Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * xlog.c
4 : * PostgreSQL write-ahead log manager
5 : *
6 : *
7 : * Portions Copyright (c) 1996-2019, 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/heaptoast.h"
28 : #include "access/multixact.h"
29 : #include "access/rewriteheap.h"
30 : #include "access/subtrans.h"
31 : #include "access/timeline.h"
32 : #include "access/transam.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 "common/controldata_utils.h"
44 : #include "miscadmin.h"
45 : #include "pg_trace.h"
46 : #include "pgstat.h"
47 : #include "port/atomics.h"
48 : #include "postmaster/bgwriter.h"
49 : #include "postmaster/startup.h"
50 : #include "postmaster/walwriter.h"
51 : #include "replication/basebackup.h"
52 : #include "replication/logical.h"
53 : #include "replication/origin.h"
54 : #include "replication/slot.h"
55 : #include "replication/snapbuild.h"
56 : #include "replication/walreceiver.h"
57 : #include "replication/walsender.h"
58 : #include "storage/bufmgr.h"
59 : #include "storage/fd.h"
60 : #include "storage/ipc.h"
61 : #include "storage/large_object.h"
62 : #include "storage/latch.h"
63 : #include "storage/pmsignal.h"
64 : #include "storage/predicate.h"
65 : #include "storage/proc.h"
66 : #include "storage/procarray.h"
67 : #include "storage/reinit.h"
68 : #include "storage/smgr.h"
69 : #include "storage/spin.h"
70 : #include "storage/sync.h"
71 : #include "utils/builtins.h"
72 : #include "utils/guc.h"
73 : #include "utils/memutils.h"
74 : #include "utils/ps_status.h"
75 : #include "utils/relmapper.h"
76 : #include "utils/snapmgr.h"
77 : #include "utils/timestamp.h"
78 :
79 : extern uint32 bootstrap_data_checksum_version;
80 :
81 : /* Unsupported old recovery command file names (relative to $PGDATA) */
82 : #define RECOVERY_COMMAND_FILE "recovery.conf"
83 : #define RECOVERY_COMMAND_DONE "recovery.done"
84 :
85 : /* User-settable parameters */
86 : int max_wal_size_mb = 1024; /* 1 GB */
87 : int min_wal_size_mb = 80; /* 80 MB */
88 : int wal_keep_segments = 0;
89 : int XLOGbuffers = -1;
90 : int XLogArchiveTimeout = 0;
91 : int XLogArchiveMode = ARCHIVE_MODE_OFF;
92 : char *XLogArchiveCommand = NULL;
93 : bool EnableHotStandby = false;
94 : bool fullPageWrites = true;
95 : bool wal_log_hints = false;
96 : bool wal_compression = false;
97 : char *wal_consistency_checking_string = NULL;
98 : bool *wal_consistency_checking = NULL;
99 : bool wal_init_zero = true;
100 : bool wal_recycle = true;
101 : bool log_checkpoints = false;
102 : int sync_method = DEFAULT_SYNC_METHOD;
103 : int wal_level = WAL_LEVEL_MINIMAL;
104 : int CommitDelay = 0; /* precommit delay in microseconds */
105 : int CommitSiblings = 5; /* # concurrent xacts needed to sleep */
106 : int wal_retrieve_retry_interval = 5000;
107 :
108 : #ifdef WAL_DEBUG
109 : bool XLOG_DEBUG = false;
110 : #endif
111 :
112 : int wal_segment_size = DEFAULT_XLOG_SEG_SIZE;
113 :
114 : /*
115 : * Number of WAL insertion locks to use. A higher value allows more insertions
116 : * to happen concurrently, but adds some CPU overhead to flushing the WAL,
117 : * which needs to iterate all the locks.
118 : */
119 : #define NUM_XLOGINSERT_LOCKS 8
120 :
121 : /*
122 : * Max distance from last checkpoint, before triggering a new xlog-based
123 : * checkpoint.
124 : */
125 : int CheckPointSegments;
126 :
127 : /* Estimated distance between checkpoints, in bytes */
128 : static double CheckPointDistanceEstimate = 0;
129 : static double PrevCheckPointDistance = 0;
130 :
131 : /*
132 : * GUC support
133 : */
134 : const struct config_enum_entry sync_method_options[] = {
135 : {"fsync", SYNC_METHOD_FSYNC, false},
136 : #ifdef HAVE_FSYNC_WRITETHROUGH
137 : {"fsync_writethrough", SYNC_METHOD_FSYNC_WRITETHROUGH, false},
138 : #endif
139 : #ifdef HAVE_FDATASYNC
140 : {"fdatasync", SYNC_METHOD_FDATASYNC, false},
141 : #endif
142 : #ifdef OPEN_SYNC_FLAG
143 : {"open_sync", SYNC_METHOD_OPEN, false},
144 : #endif
145 : #ifdef OPEN_DATASYNC_FLAG
146 : {"open_datasync", SYNC_METHOD_OPEN_DSYNC, false},
147 : #endif
148 : {NULL, 0, false}
149 : };
150 :
151 :
152 : /*
153 : * Although only "on", "off", and "always" are documented,
154 : * we accept all the likely variants of "on" and "off".
155 : */
156 : const struct config_enum_entry archive_mode_options[] = {
157 : {"always", ARCHIVE_MODE_ALWAYS, false},
158 : {"on", ARCHIVE_MODE_ON, false},
159 : {"off", ARCHIVE_MODE_OFF, false},
160 : {"true", ARCHIVE_MODE_ON, true},
161 : {"false", ARCHIVE_MODE_OFF, true},
162 : {"yes", ARCHIVE_MODE_ON, true},
163 : {"no", ARCHIVE_MODE_OFF, true},
164 : {"1", ARCHIVE_MODE_ON, true},
165 : {"0", ARCHIVE_MODE_OFF, true},
166 : {NULL, 0, false}
167 : };
168 :
169 : const struct config_enum_entry recovery_target_action_options[] = {
170 : {"pause", RECOVERY_TARGET_ACTION_PAUSE, false},
171 : {"promote", RECOVERY_TARGET_ACTION_PROMOTE, false},
172 : {"shutdown", RECOVERY_TARGET_ACTION_SHUTDOWN, false},
173 : {NULL, 0, false}
174 : };
175 :
176 : /*
177 : * Statistics for current checkpoint are collected in this global struct.
178 : * Because only the checkpointer or a stand-alone backend can perform
179 : * checkpoints, this will be unused in normal backends.
180 : */
181 : CheckpointStatsData CheckpointStats;
182 :
183 : /*
184 : * ThisTimeLineID will be same in all backends --- it identifies current
185 : * WAL timeline for the database system.
186 : */
187 : TimeLineID ThisTimeLineID = 0;
188 :
189 : /*
190 : * Are we doing recovery from XLOG?
191 : *
192 : * This is only ever true in the startup process; it should be read as meaning
193 : * "this process is replaying WAL records", rather than "the system is in
194 : * recovery mode". It should be examined primarily by functions that need
195 : * to act differently when called from a WAL redo function (e.g., to skip WAL
196 : * logging). To check whether the system is in recovery regardless of which
197 : * process you're running in, use RecoveryInProgress() but only after shared
198 : * memory startup and lock initialization.
199 : */
200 : bool InRecovery = false;
201 :
202 : /* Are we in Hot Standby mode? Only valid in startup process, see xlog.h */
203 : HotStandbyState standbyState = STANDBY_DISABLED;
204 :
205 : static XLogRecPtr LastRec;
206 :
207 : /* Local copy of WalRcv->receivedUpto */
208 : static XLogRecPtr receivedUpto = 0;
209 : static TimeLineID receiveTLI = 0;
210 :
211 : /*
212 : * During recovery, lastFullPageWrites keeps track of full_page_writes that
213 : * the replayed WAL records indicate. It's initialized with full_page_writes
214 : * that the recovery starting checkpoint record indicates, and then updated
215 : * each time XLOG_FPW_CHANGE record is replayed.
216 : */
217 : static bool lastFullPageWrites;
218 :
219 : /*
220 : * Local copy of SharedRecoveryInProgress variable. True actually means "not
221 : * known, need to check the shared state".
222 : */
223 : static bool LocalRecoveryInProgress = true;
224 :
225 : /*
226 : * Local copy of SharedHotStandbyActive variable. False actually means "not
227 : * known, need to check the shared state".
228 : */
229 : static bool LocalHotStandbyActive = false;
230 :
231 : /*
232 : * Local state for XLogInsertAllowed():
233 : * 1: unconditionally allowed to insert XLOG
234 : * 0: unconditionally not allowed to insert XLOG
235 : * -1: must check RecoveryInProgress(); disallow until it is false
236 : * Most processes start with -1 and transition to 1 after seeing that recovery
237 : * is not in progress. But we can also force the value for special cases.
238 : * The coding in XLogInsertAllowed() depends on the first two of these states
239 : * being numerically the same as bool true and false.
240 : */
241 : static int LocalXLogInsertAllowed = -1;
242 :
243 : /*
244 : * When ArchiveRecoveryRequested is set, archive recovery was requested,
245 : * ie. signal files were present. When InArchiveRecovery is set, we are
246 : * currently recovering using offline XLOG archives. These variables are only
247 : * valid in the startup process.
248 : *
249 : * When ArchiveRecoveryRequested is true, but InArchiveRecovery is false, we're
250 : * currently performing crash recovery using only XLOG files in pg_wal, but
251 : * will switch to using offline XLOG archives as soon as we reach the end of
252 : * WAL in pg_wal.
253 : */
254 : bool ArchiveRecoveryRequested = false;
255 : bool InArchiveRecovery = false;
256 :
257 : static bool standby_signal_file_found = false;
258 : static bool recovery_signal_file_found = false;
259 :
260 : /* Was the last xlog file restored from archive, or local? */
261 : static bool restoredFromArchive = false;
262 :
263 : /* Buffers dedicated to consistency checks of size BLCKSZ */
264 : static char *replay_image_masked = NULL;
265 : static char *master_image_masked = NULL;
266 :
267 : /* options formerly taken from recovery.conf for archive recovery */
268 : char *recoveryRestoreCommand = NULL;
269 : char *recoveryEndCommand = NULL;
270 : char *archiveCleanupCommand = NULL;
271 : RecoveryTargetType recoveryTarget = RECOVERY_TARGET_UNSET;
272 : bool recoveryTargetInclusive = true;
273 : int recoveryTargetAction = RECOVERY_TARGET_ACTION_PAUSE;
274 : TransactionId recoveryTargetXid;
275 : char *recovery_target_time_string;
276 : static TimestampTz recoveryTargetTime;
277 : const char *recoveryTargetName;
278 : XLogRecPtr recoveryTargetLSN;
279 : int recovery_min_apply_delay = 0;
280 : TimestampTz recoveryDelayUntilTime;
281 :
282 : /* options formerly taken from recovery.conf for XLOG streaming */
283 : bool StandbyModeRequested = false;
284 : char *PrimaryConnInfo = NULL;
285 : char *PrimarySlotName = NULL;
286 : char *PromoteTriggerFile = NULL;
287 :
288 : /* are we currently in standby mode? */
289 : bool StandbyMode = false;
290 :
291 : /* whether request for fast promotion has been made yet */
292 : static bool fast_promote = false;
293 :
294 : /*
295 : * if recoveryStopsBefore/After returns true, it saves information of the stop
296 : * point here
297 : */
298 : static TransactionId recoveryStopXid;
299 : static TimestampTz recoveryStopTime;
300 : static XLogRecPtr recoveryStopLSN;
301 : static char recoveryStopName[MAXFNAMELEN];
302 : static bool recoveryStopAfter;
303 :
304 : /*
305 : * During normal operation, the only timeline we care about is ThisTimeLineID.
306 : * During recovery, however, things are more complicated. To simplify life
307 : * for rmgr code, we keep ThisTimeLineID set to the "current" timeline as we
308 : * scan through the WAL history (that is, it is the line that was active when
309 : * the currently-scanned WAL record was generated). We also need these
310 : * timeline values:
311 : *
312 : * recoveryTargetTimeLineGoal: what the user requested, if any
313 : *
314 : * recoveryTargetTLIRequested: numeric value of requested timeline, if constant
315 : *
316 : * recoveryTargetTLI: the currently understood target timeline; changes
317 : *
318 : * expectedTLEs: a list of TimeLineHistoryEntries for recoveryTargetTLI and the timelines of
319 : * its known parents, newest first (so recoveryTargetTLI is always the
320 : * first list member). Only these TLIs are expected to be seen in the WAL
321 : * segments we read, and indeed only these TLIs will be considered as
322 : * candidate WAL files to open at all.
323 : *
324 : * curFileTLI: the TLI appearing in the name of the current input WAL file.
325 : * (This is not necessarily the same as ThisTimeLineID, because we could
326 : * be scanning data that was copied from an ancestor timeline when the current
327 : * file was created.) During a sequential scan we do not allow this value
328 : * to decrease.
329 : */
330 : RecoveryTargetTimeLineGoal recoveryTargetTimeLineGoal = RECOVERY_TARGET_TIMELINE_LATEST;
331 : TimeLineID recoveryTargetTLIRequested = 0;
332 : TimeLineID recoveryTargetTLI = 0;
333 : static List *expectedTLEs;
334 : static TimeLineID curFileTLI;
335 :
336 : /*
337 : * ProcLastRecPtr points to the start of the last XLOG record inserted by the
338 : * current backend. It is updated for all inserts. XactLastRecEnd points to
339 : * end+1 of the last record, and is reset when we end a top-level transaction,
340 : * or start a new one; so it can be used to tell if the current transaction has
341 : * created any XLOG records.
342 : *
343 : * While in parallel mode, this may not be fully up to date. When committing,
344 : * a transaction can assume this covers all xlog records written either by the
345 : * user backend or by any parallel worker which was present at any point during
346 : * the transaction. But when aborting, or when still in parallel mode, other
347 : * parallel backends may have written WAL records at later LSNs than the value
348 : * stored here. The parallel leader advances its own copy, when necessary,
349 : * in WaitForParallelWorkersToFinish.
350 : */
351 : XLogRecPtr ProcLastRecPtr = InvalidXLogRecPtr;
352 : XLogRecPtr XactLastRecEnd = InvalidXLogRecPtr;
353 : XLogRecPtr XactLastCommitEnd = InvalidXLogRecPtr;
354 :
355 : /*
356 : * RedoRecPtr is this backend's local copy of the REDO record pointer
357 : * (which is almost but not quite the same as a pointer to the most recent
358 : * CHECKPOINT record). We update this from the shared-memory copy,
359 : * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
360 : * hold an insertion lock). See XLogInsertRecord for details. We are also
361 : * allowed to update from XLogCtl->RedoRecPtr if we hold the info_lck;
362 : * see GetRedoRecPtr. A freshly spawned backend obtains the value during
363 : * InitXLOGAccess.
364 : */
365 : static XLogRecPtr RedoRecPtr;
366 :
367 : /*
368 : * doPageWrites is this backend's local copy of (forcePageWrites ||
369 : * fullPageWrites). It is used together with RedoRecPtr to decide whether
370 : * a full-page image of a page need to be taken.
371 : */
372 : static bool doPageWrites;
373 :
374 : /* Has the recovery code requested a walreceiver wakeup? */
375 : static bool doRequestWalReceiverReply;
376 :
377 : /*
378 : * RedoStartLSN points to the checkpoint's REDO location which is specified
379 : * in a backup label file, backup history file or control file. In standby
380 : * mode, XLOG streaming usually starts from the position where an invalid
381 : * record was found. But if we fail to read even the initial checkpoint
382 : * record, we use the REDO location instead of the checkpoint location as
383 : * the start position of XLOG streaming. Otherwise we would have to jump
384 : * backwards to the REDO location after reading the checkpoint record,
385 : * because the REDO record can precede the checkpoint record.
386 : */
387 : static XLogRecPtr RedoStartLSN = InvalidXLogRecPtr;
388 :
389 : /*----------
390 : * Shared-memory data structures for XLOG control
391 : *
392 : * LogwrtRqst indicates a byte position that we need to write and/or fsync
393 : * the log up to (all records before that point must be written or fsynced).
394 : * LogwrtResult indicates the byte positions we have already written/fsynced.
395 : * These structs are identical but are declared separately to indicate their
396 : * slightly different functions.
397 : *
398 : * To read XLogCtl->LogwrtResult, you must hold either info_lck or
399 : * WALWriteLock. To update it, you need to hold both locks. The point of
400 : * this arrangement is that the value can be examined by code that already
401 : * holds WALWriteLock without needing to grab info_lck as well. In addition
402 : * to the shared variable, each backend has a private copy of LogwrtResult,
403 : * which is updated when convenient.
404 : *
405 : * The request bookkeeping is simpler: there is a shared XLogCtl->LogwrtRqst
406 : * (protected by info_lck), but we don't need to cache any copies of it.
407 : *
408 : * info_lck is only held long enough to read/update the protected variables,
409 : * so it's a plain spinlock. The other locks are held longer (potentially
410 : * over I/O operations), so we use LWLocks for them. These locks are:
411 : *
412 : * WALBufMappingLock: must be held to replace a page in the WAL buffer cache.
413 : * It is only held while initializing and changing the mapping. If the
414 : * contents of the buffer being replaced haven't been written yet, the mapping
415 : * lock is released while the write is done, and reacquired afterwards.
416 : *
417 : * WALWriteLock: must be held to write WAL buffers to disk (XLogWrite or
418 : * XLogFlush).
419 : *
420 : * ControlFileLock: must be held to read/update control file or create
421 : * new log file.
422 : *
423 : * CheckpointLock: must be held to do a checkpoint or restartpoint (ensures
424 : * only one checkpointer at a time; currently, with all checkpoints done by
425 : * the checkpointer, this is just pro forma).
426 : *
427 : *----------
428 : */
429 :
430 : typedef struct XLogwrtRqst
431 : {
432 : XLogRecPtr Write; /* last byte + 1 to write out */
433 : XLogRecPtr Flush; /* last byte + 1 to flush */
434 : } XLogwrtRqst;
435 :
436 : typedef struct XLogwrtResult
437 : {
438 : XLogRecPtr Write; /* last byte + 1 written out */
439 : XLogRecPtr Flush; /* last byte + 1 flushed */
440 : } XLogwrtResult;
441 :
442 : /*
443 : * Inserting to WAL is protected by a small fixed number of WAL insertion
444 : * locks. To insert to the WAL, you must hold one of the locks - it doesn't
445 : * matter which one. To lock out other concurrent insertions, you must hold
446 : * of them. Each WAL insertion lock consists of a lightweight lock, plus an
447 : * indicator of how far the insertion has progressed (insertingAt).
448 : *
449 : * The insertingAt values are read when a process wants to flush WAL from
450 : * the in-memory buffers to disk, to check that all the insertions to the
451 : * region the process is about to write out have finished. You could simply
452 : * wait for all currently in-progress insertions to finish, but the
453 : * insertingAt indicator allows you to ignore insertions to later in the WAL,
454 : * so that you only wait for the insertions that are modifying the buffers
455 : * you're about to write out.
456 : *
457 : * This isn't just an optimization. If all the WAL buffers are dirty, an
458 : * inserter that's holding a WAL insert lock might need to evict an old WAL
459 : * buffer, which requires flushing the WAL. If it's possible for an inserter
460 : * to block on another inserter unnecessarily, deadlock can arise when two
461 : * inserters holding a WAL insert lock wait for each other to finish their
462 : * insertion.
463 : *
464 : * Small WAL records that don't cross a page boundary never update the value,
465 : * the WAL record is just copied to the page and the lock is released. But
466 : * to avoid the deadlock-scenario explained above, the indicator is always
467 : * updated before sleeping while holding an insertion lock.
468 : *
469 : * lastImportantAt contains the LSN of the last important WAL record inserted
470 : * using a given lock. This value is used to detect if there has been
471 : * important WAL activity since the last time some action, like a checkpoint,
472 : * was performed - allowing to not repeat the action if not. The LSN is
473 : * updated for all insertions, unless the XLOG_MARK_UNIMPORTANT flag was
474 : * set. lastImportantAt is never cleared, only overwritten by the LSN of newer
475 : * records. Tracking the WAL activity directly in WALInsertLock has the
476 : * advantage of not needing any additional locks to update the value.
477 : */
478 : typedef struct
479 : {
480 : LWLock lock;
481 : XLogRecPtr insertingAt;
482 : XLogRecPtr lastImportantAt;
483 : } WALInsertLock;
484 :
485 : /*
486 : * All the WAL insertion locks are allocated as an array in shared memory. We
487 : * force the array stride to be a power of 2, which saves a few cycles in
488 : * indexing, but more importantly also ensures that individual slots don't
489 : * cross cache line boundaries. (Of course, we have to also ensure that the
490 : * array start address is suitably aligned.)
491 : */
492 : typedef union WALInsertLockPadded
493 : {
494 : WALInsertLock l;
495 : char pad[PG_CACHE_LINE_SIZE];
496 : } WALInsertLockPadded;
497 :
498 : /*
499 : * State of an exclusive backup, necessary to control concurrent activities
500 : * across sessions when working on exclusive backups.
501 : *
502 : * EXCLUSIVE_BACKUP_NONE means that there is no exclusive backup actually
503 : * running, to be more precise pg_start_backup() is not being executed for
504 : * an exclusive backup and there is no exclusive backup in progress.
505 : * EXCLUSIVE_BACKUP_STARTING means that pg_start_backup() is starting an
506 : * exclusive backup.
507 : * EXCLUSIVE_BACKUP_IN_PROGRESS means that pg_start_backup() has finished
508 : * running and an exclusive backup is in progress. pg_stop_backup() is
509 : * needed to finish it.
510 : * EXCLUSIVE_BACKUP_STOPPING means that pg_stop_backup() is stopping an
511 : * exclusive backup.
512 : */
513 : typedef enum ExclusiveBackupState
514 : {
515 : EXCLUSIVE_BACKUP_NONE = 0,
516 : EXCLUSIVE_BACKUP_STARTING,
517 : EXCLUSIVE_BACKUP_IN_PROGRESS,
518 : EXCLUSIVE_BACKUP_STOPPING
519 : } ExclusiveBackupState;
520 :
521 : /*
522 : * Session status of running backup, used for sanity checks in SQL-callable
523 : * functions to start and stop backups.
524 : */
525 : static SessionBackupState sessionBackupState = SESSION_BACKUP_NONE;
526 :
527 : /*
528 : * Shared state data for WAL insertion.
529 : */
530 : typedef struct XLogCtlInsert
531 : {
532 : slock_t insertpos_lck; /* protects CurrBytePos and PrevBytePos */
533 :
534 : /*
535 : * CurrBytePos is the end of reserved WAL. The next record will be
536 : * inserted at that position. PrevBytePos is the start position of the
537 : * previously inserted (or rather, reserved) record - it is copied to the
538 : * prev-link of the next record. These are stored as "usable byte
539 : * positions" rather than XLogRecPtrs (see XLogBytePosToRecPtr()).
540 : */
541 : uint64 CurrBytePos;
542 : uint64 PrevBytePos;
543 :
544 : /*
545 : * Make sure the above heavily-contended spinlock and byte positions are
546 : * on their own cache line. In particular, the RedoRecPtr and full page
547 : * write variables below should be on a different cache line. They are
548 : * read on every WAL insertion, but updated rarely, and we don't want
549 : * those reads to steal the cache line containing Curr/PrevBytePos.
550 : */
551 : char pad[PG_CACHE_LINE_SIZE];
552 :
553 : /*
554 : * fullPageWrites is the master copy used by all backends to determine
555 : * whether to write full-page to WAL, instead of using process-local one.
556 : * This is required because, when full_page_writes is changed by SIGHUP,
557 : * we must WAL-log it before it actually affects WAL-logging by backends.
558 : * Checkpointer sets at startup or after SIGHUP.
559 : *
560 : * To read these fields, you must hold an insertion lock. To modify them,
561 : * you must hold ALL the locks.
562 : */
563 : XLogRecPtr RedoRecPtr; /* current redo point for insertions */
564 : bool forcePageWrites; /* forcing full-page writes for PITR? */
565 : bool fullPageWrites;
566 :
567 : /*
568 : * exclusiveBackupState indicates the state of an exclusive backup (see
569 : * comments of ExclusiveBackupState for more details). nonExclusiveBackups
570 : * is a counter indicating the number of streaming base backups currently
571 : * in progress. forcePageWrites is set to true when either of these is
572 : * non-zero. lastBackupStart is the latest checkpoint redo location used
573 : * as a starting point for an online backup.
574 : */
575 : ExclusiveBackupState exclusiveBackupState;
576 : int nonExclusiveBackups;
577 : XLogRecPtr lastBackupStart;
578 :
579 : /*
580 : * WAL insertion locks.
581 : */
582 : WALInsertLockPadded *WALInsertLocks;
583 : } XLogCtlInsert;
584 :
585 : /*
586 : * Total shared-memory state for XLOG.
587 : */
588 : typedef struct XLogCtlData
589 : {
590 : XLogCtlInsert Insert;
591 :
592 : /* Protected by info_lck: */
593 : XLogwrtRqst LogwrtRqst;
594 : XLogRecPtr RedoRecPtr; /* a recent copy of Insert->RedoRecPtr */
595 : FullTransactionId ckptFullXid; /* nextFullXid of latest checkpoint */
596 : XLogRecPtr asyncXactLSN; /* LSN of newest async commit/abort */
597 : XLogRecPtr replicationSlotMinLSN; /* oldest LSN needed by any slot */
598 :
599 : XLogSegNo lastRemovedSegNo; /* latest removed/recycled XLOG segment */
600 :
601 : /* Fake LSN counter, for unlogged relations. Protected by ulsn_lck. */
602 : XLogRecPtr unloggedLSN;
603 : slock_t ulsn_lck;
604 :
605 : /* Time and LSN of last xlog segment switch. Protected by WALWriteLock. */
606 : pg_time_t lastSegSwitchTime;
607 : XLogRecPtr lastSegSwitchLSN;
608 :
609 : /*
610 : * Protected by info_lck and WALWriteLock (you must hold either lock to
611 : * read it, but both to update)
612 : */
613 : XLogwrtResult LogwrtResult;
614 :
615 : /*
616 : * Latest initialized page in the cache (last byte position + 1).
617 : *
618 : * To change the identity of a buffer (and InitializedUpTo), you need to
619 : * hold WALBufMappingLock. To change the identity of a buffer that's
620 : * still dirty, the old page needs to be written out first, and for that
621 : * you need WALWriteLock, and you need to ensure that there are no
622 : * in-progress insertions to the page by calling
623 : * WaitXLogInsertionsToFinish().
624 : */
625 : XLogRecPtr InitializedUpTo;
626 :
627 : /*
628 : * These values do not change after startup, although the pointed-to pages
629 : * and xlblocks values certainly do. xlblocks values are protected by
630 : * WALBufMappingLock.
631 : */
632 : char *pages; /* buffers for unwritten XLOG pages */
633 : XLogRecPtr *xlblocks; /* 1st byte ptr-s + XLOG_BLCKSZ */
634 : int XLogCacheBlck; /* highest allocated xlog buffer index */
635 :
636 : /*
637 : * Shared copy of ThisTimeLineID. Does not change after end-of-recovery.
638 : * If we created a new timeline when the system was started up,
639 : * PrevTimeLineID is the old timeline's ID that we forked off from.
640 : * Otherwise it's equal to ThisTimeLineID.
641 : */
642 : TimeLineID ThisTimeLineID;
643 : TimeLineID PrevTimeLineID;
644 :
645 : /*
646 : * SharedRecoveryInProgress indicates if we're still in crash or archive
647 : * recovery. Protected by info_lck.
648 : */
649 : bool SharedRecoveryInProgress;
650 :
651 : /*
652 : * SharedHotStandbyActive indicates if we're still in crash or archive
653 : * recovery. Protected by info_lck.
654 : */
655 : bool SharedHotStandbyActive;
656 :
657 : /*
658 : * WalWriterSleeping indicates whether the WAL writer is currently in
659 : * low-power mode (and hence should be nudged if an async commit occurs).
660 : * Protected by info_lck.
661 : */
662 : bool WalWriterSleeping;
663 :
664 : /*
665 : * recoveryWakeupLatch is used to wake up the startup process to continue
666 : * WAL replay, if it is waiting for WAL to arrive or failover trigger file
667 : * to appear.
668 : */
669 : Latch recoveryWakeupLatch;
670 :
671 : /*
672 : * During recovery, we keep a copy of the latest checkpoint record here.
673 : * lastCheckPointRecPtr points to start of checkpoint record and
674 : * lastCheckPointEndPtr points to end+1 of checkpoint record. Used by the
675 : * checkpointer when it wants to create a restartpoint.
676 : *
677 : * Protected by info_lck.
678 : */
679 : XLogRecPtr lastCheckPointRecPtr;
680 : XLogRecPtr lastCheckPointEndPtr;
681 : CheckPoint lastCheckPoint;
682 :
683 : /*
684 : * lastReplayedEndRecPtr points to end+1 of the last record successfully
685 : * replayed. When we're currently replaying a record, ie. in a redo
686 : * function, replayEndRecPtr points to the end+1 of the record being
687 : * replayed, otherwise it's equal to lastReplayedEndRecPtr.
688 : */
689 : XLogRecPtr lastReplayedEndRecPtr;
690 : TimeLineID lastReplayedTLI;
691 : XLogRecPtr replayEndRecPtr;
692 : TimeLineID replayEndTLI;
693 : /* timestamp of last COMMIT/ABORT record replayed (or being replayed) */
694 : TimestampTz recoveryLastXTime;
695 :
696 : /*
697 : * timestamp of when we started replaying the current chunk of WAL data,
698 : * only relevant for replication or archive recovery
699 : */
700 : TimestampTz currentChunkStartTime;
701 : /* Are we requested to pause recovery? */
702 : bool recoveryPause;
703 :
704 : /*
705 : * lastFpwDisableRecPtr points to the start of the last replayed
706 : * XLOG_FPW_CHANGE record that instructs full_page_writes is disabled.
707 : */
708 : XLogRecPtr lastFpwDisableRecPtr;
709 :
710 : slock_t info_lck; /* locks shared variables shown above */
711 : } XLogCtlData;
712 :
713 : static XLogCtlData *XLogCtl = NULL;
714 :
715 : /* a private copy of XLogCtl->Insert.WALInsertLocks, for convenience */
716 : static WALInsertLockPadded *WALInsertLocks = NULL;
717 :
718 : /*
719 : * We maintain an image of pg_control in shared memory.
720 : */
721 : static ControlFileData *ControlFile = NULL;
722 :
723 : /*
724 : * Calculate the amount of space left on the page after 'endptr'. Beware
725 : * multiple evaluation!
726 : */
727 : #define INSERT_FREESPACE(endptr) \
728 : (((endptr) % XLOG_BLCKSZ == 0) ? 0 : (XLOG_BLCKSZ - (endptr) % XLOG_BLCKSZ))
729 :
730 : /* Macro to advance to next buffer index. */
731 : #define NextBufIdx(idx) \
732 : (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))
733 :
734 : /*
735 : * XLogRecPtrToBufIdx returns the index of the WAL buffer that holds, or
736 : * would hold if it was in cache, the page containing 'recptr'.
737 : */
738 : #define XLogRecPtrToBufIdx(recptr) \
739 : (((recptr) / XLOG_BLCKSZ) % (XLogCtl->XLogCacheBlck + 1))
740 :
741 : /*
742 : * These are the number of bytes in a WAL page usable for WAL data.
743 : */
744 : #define UsableBytesInPage (XLOG_BLCKSZ - SizeOfXLogShortPHD)
745 :
746 : /* Convert min_wal_size_mb and max_wal_size_mb to equivalent segment count */
747 : #define ConvertToXSegs(x, segsize) \
748 : (x / ((segsize) / (1024 * 1024)))
749 :
750 : /* The number of bytes in a WAL segment usable for WAL data. */
751 : static int UsableBytesInSegment;
752 :
753 : /*
754 : * Private, possibly out-of-date copy of shared LogwrtResult.
755 : * See discussion above.
756 : */
757 : static XLogwrtResult LogwrtResult = {0, 0};
758 :
759 : /*
760 : * Codes indicating where we got a WAL file from during recovery, or where
761 : * to attempt to get one.
762 : */
763 : typedef enum
764 : {
765 : XLOG_FROM_ANY = 0, /* request to read WAL from any source */
766 : XLOG_FROM_ARCHIVE, /* restored using restore_command */
767 : XLOG_FROM_PG_WAL, /* existing file in pg_wal */
768 : XLOG_FROM_STREAM /* streamed from master */
769 : } XLogSource;
770 :
771 : /* human-readable names for XLogSources, for debugging output */
772 : static const char *const xlogSourceNames[] = {"any", "archive", "pg_wal", "stream"};
773 :
774 : /*
775 : * openLogFile is -1 or a kernel FD for an open log file segment.
776 : * openLogSegNo identifies the segment. These variables are only used to
777 : * write the XLOG, and so will normally refer to the active segment.
778 : */
779 : static int openLogFile = -1;
780 : static XLogSegNo openLogSegNo = 0;
781 :
782 : /*
783 : * These variables are used similarly to the ones above, but for reading
784 : * the XLOG. Note, however, that readOff generally represents the offset
785 : * of the page just read, not the seek position of the FD itself, which
786 : * will be just past that page. readLen indicates how much of the current
787 : * page has been read into readBuf, and readSource indicates where we got
788 : * the currently open file from.
789 : */
790 : static int readFile = -1;
791 : static XLogSegNo readSegNo = 0;
792 : static uint32 readOff = 0;
793 : static uint32 readLen = 0;
794 : static XLogSource readSource = 0; /* XLOG_FROM_* code */
795 :
796 : /*
797 : * Keeps track of which source we're currently reading from. This is
798 : * different from readSource in that this is always set, even when we don't
799 : * currently have a WAL file open. If lastSourceFailed is set, our last
800 : * attempt to read from currentSource failed, and we should try another source
801 : * next.
802 : */
803 : static XLogSource currentSource = 0; /* XLOG_FROM_* code */
804 : static bool lastSourceFailed = false;
805 :
806 : typedef struct XLogPageReadPrivate
807 : {
808 : int emode;
809 : bool fetching_ckpt; /* are we fetching a checkpoint record? */
810 : bool randAccess;
811 : } XLogPageReadPrivate;
812 :
813 : /*
814 : * These variables track when we last obtained some WAL data to process,
815 : * and where we got it from. (XLogReceiptSource is initially the same as
816 : * readSource, but readSource gets reset to zero when we don't have data
817 : * to process right now. It is also different from currentSource, which
818 : * also changes when we try to read from a source and fail, while
819 : * XLogReceiptSource tracks where we last successfully read some WAL.)
820 : */
821 : static TimestampTz XLogReceiptTime = 0;
822 : static XLogSource XLogReceiptSource = 0; /* XLOG_FROM_* code */
823 :
824 : /* State information for XLOG reading */
825 : static XLogRecPtr ReadRecPtr; /* start of last record read */
826 : static XLogRecPtr EndRecPtr; /* end+1 of last record read */
827 :
828 : /*
829 : * Local copies of equivalent fields in the control file. When running
830 : * crash recovery, minRecoveryPoint is set to InvalidXLogRecPtr as we
831 : * expect to replay all the WAL available, and updateMinRecoveryPoint is
832 : * switched to false to prevent any updates while replaying records.
833 : * Those values are kept consistent as long as crash recovery runs.
834 : */
835 : static XLogRecPtr minRecoveryPoint;
836 : static TimeLineID minRecoveryPointTLI;
837 : static bool updateMinRecoveryPoint = true;
838 :
839 : /*
840 : * Have we reached a consistent database state? In crash recovery, we have
841 : * to replay all the WAL, so reachedConsistency is never set. During archive
842 : * recovery, the database is consistent once minRecoveryPoint is reached.
843 : */
844 : bool reachedConsistency = false;
845 :
846 : static bool InRedo = false;
847 :
848 : /* Have we launched bgwriter during recovery? */
849 : static bool bgwriterLaunched = false;
850 :
851 : /* For WALInsertLockAcquire/Release functions */
852 : static int MyLockNo = 0;
853 : static bool holdingAllLocks = false;
854 :
855 : #ifdef WAL_DEBUG
856 : static MemoryContext walDebugCxt = NULL;
857 : #endif
858 :
859 : static void readRecoverySignalFile(void);
860 : static void validateRecoveryParameters(void);
861 : static void exitArchiveRecovery(TimeLineID endTLI, XLogRecPtr endOfLog);
862 : static bool recoveryStopsBefore(XLogReaderState *record);
863 : static bool recoveryStopsAfter(XLogReaderState *record);
864 : static void recoveryPausesHere(void);
865 : static bool recoveryApplyDelay(XLogReaderState *record);
866 : static void SetLatestXTime(TimestampTz xtime);
867 : static void SetCurrentChunkStartTime(TimestampTz xtime);
868 : static void CheckRequiredParameterValues(void);
869 : static void XLogReportParameters(void);
870 : static void checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI,
871 : TimeLineID prevTLI);
872 : static void LocalSetXLogInsertAllowed(void);
873 : static void CreateEndOfRecoveryRecord(void);
874 : static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
875 : static void KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo);
876 : static XLogRecPtr XLogGetReplicationSlotMinimumLSN(void);
877 :
878 : static void AdvanceXLInsertBuffer(XLogRecPtr upto, bool opportunistic);
879 : static bool XLogCheckpointNeeded(XLogSegNo new_segno);
880 : static void XLogWrite(XLogwrtRqst WriteRqst, bool flexible);
881 : static bool InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
882 : bool find_free, XLogSegNo max_segno,
883 : bool use_lock);
884 : static int XLogFileRead(XLogSegNo segno, int emode, TimeLineID tli,
885 : int source, bool notfoundOk);
886 : static int XLogFileReadAnyTLI(XLogSegNo segno, int emode, int source);
887 : static int XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr,
888 : int reqLen, XLogRecPtr targetRecPtr, char *readBuf);
889 : static bool WaitForWALToBecomeAvailable(XLogRecPtr RecPtr, bool randAccess,
890 : bool fetching_ckpt, XLogRecPtr tliRecPtr);
891 : static int emode_for_corrupt_record(int emode, XLogRecPtr RecPtr);
892 : static void XLogFileClose(void);
893 : static void PreallocXlogFiles(XLogRecPtr endptr);
894 : static void RemoveTempXlogFiles(void);
895 : static void RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr RedoRecPtr, XLogRecPtr endptr);
896 : static void RemoveXlogFile(const char *segname, XLogRecPtr RedoRecPtr, XLogRecPtr endptr);
897 : static void UpdateLastRemovedPtr(char *filename);
898 : static void ValidateXLOGDirectoryStructure(void);
899 : static void CleanupBackupHistory(void);
900 : static void UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force);
901 : static XLogRecord *ReadRecord(XLogReaderState *xlogreader, XLogRecPtr RecPtr,
902 : int emode, bool fetching_ckpt);
903 : static void CheckRecoveryConsistency(void);
904 : static XLogRecord *ReadCheckpointRecord(XLogReaderState *xlogreader,
905 : XLogRecPtr RecPtr, int whichChkpt, bool report);
906 : static bool rescanLatestTimeLine(void);
907 : static void WriteControlFile(void);
908 : static void ReadControlFile(void);
909 : static char *str_time(pg_time_t tnow);
910 : static bool CheckForStandbyTrigger(void);
911 :
912 : #ifdef WAL_DEBUG
913 : static void xlog_outrec(StringInfo buf, XLogReaderState *record);
914 : #endif
915 : static void xlog_outdesc(StringInfo buf, XLogReaderState *record);
916 : static void pg_start_backup_callback(int code, Datum arg);
917 : static void pg_stop_backup_callback(int code, Datum arg);
918 : static bool read_backup_label(XLogRecPtr *checkPointLoc,
919 : bool *backupEndRequired, bool *backupFromStandby);
920 : static bool read_tablespace_map(List **tablespaces);
921 :
922 : static void rm_redo_error_callback(void *arg);
923 : static int get_sync_bit(int method);
924 :
925 : static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch,
926 : XLogRecData *rdata,
927 : XLogRecPtr StartPos, XLogRecPtr EndPos);
928 : static void ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos,
929 : XLogRecPtr *EndPos, XLogRecPtr *PrevPtr);
930 : static bool ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos,
931 : XLogRecPtr *PrevPtr);
932 : static XLogRecPtr WaitXLogInsertionsToFinish(XLogRecPtr upto);
933 : static char *GetXLogBuffer(XLogRecPtr ptr);
934 : static XLogRecPtr XLogBytePosToRecPtr(uint64 bytepos);
935 : static XLogRecPtr XLogBytePosToEndRecPtr(uint64 bytepos);
936 : static uint64 XLogRecPtrToBytePos(XLogRecPtr ptr);
937 : static void checkXLogConsistency(XLogReaderState *record);
938 :
939 : static void WALInsertLockAcquire(void);
940 : static void WALInsertLockAcquireExclusive(void);
941 : static void WALInsertLockRelease(void);
942 : static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
943 :
944 : /*
945 : * Insert an XLOG record represented by an already-constructed chain of data
946 : * chunks. This is a low-level routine; to construct the WAL record header
947 : * and data, use the higher-level routines in xloginsert.c.
948 : *
949 : * If 'fpw_lsn' is valid, it is the oldest LSN among the pages that this
950 : * WAL record applies to, that were not included in the record as full page
951 : * images. If fpw_lsn <= RedoRecPtr, the function does not perform the
952 : * insertion and returns InvalidXLogRecPtr. The caller can then recalculate
953 : * which pages need a full-page image, and retry. If fpw_lsn is invalid, the
954 : * record is always inserted.
955 : *
956 : * 'flags' gives more in-depth control on the record being inserted. See
957 : * XLogSetRecordFlags() for details.
958 : *
959 : * The first XLogRecData in the chain must be for the record header, and its
960 : * data must be MAXALIGNed. XLogInsertRecord fills in the xl_prev and
961 : * xl_crc fields in the header, the rest of the header must already be filled
962 : * by the caller.
963 : *
964 : * Returns XLOG pointer to end of record (beginning of next record).
965 : * This can be used as LSN for data pages affected by the logged action.
966 : * (LSN is the XLOG point up to which the XLOG must be flushed to disk
967 : * before the data page can be written out. This implements the basic
968 : * WAL rule "write the log before the data".)
969 : */
970 : XLogRecPtr
971 29477740 : XLogInsertRecord(XLogRecData *rdata,
972 : XLogRecPtr fpw_lsn,
973 : uint8 flags)
974 : {
975 29477740 : XLogCtlInsert *Insert = &XLogCtl->Insert;
976 : pg_crc32c rdata_crc;
977 : bool inserted;
978 29477740 : XLogRecord *rechdr = (XLogRecord *) rdata->data;
979 29477740 : uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
980 29477740 : bool isLogSwitch = (rechdr->xl_rmid == RM_XLOG_ID &&
981 : info == XLOG_SWITCH);
982 : XLogRecPtr StartPos;
983 : XLogRecPtr EndPos;
984 29477740 : bool prevDoPageWrites = doPageWrites;
985 :
986 : /* we assume that all of the record header is in the first chunk */
987 : Assert(rdata->len >= SizeOfXLogRecord);
988 :
989 : /* cross-check on whether we should be here or not */
990 29477740 : if (!XLogInsertAllowed())
991 0 : elog(ERROR, "cannot make new WAL entries during recovery");
992 :
993 : /*----------
994 : *
995 : * We have now done all the preparatory work we can without holding a
996 : * lock or modifying shared state. From here on, inserting the new WAL
997 : * record to the shared WAL buffer cache is a two-step process:
998 : *
999 : * 1. Reserve the right amount of space from the WAL. The current head of
1000 : * reserved space is kept in Insert->CurrBytePos, and is protected by
1001 : * insertpos_lck.
1002 : *
1003 : * 2. Copy the record to the reserved WAL space. This involves finding the
1004 : * correct WAL buffer containing the reserved space, and copying the
1005 : * record in place. This can be done concurrently in multiple processes.
1006 : *
1007 : * To keep track of which insertions are still in-progress, each concurrent
1008 : * inserter acquires an insertion lock. In addition to just indicating that
1009 : * an insertion is in progress, the lock tells others how far the inserter
1010 : * has progressed. There is a small fixed number of insertion locks,
1011 : * determined by NUM_XLOGINSERT_LOCKS. When an inserter crosses a page
1012 : * boundary, it updates the value stored in the lock to the how far it has
1013 : * inserted, to allow the previous buffer to be flushed.
1014 : *
1015 : * Holding onto an insertion lock also protects RedoRecPtr and
1016 : * fullPageWrites from changing until the insertion is finished.
1017 : *
1018 : * Step 2 can usually be done completely in parallel. If the required WAL
1019 : * page is not initialized yet, you have to grab WALBufMappingLock to
1020 : * initialize it, but the WAL writer tries to do that ahead of insertions
1021 : * to avoid that from happening in the critical path.
1022 : *
1023 : *----------
1024 : */
1025 29477740 : START_CRIT_SECTION();
1026 29477740 : if (isLogSwitch)
1027 192 : WALInsertLockAcquireExclusive();
1028 : else
1029 29477548 : WALInsertLockAcquire();
1030 :
1031 : /*
1032 : * Check to see if my copy of RedoRecPtr is out of date. If so, may have
1033 : * to go back and have the caller recompute everything. This can only
1034 : * happen just after a checkpoint, so it's better to be slow in this case
1035 : * and fast otherwise.
1036 : *
1037 : * Also check to see if fullPageWrites or forcePageWrites was just turned
1038 : * on; if we weren't already doing full-page writes then go back and
1039 : * recompute.
1040 : *
1041 : * If we aren't doing full-page writes then RedoRecPtr doesn't actually
1042 : * affect the contents of the XLOG record, so we'll update our local copy
1043 : * but not force a recomputation. (If doPageWrites was just turned off,
1044 : * we could recompute the record without full pages, but we choose not to
1045 : * bother.)
1046 : */
1047 29477740 : if (RedoRecPtr != Insert->RedoRecPtr)
1048 : {
1049 : Assert(RedoRecPtr < Insert->RedoRecPtr);
1050 666 : RedoRecPtr = Insert->RedoRecPtr;
1051 : }
1052 29477740 : doPageWrites = (Insert->fullPageWrites || Insert->forcePageWrites);
1053 :
1054 58806252 : if (doPageWrites &&
1055 58657024 : (!prevDoPageWrites ||
1056 28366490 : (fpw_lsn != InvalidXLogRecPtr && fpw_lsn <= RedoRecPtr)))
1057 : {
1058 : /*
1059 : * Oops, some buffer now needs to be backed up that the caller didn't
1060 : * back up. Start over.
1061 : */
1062 26 : WALInsertLockRelease();
1063 26 : END_CRIT_SECTION();
1064 26 : return InvalidXLogRecPtr;
1065 : }
1066 :
1067 : /*
1068 : * Reserve space for the record in the WAL. This also sets the xl_prev
1069 : * pointer.
1070 : */
1071 29477714 : if (isLogSwitch)
1072 192 : inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
1073 : else
1074 : {
1075 29477522 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
1076 : &rechdr->xl_prev);
1077 29477522 : inserted = true;
1078 : }
1079 :
1080 29477714 : if (inserted)
1081 : {
1082 : /*
1083 : * Now that xl_prev has been filled in, calculate CRC of the record
1084 : * header.
1085 : */
1086 29477690 : rdata_crc = rechdr->xl_crc;
1087 29477690 : COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
1088 29477690 : FIN_CRC32C(rdata_crc);
1089 29477690 : rechdr->xl_crc = rdata_crc;
1090 :
1091 : /*
1092 : * All the record data, including the header, is now ready to be
1093 : * inserted. Copy the record in the space reserved.
1094 : */
1095 29477690 : CopyXLogRecordToWAL(rechdr->xl_tot_len, isLogSwitch, rdata,
1096 : StartPos, EndPos);
1097 :
1098 : /*
1099 : * Unless record is flagged as not important, update LSN of last
1100 : * important record in the current slot. When holding all locks, just
1101 : * update the first one.
1102 : */
1103 29477690 : if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
1104 : {
1105 29338020 : int lockno = holdingAllLocks ? 0 : MyLockNo;
1106 :
1107 29338020 : WALInsertLocks[lockno].l.lastImportantAt = StartPos;
1108 : }
1109 : }
1110 : else
1111 : {
1112 : /*
1113 : * This was an xlog-switch record, but the current insert location was
1114 : * already exactly at the beginning of a segment, so there was no need
1115 : * to do anything.
1116 : */
1117 : }
1118 :
1119 : /*
1120 : * Done! Let others know that we're finished.
1121 : */
1122 29477714 : WALInsertLockRelease();
1123 :
1124 29477714 : MarkCurrentTransactionIdLoggedIfAny();
1125 :
1126 29477714 : END_CRIT_SECTION();
1127 :
1128 : /*
1129 : * Update shared LogwrtRqst.Write, if we crossed page boundary.
1130 : */
1131 29477714 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
1132 : {
1133 457296 : SpinLockAcquire(&XLogCtl->info_lck);
1134 : /* advance global request to include new block(s) */
1135 457296 : if (XLogCtl->LogwrtRqst.Write < EndPos)
1136 457234 : XLogCtl->LogwrtRqst.Write = EndPos;
1137 : /* update local result copy while I have the chance */
1138 457296 : LogwrtResult = XLogCtl->LogwrtResult;
1139 457296 : SpinLockRelease(&XLogCtl->info_lck);
1140 : }
1141 :
1142 : /*
1143 : * If this was an XLOG_SWITCH record, flush the record and the empty
1144 : * padding space that fills the rest of the segment, and perform
1145 : * end-of-segment actions (eg, notifying archiver).
1146 : */
1147 29477714 : if (isLogSwitch)
1148 : {
1149 : TRACE_POSTGRESQL_WAL_SWITCH();
1150 192 : XLogFlush(EndPos);
1151 :
1152 : /*
1153 : * Even though we reserved the rest of the segment for us, which is
1154 : * reflected in EndPos, we return a pointer to just the end of the
1155 : * xlog-switch record.
1156 : */
1157 192 : if (inserted)
1158 : {
1159 168 : EndPos = StartPos + SizeOfXLogRecord;
1160 168 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
1161 : {
1162 0 : uint64 offset = XLogSegmentOffset(EndPos, wal_segment_size);
1163 :
1164 0 : if (offset == EndPos % XLOG_BLCKSZ)
1165 0 : EndPos += SizeOfXLogLongPHD;
1166 : else
1167 0 : EndPos += SizeOfXLogShortPHD;
1168 : }
1169 : }
1170 : }
1171 :
1172 : #ifdef WAL_DEBUG
1173 : if (XLOG_DEBUG)
1174 : {
1175 : static XLogReaderState *debug_reader = NULL;
1176 : StringInfoData buf;
1177 : StringInfoData recordBuf;
1178 : char *errormsg = NULL;
1179 : MemoryContext oldCxt;
1180 :
1181 : oldCxt = MemoryContextSwitchTo(walDebugCxt);
1182 :
1183 : initStringInfo(&buf);
1184 : appendStringInfo(&buf, "INSERT @ %X/%X: ",
1185 : (uint32) (EndPos >> 32), (uint32) EndPos);
1186 :
1187 : /*
1188 : * We have to piece together the WAL record data from the XLogRecData
1189 : * entries, so that we can pass it to the rm_desc function as one
1190 : * contiguous chunk.
1191 : */
1192 : initStringInfo(&recordBuf);
1193 : for (; rdata != NULL; rdata = rdata->next)
1194 : appendBinaryStringInfo(&recordBuf, rdata->data, rdata->len);
1195 :
1196 : if (!debug_reader)
1197 : debug_reader = XLogReaderAllocate(wal_segment_size, NULL,
1198 : NULL, NULL);
1199 :
1200 : if (!debug_reader)
1201 : {
1202 : appendStringInfoString(&buf, "error decoding record: out of memory");
1203 : }
1204 : else if (!DecodeXLogRecord(debug_reader, (XLogRecord *) recordBuf.data,
1205 : &errormsg))
1206 : {
1207 : appendStringInfo(&buf, "error decoding record: %s",
1208 : errormsg ? errormsg : "no error message");
1209 : }
1210 : else
1211 : {
1212 : appendStringInfoString(&buf, " - ");
1213 : xlog_outdesc(&buf, debug_reader);
1214 : }
1215 : elog(LOG, "%s", buf.data);
1216 :
1217 : pfree(buf.data);
1218 : pfree(recordBuf.data);
1219 : MemoryContextSwitchTo(oldCxt);
1220 : }
1221 : #endif
1222 :
1223 : /*
1224 : * Update our global variables
1225 : */
1226 29477714 : ProcLastRecPtr = StartPos;
1227 29477714 : XactLastRecEnd = EndPos;
1228 :
1229 29477714 : return EndPos;
1230 : }
1231 :
1232 : /*
1233 : * Reserves the right amount of space for a record of given size from the WAL.
1234 : * *StartPos is set to the beginning of the reserved section, *EndPos to
1235 : * its end+1. *PrevPtr is set to the beginning of the previous record; it is
1236 : * used to set the xl_prev of this record.
1237 : *
1238 : * This is the performance critical part of XLogInsert that must be serialized
1239 : * across backends. The rest can happen mostly in parallel. Try to keep this
1240 : * section as short as possible, insertpos_lck can be heavily contended on a
1241 : * busy system.
1242 : *
1243 : * NB: The space calculation here must match the code in CopyXLogRecordToWAL,
1244 : * where we actually copy the record to the reserved space.
1245 : */
1246 : static void
1247 29477522 : ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos,
1248 : XLogRecPtr *PrevPtr)
1249 : {
1250 29477522 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1251 : uint64 startbytepos;
1252 : uint64 endbytepos;
1253 : uint64 prevbytepos;
1254 :
1255 29477522 : size = MAXALIGN(size);
1256 :
1257 : /* All (non xlog-switch) records should contain data. */
1258 : Assert(size > SizeOfXLogRecord);
1259 :
1260 : /*
1261 : * The duration the spinlock needs to be held is minimized by minimizing
1262 : * the calculations that have to be done while holding the lock. The
1263 : * current tip of reserved WAL is kept in CurrBytePos, as a byte position
1264 : * that only counts "usable" bytes in WAL, that is, it excludes all WAL
1265 : * page headers. The mapping between "usable" byte positions and physical
1266 : * positions (XLogRecPtrs) can be done outside the locked region, and
1267 : * because the usable byte position doesn't include any headers, reserving
1268 : * X bytes from WAL is almost as simple as "CurrBytePos += X".
1269 : */
1270 29477522 : SpinLockAcquire(&Insert->insertpos_lck);
1271 :
1272 29477522 : startbytepos = Insert->CurrBytePos;
1273 29477522 : endbytepos = startbytepos + size;
1274 29477522 : prevbytepos = Insert->PrevBytePos;
1275 29477522 : Insert->CurrBytePos = endbytepos;
1276 29477522 : Insert->PrevBytePos = startbytepos;
1277 :
1278 29477522 : SpinLockRelease(&Insert->insertpos_lck);
1279 :
1280 29477522 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1281 29477522 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1282 29477522 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1283 :
1284 : /*
1285 : * Check that the conversions between "usable byte positions" and
1286 : * XLogRecPtrs work consistently in both directions.
1287 : */
1288 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1289 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1290 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1291 29477522 : }
1292 :
1293 : /*
1294 : * Like ReserveXLogInsertLocation(), but for an xlog-switch record.
1295 : *
1296 : * A log-switch record is handled slightly differently. The rest of the
1297 : * segment will be reserved for this insertion, as indicated by the returned
1298 : * *EndPos value. However, if we are already at the beginning of the current
1299 : * segment, *StartPos and *EndPos are set to the current location without
1300 : * reserving any space, and the function returns false.
1301 : */
1302 : static bool
1303 192 : ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
1304 : {
1305 192 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1306 : uint64 startbytepos;
1307 : uint64 endbytepos;
1308 : uint64 prevbytepos;
1309 192 : uint32 size = MAXALIGN(SizeOfXLogRecord);
1310 : XLogRecPtr ptr;
1311 : uint32 segleft;
1312 :
1313 : /*
1314 : * These calculations are a bit heavy-weight to be done while holding a
1315 : * spinlock, but since we're holding all the WAL insertion locks, there
1316 : * are no other inserters competing for it. GetXLogInsertRecPtr() does
1317 : * compete for it, but that's not called very frequently.
1318 : */
1319 192 : SpinLockAcquire(&Insert->insertpos_lck);
1320 :
1321 192 : startbytepos = Insert->CurrBytePos;
1322 :
1323 192 : ptr = XLogBytePosToEndRecPtr(startbytepos);
1324 192 : if (XLogSegmentOffset(ptr, wal_segment_size) == 0)
1325 : {
1326 24 : SpinLockRelease(&Insert->insertpos_lck);
1327 24 : *EndPos = *StartPos = ptr;
1328 24 : return false;
1329 : }
1330 :
1331 168 : endbytepos = startbytepos + size;
1332 168 : prevbytepos = Insert->PrevBytePos;
1333 :
1334 168 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1335 168 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1336 :
1337 168 : segleft = wal_segment_size - XLogSegmentOffset(*EndPos, wal_segment_size);
1338 168 : if (segleft != wal_segment_size)
1339 : {
1340 : /* consume the rest of the segment */
1341 168 : *EndPos += segleft;
1342 168 : endbytepos = XLogRecPtrToBytePos(*EndPos);
1343 : }
1344 168 : Insert->CurrBytePos = endbytepos;
1345 168 : Insert->PrevBytePos = startbytepos;
1346 :
1347 168 : SpinLockRelease(&Insert->insertpos_lck);
1348 :
1349 168 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1350 :
1351 : Assert(XLogSegmentOffset(*EndPos, wal_segment_size) == 0);
1352 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1353 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1354 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1355 :
1356 168 : return true;
1357 : }
1358 :
1359 : /*
1360 : * Checks whether the current buffer page and backup page stored in the
1361 : * WAL record are consistent or not. Before comparing the two pages, a
1362 : * masking can be applied to the pages to ignore certain areas like hint bits,
1363 : * unused space between pd_lower and pd_upper among other things. This
1364 : * function should be called once WAL replay has been completed for a
1365 : * given record.
1366 : */
1367 : static void
1368 0 : checkXLogConsistency(XLogReaderState *record)
1369 : {
1370 0 : RmgrId rmid = XLogRecGetRmid(record);
1371 : RelFileNode rnode;
1372 : ForkNumber forknum;
1373 : BlockNumber blkno;
1374 : int block_id;
1375 :
1376 : /* Records with no backup blocks have no need for consistency checks. */
1377 0 : if (!XLogRecHasAnyBlockRefs(record))
1378 0 : return;
1379 :
1380 : Assert((XLogRecGetInfo(record) & XLR_CHECK_CONSISTENCY) != 0);
1381 :
1382 0 : for (block_id = 0; block_id <= record->max_block_id; block_id++)
1383 : {
1384 : Buffer buf;
1385 : Page page;
1386 :
1387 0 : if (!XLogRecGetBlockTag(record, block_id, &rnode, &forknum, &blkno))
1388 : {
1389 : /*
1390 : * WAL record doesn't contain a block reference with the given id.
1391 : * Do nothing.
1392 : */
1393 0 : continue;
1394 : }
1395 :
1396 : Assert(XLogRecHasBlockImage(record, block_id));
1397 :
1398 0 : if (XLogRecBlockImageApply(record, block_id))
1399 : {
1400 : /*
1401 : * WAL record has already applied the page, so bypass the
1402 : * consistency check as that would result in comparing the full
1403 : * page stored in the record with itself.
1404 : */
1405 0 : continue;
1406 : }
1407 :
1408 : /*
1409 : * Read the contents from the current buffer and store it in a
1410 : * temporary page.
1411 : */
1412 0 : buf = XLogReadBufferExtended(rnode, forknum, blkno,
1413 : RBM_NORMAL_NO_LOG);
1414 0 : if (!BufferIsValid(buf))
1415 0 : continue;
1416 :
1417 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
1418 0 : page = BufferGetPage(buf);
1419 :
1420 : /*
1421 : * Take a copy of the local page where WAL has been applied to have a
1422 : * comparison base before masking it...
1423 : */
1424 0 : memcpy(replay_image_masked, page, BLCKSZ);
1425 :
1426 : /* No need for this page anymore now that a copy is in. */
1427 0 : UnlockReleaseBuffer(buf);
1428 :
1429 : /*
1430 : * If the block LSN is already ahead of this WAL record, we can't
1431 : * expect contents to match. This can happen if recovery is
1432 : * restarted.
1433 : */
1434 0 : if (PageGetLSN(replay_image_masked) > record->EndRecPtr)
1435 0 : continue;
1436 :
1437 : /*
1438 : * Read the contents from the backup copy, stored in WAL record and
1439 : * store it in a temporary page. There is no need to allocate a new
1440 : * page here, a local buffer is fine to hold its contents and a mask
1441 : * can be directly applied on it.
1442 : */
1443 0 : if (!RestoreBlockImage(record, block_id, master_image_masked))
1444 0 : elog(ERROR, "failed to restore block image");
1445 :
1446 : /*
1447 : * If masking function is defined, mask both the master and replay
1448 : * images
1449 : */
1450 0 : if (RmgrTable[rmid].rm_mask != NULL)
1451 : {
1452 0 : RmgrTable[rmid].rm_mask(replay_image_masked, blkno);
1453 0 : RmgrTable[rmid].rm_mask(master_image_masked, blkno);
1454 : }
1455 :
1456 : /* Time to compare the master and replay images. */
1457 0 : if (memcmp(replay_image_masked, master_image_masked, BLCKSZ) != 0)
1458 : {
1459 0 : elog(FATAL,
1460 : "inconsistent page found, rel %u/%u/%u, forknum %u, blkno %u",
1461 : rnode.spcNode, rnode.dbNode, rnode.relNode,
1462 : forknum, blkno);
1463 : }
1464 : }
1465 : }
1466 :
1467 : /*
1468 : * Subroutine of XLogInsertRecord. Copies a WAL record to an already-reserved
1469 : * area in the WAL.
1470 : */
1471 : static void
1472 29477690 : CopyXLogRecordToWAL(int write_len, bool isLogSwitch, XLogRecData *rdata,
1473 : XLogRecPtr StartPos, XLogRecPtr EndPos)
1474 : {
1475 : char *currpos;
1476 : int freespace;
1477 : int written;
1478 : XLogRecPtr CurrPos;
1479 : XLogPageHeader pagehdr;
1480 :
1481 : /*
1482 : * Get a pointer to the right place in the right WAL buffer to start
1483 : * inserting to.
1484 : */
1485 29477690 : CurrPos = StartPos;
1486 29477690 : currpos = GetXLogBuffer(CurrPos);
1487 29477690 : freespace = INSERT_FREESPACE(CurrPos);
1488 :
1489 : /*
1490 : * there should be enough space for at least the first field (xl_tot_len)
1491 : * on this page.
1492 : */
1493 : Assert(freespace >= sizeof(uint32));
1494 :
1495 : /* Copy record data */
1496 29477690 : written = 0;
1497 161638350 : while (rdata != NULL)
1498 : {
1499 102682970 : char *rdata_data = rdata->data;
1500 102682970 : int rdata_len = rdata->len;
1501 :
1502 205797994 : while (rdata_len > freespace)
1503 : {
1504 : /*
1505 : * Write what fits on this page, and continue on the next page.
1506 : */
1507 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || freespace == 0);
1508 432054 : memcpy(currpos, rdata_data, freespace);
1509 432054 : rdata_data += freespace;
1510 432054 : rdata_len -= freespace;
1511 432054 : written += freespace;
1512 432054 : CurrPos += freespace;
1513 :
1514 : /*
1515 : * Get pointer to beginning of next page, and set the xlp_rem_len
1516 : * in the page header. Set XLP_FIRST_IS_CONTRECORD.
1517 : *
1518 : * It's safe to set the contrecord flag and xlp_rem_len without a
1519 : * lock on the page. All the other flags were already set when the
1520 : * page was initialized, in AdvanceXLInsertBuffer, and we're the
1521 : * only backend that needs to set the contrecord flag.
1522 : */
1523 432054 : currpos = GetXLogBuffer(CurrPos);
1524 432054 : pagehdr = (XLogPageHeader) currpos;
1525 432054 : pagehdr->xlp_rem_len = write_len - written;
1526 432054 : pagehdr->xlp_info |= XLP_FIRST_IS_CONTRECORD;
1527 :
1528 : /* skip over the page header */
1529 432054 : if (XLogSegmentOffset(CurrPos, wal_segment_size) == 0)
1530 : {
1531 546 : CurrPos += SizeOfXLogLongPHD;
1532 546 : currpos += SizeOfXLogLongPHD;
1533 : }
1534 : else
1535 : {
1536 431508 : CurrPos += SizeOfXLogShortPHD;
1537 431508 : currpos += SizeOfXLogShortPHD;
1538 : }
1539 432054 : freespace = INSERT_FREESPACE(CurrPos);
1540 : }
1541 :
1542 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || rdata_len == 0);
1543 102682970 : memcpy(currpos, rdata_data, rdata_len);
1544 102682970 : currpos += rdata_len;
1545 102682970 : CurrPos += rdata_len;
1546 102682970 : freespace -= rdata_len;
1547 102682970 : written += rdata_len;
1548 :
1549 102682970 : rdata = rdata->next;
1550 : }
1551 : Assert(written == write_len);
1552 :
1553 : /*
1554 : * If this was an xlog-switch, it's not enough to write the switch record,
1555 : * we also have to consume all the remaining space in the WAL segment. We
1556 : * have already reserved that space, but we need to actually fill it.
1557 : */
1558 29477858 : if (isLogSwitch && XLogSegmentOffset(CurrPos, wal_segment_size) != 0)
1559 : {
1560 : /* An xlog-switch record doesn't contain any data besides the header */
1561 : Assert(write_len == SizeOfXLogRecord);
1562 :
1563 : /* Assert that we did reserve the right amount of space */
1564 : Assert(XLogSegmentOffset(EndPos, wal_segment_size) == 0);
1565 :
1566 : /* Use up all the remaining space on the current page */
1567 168 : CurrPos += freespace;
1568 :
1569 : /*
1570 : * Cause all remaining pages in the segment to be flushed, leaving the
1571 : * XLog position where it should be, at the start of the next segment.
1572 : * We do this one page at a time, to make sure we don't deadlock
1573 : * against ourselves if wal_buffers < wal_segment_size.
1574 : */
1575 304718 : while (CurrPos < EndPos)
1576 : {
1577 : /*
1578 : * The minimal action to flush the page would be to call
1579 : * WALInsertLockUpdateInsertingAt(CurrPos) followed by
1580 : * AdvanceXLInsertBuffer(...). The page would be left initialized
1581 : * mostly to zeros, except for the page header (always the short
1582 : * variant, as this is never a segment's first page).
1583 : *
1584 : * The large vistas of zeros are good for compressibility, but the
1585 : * headers interrupting them every XLOG_BLCKSZ (with values that
1586 : * differ from page to page) are not. The effect varies with
1587 : * compression tool, but bzip2 for instance compresses about an
1588 : * order of magnitude worse if those headers are left in place.
1589 : *
1590 : * Rather than complicating AdvanceXLInsertBuffer itself (which is
1591 : * called in heavily-loaded circumstances as well as this lightly-
1592 : * loaded one) with variant behavior, we just use GetXLogBuffer
1593 : * (which itself calls the two methods we need) to get the pointer
1594 : * and zero most of the page. Then we just zero the page header.
1595 : */
1596 304382 : currpos = GetXLogBuffer(CurrPos);
1597 304382 : MemSet(currpos, 0, SizeOfXLogShortPHD);
1598 :
1599 304382 : CurrPos += XLOG_BLCKSZ;
1600 : }
1601 : }
1602 : else
1603 : {
1604 : /* Align the end position, so that the next record starts aligned */
1605 29477522 : CurrPos = MAXALIGN64(CurrPos);
1606 : }
1607 :
1608 29477690 : if (CurrPos != EndPos)
1609 0 : elog(PANIC, "space reserved for WAL record does not match what was written");
1610 29477690 : }
1611 :
1612 : /*
1613 : * Acquire a WAL insertion lock, for inserting to WAL.
1614 : */
1615 : static void
1616 29477548 : WALInsertLockAcquire(void)
1617 : {
1618 : bool immed;
1619 :
1620 : /*
1621 : * It doesn't matter which of the WAL insertion locks we acquire, so try
1622 : * the one we used last time. If the system isn't particularly busy, it's
1623 : * a good bet that it's still available, and it's good to have some
1624 : * affinity to a particular lock so that you don't unnecessarily bounce
1625 : * cache lines between processes when there's no contention.
1626 : *
1627 : * If this is the first time through in this backend, pick a lock
1628 : * (semi-)randomly. This allows the locks to be used evenly if you have a
1629 : * lot of very short connections.
1630 : */
1631 : static int lockToTry = -1;
1632 :
1633 29477548 : if (lockToTry == -1)
1634 6036 : lockToTry = MyProc->pgprocno % NUM_XLOGINSERT_LOCKS;
1635 29477548 : MyLockNo = lockToTry;
1636 :
1637 : /*
1638 : * The insertingAt value is initially set to 0, as we don't know our
1639 : * insert location yet.
1640 : */
1641 29477548 : immed = LWLockAcquire(&WALInsertLocks[MyLockNo].l.lock, LW_EXCLUSIVE);
1642 29477548 : if (!immed)
1643 : {
1644 : /*
1645 : * If we couldn't get the lock immediately, try another lock next
1646 : * time. On a system with more insertion locks than concurrent
1647 : * inserters, this causes all the inserters to eventually migrate to a
1648 : * lock that no-one else is using. On a system with more inserters
1649 : * than locks, it still helps to distribute the inserters evenly
1650 : * across the locks.
1651 : */
1652 144 : lockToTry = (lockToTry + 1) % NUM_XLOGINSERT_LOCKS;
1653 : }
1654 29477548 : }
1655 :
1656 : /*
1657 : * Acquire all WAL insertion locks, to prevent other backends from inserting
1658 : * to WAL.
1659 : */
1660 : static void
1661 3418 : WALInsertLockAcquireExclusive(void)
1662 : {
1663 : int i;
1664 :
1665 : /*
1666 : * When holding all the locks, all but the last lock's insertingAt
1667 : * indicator is set to 0xFFFFFFFFFFFFFFFF, which is higher than any real
1668 : * XLogRecPtr value, to make sure that no-one blocks waiting on those.
1669 : */
1670 27344 : for (i = 0; i < NUM_XLOGINSERT_LOCKS - 1; i++)
1671 : {
1672 23926 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1673 23926 : LWLockUpdateVar(&WALInsertLocks[i].l.lock,
1674 23926 : &WALInsertLocks[i].l.insertingAt,
1675 : PG_UINT64_MAX);
1676 : }
1677 : /* Variable value reset to 0 at release */
1678 3418 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1679 :
1680 3418 : holdingAllLocks = true;
1681 3418 : }
1682 :
1683 : /*
1684 : * Release our insertion lock (or locks, if we're holding them all).
1685 : *
1686 : * NB: Reset all variables to 0, so they cause LWLockWaitForVar to block the
1687 : * next time the lock is acquired.
1688 : */
1689 : static void
1690 29480966 : WALInsertLockRelease(void)
1691 : {
1692 29480966 : if (holdingAllLocks)
1693 : {
1694 : int i;
1695 :
1696 30762 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1697 27344 : LWLockReleaseClearVar(&WALInsertLocks[i].l.lock,
1698 27344 : &WALInsertLocks[i].l.insertingAt,
1699 : 0);
1700 :
1701 3418 : holdingAllLocks = false;
1702 : }
1703 : else
1704 : {
1705 29477548 : LWLockReleaseClearVar(&WALInsertLocks[MyLockNo].l.lock,
1706 29477548 : &WALInsertLocks[MyLockNo].l.insertingAt,
1707 : 0);
1708 : }
1709 29480966 : }
1710 :
1711 : /*
1712 : * Update our insertingAt value, to let others know that we've finished
1713 : * inserting up to that point.
1714 : */
1715 : static void
1716 591516 : WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt)
1717 : {
1718 591516 : if (holdingAllLocks)
1719 : {
1720 : /*
1721 : * We use the last lock to mark our actual position, see comments in
1722 : * WALInsertLockAcquireExclusive.
1723 : */
1724 302330 : LWLockUpdateVar(&WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.lock,
1725 302330 : &WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.insertingAt,
1726 : insertingAt);
1727 : }
1728 : else
1729 289186 : LWLockUpdateVar(&WALInsertLocks[MyLockNo].l.lock,
1730 289186 : &WALInsertLocks[MyLockNo].l.insertingAt,
1731 : insertingAt);
1732 591516 : }
1733 :
1734 : /*
1735 : * Wait for any WAL insertions < upto to finish.
1736 : *
1737 : * Returns the location of the oldest insertion that is still in-progress.
1738 : * Any WAL prior to that point has been fully copied into WAL buffers, and
1739 : * can be flushed out to disk. Because this waits for any insertions older
1740 : * than 'upto' to finish, the return value is always >= 'upto'.
1741 : *
1742 : * Note: When you are about to write out WAL, you must call this function
1743 : * *before* acquiring WALWriteLock, to avoid deadlocks. This function might
1744 : * need to wait for an insertion to finish (or at least advance to next
1745 : * uninitialized page), and the inserter might need to evict an old WAL buffer
1746 : * to make room for a new one, which in turn requires WALWriteLock.
1747 : */
1748 : static XLogRecPtr
1749 534096 : WaitXLogInsertionsToFinish(XLogRecPtr upto)
1750 : {
1751 : uint64 bytepos;
1752 : XLogRecPtr reservedUpto;
1753 : XLogRecPtr finishedUpto;
1754 534096 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1755 : int i;
1756 :
1757 534096 : if (MyProc == NULL)
1758 0 : elog(PANIC, "cannot wait without a PGPROC structure");
1759 :
1760 : /* Read the current insert position */
1761 534096 : SpinLockAcquire(&Insert->insertpos_lck);
1762 534096 : bytepos = Insert->CurrBytePos;
1763 534096 : SpinLockRelease(&Insert->insertpos_lck);
1764 534096 : reservedUpto = XLogBytePosToEndRecPtr(bytepos);
1765 :
1766 : /*
1767 : * No-one should request to flush a piece of WAL that hasn't even been
1768 : * reserved yet. However, it can happen if there is a block with a bogus
1769 : * LSN on disk, for example. XLogFlush checks for that situation and
1770 : * complains, but only after the flush. Here we just assume that to mean
1771 : * that all WAL that has been reserved needs to be finished. In this
1772 : * corner-case, the return value can be smaller than 'upto' argument.
1773 : */
1774 534096 : if (upto > reservedUpto)
1775 : {
1776 0 : elog(LOG, "request to flush past end of generated WAL; request %X/%X, currpos %X/%X",
1777 : (uint32) (upto >> 32), (uint32) upto,
1778 : (uint32) (reservedUpto >> 32), (uint32) reservedUpto);
1779 0 : upto = reservedUpto;
1780 : }
1781 :
1782 : /*
1783 : * Loop through all the locks, sleeping on any in-progress insert older
1784 : * than 'upto'.
1785 : *
1786 : * finishedUpto is our return value, indicating the point upto which all
1787 : * the WAL insertions have been finished. Initialize it to the head of
1788 : * reserved WAL, and as we iterate through the insertion locks, back it
1789 : * out for any insertion that's still in progress.
1790 : */
1791 534096 : finishedUpto = reservedUpto;
1792 4806864 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1793 : {
1794 4272768 : XLogRecPtr insertingat = InvalidXLogRecPtr;
1795 :
1796 : do
1797 : {
1798 : /*
1799 : * See if this insertion is in progress. LWLockWaitForVar will
1800 : * wait for the lock to be released, or for the 'value' to be set
1801 : * by a LWLockUpdateVar call. When a lock is initially acquired,
1802 : * its value is 0 (InvalidXLogRecPtr), which means that we don't
1803 : * know where it's inserting yet. We will have to wait for it. If
1804 : * it's a small insertion, the record will most likely fit on the
1805 : * same page and the inserter will release the lock without ever
1806 : * calling LWLockUpdateVar. But if it has to sleep, it will
1807 : * advertise the insertion point with LWLockUpdateVar before
1808 : * sleeping.
1809 : */
1810 8545904 : if (LWLockWaitForVar(&WALInsertLocks[i].l.lock,
1811 4272952 : &WALInsertLocks[i].l.insertingAt,
1812 : insertingat, &insertingat))
1813 : {
1814 : /* the lock was free, so no insertion in progress */
1815 2098512 : insertingat = InvalidXLogRecPtr;
1816 2098512 : break;
1817 : }
1818 :
1819 : /*
1820 : * This insertion is still in progress. Have to wait, unless the
1821 : * inserter has proceeded past 'upto'.
1822 : */
1823 2174440 : } while (insertingat < upto);
1824 :
1825 4272768 : if (insertingat != InvalidXLogRecPtr && insertingat < finishedUpto)
1826 284538 : finishedUpto = insertingat;
1827 : }
1828 534096 : return finishedUpto;
1829 : }
1830 :
1831 : /*
1832 : * Get a pointer to the right location in the WAL buffer containing the
1833 : * given XLogRecPtr.
1834 : *
1835 : * If the page is not initialized yet, it is initialized. That might require
1836 : * evicting an old dirty buffer from the buffer cache, which means I/O.
1837 : *
1838 : * The caller must ensure that the page containing the requested location
1839 : * isn't evicted yet, and won't be evicted. The way to ensure that is to
1840 : * hold onto a WAL insertion lock with the insertingAt position set to
1841 : * something <= ptr. GetXLogBuffer() will update insertingAt if it needs
1842 : * to evict an old page from the buffer. (This means that once you call
1843 : * GetXLogBuffer() with a given 'ptr', you must not access anything before
1844 : * that point anymore, and must not call GetXLogBuffer() with an older 'ptr'
1845 : * later, because older buffers might be recycled already)
1846 : */
1847 : static char *
1848 30214126 : GetXLogBuffer(XLogRecPtr ptr)
1849 : {
1850 : int idx;
1851 : XLogRecPtr endptr;
1852 : static uint64 cachedPage = 0;
1853 : static char *cachedPos = NULL;
1854 : XLogRecPtr expectedEndPtr;
1855 :
1856 : /*
1857 : * Fast path for the common case that we need to access again the same
1858 : * page as last time.
1859 : */
1860 30214126 : if (ptr / XLOG_BLCKSZ == cachedPage)
1861 : {
1862 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1863 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1864 29359044 : return cachedPos + ptr % XLOG_BLCKSZ;
1865 : }
1866 :
1867 : /*
1868 : * The XLog buffer cache is organized so that a page is always loaded to a
1869 : * particular buffer. That way we can easily calculate the buffer a given
1870 : * page must be loaded into, from the XLogRecPtr alone.
1871 : */
1872 855082 : idx = XLogRecPtrToBufIdx(ptr);
1873 :
1874 : /*
1875 : * See what page is loaded in the buffer at the moment. It could be the
1876 : * page we're looking for, or something older. It can't be anything newer
1877 : * - that would imply the page we're looking for has already been written
1878 : * out to disk and evicted, and the caller is responsible for making sure
1879 : * that doesn't happen.
1880 : *
1881 : * However, we don't hold a lock while we read the value. If someone has
1882 : * just initialized the page, it's possible that we get a "torn read" of
1883 : * the XLogRecPtr if 64-bit fetches are not atomic on this platform. In
1884 : * that case we will see a bogus value. That's ok, we'll grab the mapping
1885 : * lock (in AdvanceXLInsertBuffer) and retry if we see anything else than
1886 : * the page we're looking for. But it means that when we do this unlocked
1887 : * read, we might see a value that appears to be ahead of the page we're
1888 : * looking for. Don't PANIC on that, until we've verified the value while
1889 : * holding the lock.
1890 : */
1891 855082 : expectedEndPtr = ptr;
1892 855082 : expectedEndPtr += XLOG_BLCKSZ - ptr % XLOG_BLCKSZ;
1893 :
1894 855082 : endptr = XLogCtl->xlblocks[idx];
1895 855082 : if (expectedEndPtr != endptr)
1896 : {
1897 : XLogRecPtr initializedUpto;
1898 :
1899 : /*
1900 : * Before calling AdvanceXLInsertBuffer(), which can block, let others
1901 : * know how far we're finished with inserting the record.
1902 : *
1903 : * NB: If 'ptr' points to just after the page header, advertise a
1904 : * position at the beginning of the page rather than 'ptr' itself. If
1905 : * there are no other insertions running, someone might try to flush
1906 : * up to our advertised location. If we advertised a position after
1907 : * the page header, someone might try to flush the page header, even
1908 : * though page might actually not be initialized yet. As the first
1909 : * inserter on the page, we are effectively responsible for making
1910 : * sure that it's initialized, before we let insertingAt to move past
1911 : * the page header.
1912 : */
1913 609436 : if (ptr % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
1914 17920 : XLogSegmentOffset(ptr, wal_segment_size) > XLOG_BLCKSZ)
1915 17920 : initializedUpto = ptr - SizeOfXLogShortPHD;
1916 573830 : else if (ptr % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
1917 234 : XLogSegmentOffset(ptr, wal_segment_size) < XLOG_BLCKSZ)
1918 140 : initializedUpto = ptr - SizeOfXLogLongPHD;
1919 : else
1920 573456 : initializedUpto = ptr;
1921 :
1922 591516 : WALInsertLockUpdateInsertingAt(initializedUpto);
1923 :
1924 591516 : AdvanceXLInsertBuffer(ptr, false);
1925 591516 : endptr = XLogCtl->xlblocks[idx];
1926 :
1927 591516 : if (expectedEndPtr != endptr)
1928 0 : elog(PANIC, "could not find WAL buffer for %X/%X",
1929 : (uint32) (ptr >> 32), (uint32) ptr);
1930 : }
1931 : else
1932 : {
1933 : /*
1934 : * Make sure the initialization of the page is visible to us, and
1935 : * won't arrive later to overwrite the WAL data we write on the page.
1936 : */
1937 263566 : pg_memory_barrier();
1938 : }
1939 :
1940 : /*
1941 : * Found the buffer holding this page. Return a pointer to the right
1942 : * offset within the page.
1943 : */
1944 855082 : cachedPage = ptr / XLOG_BLCKSZ;
1945 855082 : cachedPos = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1946 :
1947 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1948 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1949 :
1950 855082 : return cachedPos + ptr % XLOG_BLCKSZ;
1951 : }
1952 :
1953 : /*
1954 : * Converts a "usable byte position" to XLogRecPtr. A usable byte position
1955 : * is the position starting from the beginning of WAL, excluding all WAL
1956 : * page headers.
1957 : */
1958 : static XLogRecPtr
1959 58982110 : XLogBytePosToRecPtr(uint64 bytepos)
1960 : {
1961 : uint64 fullsegs;
1962 : uint64 fullpages;
1963 : uint64 bytesleft;
1964 : uint32 seg_offset;
1965 : XLogRecPtr result;
1966 :
1967 58982110 : fullsegs = bytepos / UsableBytesInSegment;
1968 58982110 : bytesleft = bytepos % UsableBytesInSegment;
1969 :
1970 58982110 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1971 : {
1972 : /* fits on first page of segment */
1973 94566 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1974 : }
1975 : else
1976 : {
1977 : /* account for the first page on segment with long header */
1978 58887544 : seg_offset = XLOG_BLCKSZ;
1979 58887544 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1980 :
1981 58887544 : fullpages = bytesleft / UsableBytesInPage;
1982 58887544 : bytesleft = bytesleft % UsableBytesInPage;
1983 :
1984 58887544 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1985 : }
1986 :
1987 58982110 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1988 :
1989 58982110 : return result;
1990 : }
1991 :
1992 : /*
1993 : * Like XLogBytePosToRecPtr, but if the position is at a page boundary,
1994 : * returns a pointer to the beginning of the page (ie. before page header),
1995 : * not to where the first xlog record on that page would go to. This is used
1996 : * when converting a pointer to the end of a record.
1997 : */
1998 : static XLogRecPtr
1999 30011978 : XLogBytePosToEndRecPtr(uint64 bytepos)
2000 : {
2001 : uint64 fullsegs;
2002 : uint64 fullpages;
2003 : uint64 bytesleft;
2004 : uint32 seg_offset;
2005 : XLogRecPtr result;
2006 :
2007 30011978 : fullsegs = bytepos / UsableBytesInSegment;
2008 30011978 : bytesleft = bytepos % UsableBytesInSegment;
2009 :
2010 30011978 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
2011 : {
2012 : /* fits on first page of segment */
2013 317366 : if (bytesleft == 0)
2014 270178 : seg_offset = 0;
2015 : else
2016 47188 : seg_offset = bytesleft + SizeOfXLogLongPHD;
2017 : }
2018 : else
2019 : {
2020 : /* account for the first page on segment with long header */
2021 29694612 : seg_offset = XLOG_BLCKSZ;
2022 29694612 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
2023 :
2024 29694612 : fullpages = bytesleft / UsableBytesInPage;
2025 29694612 : bytesleft = bytesleft % UsableBytesInPage;
2026 :
2027 29694612 : if (bytesleft == 0)
2028 30790 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft;
2029 : else
2030 29663822 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
2031 : }
2032 :
2033 30011978 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
2034 :
2035 30011978 : return result;
2036 : }
2037 :
2038 : /*
2039 : * Convert an XLogRecPtr to a "usable byte position".
2040 : */
2041 : static uint64
2042 2508 : XLogRecPtrToBytePos(XLogRecPtr ptr)
2043 : {
2044 : uint64 fullsegs;
2045 : uint32 fullpages;
2046 : uint32 offset;
2047 : uint64 result;
2048 :
2049 2508 : XLByteToSeg(ptr, fullsegs, wal_segment_size);
2050 :
2051 2508 : fullpages = (XLogSegmentOffset(ptr, wal_segment_size)) / XLOG_BLCKSZ;
2052 2508 : offset = ptr % XLOG_BLCKSZ;
2053 :
2054 2508 : if (fullpages == 0)
2055 : {
2056 892 : result = fullsegs * UsableBytesInSegment;
2057 892 : if (offset > 0)
2058 : {
2059 : Assert(offset >= SizeOfXLogLongPHD);
2060 706 : result += offset - SizeOfXLogLongPHD;
2061 : }
2062 : }
2063 : else
2064 : {
2065 3232 : result = fullsegs * UsableBytesInSegment +
2066 3232 : (XLOG_BLCKSZ - SizeOfXLogLongPHD) + /* account for first page */
2067 1616 : (fullpages - 1) * UsableBytesInPage; /* full pages */
2068 1616 : if (offset > 0)
2069 : {
2070 : Assert(offset >= SizeOfXLogShortPHD);
2071 1616 : result += offset - SizeOfXLogShortPHD;
2072 : }
2073 : }
2074 :
2075 2508 : return result;
2076 : }
2077 :
2078 : /*
2079 : * Initialize XLOG buffers, writing out old buffers if they still contain
2080 : * unwritten data, upto the page containing 'upto'. Or if 'opportunistic' is
2081 : * true, initialize as many pages as we can without having to write out
2082 : * unwritten data. Any new pages are initialized to zeros, with pages headers
2083 : * initialized properly.
2084 : */
2085 : static void
2086 600386 : AdvanceXLInsertBuffer(XLogRecPtr upto, bool opportunistic)
2087 : {
2088 600386 : XLogCtlInsert *Insert = &XLogCtl->Insert;
2089 : int nextidx;
2090 : XLogRecPtr OldPageRqstPtr;
2091 : XLogwrtRqst WriteRqst;
2092 600386 : XLogRecPtr NewPageEndPtr = InvalidXLogRecPtr;
2093 : XLogRecPtr NewPageBeginPtr;
2094 : XLogPageHeader NewPage;
2095 600386 : int npages = 0;
2096 :
2097 600386 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2098 :
2099 : /*
2100 : * Now that we have the lock, check if someone initialized the page
2101 : * already.
2102 : */
2103 2290204 : while (upto >= XLogCtl->InitializedUpTo || opportunistic)
2104 : {
2105 1098302 : nextidx = XLogRecPtrToBufIdx(XLogCtl->InitializedUpTo);
2106 :
2107 : /*
2108 : * Get ending-offset of the buffer page we need to replace (this may
2109 : * be zero if the buffer hasn't been used yet). Fall through if it's
2110 : * already written out.
2111 : */
2112 1098302 : OldPageRqstPtr = XLogCtl->xlblocks[nextidx];
2113 1098302 : if (LogwrtResult.Write < OldPageRqstPtr)
2114 : {
2115 : /*
2116 : * Nope, got work to do. If we just want to pre-initialize as much
2117 : * as we can without flushing, give up now.
2118 : */
2119 293642 : if (opportunistic)
2120 8870 : break;
2121 :
2122 : /* Before waiting, get info_lck and update LogwrtResult */
2123 284772 : SpinLockAcquire(&XLogCtl->info_lck);
2124 284772 : if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
2125 269942 : XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
2126 284772 : LogwrtResult = XLogCtl->LogwrtResult;
2127 284772 : SpinLockRelease(&XLogCtl->info_lck);
2128 :
2129 : /*
2130 : * Now that we have an up-to-date LogwrtResult value, see if we
2131 : * still need to write it or if someone else already did.
2132 : */
2133 284772 : if (LogwrtResult.Write < OldPageRqstPtr)
2134 : {
2135 : /*
2136 : * Must acquire write lock. Release WALBufMappingLock first,
2137 : * to make sure that all insertions that we need to wait for
2138 : * can finish (up to this same position). Otherwise we risk
2139 : * deadlock.
2140 : */
2141 284442 : LWLockRelease(WALBufMappingLock);
2142 :
2143 284442 : WaitXLogInsertionsToFinish(OldPageRqstPtr);
2144 :
2145 284442 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
2146 :
2147 284442 : LogwrtResult = XLogCtl->LogwrtResult;
2148 284442 : if (LogwrtResult.Write >= OldPageRqstPtr)
2149 : {
2150 : /* OK, someone wrote it already */
2151 66 : LWLockRelease(WALWriteLock);
2152 : }
2153 : else
2154 : {
2155 : /* Have to write it ourselves */
2156 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_START();
2157 284376 : WriteRqst.Write = OldPageRqstPtr;
2158 284376 : WriteRqst.Flush = 0;
2159 284376 : XLogWrite(WriteRqst, false);
2160 284376 : LWLockRelease(WALWriteLock);
2161 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_DONE();
2162 : }
2163 : /* Re-acquire WALBufMappingLock and retry */
2164 284442 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2165 284442 : continue;
2166 : }
2167 : }
2168 :
2169 : /*
2170 : * Now the next buffer slot is free and we can set it up to be the
2171 : * next output page.
2172 : */
2173 804990 : NewPageBeginPtr = XLogCtl->InitializedUpTo;
2174 804990 : NewPageEndPtr = NewPageBeginPtr + XLOG_BLCKSZ;
2175 :
2176 : Assert(XLogRecPtrToBufIdx(NewPageBeginPtr) == nextidx);
2177 :
2178 804990 : NewPage = (XLogPageHeader) (XLogCtl->pages + nextidx * (Size) XLOG_BLCKSZ);
2179 :
2180 : /*
2181 : * Be sure to re-zero the buffer so that bytes beyond what we've
2182 : * written will look like zeroes and not valid XLOG records...
2183 : */
2184 804990 : MemSet((char *) NewPage, 0, XLOG_BLCKSZ);
2185 :
2186 : /*
2187 : * Fill the new page's header
2188 : */
2189 804990 : NewPage->xlp_magic = XLOG_PAGE_MAGIC;
2190 :
2191 : /* NewPage->xlp_info = 0; */ /* done by memset */
2192 804990 : NewPage->xlp_tli = ThisTimeLineID;
2193 804990 : NewPage->xlp_pageaddr = NewPageBeginPtr;
2194 :
2195 : /* NewPage->xlp_rem_len = 0; */ /* done by memset */
2196 :
2197 : /*
2198 : * If online backup is not in progress, mark the header to indicate
2199 : * that WAL records beginning in this page have removable backup
2200 : * blocks. This allows the WAL archiver to know whether it is safe to
2201 : * compress archived WAL data by transforming full-block records into
2202 : * the non-full-block format. It is sufficient to record this at the
2203 : * page level because we force a page switch (in fact a segment
2204 : * switch) when starting a backup, so the flag will be off before any
2205 : * records can be written during the backup. At the end of a backup,
2206 : * the last page will be marked as all unsafe when perhaps only part
2207 : * is unsafe, but at worst the archiver would miss the opportunity to
2208 : * compress a few records.
2209 : */
2210 804990 : if (!Insert->forcePageWrites)
2211 697460 : NewPage->xlp_info |= XLP_BKP_REMOVABLE;
2212 :
2213 : /*
2214 : * If first page of an XLOG segment file, make it a long header.
2215 : */
2216 804990 : if ((XLogSegmentOffset(NewPage->xlp_pageaddr, wal_segment_size)) == 0)
2217 : {
2218 732 : XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
2219 :
2220 732 : NewLongPage->xlp_sysid = ControlFile->system_identifier;
2221 732 : NewLongPage->xlp_seg_size = wal_segment_size;
2222 732 : NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
2223 732 : NewPage->xlp_info |= XLP_LONG_HEADER;
2224 : }
2225 :
2226 : /*
2227 : * Make sure the initialization of the page becomes visible to others
2228 : * before the xlblocks update. GetXLogBuffer() reads xlblocks without
2229 : * holding a lock.
2230 : */
2231 804990 : pg_write_barrier();
2232 :
2233 804990 : *((volatile XLogRecPtr *) &XLogCtl->xlblocks[nextidx]) = NewPageEndPtr;
2234 :
2235 804990 : XLogCtl->InitializedUpTo = NewPageEndPtr;
2236 :
2237 804990 : npages++;
2238 : }
2239 600386 : LWLockRelease(WALBufMappingLock);
2240 :
2241 : #ifdef WAL_DEBUG
2242 : if (XLOG_DEBUG && npages > 0)
2243 : {
2244 : elog(DEBUG1, "initialized %d pages, up to %X/%X",
2245 : npages, (uint32) (NewPageEndPtr >> 32), (uint32) NewPageEndPtr);
2246 : }
2247 : #endif
2248 600386 : }
2249 :
2250 : /*
2251 : * Calculate CheckPointSegments based on max_wal_size_mb and
2252 : * checkpoint_completion_target.
2253 : */
2254 : static void
2255 7640 : CalculateCheckpointSegments(void)
2256 : {
2257 : double target;
2258 :
2259 : /*-------
2260 : * Calculate the distance at which to trigger a checkpoint, to avoid
2261 : * exceeding max_wal_size_mb. This is based on two assumptions:
2262 : *
2263 : * a) we keep WAL for only one checkpoint cycle (prior to PG11 we kept
2264 : * WAL for two checkpoint cycles to allow us to recover from the
2265 : * secondary checkpoint if the first checkpoint failed, though we
2266 : * only did this on the master anyway, not on standby. Keeping just
2267 : * one checkpoint simplifies processing and reduces disk space in
2268 : * many smaller databases.)
2269 : * b) during checkpoint, we consume checkpoint_completion_target *
2270 : * number of segments consumed between checkpoints.
2271 : *-------
2272 : */
2273 15280 : target = (double) ConvertToXSegs(max_wal_size_mb, wal_segment_size) /
2274 7640 : (1.0 + CheckPointCompletionTarget);
2275 :
2276 : /* round down */
2277 7640 : CheckPointSegments = (int) target;
2278 :
2279 7640 : if (CheckPointSegments < 1)
2280 0 : CheckPointSegments = 1;
2281 7640 : }
2282 :
2283 : void
2284 6388 : assign_max_wal_size(int newval, void *extra)
2285 : {
2286 6388 : max_wal_size_mb = newval;
2287 6388 : CalculateCheckpointSegments();
2288 6388 : }
2289 :
2290 : void
2291 0 : assign_checkpoint_completion_target(double newval, void *extra)
2292 : {
2293 0 : CheckPointCompletionTarget = newval;
2294 0 : CalculateCheckpointSegments();
2295 0 : }
2296 :
2297 : /*
2298 : * At a checkpoint, how many WAL segments to recycle as preallocated future
2299 : * XLOG segments? Returns the highest segment that should be preallocated.
2300 : */
2301 : static XLogSegNo
2302 660 : XLOGfileslop(XLogRecPtr RedoRecPtr)
2303 : {
2304 : XLogSegNo minSegNo;
2305 : XLogSegNo maxSegNo;
2306 : double distance;
2307 : XLogSegNo recycleSegNo;
2308 :
2309 : /*
2310 : * Calculate the segment numbers that min_wal_size_mb and max_wal_size_mb
2311 : * correspond to. Always recycle enough segments to meet the minimum, and
2312 : * remove enough segments to stay below the maximum.
2313 : */
2314 1320 : minSegNo = RedoRecPtr / wal_segment_size +
2315 660 : ConvertToXSegs(min_wal_size_mb, wal_segment_size) - 1;
2316 1320 : maxSegNo = RedoRecPtr / wal_segment_size +
2317 660 : ConvertToXSegs(max_wal_size_mb, wal_segment_size) - 1;
2318 :
2319 : /*
2320 : * Between those limits, recycle enough segments to get us through to the
2321 : * estimated end of next checkpoint.
2322 : *
2323 : * To estimate where the next checkpoint will finish, assume that the
2324 : * system runs steadily consuming CheckPointDistanceEstimate bytes between
2325 : * every checkpoint.
2326 : */
2327 660 : distance = (1.0 + CheckPointCompletionTarget) * CheckPointDistanceEstimate;
2328 : /* add 10% for good measure. */
2329 660 : distance *= 1.10;
2330 :
2331 660 : recycleSegNo = (XLogSegNo) ceil(((double) RedoRecPtr + distance) /
2332 : wal_segment_size);
2333 :
2334 660 : if (recycleSegNo < minSegNo)
2335 62 : recycleSegNo = minSegNo;
2336 660 : if (recycleSegNo > maxSegNo)
2337 466 : recycleSegNo = maxSegNo;
2338 :
2339 660 : return recycleSegNo;
2340 : }
2341 :
2342 : /*
2343 : * Check whether we've consumed enough xlog space that a checkpoint is needed.
2344 : *
2345 : * new_segno indicates a log file that has just been filled up (or read
2346 : * during recovery). We measure the distance from RedoRecPtr to new_segno
2347 : * and see if that exceeds CheckPointSegments.
2348 : *
2349 : * Note: it is caller's responsibility that RedoRecPtr is up-to-date.
2350 : */
2351 : static bool
2352 810 : XLogCheckpointNeeded(XLogSegNo new_segno)
2353 : {
2354 : XLogSegNo old_segno;
2355 :
2356 810 : XLByteToSeg(RedoRecPtr, old_segno, wal_segment_size);
2357 :
2358 810 : if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
2359 30 : return true;
2360 780 : return false;
2361 : }
2362 :
2363 : /*
2364 : * Write and/or fsync the log at least as far as WriteRqst indicates.
2365 : *
2366 : * If flexible == true, we don't have to write as far as WriteRqst, but
2367 : * may stop at any convenient boundary (such as a cache or logfile boundary).
2368 : * This option allows us to avoid uselessly issuing multiple writes when a
2369 : * single one would do.
2370 : *
2371 : * Must be called with WALWriteLock held. WaitXLogInsertionsToFinish(WriteRqst)
2372 : * must be called before grabbing the lock, to make sure the data is ready to
2373 : * write.
2374 : */
2375 : static void
2376 529600 : XLogWrite(XLogwrtRqst WriteRqst, bool flexible)
2377 : {
2378 : bool ispartialpage;
2379 : bool last_iteration;
2380 : bool finishing_seg;
2381 : bool use_existent;
2382 : int curridx;
2383 : int npages;
2384 : int startidx;
2385 : uint32 startoffset;
2386 :
2387 : /* We should always be inside a critical section here */
2388 : Assert(CritSectionCount > 0);
2389 :
2390 : /*
2391 : * Update local LogwrtResult (caller probably did this already, but...)
2392 : */
2393 529600 : LogwrtResult = XLogCtl->LogwrtResult;
2394 :
2395 : /*
2396 : * Since successive pages in the xlog cache are consecutively allocated,
2397 : * we can usually gather multiple pages together and issue just one
2398 : * write() call. npages is the number of pages we have determined can be
2399 : * written together; startidx is the cache block index of the first one,
2400 : * and startoffset is the file offset at which it should go. The latter
2401 : * two variables are only valid when npages > 0, but we must initialize
2402 : * all of them to keep the compiler quiet.
2403 : */
2404 529600 : npages = 0;
2405 529600 : startidx = 0;
2406 529600 : startoffset = 0;
2407 :
2408 : /*
2409 : * Within the loop, curridx is the cache block index of the page to
2410 : * consider writing. Begin at the buffer containing the next unwritten
2411 : * page, or last partially written page.
2412 : */
2413 529600 : curridx = XLogRecPtrToBufIdx(LogwrtResult.Write);
2414 :
2415 1821252 : while (LogwrtResult.Write < WriteRqst.Write)
2416 : {
2417 : /*
2418 : * Make sure we're not ahead of the insert process. This could happen
2419 : * if we're passed a bogus WriteRqst.Write that is past the end of the
2420 : * last page that's been initialized by AdvanceXLInsertBuffer.
2421 : */
2422 1006958 : XLogRecPtr EndPtr = XLogCtl->xlblocks[curridx];
2423 :
2424 1006958 : if (LogwrtResult.Write >= EndPtr)
2425 0 : elog(PANIC, "xlog write request %X/%X is past end of log %X/%X",
2426 : (uint32) (LogwrtResult.Write >> 32),
2427 : (uint32) LogwrtResult.Write,
2428 : (uint32) (EndPtr >> 32), (uint32) EndPtr);
2429 :
2430 : /* Advance LogwrtResult.Write to end of current buffer page */
2431 1006958 : LogwrtResult.Write = EndPtr;
2432 1006958 : ispartialpage = WriteRqst.Write < LogwrtResult.Write;
2433 :
2434 1006958 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2435 : wal_segment_size))
2436 : {
2437 : /*
2438 : * Switch to new logfile segment. We cannot have any pending
2439 : * pages here (since we dump what we have at segment end).
2440 : */
2441 : Assert(npages == 0);
2442 7512 : if (openLogFile >= 0)
2443 1948 : XLogFileClose();
2444 7512 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2445 : wal_segment_size);
2446 :
2447 : /* create/use new log file */
2448 7512 : use_existent = true;
2449 7512 : openLogFile = XLogFileInit(openLogSegNo, &use_existent, true);
2450 : }
2451 :
2452 : /* Make sure we have the current logfile open */
2453 1006958 : if (openLogFile < 0)
2454 : {
2455 0 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2456 : wal_segment_size);
2457 0 : openLogFile = XLogFileOpen(openLogSegNo);
2458 : }
2459 :
2460 : /* Add current page to the set of pending pages-to-dump */
2461 1006958 : if (npages == 0)
2462 : {
2463 : /* first of group */
2464 530820 : startidx = curridx;
2465 530820 : startoffset = XLogSegmentOffset(LogwrtResult.Write - XLOG_BLCKSZ,
2466 : wal_segment_size);
2467 : }
2468 1006958 : npages++;
2469 :
2470 : /*
2471 : * Dump the set if this will be the last loop iteration, or if we are
2472 : * at the last page of the cache area (since the next page won't be
2473 : * contiguous in memory), or if we are at the end of the logfile
2474 : * segment.
2475 : */
2476 1006958 : last_iteration = WriteRqst.Write <= LogwrtResult.Write;
2477 :
2478 1773756 : finishing_seg = !ispartialpage &&
2479 766798 : (startoffset + npages * XLOG_BLCKSZ) >= wal_segment_size;
2480 :
2481 1484512 : if (last_iteration ||
2482 953692 : curridx == XLogCtl->XLogCacheBlck ||
2483 : finishing_seg)
2484 : {
2485 : char *from;
2486 : Size nbytes;
2487 : Size nleft;
2488 : int written;
2489 :
2490 : /* OK to write the page(s) */
2491 530820 : from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
2492 530820 : nbytes = npages * (Size) XLOG_BLCKSZ;
2493 530820 : nleft = nbytes;
2494 : do
2495 : {
2496 530820 : errno = 0;
2497 530820 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
2498 530820 : written = pg_pwrite(openLogFile, from, nleft, startoffset);
2499 530820 : pgstat_report_wait_end();
2500 530820 : if (written <= 0)
2501 : {
2502 0 : if (errno == EINTR)
2503 0 : continue;
2504 0 : ereport(PANIC,
2505 : (errcode_for_file_access(),
2506 : errmsg("could not write to log file %s "
2507 : "at offset %u, length %zu: %m",
2508 : XLogFileNameP(ThisTimeLineID, openLogSegNo),
2509 : startoffset, nleft)));
2510 : }
2511 530820 : nleft -= written;
2512 530820 : from += written;
2513 530820 : startoffset += written;
2514 530820 : } while (nleft > 0);
2515 :
2516 530820 : npages = 0;
2517 :
2518 : /*
2519 : * If we just wrote the whole last page of a logfile segment,
2520 : * fsync the segment immediately. This avoids having to go back
2521 : * and re-open prior segments when an fsync request comes along
2522 : * later. Doing it here ensures that one and only one backend will
2523 : * perform this fsync.
2524 : *
2525 : * This is also the right place to notify the Archiver that the
2526 : * segment is ready to copy to archival storage, and to update the
2527 : * timer for archive_timeout, and to signal for a checkpoint if
2528 : * too many logfile segments have been used since the last
2529 : * checkpoint.
2530 : */
2531 530820 : if (finishing_seg)
2532 : {
2533 750 : issue_xlog_fsync(openLogFile, openLogSegNo);
2534 :
2535 : /* signal that we need to wakeup walsenders later */
2536 750 : WalSndWakeupRequest();
2537 :
2538 750 : LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
2539 :
2540 750 : if (XLogArchivingActive())
2541 16 : XLogArchiveNotifySeg(openLogSegNo);
2542 :
2543 750 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
2544 750 : XLogCtl->lastSegSwitchLSN = LogwrtResult.Flush;
2545 :
2546 : /*
2547 : * Request a checkpoint if we've consumed too much xlog since
2548 : * the last one. For speed, we first check using the local
2549 : * copy of RedoRecPtr, which might be out of date; if it looks
2550 : * like a checkpoint is needed, forcibly update RedoRecPtr and
2551 : * recheck.
2552 : */
2553 750 : if (IsUnderPostmaster && XLogCheckpointNeeded(openLogSegNo))
2554 : {
2555 20 : (void) GetRedoRecPtr();
2556 20 : if (XLogCheckpointNeeded(openLogSegNo))
2557 10 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
2558 : }
2559 : }
2560 : }
2561 :
2562 1006958 : if (ispartialpage)
2563 : {
2564 : /* Only asked to write a partial page */
2565 240160 : LogwrtResult.Write = WriteRqst.Write;
2566 240160 : break;
2567 : }
2568 766798 : curridx = NextBufIdx(curridx);
2569 :
2570 : /* If flexible, break out of loop as soon as we wrote something */
2571 766798 : if (flexible && npages == 0)
2572 4746 : break;
2573 : }
2574 :
2575 : Assert(npages == 0);
2576 :
2577 : /*
2578 : * If asked to flush, do so
2579 : */
2580 770952 : if (LogwrtResult.Flush < WriteRqst.Flush &&
2581 241352 : LogwrtResult.Flush < LogwrtResult.Write)
2582 :
2583 : {
2584 : /*
2585 : * Could get here without iterating above loop, in which case we might
2586 : * have no open file or the wrong one. However, we do not need to
2587 : * fsync more than one file.
2588 : */
2589 482612 : if (sync_method != SYNC_METHOD_OPEN &&
2590 241306 : sync_method != SYNC_METHOD_OPEN_DSYNC)
2591 : {
2592 482604 : if (openLogFile >= 0 &&
2593 241298 : !XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2594 : wal_segment_size))
2595 0 : XLogFileClose();
2596 241306 : if (openLogFile < 0)
2597 : {
2598 8 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2599 : wal_segment_size);
2600 8 : openLogFile = XLogFileOpen(openLogSegNo);
2601 : }
2602 :
2603 241306 : issue_xlog_fsync(openLogFile, openLogSegNo);
2604 : }
2605 :
2606 : /* signal that we need to wakeup walsenders later */
2607 241306 : WalSndWakeupRequest();
2608 :
2609 241306 : LogwrtResult.Flush = LogwrtResult.Write;
2610 : }
2611 :
2612 : /*
2613 : * Update shared-memory status
2614 : *
2615 : * We make sure that the shared 'request' values do not fall behind the
2616 : * 'result' values. This is not absolutely essential, but it saves some
2617 : * code in a couple of places.
2618 : */
2619 : {
2620 529600 : SpinLockAcquire(&XLogCtl->info_lck);
2621 529600 : XLogCtl->LogwrtResult = LogwrtResult;
2622 529600 : if (XLogCtl->LogwrtRqst.Write < LogwrtResult.Write)
2623 231610 : XLogCtl->LogwrtRqst.Write = LogwrtResult.Write;
2624 529600 : if (XLogCtl->LogwrtRqst.Flush < LogwrtResult.Flush)
2625 241612 : XLogCtl->LogwrtRqst.Flush = LogwrtResult.Flush;
2626 529600 : SpinLockRelease(&XLogCtl->info_lck);
2627 : }
2628 529600 : }
2629 :
2630 : /*
2631 : * Record the LSN for an asynchronous transaction commit/abort
2632 : * and nudge the WALWriter if there is work for it to do.
2633 : * (This should not be called for synchronous commits.)
2634 : */
2635 : void
2636 59238 : XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN)
2637 : {
2638 59238 : XLogRecPtr WriteRqstPtr = asyncXactLSN;
2639 : bool sleeping;
2640 :
2641 59238 : SpinLockAcquire(&XLogCtl->info_lck);
2642 59238 : LogwrtResult = XLogCtl->LogwrtResult;
2643 59238 : sleeping = XLogCtl->WalWriterSleeping;
2644 59238 : if (XLogCtl->asyncXactLSN < asyncXactLSN)
2645 58982 : XLogCtl->asyncXactLSN = asyncXactLSN;
2646 59238 : SpinLockRelease(&XLogCtl->info_lck);
2647 :
2648 : /*
2649 : * If the WALWriter is sleeping, we should kick it to make it come out of
2650 : * low-power mode. Otherwise, determine whether there's a full page of
2651 : * WAL available to write.
2652 : */
2653 59238 : if (!sleeping)
2654 : {
2655 : /* back off to last completed page boundary */
2656 59216 : WriteRqstPtr -= WriteRqstPtr % XLOG_BLCKSZ;
2657 :
2658 : /* if we have already flushed that far, we're done */
2659 59216 : if (WriteRqstPtr <= LogwrtResult.Flush)
2660 12546 : return;
2661 : }
2662 :
2663 : /*
2664 : * Nudge the WALWriter: it has a full page of WAL to write, or we want it
2665 : * to come out of low-power mode so that this async commit will reach disk
2666 : * within the expected amount of time.
2667 : */
2668 46692 : if (ProcGlobal->walwriterLatch)
2669 11916 : SetLatch(ProcGlobal->walwriterLatch);
2670 : }
2671 :
2672 : /*
2673 : * Record the LSN up to which we can remove WAL because it's not required by
2674 : * any replication slot.
2675 : */
2676 : void
2677 1968 : XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn)
2678 : {
2679 1968 : SpinLockAcquire(&XLogCtl->info_lck);
2680 1968 : XLogCtl->replicationSlotMinLSN = lsn;
2681 1968 : SpinLockRelease(&XLogCtl->info_lck);
2682 1968 : }
2683 :
2684 :
2685 : /*
2686 : * Return the oldest LSN we must retain to satisfy the needs of some
2687 : * replication slot.
2688 : */
2689 : static XLogRecPtr
2690 2832 : XLogGetReplicationSlotMinimumLSN(void)
2691 : {
2692 : XLogRecPtr retval;
2693 :
2694 2832 : SpinLockAcquire(&XLogCtl->info_lck);
2695 2832 : retval = XLogCtl->replicationSlotMinLSN;
2696 2832 : SpinLockRelease(&XLogCtl->info_lck);
2697 :
2698 2832 : return retval;
2699 : }
2700 :
2701 : /*
2702 : * Advance minRecoveryPoint in control file.
2703 : *
2704 : * If we crash during recovery, we must reach this point again before the
2705 : * database is consistent.
2706 : *
2707 : * If 'force' is true, 'lsn' argument is ignored. Otherwise, minRecoveryPoint
2708 : * is only updated if it's not already greater than or equal to 'lsn'.
2709 : */
2710 : static void
2711 3848 : UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
2712 : {
2713 : /* Quick check using our local copy of the variable */
2714 3848 : if (!updateMinRecoveryPoint || (!force && lsn <= minRecoveryPoint))
2715 3502 : return;
2716 :
2717 : /*
2718 : * An invalid minRecoveryPoint means that we need to recover all the WAL,
2719 : * i.e., we're doing crash recovery. We never modify the control file's
2720 : * value in that case, so we can short-circuit future checks here too. The
2721 : * local values of minRecoveryPoint and minRecoveryPointTLI should not be
2722 : * updated until crash recovery finishes. We only do this for the startup
2723 : * process as it should not update its own reference of minRecoveryPoint
2724 : * until it has finished crash recovery to make sure that all WAL
2725 : * available is replayed in this case. This also saves from extra locks
2726 : * taken on the control file from the startup process.
2727 : */
2728 346 : if (XLogRecPtrIsInvalid(minRecoveryPoint) && InRecovery)
2729 : {
2730 10 : updateMinRecoveryPoint = false;
2731 10 : return;
2732 : }
2733 :
2734 336 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2735 :
2736 : /* update local copy */
2737 336 : minRecoveryPoint = ControlFile->minRecoveryPoint;
2738 336 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
2739 :
2740 336 : if (XLogRecPtrIsInvalid(minRecoveryPoint))
2741 0 : updateMinRecoveryPoint = false;
2742 336 : else if (force || minRecoveryPoint < lsn)
2743 : {
2744 : XLogRecPtr newMinRecoveryPoint;
2745 : TimeLineID newMinRecoveryPointTLI;
2746 :
2747 : /*
2748 : * To avoid having to update the control file too often, we update it
2749 : * all the way to the last record being replayed, even though 'lsn'
2750 : * would suffice for correctness. This also allows the 'force' case
2751 : * to not need a valid 'lsn' value.
2752 : *
2753 : * Another important reason for doing it this way is that the passed
2754 : * 'lsn' value could be bogus, i.e., past the end of available WAL, if
2755 : * the caller got it from a corrupted heap page. Accepting such a
2756 : * value as the min recovery point would prevent us from coming up at
2757 : * all. Instead, we just log a warning and continue with recovery.
2758 : * (See also the comments about corrupt LSNs in XLogFlush.)
2759 : */
2760 288 : SpinLockAcquire(&XLogCtl->info_lck);
2761 288 : newMinRecoveryPoint = XLogCtl->replayEndRecPtr;
2762 288 : newMinRecoveryPointTLI = XLogCtl->replayEndTLI;
2763 288 : SpinLockRelease(&XLogCtl->info_lck);
2764 :
2765 288 : if (!force && newMinRecoveryPoint < lsn)
2766 0 : elog(WARNING,
2767 : "xlog min recovery request %X/%X is past current point %X/%X",
2768 : (uint32) (lsn >> 32), (uint32) lsn,
2769 : (uint32) (newMinRecoveryPoint >> 32),
2770 : (uint32) newMinRecoveryPoint);
2771 :
2772 : /* update control file */
2773 288 : if (ControlFile->minRecoveryPoint < newMinRecoveryPoint)
2774 : {
2775 274 : ControlFile->minRecoveryPoint = newMinRecoveryPoint;
2776 274 : ControlFile->minRecoveryPointTLI = newMinRecoveryPointTLI;
2777 274 : UpdateControlFile();
2778 274 : minRecoveryPoint = newMinRecoveryPoint;
2779 274 : minRecoveryPointTLI = newMinRecoveryPointTLI;
2780 :
2781 274 : ereport(DEBUG2,
2782 : (errmsg("updated min recovery point to %X/%X on timeline %u",
2783 : (uint32) (minRecoveryPoint >> 32),
2784 : (uint32) minRecoveryPoint,
2785 : newMinRecoveryPointTLI)));
2786 : }
2787 : }
2788 336 : LWLockRelease(ControlFileLock);
2789 : }
2790 :
2791 : /*
2792 : * Ensure that all XLOG data through the given position is flushed to disk.
2793 : *
2794 : * NOTE: this differs from XLogWrite mainly in that the WALWriteLock is not
2795 : * already held, and we try to avoid acquiring it if possible.
2796 : */
2797 : void
2798 711440 : XLogFlush(XLogRecPtr record)
2799 : {
2800 : XLogRecPtr WriteRqstPtr;
2801 : XLogwrtRqst WriteRqst;
2802 :
2803 : /*
2804 : * During REDO, we are reading not writing WAL. Therefore, instead of
2805 : * trying to flush the WAL, we should update minRecoveryPoint instead. We
2806 : * test XLogInsertAllowed(), not InRecovery, because we need checkpointer
2807 : * to act this way too, and because when it tries to write the
2808 : * end-of-recovery checkpoint, it should indeed flush.
2809 : */
2810 711440 : if (!XLogInsertAllowed())
2811 : {
2812 3784 : UpdateMinRecoveryPoint(record, false);
2813 3784 : return;
2814 : }
2815 :
2816 : /* Quick exit if already known flushed */
2817 707656 : if (record <= LogwrtResult.Flush)
2818 466308 : return;
2819 :
2820 : #ifdef WAL_DEBUG
2821 : if (XLOG_DEBUG)
2822 : elog(LOG, "xlog flush request %X/%X; write %X/%X; flush %X/%X",
2823 : (uint32) (record >> 32), (uint32) record,
2824 : (uint32) (LogwrtResult.Write >> 32), (uint32) LogwrtResult.Write,
2825 : (uint32) (LogwrtResult.Flush >> 32), (uint32) LogwrtResult.Flush);
2826 : #endif
2827 :
2828 241348 : START_CRIT_SECTION();
2829 :
2830 : /*
2831 : * Since fsync is usually a horribly expensive operation, we try to
2832 : * piggyback as much data as we can on each fsync: if we see any more data
2833 : * entered into the xlog buffer, we'll write and fsync that too, so that
2834 : * the final value of LogwrtResult.Flush is as large as possible. This
2835 : * gives us some chance of avoiding another fsync immediately after.
2836 : */
2837 :
2838 : /* initialize to given target; may increase below */
2839 241348 : WriteRqstPtr = record;
2840 :
2841 : /*
2842 : * Now wait until we get the write lock, or someone else does the flush
2843 : * for us.
2844 : */
2845 : for (;;)
2846 520 : {
2847 : XLogRecPtr insertpos;
2848 :
2849 : /* read LogwrtResult and update local state */
2850 241868 : SpinLockAcquire(&XLogCtl->info_lck);
2851 241868 : if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2852 2068 : WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2853 241868 : LogwrtResult = XLogCtl->LogwrtResult;
2854 241868 : SpinLockRelease(&XLogCtl->info_lck);
2855 :
2856 : /* done already? */
2857 241868 : if (record <= LogwrtResult.Flush)
2858 1084 : break;
2859 :
2860 : /*
2861 : * Before actually performing the write, wait for all in-flight
2862 : * insertions to the pages we're about to write to finish.
2863 : */
2864 240784 : insertpos = WaitXLogInsertionsToFinish(WriteRqstPtr);
2865 :
2866 : /*
2867 : * Try to get the write lock. If we can't get it immediately, wait
2868 : * until it's released, and recheck if we still need to do the flush
2869 : * or if the backend that held the lock did it for us already. This
2870 : * helps to maintain a good rate of group committing when the system
2871 : * is bottlenecked by the speed of fsyncing.
2872 : */
2873 240784 : if (!LWLockAcquireOrWait(WALWriteLock, LW_EXCLUSIVE))
2874 : {
2875 : /*
2876 : * The lock is now free, but we didn't acquire it yet. Before we
2877 : * do, loop back to check if someone else flushed the record for
2878 : * us already.
2879 : */
2880 520 : continue;
2881 : }
2882 :
2883 : /* Got the lock; recheck whether request is satisfied */
2884 240264 : LogwrtResult = XLogCtl->LogwrtResult;
2885 240264 : if (record <= LogwrtResult.Flush)
2886 : {
2887 150 : LWLockRelease(WALWriteLock);
2888 150 : break;
2889 : }
2890 :
2891 : /*
2892 : * Sleep before flush! By adding a delay here, we may give further
2893 : * backends the opportunity to join the backlog of group commit
2894 : * followers; this can significantly improve transaction throughput,
2895 : * at the risk of increasing transaction latency.
2896 : *
2897 : * We do not sleep if enableFsync is not turned on, nor if there are
2898 : * fewer than CommitSiblings other backends with active transactions.
2899 : */
2900 240114 : if (CommitDelay > 0 && enableFsync &&
2901 0 : MinimumActiveBackends(CommitSiblings))
2902 : {
2903 0 : pg_usleep(CommitDelay);
2904 :
2905 : /*
2906 : * Re-check how far we can now flush the WAL. It's generally not
2907 : * safe to call WaitXLogInsertionsToFinish while holding
2908 : * WALWriteLock, because an in-progress insertion might need to
2909 : * also grab WALWriteLock to make progress. But we know that all
2910 : * the insertions up to insertpos have already finished, because
2911 : * that's what the earlier WaitXLogInsertionsToFinish() returned.
2912 : * We're only calling it again to allow insertpos to be moved
2913 : * further forward, not to actually wait for anyone.
2914 : */
2915 0 : insertpos = WaitXLogInsertionsToFinish(insertpos);
2916 : }
2917 :
2918 : /* try to write/flush later additions to XLOG as well */
2919 240114 : WriteRqst.Write = insertpos;
2920 240114 : WriteRqst.Flush = insertpos;
2921 :
2922 240114 : XLogWrite(WriteRqst, false);
2923 :
2924 240114 : LWLockRelease(WALWriteLock);
2925 : /* done */
2926 240114 : break;
2927 : }
2928 :
2929 241348 : END_CRIT_SECTION();
2930 :
2931 : /* wake up walsenders now that we've released heavily contended locks */
2932 241348 : WalSndWakeupProcessRequests();
2933 :
2934 : /*
2935 : * If we still haven't flushed to the request point then we have a
2936 : * problem; most likely, the requested flush point is past end of XLOG.
2937 : * This has been seen to occur when a disk page has a corrupted LSN.
2938 : *
2939 : * Formerly we treated this as a PANIC condition, but that hurts the
2940 : * system's robustness rather than helping it: we do not want to take down
2941 : * the whole system due to corruption on one data page. In particular, if
2942 : * the bad page is encountered again during recovery then we would be
2943 : * unable to restart the database at all! (This scenario actually
2944 : * happened in the field several times with 7.1 releases.) As of 8.4, bad
2945 : * LSNs encountered during recovery are UpdateMinRecoveryPoint's problem;
2946 : * the only time we can reach here during recovery is while flushing the
2947 : * end-of-recovery checkpoint record, and we don't expect that to have a
2948 : * bad LSN.
2949 : *
2950 : * Note that for calls from xact.c, the ERROR will be promoted to PANIC
2951 : * since xact.c calls this routine inside a critical section. However,
2952 : * calls from bufmgr.c are not within critical sections and so we will not
2953 : * force a restart for a bad LSN on a data page.
2954 : */
2955 241348 : if (LogwrtResult.Flush < record)
2956 0 : elog(ERROR,
2957 : "xlog flush request %X/%X is not satisfied --- flushed only to %X/%X",
2958 : (uint32) (record >> 32), (uint32) record,
2959 : (uint32) (LogwrtResult.Flush >> 32), (uint32) LogwrtResult.Flush);
2960 : }
2961 :
2962 : /*
2963 : * Write & flush xlog, but without specifying exactly where to.
2964 : *
2965 : * We normally write only completed blocks; but if there is nothing to do on
2966 : * that basis, we check for unwritten async commits in the current incomplete
2967 : * block, and write through the latest one of those. Thus, if async commits
2968 : * are not being used, we will write complete blocks only.
2969 : *
2970 : * If, based on the above, there's anything to write we do so immediately. But
2971 : * to avoid calling fsync, fdatasync et. al. at a rate that'd impact
2972 : * concurrent IO, we only flush WAL every wal_writer_delay ms, or if there's
2973 : * more than wal_writer_flush_after unflushed blocks.
2974 : *
2975 : * We can guarantee that async commits reach disk after at most three
2976 : * wal_writer_delay cycles. (When flushing complete blocks, we allow XLogWrite
2977 : * to write "flexibly", meaning it can stop at the end of the buffer ring;
2978 : * this makes a difference only with very high load or long wal_writer_delay,
2979 : * but imposes one extra cycle for the worst case for async commits.)
2980 : *
2981 : * This routine is invoked periodically by the background walwriter process.
2982 : *
2983 : * Returns true if there was any work to do, even if we skipped flushing due
2984 : * to wal_writer_delay/wal_writer_flush_after.
2985 : */
2986 : bool
2987 16262 : XLogBackgroundFlush(void)
2988 : {
2989 : XLogwrtRqst WriteRqst;
2990 16262 : bool flexible = true;
2991 : static TimestampTz lastflush;
2992 : TimestampTz now;
2993 : int flushbytes;
2994 :
2995 : /* XLOG doesn't need flushing during recovery */
2996 16262 : if (RecoveryInProgress())
2997 0 : return false;
2998 :
2999 : /* read LogwrtResult and update local state */
3000 16262 : SpinLockAcquire(&XLogCtl->info_lck);
3001 16262 : LogwrtResult = XLogCtl->LogwrtResult;
3002 16262 : WriteRqst = XLogCtl->LogwrtRqst;
3003 16262 : SpinLockRelease(&XLogCtl->info_lck);
3004 :
3005 : /* back off to last completed page boundary */
3006 16262 : WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
3007 :
3008 : /* if we have already flushed that far, consider async commit records */
3009 16262 : if (WriteRqst.Write <= LogwrtResult.Flush)
3010 : {
3011 7722 : SpinLockAcquire(&XLogCtl->info_lck);
3012 7722 : WriteRqst.Write = XLogCtl->asyncXactLSN;
3013 7722 : SpinLockRelease(&XLogCtl->info_lck);
3014 7722 : flexible = false; /* ensure it all gets written */
3015 : }
3016 :
3017 : /*
3018 : * If already known flushed, we're done. Just need to check if we are
3019 : * holding an open file handle to a logfile that's no longer in use,
3020 : * preventing the file from being deleted.
3021 : */
3022 16262 : if (WriteRqst.Write <= LogwrtResult.Flush)
3023 : {
3024 7392 : if (openLogFile >= 0)
3025 : {
3026 4054 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
3027 : wal_segment_size))
3028 : {
3029 80 : XLogFileClose();
3030 : }
3031 : }
3032 7392 : return false;
3033 : }
3034 :
3035 : /*
3036 : * Determine how far to flush WAL, based on the wal_writer_delay and
3037 : * wal_writer_flush_after GUCs.
3038 : */
3039 8870 : now = GetCurrentTimestamp();
3040 8870 : flushbytes =
3041 8870 : WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
3042 :
3043 8870 : if (WalWriterFlushAfter == 0 || lastflush == 0)
3044 : {
3045 : /* first call, or block based limits disabled */
3046 140 : WriteRqst.Flush = WriteRqst.Write;
3047 140 : lastflush = now;
3048 : }
3049 8730 : else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
3050 : {
3051 : /*
3052 : * Flush the writes at least every WalWriterDelay ms. This is
3053 : * important to bound the amount of time it takes for an asynchronous
3054 : * commit to hit disk.
3055 : */
3056 1276 : WriteRqst.Flush = WriteRqst.Write;
3057 1276 : lastflush = now;
3058 : }
3059 7454 : else if (flushbytes >= WalWriterFlushAfter)
3060 : {
3061 : /* exceeded wal_writer_flush_after blocks, flush */
3062 10 : WriteRqst.Flush = WriteRqst.Write;
3063 10 : lastflush = now;
3064 : }
3065 : else
3066 : {
3067 : /* no flushing, this time round */
3068 7444 : WriteRqst.Flush = 0;
3069 : }
3070 :
3071 : #ifdef WAL_DEBUG
3072 : if (XLOG_DEBUG)
3073 : elog(LOG, "xlog bg flush request write %X/%X; flush: %X/%X, current is write %X/%X; flush %X/%X",
3074 : (uint32) (WriteRqst.Write >> 32), (uint32) WriteRqst.Write,
3075 : (uint32) (WriteRqst.Flush >> 32), (uint32) WriteRqst.Flush,
3076 : (uint32) (LogwrtResult.Write >> 32), (uint32) LogwrtResult.Write,
3077 : (uint32) (LogwrtResult.Flush >> 32), (uint32) LogwrtResult.Flush);
3078 : #endif
3079 :
3080 8870 : START_CRIT_SECTION();
3081 :
3082 : /* now wait for any in-progress insertions to finish and get write lock */
3083 8870 : WaitXLogInsertionsToFinish(WriteRqst.Write);
3084 8870 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
3085 8870 : LogwrtResult = XLogCtl->LogwrtResult;
3086 12666 : if (WriteRqst.Write > LogwrtResult.Write ||
3087 3796 : WriteRqst.Flush > LogwrtResult.Flush)
3088 : {
3089 5110 : XLogWrite(WriteRqst, flexible);
3090 : }
3091 8870 : LWLockRelease(WALWriteLock);
3092 :
3093 8870 : END_CRIT_SECTION();
3094 :
3095 : /* wake up walsenders now that we've released heavily contended locks */
3096 8870 : WalSndWakeupProcessRequests();
3097 :
3098 : /*
3099 : * Great, done. To take some work off the critical path, try to initialize
3100 : * as many of the no-longer-needed WAL buffers for future use as we can.
3101 : */
3102 8870 : AdvanceXLInsertBuffer(InvalidXLogRecPtr, true);
3103 :
3104 : /*
3105 : * If we determined that we need to write data, but somebody else
3106 : * wrote/flushed already, it should be considered as being active, to
3107 : * avoid hibernating too early.
3108 : */
3109 8870 : return true;
3110 : }
3111 :
3112 : /*
3113 : * Test whether XLOG data has been flushed up to (at least) the given position.
3114 : *
3115 : * Returns true if a flush is still needed. (It may be that someone else
3116 : * is already in process of flushing that far, however.)
3117 : */
3118 : bool
3119 15411688 : XLogNeedsFlush(XLogRecPtr record)
3120 : {
3121 : /*
3122 : * During recovery, we don't flush WAL but update minRecoveryPoint
3123 : * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3124 : * would need to be updated.
3125 : */
3126 15411688 : if (RecoveryInProgress())
3127 : {
3128 : /*
3129 : * An invalid minRecoveryPoint means that we need to recover all the
3130 : * WAL, i.e., we're doing crash recovery. We never modify the control
3131 : * file's value in that case, so we can short-circuit future checks
3132 : * here too. This triggers a quick exit path for the startup process,
3133 : * which cannot update its local copy of minRecoveryPoint as long as
3134 : * it has not replayed all WAL available when doing crash recovery.
3135 : */
3136 112860 : if (XLogRecPtrIsInvalid(minRecoveryPoint) && InRecovery)
3137 0 : updateMinRecoveryPoint = false;
3138 :
3139 : /* Quick exit if already known to be updated or cannot be updated */
3140 112860 : if (record <= minRecoveryPoint || !updateMinRecoveryPoint)
3141 100072 : return false;
3142 :
3143 : /*
3144 : * Update local copy of minRecoveryPoint. But if the lock is busy,
3145 : * just return a conservative guess.
3146 : */
3147 12788 : if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
3148 0 : return true;
3149 12788 : minRecoveryPoint = ControlFile->minRecoveryPoint;
3150 12788 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
3151 12788 : LWLockRelease(ControlFileLock);
3152 :
3153 : /*
3154 : * Check minRecoveryPoint for any other process than the startup
3155 : * process doing crash recovery, which should not update the control
3156 : * file value if crash recovery is still running.
3157 : */
3158 12788 : if (XLogRecPtrIsInvalid(minRecoveryPoint))
3159 0 : updateMinRecoveryPoint = false;
3160 :
3161 : /* check again */
3162 12788 : if (record <= minRecoveryPoint || !updateMinRecoveryPoint)
3163 22 : return false;
3164 : else
3165 12766 : return true;
3166 : }
3167 :
3168 : /* Quick exit if already known flushed */
3169 15298828 : if (record <= LogwrtResult.Flush)
3170 15222858 : return false;
3171 :
3172 : /* read LogwrtResult and update local state */
3173 75970 : SpinLockAcquire(&XLogCtl->info_lck);
3174 75970 : LogwrtResult = XLogCtl->LogwrtResult;
3175 75970 : SpinLockRelease(&XLogCtl->info_lck);
3176 :
3177 : /* check again */
3178 75970 : if (record <= LogwrtResult.Flush)
3179 970 : return false;
3180 :
3181 75000 : return true;
3182 : }
3183 :
3184 : /*
3185 : * Create a new XLOG file segment, or open a pre-existing one.
3186 : *
3187 : * logsegno: identify segment to be created/opened.
3188 : *
3189 : * *use_existent: if true, OK to use a pre-existing file (else, any
3190 : * pre-existing file will be deleted). On return, true if a pre-existing
3191 : * file was used.
3192 : *
3193 : * use_lock: if true, acquire ControlFileLock while moving file into
3194 : * place. This should be true except during bootstrap log creation. The
3195 : * caller must *not* hold the lock at call.
3196 : *
3197 : * Returns FD of opened file.
3198 : *
3199 : * Note: errors here are ERROR not PANIC because we might or might not be
3200 : * inside a critical section (eg, during checkpoint there is no reason to
3201 : * take down the system on failure). They will promote to PANIC if we are
3202 : * in a critical section.
3203 : */
3204 : int
3205 7992 : XLogFileInit(XLogSegNo logsegno, bool *use_existent, bool use_lock)
3206 : {
3207 : char path[MAXPGPATH];
3208 : char tmppath[MAXPGPATH];
3209 : PGAlignedXLogBlock zbuffer;
3210 : XLogSegNo installed_segno;
3211 : XLogSegNo max_segno;
3212 : int fd;
3213 : int nbytes;
3214 : int save_errno;
3215 :
3216 7992 : XLogFilePath(path, ThisTimeLineID, logsegno, wal_segment_size);
3217 :
3218 : /*
3219 : * Try to use existent file (checkpoint maker may have created it already)
3220 : */
3221 7992 : if (*use_existent)
3222 : {
3223 7670 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | get_sync_bit(sync_method));
3224 7670 : if (fd < 0)
3225 : {
3226 348 : if (errno != ENOENT)
3227 0 : ereport(ERROR,
3228 : (errcode_for_file_access(),
3229 : errmsg("could not open file \"%s\": %m", path)));
3230 : }
3231 : else
3232 7322 : return fd;
3233 : }
3234 :
3235 : /*
3236 : * Initialize an empty (all zeroes) segment. NOTE: it is possible that
3237 : * another process is doing the same thing. If so, we will end up
3238 : * pre-creating an extra log segment. That seems OK, and better than
3239 : * holding the lock throughout this lengthy process.
3240 : */
3241 670 : elog(DEBUG2, "creating and filling new WAL file");
3242 :
3243 670 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3244 :
3245 670 : unlink(tmppath);
3246 :
3247 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3248 670 : fd = BasicOpenFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
3249 670 : if (fd < 0)
3250 0 : ereport(ERROR,
3251 : (errcode_for_file_access(),
3252 : errmsg("could not create file \"%s\": %m", tmppath)));
3253 :
3254 670 : memset(zbuffer.data, 0, XLOG_BLCKSZ);
3255 :
3256 670 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_WRITE);
3257 670 : save_errno = 0;
3258 670 : if (wal_init_zero)
3259 : {
3260 : /*
3261 : * Zero-fill the file. With this setting, we do this the hard way to
3262 : * ensure that all the file space has really been allocated. On
3263 : * platforms that allow "holes" in files, just seeking to the end
3264 : * doesn't allocate intermediate space. This way, we know that we
3265 : * have all the space and (after the fsync below) that all the
3266 : * indirect blocks are down on disk. Therefore, fdatasync(2) or
3267 : * O_DSYNC will be sufficient to sync future writes to the log file.
3268 : */
3269 1038750 : for (nbytes = 0; nbytes < wal_segment_size; nbytes += XLOG_BLCKSZ)
3270 : {
3271 1038080 : errno = 0;
3272 1038080 : if (write(fd, zbuffer.data, XLOG_BLCKSZ) != XLOG_BLCKSZ)
3273 : {
3274 : /* if write didn't set errno, assume no disk space */
3275 0 : save_errno = errno ? errno : ENOSPC;
3276 0 : break;
3277 : }
3278 : }
3279 : }
3280 : else
3281 : {
3282 : /*
3283 : * Otherwise, seeking to the end and writing a solitary byte is
3284 : * enough.
3285 : */
3286 0 : errno = 0;
3287 0 : if (pg_pwrite(fd, zbuffer.data, 1, wal_segment_size - 1) != 1)
3288 : {
3289 : /* if write didn't set errno, assume no disk space */
3290 0 : save_errno = errno ? errno : ENOSPC;
3291 : }
3292 : }
3293 670 : pgstat_report_wait_end();
3294 :
3295 670 : if (save_errno)
3296 : {
3297 : /*
3298 : * If we fail to make the file, delete it to release disk space
3299 : */
3300 0 : unlink(tmppath);
3301 :
3302 0 : close(fd);
3303 :
3304 0 : errno = save_errno;
3305 :
3306 0 : ereport(ERROR,
3307 : (errcode_for_file_access(),
3308 : errmsg("could not write to file \"%s\": %m", tmppath)));
3309 : }
3310 :
3311 670 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_SYNC);
3312 670 : if (pg_fsync(fd) != 0)
3313 : {
3314 0 : int save_errno = errno;
3315 :
3316 0 : close(fd);
3317 0 : errno = save_errno;
3318 0 : ereport(ERROR,
3319 : (errcode_for_file_access(),
3320 : errmsg("could not fsync file \"%s\": %m", tmppath)));
3321 : }
3322 670 : pgstat_report_wait_end();
3323 :
3324 670 : if (close(fd) != 0)
3325 0 : ereport(ERROR,
3326 : (errcode_for_file_access(),
3327 : errmsg("could not close file \"%s\": %m", tmppath)));
3328 :
3329 : /*
3330 : * Now move the segment into place with its final name.
3331 : *
3332 : * If caller didn't want to use a pre-existing file, get rid of any
3333 : * pre-existing file. Otherwise, cope with possibility that someone else
3334 : * has created the file while we were filling ours: if so, use ours to
3335 : * pre-create a future log segment.
3336 : */
3337 670 : installed_segno = logsegno;
3338 :
3339 : /*
3340 : * XXX: What should we use as max_segno? We used to use XLOGfileslop when
3341 : * that was a constant, but that was always a bit dubious: normally, at a
3342 : * checkpoint, XLOGfileslop was the offset from the checkpoint record, but
3343 : * here, it was the offset from the insert location. We can't do the
3344 : * normal XLOGfileslop calculation here because we don't have access to
3345 : * the prior checkpoint's redo location. So somewhat arbitrarily, just use
3346 : * CheckPointSegments.
3347 : */
3348 670 : max_segno = logsegno + CheckPointSegments;
3349 1340 : if (!InstallXLogFileSegment(&installed_segno, tmppath,
3350 670 : *use_existent, max_segno,
3351 : use_lock))
3352 : {
3353 : /*
3354 : * No need for any more future segments, or InstallXLogFileSegment()
3355 : * failed to rename the file into place. If the rename failed, opening
3356 : * the file below will fail.
3357 : */
3358 0 : unlink(tmppath);
3359 : }
3360 :
3361 : /* Set flag to tell caller there was no existent file */
3362 670 : *use_existent = false;
3363 :
3364 : /* Now open original target segment (might not be file I just made) */
3365 670 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | get_sync_bit(sync_method));
3366 670 : if (fd < 0)
3367 0 : ereport(ERROR,
3368 : (errcode_for_file_access(),
3369 : errmsg("could not open file \"%s\": %m", path)));
3370 :
3371 670 : elog(DEBUG2, "done creating and filling new WAL file");
3372 :
3373 670 : return fd;
3374 : }
3375 :
3376 : /*
3377 : * Create a new XLOG file segment by copying a pre-existing one.
3378 : *
3379 : * destsegno: identify segment to be created.
3380 : *
3381 : * srcTLI, srcsegno: identify segment to be copied (could be from
3382 : * a different timeline)
3383 : *
3384 : * upto: how much of the source file to copy (the rest is filled with
3385 : * zeros)
3386 : *
3387 : * Currently this is only used during recovery, and so there are no locking
3388 : * considerations. But we should be just as tense as XLogFileInit to avoid
3389 : * emplacing a bogus file.
3390 : */
3391 : static void
3392 34 : XLogFileCopy(XLogSegNo destsegno, TimeLineID srcTLI, XLogSegNo srcsegno,
3393 : int upto)
3394 : {
3395 : char path[MAXPGPATH];
3396 : char tmppath[MAXPGPATH];
3397 : PGAlignedXLogBlock buffer;
3398 : int srcfd;
3399 : int fd;
3400 : int nbytes;
3401 :
3402 : /*
3403 : * Open the source file
3404 : */
3405 34 : XLogFilePath(path, srcTLI, srcsegno, wal_segment_size);
3406 34 : srcfd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
3407 34 : if (srcfd < 0)
3408 0 : ereport(ERROR,
3409 : (errcode_for_file_access(),
3410 : errmsg("could not open file \"%s\": %m", path)));
3411 :
3412 : /*
3413 : * Copy into a temp file name.
3414 : */
3415 34 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3416 :
3417 34 : unlink(tmppath);
3418 :
3419 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3420 34 : fd = OpenTransientFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
3421 34 : if (fd < 0)
3422 0 : ereport(ERROR,
3423 : (errcode_for_file_access(),
3424 : errmsg("could not create file \"%s\": %m", tmppath)));
3425 :
3426 : /*
3427 : * Do the data copying.
3428 : */
3429 69666 : for (nbytes = 0; nbytes < wal_segment_size; nbytes += sizeof(buffer))
3430 : {
3431 : int nread;
3432 :
3433 69632 : nread = upto - nbytes;
3434 :
3435 : /*
3436 : * The part that is not read from the source file is filled with
3437 : * zeros.
3438 : */
3439 69632 : if (nread < sizeof(buffer))
3440 34 : memset(buffer.data, 0, sizeof(buffer));
3441 :
3442 69632 : if (nread > 0)
3443 : {
3444 : int r;
3445 :
3446 1566 : if (nread > sizeof(buffer))
3447 1532 : nread = sizeof(buffer);
3448 1566 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_READ);
3449 1566 : r = read(srcfd, buffer.data, nread);
3450 1566 : if (r != nread)
3451 : {
3452 0 : if (r < 0)
3453 0 : ereport(ERROR,
3454 : (errcode_for_file_access(),
3455 : errmsg("could not read file \"%s\": %m",
3456 : path)));
3457 : else
3458 0 : ereport(ERROR,
3459 : (errcode(ERRCODE_DATA_CORRUPTED),
3460 : errmsg("could not read file \"%s\": read %d of %zu",
3461 : path, r, (Size) nread)));
3462 : }
3463 1566 : pgstat_report_wait_end();
3464 : }
3465 69632 : errno = 0;
3466 69632 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_WRITE);
3467 69632 : if ((int) write(fd, buffer.data, sizeof(buffer)) != (int) sizeof(buffer))
3468 : {
3469 0 : int save_errno = errno;
3470 :
3471 : /*
3472 : * If we fail to make the file, delete it to release disk space
3473 : */
3474 0 : unlink(tmppath);
3475 : /* if write didn't set errno, assume problem is no disk space */
3476 0 : errno = save_errno ? save_errno : ENOSPC;
3477 :
3478 0 : ereport(ERROR,
3479 : (errcode_for_file_access(),
3480 : errmsg("could not write to file \"%s\": %m", tmppath)));
3481 : }
3482 69632 : pgstat_report_wait_end();
3483 : }
3484 :
3485 34 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_SYNC);
3486 34 : if (pg_fsync(fd) != 0)
3487 0 : ereport(data_sync_elevel(ERROR),
3488 : (errcode_for_file_access(),
3489 : errmsg("could not fsync file \"%s\": %m", tmppath)));
3490 34 : pgstat_report_wait_end();
3491 :
3492 34 : if (CloseTransientFile(fd) != 0)
3493 0 : ereport(ERROR,
3494 : (errcode_for_file_access(),
3495 : errmsg("could not close file \"%s\": %m", tmppath)));
3496 :
3497 34 : if (CloseTransientFile(srcfd) != 0)
3498 0 : ereport(ERROR,
3499 : (errcode_for_file_access(),
3500 : errmsg("could not close file \"%s\": %m", path)));
3501 :
3502 : /*
3503 : * Now move the segment into place with its final name.
3504 : */
3505 34 : if (!InstallXLogFileSegment(&destsegno, tmppath, false, 0, false))
3506 0 : elog(ERROR, "InstallXLogFileSegment should not have failed");
3507 34 : }
3508 :
3509 : /*
3510 : * Install a new XLOG segment file as a current or future log segment.
3511 : *
3512 : * This is used both to install a newly-created segment (which has a temp
3513 : * filename while it's being created) and to recycle an old segment.
3514 : *
3515 : * *segno: identify segment to install as (or first possible target).
3516 : * When find_free is true, this is modified on return to indicate the
3517 : * actual installation location or last segment searched.
3518 : *
3519 : * tmppath: initial name of file to install. It will be renamed into place.
3520 : *
3521 : * find_free: if true, install the new segment at the first empty segno
3522 : * number at or after the passed numbers. If false, install the new segment
3523 : * exactly where specified, deleting any existing segment file there.
3524 : *
3525 : * max_segno: maximum segment number to install the new file as. Fail if no
3526 : * free slot is found between *segno and max_segno. (Ignored when find_free
3527 : * is false.)
3528 : *
3529 : * use_lock: if true, acquire ControlFileLock while moving file into
3530 : * place. This should be true except during bootstrap log creation. The
3531 : * caller must *not* hold the lock at call.
3532 : *
3533 : * Returns true if the file was installed successfully. false indicates that
3534 : * max_segno limit was exceeded, or an error occurred while renaming the
3535 : * file into place.
3536 : */
3537 : static bool
3538 1386 : InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
3539 : bool find_free, XLogSegNo max_segno,
3540 : bool use_lock)
3541 : {
3542 : char path[MAXPGPATH];
3543 : struct stat stat_buf;
3544 :
3545 1386 : XLogFilePath(path, ThisTimeLineID, *segno, wal_segment_size);
3546 :
3547 : /*
3548 : * We want to be sure that only one process does this at a time.
3549 : */
3550 1386 : if (use_lock)
3551 1030 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
3552 :
3553 1386 : if (!find_free)
3554 : {
3555 : /* Force installation: get rid of any pre-existing segment file */
3556 356 : durable_unlink(path, DEBUG1);
3557 : }
3558 : else
3559 : {
3560 : /* Find a free slot to put it in */
3561 14250 : while (stat(path, &stat_buf) == 0)
3562 : {
3563 12200 : if ((*segno) >= max_segno)
3564 : {
3565 : /* Failed to find a free slot within specified range */
3566 10 : if (use_lock)
3567 10 : LWLockRelease(ControlFileLock);
3568 10 : return false;
3569 : }
3570 12190 : (*segno)++;
3571 12190 : XLogFilePath(path, ThisTimeLineID, *segno, wal_segment_size);
3572 : }
3573 : }
3574 :
3575 : /*
3576 : * Perform the rename using link if available, paranoidly trying to avoid
3577 : * overwriting an existing file (there shouldn't be one).
3578 : */
3579 1376 : if (durable_link_or_rename(tmppath, path, LOG) != 0)
3580 : {
3581 0 : if (use_lock)
3582 0 : LWLockRelease(ControlFileLock);
3583 : /* durable_link_or_rename already emitted log message */
3584 0 : return false;
3585 : }
3586 :
3587 1376 : if (use_lock)
3588 1020 : LWLockRelease(ControlFileLock);
3589 :
3590 1376 : return true;
3591 : }
3592 :
3593 : /*
3594 : * Open a pre-existing logfile segment for writing.
3595 : */
3596 : int
3597 8 : XLogFileOpen(XLogSegNo segno)
3598 : {
3599 : char path[MAXPGPATH];
3600 : int fd;
3601 :
3602 8 : XLogFilePath(path, ThisTimeLineID, segno, wal_segment_size);
3603 :
3604 8 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | get_sync_bit(sync_method));
3605 8 : if (fd < 0)
3606 0 : ereport(PANIC,
3607 : (errcode_for_file_access(),
3608 : errmsg("could not open file \"%s\": %m", path)));
3609 :
3610 8 : return fd;
3611 : }
3612 :
3613 : /*
3614 : * Open a logfile segment for reading (during recovery).
3615 : *
3616 : * If source == XLOG_FROM_ARCHIVE, the segment is retrieved from archive.
3617 : * Otherwise, it's assumed to be already available in pg_wal.
3618 : */
3619 : static int
3620 1988 : XLogFileRead(XLogSegNo segno, int emode, TimeLineID tli,
3621 : int source, bool notfoundOk)
3622 : {
3623 : char xlogfname[MAXFNAMELEN];
3624 : char activitymsg[MAXFNAMELEN + 16];
3625 : char path[MAXPGPATH];
3626 : int fd;
3627 :
3628 1988 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
3629 :
3630 1988 : switch (source)
3631 : {
3632 : case XLOG_FROM_ARCHIVE:
3633 : /* Report recovery progress in PS display */
3634 342 : snprintf(activitymsg, sizeof(activitymsg), "waiting for %s",
3635 : xlogfname);
3636 342 : set_ps_display(activitymsg, false);
3637 :
3638 342 : restoredFromArchive = RestoreArchivedFile(path, xlogfname,
3639 : "RECOVERYXLOG",
3640 : wal_segment_size,
3641 : InRedo);
3642 342 : if (!restoredFromArchive)
3643 328 : return -1;
3644 14 : break;
3645 :
3646 : case XLOG_FROM_PG_WAL:
3647 : case XLOG_FROM_STREAM:
3648 1646 : XLogFilePath(path, tli, segno, wal_segment_size);
3649 1646 : restoredFromArchive = false;
3650 1646 : break;
3651 :
3652 : default:
3653 0 : elog(ERROR, "invalid XLogFileRead source %d", source);
3654 : }
3655 :
3656 : /*
3657 : * If the segment was fetched from archival storage, replace the existing
3658 : * xlog segment (if any) with the archival version.
3659 : */
3660 1660 : if (source == XLOG_FROM_ARCHIVE)
3661 : {
3662 14 : KeepFileRestoredFromArchive(path, xlogfname);
3663 :
3664 : /*
3665 : * Set path to point at the new file in pg_wal.
3666 : */
3667 14 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlogfname);
3668 : }
3669 :
3670 1660 : fd = BasicOpenFile(path, O_RDONLY | PG_BINARY);
3671 1660 : if (fd >= 0)
3672 : {
3673 : /* Success! */
3674 1542 : curFileTLI = tli;
3675 :
3676 : /* Report recovery progress in PS display */
3677 1542 : snprintf(activitymsg, sizeof(activitymsg), "recovering %s",
3678 : xlogfname);
3679 1542 : set_ps_display(activitymsg, false);
3680 :
3681 : /* Track source of data in assorted state variables */
3682 1542 : readSource = source;
3683 1542 : XLogReceiptSource = source;
3684 : /* In FROM_STREAM case, caller tracks receipt time, not me */
3685 1542 : if (source != XLOG_FROM_STREAM)
3686 1438 : XLogReceiptTime = GetCurrentTimestamp();
3687 :
3688 1542 : return fd;
3689 : }
3690 118 : if (errno != ENOENT || !notfoundOk) /* unexpected failure? */
3691 0 : ereport(PANIC,
3692 : (errcode_for_file_access(),
3693 : errmsg("could not open file \"%s\": %m", path)));
3694 118 : return -1;
3695 : }
3696 :
3697 : /*
3698 : * Open a logfile segment for reading (during recovery).
3699 : *
3700 : * This version searches for the segment with any TLI listed in expectedTLEs.
3701 : */
3702 : static int
3703 1536 : XLogFileReadAnyTLI(XLogSegNo segno, int emode, int source)
3704 : {
3705 : char path[MAXPGPATH];
3706 : ListCell *cell;
3707 : int fd;
3708 : List *tles;
3709 :
3710 : /*
3711 : * Loop looking for a suitable timeline ID: we might need to read any of
3712 : * the timelines listed in expectedTLEs.
3713 : *
3714 : * We expect curFileTLI on entry to be the TLI of the preceding file in
3715 : * sequence, or 0 if there was no predecessor. We do not allow curFileTLI
3716 : * to go backwards; this prevents us from picking up the wrong file when a
3717 : * parent timeline extends to higher segment numbers than the child we
3718 : * want to read.
3719 : *
3720 : * If we haven't read the timeline history file yet, read it now, so that
3721 : * we know which TLIs to scan. We don't save the list in expectedTLEs,
3722 : * however, unless we actually find a valid segment. That way if there is
3723 : * neither a timeline history file nor a WAL segment in the archive, and
3724 : * streaming replication is set up, we'll read the timeline history file
3725 : * streamed from the master when we start streaming, instead of recovering
3726 : * with a dummy history generated here.
3727 : */
3728 1536 : if (expectedTLEs)
3729 344 : tles = expectedTLEs;
3730 : else
3731 1192 : tles = readTimeLineHistory(recoveryTargetTLI);
3732 :
3733 1654 : foreach(cell, tles)
3734 : {
3735 1556 : TimeLineID tli = ((TimeLineHistoryEntry *) lfirst(cell))->tli;
3736 :
3737 1556 : if (tli < curFileTLI)
3738 0 : break; /* don't bother looking at too-old TLIs */
3739 :
3740 1556 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_ARCHIVE)
3741 : {
3742 342 : fd = XLogFileRead(segno, emode, tli,
3743 : XLOG_FROM_ARCHIVE, true);
3744 342 : if (fd != -1)
3745 : {
3746 14 : elog(DEBUG1, "got WAL segment from archive");
3747 14 : if (!expectedTLEs)
3748 6 : expectedTLEs = tles;
3749 1452 : return fd;
3750 : }
3751 : }
3752 :
3753 1542 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_PG_WAL)
3754 : {
3755 1542 : fd = XLogFileRead(segno, emode, tli,
3756 : XLOG_FROM_PG_WAL, true);
3757 1542 : if (fd != -1)
3758 : {
3759 1424 : if (!expectedTLEs)
3760 1186 : expectedTLEs = tles;
3761 1424 : return fd;
3762 : }
3763 : }
3764 : }
3765 :
3766 : /* Couldn't find it. For simplicity, complain about front timeline */
3767 98 : XLogFilePath(path, recoveryTargetTLI, segno, wal_segment_size);
3768 98 : errno = ENOENT;
3769 100 : ereport(emode,
3770 : (errcode_for_file_access(),
3771 : errmsg("could not open file \"%s\": %m", path)));
3772 98 : return -1;
3773 : }
3774 :
3775 : /*
3776 : * Close the current logfile segment for writing.
3777 : */
3778 : static void
3779 2028 : XLogFileClose(void)
3780 : {
3781 : Assert(openLogFile >= 0);
3782 :
3783 : /*
3784 : * WAL segment files will not be re-read in normal operation, so we advise
3785 : * the OS to release any cached pages. But do not do so if WAL archiving
3786 : * or streaming is active, because archiver and walsender process could
3787 : * use the cache to read the WAL segment.
3788 : */
3789 : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
3790 2028 : if (!XLogIsNeeded())
3791 36 : (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
3792 : #endif
3793 :
3794 2028 : if (close(openLogFile) != 0)
3795 0 : ereport(PANIC,
3796 : (errcode_for_file_access(),
3797 : errmsg("could not close file \"%s\": %m",
3798 : XLogFileNameP(ThisTimeLineID, openLogSegNo))));
3799 2028 : openLogFile = -1;
3800 2028 : }
3801 :
3802 : /*
3803 : * Preallocate log files beyond the specified log endpoint.
3804 : *
3805 : * XXX this is currently extremely conservative, since it forces only one
3806 : * future log segment to exist, and even that only if we are 75% done with
3807 : * the current one. This is only appropriate for very low-WAL-volume systems.
3808 : * High-volume systems will be OK once they've built up a sufficient set of
3809 : * recycled log segments, but the startup transient is likely to include
3810 : * a lot of segment creations by foreground processes, which is not so good.
3811 : */
3812 : static void
3813 2952 : PreallocXlogFiles(XLogRecPtr endptr)
3814 : {
3815 : XLogSegNo _logSegNo;
3816 : int lf;
3817 : bool use_existent;
3818 : uint64 offset;
3819 :
3820 2952 : XLByteToPrevSeg(endptr, _logSegNo, wal_segment_size);
3821 2952 : offset = XLogSegmentOffset(endptr - 1, wal_segment_size);
3822 2952 : if (offset >= (uint32) (0.75 * wal_segment_size))
3823 : {
3824 54 : _logSegNo++;
3825 54 : use_existent = true;
3826 54 : lf = XLogFileInit(_logSegNo, &use_existent, true);
3827 54 : close(lf);
3828 54 : if (!use_existent)
3829 2 : CheckpointStats.ckpt_segs_added++;
3830 : }
3831 2952 : }
3832 :
3833 : /*
3834 : * Throws an error if the given log segment has already been removed or
3835 : * recycled. The caller should only pass a segment that it knows to have
3836 : * existed while the server has been running, as this function always
3837 : * succeeds if no WAL segments have been removed since startup.
3838 : * 'tli' is only used in the error message.
3839 : *
3840 : * Note: this function guarantees to keep errno unchanged on return.
3841 : * This supports callers that use this to possibly deliver a better
3842 : * error message about a missing file, while still being able to throw
3843 : * a normal file-access error afterwards, if this does return.
3844 : */
3845 : void
3846 5924 : CheckXLogRemoved(XLogSegNo segno, TimeLineID tli)
3847 : {
3848 5924 : int save_errno = errno;
3849 : XLogSegNo lastRemovedSegNo;
3850 :
3851 5924 : SpinLockAcquire(&XLogCtl->info_lck);
3852 5924 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3853 5924 : SpinLockRelease(&XLogCtl->info_lck);
3854 :
3855 5924 : if (segno <= lastRemovedSegNo)
3856 : {
3857 : char filename[MAXFNAMELEN];
3858 :
3859 0 : XLogFileName(filename, tli, segno, wal_segment_size);
3860 0 : errno = save_errno;
3861 0 : ereport(ERROR,
3862 : (errcode_for_file_access(),
3863 : errmsg("requested WAL segment %s has already been removed",
3864 : filename)));
3865 : }
3866 5924 : errno = save_errno;
3867 5924 : }
3868 :
3869 : /*
3870 : * Return the last WAL segment removed, or 0 if no segment has been removed
3871 : * since startup.
3872 : *
3873 : * NB: the result can be out of date arbitrarily fast, the caller has to deal
3874 : * with that.
3875 : */
3876 : XLogSegNo
3877 326 : XLogGetLastRemovedSegno(void)
3878 : {
3879 : XLogSegNo lastRemovedSegNo;
3880 :
3881 326 : SpinLockAcquire(&XLogCtl->info_lck);
3882 326 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3883 326 : SpinLockRelease(&XLogCtl->info_lck);
3884 :
3885 326 : return lastRemovedSegNo;
3886 : }
3887 :
3888 : /*
3889 : * Update the last removed segno pointer in shared memory, to reflect
3890 : * that the given XLOG file has been removed.
3891 : */
3892 : static void
3893 660 : UpdateLastRemovedPtr(char *filename)
3894 : {
3895 : uint32 tli;
3896 : XLogSegNo segno;
3897 :
3898 660 : XLogFromFileName(filename, &tli, &segno, wal_segment_size);
3899 :
3900 660 : SpinLockAcquire(&XLogCtl->info_lck);
3901 660 : if (segno > XLogCtl->lastRemovedSegNo)
3902 164 : XLogCtl->lastRemovedSegNo = segno;
3903 660 : SpinLockRelease(&XLogCtl->info_lck);
3904 660 : }
3905 :
3906 : /*
3907 : * Remove all temporary log files in pg_wal
3908 : *
3909 : * This is called at the beginning of recovery after a previous crash,
3910 : * at a point where no other processes write fresh WAL data.
3911 : */
3912 : static void
3913 106 : RemoveTempXlogFiles(void)
3914 : {
3915 : DIR *xldir;
3916 : struct dirent *xlde;
3917 :
3918 106 : elog(DEBUG2, "removing all temporary WAL segments");
3919 :
3920 106 : xldir = AllocateDir(XLOGDIR);
3921 706 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3922 : {
3923 : char path[MAXPGPATH];
3924 :
3925 494 : if (strncmp(xlde->d_name, "xlogtemp.", 9) != 0)
3926 494 : continue;
3927 :
3928 0 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
3929 0 : unlink(path);
3930 0 : elog(DEBUG2, "removed temporary WAL segment \"%s\"", path);
3931 : }
3932 106 : FreeDir(xldir);
3933 106 : }
3934 :
3935 : /*
3936 : * Recycle or remove all log files older or equal to passed segno.
3937 : *
3938 : * endptr is current (or recent) end of xlog, and RedoRecPtr is the
3939 : * redo pointer of the last checkpoint. These are used to determine
3940 : * whether we want to recycle rather than delete no-longer-wanted log files.
3941 : */
3942 : static void
3943 2832 : RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr RedoRecPtr, XLogRecPtr endptr)
3944 : {
3945 : DIR *xldir;
3946 : struct dirent *xlde;
3947 : char lastoff[MAXFNAMELEN];
3948 :
3949 : /*
3950 : * Construct a filename of the last segment to be kept. The timeline ID
3951 : * doesn't matter, we ignore that in the comparison. (During recovery,
3952 : * ThisTimeLineID isn't set, so we can't use that.)
3953 : */
3954 2832 : XLogFileName(lastoff, 0, segno, wal_segment_size);
3955 :
3956 2832 : elog(DEBUG2, "attempting to remove WAL segments older than log file %s",
3957 : lastoff);
3958 :
3959 2832 : xldir = AllocateDir(XLOGDIR);
3960 :
3961 18748 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3962 : {
3963 : /* Ignore files that are not XLOG segments */
3964 21656 : if (!IsXLogFileName(xlde->d_name) &&
3965 8572 : !IsPartialXLogFileName(xlde->d_name))
3966 8572 : continue;
3967 :
3968 : /*
3969 : * We ignore the timeline part of the XLOG segment identifiers in
3970 : * deciding whether a segment is still needed. This ensures that we
3971 : * won't prematurely remove a segment from a parent timeline. We could
3972 : * probably be a little more proactive about removing segments of
3973 : * non-parent timelines, but that would be a whole lot more
3974 : * complicated.
3975 : *
3976 : * We use the alphanumeric sorting property of the filenames to decide
3977 : * which ones are earlier than the lastoff segment.
3978 : */
3979 4512 : if (strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
3980 : {
3981 660 : if (XLogArchiveCheckDone(xlde->d_name))
3982 : {
3983 : /* Update the last removed location in shared memory first */
3984 660 : UpdateLastRemovedPtr(xlde->d_name);
3985 :
3986 660 : RemoveXlogFile(xlde->d_name, RedoRecPtr, endptr);
3987 : }
3988 : }
3989 : }
3990 :
3991 2832 : FreeDir(xldir);
3992 2832 : }
3993 :
3994 : /*
3995 : * Remove WAL files that are not part of the given timeline's history.
3996 : *
3997 : * This is called during recovery, whenever we switch to follow a new
3998 : * timeline, and at the end of recovery when we create a new timeline. We
3999 : * wouldn't otherwise care about extra WAL files lying in pg_wal, but they
4000 : * might be leftover pre-allocated or recycled WAL segments on the old timeline
4001 : * that we haven't used yet, and contain garbage. If we just leave them in
4002 : * pg_wal, they will eventually be archived, and we can't let that happen.
4003 : * Files that belong to our timeline history are valid, because we have
4004 : * successfully replayed them, but from others we can't be sure.
4005 : *
4006 : * 'switchpoint' is the current point in WAL where we switch to new timeline,
4007 : * and 'newTLI' is the new timeline we switch to.
4008 : */
4009 : static void
4010 56 : RemoveNonParentXlogFiles(XLogRecPtr switchpoint, TimeLineID newTLI)
4011 : {
4012 : DIR *xldir;
4013 : struct dirent *xlde;
4014 : char switchseg[MAXFNAMELEN];
4015 : XLogSegNo endLogSegNo;
4016 :
4017 56 : XLByteToPrevSeg(switchpoint, endLogSegNo, wal_segment_size);
4018 :
4019 : /*
4020 : * Construct a filename of the last segment to be kept.
4021 : */
4022 56 : XLogFileName(switchseg, newTLI, endLogSegNo, wal_segment_size);
4023 :
4024 56 : elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
4025 : switchseg);
4026 :
4027 56 : xldir = AllocateDir(XLOGDIR);
4028 :
4029 532 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4030 : {
4031 : /* Ignore files that are not XLOG segments */
4032 420 : if (!IsXLogFileName(xlde->d_name))
4033 242 : continue;
4034 :
4035 : /*
4036 : * Remove files that are on a timeline older than the new one we're
4037 : * switching to, but with a segment number >= the first segment on the
4038 : * new timeline.
4039 : */
4040 296 : if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
4041 118 : strcmp(xlde->d_name + 8, switchseg + 8) > 0)
4042 : {
4043 : /*
4044 : * If the file has already been marked as .ready, however, don't
4045 : * remove it yet. It should be OK to remove it - files that are
4046 : * not part of our timeline history are not required for recovery
4047 : * - but seems safer to let them be archived and removed later.
4048 : */
4049 22 : if (!XLogArchiveIsReady(xlde->d_name))
4050 22 : RemoveXlogFile(xlde->d_name, InvalidXLogRecPtr, switchpoint);
4051 : }
4052 : }
4053 :
4054 56 : FreeDir(xldir);
4055 56 : }
4056 :
4057 : /*
4058 : * Recycle or remove a log file that's no longer needed.
4059 : *
4060 : * endptr is current (or recent) end of xlog, and RedoRecPtr is the
4061 : * redo pointer of the last checkpoint. These are used to determine
4062 : * whether we want to recycle rather than delete no-longer-wanted log files.
4063 : * If RedoRecPtr is not known, pass invalid, and the function will recycle,
4064 : * somewhat arbitrarily, 10 future segments.
4065 : */
4066 : static void
4067 682 : RemoveXlogFile(const char *segname, XLogRecPtr RedoRecPtr, XLogRecPtr endptr)
4068 : {
4069 : char path[MAXPGPATH];
4070 : #ifdef WIN32
4071 : char newpath[MAXPGPATH];
4072 : #endif
4073 : struct stat statbuf;
4074 : XLogSegNo endlogSegNo;
4075 : XLogSegNo recycleSegNo;
4076 :
4077 682 : if (wal_recycle)
4078 : {
4079 : /*
4080 : * Initialize info about where to try to recycle to.
4081 : */
4082 682 : XLByteToSeg(endptr, endlogSegNo, wal_segment_size);
4083 682 : if (RedoRecPtr == InvalidXLogRecPtr)
4084 22 : recycleSegNo = endlogSegNo + 10;
4085 : else
4086 660 : recycleSegNo = XLOGfileslop(RedoRecPtr);
4087 : }
4088 : else
4089 0 : recycleSegNo = 0; /* keep compiler quiet */
4090 :
4091 682 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", segname);
4092 :
4093 : /*
4094 : * Before deleting the file, see if it can be recycled as a future log
4095 : * segment. Only recycle normal files, pg_standby for example can create
4096 : * symbolic links pointing to a separate archive directory.
4097 : */
4098 1364 : if (wal_recycle &&
4099 1364 : endlogSegNo <= recycleSegNo &&
4100 2046 : lstat(path, &statbuf) == 0 && S_ISREG(statbuf.st_mode) &&
4101 682 : InstallXLogFileSegment(&endlogSegNo, path,
4102 : true, recycleSegNo, true))
4103 : {
4104 672 : ereport(DEBUG2,
4105 : (errmsg("recycled write-ahead log file \"%s\"",
4106 : segname)));
4107 672 : CheckpointStats.ckpt_segs_recycled++;
4108 : /* Needn't recheck that slot on future iterations */
4109 672 : endlogSegNo++;
4110 : }
4111 : else
4112 : {
4113 : /* No need for any more future segments... */
4114 : int rc;
4115 :
4116 10 : ereport(DEBUG2,
4117 : (errmsg("removing write-ahead log file \"%s\"",
4118 : segname)));
4119 :
4120 : #ifdef WIN32
4121 :
4122 : /*
4123 : * On Windows, if another process (e.g another backend) holds the file
4124 : * open in FILE_SHARE_DELETE mode, unlink will succeed, but the file
4125 : * will still show up in directory listing until the last handle is
4126 : * closed. To avoid confusing the lingering deleted file for a live
4127 : * WAL file that needs to be archived, rename it before deleting it.
4128 : *
4129 : * If another process holds the file open without FILE_SHARE_DELETE
4130 : * flag, rename will fail. We'll try again at the next checkpoint.
4131 : */
4132 : snprintf(newpath, MAXPGPATH, "%s.deleted", path);
4133 : if (rename(path, newpath) != 0)
4134 : {
4135 : ereport(LOG,
4136 : (errcode_for_file_access(),
4137 : errmsg("could not rename file \"%s\": %m",
4138 : path)));
4139 : return;
4140 : }
4141 : rc = durable_unlink(newpath, LOG);
4142 : #else
4143 10 : rc = durable_unlink(path, LOG);
4144 : #endif
4145 10 : if (rc != 0)
4146 : {
4147 : /* Message already logged by durable_unlink() */
4148 0 : return;
4149 : }
4150 10 : CheckpointStats.ckpt_segs_removed++;
4151 : }
4152 :
4153 682 : XLogArchiveCleanup(segname);
4154 : }
4155 :
4156 : /*
4157 : * Verify whether pg_wal and pg_wal/archive_status exist.
4158 : * If the latter does not exist, recreate it.
4159 : *
4160 : * It is not the goal of this function to verify the contents of these
4161 : * directories, but to help in cases where someone has performed a cluster
4162 : * copy for PITR purposes but omitted pg_wal from the copy.
4163 : *
4164 : * We could also recreate pg_wal if it doesn't exist, but a deliberate
4165 : * policy decision was made not to. It is fairly common for pg_wal to be
4166 : * a symlink, and if that was the DBA's intent then automatically making a
4167 : * plain directory would result in degraded performance with no notice.
4168 : */
4169 : static void
4170 1192 : ValidateXLOGDirectoryStructure(void)
4171 : {
4172 : char path[MAXPGPATH];
4173 : struct stat stat_buf;
4174 :
4175 : /* Check for pg_wal; if it doesn't exist, error out */
4176 2384 : if (stat(XLOGDIR, &stat_buf) != 0 ||
4177 1192 : !S_ISDIR(stat_buf.st_mode))
4178 0 : ereport(FATAL,
4179 : (errmsg("required WAL directory \"%s\" does not exist",
4180 : XLOGDIR)));
4181 :
4182 : /* Check for archive_status */
4183 1192 : snprintf(path, MAXPGPATH, XLOGDIR "/archive_status");
4184 1192 : if (stat(path, &stat_buf) == 0)
4185 : {
4186 : /* Check for weird cases where it exists but isn't a directory */
4187 1192 : if (!S_ISDIR(stat_buf.st_mode))
4188 0 : ereport(FATAL,
4189 : (errmsg("required WAL directory \"%s\" does not exist",
4190 : path)));
4191 : }
4192 : else
4193 : {
4194 0 : ereport(LOG,
4195 : (errmsg("creating missing WAL directory \"%s\"", path)));
4196 0 : if (MakePGDirectory(path) < 0)
4197 0 : ereport(FATAL,
4198 : (errmsg("could not create missing directory \"%s\": %m",
4199 : path)));
4200 : }
4201 1192 : }
4202 :
4203 : /*
4204 : * Remove previous backup history files. This also retries creation of
4205 : * .ready files for any backup history files for which XLogArchiveNotify
4206 : * failed earlier.
4207 : */
4208 : static void
4209 90 : CleanupBackupHistory(void)
4210 : {
4211 : DIR *xldir;
4212 : struct dirent *xlde;
4213 : char path[MAXPGPATH + sizeof(XLOGDIR)];
4214 :
4215 90 : xldir = AllocateDir(XLOGDIR);
4216 :
4217 802 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4218 : {
4219 622 : if (IsBackupHistoryFileName(xlde->d_name))
4220 : {
4221 90 : if (XLogArchiveCheckDone(xlde->d_name))
4222 : {
4223 84 : elog(DEBUG2, "removing WAL backup history file \"%s\"",
4224 : xlde->d_name);
4225 84 : snprintf(path, sizeof(path), XLOGDIR "/%s", xlde->d_name);
4226 84 : unlink(path);
4227 84 : XLogArchiveCleanup(xlde->d_name);
4228 : }
4229 : }
4230 : }
4231 :
4232 90 : FreeDir(xldir);
4233 90 : }
4234 :
4235 : /*
4236 : * Attempt to read an XLOG record.
4237 : *
4238 : * If RecPtr is valid, try to read a record at that position. Otherwise
4239 : * try to read a record just after the last one previously read.
4240 : *
4241 : * If no valid record is available, returns NULL, or fails if emode is PANIC.
4242 : * (emode must be either PANIC, LOG). In standby mode, retries until a valid
4243 : * record is available.
4244 : */
4245 : static XLogRecord *
4246 241050 : ReadRecord(XLogReaderState *xlogreader, XLogRecPtr RecPtr, int emode,
4247 : bool fetching_ckpt)
4248 : {
4249 : XLogRecord *record;
4250 241050 : XLogPageReadPrivate *private = (XLogPageReadPrivate *) xlogreader->private_data;
4251 :
4252 : /* Pass through parameters to XLogPageRead */
4253 241050 : private->fetching_ckpt = fetching_ckpt;
4254 241050 : private->emode = emode;
4255 241050 : private->randAccess = (RecPtr != InvalidXLogRecPtr);
4256 :
4257 : /* This is the first attempt to read this page. */
4258 241050 : lastSourceFailed = false;
4259 :
4260 : for (;;)
4261 90 : {
4262 : char *errormsg;
4263 :
4264 241140 : record = XLogReadRecord(xlogreader, RecPtr, &errormsg);
4265 241118 : ReadRecPtr = xlogreader->ReadRecPtr;
4266 241118 : EndRecPtr = xlogreader->EndRecPtr;
4267 241118 : if (record == NULL)
4268 : {
4269 184 : if (readFile >= 0)
4270 : {
4271 164 : close(readFile);
4272 164 : readFile = -1;
4273 : }
4274 :
4275 : /*
4276 : * We only end up here without a message when XLogPageRead()
4277 : * failed - in that case we already logged something. In
4278 : * StandbyMode that only happens if we have been triggered, so we
4279 : * shouldn't loop anymore in that case.
4280 : */
4281 184 : if (errormsg)
4282 164 : ereport(emode_for_corrupt_record(emode,
4283 : RecPtr ? RecPtr : EndRecPtr),
4284 : (errmsg_internal("%s", errormsg) /* already translated */ ));
4285 : }
4286 :
4287 : /*
4288 : * Check page TLI is one of the expected values.
4289 : */
4290 240934 : else if (!tliInHistory(xlogreader->latestPageTLI, expectedTLEs))
4291 : {
4292 : char fname[MAXFNAMELEN];
4293 : XLogSegNo segno;
4294 : int32 offset;
4295 :
4296 0 : XLByteToSeg(xlogreader->latestPagePtr, segno, wal_segment_size);
4297 0 : offset = XLogSegmentOffset(xlogreader->latestPagePtr,
4298 : wal_segment_size);
4299 0 : XLogFileName(fname, xlogreader->seg.ws_tli, segno,
4300 : wal_segment_size);
4301 0 : ereport(emode_for_corrupt_record(emode,
4302 : RecPtr ? RecPtr : EndRecPtr),
4303 : (errmsg("unexpected timeline ID %u in log segment %s, offset %u",
4304 : xlogreader->latestPageTLI,
4305 : fname,
4306 : offset)));
4307 0 : record = NULL;
4308 : }
4309 :
4310 241118 : if (record)
4311 : {
4312 : /* Great, got a record */
4313 481962 : return record;
4314 : }
4315 : else
4316 : {
4317 : /* No valid record available from this source */
4318 184 : lastSourceFailed = true;
4319 :
4320 : /*
4321 : * If archive recovery was requested, but we were still doing
4322 : * crash recovery, switch to archive recovery and retry using the
4323 : * offline archive. We have now replayed all the valid WAL in
4324 : * pg_wal, so we are presumably now consistent.
4325 : *
4326 : * We require that there's at least some valid WAL present in
4327 : * pg_wal, however (!fetching_ckpt). We could recover using the
4328 : * WAL from the archive, even if pg_wal is completely empty, but
4329 : * we'd have no idea how far we'd have to replay to reach
4330 : * consistency. So err on the safe side and give up.
4331 : */
4332 186 : if (!InArchiveRecovery && ArchiveRecoveryRequested &&
4333 2 : !fetching_ckpt)
4334 : {
4335 2 : ereport(DEBUG1,
4336 : (errmsg_internal("reached end of WAL in pg_wal, entering archive recovery")));
4337 2 : InArchiveRecovery = true;
4338 2 : if (StandbyModeRequested)
4339 2 : StandbyMode = true;
4340 :
4341 : /* initialize minRecoveryPoint to this record */
4342 2 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
4343 2 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
4344 2 : if (ControlFile->minRecoveryPoint < EndRecPtr)
4345 : {
4346 2 : ControlFile->minRecoveryPoint = EndRecPtr;
4347 2 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
4348 : }
4349 : /* update local copy */
4350 2 : minRecoveryPoint = ControlFile->minRecoveryPoint;
4351 2 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
4352 :
4353 : /*
4354 : * The startup process can update its local copy of
4355 : * minRecoveryPoint from this point.
4356 : */
4357 2 : updateMinRecoveryPoint = true;
4358 :
4359 2 : UpdateControlFile();
4360 2 : LWLockRelease(ControlFileLock);
4361 :
4362 2 : CheckRecoveryConsistency();
4363 :
4364 : /*
4365 : * Before we retry, reset lastSourceFailed and currentSource
4366 : * so that we will check the archive next.
4367 : */
4368 2 : lastSourceFailed = false;
4369 2 : currentSource = 0;
4370 :
4371 92 : continue;
4372 : }
4373 :
4374 : /* In standby mode, loop back to retry. Otherwise, give up. */
4375 182 : if (StandbyMode && !CheckForStandbyTrigger())
4376 88 : continue;
4377 : else
4378 94 : return NULL;
4379 : }
4380 : }
4381 : }
4382 :
4383 : /*
4384 : * Scan for new timelines that might have appeared in the archive since we
4385 : * started recovery.
4386 : *
4387 : * If there are any, the function changes recovery target TLI to the latest
4388 : * one and returns 'true'.
4389 : */
4390 : static bool
4391 150 : rescanLatestTimeLine(void)
4392 : {
4393 : List *newExpectedTLEs;
4394 : bool found;
4395 : ListCell *cell;
4396 : TimeLineID newtarget;
4397 150 : TimeLineID oldtarget = recoveryTargetTLI;
4398 150 : TimeLineHistoryEntry *currentTle = NULL;
4399 :
4400 150 : newtarget = findNewestTimeLine(recoveryTargetTLI);
4401 150 : if (newtarget == recoveryTargetTLI)
4402 : {
4403 : /* No new timelines found */
4404 144 : return false;
4405 : }
4406 :
4407 : /*
4408 : * Determine the list of expected TLIs for the new TLI
4409 : */
4410 :
4411 6 : newExpectedTLEs = readTimeLineHistory(newtarget);
4412 :
4413 : /*
4414 : * If the current timeline is not part of the history of the new timeline,
4415 : * we cannot proceed to it.
4416 : */
4417 6 : found = false;
4418 12 : foreach(cell, newExpectedTLEs)
4419 : {
4420 12 : currentTle = (TimeLineHistoryEntry *) lfirst(cell);
4421 :
4422 12 : if (currentTle->tli == recoveryTargetTLI)
4423 : {
4424 6 : found = true;
4425 6 : break;
4426 : }
4427 : }
4428 6 : if (!found)
4429 : {
4430 0 : ereport(LOG,
4431 : (errmsg("new timeline %u is not a child of database system timeline %u",
4432 : newtarget,
4433 : ThisTimeLineID)));
4434 0 : return false;
4435 : }
4436 :
4437 : /*
4438 : * The current timeline was found in the history file, but check that the
4439 : * next timeline was forked off from it *after* the current recovery
4440 : * location.
4441 : */
4442 6 : if (currentTle->end < EndRecPtr)
4443 : {
4444 0 : ereport(LOG,
4445 : (errmsg("new timeline %u forked off current database system timeline %u before current recovery point %X/%X",
4446 : newtarget,
4447 : ThisTimeLineID,
4448 : (uint32) (EndRecPtr >> 32), (uint32) EndRecPtr)));
4449 0 : return false;
4450 : }
4451 :
4452 : /* The new timeline history seems valid. Switch target */
4453 6 : recoveryTargetTLI = newtarget;
4454 6 : list_free_deep(expectedTLEs);
4455 6 : expectedTLEs = newExpectedTLEs;
4456 :
4457 : /*
4458 : * As in StartupXLOG(), try to ensure we have all the history files
4459 : * between the old target and new target in pg_wal.
4460 : */
4461 6 : restoreTimeLineHistoryFiles(oldtarget + 1, newtarget);
4462 :
4463 6 : ereport(LOG,
4464 : (errmsg("new target timeline is %u",
4465 : recoveryTargetTLI)));
4466 :
4467 6 : return true;
4468 : }
4469 :
4470 : /*
4471 : * I/O routines for pg_control
4472 : *
4473 : * *ControlFile is a buffer in shared memory that holds an image of the
4474 : * contents of pg_control. WriteControlFile() initializes pg_control
4475 : * given a preloaded buffer, ReadControlFile() loads the buffer from
4476 : * the pg_control file (during postmaster or standalone-backend startup),
4477 : * and UpdateControlFile() rewrites pg_control after we modify xlog state.
4478 : *
4479 : * For simplicity, WriteControlFile() initializes the fields of pg_control
4480 : * that are related to checking backend/database compatibility, and
4481 : * ReadControlFile() verifies they are correct. We could split out the
4482 : * I/O and compatibility-check functions, but there seems no need currently.
4483 : */
4484 : static void
4485 322 : WriteControlFile(void)
4486 : {
4487 : int fd;
4488 : char buffer[PG_CONTROL_FILE_SIZE]; /* need not be aligned */
4489 :
4490 : /*
4491 : * Ensure that the size of the pg_control data structure is sane. See the
4492 : * comments for these symbols in pg_control.h.
4493 : */
4494 : StaticAssertStmt(sizeof(ControlFileData) <= PG_CONTROL_MAX_SAFE_SIZE,
4495 : "pg_control is too large for atomic disk writes");
4496 : StaticAssertStmt(sizeof(ControlFileData) <= PG_CONTROL_FILE_SIZE,
4497 : "sizeof(ControlFileData) exceeds PG_CONTROL_FILE_SIZE");
4498 :
4499 : /*
4500 : * Initialize version and compatibility-check fields
4501 : */
4502 322 : ControlFile->pg_control_version = PG_CONTROL_VERSION;
4503 322 : ControlFile->catalog_version_no = CATALOG_VERSION_NO;
4504 :
4505 322 : ControlFile->maxAlign = MAXIMUM_ALIGNOF;
4506 322 : ControlFile->floatFormat = FLOATFORMAT_VALUE;
4507 :
4508 322 : ControlFile->blcksz = BLCKSZ;
4509 322 : ControlFile->relseg_size = RELSEG_SIZE;
4510 322 : ControlFile->xlog_blcksz = XLOG_BLCKSZ;
4511 322 : ControlFile->xlog_seg_size = wal_segment_size;
4512 :
4513 322 : ControlFile->nameDataLen = NAMEDATALEN;
4514 322 : ControlFile->indexMaxKeys = INDEX_MAX_KEYS;
4515 :
4516 322 : ControlFile->toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
4517 322 : ControlFile->loblksize = LOBLKSIZE;
4518 :
4519 322 : ControlFile->float4ByVal = FLOAT4PASSBYVAL;
4520 322 : ControlFile->float8ByVal = FLOAT8PASSBYVAL;
4521 :
4522 : /* Contents are protected with a CRC */
4523 322 : INIT_CRC32C(ControlFile->crc);
4524 322 : COMP_CRC32C(ControlFile->crc,
4525 : (char *) ControlFile,
4526 : offsetof(ControlFileData, crc));
4527 322 : FIN_CRC32C(ControlFile->crc);
4528 :
4529 : /*
4530 : * We write out PG_CONTROL_FILE_SIZE bytes into pg_control, zero-padding
4531 : * the excess over sizeof(ControlFileData). This reduces the odds of
4532 : * premature-EOF errors when reading pg_control. We'll still fail when we
4533 : * check the contents of the file, but hopefully with a more specific
4534 : * error than "couldn't read pg_control".
4535 : */
4536 322 : memset(buffer, 0, PG_CONTROL_FILE_SIZE);
4537 322 : memcpy(buffer, ControlFile, sizeof(ControlFileData));
4538 :
4539 322 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4540 : O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
4541 322 : if (fd < 0)
4542 0 : ereport(PANIC,
4543 : (errcode_for_file_access(),
4544 : errmsg("could not create file \"%s\": %m",
4545 : XLOG_CONTROL_FILE)));
4546 :
4547 322 : errno = 0;
4548 322 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE);
4549 322 : if (write(fd, buffer, PG_CONTROL_FILE_SIZE) != PG_CONTROL_FILE_SIZE)
4550 : {
4551 : /* if write didn't set errno, assume problem is no disk space */
4552 0 : if (errno == 0)
4553 0 : errno = ENOSPC;
4554 0 : ereport(PANIC,
4555 : (errcode_for_file_access(),
4556 : errmsg("could not write to file \"%s\": %m",
4557 : XLOG_CONTROL_FILE)));
4558 : }
4559 322 : pgstat_report_wait_end();
4560 :
4561 322 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC);
4562 322 : if (pg_fsync(fd) != 0)
4563 0 : ereport(PANIC,
4564 : (errcode_for_file_access(),
4565 : errmsg("could not fsync file \"%s\": %m",
4566 : XLOG_CONTROL_FILE)));
4567 322 : pgstat_report_wait_end();
4568 :
4569 322 : if (close(fd) != 0)
4570 0 : ereport(PANIC,
4571 : (errcode_for_file_access(),
4572 : errmsg("could not close file \"%s\": %m",
4573 : XLOG_CONTROL_FILE)));
4574 322 : }
4575 :
4576 : static void
4577 1252 : ReadControlFile(void)
4578 : {
4579 : pg_crc32c crc;
4580 : int fd;
4581 : static char wal_segsz_str[20];
4582 : int r;
4583 :
4584 : /*
4585 : * Read data...
4586 : */
4587 1252 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4588 : O_RDWR | PG_BINARY);
4589 1252 : if (fd < 0)
4590 0 : ereport(PANIC,
4591 : (errcode_for_file_access(),
4592 : errmsg("could not open file \"%s\": %m",
4593 : XLOG_CONTROL_FILE)));
4594 :
4595 1252 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_READ);
4596 1252 : r = read(fd, ControlFile, sizeof(ControlFileData));
4597 1252 : if (r != sizeof(ControlFileData))
4598 : {
4599 0 : if (r < 0)
4600 0 : ereport(PANIC,
4601 : (errcode_for_file_access(),
4602 : errmsg("could not read file \"%s\": %m",
4603 : XLOG_CONTROL_FILE)));
4604 : else
4605 0 : ereport(PANIC,
4606 : (errcode(ERRCODE_DATA_CORRUPTED),
4607 : errmsg("could not read file \"%s\": read %d of %zu",
4608 : XLOG_CONTROL_FILE, r, sizeof(ControlFileData))));
4609 : }
4610 1252 : pgstat_report_wait_end();
4611 :
4612 1252 : close(fd);
4613 :
4614 : /*
4615 : * Check for expected pg_control format version. If this is wrong, the
4616 : * CRC check will likely fail because we'll be checking the wrong number
4617 : * of bytes. Complaining about wrong version will probably be more
4618 : * enlightening than complaining about wrong CRC.
4619 : */
4620 :
4621 1252 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION && ControlFile->pg_control_version % 65536 == 0 && ControlFile->pg_control_version / 65536 != 0)
4622 0 : ereport(FATAL,
4623 : (errmsg("database files are incompatible with server"),
4624 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
4625 : " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
4626 : ControlFile->pg_control_version, ControlFile->pg_control_version,
4627 : PG_CONTROL_VERSION, PG_CONTROL_VERSION),
4628 : errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
4629 :
4630 1252 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION)
4631 0 : ereport(FATAL,
4632 : (errmsg("database files are incompatible with server"),
4633 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
4634 : " but the server was compiled with PG_CONTROL_VERSION %d.",
4635 : ControlFile->pg_control_version, PG_CONTROL_VERSION),
4636 : errhint("It looks like you need to initdb.")));
4637 :
4638 : /* Now check the CRC. */
4639 1252 : INIT_CRC32C(crc);
4640 1252 : COMP_CRC32C(crc,
4641 : (char *) ControlFile,
4642 : offsetof(ControlFileData, crc));
4643 1252 : FIN_CRC32C(crc);
4644 :
4645 1252 : if (!EQ_CRC32C(crc, ControlFile->crc))
4646 0 : ereport(FATAL,
4647 : (errmsg("incorrect checksum in control file")));
4648 :
4649 : /*
4650 : * Do compatibility checking immediately. If the database isn't
4651 : * compatible with the backend executable, we want to abort before we can
4652 : * possibly do any damage.
4653 : */
4654 1252 : if (ControlFile->catalog_version_no != CATALOG_VERSION_NO)
4655 0 : ereport(FATAL,
4656 : (errmsg("database files are incompatible with server"),
4657 : errdetail("The database cluster was initialized with CATALOG_VERSION_NO %d,"
4658 : " but the server was compiled with CATALOG_VERSION_NO %d.",
4659 : ControlFile->catalog_version_no, CATALOG_VERSION_NO),
4660 : errhint("It looks like you need to initdb.")));
4661 1252 : if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
4662 0 : ereport(FATAL,
4663 : (errmsg("database files are incompatible with server"),
4664 : errdetail("The database cluster was initialized with MAXALIGN %d,"
4665 : " but the server was compiled with MAXALIGN %d.",
4666 : ControlFile->maxAlign, MAXIMUM_ALIGNOF),
4667 : errhint("It looks like you need to initdb.")));
4668 1252 : if (ControlFile->floatFormat != FLOATFORMAT_VALUE)
4669 0 : ereport(FATAL,
4670 : (errmsg("database files are incompatible with server"),
4671 : errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
4672 : errhint("It looks like you need to initdb.")));
4673 1252 : if (ControlFile->blcksz != BLCKSZ)
4674 0 : ereport(FATAL,
4675 : (errmsg("database files are incompatible with server"),
4676 : errdetail("The database cluster was initialized with BLCKSZ %d,"
4677 : " but the server was compiled with BLCKSZ %d.",
4678 : ControlFile->blcksz, BLCKSZ),
4679 : errhint("It looks like you need to recompile or initdb.")));
4680 1252 : if (ControlFile->relseg_size != RELSEG_SIZE)
4681 0 : ereport(FATAL,
4682 : (errmsg("database files are incompatible with server"),
4683 : errdetail("The database cluster was initialized with RELSEG_SIZE %d,"
4684 : " but the server was compiled with RELSEG_SIZE %d.",
4685 : ControlFile->relseg_size, RELSEG_SIZE),
4686 : errhint("It looks like you need to recompile or initdb.")));
4687 1252 : if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
4688 0 : ereport(FATAL,
4689 : (errmsg("database files are incompatible with server"),
4690 : errdetail("The database cluster was initialized with XLOG_BLCKSZ %d,"
4691 : " but the server was compiled with XLOG_BLCKSZ %d.",
4692 : ControlFile->xlog_blcksz, XLOG_BLCKSZ),
4693 : errhint("It looks like you need to recompile or initdb.")));
4694 1252 : if (ControlFile->nameDataLen != NAMEDATALEN)
4695 0 : ereport(FATAL,
4696 : (errmsg("database files are incompatible with server"),
4697 : errdetail("The database cluster was initialized with NAMEDATALEN %d,"
4698 : " but the server was compiled with NAMEDATALEN %d.",
4699 : ControlFile->nameDataLen, NAMEDATALEN),
4700 : errhint("It looks like you need to recompile or initdb.")));
4701 1252 : if (ControlFile->indexMaxKeys != INDEX_MAX_KEYS)
4702 0 : ereport(FATAL,
4703 : (errmsg("database files are incompatible with server"),
4704 : errdetail("The database cluster was initialized with INDEX_MAX_KEYS %d,"
4705 : " but the server was compiled with INDEX_MAX_KEYS %d.",
4706 : ControlFile->indexMaxKeys, INDEX_MAX_KEYS),
4707 : errhint("It looks like you need to recompile or initdb.")));
4708 1252 : if (ControlFile->toast_max_chunk_size != TOAST_MAX_CHUNK_SIZE)
4709 0 : ereport(FATAL,
4710 : (errmsg("database files are incompatible with server"),
4711 : errdetail("The database cluster was initialized with TOAST_MAX_CHUNK_SIZE %d,"
4712 : " but the server was compiled with TOAST_MAX_CHUNK_SIZE %d.",
4713 : ControlFile->toast_max_chunk_size, (int) TOAST_MAX_CHUNK_SIZE),
4714 : errhint("It looks like you need to recompile or initdb.")));
4715 1252 : if (ControlFile->loblksize != LOBLKSIZE)
4716 0 : ereport(FATAL,
4717 : (errmsg("database files are incompatible with server"),
4718 : errdetail("The database cluster was initialized with LOBLKSIZE %d,"
4719 : " but the server was compiled with LOBLKSIZE %d.",
4720 : ControlFile->loblksize, (int) LOBLKSIZE),
4721 : errhint("It looks like you need to recompile or initdb.")));
4722 :
4723 : #ifdef USE_FLOAT4_BYVAL
4724 1252 : if (ControlFile->float4ByVal != true)
4725 0 : ereport(FATAL,
4726 : (errmsg("database files are incompatible with server"),
4727 : errdetail("The database cluster was initialized without USE_FLOAT4_BYVAL"
4728 : " but the server was compiled with USE_FLOAT4_BYVAL."),
4729 : errhint("It looks like you need to recompile or initdb.")));
4730 : #else
4731 : if (ControlFile->float4ByVal != false)
4732 : ereport(FATAL,
4733 : (errmsg("database files are incompatible with server"),
4734 : errdetail("The database cluster was initialized with USE_FLOAT4_BYVAL"
4735 : " but the server was compiled without USE_FLOAT4_BYVAL."),
4736 : errhint("It looks like you need to recompile or initdb.")));
4737 : #endif
4738 :
4739 : #ifdef USE_FLOAT8_BYVAL
4740 1252 : if (ControlFile->float8ByVal != true)
4741 0 : ereport(FATAL,
4742 : (errmsg("database files are incompatible with server"),
4743 : errdetail("The database cluster was initialized without USE_FLOAT8_BYVAL"
4744 : " but the server was compiled with USE_FLOAT8_BYVAL."),
4745 : errhint("It looks like you need to recompile or initdb.")));
4746 : #else
4747 : if (ControlFile->float8ByVal != false)
4748 : ereport(FATAL,
4749 : (errmsg("database files are incompatible with server"),
4750 : errdetail("The database cluster was initialized with USE_FLOAT8_BYVAL"
4751 : " but the server was compiled without USE_FLOAT8_BYVAL."),
4752 : errhint("It looks like you need to recompile or initdb.")));
4753 : #endif
4754 :
4755 1252 : wal_segment_size = ControlFile->xlog_seg_size;
4756 :
4757 1252 : if (!IsValidWalSegSize(wal_segment_size))
4758 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4759 : errmsg_plural("WAL segment size must be a power of two between 1 MB and 1 GB, but the control file specifies %d byte",
4760 : "WAL segment size must be a power of two between 1 MB and 1 GB, but the control file specifies %d bytes",
4761 : wal_segment_size,
4762 : wal_segment_size)));
4763 :
4764 1252 : snprintf(wal_segsz_str, sizeof(wal_segsz_str), "%d", wal_segment_size);
4765 1252 : SetConfigOption("wal_segment_size", wal_segsz_str, PGC_INTERNAL,
4766 : PGC_S_OVERRIDE);
4767 :
4768 : /* check and update variables dependent on wal_segment_size */
4769 1252 : if (ConvertToXSegs(min_wal_size_mb, wal_segment_size) < 2)
4770 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4771 : errmsg("\"min_wal_size\" must be at least twice \"wal_segment_size\"")));
4772 :
4773 1252 : if (ConvertToXSegs(max_wal_size_mb, wal_segment_size) < 2)
4774 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4775 : errmsg("\"max_wal_size\" must be at least twice \"wal_segment_size\"")));
4776 :
4777 1252 : UsableBytesInSegment =
4778 1252 : (wal_segment_size / XLOG_BLCKSZ * UsableBytesInPage) -
4779 : (SizeOfXLogLongPHD - SizeOfXLogShortPHD);
4780 :
4781 1252 : CalculateCheckpointSegments();
4782 :
4783 : /* Make the initdb settings visible as GUC variables, too */
4784 1252 : SetConfigOption("data_checksums", DataChecksumsEnabled() ? "yes" : "no",
4785 : PGC_INTERNAL, PGC_S_OVERRIDE);
4786 1252 : }
4787 :
4788 : /*
4789 : * Utility wrapper to update the control file. Note that the control
4790 : * file gets flushed.
4791 : */
4792 : void
4793 5900 : UpdateControlFile(void)
4794 : {
4795 5900 : update_controlfile(DataDir, ControlFile, true);
4796 5900 : }
4797 :
4798 : /*
4799 : * Returns the unique system identifier from control file.
4800 : */
4801 : uint64
4802 450 : GetSystemIdentifier(void)
4803 : {
4804 : Assert(ControlFile != NULL);
4805 450 : return ControlFile->system_identifier;
4806 : }
4807 :
4808 : /*
4809 : * Returns the random nonce from control file.
4810 : */
4811 : char *
4812 2 : GetMockAuthenticationNonce(void)
4813 : {
4814 : Assert(ControlFile != NULL);
4815 2 : return ControlFile->mock_authentication_nonce;
4816 : }
4817 :
4818 : /*
4819 : * Are checksums enabled for data pages?
4820 : */
4821 : bool
4822 13209796 : DataChecksumsEnabled(void)
4823 : {
4824 : Assert(ControlFile != NULL);
4825 13209796 : return (ControlFile->data_checksum_version > 0);
4826 : }
4827 :
4828 : /*
4829 : * Returns a fake LSN for unlogged relations.
4830 : *
4831 : * Each call generates an LSN that is greater than any previous value
4832 : * returned. The current counter value is saved and restored across clean
4833 : * shutdowns, but like unlogged relations, does not survive a crash. This can
4834 : * be used in lieu of real LSN values returned by XLogInsert, if you need an
4835 : * LSN-like increasing sequence of numbers without writing any WAL.
4836 : */
4837 : XLogRecPtr
4838 0 : GetFakeLSNForUnloggedRel(void)
4839 : {
4840 : XLogRecPtr nextUnloggedLSN;
4841 :
4842 : /* increment the unloggedLSN counter, need SpinLock */
4843 0 : SpinLockAcquire(&XLogCtl->ulsn_lck);
4844 0 : nextUnloggedLSN = XLogCtl->unloggedLSN++;
4845 0 : SpinLockRelease(&XLogCtl->ulsn_lck);
4846 :
4847 0 : return nextUnloggedLSN;
4848 : }
4849 :
4850 : /*
4851 : * Auto-tune the number of XLOG buffers.
4852 : *
4853 : * The preferred setting for wal_buffers is about 3% of shared_buffers, with
4854 : * a maximum of one XLOG segment (there is little reason to think that more
4855 : * is helpful, at least so long as we force an fsync when switching log files)
4856 : * and a minimum of 8 blocks (which was the default value prior to PostgreSQL
4857 : * 9.1, when auto-tuning was added).
4858 : *
4859 : * This should not be called until NBuffers has received its final value.
4860 : */
4861 : static int
4862 1890 : XLOGChooseNumBuffers(void)
4863 : {
4864 : int xbuffers;
4865 :
4866 1890 : xbuffers = NBuffers / 32;
4867 1890 : if (xbuffers > (wal_segment_size / XLOG_BLCKSZ))
4868 20 : xbuffers = (wal_segment_size / XLOG_BLCKSZ);
4869 1890 : if (xbuffers < 8)
4870 302 : xbuffers = 8;
4871 1890 : return xbuffers;
4872 : }
4873 :
4874 : /*
4875 : * GUC check_hook for wal_buffers
4876 : */
4877 : bool
4878 3788 : check_wal_buffers(int *newval, void **extra, GucSource source)
4879 : {
4880 : /*
4881 : * -1 indicates a request for auto-tune.
4882 : */
4883 3788 : if (*newval == -1)
4884 : {
4885 : /*
4886 : * If we haven't yet changed the boot_val default of -1, just let it
4887 : * be. We'll fix it when XLOGShmemSize is called.
4888 : */
4889 1898 : if (XLOGbuffers == -1)
4890 1898 : return true;
4891 :
4892 : /* Otherwise, substitute the auto-tune value */
4893 0 : *newval = XLOGChooseNumBuffers();
4894 : }
4895 :
4896 : /*
4897 : * We clamp manually-set values to at least 4 blocks. Prior to PostgreSQL
4898 : * 9.1, a minimum of 4 was enforced by guc.c, but since that is no longer
4899 : * the case, we just silently treat such values as a request for the
4900 : * minimum. (We could throw an error instead, but that doesn't seem very
4901 : * helpful.)
4902 : */
4903 1890 : if (*newval < 4)
4904 0 : *newval = 4;
4905 :
4906 1890 : return true;
4907 : }
4908 :
4909 : /*
4910 : * Read the control file, set respective GUCs.
4911 : *
4912 : * This is to be called during startup, including a crash recovery cycle,
4913 : * unless in bootstrap mode, where no control file yet exists. As there's no
4914 : * usable shared memory yet (its sizing can depend on the contents of the
4915 : * control file!), first store the contents in local memory. XLOGShmemInit()
4916 : * will then copy it to shared memory later.
4917 : *
4918 : * reset just controls whether previous contents are to be expected (in the
4919 : * reset case, there's a dangling pointer into old shared memory), or not.
4920 : */
4921 : void
4922 930 : LocalProcessControlFile(bool reset)
4923 : {
4924 : Assert(reset || ControlFile == NULL);
4925 930 : ControlFile = palloc(sizeof(ControlFileData));
4926 930 : ReadControlFile();
4927 930 : }
4928 :
4929 : /*
4930 : * Initialization of shared memory for XLOG
4931 : */
4932 : Size
4933 3784 : XLOGShmemSize(void)
4934 : {
4935 : Size size;
4936 :
4937 : /*
4938 : * If the value of wal_buffers is -1, use the preferred auto-tune value.
4939 : * This isn't an amazingly clean place to do this, but we must wait till
4940 : * NBuffers has received its final value, and must do it before using the
4941 : * value of XLOGbuffers to do anything important.
4942 : */
4943 3784 : if (XLOGbuffers == -1)
4944 : {
4945 : char buf[32];
4946 :
4947 1890 : snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
4948 1890 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER, PGC_S_OVERRIDE);
4949 : }
4950 : Assert(XLOGbuffers > 0);
4951 :
4952 : /* XLogCtl */
4953 3784 : size = sizeof(XLogCtlData);
4954 :
4955 : /* WAL insertion locks, plus alignment */
4956 3784 : size = add_size(size, mul_size(sizeof(WALInsertLockPadded), NUM_XLOGINSERT_LOCKS + 1));
4957 : /* xlblocks array */
4958 3784 : size = add_size(size, mul_size(sizeof(XLogRecPtr), XLOGbuffers));
4959 : /* extra alignment padding for XLOG I/O buffers */
4960 3784 : size = add_size(size, XLOG_BLCKSZ);
4961 : /* and the buffers themselves */
4962 3784 : size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4963 :
4964 : /*
4965 : * Note: we don't count ControlFileData, it comes out of the "slop factor"
4966 : * added by CreateSharedMemoryAndSemaphores. This lets us use this
4967 : * routine again below to compute the actual allocation size.
4968 : */
4969 :
4970 3784 : return size;
4971 : }
4972 :
4973 : void
4974 1890 : XLOGShmemInit(void)
4975 : {
4976 : bool foundCFile,
4977 : foundXLog;
4978 : char *allocptr;
4979 : int i;
4980 : ControlFileData *localControlFile;
4981 :
4982 : #ifdef WAL_DEBUG
4983 :
4984 : /*
4985 : * Create a memory context for WAL debugging that's exempt from the normal
4986 : * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
4987 : * an allocation fails, but wal_debug is not for production use anyway.
4988 : */
4989 : if (walDebugCxt == NULL)
4990 : {
4991 : walDebugCxt = AllocSetContextCreate(TopMemoryContext,
4992 : "WAL Debug",
4993 : ALLOCSET_DEFAULT_SIZES);
4994 : MemoryContextAllowInCriticalSection(walDebugCxt, true);
4995 : }
4996 : #endif
4997 :
4998 :
4999 1890 : XLogCtl = (XLogCtlData *)
5000 1890 : ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
5001 :
5002 1890 : localControlFile = ControlFile;
5003 1890 : ControlFile = (ControlFileData *)
5004 1890 : ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
5005 :
5006 1890 : if (foundCFile || foundXLog)
5007 : {
5008 : /* both should be present or neither */
5009 : Assert(foundCFile && foundXLog);
5010 :
5011 : /* Initialize local copy of WALInsertLocks and register the tranche */
5012 0 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks;
5013 0 : LWLockRegisterTranche(LWTRANCHE_WAL_INSERT,
5014 : "wal_insert");
5015 :
5016 0 : if (localControlFile)
5017 0 : pfree(localControlFile);
5018 0 : return;
5019 : }
5020 1890 : memset(XLogCtl, 0, sizeof(XLogCtlData));
5021 :
5022 : /*
5023 : * Already have read control file locally, unless in bootstrap mode. Move
5024 : * contents into shared memory.
5025 : */
5026 1890 : if (localControlFile)
5027 : {
5028 924 : memcpy(ControlFile, localControlFile, sizeof(ControlFileData));
5029 924 : pfree(localControlFile);
5030 : }
5031 :
5032 : /*
5033 : * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
5034 : * multiple of the alignment for same, so no extra alignment padding is
5035 : * needed here.
5036 : */
5037 1890 : allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
5038 1890 : XLogCtl->xlblocks = (XLogRecPtr *) allocptr;
5039 1890 : memset(XLogCtl->xlblocks, 0, sizeof(XLogRecPtr) * XLOGbuffers);
5040 1890 : allocptr += sizeof(XLogRecPtr) * XLOGbuffers;
5041 :
5042 :
5043 : /* WAL insertion locks. Ensure they're aligned to the full padded size */
5044 1890 : allocptr += sizeof(WALInsertLockPadded) -
5045 1890 : ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
5046 1890 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks =
5047 : (WALInsertLockPadded *) allocptr;
5048 1890 : allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
5049 :
5050 1890 : LWLockRegisterTranche(LWTRANCHE_WAL_INSERT, "wal_insert");
5051 17010 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
5052 : {
5053 15120 : LWLockInitialize(&WALInsertLocks[i].l.lock, LWTRANCHE_WAL_INSERT);
5054 15120 : WALInsertLocks[i].l.insertingAt = InvalidXLogRecPtr;
5055 15120 : WALInsertLocks[i].l.lastImportantAt = InvalidXLogRecPtr;
5056 : }
5057 :
5058 : /*
5059 : * Align the start of the page buffers to a full xlog block size boundary.
5060 : * This simplifies some calculations in XLOG insertion. It is also
5061 : * required for O_DIRECT.
5062 : */
5063 1890 : allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
5064 1890 : XLogCtl->pages = allocptr;
5065 1890 : memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
5066 :
5067 : /*
5068 : * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
5069 : * in additional info.)
5070 : */
5071 1890 : XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
5072 1890 : XLogCtl->SharedRecoveryInProgress = true;
5073 1890 : XLogCtl->SharedHotStandbyActive = false;
5074 1890 : XLogCtl->WalWriterSleeping = false;
5075 :
5076 1890 : SpinLockInit(&XLogCtl->Insert.insertpos_lck);
5077 1890 : SpinLockInit(&XLogCtl->info_lck);
5078 1890 : SpinLockInit(&XLogCtl->ulsn_lck);
5079 1890 : InitSharedLatch(&XLogCtl->recoveryWakeupLatch);
5080 : }
5081 :
5082 : /*
5083 : * This func must be called ONCE on system install. It creates pg_control
5084 : * and the initial XLOG segment.
5085 : */
5086 : void
5087 322 : BootStrapXLOG(void)
5088 : {
5089 : CheckPoint checkPoint;
5090 : char *buffer;
5091 : XLogPageHeader page;
5092 : XLogLongPageHeader longpage;
5093 : XLogRecord *record;
5094 : char *recptr;
5095 : bool use_existent;
5096 : uint64 sysidentifier;
5097 : char mock_auth_nonce[MOCK_AUTH_NONCE_LEN];
5098 : struct timeval tv;
5099 : pg_crc32c crc;
5100 :
5101 : /*
5102 : * Select a hopefully-unique system identifier code for this installation.
5103 : * We use the result of gettimeofday(), including the fractional seconds
5104 : * field, as being about as unique as we can easily get. (Think not to
5105 : * use random(), since it hasn't been seeded and there's no portable way
5106 : * to seed it other than the system clock value...) The upper half of the
5107 : * uint64 value is just the tv_sec part, while the lower half contains the
5108 : * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
5109 : * PID for a little extra uniqueness. A person knowing this encoding can
5110 : * determine the initialization time of the installation, which could
5111 : * perhaps be useful sometimes.
5112 : */
5113 322 : gettimeofday(&tv, NULL);
5114 322 : sysidentifier = ((uint64) tv.tv_sec) << 32;
5115 322 : sysidentifier |= ((uint64) tv.tv_usec) << 12;
5116 322 : sysidentifier |= getpid() & 0xFFF;
5117 :
5118 : /*
5119 : * Generate a random nonce. This is used for authentication requests that
5120 : * will fail because the user does not exist. The nonce is used to create
5121 : * a genuine-looking password challenge for the non-existent user, in lieu
5122 : * of an actual stored password.
5123 : */
5124 322 : if (!pg_strong_random(mock_auth_nonce, MOCK_AUTH_NONCE_LEN))
5125 0 : ereport(PANIC,
5126 : (errcode(ERRCODE_INTERNAL_ERROR),
5127 : errmsg("could not generate secret authorization token")));
5128 :
5129 : /* First timeline ID is always 1 */
5130 322 : ThisTimeLineID = 1;
5131 :
5132 : /* page buffer must be aligned suitably for O_DIRECT */
5133 322 : buffer = (char *) palloc(XLOG_BLCKSZ + XLOG_BLCKSZ);
5134 322 : page = (XLogPageHeader) TYPEALIGN(XLOG_BLCKSZ, buffer);
5135 322 : memset(page, 0, XLOG_BLCKSZ);
5136 :
5137 : /*
5138 : * Set up information for the initial checkpoint record
5139 : *
5140 : * The initial checkpoint record is written to the beginning of the WAL
5141 : * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
5142 : * used, so that we can use 0/0 to mean "before any valid WAL segment".
5143 : */
5144 322 : checkPoint.redo = wal_segment_size + SizeOfXLogLongPHD;
5145 322 : checkPoint.ThisTimeLineID = ThisTimeLineID;
5146 322 : checkPoint.PrevTimeLineID = ThisTimeLineID;
5147 322 : checkPoint.fullPageWrites = fullPageWrites;
5148 322 : checkPoint.nextFullXid =
5149 : FullTransactionIdFromEpochAndXid(0, FirstNormalTransactionId);
5150 322 : checkPoint.nextOid = FirstBootstrapObjectId;
5151 322 : checkPoint.nextMulti = FirstMultiXactId;
5152 322 : checkPoint.nextMultiOffset = 0;
5153 322 : checkPoint.oldestXid = FirstNormalTransactionId;
5154 322 : checkPoint.oldestXidDB = TemplateDbOid;
5155 322 : checkPoint.oldestMulti = FirstMultiXactId;
5156 322 : checkPoint.oldestMultiDB = TemplateDbOid;
5157 322 : checkPoint.oldestCommitTsXid = InvalidTransactionId;
5158 322 : checkPoint.newestCommitTsXid = InvalidTransactionId;
5159 322 : checkPoint.time = (pg_time_t) time(NULL);
5160 322 : checkPoint.oldestActiveXid = InvalidTransactionId;
5161 :
5162 322 : ShmemVariableCache->nextFullXid = checkPoint.nextFullXid;
5163 322 : ShmemVariableCache->nextOid = checkPoint.nextOid;
5164 322 : ShmemVariableCache->oidCount = 0;
5165 322 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5166 322 : AdvanceOldestClogXid(checkPoint.oldestXid);
5167 322 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5168 322 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5169 322 : SetCommitTsLimit(InvalidTransactionId, InvalidTransactionId);
5170 :
5171 : /* Set up the XLOG page header */
5172 322 : page->xlp_magic = XLOG_PAGE_MAGIC;
5173 322 : page->xlp_info = XLP_LONG_HEADER;
5174 322 : page->xlp_tli = ThisTimeLineID;
5175 322 : page->xlp_pageaddr = wal_segment_size;
5176 322 : longpage = (XLogLongPageHeader) page;
5177 322 : longpage->xlp_sysid = sysidentifier;
5178 322 : longpage->xlp_seg_size = wal_segment_size;
5179 322 : longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5180 :
5181 : /* Insert the initial checkpoint record */
5182 322 : recptr = ((char *) page + SizeOfXLogLongPHD);
5183 322 : record = (XLogRecord *) recptr;
5184 322 : record->xl_prev = 0;
5185 322 : record->xl_xid = InvalidTransactionId;
5186 322 : record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
5187 322 : record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
5188 322 : record->xl_rmid = RM_XLOG_ID;
5189 322 : recptr += SizeOfXLogRecord;
5190 : /* fill the XLogRecordDataHeaderShort struct */
5191 322 : *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
5192 322 : *(recptr++) = sizeof(checkPoint);
5193 322 : memcpy(recptr, &checkPoint, sizeof(checkPoint));
5194 322 : recptr += sizeof(checkPoint);
5195 : Assert(recptr - (char *) record == record->xl_tot_len);
5196 :
5197 322 : INIT_CRC32C(crc);
5198 322 : COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
5199 322 : COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5200 322 : FIN_CRC32C(crc);
5201 322 : record->xl_crc = crc;
5202 :
5203 : /* Create first XLOG segment file */
5204 322 : use_existent = false;
5205 322 : openLogFile = XLogFileInit(1, &use_existent, false);
5206 :
5207 : /* Write the first page with the initial record */
5208 322 : errno = 0;
5209 322 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
5210 322 : if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5211 : {
5212 : /* if write didn't set errno, assume problem is no disk space */
5213 0 : if (errno == 0)
5214 0 : errno = ENOSPC;
5215 0 : ereport(PANIC,
5216 : (errcode_for_file_access(),
5217 : errmsg("could not write bootstrap write-ahead log file: %m")));
5218 : }
5219 322 : pgstat_report_wait_end();
5220 :
5221 322 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
5222 322 : if (pg_fsync(openLogFile) != 0)
5223 0 : ereport(PANIC,
5224 : (errcode_for_file_access(),
5225 : errmsg("could not fsync bootstrap write-ahead log file: %m")));
5226 322 : pgstat_report_wait_end();
5227 :
5228 322 : if (close(openLogFile) != 0)
5229 0 : ereport(PANIC,
5230 : (errcode_for_file_access(),
5231 : errmsg("could not close bootstrap write-ahead log file: %m")));
5232 :
5233 322 : openLogFile = -1;
5234 :
5235 : /* Now create pg_control */
5236 :
5237 322 : memset(ControlFile, 0, sizeof(ControlFileData));
5238 : /* Initialize pg_control status fields */
5239 322 : ControlFile->system_identifier = sysidentifier;
5240 322 : memcpy(ControlFile->mock_authentication_nonce, mock_auth_nonce, MOCK_AUTH_NONCE_LEN);
5241 322 : ControlFile->state = DB_SHUTDOWNED;
5242 322 : ControlFile->time = checkPoint.time;
5243 322 : ControlFile->checkPoint = checkPoint.redo;
5244 322 : ControlFile->checkPointCopy = checkPoint;
5245 322 : ControlFile->unloggedLSN = FirstNormalUnloggedLSN;
5246 :
5247 : /* Set important parameter values for use when replaying WAL */
5248 322 : ControlFile->MaxConnections = MaxConnections;
5249 322 : ControlFile->max_worker_processes = max_worker_processes;
5250 322 : ControlFile->max_wal_senders = max_wal_senders;
5251 322 : ControlFile->max_prepared_xacts = max_prepared_xacts;
5252 322 : ControlFile->max_locks_per_xact = max_locks_per_xact;
5253 322 : ControlFile->wal_level = wal_level;
5254 322 : ControlFile->wal_log_hints = wal_log_hints;
5255 322 : ControlFile->track_commit_timestamp = track_commit_timestamp;
5256 322 : ControlFile->data_checksum_version = bootstrap_data_checksum_version;
5257 :
5258 : /* some additional ControlFile fields are set in WriteControlFile() */
5259 :
5260 322 : WriteControlFile();
5261 :
5262 : /* Bootstrap the commit log, too */
5263 322 : BootStrapCLOG();
5264 322 : BootStrapCommitTs();
5265 322 : BootStrapSUBTRANS();
5266 322 : BootStrapMultiXact();
5267 :
5268 322 : pfree(buffer);
5269 :
5270 : /*
5271 : * Force control file to be read - in contrast to normal processing we'd
5272 : * otherwise never run the checks and GUC related initializations therein.
5273 : */
5274 322 : ReadControlFile();
5275 322 : }
5276 :
5277 : static char *
5278 548 : str_time(pg_time_t tnow)
5279 : {
5280 : static char buf[128];
5281 :
5282 548 : pg_strftime(buf, sizeof(buf),
5283 : "%Y-%m-%d %H:%M:%S %Z",
5284 548 : pg_localtime(&tnow, log_timezone));
5285 :
5286 548 : return buf;
5287 : }
5288 :
5289 : /*
5290 : * See if there are any recovery signal files and if so, set state for
5291 : * recovery.
5292 : *
5293 : * See if there is a recovery command file (recovery.conf), and if so
5294 : * throw an ERROR since as of PG12 we no longer recognize that.
5295 : */
5296 : static void
5297 1192 : readRecoverySignalFile(void)
5298 : {
5299 : struct stat stat_buf;
5300 :
5301 1192 : if (IsBootstrapProcessingMode())
5302 1446 : return;
5303 :
5304 : /*
5305 : * Check for old recovery API file: recovery.conf
5306 : */
5307 870 : if (stat(RECOVERY_COMMAND_FILE, &stat_buf) == 0)
5308 0 : ereport(FATAL,
5309 : (errcode_for_file_access(),
5310 : errmsg("using recovery command file \"%s\" is not supported",
5311 : RECOVERY_COMMAND_FILE)));
5312 :
5313 : /*
5314 : * Remove unused .done file, if present. Ignore if absent.
5315 : */
5316 870 : unlink(RECOVERY_COMMAND_DONE);
5317 :
5318 : /*
5319 : * Check for recovery signal files and if found, fsync them since they
5320 : * represent server state information. We don't sweat too much about the
5321 : * possibility of fsync failure, however.
5322 : *
5323 : * If present, standby signal file takes precedence. If neither is present
5324 : * then we won't enter archive recovery.
5325 : */
5326 870 : if (stat(STANDBY_SIGNAL_FILE, &stat_buf) == 0)
5327 : {
5328 : int fd;
5329 :
5330 68 : fd = BasicOpenFilePerm(STANDBY_SIGNAL_FILE, O_RDWR | PG_BINARY | get_sync_bit(sync_method),
5331 : S_IRUSR | S_IWUSR);
5332 68 : if (fd >= 0)
5333 : {
5334 68 : (void) pg_fsync(fd);
5335 68 : close(fd);
5336 : }
5337 68 : standby_signal_file_found = true;
5338 : }
5339 802 : else if (stat(RECOVERY_SIGNAL_FILE, &stat_buf) == 0)
5340 : {
5341 : int fd;
5342 :
5343 0 : fd = BasicOpenFilePerm(RECOVERY_SIGNAL_FILE, O_RDWR | PG_BINARY | get_sync_bit(sync_method),
5344 : S_IRUSR | S_IWUSR);
5345 0 : if (fd >= 0)
5346 : {
5347 0 : (void) pg_fsync(fd);
5348 0 : close(fd);
5349 : }
5350 0 : recovery_signal_file_found = true;
5351 : }
5352 :
5353 870 : StandbyModeRequested = false;
5354 870 : ArchiveRecoveryRequested = false;
5355 870 : if (standby_signal_file_found)
5356 : {
5357 68 : StandbyModeRequested = true;
5358 68 : ArchiveRecoveryRequested = true;
5359 : }
5360 802 : else if (recovery_signal_file_found)
5361 : {
5362 0 : StandbyModeRequested = false;
5363 0 : ArchiveRecoveryRequested = true;
5364 : }
5365 : else
5366 802 : return;
5367 :
5368 : /*
5369 : * We don't support standby mode in standalone backends; that requires
5370 : * other processes such as the WAL receiver to be alive.
5371 : */
5372 68 : if (StandbyModeRequested && !IsUnderPostmaster)
5373 0 : ereport(FATAL,
5374 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
5375 : errmsg("standby mode is not supported by single-user servers")));
5376 : }
5377 :
5378 : static void
5379 1192 : validateRecoveryParameters(void)
5380 : {
5381 1192 : if (!ArchiveRecoveryRequested)
5382 1124 : return;
5383 :
5384 : /*
5385 : * Check for compulsory parameters
5386 : */
5387 68 : if (StandbyModeRequested)
5388 : {
5389 72 : if ((PrimaryConnInfo == NULL || strcmp(PrimaryConnInfo, "") == 0) &&
5390 8 : (recoveryRestoreCommand == NULL || strcmp(recoveryRestoreCommand, "") == 0))
5391 0 : ereport(WARNING,
5392 : (errmsg("specified neither primary_conninfo nor restore_command"),
5393 : errhint("The database server will regularly poll the pg_wal subdirectory to check for files placed there.")));
5394 : }
5395 : else
5396 : {
5397 0 : if (recoveryRestoreCommand == NULL ||
5398 0 : strcmp(recoveryRestoreCommand, "") == 0)
5399 0 : ereport(FATAL,
5400 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5401 : errmsg("must specify restore_command when standby mode is not enabled")));
5402 : }
5403 :
5404 : /*
5405 : * Override any inconsistent requests. Note that this is a change of
5406 : * behaviour in 9.5; prior to this we simply ignored a request to pause if
5407 : * hot_standby = off, which was surprising behaviour.
5408 : */
5409 136 : if (recoveryTargetAction == RECOVERY_TARGET_ACTION_PAUSE &&
5410 68 : !EnableHotStandby)
5411 0 : recoveryTargetAction = RECOVERY_TARGET_ACTION_SHUTDOWN;
5412 :
5413 : /*
5414 : * Final parsing of recovery_target_time string; see also
5415 : * check_recovery_target_time().
5416 : */
5417 68 : if (recoveryTarget == RECOVERY_TARGET_TIME)
5418 : {
5419 0 : recoveryTargetTime = DatumGetTimestampTz(DirectFunctionCall3(timestamptz_in,
5420 : CStringGetDatum(recovery_target_time_string),
5421 : ObjectIdGetDatum(InvalidOid),
5422 : Int32GetDatum(-1)));
5423 : }
5424 :
5425 : /*
5426 : * If user specified recovery_target_timeline, validate it or compute the
5427 : * "latest" value. We can't do this until after we've gotten the restore
5428 : * command and set InArchiveRecovery, because we need to fetch timeline
5429 : * history files from the archive.
5430 : */
5431 68 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
5432 : {
5433 0 : TimeLineID rtli = recoveryTargetTLIRequested;
5434 :
5435 : /* Timeline 1 does not have a history file, all else should */
5436 0 : if (rtli != 1 && !existsTimeLineHistory(rtli))
5437 0 : ereport(FATAL,
5438 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5439 : errmsg("recovery target timeline %u does not exist",
5440 : rtli)));
5441 0 : recoveryTargetTLI = rtli;
5442 : }
5443 68 : else if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
5444 : {
5445 : /* We start the "latest" search from pg_control's timeline */
5446 68 : recoveryTargetTLI = findNewestTimeLine(recoveryTargetTLI);
5447 : }
5448 : else
5449 : {
5450 : /*
5451 : * else we just use the recoveryTargetTLI as already read from
5452 : * ControlFile
5453 : */
5454 : Assert(recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_CONTROLFILE);
5455 : }
5456 : }
5457 :
5458 : /*
5459 : * Exit archive-recovery state
5460 : */
5461 : static void
5462 46 : exitArchiveRecovery(TimeLineID endTLI, XLogRecPtr endOfLog)
5463 : {
5464 : char xlogfname[MAXFNAMELEN];
5465 : XLogSegNo endLogSegNo;
5466 : XLogSegNo startLogSegNo;
5467 :
5468 : /* we always switch to a new timeline after archive recovery */
5469 : Assert(endTLI != ThisTimeLineID);
5470 :
5471 : /*
5472 : * We are no longer in archive recovery state.
5473 : */
5474 46 : InArchiveRecovery = false;
5475 :
5476 : /*
5477 : * Update min recovery point one last time.
5478 : */
5479 46 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5480 :
5481 : /*
5482 : * If the ending log segment is still open, close it (to avoid problems on
5483 : * Windows with trying to rename or delete an open file).
5484 : */
5485 46 : if (readFile >= 0)
5486 : {
5487 46 : close(readFile);
5488 46 : readFile = -1;
5489 : }
5490 :
5491 : /*
5492 : * Calculate the last segment on the old timeline, and the first segment
5493 : * on the new timeline. If the switch happens in the middle of a segment,
5494 : * they are the same, but if the switch happens exactly at a segment
5495 : * boundary, startLogSegNo will be endLogSegNo + 1.
5496 : */
5497 46 : XLByteToPrevSeg(endOfLog, endLogSegNo, wal_segment_size);
5498 46 : XLByteToSeg(endOfLog, startLogSegNo, wal_segment_size);
5499 :
5500 : /*
5501 : * Initialize the starting WAL segment for the new timeline. If the switch
5502 : * happens in the middle of a segment, copy data from the last WAL segment
5503 : * of the old timeline up to the switch point, to the starting WAL segment
5504 : * on the new timeline.
5505 : */
5506 46 : if (endLogSegNo == startLogSegNo)
5507 : {
5508 : /*
5509 : * Make a copy of the file on the new timeline.
5510 : *
5511 : * Writing WAL isn't allowed yet, so there are no locking
5512 : * considerations. But we should be just as tense as XLogFileInit to
5513 : * avoid emplacing a bogus file.
5514 : */
5515 34 : XLogFileCopy(endLogSegNo, endTLI, endLogSegNo,
5516 34 : XLogSegmentOffset(endOfLog, wal_segment_size));
5517 : }
5518 : else
5519 : {
5520 : /*
5521 : * The switch happened at a segment boundary, so just create the next
5522 : * segment on the new timeline.
5523 : */
5524 12 : bool use_existent = true;
5525 : int fd;
5526 :
5527 12 : fd = XLogFileInit(startLogSegNo, &use_existent, true);
5528 :
5529 12 : if (close(fd) != 0)
5530 0 : ereport(ERROR,
5531 : (errcode_for_file_access(),
5532 : errmsg("could not close file \"%s\": %m",
5533 : XLogFileNameP(ThisTimeLineID, startLogSegNo))));
5534 : }
5535 :
5536 : /*
5537 : * Let's just make real sure there are not .ready or .done flags posted
5538 : * for the new segment.
5539 : */
5540 46 : XLogFileName(xlogfname, ThisTimeLineID, startLogSegNo, wal_segment_size);
5541 46 : XLogArchiveCleanup(xlogfname);
5542 :
5543 : /*
5544 : * Remove the signal files out of the way, so that we don't accidentally
5545 : * re-enter archive recovery mode in a subsequent crash.
5546 : */
5547 46 : if (standby_signal_file_found)
5548 46 : durable_unlink(STANDBY_SIGNAL_FILE, FATAL);
5549 :
5550 46 : if (recovery_signal_file_found)
5551 0 : durable_unlink(RECOVERY_SIGNAL_FILE, FATAL);
5552 :
5553 46 : ereport(LOG,
5554 : (errmsg("archive recovery complete")));
5555 46 : }
5556 :
5557 : /*
5558 : * Extract timestamp from WAL record.
5559 : *
5560 : * If the record contains a timestamp, returns true, and saves the timestamp
5561 : * in *recordXtime. If the record type has no timestamp, returns false.
5562 : * Currently, only transaction commit/abort records and restore points contain
5563 : * timestamps.
5564 : */
5565 : static bool
5566 158 : getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime)
5567 : {
5568 158 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
5569 158 : uint8 xact_info = info & XLOG_XACT_OPMASK;
5570 158 : uint8 rmid = XLogRecGetRmid(record);
5571 :
5572 158 : if (rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
5573 : {
5574 0 : *recordXtime = ((xl_restore_point *) XLogRecGetData(record))->rp_time;
5575 0 : return true;
5576 : }
5577 158 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_COMMIT ||
5578 : xact_info == XLOG_XACT_COMMIT_PREPARED))
5579 : {
5580 150 : *recordXtime = ((xl_xact_commit *) XLogRecGetData(record))->xact_time;
5581 150 : return true;
5582 : }
5583 8 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_ABORT ||
5584 : xact_info == XLOG_XACT_ABORT_PREPARED))
5585 : {
5586 8 : *recordXtime = ((xl_xact_abort *) XLogRecGetData(record))->xact_time;
5587 8 : return true;
5588 : }
5589 0 : return false;
5590 : }
5591 :
5592 : /*
5593 : * For point-in-time recovery, this function decides whether we want to
5594 : * stop applying the XLOG before the current record.
5595 : *
5596 : * Returns true if we are stopping, false otherwise. If stopping, some
5597 : * information is saved in recoveryStopXid et al for use in annotating the
5598 : * new timeline's history file.
5599 : */
5600 : static bool
5601 238472 : recoveryStopsBefore(XLogReaderState *record)
5602 : {
5603 238472 : bool stopsHere = false;
5604 : uint8 xact_info;
5605 : bool isCommit;
5606 238472 : TimestampTz recordXtime = 0;
5607 : TransactionId recordXid;
5608 :
5609 : /*
5610 : * Ignore recovery target settings when not in archive recovery (meaning
5611 : * we are in crash recovery).
5612 : */
5613 238472 : if (!ArchiveRecoveryRequested)
5614 68572 : return false;
5615 :
5616 : /* Check if we should stop as soon as reaching consistency */
5617 169900 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
5618 : {
5619 0 : ereport(LOG,
5620 : (errmsg("recovery stopping after reaching consistency")));
5621 :
5622 0 : recoveryStopAfter = false;
5623 0 : recoveryStopXid = InvalidTransactionId;
5624 0 : recoveryStopLSN = InvalidXLogRecPtr;
5625 0 : recoveryStopTime = 0;
5626 0 : recoveryStopName[0] = '\0';
5627 0 : return true;
5628 : }
5629 :
5630 : /* Check if target LSN has been reached */
5631 169900 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
5632 0 : !recoveryTargetInclusive &&
5633 0 : record->ReadRecPtr >= recoveryTargetLSN)
5634 : {
5635 0 : recoveryStopAfter = false;
5636 0 : recoveryStopXid = InvalidTransactionId;
5637 0 : recoveryStopLSN = record->ReadRecPtr;
5638 0 : recoveryStopTime = 0;
5639 0 : recoveryStopName[0] = '\0';
5640 0 : ereport(LOG,
5641 : (errmsg("recovery stopping before WAL location (LSN) \"%X/%X\"",
5642 : (uint32) (recoveryStopLSN >> 32),
5643 : (uint32) recoveryStopLSN)));
5644 0 : return true;
5645 : }
5646 :
5647 : /* Otherwise we only consider stopping before COMMIT or ABORT records. */
5648 169900 : if (XLogRecGetRmid(record) != RM_XACT_ID)
5649 169690 : return false;
5650 :
5651 210 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
5652 :
5653 210 : if (xact_info == XLOG_XACT_COMMIT)
5654 : {
5655 120 : isCommit = true;
5656 120 : recordXid = XLogRecGetXid(record);
5657 : }
5658 90 : else if (xact_info == XLOG_XACT_COMMIT_PREPARED)
5659 : {
5660 30 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
5661 : xl_xact_parsed_commit parsed;
5662 :
5663 30 : isCommit = true;
5664 30 : ParseCommitRecord(XLogRecGetInfo(record),
5665 : xlrec,
5666 : &parsed);
5667 30 : recordXid = parsed.twophase_xid;
5668 : }
5669 60 : else if (xact_info == XLOG_XACT_ABORT)
5670 : {
5671 2 : isCommit = false;
5672 2 : recordXid = XLogRecGetXid(record);
5673 : }
5674 58 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
5675 : {
5676 6 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
5677 : xl_xact_parsed_abort parsed;
5678 :
5679 6 : isCommit = true;
5680 6 : ParseAbortRecord(XLogRecGetInfo(record),
5681 : xlrec,
5682 : &parsed);
5683 6 : recordXid = parsed.twophase_xid;
5684 : }
5685 : else
5686 52 : return false;
5687 :
5688 158 : if (recoveryTarget == RECOVERY_TARGET_XID && !recoveryTargetInclusive)
5689 : {
5690 : /*
5691 : * There can be only one transaction end record with this exact
5692 : * transactionid
5693 : *
5694 : * when testing for an xid, we MUST test for equality only, since
5695 : * transactions are numbered in the order they start, not the order
5696 : * they complete. A higher numbered xid will complete before you about
5697 : * 50% of the time...
5698 : */
5699 0 : stopsHere = (recordXid == recoveryTargetXid);
5700 : }
5701 :
5702 158 : if (recoveryTarget == RECOVERY_TARGET_TIME &&
5703 0 : getRecordTimestamp(record, &recordXtime))
5704 : {
5705 : /*
5706 : * There can be many transactions that share the same commit time, so
5707 : * we stop after the last one, if we are inclusive, or stop at the
5708 : * first one if we are exclusive
5709 : */
5710 0 : if (recoveryTargetInclusive)
5711 0 : stopsHere = (recordXtime > recoveryTargetTime);
5712 : else
5713 0 : stopsHere = (recordXtime >= recoveryTargetTime);
5714 : }
5715 :
5716 158 : if (stopsHere)
5717 : {
5718 0 : recoveryStopAfter = false;
5719 0 : recoveryStopXid = recordXid;
5720 0 : recoveryStopTime = recordXtime;
5721 0 : recoveryStopLSN = InvalidXLogRecPtr;
5722 0 : recoveryStopName[0] = '\0';
5723 :
5724 0 : if (isCommit)
5725 : {
5726 0 : ereport(LOG,
5727 : (errmsg("recovery stopping before commit of transaction %u, time %s",
5728 : recoveryStopXid,
5729 : timestamptz_to_str(recoveryStopTime))));
5730 : }
5731 : else
5732 : {
5733 0 : ereport(LOG,
5734 : (errmsg("recovery stopping before abort of transaction %u, time %s",
5735 : recoveryStopXid,
5736 : timestamptz_to_str(recoveryStopTime))));
5737 : }
5738 : }
5739 :
5740 158 : return stopsHere;
5741 : }
5742 :
5743 : /*
5744 : * Same as recoveryStopsBefore, but called after applying the record.
5745 : *
5746 : * We also track the timestamp of the latest applied COMMIT/ABORT
5747 : * record in XLogCtl->recoveryLastXTime.
5748 : */
5749 : static bool
5750 238472 : recoveryStopsAfter(XLogReaderState *record)
5751 : {
5752 : uint8 info;
5753 : uint8 xact_info;
5754 : uint8 rmid;
5755 : TimestampTz recordXtime;
5756 :
5757 : /*
5758 : * Ignore recovery target settings when not in archive recovery (meaning
5759 : * we are in crash recovery).
5760 : */
5761 238472 : if (!ArchiveRecoveryRequested)
5762 68572 : return false;
5763 :
5764 169900 : info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
5765 169900 : rmid = XLogRecGetRmid(record);
5766 :
5767 : /*
5768 : * There can be many restore points that share the same name; we stop at
5769 : * the first one.
5770 : */
5771 169900 : if (recoveryTarget == RECOVERY_TARGET_NAME &&
5772 0 : rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
5773 : {
5774 : xl_restore_point *recordRestorePointData;
5775 :
5776 0 : recordRestorePointData = (xl_restore_point *) XLogRecGetData(record);
5777 :
5778 0 : if (strcmp(recordRestorePointData->rp_name, recoveryTargetName) == 0)
5779 : {
5780 0 : recoveryStopAfter = true;
5781 0 : recoveryStopXid = InvalidTransactionId;
5782 0 : recoveryStopLSN = InvalidXLogRecPtr;
5783 0 : (void) getRecordTimestamp(record, &recoveryStopTime);
5784 0 : strlcpy(recoveryStopName, recordRestorePointData->rp_name, MAXFNAMELEN);
5785 :
5786 0 : ereport(LOG,
5787 : (errmsg("recovery stopping at restore point \"%s\", time %s",
5788 : recoveryStopName,
5789 : timestamptz_to_str(recoveryStopTime))));
5790 0 : return true;
5791 : }
5792 : }
5793 :
5794 : /* Check if the target LSN has been reached */
5795 169900 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
5796 0 : recoveryTargetInclusive &&
5797 0 : record->ReadRecPtr >= recoveryTargetLSN)
5798 : {
5799 0 : recoveryStopAfter = true;
5800 0 : recoveryStopXid = InvalidTransactionId;
5801 0 : recoveryStopLSN = record->ReadRecPtr;
5802 0 : recoveryStopTime = 0;
5803 0 : recoveryStopName[0] = '\0';
5804 0 : ereport(LOG,
5805 : (errmsg("recovery stopping after WAL location (LSN) \"%X/%X\"",
5806 : (uint32) (recoveryStopLSN >> 32),
5807 : (uint32) recoveryStopLSN)));
5808 0 : return true;
5809 : }
5810 :
5811 169900 : if (rmid != RM_XACT_ID)
5812 169690 : return false;
5813 :
5814 210 : xact_info = info & XLOG_XACT_OPMASK;
5815 :
5816 210 : if (xact_info == XLOG_XACT_COMMIT ||
5817 60 : xact_info == XLOG_XACT_COMMIT_PREPARED ||
5818 58 : xact_info == XLOG_XACT_ABORT ||
5819 : xact_info == XLOG_XACT_ABORT_PREPARED)
5820 : {
5821 : TransactionId recordXid;
5822 :
5823 : /* Update the last applied transaction timestamp */
5824 158 : if (getRecordTimestamp(record, &recordXtime))
5825 158 : SetLatestXTime(recordXtime);
5826 :
5827 : /* Extract the XID of the committed/aborted transaction */
5828 158 : if (xact_info == XLOG_XACT_COMMIT_PREPARED)
5829 : {
5830 30 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
5831 : xl_xact_parsed_commit parsed;
5832 :
5833 30 : ParseCommitRecord(XLogRecGetInfo(record),
5834 : xlrec,
5835 : &parsed);
5836 30 : recordXid = parsed.twophase_xid;
5837 : }
5838 128 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
5839 : {
5840 6 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
5841 : xl_xact_parsed_abort parsed;
5842 :
5843 6 : ParseAbortRecord(XLogRecGetInfo(record),
5844 : xlrec,
5845 : &parsed);
5846 6 : recordXid = parsed.twophase_xid;
5847 : }
5848 : else
5849 122 : recordXid = XLogRecGetXid(record);
5850 :
5851 : /*
5852 : * There can be only one transaction end record with this exact
5853 : * transactionid
5854 : *
5855 : * when testing for an xid, we MUST test for equality only, since
5856 : * transactions are numbered in the order they start, not the order
5857 : * they complete. A higher numbered xid will complete before you about
5858 : * 50% of the time...
5859 : */
5860 158 : if (recoveryTarget == RECOVERY_TARGET_XID && recoveryTargetInclusive &&
5861 0 : recordXid == recoveryTargetXid)
5862 : {
5863 0 : recoveryStopAfter = true;
5864 0 : recoveryStopXid = recordXid;
5865 0 : recoveryStopTime = recordXtime;
5866 0 : recoveryStopLSN = InvalidXLogRecPtr;
5867 0 : recoveryStopName[0] = '\0';
5868 :
5869 0 : if (xact_info == XLOG_XACT_COMMIT ||
5870 : xact_info == XLOG_XACT_COMMIT_PREPARED)
5871 : {
5872 0 : ereport(LOG,
5873 : (errmsg("recovery stopping after commit of transaction %u, time %s",
5874 : recoveryStopXid,
5875 : timestamptz_to_str(recoveryStopTime))));
5876 : }
5877 0 : else if (xact_info == XLOG_XACT_ABORT ||
5878 : xact_info == XLOG_XACT_ABORT_PREPARED)
5879 : {
5880 0 : ereport(LOG,
5881 : (errmsg("recovery stopping after abort of transaction %u, time %s",
5882 : recoveryStopXid,
5883 : timestamptz_to_str(recoveryStopTime))));
5884 : }
5885 0 : return true;
5886 : }
5887 : }
5888 :
5889 : /* Check if we should stop as soon as reaching consistency */
5890 210 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
5891 : {
5892 0 : ereport(LOG,
5893 : (errmsg("recovery stopping after reaching consistency")));
5894 :
5895 0 : recoveryStopAfter = true;
5896 0 : recoveryStopXid = InvalidTransactionId;
5897 0 : recoveryStopTime = 0;
5898 0 : recoveryStopLSN = InvalidXLogRecPtr;
5899 0 : recoveryStopName[0] = '\0';
5900 0 : return true;
5901 : }
5902 :
5903 210 : return false;
5904 : }
5905 :
5906 : /*
5907 : * Wait until shared recoveryPause flag is cleared.
5908 : *
5909 : * XXX Could also be done with shared latch, avoiding the pg_usleep loop.
5910 : * Probably not worth the trouble though. This state shouldn't be one that
5911 : * anyone cares about server power consumption in.
5912 : */
5913 : static void
5914 0 : recoveryPausesHere(void)
5915 : {
5916 : /* Don't pause unless users can connect! */
5917 0 : if (!LocalHotStandbyActive)
5918 0 : return;
5919 :
5920 0 : ereport(LOG,
5921 : (errmsg("recovery has paused"),
5922 : errhint("Execute pg_wal_replay_resume() to continue.")));
5923 :
5924 0 : while (RecoveryIsPaused())
5925 : {
5926 0 : pg_usleep(1000000L); /* 1000 ms */
5927 0 : HandleStartupProcInterrupts();
5928 : }
5929 : }
5930 :
5931 : bool
5932 0 : RecoveryIsPaused(void)
5933 : {
5934 : bool recoveryPause;
5935 :
5936 0 : SpinLockAcquire(&XLogCtl->info_lck);
5937 0 : recoveryPause = XLogCtl->recoveryPause;
5938 0 : SpinLockRelease(&XLogCtl->info_lck);
5939 :
5940 0 : return recoveryPause;
5941 : }
5942 :
5943 : void
5944 0 : SetRecoveryPause(bool recoveryPause)
5945 : {
5946 0 : SpinLockAcquire(&XLogCtl->info_lck);
5947 0 : XLogCtl->recoveryPause = recoveryPause;
5948 0 : SpinLockRelease(&XLogCtl->info_lck);
5949 0 : }
5950 :
5951 : /*
5952 : * When recovery_min_apply_delay is set, we wait long enough to make sure
5953 : * certain record types are applied at least that interval behind the master.
5954 : *
5955 : * Returns true if we waited.
5956 : *
5957 : * Note that the delay is calculated between the WAL record log time and
5958 : * the current time on standby. We would prefer to keep track of when this
5959 : * standby received each WAL record, which would allow a more consistent
5960 : * approach and one not affected by time synchronisation issues, but that
5961 : * is significantly more effort and complexity for little actual gain in
5962 : * usability.
5963 : */
5964 : static bool
5965 238472 : recoveryApplyDelay(XLogReaderState *record)
5966 : {
5967 : uint8 xact_info;
5968 : TimestampTz xtime;
5969 : long secs;
5970 : int microsecs;
5971 :
5972 : /* nothing to do if no delay configured */
5973 238472 : if (recovery_min_apply_delay <= 0)
5974 238472 : return false;
5975 :
5976 : /* no delay is applied on a database not yet consistent */
5977 0 : if (!reachedConsistency)
5978 0 : return false;
5979 :
5980 : /* nothing to do if crash recovery is requested */
5981 0 : if (!ArchiveRecoveryRequested)
5982 0 : return false;
5983 :
5984 : /*
5985 : * Is it a COMMIT record?
5986 : *
5987 : * We deliberately choose not to delay aborts since they have no effect on
5988 : * MVCC. We already allow replay of records that don't have a timestamp,
5989 : * so there is already opportunity for issues caused by early conflicts on
5990 : * standbys.
5991 : */
5992 0 : if (XLogRecGetRmid(record) != RM_XACT_ID)
5993 0 : return false;
5994 :
5995 0 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
5996 :
5997 0 : if (xact_info != XLOG_XACT_COMMIT &&
5998 : xact_info != XLOG_XACT_COMMIT_PREPARED)
5999 0 : return false;
6000 :
6001 0 : if (!getRecordTimestamp(record, &xtime))
6002 0 : return false;
6003 :
6004 0 : recoveryDelayUntilTime =
6005 0 : TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
6006 :
6007 : /*
6008 : * Exit without arming the latch if it's already past time to apply this
6009 : * record
6010 : */
6011 0 : TimestampDifference(GetCurrentTimestamp(), recoveryDelayUntilTime,
6012 : &secs, µsecs);
6013 0 : if (secs <= 0 && microsecs <= 0)
6014 0 : return false;
6015 :
6016 : while (true)
6017 : {
6018 0 : ResetLatch(&XLogCtl->recoveryWakeupLatch);
6019 :
6020 : /* might change the trigger file's location */
6021 0 : HandleStartupProcInterrupts();
6022 :
6023 0 : if (CheckForStandbyTrigger())
6024 0 : break;
6025 :
6026 : /*
6027 : * Wait for difference between GetCurrentTimestamp() and
6028 : * recoveryDelayUntilTime
6029 : */
6030 0 : TimestampDifference(GetCurrentTimestamp(), recoveryDelayUntilTime,
6031 : &secs, µsecs);
6032 :
6033 : /*
6034 : * NB: We're ignoring waits below recovery_min_apply_delay's
6035 : * resolution.
6036 : */
6037 0 : if (secs <= 0 && microsecs / 1000 <= 0)
6038 0 : break;
6039 :
6040 0 : elog(DEBUG2, "recovery apply delay %ld seconds, %d milliseconds",
6041 : secs, microsecs / 1000);
6042 :
6043 0 : (void) WaitLatch(&XLogCtl->recoveryWakeupLatch,
6044 : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
6045 0 : secs * 1000L + microsecs / 1000,
6046 : WAIT_EVENT_RECOVERY_APPLY_DELAY);
6047 : }
6048 0 : return true;
6049 : }
6050 :
6051 : /*
6052 : * Save timestamp of latest processed commit/abort record.
6053 : *
6054 : * We keep this in XLogCtl, not a simple static variable, so that it can be
6055 : * seen by processes other than the startup process. Note in particular
6056 : * that CreateRestartPoint is executed in the checkpointer.
6057 : */
6058 : static void
6059 158 : SetLatestXTime(TimestampTz xtime)
6060 : {
6061 158 : SpinLockAcquire(&XLogCtl->info_lck);
6062 158 : XLogCtl->recoveryLastXTime = xtime;
6063 158 : SpinLockRelease(&XLogCtl->info_lck);
6064 158 : }
6065 :
6066 : /*
6067 : * Fetch timestamp of latest processed commit/abort record.
6068 : */
6069 : TimestampTz
6070 102 : GetLatestXTime(void)
6071 : {
6072 : TimestampTz xtime;
6073 :
6074 102 : SpinLockAcquire(&XLogCtl->info_lck);
6075 102 : xtime = XLogCtl->recoveryLastXTime;
6076 102 : SpinLockRelease(&XLogCtl->info_lck);
6077 :
6078 102 : return xtime;
6079 : }
6080 :
6081 : /*
6082 : * Save timestamp of the next chunk of WAL records to apply.
6083 : *
6084 : * We keep this in XLogCtl, not a simple static variable, so that it can be
6085 : * seen by all backends.
6086 : */
6087 : static void
6088 336 : SetCurrentChunkStartTime(TimestampTz xtime)
6089 : {
6090 336 : SpinLockAcquire(&XLogCtl->info_lck);
6091 336 : XLogCtl->currentChunkStartTime = xtime;
6092 336 : SpinLockRelease(&XLogCtl->info_lck);
6093 336 : }
6094 :
6095 : /*
6096 : * Fetch timestamp of latest processed commit/abort record.
6097 : * Startup process maintains an accurate local copy in XLogReceiptTime
6098 : */
6099 : TimestampTz
6100 0 : GetCurrentChunkReplayStartTime(void)
6101 : {
6102 : TimestampTz xtime;
6103 :
6104 0 : SpinLockAcquire(&XLogCtl->info_lck);
6105 0 : xtime = XLogCtl->currentChunkStartTime;
6106 0 : SpinLockRelease(&XLogCtl->info_lck);
6107 :
6108 0 : return xtime;
6109 : }
6110 :
6111 : /*
6112 : * Returns time of receipt of current chunk of XLOG data, as well as
6113 : * whether it was received from streaming replication or from archives.
6114 : */
6115 : void
6116 0 : GetXLogReceiptTime(TimestampTz *rtime, bool *fromStream)
6117 : {
6118 : /*
6119 : * This must be executed in the startup process, since we don't export the
6120 : * relevant state to shared memory.
6121 : */
6122 : Assert(InRecovery);
6123 :
6124 0 : *rtime = XLogReceiptTime;
6125 0 : *fromStream = (XLogReceiptSource == XLOG_FROM_STREAM);
6126 0 : }
6127 :
6128 : /*
6129 : * Note that text field supplied is a parameter name and does not require
6130 : * translation
6131 : */
6132 : #define RecoveryRequiresIntParameter(param_name, currValue, minValue) \
6133 : do { \
6134 : if ((currValue) < (minValue)) \
6135 : ereport(ERROR, \
6136 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
6137 : errmsg("hot standby is not possible because " \
6138 : "%s = %d is a lower setting than on the master server " \
6139 : "(its value was %d)", \
6140 : param_name, \
6141 : currValue, \
6142 : minValue))); \
6143 : } while(0)
6144 :
6145 : /*
6146 : * Check to see if required parameters are set high enough on this server
6147 : * for various aspects of recovery operation.
6148 : *
6149 : * Note that all the parameters which this function tests need to be
6150 : * listed in Administrator's Overview section in high-availability.sgml.
6151 : * If you change them, don't forget to update the list.
6152 : */
6153 : static void
6154 130 : CheckRequiredParameterValues(void)
6155 : {
6156 : /*
6157 : * For archive recovery, the WAL must be generated with at least 'replica'
6158 : * wal_level.
6159 : */
6160 130 : if (ArchiveRecoveryRequested && ControlFile->wal_level == WAL_LEVEL_MINIMAL)
6161 : {
6162 0 : ereport(WARNING,
6163 : (errmsg("WAL was generated with wal_level=minimal, data may be missing"),
6164 : errhint("This happens if you temporarily set wal_level=minimal without taking a new base backup.")));
6165 : }
6166 :
6167 : /*
6168 : * For Hot Standby, the WAL must be generated with 'replica' mode, and we
6169 : * must have at least as many backend slots as the primary.
6170 : */
6171 130 : if (ArchiveRecoveryRequested && EnableHotStandby)
6172 : {
6173 72 : if (ControlFile->wal_level < WAL_LEVEL_REPLICA)
6174 0 : ereport(ERROR,
6175 : (errmsg("hot standby is not possible because wal_level was not set to \"replica\" or higher on the master server"),
6176 : errhint("Either set wal_level to \"replica\" on the master, or turn off hot_standby here.")));
6177 :
6178 : /* We ignore autovacuum_max_workers when we make this test. */
6179 72 : RecoveryRequiresIntParameter("max_connections",
6180 : MaxConnections,
6181 : ControlFile->MaxConnections);
6182 72 : RecoveryRequiresIntParameter("max_worker_processes",
6183 : max_worker_processes,
6184 : ControlFile->max_worker_processes);
6185 72 : RecoveryRequiresIntParameter("max_wal_senders",
6186 : max_wal_senders,
6187 : ControlFile->max_wal_senders);
6188 72 : RecoveryRequiresIntParameter("max_prepared_transactions",
6189 : max_prepared_xacts,
6190 : ControlFile->max_prepared_xacts);
6191 72 : RecoveryRequiresIntParameter("max_locks_per_transaction",
6192 : max_locks_per_xact,
6193 : ControlFile->max_locks_per_xact);
6194 : }
6195 130 : }
6196 :
6197 : /*
6198 : * This must be called ONCE during postmaster or standalone-backend startup
6199 : */
6200 : void
6201 1192 : StartupXLOG(void)
6202 : {
6203 : XLogCtlInsert *Insert;
6204 : CheckPoint checkPoint;
6205 : bool wasShutdown;
6206 1192 : bool reachedStopPoint = false;
6207 1192 : bool haveBackupLabel = false;
6208 1192 : bool haveTblspcMap = false;
6209 : XLogRecPtr RecPtr,
6210 : checkPointLoc,
6211 : EndOfLog;
6212 : TimeLineID EndOfLogTLI;
6213 : TimeLineID PrevTimeLineID;
6214 : XLogRecord *record;
6215 : TransactionId oldestActiveXID;
6216 1192 : bool backupEndRequired = false;
6217 1192 : bool backupFromStandby = false;
6218 : DBState dbstate_at_startup;
6219 : XLogReaderState *xlogreader;
6220 : XLogPageReadPrivate private;
6221 1192 : bool fast_promoted = false;
6222 : struct stat st;
6223 :
6224 : /*
6225 : * We should have an aux process resource owner to use, and we should not
6226 : * be in a transaction that's installed some other resowner.
6227 : */
6228 : Assert(AuxProcessResourceOwner != NULL);
6229 : Assert(CurrentResourceOwner == NULL ||
6230 : CurrentResourceOwner == AuxProcessResourceOwner);
6231 1192 : CurrentResourceOwner = AuxProcessResourceOwner;
6232 :
6233 : /*
6234 : * Check that contents look valid.
6235 : */
6236 1192 : if (!XRecOffIsValid(ControlFile->checkPoint))
6237 0 : ereport(FATAL,
6238 : (errmsg("control file contains invalid checkpoint location")));
6239 :
6240 1192 : switch (ControlFile->state)
6241 : {
6242 : case DB_SHUTDOWNED:
6243 : /* This is the expected case, so don't be chatty in standalone mode */
6244 1080 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
6245 : (errmsg("database system was shut down at %s",
6246 : str_time(ControlFile->time))));
6247 1080 : break;
6248 :
6249 : case DB_SHUTDOWNED_IN_RECOVERY:
6250 6 : ereport(LOG,
6251 : (errmsg("database system was shut down in recovery at %s",
6252 : str_time(ControlFile->time))));
6253 6 : break;
6254 :
6255 : case DB_SHUTDOWNING:
6256 0 : ereport(LOG,
6257 : (errmsg("database system shutdown was interrupted; last known up at %s",
6258 : str_time(ControlFile->time))));
6259 0 : break;
6260 :
6261 : case DB_IN_CRASH_RECOVERY:
6262 0 : ereport(LOG,
6263 : (errmsg("database system was interrupted while in recovery at %s",
6264 : str_time(ControlFile->time)),
6265 : errhint("This probably means that some data is corrupted and"
6266 : " you will have to use the last backup for recovery.")));
6267 0 : break;
6268 :
6269 : case DB_IN_ARCHIVE_RECOVERY:
6270 4 : ereport(LOG,
6271 : (errmsg("database system was interrupted while in recovery at log time %s",
6272 : str_time(ControlFile->checkPointCopy.time)),
6273 : errhint("If this has occurred more than once some data might be corrupted"
6274 : " and you might need to choose an earlier recovery target.")));
6275 4 : break;
6276 :
6277 : case DB_IN_PRODUCTION:
6278 102 : ereport(LOG,
6279 : (errmsg("database system was interrupted; last known up at %s",
6280 : str_time(ControlFile->time))));
6281 102 : break;
6282 :
6283 : default:
6284 0 : ereport(FATAL,
6285 : (errmsg("control file contains invalid database cluster state")));
6286 : }
6287 :
6288 : /* This is just to allow attaching to startup process with a debugger */
6289 : #ifdef XLOG_REPLAY_DELAY
6290 : if (ControlFile->state != DB_SHUTDOWNED)
6291 : pg_usleep(60000000L);
6292 : #endif
6293 :
6294 : /*
6295 : * Verify that pg_wal and pg_wal/archive_status exist. In cases where
6296 : * someone has performed a copy for PITR, these directories may have been
6297 : * excluded and need to be re-created.
6298 : */
6299 1192 : ValidateXLOGDirectoryStructure();
6300 :
6301 : /*----------
6302 : * If we previously crashed, perform a couple of actions:
6303 : * - The pg_wal directory may still include some temporary WAL segments
6304 : * used when creating a new segment, so perform some clean up to not
6305 : * bloat this path. This is done first as there is no point to sync this
6306 : * temporary data.
6307 : * - There might be data which we had written, intending to fsync it,
6308 : * but which we had not actually fsync'd yet. Therefore, a power failure
6309 : * in the near future might cause earlier unflushed writes to be lost,
6310 : * even though more recent data written to disk from here on would be
6311 : * persisted. To avoid that, fsync the entire data directory.
6312 : *---------
6313 : */
6314 1304 : if (ControlFile->state != DB_SHUTDOWNED &&
6315 112 : ControlFile->state != DB_SHUTDOWNED_IN_RECOVERY)
6316 : {
6317 106 : RemoveTempXlogFiles();
6318 106 : SyncDataDirectory();
6319 : }
6320 :
6321 : /*
6322 : * Initialize on the assumption we want to recover to the latest timeline
6323 : * that's active according to pg_control.
6324 : */
6325 2384 : if (ControlFile->minRecoveryPointTLI >
6326 1192 : ControlFile->checkPointCopy.ThisTimeLineID)
6327 2 : recoveryTargetTLI = ControlFile->minRecoveryPointTLI;
6328 : else
6329 1190 : recoveryTargetTLI = ControlFile->checkPointCopy.ThisTimeLineID;
6330 :
6331 : /*
6332 : * Check for signal files, and if so set up state for offline recovery
6333 : */
6334 1192 : readRecoverySignalFile();
6335 1192 : validateRecoveryParameters();
6336 :
6337 1192 : if (ArchiveRecoveryRequested)
6338 : {
6339 68 : if (StandbyModeRequested)
6340 68 : ereport(LOG,
6341 : (errmsg("entering standby mode")));
6342 0 : else if (recoveryTarget == RECOVERY_TARGET_XID)
6343 0 : ereport(LOG,
6344 : (errmsg("starting point-in-time recovery to XID %u",
6345 : recoveryTargetXid)));
6346 0 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
6347 0 : ereport(LOG,
6348 : (errmsg("starting point-in-time recovery to %s",
6349 : timestamptz_to_str(recoveryTargetTime))));
6350 0 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
6351 0 : ereport(LOG,
6352 : (errmsg("starting point-in-time recovery to \"%s\"",
6353 : recoveryTargetName)));
6354 0 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
6355 0 : ereport(LOG,
6356 : (errmsg("starting point-in-time recovery to WAL location (LSN) \"%X/%X\"",
6357 : (uint32) (recoveryTargetLSN >> 32),
6358 : (uint32) recoveryTargetLSN)));
6359 0 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
6360 0 : ereport(LOG,
6361 : (errmsg("starting point-in-time recovery to earliest consistent point")));
6362 : else
6363 0 : ereport(LOG,
6364 : (errmsg("starting archive recovery")));
6365 : }
6366 :
6367 : /*
6368 : * Take ownership of the wakeup latch if we're going to sleep during
6369 : * recovery.
6370 : */
6371 1192 : if (ArchiveRecoveryRequested)
6372 68 : OwnLatch(&XLogCtl->recoveryWakeupLatch);
6373 :
6374 : /* Set up XLOG reader facility */
6375 1192 : MemSet(&private, 0, sizeof(XLogPageReadPrivate));
6376 1192 : xlogreader = XLogReaderAllocate(wal_segment_size, NULL,
6377 : &XLogPageRead, &private);
6378 1192 : if (!xlogreader)
6379 0 : ereport(ERROR,
6380 : (errcode(ERRCODE_OUT_OF_MEMORY),
6381 : errmsg("out of memory"),
6382 : errdetail("Failed while allocating a WAL reading processor.")));
6383 1192 : xlogreader->system_identifier = ControlFile->system_identifier;
6384 :
6385 : /*
6386 : * Allocate two page buffers dedicated to WAL consistency checks. We do
6387 : * it this way, rather than just making static arrays, for two reasons:
6388 : * (1) no need to waste the storage in most instantiations of the backend;
6389 : * (2) a static char array isn't guaranteed to have any particular
6390 : * alignment, whereas palloc() will provide MAXALIGN'd storage.
6391 : */
6392 1192 : replay_image_masked = (char *) palloc(BLCKSZ);
6393 1192 : master_image_masked = (char *) palloc(BLCKSZ);
6394 :
6395 1192 : if (read_backup_label(&checkPointLoc, &backupEndRequired,
6396 : &backupFromStandby))
6397 : {
6398 54 : List *tablespaces = NIL;
6399 :
6400 : /*
6401 : * Archive recovery was requested, and thanks to the backup label
6402 : * file, we know how far we need to replay to reach consistency. Enter
6403 : * archive recovery directly.
6404 : */
6405 54 : InArchiveRecovery = true;
6406 54 : if (StandbyModeRequested)
6407 54 : StandbyMode = true;
6408 :
6409 : /*
6410 : * When a backup_label file is present, we want to roll forward from
6411 : * the checkpoint it identifies, rather than using pg_control.
6412 : */
6413 54 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 0, true);
6414 54 : if (record != NULL)
6415 : {
6416 54 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
6417 54 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
6418 54 : ereport(DEBUG1,
6419 : (errmsg("checkpoint record is at %X/%X",
6420 : (uint32) (checkPointLoc >> 32), (uint32) checkPointLoc)));
6421 54 : InRecovery = true; /* force recovery even if SHUTDOWNED */
6422 :
6423 : /*
6424 : * Make sure that REDO location exists. This may not be the case
6425 : * if there was a crash during an online backup, which left a
6426 : * backup_label around that references a WAL segment that's
6427 : * already been archived.
6428 : */
6429 54 : if (checkPoint.redo < checkPointLoc)
6430 : {
6431 54 : if (!ReadRecord(xlogreader, checkPoint.redo, LOG, false))
6432 0 : ereport(FATAL,
6433 : (errmsg("could not find redo location referenced by checkpoint record"),
6434 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" and add required recovery options.\n"
6435 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
6436 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
6437 : DataDir, DataDir, DataDir)));
6438 : }
6439 : }
6440 : else
6441 : {
6442 0 : ereport(FATAL,
6443 : (errmsg("could not locate required checkpoint record"),
6444 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" and add required recovery options.\n"
6445 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
6446 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
6447 : DataDir, DataDir, DataDir)));
6448 : wasShutdown = false; /* keep compiler quiet */
6449 : }
6450 :
6451 : /* read the tablespace_map file if present and create symlinks. */
6452 54 : if (read_tablespace_map(&tablespaces))
6453 : {
6454 : ListCell *lc;
6455 :
6456 0 : foreach(lc, tablespaces)
6457 : {
6458 0 : tablespaceinfo *ti = lfirst(lc);
6459 : char *linkloc;
6460 :
6461 0 : linkloc = psprintf("pg_tblspc/%s", ti->oid);
6462 :
6463 : /*
6464 : * Remove the existing symlink if any and Create the symlink
6465 : * under PGDATA.
6466 : */
6467 0 : remove_tablespace_symlink(linkloc);
6468 :
6469 0 : if (symlink(ti->path, linkloc) < 0)
6470 0 : ereport(ERROR,
6471 : (errcode_for_file_access(),
6472 : errmsg("could not create symbolic link \"%s\": %m",
6473 : linkloc)));
6474 :
6475 0 : pfree(ti->oid);
6476 0 : pfree(ti->path);
6477 0 : pfree(ti);
6478 : }
6479 :
6480 : /* set flag to delete it later */
6481 0 : haveTblspcMap = true;
6482 : }
6483 :
6484 : /* set flag to delete it later */
6485 54 : haveBackupLabel = true;
6486 : }
6487 : else
6488 : {
6489 : /*
6490 : * If tablespace_map file is present without backup_label file, there
6491 : * is no use of such file. There is no harm in retaining it, but it
6492 : * is better to get rid of the map file so that we don't have any
6493 : * redundant file in data directory and it will avoid any sort of
6494 : * confusion. It seems prudent though to just rename the file out of
6495 : * the way rather than delete it completely, also we ignore any error
6496 : * that occurs in rename operation as even if map file is present
6497 : * without backup_label file, it is harmless.
6498 : */
6499 1138 : if (stat(TABLESPACE_MAP, &st) == 0)
6500 : {
6501 0 : unlink(TABLESPACE_MAP_OLD);
6502 0 : if (durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, DEBUG1) == 0)
6503 0 : ereport(LOG,
6504 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
6505 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
6506 : errdetail("File \"%s\" was renamed to \"%s\".",
6507 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
6508 : else
6509 0 : ereport(LOG,
6510 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
6511 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
6512 : errdetail("Could not rename file \"%s\" to \"%s\": %m.",
6513 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
6514 : }
6515 :
6516 : /*
6517 : * It's possible that archive recovery was requested, but we don't
6518 : * know how far we need to replay the WAL before we reach consistency.
6519 : * This can happen for example if a base backup is taken from a
6520 : * running server using an atomic filesystem snapshot, without calling
6521 : * pg_start/stop_backup. Or if you just kill a running master server
6522 : * and put it into archive recovery by creating a recovery signal
6523 : * file.
6524 : *
6525 : * Our strategy in that case is to perform crash recovery first,
6526 : * replaying all the WAL present in pg_wal, and only enter archive
6527 : * recovery after that.
6528 : *
6529 : * But usually we already know how far we need to replay the WAL (up
6530 : * to minRecoveryPoint, up to backupEndPoint, or until we see an
6531 : * end-of-backup record), and we can enter archive recovery directly.
6532 : */
6533 1152 : if (ArchiveRecoveryRequested &&
6534 20 : (ControlFile->minRecoveryPoint != InvalidXLogRecPtr ||
6535 12 : ControlFile->backupEndRequired ||
6536 12 : ControlFile->backupEndPoint != InvalidXLogRecPtr ||
6537 6 : ControlFile->state == DB_SHUTDOWNED))
6538 : {
6539 12 : InArchiveRecovery = true;
6540 12 : if (StandbyModeRequested)
6541 12 : StandbyMode = true;
6542 : }
6543 :
6544 : /* Get the last valid checkpoint record. */
6545 1138 : checkPointLoc = ControlFile->checkPoint;
6546 1138 : RedoStartLSN = ControlFile->checkPointCopy.redo;
6547 1138 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 1, true);
6548 1138 : if (record != NULL)
6549 : {
6550 1138 : ereport(DEBUG1,
6551 : (errmsg("checkpoint record is at %X/%X",
6552 : (uint32) (checkPointLoc >> 32), (uint32) checkPointLoc)));
6553 : }
6554 : else
6555 : {
6556 : /*
6557 : * We used to attempt to go back to a secondary checkpoint record
6558 : * here, but only when not in standby mode. We now just fail if we
6559 : * can't read the last checkpoint because this allows us to
6560 : * simplify processing around checkpoints.
6561 : */
6562 0 : ereport(PANIC,
6563 : (errmsg("could not locate a valid checkpoint record")));
6564 : }
6565 1138 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
6566 1138 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
6567 : }
6568 :
6569 : /*
6570 : * Clear out any old relcache cache files. This is *necessary* if we do
6571 : * any WAL replay, since that would probably result in the cache files
6572 : * being out of sync with database reality. In theory we could leave them
6573 : * in place if the database had been cleanly shut down, but it seems
6574 : * safest to just remove them always and let them be rebuilt during the
6575 : * first backend startup. These files needs to be removed from all
6576 : * directories including pg_tblspc, however the symlinks are created only
6577 : * after reading tablespace_map file in case of archive recovery from
6578 : * backup, so needs to clear old relcache files here after creating
6579 : * symlinks.
6580 : */
6581 1192 : RelationCacheInitFileRemove();
6582 :
6583 : /*
6584 : * If the location of the checkpoint record is not on the expected
6585 : * timeline in the history of the requested timeline, we cannot proceed:
6586 : * the backup is not part of the history of the requested timeline.
6587 : */
6588 : Assert(expectedTLEs); /* was initialized by reading checkpoint
6589 : * record */
6590 2384 : if (tliOfPointInHistory(checkPointLoc, expectedTLEs) !=
6591 1192 : checkPoint.ThisTimeLineID)
6592 : {
6593 : XLogRecPtr switchpoint;
6594 :
6595 : /*
6596 : * tliSwitchPoint will throw an error if the checkpoint's timeline is
6597 : * not in expectedTLEs at all.
6598 : */
6599 0 : switchpoint = tliSwitchPoint(ControlFile->checkPointCopy.ThisTimeLineID, expectedTLEs, NULL);
6600 0 : ereport(FATAL,
6601 : (errmsg("requested timeline %u is not a child of this server's history",
6602 : recoveryTargetTLI),
6603 : 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.",
6604 : (uint32) (ControlFile->checkPoint >> 32),
6605 : (uint32) ControlFile->checkPoint,
6606 : ControlFile->checkPointCopy.ThisTimeLineID,
6607 : (uint32) (switchpoint >> 32),
6608 : (uint32) switchpoint)));
6609 : }
6610 :
6611 : /*
6612 : * The min recovery point should be part of the requested timeline's
6613 : * history, too.
6614 : */
6615 1204 : if (!XLogRecPtrIsInvalid(ControlFile->minRecoveryPoint) &&
6616 12 : tliOfPointInHistory(ControlFile->minRecoveryPoint - 1, expectedTLEs) !=
6617 12 : ControlFile->minRecoveryPointTLI)
6618 0 : ereport(FATAL,
6619 : (errmsg("requested timeline %u does not contain minimum recovery point %X/%X on timeline %u",
6620 : recoveryTargetTLI,
6621 : (uint32) (ControlFile->minRecoveryPoint >> 32),
6622 : (uint32) ControlFile->minRecoveryPoint,
6623 : ControlFile->minRecoveryPointTLI)));
6624 :
6625 1192 : LastRec = RecPtr = checkPointLoc;
6626 :
6627 1192 : ereport(DEBUG1,
6628 : (errmsg_internal("redo record is at %X/%X; shutdown %s",
6629 : (uint32) (checkPoint.redo >> 32), (uint32) checkPoint.redo,
6630 : wasShutdown ? "true" : "false")));
6631 1192 : ereport(DEBUG1,
6632 : (errmsg_internal("next transaction ID: " UINT64_FORMAT "; next OID: %u",
6633 : U64FromFullTransactionId(checkPoint.nextFullXid),
6634 : checkPoint.nextOid)));
6635 1192 : ereport(DEBUG1,
6636 : (errmsg_internal("next MultiXactId: %u; next MultiXactOffset: %u",
6637 : checkPoint.nextMulti, checkPoint.nextMultiOffset)));
6638 1192 : ereport(DEBUG1,
6639 : (errmsg_internal("oldest unfrozen transaction ID: %u, in database %u",
6640 : checkPoint.oldestXid, checkPoint.oldestXidDB)));
6641 1192 : ereport(DEBUG1,
6642 : (errmsg_internal("oldest MultiXactId: %u, in database %u",
6643 : checkPoint.oldestMulti, checkPoint.oldestMultiDB)));
6644 1192 : ereport(DEBUG1,
6645 : (errmsg_internal("commit timestamp Xid oldest/newest: %u/%u",
6646 : checkPoint.oldestCommitTsXid,
6647 : checkPoint.newestCommitTsXid)));
6648 1192 : if (!TransactionIdIsNormal(XidFromFullTransactionId(checkPoint.nextFullXid)))
6649 0 : ereport(PANIC,
6650 : (errmsg("invalid next transaction ID")));
6651 :
6652 : /* initialize shared memory variables from the checkpoint record */
6653 1192 : ShmemVariableCache->nextFullXid = checkPoint.nextFullXid;
6654 1192 : ShmemVariableCache->nextOid = checkPoint.nextOid;
6655 1192 : ShmemVariableCache->oidCount = 0;
6656 1192 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
6657 1192 : AdvanceOldestClogXid(checkPoint.oldestXid);
6658 1192 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
6659 1192 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
6660 1192 : SetCommitTsLimit(checkPoint.oldestCommitTsXid,
6661 : checkPoint.newestCommitTsXid);
6662 1192 : XLogCtl->ckptFullXid = checkPoint.nextFullXid;
6663 :
6664 : /*
6665 : * Initialize replication slots, before there's a chance to remove
6666 : * required resources.
6667 : */
6668 1192 : StartupReplicationSlots();
6669 :
6670 : /*
6671 : * Startup logical state, needs to be setup now so we have proper data
6672 : * during crash recovery.
6673 : */
6674 1192 : StartupReorderBuffer();
6675 :
6676 : /*
6677 : * Startup MultiXact. We need to do this early to be able to replay
6678 : * truncations.
6679 : */
6680 1192 : StartupMultiXact();
6681 :
6682 : /*
6683 : * Ditto for commit timestamps. Activate the facility if the setting is
6684 : * enabled in the control file, as there should be no tracking of commit
6685 : * timestamps done when the setting was disabled. This facility can be
6686 : * started or stopped when replaying a XLOG_PARAMETER_CHANGE record.
6687 : */
6688 1192 : if (ControlFile->track_commit_timestamp)
6689 16 : StartupCommitTs();
6690 :
6691 : /*
6692 : * Recover knowledge about replay progress of known replication partners.
6693 : */
6694 1192 : StartupReplicationOrigin();
6695 :
6696 : /*
6697 : * Initialize unlogged LSN. On a clean shutdown, it's restored from the
6698 : * control file. On recovery, all unlogged relations are blown away, so
6699 : * the unlogged LSN counter can be reset too.
6700 : */
6701 1192 : if (ControlFile->state == DB_SHUTDOWNED)
6702 1080 : XLogCtl->unloggedLSN = ControlFile->unloggedLSN;
6703 : else
6704 112 : XLogCtl->unloggedLSN = FirstNormalUnloggedLSN;
6705 :
6706 : /*
6707 : * We must replay WAL entries using the same TimeLineID they were created
6708 : * under, so temporarily adopt the TLI indicated by the checkpoint (see
6709 : * also xlog_redo()).
6710 : */
6711 1192 : ThisTimeLineID = checkPoint.ThisTimeLineID;
6712 :
6713 : /*
6714 : * Copy any missing timeline history files between 'now' and the recovery
6715 : * target timeline from archive to pg_wal. While we don't need those files
6716 : * ourselves - the history file of the recovery target timeline covers all
6717 : * the previous timelines in the history too - a cascading standby server
6718 : * might be interested in them. Or, if you archive the WAL from this
6719 : * server to a different archive than the master, it'd be good for all the
6720 : * history files to get archived there after failover, so that you can use
6721 : * one of the old timelines as a PITR target. Timeline history files are
6722 : * small, so it's better to copy them unnecessarily than not copy them and
6723 : * regret later.
6724 : */
6725 1192 : restoreTimeLineHistoryFiles(ThisTimeLineID, recoveryTargetTLI);
6726 :
6727 : /*
6728 : * Before running in recovery, scan pg_twophase and fill in its status to
6729 : * be able to work on entries generated by redo. Doing a scan before
6730 : * taking any recovery action has the merit to discard any 2PC files that
6731 : * are newer than the first record to replay, saving from any conflicts at
6732 : * replay. This avoids as well any subsequent scans when doing recovery
6733 : * of the on-disk two-phase data.
6734 : */
6735 1192 : restoreTwoPhaseData();
6736 :
6737 1192 : lastFullPageWrites = checkPoint.fullPageWrites;
6738 :
6739 1192 : RedoRecPtr = XLogCtl->RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
6740 1192 : doPageWrites = lastFullPageWrites;
6741 :
6742 1192 : if (RecPtr < checkPoint.redo)
6743 0 : ereport(PANIC,
6744 : (errmsg("invalid redo in checkpoint record")));
6745 :
6746 : /*
6747 : * Check whether we need to force recovery from WAL. If it appears to
6748 : * have been a clean shutdown and we did not have a recovery signal file,
6749 : * then assume no recovery needed.
6750 : */
6751 1192 : if (checkPoint.redo < RecPtr)
6752 : {
6753 82 : if (wasShutdown)
6754 0 : ereport(PANIC,
6755 : (errmsg("invalid redo record in shutdown checkpoint")));
6756 82 : InRecovery = true;
6757 : }
6758 1110 : else if (ControlFile->state != DB_SHUTDOWNED)
6759 30 : InRecovery = true;
6760 1080 : else if (ArchiveRecoveryRequested)
6761 : {
6762 : /* force recovery due to presence of recovery signal file */
6763 4 : InRecovery = true;
6764 : }
6765 :
6766 : /* REDO */
6767 1192 : if (InRecovery)
6768 : {
6769 : int rmid;
6770 :
6771 : /*
6772 : * Update pg_control to show that we are recovering and to show the
6773 : * selected checkpoint as the place we are starting from. We also mark
6774 : * pg_control with any minimum recovery stop point obtained from a
6775 : * backup history file.
6776 : */
6777 116 : dbstate_at_startup = ControlFile->state;
6778 116 : if (InArchiveRecovery)
6779 66 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
6780 : else
6781 : {
6782 50 : ereport(LOG,
6783 : (errmsg("database system was not properly shut down; "
6784 : "automatic recovery in progress")));
6785 50 : if (recoveryTargetTLI > ControlFile->checkPointCopy.ThisTimeLineID)
6786 2 : ereport(LOG,
6787 : (errmsg("crash recovery starts in timeline %u "
6788 : "and has target timeline %u",
6789 : ControlFile->checkPointCopy.ThisTimeLineID,
6790 : recoveryTargetTLI)));
6791 50 : ControlFile->state = DB_IN_CRASH_RECOVERY;
6792 : }
6793 116 : ControlFile->checkPoint = checkPointLoc;
6794 116 : ControlFile->checkPointCopy = checkPoint;
6795 116 : if (InArchiveRecovery)
6796 : {
6797 : /* initialize minRecoveryPoint if not set yet */
6798 66 : if (ControlFile->minRecoveryPoint < checkPoint.redo)
6799 : {
6800 56 : ControlFile->minRecoveryPoint = checkPoint.redo;
6801 56 : ControlFile->minRecoveryPointTLI = checkPoint.ThisTimeLineID;
6802 : }
6803 : }
6804 :
6805 : /*
6806 : * Set backupStartPoint if we're starting recovery from a base backup.
6807 : *
6808 : * Also set backupEndPoint and use minRecoveryPoint as the backup end
6809 : * location if we're starting recovery from a base backup which was
6810 : * taken from a standby. In this case, the database system status in
6811 : * pg_control must indicate that the database was already in recovery.
6812 : * Usually that will be DB_IN_ARCHIVE_RECOVERY but also can be
6813 : * DB_SHUTDOWNED_IN_RECOVERY if recovery previously was interrupted
6814 : * before reaching this point; e.g. because restore_command or
6815 : * primary_conninfo were faulty.
6816 : *
6817 : * Any other state indicates that the backup somehow became corrupted
6818 : * and we can't sensibly continue with recovery.
6819 : */
6820 116 : if (haveBackupLabel)
6821 : {
6822 54 : ControlFile->backupStartPoint = checkPoint.redo;
6823 54 : ControlFile->backupEndRequired = backupEndRequired;
6824 :
6825 54 : if (backupFromStandby)
6826 : {
6827 2 : if (dbstate_at_startup != DB_IN_ARCHIVE_RECOVERY &&
6828 : dbstate_at_startup != DB_SHUTDOWNED_IN_RECOVERY)
6829 0 : ereport(FATAL,
6830 : (errmsg("backup_label contains data inconsistent with control file"),
6831 : errhint("This means that the backup is corrupted and you will "
6832 : "have to use another backup for recovery.")));
6833 2 : ControlFile->backupEndPoint = ControlFile->minRecoveryPoint;
6834 : }
6835 : }
6836 116 : ControlFile->time = (pg_time_t) time(NULL);
6837 : /* No need to hold ControlFileLock yet, we aren't up far enough */
6838 116 : UpdateControlFile();
6839 :
6840 : /*
6841 : * Initialize our local copy of minRecoveryPoint. When doing crash
6842 : * recovery we want to replay up to the end of WAL. Particularly, in
6843 : * the case of a promoted standby minRecoveryPoint value in the
6844 : * control file is only updated after the first checkpoint. However,
6845 : * if the instance crashes before the first post-recovery checkpoint
6846 : * is completed then recovery will use a stale location causing the
6847 : * startup process to think that there are still invalid page
6848 : * references when checking for data consistency.
6849 : */
6850 116 : if (InArchiveRecovery)
6851 : {
6852 66 : minRecoveryPoint = ControlFile->minRecoveryPoint;
6853 66 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
6854 : }
6855 : else
6856 : {
6857 50 : minRecoveryPoint = InvalidXLogRecPtr;
6858 50 : minRecoveryPointTLI = 0;
6859 : }
6860 :
6861 : /*
6862 : * Reset pgstat data, because it may be invalid after recovery.
6863 : */
6864 116 : pgstat_reset_all();
6865 :
6866 : /*
6867 : * If there was a backup label file, it's done its job and the info
6868 : * has now been propagated into pg_control. We must get rid of the
6869 : * label file so that if we crash during recovery, we'll pick up at
6870 : * the latest recovery restartpoint instead of going all the way back
6871 : * to the backup start point. It seems prudent though to just rename
6872 : * the file out of the way rather than delete it completely.
6873 : */
6874 116 : if (haveBackupLabel)
6875 : {
6876 54 : unlink(BACKUP_LABEL_OLD);
6877 54 : durable_rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, FATAL);
6878 : }
6879 :
6880 : /*
6881 : * If there was a tablespace_map file, it's done its job and the
6882 : * symlinks have been created. We must get rid of the map file so
6883 : * that if we crash during recovery, we don't create symlinks again.
6884 : * It seems prudent though to just rename the file out of the way
6885 : * rather than delete it completely.
6886 : */
6887 116 : if (haveTblspcMap)
6888 : {
6889 0 : unlink(TABLESPACE_MAP_OLD);
6890 0 : durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, FATAL);
6891 : }
6892 :
6893 : /* Check that the GUCs used to generate the WAL allow recovery */
6894 116 : CheckRequiredParameterValues();
6895 :
6896 : /*
6897 : * We're in recovery, so unlogged relations may be trashed and must be
6898 : * reset. This should be done BEFORE allowing Hot Standby
6899 : * connections, so that read-only backends don't try to read whatever
6900 : * garbage is left over from before.
6901 : */
6902 116 : ResetUnloggedRelations(UNLOGGED_RELATION_CLEANUP);
6903 :
6904 : /*
6905 : * Likewise, delete any saved transaction snapshot files that got left
6906 : * behind by crashed backends.
6907 : */
6908 116 : DeleteAllExportedSnapshotFiles();
6909 :
6910 : /*
6911 : * Initialize for Hot Standby, if enabled. We won't let backends in
6912 : * yet, not until we've reached the min recovery point specified in
6913 : * control file and we've established a recovery snapshot from a
6914 : * running-xacts WAL record.
6915 : */
6916 116 : if (ArchiveRecoveryRequested && EnableHotStandby)
6917 : {
6918 : TransactionId *xids;
6919 : int nxids;
6920 :
6921 68 : ereport(DEBUG1,
6922 : (errmsg("initializing for hot standby")));
6923 :
6924 68 : InitRecoveryTransactionEnvironment();
6925 :
6926 68 : if (wasShutdown)
6927 10 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
6928 : else
6929 58 : oldestActiveXID = checkPoint.oldestActiveXid;
6930 : Assert(TransactionIdIsValid(oldestActiveXID));
6931 :
6932 : /* Tell procarray about the range of xids it has to deal with */
6933 68 : ProcArrayInitRecovery(XidFromFullTransactionId(ShmemVariableCache->nextFullXid));
6934 :
6935 : /*
6936 : * Startup commit log and subtrans only. MultiXact and commit
6937 : * timestamp have already been started up and other SLRUs are not
6938 : * maintained during recovery and need not be started yet.
6939 : */
6940 68 : StartupCLOG();
6941 68 : StartupSUBTRANS(oldestActiveXID);
6942 :
6943 : /*
6944 : * If we're beginning at a shutdown checkpoint, we know that
6945 : * nothing was running on the master at this point. So fake-up an
6946 : * empty running-xacts record and use that here and now. Recover
6947 : * additional standby state for prepared transactions.
6948 : */
6949 68 : if (wasShutdown)
6950 : {
6951 : RunningTransactionsData running;
6952 : TransactionId latestCompletedXid;
6953 :
6954 : /*
6955 : * Construct a RunningTransactions snapshot representing a
6956 : * shut down server, with only prepared transactions still
6957 : * alive. We're never overflowed at this point because all
6958 : * subxids are listed with their parent prepared transactions.
6959 : */
6960 10 : running.xcnt = nxids;
6961 10 : running.subxcnt = 0;
6962 10 : running.subxid_overflow = false;
6963 10 : running.nextXid = XidFromFullTransactionId(checkPoint.nextFullXid);
6964 10 : running.oldestRunningXid = oldestActiveXID;
6965 10 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextFullXid);
6966 10 : TransactionIdRetreat(latestCompletedXid);
6967 : Assert(TransactionIdIsNormal(latestCompletedXid));
6968 10 : running.latestCompletedXid = latestCompletedXid;
6969 10 : running.xids = xids;
6970 :
6971 10 : ProcArrayApplyRecoveryInfo(&running);
6972 :
6973 10 : StandbyRecoverPreparedTransactions();
6974 : }
6975 : }
6976 :
6977 : /* Initialize resource managers */
6978 2668 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
6979 : {
6980 2552 : if (RmgrTable[rmid].rm_startup != NULL)
6981 348 : RmgrTable[rmid].rm_startup();
6982 : }
6983 :
6984 : /*
6985 : * Initialize shared variables for tracking progress of WAL replay, as
6986 : * if we had just replayed the record before the REDO location (or the
6987 : * checkpoint record itself, if it's a shutdown checkpoint).
6988 : */
6989 116 : SpinLockAcquire(&XLogCtl->info_lck);
6990 116 : if (checkPoint.redo < RecPtr)
6991 82 : XLogCtl->replayEndRecPtr = checkPoint.redo;
6992 : else
6993 34 : XLogCtl->replayEndRecPtr = EndRecPtr;
6994 116 : XLogCtl->replayEndTLI = ThisTimeLineID;
6995 116 : XLogCtl->lastReplayedEndRecPtr = XLogCtl->replayEndRecPtr;
6996 116 : XLogCtl->lastReplayedTLI = XLogCtl->replayEndTLI;
6997 116 : XLogCtl->recoveryLastXTime = 0;
6998 116 : XLogCtl->currentChunkStartTime = 0;
6999 116 : XLogCtl->recoveryPause = false;
7000 116 : SpinLockRelease(&XLogCtl->info_lck);
7001 :
7002 : /* Also ensure XLogReceiptTime has a sane value */
7003 116 : XLogReceiptTime = GetCurrentTimestamp();
7004 :
7005 : /*
7006 : * Let postmaster know we've started redo now, so that it can launch
7007 : * checkpointer to perform restartpoints. We don't bother during
7008 : * crash recovery as restartpoints can only be performed during
7009 : * archive recovery. And we'd like to keep crash recovery simple, to
7010 : * avoid introducing bugs that could affect you when recovering after
7011 : * crash.
7012 : *
7013 : * After this point, we can no longer assume that we're the only
7014 : * process in addition to postmaster! Also, fsync requests are
7015 : * subsequently to be handled by the checkpointer, not locally.
7016 : */
7017 116 : if (ArchiveRecoveryRequested && IsUnderPostmaster)
7018 : {
7019 68 : PublishStartupProcessInformation();
7020 68 : EnableSyncRequestForwarding();
7021 68 : SendPostmasterSignal(PMSIGNAL_RECOVERY_STARTED);
7022 68 : bgwriterLaunched = true;
7023 : }
7024 :
7025 : /*
7026 : * Allow read-only connections immediately if we're consistent
7027 : * already.
7028 : */
7029 116 : CheckRecoveryConsistency();
7030 :
7031 : /*
7032 : * Find the first record that logically follows the checkpoint --- it
7033 : * might physically precede it, though.
7034 : */
7035 116 : if (checkPoint.redo < RecPtr)
7036 : {
7037 : /* back up to find the record */
7038 82 : record = ReadRecord(xlogreader, checkPoint.redo, PANIC, false);
7039 : }
7040 : else
7041 : {
7042 : /* just have to read next record after CheckPoint */
7043 34 : record = ReadRecord(xlogreader, InvalidXLogRecPtr, LOG, false);
7044 : }
7045 :
7046 116 : if (record != NULL)
7047 : {
7048 : ErrorContextCallback errcallback;
7049 : TimestampTz xtime;
7050 :
7051 108 : InRedo = true;
7052 :
7053 108 : ereport(LOG,
7054 : (errmsg("redo starts at %X/%X",
7055 : (uint32) (ReadRecPtr >> 32), (uint32) ReadRecPtr)));
7056 :
7057 : /*
7058 : * main redo apply loop
7059 : */
7060 : do
7061 : {
7062 238472 : bool switchedTLI = false;
7063 :
7064 : #ifdef WAL_DEBUG
7065 : if (XLOG_DEBUG ||
7066 : (rmid == RM_XACT_ID && trace_recovery_messages <= DEBUG2) ||
7067 : (rmid != RM_XACT_ID && trace_recovery_messages <= DEBUG3))
7068 : {
7069 : StringInfoData buf;
7070 :
7071 : initStringInfo(&buf);
7072 : appendStringInfo(&buf, "REDO @ %X/%X; LSN %X/%X: ",
7073 : (uint32) (ReadRecPtr >> 32), (uint32) ReadRecPtr,
7074 : (uint32) (EndRecPtr >> 32), (uint32) EndRecPtr);
7075 : xlog_outrec(&buf, xlogreader);
7076 : appendStringInfoString(&buf, " - ");
7077 : xlog_outdesc(&buf, xlogreader);
7078 : elog(LOG, "%s", buf.data);
7079 : pfree(buf.data);
7080 : }
7081 : #endif
7082 :
7083 : /* Handle interrupt signals of startup process */
7084 238472 : HandleStartupProcInterrupts();
7085 :
7086 : /*
7087 : * Pause WAL replay, if requested by a hot-standby session via
7088 : * SetRecoveryPause().
7089 : *
7090 : * Note that we intentionally don't take the info_lck spinlock
7091 : * here. We might therefore read a slightly stale value of
7092 : * the recoveryPause flag, but it can't be very stale (no
7093 : * worse than the last spinlock we did acquire). Since a
7094 : * pause request is a pretty asynchronous thing anyway,
7095 : * possibly responding to it one WAL record later than we
7096 : * otherwise would is a minor issue, so it doesn't seem worth
7097 : * adding another spinlock cycle to prevent that.
7098 : */
7099 238472 : if (((volatile XLogCtlData *) XLogCtl)->recoveryPause)
7100 0 : recoveryPausesHere();
7101 :
7102 : /*
7103 : * Have we reached our recovery target?
7104 : */
7105 238472 : if (recoveryStopsBefore(xlogreader))
7106 : {
7107 0 : reachedStopPoint = true; /* see below */
7108 0 : break;
7109 : }
7110 :
7111 : /*
7112 : * If we've been asked to lag the master, wait on latch until
7113 : * enough time has passed.
7114 : */
7115 238472 : if (recoveryApplyDelay(xlogreader))
7116 : {
7117 : /*
7118 : * We test for paused recovery again here. If user sets
7119 : * delayed apply, it may be because they expect to pause
7120 : * recovery in case of problems, so we must test again
7121 : * here otherwise pausing during the delay-wait wouldn't
7122 : * work.
7123 : */
7124 0 : if (((volatile XLogCtlData *) XLogCtl)->recoveryPause)
7125 0 : recoveryPausesHere();
7126 : }
7127 :
7128 : /* Setup error traceback support for ereport() */
7129 238472 : errcallback.callback = rm_redo_error_callback;
7130 238472 : errcallback.arg = (void *) xlogreader;
7131 238472 : errcallback.previous = error_context_stack;
7132 238472 : error_context_stack = &errcallback;
7133 :
7134 : /*
7135 : * ShmemVariableCache->nextFullXid must be beyond record's
7136 : * xid.
7137 : */
7138 238472 : AdvanceNextFullTransactionIdPastXid(record->xl_xid);
7139 :
7140 : /*
7141 : * Before replaying this record, check if this record causes
7142 : * the current timeline to change. The record is already
7143 : * considered to be part of the new timeline, so we update
7144 : * ThisTimeLineID before replaying it. That's important so
7145 : * that replayEndTLI, which is recorded as the minimum
7146 : * recovery point's TLI if recovery stops after this record,
7147 : * is set correctly.
7148 : */
7149 238472 : if (record->xl_rmid == RM_XLOG_ID)
7150 : {
7151 1602 : TimeLineID newTLI = ThisTimeLineID;
7152 1602 : TimeLineID prevTLI = ThisTimeLineID;
7153 1602 : uint8 info = record->xl_info & ~XLR_INFO_MASK;
7154 :
7155 1602 : if (info == XLOG_CHECKPOINT_SHUTDOWN)
7156 : {
7157 : CheckPoint checkPoint;
7158 :
7159 34 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
7160 34 : newTLI = checkPoint.ThisTimeLineID;
7161 34 : prevTLI = checkPoint.PrevTimeLineID;
7162 : }
7163 1568 : else if (info == XLOG_END_OF_RECOVERY)
7164 : {
7165 : xl_end_of_recovery xlrec;
7166 :
7167 10 : memcpy(&xlrec, XLogRecGetData(xlogreader), sizeof(xl_end_of_recovery));
7168 10 : newTLI = xlrec.ThisTimeLineID;
7169 10 : prevTLI = xlrec.PrevTimeLineID;
7170 : }
7171 :
7172 1602 : if (newTLI != ThisTimeLineID)
7173 : {
7174 : /* Check that it's OK to switch to this TLI */
7175 10 : checkTimeLineSwitch(EndRecPtr, newTLI, prevTLI);
7176 :
7177 : /* Following WAL records should be run with new TLI */
7178 10 : ThisTimeLineID = newTLI;
7179 10 : switchedTLI = true;
7180 : }
7181 : }
7182 :
7183 : /*
7184 : * Update shared replayEndRecPtr before replaying this record,
7185 : * so that XLogFlush will update minRecoveryPoint correctly.
7186 : */
7187 238472 : SpinLockAcquire(&XLogCtl->info_lck);
7188 238472 : XLogCtl->replayEndRecPtr = EndRecPtr;
7189 238472 : XLogCtl->replayEndTLI = ThisTimeLineID;
7190 238472 : SpinLockRelease(&XLogCtl->info_lck);
7191 :
7192 : /*
7193 : * If we are attempting to enter Hot Standby mode, process
7194 : * XIDs we see
7195 : */
7196 408372 : if (standbyState >= STANDBY_INITIALIZED &&
7197 169900 : TransactionIdIsValid(record->xl_xid))
7198 168986 : RecordKnownAssignedTransactionIds(record->xl_xid);
7199 :
7200 : /* Now apply the WAL record itself */
7201 238472 : RmgrTable[record->xl_rmid].rm_redo(xlogreader);
7202 :
7203 : /*
7204 : * After redo, check whether the backup pages associated with
7205 : * the WAL record are consistent with the existing pages. This
7206 : * check is done only if consistency check is enabled for this
7207 : * record.
7208 : */
7209 238472 : if ((record->xl_info & XLR_CHECK_CONSISTENCY) != 0)
7210 0 : checkXLogConsistency(xlogreader);
7211 :
7212 : /* Pop the error context stack */
7213 238472 : error_context_stack = errcallback.previous;
7214 :
7215 : /*
7216 : * Update lastReplayedEndRecPtr after this record has been
7217 : * successfully replayed.
7218 : */
7219 238472 : SpinLockAcquire(&XLogCtl->info_lck);
7220 238472 : XLogCtl->lastReplayedEndRecPtr = EndRecPtr;
7221 238472 : XLogCtl->lastReplayedTLI = ThisTimeLineID;
7222 238472 : SpinLockRelease(&XLogCtl->info_lck);
7223 :
7224 : /*
7225 : * If rm_redo called XLogRequestWalReceiverReply, then we wake
7226 : * up the receiver so that it notices the updated
7227 : * lastReplayedEndRecPtr and sends a reply to the master.
7228 : */
7229 238472 : if (doRequestWalReceiverReply)
7230 : {
7231 0 : doRequestWalReceiverReply = false;
7232 0 : WalRcvForceReply();
7233 : }
7234 :
7235 : /* Remember this record as the last-applied one */
7236 238472 : LastRec = ReadRecPtr;
7237 :
7238 : /* Allow read-only connections if we're consistent now */
7239 238472 : CheckRecoveryConsistency();
7240 :
7241 : /* Is this a timeline switch? */
7242 238472 : if (switchedTLI)
7243 : {
7244 : /*
7245 : * Before we continue on the new timeline, clean up any
7246 : * (possibly bogus) future WAL segments on the old
7247 : * timeline.
7248 : */
7249 10 : RemoveNonParentXlogFiles(EndRecPtr, ThisTimeLineID);
7250 :
7251 : /*
7252 : * Wake up any walsenders to notice that we are on a new
7253 : * timeline.
7254 : */
7255 10 : if (switchedTLI && AllowCascadeReplication())
7256 10 : WalSndWakeup();
7257 : }
7258 :
7259 : /* Exit loop if we reached inclusive recovery target */
7260 238472 : if (recoveryStopsAfter(xlogreader))
7261 : {
7262 0 : reachedStopPoint = true;
7263 0 : break;
7264 : }
7265 :
7266 : /* Else, try to fetch the next WAL record */
7267 238472 : record = ReadRecord(xlogreader, InvalidXLogRecPtr, LOG, false);
7268 238450 : } while (record != NULL);
7269 :
7270 : /*
7271 : * end of main redo apply loop
7272 : */
7273 :
7274 86 : if (reachedStopPoint)
7275 : {
7276 0 : if (!reachedConsistency)
7277 0 : ereport(FATAL,
7278 : (errmsg("requested recovery stop point is before consistent recovery point")));
7279 :
7280 : /*
7281 : * This is the last point where we can restart recovery with a
7282 : * new recovery target, if we shutdown and begin again. After
7283 : * this, Resource Managers may choose to do permanent
7284 : * corrective actions at end of recovery.
7285 : */
7286 0 : switch (recoveryTargetAction)
7287 : {
7288 : case RECOVERY_TARGET_ACTION_SHUTDOWN:
7289 :
7290 : /*
7291 : * exit with special return code to request shutdown
7292 : * of postmaster. Log messages issued from
7293 : * postmaster.
7294 : */
7295 0 : proc_exit(3);
7296 :
7297 : case RECOVERY_TARGET_ACTION_PAUSE:
7298 0 : SetRecoveryPause(true);
7299 0 : recoveryPausesHere();
7300 :
7301 : /* drop into promote */
7302 :
7303 : case RECOVERY_TARGET_ACTION_PROMOTE:
7304 0 : break;
7305 : }
7306 : }
7307 :
7308 : /* Allow resource managers to do any required cleanup. */
7309 1978 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
7310 : {
7311 1892 : if (RmgrTable[rmid].rm_cleanup != NULL)
7312 258 : RmgrTable[rmid].rm_cleanup();
7313 : }
7314 :
7315 86 : ereport(LOG,
7316 : (errmsg("redo done at %X/%X",
7317 : (uint32) (ReadRecPtr >> 32), (uint32) ReadRecPtr)));
7318 86 : xtime = GetLatestXTime();
7319 86 : if (xtime)
7320 34 : ereport(LOG,
7321 : (errmsg("last completed transaction was at log time %s",
7322 : timestamptz_to_str(xtime))));
7323 :
7324 86 : InRedo = false;
7325 : }
7326 : else
7327 : {
7328 : /* there are no WAL records following the checkpoint */
7329 8 : ereport(LOG,
7330 : (errmsg("redo is not required")));
7331 : }
7332 : }
7333 :
7334 : /*
7335 : * Kill WAL receiver, if it's still running, before we continue to write
7336 : * the startup checkpoint record. It will trump over the checkpoint and
7337 : * subsequent records if it's still alive when we start writing WAL.
7338 : */
7339 1170 : ShutdownWalRcv();
7340 :
7341 : /*
7342 : * Reset unlogged relations to the contents of their INIT fork. This is
7343 : * done AFTER recovery is complete so as to include any unlogged relations
7344 : * created during recovery, but BEFORE recovery is marked as having
7345 : * completed successfully. Otherwise we'd not retry if any of the post
7346 : * end-of-recovery steps fail.
7347 : */
7348 1170 : if (InRecovery)
7349 94 : ResetUnloggedRelations(UNLOGGED_RELATION_INIT);
7350 :
7351 : /*
7352 : * We don't need the latch anymore. It's not strictly necessary to disown
7353 : * it, but let's do it for the sake of tidiness.
7354 : */
7355 1170 : if (ArchiveRecoveryRequested)
7356 46 : DisownLatch(&XLogCtl->recoveryWakeupLatch);
7357 :
7358 : /*
7359 : * We are now done reading the xlog from stream. Turn off streaming
7360 : * recovery to force fetching the files (which would be required at end of
7361 : * recovery, e.g., timeline history file) from archive or pg_wal.
7362 : */
7363 1170 : StandbyMode = false;
7364 :
7365 : /*
7366 : * Re-fetch the last valid or last applied record, so we can identify the
7367 : * exact endpoint of what we consider the valid portion of WAL.
7368 : */
7369 1170 : record = ReadRecord(xlogreader, LastRec, PANIC, false);
7370 1170 : EndOfLog = EndRecPtr;
7371 :
7372 : /*
7373 : * EndOfLogTLI is the TLI in the filename of the XLOG segment containing
7374 : * the end-of-log. It could be different from the timeline that EndOfLog
7375 : * nominally belongs to, if there was a timeline switch in that segment,
7376 : * and we were reading the old WAL from a segment belonging to a higher
7377 : * timeline.
7378 : */
7379 1170 : EndOfLogTLI = xlogreader->seg.ws_tli;
7380 :
7381 : /*
7382 : * Complain if we did not roll forward far enough to render the backup
7383 : * dump consistent. Note: it is indeed okay to look at the local variable
7384 : * minRecoveryPoint here, even though ControlFile->minRecoveryPoint might
7385 : * be further ahead --- ControlFile->minRecoveryPoint cannot have been
7386 : * advanced beyond the WAL we processed.
7387 : */
7388 1264 : if (InRecovery &&
7389 188 : (EndOfLog < minRecoveryPoint ||
7390 94 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint)))
7391 : {
7392 : /*
7393 : * Ran off end of WAL before reaching end-of-backup WAL record, or
7394 : * minRecoveryPoint. That's usually a bad sign, indicating that you
7395 : * tried to recover from an online backup but never called
7396 : * pg_stop_backup(), or you didn't archive all the WAL up to that
7397 : * point. However, this also happens in crash recovery, if the system
7398 : * crashes while an online backup is in progress. We must not treat
7399 : * that as an error, or the database will refuse to start up.
7400 : */
7401 0 : if (ArchiveRecoveryRequested || ControlFile->backupEndRequired)
7402 : {
7403 0 : if (ControlFile->backupEndRequired)
7404 0 : ereport(FATAL,
7405 : (errmsg("WAL ends before end of online backup"),
7406 : errhint("All WAL generated while online backup was taken must be available at recovery.")));
7407 0 : else if (!XLogRecPtrIsInvalid(ControlFile->backupStartPoint))
7408 0 : ereport(FATAL,
7409 : (errmsg("WAL ends before end of online backup"),
7410 : 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.")));
7411 : else
7412 0 : ereport(FATAL,
7413 : (errmsg("WAL ends before consistent recovery point")));
7414 : }
7415 : }
7416 :
7417 : /*
7418 : * Pre-scan prepared transactions to find out the range of XIDs present.
7419 : * This information is not quite needed yet, but it is positioned here so
7420 : * as potential problems are detected before any on-disk change is done.
7421 : */
7422 1170 : oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
7423 :
7424 : /*
7425 : * Consider whether we need to assign a new timeline ID.
7426 : *
7427 : * If we are doing an archive recovery, we always assign a new ID. This
7428 : * handles a couple of issues. If we stopped short of the end of WAL
7429 : * during recovery, then we are clearly generating a new timeline and must
7430 : * assign it a unique new ID. Even if we ran to the end, modifying the
7431 : * current last segment is problematic because it may result in trying to
7432 : * overwrite an already-archived copy of that segment, and we encourage
7433 : * DBAs to make their archive_commands reject that. We can dodge the
7434 : * problem by making the new active segment have a new timeline ID.
7435 : *
7436 : * In a normal crash recovery, we can just extend the timeline we were in.
7437 : */
7438 1170 : PrevTimeLineID = ThisTimeLineID;
7439 1170 : if (ArchiveRecoveryRequested)
7440 : {
7441 : char reason[200];
7442 : char recoveryPath[MAXPGPATH];
7443 :
7444 : Assert(InArchiveRecovery);
7445 :
7446 46 : ThisTimeLineID = findNewestTimeLine(recoveryTargetTLI) + 1;
7447 46 : ereport(LOG,
7448 : (errmsg("selected new timeline ID: %u", ThisTimeLineID)));
7449 :
7450 : /*
7451 : * Create a comment for the history file to explain why and where
7452 : * timeline changed.
7453 : */
7454 46 : if (recoveryTarget == RECOVERY_TARGET_XID)
7455 0 : snprintf(reason, sizeof(reason),
7456 : "%s transaction %u",
7457 0 : recoveryStopAfter ? "after" : "before",
7458 : recoveryStopXid);
7459 46 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
7460 0 : snprintf(reason, sizeof(reason),
7461 : "%s %s\n",
7462 0 : recoveryStopAfter ? "after" : "before",
7463 : timestamptz_to_str(recoveryStopTime));
7464 46 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
7465 0 : snprintf(reason, sizeof(reason),
7466 : "%s LSN %X/%X\n",
7467 0 : recoveryStopAfter ? "after" : "before",
7468 0 : (uint32) (recoveryStopLSN >> 32),
7469 : (uint32) recoveryStopLSN);
7470 46 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
7471 0 : snprintf(reason, sizeof(reason),
7472 : "at restore point \"%s\"",
7473 : recoveryStopName);
7474 46 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
7475 0 : snprintf(reason, sizeof(reason), "reached consistency");
7476 : else
7477 46 : snprintf(reason, sizeof(reason), "no recovery target specified");
7478 :
7479 : /*
7480 : * We are now done reading the old WAL. Turn off archive fetching if
7481 : * it was active, and make a writable copy of the last WAL segment.
7482 : * (Note that we also have a copy of the last block of the old WAL in
7483 : * readBuf; we will use that below.)
7484 : */
7485 46 : exitArchiveRecovery(EndOfLogTLI, EndOfLog);
7486 :
7487 : /*
7488 : * Write the timeline history file, and have it archived. After this
7489 : * point (or rather, as soon as the file is archived), the timeline
7490 : * will appear as "taken" in the WAL archive and to any standby
7491 : * servers. If we crash before actually switching to the new
7492 : * timeline, standby servers will nevertheless think that we switched
7493 : * to the new timeline, and will try to connect to the new timeline.
7494 : * To minimize the window for that, try to do as little as possible
7495 : * between here and writing the end-of-recovery record.
7496 : */
7497 46 : writeTimeLineHistory(ThisTimeLineID, recoveryTargetTLI,
7498 : EndRecPtr, reason);
7499 :
7500 : /*
7501 : * Since there might be a partial WAL segment named RECOVERYXLOG, get
7502 : * rid of it.
7503 : */
7504 46 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYXLOG");
7505 46 : unlink(recoveryPath); /* ignore any error */
7506 :
7507 : /* Get rid of any remaining recovered timeline-history file, too */
7508 46 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYHISTORY");
7509 46 : unlink(recoveryPath); /* ignore any error */
7510 : }
7511 :
7512 : /* Save the selected TimeLineID in shared memory, too */
7513 1170 : XLogCtl->ThisTimeLineID = ThisTimeLineID;
7514 1170 : XLogCtl->PrevTimeLineID = PrevTimeLineID;
7515 :
7516 : /*
7517 : * Prepare to write WAL starting at EndOfLog location, and init xlog
7518 : * buffer cache using the block containing the last record from the
7519 : * previous incarnation.
7520 : */
7521 1170 : Insert = &XLogCtl->Insert;
7522 1170 : Insert->PrevBytePos = XLogRecPtrToBytePos(LastRec);
7523 1170 : Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
7524 :
7525 : /*
7526 : * Tricky point here: readBuf contains the *last* block that the LastRec
7527 : * record spans, not the one it starts in. The last block is indeed the
7528 : * one we want to use.
7529 : */
7530 1170 : if (EndOfLog % XLOG_BLCKSZ != 0)
7531 : {
7532 : char *page;
7533 : int len;
7534 : int firstIdx;
7535 : XLogRecPtr pageBeginPtr;
7536 :
7537 1152 : pageBeginPtr = EndOfLog - (EndOfLog % XLOG_BLCKSZ);
7538 : Assert(readOff == XLogSegmentOffset(pageBeginPtr, wal_segment_size));
7539 :
7540 1152 : firstIdx = XLogRecPtrToBufIdx(EndOfLog);
7541 :
7542 : /* Copy the valid part of the last block, and zero the rest */
7543 1152 : page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
7544 1152 : len = EndOfLog % XLOG_BLCKSZ;
7545 1152 : memcpy(page, xlogreader->readBuf, len);
7546 1152 : memset(page + len, 0, XLOG_BLCKSZ - len);
7547 :
7548 1152 : XLogCtl->xlblocks[firstIdx] = pageBeginPtr + XLOG_BLCKSZ;
7549 1152 : XLogCtl->InitializedUpTo = pageBeginPtr + XLOG_BLCKSZ;
7550 : }
7551 : else
7552 : {
7553 : /*
7554 : * There is no partial block to copy. Just set InitializedUpTo, and
7555 : * let the first attempt to insert a log record to initialize the next
7556 : * buffer.
7557 : */
7558 18 : XLogCtl->InitializedUpTo = EndOfLog;
7559 : }
7560 :
7561 1170 : LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
7562 :
7563 1170 : XLogCtl->LogwrtResult = LogwrtResult;
7564 :
7565 1170 : XLogCtl->LogwrtRqst.Write = EndOfLog;
7566 1170 : XLogCtl->LogwrtRqst.Flush = EndOfLog;
7567 :
7568 : /*
7569 : * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
7570 : * record before resource manager writes cleanup WAL records or checkpoint
7571 : * record is written.
7572 : */
7573 1170 : Insert->fullPageWrites = lastFullPageWrites;
7574 1170 : LocalSetXLogInsertAllowed();
7575 1170 : UpdateFullPageWrites();
7576 1170 : LocalXLogInsertAllowed = -1;
7577 :
7578 1170 : if (InRecovery)
7579 : {
7580 : /*
7581 : * Perform a checkpoint to update all our recovery activity to disk.
7582 : *
7583 : * Note that we write a shutdown checkpoint rather than an on-line
7584 : * one. This is not particularly critical, but since we may be
7585 : * assigning a new TLI, using a shutdown checkpoint allows us to have
7586 : * the rule that TLI only changes in shutdown checkpoints, which
7587 : * allows some extra error checking in xlog_redo.
7588 : *
7589 : * In fast promotion, only create a lightweight end-of-recovery record
7590 : * instead of a full checkpoint. A checkpoint is requested later,
7591 : * after we're fully out of recovery mode and already accepting
7592 : * queries.
7593 : */
7594 94 : if (bgwriterLaunched)
7595 : {
7596 46 : if (fast_promote)
7597 : {
7598 46 : checkPointLoc = ControlFile->checkPoint;
7599 :
7600 : /*
7601 : * Confirm the last checkpoint is available for us to recover
7602 : * from if we fail.
7603 : */
7604 46 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 1, false);
7605 46 : if (record != NULL)
7606 : {
7607 46 : fast_promoted = true;
7608 :
7609 : /*
7610 : * Insert a special WAL record to mark the end of
7611 : * recovery, since we aren't doing a checkpoint. That
7612 : * means that the checkpointer process may likely be in
7613 : * the middle of a time-smoothed restartpoint and could
7614 : * continue to be for minutes after this. That sounds
7615 : * strange, but the effect is roughly the same and it
7616 : * would be stranger to try to come out of the
7617 : * restartpoint and then checkpoint. We request a
7618 : * checkpoint later anyway, just for safety.
7619 : */
7620 46 : CreateEndOfRecoveryRecord();
7621 : }
7622 : }
7623 :
7624 46 : if (!fast_promoted)
7625 0 : RequestCheckpoint(CHECKPOINT_END_OF_RECOVERY |
7626 : CHECKPOINT_IMMEDIATE |
7627 : CHECKPOINT_WAIT);
7628 : }
7629 : else
7630 48 : CreateCheckPoint(CHECKPOINT_END_OF_RECOVERY | CHECKPOINT_IMMEDIATE);
7631 : }
7632 :
7633 1170 : if (ArchiveRecoveryRequested)
7634 : {
7635 : /*
7636 : * And finally, execute the recovery_end_command, if any.
7637 : */
7638 46 : if (recoveryEndCommand && strcmp(recoveryEndCommand, "") != 0)
7639 0 : ExecuteRecoveryCommand(recoveryEndCommand,
7640 : "recovery_end_command",
7641 : true);
7642 :
7643 : /*
7644 : * We switched to a new timeline. Clean up segments on the old
7645 : * timeline.
7646 : *
7647 : * If there are any higher-numbered segments on the old timeline,
7648 : * remove them. They might contain valid WAL, but they might also be
7649 : * pre-allocated files containing garbage. In any case, they are not
7650 : * part of the new timeline's history so we don't need them.
7651 : */
7652 46 : RemoveNonParentXlogFiles(EndOfLog, ThisTimeLineID);
7653 :
7654 : /*
7655 : * If the switch happened in the middle of a segment, what to do with
7656 : * the last, partial segment on the old timeline? If we don't archive
7657 : * it, and the server that created the WAL never archives it either
7658 : * (e.g. because it was hit by a meteor), it will never make it to the
7659 : * archive. That's OK from our point of view, because the new segment
7660 : * that we created with the new TLI contains all the WAL from the old
7661 : * timeline up to the switch point. But if you later try to do PITR to
7662 : * the "missing" WAL on the old timeline, recovery won't find it in
7663 : * the archive. It's physically present in the new file with new TLI,
7664 : * but recovery won't look there when it's recovering to the older
7665 : * timeline. On the other hand, if we archive the partial segment, and
7666 : * the original server on that timeline is still running and archives
7667 : * the completed version of the same segment later, it will fail. (We
7668 : * used to do that in 9.4 and below, and it caused such problems).
7669 : *
7670 : * As a compromise, we rename the last segment with the .partial
7671 : * suffix, and archive it. Archive recovery will never try to read
7672 : * .partial segments, so they will normally go unused. But in the odd
7673 : * PITR case, the administrator can copy them manually to the pg_wal
7674 : * directory (removing the suffix). They can be useful in debugging,
7675 : * too.
7676 : *
7677 : * If a .done or .ready file already exists for the old timeline,
7678 : * however, we had already determined that the segment is complete, so
7679 : * we can let it be archived normally. (In particular, if it was
7680 : * restored from the archive to begin with, it's expected to have a
7681 : * .done file).
7682 : */
7683 46 : if (XLogSegmentOffset(EndOfLog, wal_segment_size) != 0 &&
7684 : XLogArchivingActive())
7685 : {
7686 : char origfname[MAXFNAMELEN];
7687 : XLogSegNo endLogSegNo;
7688 :
7689 2 : XLByteToPrevSeg(EndOfLog, endLogSegNo, wal_segment_size);
7690 2 : XLogFileName(origfname, EndOfLogTLI, endLogSegNo, wal_segment_size);
7691 :
7692 2 : if (!XLogArchiveIsReadyOrDone(origfname))
7693 : {
7694 : char origpath[MAXPGPATH];
7695 : char partialfname[MAXFNAMELEN];
7696 : char partialpath[MAXPGPATH];
7697 :
7698 2 : XLogFilePath(origpath, EndOfLogTLI, endLogSegNo, wal_segment_size);
7699 2 : snprintf(partialfname, MAXFNAMELEN, "%s.partial", origfname);
7700 2 : snprintf(partialpath, MAXPGPATH, "%s.partial", origpath);
7701 :
7702 : /*
7703 : * Make sure there's no .done or .ready file for the .partial
7704 : * file.
7705 : */
7706 2 : XLogArchiveCleanup(partialfname);
7707 :
7708 2 : durable_rename(origpath, partialpath, ERROR);
7709 2 : XLogArchiveNotify(partialfname);
7710 : }
7711 : }
7712 : }
7713 :
7714 : /*
7715 : * Preallocate additional log files, if wanted.
7716 : */
7717 1170 : PreallocXlogFiles(EndOfLog);
7718 :
7719 : /*
7720 : * Okay, we're officially UP.
7721 : */
7722 1170 : InRecovery = false;
7723 :
7724 : /* start the archive_timeout timer and LSN running */
7725 1170 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
7726 1170 : XLogCtl->lastSegSwitchLSN = EndOfLog;
7727 :
7728 : /* also initialize latestCompletedXid, to nextXid - 1 */
7729 1170 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
7730 1170 : ShmemVariableCache->latestCompletedXid = XidFromFullTransactionId(ShmemVariableCache->nextFullXid);
7731 3102 : TransactionIdRetreat(ShmemVariableCache->latestCompletedXid);
7732 1170 : LWLockRelease(ProcArrayLock);
7733 :
7734 : /*
7735 : * Start up the commit log and subtrans, if not already done for hot
7736 : * standby. (commit timestamps are started below, if necessary.)
7737 : */
7738 1170 : if (standbyState == STANDBY_DISABLED)
7739 : {
7740 1124 : StartupCLOG();
7741 1124 : StartupSUBTRANS(oldestActiveXID);
7742 : }
7743 :
7744 : /*
7745 : * Perform end of recovery actions for any SLRUs that need it.
7746 : */
7747 1170 : TrimCLOG();
7748 1170 : TrimMultiXact();
7749 :
7750 : /* Reload shared-memory state for prepared transactions */
7751 1170 : RecoverPreparedTransactions();
7752 :
7753 : /*
7754 : * Shutdown the recovery environment. This must occur after
7755 : * RecoverPreparedTransactions(), see notes for lock_twophase_recover()
7756 : */
7757 1170 : if (standbyState != STANDBY_DISABLED)
7758 46 : ShutdownRecoveryTransactionEnvironment();
7759 :
7760 : /* Shut down xlogreader */
7761 1170 : if (readFile >= 0)
7762 : {
7763 1156 : close(readFile);
7764 1156 : readFile = -1;
7765 : }
7766 1170 : XLogReaderFree(xlogreader);
7767 :
7768 : /*
7769 : * If any of the critical GUCs have changed, log them before we allow
7770 : * backends to write WAL.
7771 : */
7772 1170 : LocalSetXLogInsertAllowed();
7773 1170 : XLogReportParameters();
7774 :
7775 : /*
7776 : * Local WAL inserts enabled, so it's time to finish initialization of
7777 : * commit timestamp.
7778 : */
7779 1170 : CompleteCommitTsInitialization();
7780 :
7781 : /*
7782 : * All done with end-of-recovery actions.
7783 : *
7784 : * Now allow backends to write WAL and update the control file status in
7785 : * consequence. The boolean flag allowing backends to write WAL is
7786 : * updated while holding ControlFileLock to prevent other backends to look
7787 : * at an inconsistent state of the control file in shared memory. There
7788 : * is still a small window during which backends can write WAL and the
7789 : * control file is still referring to a system not in DB_IN_PRODUCTION
7790 : * state while looking at the on-disk control file.
7791 : *
7792 : * Also, although the boolean flag to allow WAL is probably atomic in
7793 : * itself, we use the info_lck here to ensure that there are no race
7794 : * conditions concerning visibility of other recent updates to shared
7795 : * memory.
7796 : */
7797 1170 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7798 1170 : ControlFile->state = DB_IN_PRODUCTION;
7799 1170 : ControlFile->time = (pg_time_t) time(NULL);
7800 :
7801 1170 : SpinLockAcquire(&XLogCtl->info_lck);
7802 1170 : XLogCtl->SharedRecoveryInProgress = false;
7803 1170 : SpinLockRelease(&XLogCtl->info_lck);
7804 :
7805 1170 : UpdateControlFile();
7806 1170 : LWLockRelease(ControlFileLock);
7807 :
7808 : /*
7809 : * If there were cascading standby servers connected to us, nudge any wal
7810 : * sender processes to notice that we've been promoted.
7811 : */
7812 1170 : WalSndWakeup();
7813 :
7814 : /*
7815 : * If this was a fast promotion, request an (online) checkpoint now. This
7816 : * isn't required for consistency, but the last restartpoint might be far
7817 : * back, and in case of a crash, recovering from it might take a longer
7818 : * than is appropriate now that we're not in standby mode anymore.
7819 : */
7820 1170 : if (fast_promoted)
7821 46 : RequestCheckpoint(CHECKPOINT_FORCE);
7822 1170 : }
7823 :
7824 : /*
7825 : * Checks if recovery has reached a consistent state. When consistency is
7826 : * reached and we have a valid starting standby snapshot, tell postmaster
7827 : * that it can start accepting read-only connections.
7828 : */
7829 : static void
7830 238590 : CheckRecoveryConsistency(void)
7831 : {
7832 : XLogRecPtr lastReplayedEndRecPtr;
7833 :
7834 : /*
7835 : * During crash recovery, we don't reach a consistent state until we've
7836 : * replayed all the WAL.
7837 : */
7838 238590 : if (XLogRecPtrIsInvalid(minRecoveryPoint))
7839 68652 : return;
7840 :
7841 : Assert(InArchiveRecovery);
7842 :
7843 : /*
7844 : * assume that we are called in the startup process, and hence don't need
7845 : * a lock to read lastReplayedEndRecPtr
7846 : */
7847 169938 : lastReplayedEndRecPtr = XLogCtl->lastReplayedEndRecPtr;
7848 :
7849 : /*
7850 : * Have we reached the point where our base backup was completed?
7851 : */
7852 169948 : if (!XLogRecPtrIsInvalid(ControlFile->backupEndPoint) &&
7853 10 : ControlFile->backupEndPoint <= lastReplayedEndRecPtr)
7854 : {
7855 : /*
7856 : * We have reached the end of base backup, as indicated by pg_control.
7857 : * The data on disk is now consistent. Reset backupStartPoint and
7858 : * backupEndPoint, and update minRecoveryPoint to make sure we don't
7859 : * allow starting up at an earlier point even if recovery is stopped
7860 : * and restarted soon after this.
7861 : */
7862 2 : elog(DEBUG1, "end of backup reached");
7863 :
7864 2 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7865 :
7866 2 : if (ControlFile->minRecoveryPoint < lastReplayedEndRecPtr)
7867 0 : ControlFile->minRecoveryPoint = lastReplayedEndRecPtr;
7868 :
7869 2 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
7870 2 : ControlFile->backupEndPoint = InvalidXLogRecPtr;
7871 2 : ControlFile->backupEndRequired = false;
7872 2 : UpdateControlFile();
7873 :
7874 2 : LWLockRelease(ControlFileLock);
7875 : }
7876 :
7877 : /*
7878 : * Have we passed our safe starting point? Note that minRecoveryPoint is
7879 : * known to be incorrectly set if ControlFile->backupEndRequired, until
7880 : * the XLOG_BACKUP_END arrives to advise us of the correct
7881 : * minRecoveryPoint. All we know prior to that is that we're not
7882 : * consistent yet.
7883 : */
7884 170026 : if (!reachedConsistency && !ControlFile->backupEndRequired &&
7885 162 : minRecoveryPoint <= lastReplayedEndRecPtr &&
7886 74 : XLogRecPtrIsInvalid(ControlFile->backupStartPoint))
7887 : {
7888 : /*
7889 : * Check to see if the XLOG sequence contained any unresolved
7890 : * references to uninitialized pages.
7891 : */
7892 68 : XLogCheckInvalidPages();
7893 :
7894 68 : reachedConsistency = true;
7895 68 : ereport(LOG,
7896 : (errmsg("consistent recovery state reached at %X/%X",
7897 : (uint32) (lastReplayedEndRecPtr >> 32),
7898 : (uint32) lastReplayedEndRecPtr)));
7899 : }
7900 :
7901 : /*
7902 : * Have we got a valid starting snapshot that will allow queries to be
7903 : * run? If so, we can tell postmaster that the database is consistent now,
7904 : * enabling connections.
7905 : */
7906 339818 : if (standbyState == STANDBY_SNAPSHOT_READY &&
7907 170066 : !LocalHotStandbyActive &&
7908 68 : reachedConsistency &&
7909 : IsUnderPostmaster)
7910 : {
7911 68 : SpinLockAcquire(&XLogCtl->info_lck);
7912 68 : XLogCtl->SharedHotStandbyActive = true;
7913 68 : SpinLockRelease(&XLogCtl->info_lck);
7914 :
7915 68 : LocalHotStandbyActive = true;
7916 :
7917 68 : SendPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY);
7918 : }
7919 : }
7920 :
7921 : /*
7922 : * Is the system still in recovery?
7923 : *
7924 : * Unlike testing InRecovery, this works in any process that's connected to
7925 : * shared memory.
7926 : *
7927 : * As a side-effect, we initialize the local TimeLineID and RedoRecPtr
7928 : * variables the first time we see that recovery is finished.
7929 : */
7930 : bool
7931 59778584 : RecoveryInProgress(void)
7932 : {
7933 : /*
7934 : * We check shared state each time only until we leave recovery mode. We
7935 : * can't re-enter recovery, so there's no need to keep checking after the
7936 : * shared variable has once been seen false.
7937 : */
7938 59778584 : if (!LocalRecoveryInProgress)
7939 59400024 : return false;
7940 : else
7941 : {
7942 : /*
7943 : * use volatile pointer to make sure we make a fresh read of the
7944 : * shared variable.
7945 : */
7946 378560 : volatile XLogCtlData *xlogctl = XLogCtl;
7947 :
7948 378560 : LocalRecoveryInProgress = xlogctl->SharedRecoveryInProgress;
7949 :
7950 : /*
7951 : * Initialize TimeLineID and RedoRecPtr when we discover that recovery
7952 : * is finished. InitPostgres() relies upon this behaviour to ensure
7953 : * that InitXLOGAccess() is called at backend startup. (If you change
7954 : * this, see also LocalSetXLogInsertAllowed.)
7955 : */
7956 378560 : if (!LocalRecoveryInProgress)
7957 : {
7958 : /*
7959 : * If we just exited recovery, make sure we read TimeLineID and
7960 : * RedoRecPtr after SharedRecoveryInProgress (for machines with
7961 : * weak memory ordering).
7962 : */
7963 10598 : pg_memory_barrier();
7964 10598 : InitXLOGAccess();
7965 : }
7966 :
7967 : /*
7968 : * Note: We don't need a memory barrier when we're still in recovery.
7969 : * We might exit recovery immediately after return, so the caller
7970 : * can't rely on 'true' meaning that we're still in recovery anyway.
7971 : */
7972 :
7973 378560 : return LocalRecoveryInProgress;
7974 : }
7975 : }
7976 :
7977 : /*
7978 : * Is HotStandby active yet? This is only important in special backends
7979 : * since normal backends won't ever be able to connect until this returns
7980 : * true. Postmaster knows this by way of signal, not via shared memory.
7981 : *
7982 : * Unlike testing standbyState, this works in any process that's connected to
7983 : * shared memory. (And note that standbyState alone doesn't tell the truth
7984 : * anyway.)
7985 : */
7986 : bool
7987 120 : HotStandbyActive(void)
7988 : {
7989 : /*
7990 : * We check shared state each time only until Hot Standby is active. We
7991 : * can't de-activate Hot Standby, so there's no need to keep checking
7992 : * after the shared variable has once been seen true.
7993 : */
7994 120 : if (LocalHotStandbyActive)
7995 20 : return true;
7996 : else
7997 : {
7998 : /* spinlock is essential on machines with weak memory ordering! */
7999 100 : SpinLockAcquire(&XLogCtl->info_lck);
8000 100 : LocalHotStandbyActive = XLogCtl->SharedHotStandbyActive;
8001 100 : SpinLockRelease(&XLogCtl->info_lck);
8002 :
8003 100 : return LocalHotStandbyActive;
8004 : }
8005 : }
8006 :
8007 : /*
8008 : * Like HotStandbyActive(), but to be used only in WAL replay code,
8009 : * where we don't need to ask any other process what the state is.
8010 : */
8011 : bool
8012 0 : HotStandbyActiveInReplay(void)
8013 : {
8014 : Assert(AmStartupProcess() || !IsPostmasterEnvironment);
8015 0 : return LocalHotStandbyActive;
8016 : }
8017 :
8018 : /*
8019 : * Is this process allowed to insert new WAL records?
8020 : *
8021 : * Ordinarily this is essentially equivalent to !RecoveryInProgress().
8022 : * But we also have provisions for forcing the result "true" or "false"
8023 : * within specific processes regardless of the global state.
8024 : */
8025 : bool
8026 62434806 : XLogInsertAllowed(void)
8027 : {
8028 : /*
8029 : * If value is "unconditionally true" or "unconditionally false", just
8030 : * return it. This provides the normal fast path once recovery is known
8031 : * done.
8032 : */
8033 62434806 : if (LocalXLogInsertAllowed >= 0)
8034 62425828 : return (bool) LocalXLogInsertAllowed;
8035 :
8036 : /*
8037 : * Else, must check to see if we're still in recovery.
8038 : */
8039 8978 : if (RecoveryInProgress())
8040 3784 : return false;
8041 :
8042 : /*
8043 : * On exit from recovery, reset to "unconditionally true", since there is
8044 : * no need to keep checking.
8045 : */
8046 5194 : LocalXLogInsertAllowed = 1;
8047 5194 : return true;
8048 : }
8049 :
8050 : /*
8051 : * Make XLogInsertAllowed() return true in the current process only.
8052 : *
8053 : * Note: it is allowed to switch LocalXLogInsertAllowed back to -1 later,
8054 : * and even call LocalSetXLogInsertAllowed() again after that.
8055 : */
8056 : static void
8057 2434 : LocalSetXLogInsertAllowed(void)
8058 : {
8059 : Assert(LocalXLogInsertAllowed == -1);
8060 2434 : LocalXLogInsertAllowed = 1;
8061 :
8062 : /* Initialize as RecoveryInProgress() would do when switching state */
8063 2434 : InitXLOGAccess();
8064 2434 : }
8065 :
8066 : /*
8067 : * Subroutine to try to fetch and validate a prior checkpoint record.
8068 : *
8069 : * whichChkpt identifies the checkpoint (merely for reporting purposes).
8070 : * 1 for "primary", 0 for "other" (backup_label)
8071 : */
8072 : static XLogRecord *
8073 1238 : ReadCheckpointRecord(XLogReaderState *xlogreader, XLogRecPtr RecPtr,
8074 : int whichChkpt, bool report)
8075 : {
8076 : XLogRecord *record;
8077 : uint8 info;
8078 :
8079 1238 : if (!XRecOffIsValid(RecPtr))
8080 : {
8081 0 : if (!report)
8082 0 : return NULL;
8083 :
8084 0 : switch (whichChkpt)
8085 : {
8086 : case 1:
8087 0 : ereport(LOG,
8088 : (errmsg("invalid primary checkpoint link in control file")));
8089 0 : break;
8090 : default:
8091 0 : ereport(LOG,
8092 : (errmsg("invalid checkpoint link in backup_label file")));
8093 0 : break;
8094 : }
8095 0 : return NULL;
8096 : }
8097 :
8098 1238 : record = ReadRecord(xlogreader, RecPtr, LOG, true);
8099 :
8100 1238 : if (record == NULL)
8101 : {
8102 0 : if (!report)
8103 0 : return NULL;
8104 :
8105 0 : switch (whichChkpt)
8106 : {
8107 : case 1:
8108 0 : ereport(LOG,
8109 : (errmsg("invalid primary checkpoint record")));
8110 0 : break;
8111 : default:
8112 0 : ereport(LOG,
8113 : (errmsg("invalid checkpoint record")));
8114 0 : break;
8115 : }
8116 0 : return NULL;
8117 : }
8118 1238 : if (record->xl_rmid != RM_XLOG_ID)
8119 : {
8120 0 : switch (whichChkpt)
8121 : {
8122 : case 1:
8123 0 : ereport(LOG,
8124 : (errmsg("invalid resource manager ID in primary checkpoint record")));
8125 0 : break;
8126 : default:
8127 0 : ereport(LOG,
8128 : (errmsg("invalid resource manager ID in checkpoint record")));
8129 0 : break;
8130 : }
8131 0 : return NULL;
8132 : }
8133 1238 : info = record->xl_info & ~XLR_INFO_MASK;
8134 1238 : if (info != XLOG_CHECKPOINT_SHUTDOWN &&
8135 : info != XLOG_CHECKPOINT_ONLINE)
8136 : {
8137 0 : switch (whichChkpt)
8138 : {
8139 : case 1:
8140 0 : ereport(LOG,
8141 : (errmsg("invalid xl_info in primary checkpoint record")));
8142 0 : break;
8143 : default:
8144 0 : ereport(LOG,
8145 : (errmsg("invalid xl_info in checkpoint record")));
8146 0 : break;
8147 : }
8148 0 : return NULL;
8149 : }
8150 1238 : if (record->xl_tot_len != SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(CheckPoint))
8151 : {
8152 0 : switch (whichChkpt)
8153 : {
8154 : case 1:
8155 0 : ereport(LOG,
8156 : (errmsg("invalid length of primary checkpoint record")));
8157 0 : break;
8158 : default:
8159 0 : ereport(LOG,
8160 : (errmsg("invalid length of checkpoint record")));
8161 0 : break;
8162 : }
8163 0 : return NULL;
8164 : }
8165 1238 : return record;
8166 : }
8167 :
8168 : /*
8169 : * This must be called in a backend process before creating WAL records
8170 : * (except in a standalone backend, which does StartupXLOG instead). We need
8171 : * to initialize the local copies of ThisTimeLineID and RedoRecPtr.
8172 : *
8173 : * Note: before Postgres 8.0, we went to some effort to keep the postmaster
8174 : * process's copies of ThisTimeLineID and RedoRecPtr valid too. This was
8175 : * unnecessary however, since the postmaster itself never touches XLOG anyway.
8176 : */
8177 : void
8178 13378 : InitXLOGAccess(void)
8179 : {
8180 13378 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8181 :
8182 : /* ThisTimeLineID doesn't change so we need no lock to copy it */
8183 13378 : ThisTimeLineID = XLogCtl->ThisTimeLineID;
8184 : Assert(ThisTimeLineID != 0 || IsBootstrapProcessingMode());
8185 :
8186 : /* set wal_segment_size */
8187 13378 : wal_segment_size = ControlFile->xlog_seg_size;
8188 :
8189 : /* Use GetRedoRecPtr to copy the RedoRecPtr safely */
8190 13378 : (void) GetRedoRecPtr();
8191 : /* Also update our copy of doPageWrites. */
8192 13378 : doPageWrites = (Insert->fullPageWrites || Insert->forcePageWrites);
8193 :
8194 : /* Also initialize the working areas for constructing WAL records */
8195 13378 : InitXLogInsert();
8196 13378 : }
8197 :
8198 : /*
8199 : * Return the current Redo pointer from shared memory.
8200 : *
8201 : * As a side-effect, the local RedoRecPtr copy is updated.
8202 : */
8203 : XLogRecPtr
8204 158606 : GetRedoRecPtr(void)
8205 : {
8206 : XLogRecPtr ptr;
8207 :
8208 : /*
8209 : * The possibly not up-to-date copy in XlogCtl is enough. Even if we
8210 : * grabbed a WAL insertion lock to read the master copy, someone might
8211 : * update it just after we've released the lock.
8212 : */
8213 158606 : SpinLockAcquire(&XLogCtl->info_lck);
8214 158606 : ptr = XLogCtl->RedoRecPtr;
8215 158606 : SpinLockRelease(&XLogCtl->info_lck);
8216 :
8217 158606 : if (RedoRecPtr < ptr)
8218 9954 : RedoRecPtr = ptr;
8219 :
8220 158606 : return RedoRecPtr;
8221 : }
8222 :
8223 : /*
8224 : * Return information needed to decide whether a modified block needs a
8225 : * full-page image to be included in the WAL record.
8226 : *
8227 : * The returned values are cached copies from backend-private memory, and
8228 : * possibly out-of-date. XLogInsertRecord will re-check them against
8229 : * up-to-date values, while holding the WAL insert lock.
8230 : */
8231 : void
8232 29683720 : GetFullPageWriteInfo(XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
8233 : {
8234 29683720 : *RedoRecPtr_p = RedoRecPtr;
8235 29683720 : *doPageWrites_p = doPageWrites;
8236 29683720 : }
8237 :
8238 : /*
8239 : * GetInsertRecPtr -- Returns the current insert position.
8240 : *
8241 : * NOTE: The value *actually* returned is the position of the last full
8242 : * xlog page. It lags behind the real insert position by at most 1 page.
8243 : * For that, we don't need to scan through WAL insertion locks, and an
8244 : * approximation is enough for the current usage of this function.
8245 : */
8246 : XLogRecPtr
8247 1856 : GetInsertRecPtr(void)
8248 : {
8249 : XLogRecPtr recptr;
8250 :
8251 1856 : SpinLockAcquire(&XLogCtl->info_lck);
8252 1856 : recptr = XLogCtl->LogwrtRqst.Write;
8253 1856 : SpinLockRelease(&XLogCtl->info_lck);
8254 :
8255 1856 : return recptr;
8256 : }
8257 :
8258 : /*
8259 : * GetFlushRecPtr -- Returns the current flush position, ie, the last WAL
8260 : * position known to be fsync'd to disk.
8261 : */
8262 : XLogRecPtr
8263 56128 : GetFlushRecPtr(void)
8264 : {
8265 56128 : SpinLockAcquire(&XLogCtl->info_lck);
8266 56128 : LogwrtResult = XLogCtl->LogwrtResult;
8267 56128 : SpinLockRelease(&XLogCtl->info_lck);
8268 :
8269 56128 : return LogwrtResult.Flush;
8270 : }
8271 :
8272 : /*
8273 : * GetLastImportantRecPtr -- Returns the LSN of the last important record
8274 : * inserted. All records not explicitly marked as unimportant are considered
8275 : * important.
8276 : *
8277 : * The LSN is determined by computing the maximum of
8278 : * WALInsertLocks[i].lastImportantAt.
8279 : */
8280 : XLogRecPtr
8281 2872 : GetLastImportantRecPtr(void)
8282 : {
8283 2872 : XLogRecPtr res = InvalidXLogRecPtr;
8284 : int i;
8285 :
8286 25848 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
8287 : {
8288 : XLogRecPtr last_important;
8289 :
8290 : /*
8291 : * Need to take a lock to prevent torn reads of the LSN, which are
8292 : * possible on some of the supported platforms. WAL insert locks only
8293 : * support exclusive mode, so we have to use that.
8294 : */
8295 22976 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
8296 22976 : last_important = WALInsertLocks[i].l.lastImportantAt;
8297 22976 : LWLockRelease(&WALInsertLocks[i].l.lock);
8298 :
8299 22976 : if (res < last_important)
8300 3420 : res = last_important;
8301 : }
8302 :
8303 2872 : return res;
8304 : }
8305 :
8306 : /*
8307 : * Get the time and LSN of the last xlog segment switch
8308 : */
8309 : pg_time_t
8310 0 : GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
8311 : {
8312 : pg_time_t result;
8313 :
8314 : /* Need WALWriteLock, but shared lock is sufficient */
8315 0 : LWLockAcquire(WALWriteLock, LW_SHARED);
8316 0 : result = XLogCtl->lastSegSwitchTime;
8317 0 : *lastSwitchLSN = XLogCtl->lastSegSwitchLSN;
8318 0 : LWLockRelease(WALWriteLock);
8319 :
8320 0 : return result;
8321 : }
8322 :
8323 : /*
8324 : * This must be called ONCE during postmaster or standalone-backend shutdown
8325 : */
8326 : void
8327 1024 : ShutdownXLOG(int code, Datum arg)
8328 : {
8329 : /*
8330 : * We should have an aux process resource owner to use, and we should not
8331 : * be in a transaction that's installed some other resowner.
8332 : */
8333 : Assert(AuxProcessResourceOwner != NULL);
8334 : Assert(CurrentResourceOwner == NULL ||
8335 : CurrentResourceOwner == AuxProcessResourceOwner);
8336 1024 : CurrentResourceOwner = AuxProcessResourceOwner;
8337 :
8338 : /* Don't be chatty in standalone mode */
8339 1024 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
8340 : (errmsg("shutting down")));
8341 :
8342 : /*
8343 : * Signal walsenders to move to stopping state.
8344 : */
8345 1024 : WalSndInitStopping();
8346 :
8347 : /*
8348 : * Wait for WAL senders to be in stopping state. This prevents commands
8349 : * from writing new WAL.
8350 : */
8351 1024 : WalSndWaitStopping();
8352 :
8353 1024 : if (RecoveryInProgress())
8354 22 : CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
8355 : else
8356 : {
8357 : /*
8358 : * If archiving is enabled, rotate the last XLOG file so that all the
8359 : * remaining records are archived (postmaster wakes up the archiver
8360 : * process one more time at the end of shutdown). The checkpoint
8361 : * record will go to the next XLOG file and won't be archived (yet).
8362 : */
8363 1002 : if (XLogArchivingActive() && XLogArchiveCommandSet())
8364 0 : RequestXLogSwitch(false);
8365 :
8366 1002 : CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
8367 : }
8368 1024 : ShutdownCLOG();
8369 1024 : ShutdownCommitTs();
8370 1024 : ShutdownSUBTRANS();
8371 1024 : ShutdownMultiXact();
8372 1024 : }
8373 :
8374 : /*
8375 : * Log start of a checkpoint.
8376 : */
8377 : static void
8378 450 : LogCheckpointStart(int flags, bool restartpoint)
8379 : {
8380 450 : elog(LOG, "%s starting:%s%s%s%s%s%s%s%s",
8381 : restartpoint ? "restartpoint" : "checkpoint",
8382 : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
8383 : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
8384 : (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
8385 : (flags & CHECKPOINT_FORCE) ? " force" : "",
8386 : (flags & CHECKPOINT_WAIT) ? " wait" : "",
8387 : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
8388 : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
8389 : (flags & CHECKPOINT_FLUSH_ALL) ? " flush-all" : "");
8390 450 : }
8391 :
8392 : /*
8393 : * Log end of a checkpoint.
8394 : */
8395 : static void
8396 2832 : LogCheckpointEnd(bool restartpoint)
8397 : {
8398 : long write_secs,
8399 : sync_secs,
8400 : total_secs,
8401 : longest_secs,
8402 : average_secs;
8403 : int write_usecs,
8404 : sync_usecs,
8405 : total_usecs,
8406 : longest_usecs,
8407 : average_usecs;
8408 : uint64 average_sync_time;
8409 :
8410 2832 : CheckpointStats.ckpt_end_t = GetCurrentTimestamp();
8411 :
8412 2832 : TimestampDifference(CheckpointStats.ckpt_write_t,
8413 : CheckpointStats.ckpt_sync_t,
8414 : &write_secs, &write_usecs);
8415 :
8416 2832 : TimestampDifference(CheckpointStats.ckpt_sync_t,
8417 : CheckpointStats.ckpt_sync_end_t,
8418 : &sync_secs, &sync_usecs);
8419 :
8420 : /* Accumulate checkpoint timing summary data, in milliseconds. */
8421 5664 : BgWriterStats.m_checkpoint_write_time +=
8422 2832 : write_secs * 1000 + write_usecs / 1000;
8423 5664 : BgWriterStats.m_checkpoint_sync_time +=
8424 2832 : sync_secs * 1000 + sync_usecs / 1000;
8425 :
8426 : /*
8427 : * All of the published timing statistics are accounted for. Only
8428 : * continue if a log message is to be written.
8429 : */
8430 2832 : if (!log_checkpoints)
8431 2382 : return;
8432 :
8433 450 : TimestampDifference(CheckpointStats.ckpt_start_t,
8434 : CheckpointStats.ckpt_end_t,
8435 : &total_secs, &total_usecs);
8436 :
8437 : /*
8438 : * Timing values returned from CheckpointStats are in microseconds.
8439 : * Convert to the second plus microsecond form that TimestampDifference
8440 : * returns for homogeneous printing.
8441 : */
8442 450 : longest_secs = (long) (CheckpointStats.ckpt_longest_sync / 1000000);
8443 450 : longest_usecs = CheckpointStats.ckpt_longest_sync -
8444 : (uint64) longest_secs * 1000000;
8445 :
8446 450 : average_sync_time = 0;
8447 450 : if (CheckpointStats.ckpt_sync_rels > 0)
8448 0 : average_sync_time = CheckpointStats.ckpt_agg_sync_time /
8449 0 : CheckpointStats.ckpt_sync_rels;
8450 450 : average_secs = (long) (average_sync_time / 1000000);
8451 450 : average_usecs = average_sync_time - (uint64) average_secs * 1000000;
8452 :
8453 450 : elog(LOG, "%s complete: wrote %d buffers (%.1f%%); "
8454 : "%d WAL file(s) added, %d removed, %d recycled; "
8455 : "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s; "
8456 : "sync files=%d, longest=%ld.%03d s, average=%ld.%03d s; "
8457 : "distance=%d kB, estimate=%d kB",
8458 : restartpoint ? "restartpoint" : "checkpoint",
8459 : CheckpointStats.ckpt_bufs_written,
8460 : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
8461 : CheckpointStats.ckpt_segs_added,
8462 : CheckpointStats.ckpt_segs_removed,
8463 : CheckpointStats.ckpt_segs_recycled,
8464 : write_secs, write_usecs / 1000,
8465 : sync_secs, sync_usecs / 1000,
8466 : total_secs, total_usecs / 1000,
8467 : CheckpointStats.ckpt_sync_rels,
8468 : longest_secs, longest_usecs / 1000,
8469 : average_secs, average_usecs / 1000,
8470 : (int) (PrevCheckPointDistance / 1024.0),
8471 : (int) (CheckPointDistanceEstimate / 1024.0));
8472 : }
8473 :
8474 : /*
8475 : * Update the estimate of distance between checkpoints.
8476 : *
8477 : * The estimate is used to calculate the number of WAL segments to keep
8478 : * preallocated, see XLOGfileslop().
8479 : */
8480 : static void
8481 2832 : UpdateCheckPointDistanceEstimate(uint64 nbytes)
8482 : {
8483 : /*
8484 : * To estimate the number of segments consumed between checkpoints, keep a
8485 : * moving average of the amount of WAL generated in previous checkpoint
8486 : * cycles. However, if the load is bursty, with quiet periods and busy
8487 : * periods, we want to cater for the peak load. So instead of a plain
8488 : * moving average, let the average decline slowly if the previous cycle
8489 : * used less WAL than estimated, but bump it up immediately if it used
8490 : * more.
8491 : *
8492 : * When checkpoints are triggered by max_wal_size, this should converge to
8493 : * CheckpointSegments * wal_segment_size,
8494 : *
8495 : * Note: This doesn't pay any attention to what caused the checkpoint.
8496 : * Checkpoints triggered manually with CHECKPOINT command, or by e.g.
8497 : * starting a base backup, are counted the same as those created
8498 : * automatically. The slow-decline will largely mask them out, if they are
8499 : * not frequent. If they are frequent, it seems reasonable to count them
8500 : * in as any others; if you issue a manual checkpoint every 5 minutes and
8501 : * never let a timed checkpoint happen, it makes sense to base the
8502 : * preallocation on that 5 minute interval rather than whatever
8503 : * checkpoint_timeout is set to.
8504 : */
8505 2832 : PrevCheckPointDistance = nbytes;
8506 2832 : if (CheckPointDistanceEstimate < nbytes)
8507 1228 : CheckPointDistanceEstimate = nbytes;
8508 : else
8509 1604 : CheckPointDistanceEstimate =
8510 1604 : (0.90 * CheckPointDistanceEstimate + 0.10 * (double) nbytes);
8511 2832 : }
8512 :
8513 : /*
8514 : * Perform a checkpoint --- either during shutdown, or on-the-fly
8515 : *
8516 : * flags is a bitwise OR of the following:
8517 : * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
8518 : * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
8519 : * CHECKPOINT_IMMEDIATE: finish the checkpoint ASAP,
8520 : * ignoring checkpoint_completion_target parameter.
8521 : * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
8522 : * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
8523 : * CHECKPOINT_END_OF_RECOVERY).
8524 : * CHECKPOINT_FLUSH_ALL: also flush buffers of unlogged tables.
8525 : *
8526 : * Note: flags contains other bits, of interest here only for logging purposes.
8527 : * In particular note that this routine is synchronous and does not pay
8528 : * attention to CHECKPOINT_WAIT.
8529 : *
8530 : * If !shutdown then we are writing an online checkpoint. This is a very special
8531 : * kind of operation and WAL record because the checkpoint action occurs over
8532 : * a period of time yet logically occurs at just a single LSN. The logical
8533 : * position of the WAL record (redo ptr) is the same or earlier than the
8534 : * physical position. When we replay WAL we locate the checkpoint via its
8535 : * physical position then read the redo ptr and actually start replay at the
8536 : * earlier logical position. Note that we don't write *anything* to WAL at
8537 : * the logical position, so that location could be any other kind of WAL record.
8538 : * All of this mechanism allows us to continue working while we checkpoint.
8539 : * As a result, timing of actions is critical here and be careful to note that
8540 : * this function will likely take minutes to execute on a busy system.
8541 : */
8542 : void
8543 2816 : CreateCheckPoint(int flags)
8544 : {
8545 : bool shutdown;
8546 : CheckPoint checkPoint;
8547 : XLogRecPtr recptr;
8548 : XLogSegNo _logSegNo;
8549 2816 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8550 : uint32 freespace;
8551 : XLogRecPtr PriorRedoPtr;
8552 : XLogRecPtr curInsert;
8553 : XLogRecPtr last_important_lsn;
8554 : VirtualTransactionId *vxids;
8555 : int nvxids;
8556 :
8557 : /*
8558 : * An end-of-recovery checkpoint is really a shutdown checkpoint, just
8559 : * issued at a different time.
8560 : */
8561 2816 : if (flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY))
8562 1050 : shutdown = true;
8563 : else
8564 1766 : shutdown = false;
8565 :
8566 : /* sanity check */
8567 2816 : if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
8568 0 : elog(ERROR, "can't create a checkpoint during recovery");
8569 :
8570 : /*
8571 : * Initialize InitXLogInsert working areas before entering the critical
8572 : * section. Normally, this is done by the first call to
8573 : * RecoveryInProgress() or LocalSetXLogInsertAllowed(), but when creating
8574 : * an end-of-recovery checkpoint, the LocalSetXLogInsertAllowed call is
8575 : * done below in a critical section, and InitXLogInsert cannot be called
8576 : * in a critical section.
8577 : */
8578 2816 : InitXLogInsert();
8579 :
8580 : /*
8581 : * Acquire CheckpointLock to ensure only one checkpoint happens at a time.
8582 : * (This is just pro forma, since in the present system structure there is
8583 : * only one process that is allowed to issue checkpoints at any given
8584 : * time.)
8585 : */
8586 2816 : LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
8587 :
8588 : /*
8589 : * Prepare to accumulate statistics.
8590 : *
8591 : * Note: because it is possible for log_checkpoints to change while a
8592 : * checkpoint proceeds, we always accumulate stats, even if
8593 : * log_checkpoints is currently off.
8594 : */
8595 2816 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
8596 2816 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
8597 :
8598 : /*
8599 : * Use a critical section to force system panic if we have trouble.
8600 : */
8601 2816 : START_CRIT_SECTION();
8602 :
8603 2816 : if (shutdown)
8604 : {
8605 1050 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8606 1050 : ControlFile->state = DB_SHUTDOWNING;
8607 1050 : ControlFile->time = (pg_time_t) time(NULL);
8608 1050 : UpdateControlFile();
8609 1050 : LWLockRelease(ControlFileLock);
8610 : }
8611 :
8612 : /*
8613 : * Let smgr prepare for checkpoint; this has to happen before we determine
8614 : * the REDO pointer. Note that smgr must not do anything that'd have to
8615 : * be undone if we decide no checkpoint is needed.
8616 : */
8617 2816 : SyncPreCheckpoint();
8618 :
8619 : /* Begin filling in the checkpoint WAL record */
8620 2816 : MemSet(&checkPoint, 0, sizeof(checkPoint));
8621 2816 : checkPoint.time = (pg_time_t) time(NULL);
8622 :
8623 : /*
8624 : * For Hot Standby, derive the oldestActiveXid before we fix the redo
8625 : * pointer. This allows us to begin accumulating changes to assemble our
8626 : * starting snapshot of locks and transactions.
8627 : */
8628 2816 : if (!shutdown && XLogStandbyInfoActive())
8629 1728 : checkPoint.oldestActiveXid = GetOldestActiveTransactionId();
8630 : else
8631 1088 : checkPoint.oldestActiveXid = InvalidTransactionId;
8632 :
8633 : /*
8634 : * Get location of last important record before acquiring insert locks (as
8635 : * GetLastImportantRecPtr() also locks WAL locks).
8636 : */
8637 2816 : last_important_lsn = GetLastImportantRecPtr();
8638 :
8639 : /*
8640 : * We must block concurrent insertions while examining insert state to
8641 : * determine the checkpoint REDO pointer.
8642 : */
8643 2816 : WALInsertLockAcquireExclusive();
8644 2816 : curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
8645 :
8646 : /*
8647 : * If this isn't a shutdown or forced checkpoint, and if there has been no
8648 : * WAL activity requiring a checkpoint, skip it. The idea here is to
8649 : * avoid inserting duplicate checkpoints when the system is idle.
8650 : */
8651 2816 : if ((flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY |
8652 : CHECKPOINT_FORCE)) == 0)
8653 : {
8654 10 : if (last_important_lsn == ControlFile->checkPoint)
8655 : {
8656 0 : WALInsertLockRelease();
8657 0 : LWLockRelease(CheckpointLock);
8658 0 : END_CRIT_SECTION();
8659 0 : ereport(DEBUG1,
8660 : (errmsg("checkpoint skipped because system is idle")));
8661 0 : return;
8662 : }
8663 : }
8664 :
8665 : /*
8666 : * An end-of-recovery checkpoint is created before anyone is allowed to
8667 : * write WAL. To allow us to write the checkpoint record, temporarily
8668 : * enable XLogInsertAllowed. (This also ensures ThisTimeLineID is
8669 : * initialized, which we need here and in AdvanceXLInsertBuffer.)
8670 : */
8671 2816 : if (flags & CHECKPOINT_END_OF_RECOVERY)
8672 48 : LocalSetXLogInsertAllowed();
8673 :
8674 2816 : checkPoint.ThisTimeLineID = ThisTimeLineID;
8675 2816 : if (flags & CHECKPOINT_END_OF_RECOVERY)
8676 48 : checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
8677 : else
8678 2768 : checkPoint.PrevTimeLineID = ThisTimeLineID;
8679 :
8680 2816 : checkPoint.fullPageWrites = Insert->fullPageWrites;
8681 :
8682 : /*
8683 : * Compute new REDO record ptr = location of next XLOG record.
8684 : *
8685 : * NB: this is NOT necessarily where the checkpoint record itself will be,
8686 : * since other backends may insert more XLOG records while we're off doing
8687 : * the buffer flush work. Those XLOG records are logically after the
8688 : * checkpoint, even though physically before it. Got that?
8689 : */
8690 2816 : freespace = INSERT_FREESPACE(curInsert);
8691 2816 : if (freespace == 0)
8692 : {
8693 0 : if (XLogSegmentOffset(curInsert, wal_segment_size) == 0)
8694 0 : curInsert += SizeOfXLogLongPHD;
8695 : else
8696 0 : curInsert += SizeOfXLogShortPHD;
8697 : }
8698 2816 : checkPoint.redo = curInsert;
8699 :
8700 : /*
8701 : * Here we update the shared RedoRecPtr for future XLogInsert calls; this
8702 : * must be done while holding all the insertion locks.
8703 : *
8704 : * Note: if we fail to complete the checkpoint, RedoRecPtr will be left
8705 : * pointing past where it really needs to point. This is okay; the only
8706 : * consequence is that XLogInsert might back up whole buffers that it
8707 : * didn't really need to. We can't postpone advancing RedoRecPtr because
8708 : * XLogInserts that happen while we are dumping buffers must assume that
8709 : * their buffer changes are not included in the checkpoint.
8710 : */
8711 2816 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
8712 :
8713 : /*
8714 : * Now we can release the WAL insertion locks, allowing other xacts to
8715 : * proceed while we are flushing disk buffers.
8716 : */
8717 2816 : WALInsertLockRelease();
8718 :
8719 : /* Update the info_lck-protected copy of RedoRecPtr as well */
8720 2816 : SpinLockAcquire(&XLogCtl->info_lck);
8721 2816 : XLogCtl->RedoRecPtr = checkPoint.redo;
8722 2816 : SpinLockRelease(&XLogCtl->info_lck);
8723 :
8724 : /*
8725 : * If enabled, log checkpoint start. We postpone this until now so as not
8726 : * to log anything if we decided to skip the checkpoint.
8727 : */
8728 2816 : if (log_checkpoints)
8729 446 : LogCheckpointStart(flags, false);
8730 :
8731 : TRACE_POSTGRESQL_CHECKPOINT_START(flags);
8732 :
8733 : /*
8734 : * Get the other info we need for the checkpoint record.
8735 : *
8736 : * We don't need to save oldestClogXid in the checkpoint, it only matters
8737 : * for the short period in which clog is being truncated, and if we crash
8738 : * during that we'll redo the clog truncation and fix up oldestClogXid
8739 : * there.
8740 : */
8741 2816 : LWLockAcquire(XidGenLock, LW_SHARED);
8742 2816 : checkPoint.nextFullXid = ShmemVariableCache->nextFullXid;
8743 2816 : checkPoint.oldestXid = ShmemVariableCache->oldestXid;
8744 2816 : checkPoint.oldestXidDB = ShmemVariableCache->oldestXidDB;
8745 2816 : LWLockRelease(XidGenLock);
8746 :
8747 2816 : LWLockAcquire(CommitTsLock, LW_SHARED);
8748 2816 : checkPoint.oldestCommitTsXid = ShmemVariableCache->oldestCommitTsXid;
8749 2816 : checkPoint.newestCommitTsXid = ShmemVariableCache->newestCommitTsXid;
8750 2816 : LWLockRelease(CommitTsLock);
8751 :
8752 2816 : LWLockAcquire(OidGenLock, LW_SHARED);
8753 2816 : checkPoint.nextOid = ShmemVariableCache->nextOid;
8754 2816 : if (!shutdown)
8755 1766 : checkPoint.nextOid += ShmemVariableCache->oidCount;
8756 2816 : LWLockRelease(OidGenLock);
8757 :
8758 2816 : MultiXactGetCheckptMulti(shutdown,
8759 : &checkPoint.nextMulti,
8760 : &checkPoint.nextMultiOffset,
8761 : &checkPoint.oldestMulti,
8762 : &checkPoint.oldestMultiDB);
8763 :
8764 : /*
8765 : * Having constructed the checkpoint record, ensure all shmem disk buffers
8766 : * and commit-log buffers are flushed to disk.
8767 : *
8768 : * This I/O could fail for various reasons. If so, we will fail to
8769 : * complete the checkpoint, but there is no reason to force a system
8770 : * panic. Accordingly, exit critical section while doing it.
8771 : */
8772 2816 : END_CRIT_SECTION();
8773 :
8774 : /*
8775 : * In some cases there are groups of actions that must all occur on one
8776 : * side or the other of a checkpoint record. Before flushing the
8777 : * checkpoint record we must explicitly wait for any backend currently
8778 : * performing those groups of actions.
8779 : *
8780 : * One example is end of transaction, so we must wait for any transactions
8781 : * that are currently in commit critical sections. If an xact inserted
8782 : * its commit record into XLOG just before the REDO point, then a crash
8783 : * restart from the REDO point would not replay that record, which means
8784 : * that our flushing had better include the xact's update of pg_xact. So
8785 : * we wait till he's out of his commit critical section before proceeding.
8786 : * See notes in RecordTransactionCommit().
8787 : *
8788 : * Because we've already released the insertion locks, this test is a bit
8789 : * fuzzy: it is possible that we will wait for xacts we didn't really need
8790 : * to wait for. But the delay should be short and it seems better to make
8791 : * checkpoint take a bit longer than to hold off insertions longer than
8792 : * necessary. (In fact, the whole reason we have this issue is that xact.c
8793 : * does commit record XLOG insertion and clog update as two separate steps
8794 : * protected by different locks, but again that seems best on grounds of
8795 : * minimizing lock contention.)
8796 : *
8797 : * A transaction that has not yet set delayChkpt when we look cannot be at
8798 : * risk, since he's not inserted his commit record yet; and one that's
8799 : * already cleared it is not at risk either, since he's done fixing clog
8800 : * and we will correctly flush the update below. So we cannot miss any
8801 : * xacts we need to wait for.
8802 : */
8803 2816 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids);
8804 2816 : if (nvxids > 0)
8805 : {
8806 : do
8807 : {
8808 8 : pg_usleep(10000L); /* wait for 10 msec */
8809 8 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids));
8810 : }
8811 2816 : pfree(vxids);
8812 :
8813 2816 : CheckPointGuts(checkPoint.redo, flags);
8814 :
8815 : /*
8816 : * Take a snapshot of running transactions and write this to WAL. This
8817 : * allows us to reconstruct the state of running transactions during
8818 : * archive recovery, if required. Skip, if this info disabled.
8819 : *
8820 : * If we are shutting down, or Startup process is completing crash
8821 : * recovery we don't need to write running xact data.
8822 : */
8823 2816 : if (!shutdown && XLogStandbyInfoActive())
8824 1728 : LogStandbySnapshot();
8825 :
8826 2816 : START_CRIT_SECTION();
8827 :
8828 : /*
8829 : * Now insert the checkpoint record into XLOG.
8830 : */
8831 2816 : XLogBeginInsert();
8832 2816 : XLogRegisterData((char *) (&checkPoint), sizeof(checkPoint));
8833 2816 : recptr = XLogInsert(RM_XLOG_ID,
8834 : shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
8835 : XLOG_CHECKPOINT_ONLINE);
8836 :
8837 2816 : XLogFlush(recptr);
8838 :
8839 : /*
8840 : * We mustn't write any new WAL after a shutdown checkpoint, or it will be
8841 : * overwritten at next startup. No-one should even try, this just allows
8842 : * sanity-checking. In the case of an end-of-recovery checkpoint, we want
8843 : * to just temporarily disable writing until the system has exited
8844 : * recovery.
8845 : */
8846 2816 : if (shutdown)
8847 : {
8848 1050 : if (flags & CHECKPOINT_END_OF_RECOVERY)
8849 48 : LocalXLogInsertAllowed = -1; /* return to "check" state */
8850 : else
8851 1002 : LocalXLogInsertAllowed = 0; /* never again write WAL */
8852 : }
8853 :
8854 : /*
8855 : * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
8856 : * = end of actual checkpoint record.
8857 : */
8858 2816 : if (shutdown && checkPoint.redo != ProcLastRecPtr)
8859 0 : ereport(PANIC,
8860 : (errmsg("concurrent write-ahead log activity while database system is shutting down")));
8861 :
8862 : /*
8863 : * Remember the prior checkpoint's redo ptr for
8864 : * UpdateCheckPointDistanceEstimate()
8865 : */
8866 2816 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
8867 :
8868 : /*
8869 : * Update the control file.
8870 : */
8871 2816 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8872 2816 : if (shutdown)
8873 1050 : ControlFile->state = DB_SHUTDOWNED;
8874 2816 : ControlFile->checkPoint = ProcLastRecPtr;
8875 2816 : ControlFile->checkPointCopy = checkPoint;
8876 2816 : ControlFile->time = (pg_time_t) time(NULL);
8877 : /* crash recovery should always recover to the end of WAL */
8878 2816 : ControlFile->minRecoveryPoint = InvalidXLogRecPtr;
8879 2816 : ControlFile->minRecoveryPointTLI = 0;
8880 :
8881 : /*
8882 : * Persist unloggedLSN value. It's reset on crash recovery, so this goes
8883 : * unused on non-shutdown checkpoints, but seems useful to store it always
8884 : * for debugging purposes.
8885 : */
8886 2816 : SpinLockAcquire(&XLogCtl->ulsn_lck);
8887 2816 : ControlFile->unloggedLSN = XLogCtl->unloggedLSN;
8888 2816 : SpinLockRelease(&XLogCtl->ulsn_lck);
8889 :
8890 2816 : UpdateControlFile();
8891 2816 : LWLockRelease(ControlFileLock);
8892 :
8893 : /* Update shared-memory copy of checkpoint XID/epoch */
8894 2816 : SpinLockAcquire(&XLogCtl->info_lck);
8895 2816 : XLogCtl->ckptFullXid = checkPoint.nextFullXid;
8896 2816 : SpinLockRelease(&XLogCtl->info_lck);
8897 :
8898 : /*
8899 : * We are now done with critical updates; no need for system panic if we
8900 : * have trouble while fooling with old log segments.
8901 : */
8902 2816 : END_CRIT_SECTION();
8903 :
8904 : /*
8905 : * Let smgr do post-checkpoint cleanup (eg, deleting old files).
8906 : */
8907 2816 : SyncPostCheckpoint();
8908 :
8909 : /*
8910 : * Update the average distance between checkpoints if the prior checkpoint
8911 : * exists.
8912 : */
8913 2816 : if (PriorRedoPtr != InvalidXLogRecPtr)
8914 2816 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
8915 :
8916 : /*
8917 : * Delete old log files, those no longer needed for last checkpoint to
8918 : * prevent the disk holding the xlog from growing full.
8919 : */
8920 2816 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
8921 2816 : KeepLogSeg(recptr, &_logSegNo);
8922 2816 : _logSegNo--;
8923 2816 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, recptr);
8924 :
8925 : /*
8926 : * Make more log segments if needed. (Do this after recycling old log
8927 : * segments, since that may supply some of the needed files.)
8928 : */
8929 2816 : if (!shutdown)
8930 1766 : PreallocXlogFiles(recptr);
8931 :
8932 : /*
8933 : * Truncate pg_subtrans if possible. We can throw away all data before
8934 : * the oldest XMIN of any running transaction. No future transaction will
8935 : * attempt to reference any pg_subtrans entry older than that (see Asserts
8936 : * in subtrans.c). During recovery, though, we mustn't do this because
8937 : * StartupSUBTRANS hasn't been called yet.
8938 : */
8939 2816 : if (!RecoveryInProgress())
8940 2768 : TruncateSUBTRANS(GetOldestXmin(NULL, PROCARRAY_FLAGS_DEFAULT));
8941 :
8942 : /* Real work is done, but log and update stats before releasing lock. */
8943 2816 : LogCheckpointEnd(false);
8944 :
8945 : TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
8946 : NBuffers,
8947 : CheckpointStats.ckpt_segs_added,
8948 : CheckpointStats.ckpt_segs_removed,
8949 : CheckpointStats.ckpt_segs_recycled);
8950 :
8951 2816 : LWLockRelease(CheckpointLock);
8952 : }
8953 :
8954 : /*
8955 : * Mark the end of recovery in WAL though without running a full checkpoint.
8956 : * We can expect that a restartpoint is likely to be in progress as we
8957 : * do this, though we are unwilling to wait for it to complete. So be
8958 : * careful to avoid taking the CheckpointLock anywhere here.
8959 : *
8960 : * CreateRestartPoint() allows for the case where recovery may end before
8961 : * the restartpoint completes so there is no concern of concurrent behaviour.
8962 : */
8963 : static void
8964 46 : CreateEndOfRecoveryRecord(void)
8965 : {
8966 : xl_end_of_recovery xlrec;
8967 : XLogRecPtr recptr;
8968 :
8969 : /* sanity check */
8970 46 : if (!RecoveryInProgress())
8971 0 : elog(ERROR, "can only be used to end recovery");
8972 :
8973 46 : xlrec.end_time = GetCurrentTimestamp();
8974 :
8975 46 : WALInsertLockAcquireExclusive();
8976 46 : xlrec.ThisTimeLineID = ThisTimeLineID;
8977 46 : xlrec.PrevTimeLineID = XLogCtl->PrevTimeLineID;
8978 46 : WALInsertLockRelease();
8979 :
8980 46 : LocalSetXLogInsertAllowed();
8981 :
8982 46 : START_CRIT_SECTION();
8983 :
8984 46 : XLogBeginInsert();
8985 46 : XLogRegisterData((char *) &xlrec, sizeof(xl_end_of_recovery));
8986 46 : recptr = XLogInsert(RM_XLOG_ID, XLOG_END_OF_RECOVERY);
8987 :
8988 46 : XLogFlush(recptr);
8989 :
8990 : /*
8991 : * Update the control file so that crash recovery can follow the timeline
8992 : * changes to this point.
8993 : */
8994 46 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8995 46 : ControlFile->time = (pg_time_t) time(NULL);
8996 46 : ControlFile->minRecoveryPoint = recptr;
8997 46 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
8998 46 : UpdateControlFile();
8999 46 : LWLockRelease(ControlFileLock);
9000 :
9001 46 : END_CRIT_SECTION();
9002 :
9003 46 : LocalXLogInsertAllowed = -1; /* return to "check" state */
9004 46 : }
9005 :
9006 : /*
9007 : * Flush all data in shared memory to disk, and fsync
9008 : *
9009 : * This is the common code shared between regular checkpoints and
9010 : * recovery restartpoints.
9011 : */
9012 : static void
9013 2832 : CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
9014 : {
9015 2832 : CheckPointCLOG();
9016 2832 : CheckPointCommitTs();
9017 2832 : CheckPointSUBTRANS();
9018 2832 : CheckPointMultiXact();
9019 2832 : CheckPointPredicate();
9020 2832 : CheckPointRelationMap();
9021 2832 : CheckPointReplicationSlots();
9022 2832 : CheckPointSnapBuild();
9023 2832 : CheckPointLogicalRewriteHeap();
9024 2832 : CheckPointBuffers(flags); /* performs all required fsyncs */
9025 2832 : CheckPointReplicationOrigin();
9026 : /* We deliberately delay 2PC checkpointing as long as possible */
9027 2832 : CheckPointTwoPhase(checkPointRedo);
9028 2832 : }
9029 :
9030 : /*
9031 : * Save a checkpoint for recovery restart if appropriate
9032 : *
9033 : * This function is called each time a checkpoint record is read from XLOG.
9034 : * It must determine whether the checkpoint represents a safe restartpoint or
9035 : * not. If so, the checkpoint record is stashed in shared memory so that
9036 : * CreateRestartPoint can consult it. (Note that the latter function is
9037 : * executed by the checkpointer, while this one will be executed by the
9038 : * startup process.)
9039 : */
9040 : static void
9041 162 : RecoveryRestartPoint(const CheckPoint *checkPoint)
9042 : {
9043 : /*
9044 : * Also refrain from creating a restartpoint if we have seen any
9045 : * references to non-existent pages. Restarting recovery from the
9046 : * restartpoint would not see the references, so we would lose the
9047 : * cross-check that the pages belonged to a relation that was dropped
9048 : * later.
9049 : */
9050 162 : if (XLogHaveInvalidPages())
9051 : {
9052 0 : elog(trace_recovery(DEBUG2),
9053 : "could not record restart point at %X/%X because there "
9054 : "are unresolved references to invalid pages",
9055 : (uint32) (checkPoint->redo >> 32),
9056 : (uint32) checkPoint->redo);
9057 0 : return;
9058 : }
9059 :
9060 : /*
9061 : * Copy the checkpoint record to shared memory, so that checkpointer can
9062 : * work out the next time it wants to perform a restartpoint.
9063 : */
9064 162 : SpinLockAcquire(&XLogCtl->info_lck);
9065 162 : XLogCtl->lastCheckPointRecPtr = ReadRecPtr;
9066 162 : XLogCtl->lastCheckPointEndPtr = EndRecPtr;
9067 162 : XLogCtl->lastCheckPoint = *checkPoint;
9068 162 : SpinLockRelease(&XLogCtl->info_lck);
9069 : }
9070 :
9071 : /*
9072 : * Establish a restartpoint if possible.
9073 : *
9074 : * This is similar to CreateCheckPoint, but is used during WAL recovery
9075 : * to establish a point from which recovery can roll forward without
9076 : * replaying the entire recovery log.
9077 : *
9078 : * Returns true if a new restartpoint was established. We can only establish
9079 : * a restartpoint if we have replayed a safe checkpoint record since last
9080 : * restartpoint.
9081 : */
9082 : bool
9083 34 : CreateRestartPoint(int flags)
9084 : {
9085 : XLogRecPtr lastCheckPointRecPtr;
9086 : XLogRecPtr lastCheckPointEndPtr;
9087 : CheckPoint lastCheckPoint;
9088 : XLogRecPtr PriorRedoPtr;
9089 : XLogRecPtr receivePtr;
9090 : XLogRecPtr replayPtr;
9091 : TimeLineID replayTLI;
9092 : XLogRecPtr endptr;
9093 : XLogSegNo _logSegNo;
9094 : TimestampTz xtime;
9095 :
9096 : /*
9097 : * Acquire CheckpointLock to ensure only one restartpoint or checkpoint
9098 : * happens at a time.
9099 : */
9100 34 : LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
9101 :
9102 : /* Get a local copy of the last safe checkpoint record. */
9103 34 : SpinLockAcquire(&XLogCtl->info_lck);
9104 34 : lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
9105 34 : lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
9106 34 : lastCheckPoint = XLogCtl->lastCheckPoint;
9107 34 : SpinLockRelease(&XLogCtl->info_lck);
9108 :
9109 : /*
9110 : * Check that we're still in recovery mode. It's ok if we exit recovery
9111 : * mode after this check, the restart point is valid anyway.
9112 : */
9113 34 : if (!RecoveryInProgress())
9114 : {
9115 0 : ereport(DEBUG2,
9116 : (errmsg("skipping restartpoint, recovery has already ended")));
9117 0 : LWLockRelease(CheckpointLock);
9118 0 : return false;
9119 : }
9120 :
9121 : /*
9122 : * If the last checkpoint record we've replayed is already our last
9123 : * restartpoint, we can't perform a new restart point. We still update
9124 : * minRecoveryPoint in that case, so that if this is a shutdown restart
9125 : * point, we won't start up earlier than before. That's not strictly
9126 : * necessary, but when hot standby is enabled, it would be rather weird if
9127 : * the database opened up for read-only connections at a point-in-time
9128 : * before the last shutdown. Such time travel is still possible in case of
9129 : * immediate shutdown, though.
9130 : *
9131 : * We don't explicitly advance minRecoveryPoint when we do create a
9132 : * restartpoint. It's assumed that flushing the buffers will do that as a
9133 : * side-effect.
9134 : */
9135 68 : if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
9136 34 : lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
9137 : {
9138 18 : ereport(DEBUG2,
9139 : (errmsg("skipping restartpoint, already performed at %X/%X",
9140 : (uint32) (lastCheckPoint.redo >> 32),
9141 : (uint32) lastCheckPoint.redo)));
9142 :
9143 18 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
9144 18 : if (flags & CHECKPOINT_IS_SHUTDOWN)
9145 : {
9146 16 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9147 16 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
9148 16 : ControlFile->time = (pg_time_t) time(NULL);
9149 16 : UpdateControlFile();
9150 16 : LWLockRelease(ControlFileLock);
9151 : }
9152 18 : LWLockRelease(CheckpointLock);
9153 18 : return false;
9154 : }
9155 :
9156 : /*
9157 : * Update the shared RedoRecPtr so that the startup process can calculate
9158 : * the number of segments replayed since last restartpoint, and request a
9159 : * restartpoint if it exceeds CheckPointSegments.
9160 : *
9161 : * Like in CreateCheckPoint(), hold off insertions to update it, although
9162 : * during recovery this is just pro forma, because no WAL insertions are
9163 : * happening.
9164 : */
9165 16 : WALInsertLockAcquireExclusive();
9166 16 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
9167 16 : WALInsertLockRelease();
9168 :
9169 : /* Also update the info_lck-protected copy */
9170 16 : SpinLockAcquire(&XLogCtl->info_lck);
9171 16 : XLogCtl->RedoRecPtr = lastCheckPoint.redo;
9172 16 : SpinLockRelease(&XLogCtl->info_lck);
9173 :
9174 : /*
9175 : * Prepare to accumulate statistics.
9176 : *
9177 : * Note: because it is possible for log_checkpoints to change while a
9178 : * checkpoint proceeds, we always accumulate stats, even if
9179 : * log_checkpoints is currently off.
9180 : */
9181 16 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
9182 16 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
9183 :
9184 16 : if (log_checkpoints)
9185 4 : LogCheckpointStart(flags, true);
9186 :
9187 16 : CheckPointGuts(lastCheckPoint.redo, flags);
9188 :
9189 : /*
9190 : * Remember the prior checkpoint's redo ptr for
9191 : * UpdateCheckPointDistanceEstimate()
9192 : */
9193 16 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
9194 :
9195 : /*
9196 : * Update pg_control, using current time. Check that it still shows
9197 : * DB_IN_ARCHIVE_RECOVERY state and an older checkpoint, else do nothing;
9198 : * this is a quick hack to make sure nothing really bad happens if somehow
9199 : * we get here after the end-of-recovery checkpoint.
9200 : */
9201 16 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9202 32 : if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY &&
9203 16 : ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
9204 : {
9205 16 : ControlFile->checkPoint = lastCheckPointRecPtr;
9206 16 : ControlFile->checkPointCopy = lastCheckPoint;
9207 16 : ControlFile->time = (pg_time_t) time(NULL);
9208 :
9209 : /*
9210 : * Ensure minRecoveryPoint is past the checkpoint record. Normally,
9211 : * this will have happened already while writing out dirty buffers,
9212 : * but not necessarily - e.g. because no buffers were dirtied. We do
9213 : * this because a non-exclusive base backup uses minRecoveryPoint to
9214 : * determine which WAL files must be included in the backup, and the
9215 : * file (or files) containing the checkpoint record must be included,
9216 : * at a minimum. Note that for an ordinary restart of recovery there's
9217 : * no value in having the minimum recovery point any earlier than this
9218 : * anyway, because redo will begin just after the checkpoint record.
9219 : */
9220 16 : if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
9221 : {
9222 2 : ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
9223 2 : ControlFile->minRecoveryPointTLI = lastCheckPoint.ThisTimeLineID;
9224 :
9225 : /* update local copy */
9226 2 : minRecoveryPoint = ControlFile->minRecoveryPoint;
9227 2 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
9228 : }
9229 16 : if (flags & CHECKPOINT_IS_SHUTDOWN)
9230 6 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
9231 16 : UpdateControlFile();
9232 : }
9233 16 : LWLockRelease(ControlFileLock);
9234 :
9235 : /*
9236 : * Update the average distance between checkpoints/restartpoints if the
9237 : * prior checkpoint exists.
9238 : */
9239 16 : if (PriorRedoPtr != InvalidXLogRecPtr)
9240 16 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
9241 :
9242 : /*
9243 : * Delete old log files, those no longer needed for last restartpoint to
9244 : * prevent the disk holding the xlog from growing full.
9245 : */
9246 16 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
9247 :
9248 : /*
9249 : * Retreat _logSegNo using the current end of xlog replayed or received,
9250 : * whichever is later.
9251 : */
9252 16 : receivePtr = GetWalRcvWriteRecPtr(NULL, NULL);
9253 16 : replayPtr = GetXLogReplayRecPtr(&replayTLI);
9254 16 : endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
9255 16 : KeepLogSeg(endptr, &_logSegNo);
9256 16 : _logSegNo--;
9257 :
9258 : /*
9259 : * Try to recycle segments on a useful timeline. If we've been promoted
9260 : * since the beginning of this restartpoint, use the new timeline chosen
9261 : * at end of recovery (RecoveryInProgress() sets ThisTimeLineID in that
9262 : * case). If we're still in recovery, use the timeline we're currently
9263 : * replaying.
9264 : *
9265 : * There is no guarantee that the WAL segments will be useful on the
9266 : * current timeline; if recovery proceeds to a new timeline right after
9267 : * this, the pre-allocated WAL segments on this timeline will not be used,
9268 : * and will go wasted until recycled on the next restartpoint. We'll live
9269 : * with that.
9270 : */
9271 16 : if (RecoveryInProgress())
9272 16 : ThisTimeLineID = replayTLI;
9273 :
9274 16 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, endptr);
9275 :
9276 : /*
9277 : * Make more log segments if needed. (Do this after recycling old log
9278 : * segments, since that may supply some of the needed files.)
9279 : */
9280 16 : PreallocXlogFiles(endptr);
9281 :
9282 : /*
9283 : * ThisTimeLineID is normally not set when we're still in recovery.
9284 : * However, recycling/preallocating segments above needed ThisTimeLineID
9285 : * to determine which timeline to install the segments on. Reset it now,
9286 : * to restore the normal state of affairs for debugging purposes.
9287 : */
9288 16 : if (RecoveryInProgress())
9289 16 : ThisTimeLineID = 0;
9290 :
9291 : /*
9292 : * Truncate pg_subtrans if possible. We can throw away all data before
9293 : * the oldest XMIN of any running transaction. No future transaction will
9294 : * attempt to reference any pg_subtrans entry older than that (see Asserts
9295 : * in subtrans.c). When hot standby is disabled, though, we mustn't do
9296 : * this because StartupSUBTRANS hasn't been called yet.
9297 : */
9298 16 : if (EnableHotStandby)
9299 16 : TruncateSUBTRANS(GetOldestXmin(NULL, PROCARRAY_FLAGS_DEFAULT));
9300 :
9301 : /* Real work is done, but log and update before releasing lock. */
9302 16 : LogCheckpointEnd(true);
9303 :
9304 16 : xtime = GetLatestXTime();
9305 16 : ereport((log_checkpoints ? LOG : DEBUG2),
9306 : (errmsg("recovery restart point at %X/%X",
9307 : (uint32) (lastCheckPoint.redo >> 32), (uint32) lastCheckPoint.redo),
9308 : xtime ? errdetail("Last completed transaction was at log time %s.",
9309 : timestamptz_to_str(xtime)) : 0));
9310 :
9311 16 : LWLockRelease(CheckpointLock);
9312 :
9313 : /*
9314 : * Finally, execute archive_cleanup_command, if any.
9315 : */
9316 16 : if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
9317 0 : ExecuteRecoveryCommand(archiveCleanupCommand,
9318 : "archive_cleanup_command",
9319 : false);
9320 :
9321 16 : return true;
9322 : }
9323 :
9324 : /*
9325 : * Retreat *logSegNo to the last segment that we need to retain because of
9326 : * either wal_keep_segments or replication slots.
9327 : *
9328 : * This is calculated by subtracting wal_keep_segments from the given xlog
9329 : * location, recptr and by making sure that that result is below the
9330 : * requirement of replication slots.
9331 : */
9332 : static void
9333 2832 : KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
9334 : {
9335 : XLogSegNo segno;
9336 : XLogRecPtr keep;
9337 :
9338 2832 : XLByteToSeg(recptr, segno, wal_segment_size);
9339 2832 : keep = XLogGetReplicationSlotMinimumLSN();
9340 :
9341 : /* compute limit for wal_keep_segments first */
9342 2832 : if (wal_keep_segments > 0)
9343 : {
9344 : /* avoid underflow, don't go below 1 */
9345 74 : if (segno <= wal_keep_segments)
9346 74 : segno = 1;
9347 : else
9348 0 : segno = segno - wal_keep_segments;
9349 : }
9350 :
9351 : /* then check whether slots limit removal further */
9352 2832 : if (max_replication_slots > 0 && keep != InvalidXLogRecPtr)
9353 : {
9354 : XLogSegNo slotSegNo;
9355 :
9356 52 : XLByteToSeg(keep, slotSegNo, wal_segment_size);
9357 :
9358 52 : if (slotSegNo <= 0)
9359 0 : segno = 1;
9360 52 : else if (slotSegNo < segno)
9361 16 : segno = slotSegNo;
9362 : }
9363 :
9364 : /* don't delete WAL segments newer than the calculated segment */
9365 2832 : if (segno < *logSegNo)
9366 88 : *logSegNo = segno;
9367 2832 : }
9368 :
9369 : /*
9370 : * Write a NEXTOID log record
9371 : */
9372 : void
9373 1042 : XLogPutNextOid(Oid nextOid)
9374 : {
9375 1042 : XLogBeginInsert();
9376 1042 : XLogRegisterData((char *) (&nextOid), sizeof(Oid));
9377 1042 : (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
9378 :
9379 : /*
9380 : * We need not flush the NEXTOID record immediately, because any of the
9381 : * just-allocated OIDs could only reach disk as part of a tuple insert or
9382 : * update that would have its own XLOG record that must follow the NEXTOID
9383 : * record. Therefore, the standard buffer LSN interlock applied to those
9384 : * records will ensure no such OID reaches disk before the NEXTOID record
9385 : * does.
9386 : *
9387 : * Note, however, that the above statement only covers state "within" the
9388 : * database. When we use a generated OID as a file or directory name, we
9389 : * are in a sense violating the basic WAL rule, because that filesystem
9390 : * change may reach disk before the NEXTOID WAL record does. The impact
9391 : * of this is that if a database crash occurs immediately afterward, we
9392 : * might after restart re-generate the same OID and find that it conflicts
9393 : * with the leftover file or directory. But since for safety's sake we
9394 : * always loop until finding a nonconflicting filename, this poses no real
9395 : * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
9396 : */
9397 1042 : }
9398 :
9399 : /*
9400 : * Write an XLOG SWITCH record.
9401 : *
9402 : * Here we just blindly issue an XLogInsert request for the record.
9403 : * All the magic happens inside XLogInsert.
9404 : *
9405 : * The return value is either the end+1 address of the switch record,
9406 : * or the end+1 address of the prior segment if we did not need to
9407 : * write a switch record because we are already at segment start.
9408 : */
9409 : XLogRecPtr
9410 192 : RequestXLogSwitch(bool mark_unimportant)
9411 : {
9412 : XLogRecPtr RecPtr;
9413 :
9414 : /* XLOG SWITCH has no data */
9415 192 : XLogBeginInsert();
9416 :
9417 192 : if (mark_unimportant)
9418 0 : XLogSetRecordFlags(XLOG_MARK_UNIMPORTANT);
9419 192 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH);
9420 :
9421 192 : return RecPtr;
9422 : }
9423 :
9424 : /*
9425 : * Write a RESTORE POINT record
9426 : */
9427 : XLogRecPtr
9428 2 : XLogRestorePoint(const char *rpName)
9429 : {
9430 : XLogRecPtr RecPtr;
9431 : xl_restore_point xlrec;
9432 :
9433 2 : xlrec.rp_time = GetCurrentTimestamp();
9434 2 : strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
9435 :
9436 2 : XLogBeginInsert();
9437 2 : XLogRegisterData((char *) &xlrec, sizeof(xl_restore_point));
9438 :
9439 2 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
9440 :
9441 2 : ereport(LOG,
9442 : (errmsg("restore point \"%s\" created at %X/%X",
9443 : rpName, (uint32) (RecPtr >> 32), (uint32) RecPtr)));
9444 :
9445 2 : return RecPtr;
9446 : }
9447 :
9448 : /*
9449 : * Check if any of the GUC parameters that are critical for hot standby
9450 : * have changed, and update the value in pg_control file if necessary.
9451 : */
9452 : static void
9453 1170 : XLogReportParameters(void)
9454 : {
9455 2208 : if (wal_level != ControlFile->wal_level ||
9456 2024 : wal_log_hints != ControlFile->wal_log_hints ||
9457 1970 : MaxConnections != ControlFile->MaxConnections ||
9458 1968 : max_worker_processes != ControlFile->max_worker_processes ||
9459 1968 : max_wal_senders != ControlFile->max_wal_senders ||
9460 1842 : max_prepared_xacts != ControlFile->max_prepared_xacts ||
9461 1716 : max_locks_per_xact != ControlFile->max_locks_per_xact ||
9462 858 : 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 324 : if (wal_level != ControlFile->wal_level || XLogIsNeeded())
9472 : {
9473 : xl_parameter_change xlrec;
9474 : XLogRecPtr recptr;
9475 :
9476 324 : xlrec.MaxConnections = MaxConnections;
9477 324 : xlrec.max_worker_processes = max_worker_processes;
9478 324 : xlrec.max_wal_senders = max_wal_senders;
9479 324 : xlrec.max_prepared_xacts = max_prepared_xacts;
9480 324 : xlrec.max_locks_per_xact = max_locks_per_xact;
9481 324 : xlrec.wal_level = wal_level;
9482 324 : xlrec.wal_log_hints = wal_log_hints;
9483 324 : xlrec.track_commit_timestamp = track_commit_timestamp;
9484 :
9485 324 : XLogBeginInsert();
9486 324 : XLogRegisterData((char *) &xlrec, sizeof(xlrec));
9487 :
9488 324 : recptr = XLogInsert(RM_XLOG_ID, XLOG_PARAMETER_CHANGE);
9489 324 : XLogFlush(recptr);
9490 : }
9491 :
9492 324 : ControlFile->MaxConnections = MaxConnections;
9493 324 : ControlFile->max_worker_processes = max_worker_processes;
9494 324 : ControlFile->max_wal_senders = max_wal_senders;
9495 324 : ControlFile->max_prepared_xacts = max_prepared_xacts;
9496 324 : ControlFile->max_locks_per_xact = max_locks_per_xact;
9497 324 : ControlFile->wal_level = wal_level;
9498 324 : ControlFile->wal_log_hints = wal_log_hints;
9499 324 : ControlFile->track_commit_timestamp = track_commit_timestamp;
9500 324 : UpdateControlFile();
9501 : }
9502 1170 : }
9503 :
9504 : /*
9505 : * Update full_page_writes in shared memory, and write an
9506 : * XLOG_FPW_CHANGE record if necessary.
9507 : *
9508 : * Note: this function assumes there is no other process running
9509 : * concurrently that could update it.
9510 : */
9511 : void
9512 1548 : UpdateFullPageWrites(void)
9513 : {
9514 1548 : XLogCtlInsert *Insert = &XLogCtl->Insert;
9515 : bool recoveryInProgress;
9516 :
9517 : /*
9518 : * Do nothing if full_page_writes has not been changed.
9519 : *
9520 : * It's safe to check the shared full_page_writes without the lock,
9521 : * because we assume that there is no concurrently running process which
9522 : * can update it.
9523 : */
9524 1548 : if (fullPageWrites == Insert->fullPageWrites)
9525 1504 : return;
9526 :
9527 : /*
9528 : * Perform this outside critical section so that the WAL insert
9529 : * initialization done by RecoveryInProgress() doesn't trigger an
9530 : * assertion failure.
9531 : */
9532 44 : recoveryInProgress = RecoveryInProgress();
9533 :
9534 44 : START_CRIT_SECTION();
9535 :
9536 : /*
9537 : * It's always safe to take full page images, even when not strictly
9538 : * required, but not the other round. So if we're setting full_page_writes
9539 : * to true, first set it true and then write the WAL record. If we're
9540 : * setting it to false, first write the WAL record and then set the global
9541 : * flag.
9542 : */
9543 44 : if (fullPageWrites)
9544 : {
9545 42 : WALInsertLockAcquireExclusive();
9546 42 : Insert->fullPageWrites = true;
9547 42 : WALInsertLockRelease();
9548 : }
9549 :
9550 : /*
9551 : * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
9552 : * full_page_writes during archive recovery, if required.
9553 : */
9554 44 : if (XLogStandbyInfoActive() && !recoveryInProgress)
9555 : {
9556 0 : XLogBeginInsert();
9557 0 : XLogRegisterData((char *) (&fullPageWrites), sizeof(bool));
9558 :
9559 0 : XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
9560 : }
9561 :
9562 44 : if (!fullPageWrites)
9563 : {
9564 2 : WALInsertLockAcquireExclusive();
9565 2 : Insert->fullPageWrites = false;
9566 2 : WALInsertLockRelease();
9567 : }
9568 44 : END_CRIT_SECTION();
9569 : }
9570 :
9571 : /*
9572 : * Check that it's OK to switch to new timeline during recovery.
9573 : *
9574 : * 'lsn' is the address of the shutdown checkpoint record we're about to
9575 : * replay. (Currently, timeline can only change at a shutdown checkpoint).
9576 : */
9577 : static void
9578 10 : checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI, TimeLineID prevTLI)
9579 : {
9580 : /* Check that the record agrees on what the current (old) timeline is */
9581 10 : if (prevTLI != ThisTimeLineID)
9582 0 : ereport(PANIC,
9583 : (errmsg("unexpected previous timeline ID %u (current timeline ID %u) in checkpoint record",
9584 : prevTLI, ThisTimeLineID)));
9585 :
9586 : /*
9587 : * The new timeline better be in the list of timelines we expect to see,
9588 : * according to the timeline history. It should also not decrease.
9589 : */
9590 10 : if (newTLI < ThisTimeLineID || !tliInHistory(newTLI, expectedTLEs))
9591 0 : ereport(PANIC,
9592 : (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
9593 : newTLI, ThisTimeLineID)));
9594 :
9595 : /*
9596 : * If we have not yet reached min recovery point, and we're about to
9597 : * switch to a timeline greater than the timeline of the min recovery
9598 : * point: trouble. After switching to the new timeline, we could not
9599 : * possibly visit the min recovery point on the correct timeline anymore.
9600 : * This can happen if there is a newer timeline in the archive that
9601 : * branched before the timeline the min recovery point is on, and you
9602 : * attempt to do PITR to the new timeline.
9603 : */
9604 18 : if (!XLogRecPtrIsInvalid(minRecoveryPoint) &&
9605 8 : lsn < minRecoveryPoint &&
9606 0 : newTLI > minRecoveryPointTLI)
9607 0 : ereport(PANIC,
9608 : (errmsg("unexpected timeline ID %u in checkpoint record, before reaching minimum recovery point %X/%X on timeline %u",
9609 : newTLI,
9610 : (uint32) (minRecoveryPoint >> 32),
9611 : (uint32) minRecoveryPoint,
9612 : minRecoveryPointTLI)));
9613 :
9614 : /* Looks good */
9615 10 : }
9616 :
9617 : /*
9618 : * XLOG resource manager's routines
9619 : *
9620 : * Definitions of info values are in include/catalog/pg_control.h, though
9621 : * not all record types are related to control file updates.
9622 : */
9623 : void
9624 1602 : xlog_redo(XLogReaderState *record)
9625 : {
9626 1602 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
9627 1602 : XLogRecPtr lsn = record->EndRecPtr;
9628 :
9629 : /* in XLOG rmgr, backup blocks are only used by XLOG_FPI records */
9630 : Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
9631 : !XLogRecHasAnyBlockRefs(record));
9632 :
9633 1602 : if (info == XLOG_NEXTOID)
9634 : {
9635 : Oid nextOid;
9636 :
9637 : /*
9638 : * We used to try to take the maximum of ShmemVariableCache->nextOid
9639 : * and the recorded nextOid, but that fails if the OID counter wraps
9640 : * around. Since no OID allocation should be happening during replay
9641 : * anyway, better to just believe the record exactly. We still take
9642 : * OidGenLock while setting the variable, just in case.
9643 : */
9644 42 : memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
9645 42 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
9646 42 : ShmemVariableCache->nextOid = nextOid;
9647 42 : ShmemVariableCache->oidCount = 0;
9648 42 : LWLockRelease(OidGenLock);
9649 : }
9650 1560 : else if (info == XLOG_CHECKPOINT_SHUTDOWN)
9651 : {
9652 : CheckPoint checkPoint;
9653 :
9654 34 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
9655 : /* In a SHUTDOWN checkpoint, believe the counters exactly */
9656 34 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
9657 34 : ShmemVariableCache->nextFullXid = checkPoint.nextFullXid;
9658 34 : LWLockRelease(XidGenLock);
9659 34 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
9660 34 : ShmemVariableCache->nextOid = checkPoint.nextOid;
9661 34 : ShmemVariableCache->oidCount = 0;
9662 34 : LWLockRelease(OidGenLock);
9663 34 : MultiXactSetNextMXact(checkPoint.nextMulti,
9664 : checkPoint.nextMultiOffset);
9665 :
9666 34 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
9667 : checkPoint.oldestMultiDB);
9668 :
9669 : /*
9670 : * No need to set oldestClogXid here as well; it'll be set when we
9671 : * redo an xl_clog_truncate if it changed since initialization.
9672 : */
9673 34 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
9674 :
9675 : /*
9676 : * If we see a shutdown checkpoint while waiting for an end-of-backup
9677 : * record, the backup was canceled and the end-of-backup record will
9678 : * never arrive.
9679 : */
9680 68 : if (ArchiveRecoveryRequested &&
9681 34 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint) &&
9682 0 : XLogRecPtrIsInvalid(ControlFile->backupEndPoint))
9683 0 : ereport(PANIC,
9684 : (errmsg("online backup was canceled, recovery cannot continue")));
9685 :
9686 : /*
9687 : * If we see a shutdown checkpoint, we know that nothing was running
9688 : * on the master at this point. So fake-up an empty running-xacts
9689 : * record and use that here and now. Recover additional standby state
9690 : * for prepared transactions.
9691 : */
9692 34 : if (standbyState >= STANDBY_INITIALIZED)
9693 : {
9694 : TransactionId *xids;
9695 : int nxids;
9696 : TransactionId oldestActiveXID;
9697 : TransactionId latestCompletedXid;
9698 : RunningTransactionsData running;
9699 :
9700 34 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
9701 :
9702 : /*
9703 : * Construct a RunningTransactions snapshot representing a shut
9704 : * down server, with only prepared transactions still alive. We're
9705 : * never overflowed at this point because all subxids are listed
9706 : * with their parent prepared transactions.
9707 : */
9708 34 : running.xcnt = nxids;
9709 34 : running.subxcnt = 0;
9710 34 : running.subxid_overflow = false;
9711 34 : running.nextXid = XidFromFullTransactionId(checkPoint.nextFullXid);
9712 34 : running.oldestRunningXid = oldestActiveXID;
9713 34 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextFullXid);
9714 34 : TransactionIdRetreat(latestCompletedXid);
9715 : Assert(TransactionIdIsNormal(latestCompletedXid));
9716 34 : running.latestCompletedXid = latestCompletedXid;
9717 34 : running.xids = xids;
9718 :
9719 34 : ProcArrayApplyRecoveryInfo(&running);
9720 :
9721 34 : StandbyRecoverPreparedTransactions();
9722 : }
9723 :
9724 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
9725 34 : ControlFile->checkPointCopy.nextFullXid = checkPoint.nextFullXid;
9726 :
9727 : /* Update shared-memory copy of checkpoint XID/epoch */
9728 34 : SpinLockAcquire(&XLogCtl->info_lck);
9729 34 : XLogCtl->ckptFullXid = checkPoint.nextFullXid;
9730 34 : SpinLockRelease(&XLogCtl->info_lck);
9731 :
9732 : /*
9733 : * We should've already switched to the new TLI before replaying this
9734 : * record.
9735 : */
9736 34 : if (checkPoint.ThisTimeLineID != ThisTimeLineID)
9737 0 : ereport(PANIC,
9738 : (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
9739 : checkPoint.ThisTimeLineID, ThisTimeLineID)));
9740 :
9741 34 : RecoveryRestartPoint(&checkPoint);
9742 : }
9743 1526 : else if (info == XLOG_CHECKPOINT_ONLINE)
9744 : {
9745 : CheckPoint checkPoint;
9746 :
9747 128 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
9748 : /* In an ONLINE checkpoint, treat the XID counter as a minimum */
9749 128 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
9750 128 : if (FullTransactionIdPrecedes(ShmemVariableCache->nextFullXid,
9751 : checkPoint.nextFullXid))
9752 0 : ShmemVariableCache->nextFullXid = checkPoint.nextFullXid;
9753 128 : LWLockRelease(XidGenLock);
9754 :
9755 : /*
9756 : * We ignore the nextOid counter in an ONLINE checkpoint, preferring
9757 : * to track OID assignment through XLOG_NEXTOID records. The nextOid
9758 : * counter is from the start of the checkpoint and might well be stale
9759 : * compared to later XLOG_NEXTOID records. We could try to take the
9760 : * maximum of the nextOid counter and our latest value, but since
9761 : * there's no particular guarantee about the speed with which the OID
9762 : * counter wraps around, that's a risky thing to do. In any case,
9763 : * users of the nextOid counter are required to avoid assignment of
9764 : * duplicates, so that a somewhat out-of-date value should be safe.
9765 : */
9766 :
9767 : /* Handle multixact */
9768 128 : MultiXactAdvanceNextMXact(checkPoint.nextMulti,
9769 : checkPoint.nextMultiOffset);
9770 :
9771 : /*
9772 : * NB: This may perform multixact truncation when replaying WAL
9773 : * generated by an older primary.
9774 : */
9775 128 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
9776 : checkPoint.oldestMultiDB);
9777 128 : if (TransactionIdPrecedes(ShmemVariableCache->oldestXid,
9778 : checkPoint.oldestXid))
9779 0 : SetTransactionIdLimit(checkPoint.oldestXid,
9780 : checkPoint.oldestXidDB);
9781 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
9782 128 : ControlFile->checkPointCopy.nextFullXid = checkPoint.nextFullXid;
9783 :
9784 : /* Update shared-memory copy of checkpoint XID/epoch */
9785 128 : SpinLockAcquire(&XLogCtl->info_lck);
9786 128 : XLogCtl->ckptFullXid = checkPoint.nextFullXid;
9787 128 : SpinLockRelease(&XLogCtl->info_lck);
9788 :
9789 : /* TLI should not change in an on-line checkpoint */
9790 128 : if (checkPoint.ThisTimeLineID != ThisTimeLineID)
9791 0 : ereport(PANIC,
9792 : (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
9793 : checkPoint.ThisTimeLineID, ThisTimeLineID)));
9794 :
9795 128 : RecoveryRestartPoint(&checkPoint);
9796 : }
9797 1398 : else if (info == XLOG_END_OF_RECOVERY)
9798 : {
9799 : xl_end_of_recovery xlrec;
9800 :
9801 10 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
9802 :
9803 : /*
9804 : * For Hot Standby, we could treat this like a Shutdown Checkpoint,
9805 : * but this case is rarer and harder to test, so the benefit doesn't
9806 : * outweigh the potential extra cost of maintenance.
9807 : */
9808 :
9809 : /*
9810 : * We should've already switched to the new TLI before replaying this
9811 : * record.
9812 : */
9813 10 : if (xlrec.ThisTimeLineID != ThisTimeLineID)
9814 0 : ereport(PANIC,
9815 : (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
9816 : xlrec.ThisTimeLineID, ThisTimeLineID)));
9817 : }
9818 1388 : else if (info == XLOG_NOOP)
9819 : {
9820 : /* nothing to do here */
9821 : }
9822 1388 : else if (info == XLOG_SWITCH)
9823 : {
9824 : /* nothing to do here */
9825 : }
9826 1322 : else if (info == XLOG_RESTORE_POINT)
9827 : {
9828 : /* nothing to do here */
9829 : }
9830 1322 : else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
9831 : {
9832 : /*
9833 : * Full-page image (FPI) records contain nothing else but a backup
9834 : * block (or multiple backup blocks). Every block reference must
9835 : * include a full-page image - otherwise there would be no point in
9836 : * this record.
9837 : *
9838 : * No recovery conflicts are generated by these generic records - if a
9839 : * resource manager needs to generate conflicts, it has to define a
9840 : * separate WAL record type and redo routine.
9841 : *
9842 : * XLOG_FPI_FOR_HINT records are generated when a page needs to be
9843 : * WAL- logged because of a hint bit update. They are only generated
9844 : * when checksums are enabled. There is no difference in handling
9845 : * XLOG_FPI and XLOG_FPI_FOR_HINT records, they use a different info
9846 : * code just to distinguish them for statistics purposes.
9847 : */
9848 2492 : for (uint8 block_id = 0; block_id <= record->max_block_id; block_id++)
9849 : {
9850 : Buffer buffer;
9851 :
9852 1246 : if (XLogReadBufferForRedo(record, block_id, &buffer) != BLK_RESTORED)
9853 0 : elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
9854 1246 : UnlockReleaseBuffer(buffer);
9855 : }
9856 : }
9857 76 : else if (info == XLOG_BACKUP_END)
9858 : {
9859 : XLogRecPtr startpoint;
9860 :
9861 62 : memcpy(&startpoint, XLogRecGetData(record), sizeof(startpoint));
9862 :
9863 62 : if (ControlFile->backupStartPoint == startpoint)
9864 : {
9865 : /*
9866 : * We have reached the end of base backup, the point where
9867 : * pg_stop_backup() was done. The data on disk is now consistent.
9868 : * Reset backupStartPoint, and update minRecoveryPoint to make
9869 : * sure we don't allow starting up at an earlier point even if
9870 : * recovery is stopped and restarted soon after this.
9871 : */
9872 54 : elog(DEBUG1, "end of backup reached");
9873 :
9874 54 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9875 :
9876 54 : if (ControlFile->minRecoveryPoint < lsn)
9877 : {
9878 52 : ControlFile->minRecoveryPoint = lsn;
9879 52 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
9880 : }
9881 54 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
9882 54 : ControlFile->backupEndRequired = false;
9883 54 : UpdateControlFile();
9884 :
9885 54 : LWLockRelease(ControlFileLock);
9886 : }
9887 : }
9888 14 : else if (info == XLOG_PARAMETER_CHANGE)
9889 : {
9890 : xl_parameter_change xlrec;
9891 :
9892 : /* Update our copy of the parameters in pg_control */
9893 14 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
9894 :
9895 14 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9896 14 : ControlFile->MaxConnections = xlrec.MaxConnections;
9897 14 : ControlFile->max_worker_processes = xlrec.max_worker_processes;
9898 14 : ControlFile->max_wal_senders = xlrec.max_wal_senders;
9899 14 : ControlFile->max_prepared_xacts = xlrec.max_prepared_xacts;
9900 14 : ControlFile->max_locks_per_xact = xlrec.max_locks_per_xact;
9901 14 : ControlFile->wal_level = xlrec.wal_level;
9902 14 : ControlFile->wal_log_hints = xlrec.wal_log_hints;
9903 :
9904 : /*
9905 : * Update minRecoveryPoint to ensure that if recovery is aborted, we
9906 : * recover back up to this point before allowing hot standby again.
9907 : * This is important if the max_* settings are decreased, to ensure
9908 : * you don't run queries against the WAL preceding the change. The
9909 : * local copies cannot be updated as long as crash recovery is
9910 : * happening and we expect all the WAL to be replayed.
9911 : */
9912 14 : if (InArchiveRecovery)
9913 : {
9914 4 : minRecoveryPoint = ControlFile->minRecoveryPoint;
9915 4 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
9916 : }
9917 14 : if (minRecoveryPoint != InvalidXLogRecPtr && minRecoveryPoint < lsn)
9918 : {
9919 4 : ControlFile->minRecoveryPoint = lsn;
9920 4 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
9921 : }
9922 :
9923 14 : CommitTsParameterChange(xlrec.track_commit_timestamp,
9924 14 : ControlFile->track_commit_timestamp);
9925 14 : ControlFile->track_commit_timestamp = xlrec.track_commit_timestamp;
9926 :
9927 14 : UpdateControlFile();
9928 14 : LWLockRelease(ControlFileLock);
9929 :
9930 : /* Check to see if any parameter change gives a problem on recovery */
9931 14 : CheckRequiredParameterValues();
9932 : }
9933 0 : else if (info == XLOG_FPW_CHANGE)
9934 : {
9935 : bool fpw;
9936 :
9937 0 : memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
9938 :
9939 : /*
9940 : * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
9941 : * do_pg_start_backup() and do_pg_stop_backup() can check whether
9942 : * full_page_writes has been disabled during online backup.
9943 : */
9944 0 : if (!fpw)
9945 : {
9946 0 : SpinLockAcquire(&XLogCtl->info_lck);
9947 0 : if (XLogCtl->lastFpwDisableRecPtr < ReadRecPtr)
9948 0 : XLogCtl->lastFpwDisableRecPtr = ReadRecPtr;
9949 0 : SpinLockRelease(&XLogCtl->info_lck);
9950 : }
9951 :
9952 : /* Keep track of full_page_writes */
9953 0 : lastFullPageWrites = fpw;
9954 : }
9955 1602 : }
9956 :
9957 : #ifdef WAL_DEBUG
9958 :
9959 : static void
9960 : xlog_outrec(StringInfo buf, XLogReaderState *record)
9961 : {
9962 : int block_id;
9963 :
9964 : appendStringInfo(buf, "prev %X/%X; xid %u",
9965 : (uint32) (XLogRecGetPrev(record) >> 32),
9966 : (uint32) XLogRecGetPrev(record),
9967 : XLogRecGetXid(record));
9968 :
9969 : appendStringInfo(buf, "; len %u",
9970 : XLogRecGetDataLen(record));
9971 :
9972 : /* decode block references */
9973 : for (block_id = 0; block_id <= record->max_block_id; block_id++)
9974 : {
9975 : RelFileNode rnode;
9976 : ForkNumber forknum;
9977 : BlockNumber blk;
9978 :
9979 : if (!XLogRecHasBlockRef(record, block_id))
9980 : continue;
9981 :
9982 : XLogRecGetBlockTag(record, block_id, &rnode, &forknum, &blk);
9983 : if (forknum != MAIN_FORKNUM)
9984 : appendStringInfo(buf, "; blkref #%u: rel %u/%u/%u, fork %u, blk %u",
9985 : block_id,
9986 : rnode.spcNode, rnode.dbNode, rnode.relNode,
9987 : forknum,
9988 : blk);
9989 : else
9990 : appendStringInfo(buf, "; blkref #%u: rel %u/%u/%u, blk %u",
9991 : block_id,
9992 : rnode.spcNode, rnode.dbNode, rnode.relNode,
9993 : blk);
9994 : if (XLogRecHasBlockImage(record, block_id))
9995 : appendStringInfoString(buf, " FPW");
9996 : }
9997 : }
9998 : #endif /* WAL_DEBUG */
9999 :
10000 : /*
10001 : * Returns a string describing an XLogRecord, consisting of its identity
10002 : * optionally followed by a colon, a space, and a further description.
10003 : */
10004 : static void
10005 6 : xlog_outdesc(StringInfo buf, XLogReaderState *record)
10006 : {
10007 6 : RmgrId rmid = XLogRecGetRmid(record);
10008 6 : uint8 info = XLogRecGetInfo(record);
10009 : const char *id;
10010 :
10011 6 : appendStringInfoString(buf, RmgrTable[rmid].rm_name);
10012 6 : appendStringInfoChar(buf, '/');
10013 :
10014 6 : id = RmgrTable[rmid].rm_identify(info);
10015 6 : if (id == NULL)
10016 0 : appendStringInfo(buf, "UNKNOWN (%X): ", info & ~XLR_INFO_MASK);
10017 : else
10018 6 : appendStringInfo(buf, "%s: ", id);
10019 :
10020 6 : RmgrTable[rmid].rm_desc(buf, record);
10021 6 : }
10022 :
10023 :
10024 : /*
10025 : * Return the (possible) sync flag used for opening a file, depending on the
10026 : * value of the GUC wal_sync_method.
10027 : */
10028 : static int
10029 8416 : get_sync_bit(int method)
10030 : {
10031 8416 : int o_direct_flag = 0;
10032 :
10033 : /* If fsync is disabled, never open in sync mode */
10034 8416 : if (!enableFsync)
10035 8384 : return 0;
10036 :
10037 : /*
10038 : * Optimize writes by bypassing kernel cache with O_DIRECT when using
10039 : * O_SYNC/O_FSYNC and O_DSYNC. But only if archiving and streaming are
10040 : * disabled, otherwise the archive command or walsender process will read
10041 : * the WAL soon after writing it, which is guaranteed to cause a physical
10042 : * read if we bypassed the kernel cache. We also skip the
10043 : * posix_fadvise(POSIX_FADV_DONTNEED) call in XLogFileClose() for the same
10044 : * reason.
10045 : *
10046 : * Never use O_DIRECT in walreceiver process for similar reasons; the WAL
10047 : * written by walreceiver is normally read by the startup process soon
10048 : * after its written. Also, walreceiver performs unaligned writes, which
10049 : * don't work with O_DIRECT, so it is required for correctness too.
10050 : */
10051 32 : if (!XLogIsNeeded() && !AmWalReceiverProcess())
10052 0 : o_direct_flag = PG_O_DIRECT;
10053 :
10054 32 : switch (method)
10055 : {
10056 : /*
10057 : * enum values for all sync options are defined even if they are
10058 : * not supported on the current platform. But if not, they are
10059 : * not included in the enum option array, and therefore will never
10060 : * be seen here.
10061 : */
10062 : case SYNC_METHOD_FSYNC:
10063 : case SYNC_METHOD_FSYNC_WRITETHROUGH:
10064 : case SYNC_METHOD_FDATASYNC:
10065 32 : return 0;
10066 : #ifdef OPEN_SYNC_FLAG
10067 : case SYNC_METHOD_OPEN:
10068 0 : return OPEN_SYNC_FLAG | o_direct_flag;
10069 : #endif
10070 : #ifdef OPEN_DATASYNC_FLAG
10071 : case SYNC_METHOD_OPEN_DSYNC:
10072 0 : return OPEN_DATASYNC_FLAG | o_direct_flag;
10073 : #endif
10074 : default:
10075 : /* can't happen (unless we are out of sync with option array) */
10076 0 : elog(ERROR, "unrecognized wal_sync_method: %d", method);
10077 : return 0; /* silence warning */
10078 : }
10079 : }
10080 :
10081 : /*
10082 : * GUC support
10083 : */
10084 : void
10085 1898 : assign_xlog_sync_method(int new_sync_method, void *extra)
10086 : {
10087 1898 : if (sync_method != new_sync_method)
10088 : {
10089 : /*
10090 : * To ensure that no blocks escape unsynced, force an fsync on the
10091 : * currently open log segment (if any). Also, if the open flag is
10092 : * changing, close the log file so it will be reopened (with new flag
10093 : * bit) at next use.
10094 : */
10095 0 : if (openLogFile >= 0)
10096 : {
10097 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN);
10098 0 : if (pg_fsync(openLogFile) != 0)
10099 0 : ereport(PANIC,
10100 : (errcode_for_file_access(),
10101 : errmsg("could not fsync file \"%s\": %m",
10102 : XLogFileNameP(ThisTimeLineID, openLogSegNo))));
10103 0 : pgstat_report_wait_end();
10104 0 : if (get_sync_bit(sync_method) != get_sync_bit(new_sync_method))
10105 0 : XLogFileClose();
10106 : }
10107 : }
10108 1898 : }
10109 :
10110 :
10111 : /*
10112 : * Issue appropriate kind of fsync (if any) for an XLOG output file.
10113 : *
10114 : * 'fd' is a file descriptor for the XLOG file to be fsync'd.
10115 : * 'segno' is for error reporting purposes.
10116 : */
10117 : void
10118 242628 : issue_xlog_fsync(int fd, XLogSegNo segno)
10119 : {
10120 242628 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC);
10121 242628 : switch (sync_method)
10122 : {
10123 : case SYNC_METHOD_FSYNC:
10124 0 : if (pg_fsync_no_writethrough(fd) != 0)
10125 0 : ereport(PANIC,
10126 : (errcode_for_file_access(),
10127 : errmsg("could not fsync file \"%s\": %m",
10128 : XLogFileNameP(ThisTimeLineID, segno))));
10129 0 : break;
10130 : #ifdef HAVE_FSYNC_WRITETHROUGH
10131 : case SYNC_METHOD_FSYNC_WRITETHROUGH:
10132 : if (pg_fsync_writethrough(fd) != 0)
10133 : ereport(PANIC,
10134 : (errcode_for_file_access(),
10135 : errmsg("could not fsync write-through file \"%s\": %m",
10136 : XLogFileNameP(ThisTimeLineID, segno))));
10137 : break;
10138 : #endif
10139 : #ifdef HAVE_FDATASYNC
10140 : case SYNC_METHOD_FDATASYNC:
10141 242628 : if (pg_fdatasync(fd) != 0)
10142 0 : ereport(PANIC,
10143 : (errcode_for_file_access(),
10144 : errmsg("could not fdatasync file \"%s\": %m",
10145 : XLogFileNameP(ThisTimeLineID, segno))));
10146 242628 : break;
10147 : #endif
10148 : case SYNC_METHOD_OPEN:
10149 : case SYNC_METHOD_OPEN_DSYNC:
10150 : /* write synced it already */
10151 0 : break;
10152 : default:
10153 0 : elog(PANIC, "unrecognized wal_sync_method: %d", sync_method);
10154 : break;
10155 : }
10156 242628 : pgstat_report_wait_end();
10157 242628 : }
10158 :
10159 : /*
10160 : * Return the filename of given log segment, as a palloc'd string.
10161 : */
10162 : char *
10163 0 : XLogFileNameP(TimeLineID tli, XLogSegNo segno)
10164 : {
10165 0 : char *result = palloc(MAXFNAMELEN);
10166 :
10167 0 : XLogFileName(result, tli, segno, wal_segment_size);
10168 0 : return result;
10169 : }
10170 :
10171 : /*
10172 : * do_pg_start_backup
10173 : *
10174 : * Utility function called at the start of an online backup. It creates the
10175 : * necessary starting checkpoint and constructs the backup label file.
10176 : *
10177 : * There are two kind of backups: exclusive and non-exclusive. An exclusive
10178 : * backup is started with pg_start_backup(), and there can be only one active
10179 : * at a time. The backup and tablespace map files of an exclusive backup are
10180 : * written to $PGDATA/backup_label and $PGDATA/tablespace_map, and they are
10181 : * removed by pg_stop_backup().
10182 : *
10183 : * A non-exclusive backup is used for the streaming base backups (see
10184 : * src/backend/replication/basebackup.c). The difference to exclusive backups
10185 : * is that the backup label and tablespace map files are not written to disk.
10186 : * Instead, their would-be contents are returned in *labelfile and *tblspcmapfile,
10187 : * and the caller is responsible for including them in the backup archive as
10188 : * 'backup_label' and 'tablespace_map'. There can be many non-exclusive backups
10189 : * active at the same time, and they don't conflict with an exclusive backup
10190 : * either.
10191 : *
10192 : * tblspcmapfile is required mainly for tar format in windows as native windows
10193 : * utilities are not able to create symlinks while extracting files from tar.
10194 : * However for consistency, the same is used for all platforms.
10195 : *
10196 : * needtblspcmapfile is true for the cases (exclusive backup and for
10197 : * non-exclusive backup only when tar format is used for taking backup)
10198 : * when backup needs to generate tablespace_map file, it is used to
10199 : * embed escape character before newline character in tablespace path.
10200 : *
10201 : * Returns the minimum WAL location that must be present to restore from this
10202 : * backup, and the corresponding timeline ID in *starttli_p.
10203 : *
10204 : * Every successfully started non-exclusive backup must be stopped by calling
10205 : * do_pg_stop_backup() or do_pg_abort_backup().
10206 : *
10207 : * It is the responsibility of the caller of this function to verify the
10208 : * permissions of the calling user!
10209 : */
10210 : XLogRecPtr
10211 100 : do_pg_start_backup(const char *backupidstr, bool fast, TimeLineID *starttli_p,
10212 : StringInfo labelfile, List **tablespaces,
10213 : StringInfo tblspcmapfile, bool infotbssize,
10214 : bool needtblspcmapfile)
10215 : {
10216 100 : bool exclusive = (labelfile == NULL);
10217 100 : bool backup_started_in_recovery = false;
10218 : XLogRecPtr checkpointloc;
10219 : XLogRecPtr startpoint;
10220 : TimeLineID starttli;
10221 : pg_time_t stamp_time;
10222 : char strfbuf[128];
10223 : char xlogfilename[MAXFNAMELEN];
10224 : XLogSegNo _logSegNo;
10225 : struct stat stat_buf;
10226 : FILE *fp;
10227 :
10228 100 : backup_started_in_recovery = RecoveryInProgress();
10229 :
10230 : /*
10231 : * Currently only non-exclusive backup can be taken during recovery.
10232 : */
10233 100 : if (backup_started_in_recovery && exclusive)
10234 0 : ereport(ERROR,
10235 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10236 : errmsg("recovery is in progress"),
10237 : errhint("WAL control functions cannot be executed during recovery.")));
10238 :
10239 : /*
10240 : * During recovery, we don't need to check WAL level. Because, if WAL
10241 : * level is not sufficient, it's impossible to get here during recovery.
10242 : */
10243 100 : if (!backup_started_in_recovery && !XLogIsNeeded())
10244 0 : ereport(ERROR,
10245 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10246 : errmsg("WAL level not sufficient for making an online backup"),
10247 : errhint("wal_level must be set to \"replica\" or \"logical\" at server start.")));
10248 :
10249 100 : if (strlen(backupidstr) > MAXPGPATH)
10250 0 : ereport(ERROR,
10251 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
10252 : errmsg("backup label too long (max %d bytes)",
10253 : MAXPGPATH)));
10254 :
10255 : /*
10256 : * Mark backup active in shared memory. We must do full-page WAL writes
10257 : * during an on-line backup even if not doing so at other times, because
10258 : * it's quite possible for the backup dump to obtain a "torn" (partially
10259 : * written) copy of a database page if it reads the page concurrently with
10260 : * our write to the same page. This can be fixed as long as the first
10261 : * write to the page in the WAL sequence is a full-page write. Hence, we
10262 : * turn on forcePageWrites and then force a CHECKPOINT, to ensure there
10263 : * are no dirty pages in shared memory that might get dumped while the
10264 : * backup is in progress without having a corresponding WAL record. (Once
10265 : * the backup is complete, we need not force full-page writes anymore,
10266 : * since we expect that any pages not modified during the backup interval
10267 : * must have been correctly captured by the backup.)
10268 : *
10269 : * Note that forcePageWrites has no effect during an online backup from
10270 : * the standby.
10271 : *
10272 : * We must hold all the insertion locks to change the value of
10273 : * forcePageWrites, to ensure adequate interlocking against
10274 : * XLogInsertRecord().
10275 : */
10276 100 : WALInsertLockAcquireExclusive();
10277 100 : if (exclusive)
10278 : {
10279 : /*
10280 : * At first, mark that we're now starting an exclusive backup, to
10281 : * ensure that there are no other sessions currently running
10282 : * pg_start_backup() or pg_stop_backup().
10283 : */
10284 2 : if (XLogCtl->Insert.exclusiveBackupState != EXCLUSIVE_BACKUP_NONE)
10285 : {
10286 0 : WALInsertLockRelease();
10287 0 : ereport(ERROR,
10288 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10289 : errmsg("a backup is already in progress"),
10290 : errhint("Run pg_stop_backup() and try again.")));
10291 : }
10292 2 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_STARTING;
10293 : }
10294 : else
10295 98 : XLogCtl->Insert.nonExclusiveBackups++;
10296 100 : XLogCtl->Insert.forcePageWrites = true;
10297 100 : WALInsertLockRelease();
10298 :
10299 : /* Ensure we release forcePageWrites if fail below */
10300 100 : PG_ENSURE_ERROR_CLEANUP(pg_start_backup_callback, (Datum) BoolGetDatum(exclusive));
10301 : {
10302 100 : bool gotUniqueStartpoint = false;
10303 : DIR *tblspcdir;
10304 : struct dirent *de;
10305 : tablespaceinfo *ti;
10306 : int datadirpathlen;
10307 :
10308 : /*
10309 : * Force an XLOG file switch before the checkpoint, to ensure that the
10310 : * WAL segment the checkpoint is written to doesn't contain pages with
10311 : * old timeline IDs. That would otherwise happen if you called
10312 : * pg_start_backup() right after restoring from a PITR archive: the
10313 : * first WAL segment containing the startup checkpoint has pages in
10314 : * the beginning with the old timeline ID. That can cause trouble at
10315 : * recovery: we won't have a history file covering the old timeline if
10316 : * pg_wal directory was not included in the base backup and the WAL
10317 : * archive was cleared too before starting the backup.
10318 : *
10319 : * This also ensures that we have emitted a WAL page header that has
10320 : * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
10321 : * Therefore, if a WAL archiver (such as pglesslog) is trying to
10322 : * compress out removable backup blocks, it won't remove any that
10323 : * occur after this point.
10324 : *
10325 : * During recovery, we skip forcing XLOG file switch, which means that
10326 : * the backup taken during recovery is not available for the special
10327 : * recovery case described above.
10328 : */
10329 100 : if (!backup_started_in_recovery)
10330 96 : RequestXLogSwitch(false);
10331 :
10332 : do
10333 : {
10334 : bool checkpointfpw;
10335 :
10336 : /*
10337 : * Force a CHECKPOINT. Aside from being necessary to prevent torn
10338 : * page problems, this guarantees that two successive backup runs
10339 : * will have different checkpoint positions and hence different
10340 : * history file names, even if nothing happened in between.
10341 : *
10342 : * During recovery, establish a restartpoint if possible. We use
10343 : * the last restartpoint as the backup starting checkpoint. This
10344 : * means that two successive backup runs can have same checkpoint
10345 : * positions.
10346 : *
10347 : * Since the fact that we are executing do_pg_start_backup()
10348 : * during recovery means that checkpointer is running, we can use
10349 : * RequestCheckpoint() to establish a restartpoint.
10350 : *
10351 : * We use CHECKPOINT_IMMEDIATE only if requested by user (via
10352 : * passing fast = true). Otherwise this can take awhile.
10353 : */
10354 100 : RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_WAIT |
10355 : (fast ? CHECKPOINT_IMMEDIATE : 0));
10356 :
10357 : /*
10358 : * Now we need to fetch the checkpoint record location, and also
10359 : * its REDO pointer. The oldest point in WAL that would be needed
10360 : * to restore starting from the checkpoint is precisely the REDO
10361 : * pointer.
10362 : */
10363 100 : LWLockAcquire(ControlFileLock, LW_SHARED);
10364 100 : checkpointloc = ControlFile->checkPoint;
10365 100 : startpoint = ControlFile->checkPointCopy.redo;
10366 100 : starttli = ControlFile->checkPointCopy.ThisTimeLineID;
10367 100 : checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
10368 100 : LWLockRelease(ControlFileLock);
10369 :
10370 100 : if (backup_started_in_recovery)
10371 : {
10372 : XLogRecPtr recptr;
10373 :
10374 : /*
10375 : * Check to see if all WAL replayed during online backup
10376 : * (i.e., since last restartpoint used as backup starting
10377 : * checkpoint) contain full-page writes.
10378 : */
10379 4 : SpinLockAcquire(&XLogCtl->info_lck);
10380 4 : recptr = XLogCtl->lastFpwDisableRecPtr;
10381 4 : SpinLockRelease(&XLogCtl->info_lck);
10382 :
10383 4 : if (!checkpointfpw || startpoint <= recptr)
10384 0 : ereport(ERROR,
10385 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10386 : errmsg("WAL generated with full_page_writes=off was replayed "
10387 : "since last restartpoint"),
10388 : errhint("This means that the backup being taken on the standby "
10389 : "is corrupt and should not be used. "
10390 : "Enable full_page_writes and run CHECKPOINT on the master, "
10391 : "and then try an online backup again.")));
10392 :
10393 : /*
10394 : * During recovery, since we don't use the end-of-backup WAL
10395 : * record and don't write the backup history file, the
10396 : * starting WAL location doesn't need to be unique. This means
10397 : * that two base backups started at the same time might use
10398 : * the same checkpoint as starting locations.
10399 : */
10400 4 : gotUniqueStartpoint = true;
10401 : }
10402 :
10403 : /*
10404 : * If two base backups are started at the same time (in WAL sender
10405 : * processes), we need to make sure that they use different
10406 : * checkpoints as starting locations, because we use the starting
10407 : * WAL location as a unique identifier for the base backup in the
10408 : * end-of-backup WAL record and when we write the backup history
10409 : * file. Perhaps it would be better generate a separate unique ID
10410 : * for each backup instead of forcing another checkpoint, but
10411 : * taking a checkpoint right after another is not that expensive
10412 : * either because only few buffers have been dirtied yet.
10413 : */
10414 100 : WALInsertLockAcquireExclusive();
10415 100 : if (XLogCtl->Insert.lastBackupStart < startpoint)
10416 : {
10417 100 : XLogCtl->Insert.lastBackupStart = startpoint;
10418 100 : gotUniqueStartpoint = true;
10419 : }
10420 100 : WALInsertLockRelease();
10421 100 : } while (!gotUniqueStartpoint);
10422 :
10423 100 : XLByteToSeg(startpoint, _logSegNo, wal_segment_size);
10424 100 : XLogFileName(xlogfilename, starttli, _logSegNo, wal_segment_size);
10425 :
10426 : /*
10427 : * Construct tablespace_map file
10428 : */
10429 100 : if (exclusive)
10430 2 : tblspcmapfile = makeStringInfo();
10431 :
10432 100 : datadirpathlen = strlen(DataDir);
10433 :
10434 : /* Collect information about all tablespaces */
10435 100 : tblspcdir = AllocateDir("pg_tblspc");
10436 410 : while ((de = ReadDir(tblspcdir, "pg_tblspc")) != NULL)
10437 : {
10438 : char fullpath[MAXPGPATH + 10];
10439 : char linkpath[MAXPGPATH];
10440 210 : char *relpath = NULL;
10441 : int rllen;
10442 : StringInfoData buflinkpath;
10443 210 : char *s = linkpath;
10444 :
10445 : /* Skip special stuff */
10446 210 : if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0)
10447 400 : continue;
10448 :
10449 10 : snprintf(fullpath, sizeof(fullpath), "pg_tblspc/%s", de->d_name);
10450 :
10451 : #if defined(HAVE_READLINK) || defined(WIN32)
10452 10 : rllen = readlink(fullpath, linkpath, sizeof(linkpath));
10453 10 : if (rllen < 0)
10454 : {
10455 0 : ereport(WARNING,
10456 : (errmsg("could not read symbolic link \"%s\": %m",
10457 : fullpath)));
10458 0 : continue;
10459 : }
10460 10 : else if (rllen >= sizeof(linkpath))
10461 : {
10462 0 : ereport(WARNING,
10463 : (errmsg("symbolic link \"%s\" target is too long",
10464 : fullpath)));
10465 0 : continue;
10466 : }
10467 10 : linkpath[rllen] = '\0';
10468 :
10469 : /*
10470 : * Add the escape character '\\' before newline in a string to
10471 : * ensure that we can distinguish between the newline in the
10472 : * tablespace path and end of line while reading tablespace_map
10473 : * file during archive recovery.
10474 : */
10475 10 : initStringInfo(&buflinkpath);
10476 :
10477 644 : while (*s)
10478 : {
10479 624 : if ((*s == '\n' || *s == '\r') && needtblspcmapfile)
10480 0 : appendStringInfoChar(&buflinkpath, '\\');
10481 624 : appendStringInfoChar(&buflinkpath, *s++);
10482 : }
10483 :
10484 : /*
10485 : * Relpath holds the relative path of the tablespace directory
10486 : * when it's located within PGDATA, or NULL if it's located
10487 : * elsewhere.
10488 : */
10489 12 : if (rllen > datadirpathlen &&
10490 2 : strncmp(linkpath, DataDir, datadirpathlen) == 0 &&
10491 0 : IS_DIR_SEP(linkpath[datadirpathlen]))
10492 0 : relpath = linkpath + datadirpathlen + 1;
10493 :
10494 10 : ti = palloc(sizeof(tablespaceinfo));
10495 10 : ti->oid = pstrdup(de->d_name);
10496 10 : ti->path = pstrdup(buflinkpath.data);
10497 10 : ti->rpath = relpath ? pstrdup(relpath) : NULL;
10498 10 : ti->size = infotbssize ? sendTablespace(fullpath, true) : -1;
10499 :
10500 10 : if (tablespaces)
10501 10 : *tablespaces = lappend(*tablespaces, ti);
10502 :
10503 10 : appendStringInfo(tblspcmapfile, "%s %s\n", ti->oid, ti->path);
10504 :
10505 10 : pfree(buflinkpath.data);
10506 : #else
10507 :
10508 : /*
10509 : * If the platform does not have symbolic links, it should not be
10510 : * possible to have tablespaces - clearly somebody else created
10511 : * them. Warn about it and ignore.
10512 : */
10513 : ereport(WARNING,
10514 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
10515 : errmsg("tablespaces are not supported on this platform")));
10516 : #endif
10517 : }
10518 100 : FreeDir(tblspcdir);
10519 :
10520 : /*
10521 : * Construct backup label file
10522 : */
10523 100 : if (exclusive)
10524 2 : labelfile = makeStringInfo();
10525 :
10526 : /* Use the log timezone here, not the session timezone */
10527 100 : stamp_time = (pg_time_t) time(NULL);
10528 100 : pg_strftime(strfbuf, sizeof(strfbuf),
10529 : "%Y-%m-%d %H:%M:%S %Z",
10530 100 : pg_localtime(&stamp_time, log_timezone));
10531 200 : appendStringInfo(labelfile, "START WAL LOCATION: %X/%X (file %s)\n",
10532 100 : (uint32) (startpoint >> 32), (uint32) startpoint, xlogfilename);
10533 200 : appendStringInfo(labelfile, "CHECKPOINT LOCATION: %X/%X\n",
10534 100 : (uint32) (checkpointloc >> 32), (uint32) checkpointloc);
10535 100 : appendStringInfo(labelfile, "BACKUP METHOD: %s\n",
10536 : exclusive ? "pg_start_backup" : "streamed");
10537 100 : appendStringInfo(labelfile, "BACKUP FROM: %s\n",
10538 : backup_started_in_recovery ? "standby" : "master");
10539 100 : appendStringInfo(labelfile, "START TIME: %s\n", strfbuf);
10540 100 : appendStringInfo(labelfile, "LABEL: %s\n", backupidstr);
10541 100 : appendStringInfo(labelfile, "START TIMELINE: %u\n", starttli);
10542 :
10543 : /*
10544 : * Okay, write the file, or return its contents to caller.
10545 : */
10546 100 : if (exclusive)
10547 : {
10548 : /*
10549 : * Check for existing backup label --- implies a backup is already
10550 : * running. (XXX given that we checked exclusiveBackupState
10551 : * above, maybe it would be OK to just unlink any such label
10552 : * file?)
10553 : */
10554 2 : if (stat(BACKUP_LABEL_FILE, &stat_buf) != 0)
10555 : {
10556 2 : if (errno != ENOENT)
10557 0 : ereport(ERROR,
10558 : (errcode_for_file_access(),
10559 : errmsg("could not stat file \"%s\": %m",
10560 : BACKUP_LABEL_FILE)));
10561 : }
10562 : else
10563 0 : ereport(ERROR,
10564 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10565 : errmsg("a backup is already in progress"),
10566 : errhint("If you're sure there is no backup in progress, remove file \"%s\" and try again.",
10567 : BACKUP_LABEL_FILE)));
10568 :
10569 2 : fp = AllocateFile(BACKUP_LABEL_FILE, "w");
10570 :
10571 2 : if (!fp)
10572 0 : ereport(ERROR,
10573 : (errcode_for_file_access(),
10574 : errmsg("could not create file \"%s\": %m",
10575 : BACKUP_LABEL_FILE)));
10576 4 : if (fwrite(labelfile->data, labelfile->len, 1, fp) != 1 ||
10577 4 : fflush(fp) != 0 ||
10578 4 : pg_fsync(fileno(fp)) != 0 ||
10579 4 : ferror(fp) ||
10580 2 : FreeFile(fp))
10581 0 : ereport(ERROR,
10582 : (errcode_for_file_access(),
10583 : errmsg("could not write file \"%s\": %m",
10584 : BACKUP_LABEL_FILE)));
10585 : /* Allocated locally for exclusive backups, so free separately */
10586 2 : pfree(labelfile->data);
10587 2 : pfree(labelfile);
10588 :
10589 : /* Write backup tablespace_map file. */
10590 2 : if (tblspcmapfile->len > 0)
10591 : {
10592 0 : if (stat(TABLESPACE_MAP, &stat_buf) != 0)
10593 : {
10594 0 : if (errno != ENOENT)
10595 0 : ereport(ERROR,
10596 : (errcode_for_file_access(),
10597 : errmsg("could not stat file \"%s\": %m",
10598 : TABLESPACE_MAP)));
10599 : }
10600 : else
10601 0 : ereport(ERROR,
10602 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10603 : errmsg("a backup is already in progress"),
10604 : errhint("If you're sure there is no backup in progress, remove file \"%s\" and try again.",
10605 : TABLESPACE_MAP)));
10606 :
10607 0 : fp = AllocateFile(TABLESPACE_MAP, "w");
10608 :
10609 0 : if (!fp)
10610 0 : ereport(ERROR,
10611 : (errcode_for_file_access(),
10612 : errmsg("could not create file \"%s\": %m",
10613 : TABLESPACE_MAP)));
10614 0 : if (fwrite(tblspcmapfile->data, tblspcmapfile->len, 1, fp) != 1 ||
10615 0 : fflush(fp) != 0 ||
10616 0 : pg_fsync(fileno(fp)) != 0 ||
10617 0 : ferror(fp) ||
10618 0 : FreeFile(fp))
10619 0 : ereport(ERROR,
10620 : (errcode_for_file_access(),
10621 : errmsg("could not write file \"%s\": %m",
10622 : TABLESPACE_MAP)));
10623 : }
10624 :
10625 : /* Allocated locally for exclusive backups, so free separately */
10626 2 : pfree(tblspcmapfile->data);
10627 2 : pfree(tblspcmapfile);
10628 : }
10629 : }
10630 100 : PG_END_ENSURE_ERROR_CLEANUP(pg_start_backup_callback, (Datum) BoolGetDatum(exclusive));
10631 :
10632 : /*
10633 : * Mark that start phase has correctly finished for an exclusive backup.
10634 : * Session-level locks are updated as well to reflect that state.
10635 : *
10636 : * Note that CHECK_FOR_INTERRUPTS() must not occur while updating backup
10637 : * counters and session-level lock. Otherwise they can be updated
10638 : * inconsistently, and which might cause do_pg_abort_backup() to fail.
10639 : */
10640 100 : if (exclusive)
10641 : {
10642 2 : WALInsertLockAcquireExclusive();
10643 2 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_IN_PROGRESS;
10644 :
10645 : /* Set session-level lock */
10646 2 : sessionBackupState = SESSION_BACKUP_EXCLUSIVE;
10647 2 : WALInsertLockRelease();
10648 : }
10649 : else
10650 98 : sessionBackupState = SESSION_BACKUP_NON_EXCLUSIVE;
10651 :
10652 : /*
10653 : * We're done. As a convenience, return the starting WAL location.
10654 : */
10655 100 : if (starttli_p)
10656 98 : *starttli_p = starttli;
10657 100 : return startpoint;
10658 : }
10659 :
10660 : /* Error cleanup callback for pg_start_backup */
10661 : static void
10662 0 : pg_start_backup_callback(int code, Datum arg)
10663 : {
10664 0 : bool exclusive = DatumGetBool(arg);
10665 :
10666 : /* Update backup counters and forcePageWrites on failure */
10667 0 : WALInsertLockAcquireExclusive();
10668 0 : if (exclusive)
10669 : {
10670 : Assert(XLogCtl->Insert.exclusiveBackupState == EXCLUSIVE_BACKUP_STARTING);
10671 0 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_NONE;
10672 : }
10673 : else
10674 : {
10675 : Assert(XLogCtl->Insert.nonExclusiveBackups > 0);
10676 0 : XLogCtl->Insert.nonExclusiveBackups--;
10677 : }
10678 :
10679 0 : if (XLogCtl->Insert.exclusiveBackupState == EXCLUSIVE_BACKUP_NONE &&
10680 0 : XLogCtl->Insert.nonExclusiveBackups == 0)
10681 : {
10682 0 : XLogCtl->Insert.forcePageWrites = false;
|
