk, to check that all the insertions to the
436 : * region the process is about to write out have finished. You could simply
437 : * wait for all currently in-progress insertions to finish, but the
438 : * insertingAt indicator allows you to ignore insertions to later in the WAL,
439 : * so that you only wait for the insertions that are modifying the buffers
440 : * you're about to write out.
441 : *
442 : * This isn't just an optimization. If all the WAL buffers are dirty, an
443 : * inserter that's holding a WAL insert lock might need to evict an old WAL
444 : * buffer, which requires flushing the WAL. If it's possible for an inserter
445 : * to block on another inserter unnecessarily, deadlock can arise when two
446 : * inserters holding a WAL insert lock wait for each other to finish their
447 : * insertion.
448 : *
449 : * Small WAL records that don't cross a page boundary never update the value,
450 : * the WAL record is just copied to the page and the lock is released. But
451 : * to avoid the deadlock-scenario explained above, the indicator is always
452 : * updated before sleeping while holding an insertion lock.
453 : *
454 : * lastImportantAt contains the LSN of the last important WAL record inserted
455 : * using a given lock. This value is used to detect if there has been
456 : * important WAL activity since the last time some action, like a checkpoint,
457 : * was performed - allowing to not repeat the action if not. The LSN is
458 : * updated for all insertions, unless the XLOG_MARK_UNIMPORTANT flag was
459 : * set. lastImportantAt is never cleared, only overwritten by the LSN of newer
460 : * records. Tracking the WAL activity directly in WALInsertLock has the
461 : * advantage of not needing any additional locks to update the value.
462 : */
463 : typedef struct
464 : {
465 : LWLock lock;
466 : XLogRecPtr insertingAt;
467 : XLogRecPtr lastImportantAt;
468 : } WALInsertLock;
469 :
470 : /*
471 : * All the WAL insertion locks are allocated as an array in shared memory. We
472 : * force the array stride to be a power of 2, which saves a few cycles in
473 : * indexing, but more importantly also ensures that individual slots don't
474 : * cross cache line boundaries. (Of course, we have to also ensure that the
475 : * array start address is suitably aligned.)
476 : */
477 : typedef union WALInsertLockPadded
478 : {
479 : WALInsertLock l;
480 : char pad[PG_CACHE_LINE_SIZE];
481 : } WALInsertLockPadded;
482 :
483 : /*
484 : * State of an exclusive backup, necessary to control concurrent activities
485 : * across sessions when working on exclusive backups.
486 : *
487 : * EXCLUSIVE_BACKUP_NONE means that there is no exclusive backup actually
488 : * running, to be more precise pg_start_backup() is not being executed for
489 : * an exclusive backup and there is no exclusive backup in progress.
490 : * EXCLUSIVE_BACKUP_STARTING means that pg_start_backup() is starting an
491 : * exclusive backup.
492 : * EXCLUSIVE_BACKUP_IN_PROGRESS means that pg_start_backup() has finished
493 : * running and an exclusive backup is in progress. pg_stop_backup() is
494 : * needed to finish it.
495 : * EXCLUSIVE_BACKUP_STOPPING means that pg_stop_backup() is stopping an
496 : * exclusive backup.
497 : */
498 : typedef enum ExclusiveBackupState
499 : {
500 : EXCLUSIVE_BACKUP_NONE = 0,
501 : EXCLUSIVE_BACKUP_STARTING,
502 : EXCLUSIVE_BACKUP_IN_PROGRESS,
503 : EXCLUSIVE_BACKUP_STOPPING
504 : } ExclusiveBackupState;
505 :
506 : /*
507 : * Session status of running backup, used for sanity checks in SQL-callable
508 : * functions to start and stop backups.
509 : */
510 : static SessionBackupState sessionBackupState = SESSION_BACKUP_NONE;
511 :
512 : /*
513 : * Shared state data for WAL insertion.
514 : */
515 : typedef struct XLogCtlInsert
516 : {
517 : slock_t insertpos_lck; /* protects CurrBytePos and PrevBytePos */
518 :
519 : /*
520 : * CurrBytePos is the end of reserved WAL. The next record will be
521 : * inserted at that position. PrevBytePos is the start position of the
522 : * previously inserted (or rather, reserved) record - it is copied to the
523 : * prev-link of the next record. These are stored as "usable byte
524 : * positions" rather than XLogRecPtrs (see XLogBytePosToRecPtr()).
525 : */
526 : uint64 CurrBytePos;
527 : uint64 PrevBytePos;
528 :
529 : /*
530 : * Make sure the above heavily-contended spinlock and byte positions are
531 : * on their own cache line. In particular, the RedoRecPtr and full page
532 : * write variables below should be on a different cache line. They are
533 : * read on every WAL insertion, but updated rarely, and we don't want
534 : * those reads to steal the cache line containing Curr/PrevBytePos.
535 : */
536 : char pad[PG_CACHE_LINE_SIZE];
537 :
538 : /*
539 : * fullPageWrites is the master copy used by all backends to determine
540 : * whether to write full-page to WAL, instead of using process-local one.
541 : * This is required because, when full_page_writes is changed by SIGHUP,
542 : * we must WAL-log it before it actually affects WAL-logging by backends.
543 : * Checkpointer sets at startup or after SIGHUP.
544 : *
545 : * To read these fields, you must hold an insertion lock. To modify them,
546 : * you must hold ALL the locks.
547 : */
548 : XLogRecPtr RedoRecPtr; /* current redo point for insertions */
549 : bool forcePageWrites; /* forcing full-page writes for PITR? */
550 : bool fullPageWrites;
551 :
552 : /*
553 : * exclusiveBackupState indicates the state of an exclusive backup (see
554 : * comments of ExclusiveBackupState for more details). nonExclusiveBackups
555 : * is a counter indicating the number of streaming base backups currently
556 : * in progress. forcePageWrites is set to true when either of these is
557 : * non-zero. lastBackupStart is the latest checkpoint redo location used
558 : * as a starting point for an online backup.
559 : */
560 : ExclusiveBackupState exclusiveBackupState;
561 : int nonExclusiveBackups;
562 : XLogRecPtr lastBackupStart;
563 :
564 : /*
565 : * WAL insertion locks.
566 : */
567 : WALInsertLockPadded *WALInsertLocks;
568 : } XLogCtlInsert;
569 :
570 : /*
571 : * Total shared-memory state for XLOG.
572 : */
573 : typedef struct XLogCtlData
574 : {
575 : XLogCtlInsert Insert;
576 :
577 : /* Protected by info_lck: */
578 : XLogwrtRqst LogwrtRqst;
579 : XLogRecPtr RedoRecPtr; /* a recent copy of Insert->RedoRecPtr */
580 : uint32 ckptXidEpoch; /* nextXID & epoch of latest checkpoint */
581 : TransactionId ckptXid;
582 : XLogRecPtr asyncXactLSN; /* LSN of newest async commit/abort */
583 : XLogRecPtr replicationSlotMinLSN; /* oldest LSN needed by any slot */
584 :
585 : XLogSegNo lastRemovedSegNo; /* latest removed/recycled XLOG segment */
586 :
587 : /* Fake LSN counter, for unlogged relations. Protected by ulsn_lck. */
588 : XLogRecPtr unloggedLSN;
589 : slock_t ulsn_lck;
590 :
591 : /* Time and LSN of last xlog segment switch. Protected by WALWriteLock. */
592 : pg_time_t lastSegSwitchTime;
593 : XLogRecPtr lastSegSwitchLSN;
594 :
595 : /*
596 : * Protected by info_lck and WALWriteLock (you must hold either lock to
597 : * read it, but both to update)
598 : */
599 : XLogwrtResult LogwrtResult;
600 :
601 : /*
602 : * Latest initialized page in the cache (last byte position + 1).
603 : *
604 : * To change the identity of a buffer (and InitializedUpTo), you need to
605 : * hold WALBufMappingLock. To change the identity of a buffer that's
606 : * still dirty, the old page needs to be written out first, and for that
607 : * you need WALWriteLock, and you need to ensure that there are no
608 : * in-progress insertions to the page by calling
609 : * WaitXLogInsertionsToFinish().
610 : */
611 : XLogRecPtr InitializedUpTo;
612 :
613 : /*
614 : * These values do not change after startup, although the pointed-to pages
615 : * and xlblocks values certainly do. xlblock values are protected by
616 : * WALBufMappingLock.
617 : */
618 : char *pages; /* buffers for unwritten XLOG pages */
619 : XLogRecPtr *xlblocks; /* 1st byte ptr-s + XLOG_BLCKSZ */
620 : int XLogCacheBlck; /* highest allocated xlog buffer index */
621 :
622 : /*
623 : * Shared copy of ThisTimeLineID. Does not change after end-of-recovery.
624 : * If we created a new timeline when the system was started up,
625 : * PrevTimeLineID is the old timeline's ID that we forked off from.
626 : * Otherwise it's equal to ThisTimeLineID.
627 : */
628 : TimeLineID ThisTimeLineID;
629 : TimeLineID PrevTimeLineID;
630 :
631 : /*
632 : * archiveCleanupCommand is read from recovery.conf but needs to be in
633 : * shared memory so that the checkpointer process can access it.
634 : */
635 : char archiveCleanupCommand[MAXPGPATH];
636 :
637 : /*
638 : * SharedRecoveryInProgress indicates if we're still in crash or archive
639 : * recovery. Protected by info_lck.
640 : */
641 : bool SharedRecoveryInProgress;
642 :
643 : /*
644 : * SharedHotStandbyActive indicates if we're still in crash or archive
645 : * recovery. Protected by info_lck.
646 : */
647 : bool SharedHotStandbyActive;
648 :
649 : /*
650 : * WalWriterSleeping indicates whether the WAL writer is currently in
651 : * low-power mode (and hence should be nudged if an async commit occurs).
652 : * Protected by info_lck.
653 : */
654 : bool WalWriterSleeping;
655 :
656 : /*
657 : * recoveryWakeupLatch is used to wake up the startup process to continue
658 : * WAL replay, if it is waiting for WAL to arrive or failover trigger file
659 : * to appear.
660 : */
661 : Latch recoveryWakeupLatch;
662 :
663 : /*
664 : * During recovery, we keep a copy of the latest checkpoint record here.
665 : * lastCheckPointRecPtr points to start of checkpoint record and
666 : * lastCheckPointEndPtr points to end+1 of checkpoint record. Used by the
667 : * checkpointer when it wants to create a restartpoint.
668 : *
669 : * Protected by info_lck.
670 : */
671 : XLogRecPtr lastCheckPointRecPtr;
672 : XLogRecPtr lastCheckPointEndPtr;
673 : CheckPoint lastCheckPoint;
674 :
675 : /*
676 : * lastReplayedEndRecPtr points to end+1 of the last record successfully
677 : * replayed. When we're currently replaying a record, ie. in a redo
678 : * function, replayEndRecPtr points to the end+1 of the record being
679 : * replayed, otherwise it's equal to lastReplayedEndRecPtr.
680 : */
681 : XLogRecPtr lastReplayedEndRecPtr;
682 : TimeLineID lastReplayedTLI;
683 : XLogRecPtr replayEndRecPtr;
684 : TimeLineID replayEndTLI;
685 : /* timestamp of last COMMIT/ABORT record replayed (or being replayed) */
686 : TimestampTz recoveryLastXTime;
687 :
688 : /*
689 : * timestamp of when we started replaying the current chunk of WAL data,
690 : * only relevant for replication or archive recovery
691 : */
692 : TimestampTz currentChunkStartTime;
693 : /* Are we requested to pause recovery? */
694 : bool recoveryPause;
695 :
696 : /*
697 : * lastFpwDisableRecPtr points to the start of the last replayed
698 : * XLOG_FPW_CHANGE record that instructs full_page_writes is disabled.
699 : */
700 : XLogRecPtr lastFpwDisableRecPtr;
701 :
702 : slock_t info_lck; /* locks shared variables shown above */
703 : } XLogCtlData;
704 :
705 : static XLogCtlData *XLogCtl = NULL;
706 :
707 : /* a private copy of XLogCtl->Insert.WALInsertLocks, for convenience */
708 : static WALInsertLockPadded *WALInsertLocks = NULL;
709 :
710 : /*
711 : * We maintain an image of pg_control in shared memory.
712 : */
713 : static ControlFileData *ControlFile = NULL;
714 :
715 : /*
716 : * Calculate the amount of space left on the page after 'endptr'. Beware
717 : * multiple evaluation!
718 : */
719 : #define INSERT_FREESPACE(endptr) \
720 : (((endptr) % XLOG_BLCKSZ == 0) ? 0 : (XLOG_BLCKSZ - (endptr) % XLOG_BLCKSZ))
721 :
722 : /* Macro to advance to next buffer index. */
723 : #define NextBufIdx(idx) \
724 : (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))
725 :
726 : /*
727 : * XLogRecPtrToBufIdx returns the index of the WAL buffer that holds, or
728 : * would hold if it was in cache, the page containing 'recptr'.
729 : */
730 : #define XLogRecPtrToBufIdx(recptr) \
731 : (((recptr) / XLOG_BLCKSZ) % (XLogCtl->XLogCacheBlck + 1))
732 :
733 : /*
734 : * These are the number of bytes in a WAL page and segment usable for WAL data.
735 : */
736 : #define UsableBytesInPage (XLOG_BLCKSZ - SizeOfXLogShortPHD)
737 : #define UsableBytesInSegment ((XLOG_SEG_SIZE / XLOG_BLCKSZ) * UsableBytesInPage - (SizeOfXLogLongPHD - SizeOfXLogShortPHD))
738 :
739 : /* Convert min_wal_size_mb and max wal_size_mb to equivalent segment count */
740 : #define ConvertToXSegs(x) \
741 : (x / (XLOG_SEG_SIZE / (1024 * 1024)))
742 :
743 : /*
744 : * Private, possibly out-of-date copy of shared LogwrtResult.
745 : * See discussion above.
746 : */
747 : static XLogwrtResult LogwrtResult = {0, 0};
748 :
749 : /*
750 : * Codes indicating where we got a WAL file from during recovery, or where
751 : * to attempt to get one.
752 : */
753 : typedef enum
754 : {
755 : XLOG_FROM_ANY = 0, /* request to read WAL from any source */
756 : XLOG_FROM_ARCHIVE, /* restored using restore_command */
757 : XLOG_FROM_PG_WAL, /* existing file in pg_wal */
758 : XLOG_FROM_STREAM /* streamed from master */
759 : } XLogSource;
760 :
761 : /* human-readable names for XLogSources, for debugging output */
762 : static const char *xlogSourceNames[] = {"any", "archive", "pg_wal", "stream"};
763 :
764 : /*
765 : * openLogFile is -1 or a kernel FD for an open log file segment.
766 : * When it's open, openLogOff is the current seek offset in the file.
767 : * openLogSegNo identifies the segment. These variables are only
768 : * used to write the XLOG, and so will normally refer to the active segment.
769 : */
770 : static int openLogFile = -1;
771 : static XLogSegNo openLogSegNo = 0;
772 : static uint32 openLogOff = 0;
773 :
774 : /*
775 : * These variables are used similarly to the ones above, but for reading
776 : * the XLOG. Note, however, that readOff generally represents the offset
777 : * of the page just read, not the seek position of the FD itself, which
778 : * will be just past that page. readLen indicates how much of the current
779 : * page has been read into readBuf, and readSource indicates where we got
780 : * the currently open file from.
781 : */
782 : static int readFile = -1;
783 : static XLogSegNo readSegNo = 0;
784 : static uint32 readOff = 0;
785 : static uint32 readLen = 0;
786 : static XLogSource readSource = 0; /* XLOG_FROM_* code */
787 :
788 : /*
789 : * Keeps track of which source we're currently reading from. This is
790 : * different from readSource in that this is always set, even when we don't
791 : * currently have a WAL file open. If lastSourceFailed is set, our last
792 : * attempt to read from currentSource failed, and we should try another source
793 : * next.
794 : */
795 : static XLogSource currentSource = 0; /* XLOG_FROM_* code */
796 : static bool lastSourceFailed = false;
797 :
798 : typedef struct XLogPageReadPrivate
799 : {
800 : int emode;
801 : bool fetching_ckpt; /* are we fetching a checkpoint record? */
802 : bool randAccess;
803 : } XLogPageReadPrivate;
804 :
805 : /*
806 : * These variables track when we last obtained some WAL data to process,
807 : * and where we got it from. (XLogReceiptSource is initially the same as
808 : * readSource, but readSource gets reset to zero when we don't have data
809 : * to process right now. It is also different from currentSource, which
810 : * also changes when we try to read from a source and fail, while
811 : * XLogReceiptSource tracks where we last successfully read some WAL.)
812 : */
813 : static TimestampTz XLogReceiptTime = 0;
814 : static XLogSource XLogReceiptSource = 0; /* XLOG_FROM_* code */
815 :
816 : /* State information for XLOG reading */
817 : static XLogRecPtr ReadRecPtr; /* start of last record read */
818 : static XLogRecPtr EndRecPtr; /* end+1 of last record read */
819 :
820 : static XLogRecPtr minRecoveryPoint; /* local copy of
821 : * ControlFile->minRecoveryPoint */
822 : static TimeLineID minRecoveryPointTLI;
823 : static bool updateMinRecoveryPoint = true;
824 :
825 : /*
826 : * Have we reached a consistent database state? In crash recovery, we have
827 : * to replay all the WAL, so reachedConsistency is never set. During archive
828 : * recovery, the database is consistent once minRecoveryPoint is reached.
829 : */
830 : bool reachedConsistency = false;
831 :
832 : static bool InRedo = false;
833 :
834 : /* Have we launched bgwriter during recovery? */
835 : static bool bgwriterLaunched = false;
836 :
837 : /* For WALInsertLockAcquire/Release functions */
838 : static int MyLockNo = 0;
839 : static bool holdingAllLocks = false;
840 :
841 : #ifdef WAL_DEBUG
842 : static MemoryContext walDebugCxt = NULL;
843 : #endif
844 :
845 : static void readRecoveryCommandFile(void);
846 : static void exitArchiveRecovery(TimeLineID endTLI, XLogRecPtr endOfLog);
847 : static bool recoveryStopsBefore(XLogReaderState *record);
848 : static bool recoveryStopsAfter(XLogReaderState *record);
849 : static void recoveryPausesHere(void);
850 : static bool recoveryApplyDelay(XLogReaderState *record);
851 : static void SetLatestXTime(TimestampTz xtime);
852 : static void SetCurrentChunkStartTime(TimestampTz xtime);
853 : static void CheckRequiredParameterValues(void);
854 : static void XLogReportParameters(void);
855 : static void checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI,
856 : TimeLineID prevTLI);
857 : static void LocalSetXLogInsertAllowed(void);
858 : static void CreateEndOfRecoveryRecord(void);
859 : static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
860 : static void KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo);
861 : static XLogRecPtr XLogGetReplicationSlotMinimumLSN(void);
862 :
863 : static void AdvanceXLInsertBuffer(XLogRecPtr upto, bool opportunistic);
864 : static bool XLogCheckpointNeeded(XLogSegNo new_segno);
865 : static void XLogWrite(XLogwrtRqst WriteRqst, bool flexible);
866 : static bool InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
867 : bool find_free, XLogSegNo max_segno,
868 : bool use_lock);
869 : static int XLogFileRead(XLogSegNo segno, int emode, TimeLineID tli,
870 : int source, bool notfoundOk);
871 : static int XLogFileReadAnyTLI(XLogSegNo segno, int emode, int source);
872 : static int XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr,
873 : int reqLen, XLogRecPtr targetRecPtr, char *readBuf,
874 : TimeLineID *readTLI);
875 : static bool WaitForWALToBecomeAvailable(XLogRecPtr RecPtr, bool randAccess,
876 : bool fetching_ckpt, XLogRecPtr tliRecPtr);
877 : static int emode_for_corrupt_record(int emode, XLogRecPtr RecPtr);
878 : static void XLogFileClose(void);
879 : static void PreallocXlogFiles(XLogRecPtr endptr);
880 : static void RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr PriorRedoPtr, XLogRecPtr endptr);
881 : static void RemoveXlogFile(const char *segname, XLogRecPtr PriorRedoPtr, XLogRecPtr endptr);
882 : static void UpdateLastRemovedPtr(char *filename);
883 : static void ValidateXLOGDirectoryStructure(void);
884 : static void CleanupBackupHistory(void);
885 : static void UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force);
886 : static XLogRecord *ReadRecord(XLogReaderState *xlogreader, XLogRecPtr RecPtr,
887 : int emode, bool fetching_ckpt);
888 : static void CheckRecoveryConsistency(void);
889 : static XLogRecord *ReadCheckpointRecord(XLogReaderState *xlogreader,
890 : XLogRecPtr RecPtr, int whichChkpti, bool report);
891 : static bool rescanLatestTimeLine(void);
892 : static void WriteControlFile(void);
893 : static void ReadControlFile(void);
894 : static char *str_time(pg_time_t tnow);
895 : static bool CheckForStandbyTrigger(void);
896 :
897 : #ifdef WAL_DEBUG
898 : static void xlog_outrec(StringInfo buf, XLogReaderState *record);
899 : #endif
900 : static void xlog_outdesc(StringInfo buf, XLogReaderState *record);
901 : static void pg_start_backup_callback(int code, Datum arg);
902 : static void pg_stop_backup_callback(int code, Datum arg);
903 : static bool read_backup_label(XLogRecPtr *checkPointLoc,
904 : bool *backupEndRequired, bool *backupFromStandby);
905 : static bool read_tablespace_map(List **tablespaces);
906 :
907 : static void rm_redo_error_callback(void *arg);
908 : static int get_sync_bit(int method);
909 :
910 : static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch,
911 : XLogRecData *rdata,
912 : XLogRecPtr StartPos, XLogRecPtr EndPos);
913 : static void ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos,
914 : XLogRecPtr *EndPos, XLogRecPtr *PrevPtr);
915 : static bool ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos,
916 : XLogRecPtr *PrevPtr);
917 : static XLogRecPtr WaitXLogInsertionsToFinish(XLogRecPtr upto);
918 : static char *GetXLogBuffer(XLogRecPtr ptr);
919 : static XLogRecPtr XLogBytePosToRecPtr(uint64 bytepos);
920 : static XLogRecPtr XLogBytePosToEndRecPtr(uint64 bytepos);
921 : static uint64 XLogRecPtrToBytePos(XLogRecPtr ptr);
922 : static void checkXLogConsistency(XLogReaderState *record);
923 :
924 : static void WALInsertLockAcquire(void);
925 : static void WALInsertLockAcquireExclusive(void);
926 : static void WALInsertLockRelease(void);
927 : static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
928 :
929 : /*
930 : * Insert an XLOG record represented by an already-constructed chain of data
931 : * chunks. This is a low-level routine; to construct the WAL record header
932 : * and data, use the higher-level routines in xloginsert.c.
933 : *
934 : * If 'fpw_lsn' is valid, it is the oldest LSN among the pages that this
935 : * WAL record applies to, that were not included in the record as full page
936 : * images. If fpw_lsn >= RedoRecPtr, the function does not perform the
937 : * insertion and returns InvalidXLogRecPtr. The caller can then recalculate
938 : * which pages need a full-page image, and retry. If fpw_lsn is invalid, the
939 : * record is always inserted.
940 : *
941 : * 'flags' gives more in-depth control on the record being inserted. See
942 : * XLogSetRecordFlags() for details.
943 : *
944 : * The first XLogRecData in the chain must be for the record header, and its
945 : * data must be MAXALIGNed. XLogInsertRecord fills in the xl_prev and
946 : * xl_crc fields in the header, the rest of the header must already be filled
947 : * by the caller.
948 : *
949 : * Returns XLOG pointer to end of record (beginning of next record).
950 : * This can be used as LSN for data pages affected by the logged action.
951 : * (LSN is the XLOG point up to which the XLOG must be flushed to disk
952 : * before the data page can be written out. This implements the basic
953 : * WAL rule "write the log before the data".)
954 : */
955 : XLogRecPtr
956 1306714 : XLogInsertRecord(XLogRecData *rdata,
957 : XLogRecPtr fpw_lsn,
958 : uint8 flags)
959 : {
960 1306714 : XLogCtlInsert *Insert = &XLogCtl->Insert;
961 : pg_crc32c rdata_crc;
962 : bool inserted;
963 1306714 : XLogRecord *rechdr = (XLogRecord *) rdata->data;
964 1306714 : uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
965 1306714 : bool isLogSwitch = (rechdr->xl_rmid == RM_XLOG_ID &&
966 : info == XLOG_SWITCH);
967 : XLogRecPtr StartPos;
968 : XLogRecPtr EndPos;
969 :
970 : /* we assume that all of the record header is in the first chunk */
971 1306714 : Assert(rdata->len >= SizeOfXLogRecord);
972 :
973 : /* cross-check on whether we should be here or not */
974 1306714 : if (!XLogInsertAllowed())
975 0 : elog(ERROR, "cannot make new WAL entries during recovery");
976 :
977 : /*----------
978 : *
979 : * We have now done all the preparatory work we can without holding a
980 : * lock or modifying shared state. From here on, inserting the new WAL
981 : * record to the shared WAL buffer cache is a two-step process:
982 : *
983 : * 1. Reserve the right amount of space from the WAL. The current head of
984 : * reserved space is kept in Insert->CurrBytePos, and is protected by
985 : * insertpos_lck.
986 : *
987 : * 2. Copy the record to the reserved WAL space. This involves finding the
988 : * correct WAL buffer containing the reserved space, and copying the
989 : * record in place. This can be done concurrently in multiple processes.
990 : *
991 : * To keep track of which insertions are still in-progress, each concurrent
992 : * inserter acquires an insertion lock. In addition to just indicating that
993 : * an insertion is in progress, the lock tells others how far the inserter
994 : * has progressed. There is a small fixed number of insertion locks,
995 : * determined by NUM_XLOGINSERT_LOCKS. When an inserter crosses a page
996 : * boundary, it updates the value stored in the lock to the how far it has
997 : * inserted, to allow the previous buffer to be flushed.
998 : *
999 : * Holding onto an insertion lock also protects RedoRecPtr and
1000 : * fullPageWrites from changing until the insertion is finished.
1001 : *
1002 : * Step 2 can usually be done completely in parallel. If the required WAL
1003 : * page is not initialized yet, you have to grab WALBufMappingLock to
1004 : * initialize it, but the WAL writer tries to do that ahead of insertions
1005 : * to avoid that from happening in the critical path.
1006 : *
1007 : *----------
1008 : */
1009 1306714 : START_CRIT_SECTION();
1010 1306714 : if (isLogSwitch)
1011 0 : WALInsertLockAcquireExclusive();
1012 : else
1013 1306714 : WALInsertLockAcquire();
1014 :
1015 : /*
1016 : * Check to see if my copy of RedoRecPtr or doPageWrites is out of date.
1017 : * If so, may have to go back and have the caller recompute everything.
1018 : * This can only happen just after a checkpoint, so it's better to be slow
1019 : * in this case and fast otherwise.
1020 : *
1021 : * If we aren't doing full-page writes then RedoRecPtr doesn't actually
1022 : * affect the contents of the XLOG record, so we'll update our local copy
1023 : * but not force a recomputation. (If doPageWrites was just turned off,
1024 : * we could recompute the record without full pages, but we choose not to
1025 : * bother.)
1026 : */
1027 1306714 : if (RedoRecPtr != Insert->RedoRecPtr)
1028 : {
1029 5 : Assert(RedoRecPtr < Insert->RedoRecPtr);
1030 5 : RedoRecPtr = Insert->RedoRecPtr;
1031 : }
1032 1306714 : doPageWrites = (Insert->fullPageWrites || Insert->forcePageWrites);
1033 :
1034 1306714 : if (fpw_lsn != InvalidXLogRecPtr && fpw_lsn <= RedoRecPtr && doPageWrites)
1035 : {
1036 : /*
1037 : * Oops, some buffer now needs to be backed up that the caller didn't
1038 : * back up. Start over.
1039 : */
1040 0 : WALInsertLockRelease();
1041 0 : END_CRIT_SECTION();
1042 0 : return InvalidXLogRecPtr;
1043 : }
1044 :
1045 : /*
1046 : * Reserve space for the record in the WAL. This also sets the xl_prev
1047 : * pointer.
1048 : */
1049 1306714 : if (isLogSwitch)
1050 0 : inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
1051 : else
1052 : {
1053 1306714 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
1054 : &rechdr->xl_prev);
1055 1306714 : inserted = true;
1056 : }
1057 :
1058 1306714 : if (inserted)
1059 : {
1060 : /*
1061 : * Now that xl_prev has been filled in, calculate CRC of the record
1062 : * header.
1063 : */
1064 1306714 : rdata_crc = rechdr->xl_crc;
1065 1306714 : COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
1066 1306714 : FIN_CRC32C(rdata_crc);
1067 1306714 : rechdr->xl_crc = rdata_crc;
1068 :
1069 : /*
1070 : * All the record data, including the header, is now ready to be
1071 : * inserted. Copy the record in the space reserved.
1072 : */
1073 1306714 : CopyXLogRecordToWAL(rechdr->xl_tot_len, isLogSwitch, rdata,
1074 : StartPos, EndPos);
1075 :
1076 : /*
1077 : * Unless record is flagged as not important, update LSN of last
1078 : * important record in the current slot. When holding all locks, just
1079 : * update the first one.
1080 : */
1081 1306714 : if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
1082 : {
1083 1298031 : int lockno = holdingAllLocks ? 0 : MyLockNo;
1084 :
1085 1298031 : WALInsertLocks[lockno].l.lastImportantAt = StartPos;
1086 : }
1087 : }
1088 : else
1089 : {
1090 : /*
1091 : * This was an xlog-switch record, but the current insert location was
1092 : * already exactly at the beginning of a segment, so there was no need
1093 : * to do anything.
1094 : */
1095 : }
1096 :
1097 : /*
1098 : * Done! Let others know that we're finished.
1099 : */
1100 1306714 : WALInsertLockRelease();
1101 :
1102 1306714 : MarkCurrentTransactionIdLoggedIfAny();
1103 :
1104 1306714 : END_CRIT_SECTION();
1105 :
1106 : /*
1107 : * Update shared LogwrtRqst.Write, if we crossed page boundary.
1108 : */
1109 1306714 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
1110 : {
1111 20958 : SpinLockAcquire(&XLogCtl->info_lck);
1112 : /* advance global request to include new block(s) */
1113 20958 : if (XLogCtl->LogwrtRqst.Write < EndPos)
1114 20950 : XLogCtl->LogwrtRqst.Write = EndPos;
1115 : /* update local result copy while I have the chance */
1116 20958 : LogwrtResult = XLogCtl->LogwrtResult;
1117 20958 : SpinLockRelease(&XLogCtl->info_lck);
1118 : }
1119 :
1120 : /*
1121 : * If this was an XLOG_SWITCH record, flush the record and the empty
1122 : * padding space that fills the rest of the segment, and perform
1123 : * end-of-segment actions (eg, notifying archiver).
1124 : */
1125 1306714 : if (isLogSwitch)
1126 : {
1127 : TRACE_POSTGRESQL_WAL_SWITCH();
1128 0 : XLogFlush(EndPos);
1129 :
1130 : /*
1131 : * Even though we reserved the rest of the segment for us, which is
1132 : * reflected in EndPos, we return a pointer to just the end of the
1133 : * xlog-switch record.
1134 : */
1135 0 : if (inserted)
1136 : {
1137 0 : EndPos = StartPos + SizeOfXLogRecord;
1138 0 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
1139 : {
1140 0 : if (EndPos % XLOG_SEG_SIZE == EndPos % XLOG_BLCKSZ)
1141 0 : EndPos += SizeOfXLogLongPHD;
1142 : else
1143 0 : EndPos += SizeOfXLogShortPHD;
1144 : }
1145 : }
1146 : }
1147 :
1148 : #ifdef WAL_DEBUG
1149 : if (XLOG_DEBUG)
1150 : {
1151 : static XLogReaderState *debug_reader = NULL;
1152 : StringInfoData buf;
1153 : StringInfoData recordBuf;
1154 : char *errormsg = NULL;
1155 : MemoryContext oldCxt;
1156 :
1157 : oldCxt = MemoryContextSwitchTo(walDebugCxt);
1158 :
1159 : initStringInfo(&buf);
1160 : appendStringInfo(&buf, "INSERT @ %X/%X: ",
1161 : (uint32) (EndPos >> 32), (uint32) EndPos);
1162 :
1163 : /*
1164 : * We have to piece together the WAL record data from the XLogRecData
1165 : * entries, so that we can pass it to the rm_desc function as one
1166 : * contiguous chunk.
1167 : */
1168 : initStringInfo(&recordBuf);
1169 : for (; rdata != NULL; rdata = rdata->next)
1170 : appendBinaryStringInfo(&recordBuf, rdata->data, rdata->len);
1171 :
1172 : if (!debug_reader)
1173 : debug_reader = XLogReaderAllocate(NULL, NULL);
1174 :
1175 : if (!debug_reader)
1176 : {
1177 : appendStringInfoString(&buf, "error decoding record: out of memory");
1178 : }
1179 : else if (!DecodeXLogRecord(debug_reader, (XLogRecord *) recordBuf.data,
1180 : &errormsg))
1181 : {
1182 : appendStringInfo(&buf, "error decoding record: %s",
1183 : errormsg ? errormsg : "no error message");
1184 : }
1185 : else
1186 : {
1187 : appendStringInfoString(&buf, " - ");
1188 : xlog_outdesc(&buf, debug_reader);
1189 : }
1190 : elog(LOG, "%s", buf.data);
1191 :
1192 : pfree(buf.data);
1193 : pfree(recordBuf.data);
1194 : MemoryContextSwitchTo(oldCxt);
1195 : }
1196 : #endif
1197 :
1198 : /*
1199 : * Update our global variables
1200 : */
1201 1306714 : ProcLastRecPtr = StartPos;
1202 1306714 : XactLastRecEnd = EndPos;
1203 :
1204 1306714 : return EndPos;
1205 : }
1206 :
1207 : /*
1208 : * Reserves the right amount of space for a record of given size from the WAL.
1209 : * *StartPos is set to the beginning of the reserved section, *EndPos to
1210 : * its end+1. *PrevPtr is set to the beginning of the previous record; it is
1211 : * used to set the xl_prev of this record.
1212 : *
1213 : * This is the performance critical part of XLogInsert that must be serialized
1214 : * across backends. The rest can happen mostly in parallel. Try to keep this
1215 : * section as short as possible, insertpos_lck can be heavily contended on a
1216 : * busy system.
1217 : *
1218 : * NB: The space calculation here must match the code in CopyXLogRecordToWAL,
1219 : * where we actually copy the record to the reserved space.
1220 : */
1221 : static void
1222 1306714 : ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos,
1223 : XLogRecPtr *PrevPtr)
1224 : {
1225 1306714 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1226 : uint64 startbytepos;
1227 : uint64 endbytepos;
1228 : uint64 prevbytepos;
1229 :
1230 1306714 : size = MAXALIGN(size);
1231 :
1232 : /* All (non xlog-switch) records should contain data. */
1233 1306714 : Assert(size > SizeOfXLogRecord);
1234 :
1235 : /*
1236 : * The duration the spinlock needs to be held is minimized by minimizing
1237 : * the calculations that have to be done while holding the lock. The
1238 : * current tip of reserved WAL is kept in CurrBytePos, as a byte position
1239 : * that only counts "usable" bytes in WAL, that is, it excludes all WAL
1240 : * page headers. The mapping between "usable" byte positions and physical
1241 : * positions (XLogRecPtrs) can be done outside the locked region, and
1242 : * because the usable byte position doesn't include any headers, reserving
1243 : * X bytes from WAL is almost as simple as "CurrBytePos += X".
1244 : */
1245 1306714 : SpinLockAcquire(&Insert->insertpos_lck);
1246 :
1247 1306714 : startbytepos = Insert->CurrBytePos;
1248 1306714 : endbytepos = startbytepos + size;
1249 1306714 : prevbytepos = Insert->PrevBytePos;
1250 1306714 : Insert->CurrBytePos = endbytepos;
1251 1306714 : Insert->PrevBytePos = startbytepos;
1252 :
1253 1306714 : SpinLockRelease(&Insert->insertpos_lck);
1254 :
1255 1306714 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1256 1306714 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1257 1306714 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1258 :
1259 : /*
1260 : * Check that the conversions between "usable byte positions" and
1261 : * XLogRecPtrs work consistently in both directions.
1262 : */
1263 1306714 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1264 1306714 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1265 1306714 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1266 1306714 : }
1267 :
1268 : /*
1269 : * Like ReserveXLogInsertLocation(), but for an xlog-switch record.
1270 : *
1271 : * A log-switch record is handled slightly differently. The rest of the
1272 : * segment will be reserved for this insertion, as indicated by the returned
1273 : * *EndPos value. However, if we are already at the beginning of the current
1274 : * segment, *StartPos and *EndPos are set to the current location without
1275 : * reserving any space, and the function returns false.
1276 : */
1277 : static bool
1278 0 : ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
1279 : {
1280 0 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1281 : uint64 startbytepos;
1282 : uint64 endbytepos;
1283 : uint64 prevbytepos;
1284 0 : uint32 size = MAXALIGN(SizeOfXLogRecord);
1285 : XLogRecPtr ptr;
1286 : uint32 segleft;
1287 :
1288 : /*
1289 : * These calculations are a bit heavy-weight to be done while holding a
1290 : * spinlock, but since we're holding all the WAL insertion locks, there
1291 : * are no other inserters competing for it. GetXLogInsertRecPtr() does
1292 : * compete for it, but that's not called very frequently.
1293 : */
1294 0 : SpinLockAcquire(&Insert->insertpos_lck);
1295 :
1296 0 : startbytepos = Insert->CurrBytePos;
1297 :
1298 0 : ptr = XLogBytePosToEndRecPtr(startbytepos);
1299 0 : if (ptr % XLOG_SEG_SIZE == 0)
1300 : {
1301 0 : SpinLockRelease(&Insert->insertpos_lck);
1302 0 : *EndPos = *StartPos = ptr;
1303 0 : return false;
1304 : }
1305 :
1306 0 : endbytepos = startbytepos + size;
1307 0 : prevbytepos = Insert->PrevBytePos;
1308 :
1309 0 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1310 0 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1311 :
1312 0 : segleft = XLOG_SEG_SIZE - ((*EndPos) % XLOG_SEG_SIZE);
1313 0 : if (segleft != XLOG_SEG_SIZE)
1314 : {
1315 : /* consume the rest of the segment */
1316 0 : *EndPos += segleft;
1317 0 : endbytepos = XLogRecPtrToBytePos(*EndPos);
1318 : }
1319 0 : Insert->CurrBytePos = endbytepos;
1320 0 : Insert->PrevBytePos = startbytepos;
1321 :
1322 0 : SpinLockRelease(&Insert->insertpos_lck);
1323 :
1324 0 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1325 :
1326 0 : Assert((*EndPos) % XLOG_SEG_SIZE == 0);
1327 0 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1328 0 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1329 0 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1330 :
1331 0 : return true;
1332 : }
1333 :
1334 : /*
1335 : * Checks whether the current buffer page and backup page stored in the
1336 : * WAL record are consistent or not. Before comparing the two pages, a
1337 : * masking can be applied to the pages to ignore certain areas like hint bits,
1338 : * unused space between pd_lower and pd_upper among other things. This
1339 : * function should be called once WAL replay has been completed for a
1340 : * given record.
1341 : */
1342 : static void
1343 0 : checkXLogConsistency(XLogReaderState *record)
1344 : {
1345 0 : RmgrId rmid = XLogRecGetRmid(record);
1346 : RelFileNode rnode;
1347 : ForkNumber forknum;
1348 : BlockNumber blkno;
1349 : int block_id;
1350 :
1351 : /* Records with no backup blocks have no need for consistency checks. */
1352 0 : if (!XLogRecHasAnyBlockRefs(record))
1353 0 : return;
1354 :
1355 0 : Assert((XLogRecGetInfo(record) & XLR_CHECK_CONSISTENCY) != 0);
1356 :
1357 0 : for (block_id = 0; block_id <= record->max_block_id; block_id++)
1358 : {
1359 : Buffer buf;
1360 : Page page;
1361 :
1362 0 : if (!XLogRecGetBlockTag(record, block_id, &rnode, &forknum, &blkno))
1363 : {
1364 : /*
1365 : * WAL record doesn't contain a block reference with the given id.
1366 : * Do nothing.
1367 : */
1368 0 : continue;
1369 : }
1370 :
1371 0 : Assert(XLogRecHasBlockImage(record, block_id));
1372 :
1373 0 : if (XLogRecBlockImageApply(record, block_id))
1374 : {
1375 : /*
1376 : * WAL record has already applied the page, so bypass the
1377 : * consistency check as that would result in comparing the full
1378 : * page stored in the record with itself.
1379 : */
1380 0 : continue;
1381 : }
1382 :
1383 : /*
1384 : * Read the contents from the current buffer and store it in a
1385 : * temporary page.
1386 : */
1387 0 : buf = XLogReadBufferExtended(rnode, forknum, blkno,
1388 : RBM_NORMAL_NO_LOG);
1389 0 : if (!BufferIsValid(buf))
1390 0 : continue;
1391 :
1392 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
1393 0 : page = BufferGetPage(buf);
1394 :
1395 : /*
1396 : * Take a copy of the local page where WAL has been applied to have a
1397 : * comparison base before masking it...
1398 : */
1399 0 : memcpy(replay_image_masked, page, BLCKSZ);
1400 :
1401 : /* No need for this page anymore now that a copy is in. */
1402 0 : UnlockReleaseBuffer(buf);
1403 :
1404 : /*
1405 : * If the block LSN is already ahead of this WAL record, we can't
1406 : * expect contents to match. This can happen if recovery is
1407 : * restarted.
1408 : */
1409 0 : if (PageGetLSN(replay_image_masked) > record->EndRecPtr)
1410 0 : continue;
1411 :
1412 : /*
1413 : * Read the contents from the backup copy, stored in WAL record and
1414 : * store it in a temporary page. There is no need to allocate a new
1415 : * page here, a local buffer is fine to hold its contents and a mask
1416 : * can be directly applied on it.
1417 : */
1418 0 : if (!RestoreBlockImage(record, block_id, master_image_masked))
1419 0 : elog(ERROR, "failed to restore block image");
1420 :
1421 : /*
1422 : * If masking function is defined, mask both the master and replay
1423 : * images
1424 : */
1425 0 : if (RmgrTable[rmid].rm_mask != NULL)
1426 : {
1427 0 : RmgrTable[rmid].rm_mask(replay_image_masked, blkno);
1428 0 : RmgrTable[rmid].rm_mask(master_image_masked, blkno);
1429 : }
1430 :
1431 : /* Time to compare the master and replay images. */
1432 0 : if (memcmp(replay_image_masked, master_image_masked, BLCKSZ) != 0)
1433 : {
1434 0 : elog(FATAL,
1435 : "inconsistent page found, rel %u/%u/%u, forknum %u, blkno %u",
1436 : rnode.spcNode, rnode.dbNode, rnode.relNode,
1437 : forknum, blkno);
1438 : }
1439 : }
1440 : }
1441 :
1442 : /*
1443 : * Subroutine of XLogInsertRecord. Copies a WAL record to an already-reserved
1444 : * area in the WAL.
1445 : */
1446 : static void
1447 1306714 : CopyXLogRecordToWAL(int write_len, bool isLogSwitch, XLogRecData *rdata,
1448 : XLogRecPtr StartPos, XLogRecPtr EndPos)
1449 : {
1450 : char *currpos;
1451 : int freespace;
1452 : int written;
1453 : XLogRecPtr CurrPos;
1454 : XLogPageHeader pagehdr;
1455 :
1456 : /*
1457 : * Get a pointer to the right place in the right WAL buffer to start
1458 : * inserting to.
1459 : */
1460 1306714 : CurrPos = StartPos;
1461 1306714 : currpos = GetXLogBuffer(CurrPos);
1462 1306714 : freespace = INSERT_FREESPACE(CurrPos);
1463 :
1464 : /*
1465 : * there should be enough space for at least the first field (xl_tot_len)
1466 : * on this page.
1467 : */
1468 1306714 : Assert(freespace >= sizeof(uint32));
1469 :
1470 : /* Copy record data */
1471 1306714 : written = 0;
1472 7210509 : while (rdata != NULL)
1473 : {
1474 4597081 : char *rdata_data = rdata->data;
1475 4597081 : int rdata_len = rdata->len;
1476 :
1477 9214513 : while (rdata_len > freespace)
1478 : {
1479 : /*
1480 : * Write what fits on this page, and continue on the next page.
1481 : */
1482 20351 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || freespace == 0);
1483 20351 : memcpy(currpos, rdata_data, freespace);
1484 20351 : rdata_data += freespace;
1485 20351 : rdata_len -= freespace;
1486 20351 : written += freespace;
1487 20351 : CurrPos += freespace;
1488 :
1489 : /*
1490 : * Get pointer to beginning of next page, and set the xlp_rem_len
1491 : * in the page header. Set XLP_FIRST_IS_CONTRECORD.
1492 : *
1493 : * It's safe to set the contrecord flag and xlp_rem_len without a
1494 : * lock on the page. All the other flags were already set when the
1495 : * page was initialized, in AdvanceXLInsertBuffer, and we're the
1496 : * only backend that needs to set the contrecord flag.
1497 : */
1498 20351 : currpos = GetXLogBuffer(CurrPos);
1499 20351 : pagehdr = (XLogPageHeader) currpos;
1500 20351 : pagehdr->xlp_rem_len = write_len - written;
1501 20351 : pagehdr->xlp_info |= XLP_FIRST_IS_CONTRECORD;
1502 :
1503 : /* skip over the page header */
1504 20351 : if (CurrPos % XLogSegSize == 0)
1505 : {
1506 10 : CurrPos += SizeOfXLogLongPHD;
1507 10 : currpos += SizeOfXLogLongPHD;
1508 : }
1509 : else
1510 : {
1511 20341 : CurrPos += SizeOfXLogShortPHD;
1512 20341 : currpos += SizeOfXLogShortPHD;
1513 : }
1514 20351 : freespace = INSERT_FREESPACE(CurrPos);
1515 : }
1516 :
1517 4597081 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || rdata_len == 0);
1518 4597081 : memcpy(currpos, rdata_data, rdata_len);
1519 4597081 : currpos += rdata_len;
1520 4597081 : CurrPos += rdata_len;
1521 4597081 : freespace -= rdata_len;
1522 4597081 : written += rdata_len;
1523 :
1524 4597081 : rdata = rdata->next;
1525 : }
1526 1306714 : Assert(written == write_len);
1527 :
1528 : /*
1529 : * If this was an xlog-switch, it's not enough to write the switch record,
1530 : * we also have to consume all the remaining space in the WAL segment. We
1531 : * have already reserved it for us, but we still need to make sure it's
1532 : * allocated and zeroed in the WAL buffers so that when the caller (or
1533 : * someone else) does XLogWrite(), it can really write out all the zeros.
1534 : */
1535 1306714 : if (isLogSwitch && CurrPos % XLOG_SEG_SIZE != 0)
1536 : {
1537 : /* An xlog-switch record doesn't contain any data besides the header */
1538 0 : Assert(write_len == SizeOfXLogRecord);
1539 :
1540 : /*
1541 : * We do this one page at a time, to make sure we don't deadlock
1542 : * against ourselves if wal_buffers < XLOG_SEG_SIZE.
1543 : */
1544 0 : Assert(EndPos % XLogSegSize == 0);
1545 :
1546 : /* Use up all the remaining space on the first page */
1547 0 : CurrPos += freespace;
1548 :
1549 0 : while (CurrPos < EndPos)
1550 : {
1551 : /* initialize the next page (if not initialized already) */
1552 0 : WALInsertLockUpdateInsertingAt(CurrPos);
1553 0 : AdvanceXLInsertBuffer(CurrPos, false);
1554 0 : CurrPos += XLOG_BLCKSZ;
1555 : }
1556 : }
1557 : else
1558 : {
1559 : /* Align the end position, so that the next record starts aligned */
1560 1306714 : CurrPos = MAXALIGN64(CurrPos);
1561 : }
1562 :
1563 1306714 : if (CurrPos != EndPos)
1564 0 : elog(PANIC, "space reserved for WAL record does not match what was written");
1565 1306714 : }
1566 :
1567 : /*
1568 : * Acquire a WAL insertion lock, for inserting to WAL.
1569 : */
1570 : static void
1571 1306714 : WALInsertLockAcquire(void)
1572 : {
1573 : bool immed;
1574 :
1575 : /*
1576 : * It doesn't matter which of the WAL insertion locks we acquire, so try
1577 : * the one we used last time. If the system isn't particularly busy, it's
1578 : * a good bet that it's still available, and it's good to have some
1579 : * affinity to a particular lock so that you don't unnecessarily bounce
1580 : * cache lines between processes when there's no contention.
1581 : *
1582 : * If this is the first time through in this backend, pick a lock
1583 : * (semi-)randomly. This allows the locks to be used evenly if you have a
1584 : * lot of very short connections.
1585 : */
1586 : static int lockToTry = -1;
1587 :
1588 1306714 : if (lockToTry == -1)
1589 212 : lockToTry = MyProc->pgprocno % NUM_XLOGINSERT_LOCKS;
1590 1306714 : MyLockNo = lockToTry;
1591 :
1592 : /*
1593 : * The insertingAt value is initially set to 0, as we don't know our
1594 : * insert location yet.
1595 : */
1596 1306714 : immed = LWLockAcquire(&WALInsertLocks[MyLockNo].l.lock, LW_EXCLUSIVE);
1597 1306714 : if (!immed)
1598 : {
1599 : /*
1600 : * If we couldn't get the lock immediately, try another lock next
1601 : * time. On a system with more insertion locks than concurrent
1602 : * inserters, this causes all the inserters to eventually migrate to a
1603 : * lock that no-one else is using. On a system with more inserters
1604 : * than locks, it still helps to distribute the inserters evenly
1605 : * across the locks.
1606 : */
1607 45 : lockToTry = (lockToTry + 1) % NUM_XLOGINSERT_LOCKS;
1608 : }
1609 1306714 : }
1610 :
1611 : /*
1612 : * Acquire all WAL insertion locks, to prevent other backends from inserting
1613 : * to WAL.
1614 : */
1615 : static void
1616 11 : WALInsertLockAcquireExclusive(void)
1617 : {
1618 : int i;
1619 :
1620 : /*
1621 : * When holding all the locks, all but the last lock's insertingAt
1622 : * indicator is set to 0xFFFFFFFFFFFFFFFF, which is higher than any real
1623 : * XLogRecPtr value, to make sure that no-one blocks waiting on those.
1624 : */
1625 88 : for (i = 0; i < NUM_XLOGINSERT_LOCKS - 1; i++)
1626 : {
1627 77 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1628 77 : LWLockUpdateVar(&WALInsertLocks[i].l.lock,
1629 77 : &WALInsertLocks[i].l.insertingAt,
1630 : PG_UINT64_MAX);
1631 : }
1632 : /* Variable value reset to 0 at release */
1633 11 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1634 :
1635 11 : holdingAllLocks = true;
1636 11 : }
1637 :
1638 : /*
1639 : * Release our insertion lock (or locks, if we're holding them all).
1640 : *
1641 : * NB: Reset all variables to 0, so they cause LWLockWaitForVar to block the
1642 : * next time the lock is acquired.
1643 : */
1644 : static void
1645 1306725 : WALInsertLockRelease(void)
1646 : {
1647 1306725 : if (holdingAllLocks)
1648 : {
1649 : int i;
1650 :
1651 99 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1652 88 : LWLockReleaseClearVar(&WALInsertLocks[i].l.lock,
1653 88 : &WALInsertLocks[i].l.insertingAt,
1654 : 0);
1655 :
1656 11 : holdingAllLocks = false;
1657 : }
1658 : else
1659 : {
1660 1306714 : LWLockReleaseClearVar(&WALInsertLocks[MyLockNo].l.lock,
1661 1306714 : &WALInsertLocks[MyLockNo].l.insertingAt,
1662 : 0);
1663 : }
1664 1306725 : }
1665 :
1666 : /*
1667 : * Update our insertingAt value, to let others know that we've finished
1668 : * inserting up to that point.
1669 : */
1670 : static void
1671 1026 : WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt)
1672 : {
1673 1026 : if (holdingAllLocks)
1674 : {
1675 : /*
1676 : * We use the last lock to mark our actual position, see comments in
1677 : * WALInsertLockAcquireExclusive.
1678 : */
1679 0 : LWLockUpdateVar(&WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.lock,
1680 0 : &WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.insertingAt,
1681 : insertingAt);
1682 : }
1683 : else
1684 1026 : LWLockUpdateVar(&WALInsertLocks[MyLockNo].l.lock,
1685 1026 : &WALInsertLocks[MyLockNo].l.insertingAt,
1686 : insertingAt);
1687 1026 : }
1688 :
1689 : /*
1690 : * Wait for any WAL insertions < upto to finish.
1691 : *
1692 : * Returns the location of the oldest insertion that is still in-progress.
1693 : * Any WAL prior to that point has been fully copied into WAL buffers, and
1694 : * can be flushed out to disk. Because this waits for any insertions older
1695 : * than 'upto' to finish, the return value is always >= 'upto'.
1696 : *
1697 : * Note: When you are about to write out WAL, you must call this function
1698 : * *before* acquiring WALWriteLock, to avoid deadlocks. This function might
1699 : * need to wait for an insertion to finish (or at least advance to next
1700 : * uninitialized page), and the inserter might need to evict an old WAL buffer
1701 : * to make room for a new one, which in turn requires WALWriteLock.
1702 : */
1703 : static XLogRecPtr
1704 10333 : WaitXLogInsertionsToFinish(XLogRecPtr upto)
1705 : {
1706 : uint64 bytepos;
1707 : XLogRecPtr reservedUpto;
1708 : XLogRecPtr finishedUpto;
1709 10333 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1710 : int i;
1711 :
1712 10333 : if (MyProc == NULL)
1713 0 : elog(PANIC, "cannot wait without a PGPROC structure");
1714 :
1715 : /* Read the current insert position */
1716 10333 : SpinLockAcquire(&Insert->insertpos_lck);
1717 10333 : bytepos = Insert->CurrBytePos;
1718 10333 : SpinLockRelease(&Insert->insertpos_lck);
1719 10333 : reservedUpto = XLogBytePosToEndRecPtr(bytepos);
1720 :
1721 : /*
1722 : * No-one should request to flush a piece of WAL that hasn't even been
1723 : * reserved yet. However, it can happen if there is a block with a bogus
1724 : * LSN on disk, for example. XLogFlush checks for that situation and
1725 : * complains, but only after the flush. Here we just assume that to mean
1726 : * that all WAL that has been reserved needs to be finished. In this
1727 : * corner-case, the return value can be smaller than 'upto' argument.
1728 : */
1729 10333 : if (upto > reservedUpto)
1730 : {
1731 0 : elog(LOG, "request to flush past end of generated WAL; request %X/%X, currpos %X/%X",
1732 : (uint32) (upto >> 32), (uint32) upto,
1733 : (uint32) (reservedUpto >> 32), (uint32) reservedUpto);
1734 0 : upto = reservedUpto;
1735 : }
1736 :
1737 : /*
1738 : * Loop through all the locks, sleeping on any in-progress insert older
1739 : * than 'upto'.
1740 : *
1741 : * finishedUpto is our return value, indicating the point upto which all
1742 : * the WAL insertions have been finished. Initialize it to the head of
1743 : * reserved WAL, and as we iterate through the insertion locks, back it
1744 : * out for any insertion that's still in progress.
1745 : */
1746 10333 : finishedUpto = reservedUpto;
1747 92997 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1748 : {
1749 82664 : XLogRecPtr insertingat = InvalidXLogRecPtr;
1750 :
1751 : do
1752 : {
1753 : /*
1754 : * See if this insertion is in progress. LWLockWait will wait for
1755 : * the lock to be released, or for the 'value' to be set by a
1756 : * LWLockUpdateVar call. When a lock is initially acquired, its
1757 : * value is 0 (InvalidXLogRecPtr), which means that we don't know
1758 : * where it's inserting yet. We will have to wait for it. If
1759 : * it's a small insertion, the record will most likely fit on the
1760 : * same page and the inserter will release the lock without ever
1761 : * calling LWLockUpdateVar. But if it has to sleep, it will
1762 : * advertise the insertion point with LWLockUpdateVar before
1763 : * sleeping.
1764 : */
1765 165336 : if (LWLockWaitForVar(&WALInsertLocks[i].l.lock,
1766 82668 : &WALInsertLocks[i].l.insertingAt,
1767 : insertingat, &insertingat))
1768 : {
1769 : /* the lock was free, so no insertion in progress */
1770 82489 : insertingat = InvalidXLogRecPtr;
1771 82489 : break;
1772 : }
1773 :
1774 : /*
1775 : * This insertion is still in progress. Have to wait, unless the
1776 : * inserter has proceeded past 'upto'.
1777 : */
1778 179 : } while (insertingat < upto);
1779 :
1780 82664 : if (insertingat != InvalidXLogRecPtr && insertingat < finishedUpto)
1781 175 : finishedUpto = insertingat;
1782 : }
1783 10333 : return finishedUpto;
1784 : }
1785 :
1786 : /*
1787 : * Get a pointer to the right location in the WAL buffer containing the
1788 : * given XLogRecPtr.
1789 : *
1790 : * If the page is not initialized yet, it is initialized. That might require
1791 : * evicting an old dirty buffer from the buffer cache, which means I/O.
1792 : *
1793 : * The caller must ensure that the page containing the requested location
1794 : * isn't evicted yet, and won't be evicted. The way to ensure that is to
1795 : * hold onto a WAL insertion lock with the insertingAt position set to
1796 : * something <= ptr. GetXLogBuffer() will update insertingAt if it needs
1797 : * to evict an old page from the buffer. (This means that once you call
1798 : * GetXLogBuffer() with a given 'ptr', you must not access anything before
1799 : * that point anymore, and must not call GetXLogBuffer() with an older 'ptr'
1800 : * later, because older buffers might be recycled already)
1801 : */
1802 : static char *
1803 1327065 : GetXLogBuffer(XLogRecPtr ptr)
1804 : {
1805 : int idx;
1806 : XLogRecPtr endptr;
1807 : static uint64 cachedPage = 0;
1808 : static char *cachedPos = NULL;
1809 : XLogRecPtr expectedEndPtr;
1810 :
1811 : /*
1812 : * Fast path for the common case that we need to access again the same
1813 : * page as last time.
1814 : */
1815 1327065 : if (ptr / XLOG_BLCKSZ == cachedPage)
1816 : {
1817 1295596 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1818 1295596 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1819 1295596 : return cachedPos + ptr % XLOG_BLCKSZ;
1820 : }
1821 :
1822 : /*
1823 : * The XLog buffer cache is organized so that a page is always loaded to a
1824 : * particular buffer. That way we can easily calculate the buffer a given
1825 : * page must be loaded into, from the XLogRecPtr alone.
1826 : */
1827 31469 : idx = XLogRecPtrToBufIdx(ptr);
1828 :
1829 : /*
1830 : * See what page is loaded in the buffer at the moment. It could be the
1831 : * page we're looking for, or something older. It can't be anything newer
1832 : * - that would imply the page we're looking for has already been written
1833 : * out to disk and evicted, and the caller is responsible for making sure
1834 : * that doesn't happen.
1835 : *
1836 : * However, we don't hold a lock while we read the value. If someone has
1837 : * just initialized the page, it's possible that we get a "torn read" of
1838 : * the XLogRecPtr if 64-bit fetches are not atomic on this platform. In
1839 : * that case we will see a bogus value. That's ok, we'll grab the mapping
1840 : * lock (in AdvanceXLInsertBuffer) and retry if we see anything else than
1841 : * the page we're looking for. But it means that when we do this unlocked
1842 : * read, we might see a value that appears to be ahead of the page we're
1843 : * looking for. Don't PANIC on that, until we've verified the value while
1844 : * holding the lock.
1845 : */
1846 31469 : expectedEndPtr = ptr;
1847 31469 : expectedEndPtr += XLOG_BLCKSZ - ptr % XLOG_BLCKSZ;
1848 :
1849 31469 : endptr = XLogCtl->xlblocks[idx];
1850 31469 : if (expectedEndPtr != endptr)
1851 : {
1852 : XLogRecPtr initializedUpto;
1853 :
1854 : /*
1855 : * Before calling AdvanceXLInsertBuffer(), which can block, let others
1856 : * know how far we're finished with inserting the record.
1857 : *
1858 : * NB: If 'ptr' points to just after the page header, advertise a
1859 : * position at the beginning of the page rather than 'ptr' itself. If
1860 : * there are no other insertions running, someone might try to flush
1861 : * up to our advertised location. If we advertised a position after
1862 : * the page header, someone might try to flush the page header, even
1863 : * though page might actually not be initialized yet. As the first
1864 : * inserter on the page, we are effectively responsible for making
1865 : * sure that it's initialized, before we let insertingAt to move past
1866 : * the page header.
1867 : */
1868 1062 : if (ptr % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
1869 36 : ptr % XLOG_SEG_SIZE > XLOG_BLCKSZ)
1870 36 : initializedUpto = ptr - SizeOfXLogShortPHD;
1871 990 : else if (ptr % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
1872 0 : ptr % XLOG_SEG_SIZE < XLOG_BLCKSZ)
1873 0 : initializedUpto = ptr - SizeOfXLogLongPHD;
1874 : else
1875 990 : initializedUpto = ptr;
1876 :
1877 1026 : WALInsertLockUpdateInsertingAt(initializedUpto);
1878 :
1879 1026 : AdvanceXLInsertBuffer(ptr, false);
1880 1026 : endptr = XLogCtl->xlblocks[idx];
1881 :
1882 1026 : if (expectedEndPtr != endptr)
1883 0 : elog(PANIC, "could not find WAL buffer for %X/%X",
1884 : (uint32) (ptr >> 32), (uint32) ptr);
1885 : }
1886 : else
1887 : {
1888 : /*
1889 : * Make sure the initialization of the page is visible to us, and
1890 : * won't arrive later to overwrite the WAL data we write on the page.
1891 : */
1892 30443 : pg_memory_barrier();
1893 : }
1894 :
1895 : /*
1896 : * Found the buffer holding this page. Return a pointer to the right
1897 : * offset within the page.
1898 : */
1899 31469 : cachedPage = ptr / XLOG_BLCKSZ;
1900 31469 : cachedPos = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1901 :
1902 31469 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1903 31469 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1904 :
1905 31469 : return cachedPos + ptr % XLOG_BLCKSZ;
1906 : }
1907 :
1908 : /*
1909 : * Converts a "usable byte position" to XLogRecPtr. A usable byte position
1910 : * is the position starting from the beginning of WAL, excluding all WAL
1911 : * page headers.
1912 : */
1913 : static XLogRecPtr
1914 2613439 : XLogBytePosToRecPtr(uint64 bytepos)
1915 : {
1916 : uint64 fullsegs;
1917 : uint64 fullpages;
1918 : uint64 bytesleft;
1919 : uint32 seg_offset;
1920 : XLogRecPtr result;
1921 :
1922 2613439 : fullsegs = bytepos / UsableBytesInSegment;
1923 2613439 : bytesleft = bytepos % UsableBytesInSegment;
1924 :
1925 2613439 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1926 : {
1927 : /* fits on first page of segment */
1928 1493 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1929 : }
1930 : else
1931 : {
1932 : /* account for the first page on segment with long header */
1933 2611946 : seg_offset = XLOG_BLCKSZ;
1934 2611946 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1935 :
1936 2611946 : fullpages = bytesleft / UsableBytesInPage;
1937 2611946 : bytesleft = bytesleft % UsableBytesInPage;
1938 :
1939 2611946 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1940 : }
1941 :
1942 2613439 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, result);
1943 :
1944 2613439 : return result;
1945 : }
1946 :
1947 : /*
1948 : * Like XLogBytePosToRecPtr, but if the position is at a page boundary,
1949 : * returns a pointer to the beginning of the page (ie. before page header),
1950 : * not to where the first xlog record on that page would go to. This is used
1951 : * when converting a pointer to the end of a record.
1952 : */
1953 : static XLogRecPtr
1954 1317047 : XLogBytePosToEndRecPtr(uint64 bytepos)
1955 : {
1956 : uint64 fullsegs;
1957 : uint64 fullpages;
1958 : uint64 bytesleft;
1959 : uint32 seg_offset;
1960 : XLogRecPtr result;
1961 :
1962 1317047 : fullsegs = bytepos / UsableBytesInSegment;
1963 1317047 : bytesleft = bytepos % UsableBytesInSegment;
1964 :
1965 1317047 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1966 : {
1967 : /* fits on first page of segment */
1968 762 : if (bytesleft == 0)
1969 0 : seg_offset = 0;
1970 : else
1971 762 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1972 : }
1973 : else
1974 : {
1975 : /* account for the first page on segment with long header */
1976 1316285 : seg_offset = XLOG_BLCKSZ;
1977 1316285 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1978 :
1979 1316285 : fullpages = bytesleft / UsableBytesInPage;
1980 1316285 : bytesleft = bytesleft % UsableBytesInPage;
1981 :
1982 1316285 : if (bytesleft == 0)
1983 636 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft;
1984 : else
1985 1315649 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1986 : }
1987 :
1988 1317047 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, result);
1989 :
1990 1317047 : return result;
1991 : }
1992 :
1993 : /*
1994 : * Convert an XLogRecPtr to a "usable byte position".
1995 : */
1996 : static uint64
1997 3920148 : XLogRecPtrToBytePos(XLogRecPtr ptr)
1998 : {
1999 : uint64 fullsegs;
2000 : uint32 fullpages;
2001 : uint32 offset;
2002 : uint64 result;
2003 :
2004 3920148 : XLByteToSeg(ptr, fullsegs);
2005 :
2006 3920148 : fullpages = (ptr % XLOG_SEG_SIZE) / XLOG_BLCKSZ;
2007 3920148 : offset = ptr % XLOG_BLCKSZ;
2008 :
2009 3920148 : if (fullpages == 0)
2010 : {
2011 2240 : result = fullsegs * UsableBytesInSegment;
2012 2240 : if (offset > 0)
2013 : {
2014 2240 : Assert(offset >= SizeOfXLogLongPHD);
2015 2240 : result += offset - SizeOfXLogLongPHD;
2016 : }
2017 : }
2018 : else
2019 : {
2020 7835816 : result = fullsegs * UsableBytesInSegment +
2021 3917908 : (XLOG_BLCKSZ - SizeOfXLogLongPHD) + /* account for first page */
2022 3917908 : (fullpages - 1) * UsableBytesInPage; /* full pages */
2023 3917908 : if (offset > 0)
2024 : {
2025 3917278 : Assert(offset >= SizeOfXLogShortPHD);
2026 3917278 : result += offset - SizeOfXLogShortPHD;
2027 : }
2028 : }
2029 :
2030 3920148 : return result;
2031 : }
2032 :
2033 : /*
2034 : * Initialize XLOG buffers, writing out old buffers if they still contain
2035 : * unwritten data, upto the page containing 'upto'. Or if 'opportunistic' is
2036 : * true, initialize as many pages as we can without having to write out
2037 : * unwritten data. Any new pages are initialized to zeros, with pages headers
2038 : * initialized properly.
2039 : */
2040 : static void
2041 1608 : AdvanceXLInsertBuffer(XLogRecPtr upto, bool opportunistic)
2042 : {
2043 1608 : XLogCtlInsert *Insert = &XLogCtl->Insert;
2044 : int nextidx;
2045 : XLogRecPtr OldPageRqstPtr;
2046 : XLogwrtRqst WriteRqst;
2047 1608 : XLogRecPtr NewPageEndPtr = InvalidXLogRecPtr;
2048 : XLogRecPtr NewPageBeginPtr;
2049 : XLogPageHeader NewPage;
2050 1608 : int npages = 0;
2051 :
2052 1608 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2053 :
2054 : /*
2055 : * Now that we have the lock, check if someone initialized the page
2056 : * already.
2057 : */
2058 24871 : while (upto >= XLogCtl->InitializedUpTo || opportunistic)
2059 : {
2060 22237 : nextidx = XLogRecPtrToBufIdx(XLogCtl->InitializedUpTo);
2061 :
2062 : /*
2063 : * Get ending-offset of the buffer page we need to replace (this may
2064 : * be zero if the buffer hasn't been used yet). Fall through if it's
2065 : * already written out.
2066 : */
2067 22237 : OldPageRqstPtr = XLogCtl->xlblocks[nextidx];
2068 22237 : if (LogwrtResult.Write < OldPageRqstPtr)
2069 : {
2070 : /*
2071 : * Nope, got work to do. If we just want to pre-initialize as much
2072 : * as we can without flushing, give up now.
2073 : */
2074 749 : if (opportunistic)
2075 582 : break;
2076 :
2077 : /* Before waiting, get info_lck and update LogwrtResult */
2078 167 : SpinLockAcquire(&XLogCtl->info_lck);
2079 167 : if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
2080 0 : XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
2081 167 : LogwrtResult = XLogCtl->LogwrtResult;
2082 167 : SpinLockRelease(&XLogCtl->info_lck);
2083 :
2084 : /*
2085 : * Now that we have an up-to-date LogwrtResult value, see if we
2086 : * still need to write it or if someone else already did.
2087 : */
2088 167 : if (LogwrtResult.Write < OldPageRqstPtr)
2089 : {
2090 : /*
2091 : * Must acquire write lock. Release WALBufMappingLock first,
2092 : * to make sure that all insertions that we need to wait for
2093 : * can finish (up to this same position). Otherwise we risk
2094 : * deadlock.
2095 : */
2096 167 : LWLockRelease(WALBufMappingLock);
2097 :
2098 167 : WaitXLogInsertionsToFinish(OldPageRqstPtr);
2099 :
2100 167 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
2101 :
2102 167 : LogwrtResult = XLogCtl->LogwrtResult;
2103 167 : if (LogwrtResult.Write >= OldPageRqstPtr)
2104 : {
2105 : /* OK, someone wrote it already */
2106 3 : LWLockRelease(WALWriteLock);
2107 : }
2108 : else
2109 : {
2110 : /* Have to write it ourselves */
2111 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_START();
2112 164 : WriteRqst.Write = OldPageRqstPtr;
2113 164 : WriteRqst.Flush = 0;
2114 164 : XLogWrite(WriteRqst, false);
2115 164 : LWLockRelease(WALWriteLock);
2116 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_DONE();
2117 : }
2118 : /* Re-acquire WALBufMappingLock and retry */
2119 167 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2120 167 : continue;
2121 : }
2122 : }
2123 :
2124 : /*
2125 : * Now the next buffer slot is free and we can set it up to be the
2126 : * next output page.
2127 : */
2128 21488 : NewPageBeginPtr = XLogCtl->InitializedUpTo;
2129 21488 : NewPageEndPtr = NewPageBeginPtr + XLOG_BLCKSZ;
2130 :
2131 21488 : Assert(XLogRecPtrToBufIdx(NewPageBeginPtr) == nextidx);
2132 :
2133 21488 : NewPage = (XLogPageHeader) (XLogCtl->pages + nextidx * (Size) XLOG_BLCKSZ);
2134 :
2135 : /*
2136 : * Be sure to re-zero the buffer so that bytes beyond what we've
2137 : * written will look like zeroes and not valid XLOG records...
2138 : */
2139 21488 : MemSet((char *) NewPage, 0, XLOG_BLCKSZ);
2140 :
2141 : /*
2142 : * Fill the new page's header
2143 : */
2144 21488 : NewPage->xlp_magic = XLOG_PAGE_MAGIC;
2145 :
2146 : /* NewPage->xlp_info = 0; */ /* done by memset */
2147 21488 : NewPage->xlp_tli = ThisTimeLineID;
2148 21488 : NewPage->xlp_pageaddr = NewPageBeginPtr;
2149 :
2150 : /* NewPage->xlp_rem_len = 0; */ /* done by memset */
2151 :
2152 : /*
2153 : * If online backup is not in progress, mark the header to indicate
2154 : * that* WAL records beginning in this page have removable backup
2155 : * blocks. This allows the WAL archiver to know whether it is safe to
2156 : * compress archived WAL data by transforming full-block records into
2157 : * the non-full-block format. It is sufficient to record this at the
2158 : * page level because we force a page switch (in fact a segment
2159 : * switch) when starting a backup, so the flag will be off before any
2160 : * records can be written during the backup. At the end of a backup,
2161 : * the last page will be marked as all unsafe when perhaps only part
2162 : * is unsafe, but at worst the archiver would miss the opportunity to
2163 : * compress a few records.
2164 : */
2165 21488 : if (!Insert->forcePageWrites)
2166 21488 : NewPage->xlp_info |= XLP_BKP_REMOVABLE;
2167 :
2168 : /*
2169 : * If first page of an XLOG segment file, make it a long header.
2170 : */
2171 21488 : if ((NewPage->xlp_pageaddr % XLogSegSize) == 0)
2172 : {
2173 10 : XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
2174 :
2175 10 : NewLongPage->xlp_sysid = ControlFile->system_identifier;
2176 10 : NewLongPage->xlp_seg_size = XLogSegSize;
2177 10 : NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
2178 10 : NewPage->xlp_info |= XLP_LONG_HEADER;
2179 : }
2180 :
2181 : /*
2182 : * Make sure the initialization of the page becomes visible to others
2183 : * before the xlblocks update. GetXLogBuffer() reads xlblocks without
2184 : * holding a lock.
2185 : */
2186 21488 : pg_write_barrier();
2187 :
2188 21488 : *((volatile XLogRecPtr *) &XLogCtl->xlblocks[nextidx]) = NewPageEndPtr;
2189 :
2190 21488 : XLogCtl->InitializedUpTo = NewPageEndPtr;
2191 :
2192 21488 : npages++;
2193 : }
2194 1608 : LWLockRelease(WALBufMappingLock);
2195 :
2196 : #ifdef WAL_DEBUG
2197 : if (XLOG_DEBUG && npages > 0)
2198 : {
2199 : elog(DEBUG1, "initialized %d pages, up to %X/%X",
2200 : npages, (uint32) (NewPageEndPtr >> 32), (uint32) NewPageEndPtr);
2201 : }
2202 : #endif
2203 1608 : }
2204 :
2205 : /*
2206 : * Calculate CheckPointSegments based on max_wal_size_mb and
2207 : * checkpoint_completion_target.
2208 : */
2209 : static void
2210 5 : CalculateCheckpointSegments(void)
2211 : {
2212 : double target;
2213 :
2214 : /*-------
2215 : * Calculate the distance at which to trigger a checkpoint, to avoid
2216 : * exceeding max_wal_size_mb. This is based on two assumptions:
2217 : *
2218 : * a) we keep WAL for two checkpoint cycles, back to the "prev" checkpoint.
2219 : * b) during checkpoint, we consume checkpoint_completion_target *
2220 : * number of segments consumed between checkpoints.
2221 : *-------
2222 : */
2223 5 : target = (double) ConvertToXSegs(max_wal_size_mb) / (2.0 + CheckPointCompletionTarget);
2224 :
2225 : /* round down */
2226 5 : CheckPointSegments = (int) target;
2227 :
2228 5 : if (CheckPointSegments < 1)
2229 0 : CheckPointSegments = 1;
2230 5 : }
2231 :
2232 : void
2233 5 : assign_max_wal_size(int newval, void *extra)
2234 : {
2235 5 : max_wal_size_mb = newval;
2236 5 : CalculateCheckpointSegments();
2237 5 : }
2238 :
2239 : void
2240 0 : assign_checkpoint_completion_target(double newval, void *extra)
2241 : {
2242 0 : CheckPointCompletionTarget = newval;
2243 0 : CalculateCheckpointSegments();
2244 0 : }
2245 :
2246 : /*
2247 : * At a checkpoint, how many WAL segments to recycle as preallocated future
2248 : * XLOG segments? Returns the highest segment that should be preallocated.
2249 : */
2250 : static XLogSegNo
2251 0 : XLOGfileslop(XLogRecPtr PriorRedoPtr)
2252 : {
2253 : XLogSegNo minSegNo;
2254 : XLogSegNo maxSegNo;
2255 : double distance;
2256 : XLogSegNo recycleSegNo;
2257 :
2258 : /*
2259 : * Calculate the segment numbers that min_wal_size_mb and max_wal_size_mb
2260 : * correspond to. Always recycle enough segments to meet the minimum, and
2261 : * remove enough segments to stay below the maximum.
2262 : */
2263 0 : minSegNo = PriorRedoPtr / XLOG_SEG_SIZE + ConvertToXSegs(min_wal_size_mb) - 1;
2264 0 : maxSegNo = PriorRedoPtr / XLOG_SEG_SIZE + ConvertToXSegs(max_wal_size_mb) - 1;
2265 :
2266 : /*
2267 : * Between those limits, recycle enough segments to get us through to the
2268 : * estimated end of next checkpoint.
2269 : *
2270 : * To estimate where the next checkpoint will finish, assume that the
2271 : * system runs steadily consuming CheckPointDistanceEstimate bytes between
2272 : * every checkpoint.
2273 : *
2274 : * The reason this calculation is done from the prior checkpoint, not the
2275 : * one that just finished, is that this behaves better if some checkpoint
2276 : * cycles are abnormally short, like if you perform a manual checkpoint
2277 : * right after a timed one. The manual checkpoint will make almost a full
2278 : * cycle's worth of WAL segments available for recycling, because the
2279 : * segments from the prior's prior, fully-sized checkpoint cycle are no
2280 : * longer needed. However, the next checkpoint will make only few segments
2281 : * available for recycling, the ones generated between the timed
2282 : * checkpoint and the manual one right after that. If at the manual
2283 : * checkpoint we only retained enough segments to get us to the next timed
2284 : * one, and removed the rest, then at the next checkpoint we would not
2285 : * have enough segments around for recycling, to get us to the checkpoint
2286 : * after that. Basing the calculations on the distance from the prior redo
2287 : * pointer largely fixes that problem.
2288 : */
2289 0 : distance = (2.0 + CheckPointCompletionTarget) * CheckPointDistanceEstimate;
2290 : /* add 10% for good measure. */
2291 0 : distance *= 1.10;
2292 :
2293 0 : recycleSegNo = (XLogSegNo) ceil(((double) PriorRedoPtr + distance) / XLOG_SEG_SIZE);
2294 :
2295 0 : if (recycleSegNo < minSegNo)
2296 0 : recycleSegNo = minSegNo;
2297 0 : if (recycleSegNo > maxSegNo)
2298 0 : recycleSegNo = maxSegNo;
2299 :
2300 0 : return recycleSegNo;
2301 : }
2302 :
2303 : /*
2304 : * Check whether we've consumed enough xlog space that a checkpoint is needed.
2305 : *
2306 : * new_segno indicates a log file that has just been filled up (or read
2307 : * during recovery). We measure the distance from RedoRecPtr to new_segno
2308 : * and see if that exceeds CheckPointSegments.
2309 : *
2310 : * Note: it is caller's responsibility that RedoRecPtr is up-to-date.
2311 : */
2312 : static bool
2313 10 : XLogCheckpointNeeded(XLogSegNo new_segno)
2314 : {
2315 : XLogSegNo old_segno;
2316 :
2317 10 : XLByteToSeg(RedoRecPtr, old_segno);
2318 :
2319 10 : if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
2320 0 : return true;
2321 10 : return false;
2322 : }
2323 :
2324 : /*
2325 : * Write and/or fsync the log at least as far as WriteRqst indicates.
2326 : *
2327 : * If flexible == TRUE, we don't have to write as far as WriteRqst, but
2328 : * may stop at any convenient boundary (such as a cache or logfile boundary).
2329 : * This option allows us to avoid uselessly issuing multiple writes when a
2330 : * single one would do.
2331 : *
2332 : * Must be called with WALWriteLock held. WaitXLogInsertionsToFinish(WriteRqst)
2333 : * must be called before grabbing the lock, to make sure the data is ready to
2334 : * write.
2335 : */
2336 : static void
2337 10084 : XLogWrite(XLogwrtRqst WriteRqst, bool flexible)
2338 : {
2339 : bool ispartialpage;
2340 : bool last_iteration;
2341 : bool finishing_seg;
2342 : bool use_existent;
2343 : int curridx;
2344 : int npages;
2345 : int startidx;
2346 : uint32 startoffset;
2347 :
2348 : /* We should always be inside a critical section here */
2349 10084 : Assert(CritSectionCount > 0);
2350 :
2351 : /*
2352 : * Update local LogwrtResult (caller probably did this already, but...)
2353 : */
2354 10084 : LogwrtResult = XLogCtl->LogwrtResult;
2355 :
2356 : /*
2357 : * Since successive pages in the xlog cache are consecutively allocated,
2358 : * we can usually gather multiple pages together and issue just one
2359 : * write() call. npages is the number of pages we have determined can be
2360 : * written together; startidx is the cache block index of the first one,
2361 : * and startoffset is the file offset at which it should go. The latter
2362 : * two variables are only valid when npages > 0, but we must initialize
2363 : * all of them to keep the compiler quiet.
2364 : */
2365 10084 : npages = 0;
2366 10084 : startidx = 0;
2367 10084 : startoffset = 0;
2368 :
2369 : /*
2370 : * Within the loop, curridx is the cache block index of the page to
2371 : * consider writing. Begin at the buffer containing the next unwritten
2372 : * page, or last partially written page.
2373 : */
2374 10084 : curridx = XLogRecPtrToBufIdx(LogwrtResult.Write);
2375 :
2376 40671 : while (LogwrtResult.Write < WriteRqst.Write)
2377 : {
2378 : /*
2379 : * Make sure we're not ahead of the insert process. This could happen
2380 : * if we're passed a bogus WriteRqst.Write that is past the end of the
2381 : * last page that's been initialized by AdvanceXLInsertBuffer.
2382 : */
2383 30414 : XLogRecPtr EndPtr = XLogCtl->xlblocks[curridx];
2384 :
2385 30414 : if (LogwrtResult.Write >= EndPtr)
2386 0 : elog(PANIC, "xlog write request %X/%X is past end of log %X/%X",
2387 : (uint32) (LogwrtResult.Write >> 32),
2388 : (uint32) LogwrtResult.Write,
2389 : (uint32) (EndPtr >> 32), (uint32) EndPtr);
2390 :
2391 : /* Advance LogwrtResult.Write to end of current buffer page */
2392 30414 : LogwrtResult.Write = EndPtr;
2393 30414 : ispartialpage = WriteRqst.Write < LogwrtResult.Write;
2394 :
2395 30414 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo))
2396 : {
2397 : /*
2398 : * Switch to new logfile segment. We cannot have any pending
2399 : * pages here (since we dump what we have at segment end).
2400 : */
2401 256 : Assert(npages == 0);
2402 256 : if (openLogFile >= 0)
2403 51 : XLogFileClose();
2404 256 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo);
2405 :
2406 : /* create/use new log file */
2407 256 : use_existent = true;
2408 256 : openLogFile = XLogFileInit(openLogSegNo, &use_existent, true);
2409 256 : openLogOff = 0;
2410 : }
2411 :
2412 : /* Make sure we have the current logfile open */
2413 30414 : if (openLogFile < 0)
2414 : {
2415 0 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo);
2416 0 : openLogFile = XLogFileOpen(openLogSegNo);
2417 0 : openLogOff = 0;
2418 : }
2419 :
2420 : /* Add current page to the set of pending pages-to-dump */
2421 30414 : if (npages == 0)
2422 : {
2423 : /* first of group */
2424 10112 : startidx = curridx;
2425 10112 : startoffset = (LogwrtResult.Write - XLOG_BLCKSZ) % XLogSegSize;
2426 : }
2427 30414 : npages++;
2428 :
2429 : /*
2430 : * Dump the set if this will be the last loop iteration, or if we are
2431 : * at the last page of the cache area (since the next page won't be
2432 : * contiguous in memory), or if we are at the end of the logfile
2433 : * segment.
2434 : */
2435 30414 : last_iteration = WriteRqst.Write <= LogwrtResult.Write;
2436 :
2437 51395 : finishing_seg = !ispartialpage &&
2438 20981 : (startoffset + npages * XLOG_BLCKSZ) >= XLogSegSize;
2439 :
2440 50754 : if (last_iteration ||
2441 40642 : curridx == XLogCtl->XLogCacheBlck ||
2442 : finishing_seg)
2443 : {
2444 : char *from;
2445 : Size nbytes;
2446 : Size nleft;
2447 : int written;
2448 :
2449 : /* Need to seek in the file? */
2450 10112 : if (openLogOff != startoffset)
2451 : {
2452 8875 : if (lseek(openLogFile, (off_t) startoffset, SEEK_SET) < 0)
2453 0 : ereport(PANIC,
2454 : (errcode_for_file_access(),
2455 : errmsg("could not seek in log file %s to offset %u: %m",
2456 : XLogFileNameP(ThisTimeLineID, openLogSegNo),
2457 : startoffset)));
2458 8875 : openLogOff = startoffset;
2459 : }
2460 :
2461 : /* OK to write the page(s) */
2462 10112 : from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
2463 10112 : nbytes = npages * (Size) XLOG_BLCKSZ;
2464 10112 : nleft = nbytes;
2465 : do
2466 : {
2467 10112 : errno = 0;
2468 10112 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
2469 10112 : written = write(openLogFile, from, nleft);
2470 10112 : pgstat_report_wait_end();
2471 10112 : if (written <= 0)
2472 : {
2473 0 : if (errno == EINTR)
2474 0 : continue;
2475 0 : ereport(PANIC,
2476 : (errcode_for_file_access(),
2477 : errmsg("could not write to log file %s "
2478 : "at offset %u, length %zu: %m",
2479 : XLogFileNameP(ThisTimeLineID, openLogSegNo),
2480 : openLogOff, nbytes)));
2481 : }
2482 10112 : nleft -= written;
2483 10112 : from += written;
2484 10112 : } while (nleft > 0);
2485 :
2486 : /* Update state for write */
2487 10112 : openLogOff += nbytes;
2488 10112 : npages = 0;
2489 :
2490 : /*
2491 : * If we just wrote the whole last page of a logfile segment,
2492 : * fsync the segment immediately. This avoids having to go back
2493 : * and re-open prior segments when an fsync request comes along
2494 : * later. Doing it here ensures that one and only one backend will
2495 : * perform this fsync.
2496 : *
2497 : * This is also the right place to notify the Archiver that the
2498 : * segment is ready to copy to archival storage, and to update the
2499 : * timer for archive_timeout, and to signal for a checkpoint if
2500 : * too many logfile segments have been used since the last
2501 : * checkpoint.
2502 : */
2503 10112 : if (finishing_seg)
2504 : {
2505 10 : issue_xlog_fsync(openLogFile, openLogSegNo);
2506 :
2507 : /* signal that we need to wakeup walsenders later */
2508 10 : WalSndWakeupRequest();
2509 :
2510 10 : LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
2511 :
2512 10 : if (XLogArchivingActive())
2513 0 : XLogArchiveNotifySeg(openLogSegNo);
2514 :
2515 10 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
2516 10 : XLogCtl->lastSegSwitchLSN = LogwrtResult.Flush;
2517 :
2518 : /*
2519 : * Request a checkpoint if we've consumed too much xlog since
2520 : * the last one. For speed, we first check using the local
2521 : * copy of RedoRecPtr, which might be out of date; if it looks
2522 : * like a checkpoint is needed, forcibly update RedoRecPtr and
2523 : * recheck.
2524 : */
2525 10 : if (IsUnderPostmaster && XLogCheckpointNeeded(openLogSegNo))
2526 : {
2527 0 : (void) GetRedoRecPtr();
2528 0 : if (XLogCheckpointNeeded(openLogSegNo))
2529 0 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
2530 : }
2531 : }
2532 : }
2533 :
2534 30414 : if (ispartialpage)
2535 : {
2536 : /* Only asked to write a partial page */
2537 9433 : LogwrtResult.Write = WriteRqst.Write;
2538 9433 : break;
2539 : }
2540 20981 : curridx = NextBufIdx(curridx);
2541 :
2542 : /* If flexible, break out of loop as soon as we wrote something */
2543 20981 : if (flexible && npages == 0)
2544 478 : break;
2545 : }
2546 :
2547 10084 : Assert(npages == 0);
2548 :
2549 : /*
2550 : * If asked to flush, do so
2551 : */
2552 19652 : if (LogwrtResult.Flush < WriteRqst.Flush &&
2553 9568 : LogwrtResult.Flush < LogwrtResult.Write)
2554 :
2555 : {
2556 : /*
2557 : * Could get here without iterating above loop, in which case we might
2558 : * have no open file or the wrong one. However, we do not need to
2559 : * fsync more than one file.
2560 : */
2561 19134 : if (sync_method != SYNC_METHOD_OPEN &&
2562 9567 : sync_method != SYNC_METHOD_OPEN_DSYNC)
2563 : {
2564 19133 : if (openLogFile >= 0 &&
2565 9566 : !XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo))
2566 0 : XLogFileClose();
2567 9567 : if (openLogFile < 0)
2568 : {
2569 1 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo);
2570 1 : openLogFile = XLogFileOpen(openLogSegNo);
2571 1 : openLogOff = 0;
2572 : }
2573 :
2574 9567 : issue_xlog_fsync(openLogFile, openLogSegNo);
2575 : }
2576 :
2577 : /* signal that we need to wakeup walsenders later */
2578 9567 : WalSndWakeupRequest();
2579 :
2580 9567 : LogwrtResult.Flush = LogwrtResult.Write;
2581 : }
2582 :
2583 : /*
2584 : * Update shared-memory status
2585 : *
2586 : * We make sure that the shared 'request' values do not fall behind the
2587 : * 'result' values. This is not absolutely essential, but it saves some
2588 : * code in a couple of places.
2589 : */
2590 : {
2591 10084 : SpinLockAcquire(&XLogCtl->info_lck);
2592 10084 : XLogCtl->LogwrtResult = LogwrtResult;
2593 10084 : if (XLogCtl->LogwrtRqst.Write < LogwrtResult.Write)
2594 8951 : XLogCtl->LogwrtRqst.Write = LogwrtResult.Write;
2595 10084 : if (XLogCtl->LogwrtRqst.Flush < LogwrtResult.Flush)
2596 9569 : XLogCtl->LogwrtRqst.Flush = LogwrtResult.Flush;
2597 10084 : SpinLockRelease(&XLogCtl->info_lck);
2598 : }
2599 10084 : }
2600 :
2601 : /*
2602 : * Record the LSN for an asynchronous transaction commit/abort
2603 : * and nudge the WALWriter if there is work for it to do.
2604 : * (This should not be called for synchronous commits.)
2605 : */
2606 : void
2607 1580 : XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN)
2608 : {
2609 1580 : XLogRecPtr WriteRqstPtr = asyncXactLSN;
2610 : bool sleeping;
2611 :
2612 1580 : SpinLockAcquire(&XLogCtl->info_lck);
2613 1580 : LogwrtResult = XLogCtl->LogwrtResult;
2614 1580 : sleeping = XLogCtl->WalWriterSleeping;
2615 1580 : if (XLogCtl->asyncXactLSN < asyncXactLSN)
2616 1574 : XLogCtl->asyncXactLSN = asyncXactLSN;
2617 1580 : SpinLockRelease(&XLogCtl->info_lck);
2618 :
2619 : /*
2620 : * If the WALWriter is sleeping, we should kick it to make it come out of
2621 : * low-power mode. Otherwise, determine whether there's a full page of
2622 : * WAL available to write.
2623 : */
2624 1580 : if (!sleeping)
2625 : {
2626 : /* back off to last completed page boundary */
2627 1580 : WriteRqstPtr -= WriteRqstPtr % XLOG_BLCKSZ;
2628 :
2629 : /* if we have already flushed that far, we're done */
2630 1580 : if (WriteRqstPtr <= LogwrtResult.Flush)
2631 2567 : return;
2632 : }
2633 :
2634 : /*
2635 : * Nudge the WALWriter: it has a full page of WAL to write, or we want it
2636 : * to come out of low-power mode so that this async commit will reach disk
2637 : * within the expected amount of time.
2638 : */
2639 593 : if (ProcGlobal->walwriterLatch)
2640 503 : SetLatch(ProcGlobal->walwriterLatch);
2641 : }
2642 :
2643 : /*
2644 : * Record the LSN up to which we can remove WAL because it's not required by
2645 : * any replication slot.
2646 : */
2647 : void
2648 3 : XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn)
2649 : {
2650 3 : SpinLockAcquire(&XLogCtl->info_lck);
2651 3 : XLogCtl->replicationSlotMinLSN = lsn;
2652 3 : SpinLockRelease(&XLogCtl->info_lck);
2653 3 : }
2654 :
2655 :
2656 : /*
2657 : * Return the oldest LSN we must retain to satisfy the needs of some
2658 : * replication slot.
2659 : */
2660 : static XLogRecPtr
2661 11 : XLogGetReplicationSlotMinimumLSN(void)
2662 : {
2663 : XLogRecPtr retval;
2664 :
2665 11 : SpinLockAcquire(&XLogCtl->info_lck);
2666 11 : retval = XLogCtl->replicationSlotMinLSN;
2667 11 : SpinLockRelease(&XLogCtl->info_lck);
2668 :
2669 11 : return retval;
2670 : }
2671 :
2672 : /*
2673 : * Advance minRecoveryPoint in control file.
2674 : *
2675 : * If we crash during recovery, we must reach this point again before the
2676 : * database is consistent.
2677 : *
2678 : * If 'force' is true, 'lsn' argument is ignored. Otherwise, minRecoveryPoint
2679 : * is only updated if it's not already greater than or equal to 'lsn'.
2680 : */
2681 : static void
2682 0 : UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
2683 : {
2684 : /* Quick check using our local copy of the variable */
2685 0 : if (!updateMinRecoveryPoint || (!force && lsn <= minRecoveryPoint))
2686 0 : return;
2687 :
2688 0 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2689 :
2690 : /* update local copy */
2691 0 : minRecoveryPoint = ControlFile->minRecoveryPoint;
2692 0 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
2693 :
2694 : /*
2695 : * An invalid minRecoveryPoint means that we need to recover all the WAL,
2696 : * i.e., we're doing crash recovery. We never modify the control file's
2697 : * value in that case, so we can short-circuit future checks here too.
2698 : */
2699 0 : if (minRecoveryPoint == 0)
2700 0 : updateMinRecoveryPoint = false;
2701 0 : else if (force || minRecoveryPoint < lsn)
2702 : {
2703 : XLogRecPtr newMinRecoveryPoint;
2704 : TimeLineID newMinRecoveryPointTLI;
2705 :
2706 : /*
2707 : * To avoid having to update the control file too often, we update it
2708 : * all the way to the last record being replayed, even though 'lsn'
2709 : * would suffice for correctness. This also allows the 'force' case
2710 : * to not need a valid 'lsn' value.
2711 : *
2712 : * Another important reason for doing it this way is that the passed
2713 : * 'lsn' value could be bogus, i.e., past the end of available WAL, if
2714 : * the caller got it from a corrupted heap page. Accepting such a
2715 : * value as the min recovery point would prevent us from coming up at
2716 : * all. Instead, we just log a warning and continue with recovery.
2717 : * (See also the comments about corrupt LSNs in XLogFlush.)
2718 : */
2719 0 : SpinLockAcquire(&XLogCtl->info_lck);
2720 0 : newMinRecoveryPoint = XLogCtl->replayEndRecPtr;
2721 0 : newMinRecoveryPointTLI = XLogCtl->replayEndTLI;
2722 0 : SpinLockRelease(&XLogCtl->info_lck);
2723 :
2724 0 : if (!force && newMinRecoveryPoint < lsn)
2725 0 : elog(WARNING,
2726 : "xlog min recovery request %X/%X is past current point %X/%X",
2727 : (uint32) (lsn >> 32), (uint32) lsn,
2728 : (uint32) (newMinRecoveryPoint >> 32),
2729 : (uint32) newMinRecoveryPoint);
2730 :
2731 : /* update control file */
2732 0 : if (ControlFile->minRecoveryPoint < newMinRecoveryPoint)
2733 : {
2734 0 : ControlFile->minRecoveryPoint = newMinRecoveryPoint;
2735 0 : ControlFile->minRecoveryPointTLI = newMinRecoveryPointTLI;
2736 0 : UpdateControlFile();
2737 0 : minRecoveryPoint = newMinRecoveryPoint;
2738 0 : minRecoveryPointTLI = newMinRecoveryPointTLI;
2739 :
2740 0 : ereport(DEBUG2,
2741 : (errmsg("updated min recovery point to %X/%X on timeline %u",
2742 : (uint32) (minRecoveryPoint >> 32),
2743 : (uint32) minRecoveryPoint,
2744 : newMinRecoveryPointTLI)));
2745 : }
2746 : }
2747 0 : LWLockRelease(ControlFileLock);
2748 : }
2749 :
2750 : /*
2751 : * Ensure that all XLOG data through the given position is flushed to disk.
2752 : *
2753 : * NOTE: this differs from XLogWrite mainly in that the WALWriteLock is not
2754 : * already held, and we try to avoid acquiring it if possible.
2755 : */
2756 : void
2757 17690 : XLogFlush(XLogRecPtr record)
2758 : {
2759 : XLogRecPtr WriteRqstPtr;
2760 : XLogwrtRqst WriteRqst;
2761 :
2762 : /*
2763 : * During REDO, we are reading not writing WAL. Therefore, instead of
2764 : * trying to flush the WAL, we should update minRecoveryPoint instead. We
2765 : * test XLogInsertAllowed(), not InRecovery, because we need checkpointer
2766 : * to act this way too, and because when it tries to write the
2767 : * end-of-recovery checkpoint, it should indeed flush.
2768 : */
2769 17690 : if (!XLogInsertAllowed())
2770 : {
2771 0 : UpdateMinRecoveryPoint(record, false);
2772 0 : return;
2773 : }
2774 :
2775 : /* Quick exit if already known flushed */
2776 17690 : if (record <= LogwrtResult.Flush)
2777 8163 : return;
2778 :
2779 : #ifdef WAL_DEBUG
2780 : if (XLOG_DEBUG)
2781 : elog(LOG, "xlog flush request %X/%X; write %X/%X; flush %X/%X",
2782 : (uint32) (record >> 32), (uint32) record,
2783 : (uint32) (LogwrtResult.Write >> 32), (uint32) LogwrtResult.Write,
2784 : (uint32) (LogwrtResult.Flush >> 32), (uint32) LogwrtResult.Flush);
2785 : #endif
2786 :
2787 9527 : START_CRIT_SECTION();
2788 :
2789 : /*
2790 : * Since fsync is usually a horribly expensive operation, we try to
2791 : * piggyback as much data as we can on each fsync: if we see any more data
2792 : * entered into the xlog buffer, we'll write and fsync that too, so that
2793 : * the final value of LogwrtResult.Flush is as large as possible. This
2794 : * gives us some chance of avoiding another fsync immediately after.
2795 : */
2796 :
2797 : /* initialize to given target; may increase below */
2798 9527 : WriteRqstPtr = record;
2799 :
2800 : /*
2801 : * Now wait until we get the write lock, or someone else does the flush
2802 : * for us.
2803 : */
2804 : for (;;)
2805 : {
2806 : XLogRecPtr insertpos;
2807 :
2808 : /* read LogwrtResult and update local state */
2809 9640 : SpinLockAcquire(&XLogCtl->info_lck);
2810 9640 : if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2811 83 : WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2812 9640 : LogwrtResult = XLogCtl->LogwrtResult;
2813 9640 : SpinLockRelease(&XLogCtl->info_lck);
2814 :
2815 : /* done already? */
2816 9640 : if (record <= LogwrtResult.Flush)
2817 56 : break;
2818 :
2819 : /*
2820 : * Before actually performing the write, wait for all in-flight
2821 : * insertions to the pages we're about to write to finish.
2822 : */
2823 9584 : insertpos = WaitXLogInsertionsToFinish(WriteRqstPtr);
2824 :
2825 : /*
2826 : * Try to get the write lock. If we can't get it immediately, wait
2827 : * until it's released, and recheck if we still need to do the flush
2828 : * or if the backend that held the lock did it for us already. This
2829 : * helps to maintain a good rate of group committing when the system
2830 : * is bottlenecked by the speed of fsyncing.
2831 : */
2832 9584 : if (!LWLockAcquireOrWait(WALWriteLock, LW_EXCLUSIVE))
2833 : {
2834 : /*
2835 : * The lock is now free, but we didn't acquire it yet. Before we
2836 : * do, loop back to check if someone else flushed the record for
2837 : * us already.
2838 : */
2839 113 : continue;
2840 : }
2841 :
2842 : /* Got the lock; recheck whether request is satisfied */
2843 9471 : LogwrtResult = XLogCtl->LogwrtResult;
2844 9471 : if (record <= LogwrtResult.Flush)
2845 : {
2846 51 : LWLockRelease(WALWriteLock);
2847 51 : break;
2848 : }
2849 :
2850 : /*
2851 : * Sleep before flush! By adding a delay here, we may give further
2852 : * backends the opportunity to join the backlog of group commit
2853 : * followers; this can significantly improve transaction throughput,
2854 : * at the risk of increasing transaction latency.
2855 : *
2856 : * We do not sleep if enableFsync is not turned on, nor if there are
2857 : * fewer than CommitSiblings other backends with active transactions.
2858 : */
2859 9420 : if (CommitDelay > 0 && enableFsync &&
2860 0 : MinimumActiveBackends(CommitSiblings))
2861 : {
2862 0 : pg_usleep(CommitDelay);
2863 :
2864 : /*
2865 : * Re-check how far we can now flush the WAL. It's generally not
2866 : * safe to call WaitXLogInsertionsToFinish while holding
2867 : * WALWriteLock, because an in-progress insertion might need to
2868 : * also grab WALWriteLock to make progress. But we know that all
2869 : * the insertions up to insertpos have already finished, because
2870 : * that's what the earlier WaitXLogInsertionsToFinish() returned.
2871 : * We're only calling it again to allow insertpos to be moved
2872 : * further forward, not to actually wait for anyone.
2873 : */
2874 0 : insertpos = WaitXLogInsertionsToFinish(insertpos);
2875 : }
2876 :
2877 : /* try to write/flush later additions to XLOG as well */
2878 9420 : WriteRqst.Write = insertpos;
2879 9420 : WriteRqst.Flush = insertpos;
2880 :
2881 9420 : XLogWrite(WriteRqst, false);
2882 :
2883 9420 : LWLockRelease(WALWriteLock);
2884 : /* done */
2885 9420 : break;
2886 113 : }
2887 :
2888 9527 : END_CRIT_SECTION();
2889 :
2890 : /* wake up walsenders now that we've released heavily contended locks */
2891 9527 : WalSndWakeupProcessRequests();
2892 :
2893 : /*
2894 : * If we still haven't flushed to the request point then we have a
2895 : * problem; most likely, the requested flush point is past end of XLOG.
2896 : * This has been seen to occur when a disk page has a corrupted LSN.
2897 : *
2898 : * Formerly we treated this as a PANIC condition, but that hurts the
2899 : * system's robustness rather than helping it: we do not want to take down
2900 : * the whole system due to corruption on one data page. In particular, if
2901 : * the bad page is encountered again during recovery then we would be
2902 : * unable to restart the database at all! (This scenario actually
2903 : * happened in the field several times with 7.1 releases.) As of 8.4, bad
2904 : * LSNs encountered during recovery are UpdateMinRecoveryPoint's problem;
2905 : * the only time we can reach here during recovery is while flushing the
2906 : * end-of-recovery checkpoint record, and we don't expect that to have a
2907 : * bad LSN.
2908 : *
2909 : * Note that for calls from xact.c, the ERROR will be promoted to PANIC
2910 : * since xact.c calls this routine inside a critical section. However,
2911 : * calls from bufmgr.c are not within critical sections and so we will not
2912 : * force a restart for a bad LSN on a data page.
2913 : */
2914 9527 : if (LogwrtResult.Flush < record)
2915 0 : elog(ERROR,
2916 : "xlog flush request %X/%X is not satisfied --- flushed only to %X/%X",
2917 : (uint32) (record >> 32), (uint32) record,
2918 : (uint32) (LogwrtResult.Flush >> 32), (uint32) LogwrtResult.Flush);
2919 : }
2920 :
2921 : /*
2922 : * Write & flush xlog, but without specifying exactly where to.
2923 : *
2924 : * We normally write only completed blocks; but if there is nothing to do on
2925 : * that basis, we check for unwritten async commits in the current incomplete
2926 : * block, and write through the latest one of those. Thus, if async commits
2927 : * are not being used, we will write complete blocks only.
2928 : *
2929 : * If, based on the above, there's anything to write we do so immediately. But
2930 : * to avoid calling fsync, fdatasync et. al. at a rate that'd impact
2931 : * concurrent IO, we only flush WAL every wal_writer_delay ms, or if there's
2932 : * more than wal_writer_flush_after unflushed blocks.
2933 : *
2934 : * We can guarantee that async commits reach disk after at most three
2935 : * wal_writer_delay cycles. (When flushing complete blocks, we allow XLogWrite
2936 : * to write "flexibly", meaning it can stop at the end of the buffer ring;
2937 : * this makes a difference only with very high load or long wal_writer_delay,
2938 : * but imposes one extra cycle for the worst case for async commits.)
2939 : *
2940 : * This routine is invoked periodically by the background walwriter process.
2941 : *
2942 : * Returns TRUE if there was any work to do, even if we skipped flushing due
2943 : * to wal_writer_delay/wal_writer_flush_after.
2944 : */
2945 : bool
2946 904 : XLogBackgroundFlush(void)
2947 : {
2948 : XLogwrtRqst WriteRqst;
2949 904 : bool flexible = true;
2950 : static TimestampTz lastflush;
2951 : TimestampTz now;
2952 : int flushbytes;
2953 :
2954 : /* XLOG doesn't need flushing during recovery */
2955 904 : if (RecoveryInProgress())
2956 0 : return false;
2957 :
2958 : /* read LogwrtResult and update local state */
2959 904 : SpinLockAcquire(&XLogCtl->info_lck);
2960 904 : LogwrtResult = XLogCtl->LogwrtResult;
2961 904 : WriteRqst = XLogCtl->LogwrtRqst;
2962 904 : SpinLockRelease(&XLogCtl->info_lck);
2963 :
2964 : /* back off to last completed page boundary */
2965 904 : WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
2966 :
2967 : /* if we have already flushed that far, consider async commit records */
2968 904 : if (WriteRqst.Write <= LogwrtResult.Flush)
2969 : {
2970 344 : SpinLockAcquire(&XLogCtl->info_lck);
2971 344 : WriteRqst.Write = XLogCtl->asyncXactLSN;
2972 344 : SpinLockRelease(&XLogCtl->info_lck);
2973 344 : flexible = false; /* ensure it all gets written */
2974 : }
2975 :
2976 : /*
2977 : * If already known flushed, we're done. Just need to check if we are
2978 : * holding an open file handle to a logfile that's no longer in use,
2979 : * preventing the file from being deleted.
2980 : */
2981 904 : if (WriteRqst.Write <= LogwrtResult.Flush)
2982 : {
2983 322 : if (openLogFile >= 0)
2984 : {
2985 289 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo))
2986 : {
2987 2 : XLogFileClose();
2988 : }
2989 : }
2990 322 : return false;
2991 : }
2992 :
2993 : /*
2994 : * Determine how far to flush WAL, based on the wal_writer_delay and
2995 : * wal_writer_flush_after GUCs.
2996 : */
2997 582 : now = GetCurrentTimestamp();
2998 1164 : flushbytes =
2999 1164 : WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
3000 :
3001 582 : if (WalWriterFlushAfter == 0 || lastflush == 0)
3002 : {
3003 : /* first call, or block based limits disabled */
3004 1 : WriteRqst.Flush = WriteRqst.Write;
3005 1 : lastflush = now;
3006 : }
3007 581 : else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
3008 : {
3009 : /*
3010 : * Flush the writes at least every WalWriteDelay ms. This is important
3011 : * to bound the amount of time it takes for an asynchronous commit to
3012 : * hit disk.
3013 : */
3014 154 : WriteRqst.Flush = WriteRqst.Write;
3015 154 : lastflush = now;
3016 : }
3017 427 : else if (flushbytes >= WalWriterFlushAfter)
3018 : {
3019 : /* exceeded wal_writer_flush_after blocks, flush */
3020 0 : WriteRqst.Flush = WriteRqst.Write;
3021 0 : lastflush = now;
3022 : }
3023 : else
3024 : {
3025 : /* no flushing, this time round */
3026 427 : WriteRqst.Flush = 0;
3027 : }
3028 :
3029 : #ifdef WAL_DEBUG
3030 : if (XLOG_DEBUG)
3031 : elog(LOG, "xlog bg flush request write %X/%X; flush: %X/%X, current is write %X/%X; flush %X/%X",
3032 : (uint32) (WriteRqst.Write >> 32), (uint32) WriteRqst.Write,
3033 : (uint32) (WriteRqst.Flush >> 32), (uint32) WriteRqst.Flush,
3034 : (uint32) (LogwrtResult.Write >> 32), (uint32) LogwrtResult.Write,
3035 : (uint32) (LogwrtResult.Flush >> 32), (uint32) LogwrtResult.Flush);
3036 : #endif
3037 :
3038 582 : START_CRIT_SECTION();
3039 :
3040 : /* now wait for any in-progress insertions to finish and get write lock */
3041 582 : WaitXLogInsertionsToFinish(WriteRqst.Write);
3042 582 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
3043 582 : LogwrtResult = XLogCtl->LogwrtResult;
3044 664 : if (WriteRqst.Write > LogwrtResult.Write ||
3045 82 : WriteRqst.Flush > LogwrtResult.Flush)
3046 : {
3047 500 : XLogWrite(WriteRqst, flexible);
3048 : }
3049 582 : LWLockRelease(WALWriteLock);
3050 :
3051 582 : END_CRIT_SECTION();
3052 :
3053 : /* wake up walsenders now that we've released heavily contended locks */
3054 582 : WalSndWakeupProcessRequests();
3055 :
3056 : /*
3057 : * Great, done. To take some work off the critical path, try to initialize
3058 : * as many of the no-longer-needed WAL buffers for future use as we can.
3059 : */
3060 582 : AdvanceXLInsertBuffer(InvalidXLogRecPtr, true);
3061 :
3062 : /*
3063 : * If we determined that we need to write data, but somebody else
3064 : * wrote/flushed already, it should be considered as being active, to
3065 : * avoid hibernating too early.
3066 : */
3067 582 : return true;
3068 : }
3069 :
3070 : /*
3071 : * Test whether XLOG data has been flushed up to (at least) the given position.
3072 : *
3073 : * Returns true if a flush is still needed. (It may be that someone else
3074 : * is already in process of flushing that far, however.)
3075 : */
3076 : bool
3077 704592 : XLogNeedsFlush(XLogRecPtr record)
3078 : {
3079 : /*
3080 : * During recovery, we don't flush WAL but update minRecoveryPoint
3081 : * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3082 : * would need to be updated.
3083 : */
3084 704592 : if (RecoveryInProgress())
3085 : {
3086 : /* Quick exit if already known updated */
3087 0 : if (record <= minRecoveryPoint || !updateMinRecoveryPoint)
3088 0 : return false;
3089 :
3090 : /*
3091 : * Update local copy of minRecoveryPoint. But if the lock is busy,
3092 : * just return a conservative guess.
3093 : */
3094 0 : if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
3095 0 : return true;
3096 0 : minRecoveryPoint = ControlFile->minRecoveryPoint;
3097 0 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
3098 0 : LWLockRelease(ControlFileLock);
3099 :
3100 : /*
3101 : * An invalid minRecoveryPoint means that we need to recover all the
3102 : * WAL, i.e., we're doing crash recovery. We never modify the control
3103 : * file's value in that case, so we can short-circuit future checks
3104 : * here too.
3105 : */
3106 0 : if (minRecoveryPoint == 0)
3107 0 : updateMinRecoveryPoint = false;
3108 :
3109 : /* check again */
3110 0 : if (record <= minRecoveryPoint || !updateMinRecoveryPoint)
3111 0 : return false;
3112 : else
3113 0 : return true;
3114 : }
3115 :
3116 : /* Quick exit if already known flushed */
3117 704592 : if (record <= LogwrtResult.Flush)
3118 702664 : return false;
3119 :
3120 : /* read LogwrtResult and update local state */
3121 1928 : SpinLockAcquire(&XLogCtl->info_lck);
3122 1928 : LogwrtResult = XLogCtl->LogwrtResult;
3123 1928 : SpinLockRelease(&XLogCtl->info_lck);
3124 :
3125 : /* check again */
3126 1928 : if (record <= LogwrtResult.Flush)
3127 204 : return false;
3128 :
3129 1724 : return true;
3130 : }
3131 :
3132 : /*
3133 : * Create a new XLOG file segment, or open a pre-existing one.
3134 : *
3135 : * log, seg: identify segment to be created/opened.
3136 : *
3137 : * *use_existent: if TRUE, OK to use a pre-existing file (else, any
3138 : * pre-existing file will be deleted). On return, TRUE if a pre-existing
3139 : * file was used.
3140 : *
3141 : * use_lock: if TRUE, acquire ControlFileLock while moving file into
3142 : * place. This should be TRUE except during bootstrap log creation. The
3143 : * caller must *not* hold the lock at call.
3144 : *
3145 : * Returns FD of opened file.
3146 : *
3147 : * Note: errors here are ERROR not PANIC because we might or might not be
3148 : * inside a critical section (eg, during checkpoint there is no reason to
3149 : * take down the system on failure). They will promote to PANIC if we are
3150 : * in a critical section.
3151 : */
3152 : int
3153 257 : XLogFileInit(XLogSegNo logsegno, bool *use_existent, bool use_lock)
3154 : {
3155 : char path[MAXPGPATH];
3156 : char tmppath[MAXPGPATH];
3157 : char zbuffer_raw[XLOG_BLCKSZ + MAXIMUM_ALIGNOF];
3158 : char *zbuffer;
3159 : XLogSegNo installed_segno;
3160 : XLogSegNo max_segno;
3161 : int fd;
3162 : int nbytes;
3163 :
3164 257 : XLogFilePath(path, ThisTimeLineID, logsegno);
3165 :
3166 : /*
3167 : * Try to use existent file (checkpoint maker may have created it already)
3168 : */
3169 257 : if (*use_existent)
3170 : {
3171 256 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | get_sync_bit(sync_method),
3172 : S_IRUSR | S_IWUSR);
3173 256 : if (fd < 0)
3174 : {
3175 10 : if (errno != ENOENT)
3176 0 : ereport(ERROR,
3177 : (errcode_for_file_access(),
3178 : errmsg("could not open file \"%s\": %m", path)));
3179 : }
3180 : else
3181 246 : return fd;
3182 : }
3183 :
3184 : /*
3185 : * Initialize an empty (all zeroes) segment. NOTE: it is possible that
3186 : * another process is doing the same thing. If so, we will end up
3187 : * pre-creating an extra log segment. That seems OK, and better than
3188 : * holding the lock throughout this lengthy process.
3189 : */
3190 11 : elog(DEBUG2, "creating and filling new WAL file");
3191 :
3192 11 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3193 :
3194 11 : unlink(tmppath);
3195 :
3196 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3197 11 : fd = BasicOpenFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
3198 : S_IRUSR | S_IWUSR);
3199 11 : if (fd < 0)
3200 0 : ereport(ERROR,
3201 : (errcode_for_file_access(),
3202 : errmsg("could not create file \"%s\": %m", tmppath)));
3203 :
3204 : /*
3205 : * Zero-fill the file. We have to do this the hard way to ensure that all
3206 : * the file space has really been allocated --- on platforms that allow
3207 : * "holes" in files, just seeking to the end doesn't allocate intermediate
3208 : * space. This way, we know that we have all the space and (after the
3209 : * fsync below) that all the indirect blocks are down on disk. Therefore,
3210 : * fdatasync(2) or O_DSYNC will be sufficient to sync future writes to the
3211 : * log file.
3212 : *
3213 : * Note: ensure the buffer is reasonably well-aligned; this may save a few
3214 : * cycles transferring data to the kernel.
3215 : */
3216 11 : zbuffer = (char *) MAXALIGN(zbuffer_raw);
3217 11 : memset(zbuffer, 0, XLOG_BLCKSZ);
3218 22539 : for (nbytes = 0; nbytes < XLogSegSize; nbytes += XLOG_BLCKSZ)
3219 : {
3220 22528 : errno = 0;
3221 22528 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_WRITE);
3222 22528 : if ((int) write(fd, zbuffer, XLOG_BLCKSZ) != (int) XLOG_BLCKSZ)
3223 : {
3224 0 : int save_errno = errno;
3225 :
3226 : /*
3227 : * If we fail to make the file, delete it to release disk space
3228 : */
3229 0 : unlink(tmppath);
3230 :
3231 0 : close(fd);
3232 :
3233 : /* if write didn't set errno, assume problem is no disk space */
3234 0 : errno = save_errno ? save_errno : ENOSPC;
3235 :
3236 0 : ereport(ERROR,
3237 : (errcode_for_file_access(),
3238 : errmsg("could not write to file \"%s\": %m", tmppath)));
3239 : }
3240 22528 : pgstat_report_wait_end();
3241 : }
3242 :
3243 11 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_SYNC);
3244 11 : if (pg_fsync(fd) != 0)
3245 : {
3246 0 : close(fd);
3247 0 : ereport(ERROR,
3248 : (errcode_for_file_access(),
3249 : errmsg("could not fsync file \"%s\": %m", tmppath)));
3250 : }
3251 11 : pgstat_report_wait_end();
3252 :
3253 11 : if (close(fd))
3254 0 : ereport(ERROR,
3255 : (errcode_for_file_access(),
3256 : errmsg("could not close file \"%s\": %m", tmppath)));
3257 :
3258 : /*
3259 : * Now move the segment into place with its final name.
3260 : *
3261 : * If caller didn't want to use a pre-existing file, get rid of any
3262 : * pre-existing file. Otherwise, cope with possibility that someone else
3263 : * has created the file while we were filling ours: if so, use ours to
3264 : * pre-create a future log segment.
3265 : */
3266 11 : installed_segno = logsegno;
3267 :
3268 : /*
3269 : * XXX: What should we use as max_segno? We used to use XLOGfileslop when
3270 : * that was a constant, but that was always a bit dubious: normally, at a
3271 : * checkpoint, XLOGfileslop was the offset from the checkpoint record, but
3272 : * here, it was the offset from the insert location. We can't do the
3273 : * normal XLOGfileslop calculation here because we don't have access to
3274 : * the prior checkpoint's redo location. So somewhat arbitrarily, just use
3275 : * CheckPointSegments.
3276 : */
3277 11 : max_segno = logsegno + CheckPointSegments;
3278 22 : if (!InstallXLogFileSegment(&installed_segno, tmppath,
3279 11 : *use_existent, max_segno,
3280 : use_lock))
3281 : {
3282 : /*
3283 : * No need for any more future segments, or InstallXLogFileSegment()
3284 : * failed to rename the file into place. If the rename failed, opening
3285 : * the file below will fail.
3286 : */
3287 0 : unlink(tmppath);
3288 : }
3289 :
3290 : /* Set flag to tell caller there was no existent file */
3291 11 : *use_existent = false;
3292 :
3293 : /* Now open original target segment (might not be file I just made) */
3294 11 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | get_sync_bit(sync_method),
3295 : S_IRUSR | S_IWUSR);
3296 11 : if (fd < 0)
3297 0 : ereport(ERROR,
3298 : (errcode_for_file_access(),
3299 : errmsg("could not open file \"%s\": %m", path)));
3300 :
3301 11 : elog(DEBUG2, "done creating and filling new WAL file");
3302 :
3303 11 : return fd;
3304 : }
3305 :
3306 : /*
3307 : * Create a new XLOG file segment by copying a pre-existing one.
3308 : *
3309 : * destsegno: identify segment to be created.
3310 : *
3311 : * srcTLI, srcsegno: identify segment to be copied (could be from
3312 : * a different timeline)
3313 : *
3314 : * upto: how much of the source file to copy (the rest is filled with
3315 : * zeros)
3316 : *
3317 : * Currently this is only used during recovery, and so there are no locking
3318 : * considerations. But we should be just as tense as XLogFileInit to avoid
3319 : * emplacing a bogus file.
3320 : */
3321 : static void
3322 0 : XLogFileCopy(XLogSegNo destsegno, TimeLineID srcTLI, XLogSegNo srcsegno,
3323 : int upto)
3324 : {
3325 : char path[MAXPGPATH];
3326 : char tmppath[MAXPGPATH];
3327 : char buffer[XLOG_BLCKSZ];
3328 : int srcfd;
3329 : int fd;
3330 : int nbytes;
3331 :
3332 : /*
3333 : * Open the source file
3334 : */
3335 0 : XLogFilePath(path, srcTLI, srcsegno);
3336 0 : srcfd = OpenTransientFile(path, O_RDONLY | PG_BINARY, 0);
3337 0 : if (srcfd < 0)
3338 0 : ereport(ERROR,
3339 : (errcode_for_file_access(),
3340 : errmsg("could not open file \"%s\": %m", path)));
3341 :
3342 : /*
3343 : * Copy into a temp file name.
3344 : */
3345 0 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3346 :
3347 0 : unlink(tmppath);
3348 :
3349 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3350 0 : fd = OpenTransientFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
3351 : S_IRUSR | S_IWUSR);
3352 0 : if (fd < 0)
3353 0 : ereport(ERROR,
3354 : (errcode_for_file_access(),
3355 : errmsg("could not create file \"%s\": %m", tmppath)));
3356 :
3357 : /*
3358 : * Do the data copying.
3359 : */
3360 0 : for (nbytes = 0; nbytes < XLogSegSize; nbytes += sizeof(buffer))
3361 : {
3362 : int nread;
3363 :
3364 0 : nread = upto - nbytes;
3365 :
3366 : /*
3367 : * The part that is not read from the source file is filled with
3368 : * zeros.
3369 : */
3370 0 : if (nread < sizeof(buffer))
3371 0 : memset(buffer, 0, sizeof(buffer));
3372 :
3373 0 : if (nread > 0)
3374 : {
3375 0 : if (nread > sizeof(buffer))
3376 0 : nread = sizeof(buffer);
3377 0 : errno = 0;
3378 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_READ);
3379 0 : if (read(srcfd, buffer, nread) != nread)
3380 : {
3381 0 : if (errno != 0)
3382 0 : ereport(ERROR,
3383 : (errcode_for_file_access(),
3384 : errmsg("could not read file \"%s\": %m",
3385 : path)));
3386 : else
3387 0 : ereport(ERROR,
3388 : (errmsg("not enough data in file \"%s\"",
3389 : path)));
3390 : }
3391 0 : pgstat_report_wait_end();
3392 : }
3393 0 : errno = 0;
3394 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_WRITE);
3395 0 : if ((int) write(fd, buffer, sizeof(buffer)) != (int) sizeof(buffer))
3396 : {
3397 0 : int save_errno = errno;
3398 :
3399 : /*
3400 : * If we fail to make the file, delete it to release disk space
3401 : */
3402 0 : unlink(tmppath);
3403 : /* if write didn't set errno, assume problem is no disk space */
3404 0 : errno = save_errno ? save_errno : ENOSPC;
3405 :
3406 0 : ereport(ERROR,
3407 : (errcode_for_file_access(),
3408 : errmsg("could not write to file \"%s\": %m", tmppath)));
3409 : }
3410 0 : pgstat_report_wait_end();
3411 : }
3412 :
3413 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_SYNC);
3414 0 : if (pg_fsync(fd) != 0)
3415 0 : ereport(ERROR,
3416 : (errcode_for_file_access(),
3417 : errmsg("could not fsync file \"%s\": %m", tmppath)));
3418 0 : pgstat_report_wait_end();
3419 :
3420 0 : if (CloseTransientFile(fd))
3421 0 : ereport(ERROR,
3422 : (errcode_for_file_access(),
3423 : errmsg("could not close file \"%s\": %m", tmppath)));
3424 :
3425 0 : CloseTransientFile(srcfd);
3426 :
3427 : /*
3428 : * Now move the segment into place with its final name.
3429 : */
3430 0 : if (!InstallXLogFileSegment(&destsegno, tmppath, false, 0, false))
3431 0 : elog(ERROR, "InstallXLogFileSegment should not have failed");
3432 0 : }
3433 :
3434 : /*
3435 : * Install a new XLOG segment file as a current or future log segment.
3436 : *
3437 : * This is used both to install a newly-created segment (which has a temp
3438 : * filename while it's being created) and to recycle an old segment.
3439 : *
3440 : * *segno: identify segment to install as (or first possible target).
3441 : * When find_free is TRUE, this is modified on return to indicate the
3442 : * actual installation location or last segment searched.
3443 : *
3444 : * tmppath: initial name of file to install. It will be renamed into place.
3445 : *
3446 : * find_free: if TRUE, install the new segment at the first empty segno
3447 : * number at or after the passed numbers. If FALSE, install the new segment
3448 : * exactly where specified, deleting any existing segment file there.
3449 : *
3450 : * max_segno: maximum segment number to install the new file as. Fail if no
3451 : * free slot is found between *segno and max_segno. (Ignored when find_free
3452 : * is FALSE.)
3453 : *
3454 : * use_lock: if TRUE, acquire ControlFileLock while moving file into
3455 : * place. This should be TRUE except during bootstrap log creation. The
3456 : * caller must *not* hold the lock at call.
3457 : *
3458 : * Returns TRUE if the file was installed successfully. FALSE indicates that
3459 : * max_segno limit was exceeded, or an error occurred while renaming the
3460 : * file into place.
3461 : */
3462 : static bool
3463 11 : InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
3464 : bool find_free, XLogSegNo max_segno,
3465 : bool use_lock)
3466 : {
3467 : char path[MAXPGPATH];
3468 : struct stat stat_buf;
3469 :
3470 11 : XLogFilePath(path, ThisTimeLineID, *segno);
3471 :
3472 : /*
3473 : * We want to be sure that only one process does this at a time.
3474 : */
3475 11 : if (use_lock)
3476 10 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
3477 :
3478 11 : if (!find_free)
3479 : {
3480 : /* Force installation: get rid of any pre-existing segment file */
3481 1 : durable_unlink(path, DEBUG1);
3482 : }
3483 : else
3484 : {
3485 : /* Find a free slot to put it in */
3486 20 : while (stat(path, &stat_buf) == 0)
3487 : {
3488 0 : if ((*segno) >= max_segno)
3489 : {
3490 : /* Failed to find a free slot within specified range */
3491 0 : if (use_lock)
3492 0 : LWLockRelease(ControlFileLock);
3493 0 : return false;
3494 : }
3495 0 : (*segno)++;
3496 0 : XLogFilePath(path, ThisTimeLineID, *segno);
3497 : }
3498 : }
3499 :
3500 : /*
3501 : * Perform the rename using link if available, paranoidly trying to avoid
3502 : * overwriting an existing file (there shouldn't be one).
3503 : */
3504 11 : if (durable_link_or_rename(tmppath, path, LOG) != 0)
3505 : {
3506 0 : if (use_lock)
3507 0 : LWLockRelease(ControlFileLock);
3508 : /* durable_link_or_rename already emitted log message */
3509 0 : return false;
3510 : }
3511 :
3512 11 : if (use_lock)
3513 10 : LWLockRelease(ControlFileLock);
3514 :
3515 11 : return true;
3516 : }
3517 :
3518 : /*
3519 : * Open a pre-existing logfile segment for writing.
3520 : */
3521 : int
3522 1 : XLogFileOpen(XLogSegNo segno)
3523 : {
3524 : char path[MAXPGPATH];
3525 : int fd;
3526 :
3527 1 : XLogFilePath(path, ThisTimeLineID, segno);
3528 :
3529 1 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | get_sync_bit(sync_method),
3530 : S_IRUSR | S_IWUSR);
3531 1 : if (fd < 0)
3532 0 : ereport(PANIC,
3533 : (errcode_for_file_access(),
3534 : errmsg("could not open write-ahead log file \"%s\": %m", path)));
3535 :
3536 1 : return fd;
3537 : }
3538 :
3539 : /*
3540 : * Open a logfile segment for reading (during recovery).
3541 : *
3542 : * If source == XLOG_FROM_ARCHIVE, the segment is retrieved from archive.
3543 : * Otherwise, it's assumed to be already available in pg_wal.
3544 : */
3545 : static int
3546 3 : XLogFileRead(XLogSegNo segno, int emode, TimeLineID tli,
3547 : int source, bool notfoundOk)
3548 : {
3549 : char xlogfname[MAXFNAMELEN];
3550 : char activitymsg[MAXFNAMELEN + 16];
3551 : char path[MAXPGPATH];
3552 : int fd;
3553 :
3554 3 : XLogFileName(xlogfname, tli, segno);
3555 :
3556 3 : switch (source)
3557 : {
3558 : case XLOG_FROM_ARCHIVE:
3559 : /* Report recovery progress in PS display */
3560 0 : snprintf(activitymsg, sizeof(activitymsg), "waiting for %s",
3561 : xlogfname);
3562 0 : set_ps_display(activitymsg, false);
3563 :
3564 0 : restoredFromArchive = RestoreArchivedFile(path, xlogfname,
3565 : "RECOVERYXLOG",
3566 : XLogSegSize,
3567 : InRedo);
3568 0 : if (!restoredFromArchive)
3569 0 : return -1;
3570 0 : break;
3571 :
3572 : case XLOG_FROM_PG_WAL:
3573 : case XLOG_FROM_STREAM:
3574 3 : XLogFilePath(path, tli, segno);
3575 3 : restoredFromArchive = false;
3576 3 : break;
3577 :
3578 : default:
3579 0 : elog(ERROR, "invalid XLogFileRead source %d", source);
3580 : }
3581 :
3582 : /*
3583 : * If the segment was fetched from archival storage, replace the existing
3584 : * xlog segment (if any) with the archival version.
3585 : */
3586 3 : if (source == XLOG_FROM_ARCHIVE)
3587 : {
3588 0 : KeepFileRestoredFromArchive(path, xlogfname);
3589 :
3590 : /*
3591 : * Set path to point at the new file in pg_wal.
3592 : */
3593 0 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlogfname);
3594 : }
3595 :
3596 3 : fd = BasicOpenFile(path, O_RDONLY | PG_BINARY, 0);
3597 3 : if (fd >= 0)
3598 : {
3599 : /* Success! */
3600 3 : curFileTLI = tli;
3601 :
3602 : /* Report recovery progress in PS display */
3603 3 : snprintf(activitymsg, sizeof(activitymsg), "recovering %s",
3604 : xlogfname);
3605 3 : set_ps_display(activitymsg, false);
3606 :
3607 : /* Track source of data in assorted state variables */
3608 3 : readSource = source;
3609 3 : XLogReceiptSource = source;
3610 : /* In FROM_STREAM case, caller tracks receipt time, not me */
3611 3 : if (source != XLOG_FROM_STREAM)
3612 3 : XLogReceiptTime = GetCurrentTimestamp();
3613 :
3614 3 : return fd;
3615 : }
3616 0 : if (errno != ENOENT || !notfoundOk) /* unexpected failure? */
3617 0 : ereport(PANIC,
3618 : (errcode_for_file_access(),
3619 : errmsg("could not open file \"%s\": %m", path)));
3620 0 : return -1;
3621 : }
3622 :
3623 : /*
3624 : * Open a logfile segment for reading (during recovery).
3625 : *
3626 : * This version searches for the segment with any TLI listed in expectedTLEs.
3627 : */
3628 : static int
3629 3 : XLogFileReadAnyTLI(XLogSegNo segno, int emode, int source)
3630 : {
3631 : char path[MAXPGPATH];
3632 : ListCell *cell;
3633 : int fd;
3634 : List *tles;
3635 :
3636 : /*
3637 : * Loop looking for a suitable timeline ID: we might need to read any of
3638 : * the timelines listed in expectedTLEs.
3639 : *
3640 : * We expect curFileTLI on entry to be the TLI of the preceding file in
3641 : * sequence, or 0 if there was no predecessor. We do not allow curFileTLI
3642 : * to go backwards; this prevents us from picking up the wrong file when a
3643 : * parent timeline extends to higher segment numbers than the child we
3644 : * want to read.
3645 : *
3646 : * If we haven't read the timeline history file yet, read it now, so that
3647 : * we know which TLIs to scan. We don't save the list in expectedTLEs,
3648 : * however, unless we actually find a valid segment. That way if there is
3649 : * neither a timeline history file nor a WAL segment in the archive, and
3650 : * streaming replication is set up, we'll read the timeline history file
3651 : * streamed from the master when we start streaming, instead of recovering
3652 : * with a dummy history generated here.
3653 : */
3654 3 : if (expectedTLEs)
3655 0 : tles = expectedTLEs;
3656 : else
3657 3 : tles = readTimeLineHistory(recoveryTargetTLI);
3658 :
3659 3 : foreach(cell, tles)
3660 : {
3661 3 : TimeLineID tli = ((TimeLineHistoryEntry *) lfirst(cell))->tli;
3662 :
3663 3 : if (tli < curFileTLI)
3664 0 : break; /* don't bother looking at too-old TLIs */
3665 :
3666 3 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_ARCHIVE)
3667 : {
3668 0 : fd = XLogFileRead(segno, emode, tli,
3669 : XLOG_FROM_ARCHIVE, true);
3670 0 : if (fd != -1)
3671 : {
3672 0 : elog(DEBUG1, "got WAL segment from archive");
3673 0 : if (!expectedTLEs)
3674 0 : expectedTLEs = tles;
3675 0 : return fd;
3676 : }
3677 : }
3678 :
3679 3 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_PG_WAL)
3680 : {
3681 3 : fd = XLogFileRead(segno, emode, tli,
3682 : XLOG_FROM_PG_WAL, true);
3683 3 : if (fd != -1)
3684 : {
3685 3 : if (!expectedTLEs)
3686 3 : expectedTLEs = tles;
3687 3 : return fd;
3688 : }
3689 : }
3690 : }
3691 :
3692 : /* Couldn't find it. For simplicity, complain about front timeline */
3693 0 : XLogFilePath(path, recoveryTargetTLI, segno);
3694 0 : errno = ENOENT;
3695 0 : ereport(emode,
3696 : (errcode_for_file_access(),
3697 : errmsg("could not open file \"%s\": %m", path)));
3698 0 : return -1;
3699 : }
3700 :
3701 : /*
3702 : * Close the current logfile segment for writing.
3703 : */
3704 : static void
3705 53 : XLogFileClose(void)
3706 : {
3707 53 : Assert(openLogFile >= 0);
3708 :
3709 : /*
3710 : * WAL segment files will not be re-read in normal operation, so we advise
3711 : * the OS to release any cached pages. But do not do so if WAL archiving
3712 : * or streaming is active, because archiver and walsender process could
3713 : * use the cache to read the WAL segment.
3714 : */
3715 : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
3716 53 : if (!XLogIsNeeded())
3717 0 : (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
3718 : #endif
3719 :
3720 53 : if (close(openLogFile))
3721 0 : ereport(PANIC,
3722 : (errcode_for_file_access(),
3723 : errmsg("could not close log file %s: %m",
3724 : XLogFileNameP(ThisTimeLineID, openLogSegNo))));
3725 53 : openLogFile = -1;
3726 53 : }
3727 :
3728 : /*
3729 : * Preallocate log files beyond the specified log endpoint.
3730 : *
3731 : * XXX this is currently extremely conservative, since it forces only one
3732 : * future log segment to exist, and even that only if we are 75% done with
3733 : * the current one. This is only appropriate for very low-WAL-volume systems.
3734 : * High-volume systems will be OK once they've built up a sufficient set of
3735 : * recycled log segments, but the startup transient is likely to include
3736 : * a lot of segment creations by foreground processes, which is not so good.
3737 : */
3738 : static void
3739 11 : PreallocXlogFiles(XLogRecPtr endptr)
3740 : {
3741 : XLogSegNo _logSegNo;
3742 : int lf;
3743 : bool use_existent;
3744 :
3745 11 : XLByteToPrevSeg(endptr, _logSegNo);
3746 11 : if ((endptr - 1) % XLogSegSize >= (uint32) (0.75 * XLogSegSize))
3747 : {
3748 0 : _logSegNo++;
3749 0 : use_existent = true;
3750 0 : lf = XLogFileInit(_logSegNo, &use_existent, true);
3751 0 : close(lf);
3752 0 : if (!use_existent)
3753 0 : CheckpointStats.ckpt_segs_added++;
3754 : }
3755 11 : }
3756 :
3757 : /*
3758 : * Throws an error if the given log segment has already been removed or
3759 : * recycled. The caller should only pass a segment that it knows to have
3760 : * existed while the server has been running, as this function always
3761 : * succeeds if no WAL segments have been removed since startup.
3762 : * 'tli' is only used in the error message.
3763 : */
3764 : void
3765 0 : CheckXLogRemoved(XLogSegNo segno, TimeLineID tli)
3766 : {
3767 : XLogSegNo lastRemovedSegNo;
3768 :
3769 0 : SpinLockAcquire(&XLogCtl->info_lck);
3770 0 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3771 0 : SpinLockRelease(&XLogCtl->info_lck);
3772 :
3773 0 : if (segno <= lastRemovedSegNo)
3774 : {
3775 : char filename[MAXFNAMELEN];
3776 :
3777 0 : XLogFileName(filename, tli, segno);
3778 0 : ereport(ERROR,
3779 : (errcode_for_file_access(),
3780 : errmsg("requested WAL segment %s has already been removed",
3781 : filename)));
3782 : }
3783 0 : }
3784 :
3785 : /*
3786 : * Return the last WAL segment removed, or 0 if no segment has been removed
3787 : * since startup.
3788 : *
3789 : * NB: the result can be out of date arbitrarily fast, the caller has to deal
3790 : * with that.
3791 : */
3792 : XLogSegNo
3793 0 : XLogGetLastRemovedSegno(void)
3794 : {
3795 : XLogSegNo lastRemovedSegNo;
3796 :
3797 0 : SpinLockAcquire(&XLogCtl->info_lck);
3798 0 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3799 0 : SpinLockRelease(&XLogCtl->info_lck);
3800 :
3801 0 : return lastRemovedSegNo;
3802 : }
3803 :
3804 : /*
3805 : * Update the last removed segno pointer in shared memory, to reflect
3806 : * that the given XLOG file has been removed.
3807 : */
3808 : static void
3809 0 : UpdateLastRemovedPtr(char *filename)
3810 : {
3811 : uint32 tli;
3812 : XLogSegNo segno;
3813 :
3814 0 : XLogFromFileName(filename, &tli, &segno);
3815 :
3816 0 : SpinLockAcquire(&XLogCtl->info_lck);
3817 0 : if (segno > XLogCtl->lastRemovedSegNo)
3818 0 : XLogCtl->lastRemovedSegNo = segno;
3819 0 : SpinLockRelease(&XLogCtl->info_lck);
3820 0 : }
3821 :
3822 : /*
3823 : * Recycle or remove all log files older or equal to passed segno.
3824 : *
3825 : * endptr is current (or recent) end of xlog, and PriorRedoRecPtr is the
3826 : * redo pointer of the previous checkpoint. These are used to determine
3827 : * whether we want to recycle rather than delete no-longer-wanted log files.
3828 : */
3829 : static void
3830 11 : RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr PriorRedoPtr, XLogRecPtr endptr)
3831 : {
3832 : DIR *xldir;
3833 : struct dirent *xlde;
3834 : char lastoff[MAXFNAMELEN];
3835 :
3836 11 : xldir = AllocateDir(XLOGDIR);
3837 11 : if (xldir == NULL)
3838 0 : ereport(ERROR,
3839 : (errcode_for_file_access(),
3840 : errmsg("could not open write-ahead log directory \"%s\": %m",
3841 : XLOGDIR)));
3842 :
3843 : /*
3844 : * Construct a filename of the last segment to be kept. The timeline ID
3845 : * doesn't matter, we ignore that in the comparison. (During recovery,
3846 : * ThisTimeLineID isn't set, so we can't use that.)
3847 : */
3848 11 : XLogFileName(lastoff, 0, segno);
3849 :
3850 11 : elog(DEBUG2, "attempting to remove WAL segments older than log file %s",
3851 : lastoff);
3852 :
3853 76 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3854 : {
3855 : /* Ignore files that are not XLOG segments */
3856 87 : if (!IsXLogFileName(xlde->d_name) &&
3857 33 : !IsPartialXLogFileName(xlde->d_name))
3858 33 : continue;
3859 :
3860 : /*
3861 : * We ignore the timeline part of the XLOG segment identifiers in
3862 : * deciding whether a segment is still needed. This ensures that we
3863 : * won't prematurely remove a segment from a parent timeline. We could
3864 : * probably be a little more proactive about removing segments of
3865 : * non-parent timelines, but that would be a whole lot more
3866 : * complicated.
3867 : *
3868 : * We use the alphanumeric sorting property of the filenames to decide
3869 : * which ones are earlier than the lastoff segment.
3870 : */
3871 21 : if (strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
3872 : {
3873 0 : if (XLogArchiveCheckDone(xlde->d_name))
3874 : {
3875 : /* Update the last removed location in shared memory first */
3876 0 : UpdateLastRemovedPtr(xlde->d_name);
3877 :
3878 0 : RemoveXlogFile(xlde->d_name, PriorRedoPtr, endptr);
3879 : }
3880 : }
3881 : }
3882 :
3883 11 : FreeDir(xldir);
3884 11 : }
3885 :
3886 : /*
3887 : * Remove WAL files that are not part of the given timeline's history.
3888 : *
3889 : * This is called during recovery, whenever we switch to follow a new
3890 : * timeline, and at the end of recovery when we create a new timeline. We
3891 : * wouldn't otherwise care about extra WAL files lying in pg_wal, but they
3892 : * might be leftover pre-allocated or recycled WAL segments on the old timeline
3893 : * that we haven't used yet, and contain garbage. If we just leave them in
3894 : * pg_wal, they will eventually be archived, and we can't let that happen.
3895 : * Files that belong to our timeline history are valid, because we have
3896 : * successfully replayed them, but from others we can't be sure.
3897 : *
3898 : * 'switchpoint' is the current point in WAL where we switch to new timeline,
3899 : * and 'newTLI' is the new timeline we switch to.
3900 : */
3901 : static void
3902 0 : RemoveNonParentXlogFiles(XLogRecPtr switchpoint, TimeLineID newTLI)
3903 : {
3904 : DIR *xldir;
3905 : struct dirent *xlde;
3906 : char switchseg[MAXFNAMELEN];
3907 : XLogSegNo endLogSegNo;
3908 :
3909 0 : XLByteToPrevSeg(switchpoint, endLogSegNo);
3910 :
3911 0 : xldir = AllocateDir(XLOGDIR);
3912 0 : if (xldir == NULL)
3913 0 : ereport(ERROR,
3914 : (errcode_for_file_access(),
3915 : errmsg("could not open write-ahead log directory \"%s\": %m",
3916 : XLOGDIR)));
3917 :
3918 : /*
3919 : * Construct a filename of the last segment to be kept.
3920 : */
3921 0 : XLogFileName(switchseg, newTLI, endLogSegNo);
3922 :
3923 0 : elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
3924 : switchseg);
3925 :
3926 0 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3927 : {
3928 : /* Ignore files that are not XLOG segments */
3929 0 : if (!IsXLogFileName(xlde->d_name))
3930 0 : continue;
3931 :
3932 : /*
3933 : * Remove files that are on a timeline older than the new one we're
3934 : * switching to, but with a segment number >= the first segment on the
3935 : * new timeline.
3936 : */
3937 0 : if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
3938 0 : strcmp(xlde->d_name + 8, switchseg + 8) > 0)
3939 : {
3940 : /*
3941 : * If the file has already been marked as .ready, however, don't
3942 : * remove it yet. It should be OK to remove it - files that are
3943 : * not part of our timeline history are not required for recovery
3944 : * - but seems safer to let them be archived and removed later.
3945 : */
3946 0 : if (!XLogArchiveIsReady(xlde->d_name))
3947 0 : RemoveXlogFile(xlde->d_name, InvalidXLogRecPtr, switchpoint);
3948 : }
3949 : }
3950 :
3951 0 : FreeDir(xldir);
3952 0 : }
3953 :
3954 : /*
3955 : * Recycle or remove a log file that's no longer needed.
3956 : *
3957 : * endptr is current (or recent) end of xlog, and PriorRedoRecPtr is the
3958 : * redo pointer of the previous checkpoint. These are used to determine
3959 : * whether we want to recycle rather than delete no-longer-wanted log files.
3960 : * If PriorRedoRecPtr is not known, pass invalid, and the function will
3961 : * recycle, somewhat arbitrarily, 10 future segments.
3962 : */
3963 : static void
3964 0 : RemoveXlogFile(const char *segname, XLogRecPtr PriorRedoPtr, XLogRecPtr endptr)
3965 : {
3966 : char path[MAXPGPATH];
3967 : #ifdef WIN32
3968 : char newpath[MAXPGPATH];
3969 : #endif
3970 : struct stat statbuf;
3971 : XLogSegNo endlogSegNo;
3972 : XLogSegNo recycleSegNo;
3973 :
3974 : /*
3975 : * Initialize info about where to try to recycle to.
3976 : */
3977 0 : XLByteToSeg(endptr, endlogSegNo);
3978 0 : if (PriorRedoPtr == InvalidXLogRecPtr)
3979 0 : recycleSegNo = endlogSegNo + 10;
3980 : else
3981 0 : recycleSegNo = XLOGfileslop(PriorRedoPtr);
3982 :
3983 0 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", segname);
3984 :
3985 : /*
3986 : * Before deleting the file, see if it can be recycled as a future log
3987 : * segment. Only recycle normal files, pg_standby for example can create
3988 : * symbolic links pointing to a separate archive directory.
3989 : */
3990 0 : if (endlogSegNo <= recycleSegNo &&
3991 0 : lstat(path, &statbuf) == 0 && S_ISREG(statbuf.st_mode) &&
3992 0 : InstallXLogFileSegment(&endlogSegNo, path,
3993 : true, recycleSegNo, true))
3994 : {
3995 0 : ereport(DEBUG2,
3996 : (errmsg("recycled write-ahead log file \"%s\"",
3997 : segname)));
3998 0 : CheckpointStats.ckpt_segs_recycled++;
3999 : /* Needn't recheck that slot on future iterations */
4000 0 : endlogSegNo++;
4001 : }
4002 : else
4003 : {
4004 : /* No need for any more future segments... */
4005 : int rc;
4006 :
4007 0 : ereport(DEBUG2,
4008 : (errmsg("removing write-ahead log file \"%s\"",
4009 : segname)));
4010 :
4011 : #ifdef WIN32
4012 :
4013 : /*
4014 : * On Windows, if another process (e.g another backend) holds the file
4015 : * open in FILE_SHARE_DELETE mode, unlink will succeed, but the file
4016 : * will still show up in directory listing until the last handle is
4017 : * closed. To avoid confusing the lingering deleted file for a live
4018 : * WAL file that needs to be archived, rename it before deleting it.
4019 : *
4020 : * If another process holds the file open without FILE_SHARE_DELETE
4021 : * flag, rename will fail. We'll try again at the next checkpoint.
4022 : */
4023 : snprintf(newpath, MAXPGPATH, "%s.deleted", path);
4024 : if (rename(path, newpath) != 0)
4025 : {
4026 : ereport(LOG,
4027 : (errcode_for_file_access(),
4028 : errmsg("could not rename old write-ahead log file \"%s\": %m",
4029 : path)));
4030 : return;
4031 : }
4032 : rc = durable_unlink(newpath, LOG);
4033 : #else
4034 0 : rc = durable_unlink(path, LOG);
4035 : #endif
4036 0 : if (rc != 0)
4037 : {
4038 : /* Message already logged by durable_unlink() */
4039 0 : return;
4040 : }
4041 0 : CheckpointStats.ckpt_segs_removed++;
4042 : }
4043 :
4044 0 : XLogArchiveCleanup(segname);
4045 : }
4046 :
4047 : /*
4048 : * Verify whether pg_wal and pg_wal/archive_status exist.
4049 : * If the latter does not exist, recreate it.
4050 : *
4051 : * It is not the goal of this function to verify the contents of these
4052 : * directories, but to help in cases where someone has performed a cluster
4053 : * copy for PITR purposes but omitted pg_wal from the copy.
4054 : *
4055 : * We could also recreate pg_wal if it doesn't exist, but a deliberate
4056 : * policy decision was made not to. It is fairly common for pg_wal to be
4057 : * a symlink, and if that was the DBA's intent then automatically making a
4058 : * plain directory would result in degraded performance with no notice.
4059 : */
4060 : static void
4061 3 : ValidateXLOGDirectoryStructure(void)
4062 : {
4063 : char path[MAXPGPATH];
4064 : struct stat stat_buf;
4065 :
4066 : /* Check for pg_wal; if it doesn't exist, error out */
4067 6 : if (stat(XLOGDIR, &stat_buf) != 0 ||
4068 3 : !S_ISDIR(stat_buf.st_mode))
4069 0 : ereport(FATAL,
4070 : (errmsg("required WAL directory \"%s\" does not exist",
4071 : XLOGDIR)));
4072 :
4073 : /* Check for archive_status */
4074 3 : snprintf(path, MAXPGPATH, XLOGDIR "/archive_status");
4075 3 : if (stat(path, &stat_buf) == 0)
4076 : {
4077 : /* Check for weird cases where it exists but isn't a directory */
4078 3 : if (!S_ISDIR(stat_buf.st_mode))
4079 0 : ereport(FATAL,
4080 : (errmsg("required WAL directory \"%s\" does not exist",
4081 : path)));
4082 : }
4083 : else
4084 : {
4085 0 : ereport(LOG,
4086 : (errmsg("creating missing WAL directory \"%s\"", path)));
4087 0 : if (mkdir(path, S_IRWXU) < 0)
4088 0 : ereport(FATAL,
4089 : (errmsg("could not create missing directory \"%s\": %m",
4090 : path)));
4091 : }
4092 3 : }
4093 :
4094 : /*
4095 : * Remove previous backup history files. This also retries creation of
4096 : * .ready files for any backup history files for which XLogArchiveNotify
4097 : * failed earlier.
4098 : */
4099 : static void
4100 0 : CleanupBackupHistory(void)
4101 : {
4102 : DIR *xldir;
4103 : struct dirent *xlde;
4104 : char path[MAXPGPATH + sizeof(XLOGDIR)];
4105 :
4106 0 : xldir = AllocateDir(XLOGDIR);
4107 0 : if (xldir == NULL)
4108 0 : ereport(ERROR,
4109 : (errcode_for_file_access(),
4110 : errmsg("could not open write-ahead log directory \"%s\": %m",
4111 : XLOGDIR)));
4112 :
4113 0 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4114 : {
4115 0 : if (IsBackupHistoryFileName(xlde->d_name))
4116 : {
4117 0 : if (XLogArchiveCheckDone(xlde->d_name))
4118 : {
4119 0 : elog(DEBUG2, "removing WAL backup history file \"%s\"",
4120 : xlde->d_name);
4121 0 : snprintf(path, sizeof(path), XLOGDIR "/%s", xlde->d_name);
4122 0 : unlink(path);
4123 0 : XLogArchiveCleanup(xlde->d_name);
4124 : }
4125 : }
4126 : }
4127 :
4128 0 : FreeDir(xldir);
4129 0 : }
4130 :
4131 : /*
4132 : * Attempt to read an XLOG record.
4133 : *
4134 : * If RecPtr is valid, try to read a record at that position. Otherwise
4135 : * try to read a record just after the last one previously read.
4136 : *
4137 : * If no valid record is available, returns NULL, or fails if emode is PANIC.
4138 : * (emode must be either PANIC, LOG). In standby mode, retries until a valid
4139 : * record is available.
4140 : *
4141 : * The record is copied into readRecordBuf, so that on successful return,
4142 : * the returned record pointer always points there.
4143 : */
4144 : static XLogRecord *
4145 6 : ReadRecord(XLogReaderState *xlogreader, XLogRecPtr RecPtr, int emode,
4146 : bool fetching_ckpt)
4147 : {
4148 : XLogRecord *record;
4149 6 : XLogPageReadPrivate *private = (XLogPageReadPrivate *) xlogreader->private_data;
4150 :
4151 : /* Pass through parameters to XLogPageRead */
4152 6 : private->fetching_ckpt = fetching_ckpt;
4153 6 : private->emode = emode;
4154 6 : private->randAccess = (RecPtr != InvalidXLogRecPtr);
4155 :
4156 : /* This is the first attempt to read this page. */
4157 6 : lastSourceFailed = false;
4158 :
4159 : for (;;)
4160 : {
4161 : char *errormsg;
4162 :
4163 6 : record = XLogReadRecord(xlogreader, RecPtr, &errormsg);
4164 6 : ReadRecPtr = xlogreader->ReadRecPtr;
4165 6 : EndRecPtr = xlogreader->EndRecPtr;
4166 6 : if (record == NULL)
4167 : {
4168 0 : if (readFile >= 0)
4169 : {
4170 0 : close(readFile);
4171 0 : readFile = -1;
4172 : }
4173 :
4174 : /*
4175 : * We only end up here without a message when XLogPageRead()
4176 : * failed - in that case we already logged something. In
4177 : * StandbyMode that only happens if we have been triggered, so we
4178 : * shouldn't loop anymore in that case.
4179 : */
4180 0 : if (errormsg)
4181 0 : ereport(emode_for_corrupt_record(emode,
4182 : RecPtr ? RecPtr : EndRecPtr),
4183 : (errmsg_internal("%s", errormsg) /* already translated */ ));
4184 : }
4185 :
4186 : /*
4187 : * Check page TLI is one of the expected values.
4188 : */
4189 6 : else if (!tliInHistory(xlogreader->latestPageTLI, expectedTLEs))
4190 : {
4191 : char fname[MAXFNAMELEN];
4192 : XLogSegNo segno;
4193 : int32 offset;
4194 :
4195 0 : XLByteToSeg(xlogreader->latestPagePtr, segno);
4196 0 : offset = xlogreader->latestPagePtr % XLogSegSize;
4197 0 : XLogFileName(fname, xlogreader->readPageTLI, segno);
4198 0 : ereport(emode_for_corrupt_record(emode,
4199 : RecPtr ? RecPtr : EndRecPtr),
4200 : (errmsg("unexpected timeline ID %u in log segment %s, offset %u",
4201 : xlogreader->latestPageTLI,
4202 : fname,
4203 : offset)));
4204 0 : record = NULL;
4205 : }
4206 :
4207 6 : if (record)
4208 : {
4209 : /* Great, got a record */
4210 12 : return record;
4211 : }
4212 : else
4213 : {
4214 : /* No valid record available from this source */
4215 0 : lastSourceFailed = true;
4216 :
4217 : /*
4218 : * If archive recovery was requested, but we were still doing
4219 : * crash recovery, switch to archive recovery and retry using the
4220 : * offline archive. We have now replayed all the valid WAL in
4221 : * pg_wal, so we are presumably now consistent.
4222 : *
4223 : * We require that there's at least some valid WAL present in
4224 : * pg_wal, however (!fetching_ckpt). We could recover using the
4225 : * WAL from the archive, even if pg_wal is completely empty, but
4226 : * we'd have no idea how far we'd have to replay to reach
4227 : * consistency. So err on the safe side and give up.
4228 : */
4229 0 : if (!InArchiveRecovery && ArchiveRecoveryRequested &&
4230 : !fetching_ckpt)
4231 : {
4232 0 : ereport(DEBUG1,
4233 : (errmsg_internal("reached end of WAL in pg_wal, entering archive recovery")));
4234 0 : InArchiveRecovery = true;
4235 0 : if (StandbyModeRequested)
4236 0 : StandbyMode = true;
4237 :
4238 : /* initialize minRecoveryPoint to this record */
4239 0 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
4240 0 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
4241 0 : if (ControlFile->minRecoveryPoint < EndRecPtr)
4242 : {
4243 0 : ControlFile->minRecoveryPoint = EndRecPtr;
4244 0 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
4245 : }
4246 : /* update local copy */
4247 0 : minRecoveryPoint = ControlFile->minRecoveryPoint;
4248 0 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
4249 :
4250 0 : UpdateControlFile();
4251 0 : LWLockRelease(ControlFileLock);
4252 :
4253 0 : CheckRecoveryConsistency();
4254 :
4255 : /*
4256 : * Before we retry, reset lastSourceFailed and currentSource
4257 : * so that we will check the archive next.
4258 : */
4259 0 : lastSourceFailed = false;
4260 0 : currentSource = 0;
4261 :
4262 0 : continue;
4263 : }
4264 :
4265 : /* In standby mode, loop back to retry. Otherwise, give up. */
4266 0 : if (StandbyMode && !CheckForStandbyTrigger())
4267 0 : continue;
4268 : else
4269 0 : return NULL;
4270 : }
4271 0 : }
4272 : }
4273 :
4274 : /*
4275 : * Scan for new timelines that might have appeared in the archive since we
4276 : * started recovery.
4277 : *
4278 : * If there are any, the function changes recovery target TLI to the latest
4279 : * one and returns 'true'.
4280 : */
4281 : static bool
4282 0 : rescanLatestTimeLine(void)
4283 : {
4284 : List *newExpectedTLEs;
4285 : bool found;
4286 : ListCell *cell;
4287 : TimeLineID newtarget;
4288 0 : TimeLineID oldtarget = recoveryTargetTLI;
4289 0 : TimeLineHistoryEntry *currentTle = NULL;
4290 :
4291 0 : newtarget = findNewestTimeLine(recoveryTargetTLI);
4292 0 : if (newtarget == recoveryTargetTLI)
4293 : {
4294 : /* No new timelines found */
4295 0 : return false;
4296 : }
4297 :
4298 : /*
4299 : * Determine the list of expected TLIs for the new TLI
4300 : */
4301 :
4302 0 : newExpectedTLEs = readTimeLineHistory(newtarget);
4303 :
4304 : /*
4305 : * If the current timeline is not part of the history of the new timeline,
4306 : * we cannot proceed to it.
4307 : */
4308 0 : found = false;
4309 0 : foreach(cell, newExpectedTLEs)
4310 : {
4311 0 : currentTle = (TimeLineHistoryEntry *) lfirst(cell);
4312 :
4313 0 : if (currentTle->tli == recoveryTargetTLI)
4314 : {
4315 0 : found = true;
4316 0 : break;
4317 : }
4318 : }
4319 0 : if (!found)
4320 : {
4321 0 : ereport(LOG,
4322 : (errmsg("new timeline %u is not a child of database system timeline %u",
4323 : newtarget,
4324 : ThisTimeLineID)));
4325 0 : return false;
4326 : }
4327 :
4328 : /*
4329 : * The current timeline was found in the history file, but check that the
4330 : * next timeline was forked off from it *after* the current recovery
4331 : * location.
4332 : */
4333 0 : if (currentTle->end < EndRecPtr)
4334 : {
4335 0 : ereport(LOG,
4336 : (errmsg("new timeline %u forked off current database system timeline %u before current recovery point %X/%X",
4337 : newtarget,
4338 : ThisTimeLineID,
4339 : (uint32) (EndRecPtr >> 32), (uint32) EndRecPtr)));
4340 0 : return false;
4341 : }
4342 :
4343 : /* The new timeline history seems valid. Switch target */
4344 0 : recoveryTargetTLI = newtarget;
4345 0 : list_free_deep(expectedTLEs);
4346 0 : expectedTLEs = newExpectedTLEs;
4347 :
4348 : /*
4349 : * As in StartupXLOG(), try to ensure we have all the history files
4350 : * between the old target and new target in pg_wal.
4351 : */
4352 0 : restoreTimeLineHistoryFiles(oldtarget + 1, newtarget);
4353 :
4354 0 : ereport(LOG,
4355 : (errmsg("new target timeline is %u",
4356 : recoveryTargetTLI)));
4357 :
4358 0 : return true;
4359 : }
4360 :
4361 : /*
4362 : * I/O routines for pg_control
4363 : *
4364 : * *ControlFile is a buffer in shared memory that holds an image of the
4365 : * contents of pg_control. WriteControlFile() initializes pg_control
4366 : * given a preloaded buffer, ReadControlFile() loads the buffer from
4367 : * the pg_control file (during postmaster or standalone-backend startup),
4368 : * and UpdateControlFile() rewrites pg_control after we modify xlog state.
4369 : *
4370 : * For simplicity, WriteControlFile() initializes the fields of pg_control
4371 : * that are related to checking backend/database compatibility, and
4372 : * ReadControlFile() verifies they are correct. We could split out the
4373 : * I/O and compatibility-check functions, but there seems no need currently.
4374 : */
4375 : static void
4376 1 : WriteControlFile(void)
4377 : {
4378 : int fd;
4379 : char buffer[PG_CONTROL_FILE_SIZE]; /* need not be aligned */
4380 :
4381 : /*
4382 : * Ensure that the size of the pg_control data structure is sane. See the
4383 : * comments for these symbols in pg_control.h.
4384 : */
4385 : StaticAssertStmt(sizeof(ControlFileData) <= PG_CONTROL_MAX_SAFE_SIZE,
4386 : "pg_control is too large for atomic disk writes");
4387 : StaticAssertStmt(sizeof(ControlFileData) <= PG_CONTROL_FILE_SIZE,
4388 : "sizeof(ControlFileData) exceeds PG_CONTROL_FILE_SIZE");
4389 :
4390 : /*
4391 : * Initialize version and compatibility-check fields
4392 : */
4393 1 : ControlFile->pg_control_version = PG_CONTROL_VERSION;
4394 1 : ControlFile->catalog_version_no = CATALOG_VERSION_NO;
4395 :
4396 1 : ControlFile->maxAlign = MAXIMUM_ALIGNOF;
4397 1 : ControlFile->floatFormat = FLOATFORMAT_VALUE;
4398 :
4399 1 : ControlFile->blcksz = BLCKSZ;
4400 1 : ControlFile->relseg_size = RELSEG_SIZE;
4401 1 : ControlFile->xlog_blcksz = XLOG_BLCKSZ;
4402 1 : ControlFile->xlog_seg_size = XLOG_SEG_SIZE;
4403 :
4404 1 : ControlFile->nameDataLen = NAMEDATALEN;
4405 1 : ControlFile->indexMaxKeys = INDEX_MAX_KEYS;
4406 :
4407 1 : ControlFile->toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
4408 1 : ControlFile->loblksize = LOBLKSIZE;
4409 :
4410 1 : ControlFile->float4ByVal = FLOAT4PASSBYVAL;
4411 1 : ControlFile->float8ByVal = FLOAT8PASSBYVAL;
4412 :
4413 : /* Contents are protected with a CRC */
4414 1 : INIT_CRC32C(ControlFile->crc);
4415 1 : COMP_CRC32C(ControlFile->crc,
4416 : (char *) ControlFile,
4417 : offsetof(ControlFileData, crc));
4418 1 : FIN_CRC32C(ControlFile->crc);
4419 :
4420 : /*
4421 : * We write out PG_CONTROL_FILE_SIZE bytes into pg_control, zero-padding
4422 : * the excess over sizeof(ControlFileData). This reduces the odds of
4423 : * premature-EOF errors when reading pg_control. We'll still fail when we
4424 : * check the contents of the file, but hopefully with a more specific
4425 : * error than "couldn't read pg_control".
4426 : */
4427 1 : memset(buffer, 0, PG_CONTROL_FILE_SIZE);
4428 1 : memcpy(buffer, ControlFile, sizeof(ControlFileData));
4429 :
4430 1 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4431 : O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
4432 : S_IRUSR | S_IWUSR);
4433 1 : if (fd < 0)
4434 0 : ereport(PANIC,
4435 : (errcode_for_file_access(),
4436 : errmsg("could not create control file \"%s\": %m",
4437 : XLOG_CONTROL_FILE)));
4438 :
4439 1 : errno = 0;
4440 1 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE);
4441 1 : if (write(fd, buffer, PG_CONTROL_FILE_SIZE) != PG_CONTROL_FILE_SIZE)
4442 : {
4443 : /* if write didn't set errno, assume problem is no disk space */
4444 0 : if (errno == 0)
4445 0 : errno = ENOSPC;
4446 0 : ereport(PANIC,
4447 : (errcode_for_file_access(),
4448 : errmsg("could not write to control file: %m")));
4449 : }
4450 1 : pgstat_report_wait_end();
4451 :
4452 1 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC);
4453 1 : if (pg_fsync(fd) != 0)
4454 0 : ereport(PANIC,
4455 : (errcode_for_file_access(),
4456 : errmsg("could not fsync control file: %m")));
4457 1 : pgstat_report_wait_end();
4458 :
4459 1 : if (close(fd))
4460 0 : ereport(PANIC,
4461 : (errcode_for_file_access(),
4462 : errmsg("could not close control file: %m")));
4463 1 : }
4464 :
4465 : static void
4466 5 : ReadControlFile(void)
4467 : {
4468 : pg_crc32c crc;
4469 : int fd;
4470 :
4471 : /*
4472 : * Read data...
4473 : */
4474 5 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4475 : O_RDWR | PG_BINARY,
4476 : S_IRUSR | S_IWUSR);
4477 5 : if (fd < 0)
4478 0 : ereport(PANIC,
4479 : (errcode_for_file_access(),
4480 : errmsg("could not open control file \"%s\": %m",
4481 : XLOG_CONTROL_FILE)));
4482 :
4483 5 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_READ);
4484 5 : if (read(fd, ControlFile, sizeof(ControlFileData)) != sizeof(ControlFileData))
4485 0 : ereport(PANIC,
4486 : (errcode_for_file_access(),
4487 : errmsg("could not read from control file: %m")));
4488 5 : pgstat_report_wait_end();
4489 :
4490 5 : close(fd);
4491 :
4492 : /*
4493 : * Check for expected pg_control format version. If this is wrong, the
4494 : * CRC check will likely fail because we'll be checking the wrong number
4495 : * of bytes. Complaining about wrong version will probably be more
4496 : * enlightening than complaining about wrong CRC.
4497 : */
4498 :
4499 5 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION && ControlFile->pg_control_version % 65536 == 0 && ControlFile->pg_control_version / 65536 != 0)
4500 0 : ereport(FATAL,
4501 : (errmsg("database files are incompatible with server"),
4502 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
4503 : " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
4504 : ControlFile->pg_control_version, ControlFile->pg_control_version,
4505 : PG_CONTROL_VERSION, PG_CONTROL_VERSION),
4506 : errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
4507 :
4508 5 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION)
4509 0 : ereport(FATAL,
4510 : (errmsg("database files are incompatible with server"),
4511 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
4512 : " but the server was compiled with PG_CONTROL_VERSION %d.",
4513 : ControlFile->pg_control_version, PG_CONTROL_VERSION),
4514 : errhint("It looks like you need to initdb.")));
4515 :
4516 : /* Now check the CRC. */
4517 5 : INIT_CRC32C(crc);
4518 5 : COMP_CRC32C(crc,
4519 : (char *) ControlFile,
4520 : offsetof(ControlFileData, crc));
4521 5 : FIN_CRC32C(crc);
4522 :
4523 5 : if (!EQ_CRC32C(crc, ControlFile->crc))
4524 0 : ereport(FATAL,
4525 : (errmsg("incorrect checksum in control file")));
4526 :
4527 : /*
4528 : * Do compatibility checking immediately. If the database isn't
4529 : * compatible with the backend executable, we want to abort before we can
4530 : * possibly do any damage.
4531 : */
4532 5 : if (ControlFile->catalog_version_no != CATALOG_VERSION_NO)
4533 0 : ereport(FATAL,
4534 : (errmsg("database files are incompatible with server"),
4535 : errdetail("The database cluster was initialized with CATALOG_VERSION_NO %d,"
4536 : " but the server was compiled with CATALOG_VERSION_NO %d.",
4537 : ControlFile->catalog_version_no, CATALOG_VERSION_NO),
4538 : errhint("It looks like you need to initdb.")));
4539 5 : if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
4540 0 : ereport(FATAL,
4541 : (errmsg("database files are incompatible with server"),
4542 : errdetail("The database cluster was initialized with MAXALIGN %d,"
4543 : " but the server was compiled with MAXALIGN %d.",
4544 : ControlFile->maxAlign, MAXIMUM_ALIGNOF),
4545 : errhint("It looks like you need to initdb.")));
4546 5 : if (ControlFile->floatFormat != FLOATFORMAT_VALUE)
4547 0 : ereport(FATAL,
4548 : (errmsg("database files are incompatible with server"),
4549 : errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
4550 : errhint("It looks like you need to initdb.")));
4551 5 : if (ControlFile->blcksz != BLCKSZ)
4552 0 : ereport(FATAL,
4553 : (errmsg("database files are incompatible with server"),
4554 : errdetail("The database cluster was initialized with BLCKSZ %d,"
4555 : " but the server was compiled with BLCKSZ %d.",
4556 : ControlFile->blcksz, BLCKSZ),
4557 : errhint("It looks like you need to recompile or initdb.")));
4558 5 : if (ControlFile->relseg_size != RELSEG_SIZE)
4559 0 : ereport(FATAL,
4560 : (errmsg("database files are incompatible with server"),
4561 : errdetail("The database cluster was initialized with RELSEG_SIZE %d,"
4562 : " but the server was compiled with RELSEG_SIZE %d.",
4563 : ControlFile->relseg_size, RELSEG_SIZE),
4564 : errhint("It looks like you need to recompile or initdb.")));
4565 5 : if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
4566 0 : ereport(FATAL,
4567 : (errmsg("database files are incompatible with server"),
4568 : errdetail("The database cluster was initialized with XLOG_BLCKSZ %d,"
4569 : " but the server was compiled with XLOG_BLCKSZ %d.",
4570 : ControlFile->xlog_blcksz, XLOG_BLCKSZ),
4571 : errhint("It looks like you need to recompile or initdb.")));
4572 5 : if (ControlFile->xlog_seg_size != XLOG_SEG_SIZE)
4573 0 : ereport(FATAL,
4574 : (errmsg("database files are incompatible with server"),
4575 : errdetail("The database cluster was initialized with XLOG_SEG_SIZE %d,"
4576 : " but the server was compiled with XLOG_SEG_SIZE %d.",
4577 : ControlFile->xlog_seg_size, XLOG_SEG_SIZE),
4578 : errhint("It looks like you need to recompile or initdb.")));
4579 5 : if (ControlFile->nameDataLen != NAMEDATALEN)
4580 0 : ereport(FATAL,
4581 : (errmsg("database files are incompatible with server"),
4582 : errdetail("The database cluster was initialized with NAMEDATALEN %d,"
4583 : " but the server was compiled with NAMEDATALEN %d.",
4584 : ControlFile->nameDataLen, NAMEDATALEN),
4585 : errhint("It looks like you need to recompile or initdb.")));
4586 5 : if (ControlFile->indexMaxKeys != INDEX_MAX_KEYS)
4587 0 : ereport(FATAL,
4588 : (errmsg("database files are incompatible with server"),
4589 : errdetail("The database cluster was initialized with INDEX_MAX_KEYS %d,"
4590 : " but the server was compiled with INDEX_MAX_KEYS %d.",
4591 : ControlFile->indexMaxKeys, INDEX_MAX_KEYS),
4592 : errhint("It looks like you need to recompile or initdb.")));
4593 5 : if (ControlFile->toast_max_chunk_size != TOAST_MAX_CHUNK_SIZE)
4594 0 : ereport(FATAL,
4595 : (errmsg("database files are incompatible with server"),
4596 : errdetail("The database cluster was initialized with TOAST_MAX_CHUNK_SIZE %d,"
4597 : " but the server was compiled with TOAST_MAX_CHUNK_SIZE %d.",
4598 : ControlFile->toast_max_chunk_size, (int) TOAST_MAX_CHUNK_SIZE),
4599 : errhint("It looks like you need to recompile or initdb.")));
4600 5 : if (ControlFile->loblksize != LOBLKSIZE)
4601 0 : ereport(FATAL,
4602 : (errmsg("database files are incompatible with server"),
4603 : errdetail("The database cluster was initialized with LOBLKSIZE %d,"
4604 : " but the server was compiled with LOBLKSIZE %d.",
4605 : ControlFile->loblksize, (int) LOBLKSIZE),
4606 : errhint("It looks like you need to recompile or initdb.")));
4607 :
4608 : #ifdef USE_FLOAT4_BYVAL
4609 5 : if (ControlFile->float4ByVal != true)
4610 0 : ereport(FATAL,
4611 : (errmsg("database files are incompatible with server"),
4612 : errdetail("The database cluster was initialized without USE_FLOAT4_BYVAL"
4613 : " but the server was compiled with USE_FLOAT4_BYVAL."),
4614 : errhint("It looks like you need to recompile or initdb.")));
4615 : #else
4616 : if (ControlFile->float4ByVal != false)
4617 : ereport(FATAL,
4618 : (errmsg("database files are incompatible with server"),
4619 : errdetail("The database cluster was initialized with USE_FLOAT4_BYVAL"
4620 : " but the server was compiled without USE_FLOAT4_BYVAL."),
4621 : errhint("It looks like you need to recompile or initdb.")));
4622 : #endif
4623 :
4624 : #ifdef USE_FLOAT8_BYVAL
4625 : if (ControlFile->float8ByVal != true)
4626 : ereport(FATAL,
4627 : (errmsg("database files are incompatible with server"),
4628 : errdetail("The database cluster was initialized without USE_FLOAT8_BYVAL"
4629 : " but the server was compiled with USE_FLOAT8_BYVAL."),
4630 : errhint("It looks like you need to recompile or initdb.")));
4631 : #else
4632 5 : if (ControlFile->float8ByVal != false)
4633 0 : ereport(FATAL,
4634 : (errmsg("database files are incompatible with server"),
4635 : errdetail("The database cluster was initialized with USE_FLOAT8_BYVAL"
4636 : " but the server was compiled without USE_FLOAT8_BYVAL."),
4637 : errhint("It looks like you need to recompile or initdb.")));
4638 : #endif
4639 :
4640 : /* Make the initdb settings visible as GUC variables, too */
4641 5 : SetConfigOption("data_checksums", DataChecksumsEnabled() ? "yes" : "no",
4642 : PGC_INTERNAL, PGC_S_OVERRIDE);
4643 5 : }
4644 :
4645 : void
4646 18 : UpdateControlFile(void)
4647 : {
4648 : int fd;
4649 :
4650 18 : INIT_CRC32C(ControlFile->crc);
4651 18 : COMP_CRC32C(ControlFile->crc,
4652 : (char *) ControlFile,
4653 : offsetof(ControlFileData, crc));
4654 18 : FIN_CRC32C(ControlFile->crc);
4655 :
4656 18 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4657 : O_RDWR | PG_BINARY,
4658 : S_IRUSR | S_IWUSR);
4659 18 : if (fd < 0)
4660 0 : ereport(PANIC,
4661 : (errcode_for_file_access(),
4662 : errmsg("could not open control file \"%s\": %m",
4663 : XLOG_CONTROL_FILE)));
4664 :
4665 18 : errno = 0;
4666 18 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE_UPDATE);
4667 18 : if (write(fd, ControlFile, sizeof(ControlFileData)) != sizeof(ControlFileData))
4668 : {
4669 : /* if write didn't set errno, assume problem is no disk space */
4670 0 : if (errno == 0)
4671 0 : errno = ENOSPC;
4672 0 : ereport(PANIC,
4673 : (errcode_for_file_access(),
4674 : errmsg("could not write to control file: %m")));
4675 : }
4676 18 : pgstat_report_wait_end();
4677 :
4678 18 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC_UPDATE);
4679 18 : if (pg_fsync(fd) != 0)
4680 0 : ereport(PANIC,
4681 : (errcode_for_file_access(),
4682 : errmsg("could not fsync control file: %m")));
4683 18 : pgstat_report_wait_end();
4684 :
4685 18 : if (close(fd))
4686 0 : ereport(PANIC,
4687 : (errcode_for_file_access(),
4688 : errmsg("could not close control file: %m")));
4689 18 : }
4690 :
4691 : /*
4692 : * Returns the unique system identifier from control file.
4693 : */
4694 : uint64
4695 0 : GetSystemIdentifier(void)
4696 : {
4697 0 : Assert(ControlFile != NULL);
4698 0 : return ControlFile->system_identifier;
4699 : }
4700 :
4701 : /*
4702 : * Returns the random nonce from control file.
4703 : */
4704 : char *
4705 0 : GetMockAuthenticationNonce(void)
4706 : {
4707 0 : Assert(ControlFile != NULL);
4708 0 : return ControlFile->mock_authentication_nonce;
4709 : }
4710 :
4711 : /*
4712 : * Are checksums enabled for data pages?
4713 : */
4714 : bool
4715 27816 : DataChecksumsEnabled(void)
4716 : {
4717 27816 : Assert(ControlFile != NULL);
4718 27816 : return (ControlFile->data_checksum_version > 0);
4719 : }
4720 :
4721 : /*
4722 : * Returns a fake LSN for unlogged relations.
4723 : *
4724 : * Each call generates an LSN that is greater than any previous value
4725 : * returned. The current counter value is saved and restored across clean
4726 : * shutdowns, but like unlogged relations, does not survive a crash. This can
4727 : * be used in lieu of real LSN values returned by XLogInsert, if you need an
4728 : * LSN-like increasing sequence of numbers without writing any WAL.
4729 : */
4730 : XLogRecPtr
4731 0 : GetFakeLSNForUnloggedRel(void)
4732 : {
4733 : XLogRecPtr nextUnloggedLSN;
4734 :
4735 : /* increment the unloggedLSN counter, need SpinLock */
4736 0 : SpinLockAcquire(&XLogCtl->ulsn_lck);
4737 0 : nextUnloggedLSN = XLogCtl->unloggedLSN++;
4738 0 : SpinLockRelease(&XLogCtl->ulsn_lck);
4739 :
4740 0 : return nextUnloggedLSN;
4741 : }
4742 :
4743 : /*
4744 : * Auto-tune the number of XLOG buffers.
4745 : *
4746 : * The preferred setting for wal_buffers is about 3% of shared_buffers, with
4747 : * a maximum of one XLOG segment (there is little reason to think that more
4748 : * is helpful, at least so long as we force an fsync when switching log files)
4749 : * and a minimum of 8 blocks (which was the default value prior to PostgreSQL
4750 : * 9.1, when auto-tuning was added).
4751 : *
4752 : * This should not be called until NBuffers has received its final value.
4753 : */
4754 : static int
4755 5 : XLOGChooseNumBuffers(void)
4756 : {
4757 : int xbuffers;
4758 :
4759 5 : xbuffers = NBuffers / 32;
4760 5 : if (xbuffers > XLOG_SEG_SIZE / XLOG_BLCKSZ)
4761 0 : xbuffers = XLOG_SEG_SIZE / XLOG_BLCKSZ;
4762 5 : if (xbuffers < 8)
4763 0 : xbuffers = 8;
4764 5 : return xbuffers;
4765 : }
4766 :
4767 : /*
4768 : * GUC check_hook for wal_buffers
4769 : */
4770 : bool
4771 10 : check_wal_buffers(int *newval, void **extra, GucSource source)
4772 : {
4773 : /*
4774 : * -1 indicates a request for auto-tune.
4775 : */
4776 10 : if (*newval == -1)
4777 : {
4778 : /*
4779 : * If we haven't yet changed the boot_val default of -1, just let it
4780 : * be. We'll fix it when XLOGShmemSize is called.
4781 : */
4782 5 : if (XLOGbuffers == -1)
4783 5 : return true;
4784 :
4785 : /* Otherwise, substitute the auto-tune value */
4786 0 : *newval = XLOGChooseNumBuffers();
4787 : }
4788 :
4789 : /*
4790 : * We clamp manually-set values to at least 4 blocks. Prior to PostgreSQL
4791 : * 9.1, a minimum of 4 was enforced by guc.c, but since that is no longer
4792 : * the case, we just silently treat such values as a request for the
4793 : * minimum. (We could throw an error instead, but that doesn't seem very
4794 : * helpful.)
4795 : */
4796 5 : if (*newval < 4)
4797 0 : *newval = 4;
4798 :
4799 5 : return true;
4800 : }
4801 :
4802 : /*
4803 : * Initialization of shared memory for XLOG
4804 : */
4805 : Size
4806 10 : XLOGShmemSize(void)
4807 : {
4808 : Size size;
4809 :
4810 : /*
4811 : * If the value of wal_buffers is -1, use the preferred auto-tune value.
4812 : * This isn't an amazingly clean place to do this, but we must wait till
4813 : * NBuffers has received its final value, and must do it before using the
4814 : * value of XLOGbuffers to do anything important.
4815 : */
4816 10 : if (XLOGbuffers == -1)
4817 : {
4818 : char buf[32];
4819 :
4820 5 : snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
4821 5 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER, PGC_S_OVERRIDE);
4822 : }
4823 10 : Assert(XLOGbuffers > 0);
4824 :
4825 : /* XLogCtl */
4826 10 : size = sizeof(XLogCtlData);
4827 :
4828 : /* WAL insertion locks, plus alignment */
4829 10 : size = add_size(size, mul_size(sizeof(WALInsertLockPadded), NUM_XLOGINSERT_LOCKS + 1));
4830 : /* xlblocks array */
4831 10 : size = add_size(size, mul_size(sizeof(XLogRecPtr), XLOGbuffers));
4832 : /* extra alignment padding for XLOG I/O buffers */
4833 10 : size = add_size(size, XLOG_BLCKSZ);
4834 : /* and the buffers themselves */
4835 10 : size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4836 :
4837 : /*
4838 : * Note: we don't count ControlFileData, it comes out of the "slop factor"
4839 : * added by CreateSharedMemoryAndSemaphores. This lets us use this
4840 : * routine again below to compute the actual allocation size.
4841 : */
4842 :
4843 10 : return size;
4844 : }
4845 :
4846 : void
4847 5 : XLOGShmemInit(void)
4848 : {
4849 : bool foundCFile,
4850 : foundXLog;
4851 : char *allocptr;
4852 : int i;
4853 :
4854 : #ifdef WAL_DEBUG
4855 :
4856 : /*
4857 : * Create a memory context for WAL debugging that's exempt from the normal
4858 : * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
4859 : * an allocation fails, but wal_debug is not for production use anyway.
4860 : */
4861 : if (walDebugCxt == NULL)
4862 : {
4863 : walDebugCxt = AllocSetContextCreate(TopMemoryContext,
4864 : "WAL Debug",
4865 : ALLOCSET_DEFAULT_SIZES);
4866 : MemoryContextAllowInCriticalSection(walDebugCxt, true);
4867 : }
4868 : #endif
4869 :
4870 5 : ControlFile = (ControlFileData *)
4871 5 : ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
4872 5 : XLogCtl = (XLogCtlData *)
4873 5 : ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
4874 :
4875 5 : if (foundCFile || foundXLog)
4876 : {
4877 : /* both should be present or neither */
4878 0 : Assert(foundCFile && foundXLog);
4879 :
4880 : /* Initialize local copy of WALInsertLocks and register the tranche */
4881 0 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks;
4882 0 : LWLockRegisterTranche(LWTRANCHE_WAL_INSERT,
4883 : "wal_insert");
4884 5 : return;
4885 : }
4886 5 : memset(XLogCtl, 0, sizeof(XLogCtlData));
4887 :
4888 : /*
4889 : * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
4890 : * multiple of the alignment for same, so no extra alignment padding is
4891 : * needed here.
4892 : */
4893 5 : allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
4894 5 : XLogCtl->xlblocks = (XLogRecPtr *) allocptr;
4895 5 : memset(XLogCtl->xlblocks, 0, sizeof(XLogRecPtr) * XLOGbuffers);
4896 5 : allocptr += sizeof(XLogRecPtr) * XLOGbuffers;
4897 :
4898 :
4899 : /* WAL insertion locks. Ensure they're aligned to the full padded size */
4900 5 : allocptr += sizeof(WALInsertLockPadded) -
4901 5 : ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
4902 5 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks =
4903 : (WALInsertLockPadded *) allocptr;
4904 5 : allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
4905 :
4906 5 : LWLockRegisterTranche(LWTRANCHE_WAL_INSERT, "wal_insert");
4907 45 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
4908 : {
4909 40 : LWLockInitialize(&WALInsertLocks[i].l.lock, LWTRANCHE_WAL_INSERT);
4910 40 : WALInsertLocks[i].l.insertingAt = InvalidXLogRecPtr;
4911 40 : WALInsertLocks[i].l.lastImportantAt = InvalidXLogRecPtr;
4912 : }
4913 :
4914 : /*
4915 : * Align the start of the page buffers to a full xlog block size boundary.
4916 : * This simplifies some calculations in XLOG insertion. It is also
4917 : * required for O_DIRECT.
4918 : */
4919 5 : allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
4920 5 : XLogCtl->pages = allocptr;
4921 5 : memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
4922 :
4923 : /*
4924 : * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
4925 : * in additional info.)
4926 : */
4927 5 : XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
4928 5 : XLogCtl->SharedRecoveryInProgress = true;
4929 5 : XLogCtl->SharedHotStandbyActive = false;
4930 5 : XLogCtl->WalWriterSleeping = false;
4931 :
4932 5 : SpinLockInit(&XLogCtl->Insert.insertpos_lck);
4933 5 : SpinLockInit(&XLogCtl->info_lck);
4934 5 : SpinLockInit(&XLogCtl->ulsn_lck);
4935 5 : InitSharedLatch(&XLogCtl->recoveryWakeupLatch);
4936 :
4937 : /*
4938 : * If we are not in bootstrap mode, pg_control should already exist. Read
4939 : * and validate it immediately (see comments in ReadControlFile() for the
4940 : * reasons why).
4941 : */
4942 5 : if (!IsBootstrapProcessingMode())
4943 2 : ReadControlFile();
4944 : }
4945 :
4946 : /*
4947 : * This func must be called ONCE on system install. It creates pg_control
4948 : * and the initial XLOG segment.
4949 : */
4950 : void
4951 1 : BootStrapXLOG(void)
4952 : {
4953 : CheckPoint checkPoint;
4954 : char *buffer;
4955 : XLogPageHeader page;
4956 : XLogLongPageHeader longpage;
4957 : XLogRecord *record;
4958 : char *recptr;
4959 : bool use_existent;
4960 : uint64 sysidentifier;
4961 : char mock_auth_nonce[MOCK_AUTH_NONCE_LEN];
4962 : struct timeval tv;
4963 : pg_crc32c crc;
4964 :
4965 : /*
4966 : * Select a hopefully-unique system identifier code for this installation.
4967 : * We use the result of gettimeofday(), including the fractional seconds
4968 : * field, as being about as unique as we can easily get. (Think not to
4969 : * use random(), since it hasn't been seeded and there's no portable way
4970 : * to seed it other than the system clock value...) The upper half of the
4971 : * uint64 value is just the tv_sec part, while the lower half contains the
4972 : * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
4973 : * PID for a little extra uniqueness. A person knowing this encoding can
4974 : * determine the initialization time of the installation, which could
4975 : * perhaps be useful sometimes.
4976 : */
4977 1 : gettimeofday(&tv, NULL);
4978 1 : sysidentifier = ((uint64) tv.tv_sec) << 32;
4979 1 : sysidentifier |= ((uint64) tv.tv_usec) << 12;
4980 1 : sysidentifier |= getpid() & 0xFFF;
4981 :
4982 : /*
4983 : * Generate a random nonce. This is used for authentication requests that
4984 : * will fail because the user does not exist. The nonce is used to create
4985 : * a genuine-looking password challenge for the non-existent user, in lieu
4986 : * of an actual stored password.
4987 : */
4988 1 : if (!pg_backend_random(mock_auth_nonce, MOCK_AUTH_NONCE_LEN))
4989 0 : ereport(PANIC,
4990 : (errcode(ERRCODE_INTERNAL_ERROR),
4991 : errmsg("could not generate secret authorization token")));
4992 :
4993 : /* First timeline ID is always 1 */
4994 1 : ThisTimeLineID = 1;
4995 :
4996 : /* page buffer must be aligned suitably for O_DIRECT */
4997 1 : buffer = (char *) palloc(XLOG_BLCKSZ + XLOG_BLCKSZ);
4998 1 : page = (XLogPageHeader) TYPEALIGN(XLOG_BLCKSZ, buffer);
4999 1 : memset(page, 0, XLOG_BLCKSZ);
5000 :
5001 : /*
5002 : * Set up information for the initial checkpoint record
5003 : *
5004 : * The initial checkpoint record is written to the beginning of the WAL
5005 : * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
5006 : * used, so that we can use 0/0 to mean "before any valid WAL segment".
5007 : */
5008 1 : checkPoint.redo = XLogSegSize + SizeOfXLogLongPHD;
5009 1 : checkPoint.ThisTimeLineID = ThisTimeLineID;
5010 1 : checkPoint.PrevTimeLineID = ThisTimeLineID;
5011 1 : checkPoint.fullPageWrites = fullPageWrites;
5012 1 : checkPoint.nextXidEpoch = 0;
5013 1 : checkPoint.nextXid = FirstNormalTransactionId;
5014 1 : checkPoint.nextOid = FirstBootstrapObjectId;
5015 1 : checkPoint.nextMulti = FirstMultiXactId;
5016 1 : checkPoint.nextMultiOffset = 0;
5017 1 : checkPoint.oldestXid = FirstNormalTransactionId;
5018 1 : checkPoint.oldestXidDB = TemplateDbOid;
5019 1 : checkPoint.oldestMulti = FirstMultiXactId;
5020 1 : checkPoint.oldestMultiDB = TemplateDbOid;
5021 1 : checkPoint.oldestCommitTsXid = InvalidTransactionId;
5022 1 : checkPoint.newestCommitTsXid = InvalidTransactionId;
5023 1 : checkPoint.time = (pg_time_t) time(NULL);
5024 1 : checkPoint.oldestActiveXid = InvalidTransactionId;
5025 :
5026 1 : ShmemVariableCache->nextXid = checkPoint.nextXid;
5027 1 : ShmemVariableCache->nextOid = checkPoint.nextOid;
5028 1 : ShmemVariableCache->oidCount = 0;
5029 1 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5030 1 : AdvanceOldestClogXid(checkPoint.oldestXid);
5031 1 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5032 1 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5033 1 : SetCommitTsLimit(InvalidTransactionId, InvalidTransactionId);
5034 :
5035 : /* Set up the XLOG page header */
5036 1 : page->xlp_magic = XLOG_PAGE_MAGIC;
5037 1 : page->xlp_info = XLP_LONG_HEADER;
5038 1 : page->xlp_tli = ThisTimeLineID;
5039 1 : page->xlp_pageaddr = XLogSegSize;
5040 1 : longpage = (XLogLongPageHeader) page;
5041 1 : longpage->xlp_sysid = sysidentifier;
5042 1 : longpage->xlp_seg_size = XLogSegSize;
5043 1 : longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5044 :
5045 : /* Insert the initial checkpoint record */
5046 1 : recptr = ((char *) page + SizeOfXLogLongPHD);
5047 1 : record = (XLogRecord *) recptr;
5048 1 : record->xl_prev = 0;
5049 1 : record->xl_xid = InvalidTransactionId;
5050 1 : record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
5051 1 : record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
5052 1 : record->xl_rmid = RM_XLOG_ID;
5053 1 : recptr += SizeOfXLogRecord;
5054 : /* fill the XLogRecordDataHeaderShort struct */
5055 1 : *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
5056 1 : *(recptr++) = sizeof(checkPoint);
5057 1 : memcpy(recptr, &checkPoint, sizeof(checkPoint));
5058 1 : recptr += sizeof(checkPoint);
5059 1 : Assert(recptr - (char *) record == record->xl_tot_len);
5060 :
5061 1 : INIT_CRC32C(crc);
5062 1 : COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
5063 1 : COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5064 1 : FIN_CRC32C(crc);
5065 1 : record->xl_crc = crc;
5066 :
5067 : /* Create first XLOG segment file */
5068 1 : use_existent = false;
5069 1 : openLogFile = XLogFileInit(1, &use_existent, false);
5070 :
5071 : /* Write the first page with the initial record */
5072 1 : errno = 0;
5073 1 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
5074 1 : if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5075 : {
5076 : /* if write didn't set errno, assume problem is no disk space */
5077 0 : if (errno == 0)
5078 0 : errno = ENOSPC;
5079 0 : ereport(PANIC,
5080 : (errcode_for_file_access(),
5081 : errmsg("could not write bootstrap write-ahead log file: %m")));
5082 : }
5083 1 : pgstat_report_wait_end();
5084 :
5085 1 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
5086 1 : if (pg_fsync(openLogFile) != 0)
5087 0 : ereport(PANIC,
5088 : (errcode_for_file_access(),
5089 : errmsg("could not fsync bootstrap write-ahead log file: %m")));
5090 1 : pgstat_report_wait_end();
5091 :
5092 1 : if (close(openLogFile))
5093 0 : ereport(PANIC,
5094 : (errcode_for_file_access(),
5095 : errmsg("could not close bootstrap write-ahead log file: %m")));
5096 :
5097 1 : openLogFile = -1;
5098 :
5099 : /* Now create pg_control */
5100 :
5101 1 : memset(ControlFile, 0, sizeof(ControlFileData));
5102 : /* Initialize pg_control status fields */
5103 1 : ControlFile->system_identifier = sysidentifier;
5104 1 : memcpy(ControlFile->mock_authentication_nonce, mock_auth_nonce, MOCK_AUTH_NONCE_LEN);
5105 1 : ControlFile->state = DB_SHUTDOWNED;
5106 1 : ControlFile->time = checkPoint.time;
5107 1 : ControlFile->checkPoint = checkPoint.redo;
5108 1 : ControlFile->checkPointCopy = checkPoint;
5109 1 : ControlFile->unloggedLSN = 1;
5110 :
5111 : /* Set important parameter values for use when replaying WAL */
5112 1 : ControlFile->MaxConnections = MaxConnections;
5113 1 : ControlFile->max_worker_processes = max_worker_processes;
5114 1 : ControlFile->max_prepared_xacts = max_prepared_xacts;
5115 1 : ControlFile->max_locks_per_xact = max_locks_per_xact;
5116 1 : ControlFile->wal_level = wal_level;
5117 1 : ControlFile->wal_log_hints = wal_log_hints;
5118 1 : ControlFile->track_commit_timestamp = track_commit_timestamp;
5119 1 : ControlFile->data_checksum_version = bootstrap_data_checksum_version;
5120 :
5121 : /* some additional ControlFile fields are set in WriteControlFile() */
5122 :
5123 1 : WriteControlFile();
5124 :
5125 : /* Bootstrap the commit log, too */
5126 1 : BootStrapCLOG();
5127 1 : BootStrapCommitTs();
5128 1 : BootStrapSUBTRANS();
5129 1 : BootStrapMultiXact();
5130 :
5131 1 : pfree(buffer);
5132 1 : }
5133 :
5134 : static char *
5135 1 : str_time(pg_time_t tnow)
5136 : {
5137 : static char buf[128];
5138 :
5139 1 : pg_strftime(buf, sizeof(buf),
5140 : "%Y-%m-%d %H:%M:%S %Z",
5141 1 : pg_localtime(&tnow, log_timezone));
5142 :
5143 1 : return buf;
5144 : }
5145 :
5146 : /*
5147 : * See if there is a recovery command file (recovery.conf), and if so
5148 : * read in parameters for archive recovery and XLOG streaming.
5149 : *
5150 : * The file is parsed using the main configuration parser.
5151 : */
5152 : static void
5153 3 : readRecoveryCommandFile(void)
5154 : {
5155 : FILE *fd;
5156 3 : TimeLineID rtli = 0;
5157 3 : bool rtliGiven = false;
5158 : ConfigVariable *item,
5159 3 : *head = NULL,
5160 3 : *tail = NULL;
5161 3 : bool recoveryTargetActionSet = false;
5162 :
5163 :
5164 3 : fd = AllocateFile(RECOVERY_COMMAND_FILE, "r");
5165 3 : if (fd == NULL)
5166 : {
5167 3 : if (errno == ENOENT)
5168 6 : return; /* not there, so no archive recovery */
5169 0 : ereport(FATAL,
5170 : (errcode_for_file_access(),
5171 : errmsg("could not open recovery command file \"%s\": %m",
5172 : RECOVERY_COMMAND_FILE)));
5173 : }
5174 :
5175 : /*
5176 : * Since we're asking ParseConfigFp() to report errors as FATAL, there's
5177 : * no need to check the return value.
5178 : */
5179 0 : (void) ParseConfigFp(fd, RECOVERY_COMMAND_FILE, 0, FATAL, &head, &tail);
5180 :
5181 0 : FreeFile(fd);
5182 :
5183 0 : for (item = head; item; item = item->next)
5184 : {
5185 0 : if (strcmp(item->name, "restore_command") == 0)
5186 : {
5187 0 : recoveryRestoreCommand = pstrdup(item->value);
5188 0 : ereport(DEBUG2,
5189 : (errmsg_internal("restore_command = '%s'",
5190 : recoveryRestoreCommand)));
5191 : }
5192 0 : else if (strcmp(item->name, "recovery_end_command") == 0)
5193 : {
5194 0 : recoveryEndCommand = pstrdup(item->value);
5195 0 : ereport(DEBUG2,
5196 : (errmsg_internal("recovery_end_command = '%s'",
5197 : recoveryEndCommand)));
5198 : }
5199 0 : else if (strcmp(item->name, "archive_cleanup_command") == 0)
5200 : {
5201 0 : archiveCleanupCommand = pstrdup(item->value);
5202 0 : ereport(DEBUG2,
5203 : (errmsg_internal("archive_cleanup_command = '%s'",
5204 : archiveCleanupCommand)));
5205 : }
5206 0 : else if (strcmp(item->name, "recovery_target_action") == 0)
5207 : {
5208 0 : if (strcmp(item->value, "pause") == 0)
5209 0 : recoveryTargetAction = RECOVERY_TARGET_ACTION_PAUSE;
5210 0 : else if (strcmp(item->value, "promote") == 0)
5211 0 : recoveryTargetAction = RECOVERY_TARGET_ACTION_PROMOTE;
5212 0 : else if (strcmp(item->value, "shutdown") == 0)
5213 0 : recoveryTargetAction = RECOVERY_TARGET_ACTION_SHUTDOWN;
5214 : else
5215 0 : ereport(ERROR,
5216 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5217 : errmsg("invalid value for recovery parameter \"%s\": \"%s\"",
5218 : "recovery_target_action",
5219 : item->value),
5220 : errhint("Valid values are \"pause\", \"promote\", and \"shutdown\".")));
5221 :
5222 0 : ereport(DEBUG2,
5223 : (errmsg_internal("recovery_target_action = '%s'",
5224 : item->value)));
5225 :
5226 0 : recoveryTargetActionSet = true;
5227 : }
5228 0 : else if (strcmp(item->name, "recovery_target_timeline") == 0)
5229 : {
5230 0 : rtliGiven = true;
5231 0 : if (strcmp(item->value, "latest") == 0)
5232 0 : rtli = 0;
5233 : else
5234 : {
5235 0 : errno = 0;
5236 0 : rtli = (TimeLineID) strtoul(item->value, NULL, 0);
5237 0 : if (errno == EINVAL || errno == ERANGE)
5238 0 : ereport(FATAL,
5239 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5240 : errmsg("recovery_target_timeline is not a valid number: \"%s\"",
5241 : item->value)));
5242 : }
5243 0 : if (rtli)
5244 0 : ereport(DEBUG2,
5245 : (errmsg_internal("recovery_target_timeline = %u", rtli)));
5246 : else
5247 0 : ereport(DEBUG2,
5248 : (errmsg_internal("recovery_target_timeline = latest")));
5249 : }
5250 0 : else if (strcmp(item->name, "recovery_target_xid") == 0)
5251 : {
5252 0 : errno = 0;
5253 0 : recoveryTargetXid = (TransactionId) strtoul(item->value, NULL, 0);
5254 0 : if (errno == EINVAL || errno == ERANGE)
5255 0 : ereport(FATAL,
5256 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5257 : errmsg("recovery_target_xid is not a valid number: \"%s\"",
5258 : item->value)));
5259 0 : ereport(DEBUG2,
5260 : (errmsg_internal("recovery_target_xid = %u",
5261 : recoveryTargetXid)));
5262 0 : recoveryTarget = RECOVERY_TARGET_XID;
5263 : }
5264 0 : else if (strcmp(item->name, "recovery_target_time") == 0)
5265 : {
5266 0 : recoveryTarget = RECOVERY_TARGET_TIME;
5267 :
5268 : /*
5269 : * Convert the time string given by the user to TimestampTz form.
5270 : */
5271 0 : recoveryTargetTime =
5272 0 : DatumGetTimestampTz(DirectFunctionCall3(timestamptz_in,
5273 : CStringGetDatum(item->value),
5274 : ObjectIdGetDatum(InvalidOid),
5275 : Int32GetDatum(-1)));
5276 0 : ereport(DEBUG2,
5277 : (errmsg_internal("recovery_target_time = '%s'",
5278 : timestamptz_to_str(recoveryTargetTime))));
5279 : }
5280 0 : else if (strcmp(item->name, "recovery_target_name") == 0)
5281 : {
5282 0 : recoveryTarget = RECOVERY_TARGET_NAME;
5283 :
5284 0 : recoveryTargetName = pstrdup(item->value);
5285 0 : if (strlen(recoveryTargetName) >= MAXFNAMELEN)
5286 0 : ereport(FATAL,
5287 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5288 : errmsg("recovery_target_name is too long (maximum %d characters)",
5289 : MAXFNAMELEN - 1)));
5290 :
5291 0 : ereport(DEBUG2,
5292 : (errmsg_internal("recovery_target_name = '%s'",
5293 : recoveryTargetName)));
5294 : }
5295 0 : else if (strcmp(item->name, "recovery_target_lsn") == 0)
5296 : {
5297 0 : recoveryTarget = RECOVERY_TARGET_LSN;
5298 :
5299 : /*
5300 : * Convert the LSN string given by the user to XLogRecPtr form.
5301 : */
5302 0 : recoveryTargetLSN =
5303 0 : DatumGetLSN(DirectFunctionCall3(pg_lsn_in,
5304 : CStringGetDatum(item->value),
5305 : ObjectIdGetDatum(InvalidOid),
5306 : Int32GetDatum(-1)));
5307 0 : ereport(DEBUG2,
5308 : (errmsg_internal("recovery_target_lsn = '%X/%X'",
5309 : (uint32) (recoveryTargetLSN >> 32),
5310 : (uint32) recoveryTargetLSN)));
5311 : }
5312 0 : else if (strcmp(item->name, "recovery_target") == 0)
5313 : {
5314 0 : if (strcmp(item->value, "immediate") == 0)
5315 0 : recoveryTarget = RECOVERY_TARGET_IMMEDIATE;
5316 : else
5317 0 : ereport(ERROR,
5318 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5319 : errmsg("invalid value for recovery parameter \"%s\": \"%s\"",
5320 : "recovery_target",
5321 : item->value),
5322 : errhint("The only allowed value is \"immediate\".")));
5323 0 : ereport(DEBUG2,
5324 : (errmsg_internal("recovery_target = '%s'",
5325 : item->value)));
5326 : }
5327 0 : else if (strcmp(item->name, "recovery_target_inclusive") == 0)
5328 : {
5329 : /*
5330 : * does nothing if a recovery_target is not also set
5331 : */
5332 0 : if (!parse_bool(item->value, &recoveryTargetInclusive))
5333 0 : ereport(ERROR,
5334 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5335 : errmsg("parameter \"%s\" requires a Boolean value",
5336 : "recovery_target_inclusive")));
5337 0 : ereport(DEBUG2,
5338 : (errmsg_internal("recovery_target_inclusive = %s",
5339 : item->value)));
5340 : }
5341 0 : else if (strcmp(item->name, "standby_mode") == 0)
5342 : {
5343 0 : if (!parse_bool(item->value, &StandbyModeRequested))
5344 0 : ereport(ERROR,
5345 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5346 : errmsg("parameter \"%s\" requires a Boolean value",
5347 : "standby_mode")));
5348 0 : ereport(DEBUG2,
5349 : (errmsg_internal("standby_mode = '%s'", item->value)));
5350 : }
5351 0 : else if (strcmp(item->name, "primary_conninfo") == 0)
5352 : {
5353 0 : PrimaryConnInfo = pstrdup(item->value);
5354 0 : ereport(DEBUG2,
5355 : (errmsg_internal("primary_conninfo = '%s'",
5356 : PrimaryConnInfo)));
5357 : }
5358 0 : else if (strcmp(item->name, "primary_slot_name") == 0)
5359 : {
5360 0 : ReplicationSlotValidateName(item->value, ERROR);
5361 0 : PrimarySlotName = pstrdup(item->value);
5362 0 : ereport(DEBUG2,
5363 : (errmsg_internal("primary_slot_name = '%s'",
5364 : PrimarySlotName)));
5365 : }
5366 0 : else if (strcmp(item->name, "trigger_file") == 0)
5367 : {
5368 0 : TriggerFile = pstrdup(item->value);
5369 0 : ereport(DEBUG2,
5370 : (errmsg_internal("trigger_file = '%s'",
5371 : TriggerFile)));
5372 : }
5373 0 : else if (strcmp(item->name, "recovery_min_apply_delay") == 0)
5374 : {
5375 : const char *hintmsg;
5376 :
5377 0 : if (!parse_int(item->value, &recovery_min_apply_delay, GUC_UNIT_MS,
5378 : &hintmsg))
5379 0 : ereport(ERROR,
5380 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5381 : errmsg("parameter \"%s\" requires a temporal value",
5382 : "recovery_min_apply_delay"),
5383 : hintmsg ? errhint("%s", _(hintmsg)) : 0));
5384 0 : ereport(DEBUG2,
5385 : (errmsg_internal("recovery_min_apply_delay = '%s'", item->value)));
5386 : }
5387 : else
5388 0 : ereport(FATAL,
5389 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5390 : errmsg("unrecognized recovery parameter \"%s\"",
5391 : item->name)));
5392 : }
5393 :
5394 : /*
5395 : * Check for compulsory parameters
5396 : */
5397 0 : if (StandbyModeRequested)
5398 : {
5399 0 : if (PrimaryConnInfo == NULL && recoveryRestoreCommand == NULL)
5400 0 : ereport(WARNING,
5401 : (errmsg("recovery command file \"%s\" specified neither primary_conninfo nor restore_command",
5402 : RECOVERY_COMMAND_FILE),
5403 : errhint("The database server will regularly poll the pg_wal subdirectory to check for files placed there.")));
5404 : }
5405 : else
5406 : {
5407 0 : if (recoveryRestoreCommand == NULL)
5408 0 : ereport(FATAL,
5409 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5410 : errmsg("recovery command file \"%s\" must specify restore_command when standby mode is not enabled",
5411 : RECOVERY_COMMAND_FILE)));
5412 : }
5413 :
5414 : /*
5415 : * Override any inconsistent requests. Not that this is a change of
5416 : * behaviour in 9.5; prior to this we simply ignored a request to pause if
5417 : * hot_standby = off, which was surprising behaviour.
5418 : */
5419 0 : if (recoveryTargetAction == RECOVERY_TARGET_ACTION_PAUSE &&
5420 0 : recoveryTargetActionSet &&
5421 0 : !EnableHotStandby)
5422 0 : recoveryTargetAction = RECOVERY_TARGET_ACTION_SHUTDOWN;
5423 :
5424 : /*
5425 : * We don't support standby_mode in standalone backends; that requires
5426 : * other processes such as the WAL receiver to be alive.
5427 : */
5428 0 : if (StandbyModeRequested && !IsUnderPostmaster)
5429 0 : ereport(FATAL,
5430 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
5431 : errmsg("standby mode is not supported by single-user servers")));
5432 :
5433 : /* Enable fetching from archive recovery area */
5434 0 : ArchiveRecoveryRequested = true;
5435 :
5436 : /*
5437 : * If user specified recovery_target_timeline, validate it or compute the
5438 : * "latest" value. We can't do this until after we've gotten the restore
5439 : * command and set InArchiveRecovery, because we need to fetch timeline
5440 : * history files from the archive.
5441 : */
5442 0 : if (rtliGiven)
5443 : {
5444 0 : if (rtli)
5445 : {
5446 : /* Timeline 1 does not have a history file, all else should */
5447 0 : if (rtli != 1 && !existsTimeLineHistory(rtli))
5448 0 : ereport(FATAL,
5449 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5450 : errmsg("recovery target timeline %u does not exist",
5451 : rtli)));
5452 0 : recoveryTargetTLI = rtli;
5453 0 : recoveryTargetIsLatest = false;
5454 : }
5455 : else
5456 : {
5457 : /* We start the "latest" search from pg_control's timeline */
5458 0 : recoveryTargetTLI = findNewestTimeLine(recoveryTargetTLI);
5459 0 : recoveryTargetIsLatest = true;
5460 : }
5461 : }
5462 :
5463 0 : FreeConfigVariables(head);
5464 : }
5465 :
5466 : /*
5467 : * Exit archive-recovery state
5468 : */
5469 : static void
5470 0 : exitArchiveRecovery(TimeLineID endTLI, XLogRecPtr endOfLog)
5471 : {
5472 : char recoveryPath[MAXPGPATH];
5473 : char xlogfname[MAXFNAMELEN];
5474 : XLogSegNo endLogSegNo;
5475 : XLogSegNo startLogSegNo;
5476 :
5477 : /* we always switch to a new timeline after archive recovery */
5478 0 : Assert(endTLI != ThisTimeLineID);
5479 :
5480 : /*
5481 : * We are no longer in archive recovery state.
5482 : */
5483 0 : InArchiveRecovery = false;
5484 :
5485 : /*
5486 : * Update min recovery point one last time.
5487 : */
5488 0 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5489 :
5490 : /*
5491 : * If the ending log segment is still open, close it (to avoid problems on
5492 : * Windows with trying to rename or delete an open file).
5493 : */
5494 0 : if (readFile >= 0)
5495 : {
5496 0 : close(readFile);
5497 0 : readFile = -1;
5498 : }
5499 :
5500 : /*
5501 : * Calculate the last segment on the old timeline, and the first segment
5502 : * on the new timeline. If the switch happens in the middle of a segment,
5503 : * they are the same, but if the switch happens exactly at a segment
5504 : * boundary, startLogSegNo will be endLogSegNo + 1.
5505 : */
5506 0 : XLByteToPrevSeg(endOfLog, endLogSegNo);
5507 0 : XLByteToSeg(endOfLog, startLogSegNo);
5508 :
5509 : /*
5510 : * Initialize the starting WAL segment for the new timeline. If the switch
5511 : * happens in the middle of a segment, copy data from the last WAL segment
5512 : * of the old timeline up to the switch point, to the starting WAL segment
5513 : * on the new timeline.
5514 : */
5515 0 : if (endLogSegNo == startLogSegNo)
5516 : {
5517 : /*
5518 : * Make a copy of the file on the new timeline.
5519 : *
5520 : * Writing WAL isn't allowed yet, so there are no locking
5521 : * considerations. But we should be just as tense as XLogFileInit to
5522 : * avoid emplacing a bogus file.
5523 : */
5524 0 : XLogFileCopy(endLogSegNo, endTLI, endLogSegNo,
5525 : endOfLog % XLOG_SEG_SIZE);
5526 : }
5527 : else
5528 : {
5529 : /*
5530 : * The switch happened at a segment boundary, so just create the next
5531 : * segment on the new timeline.
5532 : */
5533 0 : bool use_existent = true;
5534 : int fd;
5535 :
5536 0 : fd = XLogFileInit(startLogSegNo, &use_existent, true);
5537 :
5538 0 : if (close(fd))
5539 0 : ereport(ERROR,
5540 : (errcode_for_file_access(),
5541 : errmsg("could not close log file %s: %m",
5542 : XLogFileNameP(ThisTimeLineID, startLogSegNo))));
5543 : }
5544 :
5545 : /*
5546 : * Let's just make real sure there are not .ready or .done flags posted
5547 : * for the new segment.
5548 : */
5549 0 : XLogFileName(xlogfname, ThisTimeLineID, startLogSegNo);
5550 0 : XLogArchiveCleanup(xlogfname);
5551 :
5552 : /*
5553 : * Since there might be a partial WAL segment named RECOVERYXLOG, get rid
5554 : * of it.
5555 : */
5556 0 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYXLOG");
5557 0 : unlink(recoveryPath); /* ignore any error */
5558 :
5559 : /* Get rid of any remaining recovered timeline-history file, too */
5560 0 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYHISTORY");
5561 0 : unlink(recoveryPath); /* ignore any error */
5562 :
5563 : /*
5564 : * Rename the config file out of the way, so that we don't accidentally
5565 : * re-enter archive recovery mode in a subsequent crash.
5566 : */
5567 0 : unlink(RECOVERY_COMMAND_DONE);
5568 0 : durable_rename(RECOVERY_COMMAND_FILE, RECOVERY_COMMAND_DONE, FATAL);
5569 :
5570 0 : ereport(LOG,
5571 : (errmsg("archive recovery complete")));
5572 0 : }
5573 :
5574 : /*
5575 : * Extract timestamp from WAL record.
5576 : *
5577 : * If the record contains a timestamp, returns true, and saves the timestamp
5578 : * in *recordXtime. If the record type has no timestamp, returns false.
5579 : * Currently, only transaction commit/abort records and restore points contain
5580 : * timestamps.
5581 : */
5582 : static bool
5583 0 : getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime)
5584 : {
5585 0 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
5586 0 : uint8 xact_info = info & XLOG_XACT_OPMASK;
5587 0 : uint8 rmid = XLogRecGetRmid(record);
5588 :
5589 0 : if (rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
5590 : {
5591 0 : *recordXtime = ((xl_restore_point *) XLogRecGetData(record))->rp_time;
5592 0 : return true;
5593 : }
5594 0 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_COMMIT ||
5595 : xact_info == XLOG_XACT_COMMIT_PREPARED))
5596 : {
5597 0 : *recordXtime = ((xl_xact_commit *) XLogRecGetData(record))->xact_time;
5598 0 : return true;
5599 : }
5600 0 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_ABORT ||
5601 : xact_info == XLOG_XACT_ABORT_PREPARED))
5602 : {
5603 0 : *recordXtime = ((xl_xact_abort *) XLogRecGetData(record))->xact_time;
5604 0 : return true;
5605 : }
5606 0 : return false;
5607 : }
5608 :
5609 : /*
5610 : * For point-in-time recovery, this function decides whether we want to
5611 : * stop applying the XLOG before the current record.
5612 : *
5613 : * Returns TRUE if we are stopping, FALSE otherwise. If stopping, some
5614 : * information is saved in recoveryStopXid et al for use in annotating the
5615 : * new timeline's history file.
5616 : */
5617 : static bool
5618 0 : recoveryStopsBefore(XLogReaderState *record)
5619 : {
5620 0 : bool stopsHere = false;
5621 : uint8 xact_info;
5622 : bool isCommit;
5623 0 : TimestampTz recordXtime = 0;
5624 : TransactionId recordXid;
5625 :
5626 : /* Check if we should stop as soon as reaching consistency */
5627 0 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
5628 : {
5629 0 : ereport(LOG,
5630 : (errmsg("recovery stopping after reaching consistency")));
5631 :
5632 0 : recoveryStopAfter = false;
5633 0 : recoveryStopXid = InvalidTransactionId;
5634 0 : recoveryStopLSN = InvalidXLogRecPtr;
5635 0 : recoveryStopTime = 0;
5636 0 : recoveryStopName[0] = '\0';
5637 0 : return true;
5638 : }
5639 :
5640 : /* Check if target LSN has been reached */
5641 0 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
5642 0 : !recoveryTargetInclusive &&
5643 0 : record->ReadRecPtr >= recoveryTargetLSN)
5644 : {
5645 0 : recoveryStopAfter = false;
5646 0 : recoveryStopXid = InvalidTransactionId;
5647 0 : recoveryStopLSN = record->ReadRecPtr;
5648 0 : recoveryStopTime = 0;
5649 0 : recoveryStopName[0] = '\0';
5650 0 : ereport(LOG,
5651 : (errmsg("recovery stopping before WAL location (LSN) \"%X/%X\"",
5652 : (uint32) (recoveryStopLSN >> 32),
5653 : (uint32) recoveryStopLSN)));
5654 0 : return true;
5655 : }
5656 :
5657 : /* Otherwise we only consider stopping before COMMIT or ABORT records. */
5658 0 : if (XLogRecGetRmid(record) != RM_XACT_ID)
5659 0 : return false;
5660 :
5661 0 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
5662 :
5663 0 : if (xact_info == XLOG_XACT_COMMIT)
5664 : {
5665 0 : isCommit = true;
5666 0 : recordXid = XLogRecGetXid(record);
5667 : }
5668 0 : else if (xact_info == XLOG_XACT_COMMIT_PREPARED)
5669 : {
5670 0 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
5671 : xl_xact_parsed_commit parsed;
5672 :
5673 0 : isCommit = true;
5674 0 : ParseCommitRecord(XLogRecGetInfo(record),
5675 : xlrec,
5676 : &parsed);
5677 0 : recordXid = parsed.twophase_xid;
5678 : }
5679 0 : else if (xact_info == XLOG_XACT_ABORT)
5680 : {
5681 0 : isCommit = false;
5682 0 : recordXid = XLogRecGetXid(record);
5683 : }
5684 0 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
5685 : {
5686 0 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
5687 : xl_xact_parsed_abort parsed;
5688 :
5689 0 : isCommit = true;
5690 0 : ParseAbortRecord(XLogRecGetInfo(record),
5691 : xlrec,
5692 : &parsed);
5693 0 : recordXid = parsed.twophase_xid;
5694 : }
5695 : else
5696 0 : return false;
5697 :
5698 0 : if (recoveryTarget == RECOVERY_TARGET_XID && !recoveryTargetInclusive)
5699 : {
5700 : /*
5701 : * There can be only one transaction end record with this exact
5702 : * transactionid
5703 : *
5704 : * when testing for an xid, we MUST test for equality only, since
5705 : * transactions are numbered in the order they start, not the order
5706 : * they complete. A higher numbered xid will complete before you about
5707 : * 50% of the time...
5708 : */
5709 0 : stopsHere = (recordXid == recoveryTargetXid);
5710 : }
5711 :
5712 0 : if (recoveryTarget == RECOVERY_TARGET_TIME &&
5713 0 : getRecordTimestamp(record, &recordXtime))
5714 : {
5715 : /*
5716 : * There can be many transactions that share the same commit time, so
5717 : * we stop after the last one, if we are inclusive, or stop at the
5718 : * first one if we are exclusive
5719 : */
5720 0 : if (recoveryTargetInclusive)
5721 0 : stopsHere = (recordXtime > recoveryTargetTime);
5722 : else
5723 0 : stopsHere = (recordXtime >= recoveryTargetTime);
5724 : }
5725 :
5726 0 : if (stopsHere)
5727 : {
5728 0 : recoveryStopAfter = false;
5729 0 : recoveryStopXid = recordXid;
5730 0 : recoveryStopTime = recordXtime;
5731 0 : recoveryStopLSN = InvalidXLogRecPtr;
5732 0 : recoveryStopName[0] = '\0';
5733 :
5734 0 : if (isCommit)
5735 : {
5736 0 : ereport(LOG,
5737 : (errmsg("recovery stopping before commit of transaction %u, time %s",
5738 : recoveryStopXid,
5739 : timestamptz_to_str(recoveryStopTime))));
5740 : }
5741 : else
5742 : {
5743 0 : ereport(LOG,
5744 : (errmsg("recovery stopping before abort of transaction %u, time %s",
5745 : recoveryStopXid,
5746 : timestamptz_to_str(recoveryStopTime))));
5747 : }
5748 : }
5749 :
5750 0 : return stopsHere;
5751 : }
5752 :
5753 : /*
5754 : * Same as recoveryStopsBefore, but called after applying the record.
5755 : *
5756 : * We also track the timestamp of the latest applied COMMIT/ABORT
5757 : * record in XLogCtl->recoveryLastXTime.
5758 : */
5759 : static bool
5760 0 : recoveryStopsAfter(XLogReaderState *record)
5761 : {
5762 : uint8 info;
5763 : uint8 xact_info;
5764 : uint8 rmid;
5765 : TimestampTz recordXtime;
5766 :
5767 0 : info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
5768 0 : rmid = XLogRecGetRmid(record);
5769 :
5770 : /*
5771 : * There can be many restore points that share the same name; we stop at
5772 : * the first one.
5773 : */
5774 0 : if (recoveryTarget == RECOVERY_TARGET_NAME &&
5775 0 : rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
5776 : {
5777 : xl_restore_point *recordRestorePointData;
5778 :
5779 0 : recordRestorePointData = (xl_restore_point *) XLogRecGetData(record);
5780 :
5781 0 : if (strcmp(recordRestorePointData->rp_name, recoveryTargetName) == 0)
5782 : {
5783 0 : recoveryStopAfter = true;
5784 0 : recoveryStopXid = InvalidTransactionId;
5785 0 : recoveryStopLSN = InvalidXLogRecPtr;
5786 0 : (void) getRecordTimestamp(record, &recoveryStopTime);
5787 0 : strlcpy(recoveryStopName, recordRestorePointData->rp_name, MAXFNAMELEN);
5788 :
5789 0 : ereport(LOG,
5790 : (errmsg("recovery stopping at restore point \"%s\", time %s",
5791 : recoveryStopName,
5792 : timestamptz_to_str(recoveryStopTime))));
5793 0 : return true;
5794 : }
5795 : }
5796 :
5797 : /* Check if the target LSN has been reached */
5798 0 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
5799 0 : recoveryTargetInclusive &&
5800 0 : record->ReadRecPtr >= recoveryTargetLSN)
5801 : {
5802 0 : recoveryStopAfter = true;
5803 0 : recoveryStopXid = InvalidTransactionId;
5804 0 : recoveryStopLSN = record->ReadRecPtr;
5805 0 : recoveryStopTime = 0;
5806 0 : recoveryStopName[0] = '\0';
5807 0 : ereport(LOG,
5808 : (errmsg("recovery stopping after WAL location (LSN) \"%X/%X\"",
5809 : (uint32) (recoveryStopLSN >> 32),
5810 : (uint32) recoveryStopLSN)));
5811 0 : return true;
5812 : }
5813 :
5814 0 : if (rmid != RM_XACT_ID)
5815 0 : return false;
5816 :
5817 0 : xact_info = info & XLOG_XACT_OPMASK;
5818 :
5819 0 : if (xact_info == XLOG_XACT_COMMIT ||
5820 0 : xact_info == XLOG_XACT_COMMIT_PREPARED ||
5821 0 : xact_info == XLOG_XACT_ABORT ||
5822 : xact_info == XLOG_XACT_ABORT_PREPARED)
5823 : {
5824 : TransactionId recordXid;
5825 :
5826 : /* Update the last applied transaction timestamp */
5827 0 : if (getRecordTimestamp(record, &recordXtime))
5828 0 : SetLatestXTime(recordXtime);
5829 :
5830 : /* Extract the XID of the committed/aborted transaction */
5831 0 : if (xact_info == XLOG_XACT_COMMIT_PREPARED)
5832 : {
5833 0 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
5834 : xl_xact_parsed_commit parsed;
5835 :
5836 0 : ParseCommitRecord(XLogRecGetInfo(record),
5837 : xlrec,
5838 : &parsed);
5839 0 : recordXid = parsed.twophase_xid;
5840 : }
5841 0 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
5842 : {
5843 0 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
5844 : xl_xact_parsed_abort parsed;
5845 :
5846 0 : ParseAbortRecord(XLogRecGetInfo(record),
5847 : xlrec,
5848 : &parsed);
5849 0 : recordXid = parsed.twophase_xid;
5850 : }
5851 : else
5852 0 : recordXid = XLogRecGetXid(record);
5853 :
5854 : /*
5855 : * There can be only one transaction end record with this exact
5856 : * transactionid
5857 : *
5858 : * when testing for an xid, we MUST test for equality only, since
5859 : * transactions are numbered in the order they start, not the order
5860 : * they complete. A higher numbered xid will complete before you about
5861 : * 50% of the time...
5862 : */
5863 0 : if (recoveryTarget == RECOVERY_TARGET_XID && recoveryTargetInclusive &&
5864 0 : recordXid == recoveryTargetXid)
5865 : {
5866 0 : recoveryStopAfter = true;
5867 0 : recoveryStopXid = recordXid;
5868 0 : recoveryStopTime = recordXtime;
5869 0 : recoveryStopLSN = InvalidXLogRecPtr;
5870 0 : recoveryStopName[0] = '\0';
5871 :
5872 0 : if (xact_info == XLOG_XACT_COMMIT ||
5873 : xact_info == XLOG_XACT_COMMIT_PREPARED)
5874 : {
5875 0 : ereport(LOG,
5876 : (errmsg("recovery stopping after commit of transaction %u, time %s",
5877 : recoveryStopXid,
5878 : timestamptz_to_str(recoveryStopTime))));
5879 : }
5880 0 : else if (xact_info == XLOG_XACT_ABORT ||
5881 : xact_info == XLOG_XACT_ABORT_PREPARED)
5882 : {
5883 0 : ereport(LOG,
5884 : (errmsg("recovery stopping after abort of transaction %u, time %s",
5885 : recoveryStopXid,
5886 : timestamptz_to_str(recoveryStopTime))));
5887 : }
5888 0 : return true;
5889 : }
5890 : }
5891 :
5892 : /* Check if we should stop as soon as reaching consistency */
5893 0 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
5894 : {
5895 0 : ereport(LOG,
5896 : (errmsg("recovery stopping after reaching consistency")));
5897 :
5898 0 : recoveryStopAfter = true;
5899 0 : recoveryStopXid = InvalidTransactionId;
5900 0 : recoveryStopTime = 0;
5901 0 : recoveryStopLSN = InvalidXLogRecPtr;
5902 0 : recoveryStopName[0] = '\0';
5903 0 : return true;
5904 : }
5905 :
5906 0 : return false;
5907 : }
5908 :
5909 : /*
5910 : * Wait until shared recoveryPause flag is cleared.
5911 : *
5912 : * XXX Could also be done with shared latch, avoiding the pg_usleep loop.
5913 : * Probably not worth the trouble though. This state shouldn't be one that
5914 : * anyone cares about server power consumption in.
5915 : */
5916 : static void
5917 0 : recoveryPausesHere(void)
5918 : {
5919 : /* Don't pause unless users can connect! */
5920 0 : if (!LocalHotStandbyActive)
5921 0 : return;
5922 :
5923 0 : ereport(LOG,
5924 : (errmsg("recovery has paused"),
5925 : errhint("Execute pg_wal_replay_resume() to continue.")));
5926 :
5927 0 : while (RecoveryIsPaused())
5928 : {
5929 0 : pg_usleep(1000000L); /* 1000 ms */
5930 0 : HandleStartupProcInterrupts();
5931 : }
5932 : }
5933 :
5934 : bool
5935 0 : RecoveryIsPaused(void)
5936 : {
5937 : bool recoveryPause;
5938 :
5939 0 : SpinLockAcquire(&XLogCtl->info_lck);
5940 0 : recoveryPause = XLogCtl->recoveryPause;
5941 0 : SpinLockRelease(&XLogCtl->info_lck);
5942 :
5943 0 : return recoveryPause;
5944 : }
5945 :
5946 : void
5947 0 : SetRecoveryPause(bool recoveryPause)
5948 : {
5949 0 : SpinLockAcquire(&XLogCtl->info_lck);
5950 0 : XLogCtl->recoveryPause = recoveryPause;
5951 0 : SpinLockRelease(&XLogCtl->info_lck);
5952 0 : }
5953 :
5954 : /*
5955 : * When recovery_min_apply_delay is set, we wait long enough to make sure
5956 : * certain record types are applied at least that interval behind the master.
5957 : *
5958 : * Returns true if we waited.
5959 : *
5960 : * Note that the delay is calculated between the WAL record log time and
5961 : * the current time on standby. We would prefer to keep track of when this
5962 : * standby received each WAL record, which would allow a more consistent
5963 : * approach and one not affected by time synchronisation issues, but that
5964 : * is significantly more effort and complexity for little actual gain in
5965 : * usability.
5966 : */
5967 : static bool
5968 0 : recoveryApplyDelay(XLogReaderState *record)
5969 : {
5970 : uint8 xact_info;
5971 : TimestampTz xtime;
5972 : long secs;
5973 : int microsecs;
5974 :
5975 : /* nothing to do if no delay configured */
5976 0 : if (recovery_min_apply_delay <= 0)
5977 0 : return false;
5978 :
5979 : /* no delay is applied on a database not yet consistent */
5980 0 : if (!reachedConsistency)
5981 0 : return false;
5982 :
5983 : /*
5984 : * Is it a COMMIT record?
5985 : *
5986 : * We deliberately choose not to delay aborts since they have no effect on
5987 : * MVCC. We already allow replay of records that don't have a timestamp,
5988 : * so there is already opportunity for issues caused by early conflicts on
5989 : * standbys.
5990 : */
5991 0 : if (XLogRecGetRmid(record) != RM_XACT_ID)
5992 0 : return false;
5993 :
5994 0 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
5995 :
5996 0 : if (xact_info != XLOG_XACT_COMMIT &&
5997 : xact_info != XLOG_XACT_COMMIT_PREPARED)
5998 0 : return false;
5999 :
6000 0 : if (!getRecordTimestamp(record, &xtime))
6001 0 : return false;
6002 :
6003 0 : recoveryDelayUntilTime =
6004 0 : TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
6005 :
6006 : /*
6007 : * Exit without arming the latch if it's already past time to apply this
6008 : * record
6009 : */
6010 0 : TimestampDifference(GetCurrentTimestamp(), recoveryDelayUntilTime,
6011 : &secs, µsecs);
6012 0 : if (secs <= 0 && microsecs <= 0)
6013 0 : return false;
6014 :
6015 : while (true)
6016 : {
6017 0 : ResetLatch(&XLogCtl->recoveryWakeupLatch);
6018 :
6019 : /* might change the trigger file's location */
6020 0 : HandleStartupProcInterrupts();
6021 :
6022 0 : if (CheckForStandbyTrigger())
6023 0 : break;
6024 :
6025 : /*
6026 : * Wait for difference between GetCurrentTimestamp() and
6027 : * recoveryDelayUntilTime
6028 : */
6029 0 : TimestampDifference(GetCurrentTimestamp(), recoveryDelayUntilTime,
6030 : &secs, µsecs);
6031 :
6032 : /* NB: We're ignoring waits below min_apply_delay's resolution. */
6033 0 : if (secs <= 0 && microsecs / 1000 <= 0)
6034 0 : break;
6035 :
6036 0 : elog(DEBUG2, "recovery apply delay %ld seconds, %d milliseconds",
6037 : secs, microsecs / 1000);
6038 :
6039 0 : WaitLatch(&XLogCtl->recoveryWakeupLatch,
6040 : WL_LATCH_SET | WL_TIMEOUT | WL_POSTMASTER_DEATH,
6041 0 : secs * 1000L + microsecs / 1000,
6042 : WAIT_EVENT_RECOVERY_APPLY_DELAY);
6043 0 : }
6044 0 : return true;
6045 : }
6046 :
6047 : /*
6048 : * Save timestamp of latest processed commit/abort record.
6049 : *
6050 : * We keep this in XLogCtl, not a simple static variable, so that it can be
6051 : * seen by processes other than the startup process. Note in particular
6052 : * that CreateRestartPoint is executed in the checkpointer.
6053 : */
6054 : static void
6055 0 : SetLatestXTime(TimestampTz xtime)
6056 : {
6057 0 : SpinLockAcquire(&XLogCtl->info_lck);
6058 0 : XLogCtl->recoveryLastXTime = xtime;
6059 0 : SpinLockRelease(&XLogCtl->info_lck);
6060 0 : }
6061 :
6062 : /*
6063 : * Fetch timestamp of latest processed commit/abort record.
6064 : */
6065 : TimestampTz
6066 0 : GetLatestXTime(void)
6067 : {
6068 : TimestampTz xtime;
6069 :
6070 0 : SpinLockAcquire(&XLogCtl->info_lck);
6071 0 : xtime = XLogCtl->recoveryLastXTime;
6072 0 : SpinLockRelease(&XLogCtl->info_lck);
6073 :
6074 0 : return xtime;
6075 : }
6076 :
6077 : /*
6078 : * Save timestamp of the next chunk of WAL records to apply.
6079 : *
6080 : * We keep this in XLogCtl, not a simple static variable, so that it can be
6081 : * seen by all backends.
6082 : */
6083 : static void
6084 0 : SetCurrentChunkStartTime(TimestampTz xtime)
6085 : {
6086 0 : SpinLockAcquire(&XLogCtl->info_lck);
6087 0 : XLogCtl->currentChunkStartTime = xtime;
6088 0 : SpinLockRelease(&XLogCtl->info_lck);
6089 0 : }
6090 :
6091 : /*
6092 : * Fetch timestamp of latest processed commit/abort record.
6093 : * Startup process maintains an accurate local copy in XLogReceiptTime
6094 : */
6095 : TimestampTz
6096 0 : GetCurrentChunkReplayStartTime(void)
6097 : {
6098 : TimestampTz xtime;
6099 :
6100 0 : SpinLockAcquire(&XLogCtl->info_lck);
6101 0 : xtime = XLogCtl->currentChunkStartTime;
6102 0 : SpinLockRelease(&XLogCtl->info_lck);
6103 :
6104 0 : return xtime;
6105 : }
6106 :
6107 : /*
6108 : * Returns time of receipt of current chunk of XLOG data, as well as
6109 : * whether it was received from streaming replication or from archives.
6110 : */
6111 : void
6112 0 : GetXLogReceiptTime(TimestampTz *rtime, bool *fromStream)
6113 : {
6114 : /*
6115 : * This must be executed in the startup process, since we don't export the
6116 : * relevant state to shared memory.
6117 : */
6118 0 : Assert(InRecovery);
6119 :
6120 0 : *rtime = XLogReceiptTime;
6121 0 : *fromStream = (XLogReceiptSource == XLOG_FROM_STREAM);
6122 0 : }
6123 :
6124 : /*
6125 : * Note that text field supplied is a parameter name and does not require
6126 : * translation
6127 : */
6128 : #define RecoveryRequiresIntParameter(param_name, currValue, minValue) \
6129 : do { \
6130 : if ((currValue) < (minValue)) \
6131 : ereport(ERROR, \
6132 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
6133 : errmsg("hot standby is not possible because " \
6134 : "%s = %d is a lower setting than on the master server " \
6135 : "(its value was %d)", \
6136 : param_name, \
6137 : currValue, \
6138 : minValue))); \
6139 : } while(0)
6140 :
6141 : /*
6142 : * Check to see if required parameters are set high enough on this server
6143 : * for various aspects of recovery operation.
6144 : *
6145 : * Note that all the parameters which this function tests need to be
6146 : * listed in Administrator's Overview section in high-availability.sgml.
6147 : * If you change them, don't forget to update the list.
6148 : */
6149 : static void
6150 0 : CheckRequiredParameterValues(void)
6151 : {
6152 : /*
6153 : * For archive recovery, the WAL must be generated with at least 'replica'
6154 : * wal_level.
6155 : */
6156 0 : if (ArchiveRecoveryRequested && ControlFile->wal_level == WAL_LEVEL_MINIMAL)
6157 : {
6158 0 : ereport(WARNING,
6159 : (errmsg("WAL was generated with wal_level=minimal, data may be missing"),
6160 : errhint("This happens if you temporarily set wal_level=minimal without taking a new base backup.")));
6161 : }
6162 :
6163 : /*
6164 : * For Hot Standby, the WAL must be generated with 'replica' mode, and we
6165 : * must have at least as many backend slots as the primary.
6166 : */
6167 0 : if (ArchiveRecoveryRequested && EnableHotStandby)
6168 : {
6169 0 : if (ControlFile->wal_level < WAL_LEVEL_REPLICA)
6170 0 : ereport(ERROR,
6171 : (errmsg("hot standby is not possible because wal_level was not set to \"replica\" or higher on the master server"),
6172 : errhint("Either set wal_level to \"replica\" on the master, or turn off hot_standby here.")));
6173 :
6174 : /* We ignore autovacuum_max_workers when we make this test. */
6175 0 : RecoveryRequiresIntParameter("max_connections",
6176 : MaxConnections,
6177 : ControlFile->MaxConnections);
6178 0 : RecoveryRequiresIntParameter("max_worker_processes",
6179 : max_worker_processes,
6180 : ControlFile->max_worker_processes);
6181 0 : RecoveryRequiresIntParameter("max_prepared_transactions",
6182 : max_prepared_xacts,
6183 : ControlFile->max_prepared_xacts);
6184 0 : RecoveryRequiresIntParameter("max_locks_per_transaction",
6185 : max_locks_per_xact,
6186 : ControlFile->max_locks_per_xact);
6187 : }
6188 0 : }
6189 :
6190 : /*
6191 : * This must be called ONCE during postmaster or standalone-backend startup
6192 : */
6193 : void
6194 3 : StartupXLOG(void)
6195 : {
6196 : XLogCtlInsert *Insert;
6197 : CheckPoint checkPoint;
6198 : bool wasShutdown;
6199 3 : bool reachedStopPoint = false;
6200 3 : bool haveBackupLabel = false;
6201 3 : bool haveTblspcMap = false;
6202 : XLogRecPtr RecPtr,
6203 : checkPointLoc,
6204 : EndOfLog;
6205 : TimeLineID EndOfLogTLI;
6206 : TimeLineID PrevTimeLineID;
6207 : XLogRecord *record;
6208 : TransactionId oldestActiveXID;
6209 3 : bool backupEndRequired = false;
6210 3 : bool backupFromStandby = false;
6211 : DBState dbstate_at_startup;
6212 : XLogReaderState *xlogreader;
6213 : XLogPageReadPrivate private;
6214 3 : bool fast_promoted = false;
6215 : struct stat st;
6216 :
6217 : /*
6218 : * Read control file and check XLOG status looks valid.
6219 : *
6220 : * Note: in most control paths, *ControlFile is already valid and we need
6221 : * not do ReadControlFile() here, but might as well do it to be sure.
6222 : */
6223 3 : ReadControlFile();
6224 :
6225 6 : if (ControlFile->state < DB_SHUTDOWNED ||
6226 6 : ControlFile->state > DB_IN_PRODUCTION ||
6227 3 : !XRecOffIsValid(ControlFile->checkPoint))
6228 0 : ereport(FATAL,
6229 : (errmsg("control file contains invalid data")));
6230 :
6231 3 : if (ControlFile->state == DB_SHUTDOWNED)
6232 : {
6233 : /* This is the expected case, so don't be chatty in standalone mode */
6234 3 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
6235 : (errmsg("database system was shut down at %s",
6236 : str_time(ControlFile->time))));
6237 : }
6238 0 : else if (ControlFile->state == DB_SHUTDOWNED_IN_RECOVERY)
6239 0 : ereport(LOG,
6240 : (errmsg("database system was shut down in recovery at %s",
6241 : str_time(ControlFile->time))));
6242 0 : else if (ControlFile->state == DB_SHUTDOWNING)
6243 0 : ereport(LOG,
6244 : (errmsg("database system shutdown was interrupted; last known up at %s",
6245 : str_time(ControlFile->time))));
6246 0 : else if (ControlFile->state == DB_IN_CRASH_RECOVERY)
6247 0 : ereport(LOG,
6248 : (errmsg("database system was interrupted while in recovery at %s",
6249 : str_time(ControlFile->time)),
6250 : errhint("This probably means that some data is corrupted and"
6251 : " you will have to use the last backup for recovery.")));
6252 0 : else if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY)
6253 0 : ereport(LOG,
6254 : (errmsg("database system was interrupted while in recovery at log time %s",
6255 : str_time(ControlFile->checkPointCopy.time)),
6256 : errhint("If this has occurred more than once some data might be corrupted"
6257 : " and you might need to choose an earlier recovery target.")));
6258 0 : else if (ControlFile->state == DB_IN_PRODUCTION)
6259 0 : ereport(LOG,
6260 : (errmsg("database system was interrupted; last known up at %s",
6261 : str_time(ControlFile->time))));
6262 :
6263 : /* This is just to allow attaching to startup process with a debugger */
6264 : #ifdef XLOG_REPLAY_DELAY
6265 : if (ControlFile->state != DB_SHUTDOWNED)
6266 : pg_usleep(60000000L);
6267 : #endif
6268 :
6269 : /*
6270 : * Verify that pg_wal and pg_wal/archive_status exist. In cases where
6271 : * someone has performed a copy for PITR, these directories may have been
6272 : * excluded and need to be re-created.
6273 : */
6274 3 : ValidateXLOGDirectoryStructure();
6275 :
6276 : /*
6277 : * If we previously crashed, there might be data which we had written,
6278 : * intending to fsync it, but which we had not actually fsync'd yet.
6279 : * Therefore, a power failure in the near future might cause earlier
6280 : * unflushed writes to be lost, even though more recent data written to
6281 : * disk from here on would be persisted. To avoid that, fsync the entire
6282 : * data directory.
6283 : */
6284 3 : if (ControlFile->state != DB_SHUTDOWNED &&
6285 0 : ControlFile->state != DB_SHUTDOWNED_IN_RECOVERY)
6286 0 : SyncDataDirectory();
6287 :
6288 : /*
6289 : * Initialize on the assumption we want to recover to the latest timeline
6290 : * that's active according to pg_control.
6291 : */
6292 6 : if (ControlFile->minRecoveryPointTLI >
6293 3 : ControlFile->checkPointCopy.ThisTimeLineID)
6294 0 : recoveryTargetTLI = ControlFile->minRecoveryPointTLI;
6295 : else
6296 3 : recoveryTargetTLI = ControlFile->checkPointCopy.ThisTimeLineID;
6297 :
6298 : /*
6299 : * Check for recovery control file, and if so set up state for offline
6300 : * recovery
6301 : */
6302 3 : readRecoveryCommandFile();
6303 :
6304 : /*
6305 : * Save archive_cleanup_command in shared memory so that other processes
6306 : * can see it.
6307 : */
6308 3 : strlcpy(XLogCtl->archiveCleanupCommand,
6309 3 : archiveCleanupCommand ? archiveCleanupCommand : "",
6310 : sizeof(XLogCtl->archiveCleanupCommand));
6311 :
6312 3 : if (ArchiveRecoveryRequested)
6313 : {
6314 0 : if (StandbyModeRequested)
6315 0 : ereport(LOG,
6316 : (errmsg("entering standby mode")));
6317 0 : else if (recoveryTarget == RECOVERY_TARGET_XID)
6318 0 : ereport(LOG,
6319 : (errmsg("starting point-in-time recovery to XID %u",
6320 : recoveryTargetXid)));
6321 0 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
6322 0 : ereport(LOG,
6323 : (errmsg("starting point-in-time recovery to %s",
6324 : timestamptz_to_str(recoveryTargetTime))));
6325 0 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
6326 0 : ereport(LOG,
6327 : (errmsg("starting point-in-time recovery to \"%s\"",
6328 : recoveryTargetName)));
6329 0 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
6330 0 : ereport(LOG,
6331 : (errmsg("starting point-in-time recovery to WAL location (LSN) \"%X/%X\"",
6332 : (uint32) (recoveryTargetLSN >> 32),
6333 : (uint32) recoveryTargetLSN)));
6334 0 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
6335 0 : ereport(LOG,
6336 : (errmsg("starting point-in-time recovery to earliest consistent point")));
6337 : else
6338 0 : ereport(LOG,
6339 : (errmsg("starting archive recovery")));
6340 : }
6341 :
6342 : /*
6343 : * Take ownership of the wakeup latch if we're going to sleep during
6344 : * recovery.
6345 : */
6346 3 : if (StandbyModeRequested)
6347 0 : OwnLatch(&XLogCtl->recoveryWakeupLatch);
6348 :
6349 : /* Set up XLOG reader facility */
6350 3 : MemSet(&private, 0, sizeof(XLogPageReadPrivate));
6351 3 : xlogreader = XLogReaderAllocate(&XLogPageRead, &private);
6352 3 : if (!xlogreader)
6353 0 : ereport(ERROR,
6354 : (errcode(ERRCODE_OUT_OF_MEMORY),
6355 : errmsg("out of memory"),
6356 : errdetail("Failed while allocating a WAL reading processor.")));
6357 3 : xlogreader->system_identifier = ControlFile->system_identifier;
6358 :
6359 : /*
6360 : * Allocate pages dedicated to WAL consistency checks, those had better be
6361 : * aligned.
6362 : */
6363 3 : replay_image_masked = (char *) palloc(BLCKSZ);
6364 3 : master_image_masked = (char *) palloc(BLCKSZ);
6365 :
6366 3 : if (read_backup_label(&checkPointLoc, &backupEndRequired,
6367 : &backupFromStandby))
6368 : {
6369 0 : List *tablespaces = NIL;
6370 :
6371 : /*
6372 : * Archive recovery was requested, and thanks to the backup label
6373 : * file, we know how far we need to replay to reach consistency. Enter
6374 : * archive recovery directly.
6375 : */
6376 0 : InArchiveRecovery = true;
6377 0 : if (StandbyModeRequested)
6378 0 : StandbyMode = true;
6379 :
6380 : /*
6381 : * When a backup_label file is present, we want to roll forward from
6382 : * the checkpoint it identifies, rather than using pg_control.
6383 : */
6384 0 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 0, true);
6385 0 : if (record != NULL)
6386 : {
6387 0 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
6388 0 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
6389 0 : ereport(DEBUG1,
6390 : (errmsg("checkpoint record is at %X/%X",
6391 : (uint32) (checkPointLoc >> 32), (uint32) checkPointLoc)));
6392 0 : InRecovery = true; /* force recovery even if SHUTDOWNED */
6393 :
6394 : /*
6395 : * Make sure that REDO location exists. This may not be the case
6396 : * if there was a crash during an online backup, which left a
6397 : * backup_label around that references a WAL segment that's
6398 : * already been archived.
6399 : */
6400 0 : if (checkPoint.redo < checkPointLoc)
6401 : {
6402 0 : if (!ReadRecord(xlogreader, checkPoint.redo, LOG, false))
6403 0 : ereport(FATAL,
6404 : (errmsg("could not find redo location referenced by checkpoint record"),
6405 : errhint("If you are not restoring from a backup, try removing the file \"%s/backup_label\".", DataDir)));
6406 : }
6407 : }
6408 : else
6409 : {
6410 0 : ereport(FATAL,
6411 : (errmsg("could not locate required checkpoint record"),
6412 : errhint("If you are not restoring from a backup, try removing the file \"%s/backup_label\".", DataDir)));
6413 : wasShutdown = false; /* keep compiler quiet */
6414 : }
6415 :
6416 : /* read the tablespace_map file if present and create symlinks. */
6417 0 : if (read_tablespace_map(&tablespaces))
6418 : {
6419 : ListCell *lc;
6420 :
6421 0 : foreach(lc, tablespaces)
6422 : {
6423 0 : tablespaceinfo *ti = lfirst(lc);
6424 : char *linkloc;
6425 :
6426 0 : linkloc = psprintf("pg_tblspc/%s", ti->oid);
6427 :
6428 : /*
6429 : * Remove the existing symlink if any and Create the symlink
6430 : * under PGDATA.
6431 : */
6432 0 : remove_tablespace_symlink(linkloc);
6433 :
6434 0 : if (symlink(ti->path, linkloc) < 0)
6435 0 : ereport(ERROR,
6436 : (errcode_for_file_access(),
6437 : errmsg("could not create symbolic link \"%s\": %m",
6438 : linkloc)));
6439 :
6440 0 : pfree(ti->oid);
6441 0 : pfree(ti->path);
6442 0 : pfree(ti);
6443 : }
6444 :
6445 : /* set flag to delete it later */
6446 0 : haveTblspcMap = true;
6447 : }
6448 :
6449 : /* set flag to delete it later */
6450 0 : haveBackupLabel = true;
6451 : }
6452 : else
6453 : {
6454 : /*
6455 : * If tablespace_map file is present without backup_label file, there
6456 : * is no use of such file. There is no harm in retaining it, but it
6457 : * is better to get rid of the map file so that we don't have any
6458 : * redundant file in data directory and it will avoid any sort of
6459 : * confusion. It seems prudent though to just rename the file out of
6460 : * the way rather than delete it completely, also we ignore any error
6461 : * that occurs in rename operation as even if map file is present
6462 : * without backup_label file, it is harmless.
6463 : */
6464 3 : if (stat(TABLESPACE_MAP, &st) == 0)
6465 : {
6466 0 : unlink(TABLESPACE_MAP_OLD);
6467 0 : if (durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, DEBUG1) == 0)
6468 0 : ereport(LOG,
6469 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
6470 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
6471 : errdetail("File \"%s\" was renamed to \"%s\".",
6472 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
6473 : else
6474 0 : ereport(LOG,
6475 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
6476 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
6477 : errdetail("Could not rename file \"%s\" to \"%s\": %m.",
6478 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
6479 : }
6480 :
6481 : /*
6482 : * It's possible that archive recovery was requested, but we don't
6483 : * know how far we need to replay the WAL before we reach consistency.
6484 : * This can happen for example if a base backup is taken from a
6485 : * running server using an atomic filesystem snapshot, without calling
6486 : * pg_start/stop_backup. Or if you just kill a running master server
6487 : * and put it into archive recovery by creating a recovery.conf file.
6488 : *
6489 : * Our strategy in that case is to perform crash recovery first,
6490 : * replaying all the WAL present in pg_wal, and only enter archive
6491 : * recovery after that.
6492 : *
6493 : * But usually we already know how far we need to replay the WAL (up
6494 : * to minRecoveryPoint, up to backupEndPoint, or until we see an
6495 : * end-of-backup record), and we can enter archive recovery directly.
6496 : */
6497 3 : if (ArchiveRecoveryRequested &&
6498 0 : (ControlFile->minRecoveryPoint != InvalidXLogRecPtr ||
6499 0 : ControlFile->backupEndRequired ||
6500 0 : ControlFile->backupEndPoint != InvalidXLogRecPtr ||
6501 0 : ControlFile->state == DB_SHUTDOWNED))
6502 : {
6503 0 : InArchiveRecovery = true;
6504 0 : if (StandbyModeRequested)
6505 0 : StandbyMode = true;
6506 : }
6507 :
6508 : /*
6509 : * Get the last valid checkpoint record. If the latest one according
6510 : * to pg_control is broken, try the next-to-last one.
6511 : */
6512 3 : checkPointLoc = ControlFile->checkPoint;
6513 3 : RedoStartLSN = ControlFile->checkPointCopy.redo;
6514 3 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 1, true);
6515 3 : if (record != NULL)
6516 : {
6517 3 : ereport(DEBUG1,
6518 : (errmsg("checkpoint record is at %X/%X",
6519 : (uint32) (checkPointLoc >> 32), (uint32) checkPointLoc)));
6520 : }
6521 0 : else if (StandbyMode)
6522 : {
6523 : /*
6524 : * The last valid checkpoint record required for a streaming
6525 : * recovery exists in neither standby nor the primary.
6526 : */
6527 0 : ereport(PANIC,
6528 : (errmsg("could not locate a valid checkpoint record")));
6529 : }
6530 : else
6531 : {
6532 0 : checkPointLoc = ControlFile->prevCheckPoint;
6533 0 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 2, true);
6534 0 : if (record != NULL)
6535 : {
6536 0 : ereport(LOG,
6537 : (errmsg("using previous checkpoint record at %X/%X",
6538 : (uint32) (checkPointLoc >> 32), (uint32) checkPointLoc)));
6539 0 : InRecovery = true; /* force recovery even if SHUTDOWNED */
6540 : }
6541 : else
6542 0 : ereport(PANIC,
6543 : (errmsg("could not locate a valid checkpoint record")));
6544 : }
6545 3 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
6546 3 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
6547 : }
6548 :
6549 : /*
6550 : * Clear out any old relcache cache files. This is *necessary* if we do
6551 : * any WAL replay, since that would probably result in the cache files
6552 : * being out of sync with database reality. In theory we could leave them
6553 : * in place if the database had been cleanly shut down, but it seems
6554 : * safest to just remove them always and let them be rebuilt during the
6555 : * first backend startup. These files needs to be removed from all
6556 : * directories including pg_tblspc, however the symlinks are created only
6557 : * after reading tablespace_map file in case of archive recovery from
6558 : * backup, so needs to clear old relcache files here after creating
6559 : * symlinks.
6560 : */
6561 3 : RelationCacheInitFileRemove();
6562 :
6563 : /*
6564 : * If the location of the checkpoint record is not on the expected
6565 : * timeline in the history of the requested timeline, we cannot proceed:
6566 : * the backup is not part of the history of the requested timeline.
6567 : */
6568 3 : Assert(expectedTLEs); /* was initialized by reading checkpoint
6569 : * record */
6570 6 : if (tliOfPointInHistory(checkPointLoc, expectedTLEs) !=
6571 3 : checkPoint.ThisTimeLineID)
6572 : {
6573 : XLogRecPtr switchpoint;
6574 :
6575 : /*
6576 : * tliSwitchPoint will throw an error if the checkpoint's timeline is
6577 : * not in expectedTLEs at all.
6578 : */
6579 0 : switchpoint = tliSwitchPoint(ControlFile->checkPointCopy.ThisTimeLineID, expectedTLEs, NULL);
6580 0 : ereport(FATAL,
6581 : (errmsg("requested timeline %u is not a child of this server's history",
6582 : recoveryTargetTLI),
6583 : errdetail("Latest checkpoint is at %X/%X on timeline %u, but in the history of the requested timeline, the server forked off from that timeline at %X/%X.",
6584 : (uint32) (ControlFile->checkPoint >> 32),
6585 : (uint32) ControlFile->checkPoint,
6586 : ControlFile->checkPointCopy.ThisTimeLineID,
6587 : (uint32) (switchpoint >> 32),
6588 : (uint32) switchpoint)));
6589 : }
6590 :
6591 : /*
6592 : * The min recovery point should be part of the requested timeline's
6593 : * history, too.
6594 : */
6595 3 : if (!XLogRecPtrIsInvalid(ControlFile->minRecoveryPoint) &&
6596 0 : tliOfPointInHistory(ControlFile->minRecoveryPoint - 1, expectedTLEs) !=
6597 0 : ControlFile->minRecoveryPointTLI)
6598 0 : ereport(FATAL,
6599 : (errmsg("requested timeline %u does not contain minimum recovery point %X/%X on timeline %u",
6600 : recoveryTargetTLI,
6601 : (uint32) (ControlFile->minRecoveryPoint >> 32),
6602 : (uint32) ControlFile->minRecoveryPoint,
6603 : ControlFile->minRecoveryPointTLI)));
6604 :
6605 3 : LastRec = RecPtr = checkPointLoc;
6606 :
6607 3 : ereport(DEBUG1,
6608 : (errmsg_internal("redo record is at %X/%X; shutdown %s",
6609 : (uint32) (checkPoint.redo >> 32), (uint32) checkPoint.redo,
6610 : wasShutdown ? "TRUE" : "FALSE")));
6611 3 : ereport(DEBUG1,
6612 : (errmsg_internal("next transaction ID: %u:%u; next OID: %u",
6613 : checkPoint.nextXidEpoch, checkPoint.nextXid,
6614 : checkPoint.nextOid)));
6615 3 : ereport(DEBUG1,
6616 : (errmsg_internal("next MultiXactId: %u; next MultiXactOffset: %u",
6617 : checkPoint.nextMulti, checkPoint.nextMultiOffset)));
6618 3 : ereport(DEBUG1,
6619 : (errmsg_internal("oldest unfrozen transaction ID: %u, in database %u",
6620 : checkPoint.oldestXid, checkPoint.oldestXidDB)));
6621 3 : ereport(DEBUG1,
6622 : (errmsg_internal("oldest MultiXactId: %u, in database %u",
6623 : checkPoint.oldestMulti, checkPoint.oldestMultiDB)));
6624 3 : ereport(DEBUG1,
6625 : (errmsg_internal("commit timestamp Xid oldest/newest: %u/%u",
6626 : checkPoint.oldestCommitTsXid,
6627 : checkPoint.newestCommitTsXid)));
6628 3 : if (!TransactionIdIsNormal(checkPoint.nextXid))
6629 0 : ereport(PANIC,
6630 : (errmsg("invalid next transaction ID")));
6631 :
6632 : /* initialize shared memory variables from the checkpoint record */
6633 3 : ShmemVariableCache->nextXid = checkPoint.nextXid;
6634 3 : ShmemVariableCache->nextOid = checkPoint.nextOid;
6635 3 : ShmemVariableCache->oidCount = 0;
6636 3 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
6637 3 : AdvanceOldestClogXid(checkPoint.oldestXid);
6638 3 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
6639 3 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
6640 3 : SetCommitTsLimit(checkPoint.oldestCommitTsXid,
6641 : checkPoint.newestCommitTsXid);
6642 3 : XLogCtl->ckptXidEpoch = checkPoint.nextXidEpoch;
6643 3 : XLogCtl->ckptXid = checkPoint.nextXid;
6644 :
6645 : /*
6646 : * Initialize replication slots, before there's a chance to remove
6647 : * required resources.
6648 : */
6649 3 : StartupReplicationSlots();
6650 :
6651 : /*
6652 : * Startup logical state, needs to be setup now so we have proper data
6653 : * during crash recovery.
6654 : */
6655 3 : StartupReorderBuffer();
6656 :
6657 : /*
6658 : * Startup MultiXact. We need to do this early to be able to replay
6659 : * truncations.
6660 : */
6661 3 : StartupMultiXact();
6662 :
6663 : /*
6664 : * Ditto commit timestamps. In a standby, we do it if setting is enabled
6665 : * in ControlFile; in a master we base the decision on the GUC itself.
6666 : */
6667 3 : if (ArchiveRecoveryRequested ?
6668 0 : ControlFile->track_commit_timestamp : track_commit_timestamp)
6669 0 : StartupCommitTs();
6670 :
6671 : /*
6672 : * Recover knowledge about replay progress of known replication partners.
6673 : */
6674 3 : StartupReplicationOrigin();
6675 :
6676 : /*
6677 : * Initialize unlogged LSN. On a clean shutdown, it's restored from the
6678 : * control file. On recovery, all unlogged relations are blown away, so
6679 : * the unlogged LSN counter can be reset too.
6680 : */
6681 3 : if (ControlFile->state == DB_SHUTDOWNED)
6682 3 : XLogCtl->unloggedLSN = ControlFile->unloggedLSN;
6683 : else
6684 0 : XLogCtl->unloggedLSN = 1;
6685 :
6686 : /*
6687 : * We must replay WAL entries using the same TimeLineID they were created
6688 : * under, so temporarily adopt the TLI indicated by the checkpoint (see
6689 : * also xlog_redo()).
6690 : */
6691 3 : ThisTimeLineID = checkPoint.ThisTimeLineID;
6692 :
6693 : /*
6694 : * Copy any missing timeline history files between 'now' and the recovery
6695 : * target timeline from archive to pg_wal. While we don't need those files
6696 : * ourselves - the history file of the recovery target timeline covers all
6697 : * the previous timelines in the history too - a cascading standby server
6698 : * might be interested in them. Or, if you archive the WAL from this
6699 : * server to a different archive than the master, it'd be good for all the
6700 : * history files to get archived there after failover, so that you can use
6701 : * one of the old timelines as a PITR target. Timeline history files are
6702 : * small, so it's better to copy them unnecessarily than not copy them and
6703 : * regret later.
6704 : */
6705 3 : restoreTimeLineHistoryFiles(ThisTimeLineID, recoveryTargetTLI);
6706 :
6707 : /*
6708 : * Before running in recovery, scan pg_twophase and fill in its status to
6709 : * be able to work on entries generated by redo. Doing a scan before
6710 : * taking any recovery action has the merit to discard any 2PC files that
6711 : * are newer than the first record to replay, saving from any conflicts at
6712 : * replay. This avoids as well any subsequent scans when doing recovery
6713 : * of the on-disk two-phase data.
6714 : */
6715 3 : restoreTwoPhaseData();
6716 :
6717 3 : lastFullPageWrites = checkPoint.fullPageWrites;
6718 :
6719 3 : RedoRecPtr = XLogCtl->RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
6720 3 : doPageWrites = lastFullPageWrites;
6721 :
6722 3 : if (RecPtr < checkPoint.redo)
6723 0 : ereport(PANIC,
6724 : (errmsg("invalid redo in checkpoint record")));
6725 :
6726 : /*
6727 : * Check whether we need to force recovery from WAL. If it appears to
6728 : * have been a clean shutdown and we did not have a recovery.conf file,
6729 : * then assume no recovery needed.
6730 : */
6731 3 : if (checkPoint.redo < RecPtr)
6732 : {
6733 0 : if (wasShutdown)
6734 0 : ereport(PANIC,
6735 : (errmsg("invalid redo record in shutdown checkpoint")));
6736 0 : InRecovery = true;
6737 : }
6738 3 : else if (ControlFile->state != DB_SHUTDOWNED)
6739 0 : InRecovery = true;
6740 3 : else if (ArchiveRecoveryRequested)
6741 : {
6742 : /* force recovery due to presence of recovery.conf */
6743 0 : InRecovery = true;
6744 : }
6745 :
6746 : /* REDO */
6747 3 : if (InRecovery)
6748 : {
6749 : int rmid;
6750 :
6751 : /*
6752 : * Update pg_control to show that we are recovering and to show the
6753 : * selected checkpoint as the place we are starting from. We also mark
6754 : * pg_control with any minimum recovery stop point obtained from a
6755 : * backup history file.
6756 : */
6757 0 : dbstate_at_startup = ControlFile->state;
6758 0 : if (InArchiveRecovery)
6759 0 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
6760 : else
6761 : {
6762 0 : ereport(LOG,
6763 : (errmsg("database system was not properly shut down; "
6764 : "automatic recovery in progress")));
6765 0 : if (recoveryTargetTLI > ControlFile->checkPointCopy.ThisTimeLineID)
6766 0 : ereport(LOG,
6767 : (errmsg("crash recovery starts in timeline %u "
6768 : "and has target timeline %u",
6769 : ControlFile->checkPointCopy.ThisTimeLineID,
6770 : recoveryTargetTLI)));
6771 0 : ControlFile->state = DB_IN_CRASH_RECOVERY;
6772 : }
6773 0 : ControlFile->prevCheckPoint = ControlFile->checkPoint;
6774 0 : ControlFile->checkPoint = checkPointLoc;
6775 0 : ControlFile->checkPointCopy = checkPoint;
6776 0 : if (InArchiveRecovery)
6777 : {
6778 : /* initialize minRecoveryPoint if not set yet */
6779 0 : if (ControlFile->minRecoveryPoint < checkPoint.redo)
6780 : {
6781 0 : ControlFile->minRecoveryPoint = checkPoint.redo;
6782 0 : ControlFile->minRecoveryPointTLI = checkPoint.ThisTimeLineID;
6783 : }
6784 : }
6785 :
6786 : /*
6787 : * Set backupStartPoint if we're starting recovery from a base backup.
6788 : *
6789 : * Also set backupEndPoint and use minRecoveryPoint as the backup end
6790 : * location if we're starting recovery from a base backup which was
6791 : * taken from a standby. In this case, the database system status in
6792 : * pg_control must indicate that the database was already in recovery.
6793 : * Usually that will be DB_IN_ARCHIVE_RECOVERY but also can be
6794 : * DB_SHUTDOWNED_IN_RECOVERY if recovery previously was interrupted
6795 : * before reaching this point; e.g. because restore_command or
6796 : * primary_conninfo were faulty.
6797 : *
6798 : * Any other state indicates that the backup somehow became corrupted
6799 : * and we can't sensibly continue with recovery.
6800 : */
6801 0 : if (haveBackupLabel)
6802 : {
6803 0 : ControlFile->backupStartPoint = checkPoint.redo;
6804 0 : ControlFile->backupEndRequired = backupEndRequired;
6805 :
6806 0 : if (backupFromStandby)
6807 : {
6808 0 : if (dbstate_at_startup != DB_IN_ARCHIVE_RECOVERY &&
6809 : dbstate_at_startup != DB_SHUTDOWNED_IN_RECOVERY)
6810 0 : ereport(FATAL,
6811 : (errmsg("backup_label contains data inconsistent with control file"),
6812 : errhint("This means that the backup is corrupted and you will "
6813 : "have to use another backup for recovery.")));
6814 0 : ControlFile->backupEndPoint = ControlFile->minRecoveryPoint;
6815 : }
6816 : }
6817 0 : ControlFile->time = (pg_time_t) time(NULL);
6818 : /* No need to hold ControlFileLock yet, we aren't up far enough */
6819 0 : UpdateControlFile();
6820 :
6821 : /* initialize our local copy of minRecoveryPoint */
6822 0 : minRecoveryPoint = ControlFile->minRecoveryPoint;
6823 0 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
6824 :
6825 : /*
6826 : * Reset pgstat data, because it may be invalid after recovery.
6827 : */
6828 0 : pgstat_reset_all();
6829 :
6830 : /*
6831 : * If there was a backup label file, it's done its job and the info
6832 : * has now been propagated into pg_control. We must get rid of the
6833 : * label file so that if we crash during recovery, we'll pick up at
6834 : * the latest recovery restartpoint instead of going all the way back
6835 : * to the backup start point. It seems prudent though to just rename
6836 : * the file out of the way rather than delete it completely.
6837 : */
6838 0 : if (haveBackupLabel)
6839 : {
6840 0 : unlink(BACKUP_LABEL_OLD);
6841 0 : durable_rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, FATAL);
6842 : }
6843 :
6844 : /*
6845 : * If there was a tablespace_map file, it's done its job and the
6846 : * symlinks have been created. We must get rid of the map file so
6847 : * that if we crash during recovery, we don't create symlinks again.
6848 : * It seems prudent though to just rename the file out of the way
6849 : * rather than delete it completely.
6850 : */
6851 0 : if (haveTblspcMap)
6852 : {
6853 0 : unlink(TABLESPACE_MAP_OLD);
6854 0 : durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, FATAL);
6855 : }
6856 :
6857 : /* Check that the GUCs used to generate the WAL allow recovery */
6858 0 : CheckRequiredParameterValues();
6859 :
6860 : /*
6861 : * We're in recovery, so unlogged relations may be trashed and must be
6862 : * reset. This should be done BEFORE allowing Hot Standby
6863 : * connections, so that read-only backends don't try to read whatever
6864 : * garbage is left over from before.
6865 : */
6866 0 : ResetUnloggedRelations(UNLOGGED_RELATION_CLEANUP);
6867 :
6868 : /*
6869 : * Likewise, delete any saved transaction snapshot files that got left
6870 : * behind by crashed backends.
6871 : */
6872 0 : DeleteAllExportedSnapshotFiles();
6873 :
6874 : /*
6875 : * Initialize for Hot Standby, if enabled. We won't let backends in
6876 : * yet, not until we've reached the min recovery point specified in
6877 : * control file and we've established a recovery snapshot from a
6878 : * running-xacts WAL record.
6879 : */
6880 0 : if (ArchiveRecoveryRequested && EnableHotStandby)
6881 : {
6882 : TransactionId *xids;
6883 : int nxids;
6884 :
6885 0 : ereport(DEBUG1,
6886 : (errmsg("initializing for hot standby")));
6887 :
6888 0 : InitRecoveryTransactionEnvironment();
6889 :
6890 0 : if (wasShutdown)
6891 0 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
6892 : else
6893 0 : oldestActiveXID = checkPoint.oldestActiveXid;
6894 0 : Assert(TransactionIdIsValid(oldestActiveXID));
6895 :
6896 : /* Tell procarray about the range of xids it has to deal with */
6897 0 : ProcArrayInitRecovery(ShmemVariableCache->nextXid);
6898 :
6899 : /*
6900 : * Startup commit log and subtrans only. MultiXact and commit
6901 : * timestamp have already been started up and other SLRUs are not
6902 : * maintained during recovery and need not be started yet.
6903 : */
6904 0 : StartupCLOG();
6905 0 : StartupSUBTRANS(oldestActiveXID);
6906 :
6907 : /*
6908 : * If we're beginning at a shutdown checkpoint, we know that
6909 : * nothing was running on the master at this point. So fake-up an
6910 : * empty running-xacts record and use that here and now. Recover
6911 : * additional standby state for prepared transactions.
6912 : */
6913 0 : if (wasShutdown)
6914 : {
6915 : RunningTransactionsData running;
6916 : TransactionId latestCompletedXid;
6917 :
6918 : /*
6919 : * Construct a RunningTransactions snapshot representing a
6920 : * shut down server, with only prepared transactions still
6921 : * alive. We're never overflowed at this point because all
6922 : * subxids are listed with their parent prepared transactions.
6923 : */
6924 0 : running.xcnt = nxids;
6925 0 : running.subxcnt = 0;
6926 0 : running.subxid_overflow = false;
6927 0 : running.nextXid = checkPoint.nextXid;
6928 0 : running.oldestRunningXid = oldestActiveXID;
6929 0 : latestCompletedXid = checkPoint.nextXid;
6930 0 : TransactionIdRetreat(latestCompletedXid);
6931 0 : Assert(TransactionIdIsNormal(latestCompletedXid));
6932 0 : running.latestCompletedXid = latestCompletedXid;
6933 0 : running.xids = xids;
6934 :
6935 0 : ProcArrayApplyRecoveryInfo(&running);
6936 :
6937 0 : StandbyRecoverPreparedTransactions();
6938 : }
6939 : }
6940 :
6941 : /* Initialize resource managers */
6942 0 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
6943 : {
6944 0 : if (RmgrTable[rmid].rm_startup != NULL)
6945 0 : RmgrTable[rmid].rm_startup();
6946 : }
6947 :
6948 : /*
6949 : * Initialize shared variables for tracking progress of WAL replay, as
6950 : * if we had just replayed the record before the REDO location (or the
6951 : * checkpoint record itself, if it's a shutdown checkpoint).
6952 : */
6953 0 : SpinLockAcquire(&XLogCtl->info_lck);
6954 0 : if (checkPoint.redo < RecPtr)
6955 0 : XLogCtl->replayEndRecPtr = checkPoint.redo;
6956 : else
6957 0 : XLogCtl->replayEndRecPtr = EndRecPtr;
6958 0 : XLogCtl->replayEndTLI = ThisTimeLineID;
6959 0 : XLogCtl->lastReplayedEndRecPtr = XLogCtl->replayEndRecPtr;
6960 0 : XLogCtl->lastReplayedTLI = XLogCtl->replayEndTLI;
6961 0 : XLogCtl->recoveryLastXTime = 0;
6962 0 : XLogCtl->currentChunkStartTime = 0;
6963 0 : XLogCtl->recoveryPause = false;
6964 0 : SpinLockRelease(&XLogCtl->info_lck);
6965 :
6966 : /* Also ensure XLogReceiptTime has a sane value */
6967 0 : XLogReceiptTime = GetCurrentTimestamp();
6968 :
6969 : /*
6970 : * Let postmaster know we've started redo now, so that it can launch
6971 : * checkpointer to perform restartpoints. We don't bother during
6972 : * crash recovery as restartpoints can only be performed during
6973 : * archive recovery. And we'd like to keep crash recovery simple, to
6974 : * avoid introducing bugs that could affect you when recovering after
6975 : * crash.
6976 : *
6977 : * After this point, we can no longer assume that we're the only
6978 : * process in addition to postmaster! Also, fsync requests are
6979 : * subsequently to be handled by the checkpointer, not locally.
6980 : */
6981 0 : if (ArchiveRecoveryRequested && IsUnderPostmaster)
6982 : {
6983 0 : PublishStartupProcessInformation();
6984 0 : SetForwardFsyncRequests();
6985 0 : SendPostmasterSignal(PMSIGNAL_RECOVERY_STARTED);
6986 0 : bgwriterLaunched = true;
6987 : }
6988 :
6989 : /*
6990 : * Allow read-only connections immediately if we're consistent
6991 : * already.
6992 : */
6993 0 : CheckRecoveryConsistency();
6994 :
6995 : /*
6996 : * Find the first record that logically follows the checkpoint --- it
6997 : * might physically precede it, though.
6998 : */
6999 0 : if (checkPoint.redo < RecPtr)
7000 : {
7001 : /* back up to find the record */
7002 0 : record = ReadRecord(xlogreader, checkPoint.redo, PANIC, false);
7003 : }
7004 : else
7005 : {
7006 : /* just have to read next record after CheckPoint */
7007 0 : record = ReadRecord(xlogreader, InvalidXLogRecPtr, LOG, false);
7008 : }
7009 :
7010 0 : if (record != NULL)
7011 : {
7012 : ErrorContextCallback errcallback;
7013 : TimestampTz xtime;
7014 :
7015 0 : InRedo = true;
7016 :
7017 0 : ereport(LOG,
7018 : (errmsg("redo starts at %X/%X",
7019 : (uint32) (ReadRecPtr >> 32), (uint32) ReadRecPtr)));
7020 :
7021 : /*
7022 : * main redo apply loop
7023 : */
7024 : do
7025 : {
7026 0 : bool switchedTLI = false;
7027 :
7028 : #ifdef WAL_DEBUG
7029 : if (XLOG_DEBUG ||
7030 : (rmid == RM_XACT_ID && trace_recovery_messages <= DEBUG2) ||
7031 : (rmid != RM_XACT_ID && trace_recovery_messages <= DEBUG3))
7032 : {
7033 : StringInfoData buf;
7034 :
7035 : initStringInfo(&buf);
7036 : appendStringInfo(&buf, "REDO @ %X/%X; LSN %X/%X: ",
7037 : (uint32) (ReadRecPtr >> 32), (uint32) ReadRecPtr,
7038 : (uint32) (EndRecPtr >> 32), (uint32) EndRecPtr);
7039 : xlog_outrec(&buf, xlogreader);
7040 : appendStringInfoString(&buf, " - ");
7041 : xlog_outdesc(&buf, xlogreader);
7042 : elog(LOG, "%s", buf.data);
7043 : pfree(buf.data);
7044 : }
7045 : #endif
7046 :
7047 : /* Handle interrupt signals of startup process */
7048 0 : HandleStartupProcInterrupts();
7049 :
7050 : /*
7051 : * Pause WAL replay, if requested by a hot-standby session via
7052 : * SetRecoveryPause().
7053 : *
7054 : * Note that we intentionally don't take the info_lck spinlock
7055 : * here. We might therefore read a slightly stale value of
7056 : * the recoveryPause flag, but it can't be very stale (no
7057 : * worse than the last spinlock we did acquire). Since a
7058 : * pause request is a pretty asynchronous thing anyway,
7059 : * possibly responding to it one WAL record later than we
7060 : * otherwise would is a minor issue, so it doesn't seem worth
7061 : * adding another spinlock cycle to prevent that.
7062 : */
7063 0 : if (((volatile XLogCtlData *) XLogCtl)->recoveryPause)
7064 0 : recoveryPausesHere();
7065 :
7066 : /*
7067 : * Have we reached our recovery target?
7068 : */
7069 0 : if (recoveryStopsBefore(xlogreader))
7070 : {
7071 0 : reachedStopPoint = true; /* see below */
7072 0 : break;
7073 : }
7074 :
7075 : /*
7076 : * If we've been asked to lag the master, wait on latch until
7077 : * enough time has passed.
7078 : */
7079 0 : if (recoveryApplyDelay(xlogreader))
7080 : {
7081 : /*
7082 : * We test for paused recovery again here. If user sets
7083 : * delayed apply, it may be because they expect to pause
7084 : * recovery in case of problems, so we must test again
7085 : * here otherwise pausing during the delay-wait wouldn't
7086 : * work.
7087 : */
7088 0 : if (((volatile XLogCtlData *) XLogCtl)->recoveryPause)
7089 0 : recoveryPausesHere();
7090 : }
7091 :
7092 : /* Setup error traceback support for ereport() */
7093 0 : errcallback.callback = rm_redo_error_callback;
7094 0 : errcallback.arg = (void *) xlogreader;
7095 0 : errcallback.previous = error_context_stack;
7096 0 : error_context_stack = &errcallback;
7097 :
7098 : /*
7099 : * ShmemVariableCache->nextXid must be beyond record's xid.
7100 : *
7101 : * We don't expect anyone else to modify nextXid, hence we
7102 : * don't need to hold a lock while examining it. We still
7103 : * acquire the lock to modify it, though.
7104 : */
7105 0 : if (TransactionIdFollowsOrEquals(record->xl_xid,
7106 0 : ShmemVariableCache->nextXid))
7107 : {
7108 0 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
7109 0 : ShmemVariableCache->nextXid = record->xl_xid;
7110 0 : TransactionIdAdvance(ShmemVariableCache->nextXid);
7111 0 : LWLockRelease(XidGenLock);
7112 : }
7113 :
7114 : /*
7115 : * Before replaying this record, check if this record causes
7116 : * the current timeline to change. The record is already
7117 : * considered to be part of the new timeline, so we update
7118 : * ThisTimeLineID before replaying it. That's important so
7119 : * that replayEndTLI, which is recorded as the minimum
7120 : * recovery point's TLI if recovery stops after this record,
7121 : * is set correctly.
7122 : */
7123 0 : if (record->xl_rmid == RM_XLOG_ID)
7124 : {
7125 0 : TimeLineID newTLI = ThisTimeLineID;
7126 0 : TimeLineID prevTLI = ThisTimeLineID;
7127 0 : uint8 info = record->xl_info & ~XLR_INFO_MASK;
7128 :
7129 0 : if (info == XLOG_CHECKPOINT_SHUTDOWN)
7130 : {
7131 : CheckPoint checkPoint;
7132 :
7133 0 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
7134 0 : newTLI = checkPoint.ThisTimeLineID;
7135 0 : prevTLI = checkPoint.PrevTimeLineID;
7136 : }
7137 0 : else if (info == XLOG_END_OF_RECOVERY)
7138 : {
7139 : xl_end_of_recovery xlrec;
7140 :
7141 0 : memcpy(&xlrec, XLogRecGetData(xlogreader), sizeof(xl_end_of_recovery));
7142 0 : newTLI = xlrec.ThisTimeLineID;
7143 0 : prevTLI = xlrec.PrevTimeLineID;
7144 : }
7145 :
7146 0 : if (newTLI != ThisTimeLineID)
7147 : {
7148 : /* Check that it's OK to switch to this TLI */
7149 0 : checkTimeLineSwitch(EndRecPtr, newTLI, prevTLI);
7150 :
7151 : /* Following WAL records should be run with new TLI */
7152 0 : ThisTimeLineID = newTLI;
7153 0 : switchedTLI = true;
7154 : }
7155 : }
7156 :
7157 : /*
7158 : * Update shared replayEndRecPtr before replaying this record,
7159 : * so that XLogFlush will update minRecoveryPoint correctly.
7160 : */
7161 0 : SpinLockAcquire(&XLogCtl->info_lck);
7162 0 : XLogCtl->replayEndRecPtr = EndRecPtr;
7163 0 : XLogCtl->replayEndTLI = ThisTimeLineID;
7164 0 : SpinLockRelease(&XLogCtl->info_lck);
7165 :
7166 : /*
7167 : * If we are attempting to enter Hot Standby mode, process
7168 : * XIDs we see
7169 : */
7170 0 : if (standbyState >= STANDBY_INITIALIZED &&
7171 0 : TransactionIdIsValid(record->xl_xid))
7172 0 : RecordKnownAssignedTransactionIds(record->xl_xid);
7173 :
7174 : /* Now apply the WAL record itself */
7175 0 : RmgrTable[record->xl_rmid].rm_redo(xlogreader);
7176 :
7177 : /*
7178 : * After redo, check whether the backup pages associated with
7179 : * the WAL record are consistent with the existing pages. This
7180 : * check is done only if consistency check is enabled for this
7181 : * record.
7182 : */
7183 0 : if ((record->xl_info & XLR_CHECK_CONSISTENCY) != 0)
7184 0 : checkXLogConsistency(xlogreader);
7185 :
7186 : /* Pop the error context stack */
7187 0 : error_context_stack = errcallback.previous;
7188 :
7189 : /*
7190 : * Update lastReplayedEndRecPtr after this record has been
7191 : * successfully replayed.
7192 : */
7193 0 : SpinLockAcquire(&XLogCtl->info_lck);
7194 0 : XLogCtl->lastReplayedEndRecPtr = EndRecPtr;
7195 0 : XLogCtl->lastReplayedTLI = ThisTimeLineID;
7196 0 : SpinLockRelease(&XLogCtl->info_lck);
7197 :
7198 : /*
7199 : * If rm_redo called XLogRequestWalReceiverReply, then we wake
7200 : * up the receiver so that it notices the updated
7201 : * lastReplayedEndRecPtr and sends a reply to the master.
7202 : */
7203 0 : if (doRequestWalReceiverReply)
7204 : {
7205 0 : doRequestWalReceiverReply = false;
7206 0 : WalRcvForceReply();
7207 : }
7208 :
7209 : /* Remember this record as the last-applied one */
7210 0 : LastRec = ReadRecPtr;
7211 :
7212 : /* Allow read-only connections if we're consistent now */
7213 0 : CheckRecoveryConsistency();
7214 :
7215 : /* Is this a timeline switch? */
7216 0 : if (switchedTLI)
7217 : {
7218 : /*
7219 : * Before we continue on the new timeline, clean up any
7220 : * (possibly bogus) future WAL segments on the old
7221 : * timeline.
7222 : */
7223 0 : RemoveNonParentXlogFiles(EndRecPtr, ThisTimeLineID);
7224 :
7225 : /*
7226 : * Wake up any walsenders to notice that we are on a new
7227 : * timeline.
7228 : */
7229 0 : if (switchedTLI && AllowCascadeReplication())
7230 0 : WalSndWakeup();
7231 : }
7232 :
7233 : /* Exit loop if we reached inclusive recovery target */
7234 0 : if (recoveryStopsAfter(xlogreader))
7235 : {
7236 0 : reachedStopPoint = true;
7237 0 : break;
7238 : }
7239 :
7240 : /* Else, try to fetch the next WAL record */
7241 0 : record = ReadRecord(xlogreader, InvalidXLogRecPtr, LOG, false);
7242 0 : } while (record != NULL);
7243 :
7244 : /*
7245 : * end of main redo apply loop
7246 : */
7247 :
7248 0 : if (reachedStopPoint)
7249 : {
7250 0 : if (!reachedConsistency)
7251 0 : ereport(FATAL,
7252 : (errmsg("requested recovery stop point is before consistent recovery point")));
7253 :
7254 : /*
7255 : * This is the last point where we can restart recovery with a
7256 : * new recovery target, if we shutdown and begin again. After
7257 : * this, Resource Managers may choose to do permanent
7258 : * corrective actions at end of recovery.
7259 : */
7260 0 : switch (recoveryTargetAction)
7261 : {
7262 : case RECOVERY_TARGET_ACTION_SHUTDOWN:
7263 :
7264 : /*
7265 : * exit with special return code to request shutdown
7266 : * of postmaster. Log messages issued from
7267 : * postmaster.
7268 : */
7269 0 : proc_exit(3);
7270 :
7271 : case RECOVERY_TARGET_ACTION_PAUSE:
7272 0 : SetRecoveryPause(true);
7273 0 : recoveryPausesHere();
7274 :
7275 : /* drop into promote */
7276 :
7277 : case RECOVERY_TARGET_ACTION_PROMOTE:
7278 0 : break;
7279 : }
7280 : }
7281 :
7282 : /* Allow resource managers to do any required cleanup. */
7283 0 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
7284 : {
7285 0 : if (RmgrTable[rmid].rm_cleanup != NULL)
7286 0 : RmgrTable[rmid].rm_cleanup();
7287 : }
7288 :
7289 0 : ereport(LOG,
7290 : (errmsg("redo done at %X/%X",
7291 : (uint32) (ReadRecPtr >> 32), (uint32) ReadRecPtr)));
7292 0 : xtime = GetLatestXTime();
7293 0 : if (xtime)
7294 0 : ereport(LOG,
7295 : (errmsg("last completed transaction was at log time %s",
7296 : timestamptz_to_str(xtime))));
7297 :
7298 0 : InRedo = false;
7299 : }
7300 : else
7301 : {
7302 : /* there are no WAL records following the checkpoint */
7303 0 : ereport(LOG,
7304 : (errmsg("redo is not required")));
7305 : }
7306 : }
7307 :
7308 : /*
7309 : * Kill WAL receiver, if it's still running, before we continue to write
7310 : * the startup checkpoint record. It will trump over the checkpoint and
7311 : * subsequent records if it's still alive when we start writing WAL.
7312 : */
7313 3 : ShutdownWalRcv();
7314 :
7315 : /*
7316 : * Reset unlogged relations to the contents of their INIT fork. This is
7317 : * done AFTER recovery is complete so as to include any unlogged relations
7318 : * created during recovery, but BEFORE recovery is marked as having
7319 : * completed successfully. Otherwise we'd not retry if any of the post
7320 : * end-of-recovery steps fail.
7321 : */
7322 3 : if (InRecovery)
7323 0 : ResetUnloggedRelations(UNLOGGED_RELATION_INIT);
7324 :
7325 : /*
7326 : * We don't need the latch anymore. It's not strictly necessary to disown
7327 : * it, but let's do it for the sake of tidiness.
7328 : */
7329 3 : if (StandbyModeRequested)
7330 0 : DisownLatch(&XLogCtl->recoveryWakeupLatch);
7331 :
7332 : /*
7333 : * We are now done reading the xlog from stream. Turn off streaming
7334 : * recovery to force fetching the files (which would be required at end of
7335 : * recovery, e.g., timeline history file) from archive or pg_wal.
7336 : */
7337 3 : StandbyMode = false;
7338 :
7339 : /*
7340 : * Re-fetch the last valid or last applied record, so we can identify the
7341 : * exact endpoint of what we consider the valid portion of WAL.
7342 : */
7343 3 : record = ReadRecord(xlogreader, LastRec, PANIC, false);
7344 3 : EndOfLog = EndRecPtr;
7345 :
7346 : /*
7347 : * EndOfLogTLI is the TLI in the filename of the XLOG segment containing
7348 : * the end-of-log. It could be different from the timeline that EndOfLog
7349 : * nominally belongs to, if there was a timeline switch in that segment,
7350 : * and we were reading the old WAL from a segment belonging to a higher
7351 : * timeline.
7352 : */
7353 3 : EndOfLogTLI = xlogreader->readPageTLI;
7354 :
7355 : /*
7356 : * Complain if we did not roll forward far enough to render the backup
7357 : * dump consistent. Note: it is indeed okay to look at the local variable
7358 : * minRecoveryPoint here, even though ControlFile->minRecoveryPoint might
7359 : * be further ahead --- ControlFile->minRecoveryPoint cannot have been
7360 : * advanced beyond the WAL we processed.
7361 : */
7362 3 : if (InRecovery &&
7363 0 : (EndOfLog < minRecoveryPoint ||
7364 0 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint)))
7365 : {
7366 : /*
7367 : * Ran off end of WAL before reaching end-of-backup WAL record, or
7368 : * minRecoveryPoint. That's usually a bad sign, indicating that you
7369 : * tried to recover from an online backup but never called
7370 : * pg_stop_backup(), or you didn't archive all the WAL up to that
7371 : * point. However, this also happens in crash recovery, if the system
7372 : * crashes while an online backup is in progress. We must not treat
7373 : * that as an error, or the database will refuse to start up.
7374 : */
7375 0 : if (ArchiveRecoveryRequested || ControlFile->backupEndRequired)
7376 : {
7377 0 : if (ControlFile->backupEndRequired)
7378 0 : ereport(FATAL,
7379 : (errmsg("WAL ends before end of online backup"),
7380 : errhint("All WAL generated while online backup was taken must be available at recovery.")));
7381 0 : else if (!XLogRecPtrIsInvalid(ControlFile->backupStartPoint))
7382 0 : ereport(FATAL,
7383 : (errmsg("WAL ends before end of online backup"),
7384 : errhint("Online backup started with pg_start_backup() must be ended with pg_stop_backup(), and all WAL up to that point must be available at recovery.")));
7385 : else
7386 0 : ereport(FATAL,
7387 : (errmsg("WAL ends before consistent recovery point")));
7388 : }
7389 : }
7390 :
7391 : /*
7392 : * Consider whether we need to assign a new timeline ID.
7393 : *
7394 : * If we are doing an archive recovery, we always assign a new ID. This
7395 : * handles a couple of issues. If we stopped short of the end of WAL
7396 : * during recovery, then we are clearly generating a new timeline and must
7397 : * assign it a unique new ID. Even if we ran to the end, modifying the
7398 : * current last segment is problematic because it may result in trying to
7399 : * overwrite an already-archived copy of that segment, and we encourage
7400 : * DBAs to make their archive_commands reject that. We can dodge the
7401 : * problem by making the new active segment have a new timeline ID.
7402 : *
7403 : * In a normal crash recovery, we can just extend the timeline we were in.
7404 : */
7405 3 : PrevTimeLineID = ThisTimeLineID;
7406 3 : if (ArchiveRecoveryRequested)
7407 : {
7408 : char reason[200];
7409 :
7410 0 : Assert(InArchiveRecovery);
7411 :
7412 0 : ThisTimeLineID = findNewestTimeLine(recoveryTargetTLI) + 1;
7413 0 : ereport(LOG,
7414 : (errmsg("selected new timeline ID: %u", ThisTimeLineID)));
7415 :
7416 : /*
7417 : * Create a comment for the history file to explain why and where
7418 : * timeline changed.
7419 : */
7420 0 : if (recoveryTarget == RECOVERY_TARGET_XID)
7421 0 : snprintf(reason, sizeof(reason),
7422 : "%s transaction %u",
7423 0 : recoveryStopAfter ? "after" : "before",
7424 : recoveryStopXid);
7425 0 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
7426 0 : snprintf(reason, sizeof(reason),
7427 : "%s %s\n",
7428 0 : recoveryStopAfter ? "after" : "before",
7429 : timestamptz_to_str(recoveryStopTime));
7430 0 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
7431 0 : snprintf(reason, sizeof(reason),
7432 : "%s LSN %X/%X\n",
7433 0 : recoveryStopAfter ? "after" : "before",
7434 0 : (uint32) (recoveryStopLSN >> 32),
7435 : (uint32) recoveryStopLSN);
7436 0 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
7437 0 : snprintf(reason, sizeof(reason),
7438 : "at restore point \"%s\"",
7439 : recoveryStopName);
7440 0 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
7441 0 : snprintf(reason, sizeof(reason), "reached consistency");
7442 : else
7443 0 : snprintf(reason, sizeof(reason), "no recovery target specified");
7444 :
7445 0 : writeTimeLineHistory(ThisTimeLineID, recoveryTargetTLI,
7446 : EndRecPtr, reason);
7447 : }
7448 :
7449 : /* Save the selected TimeLineID in shared memory, too */
7450 3 : XLogCtl->ThisTimeLineID = ThisTimeLineID;
7451 3 : XLogCtl->PrevTimeLineID = PrevTimeLineID;
7452 :
7453 : /*
7454 : * We are now done reading the old WAL. Turn off archive fetching if it
7455 : * was active, and make a writable copy of the last WAL segment. (Note
7456 : * that we also have a copy of the last block of the old WAL in readBuf;
7457 : * we will use that below.)
7458 : */
7459 3 : if (ArchiveRecoveryRequested)
7460 0 : exitArchiveRecovery(EndOfLogTLI, EndOfLog);
7461 :
7462 : /*
7463 : * Prepare to write WAL starting at EndOfLog location, and init xlog
7464 : * buffer cache using the block containing the last record from the
7465 : * previous incarnation.
7466 : */
7467 3 : Insert = &XLogCtl->Insert;
7468 3 : Insert->PrevBytePos = XLogRecPtrToBytePos(LastRec);
7469 3 : Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
7470 :
7471 : /*
7472 : * Tricky point here: readBuf contains the *last* block that the LastRec
7473 : * record spans, not the one it starts in. The last block is indeed the
7474 : * one we want to use.
7475 : */
7476 3 : if (EndOfLog % XLOG_BLCKSZ != 0)
7477 : {
7478 : char *page;
7479 : int len;
7480 : int firstIdx;
7481 : XLogRecPtr pageBeginPtr;
7482 :
7483 3 : pageBeginPtr = EndOfLog - (EndOfLog % XLOG_BLCKSZ);
7484 3 : Assert(readOff == pageBeginPtr % XLogSegSize);
7485 :
7486 3 : firstIdx = XLogRecPtrToBufIdx(EndOfLog);
7487 :
7488 : /* Copy the valid part of the last block, and zero the rest */
7489 3 : page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
7490 3 : len = EndOfLog % XLOG_BLCKSZ;
7491 3 : memcpy(page, xlogreader->readBuf, len);
7492 3 : memset(page + len, 0, XLOG_BLCKSZ - len);
7493 :
7494 3 : XLogCtl->xlblocks[firstIdx] = pageBeginPtr + XLOG_BLCKSZ;
7495 3 : XLogCtl->InitializedUpTo = pageBeginPtr + XLOG_BLCKSZ;
7496 : }
7497 : else
7498 : {
7499 : /*
7500 : * There is no partial block to copy. Just set InitializedUpTo, and
7501 : * let the first attempt to insert a log record to initialize the next
7502 : * buffer.
7503 : */
7504 0 : XLogCtl->InitializedUpTo = EndOfLog;
7505 : }
7506 :
7507 3 : LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
7508 :
7509 3 : XLogCtl->LogwrtResult = LogwrtResult;
7510 :
7511 3 : XLogCtl->LogwrtRqst.Write = EndOfLog;
7512 3 : XLogCtl->LogwrtRqst.Flush = EndOfLog;
7513 :
7514 : /* Pre-scan prepared transactions to find out the range of XIDs present */
7515 3 : oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
7516 :
7517 : /*
7518 : * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
7519 : * record before resource manager writes cleanup WAL records or checkpoint
7520 : * record is written.
7521 : */
7522 3 : Insert->fullPageWrites = lastFullPageWrites;
7523 3 : LocalSetXLogInsertAllowed();
7524 3 : UpdateFullPageWrites();
7525 3 : LocalXLogInsertAllowed = -1;
7526 :
7527 3 : if (InRecovery)
7528 : {
7529 : /*
7530 : * Perform a checkpoint to update all our recovery activity to disk.
7531 : *
7532 : * Note that we write a shutdown checkpoint rather than an on-line
7533 : * one. This is not particularly critical, but since we may be
7534 : * assigning a new TLI, using a shutdown checkpoint allows us to have
7535 : * the rule that TLI only changes in shutdown checkpoints, which
7536 : * allows some extra error checking in xlog_redo.
7537 : *
7538 : * In fast promotion, only create a lightweight end-of-recovery record
7539 : * instead of a full checkpoint. A checkpoint is requested later,
7540 : * after we're fully out of recovery mode and already accepting
7541 : * queries.
7542 : */
7543 0 : if (bgwriterLaunched)
7544 : {
7545 0 : if (fast_promote)
7546 : {
7547 0 : checkPointLoc = ControlFile->prevCheckPoint;
7548 :
7549 : /*
7550 : * Confirm the last checkpoint is available for us to recover
7551 : * from if we fail. Note that we don't check for the secondary
7552 : * checkpoint since that isn't available in most base backups.
7553 : */
7554 0 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 1, false);
7555 0 : if (record != NULL)
7556 : {
7557 0 : fast_promoted = true;
7558 :
7559 : /*
7560 : * Insert a special WAL record to mark the end of
7561 : * recovery, since we aren't doing a checkpoint. That
7562 : * means that the checkpointer process may likely be in
7563 : * the middle of a time-smoothed restartpoint and could
7564 : * continue to be for minutes after this. That sounds
7565 : * strange, but the effect is roughly the same and it
7566 : * would be stranger to try to come out of the
7567 : * restartpoint and then checkpoint. We request a
7568 : * checkpoint later anyway, just for safety.
7569 : */
7570 0 : CreateEndOfRecoveryRecord();
7571 : }
7572 : }
7573 :
7574 0 : if (!fast_promoted)
7575 0 : RequestCheckpoint(CHECKPOINT_END_OF_RECOVERY |
7576 : CHECKPOINT_IMMEDIATE |
7577 : CHECKPOINT_WAIT);
7578 : }
7579 : else
7580 0 : CreateCheckPoint(CHECKPOINT_END_OF_RECOVERY | CHECKPOINT_IMMEDIATE);
7581 :
7582 : /*
7583 : * And finally, execute the recovery_end_command, if any.
7584 : */
7585 0 : if (recoveryEndCommand)
7586 0 : ExecuteRecoveryCommand(recoveryEndCommand,
7587 : "recovery_end_command",
7588 : true);
7589 : }
7590 :
7591 3 : if (ArchiveRecoveryRequested)
7592 : {
7593 : /*
7594 : * We switched to a new timeline. Clean up segments on the old
7595 : * timeline.
7596 : *
7597 : * If there are any higher-numbered segments on the old timeline,
7598 : * remove them. They might contain valid WAL, but they might also be
7599 : * pre-allocated files containing garbage. In any case, they are not
7600 : * part of the new timeline's history so we don't need them.
7601 : */
7602 0 : RemoveNonParentXlogFiles(EndOfLog, ThisTimeLineID);
7603 :
7604 : /*
7605 : * If the switch happened in the middle of a segment, what to do with
7606 : * the last, partial segment on the old timeline? If we don't archive
7607 : * it, and the server that created the WAL never archives it either
7608 : * (e.g. because it was hit by a meteor), it will never make it to the
7609 : * archive. That's OK from our point of view, because the new segment
7610 : * that we created with the new TLI contains all the WAL from the old
7611 : * timeline up to the switch point. But if you later try to do PITR to
7612 : * the "missing" WAL on the old timeline, recovery won't find it in
7613 : * the archive. It's physically present in the new file with new TLI,
7614 : * but recovery won't look there when it's recovering to the older
7615 : * timeline. On the other hand, if we archive the partial segment, and
7616 : * the original server on that timeline is still running and archives
7617 : * the completed version of the same segment later, it will fail. (We
7618 : * used to do that in 9.4 and below, and it caused such problems).
7619 : *
7620 : * As a compromise, we rename the last segment with the .partial
7621 : * suffix, and archive it. Archive recovery will never try to read
7622 : * .partial segments, so they will normally go unused. But in the odd
7623 : * PITR case, the administrator can copy them manually to the pg_wal
7624 : * directory (removing the suffix). They can be useful in debugging,
7625 : * too.
7626 : *
7627 : * If a .done or .ready file already exists for the old timeline,
7628 : * however, we had already determined that the segment is complete, so
7629 : * we can let it be archived normally. (In particular, if it was
7630 : * restored from the archive to begin with, it's expected to have a
7631 : * .done file).
7632 : */
7633 0 : if (EndOfLog % XLOG_SEG_SIZE != 0 && XLogArchivingActive())
7634 : {
7635 : char origfname[MAXFNAMELEN];
7636 : XLogSegNo endLogSegNo;
7637 :
7638 0 : XLByteToPrevSeg(EndOfLog, endLogSegNo);
7639 0 : XLogFileName(origfname, EndOfLogTLI, endLogSegNo);
7640 :
7641 0 : if (!XLogArchiveIsReadyOrDone(origfname))
7642 : {
7643 : char origpath[MAXPGPATH];
7644 : char partialfname[MAXFNAMELEN];
7645 : char partialpath[MAXPGPATH];
7646 :
7647 0 : XLogFilePath(origpath, EndOfLogTLI, endLogSegNo);
7648 0 : snprintf(partialfname, MAXFNAMELEN, "%s.partial", origfname);
7649 0 : snprintf(partialpath, MAXPGPATH, "%s.partial", origpath);
7650 :
7651 : /*
7652 : * Make sure there's no .done or .ready file for the .partial
7653 : * file.
7654 : */
7655 0 : XLogArchiveCleanup(partialfname);
7656 :
7657 0 : durable_rename(origpath, partialpath, ERROR);
7658 0 : XLogArchiveNotify(partialfname);
7659 : }
7660 : }
7661 : }
7662 :
7663 : /*
7664 : * Preallocate additional log files, if wanted.
7665 : */
7666 3 : PreallocXlogFiles(EndOfLog);
7667 :
7668 : /*
7669 : * Okay, we're officially UP.
7670 : */
7671 3 : InRecovery = false;
7672 :
7673 : /* start the archive_timeout timer and LSN running */
7674 3 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
7675 3 : XLogCtl->lastSegSwitchLSN = EndOfLog;
7676 :
7677 : /* also initialize latestCompletedXid, to nextXid - 1 */
7678 3 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
7679 3 : ShmemVariableCache->latestCompletedXid = ShmemVariableCache->nextXid;
7680 9 : TransactionIdRetreat(ShmemVariableCache->latestCompletedXid);
7681 3 : LWLockRelease(ProcArrayLock);
7682 :
7683 : /*
7684 : * Start up the commit log and subtrans, if not already done for hot
7685 : * standby. (commit timestamps are started below, if necessary.)
7686 : */
7687 3 : if (standbyState == STANDBY_DISABLED)
7688 : {
7689 3 : StartupCLOG();
7690 3 : StartupSUBTRANS(oldestActiveXID);
7691 : }
7692 :
7693 : /*
7694 : * Perform end of recovery actions for any SLRUs that need it.
7695 : */
7696 3 : TrimCLOG();
7697 3 : TrimMultiXact();
7698 :
7699 : /* Reload shared-memory state for prepared transactions */
7700 3 : RecoverPreparedTransactions();
7701 :
7702 : /*
7703 : * Shutdown the recovery environment. This must occur after
7704 : * RecoverPreparedTransactions(), see notes for lock_twophase_recover()
7705 : */
7706 3 : if (standbyState != STANDBY_DISABLED)
7707 0 : ShutdownRecoveryTransactionEnvironment();
7708 :
7709 : /* Shut down xlogreader */
7710 3 : if (readFile >= 0)
7711 : {
7712 3 : close(readFile);
7713 3 : readFile = -1;
7714 : }
7715 3 : XLogReaderFree(xlogreader);
7716 :
7717 : /*
7718 : * If any of the critical GUCs have changed, log them before we allow
7719 : * backends to write WAL.
7720 : */
7721 3 : LocalSetXLogInsertAllowed();
7722 3 : XLogReportParameters();
7723 :
7724 : /*
7725 : * Local WAL inserts enabled, so it's time to finish initialization of
7726 : * commit timestamp.
7727 : */
7728 3 : CompleteCommitTsInitialization();
7729 :
7730 : /*
7731 : * All done with end-of-recovery actions.
7732 : *
7733 : * Now allow backends to write WAL and update the control file status in
7734 : * consequence. The boolean flag allowing backends to write WAL is
7735 : * updated while holding ControlFileLock to prevent other backends to look
7736 : * at an inconsistent state of the control file in shared memory. There
7737 : * is still a small window during which backends can write WAL and the
7738 : * control file is still referring to a system not in DB_IN_PRODUCTION
7739 : * state while looking at the on-disk control file.
7740 : *
7741 : * Also, although the boolean flag to allow WAL is probably atomic in
7742 : * itself, we use the info_lck here to ensure that there are no race
7743 : * conditions concerning visibility of other recent updates to shared
7744 : * memory.
7745 : */
7746 3 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7747 3 : ControlFile->state = DB_IN_PRODUCTION;
7748 3 : ControlFile->time = (pg_time_t) time(NULL);
7749 :
7750 3 : SpinLockAcquire(&XLogCtl->info_lck);
7751 3 : XLogCtl->SharedRecoveryInProgress = false;
7752 3 : SpinLockRelease(&XLogCtl->info_lck);
7753 :
7754 3 : UpdateControlFile();
7755 3 : LWLockRelease(ControlFileLock);
7756 :
7757 : /*
7758 : * If there were cascading standby servers connected to us, nudge any wal
7759 : * sender processes to notice that we've been promoted.
7760 : */
7761 3 : WalSndWakeup();
7762 :
7763 : /*
7764 : * If this was a fast promotion, request an (online) checkpoint now. This
7765 : * isn't required for consistency, but the last restartpoint might be far
7766 : * back, and in case of a crash, recovering from it might take a longer
7767 : * than is appropriate now that we're not in standby mode anymore.
7768 : */
7769 3 : if (fast_promoted)
7770 0 : RequestCheckpoint(CHECKPOINT_FORCE);
7771 3 : }
7772 :
7773 : /*
7774 : * Checks if recovery has reached a consistent state. When consistency is
7775 : * reached and we have a valid starting standby snapshot, tell postmaster
7776 : * that it can start accepting read-only connections.
7777 : */
7778 : static void
7779 0 : CheckRecoveryConsistency(void)
7780 : {
7781 : XLogRecPtr lastReplayedEndRecPtr;
7782 :
7783 : /*
7784 : * During crash recovery, we don't reach a consistent state until we've
7785 : * replayed all the WAL.
7786 : */
7787 0 : if (XLogRecPtrIsInvalid(minRecoveryPoint))
7788 0 : return;
7789 :
7790 : /*
7791 : * assume that we are called in the startup process, and hence don't need
7792 : * a lock to read lastReplayedEndRecPtr
7793 : */
7794 0 : lastReplayedEndRecPtr = XLogCtl->lastReplayedEndRecPtr;
7795 :
7796 : /*
7797 : * Have we reached the point where our base backup was completed?
7798 : */
7799 0 : if (!XLogRecPtrIsInvalid(ControlFile->backupEndPoint) &&
7800 0 : ControlFile->backupEndPoint <= lastReplayedEndRecPtr)
7801 : {
7802 : /*
7803 : * We have reached the end of base backup, as indicated by pg_control.
7804 : * The data on disk is now consistent. Reset backupStartPoint and
7805 : * backupEndPoint, and update minRecoveryPoint to make sure we don't
7806 : * allow starting up at an earlier point even if recovery is stopped
7807 : * and restarted soon after this.
7808 : */
7809 0 : elog(DEBUG1, "end of backup reached");
7810 :
7811 0 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7812 :
7813 0 : if (ControlFile->minRecoveryPoint < lastReplayedEndRecPtr)
7814 0 : ControlFile->minRecoveryPoint = lastReplayedEndRecPtr;
7815 :
7816 0 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
7817 0 : ControlFile->backupEndPoint = InvalidXLogRecPtr;
7818 0 : ControlFile->backupEndRequired = false;
7819 0 : UpdateControlFile();
7820 :
7821 0 : LWLockRelease(ControlFileLock);
7822 : }
7823 :
7824 : /*
7825 : * Have we passed our safe starting point? Note that minRecoveryPoint is
7826 : * known to be incorrectly set if ControlFile->backupEndRequired, until
7827 : * the XLOG_BACKUP_RECORD arrives to advise us of the correct
7828 : * minRecoveryPoint. All we know prior to that is that we're not
7829 : * consistent yet.
7830 : */
7831 0 : if (!reachedConsistency && !ControlFile->backupEndRequired &&
7832 0 : minRecoveryPoint <= lastReplayedEndRecPtr &&
7833 0 : XLogRecPtrIsInvalid(ControlFile->backupStartPoint))
7834 : {
7835 : /*
7836 : * Check to see if the XLOG sequence contained any unresolved
7837 : * references to uninitialized pages.
7838 : */
7839 0 : XLogCheckInvalidPages();
7840 :
7841 0 : reachedConsistency = true;
7842 0 : ereport(LOG,
7843 : (errmsg("consistent recovery state reached at %X/%X",
7844 : (uint32) (lastReplayedEndRecPtr >> 32),
7845 : (uint32) lastReplayedEndRecPtr)));
7846 : }
7847 :
7848 : /*
7849 : * Have we got a valid starting snapshot that will allow queries to be
7850 : * run? If so, we can tell postmaster that the database is consistent now,
7851 : * enabling connections.
7852 : */
7853 0 : if (standbyState == STANDBY_SNAPSHOT_READY &&
7854 0 : !LocalHotStandbyActive &&
7855 0 : reachedConsistency &&
7856 : IsUnderPostmaster)
7857 : {
7858 0 : SpinLockAcquire(&XLogCtl->info_lck);
7859 0 : XLogCtl->SharedHotStandbyActive = true;
7860 0 : SpinLockRelease(&XLogCtl->info_lck);
7861 :
7862 0 : LocalHotStandbyActive = true;
7863 :
7864 0 : SendPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY);
7865 : }
7866 : }
7867 :
7868 : /*
7869 : * Is the system still in recovery?
7870 : *
7871 : * Unlike testing InRecovery, this works in any process that's connected to
7872 : * shared memory.
7873 : *
7874 : * As a side-effect, we initialize the local TimeLineID and RedoRecPtr
7875 : * variables the first time we see that recovery is finished.
7876 : */
7877 : bool
7878 2390619 : RecoveryInProgress(void)
7879 : {
7880 : /*
7881 : * We check shared state each time only until we leave recovery mode. We
7882 : * can't re-enter recovery, so there's no need to keep checking after the
7883 : * shared variable has once been seen false.
7884 : */
7885 2390619 : if (!LocalRecoveryInProgress)
7886 2390278 : return false;
7887 : else
7888 : {
7889 : /*
7890 : * use volatile pointer to make sure we make a fresh read of the
7891 : * shared variable.
7892 : */
7893 341 : volatile XLogCtlData *xlogctl = XLogCtl;
7894 :
7895 341 : LocalRecoveryInProgress = xlogctl->SharedRecoveryInProgress;
7896 :
7897 : /*
7898 : * Initialize TimeLineID and RedoRecPtr when we discover that recovery
7899 : * is finished. InitPostgres() relies upon this behaviour to ensure
7900 : * that InitXLOGAccess() is called at backend startup. (If you change
7901 : * this, see also LocalSetXLogInsertAllowed.)
7902 : */
7903 341 : if (!LocalRecoveryInProgress)
7904 : {
7905 : /*
7906 : * If we just exited recovery, make sure we read TimeLineID and
7907 : * RedoRecPtr after SharedRecoveryInProgress (for machines with
7908 : * weak memory ordering).
7909 : */
7910 341 : pg_memory_barrier();
7911 341 : InitXLOGAccess();
7912 : }
7913 :
7914 : /*
7915 : * Note: We don't need a memory barrier when we're still in recovery.
7916 : * We might exit recovery immediately after return, so the caller
7917 : * can't rely on 'true' meaning that we're still in recovery anyway.
7918 : */
7919 :
7920 341 : return LocalRecoveryInProgress;
7921 : }
7922 : }
7923 :
7924 : /*
7925 : * Is HotStandby active yet? This is only important in special backends
7926 : * since normal backends won't ever be able to connect until this returns
7927 : * true. Postmaster knows this by way of signal, not via shared memory.
7928 : *
7929 : * Unlike testing standbyState, this works in any process that's connected to
7930 : * shared memory. (And note that standbyState alone doesn't tell the truth
7931 : * anyway.)
7932 : */
7933 : bool
7934 0 : HotStandbyActive(void)
7935 : {
7936 : /*
7937 : * We check shared state each time only until Hot Standby is active. We
7938 : * can't de-activate Hot Standby, so there's no need to keep checking
7939 : * after the shared variable has once been seen true.
7940 : */
7941 0 : if (LocalHotStandbyActive)
7942 0 : return true;
7943 : else
7944 : {
7945 : /* spinlock is essential on machines with weak memory ordering! */
7946 0 : SpinLockAcquire(&XLogCtl->info_lck);
7947 0 : LocalHotStandbyActive = XLogCtl->SharedHotStandbyActive;
7948 0 : SpinLockRelease(&XLogCtl->info_lck);
7949 :
7950 0 : return LocalHotStandbyActive;
7951 : }
7952 : }
7953 :
7954 : /*
7955 : * Like HotStandbyActive(), but to be used only in WAL replay code,
7956 : * where we don't need to ask any other process what the state is.
7957 : */
7958 : bool
7959 0 : HotStandbyActiveInReplay(void)
7960 : {
7961 0 : Assert(AmStartupProcess() || !IsPostmasterEnvironment);
7962 0 : return LocalHotStandbyActive;
7963 : }
7964 :
7965 : /*
7966 : * Is this process allowed to insert new WAL records?
7967 : *
7968 : * Ordinarily this is essentially equivalent to !RecoveryInProgress().
7969 : * But we also have provisions for forcing the result "true" or "false"
7970 : * within specific processes regardless of the global state.
7971 : */
7972 : bool
7973 2638911 : XLogInsertAllowed(void)
7974 : {
7975 : /*
7976 : * If value is "unconditionally true" or "unconditionally false", just
7977 : * return it. This provides the normal fast path once recovery is known
7978 : * done.
7979 : */
7980 2638911 : if (LocalXLogInsertAllowed >= 0)
7981 2638702 : return (bool) LocalXLogInsertAllowed;
7982 :
7983 : /*
7984 : * Else, must check to see if we're still in recovery.
7985 : */
7986 209 : if (RecoveryInProgress())
7987 0 : return false;
7988 :
7989 : /*
7990 : * On exit from recovery, reset to "unconditionally true", since there is
7991 : * no need to keep checking.
7992 : */
7993 209 : LocalXLogInsertAllowed = 1;
7994 209 : return true;
7995 : }
7996 :
7997 : /*
7998 : * Make XLogInsertAllowed() return true in the current process only.
7999 : *
8000 : * Note: it is allowed to switch LocalXLogInsertAllowed back to -1 later,
8001 : * and even call LocalSetXLogInsertAllowed() again after that.
8002 : */
8003 : static void
8004 6 : LocalSetXLogInsertAllowed(void)
8005 : {
8006 6 : Assert(LocalXLogInsertAllowed == -1);
8007 6 : LocalXLogInsertAllowed = 1;
8008 :
8009 : /* Initialize as RecoveryInProgress() would do when switching state */
8010 6 : InitXLOGAccess();
8011 6 : }
8012 :
8013 : /*
8014 : * Subroutine to try to fetch and validate a prior checkpoint record.
8015 : *
8016 : * whichChkpt identifies the checkpoint (merely for reporting purposes).
8017 : * 1 for "primary", 2 for "secondary", 0 for "other" (backup_label)
8018 : */
8019 : static XLogRecord *
8020 3 : ReadCheckpointRecord(XLogReaderState *xlogreader, XLogRecPtr RecPtr,
8021 : int whichChkpt, bool report)
8022 : {
8023 : XLogRecord *record;
8024 : uint8 info;
8025 :
8026 3 : if (!XRecOffIsValid(RecPtr))
8027 : {
8028 0 : if (!report)
8029 0 : return NULL;
8030 :
8031 0 : switch (whichChkpt)
8032 : {
8033 : case 1:
8034 0 : ereport(LOG,
8035 : (errmsg("invalid primary checkpoint link in control file")));
8036 0 : break;
8037 : case 2:
8038 0 : ereport(LOG,
8039 : (errmsg("invalid secondary checkpoint link in control file")));
8040 0 : break;
8041 : default:
8042 0 : ereport(LOG,
8043 : (errmsg("invalid checkpoint link in backup_label file")));
8044 0 : break;
8045 : }
8046 0 : return NULL;
8047 : }
8048 :
8049 3 : record = ReadRecord(xlogreader, RecPtr, LOG, true);
8050 :
8051 3 : if (record == NULL)
8052 : {
8053 0 : if (!report)
8054 0 : return NULL;
8055 :
8056 0 : switch (whichChkpt)
8057 : {
8058 : case 1:
8059 0 : ereport(LOG,
8060 : (errmsg("invalid primary checkpoint record")));
8061 0 : break;
8062 : case 2:
8063 0 : ereport(LOG,
8064 : (errmsg("invalid secondary checkpoint record")));
8065 0 : break;
8066 : default:
8067 0 : ereport(LOG,
8068 : (errmsg("invalid checkpoint record")));
8069 0 : break;
8070 : }
8071 0 : return NULL;
8072 : }
8073 3 : if (record->xl_rmid != RM_XLOG_ID)
8074 : {
8075 0 : switch (whichChkpt)
8076 : {
8077 : case 1:
8078 0 : ereport(LOG,
8079 : (errmsg("invalid resource manager ID in primary checkpoint record")));
8080 0 : break;
8081 : case 2:
8082 0 : ereport(LOG,
8083 : (errmsg("invalid resource manager ID in secondary checkpoint record")));
8084 0 : break;
8085 : default:
8086 0 : ereport(LOG,
8087 : (errmsg("invalid resource manager ID in checkpoint record")));
8088 0 : break;
8089 : }
8090 0 : return NULL;
8091 : }
8092 3 : info = record->xl_info & ~XLR_INFO_MASK;
8093 3 : if (info != XLOG_CHECKPOINT_SHUTDOWN &&
8094 : info != XLOG_CHECKPOINT_ONLINE)
8095 : {
8096 0 : switch (whichChkpt)
8097 : {
8098 : case 1:
8099 0 : ereport(LOG,
8100 : (errmsg("invalid xl_info in primary checkpoint record")));
8101 0 : break;
8102 : case 2:
8103 0 : ereport(LOG,
8104 : (errmsg("invalid xl_info in secondary checkpoint record")));
8105 0 : break;
8106 : default:
8107 0 : ereport(LOG,
8108 : (errmsg("invalid xl_info in checkpoint record")));
8109 0 : break;
8110 : }
8111 0 : return NULL;
8112 : }
8113 3 : if (record->xl_tot_len != SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(CheckPoint))
8114 : {
8115 0 : switch (whichChkpt)
8116 : {
8117 : case 1:
8118 0 : ereport(LOG,
8119 : (errmsg("invalid length of primary checkpoint record")));
8120 0 : break;
8121 : case 2:
8122 0 : ereport(LOG,
8123 : (errmsg("invalid length of secondary checkpoint record")));
8124 0 : break;
8125 : default:
8126 0 : ereport(LOG,
8127 : (errmsg("invalid length of checkpoint record")));
8128 0 : break;
8129 : }
8130 0 : return NULL;
8131 : }
8132 3 : return record;
8133 : }
8134 :
8135 : /*
8136 : * This must be called in a backend process before creating WAL records
8137 : * (except in a standalone backend, which does StartupXLOG instead). We need
8138 : * to initialize the local copies of ThisTimeLineID and RedoRecPtr.
8139 : *
8140 : * Note: before Postgres 8.0, we went to some effort to keep the postmaster
8141 : * process's copies of ThisTimeLineID and RedoRecPtr valid too. This was
8142 : * unnecessary however, since the postmaster itself never touches XLOG anyway.
8143 : */
8144 : void
8145 348 : InitXLOGAccess(void)
8146 : {
8147 348 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8148 :
8149 : /* ThisTimeLineID doesn't change so we need no lock to copy it */
8150 348 : ThisTimeLineID = XLogCtl->ThisTimeLineID;
8151 348 : Assert(ThisTimeLineID != 0 || IsBootstrapProcessingMode());
8152 :
8153 : /* Use GetRedoRecPtr to copy the RedoRecPtr safely */
8154 348 : (void) GetRedoRecPtr();
8155 : /* Also update our copy of doPageWrites. */
8156 348 : doPageWrites = (Insert->fullPageWrites || Insert->forcePageWrites);
8157 :
8158 : /* Also initialize the working areas for constructing WAL records */
8159 348 : InitXLogInsert();
8160 348 : }
8161 :
8162 : /*
8163 : * Return the current Redo pointer from shared memory.
8164 : *
8165 : * As a side-effect, the local RedoRecPtr copy is updated.
8166 : */
8167 : XLogRecPtr
8168 758 : GetRedoRecPtr(void)
8169 : {
8170 : XLogRecPtr ptr;
8171 :
8172 : /*
8173 : * The possibly not up-to-date copy in XlogCtl is enough. Even if we
8174 : * grabbed a WAL insertion lock to read the master copy, someone might
8175 : * update it just after we've released the lock.
8176 : */
8177 758 : SpinLockAcquire(&XLogCtl->info_lck);
8178 758 : ptr = XLogCtl->RedoRecPtr;
8179 758 : SpinLockRelease(&XLogCtl->info_lck);
8180 :
8181 758 : if (RedoRecPtr < ptr)
8182 339 : RedoRecPtr = ptr;
8183 :
8184 758 : return RedoRecPtr;
8185 : }
8186 :
8187 : /*
8188 : * Return information needed to decide whether a modified block needs a
8189 : * full-page image to be included in the WAL record.
8190 : *
8191 : * The returned values are cached copies from backend-private memory, and
8192 : * possibly out-of-date. XLogInsertRecord will re-check them against
8193 : * up-to-date values, while holding the WAL insert lock.
8194 : */
8195 : void
8196 1312995 : GetFullPageWriteInfo(XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
8197 : {
8198 1312995 : *RedoRecPtr_p = RedoRecPtr;
8199 1312995 : *doPageWrites_p = doPageWrites;
8200 1312995 : }
8201 :
8202 : /*
8203 : * GetInsertRecPtr -- Returns the current insert position.
8204 : *
8205 : * NOTE: The value *actually* returned is the position of the last full
8206 : * xlog page. It lags behind the real insert position by at most 1 page.
8207 : * For that, we don't need to scan through WAL insertion locks, and an
8208 : * approximation is enough for the current usage of this function.
8209 : */
8210 : XLogRecPtr
8211 4 : GetInsertRecPtr(void)
8212 : {
8213 : XLogRecPtr recptr;
8214 :
8215 4 : SpinLockAcquire(&XLogCtl->info_lck);
8216 4 : recptr = XLogCtl->LogwrtRqst.Write;
8217 4 : SpinLockRelease(&XLogCtl->info_lck);
8218 :
8219 4 : return recptr;
8220 : }
8221 :
8222 : /*
8223 : * GetFlushRecPtr -- Returns the current flush position, ie, the last WAL
8224 : * position known to be fsync'd to disk.
8225 : */
8226 : XLogRecPtr
8227 13 : GetFlushRecPtr(void)
8228 : {
8229 13 : SpinLockAcquire(&XLogCtl->info_lck);
8230 13 : LogwrtResult = XLogCtl->LogwrtResult;
8231 13 : SpinLockRelease(&XLogCtl->info_lck);
8232 :
8233 13 : return LogwrtResult.Flush;
8234 : }
8235 :
8236 : /*
8237 : * GetLastImportantRecPtr -- Returns the LSN of the last important record
8238 : * inserted. All records not explicitly marked as unimportant are considered
8239 : * important.
8240 : *
8241 : * The LSN is determined by computing the maximum of
8242 : * WALInsertLocks[i].lastImportantAt.
8243 : */
8244 : XLogRecPtr
8245 17 : GetLastImportantRecPtr(void)
8246 : {
8247 17 : XLogRecPtr res = InvalidXLogRecPtr;
8248 : int i;
8249 :
8250 153 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
8251 : {
8252 : XLogRecPtr last_important;
8253 :
8254 : /*
8255 : * Need to take a lock to prevent torn reads of the LSN, which are
8256 : * possible on some of the supported platforms. WAL insert locks only
8257 : * support exclusive mode, so we have to use that.
8258 : */
8259 136 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
8260 136 : last_important = WALInsertLocks[i].l.lastImportantAt;
8261 136 : LWLockRelease(&WALInsertLocks[i].l.lock);
8262 :
8263 136 : if (res < last_important)
8264 33 : res = last_important;
8265 : }
8266 :
8267 17 : return res;
8268 : }
8269 :
8270 : /*
8271 : * Get the time and LSN of the last xlog segment switch
8272 : */
8273 : pg_time_t
8274 0 : GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
8275 : {
8276 : pg_time_t result;
8277 :
8278 : /* Need WALWriteLock, but shared lock is sufficient */
8279 0 : LWLockAcquire(WALWriteLock, LW_SHARED);
8280 0 : result = XLogCtl->lastSegSwitchTime;
8281 0 : *lastSwitchLSN = XLogCtl->lastSegSwitchLSN;
8282 0 : LWLockRelease(WALWriteLock);
8283 :
8284 0 : return result;
8285 : }
8286 :
8287 : /*
8288 : * GetNextXidAndEpoch - get the current nextXid value and associated epoch
8289 : *
8290 : * This is exported for use by code that would like to have 64-bit XIDs.
8291 : * We don't really support such things, but all XIDs within the system
8292 : * can be presumed "close to" the result, and thus the epoch associated
8293 : * with them can be determined.
8294 : */
8295 : void
8296 298 : GetNextXidAndEpoch(TransactionId *xid, uint32 *epoch)
8297 : {
8298 : uint32 ckptXidEpoch;
8299 : TransactionId ckptXid;
8300 : TransactionId nextXid;
8301 :
8302 : /* Must read checkpoint info first, else have race condition */
8303 298 : SpinLockAcquire(&XLogCtl->info_lck);
8304 298 : ckptXidEpoch = XLogCtl->ckptXidEpoch;
8305 298 : ckptXid = XLogCtl->ckptXid;
8306 298 : SpinLockRelease(&XLogCtl->info_lck);
8307 :
8308 : /* Now fetch current nextXid */
8309 298 : nextXid = ReadNewTransactionId();
8310 :
8311 : /*
8312 : * nextXid is certainly logically later than ckptXid. So if it's
8313 : * numerically less, it must have wrapped into the next epoch.
8314 : */
8315 298 : if (nextXid < ckptXid)
8316 0 : ckptXidEpoch++;
8317 :
8318 298 : *xid = nextXid;
8319 298 : *epoch = ckptXidEpoch;
8320 298 : }
8321 :
8322 : /*
8323 : * This must be called ONCE during postmaster or standalone-backend shutdown
8324 : */
8325 : void
8326 3 : ShutdownXLOG(int code, Datum arg)
8327 : {
8328 : /* Don't be chatty in standalone mode */
8329 3 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
8330 : (errmsg("shutting down")));
8331 :
8332 : /*
8333 : * Signal walsenders to move to stopping state.
8334 : */
8335 3 : WalSndInitStopping();
8336 :
8337 : /*
8338 : * Wait for WAL senders to be in stopping state. This prevents commands
8339 : * from writing new WAL.
8340 : */
8341 3 : WalSndWaitStopping();
8342 :
8343 3 : if (RecoveryInProgress())
8344 0 : CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
8345 : else
8346 : {
8347 : /*
8348 : * If archiving is enabled, rotate the last XLOG file so that all the
8349 : * remaining records are archived (postmaster wakes up the archiver
8350 : * process one more time at the end of shutdown). The checkpoint
8351 : * record will go to the next XLOG file and won't be archived (yet).
8352 : */
8353 3 : if (XLogArchivingActive() && XLogArchiveCommandSet())
8354 0 : RequestXLogSwitch(false);
8355 :
8356 3 : CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
8357 : }
8358 3 : ShutdownCLOG();
8359 3 : ShutdownCommitTs();
8360 3 : ShutdownSUBTRANS();
8361 3 : ShutdownMultiXact();
8362 3 : }
8363 :
8364 : /*
8365 : * Log start of a checkpoint.
8366 : */
8367 : static void
8368 5 : LogCheckpointStart(int flags, bool restartpoint)
8369 : {
8370 5 : elog(LOG, "%s starting:%s%s%s%s%s%s%s%s",
8371 : restartpoint ? "restartpoint" : "checkpoint",
8372 : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
8373 : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
8374 : (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
8375 : (flags & CHECKPOINT_FORCE) ? " force" : "",
8376 : (flags & CHECKPOINT_WAIT) ? " wait" : "",
8377 : (flags & CHECKPOINT_CAUSE_XLOG) ? " xlog" : "",
8378 : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
8379 : (flags & CHECKPOINT_FLUSH_ALL) ? " flush-all" : "");
8380 5 : }
8381 :
8382 : /*
8383 : * Log end of a checkpoint.
8384 : */
8385 : static void
8386 11 : LogCheckpointEnd(bool restartpoint)
8387 : {
8388 : long write_secs,
8389 : sync_secs,
8390 : total_secs,
8391 : longest_secs,
8392 : average_secs;
8393 : int write_usecs,
8394 : sync_usecs,
8395 : total_usecs,
8396 : longest_usecs,
8397 : average_usecs;
8398 : uint64 average_sync_time;
8399 :
8400 11 : CheckpointStats.ckpt_end_t = GetCurrentTimestamp();
8401 :
8402 11 : TimestampDifference(CheckpointStats.ckpt_write_t,
8403 : CheckpointStats.ckpt_sync_t,
8404 : &write_secs, &write_usecs);
8405 :
8406 11 : TimestampDifference(CheckpointStats.ckpt_sync_t,
8407 : CheckpointStats.ckpt_sync_end_t,
8408 : &sync_secs, &sync_usecs);
8409 :
8410 : /* Accumulate checkpoint timing summary data, in milliseconds. */
8411 22 : BgWriterStats.m_checkpoint_write_time +=
8412 11 : write_secs * 1000 + write_usecs / 1000;
8413 22 : BgWriterStats.m_checkpoint_sync_time +=
8414 11 : sync_secs * 1000 + sync_usecs / 1000;
8415 :
8416 : /*
8417 : * All of the published timing statistics are accounted for. Only
8418 : * continue if a log message is to be written.
8419 : */
8420 11 : if (!log_checkpoints)
8421 17 : return;
8422 :
8423 5 : TimestampDifference(CheckpointStats.ckpt_start_t,
8424 : CheckpointStats.ckpt_end_t,
8425 : &total_secs, &total_usecs);
8426 :
8427 : /*
8428 : * Timing values returned from CheckpointStats are in microseconds.
8429 : * Convert to the second plus microsecond form that TimestampDifference
8430 : * returns for homogeneous printing.
8431 : */
8432 5 : longest_secs = (long) (CheckpointStats.ckpt_longest_sync / 1000000);
8433 10 : longest_usecs = CheckpointStats.ckpt_longest_sync -
8434 5 : (uint64) longest_secs * 1000000;
8435 :
8436 5 : average_sync_time = 0;
8437 5 : if (CheckpointStats.ckpt_sync_rels > 0)
8438 0 : average_sync_time = CheckpointStats.ckpt_agg_sync_time /
8439 0 : CheckpointStats.ckpt_sync_rels;
8440 5 : average_secs = (long) (average_sync_time / 1000000);
8441 5 : average_usecs = average_sync_time - (uint64) average_secs * 1000000;
8442 :
8443 5 : elog(LOG, "%s complete: wrote %d buffers (%.1f%%); "
8444 : "%d WAL file(s) added, %d removed, %d recycled; "
8445 : "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s; "
8446 : "sync files=%d, longest=%ld.%03d s, average=%ld.%03d s; "
8447 : "distance=%d kB, estimate=%d kB",
8448 : restartpoint ? "restartpoint" : "checkpoint",
8449 : CheckpointStats.ckpt_bufs_written,
8450 : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
8451 : CheckpointStats.ckpt_segs_added,
8452 : CheckpointStats.ckpt_segs_removed,
8453 : CheckpointStats.ckpt_segs_recycled,
8454 : write_secs, write_usecs / 1000,
8455 : sync_secs, sync_usecs / 1000,
8456 : total_secs, total_usecs / 1000,
8457 : CheckpointStats.ckpt_sync_rels,
8458 : longest_secs, longest_usecs / 1000,
8459 : average_secs, average_usecs / 1000,
8460 : (int) (PrevCheckPointDistance / 1024.0),
8461 : (int) (CheckPointDistanceEstimate / 1024.0));
8462 : }
8463 :
8464 : /*
8465 : * Update the estimate of distance between checkpoints.
8466 : *
8467 : * The estimate is used to calculate the number of WAL segments to keep
8468 : * preallocated, see XLOGFileSlop().
8469 : */
8470 : static void
8471 11 : UpdateCheckPointDistanceEstimate(uint64 nbytes)
8472 : {
8473 : /*
8474 : * To estimate the number of segments consumed between checkpoints, keep a
8475 : * moving average of the amount of WAL generated in previous checkpoint
8476 : * cycles. However, if the load is bursty, with quiet periods and busy
8477 : * periods, we want to cater for the peak load. So instead of a plain
8478 : * moving average, let the average decline slowly if the previous cycle
8479 : * used less WAL than estimated, but bump it up immediately if it used
8480 : * more.
8481 : *
8482 : * When checkpoints are triggered by max_wal_size, this should converge to
8483 : * CheckpointSegments * XLOG_SEG_SIZE,
8484 : *
8485 : * Note: This doesn't pay any attention to what caused the checkpoint.
8486 : * Checkpoints triggered manually with CHECKPOINT command, or by e.g.
8487 : * starting a base backup, are counted the same as those created
8488 : * automatically. The slow-decline will largely mask them out, if they are
8489 : * not frequent. If they are frequent, it seems reasonable to count them
8490 : * in as any others; if you issue a manual checkpoint every 5 minutes and
8491 : * never let a timed checkpoint happen, it makes sense to base the
8492 : * preallocation on that 5 minute interval rather than whatever
8493 : * checkpoint_timeout is set to.
8494 : */
8495 11 : PrevCheckPointDistance = nbytes;
8496 11 : if (CheckPointDistanceEstimate < nbytes)
8497 5 : CheckPointDistanceEstimate = nbytes;
8498 : else
8499 12 : CheckPointDistanceEstimate =
8500 6 : (0.90 * CheckPointDistanceEstimate + 0.10 * (double) nbytes);
8501 11 : }
8502 :
8503 : /*
8504 : * Perform a checkpoint --- either during shutdown, or on-the-fly
8505 : *
8506 : * flags is a bitwise OR of the following:
8507 : * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
8508 : * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
8509 : * CHECKPOINT_IMMEDIATE: finish the checkpoint ASAP,
8510 : * ignoring checkpoint_completion_target parameter.
8511 : * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
8512 : * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
8513 : * CHECKPOINT_END_OF_RECOVERY).
8514 : * CHECKPOINT_FLUSH_ALL: also flush buffers of unlogged tables.
8515 : *
8516 : * Note: flags contains other bits, of interest here only for logging purposes.
8517 : * In particular note that this routine is synchronous and does not pay
8518 : * attention to CHECKPOINT_WAIT.
8519 : *
8520 : * If !shutdown then we are writing an online checkpoint. This is a very special
8521 : * kind of operation and WAL record because the checkpoint action occurs over
8522 : * a period of time yet logically occurs at just a single LSN. The logical
8523 : * position of the WAL record (redo ptr) is the same or earlier than the
8524 : * physical position. When we replay WAL we locate the checkpoint via its
8525 : * physical position then read the redo ptr and actually start replay at the
8526 : * earlier logical position. Note that we don't write *anything* to WAL at
8527 : * the logical position, so that location could be any other kind of WAL record.
8528 : * All of this mechanism allows us to continue working while we checkpoint.
8529 : * As a result, timing of actions is critical here and be careful to note that
8530 : * this function will likely take minutes to execute on a busy system.
8531 : */
8532 : void
8533 11 : CreateCheckPoint(int flags)
8534 : {
8535 : bool shutdown;
8536 : CheckPoint checkPoint;
8537 : XLogRecPtr recptr;
8538 11 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8539 : uint32 freespace;
8540 : XLogRecPtr PriorRedoPtr;
8541 : XLogRecPtr curInsert;
8542 : XLogRecPtr last_important_lsn;
8543 : VirtualTransactionId *vxids;
8544 : int nvxids;
8545 :
8546 : /*
8547 : * An end-of-recovery checkpoint is really a shutdown checkpoint, just
8548 : * issued at a different time.
8549 : */
8550 11 : if (flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY))
8551 3 : shutdown = true;
8552 : else
8553 8 : shutdown = false;
8554 :
8555 : /* sanity check */
8556 11 : if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
8557 0 : elog(ERROR, "can't create a checkpoint during recovery");
8558 :
8559 : /*
8560 : * Initialize InitXLogInsert working areas before entering the critical
8561 : * section. Normally, this is done by the first call to
8562 : * RecoveryInProgress() or LocalSetXLogInsertAllowed(), but when creating
8563 : * an end-of-recovery checkpoint, the LocalSetXLogInsertAllowed call is
8564 : * done below in a critical section, and InitXLogInsert cannot be called
8565 : * in a critical section.
8566 : */
8567 11 : InitXLogInsert();
8568 :
8569 : /*
8570 : * Acquire CheckpointLock to ensure only one checkpoint happens at a time.
8571 : * (This is just pro forma, since in the present system structure there is
8572 : * only one process that is allowed to issue checkpoints at any given
8573 : * time.)
8574 : */
8575 11 : LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
8576 :
8577 : /*
8578 : * Prepare to accumulate statistics.
8579 : *
8580 : * Note: because it is possible for log_checkpoints to change while a
8581 : * checkpoint proceeds, we always accumulate stats, even if
8582 : * log_checkpoints is currently off.
8583 : */
8584 11 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
8585 11 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
8586 :
8587 : /*
8588 : * Use a critical section to force system panic if we have trouble.
8589 : */
8590 11 : START_CRIT_SECTION();
8591 :
8592 11 : if (shutdown)
8593 : {
8594 3 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8595 3 : ControlFile->state = DB_SHUTDOWNING;
8596 3 : ControlFile->time = (pg_time_t) time(NULL);
8597 3 : UpdateControlFile();
8598 3 : LWLockRelease(ControlFileLock);
8599 : }
8600 :
8601 : /*
8602 : * Let smgr prepare for checkpoint; this has to happen before we determine
8603 : * the REDO pointer. Note that smgr must not do anything that'd have to
8604 : * be undone if we decide no checkpoint is needed.
8605 : */
8606 11 : smgrpreckpt();
8607 :
8608 : /* Begin filling in the checkpoint WAL record */
8609 11 : MemSet(&checkPoint, 0, sizeof(checkPoint));
8610 11 : checkPoint.time = (pg_time_t) time(NULL);
8611 :
8612 : /*
8613 : * For Hot Standby, derive the oldestActiveXid before we fix the redo
8614 : * pointer. This allows us to begin accumulating changes to assemble our
8615 : * starting snapshot of locks and transactions.
8616 : */
8617 11 : if (!shutdown && XLogStandbyInfoActive())
8618 8 : checkPoint.oldestActiveXid = GetOldestActiveTransactionId();
8619 : else
8620 3 : checkPoint.oldestActiveXid = InvalidTransactionId;
8621 :
8622 : /*
8623 : * Get location of last important record before acquiring insert locks (as
8624 : * GetLastImportantRecPtr() also locks WAL locks).
8625 : */
8626 11 : last_important_lsn = GetLastImportantRecPtr();
8627 :
8628 : /*
8629 : * We must block concurrent insertions while examining insert state to
8630 : * determine the checkpoint REDO pointer.
8631 : */
8632 11 : WALInsertLockAcquireExclusive();
8633 11 : curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
8634 :
8635 : /*
8636 : * If this isn't a shutdown or forced checkpoint, and if there has been no
8637 : * WAL activity requiring a checkpoint, skip it. The idea here is to
8638 : * avoid inserting duplicate checkpoints when the system is idle.
8639 : */
8640 11 : if ((flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY |
8641 : CHECKPOINT_FORCE)) == 0)
8642 : {
8643 0 : if (last_important_lsn == ControlFile->checkPoint)
8644 : {
8645 0 : WALInsertLockRelease();
8646 0 : LWLockRelease(CheckpointLock);
8647 0 : END_CRIT_SECTION();
8648 0 : ereport(DEBUG1,
8649 : (errmsg("checkpoint skipped due to an idle system")));
8650 11 : return;
8651 : }
8652 : }
8653 :
8654 : /*
8655 : * An end-of-recovery checkpoint is created before anyone is allowed to
8656 : * write WAL. To allow us to write the checkpoint record, temporarily
8657 : * enable XLogInsertAllowed. (This also ensures ThisTimeLineID is
8658 : * initialized, which we need here and in AdvanceXLInsertBuffer.)
8659 : */
8660 11 : if (flags & CHECKPOINT_END_OF_RECOVERY)
8661 0 : LocalSetXLogInsertAllowed();
8662 :
8663 11 : checkPoint.ThisTimeLineID = ThisTimeLineID;
8664 11 : if (flags & CHECKPOINT_END_OF_RECOVERY)
8665 0 : checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
8666 : else
8667 11 : checkPoint.PrevTimeLineID = ThisTimeLineID;
8668 :
8669 11 : checkPoint.fullPageWrites = Insert->fullPageWrites;
8670 :
8671 : /*
8672 : * Compute new REDO record ptr = location of next XLOG record.
8673 : *
8674 : * NB: this is NOT necessarily where the checkpoint record itself will be,
8675 : * since other backends may insert more XLOG records while we're off doing
8676 : * the buffer flush work. Those XLOG records are logically after the
8677 : * checkpoint, even though physically before it. Got that?
8678 : */
8679 11 : freespace = INSERT_FREESPACE(curInsert);
8680 11 : if (freespace == 0)
8681 : {
8682 0 : if (curInsert % XLogSegSize == 0)
8683 0 : curInsert += SizeOfXLogLongPHD;
8684 : else
8685 0 : curInsert += SizeOfXLogShortPHD;
8686 : }
8687 11 : checkPoint.redo = curInsert;
8688 :
8689 : /*
8690 : * Here we update the shared RedoRecPtr for future XLogInsert calls; this
8691 : * must be done while holding all the insertion locks.
8692 : *
8693 : * Note: if we fail to complete the checkpoint, RedoRecPtr will be left
8694 : * pointing past where it really needs to point. This is okay; the only
8695 : * consequence is that XLogInsert might back up whole buffers that it
8696 : * didn't really need to. We can't postpone advancing RedoRecPtr because
8697 : * XLogInserts that happen while we are dumping buffers must assume that
8698 : * their buffer changes are not included in the checkpoint.
8699 : */
8700 11 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
8701 :
8702 : /*
8703 : * Now we can release the WAL insertion locks, allowing other xacts to
8704 : * proceed while we are flushing disk buffers.
8705 : */
8706 11 : WALInsertLockRelease();
8707 :
8708 : /* Update the info_lck-protected copy of RedoRecPtr as well */
8709 11 : SpinLockAcquire(&XLogCtl->info_lck);
8710 11 : XLogCtl->RedoRecPtr = checkPoint.redo;
8711 11 : SpinLockRelease(&XLogCtl->info_lck);
8712 :
8713 : /*
8714 : * If enabled, log checkpoint start. We postpone this until now so as not
8715 : * to log anything if we decided to skip the checkpoint.
8716 : */
8717 11 : if (log_checkpoints)
8718 5 : LogCheckpointStart(flags, false);
8719 :
8720 : TRACE_POSTGRESQL_CHECKPOINT_START(flags);
8721 :
8722 : /*
8723 : * Get the other info we need for the checkpoint record.
8724 : *
8725 : * We don't need to save oldestClogXid in the checkpoint, it only matters
8726 : * for the short period in which clog is being truncated, and if we crash
8727 : * during that we'll redo the clog truncation and fix up oldestClogXid
8728 : * there.
8729 : */
8730 11 : LWLockAcquire(XidGenLock, LW_SHARED);
8731 11 : checkPoint.nextXid = ShmemVariableCache->nextXid;
8732 11 : checkPoint.oldestXid = ShmemVariableCache->oldestXid;
8733 11 : checkPoint.oldestXidDB = ShmemVariableCache->oldestXidDB;
8734 11 : LWLockRelease(XidGenLock);
8735 :
8736 11 : LWLockAcquire(CommitTsLock, LW_SHARED);
8737 11 : checkPoint.oldestCommitTsXid = ShmemVariableCache->oldestCommitTsXid;
8738 11 : checkPoint.newestCommitTsXid = ShmemVariableCache->newestCommitTsXid;
8739 11 : LWLockRelease(CommitTsLock);
8740 :
8741 : /* Increase XID epoch if we've wrapped around since last checkpoint */
8742 11 : checkPoint.nextXidEpoch = ControlFile->checkPointCopy.nextXidEpoch;
8743 11 : if (checkPoint.nextXid < ControlFile->checkPointCopy.nextXid)
8744 0 : checkPoint.nextXidEpoch++;
8745 :
8746 11 : LWLockAcquire(OidGenLock, LW_SHARED);
8747 11 : checkPoint.nextOid = ShmemVariableCache->nextOid;
8748 11 : if (!shutdown)
8749 8 : checkPoint.nextOid += ShmemVariableCache->oidCount;
8750 11 : LWLockRelease(OidGenLock);
8751 :
8752 11 : MultiXactGetCheckptMulti(shutdown,
8753 : &checkPoint.nextMulti,
8754 : &checkPoint.nextMultiOffset,
8755 : &checkPoint.oldestMulti,
8756 : &checkPoint.oldestMultiDB);
8757 :
8758 : /*
8759 : * Having constructed the checkpoint record, ensure all shmem disk buffers
8760 : * and commit-log buffers are flushed to disk.
8761 : *
8762 : * This I/O could fail for various reasons. If so, we will fail to
8763 : * complete the checkpoint, but there is no reason to force a system
8764 : * panic. Accordingly, exit critical section while doing it.
8765 : */
8766 11 : END_CRIT_SECTION();
8767 :
8768 : /*
8769 : * In some cases there are groups of actions that must all occur on one
8770 : * side or the other of a checkpoint record. Before flushing the
8771 : * checkpoint record we must explicitly wait for any backend currently
8772 : * performing those groups of actions.
8773 : *
8774 : * One example is end of transaction, so we must wait for any transactions
8775 : * that are currently in commit critical sections. If an xact inserted
8776 : * its commit record into XLOG just before the REDO point, then a crash
8777 : * restart from the REDO point would not replay that record, which means
8778 : * that our flushing had better include the xact's update of pg_xact. So
8779 : * we wait till he's out of his commit critical section before proceeding.
8780 : * See notes in RecordTransactionCommit().
8781 : *
8782 : * Because we've already released the insertion locks, this test is a bit
8783 : * fuzzy: it is possible that we will wait for xacts we didn't really need
8784 : * to wait for. But the delay should be short and it seems better to make
8785 : * checkpoint take a bit longer than to hold off insertions longer than
8786 : * necessary. (In fact, the whole reason we have this issue is that xact.c
8787 : * does commit record XLOG insertion and clog update as two separate steps
8788 : * protected by different locks, but again that seems best on grounds of
8789 : * minimizing lock contention.)
8790 : *
8791 : * A transaction that has not yet set delayChkpt when we look cannot be at
8792 : * risk, since he's not inserted his commit record yet; and one that's
8793 : * already cleared it is not at risk either, since he's done fixing clog
8794 : * and we will correctly flush the update below. So we cannot miss any
8795 : * xacts we need to wait for.
8796 : */
8797 11 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids);
8798 11 : if (nvxids > 0)
8799 : {
8800 : do
8801 : {
8802 0 : pg_usleep(10000L); /* wait for 10 msec */
8803 0 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids));
8804 : }
8805 11 : pfree(vxids);
8806 :
8807 11 : CheckPointGuts(checkPoint.redo, flags);
8808 :
8809 : /*
8810 : * Take a snapshot of running transactions and write this to WAL. This
8811 : * allows us to reconstruct the state of running transactions during
8812 : * archive recovery, if required. Skip, if this info disabled.
8813 : *
8814 : * If we are shutting down, or Startup process is completing crash
8815 : * recovery we don't need to write running xact data.
8816 : */
8817 11 : if (!shutdown && XLogStandbyInfoActive())
8818 8 : LogStandbySnapshot();
8819 :
8820 11 : START_CRIT_SECTION();
8821 :
8822 : /*
8823 : * Now insert the checkpoint record into XLOG.
8824 : */
8825 11 : XLogBeginInsert();
8826 11 : XLogRegisterData((char *) (&checkPoint), sizeof(checkPoint));
8827 11 : recptr = XLogInsert(RM_XLOG_ID,
8828 : shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
8829 : XLOG_CHECKPOINT_ONLINE);
8830 :
8831 11 : XLogFlush(recptr);
8832 :
8833 : /*
8834 : * We mustn't write any new WAL after a shutdown checkpoint, or it will be
8835 : * overwritten at next startup. No-one should even try, this just allows
8836 : * sanity-checking. In the case of an end-of-recovery checkpoint, we want
8837 : * to just temporarily disable writing until the system has exited
8838 : * recovery.
8839 : */
8840 11 : if (shutdown)
8841 : {
8842 3 : if (flags & CHECKPOINT_END_OF_RECOVERY)
8843 0 : LocalXLogInsertAllowed = -1; /* return to "check" state */
8844 : else
8845 3 : LocalXLogInsertAllowed = 0; /* never again write WAL */
8846 : }
8847 :
8848 : /*
8849 : * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
8850 : * = end of actual checkpoint record.
8851 : */
8852 11 : if (shutdown && checkPoint.redo != ProcLastRecPtr)
8853 0 : ereport(PANIC,
8854 : (errmsg("concurrent write-ahead log activity while database system is shutting down")));
8855 :
8856 : /*
8857 : * Remember the prior checkpoint's redo pointer, used later to determine
8858 : * the point where the log can be truncated.
8859 : */
8860 11 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
8861 :
8862 : /*
8863 : * Update the control file.
8864 : */
8865 11 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8866 11 : if (shutdown)
8867 3 : ControlFile->state = DB_SHUTDOWNED;
8868 11 : ControlFile->prevCheckPoint = ControlFile->checkPoint;
8869 11 : ControlFile->checkPoint = ProcLastRecPtr;
8870 11 : ControlFile->checkPointCopy = checkPoint;
8871 11 : ControlFile->time = (pg_time_t) time(NULL);
8872 : /* crash recovery should always recover to the end of WAL */
8873 11 : ControlFile->minRecoveryPoint = InvalidXLogRecPtr;
8874 11 : ControlFile->minRecoveryPointTLI = 0;
8875 :
8876 : /*
8877 : * Persist unloggedLSN value. It's reset on crash recovery, so this goes
8878 : * unused on non-shutdown checkpoints, but seems useful to store it always
8879 : * for debugging purposes.
8880 : */
8881 11 : SpinLockAcquire(&XLogCtl->ulsn_lck);
8882 11 : ControlFile->unloggedLSN = XLogCtl->unloggedLSN;
8883 11 : SpinLockRelease(&XLogCtl->ulsn_lck);
8884 :
8885 11 : UpdateControlFile();
8886 11 : LWLockRelease(ControlFileLock);
8887 :
8888 : /* Update shared-memory copy of checkpoint XID/epoch */
8889 11 : SpinLockAcquire(&XLogCtl->info_lck);
8890 11 : XLogCtl->ckptXidEpoch = checkPoint.nextXidEpoch;
8891 11 : XLogCtl->ckptXid = checkPoint.nextXid;
8892 11 : SpinLockRelease(&XLogCtl->info_lck);
8893 :
8894 : /*
8895 : * We are now done with critical updates; no need for system panic if we
8896 : * have trouble while fooling with old log segments.
8897 : */
8898 11 : END_CRIT_SECTION();
8899 :
8900 : /*
8901 : * Let smgr do post-checkpoint cleanup (eg, deleting old files).
8902 : */
8903 11 : smgrpostckpt();
8904 :
8905 : /*
8906 : * Delete old log files (those no longer needed even for previous
8907 : * checkpoint or the standbys in XLOG streaming).
8908 : */
8909 11 : if (PriorRedoPtr != InvalidXLogRecPtr)
8910 : {
8911 : XLogSegNo _logSegNo;
8912 :
8913 : /* Update the average distance between checkpoints. */
8914 11 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
8915 :
8916 11 : XLByteToSeg(PriorRedoPtr, _logSegNo);
8917 11 : KeepLogSeg(recptr, &_logSegNo);
8918 11 : _logSegNo--;
8919 11 : RemoveOldXlogFiles(_logSegNo, PriorRedoPtr, recptr);
8920 : }
8921 :
8922 : /*
8923 : * Make more log segments if needed. (Do this after recycling old log
8924 : * segments, since that may supply some of the needed files.)
8925 : */
8926 11 : if (!shutdown)
8927 8 : PreallocXlogFiles(recptr);
8928 :
8929 : /*
8930 : * Truncate pg_subtrans if possible. We can throw away all data before
8931 : * the oldest XMIN of any running transaction. No future transaction will
8932 : * attempt to reference any pg_subtrans entry older than that (see Asserts
8933 : * in subtrans.c). During recovery, though, we mustn't do this because
8934 : * StartupSUBTRANS hasn't been called yet.
8935 : */
8936 11 : if (!RecoveryInProgress())
8937 11 : TruncateSUBTRANS(GetOldestXmin(NULL, PROCARRAY_FLAGS_DEFAULT));
8938 :
8939 : /* Real work is done, but log and update stats before releasing lock. */
8940 11 : LogCheckpointEnd(false);
8941 :
8942 : TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
8943 : NBuffers,
8944 : CheckpointStats.ckpt_segs_added,
8945 : CheckpointStats.ckpt_segs_removed,
8946 : CheckpointStats.ckpt_segs_recycled);
8947 :
8948 11 : LWLockRelease(CheckpointLock);
8949 : }
8950 :
8951 : /*
8952 : * Mark the end of recovery in WAL though without running a full checkpoint.
8953 : * We can expect that a restartpoint is likely to be in progress as we
8954 : * do this, though we are unwilling to wait for it to complete. So be
8955 : * careful to avoid taking the CheckpointLock anywhere here.
8956 : *
8957 : * CreateRestartPoint() allows for the case where recovery may end before
8958 : * the restartpoint completes so there is no concern of concurrent behaviour.
8959 : */
8960 : static void
8961 0 : CreateEndOfRecoveryRecord(void)
8962 : {
8963 : xl_end_of_recovery xlrec;
8964 : XLogRecPtr recptr;
8965 :
8966 : /* sanity check */
8967 0 : if (!RecoveryInProgress())
8968 0 : elog(ERROR, "can only be used to end recovery");
8969 :
8970 0 : xlrec.end_time = GetCurrentTimestamp();
8971 :
8972 0 : WALInsertLockAcquireExclusive();
8973 0 : xlrec.ThisTimeLineID = ThisTimeLineID;
8974 0 : xlrec.PrevTimeLineID = XLogCtl->PrevTimeLineID;
8975 0 : WALInsertLockRelease();
8976 :
8977 0 : LocalSetXLogInsertAllowed();
8978 :
8979 0 : START_CRIT_SECTION();
8980 :
8981 0 : XLogBeginInsert();
8982 0 : XLogRegisterData((char *) &xlrec, sizeof(xl_end_of_recovery));
8983 0 : recptr = XLogInsert(RM_XLOG_ID, XLOG_END_OF_RECOVERY);
8984 :
8985 0 : XLogFlush(recptr);
8986 :
8987 : /*
8988 : * Update the control file so that crash recovery can follow the timeline
8989 : * changes to this point.
8990 : */
8991 0 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8992 0 : ControlFile->time = (pg_time_t) time(NULL);
8993 0 : ControlFile->minRecoveryPoint = recptr;
8994 0 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
8995 0 : UpdateControlFile();
8996 0 : LWLockRelease(ControlFileLock);
8997 :
8998 0 : END_CRIT_SECTION();
8999 :
9000 0 : LocalXLogInsertAllowed = -1; /* return to "check" state */
9001 0 : }
9002 :
9003 : /*
9004 : * Flush all data in shared memory to disk, and fsync
9005 : *
9006 : * This is the common code shared between regular checkpoints and
9007 : * recovery restartpoints.
9008 : */
9009 : static void
9010 11 : CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
9011 : {
9012 11 : CheckPointCLOG();
9013 11 : CheckPointCommitTs();
9014 11 : CheckPointSUBTRANS();
9015 11 : CheckPointMultiXact();
9016 11 : CheckPointPredicate();
9017 11 : CheckPointRelationMap();
9018 11 : CheckPointReplicationSlots();
9019 11 : CheckPointSnapBuild();
9020 11 : CheckPointLogicalRewriteHeap();
9021 11 : CheckPointBuffers(flags); /* performs all required fsyncs */
9022 11 : CheckPointReplicationOrigin();
9023 : /* We deliberately delay 2PC checkpointing as long as possible */
9024 11 : CheckPointTwoPhase(checkPointRedo);
9025 11 : }
9026 :
9027 : /*
9028 : * Save a checkpoint for recovery restart if appropriate
9029 : *
9030 : * This function is called each time a checkpoint record is read from XLOG.
9031 : * It must determine whether the checkpoint represents a safe restartpoint or
9032 : * not. If so, the checkpoint record is stashed in shared memory so that
9033 : * CreateRestartPoint can consult it. (Note that the latter function is
9034 : * executed by the checkpointer, while this one will be executed by the
9035 : * startup process.)
9036 : */
9037 : static void
9038 0 : RecoveryRestartPoint(const CheckPoint *checkPoint)
9039 : {
9040 : /*
9041 : * Also refrain from creating a restartpoint if we have seen any
9042 : * references to non-existent pages. Restarting recovery from the
9043 : * restartpoint would not see the references, so we would lose the
9044 : * cross-check that the pages belonged to a relation that was dropped
9045 : * later.
9046 : */
9047 0 : if (XLogHaveInvalidPages())
9048 : {
9049 0 : elog(trace_recovery(DEBUG2),
9050 : "could not record restart point at %X/%X because there "
9051 : "are unresolved references to invalid pages",
9052 : (uint32) (checkPoint->redo >> 32),
9053 : (uint32) checkPoint->redo);
9054 0 : return;
9055 : }
9056 :
9057 : /*
9058 : * Copy the checkpoint record to shared memory, so that checkpointer can
9059 : * work out the next time it wants to perform a restartpoint.
9060 : */
9061 0 : SpinLockAcquire(&XLogCtl->info_lck);
9062 0 : XLogCtl->lastCheckPointRecPtr = ReadRecPtr;
9063 0 : XLogCtl->lastCheckPointEndPtr = EndRecPtr;
9064 0 : XLogCtl->lastCheckPoint = *checkPoint;
9065 0 : SpinLockRelease(&XLogCtl->info_lck);
9066 : }
9067 :
9068 : /*
9069 : * Establish a restartpoint if possible.
9070 : *
9071 : * This is similar to CreateCheckPoint, but is used during WAL recovery
9072 : * to establish a point from which recovery can roll forward without
9073 : * replaying the entire recovery log.
9074 : *
9075 : * Returns true if a new restartpoint was established. We can only establish
9076 : * a restartpoint if we have replayed a safe checkpoint record since last
9077 : * restartpoint.
9078 : */
9079 : bool
9080 0 : CreateRestartPoint(int flags)
9081 : {
9082 : XLogRecPtr lastCheckPointRecPtr;
9083 : XLogRecPtr lastCheckPointEndPtr;
9084 : CheckPoint lastCheckPoint;
9085 : XLogRecPtr PriorRedoPtr;
9086 : TimestampTz xtime;
9087 :
9088 : /*
9089 : * Acquire CheckpointLock to ensure only one restartpoint or checkpoint
9090 : * happens at a time.
9091 : */
9092 0 : LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
9093 :
9094 : /* Get a local copy of the last safe checkpoint record. */
9095 0 : SpinLockAcquire(&XLogCtl->info_lck);
9096 0 : lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
9097 0 : lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
9098 0 : lastCheckPoint = XLogCtl->lastCheckPoint;
9099 0 : SpinLockRelease(&XLogCtl->info_lck);
9100 :
9101 : /*
9102 : * Check that we're still in recovery mode. It's ok if we exit recovery
9103 : * mode after this check, the restart point is valid anyway.
9104 : */
9105 0 : if (!RecoveryInProgress())
9106 : {
9107 0 : ereport(DEBUG2,
9108 : (errmsg("skipping restartpoint, recovery has already ended")));
9109 0 : LWLockRelease(CheckpointLock);
9110 0 : return false;
9111 : }
9112 :
9113 : /*
9114 : * If the last checkpoint record we've replayed is already our last
9115 : * restartpoint, we can't perform a new restart point. We still update
9116 : * minRecoveryPoint in that case, so that if this is a shutdown restart
9117 : * point, we won't start up earlier than before. That's not strictly
9118 : * necessary, but when hot standby is enabled, it would be rather weird if
9119 : * the database opened up for read-only connections at a point-in-time
9120 : * before the last shutdown. Such time travel is still possible in case of
9121 : * immediate shutdown, though.
9122 : *
9123 : * We don't explicitly advance minRecoveryPoint when we do create a
9124 : * restartpoint. It's assumed that flushing the buffers will do that as a
9125 : * side-effect.
9126 : */
9127 0 : if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
9128 0 : lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
9129 : {
9130 0 : ereport(DEBUG2,
9131 : (errmsg("skipping restartpoint, already performed at %X/%X",
9132 : (uint32) (lastCheckPoint.redo >> 32),
9133 : (uint32) lastCheckPoint.redo)));
9134 :
9135 0 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
9136 0 : if (flags & CHECKPOINT_IS_SHUTDOWN)
9137 : {
9138 0 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9139 0 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
9140 0 : ControlFile->time = (pg_time_t) time(NULL);
9141 0 : UpdateControlFile();
9142 0 : LWLockRelease(ControlFileLock);
9143 : }
9144 0 : LWLockRelease(CheckpointLock);
9145 0 : return false;
9146 : }
9147 :
9148 : /*
9149 : * Update the shared RedoRecPtr so that the startup process can calculate
9150 : * the number of segments replayed since last restartpoint, and request a
9151 : * restartpoint if it exceeds CheckPointSegments.
9152 : *
9153 : * Like in CreateCheckPoint(), hold off insertions to update it, although
9154 : * during recovery this is just pro forma, because no WAL insertions are
9155 : * happening.
9156 : */
9157 0 : WALInsertLockAcquireExclusive();
9158 0 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
9159 0 : WALInsertLockRelease();
9160 :
9161 : /* Also update the info_lck-protected copy */
9162 0 : SpinLockAcquire(&XLogCtl->info_lck);
9163 0 : XLogCtl->RedoRecPtr = lastCheckPoint.redo;
9164 0 : SpinLockRelease(&XLogCtl->info_lck);
9165 :
9166 : /*
9167 : * Prepare to accumulate statistics.
9168 : *
9169 : * Note: because it is possible for log_checkpoints to change while a
9170 : * checkpoint proceeds, we always accumulate stats, even if
9171 : * log_checkpoints is currently off.
9172 : */
9173 0 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
9174 0 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
9175 :
9176 0 : if (log_checkpoints)
9177 0 : LogCheckpointStart(flags, true);
9178 :
9179 0 : CheckPointGuts(lastCheckPoint.redo, flags);
9180 :
9181 : /*
9182 : * Remember the prior checkpoint's redo pointer, used later to determine
9183 : * the point at which we can truncate the log.
9184 : */
9185 0 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
9186 :
9187 : /*
9188 : * Update pg_control, using current time. Check that it still shows
9189 : * IN_ARCHIVE_RECOVERY state and an older checkpoint, else do nothing;
9190 : * this is a quick hack to make sure nothing really bad happens if somehow
9191 : * we get here after the end-of-recovery checkpoint.
9192 : */
9193 0 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9194 0 : if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY &&
9195 0 : ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
9196 : {
9197 0 : ControlFile->prevCheckPoint = ControlFile->checkPoint;
9198 0 : ControlFile->checkPoint = lastCheckPointRecPtr;
9199 0 : ControlFile->checkPointCopy = lastCheckPoint;
9200 0 : ControlFile->time = (pg_time_t) time(NULL);
9201 :
9202 : /*
9203 : * Ensure minRecoveryPoint is past the checkpoint record. Normally,
9204 : * this will have happened already while writing out dirty buffers,
9205 : * but not necessarily - e.g. because no buffers were dirtied. We do
9206 : * this because a non-exclusive base backup uses minRecoveryPoint to
9207 : * determine which WAL files must be included in the backup, and the
9208 : * file (or files) containing the checkpoint record must be included,
9209 : * at a minimum. Note that for an ordinary restart of recovery there's
9210 : * no value in having the minimum recovery point any earlier than this
9211 : * anyway, because redo will begin just after the checkpoint record.
9212 : */
9213 0 : if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
9214 : {
9215 0 : ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
9216 0 : ControlFile->minRecoveryPointTLI = lastCheckPoint.ThisTimeLineID;
9217 :
9218 : /* update local copy */
9219 0 : minRecoveryPoint = ControlFile->minRecoveryPoint;
9220 0 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
9221 : }
9222 0 : if (flags & CHECKPOINT_IS_SHUTDOWN)
9223 0 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
9224 0 : UpdateControlFile();
9225 : }
9226 0 : LWLockRelease(ControlFileLock);
9227 :
9228 : /*
9229 : * Delete old log files (those no longer needed even for previous
9230 : * checkpoint/restartpoint) to prevent the disk holding the xlog from
9231 : * growing full.
9232 : */
9233 0 : if (PriorRedoPtr != InvalidXLogRecPtr)
9234 : {
9235 : XLogRecPtr receivePtr;
9236 : XLogRecPtr replayPtr;
9237 : TimeLineID replayTLI;
9238 : XLogRecPtr endptr;
9239 : XLogSegNo _logSegNo;
9240 :
9241 : /* Update the average distance between checkpoints/restartpoints. */
9242 0 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
9243 :
9244 0 : XLByteToSeg(PriorRedoPtr, _logSegNo);
9245 :
9246 : /*
9247 : * Get the current end of xlog replayed or received, whichever is
9248 : * later.
9249 : */
9250 0 : receivePtr = GetWalRcvWriteRecPtr(NULL, NULL);
9251 0 : replayPtr = GetXLogReplayRecPtr(&replayTLI);
9252 0 : endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
9253 :
9254 0 : KeepLogSeg(endptr, &_logSegNo);
9255 0 : _logSegNo--;
9256 :
9257 : /*
9258 : * Try to recycle segments on a useful timeline. If we've been
9259 : * promoted since the beginning of this restartpoint, use the new
9260 : * timeline chosen at end of recovery (RecoveryInProgress() sets
9261 : * ThisTimeLineID in that case). If we're still in recovery, use the
9262 : * timeline we're currently replaying.
9263 : *
9264 : * There is no guarantee that the WAL segments will be useful on the
9265 : * current timeline; if recovery proceeds to a new timeline right
9266 : * after this, the pre-allocated WAL segments on this timeline will
9267 : * not be used, and will go wasted until recycled on the next
9268 : * restartpoint. We'll live with that.
9269 : */
9270 0 : if (RecoveryInProgress())
9271 0 : ThisTimeLineID = replayTLI;
9272 :
9273 0 : RemoveOldXlogFiles(_logSegNo, PriorRedoPtr, endptr);
9274 :
9275 : /*
9276 : * Make more log segments if needed. (Do this after recycling old log
9277 : * segments, since that may supply some of the needed files.)
9278 : */
9279 0 : PreallocXlogFiles(endptr);
9280 :
9281 : /*
9282 : * ThisTimeLineID is normally not set when we're still in recovery.
9283 : * However, recycling/preallocating segments above needed
9284 : * ThisTimeLineID to determine which timeline to install the segments
9285 : * on. Reset it now, to restore the normal state of affairs for
9286 : * debugging purposes.
9287 : */
9288 0 : if (RecoveryInProgress())
9289 0 : ThisTimeLineID = 0;
9290 : }
9291 :
9292 : /*
9293 : * Truncate pg_subtrans if possible. We can throw away all data before
9294 : * the oldest XMIN of any running transaction. No future transaction will
9295 : * attempt to reference any pg_subtrans entry older than that (see Asserts
9296 : * in subtrans.c). When hot standby is disabled, though, we mustn't do
9297 : * this because StartupSUBTRANS hasn't been called yet.
9298 : */
9299 0 : if (EnableHotStandby)
9300 0 : TruncateSUBTRANS(GetOldestXmin(NULL, PROCARRAY_FLAGS_DEFAULT));
9301 :
9302 : /* Real work is done, but log and update before releasing lock. */
9303 0 : LogCheckpointEnd(true);
9304 :
9305 0 : xtime = GetLatestXTime();
9306 0 : ereport((log_checkpoints ? LOG : DEBUG2),
9307 : (errmsg("recovery restart point at %X/%X",
9308 : (uint32) (lastCheckPoint.redo >> 32), (uint32) lastCheckPoint.redo),
9309 : xtime ? errdetail("last completed transaction was at log time %s",
9310 : timestamptz_to_str(xtime)) : 0));
9311 :
9312 0 : LWLockRelease(CheckpointLock);
9313 :
9314 : /*
9315 : * Finally, execute archive_cleanup_command, if any.
9316 : */
9317 0 : if (XLogCtl->archiveCleanupCommand[0])
9318 0 : ExecuteRecoveryCommand(XLogCtl->archiveCleanupCommand,
9319 : "archive_cleanup_command",
9320 : false);
9321 :
9322 0 : return true;
9323 : }
9324 :
9325 : /*
9326 : * Retreat *logSegNo to the last segment that we need to retain because of
9327 : * either wal_keep_segments or replication slots.
9328 : *
9329 : * This is calculated by subtracting wal_keep_segments from the given xlog
9330 : * location, recptr and by making sure that that result is below the
9331 : * requirement of replication slots.
9332 : */
9333 : static void
9334 11 : KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
9335 : {
9336 : XLogSegNo segno;
9337 : XLogRecPtr keep;
9338 :
9339 11 : XLByteToSeg(recptr, segno);
9340 11 : keep = XLogGetReplicationSlotMinimumLSN();
9341 :
9342 : /* compute limit for wal_keep_segments first */
9343 11 : if (wal_keep_segments > 0)
9344 : {
9345 : /* avoid underflow, don't go below 1 */
9346 0 : if (segno <= wal_keep_segments)
9347 0 : segno = 1;
9348 : else
9349 0 : segno = segno - wal_keep_segments;
9350 : }
9351 :
9352 : /* then check whether slots limit removal further */
9353 11 : if (max_replication_slots > 0 && keep != InvalidXLogRecPtr)
9354 : {
9355 : XLogSegNo slotSegNo;
9356 :
9357 0 : XLByteToSeg(keep, slotSegNo);
9358 :
9359 0 : if (slotSegNo <= 0)
9360 0 : segno = 1;
9361 0 : else if (slotSegNo < segno)
9362 0 : segno = slotSegNo;
9363 : }
9364 :
9365 : /* don't delete WAL segments newer than the calculated segment */
9366 11 : if (segno < *logSegNo)
9367 0 : *logSegNo = segno;
9368 11 : }
9369 :
9370 : /*
9371 : * Write a NEXTOID log record
9372 : */
9373 : void
9374 6 : XLogPutNextOid(Oid nextOid)
9375 : {
9376 6 : XLogBeginInsert();
9377 6 : XLogRegisterData((char *) (&nextOid), sizeof(Oid));
9378 6 : (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
9379 :
9380 : /*
9381 : * We need not flush the NEXTOID record immediately, because any of the
9382 : * just-allocated OIDs could only reach disk as part of a tuple insert or
9383 : * update that would have its own XLOG record that must follow the NEXTOID
9384 : * record. Therefore, the standard buffer LSN interlock applied to those
9385 : * records will ensure no such OID reaches disk before the NEXTOID record
9386 : * does.
9387 : *
9388 : * Note, however, that the above statement only covers state "within" the
9389 : * database. When we use a generated OID as a file or directory name, we
9390 : * are in a sense violating the basic WAL rule, because that filesystem
9391 : * change may reach disk before the NEXTOID WAL record does. The impact
9392 : * of this is that if a database crash occurs immediately afterward, we
9393 : * might after restart re-generate the same OID and find that it conflicts
9394 : * with the leftover file or directory. But since for safety's sake we
9395 : * always loop until finding a nonconflicting filename, this poses no real
9396 : * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
9397 : */
9398 6 : }
9399 :
9400 : /*
9401 : * Write an XLOG SWITCH record.
9402 : *
9403 : * Here we just blindly issue an XLogInsert request for the record.
9404 : * All the magic happens inside XLogInsert.
9405 : *
9406 : * The return value is either the end+1 address of the switch record,
9407 : * or the end+1 address of the prior segment if we did not need to
9408 : * write a switch record because we are already at segment start.
9409 : */
9410 : XLogRecPtr
9411 0 : RequestXLogSwitch(bool mark_unimportant)
9412 : {
9413 : XLogRecPtr RecPtr;
9414 :
9415 : /* XLOG SWITCH has no data */
9416 0 : XLogBeginInsert();
9417 :
9418 0 : if (mark_unimportant)
9419 0 : XLogSetRecordFlags(XLOG_MARK_UNIMPORTANT);
9420 0 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH);
9421 :
9422 0 : return RecPtr;
9423 : }
9424 :
9425 : /*
9426 : * Write a RESTORE POINT record
9427 : */
9428 : XLogRecPtr
9429 0 : XLogRestorePoint(const char *rpName)
9430 : {
9431 : XLogRecPtr RecPtr;
9432 : xl_restore_point xlrec;
9433 :
9434 0 : xlrec.rp_time = GetCurrentTimestamp();
9435 0 : strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
9436 :
9437 0 : XLogBeginInsert();
9438 0 : XLogRegisterData((char *) &xlrec, sizeof(xl_restore_point));
9439 :
9440 0 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
9441 :
9442 0 : ereport(LOG,
9443 : (errmsg("restore point \"%s\" created at %X/%X",
9444 : rpName, (uint32) (RecPtr >> 32), (uint32) RecPtr)));
9445 :
9446 0 : return RecPtr;
9447 : }
9448 :
9449 : /*
9450 : * Check if any of the GUC parameters that are critical for hot standby
9451 : * have changed, and update the value in pg_control file if necessary.
9452 : */
9453 : static void
9454 3 : XLogReportParameters(void)
9455 : {
9456 6 : if (wal_level != ControlFile->wal_level ||
9457 6 : wal_log_hints != ControlFile->wal_log_hints ||
9458 6 : MaxConnections != ControlFile->MaxConnections ||
9459 6 : max_worker_processes != ControlFile->max_worker_processes ||
9460 5 : max_prepared_xacts != ControlFile->max_prepared_xacts ||
9461 4 : max_locks_per_xact != ControlFile->max_locks_per_xact ||
9462 2 : track_commit_timestamp != ControlFile->track_commit_timestamp)
9463 : {
9464 : /*
9465 : * The change in number of backend slots doesn't need to be WAL-logged
9466 : * if archiving is not enabled, as you can't start archive recovery
9467 : * with wal_level=minimal anyway. We don't really care about the
9468 : * values in pg_control either if wal_level=minimal, but seems better
9469 : * to keep them up-to-date to avoid confusion.
9470 : */
9471 1 : if (wal_level != ControlFile->wal_level || XLogIsNeeded())
9472 : {
9473 : xl_parameter_change xlrec;
9474 : XLogRecPtr recptr;
9475 :
9476 1 : xlrec.MaxConnections = MaxConnections;
9477 1 : xlrec.max_worker_processes = max_worker_processes;
9478 1 : xlrec.max_prepared_xacts = max_prepared_xacts;
9479 1 : xlrec.max_locks_per_xact = max_locks_per_xact;
9480 1 : xlrec.wal_level = wal_level;
9481 1 : xlrec.wal_log_hints = wal_log_hints;
9482 1 : xlrec.track_commit_timestamp = track_commit_timestamp;
9483 :
9484 1 : XLogBeginInsert();
9485 1 : XLogRegisterData((char *) &xlrec, sizeof(xlrec));
9486 :
9487 1 : recptr = XLogInsert(RM_XLOG_ID, XLOG_PARAMETER_CHANGE);
9488 1 : XLogFlush(recptr);
9489 : }
9490 :
9491 1 : ControlFile->MaxConnections = MaxConnections;
9492 1 : ControlFile->max_worker_processes = max_worker_processes;
9493 1 : ControlFile->max_prepared_xacts = max_prepared_xacts;
9494 1 : ControlFile->max_locks_per_xact = max_locks_per_xact;
9495 1 : ControlFile->wal_level = wal_level;
9496 1 : ControlFile->wal_log_hints = wal_log_hints;
9497 1 : ControlFile->track_commit_timestamp = track_commit_timestamp;
9498 1 : UpdateControlFile();
9499 : }
9500 3 : }
9501 :
9502 : /*
9503 : * Update full_page_writes in shared memory, and write an
9504 : * XLOG_FPW_CHANGE record if necessary.
9505 : *
9506 : * Note: this function assumes there is no other process running
9507 : * concurrently that could update it.
9508 : */
9509 : void
9510 4 : UpdateFullPageWrites(void)
9511 : {
9512 4 : XLogCtlInsert *Insert = &XLogCtl->Insert;
9513 :
9514 : /*
9515 : * Do nothing if full_page_writes has not been changed.
9516 : *
9517 : * It's safe to check the shared full_page_writes without the lock,
9518 : * because we assume that there is no concurrently running process which
9519 : * can update it.
9520 : */
9521 4 : if (fullPageWrites == Insert->fullPageWrites)
9522 8 : return;
9523 :
9524 0 : START_CRIT_SECTION();
9525 :
9526 : /*
9527 : * It's always safe to take full page images, even when not strictly
9528 : * required, but not the other round. So if we're setting full_page_writes
9529 : * to true, first set it true and then write the WAL record. If we're
9530 : * setting it to false, first write the WAL record and then set the global
9531 : * flag.
9532 : */
9533 0 : if (fullPageWrites)
9534 : {
9535 0 : WALInsertLockAcquireExclusive();
9536 0 : Insert->fullPageWrites = true;
9537 0 : WALInsertLockRelease();
9538 : }
9539 :
9540 : /*
9541 : * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
9542 : * full_page_writes during archive recovery, if required.
9543 : */
9544 0 : if (XLogStandbyInfoActive() && !RecoveryInProgress())
9545 : {
9546 0 : XLogBeginInsert();
9547 0 : XLogRegisterData((char *) (&fullPageWrites), sizeof(bool));
9548 :
9549 0 : XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
9550 : }
9551 :
9552 0 : if (!fullPageWrites)
9553 : {
9554 0 : WALInsertLockAcquireExclusive();
9555 0 : Insert->fullPageWrites = false;
9556 0 : WALInsertLockRelease();
9557 : }
9558 0 : END_CRIT_SECTION();
9559 : }
9560 :
9561 : /*
9562 : * Check that it's OK to switch to new timeline during recovery.
9563 : *
9564 : * 'lsn' is the address of the shutdown checkpoint record we're about to
9565 : * replay. (Currently, timeline can only change at a shutdown checkpoint).
9566 : */
9567 : static void
9568 0 : checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI, TimeLineID prevTLI)
9569 : {
9570 : /* Check that the record agrees on what the current (old) timeline is */
9571 0 : if (prevTLI != ThisTimeLineID)
9572 0 : ereport(PANIC,
9573 : (errmsg("unexpected previous timeline ID %u (current timeline ID %u) in checkpoint record",
9574 : prevTLI, ThisTimeLineID)));
9575 :
9576 : /*
9577 : * The new timeline better be in the list of timelines we expect to see,
9578 : * according to the timeline history. It should also not decrease.
9579 : */
9580 0 : if (newTLI < ThisTimeLineID || !tliInHistory(newTLI, expectedTLEs))
9581 0 : ereport(PANIC,
9582 : (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
9583 : newTLI, ThisTimeLineID)));
9584 :
9585 : /*
9586 : * If we have not yet reached min recovery point, and we're about to
9587 : * switch to a timeline greater than the timeline of the min recovery
9588 : * point: trouble. After switching to the new timeline, we could not
9589 : * possibly visit the min recovery point on the correct timeline anymore.
9590 : * This can happen if there is a newer timeline in the archive that
9591 : * branched before the timeline the min recovery point is on, and you
9592 : * attempt to do PITR to the new timeline.
9593 : */
9594 0 : if (!XLogRecPtrIsInvalid(minRecoveryPoint) &&
9595 0 : lsn < minRecoveryPoint &&
9596 0 : newTLI > minRecoveryPointTLI)
9597 0 : ereport(PANIC,
9598 : (errmsg("unexpected timeline ID %u in checkpoint record, before reaching minimum recovery point %X/%X on timeline %u",
9599 : newTLI,
9600 : (uint32) (minRecoveryPoint >> 32),
9601 : (uint32) minRecoveryPoint,
9602 : minRecoveryPointTLI)));
9603 :
9604 : /* Looks good */
9605 0 : }
9606 :
9607 : /*
9608 : * XLOG resource manager's routines
9609 : *
9610 : * Definitions of info values are in include/catalog/pg_control.h, though
9611 : * not all record types are related to control file updates.
9612 : */
9613 : void
9614 0 : xlog_redo(XLogReaderState *record)
9615 : {
9616 0 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
9617 0 : XLogRecPtr lsn = record->EndRecPtr;
9618 :
9619 : /* in XLOG rmgr, backup blocks are only used by XLOG_FPI records */
9620 0 : Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
9621 : !XLogRecHasAnyBlockRefs(record));
9622 :
9623 0 : if (info == XLOG_NEXTOID)
9624 : {
9625 : Oid nextOid;
9626 :
9627 : /*
9628 : * We used to try to take the maximum of ShmemVariableCache->nextOid
9629 : * and the recorded nextOid, but that fails if the OID counter wraps
9630 : * around. Since no OID allocation should be happening during replay
9631 : * anyway, better to just believe the record exactly. We still take
9632 : * OidGenLock while setting the variable, just in case.
9633 : */
9634 0 : memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
9635 0 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
9636 0 : ShmemVariableCache->nextOid = nextOid;
9637 0 : ShmemVariableCache->oidCount = 0;
9638 0 : LWLockRelease(OidGenLock);
9639 : }
9640 0 : else if (info == XLOG_CHECKPOINT_SHUTDOWN)
9641 : {
9642 : CheckPoint checkPoint;
9643 :
9644 0 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
9645 : /* In a SHUTDOWN checkpoint, believe the counters exactly */
9646 0 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
9647 0 : ShmemVariableCache->nextXid = checkPoint.nextXid;
9648 0 : LWLockRelease(XidGenLock);
9649 0 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
9650 0 : ShmemVariableCache->nextOid = checkPoint.nextOid;
9651 0 : ShmemVariableCache->oidCount = 0;
9652 0 : LWLockRelease(OidGenLock);
9653 0 : MultiXactSetNextMXact(checkPoint.nextMulti,
9654 : checkPoint.nextMultiOffset);
9655 :
9656 0 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
9657 : checkPoint.oldestMultiDB);
9658 :
9659 : /*
9660 : * No need to set oldestClogXid here as well; it'll be set when we
9661 : * redo an xl_clog_truncate if it changed since initialization.
9662 : */
9663 0 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
9664 :
9665 : /*
9666 : * If we see a shutdown checkpoint while waiting for an end-of-backup
9667 : * record, the backup was canceled and the end-of-backup record will
9668 : * never arrive.
9669 : */
9670 0 : if (ArchiveRecoveryRequested &&
9671 0 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint) &&
9672 0 : XLogRecPtrIsInvalid(ControlFile->backupEndPoint))
9673 0 : ereport(PANIC,
9674 : (errmsg("online backup was canceled, recovery cannot continue")));
9675 :
9676 : /*
9677 : * If we see a shutdown checkpoint, we know that nothing was running
9678 : * on the master at this point. So fake-up an empty running-xacts
9679 : * record and use that here and now. Recover additional standby state
9680 : * for prepared transactions.
9681 : */
9682 0 : if (standbyState >= STANDBY_INITIALIZED)
9683 : {
9684 : TransactionId *xids;
9685 : int nxids;
9686 : TransactionId oldestActiveXID;
9687 : TransactionId latestCompletedXid;
9688 : RunningTransactionsData running;
9689 :
9690 0 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
9691 :
9692 : /*
9693 : * Construct a RunningTransactions snapshot representing a shut
9694 : * down server, with only prepared transactions still alive. We're
9695 : * never overflowed at this point because all subxids are listed
9696 : * with their parent prepared transactions.
9697 : */
9698 0 : running.xcnt = nxids;
9699 0 : running.subxcnt = 0;
9700 0 : running.subxid_overflow = false;
9701 0 : running.nextXid = checkPoint.nextXid;
9702 0 : running.oldestRunningXid = oldestActiveXID;
9703 0 : latestCompletedXid = checkPoint.nextXid;
9704 0 : TransactionIdRetreat(latestCompletedXid);
9705 0 : Assert(TransactionIdIsNormal(latestCompletedXid));
9706 0 : running.latestCompletedXid = latestCompletedXid;
9707 0 : running.xids = xids;
9708 :
9709 0 : ProcArrayApplyRecoveryInfo(&running);
9710 :
9711 0 : StandbyRecoverPreparedTransactions();
9712 : }
9713 :
9714 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
9715 0 : ControlFile->checkPointCopy.nextXidEpoch = checkPoint.nextXidEpoch;
9716 0 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
9717 :
9718 : /* Update shared-memory copy of checkpoint XID/epoch */
9719 0 : SpinLockAcquire(&XLogCtl->info_lck);
9720 0 : XLogCtl->ckptXidEpoch = checkPoint.nextXidEpoch;
9721 0 : XLogCtl->ckptXid = checkPoint.nextXid;
9722 0 : SpinLockRelease(&XLogCtl->info_lck);
9723 :
9724 : /*
9725 : * We should've already switched to the new TLI before replaying this
9726 : * record.
9727 : */
9728 0 : if (checkPoint.ThisTimeLineID != ThisTimeLineID)
9729 0 : ereport(PANIC,
9730 : (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
9731 : checkPoint.ThisTimeLineID, ThisTimeLineID)));
9732 :
9733 0 : RecoveryRestartPoint(&checkPoint);
9734 : }
9735 0 : else if (info == XLOG_CHECKPOINT_ONLINE)
9736 : {
9737 : CheckPoint checkPoint;
9738 :
9739 0 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
9740 : /* In an ONLINE checkpoint, treat the XID counter as a minimum */
9741 0 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
9742 0 : if (TransactionIdPrecedes(ShmemVariableCache->nextXid,
9743 : checkPoint.nextXid))
9744 0 : ShmemVariableCache->nextXid = checkPoint.nextXid;
9745 0 : LWLockRelease(XidGenLock);
9746 : /* ... but still treat OID counter as exact */
9747 0 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
9748 0 : ShmemVariableCache->nextOid = checkPoint.nextOid;
9749 0 : ShmemVariableCache->oidCount = 0;
9750 0 : LWLockRelease(OidGenLock);
9751 0 : MultiXactAdvanceNextMXact(checkPoint.nextMulti,
9752 : checkPoint.nextMultiOffset);
9753 :
9754 : /*
9755 : * NB: This may perform multixact truncation when replaying WAL
9756 : * generated by an older primary.
9757 : */
9758 0 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
9759 : checkPoint.oldestMultiDB);
9760 0 : if (TransactionIdPrecedes(ShmemVariableCache->oldestXid,
9761 : checkPoint.oldestXid))
9762 0 : SetTransactionIdLimit(checkPoint.oldestXid,
9763 : checkPoint.oldestXidDB);
9764 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
9765 0 : ControlFile->checkPointCopy.nextXidEpoch = checkPoint.nextXidEpoch;
9766 0 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
9767 :
9768 : /* Update shared-memory copy of checkpoint XID/epoch */
9769 0 : SpinLockAcquire(&XLogCtl->info_lck);
9770 0 : XLogCtl->ckptXidEpoch = checkPoint.nextXidEpoch;
9771 0 : XLogCtl->ckptXid = checkPoint.nextXid;
9772 0 : SpinLockRelease(&XLogCtl->info_lck);
9773 :
9774 : /* TLI should not change in an on-line checkpoint */
9775 0 : if (checkPoint.ThisTimeLineID != ThisTimeLineID)
9776 0 : ereport(PANIC,
9777 : (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
9778 : checkPoint.ThisTimeLineID, ThisTimeLineID)));
9779 :
9780 0 : RecoveryRestartPoint(&checkPoint);
9781 : }
9782 0 : else if (info == XLOG_END_OF_RECOVERY)
9783 : {
9784 : xl_end_of_recovery xlrec;
9785 :
9786 0 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
9787 :
9788 : /*
9789 : * For Hot Standby, we could treat this like a Shutdown Checkpoint,
9790 : * but this case is rarer and harder to test, so the benefit doesn't
9791 : * outweigh the potential extra cost of maintenance.
9792 : */
9793 :
9794 : /*
9795 : * We should've already switched to the new TLI before replaying this
9796 : * record.
9797 : */
9798 0 : if (xlrec.ThisTimeLineID != ThisTimeLineID)
9799 0 : ereport(PANIC,
9800 : (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
9801 : xlrec.ThisTimeLineID, ThisTimeLineID)));
9802 : }
9803 0 : else if (info == XLOG_NOOP)
9804 : {
9805 : /* nothing to do here */
9806 : }
9807 0 : else if (info == XLOG_SWITCH)
9808 : {
9809 : /* nothing to do here */
9810 : }
9811 0 : else if (info == XLOG_RESTORE_POINT)
9812 : {
9813 : /* nothing to do here */
9814 : }
9815 0 : else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
9816 0 : {
9817 : Buffer buffer;
9818 :
9819 : /*
9820 : * Full-page image (FPI) records contain nothing else but a backup
9821 : * block. The block reference must include a full-page image -
9822 : * otherwise there would be no point in this record.
9823 : *
9824 : * No recovery conflicts are generated by these generic records - if a
9825 : * resource manager needs to generate conflicts, it has to define a
9826 : * separate WAL record type and redo routine.
9827 : *
9828 : * XLOG_FPI_FOR_HINT records are generated when a page needs to be
9829 : * WAL- logged because of a hint bit update. They are only generated
9830 : * when checksums are enabled. There is no difference in handling
9831 : * XLOG_FPI and XLOG_FPI_FOR_HINT records, they use a different info
9832 : * code just to distinguish them for statistics purposes.
9833 : */
9834 0 : if (XLogReadBufferForRedo(record, 0, &buffer) != BLK_RESTORED)
9835 0 : elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
9836 0 : UnlockReleaseBuffer(buffer);
9837 : }
9838 0 : else if (info == XLOG_BACKUP_END)
9839 : {
9840 : XLogRecPtr startpoint;
9841 :
9842 0 : memcpy(&startpoint, XLogRecGetData(record), sizeof(startpoint));
9843 :
9844 0 : if (ControlFile->backupStartPoint == startpoint)
9845 : {
9846 : /*
9847 : * We have reached the end of base backup, the point where
9848 : * pg_stop_backup() was done. The data on disk is now consistent.
9849 : * Reset backupStartPoint, and update minRecoveryPoint to make
9850 : * sure we don't allow starting up at an earlier point even if
9851 : * recovery is stopped and restarted soon after this.
9852 : */
9853 0 : elog(DEBUG1, "end of backup reached");
9854 :
9855 0 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9856 :
9857 0 : if (ControlFile->minRecoveryPoint < lsn)
9858 : {
9859 0 : ControlFile->minRecoveryPoint = lsn;
9860 0 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
9861 : }
9862 0 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
9863 0 : ControlFile->backupEndRequired = false;
9864 0 : UpdateControlFile();
9865 :
9866 0 : LWLockRelease(ControlFileLock);
9867 : }
9868 : }
9869 0 : else if (info == XLOG_PARAMETER_CHANGE)
9870 : {
9871 : xl_parameter_change xlrec;
9872 :
9873 : /* Update our copy of the parameters in pg_control */
9874 0 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
9875 :
9876 0 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9877 0 : ControlFile->MaxConnections = xlrec.MaxConnections;
9878 0 : ControlFile->max_worker_processes = xlrec.max_worker_processes;
9879 0 : ControlFile->max_prepared_xacts = xlrec.max_prepared_xacts;
9880 0 : ControlFile->max_locks_per_xact = xlrec.max_locks_per_xact;
9881 0 : ControlFile->wal_level = xlrec.wal_level;
9882 0 : ControlFile->wal_log_hints = xlrec.wal_log_hints;
9883 :
9884 : /*
9885 : * Update minRecoveryPoint to ensure that if recovery is aborted, we
9886 : * recover back up to this point before allowing hot standby again.
9887 : * This is important if the max_* settings are decreased, to ensure
9888 : * you don't run queries against the WAL preceding the change.
9889 : */
9890 0 : minRecoveryPoint = ControlFile->minRecoveryPoint;
9891 0 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
9892 0 : if (minRecoveryPoint != 0 && minRecoveryPoint < lsn)
9893 : {
9894 0 : ControlFile->minRecoveryPoint = lsn;
9895 0 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
9896 : }
9897 :
9898 0 : CommitTsParameterChange(xlrec.track_commit_timestamp,
9899 0 : ControlFile->track_commit_timestamp);
9900 0 : ControlFile->track_commit_timestamp = xlrec.track_commit_timestamp;
9901 :
9902 0 : UpdateControlFile();
9903 0 : LWLockRelease(ControlFileLock);
9904 :
9905 : /* Check to see if any changes to max_connections give problems */
9906 0 : CheckRequiredParameterValues();
9907 : }
9908 0 : else if (info == XLOG_FPW_CHANGE)
9909 : {
9910 : bool fpw;
9911 :
9912 0 : memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
9913 :
9914 : /*
9915 : * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
9916 : * do_pg_start_backup() and do_pg_stop_backup() can check whether
9917 : * full_page_writes has been disabled during online backup.
9918 : */
9919 0 : if (!fpw)
9920 : {
9921 0 : SpinLockAcquire(&XLogCtl->info_lck);
9922 0 : if (XLogCtl->lastFpwDisableRecPtr < ReadRecPtr)
9923 0 : XLogCtl->lastFpwDisableRecPtr = ReadRecPtr;
9924 0 : SpinLockRelease(&XLogCtl->info_lck);
9925 : }
9926 :
9927 : /* Keep track of full_page_writes */
9928 0 : lastFullPageWrites = fpw;
9929 : }
9930 0 : }
9931 :
9932 : #ifdef WAL_DEBUG
9933 :
9934 : static void
9935 : xlog_outrec(StringInfo buf, XLogReaderState *record)
9936 : {
9937 : int block_id;
9938 :
9939 : appendStringInfo(buf, "prev %X/%X; xid %u",
9940 : (uint32) (XLogRecGetPrev(record) >> 32),
9941 : (uint32) XLogRecGetPrev(record),
9942 : XLogRecGetXid(record));
9943 :
9944 : appendStringInfo(buf, "; len %u",
9945 : XLogRecGetDataLen(record));
9946 :
9947 : /* decode block references */
9948 : for (block_id = 0; block_id <= record->max_block_id; block_id++)
9949 : {
9950 : RelFileNode rnode;
9951 : ForkNumber forknum;
9952 : BlockNumber blk;
9953 :
9954 : if (!XLogRecHasBlockRef(record, block_id))
9955 : continue;
9956 :
9957 : XLogRecGetBlockTag(record, block_id, &rnode, &forknum, &blk);
9958 : if (forknum != MAIN_FORKNUM)
9959 : appendStringInfo(buf, "; blkref #%u: rel %u/%u/%u, fork %u, blk %u",
9960 : block_id,
9961 : rnode.spcNode, rnode.dbNode, rnode.relNode,
9962 : forknum,
9963 : blk);
9964 : else
9965 : appendStringInfo(buf, "; blkref #%u: rel %u/%u/%u, blk %u",
9966 : block_id,
9967 : rnode.spcNode, rnode.dbNode, rnode.relNode,
9968 : blk);
9969 : if (XLogRecHasBlockImage(record, block_id))
9970 : appendStringInfoString(buf, " FPW");
9971 : }
9972 : }
9973 : #endif /* WAL_DEBUG */
9974 :
9975 : /*
9976 : * Returns a string describing an XLogRecord, consisting of its identity
9977 : * optionally followed by a colon, a space, and a further description.
9978 : */
9979 : static void
9980 0 : xlog_outdesc(StringInfo buf, XLogReaderState *record)
9981 : {
9982 0 : RmgrId rmid = XLogRecGetRmid(record);
9983 0 : uint8 info = XLogRecGetInfo(record);
9984 : const char *id;
9985 :
9986 0 : appendStringInfoString(buf, RmgrTable[rmid].rm_name);
9987 0 : appendStringInfoChar(buf, '/');
9988 :
9989 0 : id = RmgrTable[rmid].rm_identify(info);
9990 0 : if (id == NULL)
9991 0 : appendStringInfo(buf, "UNKNOWN (%X): ", info & ~XLR_INFO_MASK);
9992 : else
9993 0 : appendStringInfo(buf, "%s: ", id);
9994 :
9995 0 : RmgrTable[rmid].rm_desc(buf, record);
9996 0 : }
9997 :
9998 :
9999 : /*
10000 : * Return the (possible) sync flag used for opening a file, depending on the
10001 : * value of the GUC wal_sync_method.
10002 : */
10003 : static int
10004 268 : get_sync_bit(int method)
10005 : {
10006 268 : int o_direct_flag = 0;
10007 :
10008 : /* If fsync is disabled, never open in sync mode */
10009 268 : if (!enableFsync)
10010 268 : return 0;
10011 :
10012 : /*
10013 : * Optimize writes by bypassing kernel cache with O_DIRECT when using
10014 : * O_SYNC/O_FSYNC and O_DSYNC. But only if archiving and streaming are
10015 : * disabled, otherwise the archive command or walsender process will read
10016 : * the WAL soon after writing it, which is guaranteed to cause a physical
10017 : * read if we bypassed the kernel cache. We also skip the
10018 : * posix_fadvise(POSIX_FADV_DONTNEED) call in XLogFileClose() for the same
10019 : * reason.
10020 : *
10021 : * Never use O_DIRECT in walreceiver process for similar reasons; the WAL
10022 : * written by walreceiver is normally read by the startup process soon
10023 : * after its written. Also, walreceiver performs unaligned writes, which
10024 : * don't work with O_DIRECT, so it is required for correctness too.
10025 : */
10026 0 : if (!XLogIsNeeded() && !AmWalReceiverProcess())
10027 0 : o_direct_flag = PG_O_DIRECT;
10028 :
10029 0 : switch (method)
10030 : {
10031 : /*
10032 : * enum values for all sync options are defined even if they are
10033 : * not supported on the current platform. But if not, they are
10034 : * not included in the enum option array, and therefore will never
10035 : * be seen here.
10036 : */
10037 : case SYNC_METHOD_FSYNC:
10038 : case SYNC_METHOD_FSYNC_WRITETHROUGH:
10039 : case SYNC_METHOD_FDATASYNC:
10040 0 : return 0;
10041 : #ifdef OPEN_SYNC_FLAG
10042 : case SYNC_METHOD_OPEN:
10043 0 : return OPEN_SYNC_FLAG | o_direct_flag;
10044 : #endif
10045 : #ifdef OPEN_DATASYNC_FLAG
10046 : case SYNC_METHOD_OPEN_DSYNC:
10047 0 : return OPEN_DATASYNC_FLAG | o_direct_flag;
10048 : #endif
10049 : default:
10050 : /* can't happen (unless we are out of sync with option array) */
10051 0 : elog(ERROR, "unrecognized wal_sync_method: %d", method);
10052 : return 0; /* silence warning */
10053 : }
10054 : }
10055 :
10056 : /*
10057 : * GUC support
10058 : */
10059 : void
10060 5 : assign_xlog_sync_method(int new_sync_method, void *extra)
10061 : {
10062 5 : if (sync_method != new_sync_method)
10063 : {
10064 : /*
10065 : * To ensure that no blocks escape unsynced, force an fsync on the
10066 : * currently open log segment (if any). Also, if the open flag is
10067 : * changing, close the log file so it will be reopened (with new flag
10068 : * bit) at next use.
10069 : */
10070 0 : if (openLogFile >= 0)
10071 : {
10072 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN);
10073 0 : if (pg_fsync(openLogFile) != 0)
10074 0 : ereport(PANIC,
10075 : (errcode_for_file_access(),
10076 : errmsg("could not fsync log segment %s: %m",
10077 : XLogFileNameP(ThisTimeLineID, openLogSegNo))));
10078 0 : pgstat_report_wait_end();
10079 0 : if (get_sync_bit(sync_method) != get_sync_bit(new_sync_method))
10080 0 : XLogFileClose();
10081 : }
10082 : }
10083 5 : }
10084 :
10085 :
10086 : /*
10087 : * Issue appropriate kind of fsync (if any) for an XLOG output file.
10088 : *
10089 : * 'fd' is a file descriptor for the XLOG file to be fsync'd.
10090 : * 'log' and 'seg' are for error reporting purposes.
10091 : */
10092 : void
10093 9577 : issue_xlog_fsync(int fd, XLogSegNo segno)
10094 : {
10095 9577 : switch (sync_method)
10096 : {
10097 : case SYNC_METHOD_FSYNC:
10098 0 : if (pg_fsync_no_writethrough(fd) != 0)
10099 0 : ereport(PANIC,
10100 : (errcode_for_file_access(),
10101 : errmsg("could not fsync log file %s: %m",
10102 : XLogFileNameP(ThisTimeLineID, segno))));
10103 0 : break;
10104 : #ifdef HAVE_FSYNC_WRITETHROUGH
10105 : case SYNC_METHOD_FSYNC_WRITETHROUGH:
10106 : if (pg_fsync_writethrough(fd) != 0)
10107 : ereport(PANIC,
10108 : (errcode_for_file_access(),
10109 : errmsg("could not fsync write-through log file %s: %m",
10110 : XLogFileNameP(ThisTimeLineID, segno))));
10111 : break;
10112 : #endif
10113 : #ifdef HAVE_FDATASYNC
10114 : case SYNC_METHOD_FDATASYNC:
10115 9577 : if (pg_fdatasync(fd) != 0)
10116 0 : ereport(PANIC,
10117 : (errcode_for_file_access(),
10118 : errmsg("could not fdatasync log file %s: %m",
10119 : XLogFileNameP(ThisTimeLineID, segno))));
10120 9577 : break;
10121 : #endif
10122 : case SYNC_METHOD_OPEN:
10123 : case SYNC_METHOD_OPEN_DSYNC:
10124 : /* write synced it already */
10125 0 : break;
10126 : default:
10127 0 : elog(PANIC, "unrecognized wal_sync_method: %d", sync_method);
10128 : break;
10129 : }
10130 9577 : }
10131 :
10132 : /*
10133 : * Return the filename of given log segment, as a palloc'd string.
10134 : */
10135 : char *
10136 0 : XLogFileNameP(TimeLineID tli, XLogSegNo segno)
10137 : {
10138 0 : char *result = palloc(MAXFNAMELEN);
10139 :
10140 0 : XLogFileName(result, tli, segno);
10141 0 : return result;
10142 : }
10143 :
10144 : /*
10145 : * do_pg_start_backup is the workhorse of the user-visible pg_start_backup()
10146 : * function. It creates the necessary starting checkpoint and constructs the
10147 : * backup label file.
10148 : *
10149 : * There are two kind of backups: exclusive and non-exclusive. An exclusive
10150 : * backup is started with pg_start_backup(), and there can be only one active
10151 : * at a time. The backup and tablespace map files of an exclusive backup are
10152 : * written to $PGDATA/backup_label and $PGDATA/tablespace_map, and they are
10153 : * removed by pg_stop_backup().
10154 : *
10155 : * A non-exclusive backup is used for the streaming base backups (see
10156 : * src/backend/replication/basebackup.c). The difference to exclusive backups
10157 : * is that the backup label and tablespace map files are not written to disk.
10158 : * Instead, their would-be contents are returned in *labelfile and *tblspcmapfile,
10159 : * and the caller is responsible for including them in the backup archive as
10160 : * 'backup_label' and 'tablespace_map'. There can be many non-exclusive backups
10161 : * active at the same time, and they don't conflict with an exclusive backup
10162 : * either.
10163 : *
10164 : * tblspcmapfile is required mainly for tar format in windows as native windows
10165 : * utilities are not able to create symlinks while extracting files from tar.
10166 : * However for consistency, the same is used for all platforms.
10167 : *
10168 : * needtblspcmapfile is true for the cases (exclusive backup and for
10169 : * non-exclusive backup only when tar format is used for taking backup)
10170 : * when backup needs to generate tablespace_map file, it is used to
10171 : * embed escape character before newline character in tablespace path.
10172 : *
10173 : * Returns the minimum WAL location that must be present to restore from this
10174 : * backup, and the corresponding timeline ID in *starttli_p.
10175 : *
10176 : * Every successfully started non-exclusive backup must be stopped by calling
10177 : * do_pg_stop_backup() or do_pg_abort_backup().
10178 : *
10179 : * It is the responsibility of the caller of this function to verify the
10180 : * permissions of the calling user!
10181 : */
10182 : XLogRecPtr
10183 0 : do_pg_start_backup(const char *backupidstr, bool fast, TimeLineID *starttli_p,
10184 : StringInfo labelfile, DIR *tblspcdir, List **tablespaces,
10185 : StringInfo tblspcmapfile, bool infotbssize,
10186 : bool needtblspcmapfile)
10187 : {
10188 0 : bool exclusive = (labelfile == NULL);
10189 0 : bool backup_started_in_recovery = false;
10190 : XLogRecPtr checkpointloc;
10191 : XLogRecPtr startpoint;
10192 : TimeLineID starttli;
10193 : pg_time_t stamp_time;
10194 : char strfbuf[128];
10195 : char xlogfilename[MAXFNAMELEN];
10196 : XLogSegNo _logSegNo;
10197 : struct stat stat_buf;
10198 : FILE *fp;
10199 :
10200 0 : backup_started_in_recovery = RecoveryInProgress();
10201 :
10202 : /*
10203 : * Currently only non-exclusive backup can be taken during recovery.
10204 : */
10205 0 : if (backup_started_in_recovery && exclusive)
10206 0 : ereport(ERROR,
10207 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10208 : errmsg("recovery is in progress"),
10209 : errhint("WAL control functions cannot be executed during recovery.")));
10210 :
10211 : /*
10212 : * During recovery, we don't need to check WAL level. Because, if WAL
10213 : * level is not sufficient, it's impossible to get here during recovery.
10214 : */
10215 0 : if (!backup_started_in_recovery && !XLogIsNeeded())
10216 0 : ereport(ERROR,
10217 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10218 : errmsg("WAL level not sufficient for making an online backup"),
10219 : errhint("wal_level must be set to \"replica\" or \"logical\" at server start.")));
10220 :
10221 0 : if (strlen(backupidstr) > MAXPGPATH)
10222 0 : ereport(ERROR,
10223 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
10224 : errmsg("backup label too long (max %d bytes)",
10225 : MAXPGPATH)));
10226 :
10227 : /*
10228 : * Mark backup active in shared memory. We must do full-page WAL writes
10229 : * during an on-line backup even if not doing so at other times, because
10230 : * it's quite possible for the backup dump to obtain a "torn" (partially
10231 : * written) copy of a database page if it reads the page concurrently with
10232 : * our write to the same page. This can be fixed as long as the first
10233 : * write to the page in the WAL sequence is a full-page write. Hence, we
10234 : * turn on forcePageWrites and then force a CHECKPOINT, to ensure there
10235 : * are no dirty pages in shared memory that might get dumped while the
10236 : * backup is in progress without having a corresponding WAL record. (Once
10237 : * the backup is complete, we need not force full-page writes anymore,
10238 : * since we expect that any pages not modified during the backup interval
10239 : * must have been correctly captured by the backup.)
10240 : *
10241 : * Note that forcePageWrites has no effect during an online backup from
10242 : * the standby.
10243 : *
10244 : * We must hold all the insertion locks to change the value of
10245 : * forcePageWrites, to ensure adequate interlocking against
10246 : * XLogInsertRecord().
10247 : */
10248 0 : WALInsertLockAcquireExclusive();
10249 0 : if (exclusive)
10250 : {
10251 : /*
10252 : * At first, mark that we're now starting an exclusive backup, to
10253 : * ensure that there are no other sessions currently running
10254 : * pg_start_backup() or pg_stop_backup().
10255 : */
10256 0 : if (XLogCtl->Insert.exclusiveBackupState != EXCLUSIVE_BACKUP_NONE)
10257 : {
10258 0 : WALInsertLockRelease();
10259 0 : ereport(ERROR,
10260 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10261 : errmsg("a backup is already in progress"),
10262 : errhint("Run pg_stop_backup() and try again.")));
10263 : }
10264 0 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_STARTING;
10265 : }
10266 : else
10267 0 : XLogCtl->Insert.nonExclusiveBackups++;
10268 0 : XLogCtl->Insert.forcePageWrites = true;
10269 0 : WALInsertLockRelease();
10270 :
10271 : /* Ensure we release forcePageWrites if fail below */
10272 0 : PG_ENSURE_ERROR_CLEANUP(pg_start_backup_callback, (Datum) BoolGetDatum(exclusive));
10273 : {
10274 0 : bool gotUniqueStartpoint = false;
10275 : struct dirent *de;
10276 : tablespaceinfo *ti;
10277 : int datadirpathlen;
10278 :
10279 : /*
10280 : * Force an XLOG file switch before the checkpoint, to ensure that the
10281 : * WAL segment the checkpoint is written to doesn't contain pages with
10282 : * old timeline IDs. That would otherwise happen if you called
10283 : * pg_start_backup() right after restoring from a PITR archive: the
10284 : * first WAL segment containing the startup checkpoint has pages in
10285 : * the beginning with the old timeline ID. That can cause trouble at
10286 : * recovery: we won't have a history file covering the old timeline if
10287 : * pg_wal directory was not included in the base backup and the WAL
10288 : * archive was cleared too before starting the backup.
10289 : *
10290 : * This also ensures that we have emitted a WAL page header that has
10291 : * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
10292 : * Therefore, if a WAL archiver (such as pglesslog) is trying to
10293 : * compress out removable backup blocks, it won't remove any that
10294 : * occur after this point.
10295 : *
10296 : * During recovery, we skip forcing XLOG file switch, which means that
10297 : * the backup taken during recovery is not available for the special
10298 : * recovery case described above.
10299 : */
10300 0 : if (!backup_started_in_recovery)
10301 0 : RequestXLogSwitch(false);
10302 :
10303 : do
10304 : {
10305 : bool checkpointfpw;
10306 :
10307 : /*
10308 : * Force a CHECKPOINT. Aside from being necessary to prevent torn
10309 : * page problems, this guarantees that two successive backup runs
10310 : * will have different checkpoint positions and hence different
10311 : * history file names, even if nothing happened in between.
10312 : *
10313 : * During recovery, establish a restartpoint if possible. We use
10314 : * the last restartpoint as the backup starting checkpoint. This
10315 : * means that two successive backup runs can have same checkpoint
10316 : * positions.
10317 : *
10318 : * Since the fact that we are executing do_pg_start_backup()
10319 : * during recovery means that checkpointer is running, we can use
10320 : * RequestCheckpoint() to establish a restartpoint.
10321 : *
10322 : * We use CHECKPOINT_IMMEDIATE only if requested by user (via
10323 : * passing fast = true). Otherwise this can take awhile.
10324 : */
10325 0 : RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_WAIT |
10326 : (fast ? CHECKPOINT_IMMEDIATE : 0));
10327 :
10328 : /*
10329 : * Now we need to fetch the checkpoint record location, and also
10330 : * its REDO pointer. The oldest point in WAL that would be needed
10331 : * to restore starting from the checkpoint is precisely the REDO
10332 : * pointer.
10333 : */
10334 0 : LWLockAcquire(ControlFileLock, LW_SHARED);
10335 0 : checkpointloc = ControlFile->checkPoint;
10336 0 : startpoint = ControlFile->checkPointCopy.redo;
10337 0 : starttli = ControlFile->checkPointCopy.ThisTimeLineID;
10338 0 : checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
10339 0 : LWLockRelease(ControlFileLock);
10340 :
10341 0 : if (backup_started_in_recovery)
10342 : {
10343 : XLogRecPtr recptr;
10344 :
10345 : /*
10346 : * Check to see if all WAL replayed during online backup
10347 : * (i.e., since last restartpoint used as backup starting
10348 : * checkpoint) contain full-page writes.
10349 : */
10350 0 : SpinLockAcquire(&XLogCtl->info_lck);
10351 0 : recptr = XLogCtl->lastFpwDisableRecPtr;
10352 0 : SpinLockRelease(&XLogCtl->info_lck);
10353 :
10354 0 : if (!checkpointfpw || startpoint <= recptr)
10355 0 : ereport(ERROR,
10356 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10357 : errmsg("WAL generated with full_page_writes=off was replayed "
10358 : "since last restartpoint"),
10359 : errhint("This means that the backup being taken on the standby "
10360 : "is corrupt and should not be used. "
10361 : "Enable full_page_writes and run CHECKPOINT on the master, "
10362 : "and then try an online backup again.")));
10363 :
10364 : /*
10365 : * During recovery, since we don't use the end-of-backup WAL
10366 : * record and don't write the backup history file, the
10367 : * starting WAL location doesn't need to be unique. This means
10368 : * that two base backups started at the same time might use
10369 : * the same checkpoint as starting locations.
10370 : */
10371 0 : gotUniqueStartpoint = true;
10372 : }
10373 :
10374 : /*
10375 : * If two base backups are started at the same time (in WAL sender
10376 : * processes), we need to make sure that they use different
10377 : * checkpoints as starting locations, because we use the starting
10378 : * WAL location as a unique identifier for the base backup in the
10379 : * end-of-backup WAL record and when we write the backup history
10380 : * file. Perhaps it would be better generate a separate unique ID
10381 : * for each backup instead of forcing another checkpoint, but
10382 : * taking a checkpoint right after another is not that expensive
10383 : * either because only few buffers have been dirtied yet.
10384 : */
10385 0 : WALInsertLockAcquireExclusive();
10386 0 : if (XLogCtl->Insert.lastBackupStart < startpoint)
10387 : {
10388 0 : XLogCtl->Insert.lastBackupStart = startpoint;
10389 0 : gotUniqueStartpoint = true;
10390 : }
10391 0 : WALInsertLockRelease();
10392 0 : } while (!gotUniqueStartpoint);
10393 :
10394 0 : XLByteToSeg(startpoint, _logSegNo);
10395 0 : XLogFileName(xlogfilename, starttli, _logSegNo);
10396 :
10397 : /*
10398 : * Construct tablespace_map file
10399 : */
10400 0 : if (exclusive)
10401 0 : tblspcmapfile = makeStringInfo();
10402 :
10403 0 : datadirpathlen = strlen(DataDir);
10404 :
10405 : /* Collect information about all tablespaces */
10406 0 : while ((de = ReadDir(tblspcdir, "pg_tblspc")) != NULL)
10407 : {
10408 : char fullpath[MAXPGPATH + 10];
10409 : char linkpath[MAXPGPATH];
10410 0 : char *relpath = NULL;
10411 : int rllen;
10412 : StringInfoData buflinkpath;
10413 0 : char *s = linkpath;
10414 :
10415 : /* Skip special stuff */
10416 0 : if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0)
10417 0 : continue;
10418 :
10419 0 : snprintf(fullpath, sizeof(fullpath), "pg_tblspc/%s", de->d_name);
10420 :
10421 : #if defined(HAVE_READLINK) || defined(WIN32)
10422 0 : rllen = readlink(fullpath, linkpath, sizeof(linkpath));
10423 0 : if (rllen < 0)
10424 : {
10425 0 : ereport(WARNING,
10426 : (errmsg("could not read symbolic link \"%s\": %m",
10427 : fullpath)));
10428 0 : continue;
10429 : }
10430 0 : else if (rllen >= sizeof(linkpath))
10431 : {
10432 0 : ereport(WARNING,
10433 : (errmsg("symbolic link \"%s\" target is too long",
10434 : fullpath)));
10435 0 : continue;
10436 : }
10437 0 : linkpath[rllen] = '\0';
10438 :
10439 : /*
10440 : * Add the escape character '\\' before newline in a string to
10441 : * ensure that we can distinguish between the newline in the
10442 : * tablespace path and end of line while reading tablespace_map
10443 : * file during archive recovery.
10444 : */
10445 0 : initStringInfo(&buflinkpath);
10446 :
10447 0 : while (*s)
10448 : {
10449 0 : if ((*s == '\n' || *s == '\r') && needtblspcmapfile)
10450 0 : appendStringInfoChar(&buflinkpath, '\\');
10451 0 : appendStringInfoChar(&buflinkpath, *s++);
10452 : }
10453 :
10454 :
10455 : /*
10456 : * Relpath holds the relative path of the tablespace directory
10457 : * when it's located within PGDATA, or NULL if it's located
10458 : * elsewhere.
10459 : */
10460 0 : if (rllen > datadirpathlen &&
10461 0 : strncmp(linkpath, DataDir, datadirpathlen) == 0 &&
10462 0 : IS_DIR_SEP(linkpath[datadirpathlen]))
10463 0 : relpath = linkpath + datadirpathlen + 1;
10464 :
10465 0 : ti = palloc(sizeof(tablespaceinfo));
10466 0 : ti->oid = pstrdup(de->d_name);
10467 0 : ti->path = pstrdup(buflinkpath.data);
10468 0 : ti->rpath = relpath ? pstrdup(relpath) : NULL;
10469 0 : ti->size = infotbssize ? sendTablespace(fullpath, true) : -1;
10470 :
10471 0 : if (tablespaces)
10472 0 : *tablespaces = lappend(*tablespaces, ti);
10473 :
10474 0 : appendStringInfo(tblspcmapfile, "%s %s\n", ti->oid, ti->path);
10475 :
10476 0 : pfree(buflinkpath.data);
10477 : #else
10478 :
10479 : /*
10480 : * If the platform does not have symbolic links, it should not be
10481 : * possible to have tablespaces - clearly somebody else created
10482 : * them. Warn about it and ignore.
10483 : */
10484 : ereport(WARNING,
10485 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
10486 : errmsg("tablespaces are not supported on this platform")));
10487 : #endif
10488 : }
10489 :
10490 : /*
10491 : * Construct backup label file
10492 : */
10493 0 : if (exclusive)
10494 0 : labelfile = makeStringInfo();
10495 :
10496 : /* Use the log timezone here, not the session timezone */
10497 0 : stamp_time = (pg_time_t) time(NULL);
10498 0 : pg_strftime(strfbuf, sizeof(strfbuf),
10499 : "%Y-%m-%d %H:%M:%S %Z",
10500 0 : pg_localtime(&stamp_time, log_timezone));
10501 0 : appendStringInfo(labelfile, "START WAL LOCATION: %X/%X (file %s)\n",
10502 0 : (uint32) (startpoint >> 32), (uint32) startpoint, xlogfilename);
10503 0 : appendStringInfo(labelfile, "CHECKPOINT LOCATION: %X/%X\n",
10504 0 : (uint32) (checkpointloc >> 32), (uint32) checkpointloc);
10505 0 : appendStringInfo(labelfile, "BACKUP METHOD: %s\n",
10506 : exclusive ? "pg_start_backup" : "streamed");
10507 0 : appendStringInfo(labelfile, "BACKUP FROM: %s\n",
10508 : backup_started_in_recovery ? "standby" : "master");
10509 0 : appendStringInfo(labelfile, "START TIME: %s\n", strfbuf);
10510 0 : appendStringInfo(labelfile, "LABEL: %s\n", backupidstr);
10511 :
10512 : /*
10513 : * Okay, write the file, or return its contents to caller.
10514 : */
10515 0 : if (exclusive)
10516 : {
10517 : /*
10518 : * Check for existing backup label --- implies a backup is already
10519 : * running. (XXX given that we checked exclusiveBackupState
10520 : * above, maybe it would be OK to just unlink any such label
10521 : * file?)
10522 : */
10523 0 : if (stat(BACKUP_LABEL_FILE, &stat_buf) != 0)
10524 : {
10525 0 : if (errno != ENOENT)
10526 0 : ereport(ERROR,
10527 : (errcode_for_file_access(),
10528 : errmsg("could not stat file \"%s\": %m",
10529 : BACKUP_LABEL_FILE)));
10530 : }
10531 : else
10532 0 : ereport(ERROR,
10533 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10534 : errmsg("a backup is already in progress"),
10535 : errhint("If you're sure there is no backup in progress, remove file \"%s\" and try again.",
10536 : BACKUP_LABEL_FILE)));
10537 :
10538 0 : fp = AllocateFile(BACKUP_LABEL_FILE, "w");
10539 :
10540 0 : if (!fp)
10541 0 : ereport(ERROR,
10542 : (errcode_for_file_access(),
10543 : errmsg("could not create file \"%s\": %m",
10544 : BACKUP_LABEL_FILE)));
10545 0 : if (fwrite(labelfile->data, labelfile->len, 1, fp) != 1 ||
10546 0 : fflush(fp) != 0 ||
10547 0 : pg_fsync(fileno(fp)) != 0 ||
10548 0 : ferror(fp) ||
10549 0 : FreeFile(fp))
10550 0 : ereport(ERROR,
10551 : (errcode_for_file_access(),
10552 : errmsg("could not write file \"%s\": %m",
10553 : BACKUP_LABEL_FILE)));
10554 : /* Allocated locally for exclusive backups, so free separately */
10555 0 : pfree(labelfile->data);
10556 0 : pfree(labelfile);
10557 :
10558 : /* Write backup tablespace_map file. */
10559 0 : if (tblspcmapfile->len > 0)
10560 : {
10561 0 : if (stat(TABLESPACE_MAP, &stat_buf) != 0)
10562 : {
10563 0 : if (errno != ENOENT)
10564 0 : ereport(ERROR,
10565 : (errcode_for_file_access(),
10566 : errmsg("could not stat file \"%s\": %m",
10567 : TABLESPACE_MAP)));
10568 : }
10569 : else
10570 0 : ereport(ERROR,
10571 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10572 : errmsg("a backup is already in progress"),
10573 : errhint("If you're sure there is no backup in progress, remove file \"%s\" and try again.",
10574 : TABLESPACE_MAP)));
10575 :
10576 0 : fp = AllocateFile(TABLESPACE_MAP, "w");
10577 :
10578 0 : if (!fp)
10579 0 : ereport(ERROR,
10580 : (errcode_for_file_access(),
10581 : errmsg("could not create file \"%s\": %m",
10582 : TABLESPACE_MAP)));
10583 0 : if (fwrite(tblspcmapfile->data, tblspcmapfile->len, 1, fp) != 1 ||
10584 0 : fflush(fp) != 0 ||
10585 0 : pg_fsync(fileno(fp)) != 0 ||
10586 0 : ferror(fp) ||
10587 0 : FreeFile(fp))
10588 0 : ereport(ERROR,
10589 : (errcode_for_file_access(),
10590 : errmsg("could not write file \"%s\": %m",
10591 : TABLESPACE_MAP)));
10592 : }
10593 :
10594 : /* Allocated locally for exclusive backups, so free separately */
10595 0 : pfree(tblspcmapfile->data);
10596 0 : pfree(tblspcmapfile);
10597 : }
10598 : }
10599 0 : PG_END_ENSURE_ERROR_CLEANUP(pg_start_backup_callback, (Datum) BoolGetDatum(exclusive));
10600 :
10601 : /*
10602 : * Mark that start phase has correctly finished for an exclusive backup.
10603 : * Session-level locks are updated as well to reflect that state.
10604 : */
10605 0 : if (exclusive)
10606 : {
10607 0 : WALInsertLockAcquireExclusive();
10608 0 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_IN_PROGRESS;
10609 0 : WALInsertLockRelease();
10610 0 : sessionBackupState = SESSION_BACKUP_EXCLUSIVE;
10611 : }
10612 : else
10613 0 : sessionBackupState = SESSION_BACKUP_NON_EXCLUSIVE;
10614 :
10615 : /*
10616 : * We're done. As a convenience, return the starting WAL location.
10617 : */
10618 0 : if (starttli_p)
10619 0 : *starttli_p = starttli;
10620 0 : return startpoint;
10621 : }
10622 :
10623 : /* Error cleanup callback for pg_start_backup */
10624 : static void
10625 0 : pg_start_backup_callback(int code, Datum arg)
10626 : {
10627 0 : bool exclusive = DatumGetBool(arg);
10628 :
10629 : /* Update backup counters and forcePageWrites on failure */
10630 0 : WALInsertLockAcquireExclusive();
10631 0 : if (exclusive)
10632 : {
10633 0 : Assert(XLogCtl->Insert.exclusiveBackupState == EXCLUSIVE_BACKUP_STARTING);
10634 0 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_NONE;
10635 : }
10636 : else
10637 : {
10638 0 : Assert(XLogCtl->Insert.nonExclusiveBackups > 0);
10639 0 : XLogCtl->Insert.nonExclusiveBackups--;
10640 : }
10641 :
10642 0 : if (XLogCtl->Insert.exclusiveBackupState == EXCLUSIVE_BACKUP_NONE &&
10643 0 : XLogCtl->Insert.nonExclusiveBackups == 0)
10644 : {
10645 0 : XLogCtl->Insert.forcePageWrites = false;
10646 : }
10647 0 : WALInsertLockRelease();
10648 0 : }
10649 :
10650 : /*
10651 : * Error cleanup callback for pg_stop_backup
10652 : */
10653 : static void
10654 0 : pg_stop_backup_callback(int code, Datum arg)
10655 : {
10656 0 : bool exclusive = DatumGetBool(arg);
10657 :
10658 : /* Update backup status on failure */
10659 0 : WALInsertLockAcquireExclusive();
10660 0 : if (exclusive)
10661 : {
10662 0 : Assert(XLogCtl->Insert.exclusiveBackupState == EXCLUSIVE_BACKUP_STOPPING);
10663 0 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_IN_PROGRESS;
10664 : }
10665 0 : WALInsertLockRelease();
10666 0 : }
10667 :
10668 : /*
10669 : * Utility routine to fetch the session-level status of a backup running.
10670 : */
10671 : SessionBackupState
10672 0 : get_backup_status(void)
10673 : {
10674 0 : return sessionBackupState;
10675 : }
10676 :
10677 : /*
10678 : * do_pg_stop_backup is the workhorse of the user-visible pg_stop_backup()
10679 : * function.
10680 : *
10681 : * If labelfile is NULL, this stops an exclusive backup. Otherwise this stops
10682 : * the non-exclusive backup specified by 'labelfile'.
10683 : *
10684 : * Returns the last WAL location that must be present to restore from this
10685 : * backup, and the corresponding timeline ID in *stoptli_p.
10686 : *
10687 : * It is the responsibility of the caller of this function to verify the
10688 : * permissions of the calling user!
10689 : */
10690 : XLogRecPtr
10691 0 : do_pg_stop_backup(char *labelfile, bool waitforarchive, TimeLineID *stoptli_p)
10692 : {
10693 0 : bool exclusive = (labelfile == NULL);
10694 0 : bool backup_started_in_recovery = false;
10695 : XLogRecPtr startpoint;
10696 : XLogRecPtr stoppoint;
10697 : TimeLineID stoptli;
10698 : pg_time_t stamp_time;
10699 : char strfbuf[128];
10700 : char histfilepath[MAXPGPATH];
10701 : char startxlogfilename[MAXFNAMELEN];
10702 : char stopxlogfilename[MAXFNAMELEN];
10703 : char lastxlogfilename[MAXFNAMELEN];
10704 : char histfilename[MAXFNAMELEN];
10705 : char backupfrom[20];
10706 : XLogSegNo _logSegNo;
10707 : FILE *lfp;
10708 : FILE *fp;
10709 : char ch;
10710 : int seconds_before_warning;
10711 0 : int waits = 0;
10712 0 : bool reported_waiting = false;
10713 : char *remaining;
10714 : char *ptr;
10715 : uint32 hi,
10716 : lo;
10717 :
10718 0 : backup_started_in_recovery = RecoveryInProgress();
10719 :
10720 : /*
10721 : * Currently only non-exclusive backup can be taken during recovery.
10722 : */
10723 0 : if (backup_started_in_recovery && exclusive)
10724 0 : ereport(ERROR,
10725 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10726 : errmsg("recovery is in progress"),
10727 : errhint("WAL control functions cannot be executed during recovery.")));
10728 :
10729 : /*
10730 : * During recovery, we don't need to check WAL level. Because, if WAL
10731 : * level is not sufficient, it's impossible to get here during recovery.
10732 : */
10733 0 : if (!backup_started_in_recovery && !XLogIsNeeded())
10734 0 : ereport(ERROR,
10735 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10736 : errmsg("WAL level not sufficient for making an online backup"),
10737 : errhint("wal_level must be set to \"replica\" or \"logical\" at server start.")));
10738 :
10739 0 : if (exclusive)
10740 : {
10741 : /*
10742 : * At first, mark that we're now stopping an exclusive backup, to
10743 : * ensure that there are no other sessions currently running
10744 : * pg_start_backup() or pg_stop_backup().
10745 : */
10746 0 : WALInsertLockAcquireExclusive();
10747 0 : if (XLogCtl->Insert.exclusiveBackupState != EXCLUSIVE_BACKUP_IN_PROGRESS)
10748 : {
10749 0 : WALInsertLockRelease();
10750 0 : ereport(ERROR,
10751 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10752 : errmsg("exclusive backup not in progress")));
10753 : }
10754 0 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_STOPPING;
10755 0 : WALInsertLockRelease();
10756 :
10757 : /*
10758 : * Remove backup_label. In case of failure, the state for an exclusive
10759 : * backup is switched back to in-progress.
10760 : */
10761 0 : PG_ENSURE_ERROR_CLEANUP(pg_stop_backup_callback, (Datum) BoolGetDatum(exclusive));
10762 : {
10763 : /*
10764 : * Read the existing label file into memory.
10765 : */
10766 : struct stat statbuf;
10767 : int r;
10768 :
10769 0 : if (stat(BACKUP_LABEL_FILE, &statbuf))
10770 : {
10771 : /* should not happen per the upper checks */
10772 0 : if (errno != ENOENT)
10773 0 : ereport(ERROR,
10774 : (errcode_for_file_access(),
10775 : errmsg("could not stat file \"%s\": %m",
10776 : BACKUP_LABEL_FILE)));
10777 0 : ereport(ERROR,
10778 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10779 : errmsg("a backup is not in progress")));
10780 : }
10781 :
10782 0 : lfp = AllocateFile(BACKUP_LABEL_FILE, "r");
10783 0 : if (!lfp)
10784 : {
10785 0 : ereport(ERROR,
10786 : (errcode_for_file_access(),
10787 : errmsg("could not read file \"%s\": %m",
10788 : BACKUP_LABEL_FILE)));
10789 : }
10790 0 : labelfile = palloc(statbuf.st_size + 1);
10791 0 : r = fread(labelfile, statbuf.st_size, 1, lfp);
10792 0 : labelfile[statbuf.st_size] = '\0';
10793 :
10794 : /*
10795 : * Close and remove the backup label file
10796 : */
10797 0 : if (r != 1 || ferror(lfp) || FreeFile(lfp))
10798 0 : ereport(ERROR,
10799 : (errcode_for_file_access(),
10800 : errmsg("could not read file \"%s\": %m",
10801 : BACKUP_LABEL_FILE)));
10802 0 : durable_unlink(BACKUP_LABEL_FILE, ERROR);
10803 :
10804 : /*
10805 : * Remove tablespace_map file if present, it is created only if
10806 : * there are tablespaces.
10807 : */
10808 0 : durable_unlink(TABLESPACE_MAP, DEBUG1);
10809 : }
10810 0 : PG_END_ENSURE_ERROR_CLEANUP(pg_stop_backup_callback, (Datum) BoolGetDatum(exclusive));
10811 : }
10812 :
10813 : /*
10814 : * OK to update backup counters and forcePageWrites
10815 : */
10816 0 : WALInsertLockAcquireExclusive();
10817 0 : if (exclusive)
10818 : {
10819 0 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_NONE;
10820 : }
10821 : else
10822 : {
10823 : /*
10824 : * The user-visible pg_start/stop_backup() functions that operate on
10825 : * exclusive backups can be called at any time, but for non-exclusive
10826 : * backups, it is expected that each do_pg_start_backup() call is
10827 : * matched by exactly one do_pg_stop_backup() call.
10828 : */
10829 0 : Assert(XLogCtl->Insert.nonExclusiveBackups > 0);
10830 0 : XLogCtl->Insert.nonExclusiveBackups--;
10831 : }
10832 :
10833 0 : if (XLogCtl->Insert.exclusiveBackupState == EXCLUSIVE_BACKUP_NONE &&
10834 0 : XLogCtl->Insert.nonExclusiveBackups == 0)
10835 : {
10836 0 : XLogCtl->Insert.forcePageWrites = false;
10837 : }
10838 0 : WALInsertLockRelease();
10839 :
10840 : /* Clean up session-level lock */
10841 0 : sessionBackupState = SESSION_BACKUP_NONE;
10842 :
10843 : /*
10844 : * Read and parse the START WAL LOCATION line (this code is pretty crude,
10845 : * but we are not expecting any variability in the file format).
10846 : */
10847 0 : if (sscanf(labelfile, "START WAL LOCATION: %X/%X (file %24s)%c",
10848 : &hi, &lo, startxlogfilename,
10849 0 : &ch) != 4 || ch != '\n')
10850 0 : ereport(ERROR,
10851 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10852 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
10853 0 : startpoint = ((uint64) hi) << 32 | lo;
10854 0 : remaining = strchr(labelfile, '\n') + 1; /* %n is not portable enough */
10855 :
10856 : /*
10857 : * Parse the BACKUP FROM line. If we are taking an online backup from the
10858 : * standby, we confirm that the standby has not been promoted during the
10859 : * backup.
10860 : */
10861 0 : ptr = strstr(remaining, "BACKUP FROM:");
10862 0 : if (!ptr || sscanf(ptr, "BACKUP FROM: %19s\n", backupfrom) != 1)
10863 0 : ereport(ERROR,
10864 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10865 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
10866 0 : if (strcmp(backupfrom, "standby") == 0 && !backup_started_in_recovery)
10867 0 : ereport(ERROR,
10868 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10869 : errmsg("the standby was promoted during online backup"),
10870 : errhint("This means that the backup being taken is corrupt "
10871 : "and should not be used. "
10872 : "Try taking another online backup.")));
10873 :
10874 : /*
10875 : * During recovery, we don't write an end-of-backup record. We assume that
10876 : * pg_control was backed up last and its minimum recovery point can be
10877 : * available as the backup end location. Since we don't have an
10878 : * end-of-backup record, we use the pg_control value to check whether
10879 : * we've reached the end of backup when starting recovery from this
10880 : * backup. We have no way of checking if pg_control wasn't backed up last
10881 : * however.
10882 : *
10883 : * We don't force a switch to new WAL file but it is still possible to
10884 : * wait for all the required files to be archived if waitforarchive is
10885 : * true. This is okay if we use the backup to start a standby and fetch
10886 : * the missing WAL using streaming replication. But in the case of an
10887 : * archive recovery, a user should set waitforarchive to true and wait for
10888 : * them to be archived to ensure that all the required files are
10889 : * available.
10890 : *
10891 : * We return the current minimum recovery point as the backup end
10892 : * location. Note that it can be greater than the exact backup end
10893 : * location if the minimum recovery point is updated after the backup of
10894 : * pg_control. This is harmless for current uses.
10895 : *
10896 : * XXX currently a backup history file is for informational and debug
10897 : * purposes only. It's not essential for an online backup. Furthermore,
10898 : * even if it's created, it will not be archived during recovery because
10899 : * an archiver is not invoked. So it doesn't seem worthwhile to write a
10900 : * backup history file during recovery.
10901 : */
10902 0 : if (backup_started_in_recovery)
10903 : {
10904 : XLogRecPtr recptr;
10905 :
10906 : /*
10907 : * Check to see if all WAL replayed during online backup contain
10908 : * full-page writes.
10909 : */
10910 0 : SpinLockAcquire(&XLogCtl->info_lck);
10911 0 : recptr = XLogCtl->lastFpwDisableRecPtr;
10912 0 : SpinLockRelease(&XLogCtl->info_lck);
10913 :
10914 0 : if (startpoint <= recptr)
10915 0 : ereport(ERROR,
10916 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10917 : errmsg("WAL generated with full_page_writes=off was replayed "
10918 : "during online backup"),
10919 : errhint("This means that the backup being taken on the standby "
10920 : "is corrupt and should not be used. "
10921 : "Enable full_page_writes and run CHECKPOINT on the master, "
10922 : "and then try an online backup again.")));
10923 :
10924 :
10925 0 : LWLockAcquire(ControlFileLock, LW_SHARED);
10926 0 : stoppoint = ControlFile->minRecoveryPoint;
10927 0 : stoptli = ControlFile->minRecoveryPointTLI;
10928 0 : LWLockRelease(ControlFileLock);
10929 : }
10930 : else
10931 : {
10932 : /*
10933 : * Write the backup-end xlog record
10934 : */
10935 0 : XLogBeginInsert();
10936 0 : XLogRegisterData((char *) (&startpoint), sizeof(startpoint));
10937 0 : stoppoint = XLogInsert(RM_XLOG_ID, XLOG_BACKUP_END);
10938 0 : stoptli = ThisTimeLineID;
10939 :
10940 : /*
10941 : * Force a switch to a new xlog segment file, so that the backup is
10942 : * valid as soon as archiver moves out the current segment file.
10943 : */
10944 0 : RequestXLogSwitch(false);
10945 :
10946 0 : XLByteToPrevSeg(stoppoint, _logSegNo);
10947 0 : XLogFileName(stopxlogfilename, stoptli, _logSegNo);
10948 :
10949 : /* Use the log timezone here, not the session timezone */
10950 0 : stamp_time = (pg_time_t) time(NULL);
10951 0 : pg_strftime(strfbuf, sizeof(strfbuf),
10952 : "%Y-%m-%d %H:%M:%S %Z",
10953 0 : pg_localtime(&stamp_time, log_timezone));
10954 :
10955 : /*
10956 : * Write the backup history file
10957 : */
10958 0 : XLByteToSeg(startpoint, _logSegNo);
10959 0 : BackupHistoryFilePath(histfilepath, stoptli, _logSegNo,
10960 : (uint32) (startpoint % XLogSegSize));
10961 0 : fp = AllocateFile(histfilepath, "w");
10962 0 : if (!fp)
10963 0 : ereport(ERROR,
10964 : (errcode_for_file_access(),
10965 : errmsg("could not create file \"%s\": %m",
10966 : histfilepath)));
10967 0 : fprintf(fp, "START WAL LOCATION: %X/%X (file %s)\n",
10968 0 : (uint32) (startpoint >> 32), (uint32) startpoint, startxlogfilename);
10969 0 : fprintf(fp, "STOP WAL LOCATION: %X/%X (file %s)\n",
10970 0 : (uint32) (stoppoint >> 32), (uint32) stoppoint, stopxlogfilename);
10971 : /* transfer remaining lines from label to history file */
10972 0 : fprintf(fp, "%s", remaining);
10973 0 : fprintf(fp, "STOP TIME: %s\n", strfbuf);
10974 0 : if (fflush(fp) || ferror(fp) || FreeFile(fp))
10975 0 : ereport(ERROR,
10976 : (errcode_for_file_access(),
10977 : errmsg("could not write file \"%s\": %m",
10978 : histfilepath)));
10979 :
10980 : /*
10981 : * Clean out any no-longer-needed history files. As a side effect,
10982 : * this will post a .ready file for the newly created history file,
10983 : * notifying the archiver that history file may be archived
10984 : * immediately.
10985 : */
10986 0 : CleanupBackupHistory();
10987 : }
10988 :
10989 : /*
10990 : * If archiving is enabled, wait for all the required WAL files to be
10991 : * archived before returning. If archiving isn't enabled, the required WAL
10992 : * needs to be transported via streaming replication (hopefully with
10993 : * wal_keep_segments set high enough), or some more exotic mechanism like
10994 : * polling and copying files from pg_wal with script. We have no knowledge
10995 : * of those mechanisms, so it's up to the user to ensure that he gets all
10996 : * the required WAL.
10997 : *
10998 : * We wait until both the last WAL file filled during backup and the
10999 : * history file have been archived, and assume that the alphabetic sorting
11000 : * property of the WAL files ensures any earlier WAL files are safely
11001 : * archived as well.
11002 : *
11003 : * We wait forever, since archive_command is supposed to work and we
11004 : * assume the admin wanted his backup to work completely. If you don't
11005 : * wish to wait, then either waitforarchive should be passed in as false,
11006 : * or you can set statement_timeout. Also, some notices are issued to
11007 : * clue in anyone who might be doing this interactively.
11008 : */
11009 :
11010 0 : if (waitforarchive &&
11011 0 : ((!backup_started_in_recovery && XLogArchivingActive()) ||
11012 0 : (backup_started_in_recovery && XLogArchivingAlways())))
11013 : {
11014 0 : XLByteToPrevSeg(stoppoint, _logSegNo);
11015 0 : XLogFileName(lastxlogfilename, stoptli, _logSegNo);
11016 :
11017 0 : XLByteToSeg(startpoint, _logSegNo);
11018 0 : BackupHistoryFileName(histfilename, stoptli, _logSegNo,
11019 : (uint32) (startpoint % XLogSegSize));
11020 :
11021 0 : seconds_before_warning = 60;
11022 0 : waits = 0;
11023 :
11024 0 : while (XLogArchiveIsBusy(lastxlogfilename) ||
11025 0 : XLogArchiveIsBusy(histfilename))
11026 : {
11027 0 : CHECK_FOR_INTERRUPTS();
11028 :
11029 0 : if (!reported_waiting && waits > 5)
11030 : {
11031 0 : ereport(NOTICE,
11032 : (errmsg("pg_stop_backup cleanup done, waiting for required WAL segments to be archived")));
11033 0 : reported_waiting = true;
11034 : }
11035 :
11036 0 : pg_usleep(1000000L);
11037 :
11038 0 : if (++waits >= seconds_before_warning)
11039 : {
11040 0 : seconds_before_warning *= 2; /* This wraps in >10 years... */
11041 0 : ereport(WARNING,
11042 : (errmsg("pg_stop_backup still waiting for all required WAL segments to be archived (%d seconds elapsed)",
11043 : waits),
11044 : errhint("Check that your archive_command is executing properly. "
11045 : "pg_stop_backup can be canceled safely, "
11046 : "but the database backup will not be usable without all the WAL segments.")));
11047 : }
11048 : }
11049 :
11050 0 : ereport(NOTICE,
11051 : (errmsg("pg_stop_backup complete, all required WAL segments have been archived")));
11052 : }
11053 0 : else if (waitforarchive)
11054 0 : ereport(NOTICE,
11055 : (errmsg("WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup")));
11056 :
11057 : /*
11058 : * We're done. As a convenience, return the ending WAL location.
11059 : */
11060 0 : if (stoptli_p)
11061 0 : *stoptli_p = stoptli;
11062 0 : return stoppoint;
11063 : }
11064 :
11065 :
11066 : /*
11067 : * do_pg_abort_backup: abort a running backup
11068 : *
11069 : * This does just the most basic steps of do_pg_stop_backup(), by taking the
11070 : * system out of backup mode, thus making it a lot more safe to call from
11071 : * an error handler.
11072 : *
11073 : * NB: This is only for aborting a non-exclusive backup that doesn't write
11074 : * backup_label. A backup started with pg_start_backup() needs to be finished
11075 : * with pg_stop_backup().
11076 : */
11077 : void
11078 0 : do_pg_abort_backup(void)
11079 : {
11080 0 : WALInsertLockAcquireExclusive();
11081 0 : Assert(XLogCtl->Insert.nonExclusiveBackups > 0);
11082 0 : XLogCtl->Insert.nonExclusiveBackups--;
11083 :
11084 0 : if (XLogCtl->Insert.exclusiveBackupState == EXCLUSIVE_BACKUP_NONE &&
11085 0 : XLogCtl->Insert.nonExclusiveBackups == 0)
11086 : {
11087 0 : XLogCtl->Insert.forcePageWrites = false;
11088 : }
11089 0 : WALInsertLockRelease();
11090 0 : }
11091 :
11092 : /*
11093 : * Get latest redo apply position.
11094 : *
11095 : * Exported to allow WALReceiver to read the pointer directly.
11096 : */
11097 : XLogRecPtr
11098 0 : GetXLogReplayRecPtr(TimeLineID *replayTLI)
11099 : {
11100 : XLogRecPtr recptr;
11101 : TimeLineID tli;
11102 :
11103 0 : SpinLockAcquire(&XLogCtl->info_lck);
11104 0 : recptr = XLogCtl->lastReplayedEndRecPtr;
11105 0 : tli = XLogCtl->lastReplayedTLI;
11106 0 : SpinLockRelease(&XLogCtl->info_lck);
11107 :
11108 0 : if (replayTLI)
11109 0 : *replayTLI = tli;
11110 0 : return recptr;
11111 : }
11112 :
11113 : /*
11114 : * Get latest WAL insert pointer
11115 : */
11116 : XLogRecPtr
11117 0 : GetXLogInsertRecPtr(void)
11118 : {
11119 0 : XLogCtlInsert *Insert = &XLogCtl->Insert;
11120 : uint64 current_bytepos;
11121 :
11122 0 : SpinLockAcquire(&Insert->insertpos_lck);
11123 0 : current_bytepos = Insert->CurrBytePos;
11124 0 : SpinLockRelease(&Insert->insertpos_lck);
11125 :
11126 0 : return XLogBytePosToRecPtr(current_bytepos);
11127 : }
11128 :
11129 : /*
11130 : * Get latest WAL write pointer
11131 : */
11132 : XLogRecPtr
11133 0 : GetXLogWriteRecPtr(void)
11134 : {
11135 0 : SpinLockAcquire(&XLogCtl->info_lck);
11136 0 : LogwrtResult = XLogCtl->LogwrtResult;
11137 0 : SpinLockRelease(&XLogCtl->info_lck);
11138 :
11139 0 : return LogwrtResult.Write;
11140 : }
11141 :
11142 : /*
11143 : * Returns the redo pointer of the last checkpoint or restartpoint. This is
11144 : * the oldest point in WAL that we still need, if we have to restart recovery.
11145 : */
11146 : void
11147 0 : GetOldestRestartPoint(XLogRecPtr *oldrecptr, TimeLineID *oldtli)
11148 : {
11149 0 : LWLockAcquire(ControlFileLock, LW_SHARED);
11150 0 : *oldrecptr = ControlFile->checkPointCopy.redo;
11151 0 : *oldtli = ControlFile->checkPointCopy.ThisTimeLineID;
11152 0 : LWLockRelease(ControlFileLock);
11153 0 : }
11154 :
11155 : /*
11156 : * read_backup_label: check to see if a backup_label file is present
11157 : *
11158 : * If we see a backup_label during recovery, we assume that we are recovering
11159 : * from a backup dump file, and we therefore roll forward from the checkpoint
11160 : * identified by the label file, NOT what pg_control says. This avoids the
11161 : * problem that pg_control might have been archived one or more checkpoints
11162 : * later than the start of the dump, and so if we rely on it as the start
11163 : * point, we will fail to restore a consistent database state.
11164 : *
11165 : * Returns TRUE if a backup_label was found (and fills the checkpoint
11166 : * location and its REDO location into *checkPointLoc and RedoStartLSN,
11167 : * respectively); returns FALSE if not. If this backup_label came from a
11168 : * streamed backup, *backupEndRequired is set to TRUE. If this backup_label
11169 : * was created during recovery, *backupFromStandby is set to TRUE.
11170 : */
11171 : static bool
11172 3 : read_backup_label(XLogRecPtr *checkPointLoc, bool *backupEndRequired,
11173 : bool *backupFromStandby)
11174 : {
11175 : char startxlogfilename[MAXFNAMELEN];
11176 : TimeLineID tli;
11177 : FILE *lfp;
11178 : char ch;
11179 : char backuptype[20];
11180 : char backupfrom[20];
11181 : uint32 hi,
11182 : lo;
11183 :
11184 3 : *backupEndRequired = false;
11185 3 : *backupFromStandby = false;
11186 :
11187 : /*
11188 : * See if label file is present
11189 : */
11190 3 : lfp = AllocateFile(BACKUP_LABEL_FILE, "r");
11191 3 : if (!lfp)
11192 : {
11193 3 : if (errno != ENOENT)
11194 0 : ereport(FATAL,
11195 : (errcode_for_file_access(),
11196 : errmsg("could not read file \"%s\": %m",
11197 : BACKUP_LABEL_FILE)));
11198 3 : return false; /* it's not there, all is fine */
11199 : }
11200 :
11201 : /*
11202 : * Read and parse the START WAL LOCATION and CHECKPOINT lines (this code
11203 : * is pretty crude, but we are not expecting any variability in the file
11204 : * format).
11205 : */
11206 0 : if (fscanf(lfp, "START WAL LOCATION: %X/%X (file %08X%16s)%c",
11207 0 : &hi, &lo, &tli, startxlogfilename, &ch) != 5 || ch != '\n')
11208 0 : ereport(FATAL,
11209 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
11210 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
11211 0 : RedoStartLSN = ((uint64) hi) << 32 | lo;
11212 0 : if (fscanf(lfp, "CHECKPOINT LOCATION: %X/%X%c",
11213 0 : &hi, &lo, &ch) != 3 || ch != '\n')
11214 0 : ereport(FATAL,
11215 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
11216 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
11217 0 : *checkPointLoc = ((uint64) hi) << 32 | lo;
11218 :
11219 : /*
11220 : * BACKUP METHOD and BACKUP FROM lines are new in 9.2. We can't restore
11221 : * from an older backup anyway, but since the information on it is not
11222 : * strictly required, don't error out if it's missing for some reason.
11223 : */
11224 0 : if (fscanf(lfp, "BACKUP METHOD: %19s\n", backuptype) == 1)
11225 : {
11226 0 : if (strcmp(backuptype, "streamed") == 0)
11227 0 : *backupEndRequired = true;
11228 : }
11229 :
11230 0 : if (fscanf(lfp, "BACKUP FROM: %19s\n", backupfrom) == 1)
11231 : {
11232 0 : if (strcmp(backupfrom, "standby") == 0)
11233 0 : *backupFromStandby = true;
11234 : }
11235 :
11236 0 : if (ferror(lfp) || FreeFile(lfp))
11237 0 : ereport(FATAL,
11238 : (errcode_for_file_access(),
11239 : errmsg("could not read file \"%s\": %m",
11240 : BACKUP_LABEL_FILE)));
11241 :
11242 0 : return true;
11243 : }
11244 :
11245 : /*
11246 : * read_tablespace_map: check to see if a tablespace_map file is present
11247 : *
11248 : * If we see a tablespace_map file during recovery, we assume that we are
11249 : * recovering from a backup dump file, and we therefore need to create symlinks
11250 : * as per the information present in tablespace_map file.
11251 : *
11252 : * Returns TRUE if a tablespace_map file was found (and fills the link
11253 : * information for all the tablespace links present in file); returns FALSE
11254 : * if not.
11255 : */
11256 : static bool
11257 0 : read_tablespace_map(List **tablespaces)
11258 : {
11259 : tablespaceinfo *ti;
11260 : FILE *lfp;
11261 : char tbsoid[MAXPGPATH];
11262 : char *tbslinkpath;
11263 : char str[MAXPGPATH];
11264 : int ch,
11265 0 : prev_ch = -1,
11266 0 : i = 0,
11267 : n;
11268 :
11269 : /*
11270 : * See if tablespace_map file is present
11271 : */
11272 0 : lfp = AllocateFile(TABLESPACE_MAP, "r");
11273 0 : if (!lfp)
11274 : {
11275 0 : if (errno != ENOENT)
11276 0 : ereport(FATAL,
11277 : (errcode_for_file_access(),
11278 : errmsg("could not read file \"%s\": %m",
11279 : TABLESPACE_MAP)));
11280 0 : return false; /* it's not there, all is fine */
11281 : }
11282 :
11283 : /*
11284 : * Read and parse the link name and path lines from tablespace_map file
11285 : * (this code is pretty crude, but we are not expecting any variability in
11286 : * the file format). While taking backup we embed escape character '\\'
11287 : * before newline in tablespace path, so that during reading of
11288 : * tablespace_map file, we could distinguish newline in tablespace path
11289 : * and end of line. Now while reading tablespace_map file, remove the
11290 : * escape character that has been added in tablespace path during backup.
11291 : */
11292 0 : while ((ch = fgetc(lfp)) != EOF)
11293 : {
11294 0 : if ((ch == '\n' || ch == '\r') && prev_ch != '\\')
11295 : {
11296 0 : str[i] = '\0';
11297 0 : if (sscanf(str, "%s %n", tbsoid, &n) != 1)
11298 0 : ereport(FATAL,
11299 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
11300 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
11301 0 : tbslinkpath = str + n;
11302 0 : i = 0;
11303 :
11304 0 : ti = palloc(sizeof(tablespaceinfo));
11305 0 : ti->oid = pstrdup(tbsoid);
11306 0 : ti->path = pstrdup(tbslinkpath);
11307 :
11308 0 : *tablespaces = lappend(*tablespaces, ti);
11309 0 : continue;
11310 : }
11311 0 : else if ((ch == '\n' || ch == '\r') && prev_ch == '\\')
11312 0 : str[i - 1] = ch;
11313 : else
11314 0 : str[i++] = ch;
11315 0 : prev_ch = ch;
11316 : }
11317 :
11318 0 : if (ferror(lfp) || FreeFile(lfp))
11319 0 : ereport(FATAL,
11320 : (errcode_for_file_access(),
11321 : errmsg("could not read file \"%s\": %m",
11322 : TABLESPACE_MAP)));
11323 :
11324 0 : return true;
11325 : }
11326 :
11327 : /*
11328 : * Error context callback for errors occurring during rm_redo().
11329 : */
11330 : static void
11331 0 : rm_redo_error_callback(void *arg)
11332 : {
11333 0 : XLogReaderState *record = (XLogReaderState *) arg;
11334 : StringInfoData buf;
11335 :
11336 0 : initStringInfo(&buf);
11337 0 : xlog_outdesc(&buf, record);
11338 :
11339 : /* translator: %s is a WAL record description */
11340 0 : errcontext("WAL redo at %X/%X for %s",
11341 0 : (uint32) (record->ReadRecPtr >> 32),
11342 0 : (uint32) record->ReadRecPtr,
11343 : buf.data);
11344 :
11345 0 : pfree(buf.data);
11346 0 : }
11347 :
11348 : /*
11349 : * BackupInProgress: check if online backup mode is active
11350 : *
11351 : * This is done by checking for existence of the "backup_label" file.
11352 : */
11353 : bool
11354 0 : BackupInProgress(void)
11355 : {
11356 : struct stat stat_buf;
11357 :
11358 0 : return (stat(BACKUP_LABEL_FILE, &stat_buf) == 0);
11359 : }
11360 :
11361 : /*
11362 : * CancelBackup: rename the "backup_label" and "tablespace_map"
11363 : * files to cancel backup mode
11364 : *
11365 : * If the "backup_label" file exists, it will be renamed to "backup_label.old".
11366 : * Similarly, if the "tablespace_map" file exists, it will be renamed to
11367 : * "tablespace_map.old".
11368 : *
11369 : * Note that this will render an online backup in progress
11370 : * useless. To correctly finish an online backup, pg_stop_backup must be
11371 : * called.
11372 : */
11373 : void
11374 1 : CancelBackup(void)
11375 : {
11376 : struct stat stat_buf;
11377 :
11378 : /* if the backup_label file is not there, return */
11379 1 : if (stat(BACKUP_LABEL_FILE, &stat_buf) < 0)
11380 2 : return;
11381 :
11382 : /* remove leftover file from previously canceled backup if it exists */
11383 0 : unlink(BACKUP_LABEL_OLD);
11384 :
11385 0 : if (durable_rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, DEBUG1) != 0)
11386 : {
11387 0 : ereport(WARNING,
11388 : (errcode_for_file_access(),
11389 : errmsg("online backup mode was not canceled"),
11390 : errdetail("File \"%s\" could not be renamed to \"%s\": %m.",
11391 : BACKUP_LABEL_FILE, BACKUP_LABEL_OLD)));
11392 0 : return;
11393 : }
11394 :
11395 : /* if the tablespace_map file is not there, return */
11396 0 : if (stat(TABLESPACE_MAP, &stat_buf) < 0)
11397 : {
11398 0 : ereport(LOG,
11399 : (errmsg("online backup mode canceled"),
11400 : errdetail("File \"%s\" was renamed to \"%s\".",
11401 : BACKUP_LABEL_FILE, BACKUP_LABEL_OLD)));
11402 0 : return;
11403 : }
11404 :
11405 : /* remove leftover file from previously canceled backup if it exists */
11406 0 : unlink(TABLESPACE_MAP_OLD);
11407 :
11408 0 : if (durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, DEBUG1) == 0)
11409 : {
11410 0 : ereport(LOG,
11411 : (errmsg("online backup mode canceled"),
11412 : errdetail("Files \"%s\" and \"%s\" were renamed to "
11413 : "\"%s\" and \"%s\", respectively.",
11414 : BACKUP_LABEL_FILE, TABLESPACE_MAP,
11415 : BACKUP_LABEL_OLD, TABLESPACE_MAP_OLD)));
11416 : }
11417 : else
11418 : {
11419 0 : ereport(WARNING,
11420 : (errcode_for_file_access(),
11421 : errmsg("online backup mode canceled"),
11422 : errdetail("File \"%s\" was renamed to \"%s\", but "
11423 : "file \"%s\" could not be renamed to \"%s\": %m.",
11424 : BACKUP_LABEL_FILE, BACKUP_LABEL_OLD,
11425 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
11426 : }
11427 : }
11428 :
11429 : /*
11430 : * Read the XLOG page containing RecPtr into readBuf (if not read already).
11431 : * Returns number of bytes read, if the page is read successfully, or -1
11432 : * in case of errors. When errors occur, they are ereport'ed, but only
11433 : * if they have not been previously reported.
11434 : *
11435 : * This is responsible for restoring files from archive as needed, as well
11436 : * as for waiting for the requested WAL record to arrive in standby mode.
11437 : *
11438 : * 'emode' specifies the log level used for reporting "file not found" or
11439 : * "end of WAL" situations in archive recovery, or in standby mode when a
11440 : * trigger file is found. If set to WARNING or below, XLogPageRead() returns
11441 : * false in those situations, on higher log levels the ereport() won't
11442 : * return.
11443 : *
11444 : * In standby mode, if after a successful return of XLogPageRead() the
11445 : * caller finds the record it's interested in to be broken, it should
11446 : * ereport the error with the level determined by
11447 : * emode_for_corrupt_record(), and then set lastSourceFailed
11448 : * and call XLogPageRead() again with the same arguments. This lets
11449 : * XLogPageRead() to try fetching the record from another source, or to
11450 : * sleep and retry.
11451 : */
11452 : static int
11453 5 : XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr, int reqLen,
11454 : XLogRecPtr targetRecPtr, char *readBuf, TimeLineID *readTLI)
11455 : {
11456 5 : XLogPageReadPrivate *private =
11457 : (XLogPageReadPrivate *) xlogreader->private_data;
11458 5 : int emode = private->emode;
11459 : uint32 targetPageOff;
11460 : XLogSegNo targetSegNo PG_USED_FOR_ASSERTS_ONLY;
11461 :
11462 5 : XLByteToSeg(targetPagePtr, targetSegNo);
11463 5 : targetPageOff = targetPagePtr % XLogSegSize;
11464 :
11465 : /*
11466 : * See if we need to switch to a new segment because the requested record
11467 : * is not in the currently open one.
11468 : */
11469 5 : if (readFile >= 0 && !XLByteInSeg(targetPagePtr, readSegNo))
11470 : {
11471 : /*
11472 : * Request a restartpoint if we've replayed too much xlog since the
11473 : * last one.
11474 : */
11475 0 : if (bgwriterLaunched)
11476 : {
11477 0 : if (XLogCheckpointNeeded(readSegNo))
11478 : {
11479 0 : (void) GetRedoRecPtr();
11480 0 : if (XLogCheckpointNeeded(readSegNo))
11481 0 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
11482 : }
11483 : }
11484 :
11485 0 : close(readFile);
11486 0 : readFile = -1;
11487 0 : readSource = 0;
11488 : }
11489 :
11490 5 : XLByteToSeg(targetPagePtr, readSegNo);
11491 :
11492 : retry:
11493 : /* See if we need to retrieve more data */
11494 7 : if (readFile < 0 ||
11495 2 : (readSource == XLOG_FROM_STREAM &&
11496 0 : receivedUpto < targetPagePtr + reqLen))
11497 : {
11498 6 : if (!WaitForWALToBecomeAvailable(targetPagePtr + reqLen,
11499 3 : private->randAccess,
11500 3 : private->fetching_ckpt,
11501 : targetRecPtr))
11502 : {
11503 0 : if (readFile >= 0)
11504 0 : close(readFile);
11505 0 : readFile = -1;
11506 0 : readLen = 0;
11507 0 : readSource = 0;
11508 :
11509 0 : return -1;
11510 : }
11511 : }
11512 :
11513 : /*
11514 : * At this point, we have the right segment open and if we're streaming we
11515 : * know the requested record is in it.
11516 : */
11517 5 : Assert(readFile != -1);
11518 :
11519 : /*
11520 : * If the current segment is being streamed from master, calculate how
11521 : * much of the current page we have received already. We know the
11522 : * requested record has been received, but this is for the benefit of
11523 : * future calls, to allow quick exit at the top of this function.
11524 : */
11525 5 : if (readSource == XLOG_FROM_STREAM)
11526 : {
11527 0 : if (((targetPagePtr) / XLOG_BLCKSZ) != (receivedUpto / XLOG_BLCKSZ))
11528 0 : readLen = XLOG_BLCKSZ;
11529 : else
11530 0 : readLen = receivedUpto % XLogSegSize - targetPageOff;
11531 : }
11532 : else
11533 5 : readLen = XLOG_BLCKSZ;
11534 :
11535 : /* Read the requested page */
11536 5 : readOff = targetPageOff;
11537 5 : if (lseek(readFile, (off_t) readOff, SEEK_SET) < 0)
11538 : {
11539 : char fname[MAXFNAMELEN];
11540 :
11541 0 : XLogFileName(fname, curFileTLI, readSegNo);
11542 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
11543 : (errcode_for_file_access(),
11544 : errmsg("could not seek in log segment %s to offset %u: %m",
11545 : fname, readOff)));
11546 0 : goto next_record_is_invalid;
11547 : }
11548 :
11549 5 : pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
11550 5 : if (read(readFile, readBuf, XLOG_BLCKSZ) != XLOG_BLCKSZ)
11551 : {
11552 : char fname[MAXFNAMELEN];
11553 :
11554 0 : pgstat_report_wait_end();
11555 0 : XLogFileName(fname, curFileTLI, readSegNo);
11556 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
11557 : (errcode_for_file_access(),
11558 : errmsg("could not read from log segment %s, offset %u: %m",
11559 : fname, readOff)));
11560 0 : goto next_record_is_invalid;
11561 : }
11562 5 : pgstat_report_wait_end();
11563 :
11564 5 : Assert(targetSegNo == readSegNo);
11565 5 : Assert(targetPageOff == readOff);
11566 5 : Assert(reqLen <= readLen);
11567 :
11568 5 : *readTLI = curFileTLI;
11569 5 : return readLen;
11570 :
11571 : next_record_is_invalid:
11572 0 : lastSourceFailed = true;
11573 :
11574 0 : if (readFile >= 0)
11575 0 : close(readFile);
11576 0 : readFile = -1;
11577 0 : readLen = 0;
11578 0 : readSource = 0;
11579 :
11580 : /* In standby-mode, keep trying */
11581 0 : if (StandbyMode)
11582 0 : goto retry;
11583 : else
11584 0 : return -1;
11585 : }
11586 :
11587 : /*
11588 : * Open the WAL segment containing WAL location 'RecPtr'.
11589 : *
11590 : * The segment can be fetched via restore_command, or via walreceiver having
11591 : * streamed the record, or it can already be present in pg_wal. Checking
11592 : * pg_wal is mainly for crash recovery, but it will be polled in standby mode
11593 : * too, in case someone copies a new segment directly to pg_wal. That is not
11594 : * documented or recommended, though.
11595 : *
11596 : * If 'fetching_ckpt' is true, we're fetching a checkpoint record, and should
11597 : * prepare to read WAL starting from RedoStartLSN after this.
11598 : *
11599 : * 'RecPtr' might not point to the beginning of the record we're interested
11600 : * in, it might also point to the page or segment header. In that case,
11601 : * 'tliRecPtr' is the position of the WAL record we're interested in. It is
11602 : * used to decide which timeline to stream the requested WAL from.
11603 : *
11604 : * If the record is not immediately available, the function returns false
11605 : * if we're not in standby mode. In standby mode, waits for it to become
11606 : * available.
11607 : *
11608 : * When the requested record becomes available, the function opens the file
11609 : * containing it (if not open already), and returns true. When end of standby
11610 : * mode is triggered by the user, and there is no more WAL available, returns
11611 : * false.
11612 : */
11613 : static bool
11614 3 : WaitForWALToBecomeAvailable(XLogRecPtr RecPtr, bool randAccess,
11615 : bool fetching_ckpt, XLogRecPtr tliRecPtr)
11616 : {
11617 : static TimestampTz last_fail_time = 0;
11618 : TimestampTz now;
11619 3 : bool streaming_reply_sent = false;
11620 :
11621 : /*-------
11622 : * Standby mode is implemented by a state machine:
11623 : *
11624 : * 1. Read from either archive or pg_wal (XLOG_FROM_ARCHIVE), or just
11625 : * pg_wal (XLOG_FROM_PG_WAL)
11626 : * 2. Check trigger file
11627 : * 3. Read from primary server via walreceiver (XLOG_FROM_STREAM)
11628 : * 4. Rescan timelines
11629 : * 5. Sleep wal_retrieve_retry_interval milliseconds, and loop back to 1.
11630 : *
11631 : * Failure to read from the current source advances the state machine to
11632 : * the next state.
11633 : *
11634 : * 'currentSource' indicates the current state. There are no currentSource
11635 : * values for "check trigger", "rescan timelines", and "sleep" states,
11636 : * those actions are taken when reading from the previous source fails, as
11637 : * part of advancing to the next state.
11638 : *-------
11639 : */
11640 3 : if (!InArchiveRecovery)
11641 3 : currentSource = XLOG_FROM_PG_WAL;
11642 0 : else if (currentSource == 0)
11643 0 : currentSource = XLOG_FROM_ARCHIVE;
11644 :
11645 : for (;;)
11646 : {
11647 3 : int oldSource = currentSource;
11648 :
11649 : /*
11650 : * First check if we failed to read from the current source, and
11651 : * advance the state machine if so. The failure to read might've
11652 : * happened outside this function, e.g when a CRC check fails on a
11653 : * record, or within this loop.
11654 : */
11655 3 : if (lastSourceFailed)
11656 : {
11657 0 : switch (currentSource)
11658 : {
11659 : case XLOG_FROM_ARCHIVE:
11660 : case XLOG_FROM_PG_WAL:
11661 :
11662 : /*
11663 : * Check to see if the trigger file exists. Note that we
11664 : * do this only after failure, so when you create the
11665 : * trigger file, we still finish replaying as much as we
11666 : * can from archive and pg_wal before failover.
11667 : */
11668 0 : if (StandbyMode && CheckForStandbyTrigger())
11669 : {
11670 0 : ShutdownWalRcv();
11671 0 : return false;
11672 : }
11673 :
11674 : /*
11675 : * Not in standby mode, and we've now tried the archive
11676 : * and pg_wal.
11677 : */
11678 0 : if (!StandbyMode)
11679 0 : return false;
11680 :
11681 : /*
11682 : * If primary_conninfo is set, launch walreceiver to try
11683 : * to stream the missing WAL.
11684 : *
11685 : * If fetching_ckpt is TRUE, RecPtr points to the initial
11686 : * checkpoint location. In that case, we use RedoStartLSN
11687 : * as the streaming start position instead of RecPtr, so
11688 : * that when we later jump backwards to start redo at
11689 : * RedoStartLSN, we will have the logs streamed already.
11690 : */
11691 0 : if (PrimaryConnInfo)
11692 : {
11693 : XLogRecPtr ptr;
11694 : TimeLineID tli;
11695 :
11696 0 : if (fetching_ckpt)
11697 : {
11698 0 : ptr = RedoStartLSN;
11699 0 : tli = ControlFile->checkPointCopy.ThisTimeLineID;
11700 : }
11701 : else
11702 : {
11703 0 : ptr = tliRecPtr;
11704 0 : tli = tliOfPointInHistory(tliRecPtr, expectedTLEs);
11705 :
11706 0 : if (curFileTLI > 0 && tli < curFileTLI)
11707 0 : elog(ERROR, "according to history file, WAL location %X/%X belongs to timeline %u, but previous recovered WAL file came from timeline %u",
11708 : (uint32) (ptr >> 32), (uint32) ptr,
11709 : tli, curFileTLI);
11710 : }
11711 0 : curFileTLI = tli;
11712 0 : RequestXLogStreaming(tli, ptr, PrimaryConnInfo,
11713 : PrimarySlotName);
11714 0 : receivedUpto = 0;
11715 : }
11716 :
11717 : /*
11718 : * Move to XLOG_FROM_STREAM state in either case. We'll
11719 : * get immediate failure if we didn't launch walreceiver,
11720 : * and move on to the next state.
11721 : */
11722 0 : currentSource = XLOG_FROM_STREAM;
11723 0 : break;
11724 :
11725 : case XLOG_FROM_STREAM:
11726 :
11727 : /*
11728 : * Failure while streaming. Most likely, we got here
11729 : * because streaming replication was terminated, or
11730 : * promotion was triggered. But we also get here if we
11731 : * find an invalid record in the WAL streamed from master,
11732 : * in which case something is seriously wrong. There's
11733 : * little chance that the problem will just go away, but
11734 : * PANIC is not good for availability either, especially
11735 : * in hot standby mode. So, we treat that the same as
11736 : * disconnection, and retry from archive/pg_wal again. The
11737 : * WAL in the archive should be identical to what was
11738 : * streamed, so it's unlikely that it helps, but one can
11739 : * hope...
11740 : */
11741 :
11742 : /*
11743 : * Before we leave XLOG_FROM_STREAM state, make sure that
11744 : * walreceiver is not active, so that it won't overwrite
11745 : * WAL that we restore from archive.
11746 : */
11747 0 : if (WalRcvStreaming())
11748 0 : ShutdownWalRcv();
11749 :
11750 : /*
11751 : * Before we sleep, re-scan for possible new timelines if
11752 : * we were requested to recover to the latest timeline.
11753 : */
11754 0 : if (recoveryTargetIsLatest)
11755 : {
11756 0 : if (rescanLatestTimeLine())
11757 : {
11758 0 : currentSource = XLOG_FROM_ARCHIVE;
11759 0 : break;
11760 : }
11761 : }
11762 :
11763 : /*
11764 : * XLOG_FROM_STREAM is the last state in our state
11765 : * machine, so we've exhausted all the options for
11766 : * obtaining the requested WAL. We're going to loop back
11767 : * and retry from the archive, but if it hasn't been long
11768 : * since last attempt, sleep wal_retrieve_retry_interval
11769 : * milliseconds to avoid busy-waiting.
11770 : */
11771 0 : now = GetCurrentTimestamp();
11772 0 : if (!TimestampDifferenceExceeds(last_fail_time, now,
11773 : wal_retrieve_retry_interval))
11774 : {
11775 : long secs,
11776 : wait_time;
11777 : int usecs;
11778 :
11779 0 : TimestampDifference(last_fail_time, now, &secs, &usecs);
11780 0 : wait_time = wal_retrieve_retry_interval -
11781 0 : (secs * 1000 + usecs / 1000);
11782 :
11783 0 : WaitLatch(&XLogCtl->recoveryWakeupLatch,
11784 : WL_LATCH_SET | WL_TIMEOUT | WL_POSTMASTER_DEATH,
11785 : wait_time, WAIT_EVENT_RECOVERY_WAL_STREAM);
11786 0 : ResetLatch(&XLogCtl->recoveryWakeupLatch);
11787 0 : now = GetCurrentTimestamp();
11788 : }
11789 0 : last_fail_time = now;
11790 0 : currentSource = XLOG_FROM_ARCHIVE;
11791 0 : break;
11792 :
11793 : default:
11794 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
11795 : }
11796 : }
11797 3 : else if (currentSource == XLOG_FROM_PG_WAL)
11798 : {
11799 : /*
11800 : * We just successfully read a file in pg_wal. We prefer files in
11801 : * the archive over ones in pg_wal, so try the next file again
11802 : * from the archive first.
11803 : */
11804 3 : if (InArchiveRecovery)
11805 0 : currentSource = XLOG_FROM_ARCHIVE;
11806 : }
11807 :
11808 3 : if (currentSource != oldSource)
11809 0 : elog(DEBUG2, "switched WAL source from %s to %s after %s",
11810 : xlogSourceNames[oldSource], xlogSourceNames[currentSource],
11811 : lastSourceFailed ? "failure" : "success");
11812 :
11813 : /*
11814 : * We've now handled possible failure. Try to read from the chosen
11815 : * source.
11816 : */
11817 3 : lastSourceFailed = false;
11818 :
11819 3 : switch (currentSource)
11820 : {
11821 : case XLOG_FROM_ARCHIVE:
11822 : case XLOG_FROM_PG_WAL:
11823 : /* Close any old file we might have open. */
11824 3 : if (readFile >= 0)
11825 : {
11826 0 : close(readFile);
11827 0 : readFile = -1;
11828 : }
11829 : /* Reset curFileTLI if random fetch. */
11830 3 : if (randAccess)
11831 3 : curFileTLI = 0;
11832 :
11833 : /*
11834 : * Try to restore the file from archive, or read an existing
11835 : * file from pg_wal.
11836 : */
11837 3 : readFile = XLogFileReadAnyTLI(readSegNo, DEBUG2,
11838 3 : currentSource == XLOG_FROM_ARCHIVE ? XLOG_FROM_ANY :
11839 : currentSource);
11840 3 : if (readFile >= 0)
11841 3 : return true; /* success! */
11842 :
11843 : /*
11844 : * Nope, not found in archive or pg_wal.
11845 : */
11846 0 : lastSourceFailed = true;
11847 0 : break;
11848 :
11849 : case XLOG_FROM_STREAM:
11850 : {
11851 : bool havedata;
11852 :
11853 : /*
11854 : * Check if WAL receiver is still active.
11855 : */
11856 0 : if (!WalRcvStreaming())
11857 : {
11858 0 : lastSourceFailed = true;
11859 0 : break;
11860 : }
11861 :
11862 : /*
11863 : * Walreceiver is active, so see if new data has arrived.
11864 : *
11865 : * We only advance XLogReceiptTime when we obtain fresh
11866 : * WAL from walreceiver and observe that we had already
11867 : * processed everything before the most recent "chunk"
11868 : * that it flushed to disk. In steady state where we are
11869 : * keeping up with the incoming data, XLogReceiptTime will
11870 : * be updated on each cycle. When we are behind,
11871 : * XLogReceiptTime will not advance, so the grace time
11872 : * allotted to conflicting queries will decrease.
11873 : */
11874 0 : if (RecPtr < receivedUpto)
11875 0 : havedata = true;
11876 : else
11877 : {
11878 : XLogRecPtr latestChunkStart;
11879 :
11880 0 : receivedUpto = GetWalRcvWriteRecPtr(&latestChunkStart, &receiveTLI);
11881 0 : if (RecPtr < receivedUpto && receiveTLI == curFileTLI)
11882 : {
11883 0 : havedata = true;
11884 0 : if (latestChunkStart <= RecPtr)
11885 : {
11886 0 : XLogReceiptTime = GetCurrentTimestamp();
11887 0 : SetCurrentChunkStartTime(XLogReceiptTime);
11888 : }
11889 : }
11890 : else
11891 0 : havedata = false;
11892 : }
11893 0 : if (havedata)
11894 : {
11895 : /*
11896 : * Great, streamed far enough. Open the file if it's
11897 : * not open already. Also read the timeline history
11898 : * file if we haven't initialized timeline history
11899 : * yet; it should be streamed over and present in
11900 : * pg_wal by now. Use XLOG_FROM_STREAM so that source
11901 : * info is set correctly and XLogReceiptTime isn't
11902 : * changed.
11903 : */
11904 0 : if (readFile < 0)
11905 : {
11906 0 : if (!expectedTLEs)
11907 0 : expectedTLEs = readTimeLineHistory(receiveTLI);
11908 0 : readFile = XLogFileRead(readSegNo, PANIC,
11909 : receiveTLI,
11910 : XLOG_FROM_STREAM, false);
11911 0 : Assert(readFile >= 0);
11912 : }
11913 : else
11914 : {
11915 : /* just make sure source info is correct... */
11916 0 : readSource = XLOG_FROM_STREAM;
11917 0 : XLogReceiptSource = XLOG_FROM_STREAM;
11918 0 : return true;
11919 : }
11920 0 : break;
11921 : }
11922 :
11923 : /*
11924 : * Data not here yet. Check for trigger, then wait for
11925 : * walreceiver to wake us up when new WAL arrives.
11926 : */
11927 0 : if (CheckForStandbyTrigger())
11928 : {
11929 : /*
11930 : * Note that we don't "return false" immediately here.
11931 : * After being triggered, we still want to replay all
11932 : * the WAL that was already streamed. It's in pg_wal
11933 : * now, so we just treat this as a failure, and the
11934 : * state machine will move on to replay the streamed
11935 : * WAL from pg_wal, and then recheck the trigger and
11936 : * exit replay.
11937 : */
11938 0 : lastSourceFailed = true;
11939 0 : break;
11940 : }
11941 :
11942 : /*
11943 : * Since we have replayed everything we have received so
11944 : * far and are about to start waiting for more WAL, let's
11945 : * tell the upstream server our replay location now so
11946 : * that pg_stat_replication doesn't show stale
11947 : * information.
11948 : */
11949 0 : if (!streaming_reply_sent)
11950 : {
11951 0 : WalRcvForceReply();
11952 0 : streaming_reply_sent = true;
11953 : }
11954 :
11955 : /*
11956 : * Wait for more WAL to arrive. Time out after 5 seconds
11957 : * to react to a trigger file promptly.
11958 : */
11959 0 : WaitLatch(&XLogCtl->recoveryWakeupLatch,
11960 : WL_LATCH_SET | WL_TIMEOUT | WL_POSTMASTER_DEATH,
11961 : 5000L, WAIT_EVENT_RECOVERY_WAL_ALL);
11962 0 : ResetLatch(&XLogCtl->recoveryWakeupLatch);
11963 0 : break;
11964 : }
11965 :
11966 : default:
11967 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
11968 : }
11969 :
11970 : /*
11971 : * This possibly-long loop needs to handle interrupts of startup
11972 : * process.
11973 : */
11974 0 : HandleStartupProcInterrupts();
11975 0 : }
11976 :
11977 : return false; /* not reached */
11978 : }
11979 :
11980 : /*
11981 : * Determine what log level should be used to report a corrupt WAL record
11982 : * in the current WAL page, previously read by XLogPageRead().
11983 : *
11984 : * 'emode' is the error mode that would be used to report a file-not-found
11985 : * or legitimate end-of-WAL situation. Generally, we use it as-is, but if
11986 : * we're retrying the exact same record that we've tried previously, only
11987 : * complain the first time to keep the noise down. However, we only do when
11988 : * reading from pg_wal, because we don't expect any invalid records in archive
11989 : * or in records streamed from master. Files in the archive should be complete,
11990 : * and we should never hit the end of WAL because we stop and wait for more WAL
11991 : * to arrive before replaying it.
11992 : *
11993 : * NOTE: This function remembers the RecPtr value it was last called with,
11994 : * to suppress repeated messages about the same record. Only call this when
11995 : * you are about to ereport(), or you might cause a later message to be
11996 : * erroneously suppressed.
11997 : */
11998 : static int
11999 0 : emode_for_corrupt_record(int emode, XLogRecPtr RecPtr)
12000 : {
12001 : static XLogRecPtr lastComplaint = 0;
12002 :
12003 0 : if (readSource == XLOG_FROM_PG_WAL && emode == LOG)
12004 : {
12005 0 : if (RecPtr == lastComplaint)
12006 0 : emode = DEBUG1;
12007 : else
12008 0 : lastComplaint = RecPtr;
12009 : }
12010 0 : return emode;
12011 : }
12012 :
12013 : /*
12014 : * Check to see whether the user-specified trigger file exists and whether a
12015 : * promote request has arrived. If either condition holds, return true.
12016 : */
12017 : static bool
12018 0 : CheckForStandbyTrigger(void)
12019 : {
12020 : struct stat stat_buf;
12021 : static bool triggered = false;
12022 :
12023 0 : if (triggered)
12024 0 : return true;
12025 :
12026 0 : if (IsPromoteTriggered())
12027 : {
12028 : /*
12029 : * In 9.1 and 9.2 the postmaster unlinked the promote file inside the
12030 : * signal handler. It now leaves the file in place and lets the
12031 : * Startup process do the unlink. This allows Startup to know whether
12032 : * it should create a full checkpoint before starting up (fallback
12033 : * mode). Fast promotion takes precedence.
12034 : */
12035 0 : if (stat(PROMOTE_SIGNAL_FILE, &stat_buf) == 0)
12036 : {
12037 0 : unlink(PROMOTE_SIGNAL_FILE);
12038 0 : unlink(FALLBACK_PROMOTE_SIGNAL_FILE);
12039 0 : fast_promote = true;
12040 : }
12041 0 : else if (stat(FALLBACK_PROMOTE_SIGNAL_FILE, &stat_buf) == 0)
12042 : {
12043 0 : unlink(FALLBACK_PROMOTE_SIGNAL_FILE);
12044 0 : fast_promote = false;
12045 : }
12046 :
12047 0 : ereport(LOG, (errmsg("received promote request")));
12048 :
12049 0 : ResetPromoteTriggered();
12050 0 : triggered = true;
12051 0 : return true;
12052 : }
12053 :
12054 0 : if (TriggerFile == NULL)
12055 0 : return false;
12056 :
12057 0 : if (stat(TriggerFile, &stat_buf) == 0)
12058 : {
12059 0 : ereport(LOG,
12060 : (errmsg("trigger file found: %s", TriggerFile)));
12061 0 : unlink(TriggerFile);
12062 0 : triggered = true;
12063 0 : fast_promote = true;
12064 0 : return true;
12065 : }
12066 0 : else if (errno != ENOENT)
12067 0 : ereport(ERROR,
12068 : (errcode_for_file_access(),
12069 : errmsg("could not stat trigger file \"%s\": %m",
12070 : TriggerFile)));
12071 :
12072 0 : return false;
12073 : }
12074 :
12075 : /*
12076 : * Remove the files signaling a standby promotion request.
12077 : */
12078 : void
12079 1 : RemovePromoteSignalFiles(void)
12080 : {
12081 1 : unlink(PROMOTE_SIGNAL_FILE);
12082 1 : unlink(FALLBACK_PROMOTE_SIGNAL_FILE);
12083 1 : }
12084 :
12085 : /*
12086 : * Check to see if a promote request has arrived. Should be
12087 : * called by postmaster after receiving SIGUSR1.
12088 : */
12089 : bool
12090 43 : CheckPromoteSignal(void)
12091 : {
12092 : struct stat stat_buf;
12093 :
12094 86 : if (stat(PROMOTE_SIGNAL_FILE, &stat_buf) == 0 ||
12095 43 : stat(FALLBACK_PROMOTE_SIGNAL_FILE, &stat_buf) == 0)
12096 0 : return true;
12097 :
12098 43 : return false;
12099 : }
12100 :
12101 : /*
12102 : * Wake up startup process to replay newly arrived WAL, or to notice that
12103 : * failover has been requested.
12104 : */
12105 : void
12106 0 : WakeupRecovery(void)
12107 : {
12108 0 : SetLatch(&XLogCtl->recoveryWakeupLatch);
12109 0 : }
12110 :
12111 : /*
12112 : * Update the WalWriterSleeping flag.
12113 : */
12114 : void
12115 1 : SetWalWriterSleeping(bool sleeping)
12116 : {
12117 1 : SpinLockAcquire(&XLogCtl->info_lck);
12118 1 : XLogCtl->WalWriterSleeping = sleeping;
12119 1 : SpinLockRelease(&XLogCtl->info_lck);
12120 1 : }
12121 :
12122 : /*
12123 : * Schedule a walreceiver wakeup in the main recovery loop.
12124 : */
12125 : void
12126 0 : XLogRequestWalReceiverReply(void)
12127 : {
12128 0 : doRequestWalReceiverReply = true;
12129 0 : }
|
